Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pathnode.c
4 : * Routines to manipulate pathlists and create path nodes
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/optimizer/util/pathnode.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <math.h>
18 :
19 : #include "access/htup_details.h"
20 : #include "executor/nodeSetOp.h"
21 : #include "foreign/fdwapi.h"
22 : #include "miscadmin.h"
23 : #include "nodes/extensible.h"
24 : #include "optimizer/appendinfo.h"
25 : #include "optimizer/clauses.h"
26 : #include "optimizer/cost.h"
27 : #include "optimizer/optimizer.h"
28 : #include "optimizer/pathnode.h"
29 : #include "optimizer/paths.h"
30 : #include "optimizer/planmain.h"
31 : #include "optimizer/tlist.h"
32 : #include "parser/parsetree.h"
33 : #include "utils/memutils.h"
34 : #include "utils/selfuncs.h"
35 :
36 : typedef enum
37 : {
38 : COSTS_EQUAL, /* path costs are fuzzily equal */
39 : COSTS_BETTER1, /* first path is cheaper than second */
40 : COSTS_BETTER2, /* second path is cheaper than first */
41 : COSTS_DIFFERENT, /* neither path dominates the other on cost */
42 : } PathCostComparison;
43 :
44 : /*
45 : * STD_FUZZ_FACTOR is the normal fuzz factor for compare_path_costs_fuzzily.
46 : * XXX is it worth making this user-controllable? It provides a tradeoff
47 : * between planner runtime and the accuracy of path cost comparisons.
48 : */
49 : #define STD_FUZZ_FACTOR 1.01
50 :
51 : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
52 : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
53 : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
54 : List *pathlist,
55 : RelOptInfo *child_rel);
56 : static bool pathlist_is_reparameterizable_by_child(List *pathlist,
57 : RelOptInfo *child_rel);
58 :
59 :
60 : /*****************************************************************************
61 : * MISC. PATH UTILITIES
62 : *****************************************************************************/
63 :
64 : /*
65 : * compare_path_costs
66 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
67 : * or more expensive than path2 for the specified criterion.
68 : */
69 : int
70 1124184 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
71 : {
72 : /* Number of disabled nodes, if different, trumps all else. */
73 1124184 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
74 : {
75 2712 : if (path1->disabled_nodes < path2->disabled_nodes)
76 2712 : return -1;
77 : else
78 0 : return +1;
79 : }
80 :
81 1121472 : if (criterion == STARTUP_COST)
82 : {
83 564642 : if (path1->startup_cost < path2->startup_cost)
84 344106 : return -1;
85 220536 : if (path1->startup_cost > path2->startup_cost)
86 113508 : return +1;
87 :
88 : /*
89 : * If paths have the same startup cost (not at all unlikely), order
90 : * them by total cost.
91 : */
92 107028 : if (path1->total_cost < path2->total_cost)
93 54916 : return -1;
94 52112 : if (path1->total_cost > path2->total_cost)
95 5114 : return +1;
96 : }
97 : else
98 : {
99 556830 : if (path1->total_cost < path2->total_cost)
100 519566 : return -1;
101 37264 : if (path1->total_cost > path2->total_cost)
102 10696 : return +1;
103 :
104 : /*
105 : * If paths have the same total cost, order them by startup cost.
106 : */
107 26568 : if (path1->startup_cost < path2->startup_cost)
108 2602 : return -1;
109 23966 : if (path1->startup_cost > path2->startup_cost)
110 30 : return +1;
111 : }
112 70934 : return 0;
113 : }
114 :
115 : /*
116 : * compare_fractional_path_costs
117 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
118 : * or more expensive than path2 for fetching the specified fraction
119 : * of the total tuples.
120 : *
121 : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
122 : * path with the cheaper total_cost.
123 : */
124 : int
125 6348 : compare_fractional_path_costs(Path *path1, Path *path2,
126 : double fraction)
127 : {
128 : Cost cost1,
129 : cost2;
130 :
131 : /* Number of disabled nodes, if different, trumps all else. */
132 6348 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
133 : {
134 36 : if (path1->disabled_nodes < path2->disabled_nodes)
135 36 : return -1;
136 : else
137 0 : return +1;
138 : }
139 :
140 6312 : if (fraction <= 0.0 || fraction >= 1.0)
141 1962 : return compare_path_costs(path1, path2, TOTAL_COST);
142 4350 : cost1 = path1->startup_cost +
143 4350 : fraction * (path1->total_cost - path1->startup_cost);
144 4350 : cost2 = path2->startup_cost +
145 4350 : fraction * (path2->total_cost - path2->startup_cost);
146 4350 : if (cost1 < cost2)
147 3586 : return -1;
148 764 : if (cost1 > cost2)
149 764 : return +1;
150 0 : return 0;
151 : }
152 :
153 : /*
154 : * compare_path_costs_fuzzily
155 : * Compare the costs of two paths to see if either can be said to
156 : * dominate the other.
157 : *
158 : * We use fuzzy comparisons so that add_path() can avoid keeping both of
159 : * a pair of paths that really have insignificantly different cost.
160 : *
161 : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
162 : * fraction of the smaller cost that is considered to be a significant
163 : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
164 : * be 1% of the smaller cost.
165 : *
166 : * The two paths are said to have "equal" costs if both startup and total
167 : * costs are fuzzily the same. Path1 is said to be better than path2 if
168 : * it has fuzzily better startup cost and fuzzily no worse total cost,
169 : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
170 : * Path2 is better than path1 if the reverse holds. Finally, if one path
171 : * is fuzzily better than the other on startup cost and fuzzily worse on
172 : * total cost, we just say that their costs are "different", since neither
173 : * dominates the other across the whole performance spectrum.
174 : *
175 : * This function also enforces a policy rule that paths for which the relevant
176 : * one of parent->consider_startup and parent->consider_param_startup is false
177 : * cannot survive comparisons solely on the grounds of good startup cost, so
178 : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
179 : * (But if total costs are fuzzily equal, we compare startup costs anyway,
180 : * in hopes of eliminating one path or the other.)
181 : */
182 : static PathCostComparison
183 4616218 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
184 : {
185 : #define CONSIDER_PATH_STARTUP_COST(p) \
186 : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
187 :
188 : /* Number of disabled nodes, if different, trumps all else. */
189 4616218 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
190 : {
191 31148 : if (path1->disabled_nodes < path2->disabled_nodes)
192 16888 : return COSTS_BETTER1;
193 : else
194 14260 : return COSTS_BETTER2;
195 : }
196 :
197 : /*
198 : * Check total cost first since it's more likely to be different; many
199 : * paths have zero startup cost.
200 : */
201 4585070 : if (path1->total_cost > path2->total_cost * fuzz_factor)
202 : {
203 : /* path1 fuzzily worse on total cost */
204 2402912 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
205 137456 : path2->startup_cost > path1->startup_cost * fuzz_factor)
206 : {
207 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
208 92758 : return COSTS_DIFFERENT;
209 : }
210 : /* else path2 dominates */
211 2310154 : return COSTS_BETTER2;
212 : }
213 2182158 : if (path2->total_cost > path1->total_cost * fuzz_factor)
214 : {
215 : /* path2 fuzzily worse on total cost */
216 1143914 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
217 61356 : path1->startup_cost > path2->startup_cost * fuzz_factor)
218 : {
219 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
220 40724 : return COSTS_DIFFERENT;
221 : }
222 : /* else path1 dominates */
223 1103190 : return COSTS_BETTER1;
224 : }
225 : /* fuzzily the same on total cost ... */
226 1038244 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
227 : {
228 : /* ... but path1 fuzzily worse on startup, so path2 wins */
229 378560 : return COSTS_BETTER2;
230 : }
231 659684 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
232 : {
233 : /* ... but path2 fuzzily worse on startup, so path1 wins */
234 65510 : return COSTS_BETTER1;
235 : }
236 : /* fuzzily the same on both costs */
237 594174 : return COSTS_EQUAL;
238 :
239 : #undef CONSIDER_PATH_STARTUP_COST
240 : }
241 :
242 : /*
243 : * set_cheapest
244 : * Find the minimum-cost paths from among a relation's paths,
245 : * and save them in the rel's cheapest-path fields.
246 : *
247 : * cheapest_total_path is normally the cheapest-total-cost unparameterized
248 : * path; but if there are no unparameterized paths, we assign it to be the
249 : * best (cheapest least-parameterized) parameterized path. However, only
250 : * unparameterized paths are considered candidates for cheapest_startup_path,
251 : * so that will be NULL if there are no unparameterized paths.
252 : *
253 : * The cheapest_parameterized_paths list collects all parameterized paths
254 : * that have survived the add_path() tournament for this relation. (Since
255 : * add_path ignores pathkeys for a parameterized path, these will be paths
256 : * that have best cost or best row count for their parameterization. We
257 : * may also have both a parallel-safe and a non-parallel-safe path in some
258 : * cases for the same parameterization in some cases, but this should be
259 : * relatively rare since, most typically, all paths for the same relation
260 : * will be parallel-safe or none of them will.)
261 : *
262 : * cheapest_parameterized_paths always includes the cheapest-total
263 : * unparameterized path, too, if there is one; the users of that list find
264 : * it more convenient if that's included.
265 : *
266 : * This is normally called only after we've finished constructing the path
267 : * list for the rel node.
268 : */
269 : void
270 2119902 : set_cheapest(RelOptInfo *parent_rel)
271 : {
272 : Path *cheapest_startup_path;
273 : Path *cheapest_total_path;
274 : Path *best_param_path;
275 : List *parameterized_paths;
276 : ListCell *p;
277 :
278 : Assert(IsA(parent_rel, RelOptInfo));
279 :
280 2119902 : if (parent_rel->pathlist == NIL)
281 0 : elog(ERROR, "could not devise a query plan for the given query");
282 :
283 2119902 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
284 2119902 : parameterized_paths = NIL;
285 :
286 4815508 : foreach(p, parent_rel->pathlist)
287 : {
288 2695606 : Path *path = (Path *) lfirst(p);
289 : int cmp;
290 :
291 2695606 : if (path->param_info)
292 : {
293 : /* Parameterized path, so add it to parameterized_paths */
294 142690 : parameterized_paths = lappend(parameterized_paths, path);
295 :
296 : /*
297 : * If we have an unparameterized cheapest-total, we no longer care
298 : * about finding the best parameterized path, so move on.
299 : */
300 142690 : if (cheapest_total_path)
301 30014 : continue;
302 :
303 : /*
304 : * Otherwise, track the best parameterized path, which is the one
305 : * with least total cost among those of the minimum
306 : * parameterization.
307 : */
308 112676 : if (best_param_path == NULL)
309 102754 : best_param_path = path;
310 : else
311 : {
312 9922 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
313 9922 : PATH_REQ_OUTER(best_param_path)))
314 : {
315 60 : case BMS_EQUAL:
316 : /* keep the cheaper one */
317 60 : if (compare_path_costs(path, best_param_path,
318 : TOTAL_COST) < 0)
319 0 : best_param_path = path;
320 60 : break;
321 764 : case BMS_SUBSET1:
322 : /* new path is less-parameterized */
323 764 : best_param_path = path;
324 764 : break;
325 6 : case BMS_SUBSET2:
326 : /* old path is less-parameterized, keep it */
327 6 : break;
328 9092 : case BMS_DIFFERENT:
329 :
330 : /*
331 : * This means that neither path has the least possible
332 : * parameterization for the rel. We'll sit on the old
333 : * path until something better comes along.
334 : */
335 9092 : break;
336 : }
337 : }
338 : }
339 : else
340 : {
341 : /* Unparameterized path, so consider it for cheapest slots */
342 2552916 : if (cheapest_total_path == NULL)
343 : {
344 2107152 : cheapest_startup_path = cheapest_total_path = path;
345 2107152 : continue;
346 : }
347 :
348 : /*
349 : * If we find two paths of identical costs, try to keep the
350 : * better-sorted one. The paths might have unrelated sort
351 : * orderings, in which case we can only guess which might be
352 : * better to keep, but if one is superior then we definitely
353 : * should keep that one.
354 : */
355 445764 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
356 445764 : if (cmp > 0 ||
357 378 : (cmp == 0 &&
358 378 : compare_pathkeys(cheapest_startup_path->pathkeys,
359 : path->pathkeys) == PATHKEYS_BETTER2))
360 83720 : cheapest_startup_path = path;
361 :
362 445764 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
363 445764 : if (cmp > 0 ||
364 48 : (cmp == 0 &&
365 48 : compare_pathkeys(cheapest_total_path->pathkeys,
366 : path->pathkeys) == PATHKEYS_BETTER2))
367 0 : cheapest_total_path = path;
368 : }
369 : }
370 :
371 : /* Add cheapest unparameterized path, if any, to parameterized_paths */
372 2119902 : if (cheapest_total_path)
373 2107152 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
374 :
375 : /*
376 : * If there is no unparameterized path, use the best parameterized path as
377 : * cheapest_total_path (but not as cheapest_startup_path).
378 : */
379 2119902 : if (cheapest_total_path == NULL)
380 12750 : cheapest_total_path = best_param_path;
381 : Assert(cheapest_total_path != NULL);
382 :
383 2119902 : parent_rel->cheapest_startup_path = cheapest_startup_path;
384 2119902 : parent_rel->cheapest_total_path = cheapest_total_path;
385 2119902 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
386 2119902 : }
387 :
388 : /*
389 : * add_path
390 : * Consider a potential implementation path for the specified parent rel,
391 : * and add it to the rel's pathlist if it is worthy of consideration.
392 : *
393 : * A path is worthy if it has a better sort order (better pathkeys) or
394 : * cheaper cost (as defined below), or generates fewer rows, than any
395 : * existing path that has the same or superset parameterization rels. We
396 : * also consider parallel-safe paths more worthy than others.
397 : *
398 : * Cheaper cost can mean either a cheaper total cost or a cheaper startup
399 : * cost; if one path is cheaper in one of these aspects and another is
400 : * cheaper in the other, we keep both. However, when some path type is
401 : * disabled (e.g. due to enable_seqscan=false), the number of times that
402 : * a disabled path type is used is considered to be a higher-order
403 : * component of the cost. Hence, if path A uses no disabled path type,
404 : * and path B uses 1 or more disabled path types, A is cheaper, no matter
405 : * what we estimate for the startup and total costs. The startup and total
406 : * cost essentially act as a tiebreak when comparing paths that use equal
407 : * numbers of disabled path nodes; but in practice this tiebreak is almost
408 : * always used, since normally no path types are disabled.
409 : *
410 : * In addition to possibly adding new_path, we also remove from the rel's
411 : * pathlist any old paths that are dominated by new_path --- that is,
412 : * new_path is cheaper, at least as well ordered, generates no more rows,
413 : * requires no outer rels not required by the old path, and is no less
414 : * parallel-safe.
415 : *
416 : * In most cases, a path with a superset parameterization will generate
417 : * fewer rows (since it has more join clauses to apply), so that those two
418 : * figures of merit move in opposite directions; this means that a path of
419 : * one parameterization can seldom dominate a path of another. But such
420 : * cases do arise, so we make the full set of checks anyway.
421 : *
422 : * There are two policy decisions embedded in this function, along with
423 : * its sibling add_path_precheck. First, we treat all parameterized paths
424 : * as having NIL pathkeys, so that they cannot win comparisons on the
425 : * basis of sort order. This is to reduce the number of parameterized
426 : * paths that are kept; see discussion in src/backend/optimizer/README.
427 : *
428 : * Second, we only consider cheap startup cost to be interesting if
429 : * parent_rel->consider_startup is true for an unparameterized path, or
430 : * parent_rel->consider_param_startup is true for a parameterized one.
431 : * Again, this allows discarding useless paths sooner.
432 : *
433 : * The pathlist is kept sorted by disabled_nodes and then by total_cost,
434 : * with cheaper paths at the front. Within this routine, that's simply a
435 : * speed hack: doing it that way makes it more likely that we will reject
436 : * an inferior path after a few comparisons, rather than many comparisons.
437 : * However, add_path_precheck relies on this ordering to exit early
438 : * when possible.
439 : *
440 : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
441 : * memory consumption. We dare not try to free the substructure of a Path,
442 : * since much of it may be shared with other Paths or the query tree itself;
443 : * but just recycling discarded Path nodes is a very useful savings in
444 : * a large join tree. We can recycle the List nodes of pathlist, too.
445 : *
446 : * As noted in optimizer/README, deleting a previously-accepted Path is
447 : * safe because we know that Paths of this rel cannot yet be referenced
448 : * from any other rel, such as a higher-level join. However, in some cases
449 : * it is possible that a Path is referenced by another Path for its own
450 : * rel; we must not delete such a Path, even if it is dominated by the new
451 : * Path. Currently this occurs only for IndexPath objects, which may be
452 : * referenced as children of BitmapHeapPaths as well as being paths in
453 : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
454 : *
455 : * 'parent_rel' is the relation entry to which the path corresponds.
456 : * 'new_path' is a potential path for parent_rel.
457 : *
458 : * Returns nothing, but modifies parent_rel->pathlist.
459 : */
460 : void
461 4616822 : add_path(RelOptInfo *parent_rel, Path *new_path)
462 : {
463 4616822 : bool accept_new = true; /* unless we find a superior old path */
464 4616822 : int insert_at = 0; /* where to insert new item */
465 : List *new_path_pathkeys;
466 : ListCell *p1;
467 :
468 : /*
469 : * This is a convenient place to check for query cancel --- no part of the
470 : * planner goes very long without calling add_path().
471 : */
472 4616822 : CHECK_FOR_INTERRUPTS();
473 :
474 : /* Pretend parameterized paths have no pathkeys, per comment above */
475 4616822 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
476 :
477 : /*
478 : * Loop to check proposed new path against old paths. Note it is possible
479 : * for more than one old path to be tossed out because new_path dominates
480 : * it.
481 : */
482 7139634 : foreach(p1, parent_rel->pathlist)
483 : {
484 4247266 : Path *old_path = (Path *) lfirst(p1);
485 4247266 : bool remove_old = false; /* unless new proves superior */
486 : PathCostComparison costcmp;
487 : PathKeysComparison keyscmp;
488 : BMS_Comparison outercmp;
489 :
490 : /*
491 : * Do a fuzzy cost comparison with standard fuzziness limit.
492 : */
493 4247266 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
494 : STD_FUZZ_FACTOR);
495 :
496 : /*
497 : * If the two paths compare differently for startup and total cost,
498 : * then we want to keep both, and we can skip comparing pathkeys and
499 : * required_outer rels. If they compare the same, proceed with the
500 : * other comparisons. Row count is checked last. (We make the tests
501 : * in this order because the cost comparison is most likely to turn
502 : * out "different", and the pathkeys comparison next most likely. As
503 : * explained above, row count very seldom makes a difference, so even
504 : * though it's cheap to compare there's not much point in checking it
505 : * earlier.)
506 : */
507 4247266 : if (costcmp != COSTS_DIFFERENT)
508 : {
509 : /* Similarly check to see if either dominates on pathkeys */
510 : List *old_path_pathkeys;
511 :
512 4113850 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
513 4113850 : keyscmp = compare_pathkeys(new_path_pathkeys,
514 : old_path_pathkeys);
515 4113850 : if (keyscmp != PATHKEYS_DIFFERENT)
516 : {
517 3926380 : switch (costcmp)
518 : {
519 401778 : case COSTS_EQUAL:
520 401778 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
521 401778 : PATH_REQ_OUTER(old_path));
522 401778 : if (keyscmp == PATHKEYS_BETTER1)
523 : {
524 6958 : if ((outercmp == BMS_EQUAL ||
525 6958 : outercmp == BMS_SUBSET1) &&
526 6958 : new_path->rows <= old_path->rows &&
527 6950 : new_path->parallel_safe >= old_path->parallel_safe)
528 6950 : remove_old = true; /* new dominates old */
529 : }
530 394820 : else if (keyscmp == PATHKEYS_BETTER2)
531 : {
532 20210 : if ((outercmp == BMS_EQUAL ||
533 20210 : outercmp == BMS_SUBSET2) &&
534 20210 : new_path->rows >= old_path->rows &&
535 20210 : new_path->parallel_safe <= old_path->parallel_safe)
536 20210 : accept_new = false; /* old dominates new */
537 : }
538 : else /* keyscmp == PATHKEYS_EQUAL */
539 : {
540 374610 : if (outercmp == BMS_EQUAL)
541 : {
542 : /*
543 : * Same pathkeys and outer rels, and fuzzily
544 : * the same cost, so keep just one; to decide
545 : * which, first check parallel-safety, then
546 : * rows, then do a fuzzy cost comparison with
547 : * very small fuzz limit. (We used to do an
548 : * exact cost comparison, but that results in
549 : * annoying platform-specific plan variations
550 : * due to roundoff in the cost estimates.) If
551 : * things are still tied, arbitrarily keep
552 : * only the old path. Notice that we will
553 : * keep only the old path even if the
554 : * less-fuzzy comparison decides the startup
555 : * and total costs compare differently.
