Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pathnode.c
4 : * Routines to manipulate pathlists and create path nodes
5 : *
6 : * Portions Copyright (c) 1996-2021, 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 "nodes/nodeFuncs.h"
23 : #include "optimizer/appendinfo.h"
24 : #include "optimizer/clauses.h"
25 : #include "optimizer/cost.h"
26 : #include "optimizer/optimizer.h"
27 : #include "optimizer/pathnode.h"
28 : #include "optimizer/paths.h"
29 : #include "optimizer/planmain.h"
30 : #include "optimizer/prep.h"
31 : #include "optimizer/restrictinfo.h"
32 : #include "optimizer/tlist.h"
33 : #include "parser/parsetree.h"
34 : #include "utils/lsyscache.h"
35 : #include "utils/memutils.h"
36 : #include "utils/selfuncs.h"
37 :
38 : typedef enum
39 : {
40 : COSTS_EQUAL, /* path costs are fuzzily equal */
41 : COSTS_BETTER1, /* first path is cheaper than second */
42 : COSTS_BETTER2, /* second path is cheaper than first */
43 : COSTS_DIFFERENT /* neither path dominates the other on cost */
44 : } PathCostComparison;
45 :
46 : /*
47 : * STD_FUZZ_FACTOR is the normal fuzz factor for compare_path_costs_fuzzily.
48 : * XXX is it worth making this user-controllable? It provides a tradeoff
49 : * between planner runtime and the accuracy of path cost comparisons.
50 : */
51 : #define STD_FUZZ_FACTOR 1.01
52 :
53 : static List *translate_sub_tlist(List *tlist, int relid);
54 : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
55 : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
56 : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
57 : List *pathlist,
58 : RelOptInfo *child_rel);
59 :
60 :
61 : /*****************************************************************************
62 : * MISC. PATH UTILITIES
63 : *****************************************************************************/
64 :
65 : /*
66 : * compare_path_costs
67 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
68 : * or more expensive than path2 for the specified criterion.
69 : */
70 : int
71 439796 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
72 : {
73 439796 : if (criterion == STARTUP_COST)
74 : {
75 224268 : if (path1->startup_cost < path2->startup_cost)
76 123094 : return -1;
77 101174 : if (path1->startup_cost > path2->startup_cost)
78 36032 : return +1;
79 :
80 : /*
81 : * If paths have the same startup cost (not at all unlikely), order
82 : * them by total cost.
83 : */
84 65142 : if (path1->total_cost < path2->total_cost)
85 42586 : return -1;
86 22556 : if (path1->total_cost > path2->total_cost)
87 2636 : return +1;
88 : }
89 : else
90 : {
91 215528 : if (path1->total_cost < path2->total_cost)
92 205646 : return -1;
93 9882 : if (path1->total_cost > path2->total_cost)
94 2158 : return +1;
95 :
96 : /*
97 : * If paths have the same total cost, order them by startup cost.
98 : */
99 7724 : if (path1->startup_cost < path2->startup_cost)
100 8 : return -1;
101 7716 : if (path1->startup_cost > path2->startup_cost)
102 0 : return +1;
103 : }
104 27636 : return 0;
105 : }
106 :
107 : /*
108 : * compare_path_fractional_costs
109 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
110 : * or more expensive than path2 for fetching the specified fraction
111 : * of the total tuples.
112 : *
113 : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
114 : * path with the cheaper total_cost.
115 : */
116 : int
117 2340 : compare_fractional_path_costs(Path *path1, Path *path2,
118 : double fraction)
119 : {
120 : Cost cost1,
121 : cost2;
122 :
123 2340 : if (fraction <= 0.0 || fraction >= 1.0)
124 1226 : return compare_path_costs(path1, path2, TOTAL_COST);
125 2228 : cost1 = path1->startup_cost +
126 1114 : fraction * (path1->total_cost - path1->startup_cost);
127 2228 : cost2 = path2->startup_cost +
128 1114 : fraction * (path2->total_cost - path2->startup_cost);
129 1114 : if (cost1 < cost2)
130 646 : return -1;
131 468 : if (cost1 > cost2)
132 468 : return +1;
133 0 : return 0;
134 : }
135 :
136 : /*
137 : * compare_path_costs_fuzzily
138 : * Compare the costs of two paths to see if either can be said to
139 : * dominate the other.
140 : *
141 : * We use fuzzy comparisons so that add_path() can avoid keeping both of
142 : * a pair of paths that really have insignificantly different cost.
143 : *
144 : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
145 : * fraction of the smaller cost that is considered to be a significant
146 : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
147 : * be 1% of the smaller cost.
148 : *
149 : * The two paths are said to have "equal" costs if both startup and total
150 : * costs are fuzzily the same. Path1 is said to be better than path2 if
151 : * it has fuzzily better startup cost and fuzzily no worse total cost,
152 : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
153 : * Path2 is better than path1 if the reverse holds. Finally, if one path
154 : * is fuzzily better than the other on startup cost and fuzzily worse on
155 : * total cost, we just say that their costs are "different", since neither
156 : * dominates the other across the whole performance spectrum.
157 : *
158 : * This function also enforces a policy rule that paths for which the relevant
159 : * one of parent->consider_startup and parent->consider_param_startup is false
160 : * cannot survive comparisons solely on the grounds of good startup cost, so
161 : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
162 : * (But if total costs are fuzzily equal, we compare startup costs anyway,
163 : * in hopes of eliminating one path or the other.)
164 : */
165 : static PathCostComparison
166 1932902 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
167 : {
168 : #define CONSIDER_PATH_STARTUP_COST(p) \
169 : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
170 :
171 : /*
172 : * Check total cost first since it's more likely to be different; many
173 : * paths have zero startup cost.
174 : */
175 1932902 : if (path1->total_cost > path2->total_cost * fuzz_factor)
176 : {
177 : /* path1 fuzzily worse on total cost */
178 921346 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
179 33676 : path2->startup_cost > path1->startup_cost * fuzz_factor)
180 : {
181 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
182 15848 : return COSTS_DIFFERENT;
183 : }
184 : /* else path2 dominates */
185 905498 : return COSTS_BETTER2;
186 : }
187 1011556 : if (path2->total_cost > path1->total_cost * fuzz_factor)
188 : {
189 : /* path2 fuzzily worse on total cost */
190 459218 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
191 13352 : path1->startup_cost > path2->startup_cost * fuzz_factor)
192 : {
193 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
194 7160 : return COSTS_DIFFERENT;
195 : }
196 : /* else path1 dominates */
197 452058 : return COSTS_BETTER1;
198 : }
199 : /* fuzzily the same on total cost ... */
200 552338 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
201 : {
202 : /* ... but path1 fuzzily worse on startup, so path2 wins */
203 209212 : return COSTS_BETTER2;
204 : }
205 343126 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
206 : {
207 : /* ... but path2 fuzzily worse on startup, so path1 wins */
208 46682 : return COSTS_BETTER1;
209 : }
210 : /* fuzzily the same on both costs */
211 296444 : return COSTS_EQUAL;
212 :
213 : #undef CONSIDER_PATH_STARTUP_COST
214 : }
215 :
216 : /*
217 : * set_cheapest
218 : * Find the minimum-cost paths from among a relation's paths,
219 : * and save them in the rel's cheapest-path fields.
220 : *
221 : * cheapest_total_path is normally the cheapest-total-cost unparameterized
222 : * path; but if there are no unparameterized paths, we assign it to be the
223 : * best (cheapest least-parameterized) parameterized path. However, only
224 : * unparameterized paths are considered candidates for cheapest_startup_path,
225 : * so that will be NULL if there are no unparameterized paths.
226 : *
227 : * The cheapest_parameterized_paths list collects all parameterized paths
228 : * that have survived the add_path() tournament for this relation. (Since
229 : * add_path ignores pathkeys for a parameterized path, these will be paths
230 : * that have best cost or best row count for their parameterization. We
231 : * may also have both a parallel-safe and a non-parallel-safe path in some
232 : * cases for the same parameterization in some cases, but this should be
233 : * relatively rare since, most typically, all paths for the same relation
234 : * will be parallel-safe or none of them will.)
235 : *
236 : * cheapest_parameterized_paths always includes the cheapest-total
237 : * unparameterized path, too, if there is one; the users of that list find
238 : * it more convenient if that's included.
239 : *
240 : * This is normally called only after we've finished constructing the path
241 : * list for the rel node.
242 : */
243 : void
244 1247452 : set_cheapest(RelOptInfo *parent_rel)
245 : {
246 : Path *cheapest_startup_path;
247 : Path *cheapest_total_path;
248 : Path *best_param_path;
249 : List *parameterized_paths;
250 : ListCell *p;
251 :
252 : Assert(IsA(parent_rel, RelOptInfo));
253 :
254 1247452 : if (parent_rel->pathlist == NIL)
255 0 : elog(ERROR, "could not devise a query plan for the given query");
256 :
257 1247452 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
258 1247452 : parameterized_paths = NIL;
259 :
260 2722888 : foreach(p, parent_rel->pathlist)
261 : {
262 1475436 : Path *path = (Path *) lfirst(p);
263 : int cmp;
264 :
265 1475436 : if (path->param_info)
266 : {
267 : /* Parameterized path, so add it to parameterized_paths */
268 59116 : parameterized_paths = lappend(parameterized_paths, path);
269 :
270 : /*
271 : * If we have an unparameterized cheapest-total, we no longer care
272 : * about finding the best parameterized path, so move on.
273 : */
274 59116 : if (cheapest_total_path)
275 13578 : continue;
276 :
277 : /*
278 : * Otherwise, track the best parameterized path, which is the one
279 : * with least total cost among those of the minimum
280 : * parameterization.
281 : */
282 45538 : if (best_param_path == NULL)
283 42604 : best_param_path = path;
284 : else
285 : {
286 2934 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
287 2934 : PATH_REQ_OUTER(best_param_path)))
288 : {
289 36 : case BMS_EQUAL:
290 : /* keep the cheaper one */
291 36 : if (compare_path_costs(path, best_param_path,
292 : TOTAL_COST) < 0)
293 0 : best_param_path = path;
294 36 : break;
295 88 : case BMS_SUBSET1:
296 : /* new path is less-parameterized */
297 88 : best_param_path = path;
298 88 : break;
299 70 : case BMS_SUBSET2:
300 : /* old path is less-parameterized, keep it */
301 70 : break;
302 2740 : case BMS_DIFFERENT:
303 :
304 : /*
305 : * This means that neither path has the least possible
306 : * parameterization for the rel. We'll sit on the old
307 : * path until something better comes along.
308 : */
309 2740 : break;
310 : }
311 45538 : }
312 : }
313 : else
314 : {
315 : /* Unparameterized path, so consider it for cheapest slots */
316 1416320 : if (cheapest_total_path == NULL)
317 : {
318 1245728 : cheapest_startup_path = cheapest_total_path = path;
319 1245728 : continue;
320 : }
321 :
322 : /*
323 : * If we find two paths of identical costs, try to keep the
324 : * better-sorted one. The paths might have unrelated sort
325 : * orderings, in which case we can only guess which might be
326 : * better to keep, but if one is superior then we definitely
327 : * should keep that one.
328 : */
329 170592 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
330 170592 : if (cmp > 0 ||
331 86 : (cmp == 0 &&
332 86 : compare_pathkeys(cheapest_startup_path->pathkeys,
333 : path->pathkeys) == PATHKEYS_BETTER2))
334 27138 : cheapest_startup_path = path;
335 :
336 170592 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
337 170592 : if (cmp > 0 ||
338 8 : (cmp == 0 &&
339 8 : compare_pathkeys(cheapest_total_path->pathkeys,
340 : path->pathkeys) == PATHKEYS_BETTER2))
341 0 : cheapest_total_path = path;
342 : }
343 : }
344 :
345 : /* Add cheapest unparameterized path, if any, to parameterized_paths */
346 1247452 : if (cheapest_total_path)
347 1245728 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
348 :
349 : /*
350 : * If there is no unparameterized path, use the best parameterized path as
351 : * cheapest_total_path (but not as cheapest_startup_path).
352 : */
353 1247452 : if (cheapest_total_path == NULL)
354 1724 : cheapest_total_path = best_param_path;
355 : Assert(cheapest_total_path != NULL);
356 :
357 1247452 : parent_rel->cheapest_startup_path = cheapest_startup_path;
358 1247452 : parent_rel->cheapest_total_path = cheapest_total_path;
359 1247452 : parent_rel->cheapest_unique_path = NULL; /* computed only if needed */
360 1247452 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
361 1247452 : }
362 :
363 : /*
364 : * add_path
365 : * Consider a potential implementation path for the specified parent rel,
366 : * and add it to the rel's pathlist if it is worthy of consideration.
367 : * A path is worthy if it has a better sort order (better pathkeys) or
368 : * cheaper cost (on either dimension), or generates fewer rows, than any
369 : * existing path that has the same or superset parameterization rels.
370 : * We also consider parallel-safe paths more worthy than others.
371 : *
372 : * We also remove from the rel's pathlist any old paths that are dominated
373 : * by new_path --- that is, new_path is cheaper, at least as well ordered,
374 : * generates no more rows, requires no outer rels not required by the old
375 : * path, and is no less parallel-safe.
376 : *
377 : * In most cases, a path with a superset parameterization will generate
378 : * fewer rows (since it has more join clauses to apply), so that those two
379 : * figures of merit move in opposite directions; this means that a path of
380 : * one parameterization can seldom dominate a path of another. But such
381 : * cases do arise, so we make the full set of checks anyway.
382 : *
383 : * There are two policy decisions embedded in this function, along with
384 : * its sibling add_path_precheck. First, we treat all parameterized paths
385 : * as having NIL pathkeys, so that they cannot win comparisons on the
386 : * basis of sort order. This is to reduce the number of parameterized
387 : * paths that are kept; see discussion in src/backend/optimizer/README.
388 : *
389 : * Second, we only consider cheap startup cost to be interesting if
390 : * parent_rel->consider_startup is true for an unparameterized path, or
391 : * parent_rel->consider_param_startup is true for a parameterized one.
392 : * Again, this allows discarding useless paths sooner.
393 : *
394 : * The pathlist is kept sorted by total_cost, with cheaper paths
395 : * at the front. Within this routine, that's simply a speed hack:
396 : * doing it that way makes it more likely that we will reject an inferior
397 : * path after a few comparisons, rather than many comparisons.
398 : * However, add_path_precheck relies on this ordering to exit early
399 : * when possible.
400 : *
401 : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
402 : * memory consumption. We dare not try to free the substructure of a Path,
403 : * since much of it may be shared with other Paths or the query tree itself;
404 : * but just recycling discarded Path nodes is a very useful savings in
405 : * a large join tree. We can recycle the List nodes of pathlist, too.
406 : *
407 : * As noted in optimizer/README, deleting a previously-accepted Path is
408 : * safe because we know that Paths of this rel cannot yet be referenced
409 : * from any other rel, such as a higher-level join. However, in some cases
410 : * it is possible that a Path is referenced by another Path for its own
411 : * rel; we must not delete such a Path, even if it is dominated by the new
412 : * Path. Currently this occurs only for IndexPath objects, which may be
413 : * referenced as children of BitmapHeapPaths as well as being paths in
414 : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
415 : *
416 : * 'parent_rel' is the relation entry to which the path corresponds.
417 : * 'new_path' is a potential path for parent_rel.
418 : *
419 : * Returns nothing, but modifies parent_rel->pathlist.
420 : */
421 : void
422 2254276 : add_path(RelOptInfo *parent_rel, Path *new_path)
423 : {
424 2254276 : bool accept_new = true; /* unless we find a superior old path */
425 2254276 : int insert_at = 0; /* where to insert new item */
426 : List *new_path_pathkeys;
427 : ListCell *p1;
428 :
429 : /*
430 : * This is a convenient place to check for query cancel --- no part of the
431 : * planner goes very long without calling add_path().
432 : */
433 2254276 : CHECK_FOR_INTERRUPTS();
434 :
435 : /* Pretend parameterized paths have no pathkeys, per comment above */
436 2254276 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
437 :
438 : /*
439 : * Loop to check proposed new path against old paths. Note it is possible
440 : * for more than one old path to be tossed out because new_path dominates
441 : * it.
442 : */
443 3191872 : foreach(p1, parent_rel->pathlist)
444 : {
445 1736122 : Path *old_path = (Path *) lfirst(p1);
446 1736122 : bool remove_old = false; /* unless new proves superior */
447 : PathCostComparison costcmp;
448 : PathKeysComparison keyscmp;
449 : BMS_Comparison outercmp;
450 :
451 : /*
452 : * Do a fuzzy cost comparison with standard fuzziness limit.
453 : */
454 1736122 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
455 : STD_FUZZ_FACTOR);
456 :
457 : /*
458 : * If the two paths compare differently for startup and total cost,
459 : * then we want to keep both, and we can skip comparing pathkeys and
460 : * required_outer rels. If they compare the same, proceed with the
461 : * other comparisons. Row count is checked last. (We make the tests
462 : * in this order because the cost comparison is most likely to turn
463 : * out "different", and the pathkeys comparison next most likely. As
464 : * explained above, row count very seldom makes a difference, so even
465 : * though it's cheap to compare there's not much point in checking it
466 : * earlier.)
467 : */
468 1736122 : if (costcmp != COSTS_DIFFERENT)
469 : {
470 : /* Similarly check to see if either dominates on pathkeys */
471 : List *old_path_pathkeys;
472 :
473 1713124 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
474 1713124 : keyscmp = compare_pathkeys(new_path_pathkeys,
475 : old_path_pathkeys);
476 1713124 : if (keyscmp != PATHKEYS_DIFFERENT)
477 : {
478 1643720 : switch (costcmp)
479 : {
480 209430 : case COSTS_EQUAL:
481 209430 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
482 209430 : PATH_REQ_OUTER(old_path));
483 209430 : if (keyscmp == PATHKEYS_BETTER1)
484 : {
485 1328 : if ((outercmp == BMS_EQUAL ||
486 1328 : outercmp == BMS_SUBSET1) &&
487 1328 : new_path->rows <= old_path->rows &&
488 1316 : new_path->parallel_safe >= old_path->parallel_safe)
489 1316 : remove_old = true; /* new dominates old */
490 : }
491 208102 : else if (keyscmp == PATHKEYS_BETTER2)
492 : {
493 8826 : if ((outercmp == BMS_EQUAL ||
494 8826 : outercmp == BMS_SUBSET2) &&
495 8826 : new_path->rows >= old_path->rows &&
496 8758 : new_path->parallel_safe <= old_path->parallel_safe)
497 8758 : accept_new = false; /* old dominates new */
498 : }
499 : else /* keyscmp == PATHKEYS_EQUAL */
500 : {
501 199276 : if (outercmp == BMS_EQUAL)
502 : {
503 : /*
504 : * Same pathkeys and outer rels, and fuzzily
505 : * the same cost, so keep just one; to decide
506 : * which, first check parallel-safety, then
507 : * rows, then do a fuzzy cost comparison with
508 : * very small fuzz limit. (We used to do an
509 : * exact cost comparison, but that results in
510 : * annoying platform-specific plan variations
511 : * due to roundoff in the cost estimates.) If
512 : * things are still tied, arbitrarily keep
513 : * only the old path. Notice that we will
514 : * keep only the old path even if the
515 : * less-fuzzy comparison decides the startup
516 : * and total costs compare differently.
