PostgreSQL Source Code  git master
allpaths.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * allpaths.c
4  * Routines to find possible search paths for processing a query
5  *
6  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  *
10  * IDENTIFICATION
11  * src/backend/optimizer/path/allpaths.c
12  *
13  *-------------------------------------------------------------------------
14  */
15 
16 #include "postgres.h"
17 
18 #include <limits.h>
19 #include <math.h>
20 
21 #include "access/sysattr.h"
22 #include "access/tsmapi.h"
23 #include "catalog/pg_class.h"
24 #include "catalog/pg_operator.h"
25 #include "catalog/pg_proc.h"
26 #include "foreign/fdwapi.h"
27 #include "miscadmin.h"
28 #include "nodes/makefuncs.h"
29 #include "nodes/nodeFuncs.h"
30 #ifdef OPTIMIZER_DEBUG
31 #include "nodes/print.h"
32 #endif
33 #include "optimizer/appendinfo.h"
34 #include "optimizer/clauses.h"
35 #include "optimizer/cost.h"
36 #include "optimizer/geqo.h"
37 #include "optimizer/inherit.h"
38 #include "optimizer/optimizer.h"
39 #include "optimizer/pathnode.h"
40 #include "optimizer/paths.h"
41 #include "optimizer/plancat.h"
42 #include "optimizer/planner.h"
43 #include "optimizer/restrictinfo.h"
44 #include "optimizer/tlist.h"
45 #include "parser/parse_clause.h"
46 #include "parser/parsetree.h"
48 #include "partitioning/partprune.h"
49 #include "rewrite/rewriteManip.h"
50 #include "utils/lsyscache.h"
51 
52 
53 /* results of subquery_is_pushdown_safe */
54 typedef struct pushdown_safety_info
55 {
56  bool *unsafeColumns; /* which output columns are unsafe to use */
57  bool unsafeVolatile; /* don't push down volatile quals */
58  bool unsafeLeaky; /* don't push down leaky quals */
60 
61 /* These parameters are set by GUC */
62 bool enable_geqo = false; /* just in case GUC doesn't set it */
66 
67 /* Hook for plugins to get control in set_rel_pathlist() */
69 
70 /* Hook for plugins to replace standard_join_search() */
72 
73 
75 static void set_base_rel_sizes(PlannerInfo *root);
76 static void set_base_rel_pathlists(PlannerInfo *root);
77 static void set_rel_size(PlannerInfo *root, RelOptInfo *rel,
78  Index rti, RangeTblEntry *rte);
79 static void set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
80  Index rti, RangeTblEntry *rte);
81 static void set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel,
82  RangeTblEntry *rte);
83 static void create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel);
84 static void set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel,
85  RangeTblEntry *rte);
86 static void set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
87  RangeTblEntry *rte);
88 static void set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel,
89  RangeTblEntry *rte);
91  RangeTblEntry *rte);
92 static void set_foreign_size(PlannerInfo *root, RelOptInfo *rel,
93  RangeTblEntry *rte);
94 static void set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel,
95  RangeTblEntry *rte);
96 static void set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
97  Index rti, RangeTblEntry *rte);
98 static void set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
99  Index rti, RangeTblEntry *rte);
100 static void generate_orderedappend_paths(PlannerInfo *root, RelOptInfo *rel,
101  List *live_childrels,
102  List *all_child_pathkeys,
103  List *partitioned_rels);
105  RelOptInfo *rel,
106  Relids required_outer);
107 static void accumulate_append_subpath(Path *path,
108  List **subpaths, List **special_subpaths);
109 static Path *get_singleton_append_subpath(Path *path);
110 static void set_dummy_rel_pathlist(RelOptInfo *rel);
111 static void set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
112  Index rti, RangeTblEntry *rte);
113 static void set_function_pathlist(PlannerInfo *root, RelOptInfo *rel,
114  RangeTblEntry *rte);
115 static void set_values_pathlist(PlannerInfo *root, RelOptInfo *rel,
116  RangeTblEntry *rte);
117 static void set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel,
118  RangeTblEntry *rte);
119 static void set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel,
120  RangeTblEntry *rte);
121 static void set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel,
122  RangeTblEntry *rte);
123 static void set_result_pathlist(PlannerInfo *root, RelOptInfo *rel,
124  RangeTblEntry *rte);
125 static void set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel,
126  RangeTblEntry *rte);
127 static RelOptInfo *make_rel_from_joinlist(PlannerInfo *root, List *joinlist);
128 static bool subquery_is_pushdown_safe(Query *subquery, Query *topquery,
129  pushdown_safety_info *safetyInfo);
130 static bool recurse_pushdown_safe(Node *setOp, Query *topquery,
131  pushdown_safety_info *safetyInfo);
132 static void check_output_expressions(Query *subquery,
133  pushdown_safety_info *safetyInfo);
134 static void compare_tlist_datatypes(List *tlist, List *colTypes,
135  pushdown_safety_info *safetyInfo);
136 static bool targetIsInAllPartitionLists(TargetEntry *tle, Query *query);
137 static bool qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual,
138  pushdown_safety_info *safetyInfo);
139 static void subquery_push_qual(Query *subquery,
140  RangeTblEntry *rte, Index rti, Node *qual);
141 static void recurse_push_qual(Node *setOp, Query *topquery,
142  RangeTblEntry *rte, Index rti, Node *qual);
143 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
144 
145 
146 /*
147  * make_one_rel
148  * Finds all possible access paths for executing a query, returning a
149  * single rel that represents the join of all base rels in the query.
150  */
151 RelOptInfo *
152 make_one_rel(PlannerInfo *root, List *joinlist)
153 {
154  RelOptInfo *rel;
155  Index rti;
156  double total_pages;
157 
158  /*
159  * Construct the all_baserels Relids set.
160  */
161  root->all_baserels = NULL;
162  for (rti = 1; rti < root->simple_rel_array_size; rti++)
163  {
164  RelOptInfo *brel = root->simple_rel_array[rti];
165 
166  /* there may be empty slots corresponding to non-baserel RTEs */
167  if (brel == NULL)
168  continue;
169 
170  Assert(brel->relid == rti); /* sanity check on array */
171 
172  /* ignore RTEs that are "other rels" */
173  if (brel->reloptkind != RELOPT_BASEREL)
174  continue;
175 
176  root->all_baserels = bms_add_member(root->all_baserels, brel->relid);
177  }
178 
179  /* Mark base rels as to whether we care about fast-start plans */
181 
182  /*
183  * Compute size estimates and consider_parallel flags for each base rel.
184  */
185  set_base_rel_sizes(root);
186 
187  /*
188  * We should now have size estimates for every actual table involved in
189  * the query, and we also know which if any have been deleted from the
190  * query by join removal, pruned by partition pruning, or eliminated by
191  * constraint exclusion. So we can now compute total_table_pages.
192  *
193  * Note that appendrels are not double-counted here, even though we don't
194  * bother to distinguish RelOptInfos for appendrel parents, because the
195  * parents will have pages = 0.
196  *
197  * XXX if a table is self-joined, we will count it once per appearance,
198  * which perhaps is the wrong thing ... but that's not completely clear,
199  * and detecting self-joins here is difficult, so ignore it for now.
200  */
201  total_pages = 0;
202  for (rti = 1; rti < root->simple_rel_array_size; rti++)
203  {
204  RelOptInfo *brel = root->simple_rel_array[rti];
205 
206  if (brel == NULL)
207  continue;
208 
209  Assert(brel->relid == rti); /* sanity check on array */
210 
211  if (IS_DUMMY_REL(brel))
212  continue;
213 
214  if (IS_SIMPLE_REL(brel))
215  total_pages += (double) brel->pages;
216  }
217  root->total_table_pages = total_pages;
218 
219  /*
220  * Generate access paths for each base rel.
221  */
223 
224  /*
225  * Generate access paths for the entire join tree.
226  */
227  rel = make_rel_from_joinlist(root, joinlist);
228 
229  /*
230  * The result should join all and only the query's base rels.
231  */
232  Assert(bms_equal(rel->relids, root->all_baserels));
233 
234  return rel;
235 }
236 
237 /*
238  * set_base_rel_consider_startup
239  * Set the consider_[param_]startup flags for each base-relation entry.
240  *
241  * For the moment, we only deal with consider_param_startup here; because the
242  * logic for consider_startup is pretty trivial and is the same for every base
243  * relation, we just let build_simple_rel() initialize that flag correctly to
244  * start with. If that logic ever gets more complicated it would probably
245  * be better to move it here.
246  */
247 static void
249 {
250  /*
251  * Since parameterized paths can only be used on the inside of a nestloop
252  * join plan, there is usually little value in considering fast-start
253  * plans for them. However, for relations that are on the RHS of a SEMI
254  * or ANTI join, a fast-start plan can be useful because we're only going
255  * to care about fetching one tuple anyway.
256  *
257  * To minimize growth of planning time, we currently restrict this to
258  * cases where the RHS is a single base relation, not a join; there is no
259  * provision for consider_param_startup to get set at all on joinrels.
260  * Also we don't worry about appendrels. costsize.c's costing rules for
261  * nestloop semi/antijoins don't consider such cases either.
262  */
263  ListCell *lc;
264 
265  foreach(lc, root->join_info_list)
266  {
267  SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc);
268  int varno;
269 
270  if ((sjinfo->jointype == JOIN_SEMI || sjinfo->jointype == JOIN_ANTI) &&
271  bms_get_singleton_member(sjinfo->syn_righthand, &varno))
272  {
273  RelOptInfo *rel = find_base_rel(root, varno);
274 
275  rel->consider_param_startup = true;
276  }
277  }
278 }
279 
280 /*
281  * set_base_rel_sizes
282  * Set the size estimates (rows and widths) for each base-relation entry.
283  * Also determine whether to consider parallel paths for base relations.
284  *
285  * We do this in a separate pass over the base rels so that rowcount
286  * estimates are available for parameterized path generation, and also so
287  * that each rel's consider_parallel flag is set correctly before we begin to
288  * generate paths.
289  */
290 static void
292 {
293  Index rti;
294 
295  for (rti = 1; rti < root->simple_rel_array_size; rti++)
296  {
297  RelOptInfo *rel = root->simple_rel_array[rti];
298  RangeTblEntry *rte;
299 
300  /* there may be empty slots corresponding to non-baserel RTEs */
301  if (rel == NULL)
302  continue;
303 
304  Assert(rel->relid == rti); /* sanity check on array */
305 
306  /* ignore RTEs that are "other rels" */
307  if (rel->reloptkind != RELOPT_BASEREL)
308  continue;
309 
310  rte = root->simple_rte_array[rti];
311 
312  /*
313  * If parallelism is allowable for this query in general, see whether
314  * it's allowable for this rel in particular. We have to do this
315  * before set_rel_size(), because (a) if this rel is an inheritance
316  * parent, set_append_rel_size() will use and perhaps change the rel's
317  * consider_parallel flag, and (b) for some RTE types, set_rel_size()
318  * goes ahead and makes paths immediately.
319  */
320  if (root->glob->parallelModeOK)
321  set_rel_consider_parallel(root, rel, rte);
322 
323  set_rel_size(root, rel, rti, rte);
324  }
325 }
326 
327 /*
328  * set_base_rel_pathlists
329  * Finds all paths available for scanning each base-relation entry.
330  * Sequential scan and any available indices are considered.
331  * Each useful path is attached to its relation's 'pathlist' field.
332  */
333 static void
335 {
336  Index rti;
337 
338  for (rti = 1; rti < root->simple_rel_array_size; rti++)
339  {
340  RelOptInfo *rel = root->simple_rel_array[rti];
341 
342  /* there may be empty slots corresponding to non-baserel RTEs */
343  if (rel == NULL)
344  continue;
345 
346  Assert(rel->relid == rti); /* sanity check on array */
347 
348  /* ignore RTEs that are "other rels" */
349  if (rel->reloptkind != RELOPT_BASEREL)
350  continue;
351 
352  set_rel_pathlist(root, rel, rti, root->simple_rte_array[rti]);
353  }
354 }
355 
356 /*
357  * set_rel_size
358  * Set size estimates for a base relation
359  */
360 static void
362  Index rti, RangeTblEntry *rte)
363 {
364  if (rel->reloptkind == RELOPT_BASEREL &&
365  relation_excluded_by_constraints(root, rel, rte))
366  {
367  /*
368  * We proved we don't need to scan the rel via constraint exclusion,
369  * so set up a single dummy path for it. Here we only check this for
370  * regular baserels; if it's an otherrel, CE was already checked in
371  * set_append_rel_size().
372  *
373  * In this case, we go ahead and set up the relation's path right away
374  * instead of leaving it for set_rel_pathlist to do. This is because
375  * we don't have a convention for marking a rel as dummy except by
376  * assigning a dummy path to it.
377  */
379  }
380  else if (rte->inh)
381  {
382  /* It's an "append relation", process accordingly */
383  set_append_rel_size(root, rel, rti, rte);
384  }
385  else
386  {
387  switch (rel->rtekind)
388  {
389  case RTE_RELATION:
390  if (rte->relkind == RELKIND_FOREIGN_TABLE)
391  {
392  /* Foreign table */
393  set_foreign_size(root, rel, rte);
394  }
395  else if (rte->relkind == RELKIND_PARTITIONED_TABLE)
396  {
397  /*
398  * We could get here if asked to scan a partitioned table
399  * with ONLY. In that case we shouldn't scan any of the
400  * partitions, so mark it as a dummy rel.
401  */
403  }
404  else if (rte->tablesample != NULL)
405  {
406  /* Sampled relation */
407  set_tablesample_rel_size(root, rel, rte);
408  }
409  else
410  {
411  /* Plain relation */
412  set_plain_rel_size(root, rel, rte);
413  }
414  break;
415  case RTE_SUBQUERY:
416 
417  /*
418  * Subqueries don't support making a choice between
419  * parameterized and unparameterized paths, so just go ahead
420  * and build their paths immediately.
421  */
422  set_subquery_pathlist(root, rel, rti, rte);
423  break;
424  case RTE_FUNCTION:
425  set_function_size_estimates(root, rel);
426  break;
427  case RTE_TABLEFUNC:
428  set_tablefunc_size_estimates(root, rel);
429  break;
430  case RTE_VALUES:
431  set_values_size_estimates(root, rel);
432  break;
433  case RTE_CTE:
434 
435  /*
436  * CTEs don't support making a choice between parameterized
437  * and unparameterized paths, so just go ahead and build their
438  * paths immediately.
439  */
440  if (rte->self_reference)
441  set_worktable_pathlist(root, rel, rte);
442  else
443  set_cte_pathlist(root, rel, rte);
444  break;
445  case RTE_NAMEDTUPLESTORE:
446  /* Might as well just build the path immediately */
447  set_namedtuplestore_pathlist(root, rel, rte);
448  break;
449  case RTE_RESULT:
450  /* Might as well just build the path immediately */
451  set_result_pathlist(root, rel, rte);
452  break;
453  default:
454  elog(ERROR, "unexpected rtekind: %d", (int) rel->rtekind);
455  break;
456  }
457  }
458 
459  /*
460  * We insist that all non-dummy rels have a nonzero rowcount estimate.
461  */
462  Assert(rel->rows > 0 || IS_DUMMY_REL(rel));
463 }
464 
465 /*
466  * set_rel_pathlist
467  * Build access paths for a base relation
468  */
469 static void
471  Index rti, RangeTblEntry *rte)
472 {
473  if (IS_DUMMY_REL(rel))
474  {
475  /* We already proved the relation empty, so nothing more to do */
476  }
477  else if (rte->inh)
478  {
479  /* It's an "append relation", process accordingly */
480  set_append_rel_pathlist(root, rel, rti, rte);
481  }
482  else
483  {
484  switch (rel->rtekind)
485  {
486  case RTE_RELATION:
487  if (rte->relkind == RELKIND_FOREIGN_TABLE)
488  {
489  /* Foreign table */
490  set_foreign_pathlist(root, rel, rte);
491  }
492  else if (rte->tablesample != NULL)
493  {
494  /* Sampled relation */
495  set_tablesample_rel_pathlist(root, rel, rte);
496  }
497  else
498  {
499  /* Plain relation */
500  set_plain_rel_pathlist(root, rel, rte);
501  }
502  break;
503  case RTE_SUBQUERY:
504  /* Subquery --- fully handled during set_rel_size */
505  break;
506  case RTE_FUNCTION:
507  /* RangeFunction */
508  set_function_pathlist(root, rel, rte);
509  break;
510  case RTE_TABLEFUNC:
511  /* Table Function */
512  set_tablefunc_pathlist(root, rel, rte);
513  break;
514  case RTE_VALUES:
515  /* Values list */
516  set_values_pathlist(root, rel, rte);
517  break;
518  case RTE_CTE:
519  /* CTE reference --- fully handled during set_rel_size */
520  break;
521  case RTE_NAMEDTUPLESTORE:
522  /* tuplestore reference --- fully handled during set_rel_size */
523  break;
524  case RTE_RESULT:
525  /* simple Result --- fully handled during set_rel_size */
526  break;
527  default:
528  elog(ERROR, "unexpected rtekind: %d", (int) rel->rtekind);
529  break;
530  }
531  }
532 
533  /*
534  * Allow a plugin to editorialize on the set of Paths for this base
535  * relation. It could add new paths (such as CustomPaths) by calling
536  * add_path(), or add_partial_path() if parallel aware. It could also
537  * delete or modify paths added by the core code.
538  */
540  (*set_rel_pathlist_hook) (root, rel, rti, rte);
541 
542  /*
543  * If this is a baserel, we should normally consider gathering any partial
544  * paths we may have created for it. We have to do this after calling the
545  * set_rel_pathlist_hook, else it cannot add partial paths to be included
546  * here.
547  *
548  * However, if this is an inheritance child, skip it. Otherwise, we could
549  * end up with a very large number of gather nodes, each trying to grab
550  * its own pool of workers. Instead, we'll consider gathering partial
551  * paths for the parent appendrel.
552  *
553  * Also, if this is the topmost scan/join rel (that is, the only baserel),
554  * we postpone gathering until the final scan/join targetlist is available
555  * (see grouping_planner).
556  */
557  if (rel->reloptkind == RELOPT_BASEREL &&
558  bms_membership(root->all_baserels) != BMS_SINGLETON)
559  generate_gather_paths(root, rel, false);
560 
561  /* Now find the cheapest of the paths for this rel */
562  set_cheapest(rel);
563 
564 #ifdef OPTIMIZER_DEBUG
565  debug_print_rel(root, rel);
566 #endif
567 }
568 
569 /*
570  * set_plain_rel_size
571  * Set size estimates for a plain relation (no subquery, no inheritance)
572  */
573 static void
575 {
576  /*
577  * Test any partial indexes of rel for applicability. We must do this
578  * first since partial unique indexes can affect size estimates.
579  */
580  check_index_predicates(root, rel);
581 
582  /* Mark rel with estimated output rows, width, etc */
583  set_baserel_size_estimates(root, rel);
584 }
585 
586 /*
587  * If this relation could possibly be scanned from within a worker, then set
588  * its consider_parallel flag.
589  */
590 static void
592  RangeTblEntry *rte)
593 {
594  /*
595  * The flag has previously been initialized to false, so we can just
596  * return if it becomes clear that we can't safely set it.
597  */
598  Assert(!rel->consider_parallel);
599 
600  /* Don't call this if parallelism is disallowed for the entire query. */
601  Assert(root->glob->parallelModeOK);
602 
603  /* This should only be called for baserels and appendrel children. */
604  Assert(IS_SIMPLE_REL(rel));
605 
606  /* Assorted checks based on rtekind. */
607  switch (rte->rtekind)
608  {
609  case RTE_RELATION:
610 
611  /*
612  * Currently, parallel workers can't access the leader's temporary
613  * tables. We could possibly relax this if the wrote all of its
614  * local buffers at the start of the query and made no changes
615  * thereafter (maybe we could allow hint bit changes), and if we
616  * taught the workers to read them. Writing a large number of
617  * temporary buffers could be expensive, though, and we don't have
618  * the rest of the necessary infrastructure right now anyway. So
619  * for now, bail out if we see a temporary table.