556 : */
557 369238 : if (new_path->parallel_safe >
558 369238 : old_path->parallel_safe)
559 42 : remove_old = true; /* new dominates old */
560 369196 : else if (new_path->parallel_safe <
561 369196 : old_path->parallel_safe)
562 54 : accept_new = false; /* old dominates new */
563 369142 : else if (new_path->rows < old_path->rows)
564 0 : remove_old = true; /* new dominates old */
565 369142 : else if (new_path->rows > old_path->rows)
566 190 : accept_new = false; /* old dominates new */
567 368952 : else if (compare_path_costs_fuzzily(new_path,
568 : old_path,
569 : 1.0000000001) == COSTS_BETTER1)
570 16294 : remove_old = true; /* new dominates old */
571 : else
572 352658 : accept_new = false; /* old equals or
573 : * dominates new */
574 : }
575 5372 : else if (outercmp == BMS_SUBSET1 &&
576 1208 : new_path->rows <= old_path->rows &&
577 1188 : new_path->parallel_safe >= old_path->parallel_safe)
578 1188 : remove_old = true; /* new dominates old */
579 4184 : else if (outercmp == BMS_SUBSET2 &&
580 3466 : new_path->rows >= old_path->rows &&
581 3430 : new_path->parallel_safe <= old_path->parallel_safe)
582 3430 : accept_new = false; /* old dominates new */
583 : /* else different parameterizations, keep both */
584 : }
585 401778 : break;
586 1141896 : case COSTS_BETTER1:
587 1141896 : if (keyscmp != PATHKEYS_BETTER2)
588 : {
589 774322 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
590 774322 : PATH_REQ_OUTER(old_path));
591 774322 : if ((outercmp == BMS_EQUAL ||
592 662120 : outercmp == BMS_SUBSET1) &&
593 662120 : new_path->rows <= old_path->rows &&
594 656504 : new_path->parallel_safe >= old_path->parallel_safe)
595 653826 : remove_old = true; /* new dominates old */
596 : }
597 1141896 : break;
598 2382706 : case COSTS_BETTER2:
599 2382706 : if (keyscmp != PATHKEYS_BETTER1)
600 : {
601 1515556 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
602 1515556 : PATH_REQ_OUTER(old_path));
603 1515556 : if ((outercmp == BMS_EQUAL ||
604 1427608 : outercmp == BMS_SUBSET2) &&
605 1427608 : new_path->rows >= old_path->rows &&
606 1349718 : new_path->parallel_safe <= old_path->parallel_safe)
607 1347912 : accept_new = false; /* old dominates new */
608 : }
609 2382706 : break;
610 0 : case COSTS_DIFFERENT:
611 :
612 : /*
613 : * can't get here, but keep this case to keep compiler
614 : * quiet
615 : */
616 0 : break;
617 : }
618 : }
619 : }
620 :
621 : /*
622 : * Remove current element from pathlist if dominated by new.
623 : */
624 4247266 : if (remove_old)
625 : {
626 678300 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
627 : p1);
628 :
629 : /*
630 : * Delete the data pointed-to by the deleted cell, if possible
631 : */
632 678300 : if (!IsA(old_path, IndexPath))
633 658460 : pfree(old_path);
634 : }
635 : else
636 : {
637 : /*
638 : * new belongs after this old path if it has more disabled nodes
639 : * or if it has the same number of nodes but a greater total cost
640 : */
641 3568966 : if (new_path->disabled_nodes > old_path->disabled_nodes ||
642 3554706 : (new_path->disabled_nodes == old_path->disabled_nodes &&
643 3553740 : new_path->total_cost >= old_path->total_cost))
644 2957738 : insert_at = foreach_current_index(p1) + 1;
645 : }
646 :
647 : /*
648 : * If we found an old path that dominates new_path, we can quit
649 : * scanning the pathlist; we will not add new_path, and we assume
650 : * new_path cannot dominate any other elements of the pathlist.
651 : */
652 4247266 : if (!accept_new)
653 1724454 : break;
654 : }
655 :
656 4616822 : if (accept_new)
657 : {
658 : /* Accept the new path: insert it at proper place in pathlist */
659 2892368 : parent_rel->pathlist =
660 2892368 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
661 : }
662 : else
663 : {
664 : /* Reject and recycle the new path */
665 1724454 : if (!IsA(new_path, IndexPath))
666 1622148 : pfree(new_path);
667 : }
668 4616822 : }
669 :
670 : /*
671 : * add_path_precheck
672 : * Check whether a proposed new path could possibly get accepted.
673 : * We assume we know the path's pathkeys and parameterization accurately,
674 : * and have lower bounds for its costs.
675 : *
676 : * Note that we do not know the path's rowcount, since getting an estimate for
677 : * that is too expensive to do before prechecking. We assume here that paths
678 : * of a superset parameterization will generate fewer rows; if that holds,
679 : * then paths with different parameterizations cannot dominate each other
680 : * and so we can simply ignore existing paths of another parameterization.
681 : * (In the infrequent cases where that rule of thumb fails, add_path will
682 : * get rid of the inferior path.)
683 : *
684 : * At the time this is called, we haven't actually built a Path structure,
685 : * so the required information has to be passed piecemeal.
686 : */
687 : bool
688 5266768 : add_path_precheck(RelOptInfo *parent_rel, int disabled_nodes,
689 : Cost startup_cost, Cost total_cost,
690 : List *pathkeys, Relids required_outer)
691 : {
692 : List *new_path_pathkeys;
693 : bool consider_startup;
694 : ListCell *p1;
695 :
696 : /* Pretend parameterized paths have no pathkeys, per add_path policy */
697 5266768 : new_path_pathkeys = required_outer ? NIL : pathkeys;
698 :
699 : /* Decide whether new path's startup cost is interesting */
700 5266768 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
701 :
702 6741942 : foreach(p1, parent_rel->pathlist)
703 : {
704 6396686 : Path *old_path = (Path *) lfirst(p1);
705 : PathKeysComparison keyscmp;
706 :
707 : /*
708 : * Since the pathlist is sorted by disabled_nodes and then by
709 : * total_cost, we can stop looking once we reach a path with more
710 : * disabled nodes, or the same number of disabled nodes plus a
711 : * total_cost larger than the new path's.
712 : */
713 6396686 : if (unlikely(old_path->disabled_nodes != disabled_nodes))
714 : {
715 12966 : if (disabled_nodes < old_path->disabled_nodes)
716 318 : break;
717 : }
718 6383720 : else if (total_cost <= old_path->total_cost * STD_FUZZ_FACTOR)
719 1911882 : break;
720 :
721 : /*
722 : * We are looking for an old_path with the same parameterization (and
723 : * by assumption the same rowcount) that dominates the new path on
724 : * pathkeys as well as both cost metrics. If we find one, we can
725 : * reject the new path.
726 : *
727 : * Cost comparisons here should match compare_path_costs_fuzzily.
728 : */
729 : /* new path can win on startup cost only if consider_startup */
730 4484486 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
731 2112574 : !consider_startup)
732 : {
733 : /* new path loses on cost, so check pathkeys... */
734 : List *old_path_pathkeys;
735 :
736 4381084 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
737 4381084 : keyscmp = compare_pathkeys(new_path_pathkeys,
738 : old_path_pathkeys);
739 4381084 : if (keyscmp == PATHKEYS_EQUAL ||
740 : keyscmp == PATHKEYS_BETTER2)
741 : {
742 : /* new path does not win on pathkeys... */
743 3075590 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
744 : {
745 : /* Found an old path that dominates the new one */
746 3009312 : return false;
747 : }
748 : }
749 : }
750 : }
751 :
752 2257456 : return true;
753 : }
754 :
755 : /*
756 : * add_partial_path
757 : * Like add_path, our goal here is to consider whether a path is worthy
758 : * of being kept around, but the considerations here are a bit different.
759 : * A partial path is one which can be executed in any number of workers in
760 : * parallel such that each worker will generate a subset of the path's
761 : * overall result.
762 : *
763 : * As in add_path, the partial_pathlist is kept sorted with the cheapest
764 : * total path in front. This is depended on by multiple places, which
765 : * just take the front entry as the cheapest path without searching.
766 : *
767 : * We don't generate parameterized partial paths for several reasons. Most
768 : * importantly, they're not safe to execute, because there's nothing to
769 : * make sure that a parallel scan within the parameterized portion of the
770 : * plan is running with the same value in every worker at the same time.
771 : * Fortunately, it seems unlikely to be worthwhile anyway, because having
772 : * each worker scan the entire outer relation and a subset of the inner
773 : * relation will generally be a terrible plan. The inner (parameterized)
774 : * side of the plan will be small anyway. There could be rare cases where
775 : * this wins big - e.g. if join order constraints put a 1-row relation on
776 : * the outer side of the topmost join with a parameterized plan on the inner
777 : * side - but we'll have to be content not to handle such cases until
778 : * somebody builds an executor infrastructure that can cope with them.
779 : *
780 : * Because we don't consider parameterized paths here, we also don't
781 : * need to consider the row counts as a measure of quality: every path will
782 : * produce the same number of rows. Neither do we need to consider startup
783 : * costs: parallelism is only used for plans that will be run to completion.
784 : * Therefore, this routine is much simpler than add_path: it needs to
785 : * consider only disabled nodes, pathkeys and total cost.
786 : *
787 : * As with add_path, we pfree paths that are found to be dominated by
788 : * another partial path; this requires that there be no other references to
789 : * such paths yet. Hence, GatherPaths must not be created for a rel until
790 : * we're done creating all partial paths for it. Unlike add_path, we don't
791 : * take an exception for IndexPaths as partial index paths won't be
792 : * referenced by partial BitmapHeapPaths.
793 : */
794 : void
795 306998 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
796 : {
797 306998 : bool accept_new = true; /* unless we find a superior old path */
798 306998 : int insert_at = 0; /* where to insert new item */
799 : ListCell *p1;
800 :
801 : /* Check for query cancel. */
802 306998 : CHECK_FOR_INTERRUPTS();
803 :
804 : /* Path to be added must be parallel safe. */
805 : Assert(new_path->parallel_safe);
806 :
807 : /* Relation should be OK for parallelism, too. */
808 : Assert(parent_rel->consider_parallel);
809 :
810 : /*
811 : * As in add_path, throw out any paths which are dominated by the new
812 : * path, but throw out the new path if some existing path dominates it.
813 : */
814 427766 : foreach(p1, parent_rel->partial_pathlist)
815 : {
816 251958 : Path *old_path = (Path *) lfirst(p1);
817 251958 : bool remove_old = false; /* unless new proves superior */
818 : PathKeysComparison keyscmp;
819 :
820 : /* Compare pathkeys. */
821 251958 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
822 :
823 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
824 251958 : if (keyscmp != PATHKEYS_DIFFERENT)
825 : {
826 251748 : if (unlikely(new_path->disabled_nodes != old_path->disabled_nodes))
827 : {
828 1532 : if (new_path->disabled_nodes > old_path->disabled_nodes)
829 1004 : accept_new = false;
830 : else
831 528 : remove_old = true;
832 : }
833 250216 : else if (new_path->total_cost > old_path->total_cost
834 250216 : * STD_FUZZ_FACTOR)
835 : {
836 : /* New path costs more; keep it only if pathkeys are better. */
837 123294 : if (keyscmp != PATHKEYS_BETTER1)
838 89264 : accept_new = false;
839 : }
840 126922 : else if (old_path->total_cost > new_path->total_cost
841 126922 : * STD_FUZZ_FACTOR)
842 : {
843 : /* Old path costs more; keep it only if pathkeys are better. */
844 85400 : if (keyscmp != PATHKEYS_BETTER2)
845 30048 : remove_old = true;
846 : }
847 41522 : else if (keyscmp == PATHKEYS_BETTER1)
848 : {
849 : /* Costs are about the same, new path has better pathkeys. */
850 24 : remove_old = true;
851 : }
852 41498 : else if (keyscmp == PATHKEYS_BETTER2)
853 : {
854 : /* Costs are about the same, old path has better pathkeys. */
855 2034 : accept_new = false;
856 : }
857 39464 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
858 : {
859 : /* Pathkeys are the same, and the old path costs more. */
860 576 : remove_old = true;
861 : }
862 : else
863 : {
864 : /*
865 : * Pathkeys are the same, and new path isn't materially
866 : * cheaper.
867 : */
868 38888 : accept_new = false;
869 : }
870 : }
871 :
872 : /*
873 : * Remove current element from partial_pathlist if dominated by new.
874 : */
875 251958 : if (remove_old)
876 : {
877 31176 : parent_rel->partial_pathlist =
878 31176 : foreach_delete_current(parent_rel->partial_pathlist, p1);
879 31176 : pfree(old_path);
880 : }
881 : else
882 : {
883 : /* new belongs after this old path if it has cost >= old's */
884 220782 : if (new_path->total_cost >= old_path->total_cost)
885 164258 : insert_at = foreach_current_index(p1) + 1;
886 : }
887 :
888 : /*
889 : * If we found an old path that dominates new_path, we can quit
890 : * scanning the partial_pathlist; we will not add new_path, and we
891 : * assume new_path cannot dominate any later path.
892 : */
893 251958 : if (!accept_new)
894 131190 : break;
895 : }
896 :
897 306998 : if (accept_new)
898 : {
899 : /* Accept the new path: insert it at proper place */
900 175808 : parent_rel->partial_pathlist =
901 175808 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
902 : }
903 : else
904 : {
905 : /* Reject and recycle the new path */
906 131190 : pfree(new_path);
907 : }
908 306998 : }
909 :
910 : /*
911 : * add_partial_path_precheck
912 : * Check whether a proposed new partial path could possibly get accepted.
913 : *
914 : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
915 : * since they don't matter for partial paths (see add_partial_path). But
916 : * we do want to make sure we don't add a partial path if there's already
917 : * a complete path that dominates it, since in that case the proposed path
918 : * is surely a loser.
919 : */
920 : bool
921 448746 : add_partial_path_precheck(RelOptInfo *parent_rel, int disabled_nodes,
922 : Cost total_cost, List *pathkeys)
923 : {
924 : ListCell *p1;
925 :
926 : /*
927 : * Our goal here is twofold. First, we want to find out whether this path
928 : * is clearly inferior to some existing partial path. If so, we want to
929 : * reject it immediately. Second, we want to find out whether this path
930 : * is clearly superior to some existing partial path -- at least, modulo
931 : * final cost computations. If so, we definitely want to consider it.
932 : *
933 : * Unlike add_path(), we always compare pathkeys here. This is because we
934 : * expect partial_pathlist to be very short, and getting a definitive
935 : * answer at this stage avoids the need to call add_path_precheck.
936 : */
937 572804 : foreach(p1, parent_rel->partial_pathlist)
938 : {
939 466346 : Path *old_path = (Path *) lfirst(p1);
940 : PathKeysComparison keyscmp;
941 :
942 466346 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
943 466346 : if (keyscmp != PATHKEYS_DIFFERENT)
944 : {
945 466154 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
946 : keyscmp != PATHKEYS_BETTER1)
947 342288 : return false;
948 237016 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
949 : keyscmp != PATHKEYS_BETTER2)
950 113150 : return true;
951 : }
952 : }
953 :
954 : /*
955 : * This path is neither clearly inferior to an existing partial path nor
956 : * clearly good enough that it might replace one. Compare it to
957 : * non-parallel plans. If it loses even before accounting for the cost of
958 : * the Gather node, we should definitely reject it.
959 : *
960 : * Note that we pass the total_cost to add_path_precheck twice. This is
961 : * because it's never advantageous to consider the startup cost of a
962 : * partial path; the resulting plans, if run in parallel, will be run to
963 : * completion.
964 : */
965 106458 : if (!add_path_precheck(parent_rel, disabled_nodes, total_cost, total_cost,
966 : pathkeys, NULL))
967 2826 : return false;
968 :
969 103632 : return true;
970 : }
971 :
972 :
973 : /*****************************************************************************
974 : * PATH NODE CREATION ROUTINES
975 : *****************************************************************************/
976 :
977 : /*
978 : * create_seqscan_path
979 : * Creates a path corresponding to a sequential scan, returning the
980 : * pathnode.
981 : */
982 : Path *
983 433148 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
984 : Relids required_outer, int parallel_workers)
985 : {
986 433148 : Path *pathnode = makeNode(Path);
987 :
988 433148 : pathnode->pathtype = T_SeqScan;
989 433148 : pathnode->parent = rel;
990 433148 : pathnode->pathtarget = rel->reltarget;
991 433148 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
992 : required_outer);
993 433148 : pathnode->parallel_aware = (parallel_workers > 0);
994 433148 : pathnode->parallel_safe = rel->consider_parallel;
995 433148 : pathnode->parallel_workers = parallel_workers;
996 433148 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
997 :
998 433148 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
999 :
1000 433148 : return pathnode;
1001 : }
1002 :
1003 : /*
1004 : * create_samplescan_path
1005 : * Creates a path node for a sampled table scan.
1006 : */
1007 : Path *
1008 306 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
1009 : {
1010 306 : Path *pathnode = makeNode(Path);
1011 :
1012 306 : pathnode->pathtype = T_SampleScan;
1013 306 : pathnode->parent = rel;
1014 306 : pathnode->pathtarget = rel->reltarget;
1015 306 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1016 : required_outer);
1017 306 : pathnode->parallel_aware = false;
1018 306 : pathnode->parallel_safe = rel->consider_parallel;
1019 306 : pathnode->parallel_workers = 0;
1020 306 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
1021 :
1022 306 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
1023 :
1024 306 : return pathnode;
1025 : }
1026 :
1027 : /*
1028 : * create_index_path
1029 : * Creates a path node for an index scan.
1030 : *
1031 : * 'index' is a usable index.
1032 : * 'indexclauses' is a list of IndexClause nodes representing clauses
1033 : * to be enforced as qual conditions in the scan.
1034 : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
1035 : * to be used as index ordering operators in the scan.
1036 : * 'indexorderbycols' is an integer list of index column numbers (zero based)
1037 : * the ordering operators can be used with.
1038 : * 'pathkeys' describes the ordering of the path.
1039 : * 'indexscandir' is either ForwardScanDirection or BackwardScanDirection.
1040 : * 'indexonly' is true if an index-only scan is wanted.
1041 : * 'required_outer' is the set of outer relids for a parameterized path.
1042 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1043 : * estimates of caching behavior.
1044 : * 'partial_path' is true if constructing a parallel index scan path.
1045 : *
1046 : * Returns the new path node.
1047 : */
1048 : IndexPath *
1049 802444 : create_index_path(PlannerInfo *root,
1050 : IndexOptInfo *index,
1051 : List *indexclauses,
1052 : List *indexorderbys,
1053 : List *indexorderbycols,
1054 : List *pathkeys,
1055 : ScanDirection indexscandir,
1056 : bool indexonly,
1057 : Relids required_outer,
1058 : double loop_count,
1059 : bool partial_path)
1060 : {
1061 802444 : IndexPath *pathnode = makeNode(IndexPath);
1062 802444 : RelOptInfo *rel = index->rel;
1063 :
1064 802444 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1065 802444 : pathnode->path.parent = rel;
1066 802444 : pathnode->path.pathtarget = rel->reltarget;
1067 802444 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1068 : required_outer);
1069 802444 : pathnode->path.parallel_aware = false;
1070 802444 : pathnode->path.parallel_safe = rel->consider_parallel;
1071 802444 : pathnode->path.parallel_workers = 0;
1072 802444 : pathnode->path.pathkeys = pathkeys;
1073 :
1074 802444 : pathnode->indexinfo = index;
1075 802444 : pathnode->indexclauses = indexclauses;
1076 802444 : pathnode->indexorderbys = indexorderbys;
1077 802444 : pathnode->indexorderbycols = indexorderbycols;
1078 802444 : pathnode->indexscandir = indexscandir;
1079 :
1080 802444 : cost_index(pathnode, root, loop_count, partial_path);
1081 :
1082 802444 : return pathnode;
1083 : }
1084 :
1085 : /*
1086 : * create_bitmap_heap_path
1087 : * Creates a path node for a bitmap scan.
1088 : *
1089 : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1090 : * 'required_outer' is the set of outer relids for a parameterized path.
1091 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1092 : * estimates of caching behavior.
1093 : *
1094 : * loop_count should match the value used when creating the component
1095 : * IndexPaths.
1096 : */
1097 : BitmapHeapPath *
1098 350754 : create_bitmap_heap_path(PlannerInfo *root,
1099 : RelOptInfo *rel,
1100 : Path *bitmapqual,
1101 : Relids required_outer,
1102 : double loop_count,
1103 : int parallel_degree)
1104 : {
1105 350754 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1106 :
1107 350754 : pathnode->path.pathtype = T_BitmapHeapScan;
1108 350754 : pathnode->path.parent = rel;
1109 350754 : pathnode->path.pathtarget = rel->reltarget;
1110 350754 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1111 : required_outer);
1112 350754 : pathnode->path.parallel_aware = (parallel_degree > 0);
1113 350754 : pathnode->path.parallel_safe = rel->consider_parallel;
1114 350754 : pathnode->path.parallel_workers = parallel_degree;
1115 350754 : pathnode->path.pathkeys = NIL; /* always unordered */
1116 :
1117 350754 : pathnode->bitmapqual = bitmapqual;
1118 :
1119 350754 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1120 : pathnode->path.param_info,
1121 : bitmapqual, loop_count);
1122 :
1123 350754 : return pathnode;
1124 : }
1125 :
1126 : /*
1127 : * create_bitmap_and_path
1128 : * Creates a path node representing a BitmapAnd.
