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) 1996-2020, 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"
47 #include "partitioning/partbounds.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 */
59 } pushdown_safety_info;
60 
61 /* These parameters are set by GUC */
62 bool enable_geqo = false; /* just in case GUC doesn't set it */
63 int geqo_threshold;
64 int min_parallel_table_scan_size;
65 int min_parallel_index_scan_size;
66 
67 /* Hook for plugins to get control in set_rel_pathlist() */
68 set_rel_pathlist_hook_type set_rel_pathlist_hook = NULL;
69 
70 /* Hook for plugins to replace standard_join_search() */
71 join_search_hook_type join_search_hook = NULL;
72 
73 
74 static void set_base_rel_consider_startup(PlannerInfo *root);
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);
90 static void set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
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);
104 static Path *get_cheapest_parameterized_child_path(PlannerInfo *root,
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 */
180  set_base_rel_consider_startup(root);
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  */
222  set_base_rel_pathlists(root);
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
248 set_base_rel_consider_startup(PlannerInfo *root)
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
291 set_base_rel_sizes(PlannerInfo *root)
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
334 set_base_rel_pathlists(PlannerInfo *root)
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
361 set_rel_size(PlannerInfo *root, RelOptInfo *rel,
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  */
378  set_dummy_rel_pathlist(rel);
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  */
402  set_dummy_rel_pathlist(rel);
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
470 set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
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  */
539  if (set_rel_pathlist_hook)
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_useful_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
574 set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
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
591 set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel,
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 we 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);
649  if (!rel->fdwroutine->IsForeignScanParallelSafe)
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
766 set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
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
796 create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel)
797 {
798  int parallel_workers;
799 
800  parallel_workers = compute_parallel_worker(rel, rel->pages, -1,
801  max_parallel_workers_per_gather);
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
816 set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
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
856 set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
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 ||
888  bms_membership(root->all_baserels) != BMS_SINGLETON) &&
889  !(GetTsmRoutine(rte->tablesample->tsmhandler)->repeatable_across_scans))
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
904 set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
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  * Also, make sure rel->tuples is not insane relative to rel->rows.
917  * Notably, this ensures sanity if pg_class.reltuples contains -1 and the
918  * FDW doesn't do anything to replace that.
919  */
920  rel->tuples = Max(rel->tuples, rel->rows);
921 }
922 
923 /*
924  * set_foreign_pathlist
925  * Build access paths for a foreign table RTE
926  */
927 static void
928 set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
929 {
930  /* Call the FDW's GetForeignPaths function to generate path(s) */
931  rel->fdwroutine->GetForeignPaths(root, rel, rte->relid);
932 }
933 
934 /*
935  * set_append_rel_size
936  * Set size estimates for a simple "append relation"
937  *
938  * The passed-in rel and RTE represent the entire append relation. The
939  * relation's contents are computed by appending together the output of the
940  * individual member relations. Note that in the non-partitioned inheritance
941  * case, the first member relation is actually the same table as is mentioned
942  * in the parent RTE ... but it has a different RTE and RelOptInfo. This is
943  * a good thing because their outputs are not the same size.
944  */
945 static void
946 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
947  Index rti, RangeTblEntry *rte)
948 {
949  int parentRTindex = rti;
950  bool has_live_children;
951  double parent_rows;
952  double parent_size;
953  double *parent_attrsizes;
954  int nattrs;
955  ListCell *l;
956 
957  /* Guard against stack overflow due to overly deep inheritance tree. */
958  check_stack_depth();
959 
960  Assert(IS_SIMPLE_REL(rel));
961 
962  /*
963  * Initialize partitioned_child_rels to contain this RT index.
964  *
965  * Note that during the set_append_rel_pathlist() phase, we will bubble up
966  * the indexes of partitioned relations that appear down in the tree, so
967  * that when we've created Paths for all the children, the root
968  * partitioned table's list will contain all such indexes.
969  */
970  if (rte->relkind == RELKIND_PARTITIONED_TABLE)
971  rel->partitioned_child_rels = list_make1_int(rti);
972 
973  /*
974  * If this is a partitioned baserel, set the consider_partitionwise_join
975  * flag; currently, we only consider partitionwise joins with the baserel
976  * if its targetlist doesn't contain a whole-row Var.
977  */
978  if (enable_partitionwise_join &&
979  rel->reloptkind == RELOPT_BASEREL &&
980  rte->relkind == RELKIND_PARTITIONED_TABLE &&
981  rel->attr_needed[InvalidAttrNumber - rel->min_attr] == NULL)
982  rel->consider_partitionwise_join = true;
983 
984  /*
985  * Initialize to compute size estimates for whole append relation.
986  *
987  * We handle width estimates by weighting the widths of different child
988  * rels proportionally to their number of rows. This is sensible because
989  * the use of width estimates is mainly to compute the total relation
990  * "footprint" if we have to sort or hash it. To do this, we sum the
991  * total equivalent size (in "double" arithmetic) and then divide by the
992  * total rowcount estimate. This is done separately for the total rel
993  * width and each attribute.
994  *
995  * Note: if you consider changing this logic, beware that child rels could
996  * have zero rows and/or width, if they were excluded by constraints.
997  */
998  has_live_children = false;
999  parent_rows = 0;
1000  parent_size = 0;
1001  nattrs = rel->max_attr - rel->min_attr + 1;
1002  parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
1003 
1004  foreach(l, root->append_rel_list)
1005  {
1006  AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1007  int childRTindex;
1008  RangeTblEntry *childRTE;
1009  RelOptInfo *childrel;
1010  ListCell *parentvars;
1011  ListCell *childvars;
1012 
1013  /* append_rel_list contains all append rels; ignore others */
1014  if (appinfo->parent_relid != parentRTindex)
1015  continue;
1016 
1017  childRTindex = appinfo->child_relid;
1018  childRTE = root->simple_rte_array[childRTindex];
1019 
1020  /*
1021  * The child rel's RelOptInfo was already created during
1022  * add_other_rels_to_query.
1023  */
1024  childrel = find_base_rel(root, childRTindex);
1025  Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
1026 
1027  /* We may have already proven the child to be dummy. */
1028  if (IS_DUMMY_REL(childrel))
1029  continue;
1030 
1031  /*
1032  * We have to copy the parent's targetlist and quals to the child,
1033  * with appropriate substitution of variables. However, the
1034  * baserestrictinfo quals were already copied/substituted when the
1035  * child RelOptInfo was built. So we don't need any additional setup
1036  * before applying constraint exclusion.
1037  */
1038  if (relation_excluded_by_constraints(root, childrel, childRTE))
1039  {
1040  /*
1041  * This child need not be scanned, so we can omit it from the
1042  * appendrel.
1043  */
1044  set_dummy_rel_pathlist(childrel);
1045  continue;
1046  }
1047 
1048  /*
1049  * Constraint exclusion failed, so copy the parent's join quals and
1050  * targetlist to the child, with appropriate variable substitutions.
1051  *
1052  * NB: the resulting childrel->reltarget->exprs may contain arbitrary
1053  * expressions, which otherwise would not occur in a rel's targetlist.
1054  * Code that might be looking at an appendrel child must cope with
1055  * such. (Normally, a rel's targetlist would only include Vars and
1056  * PlaceHolderVars.) XXX we do not bother to update the cost or width
1057  * fields of childrel->reltarget; not clear if that would be useful.
1058  */
1059  childrel->joininfo = (List *)
1060  adjust_appendrel_attrs(root,
1061  (Node *) rel->joininfo,
1062  1, &appinfo);
1063  childrel->reltarget->exprs = (List *)
1064  adjust_appendrel_attrs(root,
1065  (Node *) rel->reltarget->exprs,
1066  1, &appinfo);
1067 
1068  /*
1069  * We have to make child entries in the EquivalenceClass data
1070  * structures as well. This is needed either if the parent
1071  * participates in some eclass joins (because we will want to consider
1072  * inner-indexscan joins on the individual children) or if the parent
1073  * has useful pathkeys (because we should try to build MergeAppend
1074  * paths that produce those sort orderings).
1075  */
1076  if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
1077  add_child_rel_equivalences(root, appinfo, rel, childrel);
1078  childrel->has_eclass_joins = rel->has_eclass_joins;
1079 
1080  /*
1081  * Note: we could compute appropriate attr_needed data for the child's
1082  * variables, by transforming the parent's attr_needed through the
1083  * translated_vars mapping. However, currently there's no need
1084  * because attr_needed is only examined for base relations not
1085  * otherrels. So we just leave the child's attr_needed empty.
1086  */
1087 
1088  /*
1089  * If we consider partitionwise joins with the parent rel, do the same
1090  * for partitioned child rels.
1091  *
1092  * Note: here we abuse the consider_partitionwise_join flag by setting
1093  * it for child rels that are not themselves partitioned. We do so to
1094  * tell try_partitionwise_join() that the child rel is sufficiently
1095  * valid to be used as a per-partition input, even if it later gets
1096  * proven to be dummy. (It's not usable until we've set up the
1097  * reltarget and EC entries, which we just did.)
1098  */
1099  if (rel->consider_partitionwise_join)
1100  childrel->consider_partitionwise_join = true;
1101 
1102  /*
1103  * If parallelism is allowable for this query in general, see whether
1104  * it's allowable for this childrel in particular. But if we've
1105  * already decided the appendrel is not parallel-safe as a whole,
1106  * there's no point in considering parallelism for this child. For
1107  * consistency, do this before calling set_rel_size() for the child.
1108  */
1109  if (root->glob->parallelModeOK && rel->consider_parallel)
1110  set_rel_consider_parallel(root, childrel, childRTE);
1111 
1112  /*
1113  * Compute the child's size.
1114  */
1115  set_rel_size(root, childrel, childRTindex, childRTE);
1116 
1117  /*
1118  * It is possible that constraint exclusion detected a contradiction
1119  * within a child subquery, even though we didn't prove one above. If
1120  * so, we can skip this child.
1121  */
1122  if (IS_DUMMY_REL(childrel))
1123  continue;
1124 
1125  /* We have at least one live child. */
1126  has_live_children = true;
1127 
1128  /*
1129  * If any live child is not parallel-safe, treat the whole appendrel
1130  * as not parallel-safe. In future we might be able to generate plans
1131  * in which some children are farmed out to workers while others are
1132  * not; but we don't have that today, so it's a waste to consider
1133  * partial paths anywhere in the appendrel unless it's all safe.
1134  * (Child rels visited before this one will be unmarked in
1135  * set_append_rel_pathlist().)
1136  */
1137  if (!childrel->consider_parallel)
1138  rel->consider_parallel = false;
1139 
1140  /*
1141  * Accumulate size information from each live child.
1142  */
1143  Assert(childrel->rows > 0);
1144 
1145  parent_rows += childrel->rows;
1146  parent_size += childrel->reltarget->width * childrel->rows;
1147 
1148  /*
1149  * Accumulate per-column estimates too. We need not do anything for
1150  * PlaceHolderVars in the parent list. If child expression isn't a
1151  * Var, or we didn't record a width estimate for it, we have to fall
1152  * back on a datatype-based estimate.
1153  *
1154  * By construction, child's targetlist is 1-to-1 with parent's.
1155  */
1156  forboth(parentvars, rel->reltarget->exprs,
1157  childvars, childrel->reltarget->exprs)
1158  {
1159  Var *parentvar = (Var *) lfirst(parentvars);
1160  Node *childvar = (Node *) lfirst(childvars);
1161 
1162  if (IsA(parentvar, Var))
1163  {
1164  int pndx = parentvar->varattno - rel->min_attr;
1165  int32 child_width = 0;
1166 
1167  if (IsA(childvar, Var) &&
1168  ((Var *) childvar)->varno == childrel->relid)
1169  {
1170  int cndx = ((Var *) childvar)->varattno - childrel->min_attr;
1171 
1172  child_width = childrel->attr_widths[cndx];
1173  }
1174  if (child_width <= 0)
1175  child_width = get_typavgwidth(exprType(childvar),
1176  exprTypmod(childvar));
1177  Assert(child_width > 0);
1178  parent_attrsizes[pndx] += child_width * childrel->rows;
1179  }
1180  }
1181  }
1182 
1183  if (has_live_children)
1184  {
1185  /*
1186  * Save the finished size estimates.
1187  */
1188  int i;
1189 
1190  Assert(parent_rows > 0);
1191  rel->rows = parent_rows;
1192  rel->reltarget->width = rint(parent_size / parent_rows);
1193  for (i = 0; i < nattrs; i++)
1194  rel->attr_widths[i] = rint(parent_attrsizes[i] / parent_rows);
1195 
1196  /*
1197  * Set "raw tuples" count equal to "rows" for the appendrel; needed
1198  * because some places assume rel->tuples is valid for any baserel.
1199  */
1200  rel->tuples = parent_rows;
1201 
1202  /*
1203  * Note that we leave rel->pages as zero; this is important to avoid
1204  * double-counting the appendrel tree in total_table_pages.
1205  */
1206  }
1207  else
1208  {
1209  /*
1210  * All children were excluded by constraints, so mark the whole
1211  * appendrel dummy. We must do this in this phase so that the rel's
1212  * dummy-ness is visible when we generate paths for other rels.
1213  */
1214  set_dummy_rel_pathlist(rel);
1215  }
1216 
1217  pfree(parent_attrsizes);
1218 }
1219 
1220 /*
1221  * set_append_rel_pathlist
1222  * Build access paths for an "append relation"
1223  */
1224 static void
1225 set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
1226  Index rti, RangeTblEntry *rte)
1227 {
1228  int parentRTindex = rti;
1229  List *live_childrels = NIL;
1230  ListCell *l;
1231 
1232  /*
1233  * Generate access paths for each member relation, and remember the
1234  * non-dummy children.
1235  */
1236  foreach(l, root->append_rel_list)
1237  {
1238  AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
1239  int childRTindex;
1240  RangeTblEntry *childRTE;
1241  RelOptInfo *childrel;
1242 
1243  /* append_rel_list contains all append rels; ignore others */
1244  if (appinfo->parent_relid != parentRTindex)
1245  continue;
1246 
1247  /* Re-locate the child RTE and RelOptInfo */
1248  childRTindex = appinfo->child_relid;
1249  childRTE = root->simple_rte_array[childRTindex];
1250  childrel = root->simple_rel_array[childRTindex];
1251 
1252  /*
1253  * If set_append_rel_size() decided the parent appendrel was
1254  * parallel-unsafe at some point after visiting this child rel, we
1255  * need to propagate the unsafety marking down to the child, so that
1256  * we don't generate useless partial paths for it.
1257  */
1258  if (!rel->consider_parallel)
1259  childrel->consider_parallel = false;
1260 
1261  /*
1262  * Compute the child's access paths.
1263  */
1264  set_rel_pathlist(root, childrel, childRTindex, childRTE);
1265 
1266  /*
1267  * If child is dummy, ignore it.
1268  */
1269  if (IS_DUMMY_REL(childrel))
1270  continue;
1271 
1272  /* Bubble up childrel's partitioned children. */
1273  if (rel->part_scheme)
1274  rel->partitioned_child_rels =
1275  list_concat(rel->partitioned_child_rels,
1276  childrel->partitioned_child_rels);
1277 
1278  /*
1279  * Child is live, so add it to the live_childrels list for use below.
1280  */
1281  live_childrels = lappend(live_childrels, childrel);
1282  }
1283 
1284  /* Add paths to the append relation. */
1285  add_paths_to_append_rel(root, rel, live_childrels);
1286 }
1287 
1288 
1289 /*
1290  * add_paths_to_append_rel
1291  * Generate paths for the given append relation given the set of non-dummy
1292  * child rels.
1293  *
1294  * The function collects all parameterizations and orderings supported by the
1295  * non-dummy children. For every such parameterization or ordering, it creates
1296  * an append path collecting one path from each non-dummy child with given
1297  * parameterization or ordering. Similarly it collects partial paths from
1298  * non-dummy children to create partial append paths.
1299  */
1300 void
1301 add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
1302  List *live_childrels)
1303 {
1304  List *subpaths = NIL;
1305  bool subpaths_valid = true;
1306  List *partial_subpaths = NIL;
1307  List *pa_partial_subpaths = NIL;
1308  List *pa_nonpartial_subpaths = NIL;
1309  bool partial_subpaths_valid = true;
1310  bool pa_subpaths_valid;
1311  List *all_child_pathkeys = NIL;
1312  List *all_child_outers = NIL;
1313  ListCell *l;
1314  List *partitioned_rels = NIL;
1315  double partial_rows = -1;
1316 
1317  /* If appropriate, consider parallel append */
1318  pa_subpaths_valid = enable_parallel_append && rel->consider_parallel;
1319 
1320  /*
1321  * AppendPath generated for partitioned tables must record the RT indexes
1322  * of partitioned tables that are direct or indirect children of this
1323  * Append rel.
1324  *
1325  * AppendPath may be for a sub-query RTE (UNION ALL), in which case, 'rel'
1326  * itself does not represent a partitioned relation, but the child sub-
1327  * queries may contain references to partitioned relations. The loop
1328  * below will look for such children and collect them in a list to be
1329  * passed to the path creation function. (This assumes that we don't need
1330  * to look through multiple levels of subquery RTEs; if we ever do, we
1331  * could consider stuffing the list we generate here into sub-query RTE's
1332  * RelOptInfo, just like we do for partitioned rels, which would be used
1333  * when populating our parent rel with paths. For the present, that
1334  * appears to be unnecessary.)
1335  */
1336  if (rel->part_scheme != NULL)
1337  {
1338  if (IS_SIMPLE_REL(rel))
1339  partitioned_rels = list_make1(rel->partitioned_child_rels);
1340  else if (IS_JOIN_REL(rel))
1341  {
1342  int relid = -1;
1343  List *partrels = NIL;
1344 
1345  /*
1346  * For a partitioned joinrel, concatenate the component rels'
1347  * partitioned_child_rels lists.
1348  */
1349  while ((relid = bms_next_member(rel->relids, relid)) >= 0)
1350  {
1351  RelOptInfo *component;
1352 
1353  Assert(relid >= 1 && relid < root->simple_rel_array_size);
1354  component = root->simple_rel_array[relid];
1355  Assert(component->part_scheme != NULL);
1356  Assert(list_length(component->partitioned_child_rels) >= 1);
1357  partrels = list_concat(partrels,
1358  component->partitioned_child_rels);
1359  }
1360 
1361  partitioned_rels = list_make1(partrels);
1362  }
1363 
1364  Assert(list_length(partitioned_rels) >= 1);
1365  }
1366 
1367  /*
1368  * For every non-dummy child, remember the cheapest path. Also, identify
1369  * all pathkeys (orderings) and parameterizations (required_outer sets)
1370  * available for the non-dummy member relations.
1371  */
1372  foreach(l, live_childrels)
1373  {
1374  RelOptInfo *childrel = lfirst(l);
1375  ListCell *lcp;
1376  Path *cheapest_partial_path = NULL;
1377 
1378  /*
1379  * For UNION ALLs with non-empty partitioned_child_rels, accumulate
1380  * the Lists of child relations.
1381  */
1382  if (rel->rtekind == RTE_SUBQUERY && childrel->partitioned_child_rels != NIL)
1383  partitioned_rels = lappend(partitioned_rels,
1384  childrel->partitioned_child_rels);
1385 
1386  /*
1387  * If child has an unparameterized cheapest-total path, add that to
1388  * the unparameterized Append path we are constructing for the parent.
1389  * If not, there's no workable unparameterized path.
1390  *
1391  * With partitionwise aggregates, the child rel's pathlist may be
1392  * empty, so don't assume that a path exists here.