517 : */
518 394048 : if (new_path->parallel_safe >
519 197024 : old_path->parallel_safe)
520 32 : remove_old = true; /* new dominates old */
521 393984 : else if (new_path->parallel_safe <
522 196992 : old_path->parallel_safe)
523 108 : accept_new = false; /* old dominates new */
524 196884 : else if (new_path->rows < old_path->rows)
525 36 : remove_old = true; /* new dominates old */
526 196848 : else if (new_path->rows > old_path->rows)
527 68 : accept_new = false; /* old dominates new */
528 196780 : else if (compare_path_costs_fuzzily(new_path,
529 : old_path,
530 : 1.0000000001) == COSTS_BETTER1)
531 4860 : remove_old = true; /* new dominates old */
532 : else
533 191920 : accept_new = false; /* old equals or
534 : * dominates new */
535 : }
536 2252 : else if (outercmp == BMS_SUBSET1 &&
537 1490 : new_path->rows <= old_path->rows &&
538 1478 : new_path->parallel_safe >= old_path->parallel_safe)
539 1478 : remove_old = true; /* new dominates old */
540 774 : else if (outercmp == BMS_SUBSET2 &&
541 644 : new_path->rows >= old_path->rows &&
542 560 : new_path->parallel_safe <= old_path->parallel_safe)
543 560 : accept_new = false; /* old dominates new */
544 : /* else different parameterizations, keep both */
545 : }
546 209430 : break;
547 481192 : case COSTS_BETTER1:
548 481192 : if (keyscmp != PATHKEYS_BETTER2)
549 : {
550 333940 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
551 333940 : PATH_REQ_OUTER(old_path));
552 333940 : if ((outercmp == BMS_EQUAL ||
553 289410 : outercmp == BMS_SUBSET1) &&
554 289410 : new_path->rows <= old_path->rows &&
555 287670 : new_path->parallel_safe >= old_path->parallel_safe)
556 286142 : remove_old = true; /* new dominates old */
557 : }
558 481192 : break;
559 953098 : case COSTS_BETTER2:
560 953098 : if (keyscmp != PATHKEYS_BETTER1)
561 : {
562 660060 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
563 660060 : PATH_REQ_OUTER(old_path));
564 660060 : if ((outercmp == BMS_EQUAL ||
565 629704 : outercmp == BMS_SUBSET2) &&
566 629704 : new_path->rows >= old_path->rows &&
567 598536 : new_path->parallel_safe <= old_path->parallel_safe)
568 597112 : accept_new = false; /* old dominates new */
569 : }
570 953098 : break;
571 0 : case COSTS_DIFFERENT:
572 :
573 : /*
574 : * can't get here, but keep this case to keep compiler
575 : * quiet
576 : */
577 0 : break;
578 : }
579 92402 : }
580 : }
581 :
582 : /*
583 : * Remove current element from pathlist if dominated by new.
584 : */
585 1736122 : if (remove_old)
586 : {
587 293864 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
588 : p1);
589 :
590 : /*
591 : * Delete the data pointed-to by the deleted cell, if possible
592 : */
593 293864 : if (!IsA(old_path, IndexPath))
594 280364 : pfree(old_path);
595 : }
596 : else
597 : {
598 : /* new belongs after this old path if it has cost >= old's */
599 1442258 : if (new_path->total_cost >= old_path->total_cost)
600 1191304 : insert_at = foreach_current_index(p1) + 1;
601 : }
602 :
603 : /*
604 : * If we found an old path that dominates new_path, we can quit
605 : * scanning the pathlist; we will not add new_path, and we assume
606 : * new_path cannot dominate any other elements of the pathlist.
607 : */
608 1736122 : if (!accept_new)
609 798526 : break;
610 : }
611 :
612 2254276 : if (accept_new)
613 : {
614 : /* Accept the new path: insert it at proper place in pathlist */
615 1455750 : parent_rel->pathlist =
616 1455750 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
617 : }
618 : else
619 : {
620 : /* Reject and recycle the new path */
621 798526 : if (!IsA(new_path, IndexPath))
622 736626 : pfree(new_path);
623 : }
624 2254276 : }
625 :
626 : /*
627 : * add_path_precheck
628 : * Check whether a proposed new path could possibly get accepted.
629 : * We assume we know the path's pathkeys and parameterization accurately,
630 : * and have lower bounds for its costs.
631 : *
632 : * Note that we do not know the path's rowcount, since getting an estimate for
633 : * that is too expensive to do before prechecking. We assume here that paths
634 : * of a superset parameterization will generate fewer rows; if that holds,
635 : * then paths with different parameterizations cannot dominate each other
636 : * and so we can simply ignore existing paths of another parameterization.
637 : * (In the infrequent cases where that rule of thumb fails, add_path will
638 : * get rid of the inferior path.)
639 : *
640 : * At the time this is called, we haven't actually built a Path structure,
641 : * so the required information has to be passed piecemeal.
642 : */
643 : bool
644 1933632 : add_path_precheck(RelOptInfo *parent_rel,
645 : Cost startup_cost, Cost total_cost,
646 : List *pathkeys, Relids required_outer)
647 : {
648 : List *new_path_pathkeys;
649 : bool consider_startup;
650 : ListCell *p1;
651 :
652 : /* Pretend parameterized paths have no pathkeys, per add_path policy */
653 1933632 : new_path_pathkeys = required_outer ? NIL : pathkeys;
654 :
655 : /* Decide whether new path's startup cost is interesting */
656 1933632 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
657 :
658 2394594 : foreach(p1, parent_rel->pathlist)
659 : {
660 2259008 : Path *old_path = (Path *) lfirst(p1);
661 : PathKeysComparison keyscmp;
662 :
663 : /*
664 : * We are looking for an old_path with the same parameterization (and
665 : * by assumption the same rowcount) that dominates the new path on
666 : * pathkeys as well as both cost metrics. If we find one, we can
667 : * reject the new path.
668 : *
669 : * Cost comparisons here should match compare_path_costs_fuzzily.
670 : */
671 2259008 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
672 : {
673 : /* new path can win on startup cost only if consider_startup */
674 1486510 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
675 727140 : !consider_startup)
676 : {
677 : /* new path loses on cost, so check pathkeys... */
678 : List *old_path_pathkeys;
679 :
680 1472128 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
681 1472128 : keyscmp = compare_pathkeys(new_path_pathkeys,
682 : old_path_pathkeys);
683 1472128 : if (keyscmp == PATHKEYS_EQUAL ||
684 : keyscmp == PATHKEYS_BETTER2)
685 : {
686 : /* new path does not win on pathkeys... */
687 1036192 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
688 : {
689 : /* Found an old path that dominates the new one */
690 1025548 : return false;
691 : }
692 : }
693 : }
694 : }
695 : else
696 : {
697 : /*
698 : * Since the pathlist is sorted by total_cost, we can stop looking
699 : * once we reach a path with a total_cost larger than the new
700 : * path's.
701 : */
702 772498 : break;
703 : }
704 : }
705 :
706 908084 : return true;
707 : }
708 :
709 : /*
710 : * add_partial_path
711 : * Like add_path, our goal here is to consider whether a path is worthy
712 : * of being kept around, but the considerations here are a bit different.
713 : * A partial path is one which can be executed in any number of workers in
714 : * parallel such that each worker will generate a subset of the path's
715 : * overall result.
716 : *
717 : * As in add_path, the partial_pathlist is kept sorted with the cheapest
718 : * total path in front. This is depended on by multiple places, which
719 : * just take the front entry as the cheapest path without searching.
720 : *
721 : * We don't generate parameterized partial paths for several reasons. Most
722 : * importantly, they're not safe to execute, because there's nothing to
723 : * make sure that a parallel scan within the parameterized portion of the
724 : * plan is running with the same value in every worker at the same time.
725 : * Fortunately, it seems unlikely to be worthwhile anyway, because having
726 : * each worker scan the entire outer relation and a subset of the inner
727 : * relation will generally be a terrible plan. The inner (parameterized)
728 : * side of the plan will be small anyway. There could be rare cases where
729 : * this wins big - e.g. if join order constraints put a 1-row relation on
730 : * the outer side of the topmost join with a parameterized plan on the inner
731 : * side - but we'll have to be content not to handle such cases until
732 : * somebody builds an executor infrastructure that can cope with them.
733 : *
734 : * Because we don't consider parameterized paths here, we also don't
735 : * need to consider the row counts as a measure of quality: every path will
736 : * produce the same number of rows. Neither do we need to consider startup
737 : * costs: parallelism is only used for plans that will be run to completion.
738 : * Therefore, this routine is much simpler than add_path: it needs to
739 : * consider only pathkeys and total cost.
740 : *
741 : * As with add_path, we pfree paths that are found to be dominated by
742 : * another partial path; this requires that there be no other references to
743 : * such paths yet. Hence, GatherPaths must not be created for a rel until
744 : * we're done creating all partial paths for it. Unlike add_path, we don't
745 : * take an exception for IndexPaths as partial index paths won't be
746 : * referenced by partial BitmapHeapPaths.
747 : */
748 : void
749 58066 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
750 : {
751 58066 : bool accept_new = true; /* unless we find a superior old path */
752 58066 : int insert_at = 0; /* where to insert new item */
753 : ListCell *p1;
754 :
755 : /* Check for query cancel. */
756 58066 : CHECK_FOR_INTERRUPTS();
757 :
758 : /* Path to be added must be parallel safe. */
759 : Assert(new_path->parallel_safe);
760 :
761 : /* Relation should be OK for parallelism, too. */
762 : Assert(parent_rel->consider_parallel);
763 :
764 : /*
765 : * As in add_path, throw out any paths which are dominated by the new
766 : * path, but throw out the new path if some existing path dominates it.
767 : */
768 77262 : foreach(p1, parent_rel->partial_pathlist)
769 : {
770 27554 : Path *old_path = (Path *) lfirst(p1);
771 27554 : bool remove_old = false; /* unless new proves superior */
772 : PathKeysComparison keyscmp;
773 :
774 : /* Compare pathkeys. */
775 27554 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
776 :
777 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
778 27554 : if (keyscmp != PATHKEYS_DIFFERENT)
779 : {
780 27426 : if (new_path->total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
781 : {
782 : /* New path costs more; keep it only if pathkeys are better. */
783 8896 : if (keyscmp != PATHKEYS_BETTER1)
784 3634 : accept_new = false;
785 : }
786 37060 : else if (old_path->total_cost > new_path->total_cost
787 18530 : * STD_FUZZ_FACTOR)
788 : {
789 : /* Old path costs more; keep it only if pathkeys are better. */
790 13658 : if (keyscmp != PATHKEYS_BETTER2)
791 7444 : remove_old = true;
792 : }
793 4872 : else if (keyscmp == PATHKEYS_BETTER1)
794 : {
795 : /* Costs are about the same, new path has better pathkeys. */
796 0 : remove_old = true;
797 : }
798 4872 : else if (keyscmp == PATHKEYS_BETTER2)
799 : {
800 : /* Costs are about the same, old path has better pathkeys. */
801 1056 : accept_new = false;
802 : }
803 3816 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
804 : {
805 : /* Pathkeys are the same, and the old path costs more. */
806 148 : remove_old = true;
807 : }
808 : else
809 : {
810 : /*
811 : * Pathkeys are the same, and new path isn't materially
812 : * cheaper.
813 : */
814 3668 : accept_new = false;
815 : }
816 : }
817 :
818 : /*
819 : * Remove current element from partial_pathlist if dominated by new.
820 : */
821 27554 : if (remove_old)
822 : {
823 7592 : parent_rel->partial_pathlist =
824 7592 : foreach_delete_current(parent_rel->partial_pathlist, p1);
825 7592 : pfree(old_path);
826 : }
827 : else
828 : {
829 : /* new belongs after this old path if it has cost >= old's */
830 19962 : if (new_path->total_cost >= old_path->total_cost)
831 13508 : insert_at = foreach_current_index(p1) + 1;
832 : }
833 :
834 : /*
835 : * If we found an old path that dominates new_path, we can quit
836 : * scanning the partial_pathlist; we will not add new_path, and we
837 : * assume new_path cannot dominate any later path.
838 : */
839 27554 : if (!accept_new)
840 8358 : break;
841 : }
842 :
843 58066 : if (accept_new)
844 : {
845 : /* Accept the new path: insert it at proper place */
846 49708 : parent_rel->partial_pathlist =
847 49708 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
848 : }
849 : else
850 : {
851 : /* Reject and recycle the new path */
852 8358 : pfree(new_path);
853 : }
854 58066 : }
855 :
856 : /*
857 : * add_partial_path_precheck
858 : * Check whether a proposed new partial path could possibly get accepted.
859 : *
860 : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
861 : * since they don't matter for partial paths (see add_partial_path). But
862 : * we do want to make sure we don't add a partial path if there's already
863 : * a complete path that dominates it, since in that case the proposed path
864 : * is surely a loser.
865 : */
866 : bool
867 34232 : add_partial_path_precheck(RelOptInfo *parent_rel, Cost total_cost,
868 : List *pathkeys)
869 : {
870 : ListCell *p1;
871 :
872 : /*
873 : * Our goal here is twofold. First, we want to find out whether this path
874 : * is clearly inferior to some existing partial path. If so, we want to
875 : * reject it immediately. Second, we want to find out whether this path
876 : * is clearly superior to some existing partial path -- at least, modulo
877 : * final cost computations. If so, we definitely want to consider it.
878 : *
879 : * Unlike add_path(), we always compare pathkeys here. This is because we
880 : * expect partial_pathlist to be very short, and getting a definitive
881 : * answer at this stage avoids the need to call add_path_precheck.
882 : */
883 47960 : foreach(p1, parent_rel->partial_pathlist)
884 : {
885 38680 : Path *old_path = (Path *) lfirst(p1);
886 : PathKeysComparison keyscmp;
887 :
888 38680 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
889 38680 : if (keyscmp != PATHKEYS_DIFFERENT)
890 : {
891 38552 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
892 : keyscmp != PATHKEYS_BETTER1)
893 24952 : return false;
894 18832 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
895 : keyscmp != PATHKEYS_BETTER2)
896 5232 : return true;
897 : }
898 : }
899 :
900 : /*
901 : * This path is neither clearly inferior to an existing partial path nor
902 : * clearly good enough that it might replace one. Compare it to
903 : * non-parallel plans. If it loses even before accounting for the cost of
904 : * the Gather node, we should definitely reject it.
905 : *
906 : * Note that we pass the total_cost to add_path_precheck twice. This is
907 : * because it's never advantageous to consider the startup cost of a
908 : * partial path; the resulting plans, if run in parallel, will be run to
909 : * completion.
910 : */
911 9280 : if (!add_path_precheck(parent_rel, total_cost, total_cost, pathkeys,
912 : NULL))
913 252 : return false;
914 :
915 9028 : return true;
916 : }
917 :
918 :
919 : /*****************************************************************************
920 : * PATH NODE CREATION ROUTINES
921 : *****************************************************************************/
922 :
923 : /*
924 : * create_seqscan_path
925 : * Creates a path corresponding to a sequential scan, returning the
926 : * pathnode.
927 : */
928 : Path *
929 253502 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
930 : Relids required_outer, int parallel_workers)
931 : {
932 253502 : Path *pathnode = makeNode(Path);
933 :
934 253502 : pathnode->pathtype = T_SeqScan;
935 253502 : pathnode->parent = rel;
936 253502 : pathnode->pathtarget = rel->reltarget;
937 253502 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
938 : required_outer);
939 253502 : pathnode->parallel_aware = parallel_workers > 0 ? true : false;
940 253502 : pathnode->parallel_safe = rel->consider_parallel;
941 253502 : pathnode->parallel_workers = parallel_workers;
942 253502 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
943 :
944 253502 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
945 :
946 253502 : return pathnode;
947 : }
948 :
949 : /*
950 : * create_samplescan_path
951 : * Creates a path node for a sampled table scan.
952 : */
953 : Path *
954 180 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
955 : {
956 180 : Path *pathnode = makeNode(Path);
957 :
958 180 : pathnode->pathtype = T_SampleScan;
959 180 : pathnode->parent = rel;
960 180 : pathnode->pathtarget = rel->reltarget;
961 180 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
962 : required_outer);
963 180 : pathnode->parallel_aware = false;
964 180 : pathnode->parallel_safe = rel->consider_parallel;
965 180 : pathnode->parallel_workers = 0;
966 180 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
967 :
968 180 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
969 :
970 180 : return pathnode;
971 : }
972 :
973 : /*
974 : * create_index_path
975 : * Creates a path node for an index scan.
976 : *
977 : * 'index' is a usable index.
978 : * 'indexclauses' is a list of IndexClause nodes representing clauses
979 : * to be enforced as qual conditions in the scan.
980 : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
981 : * to be used as index ordering operators in the scan.
982 : * 'indexorderbycols' is an integer list of index column numbers (zero based)
983 : * the ordering operators can be used with.
984 : * 'pathkeys' describes the ordering of the path.
985 : * 'indexscandir' is ForwardScanDirection or BackwardScanDirection
986 : * for an ordered index, or NoMovementScanDirection for
987 : * an unordered index.
988 : * 'indexonly' is true if an index-only scan is wanted.
989 : * 'required_outer' is the set of outer relids for a parameterized path.
990 : * 'loop_count' is the number of repetitions of the indexscan to factor into
991 : * estimates of caching behavior.
992 : * 'partial_path' is true if constructing a parallel index scan path.
993 : *
994 : * Returns the new path node.
995 : */
996 : IndexPath *
997 404998 : create_index_path(PlannerInfo *root,
998 : IndexOptInfo *index,
999 : List *indexclauses,
1000 : List *indexorderbys,
1001 : List *indexorderbycols,
1002 : List *pathkeys,
1003 : ScanDirection indexscandir,
1004 : bool indexonly,
1005 : Relids required_outer,
1006 : double loop_count,
1007 : bool partial_path)
1008 : {
1009 404998 : IndexPath *pathnode = makeNode(IndexPath);
1010 404998 : RelOptInfo *rel = index->rel;
1011 :
1012 404998 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1013 404998 : pathnode->path.parent = rel;
1014 404998 : pathnode->path.pathtarget = rel->reltarget;
1015 404998 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1016 : required_outer);
1017 404998 : pathnode->path.parallel_aware = false;
1018 404998 : pathnode->path.parallel_safe = rel->consider_parallel;
1019 404998 : pathnode->path.parallel_workers = 0;
1020 404998 : pathnode->path.pathkeys = pathkeys;
1021 :
1022 404998 : pathnode->indexinfo = index;
1023 404998 : pathnode->indexclauses = indexclauses;
1024 404998 : pathnode->indexorderbys = indexorderbys;
1025 404998 : pathnode->indexorderbycols = indexorderbycols;
1026 404998 : pathnode->indexscandir = indexscandir;
1027 :
1028 404998 : cost_index(pathnode, root, loop_count, partial_path);
1029 :
1030 404998 : return pathnode;
1031 : }
1032 :
1033 : /*
1034 : * create_bitmap_heap_path
1035 : * Creates a path node for a bitmap scan.
1036 : *
1037 : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1038 : * 'required_outer' is the set of outer relids for a parameterized path.
1039 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1040 : * estimates of caching behavior.
1041 : *
1042 : * loop_count should match the value used when creating the component
1043 : * IndexPaths.
1044 : */
1045 : BitmapHeapPath *
1046 191250 : create_bitmap_heap_path(PlannerInfo *root,
1047 : RelOptInfo *rel,
1048 : Path *bitmapqual,
1049 : Relids required_outer,
1050 : double loop_count,
1051 : int parallel_degree)
1052 : {
1053 191250 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1054 :
1055 191250 : pathnode->path.pathtype = T_BitmapHeapScan;
1056 191250 : pathnode->path.parent = rel;
1057 191250 : pathnode->path.pathtarget = rel->reltarget;
1058 191250 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1059 : required_outer);
1060 191250 : pathnode->path.parallel_aware = parallel_degree > 0 ? true : false;
1061 191250 : pathnode->path.parallel_safe = rel->consider_parallel;
1062 191250 : pathnode->path.parallel_workers = parallel_degree;
1063 191250 : pathnode->path.pathkeys = NIL; /* always unordered */
1064 :
1065 191250 : pathnode->bitmapqual = bitmapqual;
1066 :
1067 191250 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1068 : pathnode->path.param_info,
1069 : bitmapqual, loop_count);
1070 :
1071 191250 : return pathnode;
1072 : }
1073 :
1074 : /*
1075 : * create_bitmap_and_path
1076 : * Creates a path node representing a BitmapAnd.
