Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
-- PARTITION_JOIN
2018-02-16 16:33:59 +01:00
-- Test partitionwise join between partitioned tables
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
2018-02-16 16:33:59 +01:00
-- Enable partitionwise join, which by default is disabled.
SET enable_partitionwise_join to true;
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
-- partitioned by a single column
--
CREATE TABLE prt1 (a int, b int, c varchar) PARTITION BY RANGE(a);
CREATE TABLE prt1_p1 PARTITION OF prt1 FOR VALUES FROM (0) TO (250);
CREATE TABLE prt1_p3 PARTITION OF prt1 FOR VALUES FROM (500) TO (600);
CREATE TABLE prt1_p2 PARTITION OF prt1 FOR VALUES FROM (250) TO (500);
INSERT INTO prt1 SELECT i, i % 25, to_char(i, 'FM0000') FROM generate_series(0, 599) i WHERE i % 2 = 0;
CREATE INDEX iprt1_p1_a on prt1_p1(a);
CREATE INDEX iprt1_p2_a on prt1_p2(a);
CREATE INDEX iprt1_p3_a on prt1_p3(a);
ANALYZE prt1;
CREATE TABLE prt2 (a int, b int, c varchar) PARTITION BY RANGE(b);
CREATE TABLE prt2_p1 PARTITION OF prt2 FOR VALUES FROM (0) TO (250);
CREATE TABLE prt2_p2 PARTITION OF prt2 FOR VALUES FROM (250) TO (500);
CREATE TABLE prt2_p3 PARTITION OF prt2 FOR VALUES FROM (500) TO (600);
INSERT INTO prt2 SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(0, 599) i WHERE i % 3 = 0;
CREATE INDEX iprt2_p1_b on prt2_p1(b);
CREATE INDEX iprt2_p2_b on prt2_p2(b);
CREATE INDEX iprt2_p3_b on prt2_p3(b);
ANALYZE prt2;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt2 t2 WHERE t1.a = t2.b AND t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
--------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
(21 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt2 t2 WHERE t1.a = t2.b AND t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | 0 | 0000
150 | 0150 | 150 | 0150
300 | 0300 | 300 | 0300
450 | 0450 | 450 | 0450
(4 rows)
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
-- left outer join, with whole-row reference; partitionwise join does not apply
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1, t2 FROM prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b WHERE t1.b = 0 ORDER BY t1.a, t2.b;
2018-03-29 21:47:57 +02:00
QUERY PLAN
--------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a, t2.b
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
-> Hash Right Join
Hash Cond: (t2.b = t1.a)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
-> Seq Scan on prt2_p2 t2_2
-> Seq Scan on prt2_p3 t2_3
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p3 t1_3
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
(16 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1, t2 FROM prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b WHERE t1.b = 0 ORDER BY t1.a, t2.b;
t1 | t2
--------------+--------------
(0,0,0000) | (0,0,0000)
(50,0,0050) |
(100,0,0100) |
(150,0,0150) | (0,150,0150)
(200,0,0200) |
(250,0,0250) |
(300,0,0300) | (0,300,0300)
(350,0,0350) |
(400,0,0400) |
(450,0,0450) | (0,450,0450)
(500,0,0500) |
(550,0,0550) |
(12 rows)
-- right outer join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1 RIGHT JOIN prt2 t2 ON t1.a = t2.b WHERE t2.a = 0 ORDER BY t1.a, t2.b;
2018-03-29 21:47:57 +02:00
QUERY PLAN
---------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a, t2.b
2018-03-29 21:47:57 +02:00
-> Append
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_1.a = t2_1.b)
-> Seq Scan on prt1_p1 t1_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
2018-03-29 21:47:57 +02:00
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_2.a = t2_2.b)
-> Seq Scan on prt1_p2 t1_2
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
-> Nested Loop Left Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 t2_3
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_p3_a on prt1_p3 t1_3
Index Cond: (a = t2_3.b)
2018-03-29 21:47:57 +02:00
(20 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1 RIGHT JOIN prt2 t2 ON t1.a = t2.b WHERE t2.a = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | 0 | 0000
150 | 0150 | 150 | 0150
300 | 0300 | 300 | 0300
450 | 0450 | 450 | 0450
| | 75 | 0075
| | 225 | 0225
| | 375 | 0375
| | 525 | 0525
(8 rows)
-- full outer join, with placeholder vars
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT 50 phv, * FROM prt1 WHERE prt1.b = 0) t1 FULL JOIN (SELECT 75 phv, * FROM prt2 WHERE prt2.a = 0) t2 ON (t1.a = t2.b) WHERE t1.phv = t1.a OR t2.phv = t2.b ORDER BY t1.a, t2.b;
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
QUERY PLAN
----------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a, prt2.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
-> Hash Full Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Hash Cond: (prt1_1.a = prt2_1.b)
Filter: (((50) = prt1_1.a) OR ((75) = prt2_1.b))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 prt1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 prt2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Hash Full Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Hash Cond: (prt1_2.a = prt2_2.b)
Filter: (((50) = prt1_2.a) OR ((75) = prt2_2.b))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 prt1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 prt2_2
Filter: (a = 0)
-> Hash Full Join
Hash Cond: (prt1_3.a = prt2_3.b)
Filter: (((50) = prt1_3.a) OR ((75) = prt2_3.b))
-> Seq Scan on prt1_p3 prt1_3
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_p3 prt2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
(27 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT 50 phv, * FROM prt1 WHERE prt1.b = 0) t1 FULL JOIN (SELECT 75 phv, * FROM prt2 WHERE prt2.a = 0) t2 ON (t1.a = t2.b) WHERE t1.phv = t1.a OR t2.phv = t2.b ORDER BY t1.a, t2.b;
a | c | b | c
----+------+----+------
50 | 0050 | |
| | 75 | 0075
(2 rows)
-- Join with pruned partitions from joining relations
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt2 t2 WHERE t1.a = t2.b AND t1.a < 450 AND t2.b > 250 AND t1.b = 0 ORDER BY t1.a, t2.b;
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
QUERY PLAN
-----------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
-> Hash Join
Hash Cond: (t2.b = t1.a)
-> Seq Scan on prt2_p2 t2
Filter: (b > 250)
-> Hash
-> Seq Scan on prt1_p2 t1
Filter: ((a < 450) AND (b = 0))
(9 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt2 t2 WHERE t1.a = t2.b AND t1.a < 450 AND t2.b > 250 AND t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
300 | 0300 | 300 | 0300
(1 row)
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-- Currently we can't do partitioned join if nullable-side partitions are pruned
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a < 450) t1 LEFT JOIN (SELECT * FROM prt2 WHERE b > 250) t2 ON t1.a = t2.b WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
-----------------------------------------------------------
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a, prt2.b
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Hash Right Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Hash Cond: (prt2.b = prt1.a)
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 prt2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b > 250)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 prt2_2
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
Filter: (b > 250)
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 prt1_1
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
Filter: ((a < 450) AND (b = 0))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 prt1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: ((a < 450) AND (b = 0))
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
(15 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a < 450) t1 LEFT JOIN (SELECT * FROM prt2 WHERE b > 250) t2 ON t1.a = t2.b WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | |
50 | 0050 | |
100 | 0100 | |
150 | 0150 | |
200 | 0200 | |
250 | 0250 | |
300 | 0300 | 300 | 0300
350 | 0350 | |
400 | 0400 | |
(9 rows)
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-- Currently we can't do partitioned join if nullable-side partitions are pruned
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a < 450) t1 FULL JOIN (SELECT * FROM prt2 WHERE b > 250) t2 ON t1.a = t2.b WHERE t1.b = 0 OR t2.a = 0 ORDER BY t1.a, t2.b;
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
QUERY PLAN
----------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a, prt2.b
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Hash Full Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Hash Cond: (prt1.a = prt2.b)
Filter: ((prt1.b = 0) OR (prt2.a = 0))
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 prt1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a < 450)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 prt1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a < 450)
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 prt2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b > 250)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 prt2_2
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
Filter: (b > 250)
(16 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a < 450) t1 FULL JOIN (SELECT * FROM prt2 WHERE b > 250) t2 ON t1.a = t2.b WHERE t1.b = 0 OR t2.a = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | |
50 | 0050 | |
100 | 0100 | |
150 | 0150 | |
200 | 0200 | |
250 | 0250 | |
300 | 0300 | 300 | 0300
350 | 0350 | |
400 | 0400 | |
| | 375 | 0375
| | 450 | 0450
| | 525 | 0525
(12 rows)
-- Semi-join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t2.b FROM prt2 t2 WHERE t2.a = 0) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_1.a = t2_1.b)
-> Seq Scan on prt1_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Hash Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_2.a = t2_2.b)
-> Seq Scan on prt1_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Nested Loop Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Join Filter: (t1_3.a = t2_3.b)
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Materialize
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
(24 rows)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t2.b FROM prt2 t2 WHERE t2.a = 0) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
0 | 0 | 0000
150 | 0 | 0150
300 | 0 | 0300
450 | 0 | 0450
(4 rows)
-- Anti-join with aggregates
EXPLAIN (COSTS OFF)
SELECT sum(t1.a), avg(t1.a), sum(t1.b), avg(t1.b) FROM prt1 t1 WHERE NOT EXISTS (SELECT 1 FROM prt2 t2 WHERE t1.a = t2.b);
QUERY PLAN
--------------------------------------------------
Aggregate
-> Append
-> Hash Anti Join
Hash Cond: (t1_1.a = t2_1.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Anti Join
Hash Cond: (t1_2.a = t2_2.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
-> Hash Anti Join
Hash Cond: (t1_3.a = t2_3.b)
-> Seq Scan on prt1_p3 t1_3
-> Hash
-> Seq Scan on prt2_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(17 rows)
SELECT sum(t1.a), avg(t1.a), sum(t1.b), avg(t1.b) FROM prt1 t1 WHERE NOT EXISTS (SELECT 1 FROM prt2 t2 WHERE t1.a = t2.b);
sum | avg | sum | avg
-------+----------------------+------+---------------------
60000 | 300.0000000000000000 | 2400 | 12.0000000000000000
(1 row)
-- lateral reference
EXPLAIN (COSTS OFF)
SELECT * FROM prt1 t1 LEFT JOIN LATERAL
(SELECT t2.a AS t2a, t3.a AS t3a, least(t1.a,t2.a,t3.b) FROM prt1 t2 JOIN prt2 t3 ON (t2.a = t3.b)) ss
ON t1.a = ss.t2a WHERE t1.b = 0 ORDER BY t1.a;
2018-03-29 21:47:57 +02:00
QUERY PLAN
--------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a
2018-03-29 21:47:57 +02:00
-> Append
-> Nested Loop Left Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
-> Nested Loop
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Only Scan using iprt1_p1_a on prt1_p1 t2_1
2018-03-29 21:47:57 +02:00
Index Cond: (a = t1_1.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p1_b on prt2_p1 t3_1
2018-03-29 21:47:57 +02:00
Index Cond: (b = t2_1.a)
-> Nested Loop Left Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
-> Nested Loop
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Only Scan using iprt1_p2_a on prt1_p2 t2_2
2018-03-29 21:47:57 +02:00
Index Cond: (a = t1_2.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p2_b on prt2_p2 t3_2
2018-03-29 21:47:57 +02:00
Index Cond: (b = t2_2.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Nested Loop Left Join
-> Seq Scan on prt1_p3 t1_3
Filter: (b = 0)
-> Nested Loop
-> Index Only Scan using iprt1_p3_a on prt1_p3 t2_3
Index Cond: (a = t1_3.a)
-> Index Scan using iprt2_p3_b on prt2_p3 t3_3
Index Cond: (b = t2_3.a)
2018-03-29 21:47:57 +02:00
(27 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT * FROM prt1 t1 LEFT JOIN LATERAL
(SELECT t2.a AS t2a, t3.a AS t3a, least(t1.a,t2.a,t3.b) FROM prt1 t2 JOIN prt2 t3 ON (t2.a = t3.b)) ss
ON t1.a = ss.t2a WHERE t1.b = 0 ORDER BY t1.a;
a | b | c | t2a | t3a | least
-----+---+------+-----+-----+-------
0 | 0 | 0000 | 0 | 0 | 0
50 | 0 | 0050 | | |
100 | 0 | 0100 | | |
150 | 0 | 0150 | 150 | 0 | 150
200 | 0 | 0200 | | |
250 | 0 | 0250 | | |
300 | 0 | 0300 | 300 | 0 | 300
350 | 0 | 0350 | | |
400 | 0 | 0400 | | |
450 | 0 | 0450 | 450 | 0 | 450
500 | 0 | 0500 | | |
550 | 0 | 0550 | | |
(12 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, ss.t2a, ss.t2c FROM prt1 t1 LEFT JOIN LATERAL
(SELECT t2.a AS t2a, t3.a AS t3a, t2.b t2b, t2.c t2c, least(t1.a,t2.a,t3.b) FROM prt1 t2 JOIN prt2 t3 ON (t2.a = t3.b)) ss
ON t1.c = ss.t2c WHERE (t1.b + coalesce(ss.t2b, 0)) = 0 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Left Join
Hash Cond: ((t1.c)::text = (t2.c)::text)
Filter: ((t1.b + COALESCE(t2.b, 0)) = 0)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
-> Seq Scan on prt1_p2 t1_2
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
-> Hash Join
Hash Cond: (t2_1.a = t3_1.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t3_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Join
Hash Cond: (t2_2.a = t3_2.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t2_2
-> Hash
-> Seq Scan on prt2_p2 t3_2
-> Hash Join
Hash Cond: (t2_3.a = t3_3.b)
-> Seq Scan on prt1_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 t3_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(26 rows)
SELECT t1.a, ss.t2a, ss.t2c FROM prt1 t1 LEFT JOIN LATERAL
(SELECT t2.a AS t2a, t3.a AS t3a, t2.b t2b, t2.c t2c, least(t1.a,t2.a,t3.a) FROM prt1 t2 JOIN prt2 t3 ON (t2.a = t3.b)) ss
ON t1.c = ss.t2c WHERE (t1.b + coalesce(ss.t2b, 0)) = 0 ORDER BY t1.a;
a | t2a | t2c
-----+-----+------
0 | 0 | 0000
50 | |
100 | |
150 | 150 | 0150
200 | |
250 | |
300 | 300 | 0300
350 | |
400 | |
450 | 450 | 0450
500 | |
550 | |
(12 rows)
2019-11-05 17:42:24 +01:00
-- bug with inadequate sort key representation
SET enable_partitionwise_aggregate TO true;
SET enable_hashjoin TO false;
EXPLAIN (COSTS OFF)
SELECT a, b FROM prt1 FULL JOIN prt2 p2(b,a,c) USING(a,b)
WHERE a BETWEEN 490 AND 510
GROUP BY 1, 2 ORDER BY 1, 2;
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
QUERY PLAN
-----------------------------------------------------------------------------------------------------------------
2019-11-05 17:42:24 +01:00
Group
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Group Key: (COALESCE(prt1.a, p2.a)), (COALESCE(prt1.b, p2.b))
2019-11-05 17:42:24 +01:00
-> Merge Append
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: (COALESCE(prt1.a, p2.a)), (COALESCE(prt1.b, p2.b))
2019-11-05 17:42:24 +01:00
-> Group
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Group Key: (COALESCE(prt1.a, p2.a)), (COALESCE(prt1.b, p2.b))
2019-11-05 17:42:24 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: (COALESCE(prt1.a, p2.a)), (COALESCE(prt1.b, p2.b))
2019-11-05 17:42:24 +01:00
-> Merge Full Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Merge Cond: ((prt1.a = p2.a) AND (prt1.b = p2.b))
Filter: ((COALESCE(prt1.a, p2.a) >= 490) AND (COALESCE(prt1.a, p2.a) <= 510))
2019-11-05 17:42:24 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a, prt1.b
-> Seq Scan on prt1_p1 prt1
2019-11-05 17:42:24 +01:00
-> Sort
Sort Key: p2.a, p2.b
-> Seq Scan on prt2_p1 p2
-> Group
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Group Key: (COALESCE(prt1_1.a, p2_1.a)), (COALESCE(prt1_1.b, p2_1.b))
2019-11-05 17:42:24 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: (COALESCE(prt1_1.a, p2_1.a)), (COALESCE(prt1_1.b, p2_1.b))
2019-11-05 17:42:24 +01:00
-> Merge Full Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Merge Cond: ((prt1_1.a = p2_1.a) AND (prt1_1.b = p2_1.b))
Filter: ((COALESCE(prt1_1.a, p2_1.a) >= 490) AND (COALESCE(prt1_1.a, p2_1.a) <= 510))
2019-11-05 17:42:24 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1_1.a, prt1_1.b
-> Seq Scan on prt1_p2 prt1_1
2019-11-05 17:42:24 +01:00
-> Sort
Sort Key: p2_1.a, p2_1.b
-> Seq Scan on prt2_p2 p2_1
-> Group
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Group Key: (COALESCE(prt1_2.a, p2_2.a)), (COALESCE(prt1_2.b, p2_2.b))
2019-11-05 17:42:24 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: (COALESCE(prt1_2.a, p2_2.a)), (COALESCE(prt1_2.b, p2_2.b))
2019-11-05 17:42:24 +01:00
-> Merge Full Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Merge Cond: ((prt1_2.a = p2_2.a) AND (prt1_2.b = p2_2.b))
Filter: ((COALESCE(prt1_2.a, p2_2.a) >= 490) AND (COALESCE(prt1_2.a, p2_2.a) <= 510))
2019-11-05 17:42:24 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1_2.a, prt1_2.b
-> Seq Scan on prt1_p3 prt1_2
2019-11-05 17:42:24 +01:00
-> Sort
Sort Key: p2_2.a, p2_2.b
-> Seq Scan on prt2_p3 p2_2
(43 rows)
SELECT a, b FROM prt1 FULL JOIN prt2 p2(b,a,c) USING(a,b)
WHERE a BETWEEN 490 AND 510
GROUP BY 1, 2 ORDER BY 1, 2;
a | b
-----+----
490 | 15
492 | 17
494 | 19
495 | 20
496 | 21
498 | 23
500 | 0
501 | 1
502 | 2
504 | 4
506 | 6
507 | 7
508 | 8
510 | 10
(14 rows)
RESET enable_partitionwise_aggregate;
RESET enable_hashjoin;
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
-- partitioned by expression
--
CREATE TABLE prt1_e (a int, b int, c int) PARTITION BY RANGE(((a + b)/2));
CREATE TABLE prt1_e_p1 PARTITION OF prt1_e FOR VALUES FROM (0) TO (250);
CREATE TABLE prt1_e_p2 PARTITION OF prt1_e FOR VALUES FROM (250) TO (500);
CREATE TABLE prt1_e_p3 PARTITION OF prt1_e FOR VALUES FROM (500) TO (600);
INSERT INTO prt1_e SELECT i, i, i % 25 FROM generate_series(0, 599, 2) i;
CREATE INDEX iprt1_e_p1_ab2 on prt1_e_p1(((a+b)/2));
CREATE INDEX iprt1_e_p2_ab2 on prt1_e_p2(((a+b)/2));
CREATE INDEX iprt1_e_p3_ab2 on prt1_e_p3(((a+b)/2));
ANALYZE prt1_e;
CREATE TABLE prt2_e (a int, b int, c int) PARTITION BY RANGE(((b + a)/2));
CREATE TABLE prt2_e_p1 PARTITION OF prt2_e FOR VALUES FROM (0) TO (250);
CREATE TABLE prt2_e_p2 PARTITION OF prt2_e FOR VALUES FROM (250) TO (500);
CREATE TABLE prt2_e_p3 PARTITION OF prt2_e FOR VALUES FROM (500) TO (600);
INSERT INTO prt2_e SELECT i, i, i % 25 FROM generate_series(0, 599, 3) i;
ANALYZE prt2_e;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_e t1, prt2_e t2 WHERE (t1.a + t1.b)/2 = (t2.b + t2.a)/2 AND t1.c = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.b
-> Append
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t2_1.b + t2_1.a) / 2) = ((t1_1.a + t1_1.b) / 2))
-> Seq Scan on prt2_e_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t2_2.b + t2_2.a) / 2) = ((t1_2.a + t1_2.b) / 2))
-> Seq Scan on prt2_e_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t2_3.b + t2_3.a) / 2) = ((t1_3.a + t1_3.b) / 2))
-> Seq Scan on prt2_e_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
(21 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_e t1, prt2_e t2 WHERE (t1.a + t1.b)/2 = (t2.b + t2.a)/2 AND t1.c = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+---+-----+---
0 | 0 | 0 | 0
150 | 0 | 150 | 0
300 | 0 | 300 | 0
450 | 0 | 450 | 0
(4 rows)
--
-- N-way join
--
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM prt1 t1, prt2 t2, prt1_e t3 WHERE t1.a = t2.b AND t1.a = (t3.a + t3.b)/2 AND t1.b = 0 ORDER BY t1.a, t2.b;
2018-03-29 21:47:57 +02:00
QUERY PLAN
---------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a
2018-03-29 21:47:57 +02:00
-> Append
-> Nested Loop
Join Filter: (t1_1.a = ((t3_1.a + t3_1.b) / 2))
-> Hash Join
Hash Cond: (t2_1.b = t1_1.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_e_p1_ab2 on prt1_e_p1 t3_1
2018-03-29 21:47:57 +02:00
Index Cond: (((a + b) / 2) = t2_1.b)
-> Nested Loop
Join Filter: (t1_2.a = ((t3_2.a + t3_2.b) / 2))
-> Hash Join
Hash Cond: (t2_2.b = t1_2.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_e_p2_ab2 on prt1_e_p2 t3_2
2018-03-29 21:47:57 +02:00
Index Cond: (((a + b) / 2) = t2_2.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Nested Loop
Join Filter: (t1_3.a = ((t3_3.a + t3_3.b) / 2))
-> Hash Join
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_p3 t2_3
-> Hash
-> Seq Scan on prt1_p3 t1_3
Filter: (b = 0)
-> Index Scan using iprt1_e_p3_ab2 on prt1_e_p3 t3_3
Index Cond: (((a + b) / 2) = t2_3.b)
2018-03-29 21:47:57 +02:00
(33 rows)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM prt1 t1, prt2 t2, prt1_e t3 WHERE t1.a = t2.b AND t1.a = (t3.a + t3.b)/2 AND t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c | ?column? | c
-----+------+-----+------+----------+---
0 | 0000 | 0 | 0000 | 0 | 0
150 | 0150 | 150 | 0150 | 300 | 0
300 | 0300 | 300 | 0300 | 600 | 0
450 | 0450 | 450 | 0450 | 900 | 0
(4 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM (prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b) LEFT JOIN prt1_e t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t1.b = 0 ORDER BY t1.a, t2.b, t3.a + t3.b;
QUERY PLAN
--------------------------------------------------------------
Sort
Sort Key: t1.a, t2.b, ((t3.a + t3.b))
-> Append
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t3_1.a + t3_1.b) / 2) = t1_1.a)
-> Seq Scan on prt1_e_p1 t3_1
2018-03-29 21:47:57 +02:00
-> Hash
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
2018-03-29 21:47:57 +02:00
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t3_2.a + t3_2.b) / 2) = t1_2.a)
-> Seq Scan on prt1_e_p2 t3_2
2018-03-29 21:47:57 +02:00
-> Hash
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
2018-03-29 21:47:57 +02:00
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t3_3.a + t3_3.b) / 2) = t1_3.a)
-> Seq Scan on prt1_e_p3 t3_3
2018-03-29 21:47:57 +02:00
-> Hash
-> Hash Right Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
2018-03-29 21:47:57 +02:00
(33 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM (prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b) LEFT JOIN prt1_e t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t1.b = 0 ORDER BY t1.a, t2.b, t3.a + t3.b;
a | c | b | c | ?column? | c
-----+------+-----+------+----------+---
0 | 0000 | 0 | 0000 | 0 | 0
50 | 0050 | | | 100 | 0
100 | 0100 | | | 200 | 0
150 | 0150 | 150 | 0150 | 300 | 0
200 | 0200 | | | 400 | 0
250 | 0250 | | | 500 | 0
300 | 0300 | 300 | 0300 | 600 | 0
350 | 0350 | | | 700 | 0
400 | 0400 | | | 800 | 0
450 | 0450 | 450 | 0450 | 900 | 0
500 | 0500 | | | 1000 | 0
550 | 0550 | | | 1100 | 0
(12 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM (prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b) RIGHT JOIN prt1_e t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t3.c = 0 ORDER BY t1.a, t2.b, t3.a + t3.b;
2018-03-29 21:47:57 +02:00
QUERY PLAN
-------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a, t2.b, ((t3.a + t3.b))
2018-03-29 21:47:57 +02:00
-> Append
-> Nested Loop Left Join
-> Hash Right Join
Hash Cond: (t1_1.a = ((t3_1.a + t3_1.b) / 2))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p1 t3_1
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p1_b on prt2_p1 t2_1
Refactor the representation of indexable clauses in IndexPaths.
