Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * execPartition.c
4 : * Support routines for partitioning.
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * IDENTIFICATION
10 : * src/backend/executor/execPartition.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 : #include "postgres.h"
15 :
16 : #include "access/table.h"
17 : #include "access/tableam.h"
18 : #include "catalog/index.h"
19 : #include "catalog/partition.h"
20 : #include "executor/execPartition.h"
21 : #include "executor/executor.h"
22 : #include "executor/nodeModifyTable.h"
23 : #include "foreign/fdwapi.h"
24 : #include "mb/pg_wchar.h"
25 : #include "miscadmin.h"
26 : #include "partitioning/partbounds.h"
27 : #include "partitioning/partdesc.h"
28 : #include "partitioning/partprune.h"
29 : #include "rewrite/rewriteManip.h"
30 : #include "utils/acl.h"
31 : #include "utils/injection_point.h"
32 : #include "utils/lsyscache.h"
33 : #include "utils/partcache.h"
34 : #include "utils/rls.h"
35 : #include "utils/ruleutils.h"
36 :
37 :
38 : /*-----------------------
39 : * PartitionTupleRouting - Encapsulates all information required to
40 : * route a tuple inserted into a partitioned table to one of its leaf
41 : * partitions.
42 : *
43 : * partition_root
44 : * The partitioned table that's the target of the command.
45 : *
46 : * partition_dispatch_info
47 : * Array of 'max_dispatch' elements containing a pointer to a
48 : * PartitionDispatch object for every partitioned table touched by tuple
49 : * routing. The entry for the target partitioned table is *always*
50 : * present in the 0th element of this array. See comment for
51 : * PartitionDispatchData->indexes for details on how this array is
52 : * indexed.
53 : *
54 : * nonleaf_partitions
55 : * Array of 'max_dispatch' elements containing pointers to fake
56 : * ResultRelInfo objects for nonleaf partitions, useful for checking
57 : * the partition constraint.
58 : *
59 : * num_dispatch
60 : * The current number of items stored in the 'partition_dispatch_info'
61 : * array. Also serves as the index of the next free array element for
62 : * new PartitionDispatch objects that need to be stored.
63 : *
64 : * max_dispatch
65 : * The current allocated size of the 'partition_dispatch_info' array.
66 : *
67 : * partitions
68 : * Array of 'max_partitions' elements containing a pointer to a
69 : * ResultRelInfo for every leaf partition touched by tuple routing.
70 : * Some of these are pointers to ResultRelInfos which are borrowed out of
71 : * the owning ModifyTableState node. The remainder have been built
72 : * especially for tuple routing. See comment for
73 : * PartitionDispatchData->indexes for details on how this array is
74 : * indexed.
75 : *
76 : * is_borrowed_rel
77 : * Array of 'max_partitions' booleans recording whether a given entry
78 : * in 'partitions' is a ResultRelInfo pointer borrowed from the owning
79 : * ModifyTableState node, rather than being built here.
80 : *
81 : * num_partitions
82 : * The current number of items stored in the 'partitions' array. Also
83 : * serves as the index of the next free array element for new
84 : * ResultRelInfo objects that need to be stored.
85 : *
86 : * max_partitions
87 : * The current allocated size of the 'partitions' array.
88 : *
89 : * memcxt
90 : * Memory context used to allocate subsidiary structs.
91 : *-----------------------
92 : */
93 : struct PartitionTupleRouting
94 : {
95 : Relation partition_root;
96 : PartitionDispatch *partition_dispatch_info;
97 : ResultRelInfo **nonleaf_partitions;
98 : int num_dispatch;
99 : int max_dispatch;
100 : ResultRelInfo **partitions;
101 : bool *is_borrowed_rel;
102 : int num_partitions;
103 : int max_partitions;
104 : MemoryContext memcxt;
105 : };
106 :
107 : /*-----------------------
108 : * PartitionDispatch - information about one partitioned table in a partition
109 : * hierarchy required to route a tuple to any of its partitions. A
110 : * PartitionDispatch is always encapsulated inside a PartitionTupleRouting
111 : * struct and stored inside its 'partition_dispatch_info' array.
112 : *
113 : * reldesc
114 : * Relation descriptor of the table
115 : *
116 : * key
117 : * Partition key information of the table
118 : *
119 : * keystate
120 : * Execution state required for expressions in the partition key
121 : *
122 : * partdesc
123 : * Partition descriptor of the table
124 : *
125 : * tupslot
126 : * A standalone TupleTableSlot initialized with this table's tuple
127 : * descriptor, or NULL if no tuple conversion between the parent is
128 : * required.
129 : *
130 : * tupmap
131 : * TupleConversionMap to convert from the parent's rowtype to this table's
132 : * rowtype (when extracting the partition key of a tuple just before
133 : * routing it through this table). A NULL value is stored if no tuple
134 : * conversion is required.
135 : *
136 : * indexes
137 : * Array of partdesc->nparts elements. For leaf partitions the index
138 : * corresponds to the partition's ResultRelInfo in the encapsulating
139 : * PartitionTupleRouting's partitions array. For partitioned partitions,
140 : * the index corresponds to the PartitionDispatch for it in its
141 : * partition_dispatch_info array. -1 indicates we've not yet allocated
142 : * anything in PartitionTupleRouting for the partition.
143 : *-----------------------
144 : */
145 : typedef struct PartitionDispatchData
146 : {
147 : Relation reldesc;
148 : PartitionKey key;
149 : List *keystate; /* list of ExprState */
150 : PartitionDesc partdesc;
151 : TupleTableSlot *tupslot;
152 : AttrMap *tupmap;
153 : int indexes[FLEXIBLE_ARRAY_MEMBER];
154 : } PartitionDispatchData;
155 :
156 :
157 : static ResultRelInfo *ExecInitPartitionInfo(ModifyTableState *mtstate,
158 : EState *estate, PartitionTupleRouting *proute,
159 : PartitionDispatch dispatch,
160 : ResultRelInfo *rootResultRelInfo,
161 : int partidx);
162 : static void ExecInitRoutingInfo(ModifyTableState *mtstate,
163 : EState *estate,
164 : PartitionTupleRouting *proute,
165 : PartitionDispatch dispatch,
166 : ResultRelInfo *partRelInfo,
167 : int partidx,
168 : bool is_borrowed_rel);
169 : static PartitionDispatch ExecInitPartitionDispatchInfo(EState *estate,
170 : PartitionTupleRouting *proute,
171 : Oid partoid, PartitionDispatch parent_pd,
172 : int partidx, ResultRelInfo *rootResultRelInfo);
173 : static void FormPartitionKeyDatum(PartitionDispatch pd,
174 : TupleTableSlot *slot,
175 : EState *estate,
176 : Datum *values,
177 : bool *isnull);
178 : static int get_partition_for_tuple(PartitionDispatch pd, const Datum *values,
179 : const bool *isnull);
180 : static char *ExecBuildSlotPartitionKeyDescription(Relation rel,
181 : const Datum *values,
182 : const bool *isnull,
183 : int maxfieldlen);
184 : static List *adjust_partition_colnos(List *colnos, ResultRelInfo *leaf_part_rri);
185 : static List *adjust_partition_colnos_using_map(List *colnos, AttrMap *attrMap);
186 : static PartitionPruneState *CreatePartitionPruneState(EState *estate,
187 : PartitionPruneInfo *pruneinfo,
188 : Bitmapset **all_leafpart_rtis);
189 : static void InitPartitionPruneContext(PartitionPruneContext *context,
190 : List *pruning_steps,
191 : PartitionDesc partdesc,
192 : PartitionKey partkey,
193 : PlanState *planstate,
194 : ExprContext *econtext);
195 : static void InitExecPartitionPruneContexts(PartitionPruneState *prunestate,
196 : PlanState *parent_plan,
197 : Bitmapset *initially_valid_subplans,
198 : int n_total_subplans);
199 : static void find_matching_subplans_recurse(PartitionPruningData *prunedata,
200 : PartitionedRelPruningData *pprune,
201 : bool initial_prune,
202 : Bitmapset **validsubplans,
203 : Bitmapset **validsubplan_rtis);
204 :
205 :
206 : /*
207 : * ExecSetupPartitionTupleRouting - sets up information needed during
208 : * tuple routing for partitioned tables, encapsulates it in
209 : * PartitionTupleRouting, and returns it.
210 : *
211 : * Callers must use the returned PartitionTupleRouting during calls to
212 : * ExecFindPartition(). The actual ResultRelInfo for a partition is only
213 : * allocated when the partition is found for the first time.
214 : *
215 : * The current memory context is used to allocate this struct and all
216 : * subsidiary structs that will be allocated from it later on. Typically
217 : * it should be estate->es_query_cxt.
218 : */
219 : PartitionTupleRouting *
220 2885 : ExecSetupPartitionTupleRouting(EState *estate, Relation rel)
221 : {
222 : PartitionTupleRouting *proute;
223 :
224 : /*
225 : * Here we attempt to expend as little effort as possible in setting up
226 : * the PartitionTupleRouting. Each partition's ResultRelInfo is built on
227 : * demand, only when we actually need to route a tuple to that partition.
228 : * The reason for this is that a common case is for INSERT to insert a
229 : * single tuple into a partitioned table and this must be fast.
230 : */
231 2885 : proute = palloc0_object(PartitionTupleRouting);
232 2885 : proute->partition_root = rel;
233 2885 : proute->memcxt = CurrentMemoryContext;
234 : /* Rest of members initialized by zeroing */
235 :
236 : /*
237 : * Initialize this table's PartitionDispatch object. Here we pass in the
238 : * parent as NULL as we don't need to care about any parent of the target
239 : * partitioned table.
240 : */
241 2885 : ExecInitPartitionDispatchInfo(estate, proute, RelationGetRelid(rel),
242 : NULL, 0, NULL);
243 :
244 2885 : return proute;
245 : }
246 :
247 : /*
248 : * ExecFindPartition -- Return the ResultRelInfo for the leaf partition that
249 : * the tuple contained in *slot should belong to.
250 : *
251 : * If the partition's ResultRelInfo does not yet exist in 'proute' then we set
252 : * one up or reuse one from mtstate's resultRelInfo array. When reusing a
253 : * ResultRelInfo from the mtstate we verify that the relation is a valid
254 : * target for INSERTs and initialize tuple routing information.
255 : *
256 : * rootResultRelInfo is the relation named in the query.
257 : *
258 : * estate must be non-NULL; we'll need it to compute any expressions in the
259 : * partition keys. Also, its per-tuple contexts are used as evaluation
260 : * scratch space.
261 : *
262 : * If no leaf partition is found, this routine errors out with the appropriate
263 : * error message. An error may also be raised if the found target partition
264 : * is not a valid target for an INSERT.
265 : */
266 : ResultRelInfo *
267 516745 : ExecFindPartition(ModifyTableState *mtstate,
268 : ResultRelInfo *rootResultRelInfo,
269 : PartitionTupleRouting *proute,
270 : TupleTableSlot *slot, EState *estate)
271 : {
272 516745 : PartitionDispatch *pd = proute->partition_dispatch_info;
273 : Datum values[PARTITION_MAX_KEYS];
274 : bool isnull[PARTITION_MAX_KEYS];
275 : Relation rel;
276 : PartitionDispatch dispatch;
277 : PartitionDesc partdesc;
278 516745 : ExprContext *ecxt = GetPerTupleExprContext(estate);
279 516745 : TupleTableSlot *ecxt_scantuple_saved = ecxt->ecxt_scantuple;
280 516745 : TupleTableSlot *rootslot = slot;
281 516745 : TupleTableSlot *myslot = NULL;
282 : MemoryContext oldcxt;
283 516745 : ResultRelInfo *rri = NULL;
284 :
285 : /* use per-tuple context here to avoid leaking memory */
286 516745 : oldcxt = MemoryContextSwitchTo(GetPerTupleMemoryContext(estate));
287 :
288 : /*
289 : * First check the root table's partition constraint, if any. No point in
290 : * routing the tuple if it doesn't belong in the root table itself.
291 : */
292 516745 : if (rootResultRelInfo->ri_RelationDesc->rd_rel->relispartition)
293 2251 : ExecPartitionCheck(rootResultRelInfo, slot, estate, true);
294 :
295 : /* start with the root partitioned table */
296 516729 : dispatch = pd[0];
297 1091637 : while (dispatch != NULL)
298 : {
299 575004 : int partidx = -1;
300 : bool is_leaf;
301 :
302 575004 : CHECK_FOR_INTERRUPTS();
303 :
304 575004 : rel = dispatch->reldesc;
305 575004 : partdesc = dispatch->partdesc;
306 :
307 : /*
308 : * Extract partition key from tuple. Expression evaluation machinery
309 : * that FormPartitionKeyDatum() invokes expects ecxt_scantuple to
310 : * point to the correct tuple slot. The slot might have changed from
311 : * what was used for the parent table if the table of the current
312 : * partitioning level has different tuple descriptor from the parent.
313 : * So update ecxt_scantuple accordingly.
314 : */
315 575004 : ecxt->ecxt_scantuple = slot;
316 575004 : FormPartitionKeyDatum(dispatch, slot, estate, values, isnull);
317 :
318 : /*
319 : * If this partitioned table has no partitions or no partition for
320 : * these values, error out.
321 : */
322 1149981 : if (partdesc->nparts == 0 ||
323 574983 : (partidx = get_partition_for_tuple(dispatch, values, isnull)) < 0)
324 : {
325 : char *val_desc;
326 :
327 77 : val_desc = ExecBuildSlotPartitionKeyDescription(rel,
328 : values, isnull, 64);
329 : Assert(OidIsValid(RelationGetRelid(rel)));
330 77 : ereport(ERROR,
331 : (errcode(ERRCODE_CHECK_VIOLATION),
332 : errmsg("no partition of relation \"%s\" found for row",
333 : RelationGetRelationName(rel)),
334 : val_desc ?
335 : errdetail("Partition key of the failing row contains %s.",
336 : val_desc) : 0,
337 : errtable(rel)));
338 : }
339 :
340 574921 : is_leaf = partdesc->is_leaf[partidx];
341 574921 : if (is_leaf)
342 : {
343 : /*
344 : * We've reached the leaf -- hurray, we're done. Look to see if
345 : * we've already got a ResultRelInfo for this partition.
346 : */
347 516645 : if (likely(dispatch->indexes[partidx] >= 0))
348 : {
349 : /* ResultRelInfo already built */
350 : Assert(dispatch->indexes[partidx] < proute->num_partitions);
351 512685 : rri = proute->partitions[dispatch->indexes[partidx]];
352 : }
353 : else
354 : {
355 : /*
356 : * If the partition is known in the owning ModifyTableState
357 : * node, we can re-use that ResultRelInfo instead of creating
358 : * a new one with ExecInitPartitionInfo().