1077 : */
1078 : BitmapAndPath *
1079 19658 : create_bitmap_and_path(PlannerInfo *root,
1080 : RelOptInfo *rel,
1081 : List *bitmapquals)
1082 : {
1083 19658 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1084 19658 : Relids required_outer = NULL;
1085 : ListCell *lc;
1086 :
1087 19658 : pathnode->path.pathtype = T_BitmapAnd;
1088 19658 : pathnode->path.parent = rel;
1089 19658 : pathnode->path.pathtarget = rel->reltarget;
1090 :
1091 : /*
1092 : * Identify the required outer rels as the union of what the child paths
1093 : * depend on. (Alternatively, we could insist that the caller pass this
1094 : * in, but it's more convenient and reliable to compute it here.)
1095 : */
1096 58974 : foreach(lc, bitmapquals)
1097 : {
1098 39316 : Path *bitmapqual = (Path *) lfirst(lc);
1099 :
1100 39316 : required_outer = bms_add_members(required_outer,
1101 39316 : PATH_REQ_OUTER(bitmapqual));
1102 : }
1103 19658 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1104 : required_outer);
1105 :
1106 : /*
1107 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1108 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1109 : * set the flag for this path based only on the relation-level flag,
1110 : * without actually iterating over the list of children.
1111 : */
1112 19658 : pathnode->path.parallel_aware = false;
1113 19658 : pathnode->path.parallel_safe = rel->consider_parallel;
1114 19658 : pathnode->path.parallel_workers = 0;
1115 :
1116 19658 : pathnode->path.pathkeys = NIL; /* always unordered */
1117 :
1118 19658 : pathnode->bitmapquals = bitmapquals;
1119 :
1120 : /* this sets bitmapselectivity as well as the regular cost fields: */
1121 19658 : cost_bitmap_and_node(pathnode, root);
1122 :
1123 19658 : return pathnode;
1124 : }
1125 :
1126 : /*
1127 : * create_bitmap_or_path
1128 : * Creates a path node representing a BitmapOr.
1129 : */
1130 : BitmapOrPath *
1131 422 : create_bitmap_or_path(PlannerInfo *root,
1132 : RelOptInfo *rel,
1133 : List *bitmapquals)
1134 : {
1135 422 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1136 422 : Relids required_outer = NULL;
1137 : ListCell *lc;
1138 :
1139 422 : pathnode->path.pathtype = T_BitmapOr;
1140 422 : pathnode->path.parent = rel;
1141 422 : pathnode->path.pathtarget = rel->reltarget;
1142 :
1143 : /*
1144 : * Identify the required outer rels as the union of what the child paths
1145 : * depend on. (Alternatively, we could insist that the caller pass this
1146 : * in, but it's more convenient and reliable to compute it here.)
1147 : */
1148 1306 : foreach(lc, bitmapquals)
1149 : {
1150 884 : Path *bitmapqual = (Path *) lfirst(lc);
1151 :
1152 884 : required_outer = bms_add_members(required_outer,
1153 884 : PATH_REQ_OUTER(bitmapqual));
1154 : }
1155 422 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1156 : required_outer);
1157 :
1158 : /*
1159 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1160 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1161 : * set the flag for this path based only on the relation-level flag,
1162 : * without actually iterating over the list of children.
1163 : */
1164 422 : pathnode->path.parallel_aware = false;
1165 422 : pathnode->path.parallel_safe = rel->consider_parallel;
1166 422 : pathnode->path.parallel_workers = 0;
1167 :
1168 422 : pathnode->path.pathkeys = NIL; /* always unordered */
1169 :
1170 422 : pathnode->bitmapquals = bitmapquals;
1171 :
1172 : /* this sets bitmapselectivity as well as the regular cost fields: */
1173 422 : cost_bitmap_or_node(pathnode, root);
1174 :
1175 422 : return pathnode;
1176 : }
1177 :
1178 : /*
1179 : * create_tidscan_path
1180 : * Creates a path corresponding to a scan by TID, returning the pathnode.
1181 : */
1182 : TidPath *
1183 620 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1184 : Relids required_outer)
1185 : {
1186 620 : TidPath *pathnode = makeNode(TidPath);
1187 :
1188 620 : pathnode->path.pathtype = T_TidScan;
1189 620 : pathnode->path.parent = rel;
1190 620 : pathnode->path.pathtarget = rel->reltarget;
1191 620 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1192 : required_outer);
1193 620 : pathnode->path.parallel_aware = false;
1194 620 : pathnode->path.parallel_safe = rel->consider_parallel;
1195 620 : pathnode->path.parallel_workers = 0;
1196 620 : pathnode->path.pathkeys = NIL; /* always unordered */
1197 :
1198 620 : pathnode->tidquals = tidquals;
1199 :
1200 620 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1201 : pathnode->path.param_info);
1202 :
1203 620 : return pathnode;
1204 : }
1205 :
1206 : /*
1207 : * create_append_path
1208 : * Creates a path corresponding to an Append plan, returning the
1209 : * pathnode.
1210 : *
1211 : * Note that we must handle subpaths = NIL, representing a dummy access path.
1212 : * Also, there are callers that pass root = NULL.
1213 : */
1214 : AppendPath *
1215 43046 : create_append_path(PlannerInfo *root,
1216 : RelOptInfo *rel,
1217 : List *subpaths, List *partial_subpaths,
1218 : List *pathkeys, Relids required_outer,
1219 : int parallel_workers, bool parallel_aware,
1220 : List *partitioned_rels, double rows)
1221 : {
1222 43046 : AppendPath *pathnode = makeNode(AppendPath);
1223 : ListCell *l;
1224 :
1225 : Assert(!parallel_aware || parallel_workers > 0);
1226 :
1227 43046 : pathnode->path.pathtype = T_Append;
1228 43046 : pathnode->path.parent = rel;
1229 43046 : pathnode->path.pathtarget = rel->reltarget;
1230 :
1231 : /*
1232 : * When generating an Append path for a partitioned table, there may be
1233 : * parameters that are useful so we can eliminate certain partitions
1234 : * during execution. Here we'll go all the way and fully populate the
1235 : * parameter info data as we do for normal base relations. However, we
1236 : * need only bother doing this for RELOPT_BASEREL rels, as
1237 : * RELOPT_OTHER_MEMBER_REL's Append paths are merged into the base rel's
1238 : * Append subpaths. It would do no harm to do this, we just avoid it to
1239 : * save wasting effort.
1240 : */
1241 43046 : if (partitioned_rels != NIL && root && rel->reloptkind == RELOPT_BASEREL)
1242 24696 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1243 : rel,
1244 : required_outer);
1245 : else
1246 18350 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1247 : required_outer);
1248 :
1249 43046 : pathnode->path.parallel_aware = parallel_aware;
1250 43046 : pathnode->path.parallel_safe = rel->consider_parallel;
1251 43046 : pathnode->path.parallel_workers = parallel_workers;
1252 43046 : pathnode->path.pathkeys = pathkeys;
1253 43046 : pathnode->partitioned_rels = list_copy(partitioned_rels);
1254 :
1255 : /*
1256 : * For parallel append, non-partial paths are sorted by descending total
1257 : * costs. That way, the total time to finish all non-partial paths is
1258 : * minimized. Also, the partial paths are sorted by descending startup
1259 : * costs. There may be some paths that require to do startup work by a
1260 : * single worker. In such case, it's better for workers to choose the
1261 : * expensive ones first, whereas the leader should choose the cheapest
1262 : * startup plan.
1263 : */
1264 43046 : if (pathnode->path.parallel_aware)
1265 : {
1266 : /*
1267 : * We mustn't fiddle with the order of subpaths when the Append has
1268 : * pathkeys. The order they're listed in is critical to keeping the
1269 : * pathkeys valid.
1270 : */
1271 : Assert(pathkeys == NIL);
1272 :
1273 15884 : list_sort(subpaths, append_total_cost_compare);
1274 15884 : list_sort(partial_subpaths, append_startup_cost_compare);
1275 : }
1276 43046 : pathnode->first_partial_path = list_length(subpaths);
1277 43046 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1278 :
1279 : /*
1280 : * Apply query-wide LIMIT if known and path is for sole base relation.
1281 : * (Handling this at this low level is a bit klugy.)
1282 : */
1283 43046 : if (root != NULL && bms_equal(rel->relids, root->all_baserels))
1284 26378 : pathnode->limit_tuples = root->limit_tuples;
1285 : else
1286 16668 : pathnode->limit_tuples = -1.0;
1287 :
1288 137352 : foreach(l, pathnode->subpaths)
1289 : {
1290 94306 : Path *subpath = (Path *) lfirst(l);
1291 :
1292 172398 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1293 78092 : subpath->parallel_safe;
1294 :
1295 : /* All child paths must have same parameterization */
1296 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1297 : }
1298 :
1299 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1300 :
1301 : /*
1302 : * If there's exactly one child path, the Append is a no-op and will be
1303 : * discarded later (in setrefs.c); therefore, we can inherit the child's
1304 : * size and cost, as well as its pathkeys if any (overriding whatever the
1305 : * caller might've said). Otherwise, we must do the normal costsize
1306 : * calculation.
1307 : */
1308 43046 : if (list_length(pathnode->subpaths) == 1)
1309 : {
1310 15686 : Path *child = (Path *) linitial(pathnode->subpaths);
1311 :
1312 15686 : pathnode->path.rows = child->rows;
1313 15686 : pathnode->path.startup_cost = child->startup_cost;
1314 15686 : pathnode->path.total_cost = child->total_cost;
1315 15686 : pathnode->path.pathkeys = child->pathkeys;
1316 : }
1317 : else
1318 27360 : cost_append(pathnode);
1319 :
1320 : /* If the caller provided a row estimate, override the computed value. */
1321 43046 : if (rows >= 0)
1322 348 : pathnode->path.rows = rows;
1323 :
1324 43046 : return pathnode;
1325 : }
1326 :
1327 : /*
1328 : * append_total_cost_compare
1329 : * list_sort comparator for sorting append child paths
1330 : * by total_cost descending
1331 : *
1332 : * For equal total costs, we fall back to comparing startup costs; if those
1333 : * are equal too, break ties using bms_compare on the paths' relids.
1334 : * (This is to avoid getting unpredictable results from list_sort.)
1335 : */
1336 : static int
1337 2636 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1338 : {
1339 2636 : Path *path1 = (Path *) lfirst(a);
1340 2636 : Path *path2 = (Path *) lfirst(b);
1341 : int cmp;
1342 :
1343 2636 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1344 2636 : if (cmp != 0)
1345 2410 : return -cmp;
1346 226 : return bms_compare(path1->parent->relids, path2->parent->relids);
1347 : }
1348 :
1349 : /*
1350 : * append_startup_cost_compare
1351 : * list_sort comparator for sorting append child paths
1352 : * by startup_cost descending
1353 : *
1354 : * For equal startup costs, we fall back to comparing total costs; if those
1355 : * are equal too, break ties using bms_compare on the paths' relids.
1356 : * (This is to avoid getting unpredictable results from list_sort.)
1357 : */
1358 : static int
1359 19574 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1360 : {
1361 19574 : Path *path1 = (Path *) lfirst(a);
1362 19574 : Path *path2 = (Path *) lfirst(b);
1363 : int cmp;
1364 :
1365 19574 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1366 19574 : if (cmp != 0)
1367 7682 : return -cmp;
1368 11892 : return bms_compare(path1->parent->relids, path2->parent->relids);
1369 : }
1370 :
1371 : /*
1372 : * create_merge_append_path
1373 : * Creates a path corresponding to a MergeAppend plan, returning the
1374 : * pathnode.
1375 : */
1376 : MergeAppendPath *
1377 2334 : create_merge_append_path(PlannerInfo *root,
1378 : RelOptInfo *rel,
1379 : List *subpaths,
1380 : List *pathkeys,
1381 : Relids required_outer,
1382 : List *partitioned_rels)
1383 : {
1384 2334 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1385 : Cost input_startup_cost;
1386 : Cost input_total_cost;
1387 : ListCell *l;
1388 :
1389 2334 : pathnode->path.pathtype = T_MergeAppend;
1390 2334 : pathnode->path.parent = rel;
1391 2334 : pathnode->path.pathtarget = rel->reltarget;
1392 2334 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1393 : required_outer);
1394 2334 : pathnode->path.parallel_aware = false;
1395 2334 : pathnode->path.parallel_safe = rel->consider_parallel;
1396 2334 : pathnode->path.parallel_workers = 0;
1397 2334 : pathnode->path.pathkeys = pathkeys;
1398 2334 : pathnode->partitioned_rels = list_copy(partitioned_rels);
1399 2334 : pathnode->subpaths = subpaths;
1400 :
1401 : /*
1402 : * Apply query-wide LIMIT if known and path is for sole base relation.
1403 : * (Handling this at this low level is a bit klugy.)
1404 : */
1405 2334 : if (bms_equal(rel->relids, root->all_baserels))
1406 1220 : pathnode->limit_tuples = root->limit_tuples;
1407 : else
1408 1114 : pathnode->limit_tuples = -1.0;
1409 :
1410 : /*
1411 : * Add up the sizes and costs of the input paths.
1412 : */
1413 2334 : pathnode->path.rows = 0;
1414 2334 : input_startup_cost = 0;
1415 2334 : input_total_cost = 0;
1416 8784 : foreach(l, subpaths)
1417 : {
1418 6450 : Path *subpath = (Path *) lfirst(l);
1419 :
1420 6450 : pathnode->path.rows += subpath->rows;
1421 11622 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1422 5172 : subpath->parallel_safe;
1423 :
1424 6450 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1425 : {
1426 : /* Subpath is adequately ordered, we won't need to sort it */
1427 6302 : input_startup_cost += subpath->startup_cost;
1428 6302 : input_total_cost += subpath->total_cost;
1429 : }
1430 : else
1431 : {
1432 : /* We'll need to insert a Sort node, so include cost for that */
1433 : Path sort_path; /* dummy for result of cost_sort */
1434 :
1435 444 : cost_sort(&sort_path,
1436 : root,
1437 : pathkeys,
1438 : subpath->total_cost,
1439 148 : subpath->parent->tuples,
1440 148 : subpath->pathtarget->width,
1441 : 0.0,
1442 : work_mem,
1443 : pathnode->limit_tuples);
1444 148 : input_startup_cost += sort_path.startup_cost;
1445 148 : input_total_cost += sort_path.total_cost;
1446 : }
1447 :
1448 : /* All child paths must have same parameterization */
1449 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1450 : }
1451 :
1452 : /*
1453 : * Now we can compute total costs of the MergeAppend. If there's exactly
1454 : * one child path, the MergeAppend is a no-op and will be discarded later
1455 : * (in setrefs.c); otherwise we do the normal cost calculation.
1456 : */
1457 2334 : if (list_length(subpaths) == 1)
1458 : {
1459 86 : pathnode->path.startup_cost = input_startup_cost;
1460 86 : pathnode->path.total_cost = input_total_cost;
1461 : }
1462 : else
1463 2248 : cost_merge_append(&pathnode->path, root,
1464 : pathkeys, list_length(subpaths),
1465 : input_startup_cost, input_total_cost,
1466 : pathnode->path.rows);
1467 :
1468 2334 : return pathnode;
1469 : }
1470 :
1471 : /*
1472 : * create_group_result_path
1473 : * Creates a path representing a Result-and-nothing-else plan.
1474 : *
1475 : * This is only used for degenerate grouping cases, in which we know we
1476 : * need to produce one result row, possibly filtered by a HAVING qual.
1477 : */
1478 : GroupResultPath *
1479 123576 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1480 : PathTarget *target, List *havingqual)
1481 : {
1482 123576 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1483 :
1484 123576 : pathnode->path.pathtype = T_Result;
1485 123576 : pathnode->path.parent = rel;
1486 123576 : pathnode->path.pathtarget = target;
1487 123576 : pathnode->path.param_info = NULL; /* there are no other rels... */
1488 123576 : pathnode->path.parallel_aware = false;
1489 123576 : pathnode->path.parallel_safe = rel->consider_parallel;
1490 123576 : pathnode->path.parallel_workers = 0;
1491 123576 : pathnode->path.pathkeys = NIL;
1492 123576 : pathnode->quals = havingqual;
1493 :
1494 : /*
1495 : * We can't quite use cost_resultscan() because the quals we want to
1496 : * account for are not baserestrict quals of the rel. Might as well just
1497 : * hack it here.
1498 : */
1499 123576 : pathnode->path.rows = 1;
1500 123576 : pathnode->path.startup_cost = target->cost.startup;
1501 247152 : pathnode->path.total_cost = target->cost.startup +
1502 123576 : cpu_tuple_cost + target->cost.per_tuple;
1503 :
1504 : /*
1505 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1506 : * rowcount estimate should be 1 no matter what the qual is.
1507 : */
1508 123576 : if (havingqual)
1509 : {
1510 : QualCost qual_cost;
1511 :
1512 282 : cost_qual_eval(&qual_cost, havingqual, root);
1513 : /* havingqual is evaluated once at startup */
1514 282 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1515 282 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1516 : }
1517 :
1518 123576 : return pathnode;
1519 : }
1520 :
1521 : /*
1522 : * create_material_path
1523 : * Creates a path corresponding to a Material plan, returning the
1524 : * pathnode.
1525 : */
1526 : MaterialPath *
1527 236688 : create_material_path(RelOptInfo *rel, Path *subpath)
1528 : {
1529 236688 : MaterialPath *pathnode = makeNode(MaterialPath);
1530 :
1531 : Assert(subpath->parent == rel);
1532 :
1533 236688 : pathnode->path.pathtype = T_Material;
1534 236688 : pathnode->path.parent = rel;
1535 236688 : pathnode->path.pathtarget = rel->reltarget;
1536 236688 : pathnode->path.param_info = subpath->param_info;
1537 236688 : pathnode->path.parallel_aware = false;
1538 441780 : pathnode->path.parallel_safe = rel->consider_parallel &&
1539 205092 : subpath->parallel_safe;
1540 236688 : pathnode->path.parallel_workers = subpath->parallel_workers;
1541 236688 : pathnode->path.pathkeys = subpath->pathkeys;
1542 :
1543 236688 : pathnode->subpath = subpath;
1544 :
1545 236688 : cost_material(&pathnode->path,
1546 : subpath->startup_cost,
1547 : subpath->total_cost,
1548 : subpath->rows,
1549 236688 : subpath->pathtarget->width);
1550 :
1551 236688 : return pathnode;
1552 : }
1553 :
1554 : /*
1555 : * create_unique_path
1556 : * Creates a path representing elimination of distinct rows from the
1557 : * input data. Distinct-ness is defined according to the needs of the
1558 : * semijoin represented by sjinfo. If it is not possible to identify
1559 : * how to make the data unique, NULL is returned.
1560 : *
1561 : * If used at all, this is likely to be called repeatedly on the same rel;
1562 : * and the input subpath should always be the same (the cheapest_total path
1563 : * for the rel). So we cache the result.
1564 : */
1565 : UniquePath *
1566 12842 : create_unique_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1567 : SpecialJoinInfo *sjinfo)
1568 : {
1569 : UniquePath *pathnode;
1570 : Path sort_path; /* dummy for result of cost_sort */
1571 : Path agg_path; /* dummy for result of cost_agg */
1572 : MemoryContext oldcontext;
1573 : int numCols;
1574 :
1575 : /* Caller made a mistake if subpath isn't cheapest_total ... */
1576 : Assert(subpath == rel->cheapest_total_path);
1577 : Assert(subpath->parent == rel);
1578 : /* ... or if SpecialJoinInfo is the wrong one */
1579 : Assert(sjinfo->jointype == JOIN_SEMI);
1580 : Assert(bms_equal(rel->relids, sjinfo->syn_righthand));
1581 :
1582 : /* If result already cached, return it */
1583 12842 : if (rel->cheapest_unique_path)
1584 10824 : return (UniquePath *) rel->cheapest_unique_path;
1585 :
1586 : /* If it's not possible to unique-ify, return NULL */
1587 2018 : if (!(sjinfo->semi_can_btree || sjinfo->semi_can_hash))
1588 62 : return NULL;
1589 :
1590 : /*
1591 : * When called during GEQO join planning, we are in a short-lived memory
1592 : * context. We must make sure that the path and any subsidiary data
1593 : * structures created for a baserel survive the GEQO cycle, else the
1594 : * baserel is trashed for future GEQO cycles. On the other hand, when we
1595 : * are creating those for a joinrel during GEQO, we don't want them to
1596 : * clutter the main planning context. Upshot is that the best solution is
1597 : * to explicitly allocate memory in the same context the given RelOptInfo
1598 : * is in.