1393  */
1394  if (childrel->pathlist != NIL &&
1395  childrel->cheapest_total_path->param_info == NULL)
1396  accumulate_append_subpath(childrel->cheapest_total_path,
1397  &subpaths, NULL);
1398  else
1399  subpaths_valid = false;
1400 
1401  /* Same idea, but for a partial plan. */
1402  if (childrel->partial_pathlist != NIL)
1403  {
1404  cheapest_partial_path = linitial(childrel->partial_pathlist);
1405  accumulate_append_subpath(cheapest_partial_path,
1406  &partial_subpaths, NULL);
1407  }
1408  else
1409  partial_subpaths_valid = false;
1410 
1411  /*
1412  * Same idea, but for a parallel append mixing partial and non-partial
1413  * paths.
1414  */
1415  if (pa_subpaths_valid)
1416  {
1417  Path *nppath = NULL;
1418 
1419  nppath =
1420  get_cheapest_parallel_safe_total_inner(childrel->pathlist);
1421 
1422  if (cheapest_partial_path == NULL && nppath == NULL)
1423  {
1424  /* Neither a partial nor a parallel-safe path? Forget it. */
1425  pa_subpaths_valid = false;
1426  }
1427  else if (nppath == NULL ||
1428  (cheapest_partial_path != NULL &&
1429  cheapest_partial_path->total_cost < nppath->total_cost))
1430  {
1431  /* Partial path is cheaper or the only option. */
1432  Assert(cheapest_partial_path != NULL);
1433  accumulate_append_subpath(cheapest_partial_path,
1434  &pa_partial_subpaths,
1435  &pa_nonpartial_subpaths);
1436 
1437  }
1438  else
1439  {
1440  /*
1441  * Either we've got only a non-partial path, or we think that
1442  * a single backend can execute the best non-partial path
1443  * faster than all the parallel backends working together can
1444  * execute the best partial path.
1445  *
1446  * It might make sense to be more aggressive here. Even if
1447  * the best non-partial path is more expensive than the best
1448  * partial path, it could still be better to choose the
1449  * non-partial path if there are several such paths that can
1450  * be given to different workers. For now, we don't try to
1451  * figure that out.
1452  */
1453  accumulate_append_subpath(nppath,
1454  &pa_nonpartial_subpaths,
1455  NULL);
1456  }
1457  }
1458 
1459  /*
1460  * Collect lists of all the available path orderings and
1461  * parameterizations for all the children. We use these as a
1462  * heuristic to indicate which sort orderings and parameterizations we
1463  * should build Append and MergeAppend paths for.
1464  */
1465  foreach(lcp, childrel->pathlist)
1466  {
1467  Path *childpath = (Path *) lfirst(lcp);
1468  List *childkeys = childpath->pathkeys;
1469  Relids childouter = PATH_REQ_OUTER(childpath);
1470 
1471  /* Unsorted paths don't contribute to pathkey list */
1472  if (childkeys != NIL)
1473  {
1474  ListCell *lpk;
1475  bool found = false;
1476 
1477  /* Have we already seen this ordering? */
1478  foreach(lpk, all_child_pathkeys)
1479  {
1480  List *existing_pathkeys = (List *) lfirst(lpk);
1481 
1482  if (compare_pathkeys(existing_pathkeys,
1483  childkeys) == PATHKEYS_EQUAL)
1484  {
1485  found = true;
1486  break;
1487  }
1488  }
1489  if (!found)
1490  {
1491  /* No, so add it to all_child_pathkeys */
1492  all_child_pathkeys = lappend(all_child_pathkeys,
1493  childkeys);
1494  }
1495  }
1496 
1497  /* Unparameterized paths don't contribute to param-set list */
1498  if (childouter)
1499  {
1500  ListCell *lco;
1501  bool found = false;
1502 
1503  /* Have we already seen this param set? */
1504  foreach(lco, all_child_outers)
1505  {
1506  Relids existing_outers = (Relids) lfirst(lco);
1507 
1508  if (bms_equal(existing_outers, childouter))
1509  {
1510  found = true;
1511  break;
1512  }
1513  }
1514  if (!found)
1515  {
1516  /* No, so add it to all_child_outers */
1517  all_child_outers = lappend(all_child_outers,
1518  childouter);
1519  }
1520  }
1521  }
1522  }
1523 
1524  /*
1525  * If we found unparameterized paths for all children, build an unordered,
1526  * unparameterized Append path for the rel. (Note: this is correct even
1527  * if we have zero or one live subpath due to constraint exclusion.)
1528  */
1529  if (subpaths_valid)
1530  add_path(rel, (Path *) create_append_path(root, rel, subpaths, NIL,
1531  NIL, NULL, 0, false,
1532  partitioned_rels, -1));
1533 
1534  /*
1535  * Consider an append of unordered, unparameterized partial paths. Make
1536  * it parallel-aware if possible.
1537  */
1538  if (partial_subpaths_valid && partial_subpaths != NIL)
1539  {
1540  AppendPath *appendpath;
1541  ListCell *lc;
1542  int parallel_workers = 0;
1543 
1544  /* Find the highest number of workers requested for any subpath. */
1545  foreach(lc, partial_subpaths)
1546  {
1547  Path *path = lfirst(lc);
1548 
1549  parallel_workers = Max(parallel_workers, path->parallel_workers);
1550  }
1551  Assert(parallel_workers > 0);
1552 
1553  /*
1554  * If the use of parallel append is permitted, always request at least
1555  * log2(# of children) workers. We assume it can be useful to have
1556  * extra workers in this case because they will be spread out across
1557  * the children. The precise formula is just a guess, but we don't
1558  * want to end up with a radically different answer for a table with N
1559  * partitions vs. an unpartitioned table with the same data, so the
1560  * use of some kind of log-scaling here seems to make some sense.
1561  */
1562  if (enable_parallel_append)
1563  {
1564  parallel_workers = Max(parallel_workers,
1565  fls(list_length(live_childrels)));
1566  parallel_workers = Min(parallel_workers,
1567  max_parallel_workers_per_gather);
1568  }
1569  Assert(parallel_workers > 0);
1570 
1571  /* Generate a partial append path. */
1572  appendpath = create_append_path(root, rel, NIL, partial_subpaths,
1573  NIL, NULL, parallel_workers,
1574  enable_parallel_append,
1575  partitioned_rels, -1);
1576 
1577  /*
1578  * Make sure any subsequent partial paths use the same row count
1579  * estimate.
1580  */
1581  partial_rows = appendpath->path.rows;
1582 
1583  /* Add the path. */
1584  add_partial_path(rel, (Path *) appendpath);
1585  }
1586 
1587  /*
1588  * Consider a parallel-aware append using a mix of partial and non-partial
1589  * paths. (This only makes sense if there's at least one child which has
1590  * a non-partial path that is substantially cheaper than any partial path;
1591  * otherwise, we should use the append path added in the previous step.)
1592  */
1593  if (pa_subpaths_valid && pa_nonpartial_subpaths != NIL)
1594  {
1595  AppendPath *appendpath;
1596  ListCell *lc;
1597  int parallel_workers = 0;
1598 
1599  /*
1600  * Find the highest number of workers requested for any partial
1601  * subpath.
1602  */
1603  foreach(lc, pa_partial_subpaths)
1604  {
1605  Path *path = lfirst(lc);
1606 
1607  parallel_workers = Max(parallel_workers, path->parallel_workers);
1608  }
1609 
1610  /*
1611  * Same formula here as above. It's even more important in this
1612  * instance because the non-partial paths won't contribute anything to
1613  * the planned number of parallel workers.
1614  */
1615  parallel_workers = Max(parallel_workers,
1616  fls(list_length(live_childrels)));
1617  parallel_workers = Min(parallel_workers,
1618  max_parallel_workers_per_gather);
1619  Assert(parallel_workers > 0);
1620 
1621  appendpath = create_append_path(root, rel, pa_nonpartial_subpaths,
1622  pa_partial_subpaths,
1623  NIL, NULL, parallel_workers, true,
1624  partitioned_rels, partial_rows);
1625  add_partial_path(rel, (Path *) appendpath);
1626  }
1627 
1628  /*
1629  * Also build unparameterized ordered append paths based on the collected
1630  * list of child pathkeys.
1631  */
1632  if (subpaths_valid)
1633  generate_orderedappend_paths(root, rel, live_childrels,
1634  all_child_pathkeys,
1635  partitioned_rels);
1636 
1637  /*
1638  * Build Append paths for each parameterization seen among the child rels.
1639  * (This may look pretty expensive, but in most cases of practical
1640  * interest, the child rels will expose mostly the same parameterizations,
1641  * so that not that many cases actually get considered here.)
1642  *
1643  * The Append node itself cannot enforce quals, so all qual checking must
1644  * be done in the child paths. This means that to have a parameterized
1645  * Append path, we must have the exact same parameterization for each
1646  * child path; otherwise some children might be failing to check the
1647  * moved-down quals. To make them match up, we can try to increase the
1648  * parameterization of lesser-parameterized paths.
1649  */
1650  foreach(l, all_child_outers)
1651  {
1652  Relids required_outer = (Relids) lfirst(l);
1653  ListCell *lcr;
1654 
1655  /* Select the child paths for an Append with this parameterization */
1656  subpaths = NIL;
1657  subpaths_valid = true;
1658  foreach(lcr, live_childrels)
1659  {
1660  RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1661  Path *subpath;
1662 
1663  if (childrel->pathlist == NIL)
1664  {
1665  /* failed to make a suitable path for this child */
1666  subpaths_valid = false;
1667  break;
1668  }
1669 
1670  subpath = get_cheapest_parameterized_child_path(root,
1671  childrel,
1672  required_outer);
1673  if (subpath == NULL)
1674  {
1675  /* failed to make a suitable path for this child */
1676  subpaths_valid = false;
1677  break;
1678  }
1679  accumulate_append_subpath(subpath, &subpaths, NULL);
1680  }
1681 
1682  if (subpaths_valid)
1683  add_path(rel, (Path *)
1684  create_append_path(root, rel, subpaths, NIL,
1685  NIL, required_outer, 0, false,
1686  partitioned_rels, -1));
1687  }
1688 
1689  /*
1690  * When there is only a single child relation, the Append path can inherit
1691  * any ordering available for the child rel's path, so that it's useful to
1692  * consider ordered partial paths. Above we only considered the cheapest
1693  * partial path for each child, but let's also make paths using any
1694  * partial paths that have pathkeys.
1695  */
1696  if (list_length(live_childrels) == 1)
1697  {
1698  RelOptInfo *childrel = (RelOptInfo *) linitial(live_childrels);
1699 
1700  foreach(l, childrel->partial_pathlist)
1701  {
1702  Path *path = (Path *) lfirst(l);
1703  AppendPath *appendpath;
1704 
1705  /*
1706  * Skip paths with no pathkeys. Also skip the cheapest partial
1707  * path, since we already used that above.
1708  */
1709  if (path->pathkeys == NIL ||
1710  path == linitial(childrel->partial_pathlist))
1711  continue;
1712 
1713  appendpath = create_append_path(root, rel, NIL, list_make1(path),
1714  NIL, NULL,
1715  path->parallel_workers, true,
1716  partitioned_rels, partial_rows);
1717  add_partial_path(rel, (Path *) appendpath);
1718  }
1719  }
1720 }
1721 
1722 /*
1723  * generate_orderedappend_paths
1724  * Generate ordered append paths for an append relation
1725  *
1726  * Usually we generate MergeAppend paths here, but there are some special
1727  * cases where we can generate simple Append paths, because the subpaths
1728  * can provide tuples in the required order already.
1729  *
1730  * We generate a path for each ordering (pathkey list) appearing in
1731  * all_child_pathkeys.
1732  *
1733  * We consider both cheapest-startup and cheapest-total cases, ie, for each
1734  * interesting ordering, collect all the cheapest startup subpaths and all the
1735  * cheapest total paths, and build a suitable path for each case.
1736  *
1737  * We don't currently generate any parameterized ordered paths here. While
1738  * it would not take much more code here to do so, it's very unclear that it
1739  * is worth the planning cycles to investigate such paths: there's little
1740  * use for an ordered path on the inside of a nestloop. In fact, it's likely
1741  * that the current coding of add_path would reject such paths out of hand,
1742  * because add_path gives no credit for sort ordering of parameterized paths,
1743  * and a parameterized MergeAppend is going to be more expensive than the
1744  * corresponding parameterized Append path. If we ever try harder to support
1745  * parameterized mergejoin plans, it might be worth adding support for
1746  * parameterized paths here to feed such joins. (See notes in
1747  * optimizer/README for why that might not ever happen, though.)
1748  */
1749 static void
1750 generate_orderedappend_paths(PlannerInfo *root, RelOptInfo *rel,
1751  List *live_childrels,
1752  List *all_child_pathkeys,
1753  List *partitioned_rels)
1754 {
1755  ListCell *lcp;
1756  List *partition_pathkeys = NIL;
1757  List *partition_pathkeys_desc = NIL;
1758  bool partition_pathkeys_partial = true;
1759  bool partition_pathkeys_desc_partial = true;
1760 
1761  /*
1762  * Some partitioned table setups may allow us to use an Append node
1763  * instead of a MergeAppend. This is possible in cases such as RANGE
1764  * partitioned tables where it's guaranteed that an earlier partition must
1765  * contain rows which come earlier in the sort order. To detect whether
1766  * this is relevant, build pathkey descriptions of the partition ordering,
1767  * for both forward and reverse scans.
1768  */
1769  if (rel->part_scheme != NULL && IS_SIMPLE_REL(rel) &&
1770  partitions_are_ordered(rel->boundinfo, rel->nparts))
1771  {
1772  partition_pathkeys = build_partition_pathkeys(root, rel,
1773  ForwardScanDirection,
1774  &partition_pathkeys_partial);
1775 
1776  partition_pathkeys_desc = build_partition_pathkeys(root, rel,
1777  BackwardScanDirection,
1778  &partition_pathkeys_desc_partial);
1779 
1780  /*
1781  * You might think we should truncate_useless_pathkeys here, but
1782  * allowing partition keys which are a subset of the query's pathkeys
1783  * can often be useful. For example, consider a table partitioned by
1784  * RANGE (a, b), and a query with ORDER BY a, b, c. If we have child
1785  * paths that can produce the a, b, c ordering (perhaps via indexes on
1786  * (a, b, c)) then it works to consider the appendrel output as
1787  * ordered by a, b, c.
1788  */
1789  }
1790 
1791  /* Now consider each interesting sort ordering */
1792  foreach(lcp, all_child_pathkeys)
1793  {
1794  List *pathkeys = (List *) lfirst(lcp);
1795  List *startup_subpaths = NIL;
1796  List *total_subpaths = NIL;
1797  bool startup_neq_total = false;
1798  ListCell *lcr;
1799  bool match_partition_order;
1800  bool match_partition_order_desc;
1801 
1802  /*
1803  * Determine if this sort ordering matches any partition pathkeys we
1804  * have, for both ascending and descending partition order. If the
1805  * partition pathkeys happen to be contained in pathkeys then it still
1806  * works, as described above, providing that the partition pathkeys
1807  * are complete and not just a prefix of the partition keys. (In such
1808  * cases we'll be relying on the child paths to have sorted the
1809  * lower-order columns of the required pathkeys.)
1810  */
1811  match_partition_order =
1812  pathkeys_contained_in(pathkeys, partition_pathkeys) ||
1813  (!partition_pathkeys_partial &&
1814  pathkeys_contained_in(partition_pathkeys, pathkeys));
1815 
1816  match_partition_order_desc = !match_partition_order &&
1817  (pathkeys_contained_in(pathkeys, partition_pathkeys_desc) ||
1818  (!partition_pathkeys_desc_partial &&
1819  pathkeys_contained_in(partition_pathkeys_desc, pathkeys)));
1820 
1821  /* Select the child paths for this ordering... */
1822  foreach(lcr, live_childrels)
1823  {
1824  RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
1825  Path *cheapest_startup,
1826  *cheapest_total;
1827 
1828  /* Locate the right paths, if they are available. */
1829  cheapest_startup =
1830  get_cheapest_path_for_pathkeys(childrel->pathlist,
1831  pathkeys,
1832  NULL,
1833  STARTUP_COST,
1834  false);
1835  cheapest_total =
1836  get_cheapest_path_for_pathkeys(childrel->pathlist,
1837  pathkeys,
1838  NULL,
1839  TOTAL_COST,
1840  false);
1841 
1842  /*
1843  * If we can't find any paths with the right order just use the
1844  * cheapest-total path; we'll have to sort it later.
1845  */
1846  if (cheapest_startup == NULL || cheapest_total == NULL)
1847  {
1848  cheapest_startup = cheapest_total =
1849  childrel->cheapest_total_path;
1850  /* Assert we do have an unparameterized path for this child */
1851  Assert(cheapest_total->param_info == NULL);
1852  }
1853 
1854  /*
1855  * Notice whether we actually have different paths for the
1856  * "cheapest" and "total" cases; frequently there will be no point
1857  * in two create_merge_append_path() calls.
1858  */
1859  if (cheapest_startup != cheapest_total)
1860  startup_neq_total = true;
1861 
1862  /*
1863  * Collect the appropriate child paths. The required logic varies
1864  * for the Append and MergeAppend cases.
1865  */
1866  if (match_partition_order)
1867  {
1868  /*
1869  * We're going to make a plain Append path. We don't need
1870  * most of what accumulate_append_subpath would do, but we do
1871  * want to cut out child Appends or MergeAppends if they have
1872  * just a single subpath (and hence aren't doing anything
1873  * useful).
1874  */
1875  cheapest_startup = get_singleton_append_subpath(cheapest_startup);
1876  cheapest_total = get_singleton_append_subpath(cheapest_total);
1877 
1878  startup_subpaths = lappend(startup_subpaths, cheapest_startup);
1879  total_subpaths = lappend(total_subpaths, cheapest_total);
1880  }
1881  else if (match_partition_order_desc)
1882  {
1883  /*
1884  * As above, but we need to reverse the order of the children,
1885  * because nodeAppend.c doesn't know anything about reverse
1886  * ordering and will scan the children in the order presented.
1887  */
1888  cheapest_startup = get_singleton_append_subpath(cheapest_startup);
1889  cheapest_total = get_singleton_append_subpath(cheapest_total);
1890 
1891  startup_subpaths = lcons(cheapest_startup, startup_subpaths);
1892  total_subpaths = lcons(cheapest_total, total_subpaths);
1893  }
1894  else
1895  {
1896  /*
1897  * Otherwise, rely on accumulate_append_subpath to collect the
1898  * child paths for the MergeAppend.
1899  */
1900  accumulate_append_subpath(cheapest_startup,
1901  &startup_subpaths, NULL);
1902  accumulate_append_subpath(cheapest_total,
1903  &total_subpaths, NULL);
1904  }
1905  }
1906 
1907  /* ... and build the Append or MergeAppend paths */
1908  if (match_partition_order || match_partition_order_desc)
1909  {
1910  /* We only need Append */
1911  add_path(rel, (Path *) create_append_path(root,
1912  rel,
1913  startup_subpaths,
1914  NIL,
1915  pathkeys,
1916  NULL,
1917  0,
1918  false,
1919  partitioned_rels,
1920  -1));
1921  if (startup_neq_total)
1922  add_path(rel, (Path *) create_append_path(root,
1923  rel,
1924  total_subpaths,
1925  NIL,
1926  pathkeys,
1927  NULL,
1928  0,
1929  false,
1930  partitioned_rels,
1931  -1));
1932  }
1933  else
1934  {
1935  /* We need MergeAppend */
1936  add_path(rel, (Path *) create_merge_append_path(root,
1937  rel,
1938  startup_subpaths,
1939  pathkeys,
1940  NULL,
1941  partitioned_rels));
1942  if (startup_neq_total)
1943  add_path(rel, (Path *) create_merge_append_path(root,
1944  rel,
1945  total_subpaths,
1946  pathkeys,
1947  NULL,
1948  partitioned_rels));
1949  }
1950  }
1951 }
1952 
1953 /*
1954  * get_cheapest_parameterized_child_path
1955  * Get cheapest path for this relation that has exactly the requested
1956  * parameterization.