1129 : */
1130 : BitmapAndPath *
1131 49662 : create_bitmap_and_path(PlannerInfo *root,
1132 : RelOptInfo *rel,
1133 : List *bitmapquals)
1134 : {
1135 49662 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1136 49662 : Relids required_outer = NULL;
1137 : ListCell *lc;
1138 :
1139 49662 : pathnode->path.pathtype = T_BitmapAnd;
1140 49662 : pathnode->path.parent = rel;
1141 49662 : 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 148986 : foreach(lc, bitmapquals)
1149 : {
1150 99324 : Path *bitmapqual = (Path *) lfirst(lc);
1151 :
1152 99324 : required_outer = bms_add_members(required_outer,
1153 99324 : PATH_REQ_OUTER(bitmapqual));
1154 : }
1155 49662 : 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 49662 : pathnode->path.parallel_aware = false;
1165 49662 : pathnode->path.parallel_safe = rel->consider_parallel;
1166 49662 : pathnode->path.parallel_workers = 0;
1167 :
1168 49662 : pathnode->path.pathkeys = NIL; /* always unordered */
1169 :
1170 49662 : pathnode->bitmapquals = bitmapquals;
1171 :
1172 : /* this sets bitmapselectivity as well as the regular cost fields: */
1173 49662 : cost_bitmap_and_node(pathnode, root);
1174 :
1175 49662 : return pathnode;
1176 : }
1177 :
1178 : /*
1179 : * create_bitmap_or_path
1180 : * Creates a path node representing a BitmapOr.
1181 : */
1182 : BitmapOrPath *
1183 1016 : create_bitmap_or_path(PlannerInfo *root,
1184 : RelOptInfo *rel,
1185 : List *bitmapquals)
1186 : {
1187 1016 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1188 1016 : Relids required_outer = NULL;
1189 : ListCell *lc;
1190 :
1191 1016 : pathnode->path.pathtype = T_BitmapOr;
1192 1016 : pathnode->path.parent = rel;
1193 1016 : pathnode->path.pathtarget = rel->reltarget;
1194 :
1195 : /*
1196 : * Identify the required outer rels as the union of what the child paths
1197 : * depend on. (Alternatively, we could insist that the caller pass this
1198 : * in, but it's more convenient and reliable to compute it here.)
1199 : */
1200 2850 : foreach(lc, bitmapquals)
1201 : {
1202 1834 : Path *bitmapqual = (Path *) lfirst(lc);
1203 :
1204 1834 : required_outer = bms_add_members(required_outer,
1205 1834 : PATH_REQ_OUTER(bitmapqual));
1206 : }
1207 1016 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1208 : required_outer);
1209 :
1210 : /*
1211 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1212 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1213 : * set the flag for this path based only on the relation-level flag,
1214 : * without actually iterating over the list of children.
1215 : */
1216 1016 : pathnode->path.parallel_aware = false;
1217 1016 : pathnode->path.parallel_safe = rel->consider_parallel;
1218 1016 : pathnode->path.parallel_workers = 0;
1219 :
1220 1016 : pathnode->path.pathkeys = NIL; /* always unordered */
1221 :
1222 1016 : pathnode->bitmapquals = bitmapquals;
1223 :
1224 : /* this sets bitmapselectivity as well as the regular cost fields: */
1225 1016 : cost_bitmap_or_node(pathnode, root);
1226 :
1227 1016 : return pathnode;
1228 : }
1229 :
1230 : /*
1231 : * create_tidscan_path
1232 : * Creates a path corresponding to a scan by TID, returning the pathnode.
1233 : */
1234 : TidPath *
1235 872 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1236 : Relids required_outer)
1237 : {
1238 872 : TidPath *pathnode = makeNode(TidPath);
1239 :
1240 872 : pathnode->path.pathtype = T_TidScan;
1241 872 : pathnode->path.parent = rel;
1242 872 : pathnode->path.pathtarget = rel->reltarget;
1243 872 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1244 : required_outer);
1245 872 : pathnode->path.parallel_aware = false;
1246 872 : pathnode->path.parallel_safe = rel->consider_parallel;
1247 872 : pathnode->path.parallel_workers = 0;
1248 872 : pathnode->path.pathkeys = NIL; /* always unordered */
1249 :
1250 872 : pathnode->tidquals = tidquals;
1251 :
1252 872 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1253 : pathnode->path.param_info);
1254 :
1255 872 : return pathnode;
1256 : }
1257 :
1258 : /*
1259 : * create_tidrangescan_path
1260 : * Creates a path corresponding to a scan by a range of TIDs, returning
1261 : * the pathnode.
1262 : */
1263 : TidRangePath *
1264 2052 : create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
1265 : List *tidrangequals, Relids required_outer,
1266 : int parallel_workers)
1267 : {
1268 2052 : TidRangePath *pathnode = makeNode(TidRangePath);
1269 :
1270 2052 : pathnode->path.pathtype = T_TidRangeScan;
1271 2052 : pathnode->path.parent = rel;
1272 2052 : pathnode->path.pathtarget = rel->reltarget;
1273 2052 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1274 : required_outer);
1275 2052 : pathnode->path.parallel_aware = (parallel_workers > 0);
1276 2052 : pathnode->path.parallel_safe = rel->consider_parallel;
1277 2052 : pathnode->path.parallel_workers = parallel_workers;
1278 2052 : pathnode->path.pathkeys = NIL; /* always unordered */
1279 :
1280 2052 : pathnode->tidrangequals = tidrangequals;
1281 :
1282 2052 : cost_tidrangescan(&pathnode->path, root, rel, tidrangequals,
1283 : pathnode->path.param_info);
1284 :
1285 2052 : return pathnode;
1286 : }
1287 :
1288 : /*
1289 : * create_append_path
1290 : * Creates a path corresponding to an Append plan, returning the
1291 : * pathnode.
1292 : *
1293 : * Note that we must handle subpaths = NIL, representing a dummy access path.
1294 : * Also, there are callers that pass root = NULL.
1295 : *
1296 : * 'rows', when passed as a non-negative number, will be used to overwrite the
1297 : * returned path's row estimate. Otherwise, the row estimate is calculated
1298 : * by totalling the row estimates from the 'subpaths' list.
1299 : */
1300 : AppendPath *
1301 91436 : create_append_path(PlannerInfo *root,
1302 : RelOptInfo *rel,
1303 : List *subpaths, List *partial_subpaths,
1304 : List *pathkeys, Relids required_outer,
1305 : int parallel_workers, bool parallel_aware,
1306 : double rows)
1307 : {
1308 91436 : AppendPath *pathnode = makeNode(AppendPath);
1309 : ListCell *l;
1310 :
1311 : Assert(!parallel_aware || parallel_workers > 0);
1312 :
1313 91436 : pathnode->path.pathtype = T_Append;
1314 91436 : pathnode->path.parent = rel;
1315 91436 : pathnode->path.pathtarget = rel->reltarget;
1316 :
1317 : /*
1318 : * If this is for a baserel (not a join or non-leaf partition), we prefer
1319 : * to apply get_baserel_parampathinfo to construct a full ParamPathInfo
1320 : * for the path. This supports building a Memoize path atop this path,
1321 : * and if this is a partitioned table the info may be useful for run-time
1322 : * pruning (cf make_partition_pruneinfo()).
1323 : *
1324 : * However, if we don't have "root" then that won't work and we fall back
1325 : * on the simpler get_appendrel_parampathinfo. There's no point in doing
1326 : * the more expensive thing for a dummy path, either.
1327 : */
1328 91436 : if (rel->reloptkind == RELOPT_BASEREL && root && subpaths != NIL)
1329 39898 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1330 : rel,
1331 : required_outer);
1332 : else
1333 51538 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1334 : required_outer);
1335 :
1336 91436 : pathnode->path.parallel_aware = parallel_aware;
1337 91436 : pathnode->path.parallel_safe = rel->consider_parallel;
1338 91436 : pathnode->path.parallel_workers = parallel_workers;
1339 91436 : pathnode->path.pathkeys = pathkeys;
1340 :
1341 : /*
1342 : * For parallel append, non-partial paths are sorted by descending total
1343 : * costs. That way, the total time to finish all non-partial paths is
1344 : * minimized. Also, the partial paths are sorted by descending startup
1345 : * costs. There may be some paths that require to do startup work by a
1346 : * single worker. In such case, it's better for workers to choose the
1347 : * expensive ones first, whereas the leader should choose the cheapest
1348 : * startup plan.
1349 : */
1350 91436 : if (pathnode->path.parallel_aware)
1351 : {
1352 : /*
1353 : * We mustn't fiddle with the order of subpaths when the Append has
1354 : * pathkeys. The order they're listed in is critical to keeping the
1355 : * pathkeys valid.
1356 : */
1357 : Assert(pathkeys == NIL);
1358 :
1359 32886 : list_sort(subpaths, append_total_cost_compare);
1360 32886 : list_sort(partial_subpaths, append_startup_cost_compare);
1361 : }
1362 91436 : pathnode->first_partial_path = list_length(subpaths);
1363 91436 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1364 :
1365 : /*
1366 : * Apply query-wide LIMIT if known and path is for sole base relation.
1367 : * (Handling this at this low level is a bit klugy.)
1368 : */
1369 91436 : if (root != NULL && bms_equal(rel->relids, root->all_query_rels))
1370 46054 : pathnode->limit_tuples = root->limit_tuples;
1371 : else
1372 45382 : pathnode->limit_tuples = -1.0;
1373 :
1374 310770 : foreach(l, pathnode->subpaths)
1375 : {
1376 219334 : Path *subpath = (Path *) lfirst(l);
1377 :
1378 398614 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1379 179280 : subpath->parallel_safe;
1380 :
1381 : /* All child paths must have same parameterization */
1382 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1383 : }
1384 :
1385 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1386 :
1387 : /*
1388 : * If there's exactly one child path then the output of the Append is
1389 : * necessarily ordered the same as the child's, so we can inherit the
1390 : * child's pathkeys if any, overriding whatever the caller might've said.
1391 : * Furthermore, if the child's parallel awareness matches the Append's,
1392 : * then the Append is a no-op and will be discarded later (in setrefs.c).
1393 : * Then we can inherit the child's size and cost too, effectively charging
1394 : * zero for the Append. Otherwise, we must do the normal costsize
1395 : * calculation.
1396 : */
1397 91436 : if (list_length(pathnode->subpaths) == 1)
1398 : {
1399 22326 : Path *child = (Path *) linitial(pathnode->subpaths);
1400 :
1401 22326 : if (child->parallel_aware == parallel_aware)
1402 : {
1403 21888 : pathnode->path.rows = child->rows;
1404 21888 : pathnode->path.startup_cost = child->startup_cost;
1405 21888 : pathnode->path.total_cost = child->total_cost;
1406 : }
1407 : else
1408 438 : cost_append(pathnode, root);
1409 : /* Must do this last, else cost_append complains */
1410 22326 : pathnode->path.pathkeys = child->pathkeys;
1411 : }
1412 : else
1413 69110 : cost_append(pathnode, root);
1414 :
1415 : /* If the caller provided a row estimate, override the computed value. */
1416 91436 : if (rows >= 0)
1417 576 : pathnode->path.rows = rows;
1418 :
1419 91436 : return pathnode;
1420 : }
1421 :
1422 : /*
1423 : * append_total_cost_compare
1424 : * list_sort comparator for sorting append child paths
1425 : * by total_cost descending
1426 : *
1427 : * For equal total costs, we fall back to comparing startup costs; if those
1428 : * are equal too, break ties using bms_compare on the paths' relids.
1429 : * (This is to avoid getting unpredictable results from list_sort.)
1430 : */
1431 : static int
1432 21780 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1433 : {
1434 21780 : Path *path1 = (Path *) lfirst(a);
1435 21780 : Path *path2 = (Path *) lfirst(b);
1436 : int cmp;
1437 :
1438 21780 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1439 21780 : if (cmp != 0)
1440 19412 : return -cmp;
1441 2368 : return bms_compare(path1->parent->relids, path2->parent->relids);
1442 : }
1443 :
1444 : /*
1445 : * append_startup_cost_compare
1446 : * list_sort comparator for sorting append child paths
1447 : * by startup_cost descending
1448 : *
1449 : * For equal startup costs, we fall back to comparing total costs; if those
1450 : * are equal too, break ties using bms_compare on the paths' relids.
1451 : * (This is to avoid getting unpredictable results from list_sort.)
1452 : */
1453 : static int
1454 44096 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1455 : {
1456 44096 : Path *path1 = (Path *) lfirst(a);
1457 44096 : Path *path2 = (Path *) lfirst(b);
1458 : int cmp;
1459 :
1460 44096 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1461 44096 : if (cmp != 0)
1462 21462 : return -cmp;
1463 22634 : return bms_compare(path1->parent->relids, path2->parent->relids);
1464 : }
1465 :
1466 : /*
1467 : * create_merge_append_path
1468 : * Creates a path corresponding to a MergeAppend plan, returning the
1469 : * pathnode.
1470 : */
1471 : MergeAppendPath *
1472 10036 : create_merge_append_path(PlannerInfo *root,
1473 : RelOptInfo *rel,
1474 : List *subpaths,
1475 : List *pathkeys,
1476 : Relids required_outer)
1477 : {
1478 10036 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1479 : int input_disabled_nodes;
1480 : Cost input_startup_cost;
1481 : Cost input_total_cost;
1482 : ListCell *l;
1483 :
1484 : /*
1485 : * We don't currently support parameterized MergeAppend paths, as
1486 : * explained in the comments for generate_orderedappend_paths.
1487 : */
1488 : Assert(bms_is_empty(rel->lateral_relids) && bms_is_empty(required_outer));
1489 :
1490 10036 : pathnode->path.pathtype = T_MergeAppend;
1491 10036 : pathnode->path.parent = rel;
1492 10036 : pathnode->path.pathtarget = rel->reltarget;
1493 10036 : pathnode->path.param_info = NULL;
1494 10036 : pathnode->path.parallel_aware = false;
1495 10036 : pathnode->path.parallel_safe = rel->consider_parallel;
1496 10036 : pathnode->path.parallel_workers = 0;
1497 10036 : pathnode->path.pathkeys = pathkeys;
1498 10036 : pathnode->subpaths = subpaths;
1499 :
1500 : /*
1501 : * Apply query-wide LIMIT if known and path is for sole base relation.
1502 : * (Handling this at this low level is a bit klugy.)
1503 : */
1504 10036 : if (bms_equal(rel->relids, root->all_query_rels))
1505 4638 : pathnode->limit_tuples = root->limit_tuples;
1506 : else
1507 5398 : pathnode->limit_tuples = -1.0;
1508 :
1509 : /*
1510 : * Add up the sizes and costs of the input paths.
1511 : */
1512 10036 : pathnode->path.rows = 0;
1513 10036 : input_disabled_nodes = 0;
1514 10036 : input_startup_cost = 0;
1515 10036 : input_total_cost = 0;
1516 36252 : foreach(l, subpaths)
1517 : {
1518 26216 : Path *subpath = (Path *) lfirst(l);
1519 : int presorted_keys;
1520 : Path sort_path; /* dummy for result of
1521 : * cost_sort/cost_incremental_sort */
1522 :
1523 : /* All child paths should be unparameterized */
1524 : Assert(bms_is_empty(PATH_REQ_OUTER(subpath)));
1525 :
1526 26216 : pathnode->path.rows += subpath->rows;
1527 49728 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1528 23512 : subpath->parallel_safe;
1529 :
1530 26216 : if (!pathkeys_count_contained_in(pathkeys, subpath->pathkeys,
1531 : &presorted_keys))
1532 : {
1533 : /*
1534 : * We'll need to insert a Sort node, so include costs for that. We
1535 : * choose to use incremental sort if it is enabled and there are
1536 : * presorted keys; otherwise we use full sort.
1537 : *
1538 : * We can use the parent's LIMIT if any, since we certainly won't
1539 : * pull more than that many tuples from any child.
1540 : */
1541 314 : if (enable_incremental_sort && presorted_keys > 0)
1542 : {
1543 18 : cost_incremental_sort(&sort_path,
1544 : root,
1545 : pathkeys,
1546 : presorted_keys,
1547 : subpath->disabled_nodes,
1548 : subpath->startup_cost,
1549 : subpath->total_cost,
1550 : subpath->rows,
1551 18 : subpath->pathtarget->width,
1552 : 0.0,
1553 : work_mem,
1554 : pathnode->limit_tuples);
1555 : }
1556 : else
1557 : {
1558 296 : cost_sort(&sort_path,
1559 : root,
1560 : pathkeys,
1561 : subpath->disabled_nodes,
1562 : subpath->total_cost,
1563 : subpath->rows,
1564 296 : subpath->pathtarget->width,
1565 : 0.0,
1566 : work_mem,
1567 : pathnode->limit_tuples);
1568 : }
1569 :
1570 314 : subpath = &sort_path;
1571 : }
1572 :
1573 26216 : input_disabled_nodes += subpath->disabled_nodes;
1574 26216 : input_startup_cost += subpath->startup_cost;
1575 26216 : input_total_cost += subpath->total_cost;
1576 : }
1577 :
1578 : /*
1579 : * Now we can compute total costs of the MergeAppend. If there's exactly
1580 : * one child path and its parallel awareness matches that of the
1581 : * MergeAppend, then the MergeAppend is a no-op and will be discarded
1582 : * later (in setrefs.c); otherwise we do the normal cost calculation.
1583 : */
1584 10036 : if (list_length(subpaths) == 1 &&
1585 128 : ((Path *) linitial(subpaths))->parallel_aware ==
1586 128 : pathnode->path.parallel_aware)
1587 : {
1588 128 : pathnode->path.disabled_nodes = input_disabled_nodes;
1589 128 : pathnode->path.startup_cost = input_startup_cost;
1590 128 : pathnode->path.total_cost = input_total_cost;
1591 : }
1592 : else
1593 9908 : cost_merge_append(&pathnode->path, root,
1594 : pathkeys, list_length(subpaths),
1595 : input_disabled_nodes,
1596 : input_startup_cost, input_total_cost,
1597 : pathnode->path.rows);
1598 :
1599 10036 : return pathnode;
1600 : }
1601 :
1602 : /*
1603 : * create_group_result_path
1604 : * Creates a path representing a Result-and-nothing-else plan.
1605 : *
1606 : * This is only used for degenerate grouping cases, in which we know we
1607 : * need to produce one result row, possibly filtered by a HAVING qual.
1608 : */
1609 : GroupResultPath *
1610 198534 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1611 : PathTarget *target, List *havingqual)
1612 : {
1613 198534 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1614 :
1615 198534 : pathnode->path.pathtype = T_Result;
1616 198534 : pathnode->path.parent = rel;
1617 198534 : pathnode->path.pathtarget = target;
1618 198534 : pathnode->path.param_info = NULL; /* there are no other rels... */
1619 198534 : pathnode->path.parallel_aware = false;
1620 198534 : pathnode->path.parallel_safe = rel->consider_parallel;
1621 198534 : pathnode->path.parallel_workers = 0;
1622 198534 : pathnode->path.pathkeys = NIL;
1623 198534 : pathnode->quals = havingqual;
1624 :
1625 : /*
1626 : * We can't quite use cost_resultscan() because the quals we want to
1627 : * account for are not baserestrict quals of the rel. Might as well just
1628 : * hack it here.
1629 : */
1630 198534 : pathnode->path.rows = 1;
1631 198534 : pathnode->path.startup_cost = target->cost.startup;
1632 198534 : pathnode->path.total_cost = target->cost.startup +
1633 198534 : cpu_tuple_cost + target->cost.per_tuple;
1634 :
1635 : /*
1636 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1637 : * rowcount estimate should be 1 no matter what the qual is.
1638 : */
1639 198534 : if (havingqual)
1640 : {
1641 : QualCost qual_cost;
1642 :
1643 628 : cost_qual_eval(&qual_cost, havingqual, root);
1644 : /* havingqual is evaluated once at startup */
1645 628 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1646 628 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1647 : }
1648 :
1649 198534 : return pathnode;
1650 : }
1651 :
1652 : /*
1653 : * create_material_path
1654 : * Creates a path corresponding to a Material plan, returning the
1655 : * pathnode.
1656 : */
1657 : MaterialPath *
1658 673348 : create_material_path(RelOptInfo *rel, Path *subpath)
1659 : {
1660 673348 : MaterialPath *pathnode = makeNode(MaterialPath);
1661 :
1662 : Assert(subpath->parent == rel);
1663 :
1664 673348 : pathnode->path.pathtype = T_Material;
1665 673348 : pathnode->path.parent = rel;
1666 673348 : pathnode->path.pathtarget = rel->reltarget;
1667 673348 : pathnode->path.param_info = subpath->param_info;
1668 673348 : pathnode->path.parallel_aware = false;
1669 1290006 : pathnode->path.parallel_safe = rel->consider_parallel &&
1670 616658 : subpath->parallel_safe;
1671 673348 : pathnode->path.parallel_workers = subpath->parallel_workers;
1672 673348 : pathnode->path.pathkeys = subpath->pathkeys;
1673 :
1674 673348 : pathnode->subpath = subpath;
1675 :
1676 673348 : cost_material(&pathnode->path,
1677 : subpath->disabled_nodes,
1678 : subpath->startup_cost,
1679 : subpath->total_cost,
1680 : subpath->rows,
1681 673348 : subpath->pathtarget->width);
1682 :
1683 673348 : return pathnode;
1684 : }
1685 :
1686 : /*
1687 : * create_memoize_path
1688 : * Creates a path corresponding to a Memoize plan, returning the pathnode.
