Here we add a new executor node type named "Result Cache". The planner
can include this node type in the plan to have the executor cache the
results from the inner side of parameterized nested loop joins. This
allows caching of tuples for sets of parameters so that in the event that
the node sees the same parameter values again, it can just return the
cached tuples instead of rescanning the inner side of the join all over
again. Internally, result cache uses a hash table in order to quickly
find tuples that have been previously cached.
For certain data sets, this can significantly improve the performance of
joins. The best cases for using this new node type are for join problems
where a large portion of the tuples from the inner side of the join have
no join partner on the outer side of the join. In such cases, hash join
would have to hash values that are never looked up, thus bloating the hash
table and possibly causing it to multi-batch. Merge joins would have to
skip over all of the unmatched rows. If we use a nested loop join with a
result cache, then we only cache tuples that have at least one join
partner on the outer side of the join. The benefits of using a
parameterized nested loop with a result cache increase when there are
fewer distinct values being looked up and the number of lookups of each
value is large. Also, hash probes to lookup the cache can be much faster
than the hash probe in a hash join as it's common that the result cache's
hash table is much smaller than the hash join's due to result cache only
caching useful tuples rather than all tuples from the inner side of the
join. This variation in hash probe performance is more significant when
the hash join's hash table no longer fits into the CPU's L3 cache, but the
result cache's hash table does. The apparent "random" access of hash
buckets with each hash probe can cause a poor L3 cache hit ratio for large
hash tables. Smaller hash tables generally perform better.
The hash table used for the cache limits itself to not exceeding work_mem
* hash_mem_multiplier in size. We maintain a dlist of keys for this cache
and when we're adding new tuples and realize we've exceeded the memory
budget, we evict cache entries starting with the least recently used ones
until we have enough memory to add the new tuples to the cache.
For parameterized nested loop joins, we now consider using one of these
result cache nodes in between the nested loop node and its inner node. We
determine when this might be useful based on cost, which is primarily
driven off of what the expected cache hit ratio will be. Estimating the
cache hit ratio relies on having good distinct estimates on the nested
loop's parameters.
For now, the planner will only consider using a result cache for
parameterized nested loop joins. This works for both normal joins and
also for LATERAL type joins to subqueries. It is possible to use this new
node for other uses in the future. For example, to cache results from
correlated subqueries. However, that's not done here due to some
difficulties obtaining a distinct estimation on the outer plan to
calculate the estimated cache hit ratio. Currently we plan the inner plan
before planning the outer plan so there is no good way to know if a result
cache would be useful or not since we can't estimate the number of times
the subplan will be called until the outer plan is generated.
The functionality being added here is newly introducing a dependency on
the return value of estimate_num_groups() during the join search.
Previously, during the join search, we only ever needed to perform
selectivity estimations. With this commit, we need to use
estimate_num_groups() in order to estimate what the hit ratio on the
result cache will be. In simple terms, if we expect 10 distinct values
and we expect 1000 outer rows, then we'll estimate the hit ratio to be
99%. Since cache hits are very cheap compared to scanning the underlying
nodes on the inner side of the nested loop join, then this will
significantly reduce the planner's cost for the join. However, it's
fairly easy to see here that things will go bad when estimate_num_groups()
incorrectly returns a value that's significantly lower than the actual
number of distinct values. If this happens then that may cause us to make
use of a nested loop join with a result cache instead of some other join
type, such as a merge or hash join. Our distinct estimations have been
known to be a source of trouble in the past, so the extra reliance on them
here could cause the planner to choose slower plans than it did previous
to having this feature. Distinct estimations are also fairly hard to
estimate accurately when several tables have been joined already or when a
WHERE clause filters out a set of values that are correlated to the
expressions we're estimating the number of distinct value for.
For now, the costing we perform during query planning for result caches
does put quite a bit of faith in the distinct estimations being accurate.
When these are accurate then we should generally see faster execution
times for plans containing a result cache. However, in the real world, we
may find that we need to either change the costings to put less trust in
the distinct estimations being accurate or perhaps even disable this
feature by default. There's always an element of risk when we teach the
query planner to do new tricks that it decides to use that new trick at
the wrong time and causes a regression. Users may opt to get the old
behavior by turning the feature off using the enable_resultcache GUC.
Currently, this is enabled by default. It remains to be seen if we'll
maintain that setting for the release.
Additionally, the name "Result Cache" is the best name I could think of
for this new node at the time I started writing the patch. Nobody seems
to strongly dislike the name. A few people did suggest other names but no
other name seemed to dominate in the brief discussion that there was about
names. Let's allow the beta period to see if the current name pleases
enough people. If there's some consensus on a better name, then we can
change it before the release. Please see the 2nd discussion link below
for the discussion on the "Result Cache" name.
Author: David Rowley
Reviewed-by: Andy Fan, Justin Pryzby, Zhihong Yu, Hou Zhijie
Tested-By: Konstantin Knizhnik
Discussion: https://postgr.es/m/CAApHDvrPcQyQdWERGYWx8J%2B2DLUNgXu%2BfOSbQ1UscxrunyXyrQ%40mail.gmail.com
Discussion: https://postgr.es/m/CAApHDvq=yQXr5kqhRviT2RhNKwToaWr9JAN5t+5_PzhuRJ3wvg@mail.gmail.com
This removes "Add Result Cache executor node". It seems that something
weird is going on with the tracking of cache hits and misses as
highlighted by many buildfarm animals. It's not yet clear what the
problem is as other parts of the plan indicate that the cache did work
correctly, it's just the hits and misses that were being reported as 0.
This is especially a bad time to have the buildfarm so broken, so
reverting before too many more animals go red.
Discussion: https://postgr.es/m/CAApHDvq_hydhfovm4=izgWs+C5HqEeRScjMbOgbpC-jRAeK3Yw@mail.gmail.com
Here we add a new executor node type named "Result Cache". The planner
can include this node type in the plan to have the executor cache the
results from the inner side of parameterized nested loop joins. This
allows caching of tuples for sets of parameters so that in the event that
the node sees the same parameter values again, it can just return the
cached tuples instead of rescanning the inner side of the join all over
again. Internally, result cache uses a hash table in order to quickly
find tuples that have been previously cached.
For certain data sets, this can significantly improve the performance of
joins. The best cases for using this new node type are for join problems
where a large portion of the tuples from the inner side of the join have
no join partner on the outer side of the join. In such cases, hash join
would have to hash values that are never looked up, thus bloating the hash
table and possibly causing it to multi-batch. Merge joins would have to
skip over all of the unmatched rows. If we use a nested loop join with a
result cache, then we only cache tuples that have at least one join
partner on the outer side of the join. The benefits of using a
parameterized nested loop with a result cache increase when there are
fewer distinct values being looked up and the number of lookups of each
value is large. Also, hash probes to lookup the cache can be much faster
than the hash probe in a hash join as it's common that the result cache's
hash table is much smaller than the hash join's due to result cache only
caching useful tuples rather than all tuples from the inner side of the
join. This variation in hash probe performance is more significant when
the hash join's hash table no longer fits into the CPU's L3 cache, but the
result cache's hash table does. The apparent "random" access of hash
buckets with each hash probe can cause a poor L3 cache hit ratio for large
hash tables. Smaller hash tables generally perform better.