359 : */
360 3960 : rri = ExecLookupResultRelByOid(mtstate,
361 3960 : partdesc->oids[partidx],
362 : true, false);
363 3960 : if (rri)
364 : {
365 254 : ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
366 :
367 : /* Verify this ResultRelInfo allows INSERTs */
368 254 : CheckValidResultRel(rri, CMD_INSERT,
369 : node ? node->onConflictAction : ONCONFLICT_NONE,
370 : NIL);
371 :
372 : /*
373 : * Initialize information needed to insert this and
374 : * subsequent tuples routed to this partition.
375 : */
376 254 : ExecInitRoutingInfo(mtstate, estate, proute, dispatch,
377 : rri, partidx, true);
378 : }
379 : else
380 : {
381 : /* We need to create a new one. */
382 3706 : rri = ExecInitPartitionInfo(mtstate, estate, proute,
383 : dispatch,
384 : rootResultRelInfo, partidx);
385 : }
386 : }
387 : Assert(rri != NULL);
388 :
389 : /* Signal to terminate the loop */
390 516633 : dispatch = NULL;
391 : }
392 : else
393 : {
394 : /*
395 : * Partition is a sub-partitioned table; get the PartitionDispatch
396 : */
397 58276 : if (likely(dispatch->indexes[partidx] >= 0))
398 : {
399 : /* Already built. */
400 : Assert(dispatch->indexes[partidx] < proute->num_dispatch);
401 :
402 57673 : rri = proute->nonleaf_partitions[dispatch->indexes[partidx]];
403 :
404 : /*
405 : * Move down to the next partition level and search again
406 : * until we find a leaf partition that matches this tuple
407 : */
408 57673 : dispatch = pd[dispatch->indexes[partidx]];
409 : }
410 : else
411 : {
412 : /* Not yet built. Do that now. */
413 : PartitionDispatch subdispatch;
414 :
415 : /*
416 : * Create the new PartitionDispatch. We pass the current one
417 : * in as the parent PartitionDispatch
418 : */
419 603 : subdispatch = ExecInitPartitionDispatchInfo(estate,
420 : proute,
421 603 : partdesc->oids[partidx],
422 : dispatch, partidx,
423 : mtstate->rootResultRelInfo);
424 : Assert(dispatch->indexes[partidx] >= 0 &&
425 : dispatch->indexes[partidx] < proute->num_dispatch);
426 :
427 603 : rri = proute->nonleaf_partitions[dispatch->indexes[partidx]];
428 603 : dispatch = subdispatch;
429 : }
430 :
431 : /*
432 : * Convert the tuple to the new parent's layout, if different from
433 : * the previous parent.
434 : */
435 58276 : if (dispatch->tupslot)
436 : {
437 30861 : AttrMap *map = dispatch->tupmap;
438 30861 : TupleTableSlot *tempslot = myslot;
439 :
440 30861 : myslot = dispatch->tupslot;
441 30861 : slot = execute_attr_map_slot(map, slot, myslot);
442 :
443 30861 : if (tempslot != NULL)
444 147 : ExecClearTuple(tempslot);
445 : }
446 : }
447 :
448 : /*
449 : * If this partition is the default one, we must check its partition
450 : * constraint now, which may have changed concurrently due to
451 : * partitions being added to the parent.
452 : *
453 : * (We do this here, and do not rely on ExecInsert doing it, because
454 : * we don't want to miss doing it for non-leaf partitions.)
455 : */
456 574909 : if (partidx == partdesc->boundinfo->default_index)
457 : {
458 : /*
459 : * The tuple must match the partition's layout for the constraint
460 : * expression to be evaluated successfully. If the partition is
461 : * sub-partitioned, that would already be the case due to the code
462 : * above, but for a leaf partition the tuple still matches the
463 : * parent's layout.
464 : *
465 : * Note that we have a map to convert from root to current
466 : * partition, but not from immediate parent to current partition.
467 : * So if we have to convert, do it from the root slot; if not, use
468 : * the root slot as-is.
469 : */
470 416 : if (is_leaf)
471 : {
472 394 : TupleConversionMap *map = ExecGetRootToChildMap(rri, estate);
473 :
474 394 : if (map)
475 81 : slot = execute_attr_map_slot(map->attrMap, rootslot,
476 : rri->ri_PartitionTupleSlot);
477 : else
478 313 : slot = rootslot;
479 : }
480 :
481 416 : ExecPartitionCheck(rri, slot, estate, true);
482 : }
483 : }
484 :
485 : /* Release the tuple in the lowest parent's dedicated slot. */
486 516633 : if (myslot != NULL)
487 30695 : ExecClearTuple(myslot);
488 : /* and restore ecxt's scantuple */
489 516633 : ecxt->ecxt_scantuple = ecxt_scantuple_saved;
490 516633 : MemoryContextSwitchTo(oldcxt);
491 :
492 516633 : return rri;
493 : }
494 :
495 : /*
496 : * IsIndexCompatibleAsArbiter
497 : * Return true if two indexes are identical for INSERT ON CONFLICT
498 : * purposes.
499 : *
500 : * Only indexes of the same relation are supported.
501 : */
502 : static bool
503 19 : IsIndexCompatibleAsArbiter(Relation arbiterIndexRelation,
504 : IndexInfo *arbiterIndexInfo,
505 : Relation indexRelation,
506 : IndexInfo *indexInfo)
507 : {
508 : Assert(arbiterIndexRelation->rd_index->indrelid == indexRelation->rd_index->indrelid);
509 :
510 : /* must match whether they're unique */
511 19 : if (arbiterIndexInfo->ii_Unique != indexInfo->ii_Unique)
512 0 : return false;
513 :
514 : /* No support currently for comparing exclusion indexes. */
515 19 : if (arbiterIndexInfo->ii_ExclusionOps != NULL ||
516 19 : indexInfo->ii_ExclusionOps != NULL)
517 0 : return false;
518 :
519 : /* the "nulls not distinct" criterion must match */
520 19 : if (arbiterIndexInfo->ii_NullsNotDistinct !=
521 19 : indexInfo->ii_NullsNotDistinct)
522 0 : return false;
523 :
524 : /* number of key attributes must match */
525 19 : if (arbiterIndexInfo->ii_NumIndexKeyAttrs !=
526 19 : indexInfo->ii_NumIndexKeyAttrs)
527 0 : return false;
528 :
529 26 : for (int i = 0; i < arbiterIndexInfo->ii_NumIndexKeyAttrs; i++)
530 : {
531 19 : if (arbiterIndexRelation->rd_indcollation[i] !=
532 19 : indexRelation->rd_indcollation[i])
533 12 : return false;
534 :
535 7 : if (arbiterIndexRelation->rd_opfamily[i] !=
536 7 : indexRelation->rd_opfamily[i])
537 0 : return false;
538 :
539 7 : if (arbiterIndexRelation->rd_index->indkey.values[i] !=
540 7 : indexRelation->rd_index->indkey.values[i])
541 0 : return false;
542 : }
543 :
544 7 : if (list_difference(RelationGetIndexExpressions(arbiterIndexRelation),
545 7 : RelationGetIndexExpressions(indexRelation)) != NIL)
546 0 : return false;
547 :
548 7 : if (list_difference(RelationGetIndexPredicate(arbiterIndexRelation),
549 7 : RelationGetIndexPredicate(indexRelation)) != NIL)
550 0 : return false;
551 7 : return true;
552 : }
553 :
554 : /*
555 : * ExecInitPartitionInfo
556 : * Lock the partition and initialize ResultRelInfo. Also setup other
557 : * information for the partition and store it in the next empty slot in
558 : * the proute->partitions array.
559 : *
560 : * Returns the ResultRelInfo
561 : */
562 : static ResultRelInfo *
563 3706 : ExecInitPartitionInfo(ModifyTableState *mtstate, EState *estate,
564 : PartitionTupleRouting *proute,
565 : PartitionDispatch dispatch,
566 : ResultRelInfo *rootResultRelInfo,
567 : int partidx)
568 : {
569 3706 : ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
570 3706 : Oid partOid = dispatch->partdesc->oids[partidx];
571 : Relation partrel;
572 3706 : int firstVarno = mtstate->resultRelInfo[0].ri_RangeTableIndex;
573 3706 : Relation firstResultRel = mtstate->resultRelInfo[0].ri_RelationDesc;
574 : ResultRelInfo *leaf_part_rri;
575 : MemoryContext oldcxt;
576 3706 : AttrMap *part_attmap = NULL;
577 : bool found_whole_row;
578 :
579 3706 : oldcxt = MemoryContextSwitchTo(proute->memcxt);
580 :
581 3706 : partrel = table_open(partOid, RowExclusiveLock);
582 :
583 3706 : leaf_part_rri = makeNode(ResultRelInfo);
584 3706 : InitResultRelInfo(leaf_part_rri,
585 : partrel,
586 : 0,
587 : rootResultRelInfo,
588 : estate->es_instrument);
589 :
590 : /*
591 : * Verify result relation is a valid target for an INSERT. An UPDATE of a
592 : * partition-key becomes a DELETE+INSERT operation, so this check is still
593 : * required when the operation is CMD_UPDATE.
594 : */
595 3706 : CheckValidResultRel(leaf_part_rri, CMD_INSERT,
596 : node ? node->onConflictAction : ONCONFLICT_NONE, NIL);
597 :
598 : /*
599 : * Open partition indices. The user may have asked to check for conflicts
600 : * within this leaf partition and do "nothing" instead of throwing an
601 : * error. Be prepared in that case by initializing the index information
602 : * needed by ExecInsert() to perform speculative insertions.
603 : */
604 3700 : if (partrel->rd_rel->relhasindex &&
605 1109 : leaf_part_rri->ri_IndexRelationDescs == NULL)
606 1109 : ExecOpenIndices(leaf_part_rri,
607 2109 : (node != NULL &&
608 2109 : node->onConflictAction != ONCONFLICT_NONE));
609 :
610 : /*
611 : * Build WITH CHECK OPTION constraints for the partition. Note that we
612 : * didn't build the withCheckOptionList for partitions within the planner,
613 : * but simple translation of varattnos will suffice. This only occurs for
614 : * the INSERT case or in the case of UPDATE/MERGE tuple routing where we
615 : * didn't find a result rel to reuse.
616 : */
617 3700 : if (node && node->withCheckOptionLists != NIL)
618 : {
619 : List *wcoList;
620 48 : List *wcoExprs = NIL;
621 : ListCell *ll;
622 :
623 : /*
624 : * In the case of INSERT on a partitioned table, there is only one
625 : * plan. Likewise, there is only one WCO list, not one per partition.
626 : * For UPDATE/MERGE, there are as many WCO lists as there are plans.
627 : */
628 : Assert((node->operation == CMD_INSERT &&
629 : list_length(node->withCheckOptionLists) == 1 &&
630 : list_length(node->resultRelations) == 1) ||
631 : (node->operation == CMD_UPDATE &&
632 : list_length(node->withCheckOptionLists) ==
633 : list_length(node->resultRelations)) ||
634 : (node->operation == CMD_MERGE &&
635 : list_length(node->withCheckOptionLists) ==
636 : list_length(node->resultRelations)));
637 :
638 : /*
639 : * Use the WCO list of the first plan as a reference to calculate
640 : * attno's for the WCO list of this partition. In the INSERT case,
641 : * that refers to the root partitioned table, whereas in the UPDATE
642 : * tuple routing case, that refers to the first partition in the
643 : * mtstate->resultRelInfo array. In any case, both that relation and
644 : * this partition should have the same columns, so we should be able
645 : * to map attributes successfully.
646 : */
647 48 : wcoList = linitial(node->withCheckOptionLists);
648 :
649 : /*
650 : * Convert Vars in it to contain this partition's attribute numbers.
651 : */
652 : part_attmap =
653 48 : build_attrmap_by_name(RelationGetDescr(partrel),
654 : RelationGetDescr(firstResultRel),
655 : false);
656 : wcoList = (List *)
657 48 : map_variable_attnos((Node *) wcoList,
658 : firstVarno, 0,
659 : part_attmap,
660 48 : RelationGetForm(partrel)->reltype,
661 : &found_whole_row);
662 : /* We ignore the value of found_whole_row. */
663 :
664 135 : foreach(ll, wcoList)
665 : {
666 87 : WithCheckOption *wco = lfirst_node(WithCheckOption, ll);
667 87 : ExprState *wcoExpr = ExecInitQual(castNode(List, wco->qual),
668 : &mtstate->ps);
669 :
670 87 : wcoExprs = lappend(wcoExprs, wcoExpr);
671 : }
672 :
673 48 : leaf_part_rri->ri_WithCheckOptions = wcoList;
674 48 : leaf_part_rri->ri_WithCheckOptionExprs = wcoExprs;
675 : }
676 :
677 : /*
678 : * Build the RETURNING projection for the partition. Note that we didn't
679 : * build the returningList for partitions within the planner, but simple
680 : * translation of varattnos will suffice. This only occurs for the INSERT
681 : * case or in the case of UPDATE/MERGE tuple routing where we didn't find
682 : * a result rel to reuse.
683 : */
684 3700 : if (node && node->returningLists != NIL)
685 : {
686 : TupleTableSlot *slot;
687 : ExprContext *econtext;
688 : List *returningList;
689 :
690 : /* See the comment above for WCO lists. */
691 : Assert((node->operation == CMD_INSERT &&
692 : list_length(node->returningLists) == 1 &&
693 : list_length(node->resultRelations) == 1) ||
694 : (node->operation == CMD_UPDATE &&
695 : list_length(node->returningLists) ==
696 : list_length(node->resultRelations)) ||
697 : (node->operation == CMD_MERGE &&
698 : list_length(node->returningLists) ==
699 : list_length(node->resultRelations)));
700 :
701 : /*
702 : * Use the RETURNING list of the first plan as a reference to
703 : * calculate attno's for the RETURNING list of this partition. See
704 : * the comment above for WCO lists for more details on why this is
705 : * okay.
706 : */
707 148 : returningList = linitial(node->returningLists);
708 :
709 : /*
710 : * Convert Vars in it to contain this partition's attribute numbers.
711 : */
712 148 : if (part_attmap == NULL)
713 : part_attmap =
714 148 : build_attrmap_by_name(RelationGetDescr(partrel),
715 : RelationGetDescr(firstResultRel),
716 : false);
717 : returningList = (List *)
718 148 : map_variable_attnos((Node *) returningList,
719 : firstVarno, 0,
720 : part_attmap,
721 148 : RelationGetForm(partrel)->reltype,
722 : &found_whole_row);
723 : /* We ignore the value of found_whole_row. */
724 :
725 148 : leaf_part_rri->ri_returningList = returningList;
726 :
727 : /*
728 : * Initialize the projection itself.
729 : *
730 : * Use the slot and the expression context that would have been set up
731 : * in ExecInitModifyTable() for projection's output.
