Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * pathnode.c
4 : : * Routines to manipulate pathlists and create path nodes
5 : : *
6 : : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/optimizer/util/pathnode.c
12 : : *
13 : : *-------------------------------------------------------------------------
14 : : */
15 : : #include "postgres.h"
16 : :
17 : : #include <math.h>
18 : :
19 : : #include "foreign/fdwapi.h"
20 : : #include "miscadmin.h"
21 : : #include "nodes/extensible.h"
22 : : #include "optimizer/appendinfo.h"
23 : : #include "optimizer/clauses.h"
24 : : #include "optimizer/cost.h"
25 : : #include "optimizer/optimizer.h"
26 : : #include "optimizer/pathnode.h"
27 : : #include "optimizer/paths.h"
28 : : #include "optimizer/planmain.h"
29 : : #include "optimizer/tlist.h"
30 : : #include "parser/parsetree.h"
31 : : #include "utils/memutils.h"
32 : : #include "utils/selfuncs.h"
33 : :
34 : : typedef enum
35 : : {
36 : : COSTS_EQUAL, /* path costs are fuzzily equal */
37 : : COSTS_BETTER1, /* first path is cheaper than second */
38 : : COSTS_BETTER2, /* second path is cheaper than first */
39 : : COSTS_DIFFERENT, /* neither path dominates the other on cost */
40 : : } PathCostComparison;
41 : :
42 : : /*
43 : : * STD_FUZZ_FACTOR is the normal fuzz factor for compare_path_costs_fuzzily.
44 : : * XXX is it worth making this user-controllable? It provides a tradeoff
45 : : * between planner runtime and the accuracy of path cost comparisons.
46 : : */
47 : : #define STD_FUZZ_FACTOR 1.01
48 : :
49 : : static List *translate_sub_tlist(List *tlist, int relid);
50 : : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
51 : : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
52 : : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
53 : : List *pathlist,
54 : : RelOptInfo *child_rel);
55 : : static bool pathlist_is_reparameterizable_by_child(List *pathlist,
56 : : RelOptInfo *child_rel);
57 : :
58 : :
59 : : /*****************************************************************************
60 : : * MISC. PATH UTILITIES
61 : : *****************************************************************************/
62 : :
63 : : /*
64 : : * compare_path_costs
65 : : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
66 : : * or more expensive than path2 for the specified criterion.
67 : : */
68 : : int
8825 tgl@sss.pgh.pa.us 69 :CBC 400190 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
70 : : {
71 [ + + ]: 400190 : if (criterion == STARTUP_COST)
72 : : {
73 [ + + ]: 205001 : if (path1->startup_cost < path2->startup_cost)
74 : 130286 : return -1;
75 [ + + ]: 74715 : if (path1->startup_cost > path2->startup_cost)
76 : 34652 : return +1;
77 : :
78 : : /*
79 : : * If paths have the same startup cost (not at all unlikely), order
80 : : * them by total cost.
81 : : */
82 [ + + ]: 40063 : if (path1->total_cost < path2->total_cost)
83 : 18046 : return -1;
84 [ + + ]: 22017 : if (path1->total_cost > path2->total_cost)
85 : 2215 : return +1;
86 : : }
87 : : else
88 : : {
89 [ + + ]: 195189 : if (path1->total_cost < path2->total_cost)
90 : 185868 : return -1;
91 [ + + ]: 9321 : if (path1->total_cost > path2->total_cost)
92 : 659 : return +1;
93 : :
94 : : /*
95 : : * If paths have the same total cost, order them by startup cost.
96 : : */
97 [ + + ]: 8662 : if (path1->startup_cost < path2->startup_cost)
98 : 56 : return -1;
99 [ + + ]: 8606 : if (path1->startup_cost > path2->startup_cost)
100 : 12 : return +1;
101 : : }
102 : 28396 : return 0;
103 : : }
104 : :
105 : : /*
106 : : * compare_fractional_path_costs
107 : : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
108 : : * or more expensive than path2 for fetching the specified fraction
109 : : * of the total tuples.
110 : : *
111 : : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
112 : : * path with the cheaper total_cost.
113 : : */
114 : : int
115 : 1473 : compare_fractional_path_costs(Path *path1, Path *path2,
116 : : double fraction)
117 : : {
118 : : Cost cost1,
119 : : cost2;
120 : :
121 [ + + + + ]: 1473 : if (fraction <= 0.0 || fraction >= 1.0)
122 : 936 : return compare_path_costs(path1, path2, TOTAL_COST);
123 : 537 : cost1 = path1->startup_cost +
124 : 537 : fraction * (path1->total_cost - path1->startup_cost);
125 : 537 : cost2 = path2->startup_cost +
126 : 537 : fraction * (path2->total_cost - path2->startup_cost);
127 [ + + ]: 537 : if (cost1 < cost2)
128 : 338 : return -1;
129 [ + - ]: 199 : if (cost1 > cost2)
130 : 199 : return +1;
8825 tgl@sss.pgh.pa.us 131 :UBC 0 : return 0;
132 : : }
133 : :
134 : : /*
135 : : * compare_path_costs_fuzzily
136 : : * Compare the costs of two paths to see if either can be said to
137 : : * dominate the other.
138 : : *
139 : : * We use fuzzy comparisons so that add_path() can avoid keeping both of
140 : : * a pair of paths that really have insignificantly different cost.
141 : : *
142 : : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
143 : : * fraction of the smaller cost that is considered to be a significant
144 : : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
145 : : * be 1% of the smaller cost.
146 : : *
147 : : * The two paths are said to have "equal" costs if both startup and total
148 : : * costs are fuzzily the same. Path1 is said to be better than path2 if
149 : : * it has fuzzily better startup cost and fuzzily no worse total cost,
150 : : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
151 : : * Path2 is better than path1 if the reverse holds. Finally, if one path
152 : : * is fuzzily better than the other on startup cost and fuzzily worse on
153 : : * total cost, we just say that their costs are "different", since neither
154 : : * dominates the other across the whole performance spectrum.
155 : : *
156 : : * This function also enforces a policy rule that paths for which the relevant
157 : : * one of parent->consider_startup and parent->consider_param_startup is false
158 : : * cannot survive comparisons solely on the grounds of good startup cost, so
159 : : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
160 : : * (But if total costs are fuzzily equal, we compare startup costs anyway,
161 : : * in hopes of eliminating one path or the other.)
162 : : */
163 : : static PathCostComparison
3238 tgl@sss.pgh.pa.us 164 :CBC 1743019 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
165 : : {
166 : : #define CONSIDER_PATH_STARTUP_COST(p) \
167 : : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
168 : :
169 : : /*
170 : : * Check total cost first since it's more likely to be different; many
171 : : * paths have zero startup cost.
172 : : */
4376 173 [ + + ]: 1743019 : if (path1->total_cost > path2->total_cost * fuzz_factor)
174 : : {
175 : : /* path1 fuzzily worse on total cost */
3238 176 [ + + + + ]: 901484 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
177 [ + + ]: 57445 : path2->startup_cost > path1->startup_cost * fuzz_factor)
178 : : {
179 : : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
4461 180 : 38906 : return COSTS_DIFFERENT;
181 : : }
182 : : /* else path2 dominates */
183 : 862578 : return COSTS_BETTER2;
184 : : }
4376 185 [ + + ]: 841535 : if (path2->total_cost > path1->total_cost * fuzz_factor)
186 : : {
187 : : /* path2 fuzzily worse on total cost */
3238 188 [ + + + + ]: 433510 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
189 [ + + ]: 24753 : path1->startup_cost > path2->startup_cost * fuzz_factor)
190 : : {
191 : : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
4461 192 : 16300 : return COSTS_DIFFERENT;
193 : : }
194 : : /* else path1 dominates */
195 : 417210 : return COSTS_BETTER1;
196 : : }
197 : : /* fuzzily the same on total cost ... */
3238 198 [ + + ]: 408025 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
199 : : {
200 : : /* ... but path1 fuzzily worse on startup, so path2 wins */
4461 201 : 153039 : return COSTS_BETTER2;
202 : : }
3238 203 [ + + ]: 254986 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
204 : : {
205 : : /* ... but path2 fuzzily worse on startup, so path1 wins */
4461 206 : 24137 : return COSTS_BETTER1;
207 : : }
208 : : /* fuzzily the same on both costs */
209 : 230849 : return COSTS_EQUAL;
210 : :
211 : : #undef CONSIDER_PATH_STARTUP_COST
212 : : }
213 : :
214 : : /*
215 : : * set_cheapest
216 : : * Find the minimum-cost paths from among a relation's paths,
217 : : * and save them in the rel's cheapest-path fields.
218 : : *
219 : : * cheapest_total_path is normally the cheapest-total-cost unparameterized
220 : : * path; but if there are no unparameterized paths, we assign it to be the
221 : : * best (cheapest least-parameterized) parameterized path. However, only
222 : : * unparameterized paths are considered candidates for cheapest_startup_path,
223 : : * so that will be NULL if there are no unparameterized paths.
224 : : *
225 : : * The cheapest_parameterized_paths list collects all parameterized paths
226 : : * that have survived the add_path() tournament for this relation. (Since
227 : : * add_path ignores pathkeys for a parameterized path, these will be paths
228 : : * that have best cost or best row count for their parameterization. We
229 : : * may also have both a parallel-safe and a non-parallel-safe path in some
230 : : * cases for the same parameterization in some cases, but this should be
231 : : * relatively rare since, most typically, all paths for the same relation
232 : : * will be parallel-safe or none of them will.)
233 : : *
234 : : * cheapest_parameterized_paths always includes the cheapest-total
235 : : * unparameterized path, too, if there is one; the users of that list find
236 : : * it more convenient if that's included.
237 : : *
238 : : * This is normally called only after we've finished constructing the path
239 : : * list for the rel node.
240 : : */
241 : : void
8825 242 : 944967 : set_cheapest(RelOptInfo *parent_rel)
243 : : {
244 : : Path *cheapest_startup_path;
245 : : Path *cheapest_total_path;
246 : : Path *best_param_path;
247 : : List *parameterized_paths;
248 : : ListCell *p;
249 : :
9402 bruce@momjian.us 250 [ - + ]: 944967 : Assert(IsA(parent_rel, RelOptInfo));
251 : :
4246 tgl@sss.pgh.pa.us 252 [ - + ]: 944967 : if (parent_rel->pathlist == NIL)
4246 tgl@sss.pgh.pa.us 253 [ # # ]:UBC 0 : elog(ERROR, "could not devise a query plan for the given query");
254 : :
4246 tgl@sss.pgh.pa.us 255 :CBC 944967 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
256 : 944967 : parameterized_paths = NIL;
257 : :
4461 258 [ + - + + : 2104570 : foreach(p, parent_rel->pathlist)
+ + ]
259 : : {
9715 bruce@momjian.us 260 : 1159603 : Path *path = (Path *) lfirst(p);
261 : : int cmp;
262 : :
4378 tgl@sss.pgh.pa.us 263 [ + + ]: 1159603 : if (path->param_info)
264 : : {
265 : : /* Parameterized path, so add it to parameterized_paths */
4246 266 : 55750 : parameterized_paths = lappend(parameterized_paths, path);
267 : :
268 : : /*
269 : : * If we have an unparameterized cheapest-total, we no longer care
270 : : * about finding the best parameterized path, so move on.
271 : : */
272 [ + + ]: 55750 : if (cheapest_total_path)
273 : 10254 : continue;
274 : :
275 : : /*
276 : : * Otherwise, track the best parameterized path, which is the one
277 : : * with least total cost among those of the minimum
278 : : * parameterization.
279 : : */
280 [ + + ]: 45496 : if (best_param_path == NULL)
281 : 41953 : best_param_path = path;
282 : : else
283 : : {
284 [ + - + + : 3543 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
+ + - ]
285 [ + - ]: 3543 : PATH_REQ_OUTER(best_param_path)))
286 : : {
287 : 27 : case BMS_EQUAL:
288 : : /* keep the cheaper one */
289 [ - + ]: 27 : if (compare_path_costs(path, best_param_path,
290 : : TOTAL_COST) < 0)
4246 tgl@sss.pgh.pa.us 291 :UBC 0 : best_param_path = path;
4246 tgl@sss.pgh.pa.us 292 :CBC 27 : break;
293 : 170 : case BMS_SUBSET1:
294 : : /* new path is less-parameterized */
295 : 170 : best_param_path = path;
296 : 170 : break;
297 : 2 : case BMS_SUBSET2:
298 : : /* old path is less-parameterized, keep it */
299 : 2 : break;
300 : 3344 : case BMS_DIFFERENT:
301 : :
302 : : /*
303 : : * This means that neither path has the least possible
304 : : * parameterization for the rel. We'll sit on the old
305 : : * path until something better comes along.
306 : : */
307 : 3344 : break;
308 : : }
309 : : }
310 : : }
311 : : else
312 : : {
313 : : /* Unparameterized path, so consider it for cheapest slots */
314 [ + + ]: 1103853 : if (cheapest_total_path == NULL)
315 : : {
316 : 939692 : cheapest_startup_path = cheapest_total_path = path;
317 : 939692 : continue;
318 : : }
319 : :
320 : : /*
321 : : * If we find two paths of identical costs, try to keep the
322 : : * better-sorted one. The paths might have unrelated sort
323 : : * orderings, in which case we can only guess which might be
324 : : * better to keep, but if one is superior then we definitely
325 : : * should keep that one.
326 : : */
327 : 164161 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
328 [ + + + + ]: 164161 : if (cmp > 0 ||
329 [ - + ]: 180 : (cmp == 0 &&
330 : 180 : compare_pathkeys(cheapest_startup_path->pathkeys,
331 : : path->pathkeys) == PATHKEYS_BETTER2))
332 : 28014 : cheapest_startup_path = path;
333 : :
334 : 164161 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
335 [ + - + + ]: 164161 : if (cmp > 0 ||
4246 tgl@sss.pgh.pa.us 336 [ - + ]:GBC 48 : (cmp == 0 &&
337 : 48 : compare_pathkeys(cheapest_total_path->pathkeys,
338 : : path->pathkeys) == PATHKEYS_BETTER2))
4246 tgl@sss.pgh.pa.us 339 :UBC 0 : cheapest_total_path = path;
340 : : }
341 : : }
342 : :
343 : : /* Add cheapest unparameterized path, if any, to parameterized_paths */
4246 tgl@sss.pgh.pa.us 344 [ + + ]:CBC 944967 : if (cheapest_total_path)
345 : 939692 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
346 : :
347 : : /*
348 : : * If there is no unparameterized path, use the best parameterized path as
349 : : * cheapest_total_path (but not as cheapest_startup_path).
350 : : */
351 [ + + ]: 944967 : if (cheapest_total_path == NULL)
352 : 5275 : cheapest_total_path = best_param_path;
353 [ - + ]: 944967 : Assert(cheapest_total_path != NULL);
354 : :
8825 355 : 944967 : parent_rel->cheapest_startup_path = cheapest_startup_path;
356 : 944967 : parent_rel->cheapest_total_path = cheapest_total_path;
7559 bruce@momjian.us 357 : 944967 : parent_rel->cheapest_unique_path = NULL; /* computed only if needed */
4246 tgl@sss.pgh.pa.us 358 : 944967 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
8833 359 : 944967 : }
360 : :
361 : : /*
362 : : * add_path
363 : : * Consider a potential implementation path for the specified parent rel,
364 : : * and add it to the rel's pathlist if it is worthy of consideration.
365 : : * A path is worthy if it has a better sort order (better pathkeys) or
366 : : * cheaper cost (on either dimension), or generates fewer rows, than any
367 : : * existing path that has the same or superset parameterization rels.
368 : : * We also consider parallel-safe paths more worthy than others.
369 : : *
370 : : * We also remove from the rel's pathlist any old paths that are dominated
371 : : * by new_path --- that is, new_path is cheaper, at least as well ordered,
372 : : * generates no more rows, requires no outer rels not required by the old
373 : : * path, and is no less parallel-safe.
374 : : *
375 : : * In most cases, a path with a superset parameterization will generate
376 : : * fewer rows (since it has more join clauses to apply), so that those two
377 : : * figures of merit move in opposite directions; this means that a path of
378 : : * one parameterization can seldom dominate a path of another. But such
379 : : * cases do arise, so we make the full set of checks anyway.
380 : : *
381 : : * There are two policy decisions embedded in this function, along with
382 : : * its sibling add_path_precheck. First, we treat all parameterized paths
383 : : * as having NIL pathkeys, so that they cannot win comparisons on the
384 : : * basis of sort order. This is to reduce the number of parameterized
385 : : * paths that are kept; see discussion in src/backend/optimizer/README.
386 : : *
387 : : * Second, we only consider cheap startup cost to be interesting if
388 : : * parent_rel->consider_startup is true for an unparameterized path, or
389 : : * parent_rel->consider_param_startup is true for a parameterized one.
390 : : * Again, this allows discarding useless paths sooner.
391 : : *
392 : : * The pathlist is kept sorted by total_cost, with cheaper paths
393 : : * at the front. Within this routine, that's simply a speed hack:
394 : : * doing it that way makes it more likely that we will reject an inferior
395 : : * path after a few comparisons, rather than many comparisons.
396 : : * However, add_path_precheck relies on this ordering to exit early
397 : : * when possible.
398 : : *
399 : : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
400 : : * memory consumption. We dare not try to free the substructure of a Path,
401 : : * since much of it may be shared with other Paths or the query tree itself;
402 : : * but just recycling discarded Path nodes is a very useful savings in
403 : : * a large join tree. We can recycle the List nodes of pathlist, too.
404 : : *
405 : : * As noted in optimizer/README, deleting a previously-accepted Path is
406 : : * safe because we know that Paths of this rel cannot yet be referenced
407 : : * from any other rel, such as a higher-level join. However, in some cases
408 : : * it is possible that a Path is referenced by another Path for its own
409 : : * rel; we must not delete such a Path, even if it is dominated by the new
410 : : * Path. Currently this occurs only for IndexPath objects, which may be
411 : : * referenced as children of BitmapHeapPaths as well as being paths in
412 : : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
413 : : *
414 : : * 'parent_rel' is the relation entry to which the path corresponds.
415 : : * 'new_path' is a potential path for parent_rel.
416 : : *
417 : : * Returns nothing, but modifies parent_rel->pathlist.
418 : : */
419 : : void
420 : 1834978 : add_path(RelOptInfo *parent_rel, Path *new_path)
421 : : {
2489 422 : 1834978 : bool accept_new = true; /* unless we find a superior old path */
1735 423 : 1834978 : int insert_at = 0; /* where to insert new item */
424 : : List *new_path_pathkeys;
425 : : ListCell *p1;
426 : :
427 : : /*
428 : : * This is a convenient place to check for query cancel --- no part of the
429 : : * planner goes very long without calling add_path().
430 : : */
6890 431 [ - + ]: 1834978 : CHECK_FOR_INTERRUPTS();
432 : :
433 : : /* Pretend parameterized paths have no pathkeys, per comment above */
4378 434 [ + + ]: 1834978 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
435 : :
436 : : /*
437 : : * Loop to check proposed new path against old paths. Note it is possible
438 : : * for more than one old path to be tossed out because new_path dominates
439 : : * it.
440 : : */
1735 441 [ + + + + : 2776512 : foreach(p1, parent_rel->pathlist)
+ + ]
442 : : {
8833 443 : 1596047 : Path *old_path = (Path *) lfirst(p1);
8768 bruce@momjian.us 444 : 1596047 : bool remove_old = false; /* unless new proves superior */
445 : : PathCostComparison costcmp;
446 : : PathKeysComparison keyscmp;
447 : : BMS_Comparison outercmp;
448 : :
449 : : /*
450 : : * Do a fuzzy cost comparison with standard fuzziness limit.
451 : : */
3238 tgl@sss.pgh.pa.us 452 : 1596047 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
453 : : STD_FUZZ_FACTOR);
454 : :
455 : : /*
456 : : * If the two paths compare differently for startup and total cost,
457 : : * then we want to keep both, and we can skip comparing pathkeys and
458 : : * required_outer rels. If they compare the same, proceed with the
459 : : * other comparisons. Row count is checked last. (We make the tests
460 : : * in this order because the cost comparison is most likely to turn
461 : : * out "different", and the pathkeys comparison next most likely. As
462 : : * explained above, row count very seldom makes a difference, so even
463 : : * though it's cheap to compare there's not much point in checking it
464 : : * earlier.)
465 : : */
4461 466 [ + + ]: 1596047 : if (costcmp != COSTS_DIFFERENT)
467 : : {
468 : : /* Similarly check to see if either dominates on pathkeys */
469 : : List *old_path_pathkeys;
470 : :
4378 471 [ + + ]: 1540856 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
4461 472 : 1540856 : keyscmp = compare_pathkeys(new_path_pathkeys,
473 : : old_path_pathkeys);
474 [ + + ]: 1540856 : if (keyscmp != PATHKEYS_DIFFERENT)
475 : : {
476 [ + + + - : 1469989 : switch (costcmp)
- ]
477 : : {
478 : 157162 : case COSTS_EQUAL:
4378 479 [ + + ]: 157162 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
2489 480 [ + + ]: 157162 : PATH_REQ_OUTER(old_path));
4461 481 [ + + ]: 157162 : if (keyscmp == PATHKEYS_BETTER1)
482 : : {
4378 483 [ + + + - ]: 1763 : if ((outercmp == BMS_EQUAL ||
484 : 1763 : outercmp == BMS_SUBSET1) &&
3007 rhaas@postgresql.org 485 [ + + ]: 1763 : new_path->rows <= old_path->rows &&
486 [ + - ]: 1759 : new_path->parallel_safe >= old_path->parallel_safe)
2489 tgl@sss.pgh.pa.us 487 : 1759 : remove_old = true; /* new dominates old */
488 : : }
4461 489 [ + + ]: 155399 : else if (keyscmp == PATHKEYS_BETTER2)
490 : : {
4378 491 [ + + + - ]: 6476 : if ((outercmp == BMS_EQUAL ||
492 : 6476 : outercmp == BMS_SUBSET2) &&
3007 rhaas@postgresql.org 493 [ + + ]: 6476 : new_path->rows >= old_path->rows &&
494 [ + - ]: 6403 : new_path->parallel_safe <= old_path->parallel_safe)
2489 tgl@sss.pgh.pa.us 495 : 6403 : accept_new = false; /* old dominates new */
496 : : }
497 : : else /* keyscmp == PATHKEYS_EQUAL */
498 : : {
4461 499 [ + + ]: 148923 : if (outercmp == BMS_EQUAL)
500 : : {
501 : : /*
502 : : * Same pathkeys and outer rels, and fuzzily
503 : : * the same cost, so keep just one; to decide
504 : : * which, first check parallel-safety, then
505 : : * rows, then do a fuzzy cost comparison with
506 : : * very small fuzz limit. (We used to do an
507 : : * exact cost comparison, but that results in
508 : : * annoying platform-specific plan variations
509 : : * due to roundoff in the cost estimates.) If
510 : : * things are still tied, arbitrarily keep
511 : : * only the old path. Notice that we will
512 : : * keep only the old path even if the
513 : : * less-fuzzy comparison decides the startup
514 : : * and total costs compare differently.
515 : : */
3007 rhaas@postgresql.org 516 : 147089 : if (new_path->parallel_safe >
517 [ + + ]: 147089 : old_path->parallel_safe)
518 : 24 : remove_old = true; /* new dominates old */
519 : 147065 : else if (new_path->parallel_safe <
520 [ + + ]: 147065 : old_path->parallel_safe)
521 : 71 : accept_new = false; /* old dominates new */
522 [ + + ]: 146994 : else if (new_path->rows < old_path->rows)
4378 tgl@sss.pgh.pa.us 523 : 12 : remove_old = true; /* new dominates old */
524 [ + + ]: 146982 : else if (new_path->rows > old_path->rows)
4326 bruce@momjian.us 525 : 10 : accept_new = false; /* old dominates new */
4243 tgl@sss.pgh.pa.us 526 [ + + ]: 146972 : else if (compare_path_costs_fuzzily(new_path,
527 : : old_path,
528 : : 1.0000000001) == COSTS_BETTER1)
4461 529 : 5982 : remove_old = true; /* new dominates old */
530 : : else
4326 bruce@momjian.us 531 : 140990 : accept_new = false; /* old equals or
532 : : * dominates new */
533 : : }
4378 tgl@sss.pgh.pa.us 534 [ + + ]: 1834 : else if (outercmp == BMS_SUBSET1 &&
3007 rhaas@postgresql.org 535 [ + + ]: 412 : new_path->rows <= old_path->rows &&
536 [ + - ]: 398 : new_path->parallel_safe >= old_path->parallel_safe)
2489 tgl@sss.pgh.pa.us 537 : 398 : remove_old = true; /* new dominates old */
4378 538 [ + + ]: 1436 : else if (outercmp == BMS_SUBSET2 &&
3007 rhaas@postgresql.org 539 [ + + ]: 1187 : new_path->rows >= old_path->rows &&
540 [ + - ]: 1177 : new_path->parallel_safe <= old_path->parallel_safe)
2489 tgl@sss.pgh.pa.us 541 : 1177 : accept_new = false; /* old dominates new */
542 : : /* else different parameterizations, keep both */
543 : : }
4461 544 : 157162 : break;
545 : 424830 : case COSTS_BETTER1:
546 [ + + ]: 424830 : if (keyscmp != PATHKEYS_BETTER2)
547 : : {
4378 548 [ + + ]: 297066 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
2489 549 [ + + ]: 297066 : PATH_REQ_OUTER(old_path));
4378 550 [ + + + + ]: 297066 : if ((outercmp == BMS_EQUAL ||
551 : 254523 : outercmp == BMS_SUBSET1) &&
3007 rhaas@postgresql.org 552 [ + + ]: 254523 : new_path->rows <= old_path->rows &&
553 [ + + ]: 252783 : new_path->parallel_safe >= old_path->parallel_safe)
2489 tgl@sss.pgh.pa.us 554 : 251449 : remove_old = true; /* new dominates old */
555 : : }
4461 556 : 424830 : break;
557 : 887997 : case COSTS_BETTER2:
558 [ + + ]: 887997 : if (keyscmp != PATHKEYS_BETTER1)
559 : : {
4378 560 [ + + ]: 573421 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
2489 561 [ + + ]: 573421 : PATH_REQ_OUTER(old_path));
4378 562 [ + + + + ]: 573421 : if ((outercmp == BMS_EQUAL ||
563 : 532016 : outercmp == BMS_SUBSET2) &&
3007 rhaas@postgresql.org 564 [ + + ]: 532016 : new_path->rows >= old_path->rows &&
565 [ + + ]: 506706 : new_path->parallel_safe <= old_path->parallel_safe)
2489 tgl@sss.pgh.pa.us 566 : 505862 : accept_new = false; /* old dominates new */
567 : : }
4461 568 : 887997 : break;
4461 tgl@sss.pgh.pa.us 569 :UBC 0 : case COSTS_DIFFERENT:
570 : :
571 : : /*
572 : : * can't get here, but keep this case to keep compiler
573 : : * quiet
574 : : */
575 : 0 : break;
576 : : }
577 : : }
578 : : }
579 : :
580 : : /*
581 : : * Remove current element from pathlist if dominated by new.