The hash table used for the cache limits itself to not exceeding work_mem
* hash_mem_multiplier in size. We maintain a dlist of keys for this cache
and when we're adding new tuples and realize we've exceeded the memory
budget, we evict cache entries starting with the least recently used ones
until we have enough memory to add the new tuples to the cache.
For parameterized nested loop joins, we now consider using one of these
result cache nodes in between the nested loop node and its inner node. We
determine when this might be useful based on cost, which is primarily
driven off of what the expected cache hit ratio will be. Estimating the
cache hit ratio relies on having good distinct estimates on the nested
loop's parameters.
For now, the planner will only consider using a result cache for
parameterized nested loop joins. This works for both normal joins and
also for LATERAL type joins to subqueries. It is possible to use this new
node for other uses in the future. For example, to cache results from
correlated subqueries. However, that's not done here due to some
difficulties obtaining a distinct estimation on the outer plan to
calculate the estimated cache hit ratio. Currently we plan the inner plan
before planning the outer plan so there is no good way to know if a result
cache would be useful or not since we can't estimate the number of times
the subplan will be called until the outer plan is generated.
The functionality being added here is newly introducing a dependency on
the return value of estimate_num_groups() during the join search.
Previously, during the join search, we only ever needed to perform
selectivity estimations. With this commit, we need to use
estimate_num_groups() in order to estimate what the hit ratio on the
result cache will be. In simple terms, if we expect 10 distinct values
and we expect 1000 outer rows, then we'll estimate the hit ratio to be
99%. Since cache hits are very cheap compared to scanning the underlying
nodes on the inner side of the nested loop join, then this will
significantly reduce the planner's cost for the join. However, it's
fairly easy to see here that things will go bad when estimate_num_groups()
incorrectly returns a value that's significantly lower than the actual
number of distinct values. If this happens then that may cause us to make
use of a nested loop join with a result cache instead of some other join
type, such as a merge or hash join. Our distinct estimations have been
known to be a source of trouble in the past, so the extra reliance on them
here could cause the planner to choose slower plans than it did previous
to having this feature. Distinct estimations are also fairly hard to
estimate accurately when several tables have been joined already or when a
WHERE clause filters out a set of values that are correlated to the
expressions we're estimating the number of distinct value for.
For now, the costing we perform during query planning for result caches
does put quite a bit of faith in the distinct estimations being accurate.
When these are accurate then we should generally see faster execution
times for plans containing a result cache. However, in the real world, we
may find that we need to either change the costings to put less trust in
the distinct estimations being accurate or perhaps even disable this
feature by default. There's always an element of risk when we teach the
query planner to do new tricks that it decides to use that new trick at
the wrong time and causes a regression. Users may opt to get the old
behavior by turning the feature off using the enable_resultcache GUC.
Currently, this is enabled by default. It remains to be seen if we'll
maintain that setting for the release.
Additionally, the name "Result Cache" is the best name I could think of
for this new node at the time I started writing the patch. Nobody seems
to strongly dislike the name. A few people did suggest other names but no
other name seemed to dominate in the brief discussion that there was about
names. Let's allow the beta period to see if the current name pleases
enough people. If there's some consensus on a better name, then we can
change it before the release. Please see the 2nd discussion link below
for the discussion on the "Result Cache" name.
Author: David Rowley
Reviewed-by: Andy Fan, Justin Pryzby, Zhihong Yu
Tested-By: Konstantin Knizhnik
Discussion: https://postgr.es/m/CAApHDvrPcQyQdWERGYWx8J%2B2DLUNgXu%2BfOSbQ1UscxrunyXyrQ%40mail.gmail.com
Discussion: https://postgr.es/m/CAApHDvq=yQXr5kqhRviT2RhNKwToaWr9JAN5t+5_PzhuRJ3wvg@mail.gmail.com
This patch makes two closely related sets of changes:
1. For UPDATE, the subplan of the ModifyTable node now only delivers
the new values of the changed columns (i.e., the expressions computed
in the query's SET clause) plus row identity information such as CTID.
ModifyTable must re-fetch the original tuple to merge in the old
values of any unchanged columns. The core advantage of this is that
the changed columns are uniform across all tables of an inherited or
partitioned target relation, whereas the other columns might not be.
A secondary advantage, when the UPDATE involves joins, is that less
data needs to pass through the plan tree. The disadvantage of course
is an extra fetch of each tuple to be updated. However, that seems to
be very nearly free in context; even worst-case tests don't show it to
add more than a couple percent to the total query cost. At some point
it might be interesting to combine the re-fetch with the tuple access
that ModifyTable must do anyway to mark the old tuple dead; but that
would require a good deal of refactoring and it seems it wouldn't buy
all that much, so this patch doesn't attempt it.
2. For inherited UPDATE/DELETE, instead of generating a separate
subplan for each target relation, we now generate a single subplan
that is just exactly like a SELECT's plan, then stick ModifyTable
on top of that. To let ModifyTable know which target relation a
given incoming row refers to, a tableoid junk column is added to
the row identity information. This gets rid of the horrid hack
that was inheritance_planner(), eliminating O(N^2) planning cost
and memory consumption in cases where there were many unprunable
target relations.
Point 2 of course requires point 1, so that there is a uniform
definition of the non-junk columns to be returned by the subplan.
We can't insist on uniform definition of the row identity junk
columns however, if we want to keep the ability to have both
plain and foreign tables in a partitioning hierarchy. Since
it wouldn't scale very far to have every child table have its
own row identity column, this patch includes provisions to merge
similar row identity columns into one column of the subplan result.
In particular, we can merge the whole-row Vars typically used as
row identity by FDWs into one column by pretending they are type
RECORD. (It's still okay for the actual composite Datums to be
labeled with the table's rowtype OID, though.)
There is more that can be done to file down residual inefficiencies
in this patch, but it seems to be committable now.
FDW authors should note several API changes:
* The argument list for AddForeignUpdateTargets() has changed, and so
has the method it must use for adding junk columns to the query. Call
add_row_identity_var() instead of manipulating the parse tree directly.
You might want to reconsider exactly what you're adding, too.
* PlanDirectModify() must now work a little harder to find the
ForeignScan plan node; if the foreign table is part of a partitioning
hierarchy then the ForeignScan might not be the direct child of
ModifyTable. See postgres_fdw for sample code.
* To check whether a relation is a target relation, it's no
longer sufficient to compare its relid to root->parse->resultRelation.
Instead, check it against all_result_relids or leaf_result_relids,
as appropriate.
