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 1143002 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
69 : {
70 : /* Number of disabled nodes, if different, trumps all else. */
71 1143002 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
72 : {
73 2712 : if (path1->disabled_nodes < path2->disabled_nodes)
74 2712 : return -1;
75 : else
76 0 : return +1;
77 : }
78 :
79 1140290 : if (criterion == STARTUP_COST)
80 : {
81 574748 : if (path1->startup_cost < path2->startup_cost)
82 351380 : return -1;
83 223368 : if (path1->startup_cost > path2->startup_cost)
84 111698 : return +1;
85 :
86 : /*
87 : * If paths have the same startup cost (not at all unlikely), order
88 : * them by total cost.
89 : */
90 111670 : if (path1->total_cost < path2->total_cost)
91 56600 : return -1;
92 55070 : if (path1->total_cost > path2->total_cost)
93 5282 : return +1;
94 : }
95 : else
96 : {
97 565542 : if (path1->total_cost < path2->total_cost)
98 526382 : return -1;
99 39160 : if (path1->total_cost > path2->total_cost)
100 10820 : return +1;
101 :
102 : /*
103 : * If paths have the same total cost, order them by startup cost.
104 : */
105 28340 : if (path1->startup_cost < path2->startup_cost)
106 2616 : return -1;
107 25724 : if (path1->startup_cost > path2->startup_cost)
108 10 : return +1;
109 : }
110 75502 : 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 5970 : 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 5970 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
131 : {
132 36 : if (path1->disabled_nodes < path2->disabled_nodes)
133 36 : return -1;
134 : else
135 0 : return +1;
136 : }
137 :
138 5934 : if (fraction <= 0.0 || fraction >= 1.0)
139 1746 : return compare_path_costs(path1, path2, TOTAL_COST);
140 4188 : cost1 = path1->startup_cost +
141 4188 : fraction * (path1->total_cost - path1->startup_cost);
142 4188 : cost2 = path2->startup_cost +
143 4188 : fraction * (path2->total_cost - path2->startup_cost);
144 4188 : if (cost1 < cost2)
145 3514 : return -1;
146 674 : if (cost1 > cost2)
147 674 : 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 4704300 : 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 4704300 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
188 : {
189 37502 : if (path1->disabled_nodes < path2->disabled_nodes)
190 23160 : return COSTS_BETTER1;
191 : else
192 14342 : 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 4666798 : if (path1->total_cost > path2->total_cost * fuzz_factor)
200 : {
201 : /* path1 fuzzily worse on total cost */
202 2442802 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
203 136928 : path2->startup_cost > path1->startup_cost * fuzz_factor)
204 : {
205 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
206 91938 : return COSTS_DIFFERENT;
207 : }
208 : /* else path2 dominates */
209 2350864 : return COSTS_BETTER2;
210 : }
211 2223996 : if (path2->total_cost > path1->total_cost * fuzz_factor)
212 : {
213 : /* path2 fuzzily worse on total cost */
214 1154004 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
215 60660 : path1->startup_cost > path2->startup_cost * fuzz_factor)
216 : {
217 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
218 40012 : return COSTS_DIFFERENT;
219 : }
220 : /* else path1 dominates */
221 1113992 : return COSTS_BETTER1;
222 : }
223 : /* fuzzily the same on total cost ... */
224 1069992 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
225 : {
226 : /* ... but path1 fuzzily worse on startup, so path2 wins */
227 391694 : return COSTS_BETTER2;
228 : }
229 678298 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
230 : {
231 : /* ... but path2 fuzzily worse on startup, so path1 wins */
232 67658 : return COSTS_BETTER1;
233 : }
234 : /* fuzzily the same on both costs */
235 610640 : 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 2171820 : 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 2171820 : if (parent_rel->pathlist == NIL)
279 0 : elog(ERROR, "could not devise a query plan for the given query");
280 :
281 2171820 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
282 2171820 : parameterized_paths = NIL;
283 :
284 4927704 : foreach(p, parent_rel->pathlist)
285 : {
286 2755884 : Path *path = (Path *) lfirst(p);
287 : int cmp;
288 :
289 2755884 : if (path->param_info)
290 : {
291 : /* Parameterized path, so add it to parameterized_paths */
292 144990 : 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 144990 : if (cheapest_total_path)
299 29886 : 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 115104 : if (best_param_path == NULL)
307 105114 : best_param_path = path;
308 : else
309 : {
310 9990 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
311 9990 : PATH_REQ_OUTER(best_param_path)))
312 : {
313 60 : case BMS_EQUAL:
314 : /* keep the cheaper one */
315 60 : if (compare_path_costs(path, best_param_path,
316 : TOTAL_COST) < 0)
317 0 : best_param_path = path;
318 60 : break;
319 760 : case BMS_SUBSET1:
320 : /* new path is less-parameterized */
321 760 : best_param_path = path;
322 760 : break;
323 6 : case BMS_SUBSET2:
324 : /* old path is less-parameterized, keep it */
325 6 : break;
326 9164 : 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 9164 : break;
334 : }
335 : }
336 : }
337 : else
338 : {
339 : /* Unparameterized path, so consider it for cheapest slots */
340 2610894 : if (cheapest_total_path == NULL)
341 : {
342 2159028 : cheapest_startup_path = cheapest_total_path = path;
343 2159028 : 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 451866 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
354 451866 : if (cmp > 0 ||
355 388 : (cmp == 0 &&
356 388 : compare_pathkeys(cheapest_startup_path->pathkeys,
357 : path->pathkeys) == PATHKEYS_BETTER2))
358 82090 : cheapest_startup_path = path;
359 :
360 451866 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
361 451866 : if (cmp > 0 ||
362 48 : (cmp == 0 &&
363 48 : 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 2171820 : if (cheapest_total_path)
371 2159028 : 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 2171820 : if (cheapest_total_path == NULL)
378 12792 : cheapest_total_path = best_param_path;
379 : Assert(cheapest_total_path != NULL);
380 :
381 2171820 : parent_rel->cheapest_startup_path = cheapest_startup_path;
382 2171820 : parent_rel->cheapest_total_path = cheapest_total_path;
383 2171820 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
384 2171820 : }
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 4721028 : add_path(RelOptInfo *parent_rel, Path *new_path)
460 : {
461 4721028 : bool accept_new = true; /* unless we find a superior old path */
462 4721028 : 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 4721028 : CHECK_FOR_INTERRUPTS();
471 :
472 : /* Pretend parameterized paths have no pathkeys, per comment above */
473 4721028 : 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 7281964 : foreach(p1, parent_rel->pathlist)
481 : {
482 4326216 : Path *old_path = (Path *) lfirst(p1);
483 4326216 : 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 4326216 : 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 4326216 : if (costcmp != COSTS_DIFFERENT)
506 : {
507 : /* Similarly check to see if either dominates on pathkeys */
508 : List *old_path_pathkeys;
509 :
510 4194332 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
511 4194332 : keyscmp = compare_pathkeys(new_path_pathkeys,
512 : old_path_pathkeys);
513 4194332 : if (keyscmp != PATHKEYS_DIFFERENT)
514 : {
515 4003860 : switch (costcmp)
516 : {
517 413844 : case COSTS_EQUAL:
518 413844 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
519 413844 : PATH_REQ_OUTER(old_path));
520 413844 : if (keyscmp == PATHKEYS_BETTER1)
521 : {
522 9084 : if ((outercmp == BMS_EQUAL ||
523 9084 : outercmp == BMS_SUBSET1) &&
524 9084 : new_path->rows <= old_path->rows &&
525 9076 : new_path->parallel_safe >= old_path->parallel_safe)
526 9076 : remove_old = true; /* new dominates old */
527 : }
528 404760 : else if (keyscmp == PATHKEYS_BETTER2)
529 : {
530 20690 : if ((outercmp == BMS_EQUAL ||
531 20690 : outercmp == BMS_SUBSET2) &&
532 20690 : new_path->rows >= old_path->rows &&
533 20688 : new_path->parallel_safe <= old_path->parallel_safe)
534 20688 : accept_new = false; /* old dominates new */
535 : }
536 : else /* keyscmp == PATHKEYS_EQUAL */
537 : {
538 384070 : 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 378434 : if (new_path->parallel_safe >
556 378434 : old_path->parallel_safe)
557 42 : remove_old = true; /* new dominates old */
558 378392 : else if (new_path->parallel_safe <
559 378392 : old_path->parallel_safe)
560 54 : accept_new = false; /* old dominates new */
561 378338 : else if (new_path->rows < old_path->rows)
562 36 : remove_old = true; /* new dominates old */
563 378302 : else if (new_path->rows > old_path->rows)
564 218 : accept_new = false; /* old dominates new */
565 378084 : else if (compare_path_costs_fuzzily(new_path,
566 : old_path,
567 : 1.0000000001) == COSTS_BETTER1)
568 16878 : remove_old = true; /* new dominates old */
569 : else
570 361206 : accept_new = false; /* old equals or
571 : * dominates new */
572 : }
573 5636 : else if (outercmp == BMS_SUBSET1 &&
574 1226 : new_path->rows <= old_path->rows &&
575 1206 : new_path->parallel_safe >= old_path->parallel_safe)
576 1206 : remove_old = true; /* new dominates old */
577 4430 : else if (outercmp == BMS_SUBSET2 &&
578 3658 : new_path->rows >= old_path->rows &&
579 3408 : new_path->parallel_safe <= old_path->parallel_safe)
580 3408 : accept_new = false; /* old dominates new */
581 : /* else different parameterizations, keep both */
582 : }
583 413844 : break;
584 1160356 : case COSTS_BETTER1:
585 1160356 : if (keyscmp != PATHKEYS_BETTER2)
586 : {
587 788562 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
588 788562 : PATH_REQ_OUTER(old_path));
589 788562 : if ((outercmp == BMS_EQUAL ||
590 677000 : outercmp == BMS_SUBSET1) &&
591 677000 : new_path->rows <= old_path->rows &&
592 672244 : new_path->parallel_safe >= old_path->parallel_safe)
593 669540 : remove_old = true; /* new dominates old */
594 : }
595 1160356 : break;
596 2429660 : case COSTS_BETTER2:
597 2429660 : if (keyscmp != PATHKEYS_BETTER1)
598 : {
599 1551620 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
600 1551620 : PATH_REQ_OUTER(old_path));
601 1551620 : if ((outercmp == BMS_EQUAL ||
602 1463178 : outercmp == BMS_SUBSET2) &&
603 1463178 : new_path->rows >= old_path->rows &&
604 1381526 : new_path->parallel_safe <= old_path->parallel_safe)
605 1379706 : accept_new = false; /* old dominates new */
606 : }
607 2429660 : 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 4326216 : if (remove_old)
623 : {
624 696778 : 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 696778 : if (!IsA(old_path, IndexPath))
631 671312 : 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 3629438 : if (new_path->disabled_nodes > old_path->disabled_nodes ||
640 3615096 : (new_path->disabled_nodes == old_path->disabled_nodes &&
641 3614130 : new_path->total_cost >= old_path->total_cost))
642 3015162 : 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 4326216 : if (!accept_new)
651 1765280 : break;
652 : }
653 :
654 4721028 : if (accept_new)
655 : {
656 : /* Accept the new path: insert it at proper place in pathlist */
657 2955748 : parent_rel->pathlist =
658 2955748 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
659 : }
660 : else
661 : {
662 : /* Reject and recycle the new path */
663 1765280 : if (!IsA(new_path, IndexPath))
664 1661620 : pfree(new_path);
665 : }
666 4721028 : }
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 5331652 : 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 5331652 : new_path_pathkeys = required_outer ? NIL : pathkeys;
696 :
697 : /* Decide whether new path's startup cost is interesting */
698 5331652 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
699 :
700 6820352 : foreach(p1, parent_rel->pathlist)
701 : {
702 6469378 : 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 6469378 : if (unlikely(old_path->disabled_nodes != disabled_nodes))
712 : {
713 12930 : if (disabled_nodes < old_path->disabled_nodes)
714 324 : break;
715 : }
716 6456448 : else if (total_cost <= old_path->total_cost * STD_FUZZ_FACTOR)
717 1943112 : 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 4525942 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
729 2134004 : !consider_startup)
730 : {
731 : /* new path loses on cost, so check pathkeys... */
732 : List *old_path_pathkeys;
733 :
734 4423190 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
735 4423190 : keyscmp = compare_pathkeys(new_path_pathkeys,
736 : old_path_pathkeys);
737 4423190 : if (keyscmp == PATHKEYS_EQUAL ||
738 : keyscmp == PATHKEYS_BETTER2)
739 : {
740 : /* new path does not win on pathkeys... */
741 3107260 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
742 : {
743 : /* Found an old path that dominates the new one */
744 3037242 : return false;
745 : }
746 : }
747 : }
748 : }
749 :
750 2294410 : 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 309726 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
794 : {
795 309726 : bool accept_new = true; /* unless we find a superior old path */
796 309726 : int insert_at = 0; /* where to insert new item */
797 : ListCell *p1;
798 :
799 : /* Check for query cancel. */
800 309726 : 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 430632 : foreach(p1, parent_rel->partial_pathlist)
813 : {
814 254150 : Path *old_path = (Path *) lfirst(p1);
815 254150 : bool remove_old = false; /* unless new proves superior */
816 : PathKeysComparison keyscmp;
817 :
818 : /* Compare pathkeys. */
819 254150 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
820 :
821 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
822 254150 : if (keyscmp != PATHKEYS_DIFFERENT)
823 : {
824 253904 : if (unlikely(new_path->disabled_nodes != old_path->disabled_nodes))
825 : {
826 1652 : if (new_path->disabled_nodes > old_path->disabled_nodes)
827 1004 : accept_new = false;
828 : else
829 648 : remove_old = true;
830 : }
831 252252 : else if (new_path->total_cost > old_path->total_cost
832 252252 : * STD_FUZZ_FACTOR)
833 : {
834 : /* New path costs more; keep it only if pathkeys are better. */
835 124264 : if (keyscmp != PATHKEYS_BETTER1)
836 90080 : accept_new = false;
837 : }
838 127988 : else if (old_path->total_cost > new_path->total_cost
839 127988 : * STD_FUZZ_FACTOR)
840 : {
841 : /* Old path costs more; keep it only if pathkeys are better. */
842 85138 : if (keyscmp != PATHKEYS_BETTER2)
843 29962 : remove_old = true;
844 : }
845 42850 : else if (keyscmp == PATHKEYS_BETTER1)
846 : {
847 : /* Costs are about the same, new path has better pathkeys. */
848 24 : remove_old = true;
849 : }
850 42826 : else if (keyscmp == PATHKEYS_BETTER2)
851 : {
852 : /* Costs are about the same, old path has better pathkeys. */
853 2064 : accept_new = false;
854 : }
855 40762 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
856 : {
857 : /* Pathkeys are the same, and the old path costs more. */
858 666 : remove_old = true;
859 : }
860 : else
861 : {
862 : /*
863 : * Pathkeys are the same, and new path isn't materially
864 : * cheaper.
865 : */
866 40096 : accept_new = false;
867 : }
868 : }
869 :
870 : /*
871 : * Remove current element from partial_pathlist if dominated by new.
872 : */
873 254150 : if (remove_old)
874 : {
875 31300 : parent_rel->partial_pathlist =
876 31300 : foreach_delete_current(parent_rel->partial_pathlist, p1);
877 31300 : pfree(old_path);
878 : }
879 : else
880 : {
881 : /* new belongs after this old path if it has cost >= old's */
882 222850 : if (new_path->total_cost >= old_path->total_cost)
883 166410 : 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 254150 : if (!accept_new)
892 133244 : break;
893 : }
894 :
895 309726 : if (accept_new)
896 : {
897 : /* Accept the new path: insert it at proper place */
898 176482 : parent_rel->partial_pathlist =
899 176482 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
900 : }
901 : else
902 : {
903 : /* Reject and recycle the new path */
904 133244 : pfree(new_path);
905 : }
906 309726 : }
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 448990 : 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 573092 : foreach(p1, parent_rel->partial_pathlist)
936 : {
937 466432 : Path *old_path = (Path *) lfirst(p1);
938 : PathKeysComparison keyscmp;
939 :
940 466432 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
941 466432 : if (keyscmp != PATHKEYS_DIFFERENT)
942 : {
943 466204 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
944 : keyscmp != PATHKEYS_BETTER1)
945 342330 : return false;
946 237270 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
947 : keyscmp != PATHKEYS_BETTER2)
948 113396 : 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 106660 : if (!add_path_precheck(parent_rel, disabled_nodes, total_cost, total_cost,
964 : pathkeys, NULL))
965 2936 : return false;
966 :
967 103724 : 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 445892 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
982 : Relids required_outer, int parallel_workers)
983 : {
984 445892 : Path *pathnode = makeNode(Path);
985 :
986 445892 : pathnode->pathtype = T_SeqScan;
987 445892 : pathnode->parent = rel;
988 445892 : pathnode->pathtarget = rel->reltarget;
989 445892 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
990 : required_outer);
991 445892 : pathnode->parallel_aware = (parallel_workers > 0);
992 445892 : pathnode->parallel_safe = rel->consider_parallel;
993 445892 : pathnode->parallel_workers = parallel_workers;
994 445892 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
995 :
996 445892 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
997 :
998 445892 : return pathnode;
999 : }
1000 :
1001 : /*
1002 : * create_samplescan_path
1003 : * Creates a path node for a sampled table scan.