582 : : */
7294 neilc@samurai.com 583 [ + + ]:CBC 1596047 : if (remove_old)
584 : : {
1735 tgl@sss.pgh.pa.us 585 : 259624 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
586 : : p1);
587 : :
588 : : /*
589 : : * Delete the data pointed-to by the deleted cell, if possible
590 : : */
6932 591 [ + + ]: 259624 : if (!IsA(old_path, IndexPath))
592 : 252443 : pfree(old_path);
593 : : }
594 : : else
595 : : {
596 : : /* new belongs after this old path if it has cost >= old's */
4461 597 [ + + ]: 1336423 : if (new_path->total_cost >= old_path->total_cost)
1735 598 : 1115070 : insert_at = foreach_current_index(p1) + 1;
599 : : }
600 : :
601 : : /*
602 : : * If we found an old path that dominates new_path, we can quit
603 : : * scanning the pathlist; we will not add new_path, and we assume
604 : : * new_path cannot dominate any other elements of the pathlist.
605 : : */
8768 bruce@momjian.us 606 [ + + ]: 1596047 : if (!accept_new)
8833 tgl@sss.pgh.pa.us 607 : 654513 : break;
608 : : }
609 : :
610 [ + + ]: 1834978 : if (accept_new)
611 : : {
612 : : /* Accept the new path: insert it at proper place in pathlist */
1735 613 : 1180465 : parent_rel->pathlist =
614 : 1180465 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
615 : : }
616 : : else
617 : : {
618 : : /* Reject and recycle the new path */
6932 619 [ + + ]: 654513 : if (!IsA(new_path, IndexPath))
620 : 616684 : pfree(new_path);
621 : : }
10141 scrappy@hub.org 622 : 1834978 : }
623 : :
624 : : /*
625 : : * add_path_precheck
626 : : * Check whether a proposed new path could possibly get accepted.
627 : : * We assume we know the path's pathkeys and parameterization accurately,
628 : : * and have lower bounds for its costs.
629 : : *
630 : : * Note that we do not know the path's rowcount, since getting an estimate for
631 : : * that is too expensive to do before prechecking. We assume here that paths
632 : : * of a superset parameterization will generate fewer rows; if that holds,
633 : : * then paths with different parameterizations cannot dominate each other
634 : : * and so we can simply ignore existing paths of another parameterization.
635 : : * (In the infrequent cases where that rule of thumb fails, add_path will
636 : : * get rid of the inferior path.)
637 : : *
638 : : * At the time this is called, we haven't actually built a Path structure,
639 : : * so the required information has to be passed piecemeal.
640 : : */
641 : : bool
4461 tgl@sss.pgh.pa.us 642 : 1883895 : add_path_precheck(RelOptInfo *parent_rel,
643 : : Cost startup_cost, Cost total_cost,
644 : : List *pathkeys, Relids required_outer)
645 : : {
646 : : List *new_path_pathkeys;
647 : : bool consider_startup;
648 : : ListCell *p1;
649 : :
650 : : /* Pretend parameterized paths have no pathkeys, per add_path policy */
651 [ + + ]: 1883895 : new_path_pathkeys = required_outer ? NIL : pathkeys;
652 : :
653 : : /* Decide whether new path's startup cost is interesting */
3238 654 [ + + ]: 1883895 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
655 : :
4461 656 [ + + + + : 2457669 : foreach(p1, parent_rel->pathlist)
+ + ]
657 : : {
658 : 2333520 : Path *old_path = (Path *) lfirst(p1);
659 : : PathKeysComparison keyscmp;
660 : :
661 : : /*
662 : : * We are looking for an old_path with the same parameterization (and
663 : : * by assumption the same rowcount) that dominates the new path on
664 : : * pathkeys as well as both cost metrics. If we find one, we can
665 : : * reject the new path.
666 : : *
667 : : * Cost comparisons here should match compare_path_costs_fuzzily.
668 : : */
3238 669 [ + + ]: 2333520 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
670 : : {
671 : : /* new path can win on startup cost only if consider_startup */
672 [ + + ]: 1653556 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
673 [ + + ]: 740902 : !consider_startup)
674 : : {
675 : : /* new path loses on cost, so check pathkeys... */
676 : : List *old_path_pathkeys;
677 : :
4378 678 [ + + ]: 1607951 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
4461 679 : 1607951 : keyscmp = compare_pathkeys(new_path_pathkeys,
680 : : old_path_pathkeys);
681 [ + + + + ]: 1607951 : if (keyscmp == PATHKEYS_EQUAL ||
682 : : keyscmp == PATHKEYS_BETTER2)
683 : : {
684 : : /* new path does not win on pathkeys... */
4378 685 [ + + + + ]: 1117231 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
686 : : {
687 : : /* Found an old path that dominates the new one */
4461 688 : 1079782 : return false;
689 : : }
690 : : }
691 : : }
692 : : }
693 : : else
694 : : {
695 : : /*
696 : : * Since the pathlist is sorted by total_cost, we can stop looking
697 : : * once we reach a path with a total_cost larger than the new
698 : : * path's.
699 : : */
700 : 679964 : break;
701 : : }
702 : : }
703 : :
704 : 804113 : return true;
705 : : }
706 : :
707 : : /*
708 : : * add_partial_path
709 : : * Like add_path, our goal here is to consider whether a path is worthy
710 : : * of being kept around, but the considerations here are a bit different.
711 : : * A partial path is one which can be executed in any number of workers in
712 : : * parallel such that each worker will generate a subset of the path's
713 : : * overall result.
714 : : *
715 : : * As in add_path, the partial_pathlist is kept sorted with the cheapest
716 : : * total path in front. This is depended on by multiple places, which
717 : : * just take the front entry as the cheapest path without searching.
718 : : *
719 : : * We don't generate parameterized partial paths for several reasons. Most
720 : : * importantly, they're not safe to execute, because there's nothing to
721 : : * make sure that a parallel scan within the parameterized portion of the
722 : : * plan is running with the same value in every worker at the same time.
723 : : * Fortunately, it seems unlikely to be worthwhile anyway, because having
724 : : * each worker scan the entire outer relation and a subset of the inner
725 : : * relation will generally be a terrible plan. The inner (parameterized)
726 : : * side of the plan will be small anyway. There could be rare cases where
727 : : * this wins big - e.g. if join order constraints put a 1-row relation on
728 : : * the outer side of the topmost join with a parameterized plan on the inner
729 : : * side - but we'll have to be content not to handle such cases until
730 : : * somebody builds an executor infrastructure that can cope with them.
731 : : *
732 : : * Because we don't consider parameterized paths here, we also don't
733 : : * need to consider the row counts as a measure of quality: every path will
734 : : * produce the same number of rows. Neither do we need to consider startup
735 : : * costs: parallelism is only used for plans that will be run to completion.
736 : : * Therefore, this routine is much simpler than add_path: it needs to
737 : : * consider only pathkeys and total cost.
738 : : *
739 : : * As with add_path, we pfree paths that are found to be dominated by
740 : : * another partial path; this requires that there be no other references to
741 : : * such paths yet. Hence, GatherPaths must not be created for a rel until
742 : : * we're done creating all partial paths for it. Unlike add_path, we don't
743 : : * take an exception for IndexPaths as partial index paths won't be
744 : : * referenced by partial BitmapHeapPaths.
745 : : */
746 : : void
3007 rhaas@postgresql.org 747 : 46883 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
748 : : {
2489 tgl@sss.pgh.pa.us 749 : 46883 : bool accept_new = true; /* unless we find a superior old path */
1735 750 : 46883 : int insert_at = 0; /* where to insert new item */
751 : : ListCell *p1;
752 : :
753 : : /* Check for query cancel. */
3007 rhaas@postgresql.org 754 [ - + ]: 46883 : CHECK_FOR_INTERRUPTS();
755 : :
756 : : /* Path to be added must be parallel safe. */
2181 757 [ - + ]: 46883 : Assert(new_path->parallel_safe);
758 : :
759 : : /* Relation should be OK for parallelism, too. */
760 [ - + ]: 46883 : Assert(parent_rel->consider_parallel);
761 : :
762 : : /*
763 : : * As in add_path, throw out any paths which are dominated by the new
764 : : * path, but throw out the new path if some existing path dominates it.
765 : : */
1735 tgl@sss.pgh.pa.us 766 [ + + + + : 62983 : foreach(p1, parent_rel->partial_pathlist)
+ + ]
767 : : {
3007 rhaas@postgresql.org 768 : 23494 : Path *old_path = (Path *) lfirst(p1);
769 : 23494 : bool remove_old = false; /* unless new proves superior */
770 : : PathKeysComparison keyscmp;
771 : :
772 : : /* Compare pathkeys. */
773 : 23494 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
774 : :
775 : : /* Unless pathkeys are incompatible, keep just one of the two paths. */
776 [ + + ]: 23494 : if (keyscmp != PATHKEYS_DIFFERENT)
777 : : {
778 [ + + ]: 23395 : if (new_path->total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
779 : : {
780 : : /* New path costs more; keep it only if pathkeys are better. */
781 [ + + ]: 7490 : if (keyscmp != PATHKEYS_BETTER1)
782 : 3259 : accept_new = false;
783 : : }
784 : 15905 : else if (old_path->total_cost > new_path->total_cost
785 [ + + ]: 15905 : * STD_FUZZ_FACTOR)
786 : : {
787 : : /* Old path costs more; keep it only if pathkeys are better. */
788 [ + + ]: 11595 : if (keyscmp != PATHKEYS_BETTER2)
789 : 6039 : remove_old = true;
790 : : }
791 [ - + ]: 4310 : else if (keyscmp == PATHKEYS_BETTER1)
792 : : {
793 : : /* Costs are about the same, new path has better pathkeys. */
3007 rhaas@postgresql.org 794 :UBC 0 : remove_old = true;
795 : : }
3007 rhaas@postgresql.org 796 [ + + ]:CBC 4310 : else if (keyscmp == PATHKEYS_BETTER2)
797 : : {
798 : : /* Costs are about the same, old path has better pathkeys. */
799 : 843 : accept_new = false;
800 : : }
801 [ + + ]: 3467 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
802 : : {
803 : : /* Pathkeys are the same, and the old path costs more. */
804 : 175 : remove_old = true;
805 : : }
806 : : else
807 : : {
808 : : /*
809 : : * Pathkeys are the same, and new path isn't materially
810 : : * cheaper.
811 : : */
812 : 3292 : accept_new = false;
813 : : }
814 : : }
815 : :
816 : : /*
817 : : * Remove current element from partial_pathlist if dominated by new.
818 : : */
819 [ + + ]: 23494 : if (remove_old)
820 : : {
821 : 6214 : parent_rel->partial_pathlist =
1735 tgl@sss.pgh.pa.us 822 : 6214 : foreach_delete_current(parent_rel->partial_pathlist, p1);
3007 rhaas@postgresql.org 823 : 6214 : pfree(old_path);
824 : : }
825 : : else
826 : : {
827 : : /* new belongs after this old path if it has cost >= old's */
828 [ + + ]: 17280 : if (new_path->total_cost >= old_path->total_cost)
1735 tgl@sss.pgh.pa.us 829 : 11504 : insert_at = foreach_current_index(p1) + 1;
830 : : }
831 : :
832 : : /*
833 : : * If we found an old path that dominates new_path, we can quit
834 : : * scanning the partial_pathlist; we will not add new_path, and we
835 : : * assume new_path cannot dominate any later path.
836 : : */
3007 rhaas@postgresql.org 837 [ + + ]: 23494 : if (!accept_new)
838 : 7394 : break;
839 : : }
840 : :
841 [ + + ]: 46883 : if (accept_new)
842 : : {
843 : : /* Accept the new path: insert it at proper place */
1735 tgl@sss.pgh.pa.us 844 : 39489 : parent_rel->partial_pathlist =
845 : 39489 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
846 : : }
847 : : else
848 : : {
849 : : /* Reject and recycle the new path */
3007 rhaas@postgresql.org 850 : 7394 : pfree(new_path);
851 : : }
852 : 46883 : }
853 : :
854 : : /*
855 : : * add_partial_path_precheck
856 : : * Check whether a proposed new partial path could possibly get accepted.
857 : : *
858 : : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
859 : : * since they don't matter for partial paths (see add_partial_path). But
860 : : * we do want to make sure we don't add a partial path if there's already
861 : : * a complete path that dominates it, since in that case the proposed path
862 : : * is surely a loser.
863 : : */
864 : : bool
865 : 31739 : add_partial_path_precheck(RelOptInfo *parent_rel, Cost total_cost,
866 : : List *pathkeys)
867 : : {
868 : : ListCell *p1;
869 : :
870 : : /*
871 : : * Our goal here is twofold. First, we want to find out whether this path
872 : : * is clearly inferior to some existing partial path. If so, we want to
873 : : * reject it immediately. Second, we want to find out whether this path
874 : : * is clearly superior to some existing partial path -- at least, modulo
875 : : * final cost computations. If so, we definitely want to consider it.
876 : : *
877 : : * Unlike add_path(), we always compare pathkeys here. This is because we
878 : : * expect partial_pathlist to be very short, and getting a definitive
879 : : * answer at this stage avoids the need to call add_path_precheck.
880 : : */
881 [ + + + + : 44381 : foreach(p1, parent_rel->partial_pathlist)
+ + ]
882 : : {
883 : 35423 : Path *old_path = (Path *) lfirst(p1);
884 : : PathKeysComparison keyscmp;
885 : :
886 : 35423 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
887 [ + + ]: 35423 : if (keyscmp != PATHKEYS_DIFFERENT)
888 : : {
889 [ + + + + ]: 35327 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
890 : : keyscmp != PATHKEYS_BETTER1)
891 : 22781 : return false;
892 [ + + + + ]: 17714 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
893 : : keyscmp != PATHKEYS_BETTER2)
894 : 5168 : return true;
895 : : }
896 : : }
897 : :
898 : : /*
899 : : * This path is neither clearly inferior to an existing partial path nor
900 : : * clearly good enough that it might replace one. Compare it to
901 : : * non-parallel plans. If it loses even before accounting for the cost of
902 : : * the Gather node, we should definitely reject it.
903 : : *
904 : : * Note that we pass the total_cost to add_path_precheck twice. This is
905 : : * because it's never advantageous to consider the startup cost of a
906 : : * partial path; the resulting plans, if run in parallel, will be run to
907 : : * completion.
908 : : */
909 [ + + ]: 8958 : if (!add_path_precheck(parent_rel, total_cost, total_cost, pathkeys,
910 : : NULL))
911 : 371 : return false;
912 : :
913 : 8587 : return true;
914 : : }
915 : :
916 : :
917 : : /*****************************************************************************
918 : : * PATH NODE CREATION ROUTINES
919 : : *****************************************************************************/
920 : :
921 : : /*
922 : : * create_seqscan_path
923 : : * Creates a path corresponding to a sequential scan, returning the
924 : : * pathnode.
925 : : */
926 : : Path *
3077 927 : 186469 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
928 : : Relids required_outer, int parallel_workers)
929 : : {
9715 bruce@momjian.us 930 : 186469 : Path *pathnode = makeNode(Path);
931 : :
9716 932 : 186469 : pathnode->pathtype = T_SeqScan;
933 : 186469 : pathnode->parent = rel;
2953 tgl@sss.pgh.pa.us 934 : 186469 : pathnode->pathtarget = rel->reltarget;
4378 935 : 186469 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
936 : : required_outer);
949 michael@paquier.xyz 937 : 186469 : pathnode->parallel_aware = (parallel_workers > 0);
3007 rhaas@postgresql.org 938 : 186469 : pathnode->parallel_safe = rel->consider_parallel;
2866 939 : 186469 : pathnode->parallel_workers = parallel_workers;
9008 tgl@sss.pgh.pa.us 940 : 186469 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
941 : :
3007 rhaas@postgresql.org 942 : 186469 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
943 : :
9357 bruce@momjian.us 944 : 186469 : return pathnode;
945 : : }
946 : :
947 : : /*
948 : : * create_samplescan_path
949 : : * Creates a path node for a sampled table scan.
950 : : */
951 : : Path *
3257 simon@2ndQuadrant.co 952 : 150 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
953 : : {
3249 bruce@momjian.us 954 : 150 : Path *pathnode = makeNode(Path);
955 : :
3257 simon@2ndQuadrant.co 956 : 150 : pathnode->pathtype = T_SampleScan;
957 : 150 : pathnode->parent = rel;
2953 tgl@sss.pgh.pa.us 958 : 150 : pathnode->pathtarget = rel->reltarget;
3257 simon@2ndQuadrant.co 959 : 150 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
960 : : required_outer);
3077 rhaas@postgresql.org 961 : 150 : pathnode->parallel_aware = false;
3007 962 : 150 : pathnode->parallel_safe = rel->consider_parallel;
2866 963 : 150 : pathnode->parallel_workers = 0;
3257 simon@2ndQuadrant.co 964 : 150 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
965 : :
3186 tgl@sss.pgh.pa.us 966 : 150 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
967 : :
3257 simon@2ndQuadrant.co 968 : 150 : return pathnode;
969 : : }
970 : :
971 : : /*
972 : : * create_index_path
973 : : * Creates a path node for an index scan.
974 : : *
975 : : * 'index' is a usable index.
976 : : * 'indexclauses' is a list of IndexClause nodes representing clauses
977 : : * to be enforced as qual conditions in the scan.
978 : : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
979 : : * to be used as index ordering operators in the scan.
980 : : * 'indexorderbycols' is an integer list of index column numbers (zero based)
981 : : * the ordering operators can be used with.
982 : : * 'pathkeys' describes the ordering of the path.
983 : : * 'indexscandir' is either ForwardScanDirection or BackwardScanDirection.
984 : : * 'indexonly' is true if an index-only scan is wanted.
985 : : * 'required_outer' is the set of outer relids for a parameterized path.
986 : : * 'loop_count' is the number of repetitions of the indexscan to factor into
987 : : * estimates of caching behavior.
988 : : * 'partial_path' is true if constructing a parallel index scan path.
989 : : *
990 : : * Returns the new path node.
991 : : */
992 : : IndexPath *
6888 tgl@sss.pgh.pa.us 993 : 322548 : create_index_path(PlannerInfo *root,
994 : : IndexOptInfo *index,
995 : : List *indexclauses,
996 : : List *indexorderbys,
997 : : List *indexorderbycols,
998 : : List *pathkeys,
999 : : ScanDirection indexscandir,
1000 : : bool indexonly,
1001 : : Relids required_outer,
1002 : : double loop_count,
1003 : : bool partial_path)
1004 : : {
9715 bruce@momjian.us 1005 : 322548 : IndexPath *pathnode = makeNode(IndexPath);
6932 tgl@sss.pgh.pa.us 1006 : 322548 : RelOptInfo *rel = index->rel;
1007 : :
4569 1008 [ + + ]: 322548 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
6932 1009 : 322548 : pathnode->path.parent = rel;
2953 1010 : 322548 : pathnode->path.pathtarget = rel->reltarget;
4378 1011 : 322548 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1012 : : required_outer);
3077 rhaas@postgresql.org 1013 : 322548 : pathnode->path.parallel_aware = false;
3007 1014 : 322548 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1015 : 322548 : pathnode->path.parallel_workers = 0;
8522 tgl@sss.pgh.pa.us 1016 : 322548 : pathnode->path.pathkeys = pathkeys;
1017 : :
6929 1018 : 322548 : pathnode->indexinfo = index;
4495 1019 : 322548 : pathnode->indexclauses = indexclauses;
4882 1020 : 322548 : pathnode->indexorderbys = indexorderbys;
4495 1021 : 322548 : pathnode->indexorderbycols = indexorderbycols;
8825 1022 : 322548 : pathnode->indexscandir = indexscandir;
1023 : :
2615 rhaas@postgresql.org 1024 : 322548 : cost_index(pathnode, root, loop_count, partial_path);
1025 : :
9357 bruce@momjian.us 1026 : 322548 : return pathnode;
1027 : : }
1028 : :
1029 : : /*
1030 : : * create_bitmap_heap_path
1031 : : * Creates a path node for a bitmap scan.
1032 : : *
1033 : : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1034 : : * 'required_outer' is the set of outer relids for a parameterized path.
1035 : : * 'loop_count' is the number of repetitions of the indexscan to factor into
1036 : : * estimates of caching behavior.
1037 : : *
1038 : : * loop_count should match the value used when creating the component
1039 : : * IndexPaths.
1040 : : */
1041 : : BitmapHeapPath *
6888 tgl@sss.pgh.pa.us 1042 : 145481 : create_bitmap_heap_path(PlannerInfo *root,
1043 : : RelOptInfo *rel,
1044 : : Path *bitmapqual,
1045 : : Relids required_outer,
1046 : : double loop_count,
1047 : : int parallel_degree)
1048 : : {
6935 1049 : 145481 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1050 : :
1051 : 145481 : pathnode->path.pathtype = T_BitmapHeapScan;
1052 : 145481 : pathnode->path.parent = rel;
2953 1053 : 145481 : pathnode->path.pathtarget = rel->reltarget;
4378 1054 : 145481 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1055 : : required_outer);
949 michael@paquier.xyz 1056 : 145481 : pathnode->path.parallel_aware = (parallel_degree > 0);
2960 tgl@sss.pgh.pa.us 1057 : 145481 : pathnode->path.parallel_safe = rel->consider_parallel;
2594 rhaas@postgresql.org 1058 : 145481 : pathnode->path.parallel_workers = parallel_degree;
2489 tgl@sss.pgh.pa.us 1059 : 145481 : pathnode->path.pathkeys = NIL; /* always unordered */
1060 : :
6935 1061 : 145481 : pathnode->bitmapqual = bitmapqual;
1062 : :
4378 1063 : 145481 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1064 : : pathnode->path.param_info,
1065 : : bitmapqual, loop_count);
1066 : :
6933 1067 : 145481 : return pathnode;
1068 : : }
1069 : :
1070 : : /*
1071 : : * create_bitmap_and_path
1072 : : * Creates a path node representing a BitmapAnd.
1073 : : */
1074 : : BitmapAndPath *
6888 1075 : 17038 : create_bitmap_and_path(PlannerInfo *root,
1076 : : RelOptInfo *rel,
1077 : : List *bitmapquals)
1078 : : {
6933 1079 : 17038 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1370 1080 : 17038 : Relids required_outer = NULL;
1081 : : ListCell *lc;
1082 : :
6933 1083 : 17038 : pathnode->path.pathtype = T_BitmapAnd;
1084 : 17038 : pathnode->path.parent = rel;
2953 1085 : 17038 : pathnode->path.pathtarget = rel->reltarget;
1086 : :
1087 : : /*
1088 : : * Identify the required outer rels as the union of what the child paths
1089 : : * depend on. (Alternatively, we could insist that the caller pass this
1090 : : * in, but it's more convenient and reliable to compute it here.)
1091 : : */
1370 1092 [ + - + + : 51114 : foreach(lc, bitmapquals)
+ + ]
1093 : : {
1094 : 34076 : Path *bitmapqual = (Path *) lfirst(lc);
1095 : :
1096 : 34076 : required_outer = bms_add_members(required_outer,
1097 [ + + ]: 34076 : PATH_REQ_OUTER(bitmapqual));
1098 : : }
1099 : 17038 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1100 : : required_outer);
1101 : :
1102 : : /*
1103 : : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1104 : : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1105 : : * set the flag for this path based only on the relation-level flag,
1106 : : * without actually iterating over the list of children.
1107 : : */
3077 rhaas@postgresql.org 1108 : 17038 : pathnode->path.parallel_aware = false;
3007 1109 : 17038 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1110 : 17038 : pathnode->path.parallel_workers = 0;
1111 : :
2489 tgl@sss.pgh.pa.us 1112 : 17038 : pathnode->path.pathkeys = NIL; /* always unordered */
1113 : :
6933 1114 : 17038 : pathnode->bitmapquals = bitmapquals;
1115 : :
1116 : : /* this sets bitmapselectivity as well as the regular cost fields: */
1117 : 17038 : cost_bitmap_and_node(pathnode, root);
1118 : :
1119 : 17038 : return pathnode;
1120 : : }
1121 : :
1122 : : /*
1123 : : * create_bitmap_or_path
1124 : : * Creates a path node representing a BitmapOr.
1125 : : */
1126 : : BitmapOrPath *
6888 1127 : 480 : create_bitmap_or_path(PlannerInfo *root,
1128 : : RelOptInfo *rel,
1129 : : List *bitmapquals)
1130 : : {
6933 1131 : 480 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1370 1132 : 480 : Relids required_outer = NULL;
1133 : : ListCell *lc;
1134 : :
6933 1135 : 480 : pathnode->path.pathtype = T_BitmapOr;
1136 : 480 : pathnode->path.parent = rel;
2953 1137 : 480 : pathnode->path.pathtarget = rel->reltarget;
1138 : :
1139 : : /*
1140 : : * Identify the required outer rels as the union of what the child paths
1141 : : * depend on. (Alternatively, we could insist that the caller pass this
1142 : : * in, but it's more convenient and reliable to compute it here.)
1143 : : */
1370 1144 [ + - + + : 1470 : foreach(lc, bitmapquals)
+ + ]
1145 : : {
1146 : 990 : Path *bitmapqual = (Path *) lfirst(lc);
1147 : :
1148 : 990 : required_outer = bms_add_members(required_outer,
1149 [ + + ]: 990 : PATH_REQ_OUTER(bitmapqual));
1150 : : }
1151 : 480 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1152 : : required_outer);
1153 : :
1154 : : /*
1155 : : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1156 : : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1157 : : * set the flag for this path based only on the relation-level flag,
1158 : : * without actually iterating over the list of children.
1159 : : */
3077 rhaas@postgresql.org 1160 : 480 : pathnode->path.parallel_aware = false;
3007 1161 : 480 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1162 : 480 : pathnode->path.parallel_workers = 0;
1163 : :
2489 tgl@sss.pgh.pa.us 1164 : 480 : pathnode->path.pathkeys = NIL; /* always unordered */
1165 : :
6933 1166 : 480 : pathnode->bitmapquals = bitmapquals;
1167 : :
1168 : : /* this sets bitmapselectivity as well as the regular cost fields: */
1169 : 480 : cost_bitmap_or_node(pathnode, root);
1170 : :
6935 1171 : 480 : return pathnode;
1172 : : }
1173 : :
1174 : : /*
1175 : : * create_tidscan_path
1176 : : * Creates a path corresponding to a scan by TID, returning the pathnode.
