Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * allpaths.c
4 : * Routines to find possible search paths for processing a query
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/optimizer/path/allpaths.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include <limits.h>
19 : #include <math.h>
20 :
21 : #include "access/sysattr.h"
22 : #include "access/tsmapi.h"
23 : #include "catalog/pg_class.h"
24 : #include "catalog/pg_operator.h"
25 : #include "catalog/pg_proc.h"
26 : #include "foreign/fdwapi.h"
27 : #include "miscadmin.h"
28 : #include "nodes/makefuncs.h"
29 : #include "nodes/nodeFuncs.h"
30 : #include "nodes/supportnodes.h"
31 : #ifdef OPTIMIZER_DEBUG
32 : #include "nodes/print.h"
33 : #endif
34 : #include "optimizer/appendinfo.h"
35 : #include "optimizer/clauses.h"
36 : #include "optimizer/cost.h"
37 : #include "optimizer/geqo.h"
38 : #include "optimizer/optimizer.h"
39 : #include "optimizer/pathnode.h"
40 : #include "optimizer/paths.h"
41 : #include "optimizer/plancat.h"
42 : #include "optimizer/planner.h"
43 : #include "optimizer/prep.h"
44 : #include "optimizer/tlist.h"
45 : #include "parser/parse_clause.h"
46 : #include "parser/parsetree.h"
47 : #include "partitioning/partbounds.h"
48 : #include "port/pg_bitutils.h"
49 : #include "rewrite/rewriteManip.h"
50 : #include "utils/lsyscache.h"
51 : #include "utils/selfuncs.h"
52 :
53 :
54 : /* Bitmask flags for pushdown_safety_info.unsafeFlags */
55 : #define UNSAFE_HAS_VOLATILE_FUNC (1 << 0)
56 : #define UNSAFE_HAS_SET_FUNC (1 << 1)
57 : #define UNSAFE_NOTIN_DISTINCTON_CLAUSE (1 << 2)
58 : #define UNSAFE_NOTIN_PARTITIONBY_CLAUSE (1 << 3)
59 : #define UNSAFE_TYPE_MISMATCH (1 << 4)
60 :
61 : /* results of subquery_is_pushdown_safe */
62 : typedef struct pushdown_safety_info
63 : {
64 : unsigned char *unsafeFlags; /* bitmask of reasons why this target list
65 : * column is unsafe for qual pushdown, or 0 if
66 : * no reason. */
67 : bool unsafeVolatile; /* don't push down volatile quals */
68 : bool unsafeLeaky; /* don't push down leaky quals */
69 : } pushdown_safety_info;
70 :
71 : /* Return type for qual_is_pushdown_safe */
72 : typedef enum pushdown_safe_type
73 : {
74 : PUSHDOWN_UNSAFE, /* unsafe to push qual into subquery */
75 : PUSHDOWN_SAFE, /* safe to push qual into subquery */
76 : PUSHDOWN_WINDOWCLAUSE_RUNCOND, /* unsafe, but may work as WindowClause
77 : * run condition */
78 : } pushdown_safe_type;
79 :
80 : /* These parameters are set by GUC */
81 : bool enable_geqo = false; /* just in case GUC doesn't set it */
82 : bool enable_eager_aggregate = true;
83 : int geqo_threshold;
84 : double min_eager_agg_group_size;
85 : int min_parallel_table_scan_size;
86 : int min_parallel_index_scan_size;
87 :
88 : /* Hook for plugins to get control in set_rel_pathlist() */
89 : set_rel_pathlist_hook_type set_rel_pathlist_hook = NULL;
90 :
91 : /* Hook for plugins to replace standard_join_search() */
92 : join_search_hook_type join_search_hook = NULL;
93 :
94 :
95 : static void set_base_rel_consider_startup(PlannerInfo *root);
96 : static void set_base_rel_sizes(PlannerInfo *root);
97 : static void setup_simple_grouped_rels(PlannerInfo *root);
98 : static void set_base_rel_pathlists(PlannerInfo *root);
99 : static void set_rel_size(PlannerInfo *root, RelOptInfo *rel,
100 : Index rti, RangeTblEntry *rte);
101 : static void set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
102 : Index rti, RangeTblEntry *rte);
103 : static void set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel,
104 : RangeTblEntry *rte);
105 : static void create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel);
106 : static void set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel,
107 : RangeTblEntry *rte);
108 : static void set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
109 : RangeTblEntry *rte);
110 : static void set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel,
111 : RangeTblEntry *rte);
112 : static void set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
113 : RangeTblEntry *rte);
114 : static void set_foreign_size(PlannerInfo *root, RelOptInfo *rel,
115 : RangeTblEntry *rte);
116 : static void set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel,
117 : RangeTblEntry *rte);
118 : static void set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
119 : Index rti, RangeTblEntry *rte);
120 : static void set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
121 : Index rti, RangeTblEntry *rte);
122 : static void set_grouped_rel_pathlist(PlannerInfo *root, RelOptInfo *rel);
123 : static void generate_orderedappend_paths(PlannerInfo *root, RelOptInfo *rel,
124 : List *live_childrels,
125 : List *all_child_pathkeys);
126 : static Path *get_cheapest_parameterized_child_path(PlannerInfo *root,
127 : RelOptInfo *rel,
128 : Relids required_outer);
129 : static void accumulate_append_subpath(Path *path,
130 : List **subpaths,
131 : List **special_subpaths,
132 : List **child_append_relid_sets);
133 : static Path *get_singleton_append_subpath(Path *path,
134 : List **child_append_relid_sets);
135 : static void set_dummy_rel_pathlist(RelOptInfo *rel);
136 : static void set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
137 : Index rti, RangeTblEntry *rte);
138 : static void set_function_pathlist(PlannerInfo *root, RelOptInfo *rel,
139 : RangeTblEntry *rte);
140 : static void set_values_pathlist(PlannerInfo *root, RelOptInfo *rel,
141 : RangeTblEntry *rte);
142 : static void set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel,
143 : RangeTblEntry *rte);
144 : static void set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel,
145 : RangeTblEntry *rte);
146 : static void set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel,
147 : RangeTblEntry *rte);
148 : static void set_result_pathlist(PlannerInfo *root, RelOptInfo *rel,
149 : RangeTblEntry *rte);
150 : static void set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel,
151 : RangeTblEntry *rte);
152 : static RelOptInfo *make_rel_from_joinlist(PlannerInfo *root, List *joinlist);
153 : static bool subquery_is_pushdown_safe(Query *subquery, Query *topquery,
154 : pushdown_safety_info *safetyInfo);
155 : static bool recurse_pushdown_safe(Node *setOp, Query *topquery,
156 : pushdown_safety_info *safetyInfo);
157 : static void check_output_expressions(Query *subquery,
158 : pushdown_safety_info *safetyInfo);
159 : static void compare_tlist_datatypes(List *tlist, List *colTypes,
160 : pushdown_safety_info *safetyInfo);
161 : static bool targetIsInAllPartitionLists(TargetEntry *tle, Query *query);
162 : static pushdown_safe_type qual_is_pushdown_safe(Query *subquery, Index rti,
163 : RestrictInfo *rinfo,
164 : pushdown_safety_info *safetyInfo);
165 : static void subquery_push_qual(Query *subquery,
166 : RangeTblEntry *rte, Index rti, Node *qual);
167 : static void recurse_push_qual(Node *setOp, Query *topquery,
168 : RangeTblEntry *rte, Index rti, Node *qual);
169 : static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel,
170 : Bitmapset *extra_used_attrs);
171 :
172 :
173 : /*
174 : * make_one_rel
175 : * Finds all possible access paths for executing a query, returning a
176 : * single rel that represents the join of all base rels in the query.
177 : */
178 : RelOptInfo *
179 253299 : make_one_rel(PlannerInfo *root, List *joinlist)
180 : {
181 : RelOptInfo *rel;
182 : Index rti;
183 : double total_pages;
184 :
185 : /* Mark base rels as to whether we care about fast-start plans */
186 253299 : set_base_rel_consider_startup(root);
187 :
188 : /*
189 : * Compute size estimates and consider_parallel flags for each base rel.
190 : */
191 253299 : set_base_rel_sizes(root);
192 :
193 : /*
194 : * Build grouped relations for simple rels (i.e., base or "other" member
195 : * relations) where possible.
196 : */
197 253277 : setup_simple_grouped_rels(root);
198 :
199 : /*
200 : * We should now have size estimates for every actual table involved in
201 : * the query, and we also know which if any have been deleted from the
202 : * query by join removal, pruned by partition pruning, or eliminated by
203 : * constraint exclusion. So we can now compute total_table_pages.
204 : *
205 : * Note that appendrels are not double-counted here, even though we don't
206 : * bother to distinguish RelOptInfos for appendrel parents, because the
207 : * parents will have pages = 0.
208 : *
209 : * XXX if a table is self-joined, we will count it once per appearance,
210 : * which perhaps is the wrong thing ... but that's not completely clear,
211 : * and detecting self-joins here is difficult, so ignore it for now.
212 : */
213 253277 : total_pages = 0;
214 794238 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
215 : {
216 540961 : RelOptInfo *brel = root->simple_rel_array[rti];
217 :
218 : /* there may be empty slots corresponding to non-baserel RTEs */
219 540961 : if (brel == NULL)
220 127684 : continue;
221 :
222 : Assert(brel->relid == rti); /* sanity check on array */
223 :
224 413277 : if (IS_DUMMY_REL(brel))
225 1261 : continue;
226 :
227 412016 : if (IS_SIMPLE_REL(brel))
228 412016 : total_pages += (double) brel->pages;
229 : }
230 253277 : root->total_table_pages = total_pages;
231 :
232 : /*
233 : * Generate access paths for each base rel.
234 : */
235 253277 : set_base_rel_pathlists(root);
236 :
237 : /*
238 : * Generate access paths for the entire join tree.
239 : */
240 253277 : rel = make_rel_from_joinlist(root, joinlist);
241 :
242 : /*
243 : * The result should join all and only the query's base + outer-join rels.
244 : */
245 : Assert(bms_equal(rel->relids, root->all_query_rels));
246 :
247 253277 : return rel;
248 : }
249 :
250 : /*
251 : * set_base_rel_consider_startup
252 : * Set the consider_[param_]startup flags for each base-relation entry.
253 : *
254 : * For the moment, we only deal with consider_param_startup here; because the
255 : * logic for consider_startup is pretty trivial and is the same for every base
256 : * relation, we just let build_simple_rel() initialize that flag correctly to
257 : * start with. If that logic ever gets more complicated it would probably
258 : * be better to move it here.
259 : */
260 : static void
261 253299 : set_base_rel_consider_startup(PlannerInfo *root)
262 : {
263 : /*
264 : * Since parameterized paths can only be used on the inside of a nestloop
265 : * join plan, there is usually little value in considering fast-start
266 : * plans for them. However, for relations that are on the RHS of a SEMI
267 : * or ANTI join, a fast-start plan can be useful because we're only going
268 : * to care about fetching one tuple anyway.
269 : *
270 : * To minimize growth of planning time, we currently restrict this to
271 : * cases where the RHS is a single base relation, not a join; there is no
272 : * provision for consider_param_startup to get set at all on joinrels.
273 : * Also we don't worry about appendrels. costsize.c's costing rules for
274 : * nestloop semi/antijoins don't consider such cases either.
275 : */
276 : ListCell *lc;
277 :
278 288035 : foreach(lc, root->join_info_list)
279 : {
280 34736 : SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc);
281 : int varno;
282 :
283 44348 : if ((sjinfo->jointype == JOIN_SEMI || sjinfo->jointype == JOIN_ANTI) &&
284 9612 : bms_get_singleton_member(sjinfo->syn_righthand, &varno))
285 : {
286 9433 : RelOptInfo *rel = find_base_rel(root, varno);
287 :
288 9433 : rel->consider_param_startup = true;
289 : }
290 : }
291 253299 : }
292 :
293 : /*
294 : * set_base_rel_sizes
295 : * Set the size estimates (rows and widths) for each base-relation entry.
296 : * Also determine whether to consider parallel paths for base relations.
297 : *
298 : * We do this in a separate pass over the base rels so that rowcount
299 : * estimates are available for parameterized path generation, and also so
300 : * that each rel's consider_parallel flag is set correctly before we begin to
301 : * generate paths.
302 : */
303 : static void
304 253299 : set_base_rel_sizes(PlannerInfo *root)
305 : {
306 : Index rti;
307 :
308 794261 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
309 : {
310 540984 : RelOptInfo *rel = root->simple_rel_array[rti];
311 : RangeTblEntry *rte;
312 :
313 : /* there may be empty slots corresponding to non-baserel RTEs */
314 540984 : if (rel == NULL)
315 127685 : continue;
316 :
317 : Assert(rel->relid == rti); /* sanity check on array */
318 :
319 : /* ignore RTEs that are "other rels" */
320 413299 : if (rel->reloptkind != RELOPT_BASEREL)
321 48600 : continue;
322 :
323 364699 : rte = root->simple_rte_array[rti];
324 :
325 : /*
326 : * If parallelism is allowable for this query in general, see whether
327 : * it's allowable for this rel in particular. We have to do this
328 : * before set_rel_size(), because (a) if this rel is an inheritance
329 : * parent, set_append_rel_size() will use and perhaps change the rel's
330 : * consider_parallel flag, and (b) for some RTE types, set_rel_size()
331 : * goes ahead and makes paths immediately.
332 : */
333 364699 : if (root->glob->parallelModeOK)
334 292890 : set_rel_consider_parallel(root, rel, rte);
335 :
336 364699 : set_rel_size(root, rel, rti, rte);
337 : }
338 253277 : }
339 :
340 : /*
341 : * setup_simple_grouped_rels
342 : * For each simple relation, build a grouped simple relation if eager
343 : * aggregation is possible and if this relation can produce grouped paths.
344 : */
345 : static void
346 253277 : setup_simple_grouped_rels(PlannerInfo *root)
347 : {
348 : Index rti;
349 :
350 : /*
351 : * If there are no aggregate expressions or grouping expressions, eager
352 : * aggregation is not possible.
353 : */
354 253277 : if (root->agg_clause_list == NIL ||
355 551 : root->group_expr_list == NIL)
356 252837 : return;
357 :
358 3761 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
359 : {
360 3321 : RelOptInfo *rel = root->simple_rel_array[rti];
361 :
362 : /* there may be empty slots corresponding to non-baserel RTEs */
363 3321 : if (rel == NULL)
364 1071 : continue;
365 :
366 : Assert(rel->relid == rti); /* sanity check on array */
367 : Assert(IS_SIMPLE_REL(rel)); /* sanity check on rel */
368 :
369 2250 : (void) build_simple_grouped_rel(root, rel);
370 : }
371 : }
372 :
373 : /*
374 : * set_base_rel_pathlists
375 : * Finds all paths available for scanning each base-relation entry.
376 : * Sequential scan and any available indices are considered.
377 : * Each useful path is attached to its relation's 'pathlist' field.
378 : */
379 : static void
380 253277 : set_base_rel_pathlists(PlannerInfo *root)
381 : {
382 : Index rti;
383 :
384 794238 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
385 : {
386 540961 : RelOptInfo *rel = root->simple_rel_array[rti];
387 :
388 : /* there may be empty slots corresponding to non-baserel RTEs */
389 540961 : if (rel == NULL)
390 127684 : continue;
391 :
392 : Assert(rel->relid == rti); /* sanity check on array */
393 :
394 : /* ignore RTEs that are "other rels" */
395 413277 : if (rel->reloptkind != RELOPT_BASEREL)
396 48600 : continue;
397 :
398 364677 : set_rel_pathlist(root, rel, rti, root->simple_rte_array[rti]);
399 : }
400 253277 : }
401 :
402 : /*
403 : * set_rel_size
404 : * Set size estimates for a base relation
405 : */
406 : static void
407 413027 : set_rel_size(PlannerInfo *root, RelOptInfo *rel,
408 : Index rti, RangeTblEntry *rte)
409 : {
410 777726 : if (rel->reloptkind == RELOPT_BASEREL &&
411 364699 : relation_excluded_by_constraints(root, rel, rte))
412 : {
413 : /*
414 : * We proved we don't need to scan the rel via constraint exclusion,
415 : * so set up a single dummy path for it. Here we only check this for
416 : * regular baserels; if it's an otherrel, CE was already checked in
417 : * set_append_rel_size().
418 : *
419 : * In this case, we go ahead and set up the relation's path right away
420 : * instead of leaving it for set_rel_pathlist to do. This is because
421 : * we don't have a convention for marking a rel as dummy except by
422 : * assigning a dummy path to it.
423 : */
424 631 : set_dummy_rel_pathlist(rel);
425 : }
426 412396 : else if (rte->inh)
427 : {
428 : /* It's an "append relation", process accordingly */
429 20787 : set_append_rel_size(root, rel, rti, rte);
430 : }
431 : else
432 : {
433 391609 : switch (rel->rtekind)
434 : {
435 323647 : case RTE_RELATION:
436 323647 : if (rte->relkind == RELKIND_FOREIGN_TABLE)
437 : {
438 : /* Foreign table */
439 1252 : set_foreign_size(root, rel, rte);
440 : }
441 322395 : else if (rte->relkind == RELKIND_PARTITIONED_TABLE)
442 : {
443 : /*
444 : * We could get here if asked to scan a partitioned table
445 : * with ONLY. In that case we shouldn't scan any of the
446 : * partitions, so mark it as a dummy rel.
447 : */
448 26 : set_dummy_rel_pathlist(rel);
449 : }
450 322369 : else if (rte->tablesample != NULL)
451 : {
452 : /* Sampled relation */
453 243 : set_tablesample_rel_size(root, rel, rte);
454 : }
455 : else
456 : {
457 : /* Plain relation */
458 322126 : set_plain_rel_size(root, rel, rte);
459 : }
460 323625 : break;
461 17357 : case RTE_SUBQUERY:
462 :
463 : /*
464 : * Subqueries don't support making a choice between
465 : * parameterized and unparameterized paths, so just go ahead
466 : * and build their paths immediately.
467 : */
468 17357 : set_subquery_pathlist(root, rel, rti, rte);
469 17357 : break;
470 35645 : case RTE_FUNCTION:
471 35645 : set_function_size_estimates(root, rel);
472 35645 : break;
473 519 : case RTE_TABLEFUNC:
474 519 : set_tablefunc_size_estimates(root, rel);
475 519 : break;
476 6793 : case RTE_VALUES:
477 6793 : set_values_size_estimates(root, rel);
478 6793 : break;
479 3620 : case RTE_CTE:
480 :
481 : /*
482 : * CTEs don't support making a choice between parameterized
483 : * and unparameterized paths, so just go ahead and build their
484 : * paths immediately.
485 : */
486 3620 : if (rte->self_reference)
487 692 : set_worktable_pathlist(root, rel, rte);
488 : else
489 2928 : set_cte_pathlist(root, rel, rte);
490 3620 : break;
491 433 : case RTE_NAMEDTUPLESTORE:
492 : /* Might as well just build the path immediately */
493 433 : set_namedtuplestore_pathlist(root, rel, rte);
494 433 : break;
495 3595 : case RTE_RESULT:
496 : /* Might as well just build the path immediately */
497 3595 : set_result_pathlist(root, rel, rte);
498 3595 : break;
499 0 : default:
500 0 : elog(ERROR, "unexpected rtekind: %d", (int) rel->rtekind);
501 : break;
502 : }
503 : }
504 :
505 : /*
506 : * We insist that all non-dummy rels have a nonzero rowcount estimate.
507 : */
508 : Assert(rel->rows > 0 || IS_DUMMY_REL(rel));
509 413004 : }
510 :
511 : /*
512 : * set_rel_pathlist
513 : * Build access paths for a base relation
514 : */
515 : static void
516 413129 : set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
517 : Index rti, RangeTblEntry *rte)
518 : {
519 413129 : if (IS_DUMMY_REL(rel))
520 : {
521 : /* We already proved the relation empty, so nothing more to do */
522 : }
523 411986 : else if (rte->inh)
524 : {
525 : /* It's an "append relation", process accordingly */
526 20530 : set_append_rel_pathlist(root, rel, rti, rte);
527 : }
528 : else
529 : {
530 391456 : switch (rel->rtekind)
531 : {
532 323599 : case RTE_RELATION:
533 323599 : if (rte->relkind == RELKIND_FOREIGN_TABLE)
534 : {
535 : /* Foreign table */
536 1250 : set_foreign_pathlist(root, rel, rte);
537 : }
538 322349 : else if (rte->tablesample != NULL)
539 : {
540 : /* Sampled relation */
541 243 : set_tablesample_rel_pathlist(root, rel, rte);
542 : }
543 : else
544 : {
545 : /* Plain relation */
546 322106 : set_plain_rel_pathlist(root, rel, rte);
547 : }
548 323599 : break;
549 17252 : case RTE_SUBQUERY:
550 : /* Subquery --- fully handled during set_rel_size */
551 17252 : break;
552 35645 : case RTE_FUNCTION:
553 : /* RangeFunction */
554 35645 : set_function_pathlist(root, rel, rte);
555 35645 : break;
556 519 : case RTE_TABLEFUNC:
557 : /* Table Function */
558 519 : set_tablefunc_pathlist(root, rel, rte);
559 519 : break;
560 6793 : case RTE_VALUES:
561 : /* Values list */
562 6793 : set_values_pathlist(root, rel, rte);
563 6793 : break;
564 3620 : case RTE_CTE:
565 : /* CTE reference --- fully handled during set_rel_size */
566 3620 : break;
567 433 : case RTE_NAMEDTUPLESTORE:
568 : /* tuplestore reference --- fully handled during set_rel_size */
569 433 : break;
570 3595 : case RTE_RESULT:
571 : /* simple Result --- fully handled during set_rel_size */
572 3595 : break;
573 0 : default:
574 0 : elog(ERROR, "unexpected rtekind: %d", (int) rel->rtekind);
575 : break;
576 : }
577 : }
578 :
579 : /*
580 : * Allow a plugin to editorialize on the set of Paths for this base
581 : * relation. It could add new paths (such as CustomPaths) by calling
582 : * add_path(), or add_partial_path() if parallel aware. It could also
583 : * delete or modify paths added by the core code.
584 : */
585 413129 : if (set_rel_pathlist_hook)
586 0 : (*set_rel_pathlist_hook) (root, rel, rti, rte);
587 :
588 : /*
589 : * If this is a baserel, we should normally consider gathering any partial
590 : * paths we may have created for it. We have to do this after calling the
591 : * set_rel_pathlist_hook, else it cannot add partial paths to be included
592 : * here.
593 : *
594 : * However, if this is an inheritance child, skip it. Otherwise, we could
595 : * end up with a very large number of gather nodes, each trying to grab
596 : * its own pool of workers. Instead, we'll consider gathering partial
597 : * paths for the parent appendrel.
598 : *
599 : * Also, if this is the topmost scan/join rel, we postpone gathering until
600 : * the final scan/join targetlist is available (see grouping_planner).
601 : */
602 413129 : if (rel->reloptkind == RELOPT_BASEREL &&
603 364677 : !bms_equal(rel->relids, root->all_query_rels))
604 192325 : generate_useful_gather_paths(root, rel, false);
605 :
606 : /* Now find the cheapest of the paths for this rel */
607 413129 : set_cheapest(rel);
608 :
609 : /*
610 : * If a grouped relation for this rel exists, build partial aggregation
611 : * paths for it.
612 : *
613 : * Note that this can only happen after we've called set_cheapest() for
614 : * this base rel, because we need its cheapest paths.
615 : */
616 413129 : set_grouped_rel_pathlist(root, rel);
617 :
618 : #ifdef OPTIMIZER_DEBUG
619 : pprint(rel);
620 : #endif
621 413129 : }
622 :
623 : /*
624 : * set_plain_rel_size
625 : * Set size estimates for a plain relation (no subquery, no inheritance)
626 : */
627 : static void
628 322126 : set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
629 : {
630 : /*
631 : * Test any partial indexes of rel for applicability. We must do this
632 : * first since partial unique indexes can affect size estimates.
633 : */
634 322126 : check_index_predicates(root, rel);
635 :
636 : /* Mark rel with estimated output rows, width, etc */
637 322126 : set_baserel_size_estimates(root, rel);
638 322106 : }
639 :
640 : /*
641 : * If this relation could possibly be scanned from within a worker, then set
642 : * its consider_parallel flag.
643 : */
644 : static void
645 329174 : set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel,
646 : RangeTblEntry *rte)
647 : {
648 : /*
649 : * The flag has previously been initialized to false, so we can just
650 : * return if it becomes clear that we can't safely set it.
651 : */
652 : Assert(!rel->consider_parallel);
653 :
654 : /* Don't call this if parallelism is disallowed for the entire query. */
655 : Assert(root->glob->parallelModeOK);
656 :
657 : /* This should only be called for baserels and appendrel children. */
658 : Assert(IS_SIMPLE_REL(rel));
659 :
660 : /* Assorted checks based on rtekind. */
661 329174 : switch (rte->rtekind)
662 : {
663 283406 : case RTE_RELATION:
664 :
665 : /*
666 : * Currently, parallel workers can't access the leader's temporary
667 : * tables. We could possibly relax this if we wrote all of its
668 : * local buffers at the start of the query and made no changes
669 : * thereafter (maybe we could allow hint bit changes), and if we
670 : * taught the workers to read them. Writing a large number of
671 : * temporary buffers could be expensive, though, and we don't have
672 : * the rest of the necessary infrastructure right now anyway. So
673 : * for now, bail out if we see a temporary table.