1689 : */
1690 : MemoizePath *
1691 315292 : create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1692 : List *param_exprs, List *hash_operators,
1693 : bool singlerow, bool binary_mode, Cardinality est_calls)
1694 : {
1695 315292 : MemoizePath *pathnode = makeNode(MemoizePath);
1696 :
1697 : Assert(subpath->parent == rel);
1698 :
1699 315292 : pathnode->path.pathtype = T_Memoize;
1700 315292 : pathnode->path.parent = rel;
1701 315292 : pathnode->path.pathtarget = rel->reltarget;
1702 315292 : pathnode->path.param_info = subpath->param_info;
1703 315292 : pathnode->path.parallel_aware = false;
1704 616326 : pathnode->path.parallel_safe = rel->consider_parallel &&
1705 301034 : subpath->parallel_safe;
1706 315292 : pathnode->path.parallel_workers = subpath->parallel_workers;
1707 315292 : pathnode->path.pathkeys = subpath->pathkeys;
1708 :
1709 315292 : pathnode->subpath = subpath;
1710 315292 : pathnode->hash_operators = hash_operators;
1711 315292 : pathnode->param_exprs = param_exprs;
1712 315292 : pathnode->singlerow = singlerow;
1713 315292 : pathnode->binary_mode = binary_mode;
1714 :
1715 : /*
1716 : * For now we set est_entries to 0. cost_memoize_rescan() does all the
1717 : * hard work to determine how many cache entries there are likely to be,
1718 : * so it seems best to leave it up to that function to fill this field in.
1719 : * If left at 0, the executor will make a guess at a good value.
1720 : */
1721 315292 : pathnode->est_entries = 0;
1722 :
1723 315292 : pathnode->est_calls = clamp_row_est(est_calls);
1724 :
1725 : /* These will also be set later in cost_memoize_rescan() */
1726 315292 : pathnode->est_unique_keys = 0.0;
1727 315292 : pathnode->est_hit_ratio = 0.0;
1728 :
1729 : /* we should not generate this path type when enable_memoize=false */
1730 : Assert(enable_memoize);
1731 315292 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
1732 :
1733 : /*
1734 : * Add a small additional charge for caching the first entry. All the
1735 : * harder calculations for rescans are performed in cost_memoize_rescan().
1736 : */
1737 315292 : pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
1738 315292 : pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
1739 315292 : pathnode->path.rows = subpath->rows;
1740 :
1741 315292 : return pathnode;
1742 : }
1743 :
1744 : /*
1745 : * create_gather_merge_path
1746 : *
1747 : * Creates a path corresponding to a gather merge scan, returning
1748 : * the pathnode.
1749 : */
1750 : GatherMergePath *
1751 17504 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1752 : PathTarget *target, List *pathkeys,
1753 : Relids required_outer, double *rows)
1754 : {
1755 17504 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1756 17504 : int input_disabled_nodes = 0;
1757 17504 : Cost input_startup_cost = 0;
1758 17504 : Cost input_total_cost = 0;
1759 :
1760 : Assert(subpath->parallel_safe);
1761 : Assert(pathkeys);
1762 :
1763 : /*
1764 : * The subpath should guarantee that it is adequately ordered either by
1765 : * adding an explicit sort node or by using presorted input. We cannot
1766 : * add an explicit Sort node for the subpath in createplan.c on additional
1767 : * pathkeys, because we can't guarantee the sort would be safe. For
1768 : * example, expressions may be volatile or otherwise parallel unsafe.
1769 : */
1770 17504 : if (!pathkeys_contained_in(pathkeys, subpath->pathkeys))
1771 0 : elog(ERROR, "gather merge input not sufficiently sorted");
1772 :
1773 17504 : pathnode->path.pathtype = T_GatherMerge;
1774 17504 : pathnode->path.parent = rel;
1775 17504 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1776 : required_outer);
1777 17504 : pathnode->path.parallel_aware = false;
1778 :
1779 17504 : pathnode->subpath = subpath;
1780 17504 : pathnode->num_workers = subpath->parallel_workers;
1781 17504 : pathnode->path.pathkeys = pathkeys;
1782 17504 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1783 :
1784 17504 : input_disabled_nodes += subpath->disabled_nodes;
1785 17504 : input_startup_cost += subpath->startup_cost;
1786 17504 : input_total_cost += subpath->total_cost;
1787 :
1788 17504 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1789 : input_disabled_nodes, input_startup_cost,
1790 : input_total_cost, rows);
1791 :
1792 17504 : return pathnode;
1793 : }
1794 :
1795 : /*
1796 : * create_gather_path
1797 : * Creates a path corresponding to a gather scan, returning the
1798 : * pathnode.
1799 : *
1800 : * 'rows' may optionally be set to override row estimates from other sources.
1801 : */
1802 : GatherPath *
1803 24836 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1804 : PathTarget *target, Relids required_outer, double *rows)
1805 : {
1806 24836 : GatherPath *pathnode = makeNode(GatherPath);
1807 :
1808 : Assert(subpath->parallel_safe);
1809 :
1810 24836 : pathnode->path.pathtype = T_Gather;
1811 24836 : pathnode->path.parent = rel;
1812 24836 : pathnode->path.pathtarget = target;
1813 24836 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1814 : required_outer);
1815 24836 : pathnode->path.parallel_aware = false;
1816 24836 : pathnode->path.parallel_safe = false;
1817 24836 : pathnode->path.parallel_workers = 0;
1818 24836 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1819 :
1820 24836 : pathnode->subpath = subpath;
1821 24836 : pathnode->num_workers = subpath->parallel_workers;
1822 24836 : pathnode->single_copy = false;
1823 :
1824 24836 : if (pathnode->num_workers == 0)
1825 : {
1826 0 : pathnode->path.pathkeys = subpath->pathkeys;
1827 0 : pathnode->num_workers = 1;
1828 0 : pathnode->single_copy = true;
1829 : }
1830 :
1831 24836 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
1832 :
1833 24836 : return pathnode;
1834 : }
1835 :
1836 : /*
1837 : * create_subqueryscan_path
1838 : * Creates a path corresponding to a scan of a subquery,
1839 : * returning the pathnode.
1840 : *
1841 : * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
1842 : * be trivial, ie just a fetch of all the subquery output columns in order.
1843 : * While we could determine that here, the caller can usually do it more
1844 : * efficiently (or at least amortize it over multiple calls).
1845 : */
1846 : SubqueryScanPath *
1847 55814 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1848 : bool trivial_pathtarget,
1849 : List *pathkeys, Relids required_outer)
1850 : {
1851 55814 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
1852 :
1853 55814 : pathnode->path.pathtype = T_SubqueryScan;
1854 55814 : pathnode->path.parent = rel;
1855 55814 : pathnode->path.pathtarget = rel->reltarget;
1856 55814 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1857 : required_outer);
1858 55814 : pathnode->path.parallel_aware = false;
1859 96276 : pathnode->path.parallel_safe = rel->consider_parallel &&
1860 40462 : subpath->parallel_safe;
1861 55814 : pathnode->path.parallel_workers = subpath->parallel_workers;
1862 55814 : pathnode->path.pathkeys = pathkeys;
1863 55814 : pathnode->subpath = subpath;
1864 :
1865 55814 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
1866 : trivial_pathtarget);
1867 :
1868 55814 : return pathnode;
1869 : }
1870 :
1871 : /*
1872 : * create_functionscan_path
1873 : * Creates a path corresponding to a sequential scan of a function,
1874 : * returning the pathnode.
1875 : */
1876 : Path *
1877 51838 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
1878 : List *pathkeys, Relids required_outer)
1879 : {
1880 51838 : Path *pathnode = makeNode(Path);
1881 :
1882 51838 : pathnode->pathtype = T_FunctionScan;
1883 51838 : pathnode->parent = rel;
1884 51838 : pathnode->pathtarget = rel->reltarget;
1885 51838 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1886 : required_outer);
1887 51838 : pathnode->parallel_aware = false;
1888 51838 : pathnode->parallel_safe = rel->consider_parallel;
1889 51838 : pathnode->parallel_workers = 0;
1890 51838 : pathnode->pathkeys = pathkeys;
1891 :
1892 51838 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
1893 :
1894 51838 : return pathnode;
1895 : }
1896 :
1897 : /*
1898 : * create_tablefuncscan_path
1899 : * Creates a path corresponding to a sequential scan of a table function,
1900 : * returning the pathnode.
1901 : */
1902 : Path *
1903 626 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
1904 : Relids required_outer)
1905 : {
1906 626 : Path *pathnode = makeNode(Path);
1907 :
1908 626 : pathnode->pathtype = T_TableFuncScan;
1909 626 : pathnode->parent = rel;
1910 626 : pathnode->pathtarget = rel->reltarget;
1911 626 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1912 : required_outer);
1913 626 : pathnode->parallel_aware = false;
1914 626 : pathnode->parallel_safe = rel->consider_parallel;
1915 626 : pathnode->parallel_workers = 0;
1916 626 : pathnode->pathkeys = NIL; /* result is always unordered */
1917 :
1918 626 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
1919 :
1920 626 : return pathnode;
1921 : }
1922 :
1923 : /*
1924 : * create_valuesscan_path
1925 : * Creates a path corresponding to a scan of a VALUES list,
1926 : * returning the pathnode.
1927 : */
1928 : Path *
1929 8282 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
1930 : Relids required_outer)
1931 : {
1932 8282 : Path *pathnode = makeNode(Path);
1933 :
1934 8282 : pathnode->pathtype = T_ValuesScan;
1935 8282 : pathnode->parent = rel;
1936 8282 : pathnode->pathtarget = rel->reltarget;
1937 8282 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1938 : required_outer);
1939 8282 : pathnode->parallel_aware = false;
1940 8282 : pathnode->parallel_safe = rel->consider_parallel;
1941 8282 : pathnode->parallel_workers = 0;
1942 8282 : pathnode->pathkeys = NIL; /* result is always unordered */
1943 :
1944 8282 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
1945 :
1946 8282 : return pathnode;
1947 : }
1948 :
1949 : /*
1950 : * create_ctescan_path
1951 : * Creates a path corresponding to a scan of a non-self-reference CTE,
1952 : * returning the pathnode.
1953 : */
1954 : Path *
1955 4242 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel,
1956 : List *pathkeys, Relids required_outer)
1957 : {
1958 4242 : Path *pathnode = makeNode(Path);
1959 :
1960 4242 : pathnode->pathtype = T_CteScan;
1961 4242 : pathnode->parent = rel;
1962 4242 : pathnode->pathtarget = rel->reltarget;
1963 4242 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1964 : required_outer);
1965 4242 : pathnode->parallel_aware = false;
1966 4242 : pathnode->parallel_safe = rel->consider_parallel;
1967 4242 : pathnode->parallel_workers = 0;
1968 4242 : pathnode->pathkeys = pathkeys;
1969 :
1970 4242 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
1971 :
1972 4242 : return pathnode;
1973 : }
1974 :
1975 : /*
1976 : * create_namedtuplestorescan_path
1977 : * Creates a path corresponding to a scan of a named tuplestore, returning
1978 : * the pathnode.
1979 : */
1980 : Path *
1981 474 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
1982 : Relids required_outer)
1983 : {
1984 474 : Path *pathnode = makeNode(Path);
1985 :
1986 474 : pathnode->pathtype = T_NamedTuplestoreScan;
1987 474 : pathnode->parent = rel;
1988 474 : pathnode->pathtarget = rel->reltarget;
1989 474 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1990 : required_outer);
1991 474 : pathnode->parallel_aware = false;
1992 474 : pathnode->parallel_safe = rel->consider_parallel;
1993 474 : pathnode->parallel_workers = 0;
1994 474 : pathnode->pathkeys = NIL; /* result is always unordered */
1995 :
1996 474 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
1997 :
1998 474 : return pathnode;
1999 : }
2000 :
2001 : /*
2002 : * create_resultscan_path
2003 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2004 : * returning the pathnode.
2005 : */
2006 : Path *
2007 4268 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2008 : Relids required_outer)
2009 : {
2010 4268 : Path *pathnode = makeNode(Path);
2011 :
2012 4268 : pathnode->pathtype = T_Result;
2013 4268 : pathnode->parent = rel;
2014 4268 : pathnode->pathtarget = rel->reltarget;
2015 4268 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2016 : required_outer);
2017 4268 : pathnode->parallel_aware = false;
2018 4268 : pathnode->parallel_safe = rel->consider_parallel;
2019 4268 : pathnode->parallel_workers = 0;
2020 4268 : pathnode->pathkeys = NIL; /* result is always unordered */
2021 :
2022 4268 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2023 :
2024 4268 : return pathnode;
2025 : }
2026 :
2027 : /*
2028 : * create_worktablescan_path
2029 : * Creates a path corresponding to a scan of a self-reference CTE,
2030 : * returning the pathnode.
2031 : */
2032 : Path *
2033 934 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2034 : Relids required_outer)
2035 : {
2036 934 : Path *pathnode = makeNode(Path);
2037 :
2038 934 : pathnode->pathtype = T_WorkTableScan;
2039 934 : pathnode->parent = rel;
2040 934 : pathnode->pathtarget = rel->reltarget;
2041 934 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2042 : required_outer);
2043 934 : pathnode->parallel_aware = false;
2044 934 : pathnode->parallel_safe = rel->consider_parallel;
2045 934 : pathnode->parallel_workers = 0;
2046 934 : pathnode->pathkeys = NIL; /* result is always unordered */
2047 :
2048 : /* Cost is the same as for a regular CTE scan */
2049 934 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2050 :
2051 934 : return pathnode;
2052 : }
2053 :
2054 : /*
2055 : * create_foreignscan_path
2056 : * Creates a path corresponding to a scan of a foreign base table,
2057 : * returning the pathnode.
2058 : *
2059 : * This function is never called from core Postgres; rather, it's expected
2060 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2061 : * We make the FDW supply all fields of the path, since we do not have any way
2062 : * to calculate them in core. However, there is a usually-sane default for
2063 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2064 : */
2065 : ForeignPath *
2066 3722 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2067 : PathTarget *target,
2068 : double rows, int disabled_nodes,
2069 : Cost startup_cost, Cost total_cost,
2070 : List *pathkeys,
2071 : Relids required_outer,
2072 : Path *fdw_outerpath,
2073 : List *fdw_restrictinfo,
2074 : List *fdw_private)
2075 : {
2076 3722 : ForeignPath *pathnode = makeNode(ForeignPath);
2077 :
2078 : /* Historically some FDWs were confused about when to use this */
2079 : Assert(IS_SIMPLE_REL(rel));
2080 :
2081 3722 : pathnode->path.pathtype = T_ForeignScan;
2082 3722 : pathnode->path.parent = rel;
2083 3722 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2084 3722 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2085 : required_outer);
2086 3722 : pathnode->path.parallel_aware = false;
2087 3722 : pathnode->path.parallel_safe = rel->consider_parallel;
2088 3722 : pathnode->path.parallel_workers = 0;
2089 3722 : pathnode->path.rows = rows;
2090 3722 : pathnode->path.disabled_nodes = disabled_nodes;
2091 3722 : pathnode->path.startup_cost = startup_cost;
2092 3722 : pathnode->path.total_cost = total_cost;
2093 3722 : pathnode->path.pathkeys = pathkeys;
2094 :
2095 3722 : pathnode->fdw_outerpath = fdw_outerpath;
2096 3722 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2097 3722 : pathnode->fdw_private = fdw_private;
2098 :
2099 3722 : return pathnode;
2100 : }
2101 :
2102 : /*
2103 : * create_foreign_join_path
2104 : * Creates a path corresponding to a scan of a foreign join,
2105 : * returning the pathnode.
2106 : *
2107 : * This function is never called from core Postgres; rather, it's expected
2108 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2109 : * We make the FDW supply all fields of the path, since we do not have any way
2110 : * to calculate them in core. However, there is a usually-sane default for
2111 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2112 : */
2113 : ForeignPath *
2114 1212 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2115 : PathTarget *target,
2116 : double rows, int disabled_nodes,
2117 : Cost startup_cost, Cost total_cost,
2118 : List *pathkeys,
2119 : Relids required_outer,
2120 : Path *fdw_outerpath,
2121 : List *fdw_restrictinfo,
2122 : List *fdw_private)
2123 : {
2124 1212 : ForeignPath *pathnode = makeNode(ForeignPath);
2125 :
2126 : /*
2127 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2128 : * but the API of this function doesn't support it, and existing
2129 : * extensions aren't yet trying to build such paths anyway. For the
2130 : * moment just throw an error if someone tries it; eventually we should
2131 : * revisit this.
2132 : */
2133 1212 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2134 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2135 :
2136 1212 : pathnode->path.pathtype = T_ForeignScan;
2137 1212 : pathnode->path.parent = rel;
2138 1212 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2139 1212 : pathnode->path.param_info = NULL; /* XXX see above */
2140 1212 : pathnode->path.parallel_aware = false;
2141 1212 : pathnode->path.parallel_safe = rel->consider_parallel;
2142 1212 : pathnode->path.parallel_workers = 0;
2143 1212 : pathnode->path.rows = rows;
2144 1212 : pathnode->path.disabled_nodes = disabled_nodes;
2145 1212 : pathnode->path.startup_cost = startup_cost;
2146 1212 : pathnode->path.total_cost = total_cost;
2147 1212 : pathnode->path.pathkeys = pathkeys;
2148 :
2149 1212 : pathnode->fdw_outerpath = fdw_outerpath;
2150 1212 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2151 1212 : pathnode->fdw_private = fdw_private;
2152 :
2153 1212 : return pathnode;
2154 : }
2155 :
2156 : /*
2157 : * create_foreign_upper_path
2158 : * Creates a path corresponding to an upper relation that's computed
2159 : * directly by an FDW, returning the pathnode.
2160 : *
2161 : * This function is never called from core Postgres; rather, it's expected to
2162 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2163 : * We make the FDW supply all fields of the path, since we do not have any way
2164 : * to calculate them in core. However, there is a usually-sane default for
2165 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2166 : */
2167 : ForeignPath *
2168 588 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2169 : PathTarget *target,
2170 : double rows, int disabled_nodes,
2171 : Cost startup_cost, Cost total_cost,
2172 : List *pathkeys,
2173 : Path *fdw_outerpath,
2174 : List *fdw_restrictinfo,
2175 : List *fdw_private)
2176 : {
2177 588 : ForeignPath *pathnode = makeNode(ForeignPath);
2178 :
2179 : /*
2180 : * Upper relations should never have any lateral references, since joining
2181 : * is complete.
2182 : */
2183 : Assert(bms_is_empty(rel->lateral_relids));
2184 :
2185 588 : pathnode->path.pathtype = T_ForeignScan;
2186 588 : pathnode->path.parent = rel;
2187 588 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2188 588 : pathnode->path.param_info = NULL;
2189 588 : pathnode->path.parallel_aware = false;
2190 588 : pathnode->path.parallel_safe = rel->consider_parallel;
2191 588 : pathnode->path.parallel_workers = 0;
2192 588 : pathnode->path.rows = rows;
2193 588 : pathnode->path.disabled_nodes = disabled_nodes;
2194 588 : pathnode->path.startup_cost = startup_cost;
2195 588 : pathnode->path.total_cost = total_cost;
2196 588 : pathnode->path.pathkeys = pathkeys;
2197 :
2198 588 : pathnode->fdw_outerpath = fdw_outerpath;
2199 588 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2200 588 : pathnode->fdw_private = fdw_private;
2201 :
2202 588 : return pathnode;
2203 : }
2204 :
2205 : /*
2206 : * calc_nestloop_required_outer
2207 : * Compute the required_outer set for a nestloop join path
2208 : *
2209 : * Note: when considering a child join, the inputs nonetheless use top-level
2210 : * parent relids
2211 : *
2212 : * Note: result must not share storage with either input
2213 : */
2214 : Relids
2215 3369108 : calc_nestloop_required_outer(Relids outerrelids,
2216 : Relids outer_paramrels,
2217 : Relids innerrelids,
2218 : Relids inner_paramrels)
2219 : {
2220 : Relids required_outer;
2221 :
2222 : /* inner_path can require rels from outer path, but not vice versa */
2223 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2224 : /* easy case if inner path is not parameterized */
2225 3369108 : if (!inner_paramrels)
2226 2336188 : return bms_copy(outer_paramrels);
2227 : /* else, form the union ... */
2228 1032920 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2229 : /* ... and remove any mention of now-satisfied outer rels */
2230 1032920 : required_outer = bms_del_members(required_outer,
2231 : outerrelids);
2232 1032920 : return required_outer;
2233 : }
2234 :
2235 : /*
2236 : * calc_non_nestloop_required_outer
2237 : * Compute the required_outer set for a merge or hash join path
2238 : *
2239 : * Note: result must not share storage with either input
2240 : */
2241 : Relids
2242 2267806 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2243 : {
2244 2267806 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2245 2267806 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2246 : Relids innerrelids PG_USED_FOR_ASSERTS_ONLY;
2247 : Relids outerrelids PG_USED_FOR_ASSERTS_ONLY;
2248 : Relids required_outer;
2249 :
2250 : /*
2251 : * Any parameterization of the input paths refers to topmost parents of
2252 : * the relevant relations, because reparameterize_path_by_child() hasn't
2253 : * been called yet. So we must consider topmost parents of the relations
2254 : * being joined, too, while checking for disallowed parameterization
2255 : * cases.
2256 : */
2257 2267806 : if (inner_path->parent->top_parent_relids)
2258 159758 : innerrelids = inner_path->parent->top_parent_relids;
2259 : else
2260 2108048 : innerrelids = inner_path->parent->relids;
2261 :
2262 2267806 : if (outer_path->parent->top_parent_relids)
2263 159758 : outerrelids = outer_path->parent->top_parent_relids;
2264 : else
2265 2108048 : outerrelids = outer_path->parent->relids;
2266 :
2267 : /* neither path can require rels from the other */
2268 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2269 : Assert(!bms_overlap(inner_paramrels, outerrelids));
2270 : /* form the union ... */
2271 2267806 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2272 : /* we do not need an explicit test for empty; bms_union gets it right */
2273 2267806 : return required_outer;
2274 : }
2275 :
2276 : /*
2277 : * create_nestloop_path
2278 : * Creates a pathnode corresponding to a nestloop join between two
2279 : * relations.