620  */
621  if (get_rel_persistence(rte->relid) == RELPERSISTENCE_TEMP)
622  return;
623 
624  /*
625  * Table sampling can be pushed down to workers if the sample
626  * function and its arguments are safe.
627  */
628  if (rte->tablesample != NULL)
629  {
630  char proparallel = func_parallel(rte->tablesample->tsmhandler);
631 
632  if (proparallel != PROPARALLEL_SAFE)
633  return;
634  if (!is_parallel_safe(root, (Node *) rte->tablesample->args))
635  return;
636  }
637 
638  /*
639  * Ask FDWs whether they can support performing a ForeignScan
640  * within a worker. Most often, the answer will be no. For
641  * example, if the nature of the FDW is such that it opens a TCP
642  * connection with a remote server, each parallel worker would end
643  * up with a separate connection, and these connections might not
644  * be appropriately coordinated between workers and the leader.
645  */
646  if (rte->relkind == RELKIND_FOREIGN_TABLE)
647  {
648  Assert(rel->fdwroutine);
650  return;
651  if (!rel->fdwroutine->IsForeignScanParallelSafe(root, rel, rte))
652  return;
653  }
654 
655  /*
656  * There are additional considerations for appendrels, which we'll
657  * deal with in set_append_rel_size and set_append_rel_pathlist.
658  * For now, just set consider_parallel based on the rel's own
659  * quals and targetlist.
660  */
661  break;
662 
663  case RTE_SUBQUERY:
664 
665  /*
666  * There's no intrinsic problem with scanning a subquery-in-FROM
667  * (as distinct from a SubPlan or InitPlan) in a parallel worker.
668  * If the subquery doesn't happen to have any parallel-safe paths,
669  * then flagging it as consider_parallel won't change anything,
670  * but that's true for plain tables, too. We must set
671  * consider_parallel based on the rel's own quals and targetlist,
672  * so that if a subquery path is parallel-safe but the quals and
673  * projection we're sticking onto it are not, we correctly mark
674  * the SubqueryScanPath as not parallel-safe. (Note that
675  * set_subquery_pathlist() might push some of these quals down
676  * into the subquery itself, but that doesn't change anything.)
677  *
678  * We can't push sub-select containing LIMIT/OFFSET to workers as
679  * there is no guarantee that the row order will be fully
680  * deterministic, and applying LIMIT/OFFSET will lead to
681  * inconsistent results at the top-level. (In some cases, where
682  * the result is ordered, we could relax this restriction. But it
683  * doesn't currently seem worth expending extra effort to do so.)
684  */
685  {
686  Query *subquery = castNode(Query, rte->subquery);
687 
688  if (limit_needed(subquery))
689  return;
690  }
691  break;
692 
693  case RTE_JOIN:
694  /* Shouldn't happen; we're only considering baserels here. */
695  Assert(false);
696  return;
697 
698  case RTE_FUNCTION:
699  /* Check for parallel-restricted functions. */
700  if (!is_parallel_safe(root, (Node *) rte->functions))
701  return;
702  break;
703 
704  case RTE_TABLEFUNC:
705  /* not parallel safe */
706  return;
707 
708  case RTE_VALUES:
709  /* Check for parallel-restricted functions. */
710  if (!is_parallel_safe(root, (Node *) rte->values_lists))
711  return;
712  break;
713 
714  case RTE_CTE:
715 
716  /*
717  * CTE tuplestores aren't shared among parallel workers, so we
718  * force all CTE scans to happen in the leader. Also, populating
719  * the CTE would require executing a subplan that's not available
720  * in the worker, might be parallel-restricted, and must get
721  * executed only once.
722  */
723  return;
724 
725  case RTE_NAMEDTUPLESTORE:
726 
727  /*
728  * tuplestore cannot be shared, at least without more
729  * infrastructure to support that.
730  */
731  return;
732 
733  case RTE_RESULT:
734  /* RESULT RTEs, in themselves, are no problem. */
735  break;
736  }
737 
738  /*
739  * If there's anything in baserestrictinfo that's parallel-restricted, we
740  * give up on parallelizing access to this relation. We could consider
741  * instead postponing application of the restricted quals until we're
742  * above all the parallelism in the plan tree, but it's not clear that
743  * that would be a win in very many cases, and it might be tricky to make
744  * outer join clauses work correctly. It would likely break equivalence
745  * classes, too.
746  */
747  if (!is_parallel_safe(root, (Node *) rel->baserestrictinfo))
748  return;
749 
750  /*
751  * Likewise, if the relation's outputs are not parallel-safe, give up.
752  * (Usually, they're just Vars, but sometimes they're not.)
753  */
754  if (!is_parallel_safe(root, (Node *) rel->reltarget->exprs))
755  return;
756 
757  /* We have a winner. */
758  rel->consider_parallel = true;
759 }
760 
761 /*
762  * set_plain_rel_pathlist
763  * Build access paths for a plain relation (no subquery, no inheritance)
764  */
765 static void
767 {
768  Relids required_outer;
769 
770  /*
771  * We don't support pushing join clauses into the quals of a seqscan, but
772  * it could still have required parameterization due to LATERAL refs in
773  * its tlist.
774  */
775  required_outer = rel->lateral_relids;
776 
777  /* Consider sequential scan */
778  add_path(rel, create_seqscan_path(root, rel, required_outer, 0));
779 
780  /* If appropriate, consider parallel sequential scan */
781  if (rel->consider_parallel && required_outer == NULL)
782  create_plain_partial_paths(root, rel);
783 
784  /* Consider index scans */
785  create_index_paths(root, rel);
786 
787  /* Consider TID scans */
788  create_tidscan_paths(root, rel);
789 }
790 
791 /*
792  * create_plain_partial_paths
793  * Build partial access paths for parallel scan of a plain relation
794  */
795 static void
797 {
798  int parallel_workers;
799 
800  parallel_workers = compute_parallel_worker(rel, rel->pages, -1,
802 
803  /* If any limit was set to zero, the user doesn't want a parallel scan. */
804  if (parallel_workers <= 0)
805  return;
806 
807  /* Add an unordered partial path based on a parallel sequential scan. */
808  add_partial_path(rel, create_seqscan_path(root, rel, NULL, parallel_workers));
809 }
810 
811 /*
812  * set_tablesample_rel_size
813  * Set size estimates for a sampled relation
814  */
815 static void
817 {
818  TableSampleClause *tsc = rte->tablesample;
819  TsmRoutine *tsm;
820  BlockNumber pages;
821  double tuples;
822 
823  /*
824  * Test any partial indexes of rel for applicability. We must do this
825  * first since partial unique indexes can affect size estimates.
826  */
827  check_index_predicates(root, rel);
828 
829  /*
830  * Call the sampling method's estimation function to estimate the number
831  * of pages it will read and the number of tuples it will return. (Note:
832  * we assume the function returns sane values.)
833  */
834  tsm = GetTsmRoutine(tsc->tsmhandler);
835  tsm->SampleScanGetSampleSize(root, rel, tsc->args,
836  &pages, &tuples);
837 
838  /*
839  * For the moment, because we will only consider a SampleScan path for the
840  * rel, it's okay to just overwrite the pages and tuples estimates for the
841  * whole relation. If we ever consider multiple path types for sampled
842  * rels, we'll need more complication.
843  */
844  rel->pages = pages;
845  rel->tuples = tuples;
846 
847  /* Mark rel with estimated output rows, width, etc */
848  set_baserel_size_estimates(root, rel);
849 }
850 
851 /*
852  * set_tablesample_rel_pathlist
853  * Build access paths for a sampled relation
854  */
855 static void
857 {
858  Relids required_outer;
859  Path *path;
860 
861  /*
862  * We don't support pushing join clauses into the quals of a samplescan,
863  * but it could still have required parameterization due to LATERAL refs
864  * in its tlist or TABLESAMPLE arguments.
865  */
866  required_outer = rel->lateral_relids;
867 
868  /* Consider sampled scan */
869  path = create_samplescan_path(root, rel, required_outer);
870 
871  /*
872  * If the sampling method does not support repeatable scans, we must avoid
873  * plans that would scan the rel multiple times. Ideally, we'd simply
874  * avoid putting the rel on the inside of a nestloop join; but adding such
875  * a consideration to the planner seems like a great deal of complication
876  * to support an uncommon usage of second-rate sampling methods. Instead,
877  * if there is a risk that the query might perform an unsafe join, just
878  * wrap the SampleScan in a Materialize node. We can check for joins by
879  * counting the membership of all_baserels (note that this correctly
880  * counts inheritance trees as single rels). If we're inside a subquery,
881  * we can't easily check whether a join might occur in the outer query, so
882  * just assume one is possible.
883  *
884  * GetTsmRoutine is relatively expensive compared to the other tests here,
885  * so check repeatable_across_scans last, even though that's a bit odd.
886  */
887  if ((root->query_level > 1 ||
890  {
891  path = (Path *) create_material_path(rel, path);
892  }
893 
894  add_path(rel, path);
895 
896  /* For the moment, at least, there are no other paths to consider */
897 }
898 
899 /*
900  * set_foreign_size
901  * Set size estimates for a foreign table RTE
902  */
903 static void
905 {
906  /* Mark rel with estimated output rows, width, etc */
907  set_foreign_size_estimates(root, rel);
908 
909  /* Let FDW adjust the size estimates, if it can */
910  rel->fdwroutine->GetForeignRelSize(root, rel, rte->relid);
911 
912  /* ... but do not let it set the rows estimate to zero */
913  rel->rows = clamp_row_est(rel->rows);
914 }
915 
916 /*
917  * set_foreign_pathlist
918  * Build access paths for a foreign table RTE
919  */
920 static void
922 {
923  /* Call the FDW's GetForeignPaths function to generate path(s) */
924  rel->fdwroutine->GetForeignPaths(root, rel, rte->relid);
925 }
926 
927 /*
928  * set_append_rel_size
929  * Set size estimates for a simple "append relation"
930  *
931  * The passed-in rel and RTE represent the entire append relation. The
932  * relation's contents are computed by appending together the output of the
933  * individual member relations. Note that in the non-partitioned inheritance
934  * case, the first member relation is actually the same table as is mentioned
935  * in the parent RTE ... but it has a different RTE and RelOptInfo. This is
936  * a good thing because their outputs are not the same size.
937  */
938 static void
940  Index rti, RangeTblEntry *rte)
941 {
942  int parentRTindex = rti;
943  bool has_live_children;
944  double parent_rows;
945  double parent_size;
946  double *parent_attrsizes;
947  int nattrs;
948  ListCell *l;
949 
950  /* Guard against stack overflow due to overly deep inheritance tree. */
952 
953  Assert(IS_SIMPLE_REL(rel));
954 
955  /*
956  * Initialize partitioned_child_rels to contain this RT index.
957  *
958  * Note that during the set_append_rel_pathlist() phase, we will bubble up
959  * the indexes of partitioned relations that appear down in the tree, so
960  * that when we've created Paths for all the children, the root
961  * partitioned table's list will contain all such indexes.
962  */
963  if (rte->relkind == RELKIND_PARTITIONED_TABLE)
965 
966  /*
967  * If this is a partitioned baserel, set the consider_partitionwise_join
968  * flag; currently, we only consider partitionwise joins with the baserel
969  * if its targetlist doesn't contain a whole-row Var.
970  */
972  rel->reloptkind == RELOPT_BASEREL &&
973  rte->relkind == RELKIND_PARTITIONED_TABLE &&
974  rel->attr_needed[InvalidAttrNumber - rel->min_attr] == NULL)
975  rel->consider_partitionwise_join = true;
976 
977  /*
978  * Initialize to compute size estimates for whole append relation.
979  *
980  * We handle width estimates by weighting the widths of different child
981  * rels proportionally to their number of rows. This is sensible because
982  * the use of width estimates is mainly to compute the total relation
983  * "footprint" if we have to sort or hash it. To do this, we sum the
984  * total equivalent size (in "double" arithmetic) and then divide by the
985  * total rowcount estimate. This is done separately for the total rel
986  * width and each attribute.
987  *
988  * Note: if you consider changing this logic, beware that child rels could
989  * have zero rows and/or width, if they were excluded by constraints.
990  */
991  has_live_children = false;
992  parent_rows = 0;
993  parent_size = 0;
994  nattrs = rel->max_attr - rel->min_attr + 1;
995  parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
996 
997  foreach(l, root->append_rel_list)
998  {
999  AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1000  int childRTindex;
1001  RangeTblEntry *childRTE;
1002  RelOptInfo *childrel;
1003  ListCell *parentvars;
1004  ListCell *childvars;
1005 
1006  /* append_rel_list contains all append rels; ignore others */
1007  if (appinfo->parent_relid != parentRTindex)
1008  continue;
1009 
1010  childRTindex = appinfo->child_relid;
1011  childRTE = root->simple_rte_array[childRTindex];
1012 
1013  /*
1014  * The child rel's RelOptInfo was already created during
1015  * add_other_rels_to_query.
1016  */
1017  childrel = find_base_rel(root, childRTindex);
1019 
1020  /* We may have already proven the child to be dummy. */
1021  if (IS_DUMMY_REL(childrel))
1022  continue;
1023 
1024  /*
1025  * We have to copy the parent's targetlist and quals to the child,
1026  * with appropriate substitution of variables. However, the
1027  * baserestrictinfo quals were already copied/substituted when the
1028  * child RelOptInfo was built. So we don't need any additional setup
1029  * before applying constraint exclusion.
1030  */
1031  if (relation_excluded_by_constraints(root, childrel, childRTE))
1032  {
1033  /*
1034  * This child need not be scanned, so we can omit it from the
1035  * appendrel.
1036  */
1037  set_dummy_rel_pathlist(childrel);
1038  continue;
1039  }
1040 
1041  /*
1042  * Constraint exclusion failed, so copy the parent's join quals and
1043  * targetlist to the child, with appropriate variable substitutions.
1044  *
1045  * NB: the resulting childrel->reltarget->exprs may contain arbitrary
1046  * expressions, which otherwise would not occur in a rel's targetlist.
1047  * Code that might be looking at an appendrel child must cope with
1048  * such. (Normally, a rel's targetlist would only include Vars and
1049  * PlaceHolderVars.) XXX we do not bother to update the cost or width
1050  * fields of childrel->reltarget; not clear if that would be useful.
1051  */
1052  childrel->joininfo = (List *)
1054  (Node *) rel->joininfo,
1055  1, &appinfo);
1056  childrel->reltarget->exprs = (List *)
1058  (Node *) rel->reltarget->exprs,
1059  1, &appinfo);
1060 
1061  /*
1062  * We have to make child entries in the EquivalenceClass data
1063  * structures as well. This is needed either if the parent
1064  * participates in some eclass joins (because we will want to consider
1065  * inner-indexscan joins on the individual children) or if the parent
1066  * has useful pathkeys (because we should try to build MergeAppend
1067  * paths that produce those sort orderings).
1068  */
1069  if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
1070  add_child_rel_equivalences(root, appinfo, rel, childrel);
1071  childrel->has_eclass_joins = rel->has_eclass_joins;
1072 
1073  /*
1074  * Note: we could compute appropriate attr_needed data for the child's
1075  * variables, by transforming the parent's attr_needed through the
1076  * translated_vars mapping. However, currently there's no need
1077  * because attr_needed is only examined for base relations not
1078  * otherrels. So we just leave the child's attr_needed empty.
1079  */
1080 
1081  /*
1082  * If we consider partitionwise joins with the parent rel, do the same
1083  * for partitioned child rels.
1084  *
1085  * Note: here we abuse the consider_partitionwise_join flag by setting
1086  * it for child rels that are not themselves partitioned. We do so to
1087  * tell try_partitionwise_join() that the child rel is sufficiently
1088  * valid to be used as a per-partition input, even if it later gets
1089  * proven to be dummy. (It's not usable until we've set up the
1090  * reltarget and EC entries, which we just did.)
1091  */
1092  if (rel->consider_partitionwise_join)
1093  childrel->consider_partitionwise_join = true;
1094 
1095  /*
1096  * If parallelism is allowable for this query in general, see whether
1097  * it's allowable for this childrel in particular. But if we've
1098  * already decided the appendrel is not parallel-safe as a whole,
1099  * there's no point in considering parallelism for this child. For
1100  * consistency, do this before calling set_rel_size() for the child.
1101  */
1102  if (root->glob->parallelModeOK && rel->consider_parallel)
1103  set_rel_consider_parallel(root, childrel, childRTE);
1104 
1105  /*
1106  * Compute the child's size.
1107  */
1108  set_rel_size(root, childrel, childRTindex, childRTE);
1109 
1110  /*
1111  * It is possible that constraint exclusion detected a contradiction
1112  * within a child subquery, even though we didn't prove one above. If
1113  * so, we can skip this child.
1114  */
1115  if (IS_DUMMY_REL(childrel))
1116  continue;
1117 
1118  /* We have at least one live child. */
1119  has_live_children = true;
1120 
1121  /*
1122  * If any live child is not parallel-safe, treat the whole appendrel
1123  * as not parallel-safe. In future we might be able to generate plans
1124  * in which some children are farmed out to workers while others are
1125  * not; but we don't have that today, so it's a waste to consider
1126  * partial paths anywhere in the appendrel unless it's all safe.
1127  * (Child rels visited before this one will be unmarked in
1128  * set_append_rel_pathlist().)
1129  */
1130  if (!childrel->consider_parallel)
1131  rel->consider_parallel = false;
1132 
1133  /*
1134  * Accumulate size information from each live child.
1135  */
1136  Assert(childrel->rows > 0);
1137 
1138  parent_rows += childrel->rows;
1139  parent_size += childrel->reltarget->width * childrel->rows;
1140 
1141  /*
1142  * Accumulate per-column estimates too. We need not do anything for
1143  * PlaceHolderVars in the parent list. If child expression isn't a
1144  * Var, or we didn't record a width estimate for it, we have to fall
1145  * back on a datatype-based estimate.
1146  *
1147  * By construction, child's targetlist is 1-to-1 with parent's.
1148  */
1149  forboth(parentvars, rel->reltarget->exprs,
1150  childvars, childrel->reltarget->exprs)
1151  {
1152  Var *parentvar = (Var *) lfirst(parentvars);
1153  Node *childvar = (Node *) lfirst(childvars);
1154 
1155  if (IsA(parentvar, Var))
1156  {
1157  int pndx = parentvar->varattno - rel->min_attr;
1158  int32 child_width = 0;
1159 
1160  if (IsA(childvar, Var) &&
1161  ((Var *) childvar)->varno == childrel->relid)
1162  {
1163  int cndx = ((Var *) childvar)->varattno - childrel->min_attr;
1164 
1165  child_width = childrel->attr_widths[cndx];
1166  }
1167  if (child_width <= 0)
1168  child_width = get_typavgwidth(exprType(childvar),
1169  exprTypmod(childvar));
1170  Assert(child_width > 0);
1171  parent_attrsizes[pndx] += child_width * childrel->rows;
1172  }
1173  }
1174  }
1175 
1176  if (has_live_children)
1177  {
1178  /*
1179  * Save the finished size estimates.
1180  */
1181  int i;
1182 
1183  Assert(parent_rows > 0);
1184  rel->rows = parent_rows;
1185  rel->reltarget->width = rint(parent_size / parent_rows);
1186  for (i = 0; i < nattrs; i++)
1187  rel->attr_widths[i] = rint(parent_attrsizes[i] / parent_rows);
1188 
1189  /*
1190  * Set "raw tuples" count equal to "rows" for the appendrel; needed
1191  * because some places assume rel->tuples is valid for any baserel.
1192  */
1193  rel->tuples = parent_rows;
1194 
1195  /*
1196  * Note that we leave rel->pages as zero; this is important to avoid
1197  * double-counting the appendrel tree in total_table_pages.
1198  */
1199  }
1200  else
1201  {
1202  /*
1203  * All children were excluded by constraints, so mark the whole
1204  * appendrel dummy. We must do this in this phase so that the rel's
1205  * dummy-ness is visible when we generate paths for other rels.
1206  */
1208  }
1209 
1210  pfree(parent_attrsizes);
1211 }
1212 
1213 /*
1214  * set_append_rel_pathlist
1215  * Build access paths for an "append relation"
1216  */
1217 static void
1219  Index rti, RangeTblEntry *rte)
1220 {
1221  int parentRTindex = rti;
1222  List *live_childrels = NIL;
1223  ListCell *l;
1224 
1225  /*
1226  * Generate access paths for each member relation, and remember the
1227  * non-dummy children.
