Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * joinpath.c
4 : * Routines to find all possible paths for processing a set of joins
5 : *
6 : * Portions Copyright (c) 1996-2025, 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 <math.h>
18 :
19 : #include "executor/executor.h"
20 : #include "foreign/fdwapi.h"
21 : #include "nodes/nodeFuncs.h"
22 : #include "optimizer/cost.h"
23 : #include "optimizer/optimizer.h"
24 : #include "optimizer/pathnode.h"
25 : #include "optimizer/paths.h"
26 : #include "optimizer/placeholder.h"
27 : #include "optimizer/planmain.h"
28 : #include "optimizer/restrictinfo.h"
29 : #include "utils/lsyscache.h"
30 : #include "utils/typcache.h"
31 :
32 : /* Hook for plugins to get control in add_paths_to_joinrel() */
33 : set_join_pathlist_hook_type set_join_pathlist_hook = NULL;
34 :
35 : /*
36 : * Paths parameterized by a parent rel can be considered to be parameterized
37 : * by any of its children, when we are performing partitionwise joins. These
38 : * macros simplify checking for such cases. Beware multiple eval of args.
39 : */
40 : #define PATH_PARAM_BY_PARENT(path, rel) \
41 : ((path)->param_info && bms_overlap(PATH_REQ_OUTER(path), \
42 : (rel)->top_parent_relids))
43 : #define PATH_PARAM_BY_REL_SELF(path, rel) \
44 : ((path)->param_info && bms_overlap(PATH_REQ_OUTER(path), (rel)->relids))
45 :
46 : #define PATH_PARAM_BY_REL(path, rel) \
47 : (PATH_PARAM_BY_REL_SELF(path, rel) || PATH_PARAM_BY_PARENT(path, rel))
48 :
49 : static void try_partial_mergejoin_path(PlannerInfo *root,
50 : RelOptInfo *joinrel,
51 : Path *outer_path,
52 : Path *inner_path,
53 : List *pathkeys,
54 : List *mergeclauses,
55 : List *outersortkeys,
56 : List *innersortkeys,
57 : JoinType jointype,
58 : JoinPathExtraData *extra);
59 : static void sort_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel,
60 : RelOptInfo *outerrel, RelOptInfo *innerrel,
61 : JoinType jointype, JoinPathExtraData *extra);
62 : static void match_unsorted_outer(PlannerInfo *root, RelOptInfo *joinrel,
63 : RelOptInfo *outerrel, RelOptInfo *innerrel,
64 : JoinType jointype, JoinPathExtraData *extra);
65 : static void consider_parallel_nestloop(PlannerInfo *root,
66 : RelOptInfo *joinrel,
67 : RelOptInfo *outerrel,
68 : RelOptInfo *innerrel,
69 : JoinType jointype,
70 : JoinPathExtraData *extra);
71 : static void consider_parallel_mergejoin(PlannerInfo *root,
72 : RelOptInfo *joinrel,
73 : RelOptInfo *outerrel,
74 : RelOptInfo *innerrel,
75 : JoinType jointype,
76 : JoinPathExtraData *extra,
77 : Path *inner_cheapest_total);
78 : static void hash_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel,
79 : RelOptInfo *outerrel, RelOptInfo *innerrel,
80 : JoinType jointype, JoinPathExtraData *extra);
81 : static List *select_mergejoin_clauses(PlannerInfo *root,
82 : RelOptInfo *joinrel,
83 : RelOptInfo *outerrel,
84 : RelOptInfo *innerrel,
85 : List *restrictlist,
86 : JoinType jointype,
87 : bool *mergejoin_allowed);
88 : static void generate_mergejoin_paths(PlannerInfo *root,
89 : RelOptInfo *joinrel,
90 : RelOptInfo *innerrel,
91 : Path *outerpath,
92 : JoinType jointype,
93 : JoinPathExtraData *extra,
94 : bool useallclauses,
95 : Path *inner_cheapest_total,
96 : List *merge_pathkeys,
97 : bool is_partial);
98 :
99 :
100 : /*
101 : * add_paths_to_joinrel
102 : * Given a join relation and two component rels from which it can be made,
103 : * consider all possible paths that use the two component rels as outer
104 : * and inner rel respectively. Add these paths to the join rel's pathlist
105 : * if they survive comparison with other paths (and remove any existing
106 : * paths that are dominated by these paths).
107 : *
108 : * Modifies the pathlist field of the joinrel node to contain the best
109 : * paths found so far.
110 : *
111 : * jointype is not necessarily the same as sjinfo->jointype; it might be
112 : * "flipped around" if we are considering joining the rels in the opposite
113 : * direction from what's indicated in sjinfo.
114 : *
115 : * Also, this routine accepts the special JoinTypes JOIN_UNIQUE_OUTER and
116 : * JOIN_UNIQUE_INNER to indicate that the outer or inner relation has been
117 : * unique-ified and a regular inner join should then be applied. These values
118 : * are not allowed to propagate outside this routine, however. Path cost
119 : * estimation code, as well as match_unsorted_outer, may need to recognize that
120 : * it's dealing with such a case --- the combination of nominal jointype INNER
121 : * with sjinfo->jointype == JOIN_SEMI indicates that.
122 : */
123 : void
124 644420 : add_paths_to_joinrel(PlannerInfo *root,
125 : RelOptInfo *joinrel,
126 : RelOptInfo *outerrel,
127 : RelOptInfo *innerrel,
128 : JoinType jointype,
129 : SpecialJoinInfo *sjinfo,
130 : List *restrictlist)
131 : {
132 644420 : JoinType save_jointype = jointype;
133 : JoinPathExtraData extra;
134 644420 : bool mergejoin_allowed = true;
135 : ListCell *lc;
136 : Relids joinrelids;
137 :
138 : /*
139 : * PlannerInfo doesn't contain the SpecialJoinInfos created for joins
140 : * between child relations, even if there is a SpecialJoinInfo node for
141 : * the join between the topmost parents. So, while calculating Relids set
142 : * representing the restriction, consider relids of topmost parent of
143 : * partitions.
144 : */
145 644420 : if (joinrel->reloptkind == RELOPT_OTHER_JOINREL)
146 12004 : joinrelids = joinrel->top_parent_relids;
147 : else
148 632416 : joinrelids = joinrel->relids;
149 :
150 644420 : extra.restrictlist = restrictlist;
151 644420 : extra.mergeclause_list = NIL;
152 644420 : extra.sjinfo = sjinfo;
153 644420 : extra.param_source_rels = NULL;
154 :
155 : /*
156 : * See if the inner relation is provably unique for this outer rel.
157 : *
158 : * We have some special cases: for JOIN_SEMI, it doesn't matter since the
159 : * executor can make the equivalent optimization anyway. It also doesn't
160 : * help enable use of Memoize, since a semijoin with a provably unique
161 : * inner side should have been reduced to an inner join in that case.
162 : * Therefore, we need not expend planner cycles on proofs. (For
163 : * JOIN_ANTI, although it doesn't help the executor for the same reason,
164 : * it can benefit Memoize paths.) For JOIN_UNIQUE_INNER, we must be
165 : * considering a semijoin whose inner side is not provably unique (else
166 : * reduce_unique_semijoins would've simplified it), so there's no point in
167 : * calling innerrel_is_unique. However, if the LHS covers all of the
168 : * semijoin's min_lefthand, then it's appropriate to set inner_unique
169 : * because the unique relation produced by create_unique_paths will be
170 : * unique relative to the LHS. (If we have an LHS that's only part of the
171 : * min_lefthand, that is *not* true.) For JOIN_UNIQUE_OUTER, pass
172 : * JOIN_INNER to avoid letting that value escape this module.
173 : */
174 644420 : switch (jointype)
175 : {
176 7514 : case JOIN_SEMI:
177 7514 : extra.inner_unique = false; /* well, unproven */
178 7514 : break;
179 6198 : case JOIN_UNIQUE_INNER:
180 12396 : extra.inner_unique = bms_is_subset(sjinfo->min_lefthand,
181 6198 : outerrel->relids);
182 6198 : break;
183 6198 : case JOIN_UNIQUE_OUTER:
184 6198 : extra.inner_unique = innerrel_is_unique(root,
185 : joinrel->relids,
186 : outerrel->relids,
187 : innerrel,
188 : JOIN_INNER,
189 : restrictlist,
190 : false);
191 6198 : break;
192 624510 : default:
193 624510 : extra.inner_unique = innerrel_is_unique(root,
194 : joinrel->relids,
195 : outerrel->relids,
196 : innerrel,
197 : jointype,
198 : restrictlist,
199 : false);
200 624510 : break;
201 : }
202 :
203 : /*
204 : * If the outer or inner relation has been unique-ified, handle as a plain
205 : * inner join.
206 : */
207 644420 : if (jointype == JOIN_UNIQUE_OUTER || jointype == JOIN_UNIQUE_INNER)
208 12396 : jointype = JOIN_INNER;
209 :
210 : /*
211 : * Find potential mergejoin clauses. We can skip this if we are not
212 : * interested in doing a mergejoin. However, mergejoin may be our only
213 : * way of implementing a full outer join, so override enable_mergejoin if
214 : * it's a full join.
215 : */
216 644420 : if (enable_mergejoin || jointype == JOIN_FULL)
217 640824 : extra.mergeclause_list = select_mergejoin_clauses(root,
218 : joinrel,
219 : outerrel,
220 : innerrel,
221 : restrictlist,
222 : jointype,
223 : &mergejoin_allowed);
224 :
225 : /*
226 : * If it's SEMI, ANTI, or inner_unique join, compute correction factors
227 : * for cost estimation. These will be the same for all paths.
228 : */
229 644420 : if (jointype == JOIN_SEMI || jointype == JOIN_ANTI || extra.inner_unique)
230 199218 : compute_semi_anti_join_factors(root, joinrel, outerrel, innerrel,
231 : jointype, sjinfo, restrictlist,
232 : &extra.semifactors);
233 :
234 : /*
235 : * Decide whether it's sensible to generate parameterized paths for this
236 : * joinrel, and if so, which relations such paths should require. There
237 : * is usually no need to create a parameterized result path unless there
238 : * is a join order restriction that prevents joining one of our input rels
239 : * directly to the parameter source rel instead of joining to the other
240 : * input rel. (But see allow_star_schema_join().) This restriction
241 : * reduces the number of parameterized paths we have to deal with at
242 : * higher join levels, without compromising the quality of the resulting
243 : * plan. We express the restriction as a Relids set that must overlap the
244 : * parameterization of any proposed join path. Note: param_source_rels
245 : * should contain only baserels, not OJ relids, so starting from
246 : * all_baserels not all_query_rels is correct.
247 : */
248 1350224 : foreach(lc, root->join_info_list)
249 : {
250 705804 : SpecialJoinInfo *sjinfo2 = (SpecialJoinInfo *) lfirst(lc);
251 :
252 : /*
253 : * SJ is relevant to this join if we have some part of its RHS
254 : * (possibly not all of it), and haven't yet joined to its LHS. (This
255 : * test is pretty simplistic, but should be sufficient considering the
256 : * join has already been proven legal.) If the SJ is relevant, it
257 : * presents constraints for joining to anything not in its RHS.
258 : */
259 705804 : if (bms_overlap(joinrelids, sjinfo2->min_righthand) &&
260 470352 : !bms_overlap(joinrelids, sjinfo2->min_lefthand))
261 21916 : extra.param_source_rels = bms_join(extra.param_source_rels,
262 21916 : bms_difference(root->all_baserels,
263 21916 : sjinfo2->min_righthand));
264 :
265 : /* full joins constrain both sides symmetrically */
266 710708 : if (sjinfo2->jointype == JOIN_FULL &&
267 4904 : bms_overlap(joinrelids, sjinfo2->min_lefthand) &&
268 4848 : !bms_overlap(joinrelids, sjinfo2->min_righthand))
269 672 : extra.param_source_rels = bms_join(extra.param_source_rels,
270 672 : bms_difference(root->all_baserels,
271 672 : sjinfo2->min_lefthand));
272 : }
273 :
274 : /*
275 : * However, when a LATERAL subquery is involved, there will simply not be
276 : * any paths for the joinrel that aren't parameterized by whatever the
277 : * subquery is parameterized by, unless its parameterization is resolved
278 : * within the joinrel. So we might as well allow additional dependencies
279 : * on whatever residual lateral dependencies the joinrel will have.
280 : */
281 1288840 : extra.param_source_rels = bms_add_members(extra.param_source_rels,
282 644420 : joinrel->lateral_relids);
283 :
284 : /*
285 : * 1. Consider mergejoin paths where both relations must be explicitly
286 : * sorted. Skip this if we can't mergejoin.
