Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * selfuncs.c
4 : * Selectivity functions and index cost estimation functions for
5 : * standard operators and index access methods.
6 : *
7 : * Selectivity routines are registered in the pg_operator catalog
8 : * in the "oprrest" and "oprjoin" attributes.
9 : *
10 : * Index cost functions are located via the index AM's API struct,
11 : * which is obtained from the handler function registered in pg_am.
12 : *
13 : * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
14 : * Portions Copyright (c) 1994, Regents of the University of California
15 : *
16 : *
17 : * IDENTIFICATION
18 : * src/backend/utils/adt/selfuncs.c
19 : *
20 : *-------------------------------------------------------------------------
21 : */
22 :
23 : /*----------
24 : * Operator selectivity estimation functions are called to estimate the
25 : * selectivity of WHERE clauses whose top-level operator is their operator.
26 : * We divide the problem into two cases:
27 : * Restriction clause estimation: the clause involves vars of just
28 : * one relation.
29 : * Join clause estimation: the clause involves vars of multiple rels.
30 : * Join selectivity estimation is far more difficult and usually less accurate
31 : * than restriction estimation.
32 : *
33 : * When dealing with the inner scan of a nestloop join, we consider the
34 : * join's joinclauses as restriction clauses for the inner relation, and
35 : * treat vars of the outer relation as parameters (a/k/a constants of unknown
36 : * values). So, restriction estimators need to be able to accept an argument
37 : * telling which relation is to be treated as the variable.
38 : *
39 : * The call convention for a restriction estimator (oprrest function) is
40 : *
41 : * Selectivity oprrest (PlannerInfo *root,
42 : * Oid operator,
43 : * List *args,
44 : * int varRelid);
45 : *
46 : * root: general information about the query (rtable and RelOptInfo lists
47 : * are particularly important for the estimator).
48 : * operator: OID of the specific operator in question.
49 : * args: argument list from the operator clause.
50 : * varRelid: if not zero, the relid (rtable index) of the relation to
51 : * be treated as the variable relation. May be zero if the args list
52 : * is known to contain vars of only one relation.
53 : *
54 : * This is represented at the SQL level (in pg_proc) as
55 : *
56 : * float8 oprrest (internal, oid, internal, int4);
57 : *
58 : * The result is a selectivity, that is, a fraction (0 to 1) of the rows
59 : * of the relation that are expected to produce a TRUE result for the
60 : * given operator.
61 : *
62 : * The call convention for a join estimator (oprjoin function) is similar
63 : * except that varRelid is not needed, and instead join information is
64 : * supplied:
65 : *
66 : * Selectivity oprjoin (PlannerInfo *root,
67 : * Oid operator,
68 : * List *args,
69 : * JoinType jointype,
70 : * SpecialJoinInfo *sjinfo);
71 : *
72 : * float8 oprjoin (internal, oid, internal, int2, internal);
73 : *
74 : * (Before Postgres 8.4, join estimators had only the first four of these
75 : * parameters. That signature is still allowed, but deprecated.) The
76 : * relationship between jointype and sjinfo is explained in the comments for
77 : * clause_selectivity() --- the short version is that jointype is usually
78 : * best ignored in favor of examining sjinfo.
79 : *
80 : * Join selectivity for regular inner and outer joins is defined as the
81 : * fraction (0 to 1) of the cross product of the relations that is expected
82 : * to produce a TRUE result for the given operator. For both semi and anti
83 : * joins, however, the selectivity is defined as the fraction of the left-hand
84 : * side relation's rows that are expected to have a match (ie, at least one
85 : * row with a TRUE result) in the right-hand side.
86 : *
87 : * For both oprrest and oprjoin functions, the operator's input collation OID
88 : * (if any) is passed using the standard fmgr mechanism, so that the estimator
89 : * function can fetch it with PG_GET_COLLATION(). Note, however, that all
90 : * statistics in pg_statistic are currently built using the relevant column's
91 : * collation.
92 : *----------
93 : */
94 :
95 : #include "postgres.h"
96 :
97 : #include <ctype.h>
98 : #include <math.h>
99 :
100 : #include "access/brin.h"
101 : #include "access/brin_page.h"
102 : #include "access/gin.h"
103 : #include "access/table.h"
104 : #include "access/tableam.h"
105 : #include "access/visibilitymap.h"
106 : #include "catalog/pg_am.h"
107 : #include "catalog/pg_collation.h"
108 : #include "catalog/pg_operator.h"
109 : #include "catalog/pg_statistic.h"
110 : #include "catalog/pg_statistic_ext.h"
111 : #include "executor/nodeAgg.h"
112 : #include "miscadmin.h"
113 : #include "nodes/makefuncs.h"
114 : #include "nodes/nodeFuncs.h"
115 : #include "optimizer/clauses.h"
116 : #include "optimizer/cost.h"
117 : #include "optimizer/optimizer.h"
118 : #include "optimizer/pathnode.h"
119 : #include "optimizer/paths.h"
120 : #include "optimizer/plancat.h"
121 : #include "parser/parse_clause.h"
122 : #include "parser/parsetree.h"
123 : #include "statistics/statistics.h"
124 : #include "storage/bufmgr.h"
125 : #include "utils/acl.h"
126 : #include "utils/builtins.h"
127 : #include "utils/date.h"
128 : #include "utils/datum.h"
129 : #include "utils/fmgroids.h"
130 : #include "utils/index_selfuncs.h"
131 : #include "utils/lsyscache.h"
132 : #include "utils/memutils.h"
133 : #include "utils/pg_locale.h"
134 : #include "utils/rel.h"
135 : #include "utils/selfuncs.h"
136 : #include "utils/snapmgr.h"
137 : #include "utils/spccache.h"
138 : #include "utils/syscache.h"
139 : #include "utils/timestamp.h"
140 : #include "utils/typcache.h"
141 :
142 :
143 : /* Hooks for plugins to get control when we ask for stats */
144 : get_relation_stats_hook_type get_relation_stats_hook = NULL;
145 : get_index_stats_hook_type get_index_stats_hook = NULL;
146 :
147 : static double eqsel_internal(PG_FUNCTION_ARGS, bool negate);
148 : static double eqjoinsel_inner(Oid opfuncoid, Oid collation,
149 : VariableStatData *vardata1, VariableStatData *vardata2,
150 : double nd1, double nd2,
151 : bool isdefault1, bool isdefault2,
152 : AttStatsSlot *sslot1, AttStatsSlot *sslot2,
153 : Form_pg_statistic stats1, Form_pg_statistic stats2,
154 : bool have_mcvs1, bool have_mcvs2);
155 : static double eqjoinsel_semi(Oid opfuncoid, Oid collation,
156 : VariableStatData *vardata1, VariableStatData *vardata2,
157 : double nd1, double nd2,
158 : bool isdefault1, bool isdefault2,
159 : AttStatsSlot *sslot1, AttStatsSlot *sslot2,
160 : Form_pg_statistic stats1, Form_pg_statistic stats2,
161 : bool have_mcvs1, bool have_mcvs2,
162 : RelOptInfo *inner_rel);
163 : static bool estimate_multivariate_ndistinct(PlannerInfo *root,
164 : RelOptInfo *rel, List **varinfos, double *ndistinct);
165 : static bool convert_to_scalar(Datum value, Oid valuetypid, Oid collid,
166 : double *scaledvalue,
167 : Datum lobound, Datum hibound, Oid boundstypid,
168 : double *scaledlobound, double *scaledhibound);
169 : static double convert_numeric_to_scalar(Datum value, Oid typid, bool *failure);
170 : static void convert_string_to_scalar(char *value,
171 : double *scaledvalue,
172 : char *lobound,
173 : double *scaledlobound,
174 : char *hibound,
175 : double *scaledhibound);
176 : static void convert_bytea_to_scalar(Datum value,
177 : double *scaledvalue,
178 : Datum lobound,
179 : double *scaledlobound,
180 : Datum hibound,
181 : double *scaledhibound);
182 : static double convert_one_string_to_scalar(char *value,
183 : int rangelo, int rangehi);
184 : static double convert_one_bytea_to_scalar(unsigned char *value, int valuelen,
185 : int rangelo, int rangehi);
186 : static char *convert_string_datum(Datum value, Oid typid, Oid collid,
187 : bool *failure);
188 : static double convert_timevalue_to_scalar(Datum value, Oid typid,
189 : bool *failure);
190 : static void examine_simple_variable(PlannerInfo *root, Var *var,
191 : VariableStatData *vardata);
192 : static bool get_variable_range(PlannerInfo *root, VariableStatData *vardata,
193 : Oid sortop, Oid collation,
194 : Datum *min, Datum *max);
195 : static void get_stats_slot_range(AttStatsSlot *sslot,
196 : Oid opfuncoid, FmgrInfo *opproc,
197 : Oid collation, int16 typLen, bool typByVal,
198 : Datum *min, Datum *max, bool *p_have_data);
199 : static bool get_actual_variable_range(PlannerInfo *root,
200 : VariableStatData *vardata,
201 : Oid sortop, Oid collation,
202 : Datum *min, Datum *max);
203 : static bool get_actual_variable_endpoint(Relation heapRel,
204 : Relation indexRel,
205 : ScanDirection indexscandir,
206 : ScanKey scankeys,
207 : int16 typLen,
208 : bool typByVal,
209 : TupleTableSlot *tableslot,
210 : MemoryContext outercontext,
211 : Datum *endpointDatum);
212 : static RelOptInfo *find_join_input_rel(PlannerInfo *root, Relids relids);
213 :
214 :
215 : /*
216 : * eqsel - Selectivity of "=" for any data types.
217 : *
218 : * Note: this routine is also used to estimate selectivity for some
219 : * operators that are not "=" but have comparable selectivity behavior,
220 : * such as "~=" (geometric approximate-match). Even for "=", we must
221 : * keep in mind that the left and right datatypes may differ.
222 : */
223 : Datum
224 428014 : eqsel(PG_FUNCTION_ARGS)
225 : {
226 428014 : PG_RETURN_FLOAT8((float8) eqsel_internal(fcinfo, false));
227 : }
228 :
229 : /*
230 : * Common code for eqsel() and neqsel()
231 : */
232 : static double
233 459344 : eqsel_internal(PG_FUNCTION_ARGS, bool negate)
234 : {
235 459344 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
236 459344 : Oid operator = PG_GETARG_OID(1);
237 459344 : List *args = (List *) PG_GETARG_POINTER(2);
238 459344 : int varRelid = PG_GETARG_INT32(3);
239 459344 : Oid collation = PG_GET_COLLATION();
240 : VariableStatData vardata;
241 : Node *other;
242 : bool varonleft;
243 : double selec;
244 :
245 : /*
246 : * When asked about <>, we do the estimation using the corresponding =
247 : * operator, then convert to <> via "1.0 - eq_selectivity - nullfrac".
248 : */
249 459344 : if (negate)
250 : {
251 31330 : operator = get_negator(operator);
252 31330 : if (!OidIsValid(operator))
253 : {
254 : /* Use default selectivity (should we raise an error instead?) */
255 0 : return 1.0 - DEFAULT_EQ_SEL;
256 : }
257 : }
258 :
259 : /*
260 : * If expression is not variable = something or something = variable, then
261 : * punt and return a default estimate.
262 : */
263 459344 : if (!get_restriction_variable(root, args, varRelid,
264 : &vardata, &other, &varonleft))
265 2534 : return negate ? (1.0 - DEFAULT_EQ_SEL) : DEFAULT_EQ_SEL;
266 :
267 : /*
268 : * We can do a lot better if the something is a constant. (Note: the
269 : * Const might result from estimation rather than being a simple constant
270 : * in the query.)
271 : */
272 456810 : if (IsA(other, Const))
273 205740 : selec = var_eq_const(&vardata, operator, collation,
274 205740 : ((Const *) other)->constvalue,
275 205740 : ((Const *) other)->constisnull,
276 : varonleft, negate);
277 : else
278 251070 : selec = var_eq_non_const(&vardata, operator, collation, other,
279 : varonleft, negate);
280 :
281 456810 : ReleaseVariableStats(vardata);
282 :
283 456810 : return selec;
284 : }
285 :
286 : /*
287 : * var_eq_const --- eqsel for var = const case
288 : *
289 : * This is exported so that some other estimation functions can use it.
290 : */
291 : double
292 239390 : var_eq_const(VariableStatData *vardata, Oid operator, Oid collation,
293 : Datum constval, bool constisnull,
294 : bool varonleft, bool negate)
295 : {
296 : double selec;
297 239390 : double nullfrac = 0.0;
298 : bool isdefault;
299 : Oid opfuncoid;
300 :
301 : /*
302 : * If the constant is NULL, assume operator is strict and return zero, ie,
303 : * operator will never return TRUE. (It's zero even for a negator op.)
304 : */
305 239390 : if (constisnull)
306 266 : return 0.0;
307 :
308 : /*
309 : * Grab the nullfrac for use below. Note we allow use of nullfrac
310 : * regardless of security check.
311 : */
312 239124 : if (HeapTupleIsValid(vardata->statsTuple))
313 : {
314 : Form_pg_statistic stats;
315 :
316 161642 : stats = (Form_pg_statistic) GETSTRUCT(vardata->statsTuple);
317 161642 : nullfrac = stats->stanullfrac;
318 : }
319 :
320 : /*
321 : * If we matched the var to a unique index or DISTINCT clause, assume
322 : * there is exactly one match regardless of anything else. (This is
323 : * slightly bogus, since the index or clause's equality operator might be
324 : * different from ours, but it's much more likely to be right than
325 : * ignoring the information.)
326 : */
327 239124 : if (vardata->isunique && vardata->rel && vardata->rel->tuples >= 1.0)
328 : {
329 60882 : selec = 1.0 / vardata->rel->tuples;
330 : }
331 295314 : else if (HeapTupleIsValid(vardata->statsTuple) &&
332 117072 : statistic_proc_security_check(vardata,
333 117072 : (opfuncoid = get_opcode(operator))))
334 117072 : {
335 : AttStatsSlot sslot;
336 117072 : bool match = false;
337 : int i;
338 :
339 : /*
340 : * Is the constant "=" to any of the column's most common values?
341 : * (Although the given operator may not really be "=", we will assume
342 : * that seeing whether it returns TRUE is an appropriate test. If you
343 : * don't like this, maybe you shouldn't be using eqsel for your
344 : * operator...)
345 : */
346 117072 : if (get_attstatsslot(&sslot, vardata->statsTuple,
347 : STATISTIC_KIND_MCV, InvalidOid,
348 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS))
349 : {
350 103794 : LOCAL_FCINFO(fcinfo, 2);
351 : FmgrInfo eqproc;
352 :
353 103794 : fmgr_info(opfuncoid, &eqproc);
354 :
355 : /*
356 : * Save a few cycles by setting up the fcinfo struct just once.
357 : * Using FunctionCallInvoke directly also avoids failure if the
358 : * eqproc returns NULL, though really equality functions should
359 : * never do that.
360 : */
361 103794 : InitFunctionCallInfoData(*fcinfo, &eqproc, 2, collation,
362 : NULL, NULL);
363 103794 : fcinfo->args[0].isnull = false;
364 103794 : fcinfo->args[1].isnull = false;
365 : /* be careful to apply operator right way 'round */
366 103794 : if (varonleft)
367 103768 : fcinfo->args[1].value = constval;
368 : else
369 26 : fcinfo->args[0].value = constval;
370 :
371 1352570 : for (i = 0; i < sslot.nvalues; i++)
372 : {
373 : Datum fresult;
374 :
375 1302860 : if (varonleft)
376 1302810 : fcinfo->args[0].value = sslot.values[i];
377 : else
378 50 : fcinfo->args[1].value = sslot.values[i];
379 1302860 : fcinfo->isnull = false;
380 1302860 : fresult = FunctionCallInvoke(fcinfo);
381 1302860 : if (!fcinfo->isnull && DatumGetBool(fresult))
382 : {
383 54084 : match = true;
384 54084 : break;
385 : }
386 : }
387 : }
388 : else
389 : {
390 : /* no most-common-value info available */
391 13278 : i = 0; /* keep compiler quiet */
392 : }
393 :
394 117072 : if (match)
395 : {
396 : /*
397 : * Constant is "=" to this common value. We know selectivity
398 : * exactly (or as exactly as ANALYZE could calculate it, anyway).
399 : */
400 54084 : selec = sslot.numbers[i];
401 : }
402 : else
403 : {
404 : /*
405 : * Comparison is against a constant that is neither NULL nor any
406 : * of the common values. Its selectivity cannot be more than
407 : * this:
408 : */
409 62988 : double sumcommon = 0.0;
410 : double otherdistinct;
411 :
412 1200040 : for (i = 0; i < sslot.nnumbers; i++)
413 1137052 : sumcommon += sslot.numbers[i];
414 62988 : selec = 1.0 - sumcommon - nullfrac;
415 62988 : CLAMP_PROBABILITY(selec);
416 :
417 : /*
418 : * and in fact it's probably a good deal less. We approximate that
419 : * all the not-common values share this remaining fraction
420 : * equally, so we divide by the number of other distinct values.
421 : */
422 62988 : otherdistinct = get_variable_numdistinct(vardata, &isdefault) -
423 62988 : sslot.nnumbers;
424 62988 : if (otherdistinct > 1)
425 29406 : selec /= otherdistinct;
426 :
427 : /*
428 : * Another cross-check: selectivity shouldn't be estimated as more
429 : * than the least common "most common value".
430 : */
431 62988 : if (sslot.nnumbers > 0 && selec > sslot.numbers[sslot.nnumbers - 1])
432 0 : selec = sslot.numbers[sslot.nnumbers - 1];
433 : }
434 :
435 117072 : free_attstatsslot(&sslot);
436 : }
437 : else
438 : {
439 : /*
440 : * No ANALYZE stats available, so make a guess using estimated number
441 : * of distinct values and assuming they are equally common. (The guess
442 : * is unlikely to be very good, but we do know a few special cases.)
443 : */
444 61170 : selec = 1.0 / get_variable_numdistinct(vardata, &isdefault);
445 : }
446 :
447 : /* now adjust if we wanted <> rather than = */
448 239124 : if (negate)
449 25638 : selec = 1.0 - selec - nullfrac;
450 :
451 : /* result should be in range, but make sure... */
452 239124 : CLAMP_PROBABILITY(selec);
453 :
454 239124 : return selec;
455 : }
456 :
457 : /*
458 : * var_eq_non_const --- eqsel for var = something-other-than-const case
459 : *
460 : * This is exported so that some other estimation functions can use it.
461 : */
462 : double
463 251070 : var_eq_non_const(VariableStatData *vardata, Oid operator, Oid collation,
464 : Node *other,
465 : bool varonleft, bool negate)
466 : {
467 : double selec;
468 251070 : double nullfrac = 0.0;
469 : bool isdefault;
470 :
471 : /*
472 : * Grab the nullfrac for use below.
473 : */
474 251070 : if (HeapTupleIsValid(vardata->statsTuple))
475 : {
476 : Form_pg_statistic stats;
477 :
478 169926 : stats = (Form_pg_statistic) GETSTRUCT(vardata->statsTuple);
479 169926 : nullfrac = stats->stanullfrac;
480 : }
481 :
482 : /*
483 : * If we matched the var to a unique index or DISTINCT clause, assume
484 : * there is exactly one match regardless of anything else. (This is
485 : * slightly bogus, since the index or clause's equality operator might be
486 : * different from ours, but it's much more likely to be right than
487 : * ignoring the information.)