1177 : : */
1178 : : TidPath *
4249 1179 : 378 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1180 : : Relids required_outer)
1181 : : {
8768 bruce@momjian.us 1182 : 378 : TidPath *pathnode = makeNode(TidPath);
1183 : :
8909 1184 : 378 : pathnode->path.pathtype = T_TidScan;
1185 : 378 : pathnode->path.parent = rel;
2953 tgl@sss.pgh.pa.us 1186 : 378 : pathnode->path.pathtarget = rel->reltarget;
4249 1187 : 378 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1188 : : required_outer);
3077 rhaas@postgresql.org 1189 : 378 : pathnode->path.parallel_aware = false;
3007 1190 : 378 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1191 : 378 : pathnode->path.parallel_workers = 0;
2489 tgl@sss.pgh.pa.us 1192 : 378 : pathnode->path.pathkeys = NIL; /* always unordered */
1193 : :
6714 1194 : 378 : pathnode->tidquals = tidquals;
1195 : :
4249 1196 : 378 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1197 : : pathnode->path.param_info);
1198 : :
8909 bruce@momjian.us 1199 : 378 : return pathnode;
1200 : : }
1201 : :
1202 : : /*
1203 : : * create_tidrangescan_path
1204 : : * Creates a path corresponding to a scan by a range of TIDs, returning
1205 : : * the pathnode.
1206 : : */
1207 : : TidRangePath *
1142 drowley@postgresql.o 1208 : 101 : create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
1209 : : List *tidrangequals, Relids required_outer)
1210 : : {
1211 : 101 : TidRangePath *pathnode = makeNode(TidRangePath);
1212 : :
1213 : 101 : pathnode->path.pathtype = T_TidRangeScan;
1214 : 101 : pathnode->path.parent = rel;
1215 : 101 : pathnode->path.pathtarget = rel->reltarget;
1216 : 101 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1217 : : required_outer);
1218 : 101 : pathnode->path.parallel_aware = false;
1219 : 101 : pathnode->path.parallel_safe = rel->consider_parallel;
1220 : 101 : pathnode->path.parallel_workers = 0;
1221 : 101 : pathnode->path.pathkeys = NIL; /* always unordered */
1222 : :
1223 : 101 : pathnode->tidrangequals = tidrangequals;
1224 : :
1225 : 101 : cost_tidrangescan(&pathnode->path, root, rel, tidrangequals,
1226 : : pathnode->path.param_info);
1227 : :
1228 : 101 : return pathnode;
1229 : : }
1230 : :
1231 : : /*
1232 : : * create_append_path
1233 : : * Creates a path corresponding to an Append plan, returning the
1234 : : * pathnode.
1235 : : *
1236 : : * Note that we must handle subpaths = NIL, representing a dummy access path.
1237 : : * Also, there are callers that pass root = NULL.
1238 : : *
1239 : : * 'rows', when passed as a non-negative number, will be used to overwrite the
1240 : : * returned path's row estimate. Otherwise, the row estimate is calculated
1241 : : * by totalling the row estimates from the 'subpaths' list.
1242 : : */
1243 : : AppendPath *
2199 alvherre@alvh.no-ip. 1244 : 34171 : create_append_path(PlannerInfo *root,
1245 : : RelOptInfo *rel,
1246 : : List *subpaths, List *partial_subpaths,
1247 : : List *pathkeys, Relids required_outer,
1248 : : int parallel_workers, bool parallel_aware,
1249 : : double rows)
1250 : : {
8554 tgl@sss.pgh.pa.us 1251 : 34171 : AppendPath *pathnode = makeNode(AppendPath);
1252 : : ListCell *l;
1253 : :
2322 rhaas@postgresql.org 1254 [ + + - + ]: 34171 : Assert(!parallel_aware || parallel_workers > 0);
1255 : :
8554 tgl@sss.pgh.pa.us 1256 : 34171 : pathnode->path.pathtype = T_Append;
1257 : 34171 : pathnode->path.parent = rel;
2953 1258 : 34171 : pathnode->path.pathtarget = rel->reltarget;
1259 : :
1260 : : /*
1261 : : * If this is for a baserel (not a join or non-leaf partition), we prefer
1262 : : * to apply get_baserel_parampathinfo to construct a full ParamPathInfo
1263 : : * for the path. This supports building a Memoize path atop this path,
1264 : : * and if this is a partitioned table the info may be useful for run-time
1265 : : * pruning (cf make_partition_pruneinfo()).
1266 : : *
1267 : : * However, if we don't have "root" then that won't work and we fall back
1268 : : * on the simpler get_appendrel_parampathinfo. There's no point in doing
1269 : : * the more expensive thing for a dummy path, either.
1270 : : */
395 1271 [ + + + + : 34171 : if (rel->reloptkind == RELOPT_BASEREL && root && subpaths != NIL)
+ + ]
2199 alvherre@alvh.no-ip. 1272 : 15524 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1273 : : rel,
1274 : : required_outer);
1275 : : else
1276 : 18647 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1277 : : required_outer);
1278 : :
2322 rhaas@postgresql.org 1279 : 34171 : pathnode->path.parallel_aware = parallel_aware;
3007 1280 : 34171 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1281 : 34171 : pathnode->path.parallel_workers = parallel_workers;
1836 tgl@sss.pgh.pa.us 1282 : 34171 : pathnode->path.pathkeys = pathkeys;
1283 : :
1284 : : /*
1285 : : * For parallel append, non-partial paths are sorted by descending total
1286 : : * costs. That way, the total time to finish all non-partial paths is
1287 : : * minimized. Also, the partial paths are sorted by descending startup
1288 : : * costs. There may be some paths that require to do startup work by a
1289 : : * single worker. In such case, it's better for workers to choose the
1290 : : * expensive ones first, whereas the leader should choose the cheapest
1291 : : * startup plan.
1292 : : */
2322 rhaas@postgresql.org 1293 [ + + ]: 34171 : if (pathnode->path.parallel_aware)
1294 : : {
1295 : : /*
1296 : : * We mustn't fiddle with the order of subpaths when the Append has
1297 : : * pathkeys. The order they're listed in is critical to keeping the
1298 : : * pathkeys valid.
1299 : : */
1836 tgl@sss.pgh.pa.us 1300 [ - + ]: 11496 : Assert(pathkeys == NIL);
1301 : :
1734 1302 : 11496 : list_sort(subpaths, append_total_cost_compare);
1303 : 11496 : list_sort(partial_subpaths, append_startup_cost_compare);
1304 : : }
2322 rhaas@postgresql.org 1305 : 34171 : pathnode->first_partial_path = list_length(subpaths);
1306 : 34171 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1307 : :
1308 : : /*
1309 : : * Apply query-wide LIMIT if known and path is for sole base relation.
1310 : : * (Handling this at this low level is a bit klugy.)
1311 : : */
440 tgl@sss.pgh.pa.us 1312 [ + + + + ]: 34171 : if (root != NULL && bms_equal(rel->relids, root->all_query_rels))
1836 1313 : 19140 : pathnode->limit_tuples = root->limit_tuples;
1314 : : else
1315 : 15031 : pathnode->limit_tuples = -1.0;
1316 : :
2123 akapila@postgresql.o 1317 [ + + + + : 112941 : foreach(l, pathnode->subpaths)
+ + ]
1318 : : {
8424 bruce@momjian.us 1319 : 78770 : Path *subpath = (Path *) lfirst(l);
1320 : :
3007 rhaas@postgresql.org 1321 [ + + ]: 140270 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1322 [ + + ]: 61500 : subpath->parallel_safe;
1323 : :
1324 : : /* All child paths must have same parameterization */
4378 tgl@sss.pgh.pa.us 1325 [ + + - + ]: 78770 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1326 : : }
1327 : :
2322 rhaas@postgresql.org 1328 [ + + - + ]: 34171 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1329 : :
1330 : : /*
1331 : : * If there's exactly one child path then the output of the Append is
1332 : : * necessarily ordered the same as the child's, so we can inherit the
1333 : : * child's pathkeys if any, overriding whatever the caller might've said.
1334 : : * Furthermore, if the child's parallel awareness matches the Append's,
1335 : : * then the Append is a no-op and will be discarded later (in setrefs.c).
1336 : : * Then we can inherit the child's size and cost too, effectively charging
1337 : : * zero for the Append. Otherwise, we must do the normal costsize
1338 : : * calculation.
1339 : : */
1847 tgl@sss.pgh.pa.us 1340 [ + + ]: 34171 : if (list_length(pathnode->subpaths) == 1)
1341 : : {
1342 : 10925 : Path *child = (Path *) linitial(pathnode->subpaths);
1343 : :
635 1344 [ + + ]: 10925 : if (child->parallel_aware == parallel_aware)
1345 : : {
1346 : 10706 : pathnode->path.rows = child->rows;
1347 : 10706 : pathnode->path.startup_cost = child->startup_cost;
1348 : 10706 : pathnode->path.total_cost = child->total_cost;
1349 : : }
1350 : : else
559 1351 : 219 : cost_append(pathnode);
1352 : : /* Must do this last, else cost_append complains */
1847 1353 : 10925 : pathnode->path.pathkeys = child->pathkeys;
1354 : : }
1355 : : else
559 1356 : 23246 : cost_append(pathnode);
1357 : :
1358 : : /* If the caller provided a row estimate, override the computed value. */
2322 rhaas@postgresql.org 1359 [ + + ]: 34171 : if (rows >= 0)
1360 : 300 : pathnode->path.rows = rows;
1361 : :
8554 tgl@sss.pgh.pa.us 1362 : 34171 : return pathnode;
1363 : : }
1364 : :
1365 : : /*
1366 : : * append_total_cost_compare
1367 : : * list_sort comparator for sorting append child paths
1368 : : * by total_cost descending
1369 : : *
1370 : : * For equal total costs, we fall back to comparing startup costs; if those
1371 : : * are equal too, break ties using bms_compare on the paths' relids.
1372 : : * (This is to avoid getting unpredictable results from list_sort.)
1373 : : */
1374 : : static int
1734 1375 : 991 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1376 : : {
1377 : 991 : Path *path1 = (Path *) lfirst(a);
1378 : 991 : Path *path2 = (Path *) lfirst(b);
1379 : : int cmp;
1380 : :
2287 1381 : 991 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1382 [ + + ]: 991 : if (cmp != 0)
1383 : 856 : return -cmp;
1384 : 135 : return bms_compare(path1->parent->relids, path2->parent->relids);
1385 : : }
1386 : :
1387 : : /*
1388 : : * append_startup_cost_compare
1389 : : * list_sort comparator for sorting append child paths
1390 : : * by startup_cost descending
1391 : : *
1392 : : * For equal startup costs, we fall back to comparing total costs; if those
1393 : : * are equal too, break ties using bms_compare on the paths' relids.
1394 : : * (This is to avoid getting unpredictable results from list_sort.)
1395 : : */
1396 : : static int
1734 1397 : 17211 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1398 : : {
1399 : 17211 : Path *path1 = (Path *) lfirst(a);
1400 : 17211 : Path *path2 = (Path *) lfirst(b);
1401 : : int cmp;
1402 : :
2287 1403 : 17211 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1404 [ + + ]: 17211 : if (cmp != 0)
1405 : 6652 : return -cmp;
1406 : 10559 : return bms_compare(path1->parent->relids, path2->parent->relids);
1407 : : }
1408 : :
1409 : : /*
1410 : : * create_merge_append_path
1411 : : * Creates a path corresponding to a MergeAppend plan, returning the
1412 : : * pathnode.
1413 : : */
1414 : : MergeAppendPath *
4931 1415 : 1943 : create_merge_append_path(PlannerInfo *root,
1416 : : RelOptInfo *rel,
1417 : : List *subpaths,
1418 : : List *pathkeys,
1419 : : Relids required_outer)
1420 : : {
1421 : 1943 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1422 : : Cost input_startup_cost;
1423 : : Cost input_total_cost;
1424 : : ListCell *l;
1425 : :
1426 : 1943 : pathnode->path.pathtype = T_MergeAppend;
1427 : 1943 : pathnode->path.parent = rel;
2953 1428 : 1943 : pathnode->path.pathtarget = rel->reltarget;
4378 1429 : 1943 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1430 : : required_outer);
3077 rhaas@postgresql.org 1431 : 1943 : pathnode->path.parallel_aware = false;
3007 1432 : 1943 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1433 : 1943 : pathnode->path.parallel_workers = 0;
4931 tgl@sss.pgh.pa.us 1434 : 1943 : pathnode->path.pathkeys = pathkeys;
1435 : 1943 : pathnode->subpaths = subpaths;
1436 : :
1437 : : /*
1438 : : * Apply query-wide LIMIT if known and path is for sole base relation.
1439 : : * (Handling this at this low level is a bit klugy.)
1440 : : */
440 1441 [ + + ]: 1943 : if (bms_equal(rel->relids, root->all_query_rels))
4378 1442 : 1014 : pathnode->limit_tuples = root->limit_tuples;
1443 : : else
1444 : 929 : pathnode->limit_tuples = -1.0;
1445 : :
1446 : : /*
1447 : : * Add up the sizes and costs of the input paths.
1448 : : */
4461 1449 : 1943 : pathnode->path.rows = 0;
4931 1450 : 1943 : input_startup_cost = 0;
1451 : 1943 : input_total_cost = 0;
1452 [ + - + + : 7266 : foreach(l, subpaths)
+ + ]
1453 : : {
1454 : 5323 : Path *subpath = (Path *) lfirst(l);
1455 : :
4461 1456 : 5323 : pathnode->path.rows += subpath->rows;
3007 rhaas@postgresql.org 1457 [ + + ]: 9593 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1458 [ + + ]: 4270 : subpath->parallel_safe;
1459 : :
4931 tgl@sss.pgh.pa.us 1460 [ + + ]: 5323 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1461 : : {
1462 : : /* Subpath is adequately ordered, we won't need to sort it */
1463 : 5180 : input_startup_cost += subpath->startup_cost;
1464 : 5180 : input_total_cost += subpath->total_cost;
1465 : : }
1466 : : else
1467 : : {
1468 : : /* We'll need to insert a Sort node, so include cost for that */
1469 : : Path sort_path; /* dummy for result of cost_sort */
1470 : :
1471 : 143 : cost_sort(&sort_path,
1472 : : root,
1473 : : pathkeys,
1474 : : subpath->total_cost,
1475 : : subpath->rows,
2978 1476 : 143 : subpath->pathtarget->width,
1477 : : 0.0,
1478 : : work_mem,
1479 : : pathnode->limit_tuples);
4931 1480 : 143 : input_startup_cost += sort_path.startup_cost;
1481 : 143 : input_total_cost += sort_path.total_cost;
1482 : : }
1483 : :
1484 : : /* All child paths must have same parameterization */
4378 1485 [ - + - + ]: 5323 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1486 : : }
1487 : :
1488 : : /*
1489 : : * Now we can compute total costs of the MergeAppend. If there's exactly
1490 : : * one child path and its parallel awareness matches that of the
1491 : : * MergeAppend, then the MergeAppend is a no-op and will be discarded
1492 : : * later (in setrefs.c); otherwise we do the normal cost calculation.
1493 : : */
635 1494 [ + + ]: 1943 : if (list_length(subpaths) == 1 &&
1495 : 55 : ((Path *) linitial(subpaths))->parallel_aware ==
1496 [ + - ]: 55 : pathnode->path.parallel_aware)
1497 : : {
1847 1498 : 55 : pathnode->path.startup_cost = input_startup_cost;
1499 : 55 : pathnode->path.total_cost = input_total_cost;
1500 : : }
1501 : : else
1502 : 1888 : cost_merge_append(&pathnode->path, root,
1503 : : pathkeys, list_length(subpaths),
1504 : : input_startup_cost, input_total_cost,
1505 : : pathnode->path.rows);
1506 : :
4931 1507 : 1943 : return pathnode;
1508 : : }
1509 : :
1510 : : /*
1511 : : * create_group_result_path
1512 : : * Creates a path representing a Result-and-nothing-else plan.
1513 : : *
1514 : : * This is only used for degenerate grouping cases, in which we know we
1515 : : * need to produce one result row, possibly filtered by a HAVING qual.
1516 : : */
1517 : : GroupResultPath *
1903 1518 : 104222 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1519 : : PathTarget *target, List *havingqual)
1520 : : {
1521 : 104222 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1522 : :
7830 1523 : 104222 : pathnode->path.pathtype = T_Result;
2978 1524 : 104222 : pathnode->path.parent = rel;
2960 1525 : 104222 : pathnode->path.pathtarget = target;
3973 bruce@momjian.us 1526 : 104222 : pathnode->path.param_info = NULL; /* there are no other rels... */
3077 rhaas@postgresql.org 1527 : 104222 : pathnode->path.parallel_aware = false;
3007 1528 : 104222 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 1529 : 104222 : pathnode->path.parallel_workers = 0;
6497 tgl@sss.pgh.pa.us 1530 : 104222 : pathnode->path.pathkeys = NIL;
1903 1531 : 104222 : pathnode->quals = havingqual;
1532 : :
1533 : : /*
1534 : : * We can't quite use cost_resultscan() because the quals we want to
1535 : : * account for are not baserestrict quals of the rel. Might as well just
1536 : : * hack it here.
1537 : : */
4461 1538 : 104222 : pathnode->path.rows = 1;
2960 1539 : 104222 : pathnode->path.startup_cost = target->cost.startup;
1540 : 104222 : pathnode->path.total_cost = target->cost.startup +
1541 : 104222 : cpu_tuple_cost + target->cost.per_tuple;
1542 : :
1543 : : /*
1544 : : * Add cost of qual, if any --- but we ignore its selectivity, since our
1545 : : * rowcount estimate should be 1 no matter what the qual is.
1546 : : */
1903 1547 [ + + ]: 104222 : if (havingqual)
1548 : : {
1549 : : QualCost qual_cost;
1550 : :
1551 : 258 : cost_qual_eval(&qual_cost, havingqual, root);
1552 : : /* havingqual is evaluated once at startup */
2959 1553 : 258 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1554 : 258 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1555 : : }
1556 : :
7830 1557 : 104222 : return pathnode;
1558 : : }
1559 : :
1560 : : /*
1561 : : * create_material_path
1562 : : * Creates a path corresponding to a Material plan, returning the
1563 : : * pathnode.
1564 : : */
1565 : : MaterialPath *
7806 1566 : 211095 : create_material_path(RelOptInfo *rel, Path *subpath)
1567 : : {
1568 : 211095 : MaterialPath *pathnode = makeNode(MaterialPath);
1569 : :
4378 1570 [ - + ]: 211095 : Assert(subpath->parent == rel);
1571 : :
7806 1572 : 211095 : pathnode->path.pathtype = T_Material;
1573 : 211095 : pathnode->path.parent = rel;
2953 1574 : 211095 : pathnode->path.pathtarget = rel->reltarget;
4378 1575 : 211095 : pathnode->path.param_info = subpath->param_info;
3077 rhaas@postgresql.org 1576 : 211095 : pathnode->path.parallel_aware = false;
2960 tgl@sss.pgh.pa.us 1577 [ + + ]: 399375 : pathnode->path.parallel_safe = rel->consider_parallel &&
1578 [ + + ]: 188280 : subpath->parallel_safe;
2866 rhaas@postgresql.org 1579 : 211095 : pathnode->path.parallel_workers = subpath->parallel_workers;
7806 tgl@sss.pgh.pa.us 1580 : 211095 : pathnode->path.pathkeys = subpath->pathkeys;
1581 : :
1582 : 211095 : pathnode->subpath = subpath;
1583 : :
1584 : 211095 : cost_material(&pathnode->path,
1585 : : subpath->startup_cost,
1586 : : subpath->total_cost,
1587 : : subpath->rows,
2978 1588 : 211095 : subpath->pathtarget->width);
1589 : :
7806 1590 : 211095 : return pathnode;
1591 : : }
1592 : :
1593 : : /*
1594 : : * create_memoize_path
1595 : : * Creates a path corresponding to a Memoize plan, returning the pathnode.
1596 : : */
1597 : : MemoizePath *
1005 drowley@postgresql.o 1598 : 119907 : create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1599 : : List *param_exprs, List *hash_operators,
1600 : : bool singlerow, bool binary_mode, double calls)
1601 : : {
1602 : 119907 : MemoizePath *pathnode = makeNode(MemoizePath);
1603 : :
1108 1604 [ - + ]: 119907 : Assert(subpath->parent == rel);
1605 : :
1005 1606 : 119907 : pathnode->path.pathtype = T_Memoize;
1108 1607 : 119907 : pathnode->path.parent = rel;
1608 : 119907 : pathnode->path.pathtarget = rel->reltarget;
1609 : 119907 : pathnode->path.param_info = subpath->param_info;
1610 : 119907 : pathnode->path.parallel_aware = false;
1611 [ + + ]: 234957 : pathnode->path.parallel_safe = rel->consider_parallel &&
1612 [ + + ]: 115050 : subpath->parallel_safe;
1613 : 119907 : pathnode->path.parallel_workers = subpath->parallel_workers;
1614 : 119907 : pathnode->path.pathkeys = subpath->pathkeys;
1615 : :
1616 : 119907 : pathnode->subpath = subpath;
1617 : 119907 : pathnode->hash_operators = hash_operators;
1618 : 119907 : pathnode->param_exprs = param_exprs;
1619 : 119907 : pathnode->singlerow = singlerow;
872 1620 : 119907 : pathnode->binary_mode = binary_mode;
1108 1621 : 119907 : pathnode->calls = calls;
1622 : :
1623 : : /*
1624 : : * For now we set est_entries to 0. cost_memoize_rescan() does all the
1625 : : * hard work to determine how many cache entries there are likely to be,
1626 : : * so it seems best to leave it up to that function to fill this field in.
1627 : : * If left at 0, the executor will make a guess at a good value.
1628 : : */
1629 : 119907 : pathnode->est_entries = 0;
1630 : :
1631 : : /*
1632 : : * Add a small additional charge for caching the first entry. All the
1633 : : * harder calculations for rescans are performed in cost_memoize_rescan().
1634 : : */
1635 : 119907 : pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
1636 : 119907 : pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
1637 : 119907 : pathnode->path.rows = subpath->rows;
1638 : :
1639 : 119907 : return pathnode;
1640 : : }
1641 : :
1642 : : /*
1643 : : * create_unique_path
1644 : : * Creates a path representing elimination of distinct rows from the
1645 : : * input data. Distinct-ness is defined according to the needs of the
1646 : : * semijoin represented by sjinfo. If it is not possible to identify
1647 : : * how to make the data unique, NULL is returned.
1648 : : *
1649 : : * If used at all, this is likely to be called repeatedly on the same rel;
1650 : : * and the input subpath should always be the same (the cheapest_total path
1651 : : * for the rel). So we cache the result.
1652 : : */
1653 : : UniquePath *
5722 tgl@sss.pgh.pa.us 1654 : 14734 : create_unique_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1655 : : SpecialJoinInfo *sjinfo)
1656 : : {
1657 : : UniquePath *pathnode;
1658 : : Path sort_path; /* dummy for result of cost_sort */
1659 : : Path agg_path; /* dummy for result of cost_agg */
1660 : : MemoryContext oldcontext;
1661 : : int numCols;
1662 : :
1663 : : /* Caller made a mistake if subpath isn't cheapest_total ... */
7755 1664 [ - + ]: 14734 : Assert(subpath == rel->cheapest_total_path);
4378 1665 [ - + ]: 14734 : Assert(subpath->parent == rel);
1666 : : /* ... or if SpecialJoinInfo is the wrong one */
5722 1667 [ - + ]: 14734 : Assert(sjinfo->jointype == JOIN_SEMI);
1668 [ - + ]: 14734 : Assert(bms_equal(rel->relids, sjinfo->syn_righthand));
1669 : :
1670 : : /* If result already cached, return it */
7755 1671 [ + + ]: 14734 : if (rel->cheapest_unique_path)
1672 : 12919 : return (UniquePath *) rel->cheapest_unique_path;
1673 : :
1674 : : /* If it's not possible to unique-ify, return NULL */
3322 1675 [ + + + - ]: 1815 : if (!(sjinfo->semi_can_btree || sjinfo->semi_can_hash))
5722 1676 : 60 : return NULL;
1677 : :
1678 : : /*
1679 : : * When called during GEQO join planning, we are in a short-lived memory
1680 : : * context. We must make sure that the path and any subsidiary data
1681 : : * structures created for a baserel survive the GEQO cycle, else the
1682 : : * baserel is trashed for future GEQO cycles. On the other hand, when we
1683 : : * are creating those for a joinrel during GEQO, we don't want them to
1684 : : * clutter the main planning context. Upshot is that the best solution is
1685 : : * to explicitly allocate memory in the same context the given RelOptInfo
1686 : : * is in.
1687 : : */
2327 rhaas@postgresql.org 1688 : 1755 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
1689 : :
5722 tgl@sss.pgh.pa.us 1690 : 1755 : pathnode = makeNode(UniquePath);
1691 : :
7755 1692 : 1755 : pathnode->path.pathtype = T_Unique;
1693 : 1755 : pathnode->path.parent = rel;
2953 1694 : 1755 : pathnode->path.pathtarget = rel->reltarget;
4378 1695 : 1755 : pathnode->path.param_info = subpath->param_info;
3077 rhaas@postgresql.org 1696 : 1755 : pathnode->path.parallel_aware = false;
2960 tgl@sss.pgh.pa.us 1697 [ + + ]: 3280 : pathnode->path.parallel_safe = rel->consider_parallel &&
1698 [ + + ]: 1525 : subpath->parallel_safe;
2866 rhaas@postgresql.org 1699 : 1755 : pathnode->path.parallel_workers = subpath->parallel_workers;
1700 : :
1701 : : /*
1702 : : * Assume the output is unsorted, since we don't necessarily have pathkeys
1703 : : * to represent it. (This might get overridden below.)
1704 : : */
7755 tgl@sss.pgh.pa.us 1705 : 1755 : pathnode->path.pathkeys = NIL;
1706 : :
1707 : 1755 : pathnode->subpath = subpath;
1708 : :
1709 : : /*
1710 : : * Under GEQO and when planning child joins, the sjinfo might be
1711 : : * short-lived, so we'd better make copies of data structures we extract
1712 : : * from it.
1713 : : */
51 1714 : 1755 : pathnode->in_operators = copyObject(sjinfo->semi_operators);
1715 : 1755 : pathnode->uniq_exprs = copyObject(sjinfo->semi_rhs_exprs);
1716 : :
1717 : : /*
1718 : : * If the input is a relation and it has a unique index that proves the
1719 : : * semi_rhs_exprs are unique, then we don't need to do anything. Note
1720 : : * that relation_has_unique_index_for automatically considers restriction
1721 : : * clauses for the rel, as well.