287 : */
288 644420 : if (mergejoin_allowed)
289 628664 : sort_inner_and_outer(root, joinrel, outerrel, innerrel,
290 : jointype, &extra);
291 :
292 : /*
293 : * 2. Consider paths where the outer relation need not be explicitly
294 : * sorted. This includes both nestloops and mergejoins where the outer
295 : * path is already ordered. Again, skip this if we can't mergejoin.
296 : * (That's okay because we know that nestloop can't handle
297 : * right/right-anti/right-semi/full joins at all, so it wouldn't work in
298 : * the prohibited cases either.)
299 : */
300 644420 : if (mergejoin_allowed)
301 628664 : match_unsorted_outer(root, joinrel, outerrel, innerrel,
302 : jointype, &extra);
303 :
304 : #ifdef NOT_USED
305 :
306 : /*
307 : * 3. Consider paths where the inner relation need not be explicitly
308 : * sorted. This includes mergejoins only (nestloops were already built in
309 : * match_unsorted_outer).
310 : *
311 : * Diked out as redundant 2/13/2000 -- tgl. There isn't any really
312 : * significant difference between the inner and outer side of a mergejoin,
313 : * so match_unsorted_inner creates no paths that aren't equivalent to
314 : * those made by match_unsorted_outer when add_paths_to_joinrel() is
315 : * invoked with the two rels given in the other order.
316 : */
317 : if (mergejoin_allowed)
318 : match_unsorted_inner(root, joinrel, outerrel, innerrel,
319 : jointype, &extra);
320 : #endif
321 :
322 : /*
323 : * 4. Consider paths where both outer and inner relations must be hashed
324 : * before being joined. As above, disregard enable_hashjoin for full
325 : * joins, because there may be no other alternative.
326 : */
327 644420 : if (enable_hashjoin || jointype == JOIN_FULL)
328 639088 : hash_inner_and_outer(root, joinrel, outerrel, innerrel,
329 : jointype, &extra);
330 :
331 : /*
332 : * 5. If inner and outer relations are foreign tables (or joins) belonging
333 : * to the same server and assigned to the same user to check access
334 : * permissions as, give the FDW a chance to push down joins.
335 : */
336 644420 : if (joinrel->fdwroutine &&
337 2708 : joinrel->fdwroutine->GetForeignJoinPaths)
338 2704 : joinrel->fdwroutine->GetForeignJoinPaths(root, joinrel,
339 : outerrel, innerrel,
340 : save_jointype, &extra);
341 :
342 : /*
343 : * 6. Finally, give extensions a chance to manipulate the path list. They
344 : * could add new paths (such as CustomPaths) by calling add_path(), or
345 : * add_partial_path() if parallel aware. They could also delete or modify
346 : * paths added by the core code.
347 : */
348 644420 : if (set_join_pathlist_hook)
349 0 : set_join_pathlist_hook(root, joinrel, outerrel, innerrel,
350 : save_jointype, &extra);
351 644420 : }
352 :
353 : /*
354 : * We override the param_source_rels heuristic to accept nestloop paths in
355 : * which the outer rel satisfies some but not all of the inner path's
356 : * parameterization. This is necessary to get good plans for star-schema
357 : * scenarios, in which a parameterized path for a large table may require
358 : * parameters from multiple small tables that will not get joined directly to
359 : * each other. We can handle that by stacking nestloops that have the small
360 : * tables on the outside; but this breaks the rule the param_source_rels
361 : * heuristic is based on, namely that parameters should not be passed down
362 : * across joins unless there's a join-order-constraint-based reason to do so.
363 : * So we ignore the param_source_rels restriction when this case applies.
364 : *
365 : * allow_star_schema_join() returns true if the param_source_rels restriction
366 : * should be overridden, ie, it's okay to perform this join.
367 : */
368 : static inline bool
369 272096 : allow_star_schema_join(PlannerInfo *root,
370 : Relids outerrelids,
371 : Relids inner_paramrels)
372 : {
373 : /*
374 : * It's a star-schema case if the outer rel provides some but not all of
375 : * the inner rel's parameterization.
376 : */
377 318282 : return (bms_overlap(inner_paramrels, outerrelids) &&
378 46186 : bms_nonempty_difference(inner_paramrels, outerrelids));
379 : }
380 :
381 : /*
382 : * If the parameterization is only partly satisfied by the outer rel,
383 : * the unsatisfied part can't include any outer-join relids that could
384 : * null rels of the satisfied part. That would imply that we're trying
385 : * to use a clause involving a Var with nonempty varnullingrels at
386 : * a join level where that value isn't yet computable.
387 : *
388 : * In practice, this test never finds a problem because earlier join order
389 : * restrictions prevent us from attempting a join that would cause a problem.
390 : * (That's unsurprising, because the code worked before we ever added
391 : * outer-join relids to expression relids.) It still seems worth checking
392 : * as a backstop, but we only do so in assert-enabled builds.
393 : */
394 : #ifdef USE_ASSERT_CHECKING
395 : static inline bool
396 : have_unsafe_outer_join_ref(PlannerInfo *root,
397 : Relids outerrelids,
398 : Relids inner_paramrels)
399 : {
400 : bool result = false;
401 : Relids unsatisfied = bms_difference(inner_paramrels, outerrelids);
402 : Relids satisfied = bms_intersect(inner_paramrels, outerrelids);
403 :
404 : if (bms_overlap(unsatisfied, root->outer_join_rels))
405 : {
406 : ListCell *lc;
407 :
408 : foreach(lc, root->join_info_list)
409 : {
410 : SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc);
411 :
412 : if (!bms_is_member(sjinfo->ojrelid, unsatisfied))
413 : continue; /* not relevant */
414 : if (bms_overlap(satisfied, sjinfo->min_righthand) ||
415 : (sjinfo->jointype == JOIN_FULL &&
416 : bms_overlap(satisfied, sjinfo->min_lefthand)))
417 : {
418 : result = true; /* doesn't work */
419 : break;
420 : }
421 : }
422 : }
423 :
424 : /* Waste no memory when we reject a path here */
425 : bms_free(unsatisfied);
426 : bms_free(satisfied);
427 :
428 : return result;
429 : }
430 : #endif /* USE_ASSERT_CHECKING */
431 :
432 : /*
433 : * paraminfo_get_equal_hashops
434 : * Determine if the clauses in param_info and innerrel's lateral vars
435 : * can be hashed.
436 : * Returns true if hashing is possible, otherwise false.
437 : *
438 : * Additionally, on success we collect the outer expressions and the
439 : * appropriate equality operators for each hashable parameter to innerrel.
440 : * These are returned in parallel lists in *param_exprs and *operators.
441 : * We also set *binary_mode to indicate whether strict binary matching is
442 : * required.
443 : */
444 : static bool
445 365136 : paraminfo_get_equal_hashops(PlannerInfo *root, ParamPathInfo *param_info,
446 : RelOptInfo *outerrel, RelOptInfo *innerrel,
447 : List *ph_lateral_vars, List **param_exprs,
448 : List **operators, bool *binary_mode)
449 :
450 : {
451 : List *lateral_vars;
452 : ListCell *lc;
453 :
454 365136 : *param_exprs = NIL;
455 365136 : *operators = NIL;
456 365136 : *binary_mode = false;
457 :
458 : /* Add join clauses from param_info to the hash key */
459 365136 : if (param_info != NULL)
460 : {
461 365136 : List *clauses = param_info->ppi_clauses;
462 :
463 695372 : foreach(lc, clauses)
464 : {
465 401286 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
466 : OpExpr *opexpr;
467 : Node *expr;
468 : Oid hasheqoperator;
469 :
470 401286 : opexpr = (OpExpr *) rinfo->clause;
471 :
472 : /*
473 : * Bail if the rinfo is not compatible. We need a join OpExpr
474 : * with 2 args.
475 : */
476 401286 : if (!IsA(opexpr, OpExpr) || list_length(opexpr->args) != 2 ||
477 383716 : !clause_sides_match_join(rinfo, outerrel->relids,
478 : innerrel->relids))
479 : {
480 70894 : list_free(*operators);
481 70894 : list_free(*param_exprs);
482 71050 : return false;
483 : }
484 :
485 330392 : if (rinfo->outer_is_left)
486 : {
487 183314 : expr = (Node *) linitial(opexpr->args);
488 183314 : hasheqoperator = rinfo->left_hasheqoperator;
489 : }
490 : else
491 : {
492 147078 : expr = (Node *) lsecond(opexpr->args);
493 147078 : hasheqoperator = rinfo->right_hasheqoperator;
494 : }
495 :
496 : /* can't do memoize if we can't hash the outer type */
497 330392 : if (!OidIsValid(hasheqoperator))
498 : {
499 156 : list_free(*operators);
500 156 : list_free(*param_exprs);
501 156 : return false;
502 : }
503 :
504 : /*
505 : * 'expr' may already exist as a parameter from a previous item in
506 : * ppi_clauses. No need to include it again, however we'd better
507 : * ensure we do switch into binary mode if required. See below.
508 : */
509 330236 : if (!list_member(*param_exprs, expr))
510 : {
511 330230 : *operators = lappend_oid(*operators, hasheqoperator);
512 330230 : *param_exprs = lappend(*param_exprs, expr);
513 : }
514 :
515 : /*
516 : * When the join operator is not hashable then it's possible that
517 : * the operator will be able to distinguish something that the
518 : * hash equality operator could not. For example with floating
519 : * point types -0.0 and +0.0 are classed as equal by the hash
520 : * function and equality function, but some other operator may be
521 : * able to tell those values apart. This means that we must put
522 : * memoize into binary comparison mode so that it does bit-by-bit
523 : * comparisons rather than a "logical" comparison as it would
524 : * using the hash equality operator.
525 : */
526 330236 : if (!OidIsValid(rinfo->hashjoinoperator))
527 1266 : *binary_mode = true;
528 : }
529 : }
530 :
531 : /* Now add any lateral vars to the cache key too */
532 294086 : lateral_vars = list_concat(ph_lateral_vars, innerrel->lateral_vars);
533 298844 : foreach(lc, lateral_vars)
534 : {
535 4956 : Node *expr = (Node *) lfirst(lc);
536 : TypeCacheEntry *typentry;
537 :
538 : /* Reject if there are any volatile functions in lateral vars */
539 4956 : if (contain_volatile_functions(expr))
540 : {
541 0 : list_free(*operators);
542 0 : list_free(*param_exprs);
543 198 : return false;
544 : }
545 :
546 4956 : typentry = lookup_type_cache(exprType(expr),
547 : TYPECACHE_HASH_PROC | TYPECACHE_EQ_OPR);
548 :
549 : /* can't use memoize without a valid hash proc and equals operator */
550 4956 : if (!OidIsValid(typentry->hash_proc) || !OidIsValid(typentry->eq_opr))
551 : {
552 198 : list_free(*operators);
553 198 : list_free(*param_exprs);
554 198 : return false;
555 : }
556 :
557 : /*
558 : * 'expr' may already exist as a parameter from the ppi_clauses. No
559 : * need to include it again, however we'd better ensure we do switch
560 : * into binary mode.
561 : */
562 4758 : if (!list_member(*param_exprs, expr))
563 : {
564 4308 : *operators = lappend_oid(*operators, typentry->eq_opr);
565 4308 : *param_exprs = lappend(*param_exprs, expr);
566 : }
567 :
568 : /*
569 : * We must go into binary mode as we don't have too much of an idea of
570 : * how these lateral Vars are being used. See comment above when we
571 : * set *binary_mode for the non-lateral Var case. This could be
572 : * relaxed a bit if we had the RestrictInfos and knew the operators
573 : * being used, however for cases like Vars that are arguments to
574 : * functions we must operate in binary mode as we don't have
575 : * visibility into what the function is doing with the Vars.
576 : */
577 4758 : *binary_mode = true;
578 : }
579 :
580 : /* We're okay to use memoize */
581 293888 : return true;
582 : }
583 :
584 : /*
585 : * extract_lateral_vars_from_PHVs
586 : * Extract lateral references within PlaceHolderVars that are due to be
587 : * evaluated at 'innerrelids'.
588 : */
589 : static List *
590 1205806 : extract_lateral_vars_from_PHVs(PlannerInfo *root, Relids innerrelids)
591 : {
592 1205806 : List *ph_lateral_vars = NIL;
593 : ListCell *lc;
594 :
595 : /* Nothing would be found if the query contains no LATERAL RTEs */
596 1205806 : if (!root->hasLateralRTEs)
597 1185332 : return NIL;
598 :
599 : /*
600 : * No need to consider PHVs that are due to be evaluated at joinrels,
601 : * since we do not add Memoize nodes on top of joinrel paths.