1599 : */
1600 1956 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
1601 :
1602 1956 : pathnode = makeNode(UniquePath);
1603 :
1604 1956 : pathnode->path.pathtype = T_Unique;
1605 1956 : pathnode->path.parent = rel;
1606 1956 : pathnode->path.pathtarget = rel->reltarget;
1607 1956 : pathnode->path.param_info = subpath->param_info;
1608 1956 : pathnode->path.parallel_aware = false;
1609 3752 : pathnode->path.parallel_safe = rel->consider_parallel &&
1610 1796 : subpath->parallel_safe;
1611 1956 : pathnode->path.parallel_workers = subpath->parallel_workers;
1612 :
1613 : /*
1614 : * Assume the output is unsorted, since we don't necessarily have pathkeys
1615 : * to represent it. (This might get overridden below.)
1616 : */
1617 1956 : pathnode->path.pathkeys = NIL;
1618 :
1619 1956 : pathnode->subpath = subpath;
1620 1956 : pathnode->in_operators = sjinfo->semi_operators;
1621 1956 : pathnode->uniq_exprs = sjinfo->semi_rhs_exprs;
1622 :
1623 : /*
1624 : * If the input is a relation and it has a unique index that proves the
1625 : * semi_rhs_exprs are unique, then we don't need to do anything. Note
1626 : * that relation_has_unique_index_for automatically considers restriction
1627 : * clauses for the rel, as well.
1628 : */
1629 2396 : if (rel->rtekind == RTE_RELATION && sjinfo->semi_can_btree &&
1630 440 : relation_has_unique_index_for(root, rel, NIL,
1631 : sjinfo->semi_rhs_exprs,
1632 : sjinfo->semi_operators))
1633 : {
1634 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1635 0 : pathnode->path.rows = rel->rows;
1636 0 : pathnode->path.startup_cost = subpath->startup_cost;
1637 0 : pathnode->path.total_cost = subpath->total_cost;
1638 0 : pathnode->path.pathkeys = subpath->pathkeys;
1639 :
1640 0 : rel->cheapest_unique_path = (Path *) pathnode;
1641 :
1642 0 : MemoryContextSwitchTo(oldcontext);
1643 :
1644 0 : return pathnode;
1645 : }
1646 :
1647 : /*
1648 : * If the input is a subquery whose output must be unique already, then we
1649 : * don't need to do anything. The test for uniqueness has to consider
1650 : * exactly which columns we are extracting; for example "SELECT DISTINCT
1651 : * x,y" doesn't guarantee that x alone is distinct. So we cannot check for
1652 : * this optimization unless semi_rhs_exprs consists only of simple Vars
1653 : * referencing subquery outputs. (Possibly we could do something with
1654 : * expressions in the subquery outputs, too, but for now keep it simple.)
1655 : */
1656 1956 : if (rel->rtekind == RTE_SUBQUERY)
1657 : {
1658 316 : RangeTblEntry *rte = planner_rt_fetch(rel->relid, root);
1659 :
1660 316 : if (query_supports_distinctness(rte->subquery))
1661 : {
1662 : List *sub_tlist_colnos;
1663 :
1664 292 : sub_tlist_colnos = translate_sub_tlist(sjinfo->semi_rhs_exprs,
1665 292 : rel->relid);
1666 :
1667 332 : if (sub_tlist_colnos &&
1668 40 : query_is_distinct_for(rte->subquery,
1669 : sub_tlist_colnos,
1670 : sjinfo->semi_operators))
1671 : {
1672 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1673 0 : pathnode->path.rows = rel->rows;
1674 0 : pathnode->path.startup_cost = subpath->startup_cost;
1675 0 : pathnode->path.total_cost = subpath->total_cost;
1676 0 : pathnode->path.pathkeys = subpath->pathkeys;
1677 :
1678 0 : rel->cheapest_unique_path = (Path *) pathnode;
1679 :
1680 0 : MemoryContextSwitchTo(oldcontext);
1681 :
1682 0 : return pathnode;
1683 : }
1684 : }
1685 : }
1686 :
1687 : /* Estimate number of output rows */
1688 1956 : pathnode->path.rows = estimate_num_groups(root,
1689 : sjinfo->semi_rhs_exprs,
1690 : rel->rows,
1691 : NULL);
1692 1956 : numCols = list_length(sjinfo->semi_rhs_exprs);
1693 :
1694 1956 : if (sjinfo->semi_can_btree)
1695 : {
1696 : /*
1697 : * Estimate cost for sort+unique implementation
1698 : */
1699 3912 : cost_sort(&sort_path, root, NIL,
1700 : subpath->total_cost,
1701 : rel->rows,
1702 1956 : subpath->pathtarget->width,
1703 : 0.0,
1704 : work_mem,
1705 : -1.0);
1706 :
1707 : /*
1708 : * Charge one cpu_operator_cost per comparison per input tuple. We
1709 : * assume all columns get compared at most of the tuples. (XXX
1710 : * probably this is an overestimate.) This should agree with
1711 : * create_upper_unique_path.
1712 : */
1713 1956 : sort_path.total_cost += cpu_operator_cost * rel->rows * numCols;
1714 : }
1715 :
1716 1956 : if (sjinfo->semi_can_hash)
1717 : {
1718 : /*
1719 : * Estimate the overhead per hashtable entry at 64 bytes (same as in
1720 : * planner.c).
1721 : */
1722 1956 : int hashentrysize = subpath->pathtarget->width + 64;
1723 1956 : int hash_mem = get_hash_mem();
1724 :
1725 1956 : if (hashentrysize * pathnode->path.rows > hash_mem * 1024L)
1726 : {
1727 : /*
1728 : * We should not try to hash. Hack the SpecialJoinInfo to
1729 : * remember this, in case we come through here again.
1730 : */
1731 0 : sjinfo->semi_can_hash = false;
1732 : }
1733 : else
1734 1956 : cost_agg(&agg_path, root,
1735 : AGG_HASHED, NULL,
1736 : numCols, pathnode->path.rows,
1737 : NIL,
1738 : subpath->startup_cost,
1739 : subpath->total_cost,
1740 : rel->rows,
1741 1956 : subpath->pathtarget->width);
1742 : }
1743 :
1744 1956 : if (sjinfo->semi_can_btree && sjinfo->semi_can_hash)
1745 : {
1746 3912 : if (agg_path.total_cost < sort_path.total_cost)
1747 1906 : pathnode->umethod = UNIQUE_PATH_HASH;
1748 : else
1749 50 : pathnode->umethod = UNIQUE_PATH_SORT;
1750 : }
1751 0 : else if (sjinfo->semi_can_btree)
1752 0 : pathnode->umethod = UNIQUE_PATH_SORT;
1753 0 : else if (sjinfo->semi_can_hash)
1754 0 : pathnode->umethod = UNIQUE_PATH_HASH;
1755 : else
1756 : {
1757 : /* we can get here only if we abandoned hashing above */
1758 0 : MemoryContextSwitchTo(oldcontext);
1759 0 : return NULL;
1760 : }
1761 :
1762 1956 : if (pathnode->umethod == UNIQUE_PATH_HASH)
1763 : {
1764 1906 : pathnode->path.startup_cost = agg_path.startup_cost;
1765 1906 : pathnode->path.total_cost = agg_path.total_cost;
1766 : }
1767 : else
1768 : {
1769 50 : pathnode->path.startup_cost = sort_path.startup_cost;
1770 50 : pathnode->path.total_cost = sort_path.total_cost;
1771 : }
1772 :
1773 1956 : rel->cheapest_unique_path = (Path *) pathnode;
1774 :
1775 1956 : MemoryContextSwitchTo(oldcontext);
1776 :
1777 1956 : return pathnode;
1778 : }
1779 :
1780 : /*
1781 : * create_gather_merge_path
1782 : *
1783 : * Creates a path corresponding to a gather merge scan, returning
1784 : * the pathnode.
1785 : */
1786 : GatherMergePath *
1787 8188 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1788 : PathTarget *target, List *pathkeys,
1789 : Relids required_outer, double *rows)
1790 : {
1791 8188 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1792 8188 : Cost input_startup_cost = 0;
1793 8188 : Cost input_total_cost = 0;
1794 :
1795 : Assert(subpath->parallel_safe);
1796 : Assert(pathkeys);
1797 :
1798 8188 : pathnode->path.pathtype = T_GatherMerge;
1799 8188 : pathnode->path.parent = rel;
1800 8188 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1801 : required_outer);
1802 8188 : pathnode->path.parallel_aware = false;
1803 :
1804 8188 : pathnode->subpath = subpath;
1805 8188 : pathnode->num_workers = subpath->parallel_workers;
1806 8188 : pathnode->path.pathkeys = pathkeys;
1807 8188 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1808 8188 : pathnode->path.rows += subpath->rows;
1809 :
1810 8188 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1811 : {
1812 : /* Subpath is adequately ordered, we won't need to sort it */
1813 8188 : input_startup_cost += subpath->startup_cost;
1814 8188 : input_total_cost += subpath->total_cost;
1815 : }
1816 : else
1817 : {
1818 : /* We'll need to insert a Sort node, so include cost for that */
1819 : Path sort_path; /* dummy for result of cost_sort */
1820 :
1821 0 : cost_sort(&sort_path,
1822 : root,
1823 : pathkeys,
1824 : subpath->total_cost,
1825 : subpath->rows,
1826 0 : subpath->pathtarget->width,
1827 : 0.0,
1828 : work_mem,
1829 : -1);
1830 0 : input_startup_cost += sort_path.startup_cost;
1831 0 : input_total_cost += sort_path.total_cost;
1832 : }
1833 :
1834 8188 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1835 : input_startup_cost, input_total_cost, rows);
1836 :
1837 8188 : return pathnode;
1838 : }
1839 :
1840 : /*
1841 : * translate_sub_tlist - get subquery column numbers represented by tlist
1842 : *
1843 : * The given targetlist usually contains only Vars referencing the given relid.
1844 : * Extract their varattnos (ie, the column numbers of the subquery) and return
1845 : * as an integer List.
1846 : *
1847 : * If any of the tlist items is not a simple Var, we cannot determine whether
1848 : * the subquery's uniqueness condition (if any) matches ours, so punt and
1849 : * return NIL.
1850 : */
1851 : static List *
1852 292 : translate_sub_tlist(List *tlist, int relid)
1853 : {
1854 292 : List *result = NIL;
1855 : ListCell *l;
1856 :
1857 332 : foreach(l, tlist)
1858 : {
1859 292 : Var *var = (Var *) lfirst(l);
1860 :
1861 292 : if (!var || !IsA(var, Var) ||
1862 40 : var->varno != relid)
1863 252 : return NIL; /* punt */
1864 :
1865 40 : result = lappend_int(result, var->varattno);
1866 : }
1867 40 : return result;
1868 : }
1869 :
1870 : /*
1871 : * create_gather_path
1872 : * Creates a path corresponding to a gather scan, returning the
1873 : * pathnode.
1874 : *
1875 : * 'rows' may optionally be set to override row estimates from other sources.
1876 : */
1877 : GatherPath *
1878 10974 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1879 : PathTarget *target, Relids required_outer, double *rows)
1880 : {
1881 10974 : GatherPath *pathnode = makeNode(GatherPath);
1882 :
1883 : Assert(subpath->parallel_safe);
1884 :
1885 10974 : pathnode->path.pathtype = T_Gather;
1886 10974 : pathnode->path.parent = rel;
1887 10974 : pathnode->path.pathtarget = target;
1888 10974 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1889 : required_outer);
1890 10974 : pathnode->path.parallel_aware = false;
1891 10974 : pathnode->path.parallel_safe = false;
1892 10974 : pathnode->path.parallel_workers = 0;
1893 10974 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1894 :
1895 10974 : pathnode->subpath = subpath;
1896 10974 : pathnode->num_workers = subpath->parallel_workers;
1897 10974 : pathnode->single_copy = false;
1898 :
1899 10974 : if (pathnode->num_workers == 0)
1900 : {
1901 0 : pathnode->path.pathkeys = subpath->pathkeys;
1902 0 : pathnode->num_workers = 1;
1903 0 : pathnode->single_copy = true;
1904 : }
1905 :
1906 10974 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
1907 :
1908 10974 : return pathnode;
1909 : }
1910 :
1911 : /*
1912 : * create_subqueryscan_path
1913 : * Creates a path corresponding to a scan of a subquery,
1914 : * returning the pathnode.
1915 : */
1916 : SubqueryScanPath *
1917 10202 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1918 : List *pathkeys, Relids required_outer)
1919 : {
1920 10202 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
1921 :
1922 10202 : pathnode->path.pathtype = T_SubqueryScan;
1923 10202 : pathnode->path.parent = rel;
1924 10202 : pathnode->path.pathtarget = rel->reltarget;
1925 10202 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1926 : required_outer);
1927 10202 : pathnode->path.parallel_aware = false;
1928 16054 : pathnode->path.parallel_safe = rel->consider_parallel &&
1929 5852 : subpath->parallel_safe;
1930 10202 : pathnode->path.parallel_workers = subpath->parallel_workers;
1931 10202 : pathnode->path.pathkeys = pathkeys;
1932 10202 : pathnode->subpath = subpath;
1933 :
1934 10202 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info);
1935 :
1936 10202 : return pathnode;
1937 : }
1938 :
1939 : /*
1940 : * create_functionscan_path
1941 : * Creates a path corresponding to a sequential scan of a function,
1942 : * returning the pathnode.
1943 : */
1944 : Path *
1945 34040 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
1946 : List *pathkeys, Relids required_outer)
1947 : {
1948 34040 : Path *pathnode = makeNode(Path);
1949 :
1950 34040 : pathnode->pathtype = T_FunctionScan;
1951 34040 : pathnode->parent = rel;
1952 34040 : pathnode->pathtarget = rel->reltarget;
1953 34040 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1954 : required_outer);
1955 34040 : pathnode->parallel_aware = false;
1956 34040 : pathnode->parallel_safe = rel->consider_parallel;
1957 34040 : pathnode->parallel_workers = 0;
1958 34040 : pathnode->pathkeys = pathkeys;
1959 :
1960 34040 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
1961 :
1962 34040 : return pathnode;
1963 : }
1964 :
1965 : /*
1966 : * create_tablefuncscan_path
1967 : * Creates a path corresponding to a sequential scan of a table function,
1968 : * returning the pathnode.
1969 : */
1970 : Path *
1971 144 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
1972 : Relids required_outer)
1973 : {
1974 144 : Path *pathnode = makeNode(Path);
1975 :
1976 144 : pathnode->pathtype = T_TableFuncScan;
1977 144 : pathnode->parent = rel;
1978 144 : pathnode->pathtarget = rel->reltarget;
1979 144 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1980 : required_outer);
1981 144 : pathnode->parallel_aware = false;
1982 144 : pathnode->parallel_safe = rel->consider_parallel;
1983 144 : pathnode->parallel_workers = 0;
1984 144 : pathnode->pathkeys = NIL; /* result is always unordered */
1985 :
1986 144 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
1987 :
1988 144 : return pathnode;
1989 : }
1990 :
1991 : /*
1992 : * create_valuesscan_path
1993 : * Creates a path corresponding to a scan of a VALUES list,
1994 : * returning the pathnode.
1995 : */
1996 : Path *
1997 4408 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
1998 : Relids required_outer)
1999 : {
2000 4408 : Path *pathnode = makeNode(Path);
2001 :
2002 4408 : pathnode->pathtype = T_ValuesScan;
2003 4408 : pathnode->parent = rel;
2004 4408 : pathnode->pathtarget = rel->reltarget;
2005 4408 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2006 : required_outer);
2007 4408 : pathnode->parallel_aware = false;
2008 4408 : pathnode->parallel_safe = rel->consider_parallel;
2009 4408 : pathnode->parallel_workers = 0;
2010 4408 : pathnode->pathkeys = NIL; /* result is always unordered */
2011 :
2012 4408 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
2013 :
2014 4408 : return pathnode;
2015 : }
2016 :
2017 : /*
2018 : * create_ctescan_path
2019 : * Creates a path corresponding to a scan of a non-self-reference CTE,
2020 : * returning the pathnode.
2021 : */
2022 : Path *
2023 956 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
2024 : {
2025 956 : Path *pathnode = makeNode(Path);
2026 :
2027 956 : pathnode->pathtype = T_CteScan;
2028 956 : pathnode->parent = rel;
2029 956 : pathnode->pathtarget = rel->reltarget;
2030 956 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2031 : required_outer);
2032 956 : pathnode->parallel_aware = false;
2033 956 : pathnode->parallel_safe = rel->consider_parallel;
2034 956 : pathnode->parallel_workers = 0;
2035 956 : pathnode->pathkeys = NIL; /* XXX for now, result is always unordered */
2036 :
2037 956 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2038 :
2039 956 : return pathnode;
2040 : }
2041 :
2042 : /*
2043 : * create_namedtuplestorescan_path
2044 : * Creates a path corresponding to a scan of a named tuplestore, returning
2045 : * the pathnode.
2046 : */
2047 : Path *
2048 260 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
2049 : Relids required_outer)
2050 : {
2051 260 : Path *pathnode = makeNode(Path);
2052 :
2053 260 : pathnode->pathtype = T_NamedTuplestoreScan;
2054 260 : pathnode->parent = rel;
2055 260 : pathnode->pathtarget = rel->reltarget;
2056 260 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2057 : required_outer);
2058 260 : pathnode->parallel_aware = false;
2059 260 : pathnode->parallel_safe = rel->consider_parallel;
2060 260 : pathnode->parallel_workers = 0;
2061 260 : pathnode->pathkeys = NIL; /* result is always unordered */
2062 :
2063 260 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
2064 :
2065 260 : return pathnode;
2066 : }
2067 :
2068 : /*
2069 : * create_resultscan_path
2070 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2071 : * returning the pathnode.
2072 : */
2073 : Path *
2074 746 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2075 : Relids required_outer)
2076 : {
2077 746 : Path *pathnode = makeNode(Path);
2078 :
2079 746 : pathnode->pathtype = T_Result;
2080 746 : pathnode->parent = rel;
2081 746 : pathnode->pathtarget = rel->reltarget;
2082 746 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2083 : required_outer);
2084 746 : pathnode->parallel_aware = false;
2085 746 : pathnode->parallel_safe = rel->consider_parallel;
2086 746 : pathnode->parallel_workers = 0;
2087 746 : pathnode->pathkeys = NIL; /* result is always unordered */
2088 :
2089 746 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2090 :
2091 746 : return pathnode;
2092 : }
2093 :
2094 : /*
2095 : * create_worktablescan_path
2096 : * Creates a path corresponding to a scan of a self-reference CTE,
2097 : * returning the pathnode.
2098 : */
2099 : Path *
2100 348 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2101 : Relids required_outer)
2102 : {
2103 348 : Path *pathnode = makeNode(Path);
2104 :
2105 348 : pathnode->pathtype = T_WorkTableScan;
2106 348 : pathnode->parent = rel;
2107 348 : pathnode->pathtarget = rel->reltarget;
2108 348 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2109 : required_outer);
2110 348 : pathnode->parallel_aware = false;
2111 348 : pathnode->parallel_safe = rel->consider_parallel;
2112 348 : pathnode->parallel_workers = 0;
2113 348 : pathnode->pathkeys = NIL; /* result is always unordered */
2114 :
2115 : /* Cost is the same as for a regular CTE scan */
2116 348 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2117 :
2118 348 : return pathnode;
2119 : }
2120 :
2121 : /*
2122 : * create_foreignscan_path
2123 : * Creates a path corresponding to a scan of a foreign base table,
2124 : * returning the pathnode.