In place of three separate but interrelated lists (indexclauses,
indexquals, and indexqualcols), an IndexPath now has one list
"indexclauses" of IndexClause nodes. This holds basically the same
information as before, but in a more useful format: in particular, there
is now a clear connection between an indexclause (an original restriction
clause from WHERE or JOIN/ON) and the indexquals (directly usable index
conditions) derived from it.
We also change the ground rules a bit by mandating that clause commutation,
if needed, be done up-front so that what is stored in the indexquals list
is always directly usable as an index condition. This gets rid of repeated
re-determination of which side of the clause is the indexkey during costing
and plan generation, as well as repeated lookups of the commutator
operator. To minimize the added up-front cost, the typical case of
commuting a plain OpExpr is handled by a new special-purpose function
commute_restrictinfo(). For RowCompareExprs, generating the new clause
properly commuted to begin with is not really any more complex than before,
it's just different --- and we can save doing that work twice, as the
pretty-klugy original implementation did.
Tracking the connection between original and derived clauses lets us
also track explicitly whether the derived clauses are an exact or lossy
translation of the original. This provides a cheap solution to getting
rid of unnecessary rechecks of boolean index clauses, which previously
seemed like it'd be more expensive than it was worth.
Another pleasant (IMO) side-effect is that EXPLAIN now always shows
index clauses with the indexkey on the left; this seems less confusing.
This commit leaves expand_indexqual_conditions() and some related
functions in a slightly messy state. I didn't bother to change them
any more than minimally necessary to work with the new data structure,
because all that code is going to be refactored out of existence in
a follow-on patch.
Discussion: https://postgr.es/m/22182.1549124950@sss.pgh.pa.us
2019-02-09 23:30:43 +01:00
Index Cond: (b = t1_1.a)
2018-03-29 21:47:57 +02:00
-> Nested Loop Left Join
-> Hash Right Join
Hash Cond: (t1_2.a = ((t3_2.a + t3_2.b) / 2))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p2 t3_2
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p2_b on prt2_p2 t2_2
Refactor the representation of indexable clauses in IndexPaths.
In place of three separate but interrelated lists (indexclauses,
indexquals, and indexqualcols), an IndexPath now has one list
"indexclauses" of IndexClause nodes. This holds basically the same
information as before, but in a more useful format: in particular, there
is now a clear connection between an indexclause (an original restriction
clause from WHERE or JOIN/ON) and the indexquals (directly usable index
conditions) derived from it.
We also change the ground rules a bit by mandating that clause commutation,
if needed, be done up-front so that what is stored in the indexquals list
is always directly usable as an index condition. This gets rid of repeated
re-determination of which side of the clause is the indexkey during costing
and plan generation, as well as repeated lookups of the commutator
operator. To minimize the added up-front cost, the typical case of
commuting a plain OpExpr is handled by a new special-purpose function
commute_restrictinfo(). For RowCompareExprs, generating the new clause
properly commuted to begin with is not really any more complex than before,
it's just different --- and we can save doing that work twice, as the
pretty-klugy original implementation did.
Tracking the connection between original and derived clauses lets us
also track explicitly whether the derived clauses are an exact or lossy
translation of the original. This provides a cheap solution to getting
rid of unnecessary rechecks of boolean index clauses, which previously
seemed like it'd be more expensive than it was worth.
Another pleasant (IMO) side-effect is that EXPLAIN now always shows
index clauses with the indexkey on the left; this seems less confusing.
This commit leaves expand_indexqual_conditions() and some related
functions in a slightly messy state. I didn't bother to change them
any more than minimally necessary to work with the new data structure,
because all that code is going to be refactored out of existence in
a follow-on patch.
Discussion: https://postgr.es/m/22182.1549124950@sss.pgh.pa.us
2019-02-09 23:30:43 +01:00
Index Cond: (b = t1_2.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Nested Loop Left Join
-> Hash Right Join
Hash Cond: (t1_3.a = ((t3_3.a + t3_3.b) / 2))
-> Seq Scan on prt1_p3 t1_3
-> Hash
-> Seq Scan on prt1_e_p3 t3_3
Filter: (c = 0)
-> Index Scan using iprt2_p3_b on prt2_p3 t2_3
Index Cond: (b = t1_3.a)
2018-03-29 21:47:57 +02:00
(30 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM (prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b) RIGHT JOIN prt1_e t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t3.c = 0 ORDER BY t1.a, t2.b, t3.a + t3.b;
a | c | b | c | ?column? | c
-----+------+-----+------+----------+---
0 | 0000 | 0 | 0000 | 0 | 0
50 | 0050 | | | 100 | 0
100 | 0100 | | | 200 | 0
150 | 0150 | 150 | 0150 | 300 | 0
200 | 0200 | | | 400 | 0
250 | 0250 | | | 500 | 0
300 | 0300 | 300 | 0300 | 600 | 0
350 | 0350 | | | 700 | 0
400 | 0400 | | | 800 | 0
450 | 0450 | 450 | 0450 | 900 | 0
500 | 0500 | | | 1000 | 0
550 | 0550 | | | 1100 | 0
(12 rows)
2020-04-08 04:12:14 +02:00
--
-- 3-way full join
--
EXPLAIN (COSTS OFF)
SELECT COUNT(*) FROM prt1 FULL JOIN prt2 p2(b,a,c) USING(a,b) FULL JOIN prt2 p3(b,a,c) USING (a, b)
WHERE a BETWEEN 490 AND 510;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------------------------
Aggregate
-> Append
-> Hash Full Join
Hash Cond: ((COALESCE(prt1_1.a, p2_1.a) = p3_1.a) AND (COALESCE(prt1_1.b, p2_1.b) = p3_1.b))
Filter: ((COALESCE(COALESCE(prt1_1.a, p2_1.a), p3_1.a) >= 490) AND (COALESCE(COALESCE(prt1_1.a, p2_1.a), p3_1.a) <= 510))
-> Hash Full Join
Hash Cond: ((prt1_1.a = p2_1.a) AND (prt1_1.b = p2_1.b))
-> Seq Scan on prt1_p1 prt1_1
-> Hash
-> Seq Scan on prt2_p1 p2_1
-> Hash
-> Seq Scan on prt2_p1 p3_1
-> Hash Full Join
Hash Cond: ((COALESCE(prt1_2.a, p2_2.a) = p3_2.a) AND (COALESCE(prt1_2.b, p2_2.b) = p3_2.b))
Filter: ((COALESCE(COALESCE(prt1_2.a, p2_2.a), p3_2.a) >= 490) AND (COALESCE(COALESCE(prt1_2.a, p2_2.a), p3_2.a) <= 510))
-> Hash Full Join
Hash Cond: ((prt1_2.a = p2_2.a) AND (prt1_2.b = p2_2.b))
-> Seq Scan on prt1_p2 prt1_2
-> Hash
-> Seq Scan on prt2_p2 p2_2
-> Hash
-> Seq Scan on prt2_p2 p3_2
-> Hash Full Join
Hash Cond: ((COALESCE(prt1_3.a, p2_3.a) = p3_3.a) AND (COALESCE(prt1_3.b, p2_3.b) = p3_3.b))
Filter: ((COALESCE(COALESCE(prt1_3.a, p2_3.a), p3_3.a) >= 490) AND (COALESCE(COALESCE(prt1_3.a, p2_3.a), p3_3.a) <= 510))
-> Hash Full Join
Hash Cond: ((prt1_3.a = p2_3.a) AND (prt1_3.b = p2_3.b))
-> Seq Scan on prt1_p3 prt1_3
-> Hash
-> Seq Scan on prt2_p3 p2_3
-> Hash
-> Seq Scan on prt2_p3 p3_3
(32 rows)
SELECT COUNT(*) FROM prt1 FULL JOIN prt2 p2(b,a,c) USING(a,b) FULL JOIN prt2 p3(b,a,c) USING (a, b)
WHERE a BETWEEN 490 AND 510;
count
-------
14
(1 row)
--
-- 4-way full join
--
EXPLAIN (COSTS OFF)
SELECT COUNT(*) FROM prt1 FULL JOIN prt2 p2(b,a,c) USING(a,b) FULL JOIN prt2 p3(b,a,c) USING (a, b) FULL JOIN prt1 p4 (a,b,c) USING (a, b)
WHERE a BETWEEN 490 AND 510;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Aggregate
-> Append
-> Hash Full Join
Hash Cond: ((COALESCE(COALESCE(prt1_1.a, p2_1.a), p3_1.a) = p4_1.a) AND (COALESCE(COALESCE(prt1_1.b, p2_1.b), p3_1.b) = p4_1.b))
Filter: ((COALESCE(COALESCE(COALESCE(prt1_1.a, p2_1.a), p3_1.a), p4_1.a) >= 490) AND (COALESCE(COALESCE(COALESCE(prt1_1.a, p2_1.a), p3_1.a), p4_1.a) <= 510))
-> Hash Full Join
Hash Cond: ((COALESCE(prt1_1.a, p2_1.a) = p3_1.a) AND (COALESCE(prt1_1.b, p2_1.b) = p3_1.b))
-> Hash Full Join
Hash Cond: ((prt1_1.a = p2_1.a) AND (prt1_1.b = p2_1.b))
-> Seq Scan on prt1_p1 prt1_1
-> Hash
-> Seq Scan on prt2_p1 p2_1
-> Hash
-> Seq Scan on prt2_p1 p3_1
-> Hash
-> Seq Scan on prt1_p1 p4_1
-> Hash Full Join
Hash Cond: ((COALESCE(COALESCE(prt1_2.a, p2_2.a), p3_2.a) = p4_2.a) AND (COALESCE(COALESCE(prt1_2.b, p2_2.b), p3_2.b) = p4_2.b))
Filter: ((COALESCE(COALESCE(COALESCE(prt1_2.a, p2_2.a), p3_2.a), p4_2.a) >= 490) AND (COALESCE(COALESCE(COALESCE(prt1_2.a, p2_2.a), p3_2.a), p4_2.a) <= 510))
-> Hash Full Join
Hash Cond: ((COALESCE(prt1_2.a, p2_2.a) = p3_2.a) AND (COALESCE(prt1_2.b, p2_2.b) = p3_2.b))
-> Hash Full Join
Hash Cond: ((prt1_2.a = p2_2.a) AND (prt1_2.b = p2_2.b))
-> Seq Scan on prt1_p2 prt1_2
-> Hash
-> Seq Scan on prt2_p2 p2_2
-> Hash
-> Seq Scan on prt2_p2 p3_2
-> Hash
-> Seq Scan on prt1_p2 p4_2
-> Hash Full Join
Hash Cond: ((COALESCE(COALESCE(prt1_3.a, p2_3.a), p3_3.a) = p4_3.a) AND (COALESCE(COALESCE(prt1_3.b, p2_3.b), p3_3.b) = p4_3.b))
Filter: ((COALESCE(COALESCE(COALESCE(prt1_3.a, p2_3.a), p3_3.a), p4_3.a) >= 490) AND (COALESCE(COALESCE(COALESCE(prt1_3.a, p2_3.a), p3_3.a), p4_3.a) <= 510))
-> Hash Full Join
Hash Cond: ((COALESCE(prt1_3.a, p2_3.a) = p3_3.a) AND (COALESCE(prt1_3.b, p2_3.b) = p3_3.b))
-> Hash Full Join
Hash Cond: ((prt1_3.a = p2_3.a) AND (prt1_3.b = p2_3.b))
-> Seq Scan on prt1_p3 prt1_3
-> Hash
-> Seq Scan on prt2_p3 p2_3
-> Hash
-> Seq Scan on prt2_p3 p3_3
-> Hash
-> Seq Scan on prt1_p3 p4_3
(44 rows)
SELECT COUNT(*) FROM prt1 FULL JOIN prt2 p2(b,a,c) USING(a,b) FULL JOIN prt2 p3(b,a,c) USING (a, b) FULL JOIN prt1 p4 (a,b,c) USING (a, b)
WHERE a BETWEEN 490 AND 510;
count
-------
14
(1 row)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-- Cases with non-nullable expressions in subquery results;
-- make sure these go to null as expected
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.phv, t2.b, t2.phv, t3.a + t3.b, t3.phv FROM ((SELECT 50 phv, * FROM prt1 WHERE prt1.b = 0) t1 FULL JOIN (SELECT 75 phv, * FROM prt2 WHERE prt2.a = 0) t2 ON (t1.a = t2.b)) FULL JOIN (SELECT 50 phv, * FROM prt1_e WHERE prt1_e.c = 0) t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t1.a = t1.phv OR t2.b = t2.phv OR (t3.a + t3.b)/2 = t3.phv ORDER BY t1.a, t2.b, t3.a + t3.b;
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
QUERY PLAN
------------------------------------------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a, prt2.b, ((prt1_e.a + prt1_e.b))
2018-03-29 21:47:57 +02:00
-> Append
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (prt1_1.a = ((prt1_e_1.a + prt1_e_1.b) / 2))
Filter: ((prt1_1.a = (50)) OR (prt2_1.b = (75)) OR (((prt1_e_1.a + prt1_e_1.b) / 2) = (50)))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (prt1_1.a = prt2_1.b)
-> Seq Scan on prt1_p1 prt1_1
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 prt2_1
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p1 prt1_e_1
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (prt1_2.a = ((prt1_e_2.a + prt1_e_2.b) / 2))
Filter: ((prt1_2.a = (50)) OR (prt2_2.b = (75)) OR (((prt1_e_2.a + prt1_e_2.b) / 2) = (50)))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (prt1_2.a = prt2_2.b)
-> Seq Scan on prt1_p2 prt1_2
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 prt2_2
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p2 prt1_e_2
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (prt1_3.a = ((prt1_e_3.a + prt1_e_3.b) / 2))
Filter: ((prt1_3.a = (50)) OR (prt2_3.b = (75)) OR (((prt1_e_3.a + prt1_e_3.b) / 2) = (50)))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (prt1_3.a = prt2_3.b)
-> Seq Scan on prt1_p3 prt1_3
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 prt2_3
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p3 prt1_e_3
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
(42 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.phv, t2.b, t2.phv, t3.a + t3.b, t3.phv FROM ((SELECT 50 phv, * FROM prt1 WHERE prt1.b = 0) t1 FULL JOIN (SELECT 75 phv, * FROM prt2 WHERE prt2.a = 0) t2 ON (t1.a = t2.b)) FULL JOIN (SELECT 50 phv, * FROM prt1_e WHERE prt1_e.c = 0) t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t1.a = t1.phv OR t2.b = t2.phv OR (t3.a + t3.b)/2 = t3.phv ORDER BY t1.a, t2.b, t3.a + t3.b;
a | phv | b | phv | ?column? | phv
----+-----+----+-----+----------+-----
50 | 50 | | | 100 | 50
| | 75 | 75 | |
(2 rows)
-- Semi-join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t1.b FROM prt2 t1, prt1_e t2 WHERE t1.a = 0 AND t1.b = (t2.a + t2.b)/2) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
---------------------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Nested Loop
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Join Filter: (t1_2.a = t1_5.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> HashAggregate
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Group Key: t1_5.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t2_1.a + t2_1.b) / 2) = t1_5.b)
-> Seq Scan on prt1_e_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t1_5
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_p1_a on prt1_p1 t1_2
Index Cond: (a = ((t2_1.a + t2_1.b) / 2))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Nested Loop
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Join Filter: (t1_3.a = t1_6.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> HashAggregate
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Group Key: t1_6.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (((t2_2.a + t2_2.b) / 2) = t1_6.b)
-> Seq Scan on prt1_e_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t1_6
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_p2_a on prt1_p2 t1_3
Index Cond: (a = ((t2_2.a + t2_2.b) / 2))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Nested Loop
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Join Filter: (t1_4.a = t1_7.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> HashAggregate
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Group Key: t1_7.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Nested Loop
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 t1_7
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_e_p3_ab2 on prt1_e_p3 t2_3
Index Cond: (((a + b) / 2) = t1_7.b)
-> Index Scan using iprt1_p3_a on prt1_p3 t1_4
Index Cond: (a = ((t2_3.a + t2_3.b) / 2))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
(41 rows)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t1.b FROM prt2 t1, prt1_e t2 WHERE t1.a = 0 AND t1.b = (t2.a + t2.b)/2) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
0 | 0 | 0000
150 | 0 | 0150
300 | 0 | 0300
450 | 0 | 0450
(4 rows)
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t1.b FROM prt2 t1 WHERE t1.b IN (SELECT (t1.a + t1.b)/2 FROM prt1_e t1 WHERE t1.c = 0)) AND t1.b = 0 ORDER BY t1.a;
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
QUERY PLAN
---------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a
-> Append
-> Nested Loop
-> HashAggregate
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Group Key: t1_6.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_6.b = ((t1_9.a + t1_9.b) / 2))
-> Seq Scan on prt2_p1 t1_6
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p1 t1_9
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_p1_a on prt1_p1 t1_3
Index Cond: (a = t1_6.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Nested Loop
-> HashAggregate
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Group Key: t1_7.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_7.b = ((t1_10.a + t1_10.b) / 2))
-> Seq Scan on prt2_p2 t1_7
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p2 t1_10
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_p2_a on prt1_p2 t1_4
Index Cond: (a = t1_7.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Nested Loop
2018-11-23 18:48:49 +01:00
-> HashAggregate
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Group Key: t1_8.b
2018-11-23 18:48:49 +01:00
-> Hash Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t1_8.b = ((t1_11.a + t1_11.b) / 2))
-> Seq Scan on prt2_p3 t1_8
2018-11-23 18:48:49 +01:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p3 t1_11
2018-11-23 18:48:49 +01:00
Filter: (c = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt1_p3_a on prt1_p3 t1_5
Index Cond: (a = t1_8.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
2018-11-23 18:48:49 +01:00
(39 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t1.b FROM prt2 t1 WHERE t1.b IN (SELECT (t1.a + t1.b)/2 FROM prt1_e t1 WHERE t1.c = 0)) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
0 | 0 | 0000
150 | 0 | 0150
300 | 0 | 0300
450 | 0 | 0450
(4 rows)
-- test merge joins
SET enable_hashjoin TO off;
SET enable_nestloop TO off;
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t1.b FROM prt2 t1 WHERE t1.b IN (SELECT (t1.a + t1.b)/2 FROM prt1_e t1 WHERE t1.c = 0)) AND t1.b = 0 ORDER BY t1.a;
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
QUERY PLAN
------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Merge Append
Sort Key: t1.a
-> Merge Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Merge Cond: (t1_3.a = t1_6.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: t1_3.a
-> Seq Scan on prt1_p1 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Merge Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Merge Cond: (t1_6.b = (((t1_9.a + t1_9.b) / 2)))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: t1_6.b
-> Seq Scan on prt2_p1 t1_6
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: (((t1_9.a + t1_9.b) / 2))
-> Seq Scan on prt1_e_p1 t1_9
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Merge Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Merge Cond: (t1_4.a = t1_7.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: t1_4.a
-> Seq Scan on prt1_p2 t1_4
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Merge Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Merge Cond: (t1_7.b = (((t1_10.a + t1_10.b) / 2)))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: t1_7.b
-> Seq Scan on prt2_p2 t1_7
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: (((t1_10.a + t1_10.b) / 2))
-> Seq Scan on prt1_e_p2 t1_10
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Merge Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Merge Cond: (t1_5.a = t1_8.b)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: t1_5.a
-> Seq Scan on prt1_p3 t1_5
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Merge Semi Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Merge Cond: (t1_8.b = (((t1_11.a + t1_11.b) / 2)))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: t1_8.b
-> Seq Scan on prt2_p3 t1_8
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Sort
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Sort Key: (((t1_11.a + t1_11.b) / 2))
-> Seq Scan on prt1_e_p3 t1_11
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
(47 rows)
SELECT t1.* FROM prt1 t1 WHERE t1.a IN (SELECT t1.b FROM prt2 t1 WHERE t1.b IN (SELECT (t1.a + t1.b)/2 FROM prt1_e t1 WHERE t1.c = 0)) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
0 | 0 | 0000
150 | 0 | 0150
300 | 0 | 0300
450 | 0 | 0450
(4 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM (prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b) RIGHT JOIN prt1_e t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t3.c = 0 ORDER BY t1.a, t2.b, t3.a + t3.b;
2018-03-29 21:47:57 +02:00
QUERY PLAN
----------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a, t2.b, ((t3.a + t3.b))
2018-03-29 21:47:57 +02:00
-> Append
-> Merge Left Join
Merge Cond: (t1_1.a = t2_1.b)
-> Sort
Sort Key: t1_1.a
-> Merge Left Join
Merge Cond: ((((t3_1.a + t3_1.b) / 2)) = t1_1.a)
-> Sort
Sort Key: (((t3_1.a + t3_1.b) / 2))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p1 t3_1
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
-> Sort
Sort Key: t1_1.a
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
2018-03-29 21:47:57 +02:00
-> Sort
Sort Key: t2_1.b
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
2018-03-29 21:47:57 +02:00
-> Merge Left Join
Merge Cond: (t1_2.a = t2_2.b)
-> Sort
Sort Key: t1_2.a
-> Merge Left Join
Merge Cond: ((((t3_2.a + t3_2.b) / 2)) = t1_2.a)
-> Sort
Sort Key: (((t3_2.a + t3_2.b) / 2))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_e_p2 t3_2
2018-03-29 21:47:57 +02:00
Filter: (c = 0)
-> Sort
Sort Key: t1_2.a
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
2018-03-29 21:47:57 +02:00
-> Sort
Sort Key: t2_2.b
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
-> Merge Left Join
Merge Cond: (t1_3.a = t2_3.b)
-> Sort
Sort Key: t1_3.a
-> Merge Left Join
Merge Cond: ((((t3_3.a + t3_3.b) / 2)) = t1_3.a)
-> Sort
Sort Key: (((t3_3.a + t3_3.b) / 2))
-> Seq Scan on prt1_e_p3 t3_3
Filter: (c = 0)
-> Sort
Sort Key: t1_3.a
-> Seq Scan on prt1_p3 t1_3
-> Sort
Sort Key: t2_3.b
-> Seq Scan on prt2_p3 t2_3
2018-03-29 21:47:57 +02:00
(51 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c, t3.a + t3.b, t3.c FROM (prt1 t1 LEFT JOIN prt2 t2 ON t1.a = t2.b) RIGHT JOIN prt1_e t3 ON (t1.a = (t3.a + t3.b)/2) WHERE t3.c = 0 ORDER BY t1.a, t2.b, t3.a + t3.b;
a | c | b | c | ?column? | c
-----+------+-----+------+----------+---
0 | 0000 | 0 | 0000 | 0 | 0
50 | 0050 | | | 100 | 0
100 | 0100 | | | 200 | 0
150 | 0150 | 150 | 0150 | 300 | 0
200 | 0200 | | | 400 | 0
250 | 0250 | | | 500 | 0
300 | 0300 | 300 | 0300 | 600 | 0
350 | 0350 | | | 700 | 0
400 | 0400 | | | 800 | 0
450 | 0450 | 450 | 0450 | 900 | 0
500 | 0500 | | | 1000 | 0
550 | 0550 | | | 1100 | 0
(12 rows)
-- MergeAppend on nullable column
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-- This should generate a partitionwise join, but currently fails to
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t2.b FROM (SELECT * FROM prt1 WHERE a < 450) t1 LEFT JOIN (SELECT * FROM prt2 WHERE b > 250) t2 ON t1.a = t2.b WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
-----------------------------------------------------------
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a, prt2.b
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Merge Left Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Merge Cond: (prt1.a = prt2.b)
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1.a
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 prt1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: ((a < 450) AND (b = 0))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 prt1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: ((a < 450) AND (b = 0))
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt2.b
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 prt2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b > 250)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 prt2_2
Avoid crash in partitionwise join planning under GEQO.
