Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pathnode.c
4 : * Routines to manipulate pathlists and create path nodes
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/util/pathnode.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include "access/htup_details.h"
18 : #include "executor/nodeSetOp.h"
19 : #include "foreign/fdwapi.h"
20 : #include "miscadmin.h"
21 : #include "nodes/extensible.h"
22 : #include "optimizer/appendinfo.h"
23 : #include "optimizer/clauses.h"
24 : #include "optimizer/cost.h"
25 : #include "optimizer/optimizer.h"
26 : #include "optimizer/pathnode.h"
27 : #include "optimizer/paths.h"
28 : #include "optimizer/planmain.h"
29 : #include "optimizer/tlist.h"
30 : #include "parser/parsetree.h"
31 : #include "utils/memutils.h"
32 : #include "utils/selfuncs.h"
33 :
34 : typedef enum
35 : {
36 : COSTS_EQUAL, /* path costs are fuzzily equal */
37 : COSTS_BETTER1, /* first path is cheaper than second */
38 : COSTS_BETTER2, /* second path is cheaper than first */
39 : COSTS_DIFFERENT, /* neither path dominates the other on cost */
40 : } PathCostComparison;
41 :
42 : /*
43 : * STD_FUZZ_FACTOR is the normal fuzz factor for compare_path_costs_fuzzily.
44 : * XXX is it worth making this user-controllable? It provides a tradeoff
45 : * between planner runtime and the accuracy of path cost comparisons.
46 : */
47 : #define STD_FUZZ_FACTOR 1.01
48 :
49 : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
50 : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
51 : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
52 : List *pathlist,
53 : RelOptInfo *child_rel);
54 : static bool pathlist_is_reparameterizable_by_child(List *pathlist,
55 : RelOptInfo *child_rel);
56 :
57 :
58 : /*****************************************************************************
59 : * MISC. PATH UTILITIES
60 : *****************************************************************************/
61 :
62 : /*
63 : * compare_path_costs
64 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
65 : * or more expensive than path2 for the specified criterion.
66 : */
67 : int
68 582625 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
69 : {
70 : /* Number of disabled nodes, if different, trumps all else. */
71 582625 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
72 : {
73 1356 : if (path1->disabled_nodes < path2->disabled_nodes)
74 1356 : return -1;
75 : else
76 0 : return +1;
77 : }
78 :
79 581269 : if (criterion == STARTUP_COST)
80 : {
81 292936 : if (path1->startup_cost < path2->startup_cost)
82 177942 : return -1;
83 114994 : if (path1->startup_cost > path2->startup_cost)
84 56735 : return +1;
85 :
86 : /*
87 : * If paths have the same startup cost (not at all unlikely), order
88 : * them by total cost.
89 : */
90 58259 : if (path1->total_cost < path2->total_cost)
91 29720 : return -1;
92 28539 : if (path1->total_cost > path2->total_cost)
93 2643 : return +1;
94 : }
95 : else
96 : {
97 288333 : if (path1->total_cost < path2->total_cost)
98 267689 : return -1;
99 20644 : if (path1->total_cost > path2->total_cost)
100 5470 : return +1;
101 :
102 : /*
103 : * If paths have the same total cost, order them by startup cost.
104 : */
105 15174 : if (path1->startup_cost < path2->startup_cost)
106 1341 : return -1;
107 13833 : if (path1->startup_cost > path2->startup_cost)
108 15 : return +1;
109 : }
110 39714 : return 0;
111 : }
112 :
113 : /*
114 : * compare_fractional_path_costs
115 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
116 : * or more expensive than path2 for fetching the specified fraction
117 : * of the total tuples.
118 : *
119 : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
120 : * path with the cheaper total_cost.
121 : */
122 : int
123 2990 : compare_fractional_path_costs(Path *path1, Path *path2,
124 : double fraction)
125 : {
126 : Cost cost1,
127 : cost2;
128 :
129 : /* Number of disabled nodes, if different, trumps all else. */
130 2990 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
131 : {
132 18 : if (path1->disabled_nodes < path2->disabled_nodes)
133 18 : return -1;
134 : else
135 0 : return +1;
136 : }
137 :
138 2972 : if (fraction <= 0.0 || fraction >= 1.0)
139 878 : return compare_path_costs(path1, path2, TOTAL_COST);
140 2094 : cost1 = path1->startup_cost +
141 2094 : fraction * (path1->total_cost - path1->startup_cost);
142 2094 : cost2 = path2->startup_cost +
143 2094 : fraction * (path2->total_cost - path2->startup_cost);
144 2094 : if (cost1 < cost2)
145 1757 : return -1;
146 337 : if (cost1 > cost2)
147 337 : return +1;
148 0 : return 0;
149 : }
150 :
151 : /*
152 : * compare_path_costs_fuzzily
153 : * Compare the costs of two paths to see if either can be said to
154 : * dominate the other.
155 : *
156 : * We use fuzzy comparisons so that add_path() can avoid keeping both of
157 : * a pair of paths that really have insignificantly different cost.
158 : *
159 : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
160 : * fraction of the smaller cost that is considered to be a significant
161 : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
162 : * be 1% of the smaller cost.
163 : *
164 : * The two paths are said to have "equal" costs if both startup and total
165 : * costs are fuzzily the same. Path1 is said to be better than path2 if
166 : * it has fuzzily better startup cost and fuzzily no worse total cost,
167 : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
168 : * Path2 is better than path1 if the reverse holds. Finally, if one path
169 : * is fuzzily better than the other on startup cost and fuzzily worse on
170 : * total cost, we just say that their costs are "different", since neither
171 : * dominates the other across the whole performance spectrum.
172 : *
173 : * This function also enforces a policy rule that paths for which the relevant
174 : * one of parent->consider_startup and parent->consider_param_startup is false
175 : * cannot survive comparisons solely on the grounds of good startup cost, so
176 : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
177 : * (But if total costs are fuzzily equal, we compare startup costs anyway,
178 : * in hopes of eliminating one path or the other.)
179 : */
180 : static PathCostComparison
181 2395421 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
182 : {
183 : #define CONSIDER_PATH_STARTUP_COST(p) \
184 : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
185 :
186 : /* Number of disabled nodes, if different, trumps all else. */
187 2395421 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
188 : {
189 18604 : if (path1->disabled_nodes < path2->disabled_nodes)
190 11606 : return COSTS_BETTER1;
191 : else
192 6998 : return COSTS_BETTER2;
193 : }
194 :
195 : /*
196 : * Check total cost first since it's more likely to be different; many
197 : * paths have zero startup cost.
198 : */
199 2376817 : if (path1->total_cost > path2->total_cost * fuzz_factor)
200 : {
201 : /* path1 fuzzily worse on total cost */
202 1243663 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
203 68716 : path2->startup_cost > path1->startup_cost * fuzz_factor)
204 : {
205 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
206 45974 : return COSTS_DIFFERENT;
207 : }
208 : /* else path2 dominates */
209 1197689 : return COSTS_BETTER2;
210 : }
211 1133154 : if (path2->total_cost > path1->total_cost * fuzz_factor)
212 : {
213 : /* path2 fuzzily worse on total cost */
214 583000 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
215 30410 : path1->startup_cost > path2->startup_cost * fuzz_factor)
216 : {
217 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
218 20035 : return COSTS_DIFFERENT;
219 : }
220 : /* else path1 dominates */
221 562965 : return COSTS_BETTER1;
222 : }
223 : /* fuzzily the same on total cost ... */
224 550154 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
225 : {
226 : /* ... but path1 fuzzily worse on startup, so path2 wins */
227 199613 : return COSTS_BETTER2;
228 : }
229 350541 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
230 : {
231 : /* ... but path2 fuzzily worse on startup, so path1 wins */
232 34709 : return COSTS_BETTER1;
233 : }
234 : /* fuzzily the same on both costs */
235 315832 : return COSTS_EQUAL;
236 :
237 : #undef CONSIDER_PATH_STARTUP_COST
238 : }
239 :
240 : /*
241 : * set_cheapest
242 : * Find the minimum-cost paths from among a relation's paths,
243 : * and save them in the rel's cheapest-path fields.
244 : *
245 : * cheapest_total_path is normally the cheapest-total-cost unparameterized
246 : * path; but if there are no unparameterized paths, we assign it to be the
247 : * best (cheapest least-parameterized) parameterized path. However, only
248 : * unparameterized paths are considered candidates for cheapest_startup_path,
249 : * so that will be NULL if there are no unparameterized paths.
250 : *
251 : * The cheapest_parameterized_paths list collects all parameterized paths
252 : * that have survived the add_path() tournament for this relation. (Since
253 : * add_path ignores pathkeys for a parameterized path, these will be paths
254 : * that have best cost or best row count for their parameterization. We
255 : * may also have both a parallel-safe and a non-parallel-safe path in some
256 : * cases for the same parameterization in some cases, but this should be
257 : * relatively rare since, most typically, all paths for the same relation
258 : * will be parallel-safe or none of them will.)
259 : *
260 : * cheapest_parameterized_paths always includes the cheapest-total
261 : * unparameterized path, too, if there is one; the users of that list find
262 : * it more convenient if that's included.
263 : *
264 : * This is normally called only after we've finished constructing the path
265 : * list for the rel node.
266 : */
267 : void
268 1100196 : set_cheapest(RelOptInfo *parent_rel)
269 : {
270 : Path *cheapest_startup_path;
271 : Path *cheapest_total_path;
272 : Path *best_param_path;
273 : List *parameterized_paths;
274 : ListCell *p;
275 :
276 : Assert(IsA(parent_rel, RelOptInfo));
277 :
278 1100196 : if (parent_rel->pathlist == NIL)
279 0 : elog(ERROR, "could not devise a query plan for the given query");
280 :
281 1100196 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
282 1100196 : parameterized_paths = NIL;
283 :
284 2497027 : foreach(p, parent_rel->pathlist)
285 : {
286 1396831 : Path *path = (Path *) lfirst(p);
287 : int cmp;
288 :
289 1396831 : if (path->param_info)
290 : {
291 : /* Parameterized path, so add it to parameterized_paths */
292 73484 : parameterized_paths = lappend(parameterized_paths, path);
293 :
294 : /*
295 : * If we have an unparameterized cheapest-total, we no longer care
296 : * about finding the best parameterized path, so move on.
297 : */
298 73484 : if (cheapest_total_path)
299 15111 : continue;
300 :
301 : /*
302 : * Otherwise, track the best parameterized path, which is the one
303 : * with least total cost among those of the minimum
304 : * parameterization.
305 : */
306 58373 : if (best_param_path == NULL)
307 53192 : best_param_path = path;
308 : else
309 : {
310 5181 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
311 5181 : PATH_REQ_OUTER(best_param_path)))
312 : {
313 30 : case BMS_EQUAL:
314 : /* keep the cheaper one */
315 30 : if (compare_path_costs(path, best_param_path,
316 : TOTAL_COST) < 0)
317 0 : best_param_path = path;
318 30 : break;
319 387 : case BMS_SUBSET1:
320 : /* new path is less-parameterized */
321 387 : best_param_path = path;
322 387 : break;
323 3 : case BMS_SUBSET2:
324 : /* old path is less-parameterized, keep it */
325 3 : break;
326 4761 : case BMS_DIFFERENT:
327 :
328 : /*
329 : * This means that neither path has the least possible
330 : * parameterization for the rel. We'll sit on the old
331 : * path until something better comes along.
332 : */
333 4761 : break;
334 : }
335 : }
336 : }
337 : else
338 : {
339 : /* Unparameterized path, so consider it for cheapest slots */
340 1323347 : if (cheapest_total_path == NULL)
341 : {
342 1093639 : cheapest_startup_path = cheapest_total_path = path;
343 1093639 : continue;
344 : }
345 :
346 : /*
347 : * If we find two paths of identical costs, try to keep the
348 : * better-sorted one. The paths might have unrelated sort
349 : * orderings, in which case we can only guess which might be
350 : * better to keep, but if one is superior then we definitely
351 : * should keep that one.
352 : */
353 229708 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
354 229708 : if (cmp > 0 ||
355 195 : (cmp == 0 &&
356 195 : compare_pathkeys(cheapest_startup_path->pathkeys,
357 : path->pathkeys) == PATHKEYS_BETTER2))
358 41787 : cheapest_startup_path = path;
359 :
360 229708 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
361 229708 : if (cmp > 0 ||
362 24 : (cmp == 0 &&
363 24 : compare_pathkeys(cheapest_total_path->pathkeys,
364 : path->pathkeys) == PATHKEYS_BETTER2))
365 0 : cheapest_total_path = path;
366 : }
367 : }
368 :
369 : /* Add cheapest unparameterized path, if any, to parameterized_paths */
370 1100196 : if (cheapest_total_path)
371 1093639 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
372 :
373 : /*
374 : * If there is no unparameterized path, use the best parameterized path as
375 : * cheapest_total_path (but not as cheapest_startup_path).
376 : */
377 1100196 : if (cheapest_total_path == NULL)
378 6557 : cheapest_total_path = best_param_path;
379 : Assert(cheapest_total_path != NULL);
380 :
381 1100196 : parent_rel->cheapest_startup_path = cheapest_startup_path;
382 1100196 : parent_rel->cheapest_total_path = cheapest_total_path;
383 1100196 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
384 1100196 : }
385 :
386 : /*
387 : * add_path
388 : * Consider a potential implementation path for the specified parent rel,
389 : * and add it to the rel's pathlist if it is worthy of consideration.
390 : *
391 : * A path is worthy if it has a better sort order (better pathkeys) or
392 : * cheaper cost (as defined below), or generates fewer rows, than any
393 : * existing path that has the same or superset parameterization rels. We
394 : * also consider parallel-safe paths more worthy than others.
395 : *
396 : * Cheaper cost can mean either a cheaper total cost or a cheaper startup
397 : * cost; if one path is cheaper in one of these aspects and another is
398 : * cheaper in the other, we keep both. However, when some path type is
399 : * disabled (e.g. due to enable_seqscan=false), the number of times that
400 : * a disabled path type is used is considered to be a higher-order
401 : * component of the cost. Hence, if path A uses no disabled path type,
402 : * and path B uses 1 or more disabled path types, A is cheaper, no matter
403 : * what we estimate for the startup and total costs. The startup and total
404 : * cost essentially act as a tiebreak when comparing paths that use equal
405 : * numbers of disabled path nodes; but in practice this tiebreak is almost
406 : * always used, since normally no path types are disabled.
407 : *
408 : * In addition to possibly adding new_path, we also remove from the rel's
409 : * pathlist any old paths that are dominated by new_path --- that is,
410 : * new_path is cheaper, at least as well ordered, generates no more rows,
411 : * requires no outer rels not required by the old path, and is no less
412 : * parallel-safe.
413 : *
414 : * In most cases, a path with a superset parameterization will generate
415 : * fewer rows (since it has more join clauses to apply), so that those two
416 : * figures of merit move in opposite directions; this means that a path of
417 : * one parameterization can seldom dominate a path of another. But such
418 : * cases do arise, so we make the full set of checks anyway.
419 : *
420 : * There are two policy decisions embedded in this function, along with
421 : * its sibling add_path_precheck. First, we treat all parameterized paths
422 : * as having NIL pathkeys, so that they cannot win comparisons on the
423 : * basis of sort order. This is to reduce the number of parameterized
424 : * paths that are kept; see discussion in src/backend/optimizer/README.
425 : *
426 : * Second, we only consider cheap startup cost to be interesting if
427 : * parent_rel->consider_startup is true for an unparameterized path, or
428 : * parent_rel->consider_param_startup is true for a parameterized one.
429 : * Again, this allows discarding useless paths sooner.
430 : *
431 : * The pathlist is kept sorted by disabled_nodes and then by total_cost,
432 : * with cheaper paths at the front. Within this routine, that's simply a
433 : * speed hack: doing it that way makes it more likely that we will reject
434 : * an inferior path after a few comparisons, rather than many comparisons.
435 : * However, add_path_precheck relies on this ordering to exit early
436 : * when possible.
437 : *
438 : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
439 : * memory consumption. We dare not try to free the substructure of a Path,
440 : * since much of it may be shared with other Paths or the query tree itself;
441 : * but just recycling discarded Path nodes is a very useful savings in
442 : * a large join tree. We can recycle the List nodes of pathlist, too.
443 : *
444 : * As noted in optimizer/README, deleting a previously-accepted Path is
445 : * safe because we know that Paths of this rel cannot yet be referenced
446 : * from any other rel, such as a higher-level join. However, in some cases
447 : * it is possible that a Path is referenced by another Path for its own
448 : * rel; we must not delete such a Path, even if it is dominated by the new
449 : * Path. Currently this occurs only for IndexPath objects, which may be
450 : * referenced as children of BitmapHeapPaths as well as being paths in
451 : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
452 : *
453 : * 'parent_rel' is the relation entry to which the path corresponds.
454 : * 'new_path' is a potential path for parent_rel.
455 : *
456 : * Returns nothing, but modifies parent_rel->pathlist.
457 : */
458 : void
459 2398374 : add_path(RelOptInfo *parent_rel, Path *new_path)
460 : {
461 2398374 : bool accept_new = true; /* unless we find a superior old path */
462 2398374 : int insert_at = 0; /* where to insert new item */
463 : List *new_path_pathkeys;
464 : ListCell *p1;
465 :
466 : /*
467 : * This is a convenient place to check for query cancel --- no part of the
468 : * planner goes very long without calling add_path().
469 : */
470 2398374 : CHECK_FOR_INTERRUPTS();
471 :
472 : /* Pretend parameterized paths have no pathkeys, per comment above */
473 2398374 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
474 :
475 : /*
476 : * Loop to check proposed new path against old paths. Note it is possible
477 : * for more than one old path to be tossed out because new_path dominates
478 : * it.
479 : */
480 3697137 : foreach(p1, parent_rel->pathlist)
481 : {
482 2199964 : Path *old_path = (Path *) lfirst(p1);
483 2199964 : bool remove_old = false; /* unless new proves superior */
484 : PathCostComparison costcmp;
485 : PathKeysComparison keyscmp;
486 : BMS_Comparison outercmp;
487 :
488 : /*
489 : * Do a fuzzy cost comparison with standard fuzziness limit.
490 : */
491 2199964 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
492 : STD_FUZZ_FACTOR);
493 :
494 : /*
495 : * If the two paths compare differently for startup and total cost,
496 : * then we want to keep both, and we can skip comparing pathkeys and
497 : * required_outer rels. If they compare the same, proceed with the
498 : * other comparisons. Row count is checked last. (We make the tests
499 : * in this order because the cost comparison is most likely to turn
500 : * out "different", and the pathkeys comparison next most likely. As
501 : * explained above, row count very seldom makes a difference, so even
502 : * though it's cheap to compare there's not much point in checking it
503 : * earlier.)
504 : */
505 2199964 : if (costcmp != COSTS_DIFFERENT)
506 : {
507 : /* Similarly check to see if either dominates on pathkeys */
508 : List *old_path_pathkeys;
509 :
510 2133988 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
511 2133988 : keyscmp = compare_pathkeys(new_path_pathkeys,
512 : old_path_pathkeys);
513 2133988 : if (keyscmp != PATHKEYS_DIFFERENT)
514 : {
515 2036641 : switch (costcmp)
516 : {
517 213390 : case COSTS_EQUAL:
518 213390 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
519 213390 : PATH_REQ_OUTER(old_path));
520 213390 : if (keyscmp == PATHKEYS_BETTER1)
521 : {
522 4566 : if ((outercmp == BMS_EQUAL ||
523 4566 : outercmp == BMS_SUBSET1) &&
524 4566 : new_path->rows <= old_path->rows &&
525 4562 : new_path->parallel_safe >= old_path->parallel_safe)
526 4562 : remove_old = true; /* new dominates old */
527 : }
528 208824 : else if (keyscmp == PATHKEYS_BETTER2)
529 : {
530 10391 : if ((outercmp == BMS_EQUAL ||
531 10391 : outercmp == BMS_SUBSET2) &&
532 10391 : new_path->rows >= old_path->rows &&
533 10389 : new_path->parallel_safe <= old_path->parallel_safe)
534 10389 : accept_new = false; /* old dominates new */
535 : }
536 : else /* keyscmp == PATHKEYS_EQUAL */
537 : {
538 198433 : if (outercmp == BMS_EQUAL)
539 : {
540 : /*
541 : * Same pathkeys and outer rels, and fuzzily
542 : * the same cost, so keep just one; to decide
543 : * which, first check parallel-safety, then
544 : * rows, then do a fuzzy cost comparison with
545 : * very small fuzz limit. (We used to do an
546 : * exact cost comparison, but that results in
547 : * annoying platform-specific plan variations
548 : * due to roundoff in the cost estimates.) If
549 : * things are still tied, arbitrarily keep
550 : * only the old path. Notice that we will
551 : * keep only the old path even if the
552 : * less-fuzzy comparison decides the startup
553 : * and total costs compare differently.