2280 : *
2281 : * 'joinrel' is the join relation.
2282 : * 'jointype' is the type of join required
2283 : * 'workspace' is the result from initial_cost_nestloop
2284 : * 'extra' contains various information about the join
2285 : * 'outer_path' is the outer path
2286 : * 'inner_path' is the inner path
2287 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2288 : * 'pathkeys' are the path keys of the new join path
2289 : * 'required_outer' is the set of required outer rels
2290 : *
2291 : * Returns the resulting path node.
2292 : */
2293 : NestPath *
2294 1459232 : create_nestloop_path(PlannerInfo *root,
2295 : RelOptInfo *joinrel,
2296 : JoinType jointype,
2297 : JoinCostWorkspace *workspace,
2298 : JoinPathExtraData *extra,
2299 : Path *outer_path,
2300 : Path *inner_path,
2301 : List *restrict_clauses,
2302 : List *pathkeys,
2303 : Relids required_outer)
2304 : {
2305 1459232 : NestPath *pathnode = makeNode(NestPath);
2306 1459232 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2307 : Relids outerrelids;
2308 :
2309 : /*
2310 : * Paths are parameterized by top-level parents, so run parameterization
2311 : * tests on the parent relids.
2312 : */
2313 1459232 : if (outer_path->parent->top_parent_relids)
2314 76102 : outerrelids = outer_path->parent->top_parent_relids;
2315 : else
2316 1383130 : outerrelids = outer_path->parent->relids;
2317 :
2318 : /*
2319 : * If the inner path is parameterized by the outer, we must drop any
2320 : * restrict_clauses that are due to be moved into the inner path. We have
2321 : * to do this now, rather than postpone the work till createplan time,
2322 : * because the restrict_clauses list can affect the size and cost
2323 : * estimates for this path. We detect such clauses by checking for serial
2324 : * number match to clauses already enforced in the inner path.
2325 : */
2326 1459232 : if (bms_overlap(inner_req_outer, outerrelids))
2327 : {
2328 395504 : Bitmapset *enforced_serials = get_param_path_clause_serials(inner_path);
2329 395504 : List *jclauses = NIL;
2330 : ListCell *lc;
2331 :
2332 877286 : foreach(lc, restrict_clauses)
2333 : {
2334 481782 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2335 :
2336 481782 : if (!bms_is_member(rinfo->rinfo_serial, enforced_serials))
2337 60006 : jclauses = lappend(jclauses, rinfo);
2338 : }
2339 395504 : restrict_clauses = jclauses;
2340 : }
2341 :
2342 1459232 : pathnode->jpath.path.pathtype = T_NestLoop;
2343 1459232 : pathnode->jpath.path.parent = joinrel;
2344 1459232 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2345 1459232 : pathnode->jpath.path.param_info =
2346 1459232 : get_joinrel_parampathinfo(root,
2347 : joinrel,
2348 : outer_path,
2349 : inner_path,
2350 : extra->sjinfo,
2351 : required_outer,
2352 : &restrict_clauses);
2353 1459232 : pathnode->jpath.path.parallel_aware = false;
2354 4249868 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2355 1459232 : outer_path->parallel_safe && inner_path->parallel_safe;
2356 : /* This is a foolish way to estimate parallel_workers, but for now... */
2357 1459232 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2358 1459232 : pathnode->jpath.path.pathkeys = pathkeys;
2359 1459232 : pathnode->jpath.jointype = jointype;
2360 1459232 : pathnode->jpath.inner_unique = extra->inner_unique;
2361 1459232 : pathnode->jpath.outerjoinpath = outer_path;
2362 1459232 : pathnode->jpath.innerjoinpath = inner_path;
2363 1459232 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2364 :
2365 1459232 : final_cost_nestloop(root, pathnode, workspace, extra);
2366 :
2367 1459232 : return pathnode;
2368 : }
2369 :
2370 : /*
2371 : * create_mergejoin_path
2372 : * Creates a pathnode corresponding to a mergejoin join between
2373 : * two relations
2374 : *
2375 : * 'joinrel' is the join relation
2376 : * 'jointype' is the type of join required
2377 : * 'workspace' is the result from initial_cost_mergejoin
2378 : * 'extra' contains various information about the join
2379 : * 'outer_path' is the outer path
2380 : * 'inner_path' is the inner path
2381 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2382 : * 'pathkeys' are the path keys of the new join path
2383 : * 'required_outer' is the set of required outer rels
2384 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2385 : * (this should be a subset of the restrict_clauses list)
2386 : * 'outersortkeys' are the sort varkeys for the outer relation
2387 : * 'innersortkeys' are the sort varkeys for the inner relation
2388 : * 'outer_presorted_keys' is the number of presorted keys of the outer path
2389 : */
2390 : MergePath *
2391 458072 : create_mergejoin_path(PlannerInfo *root,
2392 : RelOptInfo *joinrel,
2393 : JoinType jointype,
2394 : JoinCostWorkspace *workspace,
2395 : JoinPathExtraData *extra,
2396 : Path *outer_path,
2397 : Path *inner_path,
2398 : List *restrict_clauses,
2399 : List *pathkeys,
2400 : Relids required_outer,
2401 : List *mergeclauses,
2402 : List *outersortkeys,
2403 : List *innersortkeys,
2404 : int outer_presorted_keys)
2405 : {
2406 458072 : MergePath *pathnode = makeNode(MergePath);
2407 :
2408 458072 : pathnode->jpath.path.pathtype = T_MergeJoin;
2409 458072 : pathnode->jpath.path.parent = joinrel;
2410 458072 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2411 458072 : pathnode->jpath.path.param_info =
2412 458072 : get_joinrel_parampathinfo(root,
2413 : joinrel,
2414 : outer_path,
2415 : inner_path,
2416 : extra->sjinfo,
2417 : required_outer,
2418 : &restrict_clauses);
2419 458072 : pathnode->jpath.path.parallel_aware = false;
2420 1336828 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2421 458072 : outer_path->parallel_safe && inner_path->parallel_safe;
2422 : /* This is a foolish way to estimate parallel_workers, but for now... */
2423 458072 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2424 458072 : pathnode->jpath.path.pathkeys = pathkeys;
2425 458072 : pathnode->jpath.jointype = jointype;
2426 458072 : pathnode->jpath.inner_unique = extra->inner_unique;
2427 458072 : pathnode->jpath.outerjoinpath = outer_path;
2428 458072 : pathnode->jpath.innerjoinpath = inner_path;
2429 458072 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2430 458072 : pathnode->path_mergeclauses = mergeclauses;
2431 458072 : pathnode->outersortkeys = outersortkeys;
2432 458072 : pathnode->innersortkeys = innersortkeys;
2433 458072 : pathnode->outer_presorted_keys = outer_presorted_keys;
2434 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2435 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2436 :
2437 458072 : final_cost_mergejoin(root, pathnode, workspace, extra);
2438 :
2439 458072 : return pathnode;
2440 : }
2441 :
2442 : /*
2443 : * create_hashjoin_path
2444 : * Creates a pathnode corresponding to a hash join between two relations.
2445 : *
2446 : * 'joinrel' is the join relation
2447 : * 'jointype' is the type of join required
2448 : * 'workspace' is the result from initial_cost_hashjoin
2449 : * 'extra' contains various information about the join
2450 : * 'outer_path' is the cheapest outer path
2451 : * 'inner_path' is the cheapest inner path
2452 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2453 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2454 : * 'required_outer' is the set of required outer rels
2455 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2456 : * (this should be a subset of the restrict_clauses list)
2457 : */
2458 : HashPath *
2459 453302 : create_hashjoin_path(PlannerInfo *root,
2460 : RelOptInfo *joinrel,
2461 : JoinType jointype,
2462 : JoinCostWorkspace *workspace,
2463 : JoinPathExtraData *extra,
2464 : Path *outer_path,
2465 : Path *inner_path,
2466 : bool parallel_hash,
2467 : List *restrict_clauses,
2468 : Relids required_outer,
2469 : List *hashclauses)
2470 : {
2471 453302 : HashPath *pathnode = makeNode(HashPath);
2472 :
2473 453302 : pathnode->jpath.path.pathtype = T_HashJoin;
2474 453302 : pathnode->jpath.path.parent = joinrel;
2475 453302 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2476 453302 : pathnode->jpath.path.param_info =
2477 453302 : get_joinrel_parampathinfo(root,
2478 : joinrel,
2479 : outer_path,
2480 : inner_path,
2481 : extra->sjinfo,
2482 : required_outer,
2483 : &restrict_clauses);
2484 453302 : pathnode->jpath.path.parallel_aware =
2485 453302 : joinrel->consider_parallel && parallel_hash;
2486 1323354 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2487 453302 : outer_path->parallel_safe && inner_path->parallel_safe;
2488 : /* This is a foolish way to estimate parallel_workers, but for now... */
2489 453302 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2490 :
2491 : /*
2492 : * A hashjoin never has pathkeys, since its output ordering is
2493 : * unpredictable due to possible batching. XXX If the inner relation is
2494 : * small enough, we could instruct the executor that it must not batch,
2495 : * and then we could assume that the output inherits the outer relation's
2496 : * ordering, which might save a sort step. However there is considerable
2497 : * downside if our estimate of the inner relation size is badly off. For
2498 : * the moment we don't risk it. (Note also that if we wanted to take this
2499 : * seriously, joinpath.c would have to consider many more paths for the
2500 : * outer rel than it does now.)
2501 : */
2502 453302 : pathnode->jpath.path.pathkeys = NIL;
2503 453302 : pathnode->jpath.jointype = jointype;
2504 453302 : pathnode->jpath.inner_unique = extra->inner_unique;
2505 453302 : pathnode->jpath.outerjoinpath = outer_path;
2506 453302 : pathnode->jpath.innerjoinpath = inner_path;
2507 453302 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2508 453302 : pathnode->path_hashclauses = hashclauses;
2509 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2510 :
2511 453302 : final_cost_hashjoin(root, pathnode, workspace, extra);
2512 :
2513 453302 : return pathnode;
2514 : }
2515 :
2516 : /*
2517 : * create_projection_path
2518 : * Creates a pathnode that represents performing a projection.
2519 : *
2520 : * 'rel' is the parent relation associated with the result
2521 : * 'subpath' is the path representing the source of data
2522 : * 'target' is the PathTarget to be computed
2523 : */
2524 : ProjectionPath *
2525 420106 : create_projection_path(PlannerInfo *root,
2526 : RelOptInfo *rel,
2527 : Path *subpath,
2528 : PathTarget *target)
2529 : {
2530 420106 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2531 : PathTarget *oldtarget;
2532 :
2533 : /*
2534 : * We mustn't put a ProjectionPath directly above another; it's useless
2535 : * and will confuse create_projection_plan. Rather than making sure all
2536 : * callers handle that, let's implement it here, by stripping off any
2537 : * ProjectionPath in what we're given. Given this rule, there won't be
2538 : * more than one.
2539 : */
2540 420106 : if (IsA(subpath, ProjectionPath))
2541 : {
2542 12 : ProjectionPath *subpp = (ProjectionPath *) subpath;
2543 :
2544 : Assert(subpp->path.parent == rel);
2545 12 : subpath = subpp->subpath;
2546 : Assert(!IsA(subpath, ProjectionPath));
2547 : }
2548 :
2549 420106 : pathnode->path.pathtype = T_Result;
2550 420106 : pathnode->path.parent = rel;
2551 420106 : pathnode->path.pathtarget = target;
2552 420106 : pathnode->path.param_info = subpath->param_info;
2553 420106 : pathnode->path.parallel_aware = false;
2554 980072 : pathnode->path.parallel_safe = rel->consider_parallel &&
2555 551956 : subpath->parallel_safe &&
2556 131850 : is_parallel_safe(root, (Node *) target->exprs);
2557 420106 : pathnode->path.parallel_workers = subpath->parallel_workers;
2558 : /* Projection does not change the sort order */
2559 420106 : pathnode->path.pathkeys = subpath->pathkeys;
2560 :
2561 420106 : pathnode->subpath = subpath;
2562 :
2563 : /*
2564 : * We might not need a separate Result node. If the input plan node type
2565 : * can project, we can just tell it to project something else. Or, if it
2566 : * can't project but the desired target has the same expression list as
2567 : * what the input will produce anyway, we can still give it the desired
2568 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2569 : * Note: in the latter case, create_projection_plan has to recheck our
2570 : * conclusion; see comments therein.
2571 : */
2572 420106 : oldtarget = subpath->pathtarget;
2573 436456 : if (is_projection_capable_path(subpath) ||
2574 16350 : equal(oldtarget->exprs, target->exprs))
2575 : {
2576 : /* No separate Result node needed */
2577 405638 : pathnode->dummypp = true;
2578 :
2579 : /*
2580 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2581 : */
2582 405638 : pathnode->path.rows = subpath->rows;
2583 405638 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2584 405638 : pathnode->path.startup_cost = subpath->startup_cost +
2585 405638 : (target->cost.startup - oldtarget->cost.startup);
2586 405638 : pathnode->path.total_cost = subpath->total_cost +
2587 405638 : (target->cost.startup - oldtarget->cost.startup) +
2588 405638 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2589 : }
2590 : else
2591 : {
2592 : /* We really do need the Result node */
2593 14468 : pathnode->dummypp = false;
2594 :
2595 : /*
2596 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2597 : * evaluating the tlist. There is no qual to worry about.
2598 : */
2599 14468 : pathnode->path.rows = subpath->rows;
2600 14468 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2601 14468 : pathnode->path.startup_cost = subpath->startup_cost +
2602 14468 : target->cost.startup;
2603 14468 : pathnode->path.total_cost = subpath->total_cost +
2604 14468 : target->cost.startup +
2605 14468 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2606 : }
2607 :
2608 420106 : return pathnode;
2609 : }
2610 :
2611 : /*
2612 : * apply_projection_to_path
2613 : * Add a projection step, or just apply the target directly to given path.
2614 : *
2615 : * This has the same net effect as create_projection_path(), except that if
2616 : * a separate Result plan node isn't needed, we just replace the given path's
2617 : * pathtarget with the desired one. This must be used only when the caller
2618 : * knows that the given path isn't referenced elsewhere and so can be modified
2619 : * in-place.
2620 : *
2621 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2622 : * new target down to its input as well; this is a yet more invasive
2623 : * modification of the input path, which create_projection_path() can't do.
2624 : *
2625 : * Note that we mustn't change the source path's parent link; so when it is
2626 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2627 : * not caused any trouble.
2628 : *
2629 : * 'rel' is the parent relation associated with the result
2630 : * 'path' is the path representing the source of data
2631 : * 'target' is the PathTarget to be computed
2632 : */
2633 : Path *
2634 13832 : apply_projection_to_path(PlannerInfo *root,
2635 : RelOptInfo *rel,
2636 : Path *path,
2637 : PathTarget *target)
2638 : {
2639 : QualCost oldcost;
2640 :
2641 : /*
2642 : * If given path can't project, we might need a Result node, so make a
2643 : * separate ProjectionPath.
2644 : */
2645 13832 : if (!is_projection_capable_path(path))
2646 1482 : return (Path *) create_projection_path(root, rel, path, target);
2647 :
2648 : /*
2649 : * We can just jam the desired tlist into the existing path, being sure to
2650 : * update its cost estimates appropriately.
2651 : */
2652 12350 : oldcost = path->pathtarget->cost;
2653 12350 : path->pathtarget = target;
2654 :
2655 12350 : path->startup_cost += target->cost.startup - oldcost.startup;
2656 12350 : path->total_cost += target->cost.startup - oldcost.startup +
2657 12350 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2658 :
2659 : /*
2660 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2661 : * arrange for the subpath to return the required target list so that
2662 : * workers can help project. But if there is something that is not
2663 : * parallel-safe in the target expressions, then we can't.
2664 : */
2665 12374 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
2666 24 : is_parallel_safe(root, (Node *) target->exprs))
2667 : {
2668 : /*
2669 : * We always use create_projection_path here, even if the subpath is
2670 : * projection-capable, so as to avoid modifying the subpath in place.
2671 : * It seems unlikely at present that there could be any other
2672 : * references to the subpath, but better safe than sorry.
2673 : *
2674 : * Note that we don't change the parallel path's cost estimates; it
2675 : * might be appropriate to do so, to reflect the fact that the bulk of
2676 : * the target evaluation will happen in workers.
2677 : */
2678 24 : if (IsA(path, GatherPath))
2679 : {
2680 0 : GatherPath *gpath = (GatherPath *) path;
2681 :
2682 0 : gpath->subpath = (Path *)
2683 0 : create_projection_path(root,
2684 0 : gpath->subpath->parent,
2685 : gpath->subpath,
2686 : target);
2687 : }
2688 : else
2689 : {
2690 24 : GatherMergePath *gmpath = (GatherMergePath *) path;
2691 :
2692 24 : gmpath->subpath = (Path *)
2693 24 : create_projection_path(root,
2694 24 : gmpath->subpath->parent,
2695 : gmpath->subpath,
2696 : target);
2697 : }
2698 : }
2699 12326 : else if (path->parallel_safe &&
2700 4626 : !is_parallel_safe(root, (Node *) target->exprs))
2701 : {
2702 : /*
2703 : * We're inserting a parallel-restricted target list into a path
2704 : * currently marked parallel-safe, so we have to mark it as no longer
2705 : * safe.
2706 : */
2707 12 : path->parallel_safe = false;
2708 : }
2709 :
2710 12350 : return path;
2711 : }
2712 :
2713 : /*
2714 : * create_set_projection_path
2715 : * Creates a pathnode that represents performing a projection that
2716 : * includes set-returning functions.
2717 : *
2718 : * 'rel' is the parent relation associated with the result
2719 : * 'subpath' is the path representing the source of data
2720 : * 'target' is the PathTarget to be computed
2721 : */
2722 : ProjectSetPath *
2723 11994 : create_set_projection_path(PlannerInfo *root,
2724 : RelOptInfo *rel,
2725 : Path *subpath,
2726 : PathTarget *target)
2727 : {
2728 11994 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2729 : double tlist_rows;
2730 : ListCell *lc;
2731 :
2732 11994 : pathnode->path.pathtype = T_ProjectSet;
2733 11994 : pathnode->path.parent = rel;
2734 11994 : pathnode->path.pathtarget = target;
2735 : /* For now, assume we are above any joins, so no parameterization */
2736 11994 : pathnode->path.param_info = NULL;
2737 11994 : pathnode->path.parallel_aware = false;
2738 28434 : pathnode->path.parallel_safe = rel->consider_parallel &&
2739 16404 : subpath->parallel_safe &&
2740 4410 : is_parallel_safe(root, (Node *) target->exprs);
2741 11994 : pathnode->path.parallel_workers = subpath->parallel_workers;
2742 : /* Projection does not change the sort order XXX? */
2743 11994 : pathnode->path.pathkeys = subpath->pathkeys;
2744 :
2745 11994 : pathnode->subpath = subpath;
2746 :
2747 : /*
2748 : * Estimate number of rows produced by SRFs for each row of input; if
2749 : * there's more than one in this node, use the maximum.
2750 : */
2751 11994 : tlist_rows = 1;
2752 25984 : foreach(lc, target->exprs)
2753 : {
2754 13990 : Node *node = (Node *) lfirst(lc);
2755 : double itemrows;
2756 :
2757 13990 : itemrows = expression_returns_set_rows(root, node);
2758 13990 : if (tlist_rows < itemrows)
2759 11562 : tlist_rows = itemrows;
2760 : }
2761 :
2762 : /*
2763 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2764 : * per input row, and half of cpu_tuple_cost for each added output row.
2765 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2766 : * this estimate later.
2767 : */
2768 11994 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2769 11994 : pathnode->path.rows = subpath->rows * tlist_rows;
2770 11994 : pathnode->path.startup_cost = subpath->startup_cost +
2771 11994 : target->cost.startup;
2772 11994 : pathnode->path.total_cost = subpath->total_cost +
2773 11994 : target->cost.startup +
2774 11994 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2775 11994 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2776 :
2777 11994 : return pathnode;
2778 : }
2779 :
2780 : /*
2781 : * create_incremental_sort_path
2782 : * Creates a pathnode that represents performing an incremental sort.