602 : */
603 20474 : if (bms_membership(innerrelids) == BMS_MULTIPLE)
604 6540 : return NIL;
605 :
606 16798 : foreach(lc, root->placeholder_list)
607 : {
608 2864 : PlaceHolderInfo *phinfo = (PlaceHolderInfo *) lfirst(lc);
609 : List *vars;
610 : ListCell *cell;
611 :
612 : /* PHV is uninteresting if no lateral refs */
613 2864 : if (phinfo->ph_lateral == NULL)
614 1638 : continue;
615 :
616 : /* PHV is uninteresting if not due to be evaluated at innerrelids */
617 1226 : if (!bms_equal(phinfo->ph_eval_at, innerrelids))
618 982 : continue;
619 :
620 : /*
621 : * If the PHV does not reference any rels in innerrelids, use its
622 : * contained expression as a cache key rather than extracting the
623 : * Vars/PHVs from it and using those. This can be beneficial in cases
624 : * where the expression results in fewer distinct values to cache
625 : * tuples for.
626 : */
627 244 : if (!bms_overlap(pull_varnos(root, (Node *) phinfo->ph_var->phexpr),
628 : innerrelids))
629 : {
630 232 : ph_lateral_vars = lappend(ph_lateral_vars, phinfo->ph_var->phexpr);
631 232 : continue;
632 : }
633 :
634 : /* Fetch Vars and PHVs of lateral references within PlaceHolderVars */
635 12 : vars = pull_vars_of_level((Node *) phinfo->ph_var->phexpr, 0);
636 36 : foreach(cell, vars)
637 : {
638 24 : Node *node = (Node *) lfirst(cell);
639 :
640 24 : if (IsA(node, Var))
641 : {
642 12 : Var *var = (Var *) node;
643 :
644 : Assert(var->varlevelsup == 0);
645 :
646 12 : if (bms_is_member(var->varno, phinfo->ph_lateral))
647 0 : ph_lateral_vars = lappend(ph_lateral_vars, node);
648 : }
649 12 : else if (IsA(node, PlaceHolderVar))
650 : {
651 12 : PlaceHolderVar *phv = (PlaceHolderVar *) node;
652 :
653 : Assert(phv->phlevelsup == 0);
654 :
655 12 : if (bms_is_subset(find_placeholder_info(root, phv)->ph_eval_at,
656 12 : phinfo->ph_lateral))
657 12 : ph_lateral_vars = lappend(ph_lateral_vars, node);
658 : }
659 : else
660 : Assert(false);
661 : }
662 :
663 12 : list_free(vars);
664 : }
665 :
666 13934 : return ph_lateral_vars;
667 : }
668 :
669 : /*
670 : * get_memoize_path
671 : * If possible, make and return a Memoize path atop of 'inner_path'.
672 : * Otherwise return NULL.
673 : *
674 : * Note that currently we do not add Memoize nodes on top of join relation
675 : * paths. This is because the ParamPathInfos for join relation paths do not
676 : * maintain ppi_clauses, as the set of relevant clauses varies depending on how
677 : * the join is formed. In addition, joinrels do not maintain lateral_vars. So
678 : * we do not have a way to extract cache keys from joinrels.
679 : */
680 : static Path *
681 1706802 : get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
682 : RelOptInfo *outerrel, Path *inner_path,
683 : Path *outer_path, JoinType jointype,
684 : JoinPathExtraData *extra)
685 : {
686 : List *param_exprs;
687 : List *hash_operators;
688 : ListCell *lc;
689 : bool binary_mode;
690 : List *ph_lateral_vars;
691 :
692 : /* Obviously not if it's disabled */
693 1706802 : if (!enable_memoize)
694 412 : return NULL;
695 :
696 : /*
697 : * We can safely not bother with all this unless we expect to perform more
698 : * than one inner scan. The first scan is always going to be a cache
699 : * miss. This would likely fail later anyway based on costs, so this is
700 : * really just to save some wasted effort.
701 : */
702 1706390 : if (outer_path->parent->rows < 2)
703 500584 : return NULL;
704 :
705 : /*
706 : * Extract lateral Vars/PHVs within PlaceHolderVars that are due to be
707 : * evaluated at innerrel. These lateral Vars/PHVs could be used as
708 : * memoize cache keys.
709 : */
710 1205806 : ph_lateral_vars = extract_lateral_vars_from_PHVs(root, innerrel->relids);
711 :
712 : /*
713 : * We can only have a memoize node when there's some kind of cache key,
714 : * either parameterized path clauses or lateral Vars. No cache key sounds
715 : * more like something a Materialize node might be more useful for.
716 : */
717 1205806 : if ((inner_path->param_info == NULL ||
718 495812 : inner_path->param_info->ppi_clauses == NIL) &&
719 735830 : innerrel->lateral_vars == NIL &&
720 : ph_lateral_vars == NIL)
721 732526 : return NULL;
722 :
723 : /*
724 : * Currently we don't do this for SEMI and ANTI joins, because nested loop
725 : * SEMI/ANTI joins don't scan the inner node to completion, which means
726 : * memoize cannot mark the cache entry as complete. Nor can we mark the
727 : * cache entry as complete after fetching the first inner tuple, because
728 : * if that tuple and the current outer tuple don't satisfy the join
729 : * clauses, a second inner tuple that satisfies the parameters would find
730 : * the cache entry already marked as complete. The only exception is when
731 : * the inner relation is provably unique, as in that case, there won't be
732 : * a second matching tuple and we can safely mark the cache entry as
733 : * complete after fetching the first inner tuple. Note that in such
734 : * cases, the SEMI join should have been reduced to an inner join by
735 : * reduce_unique_semijoins.
736 : */
737 473280 : if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
738 10070 : !extra->inner_unique)
739 6340 : return NULL;
740 :
741 : /*
742 : * Memoize normally marks cache entries as complete when it runs out of
743 : * tuples to read from its subplan. However, with unique joins, Nested
744 : * Loop will skip to the next outer tuple after finding the first matching
745 : * inner tuple. This means that we may not read the inner side of the
746 : * join to completion which leaves no opportunity to mark the cache entry
747 : * as complete. To work around that, when the join is unique we
748 : * automatically mark cache entries as complete after fetching the first
749 : * tuple. This works when the entire join condition is parameterized.
750 : * Otherwise, when the parameterization is only a subset of the join
751 : * condition, we can't be sure which part of it causes the join to be
752 : * unique. This means there are no guarantees that only 1 tuple will be
753 : * read. We cannot mark the cache entry as complete after reading the
754 : * first tuple without that guarantee. This means the scope of Memoize
755 : * node's usefulness is limited to only outer rows that have no join
756 : * partner as this is the only case where Nested Loop would exhaust the
757 : * inner scan of a unique join. Since the scope is limited to that, we
758 : * just don't bother making a memoize path in this case.
759 : *
760 : * Lateral vars needn't be considered here as they're not considered when
761 : * determining if the join is unique.
762 : */
763 466940 : if (extra->inner_unique)
764 : {
765 : Bitmapset *ppi_serials;
766 :
767 280332 : if (inner_path->param_info == NULL)
768 0 : return NULL;
769 :
770 280332 : ppi_serials = inner_path->param_info->ppi_serials;
771 :
772 674486 : foreach_node(RestrictInfo, rinfo, extra->restrictlist)
773 : {
774 317402 : if (!bms_is_member(rinfo->rinfo_serial, ppi_serials))
775 101790 : return NULL;
776 : }
777 : }
778 :
779 : /*
780 : * We can't use a memoize node if there are volatile functions in the
781 : * inner rel's target list or restrict list. A cache hit could reduce the
782 : * number of calls to these functions.
783 : */
784 365150 : if (contain_volatile_functions((Node *) innerrel->reltarget))
785 0 : return NULL;
786 :
787 604028 : foreach(lc, innerrel->baserestrictinfo)
788 : {
789 238890 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
790 :
791 238890 : if (contain_volatile_functions((Node *) rinfo))
792 12 : return NULL;
793 : }
794 :
795 : /*
796 : * Also check the parameterized path restrictinfos for volatile functions.
797 : * Indexed functions must be immutable so shouldn't have any volatile
798 : * functions, however, with a lateral join the inner scan may not be an
799 : * index scan.
800 : */
801 365138 : if (inner_path->param_info != NULL)
802 : {
803 791258 : foreach(lc, inner_path->param_info->ppi_clauses)
804 : {
805 426122 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
806 :
807 426122 : if (contain_volatile_functions((Node *) rinfo))
808 2 : return NULL;
809 : }
810 : }
811 :
812 : /* Check if we have hash ops for each parameter to the path */
813 365136 : if (paraminfo_get_equal_hashops(root,
814 : inner_path->param_info,
815 365136 : outerrel->top_parent ?
816 : outerrel->top_parent : outerrel,
817 : innerrel,
818 : ph_lateral_vars,
819 : ¶m_exprs,
820 : &hash_operators,
821 : &binary_mode))
822 : {
823 293888 : return (Path *) create_memoize_path(root,
824 : innerrel,
825 : inner_path,
826 : param_exprs,
827 : hash_operators,
828 293888 : extra->inner_unique,
829 : binary_mode,
830 : outer_path->rows);
831 : }
832 :
833 71248 : return NULL;
834 : }
835 :
836 : /*
837 : * try_nestloop_path
838 : * Consider a nestloop join path; if it appears useful, push it into
839 : * the joinrel's pathlist via add_path().
840 : */
841 : static void
842 2944738 : try_nestloop_path(PlannerInfo *root,
843 : RelOptInfo *joinrel,
844 : Path *outer_path,
845 : Path *inner_path,
846 : List *pathkeys,
847 : JoinType jointype,
848 : JoinPathExtraData *extra)
849 : {
850 : Relids required_outer;
851 : JoinCostWorkspace workspace;
852 2944738 : RelOptInfo *innerrel = inner_path->parent;
853 2944738 : RelOptInfo *outerrel = outer_path->parent;
854 : Relids innerrelids;
855 : Relids outerrelids;
856 2944738 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
857 2944738 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
858 :
859 : /*
860 : * If we are forming an outer join at this join, it's nonsensical to use
861 : * an input path that uses the outer join as part of its parameterization.
862 : * (This can happen despite our join order restrictions, since those apply
863 : * to what is in an input relation not what its parameters are.)
864 : */
865 3616510 : if (extra->sjinfo->ojrelid != 0 &&
866 1343544 : (bms_is_member(extra->sjinfo->ojrelid, inner_paramrels) ||
867 671772 : bms_is_member(extra->sjinfo->ojrelid, outer_paramrels)))
868 271092 : return;
869 :
870 : /*
871 : * Any parameterization of the input paths refers to topmost parents of
872 : * the relevant relations, because reparameterize_path_by_child() hasn't
873 : * been called yet. So we must consider topmost parents of the relations
874 : * being joined, too, while determining parameterization of the result and
875 : * checking for disallowed parameterization cases.
876 : */
877 2944618 : if (innerrel->top_parent_relids)
878 41380 : innerrelids = innerrel->top_parent_relids;
879 : else
880 2903238 : innerrelids = innerrel->relids;
881 :
882 2944618 : if (outerrel->top_parent_relids)
883 41380 : outerrelids = outerrel->top_parent_relids;
884 : else
885 2903238 : outerrelids = outerrel->relids;
886 :
887 : /*
888 : * Check to see if proposed path is still parameterized, and reject if the
889 : * parameterization wouldn't be sensible --- unless allow_star_schema_join
890 : * says to allow it anyway.
891 : */
892 2944618 : required_outer = calc_nestloop_required_outer(outerrelids, outer_paramrels,
893 : innerrelids, inner_paramrels);
894 2944618 : if (required_outer &&
895 308980 : !bms_overlap(required_outer, extra->param_source_rels) &&
896 272096 : !allow_star_schema_join(root, outerrelids, inner_paramrels))
897 : {
898 : /* Waste no memory when we reject a path here */
899 270972 : bms_free(required_outer);
900 270972 : return;
901 : }
902 :
903 : /* If we got past that, we shouldn't have any unsafe outer-join refs */
904 : Assert(!have_unsafe_outer_join_ref(root, outerrelids, inner_paramrels));
905 :
906 : /*
907 : * If the inner path is parameterized, it is parameterized by the topmost
908 : * parent of the outer rel, not the outer rel itself. We will need to
909 : * translate the parameterization, if this path is chosen, during
910 : * create_plan(). Here we just check whether we will be able to perform
911 : * the translation, and if not avoid creating a nestloop path.