732 : */
733 : Assert(mtstate->ps.ps_ResultTupleSlot != NULL);
734 148 : slot = mtstate->ps.ps_ResultTupleSlot;
735 : Assert(mtstate->ps.ps_ExprContext != NULL);
736 148 : econtext = mtstate->ps.ps_ExprContext;
737 148 : leaf_part_rri->ri_projectReturning =
738 148 : ExecBuildProjectionInfo(returningList, econtext, slot,
739 : &mtstate->ps, RelationGetDescr(partrel));
740 : }
741 :
742 : /* Set up information needed for routing tuples to the partition. */
743 3700 : ExecInitRoutingInfo(mtstate, estate, proute, dispatch,
744 : leaf_part_rri, partidx, false);
745 :
746 : /*
747 : * If there is an ON CONFLICT clause, initialize state for it.
748 : */
749 3700 : if (node && node->onConflictAction != ONCONFLICT_NONE)
750 : {
751 169 : TupleDesc partrelDesc = RelationGetDescr(partrel);
752 169 : ExprContext *econtext = mtstate->ps.ps_ExprContext;
753 169 : List *arbiterIndexes = NIL;
754 169 : int additional_arbiters = 0;
755 :
756 : /*
757 : * If there is a list of arbiter indexes, map it to a list of indexes
758 : * in the partition. We also add any "identical indexes" to any of
759 : * those, to cover the case where one of them is concurrently being
760 : * reindexed.
761 : */
762 169 : if (rootResultRelInfo->ri_onConflictArbiterIndexes != NIL)
763 : {
764 141 : List *unparented_idxs = NIL,
765 141 : *arbiters_listidxs = NIL,
766 141 : *ancestors_seen = NIL;
767 :
768 301 : for (int listidx = 0; listidx < leaf_part_rri->ri_NumIndices; listidx++)
769 : {
770 : Oid indexoid;
771 : List *ancestors;
772 :
773 : /*
774 : * If one of this index's ancestors is in the root's arbiter
775 : * list, then use this index as arbiter for this partition.
776 : * Otherwise, if this index has no parent, track it for later,
777 : * in case REINDEX CONCURRENTLY is working on one of the
778 : * arbiters.
779 : *
780 : * However, if two indexes appear to have the same parent,
781 : * treat the second of these as if it had no parent. This
782 : * sounds counterintuitive, but it can happen if a transaction
783 : * running REINDEX CONCURRENTLY commits right between those
784 : * two indexes are checked by another process in this loop.
785 : * This will have the effect of also treating that second
786 : * index as arbiter.
787 : *
788 : * XXX get_partition_ancestors scans pg_inherits, which is not
789 : * only slow, but also means the catalog snapshot can get
790 : * invalidated each time through the loop (cf.
791 : * GetNonHistoricCatalogSnapshot). Consider a syscache or
792 : * some other way to cache?
793 : */
794 160 : indexoid = RelationGetRelid(leaf_part_rri->ri_IndexRelationDescs[listidx]);
795 160 : ancestors = get_partition_ancestors(indexoid);
796 160 : INJECTION_POINT("exec-init-partition-after-get-partition-ancestors", NULL);
797 :
798 160 : if (ancestors != NIL &&
799 142 : !list_member_oid(ancestors_seen, linitial_oid(ancestors)))
800 : {
801 282 : foreach_oid(parent_idx, rootResultRelInfo->ri_onConflictArbiterIndexes)
802 : {
803 141 : if (list_member_oid(ancestors, parent_idx))
804 : {
805 141 : ancestors_seen = lappend_oid(ancestors_seen, linitial_oid(ancestors));
806 141 : arbiterIndexes = lappend_oid(arbiterIndexes, indexoid);
807 141 : arbiters_listidxs = lappend_int(arbiters_listidxs, listidx);
808 141 : break;
809 : }
810 : }
811 : }
812 : else
813 19 : unparented_idxs = lappend_int(unparented_idxs, listidx);
814 :
815 160 : list_free(ancestors);
816 : }
817 :
818 : /*
819 : * If we found any indexes with no ancestors, it's possible that
820 : * some arbiter index is undergoing concurrent reindex. Match all
821 : * unparented indexes against arbiters; add unparented matching
822 : * ones as "additional arbiters".
823 : *
824 : * This is critical so that all concurrent transactions use the
825 : * same set as arbiters during REINDEX CONCURRENTLY, to avoid
826 : * spurious "duplicate key" errors.
827 : */
828 141 : if (unparented_idxs && arbiterIndexes)
829 : {
830 57 : foreach_int(unparented_i, unparented_idxs)
831 : {
832 : Relation unparented_rel;
833 : IndexInfo *unparented_ii;
834 :
835 19 : unparented_rel = leaf_part_rri->ri_IndexRelationDescs[unparented_i];
836 19 : unparented_ii = leaf_part_rri->ri_IndexRelationInfo[unparented_i];
837 :
838 : Assert(!list_member_oid(arbiterIndexes,
839 : unparented_rel->rd_index->indexrelid));
840 :
841 : /* Ignore indexes not ready */
842 19 : if (!unparented_ii->ii_ReadyForInserts)
843 0 : continue;
844 :
845 50 : foreach_int(arbiter_i, arbiters_listidxs)
846 : {
847 : Relation arbiter_rel;
848 : IndexInfo *arbiter_ii;
849 :
850 19 : arbiter_rel = leaf_part_rri->ri_IndexRelationDescs[arbiter_i];
851 19 : arbiter_ii = leaf_part_rri->ri_IndexRelationInfo[arbiter_i];
852 :
853 : /*
854 : * If the non-ancestor index is compatible with the
855 : * arbiter, use the non-ancestor as arbiter too.
856 : */
857 19 : if (IsIndexCompatibleAsArbiter(arbiter_rel,
858 : arbiter_ii,
859 : unparented_rel,
860 : unparented_ii))
861 : {
862 7 : arbiterIndexes = lappend_oid(arbiterIndexes,
863 7 : unparented_rel->rd_index->indexrelid);
864 7 : additional_arbiters++;
865 7 : break;
866 : }
867 : }
868 : }
869 : }
870 141 : list_free(unparented_idxs);
871 141 : list_free(arbiters_listidxs);
872 141 : list_free(ancestors_seen);
873 : }
874 :
875 : /*
876 : * We expect to find as many arbiter indexes on this partition as the
877 : * root has, plus however many "additional arbiters" (to wit: those
878 : * being concurrently rebuilt) we found.
879 : */
880 169 : if (list_length(rootResultRelInfo->ri_onConflictArbiterIndexes) !=
881 169 : list_length(arbiterIndexes) - additional_arbiters)
882 0 : elog(ERROR, "invalid arbiter index list");
883 169 : leaf_part_rri->ri_onConflictArbiterIndexes = arbiterIndexes;
884 :
885 : /*
886 : * In the DO UPDATE and DO SELECT cases, we have some more state to
887 : * initialize.
888 : */
889 169 : if (node->onConflictAction == ONCONFLICT_UPDATE ||
890 76 : node->onConflictAction == ONCONFLICT_SELECT)
891 : {
892 135 : OnConflictActionState *onconfl = makeNode(OnConflictActionState);
893 : TupleConversionMap *map;
894 :
895 135 : map = ExecGetRootToChildMap(leaf_part_rri, estate);
896 :
897 : Assert(node->onConflictSet != NIL ||
898 : node->onConflictAction == ONCONFLICT_SELECT);
899 : Assert(rootResultRelInfo->ri_onConflict != NULL);
900 :
901 135 : leaf_part_rri->ri_onConflict = onconfl;
902 :
903 : /* Lock strength for DO SELECT [FOR UPDATE/SHARE] */
904 135 : onconfl->oc_LockStrength =
905 135 : rootResultRelInfo->ri_onConflict->oc_LockStrength;
906 :
907 : /*
908 : * Need a separate existing slot for each partition, as the
909 : * partition could be of a different AM, even if the tuple
910 : * descriptors match.
911 : */
912 135 : onconfl->oc_Existing =
913 135 : table_slot_create(leaf_part_rri->ri_RelationDesc,
914 135 : &mtstate->ps.state->es_tupleTable);
915 :
916 : /*
917 : * If the partition's tuple descriptor matches exactly the root
918 : * parent (the common case), we can re-use most of the parent's ON
919 : * CONFLICT action state, skipping a bunch of work. Otherwise, we
920 : * need to create state specific to this partition.
921 : */
922 135 : if (map == NULL)
923 : {
924 : /*
925 : * It's safe to reuse these from the partition root, as we
926 : * only process one tuple at a time (therefore we won't
927 : * overwrite needed data in slots), and the results of any
928 : * projections are independent of the underlying storage.
929 : * Projections and where clauses themselves don't store state
930 : * / are independent of the underlying storage.
931 : */
932 73 : onconfl->oc_ProjSlot =
933 73 : rootResultRelInfo->ri_onConflict->oc_ProjSlot;
934 73 : onconfl->oc_ProjInfo =
935 73 : rootResultRelInfo->ri_onConflict->oc_ProjInfo;
936 73 : onconfl->oc_WhereClause =
937 73 : rootResultRelInfo->ri_onConflict->oc_WhereClause;
938 : }
939 : else
940 : {
941 : /*
942 : * For ON CONFLICT DO UPDATE, translate expressions in
943 : * onConflictSet to account for different attribute numbers.
944 : * For that, map partition varattnos twice: first to catch the
945 : * EXCLUDED pseudo-relation (INNER_VAR), and second to handle
946 : * the main target relation (firstVarno).
947 : */
948 62 : if (node->onConflictAction == ONCONFLICT_UPDATE)
949 : {
950 : List *onconflset;
951 : List *onconflcols;
952 :
953 38 : onconflset = copyObject(node->onConflictSet);
954 38 : if (part_attmap == NULL)
955 : part_attmap =
956 35 : build_attrmap_by_name(RelationGetDescr(partrel),
957 : RelationGetDescr(firstResultRel),
958 : false);
959 : onconflset = (List *)
960 38 : map_variable_attnos((Node *) onconflset,
961 : INNER_VAR, 0,
962 : part_attmap,
963 38 : RelationGetForm(partrel)->reltype,
964 : &found_whole_row);
965 : /* We ignore the value of found_whole_row. */
966 : onconflset = (List *)
967 38 : map_variable_attnos((Node *) onconflset,
968 : firstVarno, 0,
969 : part_attmap,
970 38 : RelationGetForm(partrel)->reltype,
971 : &found_whole_row);
972 : /* We ignore the value of found_whole_row. */
973 :
974 : /*
975 : * Finally, adjust the target colnos to match the
976 : * partition.
977 : */
978 38 : onconflcols = adjust_partition_colnos(node->onConflictCols,
979 : leaf_part_rri);
980 :
981 : /* create the tuple slot for the UPDATE SET projection */
982 38 : onconfl->oc_ProjSlot =
983 38 : table_slot_create(partrel,
984 38 : &mtstate->ps.state->es_tupleTable);
985 :
986 : /* build UPDATE SET projection state */
987 38 : onconfl->oc_ProjInfo =
988 38 : ExecBuildUpdateProjection(onconflset,
989 : true,
990 : onconflcols,
991 : partrelDesc,
992 : econtext,
993 : onconfl->oc_ProjSlot,
994 : &mtstate->ps);
995 : }
996 :
997 : /*
998 : * For both ON CONFLICT DO UPDATE and ON CONFLICT DO SELECT,
999 : * there may be a WHERE clause. If so, initialize state where
1000 : * it will be evaluated, mapping the attribute numbers
1001 : * appropriately. As with onConflictSet, we need to map
1002 : * partition varattnos twice, to catch both the EXCLUDED
1003 : * pseudo-relation (INNER_VAR), and the main target relation
1004 : * (firstVarno).
1005 : */
1006 62 : if (node->onConflictWhere)
1007 : {
1008 : List *clause;
1009 :
1010 27 : if (part_attmap == NULL)
1011 : part_attmap =
1012 0 : build_attrmap_by_name(RelationGetDescr(partrel),
1013 : RelationGetDescr(firstResultRel),
1014 : false);
1015 :
1016 27 : clause = copyObject((List *) node->onConflictWhere);
1017 : clause = (List *)
1018 27 : map_variable_attnos((Node *) clause,
1019 : INNER_VAR, 0,
1020 : part_attmap,
1021 27 : RelationGetForm(partrel)->reltype,
1022 : &found_whole_row);
1023 : /* We ignore the value of found_whole_row. */
1024 : clause = (List *)
1025 27 : map_variable_attnos((Node *) clause,
1026 : firstVarno, 0,
1027 : part_attmap,
1028 27 : RelationGetForm(partrel)->reltype,
1029 : &found_whole_row);
1030 : /* We ignore the value of found_whole_row. */
1031 27 : onconfl->oc_WhereClause =
1032 27 : ExecInitQual(clause, &mtstate->ps);
1033 : }
1034 : }
1035 : }
1036 : }
1037 :
1038 : /*
1039 : * Since we've just initialized this ResultRelInfo, it's not in any list
1040 : * attached to the estate as yet. Add it, so that it can be found later.
1041 : *
1042 : * Note that the entries in this list appear in no predetermined order,
1043 : * because partition result rels are initialized as and when they're
1044 : * needed.
1045 : */
1046 3700 : MemoryContextSwitchTo(estate->es_query_cxt);
1047 3700 : estate->es_tuple_routing_result_relations =
1048 3700 : lappend(estate->es_tuple_routing_result_relations,
1049 : leaf_part_rri);
1050 :
1051 : /*
1052 : * Initialize information about this partition that's needed to handle
1053 : * MERGE. We take the "first" result relation's mergeActionList as
1054 : * reference and make copy for this relation, converting stuff that
1055 : * references attribute numbers to match this relation's.
1056 : *
1057 : * This duplicates much of the logic in ExecInitMerge(), so if something
1058 : * changes there, look here too.
1059 : */
1060 3700 : if (node && node->operation == CMD_MERGE)
1061 : {
1062 12 : List *firstMergeActionList = linitial(node->mergeActionLists);
1063 : ListCell *lc;
1064 12 : ExprContext *econtext = mtstate->ps.ps_ExprContext;
1065 : Node *joinCondition;
1066 :
1067 12 : if (part_attmap == NULL)
1068 : part_attmap =
1069 6 : build_attrmap_by_name(RelationGetDescr(partrel),
1070 : RelationGetDescr(firstResultRel),
1071 : false);
1072 :
1073 12 : if (unlikely(!leaf_part_rri->ri_projectNewInfoValid))
1074 12 : ExecInitMergeTupleSlots(mtstate, leaf_part_rri);
1075 :
1076 : /* Initialize state for join condition checking. */
1077 : joinCondition =
1078 12 : map_variable_attnos(linitial(node->mergeJoinConditions),
1079 : firstVarno, 0,
1080 : part_attmap,
1081 12 : RelationGetForm(partrel)->reltype,
1082 : &found_whole_row);
1083 : /* We ignore the value of found_whole_row. */
1084 12 : leaf_part_rri->ri_MergeJoinCondition =
1085 12 : ExecInitQual((List *) joinCondition, &mtstate->ps);
1086 :
1087 30 : foreach(lc, firstMergeActionList)
1088 : {
1089 : /* Make a copy for this relation to be safe. */
1090 18 : MergeAction *action = copyObject(lfirst(lc));
1091 : MergeActionState *action_state;
1092 :
1093 : /* Generate the action's state for this relation */
1094 18 : action_state = makeNode(MergeActionState);
1095 18 : action_state->mas_action = action;
1096 :
1097 : /* And put the action in the appropriate list */
1098 36 : leaf_part_rri->ri_MergeActions[action->matchKind] =
1099 18 : lappend(leaf_part_rri->ri_MergeActions[action->matchKind],
1100 : action_state);
1101 :
1102 18 : switch (action->commandType)
1103 : {
1104 6 : case CMD_INSERT:
1105 :
1106 : /*
1107 : * ExecCheckPlanOutput() already done on the targetlist
1108 : * when "first" result relation initialized and it is same
1109 : * for all result relations.