488 : */
489 251070 : if (vardata->isunique && vardata->rel && vardata->rel->tuples >= 1.0)
490 : {
491 97482 : selec = 1.0 / vardata->rel->tuples;
492 : }
493 153588 : else if (HeapTupleIsValid(vardata->statsTuple))
494 : {
495 : double ndistinct;
496 : AttStatsSlot sslot;
497 :
498 : /*
499 : * Search is for a value that we do not know a priori, but we will
500 : * assume it is not NULL. Estimate the selectivity as non-null
501 : * fraction divided by number of distinct values, so that we get a
502 : * result averaged over all possible values whether common or
503 : * uncommon. (Essentially, we are assuming that the not-yet-known
504 : * comparison value is equally likely to be any of the possible
505 : * values, regardless of their frequency in the table. Is that a good
506 : * idea?)
507 : */
508 86512 : selec = 1.0 - nullfrac;
509 86512 : ndistinct = get_variable_numdistinct(vardata, &isdefault);
510 86512 : if (ndistinct > 1)
511 83276 : selec /= ndistinct;
512 :
513 : /*
514 : * Cross-check: selectivity should never be estimated as more than the
515 : * most common value's.
516 : */
517 86512 : if (get_attstatsslot(&sslot, vardata->statsTuple,
518 : STATISTIC_KIND_MCV, InvalidOid,
519 : ATTSTATSSLOT_NUMBERS))
520 : {
521 70358 : if (sslot.nnumbers > 0 && selec > sslot.numbers[0])
522 354 : selec = sslot.numbers[0];
523 70358 : free_attstatsslot(&sslot);
524 : }
525 : }
526 : else
527 : {
528 : /*
529 : * No ANALYZE stats available, so make a guess using estimated number
530 : * of distinct values and assuming they are equally common. (The guess
531 : * is unlikely to be very good, but we do know a few special cases.)
532 : */
533 67076 : selec = 1.0 / get_variable_numdistinct(vardata, &isdefault);
534 : }
535 :
536 : /* now adjust if we wanted <> rather than = */
537 251070 : if (negate)
538 4442 : selec = 1.0 - selec - nullfrac;
539 :
540 : /* result should be in range, but make sure... */
541 251070 : CLAMP_PROBABILITY(selec);
542 :
543 251070 : return selec;
544 : }
545 :
546 : /*
547 : * neqsel - Selectivity of "!=" for any data types.
548 : *
549 : * This routine is also used for some operators that are not "!="
550 : * but have comparable selectivity behavior. See above comments
551 : * for eqsel().
552 : */
553 : Datum
554 31330 : neqsel(PG_FUNCTION_ARGS)
555 : {
556 31330 : PG_RETURN_FLOAT8((float8) eqsel_internal(fcinfo, true));
557 : }
558 :
559 : /*
560 : * scalarineqsel - Selectivity of "<", "<=", ">", ">=" for scalars.
561 : *
562 : * This is the guts of scalarltsel/scalarlesel/scalargtsel/scalargesel.
563 : * The isgt and iseq flags distinguish which of the four cases apply.
564 : *
565 : * The caller has commuted the clause, if necessary, so that we can treat
566 : * the variable as being on the left. The caller must also make sure that
567 : * the other side of the clause is a non-null Const, and dissect that into
568 : * a value and datatype. (This definition simplifies some callers that
569 : * want to estimate against a computed value instead of a Const node.)
570 : *
571 : * This routine works for any datatype (or pair of datatypes) known to
572 : * convert_to_scalar(). If it is applied to some other datatype,
573 : * it will return an approximate estimate based on assuming that the constant
574 : * value falls in the middle of the bin identified by binary search.
575 : */
576 : static double
577 217698 : scalarineqsel(PlannerInfo *root, Oid operator, bool isgt, bool iseq,
578 : Oid collation,
579 : VariableStatData *vardata, Datum constval, Oid consttype)
580 : {
581 : Form_pg_statistic stats;
582 : FmgrInfo opproc;
583 : double mcv_selec,
584 : hist_selec,
585 : sumcommon;
586 : double selec;
587 :
588 217698 : if (!HeapTupleIsValid(vardata->statsTuple))
589 : {
590 : /*
591 : * No stats are available. Typically this means we have to fall back
592 : * on the default estimate; but if the variable is CTID then we can
593 : * make an estimate based on comparing the constant to the table size.
594 : */
595 22588 : if (vardata->var && IsA(vardata->var, Var) &&
596 14136 : ((Var *) vardata->var)->varattno == SelfItemPointerAttributeNumber)
597 : {
598 : ItemPointer itemptr;
599 : double block;
600 : double density;
601 :
602 : /*
603 : * If the relation's empty, we're going to include all of it.
604 : * (This is mostly to avoid divide-by-zero below.)
605 : */
606 214 : if (vardata->rel->pages == 0)
607 0 : return 1.0;
608 :
609 214 : itemptr = (ItemPointer) DatumGetPointer(constval);
610 214 : block = ItemPointerGetBlockNumberNoCheck(itemptr);
611 :
612 : /*
613 : * Determine the average number of tuples per page (density).
614 : *
615 : * Since the last page will, on average, be only half full, we can
616 : * estimate it to have half as many tuples as earlier pages. So
617 : * give it half the weight of a regular page.
618 : */
619 214 : density = vardata->rel->tuples / (vardata->rel->pages - 0.5);
620 :
621 : /* If target is the last page, use half the density. */
622 214 : if (block >= vardata->rel->pages - 1)
623 30 : density *= 0.5;
624 :
625 : /*
626 : * Using the average tuples per page, calculate how far into the
627 : * page the itemptr is likely to be and adjust block accordingly,
628 : * by adding that fraction of a whole block (but never more than a
629 : * whole block, no matter how high the itemptr's offset is). Here
630 : * we are ignoring the possibility of dead-tuple line pointers,
631 : * which is fairly bogus, but we lack the info to do better.
632 : */
633 214 : if (density > 0.0)
634 : {
635 214 : OffsetNumber offset = ItemPointerGetOffsetNumberNoCheck(itemptr);
636 :
637 214 : block += Min(offset / density, 1.0);
638 : }
639 :
640 : /*
641 : * Convert relative block number to selectivity. Again, the last
642 : * page has only half weight.
643 : */
644 214 : selec = block / (vardata->rel->pages - 0.5);
645 :
646 : /*
647 : * The calculation so far gave us a selectivity for the "<=" case.
648 : * We'll have one fewer tuple for "<" and one additional tuple for
649 : * ">=", the latter of which we'll reverse the selectivity for
650 : * below, so we can simply subtract one tuple for both cases. The
651 : * cases that need this adjustment can be identified by iseq being
652 : * equal to isgt.
653 : */
654 214 : if (iseq == isgt && vardata->rel->tuples >= 1.0)
655 102 : selec -= (1.0 / vardata->rel->tuples);
656 :
657 : /* Finally, reverse the selectivity for the ">", ">=" cases. */
658 214 : if (isgt)
659 100 : selec = 1.0 - selec;
660 :
661 214 : CLAMP_PROBABILITY(selec);
662 214 : return selec;
663 : }
664 :
665 : /* no stats available, so default result */
666 22374 : return DEFAULT_INEQ_SEL;
667 : }
668 195110 : stats = (Form_pg_statistic) GETSTRUCT(vardata->statsTuple);
669 :
670 195110 : fmgr_info(get_opcode(operator), &opproc);
671 :
672 : /*
673 : * If we have most-common-values info, add up the fractions of the MCV
674 : * entries that satisfy MCV OP CONST. These fractions contribute directly
675 : * to the result selectivity. Also add up the total fraction represented
676 : * by MCV entries.
677 : */
678 195110 : mcv_selec = mcv_selectivity(vardata, &opproc, collation, constval, true,
679 : &sumcommon);
680 :
681 : /*
682 : * If there is a histogram, determine which bin the constant falls in, and
683 : * compute the resulting contribution to selectivity.
684 : */
685 195110 : hist_selec = ineq_histogram_selectivity(root, vardata,
686 : operator, &opproc, isgt, iseq,
687 : collation,
688 : constval, consttype);
689 :
690 : /*
691 : * Now merge the results from the MCV and histogram calculations,
692 : * realizing that the histogram covers only the non-null values that are
693 : * not listed in MCV.
694 : */
695 195110 : selec = 1.0 - stats->stanullfrac - sumcommon;
696 :
697 195110 : if (hist_selec >= 0.0)
698 155596 : selec *= hist_selec;
699 : else
700 : {
701 : /*
702 : * If no histogram but there are values not accounted for by MCV,
703 : * arbitrarily assume half of them will match.
704 : */
705 39514 : selec *= 0.5;
706 : }
707 :
708 195110 : selec += mcv_selec;
709 :
710 : /* result should be in range, but make sure... */
711 195110 : CLAMP_PROBABILITY(selec);
712 :
713 195110 : return selec;
714 : }
715 :
716 : /*
717 : * mcv_selectivity - Examine the MCV list for selectivity estimates
718 : *
719 : * Determine the fraction of the variable's MCV population that satisfies
720 : * the predicate (VAR OP CONST), or (CONST OP VAR) if !varonleft. Also
721 : * compute the fraction of the total column population represented by the MCV
722 : * list. This code will work for any boolean-returning predicate operator.
723 : *
724 : * The function result is the MCV selectivity, and the fraction of the
725 : * total population is returned into *sumcommonp. Zeroes are returned
726 : * if there is no MCV list.
727 : */
728 : double
729 199580 : mcv_selectivity(VariableStatData *vardata, FmgrInfo *opproc, Oid collation,
730 : Datum constval, bool varonleft,
731 : double *sumcommonp)
732 : {
733 : double mcv_selec,
734 : sumcommon;
735 : AttStatsSlot sslot;
736 : int i;
737 :
738 199580 : mcv_selec = 0.0;
739 199580 : sumcommon = 0.0;
740 :
741 396658 : if (HeapTupleIsValid(vardata->statsTuple) &&
742 394072 : statistic_proc_security_check(vardata, opproc->fn_oid) &&
743 196994 : get_attstatsslot(&sslot, vardata->statsTuple,
744 : STATISTIC_KIND_MCV, InvalidOid,
745 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS))
746 : {
747 91434 : LOCAL_FCINFO(fcinfo, 2);
748 :
749 : /*
750 : * We invoke the opproc "by hand" so that we won't fail on NULL
751 : * results. Such cases won't arise for normal comparison functions,
752 : * but generic_restriction_selectivity could perhaps be used with
753 : * operators that can return NULL. A small side benefit is to not
754 : * need to re-initialize the fcinfo struct from scratch each time.
755 : */
756 91434 : InitFunctionCallInfoData(*fcinfo, opproc, 2, collation,
757 : NULL, NULL);
758 91434 : fcinfo->args[0].isnull = false;
759 91434 : fcinfo->args[1].isnull = false;
760 : /* be careful to apply operator right way 'round */
761 91434 : if (varonleft)
762 91434 : fcinfo->args[1].value = constval;
763 : else
764 0 : fcinfo->args[0].value = constval;
765 :
766 2656600 : for (i = 0; i < sslot.nvalues; i++)
767 : {
768 : Datum fresult;
769 :
770 2565166 : if (varonleft)
771 2565166 : fcinfo->args[0].value = sslot.values[i];
772 : else
773 0 : fcinfo->args[1].value = sslot.values[i];
774 2565166 : fcinfo->isnull = false;
775 2565166 : fresult = FunctionCallInvoke(fcinfo);
776 2565166 : if (!fcinfo->isnull && DatumGetBool(fresult))
777 1195372 : mcv_selec += sslot.numbers[i];
778 2565166 : sumcommon += sslot.numbers[i];
779 : }
780 91434 : free_attstatsslot(&sslot);
781 : }
782 :
783 199580 : *sumcommonp = sumcommon;
784 199580 : return mcv_selec;
785 : }
786 :
787 : /*
788 : * histogram_selectivity - Examine the histogram for selectivity estimates
789 : *
790 : * Determine the fraction of the variable's histogram entries that satisfy
791 : * the predicate (VAR OP CONST), or (CONST OP VAR) if !varonleft.
792 : *
793 : * This code will work for any boolean-returning predicate operator, whether
794 : * or not it has anything to do with the histogram sort operator. We are
795 : * essentially using the histogram just as a representative sample. However,
796 : * small histograms are unlikely to be all that representative, so the caller
797 : * should be prepared to fall back on some other estimation approach when the
798 : * histogram is missing or very small. It may also be prudent to combine this
799 : * approach with another one when the histogram is small.
800 : *
801 : * If the actual histogram size is not at least min_hist_size, we won't bother
802 : * to do the calculation at all. Also, if the n_skip parameter is > 0, we
803 : * ignore the first and last n_skip histogram elements, on the grounds that
804 : * they are outliers and hence not very representative. Typical values for
805 : * these parameters are 10 and 1.
806 : *
807 : * The function result is the selectivity, or -1 if there is no histogram
808 : * or it's smaller than min_hist_size.
809 : *
810 : * The output parameter *hist_size receives the actual histogram size,
811 : * or zero if no histogram. Callers may use this number to decide how
812 : * much faith to put in the function result.
813 : *
814 : * Note that the result disregards both the most-common-values (if any) and
815 : * null entries. The caller is expected to combine this result with
816 : * statistics for those portions of the column population. It may also be
817 : * prudent to clamp the result range, ie, disbelieve exact 0 or 1 outputs.
818 : */
819 : double
820 4470 : histogram_selectivity(VariableStatData *vardata,
821 : FmgrInfo *opproc, Oid collation,
822 : Datum constval, bool varonleft,
823 : int min_hist_size, int n_skip,
824 : int *hist_size)
825 : {
826 : double result;
827 : AttStatsSlot sslot;
828 :
829 : /* check sanity of parameters */
830 : Assert(n_skip >= 0);
831 : Assert(min_hist_size > 2 * n_skip);
832 :
833 6438 : if (HeapTupleIsValid(vardata->statsTuple) &&
834 3930 : statistic_proc_security_check(vardata, opproc->fn_oid) &&
835 1962 : get_attstatsslot(&sslot, vardata->statsTuple,
836 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
837 : ATTSTATSSLOT_VALUES))
838 : {
839 1868 : *hist_size = sslot.nvalues;
840 1868 : if (sslot.nvalues >= min_hist_size)
841 : {
842 1364 : LOCAL_FCINFO(fcinfo, 2);
843 1364 : int nmatch = 0;
844 : int i;
845 :
846 : /*
847 : * We invoke the opproc "by hand" so that we won't fail on NULL
848 : * results. Such cases won't arise for normal comparison
849 : * functions, but generic_restriction_selectivity could perhaps be
850 : * used with operators that can return NULL. A small side benefit
851 : * is to not need to re-initialize the fcinfo struct from scratch
852 : * each time.
853 : */
854 1364 : InitFunctionCallInfoData(*fcinfo, opproc, 2, collation,
855 : NULL, NULL);
856 1364 : fcinfo->args[0].isnull = false;
857 1364 : fcinfo->args[1].isnull = false;
858 : /* be careful to apply operator right way 'round */
859 1364 : if (varonleft)
860 1364 : fcinfo->args[1].value = constval;
861 : else
862 0 : fcinfo->args[0].value = constval;
863 :
864 120656 : for (i = n_skip; i < sslot.nvalues - n_skip; i++)
865 : {
866 : Datum fresult;
867 :
868 119292 : if (varonleft)
869 119292 : fcinfo->args[0].value = sslot.values[i];
870 : else
871 0 : fcinfo->args[1].value = sslot.values[i];
872 119292 : fcinfo->isnull = false;
873 119292 : fresult = FunctionCallInvoke(fcinfo);
874 119292 : if (!fcinfo->isnull && DatumGetBool(fresult))
875 4450 : nmatch++;
876 : }
877 1364 : result = ((double) nmatch) / ((double) (sslot.nvalues - 2 * n_skip));
878 : }
879 : else
880 504 : result = -1;
881 1868 : free_attstatsslot(&sslot);
882 : }
883 : else
884 : {
885 2602 : *hist_size = 0;
886 2602 : result = -1;
887 : }
888 :
889 4470 : return result;
890 : }
891 :
892 : /*
893 : * generic_restriction_selectivity - Selectivity for almost anything
894 : *
895 : * This function estimates selectivity for operators that we don't have any
896 : * special knowledge about, but are on data types that we collect standard
897 : * MCV and/or histogram statistics for. (Additional assumptions are that
898 : * the operator is strict and immutable, or at least stable.)
899 : *
900 : * If we have "VAR OP CONST" or "CONST OP VAR", selectivity is estimated by
901 : * applying the operator to each element of the column's MCV and/or histogram
902 : * stats, and merging the results using the assumption that the histogram is
903 : * a reasonable random sample of the column's non-MCV population. Note that
904 : * if the operator's semantics are related to the histogram ordering, this
905 : * might not be such a great assumption; other functions such as
906 : * scalarineqsel() are probably a better match in such cases.
907 : *
908 : * Otherwise, fall back to the default selectivity provided by the caller.
909 : */
910 : double
911 1106 : generic_restriction_selectivity(PlannerInfo *root, Oid oproid, Oid collation,
912 : List *args, int varRelid,
913 : double default_selectivity)
914 : {
915 : double selec;
916 : VariableStatData vardata;
917 : Node *other;
918 : bool varonleft;
919 :
920 : /*
921 : * If expression is not variable OP something or something OP variable,
922 : * then punt and return the default estimate.
923 : */
924 1106 : if (!get_restriction_variable(root, args, varRelid,
925 : &vardata, &other, &varonleft))
926 0 : return default_selectivity;
927 :
928 : /*
929 : * If the something is a NULL constant, assume operator is strict and
930 : * return zero, ie, operator will never return TRUE.
931 : */
932 1106 : if (IsA(other, Const) &&
933 1106 : ((Const *) other)->constisnull)
934 : {
935 0 : ReleaseVariableStats(vardata);
936 0 : return 0.0;
937 : }
938 :
939 1106 : if (IsA(other, Const))
940 : {
941 : /* Variable is being compared to a known non-null constant */
942 1106 : Datum constval = ((Const *) other)->constvalue;
943 : FmgrInfo opproc;
944 : double mcvsum;
945 : double mcvsel;
946 : double nullfrac;
947 : int hist_size;
948 :
949 1106 : fmgr_info(get_opcode(oproid), &opproc);
950 :
951 : /*
952 : * Calculate the selectivity for the column's most common values.
953 : */
954 1106 : mcvsel = mcv_selectivity(&vardata, &opproc, collation,
955 : constval, varonleft,
956 : &mcvsum);
957 :
958 : /*
959 : * If the histogram is large enough, see what fraction of it matches
960 : * the query, and assume that's representative of the non-MCV
961 : * population. Otherwise use the default selectivity for the non-MCV
962 : * population.
963 : */
964 1106 : selec = histogram_selectivity(&vardata, &opproc, collation,
965 : constval, varonleft,
966 : 10, 1, &hist_size);
967 1106 : if (selec < 0)
968 : {
969 : /* Nope, fall back on default */
970 1106 : selec = default_selectivity;
971 : }
972 0 : else if (hist_size < 100)
973 : {
974 : /*
975 : * For histogram sizes from 10 to 100, we combine the histogram
976 : * and default selectivities, putting increasingly more trust in
977 : * the histogram for larger sizes.