912 : */
913 2673646 : if (PATH_PARAM_BY_PARENT(inner_path, outer_path->parent) &&
914 13336 : !path_is_reparameterizable_by_child(inner_path, outer_path->parent))
915 : {
916 0 : bms_free(required_outer);
917 0 : return;
918 : }
919 :
920 : /*
921 : * Do a precheck to quickly eliminate obviously-inferior paths. We
922 : * calculate a cheap lower bound on the path's cost and then use
923 : * add_path_precheck() to see if the path is clearly going to be dominated
924 : * by some existing path for the joinrel. If not, do the full pushup with
925 : * creating a fully valid path structure and submitting it to add_path().
926 : * The latter two steps are expensive enough to make this two-phase
927 : * methodology worthwhile.
928 : */
929 2673646 : initial_cost_nestloop(root, &workspace, jointype,
930 : outer_path, inner_path, extra);
931 :
932 2673646 : if (add_path_precheck(joinrel, workspace.disabled_nodes,
933 : workspace.startup_cost, workspace.total_cost,
934 : pathkeys, required_outer))
935 : {
936 1311432 : add_path(joinrel, (Path *)
937 1311432 : create_nestloop_path(root,
938 : joinrel,
939 : jointype,
940 : &workspace,
941 : extra,
942 : outer_path,
943 : inner_path,
944 : extra->restrictlist,
945 : pathkeys,
946 : required_outer));
947 : }
948 : else
949 : {
950 : /* Waste no memory when we reject a path here */
951 1362214 : bms_free(required_outer);
952 : }
953 : }
954 :
955 : /*
956 : * try_partial_nestloop_path
957 : * Consider a partial nestloop join path; if it appears useful, push it into
958 : * the joinrel's partial_pathlist via add_partial_path().
959 : */
960 : static void
961 45050 : try_partial_nestloop_path(PlannerInfo *root,
962 : RelOptInfo *joinrel,
963 : Path *outer_path,
964 : Path *inner_path,
965 : List *pathkeys,
966 : JoinType jointype,
967 : JoinPathExtraData *extra)
968 : {
969 : JoinCostWorkspace workspace;
970 :
971 : /*
972 : * If the inner path is parameterized, the parameterization must be fully
973 : * satisfied by the proposed outer path. Parameterized partial paths are
974 : * not supported. The caller should already have verified that no lateral
975 : * rels are required here.
976 : */
977 : Assert(bms_is_empty(joinrel->lateral_relids));
978 : Assert(bms_is_empty(PATH_REQ_OUTER(outer_path)));
979 45050 : if (inner_path->param_info != NULL)
980 : {
981 15694 : Relids inner_paramrels = inner_path->param_info->ppi_req_outer;
982 15694 : RelOptInfo *outerrel = outer_path->parent;
983 : Relids outerrelids;
984 :
985 : /*
986 : * The inner and outer paths are parameterized, if at all, by the top
987 : * level parents, not the child relations, so we must use those relids
988 : * for our parameterization tests.
989 : */
990 15694 : if (outerrel->top_parent_relids)
991 11592 : outerrelids = outerrel->top_parent_relids;
992 : else
993 4102 : outerrelids = outerrel->relids;
994 :
995 15694 : if (!bms_is_subset(inner_paramrels, outerrelids))
996 31600 : return;
997 : }
998 :
999 : /*
1000 : * If the inner path is parameterized, it is parameterized by the topmost
1001 : * parent of the outer rel, not the outer rel itself. We will need to
1002 : * translate the parameterization, if this path is chosen, during
1003 : * create_plan(). Here we just check whether we will be able to perform
1004 : * the translation, and if not avoid creating a nestloop path.
1005 : */
1006 43708 : if (PATH_PARAM_BY_PARENT(inner_path, outer_path->parent) &&
1007 10620 : !path_is_reparameterizable_by_child(inner_path, outer_path->parent))
1008 0 : return;
1009 :
1010 : /*
1011 : * Before creating a path, get a quick lower bound on what it is likely to
1012 : * cost. Bail out right away if it looks terrible.
1013 : */
1014 43708 : initial_cost_nestloop(root, &workspace, jointype,
1015 : outer_path, inner_path, extra);
1016 43708 : if (!add_partial_path_precheck(joinrel, workspace.disabled_nodes,
1017 : workspace.total_cost, pathkeys))
1018 30258 : return;
1019 :
1020 : /* Might be good enough to be worth trying, so let's try it. */
1021 13450 : add_partial_path(joinrel, (Path *)
1022 13450 : create_nestloop_path(root,
1023 : joinrel,
1024 : jointype,
1025 : &workspace,
1026 : extra,
1027 : outer_path,
1028 : inner_path,
1029 : extra->restrictlist,
1030 : pathkeys,
1031 : NULL));
1032 : }
1033 :
1034 : /*
1035 : * try_mergejoin_path
1036 : * Consider a merge join path; if it appears useful, push it into
1037 : * the joinrel's pathlist via add_path().
1038 : */
1039 : static void
1040 1207786 : try_mergejoin_path(PlannerInfo *root,
1041 : RelOptInfo *joinrel,
1042 : Path *outer_path,
1043 : Path *inner_path,
1044 : List *pathkeys,
1045 : List *mergeclauses,
1046 : List *outersortkeys,
1047 : List *innersortkeys,
1048 : JoinType jointype,
1049 : JoinPathExtraData *extra,
1050 : bool is_partial)
1051 : {
1052 : Relids required_outer;
1053 1207786 : int outer_presorted_keys = 0;
1054 : JoinCostWorkspace workspace;
1055 :
1056 1207786 : if (is_partial)
1057 : {
1058 7478 : try_partial_mergejoin_path(root,
1059 : joinrel,
1060 : outer_path,
1061 : inner_path,
1062 : pathkeys,
1063 : mergeclauses,
1064 : outersortkeys,
1065 : innersortkeys,
1066 : jointype,
1067 : extra);
1068 44390 : return;
1069 : }
1070 :
1071 : /*
1072 : * If we are forming an outer join at this join, it's nonsensical to use
1073 : * an input path that uses the outer join as part of its parameterization.
1074 : * (This can happen despite our join order restrictions, since those apply
1075 : * to what is in an input relation not what its parameters are.)
1076 : */
1077 1575862 : if (extra->sjinfo->ojrelid != 0 &&
1078 751108 : (bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(inner_path)) ||
1079 375554 : bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(outer_path))))
1080 12 : return;
1081 :
1082 : /*
1083 : * Check to see if proposed path is still parameterized, and reject if the
1084 : * parameterization wouldn't be sensible.
1085 : */
1086 1200296 : required_outer = calc_non_nestloop_required_outer(outer_path,
1087 : inner_path);
1088 1200296 : if (required_outer &&
1089 39726 : !bms_overlap(required_outer, extra->param_source_rels))
1090 : {
1091 : /* Waste no memory when we reject a path here */
1092 36900 : bms_free(required_outer);
1093 36900 : return;
1094 : }
1095 :
1096 : /*
1097 : * If the given paths are already well enough ordered, we can skip doing
1098 : * an explicit sort.
1099 : *
1100 : * We need to determine the number of presorted keys of the outer path to
1101 : * decide whether explicit incremental sort can be applied when
1102 : * outersortkeys is not NIL. We do not need to do the same for the inner
1103 : * path though, as incremental sort currently does not support
1104 : * mark/restore.
1105 : */
1106 1733284 : if (outersortkeys &&
1107 569888 : pathkeys_count_contained_in(outersortkeys, outer_path->pathkeys,
1108 : &outer_presorted_keys))
1109 15880 : outersortkeys = NIL;
1110 2104406 : if (innersortkeys &&
1111 941010 : pathkeys_contained_in(innersortkeys, inner_path->pathkeys))
1112 27202 : innersortkeys = NIL;
1113 :
1114 : /*
1115 : * See comments in try_nestloop_path().
1116 : */
1117 1163396 : initial_cost_mergejoin(root, &workspace, jointype, mergeclauses,
1118 : outer_path, inner_path,
1119 : outersortkeys, innersortkeys,
1120 : outer_presorted_keys,
1121 : extra);
1122 :
1123 1163396 : if (add_path_precheck(joinrel, workspace.disabled_nodes,
1124 : workspace.startup_cost, workspace.total_cost,
1125 : pathkeys, required_outer))
1126 : {
1127 300960 : add_path(joinrel, (Path *)
1128 300960 : create_mergejoin_path(root,
1129 : joinrel,
1130 : jointype,
1131 : &workspace,
1132 : extra,
1133 : outer_path,
1134 : inner_path,
1135 : extra->restrictlist,
1136 : pathkeys,
1137 : required_outer,
1138 : mergeclauses,
1139 : outersortkeys,
1140 : innersortkeys,
1141 : outer_presorted_keys));
1142 : }
1143 : else
1144 : {
1145 : /* Waste no memory when we reject a path here */
1146 862436 : bms_free(required_outer);
1147 : }
1148 : }
1149 :
1150 : /*
1151 : * try_partial_mergejoin_path
1152 : * Consider a partial merge join path; if it appears useful, push it into
1153 : * the joinrel's pathlist via add_partial_path().
1154 : */
1155 : static void
1156 20942 : try_partial_mergejoin_path(PlannerInfo *root,
1157 : RelOptInfo *joinrel,
1158 : Path *outer_path,
1159 : Path *inner_path,
1160 : List *pathkeys,
1161 : List *mergeclauses,
1162 : List *outersortkeys,
1163 : List *innersortkeys,
1164 : JoinType jointype,
1165 : JoinPathExtraData *extra)
1166 : {
1167 20942 : int outer_presorted_keys = 0;
1168 : JoinCostWorkspace workspace;
1169 :
1170 : /*
1171 : * See comments in try_partial_hashjoin_path().
1172 : */
1173 : Assert(bms_is_empty(joinrel->lateral_relids));
1174 : Assert(bms_is_empty(PATH_REQ_OUTER(outer_path)));
1175 20942 : if (!bms_is_empty(PATH_REQ_OUTER(inner_path)))
1176 10844 : return;
1177 :
1178 : /*
1179 : * If the given paths are already well enough ordered, we can skip doing
1180 : * an explicit sort.
1181 : *
1182 : * We need to determine the number of presorted keys of the outer path to
1183 : * decide whether explicit incremental sort can be applied when
1184 : * outersortkeys is not NIL. We do not need to do the same for the inner
1185 : * path though, as incremental sort currently does not support
1186 : * mark/restore.
1187 : */
1188 34406 : if (outersortkeys &&
1189 13464 : pathkeys_count_contained_in(outersortkeys, outer_path->pathkeys,
1190 : &outer_presorted_keys))
1191 192 : outersortkeys = NIL;
1192 38646 : if (innersortkeys &&
1193 17704 : pathkeys_contained_in(innersortkeys, inner_path->pathkeys))
1194 612 : innersortkeys = NIL;
1195 :
1196 : /*
1197 : * See comments in try_partial_nestloop_path().
1198 : */
1199 20942 : initial_cost_mergejoin(root, &workspace, jointype, mergeclauses,
1200 : outer_path, inner_path,
1201 : outersortkeys, innersortkeys,
1202 : outer_presorted_keys,
1203 : extra);
1204 :
1205 20942 : if (!add_partial_path_precheck(joinrel, workspace.disabled_nodes,
1206 : workspace.total_cost, pathkeys))
1207 10844 : return;
1208 :
1209 : /* Might be good enough to be worth trying, so let's try it. */
1210 10098 : add_partial_path(joinrel, (Path *)
1211 10098 : create_mergejoin_path(root,
1212 : joinrel,
1213 : jointype,
1214 : &workspace,
1215 : extra,
1216 : outer_path,
1217 : inner_path,
1218 : extra->restrictlist,
1219 : pathkeys,
1220 : NULL,
1221 : mergeclauses,
1222 : outersortkeys,
1223 : innersortkeys,
1224 : outer_presorted_keys));
1225 : }
1226 :
1227 : /*
1228 : * try_hashjoin_path
1229 : * Consider a hash join path; if it appears useful, push it into
1230 : * the joinrel's pathlist via add_path().
1231 : */
1232 : static void
1233 724784 : try_hashjoin_path(PlannerInfo *root,
1234 : RelOptInfo *joinrel,
1235 : Path *outer_path,
1236 : Path *inner_path,
1237 : List *hashclauses,
1238 : JoinType jointype,
1239 : JoinPathExtraData *extra)
1240 : {
1241 : Relids required_outer;
1242 : JoinCostWorkspace workspace;
1243 :
1244 : /*
1245 : * If we are forming an outer join at this join, it's nonsensical to use
1246 : * an input path that uses the outer join as part of its parameterization.
1247 : * (This can happen despite our join order restrictions, since those apply
1248 : * to what is in an input relation not what its parameters are.)