1004 : */
1005 : Path *
1006 306 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
1007 : {
1008 306 : Path *pathnode = makeNode(Path);
1009 :
1010 306 : pathnode->pathtype = T_SampleScan;
1011 306 : pathnode->parent = rel;
1012 306 : pathnode->pathtarget = rel->reltarget;
1013 306 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1014 : required_outer);
1015 306 : pathnode->parallel_aware = false;
1016 306 : pathnode->parallel_safe = rel->consider_parallel;
1017 306 : pathnode->parallel_workers = 0;
1018 306 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
1019 :
1020 306 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
1021 :
1022 306 : 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 822430 : 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 822430 : IndexPath *pathnode = makeNode(IndexPath);
1060 822430 : RelOptInfo *rel = index->rel;
1061 :
1062 822430 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1063 822430 : pathnode->path.parent = rel;
1064 822430 : pathnode->path.pathtarget = rel->reltarget;
1065 822430 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1066 : required_outer);
1067 822430 : pathnode->path.parallel_aware = false;
1068 822430 : pathnode->path.parallel_safe = rel->consider_parallel;
1069 822430 : pathnode->path.parallel_workers = 0;
1070 822430 : pathnode->path.pathkeys = pathkeys;
1071 :
1072 822430 : pathnode->indexinfo = index;
1073 822430 : pathnode->indexclauses = indexclauses;
1074 822430 : pathnode->indexorderbys = indexorderbys;
1075 822430 : pathnode->indexorderbycols = indexorderbycols;
1076 822430 : pathnode->indexscandir = indexscandir;
1077 :
1078 822430 : cost_index(pathnode, root, loop_count, partial_path);
1079 :
1080 822430 : 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 362612 : 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 362612 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1104 :
1105 362612 : pathnode->path.pathtype = T_BitmapHeapScan;
1106 362612 : pathnode->path.parent = rel;
1107 362612 : pathnode->path.pathtarget = rel->reltarget;
1108 362612 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1109 : required_outer);
1110 362612 : pathnode->path.parallel_aware = (parallel_degree > 0);
1111 362612 : pathnode->path.parallel_safe = rel->consider_parallel;
1112 362612 : pathnode->path.parallel_workers = parallel_degree;
1113 362612 : pathnode->path.pathkeys = NIL; /* always unordered */
1114 :
1115 362612 : pathnode->bitmapqual = bitmapqual;
1116 :
1117 362612 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1118 : pathnode->path.param_info,
1119 : bitmapqual, loop_count);
1120 :
1121 362612 : return pathnode;
1122 : }
1123 :
1124 : /*
1125 : * create_bitmap_and_path
1126 : * Creates a path node representing a BitmapAnd.
1127 : */
1128 : BitmapAndPath *
1129 50660 : create_bitmap_and_path(PlannerInfo *root,
1130 : RelOptInfo *rel,
1131 : List *bitmapquals)
1132 : {
1133 50660 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1134 50660 : Relids required_outer = NULL;
1135 : ListCell *lc;
1136 :
1137 50660 : pathnode->path.pathtype = T_BitmapAnd;
1138 50660 : pathnode->path.parent = rel;
1139 50660 : 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 151980 : foreach(lc, bitmapquals)
1147 : {
1148 101320 : Path *bitmapqual = (Path *) lfirst(lc);
1149 :
1150 101320 : required_outer = bms_add_members(required_outer,
1151 101320 : PATH_REQ_OUTER(bitmapqual));
1152 : }
1153 50660 : 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 50660 : pathnode->path.parallel_aware = false;
1163 50660 : pathnode->path.parallel_safe = rel->consider_parallel;
1164 50660 : pathnode->path.parallel_workers = 0;
1165 :
1166 50660 : pathnode->path.pathkeys = NIL; /* always unordered */
1167 :
1168 50660 : pathnode->bitmapquals = bitmapquals;
1169 :
1170 : /* this sets bitmapselectivity as well as the regular cost fields: */
1171 50660 : cost_bitmap_and_node(pathnode, root);
1172 :
1173 50660 : return pathnode;
1174 : }
1175 :
1176 : /*
1177 : * create_bitmap_or_path
1178 : * Creates a path node representing a BitmapOr.
1179 : */
1180 : BitmapOrPath *
1181 1040 : create_bitmap_or_path(PlannerInfo *root,
1182 : RelOptInfo *rel,
1183 : List *bitmapquals)
1184 : {
1185 1040 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1186 1040 : Relids required_outer = NULL;
1187 : ListCell *lc;
1188 :
1189 1040 : pathnode->path.pathtype = T_BitmapOr;
1190 1040 : pathnode->path.parent = rel;
1191 1040 : 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 2922 : foreach(lc, bitmapquals)
1199 : {
1200 1882 : Path *bitmapqual = (Path *) lfirst(lc);
1201 :
1202 1882 : required_outer = bms_add_members(required_outer,
1203 1882 : PATH_REQ_OUTER(bitmapqual));
1204 : }
1205 1040 : 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 1040 : pathnode->path.parallel_aware = false;
1215 1040 : pathnode->path.parallel_safe = rel->consider_parallel;
1216 1040 : pathnode->path.parallel_workers = 0;
1217 :
1218 1040 : pathnode->path.pathkeys = NIL; /* always unordered */
1219 :
1220 1040 : pathnode->bitmapquals = bitmapquals;
1221 :
1222 : /* this sets bitmapselectivity as well as the regular cost fields: */
1223 1040 : cost_bitmap_or_node(pathnode, root);
1224 :
1225 1040 : 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 872 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1234 : Relids required_outer)
1235 : {
1236 872 : TidPath *pathnode = makeNode(TidPath);
1237 :
1238 872 : pathnode->path.pathtype = T_TidScan;
1239 872 : pathnode->path.parent = rel;
1240 872 : pathnode->path.pathtarget = rel->reltarget;
1241 872 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1242 : required_outer);
1243 872 : pathnode->path.parallel_aware = false;
1244 872 : pathnode->path.parallel_safe = rel->consider_parallel;
1245 872 : pathnode->path.parallel_workers = 0;
1246 872 : pathnode->path.pathkeys = NIL; /* always unordered */
1247 :
1248 872 : pathnode->tidquals = tidquals;
1249 :
1250 872 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1251 : pathnode->path.param_info);
1252 :
1253 872 : 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 2052 : create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
1263 : List *tidrangequals, Relids required_outer,
1264 : int parallel_workers)
1265 : {
1266 2052 : TidRangePath *pathnode = makeNode(TidRangePath);
1267 :
1268 2052 : pathnode->path.pathtype = T_TidRangeScan;
1269 2052 : pathnode->path.parent = rel;
1270 2052 : pathnode->path.pathtarget = rel->reltarget;
1271 2052 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1272 : required_outer);
1273 2052 : pathnode->path.parallel_aware = (parallel_workers > 0);
1274 2052 : pathnode->path.parallel_safe = rel->consider_parallel;
1275 2052 : pathnode->path.parallel_workers = parallel_workers;
1276 2052 : pathnode->path.pathkeys = NIL; /* always unordered */
1277 :
1278 2052 : pathnode->tidrangequals = tidrangequals;
1279 :
1280 2052 : cost_tidrangescan(&pathnode->path, root, rel, tidrangequals,
1281 : pathnode->path.param_info);
1282 :
1283 2052 : 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 93272 : create_append_path(PlannerInfo *root,
1300 : RelOptInfo *rel,
1301 : List *subpaths, List *partial_subpaths,
1302 : List *pathkeys, Relids required_outer,
1303 : int parallel_workers, bool parallel_aware,
1304 : double rows)
1305 : {
1306 93272 : AppendPath *pathnode = makeNode(AppendPath);
1307 : ListCell *l;
1308 :
1309 : Assert(!parallel_aware || parallel_workers > 0);
1310 :
1311 93272 : pathnode->path.pathtype = T_Append;
1312 93272 : pathnode->path.parent = rel;
1313 93272 : pathnode->path.pathtarget = rel->reltarget;
1314 :
1315 : /*
1316 : * If this is for a baserel (not a join or non-leaf partition), we prefer
1317 : * to apply get_baserel_parampathinfo to construct a full ParamPathInfo
1318 : * for the path. This supports building a Memoize path atop this path,
1319 : * and if this is a partitioned table the info may be useful for run-time
1320 : * pruning (cf make_partition_pruneinfo()).
1321 : *
1322 : * However, if we don't have "root" then that won't work and we fall back
1323 : * on the simpler get_appendrel_parampathinfo. There's no point in doing
1324 : * the more expensive thing for a dummy path, either.
1325 : */
1326 93272 : if (rel->reloptkind == RELOPT_BASEREL && root && subpaths != NIL)
1327 40894 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1328 : rel,
1329 : required_outer);
1330 : else
1331 52378 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1332 : required_outer);
1333 :
1334 93272 : pathnode->path.parallel_aware = parallel_aware;
1335 93272 : pathnode->path.parallel_safe = rel->consider_parallel;
1336 93272 : pathnode->path.parallel_workers = parallel_workers;
1337 93272 : pathnode->path.pathkeys = pathkeys;
1338 :
1339 : /*
1340 : * For parallel append, non-partial paths are sorted by descending total
1341 : * costs. That way, the total time to finish all non-partial paths is
1342 : * minimized. Also, the partial paths are sorted by descending startup
1343 : * costs. There may be some paths that require to do startup work by a
1344 : * single worker. In such case, it's better for workers to choose the
1345 : * expensive ones first, whereas the leader should choose the cheapest
1346 : * startup plan.
1347 : */
1348 93272 : if (pathnode->path.parallel_aware)
1349 : {
1350 : /*
1351 : * We mustn't fiddle with the order of subpaths when the Append has
1352 : * pathkeys. The order they're listed in is critical to keeping the
1353 : * pathkeys valid.
1354 : */
1355 : Assert(pathkeys == NIL);
1356 :
1357 33608 : list_sort(subpaths, append_total_cost_compare);
1358 33608 : list_sort(partial_subpaths, append_startup_cost_compare);
1359 : }
1360 93272 : pathnode->first_partial_path = list_length(subpaths);
1361 93272 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1362 :
1363 : /*
1364 : * Apply query-wide LIMIT if known and path is for sole base relation.
1365 : * (Handling this at this low level is a bit klugy.)
1366 : */
1367 93272 : if (root != NULL && bms_equal(rel->relids, root->all_query_rels))
1368 47306 : pathnode->limit_tuples = root->limit_tuples;
1369 : else
1370 45966 : pathnode->limit_tuples = -1.0;
1371 :
1372 317206 : foreach(l, pathnode->subpaths)
1373 : {
1374 223934 : Path *subpath = (Path *) lfirst(l);
1375 :
1376 407248 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1377 183314 : subpath->parallel_safe;
1378 :
1379 : /* All child paths must have same parameterization */
1380 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1381 : }
1382 :
1383 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1384 :
1385 : /*
1386 : * If there's exactly one child path then the output of the Append is
1387 : * necessarily ordered the same as the child's, so we can inherit the
1388 : * child's pathkeys if any, overriding whatever the caller might've said.
1389 : * Furthermore, if the child's parallel awareness matches the Append's,
1390 : * then the Append is a no-op and will be discarded later (in setrefs.c).
1391 : * Then we can inherit the child's size and cost too, effectively charging
1392 : * zero for the Append. Otherwise, we must do the normal costsize
1393 : * calculation.
1394 : */
1395 93272 : if (list_length(pathnode->subpaths) == 1)
1396 : {
1397 22914 : Path *child = (Path *) linitial(pathnode->subpaths);
1398 :
1399 22914 : if (child->parallel_aware == parallel_aware)
1400 : {
1401 22458 : pathnode->path.rows = child->rows;
1402 22458 : pathnode->path.startup_cost = child->startup_cost;
1403 22458 : pathnode->path.total_cost = child->total_cost;
1404 : }
1405 : else
1406 456 : cost_append(pathnode, root);
1407 : /* Must do this last, else cost_append complains */
1408 22914 : pathnode->path.pathkeys = child->pathkeys;
1409 : }
1410 : else
1411 70358 : cost_append(pathnode, root);
1412 :
1413 : /* If the caller provided a row estimate, override the computed value. */
1414 93272 : if (rows >= 0)
1415 576 : pathnode->path.rows = rows;
1416 :
1417 93272 : return pathnode;
1418 : }
1419 :
1420 : /*
1421 : * append_total_cost_compare
1422 : * list_sort comparator for sorting append child paths
1423 : * by total_cost descending
1424 : *
1425 : * For equal total costs, we fall back to comparing startup costs; if those
1426 : * are equal too, break ties using bms_compare on the paths' relids.
1427 : * (This is to avoid getting unpredictable results from list_sort.)
1428 : */
1429 : static int
1430 22046 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1431 : {
1432 22046 : Path *path1 = (Path *) lfirst(a);
1433 22046 : Path *path2 = (Path *) lfirst(b);
1434 : int cmp;
1435 :
1436 22046 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1437 22046 : if (cmp != 0)
1438 19662 : return -cmp;
1439 2384 : return bms_compare(path1->parent->relids, path2->parent->relids);
1440 : }
1441 :
1442 : /*
1443 : * append_startup_cost_compare
1444 : * list_sort comparator for sorting append child paths
1445 : * by startup_cost descending
1446 : *
1447 : * For equal startup costs, we fall back to comparing total costs; if those
1448 : * are equal too, break ties using bms_compare on the paths' relids.
1449 : * (This is to avoid getting unpredictable results from list_sort.)
1450 : */
1451 : static int
1452 45564 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1453 : {
1454 45564 : Path *path1 = (Path *) lfirst(a);
1455 45564 : Path *path2 = (Path *) lfirst(b);
1456 : int cmp;
1457 :
1458 45564 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1459 45564 : if (cmp != 0)
1460 21906 : return -cmp;
1461 23658 : return bms_compare(path1->parent->relids, path2->parent->relids);
1462 : }
1463 :
1464 : /*
1465 : * create_merge_append_path
1466 : * Creates a path corresponding to a MergeAppend plan, returning the
1467 : * pathnode.
1468 : */
1469 : MergeAppendPath *
1470 10028 : create_merge_append_path(PlannerInfo *root,
1471 : RelOptInfo *rel,
1472 : List *subpaths,
1473 : List *pathkeys,
1474 : Relids required_outer)
1475 : {
1476 10028 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1477 : int input_disabled_nodes;
1478 : Cost input_startup_cost;
1479 : Cost input_total_cost;
1480 : ListCell *l;
1481 :
1482 : /*
1483 : * We don't currently support parameterized MergeAppend paths, as
1484 : * explained in the comments for generate_orderedappend_paths.
1485 : */
1486 : Assert(bms_is_empty(rel->lateral_relids) && bms_is_empty(required_outer));
1487 :
1488 10028 : pathnode->path.pathtype = T_MergeAppend;
1489 10028 : pathnode->path.parent = rel;
1490 10028 : pathnode->path.pathtarget = rel->reltarget;
1491 10028 : pathnode->path.param_info = NULL;
1492 10028 : pathnode->path.parallel_aware = false;
1493 10028 : pathnode->path.parallel_safe = rel->consider_parallel;
1494 10028 : pathnode->path.parallel_workers = 0;
1495 10028 : pathnode->path.pathkeys = pathkeys;
1496 10028 : pathnode->subpaths = subpaths;
1497 :
1498 : /*
1499 : * Apply query-wide LIMIT if known and path is for sole base relation.
1500 : * (Handling this at this low level is a bit klugy.)
1501 : */
1502 10028 : if (bms_equal(rel->relids, root->all_query_rels))
1503 4614 : pathnode->limit_tuples = root->limit_tuples;
1504 : else
1505 5414 : pathnode->limit_tuples = -1.0;
1506 :
1507 : /*
1508 : * Add up the sizes and costs of the input paths.
1509 : */
1510 10028 : pathnode->path.rows = 0;
1511 10028 : input_disabled_nodes = 0;
1512 10028 : input_startup_cost = 0;
1513 10028 : input_total_cost = 0;
1514 36156 : foreach(l, subpaths)
1515 : {
1516 26128 : Path *subpath = (Path *) lfirst(l);
1517 : int presorted_keys;
1518 : Path sort_path; /* dummy for result of
1519 : * cost_sort/cost_incremental_sort */
1520 :
1521 : /* All child paths should be unparameterized */
1522 : Assert(bms_is_empty(PATH_REQ_OUTER(subpath)));
1523 :
1524 26128 : pathnode->path.rows += subpath->rows;
1525 49600 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1526 23472 : subpath->parallel_safe;
1527 :
1528 26128 : if (!pathkeys_count_contained_in(pathkeys, subpath->pathkeys,
1529 : &presorted_keys))
1530 : {
1531 : /*
1532 : * We'll need to insert a Sort node, so include costs for that. We
1533 : * choose to use incremental sort if it is enabled and there are
1534 : * presorted keys; otherwise we use full sort.
1535 : *
1536 : * We can use the parent's LIMIT if any, since we certainly won't
1537 : * pull more than that many tuples from any child.
1538 : */
1539 314 : if (enable_incremental_sort && presorted_keys > 0)
1540 : {
1541 18 : cost_incremental_sort(&sort_path,
1542 : root,
1543 : pathkeys,
1544 : presorted_keys,
1545 : subpath->disabled_nodes,
1546 : subpath->startup_cost,
1547 : subpath->total_cost,
1548 : subpath->rows,
1549 18 : subpath->pathtarget->width,
1550 : 0.0,
1551 : work_mem,
1552 : pathnode->limit_tuples);
1553 : }
1554 : else
1555 : {
1556 296 : cost_sort(&sort_path,
1557 : root,
1558 : pathkeys,
1559 : subpath->disabled_nodes,
1560 : subpath->total_cost,
1561 : subpath->rows,
1562 296 : subpath->pathtarget->width,
1563 : 0.0,
1564 : work_mem,
1565 : pathnode->limit_tuples);
1566 : }
1567 :
1568 314 : subpath = &sort_path;
1569 : }
1570 :
1571 26128 : input_disabled_nodes += subpath->disabled_nodes;
1572 26128 : input_startup_cost += subpath->startup_cost;
1573 26128 : input_total_cost += subpath->total_cost;
1574 : }
1575 :
1576 : /*
1577 : * Now we can compute total costs of the MergeAppend. If there's exactly
1578 : * one child path and its parallel awareness matches that of the
1579 : * MergeAppend, then the MergeAppend is a no-op and will be discarded
1580 : * later (in setrefs.c); otherwise we do the normal cost calculation.
1581 : */
1582 10028 : if (list_length(subpaths) == 1 &&
1583 128 : ((Path *) linitial(subpaths))->parallel_aware ==
1584 128 : pathnode->path.parallel_aware)
1585 : {
1586 128 : pathnode->path.disabled_nodes = input_disabled_nodes;
1587 128 : pathnode->path.startup_cost = input_startup_cost;
1588 128 : pathnode->path.total_cost = input_total_cost;
1589 : }
1590 : else
1591 9900 : cost_merge_append(&pathnode->path, root,
1592 : pathkeys, list_length(subpaths),
1593 : input_disabled_nodes,
1594 : input_startup_cost, input_total_cost,
1595 : pathnode->path.rows);
1596 :
1597 10028 : return pathnode;
1598 : }
1599 :
1600 : /*
1601 : * create_group_result_path
1602 : * Creates a path representing a Result-and-nothing-else plan.
1603 : *
1604 : * This is only used for degenerate grouping cases, in which we know we
1605 : * need to produce one result row, possibly filtered by a HAVING qual.