1957  *
1958  * Returns NULL if unable to create such a path.
1959  */
1960 static Path *
1961 get_cheapest_parameterized_child_path(PlannerInfo *root, RelOptInfo *rel,
1962  Relids required_outer)
1963 {
1964  Path *cheapest;
1965  ListCell *lc;
1966 
1967  /*
1968  * Look up the cheapest existing path with no more than the needed
1969  * parameterization. If it has exactly the needed parameterization, we're
1970  * done.
1971  */
1972  cheapest = get_cheapest_path_for_pathkeys(rel->pathlist,
1973  NIL,
1974  required_outer,
1975  TOTAL_COST,
1976  false);
1977  Assert(cheapest != NULL);
1978  if (bms_equal(PATH_REQ_OUTER(cheapest), required_outer))
1979  return cheapest;
1980 
1981  /*
1982  * Otherwise, we can "reparameterize" an existing path to match the given
1983  * parameterization, which effectively means pushing down additional
1984  * joinquals to be checked within the path's scan. However, some existing
1985  * paths might check the available joinquals already while others don't;
1986  * therefore, it's not clear which existing path will be cheapest after
1987  * reparameterization. We have to go through them all and find out.
1988  */
1989  cheapest = NULL;
1990  foreach(lc, rel->pathlist)
1991  {
1992  Path *path = (Path *) lfirst(lc);
1993 
1994  /* Can't use it if it needs more than requested parameterization */
1995  if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
1996  continue;
1997 
1998  /*
1999  * Reparameterization can only increase the path's cost, so if it's
2000  * already more expensive than the current cheapest, forget it.
2001  */
2002  if (cheapest != NULL &&
2003  compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
2004  continue;
2005 
2006  /* Reparameterize if needed, then recheck cost */
2007  if (!bms_equal(PATH_REQ_OUTER(path), required_outer))
2008  {
2009  path = reparameterize_path(root, path, required_outer, 1.0);
2010  if (path == NULL)
2011  continue; /* failed to reparameterize this one */
2012  Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));
2013 
2014  if (cheapest != NULL &&
2015  compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
2016  continue;
2017  }
2018 
2019  /* We have a new best path */
2020  cheapest = path;
2021  }
2022 
2023  /* Return the best path, or NULL if we found no suitable candidate */
2024  return cheapest;
2025 }
2026 
2027 /*
2028  * accumulate_append_subpath
2029  * Add a subpath to the list being built for an Append or MergeAppend.
2030  *
2031  * It's possible that the child is itself an Append or MergeAppend path, in
2032  * which case we can "cut out the middleman" and just add its child paths to
2033  * our own list. (We don't try to do this earlier because we need to apply
2034  * both levels of transformation to the quals.)
2035  *
2036  * Note that if we omit a child MergeAppend in this way, we are effectively
2037  * omitting a sort step, which seems fine: if the parent is to be an Append,
2038  * its result would be unsorted anyway, while if the parent is to be a
2039  * MergeAppend, there's no point in a separate sort on a child.
2040  *
2041  * Normally, either path is a partial path and subpaths is a list of partial
2042  * paths, or else path is a non-partial plan and subpaths is a list of those.
2043  * However, if path is a parallel-aware Append, then we add its partial path
2044  * children to subpaths and the rest to special_subpaths. If the latter is
2045  * NULL, we don't flatten the path at all (unless it contains only partial
2046  * paths).
2047  */
2048 static void
2049 accumulate_append_subpath(Path *path, List **subpaths, List **special_subpaths)
2050 {
2051  if (IsA(path, AppendPath))
2052  {
2053  AppendPath *apath = (AppendPath *) path;
2054 
2055  if (!apath->path.parallel_aware || apath->first_partial_path == 0)
2056  {
2057  *subpaths = list_concat(*subpaths, apath->subpaths);
2058  return;
2059  }
2060  else if (special_subpaths != NULL)
2061  {
2062  List *new_special_subpaths;
2063 
2064  /* Split Parallel Append into partial and non-partial subpaths */
2065  *subpaths = list_concat(*subpaths,
2066  list_copy_tail(apath->subpaths,
2067  apath->first_partial_path));
2068  new_special_subpaths =
2069  list_truncate(list_copy(apath->subpaths),
2070  apath->first_partial_path);
2071  *special_subpaths = list_concat(*special_subpaths,
2072  new_special_subpaths);
2073  return;
2074  }
2075  }
2076  else if (IsA(path, MergeAppendPath))
2077  {
2078  MergeAppendPath *mpath = (MergeAppendPath *) path;
2079 
2080  *subpaths = list_concat(*subpaths, mpath->subpaths);
2081  return;
2082  }
2083 
2084  *subpaths = lappend(*subpaths, path);
2085 }
2086 
2087 /*
2088  * get_singleton_append_subpath
2089  * Returns the single subpath of an Append/MergeAppend, or just
2090  * return 'path' if it's not a single sub-path Append/MergeAppend.
2091  *
2092  * Note: 'path' must not be a parallel-aware path.
2093  */
2094 static Path *
2095 get_singleton_append_subpath(Path *path)
2096 {
2097  Assert(!path->parallel_aware);
2098 
2099  if (IsA(path, AppendPath))
2100  {
2101  AppendPath *apath = (AppendPath *) path;
2102 
2103  if (list_length(apath->subpaths) == 1)
2104  return (Path *) linitial(apath->subpaths);
2105  }
2106  else if (IsA(path, MergeAppendPath))
2107  {
2108  MergeAppendPath *mpath = (MergeAppendPath *) path;
2109 
2110  if (list_length(mpath->subpaths) == 1)
2111  return (Path *) linitial(mpath->subpaths);
2112  }
2113 
2114  return path;
2115 }
2116 
2117 /*
2118  * set_dummy_rel_pathlist
2119  * Build a dummy path for a relation that's been excluded by constraints
2120  *
2121  * Rather than inventing a special "dummy" path type, we represent this as an
2122  * AppendPath with no members (see also IS_DUMMY_APPEND/IS_DUMMY_REL macros).
2123  *
2124  * (See also mark_dummy_rel, which does basically the same thing, but is
2125  * typically used to change a rel into dummy state after we already made
2126  * paths for it.)
2127  */
2128 static void
2129 set_dummy_rel_pathlist(RelOptInfo *rel)
2130 {
2131  /* Set dummy size estimates --- we leave attr_widths[] as zeroes */
2132  rel->rows = 0;
2133  rel->reltarget->width = 0;
2134 
2135  /* Discard any pre-existing paths; no further need for them */
2136  rel->pathlist = NIL;
2137  rel->partial_pathlist = NIL;
2138 
2139  /* Set up the dummy path */
2140  add_path(rel, (Path *) create_append_path(NULL, rel, NIL, NIL,
2141  NIL, rel->lateral_relids,
2142  0, false, NIL, -1));
2143 
2144  /*
2145  * We set the cheapest-path fields immediately, just in case they were
2146  * pointing at some discarded path. This is redundant when we're called
2147  * from set_rel_size(), but not when called from elsewhere, and doing it
2148  * twice is harmless anyway.
2149  */
2150  set_cheapest(rel);
2151 }
2152 
2153 /* quick-and-dirty test to see if any joining is needed */
2154 static bool
2155 has_multiple_baserels(PlannerInfo *root)
2156 {
2157  int num_base_rels = 0;
2158  Index rti;
2159 
2160  for (rti = 1; rti < root->simple_rel_array_size; rti++)
2161  {
2162  RelOptInfo *brel = root->simple_rel_array[rti];
2163 
2164  if (brel == NULL)
2165  continue;
2166 
2167  /* ignore RTEs that are "other rels" */
2168  if (brel->reloptkind == RELOPT_BASEREL)
2169  if (++num_base_rels > 1)
2170  return true;
2171  }
2172  return false;
2173 }
2174 
2175 /*
2176  * set_subquery_pathlist
2177  * Generate SubqueryScan access paths for a subquery RTE
2178  *
2179  * We don't currently support generating parameterized paths for subqueries
2180  * by pushing join clauses down into them; it seems too expensive to re-plan
2181  * the subquery multiple times to consider different alternatives.
2182  * (XXX that could stand to be reconsidered, now that we use Paths.)
2183  * So the paths made here will be parameterized if the subquery contains
2184  * LATERAL references, otherwise not. As long as that's true, there's no need
2185  * for a separate set_subquery_size phase: just make the paths right away.
2186  */
2187 static void
2188 set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
2189  Index rti, RangeTblEntry *rte)
2190 {
2191  Query *parse = root->parse;
2192  Query *subquery = rte->subquery;
2193  Relids required_outer;
2194  pushdown_safety_info safetyInfo;
2195  double tuple_fraction;
2196  RelOptInfo *sub_final_rel;
2197  ListCell *lc;
2198 
2199  /*
2200  * Must copy the Query so that planning doesn't mess up the RTE contents
2201  * (really really need to fix the planner to not scribble on its input,
2202  * someday ... but see remove_unused_subquery_outputs to start with).
2203  */
2204  subquery = copyObject(subquery);
2205 
2206  /*
2207  * If it's a LATERAL subquery, it might contain some Vars of the current
2208  * query level, requiring it to be treated as parameterized, even though
2209  * we don't support pushing down join quals into subqueries.
2210  */
2211  required_outer = rel->lateral_relids;
2212 
2213  /*
2214  * Zero out result area for subquery_is_pushdown_safe, so that it can set
2215  * flags as needed while recursing. In particular, we need a workspace
2216  * for keeping track of unsafe-to-reference columns. unsafeColumns[i]
2217  * will be set true if we find that output column i of the subquery is
2218  * unsafe to use in a pushed-down qual.
2219  */
2220  memset(&safetyInfo, 0, sizeof(safetyInfo));
2221  safetyInfo.unsafeColumns = (bool *)
2222  palloc0((list_length(subquery->targetList) + 1) * sizeof(bool));
2223 
2224  /*
2225  * If the subquery has the "security_barrier" flag, it means the subquery
2226  * originated from a view that must enforce row level security. Then we
2227  * must not push down quals that contain leaky functions. (Ideally this
2228  * would be checked inside subquery_is_pushdown_safe, but since we don't
2229  * currently pass the RTE to that function, we must do it here.)
2230  */
2231  safetyInfo.unsafeLeaky = rte->security_barrier;
2232 
2233  /*
2234  * If there are any restriction clauses that have been attached to the
2235  * subquery relation, consider pushing them down to become WHERE or HAVING
2236  * quals of the subquery itself. This transformation is useful because it
2237  * may allow us to generate a better plan for the subquery than evaluating
2238  * all the subquery output rows and then filtering them.
2239  *
2240  * There are several cases where we cannot push down clauses. Restrictions
2241  * involving the subquery are checked by subquery_is_pushdown_safe().
2242  * Restrictions on individual clauses are checked by
2243  * qual_is_pushdown_safe(). Also, we don't want to push down
2244  * pseudoconstant clauses; better to have the gating node above the
2245  * subquery.
2246  *
2247  * Non-pushed-down clauses will get evaluated as qpquals of the
2248  * SubqueryScan node.
2249  *
2250  * XXX Are there any cases where we want to make a policy decision not to
2251  * push down a pushable qual, because it'd result in a worse plan?
2252  */
2253  if (rel->baserestrictinfo != NIL &&
2254  subquery_is_pushdown_safe(subquery, subquery, &safetyInfo))
2255  {
2256  /* OK to consider pushing down individual quals */
2257  List *upperrestrictlist = NIL;
2258  ListCell *l;
2259 
2260  foreach(l, rel->baserestrictinfo)
2261  {
2262  RestrictInfo *rinfo = (RestrictInfo *) lfirst(l);
2263  Node *clause = (Node *) rinfo->clause;
2264 
2265  if (!rinfo->pseudoconstant &&
2266  qual_is_pushdown_safe(subquery, rti, clause, &safetyInfo))
2267  {
2268  /* Push it down */
2269  subquery_push_qual(subquery, rte, rti, clause);
2270  }
2271  else
2272  {
2273  /* Keep it in the upper query */
2274  upperrestrictlist = lappend(upperrestrictlist, rinfo);
2275  }
2276  }
2277  rel->baserestrictinfo = upperrestrictlist;
2278  /* We don't bother recomputing baserestrict_min_security */
2279  }
2280 
2281  pfree(safetyInfo.unsafeColumns);
2282 
2283  /*
2284  * The upper query might not use all the subquery's output columns; if
2285  * not, we can simplify.
2286  */
2287  remove_unused_subquery_outputs(subquery, rel);
2288 
2289  /*
2290  * We can safely pass the outer tuple_fraction down to the subquery if the
2291  * outer level has no joining, aggregation, or sorting to do. Otherwise
2292  * we'd better tell the subquery to plan for full retrieval. (XXX This
2293  * could probably be made more intelligent ...)
2294  */
2295  if (parse->hasAggs ||
2296  parse->groupClause ||
2297  parse->groupingSets ||
2298  parse->havingQual ||
2299  parse->distinctClause ||
2300  parse->sortClause ||
2301  has_multiple_baserels(root))
2302  tuple_fraction = 0.0; /* default case */
2303  else
2304  tuple_fraction = root->tuple_fraction;
2305 
2306  /* plan_params should not be in use in current query level */
2307  Assert(root->plan_params == NIL);
2308 
2309  /* Generate a subroot and Paths for the subquery */
2310  rel->subroot = subquery_planner(root->glob, subquery,
2311  root,
2312  false, tuple_fraction);
2313 
2314  /* Isolate the params needed by this specific subplan */
2315  rel->subplan_params = root->plan_params;
2316  root->plan_params = NIL;
2317 
2318  /*
2319  * It's possible that constraint exclusion proved the subquery empty. If
2320  * so, it's desirable to produce an unadorned dummy path so that we will
2321  * recognize appropriate optimizations at this query level.
2322  */
2323  sub_final_rel = fetch_upper_rel(rel->subroot, UPPERREL_FINAL, NULL);
2324 
2325  if (IS_DUMMY_REL(sub_final_rel))
2326  {
2327  set_dummy_rel_pathlist(rel);
2328  return;
2329  }
2330 
2331  /*
2332  * Mark rel with estimated output rows, width, etc. Note that we have to
2333  * do this before generating outer-query paths, else cost_subqueryscan is
2334  * not happy.
2335  */
2336  set_subquery_size_estimates(root, rel);
2337 
2338  /*
2339  * For each Path that subquery_planner produced, make a SubqueryScanPath
2340  * in the outer query.
2341  */
2342  foreach(lc, sub_final_rel->pathlist)
2343  {
2344  Path *subpath = (Path *) lfirst(lc);
2345  List *pathkeys;
2346 
2347  /* Convert subpath's pathkeys to outer representation */
2348  pathkeys = convert_subquery_pathkeys(root,
2349  rel,
2350  subpath->pathkeys,
2351  make_tlist_from_pathtarget(subpath->pathtarget));
2352 
2353  /* Generate outer path using this subpath */
2354  add_path(rel, (Path *)
2355  create_subqueryscan_path(root, rel, subpath,
2356  pathkeys, required_outer));
2357  }
2358 
2359  /* If outer rel allows parallelism, do same for partial paths. */
2360  if (rel->consider_parallel && bms_is_empty(required_outer))
2361  {
2362  /* If consider_parallel is false, there should be no partial paths. */
2363  Assert(sub_final_rel->consider_parallel ||
2364  sub_final_rel->partial_pathlist == NIL);
2365 
2366  /* Same for partial paths. */
2367  foreach(lc, sub_final_rel->partial_pathlist)
2368  {
2369  Path *subpath = (Path *) lfirst(lc);
2370  List *pathkeys;
2371 
2372  /* Convert subpath's pathkeys to outer representation */
2373  pathkeys = convert_subquery_pathkeys(root,
2374  rel,
2375  subpath->pathkeys,
2376  make_tlist_from_pathtarget(subpath->pathtarget));
2377 
2378  /* Generate outer path using this subpath */
2379  add_partial_path(rel, (Path *)
2380  create_subqueryscan_path(root, rel, subpath,
2381  pathkeys,
2382  required_outer));
2383  }
2384  }
2385 }
2386 
2387 /*
2388  * set_function_pathlist
2389  * Build the (single) access path for a function RTE
2390  */
2391 static void
2392 set_function_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2393 {
2394  Relids required_outer;
2395  List *pathkeys = NIL;
2396 
2397  /*
2398  * We don't support pushing join clauses into the quals of a function
2399  * scan, but it could still have required parameterization due to LATERAL
2400  * refs in the function expression.
2401  */
2402  required_outer = rel->lateral_relids;
2403 
2404  /*
2405  * The result is considered unordered unless ORDINALITY was used, in which
2406  * case it is ordered by the ordinal column (the last one). See if we
2407  * care, by checking for uses of that Var in equivalence classes.
2408  */
2409  if (rte->funcordinality)
2410  {
2411  AttrNumber ordattno = rel->max_attr;
2412  Var *var = NULL;
2413  ListCell *lc;
2414 
2415  /*
2416  * Is there a Var for it in rel's targetlist? If not, the query did
2417  * not reference the ordinality column, or at least not in any way
2418  * that would be interesting for sorting.
2419  */
2420  foreach(lc, rel->reltarget->exprs)
2421  {
2422  Var *node = (Var *) lfirst(lc);
2423 
2424  /* checking varno/varlevelsup is just paranoia */
2425  if (IsA(node, Var) &&
2426  node->varattno == ordattno &&
2427  node->varno == rel->relid &&
2428  node->varlevelsup == 0)
2429  {
2430  var = node;
2431  break;
2432  }
2433  }
2434 
2435  /*
2436  * Try to build pathkeys for this Var with int8 sorting. We tell
2437  * build_expression_pathkey not to build any new equivalence class; if
2438  * the Var isn't already mentioned in some EC, it means that nothing
2439  * cares about the ordering.
2440  */
2441  if (var)
2442  pathkeys = build_expression_pathkey(root,
2443  (Expr *) var,
2444  NULL, /* below outer joins */
2445  Int8LessOperator,
2446  rel->relids,
2447  false);
2448  }
2449 
2450  /* Generate appropriate path */
2451  add_path(rel, create_functionscan_path(root, rel,
2452  pathkeys, required_outer));
2453 }
2454 
2455 /*
2456  * set_values_pathlist
2457  * Build the (single) access path for a VALUES RTE
2458  */
2459 static void
2460 set_values_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2461 {
2462  Relids required_outer;
2463 
2464  /*
2465  * We don't support pushing join clauses into the quals of a values scan,
2466  * but it could still have required parameterization due to LATERAL refs
2467  * in the values expressions.
2468  */
2469  required_outer = rel->lateral_relids;
2470 
2471  /* Generate appropriate path */
2472  add_path(rel, create_valuesscan_path(root, rel, required_outer));
2473 }
2474 
2475 /*
2476  * set_tablefunc_pathlist
2477  * Build the (single) access path for a table func RTE
2478  */
2479 static void
2480 set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2481 {
2482  Relids required_outer;
2483 
2484  /*
2485  * We don't support pushing join clauses into the quals of a tablefunc
2486  * scan, but it could still have required parameterization due to LATERAL
2487  * refs in the function expression.