978 : */
979 0 : double hist_weight = hist_size / 100.0;
980 :
981 0 : selec = selec * hist_weight +
982 0 : default_selectivity * (1.0 - hist_weight);
983 : }
984 :
985 : /* In any case, don't believe extremely small or large estimates. */
986 1106 : if (selec < 0.0001)
987 0 : selec = 0.0001;
988 1106 : else if (selec > 0.9999)
989 0 : selec = 0.9999;
990 :
991 : /* Don't forget to account for nulls. */
992 1106 : if (HeapTupleIsValid(vardata.statsTuple))
993 84 : nullfrac = ((Form_pg_statistic) GETSTRUCT(vardata.statsTuple))->stanullfrac;
994 : else
995 1022 : nullfrac = 0.0;
996 :
997 : /*
998 : * Now merge the results from the MCV and histogram calculations,
999 : * realizing that the histogram covers only the non-null values that
1000 : * are not listed in MCV.
1001 : */
1002 1106 : selec *= 1.0 - nullfrac - mcvsum;
1003 1106 : selec += mcvsel;
1004 : }
1005 : else
1006 : {
1007 : /* Comparison value is not constant, so we can't do anything */
1008 0 : selec = default_selectivity;
1009 : }
1010 :
1011 1106 : ReleaseVariableStats(vardata);
1012 :
1013 : /* result should be in range, but make sure... */
1014 1106 : CLAMP_PROBABILITY(selec);
1015 :
1016 1106 : return selec;
1017 : }
1018 :
1019 : /*
1020 : * ineq_histogram_selectivity - Examine the histogram for scalarineqsel
1021 : *
1022 : * Determine the fraction of the variable's histogram population that
1023 : * satisfies the inequality condition, ie, VAR < (or <=, >, >=) CONST.
1024 : * The isgt and iseq flags distinguish which of the four cases apply.
1025 : *
1026 : * While opproc could be looked up from the operator OID, common callers
1027 : * also need to call it separately, so we make the caller pass both.
1028 : *
1029 : * Returns -1 if there is no histogram (valid results will always be >= 0).
1030 : *
1031 : * Note that the result disregards both the most-common-values (if any) and
1032 : * null entries. The caller is expected to combine this result with
1033 : * statistics for those portions of the column population.
1034 : *
1035 : * This is exported so that some other estimation functions can use it.
1036 : */
1037 : double
1038 197374 : ineq_histogram_selectivity(PlannerInfo *root,
1039 : VariableStatData *vardata,
1040 : Oid opoid, FmgrInfo *opproc, bool isgt, bool iseq,
1041 : Oid collation,
1042 : Datum constval, Oid consttype)
1043 : {
1044 : double hist_selec;
1045 : AttStatsSlot sslot;
1046 :
1047 197374 : hist_selec = -1.0;
1048 :
1049 : /*
1050 : * Someday, ANALYZE might store more than one histogram per rel/att,
1051 : * corresponding to more than one possible sort ordering defined for the
1052 : * column type. Right now, we know there is only one, so just grab it and
1053 : * see if it matches the query.
1054 : *
1055 : * Note that we can't use opoid as search argument; the staop appearing in
1056 : * pg_statistic will be for the relevant '<' operator, but what we have
1057 : * might be some other inequality operator such as '>='. (Even if opoid
1058 : * is a '<' operator, it could be cross-type.) Hence we must use
1059 : * comparison_ops_are_compatible() to see if the operators match.
1060 : */
1061 393822 : if (HeapTupleIsValid(vardata->statsTuple) &&
1062 392818 : statistic_proc_security_check(vardata, opproc->fn_oid) &&
1063 196370 : get_attstatsslot(&sslot, vardata->statsTuple,
1064 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
1065 : ATTSTATSSLOT_VALUES))
1066 : {
1067 156932 : if (sslot.nvalues > 1 &&
1068 313812 : sslot.stacoll == collation &&
1069 156880 : comparison_ops_are_compatible(sslot.staop, opoid))
1070 156772 : {
1071 : /*
1072 : * Use binary search to find the desired location, namely the
1073 : * right end of the histogram bin containing the comparison value,
1074 : * which is the leftmost entry for which the comparison operator
1075 : * succeeds (if isgt) or fails (if !isgt).
1076 : *
1077 : * In this loop, we pay no attention to whether the operator iseq
1078 : * or not; that detail will be mopped up below. (We cannot tell,
1079 : * anyway, whether the operator thinks the values are equal.)
1080 : *
1081 : * If the binary search accesses the first or last histogram
1082 : * entry, we try to replace that endpoint with the true column min
1083 : * or max as found by get_actual_variable_range(). This
1084 : * ameliorates misestimates when the min or max is moving as a
1085 : * result of changes since the last ANALYZE. Note that this could
1086 : * result in effectively including MCVs into the histogram that
1087 : * weren't there before, but we don't try to correct for that.
1088 : */
1089 : double histfrac;
1090 156772 : int lobound = 0; /* first possible slot to search */
1091 156772 : int hibound = sslot.nvalues; /* last+1 slot to search */
1092 156772 : bool have_end = false;
1093 :
1094 : /*
1095 : * If there are only two histogram entries, we'll want up-to-date
1096 : * values for both. (If there are more than two, we need at most
1097 : * one of them to be updated, so we deal with that within the
1098 : * loop.)
1099 : */
1100 156772 : if (sslot.nvalues == 2)
1101 1018 : have_end = get_actual_variable_range(root,
1102 : vardata,
1103 : sslot.staop,
1104 : collation,
1105 : &sslot.values[0],
1106 1018 : &sslot.values[1]);
1107 :
1108 1065574 : while (lobound < hibound)
1109 : {
1110 908802 : int probe = (lobound + hibound) / 2;
1111 : bool ltcmp;
1112 :
1113 : /*
1114 : * If we find ourselves about to compare to the first or last
1115 : * histogram entry, first try to replace it with the actual
1116 : * current min or max (unless we already did so above).
1117 : */
1118 908802 : if (probe == 0 && sslot.nvalues > 2)
1119 74708 : have_end = get_actual_variable_range(root,
1120 : vardata,
1121 : sslot.staop,
1122 : collation,
1123 : &sslot.values[0],
1124 : NULL);
1125 834094 : else if (probe == sslot.nvalues - 1 && sslot.nvalues > 2)
1126 54500 : have_end = get_actual_variable_range(root,
1127 : vardata,
1128 : sslot.staop,
1129 : collation,
1130 : NULL,
1131 54500 : &sslot.values[probe]);
1132 :
1133 908802 : ltcmp = DatumGetBool(FunctionCall2Coll(opproc,
1134 : collation,
1135 : sslot.values[probe],
1136 : constval));
1137 908802 : if (isgt)
1138 42870 : ltcmp = !ltcmp;
1139 908802 : if (ltcmp)
1140 351046 : lobound = probe + 1;
1141 : else
1142 557756 : hibound = probe;
1143 : }
1144 :
1145 156772 : if (lobound <= 0)
1146 : {
1147 : /*
1148 : * Constant is below lower histogram boundary. More
1149 : * precisely, we have found that no entry in the histogram
1150 : * satisfies the inequality clause (if !isgt) or they all do
1151 : * (if isgt). We estimate that that's true of the entire
1152 : * table, so set histfrac to 0.0 (which we'll flip to 1.0
1153 : * below, if isgt).
1154 : */
1155 67902 : histfrac = 0.0;
1156 : }
1157 88870 : else if (lobound >= sslot.nvalues)
1158 : {
1159 : /*
1160 : * Inverse case: constant is above upper histogram boundary.
1161 : */
1162 24266 : histfrac = 1.0;
1163 : }
1164 : else
1165 : {
1166 : /* We have values[i-1] <= constant <= values[i]. */
1167 64604 : int i = lobound;
1168 64604 : double eq_selec = 0;
1169 : double val,
1170 : high,
1171 : low;
1172 : double binfrac;
1173 :
1174 : /*
1175 : * In the cases where we'll need it below, obtain an estimate
1176 : * of the selectivity of "x = constval". We use a calculation
1177 : * similar to what var_eq_const() does for a non-MCV constant,
1178 : * ie, estimate that all distinct non-MCV values occur equally
1179 : * often. But multiplication by "1.0 - sumcommon - nullfrac"
1180 : * will be done by our caller, so we shouldn't do that here.
1181 : * Therefore we can't try to clamp the estimate by reference
1182 : * to the least common MCV; the result would be too small.
1183 : *
1184 : * Note: since this is effectively assuming that constval
1185 : * isn't an MCV, it's logically dubious if constval in fact is
1186 : * one. But we have to apply *some* correction for equality,
1187 : * and anyway we cannot tell if constval is an MCV, since we
1188 : * don't have a suitable equality operator at hand.
1189 : */
1190 64604 : if (i == 1 || isgt == iseq)
1191 : {
1192 : double otherdistinct;
1193 : bool isdefault;
1194 : AttStatsSlot mcvslot;
1195 :
1196 : /* Get estimated number of distinct values */
1197 18700 : otherdistinct = get_variable_numdistinct(vardata,
1198 : &isdefault);
1199 :
1200 : /* Subtract off the number of known MCVs */
1201 18700 : if (get_attstatsslot(&mcvslot, vardata->statsTuple,
1202 : STATISTIC_KIND_MCV, InvalidOid,
1203 : ATTSTATSSLOT_NUMBERS))
1204 : {
1205 3494 : otherdistinct -= mcvslot.nnumbers;
1206 3494 : free_attstatsslot(&mcvslot);
1207 : }
1208 :
1209 : /* If result doesn't seem sane, leave eq_selec at 0 */
1210 18700 : if (otherdistinct > 1)
1211 18700 : eq_selec = 1.0 / otherdistinct;
1212 : }
1213 :
1214 : /*
1215 : * Convert the constant and the two nearest bin boundary
1216 : * values to a uniform comparison scale, and do a linear
1217 : * interpolation within this bin.
1218 : */
1219 64604 : if (convert_to_scalar(constval, consttype, collation,
1220 : &val,
1221 64604 : sslot.values[i - 1], sslot.values[i],
1222 : vardata->vartype,
1223 : &low, &high))
1224 : {
1225 64604 : if (high <= low)
1226 : {
1227 : /* cope if bin boundaries appear identical */
1228 0 : binfrac = 0.5;
1229 : }
1230 64604 : else if (val <= low)
1231 15068 : binfrac = 0.0;
1232 49536 : else if (val >= high)
1233 1954 : binfrac = 1.0;
1234 : else
1235 : {
1236 47582 : binfrac = (val - low) / (high - low);
1237 :
1238 : /*
1239 : * Watch out for the possibility that we got a NaN or
1240 : * Infinity from the division. This can happen
1241 : * despite the previous checks, if for example "low"
1242 : * is -Infinity.
1243 : */
1244 47582 : if (isnan(binfrac) ||
1245 47582 : binfrac < 0.0 || binfrac > 1.0)
1246 0 : binfrac = 0.5;
1247 : }
1248 : }
1249 : else
1250 : {
1251 : /*
1252 : * Ideally we'd produce an error here, on the grounds that
1253 : * the given operator shouldn't have scalarXXsel
1254 : * registered as its selectivity func unless we can deal
1255 : * with its operand types. But currently, all manner of
1256 : * stuff is invoking scalarXXsel, so give a default
1257 : * estimate until that can be fixed.
1258 : */
1259 0 : binfrac = 0.5;
1260 : }
1261 :
1262 : /*
1263 : * Now, compute the overall selectivity across the values
1264 : * represented by the histogram. We have i-1 full bins and
1265 : * binfrac partial bin below the constant.
1266 : */
1267 64604 : histfrac = (double) (i - 1) + binfrac;
1268 64604 : histfrac /= (double) (sslot.nvalues - 1);
1269 :
1270 : /*
1271 : * At this point, histfrac is an estimate of the fraction of
1272 : * the population represented by the histogram that satisfies
1273 : * "x <= constval". Somewhat remarkably, this statement is
1274 : * true regardless of which operator we were doing the probes
1275 : * with, so long as convert_to_scalar() delivers reasonable
1276 : * results. If the probe constant is equal to some histogram
1277 : * entry, we would have considered the bin to the left of that
1278 : * entry if probing with "<" or ">=", or the bin to the right
1279 : * if probing with "<=" or ">"; but binfrac would have come
1280 : * out as 1.0 in the first case and 0.0 in the second, leading
1281 : * to the same histfrac in either case. For probe constants
1282 : * between histogram entries, we find the same bin and get the
1283 : * same estimate with any operator.
1284 : *
1285 : * The fact that the estimate corresponds to "x <= constval"
1286 : * and not "x < constval" is because of the way that ANALYZE
1287 : * constructs the histogram: each entry is, effectively, the
1288 : * rightmost value in its sample bucket. So selectivity
1289 : * values that are exact multiples of 1/(histogram_size-1)
1290 : * should be understood as estimates including a histogram
1291 : * entry plus everything to its left.
1292 : *
1293 : * However, that breaks down for the first histogram entry,
1294 : * which necessarily is the leftmost value in its sample
1295 : * bucket. That means the first histogram bin is slightly
1296 : * narrower than the rest, by an amount equal to eq_selec.
1297 : * Another way to say that is that we want "x <= leftmost" to
1298 : * be estimated as eq_selec not zero. So, if we're dealing
1299 : * with the first bin (i==1), rescale to make that true while
1300 : * adjusting the rest of that bin linearly.
1301 : */
1302 64604 : if (i == 1)
1303 7428 : histfrac += eq_selec * (1.0 - binfrac);
1304 :
1305 : /*
1306 : * "x <= constval" is good if we want an estimate for "<=" or
1307 : * ">", but if we are estimating for "<" or ">=", we now need
1308 : * to decrease the estimate by eq_selec.
1309 : */
1310 64604 : if (isgt == iseq)
1311 17104 : histfrac -= eq_selec;
1312 : }
1313 :
1314 : /*
1315 : * Now the estimate is finished for "<" and "<=" cases. If we are
1316 : * estimating for ">" or ">=", flip it.
1317 : */
1318 156772 : hist_selec = isgt ? (1.0 - histfrac) : histfrac;
1319 :
1320 : /*
1321 : * The histogram boundaries are only approximate to begin with,
1322 : * and may well be out of date anyway. Therefore, don't believe
1323 : * extremely small or large selectivity estimates --- unless we
1324 : * got actual current endpoint values from the table, in which
1325 : * case just do the usual sanity clamp. Somewhat arbitrarily, we
1326 : * set the cutoff for other cases at a hundredth of the histogram
1327 : * resolution.
1328 : */
1329 156772 : if (have_end)
1330 83222 : CLAMP_PROBABILITY(hist_selec);
1331 : else
1332 : {
1333 73550 : double cutoff = 0.01 / (double) (sslot.nvalues - 1);
1334 :
1335 73550 : if (hist_selec < cutoff)
1336 23740 : hist_selec = cutoff;
1337 49810 : else if (hist_selec > 1.0 - cutoff)
1338 18108 : hist_selec = 1.0 - cutoff;
1339 : }
1340 : }
1341 160 : else if (sslot.nvalues > 1)
1342 : {
1343 : /*
1344 : * If we get here, we have a histogram but it's not sorted the way
1345 : * we want. Do a brute-force search to see how many of the
1346 : * entries satisfy the comparison condition, and take that
1347 : * fraction as our estimate. (This is identical to the inner loop
1348 : * of histogram_selectivity; maybe share code?)
1349 : */
1350 160 : LOCAL_FCINFO(fcinfo, 2);
1351 160 : int nmatch = 0;
1352 :
1353 160 : InitFunctionCallInfoData(*fcinfo, opproc, 2, collation,
1354 : NULL, NULL);
1355 160 : fcinfo->args[0].isnull = false;
1356 160 : fcinfo->args[1].isnull = false;
1357 160 : fcinfo->args[1].value = constval;
1358 961652 : for (int i = 0; i < sslot.nvalues; i++)
1359 : {
1360 : Datum fresult;
1361 :
1362 961492 : fcinfo->args[0].value = sslot.values[i];
1363 961492 : fcinfo->isnull = false;
1364 961492 : fresult = FunctionCallInvoke(fcinfo);
1365 961492 : if (!fcinfo->isnull && DatumGetBool(fresult))
1366 1712 : nmatch++;
1367 : }
1368 160 : hist_selec = ((double) nmatch) / ((double) sslot.nvalues);
1369 :
1370 : /*
1371 : * As above, clamp to a hundredth of the histogram resolution.
1372 : * This case is surely even less trustworthy than the normal one,
1373 : * so we shouldn't believe exact 0 or 1 selectivity. (Maybe the
1374 : * clamp should be more restrictive in this case?)
1375 : */
1376 : {
1377 160 : double cutoff = 0.01 / (double) (sslot.nvalues - 1);
1378 :
1379 160 : if (hist_selec < cutoff)
1380 0 : hist_selec = cutoff;
1381 160 : else if (hist_selec > 1.0 - cutoff)
1382 0 : hist_selec = 1.0 - cutoff;
1383 : }
1384 : }
1385 :
1386 156932 : free_attstatsslot(&sslot);
1387 : }
1388 :
1389 197374 : return hist_selec;
1390 : }
1391 :
1392 : /*
1393 : * Common wrapper function for the selectivity estimators that simply
1394 : * invoke scalarineqsel().
1395 : */
1396 : static Datum
1397 39178 : scalarineqsel_wrapper(PG_FUNCTION_ARGS, bool isgt, bool iseq)
1398 : {
1399 39178 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
1400 39178 : Oid operator = PG_GETARG_OID(1);
1401 39178 : List *args = (List *) PG_GETARG_POINTER(2);
1402 39178 : int varRelid = PG_GETARG_INT32(3);
1403 39178 : Oid collation = PG_GET_COLLATION();
1404 : VariableStatData vardata;
1405 : Node *other;
1406 : bool varonleft;
1407 : Datum constval;
1408 : Oid consttype;
1409 : double selec;
1410 :
1411 : /*
1412 : * If expression is not variable op something or something op variable,
1413 : * then punt and return a default estimate.
1414 : */
1415 39178 : if (!get_restriction_variable(root, args, varRelid,
1416 : &vardata, &other, &varonleft))
1417 342 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
1418 :
1419 : /*
1420 : * Can't do anything useful if the something is not a constant, either.
1421 : */
1422 38836 : if (!IsA(other, Const))
1423 : {
1424 2300 : ReleaseVariableStats(vardata);
1425 2300 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
1426 : }
1427 :
1428 : /*
1429 : * If the constant is NULL, assume operator is strict and return zero, ie,
1430 : * operator will never return TRUE.
1431 : */
1432 36536 : if (((Const *) other)->constisnull)
1433 : {
1434 54 : ReleaseVariableStats(vardata);
1435 54 : PG_RETURN_FLOAT8(0.0);
1436 : }
1437 36482 : constval = ((Const *) other)->constvalue;
1438 36482 : consttype = ((Const *) other)->consttype;
1439 :
1440 : /*
1441 : * Force the var to be on the left to simplify logic in scalarineqsel.
1442 : */
1443 36482 : if (!varonleft)
1444 : {
1445 354 : operator = get_commutator(operator);
1446 354 : if (!operator)
1447 : {
1448 : /* Use default selectivity (should we raise an error instead?) */
1449 0 : ReleaseVariableStats(vardata);
1450 0 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
1451 : }
1452 354 : isgt = !isgt;
1453 : }
1454 :
1455 : /* The rest of the work is done by scalarineqsel(). */
1456 36482 : selec = scalarineqsel(root, operator, isgt, iseq, collation,
1457 : &vardata, constval, consttype);
1458 :
1459 36482 : ReleaseVariableStats(vardata);
1460 :
1461 36482 : PG_RETURN_FLOAT8((float8) selec);
1462 : }
1463 :
1464 : /*
1465 : * scalarltsel - Selectivity of "<" for scalars.