1722 : : */
3322 1723 [ + + + - : 2198 : if (rel->rtekind == RTE_RELATION && sjinfo->semi_can_btree &&
- + ]
4554 1724 : 443 : relation_has_unique_index_for(root, rel, NIL,
1725 : : sjinfo->semi_rhs_exprs,
1726 : : sjinfo->semi_operators))
1727 : : {
4554 tgl@sss.pgh.pa.us 1728 :UBC 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
4461 1729 : 0 : pathnode->path.rows = rel->rows;
4554 1730 : 0 : pathnode->path.startup_cost = subpath->startup_cost;
1731 : 0 : pathnode->path.total_cost = subpath->total_cost;
1732 : 0 : pathnode->path.pathkeys = subpath->pathkeys;
1733 : :
1734 : 0 : rel->cheapest_unique_path = (Path *) pathnode;
1735 : :
1736 : 0 : MemoryContextSwitchTo(oldcontext);
1737 : :
1738 : 0 : return pathnode;
1739 : : }
1740 : :
1741 : : /*
1742 : : * If the input is a subquery whose output must be unique already, then we
1743 : : * don't need to do anything. The test for uniqueness has to consider
1744 : : * exactly which columns we are extracting; for example "SELECT DISTINCT
1745 : : * x,y" doesn't guarantee that x alone is distinct. So we cannot check for
1746 : : * this optimization unless semi_rhs_exprs consists only of simple Vars
1747 : : * referencing subquery outputs. (Possibly we could do something with
1748 : : * expressions in the subquery outputs, too, but for now keep it simple.)
1749 : : */
5722 tgl@sss.pgh.pa.us 1750 [ + + ]:CBC 1755 : if (rel->rtekind == RTE_SUBQUERY)
1751 : : {
6203 1752 [ + - ]: 323 : RangeTblEntry *rte = planner_rt_fetch(rel->relid, root);
1753 : :
3561 1754 [ + + ]: 323 : if (query_supports_distinctness(rte->subquery))
1755 : : {
1756 : : List *sub_tlist_colnos;
1757 : :
3322 1758 : 305 : sub_tlist_colnos = translate_sub_tlist(sjinfo->semi_rhs_exprs,
1759 : 305 : rel->relid);
1760 : :
3561 1761 [ + + - + ]: 366 : if (sub_tlist_colnos &&
1762 : 61 : query_is_distinct_for(rte->subquery,
1763 : : sub_tlist_colnos,
1764 : : sjinfo->semi_operators))
1765 : : {
3561 tgl@sss.pgh.pa.us 1766 :UBC 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1767 : 0 : pathnode->path.rows = rel->rows;
1768 : 0 : pathnode->path.startup_cost = subpath->startup_cost;
1769 : 0 : pathnode->path.total_cost = subpath->total_cost;
1770 : 0 : pathnode->path.pathkeys = subpath->pathkeys;
1771 : :
1772 : 0 : rel->cheapest_unique_path = (Path *) pathnode;
1773 : :
1774 : 0 : MemoryContextSwitchTo(oldcontext);
1775 : :
1776 : 0 : return pathnode;
1777 : : }
1778 : : }
1779 : : }
1780 : :
1781 : : /* Estimate number of output rows */
3322 tgl@sss.pgh.pa.us 1782 :CBC 1755 : pathnode->path.rows = estimate_num_groups(root,
1783 : : sjinfo->semi_rhs_exprs,
1784 : : rel->rows,
1785 : : NULL,
1786 : : NULL);
1787 : 1755 : numCols = list_length(sjinfo->semi_rhs_exprs);
1788 : :
1789 [ + - ]: 1755 : if (sjinfo->semi_can_btree)
1790 : : {
1791 : : /*
1792 : : * Estimate cost for sort+unique implementation
1793 : : */
5722 1794 : 1755 : cost_sort(&sort_path, root, NIL,
1795 : : subpath->total_cost,
1796 : : rel->rows,
2978 1797 : 1755 : subpath->pathtarget->width,
1798 : : 0.0,
1799 : : work_mem,
1800 : : -1.0);
1801 : :
1802 : : /*
1803 : : * Charge one cpu_operator_cost per comparison per input tuple. We
1804 : : * assume all columns get compared at most of the tuples. (XXX
1805 : : * probably this is an overestimate.) This should agree with
1806 : : * create_upper_unique_path.
1807 : : */
5722 1808 : 1755 : sort_path.total_cost += cpu_operator_cost * rel->rows * numCols;
1809 : : }
1810 : :
3322 1811 [ + - ]: 1755 : if (sjinfo->semi_can_hash)
1812 : : {
1813 : : /*
1814 : : * Estimate the overhead per hashtable entry at 64 bytes (same as in
1815 : : * planner.c).
1816 : : */
2978 1817 : 1755 : int hashentrysize = subpath->pathtarget->width + 64;
1818 : :
994 1819 [ - + ]: 1755 : if (hashentrysize * pathnode->path.rows > get_hash_memory_limit())
1820 : : {
1821 : : /*
1822 : : * We should not try to hash. Hack the SpecialJoinInfo to
1823 : : * remember this, in case we come through here again.
1824 : : */
3322 tgl@sss.pgh.pa.us 1825 :UBC 0 : sjinfo->semi_can_hash = false;
1826 : : }
1827 : : else
7753 tgl@sss.pgh.pa.us 1828 :CBC 1755 : cost_agg(&agg_path, root,
1829 : : AGG_HASHED, NULL,
1830 : : numCols, pathnode->path.rows,
1831 : : NIL,
1832 : : subpath->startup_cost,
1833 : : subpath->total_cost,
1834 : : rel->rows,
1488 jdavis@postgresql.or 1835 : 1755 : subpath->pathtarget->width);
1836 : : }
1837 : :
3322 tgl@sss.pgh.pa.us 1838 [ + - + - ]: 1755 : if (sjinfo->semi_can_btree && sjinfo->semi_can_hash)
1839 : : {
5722 1840 [ + + ]: 1755 : if (agg_path.total_cost < sort_path.total_cost)
1841 : 1688 : pathnode->umethod = UNIQUE_PATH_HASH;
1842 : : else
1843 : 67 : pathnode->umethod = UNIQUE_PATH_SORT;
1844 : : }
3322 tgl@sss.pgh.pa.us 1845 [ # # ]:UBC 0 : else if (sjinfo->semi_can_btree)
5722 1846 : 0 : pathnode->umethod = UNIQUE_PATH_SORT;
3322 1847 [ # # ]: 0 : else if (sjinfo->semi_can_hash)
5722 1848 : 0 : pathnode->umethod = UNIQUE_PATH_HASH;
1849 : : else
1850 : : {
1851 : : /* we can get here only if we abandoned hashing above */
3322 1852 : 0 : MemoryContextSwitchTo(oldcontext);
1853 : 0 : return NULL;
1854 : : }
1855 : :
7405 tgl@sss.pgh.pa.us 1856 [ + + ]:CBC 1755 : if (pathnode->umethod == UNIQUE_PATH_HASH)
1857 : : {
7753 1858 : 1688 : pathnode->path.startup_cost = agg_path.startup_cost;
1859 : 1688 : pathnode->path.total_cost = agg_path.total_cost;
1860 : : }
1861 : : else
1862 : : {
1863 : 67 : pathnode->path.startup_cost = sort_path.startup_cost;
1864 : 67 : pathnode->path.total_cost = sort_path.total_cost;
1865 : : }
1866 : :
7755 1867 : 1755 : rel->cheapest_unique_path = (Path *) pathnode;
1868 : :
5722 1869 : 1755 : MemoryContextSwitchTo(oldcontext);
1870 : :
7755 1871 : 1755 : return pathnode;
1872 : : }
1873 : :
1874 : : /*
1875 : : * create_gather_merge_path
1876 : : *
1877 : : * Creates a path corresponding to a gather merge scan, returning
1878 : : * the pathnode.
1879 : : */
1880 : : GatherMergePath *
2593 rhaas@postgresql.org 1881 : 4904 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1882 : : PathTarget *target, List *pathkeys,
1883 : : Relids required_outer, double *rows)
1884 : : {
1885 : 4904 : GatherMergePath *pathnode = makeNode(GatherMergePath);
2524 bruce@momjian.us 1886 : 4904 : Cost input_startup_cost = 0;
1887 : 4904 : Cost input_total_cost = 0;
1888 : :
2593 rhaas@postgresql.org 1889 [ - + ]: 4904 : Assert(subpath->parallel_safe);
1890 [ - + ]: 4904 : Assert(pathkeys);
1891 : :
1892 : 4904 : pathnode->path.pathtype = T_GatherMerge;
1893 : 4904 : pathnode->path.parent = rel;
1894 : 4904 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1895 : : required_outer);
1896 : 4904 : pathnode->path.parallel_aware = false;
1897 : :
1898 : 4904 : pathnode->subpath = subpath;
1899 : 4904 : pathnode->num_workers = subpath->parallel_workers;
1900 : 4904 : pathnode->path.pathkeys = pathkeys;
1901 [ - + ]: 4904 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1902 : 4904 : pathnode->path.rows += subpath->rows;
1903 : :
1904 [ + - ]: 4904 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1905 : : {
1906 : : /* Subpath is adequately ordered, we won't need to sort it */
1907 : 4904 : input_startup_cost += subpath->startup_cost;
1908 : 4904 : input_total_cost += subpath->total_cost;
1909 : : }
1910 : : else
1911 : : {
1912 : : /* We'll need to insert a Sort node, so include cost for that */
1913 : : Path sort_path; /* dummy for result of cost_sort */
1914 : :
2593 rhaas@postgresql.org 1915 :UBC 0 : cost_sort(&sort_path,
1916 : : root,
1917 : : pathkeys,
1918 : : subpath->total_cost,
1919 : : subpath->rows,
1920 : 0 : subpath->pathtarget->width,
1921 : : 0.0,
1922 : : work_mem,
1923 : : -1);
1924 : 0 : input_startup_cost += sort_path.startup_cost;
1925 : 0 : input_total_cost += sort_path.total_cost;
1926 : : }
1927 : :
2593 rhaas@postgresql.org 1928 :CBC 4904 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1929 : : input_startup_cost, input_total_cost, rows);
1930 : :
1931 : 4904 : return pathnode;
1932 : : }
1933 : :
1934 : : /*
1935 : : * translate_sub_tlist - get subquery column numbers represented by tlist
1936 : : *
1937 : : * The given targetlist usually contains only Vars referencing the given relid.
1938 : : * Extract their varattnos (ie, the column numbers of the subquery) and return
1939 : : * as an integer List.
1940 : : *
1941 : : * If any of the tlist items is not a simple Var, we cannot determine whether
1942 : : * the subquery's uniqueness condition (if any) matches ours, so punt and
1943 : : * return NIL.
1944 : : */
1945 : : static List *
2960 tgl@sss.pgh.pa.us 1946 : 305 : translate_sub_tlist(List *tlist, int relid)
1947 : : {
1948 : 305 : List *result = NIL;
1949 : : ListCell *l;
1950 : :
1951 [ + - + + : 366 : foreach(l, tlist)
+ + ]
1952 : : {
1953 : 305 : Var *var = (Var *) lfirst(l);
1954 : :
1955 [ + - + + ]: 305 : if (!var || !IsA(var, Var) ||
1956 [ - + ]: 61 : var->varno != relid)
1957 : 244 : return NIL; /* punt */
1958 : :
1959 : 61 : result = lappend_int(result, var->varattno);
1960 : : }
1961 : 61 : return result;
1962 : : }
1963 : :
1964 : : /*
1965 : : * create_gather_path
1966 : : * Creates a path corresponding to a gather scan, returning the
1967 : : * pathnode.
1968 : : *
1969 : : * 'rows' may optionally be set to override row estimates from other sources.
1970 : : */
1971 : : GatherPath *
3119 rhaas@postgresql.org 1972 : 8068 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1973 : : PathTarget *target, Relids required_outer, double *rows)
1974 : : {
1975 : 8068 : GatherPath *pathnode = makeNode(GatherPath);
1976 : :
3007 1977 [ - + ]: 8068 : Assert(subpath->parallel_safe);
1978 : :
3119 1979 : 8068 : pathnode->path.pathtype = T_Gather;
1980 : 8068 : pathnode->path.parent = rel;
2946 1981 : 8068 : pathnode->path.pathtarget = target;
3119 1982 : 8068 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1983 : : required_outer);
3077 1984 : 8068 : pathnode->path.parallel_aware = false;
3007 1985 : 8068 : pathnode->path.parallel_safe = false;
2571 1986 : 8068 : pathnode->path.parallel_workers = 0;
2489 tgl@sss.pgh.pa.us 1987 : 8068 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1988 : :
3119 rhaas@postgresql.org 1989 : 8068 : pathnode->subpath = subpath;
2571 1990 : 8068 : pathnode->num_workers = subpath->parallel_workers;
3007 1991 : 8068 : pathnode->single_copy = false;
1992 : :
2571 1993 [ - + ]: 8068 : if (pathnode->num_workers == 0)
1994 : : {
3007 rhaas@postgresql.org 1995 :UBC 0 : pathnode->path.pathkeys = subpath->pathkeys;
2571 1996 : 0 : pathnode->num_workers = 1;
3007 1997 : 0 : pathnode->single_copy = true;
1998 : : }
1999 : :
2946 rhaas@postgresql.org 2000 :CBC 8068 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
2001 : :
3119 2002 : 8068 : return pathnode;
2003 : : }
2004 : :
2005 : : /*
2006 : : * create_subqueryscan_path
2007 : : * Creates a path corresponding to a scan of a subquery,
2008 : : * returning the pathnode.
2009 : : *
2010 : : * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
2011 : : * be trivial, ie just a fetch of all the subquery output columns in order.
2012 : : * While we could determine that here, the caller can usually do it more
2013 : : * efficiently (or at least amortize it over multiple calls).
2014 : : */
2015 : : SubqueryScanPath *
2960 tgl@sss.pgh.pa.us 2016 : 19387 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
2017 : : bool trivial_pathtarget,
2018 : : List *pathkeys, Relids required_outer)
2019 : : {
2020 : 19387 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
2021 : :
2022 : 19387 : pathnode->path.pathtype = T_SubqueryScan;
2023 : 19387 : pathnode->path.parent = rel;
2953 2024 : 19387 : pathnode->path.pathtarget = rel->reltarget;
2960 2025 : 19387 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2026 : : required_outer);
2027 : 19387 : pathnode->path.parallel_aware = false;
2028 [ + + ]: 31930 : pathnode->path.parallel_safe = rel->consider_parallel &&
2029 [ + + ]: 12543 : subpath->parallel_safe;
2866 rhaas@postgresql.org 2030 : 19387 : pathnode->path.parallel_workers = subpath->parallel_workers;
2960 tgl@sss.pgh.pa.us 2031 : 19387 : pathnode->path.pathkeys = pathkeys;
2032 : 19387 : pathnode->subpath = subpath;
2033 : :
635 2034 : 19387 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
2035 : : trivial_pathtarget);
2036 : :
8598 2037 : 19387 : return pathnode;
2038 : : }
2039 : :
2040 : : /*
2041 : : * create_functionscan_path
2042 : : * Creates a path corresponding to a sequential scan of a function,
2043 : : * returning the pathnode.
2044 : : */
2045 : : Path *
4268 2046 : 21525 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
2047 : : List *pathkeys, Relids required_outer)
2048 : : {
8008 2049 : 21525 : Path *pathnode = makeNode(Path);
2050 : :
2051 : 21525 : pathnode->pathtype = T_FunctionScan;
2052 : 21525 : pathnode->parent = rel;
2953 2053 : 21525 : pathnode->pathtarget = rel->reltarget;
4268 2054 : 21525 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2055 : : required_outer);
3077 rhaas@postgresql.org 2056 : 21525 : pathnode->parallel_aware = false;
3007 2057 : 21525 : pathnode->parallel_safe = rel->consider_parallel;
2866 2058 : 21525 : pathnode->parallel_workers = 0;
3797 tgl@sss.pgh.pa.us 2059 : 21525 : pathnode->pathkeys = pathkeys;
2060 : :
4268 2061 : 21525 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
2062 : :
2594 alvherre@alvh.no-ip. 2063 : 21525 : return pathnode;
2064 : : }
2065 : :
2066 : : /*
2067 : : * create_tablefuncscan_path
2068 : : * Creates a path corresponding to a sequential scan of a table function,
2069 : : * returning the pathnode.
2070 : : */
2071 : : Path *
2072 : 254 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
2073 : : Relids required_outer)
2074 : : {
2075 : 254 : Path *pathnode = makeNode(Path);
2076 : :
2077 : 254 : pathnode->pathtype = T_TableFuncScan;
2078 : 254 : pathnode->parent = rel;
2079 : 254 : pathnode->pathtarget = rel->reltarget;
2080 : 254 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2081 : : required_outer);
2082 : 254 : pathnode->parallel_aware = false;
2083 : 254 : pathnode->parallel_safe = rel->consider_parallel;
2084 : 254 : pathnode->parallel_workers = 0;
2085 : 254 : pathnode->pathkeys = NIL; /* result is always unordered */
2086 : :
2087 : 254 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
2088 : :
8008 tgl@sss.pgh.pa.us 2089 : 254 : return pathnode;
2090 : : }
2091 : :
2092 : : /*
2093 : : * create_valuesscan_path
2094 : : * Creates a path corresponding to a scan of a VALUES list,
2095 : : * returning the pathnode.
2096 : : */
2097 : : Path *
4263 2098 : 3858 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
2099 : : Relids required_outer)
2100 : : {
6465 mail@joeconway.com 2101 : 3858 : Path *pathnode = makeNode(Path);
2102 : :
2103 : 3858 : pathnode->pathtype = T_ValuesScan;
2104 : 3858 : pathnode->parent = rel;
2953 tgl@sss.pgh.pa.us 2105 : 3858 : pathnode->pathtarget = rel->reltarget;
4263 2106 : 3858 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2107 : : required_outer);
3077 rhaas@postgresql.org 2108 : 3858 : pathnode->parallel_aware = false;
3007 2109 : 3858 : pathnode->parallel_safe = rel->consider_parallel;
2866 2110 : 3858 : pathnode->parallel_workers = 0;
6465 mail@joeconway.com 2111 : 3858 : pathnode->pathkeys = NIL; /* result is always unordered */
2112 : :
4263 tgl@sss.pgh.pa.us 2113 : 3858 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
2114 : :
6465 mail@joeconway.com 2115 : 3858 : return pathnode;
2116 : : }
2117 : :
2118 : : /*
2119 : : * create_ctescan_path
2120 : : * Creates a path corresponding to a scan of a non-self-reference CTE,
2121 : : * returning the pathnode.
2122 : : */
2123 : : Path *
19 tgl@sss.pgh.pa.us 2124 :GNC 1603 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel,
2125 : : List *pathkeys, Relids required_outer)
2126 : : {
5671 tgl@sss.pgh.pa.us 2127 :CBC 1603 : Path *pathnode = makeNode(Path);
2128 : :
2129 : 1603 : pathnode->pathtype = T_CteScan;
2130 : 1603 : pathnode->parent = rel;
2953 2131 : 1603 : pathnode->pathtarget = rel->reltarget;
4249 2132 : 1603 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2133 : : required_outer);
3077 rhaas@postgresql.org 2134 : 1603 : pathnode->parallel_aware = false;
3007 2135 : 1603 : pathnode->parallel_safe = rel->consider_parallel;
2866 2136 : 1603 : pathnode->parallel_workers = 0;
19 tgl@sss.pgh.pa.us 2137 :GNC 1603 : pathnode->pathkeys = pathkeys;
2138 : :
4249 tgl@sss.pgh.pa.us 2139 :CBC 1603 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2140 : :
5671 2141 : 1603 : return pathnode;
2142 : : }
2143 : :
2144 : : /*
2145 : : * create_namedtuplestorescan_path
2146 : : * Creates a path corresponding to a scan of a named tuplestore, returning
2147 : : * the pathnode.
2148 : : */
2149 : : Path *
2571 kgrittn@postgresql.o 2150 : 223 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
2151 : : Relids required_outer)
2152 : : {
2153 : 223 : Path *pathnode = makeNode(Path);
2154 : :
2155 : 223 : pathnode->pathtype = T_NamedTuplestoreScan;
2156 : 223 : pathnode->parent = rel;
2157 : 223 : pathnode->pathtarget = rel->reltarget;
2158 : 223 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2159 : : required_outer);
2160 : 223 : pathnode->parallel_aware = false;
2161 : 223 : pathnode->parallel_safe = rel->consider_parallel;
2162 : 223 : pathnode->parallel_workers = 0;
2163 : 223 : pathnode->pathkeys = NIL; /* result is always unordered */
2164 : :
2165 : 223 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
2166 : :
2167 : 223 : return pathnode;
2168 : : }
2169 : :
2170 : : /*
2171 : : * create_resultscan_path
2172 : : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2173 : : * returning the pathnode.
2174 : : */
2175 : : Path *
1903 tgl@sss.pgh.pa.us 2176 : 805 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2177 : : Relids required_outer)
2178 : : {
2179 : 805 : Path *pathnode = makeNode(Path);
2180 : :
2181 : 805 : pathnode->pathtype = T_Result;
2182 : 805 : pathnode->parent = rel;
2183 : 805 : pathnode->pathtarget = rel->reltarget;
2184 : 805 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2185 : : required_outer);
2186 : 805 : pathnode->parallel_aware = false;
2187 : 805 : pathnode->parallel_safe = rel->consider_parallel;
2188 : 805 : pathnode->parallel_workers = 0;
2189 : 805 : pathnode->pathkeys = NIL; /* result is always unordered */
2190 : :
2191 : 805 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2192 : :
2193 : 805 : return pathnode;
2194 : : }
2195 : :
2196 : : /*
2197 : : * create_worktablescan_path
2198 : : * Creates a path corresponding to a scan of a self-reference CTE,
2199 : : * returning the pathnode.
2200 : : */
2201 : : Path *
4249 2202 : 406 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2203 : : Relids required_outer)
2204 : : {
5671 2205 : 406 : Path *pathnode = makeNode(Path);
2206 : :
2207 : 406 : pathnode->pathtype = T_WorkTableScan;
2208 : 406 : pathnode->parent = rel;
2953 2209 : 406 : pathnode->pathtarget = rel->reltarget;
4249 2210 : 406 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2211 : : required_outer);
3077 rhaas@postgresql.org 2212 : 406 : pathnode->parallel_aware = false;
3007 2213 : 406 : pathnode->parallel_safe = rel->consider_parallel;
2866 2214 : 406 : pathnode->parallel_workers = 0;
5671 tgl@sss.pgh.pa.us 2215 : 406 : pathnode->pathkeys = NIL; /* result is always unordered */
2216 : :
2217 : : /* Cost is the same as for a regular CTE scan */
4249 2218 : 406 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2219 : :
5671 2220 : 406 : return pathnode;
2221 : : }
2222 : :
2223 : : /*
2224 : : * create_foreignscan_path
2225 : : * Creates a path corresponding to a scan of a foreign base table,
2226 : : * returning the pathnode.
2227 : : *
2228 : : * This function is never called from core Postgres; rather, it's expected
2229 : : * to be called by the GetForeignPaths function of a foreign data wrapper.
2230 : : * We make the FDW supply all fields of the path, since we do not have any way
2231 : : * to calculate them in core. However, there is a usually-sane default for
2232 : : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2233 : : */
2234 : : ForeignPath *
4423 2235 : 1776 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2236 : : PathTarget *target,
2237 : : double rows, Cost startup_cost, Cost total_cost,
2238 : : List *pathkeys,
2239 : : Relids required_outer,
2240 : : Path *fdw_outerpath,
2241 : : List *fdw_restrictinfo,
2242 : : List *fdw_private)
2243 : : {
4802 2244 : 1776 : ForeignPath *pathnode = makeNode(ForeignPath);
2245 : :
2246 : : /* Historically some FDWs were confused about when to use this */
1893 2247 [ + + - + ]: 1776 : Assert(IS_SIMPLE_REL(rel));
2248 : :
4802 2249 : 1776 : pathnode->path.pathtype = T_ForeignScan;
2250 : 1776 : pathnode->path.parent = rel;
2953 2251 [ + - ]: 1776 : pathnode->path.pathtarget = target ? target : rel->reltarget;
4378 2252 : 1776 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2253 : : required_outer);
3077 rhaas@postgresql.org 2254 : 1776 : pathnode->path.parallel_aware = false;
3007 2255 : 1776 : pathnode->path.parallel_safe = rel->consider_parallel;
1893 tgl@sss.pgh.pa.us 2256 : 1776 : pathnode->path.parallel_workers = 0;
2257 : 1776 : pathnode->path.rows = rows;
2258 : 1776 : pathnode->path.startup_cost = startup_cost;
2259 : 1776 : pathnode->path.total_cost = total_cost;
2260 : 1776 : pathnode->path.pathkeys = pathkeys;
2261 : :
2262 : 1776 : pathnode->fdw_outerpath = fdw_outerpath;
243 efujita@postgresql.o 2263 :GNC 1776 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
1893 tgl@sss.pgh.pa.us 2264 :CBC 1776 : pathnode->fdw_private = fdw_private;
2265 : :
2266 : 1776 : return pathnode;
2267 : : }
2268 : :
2269 : : /*
2270 : : * create_foreign_join_path
2271 : : * Creates a path corresponding to a scan of a foreign join,
2272 : : * returning the pathnode.
2273 : : *
2274 : : * This function is never called from core Postgres; rather, it's expected
2275 : : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2276 : : * We make the FDW supply all fields of the path, since we do not have any way
2277 : : * to calculate them in core. However, there is a usually-sane default for
2278 : : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2279 : : */
2280 : : ForeignPath *
2281 : 560 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2282 : : PathTarget *target,
2283 : : double rows, Cost startup_cost, Cost total_cost,
2284 : : List *pathkeys,
2285 : : Relids required_outer,
2286 : : Path *fdw_outerpath,
2287 : : List *fdw_restrictinfo,
2288 : : List *fdw_private)
2289 : : {
2290 : 560 : ForeignPath *pathnode = makeNode(ForeignPath);
2291 : :
2292 : : /*
2293 : : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2294 : : * but the API of this function doesn't support it, and existing
2295 : : * extensions aren't yet trying to build such paths anyway. For the
2296 : : * moment just throw an error if someone tries it; eventually we should
2297 : : * revisit this.
2298 : : */
2299 [ + - - + ]: 560 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
1893 tgl@sss.pgh.pa.us 2300 [ # # ]:UBC 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2301 : :
1893 tgl@sss.pgh.pa.us 2302 :CBC 560 : pathnode->path.pathtype = T_ForeignScan;
2303 : 560 : pathnode->path.parent = rel;
2304 [ + - ]: 560 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2305 : 560 : pathnode->path.param_info = NULL; /* XXX see above */
2306 : 560 : pathnode->path.parallel_aware = false;
2307 : 560 : pathnode->path.parallel_safe = rel->consider_parallel;
2308 : 560 : pathnode->path.parallel_workers = 0;
2309 : 560 : pathnode->path.rows = rows;
2310 : 560 : pathnode->path.startup_cost = startup_cost;
2311 : 560 : pathnode->path.total_cost = total_cost;
2312 : 560 : pathnode->path.pathkeys = pathkeys;
2313 : :
2314 : 560 : pathnode->fdw_outerpath = fdw_outerpath;
243 efujita@postgresql.o 2315 :GNC 560 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
1893 tgl@sss.pgh.pa.us 2316 :CBC 560 : pathnode->fdw_private = fdw_private;
2317 : :
2318 : 560 : return pathnode;
2319 : : }
2320 : :
2321 : : /*
2322 : : * create_foreign_upper_path
2323 : : * Creates a path corresponding to an upper relation that's computed
2324 : : * directly by an FDW, returning the pathnode.
