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) 1996-2017, 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 "nodes/makefuncs.h"
28 : #include "nodes/nodeFuncs.h"
29 : #ifdef OPTIMIZER_DEBUG
30 : #include "nodes/print.h"
31 : #endif
32 : #include "optimizer/clauses.h"
33 : #include "optimizer/cost.h"
34 : #include "optimizer/geqo.h"
35 : #include "optimizer/pathnode.h"
36 : #include "optimizer/paths.h"
37 : #include "optimizer/plancat.h"
38 : #include "optimizer/planner.h"
39 : #include "optimizer/prep.h"
40 : #include "optimizer/restrictinfo.h"
41 : #include "optimizer/tlist.h"
42 : #include "optimizer/var.h"
43 : #include "parser/parse_clause.h"
44 : #include "parser/parsetree.h"
45 : #include "rewrite/rewriteManip.h"
46 : #include "utils/lsyscache.h"
47 :
48 :
49 : /* results of subquery_is_pushdown_safe */
50 : typedef struct pushdown_safety_info
51 : {
52 : bool *unsafeColumns; /* which output columns are unsafe to use */
53 : bool unsafeVolatile; /* don't push down volatile quals */
54 : bool unsafeLeaky; /* don't push down leaky quals */
55 : } pushdown_safety_info;
56 :
57 : /* These parameters are set by GUC */
58 : bool enable_geqo = false; /* just in case GUC doesn't set it */
59 : int geqo_threshold;
60 : int min_parallel_table_scan_size;
61 : int min_parallel_index_scan_size;
62 :
63 : /* Hook for plugins to get control in set_rel_pathlist() */
64 : set_rel_pathlist_hook_type set_rel_pathlist_hook = NULL;
65 :
66 : /* Hook for plugins to replace standard_join_search() */
67 : join_search_hook_type join_search_hook = NULL;
68 :
69 :
70 : static void set_base_rel_consider_startup(PlannerInfo *root);
71 : static void set_base_rel_sizes(PlannerInfo *root);
72 : static void set_base_rel_pathlists(PlannerInfo *root);
73 : static void set_rel_size(PlannerInfo *root, RelOptInfo *rel,
74 : Index rti, RangeTblEntry *rte);
75 : static void set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
76 : Index rti, RangeTblEntry *rte);
77 : static void set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel,
78 : RangeTblEntry *rte);
79 : static void create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel);
80 : static void set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel,
81 : RangeTblEntry *rte);
82 : static void set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
83 : RangeTblEntry *rte);
84 : static void set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel,
85 : RangeTblEntry *rte);
86 : static void set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
87 : RangeTblEntry *rte);
88 : static void set_foreign_size(PlannerInfo *root, RelOptInfo *rel,
89 : RangeTblEntry *rte);
90 : static void set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel,
91 : RangeTblEntry *rte);
92 : static void set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
93 : Index rti, RangeTblEntry *rte);
94 : static void set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
95 : Index rti, RangeTblEntry *rte);
96 : static void generate_mergeappend_paths(PlannerInfo *root, RelOptInfo *rel,
97 : List *live_childrels,
98 : List *all_child_pathkeys,
99 : List *partitioned_rels);
100 : static Path *get_cheapest_parameterized_child_path(PlannerInfo *root,
101 : RelOptInfo *rel,
102 : Relids required_outer);
103 : static List *accumulate_append_subpath(List *subpaths, Path *path);
104 : static void set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
105 : Index rti, RangeTblEntry *rte);
106 : static void set_function_pathlist(PlannerInfo *root, RelOptInfo *rel,
107 : RangeTblEntry *rte);
108 : static void set_values_pathlist(PlannerInfo *root, RelOptInfo *rel,
109 : RangeTblEntry *rte);
110 : static void set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel,
111 : RangeTblEntry *rte);
112 : static void set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel,
113 : RangeTblEntry *rte);
114 : static void set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel,
115 : RangeTblEntry *rte);
116 : static void set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel,
117 : RangeTblEntry *rte);
118 : static RelOptInfo *make_rel_from_joinlist(PlannerInfo *root, List *joinlist);
119 : static bool subquery_is_pushdown_safe(Query *subquery, Query *topquery,
120 : pushdown_safety_info *safetyInfo);
121 : static bool recurse_pushdown_safe(Node *setOp, Query *topquery,
122 : pushdown_safety_info *safetyInfo);
123 : static void check_output_expressions(Query *subquery,
124 : pushdown_safety_info *safetyInfo);
125 : static void compare_tlist_datatypes(List *tlist, List *colTypes,
126 : pushdown_safety_info *safetyInfo);
127 : static bool targetIsInAllPartitionLists(TargetEntry *tle, Query *query);
128 : static bool qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual,
129 : pushdown_safety_info *safetyInfo);
130 : static void subquery_push_qual(Query *subquery,
131 : RangeTblEntry *rte, Index rti, Node *qual);
132 : static void recurse_push_qual(Node *setOp, Query *topquery,
133 : RangeTblEntry *rte, Index rti, Node *qual);
134 : static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
135 : static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
136 : List *live_childrels);
137 :
138 :
139 : /*
140 : * make_one_rel
141 : * Finds all possible access paths for executing a query, returning a
142 : * single rel that represents the join of all base rels in the query.
143 : */
144 : RelOptInfo *
145 13818 : make_one_rel(PlannerInfo *root, List *joinlist)
146 : {
147 : RelOptInfo *rel;
148 : Index rti;
149 :
150 : /*
151 : * Construct the all_baserels Relids set.
152 : */
153 13818 : root->all_baserels = NULL;
154 40855 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
155 : {
156 27037 : RelOptInfo *brel = root->simple_rel_array[rti];
157 :
158 : /* there may be empty slots corresponding to non-baserel RTEs */
159 27037 : if (brel == NULL)
160 7514 : continue;
161 :
162 19523 : Assert(brel->relid == rti); /* sanity check on array */
163 :
164 : /* ignore RTEs that are "other rels" */
165 19523 : if (brel->reloptkind != RELOPT_BASEREL)
166 1983 : continue;
167 :
168 17540 : root->all_baserels = bms_add_member(root->all_baserels, brel->relid);
169 : }
170 :
171 : /* Mark base rels as to whether we care about fast-start plans */
172 13818 : set_base_rel_consider_startup(root);
173 :
174 : /*
175 : * Compute size estimates and consider_parallel flags for each base rel,
176 : * then generate access paths.
177 : */
178 13818 : set_base_rel_sizes(root);
179 13818 : set_base_rel_pathlists(root);
180 :
181 : /*
182 : * Generate access paths for the entire join tree.
183 : */
184 13818 : rel = make_rel_from_joinlist(root, joinlist);
185 :
186 : /*
187 : * The result should join all and only the query's base rels.
188 : */
189 13818 : Assert(bms_equal(rel->relids, root->all_baserels));
190 :
191 13818 : return rel;
192 : }
193 :
194 : /*
195 : * set_base_rel_consider_startup
196 : * Set the consider_[param_]startup flags for each base-relation entry.
197 : *
198 : * For the moment, we only deal with consider_param_startup here; because the
199 : * logic for consider_startup is pretty trivial and is the same for every base
200 : * relation, we just let build_simple_rel() initialize that flag correctly to
201 : * start with. If that logic ever gets more complicated it would probably
202 : * be better to move it here.
203 : */
204 : static void
205 13818 : set_base_rel_consider_startup(PlannerInfo *root)
206 : {
207 : /*
208 : * Since parameterized paths can only be used on the inside of a nestloop
209 : * join plan, there is usually little value in considering fast-start
210 : * plans for them. However, for relations that are on the RHS of a SEMI
211 : * or ANTI join, a fast-start plan can be useful because we're only going
212 : * to care about fetching one tuple anyway.
213 : *
214 : * To minimize growth of planning time, we currently restrict this to
215 : * cases where the RHS is a single base relation, not a join; there is no
216 : * provision for consider_param_startup to get set at all on joinrels.
217 : * Also we don't worry about appendrels. costsize.c's costing rules for
218 : * nestloop semi/antijoins don't consider such cases either.
219 : */
220 : ListCell *lc;
221 :
222 14913 : foreach(lc, root->join_info_list)
223 : {
224 1095 : SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc);
225 : int varno;
226 :
227 1384 : if ((sjinfo->jointype == JOIN_SEMI || sjinfo->jointype == JOIN_ANTI) &&
228 289 : bms_get_singleton_member(sjinfo->syn_righthand, &varno))
229 : {
230 276 : RelOptInfo *rel = find_base_rel(root, varno);
231 :
232 276 : rel->consider_param_startup = true;
233 : }
234 : }
235 13818 : }
236 :
237 : /*
238 : * set_base_rel_sizes
239 : * Set the size estimates (rows and widths) for each base-relation entry.
240 : * Also determine whether to consider parallel paths for base relations.
241 : *
242 : * We do this in a separate pass over the base rels so that rowcount
243 : * estimates are available for parameterized path generation, and also so
244 : * that each rel's consider_parallel flag is set correctly before we begin to
245 : * generate paths.
246 : */
247 : static void
248 13818 : set_base_rel_sizes(PlannerInfo *root)
249 : {
250 : Index rti;
251 :
252 40855 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
253 : {
254 27037 : RelOptInfo *rel = root->simple_rel_array[rti];
255 : RangeTblEntry *rte;
256 :
257 : /* there may be empty slots corresponding to non-baserel RTEs */
258 27037 : if (rel == NULL)
259 7514 : continue;
260 :
261 19523 : Assert(rel->relid == rti); /* sanity check on array */
262 :
263 : /* ignore RTEs that are "other rels" */
264 19523 : if (rel->reloptkind != RELOPT_BASEREL)
265 1983 : continue;
266 :
267 17540 : rte = root->simple_rte_array[rti];
268 :
269 : /*
270 : * If parallelism is allowable for this query in general, see whether
271 : * it's allowable for this rel in particular. We have to do this
272 : * before set_rel_size(), because (a) if this rel is an inheritance
273 : * parent, set_append_rel_size() will use and perhaps change the rel's
274 : * consider_parallel flag, and (b) for some RTE types, set_rel_size()
275 : * goes ahead and makes paths immediately.
276 : */
277 17540 : if (root->glob->parallelModeOK)
278 13944 : set_rel_consider_parallel(root, rel, rte);
279 :
280 17540 : set_rel_size(root, rel, rti, rte);
281 : }
282 13818 : }
283 :
284 : /*
285 : * set_base_rel_pathlists
286 : * Finds all paths available for scanning each base-relation entry.
287 : * Sequential scan and any available indices are considered.
288 : * Each useful path is attached to its relation's 'pathlist' field.
289 : */
290 : static void
291 13818 : set_base_rel_pathlists(PlannerInfo *root)
292 : {
293 : Index rti;
294 :
295 40855 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
296 : {
297 27037 : RelOptInfo *rel = root->simple_rel_array[rti];
298 :
299 : /* there may be empty slots corresponding to non-baserel RTEs */
300 27037 : if (rel == NULL)
301 7514 : continue;
302 :
303 19523 : Assert(rel->relid == rti); /* sanity check on array */
304 :
305 : /* ignore RTEs that are "other rels" */
306 19523 : if (rel->reloptkind != RELOPT_BASEREL)
307 1983 : continue;
308 :
309 17540 : set_rel_pathlist(root, rel, rti, root->simple_rte_array[rti]);
310 : }
311 13818 : }
312 :
313 : /*
314 : * set_rel_size
315 : * Set size estimates for a base relation
316 : */
317 : static void
318 19014 : set_rel_size(PlannerInfo *root, RelOptInfo *rel,
319 : Index rti, RangeTblEntry *rte)
320 : {
321 36554 : if (rel->reloptkind == RELOPT_BASEREL &&
322 17540 : relation_excluded_by_constraints(root, rel, rte))
323 : {
324 : /*
325 : * We proved we don't need to scan the rel via constraint exclusion,
326 : * so set up a single dummy path for it. Here we only check this for
327 : * regular baserels; if it's an otherrel, CE was already checked in
328 : * set_append_rel_size().
329 : *
330 : * In this case, we go ahead and set up the relation's path right away
331 : * instead of leaving it for set_rel_pathlist to do. This is because
332 : * we don't have a convention for marking a rel as dummy except by
333 : * assigning a dummy path to it.
334 : */
335 31 : set_dummy_rel_pathlist(rel);
336 : }
337 18983 : else if (rte->inh)
338 : {
339 : /* It's an "append relation", process accordingly */
340 589 : set_append_rel_size(root, rel, rti, rte);
341 : }
342 : else
343 : {
344 18394 : switch (rel->rtekind)
345 : {
346 : case RTE_RELATION:
347 15547 : if (rte->relkind == RELKIND_FOREIGN_TABLE)
348 : {
349 : /* Foreign table */
350 0 : set_foreign_size(root, rel, rte);
351 : }
352 15547 : else if (rte->relkind == RELKIND_PARTITIONED_TABLE)
353 : {
354 : /*
355 : * A partitioned table without leaf partitions is marked
356 : * as a dummy rel.
357 : */
358 0 : set_dummy_rel_pathlist(rel);
359 : }
360 15547 : else if (rte->tablesample != NULL)
361 : {
362 : /* Sampled relation */
363 36 : set_tablesample_rel_size(root, rel, rte);
364 : }
365 : else
366 : {
367 : /* Plain relation */
368 15511 : set_plain_rel_size(root, rel, rte);
369 : }
370 15547 : break;
371 : case RTE_SUBQUERY:
372 :
373 : /*
374 : * Subqueries don't support making a choice between
375 : * parameterized and unparameterized paths, so just go ahead
376 : * and build their paths immediately.
377 : */
378 790 : set_subquery_pathlist(root, rel, rti, rte);
379 790 : break;
380 : case RTE_FUNCTION:
381 1327 : set_function_size_estimates(root, rel);
382 1327 : break;
383 : case RTE_TABLEFUNC:
384 22 : set_tablefunc_size_estimates(root, rel);
385 22 : break;
386 : case RTE_VALUES:
387 463 : set_values_size_estimates(root, rel);
388 463 : break;
389 : case RTE_CTE:
390 :
391 : /*
392 : * CTEs don't support making a choice between parameterized
393 : * and unparameterized paths, so just go ahead and build their
394 : * paths immediately.
395 : */
396 202 : if (rte->self_reference)
397 40 : set_worktable_pathlist(root, rel, rte);
398 : else
399 162 : set_cte_pathlist(root, rel, rte);
400 202 : break;
401 : case RTE_NAMEDTUPLESTORE:
402 43 : set_namedtuplestore_pathlist(root, rel, rte);
403 43 : break;
404 : default:
405 0 : elog(ERROR, "unexpected rtekind: %d", (int) rel->rtekind);
406 : break;
407 : }
408 : }
409 :
410 : /*
411 : * We insist that all non-dummy rels have a nonzero rowcount estimate.
412 : */
413 19014 : Assert(rel->rows > 0 || IS_DUMMY_REL(rel));
414 19014 : }
415 :
416 : /*
417 : * set_rel_pathlist
418 : * Build access paths for a base relation
419 : */
420 : static void
421 19087 : set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
422 : Index rti, RangeTblEntry *rte)
423 : {
424 19087 : if (IS_DUMMY_REL(rel))
425 : {
426 : /* We already proved the relation empty, so nothing more to do */
427 : }
428 18975 : else if (rte->inh)
429 : {
430 : /* It's an "append relation", process accordingly */
431 584 : set_append_rel_pathlist(root, rel, rti, rte);
432 : }
433 : else
434 : {
435 18391 : switch (rel->rtekind)
436 : {
437 : case RTE_RELATION:
438 15547 : if (rte->relkind == RELKIND_FOREIGN_TABLE)
439 : {
440 : /* Foreign table */
441 0 : set_foreign_pathlist(root, rel, rte);
442 : }
443 15547 : else if (rte->tablesample != NULL)
444 : {
445 : /* Sampled relation */
446 36 : set_tablesample_rel_pathlist(root, rel, rte);
447 : }
448 : else
449 : {
450 : /* Plain relation */
451 15511 : set_plain_rel_pathlist(root, rel, rte);
452 : }
453 15547 : break;
454 : case RTE_SUBQUERY:
455 : /* Subquery --- fully handled during set_rel_size */
456 787 : break;
457 : case RTE_FUNCTION:
458 : /* RangeFunction */
459 1327 : set_function_pathlist(root, rel, rte);
460 1327 : break;
461 : case RTE_TABLEFUNC:
462 : /* Table Function */
463 22 : set_tablefunc_pathlist(root, rel, rte);
464 22 : break;
465 : case RTE_VALUES:
466 : /* Values list */
467 463 : set_values_pathlist(root, rel, rte);
468 463 : break;
469 : case RTE_CTE:
470 : /* CTE reference --- fully handled during set_rel_size */
471 202 : break;
472 : case RTE_NAMEDTUPLESTORE:
473 : /* tuplestore reference --- fully handled during set_rel_size */
474 43 : break;
475 : default:
476 0 : elog(ERROR, "unexpected rtekind: %d", (int) rel->rtekind);
477 : break;
478 : }
479 : }
480 :
481 : /*
482 : * If this is a baserel, consider gathering any partial paths we may have
483 : * created for it. (If we tried to gather inheritance children, we could
484 : * end up with a very large number of gather nodes, each trying to grab
485 : * its own pool of workers, so don't do this for otherrels. Instead,
486 : * we'll consider gathering partial paths for the parent appendrel.)