1466 : */
1467 : Datum
1468 14946 : scalarltsel(PG_FUNCTION_ARGS)
1469 : {
1470 14946 : return scalarineqsel_wrapper(fcinfo, false, false);
1471 : }
1472 :
1473 : /*
1474 : * scalarlesel - Selectivity of "<=" for scalars.
1475 : */
1476 : Datum
1477 4084 : scalarlesel(PG_FUNCTION_ARGS)
1478 : {
1479 4084 : return scalarineqsel_wrapper(fcinfo, false, true);
1480 : }
1481 :
1482 : /*
1483 : * scalargtsel - Selectivity of ">" for scalars.
1484 : */
1485 : Datum
1486 13818 : scalargtsel(PG_FUNCTION_ARGS)
1487 : {
1488 13818 : return scalarineqsel_wrapper(fcinfo, true, false);
1489 : }
1490 :
1491 : /*
1492 : * scalargesel - Selectivity of ">=" for scalars.
1493 : */
1494 : Datum
1495 6330 : scalargesel(PG_FUNCTION_ARGS)
1496 : {
1497 6330 : return scalarineqsel_wrapper(fcinfo, true, true);
1498 : }
1499 :
1500 : /*
1501 : * boolvarsel - Selectivity of Boolean variable.
1502 : *
1503 : * This can actually be called on any boolean-valued expression. If it
1504 : * involves only Vars of the specified relation, and if there are statistics
1505 : * about the Var or expression (the latter is possible if it's indexed) then
1506 : * we'll produce a real estimate; otherwise it's just a default.
1507 : */
1508 : Selectivity
1509 32764 : boolvarsel(PlannerInfo *root, Node *arg, int varRelid)
1510 : {
1511 : VariableStatData vardata;
1512 : double selec;
1513 :
1514 32764 : examine_variable(root, arg, varRelid, &vardata);
1515 32764 : if (HeapTupleIsValid(vardata.statsTuple))
1516 : {
1517 : /*
1518 : * A boolean variable V is equivalent to the clause V = 't', so we
1519 : * compute the selectivity as if that is what we have.
1520 : */
1521 27722 : selec = var_eq_const(&vardata, BooleanEqualOperator, InvalidOid,
1522 : BoolGetDatum(true), false, true, false);
1523 : }
1524 : else
1525 : {
1526 : /* Otherwise, the default estimate is 0.5 */
1527 5042 : selec = 0.5;
1528 : }
1529 32764 : ReleaseVariableStats(vardata);
1530 32764 : return selec;
1531 : }
1532 :
1533 : /*
1534 : * booltestsel - Selectivity of BooleanTest Node.
1535 : */
1536 : Selectivity
1537 192 : booltestsel(PlannerInfo *root, BoolTestType booltesttype, Node *arg,
1538 : int varRelid, JoinType jointype, SpecialJoinInfo *sjinfo)
1539 : {
1540 : VariableStatData vardata;
1541 : double selec;
1542 :
1543 192 : examine_variable(root, arg, varRelid, &vardata);
1544 :
1545 192 : if (HeapTupleIsValid(vardata.statsTuple))
1546 : {
1547 : Form_pg_statistic stats;
1548 : double freq_null;
1549 : AttStatsSlot sslot;
1550 :
1551 0 : stats = (Form_pg_statistic) GETSTRUCT(vardata.statsTuple);
1552 0 : freq_null = stats->stanullfrac;
1553 :
1554 0 : if (get_attstatsslot(&sslot, vardata.statsTuple,
1555 : STATISTIC_KIND_MCV, InvalidOid,
1556 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS)
1557 0 : && sslot.nnumbers > 0)
1558 0 : {
1559 : double freq_true;
1560 : double freq_false;
1561 :
1562 : /*
1563 : * Get first MCV frequency and derive frequency for true.
1564 : */
1565 0 : if (DatumGetBool(sslot.values[0]))
1566 0 : freq_true = sslot.numbers[0];
1567 : else
1568 0 : freq_true = 1.0 - sslot.numbers[0] - freq_null;
1569 :
1570 : /*
1571 : * Next derive frequency for false. Then use these as appropriate
1572 : * to derive frequency for each case.
1573 : */
1574 0 : freq_false = 1.0 - freq_true - freq_null;
1575 :
1576 0 : switch (booltesttype)
1577 : {
1578 0 : case IS_UNKNOWN:
1579 : /* select only NULL values */
1580 0 : selec = freq_null;
1581 0 : break;
1582 0 : case IS_NOT_UNKNOWN:
1583 : /* select non-NULL values */
1584 0 : selec = 1.0 - freq_null;
1585 0 : break;
1586 0 : case IS_TRUE:
1587 : /* select only TRUE values */
1588 0 : selec = freq_true;
1589 0 : break;
1590 0 : case IS_NOT_TRUE:
1591 : /* select non-TRUE values */
1592 0 : selec = 1.0 - freq_true;
1593 0 : break;
1594 0 : case IS_FALSE:
1595 : /* select only FALSE values */
1596 0 : selec = freq_false;
1597 0 : break;
1598 0 : case IS_NOT_FALSE:
1599 : /* select non-FALSE values */
1600 0 : selec = 1.0 - freq_false;
1601 0 : break;
1602 0 : default:
1603 0 : elog(ERROR, "unrecognized booltesttype: %d",
1604 : (int) booltesttype);
1605 : selec = 0.0; /* Keep compiler quiet */
1606 : break;
1607 : }
1608 :
1609 0 : free_attstatsslot(&sslot);
1610 : }
1611 : else
1612 : {
1613 : /*
1614 : * No most-common-value info available. Still have null fraction
1615 : * information, so use it for IS [NOT] UNKNOWN. Otherwise adjust
1616 : * for null fraction and assume a 50-50 split of TRUE and FALSE.
1617 : */
1618 0 : switch (booltesttype)
1619 : {
1620 0 : case IS_UNKNOWN:
1621 : /* select only NULL values */
1622 0 : selec = freq_null;
1623 0 : break;
1624 0 : case IS_NOT_UNKNOWN:
1625 : /* select non-NULL values */
1626 0 : selec = 1.0 - freq_null;
1627 0 : break;
1628 0 : case IS_TRUE:
1629 : case IS_FALSE:
1630 : /* Assume we select half of the non-NULL values */
1631 0 : selec = (1.0 - freq_null) / 2.0;
1632 0 : break;
1633 0 : case IS_NOT_TRUE:
1634 : case IS_NOT_FALSE:
1635 : /* Assume we select NULLs plus half of the non-NULLs */
1636 : /* equiv. to freq_null + (1.0 - freq_null) / 2.0 */
1637 0 : selec = (freq_null + 1.0) / 2.0;
1638 0 : break;
1639 0 : default:
1640 0 : elog(ERROR, "unrecognized booltesttype: %d",
1641 : (int) booltesttype);
1642 : selec = 0.0; /* Keep compiler quiet */
1643 : break;
1644 : }
1645 : }
1646 : }
1647 : else
1648 : {
1649 : /*
1650 : * If we can't get variable statistics for the argument, perhaps
1651 : * clause_selectivity can do something with it. We ignore the
1652 : * possibility of a NULL value when using clause_selectivity, and just
1653 : * assume the value is either TRUE or FALSE.
1654 : */
1655 192 : switch (booltesttype)
1656 : {
1657 18 : case IS_UNKNOWN:
1658 18 : selec = DEFAULT_UNK_SEL;
1659 18 : break;
1660 18 : case IS_NOT_UNKNOWN:
1661 18 : selec = DEFAULT_NOT_UNK_SEL;
1662 18 : break;
1663 48 : case IS_TRUE:
1664 : case IS_NOT_FALSE:
1665 48 : selec = (double) clause_selectivity(root, arg,
1666 : varRelid,
1667 : jointype, sjinfo);
1668 48 : break;
1669 108 : case IS_FALSE:
1670 : case IS_NOT_TRUE:
1671 108 : selec = 1.0 - (double) clause_selectivity(root, arg,
1672 : varRelid,
1673 : jointype, sjinfo);
1674 108 : break;
1675 0 : default:
1676 0 : elog(ERROR, "unrecognized booltesttype: %d",
1677 : (int) booltesttype);
1678 : selec = 0.0; /* Keep compiler quiet */
1679 : break;
1680 : }
1681 : }
1682 :
1683 192 : ReleaseVariableStats(vardata);
1684 :
1685 : /* result should be in range, but make sure... */
1686 192 : CLAMP_PROBABILITY(selec);
1687 :
1688 192 : return (Selectivity) selec;
1689 : }
1690 :
1691 : /*
1692 : * nulltestsel - Selectivity of NullTest Node.
1693 : */
1694 : Selectivity
1695 18124 : nulltestsel(PlannerInfo *root, NullTestType nulltesttype, Node *arg,
1696 : int varRelid, JoinType jointype, SpecialJoinInfo *sjinfo)
1697 : {
1698 : VariableStatData vardata;
1699 : double selec;
1700 :
1701 18124 : examine_variable(root, arg, varRelid, &vardata);
1702 :
1703 18124 : if (HeapTupleIsValid(vardata.statsTuple))
1704 : {
1705 : Form_pg_statistic stats;
1706 : double freq_null;
1707 :
1708 7710 : stats = (Form_pg_statistic) GETSTRUCT(vardata.statsTuple);
1709 7710 : freq_null = stats->stanullfrac;
1710 :
1711 7710 : switch (nulltesttype)
1712 : {
1713 6128 : case IS_NULL:
1714 :
1715 : /*
1716 : * Use freq_null directly.
1717 : */
1718 6128 : selec = freq_null;
1719 6128 : break;
1720 1582 : case IS_NOT_NULL:
1721 :
1722 : /*
1723 : * Select not unknown (not null) values. Calculate from
1724 : * freq_null.
1725 : */
1726 1582 : selec = 1.0 - freq_null;
1727 1582 : break;
1728 0 : default:
1729 0 : elog(ERROR, "unrecognized nulltesttype: %d",
1730 : (int) nulltesttype);
1731 : return (Selectivity) 0; /* keep compiler quiet */
1732 : }
1733 : }
1734 10414 : else if (vardata.var && IsA(vardata.var, Var) &&
1735 9940 : ((Var *) vardata.var)->varattno < 0)
1736 : {
1737 : /*
1738 : * There are no stats for system columns, but we know they are never
1739 : * NULL.
1740 : */
1741 84 : selec = (nulltesttype == IS_NULL) ? 0.0 : 1.0;
1742 : }
1743 : else
1744 : {
1745 : /*
1746 : * No ANALYZE stats available, so make a guess
1747 : */
1748 10330 : switch (nulltesttype)
1749 : {
1750 1836 : case IS_NULL:
1751 1836 : selec = DEFAULT_UNK_SEL;
1752 1836 : break;
1753 8494 : case IS_NOT_NULL:
1754 8494 : selec = DEFAULT_NOT_UNK_SEL;
1755 8494 : break;
1756 0 : default:
1757 0 : elog(ERROR, "unrecognized nulltesttype: %d",
1758 : (int) nulltesttype);
1759 : return (Selectivity) 0; /* keep compiler quiet */
1760 : }
1761 : }
1762 :
1763 18124 : ReleaseVariableStats(vardata);
1764 :
1765 : /* result should be in range, but make sure... */
1766 18124 : CLAMP_PROBABILITY(selec);
1767 :
1768 18124 : return (Selectivity) selec;
1769 : }
1770 :
1771 : /*
1772 : * strip_array_coercion - strip binary-compatible relabeling from an array expr
1773 : *
1774 : * For array values, the parser normally generates ArrayCoerceExpr conversions,
1775 : * but it seems possible that RelabelType might show up. Also, the planner
1776 : * is not currently tense about collapsing stacked ArrayCoerceExpr nodes,
1777 : * so we need to be ready to deal with more than one level.
1778 : */
1779 : static Node *
1780 91948 : strip_array_coercion(Node *node)
1781 : {
1782 : for (;;)
1783 : {
1784 91948 : if (node && IsA(node, ArrayCoerceExpr))
1785 36 : {
1786 1364 : ArrayCoerceExpr *acoerce = (ArrayCoerceExpr *) node;
1787 :
1788 : /*
1789 : * If the per-element expression is just a RelabelType on top of
1790 : * CaseTestExpr, then we know it's a binary-compatible relabeling.
1791 : */
1792 1364 : if (IsA(acoerce->elemexpr, RelabelType) &&
1793 36 : IsA(((RelabelType *) acoerce->elemexpr)->arg, CaseTestExpr))
1794 36 : node = (Node *) acoerce->arg;
1795 : else
1796 : break;
1797 : }
1798 90584 : else if (node && IsA(node, RelabelType))
1799 : {
1800 : /* We don't really expect this case, but may as well cope */
1801 0 : node = (Node *) ((RelabelType *) node)->arg;
1802 : }
1803 : else
1804 : break;
1805 : }
1806 91912 : return node;
1807 : }
1808 :
1809 : /*
1810 : * scalararraysel - Selectivity of ScalarArrayOpExpr Node.
1811 : */
1812 : Selectivity
1813 14208 : scalararraysel(PlannerInfo *root,
1814 : ScalarArrayOpExpr *clause,
1815 : bool is_join_clause,
1816 : int varRelid,
1817 : JoinType jointype,
1818 : SpecialJoinInfo *sjinfo)
1819 : {
1820 14208 : Oid operator = clause->opno;
1821 14208 : bool useOr = clause->useOr;
1822 14208 : bool isEquality = false;
1823 14208 : bool isInequality = false;
1824 : Node *leftop;
1825 : Node *rightop;
1826 : Oid nominal_element_type;
1827 : Oid nominal_element_collation;
1828 : TypeCacheEntry *typentry;
1829 : RegProcedure oprsel;
1830 : FmgrInfo oprselproc;
1831 : Selectivity s1;
1832 : Selectivity s1disjoint;
1833 :
1834 : /* First, deconstruct the expression */
1835 : Assert(list_length(clause->args) == 2);
1836 14208 : leftop = (Node *) linitial(clause->args);
1837 14208 : rightop = (Node *) lsecond(clause->args);
1838 :
1839 : /* aggressively reduce both sides to constants */
1840 14208 : leftop = estimate_expression_value(root, leftop);
1841 14208 : rightop = estimate_expression_value(root, rightop);
1842 :
1843 : /* get nominal (after relabeling) element type of rightop */
1844 14208 : nominal_element_type = get_base_element_type(exprType(rightop));
1845 14208 : if (!OidIsValid(nominal_element_type))
1846 0 : return (Selectivity) 0.5; /* probably shouldn't happen */
1847 : /* get nominal collation, too, for generating constants */
1848 14208 : nominal_element_collation = exprCollation(rightop);
1849 :
1850 : /* look through any binary-compatible relabeling of rightop */
1851 14208 : rightop = strip_array_coercion(rightop);
1852 :
1853 : /*
1854 : * Detect whether the operator is the default equality or inequality
1855 : * operator of the array element type.
1856 : */
1857 14208 : typentry = lookup_type_cache(nominal_element_type, TYPECACHE_EQ_OPR);
1858 14208 : if (OidIsValid(typentry->eq_opr))
1859 : {
1860 14208 : if (operator == typentry->eq_opr)
1861 12358 : isEquality = true;
1862 1850 : else if (get_negator(operator) == typentry->eq_opr)
1863 1370 : isInequality = true;
1864 : }
1865 :
1866 : /*
1867 : * If it is equality or inequality, we might be able to estimate this as a
1868 : * form of array containment; for instance "const = ANY(column)" can be
1869 : * treated as "ARRAY[const] <@ column". scalararraysel_containment tries
1870 : * that, and returns the selectivity estimate if successful, or -1 if not.
1871 : */
1872 14208 : if ((isEquality || isInequality) && !is_join_clause)
1873 : {
1874 13728 : s1 = scalararraysel_containment(root, leftop, rightop,
1875 : nominal_element_type,
1876 : isEquality, useOr, varRelid);
1877 13728 : if (s1 >= 0.0)
1878 112 : return s1;
1879 : }
1880 :
1881 : /*
1882 : * Look up the underlying operator's selectivity estimator. Punt if it
1883 : * hasn't got one.
1884 : */
1885 14096 : if (is_join_clause)
1886 0 : oprsel = get_oprjoin(operator);
1887 : else
1888 14096 : oprsel = get_oprrest(operator);
1889 14096 : if (!oprsel)
1890 0 : return (Selectivity) 0.5;
1891 14096 : fmgr_info(oprsel, &oprselproc);
1892 :
1893 : /*
1894 : * In the array-containment check above, we must only believe that an
1895 : * operator is equality or inequality if it is the default btree equality
1896 : * operator (or its negator) for the element type, since those are the
1897 : * operators that array containment will use. But in what follows, we can
1898 : * be a little laxer, and also believe that any operators using eqsel() or
1899 : * neqsel() as selectivity estimator act like equality or inequality.
1900 : */
1901 14096 : if (oprsel == F_EQSEL || oprsel == F_EQJOINSEL)
1902 12410 : isEquality = true;
1903 1686 : else if (oprsel == F_NEQSEL || oprsel == F_NEQJOINSEL)
1904 1260 : isInequality = true;
1905 :
1906 : /*
1907 : * We consider three cases:
1908 : *
1909 : * 1. rightop is an Array constant: deconstruct the array, apply the
1910 : * operator's selectivity function for each array element, and merge the
1911 : * results in the same way that clausesel.c does for AND/OR combinations.
1912 : *
1913 : * 2. rightop is an ARRAY[] construct: apply the operator's selectivity
1914 : * function for each element of the ARRAY[] construct, and merge.
1915 : *
1916 : * 3. otherwise, make a guess ...
1917 : */
1918 14096 : if (rightop && IsA(rightop, Const))
1919 11654 : {
1920 11672 : Datum arraydatum = ((Const *) rightop)->constvalue;
1921 11672 : bool arrayisnull = ((Const *) rightop)->constisnull;
1922 : ArrayType *arrayval;
1923 : int16 elmlen;
1924 : bool elmbyval;
1925 : char elmalign;
1926 : int num_elems;
1927 : Datum *elem_values;
1928 : bool *elem_nulls;
1929 : int i;
1930 :
1931 11672 : if (arrayisnull) /* qual can't succeed if null array */
1932 18 : return (Selectivity) 0.0;
1933 11654 : arrayval = DatumGetArrayTypeP(arraydatum);
1934 11654 : get_typlenbyvalalign(ARR_ELEMTYPE(arrayval),
1935 : &elmlen, &elmbyval, &elmalign);
1936 11654 : deconstruct_array(arrayval,
1937 : ARR_ELEMTYPE(arrayval),
1938 : elmlen, elmbyval, elmalign,
1939 : &elem_values, &elem_nulls, &num_elems);
1940 :
1941 : /*
1942 : * For generic operators, we assume the probability of success is
1943 : * independent for each array element. But for "= ANY" or "<> ALL",
1944 : * if the array elements are distinct (which'd typically be the case)
1945 : * then the probabilities are disjoint, and we should just sum them.
1946 : *
1947 : * If we were being really tense we would try to confirm that the
1948 : * elements are all distinct, but that would be expensive and it
1949 : * doesn't seem to be worth the cycles; it would amount to penalizing
1950 : * well-written queries in favor of poorly-written ones. However, we
1951 : * do protect ourselves a little bit by checking whether the
1952 : * disjointness assumption leads to an impossible (out of range)
1953 : * probability; if so, we fall back to the normal calculation.