2325 : : *
2326 : : * This function is never called from core Postgres; rather, it's expected to
2327 : : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2328 : : * We make the FDW supply all fields of the path, since we do not have any way
2329 : : * to calculate them in core. However, there is a usually-sane default for
2330 : : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2331 : : */
2332 : : ForeignPath *
2333 : 285 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2334 : : PathTarget *target,
2335 : : double rows, Cost startup_cost, Cost total_cost,
2336 : : List *pathkeys,
2337 : : Path *fdw_outerpath,
2338 : : List *fdw_restrictinfo,
2339 : : List *fdw_private)
2340 : : {
2341 : 285 : ForeignPath *pathnode = makeNode(ForeignPath);
2342 : :
2343 : : /*
2344 : : * Upper relations should never have any lateral references, since joining
2345 : : * is complete.
2346 : : */
2347 [ - + ]: 285 : Assert(bms_is_empty(rel->lateral_relids));
2348 : :
2349 : 285 : pathnode->path.pathtype = T_ForeignScan;
2350 : 285 : pathnode->path.parent = rel;
2351 [ - + ]: 285 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2352 : 285 : pathnode->path.param_info = NULL;
2353 : 285 : pathnode->path.parallel_aware = false;
2354 : 285 : pathnode->path.parallel_safe = rel->consider_parallel;
2866 rhaas@postgresql.org 2355 : 285 : pathnode->path.parallel_workers = 0;
4423 tgl@sss.pgh.pa.us 2356 : 285 : pathnode->path.rows = rows;
2357 : 285 : pathnode->path.startup_cost = startup_cost;
2358 : 285 : pathnode->path.total_cost = total_cost;
2359 : 285 : pathnode->path.pathkeys = pathkeys;
2360 : :
3050 rhaas@postgresql.org 2361 : 285 : pathnode->fdw_outerpath = fdw_outerpath;
243 efujita@postgresql.o 2362 :GNC 285 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
4423 tgl@sss.pgh.pa.us 2363 :CBC 285 : pathnode->fdw_private = fdw_private;
2364 : :
4802 2365 : 285 : return pathnode;
2366 : : }
2367 : :
2368 : : /*
2369 : : * calc_nestloop_required_outer
2370 : : * Compute the required_outer set for a nestloop join path
2371 : : *
2372 : : * Note: when considering a child join, the inputs nonetheless use top-level
2373 : : * parent relids
2374 : : *
2375 : : * Note: result must not share storage with either input
2376 : : */
2377 : : Relids
2434 rhaas@postgresql.org 2378 : 1221841 : calc_nestloop_required_outer(Relids outerrelids,
2379 : : Relids outer_paramrels,
2380 : : Relids innerrelids,
2381 : : Relids inner_paramrels)
2382 : : {
2383 : : Relids required_outer;
2384 : :
2385 : : /* inner_path can require rels from outer path, but not vice versa */
2386 [ - + ]: 1221841 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2387 : : /* easy case if inner path is not parameterized */
4378 tgl@sss.pgh.pa.us 2388 [ + + ]: 1221841 : if (!inner_paramrels)
2389 : 830306 : return bms_copy(outer_paramrels);
2390 : : /* else, form the union ... */
2391 : 391535 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2392 : : /* ... and remove any mention of now-satisfied outer rels */
4461 2393 : 391535 : required_outer = bms_del_members(required_outer,
2394 : : outerrelids);
2395 : 391535 : return required_outer;
2396 : : }
2397 : :
2398 : : /*
2399 : : * calc_non_nestloop_required_outer
2400 : : * Compute the required_outer set for a merge or hash join path
2401 : : *
2402 : : * Note: result must not share storage with either input
2403 : : */
2404 : : Relids
2405 : 810997 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2406 : : {
4326 bruce@momjian.us 2407 [ + + ]: 810997 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2408 [ + + ]: 810997 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2409 : : Relids innerrelids PG_USED_FOR_ASSERTS_ONLY;
2410 : : Relids outerrelids PG_USED_FOR_ASSERTS_ONLY;
2411 : : Relids required_outer;
2412 : :
2413 : : /*
2414 : : * Any parameterization of the input paths refers to topmost parents of
2415 : : * the relevant relations, because reparameterize_path_by_child() hasn't
2416 : : * been called yet. So we must consider topmost parents of the relations
2417 : : * being joined, too, while checking for disallowed parameterization
2418 : : * cases.
2419 : : */
95 tgl@sss.pgh.pa.us 2420 [ + + ]:GNC 810997 : if (inner_path->parent->top_parent_relids)
2421 : 17218 : innerrelids = inner_path->parent->top_parent_relids;
2422 : : else
2423 : 793779 : innerrelids = inner_path->parent->relids;
2424 : :
2425 [ + + ]: 810997 : if (outer_path->parent->top_parent_relids)
2426 : 17218 : outerrelids = outer_path->parent->top_parent_relids;
2427 : : else
2428 : 793779 : outerrelids = outer_path->parent->relids;
2429 : :
2430 : : /* neither path can require rels from the other */
2431 [ - + ]: 810997 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2432 [ - + ]: 810997 : Assert(!bms_overlap(inner_paramrels, outerrelids));
2433 : : /* form the union ... */
4378 tgl@sss.pgh.pa.us 2434 :CBC 810997 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2435 : : /* we do not need an explicit test for empty; bms_union gets it right */
4461 2436 : 810997 : return required_outer;
2437 : : }
2438 : :
2439 : : /*
2440 : : * create_nestloop_path
2441 : : * Creates a pathnode corresponding to a nestloop join between two
2442 : : * relations.
2443 : : *
2444 : : * 'joinrel' is the join relation.
2445 : : * 'jointype' is the type of join required
2446 : : * 'workspace' is the result from initial_cost_nestloop
2447 : : * 'extra' contains various information about the join
2448 : : * 'outer_path' is the outer path
2449 : : * 'inner_path' is the inner path
2450 : : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2451 : : * 'pathkeys' are the path keys of the new join path
2452 : : * 'required_outer' is the set of required outer rels
2453 : : *
2454 : : * Returns the resulting path node.
2455 : : */
2456 : : NestPath *
6888 2457 : 568941 : create_nestloop_path(PlannerInfo *root,
2458 : : RelOptInfo *joinrel,
2459 : : JoinType jointype,
2460 : : JoinCostWorkspace *workspace,
2461 : : JoinPathExtraData *extra,
2462 : : Path *outer_path,
2463 : : Path *inner_path,
2464 : : List *restrict_clauses,
2465 : : List *pathkeys,
2466 : : Relids required_outer)
2467 : : {
9193 bruce@momjian.us 2468 : 568941 : NestPath *pathnode = makeNode(NestPath);
4378 tgl@sss.pgh.pa.us 2469 [ + + ]: 568941 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2470 : : Relids outerrelids;
2471 : :
2472 : : /*
2473 : : * Paths are parameterized by top-level parents, so run parameterization
2474 : : * tests on the parent relids.
2475 : : */
26 tgl@sss.pgh.pa.us 2476 [ + + ]:GNC 568941 : if (outer_path->parent->top_parent_relids)
2477 : 7673 : outerrelids = outer_path->parent->top_parent_relids;
2478 : : else
2479 : 561268 : outerrelids = outer_path->parent->relids;
2480 : :
2481 : : /*
2482 : : * If the inner path is parameterized by the outer, we must drop any
2483 : : * restrict_clauses that are due to be moved into the inner path. We have
2484 : : * to do this now, rather than postpone the work till createplan time,
2485 : : * because the restrict_clauses list can affect the size and cost
2486 : : * estimates for this path. We detect such clauses by checking for serial
2487 : : * number match to clauses already enforced in the inner path.
2488 : : */
2489 [ + + ]: 568941 : if (bms_overlap(inner_req_outer, outerrelids))
2490 : : {
440 tgl@sss.pgh.pa.us 2491 :CBC 159416 : Bitmapset *enforced_serials = get_param_path_clause_serials(inner_path);
4378 2492 : 159416 : List *jclauses = NIL;
2493 : : ListCell *lc;
2494 : :
2495 [ + + + + : 346704 : foreach(lc, restrict_clauses)
+ + ]
2496 : : {
2497 : 187288 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2498 : :
440 2499 [ + + ]: 187288 : if (!bms_is_member(rinfo->rinfo_serial, enforced_serials))
4461 2500 : 23787 : jclauses = lappend(jclauses, rinfo);
2501 : : }
4378 2502 : 159416 : restrict_clauses = jclauses;
2503 : : }
2504 : :
980 peter@eisentraut.org 2505 : 568941 : pathnode->jpath.path.pathtype = T_NestLoop;
2506 : 568941 : pathnode->jpath.path.parent = joinrel;
2507 : 568941 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2508 : 568941 : pathnode->jpath.path.param_info =
4378 tgl@sss.pgh.pa.us 2509 : 568941 : get_joinrel_parampathinfo(root,
2510 : : joinrel,
2511 : : outer_path,
2512 : : inner_path,
2513 : : extra->sjinfo,
2514 : : required_outer,
2515 : : &restrict_clauses);
980 peter@eisentraut.org 2516 : 568941 : pathnode->jpath.path.parallel_aware = false;
2517 : 1657542 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
3007 rhaas@postgresql.org 2518 [ + + + + : 568941 : outer_path->parallel_safe && inner_path->parallel_safe;
+ + ]
2519 : : /* This is a foolish way to estimate parallel_workers, but for now... */
980 peter@eisentraut.org 2520 : 568941 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2521 : 568941 : pathnode->jpath.path.pathkeys = pathkeys;
2522 : 568941 : pathnode->jpath.jointype = jointype;
2523 : 568941 : pathnode->jpath.inner_unique = extra->inner_unique;
2524 : 568941 : pathnode->jpath.outerjoinpath = outer_path;
2525 : 568941 : pathnode->jpath.innerjoinpath = inner_path;
2526 : 568941 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2527 : :
2564 tgl@sss.pgh.pa.us 2528 : 568941 : final_cost_nestloop(root, pathnode, workspace, extra);
2529 : :
9357 bruce@momjian.us 2530 : 568941 : return pathnode;
2531 : : }
2532 : :
2533 : : /*
2534 : : * create_mergejoin_path
2535 : : * Creates a pathnode corresponding to a mergejoin join between
2536 : : * two relations
2537 : : *
2538 : : * 'joinrel' is the join relation
2539 : : * 'jointype' is the type of join required
2540 : : * 'workspace' is the result from initial_cost_mergejoin
2541 : : * 'extra' contains various information about the join
2542 : : * 'outer_path' is the outer path
2543 : : * 'inner_path' is the inner path
2544 : : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2545 : : * 'pathkeys' are the path keys of the new join path
2546 : : * 'required_outer' is the set of required outer rels
2547 : : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2548 : : * (this should be a subset of the restrict_clauses list)
2549 : : * 'outersortkeys' are the sort varkeys for the outer relation
2550 : : * 'innersortkeys' are the sort varkeys for the inner relation
2551 : : */
2552 : : MergePath *
6888 tgl@sss.pgh.pa.us 2553 : 128228 : create_mergejoin_path(PlannerInfo *root,
2554 : : RelOptInfo *joinrel,
2555 : : JoinType jointype,
2556 : : JoinCostWorkspace *workspace,
2557 : : JoinPathExtraData *extra,
2558 : : Path *outer_path,
2559 : : Path *inner_path,
2560 : : List *restrict_clauses,
2561 : : List *pathkeys,
2562 : : Relids required_outer,
2563 : : List *mergeclauses,
2564 : : List *outersortkeys,
2565 : : List *innersortkeys)
2566 : : {
9715 bruce@momjian.us 2567 : 128228 : MergePath *pathnode = makeNode(MergePath);
2568 : :
9716 2569 : 128228 : pathnode->jpath.path.pathtype = T_MergeJoin;
2570 : 128228 : pathnode->jpath.path.parent = joinrel;
2953 tgl@sss.pgh.pa.us 2571 : 128228 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
4378 2572 : 128228 : pathnode->jpath.path.param_info =
2573 : 128228 : get_joinrel_parampathinfo(root,
2574 : : joinrel,
2575 : : outer_path,
2576 : : inner_path,
2577 : : extra->sjinfo,
2578 : : required_outer,
2579 : : &restrict_clauses);
3077 rhaas@postgresql.org 2580 : 128228 : pathnode->jpath.path.parallel_aware = false;
3007 2581 : 370214 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2582 [ + + + + : 128228 : outer_path->parallel_safe && inner_path->parallel_safe;
+ + ]
2583 : : /* This is a foolish way to estimate parallel_workers, but for now... */
2866 2584 : 128228 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
4461 tgl@sss.pgh.pa.us 2585 : 128228 : pathnode->jpath.path.pathkeys = pathkeys;
8615 2586 : 128228 : pathnode->jpath.jointype = jointype;
2564 2587 : 128228 : pathnode->jpath.inner_unique = extra->inner_unique;
9716 bruce@momjian.us 2588 : 128228 : pathnode->jpath.outerjoinpath = outer_path;
2589 : 128228 : pathnode->jpath.innerjoinpath = inner_path;
8833 tgl@sss.pgh.pa.us 2590 : 128228 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
9716 bruce@momjian.us 2591 : 128228 : pathnode->path_mergeclauses = mergeclauses;
2592 : 128228 : pathnode->outersortkeys = outersortkeys;
2593 : 128228 : pathnode->innersortkeys = innersortkeys;
2594 : : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2595 : : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2596 : :
2564 tgl@sss.pgh.pa.us 2597 : 128228 : final_cost_mergejoin(root, pathnode, workspace, extra);
2598 : :
9357 bruce@momjian.us 2599 : 128228 : return pathnode;
2600 : : }
2601 : :
2602 : : /*
2603 : : * create_hashjoin_path
2604 : : * Creates a pathnode corresponding to a hash join between two relations.
2605 : : *
2606 : : * 'joinrel' is the join relation
2607 : : * 'jointype' is the type of join required
2608 : : * 'workspace' is the result from initial_cost_hashjoin
2609 : : * 'extra' contains various information about the join
2610 : : * 'outer_path' is the cheapest outer path
2611 : : * 'inner_path' is the cheapest inner path
2612 : : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2613 : : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2614 : : * 'required_outer' is the set of required outer rels
2615 : : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2616 : : * (this should be a subset of the restrict_clauses list)
2617 : : */
2618 : : HashPath *
6888 tgl@sss.pgh.pa.us 2619 : 112112 : create_hashjoin_path(PlannerInfo *root,
2620 : : RelOptInfo *joinrel,
2621 : : JoinType jointype,
2622 : : JoinCostWorkspace *workspace,
2623 : : JoinPathExtraData *extra,
2624 : : Path *outer_path,
2625 : : Path *inner_path,
2626 : : bool parallel_hash,
2627 : : List *restrict_clauses,
2628 : : Relids required_outer,
2629 : : List *hashclauses)
2630 : : {
9715 bruce@momjian.us 2631 : 112112 : HashPath *pathnode = makeNode(HashPath);
2632 : :
9716 2633 : 112112 : pathnode->jpath.path.pathtype = T_HashJoin;
2634 : 112112 : pathnode->jpath.path.parent = joinrel;
2953 tgl@sss.pgh.pa.us 2635 : 112112 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
4378 2636 : 112112 : pathnode->jpath.path.param_info =
2637 : 112112 : get_joinrel_parampathinfo(root,
2638 : : joinrel,
2639 : : outer_path,
2640 : : inner_path,
2641 : : extra->sjinfo,
2642 : : required_outer,
2643 : : &restrict_clauses);
2307 andres@anarazel.de 2644 : 112112 : pathnode->jpath.path.parallel_aware =
2645 [ + + + + ]: 112112 : joinrel->consider_parallel && parallel_hash;
3007 rhaas@postgresql.org 2646 : 322390 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2647 [ + + + + : 112112 : outer_path->parallel_safe && inner_path->parallel_safe;
+ + ]
2648 : : /* This is a foolish way to estimate parallel_workers, but for now... */
2866 2649 : 112112 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2650 : :
2651 : : /*
2652 : : * A hashjoin never has pathkeys, since its output ordering is
2653 : : * unpredictable due to possible batching. XXX If the inner relation is
2654 : : * small enough, we could instruct the executor that it must not batch,
2655 : : * and then we could assume that the output inherits the outer relation's
2656 : : * ordering, which might save a sort step. However there is considerable
2657 : : * downside if our estimate of the inner relation size is badly off. For
2658 : : * the moment we don't risk it. (Note also that if we wanted to take this
2659 : : * seriously, joinpath.c would have to consider many more paths for the
2660 : : * outer rel than it does now.)
2661 : : */
9008 tgl@sss.pgh.pa.us 2662 : 112112 : pathnode->jpath.path.pathkeys = NIL;
4461 2663 : 112112 : pathnode->jpath.jointype = jointype;
2564 2664 : 112112 : pathnode->jpath.inner_unique = extra->inner_unique;
4461 2665 : 112112 : pathnode->jpath.outerjoinpath = outer_path;
2666 : 112112 : pathnode->jpath.innerjoinpath = inner_path;
2667 : 112112 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
9716 bruce@momjian.us 2668 : 112112 : pathnode->path_hashclauses = hashclauses;
2669 : : /* final_cost_hashjoin will fill in pathnode->num_batches */
2670 : :
2564 tgl@sss.pgh.pa.us 2671 : 112112 : final_cost_hashjoin(root, pathnode, workspace, extra);
2672 : :
9357 bruce@momjian.us 2673 : 112112 : return pathnode;
2674 : : }
2675 : :
2676 : : /*
2677 : : * create_projection_path
2678 : : * Creates a pathnode that represents performing a projection.
2679 : : *
2680 : : * 'rel' is the parent relation associated with the result
2681 : : * 'subpath' is the path representing the source of data
2682 : : * 'target' is the PathTarget to be computed
2683 : : */
2684 : : ProjectionPath *
2960 tgl@sss.pgh.pa.us 2685 : 186265 : create_projection_path(PlannerInfo *root,
2686 : : RelOptInfo *rel,
2687 : : Path *subpath,
2688 : : PathTarget *target)
2689 : : {
2690 : 186265 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2691 : : PathTarget *oldtarget;
2692 : :
2693 : : /*
2694 : : * We mustn't put a ProjectionPath directly above another; it's useless
2695 : : * and will confuse create_projection_plan. Rather than making sure all
2696 : : * callers handle that, let's implement it here, by stripping off any
2697 : : * ProjectionPath in what we're given. Given this rule, there won't be
2698 : : * more than one.
2699 : : */
1049 2700 [ + + ]: 186265 : if (IsA(subpath, ProjectionPath))
2701 : : {
2702 : 6 : ProjectionPath *subpp = (ProjectionPath *) subpath;
2703 : :
2704 [ - + ]: 6 : Assert(subpp->path.parent == rel);
2705 : 6 : subpath = subpp->subpath;
2706 [ - + ]: 6 : Assert(!IsA(subpath, ProjectionPath));
2707 : : }
2708 : :
2960 2709 : 186265 : pathnode->path.pathtype = T_Result;
2710 : 186265 : pathnode->path.parent = rel;
2711 : 186265 : pathnode->path.pathtarget = target;
2712 : : /* For now, assume we are above any joins, so no parameterization */
2713 : 186265 : pathnode->path.param_info = NULL;
2714 : 186265 : pathnode->path.parallel_aware = false;
2715 : 417204 : pathnode->path.parallel_safe = rel->consider_parallel &&
2844 rhaas@postgresql.org 2716 [ + + + + : 230630 : subpath->parallel_safe &&
+ - ]
2795 tgl@sss.pgh.pa.us 2717 : 44365 : is_parallel_safe(root, (Node *) target->exprs);
2866 rhaas@postgresql.org 2718 : 186265 : pathnode->path.parallel_workers = subpath->parallel_workers;
2719 : : /* Projection does not change the sort order */
2960 tgl@sss.pgh.pa.us 2720 : 186265 : pathnode->path.pathkeys = subpath->pathkeys;
2721 : :
2722 : 186265 : pathnode->subpath = subpath;
2723 : :
2724 : : /*
2725 : : * We might not need a separate Result node. If the input plan node type
2726 : : * can project, we can just tell it to project something else. Or, if it
2727 : : * can't project but the desired target has the same expression list as
2728 : : * what the input will produce anyway, we can still give it the desired
2729 : : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2730 : : * Note: in the latter case, create_projection_plan has to recheck our
2731 : : * conclusion; see comments therein.
2732 : : */
1049 2733 : 186265 : oldtarget = subpath->pathtarget;
2854 2734 [ + + + + ]: 193473 : if (is_projection_capable_path(subpath) ||
2735 : 7208 : equal(oldtarget->exprs, target->exprs))
2736 : : {
2737 : : /* No separate Result node needed */
2738 : 185241 : pathnode->dummypp = true;
2739 : :
2740 : : /*
2741 : : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2742 : : */
2743 : 185241 : pathnode->path.rows = subpath->rows;
2744 : 185241 : pathnode->path.startup_cost = subpath->startup_cost +
2745 : 185241 : (target->cost.startup - oldtarget->cost.startup);
2746 : 185241 : pathnode->path.total_cost = subpath->total_cost +
2747 : 185241 : (target->cost.startup - oldtarget->cost.startup) +
2748 : 185241 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2749 : : }
2750 : : else
2751 : : {
2752 : : /* We really do need the Result node */
2753 : 1024 : pathnode->dummypp = false;
2754 : :
2755 : : /*
2756 : : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2757 : : * evaluating the tlist. There is no qual to worry about.
2758 : : */
2759 : 1024 : pathnode->path.rows = subpath->rows;
2760 : 1024 : pathnode->path.startup_cost = subpath->startup_cost +
2761 : 1024 : target->cost.startup;
2762 : 1024 : pathnode->path.total_cost = subpath->total_cost +
2763 : 1024 : target->cost.startup +
2764 : 1024 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2765 : : }
2766 : :
2960 2767 : 186265 : return pathnode;
2768 : : }
2769 : :
2770 : : /*
2771 : : * apply_projection_to_path
2772 : : * Add a projection step, or just apply the target directly to given path.
2773 : : *
2774 : : * This has the same net effect as create_projection_path(), except that if
2775 : : * a separate Result plan node isn't needed, we just replace the given path's
2776 : : * pathtarget with the desired one. This must be used only when the caller
2777 : : * knows that the given path isn't referenced elsewhere and so can be modified
2778 : : * in-place.
2779 : : *
2780 : : * If the input path is a GatherPath or GatherMergePath, we try to push the
2781 : : * new target down to its input as well; this is a yet more invasive
2782 : : * modification of the input path, which create_projection_path() can't do.
2783 : : *
2784 : : * Note that we mustn't change the source path's parent link; so when it is
2785 : : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2786 : : * not caused any trouble.
2787 : : *
2788 : : * 'rel' is the parent relation associated with the result
2789 : : * 'path' is the path representing the source of data
2790 : : * 'target' is the PathTarget to be computed
2791 : : */
2792 : : Path *
2793 : 13849 : apply_projection_to_path(PlannerInfo *root,
2794 : : RelOptInfo *rel,
2795 : : Path *path,
2796 : : PathTarget *target)
2797 : : {
2798 : : QualCost oldcost;
2799 : :
2800 : : /*
2801 : : * If given path can't project, we might need a Result node, so make a
2802 : : * separate ProjectionPath.
2803 : : */
2854 2804 [ + + ]: 13849 : if (!is_projection_capable_path(path))
2960 2805 : 6657 : return (Path *) create_projection_path(root, rel, path, target);
2806 : :
2807 : : /*
2808 : : * We can just jam the desired tlist into the existing path, being sure to
2809 : : * update its cost estimates appropriately.
2810 : : */
2811 : 7192 : oldcost = path->pathtarget->cost;
2812 : 7192 : path->pathtarget = target;
2813 : :
2814 : 7192 : path->startup_cost += target->cost.startup - oldcost.startup;
2815 : 7192 : path->total_cost += target->cost.startup - oldcost.startup +
2816 : 7192 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2817 : :
2818 : : /*
2819 : : * If the path happens to be a Gather or GatherMerge path, we'd like to
2820 : : * arrange for the subpath to return the required target list so that
2821 : : * workers can help project. But if there is something that is not
2822 : : * parallel-safe in the target expressions, then we can't.
2823 : : */
1429 2824 [ + - + + : 7419 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
+ - ]
2795 2825 : 227 : is_parallel_safe(root, (Node *) target->exprs))
2826 : : {
2827 : : /*
2828 : : * We always use create_projection_path here, even if the subpath is
2829 : : * projection-capable, so as to avoid modifying the subpath in place.
2830 : : * It seems unlikely at present that there could be any other
2831 : : * references to the subpath, but better safe than sorry.
2832 : : *
2833 : : * Note that we don't change the parallel path's cost estimates; it
2834 : : * might be appropriate to do so, to reflect the fact that the bulk of
2835 : : * the target evaluation will happen in workers.
2836 : : */
2344 rhaas@postgresql.org 2837 [ - + ]: 227 : if (IsA(path, GatherPath))
2838 : : {
2344 rhaas@postgresql.org 2839 :UBC 0 : GatherPath *gpath = (GatherPath *) path;
2840 : :
2841 : 0 : gpath->subpath = (Path *)
2842 : 0 : create_projection_path(root,
2843 : 0 : gpath->subpath->parent,
2844 : : gpath->subpath,
2845 : : target);
2846 : : }
2847 : : else
2848 : : {
2344 rhaas@postgresql.org 2849 :CBC 227 : GatherMergePath *gmpath = (GatherMergePath *) path;
2850 : :
2851 : 227 : gmpath->subpath = (Path *)
2852 : 227 : create_projection_path(root,
2853 : 227 : gmpath->subpath->parent,
2854 : : gmpath->subpath,
2855 : : target);
2856 : : }
2857 : : }
2844 2858 [ + + ]: 6965 : else if (path->parallel_safe &&
2795 tgl@sss.pgh.pa.us 2859 [ + + ]: 3027 : !is_parallel_safe(root, (Node *) target->exprs))
2860 : : {
2861 : : /*
2862 : : * We're inserting a parallel-restricted target list into a path
2863 : : * currently marked parallel-safe, so we have to mark it as no longer
2864 : : * safe.
2865 : : */
2844 rhaas@postgresql.org 2866 : 6 : path->parallel_safe = false;
2867 : : }
2868 : :
2960 tgl@sss.pgh.pa.us 2869 : 7192 : return path;
2870 : : }
2871 : :
2872 : : /*
2873 : : * create_set_projection_path
2874 : : * Creates a pathnode that represents performing a projection that
2875 : : * includes set-returning functions.