Amit Langote and Tom Lane
Discussion: https://postgr.es/m/CA+HiwqHpHdqdDn48yCEhynnniahH78rwcrv1rEX65-fsZGBOLQ@mail.gmail.com
This implements asynchronous execution, which runs multiple parts of a
non-parallel-aware Append concurrently rather than serially to improve
performance when possible. Currently, the only node type that can be
run concurrently is a ForeignScan that is an immediate child of such an
Append. In the case where such ForeignScans access data on different
remote servers, this would run those ForeignScans concurrently, and
overlap the remote operations to be performed simultaneously, so it'll
improve the performance especially when the operations involve
time-consuming ones such as remote join and remote aggregation.
We may extend this to other node types such as joins or aggregates over
ForeignScans in the future.
This also adds the support for postgres_fdw, which is enabled by the
table-level/server-level option "async_capable". The default is false.
Robert Haas, Kyotaro Horiguchi, Thomas Munro, and myself. This commit
is mostly based on the patch proposed by Robert Haas, but also uses
stuff from the patch proposed by Kyotaro Horiguchi and from the patch
proposed by Thomas Munro. Reviewed by Kyotaro Horiguchi, Konstantin
Knizhnik, Andrey Lepikhov, Movead Li, Thomas Munro, Justin Pryzby, and
others.
Discussion: https://postgr.es/m/CA%2BTgmoaXQEt4tZ03FtQhnzeDEMzBck%2BLrni0UWHVVgOTnA6C1w%40mail.gmail.com
Discussion: https://postgr.es/m/CA%2BhUKGLBRyu0rHrDCMC4%3DRn3252gogyp1SjOgG8SEKKZv%3DFwfQ%40mail.gmail.com
Discussion: https://postgr.es/m/20200228.170650.667613673625155850.horikyota.ntt%40gmail.com
To allow inserts in parallel-mode this feature has to ensure that all the
constraints, triggers, etc. are parallel-safe for the partition hierarchy
which is costly and we need to find a better way to do that. Additionally,
we could have used existing cached information in some cases like indexes,
domains, etc. to determine the parallel-safety.
List of commits reverted, in reverse chronological order:
ed62d3737c Doc: Update description for parallel insert reloption.
c8f78b6161 Add a new GUC and a reloption to enable inserts in parallel-mode.
c5be48f092 Improve FK trigger parallel-safety check added by 05c8482f7f.
e2cda3c20a Fix use of relcache TriggerDesc field introduced by commit 05c8482f7f.
e4e87a32cc Fix valgrind issue in commit 05c8482f7f.
05c8482f7f Enable parallel SELECT for "INSERT INTO ... SELECT ...".
Discussion: https://postgr.es/m/E1lMiB9-0001c3-SY@gemulon.postgresql.org
Parallel SELECT can't be utilized for INSERT in the following cases:
- INSERT statement uses the ON CONFLICT DO UPDATE clause
- Target table has a parallel-unsafe: trigger, index expression or
predicate, column default expression or check constraint
- Target table has a parallel-unsafe domain constraint on any column
- Target table is a partitioned table with a parallel-unsafe partition key
expression or support function
The planner is updated to perform additional parallel-safety checks for
the cases listed above, for determining whether it is safe to run INSERT
in parallel-mode with an underlying parallel SELECT. The planner will
consider using parallel SELECT for "INSERT INTO ... SELECT ...", provided
nothing unsafe is found from the additional parallel-safety checks, or
from the existing parallel-safety checks for SELECT.
While checking parallel-safety, we need to check it for all the partitions
on the table which can be costly especially when we decide not to use a
parallel plan. So, in a separate patch, we will introduce a GUC and or a
reloption to enable/disable parallelism for Insert statements.
Prior to entering parallel-mode for the execution of INSERT with parallel
SELECT, a TransactionId is acquired and assigned to the current
transaction state. This is necessary to prevent the INSERT from attempting
to assign the TransactionId whilst in parallel-mode, which is not allowed.
This approach has a disadvantage in that if the underlying SELECT does not
return any rows, then the TransactionId is not used, however that
shouldn't happen in practice in many cases.
Author: Greg Nancarrow, Amit Langote, Amit Kapila
Reviewed-by: Amit Langote, Hou Zhijie, Takayuki Tsunakawa, Antonin Houska, Bharath Rupireddy, Dilip Kumar, Vignesh C, Zhihong Yu, Amit Kapila
Tested-by: Tang, Haiying
Discussion: https://postgr.es/m/CAJcOf-cXnB5cnMKqWEp2E2z7Mvcd04iLVmV=qpFJrR3AcrTS3g@mail.gmail.com
Discussion: https://postgr.es/m/CAJcOf-fAdj=nDKMsRhQzndm-O13NY4dL6xGcEvdX5Xvbbi0V7g@mail.gmail.com
This adds a new executor node named TID Range Scan. The query planner
will generate paths for TID Range scans when quals are discovered on base
relations which search for ranges on the table's ctid column. These
ranges may be open at either end. For example, WHERE ctid >= '(10,0)';
will return all tuples on page 10 and over.
To support this, two new optional callback functions have been added to
table AM. scan_set_tidrange is used to set the scan range to just the
given range of TIDs. scan_getnextslot_tidrange fetches the next tuple
in the given range.
For AMs were scanning ranges of TIDs would not make sense, these functions
can be set to NULL in the TableAmRoutine. The query planner won't
generate TID Range Scan Paths in that case.
Author: Edmund Horner, David Rowley
Reviewed-by: David Rowley, Tomas Vondra, Tom Lane, Andres Freund, Zhihong Yu
Discussion: https://postgr.es/m/CAMyN-kB-nFTkF=VA_JPwFNo08S0d-Yk0F741S2B7LDmYAi8eyA@mail.gmail.com
FDWs that can perform an UPDATE/DELETE remotely using the "direct modify"
set of APIs need to access the ResultRelInfo of the target table. That's
currently available in EState.es_result_relation_info, but the next
commit will remove that field.
This commit adds a new resultRelation field in ForeignScan, to store the
target relation's RT index, and the corresponding ResultRelInfo in
ForeignScanState. The FDW's PlanDirectModify callback is expected to set
'resultRelation' along with 'operation'. The core code doesn't need them
for anything, they are for the convenience of FDW's Begin- and
IterateDirectModify callbacks.
Authors: Amit Langote, Etsuro Fujita
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqGEmiib8FLiHMhKB%2BCH5dRgHSLc5N5wnvc4kym%2BZYpQEQ%40mail.gmail.com
Instead of allocating all the ResultRelInfos upfront in one big array,
allocate them in ExecInitModifyTable(). es_result_relations is now an
array of ResultRelInfo pointers, rather than an array of structs, and it
is indexed by the RT index.
This simplifies things: we get rid of the separate concept of a "result
rel index", and don't need to set it in setrefs.c anymore. This also
allows follow-up optimizations (not included in this commit yet) to skip
initializing ResultRelInfos for target relations that were not needed at
runtime, and removal of the es_result_relation_info pointer.
The EState arrays of regular result rels and root result rels are merged
into one array. Similarly, the resultRelations and rootResultRelations
lists in PlannedStmt are merged into one. It's not actually clear to me
why they were kept separate in the first place, but now that the
es_result_relations array is indexed by RT index, it certainly seems
pointless.
