PostgreSQL Source Code git master
Loading...
Searching...
No Matches
joinpath.c
Go to the documentation of this file.
1/*-------------------------------------------------------------------------
2 *
3 * joinpath.c
4 * Routines to find all possible paths for processing a set of joins
5 *
6 * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
8 *
9 *
10 * IDENTIFICATION
11 * src/backend/optimizer/path/joinpath.c
12 *
13 *-------------------------------------------------------------------------
14 */
15#include "postgres.h"
16
17#include "executor/executor.h"
18#include "foreign/fdwapi.h"
19#include "nodes/nodeFuncs.h"
20#include "optimizer/cost.h"
21#include "optimizer/optimizer.h"
22#include "optimizer/pathnode.h"
23#include "optimizer/paths.h"
24#include "optimizer/placeholder.h"
25#include "optimizer/planmain.h"
26#include "optimizer/restrictinfo.h"
27#include "utils/lsyscache.h"
28#include "utils/typcache.h"
29
30/* Hooks for plugins to get control in add_paths_to_joinrel() */
31set_join_pathlist_hook_type set_join_pathlist_hook = NULL;
32join_path_setup_hook_type join_path_setup_hook = NULL;
33
34/*
35 * Paths parameterized by a parent rel can be considered to be parameterized
36 * by any of its children, when we are performing partitionwise joins. These
37 * macros simplify checking for such cases. Beware multiple eval of args.
38 */
39#define PATH_PARAM_BY_PARENT(path, rel) \
40 ((path)->param_info && bms_overlap(PATH_REQ_OUTER(path), \
41 (rel)->top_parent_relids))
42#define PATH_PARAM_BY_REL_SELF(path, rel) \
43 ((path)->param_info && bms_overlap(PATH_REQ_OUTER(path), (rel)->relids))
44
45#define PATH_PARAM_BY_REL(path, rel) \
46 (PATH_PARAM_BY_REL_SELF(path, rel) || PATH_PARAM_BY_PARENT(path, rel))
47
48static void try_partial_mergejoin_path(PlannerInfo *root,
49 RelOptInfo *joinrel,
50 Path *outer_path,
51 Path *inner_path,
52 List *pathkeys,
53 List *mergeclauses,
54 List *outersortkeys,
55 List *innersortkeys,
56 JoinType jointype,
57 JoinPathExtraData *extra);
58static void sort_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel,
59 RelOptInfo *outerrel, RelOptInfo *innerrel,
60 JoinType jointype, JoinPathExtraData *extra);
61static void match_unsorted_outer(PlannerInfo *root, RelOptInfo *joinrel,
62 RelOptInfo *outerrel, RelOptInfo *innerrel,
63 JoinType jointype, JoinPathExtraData *extra);
64static void consider_parallel_nestloop(PlannerInfo *root,
65 RelOptInfo *joinrel,
66 RelOptInfo *outerrel,
67 RelOptInfo *innerrel,
68 JoinType jointype,
69 JoinPathExtraData *extra);
70static void consider_parallel_mergejoin(PlannerInfo *root,
71 RelOptInfo *joinrel,
72 RelOptInfo *outerrel,
73 RelOptInfo *innerrel,
74 JoinType jointype,
75 JoinPathExtraData *extra,
76 Path *inner_cheapest_total);
77static void hash_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel,
78 RelOptInfo *outerrel, RelOptInfo *innerrel,
79 JoinType jointype, JoinPathExtraData *extra);
80static List *select_mergejoin_clauses(PlannerInfo *root,
81 RelOptInfo *joinrel,
82 RelOptInfo *outerrel,
83 RelOptInfo *innerrel,
84 List *restrictlist,
85 JoinType jointype,
86 bool *mergejoin_allowed);
87static void generate_mergejoin_paths(PlannerInfo *root,
88 RelOptInfo *joinrel,
89 RelOptInfo *innerrel,
90 Path *outerpath,
91 JoinType jointype,
92 JoinPathExtraData *extra,
93 bool useallclauses,
94 Path *inner_cheapest_total,
95 List *merge_pathkeys,
96 bool is_partial);
97
98
99/*
100 * add_paths_to_joinrel
101 * Given a join relation and two component rels from which it can be made,
102 * consider all possible paths that use the two component rels as outer
103 * and inner rel respectively. Add these paths to the join rel's pathlist
104 * if they survive comparison with other paths (and remove any existing
105 * paths that are dominated by these paths).
106 *
107 * Modifies the pathlist field of the joinrel node to contain the best
108 * paths found so far.
109 *
110 * jointype is not necessarily the same as sjinfo->jointype; it might be
111 * "flipped around" if we are considering joining the rels in the opposite
112 * direction from what's indicated in sjinfo.
113 *
114 * Also, this routine accepts the special JoinTypes JOIN_UNIQUE_OUTER and
115 * JOIN_UNIQUE_INNER to indicate that the outer or inner relation has been
116 * unique-ified and a regular inner join should then be applied. These values
117 * are not allowed to propagate outside this routine, however. Path cost
118 * estimation code, as well as match_unsorted_outer, may need to recognize that
119 * it's dealing with such a case --- the combination of nominal jointype INNER
120 * with sjinfo->jointype == JOIN_SEMI indicates that.
121 */
122void
123add_paths_to_joinrel(PlannerInfo *root,
124 RelOptInfo *joinrel,
125 RelOptInfo *outerrel,
126 RelOptInfo *innerrel,
127 JoinType jointype,
128 SpecialJoinInfo *sjinfo,
129 List *restrictlist)
130{
131 JoinType save_jointype = jointype;
132 JoinPathExtraData extra;
133 bool mergejoin_allowed = true;
134 ListCell *lc;
135 Relids joinrelids;
136
137 /*
138 * PlannerInfo doesn't contain the SpecialJoinInfos created for joins
139 * between child relations, even if there is a SpecialJoinInfo node for
140 * the join between the topmost parents. So, while calculating Relids set
141 * representing the restriction, consider relids of topmost parent of
142 * partitions.
143 */
144 if (joinrel->reloptkind == RELOPT_OTHER_JOINREL)
145 joinrelids = joinrel->top_parent_relids;
146 else
147 joinrelids = joinrel->relids;
148
149 extra.restrictlist = restrictlist;
150 extra.mergeclause_list = NIL;
151 extra.sjinfo = sjinfo;
152 extra.param_source_rels = NULL;
153 extra.pgs_mask = joinrel->pgs_mask;
154
155 /*
156 * Give extensions a chance to take control. In particular, an extension
157 * might want to modify extra.pgs_mask. It's possible to override pgs_mask
158 * on a query-wide basis using join_search_hook, or for a particular
159 * relation using joinrel_setup_hook, but extensions that want to provide
160 * different advice for the same joinrel based on the choice of innerrel
161 * and outerrel will need to use this hook.
162 *
163 * A very simple way for an extension to use this hook is to set
164 * extra.pgs_mask &= ~PGS_JOIN_ANY, if it simply doesn't want any of the
165 * paths generated by this call to add_paths_to_joinrel() to be selected.
166 * An extension could use this technique to constrain the join order,
167 * since it could thereby arrange to reject all paths from join orders
168 * that it does not like. An extension can also selectively clear bits
169 * from extra.pgs_mask to rule out specific techniques for specific joins,
170 * or could even set additional bits to re-allow methods disabled at some
171 * higher level.
172 *
173 * NB: Below this point, this function should be careful to reference
174 * extra.pgs_mask rather than rel->pgs_mask to avoid disregarding any
175 * changes made by the hook we're about to call.
176 */
177 if (join_path_setup_hook)
178 join_path_setup_hook(root, joinrel, outerrel, innerrel,
179 jointype, &extra);
180
181 /*
182 * See if the inner relation is provably unique for this outer rel.
183 *
184 * We have some special cases: for JOIN_SEMI, it doesn't matter since the
185 * executor can make the equivalent optimization anyway. It also doesn't
186 * help enable use of Memoize, since a semijoin with a provably unique
187 * inner side should have been reduced to an inner join in that case.
188 * Therefore, we need not expend planner cycles on proofs. (For
189 * JOIN_ANTI, although it doesn't help the executor for the same reason,
190 * it can benefit Memoize paths.) For JOIN_UNIQUE_INNER, we must be
191 * considering a semijoin whose inner side is not provably unique (else
192 * reduce_unique_semijoins would've simplified it), so there's no point in
193 * calling innerrel_is_unique. However, if the LHS covers all of the
194 * semijoin's min_lefthand, then it's appropriate to set inner_unique
195 * because the unique relation produced by create_unique_paths will be
196 * unique relative to the LHS. (If we have an LHS that's only part of the
197 * min_lefthand, that is *not* true.) For JOIN_UNIQUE_OUTER, pass
198 * JOIN_INNER to avoid letting that value escape this module.
199 */
200 switch (jointype)
201 {
202 case JOIN_SEMI:
203 extra.inner_unique = false; /* well, unproven */
204 break;
205 case JOIN_UNIQUE_INNER:
206 extra.inner_unique = bms_is_subset(sjinfo->min_lefthand,
207 outerrel->relids);
208 break;
209 case JOIN_UNIQUE_OUTER:
210 extra.inner_unique = innerrel_is_unique(root,
211 joinrel->relids,
212 outerrel->relids,
213 innerrel,
214 JOIN_INNER,
215 restrictlist,
216 false);
217 break;
218 default:
219 extra.inner_unique = innerrel_is_unique(root,
220 joinrel->relids,
221 outerrel->relids,
222 innerrel,
223 jointype,
224 restrictlist,
225 false);
226 break;
227 }
228
229 /*
230 * If the outer or inner relation has been unique-ified, handle as a plain
231 * inner join.
232 */
233 if (jointype == JOIN_UNIQUE_OUTER || jointype == JOIN_UNIQUE_INNER)
234 jointype = JOIN_INNER;
235
236 /*
237 * Find potential mergejoin clauses. We can skip this if we are not
238 * interested in doing a mergejoin. However, mergejoin may be our only
239 * way of implementing a full outer join, so in that case we don't care
240 * whether mergejoins are disabled.
241 */
242 if ((extra.pgs_mask & PGS_MERGEJOIN_ANY) != 0 || jointype == JOIN_FULL)
243 extra.mergeclause_list = select_mergejoin_clauses(root,
244 joinrel,
245 outerrel,
246 innerrel,
247 restrictlist,
248 jointype,
249 &mergejoin_allowed);
250
251 /*
252 * If it's SEMI, ANTI, or inner_unique join, compute correction factors
253 * for cost estimation. These will be the same for all paths.
254 */
255 if (jointype == JOIN_SEMI || jointype == JOIN_ANTI || extra.inner_unique)
256 compute_semi_anti_join_factors(root, joinrel, outerrel, innerrel,
257 jointype, sjinfo, restrictlist,
258 &extra.semifactors);
259
260 /*
261 * Decide whether it's sensible to generate parameterized paths for this
262 * joinrel, and if so, which relations such paths should require. There
263 * is usually no need to create a parameterized result path unless there
264 * is a join order restriction that prevents joining one of our input rels
265 * directly to the parameter source rel instead of joining to the other
266 * input rel. (But see allow_star_schema_join().) This restriction
267 * reduces the number of parameterized paths we have to deal with at
268 * higher join levels, without compromising the quality of the resulting
269 * plan. We express the restriction as a Relids set that must overlap the
270 * parameterization of any proposed join path. Note: param_source_rels
271 * should contain only baserels, not OJ relids, so starting from
272 * all_baserels not all_query_rels is correct.
273 */
274 foreach(lc, root->join_info_list)
275 {
276 SpecialJoinInfo *sjinfo2 = (SpecialJoinInfo *) lfirst(lc);
277
278 /*
279 * SJ is relevant to this join if we have some part of its RHS
280 * (possibly not all of it), and haven't yet joined to its LHS. (This
281 * test is pretty simplistic, but should be sufficient considering the
282 * join has already been proven legal.) If the SJ is relevant, it
283 * presents constraints for joining to anything not in its RHS.
284 */
285 if (bms_overlap(joinrelids, sjinfo2->min_righthand) &&
286 !bms_overlap(joinrelids, sjinfo2->min_lefthand))
287 extra.param_source_rels = bms_join(extra.param_source_rels,
288 bms_difference(root->all_baserels,
289 sjinfo2->min_righthand));
290
291 /* full joins constrain both sides symmetrically */
292 if (sjinfo2->jointype == JOIN_FULL &&
293 bms_overlap(joinrelids, sjinfo2->min_lefthand) &&
294 !bms_overlap(joinrelids, sjinfo2->min_righthand))
295 extra.param_source_rels = bms_join(extra.param_source_rels,
296 bms_difference(root->all_baserels,
297 sjinfo2->min_lefthand));
298 }
299
300 /*
301 * However, when a LATERAL subquery is involved, there will simply not be
302 * any paths for the joinrel that aren't parameterized by whatever the
303 * subquery is parameterized by, unless its parameterization is resolved
304 * within the joinrel. So we might as well allow additional dependencies
305 * on whatever residual lateral dependencies the joinrel will have.
306 */
307 extra.param_source_rels = bms_add_members(extra.param_source_rels,
308 joinrel->lateral_relids);
309
310 /*
311 * 1. Consider mergejoin paths where both relations must be explicitly
312 * sorted. Skip this if we can't mergejoin.
313 */
314 if (mergejoin_allowed)
315 sort_inner_and_outer(root, joinrel, outerrel, innerrel,
316 jointype, &extra);
317
318 /*
319 * 2. Consider paths where the outer relation need not be explicitly
320 * sorted. This includes both nestloops and mergejoins where the outer
321 * path is already ordered. Again, skip this if we can't mergejoin.
322 * (That's okay because we know that nestloop can't handle
323 * right/right-anti/right-semi/full joins at all, so it wouldn't work in
324 * the prohibited cases either.)
325 */
326 if (mergejoin_allowed)
327 match_unsorted_outer(root, joinrel, outerrel, innerrel,
328 jointype, &extra);
329
330#ifdef NOT_USED
331
332 /*
333 * 3. Consider paths where the inner relation need not be explicitly
334 * sorted. This includes mergejoins only (nestloops were already built in
335 * match_unsorted_outer).