1606 : */
1607 : GroupResultPath *
1608 195058 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1609 : PathTarget *target, List *havingqual)
1610 : {
1611 195058 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1612 :
1613 195058 : pathnode->path.pathtype = T_Result;
1614 195058 : pathnode->path.parent = rel;
1615 195058 : pathnode->path.pathtarget = target;
1616 195058 : pathnode->path.param_info = NULL; /* there are no other rels... */
1617 195058 : pathnode->path.parallel_aware = false;
1618 195058 : pathnode->path.parallel_safe = rel->consider_parallel;
1619 195058 : pathnode->path.parallel_workers = 0;
1620 195058 : pathnode->path.pathkeys = NIL;
1621 195058 : pathnode->quals = havingqual;
1622 :
1623 : /*
1624 : * We can't quite use cost_resultscan() because the quals we want to
1625 : * account for are not baserestrict quals of the rel. Might as well just
1626 : * hack it here.
1627 : */
1628 195058 : pathnode->path.rows = 1;
1629 195058 : pathnode->path.startup_cost = target->cost.startup;
1630 195058 : pathnode->path.total_cost = target->cost.startup +
1631 195058 : cpu_tuple_cost + target->cost.per_tuple;
1632 :
1633 : /*
1634 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1635 : * rowcount estimate should be 1 no matter what the qual is.
1636 : */
1637 195058 : if (havingqual)
1638 : {
1639 : QualCost qual_cost;
1640 :
1641 632 : cost_qual_eval(&qual_cost, havingqual, root);
1642 : /* havingqual is evaluated once at startup */
1643 632 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1644 632 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1645 : }
1646 :
1647 195058 : return pathnode;
1648 : }
1649 :
1650 : /*
1651 : * create_material_path
1652 : * Creates a path corresponding to a Material plan, returning the
1653 : * pathnode.
1654 : */
1655 : MaterialPath *
1656 681474 : create_material_path(RelOptInfo *rel, Path *subpath)
1657 : {
1658 681474 : MaterialPath *pathnode = makeNode(MaterialPath);
1659 :
1660 : Assert(subpath->parent == rel);
1661 :
1662 681474 : pathnode->path.pathtype = T_Material;
1663 681474 : pathnode->path.parent = rel;
1664 681474 : pathnode->path.pathtarget = rel->reltarget;
1665 681474 : pathnode->path.param_info = subpath->param_info;
1666 681474 : pathnode->path.parallel_aware = false;
1667 1305566 : pathnode->path.parallel_safe = rel->consider_parallel &&
1668 624092 : subpath->parallel_safe;
1669 681474 : pathnode->path.parallel_workers = subpath->parallel_workers;
1670 681474 : pathnode->path.pathkeys = subpath->pathkeys;
1671 :
1672 681474 : pathnode->subpath = subpath;
1673 :
1674 681474 : cost_material(&pathnode->path,
1675 : subpath->disabled_nodes,
1676 : subpath->startup_cost,
1677 : subpath->total_cost,
1678 : subpath->rows,
1679 681474 : subpath->pathtarget->width);
1680 :
1681 681474 : return pathnode;
1682 : }
1683 :
1684 : /*
1685 : * create_memoize_path
1686 : * Creates a path corresponding to a Memoize plan, returning the pathnode.
1687 : */
1688 : MemoizePath *
1689 318402 : create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1690 : List *param_exprs, List *hash_operators,
1691 : bool singlerow, bool binary_mode, Cardinality est_calls)
1692 : {
1693 318402 : MemoizePath *pathnode = makeNode(MemoizePath);
1694 :
1695 : Assert(subpath->parent == rel);
1696 :
1697 318402 : pathnode->path.pathtype = T_Memoize;
1698 318402 : pathnode->path.parent = rel;
1699 318402 : pathnode->path.pathtarget = rel->reltarget;
1700 318402 : pathnode->path.param_info = subpath->param_info;
1701 318402 : pathnode->path.parallel_aware = false;
1702 622412 : pathnode->path.parallel_safe = rel->consider_parallel &&
1703 304010 : subpath->parallel_safe;
1704 318402 : pathnode->path.parallel_workers = subpath->parallel_workers;
1705 318402 : pathnode->path.pathkeys = subpath->pathkeys;
1706 :
1707 318402 : pathnode->subpath = subpath;
1708 318402 : pathnode->hash_operators = hash_operators;
1709 318402 : pathnode->param_exprs = param_exprs;
1710 318402 : pathnode->singlerow = singlerow;
1711 318402 : pathnode->binary_mode = binary_mode;
1712 :
1713 : /*
1714 : * For now we set est_entries to 0. cost_memoize_rescan() does all the
1715 : * hard work to determine how many cache entries there are likely to be,
1716 : * so it seems best to leave it up to that function to fill this field in.
1717 : * If left at 0, the executor will make a guess at a good value.
1718 : */
1719 318402 : pathnode->est_entries = 0;
1720 :
1721 318402 : pathnode->est_calls = clamp_row_est(est_calls);
1722 :
1723 : /* These will also be set later in cost_memoize_rescan() */
1724 318402 : pathnode->est_unique_keys = 0.0;
1725 318402 : pathnode->est_hit_ratio = 0.0;
1726 :
1727 : /* we should not generate this path type when enable_memoize=false */
1728 : Assert(enable_memoize);
1729 318402 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
1730 :
1731 : /*
1732 : * Add a small additional charge for caching the first entry. All the
1733 : * harder calculations for rescans are performed in cost_memoize_rescan().
1734 : */
1735 318402 : pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
1736 318402 : pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
1737 318402 : pathnode->path.rows = subpath->rows;
1738 :
1739 318402 : return pathnode;
1740 : }
1741 :
1742 : /*
1743 : * create_gather_merge_path
1744 : *
1745 : * Creates a path corresponding to a gather merge scan, returning
1746 : * the pathnode.
1747 : */
1748 : GatherMergePath *
1749 18894 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1750 : PathTarget *target, List *pathkeys,
1751 : Relids required_outer, double *rows)
1752 : {
1753 18894 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1754 18894 : int input_disabled_nodes = 0;
1755 18894 : Cost input_startup_cost = 0;
1756 18894 : Cost input_total_cost = 0;
1757 :
1758 : Assert(subpath->parallel_safe);
1759 : Assert(pathkeys);
1760 :
1761 : /*
1762 : * The subpath should guarantee that it is adequately ordered either by
1763 : * adding an explicit sort node or by using presorted input. We cannot
1764 : * add an explicit Sort node for the subpath in createplan.c on additional
1765 : * pathkeys, because we can't guarantee the sort would be safe. For
1766 : * example, expressions may be volatile or otherwise parallel unsafe.
1767 : */
1768 18894 : if (!pathkeys_contained_in(pathkeys, subpath->pathkeys))
1769 0 : elog(ERROR, "gather merge input not sufficiently sorted");
1770 :
1771 18894 : pathnode->path.pathtype = T_GatherMerge;
1772 18894 : pathnode->path.parent = rel;
1773 18894 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1774 : required_outer);
1775 18894 : pathnode->path.parallel_aware = false;
1776 :
1777 18894 : pathnode->subpath = subpath;
1778 18894 : pathnode->num_workers = subpath->parallel_workers;
1779 18894 : pathnode->path.pathkeys = pathkeys;
1780 18894 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1781 :
1782 18894 : input_disabled_nodes += subpath->disabled_nodes;
1783 18894 : input_startup_cost += subpath->startup_cost;
1784 18894 : input_total_cost += subpath->total_cost;
1785 :
1786 18894 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1787 : input_disabled_nodes, input_startup_cost,
1788 : input_total_cost, rows);
1789 :
1790 18894 : return pathnode;
1791 : }
1792 :
1793 : /*
1794 : * create_gather_path
1795 : * Creates a path corresponding to a gather scan, returning the
1796 : * pathnode.
1797 : *
1798 : * 'rows' may optionally be set to override row estimates from other sources.
1799 : */
1800 : GatherPath *
1801 25432 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1802 : PathTarget *target, Relids required_outer, double *rows)
1803 : {
1804 25432 : GatherPath *pathnode = makeNode(GatherPath);
1805 :
1806 : Assert(subpath->parallel_safe);
1807 :
1808 25432 : pathnode->path.pathtype = T_Gather;
1809 25432 : pathnode->path.parent = rel;
1810 25432 : pathnode->path.pathtarget = target;
1811 25432 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1812 : required_outer);
1813 25432 : pathnode->path.parallel_aware = false;
1814 25432 : pathnode->path.parallel_safe = false;
1815 25432 : pathnode->path.parallel_workers = 0;
1816 25432 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1817 :
1818 25432 : pathnode->subpath = subpath;
1819 25432 : pathnode->num_workers = subpath->parallel_workers;
1820 25432 : pathnode->single_copy = false;
1821 :
1822 25432 : if (pathnode->num_workers == 0)
1823 : {
1824 0 : pathnode->path.pathkeys = subpath->pathkeys;
1825 0 : pathnode->num_workers = 1;
1826 0 : pathnode->single_copy = true;
1827 : }
1828 :
1829 25432 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
1830 :
1831 25432 : return pathnode;
1832 : }
1833 :
1834 : /*
1835 : * create_subqueryscan_path
1836 : * Creates a path corresponding to a scan of a subquery,
1837 : * returning the pathnode.
1838 : *
1839 : * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
1840 : * be trivial, ie just a fetch of all the subquery output columns in order.
1841 : * While we could determine that here, the caller can usually do it more
1842 : * efficiently (or at least amortize it over multiple calls).
1843 : */
1844 : SubqueryScanPath *
1845 62372 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1846 : bool trivial_pathtarget,
1847 : List *pathkeys, Relids required_outer)
1848 : {
1849 62372 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
1850 :
1851 62372 : pathnode->path.pathtype = T_SubqueryScan;
1852 62372 : pathnode->path.parent = rel;
1853 62372 : pathnode->path.pathtarget = rel->reltarget;
1854 62372 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1855 : required_outer);
1856 62372 : pathnode->path.parallel_aware = false;
1857 103590 : pathnode->path.parallel_safe = rel->consider_parallel &&
1858 41218 : subpath->parallel_safe;
1859 62372 : pathnode->path.parallel_workers = subpath->parallel_workers;
1860 62372 : pathnode->path.pathkeys = pathkeys;
1861 62372 : pathnode->subpath = subpath;
1862 :
1863 62372 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
1864 : trivial_pathtarget);
1865 :
1866 62372 : return pathnode;
1867 : }
1868 :
1869 : /*
1870 : * create_functionscan_path
1871 : * Creates a path corresponding to a sequential scan of a function,
1872 : * returning the pathnode.
1873 : */
1874 : Path *
1875 52698 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
1876 : List *pathkeys, Relids required_outer)
1877 : {
1878 52698 : Path *pathnode = makeNode(Path);
1879 :
1880 52698 : pathnode->pathtype = T_FunctionScan;
1881 52698 : pathnode->parent = rel;
1882 52698 : pathnode->pathtarget = rel->reltarget;
1883 52698 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1884 : required_outer);
1885 52698 : pathnode->parallel_aware = false;
1886 52698 : pathnode->parallel_safe = rel->consider_parallel;
1887 52698 : pathnode->parallel_workers = 0;
1888 52698 : pathnode->pathkeys = pathkeys;
1889 :
1890 52698 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
1891 :
1892 52698 : return pathnode;
1893 : }
1894 :
1895 : /*
1896 : * create_tablefuncscan_path
1897 : * Creates a path corresponding to a sequential scan of a table function,
1898 : * returning the pathnode.
1899 : */
1900 : Path *
1901 626 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
1902 : Relids required_outer)
1903 : {
1904 626 : Path *pathnode = makeNode(Path);
1905 :
1906 626 : pathnode->pathtype = T_TableFuncScan;
1907 626 : pathnode->parent = rel;
1908 626 : pathnode->pathtarget = rel->reltarget;
1909 626 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1910 : required_outer);
1911 626 : pathnode->parallel_aware = false;
1912 626 : pathnode->parallel_safe = rel->consider_parallel;
1913 626 : pathnode->parallel_workers = 0;
1914 626 : pathnode->pathkeys = NIL; /* result is always unordered */
1915 :
1916 626 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
1917 :
1918 626 : return pathnode;
1919 : }
1920 :
1921 : /*
1922 : * create_valuesscan_path
1923 : * Creates a path corresponding to a scan of a VALUES list,
1924 : * returning the pathnode.
1925 : */
1926 : Path *
1927 8514 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
1928 : Relids required_outer)
1929 : {
1930 8514 : Path *pathnode = makeNode(Path);
1931 :
1932 8514 : pathnode->pathtype = T_ValuesScan;
1933 8514 : pathnode->parent = rel;
1934 8514 : pathnode->pathtarget = rel->reltarget;
1935 8514 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1936 : required_outer);
1937 8514 : pathnode->parallel_aware = false;
1938 8514 : pathnode->parallel_safe = rel->consider_parallel;
1939 8514 : pathnode->parallel_workers = 0;
1940 8514 : pathnode->pathkeys = NIL; /* result is always unordered */
1941 :
1942 8514 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
1943 :
1944 8514 : return pathnode;
1945 : }
1946 :
1947 : /*
1948 : * create_ctescan_path
1949 : * Creates a path corresponding to a scan of a non-self-reference CTE,
1950 : * returning the pathnode.
1951 : */
1952 : Path *
1953 4348 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel,
1954 : List *pathkeys, Relids required_outer)
1955 : {
1956 4348 : Path *pathnode = makeNode(Path);
1957 :
1958 4348 : pathnode->pathtype = T_CteScan;
1959 4348 : pathnode->parent = rel;
1960 4348 : pathnode->pathtarget = rel->reltarget;
1961 4348 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1962 : required_outer);
1963 4348 : pathnode->parallel_aware = false;
1964 4348 : pathnode->parallel_safe = rel->consider_parallel;
1965 4348 : pathnode->parallel_workers = 0;
1966 4348 : pathnode->pathkeys = pathkeys;
1967 :
1968 4348 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
1969 :
1970 4348 : return pathnode;
1971 : }
1972 :
1973 : /*
1974 : * create_namedtuplestorescan_path
1975 : * Creates a path corresponding to a scan of a named tuplestore, returning
1976 : * the pathnode.
1977 : */
1978 : Path *
1979 482 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
1980 : Relids required_outer)
1981 : {
1982 482 : Path *pathnode = makeNode(Path);
1983 :
1984 482 : pathnode->pathtype = T_NamedTuplestoreScan;
1985 482 : pathnode->parent = rel;
1986 482 : pathnode->pathtarget = rel->reltarget;
1987 482 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1988 : required_outer);
1989 482 : pathnode->parallel_aware = false;
1990 482 : pathnode->parallel_safe = rel->consider_parallel;
1991 482 : pathnode->parallel_workers = 0;
1992 482 : pathnode->pathkeys = NIL; /* result is always unordered */
1993 :
1994 482 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
1995 :
1996 482 : return pathnode;
1997 : }
1998 :
1999 : /*
2000 : * create_resultscan_path
2001 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2002 : * returning the pathnode.
2003 : */
2004 : Path *
2005 4346 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2006 : Relids required_outer)
2007 : {
2008 4346 : Path *pathnode = makeNode(Path);
2009 :
2010 4346 : pathnode->pathtype = T_Result;
2011 4346 : pathnode->parent = rel;
2012 4346 : pathnode->pathtarget = rel->reltarget;
2013 4346 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2014 : required_outer);
2015 4346 : pathnode->parallel_aware = false;
2016 4346 : pathnode->parallel_safe = rel->consider_parallel;
2017 4346 : pathnode->parallel_workers = 0;
2018 4346 : pathnode->pathkeys = NIL; /* result is always unordered */
2019 :
2020 4346 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2021 :
2022 4346 : return pathnode;
2023 : }
2024 :
2025 : /*
2026 : * create_worktablescan_path
2027 : * Creates a path corresponding to a scan of a self-reference CTE,
2028 : * returning the pathnode.
2029 : */
2030 : Path *
2031 938 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2032 : Relids required_outer)
2033 : {
2034 938 : Path *pathnode = makeNode(Path);
2035 :
2036 938 : pathnode->pathtype = T_WorkTableScan;
2037 938 : pathnode->parent = rel;
2038 938 : pathnode->pathtarget = rel->reltarget;
2039 938 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2040 : required_outer);
2041 938 : pathnode->parallel_aware = false;
2042 938 : pathnode->parallel_safe = rel->consider_parallel;
2043 938 : pathnode->parallel_workers = 0;
2044 938 : pathnode->pathkeys = NIL; /* result is always unordered */
2045 :
2046 : /* Cost is the same as for a regular CTE scan */
2047 938 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2048 :
2049 938 : return pathnode;
2050 : }
2051 :
2052 : /*
2053 : * create_foreignscan_path
2054 : * Creates a path corresponding to a scan of a foreign base table,
2055 : * returning the pathnode.
2056 : *
2057 : * This function is never called from core Postgres; rather, it's expected
2058 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2059 : * We make the FDW supply all fields of the path, since we do not have any way
2060 : * to calculate them in core. However, there is a usually-sane default for
2061 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2062 : */
2063 : ForeignPath *
2064 3722 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2065 : PathTarget *target,
2066 : double rows, int disabled_nodes,
2067 : Cost startup_cost, Cost total_cost,
2068 : List *pathkeys,
2069 : Relids required_outer,
2070 : Path *fdw_outerpath,
2071 : List *fdw_restrictinfo,
2072 : List *fdw_private)
2073 : {
2074 3722 : ForeignPath *pathnode = makeNode(ForeignPath);
2075 :
2076 : /* Historically some FDWs were confused about when to use this */
2077 : Assert(IS_SIMPLE_REL(rel));
2078 :
2079 3722 : pathnode->path.pathtype = T_ForeignScan;
2080 3722 : pathnode->path.parent = rel;
2081 3722 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2082 3722 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2083 : required_outer);
2084 3722 : pathnode->path.parallel_aware = false;
2085 3722 : pathnode->path.parallel_safe = rel->consider_parallel;
2086 3722 : pathnode->path.parallel_workers = 0;
2087 3722 : pathnode->path.rows = rows;
2088 3722 : pathnode->path.disabled_nodes = disabled_nodes;
2089 3722 : pathnode->path.startup_cost = startup_cost;
2090 3722 : pathnode->path.total_cost = total_cost;
2091 3722 : pathnode->path.pathkeys = pathkeys;
2092 :
2093 3722 : pathnode->fdw_outerpath = fdw_outerpath;
2094 3722 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2095 3722 : pathnode->fdw_private = fdw_private;
2096 :
2097 3722 : return pathnode;
2098 : }
2099 :
2100 : /*
2101 : * create_foreign_join_path
2102 : * Creates a path corresponding to a scan of a foreign join,
2103 : * returning the pathnode.