487 : */
488 19087 : if (rel->reloptkind == RELOPT_BASEREL)
489 17540 : generate_gather_paths(root, rel);
490 :
491 : /*
492 : * Allow a plugin to editorialize on the set of Paths for this base
493 : * relation. It could add new paths (such as CustomPaths) by calling
494 : * add_path(), or delete or modify paths added by the core code.
495 : */
496 19087 : if (set_rel_pathlist_hook)
497 0 : (*set_rel_pathlist_hook) (root, rel, rti, rte);
498 :
499 : /* Now find the cheapest of the paths for this rel */
500 19087 : set_cheapest(rel);
501 :
502 : #ifdef OPTIMIZER_DEBUG
503 : debug_print_rel(root, rel);
504 : #endif
505 19087 : }
506 :
507 : /*
508 : * set_plain_rel_size
509 : * Set size estimates for a plain relation (no subquery, no inheritance)
510 : */
511 : static void
512 15511 : set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
513 : {
514 : /*
515 : * Test any partial indexes of rel for applicability. We must do this
516 : * first since partial unique indexes can affect size estimates.
517 : */
518 15511 : check_index_predicates(root, rel);
519 :
520 : /* Mark rel with estimated output rows, width, etc */
521 15511 : set_baserel_size_estimates(root, rel);
522 15511 : }
523 :
524 : /*
525 : * If this relation could possibly be scanned from within a worker, then set
526 : * its consider_parallel flag.
527 : */
528 : static void
529 14992 : set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel,
530 : RangeTblEntry *rte)
531 : {
532 : /*
533 : * The flag has previously been initialized to false, so we can just
534 : * return if it becomes clear that we can't safely set it.
535 : */
536 14992 : Assert(!rel->consider_parallel);
537 :
538 : /* Don't call this if parallelism is disallowed for the entire query. */
539 14992 : Assert(root->glob->parallelModeOK);
540 :
541 : /* This should only be called for baserels and appendrel children. */
542 14992 : Assert(IS_SIMPLE_REL(rel));
543 :
544 : /* Assorted checks based on rtekind. */
545 14992 : switch (rte->rtekind)
546 : {
547 : case RTE_RELATION:
548 :
549 : /*
550 : * Currently, parallel workers can't access the leader's temporary
551 : * tables. We could possibly relax this if the wrote all of its
552 : * local buffers at the start of the query and made no changes
553 : * thereafter (maybe we could allow hint bit changes), and if we
554 : * taught the workers to read them. Writing a large number of
555 : * temporary buffers could be expensive, though, and we don't have
556 : * the rest of the necessary infrastructure right now anyway. So
557 : * for now, bail out if we see a temporary table.
558 : */
559 12797 : if (get_rel_persistence(rte->relid) == RELPERSISTENCE_TEMP)
560 667 : return;
561 :
562 : /*
563 : * Table sampling can be pushed down to workers if the sample
564 : * function and its arguments are safe.
565 : */
566 12130 : if (rte->tablesample != NULL)
567 : {
568 36 : char proparallel = func_parallel(rte->tablesample->tsmhandler);
569 :
570 36 : if (proparallel != PROPARALLEL_SAFE)
571 0 : return;
572 36 : if (!is_parallel_safe(root, (Node *) rte->tablesample->args))
573 2 : return;
574 : }
575 :
576 : /*
577 : * Ask FDWs whether they can support performing a ForeignScan
578 : * within a worker. Most often, the answer will be no. For
579 : * example, if the nature of the FDW is such that it opens a TCP
580 : * connection with a remote server, each parallel worker would end
581 : * up with a separate connection, and these connections might not
582 : * be appropriately coordinated between workers and the leader.
583 : */
584 12128 : if (rte->relkind == RELKIND_FOREIGN_TABLE)
585 : {
586 0 : Assert(rel->fdwroutine);
587 0 : if (!rel->fdwroutine->IsForeignScanParallelSafe)
588 0 : return;
589 0 : if (!rel->fdwroutine->IsForeignScanParallelSafe(root, rel, rte))
590 0 : return;
591 : }
592 :
593 : /*
594 : * There are additional considerations for appendrels, which we'll
595 : * deal with in set_append_rel_size and set_append_rel_pathlist.
596 : * For now, just set consider_parallel based on the rel's own
597 : * quals and targetlist.
598 : */
599 12128 : break;
600 :
601 : case RTE_SUBQUERY:
602 :
603 : /*
604 : * There's no intrinsic problem with scanning a subquery-in-FROM
605 : * (as distinct from a SubPlan or InitPlan) in a parallel worker.
606 : * If the subquery doesn't happen to have any parallel-safe paths,
607 : * then flagging it as consider_parallel won't change anything,
608 : * but that's true for plain tables, too. We must set
609 : * consider_parallel based on the rel's own quals and targetlist,
610 : * so that if a subquery path is parallel-safe but the quals and
611 : * projection we're sticking onto it are not, we correctly mark
612 : * the SubqueryScanPath as not parallel-safe. (Note that
613 : * set_subquery_pathlist() might push some of these quals down
614 : * into the subquery itself, but that doesn't change anything.)
615 : */
616 874 : break;
617 :
618 : case RTE_JOIN:
619 : /* Shouldn't happen; we're only considering baserels here. */
620 0 : Assert(false);
621 : return;
622 :
623 : case RTE_FUNCTION:
624 : /* Check for parallel-restricted functions. */
625 896 : if (!is_parallel_safe(root, (Node *) rte->functions))
626 451 : return;
627 445 : break;
628 :
629 : case RTE_TABLEFUNC:
630 : /* not parallel safe */
631 22 : return;
632 :
633 : case RTE_VALUES:
634 : /* Check for parallel-restricted functions. */
635 236 : if (!is_parallel_safe(root, (Node *) rte->values_lists))
636 5 : return;
637 231 : break;
638 :
639 : case RTE_CTE:
640 :
641 : /*
642 : * CTE tuplestores aren't shared among parallel workers, so we
643 : * force all CTE scans to happen in the leader. Also, populating
644 : * the CTE would require executing a subplan that's not available
645 : * in the worker, might be parallel-restricted, and must get
646 : * executed only once.
647 : */
648 125 : return;
649 :
650 : case RTE_NAMEDTUPLESTORE:
651 :
652 : /*
653 : * tuplestore cannot be shared, at least without more
654 : * infrastructure to support that.
655 : */
656 42 : return;
657 : }
658 :
659 : /*
660 : * If there's anything in baserestrictinfo that's parallel-restricted, we
661 : * give up on parallelizing access to this relation. We could consider
662 : * instead postponing application of the restricted quals until we're
663 : * above all the parallelism in the plan tree, but it's not clear that
664 : * that would be a win in very many cases, and it might be tricky to make
665 : * outer join clauses work correctly. It would likely break equivalence
666 : * classes, too.
667 : */
668 13678 : if (!is_parallel_safe(root, (Node *) rel->baserestrictinfo))
669 1655 : return;
670 :
671 : /*
672 : * Likewise, if the relation's outputs are not parallel-safe, give up.
673 : * (Usually, they're just Vars, but sometimes they're not.)
674 : */
675 12023 : if (!is_parallel_safe(root, (Node *) rel->reltarget->exprs))
676 2 : return;
677 :
678 : /* We have a winner. */
679 12021 : rel->consider_parallel = true;
680 : }
681 :
682 : /*
683 : * set_plain_rel_pathlist
684 : * Build access paths for a plain relation (no subquery, no inheritance)
685 : */
686 : static void
687 15511 : set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
688 : {
689 : Relids required_outer;
690 :
691 : /*
692 : * We don't support pushing join clauses into the quals of a seqscan, but
693 : * it could still have required parameterization due to LATERAL refs in
694 : * its tlist.
695 : */
696 15511 : required_outer = rel->lateral_relids;
697 :
698 : /* Consider sequential scan */
699 15511 : add_path(rel, create_seqscan_path(root, rel, required_outer, 0));
700 :
701 : /* If appropriate, consider parallel sequential scan */
702 15511 : if (rel->consider_parallel && required_outer == NULL)
703 10254 : create_plain_partial_paths(root, rel);
704 :
705 : /* Consider index scans */
706 15511 : create_index_paths(root, rel);
707 :
708 : /* Consider TID scans */
709 15511 : create_tidscan_paths(root, rel);
710 15511 : }
711 :
712 : /*
713 : * create_plain_partial_paths
714 : * Build partial access paths for parallel scan of a plain relation
715 : */
716 : static void
717 10254 : create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel)
718 : {
719 : int parallel_workers;
720 :
721 10254 : parallel_workers = compute_parallel_worker(rel, rel->pages, -1);
722 :
723 : /* If any limit was set to zero, the user doesn't want a parallel scan. */
724 10254 : if (parallel_workers <= 0)
725 19854 : return;
726 :
727 : /* Add an unordered partial path based on a parallel sequential scan. */
728 654 : add_partial_path(rel, create_seqscan_path(root, rel, NULL, parallel_workers));
729 : }
730 :
731 : /*
732 : * set_tablesample_rel_size
733 : * Set size estimates for a sampled relation
734 : */
735 : static void
736 36 : set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
737 : {
738 36 : TableSampleClause *tsc = rte->tablesample;
739 : TsmRoutine *tsm;
740 : BlockNumber pages;
741 : double tuples;
742 :
743 : /*
744 : * Test any partial indexes of rel for applicability. We must do this
745 : * first since partial unique indexes can affect size estimates.
746 : */
747 36 : check_index_predicates(root, rel);
748 :
749 : /*
750 : * Call the sampling method's estimation function to estimate the number
751 : * of pages it will read and the number of tuples it will return. (Note:
752 : * we assume the function returns sane values.)
753 : */
754 36 : tsm = GetTsmRoutine(tsc->tsmhandler);
755 36 : tsm->SampleScanGetSampleSize(root, rel, tsc->args,
756 : &pages, &tuples);
757 :
758 : /*
759 : * For the moment, because we will only consider a SampleScan path for the
760 : * rel, it's okay to just overwrite the pages and tuples estimates for the
761 : * whole relation. If we ever consider multiple path types for sampled
762 : * rels, we'll need more complication.
763 : */
764 36 : rel->pages = pages;
765 36 : rel->tuples = tuples;
766 :
767 : /* Mark rel with estimated output rows, width, etc */
768 36 : set_baserel_size_estimates(root, rel);
769 36 : }
770 :
771 : /*
772 : * set_tablesample_rel_pathlist
773 : * Build access paths for a sampled relation
774 : */
775 : static void
776 36 : set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
777 : {
778 : Relids required_outer;
779 : Path *path;
780 :
781 : /*
782 : * We don't support pushing join clauses into the quals of a samplescan,
783 : * but it could still have required parameterization due to LATERAL refs
784 : * in its tlist or TABLESAMPLE arguments.
785 : */
786 36 : required_outer = rel->lateral_relids;
787 :
788 : /* Consider sampled scan */
789 36 : path = create_samplescan_path(root, rel, required_outer);
790 :
791 : /*
792 : * If the sampling method does not support repeatable scans, we must avoid
793 : * plans that would scan the rel multiple times. Ideally, we'd simply
794 : * avoid putting the rel on the inside of a nestloop join; but adding such
795 : * a consideration to the planner seems like a great deal of complication
796 : * to support an uncommon usage of second-rate sampling methods. Instead,
797 : * if there is a risk that the query might perform an unsafe join, just
798 : * wrap the SampleScan in a Materialize node. We can check for joins by
799 : * counting the membership of all_baserels (note that this correctly
800 : * counts inheritance trees as single rels). If we're inside a subquery,
801 : * we can't easily check whether a join might occur in the outer query, so
802 : * just assume one is possible.
803 : *
804 : * GetTsmRoutine is relatively expensive compared to the other tests here,
805 : * so check repeatable_across_scans last, even though that's a bit odd.
806 : */
807 70 : if ((root->query_level > 1 ||
808 39 : bms_membership(root->all_baserels) != BMS_SINGLETON) &&
809 5 : !(GetTsmRoutine(rte->tablesample->tsmhandler)->repeatable_across_scans))
810 : {
811 0 : path = (Path *) create_material_path(rel, path);
812 : }
813 :
814 36 : add_path(rel, path);
815 :
816 : /* For the moment, at least, there are no other paths to consider */
817 36 : }
818 :
819 : /*
820 : * set_foreign_size
821 : * Set size estimates for a foreign table RTE
822 : */
823 : static void
824 0 : set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
825 : {
826 : /* Mark rel with estimated output rows, width, etc */
827 0 : set_foreign_size_estimates(root, rel);
828 :
829 : /* Let FDW adjust the size estimates, if it can */
830 0 : rel->fdwroutine->GetForeignRelSize(root, rel, rte->relid);
831 :
832 : /* ... but do not let it set the rows estimate to zero */
833 0 : rel->rows = clamp_row_est(rel->rows);
834 0 : }
835 :
836 : /*
837 : * set_foreign_pathlist
838 : * Build access paths for a foreign table RTE
839 : */
840 : static void
841 0 : set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
842 : {
843 : /* Call the FDW's GetForeignPaths function to generate path(s) */
844 0 : rel->fdwroutine->GetForeignPaths(root, rel, rte->relid);
845 0 : }
846 :
847 : /*
848 : * set_append_rel_size
849 : * Set size estimates for a simple "append relation"
850 : *
851 : * The passed-in rel and RTE represent the entire append relation. The
852 : * relation's contents are computed by appending together the output of the
853 : * individual member relations. Note that in the non-partitioned inheritance
854 : * case, the first member relation is actually the same table as is mentioned
855 : * in the parent RTE ... but it has a different RTE and RelOptInfo. This is
856 : * a good thing because their outputs are not the same size.
857 : */
858 : static void
859 589 : set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
860 : Index rti, RangeTblEntry *rte)
861 : {
862 589 : int parentRTindex = rti;
863 : bool has_live_children;
864 : double parent_rows;
865 : double parent_size;
866 : double *parent_attrsizes;
867 : int nattrs;
868 : ListCell *l;
869 :
870 589 : Assert(IS_SIMPLE_REL(rel));
871 :
872 : /*
873 : * Initialize to compute size estimates for whole append relation.
874 : *
875 : * We handle width estimates by weighting the widths of different child
876 : * rels proportionally to their number of rows. This is sensible because
877 : * the use of width estimates is mainly to compute the total relation
878 : * "footprint" if we have to sort or hash it. To do this, we sum the
879 : * total equivalent size (in "double" arithmetic) and then divide by the
880 : * total rowcount estimate. This is done separately for the total rel
881 : * width and each attribute.
882 : *
883 : * Note: if you consider changing this logic, beware that child rels could
884 : * have zero rows and/or width, if they were excluded by constraints.
885 : */
886 589 : has_live_children = false;
887 589 : parent_rows = 0;
888 589 : parent_size = 0;
889 589 : nattrs = rel->max_attr - rel->min_attr + 1;
890 589 : parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
891 :
892 2495 : foreach(l, root->append_rel_list)
893 : {
894 1906 : AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
895 : int childRTindex;
896 : RangeTblEntry *childRTE;
897 : RelOptInfo *childrel;
898 : List *childquals;
899 : Index cq_min_security;
900 : bool have_const_false_cq;
901 : ListCell *parentvars;
902 : ListCell *childvars;
903 : ListCell *lc;
904 :
905 : /* append_rel_list contains all append rels; ignore others */
906 1906 : if (appinfo->parent_relid != parentRTindex)
907 774 : continue;
908 :
909 1566 : childRTindex = appinfo->child_relid;
910 1566 : childRTE = root->simple_rte_array[childRTindex];
911 :
912 : /*
913 : * The child rel's RelOptInfo was already created during
914 : * add_base_rels_to_query.
915 : */
916 1566 : childrel = find_base_rel(root, childRTindex);
917 1566 : Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
918 :
919 : /*
920 : * We have to copy the parent's targetlist and quals to the child,
921 : * with appropriate substitution of variables. However, only the
922 : * baserestrictinfo quals are needed before we can check for
923 : * constraint exclusion; so do that first and then check to see if we
924 : * can disregard this child.