336 *
337 * Diked out as redundant 2/13/2000 -- tgl. There isn't any really
338 * significant difference between the inner and outer side of a mergejoin,
339 * so match_unsorted_inner creates no paths that aren't equivalent to
340 * those made by match_unsorted_outer when add_paths_to_joinrel() is
341 * invoked with the two rels given in the other order.
342 */
343 if (mergejoin_allowed)
344 match_unsorted_inner(root, joinrel, outerrel, innerrel,
345 jointype, &extra);
346#endif
347
348 /*
349 * 4. Consider paths where both outer and inner relations must be hashed
350 * before being joined. As above, when it's a full join, we must try this
351 * even when the path type is disabled, because it may be our only option.
352 */
353 if ((extra.pgs_mask & PGS_HASHJOIN) != 0 || jointype == JOIN_FULL)
354 hash_inner_and_outer(root, joinrel, outerrel, innerrel,
355 jointype, &extra);
356
357 /*
358 * 5. If inner and outer relations are foreign tables (or joins) belonging
359 * to the same server and assigned to the same user to check access
360 * permissions as, give the FDW a chance to push down joins.
361 */
362 if ((extra.pgs_mask & PGS_FOREIGNJOIN) != 0 && joinrel->fdwroutine &&
363 joinrel->fdwroutine->GetForeignJoinPaths)
364 joinrel->fdwroutine->GetForeignJoinPaths(root, joinrel,
365 outerrel, innerrel,
366 save_jointype, &extra);
367
368 /*
369 * 6. Finally, give extensions a chance to manipulate the path list. They
370 * could add new paths (such as CustomPaths) by calling add_path(), or
371 * add_partial_path() if parallel aware.
372 *
373 * In theory, extensions could also use this hook to delete or modify
374 * paths added by the core code, but in practice this is difficult to make
375 * work, since it's too late to get back any paths that have already been
376 * discarded by add_path() or add_partial_path(). If you're trying to
377 * suppress paths, consider using join_path_setup_hook instead.
378 */
379 if (set_join_pathlist_hook)
380 set_join_pathlist_hook(root, joinrel, outerrel, innerrel,
381 save_jointype, &extra);
382}
383
384/*
385 * We override the param_source_rels heuristic to accept nestloop paths in
386 * which the outer rel satisfies some but not all of the inner path's
387 * parameterization. This is necessary to get good plans for star-schema
388 * scenarios, in which a parameterized path for a large table may require
389 * parameters from multiple small tables that will not get joined directly to
390 * each other. We can handle that by stacking nestloops that have the small
391 * tables on the outside; but this breaks the rule the param_source_rels
392 * heuristic is based on, namely that parameters should not be passed down
393 * across joins unless there's a join-order-constraint-based reason to do so.
394 * So we ignore the param_source_rels restriction when this case applies.
395 *
396 * allow_star_schema_join() returns true if the param_source_rels restriction
397 * should be overridden, ie, it's okay to perform this join.
398 */
399static inline bool
400allow_star_schema_join(PlannerInfo *root,
401 Relids outerrelids,
402 Relids inner_paramrels)
403{
404 /*
405 * It's a star-schema case if the outer rel provides some but not all of
406 * the inner rel's parameterization.
407 */
408 return (bms_overlap(inner_paramrels, outerrelids) &&
409 bms_nonempty_difference(inner_paramrels, outerrelids));
410}
411
412/*
413 * If the parameterization is only partly satisfied by the outer rel,
414 * the unsatisfied part can't include any outer-join relids that could
415 * null rels of the satisfied part. That would imply that we're trying
416 * to use a clause involving a Var with nonempty varnullingrels at
417 * a join level where that value isn't yet computable.
418 *
419 * In practice, this test never finds a problem because earlier join order
420 * restrictions prevent us from attempting a join that would cause a problem.
421 * (That's unsurprising, because the code worked before we ever added
422 * outer-join relids to expression relids.) It still seems worth checking
423 * as a backstop, but we only do so in assert-enabled builds.
424 */
425#ifdef USE_ASSERT_CHECKING
426static inline bool
427have_unsafe_outer_join_ref(PlannerInfo *root,
428 Relids outerrelids,
429 Relids inner_paramrels)
430{
431 bool result = false;
432 Relids unsatisfied = bms_difference(inner_paramrels, outerrelids);
433 Relids satisfied = bms_intersect(inner_paramrels, outerrelids);
434
435 if (bms_overlap(unsatisfied, root->outer_join_rels))
436 {
437 ListCell *lc;
438
439 foreach(lc, root->join_info_list)
440 {
441 SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc);
442
443 if (!bms_is_member(sjinfo->ojrelid, unsatisfied))
444 continue; /* not relevant */
445 if (bms_overlap(satisfied, sjinfo->min_righthand) ||
446 (sjinfo->jointype == JOIN_FULL &&
447 bms_overlap(satisfied, sjinfo->min_lefthand)))
448 {
449 result = true; /* doesn't work */
450 break;
451 }
452 }
453 }
454
455 /* Waste no memory when we reject a path here */
456 bms_free(unsatisfied);
457 bms_free(satisfied);
458
459 return result;
460}
461#endif /* USE_ASSERT_CHECKING */
462
463/*
464 * paraminfo_get_equal_hashops
465 * Determine if the clauses in param_info and innerrel's lateral vars
466 * can be hashed.
467 * Returns true if hashing is possible, otherwise false.
468 *
469 * Additionally, on success we collect the outer expressions and the
470 * appropriate equality operators for each hashable parameter to innerrel.
471 * These are returned in parallel lists in *param_exprs and *operators.
472 * We also set *binary_mode to indicate whether strict binary matching is
473 * required.
474 */
475static bool
476paraminfo_get_equal_hashops(PlannerInfo *root, ParamPathInfo *param_info,
477 RelOptInfo *outerrel, RelOptInfo *innerrel,
478 List *ph_lateral_vars, List **param_exprs,
479 List **operators, bool *binary_mode)
480
481{
482 List *lateral_vars;
483 ListCell *lc;
484
485 *param_exprs = NIL;
486 *operators = NIL;
487 *binary_mode = false;
488
489 /* Add join clauses from param_info to the hash key */
490 if (param_info != NULL)
491 {
492 List *clauses = param_info->ppi_clauses;
493
494 foreach(lc, clauses)
495 {
496 RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
497 OpExpr *opexpr;
498 Node *expr;
499 Oid hasheqoperator;
500
501 opexpr = (OpExpr *) rinfo->clause;
502
503 /*
504 * Bail if the rinfo is not compatible. We need a join OpExpr
505 * with 2 args.
506 */
507 if (!IsA(opexpr, OpExpr) || list_length(opexpr->args) != 2 ||
508 !clause_sides_match_join(rinfo, outerrel->relids,
509 innerrel->relids))
510 {
511 list_free(*operators);
512 list_free(*param_exprs);
513 return false;
514 }
515
516 if (rinfo->outer_is_left)
517 {
518 expr = (Node *) linitial(opexpr->args);
519 hasheqoperator = rinfo->left_hasheqoperator;
520 }
521 else
522 {
523 expr = (Node *) lsecond(opexpr->args);
524 hasheqoperator = rinfo->right_hasheqoperator;
525 }
526
527 /* can't do memoize if we can't hash the outer type */
528 if (!OidIsValid(hasheqoperator))
529 {
530 list_free(*operators);
531 list_free(*param_exprs);
532 return false;
533 }
534
535 /*
536 * 'expr' may already exist as a parameter from a previous item in
537 * ppi_clauses. No need to include it again, however we'd better
538 * ensure we do switch into binary mode if required. See below.
539 */
540 if (!list_member(*param_exprs, expr))
541 {
542 *operators = lappend_oid(*operators, hasheqoperator);
543 *param_exprs = lappend(*param_exprs, expr);
544 }
545
546 /*
547 * When the join operator is not hashable then it's possible that
548 * the operator will be able to distinguish something that the
549 * hash equality operator could not. For example with floating
550 * point types -0.0 and +0.0 are classed as equal by the hash
551 * function and equality function, but some other operator may be
552 * able to tell those values apart. This means that we must put
553 * memoize into binary comparison mode so that it does bit-by-bit
554 * comparisons rather than a "logical" comparison as it would
555 * using the hash equality operator.
556 */
557 if (!OidIsValid(rinfo->hashjoinoperator))
558 *binary_mode = true;
559 }
560 }
561
562 /* Now add any lateral vars to the cache key too */
563 lateral_vars = list_concat(ph_lateral_vars, innerrel->lateral_vars);
564 foreach(lc, lateral_vars)
565 {
566 Node *expr = (Node *) lfirst(lc);
567 TypeCacheEntry *typentry;
568
569 /* Reject if there are any volatile functions in lateral vars */
570 if (contain_volatile_functions(expr))
571 {
572 list_free(*operators);
573 list_free(*param_exprs);
574 return false;
575 }
576
577 typentry = lookup_type_cache(exprType(expr),
578 TYPECACHE_HASH_PROC | TYPECACHE_EQ_OPR);
579
580 /* can't use memoize without a valid hash proc and equals operator */
581 if (!OidIsValid(typentry->hash_proc) || !OidIsValid(typentry->eq_opr))
582 {
583 list_free(*operators);
584 list_free(*param_exprs);
585 return false;
586 }
587
588 /*
589 * 'expr' may already exist as a parameter from the ppi_clauses. No
590 * need to include it again, however we'd better ensure we do switch
591 * into binary mode.
592 */
593 if (!list_member(*param_exprs, expr))
594 {
595 *operators = lappend_oid(*operators, typentry->eq_opr);
596 *param_exprs = lappend(*param_exprs, expr);
597 }
598
599 /*
600 * We must go into binary mode as we don't have too much of an idea of
601 * how these lateral Vars are being used. See comment above when we
602 * set *binary_mode for the non-lateral Var case. This could be
603 * relaxed a bit if we had the RestrictInfos and knew the operators
604 * being used, however for cases like Vars that are arguments to
605 * functions we must operate in binary mode as we don't have
606 * visibility into what the function is doing with the Vars.
607 */
608 *binary_mode = true;
609 }
610
611 /* We're okay to use memoize */
612 return true;
613}
614
615/*
616 * extract_lateral_vars_from_PHVs
617 * Extract lateral references within PlaceHolderVars that are due to be
618 * evaluated at 'innerrelids'.
619 */
620static List *
621extract_lateral_vars_from_PHVs(PlannerInfo *root, Relids innerrelids)
622{
623 List *ph_lateral_vars = NIL;
624 ListCell *lc;
625
626 /* Nothing would be found if the query contains no LATERAL RTEs */
627 if (!root->hasLateralRTEs)
628 return NIL;
629
630 /*
631 * No need to consider PHVs that are due to be evaluated at joinrels,
632 * since we do not add Memoize nodes on top of joinrel paths.
633 */
634 if (bms_membership(innerrelids) == BMS_MULTIPLE)
635 return NIL;
636
637 foreach(lc, root->placeholder_list)
638 {
639 PlaceHolderInfo *phinfo = (PlaceHolderInfo *) lfirst(lc);
640 List *vars;
641 ListCell *cell;
642
643 /* PHV is uninteresting if no lateral refs */
644 if (phinfo->ph_lateral == NULL)
645 continue;
646
647 /* PHV is uninteresting if not due to be evaluated at innerrelids */
648 if (!bms_equal(phinfo->ph_eval_at, innerrelids))
649 continue;
650
651 /*
652 * If the PHV does not reference any rels in innerrelids, use its
653 * contained expression as a cache key rather than extracting the
654 * Vars/PHVs from it and using those. This can be beneficial in cases
655 * where the expression results in fewer distinct values to cache
656 * tuples for.
657 */
658 if (!bms_overlap(pull_varnos(root, (Node *) phinfo->ph_var->phexpr),
659 innerrelids))
660 {
661 ph_lateral_vars = lappend(ph_lateral_vars, phinfo->ph_var->phexpr);
662 continue;
663 }
664
665 /* Fetch Vars and PHVs of lateral references within PlaceHolderVars */
666 vars = pull_vars_of_level((Node *) phinfo->ph_var->phexpr, 0);
667 foreach(cell, vars)
668 {
669 Node *node = (Node *) lfirst(cell);
670
671 if (IsA(node, Var))
672 {
673 Var *var = (Var *) node;
674
675 Assert(var->varlevelsup == 0);
676
677 if (bms_is_member(var->varno, phinfo->ph_lateral))
678 ph_lateral_vars = lappend(ph_lateral_vars, node);
679 }
680 else if (IsA(node, PlaceHolderVar))
681 {
682 PlaceHolderVar *phv = (PlaceHolderVar *) node;
683
684 Assert(phv->phlevelsup == 0);
685
686 if (bms_is_subset(find_placeholder_info(root, phv)->ph_eval_at,
687 phinfo->ph_lateral))
688 ph_lateral_vars = lappend(ph_lateral_vars, node);
689 }
690 else
691 Assert(false);
692 }
693
694 list_free(vars);
695 }
696
697 return ph_lateral_vars;
698}
699
700/*
701 * get_memoize_path
702 * If possible, make and return a Memoize path atop of 'inner_path'.
703 * Otherwise return NULL.
704 *
705 * Note that currently we do not add Memoize nodes on top of join relation
706 * paths. This is because the ParamPathInfos for join relation paths do not
707 * maintain ppi_clauses, as the set of relevant clauses varies depending on how
708 * the join is formed. In addition, joinrels do not maintain lateral_vars. So
709 * we do not have a way to extract cache keys from joinrels.
710 */
711static Path *
712get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
713 RelOptInfo *outerrel, Path *inner_path,
714 Path *outer_path, JoinType jointype,
715 JoinPathExtraData *extra)
716{
717 List *param_exprs;
718 List *hash_operators;
719 ListCell *lc;
720 bool binary_mode;
721 List *ph_lateral_vars;
722
723 /* Obviously not if it's disabled */
724 if ((extra->pgs_mask & PGS_NESTLOOP_MEMOIZE) == 0)
725 return NULL;
726
727 /*
728 * We can safely not bother with all this unless we expect to perform more
729 * than one inner scan. The first scan is always going to be a cache
730 * miss. This would likely fail later anyway based on costs, so this is
731 * really just to save some wasted effort.