674 : */
675 283406 : if (get_rel_persistence(rte->relid) == RELPERSISTENCE_TEMP)
676 7433 : return;
677 :
678 : /*
679 : * Table sampling can be pushed down to workers if the sample
680 : * function and its arguments are safe.
681 : */
682 275973 : if (rte->tablesample != NULL)
683 : {
684 263 : char proparallel = func_parallel(rte->tablesample->tsmhandler);
685 :
686 263 : if (proparallel != PROPARALLEL_SAFE)
687 18 : return;
688 245 : if (!is_parallel_safe(root, (Node *) rte->tablesample->args))
689 10 : return;
690 : }
691 :
692 : /*
693 : * Ask FDWs whether they can support performing a ForeignScan
694 : * within a worker. Most often, the answer will be no. For
695 : * example, if the nature of the FDW is such that it opens a TCP
696 : * connection with a remote server, each parallel worker would end
697 : * up with a separate connection, and these connections might not
698 : * be appropriately coordinated between workers and the leader.
699 : */
700 275945 : if (rte->relkind == RELKIND_FOREIGN_TABLE)
701 : {
702 : Assert(rel->fdwroutine);
703 797 : if (!rel->fdwroutine->IsForeignScanParallelSafe)
704 758 : return;
705 39 : if (!rel->fdwroutine->IsForeignScanParallelSafe(root, rel, rte))
706 0 : return;
707 : }
708 :
709 : /*
710 : * There are additional considerations for appendrels, which we'll
711 : * deal with in set_append_rel_size and set_append_rel_pathlist.
712 : * For now, just set consider_parallel based on the rel's own
713 : * quals and targetlist.
714 : */
715 275187 : break;
716 :
717 16306 : case RTE_SUBQUERY:
718 :
719 : /*
720 : * There's no intrinsic problem with scanning a subquery-in-FROM
721 : * (as distinct from a SubPlan or InitPlan) in a parallel worker.
722 : * If the subquery doesn't happen to have any parallel-safe paths,
723 : * then flagging it as consider_parallel won't change anything,
724 : * but that's true for plain tables, too. We must set
725 : * consider_parallel based on the rel's own quals and targetlist,
726 : * so that if a subquery path is parallel-safe but the quals and
727 : * projection we're sticking onto it are not, we correctly mark
728 : * the SubqueryScanPath as not parallel-safe. (Note that
729 : * set_subquery_pathlist() might push some of these quals down
730 : * into the subquery itself, but that doesn't change anything.)
731 : *
732 : * We can't push sub-select containing LIMIT/OFFSET to workers as
733 : * there is no guarantee that the row order will be fully
734 : * deterministic, and applying LIMIT/OFFSET will lead to
735 : * inconsistent results at the top-level. (In some cases, where
736 : * the result is ordered, we could relax this restriction. But it
737 : * doesn't currently seem worth expending extra effort to do so.)
738 : */
739 : {
740 16306 : Query *subquery = castNode(Query, rte->subquery);
741 :
742 16306 : if (limit_needed(subquery))
743 444 : return;
744 : }
745 15862 : break;
746 :
747 0 : case RTE_JOIN:
748 : /* Shouldn't happen; we're only considering baserels here. */
749 : Assert(false);
750 0 : return;
751 :
752 20137 : case RTE_FUNCTION:
753 : /* Check for parallel-restricted functions. */
754 20137 : if (!is_parallel_safe(root, (Node *) rte->functions))
755 8822 : return;
756 11315 : break;
757 :
758 519 : case RTE_TABLEFUNC:
759 : /* not parallel safe */
760 519 : return;
761 :
762 2381 : case RTE_VALUES:
763 : /* Check for parallel-restricted functions. */
764 2381 : if (!is_parallel_safe(root, (Node *) rte->values_lists))
765 10 : return;
766 2371 : break;
767 :
768 2759 : case RTE_CTE:
769 :
770 : /*
771 : * CTE tuplestores aren't shared among parallel workers, so we
772 : * force all CTE scans to happen in the leader. Also, populating
773 : * the CTE would require executing a subplan that's not available
774 : * in the worker, might be parallel-restricted, and must get
775 : * executed only once.
776 : */
777 2759 : return;
778 :
779 411 : case RTE_NAMEDTUPLESTORE:
780 :
781 : /*
782 : * tuplestore cannot be shared, at least without more
783 : * infrastructure to support that.
784 : */
785 411 : return;
786 :
787 3255 : case RTE_RESULT:
788 : /* RESULT RTEs, in themselves, are no problem. */
789 3255 : break;
790 :
791 0 : case RTE_GRAPH_TABLE:
792 :
793 : /*
794 : * Shouldn't happen since these are replaced by subquery RTEs when
795 : * rewriting queries.
796 : */
797 : Assert(false);
798 0 : return;
799 :
800 0 : case RTE_GROUP:
801 : /* Shouldn't happen; we're only considering baserels here. */
802 : Assert(false);
803 0 : return;
804 : }
805 :
806 : /*
807 : * If there's anything in baserestrictinfo that's parallel-restricted, we
808 : * give up on parallelizing access to this relation. We could consider
809 : * instead postponing application of the restricted quals until we're
810 : * above all the parallelism in the plan tree, but it's not clear that
811 : * that would be a win in very many cases, and it might be tricky to make
812 : * outer join clauses work correctly. It would likely break equivalence
813 : * classes, too.
814 : */
815 307990 : if (!is_parallel_safe(root, (Node *) rel->baserestrictinfo))
816 21830 : return;
817 :
818 : /*
819 : * Likewise, if the relation's outputs are not parallel-safe, give up.
820 : * (Usually, they're just Vars, but sometimes they're not.)
821 : */
822 286160 : if (!is_parallel_safe(root, (Node *) rel->reltarget->exprs))
823 15 : return;
824 :
825 : /* We have a winner. */
826 286145 : rel->consider_parallel = true;
827 : }
828 :
829 : /*
830 : * set_plain_rel_pathlist
831 : * Build access paths for a plain relation (no subquery, no inheritance)
832 : */
833 : static void
834 322106 : set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
835 : {
836 : Relids required_outer;
837 :
838 : /*
839 : * We don't support pushing join clauses into the quals of a seqscan, but
840 : * it could still have required parameterization due to LATERAL refs in
841 : * its tlist.
842 : */
843 322106 : required_outer = rel->lateral_relids;
844 :
845 : /*
846 : * Consider TID scans.
847 : *
848 : * If create_tidscan_paths returns true, then a TID scan path is forced.
849 : * This happens when rel->baserestrictinfo contains CurrentOfExpr, because
850 : * the executor can't handle any other type of path for such queries.
851 : * Hence, we return without adding any other paths.
852 : */
853 322106 : if (create_tidscan_paths(root, rel))
854 334 : return;
855 :
856 : /* Consider sequential scan */
857 321772 : add_path(rel, create_seqscan_path(root, rel, required_outer, 0));
858 :
859 : /* If appropriate, consider parallel sequential scan */
860 321772 : if (rel->consider_parallel && required_outer == NULL)
861 242601 : create_plain_partial_paths(root, rel);
862 :
863 : /* Consider index scans */
864 321772 : create_index_paths(root, rel);
865 : }
866 :
867 : /*
868 : * create_plain_partial_paths
869 : * Build partial access paths for parallel scan of a plain relation
870 : */
871 : static void
872 242601 : create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel)
873 : {
874 : int parallel_workers;
875 :
876 242601 : parallel_workers = compute_parallel_worker(rel, rel->pages, -1,
877 : max_parallel_workers_per_gather);
878 :
879 : /* If any limit was set to zero, the user doesn't want a parallel scan. */
880 242601 : if (parallel_workers <= 0)
881 218441 : return;
882 :
883 : /* Add an unordered partial path based on a parallel sequential scan. */
884 24160 : add_partial_path(rel, create_seqscan_path(root, rel, NULL, parallel_workers));
885 : }
886 :
887 : /*
888 : * set_tablesample_rel_size
889 : * Set size estimates for a sampled relation
890 : */
891 : static void
892 243 : set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
893 : {
894 243 : TableSampleClause *tsc = rte->tablesample;
895 : TsmRoutine *tsm;
896 : BlockNumber pages;
897 : double tuples;
898 :
899 : /*
900 : * Test any partial indexes of rel for applicability. We must do this
901 : * first since partial unique indexes can affect size estimates.
902 : */
903 243 : check_index_predicates(root, rel);
904 :
905 : /*
906 : * Call the sampling method's estimation function to estimate the number
907 : * of pages it will read and the number of tuples it will return. (Note:
908 : * we assume the function returns sane values.)
909 : */
910 243 : tsm = GetTsmRoutine(tsc->tsmhandler);
911 243 : tsm->SampleScanGetSampleSize(root, rel, tsc->args,
912 : &pages, &tuples);
913 :
914 : /*
915 : * For the moment, because we will only consider a SampleScan path for the
916 : * rel, it's okay to just overwrite the pages and tuples estimates for the
917 : * whole relation. If we ever consider multiple path types for sampled
918 : * rels, we'll need more complication.
919 : */
920 243 : rel->pages = pages;
921 243 : rel->tuples = tuples;
922 :
923 : /* Mark rel with estimated output rows, width, etc */
924 243 : set_baserel_size_estimates(root, rel);
925 243 : }
926 :
927 : /*
928 : * set_tablesample_rel_pathlist
929 : * Build access paths for a sampled relation
930 : */
931 : static void
932 243 : set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
933 : {
934 : Relids required_outer;
935 : Path *path;
936 :
937 : /*
938 : * We don't support pushing join clauses into the quals of a samplescan,
939 : * but it could still have required parameterization due to LATERAL refs
940 : * in its tlist or TABLESAMPLE arguments.
941 : */
942 243 : required_outer = rel->lateral_relids;
943 :
944 : /* Consider sampled scan */
945 243 : path = create_samplescan_path(root, rel, required_outer);
946 :
947 : /*
948 : * If the sampling method does not support repeatable scans, we must avoid
949 : * plans that would scan the rel multiple times. Ideally, we'd simply
950 : * avoid putting the rel on the inside of a nestloop join; but adding such
951 : * a consideration to the planner seems like a great deal of complication
952 : * to support an uncommon usage of second-rate sampling methods. Instead,
953 : * if there is a risk that the query might perform an unsafe join, just
954 : * wrap the SampleScan in a Materialize node. We can check for joins by
955 : * counting the membership of all_query_rels (note that this correctly
956 : * counts inheritance trees as single rels). If we're inside a subquery,
957 : * we can't easily check whether a join might occur in the outer query, so
958 : * just assume one is possible.
959 : *
960 : * GetTsmRoutine is relatively expensive compared to the other tests here,
961 : * so check repeatable_across_scans last, even though that's a bit odd.
962 : */
963 469 : if ((root->query_level > 1 ||
964 226 : bms_membership(root->all_query_rels) != BMS_SINGLETON) &&
965 77 : !(GetTsmRoutine(rte->tablesample->tsmhandler)->repeatable_across_scans))
966 : {
967 4 : path = (Path *) create_material_path(rel, path, true);
968 : }
969 :
970 243 : add_path(rel, path);
971 :
972 : /* For the moment, at least, there are no other paths to consider */
973 243 : }
974 :
975 : /*
976 : * set_foreign_size
977 : * Set size estimates for a foreign table RTE
978 : */
979 : static void
980 1252 : set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
981 : {
982 : /* Mark rel with estimated output rows, width, etc */
983 1252 : set_foreign_size_estimates(root, rel);
984 :
985 : /* Let FDW adjust the size estimates, if it can */
986 1252 : rel->fdwroutine->GetForeignRelSize(root, rel, rte->relid);
987 :
988 : /* ... but do not let it set the rows estimate to zero */
989 1250 : rel->rows = clamp_row_est(rel->rows);
990 :
991 : /*
992 : * Also, make sure rel->tuples is not insane relative to rel->rows.
993 : * Notably, this ensures sanity if pg_class.reltuples contains -1 and the
994 : * FDW doesn't do anything to replace that.
995 : */
996 1250 : rel->tuples = Max(rel->tuples, rel->rows);
997 1250 : }
998 :
999 : /*
1000 : * set_foreign_pathlist
1001 : * Build access paths for a foreign table RTE
1002 : */
1003 : static void
1004 1250 : set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
1005 : {
1006 : /* Call the FDW's GetForeignPaths function to generate path(s) */
1007 1250 : rel->fdwroutine->GetForeignPaths(root, rel, rte->relid);
1008 1250 : }
1009 :
1010 : /*
1011 : * set_append_rel_size
1012 : * Set size estimates for a simple "append relation"
1013 : *
1014 : * The passed-in rel and RTE represent the entire append relation. The
1015 : * relation's contents are computed by appending together the output of the
1016 : * individual member relations. Note that in the non-partitioned inheritance
1017 : * case, the first member relation is actually the same table as is mentioned
1018 : * in the parent RTE ... but it has a different RTE and RelOptInfo. This is
1019 : * a good thing because their outputs are not the same size.
1020 : */
1021 : static void
1022 20787 : set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
1023 : Index rti, RangeTblEntry *rte)
1024 : {
1025 20787 : int parentRTindex = rti;
1026 : bool has_live_children;
1027 : double parent_tuples;
1028 : double parent_rows;
1029 : double parent_size;
1030 : double *parent_attrsizes;
1031 : int nattrs;
1032 : ListCell *l;
1033 :
1034 : /* Guard against stack overflow due to overly deep inheritance tree. */
1035 20787 : check_stack_depth();
1036 :
1037 : Assert(IS_SIMPLE_REL(rel));
1038 :
1039 : /*
1040 : * If this is a partitioned baserel, set the consider_partitionwise_join
1041 : * flag; currently, we only consider partitionwise joins with the baserel
1042 : * if its targetlist doesn't contain a whole-row Var.
1043 : */
1044 20787 : if (enable_partitionwise_join &&
1045 4159 : rel->reloptkind == RELOPT_BASEREL &&
1046 3309 : rte->relkind == RELKIND_PARTITIONED_TABLE &&
1047 3309 : bms_is_empty(rel->attr_needed[InvalidAttrNumber - rel->min_attr]))
1048 3251 : rel->consider_partitionwise_join = true;
1049 :
1050 : /*
1051 : * Initialize to compute size estimates for whole append relation.
1052 : *
1053 : * We handle tuples estimates by setting "tuples" to the total number of
1054 : * tuples accumulated from each live child, rather than using "rows".
1055 : * Although an appendrel itself doesn't directly enforce any quals, its
1056 : * child relations may. Therefore, setting "tuples" equal to "rows" for
1057 : * an appendrel isn't always appropriate, and can lead to inaccurate cost
1058 : * estimates. For example, when estimating the number of distinct values
1059 : * from an appendrel, we would be unable to adjust the estimate based on
1060 : * the restriction selectivity (see estimate_num_groups).
1061 : *
1062 : * We handle width estimates by weighting the widths of different child
1063 : * rels proportionally to their number of rows. This is sensible because
1064 : * the use of width estimates is mainly to compute the total relation
1065 : * "footprint" if we have to sort or hash it. To do this, we sum the
1066 : * total equivalent size (in "double" arithmetic) and then divide by the
1067 : * total rowcount estimate. This is done separately for the total rel
1068 : * width and each attribute.
1069 : *
1070 : * Note: if you consider changing this logic, beware that child rels could
1071 : * have zero rows and/or width, if they were excluded by constraints.
1072 : */
1073 20787 : has_live_children = false;
1074 20787 : parent_tuples = 0;
1075 20787 : parent_rows = 0;
1076 20787 : parent_size = 0;
1077 20787 : nattrs = rel->max_attr - rel->min_attr + 1;
1078 20787 : parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
1079 :
1080 112735 : foreach(l, root->append_rel_list)
1081 : {
1082 91949 : AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1083 : int childRTindex;
1084 : RangeTblEntry *childRTE;
1085 : RelOptInfo *childrel;
1086 : List *childrinfos;
1087 : ListCell *parentvars;
1088 : ListCell *childvars;
1089 : ListCell *lc;
1090 :
1091 : /* append_rel_list contains all append rels; ignore others */
1092 91949 : if (appinfo->parent_relid != parentRTindex)
1093 43756 : continue;
1094 :
1095 48508 : childRTindex = appinfo->child_relid;
1096 48508 : childRTE = root->simple_rte_array[childRTindex];
1097 :
1098 : /*
1099 : * The child rel's RelOptInfo was already created during
1100 : * add_other_rels_to_query.
1101 : */
1102 48508 : childrel = find_base_rel(root, childRTindex);
1103 : Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
1104 :
1105 : /* We may have already proven the child to be dummy. */
1106 48508 : if (IS_DUMMY_REL(childrel))
1107 20 : continue;
1108 :
1109 : /*
1110 : * We have to copy the parent's targetlist and quals to the child,
1111 : * with appropriate substitution of variables. However, the
1112 : * baserestrictinfo quals were already copied/substituted when the
1113 : * child RelOptInfo was built. So we don't need any additional setup
1114 : * before applying constraint exclusion.
1115 : */
1116 48488 : if (relation_excluded_by_constraints(root, childrel, childRTE))
1117 : {
1118 : /*
1119 : * This child need not be scanned, so we can omit it from the
1120 : * appendrel.
1121 : */
1122 160 : set_dummy_rel_pathlist(childrel);
1123 160 : continue;
1124 : }
1125 :
1126 : /*
1127 : * Constraint exclusion failed, so copy the parent's join quals and
1128 : * targetlist to the child, with appropriate variable substitutions.
1129 : *
1130 : * We skip join quals that came from above outer joins that can null
1131 : * this rel, since they would be of no value while generating paths
1132 : * for the child. This saves some effort while processing the child
1133 : * rel, and it also avoids an implementation restriction in
1134 : * adjust_appendrel_attrs (it can't apply nullingrels to a non-Var).
1135 : */
1136 48328 : childrinfos = NIL;
1137 59430 : foreach(lc, rel->joininfo)
1138 : {
1139 11102 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
1140 :
1141 11102 : if (!bms_overlap(rinfo->clause_relids, rel->nulling_relids))
1142 9167 : childrinfos = lappend(childrinfos,
1143 9167 : adjust_appendrel_attrs(root,
1144 : (Node *) rinfo,
1145 : 1, &appinfo));
1146 : }
1147 48328 : childrel->joininfo = childrinfos;
1148 :
1149 : /*
1150 : * Now for the child's targetlist.
1151 : *
1152 : * NB: the resulting childrel->reltarget->exprs may contain arbitrary
1153 : * expressions, which otherwise would not occur in a rel's targetlist.
1154 : * Code that might be looking at an appendrel child must cope with
1155 : * such. (Normally, a rel's targetlist would only include Vars and
1156 : * PlaceHolderVars.) XXX we do not bother to update the cost or width
1157 : * fields of childrel->reltarget; not clear if that would be useful.
1158 : */
1159 96656 : childrel->reltarget->exprs = (List *)
1160 48328 : adjust_appendrel_attrs(root,
1161 48328 : (Node *) rel->reltarget->exprs,
1162 : 1, &appinfo);
1163 :
1164 : /*
1165 : * We have to make child entries in the EquivalenceClass data
1166 : * structures as well. This is needed either if the parent
1167 : * participates in some eclass joins (because we will want to consider
1168 : * inner-indexscan joins on the individual children) or if the parent
1169 : * has useful pathkeys (because we should try to build MergeAppend
1170 : * paths that produce those sort orderings).
1171 : */
1172 48328 : if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
1173 29962 : add_child_rel_equivalences(root, appinfo, rel, childrel);
1174 48328 : childrel->has_eclass_joins = rel->has_eclass_joins;
1175 :
1176 : /*
1177 : * Note: we could compute appropriate attr_needed data for the child's
1178 : * variables, by transforming the parent's attr_needed through the
1179 : * translated_vars mapping. However, currently there's no need
1180 : * because attr_needed is only examined for base relations not
1181 : * otherrels. So we just leave the child's attr_needed empty.
1182 : */
1183 :
1184 : /*
1185 : * If we consider partitionwise joins with the parent rel, do the same
1186 : * for partitioned child rels.
1187 : *
1188 : * Note: here we abuse the consider_partitionwise_join flag by setting
1189 : * it for child rels that are not themselves partitioned. We do so to
1190 : * tell try_partitionwise_join() that the child rel is sufficiently
1191 : * valid to be used as a per-partition input, even if it later gets
1192 : * proven to be dummy. (It's not usable until we've set up the
1193 : * reltarget and EC entries, which we just did.)
1194 : */
1195 48328 : if (rel->consider_partitionwise_join)
1196 11007 : childrel->consider_partitionwise_join = true;
1197 :
1198 : /*
1199 : * If parallelism is allowable for this query in general, see whether
1200 : * it's allowable for this childrel in particular. But if we've
1201 : * already decided the appendrel is not parallel-safe as a whole,
1202 : * there's no point in considering parallelism for this child. For
1203 : * consistency, do this before calling set_rel_size() for the child.
1204 : */
1205 48328 : if (root->glob->parallelModeOK && rel->consider_parallel)
1206 36284 : set_rel_consider_parallel(root, childrel, childRTE);
1207 :
1208 : /*
1209 : * Compute the child's size.
1210 : */
1211 48328 : set_rel_size(root, childrel, childRTindex, childRTE);
1212 :
1213 : /*
1214 : * It is possible that constraint exclusion detected a contradiction
1215 : * within a child subquery, even though we didn't prove one above. If
1216 : * so, we can skip this child.
1217 : */
1218 48327 : if (IS_DUMMY_REL(childrel))
1219 135 : continue;
1220 :
1221 : /* We have at least one live child. */
1222 48192 : has_live_children = true;
1223 :
1224 : /*
1225 : * If any live child is not parallel-safe, treat the whole appendrel
1226 : * as not parallel-safe. In future we might be able to generate plans
1227 : * in which some children are farmed out to workers while others are
1228 : * not; but we don't have that today, so it's a waste to consider
1229 : * partial paths anywhere in the appendrel unless it's all safe.
1230 : * (Child rels visited before this one will be unmarked in
1231 : * set_append_rel_pathlist().)
1232 : */
1233 48192 : if (!childrel->consider_parallel)
1234 12426 : rel->consider_parallel = false;
1235 :
1236 : /*
1237 : * Accumulate size information from each live child.
1238 : */
1239 : Assert(childrel->rows > 0);
1240 :
1241 48192 : parent_tuples += childrel->tuples;
1242 48192 : parent_rows += childrel->rows;
1243 48192 : parent_size += childrel->reltarget->width * childrel->rows;
1244 :
1245 : /*
1246 : * Accumulate per-column estimates too. We need not do anything for
1247 : * PlaceHolderVars in the parent list. If child expression isn't a
1248 : * Var, or we didn't record a width estimate for it, we have to fall
1249 : * back on a datatype-based estimate.
1250 : *
1251 : * By construction, child's targetlist is 1-to-1 with parent's.
1252 : */
1253 157519 : forboth(parentvars, rel->reltarget->exprs,
1254 : childvars, childrel->reltarget->exprs)
1255 : {
1256 109327 : Var *parentvar = (Var *) lfirst(parentvars);
1257 109327 : Node *childvar = (Node *) lfirst(childvars);
1258 :
1259 109327 : if (IsA(parentvar, Var) && parentvar->varno == parentRTindex)
1260 : {
1261 98953 : int pndx = parentvar->varattno - rel->min_attr;
1262 98953 : int32 child_width = 0;
1263 :
1264 98953 : if (IsA(childvar, Var) &&
1265 94744 : ((Var *) childvar)->varno == childrel->relid)
1266 : {
1267 94669 : int cndx = ((Var *) childvar)->varattno - childrel->min_attr;
1268 :
1269 94669 : child_width = childrel->attr_widths[cndx];
1270 : }
1271 98953 : if (child_width <= 0)
1272 4284 : child_width = get_typavgwidth(exprType(childvar),
1273 : exprTypmod(childvar));
1274 : Assert(child_width > 0);
1275 98953 : parent_attrsizes[pndx] += child_width * childrel->rows;
1276 : }
1277 : }
1278 : }
1279 :
1280 20786 : if (has_live_children)
1281 : {
1282 : /*
1283 : * Save the finished size estimates.
1284 : */
1285 : int i;
1286 :
1287 : Assert(parent_rows > 0);
1288 20530 : rel->tuples = parent_tuples;
1289 20530 : rel->rows = parent_rows;
1290 20530 : rel->reltarget->width = rint(parent_size / parent_rows);
1291 189086 : for (i = 0; i < nattrs; i++)
1292 168556 : rel->attr_widths[i] = rint(parent_attrsizes[i] / parent_rows);
1293 :
1294 : /*
1295 : * Note that we leave rel->pages as zero; this is important to avoid
1296 : * double-counting the appendrel tree in total_table_pages.