1110 : */
1111 6 : action_state->mas_proj =
1112 6 : ExecBuildProjectionInfo(action->targetList, econtext,
1113 : leaf_part_rri->ri_newTupleSlot,
1114 : &mtstate->ps,
1115 : RelationGetDescr(partrel));
1116 6 : break;
1117 9 : case CMD_UPDATE:
1118 :
1119 : /*
1120 : * Convert updateColnos from "first" result relation
1121 : * attribute numbers to this result rel's.
1122 : */
1123 9 : if (part_attmap)
1124 9 : action->updateColnos =
1125 9 : adjust_partition_colnos_using_map(action->updateColnos,
1126 : part_attmap);
1127 9 : action_state->mas_proj =
1128 9 : ExecBuildUpdateProjection(action->targetList,
1129 : true,
1130 : action->updateColnos,
1131 9 : RelationGetDescr(leaf_part_rri->ri_RelationDesc),
1132 : econtext,
1133 : leaf_part_rri->ri_newTupleSlot,
1134 : NULL);
1135 9 : break;
1136 3 : case CMD_DELETE:
1137 : case CMD_NOTHING:
1138 : /* Nothing to do */
1139 3 : break;
1140 :
1141 0 : default:
1142 0 : elog(ERROR, "unknown action in MERGE WHEN clause");
1143 : }
1144 :
1145 : /* found_whole_row intentionally ignored. */
1146 18 : action->qual =
1147 18 : map_variable_attnos(action->qual,
1148 : firstVarno, 0,
1149 : part_attmap,
1150 18 : RelationGetForm(partrel)->reltype,
1151 : &found_whole_row);
1152 18 : action_state->mas_whenqual =
1153 18 : ExecInitQual((List *) action->qual, &mtstate->ps);
1154 : }
1155 : }
1156 3700 : MemoryContextSwitchTo(oldcxt);
1157 :
1158 3700 : return leaf_part_rri;
1159 : }
1160 :
1161 : /*
1162 : * ExecInitRoutingInfo
1163 : * Set up information needed for translating tuples between root
1164 : * partitioned table format and partition format, and keep track of it
1165 : * in PartitionTupleRouting.
1166 : */
1167 : static void
1168 3954 : ExecInitRoutingInfo(ModifyTableState *mtstate,
1169 : EState *estate,
1170 : PartitionTupleRouting *proute,
1171 : PartitionDispatch dispatch,
1172 : ResultRelInfo *partRelInfo,
1173 : int partidx,
1174 : bool is_borrowed_rel)
1175 : {
1176 : MemoryContext oldcxt;
1177 : int rri_index;
1178 :
1179 3954 : oldcxt = MemoryContextSwitchTo(proute->memcxt);
1180 :
1181 : /*
1182 : * Set up tuple conversion between root parent and the partition if the
1183 : * two have different rowtypes. If conversion is indeed required, also
1184 : * initialize a slot dedicated to storing this partition's converted
1185 : * tuples. Various operations that are applied to tuples after routing,
1186 : * such as checking constraints, will refer to this slot.
1187 : */
1188 3954 : if (ExecGetRootToChildMap(partRelInfo, estate) != NULL)
1189 : {
1190 701 : Relation partrel = partRelInfo->ri_RelationDesc;
1191 :
1192 : /*
1193 : * This pins the partition's TupleDesc, which will be released at the
1194 : * end of the command.
1195 : */
1196 701 : partRelInfo->ri_PartitionTupleSlot =
1197 701 : table_slot_create(partrel, &estate->es_tupleTable);
1198 : }
1199 : else
1200 3253 : partRelInfo->ri_PartitionTupleSlot = NULL;
1201 :
1202 : /*
1203 : * If the partition is a foreign table, let the FDW init itself for
1204 : * routing tuples to the partition.
1205 : */
1206 3954 : if (partRelInfo->ri_FdwRoutine != NULL &&
1207 46 : partRelInfo->ri_FdwRoutine->BeginForeignInsert != NULL)
1208 46 : partRelInfo->ri_FdwRoutine->BeginForeignInsert(mtstate, partRelInfo);
1209 :
1210 : /*
1211 : * Determine if the FDW supports batch insert and determine the batch size
1212 : * (a FDW may support batching, but it may be disabled for the
1213 : * server/table or for this particular query).
1214 : *
1215 : * If the FDW does not support batching, we set the batch size to 1.
1216 : */
1217 3948 : if (partRelInfo->ri_FdwRoutine != NULL &&
1218 40 : partRelInfo->ri_FdwRoutine->GetForeignModifyBatchSize &&
1219 40 : partRelInfo->ri_FdwRoutine->ExecForeignBatchInsert)
1220 40 : partRelInfo->ri_BatchSize =
1221 40 : partRelInfo->ri_FdwRoutine->GetForeignModifyBatchSize(partRelInfo);
1222 : else
1223 3908 : partRelInfo->ri_BatchSize = 1;
1224 :
1225 : Assert(partRelInfo->ri_BatchSize >= 1);
1226 :
1227 3948 : partRelInfo->ri_CopyMultiInsertBuffer = NULL;
1228 :
1229 : /*
1230 : * Keep track of it in the PartitionTupleRouting->partitions array.
1231 : */
1232 : Assert(dispatch->indexes[partidx] == -1);
1233 :
1234 3948 : rri_index = proute->num_partitions++;
1235 :
1236 : /* Allocate or enlarge the array, as needed */
1237 3948 : if (proute->num_partitions >= proute->max_partitions)
1238 : {
1239 2728 : if (proute->max_partitions == 0)
1240 : {
1241 2722 : proute->max_partitions = 8;
1242 2722 : proute->partitions = palloc_array(ResultRelInfo *, proute->max_partitions);
1243 2722 : proute->is_borrowed_rel = palloc_array(bool, proute->max_partitions);
1244 : }
1245 : else
1246 : {
1247 6 : proute->max_partitions *= 2;
1248 6 : proute->partitions = (ResultRelInfo **)
1249 6 : repalloc(proute->partitions, sizeof(ResultRelInfo *) *
1250 6 : proute->max_partitions);
1251 6 : proute->is_borrowed_rel = (bool *)
1252 6 : repalloc(proute->is_borrowed_rel, sizeof(bool) *
1253 6 : proute->max_partitions);
1254 : }
1255 : }
1256 :
1257 3948 : proute->partitions[rri_index] = partRelInfo;
1258 3948 : proute->is_borrowed_rel[rri_index] = is_borrowed_rel;
1259 3948 : dispatch->indexes[partidx] = rri_index;
1260 :
1261 3948 : MemoryContextSwitchTo(oldcxt);
1262 3948 : }
1263 :
1264 : /*
1265 : * ExecInitPartitionDispatchInfo
1266 : * Lock the partitioned table (if not locked already) and initialize
1267 : * PartitionDispatch for a partitioned table and store it in the next
1268 : * available slot in the proute->partition_dispatch_info array. Also,
1269 : * record the index into this array in the parent_pd->indexes[] array in
1270 : * the partidx element so that we can properly retrieve the newly created
1271 : * PartitionDispatch later.
1272 : */
1273 : static PartitionDispatch
1274 3488 : ExecInitPartitionDispatchInfo(EState *estate,
1275 : PartitionTupleRouting *proute, Oid partoid,
1276 : PartitionDispatch parent_pd, int partidx,
1277 : ResultRelInfo *rootResultRelInfo)
1278 : {
1279 : Relation rel;
1280 : PartitionDesc partdesc;
1281 : PartitionDispatch pd;
1282 : int dispatchidx;
1283 : MemoryContext oldcxt;
1284 :
1285 : /*
1286 : * For data modification, it is better that executor does not include
1287 : * partitions being detached, except when running in snapshot-isolation
1288 : * mode. This means that a read-committed transaction immediately gets a
1289 : * "no partition for tuple" error when a tuple is inserted into a
1290 : * partition that's being detached concurrently, but a transaction in
1291 : * repeatable-read mode can still use such a partition.
1292 : */
1293 3488 : if (estate->es_partition_directory == NULL)
1294 2867 : estate->es_partition_directory =
1295 2867 : CreatePartitionDirectory(estate->es_query_cxt,
1296 : !IsolationUsesXactSnapshot());
1297 :
1298 3488 : oldcxt = MemoryContextSwitchTo(proute->memcxt);
1299 :
1300 : /*
1301 : * Only sub-partitioned tables need to be locked here. The root
1302 : * partitioned table will already have been locked as it's referenced in
1303 : * the query's rtable.
1304 : */
1305 3488 : if (partoid != RelationGetRelid(proute->partition_root))
1306 603 : rel = table_open(partoid, RowExclusiveLock);
1307 : else
1308 2885 : rel = proute->partition_root;
1309 3488 : partdesc = PartitionDirectoryLookup(estate->es_partition_directory, rel);
1310 :
1311 3488 : pd = (PartitionDispatch) palloc(offsetof(PartitionDispatchData, indexes) +
1312 3488 : partdesc->nparts * sizeof(int));
1313 3488 : pd->reldesc = rel;
1314 3488 : pd->key = RelationGetPartitionKey(rel);
1315 3488 : pd->keystate = NIL;
1316 3488 : pd->partdesc = partdesc;
1317 3488 : if (parent_pd != NULL)
1318 : {
1319 603 : TupleDesc tupdesc = RelationGetDescr(rel);
1320 :
1321 : /*
1322 : * For sub-partitioned tables where the column order differs from its
1323 : * direct parent partitioned table, we must store a tuple table slot
1324 : * initialized with its tuple descriptor and a tuple conversion map to
1325 : * convert a tuple from its parent's rowtype to its own. This is to
1326 : * make sure that we are looking at the correct row using the correct
1327 : * tuple descriptor when computing its partition key for tuple
1328 : * routing.
1329 : */
1330 603 : pd->tupmap = build_attrmap_by_name_if_req(RelationGetDescr(parent_pd->reldesc),
1331 : tupdesc,
1332 : false);
1333 603 : pd->tupslot = pd->tupmap ?
1334 603 : MakeSingleTupleTableSlot(tupdesc, &TTSOpsVirtual) : NULL;
1335 : }
1336 : else
1337 : {
1338 : /* Not required for the root partitioned table */
1339 2885 : pd->tupmap = NULL;
1340 2885 : pd->tupslot = NULL;
1341 : }
1342 :
1343 : /*
1344 : * Initialize with -1 to signify that the corresponding partition's
1345 : * ResultRelInfo or PartitionDispatch has not been created yet.
1346 : */
1347 3488 : memset(pd->indexes, -1, sizeof(int) * partdesc->nparts);
1348 :
1349 : /* Track in PartitionTupleRouting for later use */
1350 3488 : dispatchidx = proute->num_dispatch++;
1351 :
1352 : /* Allocate or enlarge the array, as needed */
1353 3488 : if (proute->num_dispatch >= proute->max_dispatch)
1354 : {
1355 2885 : if (proute->max_dispatch == 0)
1356 : {
1357 2885 : proute->max_dispatch = 4;
1358 2885 : proute->partition_dispatch_info = palloc_array(PartitionDispatch, proute->max_dispatch);
1359 2885 : proute->nonleaf_partitions = palloc_array(ResultRelInfo *, proute->max_dispatch);
1360 : }
1361 : else
1362 : {
1363 0 : proute->max_dispatch *= 2;
1364 0 : proute->partition_dispatch_info = (PartitionDispatch *)
1365 0 : repalloc(proute->partition_dispatch_info,
1366 0 : sizeof(PartitionDispatch) * proute->max_dispatch);
1367 0 : proute->nonleaf_partitions = (ResultRelInfo **)
1368 0 : repalloc(proute->nonleaf_partitions,
1369 0 : sizeof(ResultRelInfo *) * proute->max_dispatch);
1370 : }
1371 : }
1372 3488 : proute->partition_dispatch_info[dispatchidx] = pd;
1373 :
1374 : /*
1375 : * If setting up a PartitionDispatch for a sub-partitioned table, we may
1376 : * also need a minimally valid ResultRelInfo for checking the partition
1377 : * constraint later; set that up now.
1378 : */
1379 3488 : if (parent_pd)
1380 : {
1381 603 : ResultRelInfo *rri = makeNode(ResultRelInfo);
1382 :
1383 603 : InitResultRelInfo(rri, rel, 0, rootResultRelInfo, 0);
1384 603 : proute->nonleaf_partitions[dispatchidx] = rri;
1385 : }
1386 : else
1387 2885 : proute->nonleaf_partitions[dispatchidx] = NULL;
1388 :
1389 : /*
1390 : * Finally, if setting up a PartitionDispatch for a sub-partitioned table,
1391 : * install a downlink in the parent to allow quick descent.
1392 : */
1393 3488 : if (parent_pd)
1394 : {
1395 : Assert(parent_pd->indexes[partidx] == -1);
1396 603 : parent_pd->indexes[partidx] = dispatchidx;
1397 : }
1398 :
1399 3488 : MemoryContextSwitchTo(oldcxt);
1400 :
1401 3488 : return pd;
1402 : }
1403 :
1404 : /*
1405 : * ExecCleanupTupleRouting -- Clean up objects allocated for partition tuple
1406 : * routing.
1407 : *
1408 : * Close all the partitioned tables, leaf partitions, and their indices.
1409 : */
1410 : void
1411 2462 : ExecCleanupTupleRouting(ModifyTableState *mtstate,
1412 : PartitionTupleRouting *proute)
1413 : {
1414 : int i;
1415 :
1416 : /*
1417 : * Remember, proute->partition_dispatch_info[0] corresponds to the root
1418 : * partitioned table, which we must not try to close, because it is the
1419 : * main target table of the query that will be closed by callers such as
1420 : * ExecEndPlan() or DoCopy(). Also, tupslot is NULL for the root
1421 : * partitioned table.