2125 : *
2126 : * This function is never called from core Postgres; rather, it's expected
2127 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2128 : * We make the FDW supply all fields of the path, since we do not have any way
2129 : * to calculate them in core. However, there is a usually-sane default for
2130 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2131 : */
2132 : ForeignPath *
2133 2770 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2134 : PathTarget *target,
2135 : double rows, Cost startup_cost, Cost total_cost,
2136 : List *pathkeys,
2137 : Relids required_outer,
2138 : Path *fdw_outerpath,
2139 : List *fdw_private)
2140 : {
2141 2770 : ForeignPath *pathnode = makeNode(ForeignPath);
2142 :
2143 : /* Historically some FDWs were confused about when to use this */
2144 : Assert(IS_SIMPLE_REL(rel));
2145 :
2146 2770 : pathnode->path.pathtype = T_ForeignScan;
2147 2770 : pathnode->path.parent = rel;
2148 2770 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2149 2770 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2150 : required_outer);
2151 2770 : pathnode->path.parallel_aware = false;
2152 2770 : pathnode->path.parallel_safe = rel->consider_parallel;
2153 2770 : pathnode->path.parallel_workers = 0;
2154 2770 : pathnode->path.rows = rows;
2155 2770 : pathnode->path.startup_cost = startup_cost;
2156 2770 : pathnode->path.total_cost = total_cost;
2157 2770 : pathnode->path.pathkeys = pathkeys;
2158 :
2159 2770 : pathnode->fdw_outerpath = fdw_outerpath;
2160 2770 : pathnode->fdw_private = fdw_private;
2161 :
2162 2770 : return pathnode;
2163 : }
2164 :
2165 : /*
2166 : * create_foreign_join_path
2167 : * Creates a path corresponding to a scan of a foreign join,
2168 : * returning the pathnode.
2169 : *
2170 : * This function is never called from core Postgres; rather, it's expected
2171 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2172 : * We make the FDW supply all fields of the path, since we do not have any way
2173 : * to calculate them in core. However, there is a usually-sane default for
2174 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2175 : */
2176 : ForeignPath *
2177 702 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2178 : PathTarget *target,
2179 : double rows, Cost startup_cost, Cost total_cost,
2180 : List *pathkeys,
2181 : Relids required_outer,
2182 : Path *fdw_outerpath,
2183 : List *fdw_private)
2184 : {
2185 702 : ForeignPath *pathnode = makeNode(ForeignPath);
2186 :
2187 : /*
2188 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2189 : * but the API of this function doesn't support it, and existing
2190 : * extensions aren't yet trying to build such paths anyway. For the
2191 : * moment just throw an error if someone tries it; eventually we should
2192 : * revisit this.
2193 : */
2194 702 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2195 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2196 :
2197 702 : pathnode->path.pathtype = T_ForeignScan;
2198 702 : pathnode->path.parent = rel;
2199 702 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2200 702 : pathnode->path.param_info = NULL; /* XXX see above */
2201 702 : pathnode->path.parallel_aware = false;
2202 702 : pathnode->path.parallel_safe = rel->consider_parallel;
2203 702 : pathnode->path.parallel_workers = 0;
2204 702 : pathnode->path.rows = rows;
2205 702 : pathnode->path.startup_cost = startup_cost;
2206 702 : pathnode->path.total_cost = total_cost;
2207 702 : pathnode->path.pathkeys = pathkeys;
2208 :
2209 702 : pathnode->fdw_outerpath = fdw_outerpath;
2210 702 : pathnode->fdw_private = fdw_private;
2211 :
2212 702 : return pathnode;
2213 : }
2214 :
2215 : /*
2216 : * create_foreign_upper_path
2217 : * Creates a path corresponding to an upper relation that's computed
2218 : * directly by an FDW, returning the pathnode.
2219 : *
2220 : * This function is never called from core Postgres; rather, it's expected to
2221 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2222 : * We make the FDW supply all fields of the path, since we do not have any way
2223 : * to calculate them in core. However, there is a usually-sane default for
2224 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2225 : */
2226 : ForeignPath *
2227 480 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2228 : PathTarget *target,
2229 : double rows, Cost startup_cost, Cost total_cost,
2230 : List *pathkeys,
2231 : Path *fdw_outerpath,
2232 : List *fdw_private)
2233 : {
2234 480 : ForeignPath *pathnode = makeNode(ForeignPath);
2235 :
2236 : /*
2237 : * Upper relations should never have any lateral references, since joining
2238 : * is complete.
2239 : */
2240 : Assert(bms_is_empty(rel->lateral_relids));
2241 :
2242 480 : pathnode->path.pathtype = T_ForeignScan;
2243 480 : pathnode->path.parent = rel;
2244 480 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2245 480 : pathnode->path.param_info = NULL;
2246 480 : pathnode->path.parallel_aware = false;
2247 480 : pathnode->path.parallel_safe = rel->consider_parallel;
2248 480 : pathnode->path.parallel_workers = 0;
2249 480 : pathnode->path.rows = rows;
2250 480 : pathnode->path.startup_cost = startup_cost;
2251 480 : pathnode->path.total_cost = total_cost;
2252 480 : pathnode->path.pathkeys = pathkeys;
2253 :
2254 480 : pathnode->fdw_outerpath = fdw_outerpath;
2255 480 : pathnode->fdw_private = fdw_private;
2256 :
2257 480 : return pathnode;
2258 : }
2259 :
2260 : /*
2261 : * calc_nestloop_required_outer
2262 : * Compute the required_outer set for a nestloop join path
2263 : *
2264 : * Note: result must not share storage with either input
2265 : */
2266 : Relids
2267 1204588 : calc_nestloop_required_outer(Relids outerrelids,
2268 : Relids outer_paramrels,
2269 : Relids innerrelids,
2270 : Relids inner_paramrels)
2271 : {
2272 : Relids required_outer;
2273 :
2274 : /* inner_path can require rels from outer path, but not vice versa */
2275 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2276 : /* easy case if inner path is not parameterized */
2277 1204588 : if (!inner_paramrels)
2278 937494 : return bms_copy(outer_paramrels);
2279 : /* else, form the union ... */
2280 267094 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2281 : /* ... and remove any mention of now-satisfied outer rels */
2282 267094 : required_outer = bms_del_members(required_outer,
2283 : outerrelids);
2284 : /* maintain invariant that required_outer is exactly NULL if empty */
2285 267094 : if (bms_is_empty(required_outer))
2286 : {
2287 214494 : bms_free(required_outer);
2288 214494 : required_outer = NULL;
2289 : }
2290 267094 : return required_outer;
2291 : }
2292 :
2293 : /*
2294 : * calc_non_nestloop_required_outer
2295 : * Compute the required_outer set for a merge or hash join path
2296 : *
2297 : * Note: result must not share storage with either input
2298 : */
2299 : Relids
2300 859948 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2301 : {
2302 859948 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2303 859948 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2304 : Relids required_outer;
2305 :
2306 : /* neither path can require rels from the other */
2307 : Assert(!bms_overlap(outer_paramrels, inner_path->parent->relids));
2308 : Assert(!bms_overlap(inner_paramrels, outer_path->parent->relids));
2309 : /* form the union ... */
2310 859948 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2311 : /* we do not need an explicit test for empty; bms_union gets it right */
2312 859948 : return required_outer;
2313 : }
2314 :
2315 : /*
2316 : * create_nestloop_path
2317 : * Creates a pathnode corresponding to a nestloop join between two
2318 : * relations.
2319 : *
2320 : * 'joinrel' is the join relation.
2321 : * 'jointype' is the type of join required
2322 : * 'workspace' is the result from initial_cost_nestloop
2323 : * 'extra' contains various information about the join
2324 : * 'outer_path' is the outer path
2325 : * 'inner_path' is the inner path
2326 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2327 : * 'pathkeys' are the path keys of the new join path
2328 : * 'required_outer' is the set of required outer rels
2329 : *
2330 : * Returns the resulting path node.
2331 : */
2332 : NestPath *
2333 617508 : create_nestloop_path(PlannerInfo *root,
2334 : RelOptInfo *joinrel,
2335 : JoinType jointype,
2336 : JoinCostWorkspace *workspace,
2337 : JoinPathExtraData *extra,
2338 : Path *outer_path,
2339 : Path *inner_path,
2340 : List *restrict_clauses,
2341 : List *pathkeys,
2342 : Relids required_outer)
2343 : {
2344 617508 : NestPath *pathnode = makeNode(NestPath);
2345 617508 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2346 :
2347 : /*
2348 : * If the inner path is parameterized by the outer, we must drop any
2349 : * restrict_clauses that are due to be moved into the inner path. We have
2350 : * to do this now, rather than postpone the work till createplan time,
2351 : * because the restrict_clauses list can affect the size and cost
2352 : * estimates for this path.
2353 : */
2354 617508 : if (bms_overlap(inner_req_outer, outer_path->parent->relids))
2355 : {
2356 107080 : Relids inner_and_outer = bms_union(inner_path->parent->relids,
2357 : inner_req_outer);
2358 107080 : List *jclauses = NIL;
2359 : ListCell *lc;
2360 :
2361 222010 : foreach(lc, restrict_clauses)
2362 : {
2363 114930 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2364 :
2365 114930 : if (!join_clause_is_movable_into(rinfo,
2366 114930 : inner_path->parent->relids,
2367 : inner_and_outer))
2368 6314 : jclauses = lappend(jclauses, rinfo);
2369 : }
2370 107080 : restrict_clauses = jclauses;
2371 : }
2372 :
2373 617508 : pathnode->path.pathtype = T_NestLoop;
2374 617508 : pathnode->path.parent = joinrel;
2375 617508 : pathnode->path.pathtarget = joinrel->reltarget;
2376 617508 : pathnode->path.param_info =
2377 617508 : get_joinrel_parampathinfo(root,
2378 : joinrel,
2379 : outer_path,
2380 : inner_path,
2381 : extra->sjinfo,
2382 : required_outer,
2383 : &restrict_clauses);
2384 617508 : pathnode->path.parallel_aware = false;
2385 1768632 : pathnode->path.parallel_safe = joinrel->consider_parallel &&
2386 617508 : outer_path->parallel_safe && inner_path->parallel_safe;
2387 : /* This is a foolish way to estimate parallel_workers, but for now... */
2388 617508 : pathnode->path.parallel_workers = outer_path->parallel_workers;
2389 617508 : pathnode->path.pathkeys = pathkeys;
2390 617508 : pathnode->jointype = jointype;
2391 617508 : pathnode->inner_unique = extra->inner_unique;
2392 617508 : pathnode->outerjoinpath = outer_path;
2393 617508 : pathnode->innerjoinpath = inner_path;
2394 617508 : pathnode->joinrestrictinfo = restrict_clauses;
2395 :
2396 617508 : final_cost_nestloop(root, pathnode, workspace, extra);
2397 :
2398 617508 : return pathnode;
2399 : }
2400 :
2401 : /*
2402 : * create_mergejoin_path
2403 : * Creates a pathnode corresponding to a mergejoin join between
2404 : * two relations
2405 : *
2406 : * 'joinrel' is the join relation
2407 : * 'jointype' is the type of join required
2408 : * 'workspace' is the result from initial_cost_mergejoin
2409 : * 'extra' contains various information about the join
2410 : * 'outer_path' is the outer path
2411 : * 'inner_path' is the inner path
2412 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2413 : * 'pathkeys' are the path keys of the new join path
2414 : * 'required_outer' is the set of required outer rels
2415 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2416 : * (this should be a subset of the restrict_clauses list)
2417 : * 'outersortkeys' are the sort varkeys for the outer relation
2418 : * 'innersortkeys' are the sort varkeys for the inner relation
2419 : */
2420 : MergePath *
2421 142338 : create_mergejoin_path(PlannerInfo *root,
2422 : RelOptInfo *joinrel,
2423 : JoinType jointype,
2424 : JoinCostWorkspace *workspace,
2425 : JoinPathExtraData *extra,
2426 : Path *outer_path,
2427 : Path *inner_path,
2428 : List *restrict_clauses,
2429 : List *pathkeys,
2430 : Relids required_outer,
2431 : List *mergeclauses,
2432 : List *outersortkeys,
2433 : List *innersortkeys)
2434 : {
2435 142338 : MergePath *pathnode = makeNode(MergePath);
2436 :
2437 142338 : pathnode->jpath.path.pathtype = T_MergeJoin;
2438 142338 : pathnode->jpath.path.parent = joinrel;
2439 142338 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2440 142338 : pathnode->jpath.path.param_info =
2441 142338 : get_joinrel_parampathinfo(root,
2442 : joinrel,
2443 : outer_path,
2444 : inner_path,
2445 : extra->sjinfo,
2446 : required_outer,
2447 : &restrict_clauses);
2448 142338 : pathnode->jpath.path.parallel_aware = false;
2449 396442 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2450 142338 : outer_path->parallel_safe && inner_path->parallel_safe;
2451 : /* This is a foolish way to estimate parallel_workers, but for now... */
2452 142338 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2453 142338 : pathnode->jpath.path.pathkeys = pathkeys;
2454 142338 : pathnode->jpath.jointype = jointype;
2455 142338 : pathnode->jpath.inner_unique = extra->inner_unique;
2456 142338 : pathnode->jpath.outerjoinpath = outer_path;
2457 142338 : pathnode->jpath.innerjoinpath = inner_path;
2458 142338 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2459 142338 : pathnode->path_mergeclauses = mergeclauses;
2460 142338 : pathnode->outersortkeys = outersortkeys;
2461 142338 : pathnode->innersortkeys = innersortkeys;
2462 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2463 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2464 :
2465 142338 : final_cost_mergejoin(root, pathnode, workspace, extra);
2466 :
2467 142338 : return pathnode;
2468 : }
2469 :
2470 : /*
2471 : * create_hashjoin_path
2472 : * Creates a pathnode corresponding to a hash join between two relations.
2473 : *
2474 : * 'joinrel' is the join relation
2475 : * 'jointype' is the type of join required
2476 : * 'workspace' is the result from initial_cost_hashjoin
2477 : * 'extra' contains various information about the join
2478 : * 'outer_path' is the cheapest outer path
2479 : * 'inner_path' is the cheapest inner path
2480 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2481 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2482 : * 'required_outer' is the set of required outer rels
2483 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2484 : * (this should be a subset of the restrict_clauses list)
2485 : */
2486 : HashPath *
2487 153470 : create_hashjoin_path(PlannerInfo *root,
2488 : RelOptInfo *joinrel,
2489 : JoinType jointype,
2490 : JoinCostWorkspace *workspace,
2491 : JoinPathExtraData *extra,
2492 : Path *outer_path,
2493 : Path *inner_path,
2494 : bool parallel_hash,
2495 : List *restrict_clauses,
2496 : Relids required_outer,
2497 : List *hashclauses)
2498 : {
2499 153470 : HashPath *pathnode = makeNode(HashPath);
2500 :
2501 153470 : pathnode->jpath.path.pathtype = T_HashJoin;
2502 153470 : pathnode->jpath.path.parent = joinrel;
2503 153470 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2504 153470 : pathnode->jpath.path.param_info =
2505 153470 : get_joinrel_parampathinfo(root,
2506 : joinrel,
2507 : outer_path,
2508 : inner_path,
2509 : extra->sjinfo,
2510 : required_outer,
2511 : &restrict_clauses);
2512 153470 : pathnode->jpath.path.parallel_aware =
2513 153470 : joinrel->consider_parallel && parallel_hash;
2514 426804 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2515 153470 : outer_path->parallel_safe && inner_path->parallel_safe;
2516 : /* This is a foolish way to estimate parallel_workers, but for now... */
2517 153470 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2518 :
2519 : /*
2520 : * A hashjoin never has pathkeys, since its output ordering is
2521 : * unpredictable due to possible batching. XXX If the inner relation is
2522 : * small enough, we could instruct the executor that it must not batch,
2523 : * and then we could assume that the output inherits the outer relation's
2524 : * ordering, which might save a sort step. However there is considerable
2525 : * downside if our estimate of the inner relation size is badly off. For
2526 : * the moment we don't risk it. (Note also that if we wanted to take this
2527 : * seriously, joinpath.c would have to consider many more paths for the
2528 : * outer rel than it does now.)
2529 : */
2530 153470 : pathnode->jpath.path.pathkeys = NIL;
2531 153470 : pathnode->jpath.jointype = jointype;
2532 153470 : pathnode->jpath.inner_unique = extra->inner_unique;
2533 153470 : pathnode->jpath.outerjoinpath = outer_path;
2534 153470 : pathnode->jpath.innerjoinpath = inner_path;
2535 153470 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2536 153470 : pathnode->path_hashclauses = hashclauses;
2537 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2538 :
2539 153470 : final_cost_hashjoin(root, pathnode, workspace, extra);
2540 :
2541 153470 : return pathnode;
2542 : }
2543 :
2544 : /*
2545 : * create_projection_path
2546 : * Creates a pathnode that represents performing a projection.
2547 : *
2548 : * 'rel' is the parent relation associated with the result
2549 : * 'subpath' is the path representing the source of data
2550 : * 'target' is the PathTarget to be computed
2551 : */
2552 : ProjectionPath *
2553 253408 : create_projection_path(PlannerInfo *root,
2554 : RelOptInfo *rel,
2555 : Path *subpath,
2556 : PathTarget *target)
2557 : {
2558 253408 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2559 253408 : PathTarget *oldtarget = subpath->pathtarget;
2560 :
2561 253408 : pathnode->path.pathtype = T_Result;
2562 253408 : pathnode->path.parent = rel;
2563 253408 : pathnode->path.pathtarget = target;
2564 : /* For now, assume we are above any joins, so no parameterization */
2565 253408 : pathnode->path.param_info = NULL;
2566 253408 : pathnode->path.parallel_aware = false;
2567 544194 : pathnode->path.parallel_safe = rel->consider_parallel &&
2568 290014 : subpath->parallel_safe &&
2569 36606 : is_parallel_safe(root, (Node *) target->exprs);
2570 253408 : pathnode->path.parallel_workers = subpath->parallel_workers;
2571 : /* Projection does not change the sort order */
2572 253408 : pathnode->path.pathkeys = subpath->pathkeys;
2573 :
2574 253408 : pathnode->subpath = subpath;
2575 :
2576 : /*
2577 : * We might not need a separate Result node. If the input plan node type
2578 : * can project, we can just tell it to project something else. Or, if it
2579 : * can't project but the desired target has the same expression list as
2580 : * what the input will produce anyway, we can still give it the desired
2581 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2582 : * Note: in the latter case, create_projection_plan has to recheck our
2583 : * conclusion; see comments therein.
2584 : */
2585 263546 : if (is_projection_capable_path(subpath) ||
2586 10138 : equal(oldtarget->exprs, target->exprs))
2587 : {
2588 : /* No separate Result node needed */
2589 252166 : pathnode->dummypp = true;
2590 :
2591 : /*
2592 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2593 : */
2594 252166 : pathnode->path.rows = subpath->rows;
2595 504332 : pathnode->path.startup_cost = subpath->startup_cost +
2596 252166 : (target->cost.startup - oldtarget->cost.startup);
2597 756498 : pathnode->path.total_cost = subpath->total_cost +
2598 504332 : (target->cost.startup - oldtarget->cost.startup) +
2599 252166 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2600 : }
2601 : else
2602 : {
2603 : /* We really do need the Result node */
2604 1242 : pathnode->dummypp = false;
2605 :
2606 : /*
2607 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2608 : * evaluating the tlist. There is no qual to worry about.
