Skip to content

Commit d3cc37f

Browse files
robertmhaasrobertmhaas
committed
Don't scan partitioned tables.
Partitioned tables do not contain any data; only their unpartitioned descendents need to be scanned. However, the partitioned tables still need to be locked, even though they're not scanned. To make that work, Append and MergeAppend relations now need to carry a list of (unscanned) partitioned relations that must be locked, and InitPlan must lock all partitioned result relations. Aside from the obvious advantage of avoiding some work at execution time, this has two other advantages. First, it may improve the planner's decision-making in some cases since the empty relation might throw things off. Second, it paves the way to getting rid of the storage for partitioned tables altogether. Amit Langote, reviewed by me. Discussion: http://postgr.es/m/6837c359-45c4-8044-34d1-736756335a15@lab.ntt.co.jp
1 parent d5286aa commit d3cc37f

27 files changed

+537
-164
lines changed

‎src/backend/executor/execMain.c

Lines changed: 21 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -844,6 +844,22 @@ InitPlan(QueryDesc *queryDesc, int eflags)
844844
estate->es_num_result_relations = numResultRelations;
845845
/* es_result_relation_info is NULL except when within ModifyTable */
846846
estate->es_result_relation_info = NULL;
847+
848+
/*
849+
* In the partitioned result relation case, lock the non-leaf result
850+
* relations too. We don't however need ResultRelInfos for them.
851+
*/
852+
if (plannedstmt->nonleafResultRelations)
853+
{
854+
foreach(l, plannedstmt->nonleafResultRelations)
855+
{
856+
Index resultRelationIndex = lfirst_int(l);
857+
Oid resultRelationOid;
858+
859+
resultRelationOid = getrelid(resultRelationIndex, rangeTable);
860+
LockRelationOid(resultRelationOid, RowExclusiveLock);
861+
}
862+
}
847863
}
848864
else
849865
{
@@ -858,7 +874,11 @@ InitPlan(QueryDesc *queryDesc, int eflags)
858874
/*
859875
* Similarly, we have to lock relations selected FOR [KEY] UPDATE/SHARE
860876
* before we initialize the plan tree, else we'd be risking lock upgrades.
861-
* While we are at it, build the ExecRowMark list.
877+
* While we are at it, build the ExecRowMark list. Any partitioned child
878+
* tables are ignored here (because isParent=true) and will be locked by
879+
* the first Append or MergeAppend node that references them. (Note that
880+
* the RowMarks corresponding to partitioned child tables are present in
881+
* the same list as the rest, i.e., plannedstmt->rowMarks.)
862882
*/
863883
estate->es_rowMarks = NIL;
864884
foreach(l, plannedstmt->rowMarks)

‎src/backend/executor/execParallel.c

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -161,6 +161,7 @@ ExecSerializePlan(Plan *plan, EState *estate)
161161
pstmt->planTree = plan;
162162
pstmt->rtable = estate->es_range_table;
163163
pstmt->resultRelations = NIL;
164+
pstmt->nonleafResultRelations = NIL;
164165
pstmt->subplans = estate->es_plannedstmt->subplans;
165166
pstmt->rewindPlanIDs = NULL;
166167
pstmt->rowMarks = NIL;

‎src/backend/executor/execUtils.c

Lines changed: 56 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -43,6 +43,7 @@
4343
#include "executor/executor.h"
4444
#include "nodes/nodeFuncs.h"
4545
#include "parser/parsetree.h"
46+
#include "storage/lmgr.h"
4647
#include "utils/memutils.h"
4748
#include "utils/rel.h"
4849

@@ -953,3 +954,58 @@ ShutdownExprContext(ExprContext *econtext, bool isCommit)
953954