732 */
733 if (outer_path->parent->rows < 2)
734 return NULL;
735
736 /*
737 * Extract lateral Vars/PHVs within PlaceHolderVars that are due to be
738 * evaluated at innerrel. These lateral Vars/PHVs could be used as
739 * memoize cache keys.
740 */
741 ph_lateral_vars = extract_lateral_vars_from_PHVs(root, innerrel->relids);
742
743 /*
744 * We can only have a memoize node when there's some kind of cache key,
745 * either parameterized path clauses or lateral Vars. No cache key sounds
746 * more like something a Materialize node might be more useful for.
747 */
748 if ((inner_path->param_info == NULL ||
749 inner_path->param_info->ppi_clauses == NIL) &&
750 innerrel->lateral_vars == NIL &&
751 ph_lateral_vars == NIL)
752 return NULL;
753
754 /*
755 * Currently we don't do this for SEMI and ANTI joins, because nested loop
756 * SEMI/ANTI joins don't scan the inner node to completion, which means
757 * memoize cannot mark the cache entry as complete. Nor can we mark the
758 * cache entry as complete after fetching the first inner tuple, because
759 * if that tuple and the current outer tuple don't satisfy the join
760 * clauses, a second inner tuple that satisfies the parameters would find
761 * the cache entry already marked as complete. The only exception is when
762 * the inner relation is provably unique, as in that case, there won't be
763 * a second matching tuple and we can safely mark the cache entry as
764 * complete after fetching the first inner tuple. Note that in such
765 * cases, the SEMI join should have been reduced to an inner join by
766 * reduce_unique_semijoins.
767 */
768 if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
769 !extra->inner_unique)
770 return NULL;
771
772 /*
773 * Memoize normally marks cache entries as complete when it runs out of
774 * tuples to read from its subplan. However, with unique joins, Nested
775 * Loop will skip to the next outer tuple after finding the first matching
776 * inner tuple. This means that we may not read the inner side of the
777 * join to completion which leaves no opportunity to mark the cache entry
778 * as complete. To work around that, when the join is unique we
779 * automatically mark cache entries as complete after fetching the first
780 * tuple. This works when the entire join condition is parameterized.
781 * Otherwise, when the parameterization is only a subset of the join
782 * condition, we can't be sure which part of it causes the join to be
783 * unique. This means there are no guarantees that only 1 tuple will be
784 * read. We cannot mark the cache entry as complete after reading the
785 * first tuple without that guarantee. This means the scope of Memoize
786 * node's usefulness is limited to only outer rows that have no join
787 * partner as this is the only case where Nested Loop would exhaust the
788 * inner scan of a unique join. Since the scope is limited to that, we
789 * just don't bother making a memoize path in this case.
790 *
791 * Lateral vars needn't be considered here as they're not considered when
792 * determining if the join is unique.
793 */
794 if (extra->inner_unique)
795 {
796 Bitmapset *ppi_serials;
797
798 if (inner_path->param_info == NULL)
799 return NULL;
800
801 ppi_serials = inner_path->param_info->ppi_serials;
802
803 foreach_node(RestrictInfo, rinfo, extra->restrictlist)
804 {
805 if (!bms_is_member(rinfo->rinfo_serial, ppi_serials))
806 return NULL;
807 }
808 }
809
810 /*
811 * We can't use a memoize node if there are volatile functions in the
812 * inner rel's target list or restrict list. A cache hit could reduce the
813 * number of calls to these functions.
814 */
815 if (contain_volatile_functions((Node *) innerrel->reltarget))
816 return NULL;
817
818 foreach(lc, innerrel->baserestrictinfo)
819 {
820 RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
821
822 if (contain_volatile_functions((Node *) rinfo))
823 return NULL;
824 }
825
826 /*
827 * Also check the parameterized path restrictinfos for volatile functions.
828 * Indexed functions must be immutable so shouldn't have any volatile
829 * functions, however, with a lateral join the inner scan may not be an
830 * index scan.
831 */
832 if (inner_path->param_info != NULL)
833 {
834 foreach(lc, inner_path->param_info->ppi_clauses)
835 {
836 RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
837
838 if (contain_volatile_functions((Node *) rinfo))
839 return NULL;
840 }
841 }
842
843 /* Check if we have hash ops for each parameter to the path */
844 if (paraminfo_get_equal_hashops(root,
845 inner_path->param_info,
846 outerrel->top_parent ?
847 outerrel->top_parent : outerrel,
848 innerrel,
849 ph_lateral_vars,
850 &param_exprs,
851 &hash_operators,
852 &binary_mode))
853 {
854 return (Path *) create_memoize_path(root,
855 innerrel,
856 inner_path,
857 param_exprs,
858 hash_operators,
859 extra->inner_unique,
860 binary_mode,
861 outer_path->rows);
862 }
863
864 return NULL;
865}
866
867/*
868 * try_nestloop_path
869 * Consider a nestloop join path; if it appears useful, push it into
870 * the joinrel's pathlist via add_path().
871 */
872static void
873try_nestloop_path(PlannerInfo *root,
874 RelOptInfo *joinrel,
875 Path *outer_path,
876 Path *inner_path,
877 List *pathkeys,
878 JoinType jointype,
879 uint64 nestloop_subtype,
880 JoinPathExtraData *extra)
881{
882 Relids required_outer;
883 JoinCostWorkspace workspace;
884 RelOptInfo *innerrel = inner_path->parent;
885 RelOptInfo *outerrel = outer_path->parent;
886 Relids innerrelids;
887 Relids outerrelids;
888 Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
889 Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
890
891 /*
892 * If we are forming an outer join at this join, it's nonsensical to use
893 * an input path that uses the outer join as part of its parameterization.
894 * (This can happen despite our join order restrictions, since those apply
895 * to what is in an input relation not what its parameters are.)
896 */
897 if (extra->sjinfo->ojrelid != 0 &&
898 (bms_is_member(extra->sjinfo->ojrelid, inner_paramrels) ||
899 bms_is_member(extra->sjinfo->ojrelid, outer_paramrels)))
900 return;
901
902 /*
903 * Any parameterization of the input paths refers to topmost parents of
904 * the relevant relations, because reparameterize_path_by_child() hasn't
905 * been called yet. So we must consider topmost parents of the relations
906 * being joined, too, while determining parameterization of the result and
907 * checking for disallowed parameterization cases.
908 */
909 if (innerrel->top_parent_relids)
910 innerrelids = innerrel->top_parent_relids;
911 else
912 innerrelids = innerrel->relids;
913
914 if (outerrel->top_parent_relids)
915 outerrelids = outerrel->top_parent_relids;
916 else
917 outerrelids = outerrel->relids;
918
919 /*
920 * Check to see if proposed path is still parameterized, and reject if the
921 * parameterization wouldn't be sensible --- unless allow_star_schema_join
922 * says to allow it anyway.
923 */
924 required_outer = calc_nestloop_required_outer(outerrelids, outer_paramrels,
925 innerrelids, inner_paramrels);
926 if (required_outer &&
927 !bms_overlap(required_outer, extra->param_source_rels) &&
928 !allow_star_schema_join(root, outerrelids, inner_paramrels))
929 {
930 /* Waste no memory when we reject a path here */
931 bms_free(required_outer);
932 return;
933 }
934
935 /* If we got past that, we shouldn't have any unsafe outer-join refs */
936 Assert(!have_unsafe_outer_join_ref(root, outerrelids, inner_paramrels));
937
938 /*
939 * If the inner path is parameterized, it is parameterized by the topmost
940 * parent of the outer rel, not the outer rel itself. We will need to
941 * translate the parameterization, if this path is chosen, during
942 * create_plan(). Here we just check whether we will be able to perform
943 * the translation, and if not avoid creating a nestloop path.
944 */
945 if (PATH_PARAM_BY_PARENT(inner_path, outer_path->parent) &&
946 !path_is_reparameterizable_by_child(inner_path, outer_path->parent))
947 {
948 bms_free(required_outer);
949 return;
950 }
951
952 /*
953 * Do a precheck to quickly eliminate obviously-inferior paths. We
954 * calculate a cheap lower bound on the path's cost and then use
955 * add_path_precheck() to see if the path is clearly going to be dominated
956 * by some existing path for the joinrel. If not, do the full pushup with
957 * creating a fully valid path structure and submitting it to add_path().
958 * The latter two steps are expensive enough to make this two-phase
959 * methodology worthwhile.
960 */
961 initial_cost_nestloop(root, &workspace, jointype,
962 nestloop_subtype | PGS_CONSIDER_NONPARTIAL,
963 outer_path, inner_path, extra);
964
965 if (add_path_precheck(joinrel, workspace.disabled_nodes,
966 workspace.startup_cost, workspace.total_cost,
967 pathkeys, required_outer))
968 {
969 add_path(joinrel, (Path *)
970 create_nestloop_path(root,
971 joinrel,
972 jointype,
973 &workspace,
974 extra,
975 outer_path,
976 inner_path,
977 extra->restrictlist,
978 pathkeys,
979 required_outer));
980 }
981 else
982 {
983 /* Waste no memory when we reject a path here */
984 bms_free(required_outer);
985 }
986}
987
988/*
989 * try_partial_nestloop_path
990 * Consider a partial nestloop join path; if it appears useful, push it into
991 * the joinrel's partial_pathlist via add_partial_path().
992 */
993static void
994try_partial_nestloop_path(PlannerInfo *root,
995 RelOptInfo *joinrel,
996 Path *outer_path,
997 Path *inner_path,
998 List *pathkeys,
999 JoinType jointype,
1000 uint64 nestloop_subtype,
1001 JoinPathExtraData *extra)
1002{
1003 JoinCostWorkspace workspace;
1004
1005 /*
1006 * If the inner path is parameterized, the parameterization must be fully
1007 * satisfied by the proposed outer path. Parameterized partial paths are
1008 * not supported. The caller should already have verified that no lateral
1009 * rels are required here.
1010 */
1011 Assert(bms_is_empty(joinrel->lateral_relids));
1012 Assert(bms_is_empty(PATH_REQ_OUTER(outer_path)));
1013 if (inner_path->param_info != NULL)
1014 {
1015 Relids inner_paramrels = inner_path->param_info->ppi_req_outer;
1016 RelOptInfo *outerrel = outer_path->parent;
1017 Relids outerrelids;
1018
1019 /*
1020 * The inner and outer paths are parameterized, if at all, by the top
1021 * level parents, not the child relations, so we must use those relids
1022 * for our parameterization tests.
1023 */
1024 if (outerrel->top_parent_relids)
1025 outerrelids = outerrel->top_parent_relids;
1026 else
1027 outerrelids = outerrel->relids;
1028
1029 if (!bms_is_subset(inner_paramrels, outerrelids))
1030 return;
1031 }
1032
1033 /*
1034 * If the inner path is parameterized, it is parameterized by the topmost
1035 * parent of the outer rel, not the outer rel itself. We will need to
1036 * translate the parameterization, if this path is chosen, during
1037 * create_plan(). Here we just check whether we will be able to perform
1038 * the translation, and if not avoid creating a nestloop path.
1039 */
1040 if (PATH_PARAM_BY_PARENT(inner_path, outer_path->parent) &&
1041 !path_is_reparameterizable_by_child(inner_path, outer_path->parent))
1042 return;
1043
1044 /*
1045 * Before creating a path, get a quick lower bound on what it is likely to
1046 * cost. Bail out right away if it looks terrible.
1047 */
1048 initial_cost_nestloop(root, &workspace, jointype, nestloop_subtype,
1049 outer_path, inner_path, extra);
1050 if (!add_partial_path_precheck(joinrel, workspace.disabled_nodes,
1051 workspace.total_cost, pathkeys))
1052 return;
1053
1054 /* Might be good enough to be worth trying, so let's try it. */
1055 add_partial_path(joinrel, (Path *)
1056 create_nestloop_path(root,
1057 joinrel,
1058 jointype,
1059 &workspace,
1060 extra,
1061 outer_path,
1062 inner_path,
1063 extra->restrictlist,
1064 pathkeys,
1065 NULL));
1066}
1067
1068/*
1069 * try_mergejoin_path
1070 * Consider a merge join path; if it appears useful, push it into
1071 * the joinrel's pathlist via add_path().
1072 */
1073static void
1074try_mergejoin_path(PlannerInfo *root,
1075 RelOptInfo *joinrel,
1076 Path *outer_path,
1077 Path *inner_path,
1078 List *pathkeys,
1079 List *mergeclauses,
1080 List *outersortkeys,
1081 List *innersortkeys,
1082 JoinType jointype,
1083 JoinPathExtraData *extra,
1084 bool is_partial)
1085{
1086 Relids required_outer;
1087 int outer_presorted_keys = 0;
1088 JoinCostWorkspace workspace;
1089
1090 if (is_partial)
1091 {
1092 try_partial_mergejoin_path(root,
1093 joinrel,
1094 outer_path,
1095 inner_path,
1096 pathkeys,
1097 mergeclauses,
1098 outersortkeys,
1099 innersortkeys,
1100 jointype,
1101 extra);
1102 return;
1103 }
1104
1105 /*
1106 * If we are forming an outer join at this join, it's nonsensical to use
1107 * an input path that uses the outer join as part of its parameterization.
1108 * (This can happen despite our join order restrictions, since those apply
1109 * to what is in an input relation not what its parameters are.)
1110 */
1111 if (extra->sjinfo->ojrelid != 0 &&
1112 (bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(inner_path)) ||
1113 bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(outer_path))))
1114 return;
1115
1116 /*
1117 * Check to see if proposed path is still parameterized, and reject if the
1118 * parameterization wouldn't be sensible.
1119 */
1120 required_outer = calc_non_nestloop_required_outer(outer_path,
1121 inner_path);
1122 if (required_outer &&
1123 !bms_overlap(required_outer, extra->param_source_rels))
1124 {
1125 /* Waste no memory when we reject a path here */
1126 bms_free(required_outer);
1127 return;
1128 }
1129
1130 /*
1131 * If the given paths are already well enough ordered, we can skip doing
1132 * an explicit sort.
1133 *
1134 * We need to determine the number of presorted keys of the outer path to
1135 * decide whether explicit incremental sort can be applied when
1136 * outersortkeys is not NIL. We do not need to do the same for the inner
1137 * path though, as incremental sort currently does not support
1138 * mark/restore.