2104 : *
2105 : * This function is never called from core Postgres; rather, it's expected
2106 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2107 : * We make the FDW supply all fields of the path, since we do not have any way
2108 : * to calculate them in core. However, there is a usually-sane default for
2109 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2110 : */
2111 : ForeignPath *
2112 1212 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2113 : PathTarget *target,
2114 : double rows, int disabled_nodes,
2115 : Cost startup_cost, Cost total_cost,
2116 : List *pathkeys,
2117 : Relids required_outer,
2118 : Path *fdw_outerpath,
2119 : List *fdw_restrictinfo,
2120 : List *fdw_private)
2121 : {
2122 1212 : ForeignPath *pathnode = makeNode(ForeignPath);
2123 :
2124 : /*
2125 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2126 : * but the API of this function doesn't support it, and existing
2127 : * extensions aren't yet trying to build such paths anyway. For the
2128 : * moment just throw an error if someone tries it; eventually we should
2129 : * revisit this.
2130 : */
2131 1212 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2132 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2133 :
2134 1212 : pathnode->path.pathtype = T_ForeignScan;
2135 1212 : pathnode->path.parent = rel;
2136 1212 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2137 1212 : pathnode->path.param_info = NULL; /* XXX see above */
2138 1212 : pathnode->path.parallel_aware = false;
2139 1212 : pathnode->path.parallel_safe = rel->consider_parallel;
2140 1212 : pathnode->path.parallel_workers = 0;
2141 1212 : pathnode->path.rows = rows;
2142 1212 : pathnode->path.disabled_nodes = disabled_nodes;
2143 1212 : pathnode->path.startup_cost = startup_cost;
2144 1212 : pathnode->path.total_cost = total_cost;
2145 1212 : pathnode->path.pathkeys = pathkeys;
2146 :
2147 1212 : pathnode->fdw_outerpath = fdw_outerpath;
2148 1212 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2149 1212 : pathnode->fdw_private = fdw_private;
2150 :
2151 1212 : return pathnode;
2152 : }
2153 :
2154 : /*
2155 : * create_foreign_upper_path
2156 : * Creates a path corresponding to an upper relation that's computed
2157 : * directly by an FDW, returning the pathnode.
2158 : *
2159 : * This function is never called from core Postgres; rather, it's expected to
2160 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2161 : * We make the FDW supply all fields of the path, since we do not have any way
2162 : * to calculate them in core. However, there is a usually-sane default for
2163 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2164 : */
2165 : ForeignPath *
2166 588 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2167 : PathTarget *target,
2168 : double rows, int disabled_nodes,
2169 : Cost startup_cost, Cost total_cost,
2170 : List *pathkeys,
2171 : Path *fdw_outerpath,
2172 : List *fdw_restrictinfo,
2173 : List *fdw_private)
2174 : {
2175 588 : ForeignPath *pathnode = makeNode(ForeignPath);
2176 :
2177 : /*
2178 : * Upper relations should never have any lateral references, since joining
2179 : * is complete.
2180 : */
2181 : Assert(bms_is_empty(rel->lateral_relids));
2182 :
2183 588 : pathnode->path.pathtype = T_ForeignScan;
2184 588 : pathnode->path.parent = rel;
2185 588 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2186 588 : pathnode->path.param_info = NULL;
2187 588 : pathnode->path.parallel_aware = false;
2188 588 : pathnode->path.parallel_safe = rel->consider_parallel;
2189 588 : pathnode->path.parallel_workers = 0;
2190 588 : pathnode->path.rows = rows;
2191 588 : pathnode->path.disabled_nodes = disabled_nodes;
2192 588 : pathnode->path.startup_cost = startup_cost;
2193 588 : pathnode->path.total_cost = total_cost;
2194 588 : pathnode->path.pathkeys = pathkeys;
2195 :
2196 588 : pathnode->fdw_outerpath = fdw_outerpath;
2197 588 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2198 588 : pathnode->fdw_private = fdw_private;
2199 :
2200 588 : return pathnode;
2201 : }
2202 :
2203 : /*
2204 : * calc_nestloop_required_outer
2205 : * Compute the required_outer set for a nestloop join path
2206 : *
2207 : * Note: when considering a child join, the inputs nonetheless use top-level
2208 : * parent relids
2209 : *
2210 : * Note: result must not share storage with either input
2211 : */
2212 : Relids
2213 3411568 : calc_nestloop_required_outer(Relids outerrelids,
2214 : Relids outer_paramrels,
2215 : Relids innerrelids,
2216 : Relids inner_paramrels)
2217 : {
2218 : Relids required_outer;
2219 :
2220 : /* inner_path can require rels from outer path, but not vice versa */
2221 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2222 : /* easy case if inner path is not parameterized */
2223 3411568 : if (!inner_paramrels)
2224 2367248 : return bms_copy(outer_paramrels);
2225 : /* else, form the union ... */
2226 1044320 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2227 : /* ... and remove any mention of now-satisfied outer rels */
2228 1044320 : required_outer = bms_del_members(required_outer,
2229 : outerrelids);
2230 1044320 : return required_outer;
2231 : }
2232 :
2233 : /*
2234 : * calc_non_nestloop_required_outer
2235 : * Compute the required_outer set for a merge or hash join path
2236 : *
2237 : * Note: result must not share storage with either input
2238 : */
2239 : Relids
2240 2293426 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2241 : {
2242 2293426 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2243 2293426 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2244 : Relids innerrelids PG_USED_FOR_ASSERTS_ONLY;
2245 : Relids outerrelids PG_USED_FOR_ASSERTS_ONLY;
2246 : Relids required_outer;
2247 :
2248 : /*
2249 : * Any parameterization of the input paths refers to topmost parents of
2250 : * the relevant relations, because reparameterize_path_by_child() hasn't
2251 : * been called yet. So we must consider topmost parents of the relations
2252 : * being joined, too, while checking for disallowed parameterization
2253 : * cases.
2254 : */
2255 2293426 : if (inner_path->parent->top_parent_relids)
2256 159170 : innerrelids = inner_path->parent->top_parent_relids;
2257 : else
2258 2134256 : innerrelids = inner_path->parent->relids;
2259 :
2260 2293426 : if (outer_path->parent->top_parent_relids)
2261 159170 : outerrelids = outer_path->parent->top_parent_relids;
2262 : else
2263 2134256 : outerrelids = outer_path->parent->relids;
2264 :
2265 : /* neither path can require rels from the other */
2266 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2267 : Assert(!bms_overlap(inner_paramrels, outerrelids));
2268 : /* form the union ... */
2269 2293426 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2270 : /* we do not need an explicit test for empty; bms_union gets it right */
2271 2293426 : return required_outer;
2272 : }
2273 :
2274 : /*
2275 : * create_nestloop_path
2276 : * Creates a pathnode corresponding to a nestloop join between two
2277 : * relations.
2278 : *
2279 : * 'joinrel' is the join relation.
2280 : * 'jointype' is the type of join required
2281 : * 'workspace' is the result from initial_cost_nestloop
2282 : * 'extra' contains various information about the join
2283 : * 'outer_path' is the outer path
2284 : * 'inner_path' is the inner path
2285 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2286 : * 'pathkeys' are the path keys of the new join path
2287 : * 'required_outer' is the set of required outer rels
2288 : *
2289 : * Returns the resulting path node.
2290 : */
2291 : NestPath *
2292 1482318 : create_nestloop_path(PlannerInfo *root,
2293 : RelOptInfo *joinrel,
2294 : JoinType jointype,
2295 : JoinCostWorkspace *workspace,
2296 : JoinPathExtraData *extra,
2297 : Path *outer_path,
2298 : Path *inner_path,
2299 : List *restrict_clauses,
2300 : List *pathkeys,
2301 : Relids required_outer)
2302 : {
2303 1482318 : NestPath *pathnode = makeNode(NestPath);
2304 1482318 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2305 : Relids outerrelids;
2306 :
2307 : /*
2308 : * Paths are parameterized by top-level parents, so run parameterization
2309 : * tests on the parent relids.
2310 : */
2311 1482318 : if (outer_path->parent->top_parent_relids)
2312 76498 : outerrelids = outer_path->parent->top_parent_relids;
2313 : else
2314 1405820 : outerrelids = outer_path->parent->relids;
2315 :
2316 : /*
2317 : * If the inner path is parameterized by the outer, we must drop any
2318 : * restrict_clauses that are due to be moved into the inner path. We have
2319 : * to do this now, rather than postpone the work till createplan time,
2320 : * because the restrict_clauses list can affect the size and cost
2321 : * estimates for this path. We detect such clauses by checking for serial
2322 : * number match to clauses already enforced in the inner path.
2323 : */
2324 1482318 : if (bms_overlap(inner_req_outer, outerrelids))
2325 : {
2326 401612 : Bitmapset *enforced_serials = get_param_path_clause_serials(inner_path);
2327 401612 : List *jclauses = NIL;
2328 : ListCell *lc;
2329 :
2330 892006 : foreach(lc, restrict_clauses)
2331 : {
2332 490394 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2333 :
2334 490394 : if (!bms_is_member(rinfo->rinfo_serial, enforced_serials))
2335 61656 : jclauses = lappend(jclauses, rinfo);
2336 : }
2337 401612 : restrict_clauses = jclauses;
2338 : }
2339 :
2340 1482318 : pathnode->jpath.path.pathtype = T_NestLoop;
2341 1482318 : pathnode->jpath.path.parent = joinrel;
2342 1482318 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2343 1482318 : pathnode->jpath.path.param_info =
2344 1482318 : get_joinrel_parampathinfo(root,
2345 : joinrel,
2346 : outer_path,
2347 : inner_path,
2348 : extra->sjinfo,
2349 : required_outer,
2350 : &restrict_clauses);
2351 1482318 : pathnode->jpath.path.parallel_aware = false;
2352 4317208 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2353 1482318 : outer_path->parallel_safe && inner_path->parallel_safe;
2354 : /* This is a foolish way to estimate parallel_workers, but for now... */
2355 1482318 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2356 1482318 : pathnode->jpath.path.pathkeys = pathkeys;
2357 1482318 : pathnode->jpath.jointype = jointype;
2358 1482318 : pathnode->jpath.inner_unique = extra->inner_unique;
2359 1482318 : pathnode->jpath.outerjoinpath = outer_path;
2360 1482318 : pathnode->jpath.innerjoinpath = inner_path;
2361 1482318 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2362 :
2363 1482318 : final_cost_nestloop(root, pathnode, workspace, extra);
2364 :
2365 1482318 : return pathnode;
2366 : }
2367 :
2368 : /*
2369 : * create_mergejoin_path
2370 : * Creates a pathnode corresponding to a mergejoin join between
2371 : * two relations
2372 : *
2373 : * 'joinrel' is the join relation
2374 : * 'jointype' is the type of join required
2375 : * 'workspace' is the result from initial_cost_mergejoin
2376 : * 'extra' contains various information about the join
2377 : * 'outer_path' is the outer path
2378 : * 'inner_path' is the inner path
2379 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2380 : * 'pathkeys' are the path keys of the new join path
2381 : * 'required_outer' is the set of required outer rels
2382 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2383 : * (this should be a subset of the restrict_clauses list)
2384 : * 'outersortkeys' are the sort varkeys for the outer relation
2385 : * 'innersortkeys' are the sort varkeys for the inner relation
2386 : * 'outer_presorted_keys' is the number of presorted keys of the outer path
2387 : */
2388 : MergePath *
2389 466682 : create_mergejoin_path(PlannerInfo *root,
2390 : RelOptInfo *joinrel,
2391 : JoinType jointype,
2392 : JoinCostWorkspace *workspace,
2393 : JoinPathExtraData *extra,
2394 : Path *outer_path,
2395 : Path *inner_path,
2396 : List *restrict_clauses,
2397 : List *pathkeys,
2398 : Relids required_outer,
2399 : List *mergeclauses,
2400 : List *outersortkeys,
2401 : List *innersortkeys,
2402 : int outer_presorted_keys)
2403 : {
2404 466682 : MergePath *pathnode = makeNode(MergePath);
2405 :
2406 466682 : pathnode->jpath.path.pathtype = T_MergeJoin;
2407 466682 : pathnode->jpath.path.parent = joinrel;
2408 466682 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2409 466682 : pathnode->jpath.path.param_info =
2410 466682 : get_joinrel_parampathinfo(root,
2411 : joinrel,
2412 : outer_path,
2413 : inner_path,
2414 : extra->sjinfo,
2415 : required_outer,
2416 : &restrict_clauses);
2417 466682 : pathnode->jpath.path.parallel_aware = false;
2418 1362744 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2419 466682 : outer_path->parallel_safe && inner_path->parallel_safe;
2420 : /* This is a foolish way to estimate parallel_workers, but for now... */
2421 466682 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2422 466682 : pathnode->jpath.path.pathkeys = pathkeys;
2423 466682 : pathnode->jpath.jointype = jointype;
2424 466682 : pathnode->jpath.inner_unique = extra->inner_unique;
2425 466682 : pathnode->jpath.outerjoinpath = outer_path;
2426 466682 : pathnode->jpath.innerjoinpath = inner_path;
2427 466682 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2428 466682 : pathnode->path_mergeclauses = mergeclauses;
2429 466682 : pathnode->outersortkeys = outersortkeys;
2430 466682 : pathnode->innersortkeys = innersortkeys;
2431 466682 : pathnode->outer_presorted_keys = outer_presorted_keys;
2432 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2433 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2434 :
2435 466682 : final_cost_mergejoin(root, pathnode, workspace, extra);
2436 :
2437 466682 : return pathnode;
2438 : }
2439 :
2440 : /*
2441 : * create_hashjoin_path
2442 : * Creates a pathnode corresponding to a hash join between two relations.
2443 : *
2444 : * 'joinrel' is the join relation
2445 : * 'jointype' is the type of join required
2446 : * 'workspace' is the result from initial_cost_hashjoin
2447 : * 'extra' contains various information about the join
2448 : * 'outer_path' is the cheapest outer path
2449 : * 'inner_path' is the cheapest inner path
2450 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2451 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2452 : * 'required_outer' is the set of required outer rels
2453 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2454 : * (this should be a subset of the restrict_clauses list)
2455 : */
2456 : HashPath *
2457 458806 : create_hashjoin_path(PlannerInfo *root,
2458 : RelOptInfo *joinrel,
2459 : JoinType jointype,
2460 : JoinCostWorkspace *workspace,
2461 : JoinPathExtraData *extra,
2462 : Path *outer_path,
2463 : Path *inner_path,
2464 : bool parallel_hash,
2465 : List *restrict_clauses,
2466 : Relids required_outer,
2467 : List *hashclauses)
2468 : {
2469 458806 : HashPath *pathnode = makeNode(HashPath);
2470 :
2471 458806 : pathnode->jpath.path.pathtype = T_HashJoin;
2472 458806 : pathnode->jpath.path.parent = joinrel;
2473 458806 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2474 458806 : pathnode->jpath.path.param_info =
2475 458806 : get_joinrel_parampathinfo(root,
2476 : joinrel,
2477 : outer_path,
2478 : inner_path,
2479 : extra->sjinfo,
2480 : required_outer,
2481 : &restrict_clauses);
2482 458806 : pathnode->jpath.path.parallel_aware =
2483 458806 : joinrel->consider_parallel && parallel_hash;
2484 1339284 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2485 458806 : outer_path->parallel_safe && inner_path->parallel_safe;
2486 : /* This is a foolish way to estimate parallel_workers, but for now... */
2487 458806 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2488 :
2489 : /*
2490 : * A hashjoin never has pathkeys, since its output ordering is
2491 : * unpredictable due to possible batching. XXX If the inner relation is
2492 : * small enough, we could instruct the executor that it must not batch,
2493 : * and then we could assume that the output inherits the outer relation's
2494 : * ordering, which might save a sort step. However there is considerable
2495 : * downside if our estimate of the inner relation size is badly off. For
2496 : * the moment we don't risk it. (Note also that if we wanted to take this
2497 : * seriously, joinpath.c would have to consider many more paths for the
2498 : * outer rel than it does now.)
2499 : */
2500 458806 : pathnode->jpath.path.pathkeys = NIL;
2501 458806 : pathnode->jpath.jointype = jointype;
2502 458806 : pathnode->jpath.inner_unique = extra->inner_unique;
2503 458806 : pathnode->jpath.outerjoinpath = outer_path;
2504 458806 : pathnode->jpath.innerjoinpath = inner_path;
2505 458806 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2506 458806 : pathnode->path_hashclauses = hashclauses;
2507 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2508 :
2509 458806 : final_cost_hashjoin(root, pathnode, workspace, extra);
2510 :
2511 458806 : return pathnode;
2512 : }
2513 :
2514 : /*
2515 : * create_projection_path
2516 : * Creates a pathnode that represents performing a projection.
2517 : *
2518 : * 'rel' is the parent relation associated with the result
2519 : * 'subpath' is the path representing the source of data
2520 : * 'target' is the PathTarget to be computed
2521 : */
2522 : ProjectionPath *
2523 428674 : create_projection_path(PlannerInfo *root,
2524 : RelOptInfo *rel,
2525 : Path *subpath,
2526 : PathTarget *target)
2527 : {
2528 428674 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2529 : PathTarget *oldtarget;
2530 :
2531 : /*
2532 : * We mustn't put a ProjectionPath directly above another; it's useless
2533 : * and will confuse create_projection_plan. Rather than making sure all
2534 : * callers handle that, let's implement it here, by stripping off any
2535 : * ProjectionPath in what we're given. Given this rule, there won't be
2536 : * more than one.
2537 : */
2538 428674 : if (IsA(subpath, ProjectionPath))
2539 : {
2540 12 : ProjectionPath *subpp = (ProjectionPath *) subpath;
2541 :
2542 : Assert(subpp->path.parent == rel);
2543 12 : subpath = subpp->subpath;
2544 : Assert(!IsA(subpath, ProjectionPath));
2545 : }
2546 :
2547 428674 : pathnode->path.pathtype = T_Result;
2548 428674 : pathnode->path.parent = rel;
2549 428674 : pathnode->path.pathtarget = target;
2550 428674 : pathnode->path.param_info = subpath->param_info;
2551 428674 : pathnode->path.parallel_aware = false;
2552 1001830 : pathnode->path.parallel_safe = rel->consider_parallel &&
2553 564858 : subpath->parallel_safe &&
2554 136184 : is_parallel_safe(root, (Node *) target->exprs);
2555 428674 : pathnode->path.parallel_workers = subpath->parallel_workers;
2556 : /* Projection does not change the sort order */
2557 428674 : pathnode->path.pathkeys = subpath->pathkeys;
2558 :
2559 428674 : pathnode->subpath = subpath;
2560 :
2561 : /*
2562 : * We might not need a separate Result node. If the input plan node type
2563 : * can project, we can just tell it to project something else. Or, if it
2564 : * can't project but the desired target has the same expression list as
2565 : * what the input will produce anyway, we can still give it the desired
2566 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2567 : * Note: in the latter case, create_projection_plan has to recheck our
2568 : * conclusion; see comments therein.