2488  */
2489  required_outer = rel->lateral_relids;
2490 
2491  /* Generate appropriate path */
2492  add_path(rel, create_tablefuncscan_path(root, rel,
2493  required_outer));
2494 }
2495 
2496 /*
2497  * set_cte_pathlist
2498  * Build the (single) access path for a non-self-reference CTE RTE
2499  *
2500  * There's no need for a separate set_cte_size phase, since we don't
2501  * support join-qual-parameterized paths for CTEs.
2502  */
2503 static void
2504 set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2505 {
2506  Plan *cteplan;
2507  PlannerInfo *cteroot;
2508  Index levelsup;
2509  int ndx;
2510  ListCell *lc;
2511  int plan_id;
2512  Relids required_outer;
2513 
2514  /*
2515  * Find the referenced CTE, and locate the plan previously made for it.
2516  */
2517  levelsup = rte->ctelevelsup;
2518  cteroot = root;
2519  while (levelsup-- > 0)
2520  {
2521  cteroot = cteroot->parent_root;
2522  if (!cteroot) /* shouldn't happen */
2523  elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2524  }
2525 
2526  /*
2527  * Note: cte_plan_ids can be shorter than cteList, if we are still working
2528  * on planning the CTEs (ie, this is a side-reference from another CTE).
2529  * So we mustn't use forboth here.
2530  */
2531  ndx = 0;
2532  foreach(lc, cteroot->parse->cteList)
2533  {
2534  CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
2535 
2536  if (strcmp(cte->ctename, rte->ctename) == 0)
2537  break;
2538  ndx++;
2539  }
2540  if (lc == NULL) /* shouldn't happen */
2541  elog(ERROR, "could not find CTE \"%s\"", rte->ctename);
2542  if (ndx >= list_length(cteroot->cte_plan_ids))
2543  elog(ERROR, "could not find plan for CTE \"%s\"", rte->ctename);
2544  plan_id = list_nth_int(cteroot->cte_plan_ids, ndx);
2545  Assert(plan_id > 0);
2546  cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2547 
2548  /* Mark rel with estimated output rows, width, etc */
2549  set_cte_size_estimates(root, rel, cteplan->plan_rows);
2550 
2551  /*
2552  * We don't support pushing join clauses into the quals of a CTE scan, but
2553  * it could still have required parameterization due to LATERAL refs in
2554  * its tlist.
2555  */
2556  required_outer = rel->lateral_relids;
2557 
2558  /* Generate appropriate path */
2559  add_path(rel, create_ctescan_path(root, rel, required_outer));
2560 }
2561 
2562 /*
2563  * set_namedtuplestore_pathlist
2564  * Build the (single) access path for a named tuplestore RTE
2565  *
2566  * There's no need for a separate set_namedtuplestore_size phase, since we
2567  * don't support join-qual-parameterized paths for tuplestores.
2568  */
2569 static void
2570 set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel,
2571  RangeTblEntry *rte)
2572 {
2573  Relids required_outer;
2574 
2575  /* Mark rel with estimated output rows, width, etc */
2576  set_namedtuplestore_size_estimates(root, rel);
2577 
2578  /*
2579  * We don't support pushing join clauses into the quals of a tuplestore
2580  * scan, but it could still have required parameterization due to LATERAL
2581  * refs in its tlist.
2582  */
2583  required_outer = rel->lateral_relids;
2584 
2585  /* Generate appropriate path */
2586  add_path(rel, create_namedtuplestorescan_path(root, rel, required_outer));
2587 
2588  /* Select cheapest path (pretty easy in this case...) */
2589  set_cheapest(rel);
2590 }
2591 
2592 /*
2593  * set_result_pathlist
2594  * Build the (single) access path for an RTE_RESULT RTE
2595  *
2596  * There's no need for a separate set_result_size phase, since we
2597  * don't support join-qual-parameterized paths for these RTEs.
2598  */
2599 static void
2600 set_result_pathlist(PlannerInfo *root, RelOptInfo *rel,
2601  RangeTblEntry *rte)
2602 {
2603  Relids required_outer;
2604 
2605  /* Mark rel with estimated output rows, width, etc */
2606  set_result_size_estimates(root, rel);
2607 
2608  /*
2609  * We don't support pushing join clauses into the quals of a Result scan,
2610  * but it could still have required parameterization due to LATERAL refs
2611  * in its tlist.
2612  */
2613  required_outer = rel->lateral_relids;
2614 
2615  /* Generate appropriate path */
2616  add_path(rel, create_resultscan_path(root, rel, required_outer));
2617 
2618  /* Select cheapest path (pretty easy in this case...) */
2619  set_cheapest(rel);
2620 }
2621 
2622 /*
2623  * set_worktable_pathlist
2624  * Build the (single) access path for a self-reference CTE RTE
2625  *
2626  * There's no need for a separate set_worktable_size phase, since we don't
2627  * support join-qual-parameterized paths for CTEs.
2628  */
2629 static void
2630 set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
2631 {
2632  Path *ctepath;
2633  PlannerInfo *cteroot;
2634  Index levelsup;
2635  Relids required_outer;
2636 
2637  /*
2638  * We need to find the non-recursive term's path, which is in the plan
2639  * level that's processing the recursive UNION, which is one level *below*
2640  * where the CTE comes from.
2641  */
2642  levelsup = rte->ctelevelsup;
2643  if (levelsup == 0) /* shouldn't happen */
2644  elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2645  levelsup--;
2646  cteroot = root;
2647  while (levelsup-- > 0)
2648  {
2649  cteroot = cteroot->parent_root;
2650  if (!cteroot) /* shouldn't happen */
2651  elog(ERROR, "bad levelsup for CTE \"%s\"", rte->ctename);
2652  }
2653  ctepath = cteroot->non_recursive_path;
2654  if (!ctepath) /* shouldn't happen */
2655  elog(ERROR, "could not find path for CTE \"%s\"", rte->ctename);
2656 
2657  /* Mark rel with estimated output rows, width, etc */
2658  set_cte_size_estimates(root, rel, ctepath->rows);
2659 
2660  /*
2661  * We don't support pushing join clauses into the quals of a worktable
2662  * scan, but it could still have required parameterization due to LATERAL
2663  * refs in its tlist. (I'm not sure this is actually possible given the
2664  * restrictions on recursive references, but it's easy enough to support.)
2665  */
2666  required_outer = rel->lateral_relids;
2667 
2668  /* Generate appropriate path */
2669  add_path(rel, create_worktablescan_path(root, rel, required_outer));
2670 }
2671 
2672 /*
2673  * generate_gather_paths
2674  * Generate parallel access paths for a relation by pushing a Gather or
2675  * Gather Merge on top of a partial path.
2676  *
2677  * This must not be called until after we're done creating all partial paths
2678  * for the specified relation. (Otherwise, add_partial_path might delete a
2679  * path that some GatherPath or GatherMergePath has a reference to.)
2680  *
2681  * If we're generating paths for a scan or join relation, override_rows will
2682  * be false, and we'll just use the relation's size estimate. When we're
2683  * being called for a partially-grouped path, though, we need to override
2684  * the rowcount estimate. (It's not clear that the particular value we're
2685  * using here is actually best, but the underlying rel has no estimate so
2686  * we must do something.)
2687  */
2688 void
2689 generate_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
2690 {
2691  Path *cheapest_partial_path;
2692  Path *simple_gather_path;
2693  ListCell *lc;
2694  double rows;
2695  double *rowsp = NULL;
2696 
2697  /* If there are no partial paths, there's nothing to do here. */
2698  if (rel->partial_pathlist == NIL)
2699  return;
2700 
2701  /* Should we override the rel's rowcount estimate? */
2702  if (override_rows)
2703  rowsp = &rows;
2704 
2705  /*
2706  * The output of Gather is always unsorted, so there's only one partial
2707  * path of interest: the cheapest one. That will be the one at the front
2708  * of partial_pathlist because of the way add_partial_path works.
2709  */
2710  cheapest_partial_path = linitial(rel->partial_pathlist);
2711  rows =
2712  cheapest_partial_path->rows * cheapest_partial_path->parallel_workers;
2713  simple_gather_path = (Path *)
2714  create_gather_path(root, rel, cheapest_partial_path, rel->reltarget,
2715  NULL, rowsp);
2716  add_path(rel, simple_gather_path);
2717 
2718  /*
2719  * For each useful ordering, we can consider an order-preserving Gather
2720  * Merge.
2721  */
2722  foreach(lc, rel->partial_pathlist)
2723  {
2724  Path *subpath = (Path *) lfirst(lc);
2725  GatherMergePath *path;
2726 
2727  if (subpath->pathkeys == NIL)
2728  continue;
2729 
2730  rows = subpath->rows * subpath->parallel_workers;
2731  path = create_gather_merge_path(root, rel, subpath, rel->reltarget,
2732  subpath->pathkeys, NULL, rowsp);
2733  add_path(rel, &path->path);
2734  }
2735 }
2736 
2737 /*
2738  * get_useful_pathkeys_for_relation
2739  * Determine which orderings of a relation might be useful.
2740  *
2741  * Getting data in sorted order can be useful either because the requested
2742  * order matches the final output ordering for the overall query we're
2743  * planning, or because it enables an efficient merge join. Here, we try
2744  * to figure out which pathkeys to consider.
2745  *
2746  * This allows us to do incremental sort on top of an index scan under a gather
2747  * merge node, i.e. parallelized.
2748  *
2749  * XXX At the moment this can only ever return a list with a single element,
2750  * because it looks at query_pathkeys only. So we might return the pathkeys
2751  * directly, but it seems plausible we'll want to consider other orderings
2752  * in the future. For example, we might want to consider pathkeys useful for
2753  * merge joins.
2754  */
2755 static List *
2756 get_useful_pathkeys_for_relation(PlannerInfo *root, RelOptInfo *rel)
2757 {
2758  List *useful_pathkeys_list = NIL;
2759 
2760  /*
2761  * Considering query_pathkeys is always worth it, because it might allow
2762  * us to avoid a total sort when we have a partially presorted path
2763  * available.
2764  */
2765  if (root->query_pathkeys)
2766  {
2767  ListCell *lc;
2768  int npathkeys = 0; /* useful pathkeys */
2769 
2770  foreach(lc, root->query_pathkeys)
2771  {
2772  PathKey *pathkey = (PathKey *) lfirst(lc);
2773  EquivalenceClass *pathkey_ec = pathkey->pk_eclass;
2774 
2775  /*
2776  * We can only build an Incremental Sort for pathkeys which
2777  * contain an EC member in the current relation, so ignore any
2778  * suffix of the list as soon as we find a pathkey without an EC
2779  * member the relation.
2780  *
2781  * By still returning the prefix of the pathkeys list that does
2782  * meet criteria of EC membership in the current relation, we
2783  * enable not just an incremental sort on the entirety of
2784  * query_pathkeys but also incremental sort below a JOIN.
2785  */
2786  if (!find_em_expr_for_rel(pathkey_ec, rel))
2787  break;
2788 
2789  npathkeys++;
2790  }
2791 
2792  /*
2793  * The whole query_pathkeys list matches, so append it directly, to
2794  * allow comparing pathkeys easily by comparing list pointer. If we
2795  * have to truncate the pathkeys, we gotta do a copy though.
2796  */
2797  if (npathkeys == list_length(root->query_pathkeys))
2798  useful_pathkeys_list = lappend(useful_pathkeys_list,
2799  root->query_pathkeys);
2800  else if (npathkeys > 0)
2801  useful_pathkeys_list = lappend(useful_pathkeys_list,
2802  list_truncate(list_copy(root->query_pathkeys),
2803  npathkeys));
2804  }
2805 
2806  return useful_pathkeys_list;
2807 }
2808 
2809 /*
2810  * generate_useful_gather_paths
2811  * Generate parallel access paths for a relation by pushing a Gather or
2812  * Gather Merge on top of a partial path.
2813  *
2814  * Unlike plain generate_gather_paths, this looks both at pathkeys of input
2815  * paths (aiming to preserve the ordering), but also considers ordering that
2816  * might be useful for nodes above the gather merge node, and tries to add
2817  * a sort (regular or incremental) to provide that.
2818  */
2819 void
2820 generate_useful_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
2821 {
2822  ListCell *lc;
2823  double rows;
2824  double *rowsp = NULL;
2825  List *useful_pathkeys_list = NIL;
2826  Path *cheapest_partial_path = NULL;
2827 
2828  /* If there are no partial paths, there's nothing to do here. */
2829  if (rel->partial_pathlist == NIL)
2830  return;
2831 
2832  /* Should we override the rel's rowcount estimate? */
2833  if (override_rows)
2834  rowsp = &rows;
2835 
2836  /* generate the regular gather (merge) paths */
2837  generate_gather_paths(root, rel, override_rows);
2838 
2839  /* consider incremental sort for interesting orderings */
2840  useful_pathkeys_list = get_useful_pathkeys_for_relation(root, rel);
2841 
2842  /* used for explicit (full) sort paths */
2843  cheapest_partial_path = linitial(rel->partial_pathlist);
2844 
2845  /*
2846  * Consider incremental sort paths for each interesting ordering.
2847  */
2848  foreach(lc, useful_pathkeys_list)
2849  {
2850  List *useful_pathkeys = lfirst(lc);
2851  ListCell *lc2;
2852  bool is_sorted;
2853  int presorted_keys;
2854 
2855  foreach(lc2, rel->partial_pathlist)
2856  {
2857  Path *subpath = (Path *) lfirst(lc2);
2858  GatherMergePath *path;
2859 
2860  /*
2861  * If the path has no ordering at all, then we can't use either
2862  * incremental sort or rely on implict sorting with a gather
2863  * merge.
2864  */
2865  if (subpath->pathkeys == NIL)
2866  continue;
2867 
2868  is_sorted = pathkeys_count_contained_in(useful_pathkeys,
2869  subpath->pathkeys,
2870  &presorted_keys);
2871 
2872  /*
2873  * We don't need to consider the case where a subpath is already
2874  * fully sorted because generate_gather_paths already creates a
2875  * gather merge path for every subpath that has pathkeys present.
2876  *
2877  * But since the subpath is already sorted, we know we don't need
2878  * to consider adding a sort (other either kind) on top of it, so
2879  * we can continue here.
2880  */
2881  if (is_sorted)
2882  continue;
2883 
2884  /*
2885  * Consider regular sort for the cheapest partial path (for each
2886  * useful pathkeys). We know the path is not sorted, because we'd
2887  * not get here otherwise.
2888  *
2889  * This is not redundant with the gather paths created in
2890  * generate_gather_paths, because that doesn't generate ordered
2891  * output. Here we add an explicit sort to match the useful
2892  * ordering.
2893  */
2894  if (cheapest_partial_path == subpath)
2895  {
2896  Path *tmp;
2897 
2898  tmp = (Path *) create_sort_path(root,
2899  rel,
2900  subpath,
2901  useful_pathkeys,
2902  -1.0);
2903 
2904  rows = tmp->rows * tmp->parallel_workers;
2905 
2906  path = create_gather_merge_path(root, rel,
2907  tmp,
2908  rel->reltarget,
2909  tmp->pathkeys,
2910  NULL,
2911  rowsp);
2912 
2913  add_path(rel, &path->path);
2914 
2915  /* Fall through */
2916  }
2917 
2918  /*
2919  * Consider incremental sort, but only when the subpath is already
2920  * partially sorted on a pathkey prefix.
2921  */
2922  if (enable_incremental_sort && presorted_keys > 0)
2923  {
2924  Path *tmp;
2925 
2926  /*
2927  * We should have already excluded pathkeys of length 1
2928  * because then presorted_keys > 0 would imply is_sorted was
2929  * true.
2930  */
2931  Assert(list_length(useful_pathkeys) != 1);
2932 
2933  tmp = (Path *) create_incremental_sort_path(root,
2934  rel,
2935  subpath,
2936  useful_pathkeys,
2937  presorted_keys,
2938  -1);
2939 
2940  path = create_gather_merge_path(root, rel,
2941  tmp,
2942  rel->reltarget,
2943  tmp->pathkeys,
2944  NULL,
2945  rowsp);
2946 
2947  add_path(rel, &path->path);
2948  }
2949  }
2950  }
2951 }
2952 
2953 /*
2954  * make_rel_from_joinlist
2955  * Build access paths using a "joinlist" to guide the join path search.
2956  *
2957  * See comments for deconstruct_jointree() for definition of the joinlist
2958  * data structure.
2959  */
2960 static RelOptInfo *
2961 make_rel_from_joinlist(PlannerInfo *root, List *joinlist)
2962 {
2963  int levels_needed;
2964  List *initial_rels;
2965  ListCell *jl;
2966 
2967  /*
2968  * Count the number of child joinlist nodes. This is the depth of the
2969  * dynamic-programming algorithm we must employ to consider all ways of
2970  * joining the child nodes.
2971  */
2972  levels_needed = list_length(joinlist);
2973 
2974  if (levels_needed <= 0)
2975  return NULL; /* nothing to do? */
2976 
2977  /*
2978  * Construct a list of rels corresponding to the child joinlist nodes.
2979  * This may contain both base rels and rels constructed according to
2980  * sub-joinlists.
2981  */
2982  initial_rels = NIL;
2983  foreach(jl, joinlist)
2984  {
2985  Node *jlnode = (Node *) lfirst(jl);
2986  RelOptInfo *thisrel;
2987 
2988  if (IsA(jlnode, RangeTblRef))
2989  {
2990  int varno = ((RangeTblRef *) jlnode)->rtindex;
2991 
2992  thisrel = find_base_rel(root, varno);
2993  }
2994  else if (IsA(jlnode, List))
2995  {
2996  /* Recurse to handle subproblem */
2997  thisrel = make_rel_from_joinlist(root, (List *) jlnode);
2998  }
2999  else
3000  {
3001  elog(ERROR, "unrecognized joinlist node type: %d",
3002  (int) nodeTag(jlnode));
3003  thisrel = NULL; /* keep compiler quiet */
3004  }
3005 
3006  initial_rels = lappend(initial_rels, thisrel);
3007  }
3008 
3009  if (levels_needed == 1)
3010  {
3011  /*
3012  * Single joinlist node, so we're done.
3013  */
3014  return (RelOptInfo *) linitial(initial_rels);
3015  }
3016  else
3017  {
3018  /*
3019  * Consider the different orders in which we could join the rels,
3020  * using a plugin, GEQO, or the regular join search code.
3021  *
3022  * We put the initial_rels list into a PlannerInfo field because
3023  * has_legal_joinclause() needs to look at it (ugly :-().
3024  */
3025  root->initial_rels = initial_rels;
3026 
3027  if (join_search_hook)
3028  return (*join_search_hook) (root, levels_needed, initial_rels);
3029  else if (enable_geqo && levels_needed >= geqo_threshold)
3030  return geqo(root, levels_needed, initial_rels);
3031  else
3032  return standard_join_search(root, levels_needed, initial_rels);
3033  }
3034 }
3035 
3036 /*
3037  * standard_join_search
3038  * Find possible joinpaths for a query by successively finding ways
3039  * to join component relations into join relations.
3040  *
3041  * 'levels_needed' is the number of iterations needed, ie, the number of
3042  * independent jointree items in the query. This is > 1.
3043  *
3044  * 'initial_rels' is a list of RelOptInfo nodes for each independent
3045  * jointree item. These are the components to be joined together.
3046  * Note that levels_needed == list_length(initial_rels).
3047  *
3048  * Returns the final level of join relations, i.e., the relation that is
3049  * the result of joining all the original relations together.
3050  * At least one implementation path must be provided for this relation and
3051  * all required sub-relations.