2876 : : *
2877 : : * 'rel' is the parent relation associated with the result
2878 : : * 'subpath' is the path representing the source of data
2879 : : * 'target' is the PathTarget to be computed
2880 : : */
2881 : : ProjectSetPath *
2643 andres@anarazel.de 2882 : 4247 : create_set_projection_path(PlannerInfo *root,
2883 : : RelOptInfo *rel,
2884 : : Path *subpath,
2885 : : PathTarget *target)
2886 : : {
2887 : 4247 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2888 : : double tlist_rows;
2889 : : ListCell *lc;
2890 : :
2891 : 4247 : pathnode->path.pathtype = T_ProjectSet;
2892 : 4247 : pathnode->path.parent = rel;
2893 : 4247 : pathnode->path.pathtarget = target;
2894 : : /* For now, assume we are above any joins, so no parameterization */
2895 : 4247 : pathnode->path.param_info = NULL;
2896 : 4247 : pathnode->path.parallel_aware = false;
2897 : 9351 : pathnode->path.parallel_safe = rel->consider_parallel &&
2898 [ + + + + : 5086 : subpath->parallel_safe &&
+ - ]
2899 : 839 : is_parallel_safe(root, (Node *) target->exprs);
2900 : 4247 : pathnode->path.parallel_workers = subpath->parallel_workers;
2901 : : /* Projection does not change the sort order XXX? */
2902 : 4247 : pathnode->path.pathkeys = subpath->pathkeys;
2903 : :
2904 : 4247 : pathnode->subpath = subpath;
2905 : :
2906 : : /*
2907 : : * Estimate number of rows produced by SRFs for each row of input; if
2908 : : * there's more than one in this node, use the maximum.
2909 : : */
2910 : 4247 : tlist_rows = 1;
2911 [ + - + + : 9427 : foreach(lc, target->exprs)
+ + ]
2912 : : {
2913 : 5180 : Node *node = (Node *) lfirst(lc);
2914 : : double itemrows;
2915 : :
1891 tgl@sss.pgh.pa.us 2916 : 5180 : itemrows = expression_returns_set_rows(root, node);
2643 andres@anarazel.de 2917 [ + + ]: 5180 : if (tlist_rows < itemrows)
2918 : 4122 : tlist_rows = itemrows;
2919 : : }
2920 : :
2921 : : /*
2922 : : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2923 : : * per input row, and half of cpu_tuple_cost for each added output row.
2924 : : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2925 : : * this estimate later.
2926 : : */
2927 : 4247 : pathnode->path.rows = subpath->rows * tlist_rows;
2928 : 4247 : pathnode->path.startup_cost = subpath->startup_cost +
2929 : 4247 : target->cost.startup;
2930 : 4247 : pathnode->path.total_cost = subpath->total_cost +
2931 : 4247 : target->cost.startup +
2932 : 4247 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2933 : 4247 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2934 : :
2935 : 4247 : return pathnode;
2936 : : }
2937 : :
2938 : : /*
2939 : : * create_incremental_sort_path
2940 : : * Creates a pathnode that represents performing an incremental sort.
2941 : : *
2942 : : * 'rel' is the parent relation associated with the result
2943 : : * 'subpath' is the path representing the source of data
2944 : : * 'pathkeys' represents the desired sort order
2945 : : * 'presorted_keys' is the number of keys by which the input path is
2946 : : * already sorted
2947 : : * 'limit_tuples' is the estimated bound on the number of output tuples,
2948 : : * or -1 if no LIMIT or couldn't estimate
2949 : : */
2950 : : IncrementalSortPath *
1469 tomas.vondra@postgre 2951 : 3837 : create_incremental_sort_path(PlannerInfo *root,
2952 : : RelOptInfo *rel,
2953 : : Path *subpath,
2954 : : List *pathkeys,
2955 : : int presorted_keys,
2956 : : double limit_tuples)
2957 : : {
2958 : 3837 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
2959 : 3837 : SortPath *pathnode = &sort->spath;
2960 : :
2961 : 3837 : pathnode->path.pathtype = T_IncrementalSort;
2962 : 3837 : pathnode->path.parent = rel;
2963 : : /* Sort doesn't project, so use source path's pathtarget */
2964 : 3837 : pathnode->path.pathtarget = subpath->pathtarget;
2965 : : /* For now, assume we are above any joins, so no parameterization */
2966 : 3837 : pathnode->path.param_info = NULL;
2967 : 3837 : pathnode->path.parallel_aware = false;
2968 [ + + ]: 5579 : pathnode->path.parallel_safe = rel->consider_parallel &&
2969 [ + + ]: 1742 : subpath->parallel_safe;
2970 : 3837 : pathnode->path.parallel_workers = subpath->parallel_workers;
2971 : 3837 : pathnode->path.pathkeys = pathkeys;
2972 : :
2973 : 3837 : pathnode->subpath = subpath;
2974 : :
2975 : 3837 : cost_incremental_sort(&pathnode->path,
2976 : : root, pathkeys, presorted_keys,
2977 : : subpath->startup_cost,
2978 : : subpath->total_cost,
2979 : : subpath->rows,
2980 : 3837 : subpath->pathtarget->width,
2981 : : 0.0, /* XXX comparison_cost shouldn't be 0? */
2982 : : work_mem, limit_tuples);
2983 : :
2984 : 3837 : sort->nPresortedCols = presorted_keys;
2985 : :
1231 tgl@sss.pgh.pa.us 2986 : 3837 : return sort;
2987 : : }
2988 : :
2989 : : /*
2990 : : * create_sort_path
2991 : : * Creates a pathnode that represents performing an explicit sort.
2992 : : *
2993 : : * 'rel' is the parent relation associated with the result
2994 : : * 'subpath' is the path representing the source of data
2995 : : * 'pathkeys' represents the desired sort order
2996 : : * 'limit_tuples' is the estimated bound on the number of output tuples,
2997 : : * or -1 if no LIMIT or couldn't estimate
2998 : : */
2999 : : SortPath *
2960 3000 : 40268 : create_sort_path(PlannerInfo *root,
3001 : : RelOptInfo *rel,
3002 : : Path *subpath,
3003 : : List *pathkeys,
3004 : : double limit_tuples)
3005 : : {
3006 : 40268 : SortPath *pathnode = makeNode(SortPath);
3007 : :
3008 : 40268 : pathnode->path.pathtype = T_Sort;
3009 : 40268 : pathnode->path.parent = rel;
3010 : : /* Sort doesn't project, so use source path's pathtarget */
3011 : 40268 : pathnode->path.pathtarget = subpath->pathtarget;
3012 : : /* For now, assume we are above any joins, so no parameterization */
3013 : 40268 : pathnode->path.param_info = NULL;
3014 : 40268 : pathnode->path.parallel_aware = false;
3015 [ + + ]: 68085 : pathnode->path.parallel_safe = rel->consider_parallel &&
3016 [ + + ]: 27817 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3017 : 40268 : pathnode->path.parallel_workers = subpath->parallel_workers;
2960 tgl@sss.pgh.pa.us 3018 : 40268 : pathnode->path.pathkeys = pathkeys;
3019 : :
3020 : 40268 : pathnode->subpath = subpath;
3021 : :
3022 : 40268 : cost_sort(&pathnode->path, root, pathkeys,
3023 : : subpath->total_cost,
3024 : : subpath->rows,
3025 : 40268 : subpath->pathtarget->width,
3026 : : 0.0, /* XXX comparison_cost shouldn't be 0? */
3027 : : work_mem, limit_tuples);
3028 : :
3029 : 40268 : return pathnode;
3030 : : }
3031 : :
3032 : : /*
3033 : : * create_group_path
3034 : : * Creates a pathnode that represents performing grouping of presorted input
3035 : : *
3036 : : * 'rel' is the parent relation associated with the result
3037 : : * 'subpath' is the path representing the source of data
3038 : : * 'target' is the PathTarget to be computed
3039 : : * 'groupClause' is a list of SortGroupClause's representing the grouping
3040 : : * 'qual' is the HAVING quals if any
3041 : : * 'numGroups' is the estimated number of groups
3042 : : */
3043 : : GroupPath *
3044 : 559 : create_group_path(PlannerInfo *root,
3045 : : RelOptInfo *rel,
3046 : : Path *subpath,
3047 : : List *groupClause,
3048 : : List *qual,
3049 : : double numGroups)
3050 : : {
3051 : 559 : GroupPath *pathnode = makeNode(GroupPath);
2217 rhaas@postgresql.org 3052 : 559 : PathTarget *target = rel->reltarget;
3053 : :
2960 tgl@sss.pgh.pa.us 3054 : 559 : pathnode->path.pathtype = T_Group;
3055 : 559 : pathnode->path.parent = rel;
3056 : 559 : pathnode->path.pathtarget = target;
3057 : : /* For now, assume we are above any joins, so no parameterization */
3058 : 559 : pathnode->path.param_info = NULL;
3059 : 559 : pathnode->path.parallel_aware = false;
3060 [ + + ]: 922 : pathnode->path.parallel_safe = rel->consider_parallel &&
3061 [ + + ]: 363 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3062 : 559 : pathnode->path.parallel_workers = subpath->parallel_workers;
3063 : : /* Group doesn't change sort ordering */
2960 tgl@sss.pgh.pa.us 3064 : 559 : pathnode->path.pathkeys = subpath->pathkeys;
3065 : :
3066 : 559 : pathnode->subpath = subpath;
3067 : :
3068 : 559 : pathnode->groupClause = groupClause;
3069 : 559 : pathnode->qual = qual;
3070 : :
3071 : 559 : cost_group(&pathnode->path, root,
3072 : : list_length(groupClause),
3073 : : numGroups,
3074 : : qual,
3075 : : subpath->startup_cost, subpath->total_cost,
3076 : : subpath->rows);
3077 : :
3078 : : /* add tlist eval cost for each output row */
3079 : 559 : pathnode->path.startup_cost += target->cost.startup;
3080 : 559 : pathnode->path.total_cost += target->cost.startup +
3081 : 559 : target->cost.per_tuple * pathnode->path.rows;
3082 : :
3083 : 559 : return pathnode;
3084 : : }
3085 : :
3086 : : /*
3087 : : * create_upper_unique_path
3088 : : * Creates a pathnode that represents performing an explicit Unique step
3089 : : * on presorted input.
3090 : : *
3091 : : * This produces a Unique plan node, but the use-case is so different from
3092 : : * create_unique_path that it doesn't seem worth trying to merge the two.
3093 : : *
3094 : : * 'rel' is the parent relation associated with the result
3095 : : * 'subpath' is the path representing the source of data
3096 : : * 'numCols' is the number of grouping columns
3097 : : * 'numGroups' is the estimated number of groups
3098 : : *
3099 : : * The input path must be sorted on the grouping columns, plus possibly
3100 : : * additional columns; so the first numCols pathkeys are the grouping columns
3101 : : */
3102 : : UpperUniquePath *
3103 : 3314 : create_upper_unique_path(PlannerInfo *root,
3104 : : RelOptInfo *rel,
3105 : : Path *subpath,
3106 : : int numCols,
3107 : : double numGroups)
3108 : : {
3109 : 3314 : UpperUniquePath *pathnode = makeNode(UpperUniquePath);
3110 : :
3111 : 3314 : pathnode->path.pathtype = T_Unique;
3112 : 3314 : pathnode->path.parent = rel;
3113 : : /* Unique doesn't project, so use source path's pathtarget */
3114 : 3314 : pathnode->path.pathtarget = subpath->pathtarget;
3115 : : /* For now, assume we are above any joins, so no parameterization */
3116 : 3314 : pathnode->path.param_info = NULL;
3117 : 3314 : pathnode->path.parallel_aware = false;
3118 [ + + ]: 5083 : pathnode->path.parallel_safe = rel->consider_parallel &&
3119 [ + + ]: 1769 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3120 : 3314 : pathnode->path.parallel_workers = subpath->parallel_workers;
3121 : : /* Unique doesn't change the input ordering */
2960 tgl@sss.pgh.pa.us 3122 : 3314 : pathnode->path.pathkeys = subpath->pathkeys;
3123 : :
3124 : 3314 : pathnode->subpath = subpath;
3125 : 3314 : pathnode->numkeys = numCols;
3126 : :
3127 : : /*
3128 : : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3129 : : * all columns get compared at most of the tuples. (XXX probably this is
3130 : : * an overestimate.)
3131 : : */
3132 : 3314 : pathnode->path.startup_cost = subpath->startup_cost;
3133 : 3314 : pathnode->path.total_cost = subpath->total_cost +
3134 : 3314 : cpu_operator_cost * subpath->rows * numCols;
3135 : 3314 : pathnode->path.rows = numGroups;
3136 : :
3137 : 3314 : return pathnode;
3138 : : }
3139 : :
3140 : : /*
3141 : : * create_agg_path
3142 : : * Creates a pathnode that represents performing aggregation/grouping
3143 : : *
3144 : : * 'rel' is the parent relation associated with the result
3145 : : * 'subpath' is the path representing the source of data
3146 : : * 'target' is the PathTarget to be computed
3147 : : * 'aggstrategy' is the Agg node's basic implementation strategy
3148 : : * 'aggsplit' is the Agg node's aggregate-splitting mode
3149 : : * 'groupClause' is a list of SortGroupClause's representing the grouping
3150 : : * 'qual' is the HAVING quals if any
3151 : : * 'aggcosts' contains cost info about the aggregate functions to be computed
3152 : : * 'numGroups' is the estimated number of groups (1 if not grouping)
3153 : : */
3154 : : AggPath *
3155 : 26687 : create_agg_path(PlannerInfo *root,
3156 : : RelOptInfo *rel,
3157 : : Path *subpath,
3158 : : PathTarget *target,
3159 : : AggStrategy aggstrategy,
3160 : : AggSplit aggsplit,
3161 : : List *groupClause,
3162 : : List *qual,
3163 : : const AggClauseCosts *aggcosts,
3164 : : double numGroups)
3165 : : {
3166 : 26687 : AggPath *pathnode = makeNode(AggPath);
3167 : :
3168 : 26687 : pathnode->path.pathtype = T_Agg;
3169 : 26687 : pathnode->path.parent = rel;
3170 : 26687 : pathnode->path.pathtarget = target;
3171 : : /* For now, assume we are above any joins, so no parameterization */
3172 : 26687 : pathnode->path.param_info = NULL;
3173 : 26687 : pathnode->path.parallel_aware = false;
3174 [ + + ]: 45563 : pathnode->path.parallel_safe = rel->consider_parallel &&
3175 [ + + ]: 18876 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3176 : 26687 : pathnode->path.parallel_workers = subpath->parallel_workers;
3177 : :
2960 tgl@sss.pgh.pa.us 3178 [ + + ]: 26687 : if (aggstrategy == AGG_SORTED)
3179 : : {
3180 : : /*
3181 : : * Attempt to preserve the order of the subpath. Additional pathkeys
3182 : : * may have been added in adjust_group_pathkeys_for_groupagg() to
3183 : : * support ORDER BY / DISTINCT aggregates. Pathkeys added there
3184 : : * belong to columns within the aggregate function, so we must strip
3185 : : * these additional pathkeys off as those columns are unavailable
3186 : : * above the aggregate node.
3187 : : */
188 drowley@postgresql.o 3188 [ + + ]: 3672 : if (list_length(subpath->pathkeys) > root->num_groupby_pathkeys)
3189 : 163 : pathnode->path.pathkeys = list_copy_head(subpath->pathkeys,
3190 : : root->num_groupby_pathkeys);
3191 : : else
3192 : 3509 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3193 : : }
3194 : : else
2960 tgl@sss.pgh.pa.us 3195 : 23015 : pathnode->path.pathkeys = NIL; /* output is unordered */
3196 : :
3197 : 26687 : pathnode->subpath = subpath;
3198 : :
3199 : 26687 : pathnode->aggstrategy = aggstrategy;
2849 3200 : 26687 : pathnode->aggsplit = aggsplit;
2960 3201 : 26687 : pathnode->numGroups = numGroups;
1508 jdavis@postgresql.or 3202 [ + + ]: 26687 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
2960 tgl@sss.pgh.pa.us 3203 : 26687 : pathnode->groupClause = groupClause;
3204 : 26687 : pathnode->qual = qual;
3205 : :
3206 : 26687 : cost_agg(&pathnode->path, root,
3207 : : aggstrategy, aggcosts,
3208 : : list_length(groupClause), numGroups,
3209 : : qual,
3210 : : subpath->startup_cost, subpath->total_cost,
1488 jdavis@postgresql.or 3211 : 26687 : subpath->rows, subpath->pathtarget->width);
3212 : :
3213 : : /* add tlist eval cost for each output row */
2960 tgl@sss.pgh.pa.us 3214 : 26687 : pathnode->path.startup_cost += target->cost.startup;
3215 : 26687 : pathnode->path.total_cost += target->cost.startup +
3216 : 26687 : target->cost.per_tuple * pathnode->path.rows;
3217 : :
3218 : 26687 : return pathnode;
3219 : : }
3220 : :
3221 : : /*
3222 : : * create_groupingsets_path
3223 : : * Creates a pathnode that represents performing GROUPING SETS aggregation
3224 : : *
3225 : : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3226 : : * The input path's result must be sorted to match the last entry in
3227 : : * rollup_groupclauses.
3228 : : *
3229 : : * 'rel' is the parent relation associated with the result
3230 : : * 'subpath' is the path representing the source of data
3231 : : * 'target' is the PathTarget to be computed
3232 : : * 'having_qual' is the HAVING quals if any
3233 : : * 'rollups' is a list of RollupData nodes
3234 : : * 'agg_costs' contains cost info about the aggregate functions to be computed
3235 : : */
3236 : : GroupingSetsPath *
3237 : 896 : create_groupingsets_path(PlannerInfo *root,
3238 : : RelOptInfo *rel,
3239 : : Path *subpath,
3240 : : List *having_qual,
3241 : : AggStrategy aggstrategy,
3242 : : List *rollups,
3243 : : const AggClauseCosts *agg_costs)
3244 : : {
3245 : 896 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
2217 rhaas@postgresql.org 3246 : 896 : PathTarget *target = rel->reltarget;
3247 : : ListCell *lc;
2575 rhodiumtoad@postgres 3248 : 896 : bool is_first = true;
3249 : 896 : bool is_first_sort = true;
3250 : :
3251 : : /* The topmost generated Plan node will be an Agg */
2960 tgl@sss.pgh.pa.us 3252 : 896 : pathnode->path.pathtype = T_Agg;
3253 : 896 : pathnode->path.parent = rel;
3254 : 896 : pathnode->path.pathtarget = target;
3255 : 896 : pathnode->path.param_info = subpath->param_info;
3256 : 896 : pathnode->path.parallel_aware = false;
3257 [ + + ]: 1307 : pathnode->path.parallel_safe = rel->consider_parallel &&
3258 [ + - ]: 411 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3259 : 896 : pathnode->path.parallel_workers = subpath->parallel_workers;
2960 tgl@sss.pgh.pa.us 3260 : 896 : pathnode->subpath = subpath;
3261 : :
3262 : : /*
3263 : : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3264 : : * to AGG_HASHED, here if possible.
3265 : : */
2575 rhodiumtoad@postgres 3266 [ + + + + ]: 1278 : if (aggstrategy == AGG_SORTED &&
3267 : 382 : list_length(rollups) == 1 &&
3268 [ + + ]: 181 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3269 : 21 : aggstrategy = AGG_PLAIN;
3270 : :
3271 [ + + - + ]: 1318 : if (aggstrategy == AGG_MIXED &&
3272 : 422 : list_length(rollups) == 1)
2575 rhodiumtoad@postgres 3273 :UBC 0 : aggstrategy = AGG_HASHED;
3274 : :
3275 : : /*
3276 : : * Output will be in sorted order by group_pathkeys if, and only if, there
3277 : : * is a single rollup operation on a non-empty list of grouping
3278 : : * expressions.
3279 : : */
2575 rhodiumtoad@postgres 3280 [ + + + + ]:CBC 896 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
2960 tgl@sss.pgh.pa.us 3281 : 160 : pathnode->path.pathkeys = root->group_pathkeys;
3282 : : else
3283 : 736 : pathnode->path.pathkeys = NIL;
3284 : :
2575 rhodiumtoad@postgres 3285 : 896 : pathnode->aggstrategy = aggstrategy;
3286 : 896 : pathnode->rollups = rollups;
2960 tgl@sss.pgh.pa.us 3287 : 896 : pathnode->qual = having_qual;
1508 jdavis@postgresql.or 3288 [ + - ]: 896 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3289 : :
2575 rhodiumtoad@postgres 3290 [ - + ]: 896 : Assert(rollups != NIL);
3291 [ + + - + ]: 896 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3292 [ + + - + ]: 896 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3293 : :
3294 [ + - + + : 3198 : foreach(lc, rollups)
+ + ]
3295 : : {
3296 : 2302 : RollupData *rollup = lfirst(lc);
3297 : 2302 : List *gsets = rollup->gsets;
3298 : 2302 : int numGroupCols = list_length(linitial(gsets));
3299 : :
3300 : : /*
3301 : : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3302 : : * (already-sorted) input, and following ones do their own sort.
3303 : : *
3304 : : * In AGG_HASHED mode, there is one rollup for each grouping set.
3305 : : *
3306 : : * In AGG_MIXED mode, the first rollups are hashed, the first
3307 : : * non-hashed one takes the (already-sorted) input, and following ones
3308 : : * do their own sort.
3309 : : */
3310 [ + + ]: 2302 : if (is_first)
3311 : : {
3312 : 896 : cost_agg(&pathnode->path, root,
3313 : : aggstrategy,
3314 : : agg_costs,
3315 : : numGroupCols,
3316 : : rollup->numGroups,
3317 : : having_qual,
3318 : : subpath->startup_cost,
3319 : : subpath->total_cost,
3320 : : subpath->rows,
1488 jdavis@postgresql.or 3321 : 896 : subpath->pathtarget->width);
2575 rhodiumtoad@postgres 3322 : 896 : is_first = false;
3323 [ + + ]: 896 : if (!rollup->is_hashed)
3324 : 382 : is_first_sort = false;
3325 : : }
3326 : : else
3327 : : {
3328 : : Path sort_path; /* dummy for result of cost_sort */
3329 : : Path agg_path; /* dummy for result of cost_agg */
3330 : :
3331 [ + + + + ]: 1406 : if (rollup->is_hashed || is_first_sort)
3332 : : {
3333 : : /*
3334 : : * Account for cost of aggregation, but don't charge input
3335 : : * cost again
3336 : : */
3337 : 1079 : cost_agg(&agg_path, root,
3338 : 1079 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3339 : : agg_costs,
3340 : : numGroupCols,
3341 : : rollup->numGroups,
3342 : : having_qual,
3343 : : 0.0, 0.0,
3344 : : subpath->rows,
1488 jdavis@postgresql.or 3345 [ + + ]: 1079 : subpath->pathtarget->width);
2575 rhodiumtoad@postgres 3346 [ + + ]: 1079 : if (!rollup->is_hashed)
3347 : 422 : is_first_sort = false;
3348 : : }
3349 : : else
3350 : : {
3351 : : /* Account for cost of sort, but don't charge input cost again */
3352 : 327 : cost_sort(&sort_path, root, NIL,
3353 : : 0.0,
3354 : : subpath->rows,
3355 : 327 : subpath->pathtarget->width,
3356 : : 0.0,
3357 : : work_mem,
3358 : : -1.0);
3359 : :
3360 : : /* Account for cost of aggregation */
3361 : :
3362 : 327 : cost_agg(&agg_path, root,
3363 : : AGG_SORTED,
3364 : : agg_costs,
3365 : : numGroupCols,
3366 : : rollup->numGroups,
3367 : : having_qual,
3368 : : sort_path.startup_cost,
3369 : : sort_path.total_cost,
3370 : : sort_path.rows,
1488 jdavis@postgresql.or 3371 : 327 : subpath->pathtarget->width);
3372 : : }
3373 : :
2960 tgl@sss.pgh.pa.us 3374 : 1406 : pathnode->path.total_cost += agg_path.total_cost;
3375 : 1406 : pathnode->path.rows += agg_path.rows;
3376 : : }
3377 : : }
3378 : :
3379 : : /* add tlist eval cost for each output row */
3380 : 896 : pathnode->path.startup_cost += target->cost.startup;
3381 : 896 : pathnode->path.total_cost += target->cost.startup +
3382 : 896 : target->cost.per_tuple * pathnode->path.rows;
3383 : :
3384 : 896 : return pathnode;
3385 : : }
3386 : :
3387 : : /*
3388 : : * create_minmaxagg_path
3389 : : * Creates a pathnode that represents computation of MIN/MAX aggregates
3390 : : *
3391 : : * 'rel' is the parent relation associated with the result
3392 : : * 'target' is the PathTarget to be computed
3393 : : * 'mmaggregates' is a list of MinMaxAggInfo structs
3394 : : * 'quals' is the HAVING quals if any
3395 : : */
3396 : : MinMaxAggPath *
3397 : 193 : create_minmaxagg_path(PlannerInfo *root,
3398 : : RelOptInfo *rel,
3399 : : PathTarget *target,
3400 : : List *mmaggregates,
3401 : : List *quals)
3402 : : {
3403 : 193 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3404 : : Cost initplan_cost;
3405 : : ListCell *lc;
3406 : :
3407 : : /* The topmost generated Plan node will be a Result */
3408 : 193 : pathnode->path.pathtype = T_Result;
3409 : 193 : pathnode->path.parent = rel;
3410 : 193 : pathnode->path.pathtarget = target;
3411 : : /* For now, assume we are above any joins, so no parameterization */
3412 : 193 : pathnode->path.param_info = NULL;
3413 : 193 : pathnode->path.parallel_aware = false;
276 tgl@sss.pgh.pa.us 3414 :GNC 193 : pathnode->path.parallel_safe = true; /* might change below */
2866 rhaas@postgresql.org 3415 :CBC 193 : pathnode->path.parallel_workers = 0;
3416 : : /* Result is one unordered row */
2960 tgl@sss.pgh.pa.us 3417 : 193 : pathnode->path.rows = 1;
3418 : 193 : pathnode->path.pathkeys = NIL;
3419 : :
3420 : 193 : pathnode->mmaggregates = mmaggregates;
3421 : 193 : pathnode->quals = quals;
3422 : :
3423 : : /* Calculate cost of all the initplans, and check parallel safety */
3424 : 193 : initplan_cost = 0;
3425 [ + - + + : 404 : foreach(lc, mmaggregates)
+ + ]
3426 : : {
3427 : 211 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3428 : :
3429 : 211 : initplan_cost += mminfo->pathcost;
276 tgl@sss.pgh.pa.us 3430 [ + + ]:GNC 211 : if (!mminfo->path->parallel_safe)
3431 : 43 : pathnode->path.parallel_safe = false;
3432 : : }
3433 : :
3434 : : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
2960 tgl@sss.pgh.pa.us 3435 :CBC 193 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3436 : 193 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3437 : 193 : target->cost.per_tuple + cpu_tuple_cost;
3438 : :
3439 : : /*
3440 : : * Add cost of qual, if any --- but we ignore its selectivity, since our
3441 : : * rowcount estimate should be 1 no matter what the qual is.