1228  */
1229  foreach(l, root->append_rel_list)
1230  {
1231  AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1232  int childRTindex;
1233  RangeTblEntry *childRTE;
1234  RelOptInfo *childrel;
1235 
1236  /* append_rel_list contains all append rels; ignore others */
1237  if (appinfo->parent_relid != parentRTindex)
1238  continue;
1239 
1240  /* Re-locate the child RTE and RelOptInfo */
1241  childRTindex = appinfo->child_relid;
1242  childRTE = root->simple_rte_array[childRTindex];
1243  childrel = root->simple_rel_array[childRTindex];
1244 
1245  /*
1246  * If set_append_rel_size() decided the parent appendrel was
1247  * parallel-unsafe at some point after visiting this child rel, we
1248  * need to propagate the unsafety marking down to the child, so that
1249  * we don't generate useless partial paths for it.
1250  */
1251  if (!rel->consider_parallel)
1252  childrel->consider_parallel = false;
1253 
1254  /*
1255  * Compute the child's access paths.
1256  */
1257  set_rel_pathlist(root, childrel, childRTindex, childRTE);
1258 
1259  /*
1260  * If child is dummy, ignore it.
1261  */
1262  if (IS_DUMMY_REL(childrel))
1263  continue;
1264 
1265  /* Bubble up childrel's partitioned children. */
1266  if (rel->part_scheme)
1267  rel->partitioned_child_rels =
1269  childrel->partitioned_child_rels);
1270 
1271  /*
1272  * Child is live, so add it to the live_childrels list for use below.
1273  */
1274  live_childrels = lappend(live_childrels, childrel);
1275  }
1276 
1277  /* Add paths to the append relation. */
1278  add_paths_to_append_rel(root, rel, live_childrels);
1279 }
1280 
1281 
1282 /*
1283  * add_paths_to_append_rel
1284  * Generate paths for the given append relation given the set of non-dummy
1285  * child rels.
1286  *
1287  * The function collects all parameterizations and orderings supported by the
1288  * non-dummy children. For every such parameterization or ordering, it creates
1289  * an append path collecting one path from each non-dummy child with given
1290  * parameterization or ordering. Similarly it collects partial paths from
1291  * non-dummy children to create partial append paths.
1292  */
1293 void
1295  List *live_childrels)
1296 {
1297  List *subpaths = NIL;
1298  bool subpaths_valid = true;
1299  List *partial_subpaths = NIL;
1300  List *pa_partial_subpaths = NIL;
1301  List *pa_nonpartial_subpaths = NIL;
1302  bool partial_subpaths_valid = true;
1303  bool pa_subpaths_valid;
1304  List *all_child_pathkeys = NIL;
1305  List *all_child_outers = NIL;
1306  ListCell *l;
1307  List *partitioned_rels = NIL;
1308  double partial_rows = -1;
1309 
1310  /* If appropriate, consider parallel append */
1311  pa_subpaths_valid = enable_parallel_append && rel->consider_parallel;
1312 
1313  /*
1314  * AppendPath generated for partitioned tables must record the RT indexes
1315  * of partitioned tables that are direct or indirect children of this
1316  * Append rel.
1317  *
1318  * AppendPath may be for a sub-query RTE (UNION ALL), in which case, 'rel'
1319  * itself does not represent a partitioned relation, but the child sub-
1320  * queries may contain references to partitioned relations. The loop
1321  * below will look for such children and collect them in a list to be
1322  * passed to the path creation function. (This assumes that we don't need
1323  * to look through multiple levels of subquery RTEs; if we ever do, we
1324  * could consider stuffing the list we generate here into sub-query RTE's
1325  * RelOptInfo, just like we do for partitioned rels, which would be used
1326  * when populating our parent rel with paths. For the present, that
1327  * appears to be unnecessary.)
1328  */
1329  if (rel->part_scheme != NULL)
1330  {
1331  if (IS_SIMPLE_REL(rel))
1332  partitioned_rels = list_make1(rel->partitioned_child_rels);
1333  else if (IS_JOIN_REL(rel))
1334  {
1335  int relid = -1;
1336  List *partrels = NIL;
1337 
1338  /*
1339  * For a partitioned joinrel, concatenate the component rels'
1340  * partitioned_child_rels lists.
1341  */
1342  while ((relid = bms_next_member(rel->relids, relid)) >= 0)
1343  {
1344  RelOptInfo *component;
1345 
1346  Assert(relid >= 1 && relid < root->simple_rel_array_size);
1347  component = root->simple_rel_array[relid];
1348  Assert(component->part_scheme != NULL);
1349  Assert(list_length(component->partitioned_child_rels) >= 1);
1350  partrels = list_concat(partrels,
1351  component->partitioned_child_rels);
1352  }
1353 
1354  partitioned_rels = list_make1(partrels);
1355  }
1356 
1357  Assert(list_length(partitioned_rels) >= 1);
1358  }
1359 
1360  /*
1361  * For every non-dummy child, remember the cheapest path. Also, identify
1362  * all pathkeys (orderings) and parameterizations (required_outer sets)
1363  * available for the non-dummy member relations.
1364  */
1365  foreach(l, live_childrels)
1366  {
1367  RelOptInfo *childrel = lfirst(l);
1368  ListCell *lcp;
1369  Path *cheapest_partial_path = NULL;
1370 
1371  /*
1372  * For UNION ALLs with non-empty partitioned_child_rels, accumulate
1373  * the Lists of child relations.
1374  */
1375  if (rel->rtekind == RTE_SUBQUERY && childrel->partitioned_child_rels != NIL)
1376  partitioned_rels = lappend(partitioned_rels,
1377  childrel->partitioned_child_rels);
1378 
1379  /*
1380  * If child has an unparameterized cheapest-total path, add that to
1381  * the unparameterized Append path we are constructing for the parent.
1382  * If not, there's no workable unparameterized path.
1383  *
1384  * With partitionwise aggregates, the child rel's pathlist may be
1385  * empty, so don't assume that a path exists here.
1386  */
1387  if (childrel->pathlist != NIL &&
1388  childrel->cheapest_total_path->param_info == NULL)
1390  &subpaths, NULL);
1391  else
1392  subpaths_valid = false;
1393 
1394  /* Same idea, but for a partial plan. */
1395  if (childrel->partial_pathlist != NIL)
1396  {
1397  cheapest_partial_path = linitial(childrel->partial_pathlist);
1398  accumulate_append_subpath(cheapest_partial_path,
1399  &partial_subpaths, NULL);
1400  }
1401  else
1402  partial_subpaths_valid = false;
1403 
1404  /*
1405  * Same idea, but for a parallel append mixing partial and non-partial
1406  * paths.
1407  */
1408  if (pa_subpaths_valid)
1409  {
1410  Path *nppath = NULL;
1411 
1412  nppath =
1414 
1415  if (cheapest_partial_path == NULL && nppath == NULL)
1416  {
1417  /* Neither a partial nor a parallel-safe path? Forget it. */
1418  pa_subpaths_valid = false;
1419  }
1420  else if (nppath == NULL ||
1421  (cheapest_partial_path != NULL &&
1422  cheapest_partial_path->total_cost < nppath->total_cost))
1423  {
1424  /* Partial path is cheaper or the only option. */
1425  Assert(cheapest_partial_path != NULL);
1426  accumulate_append_subpath(cheapest_partial_path,
1427  &pa_partial_subpaths,
1428  &pa_nonpartial_subpaths);
1429 
1430  }
1431  else
1432  {
1433  /*
1434  * Either we've got only a non-partial path, or we think that
1435  * a single backend can execute the best non-partial path
1436  * faster than all the parallel backends working together can
1437  * execute the best partial path.
1438  *
1439  * It might make sense to be more aggressive here. Even if
1440  * the best non-partial path is more expensive than the best
1441  * partial path, it could still be better to choose the
1442  * non-partial path if there are several such paths that can
1443  * be given to different workers. For now, we don't try to
1444  * figure that out.
1445  */
1447  &pa_nonpartial_subpaths,
1448  NULL);
1449  }
1450  }
1451 
1452  /*
1453  * Collect lists of all the available path orderings and
1454  * parameterizations for all the children. We use these as a
1455  * heuristic to indicate which sort orderings and parameterizations we
1456  * should build Append and MergeAppend paths for.
1457  */
1458  foreach(lcp, childrel->pathlist)
1459  {
1460  Path *childpath = (Path *) lfirst(lcp);
1461  List *childkeys = childpath->pathkeys;
1462  Relids childouter = PATH_REQ_OUTER(childpath);
1463 
1464  /* Unsorted paths don't contribute to pathkey list */
1465  if (childkeys != NIL)
1466  {
1467  ListCell *lpk;
1468  bool found = false;
1469 
1470  /* Have we already seen this ordering? */
1471  foreach(lpk, all_child_pathkeys)
1472  {
1473  List *existing_pathkeys = (List *) lfirst(lpk);
1474 
1475  if (compare_pathkeys(existing_pathkeys,
1476  childkeys) == PATHKEYS_EQUAL)
1477  {
1478  found = true;
1479  break;
1480  }
1481  }
1482  if (!found)
1483  {
1484  /* No, so add it to all_child_pathkeys */
1485  all_child_pathkeys = lappend(all_child_pathkeys,
1486  childkeys);
1487  }
1488  }
1489 
1490  /* Unparameterized paths don't contribute to param-set list */
1491  if (childouter)
1492  {
1493  ListCell *lco;
1494  bool found = false;
1495 
1496  /* Have we already seen this param set? */
1497  foreach(lco, all_child_outers)
1498  {
1499  Relids existing_outers = (Relids) lfirst(lco);
1500 
1501  if (bms_equal(existing_outers, childouter))
1502  {
1503  found = true;
1504  break;
1505  }
1506  }
1507  if (!found)
1508  {
1509  /* No, so add it to all_child_outers */
1510  all_child_outers = lappend(all_child_outers,
1511  childouter);
1512  }
1513  }
1514  }
1515  }
1516 
1517  /*
1518  * If we found unparameterized paths for all children, build an unordered,
1519  * unparameterized Append path for the rel. (Note: this is correct even
1520  * if we have zero or one live subpath due to constraint exclusion.)
1521  */
1522  if (subpaths_valid)
1523  add_path(rel, (Path *) create_append_path(root, rel, subpaths, NIL,
1524  NIL, NULL, 0, false,
1525  partitioned_rels, -1));
1526 
1527  /*
1528  * Consider an append of unordered, unparameterized partial paths. Make
1529  * it parallel-aware if possible.
1530  */
1531  if (partial_subpaths_valid && partial_subpaths != NIL)
1532  {
1533  AppendPath *appendpath;
1534  ListCell *lc;
1535  int parallel_workers = 0;
1536 
1537  /* Find the highest number of workers requested for any subpath. */
1538  foreach(lc, partial_subpaths)
1539  {
1540  Path *path = lfirst(lc);
1541 
1542  parallel_workers = Max(parallel_workers, path->parallel_workers);
1543  }
1544  Assert(parallel_workers > 0);
1545 
1546  /*
1547  * If the use of parallel append is permitted, always request at least
1548  * log2(# of children) workers. We assume it can be useful to have
1549  * extra workers in this case because they will be spread out across
1550  * the children. The precise formula is just a guess, but we don't
1551  * want to end up with a radically different answer for a table with N
1552  * partitions vs. an unpartitioned table with the same data, so the
1553  * use of some kind of log-scaling here seems to make some sense.
1554  */
1556  {
1557  parallel_workers = Max(parallel_workers,
1558  fls(list_length(live_childrels)));
1559  parallel_workers = Min(parallel_workers,
1561  }
1562  Assert(parallel_workers > 0);
1563 
1564  /* Generate a partial append path. */
1565  appendpath = create_append_path(root, rel, NIL, partial_subpaths,
1566  NIL, NULL, parallel_workers,
1568  partitioned_rels, -1);
1569 
1570  /*
1571  * Make sure any subsequent partial paths use the same row count
1572  * estimate.
1573  */
1574  partial_rows = appendpath->path.rows;
1575 
1576  /* Add the path. */
1577  add_partial_path(rel, (Path *) appendpath);
1578  }
1579 
1580  /*
1581  * Consider a parallel-aware append using a mix of partial and non-partial
1582  * paths. (This only makes sense if there's at least one child which has
1583  * a non-partial path that is substantially cheaper than any partial path;
1584  * otherwise, we should use the append path added in the previous step.)
1585  */
1586  if (pa_subpaths_valid && pa_nonpartial_subpaths != NIL)
1587  {
1588  AppendPath *appendpath;
1589  ListCell *lc;
1590  int parallel_workers = 0;
1591 
1592  /*
1593  * Find the highest number of workers requested for any partial
1594  * subpath.
1595  */
1596  foreach(lc, pa_partial_subpaths)
1597  {
1598  Path *path = lfirst(lc);
1599 
1600  parallel_workers = Max(parallel_workers, path->parallel_workers);
1601  }
1602 
1603  /*
1604  * Same formula here as above. It's even more important in this
1605  * instance because the non-partial paths won't contribute anything to
1606  * the planned number of parallel workers.
1607  */
1608  parallel_workers = Max(parallel_workers,
1609  fls(list_length(live_childrels)));
1610  parallel_workers = Min(parallel_workers,
1612  Assert(parallel_workers > 0);
1613 
1614  appendpath = create_append_path(root, rel, pa_nonpartial_subpaths,
1615  pa_partial_subpaths,
1616  NIL, NULL, parallel_workers, true,
1617  partitioned_rels, partial_rows);
1618  add_partial_path(rel, (Path *) appendpath);
1619  }
1620 
1621  /*
1622  * Also build unparameterized ordered append paths based on the collected
1623  * list of child pathkeys.
1624  */
1625  if (subpaths_valid)
1626  generate_orderedappend_paths(root, rel, live_childrels,
1627  all_child_pathkeys,
1628  partitioned_rels);
1629 
1630  /*
1631  * Build Append paths for each parameterization seen among the child rels.
1632  * (This may look pretty expensive, but in most cases of practical
1633  * interest, the child rels will expose mostly the same parameterizations,
1634  * so that not that many cases actually get considered here.)
1635  *
1636  * The Append node itself cannot enforce quals, so all qual checking must
1637  * be done in the child paths. This means that to have a parameterized
1638  * Append path, we must have the exact same parameterization for each
1639  * child path; otherwise some children might be failing to check the
1640  * moved-down quals. To make them match up, we can try to increase the
1641  * parameterization of lesser-parameterized paths.
1642  */
1643  foreach(l, all_child_outers)
1644  {
1645  Relids required_outer = (Relids) lfirst(l);
1646  ListCell *lcr;
1647 
1648  /* Select the child paths for an Append with this parameterization */
1649  subpaths = NIL;
1650  subpaths_valid = true;
1651  foreach(lcr, live_childrels)
1652  {
1653  RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1654  Path *subpath;
1655 
1656  if (childrel->pathlist == NIL)
1657  {
1658  /* failed to make a suitable path for this child */
1659  subpaths_valid = false;
1660  break;
1661  }
1662 
1664  childrel,
1665  required_outer);
1666  if (subpath == NULL)
1667  {
1668  /* failed to make a suitable path for this child */
1669  subpaths_valid = false;
1670  break;
1671  }
1672  accumulate_append_subpath(subpath, &subpaths, NULL);
1673  }
1674 
1675  if (subpaths_valid)
1676  add_path(rel, (Path *)
1677  create_append_path(root, rel, subpaths, NIL,
1678  NIL, required_outer, 0, false,
1679  partitioned_rels, -1));
1680  }
1681 
1682  /*
1683  * When there is only a single child relation, the Append path can inherit
1684  * any ordering available for the child rel's path, so that it's useful to
1685  * consider ordered partial paths. Above we only considered the cheapest
1686  * partial path for each child, but let's also make paths using any
1687  * partial paths that have pathkeys.
1688  */
1689  if (list_length(live_childrels) == 1)
1690  {
1691  RelOptInfo *childrel = (RelOptInfo *) linitial(live_childrels);
1692 
1693  foreach(l, childrel->partial_pathlist)
1694  {
1695  Path *path = (Path *) lfirst(l);
1696  AppendPath *appendpath;
1697 
1698  /*
1699  * Skip paths with no pathkeys. Also skip the cheapest partial
1700  * path, since we already used that above.
1701  */
1702  if (path->pathkeys == NIL ||
1703  path == linitial(childrel->partial_pathlist))
1704  continue;
1705 
1706  appendpath = create_append_path(root, rel, NIL, list_make1(path),
1707  NIL, NULL,
1708  path->parallel_workers, true,
1709  partitioned_rels, partial_rows);
1710  add_partial_path(rel, (Path *) appendpath);
1711  }
1712  }
1713 }
1714 
1715 /*
1716  * generate_orderedappend_paths
1717  * Generate ordered append paths for an append relation
1718  *
1719  * Usually we generate MergeAppend paths here, but there are some special
1720  * cases where we can generate simple Append paths, because the subpaths
1721  * can provide tuples in the required order already.
1722  *
1723  * We generate a path for each ordering (pathkey list) appearing in
1724  * all_child_pathkeys.
1725  *
1726  * We consider both cheapest-startup and cheapest-total cases, ie, for each
1727  * interesting ordering, collect all the cheapest startup subpaths and all the
1728  * cheapest total paths, and build a suitable path for each case.
1729  *
1730  * We don't currently generate any parameterized ordered paths here. While
1731  * it would not take much more code here to do so, it's very unclear that it
1732  * is worth the planning cycles to investigate such paths: there's little
1733  * use for an ordered path on the inside of a nestloop. In fact, it's likely
1734  * that the current coding of add_path would reject such paths out of hand,
1735  * because add_path gives no credit for sort ordering of parameterized paths,
1736  * and a parameterized MergeAppend is going to be more expensive than the
1737  * corresponding parameterized Append path. If we ever try harder to support
1738  * parameterized mergejoin plans, it might be worth adding support for
1739  * parameterized paths here to feed such joins. (See notes in
1740  * optimizer/README for why that might not ever happen, though.)
1741  */
1742 static void
1744  List *live_childrels,
1745  List *all_child_pathkeys,
1746  List *partitioned_rels)
1747 {
1748  ListCell *lcp;
1749  List *partition_pathkeys = NIL;
1750  List *partition_pathkeys_desc = NIL;
1751  bool partition_pathkeys_partial = true;
1752  bool partition_pathkeys_desc_partial = true;
1753 
1754  /*
1755  * Some partitioned table setups may allow us to use an Append node
1756  * instead of a MergeAppend. This is possible in cases such as RANGE
1757  * partitioned tables where it's guaranteed that an earlier partition must
1758  * contain rows which come earlier in the sort order. To detect whether
1759  * this is relevant, build pathkey descriptions of the partition ordering,
1760  * for both forward and reverse scans.
1761  */
1762  if (rel->part_scheme != NULL && IS_SIMPLE_REL(rel) &&
1764  {
1765  partition_pathkeys = build_partition_pathkeys(root, rel,
1767  &partition_pathkeys_partial);
1768 
1769  partition_pathkeys_desc = build_partition_pathkeys(root, rel,
1771  &partition_pathkeys_desc_partial);
1772 
1773  /*
1774  * You might think we should truncate_useless_pathkeys here, but
1775  * allowing partition keys which are a subset of the query's pathkeys
1776  * can often be useful. For example, consider a table partitioned by
1777  * RANGE (a, b), and a query with ORDER BY a, b, c. If we have child
1778  * paths that can produce the a, b, c ordering (perhaps via indexes on
1779  * (a, b, c)) then it works to consider the appendrel output as
1780  * ordered by a, b, c.