The PlannedStmt->resultRelations list is now only needed for
ExecRelationIsTargetRelation(). One visible effect of this change is that
ExecRelationIsTargetRelation() will now return 'true' also for the
partition root, if a partitioned table is updated. That seems like a good
thing, although the function isn't used in core code, and I don't see any
reason for an FDW to call it on a partition root.
Author: Amit Langote
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqGEmiib8FLiHMhKB%2BCH5dRgHSLc5N5wnvc4kym%2BZYpQEQ%40mail.gmail.com
Includes some manual cleanup of places that pgindent messed up,
most of which weren't per project style anyway.
Notably, it seems some people didn't absorb the style rules of
commit c9d297751, because there were a bunch of new occurrences
of function calls with a newline just after the left paren, all
with faulty expectations about how the rest of the call would get
indented.
WITH TIES is an option to the FETCH FIRST N ROWS clause (the SQL
standard's spelling of LIMIT), where you additionally get rows that
compare equal to the last of those N rows by the columns in the
mandatory ORDER BY clause.
There was a proposal by Andrew Gierth to implement this functionality in
a more powerful way that would yield more features, but the other patch
had not been finished at this time, so we decided to use this one for
now in the spirit of incremental development.
Author: Surafel Temesgen <surafel3000@gmail.com>
Reviewed-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Discussion: https://postgr.es/m/CALAY4q9ky7rD_A4vf=FVQvCGngm3LOes-ky0J6euMrg=_Se+ag@mail.gmail.com
Discussion: https://postgr.es/m/87o8wvz253.fsf@news-spur.riddles.org.uk
Incremental Sort is an optimized variant of multikey sort for cases when
the input is already sorted by a prefix of the requested sort keys. For
example when the relation is already sorted by (key1, key2) and we need
to sort it by (key1, key2, key3) we can simply split the input rows into
groups having equal values in (key1, key2), and only sort/compare the
remaining column key3.
This has a number of benefits:
- Reduced memory consumption, because only a single group (determined by
values in the sorted prefix) needs to be kept in memory. This may also
eliminate the need to spill to disk.
- Lower startup cost, because Incremental Sort produce results after each
prefix group, which is beneficial for plans where startup cost matters
(like for example queries with LIMIT clause).
We consider both Sort and Incremental Sort, and decide based on costing.
The implemented algorithm operates in two different modes:
- Fetching a minimum number of tuples without check of equality on the
prefix keys, and sorting on all columns when safe.
- Fetching all tuples for a single prefix group and then sorting by
comparing only the remaining (non-prefix) keys.
We always start in the first mode, and employ a heuristic to switch into
the second mode if we believe it's beneficial - the goal is to minimize
the number of unnecessary comparions while keeping memory consumption
below work_mem.
This is a very old patch series. The idea was originally proposed by
Alexander Korotkov back in 2013, and then revived in 2017. In 2018 the
patch was taken over by James Coleman, who wrote and rewrote most of the
current code.
There were many reviewers/contributors since 2013 - I've done my best to
pick the most active ones, and listed them in this commit message.
Author: James Coleman, Alexander Korotkov
Reviewed-by: Tomas Vondra, Andreas Karlsson, Marti Raudsepp, Peter Geoghegan, Robert Haas, Thomas Munro, Antonin Houska, Andres Freund, Alexander Kuzmenkov
Discussion: https://postgr.es/m/CAPpHfdscOX5an71nHd8WSUH6GNOCf=V7wgDaTXdDd9=goN-gfA@mail.gmail.com
Discussion: https://postgr.es/m/CAPpHfds1waRZ=NOmueYq0sx1ZSCnt+5QJvizT8ndT2=etZEeAQ@mail.gmail.com
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
In 5f32b29c18 I changed the creation of HashState.hashkeys to
actually use HashState as the parent (instead of HashJoinState, which
was incorrect, as they were executed below HashState), to fix the
problem of hashkeys expressions otherwise relying on slot types
appropriate for HashJoinState, rather than HashState as would be
correct. That reliance was only introduced in 12, which is why it
previously worked to use HashJoinState as the parent (although I'd be
unsurprised if there were problematic cases).
Unfortunately that's not a sufficient solution, because before this
commit, the to-be-hashed expressions referenced inner/outer as
appropriate for the HashJoin, not Hash. That didn't have obvious bad
consequences, because the slots containing the tuples were put into
ecxt_innertuple when hashing a tuple for HashState (even though Hash
doesn't have an inner plan).
There are less common cases where this can cause visible problems
however (rather than just confusion when inspecting such executor
trees). E.g. "ERROR: bogus varno: 65000", when explaining queries
containing a HashJoin where the subsidiary Hash node's hash keys
reference a subplan. While normally hashkeys aren't displayed by
EXPLAIN, if one of those expressions references a subplan, that
subplan may be printed as part of the Hash node - which then failed
because an inner plan was referenced, and Hash doesn't have that.
It seems quite possible that there's other broken cases, too.
Fix the problem by properly splitting the expression for the HashJoin
and Hash nodes at plan time, and have them reference the proper
subsidiary node. While other workarounds are possible, fixing this
correctly seems easy enough. It was a pretty ugly hack to have
ExecInitHashJoin put the expression into the already initialized
HashState, in the first place.
I decided to not just split inner/outer hashkeys inside
make_hashjoin(), but also to separate out hashoperators and
hashcollations at plan time. Otherwise we would have ended up having
two very similar loops, one at plan time and the other during executor
startup. The work seems to more appropriately belong to plan time,
anyway.
Reported-By: Nikita Glukhov, Alexander Korotkov
Author: Andres Freund
Reviewed-By: Tom Lane, in an earlier version
Discussion: https://postgr.es/m/CAPpHfdvGVegF_TKKRiBrSmatJL2dR9uwFCuR+teQ_8tEXU8mxg@mail.gmail.com
Backpatch: 12-
This is still using the 2.0 version of pg_bsd_indent.
I thought it would be good to commit this separately,
so as to document the differences between 2.0 and 2.1 behavior.
Discussion: https://postgr.es/m/16296.1558103386@sss.pgh.pa.us
Previously, gen_partprune_steps() always built executor pruning steps
using all suitable clauses, including those containing PARAM_EXEC
Params. This meant that the pruning steps were only completely safe
for executor run-time (scan start) pruning. To prune at executor
startup, we had to ignore the steps involving exec Params. But this
doesn't really work in general, since there may be logic changes
needed as well --- for example, pruning according to the last operator's
btree strategy is the wrong thing if we're not applying that operator.
The rules embodied in gen_partprune_steps() and its minions are
sufficiently complicated that tracking their incremental effects in
other logic seems quite impractical.
Short of a complete redesign, the only safe fix seems to be to run
gen_partprune_steps() twice, once to create executor startup pruning
steps and then again for run-time pruning steps. We can save a few
cycles however by noting during the first scan whether we rejected
any clauses because they involved exec Params --- if not, we don't
need to do the second scan.
In support of this, refactor the internal APIs in partprune.c to make
more use of passing information in the GeneratePruningStepsContext
struct, rather than as separate arguments.