554 : */
555 195632 : if (new_path->parallel_safe >
556 195632 : old_path->parallel_safe)
557 21 : remove_old = true; /* new dominates old */
558 195611 : else if (new_path->parallel_safe <
559 195611 : old_path->parallel_safe)
560 27 : accept_new = false; /* old dominates new */
561 195584 : else if (new_path->rows < old_path->rows)
562 18 : remove_old = true; /* new dominates old */
563 195566 : else if (new_path->rows > old_path->rows)
564 109 : accept_new = false; /* old dominates new */
565 195457 : else if (compare_path_costs_fuzzily(new_path,
566 : old_path,
567 : 1.0000000001) == COSTS_BETTER1)
568 8440 : remove_old = true; /* new dominates old */
569 : else
570 187017 : accept_new = false; /* old equals or
571 : * dominates new */
572 : }
573 2801 : else if (outercmp == BMS_SUBSET1 &&
574 616 : new_path->rows <= old_path->rows &&
575 606 : new_path->parallel_safe >= old_path->parallel_safe)
576 606 : remove_old = true; /* new dominates old */
577 2195 : else if (outercmp == BMS_SUBSET2 &&
578 1826 : new_path->rows >= old_path->rows &&
579 1705 : new_path->parallel_safe <= old_path->parallel_safe)
580 1705 : accept_new = false; /* old dominates new */
581 : /* else different parameterizations, keep both */
582 : }
583 213390 : break;
584 586973 : case COSTS_BETTER1:
585 586973 : if (keyscmp != PATHKEYS_BETTER2)
586 : {
587 398811 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
588 398811 : PATH_REQ_OUTER(old_path));
589 398811 : if ((outercmp == BMS_EQUAL ||
590 342157 : outercmp == BMS_SUBSET1) &&
591 342157 : new_path->rows <= old_path->rows &&
592 339781 : new_path->parallel_safe >= old_path->parallel_safe)
593 338423 : remove_old = true; /* new dominates old */
594 : }
595 586973 : break;
596 1236278 : case COSTS_BETTER2:
597 1236278 : if (keyscmp != PATHKEYS_BETTER1)
598 : {
599 789108 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
600 789108 : PATH_REQ_OUTER(old_path));
601 789108 : if ((outercmp == BMS_EQUAL ||
602 743977 : outercmp == BMS_SUBSET2) &&
603 743977 : new_path->rows >= old_path->rows &&
604 702873 : new_path->parallel_safe <= old_path->parallel_safe)
605 701954 : accept_new = false; /* old dominates new */
606 : }
607 1236278 : break;
608 0 : case COSTS_DIFFERENT:
609 :
610 : /*
611 : * can't get here, but keep this case to keep compiler
612 : * quiet
613 : */
614 0 : break;
615 : }
616 : }
617 : }
618 :
619 : /*
620 : * Remove current element from pathlist if dominated by new.
621 : */
622 2199964 : if (remove_old)
623 : {
624 352070 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
625 : p1);
626 :
627 : /*
628 : * Delete the data pointed-to by the deleted cell, if possible
629 : */
630 352070 : if (!IsA(old_path, IndexPath))
631 339251 : pfree(old_path);
632 : }
633 : else
634 : {
635 : /*
636 : * new belongs after this old path if it has more disabled nodes
637 : * or if it has the same number of nodes but a greater total cost
638 : */
639 1847894 : if (new_path->disabled_nodes > old_path->disabled_nodes ||
640 1840896 : (new_path->disabled_nodes == old_path->disabled_nodes &&
641 1840411 : new_path->total_cost >= old_path->total_cost))
642 1538783 : insert_at = foreach_current_index(p1) + 1;
643 : }
644 :
645 : /*
646 : * If we found an old path that dominates new_path, we can quit
647 : * scanning the pathlist; we will not add new_path, and we assume
648 : * new_path cannot dominate any other elements of the pathlist.
649 : */
650 2199964 : if (!accept_new)
651 901201 : break;
652 : }
653 :
654 2398374 : if (accept_new)
655 : {
656 : /* Accept the new path: insert it at proper place in pathlist */
657 1497173 : parent_rel->pathlist =
658 1497173 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
659 : }
660 : else
661 : {
662 : /* Reject and recycle the new path */
663 901201 : if (!IsA(new_path, IndexPath))
664 848720 : pfree(new_path);
665 : }
666 2398374 : }
667 :
668 : /*
669 : * add_path_precheck
670 : * Check whether a proposed new path could possibly get accepted.
671 : * We assume we know the path's pathkeys and parameterization accurately,
672 : * and have lower bounds for its costs.
673 : *
674 : * Note that we do not know the path's rowcount, since getting an estimate for
675 : * that is too expensive to do before prechecking. We assume here that paths
676 : * of a superset parameterization will generate fewer rows; if that holds,
677 : * then paths with different parameterizations cannot dominate each other
678 : * and so we can simply ignore existing paths of another parameterization.
679 : * (In the infrequent cases where that rule of thumb fails, add_path will
680 : * get rid of the inferior path.)
681 : *
682 : * At the time this is called, we haven't actually built a Path structure,
683 : * so the required information has to be passed piecemeal.
684 : */
685 : bool
686 2740240 : add_path_precheck(RelOptInfo *parent_rel, int disabled_nodes,
687 : Cost startup_cost, Cost total_cost,
688 : List *pathkeys, Relids required_outer)
689 : {
690 : List *new_path_pathkeys;
691 : bool consider_startup;
692 : ListCell *p1;
693 :
694 : /* Pretend parameterized paths have no pathkeys, per add_path policy */
695 2740240 : new_path_pathkeys = required_outer ? NIL : pathkeys;
696 :
697 : /* Decide whether new path's startup cost is interesting */
698 2740240 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
699 :
700 3492850 : foreach(p1, parent_rel->pathlist)
701 : {
702 3313142 : Path *old_path = (Path *) lfirst(p1);
703 : PathKeysComparison keyscmp;
704 :
705 : /*
706 : * Since the pathlist is sorted by disabled_nodes and then by
707 : * total_cost, we can stop looking once we reach a path with more
708 : * disabled nodes, or the same number of disabled nodes plus a
709 : * total_cost larger than the new path's.
710 : */
711 3313142 : if (unlikely(old_path->disabled_nodes != disabled_nodes))
712 : {
713 4350 : if (disabled_nodes < old_path->disabled_nodes)
714 164 : break;
715 : }
716 3308792 : else if (total_cost <= old_path->total_cost * STD_FUZZ_FACTOR)
717 990826 : break;
718 :
719 : /*
720 : * We are looking for an old_path with the same parameterization (and
721 : * by assumption the same rowcount) that dominates the new path on
722 : * pathkeys as well as both cost metrics. If we find one, we can
723 : * reject the new path.
724 : *
725 : * Cost comparisons here should match compare_path_costs_fuzzily.
726 : */
727 : /* new path can win on startup cost only if consider_startup */
728 2322152 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
729 1096609 : !consider_startup)
730 : {
731 : /* new path loses on cost, so check pathkeys... */
732 : List *old_path_pathkeys;
733 :
734 2270922 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
735 2270922 : keyscmp = compare_pathkeys(new_path_pathkeys,
736 : old_path_pathkeys);
737 2270922 : if (keyscmp == PATHKEYS_EQUAL ||
738 : keyscmp == PATHKEYS_BETTER2)
739 : {
740 : /* new path does not win on pathkeys... */
741 1604486 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
742 : {
743 : /* Found an old path that dominates the new one */
744 1569542 : return false;
745 : }
746 : }
747 : }
748 : }
749 :
750 1170698 : return true;
751 : }
752 :
753 : /*
754 : * add_partial_path
755 : * Like add_path, our goal here is to consider whether a path is worthy
756 : * of being kept around, but the considerations here are a bit different.
757 : * A partial path is one which can be executed in any number of workers in
758 : * parallel such that each worker will generate a subset of the path's
759 : * overall result.
760 : *
761 : * As in add_path, the partial_pathlist is kept sorted with the cheapest
762 : * total path in front. This is depended on by multiple places, which
763 : * just take the front entry as the cheapest path without searching.
764 : *
765 : * We don't generate parameterized partial paths for several reasons. Most
766 : * importantly, they're not safe to execute, because there's nothing to
767 : * make sure that a parallel scan within the parameterized portion of the
768 : * plan is running with the same value in every worker at the same time.
769 : * Fortunately, it seems unlikely to be worthwhile anyway, because having
770 : * each worker scan the entire outer relation and a subset of the inner
771 : * relation will generally be a terrible plan. The inner (parameterized)
772 : * side of the plan will be small anyway. There could be rare cases where
773 : * this wins big - e.g. if join order constraints put a 1-row relation on
774 : * the outer side of the topmost join with a parameterized plan on the inner
775 : * side - but we'll have to be content not to handle such cases until
776 : * somebody builds an executor infrastructure that can cope with them.
777 : *
778 : * Because we don't consider parameterized paths here, we also don't
779 : * need to consider the row counts as a measure of quality: every path will
780 : * produce the same number of rows. Neither do we need to consider startup
781 : * costs: parallelism is only used for plans that will be run to completion.
782 : * Therefore, this routine is much simpler than add_path: it needs to
783 : * consider only disabled nodes, pathkeys and total cost.
784 : *
785 : * As with add_path, we pfree paths that are found to be dominated by
786 : * another partial path; this requires that there be no other references to
787 : * such paths yet. Hence, GatherPaths must not be created for a rel until
788 : * we're done creating all partial paths for it. Unlike add_path, we don't
789 : * take an exception for IndexPaths as partial index paths won't be
790 : * referenced by partial BitmapHeapPaths.
791 : */
792 : void
793 155849 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
794 : {
795 155849 : bool accept_new = true; /* unless we find a superior old path */
796 155849 : int insert_at = 0; /* where to insert new item */
797 : ListCell *p1;
798 :
799 : /* Check for query cancel. */
800 155849 : CHECK_FOR_INTERRUPTS();
801 :
802 : /* Path to be added must be parallel safe. */
803 : Assert(new_path->parallel_safe);
804 :
805 : /* Relation should be OK for parallelism, too. */
806 : Assert(parent_rel->consider_parallel);
807 :
808 : /*
809 : * As in add_path, throw out any paths which are dominated by the new
810 : * path, but throw out the new path if some existing path dominates it.
811 : */
812 216653 : foreach(p1, parent_rel->partial_pathlist)
813 : {
814 127357 : Path *old_path = (Path *) lfirst(p1);
815 127357 : bool remove_old = false; /* unless new proves superior */
816 : PathKeysComparison keyscmp;
817 :
818 : /* Compare pathkeys. */
819 127357 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
820 :
821 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
822 127357 : if (keyscmp != PATHKEYS_DIFFERENT)
823 : {
824 127234 : if (unlikely(new_path->disabled_nodes != old_path->disabled_nodes))
825 : {
826 1096 : if (new_path->disabled_nodes > old_path->disabled_nodes)
827 451 : accept_new = false;
828 : else
829 645 : remove_old = true;
830 : }
831 126138 : else if (new_path->total_cost > old_path->total_cost
832 126138 : * STD_FUZZ_FACTOR)
833 : {
834 : /* New path costs more; keep it only if pathkeys are better. */
835 62131 : if (keyscmp != PATHKEYS_BETTER1)
836 45022 : accept_new = false;
837 : }
838 64007 : else if (old_path->total_cost > new_path->total_cost
839 64007 : * STD_FUZZ_FACTOR)
840 : {
841 : /* Old path costs more; keep it only if pathkeys are better. */
842 42592 : if (keyscmp != PATHKEYS_BETTER2)
843 15002 : remove_old = true;
844 : }
845 21415 : else if (keyscmp == PATHKEYS_BETTER1)
846 : {
847 : /* Costs are about the same, new path has better pathkeys. */
848 12 : remove_old = true;
849 : }
850 21403 : else if (keyscmp == PATHKEYS_BETTER2)
851 : {
852 : /* Costs are about the same, old path has better pathkeys. */
853 1032 : accept_new = false;
854 : }
855 20371 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
856 : {
857 : /* Pathkeys are the same, and the old path costs more. */
858 323 : remove_old = true;
859 : }
860 : else
861 : {
862 : /*
863 : * Pathkeys are the same, and new path isn't materially
864 : * cheaper.
865 : */
866 20048 : accept_new = false;
867 : }
868 : }
869 :
870 : /*
871 : * Remove current element from partial_pathlist if dominated by new.
872 : */
873 127357 : if (remove_old)
874 : {
875 15982 : parent_rel->partial_pathlist =
876 15982 : foreach_delete_current(parent_rel->partial_pathlist, p1);
877 15982 : pfree(old_path);
878 : }
879 : else
880 : {
881 : /* new belongs after this old path if it has cost >= old's */
882 111375 : if (new_path->total_cost >= old_path->total_cost)
883 83187 : insert_at = foreach_current_index(p1) + 1;
884 : }
885 :
886 : /*
887 : * If we found an old path that dominates new_path, we can quit
888 : * scanning the partial_pathlist; we will not add new_path, and we
889 : * assume new_path cannot dominate any later path.
890 : */
891 127357 : if (!accept_new)
892 66553 : break;
893 : }
894 :
895 155849 : if (accept_new)
896 : {
897 : /* Accept the new path: insert it at proper place */
898 89296 : parent_rel->partial_pathlist =
899 89296 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
900 : }
901 : else
902 : {
903 : /* Reject and recycle the new path */
904 66553 : pfree(new_path);
905 : }
906 155849 : }
907 :
908 : /*
909 : * add_partial_path_precheck
910 : * Check whether a proposed new partial path could possibly get accepted.
911 : *
912 : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
913 : * since they don't matter for partial paths (see add_partial_path). But
914 : * we do want to make sure we don't add a partial path if there's already
915 : * a complete path that dominates it, since in that case the proposed path
916 : * is surely a loser.
917 : */
918 : bool
919 224086 : add_partial_path_precheck(RelOptInfo *parent_rel, int disabled_nodes,
920 : Cost total_cost, List *pathkeys)
921 : {
922 : ListCell *p1;
923 :
924 : /*
925 : * Our goal here is twofold. First, we want to find out whether this path
926 : * is clearly inferior to some existing partial path. If so, we want to
927 : * reject it immediately. Second, we want to find out whether this path
928 : * is clearly superior to some existing partial path -- at least, modulo
929 : * final cost computations. If so, we definitely want to consider it.
930 : *
931 : * Unlike add_path(), we always compare pathkeys here. This is because we
932 : * expect partial_pathlist to be very short, and getting a definitive
933 : * answer at this stage avoids the need to call add_path_precheck.
934 : */
935 286026 : foreach(p1, parent_rel->partial_pathlist)
936 : {
937 232851 : Path *old_path = (Path *) lfirst(p1);
938 : PathKeysComparison keyscmp;
939 :
940 232851 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
941 232851 : if (keyscmp != PATHKEYS_DIFFERENT)
942 : {
943 232737 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
944 : keyscmp != PATHKEYS_BETTER1)
945 170911 : return false;
946 118449 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
947 : keyscmp != PATHKEYS_BETTER2)
948 56623 : return true;
949 : }
950 : }
951 :
952 : /*
953 : * This path is neither clearly inferior to an existing partial path nor
954 : * clearly good enough that it might replace one. Compare it to
955 : * non-parallel plans. If it loses even before accounting for the cost of
956 : * the Gather node, we should definitely reject it.
957 : *
958 : * Note that we pass the total_cost to add_path_precheck twice. This is
959 : * because it's never advantageous to consider the startup cost of a
960 : * partial path; the resulting plans, if run in parallel, will be run to
961 : * completion.
962 : */
963 53175 : if (!add_path_precheck(parent_rel, disabled_nodes, total_cost, total_cost,
964 : pathkeys, NULL))
965 1281 : return false;
966 :
967 51894 : return true;
968 : }
969 :
970 :
971 : /*****************************************************************************
972 : * PATH NODE CREATION ROUTINES
973 : *****************************************************************************/
974 :
975 : /*
976 : * create_seqscan_path
977 : * Creates a path corresponding to a sequential scan, returning the
978 : * pathnode.
979 : */
980 : Path *
981 224803 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
982 : Relids required_outer, int parallel_workers)
983 : {
984 224803 : Path *pathnode = makeNode(Path);
985 :
986 224803 : pathnode->pathtype = T_SeqScan;
987 224803 : pathnode->parent = rel;
988 224803 : pathnode->pathtarget = rel->reltarget;
989 224803 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
990 : required_outer);
991 224803 : pathnode->parallel_aware = (parallel_workers > 0);
992 224803 : pathnode->parallel_safe = rel->consider_parallel;
993 224803 : pathnode->parallel_workers = parallel_workers;
994 224803 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
995 :
996 224803 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
997 :
998 224803 : return pathnode;
999 : }
1000 :
1001 : /*
1002 : * create_samplescan_path
1003 : * Creates a path node for a sampled table scan.
1004 : */
1005 : Path *
1006 153 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
1007 : {
1008 153 : Path *pathnode = makeNode(Path);
1009 :
1010 153 : pathnode->pathtype = T_SampleScan;
1011 153 : pathnode->parent = rel;
1012 153 : pathnode->pathtarget = rel->reltarget;
1013 153 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1014 : required_outer);
1015 153 : pathnode->parallel_aware = false;
1016 153 : pathnode->parallel_safe = rel->consider_parallel;
1017 153 : pathnode->parallel_workers = 0;
1018 153 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
1019 :
1020 153 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
1021 :
1022 153 : return pathnode;
1023 : }
1024 :
1025 : /*
1026 : * create_index_path
1027 : * Creates a path node for an index scan.
1028 : *
1029 : * 'index' is a usable index.
1030 : * 'indexclauses' is a list of IndexClause nodes representing clauses
1031 : * to be enforced as qual conditions in the scan.
1032 : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
1033 : * to be used as index ordering operators in the scan.
1034 : * 'indexorderbycols' is an integer list of index column numbers (zero based)
1035 : * the ordering operators can be used with.
1036 : * 'pathkeys' describes the ordering of the path.
1037 : * 'indexscandir' is either ForwardScanDirection or BackwardScanDirection.
1038 : * 'indexonly' is true if an index-only scan is wanted.
1039 : * 'required_outer' is the set of outer relids for a parameterized path.
1040 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1041 : * estimates of caching behavior.
1042 : * 'partial_path' is true if constructing a parallel index scan path.
1043 : *
1044 : * Returns the new path node.
1045 : */
1046 : IndexPath *
1047 415808 : create_index_path(PlannerInfo *root,
1048 : IndexOptInfo *index,
1049 : List *indexclauses,
1050 : List *indexorderbys,
1051 : List *indexorderbycols,
1052 : List *pathkeys,
1053 : ScanDirection indexscandir,
1054 : bool indexonly,
1055 : Relids required_outer,
1056 : double loop_count,
1057 : bool partial_path)
1058 : {
1059 415808 : IndexPath *pathnode = makeNode(IndexPath);
1060 415808 : RelOptInfo *rel = index->rel;
1061 :
1062 415808 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1063 415808 : pathnode->path.parent = rel;
1064 415808 : pathnode->path.pathtarget = rel->reltarget;
1065 415808 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1066 : required_outer);
1067 415808 : pathnode->path.parallel_aware = false;
1068 415808 : pathnode->path.parallel_safe = rel->consider_parallel;
1069 415808 : pathnode->path.parallel_workers = 0;
1070 415808 : pathnode->path.pathkeys = pathkeys;
1071 :
1072 415808 : pathnode->indexinfo = index;
1073 415808 : pathnode->indexclauses = indexclauses;
1074 415808 : pathnode->indexorderbys = indexorderbys;
1075 415808 : pathnode->indexorderbycols = indexorderbycols;
1076 415808 : pathnode->indexscandir = indexscandir;
1077 :
1078 415808 : cost_index(pathnode, root, loop_count, partial_path);
1079 :
1080 415808 : return pathnode;
1081 : }
1082 :
1083 : /*
1084 : * create_bitmap_heap_path
1085 : * Creates a path node for a bitmap scan.
1086 : *
1087 : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1088 : * 'required_outer' is the set of outer relids for a parameterized path.
1089 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1090 : * estimates of caching behavior.
1091 : *
1092 : * loop_count should match the value used when creating the component
1093 : * IndexPaths.
1094 : */
1095 : BitmapHeapPath *
1096 183561 : create_bitmap_heap_path(PlannerInfo *root,
1097 : RelOptInfo *rel,
1098 : Path *bitmapqual,
1099 : Relids required_outer,
1100 : double loop_count,
1101 : int parallel_degree)
1102 : {
1103 183561 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1104 :
1105 183561 : pathnode->path.pathtype = T_BitmapHeapScan;
1106 183561 : pathnode->path.parent = rel;
1107 183561 : pathnode->path.pathtarget = rel->reltarget;
1108 183561 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1109 : required_outer);
1110 183561 : pathnode->path.parallel_aware = (parallel_degree > 0);
1111 183561 : pathnode->path.parallel_safe = rel->consider_parallel;
1112 183561 : pathnode->path.parallel_workers = parallel_degree;
1113 183561 : pathnode->path.pathkeys = NIL; /* always unordered */
1114 :
1115 183561 : pathnode->bitmapqual = bitmapqual;
1116 :
1117 183561 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1118 : pathnode->path.param_info,
1119 : bitmapqual, loop_count);
1120 :
1121 183561 : return pathnode;
1122 : }
1123 :
1124 : /*
1125 : * create_bitmap_and_path
1126 : * Creates a path node representing a BitmapAnd.
1127 : */
1128 : BitmapAndPath *
1129 25908 : create_bitmap_and_path(PlannerInfo *root,
1130 : RelOptInfo *rel,
1131 : List *bitmapquals)
1132 : {
1133 25908 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1134 25908 : Relids required_outer = NULL;
1135 : ListCell *lc;
1136 :
1137 25908 : pathnode->path.pathtype = T_BitmapAnd;
1138 25908 : pathnode->path.parent = rel;
1139 25908 : pathnode->path.pathtarget = rel->reltarget;
1140 :
1141 : /*
1142 : * Identify the required outer rels as the union of what the child paths
1143 : * depend on. (Alternatively, we could insist that the caller pass this
1144 : * in, but it's more convenient and reliable to compute it here.)