1139 */
1140 if (outersortkeys &&
1141 pathkeys_count_contained_in(outersortkeys, outer_path->pathkeys,
1142 &outer_presorted_keys))
1143 outersortkeys = NIL;
1144 if (innersortkeys &&
1145 pathkeys_contained_in(innersortkeys, inner_path->pathkeys))
1146 innersortkeys = NIL;
1147
1148 /*
1149 * See comments in try_nestloop_path().
1150 */
1151 initial_cost_mergejoin(root, &workspace, jointype, mergeclauses,
1152 outer_path, inner_path,
1153 outersortkeys, innersortkeys,
1154 outer_presorted_keys,
1155 extra);
1156
1157 if (add_path_precheck(joinrel, workspace.disabled_nodes,
1158 workspace.startup_cost, workspace.total_cost,
1159 pathkeys, required_outer))
1160 {
1161 add_path(joinrel, (Path *)
1162 create_mergejoin_path(root,
1163 joinrel,
1164 jointype,
1165 &workspace,
1166 extra,
1167 outer_path,
1168 inner_path,
1169 extra->restrictlist,
1170 pathkeys,
1171 required_outer,
1172 mergeclauses,
1173 outersortkeys,
1174 innersortkeys,
1175 outer_presorted_keys));
1176 }
1177 else
1178 {
1179 /* Waste no memory when we reject a path here */
1180 bms_free(required_outer);
1181 }
1182}
1183
1184/*
1185 * try_partial_mergejoin_path
1186 * Consider a partial merge join path; if it appears useful, push it into
1187 * the joinrel's pathlist via add_partial_path().
1188 */
1189static void
1190try_partial_mergejoin_path(PlannerInfo *root,
1191 RelOptInfo *joinrel,
1192 Path *outer_path,
1193 Path *inner_path,
1194 List *pathkeys,
1195 List *mergeclauses,
1196 List *outersortkeys,
1197 List *innersortkeys,
1198 JoinType jointype,
1199 JoinPathExtraData *extra)
1200{
1201 int outer_presorted_keys = 0;
1202 JoinCostWorkspace workspace;
1203
1204 /*
1205 * See comments in try_partial_hashjoin_path().
1206 */
1207 Assert(bms_is_empty(joinrel->lateral_relids));
1208 Assert(bms_is_empty(PATH_REQ_OUTER(outer_path)));
1209 if (!bms_is_empty(PATH_REQ_OUTER(inner_path)))
1210 return;
1211
1212 /*
1213 * If the given paths are already well enough ordered, we can skip doing
1214 * an explicit sort.
1215 *
1216 * We need to determine the number of presorted keys of the outer path to
1217 * decide whether explicit incremental sort can be applied when
1218 * outersortkeys is not NIL. We do not need to do the same for the inner
1219 * path though, as incremental sort currently does not support
1220 * mark/restore.
1221 */
1222 if (outersortkeys &&
1223 pathkeys_count_contained_in(outersortkeys, outer_path->pathkeys,
1224 &outer_presorted_keys))
1225 outersortkeys = NIL;
1226 if (innersortkeys &&
1227 pathkeys_contained_in(innersortkeys, inner_path->pathkeys))
1228 innersortkeys = NIL;
1229
1230 /*
1231 * See comments in try_partial_nestloop_path().
1232 */
1233 initial_cost_mergejoin(root, &workspace, jointype, mergeclauses,
1234 outer_path, inner_path,
1235 outersortkeys, innersortkeys,
1236 outer_presorted_keys,
1237 extra);
1238
1239 if (!add_partial_path_precheck(joinrel, workspace.disabled_nodes,
1240 workspace.total_cost, pathkeys))
1241 return;
1242
1243 /* Might be good enough to be worth trying, so let's try it. */
1244 add_partial_path(joinrel, (Path *)
1245 create_mergejoin_path(root,
1246 joinrel,
1247 jointype,
1248 &workspace,
1249 extra,
1250 outer_path,
1251 inner_path,
1252 extra->restrictlist,
1253 pathkeys,
1254 NULL,
1255 mergeclauses,
1256 outersortkeys,
1257 innersortkeys,
1258 outer_presorted_keys));
1259}
1260
1261/*
1262 * try_hashjoin_path
1263 * Consider a hash join path; if it appears useful, push it into
1264 * the joinrel's pathlist via add_path().
1265 */
1266static void
1267try_hashjoin_path(PlannerInfo *root,
1268 RelOptInfo *joinrel,
1269 Path *outer_path,
1270 Path *inner_path,
1271 List *hashclauses,
1272 JoinType jointype,
1273 JoinPathExtraData *extra)
1274{
1275 Relids required_outer;
1276 JoinCostWorkspace workspace;
1277
1278 /*
1279 * If we are forming an outer join at this join, it's nonsensical to use
1280 * an input path that uses the outer join as part of its parameterization.
1281 * (This can happen despite our join order restrictions, since those apply
1282 * to what is in an input relation not what its parameters are.)
1283 */
1284 if (extra->sjinfo->ojrelid != 0 &&
1285 (bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(inner_path)) ||
1286 bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(outer_path))))
1287 return;
1288
1289 /*
1290 * Check to see if proposed path is still parameterized, and reject if the
1291 * parameterization wouldn't be sensible.
1292 */
1293 required_outer = calc_non_nestloop_required_outer(outer_path,
1294 inner_path);
1295 if (required_outer &&
1296 !bms_overlap(required_outer, extra->param_source_rels))
1297 {
1298 /* Waste no memory when we reject a path here */
1299 bms_free(required_outer);
1300 return;
1301 }
1302
1303 /*
1304 * See comments in try_nestloop_path(). Also note that hashjoin paths
1305 * never have any output pathkeys, per comments in create_hashjoin_path.
1306 */
1307 initial_cost_hashjoin(root, &workspace, jointype, hashclauses,
1308 outer_path, inner_path, extra, false);
1309
1310 if (add_path_precheck(joinrel, workspace.disabled_nodes,
1311 workspace.startup_cost, workspace.total_cost,
1312 NIL, required_outer))
1313 {
1314 add_path(joinrel, (Path *)
1315 create_hashjoin_path(root,
1316 joinrel,
1317 jointype,
1318 &workspace,
1319 extra,
1320 outer_path,
1321 inner_path,
1322 false, /* parallel_hash */
1323 extra->restrictlist,
1324 required_outer,
1325 hashclauses));
1326 }
1327 else
1328 {
1329 /* Waste no memory when we reject a path here */
1330 bms_free(required_outer);
1331 }
1332}
1333
1334/*
1335 * try_partial_hashjoin_path
1336 * Consider a partial hashjoin join path; if it appears useful, push it into
1337 * the joinrel's partial_pathlist via add_partial_path().
1338 * The outer side is partial. If parallel_hash is true, then the inner path
1339 * must be partial and will be run in parallel to create one or more shared
1340 * hash tables; otherwise the inner path must be complete and a copy of it
1341 * is run in every process to create separate identical private hash tables.
1342 */
1343static void
1344try_partial_hashjoin_path(PlannerInfo *root,
1345 RelOptInfo *joinrel,
1346 Path *outer_path,
1347 Path *inner_path,
1348 List *hashclauses,
1349 JoinType jointype,
1350 JoinPathExtraData *extra,
1351 bool parallel_hash)
1352{
1353 JoinCostWorkspace workspace;
1354
1355 /*
1356 * If the inner path is parameterized, we can't use a partial hashjoin.
1357 * Parameterized partial paths are not supported. The caller should
1358 * already have verified that no lateral rels are required here.
1359 */
1360 Assert(bms_is_empty(joinrel->lateral_relids));
1361 Assert(bms_is_empty(PATH_REQ_OUTER(outer_path)));
1362 if (!bms_is_empty(PATH_REQ_OUTER(inner_path)))
1363 return;
1364
1365 /*
1366 * Before creating a path, get a quick lower bound on what it is likely to
1367 * cost. Bail out right away if it looks terrible.
1368 */
1369 initial_cost_hashjoin(root, &workspace, jointype, hashclauses,
1370 outer_path, inner_path, extra, parallel_hash);
1371 if (!add_partial_path_precheck(joinrel, workspace.disabled_nodes,
1372 workspace.total_cost, NIL))
1373 return;
1374
1375 /* Might be good enough to be worth trying, so let's try it. */
1376 add_partial_path(joinrel, (Path *)
1377 create_hashjoin_path(root,
1378 joinrel,
1379 jointype,
1380 &workspace,
1381 extra,
1382 outer_path,
1383 inner_path,
1384 parallel_hash,
1385 extra->restrictlist,
1386 NULL,
1387 hashclauses));
1388}
1389
1390/*
1391 * sort_inner_and_outer
1392 * Create mergejoin join paths by explicitly sorting both the outer and
1393 * inner join relations on each available merge ordering.
1394 *
1395 * 'joinrel' is the join relation
1396 * 'outerrel' is the outer join relation
1397 * 'innerrel' is the inner join relation
1398 * 'jointype' is the type of join to do
1399 * 'extra' contains additional input values
1400 */
1401static void
1402sort_inner_and_outer(PlannerInfo *root,
1403 RelOptInfo *joinrel,
1404 RelOptInfo *outerrel,
1405 RelOptInfo *innerrel,
1406 JoinType jointype,
1407 JoinPathExtraData *extra)
1408{
1409 Path *outer_path;
1410 Path *inner_path;
1411 Path *cheapest_partial_outer = NULL;
1412 Path *cheapest_safe_inner = NULL;
1413 List *all_pathkeys;
1414 ListCell *l;
1415
1416 /* Nothing to do if there are no available mergejoin clauses */
1417 if (extra->mergeclause_list == NIL)
1418 return;
1419
1420 /*
1421 * We only consider the cheapest-total-cost input paths, since we are
1422 * assuming here that a sort is required. We will consider
1423 * cheapest-startup-cost input paths later, and only if they don't need a
1424 * sort.
1425 *
1426 * This function intentionally does not consider parameterized input
1427 * paths, except when the cheapest-total is parameterized. If we did so,
1428 * we'd have a combinatorial explosion of mergejoin paths of dubious
1429 * value. This interacts with decisions elsewhere that also discriminate
1430 * against mergejoins with parameterized inputs; see comments in
1431 * src/backend/optimizer/README.
1432 */
1433 outer_path = outerrel->cheapest_total_path;
1434 inner_path = innerrel->cheapest_total_path;
1435
1436 /*
1437 * If either cheapest-total path is parameterized by the other rel, we
1438 * can't use a mergejoin. (There's no use looking for alternative input
1439 * paths, since these should already be the least-parameterized available
1440 * paths.)
1441 */
1442 if (PATH_PARAM_BY_REL(outer_path, innerrel) ||
1443 PATH_PARAM_BY_REL(inner_path, outerrel))
1444 return;
1445
1446 /*
1447 * If the joinrel is parallel-safe, we may be able to consider a partial
1448 * merge join. However, we can't handle JOIN_FULL, JOIN_RIGHT and
1449 * JOIN_RIGHT_ANTI, because they can produce false null extended rows.
1450 * Also, the resulting path must not be parameterized.
1451 */
1452 if (joinrel->consider_parallel &&
1453 jointype != JOIN_FULL &&
1454 jointype != JOIN_RIGHT &&
1455 jointype != JOIN_RIGHT_ANTI &&
1456 outerrel->partial_pathlist != NIL &&
1457 bms_is_empty(joinrel->lateral_relids))
1458 {
1459 cheapest_partial_outer = (Path *) linitial(outerrel->partial_pathlist);
1460
1461 if (inner_path->parallel_safe)
1462 cheapest_safe_inner = inner_path;
1463 else
1464 cheapest_safe_inner =
1465 get_cheapest_parallel_safe_total_inner(innerrel->pathlist);
1466 }
1467
1468 /*
1469 * Each possible ordering of the available mergejoin clauses will generate
1470 * a differently-sorted result path at essentially the same cost. We have
1471 * no basis for choosing one over another at this level of joining, but
1472 * some sort orders may be more useful than others for higher-level
1473 * mergejoins, so it's worth considering multiple orderings.
1474 *
1475 * Actually, it's not quite true that every mergeclause ordering will
1476 * generate a different path order, because some of the clauses may be
1477 * partially redundant (refer to the same EquivalenceClasses). Therefore,
1478 * what we do is convert the mergeclause list to a list of canonical
1479 * pathkeys, and then consider different orderings of the pathkeys.
1480 *
1481 * Generating a path for *every* permutation of the pathkeys doesn't seem
1482 * like a winning strategy; the cost in planning time is too high. For
1483 * now, we generate one path for each pathkey, listing that pathkey first
1484 * and the rest in random order. This should allow at least a one-clause
1485 * mergejoin without re-sorting against any other possible mergejoin
1486 * partner path. But if we've not guessed the right ordering of secondary
1487 * keys, we may end up evaluating clauses as qpquals when they could have
1488 * been done as mergeclauses. (In practice, it's rare that there's more
1489 * than two or three mergeclauses, so expending a huge amount of thought
1490 * on that is probably not worth it.)
1491 *
1492 * The pathkey order returned by select_outer_pathkeys_for_merge() has
1493 * some heuristics behind it (see that function), so be sure to try it
1494 * exactly as-is as well as making variants.
1495 */
1496 all_pathkeys = select_outer_pathkeys_for_merge(root,
1497 extra->mergeclause_list,
1498 joinrel);
1499
1500 foreach(l, all_pathkeys)
1501 {
1502 PathKey *front_pathkey = (PathKey *) lfirst(l);
1503 List *cur_mergeclauses;
1504 List *outerkeys;
1505 List *innerkeys;
1506 List *merge_pathkeys;
1507
1508 /* Make a pathkey list with this guy first */
1509 if (l != list_head(all_pathkeys))
1510 outerkeys = lcons(front_pathkey,
1511 list_delete_nth_cell(list_copy(all_pathkeys),
1512 foreach_current_index(l)));
1513 else
1514 outerkeys = all_pathkeys; /* no work at first one... */
1515
1516 /* Sort the mergeclauses into the corresponding ordering */
1517 cur_mergeclauses =
1518 find_mergeclauses_for_outer_pathkeys(root,
1519 outerkeys,
1520 extra->mergeclause_list);
1521
1522 /* Should have used them all... */
1523 Assert(list_length(cur_mergeclauses) == list_length(extra->mergeclause_list));
1524
1525 /* Build sort pathkeys for the inner side */
1526 innerkeys = make_inner_pathkeys_for_merge(root,
1527 cur_mergeclauses,
1528 outerkeys);
1529
1530 /* Build pathkeys representing output sort order */
1531 merge_pathkeys = build_join_pathkeys(root, joinrel, jointype,
1532 outerkeys);
1533
1534 /*
1535 * And now we can make the path.