1297 : */
1298 : }
1299 : else
1300 : {
1301 : /*
1302 : * All children were excluded by constraints, so mark the whole
1303 : * appendrel dummy. We must do this in this phase so that the rel's
1304 : * dummy-ness is visible when we generate paths for other rels.
1305 : */
1306 256 : set_dummy_rel_pathlist(rel);
1307 : }
1308 :
1309 20786 : pfree(parent_attrsizes);
1310 20786 : }
1311 :
1312 : /*
1313 : * set_append_rel_pathlist
1314 : * Build access paths for an "append relation"
1315 : */
1316 : static void
1317 20530 : set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
1318 : Index rti, RangeTblEntry *rte)
1319 : {
1320 20530 : int parentRTindex = rti;
1321 20530 : List *live_childrels = NIL;
1322 : ListCell *l;
1323 :
1324 : /*
1325 : * Generate access paths for each member relation, and remember the
1326 : * non-dummy children.
1327 : */
1328 112073 : foreach(l, root->append_rel_list)
1329 : {
1330 91543 : AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1331 : int childRTindex;
1332 : RangeTblEntry *childRTE;
1333 : RelOptInfo *childrel;
1334 :
1335 : /* append_rel_list contains all append rels; ignore others */
1336 91543 : if (appinfo->parent_relid != parentRTindex)
1337 43091 : continue;
1338 :
1339 : /* Re-locate the child RTE and RelOptInfo */
1340 48452 : childRTindex = appinfo->child_relid;
1341 48452 : childRTE = root->simple_rte_array[childRTindex];
1342 48452 : childrel = root->simple_rel_array[childRTindex];
1343 :
1344 : /*
1345 : * If set_append_rel_size() decided the parent appendrel was
1346 : * parallel-unsafe at some point after visiting this child rel, we
1347 : * need to propagate the unsafety marking down to the child, so that
1348 : * we don't generate useless partial paths for it.
1349 : */
1350 48452 : if (!rel->consider_parallel)
1351 12595 : childrel->consider_parallel = false;
1352 :
1353 : /*
1354 : * Compute the child's access paths.
1355 : */
1356 48452 : set_rel_pathlist(root, childrel, childRTindex, childRTE);
1357 :
1358 : /*
1359 : * If child is dummy, ignore it.
1360 : */
1361 48452 : if (IS_DUMMY_REL(childrel))
1362 260 : continue;
1363 :
1364 : /*
1365 : * Child is live, so add it to the live_childrels list for use below.
1366 : */
1367 48192 : live_childrels = lappend(live_childrels, childrel);
1368 : }
1369 :
1370 : /* Add paths to the append relation. */
1371 20530 : add_paths_to_append_rel(root, rel, live_childrels);
1372 20530 : }
1373 :
1374 : /*
1375 : * set_grouped_rel_pathlist
1376 : * If a grouped relation for the given 'rel' exists, build partial
1377 : * aggregation paths for it.
1378 : */
1379 : static void
1380 413129 : set_grouped_rel_pathlist(PlannerInfo *root, RelOptInfo *rel)
1381 : {
1382 : RelOptInfo *grouped_rel;
1383 :
1384 : /*
1385 : * If there are no aggregate expressions or grouping expressions, eager
1386 : * aggregation is not possible.
1387 : */
1388 413129 : if (root->agg_clause_list == NIL ||
1389 2774 : root->group_expr_list == NIL)
1390 410879 : return;
1391 :
1392 : /* Add paths to the grouped base relation if one exists. */
1393 2250 : grouped_rel = rel->grouped_rel;
1394 2250 : if (grouped_rel)
1395 : {
1396 : Assert(IS_GROUPED_REL(grouped_rel));
1397 :
1398 467 : generate_grouped_paths(root, grouped_rel, rel);
1399 467 : set_cheapest(grouped_rel);
1400 : }
1401 : }
1402 :
1403 :
1404 : /*
1405 : * add_paths_to_append_rel
1406 : * Generate paths for the given append relation given the set of non-dummy
1407 : * child rels.
1408 : *
1409 : * The function collects all parameterizations and orderings supported by the
1410 : * non-dummy children. For every such parameterization or ordering, it creates
1411 : * an append path collecting one path from each non-dummy child with given
1412 : * parameterization or ordering. Similarly it collects partial paths from
1413 : * non-dummy children to create partial append paths.
1414 : */
1415 : void
1416 37364 : add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
1417 : List *live_childrels)
1418 : {
1419 37364 : AppendPathInput unparameterized = {0};
1420 37364 : AppendPathInput startup = {0};
1421 37364 : AppendPathInput partial_only = {0};
1422 37364 : AppendPathInput parallel_append = {0};
1423 37364 : bool unparameterized_valid = true;
1424 37364 : bool startup_valid = true;
1425 37364 : bool partial_only_valid = true;
1426 37364 : bool parallel_append_valid = true;
1427 37364 : List *all_child_pathkeys = NIL;
1428 37364 : List *all_child_outers = NIL;
1429 : ListCell *l;
1430 37364 : double partial_rows = -1;
1431 :
1432 : /* If appropriate, consider parallel append */
1433 37364 : parallel_append_valid = enable_parallel_append && rel->consider_parallel;
1434 :
1435 : /*
1436 : * For every non-dummy child, remember the cheapest path. Also, identify
1437 : * all pathkeys (orderings) and parameterizations (required_outer sets)
1438 : * available for the non-dummy member relations.
1439 : */
1440 123118 : foreach(l, live_childrels)
1441 : {
1442 85754 : RelOptInfo *childrel = lfirst(l);
1443 : ListCell *lcp;
1444 85754 : Path *cheapest_partial_path = NULL;
1445 :
1446 : /*
1447 : * If child has an unparameterized cheapest-total path, add that to
1448 : * the unparameterized Append path we are constructing for the parent.
1449 : * If not, there's no workable unparameterized path.
1450 : *
1451 : * With partitionwise aggregates, the child rel's pathlist may be
1452 : * empty, so don't assume that a path exists here.
1453 : */
1454 85754 : if (childrel->pathlist != NIL &&
1455 85754 : childrel->cheapest_total_path->param_info == NULL)
1456 85124 : accumulate_append_subpath(childrel->cheapest_total_path,
1457 : &unparameterized.subpaths, NULL, &unparameterized.child_append_relid_sets);
1458 : else
1459 630 : unparameterized_valid = false;
1460 :
1461 : /*
1462 : * When the planner is considering cheap startup plans, we'll also
1463 : * collect all the cheapest_startup_paths (if set) and build an
1464 : * AppendPath containing those as subpaths.
1465 : */
1466 85754 : if (rel->consider_startup && childrel->cheapest_startup_path != NULL)
1467 1425 : {
1468 : Path *cheapest_path;
1469 :
1470 : /*
1471 : * With an indication of how many tuples the query should provide,
1472 : * the optimizer tries to choose the path optimal for that
1473 : * specific number of tuples.
1474 : */
1475 1425 : if (root->tuple_fraction > 0.0)
1476 : cheapest_path =
1477 1425 : get_cheapest_fractional_path(childrel,
1478 : root->tuple_fraction);
1479 : else
1480 0 : cheapest_path = childrel->cheapest_startup_path;
1481 :
1482 : /* cheapest_startup_path must not be a parameterized path. */
1483 : Assert(cheapest_path->param_info == NULL);
1484 1425 : accumulate_append_subpath(cheapest_path,
1485 : &startup.subpaths,
1486 : NULL,
1487 : &startup.child_append_relid_sets);
1488 : }
1489 : else
1490 84329 : startup_valid = false;
1491 :
1492 :
1493 : /* Same idea, but for a partial plan. */
1494 85754 : if (childrel->partial_pathlist != NIL)
1495 : {
1496 53874 : cheapest_partial_path = linitial(childrel->partial_pathlist);
1497 53874 : accumulate_append_subpath(cheapest_partial_path,
1498 : &partial_only.partial_subpaths, NULL,
1499 : &partial_only.child_append_relid_sets);
1500 : }
1501 : else
1502 31880 : partial_only_valid = false;
1503 :
1504 : /*
1505 : * Same idea, but for a parallel append mixing partial and non-partial
1506 : * paths.
1507 : */
1508 85754 : if (parallel_append_valid)
1509 : {
1510 65339 : Path *nppath = NULL;
1511 :
1512 : nppath =
1513 65339 : get_cheapest_parallel_safe_total_inner(childrel->pathlist);
1514 :
1515 65339 : if (cheapest_partial_path == NULL && nppath == NULL)
1516 : {
1517 : /* Neither a partial nor a parallel-safe path? Forget it. */
1518 413 : parallel_append_valid = false;
1519 : }
1520 64926 : else if (nppath == NULL ||
1521 53499 : (cheapest_partial_path != NULL &&
1522 53499 : cheapest_partial_path->total_cost < nppath->total_cost))
1523 : {
1524 : /* Partial path is cheaper or the only option. */
1525 : Assert(cheapest_partial_path != NULL);
1526 53302 : accumulate_append_subpath(cheapest_partial_path,
1527 : ¶llel_append.partial_subpaths,
1528 : ¶llel_append.subpaths,
1529 : ¶llel_append.child_append_relid_sets);
1530 : }
1531 : else
1532 : {
1533 : /*
1534 : * Either we've got only a non-partial path, or we think that
1535 : * a single backend can execute the best non-partial path
1536 : * faster than all the parallel backends working together can
1537 : * execute the best partial path.
1538 : *
1539 : * It might make sense to be more aggressive here. Even if
1540 : * the best non-partial path is more expensive than the best
1541 : * partial path, it could still be better to choose the
1542 : * non-partial path if there are several such paths that can
1543 : * be given to different workers. For now, we don't try to
1544 : * figure that out.
1545 : */
1546 11624 : accumulate_append_subpath(nppath,
1547 : ¶llel_append.subpaths,
1548 : NULL,
1549 : ¶llel_append.child_append_relid_sets);
1550 : }
1551 : }
1552 :
1553 : /*
1554 : * Collect lists of all the available path orderings and
1555 : * parameterizations for all the children. We use these as a
1556 : * heuristic to indicate which sort orderings and parameterizations we
1557 : * should build Append and MergeAppend paths for.
1558 : */
1559 199892 : foreach(lcp, childrel->pathlist)
1560 : {
1561 114138 : Path *childpath = (Path *) lfirst(lcp);
1562 114138 : List *childkeys = childpath->pathkeys;
1563 114138 : Relids childouter = PATH_REQ_OUTER(childpath);
1564 :
1565 : /* Unsorted paths don't contribute to pathkey list */
1566 114138 : if (childkeys != NIL)
1567 : {
1568 : ListCell *lpk;
1569 28680 : bool found = false;
1570 :
1571 : /* Have we already seen this ordering? */
1572 28833 : foreach(lpk, all_child_pathkeys)
1573 : {
1574 19540 : List *existing_pathkeys = (List *) lfirst(lpk);
1575 :
1576 19540 : if (compare_pathkeys(existing_pathkeys,
1577 : childkeys) == PATHKEYS_EQUAL)
1578 : {
1579 19387 : found = true;
1580 19387 : break;
1581 : }
1582 : }
1583 28680 : if (!found)
1584 : {
1585 : /* No, so add it to all_child_pathkeys */
1586 9293 : all_child_pathkeys = lappend(all_child_pathkeys,
1587 : childkeys);
1588 : }
1589 : }
1590 :
1591 : /* Unparameterized paths don't contribute to param-set list */
1592 114138 : if (childouter)
1593 : {
1594 : ListCell *lco;
1595 5693 : bool found = false;
1596 :
1597 : /* Have we already seen this param set? */
1598 6341 : foreach(lco, all_child_outers)
1599 : {
1600 4149 : Relids existing_outers = (Relids) lfirst(lco);
1601 :
1602 4149 : if (bms_equal(existing_outers, childouter))
1603 : {
1604 3501 : found = true;
1605 3501 : break;
1606 : }
1607 : }
1608 5693 : if (!found)
1609 : {
1610 : /* No, so add it to all_child_outers */
1611 2192 : all_child_outers = lappend(all_child_outers,
1612 : childouter);
1613 : }
1614 : }
1615 : }
1616 : }
1617 :
1618 : /*
1619 : * If we found unparameterized paths for all children, build an unordered,
1620 : * unparameterized Append path for the rel. (Note: this is correct even
1621 : * if we have zero or one live subpath due to constraint exclusion.)
1622 : */
1623 37364 : if (unparameterized_valid)
1624 37094 : add_path(rel, (Path *) create_append_path(root, rel, unparameterized,
1625 : NIL, NULL, 0, false,
1626 : -1));
1627 :
1628 : /* build an AppendPath for the cheap startup paths, if valid */
1629 37364 : if (startup_valid)
1630 575 : add_path(rel, (Path *) create_append_path(root, rel, startup,
1631 : NIL, NULL, 0, false, -1));
1632 :
1633 : /*
1634 : * Consider an append of unordered, unparameterized partial paths. Make
1635 : * it parallel-aware if possible.
1636 : */
1637 37364 : if (partial_only_valid && partial_only.partial_subpaths != NIL)
1638 : {
1639 : AppendPath *appendpath;
1640 : ListCell *lc;
1641 21953 : int parallel_workers = 0;
1642 :
1643 : /* Find the highest number of workers requested for any subpath. */
1644 79485 : foreach(lc, partial_only.partial_subpaths)
1645 : {
1646 57532 : Path *path = lfirst(lc);
1647 :
1648 57532 : parallel_workers = Max(parallel_workers, path->parallel_workers);
1649 : }
1650 : Assert(parallel_workers > 0);
1651 :
1652 : /*
1653 : * If the use of parallel append is permitted, always request at least
1654 : * log2(# of children) workers. We assume it can be useful to have
1655 : * extra workers in this case because they will be spread out across
1656 : * the children. The precise formula is just a guess, but we don't
1657 : * want to end up with a radically different answer for a table with N
1658 : * partitions vs. an unpartitioned table with the same data, so the
1659 : * use of some kind of log-scaling here seems to make some sense.
1660 : */
1661 21953 : if (enable_parallel_append)
1662 : {
1663 21913 : parallel_workers = Max(parallel_workers,
1664 : pg_leftmost_one_pos32(list_length(live_childrels)) + 1);
1665 21913 : parallel_workers = Min(parallel_workers,
1666 : max_parallel_workers_per_gather);
1667 : }
1668 : Assert(parallel_workers > 0);
1669 :
1670 : /* Generate a partial append path. */
1671 21953 : appendpath = create_append_path(root, rel, partial_only,
1672 : NIL, NULL, parallel_workers,
1673 : enable_parallel_append,
1674 : -1);
1675 :
1676 : /*
1677 : * Make sure any subsequent partial paths use the same row count
1678 : * estimate.
1679 : */
1680 21953 : partial_rows = appendpath->path.rows;
1681 :
1682 : /* Add the path. */
1683 21953 : add_partial_path(rel, (Path *) appendpath);
1684 : }
1685 :
1686 : /*
1687 : * Consider a parallel-aware append using a mix of partial and non-partial
1688 : * paths. (This only makes sense if there's at least one child which has
1689 : * a non-partial path that is substantially cheaper than any partial path;
1690 : * otherwise, we should use the append path added in the previous step.)
1691 : */
1692 37364 : if (parallel_append_valid && parallel_append.subpaths != NIL)
1693 : {
1694 : AppendPath *appendpath;
1695 : ListCell *lc;
1696 4257 : int parallel_workers = 0;
1697 :
1698 : /*
1699 : * Find the highest number of workers requested for any partial
1700 : * subpath.
1701 : */
1702 5002 : foreach(lc, parallel_append.partial_subpaths)
1703 : {
1704 745 : Path *path = lfirst(lc);
1705 :
1706 745 : parallel_workers = Max(parallel_workers, path->parallel_workers);
1707 : }
1708 :
1709 : /*
1710 : * Same formula here as above. It's even more important in this
1711 : * instance because the non-partial paths won't contribute anything to
1712 : * the planned number of parallel workers.
1713 : */
1714 4257 : parallel_workers = Max(parallel_workers,
1715 : pg_leftmost_one_pos32(list_length(live_childrels)) + 1);
1716 4257 : parallel_workers = Min(parallel_workers,
1717 : max_parallel_workers_per_gather);
1718 : Assert(parallel_workers > 0);
1719 :
1720 4257 : appendpath = create_append_path(root, rel, parallel_append,
1721 : NIL, NULL, parallel_workers, true,
1722 : partial_rows);
1723 4257 : add_partial_path(rel, (Path *) appendpath);
1724 : }
1725 :
1726 : /*
1727 : * Also build unparameterized ordered append paths based on the collected
1728 : * list of child pathkeys.
1729 : */
1730 37364 : if (unparameterized_valid)
1731 37094 : generate_orderedappend_paths(root, rel, live_childrels,
1732 : all_child_pathkeys);
1733 :
1734 : /*
1735 : * Build Append paths for each parameterization seen among the child rels.
1736 : * (This may look pretty expensive, but in most cases of practical
1737 : * interest, the child rels will expose mostly the same parameterizations,
1738 : * so that not that many cases actually get considered here.)
1739 : *
1740 : * The Append node itself cannot enforce quals, so all qual checking must
1741 : * be done in the child paths. This means that to have a parameterized
1742 : * Append path, we must have the exact same parameterization for each
1743 : * child path; otherwise some children might be failing to check the
1744 : * moved-down quals. To make them match up, we can try to increase the
1745 : * parameterization of lesser-parameterized paths.
1746 : */
1747 39556 : foreach(l, all_child_outers)
1748 : {
1749 2192 : Relids required_outer = (Relids) lfirst(l);
1750 : ListCell *lcr;
1751 2192 : AppendPathInput parameterized = {0};
1752 2192 : bool parameterized_valid = true;
1753 :
1754 : /* Select the child paths for an Append with this parameterization */
1755 7984 : foreach(lcr, live_childrels)
1756 : {
1757 5802 : RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1758 : Path *subpath;
1759 :
1760 5802 : if (childrel->pathlist == NIL)
1761 : {
1762 : /* failed to make a suitable path for this child */
1763 0 : parameterized_valid = false;
1764 0 : break;
1765 : }
1766 :
1767 5802 : subpath = get_cheapest_parameterized_child_path(root,
1768 : childrel,
1769 : required_outer);
1770 5802 : if (subpath == NULL)
1771 : {
1772 : /* failed to make a suitable path for this child */
1773 10 : parameterized_valid = false;
1774 10 : break;
1775 : }
1776 5792 : accumulate_append_subpath(subpath, ¶meterized.subpaths, NULL,
1777 : ¶meterized.child_append_relid_sets);
1778 : }
1779 :
1780 2192 : if (parameterized_valid)
1781 2182 : add_path(rel, (Path *)
1782 2182 : create_append_path(root, rel, parameterized,
1783 : NIL, required_outer, 0, false,
1784 : -1));
1785 : }
1786 :
1787 : /*
1788 : * When there is only a single child relation, the Append path can inherit
1789 : * any ordering available for the child rel's path, so that it's useful to
1790 : * consider ordered partial paths. Above we only considered the cheapest
1791 : * partial path for each child, but let's also make paths using any
1792 : * partial paths that have pathkeys.
1793 : */
1794 37364 : if (list_length(live_childrels) == 1)
1795 : {
1796 10355 : RelOptInfo *childrel = (RelOptInfo *) linitial(live_childrels);
1797 :
1798 : /* skip the cheapest partial path, since we already used that above */
1799 10504 : for_each_from(l, childrel->partial_pathlist, 1)
1800 : {
1801 149 : Path *path = (Path *) lfirst(l);
1802 : AppendPath *appendpath;
1803 149 : AppendPathInput append = {0};
1804 :
1805 : /* skip paths with no pathkeys. */
1806 149 : if (path->pathkeys == NIL)
1807 0 : continue;
1808 :
1809 149 : append.partial_subpaths = list_make1(path);
1810 149 : appendpath = create_append_path(root, rel, append, NIL, NULL,
1811 : path->parallel_workers, true,
1812 : partial_rows);
1813 149 : add_partial_path(rel, (Path *) appendpath);
1814 : }
1815 : }
1816 37364 : }
1817 :
1818 : /*
1819 : * generate_orderedappend_paths
1820 : * Generate ordered append paths for an append relation
1821 : *
1822 : * Usually we generate MergeAppend paths here, but there are some special
1823 : * cases where we can generate simple Append paths, because the subpaths
1824 : * can provide tuples in the required order already.
1825 : *
1826 : * We generate a path for each ordering (pathkey list) appearing in
1827 : * all_child_pathkeys.
1828 : *
1829 : * We consider the cheapest-startup and cheapest-total cases, and also the
1830 : * cheapest-fractional case when not all tuples need to be retrieved. For each
1831 : * interesting ordering, we collect all the cheapest startup subpaths, all the
1832 : * cheapest total paths, and, if applicable, all the cheapest fractional paths,
1833 : * and build a suitable path for each case.
1834 : *
1835 : * We don't currently generate any parameterized ordered paths here. While
1836 : * it would not take much more code here to do so, it's very unclear that it
1837 : * is worth the planning cycles to investigate such paths: there's little
1838 : * use for an ordered path on the inside of a nestloop. In fact, it's likely
1839 : * that the current coding of add_path would reject such paths out of hand,
1840 : * because add_path gives no credit for sort ordering of parameterized paths,
1841 : * and a parameterized MergeAppend is going to be more expensive than the
1842 : * corresponding parameterized Append path. If we ever try harder to support
1843 : * parameterized mergejoin plans, it might be worth adding support for
1844 : * parameterized paths here to feed such joins. (See notes in
1845 : * optimizer/README for why that might not ever happen, though.)
1846 : */
1847 : static void
1848 37094 : generate_orderedappend_paths(PlannerInfo *root, RelOptInfo *rel,
1849 : List *live_childrels,
1850 : List *all_child_pathkeys)
1851 : {
1852 : ListCell *lcp;
1853 37094 : List *partition_pathkeys = NIL;
1854 37094 : List *partition_pathkeys_desc = NIL;
1855 37094 : bool partition_pathkeys_partial = true;
1856 37094 : bool partition_pathkeys_desc_partial = true;
1857 :
1858 : /*
1859 : * Some partitioned table setups may allow us to use an Append node
1860 : * instead of a MergeAppend. This is possible in cases such as RANGE
1861 : * partitioned tables where it's guaranteed that an earlier partition must
1862 : * contain rows which come earlier in the sort order. To detect whether
1863 : * this is relevant, build pathkey descriptions of the partition ordering,
1864 : * for both forward and reverse scans.
1865 : */
1866 59350 : if (rel->part_scheme != NULL && IS_SIMPLE_REL(rel) &&
1867 22256 : partitions_are_ordered(rel->boundinfo, rel->live_parts))
1868 : {
1869 18252 : partition_pathkeys = build_partition_pathkeys(root, rel,
1870 : ForwardScanDirection,
1871 : &partition_pathkeys_partial);
1872 :
1873 18252 : partition_pathkeys_desc = build_partition_pathkeys(root, rel,
1874 : BackwardScanDirection,
1875 : &partition_pathkeys_desc_partial);
1876 :
1877 : /*
1878 : * You might think we should truncate_useless_pathkeys here, but
1879 : * allowing partition keys which are a subset of the query's pathkeys
1880 : * can often be useful. For example, consider a table partitioned by
1881 : * RANGE (a, b), and a query with ORDER BY a, b, c. If we have child
1882 : * paths that can produce the a, b, c ordering (perhaps via indexes on
1883 : * (a, b, c)) then it works to consider the appendrel output as
1884 : * ordered by a, b, c.
1885 : */
1886 : }
1887 :
1888 : /* Now consider each interesting sort ordering */
1889 46337 : foreach(lcp, all_child_pathkeys)
1890 : {
1891 9243 : List *pathkeys = (List *) lfirst(lcp);
1892 9243 : AppendPathInput startup = {0};
1893 9243 : AppendPathInput total = {0};
1894 9243 : AppendPathInput fractional = {0};
1895 9243 : bool startup_neq_total = false;
1896 9243 : bool fraction_neq_total = false;
1897 : bool match_partition_order;
1898 : bool match_partition_order_desc;
1899 : int end_index;
1900 : int first_index;
1901 : int direction;
1902 :
1903 : /*
1904 : * Determine if this sort ordering matches any partition pathkeys we
1905 : * have, for both ascending and descending partition order. If the
1906 : * partition pathkeys happen to be contained in pathkeys then it still
1907 : * works, as described above, providing that the partition pathkeys
1908 : * are complete and not just a prefix of the partition keys. (In such
1909 : * cases we'll be relying on the child paths to have sorted the
1910 : * lower-order columns of the required pathkeys.)