925 : *
926 : * The child rel's targetlist might contain non-Var expressions, which
927 : * means that substitution into the quals could produce opportunities
928 : * for const-simplification, and perhaps even pseudoconstant quals.
929 : * Therefore, transform each RestrictInfo separately to see if it
930 : * reduces to a constant or pseudoconstant. (We must process them
931 : * separately to keep track of the security level of each qual.)
932 : */
933 1566 : childquals = NIL;
934 1566 : cq_min_security = UINT_MAX;
935 1566 : have_const_false_cq = false;
936 2358 : foreach(lc, rel->baserestrictinfo)
937 : {
938 793 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
939 : Node *childqual;
940 : ListCell *lc2;
941 :
942 793 : Assert(IsA(rinfo, RestrictInfo));
943 793 : childqual = adjust_appendrel_attrs(root,
944 793 : (Node *) rinfo->clause,
945 : 1, &appinfo);
946 793 : childqual = eval_const_expressions(root, childqual);
947 : /* check for flat-out constant */
948 793 : if (childqual && IsA(childqual, Const))
949 : {
950 8 : if (((Const *) childqual)->constisnull ||
951 4 : !DatumGetBool(((Const *) childqual)->constvalue))
952 : {
953 : /* Restriction reduces to constant FALSE or NULL */
954 1 : have_const_false_cq = true;
955 1 : break;
956 : }
957 : /* Restriction reduces to constant TRUE, so drop it */
958 3 : continue;
959 : }
960 : /* might have gotten an AND clause, if so flatten it */
961 1580 : foreach(lc2, make_ands_implicit((Expr *) childqual))
962 : {
963 791 : Node *onecq = (Node *) lfirst(lc2);
964 : bool pseudoconstant;
965 :
966 : /* check for pseudoconstant (no Vars or volatile functions) */
967 791 : pseudoconstant =
968 791 : !contain_vars_of_level(onecq, 0) &&
969 0 : !contain_volatile_functions(onecq);
970 791 : if (pseudoconstant)
971 : {
972 : /* tell createplan.c to check for gating quals */
973 0 : root->hasPseudoConstantQuals = true;
974 : }
975 : /* reconstitute RestrictInfo with appropriate properties */
976 791 : childquals = lappend(childquals,
977 2373 : make_restrictinfo((Expr *) onecq,
978 791 : rinfo->is_pushed_down,
979 791 : rinfo->outerjoin_delayed,
980 : pseudoconstant,
981 : rinfo->security_level,
982 : NULL, NULL, NULL));
983 : /* track minimum security level among child quals */
984 791 : cq_min_security = Min(cq_min_security, rinfo->security_level);
985 : }
986 : }
987 :
988 : /*
989 : * In addition to the quals inherited from the parent, we might have
990 : * securityQuals associated with this particular child node.
991 : * (Currently this can only happen in appendrels originating from
992 : * UNION ALL; inheritance child tables don't have their own
993 : * securityQuals, see expand_inherited_rtentry().) Pull any such
994 : * securityQuals up into the baserestrictinfo for the child. This is
995 : * similar to process_security_barrier_quals() for the parent rel,
996 : * except that we can't make any general deductions from such quals,
997 : * since they don't hold for the whole appendrel.
998 : */
999 1566 : if (childRTE->securityQuals)
1000 : {
1001 2 : Index security_level = 0;
1002 :
1003 4 : foreach(lc, childRTE->securityQuals)
1004 : {
1005 2 : List *qualset = (List *) lfirst(lc);
1006 : ListCell *lc2;
1007 :
1008 4 : foreach(lc2, qualset)
1009 : {
1010 2 : Expr *qual = (Expr *) lfirst(lc2);
1011 :
1012 : /* not likely that we'd see constants here, so no check */
1013 2 : childquals = lappend(childquals,
1014 2 : make_restrictinfo(qual,
1015 : true, false, false,
1016 : security_level,
1017 : NULL, NULL, NULL));
1018 2 : cq_min_security = Min(cq_min_security, security_level);
1019 : }
1020 2 : security_level++;
1021 : }
1022 2 : Assert(security_level <= root->qual_security_level);
1023 : }
1024 :
1025 : /*
1026 : * OK, we've got all the baserestrictinfo quals for this child.
1027 : */
1028 1566 : childrel->baserestrictinfo = childquals;
1029 1566 : childrel->baserestrict_min_security = cq_min_security;
1030 :
1031 1566 : if (have_const_false_cq)
1032 : {
1033 : /*
1034 : * Some restriction clause reduced to constant FALSE or NULL after
1035 : * substitution, so this child need not be scanned.
1036 : */
1037 1 : set_dummy_rel_pathlist(childrel);
1038 1 : continue;
1039 : }
1040 :
1041 1565 : if (relation_excluded_by_constraints(root, childrel, childRTE))
1042 : {
1043 : /*
1044 : * This child need not be scanned, so we can omit it from the
1045 : * appendrel.
1046 : */
1047 91 : set_dummy_rel_pathlist(childrel);
1048 91 : continue;
1049 : }
1050 :
1051 : /*
1052 : * CE failed, so finish copying/modifying targetlist and join quals.
1053 : *
1054 : * NB: the resulting childrel->reltarget->exprs may contain arbitrary
1055 : * expressions, which otherwise would not occur in a rel's targetlist.
1056 : * Code that might be looking at an appendrel child must cope with
1057 : * such. (Normally, a rel's targetlist would only include Vars and
1058 : * PlaceHolderVars.) XXX we do not bother to update the cost or width
1059 : * fields of childrel->reltarget; not clear if that would be useful.
1060 : */
1061 1474 : childrel->joininfo = (List *)
1062 1474 : adjust_appendrel_attrs(root,
1063 1474 : (Node *) rel->joininfo,
1064 : 1, &appinfo);
1065 2948 : childrel->reltarget->exprs = (List *)
1066 1474 : adjust_appendrel_attrs(root,
1067 1474 : (Node *) rel->reltarget->exprs,
1068 : 1, &appinfo);
1069 :
1070 : /*
1071 : * We have to make child entries in the EquivalenceClass data
1072 : * structures as well. This is needed either if the parent
1073 : * participates in some eclass joins (because we will want to consider
1074 : * inner-indexscan joins on the individual children) or if the parent
1075 : * has useful pathkeys (because we should try to build MergeAppend
1076 : * paths that produce those sort orderings).
1077 : */
1078 1474 : if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
1079 845 : add_child_rel_equivalences(root, appinfo, rel, childrel);
1080 1474 : childrel->has_eclass_joins = rel->has_eclass_joins;
1081 :
1082 : /*
1083 : * Note: we could compute appropriate attr_needed data for the child's
1084 : * variables, by transforming the parent's attr_needed through the
1085 : * translated_vars mapping. However, currently there's no need
1086 : * because attr_needed is only examined for base relations not
1087 : * otherrels. So we just leave the child's attr_needed empty.
1088 : */
1089 :
1090 : /*
1091 : * If parallelism is allowable for this query in general, see whether
1092 : * it's allowable for this childrel in particular. But if we've
1093 : * already decided the appendrel is not parallel-safe as a whole,
1094 : * there's no point in considering parallelism for this child. For
1095 : * consistency, do this before calling set_rel_size() for the child.
1096 : */
1097 1474 : if (root->glob->parallelModeOK && rel->consider_parallel)
1098 1048 : set_rel_consider_parallel(root, childrel, childRTE);
1099 :
1100 : /*
1101 : * Compute the child's size.
1102 : */
1103 1474 : set_rel_size(root, childrel, childRTindex, childRTE);
1104 :
1105 : /*
1106 : * It is possible that constraint exclusion detected a contradiction
1107 : * within a child subquery, even though we didn't prove one above. If
1108 : * so, we can skip this child.
1109 : */
1110 1474 : if (IS_DUMMY_REL(childrel))
1111 2 : continue;
1112 :
1113 : /* We have at least one live child. */
1114 1472 : has_live_children = true;
1115 :
1116 : /*
1117 : * If any live child is not parallel-safe, treat the whole appendrel
1118 : * as not parallel-safe. In future we might be able to generate plans
1119 : * in which some children are farmed out to workers while others are
1120 : * not; but we don't have that today, so it's a waste to consider
1121 : * partial paths anywhere in the appendrel unless it's all safe.
1122 : * (Child rels visited before this one will be unmarked in
1123 : * set_append_rel_pathlist().)
1124 : */
1125 1472 : if (!childrel->consider_parallel)
1126 434 : rel->consider_parallel = false;
1127 :
1128 : /*
1129 : * Accumulate size information from each live child.
1130 : */
1131 1472 : Assert(childrel->rows > 0);
1132 :
1133 1472 : parent_rows += childrel->rows;
1134 1472 : parent_size += childrel->reltarget->width * childrel->rows;
1135 :
1136 : /*
1137 : * Accumulate per-column estimates too. We need not do anything for
1138 : * PlaceHolderVars in the parent list. If child expression isn't a
1139 : * Var, or we didn't record a width estimate for it, we have to fall
1140 : * back on a datatype-based estimate.
1141 : *
1142 : * By construction, child's targetlist is 1-to-1 with parent's.
1143 : */
1144 4238 : forboth(parentvars, rel->reltarget->exprs,
1145 : childvars, childrel->reltarget->exprs)
1146 : {
1147 2766 : Var *parentvar = (Var *) lfirst(parentvars);
1148 2766 : Node *childvar = (Node *) lfirst(childvars);
1149 :
1150 2766 : if (IsA(parentvar, Var))
1151 : {
1152 2676 : int pndx = parentvar->varattno - rel->min_attr;
1153 2676 : int32 child_width = 0;
1154 :
1155 5258 : if (IsA(childvar, Var) &&
1156 2582 : ((Var *) childvar)->varno == childrel->relid)
1157 : {
1158 2579 : int cndx = ((Var *) childvar)->varattno - childrel->min_attr;
1159 :
1160 2579 : child_width = childrel->attr_widths[cndx];
1161 : }
1162 2676 : if (child_width <= 0)
1163 97 : child_width = get_typavgwidth(exprType(childvar),
1164 : exprTypmod(childvar));
1165 2676 : Assert(child_width > 0);
1166 2676 : parent_attrsizes[pndx] += child_width * childrel->rows;
1167 : }
1168 : }
1169 : }
1170 :
1171 589 : if (has_live_children)
1172 : {
1173 : /*
1174 : * Save the finished size estimates.
1175 : */
1176 : int i;
1177 :
1178 584 : Assert(parent_rows > 0);
1179 584 : rel->rows = parent_rows;
1180 584 : rel->reltarget->width = rint(parent_size / parent_rows);
1181 4464 : for (i = 0; i < nattrs; i++)
1182 3880 : rel->attr_widths[i] = rint(parent_attrsizes[i] / parent_rows);
1183 :
1184 : /*
1185 : * Set "raw tuples" count equal to "rows" for the appendrel; needed
1186 : * because some places assume rel->tuples is valid for any baserel.
1187 : */
1188 584 : rel->tuples = parent_rows;
1189 : }
1190 : else
1191 : {
1192 : /*
1193 : * All children were excluded by constraints, so mark the whole
1194 : * appendrel dummy. We must do this in this phase so that the rel's
1195 : * dummy-ness is visible when we generate paths for other rels.
1196 : */
1197 5 : set_dummy_rel_pathlist(rel);
1198 : }
1199 :
1200 589 : pfree(parent_attrsizes);
1201 589 : }
1202 :
1203 : /*
1204 : * set_append_rel_pathlist
1205 : * Build access paths for an "append relation"
1206 : */
1207 : static void
1208 584 : set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
1209 : Index rti, RangeTblEntry *rte)
1210 : {
1211 584 : int parentRTindex = rti;
1212 584 : List *live_childrels = NIL;
1213 : ListCell *l;
1214 :
1215 : /*
1216 : * Generate access paths for each member relation, and remember the
1217 : * non-dummy children.
1218 : */
1219 2469 : foreach(l, root->append_rel_list)
1220 : {
1221 1885 : AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1222 : int childRTindex;
1223 : RangeTblEntry *childRTE;
1224 : RelOptInfo *childrel;
1225 :
1226 : /* append_rel_list contains all append rels; ignore others */
1227 1885 : if (appinfo->parent_relid != parentRTindex)
1228 338 : continue;
1229 :
1230 : /* Re-locate the child RTE and RelOptInfo */
1231 1547 : childRTindex = appinfo->child_relid;
1232 1547 : childRTE = root->simple_rte_array[childRTindex];
1233 1547 : childrel = root->simple_rel_array[childRTindex];
1234 :
1235 : /*
1236 : * If set_append_rel_size() decided the parent appendrel was
1237 : * parallel-unsafe at some point after visiting this child rel, we
1238 : * need to propagate the unsafety marking down to the child, so that
1239 : * we don't generate useless partial paths for it.
1240 : */
1241 1547 : if (!rel->consider_parallel)
1242 445 : childrel->consider_parallel = false;
1243 :
1244 : /*
1245 : * Compute the child's access paths.
1246 : */
1247 1547 : set_rel_pathlist(root, childrel, childRTindex, childRTE);
1248 :
1249 : /*
1250 : * If child is dummy, ignore it.
1251 : */
1252 1547 : if (IS_DUMMY_REL(childrel))
1253 75 : continue;
1254 :
1255 : /*
1256 : * Child is live, so add it to the live_childrels list for use below.
1257 : */
1258 1472 : live_childrels = lappend(live_childrels, childrel);
1259 : }
1260 :
1261 : /* Add paths to the "append" relation. */
1262 584 : add_paths_to_append_rel(root, rel, live_childrels);
1263 584 : }
1264 :
1265 :
1266 : /*
1267 : * add_paths_to_append_rel
1268 : * Generate paths for given "append" relation given the set of non-dummy
1269 : * child rels.
1270 : *
1271 : * The function collects all parameterizations and orderings supported by the
1272 : * non-dummy children. For every such parameterization or ordering, it creates
1273 : * an append path collecting one path from each non-dummy child with given
1274 : * parameterization or ordering. Similarly it collects partial paths from
1275 : * non-dummy children to create partial append paths.
1276 : */
1277 : static void
1278 584 : add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
1279 : List *live_childrels)
1280 : {
1281 584 : List *subpaths = NIL;
1282 584 : bool subpaths_valid = true;
1283 584 : List *partial_subpaths = NIL;
1284 584 : bool partial_subpaths_valid = true;
1285 584 : List *all_child_pathkeys = NIL;
1286 584 : List *all_child_outers = NIL;
1287 : ListCell *l;
1288 584 : List *partitioned_rels = NIL;
1289 : RangeTblEntry *rte;
1290 :
1291 584 : rte = planner_rt_fetch(rel->relid, root);
1292 584 : if (rte->relkind == RELKIND_PARTITIONED_TABLE)
1293 : {
1294 62 : partitioned_rels = get_partitioned_child_rels(root, rel->relid);
1295 : /* The root partitioned table is included as a child rel */
1296 62 : Assert(list_length(partitioned_rels) >= 1);
1297 : }
1298 :
1299 : /*
1300 : * For every non-dummy child, remember the cheapest path. Also, identify
1301 : * all pathkeys (orderings) and parameterizations (required_outer sets)
1302 : * available for the non-dummy member relations.
1303 : */
1304 2056 : foreach(l, live_childrels)
1305 : {
1306 1472 : RelOptInfo *childrel = lfirst(l);
1307 : ListCell *lcp;
1308 :
1309 : /*
1310 : * If child has an unparameterized cheapest-total path, add that to
1311 : * the unparameterized Append path we are constructing for the parent.
1312 : * If not, there's no workable unparameterized path.
1313 : */
1314 1472 : if (childrel->cheapest_total_path->param_info == NULL)
1315 1462 : subpaths = accumulate_append_subpath(subpaths,
1316 1462 : childrel->cheapest_total_path);
1317 : else
1318 10 : subpaths_valid = false;
1319 :
1320 : /* Same idea, but for a partial plan. */
1321 1472 : if (childrel->partial_pathlist != NIL)
1322 618 : partial_subpaths = accumulate_append_subpath(partial_subpaths,
1323 618 : linitial(childrel->partial_pathlist));
1324 : else
1325 854 : partial_subpaths_valid = false;
1326 :
1327 : /*
1328 : * Collect lists of all the available path orderings and
1329 : * parameterizations for all the children. We use these as a
1330 : * heuristic to indicate which sort orderings and parameterizations we
1331 : * should build Append and MergeAppend paths for.