1954 : */
1955 11654 : s1 = s1disjoint = (useOr ? 0.0 : 1.0);
1956 :
1957 48450 : for (i = 0; i < num_elems; i++)
1958 : {
1959 : List *args;
1960 : Selectivity s2;
1961 :
1962 36796 : args = list_make2(leftop,
1963 : makeConst(nominal_element_type,
1964 : -1,
1965 : nominal_element_collation,
1966 : elmlen,
1967 : elem_values[i],
1968 : elem_nulls[i],
1969 : elmbyval));
1970 36796 : if (is_join_clause)
1971 0 : s2 = DatumGetFloat8(FunctionCall5Coll(&oprselproc,
1972 : clause->inputcollid,
1973 : PointerGetDatum(root),
1974 : ObjectIdGetDatum(operator),
1975 : PointerGetDatum(args),
1976 : Int16GetDatum(jointype),
1977 : PointerGetDatum(sjinfo)));
1978 : else
1979 36796 : s2 = DatumGetFloat8(FunctionCall4Coll(&oprselproc,
1980 : clause->inputcollid,
1981 : PointerGetDatum(root),
1982 : ObjectIdGetDatum(operator),
1983 : PointerGetDatum(args),
1984 : Int32GetDatum(varRelid)));
1985 :
1986 36796 : if (useOr)
1987 : {
1988 32828 : s1 = s1 + s2 - s1 * s2;
1989 32828 : if (isEquality)
1990 31964 : s1disjoint += s2;
1991 : }
1992 : else
1993 : {
1994 3968 : s1 = s1 * s2;
1995 3968 : if (isInequality)
1996 3656 : s1disjoint += s2 - 1.0;
1997 : }
1998 : }
1999 :
2000 : /* accept disjoint-probability estimate if in range */
2001 11654 : if ((useOr ? isEquality : isInequality) &&
2002 11078 : s1disjoint >= 0.0 && s1disjoint <= 1.0)
2003 11030 : s1 = s1disjoint;
2004 : }
2005 2424 : else if (rightop && IsA(rightop, ArrayExpr) &&
2006 104 : !((ArrayExpr *) rightop)->multidims)
2007 104 : {
2008 104 : ArrayExpr *arrayexpr = (ArrayExpr *) rightop;
2009 : int16 elmlen;
2010 : bool elmbyval;
2011 : ListCell *l;
2012 :
2013 104 : get_typlenbyval(arrayexpr->element_typeid,
2014 : &elmlen, &elmbyval);
2015 :
2016 : /*
2017 : * We use the assumption of disjoint probabilities here too, although
2018 : * the odds of equal array elements are rather higher if the elements
2019 : * are not all constants (which they won't be, else constant folding
2020 : * would have reduced the ArrayExpr to a Const). In this path it's
2021 : * critical to have the sanity check on the s1disjoint estimate.
2022 : */
2023 104 : s1 = s1disjoint = (useOr ? 0.0 : 1.0);
2024 :
2025 368 : foreach(l, arrayexpr->elements)
2026 : {
2027 264 : Node *elem = (Node *) lfirst(l);
2028 : List *args;
2029 : Selectivity s2;
2030 :
2031 : /*
2032 : * Theoretically, if elem isn't of nominal_element_type we should
2033 : * insert a RelabelType, but it seems unlikely that any operator
2034 : * estimation function would really care ...
2035 : */
2036 264 : args = list_make2(leftop, elem);
2037 264 : if (is_join_clause)
2038 0 : s2 = DatumGetFloat8(FunctionCall5Coll(&oprselproc,
2039 : clause->inputcollid,
2040 : PointerGetDatum(root),
2041 : ObjectIdGetDatum(operator),
2042 : PointerGetDatum(args),
2043 : Int16GetDatum(jointype),
2044 : PointerGetDatum(sjinfo)));
2045 : else
2046 264 : s2 = DatumGetFloat8(FunctionCall4Coll(&oprselproc,
2047 : clause->inputcollid,
2048 : PointerGetDatum(root),
2049 : ObjectIdGetDatum(operator),
2050 : PointerGetDatum(args),
2051 : Int32GetDatum(varRelid)));
2052 :
2053 264 : if (useOr)
2054 : {
2055 264 : s1 = s1 + s2 - s1 * s2;
2056 264 : if (isEquality)
2057 264 : s1disjoint += s2;
2058 : }
2059 : else
2060 : {
2061 0 : s1 = s1 * s2;
2062 0 : if (isInequality)
2063 0 : s1disjoint += s2 - 1.0;
2064 : }
2065 : }
2066 :
2067 : /* accept disjoint-probability estimate if in range */
2068 104 : if ((useOr ? isEquality : isInequality) &&
2069 104 : s1disjoint >= 0.0 && s1disjoint <= 1.0)
2070 104 : s1 = s1disjoint;
2071 : }
2072 : else
2073 : {
2074 : CaseTestExpr *dummyexpr;
2075 : List *args;
2076 : Selectivity s2;
2077 : int i;
2078 :
2079 : /*
2080 : * We need a dummy rightop to pass to the operator selectivity
2081 : * routine. It can be pretty much anything that doesn't look like a
2082 : * constant; CaseTestExpr is a convenient choice.
2083 : */
2084 2320 : dummyexpr = makeNode(CaseTestExpr);
2085 2320 : dummyexpr->typeId = nominal_element_type;
2086 2320 : dummyexpr->typeMod = -1;
2087 2320 : dummyexpr->collation = clause->inputcollid;
2088 2320 : args = list_make2(leftop, dummyexpr);
2089 2320 : if (is_join_clause)
2090 0 : s2 = DatumGetFloat8(FunctionCall5Coll(&oprselproc,
2091 : clause->inputcollid,
2092 : PointerGetDatum(root),
2093 : ObjectIdGetDatum(operator),
2094 : PointerGetDatum(args),
2095 : Int16GetDatum(jointype),
2096 : PointerGetDatum(sjinfo)));
2097 : else
2098 2320 : s2 = DatumGetFloat8(FunctionCall4Coll(&oprselproc,
2099 : clause->inputcollid,
2100 : PointerGetDatum(root),
2101 : ObjectIdGetDatum(operator),
2102 : PointerGetDatum(args),
2103 : Int32GetDatum(varRelid)));
2104 2320 : s1 = useOr ? 0.0 : 1.0;
2105 :
2106 : /*
2107 : * Arbitrarily assume 10 elements in the eventual array value (see
2108 : * also estimate_array_length). We don't risk an assumption of
2109 : * disjoint probabilities here.
2110 : */
2111 25520 : for (i = 0; i < 10; i++)
2112 : {
2113 23200 : if (useOr)
2114 23200 : s1 = s1 + s2 - s1 * s2;
2115 : else
2116 0 : s1 = s1 * s2;
2117 : }
2118 : }
2119 :
2120 : /* result should be in range, but make sure... */
2121 14078 : CLAMP_PROBABILITY(s1);
2122 :
2123 14078 : return s1;
2124 : }
2125 :
2126 : /*
2127 : * Estimate number of elements in the array yielded by an expression.
2128 : *
2129 : * It's important that this agree with scalararraysel.
2130 : */
2131 : int
2132 77704 : estimate_array_length(Node *arrayexpr)
2133 : {
2134 : /* look through any binary-compatible relabeling of arrayexpr */
2135 77704 : arrayexpr = strip_array_coercion(arrayexpr);
2136 :
2137 77704 : if (arrayexpr && IsA(arrayexpr, Const))
2138 : {
2139 36410 : Datum arraydatum = ((Const *) arrayexpr)->constvalue;
2140 36410 : bool arrayisnull = ((Const *) arrayexpr)->constisnull;
2141 : ArrayType *arrayval;
2142 :
2143 36410 : if (arrayisnull)
2144 36 : return 0;
2145 36374 : arrayval = DatumGetArrayTypeP(arraydatum);
2146 36374 : return ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval));
2147 : }
2148 41294 : else if (arrayexpr && IsA(arrayexpr, ArrayExpr) &&
2149 454 : !((ArrayExpr *) arrayexpr)->multidims)
2150 : {
2151 454 : return list_length(((ArrayExpr *) arrayexpr)->elements);
2152 : }
2153 : else
2154 : {
2155 : /* default guess --- see also scalararraysel */
2156 40840 : return 10;
2157 : }
2158 : }
2159 :
2160 : /*
2161 : * rowcomparesel - Selectivity of RowCompareExpr Node.
2162 : *
2163 : * We estimate RowCompare selectivity by considering just the first (high
2164 : * order) columns, which makes it equivalent to an ordinary OpExpr. While
2165 : * this estimate could be refined by considering additional columns, it
2166 : * seems unlikely that we could do a lot better without multi-column
2167 : * statistics.
2168 : */
2169 : Selectivity
2170 156 : rowcomparesel(PlannerInfo *root,
2171 : RowCompareExpr *clause,
2172 : int varRelid, JoinType jointype, SpecialJoinInfo *sjinfo)
2173 : {
2174 : Selectivity s1;
2175 156 : Oid opno = linitial_oid(clause->opnos);
2176 156 : Oid inputcollid = linitial_oid(clause->inputcollids);
2177 : List *opargs;
2178 : bool is_join_clause;
2179 :
2180 : /* Build equivalent arg list for single operator */
2181 156 : opargs = list_make2(linitial(clause->largs), linitial(clause->rargs));
2182 :
2183 : /*
2184 : * Decide if it's a join clause. This should match clausesel.c's
2185 : * treat_as_join_clause(), except that we intentionally consider only the
2186 : * leading columns and not the rest of the clause.
2187 : */
2188 156 : if (varRelid != 0)
2189 : {
2190 : /*
2191 : * Caller is forcing restriction mode (eg, because we are examining an
2192 : * inner indexscan qual).
2193 : */
2194 54 : is_join_clause = false;
2195 : }
2196 102 : else if (sjinfo == NULL)
2197 : {
2198 : /*
2199 : * It must be a restriction clause, since it's being evaluated at a
2200 : * scan node.
2201 : */
2202 90 : is_join_clause = false;
2203 : }
2204 : else
2205 : {
2206 : /*
2207 : * Otherwise, it's a join if there's more than one relation used.
2208 : */
2209 12 : is_join_clause = (NumRelids(root, (Node *) opargs) > 1);
2210 : }
2211 :
2212 156 : if (is_join_clause)
2213 : {
2214 : /* Estimate selectivity for a join clause. */
2215 12 : s1 = join_selectivity(root, opno,
2216 : opargs,
2217 : inputcollid,
2218 : jointype,
2219 : sjinfo);
2220 : }
2221 : else
2222 : {
2223 : /* Estimate selectivity for a restriction clause. */
2224 144 : s1 = restriction_selectivity(root, opno,
2225 : opargs,
2226 : inputcollid,
2227 : varRelid);
2228 : }
2229 :
2230 156 : return s1;
2231 : }
2232 :
2233 : /*
2234 : * eqjoinsel - Join selectivity of "="
2235 : */
2236 : Datum
2237 165494 : eqjoinsel(PG_FUNCTION_ARGS)
2238 : {
2239 165494 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
2240 165494 : Oid operator = PG_GETARG_OID(1);
2241 165494 : List *args = (List *) PG_GETARG_POINTER(2);
2242 :
2243 : #ifdef NOT_USED
2244 : JoinType jointype = (JoinType) PG_GETARG_INT16(3);
2245 : #endif
2246 165494 : SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) PG_GETARG_POINTER(4);
2247 165494 : Oid collation = PG_GET_COLLATION();
2248 : double selec;
2249 : double selec_inner;
2250 : VariableStatData vardata1;
2251 : VariableStatData vardata2;
2252 : double nd1;
2253 : double nd2;
2254 : bool isdefault1;
2255 : bool isdefault2;
2256 : Oid opfuncoid;
2257 : AttStatsSlot sslot1;
2258 : AttStatsSlot sslot2;
2259 165494 : Form_pg_statistic stats1 = NULL;
2260 165494 : Form_pg_statistic stats2 = NULL;
2261 165494 : bool have_mcvs1 = false;
2262 165494 : bool have_mcvs2 = false;
2263 : bool join_is_reversed;
2264 : RelOptInfo *inner_rel;
2265 :
2266 165494 : get_join_variables(root, args, sjinfo,
2267 : &vardata1, &vardata2, &join_is_reversed);
2268 :
2269 165494 : nd1 = get_variable_numdistinct(&vardata1, &isdefault1);
2270 165494 : nd2 = get_variable_numdistinct(&vardata2, &isdefault2);
2271 :
2272 165494 : opfuncoid = get_opcode(operator);
2273 :
2274 165494 : memset(&sslot1, 0, sizeof(sslot1));
2275 165494 : memset(&sslot2, 0, sizeof(sslot2));
2276 :
2277 165494 : if (HeapTupleIsValid(vardata1.statsTuple))
2278 : {
2279 : /* note we allow use of nullfrac regardless of security check */
2280 109856 : stats1 = (Form_pg_statistic) GETSTRUCT(vardata1.statsTuple);
2281 109856 : if (statistic_proc_security_check(&vardata1, opfuncoid))
2282 109856 : have_mcvs1 = get_attstatsslot(&sslot1, vardata1.statsTuple,
2283 : STATISTIC_KIND_MCV, InvalidOid,
2284 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS);
2285 : }
2286 :
2287 165494 : if (HeapTupleIsValid(vardata2.statsTuple))
2288 : {
2289 : /* note we allow use of nullfrac regardless of security check */
2290 106666 : stats2 = (Form_pg_statistic) GETSTRUCT(vardata2.statsTuple);
2291 106666 : if (statistic_proc_security_check(&vardata2, opfuncoid))
2292 106666 : have_mcvs2 = get_attstatsslot(&sslot2, vardata2.statsTuple,
2293 : STATISTIC_KIND_MCV, InvalidOid,
2294 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS);
2295 : }
2296 :
2297 : /* We need to compute the inner-join selectivity in all cases */
2298 165494 : selec_inner = eqjoinsel_inner(opfuncoid, collation,
2299 : &vardata1, &vardata2,
2300 : nd1, nd2,
2301 : isdefault1, isdefault2,
2302 : &sslot1, &sslot2,
2303 : stats1, stats2,
2304 : have_mcvs1, have_mcvs2);
2305 :
2306 165494 : switch (sjinfo->jointype)
2307 : {
2308 158656 : case JOIN_INNER:
2309 : case JOIN_LEFT:
2310 : case JOIN_FULL:
2311 158656 : selec = selec_inner;
2312 158656 : break;
2313 6838 : case JOIN_SEMI:
2314 : case JOIN_ANTI:
2315 :
2316 : /*
2317 : * Look up the join's inner relation. min_righthand is sufficient
2318 : * information because neither SEMI nor ANTI joins permit any
2319 : * reassociation into or out of their RHS, so the righthand will
2320 : * always be exactly that set of rels.
2321 : */
2322 6838 : inner_rel = find_join_input_rel(root, sjinfo->min_righthand);
2323 :
2324 6838 : if (!join_is_reversed)
2325 2584 : selec = eqjoinsel_semi(opfuncoid, collation,
2326 : &vardata1, &vardata2,
2327 : nd1, nd2,
2328 : isdefault1, isdefault2,
2329 : &sslot1, &sslot2,
2330 : stats1, stats2,
2331 : have_mcvs1, have_mcvs2,
2332 : inner_rel);
2333 : else
2334 : {
2335 4254 : Oid commop = get_commutator(operator);
2336 4254 : Oid commopfuncoid = OidIsValid(commop) ? get_opcode(commop) : InvalidOid;
2337 :
2338 4254 : selec = eqjoinsel_semi(commopfuncoid, collation,
2339 : &vardata2, &vardata1,
2340 : nd2, nd1,
2341 : isdefault2, isdefault1,
2342 : &sslot2, &sslot1,
2343 : stats2, stats1,
2344 : have_mcvs2, have_mcvs1,
2345 : inner_rel);
2346 : }
2347 :
2348 : /*
2349 : * We should never estimate the output of a semijoin to be more
2350 : * rows than we estimate for an inner join with the same input
2351 : * rels and join condition; it's obviously impossible for that to
2352 : * happen. The former estimate is N1 * Ssemi while the latter is
2353 : * N1 * N2 * Sinner, so we may clamp Ssemi <= N2 * Sinner. Doing
2354 : * this is worthwhile because of the shakier estimation rules we
2355 : * use in eqjoinsel_semi, particularly in cases where it has to
2356 : * punt entirely.
2357 : */
2358 6838 : selec = Min(selec, inner_rel->rows * selec_inner);
2359 6838 : break;
2360 0 : default:
2361 : /* other values not expected here */
2362 0 : elog(ERROR, "unrecognized join type: %d",
2363 : (int) sjinfo->jointype);
2364 : selec = 0; /* keep compiler quiet */
2365 : break;
2366 : }
2367 :
2368 165494 : free_attstatsslot(&sslot1);
2369 165494 : free_attstatsslot(&sslot2);
2370 :
2371 165494 : ReleaseVariableStats(vardata1);
2372 165494 : ReleaseVariableStats(vardata2);
2373 :
2374 165494 : CLAMP_PROBABILITY(selec);
2375 :
2376 165494 : PG_RETURN_FLOAT8((float8) selec);
2377 : }
2378 :
2379 : /*
2380 : * eqjoinsel_inner --- eqjoinsel for normal inner join
2381 : *
2382 : * We also use this for LEFT/FULL outer joins; it's not presently clear
2383 : * that it's worth trying to distinguish them here.
2384 : */
2385 : static double
2386 165494 : eqjoinsel_inner(Oid opfuncoid, Oid collation,
2387 : VariableStatData *vardata1, VariableStatData *vardata2,
2388 : double nd1, double nd2,
2389 : bool isdefault1, bool isdefault2,
2390 : AttStatsSlot *sslot1, AttStatsSlot *sslot2,
2391 : Form_pg_statistic stats1, Form_pg_statistic stats2,
2392 : bool have_mcvs1, bool have_mcvs2)
2393 : {
2394 : double selec;
2395 :
2396 165494 : if (have_mcvs1 && have_mcvs2)
2397 14358 : {
2398 : /*
2399 : * We have most-common-value lists for both relations. Run through
2400 : * the lists to see which MCVs actually join to each other with the
2401 : * given operator. This allows us to determine the exact join
2402 : * selectivity for the portion of the relations represented by the MCV
2403 : * lists. We still have to estimate for the remaining population, but
2404 : * in a skewed distribution this gives us a big leg up in accuracy.
2405 : * For motivation see the analysis in Y. Ioannidis and S.
2406 : * Christodoulakis, "On the propagation of errors in the size of join
2407 : * results", Technical Report 1018, Computer Science Dept., University
2408 : * of Wisconsin, Madison, March 1991 (available from ftp.cs.wisc.edu).
2409 : */
2410 14358 : LOCAL_FCINFO(fcinfo, 2);
2411 : FmgrInfo eqproc;
2412 : bool *hasmatch1;
2413 : bool *hasmatch2;
2414 14358 : double nullfrac1 = stats1->stanullfrac;
2415 14358 : double nullfrac2 = stats2->stanullfrac;
2416 : double matchprodfreq,
2417 : matchfreq1,
2418 : matchfreq2,
2419 : unmatchfreq1,
2420 : unmatchfreq2,
2421 : otherfreq1,
2422 : otherfreq2,
2423 : totalsel1,
2424 : totalsel2;
2425 : int i,
2426 : nmatches;
2427 :
2428 14358 : fmgr_info(opfuncoid, &eqproc);
2429 :
2430 : /*
2431 : * Save a few cycles by setting up the fcinfo struct just once. Using
2432 : * FunctionCallInvoke directly also avoids failure if the eqproc
2433 : * returns NULL, though really equality functions should never do
2434 : * that.