2783 : *
2784 : * 'rel' is the parent relation associated with the result
2785 : * 'subpath' is the path representing the source of data
2786 : * 'pathkeys' represents the desired sort order
2787 : * 'presorted_keys' is the number of keys by which the input path is
2788 : * already sorted
2789 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2790 : * or -1 if no LIMIT or couldn't estimate
2791 : */
2792 : IncrementalSortPath *
2793 10358 : create_incremental_sort_path(PlannerInfo *root,
2794 : RelOptInfo *rel,
2795 : Path *subpath,
2796 : List *pathkeys,
2797 : int presorted_keys,
2798 : double limit_tuples)
2799 : {
2800 10358 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
2801 10358 : SortPath *pathnode = &sort->spath;
2802 :
2803 10358 : pathnode->path.pathtype = T_IncrementalSort;
2804 10358 : pathnode->path.parent = rel;
2805 : /* Sort doesn't project, so use source path's pathtarget */
2806 10358 : pathnode->path.pathtarget = subpath->pathtarget;
2807 10358 : pathnode->path.param_info = subpath->param_info;
2808 10358 : pathnode->path.parallel_aware = false;
2809 15828 : pathnode->path.parallel_safe = rel->consider_parallel &&
2810 5470 : subpath->parallel_safe;
2811 10358 : pathnode->path.parallel_workers = subpath->parallel_workers;
2812 10358 : pathnode->path.pathkeys = pathkeys;
2813 :
2814 10358 : pathnode->subpath = subpath;
2815 :
2816 10358 : cost_incremental_sort(&pathnode->path,
2817 : root, pathkeys, presorted_keys,
2818 : subpath->disabled_nodes,
2819 : subpath->startup_cost,
2820 : subpath->total_cost,
2821 : subpath->rows,
2822 10358 : subpath->pathtarget->width,
2823 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2824 : work_mem, limit_tuples);
2825 :
2826 10358 : sort->nPresortedCols = presorted_keys;
2827 :
2828 10358 : return sort;
2829 : }
2830 :
2831 : /*
2832 : * create_sort_path
2833 : * Creates a pathnode that represents performing an explicit sort.
2834 : *
2835 : * 'rel' is the parent relation associated with the result
2836 : * 'subpath' is the path representing the source of data
2837 : * 'pathkeys' represents the desired sort order
2838 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2839 : * or -1 if no LIMIT or couldn't estimate
2840 : */
2841 : SortPath *
2842 116508 : create_sort_path(PlannerInfo *root,
2843 : RelOptInfo *rel,
2844 : Path *subpath,
2845 : List *pathkeys,
2846 : double limit_tuples)
2847 : {
2848 116508 : SortPath *pathnode = makeNode(SortPath);
2849 :
2850 116508 : pathnode->path.pathtype = T_Sort;
2851 116508 : pathnode->path.parent = rel;
2852 : /* Sort doesn't project, so use source path's pathtarget */
2853 116508 : pathnode->path.pathtarget = subpath->pathtarget;
2854 116508 : pathnode->path.param_info = subpath->param_info;
2855 116508 : pathnode->path.parallel_aware = false;
2856 203036 : pathnode->path.parallel_safe = rel->consider_parallel &&
2857 86528 : subpath->parallel_safe;
2858 116508 : pathnode->path.parallel_workers = subpath->parallel_workers;
2859 116508 : pathnode->path.pathkeys = pathkeys;
2860 :
2861 116508 : pathnode->subpath = subpath;
2862 :
2863 116508 : cost_sort(&pathnode->path, root, pathkeys,
2864 : subpath->disabled_nodes,
2865 : subpath->total_cost,
2866 : subpath->rows,
2867 116508 : subpath->pathtarget->width,
2868 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2869 : work_mem, limit_tuples);
2870 :
2871 116508 : return pathnode;
2872 : }
2873 :
2874 : /*
2875 : * create_group_path
2876 : * Creates a pathnode that represents performing grouping of presorted input
2877 : *
2878 : * 'rel' is the parent relation associated with the result
2879 : * 'subpath' is the path representing the source of data
2880 : * 'target' is the PathTarget to be computed
2881 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2882 : * 'qual' is the HAVING quals if any
2883 : * 'numGroups' is the estimated number of groups
2884 : */
2885 : GroupPath *
2886 1226 : create_group_path(PlannerInfo *root,
2887 : RelOptInfo *rel,
2888 : Path *subpath,
2889 : List *groupClause,
2890 : List *qual,
2891 : double numGroups)
2892 : {
2893 1226 : GroupPath *pathnode = makeNode(GroupPath);
2894 1226 : PathTarget *target = rel->reltarget;
2895 :
2896 1226 : pathnode->path.pathtype = T_Group;
2897 1226 : pathnode->path.parent = rel;
2898 1226 : pathnode->path.pathtarget = target;
2899 : /* For now, assume we are above any joins, so no parameterization */
2900 1226 : pathnode->path.param_info = NULL;
2901 1226 : pathnode->path.parallel_aware = false;
2902 1970 : pathnode->path.parallel_safe = rel->consider_parallel &&
2903 744 : subpath->parallel_safe;
2904 1226 : pathnode->path.parallel_workers = subpath->parallel_workers;
2905 : /* Group doesn't change sort ordering */
2906 1226 : pathnode->path.pathkeys = subpath->pathkeys;
2907 :
2908 1226 : pathnode->subpath = subpath;
2909 :
2910 1226 : pathnode->groupClause = groupClause;
2911 1226 : pathnode->qual = qual;
2912 :
2913 1226 : cost_group(&pathnode->path, root,
2914 : list_length(groupClause),
2915 : numGroups,
2916 : qual,
2917 : subpath->disabled_nodes,
2918 : subpath->startup_cost, subpath->total_cost,
2919 : subpath->rows);
2920 :
2921 : /* add tlist eval cost for each output row */
2922 1226 : pathnode->path.startup_cost += target->cost.startup;
2923 1226 : pathnode->path.total_cost += target->cost.startup +
2924 1226 : target->cost.per_tuple * pathnode->path.rows;
2925 :
2926 1226 : return pathnode;
2927 : }
2928 :
2929 : /*
2930 : * create_unique_path
2931 : * Creates a pathnode that represents performing an explicit Unique step
2932 : * on presorted input.
2933 : *
2934 : * 'rel' is the parent relation associated with the result
2935 : * 'subpath' is the path representing the source of data
2936 : * 'numCols' is the number of grouping columns
2937 : * 'numGroups' is the estimated number of groups
2938 : *
2939 : * The input path must be sorted on the grouping columns, plus possibly
2940 : * additional columns; so the first numCols pathkeys are the grouping columns
2941 : */
2942 : UniquePath *
2943 22394 : create_unique_path(PlannerInfo *root,
2944 : RelOptInfo *rel,
2945 : Path *subpath,
2946 : int numCols,
2947 : double numGroups)
2948 : {
2949 22394 : UniquePath *pathnode = makeNode(UniquePath);
2950 :
2951 22394 : pathnode->path.pathtype = T_Unique;
2952 22394 : pathnode->path.parent = rel;
2953 : /* Unique doesn't project, so use source path's pathtarget */
2954 22394 : pathnode->path.pathtarget = subpath->pathtarget;
2955 22394 : pathnode->path.param_info = subpath->param_info;
2956 22394 : pathnode->path.parallel_aware = false;
2957 40538 : pathnode->path.parallel_safe = rel->consider_parallel &&
2958 18144 : subpath->parallel_safe;
2959 22394 : pathnode->path.parallel_workers = subpath->parallel_workers;
2960 : /* Unique doesn't change the input ordering */
2961 22394 : pathnode->path.pathkeys = subpath->pathkeys;
2962 :
2963 22394 : pathnode->subpath = subpath;
2964 22394 : pathnode->numkeys = numCols;
2965 :
2966 : /*
2967 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
2968 : * all columns get compared at most of the tuples. (XXX probably this is
2969 : * an overestimate.)
2970 : */
2971 22394 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2972 22394 : pathnode->path.startup_cost = subpath->startup_cost;
2973 22394 : pathnode->path.total_cost = subpath->total_cost +
2974 22394 : cpu_operator_cost * subpath->rows * numCols;
2975 22394 : pathnode->path.rows = numGroups;
2976 :
2977 22394 : return pathnode;
2978 : }
2979 :
2980 : /*
2981 : * create_agg_path
2982 : * Creates a pathnode that represents performing aggregation/grouping
2983 : *
2984 : * 'rel' is the parent relation associated with the result
2985 : * 'subpath' is the path representing the source of data
2986 : * 'target' is the PathTarget to be computed
2987 : * 'aggstrategy' is the Agg node's basic implementation strategy
2988 : * 'aggsplit' is the Agg node's aggregate-splitting mode
2989 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2990 : * 'qual' is the HAVING quals if any
2991 : * 'aggcosts' contains cost info about the aggregate functions to be computed
2992 : * 'numGroups' is the estimated number of groups (1 if not grouping)
2993 : */
2994 : AggPath *
2995 81002 : create_agg_path(PlannerInfo *root,
2996 : RelOptInfo *rel,
2997 : Path *subpath,
2998 : PathTarget *target,
2999 : AggStrategy aggstrategy,
3000 : AggSplit aggsplit,
3001 : List *groupClause,
3002 : List *qual,
3003 : const AggClauseCosts *aggcosts,
3004 : double numGroups)
3005 : {
3006 81002 : AggPath *pathnode = makeNode(AggPath);
3007 :
3008 81002 : pathnode->path.pathtype = T_Agg;
3009 81002 : pathnode->path.parent = rel;
3010 81002 : pathnode->path.pathtarget = target;
3011 81002 : pathnode->path.param_info = subpath->param_info;
3012 81002 : pathnode->path.parallel_aware = false;
3013 141688 : pathnode->path.parallel_safe = rel->consider_parallel &&
3014 60686 : subpath->parallel_safe;
3015 81002 : pathnode->path.parallel_workers = subpath->parallel_workers;
3016 :
3017 81002 : if (aggstrategy == AGG_SORTED)
3018 : {
3019 : /*
3020 : * Attempt to preserve the order of the subpath. Additional pathkeys
3021 : * may have been added in adjust_group_pathkeys_for_groupagg() to
3022 : * support ORDER BY / DISTINCT aggregates. Pathkeys added there
3023 : * belong to columns within the aggregate function, so we must strip
3024 : * these additional pathkeys off as those columns are unavailable
3025 : * above the aggregate node.
3026 : */
3027 14224 : if (list_length(subpath->pathkeys) > root->num_groupby_pathkeys)
3028 808 : pathnode->path.pathkeys = list_copy_head(subpath->pathkeys,
3029 : root->num_groupby_pathkeys);
3030 : else
3031 13416 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3032 : }
3033 : else
3034 66778 : pathnode->path.pathkeys = NIL; /* output is unordered */
3035 :
3036 81002 : pathnode->subpath = subpath;
3037 :
3038 81002 : pathnode->aggstrategy = aggstrategy;
3039 81002 : pathnode->aggsplit = aggsplit;
3040 81002 : pathnode->numGroups = numGroups;
3041 81002 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
3042 81002 : pathnode->groupClause = groupClause;
3043 81002 : pathnode->qual = qual;
3044 :
3045 81002 : cost_agg(&pathnode->path, root,
3046 : aggstrategy, aggcosts,
3047 : list_length(groupClause), numGroups,
3048 : qual,
3049 : subpath->disabled_nodes,
3050 : subpath->startup_cost, subpath->total_cost,
3051 81002 : subpath->rows, subpath->pathtarget->width);
3052 :
3053 : /* add tlist eval cost for each output row */
3054 81002 : pathnode->path.startup_cost += target->cost.startup;
3055 81002 : pathnode->path.total_cost += target->cost.startup +
3056 81002 : target->cost.per_tuple * pathnode->path.rows;
3057 :
3058 81002 : return pathnode;
3059 : }
3060 :
3061 : /*
3062 : * create_groupingsets_path
3063 : * Creates a pathnode that represents performing GROUPING SETS aggregation
3064 : *
3065 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3066 : * The input path's result must be sorted to match the last entry in
3067 : * rollup_groupclauses.
3068 : *
3069 : * 'rel' is the parent relation associated with the result
3070 : * 'subpath' is the path representing the source of data
3071 : * 'target' is the PathTarget to be computed
3072 : * 'having_qual' is the HAVING quals if any
3073 : * 'rollups' is a list of RollupData nodes
3074 : * 'agg_costs' contains cost info about the aggregate functions to be computed
3075 : */
3076 : GroupingSetsPath *
3077 2230 : create_groupingsets_path(PlannerInfo *root,
3078 : RelOptInfo *rel,
3079 : Path *subpath,
3080 : List *having_qual,
3081 : AggStrategy aggstrategy,
3082 : List *rollups,
3083 : const AggClauseCosts *agg_costs)
3084 : {
3085 2230 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
3086 2230 : PathTarget *target = rel->reltarget;
3087 : ListCell *lc;
3088 2230 : bool is_first = true;
3089 2230 : bool is_first_sort = true;
3090 :
3091 : /* The topmost generated Plan node will be an Agg */
3092 2230 : pathnode->path.pathtype = T_Agg;
3093 2230 : pathnode->path.parent = rel;
3094 2230 : pathnode->path.pathtarget = target;
3095 2230 : pathnode->path.param_info = subpath->param_info;
3096 2230 : pathnode->path.parallel_aware = false;
3097 3220 : pathnode->path.parallel_safe = rel->consider_parallel &&
3098 990 : subpath->parallel_safe;
3099 2230 : pathnode->path.parallel_workers = subpath->parallel_workers;
3100 2230 : pathnode->subpath = subpath;
3101 :
3102 : /*
3103 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3104 : * to AGG_HASHED, here if possible.
3105 : */
3106 3192 : if (aggstrategy == AGG_SORTED &&
3107 962 : list_length(rollups) == 1 &&
3108 500 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3109 60 : aggstrategy = AGG_PLAIN;
3110 :
3111 3170 : if (aggstrategy == AGG_MIXED &&
3112 940 : list_length(rollups) == 1)
3113 0 : aggstrategy = AGG_HASHED;
3114 :
3115 : /*
3116 : * Output will be in sorted order by group_pathkeys if, and only if, there
3117 : * is a single rollup operation on a non-empty list of grouping
3118 : * expressions.
3119 : */
3120 2230 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3121 440 : pathnode->path.pathkeys = root->group_pathkeys;
3122 : else
3123 1790 : pathnode->path.pathkeys = NIL;
3124 :
3125 2230 : pathnode->aggstrategy = aggstrategy;
3126 2230 : pathnode->rollups = rollups;
3127 2230 : pathnode->qual = having_qual;
3128 2230 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3129 :
3130 : Assert(rollups != NIL);
3131 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3132 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3133 :
3134 7680 : foreach(lc, rollups)
3135 : {
3136 5450 : RollupData *rollup = lfirst(lc);
3137 5450 : List *gsets = rollup->gsets;
3138 5450 : int numGroupCols = list_length(linitial(gsets));
3139 :
3140 : /*
3141 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3142 : * (already-sorted) input, and following ones do their own sort.
3143 : *
3144 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3145 : *
3146 : * In AGG_MIXED mode, the first rollups are hashed, the first
3147 : * non-hashed one takes the (already-sorted) input, and following ones
3148 : * do their own sort.
3149 : */
3150 5450 : if (is_first)
3151 : {
3152 2230 : cost_agg(&pathnode->path, root,
3153 : aggstrategy,
3154 : agg_costs,
3155 : numGroupCols,
3156 : rollup->numGroups,
3157 : having_qual,
3158 : subpath->disabled_nodes,
3159 : subpath->startup_cost,
3160 : subpath->total_cost,
3161 : subpath->rows,
3162 2230 : subpath->pathtarget->width);
3163 2230 : is_first = false;
3164 2230 : if (!rollup->is_hashed)
3165 962 : is_first_sort = false;
3166 : }
3167 : else
3168 : {
3169 : Path sort_path; /* dummy for result of cost_sort */
3170 : Path agg_path; /* dummy for result of cost_agg */
3171 :
3172 3220 : if (rollup->is_hashed || is_first_sort)
3173 : {
3174 : /*
3175 : * Account for cost of aggregation, but don't charge input
3176 : * cost again
3177 : */
3178 2470 : cost_agg(&agg_path, root,
3179 2470 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3180 : agg_costs,
3181 : numGroupCols,
3182 : rollup->numGroups,
3183 : having_qual,
3184 : 0, 0.0, 0.0,
3185 : subpath->rows,
3186 2470 : subpath->pathtarget->width);
3187 2470 : if (!rollup->is_hashed)
3188 940 : is_first_sort = false;
3189 : }
3190 : else
3191 : {
3192 : /* Account for cost of sort, but don't charge input cost again */
3193 750 : cost_sort(&sort_path, root, NIL, 0,
3194 : 0.0,
3195 : subpath->rows,
3196 750 : subpath->pathtarget->width,
3197 : 0.0,
3198 : work_mem,
3199 : -1.0);
3200 :
3201 : /* Account for cost of aggregation */
3202 :
3203 750 : cost_agg(&agg_path, root,
3204 : AGG_SORTED,
3205 : agg_costs,
3206 : numGroupCols,
3207 : rollup->numGroups,
3208 : having_qual,
3209 : sort_path.disabled_nodes,
3210 : sort_path.startup_cost,
3211 : sort_path.total_cost,
3212 : sort_path.rows,
3213 750 : subpath->pathtarget->width);
3214 : }
3215 :
3216 3220 : pathnode->path.disabled_nodes += agg_path.disabled_nodes;
3217 3220 : pathnode->path.total_cost += agg_path.total_cost;
3218 3220 : pathnode->path.rows += agg_path.rows;
3219 : }
3220 : }
3221 :
3222 : /* add tlist eval cost for each output row */
3223 2230 : pathnode->path.startup_cost += target->cost.startup;
3224 2230 : pathnode->path.total_cost += target->cost.startup +
3225 2230 : target->cost.per_tuple * pathnode->path.rows;
3226 :
3227 2230 : return pathnode;
3228 : }
3229 :
3230 : /*
3231 : * create_minmaxagg_path
3232 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3233 : *
3234 : * 'rel' is the parent relation associated with the result
3235 : * 'target' is the PathTarget to be computed
3236 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3237 : * 'quals' is the HAVING quals if any
3238 : */
3239 : MinMaxAggPath *
3240 422 : create_minmaxagg_path(PlannerInfo *root,
3241 : RelOptInfo *rel,
3242 : PathTarget *target,
3243 : List *mmaggregates,
3244 : List *quals)
3245 : {
3246 422 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3247 : Cost initplan_cost;
3248 422 : int initplan_disabled_nodes = 0;
3249 : ListCell *lc;
3250 :
3251 : /* The topmost generated Plan node will be a Result */
3252 422 : pathnode->path.pathtype = T_Result;
3253 422 : pathnode->path.parent = rel;
3254 422 : pathnode->path.pathtarget = target;
3255 : /* For now, assume we are above any joins, so no parameterization */
3256 422 : pathnode->path.param_info = NULL;
3257 422 : pathnode->path.parallel_aware = false;
3258 422 : pathnode->path.parallel_safe = true; /* might change below */
3259 422 : pathnode->path.parallel_workers = 0;
3260 : /* Result is one unordered row */
3261 422 : pathnode->path.rows = 1;
3262 422 : pathnode->path.pathkeys = NIL;
3263 :
3264 422 : pathnode->mmaggregates = mmaggregates;
3265 422 : pathnode->quals = quals;
3266 :
3267 : /* Calculate cost of all the initplans, and check parallel safety */
3268 422 : initplan_cost = 0;
3269 880 : foreach(lc, mmaggregates)
3270 : {
3271 458 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3272 :
3273 458 : initplan_disabled_nodes += mminfo->path->disabled_nodes;
3274 458 : initplan_cost += mminfo->pathcost;
3275 458 : if (!mminfo->path->parallel_safe)
3276 110 : pathnode->path.parallel_safe = false;
3277 : }
3278 :
3279 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3280 422 : pathnode->path.disabled_nodes = initplan_disabled_nodes;
3281 422 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3282 422 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3283 422 : target->cost.per_tuple + cpu_tuple_cost;
3284 :
3285 : /*
3286 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3287 : * rowcount estimate should be 1 no matter what the qual is.
3288 : */
3289 422 : if (quals)
3290 : {
3291 : QualCost qual_cost;
3292 :
3293 0 : cost_qual_eval(&qual_cost, quals, root);
3294 0 : pathnode->path.startup_cost += qual_cost.startup;
3295 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3296 : }
3297 :
3298 : /*
3299 : * If the initplans were all parallel-safe, also check safety of the
3300 : * target and quals. (The Result node itself isn't parallelizable, but if
3301 : * we are in a subquery then it can be useful for the outer query to know
3302 : * that this one is parallel-safe.)
3303 : */
3304 422 : if (pathnode->path.parallel_safe)
3305 312 : pathnode->path.parallel_safe =
3306 624 : is_parallel_safe(root, (Node *) target->exprs) &&
3307 312 : is_parallel_safe(root, (Node *) quals);
3308 :
3309 422 : return pathnode;
3310 : }
3311 :
3312 : /*
3313 : * create_windowagg_path
3314 : * Creates a pathnode that represents computation of window functions
3315 : *
3316 : * 'rel' is the parent relation associated with the result
3317 : * 'subpath' is the path representing the source of data
3318 : * 'target' is the PathTarget to be computed
3319 : * 'windowFuncs' is a list of WindowFunc structs
3320 : * 'runCondition' is a list of OpExprs to short-circuit WindowAgg execution
3321 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3322 : * 'qual' WindowClause.runconditions from lower-level WindowAggPaths.
3323 : * Must always be NIL when topwindow == false
3324 : * 'topwindow' pass as true only for the top-level WindowAgg. False for all
3325 : * intermediate WindowAggs.
3326 : *
3327 : * The input must be sorted according to the WindowClause's PARTITION keys
3328 : * plus ORDER BY keys.
