Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitdeb71fa

Browse files
committed
Fix costing for parallel aggregation.
The original patch kind of ignored the fact that we were doing somethingdifferent from a costing point of view, but nobody noticed. This patchfixes that oversight.David Rowley
1 parent46d73e0 commitdeb71fa

File tree

3 files changed

+92
-25
lines changed

3 files changed

+92
-25
lines changed

‎src/backend/optimizer/plan/planner.c

Lines changed: 31 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -3262,6 +3262,8 @@ create_grouping_paths(PlannerInfo *root,
32623262
RelOptInfo*grouped_rel;
32633263
PathTarget*partial_grouping_target=NULL;
32643264
AggClauseCostsagg_costs;
3265+
AggClauseCostsagg_partial_costs;/* parallel only */
3266+
AggClauseCostsagg_final_costs;/* parallel only */
32653267
Sizehashaggtablesize;
32663268
doubledNumGroups;
32673269
doubledNumPartialGroups=0;
@@ -3346,8 +3348,10 @@ create_grouping_paths(PlannerInfo *root,
33463348
MemSet(&agg_costs,0,sizeof(AggClauseCosts));
33473349
if (parse->hasAggs)
33483350
{
3349-
count_agg_clauses(root, (Node*)target->exprs,&agg_costs);
3350-
count_agg_clauses(root,parse->havingQual,&agg_costs);
3351+
count_agg_clauses(root, (Node*)target->exprs,&agg_costs, true,
3352+
false, false);
3353+
count_agg_clauses(root,parse->havingQual,&agg_costs, true, false,
3354+
false);
33513355
}
33523356

33533357
/*
@@ -3422,6 +3426,25 @@ create_grouping_paths(PlannerInfo *root,
34223426
NIL,
34233427
NIL);
34243428

3429+
/*
3430+
* Collect statistics about aggregates for estimating costs of
3431+
* performing aggregation in parallel.
3432+
*/
3433+
MemSet(&agg_partial_costs,0,sizeof(AggClauseCosts));
3434+
MemSet(&agg_final_costs,0,sizeof(AggClauseCosts));
3435+
if (parse->hasAggs)
3436+
{
3437+
/* partial phase */
3438+
count_agg_clauses(root, (Node*)partial_grouping_target->exprs,
3439+
&agg_partial_costs, false, false, true);
3440+
3441+
/* final phase */
3442+
count_agg_clauses(root, (Node*)target->exprs,&agg_final_costs,
3443+
true, true, true);
3444+
count_agg_clauses(root,parse->havingQual,&agg_final_costs, true,
3445+
true, true);
3446+
}
3447+
34253448
if (can_sort)
34263449
{
34273450
/* Checked in set_grouped_rel_consider_parallel() */
@@ -3457,7 +3480,7 @@ create_grouping_paths(PlannerInfo *root,
34573480
parse->groupClause ?AGG_SORTED :AGG_PLAIN,
34583481
parse->groupClause,
34593482
NIL,
3460-
&agg_costs,
3483+
&agg_partial_costs,
34613484
dNumPartialGroups,
34623485
false,
34633486
false,
@@ -3482,7 +3505,7 @@ create_grouping_paths(PlannerInfo *root,
34823505

34833506
hashaggtablesize=
34843507
estimate_hashagg_tablesize(cheapest_partial_path,
3485-
&agg_costs,
3508+
&agg_partial_costs,
34863509
dNumPartialGroups);
34873510

34883511
/*
@@ -3499,7 +3522,7 @@ create_grouping_paths(PlannerInfo *root,
34993522
AGG_HASHED,
35003523
parse->groupClause,
35013524
NIL,
3502-
&agg_costs,
3525+
&agg_partial_costs,
35033526
dNumPartialGroups,
35043527
false,
35053528
false,
@@ -3631,7 +3654,7 @@ create_grouping_paths(PlannerInfo *root,
36313654
parse->groupClause ?AGG_SORTED :AGG_PLAIN,
36323655
parse->groupClause,
36333656
(List*)parse->havingQual,
3634-
&agg_costs,
3657+
&agg_final_costs,
36353658
dNumGroups,
36363659
true,
36373660
true,
@@ -3691,7 +3714,7 @@ create_grouping_paths(PlannerInfo *root,
36913714
Path*path= (Path*)linitial(grouped_rel->partial_pathlist);
36923715

36933716
hashaggtablesize=estimate_hashagg_tablesize(path,
3694-
&agg_costs,
3717+
&agg_final_costs,
36953718
dNumGroups);
36963719

36973720
if (hashaggtablesize<work_mem*1024L)
@@ -3713,7 +3736,7 @@ create_grouping_paths(PlannerInfo *root,
37133736
AGG_HASHED,
37143737
parse->groupClause,
37153738
(List*)parse->havingQual,
3716-
&agg_costs,
3739+
&agg_final_costs,
37173740
dNumGroups,
37183741
true,
37193742
true,

‎src/backend/optimizer/util/clauses.c

Lines changed: 59 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -61,6 +61,9 @@ typedef struct
6161
{
6262
PlannerInfo*root;
6363
AggClauseCosts*costs;
64+
boolfinalizeAggs;
65+
boolcombineStates;
66+
boolserialStates;
6467
}count_agg_clauses_context;
6568

6669
typedefstruct
@@ -540,12 +543,16 @@ contain_agg_clause_walker(Node *node, void *context)
540543
* are no subqueries. There mustn't be outer-aggregate references either.
541544
*/
542545
void
543-
count_agg_clauses(PlannerInfo*root,Node*clause,AggClauseCosts*costs)
546+
count_agg_clauses(PlannerInfo*root,Node*clause,AggClauseCosts*costs,
547+
boolfinalizeAggs,boolcombineStates,boolserialStates)
544548
{
545549
count_agg_clauses_contextcontext;
546550