3052  *
3053  * To support loadable plugins that modify planner behavior by changing the
3054  * join searching algorithm, we provide a hook variable that lets a plugin
3055  * replace or supplement this function. Any such hook must return the same
3056  * final join relation as the standard code would, but it might have a
3057  * different set of implementation paths attached, and only the sub-joinrels
3058  * needed for these paths need have been instantiated.
3059  *
3060  * Note to plugin authors: the functions invoked during standard_join_search()
3061  * modify root->join_rel_list and root->join_rel_hash. If you want to do more
3062  * than one join-order search, you'll probably need to save and restore the
3063  * original states of those data structures. See geqo_eval() for an example.
3064  */
3065 RelOptInfo *
3066 standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
3067 {
3068  int lev;
3069  RelOptInfo *rel;
3070 
3071  /*
3072  * This function cannot be invoked recursively within any one planning
3073  * problem, so join_rel_level[] can't be in use already.
3074  */
3075  Assert(root->join_rel_level == NULL);
3076 
3077  /*
3078  * We employ a simple "dynamic programming" algorithm: we first find all
3079  * ways to build joins of two jointree items, then all ways to build joins
3080  * of three items (from two-item joins and single items), then four-item
3081  * joins, and so on until we have considered all ways to join all the
3082  * items into one rel.
3083  *
3084  * root->join_rel_level[j] is a list of all the j-item rels. Initially we
3085  * set root->join_rel_level[1] to represent all the single-jointree-item
3086  * relations.
3087  */
3088  root->join_rel_level = (List **) palloc0((levels_needed + 1) * sizeof(List *));
3089 
3090  root->join_rel_level[1] = initial_rels;
3091 
3092  for (lev = 2; lev <= levels_needed; lev++)
3093  {
3094  ListCell *lc;
3095 
3096  /*
3097  * Determine all possible pairs of relations to be joined at this
3098  * level, and build paths for making each one from every available
3099  * pair of lower-level relations.
3100  */
3101  join_search_one_level(root, lev);
3102 
3103  /*
3104  * Run generate_partitionwise_join_paths() and generate_gather_paths()
3105  * for each just-processed joinrel. We could not do this earlier
3106  * because both regular and partial paths can get added to a
3107  * particular joinrel at multiple times within join_search_one_level.
3108  *
3109  * After that, we're done creating paths for the joinrel, so run
3110  * set_cheapest().
3111  */
3112  foreach(lc, root->join_rel_level[lev])
3113  {
3114  rel = (RelOptInfo *) lfirst(lc);
3115 
3116  /* Create paths for partitionwise joins. */
3117  generate_partitionwise_join_paths(root, rel);
3118 
3119  /*
3120  * Except for the topmost scan/join rel, consider gathering
3121  * partial paths. We'll do the same for the topmost scan/join rel
3122  * once we know the final targetlist (see grouping_planner).
3123  */
3124  if (lev < levels_needed)
3125  generate_useful_gather_paths(root, rel, false);
3126 
3127  /* Find and save the cheapest paths for this rel */
3128  set_cheapest(rel);
3129 
3130 #ifdef OPTIMIZER_DEBUG
3131  debug_print_rel(root, rel);
3132 #endif
3133  }
3134  }
3135 
3136  /*
3137  * We should have a single rel at the final level.
3138  */
3139  if (root->join_rel_level[levels_needed] == NIL)
3140  elog(ERROR, "failed to build any %d-way joins", levels_needed);
3141  Assert(list_length(root->join_rel_level[levels_needed]) == 1);
3142 
3143  rel = (RelOptInfo *) linitial(root->join_rel_level[levels_needed]);
3144 
3145  root->join_rel_level = NULL;
3146 
3147  return rel;
3148 }
3149 
3150 /*****************************************************************************
3151  * PUSHING QUALS DOWN INTO SUBQUERIES
3152  *****************************************************************************/
3153 
3154 /*
3155  * subquery_is_pushdown_safe - is a subquery safe for pushing down quals?
3156  *
3157  * subquery is the particular component query being checked. topquery
3158  * is the top component of a set-operations tree (the same Query if no
3159  * set-op is involved).
3160  *
3161  * Conditions checked here:
3162  *
3163  * 1. If the subquery has a LIMIT clause, we must not push down any quals,
3164  * since that could change the set of rows returned.
3165  *
3166  * 2. If the subquery contains EXCEPT or EXCEPT ALL set ops we cannot push
3167  * quals into it, because that could change the results.
3168  *
3169  * 3. If the subquery uses DISTINCT, we cannot push volatile quals into it.
3170  * This is because upper-level quals should semantically be evaluated only
3171  * once per distinct row, not once per original row, and if the qual is
3172  * volatile then extra evaluations could change the results. (This issue
3173  * does not apply to other forms of aggregation such as GROUP BY, because
3174  * when those are present we push into HAVING not WHERE, so that the quals
3175  * are still applied after aggregation.)
3176  *
3177  * 4. If the subquery contains window functions, we cannot push volatile quals
3178  * into it. The issue here is a bit different from DISTINCT: a volatile qual
3179  * might succeed for some rows of a window partition and fail for others,
3180  * thereby changing the partition contents and thus the window functions'
3181  * results for rows that remain.
3182  *
3183  * 5. If the subquery contains any set-returning functions in its targetlist,
3184  * we cannot push volatile quals into it. That would push them below the SRFs
3185  * and thereby change the number of times they are evaluated. Also, a
3186  * volatile qual could succeed for some SRF output rows and fail for others,
3187  * a behavior that cannot occur if it's evaluated before SRF expansion.
3188  *
3189  * 6. If the subquery has nonempty grouping sets, we cannot push down any
3190  * quals. The concern here is that a qual referencing a "constant" grouping
3191  * column could get constant-folded, which would be improper because the value
3192  * is potentially nullable by grouping-set expansion. This restriction could
3193  * be removed if we had a parsetree representation that shows that such
3194  * grouping columns are not really constant. (There are other ideas that
3195  * could be used to relax this restriction, but that's the approach most
3196  * likely to get taken in the future. Note that there's not much to be gained
3197  * so long as subquery_planner can't move HAVING clauses to WHERE within such
3198  * a subquery.)
3199  *
3200  * In addition, we make several checks on the subquery's output columns to see
3201  * if it is safe to reference them in pushed-down quals. If output column k
3202  * is found to be unsafe to reference, we set safetyInfo->unsafeColumns[k]
3203  * to true, but we don't reject the subquery overall since column k might not
3204  * be referenced by some/all quals. The unsafeColumns[] array will be
3205  * consulted later by qual_is_pushdown_safe(). It's better to do it this way
3206  * than to make the checks directly in qual_is_pushdown_safe(), because when
3207  * the subquery involves set operations we have to check the output
3208  * expressions in each arm of the set op.
3209  *
3210  * Note: pushing quals into a DISTINCT subquery is theoretically dubious:
3211  * we're effectively assuming that the quals cannot distinguish values that
3212  * the DISTINCT's equality operator sees as equal, yet there are many
3213  * counterexamples to that assumption. However use of such a qual with a
3214  * DISTINCT subquery would be unsafe anyway, since there's no guarantee which
3215  * "equal" value will be chosen as the output value by the DISTINCT operation.
3216  * So we don't worry too much about that. Another objection is that if the
3217  * qual is expensive to evaluate, running it for each original row might cost
3218  * more than we save by eliminating rows before the DISTINCT step. But it
3219  * would be very hard to estimate that at this stage, and in practice pushdown
3220  * seldom seems to make things worse, so we ignore that problem too.
3221  *
3222  * Note: likewise, pushing quals into a subquery with window functions is a
3223  * bit dubious: the quals might remove some rows of a window partition while
3224  * leaving others, causing changes in the window functions' results for the
3225  * surviving rows. We insist that such a qual reference only partitioning
3226  * columns, but again that only protects us if the qual does not distinguish
3227  * values that the partitioning equality operator sees as equal. The risks
3228  * here are perhaps larger than for DISTINCT, since no de-duplication of rows
3229  * occurs and thus there is no theoretical problem with such a qual. But
3230  * we'll do this anyway because the potential performance benefits are very
3231  * large, and we've seen no field complaints about the longstanding comparable
3232  * behavior with DISTINCT.
3233  */
3234 static bool
3235 subquery_is_pushdown_safe(Query *subquery, Query *topquery,
3236  pushdown_safety_info *safetyInfo)
3237 {
3238  SetOperationStmt *topop;
3239 
3240  /* Check point 1 */
3241  if (subquery->limitOffset != NULL || subquery->limitCount != NULL)
3242  return false;
3243 
3244  /* Check point 6 */
3245  if (subquery->groupClause && subquery->groupingSets)
3246  return false;
3247 
3248  /* Check points 3, 4, and 5 */
3249  if (subquery->distinctClause ||
3250  subquery->hasWindowFuncs ||
3251  subquery->hasTargetSRFs)
3252  safetyInfo->unsafeVolatile = true;
3253 
3254  /*
3255  * If we're at a leaf query, check for unsafe expressions in its target
3256  * list, and mark any unsafe ones in unsafeColumns[]. (Non-leaf nodes in
3257  * setop trees have only simple Vars in their tlists, so no need to check
3258  * them.)
3259  */
3260  if (subquery->setOperations == NULL)
3261  check_output_expressions(subquery, safetyInfo);
3262 
3263  /* Are we at top level, or looking at a setop component? */
3264  if (subquery == topquery)
3265  {
3266  /* Top level, so check any component queries */
3267  if (subquery->setOperations != NULL)
3268  if (!recurse_pushdown_safe(subquery->setOperations, topquery,
3269  safetyInfo))
3270  return false;
3271  }
3272  else
3273  {
3274  /* Setop component must not have more components (too weird) */
3275  if (subquery->setOperations != NULL)
3276  return false;
3277  /* Check whether setop component output types match top level */
3278  topop = castNode(SetOperationStmt, topquery->setOperations);
3279  Assert(topop);
3280  compare_tlist_datatypes(subquery->targetList,
3281  topop->colTypes,
3282  safetyInfo);
3283  }
3284  return true;
3285 }
3286 
3287 /*
3288  * Helper routine to recurse through setOperations tree
3289  */
3290 static bool
3291 recurse_pushdown_safe(Node *setOp, Query *topquery,
3292  pushdown_safety_info *safetyInfo)
3293 {
3294  if (IsA(setOp, RangeTblRef))
3295  {
3296  RangeTblRef *rtr = (RangeTblRef *) setOp;
3297  RangeTblEntry *rte = rt_fetch(rtr->rtindex, topquery->rtable);
3298  Query *subquery = rte->subquery;
3299 
3300  Assert(subquery != NULL);
3301  return subquery_is_pushdown_safe(subquery, topquery, safetyInfo);
3302  }
3303  else if (IsA(setOp, SetOperationStmt))
3304  {
3305  SetOperationStmt *op = (SetOperationStmt *) setOp;
3306 
3307  /* EXCEPT is no good (point 2 for subquery_is_pushdown_safe) */
3308  if (op->op == SETOP_EXCEPT)
3309  return false;
3310  /* Else recurse */
3311  if (!recurse_pushdown_safe(op->larg, topquery, safetyInfo))
3312  return false;
3313  if (!recurse_pushdown_safe(op->rarg, topquery, safetyInfo))
3314  return false;
3315  }
3316  else
3317  {
3318  elog(ERROR, "unrecognized node type: %d",
3319  (int) nodeTag(setOp));
3320  }
3321  return true;
3322 }
3323 
3324 /*
3325  * check_output_expressions - check subquery's output expressions for safety
3326  *
3327  * There are several cases in which it's unsafe to push down an upper-level
3328  * qual if it references a particular output column of a subquery. We check
3329  * each output column of the subquery and set unsafeColumns[k] to true if
3330  * that column is unsafe for a pushed-down qual to reference. The conditions
3331  * checked here are:
3332  *
3333  * 1. We must not push down any quals that refer to subselect outputs that
3334  * return sets, else we'd introduce functions-returning-sets into the
3335  * subquery's WHERE/HAVING quals.
3336  *
3337  * 2. We must not push down any quals that refer to subselect outputs that
3338  * contain volatile functions, for fear of introducing strange results due
3339  * to multiple evaluation of a volatile function.
3340  *
3341  * 3. If the subquery uses DISTINCT ON, we must not push down any quals that
3342  * refer to non-DISTINCT output columns, because that could change the set
3343  * of rows returned. (This condition is vacuous for DISTINCT, because then
3344  * there are no non-DISTINCT output columns, so we needn't check. Note that
3345  * subquery_is_pushdown_safe already reported that we can't use volatile
3346  * quals if there's DISTINCT or DISTINCT ON.)
3347  *
3348  * 4. If the subquery has any window functions, we must not push down quals
3349  * that reference any output columns that are not listed in all the subquery's
3350  * window PARTITION BY clauses. We can push down quals that use only
3351  * partitioning columns because they should succeed or fail identically for
3352  * every row of any one window partition, and totally excluding some
3353  * partitions will not change a window function's results for remaining
3354  * partitions. (Again, this also requires nonvolatile quals, but
3355  * subquery_is_pushdown_safe handles that.)
3356  */
3357 static void
3358 check_output_expressions(Query *subquery, pushdown_safety_info *safetyInfo)
3359 {
3360  ListCell *lc;
3361 
3362  foreach(lc, subquery->targetList)
3363  {
3364  TargetEntry *tle = (TargetEntry *) lfirst(lc);
3365 
3366  if (tle->resjunk)
3367  continue; /* ignore resjunk columns */
3368 
3369  /* We need not check further if output col is already known unsafe */
3370  if (safetyInfo->unsafeColumns[tle->resno])
3371  continue;
3372 
3373  /* Functions returning sets are unsafe (point 1) */
3374  if (subquery->hasTargetSRFs &&
3375  expression_returns_set((Node *) tle->expr))
3376  {
3377  safetyInfo->unsafeColumns[tle->resno] = true;
3378  continue;
3379  }
3380 
3381  /* Volatile functions are unsafe (point 2) */
3382  if (contain_volatile_functions((Node *) tle->expr))
3383  {
3384  safetyInfo->unsafeColumns[tle->resno] = true;
3385  continue;
3386  }
3387 
3388  /* If subquery uses DISTINCT ON, check point 3 */
3389  if (subquery->hasDistinctOn &&
3390  !targetIsInSortList(tle, InvalidOid, subquery->distinctClause))
3391  {
3392  /* non-DISTINCT column, so mark it unsafe */
3393  safetyInfo->unsafeColumns[tle->resno] = true;
3394  continue;
3395  }
3396 
3397  /* If subquery uses window functions, check point 4 */
3398  if (subquery->hasWindowFuncs &&
3399  !targetIsInAllPartitionLists(tle, subquery))
3400  {
3401  /* not present in all PARTITION BY clauses, so mark it unsafe */
3402  safetyInfo->unsafeColumns[tle->resno] = true;
3403  continue;
3404  }
3405  }
3406 }
3407 
3408 /*
3409  * For subqueries using UNION/UNION ALL/INTERSECT/INTERSECT ALL, we can
3410  * push quals into each component query, but the quals can only reference
3411  * subquery columns that suffer no type coercions in the set operation.
3412  * Otherwise there are possible semantic gotchas. So, we check the
3413  * component queries to see if any of them have output types different from
3414  * the top-level setop outputs. unsafeColumns[k] is set true if column k
3415  * has different type in any component.
3416  *
3417  * We don't have to care about typmods here: the only allowed difference
3418  * between set-op input and output typmods is input is a specific typmod
3419  * and output is -1, and that does not require a coercion.
3420  *
3421  * tlist is a subquery tlist.
3422  * colTypes is an OID list of the top-level setop's output column types.
3423  * safetyInfo->unsafeColumns[] is the result array.
3424  */
3425 static void
3426 compare_tlist_datatypes(List *tlist, List *colTypes,
3427  pushdown_safety_info *safetyInfo)
3428 {
3429  ListCell *l;
3430  ListCell *colType = list_head(colTypes);
3431 
3432  foreach(l, tlist)
3433  {
3434  TargetEntry *tle = (TargetEntry *) lfirst(l);
3435 
3436  if (tle->resjunk)
3437  continue; /* ignore resjunk columns */
3438  if (colType == NULL)
3439  elog(ERROR, "wrong number of tlist entries");
3440  if (exprType((Node *) tle->expr) != lfirst_oid(colType))
3441  safetyInfo->unsafeColumns[tle->resno] = true;
3442  colType = lnext(colTypes, colType);
3443  }
3444  if (colType != NULL)
3445  elog(ERROR, "wrong number of tlist entries");
3446 }
3447 
3448 /*
3449  * targetIsInAllPartitionLists
3450  * True if the TargetEntry is listed in the PARTITION BY clause
3451  * of every window defined in the query.
3452  *
3453  * It would be safe to ignore windows not actually used by any window
3454  * function, but it's not easy to get that info at this stage; and it's
3455  * unlikely to be useful to spend any extra cycles getting it, since
3456  * unreferenced window definitions are probably infrequent in practice.
3457  */
3458 static bool
3459 targetIsInAllPartitionLists(TargetEntry *tle, Query *query)
3460 {
3461  ListCell *lc;
3462 
3463  foreach(lc, query->windowClause)
3464  {
3465  WindowClause *wc = (WindowClause *) lfirst(lc);
3466 
3467  if (!targetIsInSortList(tle, InvalidOid, wc->partitionClause))
3468  return false;
3469  }
3470  return true;
3471 }
3472 
3473 /*
3474  * qual_is_pushdown_safe - is a particular qual safe to push down?
3475  *
3476  * qual is a restriction clause applying to the given subquery (whose RTE
3477  * has index rti in the parent query).
3478  *
3479  * Conditions checked here:
3480  *
3481  * 1. The qual must not contain any SubPlans (mainly because I'm not sure
3482  * it will work correctly: SubLinks will already have been transformed into
3483  * SubPlans in the qual, but not in the subquery). Note that SubLinks that
3484  * transform to initplans are safe, and will be accepted here because what
3485  * we'll see in the qual is just a Param referencing the initplan output.
3486  *
3487  * 2. If unsafeVolatile is set, the qual must not contain any volatile
3488  * functions.
3489  *
3490  * 3. If unsafeLeaky is set, the qual must not contain any leaky functions
3491  * that are passed Var nodes, and therefore might reveal values from the
3492  * subquery as side effects.
3493  *
3494  * 4. The qual must not refer to the whole-row output of the subquery
3495  * (since there is no easy way to name that within the subquery itself).
3496  *
3497  * 5. The qual must not refer to any subquery output columns that were
3498  * found to be unsafe to reference by subquery_is_pushdown_safe().
3499  */
3500 static bool
3501 qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual,
3502  pushdown_safety_info *safetyInfo)
3503 {
3504  bool safe = true;
3505  List *vars;
3506  ListCell *vl;
3507 
3508  /* Refuse subselects (point 1) */
3509  if (contain_subplans(qual))
3510  return false;
3511 
3512  /* Refuse volatile quals if we found they'd be unsafe (point 2) */
3513  if (safetyInfo->unsafeVolatile &&
3514  contain_volatile_functions(qual))
3515  return false;
3516 
3517  /* Refuse leaky quals if told to (point 3) */
3518  if (safetyInfo->unsafeLeaky &&
3519  contain_leaked_vars(qual))
3520  return false;
3521 
3522  /*
3523  * It would be unsafe to push down window function calls, but at least for
3524  * the moment we could never see any in a qual anyhow. (The same applies
3525  * to aggregates, which we check for in pull_var_clause below.)