1781  */
1782  }
1783 
1784  /* Now consider each interesting sort ordering */
1785  foreach(lcp, all_child_pathkeys)
1786  {
1787  List *pathkeys = (List *) lfirst(lcp);
1788  List *startup_subpaths = NIL;
1789  List *total_subpaths = NIL;
1790  bool startup_neq_total = false;
1791  ListCell *lcr;
1792  bool match_partition_order;
1793  bool match_partition_order_desc;
1794 
1795  /*
1796  * Determine if this sort ordering matches any partition pathkeys we
1797  * have, for both ascending and descending partition order. If the
1798  * partition pathkeys happen to be contained in pathkeys then it still
1799  * works, as described above, providing that the partition pathkeys
1800  * are complete and not just a prefix of the partition keys. (In such
1801  * cases we'll be relying on the child paths to have sorted the
1802  * lower-order columns of the required pathkeys.)
1803  */
1804  match_partition_order =
1805  pathkeys_contained_in(pathkeys, partition_pathkeys) ||
1806  (!partition_pathkeys_partial &&
1807  pathkeys_contained_in(partition_pathkeys, pathkeys));
1808 
1809  match_partition_order_desc = !match_partition_order &&
1810  (pathkeys_contained_in(pathkeys, partition_pathkeys_desc) ||
1811  (!partition_pathkeys_desc_partial &&
1812  pathkeys_contained_in(partition_pathkeys_desc, pathkeys)));
1813 
1814  /* Select the child paths for this ordering... */
1815  foreach(lcr, live_childrels)
1816  {
1817  RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1818  Path *cheapest_startup,
1819  *cheapest_total;
1820 
1821  /* Locate the right paths, if they are available. */
1822  cheapest_startup =
1824  pathkeys,
1825  NULL,
1826  STARTUP_COST,
1827  false);
1828  cheapest_total =
1830  pathkeys,
1831  NULL,
1832  TOTAL_COST,
1833  false);
1834 
1835  /*
1836  * If we can't find any paths with the right order just use the
1837  * cheapest-total path; we'll have to sort it later.
1838  */
1839  if (cheapest_startup == NULL || cheapest_total == NULL)
1840  {
1841  cheapest_startup = cheapest_total =
1842  childrel->cheapest_total_path;
1843  /* Assert we do have an unparameterized path for this child */
1844  Assert(cheapest_total->param_info == NULL);
1845  }
1846 
1847  /*
1848  * Notice whether we actually have different paths for the
1849  * "cheapest" and "total" cases; frequently there will be no point
1850  * in two create_merge_append_path() calls.
1851  */
1852  if (cheapest_startup != cheapest_total)
1853  startup_neq_total = true;
1854 
1855  /*
1856  * Collect the appropriate child paths. The required logic varies
1857  * for the Append and MergeAppend cases.
1858  */
1859  if (match_partition_order)
1860  {
1861  /*
1862  * We're going to make a plain Append path. We don't need
1863  * most of what accumulate_append_subpath would do, but we do
1864  * want to cut out child Appends or MergeAppends if they have
1865  * just a single subpath (and hence aren't doing anything
1866  * useful).
1867  */
1868  cheapest_startup = get_singleton_append_subpath(cheapest_startup);
1869  cheapest_total = get_singleton_append_subpath(cheapest_total);
1870 
1871  startup_subpaths = lappend(startup_subpaths, cheapest_startup);
1872  total_subpaths = lappend(total_subpaths, cheapest_total);
1873  }
1874  else if (match_partition_order_desc)
1875  {
1876  /*
1877  * As above, but we need to reverse the order of the children,
1878  * because nodeAppend.c doesn't know anything about reverse
1879  * ordering and will scan the children in the order presented.
1880  */
1881  cheapest_startup = get_singleton_append_subpath(cheapest_startup);
1882  cheapest_total = get_singleton_append_subpath(cheapest_total);
1883 
1884  startup_subpaths = lcons(cheapest_startup, startup_subpaths);
1885  total_subpaths = lcons(cheapest_total, total_subpaths);
1886  }
1887  else
1888  {
1889  /*
1890  * Otherwise, rely on accumulate_append_subpath to collect the
1891  * child paths for the MergeAppend.
1892  */
1893  accumulate_append_subpath(cheapest_startup,
1894  &startup_subpaths, NULL);
1895  accumulate_append_subpath(cheapest_total,
1896  &total_subpaths, NULL);
1897  }
1898  }
1899 
1900  /* ... and build the Append or MergeAppend paths */
1901  if (match_partition_order || match_partition_order_desc)
1902  {
1903  /* We only need Append */
1904  add_path(rel, (Path *) create_append_path(root,
1905  rel,
1906  startup_subpaths,
1907  NIL,
1908  pathkeys,
1909  NULL,
1910  0,
1911  false,
1912  partitioned_rels,
1913  -1));
1914  if (startup_neq_total)
1915  add_path(rel, (Path *) create_append_path(root,
1916  rel,
1917  total_subpaths,
1918  NIL,
1919  pathkeys,
1920  NULL,
1921  0,
1922  false,
1923  partitioned_rels,
1924  -1));
1925  }
1926  else
1927  {
1928  /* We need MergeAppend */
1929  add_path(rel, (Path *) create_merge_append_path(root,
1930  rel,
1931  startup_subpaths,
1932  pathkeys,
1933  NULL,
1934  partitioned_rels));
1935  if (startup_neq_total)
1936  add_path(rel, (Path *) create_merge_append_path(root,
1937  rel,
1938  total_subpaths,
1939  pathkeys,
1940  NULL,
1941  partitioned_rels));
1942  }
1943  }
1944 }
1945 
1946 /*
1947  * get_cheapest_parameterized_child_path
1948  * Get cheapest path for this relation that has exactly the requested
1949  * parameterization.
1950  *
1951  * Returns NULL if unable to create such a path.
1952  */
1953 static Path *
1955  Relids required_outer)
1956 {
1957  Path *cheapest;
1958  ListCell *lc;
1959 
1960  /*
1961  * Look up the cheapest existing path with no more than the needed
1962  * parameterization. If it has exactly the needed parameterization, we're
1963  * done.
1964  */
1965  cheapest = get_cheapest_path_for_pathkeys(rel->pathlist,
1966  NIL,
1967  required_outer,
1968  TOTAL_COST,
1969  false);
1970  Assert(cheapest != NULL);
1971  if (bms_equal(PATH_REQ_OUTER(cheapest), required_outer))
1972  return cheapest;
1973 
1974  /*
1975  * Otherwise, we can "reparameterize" an existing path to match the given
1976  * parameterization, which effectively means pushing down additional
1977  * joinquals to be checked within the path's scan. However, some existing
1978  * paths might check the available joinquals already while others don't;
1979  * therefore, it's not clear which existing path will be cheapest after
1980  * reparameterization. We have to go through them all and find out.
1981  */
1982  cheapest = NULL;
1983  foreach(lc, rel->pathlist)
1984  {
1985  Path *path = (Path *) lfirst(lc);
1986 
1987  /* Can't use it if it needs more than requested parameterization */
1988  if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
1989  continue;
1990 
1991  /*
1992  * Reparameterization can only increase the path's cost, so if it's
1993  * already more expensive than the current cheapest, forget it.
1994  */
1995  if (cheapest != NULL &&
1996  compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
1997  continue;
1998 
1999  /* Reparameterize if needed, then recheck cost */
2000  if (!bms_equal(PATH_REQ_OUTER(path), required_outer))
2001  {
2002  path = reparameterize_path(root, path, required_outer, 1.0);
2003  if (path == NULL)
2004  continue; /* failed to reparameterize this one */
2005  Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));
2006 
2007  if (cheapest != NULL &&
2008  compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
2009  continue;
2010  }
2011 
2012  /* We have a new best path */
2013  cheapest = path;
2014  }
2015 
2016  /* Return the best path, or NULL if we found no suitable candidate */
2017  return cheapest;
2018 }
2019 
2020 /*
2021  * accumulate_append_subpath
2022  * Add a subpath to the list being built for an Append or MergeAppend.
2023  *
2024  * It's possible that the child is itself an Append or MergeAppend path, in
2025  * which case we can "cut out the middleman" and just add its child paths to
2026  * our own list. (We don't try to do this earlier because we need to apply
2027  * both levels of transformation to the quals.)
2028  *
2029  * Note that if we omit a child MergeAppend in this way, we are effectively
2030  * omitting a sort step, which seems fine: if the parent is to be an Append,
2031  * its result would be unsorted anyway, while if the parent is to be a
2032  * MergeAppend, there's no point in a separate sort on a child.
2033  *
2034  * Normally, either path is a partial path and subpaths is a list of partial
2035  * paths, or else path is a non-partial plan and subpaths is a list of those.
2036  * However, if path is a parallel-aware Append, then we add its partial path
2037  * children to subpaths and the rest to special_subpaths. If the latter is
2038  * NULL, we don't flatten the path at all (unless it contains only partial
2039  * paths).
2040  */
2041 static void
2042 accumulate_append_subpath(Path *path, List **subpaths, List **special_subpaths)
2043 {
2044  if (IsA(path, AppendPath))
2045  {
2046  AppendPath *apath = (AppendPath *) path;
2047 
2048  if (!apath->path.parallel_aware || apath->first_partial_path == 0)
2049  {
2050  *subpaths = list_concat(*subpaths, apath->subpaths);
2051  return;
2052  }
2053  else if (special_subpaths != NULL)
2054  {
2055  List *new_special_subpaths;
2056 
2057  /* Split Parallel Append into partial and non-partial subpaths */
2058  *subpaths = list_concat(*subpaths,
2059  list_copy_tail(apath->subpaths,
2060  apath->first_partial_path));
2061  new_special_subpaths =
2063  apath->first_partial_path);
2064  *special_subpaths = list_concat(*special_subpaths,
2065  new_special_subpaths);
2066  return;
2067  }
2068  }
2069  else if (IsA(path, MergeAppendPath))
2070  {
2071  MergeAppendPath *mpath = (MergeAppendPath *) path;
2072 
2073  *subpaths = list_concat(*subpaths, mpath->subpaths);
2074  return;
2075  }
2076 
2077  *subpaths = lappend(*subpaths, path);
2078 }
2079 
2080 /*
2081  * get_singleton_append_subpath
2082  * Returns the single subpath of an Append/MergeAppend, or just
2083  * return 'path' if it's not a single sub-path Append/MergeAppend.
2084  *
2085  * Note: 'path' must not be a parallel-aware path.
2086  */
2087 static Path *
2089 {
2090  Assert(!path->parallel_aware);
2091 
2092  if (IsA(path, AppendPath))
2093  {
2094  AppendPath *apath = (AppendPath *) path;
2095 
2096  if (list_length(apath->subpaths) == 1)
2097  return (Path *) linitial(apath->subpaths);
2098  }
2099  else if (IsA(path, MergeAppendPath))
2100  {
2101  MergeAppendPath *mpath = (MergeAppendPath *) path;
2102 
2103  if (list_length(mpath->subpaths) == 1)
2104  return (Path *) linitial(mpath->subpaths);
2105  }
2106 
2107  return path;
2108 }
2109 
2110 /*
2111  * set_dummy_rel_pathlist
2112  * Build a dummy path for a relation that's been excluded by constraints
2113  *
2114  * Rather than inventing a special "dummy" path type, we represent this as an
2115  * AppendPath with no members (see also IS_DUMMY_APPEND/IS_DUMMY_REL macros).
2116  *
2117  * (See also mark_dummy_rel, which does basically the same thing, but is
2118  * typically used to change a rel into dummy state after we already made
2119  * paths for it.)
2120  */
2121 static void
2123 {
2124  /* Set dummy size estimates --- we leave attr_widths[] as zeroes */
2125  rel->rows = 0;
2126  rel->reltarget->width = 0;
2127 
2128  /* Discard any pre-existing paths; no further need for them */
2129  rel->pathlist = NIL;
2130  rel->partial_pathlist = NIL;
2131 
2132  /* Set up the dummy path */
2133  add_path(rel, (Path *) create_append_path(NULL, rel, NIL, NIL,
2134  NIL, rel->lateral_relids,
2135  0, false, NIL, -1));
2136 
2137  /*
2138  * We set the cheapest-path fields immediately, just in case they were
2139  * pointing at some discarded path. This is redundant when we're called
2140  * from set_rel_size(), but not when called from elsewhere, and doing it
2141  * twice is harmless anyway.
2142  */
2143  set_cheapest(rel);
2144 }
2145 
2146 /* quick-and-dirty test to see if any joining is needed */
2147 static bool
2149 {
2150  int num_base_rels = 0;
2151  Index rti;
2152 
2153  for (rti = 1; rti < root->simple_rel_array_size; rti++)
2154  {
2155  RelOptInfo *brel = root->simple_rel_array[rti];
2156 
2157  if (brel == NULL)
2158  continue;
2159 
2160  /* ignore RTEs that are "other rels" */
2161  if (brel->reloptkind == RELOPT_BASEREL)
2162  if (++num_base_rels > 1)
2163  return true;
2164  }
2165  return false;
2166 }
2167 
2168 /*
2169  * set_subquery_pathlist
2170  * Generate SubqueryScan access paths for a subquery RTE
2171  *
2172  * We don't currently support generating parameterized paths for subqueries
2173  * by pushing join clauses down into them; it seems too expensive to re-plan
2174  * the subquery multiple times to consider different alternatives.
2175  * (XXX that could stand to be reconsidered, now that we use Paths.)
2176  * So the paths made here will be parameterized if the subquery contains
2177  * LATERAL references, otherwise not. As long as that's true, there's no need
2178  * for a separate set_subquery_size phase: just make the paths right away.
2179  */
2180 static void
2182  Index rti, RangeTblEntry *rte)
2183 {
2184  Query *parse = root->parse;
2185  Query *subquery = rte->subquery;
2186  Relids required_outer;
2187  pushdown_safety_info safetyInfo;
2188  double tuple_fraction;
2189  RelOptInfo *sub_final_rel;
2190  ListCell *lc;
2191 
2192  /*
2193  * Must copy the Query so that planning doesn't mess up the RTE contents
2194  * (really really need to fix the planner to not scribble on its input,
2195  * someday ... but see remove_unused_subquery_outputs to start with).
2196  */
2197  subquery = copyObject(subquery);
2198 
2199  /*
2200  * If it's a LATERAL subquery, it might contain some Vars of the current
2201  * query level, requiring it to be treated as parameterized, even though
2202  * we don't support pushing down join quals into subqueries.
2203  */
2204  required_outer = rel->lateral_relids;
2205 
2206  /*
2207  * Zero out result area for subquery_is_pushdown_safe, so that it can set
2208  * flags as needed while recursing. In particular, we need a workspace
2209  * for keeping track of unsafe-to-reference columns. unsafeColumns[i]
2210  * will be set true if we find that output column i of the subquery is
2211  * unsafe to use in a pushed-down qual.
2212  */
2213  memset(&safetyInfo, 0, sizeof(safetyInfo));
2214  safetyInfo.unsafeColumns = (bool *)
2215  palloc0((list_length(subquery->targetList) + 1) * sizeof(bool));
2216 
2217  /*
2218  * If the subquery has the "security_barrier" flag, it means the subquery
2219  * originated from a view that must enforce row level security. Then we
2220  * must not push down quals that contain leaky functions. (Ideally this
2221  * would be checked inside subquery_is_pushdown_safe, but since we don't
2222  * currently pass the RTE to that function, we must do it here.)
2223  */
2224  safetyInfo.unsafeLeaky = rte->security_barrier;
2225 
2226  /*
2227  * If there are any restriction clauses that have been attached to the
2228  * subquery relation, consider pushing them down to become WHERE or HAVING
2229  * quals of the subquery itself. This transformation is useful because it
2230  * may allow us to generate a better plan for the subquery than evaluating
2231  * all the subquery output rows and then filtering them.
2232  *
2233  * There are several cases where we cannot push down clauses. Restrictions
2234  * involving the subquery are checked by subquery_is_pushdown_safe().
2235  * Restrictions on individual clauses are checked by
2236  * qual_is_pushdown_safe(). Also, we don't want to push down
2237  * pseudoconstant clauses; better to have the gating node above the
2238  * subquery.
2239  *
2240  * Non-pushed-down clauses will get evaluated as qpquals of the
2241  * SubqueryScan node.
2242  *
2243  * XXX Are there any cases where we want to make a policy decision not to
2244  * push down a pushable qual, because it'd result in a worse plan?
2245  */
2246  if (rel->baserestrictinfo != NIL &&
2247  subquery_is_pushdown_safe(subquery, subquery, &safetyInfo))
2248  {
2249  /* OK to consider pushing down individual quals */
2250  List *upperrestrictlist = NIL;
2251  ListCell *l;
2252 
2253  foreach(l, rel->baserestrictinfo)
2254  {
2255  RestrictInfo *rinfo = (RestrictInfo *) lfirst(l);
2256  Node *clause = (Node *) rinfo->clause;
2257 
2258  if (!rinfo->pseudoconstant &&
2259  qual_is_pushdown_safe(subquery, rti, clause, &safetyInfo))
2260  {
2261  /* Push it down */
2262  subquery_push_qual(subquery, rte, rti, clause);
2263  }
2264  else
2265  {
2266  /* Keep it in the upper query */
2267  upperrestrictlist = lappend(upperrestrictlist, rinfo);
2268  }
2269  }
2270  rel->baserestrictinfo = upperrestrictlist;
2271  /* We don't bother recomputing baserestrict_min_security */
2272  }
2273 
2274  pfree(safetyInfo.unsafeColumns);
2275 
2276  /*
2277  * The upper query might not use all the subquery's output columns; if
2278  * not, we can simplify.
2279  */
2280  remove_unused_subquery_outputs(subquery, rel);
2281 
2282  /*
2283  * We can safely pass the outer tuple_fraction down to the subquery if the
2284  * outer level has no joining, aggregation, or sorting to do. Otherwise
2285  * we'd better tell the subquery to plan for full retrieval. (XXX This
2286  * could probably be made more intelligent ...)
2287  */
2288  if (parse->hasAggs ||
2289  parse->groupClause ||
2290  parse->groupingSets ||
2291  parse->havingQual ||
2292  parse->distinctClause ||
2293  parse->sortClause ||
2294  has_multiple_baserels(root))
2295  tuple_fraction = 0.0; /* default case */
2296  else
2297  tuple_fraction = root->tuple_fraction;
2298 
2299  /* plan_params should not be in use in current query level */
2300  Assert(root->plan_params == NIL);
2301 
2302  /* Generate a subroot and Paths for the subquery */
2303  rel->subroot = subquery_planner(root->glob, subquery,
2304  root,
2305  false, tuple_fraction);
2306 
2307  /* Isolate the params needed by this specific subplan */
2308  rel->subplan_params = root->plan_params;
2309  root->plan_params = NIL;
2310 
2311  /*
2312  * It's possible that constraint exclusion proved the subquery empty. If
2313  * so, it's desirable to produce an unadorned dummy path so that we will
2314  * recognize appropriate optimizations at this query level.
2315  */
2316  sub_final_rel = fetch_upper_rel(rel->subroot, UPPERREL_FINAL, NULL);
2317 
2318  if (IS_DUMMY_REL(sub_final_rel))
2319  {
2321  return;
2322  }
2323 
2324  /*
2325  * Mark rel with estimated output rows, width, etc. Note that we have to
2326  * do this before generating outer-query paths, else cost_subqueryscan is
2327  * not happy.
2328  */
2329  set_subquery_size_estimates(root, rel);
2330 
2331  /*
2332  * For each Path that subquery_planner produced, make a SubqueryScanPath
2333  * in the outer query.
2334  */
2335  foreach(lc, sub_final_rel->pathlist)
2336  {
2337  Path *subpath = (Path *) lfirst(lc);
2338  List *pathkeys;
2339 
2340  /* Convert subpath's pathkeys to outer representation */
2341  pathkeys = convert_subquery_pathkeys(root,
2342  rel,
2343  subpath->pathkeys,
2345 
2346  /* Generate outer path using this subpath */
2347  add_path(rel, (Path *)
2348  create_subqueryscan_path(root, rel, subpath,
2349  pathkeys, required_outer));
2350  }
2351 
2352  /* If outer rel allows parallelism, do same for partial paths. */
2353  if (rel->consider_parallel && bms_is_empty(required_outer))
2354  {
2355  /* If consider_parallel is false, there should be no partial paths. */
2356  Assert(sub_final_rel->consider_parallel ||
2357  sub_final_rel->partial_pathlist == NIL);
2358 
2359  /* Same for partial paths. */
2360  foreach(lc, sub_final_rel->partial_pathlist)
2361  {
2362  Path *subpath = (Path *) lfirst(lc);
2363  List *pathkeys;
2364 
2365  /* Convert subpath's pathkeys to outer representation */
2366  pathkeys = convert_subquery_pathkeys(root,
2367  rel,
2368  subpath->pathkeys,
2370 
2371  /* Generate outer path using this subpath */
2372  add_partial_path(rel, (Path *)
2373  create_subqueryscan_path(root, rel, subpath,
2374  pathkeys,
2375  required_outer));
2376  }
2377  }
2378 }
2379 
2380 /*
2381  * set_function_pathlist
2382  * Build the (single) access path for a function RTE
2383  */
2384 static void
2386 {
2387  Relids required_outer;
2388  List *pathkeys = NIL;
2389 
2390  /*
2391  * We don't support pushing join clauses into the quals of a function
2392  * scan, but it could still have required parameterization due to LATERAL
2393  * refs in the function expression.