1249 : */
1250 976424 : if (extra->sjinfo->ojrelid != 0 &&
1251 503250 : (bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(inner_path)) ||
1252 251610 : bms_is_member(extra->sjinfo->ojrelid, PATH_REQ_OUTER(outer_path))))
1253 106872 : return;
1254 :
1255 : /*
1256 : * Check to see if proposed path is still parameterized, and reject if the
1257 : * parameterization wouldn't be sensible.
1258 : */
1259 724724 : required_outer = calc_non_nestloop_required_outer(outer_path,
1260 : inner_path);
1261 724724 : if (required_outer &&
1262 121178 : !bms_overlap(required_outer, extra->param_source_rels))
1263 : {
1264 : /* Waste no memory when we reject a path here */
1265 106812 : bms_free(required_outer);
1266 106812 : return;
1267 : }
1268 :
1269 : /*
1270 : * See comments in try_nestloop_path(). Also note that hashjoin paths
1271 : * never have any output pathkeys, per comments in create_hashjoin_path.
1272 : */
1273 617912 : initial_cost_hashjoin(root, &workspace, jointype, hashclauses,
1274 : outer_path, inner_path, extra, false);
1275 :
1276 617912 : if (add_path_precheck(joinrel, workspace.disabled_nodes,
1277 : workspace.startup_cost, workspace.total_cost,
1278 : NIL, required_outer))
1279 : {
1280 268148 : add_path(joinrel, (Path *)
1281 268148 : create_hashjoin_path(root,
1282 : joinrel,
1283 : jointype,
1284 : &workspace,
1285 : extra,
1286 : outer_path,
1287 : inner_path,
1288 : false, /* parallel_hash */
1289 : extra->restrictlist,
1290 : required_outer,
1291 : hashclauses));
1292 : }
1293 : else
1294 : {
1295 : /* Waste no memory when we reject a path here */
1296 349764 : bms_free(required_outer);
1297 : }
1298 : }
1299 :
1300 : /*
1301 : * try_partial_hashjoin_path
1302 : * Consider a partial hashjoin join path; if it appears useful, push it into
1303 : * the joinrel's partial_pathlist via add_partial_path().
1304 : * The outer side is partial. If parallel_hash is true, then the inner path
1305 : * must be partial and will be run in parallel to create one or more shared
1306 : * hash tables; otherwise the inner path must be complete and a copy of it
1307 : * is run in every process to create separate identical private hash tables.
1308 : */
1309 : static void
1310 22346 : try_partial_hashjoin_path(PlannerInfo *root,
1311 : RelOptInfo *joinrel,
1312 : Path *outer_path,
1313 : Path *inner_path,
1314 : List *hashclauses,
1315 : JoinType jointype,
1316 : JoinPathExtraData *extra,
1317 : bool parallel_hash)
1318 : {
1319 : JoinCostWorkspace workspace;
1320 :
1321 : /*
1322 : * If the inner path is parameterized, we can't use a partial hashjoin.
1323 : * Parameterized partial paths are not supported. The caller should
1324 : * already have verified that no lateral rels are required here.
1325 : */
1326 : Assert(bms_is_empty(joinrel->lateral_relids));
1327 : Assert(bms_is_empty(PATH_REQ_OUTER(outer_path)));
1328 22346 : if (!bms_is_empty(PATH_REQ_OUTER(inner_path)))
1329 10568 : return;
1330 :
1331 : /*
1332 : * Before creating a path, get a quick lower bound on what it is likely to
1333 : * cost. Bail out right away if it looks terrible.
1334 : */
1335 22346 : initial_cost_hashjoin(root, &workspace, jointype, hashclauses,
1336 : outer_path, inner_path, extra, parallel_hash);
1337 22346 : if (!add_partial_path_precheck(joinrel, workspace.disabled_nodes,
1338 : workspace.total_cost, NIL))
1339 10568 : return;
1340 :
1341 : /* Might be good enough to be worth trying, so let's try it. */
1342 11778 : add_partial_path(joinrel, (Path *)
1343 11778 : create_hashjoin_path(root,
1344 : joinrel,
1345 : jointype,
1346 : &workspace,
1347 : extra,
1348 : outer_path,
1349 : inner_path,
1350 : parallel_hash,
1351 : extra->restrictlist,
1352 : NULL,
1353 : hashclauses));
1354 : }
1355 :
1356 : /*
1357 : * sort_inner_and_outer
1358 : * Create mergejoin join paths by explicitly sorting both the outer and
1359 : * inner join relations on each available merge ordering.
1360 : *
1361 : * 'joinrel' is the join relation
1362 : * 'outerrel' is the outer join relation
1363 : * 'innerrel' is the inner join relation
1364 : * 'jointype' is the type of join to do
1365 : * 'extra' contains additional input values
1366 : */
1367 : static void
1368 628664 : sort_inner_and_outer(PlannerInfo *root,
1369 : RelOptInfo *joinrel,
1370 : RelOptInfo *outerrel,
1371 : RelOptInfo *innerrel,
1372 : JoinType jointype,
1373 : JoinPathExtraData *extra)
1374 : {
1375 : Path *outer_path;
1376 : Path *inner_path;
1377 628664 : Path *cheapest_partial_outer = NULL;
1378 628664 : Path *cheapest_safe_inner = NULL;
1379 : List *all_pathkeys;
1380 : ListCell *l;
1381 :
1382 : /* Nothing to do if there are no available mergejoin clauses */
1383 628664 : if (extra->mergeclause_list == NIL)
1384 115354 : return;
1385 :
1386 : /*
1387 : * We only consider the cheapest-total-cost input paths, since we are
1388 : * assuming here that a sort is required. We will consider
1389 : * cheapest-startup-cost input paths later, and only if they don't need a
1390 : * sort.
1391 : *
1392 : * This function intentionally does not consider parameterized input
1393 : * paths, except when the cheapest-total is parameterized. If we did so,
1394 : * we'd have a combinatorial explosion of mergejoin paths of dubious
1395 : * value. This interacts with decisions elsewhere that also discriminate
1396 : * against mergejoins with parameterized inputs; see comments in
1397 : * src/backend/optimizer/README.
1398 : */
1399 513310 : outer_path = outerrel->cheapest_total_path;
1400 513310 : inner_path = innerrel->cheapest_total_path;
1401 :
1402 : /*
1403 : * If either cheapest-total path is parameterized by the other rel, we
1404 : * can't use a mergejoin. (There's no use looking for alternative input
1405 : * paths, since these should already be the least-parameterized available
1406 : * paths.)
1407 : */
1408 513310 : if (PATH_PARAM_BY_REL(outer_path, innerrel) ||
1409 512418 : PATH_PARAM_BY_REL(inner_path, outerrel))
1410 1820 : return;
1411 :
1412 : /*
1413 : * If the joinrel is parallel-safe, we may be able to consider a partial
1414 : * merge join. However, we can't handle JOIN_FULL, JOIN_RIGHT and
1415 : * JOIN_RIGHT_ANTI, because they can produce false null extended rows.
1416 : * Also, the resulting path must not be parameterized.
1417 : */
1418 511490 : if (joinrel->consider_parallel &&
1419 451094 : jointype != JOIN_FULL &&
1420 370310 : jointype != JOIN_RIGHT &&
1421 368854 : jointype != JOIN_RIGHT_ANTI &&
1422 368854 : outerrel->partial_pathlist != NIL &&
1423 9908 : bms_is_empty(joinrel->lateral_relids))
1424 : {
1425 9908 : cheapest_partial_outer = (Path *) linitial(outerrel->partial_pathlist);
1426 :
1427 9908 : if (inner_path->parallel_safe)
1428 9788 : cheapest_safe_inner = inner_path;
1429 : else
1430 : cheapest_safe_inner =
1431 120 : get_cheapest_parallel_safe_total_inner(innerrel->pathlist);
1432 : }
1433 :
1434 : /*
1435 : * Each possible ordering of the available mergejoin clauses will generate
1436 : * a differently-sorted result path at essentially the same cost. We have
1437 : * no basis for choosing one over another at this level of joining, but
1438 : * some sort orders may be more useful than others for higher-level
1439 : * mergejoins, so it's worth considering multiple orderings.
1440 : *
1441 : * Actually, it's not quite true that every mergeclause ordering will
1442 : * generate a different path order, because some of the clauses may be
1443 : * partially redundant (refer to the same EquivalenceClasses). Therefore,
1444 : * what we do is convert the mergeclause list to a list of canonical
1445 : * pathkeys, and then consider different orderings of the pathkeys.
1446 : *
1447 : * Generating a path for *every* permutation of the pathkeys doesn't seem
1448 : * like a winning strategy; the cost in planning time is too high. For
1449 : * now, we generate one path for each pathkey, listing that pathkey first
1450 : * and the rest in random order. This should allow at least a one-clause
1451 : * mergejoin without re-sorting against any other possible mergejoin
1452 : * partner path. But if we've not guessed the right ordering of secondary
1453 : * keys, we may end up evaluating clauses as qpquals when they could have
1454 : * been done as mergeclauses. (In practice, it's rare that there's more
1455 : * than two or three mergeclauses, so expending a huge amount of thought
1456 : * on that is probably not worth it.)
1457 : *
1458 : * The pathkey order returned by select_outer_pathkeys_for_merge() has
1459 : * some heuristics behind it (see that function), so be sure to try it
1460 : * exactly as-is as well as making variants.
1461 : */
1462 511490 : all_pathkeys = select_outer_pathkeys_for_merge(root,
1463 : extra->mergeclause_list,
1464 : joinrel);
1465 :
1466 1081378 : foreach(l, all_pathkeys)
1467 : {
1468 569888 : PathKey *front_pathkey = (PathKey *) lfirst(l);
1469 : List *cur_mergeclauses;
1470 : List *outerkeys;
1471 : List *innerkeys;
1472 : List *merge_pathkeys;
1473 :
1474 : /* Make a pathkey list with this guy first */
1475 569888 : if (l != list_head(all_pathkeys))
1476 58398 : outerkeys = lcons(front_pathkey,
1477 : list_delete_nth_cell(list_copy(all_pathkeys),
1478 : foreach_current_index(l)));
1479 : else
1480 511490 : outerkeys = all_pathkeys; /* no work at first one... */
1481 :
1482 : /* Sort the mergeclauses into the corresponding ordering */
1483 : cur_mergeclauses =
1484 569888 : find_mergeclauses_for_outer_pathkeys(root,
1485 : outerkeys,
1486 : extra->mergeclause_list);
1487 :
1488 : /* Should have used them all... */
1489 : Assert(list_length(cur_mergeclauses) == list_length(extra->mergeclause_list));
1490 :
1491 : /* Build sort pathkeys for the inner side */
1492 569888 : innerkeys = make_inner_pathkeys_for_merge(root,
1493 : cur_mergeclauses,
1494 : outerkeys);
1495 :
1496 : /* Build pathkeys representing output sort order */
1497 569888 : merge_pathkeys = build_join_pathkeys(root, joinrel, jointype,
1498 : outerkeys);
1499 :
1500 : /*
1501 : * And now we can make the path.
1502 : *
1503 : * Note: it's possible that the cheapest paths will already be sorted
1504 : * properly. try_mergejoin_path will detect that case and suppress an
1505 : * explicit sort step, so we needn't do so here.
1506 : */
1507 569888 : try_mergejoin_path(root,
1508 : joinrel,
1509 : outer_path,
1510 : inner_path,
1511 : merge_pathkeys,
1512 : cur_mergeclauses,
1513 : outerkeys,
1514 : innerkeys,
1515 : jointype,
1516 : extra,
1517 : false);
1518 :
1519 : /*
1520 : * If we have partial outer and parallel safe inner path then try
1521 : * partial mergejoin path.
1522 : */
1523 569888 : if (cheapest_partial_outer && cheapest_safe_inner)
1524 13464 : try_partial_mergejoin_path(root,
1525 : joinrel,
1526 : cheapest_partial_outer,
1527 : cheapest_safe_inner,
1528 : merge_pathkeys,
1529 : cur_mergeclauses,
1530 : outerkeys,
1531 : innerkeys,
1532 : jointype,
1533 : extra);
1534 : }
1535 : }
1536 :
1537 : /*
1538 : * generate_mergejoin_paths
1539 : * Creates possible mergejoin paths for input outerpath.
1540 : *
1541 : * We generate mergejoins if mergejoin clauses are available. We have
1542 : * two ways to generate the inner path for a mergejoin: sort the cheapest
1543 : * inner path, or use an inner path that is already suitably ordered for the
1544 : * merge. If we have several mergeclauses, it could be that there is no inner
1545 : * path (or only a very expensive one) for the full list of mergeclauses, but
1546 : * better paths exist if we truncate the mergeclause list (thereby discarding
1547 : * some sort key requirements). So, we consider truncations of the
1548 : * mergeclause list as well as the full list. (Ideally we'd consider all
1549 : * subsets of the mergeclause list, but that seems way too expensive.)