3526  */
3527  Assert(!contain_window_function(qual));
3528 
3529  /*
3530  * Examine all Vars used in clause. Since it's a restriction clause, all
3531  * such Vars must refer to subselect output columns ... unless this is
3532  * part of a LATERAL subquery, in which case there could be lateral
3533  * references.
3534  */
3535  vars = pull_var_clause(qual, PVC_INCLUDE_PLACEHOLDERS);
3536  foreach(vl, vars)
3537  {
3538  Var *var = (Var *) lfirst(vl);
3539 
3540  /*
3541  * XXX Punt if we find any PlaceHolderVars in the restriction clause.
3542  * It's not clear whether a PHV could safely be pushed down, and even
3543  * less clear whether such a situation could arise in any cases of
3544  * practical interest anyway. So for the moment, just refuse to push
3545  * down.
3546  */
3547  if (!IsA(var, Var))
3548  {
3549  safe = false;
3550  break;
3551  }
3552 
3553  /*
3554  * Punt if we find any lateral references. It would be safe to push
3555  * these down, but we'd have to convert them into outer references,
3556  * which subquery_push_qual lacks the infrastructure to do. The case
3557  * arises so seldom that it doesn't seem worth working hard on.
3558  */
3559  if (var->varno != rti)
3560  {
3561  safe = false;
3562  break;
3563  }
3564 
3565  /* Subqueries have no system columns */
3566  Assert(var->varattno >= 0);
3567 
3568  /* Check point 4 */
3569  if (var->varattno == 0)
3570  {
3571  safe = false;
3572  break;
3573  }
3574 
3575  /* Check point 5 */
3576  if (safetyInfo->unsafeColumns[var->varattno])
3577  {
3578  safe = false;
3579  break;
3580  }
3581  }
3582 
3583  list_free(vars);
3584 
3585  return safe;
3586 }
3587 
3588 /*
3589  * subquery_push_qual - push down a qual that we have determined is safe
3590  */
3591 static void
3592 subquery_push_qual(Query *subquery, RangeTblEntry *rte, Index rti, Node *qual)
3593 {
3594  if (subquery->setOperations != NULL)
3595  {
3596  /* Recurse to push it separately to each component query */
3597  recurse_push_qual(subquery->setOperations, subquery,
3598  rte, rti, qual);
3599  }
3600  else
3601  {
3602  /*
3603  * We need to replace Vars in the qual (which must refer to outputs of
3604  * the subquery) with copies of the subquery's targetlist expressions.
3605  * Note that at this point, any uplevel Vars in the qual should have
3606  * been replaced with Params, so they need no work.
3607  *
3608  * This step also ensures that when we are pushing into a setop tree,
3609  * each component query gets its own copy of the qual.
3610  */
3611  qual = ReplaceVarsFromTargetList(qual, rti, 0, rte,
3612  subquery->targetList,
3613  REPLACEVARS_REPORT_ERROR, 0,
3614  &subquery->hasSubLinks);
3615 
3616  /*
3617  * Now attach the qual to the proper place: normally WHERE, but if the
3618  * subquery uses grouping or aggregation, put it in HAVING (since the
3619  * qual really refers to the group-result rows).
3620  */
3621  if (subquery->hasAggs || subquery->groupClause || subquery->groupingSets || subquery->havingQual)
3622  subquery->havingQual = make_and_qual(subquery->havingQual, qual);
3623  else
3624  subquery->jointree->quals =
3625  make_and_qual(subquery->jointree->quals, qual);
3626 
3627  /*
3628  * We need not change the subquery's hasAggs or hasSubLinks flags,
3629  * since we can't be pushing down any aggregates that weren't there
3630  * before, and we don't push down subselects at all.
3631  */
3632  }
3633 }
3634 
3635 /*
3636  * Helper routine to recurse through setOperations tree
3637  */
3638 static void
3639 recurse_push_qual(Node *setOp, Query *topquery,
3640  RangeTblEntry *rte, Index rti, Node *qual)
3641 {
3642  if (IsA(setOp, RangeTblRef))
3643  {
3644  RangeTblRef *rtr = (RangeTblRef *) setOp;
3645  RangeTblEntry *subrte = rt_fetch(rtr->rtindex, topquery->rtable);
3646  Query *subquery = subrte->subquery;
3647 
3648  Assert(subquery != NULL);
3649  subquery_push_qual(subquery, rte, rti, qual);
3650  }
3651  else if (IsA(setOp, SetOperationStmt))
3652  {
3653  SetOperationStmt *op = (SetOperationStmt *) setOp;
3654 
3655  recurse_push_qual(op->larg, topquery, rte, rti, qual);
3656  recurse_push_qual(op->rarg, topquery, rte, rti, qual);
3657  }
3658  else
3659  {
3660  elog(ERROR, "unrecognized node type: %d",
3661  (int) nodeTag(setOp));
3662  }
3663 }
3664 
3665 /*****************************************************************************
3666  * SIMPLIFYING SUBQUERY TARGETLISTS
3667  *****************************************************************************/
3668 
3669 /*
3670  * remove_unused_subquery_outputs
3671  * Remove subquery targetlist items we don't need
3672  *
3673  * It's possible, even likely, that the upper query does not read all the
3674  * output columns of the subquery. We can remove any such outputs that are
3675  * not needed by the subquery itself (e.g., as sort/group columns) and do not
3676  * affect semantics otherwise (e.g., volatile functions can't be removed).
3677  * This is useful not only because we might be able to remove expensive-to-
3678  * compute expressions, but because deletion of output columns might allow
3679  * optimizations such as join removal to occur within the subquery.
3680  *
3681  * To avoid affecting column numbering in the targetlist, we don't physically
3682  * remove unused tlist entries, but rather replace their expressions with NULL
3683  * constants. This is implemented by modifying subquery->targetList.
3684  */
3685 static void
3686 remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel)
3687 {
3688  Bitmapset *attrs_used = NULL;
3689  ListCell *lc;
3690 
3691  /*
3692  * Do nothing if subquery has UNION/INTERSECT/EXCEPT: in principle we
3693  * could update all the child SELECTs' tlists, but it seems not worth the
3694  * trouble presently.
3695  */
3696  if (subquery->setOperations)
3697  return;
3698 
3699  /*
3700  * If subquery has regular DISTINCT (not DISTINCT ON), we're wasting our
3701  * time: all its output columns must be used in the distinctClause.
3702  */
3703  if (subquery->distinctClause && !subquery->hasDistinctOn)
3704  return;
3705 
3706  /*
3707  * Collect a bitmap of all the output column numbers used by the upper
3708  * query.
3709  *
3710  * Add all the attributes needed for joins or final output. Note: we must
3711  * look at rel's targetlist, not the attr_needed data, because attr_needed
3712  * isn't computed for inheritance child rels, cf set_append_rel_size().
3713  * (XXX might be worth changing that sometime.)
3714  */
3715  pull_varattnos((Node *) rel->reltarget->exprs, rel->relid, &attrs_used);
3716 
3717  /* Add all the attributes used by un-pushed-down restriction clauses. */
3718  foreach(lc, rel->baserestrictinfo)
3719  {
3720  RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
3721 
3722  pull_varattnos((Node *) rinfo->clause, rel->relid, &attrs_used);
3723  }
3724 
3725  /*
3726  * If there's a whole-row reference to the subquery, we can't remove
3727  * anything.
3728  */
3729  if (bms_is_member(0 - FirstLowInvalidHeapAttributeNumber, attrs_used))
3730  return;
3731 
3732  /*
3733  * Run through the tlist and zap entries we don't need. It's okay to
3734  * modify the tlist items in-place because set_subquery_pathlist made a
3735  * copy of the subquery.
3736  */
3737  foreach(lc, subquery->targetList)
3738  {
3739  TargetEntry *tle = (TargetEntry *) lfirst(lc);
3740  Node *texpr = (Node *) tle->expr;
3741 
3742  /*
3743  * If it has a sortgroupref number, it's used in some sort/group
3744  * clause so we'd better not remove it. Also, don't remove any
3745  * resjunk columns, since their reason for being has nothing to do
3746  * with anybody reading the subquery's output. (It's likely that
3747  * resjunk columns in a sub-SELECT would always have ressortgroupref
3748  * set, but even if they don't, it seems imprudent to remove them.)
3749  */
3750  if (tle->ressortgroupref || tle->resjunk)
3751  continue;
3752 
3753  /*
3754  * If it's used by the upper query, we can't remove it.
3755  */
3756  if (bms_is_member(tle->resno - FirstLowInvalidHeapAttributeNumber,
3757  attrs_used))
3758  continue;
3759 
3760  /*
3761  * If it contains a set-returning function, we can't remove it since
3762  * that could change the number of rows returned by the subquery.
3763  */
3764  if (subquery->hasTargetSRFs &&
3765  expression_returns_set(texpr))
3766  continue;
3767 
3768  /*
3769  * If it contains volatile functions, we daren't remove it for fear
3770  * that the user is expecting their side-effects to happen.
3771  */
3772  if (contain_volatile_functions(texpr))
3773  continue;
3774 
3775  /*
3776  * OK, we don't need it. Replace the expression with a NULL constant.
3777  * Preserve the exposed type of the expression, in case something
3778  * looks at the rowtype of the subquery's result.
3779  */
3780  tle->expr = (Expr *) makeNullConst(exprType(texpr),
3781  exprTypmod(texpr),
3782  exprCollation(texpr));
3783  }
3784 }
3785 
3786 /*
3787  * create_partial_bitmap_paths
3788  * Build partial bitmap heap path for the relation
3789  */
3790 void
3791 create_partial_bitmap_paths(PlannerInfo *root, RelOptInfo *rel,
3792  Path *bitmapqual)
3793 {
3794  int parallel_workers;
3795  double pages_fetched;
3796 
3797  /* Compute heap pages for bitmap heap scan */
3798  pages_fetched = compute_bitmap_pages(root, rel, bitmapqual, 1.0,
3799  NULL, NULL);
3800 
3801  parallel_workers = compute_parallel_worker(rel, pages_fetched, -1,
3802  max_parallel_workers_per_gather);
3803 
3804  if (parallel_workers <= 0)
3805  return;
3806 
3807  add_partial_path(rel, (Path *) create_bitmap_heap_path(root, rel,
3808  bitmapqual, rel->lateral_relids, 1.0, parallel_workers));
3809 }
3810 
3811 /*
3812  * Compute the number of parallel workers that should be used to scan a
3813  * relation. We compute the parallel workers based on the size of the heap to
3814  * be scanned and the size of the index to be scanned, then choose a minimum
3815  * of those.
3816  *
3817  * "heap_pages" is the number of pages from the table that we expect to scan, or
3818  * -1 if we don't expect to scan any.
3819  *
3820  * "index_pages" is the number of pages from the index that we expect to scan, or
3821  * -1 if we don't expect to scan any.
3822  *
3823  * "max_workers" is caller's limit on the number of workers. This typically
3824  * comes from a GUC.
3825  */
3826 int
3827 compute_parallel_worker(RelOptInfo *rel, double heap_pages, double index_pages,
3828  int max_workers)
3829 {
3830  int parallel_workers = 0;
3831 
3832  /*
3833  * If the user has set the parallel_workers reloption, use that; otherwise
3834  * select a default number of workers.
3835  */
3836  if (rel->rel_parallel_workers != -1)
3837  parallel_workers = rel->rel_parallel_workers;
3838  else
3839  {
3840  /*
3841  * If the number of pages being scanned is insufficient to justify a
3842  * parallel scan, just return zero ... unless it's an inheritance
3843  * child. In that case, we want to generate a parallel path here
3844  * anyway. It might not be worthwhile just for this relation, but
3845  * when combined with all of its inheritance siblings it may well pay
3846  * off.
3847  */
3848  if (rel->reloptkind == RELOPT_BASEREL &&
3849  ((heap_pages >= 0 && heap_pages < min_parallel_table_scan_size) ||
3850  (index_pages >= 0 && index_pages < min_parallel_index_scan_size)))
3851  return 0;
3852 
3853  if (heap_pages >= 0)
3854  {
3855  int heap_parallel_threshold;
3856  int heap_parallel_workers = 1;
3857 
3858  /*
3859  * Select the number of workers based on the log of the size of
3860  * the relation. This probably needs to be a good deal more
3861  * sophisticated, but we need something here for now. Note that
3862  * the upper limit of the min_parallel_table_scan_size GUC is
3863  * chosen to prevent overflow here.
3864  */
3865  heap_parallel_threshold = Max(min_parallel_table_scan_size, 1);
3866  while (heap_pages >= (BlockNumber) (heap_parallel_threshold * 3))
3867  {
3868  heap_parallel_workers++;
3869  heap_parallel_threshold *= 3;
3870  if (heap_parallel_threshold > INT_MAX / 3)
3871  break; /* avoid overflow */
3872  }
3873 
3874  parallel_workers = heap_parallel_workers;
3875  }
3876 
3877  if (index_pages >= 0)
3878  {
3879  int index_parallel_workers = 1;
3880  int index_parallel_threshold;
3881 
3882  /* same calculation as for heap_pages above */
3883  index_parallel_threshold = Max(min_parallel_index_scan_size, 1);
3884  while (index_pages >= (BlockNumber) (index_parallel_threshold * 3))
3885  {
3886  index_parallel_workers++;
3887  index_parallel_threshold *= 3;
3888  if (index_parallel_threshold > INT_MAX / 3)
3889  break; /* avoid overflow */
3890  }
3891 
3892  if (parallel_workers > 0)
3893  parallel_workers = Min(parallel_workers, index_parallel_workers);
3894  else
3895  parallel_workers = index_parallel_workers;
3896  }
3897  }
3898 
3899  /* In no case use more than caller supplied maximum number of workers */
3900  parallel_workers = Min(parallel_workers, max_workers);
3901 
3902  return parallel_workers;
3903 }
3904 
3905 /*
3906  * generate_partitionwise_join_paths
3907  * Create paths representing partitionwise join for given partitioned
3908  * join relation.
3909  *
3910  * This must not be called until after we are done adding paths for all
3911  * child-joins. Otherwise, add_path might delete a path to which some path
3912  * generated here has a reference.
