Here we adjust the partial path generation for parallel DISTINCT queries
to add Sort nodes on top of any unsorted partial distinct paths.
This increases the likelihood of the planner pushing a Sort below a Gather
Merge which enables the final phase of the parallel distinct to be
implemented using a Unique node in more cases.
Sorting the partial distinct paths is particularly useful when the
DISTINCT query has an ORDER BY and LIMIT clause as this can allow cheaper
plans by having the workers Hash Aggregate then Sort before feeding the
results into the Gather Merge. The non-parallel portion of the plan then
becomes very cheap as it leaves only Unique and Limit to do in the leader
process.
Author: Richard Guo
Reviewed-by: David Rowley
Discussion: https://postgr.es/m/CAMbWs48u9VoVOouJsys1qOaC9WVGVmBa+wT1dx8KvxF5GPzezA@mail.gmail.com
Similar to what was done in 5543677ec for non-parallel DISTINCT, apply
the same optimization when the distinct_pathkeys are empty for the
partial paths too.
This can be faster than the non-parallel version when the first row
matching the WHERE clause of the query takes a while to find. Parallel
workers could speed that process up considerably.
Author: Richard Guo
Reviewed-by: David Rowley
Discussion: https://postgr.es/m/CAMbWs49JC0qvfUbzs-TVzgMpSSBiMJ_6sN=BaA9iohBgYkr=LA@mail.gmail.com
Avoid explicitly grouping by columns that we know are redundant
for sorting, for example we need group by only one of x and y in
SELECT ... WHERE x = y GROUP BY x, y
This comes up more often than you might think, as shown by the
changes in the regression tests. It's nearly free to detect too,
since we are just piggybacking on the existing logic that detects
redundant pathkeys. (In some of the existing plans that change,
it's visible that a sort step preceding the grouping step already
didn't bother to sort by the redundant column, making the old plan
a bit silly-looking.)
To do this, build processed_groupClause and processed_distinctClause
lists that omit any provably-redundant sort items, and consult those
not the originals where relevant. This means that within the
planner, one should usually consult root->processed_groupClause or
root->processed_distinctClause if one wants to know which columns
are to be grouped on; but to check whether grouping or distinct-ing
is happening at all, check non-NIL-ness of parse->groupClause or
parse->distinctClause. This is comparable to longstanding rules
about handling the HAVING clause, so I don't think it'll be a huge
maintenance problem.
nodeAgg.c also needs minor mods, because it's now possible to generate
AGG_PLAIN and AGG_SORTED Agg nodes with zero grouping columns.
Patch by me; thanks to Richard Guo and David Rowley for review.
Discussion: https://postgr.es/m/185315.1672179489@sss.pgh.pa.us
Prior to this, we only considered a full sort on the cheapest input path
and uniquifying any path which was already sorted in the required sort
order. Here we adjust create_final_distinct_paths() so that it also
adds an Incremental Sort path on any path which has presorted keys.
Additionally, this adjusts the parallel distinct code so that we now
consider sorting the cheapest partial path and incrementally sorting any
partial paths with presorted keys. Previously we didn't consider any
sorting for parallel distinct and only added a unique path atop any path
which had the required pathkeys already.
Author: David Rowley
Reviewed-by: Richard Guo
Discussion: https://postgr.es/m/CAApHDvo8Lz2H=42urBbfP65LTcEUOh288MT7DsG2_EWtW1AXHQ@mail.gmail.com
When all of the query's DISTINCT pathkeys have been marked as redundant
due to EquivalenceClasses existing which contain constants, we can just
implement the DISTINCT operation on a query by just limiting the number of
returned rows to 1 instead of performing a Unique on all of the matching
(duplicate) rows.
This applies in cases such as:
SELECT DISTINCT col,col2 FROM tab WHERE col = 1 AND col2 = 10;
If there are any matching rows, then they must all be {1,10}. There's no
point in fetching all of those and running a Unique operator on them to
leave only a single row. Here we effectively just find the first row and
then stop. We are obviously unable to apply this optimization if either
the col = 1 or col2 = 10 were missing from the WHERE clause or if there
were any additional columns in the SELECT clause.
Such queries are probably not all that common, but detecting when we can
apply this optimization amounts to checking if the distinct_pathkeys are
NULL, which is very cheap indeed.
Nothing is done here to check if the query already has a LIMIT clause. If
it does then the plan may end up with 2 Limits nodes. There's no harm in
that and it's probably not worth the complexity to unify them into a
single Limit node.