2609 : */
2610 1242 : pathnode->path.rows = subpath->rows;
2611 2484 : pathnode->path.startup_cost = subpath->startup_cost +
2612 1242 : target->cost.startup;
2613 3726 : pathnode->path.total_cost = subpath->total_cost +
2614 2484 : target->cost.startup +
2615 1242 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2616 : }
2617 :
2618 253408 : return pathnode;
2619 : }
2620 :
2621 : /*
2622 : * apply_projection_to_path
2623 : * Add a projection step, or just apply the target directly to given path.
2624 : *
2625 : * This has the same net effect as create_projection_path(), except that if
2626 : * a separate Result plan node isn't needed, we just replace the given path's
2627 : * pathtarget with the desired one. This must be used only when the caller
2628 : * knows that the given path isn't referenced elsewhere and so can be modified
2629 : * in-place.
2630 : *
2631 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2632 : * new target down to its input as well; this is a yet more invasive
2633 : * modification of the input path, which create_projection_path() can't do.
2634 : *
2635 : * Note that we mustn't change the source path's parent link; so when it is
2636 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2637 : * not caused any trouble.
2638 : *
2639 : * 'rel' is the parent relation associated with the result
2640 : * 'path' is the path representing the source of data
2641 : * 'target' is the PathTarget to be computed
2642 : */
2643 : Path *
2644 17012 : apply_projection_to_path(PlannerInfo *root,
2645 : RelOptInfo *rel,
2646 : Path *path,
2647 : PathTarget *target)
2648 : {
2649 : QualCost oldcost;
2650 :
2651 : /*
2652 : * If given path can't project, we might need a Result node, so make a
2653 : * separate ProjectionPath.
2654 : */
2655 17012 : if (!is_projection_capable_path(path))
2656 8242 : return (Path *) create_projection_path(root, rel, path, target);
2657 :
2658 : /*
2659 : * We can just jam the desired tlist into the existing path, being sure to
2660 : * update its cost estimates appropriately.
2661 : */
2662 8770 : oldcost = path->pathtarget->cost;
2663 8770 : path->pathtarget = target;
2664 :
2665 8770 : path->startup_cost += target->cost.startup - oldcost.startup;
2666 17540 : path->total_cost += target->cost.startup - oldcost.startup +
2667 8770 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2668 :
2669 : /*
2670 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2671 : * arrange for the subpath to return the required target list so that
2672 : * workers can help project. But if there is something that is not
2673 : * parallel-safe in the target expressions, then we can't.
2674 : */
2675 10268 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
2676 1498 : is_parallel_safe(root, (Node *) target->exprs))
2677 : {
2678 : /*
2679 : * We always use create_projection_path here, even if the subpath is
2680 : * projection-capable, so as to avoid modifying the subpath in place.
2681 : * It seems unlikely at present that there could be any other
2682 : * references to the subpath, but better safe than sorry.
2683 : *
2684 : * Note that we don't change the parallel path's cost estimates; it
2685 : * might be appropriate to do so, to reflect the fact that the bulk of
2686 : * the target evaluation will happen in workers.
2687 : */
2688 2996 : if (IsA(path, GatherPath))
2689 : {
2690 0 : GatherPath *gpath = (GatherPath *) path;
2691 :
2692 0 : gpath->subpath = (Path *)
2693 0 : create_projection_path(root,
2694 0 : gpath->subpath->parent,
2695 : gpath->subpath,
2696 : target);
2697 : }
2698 : else
2699 : {
2700 1498 : GatherMergePath *gmpath = (GatherMergePath *) path;
2701 :
2702 1498 : gmpath->subpath = (Path *)
2703 2996 : create_projection_path(root,
2704 1498 : gmpath->subpath->parent,
2705 : gmpath->subpath,
2706 : target);
2707 : }
2708 : }
2709 7272 : else if (path->parallel_safe &&
2710 2842 : !is_parallel_safe(root, (Node *) target->exprs))
2711 : {
2712 : /*
2713 : * We're inserting a parallel-restricted target list into a path
2714 : * currently marked parallel-safe, so we have to mark it as no longer
2715 : * safe.
2716 : */
2717 8 : path->parallel_safe = false;
2718 : }
2719 :
2720 8770 : return path;
2721 : }
2722 :
2723 : /*
2724 : * create_set_projection_path
2725 : * Creates a pathnode that represents performing a projection that
2726 : * includes set-returning functions.
2727 : *
2728 : * 'rel' is the parent relation associated with the result
2729 : * 'subpath' is the path representing the source of data
2730 : * 'target' is the PathTarget to be computed
2731 : */
2732 : ProjectSetPath *
2733 4110 : create_set_projection_path(PlannerInfo *root,
2734 : RelOptInfo *rel,
2735 : Path *subpath,
2736 : PathTarget *target)
2737 : {
2738 4110 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2739 : double tlist_rows;
2740 : ListCell *lc;
2741 :
2742 4110 : pathnode->path.pathtype = T_ProjectSet;
2743 4110 : pathnode->path.parent = rel;
2744 4110 : pathnode->path.pathtarget = target;
2745 : /* For now, assume we are above any joins, so no parameterization */
2746 4110 : pathnode->path.param_info = NULL;
2747 4110 : pathnode->path.parallel_aware = false;
2748 8686 : pathnode->path.parallel_safe = rel->consider_parallel &&
2749 4556 : subpath->parallel_safe &&
2750 446 : is_parallel_safe(root, (Node *) target->exprs);
2751 4110 : pathnode->path.parallel_workers = subpath->parallel_workers;
2752 : /* Projection does not change the sort order XXX? */
2753 4110 : pathnode->path.pathkeys = subpath->pathkeys;
2754 :
2755 4110 : pathnode->subpath = subpath;
2756 :
2757 : /*
2758 : * Estimate number of rows produced by SRFs for each row of input; if
2759 : * there's more than one in this node, use the maximum.
2760 : */
2761 4110 : tlist_rows = 1;
2762 9414 : foreach(lc, target->exprs)
2763 : {
2764 5304 : Node *node = (Node *) lfirst(lc);
2765 : double itemrows;
2766 :
2767 5304 : itemrows = expression_returns_set_rows(root, node);
2768 5304 : if (tlist_rows < itemrows)
2769 3980 : tlist_rows = itemrows;
2770 : }
2771 :
2772 : /*
2773 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2774 : * per input row, and half of cpu_tuple_cost for each added output row.
2775 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2776 : * this estimate later.
2777 : */
2778 4110 : pathnode->path.rows = subpath->rows * tlist_rows;
2779 8220 : pathnode->path.startup_cost = subpath->startup_cost +
2780 4110 : target->cost.startup;
2781 12330 : pathnode->path.total_cost = subpath->total_cost +
2782 8220 : target->cost.startup +
2783 8220 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2784 4110 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2785 :
2786 4110 : return pathnode;
2787 : }
2788 :
2789 : /*
2790 : * create_incremental_sort_path
2791 : * Creates a pathnode that represents performing an incremental sort.
2792 : *
2793 : * 'rel' is the parent relation associated with the result
2794 : * 'subpath' is the path representing the source of data
2795 : * 'pathkeys' represents the desired sort order
2796 : * 'presorted_keys' is the number of keys by which the input path is
2797 : * already sorted
2798 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2799 : * or -1 if no LIMIT or couldn't estimate
2800 : */
2801 : IncrementalSortPath *
2802 1610 : create_incremental_sort_path(PlannerInfo *root,
2803 : RelOptInfo *rel,
2804 : Path *subpath,
2805 : List *pathkeys,
2806 : int presorted_keys,
2807 : double limit_tuples)
2808 : {
2809 1610 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
2810 1610 : SortPath *pathnode = &sort->spath;
2811 :
2812 1610 : pathnode->path.pathtype = T_IncrementalSort;
2813 1610 : pathnode->path.parent = rel;
2814 : /* Sort doesn't project, so use source path's pathtarget */
2815 1610 : pathnode->path.pathtarget = subpath->pathtarget;
2816 : /* For now, assume we are above any joins, so no parameterization */
2817 1610 : pathnode->path.param_info = NULL;
2818 1610 : pathnode->path.parallel_aware = false;
2819 2928 : pathnode->path.parallel_safe = rel->consider_parallel &&
2820 1318 : subpath->parallel_safe;
2821 1610 : pathnode->path.parallel_workers = subpath->parallel_workers;
2822 1610 : pathnode->path.pathkeys = pathkeys;
2823 :
2824 1610 : pathnode->subpath = subpath;
2825 :
2826 3220 : cost_incremental_sort(&pathnode->path,
2827 : root, pathkeys, presorted_keys,
2828 : subpath->startup_cost,
2829 : subpath->total_cost,
2830 : subpath->rows,
2831 1610 : subpath->pathtarget->width,
2832 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2833 : work_mem, limit_tuples);
2834 :
2835 1610 : sort->nPresortedCols = presorted_keys;
2836 :
2837 1610 : return sort;
2838 : }
2839 :
2840 : /*
2841 : * create_sort_path
2842 : * Creates a pathnode that represents performing an explicit sort.
2843 : *
2844 : * 'rel' is the parent relation associated with the result
2845 : * 'subpath' is the path representing the source of data
2846 : * 'pathkeys' represents the desired sort order
2847 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2848 : * or -1 if no LIMIT or couldn't estimate
2849 : */
2850 : SortPath *
2851 43470 : create_sort_path(PlannerInfo *root,
2852 : RelOptInfo *rel,
2853 : Path *subpath,
2854 : List *pathkeys,
2855 : double limit_tuples)
2856 : {
2857 43470 : SortPath *pathnode = makeNode(SortPath);
2858 :
2859 43470 : pathnode->path.pathtype = T_Sort;
2860 43470 : pathnode->path.parent = rel;
2861 : /* Sort doesn't project, so use source path's pathtarget */
2862 43470 : pathnode->path.pathtarget = subpath->pathtarget;
2863 : /* For now, assume we are above any joins, so no parameterization */
2864 43470 : pathnode->path.param_info = NULL;
2865 43470 : pathnode->path.parallel_aware = false;
2866 69586 : pathnode->path.parallel_safe = rel->consider_parallel &&
2867 26116 : subpath->parallel_safe;
2868 43470 : pathnode->path.parallel_workers = subpath->parallel_workers;
2869 43470 : pathnode->path.pathkeys = pathkeys;
2870 :
2871 43470 : pathnode->subpath = subpath;
2872 :
2873 86940 : cost_sort(&pathnode->path, root, pathkeys,
2874 : subpath->total_cost,
2875 : subpath->rows,
2876 43470 : subpath->pathtarget->width,
2877 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2878 : work_mem, limit_tuples);
2879 :
2880 43470 : return pathnode;
2881 : }
2882 :
2883 : /*
2884 : * create_group_path
2885 : * Creates a pathnode that represents performing grouping of presorted input
2886 : *
2887 : * 'rel' is the parent relation associated with the result
2888 : * 'subpath' is the path representing the source of data
2889 : * 'target' is the PathTarget to be computed
2890 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2891 : * 'qual' is the HAVING quals if any
2892 : * 'numGroups' is the estimated number of groups
2893 : */
2894 : GroupPath *
2895 604 : create_group_path(PlannerInfo *root,
2896 : RelOptInfo *rel,
2897 : Path *subpath,
2898 : List *groupClause,
2899 : List *qual,
2900 : double numGroups)
2901 : {
2902 604 : GroupPath *pathnode = makeNode(GroupPath);
2903 604 : PathTarget *target = rel->reltarget;
2904 :
2905 604 : pathnode->path.pathtype = T_Group;
2906 604 : pathnode->path.parent = rel;
2907 604 : pathnode->path.pathtarget = target;
2908 : /* For now, assume we are above any joins, so no parameterization */
2909 604 : pathnode->path.param_info = NULL;
2910 604 : pathnode->path.parallel_aware = false;
2911 946 : pathnode->path.parallel_safe = rel->consider_parallel &&
2912 342 : subpath->parallel_safe;
2913 604 : pathnode->path.parallel_workers = subpath->parallel_workers;
2914 : /* Group doesn't change sort ordering */
2915 604 : pathnode->path.pathkeys = subpath->pathkeys;
2916 :
2917 604 : pathnode->subpath = subpath;
2918 :
2919 604 : pathnode->groupClause = groupClause;
2920 604 : pathnode->qual = qual;
2921 :
2922 604 : cost_group(&pathnode->path, root,
2923 : list_length(groupClause),
2924 : numGroups,
2925 : qual,
2926 : subpath->startup_cost, subpath->total_cost,
2927 : subpath->rows);
2928 :
2929 : /* add tlist eval cost for each output row */
2930 604 : pathnode->path.startup_cost += target->cost.startup;
2931 1208 : pathnode->path.total_cost += target->cost.startup +
2932 604 : target->cost.per_tuple * pathnode->path.rows;
2933 :
2934 604 : return pathnode;
2935 : }
2936 :
2937 : /*
2938 : * create_upper_unique_path
2939 : * Creates a pathnode that represents performing an explicit Unique step
2940 : * on presorted input.
2941 : *
2942 : * This produces a Unique plan node, but the use-case is so different from
2943 : * create_unique_path that it doesn't seem worth trying to merge the two.
2944 : *
2945 : * 'rel' is the parent relation associated with the result
2946 : * 'subpath' is the path representing the source of data
2947 : * 'numCols' is the number of grouping columns
2948 : * 'numGroups' is the estimated number of groups
2949 : *
2950 : * The input path must be sorted on the grouping columns, plus possibly
2951 : * additional columns; so the first numCols pathkeys are the grouping columns
2952 : */
2953 : UpperUniquePath *
2954 688 : create_upper_unique_path(PlannerInfo *root,
2955 : RelOptInfo *rel,
2956 : Path *subpath,
2957 : int numCols,
2958 : double numGroups)
2959 : {
2960 688 : UpperUniquePath *pathnode = makeNode(UpperUniquePath);
2961 :
2962 688 : pathnode->path.pathtype = T_Unique;
2963 688 : pathnode->path.parent = rel;
2964 : /* Unique doesn't project, so use source path's pathtarget */
2965 688 : pathnode->path.pathtarget = subpath->pathtarget;
2966 : /* For now, assume we are above any joins, so no parameterization */
2967 688 : pathnode->path.param_info = NULL;
2968 688 : pathnode->path.parallel_aware = false;
2969 1122 : pathnode->path.parallel_safe = rel->consider_parallel &&
2970 434 : subpath->parallel_safe;
2971 688 : pathnode->path.parallel_workers = subpath->parallel_workers;
2972 : /* Unique doesn't change the input ordering */
2973 688 : pathnode->path.pathkeys = subpath->pathkeys;
2974 :
2975 688 : pathnode->subpath = subpath;
2976 688 : pathnode->numkeys = numCols;
2977 :
2978 : /*
2979 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
2980 : * all columns get compared at most of the tuples. (XXX probably this is
2981 : * an overestimate.)
2982 : */
2983 688 : pathnode->path.startup_cost = subpath->startup_cost;
2984 1376 : pathnode->path.total_cost = subpath->total_cost +
2985 688 : cpu_operator_cost * subpath->rows * numCols;
2986 688 : pathnode->path.rows = numGroups;
2987 :
2988 688 : return pathnode;
2989 : }
2990 :
2991 : /*
2992 : * create_agg_path
2993 : * Creates a pathnode that represents performing aggregation/grouping
2994 : *
2995 : * 'rel' is the parent relation associated with the result
2996 : * 'subpath' is the path representing the source of data
2997 : * 'target' is the PathTarget to be computed
2998 : * 'aggstrategy' is the Agg node's basic implementation strategy
2999 : * 'aggsplit' is the Agg node's aggregate-splitting mode
3000 : * 'groupClause' is a list of SortGroupClause's representing the grouping
3001 : * 'qual' is the HAVING quals if any
3002 : * 'aggcosts' contains cost info about the aggregate functions to be computed
3003 : * 'numGroups' is the estimated number of groups (1 if not grouping)
3004 : */
3005 : AggPath *
3006 37570 : create_agg_path(PlannerInfo *root,
3007 : RelOptInfo *rel,
3008 : Path *subpath,
3009 : PathTarget *target,
3010 : AggStrategy aggstrategy,
3011 : AggSplit aggsplit,
3012 : List *groupClause,
3013 : List *qual,
3014 : const AggClauseCosts *aggcosts,
3015 : double numGroups)
3016 : {
3017 37570 : AggPath *pathnode = makeNode(AggPath);
3018 :
3019 37570 : pathnode->path.pathtype = T_Agg;
3020 37570 : pathnode->path.parent = rel;
3021 37570 : pathnode->path.pathtarget = target;
3022 : /* For now, assume we are above any joins, so no parameterization */
3023 37570 : pathnode->path.param_info = NULL;
3024 37570 : pathnode->path.parallel_aware = false;
3025 55186 : pathnode->path.parallel_safe = rel->consider_parallel &&
3026 17616 : subpath->parallel_safe;
3027 37570 : pathnode->path.parallel_workers = subpath->parallel_workers;
3028 37570 : if (aggstrategy == AGG_SORTED)
3029 4426 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3030 : else
3031 33144 : pathnode->path.pathkeys = NIL; /* output is unordered */
3032 37570 : pathnode->subpath = subpath;
3033 :
3034 37570 : pathnode->aggstrategy = aggstrategy;
3035 37570 : pathnode->aggsplit = aggsplit;
3036 37570 : pathnode->numGroups = numGroups;
3037 37570 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
3038 37570 : pathnode->groupClause = groupClause;
3039 37570 : pathnode->qual = qual;
3040 :
3041 37570 : cost_agg(&pathnode->path, root,
3042 : aggstrategy, aggcosts,
3043 : list_length(groupClause), numGroups,
3044 : qual,
3045 : subpath->startup_cost, subpath->total_cost,
3046 37570 : subpath->rows, subpath->pathtarget->width);
3047 :
3048 : /* add tlist eval cost for each output row */
3049 37570 : pathnode->path.startup_cost += target->cost.startup;
3050 75140 : pathnode->path.total_cost += target->cost.startup +
3051 37570 : target->cost.per_tuple * pathnode->path.rows;
3052 :
3053 37570 : return pathnode;
3054 : }
3055 :
3056 : /*
3057 : * create_groupingsets_path
3058 : * Creates a pathnode that represents performing GROUPING SETS aggregation
3059 : *
3060 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3061 : * The input path's result must be sorted to match the last entry in
3062 : * rollup_groupclauses.
3063 : *
3064 : * 'rel' is the parent relation associated with the result
3065 : * 'subpath' is the path representing the source of data
3066 : * 'target' is the PathTarget to be computed
3067 : * 'having_qual' is the HAVING quals if any
3068 : * 'rollups' is a list of RollupData nodes
3069 : * 'agg_costs' contains cost info about the aggregate functions to be computed
3070 : * 'numGroups' is the estimated total number of groups
3071 : */
3072 : GroupingSetsPath *
3073 1128 : create_groupingsets_path(PlannerInfo *root,
3074 : RelOptInfo *rel,
3075 : Path *subpath,
3076 : List *having_qual,
3077 : AggStrategy aggstrategy,
3078 : List *rollups,
3079 : const AggClauseCosts *agg_costs,
3080 : double numGroups)
3081 : {
3082 1128 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
3083 1128 : PathTarget *target = rel->reltarget;
3084 : ListCell *lc;
3085 1128 : bool is_first = true;
3086 1128 : bool is_first_sort = true;
3087 :
3088 : /* The topmost generated Plan node will be an Agg */
3089 1128 : pathnode->path.pathtype = T_Agg;
3090 1128 : pathnode->path.parent = rel;
3091 1128 : pathnode->path.pathtarget = target;
3092 1128 : pathnode->path.param_info = subpath->param_info;
3093 1128 : pathnode->path.parallel_aware = false;
3094 1620 : pathnode->path.parallel_safe = rel->consider_parallel &&
3095 492 : subpath->parallel_safe;
3096 1128 : pathnode->path.parallel_workers = subpath->parallel_workers;
3097 1128 : pathnode->subpath = subpath;
3098 :
3099 : /*
3100 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3101 : * to AGG_HASHED, here if possible.