1536 *
1537 * Note: it's possible that the cheapest paths will already be sorted
1538 * properly. try_mergejoin_path will detect that case and suppress an
1539 * explicit sort step, so we needn't do so here.
1540 */
1541 try_mergejoin_path(root,
1542 joinrel,
1543 outer_path,
1544 inner_path,
1545 merge_pathkeys,
1546 cur_mergeclauses,
1547 outerkeys,
1548 innerkeys,
1549 jointype,
1550 extra,
1551 false);
1552
1553 /*
1554 * If we have partial outer and parallel safe inner path then try
1555 * partial mergejoin path.
1556 */
1557 if (cheapest_partial_outer && cheapest_safe_inner)
1558 try_partial_mergejoin_path(root,
1559 joinrel,
1560 cheapest_partial_outer,
1561 cheapest_safe_inner,
1562 merge_pathkeys,
1563 cur_mergeclauses,
1564 outerkeys,
1565 innerkeys,
1566 jointype,
1567 extra);
1568 }
1569}
1570
1571/*
1572 * generate_mergejoin_paths
1573 * Creates possible mergejoin paths for input outerpath.
1574 *
1575 * We generate mergejoins if mergejoin clauses are available. We have
1576 * two ways to generate the inner path for a mergejoin: sort the cheapest
1577 * inner path, or use an inner path that is already suitably ordered for the
1578 * merge. If we have several mergeclauses, it could be that there is no inner
1579 * path (or only a very expensive one) for the full list of mergeclauses, but
1580 * better paths exist if we truncate the mergeclause list (thereby discarding
1581 * some sort key requirements). So, we consider truncations of the
1582 * mergeclause list as well as the full list. (Ideally we'd consider all
1583 * subsets of the mergeclause list, but that seems way too expensive.)
1584 */
1585static void
1586generate_mergejoin_paths(PlannerInfo *root,
1587 RelOptInfo *joinrel,
1588 RelOptInfo *innerrel,
1589 Path *outerpath,
1590 JoinType jointype,
1591 JoinPathExtraData *extra,
1592 bool useallclauses,
1593 Path *inner_cheapest_total,
1594 List *merge_pathkeys,
1595 bool is_partial)
1596{
1597 List *mergeclauses;
1598 List *innersortkeys;
1599 List *trialsortkeys;
1600 Path *cheapest_startup_inner;
1601 Path *cheapest_total_inner;
1602 int num_sortkeys;
1603 int sortkeycnt;
1604
1605 /* Look for useful mergeclauses (if any) */
1606 mergeclauses =
1607 find_mergeclauses_for_outer_pathkeys(root,
1608 outerpath->pathkeys,
1609 extra->mergeclause_list);
1610
1611 /*
1612 * Done with this outer path if no chance for a mergejoin.
1613 *
1614 * Special corner case: for "x FULL JOIN y ON true", there will be no join
1615 * clauses at all. Ordinarily we'd generate a clauseless nestloop path,
1616 * but since mergejoin is our only join type that supports FULL JOIN
1617 * without any join clauses, it's necessary to generate a clauseless
1618 * mergejoin path instead.
1619 */
1620 if (mergeclauses == NIL)
1621 {
1622 if (jointype == JOIN_FULL)
1623 /* okay to try for mergejoin */ ;
1624 else
1625 return;
1626 }
1627 if (useallclauses &&
1628 list_length(mergeclauses) != list_length(extra->mergeclause_list))
1629 return;
1630
1631 /* Compute the required ordering of the inner path */
1632 innersortkeys = make_inner_pathkeys_for_merge(root,
1633 mergeclauses,
1634 outerpath->pathkeys);
1635
1636 /*
1637 * Generate a mergejoin on the basis of sorting the cheapest inner. Since
1638 * a sort will be needed, only cheapest total cost matters. (But
1639 * try_mergejoin_path will do the right thing if inner_cheapest_total is
1640 * already correctly sorted.)
1641 */
1642 try_mergejoin_path(root,
1643 joinrel,
1644 outerpath,
1645 inner_cheapest_total,
1646 merge_pathkeys,
1647 mergeclauses,
1648 NIL,
1649 innersortkeys,
1650 jointype,
1651 extra,
1652 is_partial);
1653
1654 /*
1655 * Look for presorted inner paths that satisfy the innersortkey list ---
1656 * or any truncation thereof, if we are allowed to build a mergejoin using
1657 * a subset of the merge clauses. Here, we consider both cheap startup
1658 * cost and cheap total cost.
1659 *
1660 * Currently we do not consider parameterized inner paths here. This
1661 * interacts with decisions elsewhere that also discriminate against
1662 * mergejoins with parameterized inputs; see comments in
1663 * src/backend/optimizer/README.
1664 *
1665 * As we shorten the sortkey list, we should consider only paths that are
1666 * strictly cheaper than (in particular, not the same as) any path found
1667 * in an earlier iteration. Otherwise we'd be intentionally using fewer
1668 * merge keys than a given path allows (treating the rest as plain
1669 * joinquals), which is unlikely to be a good idea. Also, eliminating
1670 * paths here on the basis of compare_path_costs is a lot cheaper than
1671 * building the mergejoin path only to throw it away.
1672 *
1673 * If inner_cheapest_total is well enough sorted to have not required a
1674 * sort in the path made above, we shouldn't make a duplicate path with
1675 * it, either. We handle that case with the same logic that handles the
1676 * previous consideration, by initializing the variables that track
1677 * cheapest-so-far properly. Note that we do NOT reject
1678 * inner_cheapest_total if we find it matches some shorter set of
1679 * pathkeys. That case corresponds to using fewer mergekeys to avoid
1680 * sorting inner_cheapest_total, whereas we did sort it above, so the
1681 * plans being considered are different.
1682 */
1683 if (pathkeys_contained_in(innersortkeys,
1684 inner_cheapest_total->pathkeys))
1685 {
1686 /* inner_cheapest_total didn't require a sort */
1687 cheapest_startup_inner = inner_cheapest_total;
1688 cheapest_total_inner = inner_cheapest_total;
1689 }
1690 else
1691 {
1692 /* it did require a sort, at least for the full set of keys */
1693 cheapest_startup_inner = NULL;
1694 cheapest_total_inner = NULL;
1695 }
1696 num_sortkeys = list_length(innersortkeys);
1697 if (num_sortkeys > 1 && !useallclauses)
1698 trialsortkeys = list_copy(innersortkeys); /* need modifiable copy */
1699 else
1700 trialsortkeys = innersortkeys; /* won't really truncate */
1701
1702 for (sortkeycnt = num_sortkeys; sortkeycnt > 0; sortkeycnt--)
1703 {
1704 Path *innerpath;
1705 List *newclauses = NIL;
1706
1707 /*
1708 * Look for an inner path ordered well enough for the first
1709 * 'sortkeycnt' innersortkeys. NB: trialsortkeys list is modified
1710 * destructively, which is why we made a copy...
1711 */
1712 trialsortkeys = list_truncate(trialsortkeys, sortkeycnt);
1713 innerpath = get_cheapest_path_for_pathkeys(innerrel->pathlist,
1714 trialsortkeys,
1715 NULL,
1716 TOTAL_COST,
1717 is_partial);
1718 if (innerpath != NULL &&
1719 (cheapest_total_inner == NULL ||
1720 compare_path_costs(innerpath, cheapest_total_inner,
1721 TOTAL_COST) < 0))
1722 {
1723 /* Found a cheap (or even-cheaper) sorted path */
1724 /* Select the right mergeclauses, if we didn't already */
1725 if (sortkeycnt < num_sortkeys)
1726 {
1727 newclauses =
1728 trim_mergeclauses_for_inner_pathkeys(root,
1729 mergeclauses,
1730 trialsortkeys);
1731 Assert(newclauses != NIL);
1732 }
1733 else
1734 newclauses = mergeclauses;
1735 try_mergejoin_path(root,
1736 joinrel,
1737 outerpath,
1738 innerpath,
1739 merge_pathkeys,
1740 newclauses,
1741 NIL,
1742 NIL,
1743 jointype,
1744 extra,
1745 is_partial);
1746 cheapest_total_inner = innerpath;
1747 }
1748 /* Same on the basis of cheapest startup cost ... */
1749 innerpath = get_cheapest_path_for_pathkeys(innerrel->pathlist,
1750 trialsortkeys,
1751 NULL,
1752 STARTUP_COST,
1753 is_partial);
1754 if (innerpath != NULL &&
1755 (cheapest_startup_inner == NULL ||
1756 compare_path_costs(innerpath, cheapest_startup_inner,
1757 STARTUP_COST) < 0))
1758 {
1759 /* Found a cheap (or even-cheaper) sorted path */
1760 if (innerpath != cheapest_total_inner)
1761 {
1762 /*
1763 * Avoid rebuilding clause list if we already made one; saves
1764 * memory in big join trees...
1765 */
1766 if (newclauses == NIL)
1767 {
1768 if (sortkeycnt < num_sortkeys)
1769 {
1770 newclauses =
1771 trim_mergeclauses_for_inner_pathkeys(root,
1772 mergeclauses,
1773 trialsortkeys);
1774 Assert(newclauses != NIL);
1775 }
1776 else
1777 newclauses = mergeclauses;
1778 }
1779 try_mergejoin_path(root,
1780 joinrel,
1781 outerpath,
1782 innerpath,
1783 merge_pathkeys,
1784 newclauses,
1785 NIL,
1786 NIL,
1787 jointype,
1788 extra,
1789 is_partial);
1790 }
1791 cheapest_startup_inner = innerpath;
1792 }
1793
1794 /*
1795 * Don't consider truncated sortkeys if we need all clauses.
1796 */
1797 if (useallclauses)
1798 break;
1799 }
1800}
1801
1802/*
1803 * match_unsorted_outer
1804 * Creates possible join paths for processing a single join relation
1805 * 'joinrel' by employing either iterative substitution or
1806 * mergejoining on each of its possible outer paths (considering
1807 * only outer paths that are already ordered well enough for merging).
1808 *
1809 * We always generate a nestloop path for each available outer path.
1810 * In fact we may generate as many as five: one on the cheapest-total-cost
1811 * inner path, one on the same with materialization, one on the
1812 * cheapest-startup-cost inner path (if different), one on the
1813 * cheapest-total inner-indexscan path (if any), and one on the
1814 * cheapest-startup inner-indexscan path (if different).
1815 *
1816 * We also consider mergejoins if mergejoin clauses are available. See
1817 * detailed comments in generate_mergejoin_paths.
1818 *
1819 * 'joinrel' is the join relation
1820 * 'outerrel' is the outer join relation
1821 * 'innerrel' is the inner join relation
1822 * 'jointype' is the type of join to do
1823 * 'extra' contains additional input values
1824 */
1825static void
1826match_unsorted_outer(PlannerInfo *root,
1827 RelOptInfo *joinrel,
1828 RelOptInfo *outerrel,
1829 RelOptInfo *innerrel,
1830 JoinType jointype,
1831 JoinPathExtraData *extra)
1832{
1833 bool nestjoinOK;
1834 bool useallclauses;
1835 Path *inner_cheapest_total = innerrel->cheapest_total_path;
1836 Path *matpath = NULL;
1837 ListCell *lc1;
1838
1839 /*
1840 * For now we do not support RIGHT_SEMI join in mergejoin or nestloop
1841 * join.
1842 */
1843 if (jointype == JOIN_RIGHT_SEMI)
1844 return;
1845
1846 /*
1847 * Nestloop only supports inner, left, semi, and anti joins. Also, if we
1848 * are doing a right, right-anti or full mergejoin, we must use *all* the
1849 * mergeclauses as join clauses, else we will not have a valid plan.
1850 * (Although these two flags are currently inverses, keep them separate
1851 * for clarity and possible future changes.)
1852 */
1853 switch (jointype)
1854 {
1855 case JOIN_INNER:
1856 case JOIN_LEFT:
1857 case JOIN_SEMI:
1858 case JOIN_ANTI:
1859 nestjoinOK = true;
1860 useallclauses = false;
1861 break;
1862 case JOIN_RIGHT:
1863 case JOIN_RIGHT_ANTI:
1864 case JOIN_FULL:
1865 nestjoinOK = false;
1866 useallclauses = true;
1867 break;
1868 default:
1869 elog(ERROR, "unrecognized join type: %d",
1870 (int) jointype);
1871 nestjoinOK = false; /* keep compiler quiet */
1872 useallclauses = false;
1873 break;
1874 }
1875
1876 /*
1877 * If inner_cheapest_total is parameterized by the outer rel, ignore it;
1878 * we will consider it below as a member of cheapest_parameterized_paths,
1879 * but the other possibilities considered in this routine aren't usable.
1880 *
1881 * Furthermore, if the inner side is a unique-ified relation, we cannot
1882 * generate any valid paths here, because the inner rel's dependency on
1883 * the outer rel makes unique-ification meaningless.
1884 */
1885 if (PATH_PARAM_BY_REL(inner_cheapest_total, outerrel))
1886 {
1887 inner_cheapest_total = NULL;
1888
1889 if (RELATION_WAS_MADE_UNIQUE(innerrel, extra->sjinfo, jointype))
1890 return;
1891 }
1892
1893 if (nestjoinOK)
1894 {
1895 /*
1896 * Consider materializing the cheapest inner path, unless that is
1897 * disabled or the path in question materializes its output anyway.
1898 *
1899 * At present, we only consider materialization for non-partial outer
1900 * paths, so it's correct to test PGS_CONSIDER_NONPARTIAL here. If we
1901 * ever want to consider materialization for partial paths, we'll need
1902 * to create matpath whenever PGS_NESTLOOP_MATERIALIZE is set, use it
1903 * for partial paths either way, and use it for non-partial paths only
1904 * when PGS_CONSIDER_NONPARTIAL is also set.