1145 : */
1146 77724 : foreach(lc, bitmapquals)
1147 : {
1148 51816 : Path *bitmapqual = (Path *) lfirst(lc);
1149 :
1150 51816 : required_outer = bms_add_members(required_outer,
1151 51816 : PATH_REQ_OUTER(bitmapqual));
1152 : }
1153 25908 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1154 : required_outer);
1155 :
1156 : /*
1157 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1158 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1159 : * set the flag for this path based only on the relation-level flag,
1160 : * without actually iterating over the list of children.
1161 : */
1162 25908 : pathnode->path.parallel_aware = false;
1163 25908 : pathnode->path.parallel_safe = rel->consider_parallel;
1164 25908 : pathnode->path.parallel_workers = 0;
1165 :
1166 25908 : pathnode->path.pathkeys = NIL; /* always unordered */
1167 :
1168 25908 : pathnode->bitmapquals = bitmapquals;
1169 :
1170 : /* this sets bitmapselectivity as well as the regular cost fields: */
1171 25908 : cost_bitmap_and_node(pathnode, root);
1172 :
1173 25908 : return pathnode;
1174 : }
1175 :
1176 : /*
1177 : * create_bitmap_or_path
1178 : * Creates a path node representing a BitmapOr.
1179 : */
1180 : BitmapOrPath *
1181 520 : create_bitmap_or_path(PlannerInfo *root,
1182 : RelOptInfo *rel,
1183 : List *bitmapquals)
1184 : {
1185 520 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1186 520 : Relids required_outer = NULL;
1187 : ListCell *lc;
1188 :
1189 520 : pathnode->path.pathtype = T_BitmapOr;
1190 520 : pathnode->path.parent = rel;
1191 520 : pathnode->path.pathtarget = rel->reltarget;
1192 :
1193 : /*
1194 : * Identify the required outer rels as the union of what the child paths
1195 : * depend on. (Alternatively, we could insist that the caller pass this
1196 : * in, but it's more convenient and reliable to compute it here.)
1197 : */
1198 1461 : foreach(lc, bitmapquals)
1199 : {
1200 941 : Path *bitmapqual = (Path *) lfirst(lc);
1201 :
1202 941 : required_outer = bms_add_members(required_outer,
1203 941 : PATH_REQ_OUTER(bitmapqual));
1204 : }
1205 520 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1206 : required_outer);
1207 :
1208 : /*
1209 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1210 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1211 : * set the flag for this path based only on the relation-level flag,
1212 : * without actually iterating over the list of children.
1213 : */
1214 520 : pathnode->path.parallel_aware = false;
1215 520 : pathnode->path.parallel_safe = rel->consider_parallel;
1216 520 : pathnode->path.parallel_workers = 0;
1217 :
1218 520 : pathnode->path.pathkeys = NIL; /* always unordered */
1219 :
1220 520 : pathnode->bitmapquals = bitmapquals;
1221 :
1222 : /* this sets bitmapselectivity as well as the regular cost fields: */
1223 520 : cost_bitmap_or_node(pathnode, root);
1224 :
1225 520 : return pathnode;
1226 : }
1227 :
1228 : /*
1229 : * create_tidscan_path
1230 : * Creates a path corresponding to a scan by TID, returning the pathnode.
1231 : */
1232 : TidPath *
1233 436 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1234 : Relids required_outer)
1235 : {
1236 436 : TidPath *pathnode = makeNode(TidPath);
1237 :
1238 436 : pathnode->path.pathtype = T_TidScan;
1239 436 : pathnode->path.parent = rel;
1240 436 : pathnode->path.pathtarget = rel->reltarget;
1241 436 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1242 : required_outer);
1243 436 : pathnode->path.parallel_aware = false;
1244 436 : pathnode->path.parallel_safe = rel->consider_parallel;
1245 436 : pathnode->path.parallel_workers = 0;
1246 436 : pathnode->path.pathkeys = NIL; /* always unordered */
1247 :
1248 436 : pathnode->tidquals = tidquals;
1249 :
1250 436 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1251 : pathnode->path.param_info);
1252 :
1253 436 : return pathnode;
1254 : }
1255 :
1256 : /*
1257 : * create_tidrangescan_path
1258 : * Creates a path corresponding to a scan by a range of TIDs, returning
1259 : * the pathnode.
1260 : */
1261 : TidRangePath *
1262 1026 : create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
1263 : List *tidrangequals, Relids required_outer,
1264 : int parallel_workers)
1265 : {
1266 1026 : TidRangePath *pathnode = makeNode(TidRangePath);
1267 :
1268 1026 : pathnode->path.pathtype = T_TidRangeScan;
1269 1026 : pathnode->path.parent = rel;
1270 1026 : pathnode->path.pathtarget = rel->reltarget;
1271 1026 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1272 : required_outer);
1273 1026 : pathnode->path.parallel_aware = (parallel_workers > 0);
1274 1026 : pathnode->path.parallel_safe = rel->consider_parallel;
1275 1026 : pathnode->path.parallel_workers = parallel_workers;
1276 1026 : pathnode->path.pathkeys = NIL; /* always unordered */
1277 :
1278 1026 : pathnode->tidrangequals = tidrangequals;
1279 :
1280 1026 : cost_tidrangescan(&pathnode->path, root, rel, tidrangequals,
1281 : pathnode->path.param_info);
1282 :
1283 1026 : return pathnode;
1284 : }
1285 :
1286 : /*
1287 : * create_append_path
1288 : * Creates a path corresponding to an Append plan, returning the
1289 : * pathnode.
1290 : *
1291 : * Note that we must handle subpaths = NIL, representing a dummy access path.
1292 : * Also, there are callers that pass root = NULL.
1293 : *
1294 : * 'rows', when passed as a non-negative number, will be used to overwrite the
1295 : * returned path's row estimate. Otherwise, the row estimate is calculated
1296 : * by totalling the row estimates from the 'subpaths' list.
1297 : */
1298 : AppendPath *
1299 46678 : create_append_path(PlannerInfo *root,
1300 : RelOptInfo *rel,
1301 : AppendPathInput input,
1302 : List *pathkeys, Relids required_outer,
1303 : int parallel_workers, bool parallel_aware,
1304 : double rows)
1305 : {
1306 46678 : AppendPath *pathnode = makeNode(AppendPath);
1307 : ListCell *l;
1308 :
1309 : Assert(!parallel_aware || parallel_workers > 0);
1310 :
1311 46678 : pathnode->child_append_relid_sets = input.child_append_relid_sets;
1312 46678 : pathnode->path.pathtype = T_Append;
1313 46678 : pathnode->path.parent = rel;
1314 46678 : pathnode->path.pathtarget = rel->reltarget;
1315 :
1316 : /*
1317 : * If this is for a baserel (not a join or non-leaf partition), we prefer
1318 : * to apply get_baserel_parampathinfo to construct a full ParamPathInfo
1319 : * for the path. This supports building a Memoize path atop this path,
1320 : * and if this is a partitioned table the info may be useful for run-time
1321 : * pruning (cf make_partition_pruneinfo()).
1322 : *
1323 : * However, if we don't have "root" then that won't work and we fall back
1324 : * on the simpler get_appendrel_parampathinfo. There's no point in doing
1325 : * the more expensive thing for a dummy path, either.
1326 : */
1327 46678 : if (rel->reloptkind == RELOPT_BASEREL && root && input.subpaths != NIL)
1328 20455 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1329 : rel,
1330 : required_outer);
1331 : else
1332 26223 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1333 : required_outer);
1334 :
1335 46678 : pathnode->path.parallel_aware = parallel_aware;
1336 46678 : pathnode->path.parallel_safe = rel->consider_parallel;
1337 46678 : pathnode->path.parallel_workers = parallel_workers;
1338 46678 : pathnode->path.pathkeys = pathkeys;
1339 :
1340 : /*
1341 : * For parallel append, non-partial paths are sorted by descending total
1342 : * costs. That way, the total time to finish all non-partial paths is
1343 : * minimized. Also, the partial paths are sorted by descending startup
1344 : * costs. There may be some paths that require to do startup work by a
1345 : * single worker. In such case, it's better for workers to choose the
1346 : * expensive ones first, whereas the leader should choose the cheapest
1347 : * startup plan.
1348 : */
1349 46678 : if (pathnode->path.parallel_aware)
1350 : {
1351 : /*
1352 : * We mustn't fiddle with the order of subpaths when the Append has
1353 : * pathkeys. The order they're listed in is critical to keeping the
1354 : * pathkeys valid.
1355 : */
1356 : Assert(pathkeys == NIL);
1357 :
1358 16805 : list_sort(input.subpaths, append_total_cost_compare);
1359 16805 : list_sort(input.partial_subpaths, append_startup_cost_compare);
1360 : }
1361 46678 : pathnode->first_partial_path = list_length(input.subpaths);
1362 46678 : pathnode->subpaths = list_concat(input.subpaths, input.partial_subpaths);
1363 :
1364 : /*
1365 : * Apply query-wide LIMIT if known and path is for sole base relation.
1366 : * (Handling this at this low level is a bit klugy.)
1367 : */
1368 46678 : if (root != NULL && bms_equal(rel->relids, root->all_query_rels))
1369 23645 : pathnode->limit_tuples = root->limit_tuples;
1370 : else
1371 23033 : pathnode->limit_tuples = -1.0;
1372 :
1373 158762 : foreach(l, pathnode->subpaths)
1374 : {
1375 112084 : Path *subpath = (Path *) lfirst(l);
1376 :
1377 203846 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1378 203846 : subpath->parallel_safe;
1379 :
1380 : /* All child paths must have same parameterization */
1381 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1382 : }
1383 :
1384 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1385 :
1386 : /*
1387 : * If there's exactly one child path then the output of the Append is
1388 : * necessarily ordered the same as the child's, so we can inherit the
1389 : * child's pathkeys if any, overriding whatever the caller might've said.
1390 : * Furthermore, if the child's parallel awareness matches the Append's,
1391 : * then the Append is a no-op and will be discarded later (in setrefs.c).
1392 : * Then we can inherit the child's size and cost too, effectively charging
1393 : * zero for the Append. Otherwise, we must do the normal costsize
1394 : * calculation.
1395 : */
1396 46678 : if (list_length(pathnode->subpaths) == 1)
1397 : {
1398 11441 : Path *child = (Path *) linitial(pathnode->subpaths);
1399 :
1400 11441 : if (child->parallel_aware == parallel_aware)
1401 : {
1402 11213 : pathnode->path.rows = child->rows;
1403 11213 : pathnode->path.startup_cost = child->startup_cost;
1404 11213 : pathnode->path.total_cost = child->total_cost;
1405 : }
1406 : else
1407 228 : cost_append(pathnode, root);
1408 : /* Must do this last, else cost_append complains */
1409 11441 : pathnode->path.pathkeys = child->pathkeys;
1410 : }
1411 : else
1412 35237 : cost_append(pathnode, root);
1413 :
1414 : /* If the caller provided a row estimate, override the computed value. */
1415 46678 : if (rows >= 0)
1416 294 : pathnode->path.rows = rows;
1417 :
1418 46678 : return pathnode;
1419 : }
1420 :
1421 : /*
1422 : * append_total_cost_compare
1423 : * list_sort comparator for sorting append child paths
1424 : * by total_cost descending
1425 : *
1426 : * For equal total costs, we fall back to comparing startup costs; if those
1427 : * are equal too, break ties using bms_compare on the paths' relids.
1428 : * (This is to avoid getting unpredictable results from list_sort.)
1429 : */
1430 : static int
1431 11044 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1432 : {
1433 11044 : Path *path1 = (Path *) lfirst(a);
1434 11044 : Path *path2 = (Path *) lfirst(b);
1435 : int cmp;
1436 :
1437 11044 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1438 11044 : if (cmp != 0)
1439 9839 : return -cmp;
1440 1205 : return bms_compare(path1->parent->relids, path2->parent->relids);
1441 : }
1442 :
1443 : /*
1444 : * append_startup_cost_compare
1445 : * list_sort comparator for sorting append child paths
1446 : * by startup_cost descending
1447 : *
1448 : * For equal startup costs, we fall back to comparing total costs; if those
1449 : * are equal too, break ties using bms_compare on the paths' relids.
1450 : * (This is to avoid getting unpredictable results from list_sort.)
1451 : */
1452 : static int
1453 22808 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1454 : {
1455 22808 : Path *path1 = (Path *) lfirst(a);
1456 22808 : Path *path2 = (Path *) lfirst(b);
1457 : int cmp;
1458 :
1459 22808 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1460 22808 : if (cmp != 0)
1461 10955 : return -cmp;
1462 11853 : return bms_compare(path1->parent->relids, path2->parent->relids);
1463 : }
1464 :
1465 : /*
1466 : * create_merge_append_path
1467 : * Creates a path corresponding to a MergeAppend plan, returning the
1468 : * pathnode.
1469 : */
1470 : MergeAppendPath *
1471 5014 : create_merge_append_path(PlannerInfo *root,
1472 : RelOptInfo *rel,
1473 : List *subpaths,
1474 : List *child_append_relid_sets,
1475 : List *pathkeys,
1476 : Relids required_outer)
1477 : {
1478 5014 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1479 : int input_disabled_nodes;
1480 : Cost input_startup_cost;
1481 : Cost input_total_cost;
1482 : ListCell *l;
1483 :
1484 : /*
1485 : * We don't currently support parameterized MergeAppend paths, as
1486 : * explained in the comments for generate_orderedappend_paths.
1487 : */
1488 : Assert(bms_is_empty(rel->lateral_relids) && bms_is_empty(required_outer));
1489 :
1490 5014 : pathnode->child_append_relid_sets = child_append_relid_sets;
1491 5014 : pathnode->path.pathtype = T_MergeAppend;
1492 5014 : pathnode->path.parent = rel;
1493 5014 : pathnode->path.pathtarget = rel->reltarget;
1494 5014 : pathnode->path.param_info = NULL;
1495 5014 : pathnode->path.parallel_aware = false;
1496 5014 : pathnode->path.parallel_safe = rel->consider_parallel;
1497 5014 : pathnode->path.parallel_workers = 0;
1498 5014 : pathnode->path.pathkeys = pathkeys;
1499 5014 : pathnode->subpaths = subpaths;
1500 :
1501 : /*
1502 : * Apply query-wide LIMIT if known and path is for sole base relation.
1503 : * (Handling this at this low level is a bit klugy.)
1504 : */
1505 5014 : if (bms_equal(rel->relids, root->all_query_rels))
1506 2307 : pathnode->limit_tuples = root->limit_tuples;
1507 : else
1508 2707 : pathnode->limit_tuples = -1.0;
1509 :
1510 : /*
1511 : * Add up the sizes and costs of the input paths.
1512 : */
1513 5014 : pathnode->path.rows = 0;
1514 5014 : input_disabled_nodes = 0;
1515 5014 : input_startup_cost = 0;
1516 5014 : input_total_cost = 0;
1517 18078 : foreach(l, subpaths)
1518 : {
1519 13064 : Path *subpath = (Path *) lfirst(l);
1520 : int presorted_keys;
1521 : Path sort_path; /* dummy for result of
1522 : * cost_sort/cost_incremental_sort */
1523 :
1524 : /* All child paths should be unparameterized */
1525 : Assert(bms_is_empty(PATH_REQ_OUTER(subpath)));
1526 :
1527 13064 : pathnode->path.rows += subpath->rows;
1528 24800 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1529 11736 : subpath->parallel_safe;
1530 :
1531 13064 : if (!pathkeys_count_contained_in(pathkeys, subpath->pathkeys,
1532 : &presorted_keys))
1533 : {
1534 : /*
1535 : * We'll need to insert a Sort node, so include costs for that. We
1536 : * choose to use incremental sort if it is enabled and there are
1537 : * presorted keys; otherwise we use full sort.
1538 : *
1539 : * We can use the parent's LIMIT if any, since we certainly won't
1540 : * pull more than that many tuples from any child.
1541 : */
1542 157 : if (enable_incremental_sort && presorted_keys > 0)
1543 : {
1544 9 : cost_incremental_sort(&sort_path,
1545 : root,
1546 : pathkeys,
1547 : presorted_keys,
1548 : subpath->disabled_nodes,
1549 : subpath->startup_cost,
1550 : subpath->total_cost,
1551 : subpath->rows,
1552 9 : subpath->pathtarget->width,
1553 : 0.0,
1554 : work_mem,
1555 : pathnode->limit_tuples);
1556 : }
1557 : else
1558 : {
1559 148 : cost_sort(&sort_path,
1560 : root,
1561 : pathkeys,
1562 : subpath->disabled_nodes,
1563 : subpath->total_cost,
1564 : subpath->rows,
1565 148 : subpath->pathtarget->width,
1566 : 0.0,
1567 : work_mem,
1568 : pathnode->limit_tuples);
1569 : }
1570 :
1571 157 : subpath = &sort_path;
1572 : }
1573 :
1574 13064 : input_disabled_nodes += subpath->disabled_nodes;
1575 13064 : input_startup_cost += subpath->startup_cost;
1576 13064 : input_total_cost += subpath->total_cost;
1577 : }
1578 :
1579 : /*
1580 : * Now we can compute total costs of the MergeAppend. If there's exactly
1581 : * one child path and its parallel awareness matches that of the
1582 : * MergeAppend, then the MergeAppend is a no-op and will be discarded
1583 : * later (in setrefs.c); otherwise we do the normal cost calculation.
1584 : */
1585 5014 : if (list_length(subpaths) == 1 &&
1586 64 : ((Path *) linitial(subpaths))->parallel_aware ==
1587 64 : pathnode->path.parallel_aware)
1588 : {
1589 64 : pathnode->path.disabled_nodes = input_disabled_nodes;
1590 64 : pathnode->path.startup_cost = input_startup_cost;
1591 64 : pathnode->path.total_cost = input_total_cost;
1592 : }
1593 : else
1594 4950 : cost_merge_append(&pathnode->path, root,
1595 : pathkeys, list_length(subpaths),
1596 : input_disabled_nodes,
1597 : input_startup_cost, input_total_cost,
1598 : pathnode->path.rows);
1599 :
1600 5014 : return pathnode;
1601 : }
1602 :
1603 : /*
1604 : * create_group_result_path
1605 : * Creates a path representing a Result-and-nothing-else plan.
1606 : *
1607 : * This is only used for degenerate grouping cases, in which we know we
1608 : * need to produce one result row, possibly filtered by a HAVING qual.
1609 : */
1610 : GroupResultPath *
1611 99805 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1612 : PathTarget *target, List *havingqual)
1613 : {
1614 99805 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1615 :
1616 99805 : pathnode->path.pathtype = T_Result;
1617 99805 : pathnode->path.parent = rel;
1618 99805 : pathnode->path.pathtarget = target;
1619 99805 : pathnode->path.param_info = NULL; /* there are no other rels... */
1620 99805 : pathnode->path.parallel_aware = false;
1621 99805 : pathnode->path.parallel_safe = rel->consider_parallel;
1622 99805 : pathnode->path.parallel_workers = 0;
1623 99805 : pathnode->path.pathkeys = NIL;
1624 99805 : pathnode->quals = havingqual;
1625 :
1626 : /*
1627 : * We can't quite use cost_resultscan() because the quals we want to
1628 : * account for are not baserestrict quals of the rel. Might as well just
1629 : * hack it here.
1630 : */
1631 99805 : pathnode->path.rows = 1;
1632 99805 : pathnode->path.startup_cost = target->cost.startup;
1633 99805 : pathnode->path.total_cost = target->cost.startup +
1634 99805 : cpu_tuple_cost + target->cost.per_tuple;
1635 :
1636 : /*
1637 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1638 : * rowcount estimate should be 1 no matter what the qual is.
1639 : */
1640 99805 : if (havingqual)
1641 : {
1642 : QualCost qual_cost;
1643 :
1644 316 : cost_qual_eval(&qual_cost, havingqual, root);
1645 : /* havingqual is evaluated once at startup */
1646 316 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1647 316 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1648 : }
1649 :
1650 99805 : return pathnode;
1651 : }
1652 :
1653 : /*
1654 : * create_material_path
1655 : * Creates a path corresponding to a Material plan, returning the
1656 : * pathnode.
1657 : */
1658 : MaterialPath *
1659 347367 : create_material_path(RelOptInfo *rel, Path *subpath, bool enabled)
1660 : {
1661 347367 : MaterialPath *pathnode = makeNode(MaterialPath);
1662 :
1663 : Assert(subpath->parent == rel);
1664 :
1665 347367 : pathnode->path.pathtype = T_Material;
1666 347367 : pathnode->path.parent = rel;
1667 347367 : pathnode->path.pathtarget = rel->reltarget;
1668 347367 : pathnode->path.param_info = subpath->param_info;
1669 347367 : pathnode->path.parallel_aware = false;
1670 665938 : pathnode->path.parallel_safe = rel->consider_parallel &&
1671 318571 : subpath->parallel_safe;
1672 347367 : pathnode->path.parallel_workers = subpath->parallel_workers;
1673 347367 : pathnode->path.pathkeys = subpath->pathkeys;
1674 :
1675 347367 : pathnode->subpath = subpath;
1676 :
1677 347367 : cost_material(&pathnode->path,
1678 : enabled,
1679 : subpath->disabled_nodes,
1680 : subpath->startup_cost,
1681 : subpath->total_cost,
1682 : subpath->rows,
1683 347367 : subpath->pathtarget->width);
1684 :
1685 347367 : return pathnode;
1686 : }
1687 :
1688 : /*
1689 : * create_memoize_path
1690 : * Creates a path corresponding to a Memoize plan, returning the pathnode.