3329 : */
3330 : WindowAggPath *
3331 2964 : create_windowagg_path(PlannerInfo *root,
3332 : RelOptInfo *rel,
3333 : Path *subpath,
3334 : PathTarget *target,
3335 : List *windowFuncs,
3336 : List *runCondition,
3337 : WindowClause *winclause,
3338 : List *qual,
3339 : bool topwindow)
3340 : {
3341 2964 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3342 :
3343 : /* qual can only be set for the topwindow */
3344 : Assert(qual == NIL || topwindow);
3345 :
3346 2964 : pathnode->path.pathtype = T_WindowAgg;
3347 2964 : pathnode->path.parent = rel;
3348 2964 : pathnode->path.pathtarget = target;
3349 : /* For now, assume we are above any joins, so no parameterization */
3350 2964 : pathnode->path.param_info = NULL;
3351 2964 : pathnode->path.parallel_aware = false;
3352 2964 : pathnode->path.parallel_safe = rel->consider_parallel &&
3353 0 : subpath->parallel_safe;
3354 2964 : pathnode->path.parallel_workers = subpath->parallel_workers;
3355 : /* WindowAgg preserves the input sort order */
3356 2964 : pathnode->path.pathkeys = subpath->pathkeys;
3357 :
3358 2964 : pathnode->subpath = subpath;
3359 2964 : pathnode->winclause = winclause;
3360 2964 : pathnode->qual = qual;
3361 2964 : pathnode->runCondition = runCondition;
3362 2964 : pathnode->topwindow = topwindow;
3363 :
3364 : /*
3365 : * For costing purposes, assume that there are no redundant partitioning
3366 : * or ordering columns; it's not worth the trouble to deal with that
3367 : * corner case here. So we just pass the unmodified list lengths to
3368 : * cost_windowagg.
3369 : */
3370 2964 : cost_windowagg(&pathnode->path, root,
3371 : windowFuncs,
3372 : winclause,
3373 : subpath->disabled_nodes,
3374 : subpath->startup_cost,
3375 : subpath->total_cost,
3376 : subpath->rows);
3377 :
3378 : /* add tlist eval cost for each output row */
3379 2964 : pathnode->path.startup_cost += target->cost.startup;
3380 2964 : pathnode->path.total_cost += target->cost.startup +
3381 2964 : target->cost.per_tuple * pathnode->path.rows;
3382 :
3383 2964 : return pathnode;
3384 : }
3385 :
3386 : /*
3387 : * create_setop_path
3388 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3389 : *
3390 : * 'rel' is the parent relation associated with the result
3391 : * 'leftpath' is the path representing the left-hand source of data
3392 : * 'rightpath' is the path representing the right-hand source of data
3393 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3394 : * 'strategy' is the implementation strategy (sorted or hashed)
3395 : * 'groupList' is a list of SortGroupClause's representing the grouping
3396 : * 'numGroups' is the estimated number of distinct groups in left-hand input
3397 : * 'outputRows' is the estimated number of output rows
3398 : *
3399 : * leftpath and rightpath must produce the same columns. Moreover, if
3400 : * strategy is SETOP_SORTED, leftpath and rightpath must both be sorted
3401 : * by all the grouping columns.
3402 : */
3403 : SetOpPath *
3404 1276 : create_setop_path(PlannerInfo *root,
3405 : RelOptInfo *rel,
3406 : Path *leftpath,
3407 : Path *rightpath,
3408 : SetOpCmd cmd,
3409 : SetOpStrategy strategy,
3410 : List *groupList,
3411 : double numGroups,
3412 : double outputRows)
3413 : {
3414 1276 : SetOpPath *pathnode = makeNode(SetOpPath);
3415 :
3416 1276 : pathnode->path.pathtype = T_SetOp;
3417 1276 : pathnode->path.parent = rel;
3418 1276 : pathnode->path.pathtarget = rel->reltarget;
3419 : /* For now, assume we are above any joins, so no parameterization */
3420 1276 : pathnode->path.param_info = NULL;
3421 1276 : pathnode->path.parallel_aware = false;
3422 2552 : pathnode->path.parallel_safe = rel->consider_parallel &&
3423 1276 : leftpath->parallel_safe && rightpath->parallel_safe;
3424 1276 : pathnode->path.parallel_workers =
3425 1276 : leftpath->parallel_workers + rightpath->parallel_workers;
3426 : /* SetOp preserves the input sort order if in sort mode */
3427 1276 : pathnode->path.pathkeys =
3428 1276 : (strategy == SETOP_SORTED) ? leftpath->pathkeys : NIL;
3429 :
3430 1276 : pathnode->leftpath = leftpath;
3431 1276 : pathnode->rightpath = rightpath;
3432 1276 : pathnode->cmd = cmd;
3433 1276 : pathnode->strategy = strategy;
3434 1276 : pathnode->groupList = groupList;
3435 1276 : pathnode->numGroups = numGroups;
3436 :
3437 : /*
3438 : * Compute cost estimates. As things stand, we end up with the same total
3439 : * cost in this node for sort and hash methods, but different startup
3440 : * costs. This could be refined perhaps, but it'll do for now.
3441 : */
3442 1276 : pathnode->path.disabled_nodes =
3443 1276 : leftpath->disabled_nodes + rightpath->disabled_nodes;
3444 1276 : if (strategy == SETOP_SORTED)
3445 : {
3446 : /*
3447 : * In sorted mode, we can emit output incrementally. Charge one
3448 : * cpu_operator_cost per comparison per input tuple. Like cost_group,
3449 : * we assume all columns get compared at most of the tuples.
3450 : */
3451 668 : pathnode->path.startup_cost =
3452 668 : leftpath->startup_cost + rightpath->startup_cost;
3453 668 : pathnode->path.total_cost =
3454 1336 : leftpath->total_cost + rightpath->total_cost +
3455 668 : cpu_operator_cost * (leftpath->rows + rightpath->rows) * list_length(groupList);
3456 :
3457 : /*
3458 : * Also charge a small amount per extracted tuple. Like cost_sort,
3459 : * charge only operator cost not cpu_tuple_cost, since SetOp does no
3460 : * qual-checking or projection.
3461 : */
3462 668 : pathnode->path.total_cost += cpu_operator_cost * outputRows;
3463 : }
3464 : else
3465 : {
3466 : Size hashtablesize;
3467 :
3468 : /*
3469 : * In hashed mode, we must read all the input before we can emit
3470 : * anything. Also charge comparison costs to represent the cost of
3471 : * hash table lookups.
3472 : */
3473 608 : pathnode->path.startup_cost =
3474 1216 : leftpath->total_cost + rightpath->total_cost +
3475 608 : cpu_operator_cost * (leftpath->rows + rightpath->rows) * list_length(groupList);
3476 608 : pathnode->path.total_cost = pathnode->path.startup_cost;
3477 :
3478 : /*
3479 : * Also charge a small amount per extracted tuple. Like cost_sort,
3480 : * charge only operator cost not cpu_tuple_cost, since SetOp does no
3481 : * qual-checking or projection.
3482 : */
3483 608 : pathnode->path.total_cost += cpu_operator_cost * outputRows;
3484 :
3485 : /*
3486 : * Mark the path as disabled if enable_hashagg is off. While this
3487 : * isn't exactly a HashAgg node, it seems close enough to justify
3488 : * letting that switch control it.
3489 : */
3490 608 : if (!enable_hashagg)
3491 114 : pathnode->path.disabled_nodes++;
3492 :
3493 : /*
3494 : * Also disable if it doesn't look like the hashtable will fit into
3495 : * hash_mem. (Note: reject on equality, to ensure that an estimate of
3496 : * SIZE_MAX disables hashing regardless of the hash_mem limit.)
3497 : */
3498 608 : hashtablesize = EstimateSetOpHashTableSpace(numGroups,
3499 608 : leftpath->pathtarget->width);
3500 608 : if (hashtablesize >= get_hash_memory_limit())
3501 0 : pathnode->path.disabled_nodes++;
3502 : }
3503 1276 : pathnode->path.rows = outputRows;
3504 :
3505 1276 : return pathnode;
3506 : }
3507 :
3508 : /*
3509 : * create_recursiveunion_path
3510 : * Creates a pathnode that represents a recursive UNION node
3511 : *
3512 : * 'rel' is the parent relation associated with the result
3513 : * 'leftpath' is the source of data for the non-recursive term
3514 : * 'rightpath' is the source of data for the recursive term
3515 : * 'target' is the PathTarget to be computed
3516 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3517 : * 'wtParam' is the ID of Param representing work table
3518 : * 'numGroups' is the estimated number of groups
3519 : *
3520 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3521 : */
3522 : RecursiveUnionPath *
3523 928 : create_recursiveunion_path(PlannerInfo *root,
3524 : RelOptInfo *rel,
3525 : Path *leftpath,
3526 : Path *rightpath,
3527 : PathTarget *target,
3528 : List *distinctList,
3529 : int wtParam,
3530 : double numGroups)
3531 : {
3532 928 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3533 :
3534 928 : pathnode->path.pathtype = T_RecursiveUnion;
3535 928 : pathnode->path.parent = rel;
3536 928 : pathnode->path.pathtarget = target;
3537 : /* For now, assume we are above any joins, so no parameterization */
3538 928 : pathnode->path.param_info = NULL;
3539 928 : pathnode->path.parallel_aware = false;
3540 1856 : pathnode->path.parallel_safe = rel->consider_parallel &&
3541 928 : leftpath->parallel_safe && rightpath->parallel_safe;
3542 : /* Foolish, but we'll do it like joins for now: */
3543 928 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3544 : /* RecursiveUnion result is always unsorted */
3545 928 : pathnode->path.pathkeys = NIL;
3546 :
3547 928 : pathnode->leftpath = leftpath;
3548 928 : pathnode->rightpath = rightpath;
3549 928 : pathnode->distinctList = distinctList;
3550 928 : pathnode->wtParam = wtParam;
3551 928 : pathnode->numGroups = numGroups;
3552 :
3553 928 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3554 :
3555 928 : return pathnode;
3556 : }
3557 :
3558 : /*
3559 : * create_lockrows_path
3560 : * Creates a pathnode that represents acquiring row locks
3561 : *
3562 : * 'rel' is the parent relation associated with the result
3563 : * 'subpath' is the path representing the source of data
3564 : * 'rowMarks' is a list of PlanRowMark's
3565 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3566 : */
3567 : LockRowsPath *
3568 8376 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3569 : Path *subpath, List *rowMarks, int epqParam)
3570 : {
3571 8376 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3572 :
3573 8376 : pathnode->path.pathtype = T_LockRows;
3574 8376 : pathnode->path.parent = rel;
3575 : /* LockRows doesn't project, so use source path's pathtarget */
3576 8376 : pathnode->path.pathtarget = subpath->pathtarget;
3577 : /* For now, assume we are above any joins, so no parameterization */
3578 8376 : pathnode->path.param_info = NULL;
3579 8376 : pathnode->path.parallel_aware = false;
3580 8376 : pathnode->path.parallel_safe = false;
3581 8376 : pathnode->path.parallel_workers = 0;
3582 8376 : pathnode->path.rows = subpath->rows;
3583 :
3584 : /*
3585 : * The result cannot be assumed sorted, since locking might cause the sort
3586 : * key columns to be replaced with new values.
3587 : */
3588 8376 : pathnode->path.pathkeys = NIL;
3589 :
3590 8376 : pathnode->subpath = subpath;
3591 8376 : pathnode->rowMarks = rowMarks;
3592 8376 : pathnode->epqParam = epqParam;
3593 :
3594 : /*
3595 : * We should charge something extra for the costs of row locking and
3596 : * possible refetches, but it's hard to say how much. For now, use
3597 : * cpu_tuple_cost per row.
3598 : */
3599 8376 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3600 8376 : pathnode->path.startup_cost = subpath->startup_cost;
3601 8376 : pathnode->path.total_cost = subpath->total_cost +
3602 8376 : cpu_tuple_cost * subpath->rows;
3603 :
3604 8376 : return pathnode;
3605 : }
3606 :
3607 : /*
3608 : * create_modifytable_path
3609 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE/MERGE
3610 : * mods
3611 : *
3612 : * 'rel' is the parent relation associated with the result
3613 : * 'subpath' is a Path producing source data
3614 : * 'operation' is the operation type
3615 : * 'canSetTag' is true if we set the command tag/es_processed
3616 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3617 : * 'rootRelation' is the partitioned/inherited table root RTI, or 0 if none
3618 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3619 : * 'updateColnosLists' is a list of UPDATE target column number lists
3620 : * (one sublist per rel); or NIL if not an UPDATE
3621 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3622 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3623 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3624 : * 'onconflict' is the ON CONFLICT clause, or NULL
3625 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3626 : * 'mergeActionLists' is a list of lists of MERGE actions (one per rel)
3627 : * 'mergeJoinConditions' is a list of join conditions for MERGE (one per rel)
3628 : */
3629 : ModifyTablePath *
3630 93902 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3631 : Path *subpath,
3632 : CmdType operation, bool canSetTag,
3633 : Index nominalRelation, Index rootRelation,
3634 : List *resultRelations,
3635 : List *updateColnosLists,
3636 : List *withCheckOptionLists, List *returningLists,
3637 : List *rowMarks, OnConflictExpr *onconflict,
3638 : List *mergeActionLists, List *mergeJoinConditions,
3639 : int epqParam)
3640 : {
3641 93902 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3642 :
3643 : Assert(operation == CMD_MERGE ||
3644 : (operation == CMD_UPDATE ?
3645 : list_length(resultRelations) == list_length(updateColnosLists) :
3646 : updateColnosLists == NIL));
3647 : Assert(withCheckOptionLists == NIL ||
3648 : list_length(resultRelations) == list_length(withCheckOptionLists));
3649 : Assert(returningLists == NIL ||
3650 : list_length(resultRelations) == list_length(returningLists));
3651 :
3652 93902 : pathnode->path.pathtype = T_ModifyTable;
3653 93902 : pathnode->path.parent = rel;
3654 : /* pathtarget is not interesting, just make it minimally valid */
3655 93902 : pathnode->path.pathtarget = rel->reltarget;
3656 : /* For now, assume we are above any joins, so no parameterization */
3657 93902 : pathnode->path.param_info = NULL;
3658 93902 : pathnode->path.parallel_aware = false;
3659 93902 : pathnode->path.parallel_safe = false;
3660 93902 : pathnode->path.parallel_workers = 0;
3661 93902 : pathnode->path.pathkeys = NIL;
3662 :
3663 : /*
3664 : * Compute cost & rowcount as subpath cost & rowcount (if RETURNING)
3665 : *
3666 : * Currently, we don't charge anything extra for the actual table
3667 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3668 : * expressions if any. It would only be window dressing, since
3669 : * ModifyTable is always a top-level node and there is no way for the
3670 : * costs to change any higher-level planning choices. But we might want
3671 : * to make it look better sometime.
3672 : */
3673 93902 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3674 93902 : pathnode->path.startup_cost = subpath->startup_cost;
3675 93902 : pathnode->path.total_cost = subpath->total_cost;
3676 93902 : if (returningLists != NIL)
3677 : {
3678 2994 : pathnode->path.rows = subpath->rows;
3679 :
3680 : /*
3681 : * Set width to match the subpath output. XXX this is totally wrong:
3682 : * we should return an average of the RETURNING tlist widths. But
3683 : * it's what happened historically, and improving it is a task for
3684 : * another day. (Again, it's mostly window dressing.)
3685 : */
3686 2994 : pathnode->path.pathtarget->width = subpath->pathtarget->width;
3687 : }
3688 : else
3689 : {
3690 90908 : pathnode->path.rows = 0;
3691 90908 : pathnode->path.pathtarget->width = 0;
3692 : }
3693 :
3694 93902 : pathnode->subpath = subpath;
3695 93902 : pathnode->operation = operation;
3696 93902 : pathnode->canSetTag = canSetTag;
3697 93902 : pathnode->nominalRelation = nominalRelation;
3698 93902 : pathnode->rootRelation = rootRelation;
3699 93902 : pathnode->resultRelations = resultRelations;
3700 93902 : pathnode->updateColnosLists = updateColnosLists;
3701 93902 : pathnode->withCheckOptionLists = withCheckOptionLists;
3702 93902 : pathnode->returningLists = returningLists;
3703 93902 : pathnode->rowMarks = rowMarks;
3704 93902 : pathnode->onconflict = onconflict;
3705 93902 : pathnode->epqParam = epqParam;
3706 93902 : pathnode->mergeActionLists = mergeActionLists;
3707 93902 : pathnode->mergeJoinConditions = mergeJoinConditions;
3708 :
3709 93902 : return pathnode;
3710 : }
3711 :
3712 : /*
3713 : * create_limit_path
3714 : * Creates a pathnode that represents performing LIMIT/OFFSET
3715 : *
3716 : * In addition to providing the actual OFFSET and LIMIT expressions,
3717 : * the caller must provide estimates of their values for costing purposes.
3718 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3719 : * the clause not being present, and -1 means it's present but we could
3720 : * not estimate its value.
3721 : *
3722 : * 'rel' is the parent relation associated with the result
3723 : * 'subpath' is the path representing the source of data
3724 : * 'limitOffset' is the actual OFFSET expression, or NULL
3725 : * 'limitCount' is the actual LIMIT expression, or NULL
3726 : * 'offset_est' is the estimated value of the OFFSET expression
3727 : * 'count_est' is the estimated value of the LIMIT expression
3728 : */
3729 : LimitPath *
3730 6072 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3731 : Path *subpath,
3732 : Node *limitOffset, Node *limitCount,
3733 : LimitOption limitOption,
3734 : int64 offset_est, int64 count_est)
3735 : {
3736 6072 : LimitPath *pathnode = makeNode(LimitPath);
3737 :
3738 6072 : pathnode->path.pathtype = T_Limit;
3739 6072 : pathnode->path.parent = rel;
3740 : /* Limit doesn't project, so use source path's pathtarget */
3741 6072 : pathnode->path.pathtarget = subpath->pathtarget;
3742 : /* For now, assume we are above any joins, so no parameterization */
3743 6072 : pathnode->path.param_info = NULL;
3744 6072 : pathnode->path.parallel_aware = false;
3745 8516 : pathnode->path.parallel_safe = rel->consider_parallel &&
3746 2444 : subpath->parallel_safe;
3747 6072 : pathnode->path.parallel_workers = subpath->parallel_workers;
3748 6072 : pathnode->path.rows = subpath->rows;
3749 6072 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3750 6072 : pathnode->path.startup_cost = subpath->startup_cost;
3751 6072 : pathnode->path.total_cost = subpath->total_cost;
3752 6072 : pathnode->path.pathkeys = subpath->pathkeys;
3753 6072 : pathnode->subpath = subpath;
3754 6072 : pathnode->limitOffset = limitOffset;
3755 6072 : pathnode->limitCount = limitCount;
3756 6072 : pathnode->limitOption = limitOption;
3757 :
3758 : /*
3759 : * Adjust the output rows count and costs according to the offset/limit.
3760 : */
3761 6072 : adjust_limit_rows_costs(&pathnode->path.rows,
3762 : &pathnode->path.startup_cost,
3763 : &pathnode->path.total_cost,
3764 : offset_est, count_est);
3765 :
3766 6072 : return pathnode;
3767 : }
3768 :
3769 : /*
3770 : * adjust_limit_rows_costs
3771 : * Adjust the size and cost estimates for a LimitPath node according to the
3772 : * offset/limit.
3773 : *
3774 : * This is only a cosmetic issue if we are at top level, but if we are
3775 : * building a subquery then it's important to report correct info to the outer
3776 : * planner.
3777 : *
3778 : * When the offset or count couldn't be estimated, use 10% of the estimated
3779 : * number of rows emitted from the subpath.
3780 : *
3781 : * XXX we don't bother to add eval costs of the offset/limit expressions
3782 : * themselves to the path costs. In theory we should, but in most cases those
3783 : * expressions are trivial and it's just not worth the trouble.
3784 : */
3785 : void
3786 6256 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3787 : Cost *startup_cost, /* in/out parameter */
3788 : Cost *total_cost, /* in/out parameter */
3789 : int64 offset_est,
3790 : int64 count_est)
3791 : {
3792 6256 : double input_rows = *rows;
3793 6256 : Cost input_startup_cost = *startup_cost;
3794 6256 : Cost input_total_cost = *total_cost;
3795 :
3796 6256 : if (offset_est != 0)
3797 : {
3798 : double offset_rows;
3799 :
3800 712 : if (offset_est > 0)
3801 688 : offset_rows = (double) offset_est;
3802 : else
3803 24 : offset_rows = clamp_row_est(input_rows * 0.10);
3804 712 : if (offset_rows > *rows)
3805 38 : offset_rows = *rows;
3806 712 : if (input_rows > 0)
3807 712 : *startup_cost +=
3808 712 : (input_total_cost - input_startup_cost)
3809 712 : * offset_rows / input_rows;
3810 712 : *rows -= offset_rows;
3811 712 : if (*rows < 1)
3812 46 : *rows = 1;
3813 : }
3814 :
3815 6256 : if (count_est != 0)
3816 : {
3817 : double count_rows;
3818 :
3819 6180 : if (count_est > 0)
3820 6174 : count_rows = (double) count_est;
3821 : else
3822 6 : count_rows = clamp_row_est(input_rows * 0.10);
3823 6180 : if (count_rows > *rows)
3824 248 : count_rows = *rows;
3825 6180 : if (input_rows > 0)
3826 6180 : *total_cost = *startup_cost +
3827 6180 : (input_total_cost - input_startup_cost)
3828 6180 : * count_rows / input_rows;
3829 6180 : *rows = count_rows;
3830 6180 : if (*rows < 1)
3831 0 : *rows = 1;
3832 : }
3833 6256 : }
3834 :
3835 :
3836 : /*
3837 : * reparameterize_path
3838 : * Attempt to modify a Path to have greater parameterization
3839 : *
3840 : * We use this to attempt to bring all child paths of an appendrel to the
3841 : * same parameterization level, ensuring that they all enforce the same set
3842 : * of join quals (and thus that that parameterization can be attributed to
3843 : * an append path built from such paths). Currently, only a few path types
3844 : * are supported here, though more could be added at need. We return NULL
3845 : * if we can't reparameterize the given path.