3102 : */
3103 1616 : if (aggstrategy == AGG_SORTED &&
3104 488 : list_length(rollups) == 1 &&
3105 236 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3106 28 : aggstrategy = AGG_PLAIN;
3107 :
3108 1644 : if (aggstrategy == AGG_MIXED &&
3109 516 : list_length(rollups) == 1)
3110 0 : aggstrategy = AGG_HASHED;
3111 :
3112 : /*
3113 : * Output will be in sorted order by group_pathkeys if, and only if, there
3114 : * is a single rollup operation on a non-empty list of grouping
3115 : * expressions.
3116 : */
3117 1128 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3118 208 : pathnode->path.pathkeys = root->group_pathkeys;
3119 : else
3120 920 : pathnode->path.pathkeys = NIL;
3121 :
3122 1128 : pathnode->aggstrategy = aggstrategy;
3123 1128 : pathnode->rollups = rollups;
3124 1128 : pathnode->qual = having_qual;
3125 1128 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3126 :
3127 : Assert(rollups != NIL);
3128 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3129 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3130 :
3131 3944 : foreach(lc, rollups)
3132 : {
3133 2816 : RollupData *rollup = lfirst(lc);
3134 2816 : List *gsets = rollup->gsets;
3135 2816 : int numGroupCols = list_length(linitial(gsets));
3136 :
3137 : /*
3138 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3139 : * (already-sorted) input, and following ones do their own sort.
3140 : *
3141 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3142 : *
3143 : * In AGG_MIXED mode, the first rollups are hashed, the first
3144 : * non-hashed one takes the (already-sorted) input, and following ones
3145 : * do their own sort.
3146 : */
3147 2816 : if (is_first)
3148 : {
3149 1128 : cost_agg(&pathnode->path, root,
3150 : aggstrategy,
3151 : agg_costs,
3152 : numGroupCols,
3153 : rollup->numGroups,
3154 : having_qual,
3155 : subpath->startup_cost,
3156 : subpath->total_cost,
3157 : subpath->rows,
3158 1128 : subpath->pathtarget->width);
3159 1128 : is_first = false;
3160 1128 : if (!rollup->is_hashed)
3161 488 : is_first_sort = false;
3162 : }
3163 : else
3164 : {
3165 : Path sort_path; /* dummy for result of cost_sort */
3166 : Path agg_path; /* dummy for result of cost_agg */
3167 :
3168 1688 : if (rollup->is_hashed || is_first_sort)
3169 : {
3170 : /*
3171 : * Account for cost of aggregation, but don't charge input
3172 : * cost again
3173 : */
3174 1268 : cost_agg(&agg_path, root,
3175 1268 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3176 : agg_costs,
3177 : numGroupCols,
3178 : rollup->numGroups,
3179 : having_qual,
3180 : 0.0, 0.0,
3181 : subpath->rows,
3182 1268 : subpath->pathtarget->width);
3183 1784 : if (!rollup->is_hashed)
3184 516 : is_first_sort = false;
3185 : }
3186 : else
3187 : {
3188 : /* Account for cost of sort, but don't charge input cost again */
3189 840 : cost_sort(&sort_path, root, NIL,
3190 : 0.0,
3191 : subpath->rows,
3192 420 : subpath->pathtarget->width,
3193 : 0.0,
3194 : work_mem,
3195 : -1.0);
3196 :
3197 : /* Account for cost of aggregation */
3198 :
3199 420 : cost_agg(&agg_path, root,
3200 : AGG_SORTED,
3201 : agg_costs,
3202 : numGroupCols,
3203 : rollup->numGroups,
3204 : having_qual,
3205 : sort_path.startup_cost,
3206 : sort_path.total_cost,
3207 : sort_path.rows,
3208 420 : subpath->pathtarget->width);
3209 : }
3210 :
3211 1688 : pathnode->path.total_cost += agg_path.total_cost;
3212 1688 : pathnode->path.rows += agg_path.rows;
3213 : }
3214 : }
3215 :
3216 : /* add tlist eval cost for each output row */
3217 1128 : pathnode->path.startup_cost += target->cost.startup;
3218 2256 : pathnode->path.total_cost += target->cost.startup +
3219 1128 : target->cost.per_tuple * pathnode->path.rows;
3220 :
3221 1128 : return pathnode;
3222 : }
3223 :
3224 : /*
3225 : * create_minmaxagg_path
3226 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3227 : *
3228 : * 'rel' is the parent relation associated with the result
3229 : * 'target' is the PathTarget to be computed
3230 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3231 : * 'quals' is the HAVING quals if any
3232 : */
3233 : MinMaxAggPath *
3234 398 : create_minmaxagg_path(PlannerInfo *root,
3235 : RelOptInfo *rel,
3236 : PathTarget *target,
3237 : List *mmaggregates,
3238 : List *quals)
3239 : {
3240 398 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3241 : Cost initplan_cost;
3242 : ListCell *lc;
3243 :
3244 : /* The topmost generated Plan node will be a Result */
3245 398 : pathnode->path.pathtype = T_Result;
3246 398 : pathnode->path.parent = rel;
3247 398 : pathnode->path.pathtarget = target;
3248 : /* For now, assume we are above any joins, so no parameterization */
3249 398 : pathnode->path.param_info = NULL;
3250 398 : pathnode->path.parallel_aware = false;
3251 : /* A MinMaxAggPath implies use of subplans, so cannot be parallel-safe */
3252 398 : pathnode->path.parallel_safe = false;
3253 398 : pathnode->path.parallel_workers = 0;
3254 : /* Result is one unordered row */
3255 398 : pathnode->path.rows = 1;
3256 398 : pathnode->path.pathkeys = NIL;
3257 :
3258 398 : pathnode->mmaggregates = mmaggregates;
3259 398 : pathnode->quals = quals;
3260 :
3261 : /* Calculate cost of all the initplans ... */
3262 398 : initplan_cost = 0;
3263 820 : foreach(lc, mmaggregates)
3264 : {
3265 422 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3266 :
3267 422 : initplan_cost += mminfo->pathcost;
3268 : }
3269 :
3270 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3271 398 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3272 1194 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3273 796 : target->cost.per_tuple + cpu_tuple_cost;
3274 :
3275 : /*
3276 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3277 : * rowcount estimate should be 1 no matter what the qual is.
3278 : */
3279 398 : if (quals)
3280 : {
3281 : QualCost qual_cost;
3282 :
3283 0 : cost_qual_eval(&qual_cost, quals, root);
3284 0 : pathnode->path.startup_cost += qual_cost.startup;
3285 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3286 : }
3287 :
3288 398 : return pathnode;
3289 : }
3290 :
3291 : /*
3292 : * create_windowagg_path
3293 : * Creates a pathnode that represents computation of window functions
3294 : *
3295 : * 'rel' is the parent relation associated with the result
3296 : * 'subpath' is the path representing the source of data
3297 : * 'target' is the PathTarget to be computed
3298 : * 'windowFuncs' is a list of WindowFunc structs
3299 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3300 : *
3301 : * The input must be sorted according to the WindowClause's PARTITION keys
3302 : * plus ORDER BY keys.
3303 : */
3304 : WindowAggPath *
3305 1300 : create_windowagg_path(PlannerInfo *root,
3306 : RelOptInfo *rel,
3307 : Path *subpath,
3308 : PathTarget *target,
3309 : List *windowFuncs,
3310 : WindowClause *winclause)
3311 : {
3312 1300 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3313 :
3314 1300 : pathnode->path.pathtype = T_WindowAgg;
3315 1300 : pathnode->path.parent = rel;
3316 1300 : pathnode->path.pathtarget = target;
3317 : /* For now, assume we are above any joins, so no parameterization */
3318 1300 : pathnode->path.param_info = NULL;
3319 1300 : pathnode->path.parallel_aware = false;
3320 1300 : pathnode->path.parallel_safe = rel->consider_parallel &&
3321 0 : subpath->parallel_safe;
3322 1300 : pathnode->path.parallel_workers = subpath->parallel_workers;
3323 : /* WindowAgg preserves the input sort order */
3324 1300 : pathnode->path.pathkeys = subpath->pathkeys;
3325 :
3326 1300 : pathnode->subpath = subpath;
3327 1300 : pathnode->winclause = winclause;
3328 :
3329 : /*
3330 : * For costing purposes, assume that there are no redundant partitioning
3331 : * or ordering columns; it's not worth the trouble to deal with that
3332 : * corner case here. So we just pass the unmodified list lengths to
3333 : * cost_windowagg.
3334 : */
3335 1300 : cost_windowagg(&pathnode->path, root,
3336 : windowFuncs,
3337 1300 : list_length(winclause->partitionClause),
3338 1300 : list_length(winclause->orderClause),
3339 : subpath->startup_cost,
3340 : subpath->total_cost,
3341 : subpath->rows);
3342 :
3343 : /* add tlist eval cost for each output row */
3344 1300 : pathnode->path.startup_cost += target->cost.startup;
3345 2600 : pathnode->path.total_cost += target->cost.startup +
3346 1300 : target->cost.per_tuple * pathnode->path.rows;
3347 :
3348 1300 : return pathnode;
3349 : }
3350 :
3351 : /*
3352 : * create_setop_path
3353 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3354 : *
3355 : * 'rel' is the parent relation associated with the result
3356 : * 'subpath' is the path representing the source of data
3357 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3358 : * 'strategy' is the implementation strategy (sorted or hashed)
3359 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3360 : * 'flagColIdx' is the column number where the flag column will be, if any
3361 : * 'firstFlag' is the flag value for the first input relation when hashing;
3362 : * or -1 when sorting
3363 : * 'numGroups' is the estimated number of distinct groups
3364 : * 'outputRows' is the estimated number of output rows
3365 : */
3366 : SetOpPath *
3367 340 : create_setop_path(PlannerInfo *root,
3368 : RelOptInfo *rel,
3369 : Path *subpath,
3370 : SetOpCmd cmd,
3371 : SetOpStrategy strategy,
3372 : List *distinctList,
3373 : AttrNumber flagColIdx,
3374 : int firstFlag,
3375 : double numGroups,
3376 : double outputRows)
3377 : {
3378 340 : SetOpPath *pathnode = makeNode(SetOpPath);
3379 :
3380 340 : pathnode->path.pathtype = T_SetOp;
3381 340 : pathnode->path.parent = rel;
3382 : /* SetOp doesn't project, so use source path's pathtarget */
3383 340 : pathnode->path.pathtarget = subpath->pathtarget;
3384 : /* For now, assume we are above any joins, so no parameterization */
3385 340 : pathnode->path.param_info = NULL;
3386 340 : pathnode->path.parallel_aware = false;
3387 340 : pathnode->path.parallel_safe = rel->consider_parallel &&
3388 0 : subpath->parallel_safe;
3389 340 : pathnode->path.parallel_workers = subpath->parallel_workers;
3390 : /* SetOp preserves the input sort order if in sort mode */
3391 340 : pathnode->path.pathkeys =
3392 340 : (strategy == SETOP_SORTED) ? subpath->pathkeys : NIL;
3393 :
3394 340 : pathnode->subpath = subpath;
3395 340 : pathnode->cmd = cmd;
3396 340 : pathnode->strategy = strategy;
3397 340 : pathnode->distinctList = distinctList;
3398 340 : pathnode->flagColIdx = flagColIdx;
3399 340 : pathnode->firstFlag = firstFlag;
3400 340 : pathnode->numGroups = numGroups;
3401 :
3402 : /*
3403 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3404 : * all columns get compared at most of the tuples.
3405 : */
3406 340 : pathnode->path.startup_cost = subpath->startup_cost;
3407 680 : pathnode->path.total_cost = subpath->total_cost +
3408 340 : cpu_operator_cost * subpath->rows * list_length(distinctList);
3409 340 : pathnode->path.rows = outputRows;
3410 :
3411 340 : return pathnode;
3412 : }
3413 :
3414 : /*
3415 : * create_recursiveunion_path
3416 : * Creates a pathnode that represents a recursive UNION node
3417 : *
3418 : * 'rel' is the parent relation associated with the result
3419 : * 'leftpath' is the source of data for the non-recursive term
3420 : * 'rightpath' is the source of data for the recursive term
3421 : * 'target' is the PathTarget to be computed
3422 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3423 : * 'wtParam' is the ID of Param representing work table
3424 : * 'numGroups' is the estimated number of groups
3425 : *
3426 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3427 : */
3428 : RecursiveUnionPath *
3429 344 : create_recursiveunion_path(PlannerInfo *root,
3430 : RelOptInfo *rel,
3431 : Path *leftpath,
3432 : Path *rightpath,
3433 : PathTarget *target,
3434 : List *distinctList,
3435 : int wtParam,
3436 : double numGroups)
3437 : {
3438 344 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3439 :
3440 344 : pathnode->path.pathtype = T_RecursiveUnion;
3441 344 : pathnode->path.parent = rel;
3442 344 : pathnode->path.pathtarget = target;
3443 : /* For now, assume we are above any joins, so no parameterization */
3444 344 : pathnode->path.param_info = NULL;
3445 344 : pathnode->path.parallel_aware = false;
3446 688 : pathnode->path.parallel_safe = rel->consider_parallel &&
3447 344 : leftpath->parallel_safe && rightpath->parallel_safe;
3448 : /* Foolish, but we'll do it like joins for now: */
3449 344 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3450 : /* RecursiveUnion result is always unsorted */
3451 344 : pathnode->path.pathkeys = NIL;
3452 :
3453 344 : pathnode->leftpath = leftpath;
3454 344 : pathnode->rightpath = rightpath;
3455 344 : pathnode->distinctList = distinctList;
3456 344 : pathnode->wtParam = wtParam;
3457 344 : pathnode->numGroups = numGroups;
3458 :
3459 344 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3460 :
3461 344 : return pathnode;
3462 : }
3463 :
3464 : /*
3465 : * create_lockrows_path
3466 : * Creates a pathnode that represents acquiring row locks
3467 : *
3468 : * 'rel' is the parent relation associated with the result
3469 : * 'subpath' is the path representing the source of data
3470 : * 'rowMarks' is a list of PlanRowMark's
3471 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3472 : */
3473 : LockRowsPath *
3474 5376 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3475 : Path *subpath, List *rowMarks, int epqParam)
3476 : {
3477 5376 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3478 :
3479 5376 : pathnode->path.pathtype = T_LockRows;
3480 5376 : pathnode->path.parent = rel;
3481 : /* LockRows doesn't project, so use source path's pathtarget */
3482 5376 : pathnode->path.pathtarget = subpath->pathtarget;
3483 : /* For now, assume we are above any joins, so no parameterization */
3484 5376 : pathnode->path.param_info = NULL;
3485 5376 : pathnode->path.parallel_aware = false;
3486 5376 : pathnode->path.parallel_safe = false;
3487 5376 : pathnode->path.parallel_workers = 0;
3488 5376 : pathnode->path.rows = subpath->rows;
3489 :
3490 : /*
3491 : * The result cannot be assumed sorted, since locking might cause the sort
3492 : * key columns to be replaced with new values.
3493 : */
3494 5376 : pathnode->path.pathkeys = NIL;
3495 :
3496 5376 : pathnode->subpath = subpath;
3497 5376 : pathnode->rowMarks = rowMarks;
3498 5376 : pathnode->epqParam = epqParam;
3499 :
3500 : /*
3501 : * We should charge something extra for the costs of row locking and
3502 : * possible refetches, but it's hard to say how much. For now, use
3503 : * cpu_tuple_cost per row.
3504 : */
3505 5376 : pathnode->path.startup_cost = subpath->startup_cost;
3506 10752 : pathnode->path.total_cost = subpath->total_cost +
3507 5376 : cpu_tuple_cost * subpath->rows;
3508 :
3509 5376 : return pathnode;
3510 : }
3511 :
3512 : /*
3513 : * create_modifytable_path
3514 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE mods
3515 : *
3516 : * 'rel' is the parent relation associated with the result
3517 : * 'operation' is the operation type
3518 : * 'canSetTag' is true if we set the command tag/es_processed
3519 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3520 : * 'rootRelation' is the partitioned table root RT index, or 0 if none
3521 : * 'partColsUpdated' is true if any partitioning columns are being updated,
3522 : * either from the target relation or a descendent partitioned table.
3523 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3524 : * 'subpaths' is a list of Path(s) producing source data (one per rel)
3525 : * 'subroots' is a list of PlannerInfo structs (one per rel)
3526 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3527 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3528 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3529 : * 'onconflict' is the ON CONFLICT clause, or NULL
3530 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3531 : */
3532 : ModifyTablePath *
3533 77908 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3534 : CmdType operation, bool canSetTag,
3535 : Index nominalRelation, Index rootRelation,
3536 : bool partColsUpdated,
3537 : List *resultRelations, List *subpaths,
3538 : List *subroots,
3539 : List *withCheckOptionLists, List *returningLists,
3540 : List *rowMarks, OnConflictExpr *onconflict,
3541 : int epqParam)
3542 : {
3543 77908 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3544 : double total_size;
3545 : ListCell *lc;
3546 :
3547 : Assert(list_length(resultRelations) == list_length(subpaths));
3548 : Assert(list_length(resultRelations) == list_length(subroots));
3549 : Assert(withCheckOptionLists == NIL ||
3550 : list_length(resultRelations) == list_length(withCheckOptionLists));
3551 : Assert(returningLists == NIL ||
3552 : list_length(resultRelations) == list_length(returningLists));
3553 :
3554 77908 : pathnode->path.pathtype = T_ModifyTable;
3555 77908 : pathnode->path.parent = rel;
3556 : /* pathtarget is not interesting, just make it minimally valid */
3557 77908 : pathnode->path.pathtarget = rel->reltarget;
3558 : /* For now, assume we are above any joins, so no parameterization */
3559 77908 : pathnode->path.param_info = NULL;
3560 77908 : pathnode->path.parallel_aware = false;
3561 77908 : pathnode->path.parallel_safe = false;
3562 77908 : pathnode->path.parallel_workers = 0;
3563 77908 : pathnode->path.pathkeys = NIL;
3564 :
3565 : /*
3566 : * Compute cost & rowcount as sum of subpath costs & rowcounts.
3567 : *
3568 : * Currently, we don't charge anything extra for the actual table
3569 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3570 : * expressions if any. It would only be window dressing, since
3571 : * ModifyTable is always a top-level node and there is no way for the
3572 : * costs to change any higher-level planning choices. But we might want
3573 : * to make it look better sometime.
3574 : */
3575 77908 : pathnode->path.startup_cost = 0;
3576 77908 : pathnode->path.total_cost = 0;
3577 77908 : pathnode->path.rows = 0;
3578 77908 : total_size = 0;
3579 157014 : foreach(lc, subpaths)
3580 : {
3581 79106 : Path *subpath = (Path *) lfirst(lc);
3582 :
3583 79106 : if (lc == list_head(subpaths)) /* first node? */
3584 77908 : pathnode->path.startup_cost = subpath->startup_cost;
3585 79106 : pathnode->path.total_cost += subpath->total_cost;
3586 79106 : if (returningLists != NIL)
3587 : {
3588 1828 : pathnode->path.rows += subpath->rows;
3589 1828 : total_size += subpath->pathtarget->width * subpath->rows;
3590 : }
3591 : }
3592 :
3593 : /*
3594 : * Set width to the average width of the subpath outputs. XXX this is
3595 : * totally wrong: we should return an average of the RETURNING tlist
3596 : * widths. But it's what happened historically, and improving it is a task
3597 : * for another day.