1691 : */
1692 : MemoizePath *
1693 159833 : create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1694 : List *param_exprs, List *hash_operators,
1695 : bool singlerow, bool binary_mode, Cardinality est_calls)
1696 : {
1697 159833 : MemoizePath *pathnode = makeNode(MemoizePath);
1698 :
1699 : Assert(subpath->parent == rel);
1700 :
1701 159833 : pathnode->path.pathtype = T_Memoize;
1702 159833 : pathnode->path.parent = rel;
1703 159833 : pathnode->path.pathtarget = rel->reltarget;
1704 159833 : pathnode->path.param_info = subpath->param_info;
1705 159833 : pathnode->path.parallel_aware = false;
1706 312539 : pathnode->path.parallel_safe = rel->consider_parallel &&
1707 152706 : subpath->parallel_safe;
1708 159833 : pathnode->path.parallel_workers = subpath->parallel_workers;
1709 159833 : pathnode->path.pathkeys = subpath->pathkeys;
1710 :
1711 159833 : pathnode->subpath = subpath;
1712 159833 : pathnode->hash_operators = hash_operators;
1713 159833 : pathnode->param_exprs = param_exprs;
1714 159833 : pathnode->singlerow = singlerow;
1715 159833 : pathnode->binary_mode = binary_mode;
1716 :
1717 : /*
1718 : * For now we set est_entries to 0. cost_memoize_rescan() does all the
1719 : * hard work to determine how many cache entries there are likely to be,
1720 : * so it seems best to leave it up to that function to fill this field in.
1721 : * If left at 0, the executor will make a guess at a good value.
1722 : */
1723 159833 : pathnode->est_entries = 0;
1724 :
1725 159833 : pathnode->est_calls = clamp_row_est(est_calls);
1726 :
1727 : /* These will also be set later in cost_memoize_rescan() */
1728 159833 : pathnode->est_unique_keys = 0.0;
1729 159833 : pathnode->est_hit_ratio = 0.0;
1730 :
1731 : /*
1732 : * We should not be asked to generate this path type when memoization is
1733 : * disabled, so set our count of disabled nodes equal to the subpath's
1734 : * count.
1735 : *
1736 : * It would be nice to also Assert that memoization is enabled, but the
1737 : * value of enable_memoize is not controlling: what we would need to check
1738 : * is that the JoinPathExtraData's pgs_mask included PGS_NESTLOOP_MEMOIZE.
1739 : */
1740 159833 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
1741 :
1742 : /*
1743 : * Add a small additional charge for caching the first entry. All the
1744 : * harder calculations for rescans are performed in cost_memoize_rescan().
1745 : */
1746 159833 : pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
1747 159833 : pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
1748 159833 : pathnode->path.rows = subpath->rows;
1749 :
1750 159833 : return pathnode;
1751 : }
1752 :
1753 : /*
1754 : * create_gather_merge_path
1755 : *
1756 : * Creates a path corresponding to a gather merge scan, returning
1757 : * the pathnode.
1758 : */
1759 : GatherMergePath *
1760 9478 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1761 : PathTarget *target, List *pathkeys,
1762 : Relids required_outer, double *rows)
1763 : {
1764 9478 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1765 9478 : int input_disabled_nodes = 0;
1766 9478 : Cost input_startup_cost = 0;
1767 9478 : Cost input_total_cost = 0;
1768 :
1769 : Assert(subpath->parallel_safe);
1770 : Assert(pathkeys);
1771 :
1772 : /*
1773 : * The subpath should guarantee that it is adequately ordered either by
1774 : * adding an explicit sort node or by using presorted input. We cannot
1775 : * add an explicit Sort node for the subpath in createplan.c on additional
1776 : * pathkeys, because we can't guarantee the sort would be safe. For
1777 : * example, expressions may be volatile or otherwise parallel unsafe.
1778 : */
1779 9478 : if (!pathkeys_contained_in(pathkeys, subpath->pathkeys))
1780 0 : elog(ERROR, "gather merge input not sufficiently sorted");
1781 :
1782 9478 : pathnode->path.pathtype = T_GatherMerge;
1783 9478 : pathnode->path.parent = rel;
1784 9478 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1785 : required_outer);
1786 9478 : pathnode->path.parallel_aware = false;
1787 :
1788 9478 : pathnode->subpath = subpath;
1789 9478 : pathnode->num_workers = subpath->parallel_workers;
1790 9478 : pathnode->path.pathkeys = pathkeys;
1791 9478 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1792 :
1793 9478 : input_disabled_nodes += subpath->disabled_nodes;
1794 9478 : input_startup_cost += subpath->startup_cost;
1795 9478 : input_total_cost += subpath->total_cost;
1796 :
1797 9478 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1798 : input_disabled_nodes, input_startup_cost,
1799 : input_total_cost, rows);
1800 :
1801 9478 : return pathnode;
1802 : }
1803 :
1804 : /*
1805 : * create_gather_path
1806 : * Creates a path corresponding to a gather scan, returning the
1807 : * pathnode.
1808 : *
1809 : * 'rows' may optionally be set to override row estimates from other sources.
1810 : */
1811 : GatherPath *
1812 13406 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1813 : PathTarget *target, Relids required_outer, double *rows)
1814 : {
1815 13406 : GatherPath *pathnode = makeNode(GatherPath);
1816 :
1817 : Assert(subpath->parallel_safe);
1818 :
1819 13406 : pathnode->path.pathtype = T_Gather;
1820 13406 : pathnode->path.parent = rel;
1821 13406 : pathnode->path.pathtarget = target;
1822 13406 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1823 : required_outer);
1824 13406 : pathnode->path.parallel_aware = false;
1825 13406 : pathnode->path.parallel_safe = false;
1826 13406 : pathnode->path.parallel_workers = 0;
1827 13406 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1828 :
1829 13406 : pathnode->subpath = subpath;
1830 13406 : pathnode->num_workers = subpath->parallel_workers;
1831 13406 : pathnode->single_copy = false;
1832 :
1833 13406 : if (pathnode->num_workers == 0)
1834 : {
1835 0 : pathnode->path.pathkeys = subpath->pathkeys;
1836 0 : pathnode->num_workers = 1;
1837 0 : pathnode->single_copy = true;
1838 : }
1839 :
1840 13406 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
1841 :
1842 13406 : return pathnode;
1843 : }
1844 :
1845 : /*
1846 : * create_subqueryscan_path
1847 : * Creates a path corresponding to a scan of a subquery,
1848 : * returning the pathnode.
1849 : *
1850 : * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
1851 : * be trivial, ie just a fetch of all the subquery output columns in order.
1852 : * While we could determine that here, the caller can usually do it more
1853 : * efficiently (or at least amortize it over multiple calls).
1854 : */
1855 : SubqueryScanPath *
1856 31601 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1857 : bool trivial_pathtarget,
1858 : List *pathkeys, Relids required_outer)
1859 : {
1860 31601 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
1861 :
1862 31601 : pathnode->path.pathtype = T_SubqueryScan;
1863 31601 : pathnode->path.parent = rel;
1864 31601 : pathnode->path.pathtarget = rel->reltarget;
1865 31601 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1866 : required_outer);
1867 31601 : pathnode->path.parallel_aware = false;
1868 52605 : pathnode->path.parallel_safe = rel->consider_parallel &&
1869 21004 : subpath->parallel_safe;
1870 31601 : pathnode->path.parallel_workers = subpath->parallel_workers;
1871 31601 : pathnode->path.pathkeys = pathkeys;
1872 31601 : pathnode->subpath = subpath;
1873 :
1874 31601 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
1875 : trivial_pathtarget);
1876 :
1877 31601 : return pathnode;
1878 : }
1879 :
1880 : /*
1881 : * create_functionscan_path
1882 : * Creates a path corresponding to a sequential scan of a function,
1883 : * returning the pathnode.
1884 : */
1885 : Path *
1886 26434 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
1887 : List *pathkeys, Relids required_outer)
1888 : {
1889 26434 : Path *pathnode = makeNode(Path);
1890 :
1891 26434 : pathnode->pathtype = T_FunctionScan;
1892 26434 : pathnode->parent = rel;
1893 26434 : pathnode->pathtarget = rel->reltarget;
1894 26434 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1895 : required_outer);
1896 26434 : pathnode->parallel_aware = false;
1897 26434 : pathnode->parallel_safe = rel->consider_parallel;
1898 26434 : pathnode->parallel_workers = 0;
1899 26434 : pathnode->pathkeys = pathkeys;
1900 :
1901 26434 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
1902 :
1903 26434 : return pathnode;
1904 : }
1905 :
1906 : /*
1907 : * create_tablefuncscan_path
1908 : * Creates a path corresponding to a sequential scan of a table function,
1909 : * returning the pathnode.
1910 : */
1911 : Path *
1912 313 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
1913 : Relids required_outer)
1914 : {
1915 313 : Path *pathnode = makeNode(Path);
1916 :
1917 313 : pathnode->pathtype = T_TableFuncScan;
1918 313 : pathnode->parent = rel;
1919 313 : pathnode->pathtarget = rel->reltarget;
1920 313 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1921 : required_outer);
1922 313 : pathnode->parallel_aware = false;
1923 313 : pathnode->parallel_safe = rel->consider_parallel;
1924 313 : pathnode->parallel_workers = 0;
1925 313 : pathnode->pathkeys = NIL; /* result is always unordered */
1926 :
1927 313 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
1928 :
1929 313 : return pathnode;
1930 : }
1931 :
1932 : /*
1933 : * create_valuesscan_path
1934 : * Creates a path corresponding to a scan of a VALUES list,
1935 : * returning the pathnode.
1936 : */
1937 : Path *
1938 4292 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
1939 : Relids required_outer)
1940 : {
1941 4292 : Path *pathnode = makeNode(Path);
1942 :
1943 4292 : pathnode->pathtype = T_ValuesScan;
1944 4292 : pathnode->parent = rel;
1945 4292 : pathnode->pathtarget = rel->reltarget;
1946 4292 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1947 : required_outer);
1948 4292 : pathnode->parallel_aware = false;
1949 4292 : pathnode->parallel_safe = rel->consider_parallel;
1950 4292 : pathnode->parallel_workers = 0;
1951 4292 : pathnode->pathkeys = NIL; /* result is always unordered */
1952 :
1953 4292 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
1954 :
1955 4292 : return pathnode;
1956 : }
1957 :
1958 : /*
1959 : * create_ctescan_path
1960 : * Creates a path corresponding to a scan of a non-self-reference CTE,
1961 : * returning the pathnode.
1962 : */
1963 : Path *
1964 2228 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel,
1965 : List *pathkeys, Relids required_outer)
1966 : {
1967 2228 : Path *pathnode = makeNode(Path);
1968 :
1969 2228 : pathnode->pathtype = T_CteScan;
1970 2228 : pathnode->parent = rel;
1971 2228 : pathnode->pathtarget = rel->reltarget;
1972 2228 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1973 : required_outer);
1974 2228 : pathnode->parallel_aware = false;
1975 2228 : pathnode->parallel_safe = rel->consider_parallel;
1976 2228 : pathnode->parallel_workers = 0;
1977 2228 : pathnode->pathkeys = pathkeys;
1978 :
1979 2228 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
1980 :
1981 2228 : return pathnode;
1982 : }
1983 :
1984 : /*
1985 : * create_namedtuplestorescan_path
1986 : * Creates a path corresponding to a scan of a named tuplestore, returning
1987 : * the pathnode.
1988 : */
1989 : Path *
1990 237 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
1991 : Relids required_outer)
1992 : {
1993 237 : Path *pathnode = makeNode(Path);
1994 :
1995 237 : pathnode->pathtype = T_NamedTuplestoreScan;
1996 237 : pathnode->parent = rel;
1997 237 : pathnode->pathtarget = rel->reltarget;
1998 237 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1999 : required_outer);
2000 237 : pathnode->parallel_aware = false;
2001 237 : pathnode->parallel_safe = rel->consider_parallel;
2002 237 : pathnode->parallel_workers = 0;
2003 237 : pathnode->pathkeys = NIL; /* result is always unordered */
2004 :
2005 237 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
2006 :
2007 237 : return pathnode;
2008 : }
2009 :
2010 : /*
2011 : * create_resultscan_path
2012 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2013 : * returning the pathnode.
2014 : */
2015 : Path *
2016 2176 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2017 : Relids required_outer)
2018 : {
2019 2176 : Path *pathnode = makeNode(Path);
2020 :
2021 2176 : pathnode->pathtype = T_Result;
2022 2176 : pathnode->parent = rel;
2023 2176 : pathnode->pathtarget = rel->reltarget;
2024 2176 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2025 : required_outer);
2026 2176 : pathnode->parallel_aware = false;
2027 2176 : pathnode->parallel_safe = rel->consider_parallel;
2028 2176 : pathnode->parallel_workers = 0;
2029 2176 : pathnode->pathkeys = NIL; /* result is always unordered */
2030 :
2031 2176 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2032 :
2033 2176 : return pathnode;
2034 : }
2035 :
2036 : /*
2037 : * create_worktablescan_path
2038 : * Creates a path corresponding to a scan of a self-reference CTE,
2039 : * returning the pathnode.
2040 : */
2041 : Path *
2042 471 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2043 : Relids required_outer)
2044 : {
2045 471 : Path *pathnode = makeNode(Path);
2046 :
2047 471 : pathnode->pathtype = T_WorkTableScan;
2048 471 : pathnode->parent = rel;
2049 471 : pathnode->pathtarget = rel->reltarget;
2050 471 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2051 : required_outer);
2052 471 : pathnode->parallel_aware = false;
2053 471 : pathnode->parallel_safe = rel->consider_parallel;
2054 471 : pathnode->parallel_workers = 0;
2055 471 : pathnode->pathkeys = NIL; /* result is always unordered */
2056 :
2057 : /* Cost is the same as for a regular CTE scan */
2058 471 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2059 :
2060 471 : return pathnode;
2061 : }
2062 :
2063 : /*
2064 : * create_foreignscan_path
2065 : * Creates a path corresponding to a scan of a foreign base table,
2066 : * returning the pathnode.
2067 : *
2068 : * This function is never called from core Postgres; rather, it's expected
2069 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2070 : * We make the FDW supply all fields of the path, since we do not have any way
2071 : * to calculate them in core. However, there is a usually-sane default for
2072 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2073 : */
2074 : ForeignPath *
2075 1863 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2076 : PathTarget *target,
2077 : double rows, int disabled_nodes,
2078 : Cost startup_cost, Cost total_cost,
2079 : List *pathkeys,
2080 : Relids required_outer,
2081 : Path *fdw_outerpath,
2082 : List *fdw_restrictinfo,
2083 : List *fdw_private)
2084 : {
2085 1863 : ForeignPath *pathnode = makeNode(ForeignPath);
2086 :
2087 : /* Historically some FDWs were confused about when to use this */
2088 : Assert(IS_SIMPLE_REL(rel));
2089 :
2090 1863 : pathnode->path.pathtype = T_ForeignScan;
2091 1863 : pathnode->path.parent = rel;
2092 1863 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2093 1863 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2094 : required_outer);
2095 1863 : pathnode->path.parallel_aware = false;
2096 1863 : pathnode->path.parallel_safe = rel->consider_parallel;
2097 1863 : pathnode->path.parallel_workers = 0;
2098 1863 : pathnode->path.rows = rows;
2099 1863 : pathnode->path.disabled_nodes = disabled_nodes;
2100 1863 : pathnode->path.startup_cost = startup_cost;
2101 1863 : pathnode->path.total_cost = total_cost;
2102 1863 : pathnode->path.pathkeys = pathkeys;
2103 :
2104 1863 : pathnode->fdw_outerpath = fdw_outerpath;
2105 1863 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2106 1863 : pathnode->fdw_private = fdw_private;
2107 :
2108 1863 : return pathnode;
2109 : }
2110 :
2111 : /*
2112 : * create_foreign_join_path
2113 : * Creates a path corresponding to a scan of a foreign join,
2114 : * returning the pathnode.
2115 : *
2116 : * This function is never called from core Postgres; rather, it's expected
2117 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2118 : * We make the FDW supply all fields of the path, since we do not have any way
2119 : * to calculate them in core. However, there is a usually-sane default for
2120 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2121 : */
2122 : ForeignPath *
2123 606 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2124 : PathTarget *target,
2125 : double rows, int disabled_nodes,
2126 : Cost startup_cost, Cost total_cost,
2127 : List *pathkeys,
2128 : Relids required_outer,
2129 : Path *fdw_outerpath,
2130 : List *fdw_restrictinfo,
2131 : List *fdw_private)
2132 : {
2133 606 : ForeignPath *pathnode = makeNode(ForeignPath);
2134 :
2135 : /*
2136 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2137 : * but the API of this function doesn't support it, and existing
2138 : * extensions aren't yet trying to build such paths anyway. For the
2139 : * moment just throw an error if someone tries it; eventually we should
2140 : * revisit this.
2141 : */
2142 606 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2143 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2144 :
2145 606 : pathnode->path.pathtype = T_ForeignScan;
2146 606 : pathnode->path.parent = rel;
2147 606 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2148 606 : pathnode->path.param_info = NULL; /* XXX see above */
2149 606 : pathnode->path.parallel_aware = false;
2150 606 : pathnode->path.parallel_safe = rel->consider_parallel;
2151 606 : pathnode->path.parallel_workers = 0;
2152 606 : pathnode->path.rows = rows;
2153 606 : pathnode->path.disabled_nodes = disabled_nodes;
2154 606 : pathnode->path.startup_cost = startup_cost;
2155 606 : pathnode->path.total_cost = total_cost;
2156 606 : pathnode->path.pathkeys = pathkeys;
2157 :
2158 606 : pathnode->fdw_outerpath = fdw_outerpath;
2159 606 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2160 606 : pathnode->fdw_private = fdw_private;
2161 :
2162 606 : return pathnode;
2163 : }
2164 :
2165 : /*
2166 : * create_foreign_upper_path
2167 : * Creates a path corresponding to an upper relation that's computed
2168 : * directly by an FDW, returning the pathnode.
2169 : *
2170 : * This function is never called from core Postgres; rather, it's expected to
2171 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2172 : * We make the FDW supply all fields of the path, since we do not have any way
2173 : * to calculate them in core. However, there is a usually-sane default for
2174 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2175 : */
2176 : ForeignPath *
2177 294 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2178 : PathTarget *target,
2179 : double rows, int disabled_nodes,
2180 : Cost startup_cost, Cost total_cost,
2181 : List *pathkeys,
2182 : Path *fdw_outerpath,
2183 : List *fdw_restrictinfo,
2184 : List *fdw_private)
2185 : {
2186 294 : ForeignPath *pathnode = makeNode(ForeignPath);
2187 :
2188 : /*
2189 : * Upper relations should never have any lateral references, since joining
2190 : * is complete.
2191 : */
2192 : Assert(bms_is_empty(rel->lateral_relids));
2193 :
2194 294 : pathnode->path.pathtype = T_ForeignScan;
2195 294 : pathnode->path.parent = rel;
2196 294 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2197 294 : pathnode->path.param_info = NULL;
2198 294 : pathnode->path.parallel_aware = false;
2199 294 : pathnode->path.parallel_safe = rel->consider_parallel;
2200 294 : pathnode->path.parallel_workers = 0;
2201 294 : pathnode->path.rows = rows;
2202 294 : pathnode->path.disabled_nodes = disabled_nodes;
2203 294 : pathnode->path.startup_cost = startup_cost;
2204 294 : pathnode->path.total_cost = total_cost;
2205 294 : pathnode->path.pathkeys = pathkeys;
2206 :
2207 294 : pathnode->fdw_outerpath = fdw_outerpath;
2208 294 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2209 294 : pathnode->fdw_private = fdw_private;
2210 :
2211 294 : return pathnode;
2212 : }
2213 :
2214 : /*
2215 : * calc_nestloop_required_outer
2216 : * Compute the required_outer set for a nestloop join path
2217 : *
2218 : * Note: when considering a child join, the inputs nonetheless use top-level
2219 : * parent relids
2220 : *
2221 : * Note: result must not share storage with either input
2222 : */
2223 : Relids
2224 1753616 : calc_nestloop_required_outer(Relids outerrelids,
2225 : Relids outer_paramrels,
2226 : Relids innerrelids,
2227 : Relids inner_paramrels)
2228 : {
2229 : Relids required_outer;
2230 :
2231 : /* inner_path can require rels from outer path, but not vice versa */
2232 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2233 : /* easy case if inner path is not parameterized */
2234 1753616 : if (!inner_paramrels)
2235 1220152 : return bms_copy(outer_paramrels);
2236 : /* else, form the union ... */
2237 533464 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2238 : /* ... and remove any mention of now-satisfied outer rels */
2239 533464 : required_outer = bms_del_members(required_outer,
2240 : outerrelids);
2241 533464 : return required_outer;
2242 : }
2243 :
2244 : /*
2245 : * calc_non_nestloop_required_outer
2246 : * Compute the required_outer set for a merge or hash join path
2247 : *
2248 : * Note: result must not share storage with either input
2249 : */
2250 : Relids
2251 1179447 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2252 : {
2253 1179447 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2254 1179447 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2255 : Relids innerrelids PG_USED_FOR_ASSERTS_ONLY;
2256 : Relids outerrelids PG_USED_FOR_ASSERTS_ONLY;
2257 : Relids required_outer;
2258 :
2259 : /*
2260 : * Any parameterization of the input paths refers to topmost parents of
2261 : * the relevant relations, because reparameterize_path_by_child() hasn't
2262 : * been called yet. So we must consider topmost parents of the relations
2263 : * being joined, too, while checking for disallowed parameterization
2264 : * cases.