3442 : : */
2355 3443 [ - + ]: 193 : if (quals)
3444 : : {
3445 : : QualCost qual_cost;
3446 : :
2355 tgl@sss.pgh.pa.us 3447 :UBC 0 : cost_qual_eval(&qual_cost, quals, root);
3448 : 0 : pathnode->path.startup_cost += qual_cost.startup;
3449 : 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3450 : : }
3451 : :
3452 : : /*
3453 : : * If the initplans were all parallel-safe, also check safety of the
3454 : : * target and quals. (The Result node itself isn't parallelizable, but if
3455 : : * we are in a subquery then it can be useful for the outer query to know
3456 : : * that this one is parallel-safe.)
3457 : : */
276 tgl@sss.pgh.pa.us 3458 [ + + ]:GNC 193 : if (pathnode->path.parallel_safe)
3459 : 150 : pathnode->path.parallel_safe =
3460 [ + - + - ]: 300 : is_parallel_safe(root, (Node *) target->exprs) &&
3461 : 150 : is_parallel_safe(root, (Node *) quals);
3462 : :
2960 tgl@sss.pgh.pa.us 3463 :CBC 193 : return pathnode;
3464 : : }
3465 : :
3466 : : /*
3467 : : * create_windowagg_path
3468 : : * Creates a pathnode that represents computation of window functions
3469 : : *
3470 : : * 'rel' is the parent relation associated with the result
3471 : : * 'subpath' is the path representing the source of data
3472 : : * 'target' is the PathTarget to be computed
3473 : : * 'windowFuncs' is a list of WindowFunc structs
3474 : : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3475 : : * 'qual' WindowClause.runconditions from lower-level WindowAggPaths.
3476 : : * Must always be NIL when topwindow == false
3477 : : * 'topwindow' pass as true only for the top-level WindowAgg. False for all
3478 : : * intermediate WindowAggs.
3479 : : *
3480 : : * The input must be sorted according to the WindowClause's PARTITION keys
3481 : : * plus ORDER BY keys.
3482 : : */
3483 : : WindowAggPath *
3484 : 1315 : create_windowagg_path(PlannerInfo *root,
3485 : : RelOptInfo *rel,
3486 : : Path *subpath,
3487 : : PathTarget *target,
3488 : : List *windowFuncs,
3489 : : WindowClause *winclause,
3490 : : List *qual,
3491 : : bool topwindow)
3492 : : {
3493 : 1315 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3494 : :
3495 : : /* qual can only be set for the topwindow */
737 drowley@postgresql.o 3496 [ + + - + ]: 1315 : Assert(qual == NIL || topwindow);
3497 : :
2960 tgl@sss.pgh.pa.us 3498 : 1315 : pathnode->path.pathtype = T_WindowAgg;
3499 : 1315 : pathnode->path.parent = rel;
3500 : 1315 : pathnode->path.pathtarget = target;
3501 : : /* For now, assume we are above any joins, so no parameterization */
3502 : 1315 : pathnode->path.param_info = NULL;
3503 : 1315 : pathnode->path.parallel_aware = false;
3504 [ - + ]: 1315 : pathnode->path.parallel_safe = rel->consider_parallel &&
2960 tgl@sss.pgh.pa.us 3505 [ # # ]:UBC 0 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3506 :CBC 1315 : pathnode->path.parallel_workers = subpath->parallel_workers;
3507 : : /* WindowAgg preserves the input sort order */
2960 tgl@sss.pgh.pa.us 3508 : 1315 : pathnode->path.pathkeys = subpath->pathkeys;
3509 : :
3510 : 1315 : pathnode->subpath = subpath;
3511 : 1315 : pathnode->winclause = winclause;
737 drowley@postgresql.o 3512 : 1315 : pathnode->qual = qual;
3513 : 1315 : pathnode->topwindow = topwindow;
3514 : :
3515 : : /*
3516 : : * For costing purposes, assume that there are no redundant partitioning
3517 : : * or ordering columns; it's not worth the trouble to deal with that
3518 : : * corner case here. So we just pass the unmodified list lengths to
3519 : : * cost_windowagg.
3520 : : */
2960 tgl@sss.pgh.pa.us 3521 : 1315 : cost_windowagg(&pathnode->path, root,
3522 : : windowFuncs,
3523 : : winclause,
3524 : : subpath->startup_cost,
3525 : : subpath->total_cost,
3526 : : subpath->rows);
3527 : :
3528 : : /* add tlist eval cost for each output row */
3529 : 1315 : pathnode->path.startup_cost += target->cost.startup;
3530 : 1315 : pathnode->path.total_cost += target->cost.startup +
3531 : 1315 : target->cost.per_tuple * pathnode->path.rows;
3532 : :
3533 : 1315 : return pathnode;
3534 : : }
3535 : :
3536 : : /*
3537 : : * create_setop_path
3538 : : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3539 : : *
3540 : : * 'rel' is the parent relation associated with the result
3541 : : * 'subpath' is the path representing the source of data
3542 : : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3543 : : * 'strategy' is the implementation strategy (sorted or hashed)
3544 : : * 'distinctList' is a list of SortGroupClause's representing the grouping
3545 : : * 'flagColIdx' is the column number where the flag column will be, if any
3546 : : * 'firstFlag' is the flag value for the first input relation when hashing;
3547 : : * or -1 when sorting
3548 : : * 'numGroups' is the estimated number of distinct groups
3549 : : * 'outputRows' is the estimated number of output rows
3550 : : */
3551 : : SetOpPath *
3552 : 304 : create_setop_path(PlannerInfo *root,
3553 : : RelOptInfo *rel,
3554 : : Path *subpath,
3555 : : SetOpCmd cmd,
3556 : : SetOpStrategy strategy,
3557 : : List *distinctList,
3558 : : AttrNumber flagColIdx,
3559 : : int firstFlag,
3560 : : double numGroups,
3561 : : double outputRows)
3562 : : {
3563 : 304 : SetOpPath *pathnode = makeNode(SetOpPath);
3564 : :
3565 : 304 : pathnode->path.pathtype = T_SetOp;
3566 : 304 : pathnode->path.parent = rel;
3567 : : /* SetOp doesn't project, so use source path's pathtarget */
3568 : 304 : pathnode->path.pathtarget = subpath->pathtarget;
3569 : : /* For now, assume we are above any joins, so no parameterization */
3570 : 304 : pathnode->path.param_info = NULL;
3571 : 304 : pathnode->path.parallel_aware = false;
3572 [ - + ]: 304 : pathnode->path.parallel_safe = rel->consider_parallel &&
2960 tgl@sss.pgh.pa.us 3573 [ # # ]:UBC 0 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3574 :CBC 304 : pathnode->path.parallel_workers = subpath->parallel_workers;
3575 : : /* SetOp preserves the input sort order if in sort mode */
2960 tgl@sss.pgh.pa.us 3576 : 304 : pathnode->path.pathkeys =
3577 [ + + ]: 304 : (strategy == SETOP_SORTED) ? subpath->pathkeys : NIL;
3578 : :
3579 : 304 : pathnode->subpath = subpath;
3580 : 304 : pathnode->cmd = cmd;
3581 : 304 : pathnode->strategy = strategy;
3582 : 304 : pathnode->distinctList = distinctList;
3583 : 304 : pathnode->flagColIdx = flagColIdx;
3584 : 304 : pathnode->firstFlag = firstFlag;
3585 : 304 : pathnode->numGroups = numGroups;
3586 : :
3587 : : /*
3588 : : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3589 : : * all columns get compared at most of the tuples.
3590 : : */
3591 : 304 : pathnode->path.startup_cost = subpath->startup_cost;
3592 : 608 : pathnode->path.total_cost = subpath->total_cost +
3593 : 304 : cpu_operator_cost * subpath->rows * list_length(distinctList);
3594 : 304 : pathnode->path.rows = outputRows;
3595 : :
3596 : 304 : return pathnode;
3597 : : }
3598 : :
3599 : : /*
3600 : : * create_recursiveunion_path
3601 : : * Creates a pathnode that represents a recursive UNION node
3602 : : *
3603 : : * 'rel' is the parent relation associated with the result
3604 : : * 'leftpath' is the source of data for the non-recursive term
3605 : : * 'rightpath' is the source of data for the recursive term
3606 : : * 'target' is the PathTarget to be computed
3607 : : * 'distinctList' is a list of SortGroupClause's representing the grouping
3608 : : * 'wtParam' is the ID of Param representing work table
3609 : : * 'numGroups' is the estimated number of groups
3610 : : *
3611 : : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3612 : : */
3613 : : RecursiveUnionPath *
3614 : 403 : create_recursiveunion_path(PlannerInfo *root,
3615 : : RelOptInfo *rel,
3616 : : Path *leftpath,
3617 : : Path *rightpath,
3618 : : PathTarget *target,
3619 : : List *distinctList,
3620 : : int wtParam,
3621 : : double numGroups)
3622 : : {
3623 : 403 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3624 : :
3625 : 403 : pathnode->path.pathtype = T_RecursiveUnion;
3626 : 403 : pathnode->path.parent = rel;
3627 : 403 : pathnode->path.pathtarget = target;
3628 : : /* For now, assume we are above any joins, so no parameterization */
3629 : 403 : pathnode->path.param_info = NULL;
3630 : 403 : pathnode->path.parallel_aware = false;
2960 tgl@sss.pgh.pa.us 3631 :UBC 0 : pathnode->path.parallel_safe = rel->consider_parallel &&
2960 tgl@sss.pgh.pa.us 3632 [ - + - - :CBC 403 : leftpath->parallel_safe && rightpath->parallel_safe;
- - ]
3633 : : /* Foolish, but we'll do it like joins for now: */
2866 rhaas@postgresql.org 3634 : 403 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3635 : : /* RecursiveUnion result is always unsorted */
2960 tgl@sss.pgh.pa.us 3636 : 403 : pathnode->path.pathkeys = NIL;
3637 : :
3638 : 403 : pathnode->leftpath = leftpath;
3639 : 403 : pathnode->rightpath = rightpath;
3640 : 403 : pathnode->distinctList = distinctList;
3641 : 403 : pathnode->wtParam = wtParam;
3642 : 403 : pathnode->numGroups = numGroups;
3643 : :
3644 : 403 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3645 : :
3646 : 403 : return pathnode;
3647 : : }
3648 : :
3649 : : /*
3650 : : * create_lockrows_path
3651 : : * Creates a pathnode that represents acquiring row locks
3652 : : *
3653 : : * 'rel' is the parent relation associated with the result
3654 : : * 'subpath' is the path representing the source of data
3655 : : * 'rowMarks' is a list of PlanRowMark's
3656 : : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3657 : : */
3658 : : LockRowsPath *
3659 : 4041 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3660 : : Path *subpath, List *rowMarks, int epqParam)
3661 : : {
3662 : 4041 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3663 : :
3664 : 4041 : pathnode->path.pathtype = T_LockRows;
3665 : 4041 : pathnode->path.parent = rel;
3666 : : /* LockRows doesn't project, so use source path's pathtarget */
3667 : 4041 : pathnode->path.pathtarget = subpath->pathtarget;
3668 : : /* For now, assume we are above any joins, so no parameterization */
3669 : 4041 : pathnode->path.param_info = NULL;
3670 : 4041 : pathnode->path.parallel_aware = false;
3671 : 4041 : pathnode->path.parallel_safe = false;
2866 rhaas@postgresql.org 3672 : 4041 : pathnode->path.parallel_workers = 0;
2960 tgl@sss.pgh.pa.us 3673 : 4041 : pathnode->path.rows = subpath->rows;
3674 : :
3675 : : /*
3676 : : * The result cannot be assumed sorted, since locking might cause the sort
3677 : : * key columns to be replaced with new values.
3678 : : */
3679 : 4041 : pathnode->path.pathkeys = NIL;
3680 : :
3681 : 4041 : pathnode->subpath = subpath;
3682 : 4041 : pathnode->rowMarks = rowMarks;
3683 : 4041 : pathnode->epqParam = epqParam;
3684 : :
3685 : : /*
3686 : : * We should charge something extra for the costs of row locking and
3687 : : * possible refetches, but it's hard to say how much. For now, use
3688 : : * cpu_tuple_cost per row.
3689 : : */
3690 : 4041 : pathnode->path.startup_cost = subpath->startup_cost;
3691 : 4041 : pathnode->path.total_cost = subpath->total_cost +
3692 : 4041 : cpu_tuple_cost * subpath->rows;
3693 : :
3694 : 4041 : return pathnode;
3695 : : }
3696 : :
3697 : : /*
3698 : : * create_modifytable_path
3699 : : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE/MERGE
3700 : : * mods
3701 : : *
3702 : : * 'rel' is the parent relation associated with the result
3703 : : * 'subpath' is a Path producing source data
3704 : : * 'operation' is the operation type
3705 : : * 'canSetTag' is true if we set the command tag/es_processed
3706 : : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3707 : : * 'rootRelation' is the partitioned/inherited table root RTI, or 0 if none
3708 : : * 'partColsUpdated' is true if any partitioning columns are being updated,
3709 : : * either from the target relation or a descendent partitioned table.
3710 : : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3711 : : * 'updateColnosLists' is a list of UPDATE target column number lists
3712 : : * (one sublist per rel); or NIL if not an UPDATE
3713 : : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3714 : : * 'returningLists' is a list of RETURNING tlists (one per rel)
3715 : : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3716 : : * 'onconflict' is the ON CONFLICT clause, or NULL
3717 : : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3718 : : * 'mergeActionLists' is a list of lists of MERGE actions (one per rel)
3719 : : * 'mergeJoinConditions' is a list of join conditions for MERGE (one per rel)
3720 : : */
3721 : : ModifyTablePath *
3722 : 43745 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3723 : : Path *subpath,
3724 : : CmdType operation, bool canSetTag,
3725 : : Index nominalRelation, Index rootRelation,
3726 : : bool partColsUpdated,
3727 : : List *resultRelations,
3728 : : List *updateColnosLists,
3729 : : List *withCheckOptionLists, List *returningLists,
3730 : : List *rowMarks, OnConflictExpr *onconflict,
3731 : : List *mergeActionLists, List *mergeJoinConditions,
3732 : : int epqParam)
3733 : : {
3734 : 43745 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3735 : :
748 alvherre@alvh.no-ip. 3736 [ + + + + : 43745 : Assert(operation == CMD_MERGE ||
- + ]
3737 : : (operation == CMD_UPDATE ?
3738 : : list_length(resultRelations) == list_length(updateColnosLists) :
3739 : : updateColnosLists == NIL));
2960 tgl@sss.pgh.pa.us 3740 [ + + - + ]: 43745 : Assert(withCheckOptionLists == NIL ||
3741 : : list_length(resultRelations) == list_length(withCheckOptionLists));
3742 [ + + - + ]: 43745 : Assert(returningLists == NIL ||
3743 : : list_length(resultRelations) == list_length(returningLists));
3744 : :
3745 : 43745 : pathnode->path.pathtype = T_ModifyTable;
3746 : 43745 : pathnode->path.parent = rel;
3747 : : /* pathtarget is not interesting, just make it minimally valid */
2953 3748 : 43745 : pathnode->path.pathtarget = rel->reltarget;
3749 : : /* For now, assume we are above any joins, so no parameterization */
2960 3750 : 43745 : pathnode->path.param_info = NULL;
3751 : 43745 : pathnode->path.parallel_aware = false;
3752 : 43745 : pathnode->path.parallel_safe = false;
2866 rhaas@postgresql.org 3753 : 43745 : pathnode->path.parallel_workers = 0;
2960 tgl@sss.pgh.pa.us 3754 : 43745 : pathnode->path.pathkeys = NIL;
3755 : :
3756 : : /*
3757 : : * Compute cost & rowcount as subpath cost & rowcount (if RETURNING)
3758 : : *
3759 : : * Currently, we don't charge anything extra for the actual table
3760 : : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3761 : : * expressions if any. It would only be window dressing, since
3762 : : * ModifyTable is always a top-level node and there is no way for the
3763 : : * costs to change any higher-level planning choices. But we might want
3764 : : * to make it look better sometime.
3765 : : */
1110 3766 : 43745 : pathnode->path.startup_cost = subpath->startup_cost;
3767 : 43745 : pathnode->path.total_cost = subpath->total_cost;
3768 [ + + ]: 43745 : if (returningLists != NIL)
3769 : : {
3770 : 1229 : pathnode->path.rows = subpath->rows;
3771 : :
3772 : : /*
3773 : : * Set width to match the subpath output. XXX this is totally wrong:
3774 : : * we should return an average of the RETURNING tlist widths. But
3775 : : * it's what happened historically, and improving it is a task for
3776 : : * another day. (Again, it's mostly window dressing.)
3777 : : */
3778 : 1229 : pathnode->path.pathtarget->width = subpath->pathtarget->width;
3779 : : }
3780 : : else
3781 : : {
3782 : 42516 : pathnode->path.rows = 0;
3783 : 42516 : pathnode->path.pathtarget->width = 0;
3784 : : }
3785 : :
3786 : 43745 : pathnode->subpath = subpath;
2960 3787 : 43745 : pathnode->operation = operation;
3788 : 43745 : pathnode->canSetTag = canSetTag;
3789 : 43745 : pathnode->nominalRelation = nominalRelation;
2016 3790 : 43745 : pathnode->rootRelation = rootRelation;
2277 rhaas@postgresql.org 3791 : 43745 : pathnode->partColsUpdated = partColsUpdated;
2960 tgl@sss.pgh.pa.us 3792 : 43745 : pathnode->resultRelations = resultRelations;
1110 3793 : 43745 : pathnode->updateColnosLists = updateColnosLists;
2960 3794 : 43745 : pathnode->withCheckOptionLists = withCheckOptionLists;
3795 : 43745 : pathnode->returningLists = returningLists;
3796 : 43745 : pathnode->rowMarks = rowMarks;
3797 : 43745 : pathnode->onconflict = onconflict;
3798 : 43745 : pathnode->epqParam = epqParam;
748 alvherre@alvh.no-ip. 3799 : 43745 : pathnode->mergeActionLists = mergeActionLists;
15 dean.a.rasheed@gmail 3800 :GNC 43745 : pathnode->mergeJoinConditions = mergeJoinConditions;
3801 : :
2960 tgl@sss.pgh.pa.us 3802 :CBC 43745 : return pathnode;
3803 : : }
3804 : :
3805 : : /*
3806 : : * create_limit_path
3807 : : * Creates a pathnode that represents performing LIMIT/OFFSET
3808 : : *
3809 : : * In addition to providing the actual OFFSET and LIMIT expressions,
3810 : : * the caller must provide estimates of their values for costing purposes.
3811 : : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3812 : : * the clause not being present, and -1 means it's present but we could
3813 : : * not estimate its value.
3814 : : *
3815 : : * 'rel' is the parent relation associated with the result
3816 : : * 'subpath' is the path representing the source of data
3817 : : * 'limitOffset' is the actual OFFSET expression, or NULL
3818 : : * 'limitCount' is the actual LIMIT expression, or NULL
3819 : : * 'offset_est' is the estimated value of the OFFSET expression
3820 : : * 'count_est' is the estimated value of the LIMIT expression
3821 : : */
3822 : : LimitPath *
3823 : 2799 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3824 : : Path *subpath,
3825 : : Node *limitOffset, Node *limitCount,
3826 : : LimitOption limitOption,
3827 : : int64 offset_est, int64 count_est)
3828 : : {
3829 : 2799 : LimitPath *pathnode = makeNode(LimitPath);
3830 : :
3831 : 2799 : pathnode->path.pathtype = T_Limit;
3832 : 2799 : pathnode->path.parent = rel;
3833 : : /* Limit doesn't project, so use source path's pathtarget */
3834 : 2799 : pathnode->path.pathtarget = subpath->pathtarget;
3835 : : /* For now, assume we are above any joins, so no parameterization */
3836 : 2799 : pathnode->path.param_info = NULL;
3837 : 2799 : pathnode->path.parallel_aware = false;
3838 [ + + ]: 3907 : pathnode->path.parallel_safe = rel->consider_parallel &&
3839 [ + + ]: 1108 : subpath->parallel_safe;
2866 rhaas@postgresql.org 3840 : 2799 : pathnode->path.parallel_workers = subpath->parallel_workers;
2960 tgl@sss.pgh.pa.us 3841 : 2799 : pathnode->path.rows = subpath->rows;
3842 : 2799 : pathnode->path.startup_cost = subpath->startup_cost;
3843 : 2799 : pathnode->path.total_cost = subpath->total_cost;
3844 : 2799 : pathnode->path.pathkeys = subpath->pathkeys;
3845 : 2799 : pathnode->subpath = subpath;
3846 : 2799 : pathnode->limitOffset = limitOffset;
3847 : 2799 : pathnode->limitCount = limitCount;
1468 alvherre@alvh.no-ip. 3848 : 2799 : pathnode->limitOption = limitOption;
3849 : :
3850 : : /*
3851 : : * Adjust the output rows count and costs according to the offset/limit.
3852 : : */
1839 efujita@postgresql.o 3853 : 2799 : adjust_limit_rows_costs(&pathnode->path.rows,
3854 : : &pathnode->path.startup_cost,
3855 : : &pathnode->path.total_cost,
3856 : : offset_est, count_est);
3857 : :
3858 : 2799 : return pathnode;
3859 : : }
3860 : :
3861 : : /*
3862 : : * adjust_limit_rows_costs
3863 : : * Adjust the size and cost estimates for a LimitPath node according to the
3864 : : * offset/limit.
3865 : : *
3866 : : * This is only a cosmetic issue if we are at top level, but if we are
3867 : : * building a subquery then it's important to report correct info to the outer
3868 : : * planner.
3869 : : *
3870 : : * When the offset or count couldn't be estimated, use 10% of the estimated
3871 : : * number of rows emitted from the subpath.
3872 : : *
3873 : : * XXX we don't bother to add eval costs of the offset/limit expressions
3874 : : * themselves to the path costs. In theory we should, but in most cases those
3875 : : * expressions are trivial and it's just not worth the trouble.
3876 : : */
3877 : : void
3878 : 2890 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3879 : : Cost *startup_cost, /* in/out parameter */
3880 : : Cost *total_cost, /* in/out parameter */
3881 : : int64 offset_est,
3882 : : int64 count_est)
3883 : : {
3884 : 2890 : double input_rows = *rows;
3885 : 2890 : Cost input_startup_cost = *startup_cost;
3886 : 2890 : Cost input_total_cost = *total_cost;
3887 : :
2960 tgl@sss.pgh.pa.us 3888 [ + + ]: 2890 : if (offset_est != 0)
3889 : : {
3890 : : double offset_rows;
3891 : :
3892 [ + + ]: 336 : if (offset_est > 0)
3893 : 324 : offset_rows = (double) offset_est;
3894 : : else
1839 efujita@postgresql.o 3895 : 12 : offset_rows = clamp_row_est(input_rows * 0.10);
3896 [ + + ]: 336 : if (offset_rows > *rows)
3897 : 19 : offset_rows = *rows;
3898 [ + - ]: 336 : if (input_rows > 0)
3899 : 336 : *startup_cost +=
3900 : 336 : (input_total_cost - input_startup_cost)
3901 : 336 : * offset_rows / input_rows;
3902 : 336 : *rows -= offset_rows;
3903 [ + + ]: 336 : if (*rows < 1)
3904 : 19 : *rows = 1;
3905 : : }
3906 : :
2960 tgl@sss.pgh.pa.us 3907 [ + + ]: 2890 : if (count_est != 0)
3908 : : {
3909 : : double count_rows;
3910 : :
3911 [ + + ]: 2866 : if (count_est > 0)
3912 : 2863 : count_rows = (double) count_est;
3913 : : else
1839 efujita@postgresql.o 3914 : 3 : count_rows = clamp_row_est(input_rows * 0.10);
3915 [ + + ]: 2866 : if (count_rows > *rows)
3916 : 136 : count_rows = *rows;
3917 [ + - ]: 2866 : if (input_rows > 0)
3918 : 2866 : *total_cost = *startup_cost +
3919 : 2866 : (input_total_cost - input_startup_cost)
3920 : 2866 : * count_rows / input_rows;
3921 : 2866 : *rows = count_rows;
3922 [ - + ]: 2866 : if (*rows < 1)
1839 efujita@postgresql.o 3923 :UBC 0 : *rows = 1;
3924 : : }
2960 tgl@sss.pgh.pa.us 3925 :CBC 2890 : }
3926 : :
3927 : :
3928 : : /*
3929 : : * reparameterize_path
3930 : : * Attempt to modify a Path to have greater parameterization
3931 : : *
3932 : : * We use this to attempt to bring all child paths of an appendrel to the
3933 : : * same parameterization level, ensuring that they all enforce the same set
3934 : : * of join quals (and thus that that parameterization can be attributed to
3935 : : * an append path built from such paths). Currently, only a few path types
3936 : : * are supported here, though more could be added at need. We return NULL
3937 : : * if we can't reparameterize the given path.
3938 : : *
3939 : : * Note: we intentionally do not pass created paths to add_path(); it would
3940 : : * possibly try to delete them on the grounds of being cost-inferior to the
3941 : : * paths they were made from, and we don't want that. Paths made here are
3942 : : * not necessarily of general-purpose usefulness, but they can be useful
3943 : : * as members of an append path.
3944 : : */
3945 : : Path *
4378 3946 : 154 : reparameterize_path(PlannerInfo *root, Path *path,
3947 : : Relids required_outer,
3948 : : double loop_count)
3949 : : {
3950 : 154 : RelOptInfo *rel = path->parent;
3951 : :
3952 : : /* Can only increase, not decrease, path's parameterization */
3953 [ - + - + ]: 154 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
4378 tgl@sss.pgh.pa.us 3954 :UBC 0 : return NULL;
4378 tgl@sss.pgh.pa.us 3955 [ + - - - :CBC 154 : switch (path->pathtype)
- + - - -
+ ]
3956 : : {
3957 : 114 : case T_SeqScan:
3077 rhaas@postgresql.org 3958 : 114 : return create_seqscan_path(root, rel, required_outer, 0);
3186 tgl@sss.pgh.pa.us 3959 :UBC 0 : case T_SampleScan:
3960 : 0 : return (Path *) create_samplescan_path(root, rel, required_outer);
4378 3961 : 0 : case T_IndexScan:
3962 : : case T_IndexOnlyScan:
3963 : : {
4326 bruce@momjian.us 3964 : 0 : IndexPath *ipath = (IndexPath *) path;
3965 : 0 : IndexPath *newpath = makeNode(IndexPath);
3966 : :
3967 : : /*
3968 : : * We can't use create_index_path directly, and would not want
3969 : : * to because it would re-compute the indexqual conditions
3970 : : * which is wasted effort. Instead we hack things a bit:
3971 : : * flat-copy the path node, revise its param_info, and redo
3972 : : * the cost estimate.