2394  */
2395  required_outer = rel->lateral_relids;
2396 
2397  /*
2398  * The result is considered unordered unless ORDINALITY was used, in which
2399  * case it is ordered by the ordinal column (the last one). See if we
2400  * care, by checking for uses of that Var in equivalence classes.
2401  */
2402  if (rte->funcordinality)
2403  {
2404  AttrNumber ordattno = rel->max_attr;
2405  Var *var = NULL;
2406  ListCell *lc;
2407 
2408  /*
2409  * Is there a Var for it in rel's targetlist? If not, the query did
2410  * not reference the ordinality column, or at least not in any way
2411  * that would be interesting for sorting.
2412  */
2413  foreach(lc, rel->reltarget->exprs)
2414  {
2415  Var *node = (Var *) lfirst(lc);
2416 
2417  /* checking varno/varlevelsup is just paranoia */
2418  if (IsA(node, Var) &&
2419  node->varattno == ordattno &&
2420  node->varno == rel->relid &&
2421  node->varlevelsup == 0)
2422  {
2423  var = node;
2424  break;
2425  }
2426  }
2427 
2428  /*
2429  * Try to build pathkeys for this Var with int8 sorting. We tell
2430  * build_expression_pathkey not to build any new equivalence class; if
2431  * the Var isn't already mentioned in some EC, it means that nothing
2432  * cares about the ordering.
2433  */
2434  if (var)
2435  pathkeys = build_expression_pathkey(root,
2436  (Expr *) var,
2437  NULL, /* below outer joins */
2438  Int8LessOperator,
2439  rel->relids,
2440  false);
2441  }
2442 
2443  /* Generate appropriate path */
2444  add_path(rel, create_functionscan_path(root, rel,
2445  pathkeys, required_outer));
2446 }
2447 
2448 /*
2449  * set_values_pathlist
2450  * Build the (single) access path for a VALUES RTE
2451  */
2452 static void
2454 {
2455  Relids required_outer;
2456 
2457  /*
2458  * We don't support pushing join clauses into the quals of a values scan,
2459  * but it could still have required parameterization due to LATERAL refs
2460  * in the values expressions.
2461  */
2462  required_outer = rel->lateral_relids;
2463 
2464  /* Generate appropriate path */
2465  add_path(rel, create_valuesscan_path(root, rel, required_outer));
2466 }
2467 
2468 /*
2469  * set_tablefunc_pathlist
2470  * Build the (single) access path for a table func RTE
2471  */
2472 static void
2474 {
2475  Relids required_outer;
2476 
2477  /*
2478  * We don't support pushing join clauses into the quals of a tablefunc
2479  * scan, but it could still have required parameterization due to LATERAL
2480  * refs in the function expression.
2481  */
2482  required_outer = rel->lateral_relids;
2483 
2484  /* Generate appropriate path */
2485  add_path(rel, create_tablefuncscan_path(root, rel,
2486  required_outer));
2487 }
2488 
2489 /*
2490  * set_cte_pathlist
2491  * Build the (single) access path for a non-self-reference CTE RTE
2492  *
2493  * There's no need for a separate set_cte_size phase, since we don't
2494  * support join-qual-parameterized paths for CTEs.
2495  */
2496 static void
2498 {
2499  Plan *cteplan;
2500  PlannerInfo *cteroot;
2501  Index levelsup;
2502  int ndx;
2503  ListCell *lc;
2504  int plan_id;
2505  Relids required_outer;
2506 
2507  /*
2508  * Find the referenced CTE, and locate the plan previously made for it.
2509  */
2510  levelsup = rte->ctelevelsup;
2511  cteroot = root;
2512  while (levelsup-- > 0)
2513  {
2514  cteroot = cteroot->parent_root;
2515  if (!cteroot) /* shouldn't happen */
2516  elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2517  }
2518 
2519  /*
2520  * Note: cte_plan_ids can be shorter than cteList, if we are still working
2521  * on planning the CTEs (ie, this is a side-reference from another CTE).
2522  * So we mustn't use forboth here.
2523  */
2524  ndx = 0;
2525  foreach(lc, cteroot->parse->cteList)
2526  {
2527  CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
2528 
2529  if (strcmp(cte->ctename, rte->ctename) == 0)
2530  break;
2531  ndx++;
2532  }
2533  if (lc == NULL) /* shouldn't happen */
2534  elog(ERROR, "could not find CTE \"%s\"", rte->ctename);
2535  if (ndx >= list_length(cteroot->cte_plan_ids))
2536  elog(ERROR, "could not find plan for CTE \"%s\"", rte->ctename);
2537  plan_id = list_nth_int(cteroot->cte_plan_ids, ndx);
2538  Assert(plan_id > 0);
2539  cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2540 
2541  /* Mark rel with estimated output rows, width, etc */
2542  set_cte_size_estimates(root, rel, cteplan->plan_rows);
2543 
2544  /*
2545  * We don't support pushing join clauses into the quals of a CTE scan, but
2546  * it could still have required parameterization due to LATERAL refs in
2547  * its tlist.
2548  */
2549  required_outer = rel->lateral_relids;
2550 
2551  /* Generate appropriate path */
2552  add_path(rel, create_ctescan_path(root, rel, required_outer));
2553 }
2554 
2555 /*
2556  * set_namedtuplestore_pathlist
2557  * Build the (single) access path for a named tuplestore RTE
2558  *
2559  * There's no need for a separate set_namedtuplestore_size phase, since we
2560  * don't support join-qual-parameterized paths for tuplestores.
2561  */
2562 static void
2564  RangeTblEntry *rte)
2565 {
2566  Relids required_outer;
2567 
2568  /* Mark rel with estimated output rows, width, etc */
2570 
2571  /*
2572  * We don't support pushing join clauses into the quals of a tuplestore
2573  * scan, but it could still have required parameterization due to LATERAL
2574  * refs in its tlist.
2575  */
2576  required_outer = rel->lateral_relids;
2577 
2578  /* Generate appropriate path */
2579  add_path(rel, create_namedtuplestorescan_path(root, rel, required_outer));
2580 
2581  /* Select cheapest path (pretty easy in this case...) */
2582  set_cheapest(rel);
2583 }
2584 
2585 /*
2586  * set_result_pathlist
2587  * Build the (single) access path for an RTE_RESULT RTE
2588  *
2589  * There's no need for a separate set_result_size phase, since we
2590  * don't support join-qual-parameterized paths for these RTEs.
2591  */
2592 static void
2594  RangeTblEntry *rte)
2595 {
2596  Relids required_outer;
2597 
2598  /* Mark rel with estimated output rows, width, etc */
2599  set_result_size_estimates(root, rel);
2600 
2601  /*
2602  * We don't support pushing join clauses into the quals of a Result scan,
2603  * but it could still have required parameterization due to LATERAL refs
2604  * in its tlist.
2605  */
2606  required_outer = rel->lateral_relids;
2607 
2608  /* Generate appropriate path */
2609  add_path(rel, create_resultscan_path(root, rel, required_outer));
2610 
2611  /* Select cheapest path (pretty easy in this case...) */
2612  set_cheapest(rel);
2613 }
2614 
2615 /*
2616  * set_worktable_pathlist
2617  * Build the (single) access path for a self-reference CTE RTE
2618  *
2619  * There's no need for a separate set_worktable_size phase, since we don't
2620  * support join-qual-parameterized paths for CTEs.
2621  */
2622 static void
2624 {
2625  Path *ctepath;
2626  PlannerInfo *cteroot;
2627  Index levelsup;
2628  Relids required_outer;
2629 
2630  /*
2631  * We need to find the non-recursive term's path, which is in the plan
2632  * level that's processing the recursive UNION, which is one level *below*
2633  * where the CTE comes from.
2634  */
2635  levelsup = rte->ctelevelsup;
2636  if (levelsup == 0) /* shouldn't happen */
2637  elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2638  levelsup--;
2639  cteroot = root;
2640  while (levelsup-- > 0)
2641  {
2642  cteroot = cteroot->parent_root;
2643  if (!cteroot) /* shouldn't happen */
2644  elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2645  }
2646  ctepath = cteroot->non_recursive_path;
2647  if (!ctepath) /* shouldn't happen */
2648  elog(ERROR, "could not find path for CTE \"%s\"", rte->ctename);
2649 
2650  /* Mark rel with estimated output rows, width, etc */
2651  set_cte_size_estimates(root, rel, ctepath->rows);
2652 
2653  /*
2654  * We don't support pushing join clauses into the quals of a worktable
2655  * scan, but it could still have required parameterization due to LATERAL
2656  * refs in its tlist. (I'm not sure this is actually possible given the
2657  * restrictions on recursive references, but it's easy enough to support.)
2658  */
2659  required_outer = rel->lateral_relids;
2660 
2661  /* Generate appropriate path */
2662  add_path(rel, create_worktablescan_path(root, rel, required_outer));
2663 }
2664 
2665 /*
2666  * generate_gather_paths
2667  * Generate parallel access paths for a relation by pushing a Gather or
2668  * Gather Merge on top of a partial path.
2669  *
2670  * This must not be called until after we're done creating all partial paths
2671  * for the specified relation. (Otherwise, add_partial_path might delete a
2672  * path that some GatherPath or GatherMergePath has a reference to.)
2673  *
2674  * If we're generating paths for a scan or join relation, override_rows will
2675  * be false, and we'll just use the relation's size estimate. When we're
2676  * being called for a partially-grouped path, though, we need to override
2677  * the rowcount estimate. (It's not clear that the particular value we're
2678  * using here is actually best, but the underlying rel has no estimate so
2679  * we must do something.)
2680  */
2681 void
2682 generate_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
2683 {
2684  Path *cheapest_partial_path;
2685  Path *simple_gather_path;
2686  ListCell *lc;
2687  double rows;
2688  double *rowsp = NULL;
2689 
2690  /* If there are no partial paths, there's nothing to do here. */
2691  if (rel->partial_pathlist == NIL)
2692  return;
2693 
2694  /* Should we override the rel's rowcount estimate? */
2695  if (override_rows)
2696  rowsp = &rows;
2697 
2698  /*
2699  * The output of Gather is always unsorted, so there's only one partial
2700  * path of interest: the cheapest one. That will be the one at the front
2701  * of partial_pathlist because of the way add_partial_path works.
2702  */
2703  cheapest_partial_path = linitial(rel->partial_pathlist);
2704  rows =
2705  cheapest_partial_path->rows * cheapest_partial_path->parallel_workers;
2706  simple_gather_path = (Path *)
2707  create_gather_path(root, rel, cheapest_partial_path, rel->reltarget,
2708  NULL, rowsp);
2709  add_path(rel, simple_gather_path);
2710 
2711  /*
2712  * For each useful ordering, we can consider an order-preserving Gather
2713  * Merge.
2714  */
2715  foreach(lc, rel->partial_pathlist)
2716  {
2717  Path *subpath = (Path *) lfirst(lc);
2718  GatherMergePath *path;
2719 
2720  if (subpath->pathkeys == NIL)
2721  continue;
2722 
2723  rows = subpath->rows * subpath->parallel_workers;
2724  path = create_gather_merge_path(root, rel, subpath, rel->reltarget,
2725  subpath->pathkeys, NULL, rowsp);
2726  add_path(rel, &path->path);
2727  }
2728 }
2729 
2730 /*
2731  * make_rel_from_joinlist
2732  * Build access paths using a "joinlist" to guide the join path search.
2733  *
2734  * See comments for deconstruct_jointree() for definition of the joinlist
2735  * data structure.
2736  */
2737 static RelOptInfo *
2739 {
2740  int levels_needed;
2741  List *initial_rels;
2742  ListCell *jl;
2743 
2744  /*
2745  * Count the number of child joinlist nodes. This is the depth of the
2746  * dynamic-programming algorithm we must employ to consider all ways of
2747  * joining the child nodes.
2748  */
2749  levels_needed = list_length(joinlist);
2750 
2751  if (levels_needed <= 0)
2752  return NULL; /* nothing to do? */
2753 
2754  /*
2755  * Construct a list of rels corresponding to the child joinlist nodes.
2756  * This may contain both base rels and rels constructed according to
2757  * sub-joinlists.
2758  */
2759  initial_rels = NIL;
2760  foreach(jl, joinlist)
2761  {
2762  Node *jlnode = (Node *) lfirst(jl);
2763  RelOptInfo *thisrel;
2764 
2765  if (IsA(jlnode, RangeTblRef))
2766  {
2767  int varno = ((RangeTblRef *) jlnode)->rtindex;
2768 
2769  thisrel = find_base_rel(root, varno);
2770  }
2771  else if (IsA(jlnode, List))
2772  {
2773  /* Recurse to handle subproblem */
2774  thisrel = make_rel_from_joinlist(root, (List *) jlnode);
2775  }
2776  else
2777  {
2778  elog(ERROR, "unrecognized joinlist node type: %d",
2779  (int) nodeTag(jlnode));
2780  thisrel = NULL; /* keep compiler quiet */
2781  }
2782 
2783  initial_rels = lappend(initial_rels, thisrel);
2784  }
2785 
2786  if (levels_needed == 1)
2787  {
2788  /*
2789  * Single joinlist node, so we're done.
2790  */
2791  return (RelOptInfo *) linitial(initial_rels);
2792  }
2793  else
2794  {
2795  /*
2796  * Consider the different orders in which we could join the rels,
2797  * using a plugin, GEQO, or the regular join search code.
2798  *
2799  * We put the initial_rels list into a PlannerInfo field because
2800  * has_legal_joinclause() needs to look at it (ugly :-().
2801  */
2802  root->initial_rels = initial_rels;
2803 
2804  if (join_search_hook)
2805  return (*join_search_hook) (root, levels_needed, initial_rels);
2806  else if (enable_geqo && levels_needed >= geqo_threshold)
2807  return geqo(root, levels_needed, initial_rels);
2808  else
2809  return standard_join_search(root, levels_needed, initial_rels);
2810  }
2811 }
2812 
2813 /*
2814  * standard_join_search
2815  * Find possible joinpaths for a query by successively finding ways
2816  * to join component relations into join relations.
2817  *
2818  * 'levels_needed' is the number of iterations needed, ie, the number of
2819  * independent jointree items in the query. This is > 1.
2820  *
2821  * 'initial_rels' is a list of RelOptInfo nodes for each independent
2822  * jointree item. These are the components to be joined together.
2823  * Note that levels_needed == list_length(initial_rels).
2824  *
2825  * Returns the final level of join relations, i.e., the relation that is
2826  * the result of joining all the original relations together.
2827  * At least one implementation path must be provided for this relation and
2828  * all required sub-relations.
2829  *
2830  * To support loadable plugins that modify planner behavior by changing the
2831  * join searching algorithm, we provide a hook variable that lets a plugin
2832  * replace or supplement this function. Any such hook must return the same
2833  * final join relation as the standard code would, but it might have a
2834  * different set of implementation paths attached, and only the sub-joinrels
2835  * needed for these paths need have been instantiated.
2836  *
2837  * Note to plugin authors: the functions invoked during standard_join_search()
2838  * modify root->join_rel_list and root->join_rel_hash. If you want to do more
2839  * than one join-order search, you'll probably need to save and restore the
2840  * original states of those data structures. See geqo_eval() for an example.
2841  */
2842 RelOptInfo *
2843 standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
2844 {
2845  int lev;
2846  RelOptInfo *rel;
2847 
2848  /*
2849  * This function cannot be invoked recursively within any one planning
2850  * problem, so join_rel_level[] can't be in use already.
2851  */
2852  Assert(root->join_rel_level == NULL);
2853 
2854  /*
2855  * We employ a simple "dynamic programming" algorithm: we first find all
2856  * ways to build joins of two jointree items, then all ways to build joins
2857  * of three items (from two-item joins and single items), then four-item
2858  * joins, and so on until we have considered all ways to join all the
2859  * items into one rel.
2860  *
2861  * root->join_rel_level[j] is a list of all the j-item rels. Initially we
2862  * set root->join_rel_level[1] to represent all the single-jointree-item
2863  * relations.
2864  */
2865  root->join_rel_level = (List **) palloc0((levels_needed + 1) * sizeof(List *));
2866 
2867  root->join_rel_level[1] = initial_rels;
2868 
2869  for (lev = 2; lev <= levels_needed; lev++)
2870  {
2871  ListCell *lc;
2872 
2873  /*
2874  * Determine all possible pairs of relations to be joined at this
2875  * level, and build paths for making each one from every available
2876  * pair of lower-level relations.
2877  */
2878  join_search_one_level(root, lev);
2879 
2880  /*
2881  * Run generate_partitionwise_join_paths() and generate_gather_paths()
2882  * for each just-processed joinrel. We could not do this earlier
2883  * because both regular and partial paths can get added to a
2884  * particular joinrel at multiple times within join_search_one_level.
2885  *
2886  * After that, we're done creating paths for the joinrel, so run
2887  * set_cheapest().
2888  */
2889  foreach(lc, root->join_rel_level[lev])
2890  {
2891  rel = (RelOptInfo *) lfirst(lc);
2892 
2893  /* Create paths for partitionwise joins. */
2895 
2896  /*
2897  * Except for the topmost scan/join rel, consider gathering
2898  * partial paths. We'll do the same for the topmost scan/join rel
2899  * once we know the final targetlist (see grouping_planner).
2900  */
2901  if (lev < levels_needed)
2902  generate_gather_paths(root, rel, false);
2903 
2904  /* Find and save the cheapest paths for this rel */
2905  set_cheapest(rel);
2906 
2907 #ifdef OPTIMIZER_DEBUG
2908  debug_print_rel(root, rel);
2909 #endif
2910  }
2911  }
2912 
2913  /*
2914  * We should have a single rel at the final level.
2915  */
2916  if (root->join_rel_level[levels_needed] == NIL)
2917  elog(ERROR, "failed to build any %d-way joins", levels_needed);
2918  Assert(list_length(root->join_rel_level[levels_needed]) == 1);
2919 
2920  rel = (RelOptInfo *) linitial(root->join_rel_level[levels_needed]);
2921 
2922  root->join_rel_level = NULL;
2923 
2924  return rel;
2925 }
2926 
2927 /*****************************************************************************
2928  * PUSHING QUALS DOWN INTO SUBQUERIES
2929  *****************************************************************************/
2930 
2931 /*
2932  * subquery_is_pushdown_safe - is a subquery safe for pushing down quals?
2933  *
2934  * subquery is the particular component query being checked. topquery
2935  * is the top component of a set-operations tree (the same Query if no
2936  * set-op is involved).
2937  *
2938  * Conditions checked here:
2939  *
2940  * 1. If the subquery has a LIMIT clause, we must not push down any quals,
2941  * since that could change the set of rows returned.
2942  *
2943  * 2. If the subquery contains EXCEPT or EXCEPT ALL set ops we cannot push
2944  * quals into it, because that could change the results.
2945  *
2946  * 3. If the subquery uses DISTINCT, we cannot push volatile quals into it.
2947  * This is because upper-level quals should semantically be evaluated only
2948  * once per distinct row, not once per original row, and if the qual is
2949  * volatile then extra evaluations could change the results. (This issue
2950  * does not apply to other forms of aggregation such as GROUP BY, because
2951  * when those are present we push into HAVING not WHERE, so that the quals
2952  * are still applied after aggregation.)