954955
MemoryContextSwitchTo(oldcontext);
955956
}
957+
958+
/*
959+
* ExecLockNonLeafAppendTables
960+
*
961+
* Locks, if necessary, the tables indicated by the RT indexes contained in
962+
* the partitioned_rels list. These are the non-leaf tables in the partition
963+
* tree controlled by a given Append or MergeAppend node.
964+
*/
965+
void
966+
ExecLockNonLeafAppendTables(List *partitioned_rels, EState *estate)
967+
{
968+
PlannedStmt *stmt = estate->es_plannedstmt;
969+
ListCell *lc;
970+
971+
foreach(lc, partitioned_rels)
972+
{
973+
ListCell *l;
974+
Index rti = lfirst_int(lc);
975+
bool is_result_rel = false;
976+
Oid relid = getrelid(rti, estate->es_range_table);
977+
978+
/* If this is a result relation, already locked in InitPlan */
979+
foreach(l, stmt->nonleafResultRelations)
980+
{
981+
if (rti == lfirst_int(l))
982+
{
983+
is_result_rel = true;
984+
break;
985+
}
986+
}
987+
988+
/*
989+
* Not a result relation; check if there is a RowMark that requires
990+
* taking a RowShareLock on this rel.
991+
*/
992+
if (!is_result_rel)
993+
{
994+
PlanRowMark *rc = NULL;
995+
996+
foreach(l, stmt->rowMarks)
997+
{
998+
if (((PlanRowMark *) lfirst(l))->rti == rti)
999+
{
1000+
rc = lfirst(l);
1001+
break;
1002+
}
1003+
}
1004+
1005+
if (rc && RowMarkRequiresRowShareLock(rc->markType))
1006+
LockRelationOid(relid, RowShareLock);
1007+
else
1008+
LockRelationOid(relid, AccessShareLock);
1009+
}
1010+
}
1011+
}

‎src/backend/executor/nodeAppend.c

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -128,6 +128,12 @@ ExecInitAppend(Append *node, EState *estate, int eflags)
128128
/* check for unsupported flags */
129129
Assert(!(eflags & EXEC_FLAG_MARK));
130130

131+
/*
132+
* Lock the non-leaf tables in the partition tree controlled by this
133+
* node. It's a no-op for non-partitioned parent tables.
134+
*/
135+
ExecLockNonLeafAppendTables(node->partitioned_rels, estate);
136+
131137
/*
132138
* Set up empty vector of subplan states
133139
*/

‎src/backend/executor/nodeMergeAppend.c

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -71,6 +71,12 @@ ExecInitMergeAppend(MergeAppend *node, EState *estate, int eflags)
7171
/* check for unsupported flags */
7272
Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
7373

74+
/*
75+
* Lock the non-leaf tables in the partition tree controlled by this
76+
* node. It's a no-op for non-partitioned parent tables.
77+
*/
78+
ExecLockNonLeafAppendTables(node->partitioned_rels, estate);
79+
7480
/*
7581
* Set up empty vector of subplan states
7682
*/

‎src/backend/executor/nodeModifyTable.c

Lines changed: 18 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -45,6 +45,7 @@
4545
#include "foreign/fdwapi.h"
4646
#include "miscadmin.h"
4747
#include "nodes/nodeFuncs.h"
48+
#include "parser/parsetree.h"
4849
#include "storage/bufmgr.h"
4950
#include "storage/lmgr.h"
5051
#include "utils/builtins.h"
@@ -1725,8 +1726,20 @@ ExecInitModifyTable(ModifyTable *node, EState *estate, int eflags)
17251726

17261727
estate->es_result_relation_info = saved_resultRelInfo;
17271728

1729+
/* The root table RT index is at the head of the partitioned_rels list */
1730+
if (node->partitioned_rels)
1731+
{
1732+
Index root_rti;
1733+
Oid root_oid;
1734+
1735+
root_rti = linitial_int(node->partitioned_rels);
1736+
root_oid = getrelid(root_rti, estate->es_range_table);
1737+
rel = heap_open(root_oid, NoLock); /* locked by InitPlan */
1738+
}
1739+
else
1740+
rel = mtstate->resultRelInfo->ri_RelationDesc;
1741+
17281742
/* Build state for INSERT tuple routing */
1729-
rel = mtstate->resultRelInfo->ri_RelationDesc;
17301743
if (operation == CMD_INSERT &&
17311744
rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
17321745
{
@@ -1897,6 +1910,10 @@ ExecInitModifyTable(ModifyTable *node, EState *estate, int eflags)
18971910
mtstate->ps.ps_ExprContext = NULL;
18981911
}
18991912

1913+
/* Close the root partitioned rel if we opened it above. */
1914+
if (rel != mtstate->resultRelInfo->ri_RelationDesc)
1915+
heap_close(rel, NoLock);
1916+
19001917
/*
19011918
* If needed, Initialize target list, projection and qual for ON CONFLICT
19021919
* DO UPDATE.