2435 : */
2436 14358 : InitFunctionCallInfoData(*fcinfo, &eqproc, 2, collation,
2437 : NULL, NULL);
2438 14358 : fcinfo->args[0].isnull = false;
2439 14358 : fcinfo->args[1].isnull = false;
2440 :
2441 14358 : hasmatch1 = (bool *) palloc0(sslot1->nvalues * sizeof(bool));
2442 14358 : hasmatch2 = (bool *) palloc0(sslot2->nvalues * sizeof(bool));
2443 :
2444 : /*
2445 : * Note we assume that each MCV will match at most one member of the
2446 : * other MCV list. If the operator isn't really equality, there could
2447 : * be multiple matches --- but we don't look for them, both for speed
2448 : * and because the math wouldn't add up...
2449 : */
2450 14358 : matchprodfreq = 0.0;
2451 14358 : nmatches = 0;
2452 344390 : for (i = 0; i < sslot1->nvalues; i++)
2453 : {
2454 : int j;
2455 :
2456 330032 : fcinfo->args[0].value = sslot1->values[i];
2457 :
2458 13347150 : for (j = 0; j < sslot2->nvalues; j++)
2459 : {
2460 : Datum fresult;
2461 :
2462 13137416 : if (hasmatch2[j])
2463 3552254 : continue;
2464 9585162 : fcinfo->args[1].value = sslot2->values[j];
2465 9585162 : fcinfo->isnull = false;
2466 9585162 : fresult = FunctionCallInvoke(fcinfo);
2467 9585162 : if (!fcinfo->isnull && DatumGetBool(fresult))
2468 : {
2469 120298 : hasmatch1[i] = hasmatch2[j] = true;
2470 120298 : matchprodfreq += sslot1->numbers[i] * sslot2->numbers[j];
2471 120298 : nmatches++;
2472 120298 : break;
2473 : }
2474 : }
2475 : }
2476 14358 : CLAMP_PROBABILITY(matchprodfreq);
2477 : /* Sum up frequencies of matched and unmatched MCVs */
2478 14358 : matchfreq1 = unmatchfreq1 = 0.0;
2479 344390 : for (i = 0; i < sslot1->nvalues; i++)
2480 : {
2481 330032 : if (hasmatch1[i])
2482 120298 : matchfreq1 += sslot1->numbers[i];
2483 : else
2484 209734 : unmatchfreq1 += sslot1->numbers[i];
2485 : }
2486 14358 : CLAMP_PROBABILITY(matchfreq1);
2487 14358 : CLAMP_PROBABILITY(unmatchfreq1);
2488 14358 : matchfreq2 = unmatchfreq2 = 0.0;
2489 598538 : for (i = 0; i < sslot2->nvalues; i++)
2490 : {
2491 584180 : if (hasmatch2[i])
2492 120298 : matchfreq2 += sslot2->numbers[i];
2493 : else
2494 463882 : unmatchfreq2 += sslot2->numbers[i];
2495 : }
2496 14358 : CLAMP_PROBABILITY(matchfreq2);
2497 14358 : CLAMP_PROBABILITY(unmatchfreq2);
2498 14358 : pfree(hasmatch1);
2499 14358 : pfree(hasmatch2);
2500 :
2501 : /*
2502 : * Compute total frequency of non-null values that are not in the MCV
2503 : * lists.
2504 : */
2505 14358 : otherfreq1 = 1.0 - nullfrac1 - matchfreq1 - unmatchfreq1;
2506 14358 : otherfreq2 = 1.0 - nullfrac2 - matchfreq2 - unmatchfreq2;
2507 14358 : CLAMP_PROBABILITY(otherfreq1);
2508 14358 : CLAMP_PROBABILITY(otherfreq2);
2509 :
2510 : /*
2511 : * We can estimate the total selectivity from the point of view of
2512 : * relation 1 as: the known selectivity for matched MCVs, plus
2513 : * unmatched MCVs that are assumed to match against random members of
2514 : * relation 2's non-MCV population, plus non-MCV values that are
2515 : * assumed to match against random members of relation 2's unmatched
2516 : * MCVs plus non-MCV values.
2517 : */
2518 14358 : totalsel1 = matchprodfreq;
2519 14358 : if (nd2 > sslot2->nvalues)
2520 8988 : totalsel1 += unmatchfreq1 * otherfreq2 / (nd2 - sslot2->nvalues);
2521 14358 : if (nd2 > nmatches)
2522 11946 : totalsel1 += otherfreq1 * (otherfreq2 + unmatchfreq2) /
2523 11946 : (nd2 - nmatches);
2524 : /* Same estimate from the point of view of relation 2. */
2525 14358 : totalsel2 = matchprodfreq;
2526 14358 : if (nd1 > sslot1->nvalues)
2527 8600 : totalsel2 += unmatchfreq2 * otherfreq1 / (nd1 - sslot1->nvalues);
2528 14358 : if (nd1 > nmatches)
2529 11150 : totalsel2 += otherfreq2 * (otherfreq1 + unmatchfreq1) /
2530 11150 : (nd1 - nmatches);
2531 :
2532 : /*
2533 : * Use the smaller of the two estimates. This can be justified in
2534 : * essentially the same terms as given below for the no-stats case: to
2535 : * a first approximation, we are estimating from the point of view of
2536 : * the relation with smaller nd.
2537 : */
2538 14358 : selec = (totalsel1 < totalsel2) ? totalsel1 : totalsel2;
2539 : }
2540 : else
2541 : {
2542 : /*
2543 : * We do not have MCV lists for both sides. Estimate the join
2544 : * selectivity as MIN(1/nd1,1/nd2)*(1-nullfrac1)*(1-nullfrac2). This
2545 : * is plausible if we assume that the join operator is strict and the
2546 : * non-null values are about equally distributed: a given non-null
2547 : * tuple of rel1 will join to either zero or N2*(1-nullfrac2)/nd2 rows
2548 : * of rel2, so total join rows are at most
2549 : * N1*(1-nullfrac1)*N2*(1-nullfrac2)/nd2 giving a join selectivity of
2550 : * not more than (1-nullfrac1)*(1-nullfrac2)/nd2. By the same logic it
2551 : * is not more than (1-nullfrac1)*(1-nullfrac2)/nd1, so the expression
2552 : * with MIN() is an upper bound. Using the MIN() means we estimate
2553 : * from the point of view of the relation with smaller nd (since the
2554 : * larger nd is determining the MIN). It is reasonable to assume that
2555 : * most tuples in this rel will have join partners, so the bound is
2556 : * probably reasonably tight and should be taken as-is.
2557 : *
2558 : * XXX Can we be smarter if we have an MCV list for just one side? It
2559 : * seems that if we assume equal distribution for the other side, we
2560 : * end up with the same answer anyway.
2561 : */
2562 151136 : double nullfrac1 = stats1 ? stats1->stanullfrac : 0.0;
2563 151136 : double nullfrac2 = stats2 ? stats2->stanullfrac : 0.0;
2564 :
2565 151136 : selec = (1.0 - nullfrac1) * (1.0 - nullfrac2);
2566 151136 : if (nd1 > nd2)
2567 61194 : selec /= nd1;
2568 : else
2569 89942 : selec /= nd2;
2570 : }
2571 :
2572 165494 : return selec;
2573 : }
2574 :
2575 : /*
2576 : * eqjoinsel_semi --- eqjoinsel for semi join
2577 : *
2578 : * (Also used for anti join, which we are supposed to estimate the same way.)
2579 : * Caller has ensured that vardata1 is the LHS variable.
2580 : * Unlike eqjoinsel_inner, we have to cope with opfuncoid being InvalidOid.
2581 : */
2582 : static double
2583 6838 : eqjoinsel_semi(Oid opfuncoid, Oid collation,
2584 : VariableStatData *vardata1, VariableStatData *vardata2,
2585 : double nd1, double nd2,
2586 : bool isdefault1, bool isdefault2,
2587 : AttStatsSlot *sslot1, AttStatsSlot *sslot2,
2588 : Form_pg_statistic stats1, Form_pg_statistic stats2,
2589 : bool have_mcvs1, bool have_mcvs2,
2590 : RelOptInfo *inner_rel)
2591 : {
2592 : double selec;
2593 :
2594 : /*
2595 : * We clamp nd2 to be not more than what we estimate the inner relation's
2596 : * size to be. This is intuitively somewhat reasonable since obviously
2597 : * there can't be more than that many distinct values coming from the
2598 : * inner rel. The reason for the asymmetry (ie, that we don't clamp nd1
2599 : * likewise) is that this is the only pathway by which restriction clauses
2600 : * applied to the inner rel will affect the join result size estimate,
2601 : * since set_joinrel_size_estimates will multiply SEMI/ANTI selectivity by
2602 : * only the outer rel's size. If we clamped nd1 we'd be double-counting
2603 : * the selectivity of outer-rel restrictions.
2604 : *
2605 : * We can apply this clamping both with respect to the base relation from
2606 : * which the join variable comes (if there is just one), and to the
2607 : * immediate inner input relation of the current join.
2608 : *
2609 : * If we clamp, we can treat nd2 as being a non-default estimate; it's not
2610 : * great, maybe, but it didn't come out of nowhere either. This is most
2611 : * helpful when the inner relation is empty and consequently has no stats.
2612 : */
2613 6838 : if (vardata2->rel)
2614 : {
2615 6838 : if (nd2 >= vardata2->rel->rows)
2616 : {
2617 5294 : nd2 = vardata2->rel->rows;
2618 5294 : isdefault2 = false;
2619 : }
2620 : }
2621 6838 : if (nd2 >= inner_rel->rows)
2622 : {
2623 5270 : nd2 = inner_rel->rows;
2624 5270 : isdefault2 = false;
2625 : }
2626 :
2627 6838 : if (have_mcvs1 && have_mcvs2 && OidIsValid(opfuncoid))
2628 486 : {
2629 : /*
2630 : * We have most-common-value lists for both relations. Run through
2631 : * the lists to see which MCVs actually join to each other with the
2632 : * given operator. This allows us to determine the exact join
2633 : * selectivity for the portion of the relations represented by the MCV
2634 : * lists. We still have to estimate for the remaining population, but
2635 : * in a skewed distribution this gives us a big leg up in accuracy.
2636 : */
2637 486 : LOCAL_FCINFO(fcinfo, 2);
2638 : FmgrInfo eqproc;
2639 : bool *hasmatch1;
2640 : bool *hasmatch2;
2641 486 : double nullfrac1 = stats1->stanullfrac;
2642 : double matchfreq1,
2643 : uncertainfrac,
2644 : uncertain;
2645 : int i,
2646 : nmatches,
2647 : clamped_nvalues2;
2648 :
2649 : /*
2650 : * The clamping above could have resulted in nd2 being less than
2651 : * sslot2->nvalues; in which case, we assume that precisely the nd2
2652 : * most common values in the relation will appear in the join input,
2653 : * and so compare to only the first nd2 members of the MCV list. Of
2654 : * course this is frequently wrong, but it's the best bet we can make.
2655 : */
2656 486 : clamped_nvalues2 = Min(sslot2->nvalues, nd2);
2657 :
2658 486 : fmgr_info(opfuncoid, &eqproc);
2659 :
2660 : /*
2661 : * Save a few cycles by setting up the fcinfo struct just once. Using
2662 : * FunctionCallInvoke directly also avoids failure if the eqproc
2663 : * returns NULL, though really equality functions should never do
2664 : * that.
2665 : */
2666 486 : InitFunctionCallInfoData(*fcinfo, &eqproc, 2, collation,
2667 : NULL, NULL);
2668 486 : fcinfo->args[0].isnull = false;
2669 486 : fcinfo->args[1].isnull = false;
2670 :
2671 486 : hasmatch1 = (bool *) palloc0(sslot1->nvalues * sizeof(bool));
2672 486 : hasmatch2 = (bool *) palloc0(clamped_nvalues2 * sizeof(bool));
2673 :
2674 : /*
2675 : * Note we assume that each MCV will match at most one member of the
2676 : * other MCV list. If the operator isn't really equality, there could
2677 : * be multiple matches --- but we don't look for them, both for speed
2678 : * and because the math wouldn't add up...
2679 : */
2680 486 : nmatches = 0;
2681 8624 : for (i = 0; i < sslot1->nvalues; i++)
2682 : {
2683 : int j;
2684 :
2685 8138 : fcinfo->args[0].value = sslot1->values[i];
2686 :
2687 297042 : for (j = 0; j < clamped_nvalues2; j++)
2688 : {
2689 : Datum fresult;
2690 :
2691 295454 : if (hasmatch2[j])
2692 200092 : continue;
2693 95362 : fcinfo->args[1].value = sslot2->values[j];
2694 95362 : fcinfo->isnull = false;
2695 95362 : fresult = FunctionCallInvoke(fcinfo);
2696 95362 : if (!fcinfo->isnull && DatumGetBool(fresult))
2697 : {
2698 6550 : hasmatch1[i] = hasmatch2[j] = true;
2699 6550 : nmatches++;
2700 6550 : break;
2701 : }
2702 : }
2703 : }
2704 : /* Sum up frequencies of matched MCVs */
2705 486 : matchfreq1 = 0.0;
2706 8624 : for (i = 0; i < sslot1->nvalues; i++)
2707 : {
2708 8138 : if (hasmatch1[i])
2709 6550 : matchfreq1 += sslot1->numbers[i];
2710 : }
2711 486 : CLAMP_PROBABILITY(matchfreq1);
2712 486 : pfree(hasmatch1);
2713 486 : pfree(hasmatch2);
2714 :
2715 : /*
2716 : * Now we need to estimate the fraction of relation 1 that has at
2717 : * least one join partner. We know for certain that the matched MCVs
2718 : * do, so that gives us a lower bound, but we're really in the dark
2719 : * about everything else. Our crude approach is: if nd1 <= nd2 then
2720 : * assume all non-null rel1 rows have join partners, else assume for
2721 : * the uncertain rows that a fraction nd2/nd1 have join partners. We
2722 : * can discount the known-matched MCVs from the distinct-values counts
2723 : * before doing the division.
2724 : *
2725 : * Crude as the above is, it's completely useless if we don't have
2726 : * reliable ndistinct values for both sides. Hence, if either nd1 or
2727 : * nd2 is default, punt and assume half of the uncertain rows have
2728 : * join partners.
2729 : */
2730 486 : if (!isdefault1 && !isdefault2)
2731 : {
2732 486 : nd1 -= nmatches;
2733 486 : nd2 -= nmatches;
2734 486 : if (nd1 <= nd2 || nd2 < 0)
2735 450 : uncertainfrac = 1.0;
2736 : else
2737 36 : uncertainfrac = nd2 / nd1;
2738 : }
2739 : else
2740 0 : uncertainfrac = 0.5;
2741 486 : uncertain = 1.0 - matchfreq1 - nullfrac1;
2742 486 : CLAMP_PROBABILITY(uncertain);
2743 486 : selec = matchfreq1 + uncertainfrac * uncertain;
2744 : }
2745 : else
2746 : {
2747 : /*
2748 : * Without MCV lists for both sides, we can only use the heuristic
2749 : * about nd1 vs nd2.
2750 : */
2751 6352 : double nullfrac1 = stats1 ? stats1->stanullfrac : 0.0;
2752 :
2753 6352 : if (!isdefault1 && !isdefault2)
2754 : {
2755 4642 : if (nd1 <= nd2 || nd2 < 0)
2756 2048 : selec = 1.0 - nullfrac1;
2757 : else
2758 2594 : selec = (nd2 / nd1) * (1.0 - nullfrac1);
2759 : }
2760 : else
2761 1710 : selec = 0.5 * (1.0 - nullfrac1);
2762 : }
2763 :
2764 6838 : return selec;
2765 : }
2766 :
2767 : /*
2768 : * neqjoinsel - Join selectivity of "!="
2769 : */
2770 : Datum
2771 2578 : neqjoinsel(PG_FUNCTION_ARGS)
2772 : {
2773 2578 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
2774 2578 : Oid operator = PG_GETARG_OID(1);
2775 2578 : List *args = (List *) PG_GETARG_POINTER(2);
2776 2578 : JoinType jointype = (JoinType) PG_GETARG_INT16(3);
2777 2578 : SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) PG_GETARG_POINTER(4);
2778 2578 : Oid collation = PG_GET_COLLATION();
2779 : float8 result;
2780 :
2781 2578 : if (jointype == JOIN_SEMI || jointype == JOIN_ANTI)
2782 868 : {
2783 : /*
2784 : * For semi-joins, if there is more than one distinct value in the RHS
2785 : * relation then every non-null LHS row must find a row to join since
2786 : * it can only be equal to one of them. We'll assume that there is
2787 : * always more than one distinct RHS value for the sake of stability,
2788 : * though in theory we could have special cases for empty RHS
2789 : * (selectivity = 0) and single-distinct-value RHS (selectivity =
2790 : * fraction of LHS that has the same value as the single RHS value).
2791 : *
2792 : * For anti-joins, if we use the same assumption that there is more
2793 : * than one distinct key in the RHS relation, then every non-null LHS
2794 : * row must be suppressed by the anti-join.
2795 : *
2796 : * So either way, the selectivity estimate should be 1 - nullfrac.
2797 : */
2798 : VariableStatData leftvar;
2799 : VariableStatData rightvar;
2800 : bool reversed;
2801 : HeapTuple statsTuple;
2802 : double nullfrac;
2803 :
2804 868 : get_join_variables(root, args, sjinfo, &leftvar, &rightvar, &reversed);
2805 868 : statsTuple = reversed ? rightvar.statsTuple : leftvar.statsTuple;
2806 868 : if (HeapTupleIsValid(statsTuple))
2807 704 : nullfrac = ((Form_pg_statistic) GETSTRUCT(statsTuple))->stanullfrac;
2808 : else
2809 164 : nullfrac = 0.0;
2810 868 : ReleaseVariableStats(leftvar);
2811 868 : ReleaseVariableStats(rightvar);
2812 :
2813 868 : result = 1.0 - nullfrac;
2814 : }
2815 : else
2816 : {
2817 : /*
2818 : * We want 1 - eqjoinsel() where the equality operator is the one
2819 : * associated with this != operator, that is, its negator.