3598 : */
3599 77908 : if (pathnode->path.rows > 0)
3600 1630 : total_size /= pathnode->path.rows;
3601 77908 : pathnode->path.pathtarget->width = rint(total_size);
3602 :
3603 77908 : pathnode->operation = operation;
3604 77908 : pathnode->canSetTag = canSetTag;
3605 77908 : pathnode->nominalRelation = nominalRelation;
3606 77908 : pathnode->rootRelation = rootRelation;
3607 77908 : pathnode->partColsUpdated = partColsUpdated;
3608 77908 : pathnode->resultRelations = resultRelations;
3609 77908 : pathnode->subpaths = subpaths;
3610 77908 : pathnode->subroots = subroots;
3611 77908 : pathnode->withCheckOptionLists = withCheckOptionLists;
3612 77908 : pathnode->returningLists = returningLists;
3613 77908 : pathnode->rowMarks = rowMarks;
3614 77908 : pathnode->onconflict = onconflict;
3615 77908 : pathnode->epqParam = epqParam;
3616 :
3617 77908 : return pathnode;
3618 : }
3619 :
3620 : /*
3621 : * create_limit_path
3622 : * Creates a pathnode that represents performing LIMIT/OFFSET
3623 : *
3624 : * In addition to providing the actual OFFSET and LIMIT expressions,
3625 : * the caller must provide estimates of their values for costing purposes.
3626 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3627 : * the clause not being present, and -1 means it's present but we could
3628 : * not estimate its value.
3629 : *
3630 : * 'rel' is the parent relation associated with the result
3631 : * 'subpath' is the path representing the source of data
3632 : * 'limitOffset' is the actual OFFSET expression, or NULL
3633 : * 'limitCount' is the actual LIMIT expression, or NULL
3634 : * 'offset_est' is the estimated value of the OFFSET expression
3635 : * 'count_est' is the estimated value of the LIMIT expression
3636 : */
3637 : LimitPath *
3638 4152 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3639 : Path *subpath,
3640 : Node *limitOffset, Node *limitCount,
3641 : LimitOption limitOption,
3642 : int64 offset_est, int64 count_est)
3643 : {
3644 4152 : LimitPath *pathnode = makeNode(LimitPath);
3645 :
3646 4152 : pathnode->path.pathtype = T_Limit;
3647 4152 : pathnode->path.parent = rel;
3648 : /* Limit doesn't project, so use source path's pathtarget */
3649 4152 : pathnode->path.pathtarget = subpath->pathtarget;
3650 : /* For now, assume we are above any joins, so no parameterization */
3651 4152 : pathnode->path.param_info = NULL;
3652 4152 : pathnode->path.parallel_aware = false;
3653 5768 : pathnode->path.parallel_safe = rel->consider_parallel &&
3654 1616 : subpath->parallel_safe;
3655 4152 : pathnode->path.parallel_workers = subpath->parallel_workers;
3656 4152 : pathnode->path.rows = subpath->rows;
3657 4152 : pathnode->path.startup_cost = subpath->startup_cost;
3658 4152 : pathnode->path.total_cost = subpath->total_cost;
3659 4152 : pathnode->path.pathkeys = subpath->pathkeys;
3660 4152 : pathnode->subpath = subpath;
3661 4152 : pathnode->limitOffset = limitOffset;
3662 4152 : pathnode->limitCount = limitCount;
3663 4152 : pathnode->limitOption = limitOption;
3664 :
3665 : /*
3666 : * Adjust the output rows count and costs according to the offset/limit.
3667 : */
3668 4152 : adjust_limit_rows_costs(&pathnode->path.rows,
3669 : &pathnode->path.startup_cost,
3670 : &pathnode->path.total_cost,
3671 : offset_est, count_est);
3672 :
3673 4152 : return pathnode;
3674 : }
3675 :
3676 : /*
3677 : * adjust_limit_rows_costs
3678 : * Adjust the size and cost estimates for a LimitPath node according to the
3679 : * offset/limit.
3680 : *
3681 : * This is only a cosmetic issue if we are at top level, but if we are
3682 : * building a subquery then it's important to report correct info to the outer
3683 : * planner.
3684 : *
3685 : * When the offset or count couldn't be estimated, use 10% of the estimated
3686 : * number of rows emitted from the subpath.
3687 : *
3688 : * XXX we don't bother to add eval costs of the offset/limit expressions
3689 : * themselves to the path costs. In theory we should, but in most cases those
3690 : * expressions are trivial and it's just not worth the trouble.
3691 : */
3692 : void
3693 4304 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3694 : Cost *startup_cost, /* in/out parameter */
3695 : Cost *total_cost, /* in/out parameter */
3696 : int64 offset_est,
3697 : int64 count_est)
3698 : {
3699 4304 : double input_rows = *rows;
3700 4304 : Cost input_startup_cost = *startup_cost;
3701 4304 : Cost input_total_cost = *total_cost;
3702 :
3703 4304 : if (offset_est != 0)
3704 : {
3705 : double offset_rows;
3706 :
3707 606 : if (offset_est > 0)
3708 590 : offset_rows = (double) offset_est;
3709 : else
3710 16 : offset_rows = clamp_row_est(input_rows * 0.10);
3711 606 : if (offset_rows > *rows)
3712 20 : offset_rows = *rows;
3713 606 : if (input_rows > 0)
3714 1212 : *startup_cost +=
3715 606 : (input_total_cost - input_startup_cost)
3716 606 : * offset_rows / input_rows;
3717 606 : *rows -= offset_rows;
3718 606 : if (*rows < 1)
3719 20 : *rows = 1;
3720 : }
3721 :
3722 4304 : if (count_est != 0)
3723 : {
3724 : double count_rows;
3725 :
3726 4272 : if (count_est > 0)
3727 4268 : count_rows = (double) count_est;
3728 : else
3729 4 : count_rows = clamp_row_est(input_rows * 0.10);
3730 4272 : if (count_rows > *rows)
3731 52 : count_rows = *rows;
3732 4272 : if (input_rows > 0)
3733 8544 : *total_cost = *startup_cost +
3734 4272 : (input_total_cost - input_startup_cost)
3735 4272 : * count_rows / input_rows;
3736 4272 : *rows = count_rows;
3737 4272 : if (*rows < 1)
3738 0 : *rows = 1;
3739 : }
3740 4304 : }
3741 :
3742 :
3743 : /*
3744 : * reparameterize_path
3745 : * Attempt to modify a Path to have greater parameterization
3746 : *
3747 : * We use this to attempt to bring all child paths of an appendrel to the
3748 : * same parameterization level, ensuring that they all enforce the same set
3749 : * of join quals (and thus that that parameterization can be attributed to
3750 : * an append path built from such paths). Currently, only a few path types
3751 : * are supported here, though more could be added at need. We return NULL
3752 : * if we can't reparameterize the given path.
3753 : *
3754 : * Note: we intentionally do not pass created paths to add_path(); it would
3755 : * possibly try to delete them on the grounds of being cost-inferior to the
3756 : * paths they were made from, and we don't want that. Paths made here are
3757 : * not necessarily of general-purpose usefulness, but they can be useful
3758 : * as members of an append path.
3759 : */
3760 : Path *
3761 212 : reparameterize_path(PlannerInfo *root, Path *path,
3762 : Relids required_outer,
3763 : double loop_count)
3764 : {
3765 212 : RelOptInfo *rel = path->parent;
3766 :
3767 : /* Can only increase, not decrease, path's parameterization */
3768 212 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
3769 0 : return NULL;
3770 212 : switch (path->pathtype)
3771 : {
3772 148 : case T_SeqScan:
3773 148 : return create_seqscan_path(root, rel, required_outer, 0);
3774 0 : case T_SampleScan:
3775 0 : return (Path *) create_samplescan_path(root, rel, required_outer);
3776 0 : case T_IndexScan:
3777 : case T_IndexOnlyScan:
3778 : {
3779 0 : IndexPath *ipath = (IndexPath *) path;
3780 0 : IndexPath *newpath = makeNode(IndexPath);
3781 :
3782 : /*
3783 : * We can't use create_index_path directly, and would not want
3784 : * to because it would re-compute the indexqual conditions
3785 : * which is wasted effort. Instead we hack things a bit:
3786 : * flat-copy the path node, revise its param_info, and redo
3787 : * the cost estimate.
3788 : */
3789 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3790 0 : newpath->path.param_info =
3791 0 : get_baserel_parampathinfo(root, rel, required_outer);
3792 0 : cost_index(newpath, root, loop_count, false);
3793 0 : return (Path *) newpath;
3794 : }
3795 0 : case T_BitmapHeapScan:
3796 : {
3797 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3798 :
3799 0 : return (Path *) create_bitmap_heap_path(root,
3800 : rel,
3801 : bpath->bitmapqual,
3802 : required_outer,
3803 : loop_count, 0);
3804 : }
3805 0 : case T_SubqueryScan:
3806 : {
3807 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
3808 :
3809 0 : return (Path *) create_subqueryscan_path(root,
3810 : rel,
3811 : spath->subpath,
3812 : spath->path.pathkeys,
3813 : required_outer);
3814 : }
3815 32 : case T_Result:
3816 : /* Supported only for RTE_RESULT scan paths */
3817 32 : if (IsA(path, Path))
3818 32 : return create_resultscan_path(root, rel, required_outer);
3819 0 : break;
3820 0 : case T_Append:
3821 : {
3822 0 : AppendPath *apath = (AppendPath *) path;
3823 0 : List *childpaths = NIL;
3824 0 : List *partialpaths = NIL;
3825 : int i;
3826 : ListCell *lc;
3827 :
3828 : /* Reparameterize the children */
3829 0 : i = 0;
3830 0 : foreach(lc, apath->subpaths)
3831 : {
3832 0 : Path *spath = (Path *) lfirst(lc);
3833 :
3834 0 : spath = reparameterize_path(root, spath,
3835 : required_outer,
3836 : loop_count);
3837 0 : if (spath == NULL)
3838 0 : return NULL;
3839 : /* We have to re-split the regular and partial paths */
3840 0 : if (i < apath->first_partial_path)
3841 0 : childpaths = lappend(childpaths, spath);
3842 : else
3843 0 : partialpaths = lappend(partialpaths, spath);
3844 0 : i++;
3845 : }
3846 0 : return (Path *)
3847 0 : create_append_path(root, rel, childpaths, partialpaths,
3848 : apath->path.pathkeys, required_outer,
3849 : apath->path.parallel_workers,
3850 0 : apath->path.parallel_aware,
3851 : apath->partitioned_rels,
3852 : -1);
3853 : }
3854 32 : default:
3855 32 : break;
3856 : }
3857 32 : return NULL;
3858 : }
3859 :
3860 : /*
3861 : * reparameterize_path_by_child
3862 : * Given a path parameterized by the parent of the given child relation,
3863 : * translate the path to be parameterized by the given child relation.
3864 : *
3865 : * The function creates a new path of the same type as the given path, but
3866 : * parameterized by the given child relation. Most fields from the original
3867 : * path can simply be flat-copied, but any expressions must be adjusted to
3868 : * refer to the correct varnos, and any paths must be recursively
3869 : * reparameterized. Other fields that refer to specific relids also need
3870 : * adjustment.
3871 : *
3872 : * The cost, number of rows, width and parallel path properties depend upon
3873 : * path->parent, which does not change during the translation. Hence those
3874 : * members are copied as they are.
3875 : *
3876 : * If the given path can not be reparameterized, the function returns NULL.
3877 : */
3878 : Path *
3879 3004 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
3880 : RelOptInfo *child_rel)
3881 : {
3882 :
3883 : #define FLAT_COPY_PATH(newnode, node, nodetype) \
3884 : ( (newnode) = makeNode(nodetype), \
3885 : memcpy((newnode), (node), sizeof(nodetype)) )
3886 :
3887 : #define ADJUST_CHILD_ATTRS(node) \
3888 : ((node) = \
3889 : (List *) adjust_appendrel_attrs_multilevel(root, (Node *) (node), \
3890 : child_rel->relids, \
3891 : child_rel->top_parent_relids))
3892 :
3893 : #define REPARAMETERIZE_CHILD_PATH(path) \
3894 : do { \
3895 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
3896 : if ((path) == NULL) \
3897 : return NULL; \
3898 : } while(0)
3899 :
3900 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
3901 : do { \
3902 : if ((pathlist) != NIL) \
3903 : { \
3904 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
3905 : child_rel); \
3906 : if ((pathlist) == NIL) \
3907 : return NULL; \
3908 : } \
3909 : } while(0)
3910 :
3911 : Path *new_path;
3912 : ParamPathInfo *new_ppi;
3913 : ParamPathInfo *old_ppi;
3914 : Relids required_outer;
3915 :
3916 : /*
3917 : * If the path is not parameterized by parent of the given relation, it
3918 : * doesn't need reparameterization.
3919 : */
3920 3004 : if (!path->param_info ||
3921 2948 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
3922 160 : return path;
3923 :
3924 : /*
3925 : * If possible, reparameterize the given path, making a copy.
3926 : *
3927 : * This function is currently only applied to the inner side of a nestloop
3928 : * join that is being partitioned by the partitionwise-join code. Hence,
3929 : * we need only support path types that plausibly arise in that context.
3930 : * (In particular, supporting sorted path types would be a waste of code
3931 : * and cycles: even if we translated them here, they'd just lose in
3932 : * subsequent cost comparisons.) If we do see an unsupported path type,
3933 : * that just means we won't be able to generate a partitionwise-join plan
3934 : * using that path type.
3935 : */
3936 2844 : switch (nodeTag(path))
3937 : {
3938 288 : case T_Path:
3939 288 : FLAT_COPY_PATH(new_path, path, Path);
3940 288 : break;
3941 :
3942 2024 : case T_IndexPath:
3943 : {
3944 : IndexPath *ipath;
3945 :
3946 2024 : FLAT_COPY_PATH(ipath, path, IndexPath);
3947 2024 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
3948 2024 : new_path = (Path *) ipath;
3949 : }
3950 2024 : break;
3951 :
3952 32 : case T_BitmapHeapPath:
3953 : {
3954 : BitmapHeapPath *bhpath;
3955 :
3956 32 : FLAT_COPY_PATH(bhpath, path, BitmapHeapPath);
3957 32 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
3958 32 : new_path = (Path *) bhpath;
3959 : }
3960 32 : break;
3961 :
3962 16 : case T_BitmapAndPath:
3963 : {
3964 : BitmapAndPath *bapath;
3965 :
3966 16 : FLAT_COPY_PATH(bapath, path, BitmapAndPath);
3967 16 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
3968 16 : new_path = (Path *) bapath;
3969 : }
3970 16 : break;
3971 :
3972 16 : case T_BitmapOrPath:
3973 : {
3974 : BitmapOrPath *bopath;
3975 :
3976 16 : FLAT_COPY_PATH(bopath, path, BitmapOrPath);
3977 16 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
3978 16 : new_path = (Path *) bopath;
3979 : }
3980 16 : break;
3981 :
3982 52 : case T_ForeignPath:
3983 : {
3984 : ForeignPath *fpath;
3985 : ReparameterizeForeignPathByChild_function rfpc_func;
3986 :
3987 52 : FLAT_COPY_PATH(fpath, path, ForeignPath);
3988 52 : if (fpath->fdw_outerpath)
3989 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
3990 :
3991 : /* Hand over to FDW if needed. */
3992 52 : rfpc_func =
3993 52 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
3994 52 : if (rfpc_func)
3995 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
3996 : child_rel);
3997 52 : new_path = (Path *) fpath;
3998 : }
3999 52 : break;
4000 :
4001 0 : case T_CustomPath:
4002 : {
4003 : CustomPath *cpath;
4004 :
4005 0 : FLAT_COPY_PATH(cpath, path, CustomPath);
4006 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
4007 0 : if (cpath->methods &&
4008 0 : cpath->methods->ReparameterizeCustomPathByChild)
4009 0 : cpath->custom_private =
4010 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4011 : cpath->custom_private,
4012 : child_rel);
4013 0 : new_path = (Path *) cpath;
4014 : }
4015 0 : break;
4016 :
4017 200 : case T_NestPath:
4018 : {
4019 : JoinPath *jpath;
4020 :
4021 200 : FLAT_COPY_PATH(jpath, path, NestPath);
4022 :
4023 200 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4024 200 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4025 200 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4026 200 : new_path = (Path *) jpath;
4027 : }
4028 200 : break;
4029 :
4030 24 : case T_MergePath:
4031 : {
4032 : JoinPath *jpath;
4033 : MergePath *mpath;
4034 :
4035 24 : FLAT_COPY_PATH(mpath, path, MergePath);
4036 :
4037 24 : jpath = (JoinPath *) mpath;
4038 24 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4039 24 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4040 24 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4041 24 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4042 24 : new_path = (Path *) mpath;
4043 : }
4044 24 : break;
4045 :
4046 112 : case T_HashPath:
4047 : {
4048 : JoinPath *jpath;
4049 : HashPath *hpath;
4050 :
4051 112 : FLAT_COPY_PATH(hpath, path, HashPath);
4052 :
4053 112 : jpath = (JoinPath *) hpath;
4054 112 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4055 112 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4056 112 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4057 112 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4058 112 : new_path = (Path *) hpath;
4059 : }
4060 112 : break;
4061 :
4062 80 : case T_AppendPath:
4063 : {
4064 : AppendPath *apath;
4065 :
4066 80 : FLAT_COPY_PATH(apath, path, AppendPath);
4067 80 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4068 80 : new_path = (Path *) apath;
4069 : }
4070 80 : break;
4071 :
4072 0 : case T_GatherPath:
4073 : {
4074 : GatherPath *gpath;
4075 :
4076 0 : FLAT_COPY_PATH(gpath, path, GatherPath);
4077 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4078 0 : new_path = (Path *) gpath;
4079 : }
4080 0 : break;
4081 :
4082 0 : default:
4083 :
4084 : /* We don't know how to reparameterize this path. */
4085 0 : return NULL;
4086 : }
4087 :
4088 : /*
4089 : * Adjust the parameterization information, which refers to the topmost
4090 : * parent. The topmost parent can be multiple levels away from the given
4091 : * child, hence use multi-level expression adjustment routines.
4092 : */
4093 2844 : old_ppi = new_path->param_info;
4094 : required_outer =
4095 2844 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4096 : child_rel->relids,
4097 : child_rel->top_parent_relids);
4098 :
4099 : /* If we already have a PPI for this parameterization, just return it */
4100 2844 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4101 :
4102 : /*
4103 : * If not, build a new one and link it to the list of PPIs. For the same
4104 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4105 : * context the given RelOptInfo is in.
4106 : */
4107 2844 : if (new_ppi == NULL)
4108 : {
4109 : MemoryContext oldcontext;
4110 1316 : RelOptInfo *rel = path->parent;
4111 :
4112 1316 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4113 :
4114 1316 : new_ppi = makeNode(ParamPathInfo);
4115 1316 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4116 1316 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4117 1316 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4118 1316 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4119 1316 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4120 :
4121 1316 : MemoryContextSwitchTo(oldcontext);
4122 : }
4123 2844 : bms_free(required_outer);
4124 :
4125 2844 : new_path->param_info = new_ppi;
4126 :
4127 : /*
4128 : * Adjust the path target if the parent of the outer relation is
4129 : * referenced in the targetlist. This can happen when only the parent of
4130 : * outer relation is laterally referenced in this relation.
4131 : */
4132 2844 : if (bms_overlap(path->parent->lateral_relids,
4133 2844 : child_rel->top_parent_relids))
4134 : {
4135 768 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4136 768 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4137 : }
4138 :
4139 2844 : return new_path;
4140 : }
4141 :
4142 : /*
4143 : * reparameterize_pathlist_by_child
4144 : * Helper function to reparameterize a list of paths by given child rel.
4145 : */
4146 : static List *
4147 112 : reparameterize_pathlist_by_child(PlannerInfo *root,
4148 : List *pathlist,
4149 : RelOptInfo *child_rel)
4150 : {
4151 : ListCell *lc;
4152 112 : List *result = NIL;
4153 :
4154 336 : foreach(lc, pathlist)
4155 : {
4156 224 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4157 : child_rel);
4158 :
4159 224 : if (path == NULL)
4160 : {
4161 0 : list_free(result);
4162 0 : return NIL;
4163 : }
4164 :
4165 224 : result = lappend(result, path);
4166 : }
4167 :
4168 112 : return result;
4169 : }
|