2265 : */
2266 1179447 : if (inner_path->parent->top_parent_relids)
2267 79585 : innerrelids = inner_path->parent->top_parent_relids;
2268 : else
2269 1099862 : innerrelids = inner_path->parent->relids;
2270 :
2271 1179447 : if (outer_path->parent->top_parent_relids)
2272 79585 : outerrelids = outer_path->parent->top_parent_relids;
2273 : else
2274 1099862 : outerrelids = outer_path->parent->relids;
2275 :
2276 : /* neither path can require rels from the other */
2277 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2278 : Assert(!bms_overlap(inner_paramrels, outerrelids));
2279 : /* form the union ... */
2280 1179447 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2281 : /* we do not need an explicit test for empty; bms_union gets it right */
2282 1179447 : return required_outer;
2283 : }
2284 :
2285 : /*
2286 : * create_nestloop_path
2287 : * Creates a pathnode corresponding to a nestloop join between two
2288 : * relations.
2289 : *
2290 : * 'joinrel' is the join relation.
2291 : * 'jointype' is the type of join required
2292 : * 'workspace' is the result from initial_cost_nestloop
2293 : * 'extra' contains various information about the join
2294 : * 'outer_path' is the outer path
2295 : * 'inner_path' is the inner path
2296 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2297 : * 'pathkeys' are the path keys of the new join path
2298 : * 'required_outer' is the set of required outer rels
2299 : *
2300 : * Returns the resulting path node.
2301 : */
2302 : NestPath *
2303 759644 : create_nestloop_path(PlannerInfo *root,
2304 : RelOptInfo *joinrel,
2305 : JoinType jointype,
2306 : JoinCostWorkspace *workspace,
2307 : JoinPathExtraData *extra,
2308 : Path *outer_path,
2309 : Path *inner_path,
2310 : List *restrict_clauses,
2311 : List *pathkeys,
2312 : Relids required_outer)
2313 : {
2314 759644 : NestPath *pathnode = makeNode(NestPath);
2315 759644 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2316 : Relids outerrelids;
2317 :
2318 : /*
2319 : * Paths are parameterized by top-level parents, so run parameterization
2320 : * tests on the parent relids.
2321 : */
2322 759644 : if (outer_path->parent->top_parent_relids)
2323 38177 : outerrelids = outer_path->parent->top_parent_relids;
2324 : else
2325 721467 : outerrelids = outer_path->parent->relids;
2326 :
2327 : /*
2328 : * If the inner path is parameterized by the outer, we must drop any
2329 : * restrict_clauses that are due to be moved into the inner path. We have
2330 : * to do this now, rather than postpone the work till createplan time,
2331 : * because the restrict_clauses list can affect the size and cost
2332 : * estimates for this path. We detect such clauses by checking for serial
2333 : * number match to clauses already enforced in the inner path.
2334 : */
2335 759644 : if (bms_overlap(inner_req_outer, outerrelids))
2336 : {
2337 204734 : Bitmapset *enforced_serials = get_param_path_clause_serials(inner_path);
2338 204734 : List *jclauses = NIL;
2339 : ListCell *lc;
2340 :
2341 453827 : foreach(lc, restrict_clauses)
2342 : {
2343 249093 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2344 :
2345 249093 : if (!bms_is_member(rinfo->rinfo_serial, enforced_serials))
2346 32349 : jclauses = lappend(jclauses, rinfo);
2347 : }
2348 204734 : restrict_clauses = jclauses;
2349 : }
2350 :
2351 759644 : pathnode->jpath.path.pathtype = T_NestLoop;
2352 759644 : pathnode->jpath.path.parent = joinrel;
2353 759644 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2354 759644 : pathnode->jpath.path.param_info =
2355 759644 : get_joinrel_parampathinfo(root,
2356 : joinrel,
2357 : outer_path,
2358 : inner_path,
2359 : extra->sjinfo,
2360 : required_outer,
2361 : &restrict_clauses);
2362 759644 : pathnode->jpath.path.parallel_aware = false;
2363 2212872 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2364 759644 : outer_path->parallel_safe && inner_path->parallel_safe;
2365 : /* This is a foolish way to estimate parallel_workers, but for now... */
2366 759644 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2367 759644 : pathnode->jpath.path.pathkeys = pathkeys;
2368 759644 : pathnode->jpath.jointype = jointype;
2369 759644 : pathnode->jpath.inner_unique = extra->inner_unique;
2370 759644 : pathnode->jpath.outerjoinpath = outer_path;
2371 759644 : pathnode->jpath.innerjoinpath = inner_path;
2372 759644 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2373 :
2374 759644 : final_cost_nestloop(root, pathnode, workspace, extra);
2375 :
2376 759644 : return pathnode;
2377 : }
2378 :
2379 : /*
2380 : * create_mergejoin_path
2381 : * Creates a pathnode corresponding to a mergejoin join between
2382 : * two relations
2383 : *
2384 : * 'joinrel' is the join relation
2385 : * 'jointype' is the type of join required
2386 : * 'workspace' is the result from initial_cost_mergejoin
2387 : * 'extra' contains various information about the join
2388 : * 'outer_path' is the outer path
2389 : * 'inner_path' is the inner path
2390 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2391 : * 'pathkeys' are the path keys of the new join path
2392 : * 'required_outer' is the set of required outer rels
2393 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2394 : * (this should be a subset of the restrict_clauses list)
2395 : * 'outersortkeys' are the sort varkeys for the outer relation
2396 : * 'innersortkeys' are the sort varkeys for the inner relation
2397 : * 'outer_presorted_keys' is the number of presorted keys of the outer path
2398 : */
2399 : MergePath *
2400 235454 : create_mergejoin_path(PlannerInfo *root,
2401 : RelOptInfo *joinrel,
2402 : JoinType jointype,
2403 : JoinCostWorkspace *workspace,
2404 : JoinPathExtraData *extra,
2405 : Path *outer_path,
2406 : Path *inner_path,
2407 : List *restrict_clauses,
2408 : List *pathkeys,
2409 : Relids required_outer,
2410 : List *mergeclauses,
2411 : List *outersortkeys,
2412 : List *innersortkeys,
2413 : int outer_presorted_keys)
2414 : {
2415 235454 : MergePath *pathnode = makeNode(MergePath);
2416 :
2417 235454 : pathnode->jpath.path.pathtype = T_MergeJoin;
2418 235454 : pathnode->jpath.path.parent = joinrel;
2419 235454 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2420 235454 : pathnode->jpath.path.param_info =
2421 235454 : get_joinrel_parampathinfo(root,
2422 : joinrel,
2423 : outer_path,
2424 : inner_path,
2425 : extra->sjinfo,
2426 : required_outer,
2427 : &restrict_clauses);
2428 235454 : pathnode->jpath.path.parallel_aware = false;
2429 687601 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2430 235454 : outer_path->parallel_safe && inner_path->parallel_safe;
2431 : /* This is a foolish way to estimate parallel_workers, but for now... */
2432 235454 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2433 235454 : pathnode->jpath.path.pathkeys = pathkeys;
2434 235454 : pathnode->jpath.jointype = jointype;
2435 235454 : pathnode->jpath.inner_unique = extra->inner_unique;
2436 235454 : pathnode->jpath.outerjoinpath = outer_path;
2437 235454 : pathnode->jpath.innerjoinpath = inner_path;
2438 235454 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2439 235454 : pathnode->path_mergeclauses = mergeclauses;
2440 235454 : pathnode->outersortkeys = outersortkeys;
2441 235454 : pathnode->innersortkeys = innersortkeys;
2442 235454 : pathnode->outer_presorted_keys = outer_presorted_keys;
2443 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2444 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2445 :
2446 235454 : final_cost_mergejoin(root, pathnode, workspace, extra);
2447 :
2448 235454 : return pathnode;
2449 : }
2450 :
2451 : /*
2452 : * create_hashjoin_path
2453 : * Creates a pathnode corresponding to a hash join between two relations.
2454 : *
2455 : * 'joinrel' is the join relation
2456 : * 'jointype' is the type of join required
2457 : * 'workspace' is the result from initial_cost_hashjoin
2458 : * 'extra' contains various information about the join
2459 : * 'outer_path' is the cheapest outer path
2460 : * 'inner_path' is the cheapest inner path
2461 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2462 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2463 : * 'required_outer' is the set of required outer rels
2464 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2465 : * (this should be a subset of the restrict_clauses list)
2466 : */
2467 : HashPath *
2468 232223 : create_hashjoin_path(PlannerInfo *root,
2469 : RelOptInfo *joinrel,
2470 : JoinType jointype,
2471 : JoinCostWorkspace *workspace,
2472 : JoinPathExtraData *extra,
2473 : Path *outer_path,
2474 : Path *inner_path,
2475 : bool parallel_hash,
2476 : List *restrict_clauses,
2477 : Relids required_outer,
2478 : List *hashclauses)
2479 : {
2480 232223 : HashPath *pathnode = makeNode(HashPath);
2481 :
2482 232223 : pathnode->jpath.path.pathtype = T_HashJoin;
2483 232223 : pathnode->jpath.path.parent = joinrel;
2484 232223 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2485 232223 : pathnode->jpath.path.param_info =
2486 232223 : get_joinrel_parampathinfo(root,
2487 : joinrel,
2488 : outer_path,
2489 : inner_path,
2490 : extra->sjinfo,
2491 : required_outer,
2492 : &restrict_clauses);
2493 232223 : pathnode->jpath.path.parallel_aware =
2494 232223 : joinrel->consider_parallel && parallel_hash;
2495 677812 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2496 232223 : outer_path->parallel_safe && inner_path->parallel_safe;
2497 : /* This is a foolish way to estimate parallel_workers, but for now... */
2498 232223 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2499 :
2500 : /*
2501 : * A hashjoin never has pathkeys, since its output ordering is
2502 : * unpredictable due to possible batching. XXX If the inner relation is
2503 : * small enough, we could instruct the executor that it must not batch,
2504 : * and then we could assume that the output inherits the outer relation's
2505 : * ordering, which might save a sort step. However there is considerable
2506 : * downside if our estimate of the inner relation size is badly off. For
2507 : * the moment we don't risk it. (Note also that if we wanted to take this
2508 : * seriously, joinpath.c would have to consider many more paths for the
2509 : * outer rel than it does now.)
2510 : */
2511 232223 : pathnode->jpath.path.pathkeys = NIL;
2512 232223 : pathnode->jpath.jointype = jointype;
2513 232223 : pathnode->jpath.inner_unique = extra->inner_unique;
2514 232223 : pathnode->jpath.outerjoinpath = outer_path;
2515 232223 : pathnode->jpath.innerjoinpath = inner_path;
2516 232223 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2517 232223 : pathnode->path_hashclauses = hashclauses;
2518 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2519 :
2520 232223 : final_cost_hashjoin(root, pathnode, workspace, extra);
2521 :
2522 232223 : return pathnode;
2523 : }
2524 :
2525 : /*
2526 : * create_projection_path
2527 : * Creates a pathnode that represents performing a projection.
2528 : *
2529 : * 'rel' is the parent relation associated with the result
2530 : * 'subpath' is the path representing the source of data
2531 : * 'target' is the PathTarget to be computed
2532 : */
2533 : ProjectionPath *
2534 217005 : create_projection_path(PlannerInfo *root,
2535 : RelOptInfo *rel,
2536 : Path *subpath,
2537 : PathTarget *target)
2538 : {
2539 217005 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2540 : PathTarget *oldtarget;
2541 :
2542 : /*
2543 : * We mustn't put a ProjectionPath directly above another; it's useless
2544 : * and will confuse create_projection_plan. Rather than making sure all
2545 : * callers handle that, let's implement it here, by stripping off any
2546 : * ProjectionPath in what we're given. Given this rule, there won't be
2547 : * more than one.
2548 : */
2549 217005 : if (IsA(subpath, ProjectionPath))
2550 : {
2551 12 : ProjectionPath *subpp = (ProjectionPath *) subpath;
2552 :
2553 : Assert(subpp->path.parent == rel);
2554 12 : subpath = subpp->subpath;
2555 : Assert(!IsA(subpath, ProjectionPath));
2556 : }
2557 :
2558 217005 : pathnode->path.pathtype = T_Result;
2559 217005 : pathnode->path.parent = rel;
2560 217005 : pathnode->path.pathtarget = target;
2561 217005 : pathnode->path.param_info = subpath->param_info;
2562 217005 : pathnode->path.parallel_aware = false;
2563 506611 : pathnode->path.parallel_safe = rel->consider_parallel &&
2564 285296 : subpath->parallel_safe &&
2565 68291 : is_parallel_safe(root, (Node *) target->exprs);
2566 217005 : pathnode->path.parallel_workers = subpath->parallel_workers;
2567 : /* Projection does not change the sort order */
2568 217005 : pathnode->path.pathkeys = subpath->pathkeys;
2569 :
2570 217005 : pathnode->subpath = subpath;
2571 :
2572 : /*
2573 : * We might not need a separate Result node. If the input plan node type
2574 : * can project, we can just tell it to project something else. Or, if it
2575 : * can't project but the desired target has the same expression list as
2576 : * what the input will produce anyway, we can still give it the desired
2577 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2578 : * Note: in the latter case, create_projection_plan has to recheck our
2579 : * conclusion; see comments therein.
2580 : */
2581 217005 : oldtarget = subpath->pathtarget;
2582 225638 : if (is_projection_capable_path(subpath) ||
2583 8633 : equal(oldtarget->exprs, target->exprs))
2584 : {
2585 : /* No separate Result node needed */
2586 209373 : pathnode->dummypp = true;
2587 :
2588 : /*
2589 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2590 : */
2591 209373 : pathnode->path.rows = subpath->rows;
2592 209373 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2593 209373 : pathnode->path.startup_cost = subpath->startup_cost +
2594 209373 : (target->cost.startup - oldtarget->cost.startup);
2595 209373 : pathnode->path.total_cost = subpath->total_cost +
2596 209373 : (target->cost.startup - oldtarget->cost.startup) +
2597 209373 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2598 : }
2599 : else
2600 : {
2601 : /* We really do need the Result node */
2602 7632 : pathnode->dummypp = false;
2603 :
2604 : /*
2605 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2606 : * evaluating the tlist. There is no qual to worry about.
2607 : */
2608 7632 : pathnode->path.rows = subpath->rows;
2609 7632 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2610 7632 : pathnode->path.startup_cost = subpath->startup_cost +
2611 7632 : target->cost.startup;
2612 7632 : pathnode->path.total_cost = subpath->total_cost +
2613 7632 : target->cost.startup +
2614 7632 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2615 : }
2616 :
2617 217005 : return pathnode;
2618 : }
2619 :
2620 : /*
2621 : * apply_projection_to_path
2622 : * Add a projection step, or just apply the target directly to given path.
2623 : *
2624 : * This has the same net effect as create_projection_path(), except that if
2625 : * a separate Result plan node isn't needed, we just replace the given path's
2626 : * pathtarget with the desired one. This must be used only when the caller
2627 : * knows that the given path isn't referenced elsewhere and so can be modified
2628 : * in-place.
2629 : *
2630 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2631 : * new target down to its input as well; this is a yet more invasive
2632 : * modification of the input path, which create_projection_path() can't do.
2633 : *
2634 : * Note that we mustn't change the source path's parent link; so when it is
2635 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2636 : * not caused any trouble.
2637 : *
2638 : * 'rel' is the parent relation associated with the result
2639 : * 'path' is the path representing the source of data
2640 : * 'target' is the PathTarget to be computed
2641 : */
2642 : Path *
2643 7062 : apply_projection_to_path(PlannerInfo *root,
2644 : RelOptInfo *rel,
2645 : Path *path,
2646 : PathTarget *target)
2647 : {
2648 : QualCost oldcost;
2649 :
2650 : /*
2651 : * If given path can't project, we might need a Result node, so make a
2652 : * separate ProjectionPath.
2653 : */
2654 7062 : if (!is_projection_capable_path(path))
2655 792 : return (Path *) create_projection_path(root, rel, path, target);
2656 :
2657 : /*
2658 : * We can just jam the desired tlist into the existing path, being sure to
2659 : * update its cost estimates appropriately.
2660 : */
2661 6270 : oldcost = path->pathtarget->cost;
2662 6270 : path->pathtarget = target;
2663 :
2664 6270 : path->startup_cost += target->cost.startup - oldcost.startup;
2665 6270 : path->total_cost += target->cost.startup - oldcost.startup +
2666 6270 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2667 :
2668 : /*
2669 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2670 : * arrange for the subpath to return the required target list so that
2671 : * workers can help project. But if there is something that is not
2672 : * parallel-safe in the target expressions, then we can't.
2673 : */
2674 6282 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
2675 12 : is_parallel_safe(root, (Node *) target->exprs))
2676 : {
2677 : /*
2678 : * We always use create_projection_path here, even if the subpath is
2679 : * projection-capable, so as to avoid modifying the subpath in place.
2680 : * It seems unlikely at present that there could be any other
2681 : * references to the subpath, but better safe than sorry.
2682 : *
2683 : * Note that we don't change the parallel path's cost estimates; it
2684 : * might be appropriate to do so, to reflect the fact that the bulk of
2685 : * the target evaluation will happen in workers.
2686 : */
2687 12 : if (IsA(path, GatherPath))
2688 : {
2689 0 : GatherPath *gpath = (GatherPath *) path;
2690 :
2691 0 : gpath->subpath = (Path *)
2692 0 : create_projection_path(root,
2693 0 : gpath->subpath->parent,
2694 : gpath->subpath,
2695 : target);
2696 : }
2697 : else
2698 : {
2699 12 : GatherMergePath *gmpath = (GatherMergePath *) path;
2700 :
2701 12 : gmpath->subpath = (Path *)
2702 12 : create_projection_path(root,
2703 12 : gmpath->subpath->parent,
2704 : gmpath->subpath,
2705 : target);
2706 : }
2707 : }
2708 6258 : else if (path->parallel_safe &&
2709 2383 : !is_parallel_safe(root, (Node *) target->exprs))
2710 : {
2711 : /*
2712 : * We're inserting a parallel-restricted target list into a path
2713 : * currently marked parallel-safe, so we have to mark it as no longer
2714 : * safe.
2715 : */
2716 6 : path->parallel_safe = false;
2717 : }
2718 :
2719 6270 : return path;
2720 : }
2721 :
2722 : /*
2723 : * create_set_projection_path
2724 : * Creates a pathnode that represents performing a projection that
2725 : * includes set-returning functions.
2726 : *
2727 : * 'rel' is the parent relation associated with the result
2728 : * 'subpath' is the path representing the source of data
2729 : * 'target' is the PathTarget to be computed
2730 : */
2731 : ProjectSetPath *
2732 6098 : create_set_projection_path(PlannerInfo *root,
2733 : RelOptInfo *rel,
2734 : Path *subpath,
2735 : PathTarget *target)
2736 : {
2737 6098 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2738 : double tlist_rows;
2739 : ListCell *lc;
2740 :
2741 6098 : pathnode->path.pathtype = T_ProjectSet;
2742 6098 : pathnode->path.parent = rel;
2743 6098 : pathnode->path.pathtarget = target;
2744 : /* For now, assume we are above any joins, so no parameterization */
2745 6098 : pathnode->path.param_info = NULL;
2746 6098 : pathnode->path.parallel_aware = false;
2747 14495 : pathnode->path.parallel_safe = rel->consider_parallel &&
2748 8379 : subpath->parallel_safe &&
2749 2281 : is_parallel_safe(root, (Node *) target->exprs);
2750 6098 : pathnode->path.parallel_workers = subpath->parallel_workers;
2751 : /* Projection does not change the sort order XXX? */
2752 6098 : pathnode->path.pathkeys = subpath->pathkeys;
2753 :
2754 6098 : pathnode->subpath = subpath;
2755 :
2756 : /*
2757 : * Estimate number of rows produced by SRFs for each row of input; if
2758 : * there's more than one in this node, use the maximum.
2759 : */
2760 6098 : tlist_rows = 1;
2761 13230 : foreach(lc, target->exprs)
2762 : {
2763 7132 : Node *node = (Node *) lfirst(lc);
2764 : double itemrows;
2765 :
2766 7132 : itemrows = expression_returns_set_rows(root, node);
2767 7132 : if (tlist_rows < itemrows)
2768 5869 : tlist_rows = itemrows;
2769 : }
2770 :
2771 : /*
2772 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2773 : * per input row, and half of cpu_tuple_cost for each added output row.
2774 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2775 : * this estimate later.
2776 : */
2777 6098 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2778 6098 : pathnode->path.rows = subpath->rows * tlist_rows;
2779 6098 : pathnode->path.startup_cost = subpath->startup_cost +
2780 6098 : target->cost.startup;
2781 6098 : pathnode->path.total_cost = subpath->total_cost +
2782 6098 : target->cost.startup +
2783 6098 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2784 6098 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2785 :
2786 6098 : return pathnode;
2787 : }
2788 :
2789 : /*
2790 : * create_incremental_sort_path
2791 : * Creates a pathnode that represents performing an incremental sort.