1422 : */
1423 2953 : for (i = 1; i < proute->num_dispatch; i++)
1424 : {
1425 491 : PartitionDispatch pd = proute->partition_dispatch_info[i];
1426 :
1427 491 : table_close(pd->reldesc, NoLock);
1428 :
1429 491 : if (pd->tupslot)
1430 230 : ExecDropSingleTupleTableSlot(pd->tupslot);
1431 : }
1432 :
1433 6107 : for (i = 0; i < proute->num_partitions; i++)
1434 : {
1435 3645 : ResultRelInfo *resultRelInfo = proute->partitions[i];
1436 :
1437 : /* Allow any FDWs to shut down */
1438 3645 : if (resultRelInfo->ri_FdwRoutine != NULL &&
1439 34 : resultRelInfo->ri_FdwRoutine->EndForeignInsert != NULL)
1440 34 : resultRelInfo->ri_FdwRoutine->EndForeignInsert(mtstate->ps.state,
1441 : resultRelInfo);
1442 :
1443 : /*
1444 : * Close it if it's not one of the result relations borrowed from the
1445 : * owning ModifyTableState; those will be closed by ExecEndPlan().
1446 : */
1447 3645 : if (proute->is_borrowed_rel[i])
1448 230 : continue;
1449 :
1450 3415 : ExecCloseIndices(resultRelInfo);
1451 3415 : table_close(resultRelInfo->ri_RelationDesc, NoLock);
1452 : }
1453 2462 : }
1454 :
1455 : /* ----------------
1456 : * FormPartitionKeyDatum
1457 : * Construct values[] and isnull[] arrays for the partition key
1458 : * of a tuple.
1459 : *
1460 : * pd Partition dispatch object of the partitioned table
1461 : * slot Heap tuple from which to extract partition key
1462 : * estate executor state for evaluating any partition key
1463 : * expressions (must be non-NULL)
1464 : * values Array of partition key Datums (output area)
1465 : * isnull Array of is-null indicators (output area)
1466 : *
1467 : * the ecxt_scantuple slot of estate's per-tuple expr context must point to
1468 : * the heap tuple passed in.
1469 : * ----------------
1470 : */
1471 : static void
1472 575004 : FormPartitionKeyDatum(PartitionDispatch pd,
1473 : TupleTableSlot *slot,
1474 : EState *estate,
1475 : Datum *values,
1476 : bool *isnull)
1477 : {
1478 : ListCell *partexpr_item;
1479 : int i;
1480 :
1481 575004 : if (pd->key->partexprs != NIL && pd->keystate == NIL)
1482 : {
1483 : /* Check caller has set up context correctly */
1484 : Assert(estate != NULL &&
1485 : GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
1486 :
1487 : /* First time through, set up expression evaluation state */
1488 273 : pd->keystate = ExecPrepareExprList(pd->key->partexprs, estate);
1489 : }
1490 :
1491 575004 : partexpr_item = list_head(pd->keystate);
1492 1161546 : for (i = 0; i < pd->key->partnatts; i++)
1493 : {
1494 586542 : AttrNumber keycol = pd->key->partattrs[i];
1495 : Datum datum;
1496 : bool isNull;
1497 :
1498 586542 : if (keycol != 0)
1499 : {
1500 : /* Plain column; get the value directly from the heap tuple */
1501 542724 : datum = slot_getattr(slot, keycol, &isNull);
1502 : }
1503 : else
1504 : {
1505 : /* Expression; need to evaluate it */
1506 43818 : if (partexpr_item == NULL)
1507 0 : elog(ERROR, "wrong number of partition key expressions");
1508 43818 : datum = ExecEvalExprSwitchContext((ExprState *) lfirst(partexpr_item),
1509 43818 : GetPerTupleExprContext(estate),
1510 : &isNull);
1511 43818 : partexpr_item = lnext(pd->keystate, partexpr_item);
1512 : }
1513 586542 : values[i] = datum;
1514 586542 : isnull[i] = isNull;
1515 : }
1516 :
1517 575004 : if (partexpr_item != NULL)
1518 0 : elog(ERROR, "wrong number of partition key expressions");
1519 575004 : }
1520 :
1521 : /*
1522 : * The number of times the same partition must be found in a row before we
1523 : * switch from a binary search for the given values to just checking if the
1524 : * values belong to the last found partition. This must be above 0.
1525 : */
1526 : #define PARTITION_CACHED_FIND_THRESHOLD 16
1527 :
1528 : /*
1529 : * get_partition_for_tuple
1530 : * Finds partition of relation which accepts the partition key specified
1531 : * in values and isnull.
1532 : *
1533 : * Calling this function can be quite expensive when LIST and RANGE
1534 : * partitioned tables have many partitions. This is due to the binary search
1535 : * that's done to find the correct partition. Many of the use cases for LIST
1536 : * and RANGE partitioned tables make it likely that the same partition is
1537 : * found in subsequent ExecFindPartition() calls. This is especially true for
1538 : * cases such as RANGE partitioned tables on a TIMESTAMP column where the
1539 : * partition key is the current time. When asked to find a partition for a
1540 : * RANGE or LIST partitioned table, we record the partition index and datum
1541 : * offset we've found for the given 'values' in the PartitionDesc (which is
1542 : * stored in relcache), and if we keep finding the same partition
1543 : * PARTITION_CACHED_FIND_THRESHOLD times in a row, then we'll enable caching
1544 : * logic and instead of performing a binary search to find the correct
1545 : * partition, we'll just double-check that 'values' still belong to the last
1546 : * found partition, and if so, we'll return that partition index, thus
1547 : * skipping the need for the binary search. If we fail to match the last
1548 : * partition when double checking, then we fall back on doing a binary search.
1549 : * In this case, unless we find 'values' belong to the DEFAULT partition,
1550 : * we'll reset the number of times we've hit the same partition so that we
1551 : * don't attempt to use the cache again until we've found that partition at
1552 : * least PARTITION_CACHED_FIND_THRESHOLD times in a row.
1553 : *
1554 : * For cases where the partition changes on each lookup, the amount of
1555 : * additional work required just amounts to recording the last found partition
1556 : * and bound offset then resetting the found counter. This is cheap and does
1557 : * not appear to cause any meaningful slowdowns for such cases.
1558 : *
1559 : * No caching of partitions is done when the last found partition is the
1560 : * DEFAULT or NULL partition. For the case of the DEFAULT partition, there
1561 : * is no bound offset storing the matching datum, so we cannot confirm the
1562 : * indexes match. For the NULL partition, this is just so cheap, there's no
1563 : * sense in caching.
1564 : *
1565 : * Return value is index of the partition (>= 0 and < partdesc->nparts) if one
1566 : * found or -1 if none found.
1567 : */
1568 : static int
1569 574983 : get_partition_for_tuple(PartitionDispatch pd, const Datum *values, const bool *isnull)
1570 : {
1571 574983 : int bound_offset = -1;
1572 574983 : int part_index = -1;
1573 574983 : PartitionKey key = pd->key;
1574 574983 : PartitionDesc partdesc = pd->partdesc;
1575 574983 : PartitionBoundInfo boundinfo = partdesc->boundinfo;
1576 :
1577 : /*
1578 : * In the switch statement below, when we perform a cached lookup for
1579 : * RANGE and LIST partitioned tables, if we find that the last found
1580 : * partition matches the 'values', we return the partition index right
1581 : * away. We do this instead of breaking out of the switch as we don't
1582 : * want to execute the code about the DEFAULT partition or do any updates
1583 : * for any of the cache-related fields. That would be a waste of effort
1584 : * as we already know it's not the DEFAULT partition and have no need to
1585 : * increment the number of times we found the same partition any higher
1586 : * than PARTITION_CACHED_FIND_THRESHOLD.
1587 : */
1588 :
1589 : /* Route as appropriate based on partitioning strategy. */
1590 574983 : switch (key->strategy)
1591 : {
1592 105369 : case PARTITION_STRATEGY_HASH:
1593 : {
1594 : uint64 rowHash;
1595 :
1596 : /* hash partitioning is too cheap to bother caching */
1597 105369 : rowHash = compute_partition_hash_value(key->partnatts,
1598 : key->partsupfunc,
1599 105369 : key->partcollation,
1600 : values, isnull);
1601 :
1602 : /*
1603 : * HASH partitions can't have a DEFAULT partition and we don't
1604 : * do any caching work for them, so just return the part index
1605 : */
1606 105363 : return boundinfo->indexes[rowHash % boundinfo->nindexes];
1607 : }
1608 :
1609 85659 : case PARTITION_STRATEGY_LIST:
1610 85659 : if (isnull[0])
1611 : {
1612 : /* this is far too cheap to bother doing any caching */
1613 66 : if (partition_bound_accepts_nulls(boundinfo))
1614 : {
1615 : /*
1616 : * When there is a NULL partition we just return that
1617 : * directly. We don't have a bound_offset so it's not
1618 : * valid to drop into the code after the switch which
1619 : * checks and updates the cache fields. We perhaps should
1620 : * be invalidating the details of the last cached
1621 : * partition but there's no real need to. Keeping those
1622 : * fields set gives a chance at matching to the cached
1623 : * partition on the next lookup.
1624 : */
1625 51 : return boundinfo->null_index;
1626 : }
1627 : }
1628 : else
1629 : {
1630 : bool equal;
1631 :
1632 85593 : if (partdesc->last_found_count >= PARTITION_CACHED_FIND_THRESHOLD)
1633 : {
1634 11946 : int last_datum_offset = partdesc->last_found_datum_index;
1635 11946 : Datum lastDatum = boundinfo->datums[last_datum_offset][0];
1636 : int32 cmpval;
1637 :
1638 : /* does the last found datum index match this datum? */
1639 11946 : cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
1640 11946 : key->partcollation[0],
1641 : lastDatum,
1642 : values[0]));
1643 :
1644 11946 : if (cmpval == 0)
1645 11769 : return boundinfo->indexes[last_datum_offset];
1646 :
1647 : /* fall-through and do a manual lookup */
1648 : }
1649 :
1650 73824 : bound_offset = partition_list_bsearch(key->partsupfunc,
1651 : key->partcollation,
1652 : boundinfo,
1653 : values[0], &equal);
1654 73824 : if (bound_offset >= 0 && equal)
1655 73624 : part_index = boundinfo->indexes[bound_offset];
1656 : }
1657 73839 : break;
1658 :
1659 383955 : case PARTITION_STRATEGY_RANGE:
1660 : {
1661 383955 : bool equal = false,
1662 383955 : range_partkey_has_null = false;
1663 : int i;
1664 :
1665 : /*
1666 : * No range includes NULL, so this will be accepted by the
1667 : * default partition if there is one, and otherwise rejected.
1668 : */
1669 779238 : for (i = 0; i < key->partnatts; i++)
1670 : {
1671 395310 : if (isnull[i])
1672 : {
1673 27 : range_partkey_has_null = true;
1674 27 : break;
1675 : }
1676 : }
1677 :
1678 : /* NULLs belong in the DEFAULT partition */
1679 383955 : if (range_partkey_has_null)
1680 27 : break;
1681 :
1682 383928 : if (partdesc->last_found_count >= PARTITION_CACHED_FIND_THRESHOLD)
1683 : {
1684 124827 : int last_datum_offset = partdesc->last_found_datum_index;
1685 124827 : Datum *lastDatums = boundinfo->datums[last_datum_offset];
1686 124827 : PartitionRangeDatumKind *kind = boundinfo->kind[last_datum_offset];
1687 : int32 cmpval;
1688 :
1689 : /* check if the value is >= to the lower bound */
1690 124827 : cmpval = partition_rbound_datum_cmp(key->partsupfunc,
1691 : key->partcollation,
1692 : lastDatums,
1693 : kind,
1694 : values,
1695 124827 : key->partnatts);
1696 :
1697 : /*
1698 : * If it's equal to the lower bound then no need to check
1699 : * the upper bound.
1700 : */
1701 124827 : if (cmpval == 0)
1702 124672 : return boundinfo->indexes[last_datum_offset + 1];
1703 :
1704 121878 : if (cmpval < 0 && last_datum_offset + 1 < boundinfo->ndatums)
1705 : {
1706 : /* check if the value is below the upper bound */
1707 121848 : lastDatums = boundinfo->datums[last_datum_offset + 1];
1708 121848 : kind = boundinfo->kind[last_datum_offset + 1];
1709 121848 : cmpval = partition_rbound_datum_cmp(key->partsupfunc,
1710 : key->partcollation,
1711 : lastDatums,
1712 : kind,
1713 : values,
1714 121848 : key->partnatts);
1715 :
1716 121848 : if (cmpval > 0)
1717 121723 : return boundinfo->indexes[last_datum_offset + 1];
1718 : }
1719 : /* fall-through and do a manual lookup */
1720 : }
1721 :
1722 259256 : bound_offset = partition_range_datum_bsearch(key->partsupfunc,
1723 : key->partcollation,
1724 : boundinfo,
1725 259256 : key->partnatts,
1726 : values,
1727 : &equal);
1728 :
1729 : /*
1730 : * The bound at bound_offset is less than or equal to the
1731 : * tuple value, so the bound at offset+1 is the upper bound of
1732 : * the partition we're looking for, if there actually exists
1733 : * one.
1734 : */
1735 259256 : part_index = boundinfo->indexes[bound_offset + 1];
1736 : }
1737 259256 : break;
1738 :
1739 0 : default:
1740 0 : elog(ERROR, "unexpected partition strategy: %d",
1741 : (int) key->strategy);
1742 : }
1743 :
1744 : /*
1745 : * part_index < 0 means we failed to find a partition of this parent. Use
1746 : * the default partition, if there is one.
1747 : */
1748 333122 : if (part_index < 0)
1749 : {
1750 : /*
1751 : * No need to reset the cache fields here. The next set of values
1752 : * might end up belonging to the cached partition, so leaving the
1753 : * cache alone improves the chances of a cache hit on the next lookup.
1754 : */
1755 472 : return boundinfo->default_index;
1756 : }
1757 :
1758 : /* we should only make it here when the code above set bound_offset */
1759 : Assert(bound_offset >= 0);
1760 :
1761 : /*
1762 : * Attend to the cache fields. If the bound_offset matches the last
1763 : * cached bound offset then we've found the same partition as last time,
1764 : * so bump the count by one. If all goes well, we'll eventually reach
1765 : * PARTITION_CACHED_FIND_THRESHOLD and try the cache path next time
1766 : * around. Otherwise, we'll reset the cache count back to 1 to mark that
1767 : * we've found this partition for the first time.
1768 : */
1769 332650 : if (bound_offset == partdesc->last_found_datum_index)
1770 230865 : partdesc->last_found_count++;
1771 : else
1772 : {
1773 101785 : partdesc->last_found_count = 1;
1774 101785 : partdesc->last_found_part_index = part_index;
1775 101785 : partdesc->last_found_datum_index = bound_offset;
1776 : }
1777 :
1778 332650 : return part_index;
1779 : }
1780 :
1781 : /*
1782 : * ExecBuildSlotPartitionKeyDescription
1783 : *
1784 : * This works very much like BuildIndexValueDescription() and is currently
1785 : * used for building error messages when ExecFindPartition() fails to find
1786 : * partition for a row.