This is, I hope, the last piece of our response to a bug report from
Alan Jackson. Back-patch to v11 where this code came in.
Discussion: https://postgr.es/m/FAD28A83-AC73-489E-A058-2681FA31D648@tvsquared.com
Previously, the planner created RangeTblEntry and RelOptInfo structs
for every partition of a partitioned table, even though many of them
might later be deemed uninteresting thanks to partition pruning logic.
This incurred significant overhead when there are many partitions.
Arrange to postpone creation of these data structures until after
we've processed the query enough to identify restriction quals for
the partitioned table, and then apply partition pruning before not
after creation of each partition's data structures. In this way
we need not open the partition relations at all for partitions that
the planner has no real interest in.
For queries that can be proven at plan time to access only a small
number of partitions, this patch improves the practical maximum
number of partitions from under 100 to perhaps a few thousand.
Amit Langote, reviewed at various times by Dilip Kumar, Jesper Pedersen,
Yoshikazu Imai, and David Rowley
Discussion: https://postgr.es/m/9d7c5112-cb99-6a47-d3be-cf1ee6862a1d@lab.ntt.co.jp
This adds a flag "deterministic" to collations. If that is false,
such a collation disables various optimizations that assume that
strings are equal only if they are byte-wise equal. That then allows
use cases such as case-insensitive or accent-insensitive comparisons
or handling of strings with different Unicode normal forms.
This functionality is only supported with the ICU provider. At least
glibc doesn't appear to have any locales that work in a
nondeterministic way, so it's not worth supporting this for the libc
provider.
The term "deterministic comparison" in this context is from Unicode
Technical Standard #10
(https://unicode.org/reports/tr10/#Deterministic_Comparison).
This patch makes changes in three areas:
- CREATE COLLATION DDL changes and system catalog changes to support
this new flag.
- Many executor nodes and auxiliary code are extended to track
collations. Previously, this code would just throw away collation
information, because the eventually-called user-defined functions
didn't use it since they only cared about equality, which didn't
need collation information.
- String data type functions that do equality comparisons and hashing
are changed to take the (non-)deterministic flag into account. For
comparison, this just means skipping various shortcuts and tie
breakers that use byte-wise comparison. For hashing, we first need
to convert the input string to a canonical "sort key" using the ICU
analogue of strxfrm().
Reviewed-by: Daniel Verite <daniel@manitou-mail.org>
Reviewed-by: Peter Geoghegan <pg@bowt.ie>
Discussion: https://www.postgresql.org/message-id/flat/1ccc668f-4cbc-0bef-af67-450b47cdfee7@2ndquadrant.com
We still require AccessExclusiveLock on the partition itself, because
otherwise an insert that violates the newly-imposed partition
constraint could be in progress at the same time that we're changing
that constraint; only the lock level on the parent relation is
weakened.
To make this safe, we have to cope with (at least) three separate
problems. First, relevant DDL might commit while we're in the process
of building a PartitionDesc. If so, find_inheritance_children() might
see a new partition while the RELOID system cache still has the old
partition bound cached, and even before invalidation messages have
been queued. To fix that, if we see that the pg_class tuple seems to
be missing or to have a null relpartbound, refetch the value directly
from the table. We can't get the wrong value, because DETACH PARTITION
still requires AccessExclusiveLock throughout; if we ever want to
change that, this will need more thought. In testing, I found it quite
difficult to hit even the null-relpartbound case; the race condition
is extremely tight, but the theoretical risk is there.
Second, successive calls to RelationGetPartitionDesc might not return
the same answer. The query planner will get confused if lookup up the
PartitionDesc for a particular relation does not return a consistent
answer for the entire duration of query planning. Likewise, query
execution will get confused if the same relation seems to have a
different PartitionDesc at different times. Invent a new
PartitionDirectory concept and use it to ensure consistency. This
ensures that a single invocation of either the planner or the executor
sees the same view of the PartitionDesc from beginning to end, but it
does not guarantee that the planner and the executor see the same
view. Since this allows pointers to old PartitionDesc entries to
survive even after a relcache rebuild, also postpone removing the old
PartitionDesc entry until we're certain no one is using it.
For the most part, it seems to be OK for the planner and executor to
have different views of the PartitionDesc, because the executor will
just ignore any concurrently added partitions which were unknown at
plan time; those partitions won't be part of the inheritance
expansion, but invalidation messages will trigger replanning at some
point. Normally, this happens by the time the very next command is
executed, but if the next command acquires no locks and executes a
prepared query, it can manage not to notice until a new transaction is
started. We might want to tighten that up, but it's material for a
separate patch. There would still be a small window where a query
that started just after an ATTACH PARTITION command committed might
fail to notice its results -- but only if the command starts before
the commit has been acknowledged to the user. All in all, the warts
here around serializability seem small enough to be worth accepting
for the considerable advantage of being able to add partitions without
a full table lock.
Although in general the consequences of new partitions showing up
between planning and execution are limited to the query not noticing
the new partitions, run-time partition pruning will get confused in
that case, so that's the third problem that this patch fixes.
Run-time partition pruning assumes that indexes into the PartitionDesc
are stable between planning and execution. So, add code so that if
new partitions are added between plan time and execution time, the
indexes stored in the subplan_map[] and subpart_map[] arrays within
the plan's PartitionedRelPruneInfo get adjusted accordingly. There
does not seem to be a simple way to generalize this scheme to cope
with partitions that are removed, mostly because they could then get
added back again with different bounds, but it works OK for added
partitions.
This code does not try to ensure that every backend participating in
a parallel query sees the same view of the PartitionDesc. That
currently doesn't matter, because we never pass PartitionDesc
indexes between backends. Each backend will ignore the concurrently
added partitions which it notices, and it doesn't matter if different
backends are ignoring different sets of concurrently added partitions.
If in the future that matters, for example because we allow writes in
parallel query and want all participants to do tuple routing to the same
set of partitions, the PartitionDirectory concept could be improved to
share PartitionDescs across backends. There is a draft patch to
serialize and restore PartitionDescs on the thread where this patch
was discussed, which may be a useful place to start.
Patch by me. Thanks to Alvaro Herrera, David Rowley, Simon Riggs,
Amit Langote, and Michael Paquier for discussion, and to Alvaro
Herrera for some review.
Discussion: http://postgr.es/m/CA+Tgmobt2upbSocvvDej3yzokd7AkiT+PvgFH+a9-5VV1oJNSQ@mail.gmail.com
Discussion: http://postgr.es/m/CA+TgmoZE0r9-cyA-aY6f8WFEROaDLLL7Vf81kZ8MtFCkxpeQSw@mail.gmail.com
Discussion: http://postgr.es/m/CA+TgmoY13KQZF-=HNTrt9UYWYx3_oYOQpu9ioNT49jGgiDpUEA@mail.gmail.com
Up to now we've not worried much about joins where the join key is a
relation's CTID column, reasoning that storing a table's CTIDs in some
other table would be pretty useless. However, there are use-cases for
this sort of query involving self-joins, so that argument doesn't really
hold water.