1332 : */
1333 3170 : foreach(lcp, childrel->pathlist)
1334 : {
1335 1698 : Path *childpath = (Path *) lfirst(lcp);
1336 1698 : List *childkeys = childpath->pathkeys;
1337 1698 : Relids childouter = PATH_REQ_OUTER(childpath);
1338 :
1339 : /* Unsorted paths don't contribute to pathkey list */
1340 1698 : if (childkeys != NIL)
1341 : {
1342 : ListCell *lpk;
1343 233 : bool found = false;
1344 :
1345 : /* Have we already seen this ordering? */
1346 233 : foreach(lpk, all_child_pathkeys)
1347 : {
1348 152 : List *existing_pathkeys = (List *) lfirst(lpk);
1349 :
1350 152 : if (compare_pathkeys(existing_pathkeys,
1351 : childkeys) == PATHKEYS_EQUAL)
1352 : {
1353 152 : found = true;
1354 152 : break;
1355 : }
1356 : }
1357 233 : if (!found)
1358 : {
1359 : /* No, so add it to all_child_pathkeys */
1360 81 : all_child_pathkeys = lappend(all_child_pathkeys,
1361 : childkeys);
1362 : }
1363 : }
1364 :
1365 : /* Unparameterized paths don't contribute to param-set list */
1366 1698 : if (childouter)
1367 : {
1368 : ListCell *lco;
1369 90 : bool found = false;
1370 :
1371 : /* Have we already seen this param set? */
1372 106 : foreach(lco, all_child_outers)
1373 : {
1374 53 : Relids existing_outers = (Relids) lfirst(lco);
1375 :
1376 53 : if (bms_equal(existing_outers, childouter))
1377 : {
1378 37 : found = true;
1379 37 : break;
1380 : }
1381 : }
1382 90 : if (!found)
1383 : {
1384 : /* No, so add it to all_child_outers */
1385 53 : all_child_outers = lappend(all_child_outers,
1386 : childouter);
1387 : }
1388 : }
1389 : }
1390 : }
1391 :
1392 : /*
1393 : * If we found unparameterized paths for all children, build an unordered,
1394 : * unparameterized Append path for the rel. (Note: this is correct even
1395 : * if we have zero or one live subpath due to constraint exclusion.)
1396 : */
1397 584 : if (subpaths_valid)
1398 579 : add_path(rel, (Path *) create_append_path(rel, subpaths, NULL, 0,
1399 : partitioned_rels));
1400 :
1401 : /*
1402 : * Consider an append of partial unordered, unparameterized partial paths.
1403 : */
1404 584 : if (partial_subpaths_valid)
1405 : {
1406 : AppendPath *appendpath;
1407 : ListCell *lc;
1408 199 : int parallel_workers = 0;
1409 :
1410 : /*
1411 : * Decide on the number of workers to request for this append path.
1412 : * For now, we just use the maximum value from among the members. It
1413 : * might be useful to use a higher number if the Append node were
1414 : * smart enough to spread out the workers, but it currently isn't.
1415 : */
1416 799 : foreach(lc, partial_subpaths)
1417 : {
1418 600 : Path *path = lfirst(lc);
1419 :
1420 600 : parallel_workers = Max(parallel_workers, path->parallel_workers);
1421 : }
1422 199 : Assert(parallel_workers > 0);
1423 :
1424 : /* Generate a partial append path. */
1425 199 : appendpath = create_append_path(rel, partial_subpaths, NULL,
1426 : parallel_workers, partitioned_rels);
1427 199 : add_partial_path(rel, (Path *) appendpath);
1428 : }
1429 :
1430 : /*
1431 : * Also build unparameterized MergeAppend paths based on the collected
1432 : * list of child pathkeys.
1433 : */
1434 584 : if (subpaths_valid)
1435 579 : generate_mergeappend_paths(root, rel, live_childrels,
1436 : all_child_pathkeys,
1437 : partitioned_rels);
1438 :
1439 : /*
1440 : * Build Append paths for each parameterization seen among the child rels.
1441 : * (This may look pretty expensive, but in most cases of practical
1442 : * interest, the child rels will expose mostly the same parameterizations,
1443 : * so that not that many cases actually get considered here.)
1444 : *
1445 : * The Append node itself cannot enforce quals, so all qual checking must
1446 : * be done in the child paths. This means that to have a parameterized
1447 : * Append path, we must have the exact same parameterization for each
1448 : * child path; otherwise some children might be failing to check the
1449 : * moved-down quals. To make them match up, we can try to increase the
1450 : * parameterization of lesser-parameterized paths.
1451 : */
1452 637 : foreach(l, all_child_outers)
1453 : {
1454 53 : Relids required_outer = (Relids) lfirst(l);
1455 : ListCell *lcr;
1456 :
1457 : /* Select the child paths for an Append with this parameterization */
1458 53 : subpaths = NIL;
1459 53 : subpaths_valid = true;
1460 165 : foreach(lcr, live_childrels)
1461 : {
1462 112 : RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1463 : Path *subpath;
1464 :
1465 112 : subpath = get_cheapest_parameterized_child_path(root,
1466 : childrel,
1467 : required_outer);
1468 112 : if (subpath == NULL)
1469 : {
1470 : /* failed to make a suitable path for this child */
1471 0 : subpaths_valid = false;
1472 0 : break;
1473 : }
1474 112 : subpaths = accumulate_append_subpath(subpaths, subpath);
1475 : }
1476 :
1477 53 : if (subpaths_valid)
1478 53 : add_path(rel, (Path *)
1479 53 : create_append_path(rel, subpaths, required_outer, 0,
1480 : partitioned_rels));
1481 : }
1482 584 : }
1483 :
1484 : /*
1485 : * generate_mergeappend_paths
1486 : * Generate MergeAppend paths for an append relation
1487 : *
1488 : * Generate a path for each ordering (pathkey list) appearing in
1489 : * all_child_pathkeys.
1490 : *
1491 : * We consider both cheapest-startup and cheapest-total cases, ie, for each
1492 : * interesting ordering, collect all the cheapest startup subpaths and all the
1493 : * cheapest total paths, and build a MergeAppend path for each case.
1494 : *
1495 : * We don't currently generate any parameterized MergeAppend paths. While
1496 : * it would not take much more code here to do so, it's very unclear that it
1497 : * is worth the planning cycles to investigate such paths: there's little
1498 : * use for an ordered path on the inside of a nestloop. In fact, it's likely
1499 : * that the current coding of add_path would reject such paths out of hand,
1500 : * because add_path gives no credit for sort ordering of parameterized paths,
1501 : * and a parameterized MergeAppend is going to be more expensive than the
1502 : * corresponding parameterized Append path. If we ever try harder to support
1503 : * parameterized mergejoin plans, it might be worth adding support for
1504 : * parameterized MergeAppends to feed such joins. (See notes in
1505 : * optimizer/README for why that might not ever happen, though.)
1506 : */
1507 : static void
1508 579 : generate_mergeappend_paths(PlannerInfo *root, RelOptInfo *rel,
1509 : List *live_childrels,
1510 : List *all_child_pathkeys,
1511 : List *partitioned_rels)
1512 : {
1513 : ListCell *lcp;
1514 :
1515 660 : foreach(lcp, all_child_pathkeys)
1516 : {
1517 81 : List *pathkeys = (List *) lfirst(lcp);
1518 81 : List *startup_subpaths = NIL;
1519 81 : List *total_subpaths = NIL;
1520 81 : bool startup_neq_total = false;
1521 : ListCell *lcr;
1522 :
1523 : /* Select the child paths for this ordering... */
1524 281 : foreach(lcr, live_childrels)
1525 : {
1526 200 : RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1527 : Path *cheapest_startup,
1528 : *cheapest_total;
1529 :
1530 : /* Locate the right paths, if they are available. */
1531 200 : cheapest_startup =
1532 200 : get_cheapest_path_for_pathkeys(childrel->pathlist,
1533 : pathkeys,
1534 : NULL,
1535 : STARTUP_COST,
1536 : false);
1537 200 : cheapest_total =
1538 200 : get_cheapest_path_for_pathkeys(childrel->pathlist,
1539 : pathkeys,
1540 : NULL,
1541 : TOTAL_COST,
1542 : false);
1543 :
1544 : /*
1545 : * If we can't find any paths with the right order just use the
1546 : * cheapest-total path; we'll have to sort it later.
1547 : */
1548 200 : if (cheapest_startup == NULL || cheapest_total == NULL)
1549 : {
1550 36 : cheapest_startup = cheapest_total =
1551 : childrel->cheapest_total_path;
1552 : /* Assert we do have an unparameterized path for this child */
1553 36 : Assert(cheapest_total->param_info == NULL);
1554 : }
1555 :
1556 : /*
1557 : * Notice whether we actually have different paths for the
1558 : * "cheapest" and "total" cases; frequently there will be no point
1559 : * in two create_merge_append_path() calls.
1560 : */
1561 200 : if (cheapest_startup != cheapest_total)
1562 0 : startup_neq_total = true;
1563 :
1564 200 : startup_subpaths =
1565 : accumulate_append_subpath(startup_subpaths, cheapest_startup);
1566 200 : total_subpaths =
1567 : accumulate_append_subpath(total_subpaths, cheapest_total);
1568 : }
1569 :
1570 : /* ... and build the MergeAppend paths */
1571 81 : add_path(rel, (Path *) create_merge_append_path(root,
1572 : rel,
1573 : startup_subpaths,
1574 : pathkeys,
1575 : NULL,
1576 : partitioned_rels));
1577 81 : if (startup_neq_total)
1578 0 : add_path(rel, (Path *) create_merge_append_path(root,
1579 : rel,
1580 : total_subpaths,
1581 : pathkeys,
1582 : NULL,
1583 : partitioned_rels));
1584 : }
1585 579 : }
1586 :
1587 : /*
1588 : * get_cheapest_parameterized_child_path
1589 : * Get cheapest path for this relation that has exactly the requested
1590 : * parameterization.
1591 : *
1592 : * Returns NULL if unable to create such a path.
1593 : */
1594 : static Path *
1595 112 : get_cheapest_parameterized_child_path(PlannerInfo *root, RelOptInfo *rel,
1596 : Relids required_outer)
1597 : {
1598 : Path *cheapest;
1599 : ListCell *lc;
1600 :
1601 : /*
1602 : * Look up the cheapest existing path with no more than the needed
1603 : * parameterization. If it has exactly the needed parameterization, we're
1604 : * done.
1605 : */
1606 112 : cheapest = get_cheapest_path_for_pathkeys(rel->pathlist,
1607 : NIL,
1608 : required_outer,
1609 : TOTAL_COST,
1610 : false);
1611 112 : Assert(cheapest != NULL);
1612 112 : if (bms_equal(PATH_REQ_OUTER(cheapest), required_outer))
1613 72 : return cheapest;
1614 :
1615 : /*
1616 : * Otherwise, we can "reparameterize" an existing path to match the given
1617 : * parameterization, which effectively means pushing down additional
1618 : * joinquals to be checked within the path's scan. However, some existing
1619 : * paths might check the available joinquals already while others don't;
1620 : * therefore, it's not clear which existing path will be cheapest after
1621 : * reparameterization. We have to go through them all and find out.
1622 : */
1623 40 : cheapest = NULL;
1624 118 : foreach(lc, rel->pathlist)
1625 : {
1626 78 : Path *path = (Path *) lfirst(lc);
1627 :
1628 : /* Can't use it if it needs more than requested parameterization */
1629 78 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
1630 0 : continue;
1631 :
1632 : /*
1633 : * Reparameterization can only increase the path's cost, so if it's
1634 : * already more expensive than the current cheapest, forget it.
1635 : */
1636 116 : if (cheapest != NULL &&
1637 38 : compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
1638 38 : continue;
1639 :
1640 : /* Reparameterize if needed, then recheck cost */
1641 40 : if (!bms_equal(PATH_REQ_OUTER(path), required_outer))
1642 : {
1643 40 : path = reparameterize_path(root, path, required_outer, 1.0);
1644 40 : if (path == NULL)
1645 0 : continue; /* failed to reparameterize this one */
1646 40 : Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));
1647 :
1648 40 : if (cheapest != NULL &&
1649 0 : compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
1650 0 : continue;
1651 : }
1652 :
1653 : /* We have a new best path */
1654 40 : cheapest = path;
1655 : }
1656 :
1657 : /* Return the best path, or NULL if we found no suitable candidate */
1658 40 : return cheapest;
1659 : }
1660 :
1661 : /*
1662 : * accumulate_append_subpath
1663 : * Add a subpath to the list being built for an Append or MergeAppend
1664 : *
1665 : * It's possible that the child is itself an Append or MergeAppend path, in
1666 : * which case we can "cut out the middleman" and just add its child paths to
1667 : * our own list. (We don't try to do this earlier because we need to apply
1668 : * both levels of transformation to the quals.)
1669 : *
1670 : * Note that if we omit a child MergeAppend in this way, we are effectively
1671 : * omitting a sort step, which seems fine: if the parent is to be an Append,
1672 : * its result would be unsorted anyway, while if the parent is to be a
1673 : * MergeAppend, there's no point in a separate sort on a child.
1674 : */
1675 : static List *
1676 2592 : accumulate_append_subpath(List *subpaths, Path *path)
1677 : {
1678 2592 : if (IsA(path, AppendPath))
1679 : {
1680 43 : AppendPath *apath = (AppendPath *) path;
1681 :
1682 : /* list_copy is important here to avoid sharing list substructure */
1683 43 : return list_concat(subpaths, list_copy(apath->subpaths));
1684 : }
1685 2549 : else if (IsA(path, MergeAppendPath))
1686 : {
1687 24 : MergeAppendPath *mpath = (MergeAppendPath *) path;
1688 :
1689 : /* list_copy is important here to avoid sharing list substructure */
1690 24 : return list_concat(subpaths, list_copy(mpath->subpaths));
1691 : }
1692 : else
1693 2525 : return lappend(subpaths, path);
1694 : }
1695 :
1696 : /*
1697 : * set_dummy_rel_pathlist
1698 : * Build a dummy path for a relation that's been excluded by constraints
1699 : *
1700 : * Rather than inventing a special "dummy" path type, we represent this as an
1701 : * AppendPath with no members (see also IS_DUMMY_PATH/IS_DUMMY_REL macros).
1702 : *
1703 : * This is exported because inheritance_planner() has need for it.
1704 : */
1705 : void
1706 132 : set_dummy_rel_pathlist(RelOptInfo *rel)
1707 : {
1708 : /* Set dummy size estimates --- we leave attr_widths[] as zeroes */
1709 132 : rel->rows = 0;
1710 132 : rel->reltarget->width = 0;
1711 :
1712 : /* Discard any pre-existing paths; no further need for them */
1713 132 : rel->pathlist = NIL;
1714 132 : rel->partial_pathlist = NIL;
1715 :
1716 132 : add_path(rel, (Path *) create_append_path(rel, NIL, NULL, 0, NIL));
1717 :
1718 : /*
1719 : * We set the cheapest path immediately, to ensure that IS_DUMMY_REL()
1720 : * will recognize the relation as dummy if anyone asks. This is redundant
1721 : * when we're called from set_rel_size(), but not when called from
1722 : * elsewhere, and doing it twice is harmless anyway.
1723 : */
1724 132 : set_cheapest(rel);
1725 132 : }
1726 :
1727 : /* quick-and-dirty test to see if any joining is needed */
1728 : static bool
1729 328 : has_multiple_baserels(PlannerInfo *root)
1730 : {
1731 328 : int num_base_rels = 0;
1732 : Index rti;
1733 :
1734 1016 : for (rti = 1; rti < root->simple_rel_array_size; rti++)
1735 : {
1736 841 : RelOptInfo *brel = root->simple_rel_array[rti];
1737 :
1738 841 : if (brel == NULL)
1739 283 : continue;
1740 :
1741 : /* ignore RTEs that are "other rels" */
1742 558 : if (brel->reloptkind == RELOPT_BASEREL)
1743 481 : if (++num_base_rels > 1)
1744 153 : return true;
1745 : }
1746 175 : return false;
1747 : }
1748 :
1749 : /*
1750 : * set_subquery_pathlist
1751 : * Generate SubqueryScan access paths for a subquery RTE
1752 : *
1753 : * We don't currently support generating parameterized paths for subqueries
1754 : * by pushing join clauses down into them; it seems too expensive to re-plan
1755 : * the subquery multiple times to consider different alternatives.
1756 : * (XXX that could stand to be reconsidered, now that we use Paths.)
1757 : * So the paths made here will be parameterized if the subquery contains
1758 : * LATERAL references, otherwise not. As long as that's true, there's no need
1759 : * for a separate set_subquery_size phase: just make the paths right away.