While trying to plan a partitionwise join, we may be faced with cases
where one or both input partitions for a particular segment of the join
have been pruned away. In HEAD and v11, this is problematic because
earlier processing didn't bother to make a pruned RelOptInfo fully
valid. With an upcoming patch to make partition pruning more efficient,
this'll be even more problematic because said RelOptInfo won't exist at
all.
The existing code attempts to deal with this by retroactively making the
RelOptInfo fully valid, but that causes crashes under GEQO because join
planning is done in a short-lived memory context. In v11 we could
probably have fixed this by switching to the planner's main context
while fixing up the RelOptInfo, but that idea doesn't scale well to the
upcoming patch. It would be better not to mess with the base-relation
data structures during join planning, anyway --- that's just a recipe
for order-of-operations bugs.
In many cases, though, we don't actually need the child RelOptInfo,
because if the input is certainly empty then the join segment's result
is certainly empty, so we can skip making a join plan altogether. (The
existing code ultimately arrives at the same conclusion, but only after
doing a lot more work.) This approach works except when the pruned-away
partition is on the nullable side of a LEFT, ANTI, or FULL join, and the
other side isn't pruned. But in those cases the existing code leaves a
lot to be desired anyway --- the correct output is just the result of
the unpruned side of the join, but we were emitting a useless outer join
against a dummy Result. Pending somebody writing code to handle that
more nicely, let's just abandon the partitionwise-join optimization in
such cases.
When the modified code skips making a join plan, it doesn't make a
join RelOptInfo either; this requires some upper-level code to
cope with nulls in part_rels[] arrays. We would have had to have
that anyway after the upcoming patch.
Back-patch to v11 since the crash is demonstrable there.
Discussion: https://postgr.es/m/8305.1553884377@sss.pgh.pa.us
2019-03-30 17:48:19 +01:00
Filter: (b > 250)
(18 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t2.b FROM (SELECT * FROM prt1 WHERE a < 450) t1 LEFT JOIN (SELECT * FROM prt2 WHERE b > 250) t2 ON t1.a = t2.b WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | b
-----+-----
0 |
50 |
100 |
150 |
200 |
250 |
300 | 300
350 |
400 |
(9 rows)
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
-- merge join when expression with whole-row reference needs to be sorted;
-- partitionwise join does not apply
EXPLAIN (COSTS OFF)
SELECT t1.a, t2.b FROM prt1 t1, prt2 t2 WHERE t1::text = t2::text AND t1.a = t2.b ORDER BY t1.a;
QUERY PLAN
-----------------------------------------------------------------------------------------
Merge Join
Merge Cond: ((t1.a = t2.b) AND (((((t1.*)::prt1))::text) = ((((t2.*)::prt2))::text)))
-> Sort
Sort Key: t1.a, ((((t1.*)::prt1))::text)
-> Result
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
-> Seq Scan on prt1_p2 t1_2
-> Seq Scan on prt1_p3 t1_3
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
-> Sort
Sort Key: t2.b, ((((t2.*)::prt2))::text)
-> Result
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
-> Seq Scan on prt2_p2 t2_2
-> Seq Scan on prt2_p3 t2_3
Disable support for partitionwise joins in problematic cases.
Commit f49842d, which added support for partitionwise joins, built the
child's tlist by applying adjust_appendrel_attrs() to the parent's. So in
the case where the parent's included a whole-row Var for the parent, the
child's contained a ConvertRowtypeExpr. To cope with that, that commit
added code to the planner, such as setrefs.c, but some code paths still
assumed that the tlist for a scan (or join) rel would only include Vars
and PlaceHolderVars, which was true before that commit, causing errors:
* When creating an explicit sort node for an input path for a mergejoin
path for a child join, prepare_sort_from_pathkeys() threw the 'could not
find pathkey item to sort' error.
* When deparsing a relation participating in a pushed down child join as a
subquery in contrib/postgres_fdw, get_relation_column_alias_ids() threw
the 'unexpected expression in subquery output' error.
* When performing set_plan_references() on a local join plan generated by
contrib/postgres_fdw for EvalPlanQual support for a pushed down child
join, fix_join_expr() threw the 'variable not found in subplan target
lists' error.
To fix these, two approaches have been proposed: one by Ashutosh Bapat and
one by me. While the former keeps building the child's tlist with a
ConvertRowtypeExpr, the latter builds it with a whole-row Var for the
child not to violate the planner assumption, and tries to fix it up later,
But both approaches need more work, so refuse to generate partitionwise
join paths when whole-row Vars are involved, instead. We don't need to
handle ConvertRowtypeExprs in the child's tlists for now, so this commit
also removes the changes to the planner.
Previously, partitionwise join computed attr_needed data for each child
separately, and built the child join's tlist using that data, which also
required an extra step for adding PlaceHolderVars to that tlist, but it
would be more efficient to build it from the parent join's tlist through
the adjust_appendrel_attrs() transformation. So this commit builds that
list that way, and simplifies build_joinrel_tlist() and placeholder.c as
well as part of set_append_rel_size() to basically what they were before
partitionwise join went in.
Back-patch to PG11 where partitionwise join was introduced.
Report by Rajkumar Raghuwanshi. Analysis by Ashutosh Bapat, who also
provided some of regression tests. Patch by me, reviewed by Robert Haas.
Discussion: https://postgr.es/m/CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com
2018-08-31 13:34:06 +02:00
(16 rows)
SELECT t1.a, t2.b FROM prt1 t1, prt2 t2 WHERE t1::text = t2::text AND t1.a = t2.b ORDER BY t1.a;
a | b
----+----
0 | 0
6 | 6
12 | 12
18 | 18
24 | 24
(5 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
RESET enable_hashjoin;
RESET enable_nestloop;
--
-- partitioned by multiple columns
--
CREATE TABLE prt1_m (a int, b int, c int) PARTITION BY RANGE(a, ((a + b)/2));
CREATE TABLE prt1_m_p1 PARTITION OF prt1_m FOR VALUES FROM (0, 0) TO (250, 250);
CREATE TABLE prt1_m_p2 PARTITION OF prt1_m FOR VALUES FROM (250, 250) TO (500, 500);
CREATE TABLE prt1_m_p3 PARTITION OF prt1_m FOR VALUES FROM (500, 500) TO (600, 600);
INSERT INTO prt1_m SELECT i, i, i % 25 FROM generate_series(0, 599, 2) i;
ANALYZE prt1_m;
CREATE TABLE prt2_m (a int, b int, c int) PARTITION BY RANGE(((b + a)/2), b);
CREATE TABLE prt2_m_p1 PARTITION OF prt2_m FOR VALUES FROM (0, 0) TO (250, 250);
CREATE TABLE prt2_m_p2 PARTITION OF prt2_m FOR VALUES FROM (250, 250) TO (500, 500);
CREATE TABLE prt2_m_p3 PARTITION OF prt2_m FOR VALUES FROM (500, 500) TO (600, 600);
INSERT INTO prt2_m SELECT i, i, i % 25 FROM generate_series(0, 599, 3) i;
ANALYZE prt2_m;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1_m WHERE prt1_m.c = 0) t1 FULL JOIN (SELECT * FROM prt2_m WHERE prt2_m.c = 0) t2 ON (t1.a = (t2.b + t2.a)/2 AND t2.b = (t1.a + t1.b)/2) ORDER BY t1.a, t2.b;
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1_m.a, prt2_m.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_m_1.a = ((prt2_m_1.b + prt2_m_1.a) / 2)) AND (((prt1_m_1.a + prt1_m_1.b) / 2) = prt2_m_1.b))
-> Seq Scan on prt1_m_p1 prt1_m_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_m_p1 prt2_m_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_m_2.a = ((prt2_m_2.b + prt2_m_2.a) / 2)) AND (((prt1_m_2.a + prt1_m_2.b) / 2) = prt2_m_2.b))
-> Seq Scan on prt1_m_p2 prt1_m_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_m_p2 prt2_m_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_m_3.a = ((prt2_m_3.b + prt2_m_3.a) / 2)) AND (((prt1_m_3.a + prt1_m_3.b) / 2) = prt2_m_3.b))
-> Seq Scan on prt1_m_p3 prt1_m_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_m_p3 prt2_m_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (c = 0)
(24 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1_m WHERE prt1_m.c = 0) t1 FULL JOIN (SELECT * FROM prt2_m WHERE prt2_m.c = 0) t2 ON (t1.a = (t2.b + t2.a)/2 AND t2.b = (t1.a + t1.b)/2) ORDER BY t1.a, t2.b;
a | c | b | c
-----+---+-----+---
0 | 0 | 0 | 0
50 | 0 | |
100 | 0 | |
150 | 0 | 150 | 0
200 | 0 | |
250 | 0 | |
300 | 0 | 300 | 0
350 | 0 | |
400 | 0 | |
450 | 0 | 450 | 0
500 | 0 | |
550 | 0 | |
| | 75 | 0
| | 225 | 0
| | 375 | 0
| | 525 | 0
(16 rows)
--
-- tests for list partitioned tables.
--
CREATE TABLE plt1 (a int, b int, c text) PARTITION BY LIST(c);
CREATE TABLE plt1_p1 PARTITION OF plt1 FOR VALUES IN ('0000', '0003', '0004', '0010');
CREATE TABLE plt1_p2 PARTITION OF plt1 FOR VALUES IN ('0001', '0005', '0002', '0009');
CREATE TABLE plt1_p3 PARTITION OF plt1 FOR VALUES IN ('0006', '0007', '0008', '0011');
INSERT INTO plt1 SELECT i, i, to_char(i/50, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE plt1;
CREATE TABLE plt2 (a int, b int, c text) PARTITION BY LIST(c);
CREATE TABLE plt2_p1 PARTITION OF plt2 FOR VALUES IN ('0000', '0003', '0004', '0010');
CREATE TABLE plt2_p2 PARTITION OF plt2 FOR VALUES IN ('0001', '0005', '0002', '0009');
CREATE TABLE plt2_p3 PARTITION OF plt2 FOR VALUES IN ('0006', '0007', '0008', '0011');
INSERT INTO plt2 SELECT i, i, to_char(i/50, 'FM0000') FROM generate_series(0, 599, 3) i;
ANALYZE plt2;
--
-- list partitioned by expression
--
CREATE TABLE plt1_e (a int, b int, c text) PARTITION BY LIST(ltrim(c, 'A'));
CREATE TABLE plt1_e_p1 PARTITION OF plt1_e FOR VALUES IN ('0000', '0003', '0004', '0010');
CREATE TABLE plt1_e_p2 PARTITION OF plt1_e FOR VALUES IN ('0001', '0005', '0002', '0009');
CREATE TABLE plt1_e_p3 PARTITION OF plt1_e FOR VALUES IN ('0006', '0007', '0008', '0011');
INSERT INTO plt1_e SELECT i, i, 'A' || to_char(i/50, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE plt1_e;
-- test partition matching with N-way join
EXPLAIN (COSTS OFF)
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
SELECT avg(t1.a), avg(t2.b), avg(t3.a + t3.b), t1.c, t2.c, t3.c FROM plt1 t1, plt2 t2, plt1_e t3 WHERE t1.b = t2.b AND t1.c = t2.c AND ltrim(t3.c, 'A') = t1.c GROUP BY t1.c, t2.c, t3.c ORDER BY t1.c, t2.c, t3.c;
2018-03-29 21:47:57 +02:00
QUERY PLAN
--------------------------------------------------------------------------------
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
GroupAggregate
Group Key: t1.c, t2.c, t3.c
-> Sort
Sort Key: t1.c, t3.c
2018-03-29 21:47:57 +02:00
-> Append
-> Hash Join
Hash Cond: (t1_1.c = ltrim(t3_1.c, 'A'::text))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Join
2018-03-29 21:47:57 +02:00
Hash Cond: ((t1_1.b = t2_1.b) AND (t1_1.c = t2_1.c))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt2_p1 t2_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_e_p1 t3_1
2018-03-29 21:47:57 +02:00
-> Hash Join
Hash Cond: (t1_2.c = ltrim(t3_2.c, 'A'::text))
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
-> Hash Join
2018-03-29 21:47:57 +02:00
Hash Cond: ((t1_2.b = t2_2.b) AND (t1_2.c = t2_2.c))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_p2 t1_2
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt2_p2 t2_2
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_e_p2 t3_2
-> Hash Join
Hash Cond: (t1_3.c = ltrim(t3_3.c, 'A'::text))
-> Hash Join
Hash Cond: ((t1_3.b = t2_3.b) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_p3 t1_3
-> Hash
-> Seq Scan on plt2_p3 t2_3
-> Hash
-> Seq Scan on plt1_e_p3 t3_3
2018-03-29 21:47:57 +02:00
(32 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
SELECT avg(t1.a), avg(t2.b), avg(t3.a + t3.b), t1.c, t2.c, t3.c FROM plt1 t1, plt2 t2, plt1_e t3 WHERE t1.b = t2.b AND t1.c = t2.c AND ltrim(t3.c, 'A') = t1.c GROUP BY t1.c, t2.c, t3.c ORDER BY t1.c, t2.c, t3.c;
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
avg | avg | avg | c | c | c
----------------------+----------------------+-----------------------+------+------+-------
24.0000000000000000 | 24.0000000000000000 | 48.0000000000000000 | 0000 | 0000 | A0000
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
75.0000000000000000 | 75.0000000000000000 | 148.0000000000000000 | 0001 | 0001 | A0001
123.0000000000000000 | 123.0000000000000000 | 248.0000000000000000 | 0002 | 0002 | A0002
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
174.0000000000000000 | 174.0000000000000000 | 348.0000000000000000 | 0003 | 0003 | A0003
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
225.0000000000000000 | 225.0000000000000000 | 448.0000000000000000 | 0004 | 0004 | A0004
273.0000000000000000 | 273.0000000000000000 | 548.0000000000000000 | 0005 | 0005 | A0005
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
324.0000000000000000 | 324.0000000000000000 | 648.0000000000000000 | 0006 | 0006 | A0006
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
375.0000000000000000 | 375.0000000000000000 | 748.0000000000000000 | 0007 | 0007 | A0007
423.0000000000000000 | 423.0000000000000000 | 848.0000000000000000 | 0008 | 0008 | A0008
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
474.0000000000000000 | 474.0000000000000000 | 948.0000000000000000 | 0009 | 0009 | A0009
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
525.0000000000000000 | 525.0000000000000000 | 1048.0000000000000000 | 0010 | 0010 | A0010
573.0000000000000000 | 573.0000000000000000 | 1148.0000000000000000 | 0011 | 0011 | A0011
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(12 rows)
-- joins where one of the relations is proven empty
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt2 t2 WHERE t1.a = t2.b AND t1.a = 1 AND t1.a = 2;
QUERY PLAN
--------------------------
Result
One-Time Filter: false
(2 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a = 1 AND a = 2) t1 LEFT JOIN prt2 t2 ON t1.a = t2.b;
QUERY PLAN
--------------------------
Result
One-Time Filter: false
(2 rows)
EXPLAIN (COSTS OFF)
2018-02-05 23:31:57 +01:00
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a = 1 AND a = 2) t1 RIGHT JOIN prt2 t2 ON t1.a = t2.b, prt1 t3 WHERE t2.b = t3.a;
QUERY PLAN
--------------------------------------------------
Hash Left Join
Hash Cond: (t2.b = a)
-> Append
-> Hash Join
Hash Cond: (t3_1.a = t2_1.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t3_1
2018-02-05 23:31:57 +01:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
2018-02-05 23:31:57 +01:00
-> Hash Join
Hash Cond: (t3_2.a = t2_2.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t3_2
2018-02-05 23:31:57 +01:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
-> Hash Join
Hash Cond: (t3_3.a = t2_3.b)
-> Seq Scan on prt1_p3 t3_3
-> Hash
-> Seq Scan on prt2_p3 t2_3
2018-02-05 23:31:57 +01:00
-> Hash
-> Result
One-Time Filter: false
(21 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1 WHERE a = 1 AND a = 2) t1 FULL JOIN prt2 t2 ON t1.a = t2.b WHERE t2.a = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
--------------------------------------------
Sort
Sort Key: a, t2.b
-> Hash Left Join
Hash Cond: (t2.b = a)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Hash
-> Result
One-Time Filter: false
(14 rows)
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
--
-- tests for hash partitioned tables.
--
CREATE TABLE pht1 (a int, b int, c text) PARTITION BY HASH(c);
CREATE TABLE pht1_p1 PARTITION OF pht1 FOR VALUES WITH (MODULUS 3, REMAINDER 0);
CREATE TABLE pht1_p2 PARTITION OF pht1 FOR VALUES WITH (MODULUS 3, REMAINDER 1);
CREATE TABLE pht1_p3 PARTITION OF pht1 FOR VALUES WITH (MODULUS 3, REMAINDER 2);
INSERT INTO pht1 SELECT i, i, to_char(i/50, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE pht1;
CREATE TABLE pht2 (a int, b int, c text) PARTITION BY HASH(c);
CREATE TABLE pht2_p1 PARTITION OF pht2 FOR VALUES WITH (MODULUS 3, REMAINDER 0);
CREATE TABLE pht2_p2 PARTITION OF pht2 FOR VALUES WITH (MODULUS 3, REMAINDER 1);
CREATE TABLE pht2_p3 PARTITION OF pht2 FOR VALUES WITH (MODULUS 3, REMAINDER 2);
INSERT INTO pht2 SELECT i, i, to_char(i/50, 'FM0000') FROM generate_series(0, 599, 3) i;
ANALYZE pht2;
--
-- hash partitioned by expression
--
CREATE TABLE pht1_e (a int, b int, c text) PARTITION BY HASH(ltrim(c, 'A'));
CREATE TABLE pht1_e_p1 PARTITION OF pht1_e FOR VALUES WITH (MODULUS 3, REMAINDER 0);
CREATE TABLE pht1_e_p2 PARTITION OF pht1_e FOR VALUES WITH (MODULUS 3, REMAINDER 1);
CREATE TABLE pht1_e_p3 PARTITION OF pht1_e FOR VALUES WITH (MODULUS 3, REMAINDER 2);
2018-02-22 14:51:00 +01:00
INSERT INTO pht1_e SELECT i, i, 'A' || to_char(i/50, 'FM0000') FROM generate_series(0, 299, 2) i;
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
ANALYZE pht1_e;
-- test partition matching with N-way join
EXPLAIN (COSTS OFF)
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
SELECT avg(t1.a), avg(t2.b), avg(t3.a + t3.b), t1.c, t2.c, t3.c FROM pht1 t1, pht2 t2, pht1_e t3 WHERE t1.b = t2.b AND t1.c = t2.c AND ltrim(t3.c, 'A') = t1.c GROUP BY t1.c, t2.c, t3.c ORDER BY t1.c, t2.c, t3.c;
2018-03-29 21:47:57 +02:00
QUERY PLAN
--------------------------------------------------------------------------------
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
GroupAggregate
Group Key: t1.c, t2.c, t3.c
-> Sort
Sort Key: t1.c, t3.c
2018-03-29 21:47:57 +02:00
-> Append
-> Hash Join
Hash Cond: (t1_1.c = ltrim(t3_1.c, 'A'::text))
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
-> Hash Join
2018-03-29 21:47:57 +02:00
Hash Cond: ((t1_1.b = t2_1.b) AND (t1_1.c = t2_1.c))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on pht1_p1 t1_1
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on pht2_p1 t2_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on pht1_e_p1 t3_1
2018-03-29 21:47:57 +02:00
-> Hash Join
Hash Cond: (t1_2.c = ltrim(t3_2.c, 'A'::text))
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
-> Hash Join
2018-03-29 21:47:57 +02:00
Hash Cond: ((t1_2.b = t2_2.b) AND (t1_2.c = t2_2.c))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on pht1_p2 t1_2
-> Hash
-> Seq Scan on pht2_p2 t2_2
-> Hash
-> Seq Scan on pht1_e_p2 t3_2
-> Hash Join
Hash Cond: (t1_3.c = ltrim(t3_3.c, 'A'::text))
-> Hash Join
Hash Cond: ((t1_3.b = t2_3.b) AND (t1_3.c = t2_3.c))
-> Seq Scan on pht1_p3 t1_3
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on pht2_p3 t2_3
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on pht1_e_p3 t3_3
2018-03-29 21:47:57 +02:00
(32 rows)
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
Charge cpu_tuple_cost * 0.5 for Append and MergeAppend nodes.
Previously, Append didn't charge anything at all, and MergeAppend
charged only cpu_operator_cost, about half the value used here. This
change might make MergeAppend plans slightly more likely to be chosen
than before, since this commit increases the assumed cost for Append
-- with default values -- by 0.005 per tuple but MergeAppend by only
0.0025 per tuple. Since the comparisons required by MergeAppend are
costed separately, it's not clear why MergeAppend needs to be
otherwise more expensive than Append, so hopefully this is OK.