1787 : */
1788 : static char *
1789 77 : ExecBuildSlotPartitionKeyDescription(Relation rel,
1790 : const Datum *values,
1791 : const bool *isnull,
1792 : int maxfieldlen)
1793 : {
1794 : StringInfoData buf;
1795 77 : PartitionKey key = RelationGetPartitionKey(rel);
1796 77 : int partnatts = get_partition_natts(key);
1797 : int i;
1798 77 : Oid relid = RelationGetRelid(rel);
1799 : AclResult aclresult;
1800 :
1801 77 : if (check_enable_rls(relid, InvalidOid, true) == RLS_ENABLED)
1802 0 : return NULL;
1803 :
1804 : /* If the user has table-level access, just go build the description. */
1805 77 : aclresult = pg_class_aclcheck(relid, GetUserId(), ACL_SELECT);
1806 77 : if (aclresult != ACLCHECK_OK)
1807 : {
1808 : /*
1809 : * Step through the columns of the partition key and make sure the
1810 : * user has SELECT rights on all of them.
1811 : */
1812 12 : for (i = 0; i < partnatts; i++)
1813 : {
1814 9 : AttrNumber attnum = get_partition_col_attnum(key, i);
1815 :
1816 : /*
1817 : * If this partition key column is an expression, we return no
1818 : * detail rather than try to figure out what column(s) the
1819 : * expression includes and if the user has SELECT rights on them.
1820 : */
1821 15 : if (attnum == InvalidAttrNumber ||
1822 6 : pg_attribute_aclcheck(relid, attnum, GetUserId(),
1823 : ACL_SELECT) != ACLCHECK_OK)
1824 6 : return NULL;
1825 : }
1826 : }
1827 :
1828 71 : initStringInfo(&buf);
1829 71 : appendStringInfo(&buf, "(%s) = (",
1830 : pg_get_partkeydef_columns(relid, true));
1831 :
1832 169 : for (i = 0; i < partnatts; i++)
1833 : {
1834 : char *val;
1835 : int vallen;
1836 :
1837 98 : if (isnull[i])
1838 15 : val = "null";
1839 : else
1840 : {
1841 : Oid foutoid;
1842 : bool typisvarlena;
1843 :
1844 83 : getTypeOutputInfo(get_partition_col_typid(key, i),
1845 : &foutoid, &typisvarlena);
1846 83 : val = OidOutputFunctionCall(foutoid, values[i]);
1847 : }
1848 :
1849 98 : if (i > 0)
1850 27 : appendStringInfoString(&buf, ", ");
1851 :
1852 : /* truncate if needed */
1853 98 : vallen = strlen(val);
1854 98 : if (vallen <= maxfieldlen)
1855 98 : appendBinaryStringInfo(&buf, val, vallen);
1856 : else
1857 : {
1858 0 : vallen = pg_mbcliplen(val, vallen, maxfieldlen);
1859 0 : appendBinaryStringInfo(&buf, val, vallen);
1860 0 : appendStringInfoString(&buf, "...");
1861 : }
1862 : }
1863 :
1864 71 : appendStringInfoChar(&buf, ')');
1865 :
1866 71 : return buf.data;
1867 : }
1868 :
1869 : /*
1870 : * adjust_partition_colnos
1871 : * Adjust the list of UPDATE target column numbers to account for
1872 : * attribute differences between the parent and the partition.
1873 : *
1874 : * Note: mustn't be called if no adjustment is required.
1875 : */
1876 : static List *
1877 38 : adjust_partition_colnos(List *colnos, ResultRelInfo *leaf_part_rri)
1878 : {
1879 38 : TupleConversionMap *map = ExecGetChildToRootMap(leaf_part_rri);
1880 :
1881 : Assert(map != NULL);
1882 :
1883 38 : return adjust_partition_colnos_using_map(colnos, map->attrMap);
1884 : }
1885 :
1886 : /*
1887 : * adjust_partition_colnos_using_map
1888 : * Like adjust_partition_colnos, but uses a caller-supplied map instead
1889 : * of assuming to map from the "root" result relation.
1890 : *
1891 : * Note: mustn't be called if no adjustment is required.
1892 : */
1893 : static List *
1894 47 : adjust_partition_colnos_using_map(List *colnos, AttrMap *attrMap)
1895 : {
1896 47 : List *new_colnos = NIL;
1897 : ListCell *lc;
1898 :
1899 : Assert(attrMap != NULL); /* else we shouldn't be here */
1900 :
1901 116 : foreach(lc, colnos)
1902 : {
1903 69 : AttrNumber parentattrno = lfirst_int(lc);
1904 :
1905 69 : if (parentattrno <= 0 ||
1906 69 : parentattrno > attrMap->maplen ||
1907 69 : attrMap->attnums[parentattrno - 1] == 0)
1908 0 : elog(ERROR, "unexpected attno %d in target column list",
1909 : parentattrno);
1910 69 : new_colnos = lappend_int(new_colnos,
1911 69 : attrMap->attnums[parentattrno - 1]);
1912 : }
1913 :
1914 47 : return new_colnos;
1915 : }
1916 :
1917 : /*-------------------------------------------------------------------------
1918 : * Run-Time Partition Pruning Support.
1919 : *
1920 : * The following series of functions exist to support the removal of unneeded
1921 : * subplans for queries against partitioned tables. The supporting functions
1922 : * here are designed to work with any plan type which supports an arbitrary
1923 : * number of subplans, e.g. Append, MergeAppend.
1924 : *
1925 : * When pruning involves comparison of a partition key to a constant, it's
1926 : * done by the planner. However, if we have a comparison to a non-constant
1927 : * but not volatile expression, that presents an opportunity for run-time
1928 : * pruning by the executor, allowing irrelevant partitions to be skipped
1929 : * dynamically.
1930 : *
1931 : * We must distinguish expressions containing PARAM_EXEC Params from
1932 : * expressions that don't contain those. Even though a PARAM_EXEC Param is
1933 : * considered to be a stable expression, it can change value from one plan
1934 : * node scan to the next during query execution. Stable comparison
1935 : * expressions that don't involve such Params allow partition pruning to be
1936 : * done once during executor startup. Expressions that do involve such Params
1937 : * require us to prune separately for each scan of the parent plan node.
1938 : *
1939 : * Note that pruning away unneeded subplans during executor startup has the
1940 : * added benefit of not having to initialize the unneeded subplans at all.
1941 : *
1942 : *
1943 : * Functions:
1944 : *
1945 : * ExecDoInitialPruning:
1946 : * Perform runtime "initial" pruning, if necessary, to determine the set
1947 : * of child subnodes that need to be initialized during ExecInitNode() for
1948 : * all plan nodes that contain a PartitionPruneInfo.
1949 : *
1950 : * ExecInitPartitionExecPruning:
1951 : * Updates the PartitionPruneState found at given part_prune_index in
1952 : * EState.es_part_prune_states for use during "exec" pruning if required.
1953 : * Also returns the set of subplans to initialize that would be stored at
1954 : * part_prune_index in EState.es_part_prune_results by
1955 : * ExecDoInitialPruning(). Maps in PartitionPruneState are updated to
1956 : * account for initial pruning possibly having eliminated some of the
1957 : * subplans.
1958 : *
1959 : * ExecFindMatchingSubPlans:
1960 : * Returns indexes of matching subplans after evaluating the expressions
1961 : * that are safe to evaluate at a given point. This function is first
1962 : * called during ExecDoInitialPruning() to find the initially matching
1963 : * subplans based on performing the initial pruning steps and then must be
1964 : * called again each time the value of a Param listed in
1965 : * PartitionPruneState's 'execparamids' changes.
1966 : *-------------------------------------------------------------------------
1967 : */
1968 :
1969 :
1970 : /*
1971 : * ExecDoInitialPruning
1972 : * Perform runtime "initial" pruning, if necessary, to determine the set
1973 : * of child subnodes that need to be initialized during ExecInitNode() for
1974 : * plan nodes that support partition pruning.
1975 : *
1976 : * This function iterates over each PartitionPruneInfo entry in
1977 : * estate->es_part_prune_infos. For each entry, it creates a PartitionPruneState
1978 : * and adds it to es_part_prune_states. ExecInitPartitionExecPruning() accesses
1979 : * these states through their corresponding indexes in es_part_prune_states and
1980 : * assign each state to the parent node's PlanState, from where it will be used
1981 : * for "exec" pruning.
1982 : *
1983 : * If initial pruning steps exist for a PartitionPruneInfo entry, this function
1984 : * executes those pruning steps and stores the result as a bitmapset of valid
1985 : * child subplans, identifying which subplans should be initialized for
1986 : * execution. The results are saved in estate->es_part_prune_results.
1987 : *
1988 : * If no initial pruning is performed for a given PartitionPruneInfo, a NULL
1989 : * entry is still added to es_part_prune_results to maintain alignment with
1990 : * es_part_prune_infos. This ensures that ExecInitPartitionExecPruning() can
1991 : * use the same index to retrieve the pruning results.
1992 : */
1993 : void
1994 297722 : ExecDoInitialPruning(EState *estate)
1995 : {
1996 : ListCell *lc;
1997 :
1998 298123 : foreach(lc, estate->es_part_prune_infos)
1999 : {
2000 401 : PartitionPruneInfo *pruneinfo = lfirst_node(PartitionPruneInfo, lc);
2001 : PartitionPruneState *prunestate;
2002 401 : Bitmapset *validsubplans = NULL;
2003 401 : Bitmapset *all_leafpart_rtis = NULL;
2004 401 : Bitmapset *validsubplan_rtis = NULL;
2005 :
2006 : /* Create and save the PartitionPruneState. */
2007 401 : prunestate = CreatePartitionPruneState(estate, pruneinfo,
2008 : &all_leafpart_rtis);
2009 401 : estate->es_part_prune_states = lappend(estate->es_part_prune_states,
2010 : prunestate);
2011 :
2012 : /*
2013 : * Perform initial pruning steps, if any, and save the result
2014 : * bitmapset or NULL as described in the header comment.
2015 : */
2016 401 : if (prunestate->do_initial_prune)
2017 224 : validsubplans = ExecFindMatchingSubPlans(prunestate, true,
2018 : &validsubplan_rtis);
2019 : else
2020 177 : validsubplan_rtis = all_leafpart_rtis;
2021 :
2022 401 : estate->es_unpruned_relids = bms_add_members(estate->es_unpruned_relids,
2023 : validsubplan_rtis);
2024 401 : estate->es_part_prune_results = lappend(estate->es_part_prune_results,
2025 : validsubplans);
2026 : }
2027 297722 : }
2028 :
2029 : /*
2030 : * ExecInitPartitionExecPruning
2031 : * Initialize the data structures needed for runtime "exec" partition
2032 : * pruning and return the result of initial pruning, if available.
2033 : *
2034 : * 'relids' identifies the relation to which both the parent plan and the
2035 : * PartitionPruneInfo given by 'part_prune_index' belong.
2036 : *
2037 : * On return, *initially_valid_subplans is assigned the set of indexes of
2038 : * child subplans that must be initialized along with the parent plan node.
2039 : * Initial pruning would have been performed by ExecDoInitialPruning(), if
2040 : * necessary, and the bitmapset of surviving subplans' indexes would have
2041 : * been stored as the part_prune_index'th element of
2042 : * EState.es_part_prune_results.
2043 : *
2044 : * If subplans were indeed pruned during initial pruning, the subplan_map
2045 : * arrays in the returned PartitionPruneState are re-sequenced to exclude those
2046 : * subplans, but only if the maps will be needed for subsequent execution
2047 : * pruning passes.
2048 : */
2049 : PartitionPruneState *
2050 403 : ExecInitPartitionExecPruning(PlanState *planstate,
2051 : int n_total_subplans,
2052 : int part_prune_index,
2053 : Bitmapset *relids,
2054 : Bitmapset **initially_valid_subplans)
2055 : {
2056 : PartitionPruneState *prunestate;
2057 403 : EState *estate = planstate->state;
2058 : PartitionPruneInfo *pruneinfo;
2059 :
2060 : /* Obtain the pruneinfo we need. */
2061 403 : pruneinfo = list_nth_node(PartitionPruneInfo, estate->es_part_prune_infos,
2062 : part_prune_index);
2063 :
2064 : /* Its relids better match the plan node's or the planner messed up. */
2065 403 : if (!bms_equal(relids, pruneinfo->relids))
2066 0 : elog(ERROR, "wrong pruneinfo with relids=%s found at part_prune_index=%d contained in plan node with relids=%s",
2067 : bmsToString(pruneinfo->relids), part_prune_index,
2068 : bmsToString(relids));
2069 :
2070 : /*
2071 : * The PartitionPruneState would have been created by
2072 : * ExecDoInitialPruning() and stored as the part_prune_index'th element of
2073 : * EState.es_part_prune_states.
2074 : */
2075 403 : prunestate = list_nth(estate->es_part_prune_states, part_prune_index);
2076 : Assert(prunestate != NULL);
2077 :
2078 : /* Use the result of initial pruning done by ExecDoInitialPruning(). */
2079 403 : if (prunestate->do_initial_prune)
2080 225 : *initially_valid_subplans = list_nth_node(Bitmapset,
2081 : estate->es_part_prune_results,
2082 : part_prune_index);
2083 : else
2084 : {
2085 : /* No pruning, so we'll need to initialize all subplans */
2086 : Assert(n_total_subplans > 0);
2087 178 : *initially_valid_subplans = bms_add_range(NULL, 0,
2088 : n_total_subplans - 1);
2089 : }
2090 :
2091 : /*
2092 : * The exec pruning state must also be initialized, if needed, before it
2093 : * can be used for pruning during execution.
2094 : *
2095 : * This also re-sequences subplan indexes contained in prunestate to
2096 : * account for any that were removed due to initial pruning; refer to the
2097 : * condition in InitExecPartitionPruneContexts() that is used to determine
2098 : * whether to do this. If no exec pruning needs to be done, we would thus
2099 : * leave the maps to be in an invalid state, but that's ok since that data
2100 : * won't be consulted again (cf initial Assert in
2101 : * ExecFindMatchingSubPlans).
2102 : */
2103 403 : if (prunestate->do_exec_prune)
2104 199 : InitExecPartitionPruneContexts(prunestate, planstate,
2105 : *initially_valid_subplans,
2106 : n_total_subplans);
2107 :
2108 403 : return prunestate;
2109 : }
2110 :
2111 : /*
2112 : * CreatePartitionPruneState
2113 : * Build the data structure required for calling ExecFindMatchingSubPlans
2114 : *
2115 : * This includes PartitionPruneContexts (stored in each
2116 : * PartitionedRelPruningData corresponding to a PartitionedRelPruneInfo),
2117 : * which hold the ExprStates needed to evaluate pruning expressions, and
2118 : * mapping arrays to convert partition indexes from the pruning logic
2119 : * into subplan indexes in the parent plan node's list of child subplans.