1760 : */
1761 : static void
1762 790 : set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
1763 : Index rti, RangeTblEntry *rte)
1764 : {
1765 790 : Query *parse = root->parse;
1766 790 : Query *subquery = rte->subquery;
1767 : Relids required_outer;
1768 : pushdown_safety_info safetyInfo;
1769 : double tuple_fraction;
1770 : RelOptInfo *sub_final_rel;
1771 : ListCell *lc;
1772 :
1773 : /*
1774 : * Must copy the Query so that planning doesn't mess up the RTE contents
1775 : * (really really need to fix the planner to not scribble on its input,
1776 : * someday ... but see remove_unused_subquery_outputs to start with).
1777 : */
1778 790 : subquery = copyObject(subquery);
1779 :
1780 : /*
1781 : * If it's a LATERAL subquery, it might contain some Vars of the current
1782 : * query level, requiring it to be treated as parameterized, even though
1783 : * we don't support pushing down join quals into subqueries.
1784 : */
1785 790 : required_outer = rel->lateral_relids;
1786 :
1787 : /*
1788 : * Zero out result area for subquery_is_pushdown_safe, so that it can set
1789 : * flags as needed while recursing. In particular, we need a workspace
1790 : * for keeping track of unsafe-to-reference columns. unsafeColumns[i]
1791 : * will be set TRUE if we find that output column i of the subquery is
1792 : * unsafe to use in a pushed-down qual.
1793 : */
1794 790 : memset(&safetyInfo, 0, sizeof(safetyInfo));
1795 790 : safetyInfo.unsafeColumns = (bool *)
1796 790 : palloc0((list_length(subquery->targetList) + 1) * sizeof(bool));
1797 :
1798 : /*
1799 : * If the subquery has the "security_barrier" flag, it means the subquery
1800 : * originated from a view that must enforce row level security. Then we
1801 : * must not push down quals that contain leaky functions. (Ideally this
1802 : * would be checked inside subquery_is_pushdown_safe, but since we don't
1803 : * currently pass the RTE to that function, we must do it here.)
1804 : */
1805 790 : safetyInfo.unsafeLeaky = rte->security_barrier;
1806 :
1807 : /*
1808 : * If there are any restriction clauses that have been attached to the
1809 : * subquery relation, consider pushing them down to become WHERE or HAVING
1810 : * quals of the subquery itself. This transformation is useful because it
1811 : * may allow us to generate a better plan for the subquery than evaluating
1812 : * all the subquery output rows and then filtering them.
1813 : *
1814 : * There are several cases where we cannot push down clauses. Restrictions
1815 : * involving the subquery are checked by subquery_is_pushdown_safe().
1816 : * Restrictions on individual clauses are checked by
1817 : * qual_is_pushdown_safe(). Also, we don't want to push down
1818 : * pseudoconstant clauses; better to have the gating node above the
1819 : * subquery.
1820 : *
1821 : * Non-pushed-down clauses will get evaluated as qpquals of the
1822 : * SubqueryScan node.
1823 : *
1824 : * XXX Are there any cases where we want to make a policy decision not to
1825 : * push down a pushable qual, because it'd result in a worse plan?
1826 : */
1827 930 : if (rel->baserestrictinfo != NIL &&
1828 140 : subquery_is_pushdown_safe(subquery, subquery, &safetyInfo))
1829 : {
1830 : /* OK to consider pushing down individual quals */
1831 125 : List *upperrestrictlist = NIL;
1832 : ListCell *l;
1833 :
1834 330 : foreach(l, rel->baserestrictinfo)
1835 : {
1836 205 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(l);
1837 205 : Node *clause = (Node *) rinfo->clause;
1838 :
1839 410 : if (!rinfo->pseudoconstant &&
1840 205 : qual_is_pushdown_safe(subquery, rti, clause, &safetyInfo))
1841 : {
1842 : /* Push it down */
1843 136 : subquery_push_qual(subquery, rte, rti, clause);
1844 : }
1845 : else
1846 : {
1847 : /* Keep it in the upper query */
1848 69 : upperrestrictlist = lappend(upperrestrictlist, rinfo);
1849 : }
1850 : }
1851 125 : rel->baserestrictinfo = upperrestrictlist;
1852 : /* We don't bother recomputing baserestrict_min_security */
1853 : }
1854 :
1855 790 : pfree(safetyInfo.unsafeColumns);
1856 :
1857 : /*
1858 : * The upper query might not use all the subquery's output columns; if
1859 : * not, we can simplify.
1860 : */
1861 790 : remove_unused_subquery_outputs(subquery, rel);
1862 :
1863 : /*
1864 : * We can safely pass the outer tuple_fraction down to the subquery if the
1865 : * outer level has no joining, aggregation, or sorting to do. Otherwise
1866 : * we'd better tell the subquery to plan for full retrieval. (XXX This
1867 : * could probably be made more intelligent ...)
1868 : */
1869 1553 : if (parse->hasAggs ||
1870 1525 : parse->groupClause ||
1871 1524 : parse->groupingSets ||
1872 1524 : parse->havingQual ||
1873 1523 : parse->distinctClause ||
1874 1089 : parse->sortClause ||
1875 328 : has_multiple_baserels(root))
1876 615 : tuple_fraction = 0.0; /* default case */
1877 : else
1878 175 : tuple_fraction = root->tuple_fraction;
1879 :
1880 : /* plan_params should not be in use in current query level */
1881 790 : Assert(root->plan_params == NIL);
1882 :
1883 : /* Generate a subroot and Paths for the subquery */
1884 790 : rel->subroot = subquery_planner(root->glob, subquery,
1885 : root,
1886 : false, tuple_fraction);
1887 :
1888 : /* Isolate the params needed by this specific subplan */
1889 790 : rel->subplan_params = root->plan_params;
1890 790 : root->plan_params = NIL;
1891 :
1892 : /*
1893 : * It's possible that constraint exclusion proved the subquery empty. If
1894 : * so, it's desirable to produce an unadorned dummy path so that we will
1895 : * recognize appropriate optimizations at this query level.
1896 : */
1897 790 : sub_final_rel = fetch_upper_rel(rel->subroot, UPPERREL_FINAL, NULL);
1898 :
1899 790 : if (IS_DUMMY_REL(sub_final_rel))
1900 : {
1901 3 : set_dummy_rel_pathlist(rel);
1902 793 : return;
1903 : }
1904 :
1905 : /*
1906 : * Mark rel with estimated output rows, width, etc. Note that we have to
1907 : * do this before generating outer-query paths, else cost_subqueryscan is
1908 : * not happy.
1909 : */
1910 787 : set_subquery_size_estimates(root, rel);
1911 :
1912 : /*
1913 : * For each Path that subquery_planner produced, make a SubqueryScanPath
1914 : * in the outer query.
1915 : */
1916 1610 : foreach(lc, sub_final_rel->pathlist)
1917 : {
1918 823 : Path *subpath = (Path *) lfirst(lc);
1919 : List *pathkeys;
1920 :
1921 : /* Convert subpath's pathkeys to outer representation */
1922 823 : pathkeys = convert_subquery_pathkeys(root,
1923 : rel,
1924 : subpath->pathkeys,
1925 : make_tlist_from_pathtarget(subpath->pathtarget));
1926 :
1927 : /* Generate outer path using this subpath */
1928 823 : add_path(rel, (Path *)
1929 823 : create_subqueryscan_path(root, rel, subpath,
1930 : pathkeys, required_outer));
1931 : }
1932 : }
1933 :
1934 : /*
1935 : * set_function_pathlist
1936 : * Build the (single) access path for a function RTE
1937 : */
1938 : static void
1939 1327 : set_function_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
1940 : {
1941 : Relids required_outer;
1942 1327 : List *pathkeys = NIL;
1943 :
1944 : /*
1945 : * We don't support pushing join clauses into the quals of a function
1946 : * scan, but it could still have required parameterization due to LATERAL
1947 : * refs in the function expression.
1948 : */
1949 1327 : required_outer = rel->lateral_relids;
1950 :
1951 : /*
1952 : * The result is considered unordered unless ORDINALITY was used, in which
1953 : * case it is ordered by the ordinal column (the last one). See if we
1954 : * care, by checking for uses of that Var in equivalence classes.
1955 : */
1956 1327 : if (rte->funcordinality)
1957 : {
1958 63 : AttrNumber ordattno = rel->max_attr;
1959 63 : Var *var = NULL;
1960 : ListCell *lc;
1961 :
1962 : /*
1963 : * Is there a Var for it in rel's targetlist? If not, the query did
1964 : * not reference the ordinality column, or at least not in any way
1965 : * that would be interesting for sorting.
1966 : */
1967 251 : foreach(lc, rel->reltarget->exprs)
1968 : {
1969 250 : Var *node = (Var *) lfirst(lc);
1970 :
1971 : /* checking varno/varlevelsup is just paranoia */
1972 500 : if (IsA(node, Var) &&
1973 312 : node->varattno == ordattno &&
1974 124 : node->varno == rel->relid &&
1975 62 : node->varlevelsup == 0)
1976 : {
1977 62 : var = node;
1978 62 : break;
1979 : }
1980 : }
1981 :
1982 : /*
1983 : * Try to build pathkeys for this Var with int8 sorting. We tell
1984 : * build_expression_pathkey not to build any new equivalence class; if
1985 : * the Var isn't already mentioned in some EC, it means that nothing
1986 : * cares about the ordering.
1987 : */
1988 63 : if (var)
1989 62 : pathkeys = build_expression_pathkey(root,
1990 : (Expr *) var,
1991 : NULL, /* below outer joins */
1992 : Int8LessOperator,
1993 : rel->relids,
1994 : false);
1995 : }
1996 :
1997 : /* Generate appropriate path */
1998 1327 : add_path(rel, create_functionscan_path(root, rel,
1999 : pathkeys, required_outer));
2000 1327 : }
2001 :
2002 : /*
2003 : * set_values_pathlist
2004 : * Build the (single) access path for a VALUES RTE
2005 : */
2006 : static void
2007 463 : set_values_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2008 : {
2009 : Relids required_outer;
2010 :
2011 : /*
2012 : * We don't support pushing join clauses into the quals of a values scan,
2013 : * but it could still have required parameterization due to LATERAL refs
2014 : * in the values expressions.
2015 : */
2016 463 : required_outer = rel->lateral_relids;
2017 :
2018 : /* Generate appropriate path */
2019 463 : add_path(rel, create_valuesscan_path(root, rel, required_outer));
2020 463 : }
2021 :
2022 : /*
2023 : * set_tablefunc_pathlist
2024 : * Build the (single) access path for a table func RTE
2025 : */
2026 : static void
2027 22 : set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2028 : {
2029 : Relids required_outer;
2030 :
2031 : /*
2032 : * We don't support pushing join clauses into the quals of a tablefunc
2033 : * scan, but it could still have required parameterization due to LATERAL
2034 : * refs in the function expression.
2035 : */
2036 22 : required_outer = rel->lateral_relids;
2037 :
2038 : /* Generate appropriate path */
2039 22 : add_path(rel, create_tablefuncscan_path(root, rel,
2040 : required_outer));
2041 22 : }
2042 :
2043 : /*
2044 : * set_cte_pathlist
2045 : * Build the (single) access path for a non-self-reference CTE RTE
2046 : *
2047 : * There's no need for a separate set_cte_size phase, since we don't
2048 : * support join-qual-parameterized paths for CTEs.
2049 : */
2050 : static void
2051 162 : set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2052 : {
2053 : Plan *cteplan;
2054 : PlannerInfo *cteroot;
2055 : Index levelsup;
2056 : int ndx;
2057 : ListCell *lc;
2058 : int plan_id;
2059 : Relids required_outer;
2060 :
2061 : /*
2062 : * Find the referenced CTE, and locate the plan previously made for it.
2063 : */
2064 162 : levelsup = rte->ctelevelsup;
2065 162 : cteroot = root;
2066 384 : while (levelsup-- > 0)
2067 : {
2068 60 : cteroot = cteroot->parent_root;
2069 60 : if (!cteroot) /* shouldn't happen */
2070 0 : elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2071 : }
2072 :
2073 : /*
2074 : * Note: cte_plan_ids can be shorter than cteList, if we are still working
2075 : * on planning the CTEs (ie, this is a side-reference from another CTE).
2076 : * So we mustn't use forboth here.
2077 : */
2078 162 : ndx = 0;
2079 188 : foreach(lc, cteroot->parse->cteList)
2080 : {
2081 188 : CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
2082 :
2083 188 : if (strcmp(cte->ctename, rte->ctename) == 0)
2084 162 : break;
2085 26 : ndx++;
2086 : }
2087 162 : if (lc == NULL) /* shouldn't happen */
2088 0 : elog(ERROR, "could not find CTE \"%s\"", rte->ctename);
2089 162 : if (ndx >= list_length(cteroot->cte_plan_ids))
2090 0 : elog(ERROR, "could not find plan for CTE \"%s\"", rte->ctename);
2091 162 : plan_id = list_nth_int(cteroot->cte_plan_ids, ndx);
2092 162 : Assert(plan_id > 0);
2093 162 : cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2094 :
2095 : /* Mark rel with estimated output rows, width, etc */
2096 162 : set_cte_size_estimates(root, rel, cteplan->plan_rows);
2097 :
2098 : /*
2099 : * We don't support pushing join clauses into the quals of a CTE scan, but
2100 : * it could still have required parameterization due to LATERAL refs in
2101 : * its tlist.
2102 : */
2103 162 : required_outer = rel->lateral_relids;
2104 :
2105 : /* Generate appropriate path */
2106 162 : add_path(rel, create_ctescan_path(root, rel, required_outer));
2107 162 : }
2108 :
2109 : /*
2110 : * set_namedtuplestore_pathlist
2111 : * Build the (single) access path for a named tuplestore RTE
2112 : *
2113 : * There's no need for a separate set_namedtuplestore_size phase, since we
2114 : * don't support join-qual-parameterized paths for tuplestores.
2115 : */
2116 : static void
2117 43 : set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel,
2118 : RangeTblEntry *rte)
2119 : {
2120 : Relids required_outer;
2121 :
2122 : /* Mark rel with estimated output rows, width, etc */
2123 43 : set_namedtuplestore_size_estimates(root, rel);
2124 :
2125 : /*
2126 : * We don't support pushing join clauses into the quals of a tuplestore
2127 : * scan, but it could still have required parameterization due to LATERAL
2128 : * refs in its tlist.
2129 : */
2130 43 : required_outer = rel->lateral_relids;
2131 :
2132 : /* Generate appropriate path */
2133 43 : add_path(rel, create_namedtuplestorescan_path(root, rel, required_outer));
2134 :
2135 : /* Select cheapest path (pretty easy in this case...) */
2136 43 : set_cheapest(rel);
2137 43 : }
2138 :
2139 : /*
2140 : * set_worktable_pathlist
2141 : * Build the (single) access path for a self-reference CTE RTE
2142 : *
2143 : * There's no need for a separate set_worktable_size phase, since we don't
2144 : * support join-qual-parameterized paths for CTEs.
2145 : */
2146 : static void
2147 40 : set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2148 : {
2149 : Path *ctepath;
2150 : PlannerInfo *cteroot;
2151 : Index levelsup;
2152 : Relids required_outer;
2153 :
2154 : /*
2155 : * We need to find the non-recursive term's path, which is in the plan
2156 : * level that's processing the recursive UNION, which is one level *below*
2157 : * where the CTE comes from.
2158 : */
2159 40 : levelsup = rte->ctelevelsup;
2160 40 : if (levelsup == 0) /* shouldn't happen */
2161 0 : elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2162 40 : levelsup--;
2163 40 : cteroot = root;
2164 126 : while (levelsup-- > 0)
2165 : {
2166 46 : cteroot = cteroot->parent_root;
2167 46 : if (!cteroot) /* shouldn't happen */
2168 0 : elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2169 : }
2170 40 : ctepath = cteroot->non_recursive_path;
2171 40 : if (!ctepath) /* shouldn't happen */
2172 0 : elog(ERROR, "could not find path for CTE \"%s\"", rte->ctename);
2173 :
2174 : /* Mark rel with estimated output rows, width, etc */
2175 40 : set_cte_size_estimates(root, rel, ctepath->rows);
2176 :
2177 : /*
2178 : * We don't support pushing join clauses into the quals of a worktable
2179 : * scan, but it could still have required parameterization due to LATERAL
2180 : * refs in its tlist. (I'm not sure this is actually possible given the
2181 : * restrictions on recursive references, but it's easy enough to support.)
2182 : */
2183 40 : required_outer = rel->lateral_relids;
2184 :
2185 : /* Generate appropriate path */
2186 40 : add_path(rel, create_worktablescan_path(root, rel, required_outer));
2187 40 : }
2188 :
2189 : /*
2190 : * generate_gather_paths
2191 : * Generate parallel access paths for a relation by pushing a Gather or
2192 : * Gather Merge on top of a partial path.