Prior to partition-wise join, it didn't really matter whether or not
an Append node had any cost of its own, because every plan had to use
the same number of Append or MergeAppend nodes and in the same places.
Only the relative cost of Append vs. MergeAppend made a difference.
Now, however, it is possible to avoid some of the Append nodes using a
partition-wise join, so it's worth making an effort. Pending patches
for partition-wise aggregate care too, because an Append of Aggregate
nodes will incur the Append overhead fewer times than an Aggregate
over an Append. Although in most cases this change will favor the use
of partition-wise techniques, it does the opposite when the join
cardinality is greater than the sum of the input cardinalities. Since
this situation arises in an existing regression test, I [rhaas]
adjusted it to keep the overall plan shape approximately the same.
Jeevan Chalke, per a suggestion from David Rowley. Reviewed by
Ashutosh Bapat. Some changes by me. The larger patch series of which
this patch is a part was also reviewed and tested by Antonin Houska,
Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, Rafia Sabih, and me.
Discussion: http://postgr.es/m/CAKJS1f9UXdk6ZYyqbJnjFO9a9hyHKGW7B=ZRh-rxy9qxfPA5Gw@mail.gmail.com
2018-02-22 05:09:27 +01:00
SELECT avg(t1.a), avg(t2.b), avg(t3.a + t3.b), t1.c, t2.c, t3.c FROM pht1 t1, pht2 t2, pht1_e t3 WHERE t1.b = t2.b AND t1.c = t2.c AND ltrim(t3.c, 'A') = t1.c GROUP BY t1.c, t2.c, t3.c ORDER BY t1.c, t2.c, t3.c;
2018-02-22 14:51:00 +01:00
avg | avg | avg | c | c | c
----------------------+----------------------+----------------------+------+------+-------
24.0000000000000000 | 24.0000000000000000 | 48.0000000000000000 | 0000 | 0000 | A0000
75.0000000000000000 | 75.0000000000000000 | 148.0000000000000000 | 0001 | 0001 | A0001
123.0000000000000000 | 123.0000000000000000 | 248.0000000000000000 | 0002 | 0002 | A0002
174.0000000000000000 | 174.0000000000000000 | 348.0000000000000000 | 0003 | 0003 | A0003
225.0000000000000000 | 225.0000000000000000 | 448.0000000000000000 | 0004 | 0004 | A0004
273.0000000000000000 | 273.0000000000000000 | 548.0000000000000000 | 0005 | 0005 | A0005
(6 rows)
Add hash partitioning.
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
2017-11-10 00:07:25 +01:00
2018-07-06 03:56:12 +02:00
-- test default partition behavior for range
ALTER TABLE prt1 DETACH PARTITION prt1_p3;
ALTER TABLE prt1 ATTACH PARTITION prt1_p3 DEFAULT;
ANALYZE prt1;
ALTER TABLE prt2 DETACH PARTITION prt2_p3;
ALTER TABLE prt2 ATTACH PARTITION prt2_p3 DEFAULT;
ANALYZE prt2;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt2 t2 WHERE t1.a = t2.b AND t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
--------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_p1 t2_1
2018-07-06 03:56:12 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
2018-07-06 03:56:12 +02:00
Filter: (b = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_p2 t2_2
2018-07-06 03:56:12 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
2018-07-06 03:56:12 +02:00
Filter: (b = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_p3 t2_3
2018-07-06 03:56:12 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p3 t1_3
2018-07-06 03:56:12 +02:00
Filter: (b = 0)
(21 rows)
-- test default partition behavior for list
ALTER TABLE plt1 DETACH PARTITION plt1_p3;
ALTER TABLE plt1 ATTACH PARTITION plt1_p3 DEFAULT;
ANALYZE plt1;
ALTER TABLE plt2 DETACH PARTITION plt2_p3;
ALTER TABLE plt2 ATTACH PARTITION plt2_p3 DEFAULT;
ANALYZE plt2;
EXPLAIN (COSTS OFF)
SELECT avg(t1.a), avg(t2.b), t1.c, t2.c FROM plt1 t1 RIGHT JOIN plt2 t2 ON t1.c = t2.c WHERE t1.a % 25 = 0 GROUP BY t1.c, t2.c ORDER BY t1.c, t2.c;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.c
-> HashAggregate
Group Key: t1.c, t2.c
-> Append
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_1.c = t1_1.c)
-> Seq Scan on plt2_p1 t2_1
2018-07-06 03:56:12 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_p1 t1_1
2018-07-06 03:56:12 +02:00
Filter: ((a % 25) = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_2.c = t1_2.c)
-> Seq Scan on plt2_p2 t2_2
2018-07-06 03:56:12 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_p2 t1_2
2018-07-06 03:56:12 +02:00
Filter: ((a % 25) = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_3.c = t1_3.c)
-> Seq Scan on plt2_p3 t2_3
2018-07-06 03:56:12 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_p3 t1_3
2018-07-06 03:56:12 +02:00
Filter: ((a % 25) = 0)
(23 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
2017-10-11 15:15:20 +02:00
-- multiple levels of partitioning
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
CREATE TABLE prt1_l (a int, b int, c varchar) PARTITION BY RANGE(a);
CREATE TABLE prt1_l_p1 PARTITION OF prt1_l FOR VALUES FROM (0) TO (250);
CREATE TABLE prt1_l_p2 PARTITION OF prt1_l FOR VALUES FROM (250) TO (500) PARTITION BY LIST (c);
CREATE TABLE prt1_l_p2_p1 PARTITION OF prt1_l_p2 FOR VALUES IN ('0000', '0001');
CREATE TABLE prt1_l_p2_p2 PARTITION OF prt1_l_p2 FOR VALUES IN ('0002', '0003');
CREATE TABLE prt1_l_p3 PARTITION OF prt1_l FOR VALUES FROM (500) TO (600) PARTITION BY RANGE (b);
CREATE TABLE prt1_l_p3_p1 PARTITION OF prt1_l_p3 FOR VALUES FROM (0) TO (13);
CREATE TABLE prt1_l_p3_p2 PARTITION OF prt1_l_p3 FOR VALUES FROM (13) TO (25);
INSERT INTO prt1_l SELECT i, i % 25, to_char(i % 4, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE prt1_l;
CREATE TABLE prt2_l (a int, b int, c varchar) PARTITION BY RANGE(b);
CREATE TABLE prt2_l_p1 PARTITION OF prt2_l FOR VALUES FROM (0) TO (250);
CREATE TABLE prt2_l_p2 PARTITION OF prt2_l FOR VALUES FROM (250) TO (500) PARTITION BY LIST (c);
CREATE TABLE prt2_l_p2_p1 PARTITION OF prt2_l_p2 FOR VALUES IN ('0000', '0001');
CREATE TABLE prt2_l_p2_p2 PARTITION OF prt2_l_p2 FOR VALUES IN ('0002', '0003');
CREATE TABLE prt2_l_p3 PARTITION OF prt2_l FOR VALUES FROM (500) TO (600) PARTITION BY RANGE (a);
CREATE TABLE prt2_l_p3_p1 PARTITION OF prt2_l_p3 FOR VALUES FROM (0) TO (13);
CREATE TABLE prt2_l_p3_p2 PARTITION OF prt2_l_p3 FOR VALUES FROM (13) TO (25);
INSERT INTO prt2_l SELECT i % 25, i, to_char(i % 4, 'FM0000') FROM generate_series(0, 599, 3) i;
ANALYZE prt2_l;
-- inner join, qual covering only top-level partitions
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_l t1, prt2_l t2 WHERE t1.a = t2.b AND t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
-------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_l_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_3.b = t1_3.a)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p1 t2_3
-> Seq Scan on prt2_l_p2_p2 t2_4
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p1 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p2 t1_4
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: (t2_6.b = t1_5.a)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p3_p1 t2_6
-> Seq Scan on prt2_l_p3_p2 t2_7
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p3_p1 t1_5
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
Filter: (b = 0)
(28 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_l t1, prt2_l t2 WHERE t1.a = t2.b AND t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | 0 | 0000
150 | 0002 | 150 | 0002
300 | 0000 | 300 | 0000
450 | 0002 | 450 | 0002
(4 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_l t1 LEFT JOIN prt2_l t2 ON t1.a = t2.b AND t1.c = t2.c WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.b
-> Append
-> Hash Right Join
Hash Cond: ((t2_1.b = t1_1.a) AND ((t2_1.c)::text = (t1_1.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p1 t2_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash Right Join
Hash Cond: ((t2_2.b = t1_2.a) AND ((t2_2.c)::text = (t1_2.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p1 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p1 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash Right Join
Hash Cond: ((t2_3.b = t1_3.a) AND ((t2_3.c)::text = (t1_3.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p2 t2_3
-> Hash
-> Seq Scan on prt1_l_p2_p2 t1_3
Filter: (b = 0)
-> Hash Right Join
Hash Cond: ((t2_5.b = t1_4.a) AND ((t2_5.c)::text = (t1_4.c)::text))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p3_p1 t2_5
-> Seq Scan on prt2_l_p3_p2 t2_6
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p3_p1 t1_4
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
Filter: (b = 0)
(29 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_l t1 LEFT JOIN prt2_l t2 ON t1.a = t2.b AND t1.c = t2.c WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | 0 | 0000
50 | 0002 | |
100 | 0000 | |
150 | 0002 | 150 | 0002
200 | 0000 | |
250 | 0002 | |
300 | 0000 | 300 | 0000
350 | 0002 | |
400 | 0000 | |
450 | 0002 | 450 | 0002
500 | 0000 | |
550 | 0002 | |
(12 rows)
-- right join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_l t1 RIGHT JOIN prt2_l t2 ON t1.a = t2.b AND t1.c = t2.c WHERE t2.a = 0 ORDER BY t1.a, t2.b;
2018-03-29 21:47:57 +02:00
QUERY PLAN
------------------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a, t2.b
2018-03-29 21:47:57 +02:00
-> Append
-> Hash Right Join
Hash Cond: ((t1_1.a = t2_1.b) AND ((t1_1.c)::text = (t2_1.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p1 t1_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p1 t2_1
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
-> Hash Right Join
Hash Cond: ((t1_2.a = t2_2.b) AND ((t1_2.c)::text = (t2_2.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p1 t1_2
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p1 t2_2
2018-03-29 21:47:57 +02:00
Filter: (a = 0)
-> Hash Right Join
Hash Cond: ((t1_3.a = t2_3.b) AND ((t1_3.c)::text = (t2_3.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p2 t1_3
-> Hash
-> Seq Scan on prt2_l_p2_p2 t2_3
Filter: (a = 0)
-> Hash Right Join
Hash Cond: ((t1_5.a = t2_4.b) AND ((t1_5.c)::text = (t2_4.c)::text))
2018-03-29 21:47:57 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p3_p1 t1_5
-> Seq Scan on prt1_l_p3_p2 t1_6
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p3_p1 t2_4
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
Filter: (a = 0)
(29 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_l t1 RIGHT JOIN prt2_l t2 ON t1.a = t2.b AND t1.c = t2.c WHERE t2.a = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | 0 | 0000
150 | 0002 | 150 | 0002
300 | 0000 | 300 | 0000
450 | 0002 | 450 | 0002
| | 75 | 0003
| | 225 | 0001
| | 375 | 0003
| | 525 | 0001
(8 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1_l WHERE prt1_l.b = 0) t1 FULL JOIN (SELECT * FROM prt2_l WHERE prt2_l.a = 0) t2 ON (t1.a = t2.b AND t1.c = t2.c) ORDER BY t1.a, t2.b;
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
QUERY PLAN
----------------------------------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Sort Key: prt1_l.a, prt2_l.b
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_l_1.a = prt2_l_1.b) AND ((prt1_l_1.c)::text = (prt2_l_1.c)::text))
-> Seq Scan on prt1_l_p1 prt1_l_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p1 prt2_l_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_l_2.a = prt2_l_2.b) AND ((prt1_l_2.c)::text = (prt2_l_2.c)::text))
-> Seq Scan on prt1_l_p2_p1 prt1_l_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p1 prt2_l_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_l_3.a = prt2_l_3.b) AND ((prt1_l_3.c)::text = (prt2_l_3.c)::text))
-> Seq Scan on prt1_l_p2_p2 prt1_l_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (b = 0)
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p2 prt2_l_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Filter: (a = 0)
-> Hash Full Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
Hash Cond: ((prt1_l_4.a = prt2_l_4.b) AND ((prt1_l_4.c)::text = (prt2_l_4.c)::text))
-> Seq Scan on prt1_l_p3_p1 prt1_l_4
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
Filter: (b = 0)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p3_p1 prt2_l_4
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
Filter: (a = 0)
(31 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1_l WHERE prt1_l.b = 0) t1 FULL JOIN (SELECT * FROM prt2_l WHERE prt2_l.a = 0) t2 ON (t1.a = t2.b AND t1.c = t2.c) ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
0 | 0000 | 0 | 0000
50 | 0002 | |
100 | 0000 | |
150 | 0002 | 150 | 0002
200 | 0000 | |
250 | 0002 | |
300 | 0000 | 300 | 0000
350 | 0002 | |
400 | 0000 | |
450 | 0002 | 450 | 0002
500 | 0000 | |
550 | 0002 | |
| | 75 | 0003
| | 225 | 0001
| | 375 | 0003
| | 525 | 0001
(16 rows)
2018-02-16 16:33:59 +01:00
-- lateral partitionwise join
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT * FROM prt1_l t1 LEFT JOIN LATERAL
(SELECT t2.a AS t2a, t2.c AS t2c, t2.b AS t2b, t3.b AS t3b, least(t1.a,t2.a,t3.b) FROM prt1_l t2 JOIN prt2_l t3 ON (t2.a = t3.b AND t2.c = t3.c)) ss
ON t1.a = ss.t2a AND t1.c = ss.t2c WHERE t1.b = 0 ORDER BY t1.a;
2018-03-29 21:47:57 +02:00
QUERY PLAN
-----------------------------------------------------------------------------------------------
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
Sort
Sort Key: t1.a
2018-03-29 21:47:57 +02:00
-> Append
-> Nested Loop Left Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p1 t1_1
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
-> Hash Join
Hash Cond: ((t3_1.b = t2_1.a) AND ((t3_1.c)::text = (t2_1.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p1 t3_1
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p1 t2_1
2018-03-29 21:47:57 +02:00
Filter: ((t1_1.a = a) AND ((t1_1.c)::text = (c)::text))
-> Nested Loop Left Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p1 t1_2
2018-03-29 21:47:57 +02:00
Filter: (b = 0)
-> Hash Join
Hash Cond: ((t3_2.b = t2_2.a) AND ((t3_2.c)::text = (t2_2.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p1 t3_2
2018-03-29 21:47:57 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p1 t2_2
2018-03-29 21:47:57 +02:00
Filter: ((t1_2.a = a) AND ((t1_2.c)::text = (c)::text))
-> Nested Loop Left Join
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p2_p2 t1_3
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
Filter: (b = 0)
2018-03-29 21:47:57 +02:00
-> Hash Join
Hash Cond: ((t3_3.b = t2_3.a) AND ((t3_3.c)::text = (t2_3.c)::text))
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p2_p2 t3_3
-> Hash
-> Seq Scan on prt1_l_p2_p2 t2_3
Filter: ((t1_3.a = a) AND ((t1_3.c)::text = (c)::text))
-> Nested Loop Left Join
-> Seq Scan on prt1_l_p3_p1 t1_4
Filter: (b = 0)
-> Hash Join
Hash Cond: ((t3_5.b = t2_5.a) AND ((t3_5.c)::text = (t2_5.c)::text))
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p3_p1 t3_5
-> Seq Scan on prt2_l_p3_p2 t3_6
2018-03-29 21:47:57 +02:00
-> Hash
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_l_p3_p1 t2_5
Filter: ((t1_4.a = a) AND ((t1_4.c)::text = (c)::text))
-> Seq Scan on prt1_l_p3_p2 t2_6
Filter: ((t1_4.a = a) AND ((t1_4.c)::text = (c)::text))
Suppress Append and MergeAppend plan nodes that have a single child.
If there's only one child relation, the Append or MergeAppend isn't
doing anything useful, and can be elided. It does have a purpose
during planning though, which is to serve as a buffer between parent
and child Var numbering. Therefore we keep it all the way through
to setrefs.c, and get rid of it only after fixing references in the
plan level(s) above it. This works largely the same as setrefs.c's
ancient hack to get rid of no-op SubqueryScan nodes, and can even
share some code with that.
Note the change to make setrefs.c use apply_tlist_labeling rather than
ad-hoc code. This has the effect of propagating the child's resjunk
and ressortgroupref labels, which formerly weren't propagated when
removing a SubqueryScan. Doing that is demonstrably necessary for
the [Merge]Append cases, and seems harmless for SubqueryScan, if only
because trivial_subqueryscan is afraid to collapse cases where the
resjunk marking differs. (I suspect that restriction could now be
removed, though it's unclear that it'd make any new matches possible,
since the outer query can't have references to a child resjunk column.)
David Rowley, reviewed by Alvaro Herrera and Tomas Vondra
Discussion: https://postgr.es/m/CAKJS1f_7u8ATyJ1JGTMHFoKDvZdeF-iEBhs+sM_SXowOr9cArg@mail.gmail.com
2019-03-25 20:42:35 +01:00
(44 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
SELECT * FROM prt1_l t1 LEFT JOIN LATERAL
(SELECT t2.a AS t2a, t2.c AS t2c, t2.b AS t2b, t3.b AS t3b, least(t1.a,t2.a,t3.b) FROM prt1_l t2 JOIN prt2_l t3 ON (t2.a = t3.b AND t2.c = t3.c)) ss
ON t1.a = ss.t2a AND t1.c = ss.t2c WHERE t1.b = 0 ORDER BY t1.a;
a | b | c | t2a | t2c | t2b | t3b | least
-----+---+------+-----+------+-----+-----+-------
0 | 0 | 0000 | 0 | 0000 | 0 | 0 | 0
50 | 0 | 0002 | | | | |
100 | 0 | 0000 | | | | |
150 | 0 | 0002 | 150 | 0002 | 0 | 150 | 150
200 | 0 | 0000 | | | | |
250 | 0 | 0002 | | | | |
300 | 0 | 0000 | 300 | 0000 | 0 | 300 | 300
350 | 0 | 0002 | | | | |
400 | 0 | 0000 | | | | |
450 | 0 | 0002 | 450 | 0002 | 0 | 450 | 450
500 | 0 | 0000 | | | | |
550 | 0 | 0002 | | | | |
(12 rows)
-- join with one side empty
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT * FROM prt1_l WHERE a = 1 AND a = 2) t1 RIGHT JOIN prt2_l t2 ON t1.a = t2.b AND t1.b = t2.a AND t1.c = t2.c;
QUERY PLAN
-------------------------------------------------------------------------
Hash Left Join
Hash Cond: ((t2.b = a) AND (t2.a = b) AND ((t2.c)::text = (c)::text))
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_l_p1 t2_1
-> Seq Scan on prt2_l_p2_p1 t2_2
-> Seq Scan on prt2_l_p2_p2 t2_3
-> Seq Scan on prt2_l_p3_p1 t2_4
-> Seq Scan on prt2_l_p3_p2 t2_5
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Result
One-Time Filter: false
(11 rows)
2018-08-11 21:53:20 +02:00
-- Test case to verify proper handling of subqueries in a partitioned delete.
-- The weird-looking lateral join is just there to force creation of a
-- nestloop parameter within the subquery, which exposes the problem if the
-- planner fails to make multiple copies of the subquery as appropriate.
EXPLAIN (COSTS OFF)
DELETE FROM prt1_l
WHERE EXISTS (
SELECT 1
FROM int4_tbl,
LATERAL (SELECT int4_tbl.f1 FROM int8_tbl LIMIT 2) ss
WHERE prt1_l.c IS NULL);
QUERY PLAN
---------------------------------------------------------------
Delete on prt1_l
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
Delete on prt1_l_p1 prt1_l_1
Delete on prt1_l_p3_p1 prt1_l_2
Delete on prt1_l_p3_p2 prt1_l_3
2018-08-11 21:53:20 +02:00
-> Nested Loop Semi Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
-> Seq Scan on prt1_l_p1 prt1_l_1
2018-08-11 21:53:20 +02:00
Filter: (c IS NULL)
-> Nested Loop
-> Seq Scan on int4_tbl
-> Subquery Scan on ss
-> Limit
-> Seq Scan on int8_tbl
-> Nested Loop Semi Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
-> Seq Scan on prt1_l_p3_p1 prt1_l_2
2018-08-11 21:53:20 +02:00
Filter: (c IS NULL)
-> Nested Loop
-> Seq Scan on int4_tbl
-> Subquery Scan on ss_1
-> Limit
-> Seq Scan on int8_tbl int8_tbl_1
-> Nested Loop Semi Join
Fix EXPLAIN's column alias output for mismatched child tables.
If an inheritance/partitioning parent table is assigned some column
alias names in the query, EXPLAIN mapped those aliases onto the
child tables' columns by physical position, resulting in bogus output
if a child table's columns aren't one-for-one with the parent's.
To fix, make expand_single_inheritance_child() generate a correctly
re-mapped column alias list, rather than just copying the parent
RTE's alias node. (We have to fill the alias field, not just
adjust the eref field, because ruleutils.c will ignore eref in
favor of looking at the real column names.)
This means that child tables will now always have alias fields in
plan rtables, where before they might not have. That results in
a rather substantial set of regression test output changes:
EXPLAIN will now always show child tables with aliases that match
the parent table (usually with "_N" appended for uniqueness).
But that seems like a net positive for understandability, since
the parent alias corresponds to something that actually appeared
in the original query, while the child table names didn't.
(Note that this does not change anything for cases where an explicit
table alias was written in the query for the parent table; it
just makes cases without such aliases behave similarly to that.)
Hence, while we could avoid these subsidiary changes if we made
inherit.c more complicated, we choose not to.