1911 : */
1912 9243 : match_partition_order =
1913 16563 : pathkeys_contained_in(pathkeys, partition_pathkeys) ||
1914 7484 : (!partition_pathkeys_partial &&
1915 164 : pathkeys_contained_in(partition_pathkeys, pathkeys));
1916 :
1917 23681 : match_partition_order_desc = !match_partition_order &&
1918 7234 : (pathkeys_contained_in(pathkeys, partition_pathkeys_desc) ||
1919 7252 : (!partition_pathkeys_desc_partial &&
1920 48 : pathkeys_contained_in(partition_pathkeys_desc, pathkeys)));
1921 :
1922 : /*
1923 : * When the required pathkeys match the reverse of the partition
1924 : * order, we must build the list of paths in reverse starting with the
1925 : * last matching partition first. We can get away without making any
1926 : * special cases for this in the loop below by just looping backward
1927 : * over the child relations in this case.
1928 : */
1929 9243 : if (match_partition_order_desc)
1930 : {
1931 : /* loop backward */
1932 40 : first_index = list_length(live_childrels) - 1;
1933 40 : end_index = -1;
1934 40 : direction = -1;
1935 :
1936 : /*
1937 : * Set this to true to save us having to check for
1938 : * match_partition_order_desc in the loop below.
1939 : */
1940 40 : match_partition_order = true;
1941 : }
1942 : else
1943 : {
1944 : /* for all other case, loop forward */
1945 9203 : first_index = 0;
1946 9203 : end_index = list_length(live_childrels);
1947 9203 : direction = 1;
1948 : }
1949 :
1950 : /* Select the child paths for this ordering... */
1951 33383 : for (int i = first_index; i != end_index; i += direction)
1952 : {
1953 24140 : RelOptInfo *childrel = list_nth_node(RelOptInfo, live_childrels, i);
1954 : Path *cheapest_startup,
1955 : *cheapest_total,
1956 24140 : *cheapest_fractional = NULL;
1957 :
1958 : /* Locate the right paths, if they are available. */
1959 : cheapest_startup =
1960 24140 : get_cheapest_path_for_pathkeys(childrel->pathlist,
1961 : pathkeys,
1962 : NULL,
1963 : STARTUP_COST,
1964 : false);
1965 : cheapest_total =
1966 24140 : get_cheapest_path_for_pathkeys(childrel->pathlist,
1967 : pathkeys,
1968 : NULL,
1969 : TOTAL_COST,
1970 : false);
1971 :
1972 : /*
1973 : * If we can't find any paths with the right order just use the
1974 : * cheapest-total path; we'll have to sort it later.
1975 : */
1976 24140 : if (cheapest_startup == NULL || cheapest_total == NULL)
1977 : {
1978 509 : cheapest_startup = cheapest_total =
1979 : childrel->cheapest_total_path;
1980 : /* Assert we do have an unparameterized path for this child */
1981 : Assert(cheapest_total->param_info == NULL);
1982 : }
1983 :
1984 : /*
1985 : * When building a fractional path, determine a cheapest
1986 : * fractional path for each child relation too. Looking at startup
1987 : * and total costs is not enough, because the cheapest fractional
1988 : * path may be dominated by two separate paths (one for startup,
1989 : * one for total).
1990 : *
1991 : * When needed (building fractional path), determine the cheapest
1992 : * fractional path too.
1993 : */
1994 24140 : if (root->tuple_fraction > 0)
1995 : {
1996 736 : double path_fraction = root->tuple_fraction;
1997 :
1998 : /*
1999 : * We should not have a dummy child relation here. However,
2000 : * we cannot use childrel->rows to compute the tuple fraction,
2001 : * as childrel can be an upper relation with an unset row
2002 : * estimate. Instead, we use the row estimate from the
2003 : * cheapest_total path, which should already have been forced
2004 : * to a sane value.
2005 : */
2006 : Assert(cheapest_total->rows > 0);
2007 :
2008 : /* Convert absolute limit to a path fraction */
2009 736 : if (path_fraction >= 1.0)
2010 736 : path_fraction /= cheapest_total->rows;
2011 :
2012 : cheapest_fractional =
2013 736 : get_cheapest_fractional_path_for_pathkeys(childrel->pathlist,
2014 : pathkeys,
2015 : NULL,
2016 : path_fraction);
2017 :
2018 : /*
2019 : * If we found no path with matching pathkeys, use the
2020 : * cheapest total path instead.
2021 : *
2022 : * XXX We might consider partially sorted paths too (with an
2023 : * incremental sort on top). But we'd have to build all the
2024 : * incremental paths, do the costing etc.
2025 : *
2026 : * Also, notice whether we actually have different paths for
2027 : * the "fractional" and "total" cases. This helps avoid
2028 : * generating two identical ordered append paths.
2029 : */
2030 736 : if (cheapest_fractional == NULL)
2031 34 : cheapest_fractional = cheapest_total;
2032 702 : else if (cheapest_fractional != cheapest_total)
2033 0 : fraction_neq_total = true;
2034 : }
2035 :
2036 : /*
2037 : * Notice whether we actually have different paths for the
2038 : * "cheapest" and "total" cases. This helps avoid generating two
2039 : * identical ordered append paths.
2040 : */
2041 24140 : if (cheapest_startup != cheapest_total)
2042 68 : startup_neq_total = true;
2043 :
2044 : /*
2045 : * Collect the appropriate child paths. The required logic varies
2046 : * for the Append and MergeAppend cases.
2047 : */
2048 24140 : if (match_partition_order)
2049 : {
2050 : /*
2051 : * We're going to make a plain Append path. We don't need
2052 : * most of what accumulate_append_subpath would do, but we do
2053 : * want to cut out child Appends or MergeAppends if they have
2054 : * just a single subpath (and hence aren't doing anything
2055 : * useful).
2056 : */
2057 : cheapest_startup =
2058 5472 : get_singleton_append_subpath(cheapest_startup,
2059 : &startup.child_append_relid_sets);
2060 : cheapest_total =
2061 5472 : get_singleton_append_subpath(cheapest_total,
2062 : &total.child_append_relid_sets);
2063 :
2064 5472 : startup.subpaths = lappend(startup.subpaths, cheapest_startup);
2065 5472 : total.subpaths = lappend(total.subpaths, cheapest_total);
2066 :
2067 5472 : if (cheapest_fractional)
2068 : {
2069 : cheapest_fractional =
2070 120 : get_singleton_append_subpath(cheapest_fractional,
2071 : &fractional.child_append_relid_sets);
2072 120 : fractional.subpaths =
2073 120 : lappend(fractional.subpaths, cheapest_fractional);
2074 : }
2075 : }
2076 : else
2077 : {
2078 : /*
2079 : * Otherwise, rely on accumulate_append_subpath to collect the
2080 : * child paths for the MergeAppend.
2081 : */
2082 18668 : accumulate_append_subpath(cheapest_startup,
2083 : &startup.subpaths, NULL,
2084 : &startup.child_append_relid_sets);
2085 18668 : accumulate_append_subpath(cheapest_total,
2086 : &total.subpaths, NULL,
2087 : &total.child_append_relid_sets);
2088 :
2089 18668 : if (cheapest_fractional)
2090 616 : accumulate_append_subpath(cheapest_fractional,
2091 : &fractional.subpaths, NULL,
2092 : &fractional.child_append_relid_sets);
2093 : }
2094 : }
2095 :
2096 : /* ... and build the Append or MergeAppend paths */
2097 9243 : if (match_partition_order)
2098 : {
2099 : /* We only need Append */
2100 2049 : add_path(rel, (Path *) create_append_path(root,
2101 : rel,
2102 : startup,
2103 : pathkeys,
2104 : NULL,
2105 : 0,
2106 : false,
2107 : -1));
2108 2049 : if (startup_neq_total)
2109 0 : add_path(rel, (Path *) create_append_path(root,
2110 : rel,
2111 : total,
2112 : pathkeys,
2113 : NULL,
2114 : 0,
2115 : false,
2116 : -1));
2117 :
2118 2049 : if (fractional.subpaths && fraction_neq_total)
2119 0 : add_path(rel, (Path *) create_append_path(root,
2120 : rel,
2121 : fractional,
2122 : pathkeys,
2123 : NULL,
2124 : 0,
2125 : false,
2126 : -1));
2127 : }
2128 : else
2129 : {
2130 : /* We need MergeAppend */
2131 7194 : add_path(rel, (Path *) create_merge_append_path(root,
2132 : rel,
2133 : startup.subpaths,
2134 : startup.child_append_relid_sets,
2135 : pathkeys,
2136 : NULL));
2137 7194 : if (startup_neq_total)
2138 44 : add_path(rel, (Path *) create_merge_append_path(root,
2139 : rel,
2140 : total.subpaths,
2141 : total.child_append_relid_sets,
2142 : pathkeys,
2143 : NULL));
2144 :
2145 7194 : if (fractional.subpaths && fraction_neq_total)
2146 0 : add_path(rel, (Path *) create_merge_append_path(root,
2147 : rel,
2148 : fractional.subpaths,
2149 : fractional.child_append_relid_sets,
2150 : pathkeys,
2151 : NULL));
2152 : }
2153 : }
2154 37094 : }
2155 :
2156 : /*
2157 : * get_cheapest_parameterized_child_path
2158 : * Get cheapest path for this relation that has exactly the requested
2159 : * parameterization.
2160 : *
2161 : * Returns NULL if unable to create such a path.
2162 : */
2163 : static Path *
2164 5802 : get_cheapest_parameterized_child_path(PlannerInfo *root, RelOptInfo *rel,
2165 : Relids required_outer)
2166 : {
2167 : Path *cheapest;
2168 : ListCell *lc;
2169 :
2170 : /*
2171 : * Look up the cheapest existing path with no more than the needed
2172 : * parameterization. If it has exactly the needed parameterization, we're
2173 : * done.
2174 : */
2175 5802 : cheapest = get_cheapest_path_for_pathkeys(rel->pathlist,
2176 : NIL,
2177 : required_outer,
2178 : TOTAL_COST,
2179 : false);
2180 : Assert(cheapest != NULL);
2181 5802 : if (bms_equal(PATH_REQ_OUTER(cheapest), required_outer))
2182 4962 : return cheapest;
2183 :
2184 : /*
2185 : * Otherwise, we can "reparameterize" an existing path to match the given
2186 : * parameterization, which effectively means pushing down additional
2187 : * joinquals to be checked within the path's scan. However, some existing
2188 : * paths might check the available joinquals already while others don't;
2189 : * therefore, it's not clear which existing path will be cheapest after
2190 : * reparameterization. We have to go through them all and find out.
2191 : */
2192 840 : cheapest = NULL;
2193 3298 : foreach(lc, rel->pathlist)
2194 : {
2195 2458 : Path *path = (Path *) lfirst(lc);
2196 :
2197 : /* Can't use it if it needs more than requested parameterization */
2198 2458 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
2199 148 : continue;
2200 :
2201 : /*
2202 : * Reparameterization can only increase the path's cost, so if it's
2203 : * already more expensive than the current cheapest, forget it.
2204 : */
2205 3732 : if (cheapest != NULL &&
2206 1422 : compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
2207 1372 : continue;
2208 :
2209 : /* Reparameterize if needed, then recheck cost */
2210 938 : if (!bms_equal(PATH_REQ_OUTER(path), required_outer))
2211 : {
2212 853 : path = reparameterize_path(root, path, required_outer, 1.0);
2213 853 : if (path == NULL)
2214 58 : continue; /* failed to reparameterize this one */
2215 : Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));
2216 :
2217 795 : if (cheapest != NULL &&
2218 0 : compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
2219 0 : continue;
2220 : }
2221 :
2222 : /* We have a new best path */
2223 880 : cheapest = path;
2224 : }
2225 :
2226 : /* Return the best path, or NULL if we found no suitable candidate */
2227 840 : return cheapest;
2228 : }
2229 :
2230 : /*
2231 : * accumulate_append_subpath
2232 : * Add a subpath to the list being built for an Append or MergeAppend.
2233 : *
2234 : * It's possible that the child is itself an Append or MergeAppend path, in
2235 : * which case we can "cut out the middleman" and just add its child paths to
2236 : * our own list. (We don't try to do this earlier because we need to apply
2237 : * both levels of transformation to the quals.)
2238 : *
2239 : * Note that if we omit a child MergeAppend in this way, we are effectively
2240 : * omitting a sort step, which seems fine: if the parent is to be an Append,
2241 : * its result would be unsorted anyway, while if the parent is to be a
2242 : * MergeAppend, there's no point in a separate sort on a child.
2243 : *
2244 : * Normally, either path is a partial path and subpaths is a list of partial
2245 : * paths, or else path is a non-partial plan and subpaths is a list of those.
2246 : * However, if path is a parallel-aware Append, then we add its partial path
2247 : * children to subpaths and the rest to special_subpaths. If the latter is
2248 : * NULL, we don't flatten the path at all (unless it contains only partial
2249 : * paths).
2250 : */
2251 : static void
2252 249093 : accumulate_append_subpath(Path *path, List **subpaths, List **special_subpaths,
2253 : List **child_append_relid_sets)
2254 : {
2255 249093 : if (IsA(path, AppendPath))
2256 : {
2257 13008 : AppendPath *apath = (AppendPath *) path;
2258 :
2259 13008 : if (!apath->path.parallel_aware || apath->first_partial_path == 0)
2260 : {
2261 12728 : *subpaths = list_concat(*subpaths, apath->subpaths);
2262 12728 : *child_append_relid_sets =
2263 12728 : lappend(*child_append_relid_sets, path->parent->relids);
2264 12728 : *child_append_relid_sets =
2265 12728 : list_concat(*child_append_relid_sets,
2266 12728 : apath->child_append_relid_sets);
2267 12728 : return;
2268 : }
2269 280 : else if (special_subpaths != NULL)
2270 : {
2271 : List *new_special_subpaths;
2272 :
2273 : /* Split Parallel Append into partial and non-partial subpaths */
2274 140 : *subpaths = list_concat(*subpaths,
2275 140 : list_copy_tail(apath->subpaths,
2276 : apath->first_partial_path));
2277 140 : new_special_subpaths = list_copy_head(apath->subpaths,
2278 : apath->first_partial_path);
2279 140 : *special_subpaths = list_concat(*special_subpaths,
2280 : new_special_subpaths);
2281 140 : *child_append_relid_sets =
2282 140 : lappend(*child_append_relid_sets, path->parent->relids);
2283 140 : *child_append_relid_sets =
2284 140 : list_concat(*child_append_relid_sets,
2285 140 : apath->child_append_relid_sets);
2286 140 : return;
2287 : }
2288 : }
2289 236085 : else if (IsA(path, MergeAppendPath))
2290 : {
2291 818 : MergeAppendPath *mpath = (MergeAppendPath *) path;
2292 :
2293 818 : *subpaths = list_concat(*subpaths, mpath->subpaths);
2294 818 : *child_append_relid_sets =
2295 818 : lappend(*child_append_relid_sets, path->parent->relids);
2296 818 : *child_append_relid_sets =
2297 818 : list_concat(*child_append_relid_sets,
2298 818 : mpath->child_append_relid_sets);
2299 818 : return;
2300 : }
2301 :
2302 235407 : *subpaths = lappend(*subpaths, path);
2303 : }
2304 :
2305 : /*
2306 : * get_singleton_append_subpath
2307 : * Returns the single subpath of an Append/MergeAppend, or just
2308 : * return 'path' if it's not a single sub-path Append/MergeAppend.
2309 : *
2310 : * As a side effect, whenever we return a single subpath rather than the
2311 : * original path, add the relid sets for the original path to
2312 : * child_append_relid_sets, so that those relids don't entirely disappear
2313 : * from the final plan.
2314 : *
2315 : * Note: 'path' must not be a parallel-aware path.
2316 : */
2317 : static Path *
2318 11064 : get_singleton_append_subpath(Path *path, List **child_append_relid_sets)
2319 : {
2320 : Assert(!path->parallel_aware);
2321 :
2322 11064 : if (IsA(path, AppendPath))
2323 : {
2324 328 : AppendPath *apath = (AppendPath *) path;
2325 :
2326 328 : if (list_length(apath->subpaths) == 1)
2327 : {
2328 178 : *child_append_relid_sets =
2329 178 : lappend(*child_append_relid_sets, path->parent->relids);
2330 178 : *child_append_relid_sets =
2331 178 : list_concat(*child_append_relid_sets,
2332 178 : apath->child_append_relid_sets);
2333 178 : return (Path *) linitial(apath->subpaths);
2334 : }
2335 : }
2336 10736 : else if (IsA(path, MergeAppendPath))
2337 : {
2338 290 : MergeAppendPath *mpath = (MergeAppendPath *) path;
2339 :
2340 290 : if (list_length(mpath->subpaths) == 1)
2341 : {
2342 0 : *child_append_relid_sets =
2343 0 : lappend(*child_append_relid_sets, path->parent->relids);
2344 0 : *child_append_relid_sets =
2345 0 : list_concat(*child_append_relid_sets,
2346 0 : mpath->child_append_relid_sets);
2347 0 : return (Path *) linitial(mpath->subpaths);
2348 : }
2349 : }
2350 :
2351 10886 : return path;
2352 : }
2353 :
2354 : /*
2355 : * set_dummy_rel_pathlist
2356 : * Build a dummy path for a relation that's been excluded by constraints
2357 : *
2358 : * Rather than inventing a special "dummy" path type, we represent this as an
2359 : * AppendPath with no members (see also IS_DUMMY_APPEND/IS_DUMMY_REL macros).
2360 : *
2361 : * (See also mark_dummy_rel, which does basically the same thing, but is
2362 : * typically used to change a rel into dummy state after we already made
2363 : * paths for it.)
2364 : */
2365 : static void
2366 1178 : set_dummy_rel_pathlist(RelOptInfo *rel)
2367 : {
2368 1178 : AppendPathInput in = {0};
2369 :
2370 : /* Set dummy size estimates --- we leave attr_widths[] as zeroes */
2371 1178 : rel->rows = 0;
2372 1178 : rel->reltarget->width = 0;
2373 :
2374 : /* Discard any pre-existing paths; no further need for them */
2375 1178 : rel->pathlist = NIL;
2376 1178 : rel->partial_pathlist = NIL;
2377 :
2378 : /* Set up the dummy path */
2379 1178 : add_path(rel, (Path *) create_append_path(NULL, rel, in,
2380 : NIL, rel->lateral_relids,
2381 : 0, false, -1));
2382 :
2383 : /*
2384 : * We set the cheapest-path fields immediately, just in case they were
2385 : * pointing at some discarded path. This is redundant in current usage
2386 : * because set_rel_pathlist will do it later, but it's cheap so we keep it
2387 : * for safety and consistency with mark_dummy_rel.
2388 : */
2389 1178 : set_cheapest(rel);
2390 1178 : }
2391 :
2392 : /*
2393 : * find_window_run_conditions
2394 : * Determine if 'wfunc' is really a WindowFunc and call its prosupport
2395 : * function to determine the function's monotonic properties. We then
2396 : * see if 'opexpr' can be used to short-circuit execution.
2397 : *
2398 : * For example row_number() over (order by ...) always produces a value one
2399 : * higher than the previous. If someone has a window function in a subquery
2400 : * and has a WHERE clause in the outer query to filter rows <= 10, then we may
2401 : * as well stop processing the windowagg once the row number reaches 11. Here
2402 : * we check if 'opexpr' might help us to stop doing needless extra processing
2403 : * in WindowAgg nodes.
2404 : *
2405 : * '*keep_original' is set to true if the caller should also use 'opexpr' for
2406 : * its original purpose. This is set to false if the caller can assume that
2407 : * the run condition will handle all of the required filtering.
2408 : *
2409 : * Returns true if 'opexpr' was found to be useful and was added to the
2410 : * WindowFunc's runCondition. We also set *keep_original accordingly and add
2411 : * 'attno' to *run_cond_attrs offset by FirstLowInvalidHeapAttributeNumber.
2412 : * If the 'opexpr' cannot be used then we set *keep_original to true and
2413 : * return false.
2414 : */
2415 : static bool
2416 200 : find_window_run_conditions(Query *subquery, AttrNumber attno,
2417 : WindowFunc *wfunc, OpExpr *opexpr, bool wfunc_left,
2418 : bool *keep_original, Bitmapset **run_cond_attrs)
2419 : {
2420 : Oid prosupport;
2421 : Expr *otherexpr;
2422 : SupportRequestWFuncMonotonic req;
2423 : SupportRequestWFuncMonotonic *res;
2424 : WindowClause *wclause;
2425 : List *opinfos;
2426 : OpExpr *runopexpr;
2427 : Oid runoperator;
2428 : ListCell *lc;
2429 :
2430 200 : *keep_original = true;
2431 :
2432 200 : while (IsA(wfunc, RelabelType))
2433 0 : wfunc = (WindowFunc *) ((RelabelType *) wfunc)->arg;
2434 :
2435 : /* we can only work with window functions */
2436 200 : if (!IsA(wfunc, WindowFunc))
2437 20 : return false;
2438 :
2439 : /* can't use it if there are subplans in the WindowFunc */
2440 180 : if (contain_subplans((Node *) wfunc))
2441 5 : return false;
2442 :
2443 175 : prosupport = get_func_support(wfunc->winfnoid);
2444 :
2445 : /* Check if there's a support function for 'wfunc' */
2446 175 : if (!OidIsValid(prosupport))
2447 15 : return false;
2448 :
2449 : /* get the Expr from the other side of the OpExpr */
2450 160 : if (wfunc_left)
2451 140 : otherexpr = lsecond(opexpr->args);
2452 : else
2453 20 : otherexpr = linitial(opexpr->args);
2454 :
2455 : /*
2456 : * The value being compared must not change during the evaluation of the
2457 : * window partition.
2458 : */
2459 160 : if (!is_pseudo_constant_clause((Node *) otherexpr))
2460 0 : return false;
2461 :
2462 : /* find the window clause belonging to the window function */
2463 160 : wclause = (WindowClause *) list_nth(subquery->windowClause,
2464 160 : wfunc->winref - 1);
2465 :
2466 160 : req.type = T_SupportRequestWFuncMonotonic;
2467 160 : req.window_func = wfunc;
2468 160 : req.window_clause = wclause;
2469 :
2470 : /* call the support function */
2471 : res = (SupportRequestWFuncMonotonic *)
2472 160 : DatumGetPointer(OidFunctionCall1(prosupport,
2473 : PointerGetDatum(&req)));
2474 :
2475 : /*
2476 : * Nothing to do if the function is neither monotonically increasing nor
2477 : * monotonically decreasing.
2478 : */
2479 160 : if (res == NULL || res->monotonic == MONOTONICFUNC_NONE)
2480 0 : return false;
2481 :
2482 160 : runopexpr = NULL;
2483 160 : runoperator = InvalidOid;
2484 160 : opinfos = get_op_index_interpretation(opexpr->opno);
2485 :
2486 160 : foreach(lc, opinfos)
2487 : {
2488 160 : OpIndexInterpretation *opinfo = (OpIndexInterpretation *) lfirst(lc);
2489 160 : CompareType cmptype = opinfo->cmptype;
2490 :
2491 : /* handle < / <= */
2492 160 : if (cmptype == COMPARE_LT || cmptype == COMPARE_LE)
2493 : {
2494 : /*
2495 : * < / <= is supported for monotonically increasing functions in
2496 : * the form <wfunc> op <pseudoconst> and <pseudoconst> op <wfunc>
2497 : * for monotonically decreasing functions.
2498 : */
2499 115 : if ((wfunc_left && (res->monotonic & MONOTONICFUNC_INCREASING)) ||
2500 15 : (!wfunc_left && (res->monotonic & MONOTONICFUNC_DECREASING)))
2501 : {
2502 105 : *keep_original = false;
2503 105 : runopexpr = opexpr;
2504 105 : runoperator = opexpr->opno;
2505 : }
2506 115 : break;
2507 : }
2508 : /* handle > / >= */
2509 45 : else if (cmptype == COMPARE_GT || cmptype == COMPARE_GE)
2510 : {
2511 : /*
2512 : * > / >= is supported for monotonically decreasing functions in
2513 : * the form <wfunc> op <pseudoconst> and <pseudoconst> op <wfunc>
2514 : * for monotonically increasing functions.
2515 : */
2516 15 : if ((wfunc_left && (res->monotonic & MONOTONICFUNC_DECREASING)) ||
2517 10 : (!wfunc_left && (res->monotonic & MONOTONICFUNC_INCREASING)))
2518 : {
2519 15 : *keep_original = false;
2520 15 : runopexpr = opexpr;
2521 15 : runoperator = opexpr->opno;
2522 : }
2523 15 : break;
2524 : }
2525 : /* handle = */
2526 30 : else if (cmptype == COMPARE_EQ)
2527 : {
2528 : CompareType newcmptype;
2529 :
2530 : /*
2531 : * When both monotonically increasing and decreasing then the
2532 : * return value of the window function will be the same each time.
2533 : * We can simply use 'opexpr' as the run condition without
2534 : * modifying it.
2535 : */
2536 30 : if ((res->monotonic & MONOTONICFUNC_BOTH) == MONOTONICFUNC_BOTH)
2537 : {
2538 5 : *keep_original = false;
2539 5 : runopexpr = opexpr;
2540 5 : runoperator = opexpr->opno;
2541 5 : break;
2542 : }
2543 :
2544 : /*
2545 : * When monotonically increasing we make a qual with <wfunc> <=
2546 : * <value> or <value> >= <wfunc> in order to filter out values
2547 : * which are above the value in the equality condition. For
2548 : * monotonically decreasing functions we want to filter values
2549 : * below the value in the equality condition.