3913  */
3914 void
3915 generate_partitionwise_join_paths(PlannerInfo *root, RelOptInfo *rel)
3916 {
3917  List *live_children = NIL;
3918  int cnt_parts;
3919  int num_parts;
3920  RelOptInfo **part_rels;
3921 
3922  /* Handle only join relations here. */
3923  if (!IS_JOIN_REL(rel))
3924  return;
3925 
3926  /* We've nothing to do if the relation is not partitioned. */
3927  if (!IS_PARTITIONED_REL(rel))
3928  return;
3929 
3930  /* The relation should have consider_partitionwise_join set. */
3931  Assert(rel->consider_partitionwise_join);
3932 
3933  /* Guard against stack overflow due to overly deep partition hierarchy. */
3934  check_stack_depth();
3935 
3936  num_parts = rel->nparts;
3937  part_rels = rel->part_rels;
3938 
3939  /* Collect non-dummy child-joins. */
3940  for (cnt_parts = 0; cnt_parts < num_parts; cnt_parts++)
3941  {
3942  RelOptInfo *child_rel = part_rels[cnt_parts];
3943 
3944  /* If it's been pruned entirely, it's certainly dummy. */
3945  if (child_rel == NULL)
3946  continue;
3947 
3948  /* Add partitionwise join paths for partitioned child-joins. */
3949  generate_partitionwise_join_paths(root, child_rel);
3950 
3951  set_cheapest(child_rel);
3952 
3953  /* Dummy children will not be scanned, so ignore those. */
3954  if (IS_DUMMY_REL(child_rel))
3955  continue;
3956 
3957 #ifdef OPTIMIZER_DEBUG
3958  debug_print_rel(root, child_rel);
3959 #endif
3960 
3961  live_children = lappend(live_children, child_rel);
3962  }
3963 
3964  /* If all child-joins are dummy, parent join is also dummy. */
3965  if (!live_children)
3966  {
3967  mark_dummy_rel(rel);
3968  return;
3969  }
3970 
3971  /* Build additional paths for this rel from child-join paths. */
3972  add_paths_to_append_rel(root, rel, live_children);
3973  list_free(live_children);
3974 }
3975 
3976 
3977 /*****************************************************************************
3978  * DEBUG SUPPORT
3979  *****************************************************************************/
3980 
3981 #ifdef OPTIMIZER_DEBUG
3982 
3983 static void
3984 print_relids(PlannerInfo *root, Relids relids)
3985 {
3986  int x;
3987  bool first = true;
3988 
3989  x = -1;
3990  while ((x = bms_next_member(relids, x)) >= 0)
3991  {
3992  if (!first)
3993  printf(" ");
3994  if (x < root->simple_rel_array_size &&
3995  root->simple_rte_array[x])
3996  printf("%s", root->simple_rte_array[x]->eref->aliasname);
3997  else
3998  printf("%d", x);
3999  first = false;
4000  }
4001 }
4002 
4003 static void
4004 print_restrictclauses(PlannerInfo *root, List *clauses)
4005 {
4006  ListCell *l;
4007 
4008  foreach(l, clauses)
4009  {
4010  RestrictInfo *c = lfirst(l);
4011 
4012  print_expr((Node *) c->clause, root->parse->rtable);
4013  if (lnext(clauses, l))
4014  printf(", ");
4015  }
4016 }
4017 
4018 static void
4019 print_path(PlannerInfo *root, Path *path, int indent)
4020 {
4021  const char *ptype;
4022  bool join = false;
4023  Path *subpath = NULL;
4024  int i;
4025 
4026  switch (nodeTag(path))
4027  {
4028  case T_Path:
4029  switch (path->pathtype)
4030  {
4031  case T_SeqScan:
4032  ptype = "SeqScan";
4033  break;
4034  case T_SampleScan:
4035  ptype = "SampleScan";
4036  break;
4037  case T_FunctionScan:
4038  ptype = "FunctionScan";
4039  break;
4040  case T_TableFuncScan:
4041  ptype = "TableFuncScan";
4042  break;
4043  case T_ValuesScan:
4044  ptype = "ValuesScan";
4045  break;
4046  case T_CteScan:
4047  ptype = "CteScan";
4048  break;
4049  case T_NamedTuplestoreScan:
4050  ptype = "NamedTuplestoreScan";
4051  break;
4052  case T_Result:
4053  ptype = "Result";
4054  break;
4055  case T_WorkTableScan:
4056  ptype = "WorkTableScan";
4057  break;
4058  default:
4059  ptype = "???Path";
4060  break;
4061  }
4062  break;
4063  case T_IndexPath:
4064  ptype = "IdxScan";
4065  break;
4066  case T_BitmapHeapPath:
4067  ptype = "BitmapHeapScan";
4068  break;
4069  case T_BitmapAndPath:
4070  ptype = "BitmapAndPath";
4071  break;
4072  case T_BitmapOrPath:
4073  ptype = "BitmapOrPath";
4074  break;
4075  case T_TidPath:
4076  ptype = "TidScan";
4077  break;
4078  case T_SubqueryScanPath:
4079  ptype = "SubqueryScan";
4080  break;
4081  case T_ForeignPath:
4082  ptype = "ForeignScan";
4083  break;
4084  case T_CustomPath:
4085  ptype = "CustomScan";
4086  break;
4087  case T_NestPath:
4088  ptype = "NestLoop";
4089  join = true;
4090  break;
4091  case T_MergePath:
4092  ptype = "MergeJoin";
4093  join = true;
4094  break;
4095  case T_HashPath:
4096  ptype = "HashJoin";
4097  join = true;
4098  break;
4099  case T_AppendPath:
4100  ptype = "Append";
4101  break;
4102  case T_MergeAppendPath:
4103  ptype = "MergeAppend";
4104  break;
4105  case T_GroupResultPath:
4106  ptype = "GroupResult";
4107  break;
4108  case T_MaterialPath:
4109  ptype = "Material";
4110  subpath = ((MaterialPath *) path)->subpath;
4111  break;
4112  case T_UniquePath:
4113  ptype = "Unique";
4114  subpath = ((UniquePath *) path)->subpath;
4115  break;
4116  case T_GatherPath:
4117  ptype = "Gather";
4118  subpath = ((GatherPath *) path)->subpath;
4119  break;
4120  case T_GatherMergePath:
4121  ptype = "GatherMerge";
4122  subpath = ((GatherMergePath *) path)->subpath;
4123  break;
4124  case T_ProjectionPath:
4125  ptype = "Projection";
4126  subpath = ((ProjectionPath *) path)->subpath;
4127  break;
4128  case T_ProjectSetPath:
4129  ptype = "ProjectSet";
4130  subpath = ((ProjectSetPath *) path)->subpath;
4131  break;
4132  case T_SortPath:
4133  ptype = "Sort";
4134  subpath = ((SortPath *) path)->subpath;
4135  break;
4136  case T_IncrementalSortPath:
4137  ptype = "IncrementalSort";
4138  subpath = ((SortPath *) path)->subpath;
4139  break;
4140  case T_GroupPath:
4141  ptype = "Group";
4142  subpath = ((GroupPath *) path)->subpath;
4143  break;
4144  case T_UpperUniquePath:
4145  ptype = "UpperUnique";
4146  subpath = ((UpperUniquePath *) path)->subpath;
4147  break;
4148  case T_AggPath:
4149  ptype = "Agg";
4150  subpath = ((AggPath *) path)->subpath;
4151  break;
4152  case T_GroupingSetsPath:
4153  ptype = "GroupingSets";
4154  subpath = ((GroupingSetsPath *) path)->subpath;
4155  break;
4156  case T_MinMaxAggPath:
4157  ptype = "MinMaxAgg";
4158  break;
4159  case T_WindowAggPath:
4160  ptype = "WindowAgg";
4161  subpath = ((WindowAggPath *) path)->subpath;
4162  break;
4163  case T_SetOpPath:
4164  ptype = "SetOp";
4165  subpath = ((SetOpPath *) path)->subpath;
4166  break;
4167  case T_RecursiveUnionPath:
4168  ptype = "RecursiveUnion";
4169  break;
4170  case T_LockRowsPath:
4171  ptype = "LockRows";
4172  subpath = ((LockRowsPath *) path)->subpath;
4173  break;
4174  case T_ModifyTablePath:
4175  ptype = "ModifyTable";
4176  break;
4177  case T_LimitPath:
4178  ptype = "Limit";
4179  subpath = ((LimitPath *) path)->subpath;
4180  break;
4181  default:
4182  ptype = "???Path";
4183  break;
4184  }
4185 
4186  for (i = 0; i < indent; i++)
4187  printf("\t");
4188  printf("%s", ptype);
4189 
4190  if (path->parent)
4191  {
4192  printf("(");
4193  print_relids(root, path->parent->relids);
4194  printf(")");
4195  }
4196  if (path->param_info)
4197  {
4198  printf(" required_outer (");
4199  print_relids(root, path->param_info->ppi_req_outer);
4200  printf(")");
4201  }
4202  printf(" rows=%.0f cost=%.2f..%.2f\n",
4203  path->rows, path->startup_cost, path->total_cost);
4204 
4205  if (path->pathkeys)
4206  {
4207  for (i = 0; i < indent; i++)
4208  printf("\t");
4209  printf(" pathkeys: ");
4210  print_pathkeys(path->pathkeys, root->parse->rtable);
4211  }
4212 
4213  if (join)
4214  {
4215  JoinPath *jp = (JoinPath *) path;
4216 
4217  for (i = 0; i < indent; i++)
4218  printf("\t");
4219  printf(" clauses: ");
4220  print_restrictclauses(root, jp->joinrestrictinfo);
4221  printf("\n");
4222 
4223  if (IsA(path, MergePath))
4224  {
4225  MergePath *mp = (MergePath *) path;
4226 
4227  for (i = 0; i < indent; i++)
4228  printf("\t");
4229  printf(" sortouter=%d sortinner=%d materializeinner=%d\n",
4230  ((mp->outersortkeys) ? 1 : 0),
4231  ((mp->innersortkeys) ? 1 : 0),
4232  ((mp->materialize_inner) ? 1 : 0));
4233  }
4234 
4235  print_path(root, jp->outerjoinpath, indent + 1);
4236  print_path(root, jp->innerjoinpath, indent + 1);
4237  }
4238 
4239  if (subpath)
4240  print_path(root, subpath, indent + 1);
4241 }
4242 
4243 void
4244 debug_print_rel(PlannerInfo *root, RelOptInfo *rel)
4245 {
4246  ListCell *l;
4247 
4248  printf("RELOPTINFO (");
4249  print_relids(root, rel->relids);
4250  printf("): rows=%.0f width=%d\n", rel->rows, rel->reltarget->width);
4251 
4252  if (rel->baserestrictinfo)
4253  {
4254  printf("\tbaserestrictinfo: ");
4255  print_restrictclauses(root, rel->baserestrictinfo);
4256  printf("\n");
4257  }
4258 
4259  if (rel->joininfo)
4260  {
4261  printf("\tjoininfo: ");
4262  print_restrictclauses(root, rel->joininfo);
4263  printf("\n");
4264  }
4265 
4266  printf("\tpath list:\n");
4267  foreach(l, rel->pathlist)
4268  print_path(root, lfirst(l), 1);
4269  if (rel->cheapest_parameterized_paths)
4270  {
4271  printf("\n\tcheapest parameterized paths:\n");
4272  foreach(l, rel->cheapest_parameterized_paths)
4273  print_path(root, lfirst(l), 1);
4274  }
4275  if (rel->cheapest_startup_path)
4276  {
4277  printf("\n\tcheapest startup path:\n");
4278  print_path(root, rel->cheapest_startup_path, 1);
4279  }
4280  if (rel->cheapest_total_path)
4281  {
4282  printf("\n\tcheapest total path:\n");
4283  print_path(root, rel->cheapest_total_path, 1);
4284  }
4285  printf("\n");
4286  fflush(stdout);
4287 }
4288 
4289 #endif /* OPTIMIZER_DEBUG */
RelOptInfo::has_eclass_joins
bool has_eclass_joins
Definition: pathnodes.h:734
get_cheapest_path_for_pathkeys
Path * get_cheapest_path_for_pathkeys(List *paths, List *pathkeys, Relids required_outer, CostSelector cost_criterion, bool require_parallel_safe)
Definition: pathkeys.c:404
set_subquery_size_estimates
void set_subquery_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5149
standard_join_search
RelOptInfo * standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
Definition: allpaths.c:3066
Query::limitOffset
Node * limitOffset
Definition: parsenodes.h:160
NIL
#define NIL
Definition: pg_list.h:65
contain_leaked_vars
bool contain_leaked_vars(Node *clause)
Definition: clauses.c:1361
MergePath::outersortkeys
List * outersortkeys
Definition: pathnodes.h:1583
Plan::plan_rows
double plan_rows
Definition: plannodes.h:129
make_one_rel
RelOptInfo * make_one_rel(PlannerInfo *root, List *joinlist)
Definition: allpaths.c:152
create_gather_path
GatherPath * create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, PathTarget *target, Relids required_outer, double *rows)
Definition: pathnode.c:1878
IsA
#define IsA(nodeptr, _type_)
Definition: nodes.h:579
Path::pathtarget
PathTarget * pathtarget
Definition: pathnodes.h:1146
build_expression_pathkey
List * build_expression_pathkey(PlannerInfo *root, Expr *expr, Relids nullable_relids, Oid opno, Relids rel, bool create_it)
Definition: pathkeys.c:782
PlannerInfo::parse
Query * parse
Definition: pathnodes.h:179
Var::varlevelsup
Index varlevelsup
Definition: primnodes.h:191
forboth
#define forboth(cell1, list1, cell2, list2)
Definition: pg_list.h:428
add_path
void add_path(RelOptInfo *parent_rel, Path *new_path)
Definition: pathnode.c:422
create_subqueryscan_path
SubqueryScanPath * create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, List *pathkeys, Relids required_outer)
Definition: pathnode.c:1917
join_search_hook_type
RelOptInfo *(* join_search_hook_type)(PlannerInfo *root, int levels_needed, List *initial_rels)
Definition: paths.h:45
PlannerInfo::plan_params
List * plan_params
Definition: pathnodes.h:193
enable_incremental_sort
bool enable_incremental_sort
Definition: costsize.c:131
Query::sortClause
List * sortClause
Definition: parsenodes.h:158
RelOptInfo::reloptkind
RelOptKind reloptkind
Definition: pathnodes.h:663
set_base_rel_sizes
static void set_base_rel_sizes(PlannerInfo *root)
Definition: allpaths.c:291
pg_proc.h
RelOptInfo::attr_needed
Relids * attr_needed
Definition: pathnodes.h:699
PlannerInfo::query_pathkeys
List * query_pathkeys
Definition: pathnodes.h:298
PlannerInfo::join_info_list
List * join_info_list
Definition: pathnodes.h:283
get_cheapest_parameterized_child_path
static Path * get_cheapest_parameterized_child_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: allpaths.c:1961
create_incremental_sort_path
SortPath * create_incremental_sort_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, List *pathkeys, int presorted_keys, double limit_tuples)
Definition: pathnode.c:2802
MergePath::materialize_inner
bool materialize_inner
Definition: pathnodes.h:1586
lnext
static ListCell * lnext(const List *l, const ListCell *c)
Definition: pg_list.h:330
set_cte_pathlist
static void set_cte_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2504
Query::jointree
FromExpr * jointree
Definition: parsenodes.h:138
set_foreign_size
static void set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:904
PlannerInfo::parent_root
PlannerInfo * parent_root
Definition: pathnodes.h:185
castNode
#define castNode(_type_, nodeptr)
Definition: nodes.h:597
compute_parallel_worker
int compute_parallel_worker(RelOptInfo *rel, double heap_pages, double index_pages, int max_workers)
Definition: allpaths.c:3827
build_partition_pathkeys
List * build_partition_pathkeys(PlannerInfo *root, RelOptInfo *partrel, ScanDirection scandir, bool *partialkeys)
Definition: pathkeys.c:699
exprTypmod
int32 exprTypmod(const Node *expr)
Definition: nodeFuncs.c:275
JoinPath::innerjoinpath
Path * innerjoinpath
Definition: pathnodes.h:1527
add_paths_to_append_rel
void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel, List *live_childrels)
Definition: allpaths.c:1301
set_cte_size_estimates
void set_cte_size_estimates(PlannerInfo *root, RelOptInfo *rel, double cte_rows)
Definition: costsize.c:5321
RelOptInfo::cheapest_startup_path
struct Path * cheapest_startup_path
Definition: pathnodes.h:683
create_resultscan_path
Path * create_resultscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2074
RelOptInfo::tuples
double tuples
Definition: pathnodes.h:706
RelOptInfo::baserestrictinfo
List * baserestrictinfo
Definition: pathnodes.h:728
Query::hasAggs
bool hasAggs
Definition: parsenodes.h:125
create_index_paths
void create_index_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: indxpath.c:231
Min
#define Min(x, y)
Definition: c.h:927
bms_next_member
int bms_next_member(const Bitmapset *a, int prevbit)
Definition: bitmapset.c:1043
expression_returns_set
bool expression_returns_set(Node *clause)
Definition: nodeFuncs.c:718
Path::parallel_workers
int parallel_workers
Definition: pathnodes.h:1152
set_rel_pathlist_hook
set_rel_pathlist_hook_type set_rel_pathlist_hook
Definition: allpaths.c:68
RestrictInfo::pseudoconstant
bool pseudoconstant
Definition: pathnodes.h:1994
RelOptInfo::consider_param_startup
bool consider_param_startup
Definition: pathnodes.h:673
create_material_path
MaterialPath * create_material_path(RelOptInfo *rel, Path *subpath)
Definition: pathnode.c:1527
relation_excluded_by_constraints
bool relation_excluded_by_constraints(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: plancat.c:1384
list_truncate
List * list_truncate(List *list, int new_size)
Definition: list.c:585
Path::param_info
ParamPathInfo * param_info
Definition: pathnodes.h:1148
generate_partitionwise_join_paths
void generate_partitionwise_join_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: allpaths.c:3915
Query::groupingSets
List * groupingSets
Definition: parsenodes.h:150
set_namedtuplestore_pathlist
static void set_namedtuplestore_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2570
list_copy
List * list_copy(const List *oldlist)
Definition: list.c:1403
limit_needed
bool limit_needed(Query *parse)
Definition: planner.c:3011
Node
Definition: nodes.h:528
IS_JOIN_REL
#define IS_JOIN_REL(rel)
Definition: pathnodes.h:644
list_concat
List * list_concat(List *list1, const List *list2)
Definition: list.c:515
RelOptInfo::partial_pathlist
List * partial_pathlist
Definition: pathnodes.h:682
make_tlist_from_pathtarget
List * make_tlist_from_pathtarget(PathTarget *target)
Definition: tlist.c:639
Var::varattno
AttrNumber varattno
Definition: primnodes.h:186
bms_get_singleton_member
bool bms_get_singleton_member(const Bitmapset *a, int *member)
Definition: bitmapset.c:615
pushdown_safety_info::unsafeColumns
bool * unsafeColumns
Definition: allpaths.c:56
enable_parallel_append
bool enable_parallel_append
Definition: costsize.c:140
set_append_rel_size
static void set_append_rel_size(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:946
printf
#define printf(...)
Definition: port.h:199
FirstLowInvalidHeapAttributeNumber
#define FirstLowInvalidHeapAttributeNumber
Definition: sysattr.h:27
BlockNumber
uint32 BlockNumber
Definition: block.h:31
generate_gather_paths
void generate_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
Definition: allpaths.c:2689
RelOptInfo::cheapest_parameterized_paths
List * cheapest_parameterized_paths
Definition: pathnodes.h:686
make_and_qual
Node * make_and_qual(Node *qual1, Node *qual2)
Definition: makefuncs.c:688
pull_var_clause
List * pull_var_clause(Node *node, int flags)
Definition: var.c:535
compare_pathkeys
PathKeysComparison compare_pathkeys(List *keys1, List *keys2)
Definition: pathkeys.c:285
pushdown_safety_info::unsafeLeaky
bool unsafeLeaky
Definition: allpaths.c:58
add_child_rel_equivalences
void add_child_rel_equivalences(PlannerInfo *root, AppendRelInfo *appinfo, RelOptInfo *parent_rel, RelOptInfo *child_rel)
Definition: equivclass.c:2247
contain_volatile_functions
bool contain_volatile_functions(Node *clause)
Definition: clauses.c:724
pushdown_safety_info
struct pushdown_safety_info pushdown_safety_info
list_copy_tail
List * list_copy_tail(const List *oldlist, int nskip)
Definition: list.c:1422
RangeTblEntry::funcordinality
bool funcordinality
Definition: parsenodes.h:1065
recurse_pushdown_safe
static bool recurse_pushdown_safe(Node *setOp, Query *topquery, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3291
Var
Definition: primnodes.h:181
set_result_pathlist
static void set_result_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2600
list_nth_int
static int list_nth_int(const List *list, int n)
Definition: pg_list.h:297
compare_tlist_datatypes
static void compare_tlist_datatypes(List *tlist, List *colTypes, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3426
set_result_size_estimates
void set_result_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5391
create_functionscan_path
Path * create_functionscan_path(PlannerInfo *root, RelOptInfo *rel, List *pathkeys, Relids required_outer)
Definition: pathnode.c:1945
set_base_rel_consider_startup
static void set_base_rel_consider_startup(PlannerInfo *root)
Definition: allpaths.c:248
RangeTblEntry::values_lists
List * values_lists
Definition: parsenodes.h:1075
FromExpr::quals
Node * quals
Definition: primnodes.h:1514
IS_SIMPLE_REL
#define IS_SIMPLE_REL(rel)
Definition: pathnodes.h:639
pull_varattnos
void pull_varattnos(Node *node, Index varno, Bitmapset **varattnos)
Definition: var.c:219
Query::hasDistinctOn
bool hasDistinctOn
Definition: parsenodes.h:129
int32
signed int int32
Definition: c.h:362
Query::windowClause
List * windowClause
Definition: parsenodes.h:154
Query::targetList
List * targetList
Definition: parsenodes.h:140
makeNullConst
Const * makeNullConst(Oid consttype, int32 consttypmod, Oid constcollid)
Definition: makefuncs.c:337
AppendPath::first_partial_path
int first_partial_path
Definition: pathnodes.h:1407
fls
int fls(int mask)
Definition: fls.c:55
create_valuesscan_path
Path * create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:1997
PlannerInfo::simple_rel_array
struct RelOptInfo ** simple_rel_array
Definition: pathnodes.h:203
Path::pathtype
NodeTag pathtype
Definition: pathnodes.h:1143
SpecialJoinInfo::syn_righthand
Relids syn_righthand
Definition: pathnodes.h:2179
list_make1
#define list_make1(x1)
Definition: pg_list.h:226
AppendPath::subpaths
List * subpaths
Definition: pathnodes.h:1405
RelOptInfo::subroot
PlannerInfo * subroot