2569 : */
2570 428674 : oldtarget = subpath->pathtarget;
2571 445828 : if (is_projection_capable_path(subpath) ||
2572 17154 : equal(oldtarget->exprs, target->exprs))
2573 : {
2574 : /* No separate Result node needed */
2575 413522 : pathnode->dummypp = true;
2576 :
2577 : /*
2578 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2579 : */
2580 413522 : pathnode->path.rows = subpath->rows;
2581 413522 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2582 413522 : pathnode->path.startup_cost = subpath->startup_cost +
2583 413522 : (target->cost.startup - oldtarget->cost.startup);
2584 413522 : pathnode->path.total_cost = subpath->total_cost +
2585 413522 : (target->cost.startup - oldtarget->cost.startup) +
2586 413522 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2587 : }
2588 : else
2589 : {
2590 : /* We really do need the Result node */
2591 15152 : pathnode->dummypp = false;
2592 :
2593 : /*
2594 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2595 : * evaluating the tlist. There is no qual to worry about.
2596 : */
2597 15152 : pathnode->path.rows = subpath->rows;
2598 15152 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2599 15152 : pathnode->path.startup_cost = subpath->startup_cost +
2600 15152 : target->cost.startup;
2601 15152 : pathnode->path.total_cost = subpath->total_cost +
2602 15152 : target->cost.startup +
2603 15152 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2604 : }
2605 :
2606 428674 : return pathnode;
2607 : }
2608 :
2609 : /*
2610 : * apply_projection_to_path
2611 : * Add a projection step, or just apply the target directly to given path.
2612 : *
2613 : * This has the same net effect as create_projection_path(), except that if
2614 : * a separate Result plan node isn't needed, we just replace the given path's
2615 : * pathtarget with the desired one. This must be used only when the caller
2616 : * knows that the given path isn't referenced elsewhere and so can be modified
2617 : * in-place.
2618 : *
2619 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2620 : * new target down to its input as well; this is a yet more invasive
2621 : * modification of the input path, which create_projection_path() can't do.
2622 : *
2623 : * Note that we mustn't change the source path's parent link; so when it is
2624 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2625 : * not caused any trouble.
2626 : *
2627 : * 'rel' is the parent relation associated with the result
2628 : * 'path' is the path representing the source of data
2629 : * 'target' is the PathTarget to be computed
2630 : */
2631 : Path *
2632 14004 : apply_projection_to_path(PlannerInfo *root,
2633 : RelOptInfo *rel,
2634 : Path *path,
2635 : PathTarget *target)
2636 : {
2637 : QualCost oldcost;
2638 :
2639 : /*
2640 : * If given path can't project, we might need a Result node, so make a
2641 : * separate ProjectionPath.
2642 : */
2643 14004 : if (!is_projection_capable_path(path))
2644 1484 : return (Path *) create_projection_path(root, rel, path, target);
2645 :
2646 : /*
2647 : * We can just jam the desired tlist into the existing path, being sure to
2648 : * update its cost estimates appropriately.
2649 : */
2650 12520 : oldcost = path->pathtarget->cost;
2651 12520 : path->pathtarget = target;
2652 :
2653 12520 : path->startup_cost += target->cost.startup - oldcost.startup;
2654 12520 : path->total_cost += target->cost.startup - oldcost.startup +
2655 12520 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2656 :
2657 : /*
2658 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2659 : * arrange for the subpath to return the required target list so that
2660 : * workers can help project. But if there is something that is not
2661 : * parallel-safe in the target expressions, then we can't.
2662 : */
2663 12544 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
2664 24 : is_parallel_safe(root, (Node *) target->exprs))
2665 : {
2666 : /*
2667 : * We always use create_projection_path here, even if the subpath is
2668 : * projection-capable, so as to avoid modifying the subpath in place.
2669 : * It seems unlikely at present that there could be any other
2670 : * references to the subpath, but better safe than sorry.
2671 : *
2672 : * Note that we don't change the parallel path's cost estimates; it
2673 : * might be appropriate to do so, to reflect the fact that the bulk of
2674 : * the target evaluation will happen in workers.
2675 : */
2676 24 : if (IsA(path, GatherPath))
2677 : {
2678 0 : GatherPath *gpath = (GatherPath *) path;
2679 :
2680 0 : gpath->subpath = (Path *)
2681 0 : create_projection_path(root,
2682 0 : gpath->subpath->parent,
2683 : gpath->subpath,
2684 : target);
2685 : }
2686 : else
2687 : {
2688 24 : GatherMergePath *gmpath = (GatherMergePath *) path;
2689 :
2690 24 : gmpath->subpath = (Path *)
2691 24 : create_projection_path(root,
2692 24 : gmpath->subpath->parent,
2693 : gmpath->subpath,
2694 : target);
2695 : }
2696 : }
2697 12496 : else if (path->parallel_safe &&
2698 4752 : !is_parallel_safe(root, (Node *) target->exprs))
2699 : {
2700 : /*
2701 : * We're inserting a parallel-restricted target list into a path
2702 : * currently marked parallel-safe, so we have to mark it as no longer
2703 : * safe.
2704 : */
2705 12 : path->parallel_safe = false;
2706 : }
2707 :
2708 12520 : return path;
2709 : }
2710 :
2711 : /*
2712 : * create_set_projection_path
2713 : * Creates a pathnode that represents performing a projection that
2714 : * includes set-returning functions.
2715 : *
2716 : * 'rel' is the parent relation associated with the result
2717 : * 'subpath' is the path representing the source of data
2718 : * 'target' is the PathTarget to be computed
2719 : */
2720 : ProjectSetPath *
2721 12164 : create_set_projection_path(PlannerInfo *root,
2722 : RelOptInfo *rel,
2723 : Path *subpath,
2724 : PathTarget *target)
2725 : {
2726 12164 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2727 : double tlist_rows;
2728 : ListCell *lc;
2729 :
2730 12164 : pathnode->path.pathtype = T_ProjectSet;
2731 12164 : pathnode->path.parent = rel;
2732 12164 : pathnode->path.pathtarget = target;
2733 : /* For now, assume we are above any joins, so no parameterization */
2734 12164 : pathnode->path.param_info = NULL;
2735 12164 : pathnode->path.parallel_aware = false;
2736 28900 : pathnode->path.parallel_safe = rel->consider_parallel &&
2737 16700 : subpath->parallel_safe &&
2738 4536 : is_parallel_safe(root, (Node *) target->exprs);
2739 12164 : pathnode->path.parallel_workers = subpath->parallel_workers;
2740 : /* Projection does not change the sort order XXX? */
2741 12164 : pathnode->path.pathkeys = subpath->pathkeys;
2742 :
2743 12164 : pathnode->subpath = subpath;
2744 :
2745 : /*
2746 : * Estimate number of rows produced by SRFs for each row of input; if
2747 : * there's more than one in this node, use the maximum.
2748 : */
2749 12164 : tlist_rows = 1;
2750 26390 : foreach(lc, target->exprs)
2751 : {
2752 14226 : Node *node = (Node *) lfirst(lc);
2753 : double itemrows;
2754 :
2755 14226 : itemrows = expression_returns_set_rows(root, node);
2756 14226 : if (tlist_rows < itemrows)
2757 11712 : tlist_rows = itemrows;
2758 : }
2759 :
2760 : /*
2761 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2762 : * per input row, and half of cpu_tuple_cost for each added output row.
2763 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2764 : * this estimate later.
2765 : */
2766 12164 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2767 12164 : pathnode->path.rows = subpath->rows * tlist_rows;
2768 12164 : pathnode->path.startup_cost = subpath->startup_cost +
2769 12164 : target->cost.startup;
2770 12164 : pathnode->path.total_cost = subpath->total_cost +
2771 12164 : target->cost.startup +
2772 12164 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2773 12164 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2774 :
2775 12164 : return pathnode;
2776 : }
2777 :
2778 : /*
2779 : * create_incremental_sort_path
2780 : * Creates a pathnode that represents performing an incremental sort.
2781 : *
2782 : * 'rel' is the parent relation associated with the result
2783 : * 'subpath' is the path representing the source of data
2784 : * 'pathkeys' represents the desired sort order
2785 : * 'presorted_keys' is the number of keys by which the input path is
2786 : * already sorted
2787 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2788 : * or -1 if no LIMIT or couldn't estimate
2789 : */
2790 : IncrementalSortPath *
2791 10778 : create_incremental_sort_path(PlannerInfo *root,
2792 : RelOptInfo *rel,
2793 : Path *subpath,
2794 : List *pathkeys,
2795 : int presorted_keys,
2796 : double limit_tuples)
2797 : {
2798 10778 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
2799 10778 : SortPath *pathnode = &sort->spath;
2800 :
2801 10778 : pathnode->path.pathtype = T_IncrementalSort;
2802 10778 : pathnode->path.parent = rel;
2803 : /* Sort doesn't project, so use source path's pathtarget */
2804 10778 : pathnode->path.pathtarget = subpath->pathtarget;
2805 10778 : pathnode->path.param_info = subpath->param_info;
2806 10778 : pathnode->path.parallel_aware = false;
2807 16586 : pathnode->path.parallel_safe = rel->consider_parallel &&
2808 5808 : subpath->parallel_safe;
2809 10778 : pathnode->path.parallel_workers = subpath->parallel_workers;
2810 10778 : pathnode->path.pathkeys = pathkeys;
2811 :
2812 10778 : pathnode->subpath = subpath;
2813 :
2814 10778 : cost_incremental_sort(&pathnode->path,
2815 : root, pathkeys, presorted_keys,
2816 : subpath->disabled_nodes,
2817 : subpath->startup_cost,
2818 : subpath->total_cost,
2819 : subpath->rows,
2820 10778 : subpath->pathtarget->width,
2821 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2822 : work_mem, limit_tuples);
2823 :
2824 10778 : sort->nPresortedCols = presorted_keys;
2825 :
2826 10778 : return sort;
2827 : }
2828 :
2829 : /*
2830 : * create_sort_path
2831 : * Creates a pathnode that represents performing an explicit sort.
2832 : *
2833 : * 'rel' is the parent relation associated with the result
2834 : * 'subpath' is the path representing the source of data
2835 : * 'pathkeys' represents the desired sort order
2836 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2837 : * or -1 if no LIMIT or couldn't estimate
2838 : */
2839 : SortPath *
2840 118442 : create_sort_path(PlannerInfo *root,
2841 : RelOptInfo *rel,
2842 : Path *subpath,
2843 : List *pathkeys,
2844 : double limit_tuples)
2845 : {
2846 118442 : SortPath *pathnode = makeNode(SortPath);
2847 :
2848 118442 : pathnode->path.pathtype = T_Sort;
2849 118442 : pathnode->path.parent = rel;
2850 : /* Sort doesn't project, so use source path's pathtarget */
2851 118442 : pathnode->path.pathtarget = subpath->pathtarget;
2852 118442 : pathnode->path.param_info = subpath->param_info;
2853 118442 : pathnode->path.parallel_aware = false;
2854 206664 : pathnode->path.parallel_safe = rel->consider_parallel &&
2855 88222 : subpath->parallel_safe;
2856 118442 : pathnode->path.parallel_workers = subpath->parallel_workers;
2857 118442 : pathnode->path.pathkeys = pathkeys;
2858 :
2859 118442 : pathnode->subpath = subpath;
2860 :
2861 118442 : cost_sort(&pathnode->path, root, pathkeys,
2862 : subpath->disabled_nodes,
2863 : subpath->total_cost,
2864 : subpath->rows,
2865 118442 : subpath->pathtarget->width,
2866 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2867 : work_mem, limit_tuples);
2868 :
2869 118442 : return pathnode;
2870 : }
2871 :
2872 : /*
2873 : * create_group_path
2874 : * Creates a pathnode that represents performing grouping of presorted input
2875 : *
2876 : * 'rel' is the parent relation associated with the result
2877 : * 'subpath' is the path representing the source of data
2878 : * 'target' is the PathTarget to be computed
2879 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2880 : * 'qual' is the HAVING quals if any
2881 : * 'numGroups' is the estimated number of groups
2882 : */
2883 : GroupPath *
2884 1226 : create_group_path(PlannerInfo *root,
2885 : RelOptInfo *rel,
2886 : Path *subpath,
2887 : List *groupClause,
2888 : List *qual,
2889 : double numGroups)
2890 : {
2891 1226 : GroupPath *pathnode = makeNode(GroupPath);
2892 1226 : PathTarget *target = rel->reltarget;
2893 :
2894 1226 : pathnode->path.pathtype = T_Group;
2895 1226 : pathnode->path.parent = rel;
2896 1226 : pathnode->path.pathtarget = target;
2897 : /* For now, assume we are above any joins, so no parameterization */
2898 1226 : pathnode->path.param_info = NULL;
2899 1226 : pathnode->path.parallel_aware = false;
2900 1970 : pathnode->path.parallel_safe = rel->consider_parallel &&
2901 744 : subpath->parallel_safe;
2902 1226 : pathnode->path.parallel_workers = subpath->parallel_workers;
2903 : /* Group doesn't change sort ordering */
2904 1226 : pathnode->path.pathkeys = subpath->pathkeys;
2905 :
2906 1226 : pathnode->subpath = subpath;
2907 :
2908 1226 : pathnode->groupClause = groupClause;
2909 1226 : pathnode->qual = qual;
2910 :
2911 1226 : cost_group(&pathnode->path, root,
2912 : list_length(groupClause),
2913 : numGroups,
2914 : qual,
2915 : subpath->disabled_nodes,
2916 : subpath->startup_cost, subpath->total_cost,
2917 : subpath->rows);
2918 :
2919 : /* add tlist eval cost for each output row */
2920 1226 : pathnode->path.startup_cost += target->cost.startup;
2921 1226 : pathnode->path.total_cost += target->cost.startup +
2922 1226 : target->cost.per_tuple * pathnode->path.rows;
2923 :
2924 1226 : return pathnode;
2925 : }
2926 :
2927 : /*
2928 : * create_unique_path
2929 : * Creates a pathnode that represents performing an explicit Unique step
2930 : * on presorted input.
2931 : *
2932 : * 'rel' is the parent relation associated with the result
2933 : * 'subpath' is the path representing the source of data
2934 : * 'numCols' is the number of grouping columns
2935 : * 'numGroups' is the estimated number of groups
2936 : *
2937 : * The input path must be sorted on the grouping columns, plus possibly
2938 : * additional columns; so the first numCols pathkeys are the grouping columns
2939 : */
2940 : UniquePath *
2941 22864 : create_unique_path(PlannerInfo *root,
2942 : RelOptInfo *rel,
2943 : Path *subpath,
2944 : int numCols,
2945 : double numGroups)
2946 : {
2947 22864 : UniquePath *pathnode = makeNode(UniquePath);
2948 :
2949 22864 : pathnode->path.pathtype = T_Unique;
2950 22864 : pathnode->path.parent = rel;
2951 : /* Unique doesn't project, so use source path's pathtarget */
2952 22864 : pathnode->path.pathtarget = subpath->pathtarget;
2953 22864 : pathnode->path.param_info = subpath->param_info;
2954 22864 : pathnode->path.parallel_aware = false;
2955 41380 : pathnode->path.parallel_safe = rel->consider_parallel &&
2956 18516 : subpath->parallel_safe;
2957 22864 : pathnode->path.parallel_workers = subpath->parallel_workers;
2958 : /* Unique doesn't change the input ordering */
2959 22864 : pathnode->path.pathkeys = subpath->pathkeys;
2960 :
2961 22864 : pathnode->subpath = subpath;
2962 22864 : pathnode->numkeys = numCols;
2963 :
2964 : /*
2965 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
2966 : * all columns get compared at most of the tuples. (XXX probably this is
2967 : * an overestimate.)
2968 : */
2969 22864 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2970 22864 : pathnode->path.startup_cost = subpath->startup_cost;
2971 22864 : pathnode->path.total_cost = subpath->total_cost +
2972 22864 : cpu_operator_cost * subpath->rows * numCols;
2973 22864 : pathnode->path.rows = numGroups;
2974 :
2975 22864 : return pathnode;
2976 : }
2977 :
2978 : /*
2979 : * create_agg_path
2980 : * Creates a pathnode that represents performing aggregation/grouping
2981 : *
2982 : * 'rel' is the parent relation associated with the result
2983 : * 'subpath' is the path representing the source of data
2984 : * 'target' is the PathTarget to be computed
2985 : * 'aggstrategy' is the Agg node's basic implementation strategy
2986 : * 'aggsplit' is the Agg node's aggregate-splitting mode
2987 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2988 : * 'qual' is the HAVING quals if any
2989 : * 'aggcosts' contains cost info about the aggregate functions to be computed
2990 : * 'numGroups' is the estimated number of groups (1 if not grouping)
2991 : */
2992 : AggPath *
2993 87702 : create_agg_path(PlannerInfo *root,
2994 : RelOptInfo *rel,
2995 : Path *subpath,
2996 : PathTarget *target,
2997 : AggStrategy aggstrategy,
2998 : AggSplit aggsplit,
2999 : List *groupClause,
3000 : List *qual,
3001 : const AggClauseCosts *aggcosts,
3002 : double numGroups)
3003 : {
3004 87702 : AggPath *pathnode = makeNode(AggPath);
3005 :
3006 87702 : pathnode->path.pathtype = T_Agg;
3007 87702 : pathnode->path.parent = rel;
3008 87702 : pathnode->path.pathtarget = target;
3009 87702 : pathnode->path.param_info = subpath->param_info;
3010 87702 : pathnode->path.parallel_aware = false;
3011 149248 : pathnode->path.parallel_safe = rel->consider_parallel &&
3012 61546 : subpath->parallel_safe;
3013 87702 : pathnode->path.parallel_workers = subpath->parallel_workers;
3014 :
3015 87702 : if (aggstrategy == AGG_SORTED)
3016 : {
3017 : /*
3018 : * Attempt to preserve the order of the subpath. Additional pathkeys
3019 : * may have been added in adjust_group_pathkeys_for_groupagg() to
3020 : * support ORDER BY / DISTINCT aggregates. Pathkeys added there
3021 : * belong to columns within the aggregate function, so we must strip
3022 : * these additional pathkeys off as those columns are unavailable
3023 : * above the aggregate node.