2550 : */
2551 25 : if (res->monotonic & MONOTONICFUNC_INCREASING)
2552 25 : newcmptype = wfunc_left ? COMPARE_LE : COMPARE_GE;
2553 : else
2554 0 : newcmptype = wfunc_left ? COMPARE_GE : COMPARE_LE;
2555 :
2556 : /* We must keep the original equality qual */
2557 25 : *keep_original = true;
2558 25 : runopexpr = opexpr;
2559 :
2560 : /* determine the operator to use for the WindowFuncRunCondition */
2561 25 : runoperator = get_opfamily_member_for_cmptype(opinfo->opfamily_id,
2562 : opinfo->oplefttype,
2563 : opinfo->oprighttype,
2564 : newcmptype);
2565 25 : break;
2566 : }
2567 : }
2568 :
2569 160 : if (runopexpr != NULL)
2570 : {
2571 : WindowFuncRunCondition *wfuncrc;
2572 :
2573 150 : wfuncrc = makeNode(WindowFuncRunCondition);
2574 150 : wfuncrc->opno = runoperator;
2575 150 : wfuncrc->inputcollid = runopexpr->inputcollid;
2576 150 : wfuncrc->wfunc_left = wfunc_left;
2577 150 : wfuncrc->arg = copyObject(otherexpr);
2578 :
2579 150 : wfunc->runCondition = lappend(wfunc->runCondition, wfuncrc);
2580 :
2581 : /* record that this attno was used in a run condition */
2582 150 : *run_cond_attrs = bms_add_member(*run_cond_attrs,
2583 : attno - FirstLowInvalidHeapAttributeNumber);
2584 150 : return true;
2585 : }
2586 :
2587 : /* unsupported OpExpr */
2588 10 : return false;
2589 : }
2590 :
2591 : /*
2592 : * check_and_push_window_quals
2593 : * Check if 'clause' is a qual that can be pushed into a WindowFunc
2594 : * as a 'runCondition' qual. These, when present, allow some unnecessary
2595 : * work to be skipped during execution.
2596 : *
2597 : * 'run_cond_attrs' will be populated with all targetlist resnos of subquery
2598 : * targets (offset by FirstLowInvalidHeapAttributeNumber) that we pushed
2599 : * window quals for.
2600 : *
2601 : * Returns true if the caller still must keep the original qual or false if
2602 : * the caller can safely ignore the original qual because the WindowAgg node
2603 : * will use the runCondition to stop returning tuples.
2604 : */
2605 : static bool
2606 210 : check_and_push_window_quals(Query *subquery, Node *clause,
2607 : Bitmapset **run_cond_attrs)
2608 : {
2609 210 : OpExpr *opexpr = (OpExpr *) clause;
2610 210 : bool keep_original = true;
2611 : Var *var1;
2612 : Var *var2;
2613 :
2614 : /* We're only able to use OpExprs with 2 operands */
2615 210 : if (!IsA(opexpr, OpExpr))
2616 15 : return true;
2617 :
2618 195 : if (list_length(opexpr->args) != 2)
2619 0 : return true;
2620 :
2621 : /*
2622 : * Currently, we restrict this optimization to strict OpExprs. The reason
2623 : * for this is that during execution, once the runcondition becomes false,
2624 : * we stop evaluating WindowFuncs. To avoid leaving around stale window
2625 : * function result values, we set them to NULL. Having only strict
2626 : * OpExprs here ensures that we properly filter out the tuples with NULLs
2627 : * in the top-level WindowAgg.
2628 : */
2629 195 : set_opfuncid(opexpr);
2630 195 : if (!func_strict(opexpr->opfuncid))
2631 0 : return true;
2632 :
2633 : /*
2634 : * Check for plain Vars that reference window functions in the subquery.
2635 : * If we find any, we'll ask find_window_run_conditions() if 'opexpr' can
2636 : * be used as part of the run condition.
2637 : */
2638 :
2639 : /* Check the left side of the OpExpr */
2640 195 : var1 = linitial(opexpr->args);
2641 195 : if (IsA(var1, Var) && var1->varattno > 0)
2642 : {
2643 165 : TargetEntry *tle = list_nth(subquery->targetList, var1->varattno - 1);
2644 165 : WindowFunc *wfunc = (WindowFunc *) tle->expr;
2645 :
2646 165 : if (find_window_run_conditions(subquery, tle->resno, wfunc, opexpr,
2647 : true, &keep_original, run_cond_attrs))
2648 135 : return keep_original;
2649 : }
2650 :
2651 : /* and check the right side */
2652 60 : var2 = lsecond(opexpr->args);
2653 60 : if (IsA(var2, Var) && var2->varattno > 0)
2654 : {
2655 35 : TargetEntry *tle = list_nth(subquery->targetList, var2->varattno - 1);
2656 35 : WindowFunc *wfunc = (WindowFunc *) tle->expr;
2657 :
2658 35 : if (find_window_run_conditions(subquery, tle->resno, wfunc, opexpr,
2659 : false, &keep_original, run_cond_attrs))
2660 15 : return keep_original;
2661 : }
2662 :
2663 45 : return true;
2664 : }
2665 :
2666 : /*
2667 : * set_subquery_pathlist
2668 : * Generate SubqueryScan access paths for a subquery RTE
2669 : *
2670 : * We don't currently support generating parameterized paths for subqueries
2671 : * by pushing join clauses down into them; it seems too expensive to re-plan
2672 : * the subquery multiple times to consider different alternatives.
2673 : * (XXX that could stand to be reconsidered, now that we use Paths.)
2674 : * So the paths made here will be parameterized if the subquery contains
2675 : * LATERAL references, otherwise not. As long as that's true, there's no need
2676 : * for a separate set_subquery_size phase: just make the paths right away.
2677 : */
2678 : static void
2679 17357 : set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
2680 : Index rti, RangeTblEntry *rte)
2681 : {
2682 17357 : Query *parse = root->parse;
2683 17357 : Query *subquery = rte->subquery;
2684 : bool trivial_pathtarget;
2685 : Relids required_outer;
2686 : pushdown_safety_info safetyInfo;
2687 : double tuple_fraction;
2688 : RelOptInfo *sub_final_rel;
2689 17357 : Bitmapset *run_cond_attrs = NULL;
2690 : ListCell *lc;
2691 : char *plan_name;
2692 :
2693 : /*
2694 : * Must copy the Query so that planning doesn't mess up the RTE contents
2695 : * (really really need to fix the planner to not scribble on its input,
2696 : * someday ... but see remove_unused_subquery_outputs to start with).
2697 : */
2698 17357 : subquery = copyObject(subquery);
2699 :
2700 : /*
2701 : * If it's a LATERAL subquery, it might contain some Vars of the current
2702 : * query level, requiring it to be treated as parameterized, even though
2703 : * we don't support pushing down join quals into subqueries.
2704 : */
2705 17357 : required_outer = rel->lateral_relids;
2706 :
2707 : /*
2708 : * Zero out result area for subquery_is_pushdown_safe, so that it can set
2709 : * flags as needed while recursing. In particular, we need a workspace
2710 : * for keeping track of the reasons why columns are unsafe to reference.
2711 : * These reasons are stored in the bits inside unsafeFlags[i] when we
2712 : * discover reasons that column i of the subquery is unsafe to be used in
2713 : * a pushed-down qual.
2714 : */
2715 17357 : memset(&safetyInfo, 0, sizeof(safetyInfo));
2716 17357 : safetyInfo.unsafeFlags = (unsigned char *)
2717 17357 : palloc0((list_length(subquery->targetList) + 1) * sizeof(unsigned char));
2718 :
2719 : /*
2720 : * If the subquery has the "security_barrier" flag, it means the subquery
2721 : * originated from a view that must enforce row-level security. Then we
2722 : * must not push down quals that contain leaky functions. (Ideally this
2723 : * would be checked inside subquery_is_pushdown_safe, but since we don't
2724 : * currently pass the RTE to that function, we must do it here.)
2725 : */
2726 17357 : safetyInfo.unsafeLeaky = rte->security_barrier;
2727 :
2728 : /*
2729 : * If there are any restriction clauses that have been attached to the
2730 : * subquery relation, consider pushing them down to become WHERE or HAVING
2731 : * quals of the subquery itself. This transformation is useful because it
2732 : * may allow us to generate a better plan for the subquery than evaluating
2733 : * all the subquery output rows and then filtering them.
2734 : *
2735 : * There are several cases where we cannot push down clauses. Restrictions
2736 : * involving the subquery are checked by subquery_is_pushdown_safe().
2737 : * Restrictions on individual clauses are checked by
2738 : * qual_is_pushdown_safe(). Also, we don't want to push down
2739 : * pseudoconstant clauses; better to have the gating node above the
2740 : * subquery.
2741 : *
2742 : * Non-pushed-down clauses will get evaluated as qpquals of the
2743 : * SubqueryScan node.
2744 : *
2745 : * XXX Are there any cases where we want to make a policy decision not to
2746 : * push down a pushable qual, because it'd result in a worse plan?
2747 : */
2748 19520 : if (rel->baserestrictinfo != NIL &&
2749 2163 : subquery_is_pushdown_safe(subquery, subquery, &safetyInfo))
2750 : {
2751 : /* OK to consider pushing down individual quals */
2752 2042 : List *upperrestrictlist = NIL;
2753 : ListCell *l;
2754 :
2755 5558 : foreach(l, rel->baserestrictinfo)
2756 : {
2757 3516 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(l);
2758 3516 : Node *clause = (Node *) rinfo->clause;
2759 :
2760 3516 : if (rinfo->pseudoconstant)
2761 : {
2762 2 : upperrestrictlist = lappend(upperrestrictlist, rinfo);
2763 2 : continue;
2764 : }
2765 :
2766 3514 : switch (qual_is_pushdown_safe(subquery, rti, rinfo, &safetyInfo))
2767 : {
2768 2664 : case PUSHDOWN_SAFE:
2769 : /* Push it down */
2770 2664 : subquery_push_qual(subquery, rte, rti, clause);
2771 2664 : break;
2772 :
2773 210 : case PUSHDOWN_WINDOWCLAUSE_RUNCOND:
2774 :
2775 : /*
2776 : * Since we can't push the qual down into the subquery,
2777 : * check if it happens to reference a window function. If
2778 : * so then it might be useful to use for the WindowAgg's
2779 : * runCondition.
2780 : */
2781 420 : if (!subquery->hasWindowFuncs ||
2782 210 : check_and_push_window_quals(subquery, clause,
2783 : &run_cond_attrs))
2784 : {
2785 : /*
2786 : * subquery has no window funcs or the clause is not a
2787 : * suitable window run condition qual or it is, but
2788 : * the original must also be kept in the upper query.
2789 : */
2790 85 : upperrestrictlist = lappend(upperrestrictlist, rinfo);
2791 : }
2792 210 : break;
2793 :
2794 640 : case PUSHDOWN_UNSAFE:
2795 640 : upperrestrictlist = lappend(upperrestrictlist, rinfo);
2796 640 : break;
2797 : }
2798 : }
2799 2042 : rel->baserestrictinfo = upperrestrictlist;
2800 : /* We don't bother recomputing baserestrict_min_security */
2801 : }
2802 :
2803 17357 : pfree(safetyInfo.unsafeFlags);
2804 :
2805 : /*
2806 : * The upper query might not use all the subquery's output columns; if
2807 : * not, we can simplify. Pass the attributes that were pushed down into
2808 : * WindowAgg run conditions to ensure we don't accidentally think those
2809 : * are unused.
2810 : */
2811 17357 : remove_unused_subquery_outputs(subquery, rel, run_cond_attrs);
2812 :
2813 : /*
2814 : * We can safely pass the outer tuple_fraction down to the subquery if the
2815 : * outer level has no joining, aggregation, or sorting to do. Otherwise
2816 : * we'd better tell the subquery to plan for full retrieval. (XXX This
2817 : * could probably be made more intelligent ...)
2818 : */
2819 17357 : if (parse->hasAggs ||
2820 13289 : parse->groupClause ||
2821 13274 : parse->groupingSets ||
2822 13274 : root->hasHavingQual ||
2823 13274 : parse->distinctClause ||
2824 18204 : parse->sortClause ||
2825 5264 : bms_membership(root->all_baserels) == BMS_MULTIPLE)
2826 13572 : tuple_fraction = 0.0; /* default case */
2827 : else
2828 3785 : tuple_fraction = root->tuple_fraction;
2829 :
2830 : /* plan_params should not be in use in current query level */
2831 : Assert(root->plan_params == NIL);
2832 :
2833 : /* Generate a subroot and Paths for the subquery */
2834 17357 : plan_name = choose_plan_name(root->glob, rte->eref->aliasname, false);
2835 17357 : rel->subroot = subquery_planner(root->glob, subquery, plan_name,
2836 : root, NULL, false, tuple_fraction, NULL);
2837 :
2838 : /* Isolate the params needed by this specific subplan */
2839 17357 : rel->subplan_params = root->plan_params;
2840 17357 : root->plan_params = NIL;
2841 :
2842 : /*
2843 : * It's possible that constraint exclusion proved the subquery empty. If
2844 : * so, it's desirable to produce an unadorned dummy path so that we will
2845 : * recognize appropriate optimizations at this query level.
2846 : */
2847 17357 : sub_final_rel = fetch_upper_rel(rel->subroot, UPPERREL_FINAL, NULL);
2848 :
2849 17357 : if (IS_DUMMY_REL(sub_final_rel))
2850 : {
2851 105 : set_dummy_rel_pathlist(rel);
2852 105 : return;
2853 : }
2854 :
2855 : /*
2856 : * Mark rel with estimated output rows, width, etc. Note that we have to
2857 : * do this before generating outer-query paths, else cost_subqueryscan is
2858 : * not happy.
2859 : */
2860 17252 : set_subquery_size_estimates(root, rel);
2861 :
2862 : /*
2863 : * Also detect whether the reltarget is trivial, so that we can pass that
2864 : * info to cost_subqueryscan (rather than re-deriving it multiple times).
2865 : * It's trivial if it fetches all the subplan output columns in order.
2866 : */
2867 17252 : if (list_length(rel->reltarget->exprs) != list_length(subquery->targetList))
2868 9561 : trivial_pathtarget = false;
2869 : else
2870 : {
2871 7691 : trivial_pathtarget = true;
2872 21364 : foreach(lc, rel->reltarget->exprs)
2873 : {
2874 13930 : Node *node = (Node *) lfirst(lc);
2875 : Var *var;
2876 :
2877 13930 : if (!IsA(node, Var))
2878 : {
2879 0 : trivial_pathtarget = false;
2880 0 : break;
2881 : }
2882 13930 : var = (Var *) node;
2883 13930 : if (var->varno != rti ||
2884 13930 : var->varattno != foreach_current_index(lc) + 1)
2885 : {
2886 257 : trivial_pathtarget = false;
2887 257 : break;
2888 : }
2889 : }
2890 : }
2891 :
2892 : /*
2893 : * For each Path that subquery_planner produced, make a SubqueryScanPath
2894 : * in the outer query.
2895 : */
2896 35859 : foreach(lc, sub_final_rel->pathlist)
2897 : {
2898 18607 : Path *subpath = (Path *) lfirst(lc);
2899 : List *pathkeys;
2900 :
2901 : /* Convert subpath's pathkeys to outer representation */
2902 18607 : pathkeys = convert_subquery_pathkeys(root,
2903 : rel,
2904 : subpath->pathkeys,
2905 : make_tlist_from_pathtarget(subpath->pathtarget));
2906 :
2907 : /* Generate outer path using this subpath */
2908 18607 : add_path(rel, (Path *)
2909 18607 : create_subqueryscan_path(root, rel, subpath,
2910 : trivial_pathtarget,
2911 : pathkeys, required_outer));
2912 : }
2913 :
2914 : /* If outer rel allows parallelism, do same for partial paths. */
2915 17252 : if (rel->consider_parallel && bms_is_empty(required_outer))
2916 : {
2917 : /* If consider_parallel is false, there should be no partial paths. */
2918 : Assert(sub_final_rel->consider_parallel ||
2919 : sub_final_rel->partial_pathlist == NIL);
2920 :
2921 : /* Same for partial paths. */
2922 10754 : foreach(lc, sub_final_rel->partial_pathlist)
2923 : {
2924 45 : Path *subpath = (Path *) lfirst(lc);
2925 : List *pathkeys;
2926 :
2927 : /* Convert subpath's pathkeys to outer representation */
2928 45 : pathkeys = convert_subquery_pathkeys(root,
2929 : rel,
2930 : subpath->pathkeys,
2931 : make_tlist_from_pathtarget(subpath->pathtarget));
2932 :
2933 : /* Generate outer path using this subpath */
2934 45 : add_partial_path(rel, (Path *)
2935 45 : create_subqueryscan_path(root, rel, subpath,
2936 : trivial_pathtarget,
2937 : pathkeys,
2938 : required_outer));
2939 : }
2940 : }
2941 : }
2942 :
2943 : /*
2944 : * set_function_pathlist
2945 : * Build the (single) access path for a function RTE
2946 : */
2947 : static void
2948 35645 : set_function_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2949 : {
2950 : Relids required_outer;
2951 35645 : List *pathkeys = NIL;
2952 :
2953 : /*
2954 : * We don't support pushing join clauses into the quals of a function
2955 : * scan, but it could still have required parameterization due to LATERAL
2956 : * refs in the function expression.
2957 : */
2958 35645 : required_outer = rel->lateral_relids;
2959 :
2960 : /*
2961 : * The result is considered unordered unless ORDINALITY was used, in which
2962 : * case it is ordered by the ordinal column (the last one). See if we
2963 : * care, by checking for uses of that Var in equivalence classes.
2964 : */
2965 35645 : if (rte->funcordinality)
2966 : {
2967 770 : AttrNumber ordattno = rel->max_attr;
2968 770 : Var *var = NULL;
2969 : ListCell *lc;
2970 :
2971 : /*
2972 : * Is there a Var for it in rel's targetlist? If not, the query did
2973 : * not reference the ordinality column, or at least not in any way
2974 : * that would be interesting for sorting.
2975 : */
2976 1787 : foreach(lc, rel->reltarget->exprs)
2977 : {
2978 1782 : Var *node = (Var *) lfirst(lc);
2979 :
2980 : /* checking varno/varlevelsup is just paranoia */
2981 1782 : if (IsA(node, Var) &&
2982 1782 : node->varattno == ordattno &&
2983 765 : node->varno == rel->relid &&
2984 765 : node->varlevelsup == 0)
2985 : {
2986 765 : var = node;
2987 765 : break;
2988 : }
2989 : }
2990 :
2991 : /*
2992 : * Try to build pathkeys for this Var with int8 sorting. We tell
2993 : * build_expression_pathkey not to build any new equivalence class; if
2994 : * the Var isn't already mentioned in some EC, it means that nothing
2995 : * cares about the ordering.
2996 : */
2997 770 : if (var)
2998 765 : pathkeys = build_expression_pathkey(root,
2999 : (Expr *) var,
3000 : Int8LessOperator,
3001 : rel->relids,
3002 : false);
3003 : }
3004 :
3005 : /* Generate appropriate path */
3006 35645 : add_path(rel, create_functionscan_path(root, rel,
3007 : pathkeys, required_outer));
3008 35645 : }
3009 :
3010 : /*
3011 : * set_values_pathlist
3012 : * Build the (single) access path for a VALUES RTE
3013 : */
3014 : static void
3015 6793 : set_values_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
3016 : {
3017 : Relids required_outer;
3018 :
3019 : /*
3020 : * We don't support pushing join clauses into the quals of a values scan,
3021 : * but it could still have required parameterization due to LATERAL refs
3022 : * in the values expressions.
3023 : */
3024 6793 : required_outer = rel->lateral_relids;
3025 :
3026 : /* Generate appropriate path */
3027 6793 : add_path(rel, create_valuesscan_path(root, rel, required_outer));
3028 6793 : }
3029 :
3030 : /*
3031 : * set_tablefunc_pathlist
3032 : * Build the (single) access path for a table func RTE
3033 : */
3034 : static void
3035 519 : set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
3036 : {
3037 : Relids required_outer;
3038 :
3039 : /*
3040 : * We don't support pushing join clauses into the quals of a tablefunc
3041 : * scan, but it could still have required parameterization due to LATERAL
3042 : * refs in the function expression.
3043 : */
3044 519 : required_outer = rel->lateral_relids;
3045 :
3046 : /* Generate appropriate path */
3047 519 : add_path(rel, create_tablefuncscan_path(root, rel,
3048 : required_outer));
3049 519 : }
3050 :
3051 : /*
3052 : * set_cte_pathlist
3053 : * Build the (single) access path for a non-self-reference CTE RTE
3054 : *
3055 : * There's no need for a separate set_cte_size phase, since we don't
3056 : * support join-qual-parameterized paths for CTEs.
3057 : */
3058 : static void
3059 2928 : set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
3060 : {
3061 : Path *ctepath;
3062 : Plan *cteplan;
3063 : PlannerInfo *cteroot;
3064 : Index levelsup;
3065 : List *pathkeys;
3066 : int ndx;
3067 : ListCell *lc;
3068 : int plan_id;
3069 : Relids required_outer;
3070 :
3071 : /*
3072 : * Find the referenced CTE, and locate the path and plan previously made
3073 : * for it.
3074 : */
3075 2928 : levelsup = rte->ctelevelsup;
3076 2928 : cteroot = root;
3077 4994 : while (levelsup-- > 0)
3078 : {
3079 2066 : cteroot = cteroot->parent_root;
3080 2066 : if (!cteroot) /* shouldn't happen */
3081 0 : elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
3082 : }
3083 :
3084 : /*
3085 : * Note: cte_plan_ids can be shorter than cteList, if we are still working
3086 : * on planning the CTEs (ie, this is a side-reference from another CTE).
3087 : * So we mustn't use forboth here.
3088 : */
3089 2928 : ndx = 0;
3090 3884 : foreach(lc, cteroot->parse->cteList)
3091 : {
3092 3884 : CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
3093 :
3094 3884 : if (strcmp(cte->ctename, rte->ctename) == 0)
3095 2928 : break;
3096 956 : ndx++;
3097 : }
3098 2928 : if (lc == NULL) /* shouldn't happen */
3099 0 : elog(ERROR, "could not find CTE \"%s\"", rte->ctename);
3100 2928 : if (ndx >= list_length(cteroot->cte_plan_ids))
3101 0 : elog(ERROR, "could not find plan for CTE \"%s\"", rte->ctename);
3102 2928 : plan_id = list_nth_int(cteroot->cte_plan_ids, ndx);
3103 2928 : if (plan_id <= 0)
3104 0 : elog(ERROR, "no plan was made for CTE \"%s\"", rte->ctename);
3105 :
3106 : Assert(list_length(root->glob->subpaths) == list_length(root->glob->subplans));
3107 2928 : ctepath = (Path *) list_nth(root->glob->subpaths, plan_id - 1);
3108 2928 : cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
3109 :
3110 : /* Mark rel with estimated output rows, width, etc */
3111 2928 : set_cte_size_estimates(root, rel, cteplan->plan_rows);
3112 :
3113 : /* Convert the ctepath's pathkeys to outer query's representation */
3114 2928 : pathkeys = convert_subquery_pathkeys(root,
3115 : rel,
3116 : ctepath->pathkeys,
3117 : cteplan->targetlist);
3118 :
3119 : /*
3120 : * We don't support pushing join clauses into the quals of a CTE scan, but
3121 : * it could still have required parameterization due to LATERAL refs in
3122 : * its tlist.
3123 : */
3124 2928 : required_outer = rel->lateral_relids;
3125 :
3126 : /* Generate appropriate path */
3127 2928 : add_path(rel, create_ctescan_path(root, rel, pathkeys, required_outer));
3128 2928 : }
3129 :
3130 : /*
3131 : * set_namedtuplestore_pathlist
3132 : * Build the (single) access path for a named tuplestore RTE
3133 : *
3134 : * There's no need for a separate set_namedtuplestore_size phase, since we
3135 : * don't support join-qual-parameterized paths for tuplestores.
3136 : */
3137 : static void
3138 433 : set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel,
3139 : RangeTblEntry *rte)
3140 : {
3141 : Relids required_outer;
3142 :
3143 : /* Mark rel with estimated output rows, width, etc */
3144 433 : set_namedtuplestore_size_estimates(root, rel);
3145 :
3146 : /*
3147 : * We don't support pushing join clauses into the quals of a tuplestore
3148 : * scan, but it could still have required parameterization due to LATERAL
3149 : * refs in its tlist.
3150 : */
3151 433 : required_outer = rel->lateral_relids;
3152 :
3153 : /* Generate appropriate path */
3154 433 : add_path(rel, create_namedtuplestorescan_path(root, rel, required_outer));
3155 433 : }
3156 :
3157 : /*
3158 : * set_result_pathlist
3159 : * Build the (single) access path for an RTE_RESULT RTE
3160 : *
3161 : * There's no need for a separate set_result_size phase, since we
3162 : * don't support join-qual-parameterized paths for these RTEs.