2792 : *
2793 : * 'rel' is the parent relation associated with the result
2794 : * 'subpath' is the path representing the source of data
2795 : * 'pathkeys' represents the desired sort order
2796 : * 'presorted_keys' is the number of keys by which the input path is
2797 : * already sorted
2798 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2799 : * or -1 if no LIMIT or couldn't estimate
2800 : */
2801 : IncrementalSortPath *
2802 5393 : create_incremental_sort_path(PlannerInfo *root,
2803 : RelOptInfo *rel,
2804 : Path *subpath,
2805 : List *pathkeys,
2806 : int presorted_keys,
2807 : double limit_tuples)
2808 : {
2809 5393 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
2810 5393 : SortPath *pathnode = &sort->spath;
2811 :
2812 5393 : pathnode->path.pathtype = T_IncrementalSort;
2813 5393 : pathnode->path.parent = rel;
2814 : /* Sort doesn't project, so use source path's pathtarget */
2815 5393 : pathnode->path.pathtarget = subpath->pathtarget;
2816 5393 : pathnode->path.param_info = subpath->param_info;
2817 5393 : pathnode->path.parallel_aware = false;
2818 8299 : pathnode->path.parallel_safe = rel->consider_parallel &&
2819 2906 : subpath->parallel_safe;
2820 5393 : pathnode->path.parallel_workers = subpath->parallel_workers;
2821 5393 : pathnode->path.pathkeys = pathkeys;
2822 :
2823 5393 : pathnode->subpath = subpath;
2824 :
2825 5393 : cost_incremental_sort(&pathnode->path,
2826 : root, pathkeys, presorted_keys,
2827 : subpath->disabled_nodes,
2828 : subpath->startup_cost,
2829 : subpath->total_cost,
2830 : subpath->rows,
2831 5393 : subpath->pathtarget->width,
2832 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2833 : work_mem, limit_tuples);
2834 :
2835 5393 : sort->nPresortedCols = presorted_keys;
2836 :
2837 5393 : return sort;
2838 : }
2839 :
2840 : /*
2841 : * create_sort_path
2842 : * Creates a pathnode that represents performing an explicit sort.
2843 : *
2844 : * 'rel' is the parent relation associated with the result
2845 : * 'subpath' is the path representing the source of data
2846 : * 'pathkeys' represents the desired sort order
2847 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2848 : * or -1 if no LIMIT or couldn't estimate
2849 : */
2850 : SortPath *
2851 59479 : create_sort_path(PlannerInfo *root,
2852 : RelOptInfo *rel,
2853 : Path *subpath,
2854 : List *pathkeys,
2855 : double limit_tuples)
2856 : {
2857 59479 : SortPath *pathnode = makeNode(SortPath);
2858 :
2859 59479 : pathnode->path.pathtype = T_Sort;
2860 59479 : pathnode->path.parent = rel;
2861 : /* Sort doesn't project, so use source path's pathtarget */
2862 59479 : pathnode->path.pathtarget = subpath->pathtarget;
2863 59479 : pathnode->path.param_info = subpath->param_info;
2864 59479 : pathnode->path.parallel_aware = false;
2865 103802 : pathnode->path.parallel_safe = rel->consider_parallel &&
2866 44323 : subpath->parallel_safe;
2867 59479 : pathnode->path.parallel_workers = subpath->parallel_workers;
2868 59479 : pathnode->path.pathkeys = pathkeys;
2869 :
2870 59479 : pathnode->subpath = subpath;
2871 :
2872 59479 : cost_sort(&pathnode->path, root, pathkeys,
2873 : subpath->disabled_nodes,
2874 : subpath->total_cost,
2875 : subpath->rows,
2876 59479 : subpath->pathtarget->width,
2877 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2878 : work_mem, limit_tuples);
2879 :
2880 59479 : return pathnode;
2881 : }
2882 :
2883 : /*
2884 : * create_group_path
2885 : * Creates a pathnode that represents performing grouping of presorted input
2886 : *
2887 : * 'rel' is the parent relation associated with the result
2888 : * 'subpath' is the path representing the source of data
2889 : * 'target' is the PathTarget to be computed
2890 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2891 : * 'qual' is the HAVING quals if any
2892 : * 'numGroups' is the estimated number of groups
2893 : */
2894 : GroupPath *
2895 613 : create_group_path(PlannerInfo *root,
2896 : RelOptInfo *rel,
2897 : Path *subpath,
2898 : List *groupClause,
2899 : List *qual,
2900 : double numGroups)
2901 : {
2902 613 : GroupPath *pathnode = makeNode(GroupPath);
2903 613 : PathTarget *target = rel->reltarget;
2904 :
2905 613 : pathnode->path.pathtype = T_Group;
2906 613 : pathnode->path.parent = rel;
2907 613 : pathnode->path.pathtarget = target;
2908 : /* For now, assume we are above any joins, so no parameterization */
2909 613 : pathnode->path.param_info = NULL;
2910 613 : pathnode->path.parallel_aware = false;
2911 985 : pathnode->path.parallel_safe = rel->consider_parallel &&
2912 372 : subpath->parallel_safe;
2913 613 : pathnode->path.parallel_workers = subpath->parallel_workers;
2914 : /* Group doesn't change sort ordering */
2915 613 : pathnode->path.pathkeys = subpath->pathkeys;
2916 :
2917 613 : pathnode->subpath = subpath;
2918 :
2919 613 : pathnode->groupClause = groupClause;
2920 613 : pathnode->qual = qual;
2921 :
2922 613 : cost_group(&pathnode->path, root,
2923 : list_length(groupClause),
2924 : numGroups,
2925 : qual,
2926 : subpath->disabled_nodes,
2927 : subpath->startup_cost, subpath->total_cost,
2928 : subpath->rows);
2929 :
2930 : /* add tlist eval cost for each output row */
2931 613 : pathnode->path.startup_cost += target->cost.startup;
2932 613 : pathnode->path.total_cost += target->cost.startup +
2933 613 : target->cost.per_tuple * pathnode->path.rows;
2934 :
2935 613 : return pathnode;
2936 : }
2937 :
2938 : /*
2939 : * create_unique_path
2940 : * Creates a pathnode that represents performing an explicit Unique step
2941 : * on presorted input.
2942 : *
2943 : * 'rel' is the parent relation associated with the result
2944 : * 'subpath' is the path representing the source of data
2945 : * 'numCols' is the number of grouping columns
2946 : * 'numGroups' is the estimated number of groups
2947 : *
2948 : * The input path must be sorted on the grouping columns, plus possibly
2949 : * additional columns; so the first numCols pathkeys are the grouping columns
2950 : */
2951 : UniquePath *
2952 11434 : create_unique_path(PlannerInfo *root,
2953 : RelOptInfo *rel,
2954 : Path *subpath,
2955 : int numCols,
2956 : double numGroups)
2957 : {
2958 11434 : UniquePath *pathnode = makeNode(UniquePath);
2959 :
2960 11434 : pathnode->path.pathtype = T_Unique;
2961 11434 : pathnode->path.parent = rel;
2962 : /* Unique doesn't project, so use source path's pathtarget */
2963 11434 : pathnode->path.pathtarget = subpath->pathtarget;
2964 11434 : pathnode->path.param_info = subpath->param_info;
2965 11434 : pathnode->path.parallel_aware = false;
2966 20694 : pathnode->path.parallel_safe = rel->consider_parallel &&
2967 9260 : subpath->parallel_safe;
2968 11434 : pathnode->path.parallel_workers = subpath->parallel_workers;
2969 : /* Unique doesn't change the input ordering */
2970 11434 : pathnode->path.pathkeys = subpath->pathkeys;
2971 :
2972 11434 : pathnode->subpath = subpath;
2973 11434 : pathnode->numkeys = numCols;
2974 :
2975 : /*
2976 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
2977 : * all columns get compared at most of the tuples. (XXX probably this is
2978 : * an overestimate.)
2979 : */
2980 11434 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2981 11434 : pathnode->path.startup_cost = subpath->startup_cost;
2982 11434 : pathnode->path.total_cost = subpath->total_cost +
2983 11434 : cpu_operator_cost * subpath->rows * numCols;
2984 11434 : pathnode->path.rows = numGroups;
2985 :
2986 11434 : return pathnode;
2987 : }
2988 :
2989 : /*
2990 : * create_agg_path
2991 : * Creates a pathnode that represents performing aggregation/grouping
2992 : *
2993 : * 'rel' is the parent relation associated with the result
2994 : * 'subpath' is the path representing the source of data
2995 : * 'target' is the PathTarget to be computed
2996 : * 'aggstrategy' is the Agg node's basic implementation strategy
2997 : * 'aggsplit' is the Agg node's aggregate-splitting mode
2998 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2999 : * 'qual' is the HAVING quals if any
3000 : * 'aggcosts' contains cost info about the aggregate functions to be computed
3001 : * 'numGroups' is the estimated number of groups (1 if not grouping)
3002 : */
3003 : AggPath *
3004 44743 : create_agg_path(PlannerInfo *root,
3005 : RelOptInfo *rel,
3006 : Path *subpath,
3007 : PathTarget *target,
3008 : AggStrategy aggstrategy,
3009 : AggSplit aggsplit,
3010 : List *groupClause,
3011 : List *qual,
3012 : const AggClauseCosts *aggcosts,
3013 : double numGroups)
3014 : {
3015 44743 : AggPath *pathnode = makeNode(AggPath);
3016 :
3017 44743 : pathnode->path.pathtype = T_Agg;
3018 44743 : pathnode->path.parent = rel;
3019 44743 : pathnode->path.pathtarget = target;
3020 44743 : pathnode->path.param_info = subpath->param_info;
3021 44743 : pathnode->path.parallel_aware = false;
3022 76244 : pathnode->path.parallel_safe = rel->consider_parallel &&
3023 31501 : subpath->parallel_safe;
3024 44743 : pathnode->path.parallel_workers = subpath->parallel_workers;
3025 :
3026 44743 : if (aggstrategy == AGG_SORTED)
3027 : {
3028 : /*
3029 : * Attempt to preserve the order of the subpath. Additional pathkeys
3030 : * may have been added in adjust_group_pathkeys_for_groupagg() to
3031 : * support ORDER BY / DISTINCT aggregates. Pathkeys added there
3032 : * belong to columns within the aggregate function, so we must strip
3033 : * these additional pathkeys off as those columns are unavailable
3034 : * above the aggregate node.
3035 : */
3036 7266 : if (list_length(subpath->pathkeys) > root->num_groupby_pathkeys)
3037 404 : pathnode->path.pathkeys = list_copy_head(subpath->pathkeys,
3038 : root->num_groupby_pathkeys);
3039 : else
3040 6862 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3041 : }
3042 : else
3043 37477 : pathnode->path.pathkeys = NIL; /* output is unordered */
3044 :
3045 44743 : pathnode->subpath = subpath;
3046 :
3047 44743 : pathnode->aggstrategy = aggstrategy;
3048 44743 : pathnode->aggsplit = aggsplit;
3049 44743 : pathnode->numGroups = numGroups;
3050 44743 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
3051 44743 : pathnode->groupClause = groupClause;
3052 44743 : pathnode->qual = qual;
3053 :
3054 44743 : cost_agg(&pathnode->path, root,
3055 : aggstrategy, aggcosts,
3056 : list_length(groupClause), numGroups,
3057 : qual,
3058 : subpath->disabled_nodes,
3059 : subpath->startup_cost, subpath->total_cost,
3060 44743 : subpath->rows, subpath->pathtarget->width);
3061 :
3062 : /* add tlist eval cost for each output row */
3063 44743 : pathnode->path.startup_cost += target->cost.startup;
3064 44743 : pathnode->path.total_cost += target->cost.startup +
3065 44743 : target->cost.per_tuple * pathnode->path.rows;
3066 :
3067 44743 : return pathnode;
3068 : }
3069 :
3070 : /*
3071 : * create_groupingsets_path
3072 : * Creates a pathnode that represents performing GROUPING SETS aggregation
3073 : *
3074 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3075 : * The input path's result must be sorted to match the last entry in
3076 : * rollup_groupclauses.
3077 : *
3078 : * 'rel' is the parent relation associated with the result
3079 : * 'subpath' is the path representing the source of data
3080 : * 'target' is the PathTarget to be computed
3081 : * 'having_qual' is the HAVING quals if any
3082 : * 'rollups' is a list of RollupData nodes
3083 : * 'agg_costs' contains cost info about the aggregate functions to be computed
3084 : */
3085 : GroupingSetsPath *
3086 1203 : create_groupingsets_path(PlannerInfo *root,
3087 : RelOptInfo *rel,
3088 : Path *subpath,
3089 : List *having_qual,
3090 : AggStrategy aggstrategy,
3091 : List *rollups,
3092 : const AggClauseCosts *agg_costs)
3093 : {
3094 1203 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
3095 1203 : PathTarget *target = rel->reltarget;
3096 : ListCell *lc;
3097 1203 : bool is_first = true;
3098 1203 : bool is_first_sort = true;
3099 :
3100 : /* The topmost generated Plan node will be an Agg */
3101 1203 : pathnode->path.pathtype = T_Agg;
3102 1203 : pathnode->path.parent = rel;
3103 1203 : pathnode->path.pathtarget = target;
3104 1203 : pathnode->path.param_info = subpath->param_info;
3105 1203 : pathnode->path.parallel_aware = false;
3106 1776 : pathnode->path.parallel_safe = rel->consider_parallel &&
3107 573 : subpath->parallel_safe;
3108 1203 : pathnode->path.parallel_workers = subpath->parallel_workers;
3109 1203 : pathnode->subpath = subpath;
3110 :
3111 : /*
3112 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3113 : * to AGG_HASHED, here if possible.
3114 : */
3115 1719 : if (aggstrategy == AGG_SORTED &&
3116 516 : list_length(rollups) == 1 &&
3117 267 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3118 30 : aggstrategy = AGG_PLAIN;
3119 :
3120 1720 : if (aggstrategy == AGG_MIXED &&
3121 517 : list_length(rollups) == 1)
3122 0 : aggstrategy = AGG_HASHED;
3123 :
3124 : /*
3125 : * Output will be in sorted order by group_pathkeys if, and only if, there
3126 : * is a single rollup operation on a non-empty list of grouping
3127 : * expressions.
3128 : */
3129 1203 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3130 237 : pathnode->path.pathkeys = root->group_pathkeys;
3131 : else
3132 966 : pathnode->path.pathkeys = NIL;
3133 :
3134 1203 : pathnode->aggstrategy = aggstrategy;
3135 1203 : pathnode->rollups = rollups;
3136 1203 : pathnode->qual = having_qual;
3137 1203 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3138 :
3139 : Assert(rollups != NIL);
3140 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3141 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3142 :
3143 4111 : foreach(lc, rollups)
3144 : {
3145 2908 : RollupData *rollup = lfirst(lc);
3146 2908 : List *gsets = rollup->gsets;
3147 2908 : int numGroupCols = list_length(linitial(gsets));
3148 :
3149 : /*
3150 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3151 : * (already-sorted) input, and following ones do their own sort.
3152 : *
3153 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3154 : *
3155 : * In AGG_MIXED mode, the first rollups are hashed, the first
3156 : * non-hashed one takes the (already-sorted) input, and following ones
3157 : * do their own sort.
3158 : */
3159 2908 : if (is_first)
3160 : {
3161 1203 : cost_agg(&pathnode->path, root,
3162 : aggstrategy,
3163 : agg_costs,
3164 : numGroupCols,
3165 : rollup->numGroups,
3166 : having_qual,
3167 : subpath->disabled_nodes,
3168 : subpath->startup_cost,
3169 : subpath->total_cost,
3170 : subpath->rows,
3171 1203 : subpath->pathtarget->width);
3172 1203 : is_first = false;
3173 1203 : if (!rollup->is_hashed)
3174 516 : is_first_sort = false;
3175 : }
3176 : else
3177 : {
3178 : Path sort_path; /* dummy for result of cost_sort */
3179 : Path agg_path; /* dummy for result of cost_agg */
3180 :
3181 1705 : if (rollup->is_hashed || is_first_sort)
3182 : {
3183 : /*
3184 : * Account for cost of aggregation, but don't charge input
3185 : * cost again
3186 : */
3187 1306 : cost_agg(&agg_path, root,
3188 1306 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3189 : agg_costs,
3190 : numGroupCols,
3191 : rollup->numGroups,
3192 : having_qual,
3193 : 0, 0.0, 0.0,
3194 : subpath->rows,
3195 1306 : subpath->pathtarget->width);
3196 1306 : if (!rollup->is_hashed)
3197 517 : is_first_sort = false;
3198 : }
3199 : else
3200 : {
3201 : /* Account for cost of sort, but don't charge input cost again */
3202 399 : cost_sort(&sort_path, root, NIL, 0,
3203 : 0.0,
3204 : subpath->rows,
3205 399 : subpath->pathtarget->width,
3206 : 0.0,
3207 : work_mem,
3208 : -1.0);
3209 :
3210 : /* Account for cost of aggregation */
3211 :
3212 399 : cost_agg(&agg_path, root,
3213 : AGG_SORTED,
3214 : agg_costs,
3215 : numGroupCols,
3216 : rollup->numGroups,
3217 : having_qual,
3218 : sort_path.disabled_nodes,
3219 : sort_path.startup_cost,
3220 : sort_path.total_cost,
3221 : sort_path.rows,
3222 399 : subpath->pathtarget->width);
3223 : }
3224 :
3225 1705 : pathnode->path.disabled_nodes += agg_path.disabled_nodes;
3226 1705 : pathnode->path.total_cost += agg_path.total_cost;
3227 1705 : pathnode->path.rows += agg_path.rows;
3228 : }
3229 : }
3230 :
3231 : /* add tlist eval cost for each output row */
3232 1203 : pathnode->path.startup_cost += target->cost.startup;
3233 1203 : pathnode->path.total_cost += target->cost.startup +
3234 1203 : target->cost.per_tuple * pathnode->path.rows;
3235 :
3236 1203 : return pathnode;
3237 : }
3238 :
3239 : /*
3240 : * create_minmaxagg_path
3241 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3242 : *
3243 : * 'rel' is the parent relation associated with the result
3244 : * 'target' is the PathTarget to be computed
3245 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3246 : * 'quals' is the HAVING quals if any
3247 : */
3248 : MinMaxAggPath *
3249 211 : create_minmaxagg_path(PlannerInfo *root,
3250 : RelOptInfo *rel,
3251 : PathTarget *target,
3252 : List *mmaggregates,
3253 : List *quals)
3254 : {
3255 211 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3256 : Cost initplan_cost;
3257 211 : int initplan_disabled_nodes = 0;
3258 : ListCell *lc;
3259 :
3260 : /* The topmost generated Plan node will be a Result */
3261 211 : pathnode->path.pathtype = T_Result;
3262 211 : pathnode->path.parent = rel;
3263 211 : pathnode->path.pathtarget = target;
3264 : /* For now, assume we are above any joins, so no parameterization */
3265 211 : pathnode->path.param_info = NULL;
3266 211 : pathnode->path.parallel_aware = false;
3267 211 : pathnode->path.parallel_safe = true; /* might change below */
3268 211 : pathnode->path.parallel_workers = 0;
3269 : /* Result is one unordered row */
3270 211 : pathnode->path.rows = 1;
3271 211 : pathnode->path.pathkeys = NIL;
3272 :
3273 211 : pathnode->mmaggregates = mmaggregates;
3274 211 : pathnode->quals = quals;
3275 :
3276 : /* Calculate cost of all the initplans, and check parallel safety */
3277 211 : initplan_cost = 0;
3278 440 : foreach(lc, mmaggregates)
3279 : {
3280 229 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3281 :
3282 229 : initplan_disabled_nodes += mminfo->path->disabled_nodes;
3283 229 : initplan_cost += mminfo->pathcost;
3284 229 : if (!mminfo->path->parallel_safe)
3285 55 : pathnode->path.parallel_safe = false;
3286 : }
3287 :
3288 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3289 211 : pathnode->path.disabled_nodes = initplan_disabled_nodes;
3290 211 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3291 211 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3292 211 : target->cost.per_tuple + cpu_tuple_cost;
3293 :
3294 : /*
3295 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3296 : * rowcount estimate should be 1 no matter what the qual is.
3297 : */
3298 211 : if (quals)
3299 : {
3300 : QualCost qual_cost;
3301 :
3302 0 : cost_qual_eval(&qual_cost, quals, root);
3303 0 : pathnode->path.startup_cost += qual_cost.startup;
3304 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3305 : }
3306 :
3307 : /*
3308 : * If the initplans were all parallel-safe, also check safety of the
3309 : * target and quals. (The Result node itself isn't parallelizable, but if
3310 : * we are in a subquery then it can be useful for the outer query to know
3311 : * that this one is parallel-safe.)
3312 : */
3313 211 : if (pathnode->path.parallel_safe)
3314 156 : pathnode->path.parallel_safe =
3315 312 : is_parallel_safe(root, (Node *) target->exprs) &&
3316 312 : is_parallel_safe(root, (Node *) quals);
3317 :
3318 211 : return pathnode;
3319 : }
3320 :
3321 : /*
3322 : * create_windowagg_path
3323 : * Creates a pathnode that represents computation of window functions
3324 : *
3325 : * 'rel' is the parent relation associated with the result
3326 : * 'subpath' is the path representing the source of data
3327 : * 'target' is the PathTarget to be computed
3328 : * 'windowFuncs' is a list of WindowFunc structs
3329 : * 'runCondition' is a list of OpExprs to short-circuit WindowAgg execution
3330 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3331 : * 'qual' WindowClause.runconditions from lower-level WindowAggPaths.
3332 : * Must always be NIL when topwindow == false
3333 : * 'topwindow' pass as true only for the top-level WindowAgg. False for all
3334 : * intermediate WindowAggs.
3335 : *
3336 : * The input must be sorted according to the WindowClause's PARTITION keys
3337 : * plus ORDER BY keys.