2193 : *
2194 : * This must not be called until after we're done creating all partial paths
2195 : * for the specified relation. (Otherwise, add_partial_path might delete a
2196 : * path that some GatherPath or GatherMergePath has a reference to.)
2197 : */
2198 : void
2199 22891 : generate_gather_paths(PlannerInfo *root, RelOptInfo *rel)
2200 : {
2201 : Path *cheapest_partial_path;
2202 : Path *simple_gather_path;
2203 : ListCell *lc;
2204 :
2205 : /* If there are no partial paths, there's nothing to do here. */
2206 22891 : if (rel->partial_pathlist == NIL)
2207 45524 : return;
2208 :
2209 : /*
2210 : * The output of Gather is always unsorted, so there's only one partial
2211 : * path of interest: the cheapest one. That will be the one at the front
2212 : * of partial_pathlist because of the way add_partial_path works.
2213 : */
2214 258 : cheapest_partial_path = linitial(rel->partial_pathlist);
2215 258 : simple_gather_path = (Path *)
2216 258 : create_gather_path(root, rel, cheapest_partial_path, rel->reltarget,
2217 : NULL, NULL);
2218 258 : add_path(rel, simple_gather_path);
2219 :
2220 : /*
2221 : * For each useful ordering, we can consider an order-preserving Gather
2222 : * Merge.
2223 : */
2224 517 : foreach(lc, rel->partial_pathlist)
2225 : {
2226 259 : Path *subpath = (Path *) lfirst(lc);
2227 : GatherMergePath *path;
2228 :
2229 259 : if (subpath->pathkeys == NIL)
2230 256 : continue;
2231 :
2232 3 : path = create_gather_merge_path(root, rel, subpath, rel->reltarget,
2233 : subpath->pathkeys, NULL, NULL);
2234 3 : add_path(rel, &path->path);
2235 : }
2236 : }
2237 :
2238 : /*
2239 : * make_rel_from_joinlist
2240 : * Build access paths using a "joinlist" to guide the join path search.
2241 : *
2242 : * See comments for deconstruct_jointree() for definition of the joinlist
2243 : * data structure.
2244 : */
2245 : static RelOptInfo *
2246 13938 : make_rel_from_joinlist(PlannerInfo *root, List *joinlist)
2247 : {
2248 : int levels_needed;
2249 : List *initial_rels;
2250 : ListCell *jl;
2251 :
2252 : /*
2253 : * Count the number of child joinlist nodes. This is the depth of the
2254 : * dynamic-programming algorithm we must employ to consider all ways of
2255 : * joining the child nodes.
2256 : */
2257 13938 : levels_needed = list_length(joinlist);
2258 :
2259 13938 : if (levels_needed <= 0)
2260 0 : return NULL; /* nothing to do? */
2261 :
2262 : /*
2263 : * Construct a list of rels corresponding to the child joinlist nodes.
2264 : * This may contain both base rels and rels constructed according to
2265 : * sub-joinlists.
2266 : */
2267 13938 : initial_rels = NIL;
2268 31598 : foreach(jl, joinlist)
2269 : {
2270 17660 : Node *jlnode = (Node *) lfirst(jl);
2271 : RelOptInfo *thisrel;
2272 :
2273 17660 : if (IsA(jlnode, RangeTblRef))
2274 : {
2275 17540 : int varno = ((RangeTblRef *) jlnode)->rtindex;
2276 :
2277 17540 : thisrel = find_base_rel(root, varno);
2278 : }
2279 120 : else if (IsA(jlnode, List))
2280 : {
2281 : /* Recurse to handle subproblem */
2282 120 : thisrel = make_rel_from_joinlist(root, (List *) jlnode);
2283 : }
2284 : else
2285 : {
2286 0 : elog(ERROR, "unrecognized joinlist node type: %d",
2287 : (int) nodeTag(jlnode));
2288 : thisrel = NULL; /* keep compiler quiet */
2289 : }
2290 :
2291 17660 : initial_rels = lappend(initial_rels, thisrel);
2292 : }
2293 :
2294 13938 : if (levels_needed == 1)
2295 : {
2296 : /*
2297 : * Single joinlist node, so we're done.
2298 : */
2299 10847 : return (RelOptInfo *) linitial(initial_rels);
2300 : }
2301 : else
2302 : {
2303 : /*
2304 : * Consider the different orders in which we could join the rels,
2305 : * using a plugin, GEQO, or the regular join search code.
2306 : *
2307 : * We put the initial_rels list into a PlannerInfo field because
2308 : * has_legal_joinclause() needs to look at it (ugly :-().
2309 : */
2310 3091 : root->initial_rels = initial_rels;
2311 :
2312 3091 : if (join_search_hook)
2313 0 : return (*join_search_hook) (root, levels_needed, initial_rels);
2314 3091 : else if (enable_geqo && levels_needed >= geqo_threshold)
2315 1 : return geqo(root, levels_needed, initial_rels);
2316 : else
2317 3090 : return standard_join_search(root, levels_needed, initial_rels);
2318 : }
2319 : }
2320 :
2321 : /*
2322 : * standard_join_search
2323 : * Find possible joinpaths for a query by successively finding ways
2324 : * to join component relations into join relations.
2325 : *
2326 : * 'levels_needed' is the number of iterations needed, ie, the number of
2327 : * independent jointree items in the query. This is > 1.
2328 : *
2329 : * 'initial_rels' is a list of RelOptInfo nodes for each independent
2330 : * jointree item. These are the components to be joined together.
2331 : * Note that levels_needed == list_length(initial_rels).
2332 : *
2333 : * Returns the final level of join relations, i.e., the relation that is
2334 : * the result of joining all the original relations together.
2335 : * At least one implementation path must be provided for this relation and
2336 : * all required sub-relations.
2337 : *
2338 : * To support loadable plugins that modify planner behavior by changing the
2339 : * join searching algorithm, we provide a hook variable that lets a plugin
2340 : * replace or supplement this function. Any such hook must return the same
2341 : * final join relation as the standard code would, but it might have a
2342 : * different set of implementation paths attached, and only the sub-joinrels
2343 : * needed for these paths need have been instantiated.
2344 : *
2345 : * Note to plugin authors: the functions invoked during standard_join_search()
2346 : * modify root->join_rel_list and root->join_rel_hash. If you want to do more
2347 : * than one join-order search, you'll probably need to save and restore the
2348 : * original states of those data structures. See geqo_eval() for an example.
2349 : */
2350 : RelOptInfo *
2351 3090 : standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
2352 : {
2353 : int lev;
2354 : RelOptInfo *rel;
2355 :
2356 : /*
2357 : * This function cannot be invoked recursively within any one planning
2358 : * problem, so join_rel_level[] can't be in use already.
2359 : */
2360 3090 : Assert(root->join_rel_level == NULL);
2361 :
2362 : /*
2363 : * We employ a simple "dynamic programming" algorithm: we first find all
2364 : * ways to build joins of two jointree items, then all ways to build joins
2365 : * of three items (from two-item joins and single items), then four-item
2366 : * joins, and so on until we have considered all ways to join all the
2367 : * items into one rel.
2368 : *
2369 : * root->join_rel_level[j] is a list of all the j-item rels. Initially we
2370 : * set root->join_rel_level[1] to represent all the single-jointree-item
2371 : * relations.
2372 : */
2373 3090 : root->join_rel_level = (List **) palloc0((levels_needed + 1) * sizeof(List *));
2374 :
2375 3090 : root->join_rel_level[1] = initial_rels;
2376 :
2377 6808 : for (lev = 2; lev <= levels_needed; lev++)
2378 : {
2379 : ListCell *lc;
2380 :
2381 : /*
2382 : * Determine all possible pairs of relations to be joined at this
2383 : * level, and build paths for making each one from every available
2384 : * pair of lower-level relations.
2385 : */
2386 3718 : join_search_one_level(root, lev);
2387 :
2388 : /*
2389 : * Run generate_gather_paths() for each just-processed joinrel. We
2390 : * could not do this earlier because both regular and partial paths
2391 : * can get added to a particular joinrel at multiple times within
2392 : * join_search_one_level. After that, we're done creating paths for
2393 : * the joinrel, so run set_cheapest().
2394 : */
2395 8553 : foreach(lc, root->join_rel_level[lev])
2396 : {
2397 4835 : rel = (RelOptInfo *) lfirst(lc);
2398 :
2399 : /* Create GatherPaths for any useful partial paths for rel */
2400 4835 : generate_gather_paths(root, rel);
2401 :
2402 : /* Find and save the cheapest paths for this rel */
2403 4835 : set_cheapest(rel);
2404 :
2405 : #ifdef OPTIMIZER_DEBUG
2406 : debug_print_rel(root, rel);
2407 : #endif
2408 : }
2409 : }
2410 :
2411 : /*
2412 : * We should have a single rel at the final level.
2413 : */
2414 3090 : if (root->join_rel_level[levels_needed] == NIL)
2415 0 : elog(ERROR, "failed to build any %d-way joins", levels_needed);
2416 3090 : Assert(list_length(root->join_rel_level[levels_needed]) == 1);
2417 :
2418 3090 : rel = (RelOptInfo *) linitial(root->join_rel_level[levels_needed]);
2419 :
2420 3090 : root->join_rel_level = NULL;
2421 :
2422 3090 : return rel;
2423 : }
2424 :
2425 : /*****************************************************************************
2426 : * PUSHING QUALS DOWN INTO SUBQUERIES
2427 : *****************************************************************************/
2428 :
2429 : /*
2430 : * subquery_is_pushdown_safe - is a subquery safe for pushing down quals?
2431 : *
2432 : * subquery is the particular component query being checked. topquery
2433 : * is the top component of a set-operations tree (the same Query if no
2434 : * set-op is involved).
2435 : *
2436 : * Conditions checked here:
2437 : *
2438 : * 1. If the subquery has a LIMIT clause, we must not push down any quals,
2439 : * since that could change the set of rows returned.
2440 : *
2441 : * 2. If the subquery contains EXCEPT or EXCEPT ALL set ops we cannot push
2442 : * quals into it, because that could change the results.
2443 : *
2444 : * 3. If the subquery uses DISTINCT, we cannot push volatile quals into it.
2445 : * This is because upper-level quals should semantically be evaluated only
2446 : * once per distinct row, not once per original row, and if the qual is
2447 : * volatile then extra evaluations could change the results. (This issue
2448 : * does not apply to other forms of aggregation such as GROUP BY, because
2449 : * when those are present we push into HAVING not WHERE, so that the quals
2450 : * are still applied after aggregation.)
2451 : *
2452 : * 4. If the subquery contains window functions, we cannot push volatile quals
2453 : * into it. The issue here is a bit different from DISTINCT: a volatile qual
2454 : * might succeed for some rows of a window partition and fail for others,
2455 : * thereby changing the partition contents and thus the window functions'
2456 : * results for rows that remain.
2457 : *
2458 : * 5. If the subquery contains any set-returning functions in its targetlist,
2459 : * we cannot push volatile quals into it. That would push them below the SRFs
2460 : * and thereby change the number of times they are evaluated. Also, a
2461 : * volatile qual could succeed for some SRF output rows and fail for others,
2462 : * a behavior that cannot occur if it's evaluated before SRF expansion.
2463 : *
2464 : * In addition, we make several checks on the subquery's output columns to see
2465 : * if it is safe to reference them in pushed-down quals. If output column k
2466 : * is found to be unsafe to reference, we set safetyInfo->unsafeColumns[k]
2467 : * to TRUE, but we don't reject the subquery overall since column k might not
2468 : * be referenced by some/all quals. The unsafeColumns[] array will be
2469 : * consulted later by qual_is_pushdown_safe(). It's better to do it this way
2470 : * than to make the checks directly in qual_is_pushdown_safe(), because when
2471 : * the subquery involves set operations we have to check the output
2472 : * expressions in each arm of the set op.
2473 : *
2474 : * Note: pushing quals into a DISTINCT subquery is theoretically dubious:
2475 : * we're effectively assuming that the quals cannot distinguish values that
2476 : * the DISTINCT's equality operator sees as equal, yet there are many
2477 : * counterexamples to that assumption. However use of such a qual with a
2478 : * DISTINCT subquery would be unsafe anyway, since there's no guarantee which
2479 : * "equal" value will be chosen as the output value by the DISTINCT operation.
2480 : * So we don't worry too much about that. Another objection is that if the
2481 : * qual is expensive to evaluate, running it for each original row might cost
2482 : * more than we save by eliminating rows before the DISTINCT step. But it
2483 : * would be very hard to estimate that at this stage, and in practice pushdown
2484 : * seldom seems to make things worse, so we ignore that problem too.
2485 : *
2486 : * Note: likewise, pushing quals into a subquery with window functions is a
2487 : * bit dubious: the quals might remove some rows of a window partition while
2488 : * leaving others, causing changes in the window functions' results for the
2489 : * surviving rows. We insist that such a qual reference only partitioning
2490 : * columns, but again that only protects us if the qual does not distinguish
2491 : * values that the partitioning equality operator sees as equal. The risks
2492 : * here are perhaps larger than for DISTINCT, since no de-duplication of rows
2493 : * occurs and thus there is no theoretical problem with such a qual. But
2494 : * we'll do this anyway because the potential performance benefits are very
2495 : * large, and we've seen no field complaints about the longstanding comparable
2496 : * behavior with DISTINCT.
2497 : */
2498 : static bool
2499 154 : subquery_is_pushdown_safe(Query *subquery, Query *topquery,
2500 : pushdown_safety_info *safetyInfo)
2501 : {
2502 : SetOperationStmt *topop;
2503 :
2504 : /* Check point 1 */
2505 154 : if (subquery->limitOffset != NULL || subquery->limitCount != NULL)
2506 15 : return false;
2507 :
2508 : /* Check points 3, 4, and 5 */
2509 277 : if (subquery->distinctClause ||
2510 271 : subquery->hasWindowFuncs ||
2511 133 : subquery->hasTargetSRFs)
2512 39 : safetyInfo->unsafeVolatile = true;
2513 :
2514 : /*
2515 : * If we're at a leaf query, check for unsafe expressions in its target
2516 : * list, and mark any unsafe ones in unsafeColumns[]. (Non-leaf nodes in
2517 : * setop trees have only simple Vars in their tlists, so no need to check
2518 : * them.)
2519 : */
2520 139 : if (subquery->setOperations == NULL)
2521 132 : check_output_expressions(subquery, safetyInfo);
2522 :
2523 : /* Are we at top level, or looking at a setop component? */
2524 139 : if (subquery == topquery)
2525 : {
2526 : /* Top level, so check any component queries */
2527 125 : if (subquery->setOperations != NULL)
2528 7 : if (!recurse_pushdown_safe(subquery->setOperations, topquery,
2529 : safetyInfo))
2530 0 : return false;
2531 : }
2532 : else
2533 : {
2534 : /* Setop component must not have more components (too weird) */
2535 14 : if (subquery->setOperations != NULL)
2536 0 : return false;
2537 : /* Check whether setop component output types match top level */
2538 14 : topop = castNode(SetOperationStmt, topquery->setOperations);
2539 14 : Assert(topop);
2540 14 : compare_tlist_datatypes(subquery->targetList,
2541 : topop->colTypes,
2542 : safetyInfo);
2543 : }
2544 139 : return true;
2545 : }
2546 :
2547 : /*
2548 : * Helper routine to recurse through setOperations tree
2549 : */
2550 : static bool
2551 21 : recurse_pushdown_safe(Node *setOp, Query *topquery,
2552 : pushdown_safety_info *safetyInfo)
2553 : {
2554 21 : if (IsA(setOp, RangeTblRef))
2555 : {
2556 14 : RangeTblRef *rtr = (RangeTblRef *) setOp;
2557 14 : RangeTblEntry *rte = rt_fetch(rtr->rtindex, topquery->rtable);
2558 14 : Query *subquery = rte->subquery;
2559 :
2560 14 : Assert(subquery != NULL);
2561 14 : return subquery_is_pushdown_safe(subquery, topquery, safetyInfo);
2562 : }
2563 7 : else if (IsA(setOp, SetOperationStmt))
2564 : {
2565 7 : SetOperationStmt *op = (SetOperationStmt *) setOp;
2566 :
2567 : /* EXCEPT is no good (point 2 for subquery_is_pushdown_safe) */
2568 7 : if (op->op == SETOP_EXCEPT)
2569 0 : return false;
2570 : /* Else recurse */
2571 7 : if (!recurse_pushdown_safe(op->larg, topquery, safetyInfo))
2572 0 : return false;
2573 7 : if (!recurse_pushdown_safe(op->rarg, topquery, safetyInfo))
2574 0 : return false;
2575 : }
2576 : else
2577 : {
2578 0 : elog(ERROR, "unrecognized node type: %d",
2579 : (int) nodeTag(setOp));
2580 : }
2581 7 : return true;
2582 : }
2583 :
2584 : /*
2585 : * check_output_expressions - check subquery's output expressions for safety
2586 : *
2587 : * There are several cases in which it's unsafe to push down an upper-level
2588 : * qual if it references a particular output column of a subquery. We check
2589 : * each output column of the subquery and set unsafeColumns[k] to TRUE if
2590 : * that column is unsafe for a pushed-down qual to reference. The conditions
2591 : * checked here are:
2592 : *
2593 : * 1. We must not push down any quals that refer to subselect outputs that
2594 : * return sets, else we'd introduce functions-returning-sets into the
2595 : * subquery's WHERE/HAVING quals.