3163 : */
3164 : static void
3165 3595 : set_result_pathlist(PlannerInfo *root, RelOptInfo *rel,
3166 : RangeTblEntry *rte)
3167 : {
3168 : Relids required_outer;
3169 :
3170 : /* Mark rel with estimated output rows, width, etc */
3171 3595 : set_result_size_estimates(root, rel);
3172 :
3173 : /*
3174 : * We don't support pushing join clauses into the quals of a Result scan,
3175 : * but it could still have required parameterization due to LATERAL refs
3176 : * in its tlist.
3177 : */
3178 3595 : required_outer = rel->lateral_relids;
3179 :
3180 : /* Generate appropriate path */
3181 3595 : add_path(rel, create_resultscan_path(root, rel, required_outer));
3182 3595 : }
3183 :
3184 : /*
3185 : * set_worktable_pathlist
3186 : * Build the (single) access path for a self-reference CTE RTE
3187 : *
3188 : * There's no need for a separate set_worktable_size phase, since we don't
3189 : * support join-qual-parameterized paths for CTEs.
3190 : */
3191 : static void
3192 692 : set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
3193 : {
3194 : Path *ctepath;
3195 : PlannerInfo *cteroot;
3196 : Index levelsup;
3197 : Relids required_outer;
3198 :
3199 : /*
3200 : * We need to find the non-recursive term's path, which is in the plan
3201 : * level that's processing the recursive UNION, which is one level *below*
3202 : * where the CTE comes from.
3203 : */
3204 692 : levelsup = rte->ctelevelsup;
3205 692 : if (levelsup == 0) /* shouldn't happen */
3206 0 : elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
3207 692 : levelsup--;
3208 692 : cteroot = root;
3209 1626 : while (levelsup-- > 0)
3210 : {
3211 934 : cteroot = cteroot->parent_root;
3212 934 : if (!cteroot) /* shouldn't happen */
3213 0 : elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
3214 : }
3215 692 : ctepath = cteroot->non_recursive_path;
3216 692 : if (!ctepath) /* shouldn't happen */
3217 0 : elog(ERROR, "could not find path for CTE \"%s\"", rte->ctename);
3218 :
3219 : /* Mark rel with estimated output rows, width, etc */
3220 692 : set_cte_size_estimates(root, rel, ctepath->rows);
3221 :
3222 : /*
3223 : * We don't support pushing join clauses into the quals of a worktable
3224 : * scan, but it could still have required parameterization due to LATERAL
3225 : * refs in its tlist. (I'm not sure this is actually possible given the
3226 : * restrictions on recursive references, but it's easy enough to support.)
3227 : */
3228 692 : required_outer = rel->lateral_relids;
3229 :
3230 : /* Generate appropriate path */
3231 692 : add_path(rel, create_worktablescan_path(root, rel, required_outer));
3232 692 : }
3233 :
3234 : /*
3235 : * generate_gather_paths
3236 : * Generate parallel access paths for a relation by pushing a Gather or
3237 : * Gather Merge on top of a partial path.
3238 : *
3239 : * This must not be called until after we're done creating all partial paths
3240 : * for the specified relation. (Otherwise, add_partial_path might delete a
3241 : * path that some GatherPath or GatherMergePath has a reference to.)
3242 : *
3243 : * If we're generating paths for a scan or join relation, override_rows will
3244 : * be false, and we'll just use the relation's size estimate. When we're
3245 : * being called for a partially-grouped or partially-distinct path, though, we
3246 : * need to override the rowcount estimate. (It's not clear that the
3247 : * particular value we're using here is actually best, but the underlying rel
3248 : * has no estimate so we must do something.)
3249 : */
3250 : void
3251 21875 : generate_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
3252 : {
3253 : Path *cheapest_partial_path;
3254 : Path *simple_gather_path;
3255 : ListCell *lc;
3256 : double rows;
3257 21875 : double *rowsp = NULL;
3258 :
3259 : /* If there are no partial paths, there's nothing to do here. */
3260 21875 : if (rel->partial_pathlist == NIL)
3261 0 : return;
3262 :
3263 : /* Should we override the rel's rowcount estimate? */
3264 21875 : if (override_rows)
3265 5810 : rowsp = &rows;
3266 :
3267 : /*
3268 : * The output of Gather is always unsorted, so there's only one partial
3269 : * path of interest: the cheapest one. That will be the one at the front
3270 : * of partial_pathlist because of the way add_partial_path works.
3271 : */
3272 21875 : cheapest_partial_path = linitial(rel->partial_pathlist);
3273 21875 : rows = compute_gather_rows(cheapest_partial_path);
3274 : simple_gather_path = (Path *)
3275 21875 : create_gather_path(root, rel, cheapest_partial_path, rel->reltarget,
3276 : NULL, rowsp);
3277 21875 : add_path(rel, simple_gather_path);
3278 :
3279 : /*
3280 : * For each useful ordering, we can consider an order-preserving Gather
3281 : * Merge.
3282 : */
3283 48737 : foreach(lc, rel->partial_pathlist)
3284 : {
3285 26862 : Path *subpath = (Path *) lfirst(lc);
3286 : GatherMergePath *path;
3287 :
3288 26862 : if (subpath->pathkeys == NIL)
3289 21316 : continue;
3290 :
3291 5546 : rows = compute_gather_rows(subpath);
3292 5546 : path = create_gather_merge_path(root, rel, subpath, rel->reltarget,
3293 : subpath->pathkeys, NULL, rowsp);
3294 5546 : add_path(rel, &path->path);
3295 : }
3296 : }
3297 :
3298 : /*
3299 : * get_useful_pathkeys_for_relation
3300 : * Determine which orderings of a relation might be useful.
3301 : *
3302 : * Getting data in sorted order can be useful either because the requested
3303 : * order matches the final output ordering for the overall query we're
3304 : * planning, or because it enables an efficient merge join. Here, we try
3305 : * to figure out which pathkeys to consider.
3306 : *
3307 : * This allows us to do incremental sort on top of an index scan under a gather
3308 : * merge node, i.e. parallelized.
3309 : *
3310 : * If the require_parallel_safe is true, we also require the expressions to
3311 : * be parallel safe (which allows pushing the sort below Gather Merge).
3312 : *
3313 : * XXX At the moment this can only ever return a list with a single element,
3314 : * because it looks at query_pathkeys only. So we might return the pathkeys
3315 : * directly, but it seems plausible we'll want to consider other orderings
3316 : * in the future. For example, we might want to consider pathkeys useful for
3317 : * merge joins.
3318 : */
3319 : static List *
3320 21875 : get_useful_pathkeys_for_relation(PlannerInfo *root, RelOptInfo *rel,
3321 : bool require_parallel_safe)
3322 : {
3323 21875 : List *useful_pathkeys_list = NIL;
3324 :
3325 : /*
3326 : * Considering query_pathkeys is always worth it, because it might allow
3327 : * us to avoid a total sort when we have a partially presorted path
3328 : * available or to push the total sort into the parallel portion of the
3329 : * query.
3330 : */
3331 21875 : if (root->query_pathkeys)
3332 : {
3333 : ListCell *lc;
3334 13155 : int npathkeys = 0; /* useful pathkeys */
3335 :
3336 23188 : foreach(lc, root->query_pathkeys)
3337 : {
3338 16962 : PathKey *pathkey = (PathKey *) lfirst(lc);
3339 16962 : EquivalenceClass *pathkey_ec = pathkey->pk_eclass;
3340 :
3341 : /*
3342 : * We can only build a sort for pathkeys that contain a
3343 : * safe-to-compute-early EC member computable from the current
3344 : * relation's reltarget, so ignore the remainder of the list as
3345 : * soon as we find a pathkey without such a member.
3346 : *
3347 : * It's still worthwhile to return any prefix of the pathkeys list
3348 : * that meets this requirement, as we may be able to do an
3349 : * incremental sort.
3350 : *
3351 : * If requested, ensure the sort expression is parallel-safe too.
3352 : */
3353 16962 : if (!relation_can_be_sorted_early(root, rel, pathkey_ec,
3354 : require_parallel_safe))
3355 6929 : break;
3356 :
3357 10033 : npathkeys++;
3358 : }
3359 :
3360 : /*
3361 : * The whole query_pathkeys list matches, so append it directly, to
3362 : * allow comparing pathkeys easily by comparing list pointer. If we
3363 : * have to truncate the pathkeys, we gotta do a copy though.
3364 : */
3365 13155 : if (npathkeys == list_length(root->query_pathkeys))
3366 6226 : useful_pathkeys_list = lappend(useful_pathkeys_list,
3367 6226 : root->query_pathkeys);
3368 6929 : else if (npathkeys > 0)
3369 545 : useful_pathkeys_list = lappend(useful_pathkeys_list,
3370 545 : list_copy_head(root->query_pathkeys,
3371 : npathkeys));
3372 : }
3373 :
3374 21875 : return useful_pathkeys_list;
3375 : }
3376 :
3377 : /*
3378 : * generate_useful_gather_paths
3379 : * Generate parallel access paths for a relation by pushing a Gather or
3380 : * Gather Merge on top of a partial path.
3381 : *
3382 : * Unlike plain generate_gather_paths, this looks both at pathkeys of input
3383 : * paths (aiming to preserve the ordering), but also considers ordering that
3384 : * might be useful for nodes above the gather merge node, and tries to add
3385 : * a sort (regular or incremental) to provide that.
3386 : */
3387 : void
3388 486235 : generate_useful_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
3389 : {
3390 : ListCell *lc;
3391 : double rows;
3392 486235 : double *rowsp = NULL;
3393 486235 : List *useful_pathkeys_list = NIL;
3394 486235 : Path *cheapest_partial_path = NULL;
3395 :
3396 : /* If there are no partial paths, there's nothing to do here. */
3397 486235 : if (rel->partial_pathlist == NIL)
3398 464360 : return;
3399 :
3400 : /* Should we override the rel's rowcount estimate? */
3401 21875 : if (override_rows)
3402 5810 : rowsp = &rows;
3403 :
3404 : /* generate the regular gather (merge) paths */
3405 21875 : generate_gather_paths(root, rel, override_rows);
3406 :
3407 : /* consider incremental sort for interesting orderings */
3408 21875 : useful_pathkeys_list = get_useful_pathkeys_for_relation(root, rel, true);
3409 :
3410 : /* used for explicit (full) sort paths */
3411 21875 : cheapest_partial_path = linitial(rel->partial_pathlist);
3412 :
3413 : /*
3414 : * Consider sorted paths for each interesting ordering. We generate both
3415 : * incremental and full sort.
3416 : */
3417 28646 : foreach(lc, useful_pathkeys_list)
3418 : {
3419 6771 : List *useful_pathkeys = lfirst(lc);
3420 : ListCell *lc2;
3421 : bool is_sorted;
3422 : int presorted_keys;
3423 :
3424 15771 : foreach(lc2, rel->partial_pathlist)
3425 : {
3426 9000 : Path *subpath = (Path *) lfirst(lc2);
3427 : GatherMergePath *path;
3428 :
3429 9000 : is_sorted = pathkeys_count_contained_in(useful_pathkeys,
3430 : subpath->pathkeys,
3431 : &presorted_keys);
3432 :
3433 : /*
3434 : * We don't need to consider the case where a subpath is already
3435 : * fully sorted because generate_gather_paths already creates a
3436 : * gather merge path for every subpath that has pathkeys present.
3437 : *
3438 : * But since the subpath is already sorted, we know we don't need
3439 : * to consider adding a sort (full or incremental) on top of it,
3440 : * so we can continue here.
3441 : */
3442 9000 : if (is_sorted)
3443 2380 : continue;
3444 :
3445 : /*
3446 : * Try at least sorting the cheapest path and also try
3447 : * incrementally sorting any path which is partially sorted
3448 : * already (no need to deal with paths which have presorted keys
3449 : * when incremental sort is disabled unless it's the cheapest
3450 : * input path).
3451 : */
3452 6620 : if (subpath != cheapest_partial_path &&
3453 260 : (presorted_keys == 0 || !enable_incremental_sort))
3454 73 : continue;
3455 :
3456 : /*
3457 : * Consider regular sort for any path that's not presorted or if
3458 : * incremental sort is disabled. We've no need to consider both
3459 : * sort and incremental sort on the same path. We assume that
3460 : * incremental sort is always faster when there are presorted
3461 : * keys.
3462 : *
3463 : * This is not redundant with the gather paths created in
3464 : * generate_gather_paths, because that doesn't generate ordered
3465 : * output. Here we add an explicit sort to match the useful
3466 : * ordering.
3467 : */
3468 6547 : if (presorted_keys == 0 || !enable_incremental_sort)
3469 : {
3470 6349 : subpath = (Path *) create_sort_path(root,
3471 : rel,
3472 : subpath,
3473 : useful_pathkeys,
3474 : -1.0);
3475 : }
3476 : else
3477 198 : subpath = (Path *) create_incremental_sort_path(root,
3478 : rel,
3479 : subpath,
3480 : useful_pathkeys,
3481 : presorted_keys,
3482 : -1);
3483 6547 : rows = compute_gather_rows(subpath);
3484 6547 : path = create_gather_merge_path(root, rel,
3485 : subpath,
3486 6547 : rel->reltarget,
3487 : subpath->pathkeys,
3488 : NULL,
3489 : rowsp);
3490 :
3491 6547 : add_path(rel, &path->path);
3492 : }
3493 : }
3494 : }
3495 :
3496 : /*
3497 : * generate_grouped_paths
3498 : * Generate paths for a grouped relation by adding sorted and hashed
3499 : * partial aggregation paths on top of paths of the ungrouped relation.
3500 : *
3501 : * The information needed is provided by the RelAggInfo structure stored in
3502 : * "grouped_rel".
3503 : */
3504 : void
3505 727 : generate_grouped_paths(PlannerInfo *root, RelOptInfo *grouped_rel,
3506 : RelOptInfo *rel)
3507 : {
3508 727 : RelAggInfo *agg_info = grouped_rel->agg_info;
3509 : AggClauseCosts agg_costs;
3510 : bool can_hash;
3511 : bool can_sort;
3512 727 : Path *cheapest_total_path = NULL;
3513 727 : Path *cheapest_partial_path = NULL;
3514 727 : double dNumGroups = 0;
3515 727 : double dNumPartialGroups = 0;
3516 727 : List *group_pathkeys = NIL;
3517 :
3518 727 : if (IS_DUMMY_REL(rel))
3519 : {
3520 0 : mark_dummy_rel(grouped_rel);
3521 0 : return;
3522 : }
3523 :
3524 : /*
3525 : * We push partial aggregation only to the lowest possible level in the
3526 : * join tree that is deemed useful.
3527 : */
3528 727 : if (!bms_equal(agg_info->apply_agg_at, rel->relids) ||
3529 727 : !agg_info->agg_useful)
3530 0 : return;
3531 :
3532 4362 : MemSet(&agg_costs, 0, sizeof(AggClauseCosts));
3533 727 : get_agg_clause_costs(root, AGGSPLIT_INITIAL_SERIAL, &agg_costs);
3534 :
3535 : /*
3536 : * Determine whether it's possible to perform sort-based implementations
3537 : * of grouping, and generate the pathkeys that represent the grouping
3538 : * requirements in that case.
3539 : */
3540 727 : can_sort = grouping_is_sortable(agg_info->group_clauses);
3541 727 : if (can_sort)
3542 : {
3543 : RelOptInfo *top_grouped_rel;
3544 : List *top_group_tlist;
3545 :
3546 397 : top_grouped_rel = IS_OTHER_REL(rel) ?
3547 1124 : rel->top_parent->grouped_rel : grouped_rel;
3548 : top_group_tlist =
3549 727 : make_tlist_from_pathtarget(top_grouped_rel->agg_info->target);
3550 :
3551 : group_pathkeys =
3552 727 : make_pathkeys_for_sortclauses(root, agg_info->group_clauses,
3553 : top_group_tlist);
3554 : }
3555 :
3556 : /*
3557 : * Determine whether we should consider hash-based implementations of
3558 : * grouping.
3559 : */
3560 : Assert(root->numOrderedAggs == 0);
3561 1454 : can_hash = (agg_info->group_clauses != NIL &&
3562 727 : grouping_is_hashable(agg_info->group_clauses));
3563 :
3564 : /*
3565 : * Consider whether we should generate partially aggregated non-partial
3566 : * paths. We can only do this if we have a non-partial path.
3567 : */
3568 727 : if (rel->pathlist != NIL)
3569 : {
3570 727 : cheapest_total_path = rel->cheapest_total_path;
3571 : Assert(cheapest_total_path != NULL);
3572 : }
3573 :
3574 : /*
3575 : * If parallelism is possible for grouped_rel, then we should consider
3576 : * generating partially-grouped partial paths. However, if the ungrouped
3577 : * rel has no partial paths, then we can't.
3578 : */
3579 727 : if (grouped_rel->consider_parallel && rel->partial_pathlist != NIL)
3580 : {
3581 610 : cheapest_partial_path = linitial(rel->partial_pathlist);
3582 : Assert(cheapest_partial_path != NULL);
3583 : }
3584 :
3585 : /* Estimate number of partial groups. */
3586 727 : if (cheapest_total_path != NULL)
3587 727 : dNumGroups = estimate_num_groups(root,
3588 : agg_info->group_exprs,
3589 : cheapest_total_path->rows,
3590 : NULL, NULL);
3591 727 : if (cheapest_partial_path != NULL)
3592 610 : dNumPartialGroups = estimate_num_groups(root,
3593 : agg_info->group_exprs,
3594 : cheapest_partial_path->rows,
3595 : NULL, NULL);
3596 :
3597 727 : if (can_sort && cheapest_total_path != NULL)
3598 : {
3599 : ListCell *lc;
3600 :
3601 : /*
3602 : * Use any available suitably-sorted path as input, and also consider
3603 : * sorting the cheapest-total path and incremental sort on any paths
3604 : * with presorted keys.
3605 : *
3606 : * To save planning time, we ignore parameterized input paths unless
3607 : * they are the cheapest-total path.
3608 : */
3609 1735 : foreach(lc, rel->pathlist)
3610 : {
3611 1008 : Path *input_path = (Path *) lfirst(lc);
3612 : Path *path;
3613 : bool is_sorted;
3614 : int presorted_keys;
3615 :
3616 : /*
3617 : * Ignore parameterized paths that are not the cheapest-total
3618 : * path.
3619 : */
3620 1008 : if (input_path->param_info &&
3621 : input_path != cheapest_total_path)
3622 23 : continue;
3623 :
3624 1003 : is_sorted = pathkeys_count_contained_in(group_pathkeys,
3625 : input_path->pathkeys,
3626 : &presorted_keys);
3627 :
3628 : /*
3629 : * Ignore paths that are not suitably or partially sorted, unless
3630 : * they are the cheapest total path (no need to deal with paths
3631 : * which have presorted keys when incremental sort is disabled).
3632 : */
3633 1003 : if (!is_sorted && input_path != cheapest_total_path &&
3634 122 : (presorted_keys == 0 || !enable_incremental_sort))
3635 18 : continue;
3636 :
3637 : /*
3638 : * Since the path originates from a non-grouped relation that is
3639 : * not aware of eager aggregation, we must ensure that it provides
3640 : * the correct input for partial aggregation.
3641 : */
3642 985 : path = (Path *) create_projection_path(root,
3643 : grouped_rel,
3644 : input_path,
3645 985 : agg_info->agg_input);
3646 :
3647 985 : if (!is_sorted)
3648 : {
3649 : /*
3650 : * We've no need to consider both a sort and incremental sort.
3651 : * We'll just do a sort if there are no presorted keys and an
3652 : * incremental sort when there are presorted keys.
3653 : */
3654 826 : if (presorted_keys == 0 || !enable_incremental_sort)
3655 716 : path = (Path *) create_sort_path(root,
3656 : grouped_rel,
3657 : path,
3658 : group_pathkeys,
3659 : -1.0);
3660 : else
3661 110 : path = (Path *) create_incremental_sort_path(root,
3662 : grouped_rel,
3663 : path,
3664 : group_pathkeys,
3665 : presorted_keys,
3666 : -1.0);
3667 : }
3668 :
3669 : /*
3670 : * qual is NIL because the HAVING clause cannot be evaluated until
3671 : * the final value of the aggregate is known.
3672 : */
3673 985 : path = (Path *) create_agg_path(root,
3674 : grouped_rel,
3675 : path,
3676 985 : agg_info->target,
3677 : AGG_SORTED,
3678 : AGGSPLIT_INITIAL_SERIAL,
3679 : agg_info->group_clauses,
3680 : NIL,
3681 : &agg_costs,
3682 : dNumGroups);
3683 :
3684 985 : add_path(grouped_rel, path);
3685 : }
3686 : }
3687 :
3688 727 : if (can_sort && cheapest_partial_path != NULL)
3689 : {
3690 : ListCell *lc;
3691 :
3692 : /* Similar to above logic, but for partial paths. */
3693 1394 : foreach(lc, rel->partial_pathlist)
3694 : {
3695 784 : Path *input_path = (Path *) lfirst(lc);
3696 : Path *path;
3697 : bool is_sorted;
3698 : int presorted_keys;
3699 :
3700 784 : is_sorted = pathkeys_count_contained_in(group_pathkeys,
3701 : input_path->pathkeys,
3702 : &presorted_keys);
3703 :
3704 : /*
3705 : * Ignore paths that are not suitably or partially sorted, unless
3706 : * they are the cheapest partial path (no need to deal with paths
3707 : * which have presorted keys when incremental sort is disabled).
3708 : */
3709 784 : if (!is_sorted && input_path != cheapest_partial_path &&
3710 70 : (presorted_keys == 0 || !enable_incremental_sort))
3711 0 : continue;
3712 :
3713 : /*
3714 : * Since the path originates from a non-grouped relation that is
3715 : * not aware of eager aggregation, we must ensure that it provides
3716 : * the correct input for partial aggregation.
3717 : */
3718 784 : path = (Path *) create_projection_path(root,
3719 : grouped_rel,
3720 : input_path,
3721 784 : agg_info->agg_input);
3722 :
3723 784 : if (!is_sorted)
3724 : {
3725 : /*
3726 : * We've no need to consider both a sort and incremental sort.
3727 : * We'll just do a sort if there are no presorted keys and an
3728 : * incremental sort when there are presorted keys.
3729 : */
3730 680 : if (presorted_keys == 0 || !enable_incremental_sort)
3731 610 : path = (Path *) create_sort_path(root,
3732 : grouped_rel,
3733 : path,
3734 : group_pathkeys,
3735 : -1.0);
3736 : else
3737 70 : path = (Path *) create_incremental_sort_path(root,
3738 : grouped_rel,
3739 : path,
3740 : group_pathkeys,
3741 : presorted_keys,
3742 : -1.0);
3743 : }
3744 :
3745 : /*
3746 : * qual is NIL because the HAVING clause cannot be evaluated until
3747 : * the final value of the aggregate is known.
3748 : */
3749 784 : path = (Path *) create_agg_path(root,
3750 : grouped_rel,
3751 : path,
3752 784 : agg_info->target,
3753 : AGG_SORTED,
3754 : AGGSPLIT_INITIAL_SERIAL,
3755 : agg_info->group_clauses,
3756 : NIL,
3757 : &agg_costs,
3758 : dNumPartialGroups);
3759 :
3760 784 : add_partial_path(grouped_rel, path);
3761 : }
3762 : }
3763 :
3764 : /*
3765 : * Add a partially-grouped HashAgg Path where possible
3766 : */
3767 727 : if (can_hash && cheapest_total_path != NULL)
3768 : {
3769 : Path *path;
3770 :
3771 : /*
3772 : * Since the path originates from a non-grouped relation that is not
3773 : * aware of eager aggregation, we must ensure that it provides the
3774 : * correct input for partial aggregation.
3775 : */
3776 727 : path = (Path *) create_projection_path(root,
3777 : grouped_rel,
3778 : cheapest_total_path,
3779 727 : agg_info->agg_input);
3780 :
3781 : /*
3782 : * qual is NIL because the HAVING clause cannot be evaluated until the
3783 : * final value of the aggregate is known.
3784 : */
3785 727 : path = (Path *) create_agg_path(root,
3786 : grouped_rel,
3787 : path,
3788 727 : agg_info->target,
3789 : AGG_HASHED,
3790 : AGGSPLIT_INITIAL_SERIAL,
3791 : agg_info->group_clauses,
3792 : NIL,
3793 : &agg_costs,
3794 : dNumGroups);
3795 :
3796 727 : add_path(grouped_rel, path);
3797 : }
3798 :
3799 : /*
3800 : * Now add a partially-grouped HashAgg partial Path where possible
3801 : */
3802 727 : if (can_hash && cheapest_partial_path != NULL)
3803 : {
3804 : Path *path;
3805 :
3806 : /*
3807 : * Since the path originates from a non-grouped relation that is not
3808 : * aware of eager aggregation, we must ensure that it provides the
3809 : * correct input for partial aggregation.