1905 */
1906 if ((extra->pgs_mask &
1907 (PGS_NESTLOOP_MATERIALIZE | PGS_CONSIDER_NONPARTIAL)) ==
1908 (PGS_NESTLOOP_MATERIALIZE | PGS_CONSIDER_NONPARTIAL) &&
1909 inner_cheapest_total != NULL &&
1910 !ExecMaterializesOutput(inner_cheapest_total->pathtype))
1911 matpath = (Path *)
1912 create_material_path(innerrel, inner_cheapest_total, true);
1913 }
1914
1915 foreach(lc1, outerrel->pathlist)
1916 {
1917 Path *outerpath = (Path *) lfirst(lc1);
1918 List *merge_pathkeys;
1919
1920 /*
1921 * We cannot use an outer path that is parameterized by the inner rel.
1922 */
1923 if (PATH_PARAM_BY_REL(outerpath, innerrel))
1924 continue;
1925
1926 /*
1927 * The result will have this sort order (even if it is implemented as
1928 * a nestloop, and even if some of the mergeclauses are implemented by
1929 * qpquals rather than as true mergeclauses):
1930 */
1931 merge_pathkeys = build_join_pathkeys(root, joinrel, jointype,
1932 outerpath->pathkeys);
1933
1934 if (nestjoinOK)
1935 {
1936 /*
1937 * Consider nestloop joins using this outer path and various
1938 * available paths for the inner relation. We consider the
1939 * cheapest-total paths for each available parameterization of the
1940 * inner relation, including the unparameterized case.
1941 */
1942 ListCell *lc2;
1943
1944 foreach(lc2, innerrel->cheapest_parameterized_paths)
1945 {
1946 Path *innerpath = (Path *) lfirst(lc2);
1947 Path *mpath;
1948
1949 try_nestloop_path(root,
1950 joinrel,
1951 outerpath,
1952 innerpath,
1953 merge_pathkeys,
1954 jointype,
1955 PGS_NESTLOOP_PLAIN,
1956 extra);
1957
1958 /*
1959 * Try generating a memoize path and see if that makes the
1960 * nested loop any cheaper.
1961 */
1962 mpath = get_memoize_path(root, innerrel, outerrel,
1963 innerpath, outerpath, jointype,
1964 extra);
1965 if (mpath != NULL)
1966 try_nestloop_path(root,
1967 joinrel,
1968 outerpath,
1969 mpath,
1970 merge_pathkeys,
1971 jointype,
1972 PGS_NESTLOOP_MEMOIZE,
1973 extra);
1974 }
1975
1976 /* Also consider materialized form of the cheapest inner path */
1977 if (matpath != NULL)
1978 try_nestloop_path(root,
1979 joinrel,
1980 outerpath,
1981 matpath,
1982 merge_pathkeys,
1983 jointype,
1984 PGS_NESTLOOP_MATERIALIZE,
1985 extra);
1986 }
1987
1988 /* Can't do anything else if inner rel is parameterized by outer */
1989 if (inner_cheapest_total == NULL)
1990 continue;
1991
1992 /* Generate merge join paths */
1993 generate_mergejoin_paths(root, joinrel, innerrel, outerpath,
1994 jointype, extra, useallclauses,
1995 inner_cheapest_total, merge_pathkeys,
1996 false);
1997 }
1998
1999 /*
2000 * Consider partial nestloop and mergejoin plan if outerrel has any
2001 * partial path and the joinrel is parallel-safe. However, we can't
2002 * handle joins needing lateral rels, since partial paths must not be
2003 * parameterized. Similarly, we can't handle JOIN_FULL, JOIN_RIGHT and
2004 * JOIN_RIGHT_ANTI, because they can produce false null extended rows.
2005 */
2006 if (joinrel->consider_parallel &&
2007 jointype != JOIN_FULL &&
2008 jointype != JOIN_RIGHT &&
2009 jointype != JOIN_RIGHT_ANTI &&
2010 outerrel->partial_pathlist != NIL &&
2011 bms_is_empty(joinrel->lateral_relids))
2012 {
2013 if (nestjoinOK)
2014 consider_parallel_nestloop(root, joinrel, outerrel, innerrel,
2015 jointype, extra);
2016
2017 /*
2018 * If inner_cheapest_total is NULL or non parallel-safe then find the
2019 * cheapest total parallel safe path.
2020 */
2021 if (inner_cheapest_total == NULL ||
2022 !inner_cheapest_total->parallel_safe)
2023 {
2024 inner_cheapest_total =
2025 get_cheapest_parallel_safe_total_inner(innerrel->pathlist);
2026 }
2027
2028 if (inner_cheapest_total)
2029 consider_parallel_mergejoin(root, joinrel, outerrel, innerrel,
2030 jointype, extra,
2031 inner_cheapest_total);
2032 }
2033}
2034
2035/*
2036 * consider_parallel_mergejoin
2037 * Try to build partial paths for a joinrel by joining a partial path
2038 * for the outer relation to a complete path for the inner relation.
2039 *
2040 * 'joinrel' is the join relation
2041 * 'outerrel' is the outer join relation
2042 * 'innerrel' is the inner join relation
2043 * 'jointype' is the type of join to do
2044 * 'extra' contains additional input values
2045 * 'inner_cheapest_total' cheapest total path for innerrel
2046 */
2047static void
2048consider_parallel_mergejoin(PlannerInfo *root,
2049 RelOptInfo *joinrel,
2050 RelOptInfo *outerrel,
2051 RelOptInfo *innerrel,
2052 JoinType jointype,
2053 JoinPathExtraData *extra,
2054 Path *inner_cheapest_total)
2055{
2056 ListCell *lc1;
2057
2058 /* generate merge join path for each partial outer path */
2059 foreach(lc1, outerrel->partial_pathlist)
2060 {
2061 Path *outerpath = (Path *) lfirst(lc1);
2062 List *merge_pathkeys;
2063
2064 /*
2065 * Figure out what useful ordering any paths we create will have.
2066 */
2067 merge_pathkeys = build_join_pathkeys(root, joinrel, jointype,
2068 outerpath->pathkeys);
2069
2070 generate_mergejoin_paths(root, joinrel, innerrel, outerpath, jointype,
2071 extra, false, inner_cheapest_total,
2072 merge_pathkeys, true);
2073 }
2074}
2075
2076/*
2077 * consider_parallel_nestloop
2078 * Try to build partial paths for a joinrel by joining a partial path for the
2079 * outer relation to a complete path for the inner relation.
2080 *
2081 * 'joinrel' is the join relation
2082 * 'outerrel' is the outer join relation
2083 * 'innerrel' is the inner join relation
2084 * 'jointype' is the type of join to do
2085 * 'extra' contains additional input values
2086 */
2087static void
2088consider_parallel_nestloop(PlannerInfo *root,
2089 RelOptInfo *joinrel,
2090 RelOptInfo *outerrel,
2091 RelOptInfo *innerrel,
2092 JoinType jointype,
2093 JoinPathExtraData *extra)
2094{
2095 Path *inner_cheapest_total = innerrel->cheapest_total_path;
2096 Path *matpath = NULL;
2097 ListCell *lc1;
2098
2099 /*
2100 * Consider materializing the cheapest inner path, unless: 1)
2101 * materialization is disabled here, 2) the cheapest inner path is not
2102 * parallel-safe, 3) the cheapest inner path is parameterized by the outer
2103 * rel, or 4) the cheapest inner path materializes its output anyway.
2104 */
2105 if ((extra->pgs_mask & PGS_NESTLOOP_MATERIALIZE) != 0 &&
2106 inner_cheapest_total->parallel_safe &&
2107 !PATH_PARAM_BY_REL(inner_cheapest_total, outerrel) &&
2108 !ExecMaterializesOutput(inner_cheapest_total->pathtype))
2109 {
2110 matpath = (Path *)
2111 create_material_path(innerrel, inner_cheapest_total, true);
2112 Assert(matpath->parallel_safe);
2113 }
2114
2115 foreach(lc1, outerrel->partial_pathlist)
2116 {
2117 Path *outerpath = (Path *) lfirst(lc1);
2118 List *pathkeys;
2119 ListCell *lc2;
2120
2121 /* Figure out what useful ordering any paths we create will have. */
2122 pathkeys = build_join_pathkeys(root, joinrel, jointype,
2123 outerpath->pathkeys);
2124
2125 /*
2126 * Try the cheapest parameterized paths; only those which will produce
2127 * an unparameterized path when joined to this outerrel will survive
2128 * try_partial_nestloop_path. The cheapest unparameterized path is
2129 * also in this list.
2130 */
2131 foreach(lc2, innerrel->cheapest_parameterized_paths)
2132 {
2133 Path *innerpath = (Path *) lfirst(lc2);
2134 Path *mpath;
2135
2136 /* Can't join to an inner path that is not parallel-safe */
2137 if (!innerpath->parallel_safe)
2138 continue;
2139
2140 try_partial_nestloop_path(root, joinrel, outerpath, innerpath,
2141 pathkeys, jointype,
2142 PGS_NESTLOOP_PLAIN, extra);
2143
2144 /*
2145 * Try generating a memoize path and see if that makes the nested
2146 * loop any cheaper.
2147 */
2148 mpath = get_memoize_path(root, innerrel, outerrel,
2149 innerpath, outerpath, jointype,
2150 extra);
2151 if (mpath != NULL)
2152 try_partial_nestloop_path(root, joinrel, outerpath, mpath,
2153 pathkeys, jointype,
2154 PGS_NESTLOOP_MEMOIZE, extra);
2155 }
2156
2157 /* Also consider materialized form of the cheapest inner path */
2158 if (matpath != NULL)
2159 try_partial_nestloop_path(root, joinrel, outerpath, matpath,
2160 pathkeys, jointype,
2161 PGS_NESTLOOP_MATERIALIZE, extra);
2162 }
2163}
2164
2165/*
2166 * hash_inner_and_outer
2167 * Create hashjoin join paths by explicitly hashing both the outer and
2168 * inner keys of each available hash clause.
2169 *
2170 * 'joinrel' is the join relation
2171 * 'outerrel' is the outer join relation
2172 * 'innerrel' is the inner join relation
2173 * 'jointype' is the type of join to do
2174 * 'extra' contains additional input values
2175 */
2176static void
2177hash_inner_and_outer(PlannerInfo *root,
2178 RelOptInfo *joinrel,
2179 RelOptInfo *outerrel,
2180 RelOptInfo *innerrel,
2181 JoinType jointype,
2182 JoinPathExtraData *extra)
2183{
2184 bool isouterjoin = IS_OUTER_JOIN(jointype);
2185 List *hashclauses;
2186 ListCell *l;
2187
2188 /*
2189 * We need to build only one hashclauses list for any given pair of outer
2190 * and inner relations; all of the hashable clauses will be used as keys.
2191 *
2192 * Scan the join's restrictinfo list to find hashjoinable clauses that are
2193 * usable with this pair of sub-relations.
2194 */
2195 hashclauses = NIL;
2196 foreach(l, extra->restrictlist)
2197 {
2198 RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(l);
2199
2200 /*
2201 * If processing an outer join, only use its own join clauses for
2202 * hashing. For inner joins we need not be so picky.
2203 */
2204 if (isouterjoin && RINFO_IS_PUSHED_DOWN(restrictinfo, joinrel->relids))
2205 continue;
2206
2207 if (!restrictinfo->can_join ||
2208 restrictinfo->hashjoinoperator == InvalidOid)
2209 continue; /* not hashjoinable */
2210
2211 /*
2212 * Check if clause has the form "outer op inner" or "inner op outer".
2213 */
2214 if (!clause_sides_match_join(restrictinfo, outerrel->relids,
2215 innerrel->relids))
2216 continue; /* no good for these input relations */
2217
2218 /*
2219 * If clause has the form "inner op outer", check if its operator has
2220 * valid commutator. This is necessary because hashclauses in this
2221 * form will get commuted in createplan.c to put the outer var on the
2222 * left (see get_switched_clauses). This probably shouldn't ever
2223 * fail, since hashable operators ought to have commutators, but be
2224 * paranoid.
2225 *
2226 * The clause being hashjoinable indicates that it's an OpExpr.
2227 */
2228 if (!restrictinfo->outer_is_left &&
2229 !OidIsValid(get_commutator(castNode(OpExpr, restrictinfo->clause)->opno)))
2230 continue;
2231
2232 hashclauses = lappend(hashclauses, restrictinfo);
2233 }
2234
2235 /* If we found any usable hashclauses, make paths */
2236 if (hashclauses)
2237 {
2238 /*
2239 * We consider both the cheapest-total-cost and cheapest-startup-cost
2240 * outer paths. There's no need to consider any but the
2241 * cheapest-total-cost inner path, however.
2242 */
2243 Path *cheapest_startup_outer = outerrel->cheapest_startup_path;
2244 Path *cheapest_total_outer = outerrel->cheapest_total_path;
2245 Path *cheapest_total_inner = innerrel->cheapest_total_path;
2246 ListCell *lc1;
2247 ListCell *lc2;
2248
2249 /*
2250 * If either cheapest-total path is parameterized by the other rel, we
2251 * can't use a hashjoin. (There's no use looking for alternative
2252 * input paths, since these should already be the least-parameterized
2253 * available paths.)
2254 */
2255 if (PATH_PARAM_BY_REL(cheapest_total_outer, innerrel) ||
2256 PATH_PARAM_BY_REL(cheapest_total_inner, outerrel))
2257 return;
2258
2259 /*
2260 * Consider the cheapest startup outer together with the cheapest
2261 * total inner, and then consider pairings of cheapest-total paths
2262 * including parameterized ones. There is no use in generating
2263 * parameterized paths on the basis of possibly cheap startup cost, so
2264 * this is sufficient.
2265 */
2266 if (cheapest_startup_outer != NULL)
2267 try_hashjoin_path(root,
2268 joinrel,
2269 cheapest_startup_outer,
2270 cheapest_total_inner,
2271 hashclauses,
2272 jointype,
2273 extra);
2274
2275 foreach(lc1, outerrel->cheapest_parameterized_paths)
2276 {
2277 Path *outerpath = (Path *) lfirst(lc1);
2278
2279 /*
2280 * We cannot use an outer path that is parameterized by the inner
2281 * rel.