2596 : *
2597 : * 2. We must not push down any quals that refer to subselect outputs that
2598 : * contain volatile functions, for fear of introducing strange results due
2599 : * to multiple evaluation of a volatile function.
2600 : *
2601 : * 3. If the subquery uses DISTINCT ON, we must not push down any quals that
2602 : * refer to non-DISTINCT output columns, because that could change the set
2603 : * of rows returned. (This condition is vacuous for DISTINCT, because then
2604 : * there are no non-DISTINCT output columns, so we needn't check. Note that
2605 : * subquery_is_pushdown_safe already reported that we can't use volatile
2606 : * quals if there's DISTINCT or DISTINCT ON.)
2607 : *
2608 : * 4. If the subquery has any window functions, we must not push down quals
2609 : * that reference any output columns that are not listed in all the subquery's
2610 : * window PARTITION BY clauses. We can push down quals that use only
2611 : * partitioning columns because they should succeed or fail identically for
2612 : * every row of any one window partition, and totally excluding some
2613 : * partitions will not change a window function's results for remaining
2614 : * partitions. (Again, this also requires nonvolatile quals, but
2615 : * subquery_is_pushdown_safe handles that.)
2616 : */
2617 : static void
2618 132 : check_output_expressions(Query *subquery, pushdown_safety_info *safetyInfo)
2619 : {
2620 : ListCell *lc;
2621 :
2622 893 : foreach(lc, subquery->targetList)
2623 : {
2624 761 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
2625 :
2626 761 : if (tle->resjunk)
2627 9 : continue; /* ignore resjunk columns */
2628 :
2629 : /* We need not check further if output col is already known unsafe */
2630 752 : if (safetyInfo->unsafeColumns[tle->resno])
2631 4 : continue;
2632 :
2633 : /* Functions returning sets are unsafe (point 1) */
2634 909 : if (subquery->hasTargetSRFs &&
2635 161 : expression_returns_set((Node *) tle->expr))
2636 : {
2637 82 : safetyInfo->unsafeColumns[tle->resno] = true;
2638 82 : continue;
2639 : }
2640 :
2641 : /* Volatile functions are unsafe (point 2) */
2642 666 : if (contain_volatile_functions((Node *) tle->expr))
2643 : {
2644 10 : safetyInfo->unsafeColumns[tle->resno] = true;
2645 10 : continue;
2646 : }
2647 :
2648 : /* If subquery uses DISTINCT ON, check point 3 */
2649 656 : if (subquery->hasDistinctOn &&
2650 0 : !targetIsInSortList(tle, InvalidOid, subquery->distinctClause))
2651 : {
2652 : /* non-DISTINCT column, so mark it unsafe */
2653 0 : safetyInfo->unsafeColumns[tle->resno] = true;
2654 0 : continue;
2655 : }
2656 :
2657 : /* If subquery uses window functions, check point 4 */
2658 677 : if (subquery->hasWindowFuncs &&
2659 21 : !targetIsInAllPartitionLists(tle, subquery))
2660 : {
2661 : /* not present in all PARTITION BY clauses, so mark it unsafe */
2662 20 : safetyInfo->unsafeColumns[tle->resno] = true;
2663 20 : continue;
2664 : }
2665 : }
2666 132 : }
2667 :
2668 : /*
2669 : * For subqueries using UNION/UNION ALL/INTERSECT/INTERSECT ALL, we can
2670 : * push quals into each component query, but the quals can only reference
2671 : * subquery columns that suffer no type coercions in the set operation.
2672 : * Otherwise there are possible semantic gotchas. So, we check the
2673 : * component queries to see if any of them have output types different from
2674 : * the top-level setop outputs. unsafeColumns[k] is set true if column k
2675 : * has different type in any component.
2676 : *
2677 : * We don't have to care about typmods here: the only allowed difference
2678 : * between set-op input and output typmods is input is a specific typmod
2679 : * and output is -1, and that does not require a coercion.
2680 : *
2681 : * tlist is a subquery tlist.
2682 : * colTypes is an OID list of the top-level setop's output column types.
2683 : * safetyInfo->unsafeColumns[] is the result array.
2684 : */
2685 : static void
2686 14 : compare_tlist_datatypes(List *tlist, List *colTypes,
2687 : pushdown_safety_info *safetyInfo)
2688 : {
2689 : ListCell *l;
2690 14 : ListCell *colType = list_head(colTypes);
2691 :
2692 40 : foreach(l, tlist)
2693 : {
2694 26 : TargetEntry *tle = (TargetEntry *) lfirst(l);
2695 :
2696 26 : if (tle->resjunk)
2697 0 : continue; /* ignore resjunk columns */
2698 26 : if (colType == NULL)
2699 0 : elog(ERROR, "wrong number of tlist entries");
2700 26 : if (exprType((Node *) tle->expr) != lfirst_oid(colType))
2701 0 : safetyInfo->unsafeColumns[tle->resno] = true;
2702 26 : colType = lnext(colType);
2703 : }
2704 14 : if (colType != NULL)
2705 0 : elog(ERROR, "wrong number of tlist entries");
2706 14 : }
2707 :
2708 : /*
2709 : * targetIsInAllPartitionLists
2710 : * True if the TargetEntry is listed in the PARTITION BY clause
2711 : * of every window defined in the query.
2712 : *
2713 : * It would be safe to ignore windows not actually used by any window
2714 : * function, but it's not easy to get that info at this stage; and it's
2715 : * unlikely to be useful to spend any extra cycles getting it, since
2716 : * unreferenced window definitions are probably infrequent in practice.
2717 : */
2718 : static bool
2719 21 : targetIsInAllPartitionLists(TargetEntry *tle, Query *query)
2720 : {
2721 : ListCell *lc;
2722 :
2723 23 : foreach(lc, query->windowClause)
2724 : {
2725 22 : WindowClause *wc = (WindowClause *) lfirst(lc);
2726 :
2727 22 : if (!targetIsInSortList(tle, InvalidOid, wc->partitionClause))
2728 20 : return false;
2729 : }
2730 1 : return true;
2731 : }
2732 :
2733 : /*
2734 : * qual_is_pushdown_safe - is a particular qual safe to push down?
2735 : *
2736 : * qual is a restriction clause applying to the given subquery (whose RTE
2737 : * has index rti in the parent query).
2738 : *
2739 : * Conditions checked here:
2740 : *
2741 : * 1. The qual must not contain any subselects (mainly because I'm not sure
2742 : * it will work correctly: sublinks will already have been transformed into
2743 : * subplans in the qual, but not in the subquery).
2744 : *
2745 : * 2. If unsafeVolatile is set, the qual must not contain any volatile
2746 : * functions.
2747 : *
2748 : * 3. If unsafeLeaky is set, the qual must not contain any leaky functions
2749 : * that are passed Var nodes, and therefore might reveal values from the
2750 : * subquery as side effects.
2751 : *
2752 : * 4. The qual must not refer to the whole-row output of the subquery
2753 : * (since there is no easy way to name that within the subquery itself).
2754 : *
2755 : * 5. The qual must not refer to any subquery output columns that were
2756 : * found to be unsafe to reference by subquery_is_pushdown_safe().
2757 : */
2758 : static bool
2759 205 : qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual,
2760 : pushdown_safety_info *safetyInfo)
2761 : {
2762 205 : bool safe = true;
2763 : List *vars;
2764 : ListCell *vl;
2765 :
2766 : /* Refuse subselects (point 1) */
2767 205 : if (contain_subplans(qual))
2768 10 : return false;
2769 :
2770 : /* Refuse volatile quals if we found they'd be unsafe (point 2) */
2771 259 : if (safetyInfo->unsafeVolatile &&
2772 64 : contain_volatile_functions(qual))
2773 3 : return false;
2774 :
2775 : /* Refuse leaky quals if told to (point 3) */
2776 232 : if (safetyInfo->unsafeLeaky &&
2777 40 : contain_leaked_vars(qual))
2778 20 : return false;
2779 :
2780 : /*
2781 : * It would be unsafe to push down window function calls, but at least for
2782 : * the moment we could never see any in a qual anyhow. (The same applies
2783 : * to aggregates, which we check for in pull_var_clause below.)
2784 : */
2785 172 : Assert(!contain_window_function(qual));
2786 :
2787 : /*
2788 : * Examine all Vars used in clause; since it's a restriction clause, all
2789 : * such Vars must refer to subselect output columns.
2790 : */
2791 172 : vars = pull_var_clause(qual, PVC_INCLUDE_PLACEHOLDERS);
2792 310 : foreach(vl, vars)
2793 : {
2794 174 : Var *var = (Var *) lfirst(vl);
2795 :
2796 : /*
2797 : * XXX Punt if we find any PlaceHolderVars in the restriction clause.
2798 : * It's not clear whether a PHV could safely be pushed down, and even
2799 : * less clear whether such a situation could arise in any cases of
2800 : * practical interest anyway. So for the moment, just refuse to push
2801 : * down.
2802 : */
2803 174 : if (!IsA(var, Var))
2804 : {
2805 0 : safe = false;
2806 0 : break;
2807 : }
2808 :
2809 174 : Assert(var->varno == rti);
2810 174 : Assert(var->varattno >= 0);
2811 :
2812 : /* Check point 4 */
2813 174 : if (var->varattno == 0)
2814 : {
2815 0 : safe = false;
2816 0 : break;
2817 : }
2818 :
2819 : /* Check point 5 */
2820 174 : if (safetyInfo->unsafeColumns[var->varattno])
2821 : {
2822 36 : safe = false;
2823 36 : break;
2824 : }
2825 : }
2826 :
2827 172 : list_free(vars);
2828 :
2829 172 : return safe;
2830 : }
2831 :
2832 : /*
2833 : * subquery_push_qual - push down a qual that we have determined is safe
2834 : */
2835 : static void
2836 142 : subquery_push_qual(Query *subquery, RangeTblEntry *rte, Index rti, Node *qual)
2837 : {
2838 142 : if (subquery->setOperations != NULL)
2839 : {
2840 : /* Recurse to push it separately to each component query */
2841 3 : recurse_push_qual(subquery->setOperations, subquery,
2842 : rte, rti, qual);
2843 : }
2844 : else
2845 : {
2846 : /*
2847 : * We need to replace Vars in the qual (which must refer to outputs of
2848 : * the subquery) with copies of the subquery's targetlist expressions.
2849 : * Note that at this point, any uplevel Vars in the qual should have
2850 : * been replaced with Params, so they need no work.
2851 : *
2852 : * This step also ensures that when we are pushing into a setop tree,
2853 : * each component query gets its own copy of the qual.
2854 : */
2855 139 : qual = ReplaceVarsFromTargetList(qual, rti, 0, rte,
2856 : subquery->targetList,
2857 : REPLACEVARS_REPORT_ERROR, 0,
2858 : &subquery->hasSubLinks);
2859 :
2860 : /*
2861 : * Now attach the qual to the proper place: normally WHERE, but if the
2862 : * subquery uses grouping or aggregation, put it in HAVING (since the
2863 : * qual really refers to the group-result rows).
2864 : */
2865 139 : if (subquery->hasAggs || subquery->groupClause || subquery->groupingSets || subquery->havingQual)
2866 25 : subquery->havingQual = make_and_qual(subquery->havingQual, qual);
2867 : else
2868 228 : subquery->jointree->quals =
2869 114 : make_and_qual(subquery->jointree->quals, qual);
2870 :
2871 : /*
2872 : * We need not change the subquery's hasAggs or hasSubLinks flags,
2873 : * since we can't be pushing down any aggregates that weren't there
2874 : * before, and we don't push down subselects at all.
2875 : */
2876 : }
2877 142 : }
2878 :
2879 : /*
2880 : * Helper routine to recurse through setOperations tree
2881 : */
2882 : static void
2883 9 : recurse_push_qual(Node *setOp, Query *topquery,
2884 : RangeTblEntry *rte, Index rti, Node *qual)
2885 : {
2886 9 : if (IsA(setOp, RangeTblRef))
2887 : {
2888 6 : RangeTblRef *rtr = (RangeTblRef *) setOp;
2889 6 : RangeTblEntry *subrte = rt_fetch(rtr->rtindex, topquery->rtable);
2890 6 : Query *subquery = subrte->subquery;
2891 :
2892 6 : Assert(subquery != NULL);
2893 6 : subquery_push_qual(subquery, rte, rti, qual);
2894 : }
2895 3 : else if (IsA(setOp, SetOperationStmt))
2896 : {
2897 3 : SetOperationStmt *op = (SetOperationStmt *) setOp;
2898 :
2899 3 : recurse_push_qual(op->larg, topquery, rte, rti, qual);
2900 3 : recurse_push_qual(op->rarg, topquery, rte, rti, qual);
2901 : }
2902 : else
2903 : {
2904 0 : elog(ERROR, "unrecognized node type: %d",
2905 : (int) nodeTag(setOp));
2906 : }
2907 9 : }
2908 :
2909 : /*****************************************************************************
2910 : * SIMPLIFYING SUBQUERY TARGETLISTS
2911 : *****************************************************************************/
2912 :
2913 : /*
2914 : * remove_unused_subquery_outputs
2915 : * Remove subquery targetlist items we don't need
2916 : *
2917 : * It's possible, even likely, that the upper query does not read all the
2918 : * output columns of the subquery. We can remove any such outputs that are
2919 : * not needed by the subquery itself (e.g., as sort/group columns) and do not
2920 : * affect semantics otherwise (e.g., volatile functions can't be removed).
2921 : * This is useful not only because we might be able to remove expensive-to-
2922 : * compute expressions, but because deletion of output columns might allow
2923 : * optimizations such as join removal to occur within the subquery.
2924 : *
2925 : * To avoid affecting column numbering in the targetlist, we don't physically
2926 : * remove unused tlist entries, but rather replace their expressions with NULL
2927 : * constants. This is implemented by modifying subquery->targetList.
2928 : */
2929 : static void
2930 790 : remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel)
2931 : {
2932 790 : Bitmapset *attrs_used = NULL;
2933 : ListCell *lc;
2934 :
2935 : /*
2936 : * Do nothing if subquery has UNION/INTERSECT/EXCEPT: in principle we
2937 : * could update all the child SELECTs' tlists, but it seems not worth the
2938 : * trouble presently.
2939 : */
2940 790 : if (subquery->setOperations)
2941 83 : return;
2942 :
2943 : /*
2944 : * If subquery has regular DISTINCT (not DISTINCT ON), we're wasting our
2945 : * time: all its output columns must be used in the distinctClause.
2946 : */
2947 771 : if (subquery->distinctClause && !subquery->hasDistinctOn)
2948 15 : return;
2949 :
2950 : /*
2951 : * Collect a bitmap of all the output column numbers used by the upper
2952 : * query.
2953 : *
2954 : * Add all the attributes needed for joins or final output. Note: we must
2955 : * look at rel's targetlist, not the attr_needed data, because attr_needed
2956 : * isn't computed for inheritance child rels, cf set_append_rel_size().
2957 : * (XXX might be worth changing that sometime.)
2958 : */
2959 756 : pull_varattnos((Node *) rel->reltarget->exprs, rel->relid, &attrs_used);
2960 :
2961 : /* Add all the attributes used by un-pushed-down restriction clauses. */
2962 842 : foreach(lc, rel->baserestrictinfo)
2963 : {
2964 86 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2965 :
2966 86 : pull_varattnos((Node *) rinfo->clause, rel->relid, &attrs_used);
2967 : }
2968 :
2969 : /*
2970 : * If there's a whole-row reference to the subquery, we can't remove
2971 : * anything.
2972 : */
2973 756 : if (bms_is_member(0 - FirstLowInvalidHeapAttributeNumber, attrs_used))
2974 30 : return;
2975 :
2976 : /*
2977 : * Run through the tlist and zap entries we don't need. It's okay to
2978 : * modify the tlist items in-place because set_subquery_pathlist made a
2979 : * copy of the subquery.