3810 : */
3811 610 : path = (Path *) create_projection_path(root,
3812 : grouped_rel,
3813 : cheapest_partial_path,
3814 610 : agg_info->agg_input);
3815 :
3816 : /*
3817 : * qual is NIL because the HAVING clause cannot be evaluated until the
3818 : * final value of the aggregate is known.
3819 : */
3820 610 : path = (Path *) create_agg_path(root,
3821 : grouped_rel,
3822 : path,
3823 610 : agg_info->target,
3824 : AGG_HASHED,
3825 : AGGSPLIT_INITIAL_SERIAL,
3826 : agg_info->group_clauses,
3827 : NIL,
3828 : &agg_costs,
3829 : dNumPartialGroups);
3830 :
3831 610 : add_partial_path(grouped_rel, path);
3832 : }
3833 : }
3834 :
3835 : /*
3836 : * make_rel_from_joinlist
3837 : * Build access paths using a "joinlist" to guide the join path search.
3838 : *
3839 : * See comments for deconstruct_jointree() for definition of the joinlist
3840 : * data structure.
3841 : */
3842 : static RelOptInfo *
3843 255965 : make_rel_from_joinlist(PlannerInfo *root, List *joinlist)
3844 : {
3845 : int levels_needed;
3846 : List *initial_rels;
3847 : ListCell *jl;
3848 :
3849 : /*
3850 : * Count the number of child joinlist nodes. This is the depth of the
3851 : * dynamic-programming algorithm we must employ to consider all ways of
3852 : * joining the child nodes.
3853 : */
3854 255965 : levels_needed = list_length(joinlist);
3855 :
3856 255965 : if (levels_needed <= 0)
3857 0 : return NULL; /* nothing to do? */
3858 :
3859 : /*
3860 : * Construct a list of rels corresponding to the child joinlist nodes.
3861 : * This may contain both base rels and rels constructed according to
3862 : * sub-joinlists.
3863 : */
3864 255965 : initial_rels = NIL;
3865 623330 : foreach(jl, joinlist)
3866 : {
3867 367365 : Node *jlnode = (Node *) lfirst(jl);
3868 : RelOptInfo *thisrel;
3869 :
3870 367365 : if (IsA(jlnode, RangeTblRef))
3871 : {
3872 364677 : int varno = ((RangeTblRef *) jlnode)->rtindex;
3873 :
3874 364677 : thisrel = find_base_rel(root, varno);
3875 : }
3876 2688 : else if (IsA(jlnode, List))
3877 : {
3878 : /* Recurse to handle subproblem */
3879 2688 : thisrel = make_rel_from_joinlist(root, (List *) jlnode);
3880 : }
3881 : else
3882 : {
3883 0 : elog(ERROR, "unrecognized joinlist node type: %d",
3884 : (int) nodeTag(jlnode));
3885 : thisrel = NULL; /* keep compiler quiet */
3886 : }
3887 :
3888 367365 : initial_rels = lappend(initial_rels, thisrel);
3889 : }
3890 :
3891 255965 : if (levels_needed == 1)
3892 : {
3893 : /*
3894 : * Single joinlist node, so we're done.
3895 : */
3896 174631 : return (RelOptInfo *) linitial(initial_rels);
3897 : }
3898 : else
3899 : {
3900 : /*
3901 : * Consider the different orders in which we could join the rels,
3902 : * using a plugin, GEQO, or the regular join search code.
3903 : *
3904 : * We put the initial_rels list into a PlannerInfo field because
3905 : * has_legal_joinclause() needs to look at it (ugly :-().
3906 : */
3907 81334 : root->initial_rels = initial_rels;
3908 :
3909 81334 : if (join_search_hook)
3910 0 : return (*join_search_hook) (root, levels_needed, initial_rels);
3911 81334 : else if (enable_geqo && levels_needed >= geqo_threshold)
3912 35 : return geqo(root, levels_needed, initial_rels);
3913 : else
3914 81299 : return standard_join_search(root, levels_needed, initial_rels);
3915 : }
3916 : }
3917 :
3918 : /*
3919 : * standard_join_search
3920 : * Find possible joinpaths for a query by successively finding ways
3921 : * to join component relations into join relations.
3922 : *
3923 : * 'levels_needed' is the number of iterations needed, ie, the number of
3924 : * independent jointree items in the query. This is > 1.
3925 : *
3926 : * 'initial_rels' is a list of RelOptInfo nodes for each independent
3927 : * jointree item. These are the components to be joined together.
3928 : * Note that levels_needed == list_length(initial_rels).
3929 : *
3930 : * Returns the final level of join relations, i.e., the relation that is
3931 : * the result of joining all the original relations together.
3932 : * At least one implementation path must be provided for this relation and
3933 : * all required sub-relations.
3934 : *
3935 : * To support loadable plugins that modify planner behavior by changing the
3936 : * join searching algorithm, we provide a hook variable that lets a plugin
3937 : * replace or supplement this function. Any such hook must return the same
3938 : * final join relation as the standard code would, but it might have a
3939 : * different set of implementation paths attached, and only the sub-joinrels
3940 : * needed for these paths need have been instantiated.
3941 : *
3942 : * Note to plugin authors: the functions invoked during standard_join_search()
3943 : * modify root->join_rel_list and root->join_rel_hash. If you want to do more
3944 : * than one join-order search, you'll probably need to save and restore the
3945 : * original states of those data structures. See geqo_eval() for an example.
3946 : */
3947 : RelOptInfo *
3948 81299 : standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
3949 : {
3950 : int lev;
3951 : RelOptInfo *rel;
3952 :
3953 : /*
3954 : * This function cannot be invoked recursively within any one planning
3955 : * problem, so join_rel_level[] can't be in use already.
3956 : */
3957 : Assert(root->join_rel_level == NULL);
3958 :
3959 : /*
3960 : * We employ a simple "dynamic programming" algorithm: we first find all
3961 : * ways to build joins of two jointree items, then all ways to build joins
3962 : * of three items (from two-item joins and single items), then four-item
3963 : * joins, and so on until we have considered all ways to join all the
3964 : * items into one rel.
3965 : *
3966 : * root->join_rel_level[j] is a list of all the j-item rels. Initially we
3967 : * set root->join_rel_level[1] to represent all the single-jointree-item
3968 : * relations.
3969 : */
3970 81299 : root->join_rel_level = (List **) palloc0((levels_needed + 1) * sizeof(List *));
3971 :
3972 81299 : root->join_rel_level[1] = initial_rels;
3973 :
3974 192649 : for (lev = 2; lev <= levels_needed; lev++)
3975 : {
3976 : ListCell *lc;
3977 :
3978 : /*
3979 : * Determine all possible pairs of relations to be joined at this
3980 : * level, and build paths for making each one from every available
3981 : * pair of lower-level relations.
3982 : */
3983 111350 : join_search_one_level(root, lev);
3984 :
3985 : /*
3986 : * Run generate_partitionwise_join_paths() and
3987 : * generate_useful_gather_paths() for each just-processed joinrel. We
3988 : * could not do this earlier because both regular and partial paths
3989 : * can get added to a particular joinrel at multiple times within
3990 : * join_search_one_level.
3991 : *
3992 : * After that, we're done creating paths for the joinrel, so run
3993 : * set_cheapest().
3994 : *
3995 : * In addition, we also run generate_grouped_paths() for the grouped
3996 : * relation of each just-processed joinrel, and run set_cheapest() for
3997 : * the grouped relation afterwards.
3998 : */
3999 285187 : foreach(lc, root->join_rel_level[lev])
4000 : {
4001 : bool is_top_rel;
4002 :
4003 173837 : rel = (RelOptInfo *) lfirst(lc);
4004 :
4005 173837 : is_top_rel = bms_equal(rel->relids, root->all_query_rels);
4006 :
4007 : /* Create paths for partitionwise joins. */
4008 173837 : generate_partitionwise_join_paths(root, rel);
4009 :
4010 : /*
4011 : * Except for the topmost scan/join rel, consider gathering
4012 : * partial paths. We'll do the same for the topmost scan/join rel
4013 : * once we know the final targetlist (see grouping_planner's and
4014 : * its call to apply_scanjoin_target_to_paths).
4015 : */
4016 173837 : if (!is_top_rel)
4017 92947 : generate_useful_gather_paths(root, rel, false);
4018 :
4019 : /* Find and save the cheapest paths for this rel */
4020 173837 : set_cheapest(rel);
4021 :
4022 : /*
4023 : * Except for the topmost scan/join rel, consider generating
4024 : * partial aggregation paths for the grouped relation on top of
4025 : * the paths of this rel. After that, we're done creating paths
4026 : * for the grouped relation, so run set_cheapest().
4027 : */
4028 173837 : if (rel->grouped_rel != NULL && !is_top_rel)
4029 : {
4030 60 : RelOptInfo *grouped_rel = rel->grouped_rel;
4031 :
4032 : Assert(IS_GROUPED_REL(grouped_rel));
4033 :
4034 60 : generate_grouped_paths(root, grouped_rel, rel);
4035 60 : set_cheapest(grouped_rel);
4036 : }
4037 :
4038 : #ifdef OPTIMIZER_DEBUG
4039 : pprint(rel);
4040 : #endif
4041 : }
4042 : }
4043 :
4044 : /*
4045 : * We should have a single rel at the final level.
4046 : */
4047 81299 : if (root->join_rel_level[levels_needed] == NIL)
4048 0 : elog(ERROR, "failed to build any %d-way joins", levels_needed);
4049 : Assert(list_length(root->join_rel_level[levels_needed]) == 1);
4050 :
4051 81299 : rel = (RelOptInfo *) linitial(root->join_rel_level[levels_needed]);
4052 :
4053 81299 : root->join_rel_level = NULL;
4054 :
4055 81299 : return rel;
4056 : }
4057 :
4058 : /*****************************************************************************
4059 : * PUSHING QUALS DOWN INTO SUBQUERIES
4060 : *****************************************************************************/
4061 :
4062 : /*
4063 : * subquery_is_pushdown_safe - is a subquery safe for pushing down quals?
4064 : *
4065 : * subquery is the particular component query being checked. topquery
4066 : * is the top component of a set-operations tree (the same Query if no
4067 : * set-op is involved).
4068 : *
4069 : * Conditions checked here:
4070 : *
4071 : * 1. If the subquery has a LIMIT clause, we must not push down any quals,
4072 : * since that could change the set of rows returned.
4073 : *
4074 : * 2. If the subquery contains EXCEPT or EXCEPT ALL set ops we cannot push
4075 : * quals into it, because that could change the results.
4076 : *
4077 : * 3. If the subquery uses DISTINCT, we cannot push volatile quals into it.
4078 : * This is because upper-level quals should semantically be evaluated only
4079 : * once per distinct row, not once per original row, and if the qual is
4080 : * volatile then extra evaluations could change the results. (This issue
4081 : * does not apply to other forms of aggregation such as GROUP BY, because
4082 : * when those are present we push into HAVING not WHERE, so that the quals
4083 : * are still applied after aggregation.)
4084 : *
4085 : * 4. If the subquery contains window functions, we cannot push volatile quals
4086 : * into it. The issue here is a bit different from DISTINCT: a volatile qual
4087 : * might succeed for some rows of a window partition and fail for others,
4088 : * thereby changing the partition contents and thus the window functions'
4089 : * results for rows that remain.
4090 : *
4091 : * 5. If the subquery contains any set-returning functions in its targetlist,
4092 : * we cannot push volatile quals into it. That would push them below the SRFs
4093 : * and thereby change the number of times they are evaluated. Also, a
4094 : * volatile qual could succeed for some SRF output rows and fail for others,
4095 : * a behavior that cannot occur if it's evaluated before SRF expansion.
4096 : *
4097 : * 6. If the subquery has nonempty grouping sets, we cannot push down any
4098 : * quals. The concern here is that a qual referencing a "constant" grouping
4099 : * column could get constant-folded, which would be improper because the value
4100 : * is potentially nullable by grouping-set expansion. This restriction could
4101 : * be removed if we had a parsetree representation that shows that such
4102 : * grouping columns are not really constant. (There are other ideas that
4103 : * could be used to relax this restriction, but that's the approach most
4104 : * likely to get taken in the future. Note that there's not much to be gained
4105 : * so long as subquery_planner can't move HAVING clauses to WHERE within such
4106 : * a subquery.)
4107 : *
4108 : * In addition, we make several checks on the subquery's output columns to see
4109 : * if it is safe to reference them in pushed-down quals. If output column k
4110 : * is found to be unsafe to reference, we set the reason for that inside
4111 : * safetyInfo->unsafeFlags[k], but we don't reject the subquery overall since
4112 : * column k might not be referenced by some/all quals. The unsafeFlags[]
4113 : * array will be consulted later by qual_is_pushdown_safe(). It's better to
4114 : * do it this way than to make the checks directly in qual_is_pushdown_safe(),
4115 : * because when the subquery involves set operations we have to check the
4116 : * output expressions in each arm of the set op.
4117 : *
4118 : * Note: pushing quals into a DISTINCT subquery is theoretically dubious:
4119 : * we're effectively assuming that the quals cannot distinguish values that
4120 : * the DISTINCT's equality operator sees as equal, yet there are many
4121 : * counterexamples to that assumption. However use of such a qual with a
4122 : * DISTINCT subquery would be unsafe anyway, since there's no guarantee which
4123 : * "equal" value will be chosen as the output value by the DISTINCT operation.
4124 : * So we don't worry too much about that. Another objection is that if the
4125 : * qual is expensive to evaluate, running it for each original row might cost
4126 : * more than we save by eliminating rows before the DISTINCT step. But it
4127 : * would be very hard to estimate that at this stage, and in practice pushdown
4128 : * seldom seems to make things worse, so we ignore that problem too.
4129 : *
4130 : * Note: likewise, pushing quals into a subquery with window functions is a
4131 : * bit dubious: the quals might remove some rows of a window partition while
4132 : * leaving others, causing changes in the window functions' results for the
4133 : * surviving rows. We insist that such a qual reference only partitioning
4134 : * columns, but again that only protects us if the qual does not distinguish
4135 : * values that the partitioning equality operator sees as equal. The risks
4136 : * here are perhaps larger than for DISTINCT, since no de-duplication of rows
4137 : * occurs and thus there is no theoretical problem with such a qual. But
4138 : * we'll do this anyway because the potential performance benefits are very
4139 : * large, and we've seen no field complaints about the longstanding comparable
4140 : * behavior with DISTINCT.
4141 : */
4142 : static bool
4143 2375 : subquery_is_pushdown_safe(Query *subquery, Query *topquery,
4144 : pushdown_safety_info *safetyInfo)
4145 : {
4146 : SetOperationStmt *topop;
4147 :
4148 : /* Check point 1 */
4149 2375 : if (subquery->limitOffset != NULL || subquery->limitCount != NULL)
4150 111 : return false;
4151 :
4152 : /* Check point 6 */
4153 2264 : if (subquery->groupClause && subquery->groupingSets)
4154 10 : return false;
4155 :
4156 : /* Check points 3, 4, and 5 */
4157 2254 : if (subquery->distinctClause ||
4158 2184 : subquery->hasWindowFuncs ||
4159 1963 : subquery->hasTargetSRFs)
4160 706 : safetyInfo->unsafeVolatile = true;
4161 :
4162 : /*
4163 : * If we're at a leaf query, check for unsafe expressions in its target
4164 : * list, and mark any reasons why they're unsafe in unsafeFlags[].
4165 : * (Non-leaf nodes in setop trees have only simple Vars in their tlists,
4166 : * so no need to check them.)
4167 : */
4168 2254 : if (subquery->setOperations == NULL)
4169 2148 : check_output_expressions(subquery, safetyInfo);
4170 :
4171 : /* Are we at top level, or looking at a setop component? */
4172 2254 : if (subquery == topquery)
4173 : {
4174 : /* Top level, so check any component queries */
4175 2042 : if (subquery->setOperations != NULL)
4176 106 : if (!recurse_pushdown_safe(subquery->setOperations, topquery,
4177 : safetyInfo))
4178 0 : return false;
4179 : }
4180 : else
4181 : {
4182 : /* Setop component must not have more components (too weird) */
4183 212 : if (subquery->setOperations != NULL)
4184 0 : return false;
4185 : /* Check whether setop component output types match top level */
4186 212 : topop = castNode(SetOperationStmt, topquery->setOperations);
4187 : Assert(topop);
4188 212 : compare_tlist_datatypes(subquery->targetList,
4189 : topop->colTypes,
4190 : safetyInfo);
4191 : }
4192 2254 : return true;
4193 : }
4194 :
4195 : /*
4196 : * Helper routine to recurse through setOperations tree
4197 : */
4198 : static bool
4199 318 : recurse_pushdown_safe(Node *setOp, Query *topquery,
4200 : pushdown_safety_info *safetyInfo)
4201 : {
4202 318 : if (IsA(setOp, RangeTblRef))
4203 : {
4204 212 : RangeTblRef *rtr = (RangeTblRef *) setOp;
4205 212 : RangeTblEntry *rte = rt_fetch(rtr->rtindex, topquery->rtable);
4206 212 : Query *subquery = rte->subquery;
4207 :
4208 : Assert(subquery != NULL);
4209 212 : return subquery_is_pushdown_safe(subquery, topquery, safetyInfo);
4210 : }
4211 106 : else if (IsA(setOp, SetOperationStmt))
4212 : {
4213 106 : SetOperationStmt *op = (SetOperationStmt *) setOp;
4214 :
4215 : /* EXCEPT is no good (point 2 for subquery_is_pushdown_safe) */
4216 106 : if (op->op == SETOP_EXCEPT)
4217 0 : return false;
4218 : /* Else recurse */
4219 106 : if (!recurse_pushdown_safe(op->larg, topquery, safetyInfo))
4220 0 : return false;
4221 106 : if (!recurse_pushdown_safe(op->rarg, topquery, safetyInfo))
4222 0 : return false;
4223 : }
4224 : else
4225 : {
4226 0 : elog(ERROR, "unrecognized node type: %d",
4227 : (int) nodeTag(setOp));
4228 : }
4229 106 : return true;
4230 : }
4231 :
4232 : /*
4233 : * check_output_expressions - check subquery's output expressions for safety
4234 : *
4235 : * There are several cases in which it's unsafe to push down an upper-level
4236 : * qual if it references a particular output column of a subquery. We check
4237 : * each output column of the subquery and set flags in unsafeFlags[k] when we
4238 : * see that column is unsafe for a pushed-down qual to reference. The
4239 : * conditions checked here are:
4240 : *
4241 : * 1. We must not push down any quals that refer to subselect outputs that
4242 : * return sets, else we'd introduce functions-returning-sets into the
4243 : * subquery's WHERE/HAVING quals.
4244 : *
4245 : * 2. We must not push down any quals that refer to subselect outputs that
4246 : * contain volatile functions, for fear of introducing strange results due
4247 : * to multiple evaluation of a volatile function.
4248 : *
4249 : * 3. If the subquery uses DISTINCT ON, we must not push down any quals that
4250 : * refer to non-DISTINCT output columns, because that could change the set
4251 : * of rows returned. (This condition is vacuous for DISTINCT, because then
4252 : * there are no non-DISTINCT output columns, so we needn't check. Note that
4253 : * subquery_is_pushdown_safe already reported that we can't use volatile
4254 : * quals if there's DISTINCT or DISTINCT ON.)
4255 : *
4256 : * 4. If the subquery has any window functions, we must not push down quals
4257 : * that reference any output columns that are not listed in all the subquery's
4258 : * window PARTITION BY clauses. We can push down quals that use only
4259 : * partitioning columns because they should succeed or fail identically for
4260 : * every row of any one window partition, and totally excluding some
4261 : * partitions will not change a window function's results for remaining
4262 : * partitions. (Again, this also requires nonvolatile quals, but
4263 : * subquery_is_pushdown_safe handles that.). Subquery columns marked as
4264 : * unsafe for this reason can still have WindowClause run conditions pushed
4265 : * down.
4266 : */
4267 : static void
4268 2148 : check_output_expressions(Query *subquery, pushdown_safety_info *safetyInfo)
4269 : {
4270 2148 : List *flattened_targetList = subquery->targetList;
4271 : ListCell *lc;
4272 :
4273 : /*
4274 : * We must be careful with grouping Vars and join alias Vars in the
4275 : * subquery's outputs, as they hide the underlying expressions.
4276 : *
4277 : * We need to expand grouping Vars to their underlying expressions (the
4278 : * grouping clauses) because the grouping expressions themselves might be
4279 : * volatile or set-returning. However, we do not need to expand join
4280 : * alias Vars, as their underlying structure does not introduce volatile
4281 : * or set-returning functions at the current level.
4282 : *
4283 : * In neither case do we need to recursively examine the Vars contained in
4284 : * these underlying expressions. Even if they reference outputs from
4285 : * lower-level subqueries (at any depth), those references are guaranteed
4286 : * not to expand to volatile or set-returning functions, because
4287 : * subqueries containing such functions in their targetlists are never
4288 : * pulled up.
4289 : */
4290 2148 : if (subquery->hasGroupRTE)
4291 : {
4292 : /*
4293 : * We can safely pass NULL for the root here. This function uses the
4294 : * expanded expressions solely to check for volatile or set-returning
4295 : * functions, which is independent of the Vars' nullingrels.
4296 : */
4297 : flattened_targetList = (List *)
4298 259 : flatten_group_exprs(NULL, subquery, (Node *) subquery->targetList);
4299 : }
4300 :
4301 25331 : foreach(lc, flattened_targetList)
4302 : {
4303 23183 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
4304 :
4305 23183 : if (tle->resjunk)
4306 109 : continue; /* ignore resjunk columns */
4307 :
4308 : /* Functions returning sets are unsafe (point 1) */
4309 23074 : if (subquery->hasTargetSRFs &&
4310 1200 : (safetyInfo->unsafeFlags[tle->resno] &
4311 1200 : UNSAFE_HAS_SET_FUNC) == 0 &&
4312 1200 : expression_returns_set((Node *) tle->expr))
4313 : {
4314 848 : safetyInfo->unsafeFlags[tle->resno] |= UNSAFE_HAS_SET_FUNC;
4315 848 : continue;
4316 : }
4317 :
4318 : /* Volatile functions are unsafe (point 2) */
4319 22226 : if ((safetyInfo->unsafeFlags[tle->resno] &
4320 22216 : UNSAFE_HAS_VOLATILE_FUNC) == 0 &&
4321 22216 : contain_volatile_functions((Node *) tle->expr))
4322 : {
4323 75 : safetyInfo->unsafeFlags[tle->resno] |= UNSAFE_HAS_VOLATILE_FUNC;
4324 75 : continue;
4325 : }
4326 :
4327 : /* If subquery uses DISTINCT ON, check point 3 */
4328 22151 : if (subquery->hasDistinctOn &&
4329 0 : (safetyInfo->unsafeFlags[tle->resno] &
4330 0 : UNSAFE_NOTIN_DISTINCTON_CLAUSE) == 0 &&
4331 0 : !targetIsInSortList(tle, InvalidOid, subquery->distinctClause))
4332 : {
4333 : /* non-DISTINCT column, so mark it unsafe */
4334 0 : safetyInfo->unsafeFlags[tle->resno] |= UNSAFE_NOTIN_DISTINCTON_CLAUSE;
4335 0 : continue;
4336 : }
4337 :
4338 : /* If subquery uses window functions, check point 4 */
4339 22151 : if (subquery->hasWindowFuncs &&
4340 897 : (safetyInfo->unsafeFlags[tle->resno] &
4341 1722 : UNSAFE_NOTIN_DISTINCTON_CLAUSE) == 0 &&
4342 897 : !targetIsInAllPartitionLists(tle, subquery))
4343 : {
4344 : /* not present in all PARTITION BY clauses, so mark it unsafe */
4345 825 : safetyInfo->unsafeFlags[tle->resno] |= UNSAFE_NOTIN_PARTITIONBY_CLAUSE;
4346 825 : continue;
4347 : }
4348 : }
4349 2148 : }
4350 :
4351 : /*
4352 : * For subqueries using UNION/UNION ALL/INTERSECT/INTERSECT ALL, we can
4353 : * push quals into each component query, but the quals can only reference
4354 : * subquery columns that suffer no type coercions in the set operation.
4355 : * Otherwise there are possible semantic gotchas. So, we check the
4356 : * component queries to see if any of them have output types different from
4357 : * the top-level setop outputs. We set the UNSAFE_TYPE_MISMATCH bit in
4358 : * unsafeFlags[k] if column k has different type in any component.
4359 : *
4360 : * We don't have to care about typmods here: the only allowed difference
4361 : * between set-op input and output typmods is input is a specific typmod
4362 : * and output is -1, and that does not require a coercion.
4363 : *
4364 : * tlist is a subquery tlist.
4365 : * colTypes is an OID list of the top-level setop's output column types.
4366 : * safetyInfo is the pushdown_safety_info to set unsafeFlags[] for.