‎src/backend/nodes/copyfuncs.c

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -90,6 +90,7 @@ _copyPlannedStmt(const PlannedStmt *from)
9090
COPY_NODE_FIELD(planTree);
9191
COPY_NODE_FIELD(rtable);
9292
COPY_NODE_FIELD(resultRelations);
93+
COPY_NODE_FIELD(nonleafResultRelations);
9394
COPY_NODE_FIELD(subplans);
9495
COPY_BITMAPSET_FIELD(rewindPlanIDs);
9596
COPY_NODE_FIELD(rowMarks);
@@ -200,6 +201,7 @@ _copyModifyTable(const ModifyTable *from)
200201
COPY_SCALAR_FIELD(operation);
201202
COPY_SCALAR_FIELD(canSetTag);
202203
COPY_SCALAR_FIELD(nominalRelation);
204+
COPY_NODE_FIELD(partitioned_rels);
203205
COPY_NODE_FIELD(resultRelations);
204206
COPY_SCALAR_FIELD(resultRelIndex);
205207
COPY_NODE_FIELD(plans);
@@ -235,6 +237,7 @@ _copyAppend(const Append *from)
235237
/*
236238
* copy remainder of node
237239
*/
240+
COPY_NODE_FIELD(partitioned_rels);
238241
COPY_NODE_FIELD(appendplans);
239242

240243
return newnode;
@@ -256,6 +259,7 @@ _copyMergeAppend(const MergeAppend *from)
256259
/*
257260
* copy remainder of node
258261
*/
262+
COPY_NODE_FIELD(partitioned_rels);
259263
COPY_NODE_FIELD(mergeplans);
260264
COPY_SCALAR_FIELD(numCols);
261265
COPY_POINTER_FIELD(sortColIdx, from->numCols * sizeof(AttrNumber));
@@ -2204,6 +2208,20 @@ _copyAppendRelInfo(const AppendRelInfo *from)
22042208
return newnode;
22052209
}
22062210

2211+
/*
2212+
* _copyPartitionedChildRelInfo
2213+
*/
2214+
static PartitionedChildRelInfo *
2215+
_copyPartitionedChildRelInfo(const PartitionedChildRelInfo *from)
2216+
{
2217+
PartitionedChildRelInfo *newnode = makeNode(PartitionedChildRelInfo);
2218+
2219+
COPY_SCALAR_FIELD(parent_relid);
2220+
COPY_NODE_FIELD(child_rels);
2221+
2222+
return newnode;
2223+
}
2224+
22072225
/*
22082226
* _copyPlaceHolderInfo
22092227
*/
@@ -4894,6 +4912,9 @@ copyObject(const void *from)
48944912
case T_AppendRelInfo:
48954913
retval = _copyAppendRelInfo(from);
48964914
break;
4915+
case T_PartitionedChildRelInfo:
4916+
retval = _copyPartitionedChildRelInfo(from);
4917+
break;
48974918
case T_PlaceHolderInfo:
48984919
retval = _copyPlaceHolderInfo(from);
48994920
break;

‎src/backend/nodes/equalfuncs.c

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -895,6 +895,15 @@ _equalAppendRelInfo(const AppendRelInfo *a, const AppendRelInfo *b)
895895
return true;
896896
}
897897