2282 */
2283 if (PATH_PARAM_BY_REL(outerpath, innerrel))
2284 continue;
2285
2286 foreach(lc2, innerrel->cheapest_parameterized_paths)
2287 {
2288 Path *innerpath = (Path *) lfirst(lc2);
2289
2290 /*
2291 * We cannot use an inner path that is parameterized by the
2292 * outer rel, either.
2293 */
2294 if (PATH_PARAM_BY_REL(innerpath, outerrel))
2295 continue;
2296
2297 if (outerpath == cheapest_startup_outer &&
2298 innerpath == cheapest_total_inner)
2299 continue; /* already tried it */
2300
2301 try_hashjoin_path(root,
2302 joinrel,
2303 outerpath,
2304 innerpath,
2305 hashclauses,
2306 jointype,
2307 extra);
2308 }
2309 }
2310
2311 /*
2312 * If the joinrel is parallel-safe, we may be able to consider a
2313 * partial hash join.
2314 *
2315 * However, we can't handle JOIN_RIGHT_SEMI, because the hash table is
2316 * either a shared hash table or a private hash table per backend. In
2317 * the shared case, there is no concurrency protection for the match
2318 * flags, so multiple workers could inspect and set the flags
2319 * concurrently, potentially producing incorrect results. In the
2320 * private case, each worker has its own copy of the hash table, so no
2321 * single process has all the match flags.
2322 *
2323 * Also, the resulting path must not be parameterized.
2324 */
2325 if (joinrel->consider_parallel &&
2326 jointype != JOIN_RIGHT_SEMI &&
2327 outerrel->partial_pathlist != NIL &&
2328 bms_is_empty(joinrel->lateral_relids))
2329 {
2330 Path *cheapest_partial_outer;
2331 Path *cheapest_partial_inner = NULL;
2332 Path *cheapest_safe_inner = NULL;
2333
2334 cheapest_partial_outer =
2335 (Path *) linitial(outerrel->partial_pathlist);
2336
2337 /*
2338 * Can we use a partial inner plan too, so that we can build a
2339 * shared hash table in parallel?
2340 */
2341 if (innerrel->partial_pathlist != NIL &&
2342 enable_parallel_hash)
2343 {
2344 cheapest_partial_inner =
2345 (Path *) linitial(innerrel->partial_pathlist);
2346 try_partial_hashjoin_path(root, joinrel,
2347 cheapest_partial_outer,
2348 cheapest_partial_inner,
2349 hashclauses, jointype, extra,
2350 true /* parallel_hash */ );
2351 }
2352
2353 /*
2354 * Normally, given that the joinrel is parallel-safe, the cheapest
2355 * total inner path will also be parallel-safe, but if not, we'll
2356 * have to search for the cheapest safe, unparameterized inner
2357 * path. If full, right, or right-anti join, we can't use
2358 * parallelism (building the hash table in each backend) because
2359 * no one process has all the match bits.
2360 */
2361 if (jointype == JOIN_FULL ||
2362 jointype == JOIN_RIGHT ||
2363 jointype == JOIN_RIGHT_ANTI)
2364 cheapest_safe_inner = NULL;
2365 else if (cheapest_total_inner->parallel_safe)
2366 cheapest_safe_inner = cheapest_total_inner;
2367 else
2368 cheapest_safe_inner =
2369 get_cheapest_parallel_safe_total_inner(innerrel->pathlist);
2370
2371 if (cheapest_safe_inner != NULL)
2372 try_partial_hashjoin_path(root, joinrel,
2373 cheapest_partial_outer,
2374 cheapest_safe_inner,
2375 hashclauses, jointype, extra,
2376 false /* parallel_hash */ );
2377 }
2378 }
2379}
2380
2381/*
2382 * select_mergejoin_clauses
2383 * Select mergejoin clauses that are usable for a particular join.
2384 * Returns a list of RestrictInfo nodes for those clauses.
2385 *
2386 * *mergejoin_allowed is normally set to true, but it is set to false if
2387 * this is a right-semi join, or this is a right/right-anti/full join and
2388 * there are nonmergejoinable join clauses. The executor's mergejoin
2389 * machinery cannot handle such cases, so we have to avoid generating a
2390 * mergejoin plan. (Note that this flag does NOT consider whether there are
2391 * actually any mergejoinable clauses. This is correct because in some
2392 * cases we need to build a clauseless mergejoin. Simply returning NIL is
2393 * therefore not enough to distinguish safe from unsafe cases.)
2394 *
2395 * We also mark each selected RestrictInfo to show which side is currently
2396 * being considered as outer. These are transient markings that are only
2397 * good for the duration of the current add_paths_to_joinrel() call!
2398 *
2399 * We examine each restrictinfo clause known for the join to see
2400 * if it is mergejoinable and involves vars from the two sub-relations
2401 * currently of interest.
2402 */
2403static List *
2404select_mergejoin_clauses(PlannerInfo *root,
2405 RelOptInfo *joinrel,
2406 RelOptInfo *outerrel,
2407 RelOptInfo *innerrel,
2408 List *restrictlist,
2409 JoinType jointype,
2410 bool *mergejoin_allowed)
2411{
2412 List *result_list = NIL;
2413 bool isouterjoin = IS_OUTER_JOIN(jointype);
2414 bool have_nonmergeable_joinclause = false;
2415 ListCell *l;
2416
2417 /*
2418 * For now we do not support RIGHT_SEMI join in mergejoin: the benefit of
2419 * swapping inputs tends to be small here.
2420 */
2421 if (jointype == JOIN_RIGHT_SEMI)
2422 {
2423 *mergejoin_allowed = false;
2424 return NIL;
2425 }
2426
2427 foreach(l, restrictlist)
2428 {
2429 RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(l);
2430
2431 /*
2432 * If processing an outer join, only use its own join clauses in the
2433 * merge. For inner joins we can use pushed-down clauses too. (Note:
2434 * we don't set have_nonmergeable_joinclause here because pushed-down
2435 * clauses will become otherquals not joinquals.)
2436 */
2437 if (isouterjoin && RINFO_IS_PUSHED_DOWN(restrictinfo, joinrel->relids))
2438 continue;
2439
2440 /* Check that clause is a mergeable operator clause */
2441 if (!restrictinfo->can_join ||
2442 restrictinfo->mergeopfamilies == NIL)
2443 {
2444 /*
2445 * The executor can handle extra joinquals that are constants, but
2446 * not anything else, when doing right/right-anti/full merge join.
2447 * (The reason to support constants is so we can do FULL JOIN ON
2448 * FALSE.)
2449 */
2450 if (!restrictinfo->clause || !IsA(restrictinfo->clause, Const))
2451 have_nonmergeable_joinclause = true;
2452 continue; /* not mergejoinable */
2453 }
2454
2455 /*
2456 * Check if clause has the form "outer op inner" or "inner op outer".
2457 */
2458 if (!clause_sides_match_join(restrictinfo, outerrel->relids,
2459 innerrel->relids))
2460 {
2461 have_nonmergeable_joinclause = true;
2462 continue; /* no good for these input relations */
2463 }
2464
2465 /*
2466 * If clause has the form "inner op outer", check if its operator has
2467 * valid commutator. This is necessary because mergejoin clauses in
2468 * this form will get commuted in createplan.c to put the outer var on
2469 * the left (see get_switched_clauses). This probably shouldn't ever
2470 * fail, since mergejoinable operators ought to have commutators, but
2471 * be paranoid.
2472 *
2473 * The clause being mergejoinable indicates that it's an OpExpr.
2474 */
2475 if (!restrictinfo->outer_is_left &&
2476 !OidIsValid(get_commutator(castNode(OpExpr, restrictinfo->clause)->opno)))
2477 {
2478 have_nonmergeable_joinclause = true;
2479 continue;
2480 }
2481
2482 /*
2483 * Insist that each side have a non-redundant eclass. This
2484 * restriction is needed because various bits of the planner expect
2485 * that each clause in a merge be associable with some pathkey in a
2486 * canonical pathkey list, but redundant eclasses can't appear in
2487 * canonical sort orderings. (XXX it might be worth relaxing this,
2488 * but not enough time to address it for 8.3.)
2489 */
2490 update_mergeclause_eclasses(root, restrictinfo);
2491
2492 if (EC_MUST_BE_REDUNDANT(restrictinfo->left_ec) ||
2493 EC_MUST_BE_REDUNDANT(restrictinfo->right_ec))
2494 {
2495 have_nonmergeable_joinclause = true;
2496 continue; /* can't handle redundant eclasses */
2497 }
2498
2499 result_list = lappend(result_list, restrictinfo);
2500 }
2501
2502 /*
2503 * Report whether mergejoin is allowed (see comment at top of function).
2504 */
2505 switch (jointype)
2506 {
2507 case JOIN_RIGHT:
2508 case JOIN_RIGHT_ANTI:
2509 case JOIN_FULL:
2510 *mergejoin_allowed = !have_nonmergeable_joinclause;
2511 break;
2512 default:
2513 *mergejoin_allowed = true;
2514 break;
2515 }
2516
2517 return result_list;
2518}
innerrel_is_unique
bool innerrel_is_unique(PlannerInfo *root, Relids joinrelids, Relids outerrelids, RelOptInfo *innerrel, JoinType jointype, List *restrictlist, bool force_cache)
Definition analyzejoins.c:1310
bms_difference
Bitmapset * bms_difference(const Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:346
bms_intersect
Bitmapset * bms_intersect(const Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:292
bms_equal
bool bms_equal(const Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:142
bms_is_subset
bool bms_is_subset(const Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:412
bms_free
void bms_free(Bitmapset *a)
Definition bitmapset.c:239
bms_is_member
bool bms_is_member(int x, const Bitmapset *a)
Definition bitmapset.c:510
bms_add_members
Bitmapset * bms_add_members(Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:901
bms_membership
BMS_Membership bms_membership(const Bitmapset *a)
Definition bitmapset.c:765
bms_overlap
bool bms_overlap(const Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:575
bms_join
Bitmapset * bms_join(Bitmapset *a, Bitmapset *b)
Definition bitmapset.c:1214
bms_nonempty_difference
bool bms_nonempty_difference(const Bitmapset *a, const Bitmapset *b)
Definition bitmapset.c:634
bms_is_empty
#define bms_is_empty(a)
Definition bitmapset.h:118
BMS_MULTIPLE
@ BMS_MULTIPLE
Definition bitmapset.h:73
Assert
#define Assert(condition)
Definition c.h:885
uint64
uint64_t uint64
Definition c.h:559
OidIsValid
#define OidIsValid(objectId)
Definition c.h:800
contain_volatile_functions
bool contain_volatile_functions(Node *clause)
Definition clauses.c:547
compute_semi_anti_join_factors
void compute_semi_anti_join_factors(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, SpecialJoinInfo *sjinfo, List *restrictlist, SemiAntiJoinFactors *semifactors)
Definition costsize.c:5256
initial_cost_nestloop
void initial_cost_nestloop(PlannerInfo *root, JoinCostWorkspace *workspace, JoinType jointype, uint64 enable_mask, Path *outer_path, Path *inner_path, JoinPathExtraData *extra)
Definition costsize.c:3372
initial_cost_hashjoin
void initial_cost_hashjoin(PlannerInfo *root, JoinCostWorkspace *workspace, JoinType jointype, List *hashclauses, Path *outer_path, Path *inner_path, JoinPathExtraData *extra, bool parallel_hash)
Definition costsize.c:4296
initial_cost_mergejoin
void initial_cost_mergejoin(PlannerInfo *root, JoinCostWorkspace *workspace, JoinType jointype, List *mergeclauses, Path *outer_path, Path *inner_path, List *outersortkeys, List *innersortkeys, int outer_presorted_keys, JoinPathExtraData *extra)
Definition costsize.c:3657
enable_parallel_hash
bool enable_parallel_hash
Definition costsize.c:162
ERROR
#define ERROR
Definition elog.h:39
elog
#define elog(elevel,...)