1550 : */
1551 : static void
1552 1220420 : generate_mergejoin_paths(PlannerInfo *root,
1553 : RelOptInfo *joinrel,
1554 : RelOptInfo *innerrel,
1555 : Path *outerpath,
1556 : JoinType jointype,
1557 : JoinPathExtraData *extra,
1558 : bool useallclauses,
1559 : Path *inner_cheapest_total,
1560 : List *merge_pathkeys,
1561 : bool is_partial)
1562 : {
1563 : List *mergeclauses;
1564 : List *innersortkeys;
1565 : List *trialsortkeys;
1566 : Path *cheapest_startup_inner;
1567 : Path *cheapest_total_inner;
1568 : int num_sortkeys;
1569 : int sortkeycnt;
1570 :
1571 : /* Look for useful mergeclauses (if any) */
1572 : mergeclauses =
1573 1220420 : find_mergeclauses_for_outer_pathkeys(root,
1574 : outerpath->pathkeys,
1575 : extra->mergeclause_list);
1576 :
1577 : /*
1578 : * Done with this outer path if no chance for a mergejoin.
1579 : *
1580 : * Special corner case: for "x FULL JOIN y ON true", there will be no join
1581 : * clauses at all. Ordinarily we'd generate a clauseless nestloop path,
1582 : * but since mergejoin is our only join type that supports FULL JOIN
1583 : * without any join clauses, it's necessary to generate a clauseless
1584 : * mergejoin path instead.
1585 : */
1586 1220420 : if (mergeclauses == NIL)
1587 : {
1588 814096 : if (jointype == JOIN_FULL)
1589 : /* okay to try for mergejoin */ ;
1590 : else
1591 810712 : return;
1592 : }
1593 496628 : if (useallclauses &&
1594 86920 : list_length(mergeclauses) != list_length(extra->mergeclause_list))
1595 13412 : return;
1596 :
1597 : /* Compute the required ordering of the inner path */
1598 396296 : innersortkeys = make_inner_pathkeys_for_merge(root,
1599 : mergeclauses,
1600 : outerpath->pathkeys);
1601 :
1602 : /*
1603 : * Generate a mergejoin on the basis of sorting the cheapest inner. Since
1604 : * a sort will be needed, only cheapest total cost matters. (But
1605 : * try_mergejoin_path will do the right thing if inner_cheapest_total is
1606 : * already correctly sorted.)
1607 : */
1608 396296 : try_mergejoin_path(root,
1609 : joinrel,
1610 : outerpath,
1611 : inner_cheapest_total,
1612 : merge_pathkeys,
1613 : mergeclauses,
1614 : NIL,
1615 : innersortkeys,
1616 : jointype,
1617 : extra,
1618 : is_partial);
1619 :
1620 : /*
1621 : * Look for presorted inner paths that satisfy the innersortkey list ---
1622 : * or any truncation thereof, if we are allowed to build a mergejoin using
1623 : * a subset of the merge clauses. Here, we consider both cheap startup
1624 : * cost and cheap total cost.
1625 : *
1626 : * Currently we do not consider parameterized inner paths here. This
1627 : * interacts with decisions elsewhere that also discriminate against
1628 : * mergejoins with parameterized inputs; see comments in
1629 : * src/backend/optimizer/README.
1630 : *
1631 : * As we shorten the sortkey list, we should consider only paths that are
1632 : * strictly cheaper than (in particular, not the same as) any path found
1633 : * in an earlier iteration. Otherwise we'd be intentionally using fewer
1634 : * merge keys than a given path allows (treating the rest as plain
1635 : * joinquals), which is unlikely to be a good idea. Also, eliminating
1636 : * paths here on the basis of compare_path_costs is a lot cheaper than
1637 : * building the mergejoin path only to throw it away.
1638 : *
1639 : * If inner_cheapest_total is well enough sorted to have not required a
1640 : * sort in the path made above, we shouldn't make a duplicate path with
1641 : * it, either. We handle that case with the same logic that handles the
1642 : * previous consideration, by initializing the variables that track
1643 : * cheapest-so-far properly. Note that we do NOT reject
1644 : * inner_cheapest_total if we find it matches some shorter set of
1645 : * pathkeys. That case corresponds to using fewer mergekeys to avoid
1646 : * sorting inner_cheapest_total, whereas we did sort it above, so the
1647 : * plans being considered are different.
1648 : */
1649 396296 : if (pathkeys_contained_in(innersortkeys,
1650 : inner_cheapest_total->pathkeys))
1651 : {
1652 : /* inner_cheapest_total didn't require a sort */
1653 12644 : cheapest_startup_inner = inner_cheapest_total;
1654 12644 : cheapest_total_inner = inner_cheapest_total;
1655 : }
1656 : else
1657 : {
1658 : /* it did require a sort, at least for the full set of keys */
1659 383652 : cheapest_startup_inner = NULL;
1660 383652 : cheapest_total_inner = NULL;
1661 : }
1662 396296 : num_sortkeys = list_length(innersortkeys);
1663 396296 : if (num_sortkeys > 1 && !useallclauses)
1664 15490 : trialsortkeys = list_copy(innersortkeys); /* need modifiable copy */
1665 : else
1666 380806 : trialsortkeys = innersortkeys; /* won't really truncate */
1667 :
1668 734710 : for (sortkeycnt = num_sortkeys; sortkeycnt > 0; sortkeycnt--)
1669 : {
1670 : Path *innerpath;
1671 411778 : List *newclauses = NIL;
1672 :
1673 : /*
1674 : * Look for an inner path ordered well enough for the first
1675 : * 'sortkeycnt' innersortkeys. NB: trialsortkeys list is modified
1676 : * destructively, which is why we made a copy...
1677 : */
1678 411778 : trialsortkeys = list_truncate(trialsortkeys, sortkeycnt);
1679 411778 : innerpath = get_cheapest_path_for_pathkeys(innerrel->pathlist,
1680 : trialsortkeys,
1681 : NULL,
1682 : TOTAL_COST,
1683 : is_partial);
1684 411778 : if (innerpath != NULL &&
1685 23146 : (cheapest_total_inner == NULL ||
1686 23146 : compare_path_costs(innerpath, cheapest_total_inner,
1687 : TOTAL_COST) < 0))
1688 : {
1689 : /* Found a cheap (or even-cheaper) sorted path */
1690 : /* Select the right mergeclauses, if we didn't already */
1691 238794 : if (sortkeycnt < num_sortkeys)
1692 : {
1693 : newclauses =
1694 5482 : trim_mergeclauses_for_inner_pathkeys(root,
1695 : mergeclauses,
1696 : trialsortkeys);
1697 : Assert(newclauses != NIL);
1698 : }
1699 : else
1700 233312 : newclauses = mergeclauses;
1701 238794 : try_mergejoin_path(root,
1702 : joinrel,
1703 : outerpath,
1704 : innerpath,
1705 : merge_pathkeys,
1706 : newclauses,
1707 : NIL,
1708 : NIL,
1709 : jointype,
1710 : extra,
1711 : is_partial);
1712 238794 : cheapest_total_inner = innerpath;
1713 : }
1714 : /* Same on the basis of cheapest startup cost ... */
1715 411778 : innerpath = get_cheapest_path_for_pathkeys(innerrel->pathlist,
1716 : trialsortkeys,
1717 : NULL,
1718 : STARTUP_COST,
1719 : is_partial);
1720 411778 : if (innerpath != NULL &&
1721 23146 : (cheapest_startup_inner == NULL ||
1722 23146 : compare_path_costs(innerpath, cheapest_startup_inner,
1723 : STARTUP_COST) < 0))
1724 : {
1725 : /* Found a cheap (or even-cheaper) sorted path */
1726 236282 : if (innerpath != cheapest_total_inner)
1727 : {
1728 : /*
1729 : * Avoid rebuilding clause list if we already made one; saves
1730 : * memory in big join trees...
1731 : */
1732 2808 : if (newclauses == NIL)
1733 : {
1734 102 : if (sortkeycnt < num_sortkeys)
1735 : {
1736 : newclauses =
1737 0 : trim_mergeclauses_for_inner_pathkeys(root,
1738 : mergeclauses,
1739 : trialsortkeys);
1740 : Assert(newclauses != NIL);
1741 : }
1742 : else
1743 102 : newclauses = mergeclauses;
1744 : }
1745 2808 : try_mergejoin_path(root,
1746 : joinrel,
1747 : outerpath,
1748 : innerpath,
1749 : merge_pathkeys,
1750 : newclauses,
1751 : NIL,
1752 : NIL,
1753 : jointype,
1754 : extra,
1755 : is_partial);
1756 : }
1757 236282 : cheapest_startup_inner = innerpath;
1758 : }
1759 :
1760 : /*
1761 : * Don't consider truncated sortkeys if we need all clauses.
1762 : */
1763 411778 : if (useallclauses)
1764 73364 : break;
1765 : }
1766 : }
1767 :
1768 : /*
1769 : * match_unsorted_outer
1770 : * Creates possible join paths for processing a single join relation
1771 : * 'joinrel' by employing either iterative substitution or
1772 : * mergejoining on each of its possible outer paths (considering
1773 : * only outer paths that are already ordered well enough for merging).
1774 : *
1775 : * We always generate a nestloop path for each available outer path.
1776 : * In fact we may generate as many as five: one on the cheapest-total-cost
1777 : * inner path, one on the same with materialization, one on the
1778 : * cheapest-startup-cost inner path (if different), one on the
1779 : * cheapest-total inner-indexscan path (if any), and one on the
1780 : * cheapest-startup inner-indexscan path (if different).
1781 : *
1782 : * We also consider mergejoins if mergejoin clauses are available. See
1783 : * detailed comments in generate_mergejoin_paths.
1784 : *
1785 : * 'joinrel' is the join relation
1786 : * 'outerrel' is the outer join relation
1787 : * 'innerrel' is the inner join relation
1788 : * 'jointype' is the type of join to do
1789 : * 'extra' contains additional input values
1790 : */
1791 : static void
1792 628664 : match_unsorted_outer(PlannerInfo *root,
1793 : RelOptInfo *joinrel,
1794 : RelOptInfo *outerrel,
1795 : RelOptInfo *innerrel,
1796 : JoinType jointype,
1797 : JoinPathExtraData *extra)
1798 : {
1799 : bool nestjoinOK;
1800 : bool useallclauses;
1801 628664 : Path *inner_cheapest_total = innerrel->cheapest_total_path;
1802 628664 : Path *matpath = NULL;
1803 : ListCell *lc1;
1804 :
1805 : /*
1806 : * For now we do not support RIGHT_SEMI join in mergejoin or nestloop
1807 : * join.
1808 : */
1809 628664 : if (jointype == JOIN_RIGHT_SEMI)
1810 102 : return;
1811 :
1812 : /*
1813 : * Nestloop only supports inner, left, semi, and anti joins. Also, if we
1814 : * are doing a right, right-anti or full mergejoin, we must use *all* the
1815 : * mergeclauses as join clauses, else we will not have a valid plan.
1816 : * (Although these two flags are currently inverses, keep them separate
1817 : * for clarity and possible future changes.)
1818 : */
1819 628562 : switch (jointype)
1820 : {
1821 533864 : case JOIN_INNER:
1822 : case JOIN_LEFT:
1823 : case JOIN_SEMI:
1824 : case JOIN_ANTI:
1825 533864 : nestjoinOK = true;
1826 533864 : useallclauses = false;
1827 533864 : break;
1828 94698 : case JOIN_RIGHT:
1829 : case JOIN_RIGHT_ANTI:
1830 : case JOIN_FULL:
1831 94698 : nestjoinOK = false;
1832 94698 : useallclauses = true;
1833 94698 : break;
1834 0 : default:
1835 0 : elog(ERROR, "unrecognized join type: %d",
1836 : (int) jointype);
1837 : nestjoinOK = false; /* keep compiler quiet */
1838 : useallclauses = false;
1839 : break;
1840 : }
1841 :
1842 : /*
1843 : * If inner_cheapest_total is parameterized by the outer rel, ignore it;
1844 : * we will consider it below as a member of cheapest_parameterized_paths,
1845 : * but the other possibilities considered in this routine aren't usable.
1846 : *
1847 : * Furthermore, if the inner side is a unique-ified relation, we cannot
1848 : * generate any valid paths here, because the inner rel's dependency on
1849 : * the outer rel makes unique-ification meaningless.