547551
context.root=root;
548552
context.costs=costs;
553+
context.finalizeAggs=finalizeAggs;
554+
context.combineStates=combineStates;
555+
context.serialStates=serialStates;
549556
(void)count_agg_clauses_walker(clause,&context);
550557
}
551558

@@ -562,6 +569,9 @@ count_agg_clauses_walker(Node *node, count_agg_clauses_context *context)
562569
Form_pg_aggregateaggform;
563570
Oidaggtransfn;
564571
Oidaggfinalfn;
572+
Oidaggcombinefn;
573+
Oidaggserialfn;
574+
Oidaggdeserialfn;
565575
Oidaggtranstype;
566576
int32aggtransspace;
567577
QualCostargcosts;
@@ -583,6 +593,9 @@ count_agg_clauses_walker(Node *node, count_agg_clauses_context *context)
583593
aggform= (Form_pg_aggregate)GETSTRUCT(aggTuple);
584594
aggtransfn=aggform->aggtransfn;
585595
aggfinalfn=aggform->aggfinalfn;
596+
aggcombinefn=aggform->aggcombinefn;
597+
aggserialfn=aggform->aggserialfn;
598+
aggdeserialfn=aggform->aggdeserialfn;
586599
aggtranstype=aggform->aggtranstype;
587600
aggtransspace=aggform->aggtransspace;
588601
ReleaseSysCache(aggTuple);
@@ -592,28 +605,58 @@ count_agg_clauses_walker(Node *node, count_agg_clauses_context *context)
592605
if (aggref->aggorder!=NIL||aggref->aggdistinct!=NIL)
593606
costs->numOrderedAggs++;
594607

595-
/* add component function execution costs to appropriate totals */
596-
costs->transCost.per_tuple+=get_func_cost(aggtransfn)*cpu_operator_cost;
597-
if (OidIsValid(aggfinalfn))
598-
costs->finalCost+=get_func_cost(aggfinalfn)*cpu_operator_cost;
608+
/*
609+
* Add the appropriate component function execution costs to
610+
* appropriate totals.
611+
*/
612+
if (context->combineStates)
613+
{
614+
/* charge for combining previously aggregated states */
615+
costs->transCost.per_tuple+=get_func_cost(aggcombinefn)*cpu_operator_cost;
599616

600-
/* also add the input expressions' cost to per-input-row costs */
601-
cost_qual_eval_node(&argcosts, (Node*)aggref->args,context->root);
602-
costs->transCost.startup+=argcosts.startup;
603-
costs->transCost.per_tuple+=argcosts.per_tuple;
617+
/* charge for deserialization, when appropriate */
618+
if (context->serialStates&&OidIsValid(aggdeserialfn))
619+
costs->transCost.per_tuple+=get_func_cost(aggdeserialfn)*cpu_operator_cost;
620+
}
621+
else
622+
costs->transCost.per_tuple+=get_func_cost(aggtransfn)*cpu_operator_cost;
623+
624+
if (context->finalizeAggs)
625+
{
626+
if (OidIsValid(aggfinalfn))
627+
costs->finalCost+=get_func_cost(aggfinalfn)*cpu_operator_cost;
628+
}
629+
elseif (context->serialStates)
630+
{
631+
if (OidIsValid(aggserialfn))
632+
costs->finalCost+=get_func_cost(aggserialfn)*cpu_operator_cost;
633+
}
604634

605635
/*
606-
* Add any filter's cost to per-input-row costs.
607-
*
608-
* XXX Ideally we should reduce input expression costs according to
609-
* filter selectivity, but it's not clear it's worth the trouble.
636+
* Some costs will already have been incurred by the initial aggregate
637+
* node, so we mustn't include these again.
610638
*/
611-
if (aggref->aggfilter)
639+
if (!context->combineStates)
612640
{
613-
cost_qual_eval_node(&argcosts, (Node*)aggref->aggfilter,
614-
context->root);
641+
/* add the input expressions' cost to per-input-row costs */
642+
cost_qual_eval_node(&argcosts, (Node*)aggref->args,context->root);
615643
costs->transCost.startup+=argcosts.startup;
616644
costs->transCost.per_tuple+=argcosts.per_tuple;
645+
646+
/*
647+
* Add any filter's cost to per-input-row costs.
648+
*
649+
* XXX Ideally we should reduce input expression costs according
650+
* to filter selectivity, but it's not clear it's worth the
651+
* trouble.
652+
*/
653+
if (aggref->aggfilter)
654+
{
655+
cost_qual_eval_node(&argcosts, (Node*)aggref->aggfilter,
656+
context->root);
657+
costs->transCost.startup+=argcosts.startup;
658+
costs->transCost.per_tuple+=argcosts.per_tuple;
659+
}
617660
}
618661

619662
/*

‎src/include/optimizer/clauses.h

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -67,7 +67,8 @@ extern List *make_ands_implicit(Expr *clause);
6767
externPartialAggTypeaggregates_allow_partial(Node*clause);
6868
externboolcontain_agg_clause(Node*clause);
6969
externvoidcount_agg_clauses(PlannerInfo*root,Node*clause,
70-
AggClauseCosts*costs);
70+
AggClauseCosts*costs,boolfinalizeAggs,
71+
boolcombineStates,boolserialStates);
7172

7273
externboolcontain_window_function(Node*clause);
7374
externWindowFuncLists*find_window_functions(Node*clause,IndexmaxWinRef);

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp