Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pathnode.c
4 : * Routines to manipulate pathlists and create path nodes
5 : *
6 : * Portions Copyright (c) 1996-2024, 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 <math.h>
18 :
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 List *translate_sub_tlist(List *tlist, int relid);
50 : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
51 : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
52 : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
53 : List *pathlist,
54 : RelOptInfo *child_rel);
55 : static bool pathlist_is_reparameterizable_by_child(List *pathlist,
56 : RelOptInfo *child_rel);
57 :
58 :
59 : /*****************************************************************************
60 : * MISC. PATH UTILITIES
61 : *****************************************************************************/
62 :
63 : /*
64 : * compare_path_costs
65 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
66 : * or more expensive than path2 for the specified criterion.
67 : */
68 : int
69 797406 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
70 : {
71 797406 : if (criterion == STARTUP_COST)
72 : {
73 407796 : if (path1->startup_cost < path2->startup_cost)
74 255668 : return -1;
75 152128 : if (path1->startup_cost > path2->startup_cost)
76 71240 : return +1;
77 :
78 : /*
79 : * If paths have the same startup cost (not at all unlikely), order
80 : * them by total cost.
81 : */
82 80888 : if (path1->total_cost < path2->total_cost)
83 37512 : return -1;
84 43376 : if (path1->total_cost > path2->total_cost)
85 4298 : return +1;
86 : }
87 : else
88 : {
89 389610 : if (path1->total_cost < path2->total_cost)
90 370822 : return -1;
91 18788 : if (path1->total_cost > path2->total_cost)
92 1324 : return +1;
93 :
94 : /*
95 : * If paths have the same total cost, order them by startup cost.
96 : */
97 17464 : if (path1->startup_cost < path2->startup_cost)
98 44 : return -1;
99 17420 : if (path1->startup_cost > path2->startup_cost)
100 40 : return +1;
101 : }
102 56458 : return 0;
103 : }
104 :
105 : /*
106 : * compare_fractional_path_costs
107 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
108 : * or more expensive than path2 for fetching the specified fraction
109 : * of the total tuples.
110 : *
111 : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
112 : * path with the cheaper total_cost.
113 : */
114 : int
115 3222 : compare_fractional_path_costs(Path *path1, Path *path2,
116 : double fraction)
117 : {
118 : Cost cost1,
119 : cost2;
120 :
121 3222 : if (fraction <= 0.0 || fraction >= 1.0)
122 2076 : return compare_path_costs(path1, path2, TOTAL_COST);
123 1146 : cost1 = path1->startup_cost +
124 1146 : fraction * (path1->total_cost - path1->startup_cost);
125 1146 : cost2 = path2->startup_cost +
126 1146 : fraction * (path2->total_cost - path2->startup_cost);
127 1146 : if (cost1 < cost2)
128 676 : return -1;
129 470 : if (cost1 > cost2)
130 470 : return +1;
131 0 : return 0;
132 : }
133 :
134 : /*
135 : * compare_path_costs_fuzzily
136 : * Compare the costs of two paths to see if either can be said to
137 : * dominate the other.
138 : *
139 : * We use fuzzy comparisons so that add_path() can avoid keeping both of
140 : * a pair of paths that really have insignificantly different cost.
141 : *
142 : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
143 : * fraction of the smaller cost that is considered to be a significant
144 : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
145 : * be 1% of the smaller cost.
146 : *
147 : * The two paths are said to have "equal" costs if both startup and total
148 : * costs are fuzzily the same. Path1 is said to be better than path2 if
149 : * it has fuzzily better startup cost and fuzzily no worse total cost,
150 : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
151 : * Path2 is better than path1 if the reverse holds. Finally, if one path
152 : * is fuzzily better than the other on startup cost and fuzzily worse on
153 : * total cost, we just say that their costs are "different", since neither
154 : * dominates the other across the whole performance spectrum.
155 : *
156 : * This function also enforces a policy rule that paths for which the relevant
157 : * one of parent->consider_startup and parent->consider_param_startup is false
158 : * cannot survive comparisons solely on the grounds of good startup cost, so
159 : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
160 : * (But if total costs are fuzzily equal, we compare startup costs anyway,
161 : * in hopes of eliminating one path or the other.)
162 : */
163 : static PathCostComparison
164 3460316 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
165 : {
166 : #define CONSIDER_PATH_STARTUP_COST(p) \
167 : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
168 :
169 : /*
170 : * Check total cost first since it's more likely to be different; many
171 : * paths have zero startup cost.
172 : */
173 3460316 : if (path1->total_cost > path2->total_cost * fuzz_factor)
174 : {
175 : /* path1 fuzzily worse on total cost */
176 1793276 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
177 115544 : path2->startup_cost > path1->startup_cost * fuzz_factor)
178 : {
179 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
180 78354 : return COSTS_DIFFERENT;
181 : }
182 : /* else path2 dominates */
183 1714922 : return COSTS_BETTER2;
184 : }
185 1667040 : if (path2->total_cost > path1->total_cost * fuzz_factor)
186 : {
187 : /* path2 fuzzily worse on total cost */
188 888414 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
189 50414 : path1->startup_cost > path2->startup_cost * fuzz_factor)
190 : {
191 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
192 33288 : return COSTS_DIFFERENT;
193 : }
194 : /* else path1 dominates */
195 855126 : return COSTS_BETTER1;
196 : }
197 : /* fuzzily the same on total cost ... */
198 778626 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
199 : {
200 : /* ... but path1 fuzzily worse on startup, so path2 wins */
201 292008 : return COSTS_BETTER2;
202 : }
203 486618 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
204 : {
205 : /* ... but path2 fuzzily worse on startup, so path1 wins */
206 47564 : return COSTS_BETTER1;
207 : }
208 : /* fuzzily the same on both costs */
209 439054 : return COSTS_EQUAL;
210 :
211 : #undef CONSIDER_PATH_STARTUP_COST
212 : }
213 :
214 : /*
215 : * set_cheapest
216 : * Find the minimum-cost paths from among a relation's paths,
217 : * and save them in the rel's cheapest-path fields.
218 : *
219 : * cheapest_total_path is normally the cheapest-total-cost unparameterized
220 : * path; but if there are no unparameterized paths, we assign it to be the
221 : * best (cheapest least-parameterized) parameterized path. However, only
222 : * unparameterized paths are considered candidates for cheapest_startup_path,
223 : * so that will be NULL if there are no unparameterized paths.
224 : *
225 : * The cheapest_parameterized_paths list collects all parameterized paths
226 : * that have survived the add_path() tournament for this relation. (Since
227 : * add_path ignores pathkeys for a parameterized path, these will be paths
228 : * that have best cost or best row count for their parameterization. We
229 : * may also have both a parallel-safe and a non-parallel-safe path in some
230 : * cases for the same parameterization in some cases, but this should be
231 : * relatively rare since, most typically, all paths for the same relation
232 : * will be parallel-safe or none of them will.)
233 : *
234 : * cheapest_parameterized_paths always includes the cheapest-total
235 : * unparameterized path, too, if there is one; the users of that list find
236 : * it more convenient if that's included.
237 : *
238 : * This is normally called only after we've finished constructing the path
239 : * list for the rel node.
240 : */
241 : void
242 1858688 : set_cheapest(RelOptInfo *parent_rel)
243 : {
244 : Path *cheapest_startup_path;
245 : Path *cheapest_total_path;
246 : Path *best_param_path;
247 : List *parameterized_paths;
248 : ListCell *p;
249 :
250 : Assert(IsA(parent_rel, RelOptInfo));
251 :
252 1858688 : if (parent_rel->pathlist == NIL)
253 0 : elog(ERROR, "could not devise a query plan for the given query");
254 :
255 1858688 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
256 1858688 : parameterized_paths = NIL;
257 :
258 4146628 : foreach(p, parent_rel->pathlist)
259 : {
260 2287940 : Path *path = (Path *) lfirst(p);
261 : int cmp;
262 :
263 2287940 : if (path->param_info)
264 : {
265 : /* Parameterized path, so add it to parameterized_paths */
266 112788 : parameterized_paths = lappend(parameterized_paths, path);
267 :
268 : /*
269 : * If we have an unparameterized cheapest-total, we no longer care
270 : * about finding the best parameterized path, so move on.
271 : */
272 112788 : if (cheapest_total_path)
273 20792 : continue;
274 :
275 : /*
276 : * Otherwise, track the best parameterized path, which is the one
277 : * with least total cost among those of the minimum
278 : * parameterization.
279 : */
280 91996 : if (best_param_path == NULL)
281 85060 : best_param_path = path;
282 : else
283 : {
284 6936 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
285 6936 : PATH_REQ_OUTER(best_param_path)))
286 : {
287 54 : case BMS_EQUAL:
288 : /* keep the cheaper one */
289 54 : if (compare_path_costs(path, best_param_path,
290 : TOTAL_COST) < 0)
291 0 : best_param_path = path;
292 54 : break;
293 340 : case BMS_SUBSET1:
294 : /* new path is less-parameterized */
295 340 : best_param_path = path;
296 340 : break;
297 4 : case BMS_SUBSET2:
298 : /* old path is less-parameterized, keep it */
299 4 : break;
300 6538 : case BMS_DIFFERENT:
301 :
302 : /*
303 : * This means that neither path has the least possible
304 : * parameterization for the rel. We'll sit on the old
305 : * path until something better comes along.
306 : */
307 6538 : break;
308 : }
309 91996 : }
310 : }
311 : else
312 : {
313 : /* Unparameterized path, so consider it for cheapest slots */
314 2175152 : if (cheapest_total_path == NULL)
315 : {
316 1847912 : cheapest_startup_path = cheapest_total_path = path;
317 1847912 : continue;
318 : }
319 :
320 : /*
321 : * If we find two paths of identical costs, try to keep the
322 : * better-sorted one. The paths might have unrelated sort
323 : * orderings, in which case we can only guess which might be
324 : * better to keep, but if one is superior then we definitely
325 : * should keep that one.
326 : */
327 327240 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
328 327240 : if (cmp > 0 ||
329 368 : (cmp == 0 &&
330 368 : compare_pathkeys(cheapest_startup_path->pathkeys,
331 : path->pathkeys) == PATHKEYS_BETTER2))
332 57688 : cheapest_startup_path = path;
333 :
334 327240 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
335 327240 : if (cmp > 0 ||
336 96 : (cmp == 0 &&
337 96 : compare_pathkeys(cheapest_total_path->pathkeys,
338 : path->pathkeys) == PATHKEYS_BETTER2))
339 0 : cheapest_total_path = path;
340 : }
341 : }
342 :
343 : /* Add cheapest unparameterized path, if any, to parameterized_paths */
344 1858688 : if (cheapest_total_path)
345 1847912 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
346 :
347 : /*
348 : * If there is no unparameterized path, use the best parameterized path as
349 : * cheapest_total_path (but not as cheapest_startup_path).
350 : */
351 1858688 : if (cheapest_total_path == NULL)
352 10776 : cheapest_total_path = best_param_path;
353 : Assert(cheapest_total_path != NULL);
354 :
355 1858688 : parent_rel->cheapest_startup_path = cheapest_startup_path;
356 1858688 : parent_rel->cheapest_total_path = cheapest_total_path;
357 1858688 : parent_rel->cheapest_unique_path = NULL; /* computed only if needed */
358 1858688 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
359 1858688 : }
360 :
361 : /*
362 : * add_path
363 : * Consider a potential implementation path for the specified parent rel,
364 : * and add it to the rel's pathlist if it is worthy of consideration.
365 : * A path is worthy if it has a better sort order (better pathkeys) or
366 : * cheaper cost (on either dimension), or generates fewer rows, than any
367 : * existing path that has the same or superset parameterization rels.
368 : * We also consider parallel-safe paths more worthy than others.
369 : *
370 : * We also remove from the rel's pathlist any old paths that are dominated
371 : * by new_path --- that is, new_path is cheaper, at least as well ordered,
372 : * generates no more rows, requires no outer rels not required by the old
373 : * path, and is no less parallel-safe.
374 : *
375 : * In most cases, a path with a superset parameterization will generate
376 : * fewer rows (since it has more join clauses to apply), so that those two
377 : * figures of merit move in opposite directions; this means that a path of
378 : * one parameterization can seldom dominate a path of another. But such
379 : * cases do arise, so we make the full set of checks anyway.
380 : *
381 : * There are two policy decisions embedded in this function, along with
382 : * its sibling add_path_precheck. First, we treat all parameterized paths
383 : * as having NIL pathkeys, so that they cannot win comparisons on the
384 : * basis of sort order. This is to reduce the number of parameterized
385 : * paths that are kept; see discussion in src/backend/optimizer/README.
386 : *
387 : * Second, we only consider cheap startup cost to be interesting if
388 : * parent_rel->consider_startup is true for an unparameterized path, or
389 : * parent_rel->consider_param_startup is true for a parameterized one.
390 : * Again, this allows discarding useless paths sooner.
391 : *
392 : * The pathlist is kept sorted by total_cost, with cheaper paths
393 : * at the front. Within this routine, that's simply a speed hack:
394 : * doing it that way makes it more likely that we will reject an inferior
395 : * path after a few comparisons, rather than many comparisons.
396 : * However, add_path_precheck relies on this ordering to exit early
397 : * when possible.
398 : *
399 : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
400 : * memory consumption. We dare not try to free the substructure of a Path,
401 : * since much of it may be shared with other Paths or the query tree itself;
402 : * but just recycling discarded Path nodes is a very useful savings in
403 : * a large join tree. We can recycle the List nodes of pathlist, too.
404 : *
405 : * As noted in optimizer/README, deleting a previously-accepted Path is
406 : * safe because we know that Paths of this rel cannot yet be referenced
407 : * from any other rel, such as a higher-level join. However, in some cases
408 : * it is possible that a Path is referenced by another Path for its own
409 : * rel; we must not delete such a Path, even if it is dominated by the new
410 : * Path. Currently this occurs only for IndexPath objects, which may be
411 : * referenced as children of BitmapHeapPaths as well as being paths in
412 : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
413 : *
414 : * 'parent_rel' is the relation entry to which the path corresponds.
415 : * 'new_path' is a potential path for parent_rel.
416 : *
417 : * Returns nothing, but modifies parent_rel->pathlist.
418 : */
419 : void
420 3621276 : add_path(RelOptInfo *parent_rel, Path *new_path)
421 : {
422 3621276 : bool accept_new = true; /* unless we find a superior old path */
423 3621276 : int insert_at = 0; /* where to insert new item */
424 : List *new_path_pathkeys;
425 : ListCell *p1;
426 :
427 : /*
428 : * This is a convenient place to check for query cancel --- no part of the
429 : * planner goes very long without calling add_path().
430 : */
431 3621276 : CHECK_FOR_INTERRUPTS();
432 :
433 : /* Pretend parameterized paths have no pathkeys, per comment above */
434 3621276 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
435 :
436 : /*
437 : * Loop to check proposed new path against old paths. Note it is possible
438 : * for more than one old path to be tossed out because new_path dominates
439 : * it.
440 : */
441 5525574 : foreach(p1, parent_rel->pathlist)
442 : {
443 3179956 : Path *old_path = (Path *) lfirst(p1);
444 3179956 : bool remove_old = false; /* unless new proves superior */
445 : PathCostComparison costcmp;
446 : PathKeysComparison keyscmp;
447 : BMS_Comparison outercmp;
448 :
449 : /*
450 : * Do a fuzzy cost comparison with standard fuzziness limit.
451 : */
452 3179956 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
453 : STD_FUZZ_FACTOR);
454 :
455 : /*
456 : * If the two paths compare differently for startup and total cost,
457 : * then we want to keep both, and we can skip comparing pathkeys and
458 : * required_outer rels. If they compare the same, proceed with the
459 : * other comparisons. Row count is checked last. (We make the tests
460 : * in this order because the cost comparison is most likely to turn
461 : * out "different", and the pathkeys comparison next most likely. As
462 : * explained above, row count very seldom makes a difference, so even
463 : * though it's cheap to compare there's not much point in checking it
464 : * earlier.)
465 : */
466 3179956 : if (costcmp != COSTS_DIFFERENT)
467 : {
468 : /* Similarly check to see if either dominates on pathkeys */
469 : List *old_path_pathkeys;
470 :
471 3068344 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
472 3068344 : keyscmp = compare_pathkeys(new_path_pathkeys,
473 : old_path_pathkeys);
474 3068344 : if (keyscmp != PATHKEYS_DIFFERENT)
475 : {
476 2926946 : switch (costcmp)
477 : {
478 300908 : case COSTS_EQUAL:
479 300908 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
480 300908 : PATH_REQ_OUTER(old_path));
481 300908 : if (keyscmp == PATHKEYS_BETTER1)
482 : {
483 3256 : if ((outercmp == BMS_EQUAL ||
484 3256 : outercmp == BMS_SUBSET1) &&
485 3256 : new_path->rows <= old_path->rows &&
486 3248 : new_path->parallel_safe >= old_path->parallel_safe)
487 3248 : remove_old = true; /* new dominates old */
488 : }
489 297652 : else if (keyscmp == PATHKEYS_BETTER2)
490 : {
491 13244 : if ((outercmp == BMS_EQUAL ||
492 13244 : outercmp == BMS_SUBSET2) &&
493 13244 : new_path->rows >= old_path->rows &&
494 13100 : new_path->parallel_safe <= old_path->parallel_safe)
495 13100 : accept_new = false; /* old dominates new */
496 : }
497 : else /* keyscmp == PATHKEYS_EQUAL */
498 : {
499 284408 : if (outercmp == BMS_EQUAL)
500 : {
501 : /*
502 : * Same pathkeys and outer rels, and fuzzily
503 : * the same cost, so keep just one; to decide
504 : * which, first check parallel-safety, then
505 : * rows, then do a fuzzy cost comparison with
506 : * very small fuzz limit. (We used to do an
507 : * exact cost comparison, but that results in
508 : * annoying platform-specific plan variations
509 : * due to roundoff in the cost estimates.) If
510 : * things are still tied, arbitrarily keep
511 : * only the old path. Notice that we will
512 : * keep only the old path even if the
513 : * less-fuzzy comparison decides the startup
514 : * and total costs compare differently.
515 : */
516 280590 : if (new_path->parallel_safe >
517 280590 : old_path->parallel_safe)
518 48 : remove_old = true; /* new dominates old */
519 280542 : else if (new_path->parallel_safe <
520 280542 : old_path->parallel_safe)
521 142 : accept_new = false; /* old dominates new */
522 280400 : else if (new_path->rows < old_path->rows)
523 24 : remove_old = true; /* new dominates old */
524 280376 : else if (new_path->rows > old_path->rows)
525 16 : accept_new = false; /* old dominates new */
526 280360 : else if (compare_path_costs_fuzzily(new_path,
527 : old_path,
528 : 1.0000000001) == COSTS_BETTER1)
529 12828 : remove_old = true; /* new dominates old */
530 : else
531 267532 : accept_new = false; /* old equals or
532 : * dominates new */
533 : }
534 3818 : else if (outercmp == BMS_SUBSET1 &&
535 794 : new_path->rows <= old_path->rows &&
536 766 : new_path->parallel_safe >= old_path->parallel_safe)
537 766 : remove_old = true; /* new dominates old */
538 3052 : else if (outercmp == BMS_SUBSET2 &&
539 2578 : new_path->rows >= old_path->rows &&
540 2558 : new_path->parallel_safe <= old_path->parallel_safe)
541 2558 : accept_new = false; /* old dominates new */
542 : /* else different parameterizations, keep both */
543 : }
544 300908 : break;
545 868862 : case COSTS_BETTER1:
546 868862 : if (keyscmp != PATHKEYS_BETTER2)
547 : {
548 604506 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
549 604506 : PATH_REQ_OUTER(old_path));
550 604506 : if ((outercmp == BMS_EQUAL ||
551 515640 : outercmp == BMS_SUBSET1) &&
552 515640 : new_path->rows <= old_path->rows &&
553 512208 : new_path->parallel_safe >= old_path->parallel_safe)
554 509566 : remove_old = true; /* new dominates old */
555 : }
556 868862 : break;
557 1757176 : case COSTS_BETTER2:
558 1757176 : if (keyscmp != PATHKEYS_BETTER1)
559 : {
560 1127852 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
561 1127852 : PATH_REQ_OUTER(old_path));
562 1127852 : if ((outercmp == BMS_EQUAL ||
563 1045154 : outercmp == BMS_SUBSET2) &&
564 1045154 : new_path->rows >= old_path->rows &&
565 993980 : new_path->parallel_safe <= old_path->parallel_safe)
566 992310 : accept_new = false; /* old dominates new */
567 : }
568 1757176 : break;
569 0 : case COSTS_DIFFERENT:
570 :
571 : /*
572 : * can't get here, but keep this case to keep compiler
573 : * quiet
574 : */
575 0 : break;
576 : }
577 253010 : }
578 : }
579 :
580 : /*
581 : * Remove current element from pathlist if dominated by new.
582 : */
583 3179956 : if (remove_old)
584 : {
585 526480 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
586 : p1);
587 :
588 : /*
589 : * Delete the data pointed-to by the deleted cell, if possible
590 : */
591 526480 : if (!IsA(old_path, IndexPath))
592 508896 : pfree(old_path);
593 : }
594 : else
595 : {
596 : /* new belongs after this old path if it has cost >= old's */
597 2653476 : if (new_path->total_cost >= old_path->total_cost)
598 2200016 : insert_at = foreach_current_index(p1) + 1;
599 : }
600 :
601 : /*
602 : * If we found an old path that dominates new_path, we can quit
603 : * scanning the pathlist; we will not add new_path, and we assume
604 : * new_path cannot dominate any other elements of the pathlist.
605 : */
606 3179956 : if (!accept_new)
607 1275658 : break;
608 : }
609 :
610 3621276 : if (accept_new)
611 : {
612 : /* Accept the new path: insert it at proper place in pathlist */
613 2345618 : parent_rel->pathlist =
614 2345618 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
615 : }
616 : else
617 : {
618 : /* Reject and recycle the new path */
619 1275658 : if (!IsA(new_path, IndexPath))
620 1202236 : pfree(new_path);
621 : }
622 3621276 : }
623 :
624 : /*
625 : * add_path_precheck
626 : * Check whether a proposed new path could possibly get accepted.
627 : * We assume we know the path's pathkeys and parameterization accurately,
628 : * and have lower bounds for its costs.
629 : *
630 : * Note that we do not know the path's rowcount, since getting an estimate for
631 : * that is too expensive to do before prechecking. We assume here that paths
632 : * of a superset parameterization will generate fewer rows; if that holds,
633 : * then paths with different parameterizations cannot dominate each other
634 : * and so we can simply ignore existing paths of another parameterization.
635 : * (In the infrequent cases where that rule of thumb fails, add_path will
636 : * get rid of the inferior path.)
637 : *
638 : * At the time this is called, we haven't actually built a Path structure,
639 : * so the required information has to be passed piecemeal.
640 : */
641 : bool
642 3739878 : add_path_precheck(RelOptInfo *parent_rel,
643 : Cost startup_cost, Cost total_cost,
644 : List *pathkeys, Relids required_outer)
645 : {
646 : List *new_path_pathkeys;
647 : bool consider_startup;
648 : ListCell *p1;
649 :
650 : /* Pretend parameterized paths have no pathkeys, per add_path policy */
651 3739878 : new_path_pathkeys = required_outer ? NIL : pathkeys;
652 :
653 : /* Decide whether new path's startup cost is interesting */
654 3739878 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
655 :
656 4884702 : foreach(p1, parent_rel->pathlist)
657 : {
658 4637486 : Path *old_path = (Path *) lfirst(p1);
659 : PathKeysComparison keyscmp;
660 :
661 : /*
662 : * We are looking for an old_path with the same parameterization (and
663 : * by assumption the same rowcount) that dominates the new path on
664 : * pathkeys as well as both cost metrics. If we find one, we can
665 : * reject the new path.
666 : *
667 : * Cost comparisons here should match compare_path_costs_fuzzily.
668 : */
669 4637486 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
670 : {
671 : /* new path can win on startup cost only if consider_startup */
672 3302244 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
673 1507794 : !consider_startup)
674 : {
675 : /* new path loses on cost, so check pathkeys... */
676 : List *old_path_pathkeys;
677 :
678 3211010 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
679 3211010 : keyscmp = compare_pathkeys(new_path_pathkeys,
680 : old_path_pathkeys);
681 3211010 : if (keyscmp == PATHKEYS_EQUAL ||
682 : keyscmp == PATHKEYS_BETTER2)
683 : {
684 : /* new path does not win on pathkeys... */
685 2232006 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
686 : {
687 : /* Found an old path that dominates the new one */
688 2157420 : return false;
689 : }
690 : }
691 : }
692 : }
693 : else
694 : {
695 : /*
696 : * Since the pathlist is sorted by total_cost, we can stop looking
697 : * once we reach a path with a total_cost larger than the new
698 : * path's.
699 : */
700 1335242 : break;
701 : }
702 : }
703 :
704 1582458 : return true;
705 : }
706 :
707 : /*
708 : * add_partial_path
709 : * Like add_path, our goal here is to consider whether a path is worthy
710 : * of being kept around, but the considerations here are a bit different.
711 : * A partial path is one which can be executed in any number of workers in
712 : * parallel such that each worker will generate a subset of the path's
713 : * overall result.
714 : *
715 : * As in add_path, the partial_pathlist is kept sorted with the cheapest
716 : * total path in front. This is depended on by multiple places, which
717 : * just take the front entry as the cheapest path without searching.
718 : *
719 : * We don't generate parameterized partial paths for several reasons. Most
720 : * importantly, they're not safe to execute, because there's nothing to
721 : * make sure that a parallel scan within the parameterized portion of the
722 : * plan is running with the same value in every worker at the same time.
723 : * Fortunately, it seems unlikely to be worthwhile anyway, because having
724 : * each worker scan the entire outer relation and a subset of the inner
725 : * relation will generally be a terrible plan. The inner (parameterized)
726 : * side of the plan will be small anyway. There could be rare cases where
727 : * this wins big - e.g. if join order constraints put a 1-row relation on
728 : * the outer side of the topmost join with a parameterized plan on the inner
729 : * side - but we'll have to be content not to handle such cases until
730 : * somebody builds an executor infrastructure that can cope with them.
731 : *
732 : * Because we don't consider parameterized paths here, we also don't
733 : * need to consider the row counts as a measure of quality: every path will
734 : * produce the same number of rows. Neither do we need to consider startup
735 : * costs: parallelism is only used for plans that will be run to completion.
736 : * Therefore, this routine is much simpler than add_path: it needs to
737 : * consider only pathkeys and total cost.
738 : *
739 : * As with add_path, we pfree paths that are found to be dominated by
740 : * another partial path; this requires that there be no other references to
741 : * such paths yet. Hence, GatherPaths must not be created for a rel until
742 : * we're done creating all partial paths for it. Unlike add_path, we don't
743 : * take an exception for IndexPaths as partial index paths won't be
744 : * referenced by partial BitmapHeapPaths.
745 : */
746 : void
747 92518 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
748 : {
749 92518 : bool accept_new = true; /* unless we find a superior old path */
750 92518 : int insert_at = 0; /* where to insert new item */
751 : ListCell *p1;
752 :
753 : /* Check for query cancel. */
754 92518 : CHECK_FOR_INTERRUPTS();
755 :
756 : /* Path to be added must be parallel safe. */
757 : Assert(new_path->parallel_safe);
758 :
759 : /* Relation should be OK for parallelism, too. */
760 : Assert(parent_rel->consider_parallel);
761 :
762 : /*
763 : * As in add_path, throw out any paths which are dominated by the new
764 : * path, but throw out the new path if some existing path dominates it.
765 : */
766 124564 : foreach(p1, parent_rel->partial_pathlist)
767 : {
768 46762 : Path *old_path = (Path *) lfirst(p1);
769 46762 : bool remove_old = false; /* unless new proves superior */
770 : PathKeysComparison keyscmp;
771 :
772 : /* Compare pathkeys. */
773 46762 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
774 :
775 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
776 46762 : if (keyscmp != PATHKEYS_DIFFERENT)
777 : {
778 46564 : if (new_path->total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
779 : {
780 : /* New path costs more; keep it only if pathkeys are better. */
781 14978 : if (keyscmp != PATHKEYS_BETTER1)
782 6510 : accept_new = false;
783 : }
784 31586 : else if (old_path->total_cost > new_path->total_cost
785 31586 : * STD_FUZZ_FACTOR)
786 : {
787 : /* Old path costs more; keep it only if pathkeys are better. */
788 23048 : if (keyscmp != PATHKEYS_BETTER2)
789 11930 : remove_old = true;
790 : }
791 8538 : else if (keyscmp == PATHKEYS_BETTER1)
792 : {
793 : /* Costs are about the same, new path has better pathkeys. */
794 0 : remove_old = true;
795 : }
796 8538 : else if (keyscmp == PATHKEYS_BETTER2)
797 : {
798 : /* Costs are about the same, old path has better pathkeys. */
799 1668 : accept_new = false;
800 : }
801 6870 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
802 : {
803 : /* Pathkeys are the same, and the old path costs more. */
804 332 : remove_old = true;
805 : }
806 : else
807 : {
808 : /*
809 : * Pathkeys are the same, and new path isn't materially
810 : * cheaper.
811 : */
812 6538 : accept_new = false;
813 : }
814 : }
815 :
816 : /*
817 : * Remove current element from partial_pathlist if dominated by new.
818 : */
819 46762 : if (remove_old)
820 : {
821 12262 : parent_rel->partial_pathlist =
822 12262 : foreach_delete_current(parent_rel->partial_pathlist, p1);
823 12262 : pfree(old_path);
824 : }
825 : else
826 : {
827 : /* new belongs after this old path if it has cost >= old's */
828 34500 : if (new_path->total_cost >= old_path->total_cost)
829 22956 : insert_at = foreach_current_index(p1) + 1;
830 : }
831 :
832 : /*
833 : * If we found an old path that dominates new_path, we can quit
834 : * scanning the partial_pathlist; we will not add new_path, and we
835 : * assume new_path cannot dominate any later path.
836 : */
837 46762 : if (!accept_new)
838 14716 : break;
839 : }
840 :
841 92518 : if (accept_new)
842 : {
843 : /* Accept the new path: insert it at proper place */
844 77802 : parent_rel->partial_pathlist =
845 77802 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
846 : }
847 : else
848 : {
849 : /* Reject and recycle the new path */
850 14716 : pfree(new_path);
851 : }
852 92518 : }
853 :
854 : /*
855 : * add_partial_path_precheck
856 : * Check whether a proposed new partial path could possibly get accepted.
857 : *
858 : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
859 : * since they don't matter for partial paths (see add_partial_path). But
860 : * we do want to make sure we don't add a partial path if there's already
861 : * a complete path that dominates it, since in that case the proposed path
862 : * is surely a loser.
863 : */
864 : bool
865 63342 : add_partial_path_precheck(RelOptInfo *parent_rel, Cost total_cost,
866 : List *pathkeys)
867 : {
868 : ListCell *p1;
869 :
870 : /*
871 : * Our goal here is twofold. First, we want to find out whether this path
872 : * is clearly inferior to some existing partial path. If so, we want to
873 : * reject it immediately. Second, we want to find out whether this path
874 : * is clearly superior to some existing partial path -- at least, modulo
875 : * final cost computations. If so, we definitely want to consider it.
876 : *
877 : * Unlike add_path(), we always compare pathkeys here. This is because we
878 : * expect partial_pathlist to be very short, and getting a definitive
879 : * answer at this stage avoids the need to call add_path_precheck.
880 : */
881 88598 : foreach(p1, parent_rel->partial_pathlist)
882 : {
883 70750 : Path *old_path = (Path *) lfirst(p1);
884 : PathKeysComparison keyscmp;
885 :
886 70750 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
887 70750 : if (keyscmp != PATHKEYS_DIFFERENT)
888 : {
889 70558 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
890 : keyscmp != PATHKEYS_BETTER1)
891 45494 : return false;
892 35354 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
893 : keyscmp != PATHKEYS_BETTER2)
894 10290 : return true;
895 : }
896 : }
897 :
898 : /*
899 : * This path is neither clearly inferior to an existing partial path nor
900 : * clearly good enough that it might replace one. Compare it to
901 : * non-parallel plans. If it loses even before accounting for the cost of
902 : * the Gather node, we should definitely reject it.
903 : *
904 : * Note that we pass the total_cost to add_path_precheck twice. This is
905 : * because it's never advantageous to consider the startup cost of a
906 : * partial path; the resulting plans, if run in parallel, will be run to
907 : * completion.
908 : */
909 17848 : if (!add_path_precheck(parent_rel, total_cost, total_cost, pathkeys,
910 : NULL))
911 700 : return false;
912 :
913 17148 : return true;
914 : }
915 :
916 :
917 : /*****************************************************************************
918 : * PATH NODE CREATION ROUTINES
919 : *****************************************************************************/
920 :
921 : /*
922 : * create_seqscan_path
923 : * Creates a path corresponding to a sequential scan, returning the
924 : * pathnode.
925 : */
926 : Path *
927 370752 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
928 : Relids required_outer, int parallel_workers)
929 : {
930 370752 : Path *pathnode = makeNode(Path);
931 :
932 370752 : pathnode->pathtype = T_SeqScan;
933 370752 : pathnode->parent = rel;
934 370752 : pathnode->pathtarget = rel->reltarget;
935 370752 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
936 : required_outer);
937 370752 : pathnode->parallel_aware = (parallel_workers > 0);
938 370752 : pathnode->parallel_safe = rel->consider_parallel;
939 370752 : pathnode->parallel_workers = parallel_workers;
940 370752 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
941 :
942 370752 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
943 :
944 370752 : return pathnode;
945 : }
946 :
947 : /*
948 : * create_samplescan_path
949 : * Creates a path node for a sampled table scan.
950 : */
951 : Path *
952 300 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
953 : {
954 300 : Path *pathnode = makeNode(Path);
955 :
956 300 : pathnode->pathtype = T_SampleScan;
957 300 : pathnode->parent = rel;
958 300 : pathnode->pathtarget = rel->reltarget;
959 300 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
960 : required_outer);
961 300 : pathnode->parallel_aware = false;
962 300 : pathnode->parallel_safe = rel->consider_parallel;
963 300 : pathnode->parallel_workers = 0;
964 300 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
965 :
966 300 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
967 :
968 300 : return pathnode;
969 : }
970 :
971 : /*
972 : * create_index_path
973 : * Creates a path node for an index scan.
974 : *
975 : * 'index' is a usable index.
976 : * 'indexclauses' is a list of IndexClause nodes representing clauses
977 : * to be enforced as qual conditions in the scan.
978 : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
979 : * to be used as index ordering operators in the scan.
980 : * 'indexorderbycols' is an integer list of index column numbers (zero based)
981 : * the ordering operators can be used with.
982 : * 'pathkeys' describes the ordering of the path.
983 : * 'indexscandir' is either ForwardScanDirection or BackwardScanDirection.
984 : * 'indexonly' is true if an index-only scan is wanted.
985 : * 'required_outer' is the set of outer relids for a parameterized path.
986 : * 'loop_count' is the number of repetitions of the indexscan to factor into
987 : * estimates of caching behavior.
988 : * 'partial_path' is true if constructing a parallel index scan path.
989 : *
990 : * Returns the new path node.
991 : */
992 : IndexPath *
993 643440 : create_index_path(PlannerInfo *root,
994 : IndexOptInfo *index,
995 : List *indexclauses,
996 : List *indexorderbys,
997 : List *indexorderbycols,
998 : List *pathkeys,
999 : ScanDirection indexscandir,
1000 : bool indexonly,
1001 : Relids required_outer,
1002 : double loop_count,
1003 : bool partial_path)
1004 : {
1005 643440 : IndexPath *pathnode = makeNode(IndexPath);
1006 643440 : RelOptInfo *rel = index->rel;
1007 :
1008 643440 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1009 643440 : pathnode->path.parent = rel;
1010 643440 : pathnode->path.pathtarget = rel->reltarget;
1011 643440 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1012 : required_outer);
1013 643440 : pathnode->path.parallel_aware = false;
1014 643440 : pathnode->path.parallel_safe = rel->consider_parallel;
1015 643440 : pathnode->path.parallel_workers = 0;
1016 643440 : pathnode->path.pathkeys = pathkeys;
1017 :
1018 643440 : pathnode->indexinfo = index;
1019 643440 : pathnode->indexclauses = indexclauses;
1020 643440 : pathnode->indexorderbys = indexorderbys;
1021 643440 : pathnode->indexorderbycols = indexorderbycols;
1022 643440 : pathnode->indexscandir = indexscandir;
1023 :
1024 643440 : cost_index(pathnode, root, loop_count, partial_path);
1025 :
1026 643440 : return pathnode;
1027 : }
1028 :
1029 : /*
1030 : * create_bitmap_heap_path
1031 : * Creates a path node for a bitmap scan.
1032 : *
1033 : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1034 : * 'required_outer' is the set of outer relids for a parameterized path.
1035 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1036 : * estimates of caching behavior.
1037 : *
1038 : * loop_count should match the value used when creating the component
1039 : * IndexPaths.
1040 : */
1041 : BitmapHeapPath *
1042 290220 : create_bitmap_heap_path(PlannerInfo *root,
1043 : RelOptInfo *rel,
1044 : Path *bitmapqual,
1045 : Relids required_outer,
1046 : double loop_count,
1047 : int parallel_degree)
1048 : {
1049 290220 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1050 :
1051 290220 : pathnode->path.pathtype = T_BitmapHeapScan;
1052 290220 : pathnode->path.parent = rel;
1053 290220 : pathnode->path.pathtarget = rel->reltarget;
1054 290220 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1055 : required_outer);
1056 290220 : pathnode->path.parallel_aware = (parallel_degree > 0);
1057 290220 : pathnode->path.parallel_safe = rel->consider_parallel;
1058 290220 : pathnode->path.parallel_workers = parallel_degree;
1059 290220 : pathnode->path.pathkeys = NIL; /* always unordered */
1060 :
1061 290220 : pathnode->bitmapqual = bitmapqual;
1062 :
1063 290220 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1064 : pathnode->path.param_info,
1065 : bitmapqual, loop_count);
1066 :
1067 290220 : return pathnode;
1068 : }
1069 :
1070 : /*
1071 : * create_bitmap_and_path
1072 : * Creates a path node representing a BitmapAnd.
1073 : */
1074 : BitmapAndPath *
1075 33738 : create_bitmap_and_path(PlannerInfo *root,
1076 : RelOptInfo *rel,
1077 : List *bitmapquals)
1078 : {
1079 33738 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1080 33738 : Relids required_outer = NULL;
1081 : ListCell *lc;
1082 :
1083 33738 : pathnode->path.pathtype = T_BitmapAnd;
1084 33738 : pathnode->path.parent = rel;
1085 33738 : pathnode->path.pathtarget = rel->reltarget;
1086 :
1087 : /*
1088 : * Identify the required outer rels as the union of what the child paths
1089 : * depend on. (Alternatively, we could insist that the caller pass this
1090 : * in, but it's more convenient and reliable to compute it here.)
1091 : */
1092 101214 : foreach(lc, bitmapquals)
1093 : {
1094 67476 : Path *bitmapqual = (Path *) lfirst(lc);
1095 :
1096 67476 : required_outer = bms_add_members(required_outer,
1097 67476 : PATH_REQ_OUTER(bitmapqual));
1098 : }
1099 33738 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1100 : required_outer);
1101 :
1102 : /*
1103 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1104 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1105 : * set the flag for this path based only on the relation-level flag,
1106 : * without actually iterating over the list of children.
1107 : */
1108 33738 : pathnode->path.parallel_aware = false;
1109 33738 : pathnode->path.parallel_safe = rel->consider_parallel;
1110 33738 : pathnode->path.parallel_workers = 0;
1111 :
1112 33738 : pathnode->path.pathkeys = NIL; /* always unordered */
1113 :
1114 33738 : pathnode->bitmapquals = bitmapquals;
1115 :
1116 : /* this sets bitmapselectivity as well as the regular cost fields: */
1117 33738 : cost_bitmap_and_node(pathnode, root);
1118 :
1119 33738 : return pathnode;
1120 : }
1121 :
1122 : /*
1123 : * create_bitmap_or_path
1124 : * Creates a path node representing a BitmapOr.
1125 : */
1126 : BitmapOrPath *
1127 960 : create_bitmap_or_path(PlannerInfo *root,
1128 : RelOptInfo *rel,
1129 : List *bitmapquals)
1130 : {
1131 960 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1132 960 : Relids required_outer = NULL;
1133 : ListCell *lc;
1134 :
1135 960 : pathnode->path.pathtype = T_BitmapOr;
1136 960 : pathnode->path.parent = rel;
1137 960 : pathnode->path.pathtarget = rel->reltarget;
1138 :
1139 : /*
1140 : * Identify the required outer rels as the union of what the child paths
1141 : * depend on. (Alternatively, we could insist that the caller pass this
1142 : * in, but it's more convenient and reliable to compute it here.)
1143 : */
1144 2940 : foreach(lc, bitmapquals)
1145 : {
1146 1980 : Path *bitmapqual = (Path *) lfirst(lc);
1147 :
1148 1980 : required_outer = bms_add_members(required_outer,
1149 1980 : PATH_REQ_OUTER(bitmapqual));
1150 : }
1151 960 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1152 : required_outer);
1153 :
1154 : /*
1155 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1156 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1157 : * set the flag for this path based only on the relation-level flag,
1158 : * without actually iterating over the list of children.
1159 : */
1160 960 : pathnode->path.parallel_aware = false;
1161 960 : pathnode->path.parallel_safe = rel->consider_parallel;
1162 960 : pathnode->path.parallel_workers = 0;
1163 :
1164 960 : pathnode->path.pathkeys = NIL; /* always unordered */
1165 :
1166 960 : pathnode->bitmapquals = bitmapquals;
1167 :
1168 : /* this sets bitmapselectivity as well as the regular cost fields: */
1169 960 : cost_bitmap_or_node(pathnode, root);
1170 :
1171 960 : return pathnode;
1172 : }
1173 :
1174 : /*
1175 : * create_tidscan_path
1176 : * Creates a path corresponding to a scan by TID, returning the pathnode.
1177 : */
1178 : TidPath *
1179 756 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1180 : Relids required_outer)
1181 : {
1182 756 : TidPath *pathnode = makeNode(TidPath);
1183 :
1184 756 : pathnode->path.pathtype = T_TidScan;
1185 756 : pathnode->path.parent = rel;
1186 756 : pathnode->path.pathtarget = rel->reltarget;
1187 756 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1188 : required_outer);
1189 756 : pathnode->path.parallel_aware = false;
1190 756 : pathnode->path.parallel_safe = rel->consider_parallel;
1191 756 : pathnode->path.parallel_workers = 0;
1192 756 : pathnode->path.pathkeys = NIL; /* always unordered */
1193 :
1194 756 : pathnode->tidquals = tidquals;
1195 :
1196 756 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1197 : pathnode->path.param_info);
1198 :
1199 756 : return pathnode;
1200 : }
1201 :
1202 : /*
1203 : * create_tidrangescan_path
1204 : * Creates a path corresponding to a scan by a range of TIDs, returning
1205 : * the pathnode.
1206 : */
1207 : TidRangePath *
1208 202 : create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
1209 : List *tidrangequals, Relids required_outer)
1210 : {
1211 202 : TidRangePath *pathnode = makeNode(TidRangePath);
1212 :
1213 202 : pathnode->path.pathtype = T_TidRangeScan;
1214 202 : pathnode->path.parent = rel;
1215 202 : pathnode->path.pathtarget = rel->reltarget;
1216 202 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1217 : required_outer);
1218 202 : pathnode->path.parallel_aware = false;
1219 202 : pathnode->path.parallel_safe = rel->consider_parallel;
1220 202 : pathnode->path.parallel_workers = 0;
1221 202 : pathnode->path.pathkeys = NIL; /* always unordered */
1222 :
1223 202 : pathnode->tidrangequals = tidrangequals;
1224 :
1225 202 : cost_tidrangescan(&pathnode->path, root, rel, tidrangequals,
1226 : pathnode->path.param_info);
1227 :
1228 202 : return pathnode;
1229 : }
1230 :
1231 : /*
1232 : * create_append_path
1233 : * Creates a path corresponding to an Append plan, returning the
1234 : * pathnode.
1235 : *
1236 : * Note that we must handle subpaths = NIL, representing a dummy access path.
1237 : * Also, there are callers that pass root = NULL.
1238 : *
1239 : * 'rows', when passed as a non-negative number, will be used to overwrite the
1240 : * returned path's row estimate. Otherwise, the row estimate is calculated
1241 : * by totalling the row estimates from the 'subpaths' list.
1242 : */
1243 : AppendPath *
1244 67400 : create_append_path(PlannerInfo *root,
1245 : RelOptInfo *rel,
1246 : List *subpaths, List *partial_subpaths,
1247 : List *pathkeys, Relids required_outer,
1248 : int parallel_workers, bool parallel_aware,
1249 : double rows)
1250 : {
1251 67400 : AppendPath *pathnode = makeNode(AppendPath);
1252 : ListCell *l;
1253 :
1254 : Assert(!parallel_aware || parallel_workers > 0);
1255 :
1256 67400 : pathnode->path.pathtype = T_Append;
1257 67400 : pathnode->path.parent = rel;
1258 67400 : pathnode->path.pathtarget = rel->reltarget;
1259 :
1260 : /*
1261 : * If this is for a baserel (not a join or non-leaf partition), we prefer
1262 : * to apply get_baserel_parampathinfo to construct a full ParamPathInfo
1263 : * for the path. This supports building a Memoize path atop this path,
1264 : * and if this is a partitioned table the info may be useful for run-time
1265 : * pruning (cf make_partition_pruneinfo()).
1266 : *
1267 : * However, if we don't have "root" then that won't work and we fall back
1268 : * on the simpler get_appendrel_parampathinfo. There's no point in doing
1269 : * the more expensive thing for a dummy path, either.
1270 : */
1271 67400 : if (rel->reloptkind == RELOPT_BASEREL && root && subpaths != NIL)
1272 30530 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1273 : rel,
1274 : required_outer);
1275 : else
1276 36870 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1277 : required_outer);
1278 :
1279 67400 : pathnode->path.parallel_aware = parallel_aware;
1280 67400 : pathnode->path.parallel_safe = rel->consider_parallel;
1281 67400 : pathnode->path.parallel_workers = parallel_workers;
1282 67400 : pathnode->path.pathkeys = pathkeys;
1283 :
1284 : /*
1285 : * For parallel append, non-partial paths are sorted by descending total
1286 : * costs. That way, the total time to finish all non-partial paths is
1287 : * minimized. Also, the partial paths are sorted by descending startup
1288 : * costs. There may be some paths that require to do startup work by a
1289 : * single worker. In such case, it's better for workers to choose the
1290 : * expensive ones first, whereas the leader should choose the cheapest
1291 : * startup plan.
1292 : */
1293 67400 : if (pathnode->path.parallel_aware)
1294 : {
1295 : /*
1296 : * We mustn't fiddle with the order of subpaths when the Append has
1297 : * pathkeys. The order they're listed in is critical to keeping the
1298 : * pathkeys valid.
1299 : */
1300 : Assert(pathkeys == NIL);
1301 :
1302 22610 : list_sort(subpaths, append_total_cost_compare);
1303 22610 : list_sort(partial_subpaths, append_startup_cost_compare);
1304 : }
1305 67400 : pathnode->first_partial_path = list_length(subpaths);
1306 67400 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1307 :
1308 : /*
1309 : * Apply query-wide LIMIT if known and path is for sole base relation.
1310 : * (Handling this at this low level is a bit klugy.)
1311 : */
1312 67400 : if (root != NULL && bms_equal(rel->relids, root->all_query_rels))
1313 37422 : pathnode->limit_tuples = root->limit_tuples;
1314 : else
1315 29978 : pathnode->limit_tuples = -1.0;
1316 :
1317 222156 : foreach(l, pathnode->subpaths)
1318 : {
1319 154756 : Path *subpath = (Path *) lfirst(l);
1320 :
1321 275206 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1322 120450 : subpath->parallel_safe;
1323 :
1324 : /* All child paths must have same parameterization */
1325 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1326 : }
1327 :
1328 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1329 :
1330 : /*
1331 : * If there's exactly one child path then the output of the Append is
1332 : * necessarily ordered the same as the child's, so we can inherit the
1333 : * child's pathkeys if any, overriding whatever the caller might've said.
1334 : * Furthermore, if the child's parallel awareness matches the Append's,
1335 : * then the Append is a no-op and will be discarded later (in setrefs.c).
1336 : * Then we can inherit the child's size and cost too, effectively charging
1337 : * zero for the Append. Otherwise, we must do the normal costsize
1338 : * calculation.
1339 : */
1340 67400 : if (list_length(pathnode->subpaths) == 1)
1341 : {
1342 21600 : Path *child = (Path *) linitial(pathnode->subpaths);
1343 :
1344 21600 : if (child->parallel_aware == parallel_aware)
1345 : {
1346 21168 : pathnode->path.rows = child->rows;
1347 21168 : pathnode->path.startup_cost = child->startup_cost;
1348 21168 : pathnode->path.total_cost = child->total_cost;
1349 : }
1350 : else
1351 432 : cost_append(pathnode);
1352 : /* Must do this last, else cost_append complains */
1353 21600 : pathnode->path.pathkeys = child->pathkeys;
1354 : }
1355 : else
1356 45800 : cost_append(pathnode);
1357 :
1358 : /* If the caller provided a row estimate, override the computed value. */
1359 67400 : if (rows >= 0)
1360 582 : pathnode->path.rows = rows;
1361 :
1362 67400 : return pathnode;
1363 : }
1364 :
1365 : /*
1366 : * append_total_cost_compare
1367 : * list_sort comparator for sorting append child paths
1368 : * by total_cost descending
1369 : *
1370 : * For equal total costs, we fall back to comparing startup costs; if those
1371 : * are equal too, break ties using bms_compare on the paths' relids.
1372 : * (This is to avoid getting unpredictable results from list_sort.)
1373 : */
1374 : static int
1375 1996 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1376 : {
1377 1996 : Path *path1 = (Path *) lfirst(a);
1378 1996 : Path *path2 = (Path *) lfirst(b);
1379 : int cmp;
1380 :
1381 1996 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1382 1996 : if (cmp != 0)
1383 1726 : return -cmp;
1384 270 : return bms_compare(path1->parent->relids, path2->parent->relids);
1385 : }
1386 :
1387 : /*
1388 : * append_startup_cost_compare
1389 : * list_sort comparator for sorting append child paths
1390 : * by startup_cost descending
1391 : *
1392 : * For equal startup costs, we fall back to comparing total costs; if those
1393 : * are equal too, break ties using bms_compare on the paths' relids.
1394 : * (This is to avoid getting unpredictable results from list_sort.)
1395 : */
1396 : static int
1397 33202 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1398 : {
1399 33202 : Path *path1 = (Path *) lfirst(a);
1400 33202 : Path *path2 = (Path *) lfirst(b);
1401 : int cmp;
1402 :
1403 33202 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1404 33202 : if (cmp != 0)
1405 12872 : return -cmp;
1406 20330 : return bms_compare(path1->parent->relids, path2->parent->relids);
1407 : }
1408 :
1409 : /*
1410 : * create_merge_append_path
1411 : * Creates a path corresponding to a MergeAppend plan, returning the
1412 : * pathnode.
1413 : */
1414 : MergeAppendPath *
1415 3886 : create_merge_append_path(PlannerInfo *root,
1416 : RelOptInfo *rel,
1417 : List *subpaths,
1418 : List *pathkeys,
1419 : Relids required_outer)
1420 : {
1421 3886 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1422 : Cost input_startup_cost;
1423 : Cost input_total_cost;
1424 : ListCell *l;
1425 :
1426 3886 : pathnode->path.pathtype = T_MergeAppend;
1427 3886 : pathnode->path.parent = rel;
1428 3886 : pathnode->path.pathtarget = rel->reltarget;
1429 3886 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1430 : required_outer);
1431 3886 : pathnode->path.parallel_aware = false;
1432 3886 : pathnode->path.parallel_safe = rel->consider_parallel;
1433 3886 : pathnode->path.parallel_workers = 0;
1434 3886 : pathnode->path.pathkeys = pathkeys;
1435 3886 : pathnode->subpaths = subpaths;
1436 :
1437 : /*
1438 : * Apply query-wide LIMIT if known and path is for sole base relation.
1439 : * (Handling this at this low level is a bit klugy.)
1440 : */
1441 3886 : if (bms_equal(rel->relids, root->all_query_rels))
1442 2028 : pathnode->limit_tuples = root->limit_tuples;
1443 : else
1444 1858 : pathnode->limit_tuples = -1.0;
1445 :
1446 : /*
1447 : * Add up the sizes and costs of the input paths.
1448 : */
1449 3886 : pathnode->path.rows = 0;
1450 3886 : input_startup_cost = 0;
1451 3886 : input_total_cost = 0;
1452 14532 : foreach(l, subpaths)
1453 : {
1454 10646 : Path *subpath = (Path *) lfirst(l);
1455 :
1456 10646 : pathnode->path.rows += subpath->rows;
1457 19186 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1458 8540 : subpath->parallel_safe;
1459 :
1460 10646 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1461 : {
1462 : /* Subpath is adequately ordered, we won't need to sort it */
1463 10360 : input_startup_cost += subpath->startup_cost;
1464 10360 : input_total_cost += subpath->total_cost;
1465 : }
1466 : else
1467 : {
1468 : /* We'll need to insert a Sort node, so include cost for that */
1469 : Path sort_path; /* dummy for result of cost_sort */
1470 :
1471 286 : cost_sort(&sort_path,
1472 : root,
1473 : pathkeys,
1474 : subpath->total_cost,
1475 : subpath->rows,
1476 286 : subpath->pathtarget->width,
1477 : 0.0,
1478 : work_mem,
1479 : pathnode->limit_tuples);
1480 286 : input_startup_cost += sort_path.startup_cost;
1481 286 : input_total_cost += sort_path.total_cost;
1482 : }
1483 :
1484 : /* All child paths must have same parameterization */
1485 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1486 : }
1487 :
1488 : /*
1489 : * Now we can compute total costs of the MergeAppend. If there's exactly
1490 : * one child path and its parallel awareness matches that of the
1491 : * MergeAppend, then the MergeAppend is a no-op and will be discarded
1492 : * later (in setrefs.c); otherwise we do the normal cost calculation.
1493 : */
1494 3886 : if (list_length(subpaths) == 1 &&
1495 110 : ((Path *) linitial(subpaths))->parallel_aware ==
1496 110 : pathnode->path.parallel_aware)
1497 : {
1498 110 : pathnode->path.startup_cost = input_startup_cost;
1499 110 : pathnode->path.total_cost = input_total_cost;
1500 : }
1501 : else
1502 3776 : cost_merge_append(&pathnode->path, root,
1503 : pathkeys, list_length(subpaths),
1504 : input_startup_cost, input_total_cost,
1505 : pathnode->path.rows);
1506 :
1507 3886 : return pathnode;
1508 : }
1509 :
1510 : /*
1511 : * create_group_result_path
1512 : * Creates a path representing a Result-and-nothing-else plan.
1513 : *
1514 : * This is only used for degenerate grouping cases, in which we know we
1515 : * need to produce one result row, possibly filtered by a HAVING qual.
1516 : */
1517 : GroupResultPath *
1518 203974 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1519 : PathTarget *target, List *havingqual)
1520 : {
1521 203974 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1522 :
1523 203974 : pathnode->path.pathtype = T_Result;
1524 203974 : pathnode->path.parent = rel;
1525 203974 : pathnode->path.pathtarget = target;
1526 203974 : pathnode->path.param_info = NULL; /* there are no other rels... */
1527 203974 : pathnode->path.parallel_aware = false;
1528 203974 : pathnode->path.parallel_safe = rel->consider_parallel;
1529 203974 : pathnode->path.parallel_workers = 0;
1530 203974 : pathnode->path.pathkeys = NIL;
1531 203974 : pathnode->quals = havingqual;
1532 :
1533 : /*
1534 : * We can't quite use cost_resultscan() because the quals we want to
1535 : * account for are not baserestrict quals of the rel. Might as well just
1536 : * hack it here.
1537 : */
1538 203974 : pathnode->path.rows = 1;
1539 203974 : pathnode->path.startup_cost = target->cost.startup;
1540 203974 : pathnode->path.total_cost = target->cost.startup +
1541 203974 : cpu_tuple_cost + target->cost.per_tuple;
1542 :
1543 : /*
1544 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1545 : * rowcount estimate should be 1 no matter what the qual is.
1546 : */
1547 203974 : if (havingqual)
1548 : {
1549 : QualCost qual_cost;
1550 :
1551 516 : cost_qual_eval(&qual_cost, havingqual, root);
1552 : /* havingqual is evaluated once at startup */
1553 516 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1554 516 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1555 : }
1556 :
1557 203974 : return pathnode;
1558 : }
1559 :
1560 : /*
1561 : * create_material_path
1562 : * Creates a path corresponding to a Material plan, returning the
1563 : * pathnode.
1564 : */
1565 : MaterialPath *
1566 419878 : create_material_path(RelOptInfo *rel, Path *subpath)
1567 : {
1568 419878 : MaterialPath *pathnode = makeNode(MaterialPath);
1569 :
1570 : Assert(subpath->parent == rel);
1571 :
1572 419878 : pathnode->path.pathtype = T_Material;
1573 419878 : pathnode->path.parent = rel;
1574 419878 : pathnode->path.pathtarget = rel->reltarget;
1575 419878 : pathnode->path.param_info = subpath->param_info;
1576 419878 : pathnode->path.parallel_aware = false;
1577 793754 : pathnode->path.parallel_safe = rel->consider_parallel &&
1578 373876 : subpath->parallel_safe;
1579 419878 : pathnode->path.parallel_workers = subpath->parallel_workers;
1580 419878 : pathnode->path.pathkeys = subpath->pathkeys;
1581 :
1582 419878 : pathnode->subpath = subpath;
1583 :
1584 419878 : cost_material(&pathnode->path,
1585 : subpath->startup_cost,
1586 : subpath->total_cost,
1587 : subpath->rows,
1588 419878 : subpath->pathtarget->width);
1589 :
1590 419878 : return pathnode;
1591 : }
1592 :
1593 : /*
1594 : * create_memoize_path
1595 : * Creates a path corresponding to a Memoize plan, returning the pathnode.
1596 : */
1597 : MemoizePath *
1598 240650 : create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1599 : List *param_exprs, List *hash_operators,
1600 : bool singlerow, bool binary_mode, double calls)
1601 : {
1602 240650 : MemoizePath *pathnode = makeNode(MemoizePath);
1603 :
1604 : Assert(subpath->parent == rel);
1605 :
1606 240650 : pathnode->path.pathtype = T_Memoize;
1607 240650 : pathnode->path.parent = rel;
1608 240650 : pathnode->path.pathtarget = rel->reltarget;
1609 240650 : pathnode->path.param_info = subpath->param_info;
1610 240650 : pathnode->path.parallel_aware = false;
1611 471594 : pathnode->path.parallel_safe = rel->consider_parallel &&
1612 230944 : subpath->parallel_safe;
1613 240650 : pathnode->path.parallel_workers = subpath->parallel_workers;
1614 240650 : pathnode->path.pathkeys = subpath->pathkeys;
1615 :
1616 240650 : pathnode->subpath = subpath;
1617 240650 : pathnode->hash_operators = hash_operators;
1618 240650 : pathnode->param_exprs = param_exprs;
1619 240650 : pathnode->singlerow = singlerow;
1620 240650 : pathnode->binary_mode = binary_mode;
1621 240650 : pathnode->calls = calls;
1622 :
1623 : /*
1624 : * For now we set est_entries to 0. cost_memoize_rescan() does all the
1625 : * hard work to determine how many cache entries there are likely to be,
1626 : * so it seems best to leave it up to that function to fill this field in.
1627 : * If left at 0, the executor will make a guess at a good value.
1628 : */
1629 240650 : pathnode->est_entries = 0;
1630 :
1631 : /*
1632 : * Add a small additional charge for caching the first entry. All the
1633 : * harder calculations for rescans are performed in cost_memoize_rescan().
1634 : */
1635 240650 : pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
1636 240650 : pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
1637 240650 : pathnode->path.rows = subpath->rows;
1638 :
1639 240650 : return pathnode;
1640 : }
1641 :
1642 : /*
1643 : * create_unique_path
1644 : * Creates a path representing elimination of distinct rows from the
1645 : * input data. Distinct-ness is defined according to the needs of the
1646 : * semijoin represented by sjinfo. If it is not possible to identify
1647 : * how to make the data unique, NULL is returned.
1648 : *
1649 : * If used at all, this is likely to be called repeatedly on the same rel;
1650 : * and the input subpath should always be the same (the cheapest_total path
1651 : * for the rel). So we cache the result.
1652 : */
1653 : UniquePath *
1654 29468 : create_unique_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1655 : SpecialJoinInfo *sjinfo)
1656 : {
1657 : UniquePath *pathnode;
1658 : Path sort_path; /* dummy for result of cost_sort */
1659 : Path agg_path; /* dummy for result of cost_agg */
1660 : MemoryContext oldcontext;
1661 : int numCols;
1662 :
1663 : /* Caller made a mistake if subpath isn't cheapest_total ... */
1664 : Assert(subpath == rel->cheapest_total_path);
1665 : Assert(subpath->parent == rel);
1666 : /* ... or if SpecialJoinInfo is the wrong one */
1667 : Assert(sjinfo->jointype == JOIN_SEMI);
1668 : Assert(bms_equal(rel->relids, sjinfo->syn_righthand));
1669 :
1670 : /* If result already cached, return it */
1671 29468 : if (rel->cheapest_unique_path)
1672 25838 : return (UniquePath *) rel->cheapest_unique_path;
1673 :
1674 : /* If it's not possible to unique-ify, return NULL */
1675 3630 : if (!(sjinfo->semi_can_btree || sjinfo->semi_can_hash))
1676 120 : return NULL;
1677 :
1678 : /*
1679 : * When called during GEQO join planning, we are in a short-lived memory
1680 : * context. We must make sure that the path and any subsidiary data
1681 : * structures created for a baserel survive the GEQO cycle, else the
1682 : * baserel is trashed for future GEQO cycles. On the other hand, when we
1683 : * are creating those for a joinrel during GEQO, we don't want them to
1684 : * clutter the main planning context. Upshot is that the best solution is
1685 : * to explicitly allocate memory in the same context the given RelOptInfo
1686 : * is in.
1687 : */
1688 3510 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
1689 :
1690 3510 : pathnode = makeNode(UniquePath);
1691 :
1692 3510 : pathnode->path.pathtype = T_Unique;
1693 3510 : pathnode->path.parent = rel;
1694 3510 : pathnode->path.pathtarget = rel->reltarget;
1695 3510 : pathnode->path.param_info = subpath->param_info;
1696 3510 : pathnode->path.parallel_aware = false;
1697 6560 : pathnode->path.parallel_safe = rel->consider_parallel &&
1698 3050 : subpath->parallel_safe;
1699 3510 : pathnode->path.parallel_workers = subpath->parallel_workers;
1700 :
1701 : /*
1702 : * Assume the output is unsorted, since we don't necessarily have pathkeys
1703 : * to represent it. (This might get overridden below.)
1704 : */
1705 3510 : pathnode->path.pathkeys = NIL;
1706 :
1707 3510 : pathnode->subpath = subpath;
1708 :
1709 : /*
1710 : * Under GEQO and when planning child joins, the sjinfo might be
1711 : * short-lived, so we'd better make copies of data structures we extract
1712 : * from it.
1713 : */
1714 3510 : pathnode->in_operators = copyObject(sjinfo->semi_operators);
1715 3510 : pathnode->uniq_exprs = copyObject(sjinfo->semi_rhs_exprs);
1716 :
1717 : /*
1718 : * If the input is a relation and it has a unique index that proves the
1719 : * semi_rhs_exprs are unique, then we don't need to do anything. Note
1720 : * that relation_has_unique_index_for automatically considers restriction
1721 : * clauses for the rel, as well.
1722 : */
1723 4396 : if (rel->rtekind == RTE_RELATION && sjinfo->semi_can_btree &&
1724 886 : relation_has_unique_index_for(root, rel, NIL,
1725 : sjinfo->semi_rhs_exprs,
1726 : sjinfo->semi_operators))
1727 : {
1728 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1729 0 : pathnode->path.rows = rel->rows;
1730 0 : pathnode->path.startup_cost = subpath->startup_cost;
1731 0 : pathnode->path.total_cost = subpath->total_cost;
1732 0 : pathnode->path.pathkeys = subpath->pathkeys;
1733 :
1734 0 : rel->cheapest_unique_path = (Path *) pathnode;
1735 :
1736 0 : MemoryContextSwitchTo(oldcontext);
1737 :
1738 0 : return pathnode;
1739 : }
1740 :
1741 : /*
1742 : * If the input is a subquery whose output must be unique already, then we
1743 : * don't need to do anything. The test for uniqueness has to consider
1744 : * exactly which columns we are extracting; for example "SELECT DISTINCT
1745 : * x,y" doesn't guarantee that x alone is distinct. So we cannot check for
1746 : * this optimization unless semi_rhs_exprs consists only of simple Vars
1747 : * referencing subquery outputs. (Possibly we could do something with
1748 : * expressions in the subquery outputs, too, but for now keep it simple.)
1749 : */
1750 3510 : if (rel->rtekind == RTE_SUBQUERY)
1751 : {
1752 646 : RangeTblEntry *rte = planner_rt_fetch(rel->relid, root);
1753 :
1754 646 : if (query_supports_distinctness(rte->subquery))
1755 : {
1756 : List *sub_tlist_colnos;
1757 :
1758 610 : sub_tlist_colnos = translate_sub_tlist(sjinfo->semi_rhs_exprs,
1759 610 : rel->relid);
1760 :
1761 732 : if (sub_tlist_colnos &&
1762 122 : query_is_distinct_for(rte->subquery,
1763 : sub_tlist_colnos,
1764 : sjinfo->semi_operators))
1765 : {
1766 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1767 0 : pathnode->path.rows = rel->rows;
1768 0 : pathnode->path.startup_cost = subpath->startup_cost;
1769 0 : pathnode->path.total_cost = subpath->total_cost;
1770 0 : pathnode->path.pathkeys = subpath->pathkeys;
1771 :
1772 0 : rel->cheapest_unique_path = (Path *) pathnode;
1773 :
1774 0 : MemoryContextSwitchTo(oldcontext);
1775 :
1776 0 : return pathnode;
1777 : }
1778 : }
1779 : }
1780 :
1781 : /* Estimate number of output rows */
1782 3510 : pathnode->path.rows = estimate_num_groups(root,
1783 : sjinfo->semi_rhs_exprs,
1784 : rel->rows,
1785 : NULL,
1786 : NULL);
1787 3510 : numCols = list_length(sjinfo->semi_rhs_exprs);
1788 :
1789 3510 : if (sjinfo->semi_can_btree)
1790 : {
1791 : /*
1792 : * Estimate cost for sort+unique implementation
1793 : */
1794 3510 : cost_sort(&sort_path, root, NIL,
1795 : subpath->total_cost,
1796 : rel->rows,
1797 3510 : subpath->pathtarget->width,
1798 : 0.0,
1799 : work_mem,
1800 : -1.0);
1801 :
1802 : /*
1803 : * Charge one cpu_operator_cost per comparison per input tuple. We
1804 : * assume all columns get compared at most of the tuples. (XXX
1805 : * probably this is an overestimate.) This should agree with
1806 : * create_upper_unique_path.
1807 : */
1808 3510 : sort_path.total_cost += cpu_operator_cost * rel->rows * numCols;
1809 : }
1810 :
1811 3510 : if (sjinfo->semi_can_hash)
1812 : {
1813 : /*
1814 : * Estimate the overhead per hashtable entry at 64 bytes (same as in
1815 : * planner.c).
1816 : */
1817 3510 : int hashentrysize = subpath->pathtarget->width + 64;
1818 :
1819 3510 : if (hashentrysize * pathnode->path.rows > get_hash_memory_limit())
1820 : {
1821 : /*
1822 : * We should not try to hash. Hack the SpecialJoinInfo to
1823 : * remember this, in case we come through here again.
1824 : */
1825 0 : sjinfo->semi_can_hash = false;
1826 : }
1827 : else
1828 3510 : cost_agg(&agg_path, root,
1829 : AGG_HASHED, NULL,
1830 : numCols, pathnode->path.rows,
1831 : NIL,
1832 : subpath->startup_cost,
1833 : subpath->total_cost,
1834 : rel->rows,
1835 3510 : subpath->pathtarget->width);
1836 : }
1837 :
1838 3510 : if (sjinfo->semi_can_btree && sjinfo->semi_can_hash)
1839 : {
1840 3510 : if (agg_path.total_cost < sort_path.total_cost)
1841 3376 : pathnode->umethod = UNIQUE_PATH_HASH;
1842 : else
1843 134 : pathnode->umethod = UNIQUE_PATH_SORT;
1844 : }
1845 0 : else if (sjinfo->semi_can_btree)
1846 0 : pathnode->umethod = UNIQUE_PATH_SORT;
1847 0 : else if (sjinfo->semi_can_hash)
1848 0 : pathnode->umethod = UNIQUE_PATH_HASH;
1849 : else
1850 : {
1851 : /* we can get here only if we abandoned hashing above */
1852 0 : MemoryContextSwitchTo(oldcontext);
1853 0 : return NULL;
1854 : }
1855 :
1856 3510 : if (pathnode->umethod == UNIQUE_PATH_HASH)
1857 : {
1858 3376 : pathnode->path.startup_cost = agg_path.startup_cost;
1859 3376 : pathnode->path.total_cost = agg_path.total_cost;
1860 : }
1861 : else
1862 : {
1863 134 : pathnode->path.startup_cost = sort_path.startup_cost;
1864 134 : pathnode->path.total_cost = sort_path.total_cost;
1865 : }
1866 :
1867 3510 : rel->cheapest_unique_path = (Path *) pathnode;
1868 :
1869 3510 : MemoryContextSwitchTo(oldcontext);
1870 :
1871 3510 : return pathnode;
1872 : }
1873 :
1874 : /*
1875 : * create_gather_merge_path
1876 : *
1877 : * Creates a path corresponding to a gather merge scan, returning
1878 : * the pathnode.
1879 : */
1880 : GatherMergePath *
1881 9798 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1882 : PathTarget *target, List *pathkeys,
1883 : Relids required_outer, double *rows)
1884 : {
1885 9798 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1886 9798 : Cost input_startup_cost = 0;
1887 9798 : Cost input_total_cost = 0;
1888 :
1889 : Assert(subpath->parallel_safe);
1890 : Assert(pathkeys);
1891 :
1892 9798 : pathnode->path.pathtype = T_GatherMerge;
1893 9798 : pathnode->path.parent = rel;
1894 9798 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1895 : required_outer);
1896 9798 : pathnode->path.parallel_aware = false;
1897 :
1898 9798 : pathnode->subpath = subpath;
1899 9798 : pathnode->num_workers = subpath->parallel_workers;
1900 9798 : pathnode->path.pathkeys = pathkeys;
1901 9798 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1902 9798 : pathnode->path.rows += subpath->rows;
1903 :
1904 9798 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1905 : {
1906 : /* Subpath is adequately ordered, we won't need to sort it */
1907 9798 : input_startup_cost += subpath->startup_cost;
1908 9798 : input_total_cost += subpath->total_cost;
1909 : }
1910 : else
1911 : {
1912 : /* We'll need to insert a Sort node, so include cost for that */
1913 : Path sort_path; /* dummy for result of cost_sort */
1914 :
1915 0 : cost_sort(&sort_path,
1916 : root,
1917 : pathkeys,
1918 : subpath->total_cost,
1919 : subpath->rows,
1920 0 : subpath->pathtarget->width,
1921 : 0.0,
1922 : work_mem,
1923 : -1);
1924 0 : input_startup_cost += sort_path.startup_cost;
1925 0 : input_total_cost += sort_path.total_cost;
1926 : }
1927 :
1928 9798 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1929 : input_startup_cost, input_total_cost, rows);
1930 :
1931 9798 : return pathnode;
1932 : }
1933 :
1934 : /*
1935 : * translate_sub_tlist - get subquery column numbers represented by tlist
1936 : *
1937 : * The given targetlist usually contains only Vars referencing the given relid.
1938 : * Extract their varattnos (ie, the column numbers of the subquery) and return
1939 : * as an integer List.
1940 : *
1941 : * If any of the tlist items is not a simple Var, we cannot determine whether
1942 : * the subquery's uniqueness condition (if any) matches ours, so punt and
1943 : * return NIL.
1944 : */
1945 : static List *
1946 610 : translate_sub_tlist(List *tlist, int relid)
1947 : {
1948 610 : List *result = NIL;
1949 : ListCell *l;
1950 :
1951 732 : foreach(l, tlist)
1952 : {
1953 610 : Var *var = (Var *) lfirst(l);
1954 :
1955 610 : if (!var || !IsA(var, Var) ||
1956 122 : var->varno != relid)
1957 488 : return NIL; /* punt */
1958 :
1959 122 : result = lappend_int(result, var->varattno);
1960 : }
1961 122 : return result;
1962 : }
1963 :
1964 : /*
1965 : * create_gather_path
1966 : * Creates a path corresponding to a gather scan, returning the
1967 : * pathnode.
1968 : *
1969 : * 'rows' may optionally be set to override row estimates from other sources.
1970 : */
1971 : GatherPath *
1972 15932 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1973 : PathTarget *target, Relids required_outer, double *rows)
1974 : {
1975 15932 : GatherPath *pathnode = makeNode(GatherPath);
1976 :
1977 : Assert(subpath->parallel_safe);
1978 :
1979 15932 : pathnode->path.pathtype = T_Gather;
1980 15932 : pathnode->path.parent = rel;
1981 15932 : pathnode->path.pathtarget = target;
1982 15932 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1983 : required_outer);
1984 15932 : pathnode->path.parallel_aware = false;
1985 15932 : pathnode->path.parallel_safe = false;
1986 15932 : pathnode->path.parallel_workers = 0;
1987 15932 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1988 :
1989 15932 : pathnode->subpath = subpath;
1990 15932 : pathnode->num_workers = subpath->parallel_workers;
1991 15932 : pathnode->single_copy = false;
1992 :
1993 15932 : if (pathnode->num_workers == 0)
1994 : {
1995 0 : pathnode->path.pathkeys = subpath->pathkeys;
1996 0 : pathnode->num_workers = 1;
1997 0 : pathnode->single_copy = true;
1998 : }
1999 :
2000 15932 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
2001 :
2002 15932 : return pathnode;
2003 : }
2004 :
2005 : /*
2006 : * create_subqueryscan_path
2007 : * Creates a path corresponding to a scan of a subquery,
2008 : * returning the pathnode.
2009 : *
2010 : * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
2011 : * be trivial, ie just a fetch of all the subquery output columns in order.
2012 : * While we could determine that here, the caller can usually do it more
2013 : * efficiently (or at least amortize it over multiple calls).
2014 : */
2015 : SubqueryScanPath *
2016 38754 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
2017 : bool trivial_pathtarget,
2018 : List *pathkeys, Relids required_outer)
2019 : {
2020 38754 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
2021 :
2022 38754 : pathnode->path.pathtype = T_SubqueryScan;
2023 38754 : pathnode->path.parent = rel;
2024 38754 : pathnode->path.pathtarget = rel->reltarget;
2025 38754 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2026 : required_outer);
2027 38754 : pathnode->path.parallel_aware = false;
2028 63822 : pathnode->path.parallel_safe = rel->consider_parallel &&
2029 25068 : subpath->parallel_safe;
2030 38754 : pathnode->path.parallel_workers = subpath->parallel_workers;
2031 38754 : pathnode->path.pathkeys = pathkeys;
2032 38754 : pathnode->subpath = subpath;
2033 :
2034 38754 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
2035 : trivial_pathtarget);
2036 :
2037 38754 : return pathnode;
2038 : }
2039 :
2040 : /*
2041 : * create_functionscan_path
2042 : * Creates a path corresponding to a sequential scan of a function,
2043 : * returning the pathnode.
2044 : */
2045 : Path *
2046 39782 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
2047 : List *pathkeys, Relids required_outer)
2048 : {
2049 39782 : Path *pathnode = makeNode(Path);
2050 :
2051 39782 : pathnode->pathtype = T_FunctionScan;
2052 39782 : pathnode->parent = rel;
2053 39782 : pathnode->pathtarget = rel->reltarget;
2054 39782 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2055 : required_outer);
2056 39782 : pathnode->parallel_aware = false;
2057 39782 : pathnode->parallel_safe = rel->consider_parallel;
2058 39782 : pathnode->parallel_workers = 0;
2059 39782 : pathnode->pathkeys = pathkeys;
2060 :
2061 39782 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
2062 :
2063 39782 : return pathnode;
2064 : }
2065 :
2066 : /*
2067 : * create_tablefuncscan_path
2068 : * Creates a path corresponding to a sequential scan of a table function,
2069 : * returning the pathnode.
2070 : */
2071 : Path *
2072 222 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
2073 : Relids required_outer)
2074 : {
2075 222 : Path *pathnode = makeNode(Path);
2076 :
2077 222 : pathnode->pathtype = T_TableFuncScan;
2078 222 : pathnode->parent = rel;
2079 222 : pathnode->pathtarget = rel->reltarget;
2080 222 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2081 : required_outer);
2082 222 : pathnode->parallel_aware = false;
2083 222 : pathnode->parallel_safe = rel->consider_parallel;
2084 222 : pathnode->parallel_workers = 0;
2085 222 : pathnode->pathkeys = NIL; /* result is always unordered */
2086 :
2087 222 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
2088 :
2089 222 : return pathnode;
2090 : }
2091 :
2092 : /*
2093 : * create_valuesscan_path
2094 : * Creates a path corresponding to a scan of a VALUES list,
2095 : * returning the pathnode.
2096 : */
2097 : Path *
2098 7708 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
2099 : Relids required_outer)
2100 : {
2101 7708 : Path *pathnode = makeNode(Path);
2102 :
2103 7708 : pathnode->pathtype = T_ValuesScan;
2104 7708 : pathnode->parent = rel;
2105 7708 : pathnode->pathtarget = rel->reltarget;
2106 7708 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2107 : required_outer);
2108 7708 : pathnode->parallel_aware = false;
2109 7708 : pathnode->parallel_safe = rel->consider_parallel;
2110 7708 : pathnode->parallel_workers = 0;
2111 7708 : pathnode->pathkeys = NIL; /* result is always unordered */
2112 :
2113 7708 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
2114 :
2115 7708 : return pathnode;
2116 : }
2117 :
2118 : /*
2119 : * create_ctescan_path
2120 : * Creates a path corresponding to a scan of a non-self-reference CTE,
2121 : * returning the pathnode.
2122 : */
2123 : Path *
2124 3210 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel,
2125 : List *pathkeys, Relids required_outer)
2126 : {
2127 3210 : Path *pathnode = makeNode(Path);
2128 :
2129 3210 : pathnode->pathtype = T_CteScan;
2130 3210 : pathnode->parent = rel;
2131 3210 : pathnode->pathtarget = rel->reltarget;
2132 3210 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2133 : required_outer);
2134 3210 : pathnode->parallel_aware = false;
2135 3210 : pathnode->parallel_safe = rel->consider_parallel;
2136 3210 : pathnode->parallel_workers = 0;
2137 3210 : pathnode->pathkeys = pathkeys;
2138 :
2139 3210 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2140 :
2141 3210 : return pathnode;
2142 : }
2143 :
2144 : /*
2145 : * create_namedtuplestorescan_path
2146 : * Creates a path corresponding to a scan of a named tuplestore, returning
2147 : * the pathnode.
2148 : */
2149 : Path *
2150 442 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
2151 : Relids required_outer)
2152 : {
2153 442 : Path *pathnode = makeNode(Path);
2154 :
2155 442 : pathnode->pathtype = T_NamedTuplestoreScan;
2156 442 : pathnode->parent = rel;
2157 442 : pathnode->pathtarget = rel->reltarget;
2158 442 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2159 : required_outer);
2160 442 : pathnode->parallel_aware = false;
2161 442 : pathnode->parallel_safe = rel->consider_parallel;
2162 442 : pathnode->parallel_workers = 0;
2163 442 : pathnode->pathkeys = NIL; /* result is always unordered */
2164 :
2165 442 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
2166 :
2167 442 : return pathnode;
2168 : }
2169 :
2170 : /*
2171 : * create_resultscan_path
2172 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2173 : * returning the pathnode.
2174 : */
2175 : Path *
2176 1610 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2177 : Relids required_outer)
2178 : {
2179 1610 : Path *pathnode = makeNode(Path);
2180 :
2181 1610 : pathnode->pathtype = T_Result;
2182 1610 : pathnode->parent = rel;
2183 1610 : pathnode->pathtarget = rel->reltarget;
2184 1610 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2185 : required_outer);
2186 1610 : pathnode->parallel_aware = false;
2187 1610 : pathnode->parallel_safe = rel->consider_parallel;
2188 1610 : pathnode->parallel_workers = 0;
2189 1610 : pathnode->pathkeys = NIL; /* result is always unordered */
2190 :
2191 1610 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2192 :
2193 1610 : return pathnode;
2194 : }
2195 :
2196 : /*
2197 : * create_worktablescan_path
2198 : * Creates a path corresponding to a scan of a self-reference CTE,
2199 : * returning the pathnode.
2200 : */
2201 : Path *
2202 808 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2203 : Relids required_outer)
2204 : {
2205 808 : Path *pathnode = makeNode(Path);
2206 :
2207 808 : pathnode->pathtype = T_WorkTableScan;
2208 808 : pathnode->parent = rel;
2209 808 : pathnode->pathtarget = rel->reltarget;
2210 808 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2211 : required_outer);
2212 808 : pathnode->parallel_aware = false;
2213 808 : pathnode->parallel_safe = rel->consider_parallel;
2214 808 : pathnode->parallel_workers = 0;
2215 808 : pathnode->pathkeys = NIL; /* result is always unordered */
2216 :
2217 : /* Cost is the same as for a regular CTE scan */
2218 808 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2219 :
2220 808 : return pathnode;
2221 : }
2222 :
2223 : /*
2224 : * create_foreignscan_path
2225 : * Creates a path corresponding to a scan of a foreign base table,
2226 : * returning the pathnode.
2227 : *
2228 : * This function is never called from core Postgres; rather, it's expected
2229 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2230 : * We make the FDW supply all fields of the path, since we do not have any way
2231 : * to calculate them in core. However, there is a usually-sane default for
2232 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2233 : */
2234 : ForeignPath *
2235 3542 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2236 : PathTarget *target,
2237 : double rows, Cost startup_cost, Cost total_cost,
2238 : List *pathkeys,
2239 : Relids required_outer,
2240 : Path *fdw_outerpath,
2241 : List *fdw_restrictinfo,
2242 : List *fdw_private)
2243 : {
2244 3542 : ForeignPath *pathnode = makeNode(ForeignPath);
2245 :
2246 : /* Historically some FDWs were confused about when to use this */
2247 : Assert(IS_SIMPLE_REL(rel));
2248 :
2249 3542 : pathnode->path.pathtype = T_ForeignScan;
2250 3542 : pathnode->path.parent = rel;
2251 3542 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2252 3542 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2253 : required_outer);
2254 3542 : pathnode->path.parallel_aware = false;
2255 3542 : pathnode->path.parallel_safe = rel->consider_parallel;
2256 3542 : pathnode->path.parallel_workers = 0;
2257 3542 : pathnode->path.rows = rows;
2258 3542 : pathnode->path.startup_cost = startup_cost;
2259 3542 : pathnode->path.total_cost = total_cost;
2260 3542 : pathnode->path.pathkeys = pathkeys;
2261 :
2262 3542 : pathnode->fdw_outerpath = fdw_outerpath;
2263 3542 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2264 3542 : pathnode->fdw_private = fdw_private;
2265 :
2266 3542 : return pathnode;
2267 : }
2268 :
2269 : /*
2270 : * create_foreign_join_path
2271 : * Creates a path corresponding to a scan of a foreign join,
2272 : * returning the pathnode.
2273 : *
2274 : * This function is never called from core Postgres; rather, it's expected
2275 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2276 : * We make the FDW supply all fields of the path, since we do not have any way
2277 : * to calculate them in core. However, there is a usually-sane default for
2278 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2279 : */
2280 : ForeignPath *
2281 1116 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2282 : PathTarget *target,
2283 : double rows, Cost startup_cost, Cost total_cost,
2284 : List *pathkeys,
2285 : Relids required_outer,
2286 : Path *fdw_outerpath,
2287 : List *fdw_restrictinfo,
2288 : List *fdw_private)
2289 : {
2290 1116 : ForeignPath *pathnode = makeNode(ForeignPath);
2291 :
2292 : /*
2293 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2294 : * but the API of this function doesn't support it, and existing
2295 : * extensions aren't yet trying to build such paths anyway. For the
2296 : * moment just throw an error if someone tries it; eventually we should
2297 : * revisit this.
2298 : */
2299 1116 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2300 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2301 :
2302 1116 : pathnode->path.pathtype = T_ForeignScan;
2303 1116 : pathnode->path.parent = rel;
2304 1116 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2305 1116 : pathnode->path.param_info = NULL; /* XXX see above */
2306 1116 : pathnode->path.parallel_aware = false;
2307 1116 : pathnode->path.parallel_safe = rel->consider_parallel;
2308 1116 : pathnode->path.parallel_workers = 0;
2309 1116 : pathnode->path.rows = rows;
2310 1116 : pathnode->path.startup_cost = startup_cost;
2311 1116 : pathnode->path.total_cost = total_cost;
2312 1116 : pathnode->path.pathkeys = pathkeys;
2313 :
2314 1116 : pathnode->fdw_outerpath = fdw_outerpath;
2315 1116 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2316 1116 : pathnode->fdw_private = fdw_private;
2317 :
2318 1116 : return pathnode;
2319 : }
2320 :
2321 : /*
2322 : * create_foreign_upper_path
2323 : * Creates a path corresponding to an upper relation that's computed
2324 : * directly by an FDW, returning the pathnode.
2325 : *
2326 : * This function is never called from core Postgres; rather, it's expected to
2327 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2328 : * We make the FDW supply all fields of the path, since we do not have any way
2329 : * to calculate them in core. However, there is a usually-sane default for
2330 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2331 : */
2332 : ForeignPath *
2333 570 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2334 : PathTarget *target,
2335 : double rows, Cost startup_cost, Cost total_cost,
2336 : List *pathkeys,
2337 : Path *fdw_outerpath,
2338 : List *fdw_restrictinfo,
2339 : List *fdw_private)
2340 : {
2341 570 : ForeignPath *pathnode = makeNode(ForeignPath);
2342 :
2343 : /*
2344 : * Upper relations should never have any lateral references, since joining
2345 : * is complete.
2346 : */
2347 : Assert(bms_is_empty(rel->lateral_relids));
2348 :
2349 570 : pathnode->path.pathtype = T_ForeignScan;
2350 570 : pathnode->path.parent = rel;
2351 570 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2352 570 : pathnode->path.param_info = NULL;
2353 570 : pathnode->path.parallel_aware = false;
2354 570 : pathnode->path.parallel_safe = rel->consider_parallel;
2355 570 : pathnode->path.parallel_workers = 0;
2356 570 : pathnode->path.rows = rows;
2357 570 : pathnode->path.startup_cost = startup_cost;
2358 570 : pathnode->path.total_cost = total_cost;
2359 570 : pathnode->path.pathkeys = pathkeys;
2360 :
2361 570 : pathnode->fdw_outerpath = fdw_outerpath;
2362 570 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2363 570 : pathnode->fdw_private = fdw_private;
2364 :
2365 570 : return pathnode;
2366 : }
2367 :
2368 : /*
2369 : * calc_nestloop_required_outer
2370 : * Compute the required_outer set for a nestloop join path
2371 : *
2372 : * Note: when considering a child join, the inputs nonetheless use top-level
2373 : * parent relids
2374 : *
2375 : * Note: result must not share storage with either input
2376 : */
2377 : Relids
2378 2437186 : calc_nestloop_required_outer(Relids outerrelids,
2379 : Relids outer_paramrels,
2380 : Relids innerrelids,
2381 : Relids inner_paramrels)
2382 : {
2383 : Relids required_outer;
2384 :
2385 : /* inner_path can require rels from outer path, but not vice versa */
2386 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2387 : /* easy case if inner path is not parameterized */
2388 2437186 : if (!inner_paramrels)
2389 1648630 : return bms_copy(outer_paramrels);
2390 : /* else, form the union ... */
2391 788556 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2392 : /* ... and remove any mention of now-satisfied outer rels */
2393 788556 : required_outer = bms_del_members(required_outer,
2394 : outerrelids);
2395 788556 : return required_outer;
2396 : }
2397 :
2398 : /*
2399 : * calc_non_nestloop_required_outer
2400 : * Compute the required_outer set for a merge or hash join path
2401 : *
2402 : * Note: result must not share storage with either input
2403 : */
2404 : Relids
2405 1604740 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2406 : {
2407 1604740 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2408 1604740 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2409 : Relids innerrelids PG_USED_FOR_ASSERTS_ONLY;
2410 : Relids outerrelids PG_USED_FOR_ASSERTS_ONLY;
2411 : Relids required_outer;
2412 :
2413 : /*
2414 : * Any parameterization of the input paths refers to topmost parents of
2415 : * the relevant relations, because reparameterize_path_by_child() hasn't
2416 : * been called yet. So we must consider topmost parents of the relations
2417 : * being joined, too, while checking for disallowed parameterization
2418 : * cases.
2419 : */
2420 1604740 : if (inner_path->parent->top_parent_relids)
2421 34436 : innerrelids = inner_path->parent->top_parent_relids;
2422 : else
2423 1570304 : innerrelids = inner_path->parent->relids;
2424 :
2425 1604740 : if (outer_path->parent->top_parent_relids)
2426 34436 : outerrelids = outer_path->parent->top_parent_relids;
2427 : else
2428 1570304 : outerrelids = outer_path->parent->relids;
2429 :
2430 : /* neither path can require rels from the other */
2431 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2432 : Assert(!bms_overlap(inner_paramrels, outerrelids));
2433 : /* form the union ... */
2434 1604740 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2435 : /* we do not need an explicit test for empty; bms_union gets it right */
2436 1604740 : return required_outer;
2437 : }
2438 :
2439 : /*
2440 : * create_nestloop_path
2441 : * Creates a pathnode corresponding to a nestloop join between two
2442 : * relations.
2443 : *
2444 : * 'joinrel' is the join relation.
2445 : * 'jointype' is the type of join required
2446 : * 'workspace' is the result from initial_cost_nestloop
2447 : * 'extra' contains various information about the join
2448 : * 'outer_path' is the outer path
2449 : * 'inner_path' is the inner path
2450 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2451 : * 'pathkeys' are the path keys of the new join path
2452 : * 'required_outer' is the set of required outer rels
2453 : *
2454 : * Returns the resulting path node.
2455 : */
2456 : NestPath *
2457 1122872 : create_nestloop_path(PlannerInfo *root,
2458 : RelOptInfo *joinrel,
2459 : JoinType jointype,
2460 : JoinCostWorkspace *workspace,
2461 : JoinPathExtraData *extra,
2462 : Path *outer_path,
2463 : Path *inner_path,
2464 : List *restrict_clauses,
2465 : List *pathkeys,
2466 : Relids required_outer)
2467 : {
2468 1122872 : NestPath *pathnode = makeNode(NestPath);
2469 1122872 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2470 : Relids outerrelids;
2471 :
2472 : /*
2473 : * Paths are parameterized by top-level parents, so run parameterization
2474 : * tests on the parent relids.
2475 : */
2476 1122872 : if (outer_path->parent->top_parent_relids)
2477 15346 : outerrelids = outer_path->parent->top_parent_relids;
2478 : else
2479 1107526 : outerrelids = outer_path->parent->relids;
2480 :
2481 : /*
2482 : * If the inner path is parameterized by the outer, we must drop any
2483 : * restrict_clauses that are due to be moved into the inner path. We have
2484 : * to do this now, rather than postpone the work till createplan time,
2485 : * because the restrict_clauses list can affect the size and cost
2486 : * estimates for this path. We detect such clauses by checking for serial
2487 : * number match to clauses already enforced in the inner path.
2488 : */
2489 1122872 : if (bms_overlap(inner_req_outer, outerrelids))
2490 : {
2491 315418 : Bitmapset *enforced_serials = get_param_path_clause_serials(inner_path);
2492 315418 : List *jclauses = NIL;
2493 : ListCell *lc;
2494 :
2495 686738 : foreach(lc, restrict_clauses)
2496 : {
2497 371320 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2498 :
2499 371320 : if (!bms_is_member(rinfo->rinfo_serial, enforced_serials))
2500 47724 : jclauses = lappend(jclauses, rinfo);
2501 : }
2502 315418 : restrict_clauses = jclauses;
2503 : }
2504 :
2505 1122872 : pathnode->jpath.path.pathtype = T_NestLoop;
2506 1122872 : pathnode->jpath.path.parent = joinrel;
2507 1122872 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2508 1122872 : pathnode->jpath.path.param_info =
2509 1122872 : get_joinrel_parampathinfo(root,
2510 : joinrel,
2511 : outer_path,
2512 : inner_path,
2513 : extra->sjinfo,
2514 : required_outer,
2515 : &restrict_clauses);
2516 1122872 : pathnode->jpath.path.parallel_aware = false;
2517 3270192 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2518 1122872 : outer_path->parallel_safe && inner_path->parallel_safe;
2519 : /* This is a foolish way to estimate parallel_workers, but for now... */
2520 1122872 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2521 1122872 : pathnode->jpath.path.pathkeys = pathkeys;
2522 1122872 : pathnode->jpath.jointype = jointype;
2523 1122872 : pathnode->jpath.inner_unique = extra->inner_unique;
2524 1122872 : pathnode->jpath.outerjoinpath = outer_path;
2525 1122872 : pathnode->jpath.innerjoinpath = inner_path;
2526 1122872 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2527 :
2528 1122872 : final_cost_nestloop(root, pathnode, workspace, extra);
2529 :
2530 1122872 : return pathnode;
2531 : }
2532 :
2533 : /*
2534 : * create_mergejoin_path
2535 : * Creates a pathnode corresponding to a mergejoin join between
2536 : * two relations
2537 : *
2538 : * 'joinrel' is the join relation
2539 : * 'jointype' is the type of join required
2540 : * 'workspace' is the result from initial_cost_mergejoin
2541 : * 'extra' contains various information about the join
2542 : * 'outer_path' is the outer path
2543 : * 'inner_path' is the inner path
2544 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2545 : * 'pathkeys' are the path keys of the new join path
2546 : * 'required_outer' is the set of required outer rels
2547 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2548 : * (this should be a subset of the restrict_clauses list)
2549 : * 'outersortkeys' are the sort varkeys for the outer relation
2550 : * 'innersortkeys' are the sort varkeys for the inner relation
2551 : */
2552 : MergePath *
2553 250804 : create_mergejoin_path(PlannerInfo *root,
2554 : RelOptInfo *joinrel,
2555 : JoinType jointype,
2556 : JoinCostWorkspace *workspace,
2557 : JoinPathExtraData *extra,
2558 : Path *outer_path,
2559 : Path *inner_path,
2560 : List *restrict_clauses,
2561 : List *pathkeys,
2562 : Relids required_outer,
2563 : List *mergeclauses,
2564 : List *outersortkeys,
2565 : List *innersortkeys)
2566 : {
2567 250804 : MergePath *pathnode = makeNode(MergePath);
2568 :
2569 250804 : pathnode->jpath.path.pathtype = T_MergeJoin;
2570 250804 : pathnode->jpath.path.parent = joinrel;
2571 250804 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2572 250804 : pathnode->jpath.path.param_info =
2573 250804 : get_joinrel_parampathinfo(root,
2574 : joinrel,
2575 : outer_path,
2576 : inner_path,
2577 : extra->sjinfo,
2578 : required_outer,
2579 : &restrict_clauses);
2580 250804 : pathnode->jpath.path.parallel_aware = false;
2581 723820 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2582 250804 : outer_path->parallel_safe && inner_path->parallel_safe;
2583 : /* This is a foolish way to estimate parallel_workers, but for now... */
2584 250804 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2585 250804 : pathnode->jpath.path.pathkeys = pathkeys;
2586 250804 : pathnode->jpath.jointype = jointype;
2587 250804 : pathnode->jpath.inner_unique = extra->inner_unique;
2588 250804 : pathnode->jpath.outerjoinpath = outer_path;
2589 250804 : pathnode->jpath.innerjoinpath = inner_path;
2590 250804 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2591 250804 : pathnode->path_mergeclauses = mergeclauses;
2592 250804 : pathnode->outersortkeys = outersortkeys;
2593 250804 : pathnode->innersortkeys = innersortkeys;
2594 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2595 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2596 :
2597 250804 : final_cost_mergejoin(root, pathnode, workspace, extra);
2598 :
2599 250804 : return pathnode;
2600 : }
2601 :
2602 : /*
2603 : * create_hashjoin_path
2604 : * Creates a pathnode corresponding to a hash join between two relations.
2605 : *
2606 : * 'joinrel' is the join relation
2607 : * 'jointype' is the type of join required
2608 : * 'workspace' is the result from initial_cost_hashjoin
2609 : * 'extra' contains various information about the join
2610 : * 'outer_path' is the cheapest outer path
2611 : * 'inner_path' is the cheapest inner path
2612 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2613 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2614 : * 'required_outer' is the set of required outer rels
2615 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2616 : * (this should be a subset of the restrict_clauses list)
2617 : */
2618 : HashPath *
2619 219072 : create_hashjoin_path(PlannerInfo *root,
2620 : RelOptInfo *joinrel,
2621 : JoinType jointype,
2622 : JoinCostWorkspace *workspace,
2623 : JoinPathExtraData *extra,
2624 : Path *outer_path,
2625 : Path *inner_path,
2626 : bool parallel_hash,
2627 : List *restrict_clauses,
2628 : Relids required_outer,
2629 : List *hashclauses)
2630 : {
2631 219072 : HashPath *pathnode = makeNode(HashPath);
2632 :
2633 219072 : pathnode->jpath.path.pathtype = T_HashJoin;
2634 219072 : pathnode->jpath.path.parent = joinrel;
2635 219072 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2636 219072 : pathnode->jpath.path.param_info =
2637 219072 : get_joinrel_parampathinfo(root,
2638 : joinrel,
2639 : outer_path,
2640 : inner_path,
2641 : extra->sjinfo,
2642 : required_outer,
2643 : &restrict_clauses);
2644 219072 : pathnode->jpath.path.parallel_aware =
2645 219072 : joinrel->consider_parallel && parallel_hash;
2646 630064 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2647 219072 : outer_path->parallel_safe && inner_path->parallel_safe;
2648 : /* This is a foolish way to estimate parallel_workers, but for now... */
2649 219072 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2650 :
2651 : /*
2652 : * A hashjoin never has pathkeys, since its output ordering is
2653 : * unpredictable due to possible batching. XXX If the inner relation is
2654 : * small enough, we could instruct the executor that it must not batch,
2655 : * and then we could assume that the output inherits the outer relation's
2656 : * ordering, which might save a sort step. However there is considerable
2657 : * downside if our estimate of the inner relation size is badly off. For
2658 : * the moment we don't risk it. (Note also that if we wanted to take this
2659 : * seriously, joinpath.c would have to consider many more paths for the
2660 : * outer rel than it does now.)
2661 : */
2662 219072 : pathnode->jpath.path.pathkeys = NIL;
2663 219072 : pathnode->jpath.jointype = jointype;
2664 219072 : pathnode->jpath.inner_unique = extra->inner_unique;
2665 219072 : pathnode->jpath.outerjoinpath = outer_path;
2666 219072 : pathnode->jpath.innerjoinpath = inner_path;
2667 219072 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2668 219072 : pathnode->path_hashclauses = hashclauses;
2669 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2670 :
2671 219072 : final_cost_hashjoin(root, pathnode, workspace, extra);
2672 :
2673 219072 : return pathnode;
2674 : }
2675 :
2676 : /*
2677 : * create_projection_path
2678 : * Creates a pathnode that represents performing a projection.
2679 : *
2680 : * 'rel' is the parent relation associated with the result
2681 : * 'subpath' is the path representing the source of data
2682 : * 'target' is the PathTarget to be computed
2683 : */
2684 : ProjectionPath *
2685 367776 : create_projection_path(PlannerInfo *root,
2686 : RelOptInfo *rel,
2687 : Path *subpath,
2688 : PathTarget *target)
2689 : {
2690 367776 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2691 : PathTarget *oldtarget;
2692 :
2693 : /*
2694 : * We mustn't put a ProjectionPath directly above another; it's useless
2695 : * and will confuse create_projection_plan. Rather than making sure all
2696 : * callers handle that, let's implement it here, by stripping off any
2697 : * ProjectionPath in what we're given. Given this rule, there won't be
2698 : * more than one.
2699 : */
2700 367776 : if (IsA(subpath, ProjectionPath))
2701 : {
2702 12 : ProjectionPath *subpp = (ProjectionPath *) subpath;
2703 :
2704 : Assert(subpp->path.parent == rel);
2705 12 : subpath = subpp->subpath;
2706 : Assert(!IsA(subpath, ProjectionPath));
2707 : }
2708 :
2709 367776 : pathnode->path.pathtype = T_Result;
2710 367776 : pathnode->path.parent = rel;
2711 367776 : pathnode->path.pathtarget = target;
2712 : /* For now, assume we are above any joins, so no parameterization */
2713 367776 : pathnode->path.param_info = NULL;
2714 367776 : pathnode->path.parallel_aware = false;
2715 824462 : pathnode->path.parallel_safe = rel->consider_parallel &&
2716 456064 : subpath->parallel_safe &&
2717 88288 : is_parallel_safe(root, (Node *) target->exprs);
2718 367776 : pathnode->path.parallel_workers = subpath->parallel_workers;
2719 : /* Projection does not change the sort order */
2720 367776 : pathnode->path.pathkeys = subpath->pathkeys;
2721 :
2722 367776 : pathnode->subpath = subpath;
2723 :
2724 : /*
2725 : * We might not need a separate Result node. If the input plan node type
2726 : * can project, we can just tell it to project something else. Or, if it
2727 : * can't project but the desired target has the same expression list as
2728 : * what the input will produce anyway, we can still give it the desired
2729 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2730 : * Note: in the latter case, create_projection_plan has to recheck our
2731 : * conclusion; see comments therein.
2732 : */
2733 367776 : oldtarget = subpath->pathtarget;
2734 382660 : if (is_projection_capable_path(subpath) ||
2735 14884 : equal(oldtarget->exprs, target->exprs))
2736 : {
2737 : /* No separate Result node needed */
2738 365724 : pathnode->dummypp = true;
2739 :
2740 : /*
2741 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2742 : */
2743 365724 : pathnode->path.rows = subpath->rows;
2744 365724 : pathnode->path.startup_cost = subpath->startup_cost +
2745 365724 : (target->cost.startup - oldtarget->cost.startup);
2746 365724 : pathnode->path.total_cost = subpath->total_cost +
2747 365724 : (target->cost.startup - oldtarget->cost.startup) +
2748 365724 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2749 : }
2750 : else
2751 : {
2752 : /* We really do need the Result node */
2753 2052 : pathnode->dummypp = false;
2754 :
2755 : /*
2756 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2757 : * evaluating the tlist. There is no qual to worry about.
2758 : */
2759 2052 : pathnode->path.rows = subpath->rows;
2760 2052 : pathnode->path.startup_cost = subpath->startup_cost +
2761 2052 : target->cost.startup;
2762 2052 : pathnode->path.total_cost = subpath->total_cost +
2763 2052 : target->cost.startup +
2764 2052 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2765 : }
2766 :
2767 367776 : return pathnode;
2768 : }
2769 :
2770 : /*
2771 : * apply_projection_to_path
2772 : * Add a projection step, or just apply the target directly to given path.
2773 : *
2774 : * This has the same net effect as create_projection_path(), except that if
2775 : * a separate Result plan node isn't needed, we just replace the given path's
2776 : * pathtarget with the desired one. This must be used only when the caller
2777 : * knows that the given path isn't referenced elsewhere and so can be modified
2778 : * in-place.
2779 : *
2780 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2781 : * new target down to its input as well; this is a yet more invasive
2782 : * modification of the input path, which create_projection_path() can't do.
2783 : *
2784 : * Note that we mustn't change the source path's parent link; so when it is
2785 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2786 : * not caused any trouble.
2787 : *
2788 : * 'rel' is the parent relation associated with the result
2789 : * 'path' is the path representing the source of data
2790 : * 'target' is the PathTarget to be computed
2791 : */
2792 : Path *
2793 28132 : apply_projection_to_path(PlannerInfo *root,
2794 : RelOptInfo *rel,
2795 : Path *path,
2796 : PathTarget *target)
2797 : {
2798 : QualCost oldcost;
2799 :
2800 : /*
2801 : * If given path can't project, we might need a Result node, so make a
2802 : * separate ProjectionPath.
2803 : */
2804 28132 : if (!is_projection_capable_path(path))
2805 13780 : return (Path *) create_projection_path(root, rel, path, target);
2806 :
2807 : /*
2808 : * We can just jam the desired tlist into the existing path, being sure to
2809 : * update its cost estimates appropriately.
2810 : */
2811 14352 : oldcost = path->pathtarget->cost;
2812 14352 : path->pathtarget = target;
2813 :
2814 14352 : path->startup_cost += target->cost.startup - oldcost.startup;
2815 14352 : path->total_cost += target->cost.startup - oldcost.startup +
2816 14352 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2817 :
2818 : /*
2819 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2820 : * arrange for the subpath to return the required target list so that
2821 : * workers can help project. But if there is something that is not
2822 : * parallel-safe in the target expressions, then we can't.
2823 : */
2824 14808 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
2825 456 : is_parallel_safe(root, (Node *) target->exprs))
2826 : {
2827 : /*
2828 : * We always use create_projection_path here, even if the subpath is
2829 : * projection-capable, so as to avoid modifying the subpath in place.
2830 : * It seems unlikely at present that there could be any other
2831 : * references to the subpath, but better safe than sorry.
2832 : *
2833 : * Note that we don't change the parallel path's cost estimates; it
2834 : * might be appropriate to do so, to reflect the fact that the bulk of
2835 : * the target evaluation will happen in workers.
2836 : */
2837 456 : if (IsA(path, GatherPath))
2838 : {
2839 0 : GatherPath *gpath = (GatherPath *) path;
2840 :
2841 0 : gpath->subpath = (Path *)
2842 0 : create_projection_path(root,
2843 0 : gpath->subpath->parent,
2844 : gpath->subpath,
2845 : target);
2846 : }
2847 : else
2848 : {
2849 456 : GatherMergePath *gmpath = (GatherMergePath *) path;
2850 :
2851 456 : gmpath->subpath = (Path *)
2852 456 : create_projection_path(root,
2853 456 : gmpath->subpath->parent,
2854 : gmpath->subpath,
2855 : target);
2856 : }
2857 : }
2858 13896 : else if (path->parallel_safe &&
2859 6002 : !is_parallel_safe(root, (Node *) target->exprs))
2860 : {
2861 : /*
2862 : * We're inserting a parallel-restricted target list into a path
2863 : * currently marked parallel-safe, so we have to mark it as no longer
2864 : * safe.