3024 : */
3025 14472 : if (list_length(subpath->pathkeys) > root->num_groupby_pathkeys)
3026 808 : pathnode->path.pathkeys = list_copy_head(subpath->pathkeys,
3027 : root->num_groupby_pathkeys);
3028 : else
3029 13664 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3030 : }
3031 : else
3032 73230 : pathnode->path.pathkeys = NIL; /* output is unordered */
3033 :
3034 87702 : pathnode->subpath = subpath;
3035 :
3036 87702 : pathnode->aggstrategy = aggstrategy;
3037 87702 : pathnode->aggsplit = aggsplit;
3038 87702 : pathnode->numGroups = numGroups;
3039 87702 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
3040 87702 : pathnode->groupClause = groupClause;
3041 87702 : pathnode->qual = qual;
3042 :
3043 87702 : cost_agg(&pathnode->path, root,
3044 : aggstrategy, aggcosts,
3045 : list_length(groupClause), numGroups,
3046 : qual,
3047 : subpath->disabled_nodes,
3048 : subpath->startup_cost, subpath->total_cost,
3049 87702 : subpath->rows, subpath->pathtarget->width);
3050 :
3051 : /* add tlist eval cost for each output row */
3052 87702 : pathnode->path.startup_cost += target->cost.startup;
3053 87702 : pathnode->path.total_cost += target->cost.startup +
3054 87702 : target->cost.per_tuple * pathnode->path.rows;
3055 :
3056 87702 : return pathnode;
3057 : }
3058 :
3059 : /*
3060 : * create_groupingsets_path
3061 : * Creates a pathnode that represents performing GROUPING SETS aggregation
3062 : *
3063 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3064 : * The input path's result must be sorted to match the last entry in
3065 : * rollup_groupclauses.
3066 : *
3067 : * 'rel' is the parent relation associated with the result
3068 : * 'subpath' is the path representing the source of data
3069 : * 'target' is the PathTarget to be computed
3070 : * 'having_qual' is the HAVING quals if any
3071 : * 'rollups' is a list of RollupData nodes
3072 : * 'agg_costs' contains cost info about the aggregate functions to be computed
3073 : */
3074 : GroupingSetsPath *
3075 2406 : create_groupingsets_path(PlannerInfo *root,
3076 : RelOptInfo *rel,
3077 : Path *subpath,
3078 : List *having_qual,
3079 : AggStrategy aggstrategy,
3080 : List *rollups,
3081 : const AggClauseCosts *agg_costs)
3082 : {
3083 2406 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
3084 2406 : PathTarget *target = rel->reltarget;
3085 : ListCell *lc;
3086 2406 : bool is_first = true;
3087 2406 : bool is_first_sort = true;
3088 :
3089 : /* The topmost generated Plan node will be an Agg */
3090 2406 : pathnode->path.pathtype = T_Agg;
3091 2406 : pathnode->path.parent = rel;
3092 2406 : pathnode->path.pathtarget = target;
3093 2406 : pathnode->path.param_info = subpath->param_info;
3094 2406 : pathnode->path.parallel_aware = false;
3095 3552 : pathnode->path.parallel_safe = rel->consider_parallel &&
3096 1146 : subpath->parallel_safe;
3097 2406 : pathnode->path.parallel_workers = subpath->parallel_workers;
3098 2406 : pathnode->subpath = subpath;
3099 :
3100 : /*
3101 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3102 : * to AGG_HASHED, here if possible.
3103 : */
3104 3438 : if (aggstrategy == AGG_SORTED &&
3105 1032 : list_length(rollups) == 1 &&
3106 534 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3107 60 : aggstrategy = AGG_PLAIN;
3108 :
3109 3440 : if (aggstrategy == AGG_MIXED &&
3110 1034 : list_length(rollups) == 1)
3111 0 : aggstrategy = AGG_HASHED;
3112 :
3113 : /*
3114 : * Output will be in sorted order by group_pathkeys if, and only if, there
3115 : * is a single rollup operation on a non-empty list of grouping
3116 : * expressions.
3117 : */
3118 2406 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3119 474 : pathnode->path.pathkeys = root->group_pathkeys;
3120 : else
3121 1932 : pathnode->path.pathkeys = NIL;
3122 :
3123 2406 : pathnode->aggstrategy = aggstrategy;
3124 2406 : pathnode->rollups = rollups;
3125 2406 : pathnode->qual = having_qual;
3126 2406 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3127 :
3128 : Assert(rollups != NIL);
3129 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3130 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3131 :
3132 8222 : foreach(lc, rollups)
3133 : {
3134 5816 : RollupData *rollup = lfirst(lc);
3135 5816 : List *gsets = rollup->gsets;
3136 5816 : int numGroupCols = list_length(linitial(gsets));
3137 :
3138 : /*
3139 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3140 : * (already-sorted) input, and following ones do their own sort.
3141 : *
3142 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3143 : *
3144 : * In AGG_MIXED mode, the first rollups are hashed, the first
3145 : * non-hashed one takes the (already-sorted) input, and following ones
3146 : * do their own sort.
3147 : */
3148 5816 : if (is_first)
3149 : {
3150 2406 : cost_agg(&pathnode->path, root,
3151 : aggstrategy,
3152 : agg_costs,
3153 : numGroupCols,
3154 : rollup->numGroups,
3155 : having_qual,
3156 : subpath->disabled_nodes,
3157 : subpath->startup_cost,
3158 : subpath->total_cost,
3159 : subpath->rows,
3160 2406 : subpath->pathtarget->width);
3161 2406 : is_first = false;
3162 2406 : if (!rollup->is_hashed)
3163 1032 : is_first_sort = false;
3164 : }
3165 : else
3166 : {
3167 : Path sort_path; /* dummy for result of cost_sort */
3168 : Path agg_path; /* dummy for result of cost_agg */
3169 :
3170 3410 : if (rollup->is_hashed || is_first_sort)
3171 : {
3172 : /*
3173 : * Account for cost of aggregation, but don't charge input
3174 : * cost again
3175 : */
3176 2612 : cost_agg(&agg_path, root,
3177 2612 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3178 : agg_costs,
3179 : numGroupCols,
3180 : rollup->numGroups,
3181 : having_qual,
3182 : 0, 0.0, 0.0,
3183 : subpath->rows,
3184 2612 : subpath->pathtarget->width);
3185 2612 : if (!rollup->is_hashed)
3186 1034 : is_first_sort = false;
3187 : }
3188 : else
3189 : {
3190 : /* Account for cost of sort, but don't charge input cost again */
3191 798 : cost_sort(&sort_path, root, NIL, 0,
3192 : 0.0,
3193 : subpath->rows,
3194 798 : subpath->pathtarget->width,
3195 : 0.0,
3196 : work_mem,
3197 : -1.0);
3198 :
3199 : /* Account for cost of aggregation */
3200 :
3201 798 : cost_agg(&agg_path, root,
3202 : AGG_SORTED,
3203 : agg_costs,
3204 : numGroupCols,
3205 : rollup->numGroups,
3206 : having_qual,
3207 : sort_path.disabled_nodes,
3208 : sort_path.startup_cost,
3209 : sort_path.total_cost,
3210 : sort_path.rows,
3211 798 : subpath->pathtarget->width);
3212 : }
3213 :
3214 3410 : pathnode->path.disabled_nodes += agg_path.disabled_nodes;
3215 3410 : pathnode->path.total_cost += agg_path.total_cost;
3216 3410 : pathnode->path.rows += agg_path.rows;
3217 : }
3218 : }
3219 :
3220 : /* add tlist eval cost for each output row */
3221 2406 : pathnode->path.startup_cost += target->cost.startup;
3222 2406 : pathnode->path.total_cost += target->cost.startup +
3223 2406 : target->cost.per_tuple * pathnode->path.rows;
3224 :
3225 2406 : return pathnode;
3226 : }
3227 :
3228 : /*
3229 : * create_minmaxagg_path
3230 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3231 : *
3232 : * 'rel' is the parent relation associated with the result
3233 : * 'target' is the PathTarget to be computed
3234 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3235 : * 'quals' is the HAVING quals if any
3236 : */
3237 : MinMaxAggPath *
3238 422 : create_minmaxagg_path(PlannerInfo *root,
3239 : RelOptInfo *rel,
3240 : PathTarget *target,
3241 : List *mmaggregates,
3242 : List *quals)
3243 : {
3244 422 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3245 : Cost initplan_cost;
3246 422 : int initplan_disabled_nodes = 0;
3247 : ListCell *lc;
3248 :
3249 : /* The topmost generated Plan node will be a Result */
3250 422 : pathnode->path.pathtype = T_Result;
3251 422 : pathnode->path.parent = rel;
3252 422 : pathnode->path.pathtarget = target;
3253 : /* For now, assume we are above any joins, so no parameterization */
3254 422 : pathnode->path.param_info = NULL;
3255 422 : pathnode->path.parallel_aware = false;
3256 422 : pathnode->path.parallel_safe = true; /* might change below */
3257 422 : pathnode->path.parallel_workers = 0;
3258 : /* Result is one unordered row */
3259 422 : pathnode->path.rows = 1;
3260 422 : pathnode->path.pathkeys = NIL;
3261 :
3262 422 : pathnode->mmaggregates = mmaggregates;
3263 422 : pathnode->quals = quals;
3264 :
3265 : /* Calculate cost of all the initplans, and check parallel safety */
3266 422 : initplan_cost = 0;
3267 880 : foreach(lc, mmaggregates)
3268 : {
3269 458 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3270 :
3271 458 : initplan_disabled_nodes += mminfo->path->disabled_nodes;
3272 458 : initplan_cost += mminfo->pathcost;
3273 458 : if (!mminfo->path->parallel_safe)
3274 110 : pathnode->path.parallel_safe = false;
3275 : }
3276 :
3277 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3278 422 : pathnode->path.disabled_nodes = initplan_disabled_nodes;
3279 422 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3280 422 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3281 422 : target->cost.per_tuple + cpu_tuple_cost;
3282 :
3283 : /*
3284 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3285 : * rowcount estimate should be 1 no matter what the qual is.
3286 : */
3287 422 : if (quals)
3288 : {
3289 : QualCost qual_cost;
3290 :
3291 0 : cost_qual_eval(&qual_cost, quals, root);
3292 0 : pathnode->path.startup_cost += qual_cost.startup;
3293 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3294 : }
3295 :
3296 : /*
3297 : * If the initplans were all parallel-safe, also check safety of the
3298 : * target and quals. (The Result node itself isn't parallelizable, but if
3299 : * we are in a subquery then it can be useful for the outer query to know
3300 : * that this one is parallel-safe.)
3301 : */
3302 422 : if (pathnode->path.parallel_safe)
3303 312 : pathnode->path.parallel_safe =
3304 624 : is_parallel_safe(root, (Node *) target->exprs) &&
3305 312 : is_parallel_safe(root, (Node *) quals);
3306 :
3307 422 : return pathnode;
3308 : }
3309 :
3310 : /*
3311 : * create_windowagg_path
3312 : * Creates a pathnode that represents computation of window functions
3313 : *
3314 : * 'rel' is the parent relation associated with the result
3315 : * 'subpath' is the path representing the source of data
3316 : * 'target' is the PathTarget to be computed
3317 : * 'windowFuncs' is a list of WindowFunc structs
3318 : * 'runCondition' is a list of OpExprs to short-circuit WindowAgg execution
3319 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3320 : * 'qual' WindowClause.runconditions from lower-level WindowAggPaths.
3321 : * Must always be NIL when topwindow == false
3322 : * 'topwindow' pass as true only for the top-level WindowAgg. False for all
3323 : * intermediate WindowAggs.
3324 : *
3325 : * The input must be sorted according to the WindowClause's PARTITION keys
3326 : * plus ORDER BY keys.
3327 : */
3328 : WindowAggPath *
3329 2976 : create_windowagg_path(PlannerInfo *root,
3330 : RelOptInfo *rel,
3331 : Path *subpath,
3332 : PathTarget *target,
3333 : List *windowFuncs,
3334 : List *runCondition,
3335 : WindowClause *winclause,
3336 : List *qual,
3337 : bool topwindow)
3338 : {
3339 2976 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3340 :
3341 : /* qual can only be set for the topwindow */
3342 : Assert(qual == NIL || topwindow);
3343 :
3344 2976 : pathnode->path.pathtype = T_WindowAgg;
3345 2976 : pathnode->path.parent = rel;
3346 2976 : pathnode->path.pathtarget = target;
3347 : /* For now, assume we are above any joins, so no parameterization */
3348 2976 : pathnode->path.param_info = NULL;
3349 2976 : pathnode->path.parallel_aware = false;
3350 2976 : pathnode->path.parallel_safe = rel->consider_parallel &&
3351 0 : subpath->parallel_safe;
3352 2976 : pathnode->path.parallel_workers = subpath->parallel_workers;
3353 : /* WindowAgg preserves the input sort order */
3354 2976 : pathnode->path.pathkeys = subpath->pathkeys;
3355 :
3356 2976 : pathnode->subpath = subpath;
3357 2976 : pathnode->winclause = winclause;
3358 2976 : pathnode->qual = qual;
3359 2976 : pathnode->runCondition = runCondition;
3360 2976 : pathnode->topwindow = topwindow;
3361 :
3362 : /*
3363 : * For costing purposes, assume that there are no redundant partitioning
3364 : * or ordering columns; it's not worth the trouble to deal with that
3365 : * corner case here. So we just pass the unmodified list lengths to
3366 : * cost_windowagg.
3367 : */
3368 2976 : cost_windowagg(&pathnode->path, root,
3369 : windowFuncs,
3370 : winclause,
3371 : subpath->disabled_nodes,
3372 : subpath->startup_cost,
3373 : subpath->total_cost,
3374 : subpath->rows);
3375 :
3376 : /* add tlist eval cost for each output row */
3377 2976 : pathnode->path.startup_cost += target->cost.startup;
3378 2976 : pathnode->path.total_cost += target->cost.startup +
3379 2976 : target->cost.per_tuple * pathnode->path.rows;
3380 :
3381 2976 : return pathnode;
3382 : }
3383 :
3384 : /*
3385 : * create_setop_path
3386 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3387 : *
3388 : * 'rel' is the parent relation associated with the result
3389 : * 'leftpath' is the path representing the left-hand source of data
3390 : * 'rightpath' is the path representing the right-hand source of data
3391 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3392 : * 'strategy' is the implementation strategy (sorted or hashed)
3393 : * 'groupList' is a list of SortGroupClause's representing the grouping
3394 : * 'numGroups' is the estimated number of distinct groups in left-hand input
3395 : * 'outputRows' is the estimated number of output rows
3396 : *
3397 : * leftpath and rightpath must produce the same columns. Moreover, if
3398 : * strategy is SETOP_SORTED, leftpath and rightpath must both be sorted
3399 : * by all the grouping columns.
3400 : */
3401 : SetOpPath *
3402 1276 : create_setop_path(PlannerInfo *root,
3403 : RelOptInfo *rel,
3404 : Path *leftpath,
3405 : Path *rightpath,
3406 : SetOpCmd cmd,
3407 : SetOpStrategy strategy,
3408 : List *groupList,
3409 : double numGroups,
3410 : double outputRows)
3411 : {
3412 1276 : SetOpPath *pathnode = makeNode(SetOpPath);
3413 :
3414 1276 : pathnode->path.pathtype = T_SetOp;
3415 1276 : pathnode->path.parent = rel;
3416 1276 : pathnode->path.pathtarget = rel->reltarget;
3417 : /* For now, assume we are above any joins, so no parameterization */
3418 1276 : pathnode->path.param_info = NULL;
3419 1276 : pathnode->path.parallel_aware = false;
3420 2552 : pathnode->path.parallel_safe = rel->consider_parallel &&
3421 1276 : leftpath->parallel_safe && rightpath->parallel_safe;
3422 1276 : pathnode->path.parallel_workers =
3423 1276 : leftpath->parallel_workers + rightpath->parallel_workers;
3424 : /* SetOp preserves the input sort order if in sort mode */
3425 1276 : pathnode->path.pathkeys =
3426 1276 : (strategy == SETOP_SORTED) ? leftpath->pathkeys : NIL;
3427 :
3428 1276 : pathnode->leftpath = leftpath;
3429 1276 : pathnode->rightpath = rightpath;
3430 1276 : pathnode->cmd = cmd;
3431 1276 : pathnode->strategy = strategy;
3432 1276 : pathnode->groupList = groupList;
3433 1276 : pathnode->numGroups = numGroups;
3434 :
3435 : /*
3436 : * Compute cost estimates. As things stand, we end up with the same total
3437 : * cost in this node for sort and hash methods, but different startup
3438 : * costs. This could be refined perhaps, but it'll do for now.
3439 : */
3440 1276 : pathnode->path.disabled_nodes =
3441 1276 : leftpath->disabled_nodes + rightpath->disabled_nodes;
3442 1276 : if (strategy == SETOP_SORTED)
3443 : {
3444 : /*
3445 : * In sorted mode, we can emit output incrementally. Charge one
3446 : * cpu_operator_cost per comparison per input tuple. Like cost_group,
3447 : * we assume all columns get compared at most of the tuples.
3448 : */
3449 668 : pathnode->path.startup_cost =
3450 668 : leftpath->startup_cost + rightpath->startup_cost;
3451 668 : pathnode->path.total_cost =
3452 1336 : leftpath->total_cost + rightpath->total_cost +
3453 668 : cpu_operator_cost * (leftpath->rows + rightpath->rows) * list_length(groupList);
3454 :
3455 : /*
3456 : * Also charge a small amount per extracted tuple. Like cost_sort,
3457 : * charge only operator cost not cpu_tuple_cost, since SetOp does no
3458 : * qual-checking or projection.
3459 : */
3460 668 : pathnode->path.total_cost += cpu_operator_cost * outputRows;
3461 : }
3462 : else
3463 : {
3464 : Size hashtablesize;
3465 :
3466 : /*
3467 : * In hashed mode, we must read all the input before we can emit
3468 : * anything. Also charge comparison costs to represent the cost of
3469 : * hash table lookups.