This patch allows generating plans for joins on CTID that use a nestloop
with inner TidScan, similar to what we might do with an index on the join
column. This is the most efficient way to join when the outer side of
the nestloop is expected to yield relatively few rows.
This change requires upgrading tidpath.c and the generated TidPaths
to work with RestrictInfos instead of bare qual clauses, but that's
long-postponed technical debt anyway.
Discussion: https://postgr.es/m/17443.1545435266@sss.pgh.pa.us
The comments above the PartitionedRelPruneInfo struct incorrectly
document how subplan_map and subpart_map are indexed. This seems to
have snuck in on 4e23236403.
Author: David Rowley <david.rowley@2ndquadrant.com>
In the wake of commit f2343653f, we no longer need some fields that
were used before to control executor lock acquisitions:
* PlannedStmt.nonleafResultRelations can go away entirely.
* partitioned_rels can go away from Append, MergeAppend, and ModifyTable.
However, ModifyTable still needs to know the RT index of the partition
root table if any, which was formerly kept in the first entry of that
list. Add a new field "rootRelation" to remember that. rootRelation is
partly redundant with nominalRelation, in that if it's set it will have
the same value as nominalRelation. However, the latter field has a
different purpose so it seems best to keep them distinct.
Amit Langote, reviewed by David Rowley and Jesper Pedersen,
and whacked around a bit more by me
Discussion: https://postgr.es/m/468c85d9-540e-66a2-1dde-fec2b741e688@lab.ntt.co.jp
Create an array estate->es_relations[] paralleling the es_range_table,
and store references to Relations (relcache entries) there, so that any
given RT entry is opened and closed just once per executor run. Scan
nodes typically still call ExecOpenScanRelation, but ExecCloseScanRelation
is no more; relation closing is now done centrally in ExecEndPlan.
This is slightly more complex than one would expect because of the
interactions with relcache references held in ResultRelInfo nodes.
The general convention is now that ResultRelInfo->ri_RelationDesc does
not represent a separate relcache reference and so does not need to be
explicitly closed; but there is an exception for ResultRelInfos in the
es_trig_target_relations list, which are manufactured by
ExecGetTriggerResultRel and have to be cleaned up by
ExecCleanUpTriggerState. (That much was true all along, but these
ResultRelInfos are now more different from others than they used to be.)
To allow the partition pruning logic to make use of es_relations[] rather
than having its own relcache references, adjust PartitionedRelPruneInfo
to store an RT index rather than a relation OID.
Amit Langote, reviewed by David Rowley and Jesper Pedersen,
some mods by me
Discussion: https://postgr.es/m/468c85d9-540e-66a2-1dde-fec2b741e688@lab.ntt.co.jp
The previous coding here supposed that if run-time partitioning applied to
a particular Append/MergeAppend plan, then all child plans of that node
must be members of a single partitioning hierarchy. This is totally wrong,
since an Append could be formed from a UNION ALL: we could have multiple
hierarchies sharing the same Append, or child plans that aren't part of any
hierarchy.
To fix, restructure the related plan-time and execution-time data
structures so that we can have a separate list or array for each
partitioning hierarchy. Also track subplans that are not part of any
hierarchy, and make sure they don't get pruned.
Per reports from Phil Florent and others. Back-patch to v11, since
the bug originated there.
David Rowley, with a lot of cosmetic adjustments by me; thanks also
to Amit Langote for review.
Discussion: https://postgr.es/m/HE1PR03MB17068BB27404C90B5B788BCABA7B0@HE1PR03MB1706.eurprd03.prod.outlook.com
These struct definitions were originally dropped into primnodes.h,
which is a poor choice since that's mainly intended for primitive
expression node types; these are not in that category. What they
are is auxiliary info in Plan trees, so move them to plannodes.h.
For consistency, also relocate some related code that was apparently
placed with the aid of a dartboard.
There's no interesting code changes in this commit, just reshuffling.
David Rowley and Tom Lane
Discussion: https://postgr.es/m/CAFj8pRBjrufA3ocDm8o4LPGNye9Y+pm1b9kCwode4X04CULG3g@mail.gmail.com
This reverts commits d204ef6377,
83454e3c2b and a few more commits thereafter
(complete list at the end) related to MERGE feature.
While the feature was fully functional, with sufficient test coverage and
necessary documentation, it was felt that some parts of the executor and
parse-analyzer can use a different design and it wasn't possible to do that in
the available time. So it was decided to revert the patch for PG11 and retry
again in the future.
Thanks again to all reviewers and bug reporters.
List of commits reverted, in reverse chronological order:
f1464c5380 Improve parse representation for MERGE
ddb4158579 MERGE syntax diagram correction
530e69e59b Allow cpluspluscheck to pass by renaming variable
01b88b4df5 MERGE minor errata
3af7b2b0d4 MERGE fix variable warning in non-assert builds
a5d86181ec MERGE INSERT allows only one VALUES clause
4b2d44031f MERGE post-commit review
4923550c20 Tab completion for MERGE
aa3faa3c7a WITH support in MERGE
83454e3c2b New files for MERGE
d204ef6377 MERGE SQL Command following SQL:2016
Author: Pavan Deolasee
Reviewed-by: Michael Paquier
In 499be013de support for pruning unneeded Append subnodes was added.
The logic in that commit was not correctly checking if the next subplan
was in fact a valid subplan. This could cause parallel workers processes
to be given a subplan to work on which didn't require any work.
Per code review following an otherwise unexplained regression failure in
buildfarm member Pademelon. (We haven't been able to reproduce the
failure, so this is a bit of a blind fix in terms of whether it'll
actually fix it; but it is a clear bug nonetheless).
In passing, also add a comment to explain what first_partial_plan means.
Author: David Rowley
Discussion: https://postgr.es/m/CAKJS1f_E5r05hHUVG3UmCQJ49DGKKHtN=SHybD44LdzBn+CJng@mail.gmail.com
Existing partition pruning is only able to work at plan time, for query
quals that appear in the parsed query. This is good but limiting, as
there can be parameters that appear later that can be usefully used to
further prune partitions.
This commit adds support for pruning subnodes of Append which cannot
possibly contain any matching tuples, during execution, by evaluating
Params to determine the minimum set of subnodes that can possibly match.
We support more than just simple Params in WHERE clauses. Support
additionally includes:
1. Parameterized Nested Loop Joins: The parameter from the outer side of the
join can be used to determine the minimum set of inner side partitions to
scan.
2. Initplans: Once an initplan has been executed we can then determine which
partitions match the value from the initplan.
Partition pruning is performed in two ways. When Params external to the plan
are found to match the partition key we attempt to prune away unneeded Append
subplans during the initialization of the executor. This allows us to bypass
the initialization of non-matching subplans meaning they won't appear in the
EXPLAIN or EXPLAIN ANALYZE output.