2953  *
2954  * 4. If the subquery contains window functions, we cannot push volatile quals
2955  * into it. The issue here is a bit different from DISTINCT: a volatile qual
2956  * might succeed for some rows of a window partition and fail for others,
2957  * thereby changing the partition contents and thus the window functions'
2958  * results for rows that remain.
2959  *
2960  * 5. If the subquery contains any set-returning functions in its targetlist,
2961  * we cannot push volatile quals into it. That would push them below the SRFs
2962  * and thereby change the number of times they are evaluated. Also, a
2963  * volatile qual could succeed for some SRF output rows and fail for others,
2964  * a behavior that cannot occur if it's evaluated before SRF expansion.
2965  *
2966  * In addition, we make several checks on the subquery's output columns to see
2967  * if it is safe to reference them in pushed-down quals. If output column k
2968  * is found to be unsafe to reference, we set safetyInfo->unsafeColumns[k]
2969  * to true, but we don't reject the subquery overall since column k might not
2970  * be referenced by some/all quals. The unsafeColumns[] array will be
2971  * consulted later by qual_is_pushdown_safe(). It's better to do it this way
2972  * than to make the checks directly in qual_is_pushdown_safe(), because when
2973  * the subquery involves set operations we have to check the output
2974  * expressions in each arm of the set op.
2975  *
2976  * Note: pushing quals into a DISTINCT subquery is theoretically dubious:
2977  * we're effectively assuming that the quals cannot distinguish values that
2978  * the DISTINCT's equality operator sees as equal, yet there are many
2979  * counterexamples to that assumption. However use of such a qual with a
2980  * DISTINCT subquery would be unsafe anyway, since there's no guarantee which
2981  * "equal" value will be chosen as the output value by the DISTINCT operation.
2982  * So we don't worry too much about that. Another objection is that if the
2983  * qual is expensive to evaluate, running it for each original row might cost
2984  * more than we save by eliminating rows before the DISTINCT step. But it
2985  * would be very hard to estimate that at this stage, and in practice pushdown
2986  * seldom seems to make things worse, so we ignore that problem too.
2987  *
2988  * Note: likewise, pushing quals into a subquery with window functions is a
2989  * bit dubious: the quals might remove some rows of a window partition while
2990  * leaving others, causing changes in the window functions' results for the
2991  * surviving rows. We insist that such a qual reference only partitioning
2992  * columns, but again that only protects us if the qual does not distinguish
2993  * values that the partitioning equality operator sees as equal. The risks
2994  * here are perhaps larger than for DISTINCT, since no de-duplication of rows
2995  * occurs and thus there is no theoretical problem with such a qual. But
2996  * we'll do this anyway because the potential performance benefits are very
2997  * large, and we've seen no field complaints about the longstanding comparable
2998  * behavior with DISTINCT.
2999  */
3000 static bool
3002  pushdown_safety_info *safetyInfo)
3003 {
3004  SetOperationStmt *topop;
3005 
3006  /* Check point 1 */
3007  if (subquery->limitOffset != NULL || subquery->limitCount != NULL)
3008  return false;
3009 
3010  /* Check points 3, 4, and 5 */
3011  if (subquery->distinctClause ||
3012  subquery->hasWindowFuncs ||
3013  subquery->hasTargetSRFs)
3014  safetyInfo->unsafeVolatile = true;
3015 
3016  /*
3017  * If we're at a leaf query, check for unsafe expressions in its target
3018  * list, and mark any unsafe ones in unsafeColumns[]. (Non-leaf nodes in
3019  * setop trees have only simple Vars in their tlists, so no need to check
3020  * them.)
3021  */
3022  if (subquery->setOperations == NULL)
3023  check_output_expressions(subquery, safetyInfo);
3024 
3025  /* Are we at top level, or looking at a setop component? */
3026  if (subquery == topquery)
3027  {
3028  /* Top level, so check any component queries */
3029  if (subquery->setOperations != NULL)
3030  if (!recurse_pushdown_safe(subquery->setOperations, topquery,
3031  safetyInfo))
3032  return false;
3033  }
3034  else
3035  {
3036  /* Setop component must not have more components (too weird) */
3037  if (subquery->setOperations != NULL)
3038  return false;
3039  /* Check whether setop component output types match top level */
3040  topop = castNode(SetOperationStmt, topquery->setOperations);
3041  Assert(topop);
3043  topop->colTypes,
3044  safetyInfo);
3045  }
3046  return true;
3047 }
3048 
3049 /*
3050  * Helper routine to recurse through setOperations tree
3051  */
3052 static bool
3054  pushdown_safety_info *safetyInfo)
3055 {
3056  if (IsA(setOp, RangeTblRef))
3057  {
3058  RangeTblRef *rtr = (RangeTblRef *) setOp;
3059  RangeTblEntry *rte = rt_fetch(rtr->rtindex, topquery->rtable);
3060  Query *subquery = rte->subquery;
3061 
3062  Assert(subquery != NULL);
3063  return subquery_is_pushdown_safe(subquery, topquery, safetyInfo);
3064  }
3065  else if (IsA(setOp, SetOperationStmt))
3066  {
3067  SetOperationStmt *op = (SetOperationStmt *) setOp;
3068 
3069  /* EXCEPT is no good (point 2 for subquery_is_pushdown_safe) */
3070  if (op->op == SETOP_EXCEPT)
3071  return false;
3072  /* Else recurse */
3073  if (!recurse_pushdown_safe(op->larg, topquery, safetyInfo))
3074  return false;
3075  if (!recurse_pushdown_safe(op->rarg, topquery, safetyInfo))
3076  return false;
3077  }
3078  else
3079  {
3080  elog(ERROR, "unrecognized node type: %d",
3081  (int) nodeTag(setOp));
3082  }
3083  return true;
3084 }
3085 
3086 /*
3087  * check_output_expressions - check subquery's output expressions for safety
3088  *
3089  * There are several cases in which it's unsafe to push down an upper-level
3090  * qual if it references a particular output column of a subquery. We check
3091  * each output column of the subquery and set unsafeColumns[k] to true if
3092  * that column is unsafe for a pushed-down qual to reference. The conditions
3093  * checked here are:
3094  *
3095  * 1. We must not push down any quals that refer to subselect outputs that
3096  * return sets, else we'd introduce functions-returning-sets into the
3097  * subquery's WHERE/HAVING quals.
3098  *
3099  * 2. We must not push down any quals that refer to subselect outputs that
3100  * contain volatile functions, for fear of introducing strange results due
3101  * to multiple evaluation of a volatile function.
3102  *
3103  * 3. If the subquery uses DISTINCT ON, we must not push down any quals that
3104  * refer to non-DISTINCT output columns, because that could change the set
3105  * of rows returned. (This condition is vacuous for DISTINCT, because then
3106  * there are no non-DISTINCT output columns, so we needn't check. Note that
3107  * subquery_is_pushdown_safe already reported that we can't use volatile
3108  * quals if there's DISTINCT or DISTINCT ON.)
3109  *
3110  * 4. If the subquery has any window functions, we must not push down quals
3111  * that reference any output columns that are not listed in all the subquery's
3112  * window PARTITION BY clauses. We can push down quals that use only
3113  * partitioning columns because they should succeed or fail identically for
3114  * every row of any one window partition, and totally excluding some
3115  * partitions will not change a window function's results for remaining
3116  * partitions. (Again, this also requires nonvolatile quals, but
3117  * subquery_is_pushdown_safe handles that.)
3118  */
3119 static void
3121 {
3122  ListCell *lc;
3123 
3124  foreach(lc, subquery->targetList)
3125  {
3126  TargetEntry *tle = (TargetEntry *) lfirst(lc);
3127 
3128  if (tle->resjunk)
3129  continue; /* ignore resjunk columns */
3130 
3131  /* We need not check further if output col is already known unsafe */
3132  if (safetyInfo->unsafeColumns[tle->resno])
3133  continue;
3134 
3135  /* Functions returning sets are unsafe (point 1) */
3136  if (subquery->hasTargetSRFs &&
3137  expression_returns_set((Node *) tle->expr))
3138  {
3139  safetyInfo->unsafeColumns[tle->resno] = true;
3140  continue;
3141  }
3142 
3143  /* Volatile functions are unsafe (point 2) */
3144  if (contain_volatile_functions((Node *) tle->expr))
3145  {
3146  safetyInfo->unsafeColumns[tle->resno] = true;
3147  continue;
3148  }
3149 
3150  /* If subquery uses DISTINCT ON, check point 3 */
3151  if (subquery->hasDistinctOn &&
3152  !targetIsInSortList(tle, InvalidOid, subquery->distinctClause))
3153  {
3154  /* non-DISTINCT column, so mark it unsafe */
3155  safetyInfo->unsafeColumns[tle->resno] = true;
3156  continue;
3157  }
3158 
3159  /* If subquery uses window functions, check point 4 */
3160  if (subquery->hasWindowFuncs &&
3161  !targetIsInAllPartitionLists(tle, subquery))
3162  {
3163  /* not present in all PARTITION BY clauses, so mark it unsafe */
3164  safetyInfo->unsafeColumns[tle->resno] = true;
3165  continue;
3166  }
3167  }
3168 }
3169 
3170 /*
3171  * For subqueries using UNION/UNION ALL/INTERSECT/INTERSECT ALL, we can
3172  * push quals into each component query, but the quals can only reference
3173  * subquery columns that suffer no type coercions in the set operation.
3174  * Otherwise there are possible semantic gotchas. So, we check the
3175  * component queries to see if any of them have output types different from
3176  * the top-level setop outputs. unsafeColumns[k] is set true if column k
3177  * has different type in any component.
3178  *
3179  * We don't have to care about typmods here: the only allowed difference
3180  * between set-op input and output typmods is input is a specific typmod
3181  * and output is -1, and that does not require a coercion.
3182  *
3183  * tlist is a subquery tlist.
3184  * colTypes is an OID list of the top-level setop's output column types.
3185  * safetyInfo->unsafeColumns[] is the result array.
3186  */
3187 static void
3189  pushdown_safety_info *safetyInfo)
3190 {
3191  ListCell *l;
3192  ListCell *colType = list_head(colTypes);
3193 
3194  foreach(l, tlist)
3195  {
3196  TargetEntry *tle = (TargetEntry *) lfirst(l);
3197 
3198  if (tle->resjunk)
3199  continue; /* ignore resjunk columns */
3200  if (colType == NULL)
3201  elog(ERROR, "wrong number of tlist entries");
3202  if (exprType((Node *) tle->expr) != lfirst_oid(colType))
3203  safetyInfo->unsafeColumns[tle->resno] = true;
3204  colType = lnext(colTypes, colType);
3205  }
3206  if (colType != NULL)
3207  elog(ERROR, "wrong number of tlist entries");
3208 }
3209 
3210 /*
3211  * targetIsInAllPartitionLists
3212  * True if the TargetEntry is listed in the PARTITION BY clause
3213  * of every window defined in the query.
3214  *
3215  * It would be safe to ignore windows not actually used by any window
3216  * function, but it's not easy to get that info at this stage; and it's
3217  * unlikely to be useful to spend any extra cycles getting it, since
3218  * unreferenced window definitions are probably infrequent in practice.
3219  */
3220 static bool
3222 {
3223  ListCell *lc;
3224 
3225  foreach(lc, query->windowClause)
3226  {
3227  WindowClause *wc = (WindowClause *) lfirst(lc);
3228 
3230  return false;
3231  }
3232  return true;
3233 }
3234 
3235 /*
3236  * qual_is_pushdown_safe - is a particular qual safe to push down?
3237  *
3238  * qual is a restriction clause applying to the given subquery (whose RTE
3239  * has index rti in the parent query).
3240  *
3241  * Conditions checked here:
3242  *
3243  * 1. The qual must not contain any SubPlans (mainly because I'm not sure
3244  * it will work correctly: SubLinks will already have been transformed into
3245  * SubPlans in the qual, but not in the subquery). Note that SubLinks that
3246  * transform to initplans are safe, and will be accepted here because what
3247  * we'll see in the qual is just a Param referencing the initplan output.
3248  *
3249  * 2. If unsafeVolatile is set, the qual must not contain any volatile
3250  * functions.
3251  *
3252  * 3. If unsafeLeaky is set, the qual must not contain any leaky functions
3253  * that are passed Var nodes, and therefore might reveal values from the
3254  * subquery as side effects.
3255  *
3256  * 4. The qual must not refer to the whole-row output of the subquery
3257  * (since there is no easy way to name that within the subquery itself).
3258  *
3259  * 5. The qual must not refer to any subquery output columns that were
3260  * found to be unsafe to reference by subquery_is_pushdown_safe().
3261  */
3262 static bool
3263 qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual,
3264  pushdown_safety_info *safetyInfo)
3265 {
3266  bool safe = true;
3267  List *vars;
3268  ListCell *vl;
3269 
3270  /* Refuse subselects (point 1) */
3271  if (contain_subplans(qual))
3272  return false;
3273 
3274  /* Refuse volatile quals if we found they'd be unsafe (point 2) */
3275  if (safetyInfo->unsafeVolatile &&
3277  return false;
3278 
3279  /* Refuse leaky quals if told to (point 3) */
3280  if (safetyInfo->unsafeLeaky &&
3281  contain_leaked_vars(qual))
3282  return false;
3283 
3284  /*
3285  * It would be unsafe to push down window function calls, but at least for
3286  * the moment we could never see any in a qual anyhow. (The same applies
3287  * to aggregates, which we check for in pull_var_clause below.)
3288  */
3290 
3291  /*
3292  * Examine all Vars used in clause; since it's a restriction clause, all
3293  * such Vars must refer to subselect output columns.
3294  */
3296  foreach(vl, vars)
3297  {
3298  Var *var = (Var *) lfirst(vl);
3299 
3300  /*
3301  * XXX Punt if we find any PlaceHolderVars in the restriction clause.
3302  * It's not clear whether a PHV could safely be pushed down, and even
3303  * less clear whether such a situation could arise in any cases of
3304  * practical interest anyway. So for the moment, just refuse to push
3305  * down.
3306  */
3307  if (!IsA(var, Var))
3308  {
3309  safe = false;
3310  break;
3311  }
3312 
3313  Assert(var->varno == rti);
3314  Assert(var->varattno >= 0);
3315 
3316  /* Check point 4 */
3317  if (var->varattno == 0)
3318  {
3319  safe = false;
3320  break;
3321  }
3322 
3323  /* Check point 5 */
3324  if (safetyInfo->unsafeColumns[var->varattno])
3325  {
3326  safe = false;
3327  break;
3328  }
3329  }
3330 
3331  list_free(vars);
3332 
3333  return safe;
3334 }
3335 
3336 /*
3337  * subquery_push_qual - push down a qual that we have determined is safe
3338  */
3339 static void
3340 subquery_push_qual(Query *subquery, RangeTblEntry *rte, Index rti, Node *qual)
3341 {
3342  if (subquery->setOperations != NULL)
3343  {
3344  /* Recurse to push it separately to each component query */
3345  recurse_push_qual(subquery->setOperations, subquery,
3346  rte, rti, qual);
3347  }
3348  else
3349  {
3350  /*
3351  * We need to replace Vars in the qual (which must refer to outputs of
3352  * the subquery) with copies of the subquery's targetlist expressions.
3353  * Note that at this point, any uplevel Vars in the qual should have
3354  * been replaced with Params, so they need no work.
3355  *
3356  * This step also ensures that when we are pushing into a setop tree,
3357  * each component query gets its own copy of the qual.
3358  */
3359  qual = ReplaceVarsFromTargetList(qual, rti, 0, rte,
3360  subquery->targetList,
3362  &subquery->hasSubLinks);
3363 
3364  /*
3365  * Now attach the qual to the proper place: normally WHERE, but if the
3366  * subquery uses grouping or aggregation, put it in HAVING (since the
3367  * qual really refers to the group-result rows).
3368  */
3369  if (subquery->hasAggs || subquery->groupClause || subquery->groupingSets || subquery->havingQual)
3370  subquery->havingQual = make_and_qual(subquery->havingQual, qual);
3371  else
3372  subquery->jointree->quals =
3373  make_and_qual(subquery->jointree->quals, qual);
3374 
3375  /*
3376  * We need not change the subquery's hasAggs or hasSubLinks flags,
3377  * since we can't be pushing down any aggregates that weren't there
3378  * before, and we don't push down subselects at all.
3379  */
3380  }
3381 }
3382 
3383 /*
3384  * Helper routine to recurse through setOperations tree
3385  */
3386 static void
3387 recurse_push_qual(Node *setOp, Query *topquery,
3388  RangeTblEntry *rte, Index rti, Node *qual)
3389 {
3390  if (IsA(setOp, RangeTblRef))
3391  {
3392  RangeTblRef *rtr = (RangeTblRef *) setOp;
3393  RangeTblEntry *subrte = rt_fetch(rtr->rtindex, topquery->rtable);
3394  Query *subquery = subrte->subquery;
3395 
3396  Assert(subquery != NULL);
3397  subquery_push_qual(subquery, rte, rti, qual);
3398  }
3399  else if (IsA(setOp, SetOperationStmt))
3400  {
3401  SetOperationStmt *op = (SetOperationStmt *) setOp;
3402 
3403  recurse_push_qual(op->larg, topquery, rte, rti, qual);
3404  recurse_push_qual(op->rarg, topquery, rte, rti, qual);
3405  }
3406  else
3407  {
3408  elog(ERROR, "unrecognized node type: %d",
3409  (int) nodeTag(setOp));
3410  }
3411 }
3412 
3413 /*****************************************************************************
3414  * SIMPLIFYING SUBQUERY TARGETLISTS
3415  *****************************************************************************/
3416 
3417 /*
3418  * remove_unused_subquery_outputs
3419  * Remove subquery targetlist items we don't need
3420  *
3421  * It's possible, even likely, that the upper query does not read all the
3422  * output columns of the subquery. We can remove any such outputs that are
3423  * not needed by the subquery itself (e.g., as sort/group columns) and do not
3424  * affect semantics otherwise (e.g., volatile functions can't be removed).
3425  * This is useful not only because we might be able to remove expensive-to-
3426  * compute expressions, but because deletion of output columns might allow
3427  * optimizations such as join removal to occur within the subquery.
3428  *
3429  * To avoid affecting column numbering in the targetlist, we don't physically
3430  * remove unused tlist entries, but rather replace their expressions with NULL
3431  * constants. This is implemented by modifying subquery->targetList.
3432  */
3433 static void
3435 {
3436  Bitmapset *attrs_used = NULL;
3437  ListCell *lc;
3438 
3439  /*
3440  * Do nothing if subquery has UNION/INTERSECT/EXCEPT: in principle we
3441  * could update all the child SELECTs' tlists, but it seems not worth the
3442  * trouble presently.
3443  */
3444  if (subquery->setOperations)
3445  return;
3446 
3447  /*
3448  * If subquery has regular DISTINCT (not DISTINCT ON), we're wasting our
3449  * time: all its output columns must be used in the distinctClause.
3450  */
3451  if (subquery->distinctClause && !subquery->hasDistinctOn)
3452  return;
3453 
3454  /*
3455  * Collect a bitmap of all the output column numbers used by the upper
3456  * query.
3457  *
3458  * Add all the attributes needed for joins or final output. Note: we must
3459  * look at rel's targetlist, not the attr_needed data, because attr_needed
3460  * isn't computed for inheritance child rels, cf set_append_rel_size().
3461  * (XXX might be worth changing that sometime.)
3462  */
3463  pull_varattnos((Node *) rel->reltarget->exprs, rel->relid, &attrs_used);
3464 
3465  /* Add all the attributes used by un-pushed-down restriction clauses. */
3466  foreach(lc, rel->baserestrictinfo)
3467  {
3468  RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
3469 
3470  pull_varattnos((Node *) rinfo->clause, rel->relid, &attrs_used);
3471  }
3472 
3473  /*
3474  * If there's a whole-row reference to the subquery, we can't remove
3475  * anything.
3476  */
3478  return;
3479 
3480  /*
3481  * Run through the tlist and zap entries we don't need. It's okay to
3482  * modify the tlist items in-place because set_subquery_pathlist made a
3483  * copy of the subquery.