3470 : */
3471 608 : pathnode->path.startup_cost =
3472 1216 : leftpath->total_cost + rightpath->total_cost +
3473 608 : cpu_operator_cost * (leftpath->rows + rightpath->rows) * list_length(groupList);
3474 608 : pathnode->path.total_cost = pathnode->path.startup_cost;
3475 :
3476 : /*
3477 : * Also charge a small amount per extracted tuple. Like cost_sort,
3478 : * charge only operator cost not cpu_tuple_cost, since SetOp does no
3479 : * qual-checking or projection.
3480 : */
3481 608 : pathnode->path.total_cost += cpu_operator_cost * outputRows;
3482 :
3483 : /*
3484 : * Mark the path as disabled if enable_hashagg is off. While this
3485 : * isn't exactly a HashAgg node, it seems close enough to justify
3486 : * letting that switch control it.
3487 : */
3488 608 : if (!enable_hashagg)
3489 114 : pathnode->path.disabled_nodes++;
3490 :
3491 : /*
3492 : * Also disable if it doesn't look like the hashtable will fit into
3493 : * hash_mem. (Note: reject on equality, to ensure that an estimate of
3494 : * SIZE_MAX disables hashing regardless of the hash_mem limit.)
3495 : */
3496 608 : hashtablesize = EstimateSetOpHashTableSpace(numGroups,
3497 608 : leftpath->pathtarget->width);
3498 608 : if (hashtablesize >= get_hash_memory_limit())
3499 0 : pathnode->path.disabled_nodes++;
3500 : }
3501 1276 : pathnode->path.rows = outputRows;
3502 :
3503 1276 : return pathnode;
3504 : }
3505 :
3506 : /*
3507 : * create_recursiveunion_path
3508 : * Creates a pathnode that represents a recursive UNION node
3509 : *
3510 : * 'rel' is the parent relation associated with the result
3511 : * 'leftpath' is the source of data for the non-recursive term
3512 : * 'rightpath' is the source of data for the recursive term
3513 : * 'target' is the PathTarget to be computed
3514 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3515 : * 'wtParam' is the ID of Param representing work table
3516 : * 'numGroups' is the estimated number of groups
3517 : *
3518 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3519 : */
3520 : RecursiveUnionPath *
3521 932 : create_recursiveunion_path(PlannerInfo *root,
3522 : RelOptInfo *rel,
3523 : Path *leftpath,
3524 : Path *rightpath,
3525 : PathTarget *target,
3526 : List *distinctList,
3527 : int wtParam,
3528 : double numGroups)
3529 : {
3530 932 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3531 :
3532 932 : pathnode->path.pathtype = T_RecursiveUnion;
3533 932 : pathnode->path.parent = rel;
3534 932 : pathnode->path.pathtarget = target;
3535 : /* For now, assume we are above any joins, so no parameterization */
3536 932 : pathnode->path.param_info = NULL;
3537 932 : pathnode->path.parallel_aware = false;
3538 1864 : pathnode->path.parallel_safe = rel->consider_parallel &&
3539 932 : leftpath->parallel_safe && rightpath->parallel_safe;
3540 : /* Foolish, but we'll do it like joins for now: */
3541 932 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3542 : /* RecursiveUnion result is always unsorted */
3543 932 : pathnode->path.pathkeys = NIL;
3544 :
3545 932 : pathnode->leftpath = leftpath;
3546 932 : pathnode->rightpath = rightpath;
3547 932 : pathnode->distinctList = distinctList;
3548 932 : pathnode->wtParam = wtParam;
3549 932 : pathnode->numGroups = numGroups;
3550 :
3551 932 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3552 :
3553 932 : return pathnode;
3554 : }
3555 :
3556 : /*
3557 : * create_lockrows_path
3558 : * Creates a pathnode that represents acquiring row locks
3559 : *
3560 : * 'rel' is the parent relation associated with the result
3561 : * 'subpath' is the path representing the source of data
3562 : * 'rowMarks' is a list of PlanRowMark's
3563 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3564 : */
3565 : LockRowsPath *
3566 14038 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3567 : Path *subpath, List *rowMarks, int epqParam)
3568 : {
3569 14038 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3570 :
3571 14038 : pathnode->path.pathtype = T_LockRows;
3572 14038 : pathnode->path.parent = rel;
3573 : /* LockRows doesn't project, so use source path's pathtarget */
3574 14038 : pathnode->path.pathtarget = subpath->pathtarget;
3575 : /* For now, assume we are above any joins, so no parameterization */
3576 14038 : pathnode->path.param_info = NULL;
3577 14038 : pathnode->path.parallel_aware = false;
3578 14038 : pathnode->path.parallel_safe = false;
3579 14038 : pathnode->path.parallel_workers = 0;
3580 14038 : pathnode->path.rows = subpath->rows;
3581 :
3582 : /*
3583 : * The result cannot be assumed sorted, since locking might cause the sort
3584 : * key columns to be replaced with new values.
3585 : */
3586 14038 : pathnode->path.pathkeys = NIL;
3587 :
3588 14038 : pathnode->subpath = subpath;
3589 14038 : pathnode->rowMarks = rowMarks;
3590 14038 : pathnode->epqParam = epqParam;
3591 :
3592 : /*
3593 : * We should charge something extra for the costs of row locking and
3594 : * possible refetches, but it's hard to say how much. For now, use
3595 : * cpu_tuple_cost per row.
3596 : */
3597 14038 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3598 14038 : pathnode->path.startup_cost = subpath->startup_cost;
3599 14038 : pathnode->path.total_cost = subpath->total_cost +
3600 14038 : cpu_tuple_cost * subpath->rows;
3601 :
3602 14038 : return pathnode;
3603 : }
3604 :
3605 : /*
3606 : * create_modifytable_path
3607 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE/MERGE
3608 : * mods
3609 : *
3610 : * 'rel' is the parent relation associated with the result
3611 : * 'subpath' is a Path producing source data
3612 : * 'operation' is the operation type
3613 : * 'canSetTag' is true if we set the command tag/es_processed
3614 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3615 : * 'rootRelation' is the partitioned/inherited table root RTI, or 0 if none
3616 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3617 : * 'updateColnosLists' is a list of UPDATE target column number lists
3618 : * (one sublist per rel); or NIL if not an UPDATE
3619 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3620 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3621 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3622 : * 'onconflict' is the ON CONFLICT clause, or NULL
3623 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3624 : * 'mergeActionLists' is a list of lists of MERGE actions (one per rel)
3625 : * 'mergeJoinConditions' is a list of join conditions for MERGE (one per rel)
3626 : */
3627 : ModifyTablePath *
3628 89400 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3629 : Path *subpath,
3630 : CmdType operation, bool canSetTag,
3631 : Index nominalRelation, Index rootRelation,
3632 : List *resultRelations,
3633 : List *updateColnosLists,
3634 : List *withCheckOptionLists, List *returningLists,
3635 : List *rowMarks, OnConflictExpr *onconflict,
3636 : List *mergeActionLists, List *mergeJoinConditions,
3637 : int epqParam)
3638 : {
3639 89400 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3640 :
3641 : Assert(operation == CMD_MERGE ||
3642 : (operation == CMD_UPDATE ?
3643 : list_length(resultRelations) == list_length(updateColnosLists) :
3644 : updateColnosLists == NIL));
3645 : Assert(withCheckOptionLists == NIL ||
3646 : list_length(resultRelations) == list_length(withCheckOptionLists));
3647 : Assert(returningLists == NIL ||
3648 : list_length(resultRelations) == list_length(returningLists));
3649 :
3650 89400 : pathnode->path.pathtype = T_ModifyTable;
3651 89400 : pathnode->path.parent = rel;
3652 : /* pathtarget is not interesting, just make it minimally valid */
3653 89400 : pathnode->path.pathtarget = rel->reltarget;
3654 : /* For now, assume we are above any joins, so no parameterization */
3655 89400 : pathnode->path.param_info = NULL;
3656 89400 : pathnode->path.parallel_aware = false;
3657 89400 : pathnode->path.parallel_safe = false;
3658 89400 : pathnode->path.parallel_workers = 0;
3659 89400 : pathnode->path.pathkeys = NIL;
3660 :
3661 : /*
3662 : * Compute cost & rowcount as subpath cost & rowcount (if RETURNING)
3663 : *
3664 : * Currently, we don't charge anything extra for the actual table
3665 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3666 : * expressions if any. It would only be window dressing, since
3667 : * ModifyTable is always a top-level node and there is no way for the
3668 : * costs to change any higher-level planning choices. But we might want
3669 : * to make it look better sometime.
3670 : */
3671 89400 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3672 89400 : pathnode->path.startup_cost = subpath->startup_cost;
3673 89400 : pathnode->path.total_cost = subpath->total_cost;
3674 89400 : if (returningLists != NIL)
3675 : {
3676 3048 : pathnode->path.rows = subpath->rows;
3677 :
3678 : /*
3679 : * Set width to match the subpath output. XXX this is totally wrong:
3680 : * we should return an average of the RETURNING tlist widths. But
3681 : * it's what happened historically, and improving it is a task for
3682 : * another day. (Again, it's mostly window dressing.)
3683 : */
3684 3048 : pathnode->path.pathtarget->width = subpath->pathtarget->width;
3685 : }
3686 : else
3687 : {
3688 86352 : pathnode->path.rows = 0;
3689 86352 : pathnode->path.pathtarget->width = 0;
3690 : }
3691 :
3692 89400 : pathnode->subpath = subpath;
3693 89400 : pathnode->operation = operation;
3694 89400 : pathnode->canSetTag = canSetTag;
3695 89400 : pathnode->nominalRelation = nominalRelation;
3696 89400 : pathnode->rootRelation = rootRelation;
3697 89400 : pathnode->resultRelations = resultRelations;
3698 89400 : pathnode->updateColnosLists = updateColnosLists;
3699 89400 : pathnode->withCheckOptionLists = withCheckOptionLists;
3700 89400 : pathnode->returningLists = returningLists;
3701 89400 : pathnode->rowMarks = rowMarks;
3702 89400 : pathnode->onconflict = onconflict;
3703 89400 : pathnode->epqParam = epqParam;
3704 89400 : pathnode->mergeActionLists = mergeActionLists;
3705 89400 : pathnode->mergeJoinConditions = mergeJoinConditions;
3706 :
3707 89400 : return pathnode;
3708 : }
3709 :
3710 : /*
3711 : * create_limit_path
3712 : * Creates a pathnode that represents performing LIMIT/OFFSET
3713 : *
3714 : * In addition to providing the actual OFFSET and LIMIT expressions,
3715 : * the caller must provide estimates of their values for costing purposes.
3716 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3717 : * the clause not being present, and -1 means it's present but we could
3718 : * not estimate its value.
3719 : *
3720 : * 'rel' is the parent relation associated with the result
3721 : * 'subpath' is the path representing the source of data
3722 : * 'limitOffset' is the actual OFFSET expression, or NULL
3723 : * 'limitCount' is the actual LIMIT expression, or NULL
3724 : * 'offset_est' is the estimated value of the OFFSET expression
3725 : * 'count_est' is the estimated value of the LIMIT expression
3726 : */
3727 : LimitPath *
3728 6306 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3729 : Path *subpath,
3730 : Node *limitOffset, Node *limitCount,
3731 : LimitOption limitOption,
3732 : int64 offset_est, int64 count_est)
3733 : {
3734 6306 : LimitPath *pathnode = makeNode(LimitPath);
3735 :
3736 6306 : pathnode->path.pathtype = T_Limit;
3737 6306 : pathnode->path.parent = rel;
3738 : /* Limit doesn't project, so use source path's pathtarget */
3739 6306 : pathnode->path.pathtarget = subpath->pathtarget;
3740 : /* For now, assume we are above any joins, so no parameterization */
3741 6306 : pathnode->path.param_info = NULL;
3742 6306 : pathnode->path.parallel_aware = false;
3743 8770 : pathnode->path.parallel_safe = rel->consider_parallel &&
3744 2464 : subpath->parallel_safe;
3745 6306 : pathnode->path.parallel_workers = subpath->parallel_workers;
3746 6306 : pathnode->path.rows = subpath->rows;
3747 6306 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3748 6306 : pathnode->path.startup_cost = subpath->startup_cost;
3749 6306 : pathnode->path.total_cost = subpath->total_cost;
3750 6306 : pathnode->path.pathkeys = subpath->pathkeys;
3751 6306 : pathnode->subpath = subpath;
3752 6306 : pathnode->limitOffset = limitOffset;
3753 6306 : pathnode->limitCount = limitCount;
3754 6306 : pathnode->limitOption = limitOption;
3755 :
3756 : /*
3757 : * Adjust the output rows count and costs according to the offset/limit.
3758 : */
3759 6306 : adjust_limit_rows_costs(&pathnode->path.rows,
3760 : &pathnode->path.startup_cost,
3761 : &pathnode->path.total_cost,
3762 : offset_est, count_est);
3763 :
3764 6306 : return pathnode;
3765 : }
3766 :
3767 : /*
3768 : * adjust_limit_rows_costs
3769 : * Adjust the size and cost estimates for a LimitPath node according to the
3770 : * offset/limit.
3771 : *
3772 : * This is only a cosmetic issue if we are at top level, but if we are
3773 : * building a subquery then it's important to report correct info to the outer
3774 : * planner.
3775 : *
3776 : * When the offset or count couldn't be estimated, use 10% of the estimated
3777 : * number of rows emitted from the subpath.
3778 : *
3779 : * XXX we don't bother to add eval costs of the offset/limit expressions
3780 : * themselves to the path costs. In theory we should, but in most cases those
3781 : * expressions are trivial and it's just not worth the trouble.
3782 : */
3783 : void
3784 6490 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3785 : Cost *startup_cost, /* in/out parameter */
3786 : Cost *total_cost, /* in/out parameter */
3787 : int64 offset_est,
3788 : int64 count_est)
3789 : {
3790 6490 : double input_rows = *rows;
3791 6490 : Cost input_startup_cost = *startup_cost;
3792 6490 : Cost input_total_cost = *total_cost;
3793 :
3794 6490 : if (offset_est != 0)
3795 : {
3796 : double offset_rows;
3797 :
3798 712 : if (offset_est > 0)
3799 688 : offset_rows = (double) offset_est;
3800 : else
3801 24 : offset_rows = clamp_row_est(input_rows * 0.10);
3802 712 : if (offset_rows > *rows)
3803 38 : offset_rows = *rows;
3804 712 : if (input_rows > 0)
3805 712 : *startup_cost +=
3806 712 : (input_total_cost - input_startup_cost)
3807 712 : * offset_rows / input_rows;
3808 712 : *rows -= offset_rows;
3809 712 : if (*rows < 1)
3810 46 : *rows = 1;
3811 : }
3812 :
3813 6490 : if (count_est != 0)
3814 : {
3815 : double count_rows;
3816 :
3817 6414 : if (count_est > 0)
3818 6408 : count_rows = (double) count_est;
3819 : else
3820 6 : count_rows = clamp_row_est(input_rows * 0.10);
3821 6414 : if (count_rows > *rows)
3822 248 : count_rows = *rows;
3823 6414 : if (input_rows > 0)
3824 6414 : *total_cost = *startup_cost +
3825 6414 : (input_total_cost - input_startup_cost)
3826 6414 : * count_rows / input_rows;
3827 6414 : *rows = count_rows;
3828 6414 : if (*rows < 1)
3829 0 : *rows = 1;
3830 : }
3831 6490 : }
3832 :
3833 :
3834 : /*
3835 : * reparameterize_path
3836 : * Attempt to modify a Path to have greater parameterization
3837 : *
3838 : * We use this to attempt to bring all child paths of an appendrel to the
3839 : * same parameterization level, ensuring that they all enforce the same set
3840 : * of join quals (and thus that that parameterization can be attributed to
3841 : * an append path built from such paths). Currently, only a few path types
3842 : * are supported here, though more could be added at need. We return NULL
3843 : * if we can't reparameterize the given path.
3844 : *
3845 : * Note: we intentionally do not pass created paths to add_path(); it would
3846 : * possibly try to delete them on the grounds of being cost-inferior to the
3847 : * paths they were made from, and we don't want that. Paths made here are
3848 : * not necessarily of general-purpose usefulness, but they can be useful
3849 : * as members of an append path.
3850 : */
3851 : Path *
3852 356 : reparameterize_path(PlannerInfo *root, Path *path,
3853 : Relids required_outer,
3854 : double loop_count)
3855 : {
3856 356 : RelOptInfo *rel = path->parent;
3857 :
3858 : /* Can only increase, not decrease, path's parameterization */
3859 356 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
3860 0 : return NULL;
3861 356 : switch (path->pathtype)
3862 : {
3863 264 : case T_SeqScan:
3864 264 : return create_seqscan_path(root, rel, required_outer, 0);
3865 0 : case T_SampleScan:
3866 0 : return create_samplescan_path(root, rel, required_outer);
3867 0 : case T_IndexScan:
3868 : case T_IndexOnlyScan:
3869 : {
3870 0 : IndexPath *ipath = (IndexPath *) path;
3871 0 : IndexPath *newpath = makeNode(IndexPath);
3872 :
3873 : /*
3874 : * We can't use create_index_path directly, and would not want
3875 : * to because it would re-compute the indexqual conditions
3876 : * which is wasted effort. Instead we hack things a bit:
3877 : * flat-copy the path node, revise its param_info, and redo
3878 : * the cost estimate.
3879 : */
3880 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3881 0 : newpath->path.param_info =
3882 0 : get_baserel_parampathinfo(root, rel, required_outer);
3883 0 : cost_index(newpath, root, loop_count, false);
3884 0 : return (Path *) newpath;
3885 : }
3886 0 : case T_BitmapHeapScan:
3887 : {
3888 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3889 :
3890 0 : return (Path *) create_bitmap_heap_path(root,
3891 : rel,
3892 : bpath->bitmapqual,
3893 : required_outer,
3894 : loop_count, 0);
3895 : }
3896 0 : case T_SubqueryScan:
3897 : {
3898 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
3899 0 : Path *subpath = spath->subpath;
3900 : bool trivial_pathtarget;
3901 :
3902 : /*
3903 : * If existing node has zero extra cost, we must have decided
3904 : * its target is trivial. (The converse is not true, because
3905 : * it might have a trivial target but quals to enforce; but in
3906 : * that case the new node will too, so it doesn't matter
3907 : * whether we get the right answer here.)