Discussion: https://postgr.es/m/12424.1575168015@sss.pgh.pa.us
2019-12-03 01:08:10 +01:00
-> Seq Scan on prt1_l_p3_p2 prt1_l_3
2018-08-11 21:53:20 +02:00
Filter: (c IS NULL)
-> Nested Loop
-> Seq Scan on int4_tbl
-> Subquery Scan on ss_2
-> Limit
-> Seq Scan on int8_tbl int8_tbl_2
(28 rows)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
--
-- negative testcases
--
CREATE TABLE prt1_n (a int, b int, c varchar) PARTITION BY RANGE(c);
CREATE TABLE prt1_n_p1 PARTITION OF prt1_n FOR VALUES FROM ('0000') TO ('0250');
CREATE TABLE prt1_n_p2 PARTITION OF prt1_n FOR VALUES FROM ('0250') TO ('0500');
INSERT INTO prt1_n SELECT i, i, to_char(i, 'FM0000') FROM generate_series(0, 499, 2) i;
ANALYZE prt1_n;
CREATE TABLE prt2_n (a int, b int, c text) PARTITION BY LIST(c);
CREATE TABLE prt2_n_p1 PARTITION OF prt2_n FOR VALUES IN ('0000', '0003', '0004', '0010', '0006', '0007');
CREATE TABLE prt2_n_p2 PARTITION OF prt2_n FOR VALUES IN ('0001', '0005', '0002', '0009', '0008', '0011');
INSERT INTO prt2_n SELECT i, i, to_char(i/50, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE prt2_n;
CREATE TABLE prt3_n (a int, b int, c text) PARTITION BY LIST(c);
CREATE TABLE prt3_n_p1 PARTITION OF prt3_n FOR VALUES IN ('0000', '0004', '0006', '0007');
CREATE TABLE prt3_n_p2 PARTITION OF prt3_n FOR VALUES IN ('0001', '0002', '0008', '0010');
CREATE TABLE prt3_n_p3 PARTITION OF prt3_n FOR VALUES IN ('0003', '0005', '0009', '0011');
INSERT INTO prt2_n SELECT i, i, to_char(i/50, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE prt3_n;
CREATE TABLE prt4_n (a int, b int, c text) PARTITION BY RANGE(a);
CREATE TABLE prt4_n_p1 PARTITION OF prt4_n FOR VALUES FROM (0) TO (300);
CREATE TABLE prt4_n_p2 PARTITION OF prt4_n FOR VALUES FROM (300) TO (500);
CREATE TABLE prt4_n_p3 PARTITION OF prt4_n FOR VALUES FROM (500) TO (600);
INSERT INTO prt4_n SELECT i, i, to_char(i, 'FM0000') FROM generate_series(0, 599, 2) i;
ANALYZE prt4_n;
2018-02-16 16:33:59 +01:00
-- partitionwise join can not be applied if the partition ranges differ
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt4_n t2 WHERE t1.a = t2.a;
QUERY PLAN
----------------------------------------------
Hash Join
Hash Cond: (t1.a = t2.a)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
-> Seq Scan on prt1_p2 t1_2
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt4_n_p1 t2_1
-> Seq Scan on prt4_n_p2 t2_2
-> Seq Scan on prt4_n_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(11 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1, prt4_n t2, prt2 t3 WHERE t1.a = t2.a and t1.a = t3.b;
QUERY PLAN
--------------------------------------------------------
Hash Join
Hash Cond: (t2.a = t1.a)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt4_n_p1 t2_1
-> Seq Scan on prt4_n_p2 t2_2
-> Seq Scan on prt4_n_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
-> Hash Join
Hash Cond: (t1_1.a = t3_1.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p1 t3_1
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash Join
Hash Cond: (t1_2.a = t3_2.b)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p2 t1_2
-> Hash
-> Seq Scan on prt2_p2 t3_2
-> Hash Join
Hash Cond: (t1_3.a = t3_3.b)
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_p3 t3_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(23 rows)
2018-02-16 16:33:59 +01:00
-- partitionwise join can not be applied if there are no equi-join conditions
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-- between partition keys
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1 t1 LEFT JOIN prt2 t2 ON (t1.a < t2.b);
QUERY PLAN
---------------------------------------------------------
Nested Loop Left Join
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t1_1
-> Seq Scan on prt1_p2 t1_2
-> Seq Scan on prt1_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p1_b on prt2_p1 t2_1
Refactor the representation of indexable clauses in IndexPaths.
In place of three separate but interrelated lists (indexclauses,
indexquals, and indexqualcols), an IndexPath now has one list
"indexclauses" of IndexClause nodes. This holds basically the same
information as before, but in a more useful format: in particular, there
is now a clear connection between an indexclause (an original restriction
clause from WHERE or JOIN/ON) and the indexquals (directly usable index
conditions) derived from it.
We also change the ground rules a bit by mandating that clause commutation,
if needed, be done up-front so that what is stored in the indexquals list
is always directly usable as an index condition. This gets rid of repeated
re-determination of which side of the clause is the indexkey during costing
and plan generation, as well as repeated lookups of the commutator
operator. To minimize the added up-front cost, the typical case of
commuting a plain OpExpr is handled by a new special-purpose function
commute_restrictinfo(). For RowCompareExprs, generating the new clause
properly commuted to begin with is not really any more complex than before,
it's just different --- and we can save doing that work twice, as the
pretty-klugy original implementation did.
Tracking the connection between original and derived clauses lets us
also track explicitly whether the derived clauses are an exact or lossy
translation of the original. This provides a cheap solution to getting
rid of unnecessary rechecks of boolean index clauses, which previously
seemed like it'd be more expensive than it was worth.
Another pleasant (IMO) side-effect is that EXPLAIN now always shows
index clauses with the indexkey on the left; this seems less confusing.
This commit leaves expand_indexqual_conditions() and some related
functions in a slightly messy state. I didn't bother to change them
any more than minimally necessary to work with the new data structure,
because all that code is going to be refactored out of existence in
a follow-on patch.
Discussion: https://postgr.es/m/22182.1549124950@sss.pgh.pa.us
2019-02-09 23:30:43 +01:00
Index Cond: (b > t1.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p2_b on prt2_p2 t2_2
Refactor the representation of indexable clauses in IndexPaths.
In place of three separate but interrelated lists (indexclauses,
indexquals, and indexqualcols), an IndexPath now has one list
"indexclauses" of IndexClause nodes. This holds basically the same
information as before, but in a more useful format: in particular, there
is now a clear connection between an indexclause (an original restriction
clause from WHERE or JOIN/ON) and the indexquals (directly usable index
conditions) derived from it.
We also change the ground rules a bit by mandating that clause commutation,
if needed, be done up-front so that what is stored in the indexquals list
is always directly usable as an index condition. This gets rid of repeated
re-determination of which side of the clause is the indexkey during costing
and plan generation, as well as repeated lookups of the commutator
operator. To minimize the added up-front cost, the typical case of
commuting a plain OpExpr is handled by a new special-purpose function
commute_restrictinfo(). For RowCompareExprs, generating the new clause
properly commuted to begin with is not really any more complex than before,
it's just different --- and we can save doing that work twice, as the
pretty-klugy original implementation did.
Tracking the connection between original and derived clauses lets us
also track explicitly whether the derived clauses are an exact or lossy
translation of the original. This provides a cheap solution to getting
rid of unnecessary rechecks of boolean index clauses, which previously
seemed like it'd be more expensive than it was worth.
Another pleasant (IMO) side-effect is that EXPLAIN now always shows
index clauses with the indexkey on the left; this seems less confusing.
This commit leaves expand_indexqual_conditions() and some related
functions in a slightly messy state. I didn't bother to change them
any more than minimally necessary to work with the new data structure,
because all that code is going to be refactored out of existence in
a follow-on patch.
Discussion: https://postgr.es/m/22182.1549124950@sss.pgh.pa.us
2019-02-09 23:30:43 +01:00
Index Cond: (b > t1.a)
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Index Scan using iprt2_p3_b on prt2_p3 t2_3
Refactor the representation of indexable clauses in IndexPaths.
In place of three separate but interrelated lists (indexclauses,
indexquals, and indexqualcols), an IndexPath now has one list
"indexclauses" of IndexClause nodes. This holds basically the same
information as before, but in a more useful format: in particular, there
is now a clear connection between an indexclause (an original restriction
clause from WHERE or JOIN/ON) and the indexquals (directly usable index
conditions) derived from it.
We also change the ground rules a bit by mandating that clause commutation,
if needed, be done up-front so that what is stored in the indexquals list
is always directly usable as an index condition. This gets rid of repeated
re-determination of which side of the clause is the indexkey during costing
and plan generation, as well as repeated lookups of the commutator
operator. To minimize the added up-front cost, the typical case of
commuting a plain OpExpr is handled by a new special-purpose function
commute_restrictinfo(). For RowCompareExprs, generating the new clause
properly commuted to begin with is not really any more complex than before,
it's just different --- and we can save doing that work twice, as the
pretty-klugy original implementation did.
Tracking the connection between original and derived clauses lets us
also track explicitly whether the derived clauses are an exact or lossy
translation of the original. This provides a cheap solution to getting
rid of unnecessary rechecks of boolean index clauses, which previously
seemed like it'd be more expensive than it was worth.
Another pleasant (IMO) side-effect is that EXPLAIN now always shows
index clauses with the indexkey on the left; this seems less confusing.
This commit leaves expand_indexqual_conditions() and some related
functions in a slightly messy state. I didn't bother to change them
any more than minimally necessary to work with the new data structure,
because all that code is going to be refactored out of existence in
a follow-on patch.
Discussion: https://postgr.es/m/22182.1549124950@sss.pgh.pa.us
2019-02-09 23:30:43 +01:00
Index Cond: (b > t1.a)
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(12 rows)
-- equi-join with join condition on partial keys does not qualify for
2018-02-16 16:33:59 +01:00
-- partitionwise join
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_m t1, prt2_m t2 WHERE t1.a = (t2.b + t2.a)/2;
QUERY PLAN
----------------------------------------------
Hash Join
Hash Cond: (((t2.b + t2.a) / 2) = t1.a)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_m_p1 t2_1
-> Seq Scan on prt2_m_p2 t2_2
-> Seq Scan on prt2_m_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_m_p1 t1_1
-> Seq Scan on prt1_m_p2 t1_2
-> Seq Scan on prt1_m_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(11 rows)
-- equi-join between out-of-order partition key columns does not qualify for
2018-02-16 16:33:59 +01:00
-- partitionwise join
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_m t1 LEFT JOIN prt2_m t2 ON t1.a = t2.b;
QUERY PLAN
----------------------------------------------
Hash Left Join
Hash Cond: (t1.a = t2.b)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_m_p1 t1_1
-> Seq Scan on prt1_m_p2 t1_2
-> Seq Scan on prt1_m_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_m_p1 t2_1
-> Seq Scan on prt2_m_p2 t2_2
-> Seq Scan on prt2_m_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(11 rows)
2018-02-16 16:33:59 +01:00
-- equi-join between non-key columns does not qualify for partitionwise join
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_m t1 LEFT JOIN prt2_m t2 ON t1.c = t2.c;
QUERY PLAN
----------------------------------------------
Hash Left Join
Hash Cond: (t1.c = t2.c)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_m_p1 t1_1
-> Seq Scan on prt1_m_p2 t1_2
-> Seq Scan on prt1_m_p3 t1_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_m_p1 t2_1
-> Seq Scan on prt2_m_p2 t2_2
-> Seq Scan on prt2_m_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(11 rows)
2019-12-16 09:00:15 +01:00
-- partitionwise join can not be applied for a join between list and range
-- partitioned tables
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_n t1 LEFT JOIN prt2_n t2 ON (t1.c = t2.c);
QUERY PLAN
----------------------------------------------
Hash Right Join
Hash Cond: (t2.c = (t1.c)::text)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_n_p1 t2_1
-> Seq Scan on prt2_n_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_n_p1 t1_1
-> Seq Scan on prt1_n_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(9 rows)
2019-12-16 09:00:15 +01:00
-- partitionwise join can not be applied between tables with different
-- partition lists
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_n t1 JOIN prt2_n t2 ON (t1.c = t2.c) JOIN plt1 t3 ON (t1.c = t3.c);
QUERY PLAN
----------------------------------------------------------
Hash Join
Hash Cond: (t2.c = (t1.c)::text)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt2_n_p1 t2_1
-> Seq Scan on prt2_n_p2 t2_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Hash Join
Hash Cond: (t3.c = (t1.c)::text)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on plt1_p1 t3_1
-> Seq Scan on plt1_p2 t3_2
-> Seq Scan on plt1_p3 t3_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_n_p1 t1_1
-> Seq Scan on prt1_n_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(16 rows)
2019-12-16 09:00:15 +01:00
-- partitionwise join can not be applied for a join between key column and
-- non-key column
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_n t1 FULL JOIN prt1 t2 ON (t1.c = t2.c);
QUERY PLAN
----------------------------------------------
Hash Full Join
Hash Cond: ((t2.c)::text = (t1.c)::text)
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_p1 t2_1
-> Seq Scan on prt1_p2 t2_2
-> Seq Scan on prt1_p3 t2_3
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
-> Hash
-> Append
Further adjust EXPLAIN's choices of table alias names.
This patch causes EXPLAIN to always assign a separate table alias to the
parent RTE of an append relation (inheritance set); before, such RTEs
were ignored if not actually scanned by the plan. Since the child RTEs
now always have that same alias to start with (cf. commit 55a1954da),
the net effect is that the parent RTE usually gets the alias used or
implied by the query text, and the children all get that alias with "_N"
appended. (The exception to "usually" is if there are duplicate aliases
in different subtrees of the original query; then some of those original
RTEs will also have "_N" appended.)
This results in more uniform output for partitioned-table plans than
we had before: the partitioned table itself gets the original alias,
and all child tables have aliases with "_N", rather than the previous
behavior where one of the children would get an alias without "_N".
The reason for giving the parent RTE an alias, even if it isn't scanned
by the plan, is that we now use the parent's alias to qualify Vars that
refer to an appendrel output column and appear above the Append or
MergeAppend that computes the appendrel. But below the append, Vars
refer to some one of the child relations, and are displayed that way.
This seems clearer than the old behavior where a Var that could carry
values from any child relation was displayed as if it referred to only
one of them.
While at it, change ruleutils.c so that the code paths used by EXPLAIN
deal in Plan trees not PlanState trees. This effectively reverts a
decision made in commit 1cc29fe7c, which seemed like a good idea at
the time to make ruleutils.c consistent with explain.c. However,
it's problematic because we'd really like to allow executor startup
pruning to remove all the children of an append node when possible,
leaving no child PlanState to resolve Vars against. (That's not done
here, but will be in the next patch.) This requires different handling
of subplans and initplans than before, but is otherwise a pretty
straightforward change.
Discussion: https://postgr.es/m/001001d4f44b$2a2cca50$7e865ef0$@lab.ntt.co.jp
2019-12-11 23:05:18 +01:00
-> Seq Scan on prt1_n_p1 t1_1
-> Seq Scan on prt1_n_p2 t1_2
Basic partition-wise join functionality.
Instead of joining two partitioned tables in their entirety we can, if
it is an equi-join on the partition keys, join the matching partitions
individually. This involves teaching the planner about "other join"
rels, which are related to regular join rels in the same way that
other member rels are related to baserels. This can use significantly
more CPU time and memory than regular join planning, because there may
now be a set of "other" rels not only for every base relation but also
for every join relation. In most practical cases, this probably
shouldn't be a problem, because (1) it's probably unusual to join many
tables each with many partitions using the partition keys for all
joins and (2) if you do that scenario then you probably have a big
enough machine to handle the increased memory cost of planning and (3)
the resulting plan is highly likely to be better, so what you spend in
planning you'll make up on the execution side. All the same, for now,
turn this feature off by default.
Currently, we can only perform joins between two tables whose
partitioning schemes are absolutely identical. It would be nice to
cope with other scenarios, such as extra partitions on one side or the
other with no match on the other side, but that will have to wait for
a future patch.
Ashutosh Bapat, reviewed and tested by Rajkumar Raghuwanshi, Amit
Langote, Rafia Sabih, Thomas Munro, Dilip Kumar, Antonin Houska, Amit
Khandekar, and by me. A few final adjustments by me.
Discussion: http://postgr.es/m/CAFjFpRfQ8GrQvzp3jA2wnLqrHmaXna-urjm_UY9BqXj=EaDTSA@mail.gmail.com
Discussion: http://postgr.es/m/CAFjFpRcitjfrULr5jfuKWRPsGUX0LQ0k8-yG0Qw2+1LBGNpMdw@mail.gmail.com
2017-10-06 17:11:10 +02:00
(10 rows)
Fix bitmap AND/OR scans on the inside of a nestloop partition-wise join.
reparameterize_path_by_child() failed to reparameterize BitmapAnd
and BitmapOr paths. This matters only if such a path is chosen as
the inside of a nestloop partition-wise join, where we have to pass
in parameters from the outside of the nestloop. If that did happen,
we generated a bad plan that would likely lead to crashes at execution.
This is not entirely reparameterize_path_by_child()'s fault though;
it's the victim of an ancient decision (my ancient decision, I think)
to not bother filling in param_info in BitmapAnd/Or path nodes. That
caused the function to believe that such nodes and their children
contain no parameter references and so need not be processed.
In hindsight that decision looks pretty penny-wise and pound-foolish:
while it saves a few cycles during path node setup, we do commonly
need the information later. In particular, by reversing the decision
and requiring valid param_info data in all nodes of a bitmap path
tree, we can get rid of indxpath.c's get_bitmap_tree_required_outer()
function, which computed the data on-demand. It's not unlikely that
that nets out as a savings of cycles in many scenarios. A couple
of other things in indxpath.c can be simplified as well.
While here, get rid of some cases in reparameterize_path_by_child()
that are visibly dead or useless, given that we only care about
reparameterizing paths that can be on the inside of a parameterized
nestloop. This case reminds one of the maxim that untested code
probably does not work, so I'm unwilling to leave unreachable code
in this function. (I did leave the T_Gather case in place even
though it's not reached in the regression tests. It's not very
clear to me when the planner might prefer to put Gather below
rather than above a nestloop, but at least in principle the case
might be interesting.)
Per bug #16536, originally from Arne Roland but with a test case
by Andrew Gierth. Back-patch to v11 where this code came in.