For parameters whose value is only known during the actual execution
then the pruning of these subplans must wait. Subplans which are
eliminated during this stage of pruning are still visible in the EXPLAIN
output. In order to determine if pruning has actually taken place, the
EXPLAIN ANALYZE must be viewed. If a certain Append subplan was never
executed due to the elimination of the partition then the execution
timing area will state "(never executed)". Whereas, if, for example in
the case of parameterized nested loops, the number of loops stated in
the EXPLAIN ANALYZE output for certain subplans may appear lower than
others due to the subplan having been scanned fewer times. This is due
to the list of matching subnodes having to be evaluated whenever a
parameter which was found to match the partition key changes.
This commit required some additional infrastructure that permits the
building of a data structure which is able to perform the translation of
the matching partition IDs, as returned by get_matching_partitions, into
the list index of a subpaths list, as exist in node types such as
Append, MergeAppend and ModifyTable. This allows us to translate a list
of clauses into a Bitmapset of all the subpath indexes which must be
included to satisfy the clause list.
Author: David Rowley, based on an earlier effort by Beena Emerson
Reviewers: Amit Langote, Robert Haas, Amul Sul, Rajkumar Raghuwanshi,
Jesper Pedersen
Discussion: https://postgr.es/m/CAOG9ApE16ac-_VVZVvv0gePSgkg_BwYEV1NBqZFqDR2bBE0X0A@mail.gmail.com
MERGE performs actions that modify rows in the target table
using a source table or query. MERGE provides a single SQL
statement that can conditionally INSERT/UPDATE/DELETE rows
a task that would other require multiple PL statements.
e.g.
MERGE INTO target AS t
USING source AS s
ON t.tid = s.sid
WHEN MATCHED AND t.balance > s.delta THEN
UPDATE SET balance = t.balance - s.delta
WHEN MATCHED THEN
DELETE
WHEN NOT MATCHED AND s.delta > 0 THEN
INSERT VALUES (s.sid, s.delta)
WHEN NOT MATCHED THEN
DO NOTHING;
MERGE works with regular and partitioned tables, including
column and row security enforcement, as well as support for
row, statement and transition triggers.
MERGE is optimized for OLTP and is parameterizable, though
also useful for large scale ETL/ELT. MERGE is not intended
to be used in preference to existing single SQL commands
for INSERT, UPDATE or DELETE since there is some overhead.
MERGE can be used statically from PL/pgSQL.
MERGE does not yet support inheritance, write rules,
RETURNING clauses, updatable views or foreign tables.
MERGE follows SQL Standard per the most recent SQL:2016.
Includes full tests and documentation, including full
isolation tests to demonstrate the concurrent behavior.
This version written from scratch in 2017 by Simon Riggs,
using docs and tests originally written in 2009. Later work
from Pavan Deolasee has been both complex and deep, leaving
the lead author credit now in his hands.
Extensive discussion of concurrency from Peter Geoghegan,
with thanks for the time and effort contributed.
Various issues reported via sqlsmith by Andreas Seltenreich
Authors: Pavan Deolasee, Simon Riggs
Reviewer: Peter Geoghegan, Amit Langote, Tomas Vondra, Simon Riggs
Discussion:
https://postgr.es/m/CANP8+jKitBSrB7oTgT9CY2i1ObfOt36z0XMraQc+Xrz8QB0nXA@mail.gmail.comhttps://postgr.es/m/CAH2-WzkJdBuxj9PO=2QaO9-3h3xGbQPZ34kJH=HukRekwM-GZg@mail.gmail.com
This adds simple cost based plan time decision about whether JIT
should be performed. jit_above_cost, jit_optimize_above_cost are
compared with the total cost of a plan, and if the cost is above them
JIT is performed / optimization is performed respectively.
For that PlannedStmt and EState have a jitFlags (es_jit_flags) field
that stores information about what JIT operations should be performed.
EState now also has a new es_jit field, which can store a
JitContext. When there are no errors the context is released in
standard_ExecutorEnd().
It is likely that the default values for jit_[optimize_]above_cost
will need to be adapted further, but in my test these values seem to
work reasonably.
Author: Andres Freund, with feedback by Peter Eisentraut
Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
This patch adds the ability to use "RANGE offset PRECEDING/FOLLOWING"
frame boundaries in window functions. We'd punted on that back in the
original patch to add window functions, because it was not clear how to
do it in a reasonably data-type-extensible fashion. That problem is
resolved here by adding the ability for btree operator classes to provide
an "in_range" support function that defines how to add or subtract the
RANGE offset value. Factoring it this way also allows the operator class
to avoid overflow problems near the ends of the datatype's range, if it
wishes to expend effort on that. (In the committed patch, the integer
opclasses handle that issue, but it did not seem worth the trouble to
avoid overflow failures for datetime types.)
The patch includes in_range support for the integer_ops opfamily
(int2/int4/int8) as well as the standard datetime types. Support for
other numeric types has been requested, but that seems like suitable
material for a follow-on patch.
In addition, the patch adds GROUPS mode which counts the offset in
ORDER-BY peer groups rather than rows, and it adds the frame_exclusion
options specified by SQL:2011. As far as I can see, we are now fully
up to spec on window framing options.
Existing behaviors remain unchanged, except that I changed the errcode
for a couple of existing error reports to meet the SQL spec's expectation
that negative "offset" values should be reported as SQLSTATE 22013.
Internally and in relevant parts of the documentation, we now consistently
use the terminology "offset PRECEDING/FOLLOWING" rather than "value
PRECEDING/FOLLOWING", since the term "value" is confusingly vague.
Oliver Ford, reviewed and whacked around some by me
Discussion: https://postgr.es/m/CAGMVOdu9sivPAxbNN0X+q19Sfv9edEPv=HibOJhB14TJv_RCQg@mail.gmail.com
When an UPDATE causes a row to no longer match the partition
constraint, try to move it to a different partition where it does
match the partition constraint. In essence, the UPDATE is split into
a DELETE from the old partition and an INSERT into the new one. This
can lead to surprising behavior in concurrency scenarios because
EvalPlanQual rechecks won't work as they normally did; the known
problems are documented. (There is a pending patch to improve the
situation further, but it needs more review.)
Amit Khandekar, reviewed and tested by Amit Langote, David Rowley,
Rajkumar Raghuwanshi, Dilip Kumar, Amul Sul, Thomas Munro, Álvaro
Herrera, Amit Kapila, and me. A few final revisions by me.
Discussion: http://postgr.es/m/CAJ3gD9do9o2ccQ7j7+tSgiE1REY65XRiMb=yJO3u3QhyP8EEPQ@mail.gmail.com
Introduce parallel-aware hash joins that appear in EXPLAIN plans as Parallel
Hash Join with Parallel Hash. While hash joins could already appear in
parallel queries, they were previously always parallel-oblivious and had a
partial subplan only on the outer side, meaning that the work of the inner
subplan was duplicated in every worker.
After this commit, the planner will consider using a partial subplan on the
inner side too, using the Parallel Hash node to divide the work over the
available CPU cores and combine its results in shared memory. If the join
needs to be split into multiple batches in order to respect work_mem, then
workers process different batches as much as possible and then work together
on the remaining batches.