3338 : */
3339 : WindowAggPath *
3340 1538 : create_windowagg_path(PlannerInfo *root,
3341 : RelOptInfo *rel,
3342 : Path *subpath,
3343 : PathTarget *target,
3344 : List *windowFuncs,
3345 : List *runCondition,
3346 : WindowClause *winclause,
3347 : List *qual,
3348 : bool topwindow)
3349 : {
3350 1538 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3351 :
3352 : /* qual can only be set for the topwindow */
3353 : Assert(qual == NIL || topwindow);
3354 :
3355 1538 : pathnode->path.pathtype = T_WindowAgg;
3356 1538 : pathnode->path.parent = rel;
3357 1538 : pathnode->path.pathtarget = target;
3358 : /* For now, assume we are above any joins, so no parameterization */
3359 1538 : pathnode->path.param_info = NULL;
3360 1538 : pathnode->path.parallel_aware = false;
3361 1538 : pathnode->path.parallel_safe = rel->consider_parallel &&
3362 0 : subpath->parallel_safe;
3363 1538 : pathnode->path.parallel_workers = subpath->parallel_workers;
3364 : /* WindowAgg preserves the input sort order */
3365 1538 : pathnode->path.pathkeys = subpath->pathkeys;
3366 :
3367 1538 : pathnode->subpath = subpath;
3368 1538 : pathnode->winclause = winclause;
3369 1538 : pathnode->qual = qual;
3370 1538 : pathnode->runCondition = runCondition;
3371 1538 : pathnode->topwindow = topwindow;
3372 :
3373 : /*
3374 : * For costing purposes, assume that there are no redundant partitioning
3375 : * or ordering columns; it's not worth the trouble to deal with that
3376 : * corner case here. So we just pass the unmodified list lengths to
3377 : * cost_windowagg.
3378 : */
3379 1538 : cost_windowagg(&pathnode->path, root,
3380 : windowFuncs,
3381 : winclause,
3382 : subpath->disabled_nodes,
3383 : subpath->startup_cost,
3384 : subpath->total_cost,
3385 : subpath->rows);
3386 :
3387 : /* add tlist eval cost for each output row */
3388 1538 : pathnode->path.startup_cost += target->cost.startup;
3389 1538 : pathnode->path.total_cost += target->cost.startup +
3390 1538 : target->cost.per_tuple * pathnode->path.rows;
3391 :
3392 1538 : return pathnode;
3393 : }
3394 :
3395 : /*
3396 : * create_setop_path
3397 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3398 : *
3399 : * 'rel' is the parent relation associated with the result
3400 : * 'leftpath' is the path representing the left-hand source of data
3401 : * 'rightpath' is the path representing the right-hand source of data
3402 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3403 : * 'strategy' is the implementation strategy (sorted or hashed)
3404 : * 'groupList' is a list of SortGroupClause's representing the grouping
3405 : * 'numGroups' is the estimated number of distinct groups in left-hand input
3406 : * 'outputRows' is the estimated number of output rows
3407 : *
3408 : * leftpath and rightpath must produce the same columns. Moreover, if
3409 : * strategy is SETOP_SORTED, leftpath and rightpath must both be sorted
3410 : * by all the grouping columns.
3411 : */
3412 : SetOpPath *
3413 650 : create_setop_path(PlannerInfo *root,
3414 : RelOptInfo *rel,
3415 : Path *leftpath,
3416 : Path *rightpath,
3417 : SetOpCmd cmd,
3418 : SetOpStrategy strategy,
3419 : List *groupList,
3420 : double numGroups,
3421 : double outputRows)
3422 : {
3423 650 : SetOpPath *pathnode = makeNode(SetOpPath);
3424 :
3425 650 : pathnode->path.pathtype = T_SetOp;
3426 650 : pathnode->path.parent = rel;
3427 650 : pathnode->path.pathtarget = rel->reltarget;
3428 : /* For now, assume we are above any joins, so no parameterization */
3429 650 : pathnode->path.param_info = NULL;
3430 650 : pathnode->path.parallel_aware = false;
3431 1300 : pathnode->path.parallel_safe = rel->consider_parallel &&
3432 650 : leftpath->parallel_safe && rightpath->parallel_safe;
3433 650 : pathnode->path.parallel_workers =
3434 650 : leftpath->parallel_workers + rightpath->parallel_workers;
3435 : /* SetOp preserves the input sort order if in sort mode */
3436 650 : pathnode->path.pathkeys =
3437 650 : (strategy == SETOP_SORTED) ? leftpath->pathkeys : NIL;
3438 :
3439 650 : pathnode->leftpath = leftpath;
3440 650 : pathnode->rightpath = rightpath;
3441 650 : pathnode->cmd = cmd;
3442 650 : pathnode->strategy = strategy;
3443 650 : pathnode->groupList = groupList;
3444 650 : pathnode->numGroups = numGroups;
3445 :
3446 : /*
3447 : * Compute cost estimates. As things stand, we end up with the same total
3448 : * cost in this node for sort and hash methods, but different startup
3449 : * costs. This could be refined perhaps, but it'll do for now.
3450 : */
3451 650 : pathnode->path.disabled_nodes =
3452 650 : leftpath->disabled_nodes + rightpath->disabled_nodes;
3453 650 : if (strategy == SETOP_SORTED)
3454 : {
3455 : /*
3456 : * In sorted mode, we can emit output incrementally. Charge one
3457 : * cpu_operator_cost per comparison per input tuple. Like cost_group,
3458 : * we assume all columns get compared at most of the tuples.
3459 : */
3460 340 : pathnode->path.startup_cost =
3461 340 : leftpath->startup_cost + rightpath->startup_cost;
3462 340 : pathnode->path.total_cost =
3463 680 : leftpath->total_cost + rightpath->total_cost +
3464 340 : cpu_operator_cost * (leftpath->rows + rightpath->rows) * list_length(groupList);
3465 :
3466 : /*
3467 : * Also charge a small amount per extracted tuple. Like cost_sort,
3468 : * charge only operator cost not cpu_tuple_cost, since SetOp does no
3469 : * qual-checking or projection.
3470 : */
3471 340 : pathnode->path.total_cost += cpu_operator_cost * outputRows;
3472 : }
3473 : else
3474 : {
3475 : Size hashtablesize;
3476 :
3477 : /*
3478 : * In hashed mode, we must read all the input before we can emit
3479 : * anything. Also charge comparison costs to represent the cost of
3480 : * hash table lookups.
3481 : */
3482 310 : pathnode->path.startup_cost =
3483 620 : leftpath->total_cost + rightpath->total_cost +
3484 310 : cpu_operator_cost * (leftpath->rows + rightpath->rows) * list_length(groupList);
3485 310 : pathnode->path.total_cost = pathnode->path.startup_cost;
3486 :
3487 : /*
3488 : * Also charge a small amount per extracted tuple. Like cost_sort,
3489 : * charge only operator cost not cpu_tuple_cost, since SetOp does no
3490 : * qual-checking or projection.
3491 : */
3492 310 : pathnode->path.total_cost += cpu_operator_cost * outputRows;
3493 :
3494 : /*
3495 : * Mark the path as disabled if enable_hashagg is off. While this
3496 : * isn't exactly a HashAgg node, it seems close enough to justify
3497 : * letting that switch control it.
3498 : */
3499 310 : if (!enable_hashagg)
3500 57 : pathnode->path.disabled_nodes++;
3501 :
3502 : /*
3503 : * Also disable if it doesn't look like the hashtable will fit into
3504 : * hash_mem. (Note: reject on equality, to ensure that an estimate of
3505 : * SIZE_MAX disables hashing regardless of the hash_mem limit.)
3506 : */
3507 310 : hashtablesize = EstimateSetOpHashTableSpace(numGroups,
3508 310 : leftpath->pathtarget->width);
3509 310 : if (hashtablesize >= get_hash_memory_limit())
3510 0 : pathnode->path.disabled_nodes++;
3511 : }
3512 650 : pathnode->path.rows = outputRows;
3513 :
3514 650 : return pathnode;
3515 : }
3516 :
3517 : /*
3518 : * create_recursiveunion_path
3519 : * Creates a pathnode that represents a recursive UNION node
3520 : *
3521 : * 'rel' is the parent relation associated with the result
3522 : * 'leftpath' is the source of data for the non-recursive term
3523 : * 'rightpath' is the source of data for the recursive term
3524 : * 'target' is the PathTarget to be computed
3525 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3526 : * 'wtParam' is the ID of Param representing work table
3527 : * 'numGroups' is the estimated number of groups
3528 : *
3529 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3530 : */
3531 : RecursiveUnionPath *
3532 468 : create_recursiveunion_path(PlannerInfo *root,
3533 : RelOptInfo *rel,
3534 : Path *leftpath,
3535 : Path *rightpath,
3536 : PathTarget *target,
3537 : List *distinctList,
3538 : int wtParam,
3539 : double numGroups)
3540 : {
3541 468 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3542 :
3543 468 : pathnode->path.pathtype = T_RecursiveUnion;
3544 468 : pathnode->path.parent = rel;
3545 468 : pathnode->path.pathtarget = target;
3546 : /* For now, assume we are above any joins, so no parameterization */
3547 468 : pathnode->path.param_info = NULL;
3548 468 : pathnode->path.parallel_aware = false;
3549 936 : pathnode->path.parallel_safe = rel->consider_parallel &&
3550 468 : leftpath->parallel_safe && rightpath->parallel_safe;
3551 : /* Foolish, but we'll do it like joins for now: */
3552 468 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3553 : /* RecursiveUnion result is always unsorted */
3554 468 : pathnode->path.pathkeys = NIL;
3555 :
3556 468 : pathnode->leftpath = leftpath;
3557 468 : pathnode->rightpath = rightpath;
3558 468 : pathnode->distinctList = distinctList;
3559 468 : pathnode->wtParam = wtParam;
3560 468 : pathnode->numGroups = numGroups;
3561 :
3562 468 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3563 :
3564 468 : return pathnode;
3565 : }
3566 :
3567 : /*
3568 : * create_lockrows_path
3569 : * Creates a pathnode that represents acquiring row locks
3570 : *
3571 : * 'rel' is the parent relation associated with the result
3572 : * 'subpath' is the path representing the source of data
3573 : * 'rowMarks' is a list of PlanRowMark's
3574 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3575 : */
3576 : LockRowsPath *
3577 7030 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3578 : Path *subpath, List *rowMarks, int epqParam)
3579 : {
3580 7030 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3581 :
3582 7030 : pathnode->path.pathtype = T_LockRows;
3583 7030 : pathnode->path.parent = rel;
3584 : /* LockRows doesn't project, so use source path's pathtarget */
3585 7030 : pathnode->path.pathtarget = subpath->pathtarget;
3586 : /* For now, assume we are above any joins, so no parameterization */
3587 7030 : pathnode->path.param_info = NULL;
3588 7030 : pathnode->path.parallel_aware = false;
3589 7030 : pathnode->path.parallel_safe = false;
3590 7030 : pathnode->path.parallel_workers = 0;
3591 7030 : pathnode->path.rows = subpath->rows;
3592 :
3593 : /*
3594 : * The result cannot be assumed sorted, since locking might cause the sort
3595 : * key columns to be replaced with new values.
3596 : */
3597 7030 : pathnode->path.pathkeys = NIL;
3598 :
3599 7030 : pathnode->subpath = subpath;
3600 7030 : pathnode->rowMarks = rowMarks;
3601 7030 : pathnode->epqParam = epqParam;
3602 :
3603 : /*
3604 : * We should charge something extra for the costs of row locking and
3605 : * possible refetches, but it's hard to say how much. For now, use
3606 : * cpu_tuple_cost per row.
3607 : */
3608 7030 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3609 7030 : pathnode->path.startup_cost = subpath->startup_cost;
3610 7030 : pathnode->path.total_cost = subpath->total_cost +
3611 7030 : cpu_tuple_cost * subpath->rows;
3612 :
3613 7030 : return pathnode;
3614 : }
3615 :
3616 : /*
3617 : * create_modifytable_path
3618 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE/MERGE
3619 : * mods
3620 : *
3621 : * 'rel' is the parent relation associated with the result
3622 : * 'subpath' is a Path producing source data
3623 : * 'operation' is the operation type
3624 : * 'canSetTag' is true if we set the command tag/es_processed
3625 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3626 : * 'rootRelation' is the partitioned/inherited table root RTI, or 0 if none
3627 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3628 : * 'updateColnosLists' is a list of UPDATE target column number lists
3629 : * (one sublist per rel); or NIL if not an UPDATE
3630 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3631 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3632 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3633 : * 'onconflict' is the ON CONFLICT clause, or NULL
3634 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3635 : * 'mergeActionLists' is a list of lists of MERGE actions (one per rel)
3636 : * 'mergeJoinConditions' is a list of join conditions for MERGE (one per rel)
3637 : */
3638 : ModifyTablePath *
3639 46950 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3640 : Path *subpath,
3641 : CmdType operation, bool canSetTag,
3642 : Index nominalRelation, Index rootRelation,
3643 : List *resultRelations,
3644 : List *updateColnosLists,
3645 : List *withCheckOptionLists, List *returningLists,
3646 : List *rowMarks, OnConflictExpr *onconflict,
3647 : List *mergeActionLists, List *mergeJoinConditions,
3648 : int epqParam)
3649 : {
3650 46950 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3651 :
3652 : Assert(operation == CMD_MERGE ||
3653 : (operation == CMD_UPDATE ?
3654 : list_length(resultRelations) == list_length(updateColnosLists) :
3655 : updateColnosLists == NIL));
3656 : Assert(withCheckOptionLists == NIL ||
3657 : list_length(resultRelations) == list_length(withCheckOptionLists));
3658 : Assert(returningLists == NIL ||
3659 : list_length(resultRelations) == list_length(returningLists));
3660 :
3661 46950 : pathnode->path.pathtype = T_ModifyTable;
3662 46950 : pathnode->path.parent = rel;
3663 : /* pathtarget is not interesting, just make it minimally valid */
3664 46950 : pathnode->path.pathtarget = rel->reltarget;
3665 : /* For now, assume we are above any joins, so no parameterization */
3666 46950 : pathnode->path.param_info = NULL;
3667 46950 : pathnode->path.parallel_aware = false;
3668 46950 : pathnode->path.parallel_safe = false;
3669 46950 : pathnode->path.parallel_workers = 0;
3670 46950 : pathnode->path.pathkeys = NIL;
3671 :
3672 : /*
3673 : * Compute cost & rowcount as subpath cost & rowcount (if RETURNING)
3674 : *
3675 : * Currently, we don't charge anything extra for the actual table
3676 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3677 : * expressions if any. It would only be window dressing, since
3678 : * ModifyTable is always a top-level node and there is no way for the
3679 : * costs to change any higher-level planning choices. But we might want
3680 : * to make it look better sometime.
3681 : */
3682 46950 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3683 46950 : pathnode->path.startup_cost = subpath->startup_cost;
3684 46950 : pathnode->path.total_cost = subpath->total_cost;
3685 46950 : if (returningLists != NIL)
3686 : {
3687 1710 : pathnode->path.rows = subpath->rows;
3688 :
3689 : /*
3690 : * Set width to match the subpath output. XXX this is totally wrong:
3691 : * we should return an average of the RETURNING tlist widths. But
3692 : * it's what happened historically, and improving it is a task for
3693 : * another day. (Again, it's mostly window dressing.)
3694 : */
3695 1710 : pathnode->path.pathtarget->width = subpath->pathtarget->width;
3696 : }
3697 : else
3698 : {
3699 45240 : pathnode->path.rows = 0;
3700 45240 : pathnode->path.pathtarget->width = 0;
3701 : }
3702 :
3703 46950 : pathnode->subpath = subpath;
3704 46950 : pathnode->operation = operation;
3705 46950 : pathnode->canSetTag = canSetTag;
3706 46950 : pathnode->nominalRelation = nominalRelation;
3707 46950 : pathnode->rootRelation = rootRelation;
3708 46950 : pathnode->resultRelations = resultRelations;
3709 46950 : pathnode->updateColnosLists = updateColnosLists;
3710 46950 : pathnode->withCheckOptionLists = withCheckOptionLists;
3711 46950 : pathnode->returningLists = returningLists;
3712 46950 : pathnode->rowMarks = rowMarks;
3713 46950 : pathnode->onconflict = onconflict;
3714 46950 : pathnode->epqParam = epqParam;
3715 46950 : pathnode->mergeActionLists = mergeActionLists;
3716 46950 : pathnode->mergeJoinConditions = mergeJoinConditions;
3717 :
3718 46950 : return pathnode;
3719 : }
3720 :
3721 : /*
3722 : * create_limit_path
3723 : * Creates a pathnode that represents performing LIMIT/OFFSET
3724 : *
3725 : * In addition to providing the actual OFFSET and LIMIT expressions,
3726 : * the caller must provide estimates of their values for costing purposes.
3727 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3728 : * the clause not being present, and -1 means it's present but we could
3729 : * not estimate its value.
3730 : *
3731 : * 'rel' is the parent relation associated with the result
3732 : * 'subpath' is the path representing the source of data
3733 : * 'limitOffset' is the actual OFFSET expression, or NULL
3734 : * 'limitCount' is the actual LIMIT expression, or NULL
3735 : * 'offset_est' is the estimated value of the OFFSET expression
3736 : * 'count_est' is the estimated value of the LIMIT expression
3737 : */
3738 : LimitPath *
3739 3207 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3740 : Path *subpath,
3741 : Node *limitOffset, Node *limitCount,
3742 : LimitOption limitOption,
3743 : int64 offset_est, int64 count_est)
3744 : {
3745 3207 : LimitPath *pathnode = makeNode(LimitPath);
3746 :
3747 3207 : pathnode->path.pathtype = T_Limit;
3748 3207 : pathnode->path.parent = rel;
3749 : /* Limit doesn't project, so use source path's pathtarget */
3750 3207 : pathnode->path.pathtarget = subpath->pathtarget;
3751 : /* For now, assume we are above any joins, so no parameterization */
3752 3207 : pathnode->path.param_info = NULL;
3753 3207 : pathnode->path.parallel_aware = false;
3754 4439 : pathnode->path.parallel_safe = rel->consider_parallel &&
3755 1232 : subpath->parallel_safe;
3756 3207 : pathnode->path.parallel_workers = subpath->parallel_workers;
3757 3207 : pathnode->path.rows = subpath->rows;
3758 3207 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3759 3207 : pathnode->path.startup_cost = subpath->startup_cost;
3760 3207 : pathnode->path.total_cost = subpath->total_cost;
3761 3207 : pathnode->path.pathkeys = subpath->pathkeys;
3762 3207 : pathnode->subpath = subpath;
3763 3207 : pathnode->limitOffset = limitOffset;
3764 3207 : pathnode->limitCount = limitCount;
3765 3207 : pathnode->limitOption = limitOption;
3766 :
3767 : /*
3768 : * Adjust the output rows count and costs according to the offset/limit.
3769 : */
3770 3207 : adjust_limit_rows_costs(&pathnode->path.rows,
3771 : &pathnode->path.startup_cost,
3772 : &pathnode->path.total_cost,
3773 : offset_est, count_est);
3774 :
3775 3207 : return pathnode;
3776 : }
3777 :
3778 : /*
3779 : * adjust_limit_rows_costs
3780 : * Adjust the size and cost estimates for a LimitPath node according to the
3781 : * offset/limit.
3782 : *
3783 : * This is only a cosmetic issue if we are at top level, but if we are
3784 : * building a subquery then it's important to report correct info to the outer
3785 : * planner.
3786 : *
3787 : * When the offset or count couldn't be estimated, use 10% of the estimated
3788 : * number of rows emitted from the subpath.
3789 : *
3790 : * XXX we don't bother to add eval costs of the offset/limit expressions
3791 : * themselves to the path costs. In theory we should, but in most cases those
3792 : * expressions are trivial and it's just not worth the trouble.
3793 : */
3794 : void
3795 3299 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3796 : Cost *startup_cost, /* in/out parameter */
3797 : Cost *total_cost, /* in/out parameter */
3798 : int64 offset_est,
3799 : int64 count_est)
3800 : {
3801 3299 : double input_rows = *rows;
3802 3299 : Cost input_startup_cost = *startup_cost;
3803 3299 : Cost input_total_cost = *total_cost;
3804 :
3805 3299 : if (offset_est != 0)
3806 : {
3807 : double offset_rows;
3808 :
3809 356 : if (offset_est > 0)
3810 344 : offset_rows = (double) offset_est;
3811 : else
3812 12 : offset_rows = clamp_row_est(input_rows * 0.10);
3813 356 : if (offset_rows > *rows)
3814 19 : offset_rows = *rows;
3815 356 : if (input_rows > 0)
3816 356 : *startup_cost +=
3817 356 : (input_total_cost - input_startup_cost)
3818 356 : * offset_rows / input_rows;
3819 356 : *rows -= offset_rows;
3820 356 : if (*rows < 1)
3821 23 : *rows = 1;
3822 : }
3823 :
3824 3299 : if (count_est != 0)
3825 : {
3826 : double count_rows;
3827 :
3828 3261 : if (count_est > 0)
3829 3258 : count_rows = (double) count_est;
3830 : else
3831 3 : count_rows = clamp_row_est(input_rows * 0.10);
3832 3261 : if (count_rows > *rows)
3833 124 : count_rows = *rows;
3834 3261 : if (input_rows > 0)
3835 3261 : *total_cost = *startup_cost +
3836 3261 : (input_total_cost - input_startup_cost)
3837 3261 : * count_rows / input_rows;
3838 3261 : *rows = count_rows;
3839 3261 : if (*rows < 1)
3840 0 : *rows = 1;
3841 : }
3842 3299 : }
3843 :
3844 :
3845 : /*
3846 : * reparameterize_path
3847 : * Attempt to modify a Path to have greater parameterization
3848 : *
3849 : * We use this to attempt to bring all child paths of an appendrel to the
3850 : * same parameterization level, ensuring that they all enforce the same set
3851 : * of join quals (and thus that that parameterization can be attributed to
3852 : * an append path built from such paths). Currently, only a few path types
3853 : * are supported here, though more could be added at need. We return NULL
3854 : * if we can't reparameterize the given path.