Discussion: https://postgr.es/m/16536-2213ee0b3aad41fd@postgresql.org
2020-07-15 00:56:49 +02:00
--
-- Test some other plan types in a partitionwise join (unfortunately,
-- we need larger tables to get the planner to choose these plan types)
--
create temp table prtx1 (a integer, b integer, c integer)
partition by range (a);
create temp table prtx1_1 partition of prtx1 for values from (1) to (11);
create temp table prtx1_2 partition of prtx1 for values from (11) to (21);
create temp table prtx1_3 partition of prtx1 for values from (21) to (31);
create temp table prtx2 (a integer, b integer, c integer)
partition by range (a);
create temp table prtx2_1 partition of prtx2 for values from (1) to (11);
create temp table prtx2_2 partition of prtx2 for values from (11) to (21);
create temp table prtx2_3 partition of prtx2 for values from (21) to (31);
insert into prtx1 select 1 + i%30, i, i
from generate_series(1,1000) i;
insert into prtx2 select 1 + i%30, i, i
from generate_series(1,500) i, generate_series(1,10) j;
create index on prtx2 (b);
create index on prtx2 (c);
analyze prtx1;
analyze prtx2;
explain (costs off)
select * from prtx1
where not exists (select 1 from prtx2
where prtx2.a=prtx1.a and prtx2.b=prtx1.b and prtx2.c=123)
and a<20 and c=120;
QUERY PLAN
-------------------------------------------------------------
Append
-> Nested Loop Anti Join
-> Seq Scan on prtx1_1
Filter: ((a < 20) AND (c = 120))
-> Bitmap Heap Scan on prtx2_1
Recheck Cond: ((b = prtx1_1.b) AND (c = 123))
Filter: (a = prtx1_1.a)
-> BitmapAnd
-> Bitmap Index Scan on prtx2_1_b_idx
Index Cond: (b = prtx1_1.b)
-> Bitmap Index Scan on prtx2_1_c_idx
Index Cond: (c = 123)
-> Nested Loop Anti Join
-> Seq Scan on prtx1_2
Filter: ((a < 20) AND (c = 120))
-> Bitmap Heap Scan on prtx2_2
Recheck Cond: ((b = prtx1_2.b) AND (c = 123))
Filter: (a = prtx1_2.a)
-> BitmapAnd
-> Bitmap Index Scan on prtx2_2_b_idx
Index Cond: (b = prtx1_2.b)
-> Bitmap Index Scan on prtx2_2_c_idx
Index Cond: (c = 123)
(23 rows)
select * from prtx1
where not exists (select 1 from prtx2
where prtx2.a=prtx1.a and prtx2.b=prtx1.b and prtx2.c=123)
and a<20 and c=120;
a | b | c
---+-----+-----
1 | 120 | 120
(1 row)
explain (costs off)
select * from prtx1
where not exists (select 1 from prtx2
where prtx2.a=prtx1.a and (prtx2.b=prtx1.b+1 or prtx2.c=99))
and a<20 and c=91;
QUERY PLAN
-----------------------------------------------------------------
Append
-> Nested Loop Anti Join
-> Seq Scan on prtx1_1
Filter: ((a < 20) AND (c = 91))
-> Bitmap Heap Scan on prtx2_1
Recheck Cond: ((b = (prtx1_1.b + 1)) OR (c = 99))
Filter: (a = prtx1_1.a)
-> BitmapOr
-> Bitmap Index Scan on prtx2_1_b_idx
Index Cond: (b = (prtx1_1.b + 1))
-> Bitmap Index Scan on prtx2_1_c_idx
Index Cond: (c = 99)
-> Nested Loop Anti Join
-> Seq Scan on prtx1_2
Filter: ((a < 20) AND (c = 91))
-> Bitmap Heap Scan on prtx2_2
Recheck Cond: ((b = (prtx1_2.b + 1)) OR (c = 99))
Filter: (a = prtx1_2.a)
-> BitmapOr
-> Bitmap Index Scan on prtx2_2_b_idx
Index Cond: (b = (prtx1_2.b + 1))
-> Bitmap Index Scan on prtx2_2_c_idx
Index Cond: (c = 99)
(23 rows)
select * from prtx1
where not exists (select 1 from prtx2
where prtx2.a=prtx1.a and (prtx2.b=prtx1.b+1 or prtx2.c=99))
and a<20 and c=91;
a | b | c
---+----+----
2 | 91 | 91
(1 row)
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
--
-- Test advanced partition-matching algorithm for partitioned join
--
-- Tests for range-partitioned tables
CREATE TABLE prt1_adv (a int, b int, c varchar) PARTITION BY RANGE (a);
CREATE TABLE prt1_adv_p1 PARTITION OF prt1_adv FOR VALUES FROM (100) TO (200);
CREATE TABLE prt1_adv_p2 PARTITION OF prt1_adv FOR VALUES FROM (200) TO (300);
CREATE TABLE prt1_adv_p3 PARTITION OF prt1_adv FOR VALUES FROM (300) TO (400);
CREATE INDEX prt1_adv_a_idx ON prt1_adv (a);
INSERT INTO prt1_adv SELECT i, i % 25, to_char(i, 'FM0000') FROM generate_series(100, 399) i;
ANALYZE prt1_adv;
CREATE TABLE prt2_adv (a int, b int, c varchar) PARTITION BY RANGE (b);
CREATE TABLE prt2_adv_p1 PARTITION OF prt2_adv FOR VALUES FROM (100) TO (150);
CREATE TABLE prt2_adv_p2 PARTITION OF prt2_adv FOR VALUES FROM (200) TO (300);
CREATE TABLE prt2_adv_p3 PARTITION OF prt2_adv FOR VALUES FROM (350) TO (500);
CREATE INDEX prt2_adv_b_idx ON prt2_adv (b);
INSERT INTO prt2_adv_p1 SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(100, 149) i;
INSERT INTO prt2_adv_p2 SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(200, 299) i;
INSERT INTO prt2_adv_p3 SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(350, 499) i;
ANALYZE prt2_adv;
-- inner join
2018-07-06 03:56:12 +02:00
EXPLAIN (COSTS OFF)
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
2018-07-06 03:56:12 +02:00
Sort
Sort Key: t1.a
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
-> Append
-> Hash Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash Join
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash Join
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_adv_p3 t2_3
-> Hash
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
(21 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
350 | 0350 | 350 | 0350
375 | 0375 | 375 | 0375
(8 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1_adv t1 WHERE EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Semi Join
Hash Cond: (t1_1.a = t2_1.b)
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash Semi Join
Hash Cond: (t1_2.a = t2_2.b)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash Semi Join
Hash Cond: (t1_3.a = t2_3.b)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p3 t2_3
(21 rows)
SELECT t1.* FROM prt1_adv t1 WHERE EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
100 | 0 | 0100
125 | 0 | 0125
200 | 0 | 0200
225 | 0 | 0225
250 | 0 | 0250
275 | 0 | 0275
350 | 0 | 0350
375 | 0 | 0375
(8 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a, t2.b
-> Append
-> Hash Right Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash Right Join
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash Right Join
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_adv_p3 t2_3
-> Hash
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
(21 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
150 | 0150 | |
175 | 0175 | |
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
300 | 0300 | |
325 | 0325 | |
350 | 0350 | 350 | 0350
375 | 0375 | 375 | 0375
(12 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Anti Join
Hash Cond: (t1_1.a = t2_1.b)
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash Anti Join
Hash Cond: (t1_2.a = t2_2.b)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash Anti Join
Hash Cond: (t1_3.a = t2_3.b)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p3 t2_3
(21 rows)
SELECT t1.* FROM prt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
150 | 0 | 0150
175 | 0 | 0175
300 | 0 | 0300
325 | 0 | 0325
(4 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT 175 phv, * FROM prt1_adv WHERE prt1_adv.b = 0) t1 FULL JOIN (SELECT 425 phv, * FROM prt2_adv WHERE prt2_adv.a = 0) t2 ON (t1.a = t2.b) WHERE t1.phv = t1.a OR t2.phv = t2.b ORDER BY t1.a, t2.b;
QUERY PLAN
--------------------------------------------------------------------------
Sort
Sort Key: prt1_adv.a, prt2_adv.b
-> Append
-> Hash Full Join
Hash Cond: (prt1_adv_1.a = prt2_adv_1.b)
Filter: (((175) = prt1_adv_1.a) OR ((425) = prt2_adv_1.b))
-> Seq Scan on prt1_adv_p1 prt1_adv_1
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p1 prt2_adv_1
Filter: (a = 0)
-> Hash Full Join
Hash Cond: (prt1_adv_2.a = prt2_adv_2.b)
Filter: (((175) = prt1_adv_2.a) OR ((425) = prt2_adv_2.b))
-> Seq Scan on prt1_adv_p2 prt1_adv_2
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p2 prt2_adv_2
Filter: (a = 0)
-> Hash Full Join
Hash Cond: (prt2_adv_3.b = prt1_adv_3.a)
Filter: (((175) = prt1_adv_3.a) OR ((425) = prt2_adv_3.b))
-> Seq Scan on prt2_adv_p3 prt2_adv_3
Filter: (a = 0)
-> Hash
-> Seq Scan on prt1_adv_p3 prt1_adv_3
Filter: (b = 0)
(27 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT 175 phv, * FROM prt1_adv WHERE prt1_adv.b = 0) t1 FULL JOIN (SELECT 425 phv, * FROM prt2_adv WHERE prt2_adv.a = 0) t2 ON (t1.a = t2.b) WHERE t1.phv = t1.a OR t2.phv = t2.b ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
175 | 0175 | |
| | 425 | 0425
(2 rows)
-- Test cases where one side has an extra partition
CREATE TABLE prt2_adv_extra PARTITION OF prt2_adv FOR VALUES FROM (500) TO (MAXVALUE);
INSERT INTO prt2_adv SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(500, 599) i;
ANALYZE prt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash Join
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash Join
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_adv_p3 t2_3
-> Hash
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
(21 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
350 | 0350 | 350 | 0350
375 | 0375 | 375 | 0375
(8 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1_adv t1 WHERE EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Semi Join
Hash Cond: (t1_1.a = t2_1.b)
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash Semi Join
Hash Cond: (t1_2.a = t2_2.b)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash Semi Join
Hash Cond: (t1_3.a = t2_3.b)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p3 t2_3
(21 rows)
SELECT t1.* FROM prt1_adv t1 WHERE EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
100 | 0 | 0100
125 | 0 | 0125
200 | 0 | 0200
225 | 0 | 0225
250 | 0 | 0250
275 | 0 | 0275
350 | 0 | 0350
375 | 0 | 0375
(8 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a, t2.b
-> Append
-> Hash Right Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash Right Join
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash Right Join
Hash Cond: (t2_3.b = t1_3.a)
-> Seq Scan on prt2_adv_p3 t2_3
-> Hash
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
(21 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
150 | 0150 | |
175 | 0175 | |
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
300 | 0300 | |
325 | 0325 | |
350 | 0350 | 350 | 0350
375 | 0375 | 375 | 0375
(12 rows)
-- left join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.b, t1.c, t2.a, t2.c FROM prt2_adv t1 LEFT JOIN prt1_adv t2 ON (t1.b = t2.a) WHERE t1.a = 0 ORDER BY t1.b, t2.a;
QUERY PLAN
---------------------------------------------------------
Sort
Sort Key: t1.b, t2.a
-> Hash Right Join
Hash Cond: (t2.a = t1.b)
2018-07-06 03:56:12 +02:00
-> Append
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
-> Seq Scan on prt1_adv_p1 t2_1
-> Seq Scan on prt1_adv_p2 t2_2
-> Seq Scan on prt1_adv_p3 t2_3
2018-07-06 03:56:12 +02:00
-> Hash
-> Append
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
-> Seq Scan on prt2_adv_p1 t1_1
Filter: (a = 0)
-> Seq Scan on prt2_adv_p2 t1_2
Filter: (a = 0)
-> Seq Scan on prt2_adv_p3 t1_3
Filter: (a = 0)
-> Seq Scan on prt2_adv_extra t1_4
Filter: (a = 0)
(18 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Anti Join
Hash Cond: (t1_1.a = t2_1.b)
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash Anti Join
Hash Cond: (t1_2.a = t2_2.b)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash Anti Join
Hash Cond: (t1_3.a = t2_3.b)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
-> Hash
-> Seq Scan on prt2_adv_p3 t2_3
(21 rows)
SELECT t1.* FROM prt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
a | b | c
-----+---+------
150 | 0 | 0150
175 | 0 | 0175
300 | 0 | 0300
325 | 0 | 0325
(4 rows)
-- anti join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt2_adv t1 WHERE NOT EXISTS (SELECT 1 FROM prt1_adv t2 WHERE t1.b = t2.a) AND t1.a = 0 ORDER BY t1.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.b
-> Hash Anti Join
Hash Cond: (t1.b = t2.a)
-> Append
-> Seq Scan on prt2_adv_p1 t1_1
Filter: (a = 0)
-> Seq Scan on prt2_adv_p2 t1_2
Filter: (a = 0)
-> Seq Scan on prt2_adv_p3 t1_3
Filter: (a = 0)
-> Seq Scan on prt2_adv_extra t1_4
Filter: (a = 0)
-> Hash
-> Append
-> Seq Scan on prt1_adv_p1 t2_1
-> Seq Scan on prt1_adv_p2 t2_2
-> Seq Scan on prt1_adv_p3 t2_3
(18 rows)
-- full join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT 175 phv, * FROM prt1_adv WHERE prt1_adv.b = 0) t1 FULL JOIN (SELECT 425 phv, * FROM prt2_adv WHERE prt2_adv.a = 0) t2 ON (t1.a = t2.b) WHERE t1.phv = t1.a OR t2.phv = t2.b ORDER BY t1.a, t2.b;
QUERY PLAN
----------------------------------------------------------------
Sort
Sort Key: prt1_adv.a, prt2_adv.b
-> Hash Full Join
Hash Cond: (prt2_adv.b = prt1_adv.a)
Filter: (((175) = prt1_adv.a) OR ((425) = prt2_adv.b))
-> Append
-> Seq Scan on prt2_adv_p1 prt2_adv_1
Filter: (a = 0)
-> Seq Scan on prt2_adv_p2 prt2_adv_2
Filter: (a = 0)
-> Seq Scan on prt2_adv_p3 prt2_adv_3
Filter: (a = 0)
-> Seq Scan on prt2_adv_extra prt2_adv_4
Filter: (a = 0)
-> Hash
-> Append
-> Seq Scan on prt1_adv_p1 prt1_adv_1
2018-07-06 03:56:12 +02:00
Filter: (b = 0)
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
-> Seq Scan on prt1_adv_p2 prt1_adv_2
2018-07-06 03:56:12 +02:00
Filter: (b = 0)
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
-> Seq Scan on prt1_adv_p3 prt1_adv_3
2018-07-06 03:56:12 +02:00
Filter: (b = 0)
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
(22 rows)
-- 3-way join where not every pair of relations can do partitioned join
EXPLAIN (COSTS OFF)
SELECT t1.b, t1.c, t2.a, t2.c, t3.a, t3.c FROM prt2_adv t1 LEFT JOIN prt1_adv t2 ON (t1.b = t2.a) INNER JOIN prt1_adv t3 ON (t1.b = t3.a) WHERE t1.a = 0 ORDER BY t1.b, t2.a, t3.a;
QUERY PLAN
--------------------------------------------------------------------------------
Sort
Sort Key: t1.b, t2.a
-> Append
-> Nested Loop Left Join
-> Nested Loop
-> Seq Scan on prt2_adv_p1 t1_1
Filter: (a = 0)
-> Index Scan using prt1_adv_p1_a_idx on prt1_adv_p1 t3_1
Index Cond: (a = t1_1.b)
-> Index Scan using prt1_adv_p1_a_idx on prt1_adv_p1 t2_1
Index Cond: (a = t1_1.b)
-> Hash Right Join
Hash Cond: (t2_2.a = t1_2.b)
-> Seq Scan on prt1_adv_p2 t2_2
-> Hash
-> Hash Join
Hash Cond: (t3_2.a = t1_2.b)
-> Seq Scan on prt1_adv_p2 t3_2
-> Hash
-> Seq Scan on prt2_adv_p2 t1_2
Filter: (a = 0)
-> Hash Right Join
Hash Cond: (t2_3.a = t1_3.b)
-> Seq Scan on prt1_adv_p3 t2_3
-> Hash
-> Hash Join
Hash Cond: (t3_3.a = t1_3.b)
-> Seq Scan on prt1_adv_p3 t3_3
-> Hash
-> Seq Scan on prt2_adv_p3 t1_3
Filter: (a = 0)
(31 rows)
SELECT t1.b, t1.c, t2.a, t2.c, t3.a, t3.c FROM prt2_adv t1 LEFT JOIN prt1_adv t2 ON (t1.b = t2.a) INNER JOIN prt1_adv t3 ON (t1.b = t3.a) WHERE t1.a = 0 ORDER BY t1.b, t2.a, t3.a;
b | c | a | c | a | c
-----+------+-----+------+-----+------
100 | 0100 | 100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125 | 125 | 0125
200 | 0200 | 200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275 | 275 | 0275
350 | 0350 | 350 | 0350 | 350 | 0350
375 | 0375 | 375 | 0375 | 375 | 0375
(8 rows)
DROP TABLE prt2_adv_extra;
-- Test cases where a partition on one side matches multiple partitions on
-- the other side; we currently can't do partitioned join in such cases
ALTER TABLE prt2_adv DETACH PARTITION prt2_adv_p3;
-- Split prt2_adv_p3 into two partitions so that prt1_adv_p3 matches both
CREATE TABLE prt2_adv_p3_1 PARTITION OF prt2_adv FOR VALUES FROM (350) TO (375);
CREATE TABLE prt2_adv_p3_2 PARTITION OF prt2_adv FOR VALUES FROM (375) TO (500);
INSERT INTO prt2_adv SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(350, 499) i;
ANALYZE prt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: (t2.b = t1.a)
-> Append
-> Seq Scan on prt2_adv_p1 t2_1
-> Seq Scan on prt2_adv_p2 t2_2
-> Seq Scan on prt2_adv_p3_1 t2_3
-> Seq Scan on prt2_adv_p3_2 t2_4
-> Hash
-> Append
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
(17 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1_adv t1 WHERE EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Semi Join
Hash Cond: (t1.a = t2.b)
-> Append
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
-> Hash
-> Append
-> Seq Scan on prt2_adv_p1 t2_1
-> Seq Scan on prt2_adv_p2 t2_2
-> Seq Scan on prt2_adv_p3_1 t2_3
-> Seq Scan on prt2_adv_p3_2 t2_4
(17 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a, t2.b
-> Hash Right Join
Hash Cond: (t2.b = t1.a)
-> Append
-> Seq Scan on prt2_adv_p1 t2_1
-> Seq Scan on prt2_adv_p2 t2_2
-> Seq Scan on prt2_adv_p3_1 t2_3
-> Seq Scan on prt2_adv_p3_2 t2_4
-> Hash
-> Append
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
(17 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM prt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM prt2_adv t2 WHERE t1.a = t2.b) AND t1.b = 0 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Anti Join
Hash Cond: (t1.a = t2.b)
-> Append
-> Seq Scan on prt1_adv_p1 t1_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p2 t1_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 t1_3
Filter: (b = 0)
-> Hash
-> Append
-> Seq Scan on prt2_adv_p1 t2_1
-> Seq Scan on prt2_adv_p2 t2_2
-> Seq Scan on prt2_adv_p3_1 t2_3
-> Seq Scan on prt2_adv_p3_2 t2_4
(17 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM (SELECT 175 phv, * FROM prt1_adv WHERE prt1_adv.b = 0) t1 FULL JOIN (SELECT 425 phv, * FROM prt2_adv WHERE prt2_adv.a = 0) t2 ON (t1.a = t2.b) WHERE t1.phv = t1.a OR t2.phv = t2.b ORDER BY t1.a, t2.b;
QUERY PLAN
----------------------------------------------------------------
Sort
Sort Key: prt1_adv.a, prt2_adv.b
-> Hash Full Join
Hash Cond: (prt2_adv.b = prt1_adv.a)
Filter: (((175) = prt1_adv.a) OR ((425) = prt2_adv.b))
-> Append
-> Seq Scan on prt2_adv_p1 prt2_adv_1
Filter: (a = 0)
-> Seq Scan on prt2_adv_p2 prt2_adv_2
Filter: (a = 0)
-> Seq Scan on prt2_adv_p3_1 prt2_adv_3
Filter: (a = 0)
-> Seq Scan on prt2_adv_p3_2 prt2_adv_4
Filter: (a = 0)
-> Hash
-> Append
-> Seq Scan on prt1_adv_p1 prt1_adv_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p2 prt1_adv_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 prt1_adv_3
Filter: (b = 0)
(22 rows)
DROP TABLE prt2_adv_p3_1;
DROP TABLE prt2_adv_p3_2;
ANALYZE prt2_adv;
-- Test default partitions
ALTER TABLE prt1_adv DETACH PARTITION prt1_adv_p1;
-- Change prt1_adv_p1 to the default partition
ALTER TABLE prt1_adv ATTACH PARTITION prt1_adv_p1 DEFAULT;
ALTER TABLE prt1_adv DETACH PARTITION prt1_adv_p3;
ANALYZE prt1_adv;
-- We can do partitioned join even if only one of relations has the default
-- partition
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: (t2_1.b = t1_2.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_2
Filter: (b = 0)
-> Hash Join
Hash Cond: (t2_2.b = t1_1.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_1
Filter: (b = 0)
(15 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
(6 rows)
-- Restore prt1_adv_p3
ALTER TABLE prt1_adv ATTACH PARTITION prt1_adv_p3 FOR VALUES FROM (300) TO (400);
ANALYZE prt1_adv;
-- Restore prt2_adv_p3
ALTER TABLE prt2_adv ATTACH PARTITION prt2_adv_p3 FOR VALUES FROM (350) TO (500);
ANALYZE prt2_adv;
-- Partitioned join can't be applied because the default partition of prt1_adv
-- matches prt2_adv_p1 and prt2_adv_p3
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: (t2.b = t1.a)
-> Append
-> Seq Scan on prt2_adv_p1 t2_1
-> Seq Scan on prt2_adv_p2 t2_2
-> Seq Scan on prt2_adv_p3 t2_3
-> Hash
-> Append
-> Seq Scan on prt1_adv_p2 t1_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 t1_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p1 t1_3
Filter: (b = 0)
(16 rows)
ALTER TABLE prt2_adv DETACH PARTITION prt2_adv_p3;
-- Change prt2_adv_p3 to the default partition
ALTER TABLE prt2_adv ATTACH PARTITION prt2_adv_p3 DEFAULT;
ANALYZE prt2_adv;
-- Partitioned join can't be applied because the default partition of prt1_adv
-- matches prt2_adv_p1 and prt2_adv_p3
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: (t2.b = t1.a)
-> Append
-> Seq Scan on prt2_adv_p1 t2_1
-> Seq Scan on prt2_adv_p2 t2_2
-> Seq Scan on prt2_adv_p3 t2_3
-> Hash
-> Append
-> Seq Scan on prt1_adv_p2 t1_1
Filter: (b = 0)
-> Seq Scan on prt1_adv_p3 t1_2
Filter: (b = 0)
-> Seq Scan on prt1_adv_p1 t1_3
Filter: (b = 0)
(16 rows)
DROP TABLE prt1_adv_p3;
ANALYZE prt1_adv;
DROP TABLE prt2_adv_p3;
ANALYZE prt2_adv;
CREATE TABLE prt3_adv (a int, b int, c varchar) PARTITION BY RANGE (a);
CREATE TABLE prt3_adv_p1 PARTITION OF prt3_adv FOR VALUES FROM (200) TO (300);
CREATE TABLE prt3_adv_p2 PARTITION OF prt3_adv FOR VALUES FROM (300) TO (400);
CREATE INDEX prt3_adv_a_idx ON prt3_adv (a);
INSERT INTO prt3_adv SELECT i, i % 25, to_char(i, 'FM0000') FROM generate_series(200, 399) i;
ANALYZE prt3_adv;
-- 3-way join to test the default partition of a join relation
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a, t3.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) LEFT JOIN prt3_adv t3 ON (t1.a = t3.a) WHERE t1.b = 0 ORDER BY t1.a, t2.b, t3.a;
QUERY PLAN
------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.b, t3.a
-> Append
-> Hash Right Join
Hash Cond: (t3_1.a = t1_1.a)
-> Seq Scan on prt3_adv_p1 t3_1
-> Hash
-> Hash Right Join
Hash Cond: (t2_2.b = t1_1.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_1
Filter: (b = 0)
-> Hash Right Join
Hash Cond: (t3_2.a = t1_2.a)
-> Seq Scan on prt3_adv_p2 t3_2
-> Hash
-> Hash Right Join
Hash Cond: (t2_1.b = t1_2.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_2
Filter: (b = 0)
(23 rows)
SELECT t1.a, t1.c, t2.b, t2.c, t3.a, t3.c FROM prt1_adv t1 LEFT JOIN prt2_adv t2 ON (t1.a = t2.b) LEFT JOIN prt3_adv t3 ON (t1.a = t3.a) WHERE t1.b = 0 ORDER BY t1.a, t2.b, t3.a;
a | c | b | c | a | c
-----+------+-----+------+-----+------
100 | 0100 | 100 | 0100 | |
125 | 0125 | 125 | 0125 | |