3484  */
3485  foreach(lc, subquery->targetList)
3486  {
3487  TargetEntry *tle = (TargetEntry *) lfirst(lc);
3488  Node *texpr = (Node *) tle->expr;
3489 
3490  /*
3491  * If it has a sortgroupref number, it's used in some sort/group
3492  * clause so we'd better not remove it. Also, don't remove any
3493  * resjunk columns, since their reason for being has nothing to do
3494  * with anybody reading the subquery's output. (It's likely that
3495  * resjunk columns in a sub-SELECT would always have ressortgroupref
3496  * set, but even if they don't, it seems imprudent to remove them.)
3497  */
3498  if (tle->ressortgroupref || tle->resjunk)
3499  continue;
3500 
3501  /*
3502  * If it's used by the upper query, we can't remove it.
3503  */
3505  attrs_used))
3506  continue;
3507 
3508  /*
3509  * If it contains a set-returning function, we can't remove it since
3510  * that could change the number of rows returned by the subquery.
3511  */
3512  if (subquery->hasTargetSRFs &&
3513  expression_returns_set(texpr))
3514  continue;
3515 
3516  /*
3517  * If it contains volatile functions, we daren't remove it for fear
3518  * that the user is expecting their side-effects to happen.
3519  */
3520  if (contain_volatile_functions(texpr))
3521  continue;
3522 
3523  /*
3524  * OK, we don't need it. Replace the expression with a NULL constant.
3525  * Preserve the exposed type of the expression, in case something
3526  * looks at the rowtype of the subquery's result.
3527  */
3528  tle->expr = (Expr *) makeNullConst(exprType(texpr),
3529  exprTypmod(texpr),
3530  exprCollation(texpr));
3531  }
3532 }
3533 
3534 /*
3535  * create_partial_bitmap_paths
3536  * Build partial bitmap heap path for the relation
3537  */
3538 void
3540  Path *bitmapqual)
3541 {
3542  int parallel_workers;
3543  double pages_fetched;
3544 
3545  /* Compute heap pages for bitmap heap scan */
3546  pages_fetched = compute_bitmap_pages(root, rel, bitmapqual, 1.0,
3547  NULL, NULL);
3548 
3549  parallel_workers = compute_parallel_worker(rel, pages_fetched, -1,
3551 
3552  if (parallel_workers <= 0)
3553  return;
3554 
3555  add_partial_path(rel, (Path *) create_bitmap_heap_path(root, rel,
3556  bitmapqual, rel->lateral_relids, 1.0, parallel_workers));
3557 }
3558 
3559 /*
3560  * Compute the number of parallel workers that should be used to scan a
3561  * relation. We compute the parallel workers based on the size of the heap to
3562  * be scanned and the size of the index to be scanned, then choose a minimum
3563  * of those.
3564  *
3565  * "heap_pages" is the number of pages from the table that we expect to scan, or
3566  * -1 if we don't expect to scan any.
3567  *
3568  * "index_pages" is the number of pages from the index that we expect to scan, or
3569  * -1 if we don't expect to scan any.
3570  *
3571  * "max_workers" is caller's limit on the number of workers. This typically
3572  * comes from a GUC.
3573  */
3574 int
3575 compute_parallel_worker(RelOptInfo *rel, double heap_pages, double index_pages,
3576  int max_workers)
3577 {
3578  int parallel_workers = 0;
3579 
3580  /*
3581  * If the user has set the parallel_workers reloption, use that; otherwise
3582  * select a default number of workers.
3583  */
3584  if (rel->rel_parallel_workers != -1)
3585  parallel_workers = rel->rel_parallel_workers;
3586  else
3587  {
3588  /*
3589  * If the number of pages being scanned is insufficient to justify a
3590  * parallel scan, just return zero ... unless it's an inheritance
3591  * child. In that case, we want to generate a parallel path here
3592  * anyway. It might not be worthwhile just for this relation, but
3593  * when combined with all of its inheritance siblings it may well pay
3594  * off.
3595  */
3596  if (rel->reloptkind == RELOPT_BASEREL &&
3597  ((heap_pages >= 0 && heap_pages < min_parallel_table_scan_size) ||
3598  (index_pages >= 0 && index_pages < min_parallel_index_scan_size)))
3599  return 0;
3600 
3601  if (heap_pages >= 0)
3602  {
3603  int heap_parallel_threshold;
3604  int heap_parallel_workers = 1;
3605 
3606  /*
3607  * Select the number of workers based on the log of the size of
3608  * the relation. This probably needs to be a good deal more
3609  * sophisticated, but we need something here for now. Note that
3610  * the upper limit of the min_parallel_table_scan_size GUC is
3611  * chosen to prevent overflow here.
3612  */
3613  heap_parallel_threshold = Max(min_parallel_table_scan_size, 1);
3614  while (heap_pages >= (BlockNumber) (heap_parallel_threshold * 3))
3615  {
3616  heap_parallel_workers++;
3617  heap_parallel_threshold *= 3;
3618  if (heap_parallel_threshold > INT_MAX / 3)
3619  break; /* avoid overflow */
3620  }
3621 
3622  parallel_workers = heap_parallel_workers;
3623  }
3624 
3625  if (index_pages >= 0)
3626  {
3627  int index_parallel_workers = 1;
3628  int index_parallel_threshold;
3629 
3630  /* same calculation as for heap_pages above */
3631  index_parallel_threshold = Max(min_parallel_index_scan_size, 1);
3632  while (index_pages >= (BlockNumber) (index_parallel_threshold * 3))
3633  {
3634  index_parallel_workers++;
3635  index_parallel_threshold *= 3;
3636  if (index_parallel_threshold > INT_MAX / 3)
3637  break; /* avoid overflow */
3638  }
3639 
3640  if (parallel_workers > 0)
3641  parallel_workers = Min(parallel_workers, index_parallel_workers);
3642  else
3643  parallel_workers = index_parallel_workers;
3644  }
3645  }
3646 
3647  /* In no case use more than caller supplied maximum number of workers */
3648  parallel_workers = Min(parallel_workers, max_workers);
3649 
3650  return parallel_workers;
3651 }
3652 
3653 /*
3654  * generate_partitionwise_join_paths
3655  * Create paths representing partitionwise join for given partitioned
3656  * join relation.
3657  *
3658  * This must not be called until after we are done adding paths for all
3659  * child-joins. Otherwise, add_path might delete a path to which some path
3660  * generated here has a reference.
3661  */
3662 void
3664 {
3665  List *live_children = NIL;
3666  int cnt_parts;
3667  int num_parts;
3668  RelOptInfo **part_rels;
3669 
3670  /* Handle only join relations here. */
3671  if (!IS_JOIN_REL(rel))
3672  return;
3673 
3674  /* We've nothing to do if the relation is not partitioned. */
3675  if (!IS_PARTITIONED_REL(rel))
3676  return;
3677 
3678  /* The relation should have consider_partitionwise_join set. */
3680 
3681  /* Guard against stack overflow due to overly deep partition hierarchy. */
3683 
3684  num_parts = rel->nparts;
3685  part_rels = rel->part_rels;
3686 
3687  /* Collect non-dummy child-joins. */
3688  for (cnt_parts = 0; cnt_parts < num_parts; cnt_parts++)
3689  {
3690  RelOptInfo *child_rel = part_rels[cnt_parts];
3691 
3692  /* If it's been pruned entirely, it's certainly dummy. */
3693  if (child_rel == NULL)
3694  continue;
3695 
3696  /* Add partitionwise join paths for partitioned child-joins. */
3697  generate_partitionwise_join_paths(root, child_rel);
3698 
3699  set_cheapest(child_rel);
3700 
3701  /* Dummy children will not be scanned, so ignore those. */
3702  if (IS_DUMMY_REL(child_rel))
3703  continue;
3704 
3705 #ifdef OPTIMIZER_DEBUG
3706  debug_print_rel(root, child_rel);
3707 #endif
3708 
3709  live_children = lappend(live_children, child_rel);
3710  }
3711 
3712  /* If all child-joins are dummy, parent join is also dummy. */
3713  if (!live_children)
3714  {
3715  mark_dummy_rel(rel);
3716  return;
3717  }
3718 
3719  /* Build additional paths for this rel from child-join paths. */
3720  add_paths_to_append_rel(root, rel, live_children);
3721  list_free(live_children);
3722 }
3723 
3724 
3725 /*****************************************************************************
3726  * DEBUG SUPPORT
3727  *****************************************************************************/
3728 
3729 #ifdef OPTIMIZER_DEBUG
3730 
3731 static void
3732 print_relids(PlannerInfo *root, Relids relids)
3733 {
3734  int x;
3735  bool first = true;
3736 
3737  x = -1;
3738  while ((x = bms_next_member(relids, x)) >= 0)
3739  {
3740  if (!first)
3741  printf(" ");
3742  if (x < root->simple_rel_array_size &&
3743  root->simple_rte_array[x])
3744  printf("%s", root->simple_rte_array[x]->eref->aliasname);
3745  else
3746  printf("%d", x);
3747  first = false;
3748  }
3749 }
3750 
3751 static void
3752 print_restrictclauses(PlannerInfo *root, List *clauses)
3753 {
3754  ListCell *l;
3755 
3756  foreach(l, clauses)
3757  {
3758  RestrictInfo *c = lfirst(l);
3759 
3760  print_expr((Node *) c->clause, root->parse->rtable);
3761  if (lnext(clauses, l))
3762  printf(", ");
3763  }
3764 }
3765 
3766 static void
3767 print_path(PlannerInfo *root, Path *path, int indent)
3768 {
3769  const char *ptype;
3770  bool join = false;
3771  Path *subpath = NULL;
3772  int i;
3773 
3774  switch (nodeTag(path))
3775  {
3776  case T_Path:
3777  switch (path->pathtype)
3778  {
3779  case T_SeqScan:
3780  ptype = "SeqScan";
3781  break;
3782  case T_SampleScan:
3783  ptype = "SampleScan";
3784  break;
3785  case T_FunctionScan:
3786  ptype = "FunctionScan";
3787  break;
3788  case T_TableFuncScan:
3789  ptype = "TableFuncScan";
3790  break;
3791  case T_ValuesScan:
3792  ptype = "ValuesScan";
3793  break;
3794  case T_CteScan:
3795  ptype = "CteScan";
3796  break;
3797  case T_NamedTuplestoreScan:
3798  ptype = "NamedTuplestoreScan";
3799  break;
3800  case T_Result:
3801  ptype = "Result";
3802  break;
3803  case T_WorkTableScan:
3804  ptype = "WorkTableScan";
3805  break;
3806  default:
3807  ptype = "???Path";
3808  break;
3809  }
3810  break;
3811  case T_IndexPath:
3812  ptype = "IdxScan";
3813  break;
3814  case T_BitmapHeapPath:
3815  ptype = "BitmapHeapScan";
3816  break;
3817  case T_BitmapAndPath:
3818  ptype = "BitmapAndPath";
3819  break;
3820  case T_BitmapOrPath:
3821  ptype = "BitmapOrPath";
3822  break;
3823  case T_TidPath:
3824  ptype = "TidScan";
3825  break;
3826  case T_SubqueryScanPath:
3827  ptype = "SubqueryScan";
3828  break;
3829  case T_ForeignPath:
3830  ptype = "ForeignScan";
3831  break;
3832  case T_CustomPath:
3833  ptype = "CustomScan";
3834  break;
3835  case T_NestPath:
3836  ptype = "NestLoop";
3837  join = true;
3838  break;
3839  case T_MergePath:
3840  ptype = "MergeJoin";
3841  join = true;
3842  break;
3843  case T_HashPath:
3844  ptype = "HashJoin";
3845  join = true;
3846  break;
3847  case T_AppendPath:
3848  ptype = "Append";
3849  break;
3850  case T_MergeAppendPath:
3851  ptype = "MergeAppend";
3852  break;
3853  case T_GroupResultPath:
3854  ptype = "GroupResult";
3855  break;
3856  case T_MaterialPath:
3857  ptype = "Material";
3858  subpath = ((MaterialPath *) path)->subpath;
3859  break;
3860  case T_UniquePath:
3861  ptype = "Unique";
3862  subpath = ((UniquePath *) path)->subpath;
3863  break;
3864  case T_GatherPath:
3865  ptype = "Gather";
3866  subpath = ((GatherPath *) path)->subpath;
3867  break;
3868  case T_GatherMergePath:
3869  ptype = "GatherMerge";
3870  subpath = ((GatherMergePath *) path)->subpath;
3871  break;
3872  case T_ProjectionPath:
3873  ptype = "Projection";
3874  subpath = ((ProjectionPath *) path)->subpath;
3875  break;
3876  case T_ProjectSetPath:
3877  ptype = "ProjectSet";
3878  subpath = ((ProjectSetPath *) path)->subpath;
3879  break;
3880  case T_SortPath:
3881  ptype = "Sort";
3882  subpath = ((SortPath *) path)->subpath;
3883  break;
3884  case T_GroupPath:
3885  ptype = "Group";
3886  subpath = ((GroupPath *) path)->subpath;
3887  break;
3888  case T_UpperUniquePath:
3889  ptype = "UpperUnique";
3890  subpath = ((UpperUniquePath *) path)->subpath;
3891  break;
3892  case T_AggPath:
3893  ptype = "Agg";
3894  subpath = ((AggPath *) path)->subpath;
3895  break;
3896  case T_GroupingSetsPath:
3897  ptype = "GroupingSets";
3898  subpath = ((GroupingSetsPath *) path)->subpath;
3899  break;
3900  case T_MinMaxAggPath:
3901  ptype = "MinMaxAgg";
3902  break;
3903  case T_WindowAggPath:
3904  ptype = "WindowAgg";
3905  subpath = ((WindowAggPath *) path)->subpath;
3906  break;
3907  case T_SetOpPath:
3908  ptype = "SetOp";
3909  subpath = ((SetOpPath *) path)->subpath;
3910  break;
3911  case T_RecursiveUnionPath:
3912  ptype = "RecursiveUnion";
3913  break;
3914  case T_LockRowsPath:
3915  ptype = "LockRows";
3916  subpath = ((LockRowsPath *) path)->subpath;
3917  break;
3918  case T_ModifyTablePath:
3919  ptype = "ModifyTable";
3920  break;
3921  case T_LimitPath:
3922  ptype = "Limit";
3923  subpath = ((LimitPath *) path)->subpath;
3924  break;
3925  default:
3926  ptype = "???Path";
3927  break;
3928  }
3929 
3930  for (i = 0; i < indent; i++)
3931  printf("\t");
3932  printf("%s", ptype);
3933 
3934  if (path->parent)
3935  {
3936  printf("(");
3937  print_relids(root, path->parent->relids);
3938  printf(")");
3939  }
3940  if (path->param_info)
3941  {
3942  printf(" required_outer (");
3943  print_relids(root, path->param_info->ppi_req_outer);
3944  printf(")");
3945  }
3946  printf(" rows=%.0f cost=%.2f..%.2f\n",
3947  path->rows, path->startup_cost, path->total_cost);
3948 
3949  if (path->pathkeys)
3950  {
3951  for (i = 0; i < indent; i++)
3952  printf("\t");
3953  printf(" pathkeys: ");
3954  print_pathkeys(path->pathkeys, root->parse->rtable);
3955  }
3956 
3957  if (join)
3958  {
3959  JoinPath *jp = (JoinPath *) path;
3960 
3961  for (i = 0; i < indent; i++)
3962  printf("\t");
3963  printf(" clauses: ");
3964  print_restrictclauses(root, jp->joinrestrictinfo);
3965  printf("\n");
3966 
3967  if (IsA(path, MergePath))
3968  {
3969  MergePath *mp = (MergePath *) path;
3970 
3971  for (i = 0; i < indent; i++)
3972  printf("\t");
3973  printf(" sortouter=%d sortinner=%d materializeinner=%d\n",
3974  ((mp->outersortkeys) ? 1 : 0),
3975  ((mp->innersortkeys) ? 1 : 0),
3976  ((mp->materialize_inner) ? 1 : 0));
3977  }
3978 
3979  print_path(root, jp->outerjoinpath, indent + 1);
3980  print_path(root, jp->innerjoinpath, indent + 1);
3981  }
3982 
3983  if (subpath)
3984  print_path(root, subpath, indent + 1);
3985 }
3986 
3987 void
3988 debug_print_rel(PlannerInfo *root, RelOptInfo *rel)
3989 {
3990  ListCell *l;
3991 
3992  printf("RELOPTINFO (");
3993  print_relids(root, rel->relids);
3994  printf("): rows=%.0f width=%d\n", rel->rows, rel->reltarget->width);
3995 
3996  if (rel->baserestrictinfo)
3997  {
3998  printf("\tbaserestrictinfo: ");
3999  print_restrictclauses(root, rel->baserestrictinfo);
4000  printf("\n");
4001  }
4002 
4003  if (rel->joininfo)
4004  {
4005  printf("\tjoininfo: ");
4006  print_restrictclauses(root, rel->joininfo);
4007  printf("\n");
4008  }
4009 
4010  printf("\tpath list:\n");
4011  foreach(l, rel->pathlist)
4012  print_path(root, lfirst(l), 1);
4014  {
4015  printf("\n\tcheapest parameterized paths:\n");
4016  foreach(l, rel->cheapest_parameterized_paths)
4017  print_path(root, lfirst(l), 1);
4018  }
4019  if (rel->cheapest_startup_path)
4020  {
4021  printf("\n\tcheapest startup path:\n");
4022  print_path(root, rel->cheapest_startup_path, 1);
4023  }
4024  if (rel->cheapest_total_path)
4025  {
4026  printf("\n\tcheapest total path:\n");
4027  print_path(root, rel->cheapest_total_path, 1);
4028  }
4029  printf("\n");
4030  fflush(stdout);
4031 }
4032 
4033 #endif /* OPTIMIZER_DEBUG */
bool has_eclass_joins
Definition: pathnodes.h:709
Path * get_cheapest_path_for_pathkeys(List *paths, List *pathkeys, Relids required_outer, CostSelector cost_criterion, bool require_parallel_safe)
Definition: pathkeys.c:350
void set_subquery_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:4912
RelOptInfo * standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
Definition: allpaths.c:2843
Node * limitOffset
Definition: parsenodes.h:160
#define NIL
Definition: pg_list.h:65
bool contain_leaked_vars(Node *clause)
Definition: clauses.c:1327
List * outersortkeys
Definition: pathnodes.h:1552
double plan_rows
Definition: plannodes.h:127
RelOptInfo * make_one_rel(PlannerInfo *root, List *joinlist)
Definition: allpaths.c:152
GatherPath * create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, PathTarget *target, Relids required_outer, double *rows)
Definition: pathnode.c:1845
#define IsA(nodeptr, _type_)
Definition: nodes.h:576
PathTarget * pathtarget
Definition: pathnodes.h:1115
List * build_expression_pathkey(PlannerInfo *root, Expr *expr, Relids nullable_relids, Oid opno, Relids rel, bool create_it)
Definition: pathkeys.c:728
Query * parse
Definition: pathnodes.h:177
Index varlevelsup
Definition: primnodes.h:177
#define forboth(cell1, list1, cell2, list2)
Definition: pg_list.h:419
void add_path(RelOptInfo *parent_rel, Path *new_path)
Definition: pathnode.c:423
SubqueryScanPath * create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, List *pathkeys, Relids required_outer)
Definition: pathnode.c:1884
RelOptInfo *(* join_search_hook_type)(PlannerInfo *root, int levels_needed, List *initial_rels)
Definition: paths.h:45
List * plan_params
Definition: pathnodes.h:191
List * sortClause
Definition: parsenodes.h:158
RelOptKind reloptkind
Definition: pathnodes.h:638
static void set_base_rel_sizes(PlannerInfo *root)
Definition: allpaths.c:291
Relids * attr_needed
Definition: pathnodes.h:674
List * join_info_list
Definition: pathnodes.h:281
static Path * get_cheapest_parameterized_child_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: allpaths.c:1954
bool materialize_inner
Definition: pathnodes.h:1555
static ListCell * lnext(const List *l, const ListCell *c)
Definition: pg_list.h:321
static void set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2497
FromExpr * jointree
Definition: parsenodes.h:138
static void set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:904
PlannerInfo * parent_root
Definition: pathnodes.h:183
#define castNode(_type_, nodeptr)
Definition: nodes.h:594
int compute_parallel_worker(RelOptInfo *rel, double heap_pages, double index_pages, int max_workers)
Definition: allpaths.c:3575
List * build_partition_pathkeys(PlannerInfo *root, RelOptInfo *partrel, ScanDirection scandir, bool *partialkeys)
Definition: pathkeys.c:645
int32 exprTypmod(const Node *expr)
Definition: nodeFuncs.c:276
Path * innerjoinpath
Definition: pathnodes.h:1496
void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel, List *live_childrels)
Definition: allpaths.c:1294
void set_cte_size_estimates(PlannerInfo *root, RelOptInfo *rel, double cte_rows)
Definition: costsize.c:5084
struct Path * cheapest_startup_path
Definition: pathnodes.h:658
Path * create_resultscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2041
double tuples
Definition: pathnodes.h:681
List * baserestrictinfo
Definition: pathnodes.h:703
bool hasAggs
Definition: parsenodes.h:125
void create_index_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: indxpath.c:232
#define Min(x, y)
Definition: c.h:904
int bms_next_member(const Bitmapset *a, int prevbit)
Definition: bitmapset.c:1043
bool expression_returns_set(Node *clause)
Definition: nodeFuncs.c:670
int parallel_workers
Definition: pathnodes.h:1121
set_rel_pathlist_hook_type set_rel_pathlist_hook
Definition: allpaths.c:68
bool pseudoconstant
Definition: pathnodes.h:1951
bool consider_param_startup
Definition: pathnodes.h:648
MaterialPath * create_material_path(RelOptInfo *rel, Path *subpath)
Definition: pathnode.c:1496
bool relation_excluded_by_constraints(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: plancat.c:1385
List * list_truncate(List *list, int new_size)
Definition: list.c:586
ParamPathInfo * param_info
Definition: pathnodes.h:1117
void generate_partitionwise_join_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: allpaths.c:3663
List * groupingSets
Definition: parsenodes.h:150
static void set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2563
List * list_copy(const List *oldlist)
Definition: list.c:1404
bool limit_needed(Query *parse)
Definition: planner.c:3007
Definition: nodes.h:525
#define IS_JOIN_REL(rel)
Definition: pathnodes.h:619
List * list_concat(List *list1, const List *list2)
Definition: list.c:516
List * partial_pathlist
Definition: pathnodes.h:657
List * make_tlist_from_pathtarget(PathTarget *target)
Definition: tlist.c:639
AttrNumber varattno
Definition: primnodes.h:172
bool bms_get_singleton_member(const Bitmapset *a, int *member)
Definition: bitmapset.c:615
bool * unsafeColumns
Definition: allpaths.c:56
bool enable_parallel_append
Definition: costsize.c:138
static void set_append_rel_size(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:939
#define printf(...)