Definition elog.h:226
ExecMaterializesOutput
bool ExecMaterializesOutput(NodeTag plantype)
Definition execAmi.c:635
try_hashjoin_path
static void try_hashjoin_path(PlannerInfo *root, RelOptInfo *joinrel, Path *outer_path, Path *inner_path, List *hashclauses, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:1267
select_mergejoin_clauses
static List * select_mergejoin_clauses(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, List *restrictlist, JoinType jointype, bool *mergejoin_allowed)
Definition joinpath.c:2404
sort_inner_and_outer
static void sort_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:1402
extract_lateral_vars_from_PHVs
static List * extract_lateral_vars_from_PHVs(PlannerInfo *root, Relids innerrelids)
Definition joinpath.c:621
add_paths_to_joinrel
void add_paths_to_joinrel(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, SpecialJoinInfo *sjinfo, List *restrictlist)
Definition joinpath.c:123
paraminfo_get_equal_hashops
static bool paraminfo_get_equal_hashops(PlannerInfo *root, ParamPathInfo *param_info, RelOptInfo *outerrel, RelOptInfo *innerrel, List *ph_lateral_vars, List **param_exprs, List **operators, bool *binary_mode)
Definition joinpath.c:476
get_memoize_path
static Path * get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel, RelOptInfo *outerrel, Path *inner_path, Path *outer_path, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:712
consider_parallel_nestloop
static void consider_parallel_nestloop(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:2088
join_path_setup_hook
join_path_setup_hook_type join_path_setup_hook
Definition joinpath.c:32
PATH_PARAM_BY_PARENT
#define PATH_PARAM_BY_PARENT(path, rel)
Definition joinpath.c:39
set_join_pathlist_hook
set_join_pathlist_hook_type set_join_pathlist_hook
Definition joinpath.c:31
try_nestloop_path
static void try_nestloop_path(PlannerInfo *root, RelOptInfo *joinrel, Path *outer_path, Path *inner_path, List *pathkeys, JoinType jointype, uint64 nestloop_subtype, JoinPathExtraData *extra)
Definition joinpath.c:873
try_mergejoin_path
static void try_mergejoin_path(PlannerInfo *root, RelOptInfo *joinrel, Path *outer_path, Path *inner_path, List *pathkeys, List *mergeclauses, List *outersortkeys, List *innersortkeys, JoinType jointype, JoinPathExtraData *extra, bool is_partial)
Definition joinpath.c:1074
consider_parallel_mergejoin
static void consider_parallel_mergejoin(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra, Path *inner_cheapest_total)
Definition joinpath.c:2048
generate_mergejoin_paths
static void generate_mergejoin_paths(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *innerrel, Path *outerpath, JoinType jointype, JoinPathExtraData *extra, bool useallclauses, Path *inner_cheapest_total, List *merge_pathkeys, bool is_partial)
Definition joinpath.c:1586
try_partial_nestloop_path
static void try_partial_nestloop_path(PlannerInfo *root, RelOptInfo *joinrel, Path *outer_path, Path *inner_path, List *pathkeys, JoinType jointype, uint64 nestloop_subtype, JoinPathExtraData *extra)
Definition joinpath.c:994
hash_inner_and_outer
static void hash_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:2177
try_partial_mergejoin_path
static void try_partial_mergejoin_path(PlannerInfo *root, RelOptInfo *joinrel, Path *outer_path, Path *inner_path, List *pathkeys, List *mergeclauses, List *outersortkeys, List *innersortkeys, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:1190
PATH_PARAM_BY_REL
#define PATH_PARAM_BY_REL(path, rel)
Definition joinpath.c:45
allow_star_schema_join
static bool allow_star_schema_join(PlannerInfo *root, Relids outerrelids, Relids inner_paramrels)
Definition joinpath.c:400
try_partial_hashjoin_path
static void try_partial_hashjoin_path(PlannerInfo *root, RelOptInfo *joinrel, Path *outer_path, Path *inner_path, List *hashclauses, JoinType jointype, JoinPathExtraData *extra, bool parallel_hash)
Definition joinpath.c:1344
match_unsorted_outer
static void match_unsorted_outer(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra)
Definition joinpath.c:1826
lappend
List * lappend(List *list, void *datum)
Definition list.c:339
list_delete_nth_cell
List * list_delete_nth_cell(List *list, int n)
Definition list.c:767
list_concat
List * list_concat(List *list1, const List *list2)
Definition list.c:561
list_copy
List * list_copy(const List *oldlist)
Definition list.c:1573
lappend_oid
List * lappend_oid(List *list, Oid datum)
Definition list.c:375
lcons
List * lcons(void *datum, List *list)
Definition list.c:495
list_free
void list_free(List *list)
Definition list.c:1546
list_truncate
List * list_truncate(List *list, int new_size)
Definition list.c:631
list_member
bool list_member(const List *list, const void *datum)
Definition list.c:661
get_commutator
Oid get_commutator(Oid opno)
Definition lsyscache.c:1659
exprType
Oid exprType(const Node *expr)
Definition nodeFuncs.c:42
IsA
#define IsA(nodeptr, _type_)
Definition nodes.h:164
IS_OUTER_JOIN
#define IS_OUTER_JOIN(jointype)
Definition nodes.h:348
castNode
#define castNode(_type_, nodeptr)
Definition nodes.h:182
JoinType
JoinType
Definition nodes.h:298
JOIN_SEMI
@ JOIN_SEMI
Definition nodes.h:317
JOIN_FULL
@ JOIN_FULL
Definition nodes.h:305
JOIN_INNER
@ JOIN_INNER
Definition nodes.h:303
JOIN_RIGHT
@ JOIN_RIGHT
Definition nodes.h:306
JOIN_RIGHT_SEMI
@ JOIN_RIGHT_SEMI
Definition nodes.h:319
JOIN_LEFT
@ JOIN_LEFT
Definition nodes.h:304
JOIN_UNIQUE_OUTER
@ JOIN_UNIQUE_OUTER
Definition nodes.h:326
JOIN_RIGHT_ANTI
@ JOIN_RIGHT_ANTI
Definition nodes.h:320
JOIN_UNIQUE_INNER
@ JOIN_UNIQUE_INNER
Definition nodes.h:327
JOIN_ANTI
@ JOIN_ANTI
Definition nodes.h:318
get_cheapest_path_for_pathkeys
Path * get_cheapest_path_for_pathkeys(List *paths, List *pathkeys, Relids required_outer, CostSelector cost_criterion, bool require_parallel_safe)
Definition pathkeys.c:620
pathkeys_count_contained_in
bool pathkeys_count_contained_in(List *keys1, List *keys2, int *n_common)
Definition pathkeys.c:558
make_inner_pathkeys_for_merge
List * make_inner_pathkeys_for_merge(PlannerInfo *root, List *mergeclauses, List *outer_pathkeys)
Definition pathkeys.c:1858
find_mergeclauses_for_outer_pathkeys
List * find_mergeclauses_for_outer_pathkeys(PlannerInfo *root, List *pathkeys, List *restrictinfos)
Definition pathkeys.c:1544
update_mergeclause_eclasses
void update_mergeclause_eclasses(PlannerInfo *root, RestrictInfo *restrictinfo)
Definition pathkeys.c:1510
trim_mergeclauses_for_inner_pathkeys
List * trim_mergeclauses_for_inner_pathkeys(PlannerInfo *root, List *mergeclauses, List *pathkeys)
Definition pathkeys.c:1961
select_outer_pathkeys_for_merge
List * select_outer_pathkeys_for_merge(PlannerInfo *root, List *mergeclauses, RelOptInfo *joinrel)
Definition pathkeys.c:1659
pathkeys_contained_in
bool pathkeys_contained_in(List *keys1, List *keys2)
Definition pathkeys.c:343
build_join_pathkeys
List * build_join_pathkeys(PlannerInfo *root, RelOptInfo *joinrel, JoinType jointype, List *outer_pathkeys)
Definition pathkeys.c:1295
get_cheapest_parallel_safe_total_inner
Path * get_cheapest_parallel_safe_total_inner(List *paths)
Definition pathkeys.c:699
calc_non_nestloop_required_outer
Relids calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
Definition pathnode.c:2257
path_is_reparameterizable_by_child
bool path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
Definition pathnode.c:4335
create_memoize_path
MemoizePath * create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, List *param_exprs, List *hash_operators, bool singlerow, bool binary_mode, Cardinality est_calls)
Definition pathnode.c:1699
create_material_path
MaterialPath * create_material_path(RelOptInfo *rel, Path *subpath, bool enabled)
Definition pathnode.c:1665
calc_nestloop_required_outer
Relids calc_nestloop_required_outer(Relids outerrelids, Relids outer_paramrels, Relids innerrelids, Relids inner_paramrels)
Definition pathnode.c:2230
create_hashjoin_path
HashPath * create_hashjoin_path(PlannerInfo *root, RelOptInfo *joinrel, JoinType jointype, JoinCostWorkspace *workspace, JoinPathExtraData *extra, Path *outer_path, Path *inner_path, bool parallel_hash, List *restrict_clauses, Relids required_outer, List *hashclauses)
Definition pathnode.c:2474
add_partial_path
void add_partial_path(RelOptInfo *parent_rel, Path *new_path)
Definition pathnode.c:794
add_path
void add_path(RelOptInfo *parent_rel, Path *new_path)
Definition pathnode.c:459
compare_path_costs
int compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
Definition pathnode.c:68
add_path_precheck
bool add_path_precheck(RelOptInfo *parent_rel, int disabled_nodes, Cost startup_cost, Cost total_cost, List *pathkeys, Relids required_outer)
Definition pathnode.c:686
add_partial_path_precheck
bool add_partial_path_precheck(RelOptInfo *parent_rel, int disabled_nodes, Cost total_cost, List *pathkeys)
Definition pathnode.c:925
create_mergejoin_path
MergePath * create_mergejoin_path(PlannerInfo *root, RelOptInfo *joinrel, JoinType jointype, JoinCostWorkspace *workspace, JoinPathExtraData *extra, Path *outer_path, Path *inner_path, List *restrict_clauses, List *pathkeys, Relids required_outer, List *mergeclauses, List *outersortkeys, List *innersortkeys, int outer_presorted_keys)
Definition pathnode.c:2406
create_nestloop_path
NestPath * create_nestloop_path(PlannerInfo *root, RelOptInfo *joinrel, JoinType jointype, JoinCostWorkspace *workspace, JoinPathExtraData *extra, Path *outer_path, Path *inner_path, List *restrict_clauses, List *pathkeys, Relids required_outer)
Definition pathnode.c:2309
PGS_NESTLOOP_MEMOIZE
#define PGS_NESTLOOP_MEMOIZE
Definition pathnodes.h:76
EC_MUST_BE_REDUNDANT
#define EC_MUST_BE_REDUNDANT(eclass)
Definition pathnodes.h:1670
PGS_FOREIGNJOIN
#define PGS_FOREIGNJOIN
Definition pathnodes.h:71
RINFO_IS_PUSHED_DOWN
#define RINFO_IS_PUSHED_DOWN(rinfo, joinrelids)
Definition pathnodes.h:3043
PGS_NESTLOOP_MATERIALIZE
#define PGS_NESTLOOP_MATERIALIZE
Definition pathnodes.h:75
PGS_HASHJOIN
#define PGS_HASHJOIN
Definition pathnodes.h:77
TOTAL_COST
@ TOTAL_COST
Definition pathnodes.h:111
STARTUP_COST
@ STARTUP_COST
Definition pathnodes.h:111
PGS_CONSIDER_NONPARTIAL
#define PGS_CONSIDER_NONPARTIAL
Definition pathnodes.h:84
PATH_REQ_OUTER
#define PATH_REQ_OUTER(path)
Definition pathnodes.h:2001
PGS_MERGEJOIN_ANY
#define PGS_MERGEJOIN_ANY
Definition pathnodes.h:92
RELATION_WAS_MADE_UNIQUE
#define RELATION_WAS_MADE_UNIQUE(rel, sjinfo, nominal_jointype)
Definition pathnodes.h:1238
RELOPT_OTHER_JOINREL
@ RELOPT_OTHER_JOINREL
Definition pathnodes.h:968
PGS_NESTLOOP_PLAIN
#define PGS_NESTLOOP_PLAIN
Definition pathnodes.h:74
join_path_setup_hook_type
void(* join_path_setup_hook_type)(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra)
Definition paths.h:32
set_join_pathlist_hook_type
void(* set_join_pathlist_hook_type)(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel, JoinType jointype, JoinPathExtraData *extra)
Definition paths.h:46
lfirst
#define lfirst(lc)
Definition pg_list.h:172
list_length
static int list_length(const List *l)
Definition pg_list.h:152
NIL
#define NIL
Definition pg_list.h:68
foreach_current_index
#define foreach_current_index(var_or_cell)
Definition pg_list.h:403
linitial
#define linitial(l)
Definition pg_list.h:178
lsecond
#define lsecond(l)
Definition pg_list.h:183
foreach_node
#define foreach_node(type, var, lst)
Definition pg_list.h:496
list_head
static ListCell * list_head(const List *l)
Definition pg_list.h:128
find_placeholder_info
PlaceHolderInfo * find_placeholder_info(PlannerInfo *root, PlaceHolderVar *phv)
Definition placeholder.c:83
InvalidOid
#define InvalidOid
Definition postgres_ext.h:37
Oid
unsigned int Oid
Definition postgres_ext.h:32
fb
static int fb(int x)
Definition preproc-init.c:92
root
tree ctl root
Definition radixtree.h:1857
clause_sides_match_join
static bool clause_sides_match_join(RestrictInfo *rinfo, Relids outerrelids, Relids innerrelids)
Definition restrictinfo.h:73
JoinPathExtraData::mergeclause_list
List * mergeclause_list
Definition pathnodes.h:3592
JoinPathExtraData::param_source_rels
Relids param_source_rels
Definition pathnodes.h:3596
JoinPathExtraData::semifactors
SemiAntiJoinFactors semifactors
Definition pathnodes.h:3595
JoinPathExtraData::sjinfo
SpecialJoinInfo * sjinfo
Definition pathnodes.h:3594
List
Definition pg_list.h:54
Node
Definition nodes.h:135
OpExpr::args
List * args
Definition primnodes.h:869
RelOptInfo::baserestrictinfo
List * baserestrictinfo
Definition pathnodes.h:1130
RelOptInfo::relids
Relids relids
Definition pathnodes.h:1009
RelOptInfo::reltarget
struct PathTarget * reltarget
Definition pathnodes.h:1033
RelOptInfo::lateral_vars
List * lateral_vars
Definition pathnodes.h:1075
RelOptInfo::pgs_mask
uint64 pgs_mask
Definition pathnodes.h:1027
RelOptInfo::consider_parallel
bool consider_parallel
Definition pathnodes.h:1025
RelOptInfo::top_parent_relids
Relids top_parent_relids
Definition pathnodes.h:1162
RelOptInfo::lateral_relids
Relids lateral_relids
Definition pathnodes.h:1052
RelOptInfo::cheapest_parameterized_paths
List * cheapest_parameterized_paths
Definition pathnodes.h:1043
RelOptInfo::pathlist
List * pathlist
Definition pathnodes.h:1038
RelOptInfo::reloptkind
RelOptKind reloptkind
Definition pathnodes.h:1003
RelOptInfo::cheapest_startup_path
struct Path * cheapest_startup_path
Definition pathnodes.h:1041
RelOptInfo::cheapest_total_path
struct Path * cheapest_total_path
Definition pathnodes.h:1042
RelOptInfo::partial_pathlist
List * partial_pathlist
Definition pathnodes.h:1040
RestrictInfo::clause
Expr * clause
Definition pathnodes.h:2886
SpecialJoinInfo::min_righthand
Relids min_righthand
Definition pathnodes.h:3212
SpecialJoinInfo::jointype
JoinType jointype
Definition pathnodes.h:3215
SpecialJoinInfo::min_lefthand
Relids min_lefthand
Definition pathnodes.h:3211
Var::varno
int varno
Definition primnodes.h:270
Var::varlevelsup
Index varlevelsup
Definition primnodes.h:295
lookup_type_cache
TypeCacheEntry * lookup_type_cache(Oid type_id, int flags)
Definition typcache.c:389
TYPECACHE_EQ_OPR
#define TYPECACHE_EQ_OPR
Definition typcache.h:138
TYPECACHE_HASH_PROC
#define TYPECACHE_HASH_PROC
Definition typcache.h:142
pull_varnos
Relids pull_varnos(PlannerInfo *root, Node *node)
Definition var.c:114
pull_vars_of_level
List * pull_vars_of_level(Node *node, int levelsup)
Definition var.c:339