Definition: pathnodes.h:710
contain_subplans
bool contain_subplans(Node *clause)
Definition: clauses.c:610
pathkeys_count_contained_in
bool pathkeys_count_contained_in(List *keys1, List *keys2, int *n_common)
Definition: pathkeys.c:343
is_parallel_safe
bool is_parallel_safe(PlannerInfo *root, Node *node)
Definition: clauses.c:854
RelOptInfo::lateral_relids
Relids lateral_relids
Definition: pathnodes.h:691
PlannerInfo::tuple_fraction
double tuple_fraction
Definition: pathnodes.h:336
pfree
void pfree(void *pointer)
Definition: mcxt.c:1057
create_tidscan_paths
void create_tidscan_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: tidpath.c:385
TargetEntry::resjunk
bool resjunk
Definition: primnodes.h:1417
linitial
#define linitial(l)
Definition: pg_list.h:194
FdwRoutine::IsForeignScanParallelSafe
IsForeignScanParallelSafe_function IsForeignScanParallelSafe
Definition: fdwapi.h:240
T_Result
Definition: nodes.h:46
Query::rtable
List * rtable
Definition: parsenodes.h:137
Query::distinctClause
List * distinctClause
Definition: parsenodes.h:156
PlannerInfo::all_baserels
Relids all_baserels
Definition: pathnodes.h:227
generate_useful_gather_paths
void generate_useful_gather_paths(PlannerInfo *root, RelOptInfo *rel, bool override_rows)
Definition: allpaths.c:2820
ERROR
#define ERROR
Definition: elog.h:43
WindowClause::partitionClause
List * partitionClause
Definition: parsenodes.h:1354
list_nth
static void * list_nth(const List *list, int n)
Definition: pg_list.h:286
Path::startup_cost
Cost startup_cost
Definition: pathnodes.h:1156
JoinPath::joinrestrictinfo
List * joinrestrictinfo
Definition: pathnodes.h:1529
enable_geqo
bool enable_geqo
Definition: allpaths.c:62
PlannerGlobal::parallelModeOK
bool parallelModeOK
Definition: pathnodes.h:143
IS_DUMMY_REL
#define IS_DUMMY_REL(r)
Definition: pathnodes.h:1419
Path::parent
RelOptInfo * parent
Definition: pathnodes.h:1145
bms_is_subset
bool bms_is_subset(const Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:315
fetch_upper_rel
RelOptInfo * fetch_upper_rel(PlannerInfo *root, UpperRelationKind kind, Relids relids)
Definition: relnode.c:1192
targetIsInAllPartitionLists
static bool targetIsInAllPartitionLists(TargetEntry *tle, Query *query)
Definition: allpaths.c:3459
compare_path_costs
int compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
Definition: pathnode.c:71
check_index_predicates
void check_index_predicates(PlannerInfo *root, RelOptInfo *rel)
Definition: indxpath.c:3308
RelOptInfo::cheapest_total_path
struct Path * cheapest_total_path
Definition: pathnodes.h:684
make_rel_from_joinlist
static RelOptInfo * make_rel_from_joinlist(PlannerInfo *root, List *joinlist)
Definition: allpaths.c:2961
set_append_rel_pathlist
static void set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:1225
Query::limitCount
Node * limitCount
Definition: parsenodes.h:161
PATH_REQ_OUTER
#define PATH_REQ_OUTER(path)
Definition: pathnodes.h:1164
c
char * c
Definition: preproc-cursor.c:31
set_rel_consider_parallel
static void set_rel_consider_parallel(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:591
PlannerGlobal::subplans
List * subplans
Definition: pathnodes.h:111
set_dummy_rel_pathlist
static void set_dummy_rel_pathlist(RelOptInfo *rel)
Definition: allpaths.c:2129
RelOptInfo::joininfo
List * joininfo
Definition: pathnodes.h:732
convert_subquery_pathkeys
List * convert_subquery_pathkeys(PlannerInfo *root, RelOptInfo *rel, List *subquery_pathkeys, List *subquery_tlist)
Definition: pathkeys.c:838
check_stack_depth
void check_stack_depth(void)
Definition: postgres.c:3312
PlannerInfo::glob
PlannerGlobal * glob
Definition: pathnodes.h:181
remove_unused_subquery_outputs
static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel)
Definition: allpaths.c:3686
RelOptInfo::fdwroutine
struct FdwRoutine * fdwroutine
Definition: pathnodes.h:719
set_subquery_pathlist
static void set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:2188
RelOptInfo::nparts
int nparts
Definition: pathnodes.h:744
FdwRoutine::GetForeignRelSize
GetForeignRelSize_function GetForeignRelSize
Definition: fdwapi.h:188
TargetEntry::resno
AttrNumber resno
Definition: primnodes.h:1411
set_foreign_pathlist
static void set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:928
list_head
static ListCell * list_head(const List *l)
Definition: pg_list.h:125
create_merge_append_path
MergeAppendPath * create_merge_append_path(PlannerInfo *root, RelOptInfo *rel, List *subpaths, List *pathkeys, Relids required_outer, List *partitioned_rels)
Definition: pathnode.c:1377
RelOptInfo::relids
Relids relids
Definition: pathnodes.h:666
min_parallel_index_scan_size
int min_parallel_index_scan_size
Definition: allpaths.c:65
PlannerInfo::total_table_pages
double total_table_pages
Definition: pathnodes.h:333
list_make1_int
#define list_make1_int(x1)
Definition: pg_list.h:237
PlannerInfo::simple_rel_array_size
int simple_rel_array_size
Definition: pathnodes.h:204
join_search_one_level
void join_search_one_level(PlannerInfo *root, int level)
Definition: joinrels.c:71
print_expr
void print_expr(const Node *expr, const List *rtable)
Definition: print.c:321
get_singleton_append_subpath
static Path * get_singleton_append_subpath(Path *path)
Definition: allpaths.c:2095
rt_fetch
#define rt_fetch(rangetable_index, rangetable)
Definition: parsetree.h:31
get_useful_pathkeys_for_relation
static List * get_useful_pathkeys_for_relation(PlannerInfo *root, RelOptInfo *rel)
Definition: allpaths.c:2756
get_cheapest_parallel_safe_total_inner
Path * get_cheapest_parallel_safe_total_inner(List *paths)
Definition: pathkeys.c:482
create_append_path
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:1215
RelOptInfo::relid
Index relid
Definition: pathnodes.h:694
create_worktablescan_path
Path * create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2100
contain_window_function
bool contain_window_function(Node *clause)
Definition: clauses.c:494
lappend
List * lappend(List *list, void *datum)
Definition: list.c:321
set_tablesample_rel_size
static void set_tablesample_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:816
PlannerInfo::simple_rte_array
RangeTblEntry ** simple_rte_array
Definition: pathnodes.h:211
FdwRoutine::GetForeignPaths
GetForeignPaths_function GetForeignPaths
Definition: fdwapi.h:189
RestrictInfo::clause
Expr * clause
Definition: pathnodes.h:1986
bms_is_empty
bool bms_is_empty(const Bitmapset *a)
Definition: bitmapset.c:701
set_plain_rel_pathlist
static void set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:766
TsmRoutine::SampleScanGetSampleSize
SampleScanGetSampleSize_function SampleScanGetSampleSize
Definition: tsmapi.h:68
Var::varno
Index varno
Definition: primnodes.h:184
check_output_expressions
static void check_output_expressions(Query *subquery, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3358
ReplaceVarsFromTargetList
Node * ReplaceVarsFromTargetList(Node *node, int target_varno, int sublevels_up, RangeTblEntry *target_rte, List *targetlist, ReplaceVarsNoMatchOption nomatch_option, int nomatch_varno, bool *outer_hasSubLinks)
Definition: rewriteManip.c:1514
set_cheapest
void set_cheapest(RelOptInfo *parent_rel)
Definition: pathnode.c:244
PathTarget::exprs
List * exprs
Definition: pathnodes.h:1075
qual_is_pushdown_safe
static bool qual_is_pushdown_safe(Query *subquery, Index rti, Node *qual, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3501
print_pathkeys
void print_pathkeys(const List *pathkeys, const List *rtable)
Definition: print.c:426
PVC_INCLUDE_PLACEHOLDERS
#define PVC_INCLUDE_PLACEHOLDERS
Definition: optimizer.h:175
pathkeys_contained_in
bool pathkeys_contained_in(List *keys1, List *keys2)
Definition: pathkeys.c:324
JoinPath::outerjoinpath
Path * outerjoinpath
Definition: pathnodes.h:1526
set_namedtuplestore_size_estimates
void set_namedtuplestore_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5358
RelOptInfo::consider_partitionwise_join
bool consider_partitionwise_join
Definition: pathnodes.h:737
join_search_hook
join_search_hook_type join_search_hook
Definition: allpaths.c:71
bms_membership
BMS_Membership bms_membership(const Bitmapset *a)
Definition: bitmapset.c:672
palloc0
void * palloc0(Size size)
Definition: mcxt.c:981
set_base_rel_pathlists
static void set_base_rel_pathlists(PlannerInfo *root)
Definition: allpaths.c:334
mark_dummy_rel
void mark_dummy_rel(RelOptInfo *rel)
Definition: joinrels.c:1261
create_samplescan_path
Path * create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:954
RelOptInfo::rel_parallel_workers
int rel_parallel_workers
Definition: pathnodes.h:712
PlannerInfo::append_rel_list
List * append_rel_list
Definition: pathnodes.h:290
PlannerInfo::cte_plan_ids
List * cte_plan_ids
Definition: pathnodes.h:261
create_tablefuncscan_path
Path * create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:1971
RangeTblEntry::self_reference
bool self_reference
Definition: parsenodes.h:1082
set_baserel_size_estimates
void set_baserel_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:4641
RelOptInfo::boundinfo
struct PartitionBoundInfoData * boundinfo
Definition: pathnodes.h:747
Index
unsigned int Index
Definition: c.h:482
geqo_threshold
int geqo_threshold
Definition: allpaths.c:63
RelOptInfo::rtekind
RTEKind rtekind
Definition: pathnodes.h:696
geqo
RelOptInfo * geqo(PlannerInfo *root, int number_of_rels, List *initial_rels)
Definition: geqo_main.c:67
set_tablesample_rel_pathlist
static void set_tablesample_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:856
RangeTblEntry::security_barrier
bool security_barrier
Definition: parsenodes.h:1013
set_tablefunc_pathlist
static void set_tablefunc_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2480
find_em_expr_for_rel
Expr * find_em_expr_for_rel(EquivalenceClass *ec, RelOptInfo *rel)
Definition: equivclass.c:773
get_typavgwidth
int32 get_typavgwidth(Oid typid, int32 typmod)
Definition: lsyscache.c:2466
RelOptInfo::rows
double rows
Definition: pathnodes.h:669
InvalidOid
#define InvalidOid
Definition: postgres_ext.h:36
targetIsInSortList
bool targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
Definition: parse_clause.c:3511
create_gather_merge_path
GatherMergePath * create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, PathTarget *target, List *pathkeys, Relids required_outer, double *rows)
Definition: pathnode.c:1787
create_sort_path
SortPath * create_sort_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath, List *pathkeys, double limit_tuples)
Definition: pathnode.c:2851
Path::total_cost
Cost total_cost
Definition: pathnodes.h:1157
Query::hasTargetSRFs
bool hasTargetSRFs
Definition: parsenodes.h:127
lcons
List * lcons(void *datum, List *list)
Definition: list.c:453
Path::pathkeys
List * pathkeys
Definition: pathnodes.h:1159
pg_class.h
set_rel_pathlist
static void set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:470
GetTsmRoutine
TsmRoutine * GetTsmRoutine(Oid tsmhandler)
Definition: tablesample.c:27
Max
#define Max(x, y)
Definition: c.h:921
partitions_are_ordered
bool partitions_are_ordered(PartitionBoundInfo boundinfo, int nparts)
Definition: partbounds.c:2748
RelOptInfo::pages
BlockNumber pages
Definition: pathnodes.h:705
Assert
#define Assert(condition)
Definition: c.h:745
lfirst
#define lfirst(lc)
Definition: pg_list.h:189
Alias::aliasname
char * aliasname
Definition: primnodes.h:42
Query::hasWindowFuncs
bool hasWindowFuncs
Definition: parsenodes.h:126
pg_operator.h
set_rel_pathlist_hook_type
void(* set_rel_pathlist_hook_type)(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: paths.h:29
PlannerInfo::join_rel_level
List ** join_rel_level
Definition: pathnodes.h:256
RangeTblEntry::functions
List * functions
Definition: parsenodes.h:1064
has_multiple_baserels
static bool has_multiple_baserels(PlannerInfo *root)
Definition: allpaths.c:2155
Path::rows
double rows
Definition: pathnodes.h:1155
TargetEntry::expr
Expr * expr
Definition: primnodes.h:1410
func_parallel
char func_parallel(Oid funcid)
Definition: lsyscache.c:1711
SpecialJoinInfo::jointype
JoinType jointype
Definition: pathnodes.h:2180
PathKey::pk_eclass
EquivalenceClass * pk_eclass
Definition: pathnodes.h:1042
RelOptInfo::part_rels
struct RelOptInfo ** part_rels
Definition: pathnodes.h:751
PlannerInfo::non_recursive_path
struct Path * non_recursive_path
Definition: pathnodes.h:355
enable_partitionwise_join
bool enable_partitionwise_join
Definition: costsize.c:138
subquery_push_qual
static void subquery_push_qual(Query *subquery, RangeTblEntry *rte, Index rti, Node *qual)
Definition: allpaths.c:3592
exprType
Oid exprType(const Node *expr)
Definition: nodeFuncs.c:41
list_length
static int list_length(const List *l)
Definition: pg_list.h:169
min_parallel_table_scan_size
int min_parallel_table_scan_size
Definition: allpaths.c:64
exprCollation
Oid exprCollation(const Node *expr)
Definition: nodeFuncs.c:768
SetOperationStmt::op
SetOperation op
Definition: parsenodes.h:1659
recurse_push_qual
static void recurse_push_qual(Node *setOp, Query *topquery, RangeTblEntry *rte, Index rti, Node *qual)
Definition: allpaths.c:3639
RangeTblEntry::ctelevelsup
Index ctelevelsup
Definition: parsenodes.h:1081
pushdown_safety_info::unsafeVolatile
bool unsafeVolatile
Definition: allpaths.c:57
RelOptInfo::consider_parallel
bool consider_parallel
Definition: pathnodes.h:674
bms_add_member
Bitmapset * bms_add_member(Bitmapset *a, int x)
Definition: bitmapset.c:736
MergePath::innersortkeys
List * innersortkeys
Definition: pathnodes.h:1584
TsmRoutine::repeatable_across_scans
bool repeatable_across_scans
Definition: tsmapi.h:65
Relids
Bitmapset * Relids
Definition: pathnodes.h:28
PlannerInfo::query_level
Index query_level
Definition: pathnodes.h:183
InvalidAttrNumber
#define InvalidAttrNumber
Definition: attnum.h:23
nodeTag
#define nodeTag(nodeptr)
Definition: nodes.h:533
generate_unaccent_rules.stdout
stdout
Definition: generate_unaccent_rules.py:44
pushdown_safety_info
Definition: allpaths.c:54
get_rel_persistence
char get_rel_persistence(Oid relid)
Definition: lsyscache.c:1990
RangeTblEntry::rtekind
RTEKind rtekind
Definition: parsenodes.h:977
subquery_is_pushdown_safe
static bool subquery_is_pushdown_safe(Query *subquery, Query *topquery, pushdown_safety_info *safetyInfo)
Definition: allpaths.c:3235
Query::cteList
List * cteList
Definition: parsenodes.h:135
RangeTblEntry::ctename
char * ctename
Definition: parsenodes.h:1080
Query::setOperations
Node * setOperations
Definition: parsenodes.h:166
RelOptInfo::partitioned_child_rels
List * partitioned_child_rels
Definition: pathnodes.h:756
RangeTblEntry::subquery
Query * subquery
Definition: parsenodes.h:1012
Query::groupClause
List * groupClause
Definition: parsenodes.h:148
reparameterize_path
Path * reparameterize_path(PlannerInfo *root, Path *path, Relids required_outer, double loop_count)
Definition: pathnode.c:3758
RelOptInfo::max_attr
AttrNumber max_attr
Definition: pathnodes.h:698
set_values_size_estimates
void set_values_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5289
add_partial_path
void add_partial_path(RelOptInfo *parent_rel, Path *new_path)
Definition: pathnode.c:749
IS_PARTITIONED_REL
#define IS_PARTITIONED_REL(rel)
Definition: pathnodes.h:767
Query::hasSubLinks
bool hasSubLinks
Definition: parsenodes.h:128
set_plain_rel_size
static void set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:574
list_free
void list_free(List *list)
Definition: list.c:1376
elog
#define elog(elevel,...)
Definition: elog.h:214
i
int i
Definition: preproc-comment.c:23
TargetEntry::ressortgroupref
Index ressortgroupref
Definition: primnodes.h:1413
has_useful_pathkeys
bool has_useful_pathkeys(PlannerInfo *root, RelOptInfo *rel)
Definition: pathkeys.c:1910
parse_clause.h
RelOptInfo::part_scheme
PartitionScheme part_scheme
Definition: pathnodes.h:743
Path::parallel_aware
bool parallel_aware
Definition: pathnodes.h:1150
PlannerInfo::initial_rels
List * initial_rels
Definition: pathnodes.h:308
set_tablefunc_size_estimates
void set_tablefunc_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5267
RelOptInfo::pathlist
List * pathlist
Definition: pathnodes.h:680
create_plain_partial_paths
static void create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel)
Definition: allpaths.c:796
ParamPathInfo::ppi_req_outer
Relids ppi_req_outer
Definition: pathnodes.h:1104
set_function_size_estimates
void set_function_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5229
AppendRelInfo::child_relid
Index child_relid
Definition: pathnodes.h:2231
RangeTblEntry::eref
Alias * eref
Definition: parsenodes.h:1116
find_base_rel
RelOptInfo * find_base_rel(PlannerInfo *root, int relid)
Definition: relnode.c:374
copyObject
#define copyObject(obj)
Definition: nodes.h:644
set_function_pathlist
static void set_function_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2392
Query::havingQual
Node * havingQual
Definition: parsenodes.h:152
AppendRelInfo::parent_relid
Index parent_relid
Definition: pathnodes.h:2230
RelOptInfo::attr_widths
int32 * attr_widths
Definition: pathnodes.h:700
create_partial_bitmap_paths
void create_partial_bitmap_paths(PlannerInfo *root, RelOptInfo *rel, Path *bitmapqual)
Definition: allpaths.c:3791
set_foreign_size_estimates
void set_foreign_size_estimates(PlannerInfo *root, RelOptInfo *rel)
Definition: costsize.c:5420
T_Path
Definition: nodes.h:227
clamp_row_est
double clamp_row_est(double nrows)
Definition: costsize.c:189
generate_orderedappend_paths
static void generate_orderedappend_paths(PlannerInfo *root, RelOptInfo *rel, List *live_childrels, List *all_child_pathkeys, List *partitioned_rels)
Definition: allpaths.c:1750
vars
Definition: regcomp.c:224
accumulate_append_subpath
static void accumulate_append_subpath(Path *path, List **subpaths, List **special_subpaths)
Definition: allpaths.c:2049
max_parallel_workers_per_gather
int max_parallel_workers_per_gather
Definition: costsize.c:123
List
Definition: pg_list.h:50
bms_is_member
bool bms_is_member(int x, const Bitmapset *a)
Definition: bitmapset.c:427
RelOptInfo::reltarget
struct PathTarget * reltarget
Definition: pathnodes.h:677
set_values_pathlist
static void set_values_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2460
RangeTblEntry::tablesample
struct TableSampleClause * tablesample
Definition: parsenodes.h:1007
AttrNumber
int16 AttrNumber
Definition: attnum.h:21
set_rel_size
static void set_rel_size(PlannerInfo *root, RelOptInfo *rel, Index rti, RangeTblEntry *rte)
Definition: allpaths.c:361
RelOptInfo::subplan_params
List * subplan_params
Definition: pathnodes.h:711
subquery_planner
PlannerInfo * subquery_planner(PlannerGlobal *glob, Query *parse, PlannerInfo *parent_root, bool hasRecursion, double tuple_fraction)
Definition: planner.c:597
create_seqscan_path
Path * create_seqscan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer, int parallel_workers)
Definition: pathnode.c:929
compute_bitmap_pages
double compute_bitmap_pages(PlannerInfo *root, RelOptInfo *baserel, Path *bitmapqual, int loop_count, Cost *cost, double *tuple)
Definition: costsize.c:5745
subpath
Datum subpath(PG_FUNCTION_ARGS)
Definition: ltree_op.c:241
lfirst_oid
#define lfirst_oid(lc)
Definition: pg_list.h:191
bms_equal
bool bms_equal(const Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:94
parse
static struct subre * parse(struct vars *, int, int, struct state *, struct state *)
Definition: regcomp.c:648
adjust_appendrel_attrs
Node * adjust_appendrel_attrs(PlannerInfo *root, Node *node, int nappinfos, AppendRelInfo **appinfos)
Definition: appendinfo.c:194
bool
unsigned char bool
Definition: c.h:324
create_bitmap_heap_path
BitmapHeapPath * create_bitmap_heap_path(PlannerInfo *root, RelOptInfo *rel, Path *bitmapqual, Relids required_outer, double loop_count, int parallel_degree)
Definition: pathnode.c:1046
create_namedtuplestorescan_path
Path * create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2048
set_worktable_pathlist
static void set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
Definition: allpaths.c:2630
create_ctescan_path
Path * create_ctescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
Definition: pathnode.c:2023
RelOptInfo::min_attr
AttrNumber min_attr
Definition: pathnodes.h:697