The advantages of a parallel-aware hash join over a parallel-oblivious hash
join used in a parallel query are that it:
* avoids wasting memory on duplicated hash tables
* avoids wasting disk space on duplicated batch files
* divides the work of building the hash table over the CPUs
One disadvantage is that there is some communication between the participating
CPUs which might outweigh the benefits of parallelism in the case of small
hash tables. This is avoided by the planner's existing reluctance to supply
partial plans for small scans, but it may be necessary to estimate
synchronization costs in future if that situation changes. Another is that
outer batch 0 must be written to disk if multiple batches are required.
A potential future advantage of parallel-aware hash joins is that right and
full outer joins could be supported, since there is a single set of matched
bits for each hashtable, but that is not yet implemented.
A new GUC enable_parallel_hash is defined to control the feature, defaulting
to on.
Author: Thomas Munro
Reviewed-By: Andres Freund, Robert Haas
Tested-By: Rafia Sabih, Prabhat Sahu
Discussion:
https://postgr.es/m/CAEepm=2W=cOkiZxcg6qiFQP-dHUe09aqTrEMM7yJDrHMhDv_RA@mail.gmail.comhttps://postgr.es/m/CAEepm=37HKyJ4U6XOLi=JgfSHM3o6B-GaeO-6hkOmneTDkH+Uw@mail.gmail.com
When we create an Append node, we can spread out the workers over the
subplans instead of piling on to each subplan one at a time, which
should typically be a bit more efficient, both because the startup
cost of any plan executed entirely by one worker is paid only once and
also because of reduced contention. We can also construct Append
plans using a mix of partial and non-partial subplans, which may allow
for parallelism in places that otherwise couldn't support it.
Unfortunately, this patch doesn't handle the important case of
parallelizing UNION ALL by running each branch in a separate worker;
the executor infrastructure is added here, but more planner work is
needed.
Amit Khandekar, Robert Haas, Amul Sul, reviewed and tested by
Ashutosh Bapat, Amit Langote, Rafia Sabih, Amit Kapila, and
Rajkumar Raghuwanshi.
Discussion: http://postgr.es/m/CAJ3gD9dy0K_E8r727heqXoBmWZ83HwLFwdcaSSmBQ1+S+vRuUQ@mail.gmail.com
If a PARAM_EXEC parameter is used below a Gather (Merge) but the InitPlan
that computes it is attached to or above the Gather (Merge), force the
value to be computed before starting parallelism and pass it down to all
workers. This allows us to use parallelism in cases where it previously
would have had to be rejected as unsafe. We do - in this case - lose the
optimization that the value is only computed if it's actually used. An
alternative strategy would be to have the first worker that needs the value
compute it, but one downside of that approach is that we'd then need to
select a parallel-safe path to compute the parameter value; it couldn't for
example contain a Gather (Merge) node. At some point in the future, we
might want to consider both approaches.
Independent of that consideration, there is a great deal more work that
could be done to make more kinds of PARAM_EXEC parameters parallel-safe.
This infrastructure could be used to allow a Gather (Merge) on the inner
side of a nested loop (although that's not a very appealing plan) and
cases where the InitPlan is attached below the Gather (Merge) could be
addressed as well using various techniques. But this is a good start.
Amit Kapila, reviewed and revised by me. Reviewing and testing from
Kuntal Ghosh, Haribabu Kommi, and Tushar Ahuja.
Discussion: http://postgr.es/m/CAA4eK1LV0Y1AUV4cUCdC+sYOx0Z0-8NAJ2Pd9=UKsbQ5Sr7+JQ@mail.gmail.com
Up until now, we only tracked the number of parameters, which was
sufficient to allocate an array of Datums of the appropriate size,
but not sufficient to, for example, know how to serialize a Datum
stored in one of those slots. An upcoming patch wants to do that,
so add this tracking to make it possible.
Patch by me, reviewed by Tom Lane and Amit Kapila.
Discussion: http://postgr.es/m/CA+TgmoYqpxDKn8koHdW8BEKk8FMUL0=e8m2Qe=M+r0UBjr3tuQ@mail.gmail.com
This takes advantage of the infrastructure introduced by commit
81c5e46c49 to greatly reduce the
likelihood that two different queries will end up with the same query
ID. It's still possible, of course, but whereas before it the chances
of a collision reached 25% around 50,000 queries, it will now take
more than 3 billion queries.
Backward incompatibility: Because the type exposed at the SQL level is
int8, users may now see negative query IDs in the pg_stat_statements
view (and also, query IDs more than 4 billion, which was the old
limit).
Patch by me, reviewed by Michael Paquier and Peter Geoghegan.
Discussion: http://postgr.es/m/CA+TgmobG_Kp4cBKFmsznUAaM1GWW6hhRNiZC0KjRMOOeYnz5Yw@mail.gmail.com
The ExecReScan machinery contains various optimizations for postponing
or skipping rescans of plan subtrees; for example a HashAgg node may
conclude that it can re-use the table it built before, instead of
re-reading its input subtree. But that is wrong if the input contains
a parallel-aware table scan node, since the portion of the table scanned
by the leader process is likely to vary from one rescan to the next.
This explains the timing-dependent buildfarm failures we saw after
commit a2b70c89c.
The established mechanism for showing that a plan node's output is
potentially variable is to mark it as depending on some runtime Param.
Hence, to fix this, invent a dummy Param (one that has a PARAM_EXEC
parameter number, but carries no actual value) associated with each Gather
or GatherMerge node, mark parallel-aware nodes below that node as dependent
on that Param, and arrange for ExecReScanGather[Merge] to flag that Param
as changed whenever the Gather[Merge] node is rescanned.
This solution breaks an undocumented assumption made by the parallel
executor logic, namely that all rescans of nodes below a Gather[Merge]
will happen synchronously during the ReScan of the top node itself.
But that's fundamentally contrary to the design of the ExecReScan code,
and so was doomed to fail someday anyway (even if you want to argue
that the bug being fixed here wasn't a failure of that assumption).
A follow-on patch will address that issue. In the meantime, the worst
that's expected to happen is that given very bad timing luck, the leader
might have to do all the work during a rescan, because workers think
they have nothing to do, if they are able to start up before the eventual
ReScan of the leader's parallel-aware table scan node has reset the
shared scan state.
Although this problem exists in 9.6, there does not seem to be any way
for it to manifest there. Without GatherMerge, it seems that a plan tree
that has a rescan-short-circuiting node below Gather will always also
have one above it that will short-circuit in the same cases, preventing
the Gather from being rescanned. Hence we won't take the risk of
back-patching this change into 9.6. But v10 needs it.
Discussion: https://postgr.es/m/CAA4eK1JkByysFJNh9M349u_nNjqETuEnY_y1VUc_kJiU0bxtaQ@mail.gmail.com
David Rowley
Tom Lane
Etsuro Fujita
Amit Kapila
Bruce Momjian
Heikki Linnakangas
Alvaro Herrera
Tomas Vondra
Jeff Davis
Andres Freund
Peter Eisentraut
Robert Haas
Peter Eisentraut
Simon Riggs