2820 : */
2821 1710 : Oid eqop = get_negator(operator);
2822 :
2823 1710 : if (eqop)
2824 : {
2825 : result =
2826 1710 : DatumGetFloat8(DirectFunctionCall5Coll(eqjoinsel,
2827 : collation,
2828 : PointerGetDatum(root),
2829 : ObjectIdGetDatum(eqop),
2830 : PointerGetDatum(args),
2831 : Int16GetDatum(jointype),
2832 : PointerGetDatum(sjinfo)));
2833 : }
2834 : else
2835 : {
2836 : /* Use default selectivity (should we raise an error instead?) */
2837 0 : result = DEFAULT_EQ_SEL;
2838 : }
2839 1710 : result = 1.0 - result;
2840 : }
2841 :
2842 2578 : PG_RETURN_FLOAT8(result);
2843 : }
2844 :
2845 : /*
2846 : * scalarltjoinsel - Join selectivity of "<" for scalars
2847 : */
2848 : Datum
2849 300 : scalarltjoinsel(PG_FUNCTION_ARGS)
2850 : {
2851 300 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
2852 : }
2853 :
2854 : /*
2855 : * scalarlejoinsel - Join selectivity of "<=" for scalars
2856 : */
2857 : Datum
2858 190 : scalarlejoinsel(PG_FUNCTION_ARGS)
2859 : {
2860 190 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
2861 : }
2862 :
2863 : /*
2864 : * scalargtjoinsel - Join selectivity of ">" for scalars
2865 : */
2866 : Datum
2867 210 : scalargtjoinsel(PG_FUNCTION_ARGS)
2868 : {
2869 210 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
2870 : }
2871 :
2872 : /*
2873 : * scalargejoinsel - Join selectivity of ">=" for scalars
2874 : */
2875 : Datum
2876 178 : scalargejoinsel(PG_FUNCTION_ARGS)
2877 : {
2878 178 : PG_RETURN_FLOAT8(DEFAULT_INEQ_SEL);
2879 : }
2880 :
2881 :
2882 : /*
2883 : * mergejoinscansel - Scan selectivity of merge join.
2884 : *
2885 : * A merge join will stop as soon as it exhausts either input stream.
2886 : * Therefore, if we can estimate the ranges of both input variables,
2887 : * we can estimate how much of the input will actually be read. This
2888 : * can have a considerable impact on the cost when using indexscans.
2889 : *
2890 : * Also, we can estimate how much of each input has to be read before the
2891 : * first join pair is found, which will affect the join's startup time.
2892 : *
2893 : * clause should be a clause already known to be mergejoinable. opfamily,
2894 : * strategy, and nulls_first specify the sort ordering being used.
2895 : *
2896 : * The outputs are:
2897 : * *leftstart is set to the fraction of the left-hand variable expected
2898 : * to be scanned before the first join pair is found (0 to 1).
2899 : * *leftend is set to the fraction of the left-hand variable expected
2900 : * to be scanned before the join terminates (0 to 1).
2901 : * *rightstart, *rightend similarly for the right-hand variable.
2902 : */
2903 : void
2904 86732 : mergejoinscansel(PlannerInfo *root, Node *clause,
2905 : Oid opfamily, int strategy, bool nulls_first,
2906 : Selectivity *leftstart, Selectivity *leftend,
2907 : Selectivity *rightstart, Selectivity *rightend)
2908 : {
2909 : Node *left,
2910 : *right;
2911 : VariableStatData leftvar,
2912 : rightvar;
2913 : int op_strategy;
2914 : Oid op_lefttype;
2915 : Oid op_righttype;
2916 : Oid opno,
2917 : collation,
2918 : lsortop,
2919 : rsortop,
2920 : lstatop,
2921 : rstatop,
2922 : ltop,
2923 : leop,
2924 : revltop,
2925 : revleop;
2926 : bool isgt;
2927 : Datum leftmin,
2928 : leftmax,
2929 : rightmin,
2930 : rightmax;
2931 : double selec;
2932 :
2933 : /* Set default results if we can't figure anything out. */
2934 : /* XXX should default "start" fraction be a bit more than 0? */
2935 86732 : *leftstart = *rightstart = 0.0;
2936 86732 : *leftend = *rightend = 1.0;
2937 :
2938 : /* Deconstruct the merge clause */
2939 86732 : if (!is_opclause(clause))
2940 0 : return; /* shouldn't happen */
2941 86732 : opno = ((OpExpr *) clause)->opno;
2942 86732 : collation = ((OpExpr *) clause)->inputcollid;
2943 86732 : left = get_leftop((Expr *) clause);
2944 86732 : right = get_rightop((Expr *) clause);
2945 86732 : if (!right)
2946 0 : return; /* shouldn't happen */
2947 :
2948 : /* Look for stats for the inputs */
2949 86732 : examine_variable(root, left, 0, &leftvar);
2950 86732 : examine_variable(root, right, 0, &rightvar);
2951 :
2952 : /* Extract the operator's declared left/right datatypes */
2953 86732 : get_op_opfamily_properties(opno, opfamily, false,
2954 : &op_strategy,
2955 : &op_lefttype,
2956 : &op_righttype);
2957 : Assert(op_strategy == BTEqualStrategyNumber);
2958 :
2959 : /*
2960 : * Look up the various operators we need. If we don't find them all, it
2961 : * probably means the opfamily is broken, but we just fail silently.
2962 : *
2963 : * Note: we expect that pg_statistic histograms will be sorted by the '<'
2964 : * operator, regardless of which sort direction we are considering.
2965 : */
2966 86732 : switch (strategy)
2967 : {
2968 86684 : case BTLessStrategyNumber:
2969 86684 : isgt = false;
2970 86684 : if (op_lefttype == op_righttype)
2971 : {
2972 : /* easy case */
2973 85530 : ltop = get_opfamily_member(opfamily,
2974 : op_lefttype, op_righttype,
2975 : BTLessStrategyNumber);
2976 85530 : leop = get_opfamily_member(opfamily,
2977 : op_lefttype, op_righttype,
2978 : BTLessEqualStrategyNumber);
2979 85530 : lsortop = ltop;
2980 85530 : rsortop = ltop;
2981 85530 : lstatop = lsortop;
2982 85530 : rstatop = rsortop;
2983 85530 : revltop = ltop;
2984 85530 : revleop = leop;
2985 : }
2986 : else
2987 : {
2988 1154 : ltop = get_opfamily_member(opfamily,
2989 : op_lefttype, op_righttype,
2990 : BTLessStrategyNumber);
2991 1154 : leop = get_opfamily_member(opfamily,
2992 : op_lefttype, op_righttype,
2993 : BTLessEqualStrategyNumber);
2994 1154 : lsortop = get_opfamily_member(opfamily,
2995 : op_lefttype, op_lefttype,
2996 : BTLessStrategyNumber);
2997 1154 : rsortop = get_opfamily_member(opfamily,
2998 : op_righttype, op_righttype,
2999 : BTLessStrategyNumber);
3000 1154 : lstatop = lsortop;
3001 1154 : rstatop = rsortop;
3002 1154 : revltop = get_opfamily_member(opfamily,
3003 : op_righttype, op_lefttype,
3004 : BTLessStrategyNumber);
3005 1154 : revleop = get_opfamily_member(opfamily,
3006 : op_righttype, op_lefttype,
3007 : BTLessEqualStrategyNumber);
3008 : }
3009 86684 : break;
3010 48 : case BTGreaterStrategyNumber:
3011 : /* descending-order case */
3012 48 : isgt = true;
3013 48 : if (op_lefttype == op_righttype)
3014 : {
3015 : /* easy case */
3016 48 : ltop = get_opfamily_member(opfamily,
3017 : op_lefttype, op_righttype,
3018 : BTGreaterStrategyNumber);
3019 48 : leop = get_opfamily_member(opfamily,
3020 : op_lefttype, op_righttype,
3021 : BTGreaterEqualStrategyNumber);
3022 48 : lsortop = ltop;
3023 48 : rsortop = ltop;
3024 48 : lstatop = get_opfamily_member(opfamily,
3025 : op_lefttype, op_lefttype,
3026 : BTLessStrategyNumber);
3027 48 : rstatop = lstatop;
3028 48 : revltop = ltop;
3029 48 : revleop = leop;
3030 : }
3031 : else
3032 : {
3033 0 : ltop = get_opfamily_member(opfamily,
3034 : op_lefttype, op_righttype,
3035 : BTGreaterStrategyNumber);
3036 0 : leop = get_opfamily_member(opfamily,
3037 : op_lefttype, op_righttype,
3038 : BTGreaterEqualStrategyNumber);
3039 0 : lsortop = get_opfamily_member(opfamily,
3040 : op_lefttype, op_lefttype,
3041 : BTGreaterStrategyNumber);
3042 0 : rsortop = get_opfamily_member(opfamily,
3043 : op_righttype, op_righttype,
3044 : BTGreaterStrategyNumber);
3045 0 : lstatop = get_opfamily_member(opfamily,
3046 : op_lefttype, op_lefttype,
3047 : BTLessStrategyNumber);
3048 0 : rstatop = get_opfamily_member(opfamily,
3049 : op_righttype, op_righttype,
3050 : BTLessStrategyNumber);
3051 0 : revltop = get_opfamily_member(opfamily,
3052 : op_righttype, op_lefttype,
3053 : BTGreaterStrategyNumber);
3054 0 : revleop = get_opfamily_member(opfamily,
3055 : op_righttype, op_lefttype,
3056 : BTGreaterEqualStrategyNumber);
3057 : }
3058 48 : break;
3059 0 : default:
3060 0 : goto fail; /* shouldn't get here */
3061 : }
3062 :
3063 86732 : if (!OidIsValid(lsortop) ||
3064 86732 : !OidIsValid(rsortop) ||
3065 86732 : !OidIsValid(lstatop) ||
3066 86732 : !OidIsValid(rstatop) ||
3067 86720 : !OidIsValid(ltop) ||
3068 86720 : !OidIsValid(leop) ||
3069 86720 : !OidIsValid(revltop) ||
3070 : !OidIsValid(revleop))
3071 12 : goto fail; /* insufficient info in catalogs */
3072 :
3073 : /* Try to get ranges of both inputs */
3074 86720 : if (!isgt)
3075 : {
3076 86672 : if (!get_variable_range(root, &leftvar, lstatop, collation,
3077 : &leftmin, &leftmax))
3078 30288 : goto fail; /* no range available from stats */
3079 56384 : if (!get_variable_range(root, &rightvar, rstatop, collation,
3080 : &rightmin, &rightmax))
3081 11098 : goto fail; /* no range available from stats */
3082 : }
3083 : else
3084 : {
3085 : /* need to swap the max and min */
3086 48 : if (!get_variable_range(root, &leftvar, lstatop, collation,
3087 : &leftmax, &leftmin))
3088 30 : goto fail; /* no range available from stats */
3089 18 : if (!get_variable_range(root, &rightvar, rstatop, collation,
3090 : &rightmax, &rightmin))
3091 0 : goto fail; /* no range available from stats */
3092 : }
3093 :
3094 : /*
3095 : * Now, the fraction of the left variable that will be scanned is the
3096 : * fraction that's <= the right-side maximum value. But only believe
3097 : * non-default estimates, else stick with our 1.0.
3098 : */
3099 45304 : selec = scalarineqsel(root, leop, isgt, true, collation, &leftvar,
3100 : rightmax, op_righttype);
3101 45304 : if (selec != DEFAULT_INEQ_SEL)
3102 45260 : *leftend = selec;
3103 :
3104 : /* And similarly for the right variable. */
3105 45304 : selec = scalarineqsel(root, revleop, isgt, true, collation, &rightvar,
3106 : leftmax, op_lefttype);
3107 45304 : if (selec != DEFAULT_INEQ_SEL)
3108 45298 : *rightend = selec;
3109 :
3110 : /*
3111 : * Only one of the two "end" fractions can really be less than 1.0;
3112 : * believe the smaller estimate and reset the other one to exactly 1.0. If
3113 : * we get exactly equal estimates (as can easily happen with self-joins),
3114 : * believe neither.
3115 : */
3116 45304 : if (*leftend > *rightend)
3117 23706 : *leftend = 1.0;
3118 21598 : else if (*leftend < *rightend)
3119 16054 : *rightend = 1.0;
3120 : else
3121 5544 : *leftend = *rightend = 1.0;
3122 :
3123 : /*
3124 : * Also, the fraction of the left variable that will be scanned before the
3125 : * first join pair is found is the fraction that's < the right-side
3126 : * minimum value. But only believe non-default estimates, else stick with
3127 : * our own default.
3128 : */
3129 45304 : selec = scalarineqsel(root, ltop, isgt, false, collation, &leftvar,
3130 : rightmin, op_righttype);
3131 45304 : if (selec != DEFAULT_INEQ_SEL)
3132 45304 : *leftstart = selec;
3133 :
3134 : /* And similarly for the right variable. */
3135 45304 : selec = scalarineqsel(root, revltop, isgt, false, collation, &rightvar,
3136 : leftmin, op_lefttype);
3137 45304 : if (selec != DEFAULT_INEQ_SEL)
3138 45304 : *rightstart = selec;
3139 :
3140 : /*
3141 : * Only one of the two "start" fractions can really be more than zero;
3142 : * believe the larger estimate and reset the other one to exactly 0.0. If
3143 : * we get exactly equal estimates (as can easily happen with self-joins),
3144 : * believe neither.
3145 : */
3146 45304 : if (*leftstart < *rightstart)
3147 11054 : *leftstart = 0.0;
3148 34250 : else if (*leftstart > *rightstart)
3149 21782 : *rightstart = 0.0;
3150 : else
3151 12468 : *leftstart = *rightstart = 0.0;
3152 :
3153 : /*
3154 : * If the sort order is nulls-first, we're going to have to skip over any
3155 : * nulls too. These would not have been counted by scalarineqsel, and we
3156 : * can safely add in this fraction regardless of whether we believe
3157 : * scalarineqsel's results or not. But be sure to clamp the sum to 1.0!
3158 : */
3159 45304 : if (nulls_first)
3160 : {
3161 : Form_pg_statistic stats;
3162 :
3163 18 : if (HeapTupleIsValid(leftvar.statsTuple))
3164 : {
3165 18 : stats = (Form_pg_statistic) GETSTRUCT(leftvar.statsTuple);
3166 18 : *leftstart += stats->stanullfrac;
3167 18 : CLAMP_PROBABILITY(*leftstart);
3168 18 : *leftend += stats->stanullfrac;
3169 18 : CLAMP_PROBABILITY(*leftend);
3170 : }
3171 18 : if (HeapTupleIsValid(rightvar.statsTuple))
3172 : {
3173 18 : stats = (Form_pg_statistic) GETSTRUCT(rightvar.statsTuple);
3174 18 : *rightstart += stats->stanullfrac;
3175 18 : CLAMP_PROBABILITY(*rightstart);
3176 18 : *rightend += stats->stanullfrac;
3177 18 : CLAMP_PROBABILITY(*rightend);
3178 : }
3179 : }
3180 :
3181 : /* Disbelieve start >= end, just in case that can happen */
3182 45304 : if (*leftstart >= *leftend)
3183 : {
3184 366 : *leftstart = 0.0;
3185 366 : *leftend = 1.0;
3186 : }
3187 45304 : if (*rightstart >= *rightend)
3188 : {
3189 556 : *rightstart = 0.0;
3190 556 : *rightend = 1.0;
3191 : }
3192 :
3193 44748 : fail:
3194 86732 : ReleaseVariableStats(leftvar);
3195 86732 : ReleaseVariableStats(rightvar);
3196 : }
3197 :
3198 :
3199 : /*
3200 : * matchingsel -- generic matching-operator selectivity support
3201 : *
3202 : * Use these for any operators that (a) are on data types for which we collect
3203 : * standard statistics, and (b) have behavior for which the default estimate
3204 : * (twice DEFAULT_EQ_SEL) is sane. Typically that is good for match-like
3205 : * operators.
3206 : */
3207 :
3208 : Datum
3209 1106 : matchingsel(PG_FUNCTION_ARGS)
3210 : {
3211 1106 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
3212 1106 : Oid operator = PG_GETARG_OID(1);
3213 1106 : List *args = (List *) PG_GETARG_POINTER(2);
3214 1106 : int varRelid = PG_GETARG_INT32(3);
3215 1106 : Oid collation = PG_GET_COLLATION();
3216 : double selec;
3217 :
3218 : /* Use generic restriction selectivity logic. */
3219 1106 : selec = generic_restriction_selectivity(root, operator, collation,
3220 : args, varRelid,
3221 : DEFAULT_MATCHING_SEL);
3222 :
3223 1106 : PG_RETURN_FLOAT8((float8) selec);
3224 : }
3225 :
3226 : Datum
3227 6 : matchingjoinsel(PG_FUNCTION_ARGS)
3228 : {
3229 : /* Just punt, for the moment. */
3230 6 : PG_RETURN_FLOAT8(DEFAULT_MATCHING_SEL);
3231 : }
3232 :
3233 :
3234 : /*
3235 : * Helper routine for estimate_num_groups: add an item to a list of
3236 : * GroupVarInfos, but only if it's not known equal to any of the existing
3237 : * entries.
3238 : */
3239 : typedef struct
3240 : {
3241 : Node *var; /* might be an expression, not just a Var */
3242 : RelOptInfo *rel; /* relation it belongs to */
3243 : double ndistinct; /* # distinct values */
3244 : bool isdefault; /* true if DEFAULT_NUM_DISTINCT was used */
3245 : } GroupVarInfo;
3246 :
3247 : static List *
3248 1469440 : add_unique_group_var(PlannerInfo *root, List *varinfos,
3249 : Node *var, VariableStatData *vardata)
3250 : {
3251 : GroupVarInfo *varinfo;
3252 : double ndistinct;
3253 : bool isdefault;
3254 : ListCell *lc;
3255 :
3256 1469440 : ndistinct = get_variable_numdistinct(vardata, &isdefault);
3257 :
3258 1780758 : foreach(lc, varinfos)
3259 : {
3260 315198 : varinfo = (GroupVarInfo *) lfirst(lc);
3261 :
3262 : /* Drop exact duplicates */
3263 315198 : if (equal(var, varinfo->var))
3264 3880 : return varinfos;
3265 :
3266 : /*
3267 : * Drop known-equal vars, but only if they belong to different
3268 : * relations (see comments for estimate_num_groups)
3269 : */
3270 424972 : if (vardata->rel != varinfo->rel &&
3271 113324 : exprs_known_equal(root, var, varinfo->var))
3272 : {
3273 334 : if (varinfo->ndistinct <= ndistinct)
3274 : {
3275 : /* Keep older item, forget new one */
3276 330 : return varinfos;
3277 : }
3278 : else
3279 : {
3280 : /* Delete the older item */
3281 4 : varinfos = foreach_delete_current(varinfos, lc);
3282 : }
3283 : }
3284 : }
3285 :
3286 1465560 : varinfo = (GroupVarInfo *) palloc(sizeof(GroupVarInfo));
3287 :
3288 1465560 : varinfo->var = var;
3289 1465560 : varinfo->rel = vardata->rel;
3290 1465560 : varinfo->ndistinct = ndistinct;
3291 1465560 : varinfo->isdefault = isdefault;
3292 1465560 : varinfos = lappend(varinfos, varinfo);
3293 1465560 : return varinfos;
3294 : }
3295 :
3296 : /*
3297 : * estimate_num_groups - Estimate number of groups in a grouped query
3298 : *
3299 : * Given a query having a GROUP BY clause, estimate how many groups there
3300 : * will be --- ie, the number of distinct combinations of the GROUP BY
3301 : * expressions.
3302 : *
3303 : * This routine is also used to estimate the number of rows emitted by
3304 : * a DISTINCT filtering step; that is an isomorphic problem. (Note:
3305 : * actually, we only use it for DISTINCT when there's no grouping or
3306 : * aggregation ahead of the DISTINCT.)
3307 : *
3308 : * Inputs:
3309 : * root - the query
3310 : * groupExprs - list of expressions being grouped by
3311 : * input_rows - number of rows estimated to arrive at the group/unique
3312 : * filter step
3313 : * pgset - NULL, or a List** pointing to a grouping set to filter the
3314 : * groupExprs against
3315 : *
3316 : * Outputs:
3317 : * estinfo - When passed as non-NULL, the function will set bits in the
3318 : * "flags" field in order to provide callers with additional information
3319 : * about the estimation. Currently, we only set the SELFLAG_USED_DEFAULT
3320 : * bit if we used any default values in the estimation.