898+
static bool
899+
_equalPartitionedChildRelInfo(const PartitionedChildRelInfo *a, const PartitionedChildRelInfo *b)
900+
{
901+
COMPARE_SCALAR_FIELD(parent_relid);
902+
COMPARE_NODE_FIELD(child_rels);
903+
904+
return true;
905+
}
906+
898907
static bool
899908
_equalPlaceHolderInfo(const PlaceHolderInfo *a, const PlaceHolderInfo *b)
900909
{
@@ -3104,6 +3113,9 @@ equal(const void *a, const void *b)
31043113
case T_AppendRelInfo:
31053114
retval = _equalAppendRelInfo(a, b);
31063115
break;
3116+
case T_PartitionedChildRelInfo:
3117+
retval = _equalPartitionedChildRelInfo(a, b);
3118+
break;
31073119
case T_PlaceHolderInfo:
31083120
retval = _equalPlaceHolderInfo(a, b);
31093121
break;

‎src/backend/nodes/outfuncs.c

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -252,6 +252,7 @@ _outPlannedStmt(StringInfo str, const PlannedStmt *node)
252252
WRITE_NODE_FIELD(planTree);
253253
WRITE_NODE_FIELD(rtable);
254254
WRITE_NODE_FIELD(resultRelations);
255+
WRITE_NODE_FIELD(nonleafResultRelations);
255256
WRITE_NODE_FIELD(subplans);
256257
WRITE_BITMAPSET_FIELD(rewindPlanIDs);
257258
WRITE_NODE_FIELD(rowMarks);
@@ -344,6 +345,7 @@ _outModifyTable(StringInfo str, const ModifyTable *node)
344345
WRITE_ENUM_FIELD(operation, CmdType);
345346
WRITE_BOOL_FIELD(canSetTag);
346347
WRITE_UINT_FIELD(nominalRelation);
348+
WRITE_NODE_FIELD(partitioned_rels);
347349
WRITE_NODE_FIELD(resultRelations);
348350
WRITE_INT_FIELD(resultRelIndex);
349351
WRITE_NODE_FIELD(plans);
@@ -368,6 +370,7 @@ _outAppend(StringInfo str, const Append *node)
368370

369371
_outPlanInfo(str, (const Plan *) node);
370372

373+
WRITE_NODE_FIELD(partitioned_rels);
371374
WRITE_NODE_FIELD(appendplans);
372375
}
373376

@@ -380,6 +383,7 @@ _outMergeAppend(StringInfo str, const MergeAppend *node)
380383

381384
_outPlanInfo(str, (const Plan *) node);
382385

386+
WRITE_NODE_FIELD(partitioned_rels);
383387
WRITE_NODE_FIELD(mergeplans);
384388

385389
WRITE_INT_FIELD(numCols);
@@ -1808,6 +1812,7 @@ _outAppendPath(StringInfo str, const AppendPath *node)
18081812

18091813
_outPathInfo(str, (const Path *) node);
18101814

1815+
WRITE_NODE_FIELD(partitioned_rels);
18111816
WRITE_NODE_FIELD(subpaths);
18121817
}
18131818

@@ -1818,6 +1823,7 @@ _outMergeAppendPath(StringInfo str, const MergeAppendPath *node)
18181823