2120 : *
2121 : * 'pruneinfo' is a PartitionPruneInfo as generated by
2122 : * make_partition_pruneinfo. Here we build a PartitionPruneState containing a
2123 : * PartitionPruningData for each partitioning hierarchy (i.e., each sublist of
2124 : * pruneinfo->prune_infos), each of which contains a PartitionedRelPruningData
2125 : * for each PartitionedRelPruneInfo appearing in that sublist. This two-level
2126 : * system is needed to keep from confusing the different hierarchies when a
2127 : * UNION ALL contains multiple partitioned tables as children. The data
2128 : * stored in each PartitionedRelPruningData can be re-used each time we
2129 : * re-evaluate which partitions match the pruning steps provided in each
2130 : * PartitionedRelPruneInfo.
2131 : *
2132 : * Note that only the PartitionPruneContexts for initial pruning are
2133 : * initialized here. Those required for exec pruning are initialized later in
2134 : * ExecInitPartitionExecPruning(), as they depend on the availability of the
2135 : * parent plan node's PlanState.
2136 : *
2137 : * If initial pruning steps are to be skipped (e.g., during EXPLAIN
2138 : * (GENERIC_PLAN)), *all_leafpart_rtis will be populated with the RT indexes of
2139 : * all leaf partitions whose scanning subnode is included in the parent plan
2140 : * node's list of child plans. The caller must add these RT indexes to
2141 : * estate->es_unpruned_relids.
2142 : */
2143 : static PartitionPruneState *
2144 401 : CreatePartitionPruneState(EState *estate, PartitionPruneInfo *pruneinfo,
2145 : Bitmapset **all_leafpart_rtis)
2146 : {
2147 : PartitionPruneState *prunestate;
2148 : int n_part_hierarchies;
2149 : ListCell *lc;
2150 : int i;
2151 :
2152 : /*
2153 : * Expression context that will be used by partkey_datum_from_expr() to
2154 : * evaluate expressions for comparison against partition bounds.
2155 : */
2156 401 : ExprContext *econtext = CreateExprContext(estate);
2157 :
2158 : /* For data reading, executor always includes detached partitions */
2159 401 : if (estate->es_partition_directory == NULL)
2160 377 : estate->es_partition_directory =
2161 377 : CreatePartitionDirectory(estate->es_query_cxt, false);
2162 :
2163 401 : n_part_hierarchies = list_length(pruneinfo->prune_infos);
2164 : Assert(n_part_hierarchies > 0);
2165 :
2166 : /*
2167 : * Allocate the data structure
2168 : */
2169 : prunestate = (PartitionPruneState *)
2170 401 : palloc(offsetof(PartitionPruneState, partprunedata) +
2171 : sizeof(PartitionPruningData *) * n_part_hierarchies);
2172 :
2173 : /* Save ExprContext for use during InitExecPartitionPruneContexts(). */
2174 401 : prunestate->econtext = econtext;
2175 401 : prunestate->execparamids = NULL;
2176 : /* other_subplans can change at runtime, so we need our own copy */
2177 401 : prunestate->other_subplans = bms_copy(pruneinfo->other_subplans);
2178 401 : prunestate->do_initial_prune = false; /* may be set below */
2179 401 : prunestate->do_exec_prune = false; /* may be set below */
2180 401 : prunestate->num_partprunedata = n_part_hierarchies;
2181 :
2182 : /*
2183 : * Create a short-term memory context which we'll use when making calls to
2184 : * the partition pruning functions. This avoids possible memory leaks,
2185 : * since the pruning functions call comparison functions that aren't under
2186 : * our control.
2187 : */
2188 401 : prunestate->prune_context =
2189 401 : AllocSetContextCreate(CurrentMemoryContext,
2190 : "Partition Prune",
2191 : ALLOCSET_DEFAULT_SIZES);
2192 :
2193 401 : i = 0;
2194 814 : foreach(lc, pruneinfo->prune_infos)
2195 : {
2196 413 : List *partrelpruneinfos = lfirst_node(List, lc);
2197 413 : int npartrelpruneinfos = list_length(partrelpruneinfos);
2198 : PartitionPruningData *prunedata;
2199 : ListCell *lc2;
2200 : int j;
2201 :
2202 : prunedata = (PartitionPruningData *)
2203 413 : palloc(offsetof(PartitionPruningData, partrelprunedata) +
2204 413 : npartrelpruneinfos * sizeof(PartitionedRelPruningData));
2205 413 : prunestate->partprunedata[i] = prunedata;
2206 413 : prunedata->num_partrelprunedata = npartrelpruneinfos;
2207 :
2208 413 : j = 0;
2209 1231 : foreach(lc2, partrelpruneinfos)
2210 : {
2211 818 : PartitionedRelPruneInfo *pinfo = lfirst_node(PartitionedRelPruneInfo, lc2);
2212 818 : PartitionedRelPruningData *pprune = &prunedata->partrelprunedata[j];
2213 : Relation partrel;
2214 : PartitionDesc partdesc;
2215 : PartitionKey partkey;
2216 :
2217 : /*
2218 : * We can rely on the copies of the partitioned table's partition
2219 : * key and partition descriptor appearing in its relcache entry,
2220 : * because that entry will be held open and locked for the
2221 : * duration of this executor run.
2222 : */
2223 818 : partrel = ExecGetRangeTableRelation(estate, pinfo->rtindex, false);
2224 :
2225 : /* Remember for InitExecPartitionPruneContexts(). */
2226 818 : pprune->partrel = partrel;
2227 :
2228 818 : partkey = RelationGetPartitionKey(partrel);
2229 818 : partdesc = PartitionDirectoryLookup(estate->es_partition_directory,
2230 : partrel);
2231 :
2232 : /*
2233 : * Initialize the subplan_map and subpart_map.
2234 : *
2235 : * The set of partitions that exist now might not be the same that
2236 : * existed when the plan was made. The normal case is that it is;
2237 : * optimize for that case with a quick comparison, and just copy
2238 : * the subplan_map and make subpart_map, leafpart_rti_map point to
2239 : * the ones in PruneInfo.
2240 : *
2241 : * For the case where they aren't identical, we could have more
2242 : * partitions on either side; or even exactly the same number of
2243 : * them on both but the set of OIDs doesn't match fully. Handle
2244 : * this by creating new subplan_map and subpart_map arrays that
2245 : * corresponds to the ones in the PruneInfo where the new
2246 : * partition descriptor's OIDs match. Any that don't match can be
2247 : * set to -1, as if they were pruned. By construction, both
2248 : * arrays are in partition bounds order.
2249 : */
2250 818 : pprune->nparts = partdesc->nparts;
2251 818 : pprune->subplan_map = palloc_array(int, partdesc->nparts);
2252 :
2253 818 : if (partdesc->nparts == pinfo->nparts &&
2254 817 : memcmp(partdesc->oids, pinfo->relid_map,
2255 817 : sizeof(int) * partdesc->nparts) == 0)
2256 : {
2257 756 : pprune->subpart_map = pinfo->subpart_map;
2258 756 : pprune->leafpart_rti_map = pinfo->leafpart_rti_map;
2259 756 : memcpy(pprune->subplan_map, pinfo->subplan_map,
2260 756 : sizeof(int) * pinfo->nparts);
2261 : }
2262 : else
2263 : {
2264 62 : int pd_idx = 0;
2265 : int pp_idx;
2266 :
2267 : /*
2268 : * When the partition arrays are not identical, there could be
2269 : * some new ones but it's also possible that one was removed;
2270 : * we cope with both situations by walking the arrays and
2271 : * discarding those that don't match.
2272 : *
2273 : * If the number of partitions on both sides match, it's still
2274 : * possible that one partition has been detached and another
2275 : * attached. Cope with that by creating a map that skips any
2276 : * mismatches.
2277 : */
2278 62 : pprune->subpart_map = palloc_array(int, partdesc->nparts);
2279 62 : pprune->leafpart_rti_map = palloc_array(int, partdesc->nparts);
2280 :
2281 264 : for (pp_idx = 0; pp_idx < partdesc->nparts; pp_idx++)
2282 : {
2283 : /* Skip any InvalidOid relid_map entries */
2284 312 : while (pd_idx < pinfo->nparts &&
2285 252 : !OidIsValid(pinfo->relid_map[pd_idx]))
2286 110 : pd_idx++;
2287 :
2288 202 : recheck:
2289 202 : if (pd_idx < pinfo->nparts &&
2290 142 : pinfo->relid_map[pd_idx] == partdesc->oids[pp_idx])
2291 : {
2292 : /* match... */
2293 91 : pprune->subplan_map[pp_idx] =
2294 91 : pinfo->subplan_map[pd_idx];
2295 91 : pprune->subpart_map[pp_idx] =
2296 91 : pinfo->subpart_map[pd_idx];
2297 91 : pprune->leafpart_rti_map[pp_idx] =
2298 91 : pinfo->leafpart_rti_map[pd_idx];
2299 91 : pd_idx++;
2300 91 : continue;
2301 : }
2302 :
2303 : /*
2304 : * There isn't an exact match in the corresponding
2305 : * positions of both arrays. Peek ahead in
2306 : * pinfo->relid_map to see if we have a match for the
2307 : * current partition in partdesc. Normally if a match
2308 : * exists it's just one element ahead, and it means the
2309 : * planner saw one extra partition that we no longer see
2310 : * now (its concurrent detach finished just in between);
2311 : * so we skip that one by updating pd_idx to the new
2312 : * location and jumping above. We can then continue to
2313 : * match the rest of the elements after skipping the OID
2314 : * with no match; no future matches are tried for the
2315 : * element that was skipped, because we know the arrays to
2316 : * be in the same order.
2317 : *
2318 : * If we don't see a match anywhere in the rest of the
2319 : * pinfo->relid_map array, that means we see an element
2320 : * now that the planner didn't see, so mark that one as
2321 : * pruned and move on.
2322 : */
2323 144 : for (int pd_idx2 = pd_idx + 1; pd_idx2 < pinfo->nparts; pd_idx2++)
2324 : {
2325 33 : if (pd_idx2 >= pinfo->nparts)
2326 0 : break;
2327 33 : if (pinfo->relid_map[pd_idx2] == partdesc->oids[pp_idx])
2328 : {
2329 0 : pd_idx = pd_idx2;
2330 0 : goto recheck;
2331 : }
2332 : }
2333 :
2334 111 : pprune->subpart_map[pp_idx] = -1;
2335 111 : pprune->subplan_map[pp_idx] = -1;
2336 111 : pprune->leafpart_rti_map[pp_idx] = 0;
2337 : }
2338 : }
2339 :
2340 : /* present_parts is also subject to later modification */
2341 818 : pprune->present_parts = bms_copy(pinfo->present_parts);
2342 :
2343 : /*
2344 : * Only initial_context is initialized here. exec_context is
2345 : * initialized during ExecInitPartitionExecPruning() when the
2346 : * parent plan's PlanState is available.
2347 : *
2348 : * Note that we must skip execution-time (both "init" and "exec")
2349 : * partition pruning in EXPLAIN (GENERIC_PLAN), since parameter
2350 : * values may be missing.
2351 : */
2352 818 : pprune->initial_pruning_steps = pinfo->initial_pruning_steps;
2353 818 : if (pinfo->initial_pruning_steps &&
2354 278 : !(econtext->ecxt_estate->es_top_eflags & EXEC_FLAG_EXPLAIN_GENERIC))
2355 : {
2356 275 : InitPartitionPruneContext(&pprune->initial_context,
2357 : pprune->initial_pruning_steps,
2358 : partdesc, partkey, NULL,
2359 : econtext);
2360 : /* Record whether initial pruning is needed at any level */
2361 275 : prunestate->do_initial_prune = true;
2362 : }
2363 818 : pprune->exec_pruning_steps = pinfo->exec_pruning_steps;
2364 818 : if (pinfo->exec_pruning_steps &&
2365 255 : !(econtext->ecxt_estate->es_top_eflags & EXEC_FLAG_EXPLAIN_GENERIC))
2366 : {
2367 : /* Record whether exec pruning is needed at any level */
2368 255 : prunestate->do_exec_prune = true;
2369 : }
2370 :
2371 : /*
2372 : * Accumulate the IDs of all PARAM_EXEC Params affecting the
2373 : * partitioning decisions at this plan node.
2374 : */
2375 1636 : prunestate->execparamids = bms_add_members(prunestate->execparamids,
2376 818 : pinfo->execparamids);
2377 :
2378 : /*
2379 : * Return all leaf partition indexes if we're skipping pruning in
2380 : * the EXPLAIN (GENERIC_PLAN) case.
2381 : */
2382 818 : if (pinfo->initial_pruning_steps && !prunestate->do_initial_prune)
2383 : {
2384 3 : int part_index = -1;
2385 :
2386 9 : while ((part_index = bms_next_member(pprune->present_parts,
2387 9 : part_index)) >= 0)
2388 : {
2389 6 : Index rtindex = pprune->leafpart_rti_map[part_index];
2390 :
2391 6 : if (rtindex)
2392 6 : *all_leafpart_rtis = bms_add_member(*all_leafpart_rtis,
2393 : rtindex);
2394 : }
2395 : }
2396 :
2397 818 : j++;
2398 : }
2399 413 : i++;
2400 : }
2401 :
2402 401 : return prunestate;
2403 : }
2404 :
2405 : /*
2406 : * Initialize a PartitionPruneContext for the given list of pruning steps.
2407 : */
2408 : static void
2409 531 : InitPartitionPruneContext(PartitionPruneContext *context,
2410 : List *pruning_steps,
2411 : PartitionDesc partdesc,
2412 : PartitionKey partkey,
2413 : PlanState *planstate,
2414 : ExprContext *econtext)
2415 : {
2416 : int n_steps;
2417 : int partnatts;
2418 : ListCell *lc;
2419 :
2420 531 : n_steps = list_length(pruning_steps);
2421 :
2422 531 : context->strategy = partkey->strategy;
2423 531 : context->partnatts = partnatts = partkey->partnatts;
2424 531 : context->nparts = partdesc->nparts;
2425 531 : context->boundinfo = partdesc->boundinfo;
2426 531 : context->partcollation = partkey->partcollation;
2427 531 : context->partsupfunc = partkey->partsupfunc;
2428 :
2429 : /* We'll look up type-specific support functions as needed */
2430 531 : context->stepcmpfuncs = palloc0_array(FmgrInfo, n_steps * partnatts);
2431 :
2432 531 : context->ppccontext = CurrentMemoryContext;
2433 531 : context->planstate = planstate;
2434 531 : context->exprcontext = econtext;
2435 :
2436 : /* Initialize expression state for each expression we need */
2437 531 : context->exprstates = palloc0_array(ExprState *, n_steps * partnatts);
2438 1393 : foreach(lc, pruning_steps)
2439 : {
2440 862 : PartitionPruneStepOp *step = (PartitionPruneStepOp *) lfirst(lc);
2441 862 : ListCell *lc2 = list_head(step->exprs);
2442 : int keyno;
2443 :
2444 : /* not needed for other step kinds */
2445 862 : if (!IsA(step, PartitionPruneStepOp))
2446 143 : continue;
2447 :
2448 : Assert(list_length(step->exprs) <= partnatts);
2449 :
2450 1513 : for (keyno = 0; keyno < partnatts; keyno++)
2451 : {
2452 794 : if (bms_is_member(keyno, step->nullkeys))
2453 3 : continue;
2454 :
2455 791 : if (lc2 != NULL)
2456 : {
2457 743 : Expr *expr = lfirst(lc2);
2458 :
2459 : /* not needed for Consts */
2460 743 : if (!IsA(expr, Const))
2461 : {
2462 696 : int stateidx = PruneCxtStateIdx(partnatts,
2463 : step->step.step_id,
2464 : keyno);
2465 :
2466 : /*
2467 : * When planstate is NULL, pruning_steps is known not to
2468 : * contain any expressions that depend on the parent plan.