Definition: port.h:198
#define FirstLowInvalidHeapAttributeNumber
Definition: sysattr.h:27
uint32 BlockNumber
Definition: block.h:31
void generate_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
Definition: allpaths.c:2682
List * cheapest_parameterized_paths
Definition: pathnodes.h:661
Node * make_and_qual(Node *qual1, Node *qual2)
Definition: makefuncs.c:686
List * pull_var_clause(Node *node, int flags)
Definition: var.c:535
PathKeysComparison compare_pathkeys(List *keys1, List *keys2)
Definition: pathkeys.c:285
void add_child_rel_equivalences(PlannerInfo *root, AppendRelInfo *appinfo, RelOptInfo *parent_rel, RelOptInfo *child_rel)
Definition: equivclass.c:2222
bool contain_volatile_functions(Node *clause)
Definition: clauses.c:724
struct pushdown_safety_info pushdown_safety_info
List * list_copy_tail(const List *oldlist, int nskip)
Definition: list.c:1423
bool funcordinality
Definition: parsenodes.h:1041
static bool recurse_pushdown_safe(Node *setOp, Query *topquery, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3053
Definition: primnodes.h:167
static void set_result_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2593
static int list_nth_int(const List *list, int n)
Definition: pg_list.h:288
static void compare_tlist_datatypes(List *tlist, List *colTypes, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3188
void set_result_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5154
Path * create_functionscan_path(PlannerInfo *root, RelOptInfo *rel, List *pathkeys, Relids required_outer)
Definition: pathnode.c:1912
static void set_base_rel_consider_startup(PlannerInfo *root)
Definition: allpaths.c:248
List * values_lists
Definition: parsenodes.h:1051
Node * quals
Definition: primnodes.h:1497
#define IS_SIMPLE_REL(rel)
Definition: pathnodes.h:614
void pull_varattnos(Node *node, Index varno, Bitmapset **varattnos)
Definition: var.c:219
bool hasDistinctOn
Definition: parsenodes.h:129
signed int int32
Definition: c.h:346
List * windowClause
Definition: parsenodes.h:154
List * targetList
Definition: parsenodes.h:140
Const * makeNullConst(Oid consttype, int32 consttypmod, Oid constcollid)
Definition: makefuncs.c:335
int first_partial_path
Definition: pathnodes.h:1376
int fls(int mask)
Definition: fls.c:55
Path * create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:1964
struct RelOptInfo ** simple_rel_array
Definition: pathnodes.h:201
NodeTag pathtype
Definition: pathnodes.h:1112
Relids syn_righthand
Definition: pathnodes.h:2136
#define list_make1(x1)
Definition: pg_list.h:227
List * subpaths
Definition: pathnodes.h:1374
PlannerInfo * subroot
Definition: pathnodes.h:685
bool contain_subplans(Node *clause)
Definition: clauses.c:610
bool is_parallel_safe(PlannerInfo *root, Node *node)
Definition: clauses.c:854
Relids lateral_relids
Definition: pathnodes.h:666
double tuple_fraction
Definition: pathnodes.h:334
void pfree(void *pointer)
Definition: mcxt.c:1056
void create_tidscan_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: tidpath.c:385
bool resjunk
Definition: primnodes.h:1400
#define linitial(l)
Definition: pg_list.h:195
IsForeignScanParallelSafe_function IsForeignScanParallelSafe
Definition: fdwapi.h:240
Definition: nodes.h:46
List * rtable
Definition: parsenodes.h:137
List * distinctClause
Definition: parsenodes.h:156
Relids all_baserels
Definition: pathnodes.h:225
#define ERROR
Definition: elog.h:43
List * partitionClause
Definition: parsenodes.h:1330
static void * list_nth(const List *list, int n)
Definition: pg_list.h:277
Cost startup_cost
Definition: pathnodes.h:1125
List * joinrestrictinfo
Definition: pathnodes.h:1498
bool enable_geqo
Definition: allpaths.c:62
bool parallelModeOK
Definition: pathnodes.h:141
#define IS_DUMMY_REL(r)
Definition: pathnodes.h:1388
RelOptInfo * parent
Definition: pathnodes.h:1114
bool bms_is_subset(const Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:315
RelOptInfo * fetch_upper_rel(PlannerInfo *root, UpperRelationKind kind, Relids relids)
Definition: relnode.c:1164
static bool targetIsInAllPartitionLists(TargetEntry *tle, Query *query)
Definition: allpaths.c:3221
int compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
Definition: pathnode.c:72
void check_index_predicates(PlannerInfo *root, RelOptInfo *rel)
Definition: indxpath.c:3395
struct Path * cheapest_total_path
Definition: pathnodes.h:659
static RelOptInfo * make_rel_from_joinlist(PlannerInfo *root, List *joinlist)
Definition: allpaths.c:2738
static void set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:1218
Node * limitCount
Definition: parsenodes.h:161
#define PATH_REQ_OUTER(path)
Definition: pathnodes.h:1133
char * c
static void set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:591
List * subplans
Definition: pathnodes.h:111
static void set_dummy_rel_pathlist(RelOptInfo *rel)
Definition: allpaths.c:2122
List * joininfo
Definition: pathnodes.h:707
List * convert_subquery_pathkeys(PlannerInfo *root, RelOptInfo *rel, List *subquery_pathkeys, List *subquery_tlist)
Definition: pathkeys.c:784
void check_stack_depth(void)
Definition: postgres.c:3262
PlannerGlobal * glob
Definition: pathnodes.h:179
static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel)
Definition: allpaths.c:3434
struct FdwRoutine * fdwroutine
Definition: pathnodes.h:694
static void set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:2181
int nparts
Definition: pathnodes.h:719
GetForeignRelSize_function GetForeignRelSize
Definition: fdwapi.h:188
AttrNumber resno
Definition: primnodes.h:1394
static void set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:921
static ListCell * list_head(const List *l)
Definition: pg_list.h:125
MergeAppendPath * create_merge_append_path(PlannerInfo *root, RelOptInfo *rel, List *subpaths, List *pathkeys, Relids required_outer, List *partitioned_rels)
Definition: pathnode.c:1346
Relids relids
Definition: pathnodes.h:641
int min_parallel_index_scan_size
Definition: allpaths.c:65
double total_table_pages
Definition: pathnodes.h:331
#define list_make1_int(x1)
Definition: pg_list.h:238
int simple_rel_array_size
Definition: pathnodes.h:202
double rint(double x)
Definition: rint.c:21
void join_search_one_level(PlannerInfo *root, int level)
Definition: joinrels.c:67
void print_expr(const Node *expr, const List *rtable)
Definition: print.c:321
static Path * get_singleton_append_subpath(Path *path)
Definition: allpaths.c:2088
#define rt_fetch(rangetable_index, rangetable)
Definition: parsetree.h:31
Path * get_cheapest_parallel_safe_total_inner(List *paths)
Definition: pathkeys.c:428
AppendPath * create_append_path(PlannerInfo *root, RelOptInfo *rel, List *subpaths, List *partial_subpaths, List *pathkeys, Relids required_outer, int parallel_workers, bool parallel_aware, List *partitioned_rels, double rows)
Definition: pathnode.c:1184
Index relid
Definition: pathnodes.h:669
Path * create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2067
bool contain_window_function(Node *clause)
Definition: clauses.c:494
List * lappend(List *list, void *datum)
Definition: list.c:322
static void set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:816
RangeTblEntry ** simple_rte_array
Definition: pathnodes.h:209
GetForeignPaths_function GetForeignPaths
Definition: fdwapi.h:189
Expr * clause
Definition: pathnodes.h:1943
bool bms_is_empty(const Bitmapset *a)
Definition: bitmapset.c:701
static void set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:766
SampleScanGetSampleSize_function SampleScanGetSampleSize
Definition: tsmapi.h:68
Index varno
Definition: primnodes.h:170
static void check_output_expressions(Query *subquery, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3120
Node * ReplaceVarsFromTargetList(Node *node, int target_varno, int sublevels_up, RangeTblEntry *target_rte, List *targetlist, ReplaceVarsNoMatchOption nomatch_option, int nomatch_varno, bool *outer_hasSubLinks)
void set_cheapest(RelOptInfo *parent_rel)
Definition: pathnode.c:245
List * exprs
Definition: pathnodes.h:1044
static bool qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3263
void print_pathkeys(const List *pathkeys, const List *rtable)
Definition: print.c:427
#define PVC_INCLUDE_PLACEHOLDERS
Definition: optimizer.h:174
bool pathkeys_contained_in(List *keys1, List *keys2)
Definition: pathkeys.c:324
Path * outerjoinpath
Definition: pathnodes.h:1495
void set_namedtuplestore_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5121
bool consider_partitionwise_join
Definition: pathnodes.h:712
join_search_hook_type join_search_hook
Definition: allpaths.c:71
BMS_Membership bms_membership(const Bitmapset *a)
Definition: bitmapset.c:672
void * palloc0(Size size)
Definition: mcxt.c:980
static void set_base_rel_pathlists(PlannerInfo *root)
Definition: allpaths.c:334
void mark_dummy_rel(RelOptInfo *rel)
Definition: joinrels.c:1257
Path * create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:955
int rel_parallel_workers
Definition: pathnodes.h:687
List * append_rel_list
Definition: pathnodes.h:288
List * cte_plan_ids
Definition: pathnodes.h:259
Path * create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:1938
bool self_reference
Definition: parsenodes.h:1058
void set_baserel_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:4404
struct PartitionBoundInfoData * boundinfo
Definition: pathnodes.h:720
unsigned int Index
Definition: c.h:475
int geqo_threshold
Definition: allpaths.c:63
RTEKind rtekind
Definition: pathnodes.h:671
RelOptInfo * geqo(PlannerInfo *root, int number_of_rels, List *initial_rels)
Definition: geqo_main.c:67
static void set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:856
bool security_barrier
Definition: parsenodes.h:1010
static void set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2473
int32 get_typavgwidth(Oid typid, int32 typmod)
Definition: lsyscache.c:2356
double rows
Definition: pathnodes.h:644
#define InvalidOid
Definition: postgres_ext.h:36
bool targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
GatherMergePath * create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, PathTarget *target, List *pathkeys, Relids required_outer, double *rows)
Definition: pathnode.c:1754
Cost total_cost
Definition: pathnodes.h:1126
bool hasTargetSRFs
Definition: parsenodes.h:127
List * lcons(void *datum, List *list)
Definition: list.c:454
List * pathkeys
Definition: pathnodes.h:1128
static void set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:470
TsmRoutine * GetTsmRoutine(Oid tsmhandler)
Definition: tablesample.c:27
#define Max(x, y)
Definition: c.h:898
bool partitions_are_ordered(PartitionBoundInfo boundinfo, int nparts)
Definition: partbounds.c:876
BlockNumber pages
Definition: pathnodes.h:680
#define Assert(condition)
Definition: c.h:732
#define lfirst(lc)
Definition: pg_list.h:190
char * aliasname
Definition: primnodes.h:42
bool hasWindowFuncs
Definition: parsenodes.h:126
void(* set_rel_pathlist_hook_type)(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: paths.h:29
List ** join_rel_level
Definition: pathnodes.h:254
List * functions
Definition: parsenodes.h:1040
static bool has_multiple_baserels(PlannerInfo *root)
Definition: allpaths.c:2148
double rows
Definition: pathnodes.h:1124
Expr * expr
Definition: primnodes.h:1393
char func_parallel(Oid funcid)
Definition: lsyscache.c:1601
JoinType jointype
Definition: pathnodes.h:2137
struct RelOptInfo ** part_rels
Definition: pathnodes.h:722
struct Path * non_recursive_path
Definition: pathnodes.h:352
bool enable_partitionwise_join
Definition: costsize.c:136
static void subquery_push_qual(Query *subquery, RangeTblEntry *rte, Index rti, Node *qual)
Definition: allpaths.c:3340
Oid exprType(const Node *expr)
Definition: nodeFuncs.c:42
static int list_length(const List *l)
Definition: pg_list.h:169
int min_parallel_table_scan_size
Definition: allpaths.c:64
Oid exprCollation(const Node *expr)
Definition: nodeFuncs.c:720
SetOperation op
Definition: parsenodes.h:1634
static void recurse_push_qual(Node *setOp, Query *topquery, RangeTblEntry *rte, Index rti, Node *qual)
Definition: allpaths.c:3387
Index ctelevelsup
Definition: parsenodes.h:1057
bool consider_parallel
Definition: pathnodes.h:649
Bitmapset * bms_add_member(Bitmapset *a, int x)
Definition: bitmapset.c:736
List * innersortkeys
Definition: pathnodes.h:1553
bool repeatable_across_scans
Definition: tsmapi.h:65
Bitmapset * Relids
Definition: pathnodes.h:28
Index query_level
Definition: pathnodes.h:181
#define InvalidAttrNumber
Definition: attnum.h:23
#define nodeTag(nodeptr)
Definition: nodes.h:530
char get_rel_persistence(Oid relid)
Definition: lsyscache.c:1880
RTEKind rtekind
Definition: parsenodes.h:974
static bool subquery_is_pushdown_safe(Query *subquery, Query *topquery, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3001
List * cteList
Definition: parsenodes.h:135
char * ctename
Definition: parsenodes.h:1056
Node * setOperations
Definition: parsenodes.h:165
List * partitioned_child_rels
Definition: pathnodes.h:726
Query * subquery
Definition: parsenodes.h:1009
List * groupClause
Definition: parsenodes.h:148
Path * reparameterize_path(PlannerInfo *root, Path *path, Relids required_outer, double loop_count)
Definition: pathnode.c:3667
AttrNumber max_attr
Definition: pathnodes.h:673
void set_values_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5052
void add_partial_path(RelOptInfo *parent_rel, Path *new_path)
Definition: pathnode.c:750
#define IS_PARTITIONED_REL(rel)
Definition: pathnodes.h:737
bool hasSubLinks
Definition: parsenodes.h:128
static void set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:574
void list_free(List *list)
Definition: list.c:1377
#define elog(elevel,...)
Definition: elog.h:226
int i
Index ressortgroupref
Definition: primnodes.h:1396
bool has_useful_pathkeys(PlannerInfo *root, RelOptInfo *rel)
Definition: pathkeys.c:1856
PartitionScheme part_scheme
Definition: pathnodes.h:718
bool parallel_aware
Definition: pathnodes.h:1119
List * initial_rels
Definition: pathnodes.h:306
void set_tablefunc_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5030
List * pathlist
Definition: pathnodes.h:655
static void create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: allpaths.c:796
Relids ppi_req_outer
Definition: pathnodes.h:1073
void set_function_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:4992
Index child_relid
Definition: pathnodes.h:2192
Alias * eref
Definition: parsenodes.h:1092
RelOptInfo * find_base_rel(PlannerInfo *root, int relid)
Definition: relnode.c:363
#define copyObject(obj)
Definition: nodes.h:641
static void set_function_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2385
Node * havingQual
Definition: parsenodes.h:152
Index parent_relid
Definition: pathnodes.h:2191
int32 * attr_widths
Definition: pathnodes.h:675
void create_partial_bitmap_paths(PlannerInfo *root, RelOptInfo *rel, Path *bitmapqual)
Definition: allpaths.c:3539
void set_foreign_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5183
Definition: nodes.h:226
double clamp_row_est(double nrows)
Definition: costsize.c:187
static void generate_orderedappend_paths(PlannerInfo *root, RelOptInfo *rel, List *live_childrels, List *all_child_pathkeys, List *partitioned_rels)
Definition: allpaths.c:1743
Definition: regcomp.c:224
static void accumulate_append_subpath(Path *path, List **subpaths, List **special_subpaths)
Definition: allpaths.c:2042
int max_parallel_workers_per_gather
Definition: costsize.c:122
Definition: pg_list.h:50
bool bms_is_member(int x, const Bitmapset *a)
Definition: bitmapset.c:427
struct PathTarget * reltarget
Definition: pathnodes.h:652
static void set_values_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2453
struct TableSampleClause * tablesample
Definition: parsenodes.h:1004
int16 AttrNumber
Definition: attnum.h:21
static void set_rel_size(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:361
List * subplan_params
Definition: pathnodes.h:686
PlannerInfo * subquery_planner(PlannerGlobal *glob, Query *parse, PlannerInfo *parent_root, bool hasRecursion, double tuple_fraction)
Definition: planner.c:596
Path * create_seqscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer, int parallel_workers)
Definition: pathnode.c:930
double compute_bitmap_pages(PlannerInfo *root, RelOptInfo *baserel, Path *bitmapqual, int loop_count, Cost *cost, double *tuple)
Definition: costsize.c:5508
Datum subpath(PG_FUNCTION_ARGS)
Definition: ltree_op.c:241
#define lfirst_oid(lc)
Definition: pg_list.h:192
bool bms_equal(const Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:94
static struct subre * parse(struct vars *, int, int, struct state *, struct state *)
Definition: regcomp.c:648
Node * adjust_appendrel_attrs(PlannerInfo *root, Node *node, int nappinfos, AppendRelInfo **appinfos)
Definition: appendinfo.c:180
unsigned char bool
Definition: c.h:308
BitmapHeapPath * create_bitmap_heap_path(PlannerInfo *root, RelOptInfo *rel, Path *bitmapqual, Relids required_outer, double loop_count, int parallel_degree)
Definition: pathnode.c:1047
Path * create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2015
static void set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2623
Path * create_ctescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:1990
AttrNumber min_attr
Definition: pathnodes.h:672