18191824
_outPathInfo(str, (const Path *) node);
18201825

1826+
WRITE_NODE_FIELD(partitioned_rels);
18211827
WRITE_NODE_FIELD(subpaths);
18221828
WRITE_FLOAT_FIELD(limit_tuples, "%.0f");
18231829
}
@@ -2023,6 +2029,7 @@ _outModifyTablePath(StringInfo str, const ModifyTablePath *node)
20232029
WRITE_ENUM_FIELD(operation, CmdType);
20242030
WRITE_BOOL_FIELD(canSetTag);
20252031
WRITE_UINT_FIELD(nominalRelation);
2032+
WRITE_NODE_FIELD(partitioned_rels);
20262033
WRITE_NODE_FIELD(resultRelations);
20272034
WRITE_NODE_FIELD(subpaths);
20282035
WRITE_NODE_FIELD(subroots);
@@ -2099,6 +2106,7 @@ _outPlannerGlobal(StringInfo str, const PlannerGlobal *node)
20992106
WRITE_NODE_FIELD(finalrtable);
21002107
WRITE_NODE_FIELD(finalrowmarks);
21012108
WRITE_NODE_FIELD(resultRelations);
2109+
WRITE_NODE_FIELD(nonleafResultRelations);
21022110
WRITE_NODE_FIELD(relationOids);
21032111
WRITE_NODE_FIELD(invalItems);
21042112
WRITE_INT_FIELD(nParamExec);
@@ -2137,6 +2145,7 @@ _outPlannerInfo(StringInfo str, const PlannerInfo *node)
21372145
WRITE_NODE_FIELD(full_join_clauses);
21382146
WRITE_NODE_FIELD(join_info_list);
21392147
WRITE_NODE_FIELD(append_rel_list);
2148+
WRITE_NODE_FIELD(pcinfo_list);
21402149
WRITE_NODE_FIELD(rowMarks);
21412150
WRITE_NODE_FIELD(placeholder_list);
21422151
WRITE_NODE_FIELD(fkey_list);
@@ -2419,6 +2428,15 @@ _outAppendRelInfo(StringInfo str, const AppendRelInfo *node)
24192428
WRITE_OID_FIELD(parent_reloid);
24202429
}
24212430

2431+
static void
2432+
_outPartitionedChildRelInfo(StringInfo str, const PartitionedChildRelInfo *node)
2433+
{
2434+
WRITE_NODE_TYPE("PARTITIONEDCHILDRELINFO");
2435+
2436+
WRITE_UINT_FIELD(parent_relid);
2437+
WRITE_NODE_FIELD(child_rels);
2438+
}
2439+
24222440
static void
24232441
_outPlaceHolderInfo(StringInfo str, const PlaceHolderInfo *node)
24242442
{
@@ -3906,6 +3924,9 @@ outNode(StringInfo str, const void *obj)
39063924
case T_AppendRelInfo:
39073925
_outAppendRelInfo(str, obj);
39083926
break;
3927+
case T_PartitionedChildRelInfo:
3928+
_outPartitionedChildRelInfo(str, obj);
3929+
break;
39093930
case T_PlaceHolderInfo:
39103931
_outPlaceHolderInfo(str, obj);
39113932
break;

‎src/backend/nodes/readfuncs.c

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1444,6 +1444,7 @@ _readPlannedStmt(void)
14441444
READ_NODE_FIELD(planTree);
14451445
READ_NODE_FIELD(rtable);
14461446
READ_NODE_FIELD(resultRelations);
1447+
READ_NODE_FIELD(nonleafResultRelations);
14471448
READ_NODE_FIELD(subplans);
14481449
READ_BITMAPSET_FIELD(rewindPlanIDs);
14491450
READ_NODE_FIELD(rowMarks);
@@ -1535,6 +1536,7 @@ _readModifyTable(void)
15351536
READ_ENUM_FIELD(operation, CmdType);
15361537
READ_BOOL_FIELD(canSetTag);
15371538
READ_UINT_FIELD(nominalRelation);
1539+
READ_NODE_FIELD(partitioned_rels);
15381540
READ_NODE_FIELD(resultRelations);
15391541
READ_INT_FIELD(resultRelIndex);
15401542
READ_NODE_FIELD(plans);
@@ -1564,6 +1566,7 @@ _readAppend(void)
15641566

15651567
ReadCommonPlan(&local_node->plan);
15661568

1569+
READ_NODE_FIELD(partitioned_rels);
15671570
READ_NODE_FIELD(appendplans);
15681571

15691572
READ_DONE();
@@ -1579,6 +1582,7 @@ _readMergeAppend(void)
15791582

15801583
ReadCommonPlan(&local_node->plan);
15811584

1585+
READ_NODE_FIELD(partitioned_rels);
15821586
READ_NODE_FIELD(mergeplans);
15831587
READ_INT_FIELD(numCols);
15841588
READ_ATTRNUMBER_ARRAY(sortColIdx, local_node->numCols);

0 commit comments

Comments
 (0)