3908 : */
3909 0 : trivial_pathtarget =
3910 0 : (subpath->total_cost == spath->path.total_cost);
3911 :
3912 0 : return (Path *) create_subqueryscan_path(root,
3913 : rel,
3914 : subpath,
3915 : trivial_pathtarget,
3916 : spath->path.pathkeys,
3917 : required_outer);
3918 : }
3919 60 : case T_Result:
3920 : /* Supported only for RTE_RESULT scan paths */
3921 60 : if (IsA(path, Path))
3922 60 : return create_resultscan_path(root, rel, required_outer);
3923 0 : break;
3924 0 : case T_Append:
3925 : {
3926 0 : AppendPath *apath = (AppendPath *) path;
3927 0 : List *childpaths = NIL;
3928 0 : List *partialpaths = NIL;
3929 : int i;
3930 : ListCell *lc;
3931 :
3932 : /* Reparameterize the children */
3933 0 : i = 0;
3934 0 : foreach(lc, apath->subpaths)
3935 : {
3936 0 : Path *spath = (Path *) lfirst(lc);
3937 :
3938 0 : spath = reparameterize_path(root, spath,
3939 : required_outer,
3940 : loop_count);
3941 0 : if (spath == NULL)
3942 0 : return NULL;
3943 : /* We have to re-split the regular and partial paths */
3944 0 : if (i < apath->first_partial_path)
3945 0 : childpaths = lappend(childpaths, spath);
3946 : else
3947 0 : partialpaths = lappend(partialpaths, spath);
3948 0 : i++;
3949 : }
3950 0 : return (Path *)
3951 0 : create_append_path(root, rel, childpaths, partialpaths,
3952 : apath->path.pathkeys, required_outer,
3953 : apath->path.parallel_workers,
3954 0 : apath->path.parallel_aware,
3955 : -1);
3956 : }
3957 0 : case T_Material:
3958 : {
3959 0 : MaterialPath *mpath = (MaterialPath *) path;
3960 0 : Path *spath = mpath->subpath;
3961 :
3962 0 : spath = reparameterize_path(root, spath,
3963 : required_outer,
3964 : loop_count);
3965 0 : if (spath == NULL)
3966 0 : return NULL;
3967 0 : return (Path *) create_material_path(rel, spath);
3968 : }
3969 0 : case T_Memoize:
3970 : {
3971 0 : MemoizePath *mpath = (MemoizePath *) path;
3972 0 : Path *spath = mpath->subpath;
3973 :
3974 0 : spath = reparameterize_path(root, spath,
3975 : required_outer,
3976 : loop_count);
3977 0 : if (spath == NULL)
3978 0 : return NULL;
3979 0 : return (Path *) create_memoize_path(root, rel,
3980 : spath,
3981 : mpath->param_exprs,
3982 : mpath->hash_operators,
3983 0 : mpath->singlerow,
3984 0 : mpath->binary_mode,
3985 : mpath->est_calls);
3986 : }
3987 32 : default:
3988 32 : break;
3989 : }
3990 32 : return NULL;
3991 : }
3992 :
3993 : /*
3994 : * reparameterize_path_by_child
3995 : * Given a path parameterized by the parent of the given child relation,
3996 : * translate the path to be parameterized by the given child relation.
3997 : *
3998 : * Most fields in the path are not changed, but any expressions must be
3999 : * adjusted to refer to the correct varnos, and any subpaths must be
4000 : * recursively reparameterized. Other fields that refer to specific relids
4001 : * also need adjustment.
4002 : *
4003 : * The cost, number of rows, width and parallel path properties depend upon
4004 : * path->parent, which does not change during the translation. So we need
4005 : * not change those.
4006 : *
4007 : * Currently, only a few path types are supported here, though more could be
4008 : * added at need. We return NULL if we can't reparameterize the given path.
4009 : *
4010 : * Note that this function can change referenced RangeTblEntries, RelOptInfos
4011 : * and IndexOptInfos as well as the Path structures. Therefore, it's only safe
4012 : * to call during create_plan(), when we have made a final choice of which Path
4013 : * to use for each RangeTblEntry/RelOptInfo/IndexOptInfo.
4014 : *
4015 : * Keep this code in sync with path_is_reparameterizable_by_child()!
4016 : */
4017 : Path *
4018 100038 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
4019 : RelOptInfo *child_rel)
4020 : {
4021 : Path *new_path;
4022 : ParamPathInfo *new_ppi;
4023 : ParamPathInfo *old_ppi;
4024 : Relids required_outer;
4025 :
4026 : #define ADJUST_CHILD_ATTRS(node) \
4027 : ((node) = (void *) adjust_appendrel_attrs_multilevel(root, \
4028 : (Node *) (node), \
4029 : child_rel, \
4030 : child_rel->top_parent))
4031 :
4032 : #define REPARAMETERIZE_CHILD_PATH(path) \
4033 : do { \
4034 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
4035 : if ((path) == NULL) \
4036 : return NULL; \
4037 : } while(0)
4038 :
4039 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
4040 : do { \
4041 : if ((pathlist) != NIL) \
4042 : { \
4043 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
4044 : child_rel); \
4045 : if ((pathlist) == NIL) \
4046 : return NULL; \
4047 : } \
4048 : } while(0)
4049 :
4050 : /*
4051 : * If the path is not parameterized by the parent of the given relation,
4052 : * it doesn't need reparameterization.
4053 : */
4054 100038 : if (!path->param_info ||
4055 50166 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4056 99036 : return path;
4057 :
4058 : /*
4059 : * If possible, reparameterize the given path.
4060 : *
4061 : * This function is currently only applied to the inner side of a nestloop
4062 : * join that is being partitioned by the partitionwise-join code. Hence,
4063 : * we need only support path types that plausibly arise in that context.
4064 : * (In particular, supporting sorted path types would be a waste of code
4065 : * and cycles: even if we translated them here, they'd just lose in
4066 : * subsequent cost comparisons.) If we do see an unsupported path type,
4067 : * that just means we won't be able to generate a partitionwise-join plan
4068 : * using that path type.
4069 : */
4070 1002 : switch (nodeTag(path))
4071 : {
4072 228 : case T_Path:
4073 228 : new_path = path;
4074 228 : ADJUST_CHILD_ATTRS(new_path->parent->baserestrictinfo);
4075 228 : if (path->pathtype == T_SampleScan)
4076 : {
4077 48 : Index scan_relid = path->parent->relid;
4078 : RangeTblEntry *rte;
4079 :
4080 : /* it should be a base rel with a tablesample clause... */
4081 : Assert(scan_relid > 0);
4082 48 : rte = planner_rt_fetch(scan_relid, root);
4083 : Assert(rte->rtekind == RTE_RELATION);
4084 : Assert(rte->tablesample != NULL);
4085 :
4086 48 : ADJUST_CHILD_ATTRS(rte->tablesample);
4087 : }
4088 228 : break;
4089 :
4090 522 : case T_IndexPath:
4091 : {
4092 522 : IndexPath *ipath = (IndexPath *) path;
4093 :
4094 522 : ADJUST_CHILD_ATTRS(ipath->indexinfo->indrestrictinfo);
4095 522 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
4096 522 : new_path = (Path *) ipath;
4097 : }
4098 522 : break;
4099 :
4100 48 : case T_BitmapHeapPath:
4101 : {
4102 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4103 :
4104 48 : ADJUST_CHILD_ATTRS(bhpath->path.parent->baserestrictinfo);
4105 48 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
4106 48 : new_path = (Path *) bhpath;
4107 : }
4108 48 : break;
4109 :
4110 24 : case T_BitmapAndPath:
4111 : {
4112 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4113 :
4114 24 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
4115 24 : new_path = (Path *) bapath;
4116 : }
4117 24 : break;
4118 :
4119 24 : case T_BitmapOrPath:
4120 : {
4121 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4122 :
4123 24 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
4124 24 : new_path = (Path *) bopath;
4125 : }
4126 24 : break;
4127 :
4128 0 : case T_ForeignPath:
4129 : {
4130 0 : ForeignPath *fpath = (ForeignPath *) path;
4131 : ReparameterizeForeignPathByChild_function rfpc_func;
4132 :
4133 0 : ADJUST_CHILD_ATTRS(fpath->path.parent->baserestrictinfo);
4134 0 : if (fpath->fdw_outerpath)
4135 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
4136 0 : if (fpath->fdw_restrictinfo)
4137 0 : ADJUST_CHILD_ATTRS(fpath->fdw_restrictinfo);
4138 :
4139 : /* Hand over to FDW if needed. */
4140 0 : rfpc_func =
4141 0 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
4142 0 : if (rfpc_func)
4143 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
4144 : child_rel);
4145 0 : new_path = (Path *) fpath;
4146 : }
4147 0 : break;
4148 :
4149 0 : case T_CustomPath:
4150 : {
4151 0 : CustomPath *cpath = (CustomPath *) path;
4152 :
4153 0 : ADJUST_CHILD_ATTRS(cpath->path.parent->baserestrictinfo);
4154 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
4155 0 : if (cpath->custom_restrictinfo)
4156 0 : ADJUST_CHILD_ATTRS(cpath->custom_restrictinfo);
4157 0 : if (cpath->methods &&
4158 0 : cpath->methods->ReparameterizeCustomPathByChild)
4159 0 : cpath->custom_private =
4160 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4161 : cpath->custom_private,
4162 : child_rel);
4163 0 : new_path = (Path *) cpath;
4164 : }
4165 0 : break;
4166 :
4167 36 : case T_NestPath:
4168 : {
4169 36 : NestPath *npath = (NestPath *) path;
4170 36 : JoinPath *jpath = (JoinPath *) npath;
4171 :
4172 36 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4173 36 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4174 36 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4175 36 : new_path = (Path *) npath;
4176 : }
4177 36 : break;
4178 :
4179 0 : case T_MergePath:
4180 : {
4181 0 : MergePath *mpath = (MergePath *) path;
4182 0 : JoinPath *jpath = (JoinPath *) mpath;
4183 :
4184 0 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4185 0 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4186 0 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4187 0 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4188 0 : new_path = (Path *) mpath;
4189 : }
4190 0 : break;
4191 :
4192 48 : case T_HashPath:
4193 : {
4194 48 : HashPath *hpath = (HashPath *) path;
4195 48 : JoinPath *jpath = (JoinPath *) hpath;
4196 :
4197 48 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4198 48 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4199 48 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4200 48 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4201 48 : new_path = (Path *) hpath;
4202 : }
4203 48 : break;
4204 :
4205 24 : case T_AppendPath:
4206 : {
4207 24 : AppendPath *apath = (AppendPath *) path;
4208 :
4209 24 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4210 24 : new_path = (Path *) apath;
4211 : }
4212 24 : break;
4213 :
4214 0 : case T_MaterialPath:
4215 : {
4216 0 : MaterialPath *mpath = (MaterialPath *) path;
4217 :
4218 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4219 0 : new_path = (Path *) mpath;
4220 : }
4221 0 : break;
4222 :
4223 48 : case T_MemoizePath:
4224 : {
4225 48 : MemoizePath *mpath = (MemoizePath *) path;
4226 :
4227 48 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4228 48 : ADJUST_CHILD_ATTRS(mpath->param_exprs);
4229 48 : new_path = (Path *) mpath;
4230 : }
4231 48 : break;
4232 :
4233 0 : case T_GatherPath:
4234 : {
4235 0 : GatherPath *gpath = (GatherPath *) path;
4236 :
4237 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4238 0 : new_path = (Path *) gpath;
4239 : }
4240 0 : break;
4241 :
4242 0 : default:
4243 : /* We don't know how to reparameterize this path. */
4244 0 : return NULL;
4245 : }
4246 :
4247 : /*
4248 : * Adjust the parameterization information, which refers to the topmost
4249 : * parent. The topmost parent can be multiple levels away from the given
4250 : * child, hence use multi-level expression adjustment routines.
4251 : */
4252 1002 : old_ppi = new_path->param_info;
4253 : required_outer =
4254 1002 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4255 : child_rel,
4256 1002 : child_rel->top_parent);
4257 :
4258 : /* If we already have a PPI for this parameterization, just return it */
4259 1002 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4260 :
4261 : /*
4262 : * If not, build a new one and link it to the list of PPIs. For the same
4263 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4264 : * context the given RelOptInfo is in.
4265 : */
4266 1002 : if (new_ppi == NULL)
4267 : {
4268 : MemoryContext oldcontext;
4269 858 : RelOptInfo *rel = path->parent;
4270 :
4271 858 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4272 :
4273 858 : new_ppi = makeNode(ParamPathInfo);
4274 858 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4275 858 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4276 858 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4277 858 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4278 858 : new_ppi->ppi_serials = bms_copy(old_ppi->ppi_serials);
4279 858 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4280 :
4281 858 : MemoryContextSwitchTo(oldcontext);
4282 : }
4283 1002 : bms_free(required_outer);
4284 :
4285 1002 : new_path->param_info = new_ppi;
4286 :
4287 : /*
4288 : * Adjust the path target if the parent of the outer relation is
4289 : * referenced in the targetlist. This can happen when only the parent of
4290 : * outer relation is laterally referenced in this relation.
4291 : */
4292 1002 : if (bms_overlap(path->parent->lateral_relids,
4293 1002 : child_rel->top_parent_relids))
4294 : {
4295 480 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4296 480 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4297 : }
4298 :
4299 1002 : return new_path;
4300 : }
4301 :
4302 : /*
4303 : * path_is_reparameterizable_by_child
4304 : * Given a path parameterized by the parent of the given child relation,
4305 : * see if it can be translated to be parameterized by the child relation.
4306 : *
4307 : * This must return true if and only if reparameterize_path_by_child()
4308 : * would succeed on this path. Currently it's sufficient to verify that
4309 : * the path and all of its subpaths (if any) are of the types handled by
4310 : * that function. However, subpaths that are not parameterized can be
4311 : * disregarded since they won't require translation.
4312 : */
4313 : bool
4314 37224 : path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
4315 : {
4316 : #define REJECT_IF_PATH_NOT_REPARAMETERIZABLE(path) \
4317 : do { \
4318 : if (!path_is_reparameterizable_by_child(path, child_rel)) \
4319 : return false; \
4320 : } while(0)
4321 :
4322 : #define REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(pathlist) \
4323 : do { \
4324 : if (!pathlist_is_reparameterizable_by_child(pathlist, child_rel)) \
4325 : return false; \
4326 : } while(0)
4327 :
4328 : /*
4329 : * If the path is not parameterized by the parent of the given relation,
4330 : * it doesn't need reparameterization.
4331 : */
4332 37224 : if (!path->param_info ||
4333 36816 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4334 984 : return true;
4335 :
4336 : /*
4337 : * Check that the path type is one that reparameterize_path_by_child() can
4338 : * handle, and recursively check subpaths.
4339 : */
4340 36240 : switch (nodeTag(path))
4341 : {
4342 24312 : case T_Path:
4343 : case T_IndexPath:
4344 24312 : break;
4345 :
4346 48 : case T_BitmapHeapPath:
4347 : {
4348 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4349 :
4350 48 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(bhpath->bitmapqual);
4351 : }
4352 48 : break;
4353 :
4354 24 : case T_BitmapAndPath:
4355 : {
4356 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4357 :
4358 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bapath->bitmapquals);
4359 : }
4360 24 : break;
4361 :
4362 24 : case T_BitmapOrPath:
4363 : {
4364 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4365 :
4366 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bopath->bitmapquals);
4367 : }
4368 24 : break;
4369 :
4370 148 : case T_ForeignPath:
4371 : {
4372 148 : ForeignPath *fpath = (ForeignPath *) path;
4373 :
4374 148 : if (fpath->fdw_outerpath)
4375 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(fpath->fdw_outerpath);
4376 : }
4377 148 : break;
4378 :
4379 0 : case T_CustomPath:
4380 : {
4381 0 : CustomPath *cpath = (CustomPath *) path;
4382 :
4383 0 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(cpath->custom_paths);
4384 : }
4385 0 : break;
4386 :
4387 1248 : case T_NestPath:
4388 : case T_MergePath:
4389 : case T_HashPath:
4390 : {
4391 1248 : JoinPath *jpath = (JoinPath *) path;
4392 :
4393 1248 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->outerjoinpath);
4394 1248 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->innerjoinpath);
4395 : }
4396 1248 : break;
4397 :
4398 192 : case T_AppendPath:
4399 : {
4400 192 : AppendPath *apath = (AppendPath *) path;
4401 :
4402 192 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(apath->subpaths);
4403 : }
4404 192 : break;
4405 :
4406 0 : case T_MaterialPath:
4407 : {
4408 0 : MaterialPath *mpath = (MaterialPath *) path;
4409 :
4410 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4411 : }
4412 0 : break;
4413 :
4414 10244 : case T_MemoizePath:
4415 : {
4416 10244 : MemoizePath *mpath = (MemoizePath *) path;
4417 :
4418 10244 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4419 : }
4420 10244 : break;
4421 :
4422 0 : case T_GatherPath:
4423 : {
4424 0 : GatherPath *gpath = (GatherPath *) path;
4425 :
4426 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(gpath->subpath);
4427 : }
4428 0 : break;
4429 :
4430 0 : default:
4431 : /* We don't know how to reparameterize this path. */
4432 0 : return false;
4433 : }
4434 :
4435 36240 : return true;
4436 : }
4437 :
4438 : /*
4439 : * reparameterize_pathlist_by_child
4440 : * Helper function to reparameterize a list of paths by given child rel.
4441 : *
4442 : * Returns NIL to indicate failure, so pathlist had better not be NIL.
4443 : */
4444 : static List *
4445 72 : reparameterize_pathlist_by_child(PlannerInfo *root,
4446 : List *pathlist,
4447 : RelOptInfo *child_rel)
4448 : {
4449 : ListCell *lc;
4450 72 : List *result = NIL;
4451 :
4452 216 : foreach(lc, pathlist)
4453 : {
4454 144 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4455 : child_rel);
4456 :
4457 144 : if (path == NULL)
4458 : {
4459 0 : list_free(result);
4460 0 : return NIL;
4461 : }
4462 :
4463 144 : result = lappend(result, path);
4464 : }
4465 :
4466 72 : return result;
4467 : }
4468 :
4469 : /*
4470 : * pathlist_is_reparameterizable_by_child
4471 : * Helper function to check if a list of paths can be reparameterized.
4472 : */
4473 : static bool
4474 240 : pathlist_is_reparameterizable_by_child(List *pathlist, RelOptInfo *child_rel)
4475 : {
4476 : ListCell *lc;
4477 :
4478 720 : foreach(lc, pathlist)
4479 : {
4480 480 : Path *path = (Path *) lfirst(lc);
4481 :
4482 480 : if (!path_is_reparameterizable_by_child(path, child_rel))
4483 0 : return false;
4484 : }
4485 :
4486 240 : return true;
4487 : }
|