1850 : */
1851 628562 : if (PATH_PARAM_BY_REL(inner_cheapest_total, outerrel))
1852 : {
1853 12170 : inner_cheapest_total = NULL;
1854 :
1855 12170 : if (RELATION_WAS_MADE_UNIQUE(innerrel, extra->sjinfo, jointype))
1856 36 : return;
1857 : }
1858 :
1859 628526 : if (nestjoinOK)
1860 : {
1861 : /*
1862 : * Consider materializing the cheapest inner path, unless
1863 : * enable_material is off or the path in question materializes its
1864 : * output anyway.
1865 : */
1866 533828 : if (enable_material && inner_cheapest_total != NULL &&
1867 521212 : !ExecMaterializesOutput(inner_cheapest_total->pathtype))
1868 : matpath = (Path *)
1869 494828 : create_material_path(innerrel, inner_cheapest_total);
1870 : }
1871 :
1872 2087328 : foreach(lc1, outerrel->pathlist)
1873 : {
1874 1458802 : Path *outerpath = (Path *) lfirst(lc1);
1875 : List *merge_pathkeys;
1876 :
1877 : /*
1878 : * We cannot use an outer path that is parameterized by the inner rel.
1879 : */
1880 1458802 : if (PATH_PARAM_BY_REL(outerpath, innerrel))
1881 240422 : continue;
1882 :
1883 : /*
1884 : * The result will have this sort order (even if it is implemented as
1885 : * a nestloop, and even if some of the mergeclauses are implemented by
1886 : * qpquals rather than as true mergeclauses):
1887 : */
1888 1218380 : merge_pathkeys = build_join_pathkeys(root, joinrel, jointype,
1889 : outerpath->pathkeys);
1890 :
1891 1218380 : if (nestjoinOK)
1892 : {
1893 : /*
1894 : * Consider nestloop joins using this outer path and various
1895 : * available paths for the inner relation. We consider the
1896 : * cheapest-total paths for each available parameterization of the
1897 : * inner relation, including the unparameterized case.
1898 : */
1899 : ListCell *lc2;
1900 :
1901 2721398 : foreach(lc2, innerrel->cheapest_parameterized_paths)
1902 : {
1903 1682498 : Path *innerpath = (Path *) lfirst(lc2);
1904 : Path *mpath;
1905 :
1906 1682498 : try_nestloop_path(root,
1907 : joinrel,
1908 : outerpath,
1909 : innerpath,
1910 : merge_pathkeys,
1911 : jointype,
1912 : extra);
1913 :
1914 : /*
1915 : * Try generating a memoize path and see if that makes the
1916 : * nested loop any cheaper.
1917 : */
1918 1682498 : mpath = get_memoize_path(root, innerrel, outerrel,
1919 : innerpath, outerpath, jointype,
1920 : extra);
1921 1682498 : if (mpath != NULL)
1922 287764 : try_nestloop_path(root,
1923 : joinrel,
1924 : outerpath,
1925 : mpath,
1926 : merge_pathkeys,
1927 : jointype,
1928 : extra);
1929 : }
1930 :
1931 : /* Also consider materialized form of the cheapest inner path */
1932 1038900 : if (matpath != NULL)
1933 974476 : try_nestloop_path(root,
1934 : joinrel,
1935 : outerpath,
1936 : matpath,
1937 : merge_pathkeys,
1938 : jointype,
1939 : extra);
1940 : }
1941 :
1942 : /* Can't do anything else if inner rel is parameterized by outer */
1943 1218380 : if (inner_cheapest_total == NULL)
1944 13090 : continue;
1945 :
1946 : /* Generate merge join paths */
1947 1205290 : generate_mergejoin_paths(root, joinrel, innerrel, outerpath,
1948 : jointype, extra, useallclauses,
1949 : inner_cheapest_total, merge_pathkeys,
1950 : false);
1951 : }
1952 :
1953 : /*
1954 : * Consider partial nestloop and mergejoin plan if outerrel has any
1955 : * partial path and the joinrel is parallel-safe. However, we can't
1956 : * handle joins needing lateral rels, since partial paths must not be
1957 : * parameterized. Similarly, we can't handle JOIN_FULL, JOIN_RIGHT and
1958 : * JOIN_RIGHT_ANTI, because they can produce false null extended rows.
1959 : */
1960 628526 : if (joinrel->consider_parallel &&
1961 525840 : jointype != JOIN_FULL &&
1962 443410 : jointype != JOIN_RIGHT &&
1963 441936 : jointype != JOIN_RIGHT_ANTI &&
1964 441936 : outerrel->partial_pathlist != NIL &&
1965 11542 : bms_is_empty(joinrel->lateral_relids))
1966 : {
1967 11542 : if (nestjoinOK)
1968 11542 : consider_parallel_nestloop(root, joinrel, outerrel, innerrel,
1969 : jointype, extra);
1970 :
1971 : /*
1972 : * If inner_cheapest_total is NULL or non parallel-safe then find the
1973 : * cheapest total parallel safe path.
1974 : */
1975 11542 : if (inner_cheapest_total == NULL ||
1976 11134 : !inner_cheapest_total->parallel_safe)
1977 : {
1978 : inner_cheapest_total =
1979 804 : get_cheapest_parallel_safe_total_inner(innerrel->pathlist);
1980 : }
1981 :
1982 11542 : if (inner_cheapest_total)
1983 11110 : consider_parallel_mergejoin(root, joinrel, outerrel, innerrel,
1984 : jointype, extra,
1985 : inner_cheapest_total);
1986 : }
1987 : }
1988 :
1989 : /*
1990 : * consider_parallel_mergejoin
1991 : * Try to build partial paths for a joinrel by joining a partial path
1992 : * for the outer relation to a complete path for the inner relation.
1993 : *
1994 : * 'joinrel' is the join relation
1995 : * 'outerrel' is the outer join relation
1996 : * 'innerrel' is the inner join relation
1997 : * 'jointype' is the type of join to do
1998 : * 'extra' contains additional input values
1999 : * 'inner_cheapest_total' cheapest total path for innerrel
2000 : */
2001 : static void
2002 11110 : consider_parallel_mergejoin(PlannerInfo *root,
2003 : RelOptInfo *joinrel,
2004 : RelOptInfo *outerrel,
2005 : RelOptInfo *innerrel,
2006 : JoinType jointype,
2007 : JoinPathExtraData *extra,
2008 : Path *inner_cheapest_total)
2009 : {
2010 : ListCell *lc1;
2011 :
2012 : /* generate merge join path for each partial outer path */
2013 26240 : foreach(lc1, outerrel->partial_pathlist)
2014 : {
2015 15130 : Path *outerpath = (Path *) lfirst(lc1);
2016 : List *merge_pathkeys;
2017 :
2018 : /*
2019 : * Figure out what useful ordering any paths we create will have.
2020 : */
2021 15130 : merge_pathkeys = build_join_pathkeys(root, joinrel, jointype,
2022 : outerpath->pathkeys);
2023 :
2024 15130 : generate_mergejoin_paths(root, joinrel, innerrel, outerpath, jointype,
2025 : extra, false, inner_cheapest_total,
2026 : merge_pathkeys, true);
2027 : }
2028 11110 : }
2029 :
2030 : /*
2031 : * consider_parallel_nestloop
2032 : * Try to build partial paths for a joinrel by joining a partial path for the
2033 : * outer relation to a complete path for the inner relation.
2034 : *
2035 : * 'joinrel' is the join relation
2036 : * 'outerrel' is the outer join relation
2037 : * 'innerrel' is the inner join relation
2038 : * 'jointype' is the type of join to do
2039 : * 'extra' contains additional input values
2040 : */
2041 : static void
2042 11542 : consider_parallel_nestloop(PlannerInfo *root,
2043 : RelOptInfo *joinrel,
2044 : RelOptInfo *outerrel,
2045 : RelOptInfo *innerrel,
2046 : JoinType jointype,
2047 : JoinPathExtraData *extra)
2048 : {
2049 11542 : Path *inner_cheapest_total = innerrel->cheapest_total_path;
2050 11542 : Path *matpath = NULL;
2051 : ListCell *lc1;
2052 :
2053 : /*
2054 : * Consider materializing the cheapest inner path, unless: 1)
2055 : * enable_material is off, 2) the cheapest inner path is not
2056 : * parallel-safe, 3) the cheapest inner path is parameterized by the outer
2057 : * rel, or 4) the cheapest inner path materializes its output anyway.
2058 : */
2059 11542 : if (enable_material && inner_cheapest_total->parallel_safe &&
2060 11112 : !PATH_PARAM_BY_REL(inner_cheapest_total, outerrel) &&
2061 10722 : !ExecMaterializesOutput(inner_cheapest_total->pathtype))
2062 : {
2063 : matpath = (Path *)
2064 10626 : create_material_path(innerrel, inner_cheapest_total);
2065 : Assert(matpath->parallel_safe);
2066 : }
2067 :
2068 27242 : foreach(lc1, outerrel->partial_pathlist)
2069 : {
2070 15700 : Path *outerpath = (Path *) lfirst(lc1);
2071 : List *pathkeys;
2072 : ListCell *lc2;
2073 :
2074 : /* Figure out what useful ordering any paths we create will have. */
2075 15700 : pathkeys = build_join_pathkeys(root, joinrel, jointype,
2076 : outerpath->pathkeys);
2077 :
2078 : /*
2079 : * Try the cheapest parameterized paths; only those which will produce
2080 : * an unparameterized path when joined to this outerrel will survive
2081 : * try_partial_nestloop_path. The cheapest unparameterized path is
2082 : * also in this list.
2083 : */
2084 40454 : foreach(lc2, innerrel->cheapest_parameterized_paths)
2085 : {
2086 24754 : Path *innerpath = (Path *) lfirst(lc2);
2087 : Path *mpath;
2088 :
2089 : /* Can't join to an inner path that is not parallel-safe */
2090 24754 : if (!innerpath->parallel_safe)
2091 450 : continue;
2092 :
2093 24304 : try_partial_nestloop_path(root, joinrel, outerpath, innerpath,
2094 : pathkeys, jointype, extra);
2095 :
2096 : /*
2097 : * Try generating a memoize path and see if that makes the nested
2098 : * loop any cheaper.
2099 : */
2100 24304 : mpath = get_memoize_path(root, innerrel, outerrel,
2101 : innerpath, outerpath, jointype,
2102 : extra);
2103 24304 : if (mpath != NULL)
2104 6124 : try_partial_nestloop_path(root, joinrel, outerpath, mpath,
2105 : pathkeys, jointype, extra);
2106 : }
2107 :
2108 : /* Also consider materialized form of the cheapest inner path */
2109 15700 : if (matpath != NULL)
2110 14622 : try_partial_nestloop_path(root, joinrel, outerpath, matpath,
2111 : pathkeys, jointype, extra);
2112 : }
2113 11542 : }
2114 :
2115 : /*
2116 : * hash_inner_and_outer
2117 : * Create hashjoin join paths by explicitly hashing both the outer and
2118 : * inner keys of each available hash clause.
2119 : *
2120 : * 'joinrel' is the join relation
2121 : * 'outerrel' is the outer join relation
2122 : * 'innerrel' is the inner join relation
2123 : * 'jointype' is the type of join to do
2124 : * 'extra' contains additional input values
2125 : */
2126 : static void
2127 639088 : hash_inner_and_outer(PlannerInfo *root,
2128 : RelOptInfo *joinrel,
2129 : RelOptInfo *outerrel,
2130 : RelOptInfo *innerrel,
2131 : JoinType jointype,
2132 : JoinPathExtraData *extra)
2133 : {
2134 639088 : bool isouterjoin = IS_OUTER_JOIN(jointype);
2135 : List *hashclauses;
2136 : ListCell *l;
2137 :
2138 : /*
2139 : * We need to build only one hashclauses list for any given pair of outer
2140 : * and inner relations; all of the hashable clauses will be used as keys.
2141 : *
2142 : * Scan the join's restrictinfo list to find hashjoinable clauses that are
2143 : * usable with this pair of sub-relations.
2144 : */
2145 639088 : hashclauses = NIL;
2146 1326676 : foreach(l, extra->restrictlist)
2147 : {
2148 687588 : RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(l);
2149 :
2150 : /*
2151 : * If processing an outer join, only use its own join clauses for
2152 : * hashing. For inner joins we need not be so picky.
2153 : */
2154 687588 : if (isouterjoin && RINFO_IS_PUSHED_DOWN(restrictinfo, joinrel->relids))
2155 10212 : continue;
2156 :
2157 677376 : if (!restrictinfo->can_join ||
2158 602028 : restrictinfo->hashjoinoperator == InvalidOid)
2159 93316 : continue; /* not hashjoinable */
2160 :
2161 : /*
2162 : * Check if clause has the form "outer op inner" or "inner op outer".