150 | 0150 | | | |
175 | 0175 | | | |
200 | 0200 | 200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275 | 275 | 0275
(8 rows)
DROP TABLE prt1_adv;
DROP TABLE prt2_adv;
DROP TABLE prt3_adv;
-- Test interaction of partitioned join with partition pruning
CREATE TABLE prt1_adv (a int, b int, c varchar) PARTITION BY RANGE (a);
CREATE TABLE prt1_adv_p1 PARTITION OF prt1_adv FOR VALUES FROM (100) TO (200);
CREATE TABLE prt1_adv_p2 PARTITION OF prt1_adv FOR VALUES FROM (200) TO (300);
CREATE TABLE prt1_adv_p3 PARTITION OF prt1_adv FOR VALUES FROM (300) TO (400);
CREATE INDEX prt1_adv_a_idx ON prt1_adv (a);
INSERT INTO prt1_adv SELECT i, i % 25, to_char(i, 'FM0000') FROM generate_series(100, 399) i;
ANALYZE prt1_adv;
CREATE TABLE prt2_adv (a int, b int, c varchar) PARTITION BY RANGE (b);
CREATE TABLE prt2_adv_p1 PARTITION OF prt2_adv FOR VALUES FROM (100) TO (200);
CREATE TABLE prt2_adv_p2 PARTITION OF prt2_adv FOR VALUES FROM (200) TO (400);
CREATE INDEX prt2_adv_b_idx ON prt2_adv (b);
INSERT INTO prt2_adv SELECT i % 25, i, to_char(i, 'FM0000') FROM generate_series(100, 399) i;
ANALYZE prt2_adv;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.a < 300 AND t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
-----------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_1
Filter: ((a < 300) AND (b = 0))
-> Hash Join
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_2
Filter: ((a < 300) AND (b = 0))
(15 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.a < 300 AND t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
150 | 0150 | 150 | 0150
175 | 0175 | 175 | 0175
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
(8 rows)
DROP TABLE prt1_adv_p3;
CREATE TABLE prt1_adv_default PARTITION OF prt1_adv DEFAULT;
ANALYZE prt1_adv;
CREATE TABLE prt2_adv_default PARTITION OF prt2_adv DEFAULT;
ANALYZE prt2_adv;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.a >= 100 AND t1.a < 300 AND t1.b = 0 ORDER BY t1.a, t2.b;
QUERY PLAN
--------------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_adv_p1 t2_1
-> Hash
-> Seq Scan on prt1_adv_p1 t1_1
Filter: ((a >= 100) AND (a < 300) AND (b = 0))
-> Hash Join
Hash Cond: (t2_2.b = t1_2.a)
-> Seq Scan on prt2_adv_p2 t2_2
-> Hash
-> Seq Scan on prt1_adv_p2 t1_2
Filter: ((a >= 100) AND (a < 300) AND (b = 0))
(15 rows)
SELECT t1.a, t1.c, t2.b, t2.c FROM prt1_adv t1 INNER JOIN prt2_adv t2 ON (t1.a = t2.b) WHERE t1.a >= 100 AND t1.a < 300 AND t1.b = 0 ORDER BY t1.a, t2.b;
a | c | b | c
-----+------+-----+------
100 | 0100 | 100 | 0100
125 | 0125 | 125 | 0125
150 | 0150 | 150 | 0150
175 | 0175 | 175 | 0175
200 | 0200 | 200 | 0200
225 | 0225 | 225 | 0225
250 | 0250 | 250 | 0250
275 | 0275 | 275 | 0275
(8 rows)
DROP TABLE prt1_adv;
DROP TABLE prt2_adv;
-- Tests for list-partitioned tables
CREATE TABLE plt1_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt1_adv_p1 PARTITION OF plt1_adv FOR VALUES IN ('0001', '0003');
CREATE TABLE plt1_adv_p2 PARTITION OF plt1_adv FOR VALUES IN ('0004', '0006');
CREATE TABLE plt1_adv_p3 PARTITION OF plt1_adv FOR VALUES IN ('0008', '0009');
INSERT INTO plt1_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (1, 3, 4, 6, 8, 9);
ANALYZE plt1_adv;
CREATE TABLE plt2_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt2_adv_p1 PARTITION OF plt2_adv FOR VALUES IN ('0002', '0003');
CREATE TABLE plt2_adv_p2 PARTITION OF plt2_adv FOR VALUES IN ('0004', '0006');
CREATE TABLE plt2_adv_p3 PARTITION OF plt2_adv FOR VALUES IN ('0007', '0009');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
ANALYZE plt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
9 | 0009 | 9 | 0009
(4 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Nested Loop Semi Join
Join Filter: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt2_adv_p1 t2_1
-> Nested Loop Semi Join
Join Filter: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t2_2
-> Nested Loop Semi Join
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3 t2_3
(18 rows)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
a | b | c
---+---+------
3 | 3 | 0003
4 | 4 | 0004
6 | 6 | 0006
9 | 9 | 0009
(4 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Right Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
(6 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Nested Loop Anti Join
Join Filter: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt2_adv_p1 t2_1
-> Nested Loop Anti Join
Join Filter: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t2_2
-> Nested Loop Anti Join
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3 t2_3
(18 rows)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
a | b | c
---+---+------
1 | 1 | 0001
8 | 8 | 0008
(2 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
QUERY PLAN
-----------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.a
-> Append
-> Hash Full Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
Filter: ((COALESCE(t1_1.b, 0) < 10) AND (COALESCE(t2_1.b, 0) < 10))
-> Seq Scan on plt1_adv_p1 t1_1
-> Hash
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash Full Join
Hash Cond: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
Filter: ((COALESCE(t1_2.b, 0) < 10) AND (COALESCE(t2_2.b, 0) < 10))
-> Seq Scan on plt1_adv_p2 t1_2
-> Hash
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash Full Join
Hash Cond: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
Filter: ((COALESCE(t1_3.b, 0) < 10) AND (COALESCE(t2_3.b, 0) < 10))
-> Seq Scan on plt1_adv_p3 t1_3
-> Hash
-> Seq Scan on plt2_adv_p3 t2_3
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
a | c | a | c
---+------+---+------
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
| | 2 | 0002
| | 7 | 0007
(8 rows)
-- Test cases where one side has an extra partition
CREATE TABLE plt2_adv_extra PARTITION OF plt2_adv FOR VALUES IN ('0000');
INSERT INTO plt2_adv_extra VALUES (0, 0, '0000');
ANALYZE plt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
9 | 0009 | 9 | 0009
(4 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Nested Loop Semi Join
Join Filter: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt2_adv_p1 t2_1
-> Nested Loop Semi Join
Join Filter: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t2_2
-> Nested Loop Semi Join
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3 t2_3
(18 rows)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
a | b | c
---+---+------
3 | 3 | 0003
4 | 4 | 0004
6 | 6 | 0006
9 | 9 | 0009
(4 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Right Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
(6 rows)
-- left join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt2_adv t1 LEFT JOIN plt1_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
---------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Right Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Append
-> Seq Scan on plt1_adv_p1 t2_1
-> Seq Scan on plt1_adv_p2 t2_2
-> Seq Scan on plt1_adv_p3 t2_3
-> Hash
-> Append
-> Seq Scan on plt2_adv_extra t1_1
Filter: (b < 10)
-> Seq Scan on plt2_adv_p1 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3 t1_4
Filter: (b < 10)
(18 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Nested Loop Anti Join
Join Filter: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt2_adv_p1 t2_1
-> Nested Loop Anti Join
Join Filter: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t2_2
-> Nested Loop Anti Join
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3 t2_3
(18 rows)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
a | b | c
---+---+------
1 | 1 | 0001
8 | 8 | 0008
(2 rows)
-- anti join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt2_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt1_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Anti Join
Hash Cond: ((t1.a = t2.a) AND (t1.c = t2.c))
-> Append
-> Seq Scan on plt2_adv_extra t1_1
Filter: (b < 10)
-> Seq Scan on plt2_adv_p1 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3 t1_4
Filter: (b < 10)
-> Hash
-> Append
-> Seq Scan on plt1_adv_p1 t2_1
-> Seq Scan on plt1_adv_p2 t2_2
-> Seq Scan on plt1_adv_p3 t2_3
(18 rows)
-- full join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
QUERY PLAN
-------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.a
-> Hash Full Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
Filter: ((COALESCE(t1.b, 0) < 10) AND (COALESCE(t2.b, 0) < 10))
-> Append
-> Seq Scan on plt2_adv_extra t2_1
-> Seq Scan on plt2_adv_p1 t2_2
-> Seq Scan on plt2_adv_p2 t2_3
-> Seq Scan on plt2_adv_p3 t2_4
-> Hash
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
-> Seq Scan on plt1_adv_p2 t1_2
-> Seq Scan on plt1_adv_p3 t1_3
(15 rows)
DROP TABLE plt2_adv_extra;
-- Test cases where a partition on one side matches multiple partitions on
-- the other side; we currently can't do partitioned join in such cases
ALTER TABLE plt2_adv DETACH PARTITION plt2_adv_p2;
-- Split plt2_adv_p2 into two partitions so that plt1_adv_p2 matches both
CREATE TABLE plt2_adv_p2_1 PARTITION OF plt2_adv FOR VALUES IN ('0004');
CREATE TABLE plt2_adv_p2_2 PARTITION OF plt2_adv FOR VALUES IN ('0006');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (4, 6);
ANALYZE plt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_1 t2_2
-> Seq Scan on plt2_adv_p2_2 t2_3
-> Seq Scan on plt2_adv_p3 t2_4
-> Hash
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(17 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Semi Join
Hash Cond: ((t1.a = t2.a) AND (t1.c = t2.c))
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Hash
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_1 t2_2
-> Seq Scan on plt2_adv_p2_2 t2_3
-> Seq Scan on plt2_adv_p3 t2_4
(17 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Right Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_1 t2_2
-> Seq Scan on plt2_adv_p2_2 t2_3
-> Seq Scan on plt2_adv_p3 t2_4
-> Hash
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(17 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Anti Join
Hash Cond: ((t1.a = t2.a) AND (t1.c = t2.c))
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Hash
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_1 t2_2
-> Seq Scan on plt2_adv_p2_2 t2_3
-> Seq Scan on plt2_adv_p3 t2_4
(17 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
QUERY PLAN
-------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.a
-> Hash Full Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
Filter: ((COALESCE(t1.b, 0) < 10) AND (COALESCE(t2.b, 0) < 10))
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_1 t2_2
-> Seq Scan on plt2_adv_p2_2 t2_3
-> Seq Scan on plt2_adv_p3 t2_4
-> Hash
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
-> Seq Scan on plt1_adv_p2 t1_2
-> Seq Scan on plt1_adv_p3 t1_3
(15 rows)
DROP TABLE plt2_adv_p2_1;
DROP TABLE plt2_adv_p2_2;
-- Restore plt2_adv_p2
ALTER TABLE plt2_adv ATTACH PARTITION plt2_adv_p2 FOR VALUES IN ('0004', '0006');
-- Test NULL partitions
ALTER TABLE plt1_adv DETACH PARTITION plt1_adv_p1;
-- Change plt1_adv_p1 to the NULL partition
CREATE TABLE plt1_adv_p1_null PARTITION OF plt1_adv FOR VALUES IN (NULL, '0001', '0003');
INSERT INTO plt1_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (1, 3);
INSERT INTO plt1_adv VALUES (-1, -1, NULL);
ANALYZE plt1_adv;
ALTER TABLE plt2_adv DETACH PARTITION plt2_adv_p3;
-- Change plt2_adv_p3 to the NULL partition
CREATE TABLE plt2_adv_p3_null PARTITION OF plt2_adv FOR VALUES IN (NULL, '0007', '0009');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (7, 9);
INSERT INTO plt2_adv VALUES (-1, -1, NULL);
ANALYZE plt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1_null t1_1
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3_null t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
9 | 0009 | 9 | 0009
(4 rows)
-- semi join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Semi Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
-> Seq Scan on plt1_adv_p1_null t1_1
Filter: (b < 10)
-> Hash
-> Seq Scan on plt2_adv_p1 t2_1
-> Nested Loop Semi Join
Join Filter: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t2_2
-> Nested Loop Semi Join
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3_null t2_3
(19 rows)
SELECT t1.* FROM plt1_adv t1 WHERE EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
a | b | c
---+---+------
3 | 3 | 0003
4 | 4 | 0004
6 | 6 | 0006
9 | 9 | 0009
(4 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Right Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1_null t1_1
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3_null t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
----+------+---+------
-1 | | |
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
(7 rows)
-- anti join
EXPLAIN (COSTS OFF)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Anti Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
-> Seq Scan on plt1_adv_p1_null t1_1
Filter: (b < 10)
-> Hash
-> Seq Scan on plt2_adv_p1 t2_1
-> Nested Loop Anti Join
Join Filter: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt2_adv_p2 t2_2
-> Nested Loop Anti Join
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt2_adv_p3_null t2_3
(19 rows)
SELECT t1.* FROM plt1_adv t1 WHERE NOT EXISTS (SELECT 1 FROM plt2_adv t2 WHERE t1.a = t2.a AND t1.c = t2.c) AND t1.b < 10 ORDER BY t1.a;
a | b | c
----+----+------
-1 | -1 |
1 | 1 | 0001
8 | 8 | 0008
(3 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
QUERY PLAN
-----------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.a
-> Append
-> Hash Full Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
Filter: ((COALESCE(t1_1.b, 0) < 10) AND (COALESCE(t2_1.b, 0) < 10))
-> Seq Scan on plt1_adv_p1_null t1_1
-> Hash
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash Full Join
Hash Cond: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
Filter: ((COALESCE(t1_2.b, 0) < 10) AND (COALESCE(t2_2.b, 0) < 10))
-> Seq Scan on plt1_adv_p2 t1_2
-> Hash
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash Full Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
Filter: ((COALESCE(t1_3.b, 0) < 10) AND (COALESCE(t2_3.b, 0) < 10))
-> Seq Scan on plt2_adv_p3_null t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
a | c | a | c
----+------+----+------
-1 | | |
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
| | -1 |
| | 2 | 0002
| | 7 | 0007
(10 rows)
DROP TABLE plt1_adv_p1_null;
-- Restore plt1_adv_p1
ALTER TABLE plt1_adv ATTACH PARTITION plt1_adv_p1 FOR VALUES IN ('0001', '0003');
-- Add to plt1_adv the extra NULL partition containing only NULL values as the
-- key values
CREATE TABLE plt1_adv_extra PARTITION OF plt1_adv FOR VALUES IN (NULL);
INSERT INTO plt1_adv VALUES (-1, -1, NULL);
ANALYZE plt1_adv;
DROP TABLE plt2_adv_p3_null;
-- Restore plt2_adv_p3
ALTER TABLE plt2_adv ATTACH PARTITION plt2_adv_p3 FOR VALUES IN ('0007', '0009');
ANALYZE plt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
9 | 0009 | 9 | 0009
(4 rows)
-- left join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
---------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Right Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2 t2_2
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Seq Scan on plt1_adv_extra t1_4
Filter: (b < 10)
(18 rows)
-- full join; currently we can't do partitioned join if there are no matched
-- partitions on the nullable side
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
QUERY PLAN
-------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.a
-> Hash Full Join
Hash Cond: ((t1.a = t2.a) AND (t1.c = t2.c))
Filter: ((COALESCE(t1.b, 0) < 10) AND (COALESCE(t2.b, 0) < 10))
-> Append
-> Seq Scan on plt1_adv_p1 t1_1
-> Seq Scan on plt1_adv_p2 t1_2
-> Seq Scan on plt1_adv_p3 t1_3
-> Seq Scan on plt1_adv_extra t1_4
-> Hash
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2 t2_2
-> Seq Scan on plt2_adv_p3 t2_3
(15 rows)
-- Add to plt2_adv the extra NULL partition containing only NULL values as the
-- key values
CREATE TABLE plt2_adv_extra PARTITION OF plt2_adv FOR VALUES IN (NULL);
INSERT INTO plt2_adv VALUES (-1, -1, NULL);
ANALYZE plt2_adv;
-- inner join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
(21 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
9 | 0009 | 9 | 0009
(4 rows)
-- left join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
----------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Right Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Nested Loop Left Join
Join Filter: ((t1_4.a = t2_4.a) AND (t1_4.c = t2_4.c))
-> Seq Scan on plt1_adv_extra t1_4
Filter: (b < 10)
-> Seq Scan on plt2_adv_extra t2_4
(26 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
----+------+---+------
-1 | | |
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
(7 rows)
-- full join
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
QUERY PLAN
-----------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t2.a
-> Append
-> Hash Full Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.c = t2_1.c))
Filter: ((COALESCE(t1_1.b, 0) < 10) AND (COALESCE(t2_1.b, 0) < 10))
-> Seq Scan on plt1_adv_p1 t1_1
-> Hash
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash Full Join
Hash Cond: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
Filter: ((COALESCE(t1_2.b, 0) < 10) AND (COALESCE(t2_2.b, 0) < 10))
-> Seq Scan on plt1_adv_p2 t1_2
-> Hash
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash Full Join
Hash Cond: ((t1_3.a = t2_3.a) AND (t1_3.c = t2_3.c))
Filter: ((COALESCE(t1_3.b, 0) < 10) AND (COALESCE(t2_3.b, 0) < 10))
-> Seq Scan on plt1_adv_p3 t1_3
-> Hash
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash Full Join
Hash Cond: ((t1_4.a = t2_4.a) AND (t1_4.c = t2_4.c))
Filter: ((COALESCE(t1_4.b, 0) < 10) AND (COALESCE(t2_4.b, 0) < 10))
-> Seq Scan on plt1_adv_extra t1_4
-> Hash
-> Seq Scan on plt2_adv_extra t2_4
(27 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 FULL JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE coalesce(t1.b, 0) < 10 AND coalesce(t2.b, 0) < 10 ORDER BY t1.a, t2.a;
a | c | a | c
----+------+----+------
-1 | | |
1 | 0001 | |
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
8 | 0008 | |
9 | 0009 | 9 | 0009
| | -1 |
| | 2 | 0002
| | 7 | 0007
(10 rows)
-- 3-way join to test the NULL partition of a join relation
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c, t3.a, t3.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) LEFT JOIN plt1_adv t3 ON (t1.a = t3.a AND t1.c = t3.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Right Join
Hash Cond: ((t3_1.a = t1_1.a) AND (t3_1.c = t1_1.c))
-> Seq Scan on plt1_adv_p1 t3_1
-> Hash
-> Hash Right Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_1
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t3_2.a = t1_2.a) AND (t3_2.c = t1_2.c))
-> Seq Scan on plt1_adv_p2 t3_2
-> Hash
-> Hash Right Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_2
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t3_3.a = t1_3.a) AND (t3_3.c = t1_3.c))
-> Seq Scan on plt1_adv_p3 t3_3
-> Hash
-> Hash Right Join
Hash Cond: ((t2_3.a = t1_3.a) AND (t2_3.c = t1_3.c))
-> Seq Scan on plt2_adv_p3 t2_3
-> Hash
-> Seq Scan on plt1_adv_p3 t1_3
Filter: (b < 10)
-> Nested Loop Left Join
Join Filter: ((t1_4.a = t3_4.a) AND (t1_4.c = t3_4.c))
-> Nested Loop Left Join
Join Filter: ((t1_4.a = t2_4.a) AND (t1_4.c = t2_4.c))
-> Seq Scan on plt1_adv_extra t1_4
Filter: (b < 10)
-> Seq Scan on plt2_adv_extra t2_4
-> Seq Scan on plt1_adv_extra t3_4
(41 rows)
SELECT t1.a, t1.c, t2.a, t2.c, t3.a, t3.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) LEFT JOIN plt1_adv t3 ON (t1.a = t3.a AND t1.c = t3.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c | a | c
----+------+---+------+---+------
-1 | | | | |
1 | 0001 | | | 1 | 0001
3 | 0003 | 3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006 | 6 | 0006
8 | 0008 | | | 8 | 0008
9 | 0009 | 9 | 0009 | 9 | 0009
(7 rows)
DROP TABLE plt1_adv_extra;
DROP TABLE plt2_adv_extra;
-- Test default partitions
ALTER TABLE plt1_adv DETACH PARTITION plt1_adv_p1;
-- Change plt1_adv_p1 to the default partition
ALTER TABLE plt1_adv ATTACH PARTITION plt1_adv_p1 DEFAULT;
DROP TABLE plt1_adv_p3;
ANALYZE plt1_adv;
DROP TABLE plt2_adv_p3;
ANALYZE plt2_adv;
-- We can do partitioned join even if only one of relations has the default
-- partition
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_2.a) AND (t2_1.c = t1_2.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_1.a) AND (t2_2.c = t1_1.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_1
Filter: (b < 10)
(15 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
(3 rows)
ALTER TABLE plt2_adv DETACH PARTITION plt2_adv_p2;
-- Change plt2_adv_p2 to contain '0005' in addition to '0004' and '0006' as
-- the key values
CREATE TABLE plt2_adv_p2_ext PARTITION OF plt2_adv FOR VALUES IN ('0004', '0005', '0006');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (4, 5, 6);
ANALYZE plt2_adv;
-- Partitioned join can't be applied because the default partition of plt1_adv
-- matches plt2_adv_p1 and plt2_adv_p2_ext
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_ext t2_2
-> Hash
-> Append
-> Seq Scan on plt1_adv_p2 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p1 t1_2
Filter: (b < 10)
(13 rows)
ALTER TABLE plt2_adv DETACH PARTITION plt2_adv_p2_ext;
-- Change plt2_adv_p2_ext to the default partition
ALTER TABLE plt2_adv ATTACH PARTITION plt2_adv_p2_ext DEFAULT;
ANALYZE plt2_adv;
-- Partitioned join can't be applied because the default partition of plt1_adv
-- matches plt2_adv_p1 and plt2_adv_p2_ext