2865 : */
2866 12 : path->parallel_safe = false;
2867 : }
2868 :
2869 14352 : return path;
2870 : }
2871 :
2872 : /*
2873 : * create_set_projection_path
2874 : * Creates a pathnode that represents performing a projection that
2875 : * includes set-returning functions.
2876 : *
2877 : * 'rel' is the parent relation associated with the result
2878 : * 'subpath' is the path representing the source of data
2879 : * 'target' is the PathTarget to be computed
2880 : */
2881 : ProjectSetPath *
2882 8432 : create_set_projection_path(PlannerInfo *root,
2883 : RelOptInfo *rel,
2884 : Path *subpath,
2885 : PathTarget *target)
2886 : {
2887 8432 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2888 : double tlist_rows;
2889 : ListCell *lc;
2890 :
2891 8432 : pathnode->path.pathtype = T_ProjectSet;
2892 8432 : pathnode->path.parent = rel;
2893 8432 : pathnode->path.pathtarget = target;
2894 : /* For now, assume we are above any joins, so no parameterization */
2895 8432 : pathnode->path.param_info = NULL;
2896 8432 : pathnode->path.parallel_aware = false;
2897 18574 : pathnode->path.parallel_safe = rel->consider_parallel &&
2898 10106 : subpath->parallel_safe &&
2899 1674 : is_parallel_safe(root, (Node *) target->exprs);
2900 8432 : pathnode->path.parallel_workers = subpath->parallel_workers;
2901 : /* Projection does not change the sort order XXX? */
2902 8432 : pathnode->path.pathkeys = subpath->pathkeys;
2903 :
2904 8432 : pathnode->subpath = subpath;
2905 :
2906 : /*
2907 : * Estimate number of rows produced by SRFs for each row of input; if
2908 : * there's more than one in this node, use the maximum.
2909 : */
2910 8432 : tlist_rows = 1;
2911 18728 : foreach(lc, target->exprs)
2912 : {
2913 10296 : Node *node = (Node *) lfirst(lc);
2914 : double itemrows;
2915 :
2916 10296 : itemrows = expression_returns_set_rows(root, node);
2917 10296 : if (tlist_rows < itemrows)
2918 8204 : tlist_rows = itemrows;
2919 : }
2920 :
2921 : /*
2922 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2923 : * per input row, and half of cpu_tuple_cost for each added output row.
2924 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2925 : * this estimate later.
2926 : */
2927 8432 : pathnode->path.rows = subpath->rows * tlist_rows;
2928 8432 : pathnode->path.startup_cost = subpath->startup_cost +
2929 8432 : target->cost.startup;
2930 8432 : pathnode->path.total_cost = subpath->total_cost +
2931 8432 : target->cost.startup +
2932 8432 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2933 8432 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2934 :
2935 8432 : return pathnode;
2936 : }
2937 :
2938 : /*
2939 : * create_incremental_sort_path
2940 : * Creates a pathnode that represents performing an incremental sort.
2941 : *
2942 : * 'rel' is the parent relation associated with the result
2943 : * 'subpath' is the path representing the source of data
2944 : * 'pathkeys' represents the desired sort order
2945 : * 'presorted_keys' is the number of keys by which the input path is
2946 : * already sorted
2947 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2948 : * or -1 if no LIMIT or couldn't estimate
2949 : */
2950 : IncrementalSortPath *
2951 7654 : create_incremental_sort_path(PlannerInfo *root,
2952 : RelOptInfo *rel,
2953 : Path *subpath,
2954 : List *pathkeys,
2955 : int presorted_keys,
2956 : double limit_tuples)
2957 : {
2958 7654 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
2959 7654 : SortPath *pathnode = &sort->spath;
2960 :
2961 7654 : pathnode->path.pathtype = T_IncrementalSort;
2962 7654 : pathnode->path.parent = rel;
2963 : /* Sort doesn't project, so use source path's pathtarget */
2964 7654 : pathnode->path.pathtarget = subpath->pathtarget;
2965 : /* For now, assume we are above any joins, so no parameterization */
2966 7654 : pathnode->path.param_info = NULL;
2967 7654 : pathnode->path.parallel_aware = false;
2968 11116 : pathnode->path.parallel_safe = rel->consider_parallel &&
2969 3462 : subpath->parallel_safe;
2970 7654 : pathnode->path.parallel_workers = subpath->parallel_workers;
2971 7654 : pathnode->path.pathkeys = pathkeys;
2972 :
2973 7654 : pathnode->subpath = subpath;
2974 :
2975 7654 : cost_incremental_sort(&pathnode->path,
2976 : root, pathkeys, presorted_keys,
2977 : subpath->startup_cost,
2978 : subpath->total_cost,
2979 : subpath->rows,
2980 7654 : subpath->pathtarget->width,
2981 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2982 : work_mem, limit_tuples);
2983 :
2984 7654 : sort->nPresortedCols = presorted_keys;
2985 :
2986 7654 : return sort;
2987 : }
2988 :
2989 : /*
2990 : * create_sort_path
2991 : * Creates a pathnode that represents performing an explicit sort.
2992 : *
2993 : * 'rel' is the parent relation associated with the result
2994 : * 'subpath' is the path representing the source of data
2995 : * 'pathkeys' represents the desired sort order
2996 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2997 : * or -1 if no LIMIT or couldn't estimate
2998 : */
2999 : SortPath *
3000 80606 : create_sort_path(PlannerInfo *root,
3001 : RelOptInfo *rel,
3002 : Path *subpath,
3003 : List *pathkeys,
3004 : double limit_tuples)
3005 : {
3006 80606 : SortPath *pathnode = makeNode(SortPath);
3007 :
3008 80606 : pathnode->path.pathtype = T_Sort;
3009 80606 : pathnode->path.parent = rel;
3010 : /* Sort doesn't project, so use source path's pathtarget */
3011 80606 : pathnode->path.pathtarget = subpath->pathtarget;
3012 : /* For now, assume we are above any joins, so no parameterization */
3013 80606 : pathnode->path.param_info = NULL;
3014 80606 : pathnode->path.parallel_aware = false;
3015 135902 : pathnode->path.parallel_safe = rel->consider_parallel &&
3016 55296 : subpath->parallel_safe;
3017 80606 : pathnode->path.parallel_workers = subpath->parallel_workers;
3018 80606 : pathnode->path.pathkeys = pathkeys;
3019 :
3020 80606 : pathnode->subpath = subpath;
3021 :
3022 80606 : cost_sort(&pathnode->path, root, pathkeys,
3023 : subpath->total_cost,
3024 : subpath->rows,
3025 80606 : subpath->pathtarget->width,
3026 : 0.0, /* XXX comparison_cost shouldn't be 0? */
3027 : work_mem, limit_tuples);
3028 :
3029 80606 : return pathnode;
3030 : }
3031 :
3032 : /*
3033 : * create_group_path
3034 : * Creates a pathnode that represents performing grouping of presorted input
3035 : *
3036 : * 'rel' is the parent relation associated with the result
3037 : * 'subpath' is the path representing the source of data
3038 : * 'target' is the PathTarget to be computed
3039 : * 'groupClause' is a list of SortGroupClause's representing the grouping
3040 : * 'qual' is the HAVING quals if any
3041 : * 'numGroups' is the estimated number of groups
3042 : */
3043 : GroupPath *
3044 1118 : create_group_path(PlannerInfo *root,
3045 : RelOptInfo *rel,
3046 : Path *subpath,
3047 : List *groupClause,
3048 : List *qual,
3049 : double numGroups)
3050 : {
3051 1118 : GroupPath *pathnode = makeNode(GroupPath);
3052 1118 : PathTarget *target = rel->reltarget;
3053 :
3054 1118 : pathnode->path.pathtype = T_Group;
3055 1118 : pathnode->path.parent = rel;
3056 1118 : pathnode->path.pathtarget = target;
3057 : /* For now, assume we are above any joins, so no parameterization */
3058 1118 : pathnode->path.param_info = NULL;
3059 1118 : pathnode->path.parallel_aware = false;
3060 1844 : pathnode->path.parallel_safe = rel->consider_parallel &&
3061 726 : subpath->parallel_safe;
3062 1118 : pathnode->path.parallel_workers = subpath->parallel_workers;
3063 : /* Group doesn't change sort ordering */
3064 1118 : pathnode->path.pathkeys = subpath->pathkeys;
3065 :
3066 1118 : pathnode->subpath = subpath;
3067 :
3068 1118 : pathnode->groupClause = groupClause;
3069 1118 : pathnode->qual = qual;
3070 :
3071 1118 : cost_group(&pathnode->path, root,
3072 : list_length(groupClause),
3073 : numGroups,
3074 : qual,
3075 : subpath->startup_cost, subpath->total_cost,
3076 : subpath->rows);
3077 :
3078 : /* add tlist eval cost for each output row */
3079 1118 : pathnode->path.startup_cost += target->cost.startup;
3080 1118 : pathnode->path.total_cost += target->cost.startup +
3081 1118 : target->cost.per_tuple * pathnode->path.rows;
3082 :
3083 1118 : return pathnode;
3084 : }
3085 :
3086 : /*
3087 : * create_upper_unique_path
3088 : * Creates a pathnode that represents performing an explicit Unique step
3089 : * on presorted input.
3090 : *
3091 : * This produces a Unique plan node, but the use-case is so different from
3092 : * create_unique_path that it doesn't seem worth trying to merge the two.
3093 : *
3094 : * 'rel' is the parent relation associated with the result
3095 : * 'subpath' is the path representing the source of data
3096 : * 'numCols' is the number of grouping columns
3097 : * 'numGroups' is the estimated number of groups
3098 : *
3099 : * The input path must be sorted on the grouping columns, plus possibly
3100 : * additional columns; so the first numCols pathkeys are the grouping columns
3101 : */
3102 : UpperUniquePath *
3103 6606 : create_upper_unique_path(PlannerInfo *root,
3104 : RelOptInfo *rel,
3105 : Path *subpath,
3106 : int numCols,
3107 : double numGroups)
3108 : {
3109 6606 : UpperUniquePath *pathnode = makeNode(UpperUniquePath);
3110 :
3111 6606 : pathnode->path.pathtype = T_Unique;
3112 6606 : pathnode->path.parent = rel;
3113 : /* Unique doesn't project, so use source path's pathtarget */
3114 6606 : pathnode->path.pathtarget = subpath->pathtarget;
3115 : /* For now, assume we are above any joins, so no parameterization */
3116 6606 : pathnode->path.param_info = NULL;
3117 6606 : pathnode->path.parallel_aware = false;
3118 10118 : pathnode->path.parallel_safe = rel->consider_parallel &&
3119 3512 : subpath->parallel_safe;
3120 6606 : pathnode->path.parallel_workers = subpath->parallel_workers;
3121 : /* Unique doesn't change the input ordering */
3122 6606 : pathnode->path.pathkeys = subpath->pathkeys;
3123 :
3124 6606 : pathnode->subpath = subpath;
3125 6606 : pathnode->numkeys = numCols;
3126 :
3127 : /*
3128 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3129 : * all columns get compared at most of the tuples. (XXX probably this is
3130 : * an overestimate.)
3131 : */
3132 6606 : pathnode->path.startup_cost = subpath->startup_cost;
3133 6606 : pathnode->path.total_cost = subpath->total_cost +
3134 6606 : cpu_operator_cost * subpath->rows * numCols;
3135 6606 : pathnode->path.rows = numGroups;
3136 :
3137 6606 : return pathnode;
3138 : }
3139 :
3140 : /*
3141 : * create_agg_path
3142 : * Creates a pathnode that represents performing aggregation/grouping
3143 : *
3144 : * 'rel' is the parent relation associated with the result
3145 : * 'subpath' is the path representing the source of data
3146 : * 'target' is the PathTarget to be computed
3147 : * 'aggstrategy' is the Agg node's basic implementation strategy
3148 : * 'aggsplit' is the Agg node's aggregate-splitting mode
3149 : * 'groupClause' is a list of SortGroupClause's representing the grouping
3150 : * 'qual' is the HAVING quals if any
3151 : * 'aggcosts' contains cost info about the aggregate functions to be computed
3152 : * 'numGroups' is the estimated number of groups (1 if not grouping)
3153 : */
3154 : AggPath *
3155 51320 : create_agg_path(PlannerInfo *root,
3156 : RelOptInfo *rel,
3157 : Path *subpath,
3158 : PathTarget *target,
3159 : AggStrategy aggstrategy,
3160 : AggSplit aggsplit,
3161 : List *groupClause,
3162 : List *qual,
3163 : const AggClauseCosts *aggcosts,
3164 : double numGroups)
3165 : {
3166 51320 : AggPath *pathnode = makeNode(AggPath);
3167 :
3168 51320 : pathnode->path.pathtype = T_Agg;
3169 51320 : pathnode->path.parent = rel;
3170 51320 : pathnode->path.pathtarget = target;
3171 : /* For now, assume we are above any joins, so no parameterization */
3172 51320 : pathnode->path.param_info = NULL;
3173 51320 : pathnode->path.parallel_aware = false;
3174 87104 : pathnode->path.parallel_safe = rel->consider_parallel &&
3175 35784 : subpath->parallel_safe;
3176 51320 : pathnode->path.parallel_workers = subpath->parallel_workers;
3177 :
3178 51320 : if (aggstrategy == AGG_SORTED)
3179 : {
3180 : /*
3181 : * Attempt to preserve the order of the subpath. Additional pathkeys
3182 : * may have been added in adjust_group_pathkeys_for_groupagg() to
3183 : * support ORDER BY / DISTINCT aggregates. Pathkeys added there
3184 : * belong to columns within the aggregate function, so we must strip
3185 : * these additional pathkeys off as those columns are unavailable
3186 : * above the aggregate node.
3187 : */
3188 7344 : if (list_length(subpath->pathkeys) > root->num_groupby_pathkeys)
3189 326 : pathnode->path.pathkeys = list_copy_head(subpath->pathkeys,
3190 : root->num_groupby_pathkeys);
3191 : else
3192 7018 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3193 : }
3194 : else
3195 43976 : pathnode->path.pathkeys = NIL; /* output is unordered */
3196 :
3197 51320 : pathnode->subpath = subpath;
3198 :
3199 51320 : pathnode->aggstrategy = aggstrategy;
3200 51320 : pathnode->aggsplit = aggsplit;
3201 51320 : pathnode->numGroups = numGroups;
3202 51320 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
3203 51320 : pathnode->groupClause = groupClause;
3204 51320 : pathnode->qual = qual;
3205 :
3206 51320 : cost_agg(&pathnode->path, root,
3207 : aggstrategy, aggcosts,
3208 : list_length(groupClause), numGroups,
3209 : qual,
3210 : subpath->startup_cost, subpath->total_cost,
3211 51320 : subpath->rows, subpath->pathtarget->width);
3212 :
3213 : /* add tlist eval cost for each output row */
3214 51320 : pathnode->path.startup_cost += target->cost.startup;
3215 51320 : pathnode->path.total_cost += target->cost.startup +
3216 51320 : target->cost.per_tuple * pathnode->path.rows;
3217 :
3218 51320 : return pathnode;
3219 : }
3220 :
3221 : /*
3222 : * create_groupingsets_path
3223 : * Creates a pathnode that represents performing GROUPING SETS aggregation
3224 : *
3225 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3226 : * The input path's result must be sorted to match the last entry in
3227 : * rollup_groupclauses.
3228 : *
3229 : * 'rel' is the parent relation associated with the result
3230 : * 'subpath' is the path representing the source of data
3231 : * 'target' is the PathTarget to be computed
3232 : * 'having_qual' is the HAVING quals if any
3233 : * 'rollups' is a list of RollupData nodes
3234 : * 'agg_costs' contains cost info about the aggregate functions to be computed
3235 : */
3236 : GroupingSetsPath *
3237 1792 : create_groupingsets_path(PlannerInfo *root,
3238 : RelOptInfo *rel,
3239 : Path *subpath,
3240 : List *having_qual,
3241 : AggStrategy aggstrategy,
3242 : List *rollups,
3243 : const AggClauseCosts *agg_costs)
3244 : {
3245 1792 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
3246 1792 : PathTarget *target = rel->reltarget;
3247 : ListCell *lc;
3248 1792 : bool is_first = true;
3249 1792 : bool is_first_sort = true;
3250 :
3251 : /* The topmost generated Plan node will be an Agg */
3252 1792 : pathnode->path.pathtype = T_Agg;
3253 1792 : pathnode->path.parent = rel;
3254 1792 : pathnode->path.pathtarget = target;
3255 1792 : pathnode->path.param_info = subpath->param_info;
3256 1792 : pathnode->path.parallel_aware = false;
3257 2614 : pathnode->path.parallel_safe = rel->consider_parallel &&
3258 822 : subpath->parallel_safe;
3259 1792 : pathnode->path.parallel_workers = subpath->parallel_workers;
3260 1792 : pathnode->subpath = subpath;
3261 :
3262 : /*
3263 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3264 : * to AGG_HASHED, here if possible.
3265 : */
3266 2556 : if (aggstrategy == AGG_SORTED &&
3267 764 : list_length(rollups) == 1 &&
3268 362 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3269 42 : aggstrategy = AGG_PLAIN;
3270 :
3271 2636 : if (aggstrategy == AGG_MIXED &&
3272 844 : list_length(rollups) == 1)
3273 0 : aggstrategy = AGG_HASHED;
3274 :
3275 : /*
3276 : * Output will be in sorted order by group_pathkeys if, and only if, there
3277 : * is a single rollup operation on a non-empty list of grouping
3278 : * expressions.
3279 : */
3280 1792 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3281 320 : pathnode->path.pathkeys = root->group_pathkeys;
3282 : else
3283 1472 : pathnode->path.pathkeys = NIL;
3284 :
3285 1792 : pathnode->aggstrategy = aggstrategy;
3286 1792 : pathnode->rollups = rollups;
3287 1792 : pathnode->qual = having_qual;
3288 1792 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3289 :
3290 : Assert(rollups != NIL);
3291 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3292 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3293 :
3294 6396 : foreach(lc, rollups)
3295 : {
3296 4604 : RollupData *rollup = lfirst(lc);
3297 4604 : List *gsets = rollup->gsets;
3298 4604 : int numGroupCols = list_length(linitial(gsets));
3299 :
3300 : /*
3301 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3302 : * (already-sorted) input, and following ones do their own sort.
3303 : *
3304 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3305 : *
3306 : * In AGG_MIXED mode, the first rollups are hashed, the first
3307 : * non-hashed one takes the (already-sorted) input, and following ones
3308 : * do their own sort.
3309 : */
3310 4604 : if (is_first)
3311 : {
3312 1792 : cost_agg(&pathnode->path, root,
3313 : aggstrategy,
3314 : agg_costs,
3315 : numGroupCols,
3316 : rollup->numGroups,
3317 : having_qual,
3318 : subpath->startup_cost,
3319 : subpath->total_cost,
3320 : subpath->rows,
3321 1792 : subpath->pathtarget->width);
3322 1792 : is_first = false;
3323 1792 : if (!rollup->is_hashed)
3324 764 : is_first_sort = false;
3325 : }
3326 : else
3327 : {
3328 : Path sort_path; /* dummy for result of cost_sort */
3329 : Path agg_path; /* dummy for result of cost_agg */
3330 :
3331 2812 : if (rollup->is_hashed || is_first_sort)
3332 : {
3333 : /*
3334 : * Account for cost of aggregation, but don't charge input
3335 : * cost again
3336 : */
3337 2158 : cost_agg(&agg_path, root,
3338 2158 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3339 : agg_costs,
3340 : numGroupCols,
3341 : rollup->numGroups,
3342 : having_qual,
3343 : 0.0, 0.0,
3344 : subpath->rows,
3345 2158 : subpath->pathtarget->width);
3346 2158 : if (!rollup->is_hashed)
3347 844 : is_first_sort = false;
3348 : }
3349 : else
3350 : {
3351 : /* Account for cost of sort, but don't charge input cost again */
3352 654 : cost_sort(&sort_path, root, NIL,
3353 : 0.0,
3354 : subpath->rows,
3355 654 : subpath->pathtarget->width,
3356 : 0.0,
3357 : work_mem,
3358 : -1.0);
3359 :
3360 : /* Account for cost of aggregation */
3361 :
3362 654 : cost_agg(&agg_path, root,
3363 : AGG_SORTED,
3364 : agg_costs,
3365 : numGroupCols,
3366 : rollup->numGroups,
3367 : having_qual,
3368 : sort_path.startup_cost,
3369 : sort_path.total_cost,
3370 : sort_path.rows,
3371 654 : subpath->pathtarget->width);
3372 : }
3373 :
3374 2812 : pathnode->path.total_cost += agg_path.total_cost;
3375 2812 : pathnode->path.rows += agg_path.rows;
3376 : }
3377 : }
3378 :
3379 : /* add tlist eval cost for each output row */
3380 1792 : pathnode->path.startup_cost += target->cost.startup;
3381 1792 : pathnode->path.total_cost += target->cost.startup +
3382 1792 : target->cost.per_tuple * pathnode->path.rows;
3383 :
3384 1792 : return pathnode;
3385 : }
3386 :
3387 : /*
3388 : * create_minmaxagg_path
3389 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3390 : *
3391 : * 'rel' is the parent relation associated with the result
3392 : * 'target' is the PathTarget to be computed
3393 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3394 : * 'quals' is the HAVING quals if any
3395 : */
3396 : MinMaxAggPath *
3397 386 : create_minmaxagg_path(PlannerInfo *root,
3398 : RelOptInfo *rel,
3399 : PathTarget *target,
3400 : List *mmaggregates,
3401 : List *quals)
3402 : {
3403 386 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3404 : Cost initplan_cost;
3405 : ListCell *lc;
3406 :
3407 : /* The topmost generated Plan node will be a Result */
3408 386 : pathnode->path.pathtype = T_Result;
3409 386 : pathnode->path.parent = rel;
3410 386 : pathnode->path.pathtarget = target;
3411 : /* For now, assume we are above any joins, so no parameterization */
3412 386 : pathnode->path.param_info = NULL;
3413 386 : pathnode->path.parallel_aware = false;
3414 386 : pathnode->path.parallel_safe = true; /* might change below */
3415 386 : pathnode->path.parallel_workers = 0;
3416 : /* Result is one unordered row */
3417 386 : pathnode->path.rows = 1;
3418 386 : pathnode->path.pathkeys = NIL;
3419 :
3420 386 : pathnode->mmaggregates = mmaggregates;
3421 386 : pathnode->quals = quals;
3422 :
3423 : /* Calculate cost of all the initplans, and check parallel safety */
3424 386 : initplan_cost = 0;
3425 808 : foreach(lc, mmaggregates)
3426 : {
3427 422 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3428 :
3429 422 : initplan_cost += mminfo->pathcost;
3430 422 : if (!mminfo->path->parallel_safe)
3431 86 : pathnode->path.parallel_safe = false;
3432 : }
3433 :
3434 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3435 386 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3436 386 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3437 386 : target->cost.per_tuple + cpu_tuple_cost;
3438 :
3439 : /*
3440 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3441 : * rowcount estimate should be 1 no matter what the qual is.
3442 : */
3443 386 : if (quals)
3444 : {
3445 : QualCost qual_cost;
3446 :
3447 0 : cost_qual_eval(&qual_cost, quals, root);
3448 0 : pathnode->path.startup_cost += qual_cost.startup;
3449 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3450 : }
3451 :
3452 : /*
3453 : * If the initplans were all parallel-safe, also check safety of the
3454 : * target and quals. (The Result node itself isn't parallelizable, but if
3455 : * we are in a subquery then it can be useful for the outer query to know
3456 : * that this one is parallel-safe.)
3457 : */
3458 386 : if (pathnode->path.parallel_safe)
3459 300 : pathnode->path.parallel_safe =
3460 600 : is_parallel_safe(root, (Node *) target->exprs) &&
3461 300 : is_parallel_safe(root, (Node *) quals);
3462 :
3463 386 : return pathnode;
3464 : }
3465 :
3466 : /*
3467 : * create_windowagg_path
3468 : * Creates a pathnode that represents computation of window functions
3469 : *
3470 : * 'rel' is the parent relation associated with the result
3471 : * 'subpath' is the path representing the source of data
3472 : * 'target' is the PathTarget to be computed
3473 : * 'windowFuncs' is a list of WindowFunc structs
3474 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3475 : * 'qual' WindowClause.runconditions from lower-level WindowAggPaths.
3476 : * Must always be NIL when topwindow == false
3477 : * 'topwindow' pass as true only for the top-level WindowAgg. False for all
3478 : * intermediate WindowAggs.
3479 : *
3480 : * The input must be sorted according to the WindowClause's PARTITION keys
3481 : * plus ORDER BY keys.
3482 : */
3483 : WindowAggPath *
3484 2630 : create_windowagg_path(PlannerInfo *root,
3485 : RelOptInfo *rel,
3486 : Path *subpath,
3487 : PathTarget *target,
3488 : List *windowFuncs,
3489 : WindowClause *winclause,
3490 : List *qual,
3491 : bool topwindow)
3492 : {
3493 2630 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3494 :
3495 : /* qual can only be set for the topwindow */
3496 : Assert(qual == NIL || topwindow);
3497 :
3498 2630 : pathnode->path.pathtype = T_WindowAgg;
3499 2630 : pathnode->path.parent = rel;
3500 2630 : pathnode->path.pathtarget = target;
3501 : /* For now, assume we are above any joins, so no parameterization */
3502 2630 : pathnode->path.param_info = NULL;
3503 2630 : pathnode->path.parallel_aware = false;
3504 2630 : pathnode->path.parallel_safe = rel->consider_parallel &&
3505 0 : subpath->parallel_safe;
3506 2630 : pathnode->path.parallel_workers = subpath->parallel_workers;
3507 : /* WindowAgg preserves the input sort order */
3508 2630 : pathnode->path.pathkeys = subpath->pathkeys;
3509 :
3510 2630 : pathnode->subpath = subpath;
3511 2630 : pathnode->winclause = winclause;
3512 2630 : pathnode->qual = qual;
3513 2630 : pathnode->topwindow = topwindow;
3514 :
3515 : /*
3516 : * For costing purposes, assume that there are no redundant partitioning
3517 : * or ordering columns; it's not worth the trouble to deal with that
3518 : * corner case here. So we just pass the unmodified list lengths to
3519 : * cost_windowagg.
3520 : */
3521 2630 : cost_windowagg(&pathnode->path, root,
3522 : windowFuncs,
3523 : winclause,
3524 : subpath->startup_cost,
3525 : subpath->total_cost,
3526 : subpath->rows);
3527 :
3528 : /* add tlist eval cost for each output row */
3529 2630 : pathnode->path.startup_cost += target->cost.startup;
3530 2630 : pathnode->path.total_cost += target->cost.startup +
3531 2630 : target->cost.per_tuple * pathnode->path.rows;
3532 :
3533 2630 : return pathnode;
3534 : }
3535 :
3536 : /*
3537 : * create_setop_path
3538 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3539 : *
3540 : * 'rel' is the parent relation associated with the result
3541 : * 'subpath' is the path representing the source of data
3542 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3543 : * 'strategy' is the implementation strategy (sorted or hashed)
3544 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3545 : * 'flagColIdx' is the column number where the flag column will be, if any
3546 : * 'firstFlag' is the flag value for the first input relation when hashing;
3547 : * or -1 when sorting
3548 : * 'numGroups' is the estimated number of distinct groups
3549 : * 'outputRows' is the estimated number of output rows
3550 : */
3551 : SetOpPath *
3552 600 : create_setop_path(PlannerInfo *root,
3553 : RelOptInfo *rel,
3554 : Path *subpath,
3555 : SetOpCmd cmd,
3556 : SetOpStrategy strategy,
3557 : List *distinctList,
3558 : AttrNumber flagColIdx,
3559 : int firstFlag,
3560 : double numGroups,
3561 : double outputRows)
3562 : {
3563 600 : SetOpPath *pathnode = makeNode(SetOpPath);
3564 :
3565 600 : pathnode->path.pathtype = T_SetOp;
3566 600 : pathnode->path.parent = rel;
3567 : /* SetOp doesn't project, so use source path's pathtarget */
3568 600 : pathnode->path.pathtarget = subpath->pathtarget;
3569 : /* For now, assume we are above any joins, so no parameterization */
3570 600 : pathnode->path.param_info = NULL;
3571 600 : pathnode->path.parallel_aware = false;
3572 600 : pathnode->path.parallel_safe = rel->consider_parallel &&
3573 0 : subpath->parallel_safe;
3574 600 : pathnode->path.parallel_workers = subpath->parallel_workers;
3575 : /* SetOp preserves the input sort order if in sort mode */
3576 600 : pathnode->path.pathkeys =
3577 600 : (strategy == SETOP_SORTED) ? subpath->pathkeys : NIL;
3578 :
3579 600 : pathnode->subpath = subpath;
3580 600 : pathnode->cmd = cmd;
3581 600 : pathnode->strategy = strategy;
3582 600 : pathnode->distinctList = distinctList;
3583 600 : pathnode->flagColIdx = flagColIdx;
3584 600 : pathnode->firstFlag = firstFlag;
3585 600 : pathnode->numGroups = numGroups;
3586 :
3587 : /*
3588 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3589 : * all columns get compared at most of the tuples.
3590 : */
3591 600 : pathnode->path.startup_cost = subpath->startup_cost;
3592 1200 : pathnode->path.total_cost = subpath->total_cost +
3593 600 : cpu_operator_cost * subpath->rows * list_length(distinctList);
3594 600 : pathnode->path.rows = outputRows;
3595 :
3596 600 : return pathnode;
3597 : }
3598 :
3599 : /*
3600 : * create_recursiveunion_path
3601 : * Creates a pathnode that represents a recursive UNION node
3602 : *
3603 : * 'rel' is the parent relation associated with the result
3604 : * 'leftpath' is the source of data for the non-recursive term
3605 : * 'rightpath' is the source of data for the recursive term
3606 : * 'target' is the PathTarget to be computed
3607 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3608 : * 'wtParam' is the ID of Param representing work table
3609 : * 'numGroups' is the estimated number of groups
3610 : *
3611 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3612 : */
3613 : RecursiveUnionPath *
3614 802 : create_recursiveunion_path(PlannerInfo *root,
3615 : RelOptInfo *rel,
3616 : Path *leftpath,
3617 : Path *rightpath,
3618 : PathTarget *target,
3619 : List *distinctList,
3620 : int wtParam,
3621 : double numGroups)
3622 : {
3623 802 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3624 :
3625 802 : pathnode->path.pathtype = T_RecursiveUnion;
3626 802 : pathnode->path.parent = rel;
3627 802 : pathnode->path.pathtarget = target;
3628 : /* For now, assume we are above any joins, so no parameterization */
3629 802 : pathnode->path.param_info = NULL;
3630 802 : pathnode->path.parallel_aware = false;
3631 1604 : pathnode->path.parallel_safe = rel->consider_parallel &&
3632 802 : leftpath->parallel_safe && rightpath->parallel_safe;
3633 : /* Foolish, but we'll do it like joins for now: */
3634 802 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3635 : /* RecursiveUnion result is always unsorted */
3636 802 : pathnode->path.pathkeys = NIL;
3637 :
3638 802 : pathnode->leftpath = leftpath;
3639 802 : pathnode->rightpath = rightpath;
3640 802 : pathnode->distinctList = distinctList;
3641 802 : pathnode->wtParam = wtParam;
3642 802 : pathnode->numGroups = numGroups;
3643 :
3644 802 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3645 :
3646 802 : return pathnode;
3647 : }
3648 :
3649 : /*
3650 : * create_lockrows_path
3651 : * Creates a pathnode that represents acquiring row locks
3652 : *
3653 : * 'rel' is the parent relation associated with the result
3654 : * 'subpath' is the path representing the source of data
3655 : * 'rowMarks' is a list of PlanRowMark's
3656 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3657 : */
3658 : LockRowsPath *
3659 7918 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3660 : Path *subpath, List *rowMarks, int epqParam)
3661 : {
3662 7918 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3663 :
3664 7918 : pathnode->path.pathtype = T_LockRows;
3665 7918 : pathnode->path.parent = rel;
3666 : /* LockRows doesn't project, so use source path's pathtarget */
3667 7918 : pathnode->path.pathtarget = subpath->pathtarget;
3668 : /* For now, assume we are above any joins, so no parameterization */
3669 7918 : pathnode->path.param_info = NULL;
3670 7918 : pathnode->path.parallel_aware = false;
3671 7918 : pathnode->path.parallel_safe = false;
3672 7918 : pathnode->path.parallel_workers = 0;
3673 7918 : pathnode->path.rows = subpath->rows;
3674 :
3675 : /*
3676 : * The result cannot be assumed sorted, since locking might cause the sort
3677 : * key columns to be replaced with new values.
3678 : */
3679 7918 : pathnode->path.pathkeys = NIL;
3680 :
3681 7918 : pathnode->subpath = subpath;
3682 7918 : pathnode->rowMarks = rowMarks;
3683 7918 : pathnode->epqParam = epqParam;
3684 :
3685 : /*
3686 : * We should charge something extra for the costs of row locking and
3687 : * possible refetches, but it's hard to say how much. For now, use
3688 : * cpu_tuple_cost per row.
3689 : */
3690 7918 : pathnode->path.startup_cost = subpath->startup_cost;
3691 7918 : pathnode->path.total_cost = subpath->total_cost +
3692 7918 : cpu_tuple_cost * subpath->rows;
3693 :
3694 7918 : return pathnode;
3695 : }
3696 :
3697 : /*
3698 : * create_modifytable_path
3699 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE/MERGE
3700 : * mods
3701 : *
3702 : * 'rel' is the parent relation associated with the result
3703 : * 'subpath' is a Path producing source data
3704 : * 'operation' is the operation type
3705 : * 'canSetTag' is true if we set the command tag/es_processed
3706 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3707 : * 'rootRelation' is the partitioned/inherited table root RTI, or 0 if none
3708 : * 'partColsUpdated' is true if any partitioning columns are being updated,
3709 : * either from the target relation or a descendent partitioned table.
3710 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3711 : * 'updateColnosLists' is a list of UPDATE target column number lists
3712 : * (one sublist per rel); or NIL if not an UPDATE
3713 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3714 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3715 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3716 : * 'onconflict' is the ON CONFLICT clause, or NULL
3717 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3718 : * 'mergeActionLists' is a list of lists of MERGE actions (one per rel)
3719 : */
3720 : ModifyTablePath *
3721 88382 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3722 : Path *subpath,
3723 : CmdType operation, bool canSetTag,
3724 : Index nominalRelation, Index rootRelation,
3725 : bool partColsUpdated,
3726 : List *resultRelations,
3727 : List *updateColnosLists,
3728 : List *withCheckOptionLists, List *returningLists,
3729 : List *rowMarks, OnConflictExpr *onconflict,
3730 : List *mergeActionLists, int epqParam)
3731 : {
3732 88382 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3733 :
3734 : Assert(operation == CMD_MERGE ||
3735 : (operation == CMD_UPDATE ?
3736 : list_length(resultRelations) == list_length(updateColnosLists) :
3737 : updateColnosLists == NIL));
3738 : Assert(withCheckOptionLists == NIL ||
3739 : list_length(resultRelations) == list_length(withCheckOptionLists));
3740 : Assert(returningLists == NIL ||
3741 : list_length(resultRelations) == list_length(returningLists));
3742 :
3743 88382 : pathnode->path.pathtype = T_ModifyTable;
3744 88382 : pathnode->path.parent = rel;
3745 : /* pathtarget is not interesting, just make it minimally valid */
3746 88382 : pathnode->path.pathtarget = rel->reltarget;
3747 : /* For now, assume we are above any joins, so no parameterization */
3748 88382 : pathnode->path.param_info = NULL;
3749 88382 : pathnode->path.parallel_aware = false;
3750 88382 : pathnode->path.parallel_safe = false;
3751 88382 : pathnode->path.parallel_workers = 0;
3752 88382 : pathnode->path.pathkeys = NIL;
3753 :
3754 : /*
3755 : * Compute cost & rowcount as subpath cost & rowcount (if RETURNING)
3756 : *
3757 : * Currently, we don't charge anything extra for the actual table
3758 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3759 : * expressions if any. It would only be window dressing, since
3760 : * ModifyTable is always a top-level node and there is no way for the
3761 : * costs to change any higher-level planning choices. But we might want
3762 : * to make it look better sometime.
3763 : */
3764 88382 : pathnode->path.startup_cost = subpath->startup_cost;
3765 88382 : pathnode->path.total_cost = subpath->total_cost;
3766 88382 : if (returningLists != NIL)
3767 : {
3768 2386 : pathnode->path.rows = subpath->rows;
3769 :
3770 : /*
3771 : * Set width to match the subpath output. XXX this is totally wrong:
3772 : * we should return an average of the RETURNING tlist widths. But
3773 : * it's what happened historically, and improving it is a task for
3774 : * another day. (Again, it's mostly window dressing.)
3775 : */
3776 2386 : pathnode->path.pathtarget->width = subpath->pathtarget->width;
3777 : }
3778 : else
3779 : {
3780 85996 : pathnode->path.rows = 0;
3781 85996 : pathnode->path.pathtarget->width = 0;
3782 : }
3783 :
3784 88382 : pathnode->subpath = subpath;
3785 88382 : pathnode->operation = operation;
3786 88382 : pathnode->canSetTag = canSetTag;
3787 88382 : pathnode->nominalRelation = nominalRelation;
3788 88382 : pathnode->rootRelation = rootRelation;
3789 88382 : pathnode->partColsUpdated = partColsUpdated;
3790 88382 : pathnode->resultRelations = resultRelations;
3791 88382 : pathnode->updateColnosLists = updateColnosLists;
3792 88382 : pathnode->withCheckOptionLists = withCheckOptionLists;
3793 88382 : pathnode->returningLists = returningLists;
3794 88382 : pathnode->rowMarks = rowMarks;
3795 88382 : pathnode->onconflict = onconflict;
3796 88382 : pathnode->epqParam = epqParam;
3797 88382 : pathnode->mergeActionLists = mergeActionLists;
3798 :
3799 88382 : return pathnode;
3800 : }
3801 :
3802 : /*
3803 : * create_limit_path
3804 : * Creates a pathnode that represents performing LIMIT/OFFSET
3805 : *
3806 : * In addition to providing the actual OFFSET and LIMIT expressions,
3807 : * the caller must provide estimates of their values for costing purposes.
3808 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3809 : * the clause not being present, and -1 means it's present but we could
3810 : * not estimate its value.
3811 : *
3812 : * 'rel' is the parent relation associated with the result
3813 : * 'subpath' is the path representing the source of data
3814 : * 'limitOffset' is the actual OFFSET expression, or NULL
3815 : * 'limitCount' is the actual LIMIT expression, or NULL
3816 : * 'offset_est' is the estimated value of the OFFSET expression
3817 : * 'count_est' is the estimated value of the LIMIT expression
3818 : */
3819 : LimitPath *
3820 5592 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3821 : Path *subpath,
3822 : Node *limitOffset, Node *limitCount,
3823 : LimitOption limitOption,
3824 : int64 offset_est, int64 count_est)
3825 : {
3826 5592 : LimitPath *pathnode = makeNode(LimitPath);
3827 :
3828 5592 : pathnode->path.pathtype = T_Limit;
3829 5592 : pathnode->path.parent = rel;
3830 : /* Limit doesn't project, so use source path's pathtarget */
3831 5592 : pathnode->path.pathtarget = subpath->pathtarget;
3832 : /* For now, assume we are above any joins, so no parameterization */
3833 5592 : pathnode->path.param_info = NULL;
3834 5592 : pathnode->path.parallel_aware = false;
3835 7824 : pathnode->path.parallel_safe = rel->consider_parallel &&
3836 2232 : subpath->parallel_safe;
3837 5592 : pathnode->path.parallel_workers = subpath->parallel_workers;
3838 5592 : pathnode->path.rows = subpath->rows;
3839 5592 : pathnode->path.startup_cost = subpath->startup_cost;
3840 5592 : pathnode->path.total_cost = subpath->total_cost;
3841 5592 : pathnode->path.pathkeys = subpath->pathkeys;
3842 5592 : pathnode->subpath = subpath;
3843 5592 : pathnode->limitOffset = limitOffset;
3844 5592 : pathnode->limitCount = limitCount;
3845 5592 : pathnode->limitOption = limitOption;
3846 :
3847 : /*
3848 : * Adjust the output rows count and costs according to the offset/limit.
3849 : */
3850 5592 : adjust_limit_rows_costs(&pathnode->path.rows,
3851 : &pathnode->path.startup_cost,
3852 : &pathnode->path.total_cost,
3853 : offset_est, count_est);
3854 :
3855 5592 : return pathnode;
3856 : }
3857 :
3858 : /*
3859 : * adjust_limit_rows_costs
3860 : * Adjust the size and cost estimates for a LimitPath node according to the
3861 : * offset/limit.
3862 : *
3863 : * This is only a cosmetic issue if we are at top level, but if we are
3864 : * building a subquery then it's important to report correct info to the outer
3865 : * planner.
3866 : *
3867 : * When the offset or count couldn't be estimated, use 10% of the estimated
3868 : * number of rows emitted from the subpath.
3869 : *
3870 : * XXX we don't bother to add eval costs of the offset/limit expressions
3871 : * themselves to the path costs. In theory we should, but in most cases those
3872 : * expressions are trivial and it's just not worth the trouble.
3873 : */
3874 : void
3875 5774 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3876 : Cost *startup_cost, /* in/out parameter */
3877 : Cost *total_cost, /* in/out parameter */
3878 : int64 offset_est,
3879 : int64 count_est)
3880 : {
3881 5774 : double input_rows = *rows;
3882 5774 : Cost input_startup_cost = *startup_cost;
3883 5774 : Cost input_total_cost = *total_cost;
3884 :
3885 5774 : if (offset_est != 0)
3886 : {
3887 : double offset_rows;
3888 :
3889 672 : if (offset_est > 0)
3890 648 : offset_rows = (double) offset_est;
3891 : else
3892 24 : offset_rows = clamp_row_est(input_rows * 0.10);
3893 672 : if (offset_rows > *rows)
3894 26 : offset_rows = *rows;
3895 672 : if (input_rows > 0)
3896 672 : *startup_cost +=
3897 672 : (input_total_cost - input_startup_cost)
3898 672 : * offset_rows / input_rows;
3899 672 : *rows -= offset_rows;
3900 672 : if (*rows < 1)
3901 26 : *rows = 1;
3902 : }
3903 :
3904 5774 : if (count_est != 0)
3905 : {
3906 : double count_rows;
3907 :
3908 5726 : if (count_est > 0)
3909 5720 : count_rows = (double) count_est;
3910 : else
3911 6 : count_rows = clamp_row_est(input_rows * 0.10);
3912 5726 : if (count_rows > *rows)
3913 236 : count_rows = *rows;
3914 5726 : if (input_rows > 0)
3915 5726 : *total_cost = *startup_cost +
3916 5726 : (input_total_cost - input_startup_cost)
3917 5726 : * count_rows / input_rows;
3918 5726 : *rows = count_rows;
3919 5726 : if (*rows < 1)
3920 0 : *rows = 1;
3921 : }
3922 5774 : }
3923 :
3924 :
3925 : /*
3926 : * reparameterize_path
3927 : * Attempt to modify a Path to have greater parameterization
3928 : *
3929 : * We use this to attempt to bring all child paths of an appendrel to the
3930 : * same parameterization level, ensuring that they all enforce the same set
3931 : * of join quals (and thus that that parameterization can be attributed to
3932 : * an append path built from such paths). Currently, only a few path types
3933 : * are supported here, though more could be added at need. We return NULL
3934 : * if we can't reparameterize the given path.
3935 : *
3936 : * Note: we intentionally do not pass created paths to add_path(); it would
3937 : * possibly try to delete them on the grounds of being cost-inferior to the
3938 : * paths they were made from, and we don't want that. Paths made here are
3939 : * not necessarily of general-purpose usefulness, but they can be useful
3940 : * as members of an append path.
3941 : */
3942 : Path *
3943 308 : reparameterize_path(PlannerInfo *root, Path *path,
3944 : Relids required_outer,
3945 : double loop_count)
3946 : {
3947 308 : RelOptInfo *rel = path->parent;
3948 :
3949 : /* Can only increase, not decrease, path's parameterization */
3950 308 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
3951 0 : return NULL;
3952 308 : switch (path->pathtype)
3953 : {
3954 228 : case T_SeqScan:
3955 228 : return create_seqscan_path(root, rel, required_outer, 0);
3956 0 : case T_SampleScan:
3957 0 : return (Path *) create_samplescan_path(root, rel, required_outer);
3958 0 : case T_IndexScan:
3959 : case T_IndexOnlyScan:
3960 : {
3961 0 : IndexPath *ipath = (IndexPath *) path;
3962 0 : IndexPath *newpath = makeNode(IndexPath);
3963 :
3964 : /*
3965 : * We can't use create_index_path directly, and would not want
3966 : * to because it would re-compute the indexqual conditions
3967 : * which is wasted effort. Instead we hack things a bit:
3968 : * flat-copy the path node, revise its param_info, and redo
3969 : * the cost estimate.
3970 : */
3971 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3972 0 : newpath->path.param_info =
3973 0 : get_baserel_parampathinfo(root, rel, required_outer);
3974 0 : cost_index(newpath, root, loop_count, false);
3975 0 : return (Path *) newpath;
3976 : }
3977 0 : case T_BitmapHeapScan:
3978 : {
3979 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3980 :
3981 0 : return (Path *) create_bitmap_heap_path(root,
3982 : rel,
3983 : bpath->bitmapqual,
3984 : required_outer,
3985 : loop_count, 0);
3986 : }
3987 0 : case T_SubqueryScan:
3988 : {
3989 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
3990 0 : Path *subpath = spath->subpath;
3991 : bool trivial_pathtarget;
3992 :
3993 : /*
3994 : * If existing node has zero extra cost, we must have decided
3995 : * its target is trivial. (The converse is not true, because
3996 : * it might have a trivial target but quals to enforce; but in
3997 : * that case the new node will too, so it doesn't matter
3998 : * whether we get the right answer here.)
3999 : */
4000 0 : trivial_pathtarget =
4001 0 : (subpath->total_cost == spath->path.total_cost);
4002 :
4003 0 : return (Path *) create_subqueryscan_path(root,
4004 : rel,
4005 : subpath,
4006 : trivial_pathtarget,
4007 : spath->path.pathkeys,
4008 : required_outer);
4009 : }
4010 48 : case T_Result:
4011 : /* Supported only for RTE_RESULT scan paths */
4012 48 : if (IsA(path, Path))
4013 48 : return create_resultscan_path(root, rel, required_outer);
4014 0 : break;
4015 0 : case T_Append:
4016 : {
4017 0 : AppendPath *apath = (AppendPath *) path;
4018 0 : List *childpaths = NIL;
4019 0 : List *partialpaths = NIL;
4020 : int i;
4021 : ListCell *lc;
4022 :
4023 : /* Reparameterize the children */
4024 0 : i = 0;
4025 0 : foreach(lc, apath->subpaths)
4026 : {
4027 0 : Path *spath = (Path *) lfirst(lc);
4028 :
4029 0 : spath = reparameterize_path(root, spath,
4030 : required_outer,
4031 : loop_count);
4032 0 : if (spath == NULL)
4033 0 : return NULL;
4034 : /* We have to re-split the regular and partial paths */
4035 0 : if (i < apath->first_partial_path)
4036 0 : childpaths = lappend(childpaths, spath);
4037 : else
4038 0 : partialpaths = lappend(partialpaths, spath);
4039 0 : i++;
4040 : }
4041 0 : return (Path *)
4042 0 : create_append_path(root, rel, childpaths, partialpaths,
4043 : apath->path.pathkeys, required_outer,
4044 : apath->path.parallel_workers,
4045 0 : apath->path.parallel_aware,
4046 : -1);
4047 : }
4048 0 : case T_Material:
4049 : {
4050 0 : MaterialPath *mpath = (MaterialPath *) path;
4051 0 : Path *spath = mpath->subpath;
4052 :
4053 0 : spath = reparameterize_path(root, spath,
4054 : required_outer,
4055 : loop_count);
4056 0 : if (spath == NULL)
4057 0 : return NULL;
4058 0 : return (Path *) create_material_path(rel, spath);
4059 : }
4060 0 : case T_Memoize:
4061 : {
4062 0 : MemoizePath *mpath = (MemoizePath *) path;
4063 0 : Path *spath = mpath->subpath;
4064 :
4065 0 : spath = reparameterize_path(root, spath,
4066 : required_outer,
4067 : loop_count);
4068 0 : if (spath == NULL)
4069 0 : return NULL;
4070 0 : return (Path *) create_memoize_path(root, rel,
4071 : spath,
4072 : mpath->param_exprs,
4073 : mpath->hash_operators,
4074 0 : mpath->singlerow,
4075 0 : mpath->binary_mode,
4076 : mpath->calls);
4077 : }
4078 32 : default:
4079 32 : break;
4080 : }
4081 32 : return NULL;
4082 : }
4083 :
4084 : /*
4085 : * reparameterize_path_by_child
4086 : * Given a path parameterized by the parent of the given child relation,
4087 : * translate the path to be parameterized by the given child relation.
4088 : *
4089 : * Most fields in the path are not changed, but any expressions must be
4090 : * adjusted to refer to the correct varnos, and any subpaths must be
4091 : * recursively reparameterized. Other fields that refer to specific relids
4092 : * also need adjustment.
4093 : *
4094 : * The cost, number of rows, width and parallel path properties depend upon
4095 : * path->parent, which does not change during the translation. So we need
4096 : * not change those.
4097 : *
4098 : * Currently, only a few path types are supported here, though more could be
4099 : * added at need. We return NULL if we can't reparameterize the given path.
4100 : *
4101 : * Note that this function can change referenced RangeTblEntries, RelOptInfos
4102 : * and IndexOptInfos as well as the Path structures. Therefore, it's only safe
4103 : * to call during create_plan(), when we have made a final choice of which Path
4104 : * to use for each RangeTblEntry/RelOptInfo/IndexOptInfo.
4105 : *
4106 : * Keep this code in sync with path_is_reparameterizable_by_child()!
4107 : */
4108 : Path *
4109 80270 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
4110 : RelOptInfo *child_rel)
4111 : {
4112 : Path *new_path;
4113 : ParamPathInfo *new_ppi;
4114 : ParamPathInfo *old_ppi;
4115 : Relids required_outer;
4116 :
4117 : #define ADJUST_CHILD_ATTRS(node) \
4118 : ((node) = (void *) adjust_appendrel_attrs_multilevel(root, \
4119 : (Node *) (node), \
4120 : child_rel, \
4121 : child_rel->top_parent))
4122 :
4123 : #define REPARAMETERIZE_CHILD_PATH(path) \
4124 : do { \
4125 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
4126 : if ((path) == NULL) \
4127 : return NULL; \
4128 : } while(0)
4129 :
4130 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
4131 : do { \
4132 : if ((pathlist) != NIL) \
4133 : { \
4134 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
4135 : child_rel); \
4136 : if ((pathlist) == NIL) \
4137 : return NULL; \
4138 : } \
4139 : } while(0)
4140 :
4141 : /*
4142 : * If the path is not parameterized by the parent of the given relation,
4143 : * it doesn't need reparameterization.
4144 : */
4145 80270 : if (!path->param_info ||
4146 42680 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4147 79382 : return path;
4148 :
4149 : /*
4150 : * If possible, reparameterize the given path.
4151 : *
4152 : * This function is currently only applied to the inner side of a nestloop
4153 : * join that is being partitioned by the partitionwise-join code. Hence,
4154 : * we need only support path types that plausibly arise in that context.
4155 : * (In particular, supporting sorted path types would be a waste of code
4156 : * and cycles: even if we translated them here, they'd just lose in
4157 : * subsequent cost comparisons.) If we do see an unsupported path type,
4158 : * that just means we won't be able to generate a partitionwise-join plan
4159 : * using that path type.
4160 : */
4161 888 : switch (nodeTag(path))
4162 : {
4163 228 : case T_Path:
4164 228 : new_path = path;
4165 228 : ADJUST_CHILD_ATTRS(new_path->parent->baserestrictinfo);
4166 228 : if (path->pathtype == T_SampleScan)
4167 : {
4168 48 : Index scan_relid = path->parent->relid;
4169 : RangeTblEntry *rte;
4170 :
4171 : /* it should be a base rel with a tablesample clause... */
4172 : Assert(scan_relid > 0);
4173 48 : rte = planner_rt_fetch(scan_relid, root);
4174 : Assert(rte->rtekind == RTE_RELATION);
4175 : Assert(rte->tablesample != NULL);
4176 :
4177 48 : ADJUST_CHILD_ATTRS(rte->tablesample);
4178 : }
4179 228 : break;
4180 :
4181 444 : case T_IndexPath:
4182 : {
4183 444 : IndexPath *ipath = (IndexPath *) path;
4184 :
4185 444 : ADJUST_CHILD_ATTRS(ipath->indexinfo->indrestrictinfo);
4186 444 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
4187 444 : new_path = (Path *) ipath;
4188 : }
4189 444 : break;
4190 :
4191 48 : case T_BitmapHeapPath:
4192 : {
4193 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4194 :
4195 48 : ADJUST_CHILD_ATTRS(bhpath->path.parent->baserestrictinfo);
4196 48 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
4197 48 : new_path = (Path *) bhpath;
4198 : }
4199 48 : break;
4200 :
4201 24 : case T_BitmapAndPath:
4202 : {
4203 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4204 :
4205 24 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
4206 24 : new_path = (Path *) bapath;
4207 : }
4208 24 : break;
4209 :
4210 24 : case T_BitmapOrPath:
4211 : {
4212 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4213 :
4214 24 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
4215 24 : new_path = (Path *) bopath;
4216 : }
4217 24 : break;
4218 :
4219 0 : case T_ForeignPath:
4220 : {
4221 0 : ForeignPath *fpath = (ForeignPath *) path;
4222 : ReparameterizeForeignPathByChild_function rfpc_func;
4223 :
4224 0 : ADJUST_CHILD_ATTRS(fpath->path.parent->baserestrictinfo);
4225 0 : if (fpath->fdw_outerpath)
4226 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
4227 0 : if (fpath->fdw_restrictinfo)
4228 0 : ADJUST_CHILD_ATTRS(fpath->fdw_restrictinfo);
4229 :
4230 : /* Hand over to FDW if needed. */
4231 0 : rfpc_func =
4232 0 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
4233 0 : if (rfpc_func)
4234 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
4235 : child_rel);
4236 0 : new_path = (Path *) fpath;
4237 : }
4238 0 : break;
4239 :
4240 0 : case T_CustomPath:
4241 : {
4242 0 : CustomPath *cpath = (CustomPath *) path;
4243 :
4244 0 : ADJUST_CHILD_ATTRS(cpath->path.parent->baserestrictinfo);
4245 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
4246 0 : if (cpath->custom_restrictinfo)
4247 0 : ADJUST_CHILD_ATTRS(cpath->custom_restrictinfo);
4248 0 : if (cpath->methods &&
4249 0 : cpath->methods->ReparameterizeCustomPathByChild)
4250 0 : cpath->custom_private =
4251 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4252 : cpath->custom_private,
4253 : child_rel);
4254 0 : new_path = (Path *) cpath;
4255 : }
4256 0 : break;
4257 :
4258 36 : case T_NestPath:
4259 : {
4260 36 : NestPath *npath = (NestPath *) path;
4261 36 : JoinPath *jpath = (JoinPath *) npath;
4262 :
4263 36 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4264 36 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4265 36 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4266 36 : new_path = (Path *) npath;
4267 : }
4268 36 : break;
4269 :
4270 0 : case T_MergePath:
4271 : {
4272 0 : MergePath *mpath = (MergePath *) path;
4273 0 : JoinPath *jpath = (JoinPath *) mpath;
4274 :
4275 0 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4276 0 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4277 0 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4278 0 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4279 0 : new_path = (Path *) mpath;
4280 : }
4281 0 : break;
4282 :
4283 48 : case T_HashPath:
4284 : {
4285 48 : HashPath *hpath = (HashPath *) path;
4286 48 : JoinPath *jpath = (JoinPath *) hpath;
4287 :
4288 48 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4289 48 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4290 48 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4291 48 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4292 48 : new_path = (Path *) hpath;
4293 : }
4294 48 : break;
4295 :
4296 24 : case T_AppendPath:
4297 : {
4298 24 : AppendPath *apath = (AppendPath *) path;
4299 :
4300 24 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4301 24 : new_path = (Path *) apath;
4302 : }
4303 24 : break;
4304 :
4305 0 : case T_MaterialPath:
4306 : {
4307 0 : MaterialPath *mpath = (MaterialPath *) path;
4308 :
4309 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4310 0 : new_path = (Path *) mpath;
4311 : }
4312 0 : break;
4313 :
4314 12 : case T_MemoizePath:
4315 : {
4316 12 : MemoizePath *mpath = (MemoizePath *) path;
4317 :
4318 12 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4319 12 : ADJUST_CHILD_ATTRS(mpath->param_exprs);
4320 12 : new_path = (Path *) mpath;
4321 : }
4322 12 : break;
4323 :
4324 0 : case T_GatherPath:
4325 : {
4326 0 : GatherPath *gpath = (GatherPath *) path;
4327 :
4328 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4329 0 : new_path = (Path *) gpath;
4330 : }
4331 0 : break;
4332 :
4333 0 : default:
4334 : /* We don't know how to reparameterize this path. */
4335 0 : return NULL;
4336 : }
4337 :
4338 : /*
4339 : * Adjust the parameterization information, which refers to the topmost
4340 : * parent. The topmost parent can be multiple levels away from the given
4341 : * child, hence use multi-level expression adjustment routines.
4342 : */
4343 888 : old_ppi = new_path->param_info;
4344 : required_outer =
4345 888 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4346 : child_rel,
4347 888 : child_rel->top_parent);
4348 :
4349 : /* If we already have a PPI for this parameterization, just return it */
4350 888 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4351 :
4352 : /*
4353 : * If not, build a new one and link it to the list of PPIs. For the same
4354 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4355 : * context the given RelOptInfo is in.
4356 : */
4357 888 : if (new_ppi == NULL)
4358 : {
4359 : MemoryContext oldcontext;
4360 780 : RelOptInfo *rel = path->parent;
4361 :
4362 780 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4363 :
4364 780 : new_ppi = makeNode(ParamPathInfo);
4365 780 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4366 780 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4367 780 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4368 780 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4369 780 : new_ppi->ppi_serials = bms_copy(old_ppi->ppi_serials);
4370 780 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4371 :
4372 780 : MemoryContextSwitchTo(oldcontext);
4373 : }
4374 888 : bms_free(required_outer);
4375 :
4376 888 : new_path->param_info = new_ppi;
4377 :
4378 : /*
4379 : * Adjust the path target if the parent of the outer relation is
4380 : * referenced in the targetlist. This can happen when only the parent of
4381 : * outer relation is laterally referenced in this relation.
4382 : */
4383 888 : if (bms_overlap(path->parent->lateral_relids,
4384 888 : child_rel->top_parent_relids))
4385 : {
4386 480 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4387 480 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4388 : }
4389 :
4390 888 : return new_path;
4391 : }
4392 :
4393 : /*
4394 : * path_is_reparameterizable_by_child
4395 : * Given a path parameterized by the parent of the given child relation,
4396 : * see if it can be translated to be parameterized by the child relation.
4397 : *
4398 : * This must return true if and only if reparameterize_path_by_child()
4399 : * would succeed on this path. Currently it's sufficient to verify that
4400 : * the path and all of its subpaths (if any) are of the types handled by
4401 : * that function. However, subpaths that are not parameterized can be
4402 : * disregarded since they won't require translation.
4403 : */
4404 : bool
4405 31188 : path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
4406 : {
4407 : #define REJECT_IF_PATH_NOT_REPARAMETERIZABLE(path) \
4408 : do { \
4409 : if (!path_is_reparameterizable_by_child(path, child_rel)) \
4410 : return false; \
4411 : } while(0)
4412 :
4413 : #define REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(pathlist) \
4414 : do { \
4415 : if (!pathlist_is_reparameterizable_by_child(pathlist, child_rel)) \
4416 : return false; \
4417 : } while(0)
4418 :
4419 : /*
4420 : * If the path is not parameterized by the parent of the given relation,
4421 : * it doesn't need reparameterization.
4422 : */
4423 31188 : if (!path->param_info ||
4424 30852 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4425 912 : return true;
4426 :
4427 : /*
4428 : * Check that the path type is one that reparameterize_path_by_child() can
4429 : * handle, and recursively check subpaths.
4430 : */
4431 30276 : switch (nodeTag(path))
4432 : {
4433 20208 : case T_Path:
4434 : case T_IndexPath:
4435 20208 : break;
4436 :
4437 48 : case T_BitmapHeapPath:
4438 : {
4439 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4440 :
4441 48 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(bhpath->bitmapqual);
4442 : }
4443 48 : break;
4444 :
4445 24 : case T_BitmapAndPath:
4446 : {
4447 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4448 :
4449 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bapath->bitmapquals);
4450 : }
4451 24 : break;
4452 :
4453 24 : case T_BitmapOrPath:
4454 : {
4455 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4456 :
4457 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bopath->bitmapquals);
4458 : }
4459 24 : break;
4460 :
4461 148 : case T_ForeignPath:
4462 : {
4463 148 : ForeignPath *fpath = (ForeignPath *) path;
4464 :
4465 148 : if (fpath->fdw_outerpath)
4466 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(fpath->fdw_outerpath);
4467 : }
4468 148 : break;
4469 :
4470 0 : case T_CustomPath:
4471 : {
4472 0 : CustomPath *cpath = (CustomPath *) path;
4473 :
4474 0 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(cpath->custom_paths);
4475 : }
4476 0 : break;
4477 :
4478 1176 : case T_NestPath:
4479 : case T_MergePath:
4480 : case T_HashPath:
4481 : {
4482 1176 : JoinPath *jpath = (JoinPath *) path;
4483 :
4484 1176 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->outerjoinpath);
4485 1176 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->innerjoinpath);
4486 : }
4487 1176 : break;
4488 :
4489 192 : case T_AppendPath:
4490 : {
4491 192 : AppendPath *apath = (AppendPath *) path;
4492 :
4493 192 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(apath->subpaths);
4494 : }
4495 192 : break;
4496 :
4497 0 : case T_MaterialPath:
4498 : {
4499 0 : MaterialPath *mpath = (MaterialPath *) path;
4500 :
4501 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4502 : }
4503 0 : break;
4504 :
4505 8456 : case T_MemoizePath:
4506 : {
4507 8456 : MemoizePath *mpath = (MemoizePath *) path;
4508 :
4509 8456 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4510 : }
4511 8456 : break;
4512 :
4513 0 : case T_GatherPath:
4514 : {
4515 0 : GatherPath *gpath = (GatherPath *) path;
4516 :
4517 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(gpath->subpath);
4518 : }
4519 0 : break;
4520 :
4521 0 : default:
4522 : /* We don't know how to reparameterize this path. */
4523 0 : return false;
4524 : }
4525 :
4526 30276 : return true;
4527 : }
4528 :
4529 : /*
4530 : * reparameterize_pathlist_by_child
4531 : * Helper function to reparameterize a list of paths by given child rel.
4532 : *
4533 : * Returns NIL to indicate failure, so pathlist had better not be NIL.
4534 : */
4535 : static List *
4536 72 : reparameterize_pathlist_by_child(PlannerInfo *root,
4537 : List *pathlist,
4538 : RelOptInfo *child_rel)
4539 : {
4540 : ListCell *lc;
4541 72 : List *result = NIL;
4542 :
4543 216 : foreach(lc, pathlist)
4544 : {
4545 144 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4546 : child_rel);
4547 :
4548 144 : if (path == NULL)
4549 : {
4550 0 : list_free(result);
4551 0 : return NIL;
4552 : }
4553 :
4554 144 : result = lappend(result, path);
4555 : }
4556 :
4557 72 : return result;
4558 : }
4559 :
4560 : /*
4561 : * pathlist_is_reparameterizable_by_child
4562 : * Helper function to check if a list of paths can be reparameterized.
4563 : */
4564 : static bool
4565 240 : pathlist_is_reparameterizable_by_child(List *pathlist, RelOptInfo *child_rel)
4566 : {
4567 : ListCell *lc;
4568 :
4569 720 : foreach(lc, pathlist)
4570 : {
4571 480 : Path *path = (Path *) lfirst(lc);
4572 :
4573 480 : if (!path_is_reparameterizable_by_child(path, child_rel))
4574 0 : return false;
4575 : }
4576 :
4577 240 : return true;
4578 : }
|