2163 : */
2164 584060 : if (!clause_sides_match_join(restrictinfo, outerrel->relids,
2165 : innerrel->relids))
2166 216 : continue; /* no good for these input relations */
2167 :
2168 : /*
2169 : * If clause has the form "inner op outer", check if its operator has
2170 : * valid commutator. This is necessary because hashclauses in this
2171 : * form will get commuted in createplan.c to put the outer var on the
2172 : * left (see get_switched_clauses). This probably shouldn't ever
2173 : * fail, since hashable operators ought to have commutators, but be
2174 : * paranoid.
2175 : *
2176 : * The clause being hashjoinable indicates that it's an OpExpr.
2177 : */
2178 875766 : if (!restrictinfo->outer_is_left &&
2179 291922 : !OidIsValid(get_commutator(castNode(OpExpr, restrictinfo->clause)->opno)))
2180 6 : continue;
2181 :
2182 583838 : hashclauses = lappend(hashclauses, restrictinfo);
2183 : }
2184 :
2185 : /* If we found any usable hashclauses, make paths */
2186 639088 : if (hashclauses)
2187 : {
2188 : /*
2189 : * We consider both the cheapest-total-cost and cheapest-startup-cost
2190 : * outer paths. There's no need to consider any but the
2191 : * cheapest-total-cost inner path, however.
2192 : */
2193 524622 : Path *cheapest_startup_outer = outerrel->cheapest_startup_path;
2194 524622 : Path *cheapest_total_outer = outerrel->cheapest_total_path;
2195 524622 : Path *cheapest_total_inner = innerrel->cheapest_total_path;
2196 : ListCell *lc1;
2197 : ListCell *lc2;
2198 :
2199 : /*
2200 : * If either cheapest-total path is parameterized by the other rel, we
2201 : * can't use a hashjoin. (There's no use looking for alternative
2202 : * input paths, since these should already be the least-parameterized
2203 : * available paths.)
2204 : */
2205 524622 : if (PATH_PARAM_BY_REL(cheapest_total_outer, innerrel) ||
2206 523718 : PATH_PARAM_BY_REL(cheapest_total_inner, outerrel))
2207 1808 : return;
2208 :
2209 : /*
2210 : * Consider the cheapest startup outer together with the cheapest
2211 : * total inner, and then consider pairings of cheapest-total paths
2212 : * including parameterized ones. There is no use in generating
2213 : * parameterized paths on the basis of possibly cheap startup cost, so
2214 : * this is sufficient.
2215 : */
2216 522814 : if (cheapest_startup_outer != NULL)
2217 522100 : try_hashjoin_path(root,
2218 : joinrel,
2219 : cheapest_startup_outer,
2220 : cheapest_total_inner,
2221 : hashclauses,
2222 : jointype,
2223 : extra);
2224 :
2225 1331928 : foreach(lc1, outerrel->cheapest_parameterized_paths)
2226 : {
2227 809114 : Path *outerpath = (Path *) lfirst(lc1);
2228 :
2229 : /*
2230 : * We cannot use an outer path that is parameterized by the inner
2231 : * rel.
2232 : */
2233 809114 : if (PATH_PARAM_BY_REL(outerpath, innerrel))
2234 233366 : continue;
2235 :
2236 1482932 : foreach(lc2, innerrel->cheapest_parameterized_paths)
2237 : {
2238 907184 : Path *innerpath = (Path *) lfirst(lc2);
2239 :
2240 : /*
2241 : * We cannot use an inner path that is parameterized by the
2242 : * outer rel, either.
2243 : */
2244 907184 : if (PATH_PARAM_BY_REL(innerpath, outerrel))
2245 264390 : continue;
2246 :
2247 642794 : if (outerpath == cheapest_startup_outer &&
2248 : innerpath == cheapest_total_inner)
2249 440110 : continue; /* already tried it */
2250 :
2251 202684 : try_hashjoin_path(root,
2252 : joinrel,
2253 : outerpath,
2254 : innerpath,
2255 : hashclauses,
2256 : jointype,
2257 : extra);
2258 : }
2259 : }
2260 :
2261 : /*
2262 : * If the joinrel is parallel-safe, we may be able to consider a
2263 : * partial hash join. However, the resulting path must not be
2264 : * parameterized.
2265 : */
2266 522814 : if (joinrel->consider_parallel &&
2267 464328 : outerrel->partial_pathlist != NIL &&
2268 17358 : bms_is_empty(joinrel->lateral_relids))
2269 : {
2270 : Path *cheapest_partial_outer;
2271 17358 : Path *cheapest_partial_inner = NULL;
2272 17358 : Path *cheapest_safe_inner = NULL;
2273 :
2274 17358 : cheapest_partial_outer =
2275 17358 : (Path *) linitial(outerrel->partial_pathlist);
2276 :
2277 : /*
2278 : * Can we use a partial inner plan too, so that we can build a
2279 : * shared hash table in parallel?
2280 : */
2281 17358 : if (innerrel->partial_pathlist != NIL &&
2282 : enable_parallel_hash)
2283 : {
2284 12744 : cheapest_partial_inner =
2285 12744 : (Path *) linitial(innerrel->partial_pathlist);
2286 12744 : try_partial_hashjoin_path(root, joinrel,
2287 : cheapest_partial_outer,
2288 : cheapest_partial_inner,
2289 : hashclauses, jointype, extra,
2290 : true /* parallel_hash */ );
2291 : }
2292 :
2293 : /*
2294 : * Normally, given that the joinrel is parallel-safe, the cheapest
2295 : * total inner path will also be parallel-safe, but if not, we'll
2296 : * have to search for the cheapest safe, unparameterized inner
2297 : * path. If full, right, right-semi or right-anti join, we can't
2298 : * use parallelism (building the hash table in each backend)
2299 : * because no one process has all the match bits.
2300 : */
2301 17358 : if (jointype == JOIN_FULL ||
2302 13542 : jointype == JOIN_RIGHT ||
2303 9956 : jointype == JOIN_RIGHT_SEMI ||
2304 : jointype == JOIN_RIGHT_ANTI)
2305 7732 : cheapest_safe_inner = NULL;
2306 9626 : else if (cheapest_total_inner->parallel_safe)
2307 9374 : cheapest_safe_inner = cheapest_total_inner;
2308 : else
2309 : cheapest_safe_inner =
2310 252 : get_cheapest_parallel_safe_total_inner(innerrel->pathlist);
2311 :
2312 17358 : if (cheapest_safe_inner != NULL)
2313 9602 : try_partial_hashjoin_path(root, joinrel,
2314 : cheapest_partial_outer,
2315 : cheapest_safe_inner,
2316 : hashclauses, jointype, extra,
2317 : false /* parallel_hash */ );
2318 : }
2319 : }
2320 : }
2321 :
2322 : /*
2323 : * select_mergejoin_clauses
2324 : * Select mergejoin clauses that are usable for a particular join.
2325 : * Returns a list of RestrictInfo nodes for those clauses.
2326 : *
2327 : * *mergejoin_allowed is normally set to true, but it is set to false if
2328 : * this is a right-semi join, or this is a right/right-anti/full join and
2329 : * there are nonmergejoinable join clauses. The executor's mergejoin
2330 : * machinery cannot handle such cases, so we have to avoid generating a
2331 : * mergejoin plan. (Note that this flag does NOT consider whether there are
2332 : * actually any mergejoinable clauses. This is correct because in some
2333 : * cases we need to build a clauseless mergejoin. Simply returning NIL is
2334 : * therefore not enough to distinguish safe from unsafe cases.)
2335 : *
2336 : * We also mark each selected RestrictInfo to show which side is currently
2337 : * being considered as outer. These are transient markings that are only
2338 : * good for the duration of the current add_paths_to_joinrel() call!
2339 : *
2340 : * We examine each restrictinfo clause known for the join to see
2341 : * if it is mergejoinable and involves vars from the two sub-relations
2342 : * currently of interest.
2343 : */
2344 : static List *
2345 640824 : select_mergejoin_clauses(PlannerInfo *root,
2346 : RelOptInfo *joinrel,
2347 : RelOptInfo *outerrel,
2348 : RelOptInfo *innerrel,
2349 : List *restrictlist,
2350 : JoinType jointype,
2351 : bool *mergejoin_allowed)
2352 : {
2353 640824 : List *result_list = NIL;
2354 640824 : bool isouterjoin = IS_OUTER_JOIN(jointype);
2355 640824 : bool have_nonmergeable_joinclause = false;
2356 : ListCell *l;
2357 :
2358 : /*
2359 : * For now we do not support RIGHT_SEMI join in mergejoin: the benefit of
2360 : * swapping inputs tends to be small here.
2361 : */
2362 640824 : if (jointype == JOIN_RIGHT_SEMI)
2363 : {
2364 7412 : *mergejoin_allowed = false;
2365 7412 : return NIL;
2366 : }
2367 :
2368 1316604 : foreach(l, restrictlist)
2369 : {
2370 683192 : RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(l);
2371 :
2372 : /*
2373 : * If processing an outer join, only use its own join clauses in the
2374 : * merge. For inner joins we can use pushed-down clauses too. (Note:
2375 : * we don't set have_nonmergeable_joinclause here because pushed-down
2376 : * clauses will become otherquals not joinquals.)
2377 : */
2378 683192 : if (isouterjoin && RINFO_IS_PUSHED_DOWN(restrictinfo, joinrel->relids))
2379 10236 : continue;
2380 :
2381 : /* Check that clause is a mergeable operator clause */
2382 672956 : if (!restrictinfo->can_join ||
2383 597626 : restrictinfo->mergeopfamilies == NIL)
2384 : {
2385 : /*
2386 : * The executor can handle extra joinquals that are constants, but
2387 : * not anything else, when doing right/right-anti/full merge join.
2388 : * (The reason to support constants is so we can do FULL JOIN ON
2389 : * FALSE.)
2390 : */
2391 91832 : if (!restrictinfo->clause || !IsA(restrictinfo->clause, Const))
2392 81692 : have_nonmergeable_joinclause = true;
2393 91832 : continue; /* not mergejoinable */
2394 : }
2395 :
2396 : /*
2397 : * Check if clause has the form "outer op inner" or "inner op outer".
2398 : */
2399 581124 : if (!clause_sides_match_join(restrictinfo, outerrel->relids,
2400 : innerrel->relids))
2401 : {
2402 1008 : have_nonmergeable_joinclause = true;
2403 1008 : continue; /* no good for these input relations */
2404 : }
2405 :
2406 : /*
2407 : * If clause has the form "inner op outer", check if its operator has
2408 : * valid commutator. This is necessary because mergejoin clauses in
2409 : * this form will get commuted in createplan.c to put the outer var on
2410 : * the left (see get_switched_clauses). This probably shouldn't ever
2411 : * fail, since mergejoinable operators ought to have commutators, but
2412 : * be paranoid.
2413 : *
2414 : * The clause being mergejoinable indicates that it's an OpExpr.
2415 : */
2416 867296 : if (!restrictinfo->outer_is_left &&
2417 287180 : !OidIsValid(get_commutator(castNode(OpExpr, restrictinfo->clause)->opno)))
2418 : {
2419 36 : have_nonmergeable_joinclause = true;
2420 36 : continue;
2421 : }
2422 :
2423 : /*
2424 : * Insist that each side have a non-redundant eclass. This
2425 : * restriction is needed because various bits of the planner expect
2426 : * that each clause in a merge be associable with some pathkey in a
2427 : * canonical pathkey list, but redundant eclasses can't appear in
2428 : * canonical sort orderings. (XXX it might be worth relaxing this,
2429 : * but not enough time to address it for 8.3.)
2430 : */
2431 580080 : update_mergeclause_eclasses(root, restrictinfo);
2432 :
2433 580080 : if (EC_MUST_BE_REDUNDANT(restrictinfo->left_ec) ||
2434 580032 : EC_MUST_BE_REDUNDANT(restrictinfo->right_ec))
2435 : {
2436 128 : have_nonmergeable_joinclause = true;
2437 128 : continue; /* can't handle redundant eclasses */
2438 : }
2439 :
2440 579952 : result_list = lappend(result_list, restrictinfo);
2441 : }
2442 :
2443 : /*
2444 : * Report whether mergejoin is allowed (see comment at top of function).
2445 : */
2446 633412 : switch (jointype)
2447 : {
2448 102826 : case JOIN_RIGHT:
2449 : case JOIN_RIGHT_ANTI:
2450 : case JOIN_FULL:
2451 102826 : *mergejoin_allowed = !have_nonmergeable_joinclause;
2452 102826 : break;
2453 530586 : default:
2454 530586 : *mergejoin_allowed = true;
2455 530586 : break;
2456 : }
2457 :
2458 633412 : return result_list;
2459 : }
|