3846 : *
3847 : * Note: we intentionally do not pass created paths to add_path(); it would
3848 : * possibly try to delete them on the grounds of being cost-inferior to the
3849 : * paths they were made from, and we don't want that. Paths made here are
3850 : * not necessarily of general-purpose usefulness, but they can be useful
3851 : * as members of an append path.
3852 : */
3853 : Path *
3854 356 : reparameterize_path(PlannerInfo *root, Path *path,
3855 : Relids required_outer,
3856 : double loop_count)
3857 : {
3858 356 : RelOptInfo *rel = path->parent;
3859 :
3860 : /* Can only increase, not decrease, path's parameterization */
3861 356 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
3862 0 : return NULL;
3863 356 : switch (path->pathtype)
3864 : {
3865 264 : case T_SeqScan:
3866 264 : return create_seqscan_path(root, rel, required_outer, 0);
3867 0 : case T_SampleScan:
3868 0 : return create_samplescan_path(root, rel, required_outer);
3869 0 : case T_IndexScan:
3870 : case T_IndexOnlyScan:
3871 : {
3872 0 : IndexPath *ipath = (IndexPath *) path;
3873 0 : IndexPath *newpath = makeNode(IndexPath);
3874 :
3875 : /*
3876 : * We can't use create_index_path directly, and would not want
3877 : * to because it would re-compute the indexqual conditions
3878 : * which is wasted effort. Instead we hack things a bit:
3879 : * flat-copy the path node, revise its param_info, and redo
3880 : * the cost estimate.
3881 : */
3882 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3883 0 : newpath->path.param_info =
3884 0 : get_baserel_parampathinfo(root, rel, required_outer);
3885 0 : cost_index(newpath, root, loop_count, false);
3886 0 : return (Path *) newpath;
3887 : }
3888 0 : case T_BitmapHeapScan:
3889 : {
3890 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3891 :
3892 0 : return (Path *) create_bitmap_heap_path(root,
3893 : rel,
3894 : bpath->bitmapqual,
3895 : required_outer,
3896 : loop_count, 0);
3897 : }
3898 0 : case T_SubqueryScan:
3899 : {
3900 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
3901 0 : Path *subpath = spath->subpath;
3902 : bool trivial_pathtarget;
3903 :
3904 : /*
3905 : * If existing node has zero extra cost, we must have decided
3906 : * its target is trivial. (The converse is not true, because
3907 : * it might have a trivial target but quals to enforce; but in
3908 : * that case the new node will too, so it doesn't matter
3909 : * whether we get the right answer here.)
3910 : */
3911 0 : trivial_pathtarget =
3912 0 : (subpath->total_cost == spath->path.total_cost);
3913 :
3914 0 : return (Path *) create_subqueryscan_path(root,
3915 : rel,
3916 : subpath,
3917 : trivial_pathtarget,
3918 : spath->path.pathkeys,
3919 : required_outer);
3920 : }
3921 60 : case T_Result:
3922 : /* Supported only for RTE_RESULT scan paths */
3923 60 : if (IsA(path, Path))
3924 60 : return create_resultscan_path(root, rel, required_outer);
3925 0 : break;
3926 0 : case T_Append:
3927 : {
3928 0 : AppendPath *apath = (AppendPath *) path;
3929 0 : List *childpaths = NIL;
3930 0 : List *partialpaths = NIL;
3931 : int i;
3932 : ListCell *lc;
3933 :
3934 : /* Reparameterize the children */
3935 0 : i = 0;
3936 0 : foreach(lc, apath->subpaths)
3937 : {
3938 0 : Path *spath = (Path *) lfirst(lc);
3939 :
3940 0 : spath = reparameterize_path(root, spath,
3941 : required_outer,
3942 : loop_count);
3943 0 : if (spath == NULL)
3944 0 : return NULL;
3945 : /* We have to re-split the regular and partial paths */
3946 0 : if (i < apath->first_partial_path)
3947 0 : childpaths = lappend(childpaths, spath);
3948 : else
3949 0 : partialpaths = lappend(partialpaths, spath);
3950 0 : i++;
3951 : }
3952 0 : return (Path *)
3953 0 : create_append_path(root, rel, childpaths, partialpaths,
3954 : apath->path.pathkeys, required_outer,
3955 : apath->path.parallel_workers,
3956 0 : apath->path.parallel_aware,
3957 : -1);
3958 : }
3959 0 : case T_Material:
3960 : {
3961 0 : MaterialPath *mpath = (MaterialPath *) path;
3962 0 : Path *spath = mpath->subpath;
3963 :
3964 0 : spath = reparameterize_path(root, spath,
3965 : required_outer,
3966 : loop_count);
3967 0 : if (spath == NULL)
3968 0 : return NULL;
3969 0 : return (Path *) create_material_path(rel, spath);
3970 : }
3971 0 : case T_Memoize:
3972 : {
3973 0 : MemoizePath *mpath = (MemoizePath *) path;
3974 0 : Path *spath = mpath->subpath;
3975 :
3976 0 : spath = reparameterize_path(root, spath,
3977 : required_outer,
3978 : loop_count);
3979 0 : if (spath == NULL)
3980 0 : return NULL;
3981 0 : return (Path *) create_memoize_path(root, rel,
3982 : spath,
3983 : mpath->param_exprs,
3984 : mpath->hash_operators,
3985 0 : mpath->singlerow,
3986 0 : mpath->binary_mode,
3987 : mpath->est_calls);
3988 : }
3989 32 : default:
3990 32 : break;
3991 : }
3992 32 : return NULL;
3993 : }
3994 :
3995 : /*
3996 : * reparameterize_path_by_child
3997 : * Given a path parameterized by the parent of the given child relation,
3998 : * translate the path to be parameterized by the given child relation.
3999 : *
4000 : * Most fields in the path are not changed, but any expressions must be
4001 : * adjusted to refer to the correct varnos, and any subpaths must be
4002 : * recursively reparameterized. Other fields that refer to specific relids
4003 : * also need adjustment.
4004 : *
4005 : * The cost, number of rows, width and parallel path properties depend upon
4006 : * path->parent, which does not change during the translation. So we need
4007 : * not change those.
4008 : *
4009 : * Currently, only a few path types are supported here, though more could be
4010 : * added at need. We return NULL if we can't reparameterize the given path.
4011 : *
4012 : * Note that this function can change referenced RangeTblEntries, RelOptInfos
4013 : * and IndexOptInfos as well as the Path structures. Therefore, it's only safe
4014 : * to call during create_plan(), when we have made a final choice of which Path
4015 : * to use for each RangeTblEntry/RelOptInfo/IndexOptInfo.
4016 : *
4017 : * Keep this code in sync with path_is_reparameterizable_by_child()!
4018 : */
4019 : Path *
4020 98380 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
4021 : RelOptInfo *child_rel)
4022 : {
4023 : Path *new_path;
4024 : ParamPathInfo *new_ppi;
4025 : ParamPathInfo *old_ppi;
4026 : Relids required_outer;
4027 :
4028 : #define ADJUST_CHILD_ATTRS(node) \
4029 : ((node) = (void *) adjust_appendrel_attrs_multilevel(root, \
4030 : (Node *) (node), \
4031 : child_rel, \
4032 : child_rel->top_parent))
4033 :
4034 : #define REPARAMETERIZE_CHILD_PATH(path) \
4035 : do { \
4036 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
4037 : if ((path) == NULL) \
4038 : return NULL; \
4039 : } while(0)
4040 :
4041 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
4042 : do { \
4043 : if ((pathlist) != NIL) \
4044 : { \
4045 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
4046 : child_rel); \
4047 : if ((pathlist) == NIL) \
4048 : return NULL; \
4049 : } \
4050 : } while(0)
4051 :
4052 : /*
4053 : * If the path is not parameterized by the parent of the given relation,
4054 : * it doesn't need reparameterization.
4055 : */
4056 98380 : if (!path->param_info ||
4057 49666 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4058 97354 : return path;
4059 :
4060 : /*
4061 : * If possible, reparameterize the given path.
4062 : *
4063 : * This function is currently only applied to the inner side of a nestloop
4064 : * join that is being partitioned by the partitionwise-join code. Hence,
4065 : * we need only support path types that plausibly arise in that context.
4066 : * (In particular, supporting sorted path types would be a waste of code
4067 : * and cycles: even if we translated them here, they'd just lose in
4068 : * subsequent cost comparisons.) If we do see an unsupported path type,
4069 : * that just means we won't be able to generate a partitionwise-join plan
4070 : * using that path type.
4071 : */
4072 1026 : switch (nodeTag(path))
4073 : {
4074 228 : case T_Path:
4075 228 : new_path = path;
4076 228 : ADJUST_CHILD_ATTRS(new_path->parent->baserestrictinfo);
4077 228 : if (path->pathtype == T_SampleScan)
4078 : {
4079 48 : Index scan_relid = path->parent->relid;
4080 : RangeTblEntry *rte;
4081 :
4082 : /* it should be a base rel with a tablesample clause... */
4083 : Assert(scan_relid > 0);
4084 48 : rte = planner_rt_fetch(scan_relid, root);
4085 : Assert(rte->rtekind == RTE_RELATION);
4086 : Assert(rte->tablesample != NULL);
4087 :
4088 48 : ADJUST_CHILD_ATTRS(rte->tablesample);
4089 : }
4090 228 : break;
4091 :
4092 528 : case T_IndexPath:
4093 : {
4094 528 : IndexPath *ipath = (IndexPath *) path;
4095 :
4096 528 : ADJUST_CHILD_ATTRS(ipath->indexinfo->indrestrictinfo);
4097 528 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
4098 528 : new_path = (Path *) ipath;
4099 : }
4100 528 : break;
4101 :
4102 48 : case T_BitmapHeapPath:
4103 : {
4104 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4105 :
4106 48 : ADJUST_CHILD_ATTRS(bhpath->path.parent->baserestrictinfo);
4107 48 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
4108 48 : new_path = (Path *) bhpath;
4109 : }
4110 48 : break;
4111 :
4112 24 : case T_BitmapAndPath:
4113 : {
4114 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4115 :
4116 24 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
4117 24 : new_path = (Path *) bapath;
4118 : }
4119 24 : break;
4120 :
4121 24 : case T_BitmapOrPath:
4122 : {
4123 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4124 :
4125 24 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
4126 24 : new_path = (Path *) bopath;
4127 : }
4128 24 : break;
4129 :
4130 0 : case T_ForeignPath:
4131 : {
4132 0 : ForeignPath *fpath = (ForeignPath *) path;
4133 : ReparameterizeForeignPathByChild_function rfpc_func;
4134 :
4135 0 : ADJUST_CHILD_ATTRS(fpath->path.parent->baserestrictinfo);
4136 0 : if (fpath->fdw_outerpath)
4137 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
4138 0 : if (fpath->fdw_restrictinfo)
4139 0 : ADJUST_CHILD_ATTRS(fpath->fdw_restrictinfo);
4140 :
4141 : /* Hand over to FDW if needed. */
4142 0 : rfpc_func =
4143 0 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
4144 0 : if (rfpc_func)
4145 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
4146 : child_rel);
4147 0 : new_path = (Path *) fpath;
4148 : }
4149 0 : break;
4150 :
4151 0 : case T_CustomPath:
4152 : {
4153 0 : CustomPath *cpath = (CustomPath *) path;
4154 :
4155 0 : ADJUST_CHILD_ATTRS(cpath->path.parent->baserestrictinfo);
4156 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
4157 0 : if (cpath->custom_restrictinfo)
4158 0 : ADJUST_CHILD_ATTRS(cpath->custom_restrictinfo);
4159 0 : if (cpath->methods &&
4160 0 : cpath->methods->ReparameterizeCustomPathByChild)
4161 0 : cpath->custom_private =
4162 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4163 : cpath->custom_private,
4164 : child_rel);
4165 0 : new_path = (Path *) cpath;
4166 : }
4167 0 : break;
4168 :
4169 36 : case T_NestPath:
4170 : {
4171 36 : NestPath *npath = (NestPath *) path;
4172 36 : JoinPath *jpath = (JoinPath *) npath;
4173 :
4174 36 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4175 36 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4176 36 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4177 36 : new_path = (Path *) npath;
4178 : }
4179 36 : break;
4180 :
4181 0 : case T_MergePath:
4182 : {
4183 0 : MergePath *mpath = (MergePath *) path;
4184 0 : JoinPath *jpath = (JoinPath *) mpath;
4185 :
4186 0 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4187 0 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4188 0 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4189 0 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4190 0 : new_path = (Path *) mpath;
4191 : }
4192 0 : break;
4193 :
4194 48 : case T_HashPath:
4195 : {
4196 48 : HashPath *hpath = (HashPath *) path;
4197 48 : JoinPath *jpath = (JoinPath *) hpath;
4198 :
4199 48 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4200 48 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4201 48 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4202 48 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4203 48 : new_path = (Path *) hpath;
4204 : }
4205 48 : break;
4206 :
4207 24 : case T_AppendPath:
4208 : {
4209 24 : AppendPath *apath = (AppendPath *) path;
4210 :
4211 24 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4212 24 : new_path = (Path *) apath;
4213 : }
4214 24 : break;
4215 :
4216 0 : case T_MaterialPath:
4217 : {
4218 0 : MaterialPath *mpath = (MaterialPath *) path;
4219 :
4220 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4221 0 : new_path = (Path *) mpath;
4222 : }
4223 0 : break;
4224 :
4225 66 : case T_MemoizePath:
4226 : {
4227 66 : MemoizePath *mpath = (MemoizePath *) path;
4228 :
4229 66 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4230 66 : ADJUST_CHILD_ATTRS(mpath->param_exprs);
4231 66 : new_path = (Path *) mpath;
4232 : }
4233 66 : break;
4234 :
4235 0 : case T_GatherPath:
4236 : {
4237 0 : GatherPath *gpath = (GatherPath *) path;
4238 :
4239 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4240 0 : new_path = (Path *) gpath;
4241 : }
4242 0 : break;
4243 :
4244 0 : default:
4245 : /* We don't know how to reparameterize this path. */
4246 0 : return NULL;
4247 : }
4248 :
4249 : /*
4250 : * Adjust the parameterization information, which refers to the topmost
4251 : * parent. The topmost parent can be multiple levels away from the given
4252 : * child, hence use multi-level expression adjustment routines.
4253 : */
4254 1026 : old_ppi = new_path->param_info;
4255 : required_outer =
4256 1026 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4257 : child_rel,
4258 1026 : child_rel->top_parent);
4259 :
4260 : /* If we already have a PPI for this parameterization, just return it */
4261 1026 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4262 :
4263 : /*
4264 : * If not, build a new one and link it to the list of PPIs. For the same
4265 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4266 : * context the given RelOptInfo is in.
4267 : */
4268 1026 : if (new_ppi == NULL)
4269 : {
4270 : MemoryContext oldcontext;
4271 864 : RelOptInfo *rel = path->parent;
4272 :
4273 864 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4274 :
4275 864 : new_ppi = makeNode(ParamPathInfo);
4276 864 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4277 864 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4278 864 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4279 864 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4280 864 : new_ppi->ppi_serials = bms_copy(old_ppi->ppi_serials);
4281 864 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4282 :
4283 864 : MemoryContextSwitchTo(oldcontext);
4284 : }
4285 1026 : bms_free(required_outer);
4286 :
4287 1026 : new_path->param_info = new_ppi;
4288 :
4289 : /*
4290 : * Adjust the path target if the parent of the outer relation is
4291 : * referenced in the targetlist. This can happen when only the parent of
4292 : * outer relation is laterally referenced in this relation.
4293 : */
4294 1026 : if (bms_overlap(path->parent->lateral_relids,
4295 1026 : child_rel->top_parent_relids))
4296 : {
4297 480 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4298 480 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4299 : }
4300 :
4301 1026 : return new_path;
4302 : }
4303 :
4304 : /*
4305 : * path_is_reparameterizable_by_child
4306 : * Given a path parameterized by the parent of the given child relation,
4307 : * see if it can be translated to be parameterized by the child relation.
4308 : *
4309 : * This must return true if and only if reparameterize_path_by_child()
4310 : * would succeed on this path. Currently it's sufficient to verify that
4311 : * the path and all of its subpaths (if any) are of the types handled by
4312 : * that function. However, subpaths that are not parameterized can be
4313 : * disregarded since they won't require translation.
4314 : */
4315 : bool
4316 37224 : path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
4317 : {
4318 : #define REJECT_IF_PATH_NOT_REPARAMETERIZABLE(path) \
4319 : do { \
4320 : if (!path_is_reparameterizable_by_child(path, child_rel)) \
4321 : return false; \
4322 : } while(0)
4323 :
4324 : #define REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(pathlist) \
4325 : do { \
4326 : if (!pathlist_is_reparameterizable_by_child(pathlist, child_rel)) \
4327 : return false; \
4328 : } while(0)
4329 :
4330 : /*
4331 : * If the path is not parameterized by the parent of the given relation,
4332 : * it doesn't need reparameterization.
4333 : */
4334 37224 : if (!path->param_info ||
4335 36816 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4336 984 : return true;
4337 :
4338 : /*
4339 : * Check that the path type is one that reparameterize_path_by_child() can
4340 : * handle, and recursively check subpaths.
4341 : */
4342 36240 : switch (nodeTag(path))
4343 : {
4344 24312 : case T_Path:
4345 : case T_IndexPath:
4346 24312 : break;
4347 :
4348 48 : case T_BitmapHeapPath:
4349 : {
4350 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4351 :
4352 48 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(bhpath->bitmapqual);
4353 : }
4354 48 : break;
4355 :
4356 24 : case T_BitmapAndPath:
4357 : {
4358 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4359 :
4360 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bapath->bitmapquals);
4361 : }
4362 24 : break;
4363 :
4364 24 : case T_BitmapOrPath:
4365 : {
4366 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4367 :
4368 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bopath->bitmapquals);
4369 : }
4370 24 : break;
4371 :
4372 148 : case T_ForeignPath:
4373 : {
4374 148 : ForeignPath *fpath = (ForeignPath *) path;
4375 :
4376 148 : if (fpath->fdw_outerpath)
4377 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(fpath->fdw_outerpath);
4378 : }
4379 148 : break;
4380 :
4381 0 : case T_CustomPath:
4382 : {
4383 0 : CustomPath *cpath = (CustomPath *) path;
4384 :
4385 0 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(cpath->custom_paths);
4386 : }
4387 0 : break;
4388 :
4389 1248 : case T_NestPath:
4390 : case T_MergePath:
4391 : case T_HashPath:
4392 : {
4393 1248 : JoinPath *jpath = (JoinPath *) path;
4394 :
4395 1248 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->outerjoinpath);
4396 1248 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->innerjoinpath);
4397 : }
4398 1248 : break;
4399 :
4400 192 : case T_AppendPath:
4401 : {
4402 192 : AppendPath *apath = (AppendPath *) path;
4403 :
4404 192 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(apath->subpaths);
4405 : }
4406 192 : break;
4407 :
4408 0 : case T_MaterialPath:
4409 : {
4410 0 : MaterialPath *mpath = (MaterialPath *) path;
4411 :
4412 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4413 : }
4414 0 : break;
4415 :
4416 10244 : case T_MemoizePath:
4417 : {
4418 10244 : MemoizePath *mpath = (MemoizePath *) path;
4419 :
4420 10244 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4421 : }
4422 10244 : break;
4423 :
4424 0 : case T_GatherPath:
4425 : {
4426 0 : GatherPath *gpath = (GatherPath *) path;
4427 :
4428 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(gpath->subpath);
4429 : }
4430 0 : break;
4431 :
4432 0 : default:
4433 : /* We don't know how to reparameterize this path. */
4434 0 : return false;
4435 : }
4436 :
4437 36240 : return true;
4438 : }
4439 :
4440 : /*
4441 : * reparameterize_pathlist_by_child
4442 : * Helper function to reparameterize a list of paths by given child rel.
4443 : *
4444 : * Returns NIL to indicate failure, so pathlist had better not be NIL.
4445 : */
4446 : static List *
4447 72 : reparameterize_pathlist_by_child(PlannerInfo *root,
4448 : List *pathlist,
4449 : RelOptInfo *child_rel)
4450 : {
4451 : ListCell *lc;
4452 72 : List *result = NIL;
4453 :
4454 216 : foreach(lc, pathlist)
4455 : {
4456 144 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4457 : child_rel);
4458 :
4459 144 : if (path == NULL)
4460 : {
4461 0 : list_free(result);
4462 0 : return NIL;
4463 : }
4464 :
4465 144 : result = lappend(result, path);
4466 : }
4467 :
4468 72 : return result;
4469 : }
4470 :
4471 : /*
4472 : * pathlist_is_reparameterizable_by_child
4473 : * Helper function to check if a list of paths can be reparameterized.
4474 : */
4475 : static bool
4476 240 : pathlist_is_reparameterizable_by_child(List *pathlist, RelOptInfo *child_rel)
4477 : {
4478 : ListCell *lc;
4479 :
4480 720 : foreach(lc, pathlist)
4481 : {
4482 480 : Path *path = (Path *) lfirst(lc);
4483 :
4484 480 : if (!path_is_reparameterizable_by_child(path, child_rel))
4485 0 : return false;
4486 : }
4487 :
4488 240 : return true;
4489 : }
|