3855 : *
3856 : * Note: we intentionally do not pass created paths to add_path(); it would
3857 : * possibly try to delete them on the grounds of being cost-inferior to the
3858 : * paths they were made from, and we don't want that. Paths made here are
3859 : * not necessarily of general-purpose usefulness, but they can be useful
3860 : * as members of an append path.
3861 : */
3862 : Path *
3863 178 : reparameterize_path(PlannerInfo *root, Path *path,
3864 : Relids required_outer,
3865 : double loop_count)
3866 : {
3867 178 : RelOptInfo *rel = path->parent;
3868 :
3869 : /* Can only increase, not decrease, path's parameterization */
3870 178 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
3871 0 : return NULL;
3872 178 : switch (path->pathtype)
3873 : {
3874 132 : case T_SeqScan:
3875 132 : return create_seqscan_path(root, rel, required_outer, 0);
3876 0 : case T_SampleScan:
3877 0 : return create_samplescan_path(root, rel, required_outer);
3878 0 : case T_IndexScan:
3879 : case T_IndexOnlyScan:
3880 : {
3881 0 : IndexPath *ipath = (IndexPath *) path;
3882 0 : IndexPath *newpath = makeNode(IndexPath);
3883 :
3884 : /*
3885 : * We can't use create_index_path directly, and would not want
3886 : * to because it would re-compute the indexqual conditions
3887 : * which is wasted effort. Instead we hack things a bit:
3888 : * flat-copy the path node, revise its param_info, and redo
3889 : * the cost estimate.
3890 : */
3891 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3892 0 : newpath->path.param_info =
3893 0 : get_baserel_parampathinfo(root, rel, required_outer);
3894 0 : cost_index(newpath, root, loop_count, false);
3895 0 : return (Path *) newpath;
3896 : }
3897 0 : case T_BitmapHeapScan:
3898 : {
3899 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3900 :
3901 0 : return (Path *) create_bitmap_heap_path(root,
3902 : rel,
3903 : bpath->bitmapqual,
3904 : required_outer,
3905 : loop_count, 0);
3906 : }
3907 0 : case T_SubqueryScan:
3908 : {
3909 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
3910 0 : Path *subpath = spath->subpath;
3911 : bool trivial_pathtarget;
3912 :
3913 : /*
3914 : * If existing node has zero extra cost, we must have decided
3915 : * its target is trivial. (The converse is not true, because
3916 : * it might have a trivial target but quals to enforce; but in
3917 : * that case the new node will too, so it doesn't matter
3918 : * whether we get the right answer here.)
3919 : */
3920 0 : trivial_pathtarget =
3921 0 : (subpath->total_cost == spath->path.total_cost);
3922 :
3923 0 : return (Path *) create_subqueryscan_path(root,
3924 : rel,
3925 : subpath,
3926 : trivial_pathtarget,
3927 : spath->path.pathkeys,
3928 : required_outer);
3929 : }
3930 30 : case T_Result:
3931 : /* Supported only for RTE_RESULT scan paths */
3932 30 : if (IsA(path, Path))
3933 30 : return create_resultscan_path(root, rel, required_outer);
3934 0 : break;
3935 0 : case T_Append:
3936 : {
3937 0 : AppendPath *apath = (AppendPath *) path;
3938 0 : AppendPathInput new_append = {0};
3939 : int i;
3940 : ListCell *lc;
3941 :
3942 0 : new_append.child_append_relid_sets = apath->child_append_relid_sets;
3943 :
3944 : /* Reparameterize the children */
3945 0 : i = 0;
3946 0 : foreach(lc, apath->subpaths)
3947 : {
3948 0 : Path *spath = (Path *) lfirst(lc);
3949 :
3950 0 : spath = reparameterize_path(root, spath,
3951 : required_outer,
3952 : loop_count);
3953 0 : if (spath == NULL)
3954 0 : return NULL;
3955 : /* We have to re-split the regular and partial paths */
3956 0 : if (i < apath->first_partial_path)
3957 0 : new_append.subpaths = lappend(new_append.subpaths, spath);
3958 : else
3959 0 : new_append.partial_subpaths = lappend(new_append.partial_subpaths, spath);
3960 0 : i++;
3961 : }
3962 0 : return (Path *)
3963 0 : create_append_path(root, rel, new_append,
3964 : apath->path.pathkeys, required_outer,
3965 : apath->path.parallel_workers,
3966 0 : apath->path.parallel_aware,
3967 : -1);
3968 : }
3969 0 : case T_Material:
3970 : {
3971 0 : MaterialPath *mpath = (MaterialPath *) path;
3972 0 : Path *spath = mpath->subpath;
3973 : bool enabled;
3974 :
3975 0 : spath = reparameterize_path(root, spath,
3976 : required_outer,
3977 : loop_count);
3978 0 : if (spath == NULL)
3979 0 : return NULL;
3980 0 : enabled =
3981 0 : (mpath->path.disabled_nodes <= spath->disabled_nodes);
3982 0 : return (Path *) create_material_path(rel, spath, enabled);
3983 : }
3984 0 : case T_Memoize:
3985 : {
3986 0 : MemoizePath *mpath = (MemoizePath *) path;
3987 0 : Path *spath = mpath->subpath;
3988 :
3989 0 : spath = reparameterize_path(root, spath,
3990 : required_outer,
3991 : loop_count);
3992 0 : if (spath == NULL)
3993 0 : return NULL;
3994 0 : return (Path *) create_memoize_path(root, rel,
3995 : spath,
3996 : mpath->param_exprs,
3997 : mpath->hash_operators,
3998 0 : mpath->singlerow,
3999 0 : mpath->binary_mode,
4000 : mpath->est_calls);
4001 : }
4002 16 : default:
4003 16 : break;
4004 : }
4005 16 : return NULL;
4006 : }
4007 :
4008 : /*
4009 : * reparameterize_path_by_child
4010 : * Given a path parameterized by the parent of the given child relation,
4011 : * translate the path to be parameterized by the given child relation.
4012 : *
4013 : * Most fields in the path are not changed, but any expressions must be
4014 : * adjusted to refer to the correct varnos, and any subpaths must be
4015 : * recursively reparameterized. Other fields that refer to specific relids
4016 : * also need adjustment.
4017 : *
4018 : * The cost, number of rows, width and parallel path properties depend upon
4019 : * path->parent, which does not change during the translation. So we need
4020 : * not change those.
4021 : *
4022 : * Currently, only a few path types are supported here, though more could be
4023 : * added at need. We return NULL if we can't reparameterize the given path.
4024 : *
4025 : * Note that this function can change referenced RangeTblEntries, RelOptInfos
4026 : * and IndexOptInfos as well as the Path structures. Therefore, it's only safe
4027 : * to call during create_plan(), when we have made a final choice of which Path
4028 : * to use for each RangeTblEntry/RelOptInfo/IndexOptInfo.
4029 : *
4030 : * Keep this code in sync with path_is_reparameterizable_by_child()!
4031 : */
4032 : Path *
4033 50929 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
4034 : RelOptInfo *child_rel)
4035 : {
4036 : Path *new_path;
4037 : ParamPathInfo *new_ppi;
4038 : ParamPathInfo *old_ppi;
4039 : Relids required_outer;
4040 :
4041 : #define ADJUST_CHILD_ATTRS(node) \
4042 : ((node) = (void *) adjust_appendrel_attrs_multilevel(root, \
4043 : (Node *) (node), \
4044 : child_rel, \
4045 : child_rel->top_parent))
4046 :
4047 : #define REPARAMETERIZE_CHILD_PATH(path) \
4048 : do { \
4049 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
4050 : if ((path) == NULL) \
4051 : return NULL; \
4052 : } while(0)
4053 :
4054 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
4055 : do { \
4056 : if ((pathlist) != NIL) \
4057 : { \
4058 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
4059 : child_rel); \
4060 : if ((pathlist) == NIL) \
4061 : return NULL; \
4062 : } \
4063 : } while(0)
4064 :
4065 : /*
4066 : * If the path is not parameterized by the parent of the given relation,
4067 : * it doesn't need reparameterization.
4068 : */
4069 50929 : if (!path->param_info ||
4070 25939 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4071 50428 : return path;
4072 :
4073 : /*
4074 : * If possible, reparameterize the given path.
4075 : *
4076 : * This function is currently only applied to the inner side of a nestloop
4077 : * join that is being partitioned by the partitionwise-join code. Hence,
4078 : * we need only support path types that plausibly arise in that context.
4079 : * (In particular, supporting sorted path types would be a waste of code
4080 : * and cycles: even if we translated them here, they'd just lose in
4081 : * subsequent cost comparisons.) If we do see an unsupported path type,
4082 : * that just means we won't be able to generate a partitionwise-join plan
4083 : * using that path type.
4084 : */
4085 501 : switch (nodeTag(path))
4086 : {
4087 114 : case T_Path:
4088 114 : new_path = path;
4089 114 : ADJUST_CHILD_ATTRS(new_path->parent->baserestrictinfo);
4090 114 : if (path->pathtype == T_SampleScan)
4091 : {
4092 24 : Index scan_relid = path->parent->relid;
4093 : RangeTblEntry *rte;
4094 :
4095 : /* it should be a base rel with a tablesample clause... */
4096 : Assert(scan_relid > 0);
4097 24 : rte = planner_rt_fetch(scan_relid, root);
4098 : Assert(rte->rtekind == RTE_RELATION);
4099 : Assert(rte->tablesample != NULL);
4100 :
4101 24 : ADJUST_CHILD_ATTRS(rte->tablesample);
4102 : }
4103 114 : break;
4104 :
4105 261 : case T_IndexPath:
4106 : {
4107 261 : IndexPath *ipath = (IndexPath *) path;
4108 :
4109 261 : ADJUST_CHILD_ATTRS(ipath->indexinfo->indrestrictinfo);
4110 261 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
4111 261 : new_path = (Path *) ipath;
4112 : }
4113 261 : break;
4114 :
4115 24 : case T_BitmapHeapPath:
4116 : {
4117 24 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4118 :
4119 24 : ADJUST_CHILD_ATTRS(bhpath->path.parent->baserestrictinfo);
4120 24 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
4121 24 : new_path = (Path *) bhpath;
4122 : }
4123 24 : break;
4124 :
4125 12 : case T_BitmapAndPath:
4126 : {
4127 12 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4128 :
4129 12 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
4130 12 : new_path = (Path *) bapath;
4131 : }
4132 12 : break;
4133 :
4134 12 : case T_BitmapOrPath:
4135 : {
4136 12 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4137 :
4138 12 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
4139 12 : new_path = (Path *) bopath;
4140 : }
4141 12 : break;
4142 :
4143 0 : case T_ForeignPath:
4144 : {
4145 0 : ForeignPath *fpath = (ForeignPath *) path;
4146 : ReparameterizeForeignPathByChild_function rfpc_func;
4147 :
4148 0 : ADJUST_CHILD_ATTRS(fpath->path.parent->baserestrictinfo);
4149 0 : if (fpath->fdw_outerpath)
4150 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
4151 0 : if (fpath->fdw_restrictinfo)
4152 0 : ADJUST_CHILD_ATTRS(fpath->fdw_restrictinfo);
4153 :
4154 : /* Hand over to FDW if needed. */
4155 0 : rfpc_func =
4156 0 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
4157 0 : if (rfpc_func)
4158 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
4159 : child_rel);
4160 0 : new_path = (Path *) fpath;
4161 : }
4162 0 : break;
4163 :
4164 0 : case T_CustomPath:
4165 : {
4166 0 : CustomPath *cpath = (CustomPath *) path;
4167 :
4168 0 : ADJUST_CHILD_ATTRS(cpath->path.parent->baserestrictinfo);
4169 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
4170 0 : if (cpath->custom_restrictinfo)
4171 0 : ADJUST_CHILD_ATTRS(cpath->custom_restrictinfo);
4172 0 : if (cpath->methods &&
4173 0 : cpath->methods->ReparameterizeCustomPathByChild)
4174 0 : cpath->custom_private =
4175 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4176 : cpath->custom_private,
4177 : child_rel);
4178 0 : new_path = (Path *) cpath;
4179 : }
4180 0 : break;
4181 :
4182 18 : case T_NestPath:
4183 : {
4184 18 : NestPath *npath = (NestPath *) path;
4185 18 : JoinPath *jpath = (JoinPath *) npath;
4186 :
4187 18 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4188 18 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4189 18 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4190 18 : new_path = (Path *) npath;
4191 : }
4192 18 : break;
4193 :
4194 0 : case T_MergePath:
4195 : {
4196 0 : MergePath *mpath = (MergePath *) path;
4197 0 : JoinPath *jpath = (JoinPath *) mpath;
4198 :
4199 0 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4200 0 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4201 0 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4202 0 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4203 0 : new_path = (Path *) mpath;
4204 : }
4205 0 : break;
4206 :
4207 24 : case T_HashPath:
4208 : {
4209 24 : HashPath *hpath = (HashPath *) path;
4210 24 : JoinPath *jpath = (JoinPath *) hpath;
4211 :
4212 24 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4213 24 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4214 24 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4215 24 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4216 24 : new_path = (Path *) hpath;
4217 : }
4218 24 : break;
4219 :
4220 12 : case T_AppendPath:
4221 : {
4222 12 : AppendPath *apath = (AppendPath *) path;
4223 :
4224 12 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4225 12 : new_path = (Path *) apath;
4226 : }
4227 12 : break;
4228 :
4229 0 : case T_MaterialPath:
4230 : {
4231 0 : MaterialPath *mpath = (MaterialPath *) path;
4232 :
4233 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4234 0 : new_path = (Path *) mpath;
4235 : }
4236 0 : break;
4237 :
4238 24 : case T_MemoizePath:
4239 : {
4240 24 : MemoizePath *mpath = (MemoizePath *) path;
4241 :
4242 24 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4243 24 : ADJUST_CHILD_ATTRS(mpath->param_exprs);
4244 24 : new_path = (Path *) mpath;
4245 : }
4246 24 : break;
4247 :
4248 0 : case T_GatherPath:
4249 : {
4250 0 : GatherPath *gpath = (GatherPath *) path;
4251 :
4252 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4253 0 : new_path = (Path *) gpath;
4254 : }
4255 0 : break;
4256 :
4257 0 : default:
4258 : /* We don't know how to reparameterize this path. */
4259 0 : return NULL;
4260 : }
4261 :
4262 : /*
4263 : * Adjust the parameterization information, which refers to the topmost
4264 : * parent. The topmost parent can be multiple levels away from the given
4265 : * child, hence use multi-level expression adjustment routines.
4266 : */
4267 501 : old_ppi = new_path->param_info;
4268 : required_outer =
4269 501 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4270 : child_rel,
4271 501 : child_rel->top_parent);
4272 :
4273 : /* If we already have a PPI for this parameterization, just return it */
4274 501 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4275 :
4276 : /*
4277 : * If not, build a new one and link it to the list of PPIs. For the same
4278 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4279 : * context the given RelOptInfo is in.
4280 : */
4281 501 : if (new_ppi == NULL)
4282 : {
4283 : MemoryContext oldcontext;
4284 429 : RelOptInfo *rel = path->parent;
4285 :
4286 429 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4287 :
4288 429 : new_ppi = makeNode(ParamPathInfo);
4289 429 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4290 429 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4291 429 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4292 429 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4293 429 : new_ppi->ppi_serials = bms_copy(old_ppi->ppi_serials);
4294 429 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4295 :
4296 429 : MemoryContextSwitchTo(oldcontext);
4297 : }
4298 501 : bms_free(required_outer);
4299 :
4300 501 : new_path->param_info = new_ppi;
4301 :
4302 : /*
4303 : * Adjust the path target if the parent of the outer relation is
4304 : * referenced in the targetlist. This can happen when only the parent of
4305 : * outer relation is laterally referenced in this relation.
4306 : */
4307 501 : if (bms_overlap(path->parent->lateral_relids,
4308 501 : child_rel->top_parent_relids))
4309 : {
4310 240 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4311 240 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4312 : }
4313 :
4314 501 : return new_path;
4315 : }
4316 :
4317 : /*
4318 : * path_is_reparameterizable_by_child
4319 : * Given a path parameterized by the parent of the given child relation,
4320 : * see if it can be translated to be parameterized by the child relation.
4321 : *
4322 : * This must return true if and only if reparameterize_path_by_child()
4323 : * would succeed on this path. Currently it's sufficient to verify that
4324 : * the path and all of its subpaths (if any) are of the types handled by
4325 : * that function. However, subpaths that are not parameterized can be
4326 : * disregarded since they won't require translation.
4327 : */
4328 : bool
4329 18084 : path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
4330 : {
4331 : #define REJECT_IF_PATH_NOT_REPARAMETERIZABLE(path) \
4332 : do { \
4333 : if (!path_is_reparameterizable_by_child(path, child_rel)) \
4334 : return false; \
4335 : } while(0)
4336 :
4337 : #define REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(pathlist) \
4338 : do { \
4339 : if (!pathlist_is_reparameterizable_by_child(pathlist, child_rel)) \
4340 : return false; \
4341 : } while(0)
4342 :
4343 : /*
4344 : * If the path is not parameterized by the parent of the given relation,
4345 : * it doesn't need reparameterization.
4346 : */
4347 18084 : if (!path->param_info ||
4348 17880 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4349 492 : return true;
4350 :
4351 : /*
4352 : * Check that the path type is one that reparameterize_path_by_child() can
4353 : * handle, and recursively check subpaths.
4354 : */
4355 17592 : switch (nodeTag(path))
4356 : {
4357 11892 : case T_Path:
4358 : case T_IndexPath:
4359 11892 : break;
4360 :
4361 24 : case T_BitmapHeapPath:
4362 : {
4363 24 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4364 :
4365 24 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(bhpath->bitmapqual);
4366 : }
4367 24 : break;
4368 :
4369 12 : case T_BitmapAndPath:
4370 : {
4371 12 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4372 :
4373 12 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bapath->bitmapquals);
4374 : }
4375 12 : break;
4376 :
4377 12 : case T_BitmapOrPath:
4378 : {
4379 12 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4380 :
4381 12 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bopath->bitmapquals);
4382 : }
4383 12 : break;
4384 :
4385 74 : case T_ForeignPath:
4386 : {
4387 74 : ForeignPath *fpath = (ForeignPath *) path;
4388 :
4389 74 : if (fpath->fdw_outerpath)
4390 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(fpath->fdw_outerpath);
4391 : }
4392 74 : break;
4393 :
4394 0 : case T_CustomPath:
4395 : {
4396 0 : CustomPath *cpath = (CustomPath *) path;
4397 :
4398 0 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(cpath->custom_paths);
4399 : }
4400 0 : break;
4401 :
4402 624 : case T_NestPath:
4403 : case T_MergePath:
4404 : case T_HashPath:
4405 : {
4406 624 : JoinPath *jpath = (JoinPath *) path;
4407 :
4408 624 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->outerjoinpath);
4409 624 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->innerjoinpath);
4410 : }
4411 624 : break;
4412 :
4413 96 : case T_AppendPath:
4414 : {
4415 96 : AppendPath *apath = (AppendPath *) path;
4416 :
4417 96 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(apath->subpaths);
4418 : }
4419 96 : break;
4420 :
4421 0 : case T_MaterialPath:
4422 : {
4423 0 : MaterialPath *mpath = (MaterialPath *) path;
4424 :
4425 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4426 : }
4427 0 : break;
4428 :
4429 4858 : case T_MemoizePath:
4430 : {
4431 4858 : MemoizePath *mpath = (MemoizePath *) path;
4432 :
4433 4858 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4434 : }
4435 4858 : break;
4436 :
4437 0 : case T_GatherPath:
4438 : {
4439 0 : GatherPath *gpath = (GatherPath *) path;
4440 :
4441 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(gpath->subpath);
4442 : }
4443 0 : break;
4444 :
4445 0 : default:
4446 : /* We don't know how to reparameterize this path. */
4447 0 : return false;
4448 : }
4449 :
4450 17592 : return true;
4451 : }
4452 :
4453 : /*
4454 : * reparameterize_pathlist_by_child
4455 : * Helper function to reparameterize a list of paths by given child rel.
4456 : *
4457 : * Returns NIL to indicate failure, so pathlist had better not be NIL.
4458 : */
4459 : static List *
4460 36 : reparameterize_pathlist_by_child(PlannerInfo *root,
4461 : List *pathlist,
4462 : RelOptInfo *child_rel)
4463 : {
4464 : ListCell *lc;
4465 36 : List *result = NIL;
4466 :
4467 108 : foreach(lc, pathlist)
4468 : {
4469 72 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4470 : child_rel);
4471 :
4472 72 : if (path == NULL)
4473 : {
4474 0 : list_free(result);
4475 0 : return NIL;
4476 : }
4477 :
4478 72 : result = lappend(result, path);
4479 : }
4480 :
4481 36 : return result;
4482 : }
4483 :
4484 : /*
4485 : * pathlist_is_reparameterizable_by_child
4486 : * Helper function to check if a list of paths can be reparameterized.
4487 : */
4488 : static bool
4489 120 : pathlist_is_reparameterizable_by_child(List *pathlist, RelOptInfo *child_rel)
4490 : {
4491 : ListCell *lc;
4492 :
4493 360 : foreach(lc, pathlist)
4494 : {
4495 240 : Path *path = (Path *) lfirst(lc);
4496 :
4497 240 : if (!path_is_reparameterizable_by_child(path, child_rel))
4498 0 : return false;
4499 : }
4500 :
4501 120 : return true;
4502 : }
|