2469 : * Information of any available EXTERN parameters must be
2470 : * passed explicitly in that case, which the caller must
2471 : * have made available via econtext.
2472 : */
2473 696 : if (planstate == NULL)
2474 407 : context->exprstates[stateidx] =
2475 407 : ExecInitExprWithParams(expr,
2476 : econtext->ecxt_param_list_info);
2477 : else
2478 289 : context->exprstates[stateidx] =
2479 289 : ExecInitExpr(expr, context->planstate);
2480 : }
2481 743 : lc2 = lnext(step->exprs, lc2);
2482 : }
2483 : }
2484 : }
2485 531 : }
2486 :
2487 : /*
2488 : * InitExecPartitionPruneContexts
2489 : * Initialize exec pruning contexts deferred by CreatePartitionPruneState()
2490 : *
2491 : * This function finalizes exec pruning setup for a PartitionPruneState by
2492 : * initializing contexts for pruning steps that require the parent plan's
2493 : * PlanState. It iterates over PartitionPruningData entries and sets up the
2494 : * necessary execution contexts for pruning during query execution.
2495 : *
2496 : * Also fix the mapping of partition indexes to subplan indexes contained in
2497 : * prunestate by considering the new list of subplans that survived initial
2498 : * pruning.
2499 : *
2500 : * Current values of the indexes present in PartitionPruneState count all the
2501 : * subplans that would be present before initial pruning was done. If initial
2502 : * pruning got rid of some of the subplans, any subsequent pruning passes will
2503 : * be looking at a different set of target subplans to choose from than those
2504 : * in the pre-initial-pruning set, so the maps in PartitionPruneState
2505 : * containing those indexes must be updated to reflect the new indexes of
2506 : * subplans in the post-initial-pruning set.
2507 : */
2508 : static void
2509 199 : InitExecPartitionPruneContexts(PartitionPruneState *prunestate,
2510 : PlanState *parent_plan,
2511 : Bitmapset *initially_valid_subplans,
2512 : int n_total_subplans)
2513 : {
2514 : EState *estate;
2515 199 : int *new_subplan_indexes = NULL;
2516 : Bitmapset *new_other_subplans;
2517 : int i;
2518 : int newidx;
2519 199 : bool fix_subplan_map = false;
2520 :
2521 : Assert(prunestate->do_exec_prune);
2522 : Assert(parent_plan != NULL);
2523 199 : estate = parent_plan->state;
2524 :
2525 : /*
2526 : * No need to fix subplans maps if initial pruning didn't eliminate any
2527 : * subplans.
2528 : */
2529 199 : if (bms_num_members(initially_valid_subplans) < n_total_subplans)
2530 : {
2531 24 : fix_subplan_map = true;
2532 :
2533 : /*
2534 : * First we must build a temporary array which maps old subplan
2535 : * indexes to new ones. For convenience of initialization, we use
2536 : * 1-based indexes in this array and leave pruned items as 0.
2537 : */
2538 24 : new_subplan_indexes = palloc0_array(int, n_total_subplans);
2539 24 : newidx = 1;
2540 24 : i = -1;
2541 93 : while ((i = bms_next_member(initially_valid_subplans, i)) >= 0)
2542 : {
2543 : Assert(i < n_total_subplans);
2544 69 : new_subplan_indexes[i] = newidx++;
2545 : }
2546 : }
2547 :
2548 : /*
2549 : * Now we can update each PartitionedRelPruneInfo's subplan_map with new
2550 : * subplan indexes. We must also recompute its present_parts bitmap.
2551 : */
2552 410 : for (i = 0; i < prunestate->num_partprunedata; i++)
2553 : {
2554 211 : PartitionPruningData *prunedata = prunestate->partprunedata[i];
2555 : int j;
2556 :
2557 : /*
2558 : * Within each hierarchy, we perform this loop in back-to-front order
2559 : * so that we determine present_parts for the lowest-level partitioned
2560 : * tables first. This way we can tell whether a sub-partitioned
2561 : * table's partitions were entirely pruned so we can exclude it from
2562 : * the current level's present_parts.
2563 : */
2564 650 : for (j = prunedata->num_partrelprunedata - 1; j >= 0; j--)
2565 : {
2566 439 : PartitionedRelPruningData *pprune = &prunedata->partrelprunedata[j];
2567 439 : int nparts = pprune->nparts;
2568 : int k;
2569 :
2570 : /* Initialize PartitionPruneContext for exec pruning, if needed. */
2571 439 : if (pprune->exec_pruning_steps != NIL)
2572 : {
2573 : PartitionKey partkey;
2574 : PartitionDesc partdesc;
2575 :
2576 : /*
2577 : * See the comment in CreatePartitionPruneState() regarding
2578 : * the usage of partdesc and partkey.
2579 : */
2580 256 : partkey = RelationGetPartitionKey(pprune->partrel);
2581 256 : partdesc = PartitionDirectoryLookup(estate->es_partition_directory,
2582 : pprune->partrel);
2583 :
2584 256 : InitPartitionPruneContext(&pprune->exec_context,
2585 : pprune->exec_pruning_steps,
2586 : partdesc, partkey, parent_plan,
2587 : prunestate->econtext);
2588 : }
2589 :
2590 439 : if (!fix_subplan_map)
2591 343 : continue;
2592 :
2593 : /* We just rebuild present_parts from scratch */
2594 96 : bms_free(pprune->present_parts);
2595 96 : pprune->present_parts = NULL;
2596 :
2597 354 : for (k = 0; k < nparts; k++)
2598 : {
2599 258 : int oldidx = pprune->subplan_map[k];
2600 : int subidx;
2601 :
2602 : /*
2603 : * If this partition existed as a subplan then change the old
2604 : * subplan index to the new subplan index. The new index may
2605 : * become -1 if the partition was pruned above, or it may just
2606 : * come earlier in the subplan list due to some subplans being
2607 : * removed earlier in the list. If it's a subpartition, add
2608 : * it to present_parts unless it's entirely pruned.
2609 : */
2610 258 : if (oldidx >= 0)
2611 : {
2612 : Assert(oldidx < n_total_subplans);
2613 198 : pprune->subplan_map[k] = new_subplan_indexes[oldidx] - 1;
2614 :
2615 198 : if (new_subplan_indexes[oldidx] > 0)
2616 57 : pprune->present_parts =
2617 57 : bms_add_member(pprune->present_parts, k);
2618 : }
2619 60 : else if ((subidx = pprune->subpart_map[k]) >= 0)
2620 : {
2621 : PartitionedRelPruningData *subprune;
2622 :
2623 60 : subprune = &prunedata->partrelprunedata[subidx];
2624 :
2625 60 : if (!bms_is_empty(subprune->present_parts))
2626 24 : pprune->present_parts =
2627 24 : bms_add_member(pprune->present_parts, k);
2628 : }
2629 : }
2630 : }
2631 : }
2632 :
2633 : /*
2634 : * If we fixed subplan maps, we must also recompute the other_subplans
2635 : * set, since indexes in it may change.
2636 : */
2637 199 : if (fix_subplan_map)
2638 : {
2639 24 : new_other_subplans = NULL;
2640 24 : i = -1;
2641 36 : while ((i = bms_next_member(prunestate->other_subplans, i)) >= 0)
2642 12 : new_other_subplans = bms_add_member(new_other_subplans,
2643 12 : new_subplan_indexes[i] - 1);
2644 :
2645 24 : bms_free(prunestate->other_subplans);
2646 24 : prunestate->other_subplans = new_other_subplans;
2647 :
2648 24 : pfree(new_subplan_indexes);
2649 : }
2650 199 : }
2651 :
2652 : /*
2653 : * ExecFindMatchingSubPlans
2654 : * Determine which subplans match the pruning steps detailed in
2655 : * 'prunestate' for the current comparison expression values.
2656 : *
2657 : * Pass initial_prune if PARAM_EXEC Params cannot yet be evaluated. This
2658 : * differentiates the initial executor-time pruning step from later
2659 : * runtime pruning.
2660 : *
2661 : * The caller must pass a non-NULL validsubplan_rtis during initial pruning
2662 : * to collect the RT indexes of leaf partitions whose subnodes will be
2663 : * executed. These RT indexes are later added to EState.es_unpruned_relids.
2664 : */
2665 : Bitmapset *
2666 1949 : ExecFindMatchingSubPlans(PartitionPruneState *prunestate,
2667 : bool initial_prune,
2668 : Bitmapset **validsubplan_rtis)
2669 : {
2670 1949 : Bitmapset *result = NULL;
2671 : MemoryContext oldcontext;
2672 : int i;
2673 :
2674 : /*
2675 : * Either we're here on the initial prune done during pruning
2676 : * initialization, or we're at a point where PARAM_EXEC Params can be
2677 : * evaluated *and* there are steps in which to do so.
2678 : */
2679 : Assert(initial_prune || prunestate->do_exec_prune);
2680 : Assert(validsubplan_rtis != NULL || !initial_prune);
2681 :
2682 : /*
2683 : * Switch to a temp context to avoid leaking memory in the executor's
2684 : * query-lifespan memory context.
2685 : */
2686 1949 : oldcontext = MemoryContextSwitchTo(prunestate->prune_context);
2687 :
2688 : /*
2689 : * For each hierarchy, do the pruning tests, and add nondeletable
2690 : * subplans' indexes to "result".
2691 : */
2692 3919 : for (i = 0; i < prunestate->num_partprunedata; i++)
2693 : {
2694 1970 : PartitionPruningData *prunedata = prunestate->partprunedata[i];
2695 : PartitionedRelPruningData *pprune;
2696 :
2697 : /*
2698 : * We pass the zeroth item, belonging to the root table of the
2699 : * hierarchy, and find_matching_subplans_recurse() takes care of
2700 : * recursing to other (lower-level) parents as needed.
2701 : */
2702 1970 : pprune = &prunedata->partrelprunedata[0];
2703 1970 : find_matching_subplans_recurse(prunedata, pprune, initial_prune,
2704 : &result, validsubplan_rtis);
2705 :
2706 : /*
2707 : * Expression eval may have used space in ExprContext too. Avoid
2708 : * accessing exec_context during initial pruning, as it is not valid
2709 : * at that stage.
2710 : */
2711 1970 : if (!initial_prune && pprune->exec_pruning_steps)
2712 1698 : ResetExprContext(pprune->exec_context.exprcontext);
2713 : }
2714 :
2715 : /* Add in any subplans that partition pruning didn't account for */
2716 1949 : result = bms_add_members(result, prunestate->other_subplans);
2717 :
2718 1949 : MemoryContextSwitchTo(oldcontext);
2719 :
2720 : /* Copy result out of the temp context before we reset it */
2721 1949 : result = bms_copy(result);
2722 1949 : if (validsubplan_rtis)
2723 224 : *validsubplan_rtis = bms_copy(*validsubplan_rtis);
2724 :
2725 1949 : MemoryContextReset(prunestate->prune_context);
2726 :
2727 1949 : return result;
2728 : }
2729 :
2730 : /*
2731 : * find_matching_subplans_recurse
2732 : * Recursive worker function for ExecFindMatchingSubPlans
2733 : *
2734 : * Adds valid (non-prunable) subplan IDs to *validsubplans. If
2735 : * *validsubplan_rtis is non-NULL, it also adds the RT indexes of their
2736 : * corresponding partitions, but only if they are leaf partitions.
2737 : */
2738 : static void
2739 2177 : find_matching_subplans_recurse(PartitionPruningData *prunedata,
2740 : PartitionedRelPruningData *pprune,
2741 : bool initial_prune,
2742 : Bitmapset **validsubplans,
2743 : Bitmapset **validsubplan_rtis)
2744 : {
2745 : Bitmapset *partset;
2746 : int i;
2747 :
2748 : /* Guard against stack overflow due to overly deep partition hierarchy. */
2749 2177 : check_stack_depth();
2750 :
2751 : /*
2752 : * Prune as appropriate, if we have pruning steps matching the current
2753 : * execution context. Otherwise just include all partitions at this
2754 : * level.
2755 : */
2756 2177 : if (initial_prune && pprune->initial_pruning_steps)
2757 266 : partset = get_matching_partitions(&pprune->initial_context,
2758 : pprune->initial_pruning_steps);
2759 1911 : else if (!initial_prune && pprune->exec_pruning_steps)
2760 1740 : partset = get_matching_partitions(&pprune->exec_context,
2761 : pprune->exec_pruning_steps);
2762 : else
2763 171 : partset = pprune->present_parts;
2764 :
2765 : /* Translate partset into subplan indexes */
2766 2177 : i = -1;
2767 3082 : while ((i = bms_next_member(partset, i)) >= 0)
2768 : {
2769 905 : if (pprune->subplan_map[i] >= 0)
2770 : {
2771 1394 : *validsubplans = bms_add_member(*validsubplans,
2772 697 : pprune->subplan_map[i]);
2773 :
2774 : /*
2775 : * Only report leaf partitions. Non-leaf partitions may appear
2776 : * here when they use an unflattened Append or MergeAppend.
2777 : */
2778 697 : if (validsubplan_rtis && pprune->leafpart_rti_map[i])
2779 337 : *validsubplan_rtis = bms_add_member(*validsubplan_rtis,
2780 337 : pprune->leafpart_rti_map[i]);
2781 : }
2782 : else
2783 : {
2784 208 : int partidx = pprune->subpart_map[i];
2785 :
2786 208 : if (partidx >= 0)
2787 207 : find_matching_subplans_recurse(prunedata,
2788 : &prunedata->partrelprunedata[partidx],
2789 : initial_prune, validsubplans,
2790 : validsubplan_rtis);
2791 : else
2792 : {
2793 : /*
2794 : * We get here if the planner already pruned all the sub-
2795 : * partitions for this partition. Silently ignore this
2796 : * partition in this case. The end result is the same: we
2797 : * would have pruned all partitions just the same, but we
2798 : * don't have any pruning steps to execute to verify this.
2799 : */
2800 : }
2801 : }
2802 : }
2803 2177 : }
|