4367 : */
4368 : static void
4369 212 : compare_tlist_datatypes(List *tlist, List *colTypes,
4370 : pushdown_safety_info *safetyInfo)
4371 : {
4372 : ListCell *l;
4373 212 : ListCell *colType = list_head(colTypes);
4374 :
4375 660 : foreach(l, tlist)
4376 : {
4377 448 : TargetEntry *tle = (TargetEntry *) lfirst(l);
4378 :
4379 448 : if (tle->resjunk)
4380 0 : continue; /* ignore resjunk columns */
4381 448 : if (colType == NULL)
4382 0 : elog(ERROR, "wrong number of tlist entries");
4383 448 : if (exprType((Node *) tle->expr) != lfirst_oid(colType))
4384 64 : safetyInfo->unsafeFlags[tle->resno] |= UNSAFE_TYPE_MISMATCH;
4385 448 : colType = lnext(colTypes, colType);
4386 : }
4387 212 : if (colType != NULL)
4388 0 : elog(ERROR, "wrong number of tlist entries");
4389 212 : }
4390 :
4391 : /*
4392 : * targetIsInAllPartitionLists
4393 : * True if the TargetEntry is listed in the PARTITION BY clause
4394 : * of every window defined in the query.
4395 : *
4396 : * It would be safe to ignore windows not actually used by any window
4397 : * function, but it's not easy to get that info at this stage; and it's
4398 : * unlikely to be useful to spend any extra cycles getting it, since
4399 : * unreferenced window definitions are probably infrequent in practice.
4400 : */
4401 : static bool
4402 897 : targetIsInAllPartitionLists(TargetEntry *tle, Query *query)
4403 : {
4404 : ListCell *lc;
4405 :
4406 989 : foreach(lc, query->windowClause)
4407 : {
4408 917 : WindowClause *wc = (WindowClause *) lfirst(lc);
4409 :
4410 917 : if (!targetIsInSortList(tle, InvalidOid, wc->partitionClause))
4411 825 : return false;
4412 : }
4413 72 : return true;
4414 : }
4415 :
4416 : /*
4417 : * qual_is_pushdown_safe - is a particular rinfo safe to push down?
4418 : *
4419 : * rinfo is a restriction clause applying to the given subquery (whose RTE
4420 : * has index rti in the parent query).
4421 : *
4422 : * Conditions checked here:
4423 : *
4424 : * 1. rinfo's clause must not contain any SubPlans (mainly because it's
4425 : * unclear that it will work correctly: SubLinks will already have been
4426 : * transformed into SubPlans in the qual, but not in the subquery). Note that
4427 : * SubLinks that transform to initplans are safe, and will be accepted here
4428 : * because what we'll see in the qual is just a Param referencing the initplan
4429 : * output.
4430 : *
4431 : * 2. If unsafeVolatile is set, rinfo's clause must not contain any volatile
4432 : * functions.
4433 : *
4434 : * 3. If unsafeLeaky is set, rinfo's clause must not contain any leaky
4435 : * functions that are passed Var nodes, and therefore might reveal values from
4436 : * the subquery as side effects.
4437 : *
4438 : * 4. rinfo's clause must not refer to the whole-row output of the subquery
4439 : * (since there is no easy way to name that within the subquery itself).
4440 : *
4441 : * 5. rinfo's clause must not refer to any subquery output columns that were
4442 : * found to be unsafe to reference by subquery_is_pushdown_safe().
4443 : */
4444 : static pushdown_safe_type
4445 3514 : qual_is_pushdown_safe(Query *subquery, Index rti, RestrictInfo *rinfo,
4446 : pushdown_safety_info *safetyInfo)
4447 : {
4448 3514 : pushdown_safe_type safe = PUSHDOWN_SAFE;
4449 3514 : Node *qual = (Node *) rinfo->clause;
4450 : List *vars;
4451 : ListCell *vl;
4452 :
4453 : /* Refuse subselects (point 1) */
4454 3514 : if (contain_subplans(qual))
4455 55 : return PUSHDOWN_UNSAFE;
4456 :
4457 : /* Refuse volatile quals if we found they'd be unsafe (point 2) */
4458 4272 : if (safetyInfo->unsafeVolatile &&
4459 813 : contain_volatile_functions((Node *) rinfo))
4460 15 : return PUSHDOWN_UNSAFE;
4461 :
4462 : /* Refuse leaky quals if told to (point 3) */
4463 5361 : if (safetyInfo->unsafeLeaky &&
4464 1917 : contain_leaked_vars(qual))
4465 135 : return PUSHDOWN_UNSAFE;
4466 :
4467 : /*
4468 : * Examine all Vars used in clause. Since it's a restriction clause, all
4469 : * such Vars must refer to subselect output columns ... unless this is
4470 : * part of a LATERAL subquery, in which case there could be lateral
4471 : * references.
4472 : *
4473 : * By omitting the relevant flags, this also gives us a cheap sanity check
4474 : * that no aggregates or window functions appear in the qual. Those would
4475 : * be unsafe to push down, but at least for the moment we could never see
4476 : * any in a qual anyhow.
4477 : */
4478 3309 : vars = pull_var_clause(qual, PVC_INCLUDE_PLACEHOLDERS);
4479 6273 : foreach(vl, vars)
4480 : {
4481 3399 : Var *var = (Var *) lfirst(vl);
4482 :
4483 : /*
4484 : * XXX Punt if we find any PlaceHolderVars in the restriction clause.
4485 : * It's not clear whether a PHV could safely be pushed down, and even
4486 : * less clear whether such a situation could arise in any cases of
4487 : * practical interest anyway. So for the moment, just refuse to push
4488 : * down.
4489 : */
4490 3399 : if (!IsA(var, Var))
4491 : {
4492 0 : safe = PUSHDOWN_UNSAFE;
4493 0 : break;
4494 : }
4495 :
4496 : /*
4497 : * Punt if we find any lateral references. It would be safe to push
4498 : * these down, but we'd have to convert them into outer references,
4499 : * which subquery_push_qual lacks the infrastructure to do. The case
4500 : * arises so seldom that it doesn't seem worth working hard on.
4501 : */
4502 3399 : if (var->varno != rti)
4503 : {
4504 10 : safe = PUSHDOWN_UNSAFE;
4505 10 : break;
4506 : }
4507 :
4508 : /* Subqueries have no system columns */
4509 : Assert(var->varattno >= 0);
4510 :
4511 : /* Check point 4 */
4512 3389 : if (var->varattno == 0)
4513 : {
4514 0 : safe = PUSHDOWN_UNSAFE;
4515 0 : break;
4516 : }
4517 :
4518 : /* Check point 5 */
4519 3389 : if (safetyInfo->unsafeFlags[var->varattno] != 0)
4520 : {
4521 700 : if (safetyInfo->unsafeFlags[var->varattno] &
4522 : (UNSAFE_HAS_VOLATILE_FUNC | UNSAFE_HAS_SET_FUNC |
4523 : UNSAFE_NOTIN_DISTINCTON_CLAUSE | UNSAFE_TYPE_MISMATCH))
4524 : {
4525 425 : safe = PUSHDOWN_UNSAFE;
4526 425 : break;
4527 : }
4528 : else
4529 : {
4530 : /* UNSAFE_NOTIN_PARTITIONBY_CLAUSE is ok for run conditions */
4531 275 : safe = PUSHDOWN_WINDOWCLAUSE_RUNCOND;
4532 : /* don't break, we might find another Var that's unsafe */
4533 : }
4534 : }
4535 : }
4536 :
4537 3309 : list_free(vars);
4538 :
4539 3309 : return safe;
4540 : }
4541 :
4542 : /*
4543 : * subquery_push_qual - push down a qual that we have determined is safe
4544 : */
4545 : static void
4546 2836 : subquery_push_qual(Query *subquery, RangeTblEntry *rte, Index rti, Node *qual)
4547 : {
4548 2836 : if (subquery->setOperations != NULL)
4549 : {
4550 : /* Recurse to push it separately to each component query */
4551 86 : recurse_push_qual(subquery->setOperations, subquery,
4552 : rte, rti, qual);
4553 : }
4554 : else
4555 : {
4556 : /*
4557 : * We need to replace Vars in the qual (which must refer to outputs of
4558 : * the subquery) with copies of the subquery's targetlist expressions.
4559 : * Note that at this point, any uplevel Vars in the qual should have
4560 : * been replaced with Params, so they need no work.
4561 : *
4562 : * This step also ensures that when we are pushing into a setop tree,
4563 : * each component query gets its own copy of the qual.
4564 : */
4565 2750 : qual = ReplaceVarsFromTargetList(qual, rti, 0, rte,
4566 : subquery->targetList,
4567 : subquery->resultRelation,
4568 : REPLACEVARS_REPORT_ERROR, 0,
4569 : &subquery->hasSubLinks);
4570 :
4571 : /*
4572 : * Now attach the qual to the proper place: normally WHERE, but if the
4573 : * subquery uses grouping or aggregation, put it in HAVING (since the
4574 : * qual really refers to the group-result rows).
4575 : */
4576 2750 : if (subquery->hasAggs || subquery->groupClause || subquery->groupingSets || subquery->havingQual)
4577 333 : subquery->havingQual = make_and_qual(subquery->havingQual, qual);
4578 : else
4579 2417 : subquery->jointree->quals =
4580 2417 : make_and_qual(subquery->jointree->quals, qual);
4581 :
4582 : /*
4583 : * We need not change the subquery's hasAggs or hasSubLinks flags,
4584 : * since we can't be pushing down any aggregates that weren't there
4585 : * before, and we don't push down subselects at all.
4586 : */
4587 : }
4588 2836 : }
4589 :
4590 : /*
4591 : * Helper routine to recurse through setOperations tree
4592 : */
4593 : static void
4594 258 : recurse_push_qual(Node *setOp, Query *topquery,
4595 : RangeTblEntry *rte, Index rti, Node *qual)
4596 : {
4597 258 : if (IsA(setOp, RangeTblRef))
4598 : {
4599 172 : RangeTblRef *rtr = (RangeTblRef *) setOp;
4600 172 : RangeTblEntry *subrte = rt_fetch(rtr->rtindex, topquery->rtable);
4601 172 : Query *subquery = subrte->subquery;
4602 :
4603 : Assert(subquery != NULL);
4604 172 : subquery_push_qual(subquery, rte, rti, qual);
4605 : }
4606 86 : else if (IsA(setOp, SetOperationStmt))
4607 : {
4608 86 : SetOperationStmt *op = (SetOperationStmt *) setOp;
4609 :
4610 86 : recurse_push_qual(op->larg, topquery, rte, rti, qual);
4611 86 : recurse_push_qual(op->rarg, topquery, rte, rti, qual);
4612 : }
4613 : else
4614 : {
4615 0 : elog(ERROR, "unrecognized node type: %d",
4616 : (int) nodeTag(setOp));
4617 : }
4618 258 : }
4619 :
4620 : /*****************************************************************************
4621 : * SIMPLIFYING SUBQUERY TARGETLISTS
4622 : *****************************************************************************/
4623 :
4624 : /*
4625 : * remove_unused_subquery_outputs
4626 : * Remove subquery targetlist items we don't need
4627 : *
4628 : * It's possible, even likely, that the upper query does not read all the
4629 : * output columns of the subquery. We can remove any such outputs that are
4630 : * not needed by the subquery itself (e.g., as sort/group columns) and do not
4631 : * affect semantics otherwise (e.g., volatile functions can't be removed).
4632 : * This is useful not only because we might be able to remove expensive-to-
4633 : * compute expressions, but because deletion of output columns might allow
4634 : * optimizations such as join removal to occur within the subquery.
4635 : *
4636 : * extra_used_attrs can be passed as non-NULL to mark any columns (offset by
4637 : * FirstLowInvalidHeapAttributeNumber) that we should not remove. This
4638 : * parameter is modified by the function, so callers must make a copy if they
4639 : * need to use the passed in Bitmapset after calling this function.
4640 : *
4641 : * To avoid affecting column numbering in the targetlist, we don't physically
4642 : * remove unused tlist entries, but rather replace their expressions with NULL
4643 : * constants. This is implemented by modifying subquery->targetList.
4644 : */
4645 : static void
4646 17357 : remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel,
4647 : Bitmapset *extra_used_attrs)
4648 : {
4649 : Bitmapset *attrs_used;
4650 : ListCell *lc;
4651 :
4652 : /*
4653 : * Just point directly to extra_used_attrs. No need to bms_copy as none of
4654 : * the current callers use the Bitmapset after calling this function.
4655 : */
4656 17357 : attrs_used = extra_used_attrs;
4657 :
4658 : /*
4659 : * Do nothing if subquery has UNION/INTERSECT/EXCEPT: in principle we
4660 : * could update all the child SELECTs' tlists, but it seems not worth the
4661 : * trouble presently.
4662 : */
4663 17357 : if (subquery->setOperations)
4664 1365 : return;
4665 :
4666 : /*
4667 : * If subquery has regular DISTINCT (not DISTINCT ON), we're wasting our
4668 : * time: all its output columns must be used in the distinctClause.
4669 : */
4670 16765 : if (subquery->distinctClause && !subquery->hasDistinctOn)
4671 527 : return;
4672 :
4673 : /*
4674 : * Collect a bitmap of all the output column numbers used by the upper
4675 : * query.
4676 : *
4677 : * Add all the attributes needed for joins or final output. Note: we must
4678 : * look at rel's targetlist, not the attr_needed data, because attr_needed
4679 : * isn't computed for inheritance child rels, cf set_append_rel_size().
4680 : * (XXX might be worth changing that sometime.)
4681 : */
4682 16238 : pull_varattnos((Node *) rel->reltarget->exprs, rel->relid, &attrs_used);
4683 :
4684 : /* Add all the attributes used by un-pushed-down restriction clauses. */
4685 17111 : foreach(lc, rel->baserestrictinfo)
4686 : {
4687 873 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
4688 :
4689 873 : pull_varattnos((Node *) rinfo->clause, rel->relid, &attrs_used);
4690 : }
4691 :
4692 : /*
4693 : * If there's a whole-row reference to the subquery, we can't remove
4694 : * anything.
4695 : */
4696 16238 : if (bms_is_member(0 - FirstLowInvalidHeapAttributeNumber, attrs_used))
4697 246 : return;
4698 :
4699 : /*
4700 : * Run through the tlist and zap entries we don't need. It's okay to
4701 : * modify the tlist items in-place because set_subquery_pathlist made a
4702 : * copy of the subquery.
4703 : */
4704 89565 : foreach(lc, subquery->targetList)
4705 : {
4706 73573 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
4707 73573 : Node *texpr = (Node *) tle->expr;
4708 :
4709 : /*
4710 : * If it has a sortgroupref number, it's used in some sort/group
4711 : * clause so we'd better not remove it. Also, don't remove any
4712 : * resjunk columns, since their reason for being has nothing to do
4713 : * with anybody reading the subquery's output. (It's likely that
4714 : * resjunk columns in a sub-SELECT would always have ressortgroupref
4715 : * set, but even if they don't, it seems imprudent to remove them.)
4716 : */
4717 73573 : if (tle->ressortgroupref || tle->resjunk)
4718 2481 : continue;
4719 :
4720 : /*
4721 : * If it's used by the upper query, we can't remove it.
4722 : */
4723 71092 : if (bms_is_member(tle->resno - FirstLowInvalidHeapAttributeNumber,
4724 : attrs_used))
4725 40991 : continue;
4726 :
4727 : /*
4728 : * If it contains a set-returning function, we can't remove it since
4729 : * that could change the number of rows returned by the subquery.
4730 : */
4731 30805 : if (subquery->hasTargetSRFs &&
4732 704 : expression_returns_set(texpr))
4733 474 : continue;
4734 :
4735 : /*
4736 : * If it contains volatile functions, we daren't remove it for fear
4737 : * that the user is expecting their side-effects to happen.
4738 : */
4739 29627 : if (contain_volatile_functions(texpr))
4740 26 : continue;
4741 :
4742 : /*
4743 : * OK, we don't need it. Replace the expression with a NULL constant.
4744 : * Preserve the exposed type of the expression, in case something
4745 : * looks at the rowtype of the subquery's result.
4746 : */
4747 29601 : tle->expr = (Expr *) makeNullConst(exprType(texpr),
4748 : exprTypmod(texpr),
4749 : exprCollation(texpr));
4750 : }
4751 : }
4752 :
4753 : /*
4754 : * create_partial_bitmap_paths
4755 : * Build partial bitmap heap path for the relation
4756 : */
4757 : void
4758 111174 : create_partial_bitmap_paths(PlannerInfo *root, RelOptInfo *rel,
4759 : Path *bitmapqual)
4760 : {
4761 : int parallel_workers;
4762 : double pages_fetched;
4763 :
4764 : /* Compute heap pages for bitmap heap scan */
4765 111174 : pages_fetched = compute_bitmap_pages(root, rel, bitmapqual, 1.0,
4766 : NULL, NULL);
4767 :
4768 111174 : parallel_workers = compute_parallel_worker(rel, pages_fetched, -1,
4769 : max_parallel_workers_per_gather);
4770 :
4771 111174 : if (parallel_workers <= 0)
4772 107933 : return;
4773 :
4774 3241 : add_partial_path(rel, (Path *) create_bitmap_heap_path(root, rel,
4775 : bitmapqual, rel->lateral_relids, 1.0, parallel_workers));
4776 : }
4777 :
4778 : /*
4779 : * Compute the number of parallel workers that should be used to scan a
4780 : * relation. We compute the parallel workers based on the size of the heap to
4781 : * be scanned and the size of the index to be scanned, then choose a minimum
4782 : * of those.
4783 : *
4784 : * "heap_pages" is the number of pages from the table that we expect to scan, or
4785 : * -1 if we don't expect to scan any.
4786 : *
4787 : * "index_pages" is the number of pages from the index that we expect to scan, or
4788 : * -1 if we don't expect to scan any.
4789 : *
4790 : * "max_workers" is caller's limit on the number of workers. This typically
4791 : * comes from a GUC.
4792 : */
4793 : int
4794 602921 : compute_parallel_worker(RelOptInfo *rel, double heap_pages, double index_pages,
4795 : int max_workers)
4796 : {
4797 602921 : int parallel_workers = 0;
4798 :
4799 : /*
4800 : * If the user has set the parallel_workers reloption, use that; otherwise
4801 : * select a default number of workers.
4802 : */
4803 602921 : if (rel->rel_parallel_workers != -1)
4804 2740 : parallel_workers = rel->rel_parallel_workers;
4805 : else
4806 : {
4807 : /*
4808 : * If the number of pages being scanned is insufficient to justify a
4809 : * parallel scan, just return zero ... unless it's an inheritance
4810 : * child. In that case, we want to generate a parallel path here
4811 : * anyway. It might not be worthwhile just for this relation, but
4812 : * when combined with all of its inheritance siblings it may well pay
4813 : * off.
4814 : */
4815 600181 : if (rel->reloptkind == RELOPT_BASEREL &&
4816 568162 : ((heap_pages >= 0 && heap_pages < min_parallel_table_scan_size) ||
4817 18588 : (index_pages >= 0 && index_pages < min_parallel_index_scan_size)))
4818 567407 : return 0;
4819 :
4820 32774 : if (heap_pages >= 0)
4821 : {
4822 : int heap_parallel_threshold;
4823 31216 : int heap_parallel_workers = 1;
4824 :
4825 : /*
4826 : * Select the number of workers based on the log of the size of
4827 : * the relation. This probably needs to be a good deal more
4828 : * sophisticated, but we need something here for now. Note that
4829 : * the upper limit of the min_parallel_table_scan_size GUC is
4830 : * chosen to prevent overflow here.
4831 : */
4832 31216 : heap_parallel_threshold = Max(min_parallel_table_scan_size, 1);
4833 35624 : while (heap_pages >= (BlockNumber) (heap_parallel_threshold * 3))
4834 : {
4835 4408 : heap_parallel_workers++;
4836 4408 : heap_parallel_threshold *= 3;
4837 4408 : if (heap_parallel_threshold > INT_MAX / 3)
4838 0 : break; /* avoid overflow */
4839 : }
4840 :
4841 31216 : parallel_workers = heap_parallel_workers;
4842 : }
4843 :
4844 32774 : if (index_pages >= 0)
4845 : {
4846 7410 : int index_parallel_workers = 1;
4847 : int index_parallel_threshold;
4848 :
4849 : /* same calculation as for heap_pages above */
4850 7410 : index_parallel_threshold = Max(min_parallel_index_scan_size, 1);
4851 7640 : while (index_pages >= (BlockNumber) (index_parallel_threshold * 3))
4852 : {
4853 230 : index_parallel_workers++;
4854 230 : index_parallel_threshold *= 3;
4855 230 : if (index_parallel_threshold > INT_MAX / 3)
4856 0 : break; /* avoid overflow */
4857 : }
4858 :
4859 7410 : if (parallel_workers > 0)
4860 5852 : parallel_workers = Min(parallel_workers, index_parallel_workers);
4861 : else
4862 1558 : parallel_workers = index_parallel_workers;
4863 : }
4864 : }
4865 :
4866 : /* In no case use more than caller supplied maximum number of workers */
4867 35514 : parallel_workers = Min(parallel_workers, max_workers);
4868 :
4869 35514 : return parallel_workers;
4870 : }
4871 :
4872 : /*
4873 : * generate_partitionwise_join_paths
4874 : * Create paths representing partitionwise join for given partitioned
4875 : * join relation.
4876 : *
4877 : * This must not be called until after we are done adding paths for all
4878 : * child-joins. Otherwise, add_path might delete a path to which some path
4879 : * generated here has a reference.
4880 : */
4881 : void
4882 194692 : generate_partitionwise_join_paths(PlannerInfo *root, RelOptInfo *rel)
4883 : {
4884 194692 : List *live_children = NIL;
4885 : int cnt_parts;
4886 : int num_parts;
4887 : RelOptInfo **part_rels;
4888 :
4889 : /* Handle only join relations here. */
4890 194692 : if (!IS_JOIN_REL(rel))
4891 0 : return;
4892 :
4893 : /* We've nothing to do if the relation is not partitioned. */
4894 194692 : if (!IS_PARTITIONED_REL(rel))
4895 188725 : return;
4896 :
4897 : /* The relation should have consider_partitionwise_join set. */
4898 : Assert(rel->consider_partitionwise_join);
4899 :
4900 : /* Guard against stack overflow due to overly deep partition hierarchy. */
4901 5967 : check_stack_depth();
4902 :
4903 5967 : num_parts = rel->nparts;
4904 5967 : part_rels = rel->part_rels;
4905 :
4906 : /* Collect non-dummy child-joins. */
4907 21264 : for (cnt_parts = 0; cnt_parts < num_parts; cnt_parts++)
4908 : {
4909 15297 : RelOptInfo *child_rel = part_rels[cnt_parts];
4910 :
4911 : /* If it's been pruned entirely, it's certainly dummy. */
4912 15297 : if (child_rel == NULL)
4913 52 : continue;
4914 :
4915 : /* Make partitionwise join paths for this partitioned child-join. */
4916 15245 : generate_partitionwise_join_paths(root, child_rel);
4917 :
4918 : /* If we failed to make any path for this child, we must give up. */
4919 15245 : if (child_rel->pathlist == NIL)
4920 : {
4921 : /*
4922 : * Mark the parent joinrel as unpartitioned so that later
4923 : * functions treat it correctly.
4924 : */
4925 0 : rel->nparts = 0;
4926 0 : return;
4927 : }
4928 :
4929 : /* Else, identify the cheapest path for it. */
4930 15245 : set_cheapest(child_rel);
4931 :
4932 : /* Dummy children need not be scanned, so ignore those. */
4933 15245 : if (IS_DUMMY_REL(child_rel))
4934 0 : continue;
4935 :
4936 : /*
4937 : * Except for the topmost scan/join rel, consider generating partial
4938 : * aggregation paths for the grouped relation on top of the paths of
4939 : * this partitioned child-join. After that, we're done creating paths
4940 : * for the grouped relation, so run set_cheapest().
4941 : */
4942 15245 : if (child_rel->grouped_rel != NULL &&
4943 10730 : !bms_equal(IS_OTHER_REL(rel) ?
4944 : rel->top_parent_relids : rel->relids,
4945 10730 : root->all_query_rels))
4946 : {
4947 200 : RelOptInfo *grouped_rel = child_rel->grouped_rel;
4948 :
4949 : Assert(IS_GROUPED_REL(grouped_rel));
4950 :
4951 200 : generate_grouped_paths(root, grouped_rel, child_rel);
4952 200 : set_cheapest(grouped_rel);
4953 : }
4954 :
4955 : #ifdef OPTIMIZER_DEBUG
4956 : pprint(child_rel);
4957 : #endif
4958 :
4959 15245 : live_children = lappend(live_children, child_rel);
4960 : }
4961 :
4962 : /* If all child-joins are dummy, parent join is also dummy. */
4963 5967 : if (!live_children)
4964 : {
4965 0 : mark_dummy_rel(rel);
4966 0 : return;
4967 : }
4968 :
4969 : /* Build additional paths for this rel from child-join paths. */
4970 5967 : add_paths_to_append_rel(root, rel, live_children);
4971 5967 : list_free(live_children);
4972 : }
|