Author: David Rowley
Reviewed-by: Richard Guo
Discussion: https://postgr.es/m/CAApHDvqS0j8RUWRUSgCAXxOqnYjHUXmKwspRj4GzVfOO25ByHA@mail.gmail.com
Discussion: https://postgr.es/m/MEYPR01MB7101CD5DA0A07C9DE2B74850A4239@MEYPR01MB7101.ausprd01.prod.outlook.com
The idea behind this patch is to make it possible to run individual
test scripts without running the entire core test suite. Making all
the scripts completely independent would involve a massive rewrite,
and would probably be worse for coverage of things like concurrent DDL.
So this patch just does what seems practical with limited changes.
The net effect is that any test script can be run after running
limited earlier dependencies:
* all scripts depend on test_setup
* many scripts depend on create_index
* other dependencies are few in number, and are documented in
the parallel_schedule file.
To accomplish this, I chose a small number of commonly-used tables
and moved their creation and filling into test_setup. Later scripts
are expected not to modify these tables' data contents, for fear of
affecting other scripts' results. Also, our former habit of declaring
all C functions in one place is now gone in favor of declaring them
where they're used, if that's just one script, or in test_setup if
necessary.
There's more that could be done to remove some of the remaining
inter-script dependencies, but significantly more-invasive changes
would be needed, and at least for now it doesn't seem worth it.
Discussion: https://postgr.es/m/1114748.1640383217@sss.pgh.pa.us
We've supported parallel aggregation since e06a38965. At the time, we
didn't quite get around to also adding parallel DISTINCT. So, let's do
that now.
This is implemented by introducing a two-phase DISTINCT. Phase 1 is
performed on parallel workers, rows are made distinct there either by
hashing or by sort/unique. The results from the parallel workers are
combined and the final distinct phase is performed serially to get rid of
any duplicate rows that appear due to combining rows for each of the
parallel workers.
Author: David Rowley
Reviewed-by: Zhihong Yu
Discussion: https://postgr.es/m/CAApHDvrjRxVKwQN0he79xS+9wyotFXL=RmoWqGGO2N45Farpgw@mail.gmail.com
While performing hash aggregation, track memory usage when adding new
groups to a hash table. If the memory usage exceeds work_mem, enter
"spill mode".
In spill mode, new groups are not created in the hash table(s), but
existing groups continue to be advanced if input tuples match. Tuples
that would cause a new group to be created are instead spilled to a
logical tape to be processed later.
The tuples are spilled in a partitioned fashion. When all tuples from
the outer plan are processed (either by advancing the group or
spilling the tuple), finalize and emit the groups from the hash
table. Then, create new batches of work from the spilled partitions,
and select one of the saved batches and process it (possibly spilling
recursively).
Author: Jeff Davis
Reviewed-by: Tomas Vondra, Adam Lee, Justin Pryzby, Taylor Vesely, Melanie Plageman
Discussion: https://postgr.es/m/507ac540ec7c20136364b5272acbcd4574aa76ef.camel@j-davis.com
As part of upper planner pathification (commit 3fc6e2d7f5) I redid
createplan.c's approach to the physical-tlist optimization, in which scan
nodes are allowed to return exactly the underlying table's columns so as
to save doing a projection step at runtime. The logic was intentionally
more aggressive than before about applying the optimization, which is
generally a good thing, but Andres Freund found a case in which it got
too aggressive. Namely, if any column is referenced more than once in
the parent plan node's sorting or grouping column list, we can't optimize
because then that column would need to have more than one ressortgroupref
label, and we only have space for one.
Add logic to detect this situation in use_physical_tlist(), and also add
some error checking in apply_pathtarget_labeling_to_tlist(), which this
example proves was being overly cavalier about whether what it was doing
made any sense.
The added test case exposes the problem only because we do not eliminate
duplicate grouping keys. That might be something to fix someday, but it
doesn't seem like appropriate post-beta work.
Report: <20160526021235.w4nq7k3gnheg7vit@alap3.anarazel.de>
as methods for implementing the DISTINCT step. This eliminates the former
performance gap between DISTINCT and GROUP BY, and also makes it possible
to do SELECT DISTINCT on datatypes that only support hashing not sorting.
SELECT DISTINCT ON is still always implemented by sorting; it would take
executor changes to support hashing that, and it's not clear it's worth
the trouble.
This is a release-note-worthy incompatibility from previous PG versions,
since SELECT DISTINCT can no longer be counted on to deliver sorted output
without explicitly saying ORDER BY. (Anyone who can't cope with that
can consider turning off enable_hashagg.)
Several regression test queries needed to have ORDER BY added to preserve
stable output order. I fixed the ones that manifested here, but there
might be some other cases that show up on other platforms.
David Rowley
Tom Lane
Jeff Davis
Peter Eisentraut
Neil Conway
Thomas G. Lockhart
Bruce Momjian
Marc G. Fournier