2980 : */
2981 2429 : foreach(lc, subquery->targetList)
2982 : {
2983 1703 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
2984 1703 : Node *texpr = (Node *) tle->expr;
2985 :
2986 : /*
2987 : * If it has a sortgroupref number, it's used in some sort/group
2988 : * clause so we'd better not remove it. Also, don't remove any
2989 : * resjunk columns, since their reason for being has nothing to do
2990 : * with anybody reading the subquery's output. (It's likely that
2991 : * resjunk columns in a sub-SELECT would always have ressortgroupref
2992 : * set, but even if they don't, it seems imprudent to remove them.)
2993 : */
2994 1703 : if (tle->ressortgroupref || tle->resjunk)
2995 132 : continue;
2996 :
2997 : /*
2998 : * If it's used by the upper query, we can't remove it.
2999 : */
3000 1571 : if (bms_is_member(tle->resno - FirstLowInvalidHeapAttributeNumber,
3001 : attrs_used))
3002 1293 : continue;
3003 :
3004 : /*
3005 : * If it contains a set-returning function, we can't remove it since
3006 : * that could change the number of rows returned by the subquery.
3007 : */
3008 298 : if (subquery->hasTargetSRFs &&
3009 20 : expression_returns_set(texpr))
3010 2 : continue;
3011 :
3012 : /*
3013 : * If it contains volatile functions, we daren't remove it for fear
3014 : * that the user is expecting their side-effects to happen.
3015 : */
3016 276 : if (contain_volatile_functions(texpr))
3017 2 : continue;
3018 :
3019 : /*
3020 : * OK, we don't need it. Replace the expression with a NULL constant.
3021 : * Preserve the exposed type of the expression, in case something
3022 : * looks at the rowtype of the subquery's result.
3023 : */
3024 274 : tle->expr = (Expr *) makeNullConst(exprType(texpr),
3025 : exprTypmod(texpr),
3026 : exprCollation(texpr));
3027 : }
3028 : }
3029 :
3030 : /*
3031 : * create_partial_bitmap_paths
3032 : * Build partial bitmap heap path for the relation
3033 : */
3034 : void
3035 4852 : create_partial_bitmap_paths(PlannerInfo *root, RelOptInfo *rel,
3036 : Path *bitmapqual)
3037 : {
3038 : int parallel_workers;
3039 : double pages_fetched;
3040 :
3041 : /* Compute heap pages for bitmap heap scan */
3042 4852 : pages_fetched = compute_bitmap_pages(root, rel, bitmapqual, 1.0,
3043 : NULL, NULL);
3044 :
3045 4852 : parallel_workers = compute_parallel_worker(rel, pages_fetched, -1);
3046 :
3047 4852 : if (parallel_workers <= 0)
3048 9561 : return;
3049 :
3050 143 : add_partial_path(rel, (Path *) create_bitmap_heap_path(root, rel,
3051 : bitmapqual, rel->lateral_relids, 1.0, parallel_workers));
3052 : }
3053 :
3054 : /*
3055 : * Compute the number of parallel workers that should be used to scan a
3056 : * relation. We compute the parallel workers based on the size of the heap to
3057 : * be scanned and the size of the index to be scanned, then choose a minimum
3058 : * of those.
3059 : *
3060 : * "heap_pages" is the number of pages from the table that we expect to scan, or
3061 : * -1 if we don't expect to scan any.
3062 : *
3063 : * "index_pages" is the number of pages from the index that we expect to scan, or
3064 : * -1 if we don't expect to scan any.
3065 : */
3066 : int
3067 21924 : compute_parallel_worker(RelOptInfo *rel, double heap_pages, double index_pages)
3068 : {
3069 21924 : int parallel_workers = 0;
3070 :
3071 : /*
3072 : * If the user has set the parallel_workers reloption, use that; otherwise
3073 : * select a default number of workers.
3074 : */
3075 21924 : if (rel->rel_parallel_workers != -1)
3076 61 : parallel_workers = rel->rel_parallel_workers;
3077 : else
3078 : {
3079 : /*
3080 : * If the number of pages being scanned is insufficient to justify a
3081 : * parallel scan, just return zero ... unless it's an inheritance
3082 : * child. In that case, we want to generate a parallel path here
3083 : * anyway. It might not be worthwhile just for this relation, but
3084 : * when combined with all of its inheritance siblings it may well pay
3085 : * off.
3086 : */
3087 21863 : if (rel->reloptkind == RELOPT_BASEREL &&
3088 19891 : ((heap_pages >= 0 && heap_pages < min_parallel_table_scan_size) ||
3089 1010 : (index_pages >= 0 && index_pages < min_parallel_index_scan_size)))
3090 20892 : return 0;
3091 :
3092 971 : if (heap_pages >= 0)
3093 : {
3094 : int heap_parallel_threshold;
3095 829 : int heap_parallel_workers = 1;
3096 :
3097 : /*
3098 : * Select the number of workers based on the log of the size of
3099 : * the relation. This probably needs to be a good deal more
3100 : * sophisticated, but we need something here for now. Note that
3101 : * the upper limit of the min_parallel_table_scan_size GUC is
3102 : * chosen to prevent overflow here.
3103 : */
3104 829 : heap_parallel_threshold = Max(min_parallel_table_scan_size, 1);
3105 1690 : while (heap_pages >= (BlockNumber) (heap_parallel_threshold * 3))
3106 : {
3107 32 : heap_parallel_workers++;
3108 32 : heap_parallel_threshold *= 3;
3109 32 : if (heap_parallel_threshold > INT_MAX / 3)
3110 0 : break; /* avoid overflow */
3111 : }
3112 :
3113 829 : parallel_workers = heap_parallel_workers;
3114 : }
3115 :
3116 971 : if (index_pages >= 0)
3117 : {
3118 217 : int index_parallel_workers = 1;
3119 : int index_parallel_threshold;
3120 :
3121 : /* same calculation as for heap_pages above */
3122 217 : index_parallel_threshold = Max(min_parallel_index_scan_size, 1);
3123 434 : while (index_pages >= (BlockNumber) (index_parallel_threshold * 3))
3124 : {
3125 0 : index_parallel_workers++;
3126 0 : index_parallel_threshold *= 3;
3127 0 : if (index_parallel_threshold > INT_MAX / 3)
3128 0 : break; /* avoid overflow */
3129 : }
3130 :
3131 217 : if (parallel_workers > 0)
3132 75 : parallel_workers = Min(parallel_workers, index_parallel_workers);
3133 : else
3134 142 : parallel_workers = index_parallel_workers;
3135 : }
3136 : }
3137 :
3138 : /*
3139 : * In no case use more than max_parallel_workers_per_gather workers.
3140 : */
3141 1032 : parallel_workers = Min(parallel_workers, max_parallel_workers_per_gather);
3142 :
3143 1032 : return parallel_workers;
3144 : }
3145 :
3146 :
3147 : /*****************************************************************************
3148 : * DEBUG SUPPORT
3149 : *****************************************************************************/
3150 :
3151 : #ifdef OPTIMIZER_DEBUG
3152 :
3153 : static void
3154 : print_relids(PlannerInfo *root, Relids relids)
3155 : {
3156 : int x;
3157 : bool first = true;
3158 :
3159 : x = -1;
3160 : while ((x = bms_next_member(relids, x)) >= 0)
3161 : {
3162 : if (!first)
3163 : printf(" ");
3164 : if (x < root->simple_rel_array_size &&
3165 : root->simple_rte_array[x])
3166 : printf("%s", root->simple_rte_array[x]->eref->aliasname);
3167 : else
3168 : printf("%d", x);
3169 : first = false;
3170 : }
3171 : }
3172 :
3173 : static void
3174 : print_restrictclauses(PlannerInfo *root, List *clauses)
3175 : {
3176 : ListCell *l;
3177 :
3178 : foreach(l, clauses)
3179 : {
3180 : RestrictInfo *c = lfirst(l);
3181 :
3182 : print_expr((Node *) c->clause, root->parse->rtable);
3183 : if (lnext(l))
3184 : printf(", ");
3185 : }
3186 : }
3187 :
3188 : static void
3189 : print_path(PlannerInfo *root, Path *path, int indent)
3190 : {
3191 : const char *ptype;
3192 : bool join = false;
3193 : Path *subpath = NULL;
3194 : int i;
3195 :
3196 : switch (nodeTag(path))
3197 : {
3198 : case T_Path:
3199 : switch (path->pathtype)
3200 : {
3201 : case T_SeqScan:
3202 : ptype = "SeqScan";
3203 : break;
3204 : case T_SampleScan:
3205 : ptype = "SampleScan";
3206 : break;
3207 : case T_SubqueryScan:
3208 : ptype = "SubqueryScan";
3209 : break;
3210 : case T_FunctionScan:
3211 : ptype = "FunctionScan";
3212 : break;
3213 : case T_TableFuncScan:
3214 : ptype = "TableFuncScan";
3215 : break;
3216 : case T_ValuesScan:
3217 : ptype = "ValuesScan";
3218 : break;
3219 : case T_CteScan:
3220 : ptype = "CteScan";
3221 : break;
3222 : case T_WorkTableScan:
3223 : ptype = "WorkTableScan";
3224 : break;
3225 : default:
3226 : ptype = "???Path";
3227 : break;
3228 : }
3229 : break;
3230 : case T_IndexPath:
3231 : ptype = "IdxScan";
3232 : break;
3233 : case T_BitmapHeapPath:
3234 : ptype = "BitmapHeapScan";
3235 : break;
3236 : case T_BitmapAndPath:
3237 : ptype = "BitmapAndPath";
3238 : break;
3239 : case T_BitmapOrPath:
3240 : ptype = "BitmapOrPath";
3241 : break;
3242 : case T_TidPath:
3243 : ptype = "TidScan";
3244 : break;
3245 : case T_SubqueryScanPath:
3246 : ptype = "SubqueryScanScan";
3247 : break;
3248 : case T_ForeignPath:
3249 : ptype = "ForeignScan";
3250 : break;
3251 : case T_AppendPath:
3252 : ptype = "Append";
3253 : break;
3254 : case T_MergeAppendPath:
3255 : ptype = "MergeAppend";
3256 : break;
3257 : case T_ResultPath:
3258 : ptype = "Result";
3259 : break;
3260 : case T_MaterialPath:
3261 : ptype = "Material";
3262 : subpath = ((MaterialPath *) path)->subpath;
3263 : break;
3264 : case T_UniquePath:
3265 : ptype = "Unique";
3266 : subpath = ((UniquePath *) path)->subpath;
3267 : break;
3268 : case T_GatherPath:
3269 : ptype = "Gather";
3270 : subpath = ((GatherPath *) path)->subpath;
3271 : break;
3272 : case T_ProjectionPath:
3273 : ptype = "Projection";
3274 : subpath = ((ProjectionPath *) path)->subpath;
3275 : break;
3276 : case T_ProjectSetPath:
3277 : ptype = "ProjectSet";
3278 : subpath = ((ProjectSetPath *) path)->subpath;
3279 : break;
3280 : case T_SortPath:
3281 : ptype = "Sort";
3282 : subpath = ((SortPath *) path)->subpath;
3283 : break;
3284 : case T_GroupPath:
3285 : ptype = "Group";
3286 : subpath = ((GroupPath *) path)->subpath;
3287 : break;
3288 : case T_UpperUniquePath:
3289 : ptype = "UpperUnique";
3290 : subpath = ((UpperUniquePath *) path)->subpath;
3291 : break;
3292 : case T_AggPath:
3293 : ptype = "Agg";
3294 : subpath = ((AggPath *) path)->subpath;
3295 : break;
3296 : case T_GroupingSetsPath:
3297 : ptype = "GroupingSets";
3298 : subpath = ((GroupingSetsPath *) path)->subpath;
3299 : break;
3300 : case T_MinMaxAggPath:
3301 : ptype = "MinMaxAgg";
3302 : break;
3303 : case T_WindowAggPath:
3304 : ptype = "WindowAgg";
3305 : subpath = ((WindowAggPath *) path)->subpath;
3306 : break;
3307 : case T_SetOpPath:
3308 : ptype = "SetOp";
3309 : subpath = ((SetOpPath *) path)->subpath;
3310 : break;
3311 : case T_RecursiveUnionPath:
3312 : ptype = "RecursiveUnion";
3313 : break;
3314 : case T_LockRowsPath:
3315 : ptype = "LockRows";
3316 : subpath = ((LockRowsPath *) path)->subpath;
3317 : break;
3318 : case T_ModifyTablePath:
3319 : ptype = "ModifyTable";
3320 : break;
3321 : case T_LimitPath:
3322 : ptype = "Limit";
3323 : subpath = ((LimitPath *) path)->subpath;
3324 : break;
3325 : case T_NestPath:
3326 : ptype = "NestLoop";
3327 : join = true;
3328 : break;
3329 : case T_MergePath:
3330 : ptype = "MergeJoin";
3331 : join = true;
3332 : break;
3333 : case T_HashPath:
3334 : ptype = "HashJoin";
3335 : join = true;
3336 : break;
3337 : default:
3338 : ptype = "???Path";
3339 : break;
3340 : }
3341 :
3342 : for (i = 0; i < indent; i++)
3343 : printf("\t");
3344 : printf("%s", ptype);
3345 :
3346 : if (path->parent)
3347 : {
3348 : printf("(");
3349 : print_relids(root, path->parent->relids);
3350 : printf(")");
3351 : }
3352 : if (path->param_info)
3353 : {
3354 : printf(" required_outer (");
3355 : print_relids(root, path->param_info->ppi_req_outer);
3356 : printf(")");
3357 : }
3358 : printf(" rows=%.0f cost=%.2f..%.2f\n",
3359 : path->rows, path->startup_cost, path->total_cost);
3360 :
3361 : if (path->pathkeys)
3362 : {
3363 : for (i = 0; i < indent; i++)
3364 : printf("\t");
3365 : printf(" pathkeys: ");
3366 : print_pathkeys(path->pathkeys, root->parse->rtable);
3367 : }
3368 :
3369 : if (join)
3370 : {
3371 : JoinPath *jp = (JoinPath *) path;
3372 :
3373 : for (i = 0; i < indent; i++)
3374 : printf("\t");
3375 : printf(" clauses: ");
3376 : print_restrictclauses(root, jp->joinrestrictinfo);
3377 : printf("\n");
3378 :
3379 : if (IsA(path, MergePath))
3380 : {
3381 : MergePath *mp = (MergePath *) path;
3382 :
3383 : for (i = 0; i < indent; i++)
3384 : printf("\t");
3385 : printf(" sortouter=%d sortinner=%d materializeinner=%d\n",
3386 : ((mp->outersortkeys) ? 1 : 0),
3387 : ((mp->innersortkeys) ? 1 : 0),
3388 : ((mp->materialize_inner) ? 1 : 0));
3389 : }
3390 :
3391 : print_path(root, jp->outerjoinpath, indent + 1);
3392 : print_path(root, jp->innerjoinpath, indent + 1);
3393 : }
3394 :
3395 : if (subpath)
3396 : print_path(root, subpath, indent + 1);
3397 : }
3398 :
3399 : void
3400 : debug_print_rel(PlannerInfo *root, RelOptInfo *rel)
3401 : {
3402 : ListCell *l;
3403 :
3404 : printf("RELOPTINFO (");
3405 : print_relids(root, rel->relids);
3406 : printf("): rows=%.0f width=%d\n", rel->rows, rel->reltarget->width);
3407 :
3408 : if (rel->baserestrictinfo)
3409 : {
3410 : printf("\tbaserestrictinfo: ");
3411 : print_restrictclauses(root, rel->baserestrictinfo);
3412 : printf("\n");
3413 : }
3414 :
3415 : if (rel->joininfo)
3416 : {
3417 : printf("\tjoininfo: ");
3418 : print_restrictclauses(root, rel->joininfo);
3419 : printf("\n");
3420 : }
3421 :
3422 : printf("\tpath list:\n");
3423 : foreach(l, rel->pathlist)
3424 : print_path(root, lfirst(l), 1);
3425 : if (rel->cheapest_parameterized_paths)
3426 : {
3427 : printf("\n\tcheapest parameterized paths:\n");
3428 : foreach(l, rel->cheapest_parameterized_paths)
3429 : print_path(root, lfirst(l), 1);
3430 : }
3431 : if (rel->cheapest_startup_path)
3432 : {
3433 : printf("\n\tcheapest startup path:\n");
3434 : print_path(root, rel->cheapest_startup_path, 1);
3435 : }
3436 : if (rel->cheapest_total_path)
3437 : {
3438 : printf("\n\tcheapest total path:\n");
3439 : print_path(root, rel->cheapest_total_path, 1);
3440 : }
3441 : printf("\n");
3442 : fflush(stdout);
3443 : }
3444 :
3445 : #endif /* OPTIMIZER_DEBUG */
|