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Append
-> Seq Scan on plt2_adv_p1 t2_1
-> Seq Scan on plt2_adv_p2_ext t2_2
-> Hash
-> Append
-> Seq Scan on plt1_adv_p2 t1_1
Filter: (b < 10)
-> Seq Scan on plt1_adv_p1 t1_2
Filter: (b < 10)
(13 rows)
DROP TABLE plt2_adv_p2_ext;
-- Restore plt2_adv_p2
ALTER TABLE plt2_adv ATTACH PARTITION plt2_adv_p2 FOR VALUES IN ('0004', '0006');
ANALYZE plt2_adv;
CREATE TABLE plt3_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt3_adv_p1 PARTITION OF plt3_adv FOR VALUES IN ('0004', '0006');
CREATE TABLE plt3_adv_p2 PARTITION OF plt3_adv FOR VALUES IN ('0007', '0009');
INSERT INTO plt3_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (4, 6, 7, 9);
ANALYZE plt3_adv;
-- 3-way join to test the default partition of a join relation
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c, t3.a, t3.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) LEFT JOIN plt3_adv t3 ON (t1.a = t3.a AND t1.c = t3.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Right Join
Hash Cond: ((t3_1.a = t1_1.a) AND (t3_1.c = t1_1.c))
-> Seq Scan on plt3_adv_p1 t3_1
-> Hash
-> Hash Right Join
Hash Cond: ((t2_2.a = t1_1.a) AND (t2_2.c = t1_1.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_1
Filter: (b < 10)
-> Hash Right Join
Hash Cond: ((t3_2.a = t1_2.a) AND (t3_2.c = t1_2.c))
-> Seq Scan on plt3_adv_p2 t3_2
-> Hash
-> Hash Right Join
Hash Cond: ((t2_1.a = t1_2.a) AND (t2_1.c = t1_2.c))
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_2
Filter: (b < 10)
(23 rows)
SELECT t1.a, t1.c, t2.a, t2.c, t3.a, t3.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) LEFT JOIN plt3_adv t3 ON (t1.a = t3.a AND t1.c = t3.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c | a | c
---+------+---+------+---+------
1 | 0001 | | | |
3 | 0003 | 3 | 0003 | |
4 | 0004 | 4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006 | 6 | 0006
(4 rows)
-- Test cases where one side has the default partition while the other side
-- has the NULL partition
DROP TABLE plt2_adv_p1;
-- Add the NULL partition to plt2_adv
CREATE TABLE plt2_adv_p1_null PARTITION OF plt2_adv FOR VALUES IN (NULL, '0001', '0003');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (1, 3);
INSERT INTO plt2_adv VALUES (-1, -1, NULL);
ANALYZE plt2_adv;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_2.a) AND (t2_1.c = t1_2.c))
-> Seq Scan on plt2_adv_p1_null t2_1
-> Hash
-> Seq Scan on plt1_adv_p1 t1_2
Filter: (b < 10)
-> Hash Join
Hash Cond: ((t2_2.a = t1_1.a) AND (t2_2.c = t1_1.c))
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt1_adv_p2 t1_1
Filter: (b < 10)
(15 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
1 | 0001 | 1 | 0001
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
(4 rows)
DROP TABLE plt2_adv_p1_null;
-- Add the NULL partition that contains only NULL values as the key values
CREATE TABLE plt2_adv_p1_null PARTITION OF plt2_adv FOR VALUES IN (NULL);
INSERT INTO plt2_adv VALUES (-1, -1, NULL);
ANALYZE plt2_adv;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
QUERY PLAN
------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Seq Scan on plt2_adv_p2 t2
-> Hash
-> Seq Scan on plt1_adv_p2 t1
Filter: (b < 10)
(8 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
4 | 0004 | 4 | 0004
6 | 0006 | 6 | 0006
(2 rows)
DROP TABLE plt1_adv;
DROP TABLE plt2_adv;
DROP TABLE plt3_adv;
-- Test interaction of partitioned join with partition pruning
CREATE TABLE plt1_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt1_adv_p1 PARTITION OF plt1_adv FOR VALUES IN ('0001');
CREATE TABLE plt1_adv_p2 PARTITION OF plt1_adv FOR VALUES IN ('0002');
CREATE TABLE plt1_adv_p3 PARTITION OF plt1_adv FOR VALUES IN ('0003');
CREATE TABLE plt1_adv_p4 PARTITION OF plt1_adv FOR VALUES IN (NULL, '0004', '0005');
INSERT INTO plt1_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (1, 2, 3, 4, 5);
INSERT INTO plt1_adv VALUES (-1, -1, NULL);
ANALYZE plt1_adv;
CREATE TABLE plt2_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt2_adv_p1 PARTITION OF plt2_adv FOR VALUES IN ('0001', '0002');
CREATE TABLE plt2_adv_p2 PARTITION OF plt2_adv FOR VALUES IN (NULL);
CREATE TABLE plt2_adv_p3 PARTITION OF plt2_adv FOR VALUES IN ('0003');
CREATE TABLE plt2_adv_p4 PARTITION OF plt2_adv FOR VALUES IN ('0004', '0005');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 10, 'FM0000') FROM generate_series(1, 299) i WHERE i % 10 IN (1, 2, 3, 4, 5);
INSERT INTO plt2_adv VALUES (-1, -1, NULL);
ANALYZE plt2_adv;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IN ('0003', '0004', '0005') AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
-----------------------------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p3 t2_1
-> Hash
-> Seq Scan on plt1_adv_p3 t1_1
Filter: ((b < 10) AND (c = ANY ('{0003,0004,0005}'::text[])))
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p4 t2_2
-> Hash
-> Seq Scan on plt1_adv_p4 t1_2
Filter: ((b < 10) AND (c = ANY ('{0003,0004,0005}'::text[])))
(15 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IN ('0003', '0004', '0005') AND t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
5 | 0005 | 5 | 0005
(3 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IS NULL AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Right Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Seq Scan on plt2_adv_p4 t2
-> Hash
-> Seq Scan on plt1_adv_p4 t1
Filter: ((c IS NULL) AND (b < 10))
(8 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IS NULL AND t1.b < 10 ORDER BY t1.a;
a | c | a | c
----+---+---+---
-1 | | |
(1 row)
CREATE TABLE plt1_adv_default PARTITION OF plt1_adv DEFAULT;
ANALYZE plt1_adv;
CREATE TABLE plt2_adv_default PARTITION OF plt2_adv DEFAULT;
ANALYZE plt2_adv;
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IN ('0003', '0004', '0005') AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
-----------------------------------------------------------------------------------------
Sort
Sort Key: t1.a
-> Append
-> Hash Join
Hash Cond: ((t2_1.a = t1_1.a) AND (t2_1.c = t1_1.c))
-> Seq Scan on plt2_adv_p3 t2_1
-> Hash
-> Seq Scan on plt1_adv_p3 t1_1
Filter: ((b < 10) AND (c = ANY ('{0003,0004,0005}'::text[])))
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.c = t1_2.c))
-> Seq Scan on plt2_adv_p4 t2_2
-> Hash
-> Seq Scan on plt1_adv_p4 t1_2
Filter: ((b < 10) AND (c = ANY ('{0003,0004,0005}'::text[])))
(15 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 INNER JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IN ('0003', '0004', '0005') AND t1.b < 10 ORDER BY t1.a;
a | c | a | c
---+------+---+------
3 | 0003 | 3 | 0003
4 | 0004 | 4 | 0004
5 | 0005 | 5 | 0005
(3 rows)
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IS NULL AND t1.b < 10 ORDER BY t1.a;
QUERY PLAN
--------------------------------------------------------
Sort
Sort Key: t1.a
-> Hash Right Join
Hash Cond: ((t2.a = t1.a) AND (t2.c = t1.c))
-> Seq Scan on plt2_adv_p4 t2
-> Hash
-> Seq Scan on plt1_adv_p4 t1
Filter: ((c IS NULL) AND (b < 10))
(8 rows)
SELECT t1.a, t1.c, t2.a, t2.c FROM plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE t1.c IS NULL AND t1.b < 10 ORDER BY t1.a;
a | c | a | c
----+---+---+---
-1 | | |
(1 row)
DROP TABLE plt1_adv;
DROP TABLE plt2_adv;
-- Test the process_outer_partition() code path
CREATE TABLE plt1_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt1_adv_p1 PARTITION OF plt1_adv FOR VALUES IN ('0000', '0001', '0002');
CREATE TABLE plt1_adv_p2 PARTITION OF plt1_adv FOR VALUES IN ('0003', '0004');
INSERT INTO plt1_adv SELECT i, i, to_char(i % 5, 'FM0000') FROM generate_series(0, 24) i;
ANALYZE plt1_adv;
CREATE TABLE plt2_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt2_adv_p1 PARTITION OF plt2_adv FOR VALUES IN ('0002');
CREATE TABLE plt2_adv_p2 PARTITION OF plt2_adv FOR VALUES IN ('0003', '0004');
INSERT INTO plt2_adv SELECT i, i, to_char(i % 5, 'FM0000') FROM generate_series(0, 24) i WHERE i % 5 IN (2, 3, 4);
ANALYZE plt2_adv;
CREATE TABLE plt3_adv (a int, b int, c text) PARTITION BY LIST (c);
CREATE TABLE plt3_adv_p1 PARTITION OF plt3_adv FOR VALUES IN ('0001');
CREATE TABLE plt3_adv_p2 PARTITION OF plt3_adv FOR VALUES IN ('0003', '0004');
INSERT INTO plt3_adv SELECT i, i, to_char(i % 5, 'FM0000') FROM generate_series(0, 24) i WHERE i % 5 IN (1, 3, 4);
ANALYZE plt3_adv;
-- This tests that when merging partitions from plt1_adv and plt2_adv in
-- merge_list_bounds(), process_outer_partition() returns an already-assigned
-- merged partition when re-called with plt1_adv_p1 for the second list value
2020-09-21 18:43:42 +02:00
-- '0001' of that partition
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.a, t1.c, t2.a, t2.c, t3.a, t3.c FROM (plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.c = t2.c)) FULL JOIN plt3_adv t3 ON (t1.c = t3.c) WHERE coalesce(t1.a, 0) % 5 != 3 AND coalesce(t1.a, 0) % 5 != 4 ORDER BY t1.c, t1.a, t2.a, t3.a;
QUERY PLAN
-----------------------------------------------------------------------------------------------
Sort
Sort Key: t1.c, t1.a, t2.a, t3.a
-> Append
-> Hash Full Join
Hash Cond: (t1_1.c = t3_1.c)
Filter: (((COALESCE(t1_1.a, 0) % 5) <> 3) AND ((COALESCE(t1_1.a, 0) % 5) <> 4))
-> Hash Left Join
Hash Cond: (t1_1.c = t2_1.c)
-> Seq Scan on plt1_adv_p1 t1_1
-> Hash
-> Seq Scan on plt2_adv_p1 t2_1
-> Hash
-> Seq Scan on plt3_adv_p1 t3_1
-> Hash Full Join
Hash Cond: (t1_2.c = t3_2.c)
Filter: (((COALESCE(t1_2.a, 0) % 5) <> 3) AND ((COALESCE(t1_2.a, 0) % 5) <> 4))
-> Hash Left Join
Hash Cond: (t1_2.c = t2_2.c)
-> Seq Scan on plt1_adv_p2 t1_2
-> Hash
-> Seq Scan on plt2_adv_p2 t2_2
-> Hash
-> Seq Scan on plt3_adv_p2 t3_2
(23 rows)
SELECT t1.a, t1.c, t2.a, t2.c, t3.a, t3.c FROM (plt1_adv t1 LEFT JOIN plt2_adv t2 ON (t1.c = t2.c)) FULL JOIN plt3_adv t3 ON (t1.c = t3.c) WHERE coalesce(t1.a, 0) % 5 != 3 AND coalesce(t1.a, 0) % 5 != 4 ORDER BY t1.c, t1.a, t2.a, t3.a;
a | c | a | c | a | c
----+------+----+------+----+------
0 | 0000 | | | |
5 | 0000 | | | |
10 | 0000 | | | |
15 | 0000 | | | |
20 | 0000 | | | |
1 | 0001 | | | 1 | 0001
1 | 0001 | | | 6 | 0001
1 | 0001 | | | 11 | 0001
1 | 0001 | | | 16 | 0001
1 | 0001 | | | 21 | 0001
6 | 0001 | | | 1 | 0001
6 | 0001 | | | 6 | 0001
6 | 0001 | | | 11 | 0001
6 | 0001 | | | 16 | 0001
6 | 0001 | | | 21 | 0001
11 | 0001 | | | 1 | 0001
11 | 0001 | | | 6 | 0001
11 | 0001 | | | 11 | 0001
11 | 0001 | | | 16 | 0001
11 | 0001 | | | 21 | 0001
16 | 0001 | | | 1 | 0001
16 | 0001 | | | 6 | 0001
16 | 0001 | | | 11 | 0001
16 | 0001 | | | 16 | 0001
16 | 0001 | | | 21 | 0001
21 | 0001 | | | 1 | 0001
21 | 0001 | | | 6 | 0001
21 | 0001 | | | 11 | 0001
21 | 0001 | | | 16 | 0001
21 | 0001 | | | 21 | 0001
2 | 0002 | 2 | 0002 | |
2 | 0002 | 7 | 0002 | |
2 | 0002 | 12 | 0002 | |
2 | 0002 | 17 | 0002 | |
2 | 0002 | 22 | 0002 | |
7 | 0002 | 2 | 0002 | |
7 | 0002 | 7 | 0002 | |
7 | 0002 | 12 | 0002 | |
7 | 0002 | 17 | 0002 | |
7 | 0002 | 22 | 0002 | |
12 | 0002 | 2 | 0002 | |
12 | 0002 | 7 | 0002 | |
12 | 0002 | 12 | 0002 | |
12 | 0002 | 17 | 0002 | |
12 | 0002 | 22 | 0002 | |
17 | 0002 | 2 | 0002 | |
17 | 0002 | 7 | 0002 | |
17 | 0002 | 12 | 0002 | |
17 | 0002 | 17 | 0002 | |
17 | 0002 | 22 | 0002 | |
22 | 0002 | 2 | 0002 | |
22 | 0002 | 7 | 0002 | |
22 | 0002 | 12 | 0002 | |
22 | 0002 | 17 | 0002 | |
22 | 0002 | 22 | 0002 | |
(55 rows)
DROP TABLE plt1_adv;
DROP TABLE plt2_adv;
DROP TABLE plt3_adv;
-- Tests for multi-level partitioned tables
CREATE TABLE alpha (a double precision, b int, c text) PARTITION BY RANGE (a);
CREATE TABLE alpha_neg PARTITION OF alpha FOR VALUES FROM ('-Infinity') TO (0) PARTITION BY RANGE (b);
CREATE TABLE alpha_pos PARTITION OF alpha FOR VALUES FROM (0) TO (10.0) PARTITION BY LIST (c);
CREATE TABLE alpha_neg_p1 PARTITION OF alpha_neg FOR VALUES FROM (100) TO (200);
CREATE TABLE alpha_neg_p2 PARTITION OF alpha_neg FOR VALUES FROM (200) TO (300);
CREATE TABLE alpha_neg_p3 PARTITION OF alpha_neg FOR VALUES FROM (300) TO (400);
CREATE TABLE alpha_pos_p1 PARTITION OF alpha_pos FOR VALUES IN ('0001', '0003');
CREATE TABLE alpha_pos_p2 PARTITION OF alpha_pos FOR VALUES IN ('0004', '0006');
CREATE TABLE alpha_pos_p3 PARTITION OF alpha_pos FOR VALUES IN ('0008', '0009');
INSERT INTO alpha_neg SELECT -1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(100, 399) i WHERE i % 10 IN (1, 3, 4, 6, 8, 9);
INSERT INTO alpha_pos SELECT 1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(100, 399) i WHERE i % 10 IN (1, 3, 4, 6, 8, 9);
ANALYZE alpha;
CREATE TABLE beta (a double precision, b int, c text) PARTITION BY RANGE (a);
CREATE TABLE beta_neg PARTITION OF beta FOR VALUES FROM (-10.0) TO (0) PARTITION BY RANGE (b);
CREATE TABLE beta_pos PARTITION OF beta FOR VALUES FROM (0) TO ('Infinity') PARTITION BY LIST (c);
CREATE TABLE beta_neg_p1 PARTITION OF beta_neg FOR VALUES FROM (100) TO (150);
CREATE TABLE beta_neg_p2 PARTITION OF beta_neg FOR VALUES FROM (200) TO (300);
CREATE TABLE beta_neg_p3 PARTITION OF beta_neg FOR VALUES FROM (350) TO (500);
CREATE TABLE beta_pos_p1 PARTITION OF beta_pos FOR VALUES IN ('0002', '0003');
CREATE TABLE beta_pos_p2 PARTITION OF beta_pos FOR VALUES IN ('0004', '0006');
CREATE TABLE beta_pos_p3 PARTITION OF beta_pos FOR VALUES IN ('0007', '0009');
INSERT INTO beta_neg SELECT -1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(100, 149) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
INSERT INTO beta_neg SELECT -1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(200, 299) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
INSERT INTO beta_neg SELECT -1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(350, 499) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
INSERT INTO beta_pos SELECT 1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(100, 149) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
INSERT INTO beta_pos SELECT 1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(200, 299) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
INSERT INTO beta_pos SELECT 1.0, i, to_char(i % 10, 'FM0000') FROM generate_series(350, 499) i WHERE i % 10 IN (2, 3, 4, 6, 7, 9);
ANALYZE beta;
EXPLAIN (COSTS OFF)
SELECT t1.*, t2.* FROM alpha t1 INNER JOIN beta t2 ON (t1.a = t2.a AND t1.b = t2.b) WHERE t1.b >= 125 AND t1.b < 225 ORDER BY t1.a, t1.b;
QUERY PLAN
--------------------------------------------------------------------
Sort
Sort Key: t1.a, t1.b
-> Append
-> Hash Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.b = t2_1.b))
-> Seq Scan on alpha_neg_p1 t1_1
Filter: ((b >= 125) AND (b < 225))
-> Hash
-> Seq Scan on beta_neg_p1 t2_1
-> Hash Join
Hash Cond: ((t2_2.a = t1_2.a) AND (t2_2.b = t1_2.b))
-> Seq Scan on beta_neg_p2 t2_2
-> Hash
-> Seq Scan on alpha_neg_p2 t1_2
Filter: ((b >= 125) AND (b < 225))
-> Hash Join
Hash Cond: ((t2_4.a = t1_4.a) AND (t2_4.b = t1_4.b))
-> Append
-> Seq Scan on beta_pos_p1 t2_4
-> Seq Scan on beta_pos_p2 t2_5
-> Seq Scan on beta_pos_p3 t2_6
-> Hash
-> Append
-> Seq Scan on alpha_pos_p1 t1_4
Filter: ((b >= 125) AND (b < 225))
-> Seq Scan on alpha_pos_p2 t1_5
Filter: ((b >= 125) AND (b < 225))
-> Seq Scan on alpha_pos_p3 t1_6
Filter: ((b >= 125) AND (b < 225))
(29 rows)
SELECT t1.*, t2.* FROM alpha t1 INNER JOIN beta t2 ON (t1.a = t2.a AND t1.b = t2.b) WHERE t1.b >= 125 AND t1.b < 225 ORDER BY t1.a, t1.b;
a | b | c | a | b | c
----+-----+------+----+-----+------
-1 | 126 | 0006 | -1 | 126 | 0006
-1 | 129 | 0009 | -1 | 129 | 0009
-1 | 133 | 0003 | -1 | 133 | 0003
-1 | 134 | 0004 | -1 | 134 | 0004
-1 | 136 | 0006 | -1 | 136 | 0006
-1 | 139 | 0009 | -1 | 139 | 0009
-1 | 143 | 0003 | -1 | 143 | 0003
-1 | 144 | 0004 | -1 | 144 | 0004
-1 | 146 | 0006 | -1 | 146 | 0006
-1 | 149 | 0009 | -1 | 149 | 0009
-1 | 203 | 0003 | -1 | 203 | 0003
-1 | 204 | 0004 | -1 | 204 | 0004
-1 | 206 | 0006 | -1 | 206 | 0006
-1 | 209 | 0009 | -1 | 209 | 0009
-1 | 213 | 0003 | -1 | 213 | 0003
-1 | 214 | 0004 | -1 | 214 | 0004
-1 | 216 | 0006 | -1 | 216 | 0006
-1 | 219 | 0009 | -1 | 219 | 0009
-1 | 223 | 0003 | -1 | 223 | 0003
-1 | 224 | 0004 | -1 | 224 | 0004
1 | 126 | 0006 | 1 | 126 | 0006
1 | 129 | 0009 | 1 | 129 | 0009
1 | 133 | 0003 | 1 | 133 | 0003
1 | 134 | 0004 | 1 | 134 | 0004
1 | 136 | 0006 | 1 | 136 | 0006
1 | 139 | 0009 | 1 | 139 | 0009
1 | 143 | 0003 | 1 | 143 | 0003
1 | 144 | 0004 | 1 | 144 | 0004
1 | 146 | 0006 | 1 | 146 | 0006
1 | 149 | 0009 | 1 | 149 | 0009
1 | 203 | 0003 | 1 | 203 | 0003
1 | 204 | 0004 | 1 | 204 | 0004
1 | 206 | 0006 | 1 | 206 | 0006
1 | 209 | 0009 | 1 | 209 | 0009
1 | 213 | 0003 | 1 | 213 | 0003
1 | 214 | 0004 | 1 | 214 | 0004
1 | 216 | 0006 | 1 | 216 | 0006
1 | 219 | 0009 | 1 | 219 | 0009
1 | 223 | 0003 | 1 | 223 | 0003
1 | 224 | 0004 | 1 | 224 | 0004
(40 rows)
EXPLAIN (COSTS OFF)
SELECT t1.*, t2.* FROM alpha t1 INNER JOIN beta t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE ((t1.b >= 100 AND t1.b < 110) OR (t1.b >= 200 AND t1.b < 210)) AND ((t2.b >= 100 AND t2.b < 110) OR (t2.b >= 200 AND t2.b < 210)) AND t1.c IN ('0004', '0009') ORDER BY t1.a, t1.b, t2.b;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t1.b, t2.b
-> Append
-> Hash Join
Hash Cond: ((t1_2.a = t2_2.a) AND (t1_2.c = t2_2.c))
-> Append
-> Seq Scan on alpha_neg_p1 t1_2
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Seq Scan on alpha_neg_p2 t1_3
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Hash
-> Append
-> Seq Scan on beta_neg_p1 t2_2
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
-> Seq Scan on beta_neg_p2 t2_3
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
-> Nested Loop
Join Filter: ((t1_4.a = t2_4.a) AND (t1_4.c = t2_4.c))
-> Seq Scan on alpha_pos_p2 t1_4
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Seq Scan on beta_pos_p2 t2_4
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
-> Nested Loop
Join Filter: ((t1_5.a = t2_5.a) AND (t1_5.c = t2_5.c))
-> Seq Scan on alpha_pos_p3 t1_5
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Seq Scan on beta_pos_p3 t2_5
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
(28 rows)
SELECT t1.*, t2.* FROM alpha t1 INNER JOIN beta t2 ON (t1.a = t2.a AND t1.c = t2.c) WHERE ((t1.b >= 100 AND t1.b < 110) OR (t1.b >= 200 AND t1.b < 210)) AND ((t2.b >= 100 AND t2.b < 110) OR (t2.b >= 200 AND t2.b < 210)) AND t1.c IN ('0004', '0009') ORDER BY t1.a, t1.b, t2.b;
a | b | c | a | b | c
----+-----+------+----+-----+------
-1 | 104 | 0004 | -1 | 104 | 0004
-1 | 104 | 0004 | -1 | 204 | 0004
-1 | 109 | 0009 | -1 | 109 | 0009
-1 | 109 | 0009 | -1 | 209 | 0009
-1 | 204 | 0004 | -1 | 104 | 0004
-1 | 204 | 0004 | -1 | 204 | 0004
-1 | 209 | 0009 | -1 | 109 | 0009
-1 | 209 | 0009 | -1 | 209 | 0009
1 | 104 | 0004 | 1 | 104 | 0004
1 | 104 | 0004 | 1 | 204 | 0004
1 | 109 | 0009 | 1 | 109 | 0009
1 | 109 | 0009 | 1 | 209 | 0009
1 | 204 | 0004 | 1 | 104 | 0004
1 | 204 | 0004 | 1 | 204 | 0004
1 | 209 | 0009 | 1 | 109 | 0009
1 | 209 | 0009 | 1 | 209 | 0009
2018-07-06 03:56:12 +02:00
(16 rows)
Allow partitionwise joins in more cases.
Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds. This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit. For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa. In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join. Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables. See comments in partition_bounds_merge().
Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra
Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com
2020-04-08 03:25:00 +02:00
EXPLAIN (COSTS OFF)
SELECT t1.*, t2.* FROM alpha t1 INNER JOIN beta t2 ON (t1.a = t2.a AND t1.b = t2.b AND t1.c = t2.c) WHERE ((t1.b >= 100 AND t1.b < 110) OR (t1.b >= 200 AND t1.b < 210)) AND ((t2.b >= 100 AND t2.b < 110) OR (t2.b >= 200 AND t2.b < 210)) AND t1.c IN ('0004', '0009') ORDER BY t1.a, t1.b;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------
Sort
Sort Key: t1.a, t1.b
-> Append
-> Hash Join
Hash Cond: ((t1_1.a = t2_1.a) AND (t1_1.b = t2_1.b) AND (t1_1.c = t2_1.c))
-> Seq Scan on alpha_neg_p1 t1_1
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Hash
-> Seq Scan on beta_neg_p1 t2_1
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
-> Hash Join
Hash Cond: ((t1_2.a = t2_2.a) AND (t1_2.b = t2_2.b) AND (t1_2.c = t2_2.c))
-> Seq Scan on alpha_neg_p2 t1_2
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Hash
-> Seq Scan on beta_neg_p2 t2_2
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
-> Nested Loop
Join Filter: ((t1_3.a = t2_3.a) AND (t1_3.b = t2_3.b) AND (t1_3.c = t2_3.c))
-> Seq Scan on alpha_pos_p2 t1_3
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Seq Scan on beta_pos_p2 t2_3
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
-> Nested Loop
Join Filter: ((t1_4.a = t2_4.a) AND (t1_4.b = t2_4.b) AND (t1_4.c = t2_4.c))
-> Seq Scan on alpha_pos_p3 t1_4
Filter: ((c = ANY ('{0004,0009}'::text[])) AND (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210))))
-> Seq Scan on beta_pos_p3 t2_4
Filter: (((b >= 100) AND (b < 110)) OR ((b >= 200) AND (b < 210)))
(29 rows)
SELECT t1.*, t2.* FROM alpha t1 INNER JOIN beta t2 ON (t1.a = t2.a AND t1.b = t2.b AND t1.c = t2.c) WHERE ((t1.b >= 100 AND t1.b < 110) OR (t1.b >= 200 AND t1.b < 210)) AND ((t2.b >= 100 AND t2.b < 110) OR (t2.b >= 200 AND t2.b < 210)) AND t1.c IN ('0004', '0009') ORDER BY t1.a, t1.b;
a | b | c | a | b | c
----+-----+------+----+-----+------
-1 | 104 | 0004 | -1 | 104 | 0004
-1 | 109 | 0009 | -1 | 109 | 0009
-1 | 204 | 0004 | -1 | 204 | 0004
-1 | 209 | 0009 | -1 | 209 | 0009
1 | 104 | 0004 | 1 | 104 | 0004
1 | 109 | 0009 | 1 | 109 | 0009
1 | 204 | 0004 | 1 | 204 | 0004
1 | 209 | 0009 | 1 | 209 | 0009
(8 rows)