3973 : : */
3974 : 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3975 : 0 : newpath->path.param_info =
3976 : 0 : get_baserel_parampathinfo(root, rel, required_outer);
2615 rhaas@postgresql.org 3977 : 0 : cost_index(newpath, root, loop_count, false);
4326 bruce@momjian.us 3978 : 0 : return (Path *) newpath;
3979 : : }
4378 tgl@sss.pgh.pa.us 3980 : 0 : case T_BitmapHeapScan:
3981 : : {
4326 bruce@momjian.us 3982 : 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3983 : :
3984 : 0 : return (Path *) create_bitmap_heap_path(root,
3985 : : rel,
3986 : : bpath->bitmapqual,
3987 : : required_outer,
3988 : : loop_count, 0);
3989 : : }
4378 tgl@sss.pgh.pa.us 3990 : 0 : case T_SubqueryScan:
3991 : : {
2960 3992 : 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
635 3993 : 0 : Path *subpath = spath->subpath;
3994 : : bool trivial_pathtarget;
3995 : :
3996 : : /*
3997 : : * If existing node has zero extra cost, we must have decided
3998 : : * its target is trivial. (The converse is not true, because
3999 : : * it might have a trivial target but quals to enforce; but in
4000 : : * that case the new node will too, so it doesn't matter
4001 : : * whether we get the right answer here.)
4002 : : */
4003 : 0 : trivial_pathtarget =
4004 : 0 : (subpath->total_cost == spath->path.total_cost);
4005 : :
2960 4006 : 0 : return (Path *) create_subqueryscan_path(root,
4007 : : rel,
4008 : : subpath,
4009 : : trivial_pathtarget,
4010 : : spath->path.pathkeys,
4011 : : required_outer);
4012 : : }
1903 tgl@sss.pgh.pa.us 4013 :CBC 24 : case T_Result:
4014 : : /* Supported only for RTE_RESULT scan paths */
4015 [ + - ]: 24 : if (IsA(path, Path))
4016 : 24 : return create_resultscan_path(root, rel, required_outer);
1903 tgl@sss.pgh.pa.us 4017 :UBC 0 : break;
2273 4018 : 0 : case T_Append:
4019 : : {
4020 : 0 : AppendPath *apath = (AppendPath *) path;
4021 : 0 : List *childpaths = NIL;
4022 : 0 : List *partialpaths = NIL;
4023 : : int i;
4024 : : ListCell *lc;
4025 : :
4026 : : /* Reparameterize the children */
4027 : 0 : i = 0;
4028 [ # # # # : 0 : foreach(lc, apath->subpaths)
# # ]
4029 : : {
4030 : 0 : Path *spath = (Path *) lfirst(lc);
4031 : :
4032 : 0 : spath = reparameterize_path(root, spath,
4033 : : required_outer,
4034 : : loop_count);
4035 [ # # ]: 0 : if (spath == NULL)
4036 : 0 : return NULL;
4037 : : /* We have to re-split the regular and partial paths */
4038 [ # # ]: 0 : if (i < apath->first_partial_path)
4039 : 0 : childpaths = lappend(childpaths, spath);
4040 : : else
4041 : 0 : partialpaths = lappend(partialpaths, spath);
4042 : 0 : i++;
4043 : : }
4044 : 0 : return (Path *)
2199 alvherre@alvh.no-ip. 4045 : 0 : create_append_path(root, rel, childpaths, partialpaths,
4046 : : apath->path.pathkeys, required_outer,
4047 : : apath->path.parallel_workers,
2273 tgl@sss.pgh.pa.us 4048 : 0 : apath->path.parallel_aware,
4049 : : -1);
4050 : : }
497 4051 : 0 : case T_Material:
4052 : : {
4053 : 0 : MaterialPath *mpath = (MaterialPath *) path;
4054 : 0 : Path *spath = mpath->subpath;
4055 : :
4056 : 0 : spath = reparameterize_path(root, spath,
4057 : : required_outer,
4058 : : loop_count);
4059 [ # # ]: 0 : if (spath == NULL)
4060 : 0 : return NULL;
4061 : 0 : return (Path *) create_material_path(rel, spath);
4062 : : }
1005 drowley@postgresql.o 4063 : 0 : case T_Memoize:
4064 : : {
4065 : 0 : MemoizePath *mpath = (MemoizePath *) path;
497 tgl@sss.pgh.pa.us 4066 : 0 : Path *spath = mpath->subpath;
4067 : :
4068 : 0 : spath = reparameterize_path(root, spath,
4069 : : required_outer,
4070 : : loop_count);
4071 [ # # ]: 0 : if (spath == NULL)
4072 : 0 : return NULL;
1005 drowley@postgresql.o 4073 : 0 : return (Path *) create_memoize_path(root, rel,
4074 : : spath,
4075 : : mpath->param_exprs,
4076 : : mpath->hash_operators,
4077 : 0 : mpath->singlerow,
872 4078 : 0 : mpath->binary_mode,
4079 : : mpath->calls);
4080 : : }
4378 tgl@sss.pgh.pa.us 4081 :CBC 16 : default:
4082 : 16 : break;
4083 : : }
4084 : 16 : return NULL;
4085 : : }
4086 : :
4087 : : /*
4088 : : * reparameterize_path_by_child
4089 : : * Given a path parameterized by the parent of the given child relation,
4090 : : * translate the path to be parameterized by the given child relation.
4091 : : *
4092 : : * Most fields in the path are not changed, but any expressions must be
4093 : : * adjusted to refer to the correct varnos, and any subpaths must be
4094 : : * recursively reparameterized. Other fields that refer to specific relids
4095 : : * also need adjustment.
4096 : : *
4097 : : * The cost, number of rows, width and parallel path properties depend upon
4098 : : * path->parent, which does not change during the translation. So we need
4099 : : * not change those.
4100 : : *
4101 : : * Currently, only a few path types are supported here, though more could be
4102 : : * added at need. We return NULL if we can't reparameterize the given path.
4103 : : *
4104 : : * Note that this function can change referenced RangeTblEntries, RelOptInfos
4105 : : * and IndexOptInfos as well as the Path structures. Therefore, it's only safe
4106 : : * to call during create_plan(), when we have made a final choice of which Path
4107 : : * to use for each RangeTblEntry/RelOptInfo/IndexOptInfo.
4108 : : *
4109 : : * Keep this code in sync with path_is_reparameterizable_by_child()!
4110 : : */
4111 : : Path *
2382 rhaas@postgresql.org 4112 : 40057 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
4113 : : RelOptInfo *child_rel)
4114 : : {
4115 : : Path *new_path;
4116 : : ParamPathInfo *new_ppi;
4117 : : ParamPathInfo *old_ppi;
4118 : : Relids required_outer;
4119 : :
4120 : : #define ADJUST_CHILD_ATTRS(node) \
4121 : : ((node) = (void *) adjust_appendrel_attrs_multilevel(root, \
4122 : : (Node *) (node), \
4123 : : child_rel, \
4124 : : child_rel->top_parent))
4125 : :
4126 : : #define REPARAMETERIZE_CHILD_PATH(path) \
4127 : : do { \
4128 : : (path) = reparameterize_path_by_child(root, (path), child_rel); \
4129 : : if ((path) == NULL) \
4130 : : return NULL; \
4131 : : } while(0)
4132 : :
4133 : : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
4134 : : do { \
4135 : : if ((pathlist) != NIL) \
4136 : : { \
4137 : : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
4138 : : child_rel); \
4139 : : if ((pathlist) == NIL) \
4140 : : return NULL; \
4141 : : } \
4142 : : } while(0)
4143 : :
4144 : : /*
4145 : : * If the path is not parameterized by the parent of the given relation,
4146 : : * it doesn't need reparameterization.
4147 : : */
4148 [ + + ]: 40057 : if (!path->param_info ||
4149 [ + - + + ]: 21093 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4150 : 39613 : return path;
4151 : :
4152 : : /*
4153 : : * If possible, reparameterize the given path.
4154 : : *
4155 : : * This function is currently only applied to the inner side of a nestloop
4156 : : * join that is being partitioned by the partitionwise-join code. Hence,
4157 : : * we need only support path types that plausibly arise in that context.
4158 : : * (In particular, supporting sorted path types would be a waste of code
4159 : : * and cycles: even if we translated them here, they'd just lose in
4160 : : * subsequent cost comparisons.) If we do see an unsupported path type,
4161 : : * that just means we won't be able to generate a partitionwise-join plan
4162 : : * using that path type.
4163 : : */
4164 [ + + + + : 444 : switch (nodeTag(path))
+ - - + -
+ + - + -
- ]
4165 : : {
4166 : 114 : case T_Path:
26 tgl@sss.pgh.pa.us 4167 :GNC 114 : new_path = path;
4168 : 114 : ADJUST_CHILD_ATTRS(new_path->parent->baserestrictinfo);
26 tgl@sss.pgh.pa.us 4169 [ + + ]:CBC 114 : if (path->pathtype == T_SampleScan)
4170 : : {
4171 : 24 : Index scan_relid = path->parent->relid;
4172 : : RangeTblEntry *rte;
4173 : :
4174 : : /* it should be a base rel with a tablesample clause... */
4175 [ - + ]: 24 : Assert(scan_relid > 0);
4176 [ + - ]: 24 : rte = planner_rt_fetch(scan_relid, root);
4177 [ - + ]: 24 : Assert(rte->rtekind == RTE_RELATION);
4178 [ - + ]: 24 : Assert(rte->tablesample != NULL);
4179 : :
26 tgl@sss.pgh.pa.us 4180 :GNC 24 : ADJUST_CHILD_ATTRS(rte->tablesample);
4181 : : }
2382 rhaas@postgresql.org 4182 :CBC 114 : break;
4183 : :
4184 : 222 : case T_IndexPath:
4185 : : {
26 tgl@sss.pgh.pa.us 4186 :GNC 222 : IndexPath *ipath = (IndexPath *) path;
4187 : :
4188 : 222 : ADJUST_CHILD_ATTRS(ipath->indexinfo->indrestrictinfo);
2382 rhaas@postgresql.org 4189 :CBC 222 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
4190 : 222 : new_path = (Path *) ipath;
4191 : : }
4192 : 222 : break;
4193 : :
4194 : 24 : case T_BitmapHeapPath:
4195 : : {
26 tgl@sss.pgh.pa.us 4196 :GNC 24 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4197 : :
4198 : 24 : ADJUST_CHILD_ATTRS(bhpath->path.parent->baserestrictinfo);
2382 rhaas@postgresql.org 4199 [ - + ]:CBC 24 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
4200 : 24 : new_path = (Path *) bhpath;
4201 : : }
4202 : 24 : break;
4203 : :
4204 : 12 : case T_BitmapAndPath:
4205 : : {
26 tgl@sss.pgh.pa.us 4206 :GNC 12 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4207 : :
2382 rhaas@postgresql.org 4208 [ + - - + ]:CBC 12 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
4209 : 12 : new_path = (Path *) bapath;
4210 : : }
4211 : 12 : break;
4212 : :
4213 : 12 : case T_BitmapOrPath:
4214 : : {
26 tgl@sss.pgh.pa.us 4215 :GNC 12 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4216 : :
2382 rhaas@postgresql.org 4217 [ + - - + ]:CBC 12 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
4218 : 12 : new_path = (Path *) bopath;
4219 : : }
4220 : 12 : break;
4221 : :
2382 rhaas@postgresql.org 4222 :LBC (26) : case T_ForeignPath:
4223 : : {
26 tgl@sss.pgh.pa.us 4224 :UNC 0 : ForeignPath *fpath = (ForeignPath *) path;
4225 : : ReparameterizeForeignPathByChild_function rfpc_func;
4226 : :
4227 : 0 : ADJUST_CHILD_ATTRS(fpath->path.parent->baserestrictinfo);
2382 rhaas@postgresql.org 4228 [ # # ]:LBC (26) : if (fpath->fdw_outerpath)
2382 rhaas@postgresql.org 4229 [ # # ]:UBC 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
243 efujita@postgresql.o 4230 [ # # ]:UNC 0 : if (fpath->fdw_restrictinfo)
4231 : 0 : ADJUST_CHILD_ATTRS(fpath->fdw_restrictinfo);
4232 : :
4233 : : /* Hand over to FDW if needed. */
2382 rhaas@postgresql.org 4234 :LBC (26) : rfpc_func =
4235 : (26) : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
4236 [ # # ]: (26) : if (rfpc_func)
2382 rhaas@postgresql.org 4237 :UBC 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
4238 : : child_rel);
2382 rhaas@postgresql.org 4239 :LBC (26) : new_path = (Path *) fpath;
4240 : : }
4241 : (26) : break;
4242 : :
2382 rhaas@postgresql.org 4243 :UBC 0 : case T_CustomPath:
4244 : : {
26 tgl@sss.pgh.pa.us 4245 :UNC 0 : CustomPath *cpath = (CustomPath *) path;
4246 : :
4247 : 0 : ADJUST_CHILD_ATTRS(cpath->path.parent->baserestrictinfo);
2382 rhaas@postgresql.org 4248 [ # # # # ]:UBC 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
243 efujita@postgresql.o 4249 [ # # ]:UNC 0 : if (cpath->custom_restrictinfo)
4250 : 0 : ADJUST_CHILD_ATTRS(cpath->custom_restrictinfo);
2382 rhaas@postgresql.org 4251 [ # # ]:UBC 0 : if (cpath->methods &&
4252 [ # # ]: 0 : cpath->methods->ReparameterizeCustomPathByChild)
4253 : 0 : cpath->custom_private =
4254 : 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4255 : : cpath->custom_private,
4256 : : child_rel);
4257 : 0 : new_path = (Path *) cpath;
4258 : : }
4259 : 0 : break;
4260 : :
2382 rhaas@postgresql.org 4261 :CBC 18 : case T_NestPath:
4262 : : {
26 tgl@sss.pgh.pa.us 4263 :GNC 18 : NestPath *npath = (NestPath *) path;
4264 : 18 : JoinPath *jpath = (JoinPath *) npath;
4265 : :
2382 rhaas@postgresql.org 4266 [ - + ]:CBC 18 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4267 [ - + ]: 18 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4268 : 18 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
980 peter@eisentraut.org 4269 : 18 : new_path = (Path *) npath;
4270 : : }
2382 rhaas@postgresql.org 4271 : 18 : break;
4272 : :
2382 rhaas@postgresql.org 4273 :LBC (18) : case T_MergePath:
4274 : : {
26 tgl@sss.pgh.pa.us 4275 :UNC 0 : MergePath *mpath = (MergePath *) path;
4276 : 0 : JoinPath *jpath = (JoinPath *) mpath;
4277 : :
2382 rhaas@postgresql.org 4278 [ # # ]:LBC (18) : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4279 [ # # ]: (18) : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4280 : (18) : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4281 : (18) : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4282 : (18) : new_path = (Path *) mpath;
4283 : : }
4284 : (18) : break;
4285 : :
2382 rhaas@postgresql.org 4286 :CBC 24 : case T_HashPath:
4287 : : {
26 tgl@sss.pgh.pa.us 4288 :GNC 24 : HashPath *hpath = (HashPath *) path;
4289 : 24 : JoinPath *jpath = (JoinPath *) hpath;
4290 : :
2382 rhaas@postgresql.org 4291 [ - + ]:CBC 24 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4292 [ - + ]: 24 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4293 : 24 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4294 : 24 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4295 : 24 : new_path = (Path *) hpath;
4296 : : }
4297 : 24 : break;
4298 : :
4299 : 12 : case T_AppendPath:
4300 : : {
26 tgl@sss.pgh.pa.us 4301 :GNC 12 : AppendPath *apath = (AppendPath *) path;
4302 : :
2382 rhaas@postgresql.org 4303 [ + - - + ]:CBC 12 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4304 : 12 : new_path = (Path *) apath;
4305 : : }
4306 : 12 : break;
4307 : :
497 tgl@sss.pgh.pa.us 4308 :UBC 0 : case T_MaterialPath:
4309 : : {
26 tgl@sss.pgh.pa.us 4310 :UNC 0 : MaterialPath *mpath = (MaterialPath *) path;
4311 : :
497 tgl@sss.pgh.pa.us 4312 [ # # ]:UBC 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4313 : 0 : new_path = (Path *) mpath;
4314 : : }
4315 : 0 : break;
4316 : :
1005 drowley@postgresql.o 4317 :CBC 6 : case T_MemoizePath:
4318 : : {
26 tgl@sss.pgh.pa.us 4319 :GNC 6 : MemoizePath *mpath = (MemoizePath *) path;
4320 : :
1005 drowley@postgresql.o 4321 [ - + ]:CBC 6 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
496 tgl@sss.pgh.pa.us 4322 : 6 : ADJUST_CHILD_ATTRS(mpath->param_exprs);
1005 drowley@postgresql.o 4323 : 6 : new_path = (Path *) mpath;
4324 : : }
1108 4325 : 6 : break;
4326 : :
2382 rhaas@postgresql.org 4327 :UBC 0 : case T_GatherPath:
4328 : : {
26 tgl@sss.pgh.pa.us 4329 :UNC 0 : GatherPath *gpath = (GatherPath *) path;
4330 : :
2382 rhaas@postgresql.org 4331 [ # # ]:UBC 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4332 : 0 : new_path = (Path *) gpath;
4333 : : }
4334 : 0 : break;
4335 : :
4336 : 0 : default:
4337 : : /* We don't know how to reparameterize this path. */
4338 : 0 : return NULL;
4339 : : }
4340 : :
4341 : : /*
4342 : : * Adjust the parameterization information, which refers to the topmost
4343 : : * parent. The topmost parent can be multiple levels away from the given
4344 : : * child, hence use multi-level expression adjustment routines.
4345 : : */
2382 rhaas@postgresql.org 4346 :CBC 444 : old_ppi = new_path->param_info;
4347 : : required_outer =
4348 : 444 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4349 : : child_rel,
605 tgl@sss.pgh.pa.us 4350 : 444 : child_rel->top_parent);
4351 : :
4352 : : /* If we already have a PPI for this parameterization, just return it */
2382 rhaas@postgresql.org 4353 : 444 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4354 : :
4355 : : /*
4356 : : * If not, build a new one and link it to the list of PPIs. For the same
4357 : : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4358 : : * context the given RelOptInfo is in.
4359 : : */
4360 [ + + ]: 444 : if (new_ppi == NULL)
4361 : : {
4362 : : MemoryContext oldcontext;
4363 : 390 : RelOptInfo *rel = path->parent;
4364 : :
4365 : 390 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4366 : :
4367 : 390 : new_ppi = makeNode(ParamPathInfo);
4368 : 390 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4369 : 390 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4370 : 390 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4371 : 390 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
440 tgl@sss.pgh.pa.us 4372 : 390 : new_ppi->ppi_serials = bms_copy(old_ppi->ppi_serials);
2382 rhaas@postgresql.org 4373 : 390 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4374 : :
4375 : 390 : MemoryContextSwitchTo(oldcontext);
4376 : : }
4377 : 444 : bms_free(required_outer);
4378 : :
4379 : 444 : new_path->param_info = new_ppi;
4380 : :
4381 : : /*
4382 : : * Adjust the path target if the parent of the outer relation is
4383 : : * referenced in the targetlist. This can happen when only the parent of
4384 : : * outer relation is laterally referenced in this relation.
4385 : : */
4386 [ + + ]: 444 : if (bms_overlap(path->parent->lateral_relids,
4387 : 444 : child_rel->top_parent_relids))
4388 : : {
4389 : 240 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4390 : 240 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4391 : : }
4392 : :
4393 : 444 : return new_path;
4394 : : }
4395 : :
4396 : : /*
4397 : : * path_is_reparameterizable_by_child
4398 : : * Given a path parameterized by the parent of the given child relation,
4399 : : * see if it can be translated to be parameterized by the child relation.
4400 : : *
4401 : : * This must return true if and only if reparameterize_path_by_child()
4402 : : * would succeed on this path. Currently it's sufficient to verify that
4403 : : * the path and all of its subpaths (if any) are of the types handled by
4404 : : * that function. However, subpaths that are not parameterized can be
4405 : : * disregarded since they won't require translation.
4406 : : */
4407 : : bool
26 tgl@sss.pgh.pa.us 4408 :GNC 15594 : path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
4409 : : {
4410 : : #define REJECT_IF_PATH_NOT_REPARAMETERIZABLE(path) \
4411 : : do { \
4412 : : if (!path_is_reparameterizable_by_child(path, child_rel)) \
4413 : : return false; \
4414 : : } while(0)
4415 : :
4416 : : #define REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(pathlist) \
4417 : : do { \
4418 : : if (!pathlist_is_reparameterizable_by_child(pathlist, child_rel)) \
4419 : : return false; \
4420 : : } while(0)
4421 : :
4422 : : /*
4423 : : * If the path is not parameterized by the parent of the given relation,
4424 : : * it doesn't need reparameterization.
4425 : : */
4426 [ + + ]: 15594 : if (!path->param_info ||
4427 [ + - + + ]: 15426 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4428 : 456 : return true;
4429 : :
4430 : : /*
4431 : : * Check that the path type is one that reparameterize_path_by_child() can
4432 : : * handle, and recursively check subpaths.
4433 : : */
4434 [ + + + + : 15138 : switch (nodeTag(path))
+ - + + -
+ - - ]
4435 : : {
4436 : 10104 : case T_Path:
4437 : : case T_IndexPath:
4438 : 10104 : break;
4439 : :
4440 : 24 : case T_BitmapHeapPath:
4441 : : {
4442 : 24 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4443 : :
4444 [ - + ]: 24 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(bhpath->bitmapqual);
4445 : : }
4446 : 24 : break;
4447 : :
4448 : 12 : case T_BitmapAndPath:
4449 : : {
4450 : 12 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4451 : :
4452 [ - + ]: 12 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bapath->bitmapquals);
4453 : : }
4454 : 12 : break;
4455 : :
4456 : 12 : case T_BitmapOrPath:
4457 : : {
4458 : 12 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4459 : :
4460 [ - + ]: 12 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bopath->bitmapquals);
4461 : : }
4462 : 12 : break;
4463 : :
4464 : 74 : case T_ForeignPath:
4465 : : {
4466 : 74 : ForeignPath *fpath = (ForeignPath *) path;
4467 : :
4468 [ - + ]: 74 : if (fpath->fdw_outerpath)
26 tgl@sss.pgh.pa.us 4469 [ # # ]:UNC 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(fpath->fdw_outerpath);
4470 : : }
26 tgl@sss.pgh.pa.us 4471 :GNC 74 : break;
4472 : :
26 tgl@sss.pgh.pa.us 4473 :UNC 0 : case T_CustomPath:
4474 : : {
4475 : 0 : CustomPath *cpath = (CustomPath *) path;
4476 : :
4477 [ # # ]: 0 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(cpath->custom_paths);
4478 : : }
4479 : 0 : break;
4480 : :
26 tgl@sss.pgh.pa.us 4481 :GNC 588 : case T_NestPath:
4482 : : case T_MergePath:
4483 : : case T_HashPath:
4484 : : {
4485 : 588 : JoinPath *jpath = (JoinPath *) path;
4486 : :
4487 [ - + ]: 588 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->outerjoinpath);
4488 [ - + ]: 588 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->innerjoinpath);
4489 : : }
4490 : 588 : break;
4491 : :
4492 : 96 : case T_AppendPath:
4493 : : {
4494 : 96 : AppendPath *apath = (AppendPath *) path;
4495 : :
4496 [ - + ]: 96 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(apath->subpaths);
4497 : : }
4498 : 96 : break;
4499 : :
26 tgl@sss.pgh.pa.us 4500 :UNC 0 : case T_MaterialPath:
4501 : : {
4502 : 0 : MaterialPath *mpath = (MaterialPath *) path;
4503 : :
4504 [ # # ]: 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4505 : : }
4506 : 0 : break;
4507 : :
26 tgl@sss.pgh.pa.us 4508 :GNC 4228 : case T_MemoizePath:
4509 : : {
4510 : 4228 : MemoizePath *mpath = (MemoizePath *) path;
4511 : :
4512 [ - + ]: 4228 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4513 : : }
4514 : 4228 : break;
4515 : :
26 tgl@sss.pgh.pa.us 4516 :UNC 0 : case T_GatherPath:
4517 : : {
4518 : 0 : GatherPath *gpath = (GatherPath *) path;
4519 : :
4520 [ # # ]: 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(gpath->subpath);
4521 : : }
4522 : 0 : break;
4523 : :
4524 : 0 : default:
4525 : : /* We don't know how to reparameterize this path. */
4526 : 0 : return false;
4527 : : }
4528 : :
26 tgl@sss.pgh.pa.us 4529 :GNC 15138 : return true;
4530 : : }
4531 : :
4532 : : /*
4533 : : * reparameterize_pathlist_by_child
4534 : : * Helper function to reparameterize a list of paths by given child rel.
4535 : : *
4536 : : * Returns NIL to indicate failure, so pathlist had better not be NIL.
4537 : : */
4538 : : static List *
2382 rhaas@postgresql.org 4539 :CBC 36 : reparameterize_pathlist_by_child(PlannerInfo *root,
4540 : : List *pathlist,
4541 : : RelOptInfo *child_rel)
4542 : : {
4543 : : ListCell *lc;
4544 : 36 : List *result = NIL;
4545 : :
4546 [ + - + + : 108 : foreach(lc, pathlist)
+ + ]
4547 : : {
4548 : 72 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4549 : : child_rel);
4550 : :
4551 [ - + ]: 72 : if (path == NULL)
4552 : : {
2382 rhaas@postgresql.org 4553 :LBC (18) : list_free(result);
4554 : (18) : return NIL;
4555 : : }
4556 : :
2382 rhaas@postgresql.org 4557 :CBC 72 : result = lappend(result, path);
4558 : : }
4559 : :
4560 : 36 : return result;
4561 : : }
4562 : :
4563 : : /*
4564 : : * pathlist_is_reparameterizable_by_child
4565 : : * Helper function to check if a list of paths can be reparameterized.
4566 : : */
4567 : : static bool
26 tgl@sss.pgh.pa.us 4568 :GNC 120 : pathlist_is_reparameterizable_by_child(List *pathlist, RelOptInfo *child_rel)
4569 : : {
4570 : : ListCell *lc;
4571 : :
4572 [ + - + + : 360 : foreach(lc, pathlist)
+ + ]
4573 : : {
4574 : 240 : Path *path = (Path *) lfirst(lc);
4575 : :
4576 [ - + ]: 240 : if (!path_is_reparameterizable_by_child(path, child_rel))
26 tgl@sss.pgh.pa.us 4577 :UNC 0 : return false;
4578 : : }
4579 : :
26 tgl@sss.pgh.pa.us 4580 :GNC 120 : return true;
4581 : : }
|