3321 : *
3322 : * Given the lack of any cross-correlation statistics in the system, it's
3323 : * impossible to do anything really trustworthy with GROUP BY conditions
3324 : * involving multiple Vars. We should however avoid assuming the worst
3325 : * case (all possible cross-product terms actually appear as groups) since
3326 : * very often the grouped-by Vars are highly correlated. Our current approach
3327 : * is as follows:
3328 : * 1. Expressions yielding boolean are assumed to contribute two groups,
3329 : * independently of their content, and are ignored in the subsequent
3330 : * steps. This is mainly because tests like "col IS NULL" break the
3331 : * heuristic used in step 2 especially badly.
3332 : * 2. Reduce the given expressions to a list of unique Vars used. For
3333 : * example, GROUP BY a, a + b is treated the same as GROUP BY a, b.
3334 : * It is clearly correct not to count the same Var more than once.
3335 : * It is also reasonable to treat f(x) the same as x: f() cannot
3336 : * increase the number of distinct values (unless it is volatile,
3337 : * which we consider unlikely for grouping), but it probably won't
3338 : * reduce the number of distinct values much either.
3339 : * As a special case, if a GROUP BY expression can be matched to an
3340 : * expressional index for which we have statistics, then we treat the
3341 : * whole expression as though it were just a Var.
3342 : * 3. If the list contains Vars of different relations that are known equal
3343 : * due to equivalence classes, then drop all but one of the Vars from each
3344 : * known-equal set, keeping the one with smallest estimated # of values
3345 : * (since the extra values of the others can't appear in joined rows).
3346 : * Note the reason we only consider Vars of different relations is that
3347 : * if we considered ones of the same rel, we'd be double-counting the
3348 : * restriction selectivity of the equality in the next step.
3349 : * 4. For Vars within a single source rel, we multiply together the numbers
3350 : * of values, clamp to the number of rows in the rel (divided by 10 if
3351 : * more than one Var), and then multiply by a factor based on the
3352 : * selectivity of the restriction clauses for that rel. When there's
3353 : * more than one Var, the initial product is probably too high (it's the
3354 : * worst case) but clamping to a fraction of the rel's rows seems to be a
3355 : * helpful heuristic for not letting the estimate get out of hand. (The
3356 : * factor of 10 is derived from pre-Postgres-7.4 practice.) The factor
3357 : * we multiply by to adjust for the restriction selectivity assumes that
3358 : * the restriction clauses are independent of the grouping, which may not
3359 : * be a valid assumption, but it's hard to do better.
3360 : * 5. If there are Vars from multiple rels, we repeat step 4 for each such
3361 : * rel, and multiply the results together.
3362 : * Note that rels not containing grouped Vars are ignored completely, as are
3363 : * join clauses. Such rels cannot increase the number of groups, and we
3364 : * assume such clauses do not reduce the number either (somewhat bogus,
3365 : * but we don't have the info to do better).
3366 : */
3367 : double
3368 170344 : estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
3369 : List **pgset, EstimationInfo *estinfo)
3370 : {
3371 170344 : return estimate_num_groups_incremental(root, groupExprs,
3372 : input_rows, pgset, estinfo,
3373 : NULL, 0);
3374 : }
3375 :
3376 : /*
3377 : * estimate_num_groups_incremental
3378 : * An estimate_num_groups variant, optimized for cases that are adding the
3379 : * expressions incrementally (e.g. one by one).
3380 : */
3381 : double
3382 1444270 : estimate_num_groups_incremental(PlannerInfo *root, List *groupExprs,
3383 : double input_rows,
3384 : List **pgset, EstimationInfo *estinfo,
3385 : List **cache_varinfos, int prevNExprs)
3386 : {
3387 1444270 : List *varinfos = (cache_varinfos) ? *cache_varinfos : NIL;
3388 1444270 : double srf_multiplier = 1.0;
3389 : double numdistinct;
3390 : ListCell *l;
3391 : int i,
3392 : j;
3393 :
3394 : /* Zero the estinfo output parameter, if non-NULL */
3395 1444270 : if (estinfo != NULL)
3396 152166 : memset(estinfo, 0, sizeof(EstimationInfo));
3397 :
3398 : /*
3399 : * We don't ever want to return an estimate of zero groups, as that tends
3400 : * to lead to division-by-zero and other unpleasantness. The input_rows
3401 : * estimate is usually already at least 1, but clamp it just in case it
3402 : * isn't.
3403 : */
3404 1444270 : input_rows = clamp_row_est(input_rows);
3405 :
3406 : /*
3407 : * If no grouping columns, there's exactly one group. (This can't happen
3408 : * for normal cases with GROUP BY or DISTINCT, but it is possible for
3409 : * corner cases with set operations.)
3410 : */
3411 1444270 : if (groupExprs == NIL || (pgset && list_length(*pgset) < 1))
3412 730 : return 1.0;
3413 :
3414 : /*
3415 : * Count groups derived from boolean grouping expressions. For other
3416 : * expressions, find the unique Vars used, treating an expression as a Var
3417 : * if we can find stats for it. For each one, record the statistical
3418 : * estimate of number of distinct values (total in its table, without
3419 : * regard for filtering).
3420 : */
3421 1443540 : numdistinct = 1.0;
3422 :
3423 1443540 : i = j = 0;
3424 3182686 : foreach(l, groupExprs)
3425 : {
3426 1739244 : Node *groupexpr = (Node *) lfirst(l);
3427 : double this_srf_multiplier;
3428 : VariableStatData vardata;
3429 : List *varshere;
3430 : ListCell *l2;
3431 :
3432 : /* was done on previous call */
3433 1739244 : if (cache_varinfos && j++ < prevNExprs)
3434 : {
3435 272238 : if (pgset)
3436 0 : i++; /* to keep in sync with lines below */
3437 1469450 : continue;
3438 : }
3439 :
3440 : /* is expression in this grouping set? */
3441 1467006 : if (pgset && !list_member_int(*pgset, i++))
3442 728 : continue;
3443 :
3444 : /*
3445 : * Set-returning functions in grouping columns are a bit problematic.
3446 : * The code below will effectively ignore their SRF nature and come up
3447 : * with a numdistinct estimate as though they were scalar functions.
3448 : * We compensate by scaling up the end result by the largest SRF
3449 : * rowcount estimate. (This will be an overestimate if the SRF
3450 : * produces multiple copies of any output value, but it seems best to
3451 : * assume the SRF's outputs are distinct. In any case, it's probably
3452 : * pointless to worry too much about this without much better
3453 : * estimates for SRF output rowcounts than we have today.)
3454 : */
3455 1466278 : this_srf_multiplier = expression_returns_set_rows(root, groupexpr);
3456 1466278 : if (srf_multiplier < this_srf_multiplier)
3457 316 : srf_multiplier = this_srf_multiplier;
3458 :
3459 : /* Short-circuit for expressions returning boolean */
3460 1466278 : if (exprType(groupexpr) == BOOLOID)
3461 : {
3462 3554 : numdistinct *= 2.0;
3463 3554 : continue;
3464 : }
3465 :
3466 : /*
3467 : * If examine_variable is able to deduce anything about the GROUP BY
3468 : * expression, treat it as a single variable even if it's really more
3469 : * complicated.
3470 : *
3471 : * XXX This has the consequence that if there's a statistics object on
3472 : * the expression, we don't split it into individual Vars. This
3473 : * affects our selection of statistics in
3474 : * estimate_multivariate_ndistinct, because it's probably better to
3475 : * use more accurate estimate for each expression and treat them as
3476 : * independent, than to combine estimates for the extracted variables
3477 : * when we don't know how that relates to the expressions.
3478 : */
3479 1462724 : examine_variable(root, groupexpr, 0, &vardata);
3480 1462724 : if (HeapTupleIsValid(vardata.statsTuple) || vardata.isunique)
3481 : {
3482 1189862 : varinfos = add_unique_group_var(root, varinfos,
3483 : groupexpr, &vardata);
3484 1189862 : ReleaseVariableStats(vardata);
3485 1189862 : continue;
3486 : }
3487 272862 : ReleaseVariableStats(vardata);
3488 :
3489 : /*
3490 : * Else pull out the component Vars. Handle PlaceHolderVars by
3491 : * recursing into their arguments (effectively assuming that the
3492 : * PlaceHolderVar doesn't change the number of groups, which boils
3493 : * down to ignoring the possible addition of nulls to the result set).
3494 : */
3495 272862 : varshere = pull_var_clause(groupexpr,
3496 : PVC_RECURSE_AGGREGATES |
3497 : PVC_RECURSE_WINDOWFUNCS |
3498 : PVC_RECURSE_PLACEHOLDERS);
3499 :
3500 : /*
3501 : * If we find any variable-free GROUP BY item, then either it is a
3502 : * constant (and we can ignore it) or it contains a volatile function;
3503 : * in the latter case we punt and assume that each input row will
3504 : * yield a distinct group.
3505 : */
3506 272862 : if (varshere == NIL)
3507 : {
3508 3166 : if (contain_volatile_functions(groupexpr))
3509 : {
3510 98 : if (cache_varinfos)
3511 68 : *cache_varinfos = varinfos;
3512 98 : return input_rows;
3513 : }
3514 3068 : continue;
3515 : }
3516 :
3517 : /*
3518 : * Else add variables to varinfos list
3519 : */
3520 549274 : foreach(l2, varshere)
3521 : {
3522 279578 : Node *var = (Node *) lfirst(l2);
3523 :
3524 279578 : examine_variable(root, var, 0, &vardata);
3525 279578 : varinfos = add_unique_group_var(root, varinfos, var, &vardata);
3526 279578 : ReleaseVariableStats(vardata);
3527 : }
3528 : }
3529 :
3530 1443442 : if (cache_varinfos)
3531 1273858 : *cache_varinfos = varinfos;
3532 :
3533 : /*
3534 : * If now no Vars, we must have an all-constant or all-boolean GROUP BY
3535 : * list.
3536 : */
3537 1443442 : if (varinfos == NIL)
3538 : {
3539 : /* Apply SRF multiplier as we would do in the long path */
3540 5652 : numdistinct *= srf_multiplier;
3541 : /* Round off */
3542 5652 : numdistinct = ceil(numdistinct);
3543 : /* Guard against out-of-range answers */
3544 5652 : if (numdistinct > input_rows)
3545 0 : numdistinct = input_rows;
3546 5652 : if (numdistinct < 1.0)
3547 0 : numdistinct = 1.0;
3548 5652 : return numdistinct;
3549 : }
3550 :
3551 : /*
3552 : * Group Vars by relation and estimate total numdistinct.
3553 : *
3554 : * For each iteration of the outer loop, we process the frontmost Var in
3555 : * varinfos, plus all other Vars in the same relation. We remove these
3556 : * Vars from the newvarinfos list for the next iteration. This is the
3557 : * easiest way to group Vars of same rel together.
3558 : */
3559 : do
3560 : {
3561 1541776 : GroupVarInfo *varinfo1 = (GroupVarInfo *) linitial(varinfos);
3562 1541776 : RelOptInfo *rel = varinfo1->rel;
3563 1541776 : double reldistinct = 1;
3564 1541776 : double relmaxndistinct = reldistinct;
3565 1541776 : int relvarcount = 0;
3566 1541776 : List *newvarinfos = NIL;
3567 1541776 : List *relvarinfos = NIL;
3568 :
3569 : /*
3570 : * Split the list of varinfos in two - one for the current rel, one
3571 : * for remaining Vars on other rels.
3572 : */
3573 1541776 : relvarinfos = lappend(relvarinfos, varinfo1);
3574 1854780 : for_each_from(l, varinfos, 1)
3575 : {
3576 313004 : GroupVarInfo *varinfo2 = (GroupVarInfo *) lfirst(l);
3577 :
3578 313004 : if (varinfo2->rel == varinfo1->rel)
3579 : {
3580 : /* varinfos on current rel */
3581 195074 : relvarinfos = lappend(relvarinfos, varinfo2);
3582 : }
3583 : else
3584 : {
3585 : /* not time to process varinfo2 yet */
3586 117930 : newvarinfos = lappend(newvarinfos, varinfo2);
3587 : }
3588 : }
3589 :
3590 : /*
3591 : * Get the numdistinct estimate for the Vars of this rel. We
3592 : * iteratively search for multivariate n-distinct with maximum number
3593 : * of vars; assuming that each var group is independent of the others,
3594 : * we multiply them together. Any remaining relvarinfos after no more
3595 : * multivariate matches are found are assumed independent too, so
3596 : * their individual ndistinct estimates are multiplied also.
3597 : *
3598 : * While iterating, count how many separate numdistinct values we
3599 : * apply. We apply a fudge factor below, but only if we multiplied
3600 : * more than one such values.
3601 : */
3602 3086906 : while (relvarinfos)
3603 : {
3604 : double mvndistinct;
3605 :
3606 1545130 : if (estimate_multivariate_ndistinct(root, rel, &relvarinfos,
3607 : &mvndistinct))
3608 : {
3609 6858 : reldistinct *= mvndistinct;
3610 6858 : if (relmaxndistinct < mvndistinct)
3611 6534 : relmaxndistinct = mvndistinct;
3612 6858 : relvarcount++;
3613 : }
3614 : else
3615 : {
3616 3260566 : foreach(l, relvarinfos)
3617 : {
3618 1722294 : GroupVarInfo *varinfo2 = (GroupVarInfo *) lfirst(l);
3619 :
3620 1722294 : reldistinct *= varinfo2->ndistinct;
3621 1722294 : if (relmaxndistinct < varinfo2->ndistinct)
3622 1522922 : relmaxndistinct = varinfo2->ndistinct;
3623 1722294 : relvarcount++;
3624 :
3625 : /*
3626 : * When varinfo2's isdefault is set then we'd better set
3627 : * the SELFLAG_USED_DEFAULT bit in the EstimationInfo.
3628 : */
3629 1722294 : if (estinfo != NULL && varinfo2->isdefault)
3630 10724 : estinfo->flags |= SELFLAG_USED_DEFAULT;
3631 : }
3632 :
3633 : /* we're done with this relation */
3634 1538272 : relvarinfos = NIL;
3635 : }
3636 : }
3637 :
3638 : /*
3639 : * Sanity check --- don't divide by zero if empty relation.
3640 : */
3641 : Assert(IS_SIMPLE_REL(rel));
3642 1541776 : if (rel->tuples > 0)
3643 : {
3644 : /*
3645 : * Clamp to size of rel, or size of rel / 10 if multiple Vars. The
3646 : * fudge factor is because the Vars are probably correlated but we
3647 : * don't know by how much. We should never clamp to less than the
3648 : * largest ndistinct value for any of the Vars, though, since
3649 : * there will surely be at least that many groups.
3650 : */
3651 1502898 : double clamp = rel->tuples;
3652 :
3653 1502898 : if (relvarcount > 1)
3654 : {
3655 157744 : clamp *= 0.1;
3656 157744 : if (clamp < relmaxndistinct)
3657 : {
3658 123240 : clamp = relmaxndistinct;
3659 : /* for sanity in case some ndistinct is too large: */
3660 123240 : if (clamp > rel->tuples)
3661 14752 : clamp = rel->tuples;
3662 : }
3663 : }
3664 1502898 : if (reldistinct > clamp)
3665 135108 : reldistinct = clamp;
3666 :
3667 : /*
3668 : * Update the estimate based on the restriction selectivity,
3669 : * guarding against division by zero when reldistinct is zero.
3670 : * Also skip this if we know that we are returning all rows.
3671 : */
3672 1502898 : if (reldistinct > 0 && rel->rows < rel->tuples)
3673 : {
3674 : /*
3675 : * Given a table containing N rows with n distinct values in a
3676 : * uniform distribution, if we select p rows at random then
3677 : * the expected number of distinct values selected is
3678 : *
3679 : * n * (1 - product((N-N/n-i)/(N-i), i=0..p-1))
3680 : *
3681 : * = n * (1 - (N-N/n)! / (N-N/n-p)! * (N-p)! / N!)
3682 : *
3683 : * See "Approximating block accesses in database
3684 : * organizations", S. B. Yao, Communications of the ACM,
3685 : * Volume 20 Issue 4, April 1977 Pages 260-261.
3686 : *
3687 : * Alternatively, re-arranging the terms from the factorials,
3688 : * this may be written as
3689 : *
3690 : * n * (1 - product((N-p-i)/(N-i), i=0..N/n-1))
3691 : *
3692 : * This form of the formula is more efficient to compute in
3693 : * the common case where p is larger than N/n. Additionally,
3694 : * as pointed out by Dell'Era, if i << N for all terms in the
3695 : * product, it can be approximated by
3696 : *
3697 : * n * (1 - ((N-p)/N)^(N/n))
3698 : *
3699 : * See "Expected distinct values when selecting from a bag
3700 : * without replacement", Alberto Dell'Era,
3701 : * http://www.adellera.it/investigations/distinct_balls/.
3702 : *
3703 : * The condition i << N is equivalent to n >> 1, so this is a
3704 : * good approximation when the number of distinct values in
3705 : * the table is large. It turns out that this formula also
3706 : * works well even when n is small.
3707 : */
3708 557442 : reldistinct *=
3709 557442 : (1 - pow((rel->tuples - rel->rows) / rel->tuples,
3710 557442 : rel->tuples / reldistinct));
3711 : }
3712 1502898 : reldistinct = clamp_row_est(reldistinct);
3713 :
3714 : /*
3715 : * Update estimate of total distinct groups.
3716 : */
3717 1502898 : numdistinct *= reldistinct;
3718 : }
3719 :
3720 1541776 : varinfos = newvarinfos;
3721 1541776 : } while (varinfos != NIL);
3722 :
3723 : /* Now we can account for the effects of any SRFs */
3724 1437790 : numdistinct *= srf_multiplier;
3725 :
3726 : /* Round off */
3727 1437790 : numdistinct = ceil(numdistinct);
3728 :
3729 : /* Guard against out-of-range answers */
3730 1437790 : if (numdistinct > input_rows)
3731 405840 : numdistinct = input_rows;
3732 1437790 : if (numdistinct < 1.0)
3733 0 : numdistinct = 1.0;
3734 :
3735 1437790 : return numdistinct;
3736 : }
3737 :
3738 : /*
3739 : * Estimate hash bucket statistics when the specified expression is used
3740 : * as a hash key for the given number of buckets.
3741 : *
3742 : * This attempts to determine two values:
3743 : *
3744 : * 1. The frequency of the most common value of the expression (returns
3745 : * zero into *mcv_freq if we can't get that).
3746 : *
3747 : * 2. The "bucketsize fraction", ie, average number of entries in a bucket
3748 : * divided by total tuples in relation.
3749 : *
3750 : * XXX This is really pretty bogus since we're effectively assuming that the
3751 : * distribution of hash keys will be the same after applying restriction
3752 : * clauses as it was in the underlying relation. However, we are not nearly
3753 : * smart enough to figure out how the restrict clauses might change the
3754 : * distribution, so this will have to do for now.
3755 : *
3756 : * We are passed the number of buckets the executor will use for the given
3757 : * input relation. If the data were perfectly distributed, with the same
3758 : * number of tuples going into each available bucket, then the bucketsize
3759 : * fraction would be 1/nbuckets. But this happy state of affairs will occur
3760 : * only if (a) there are at least nbuckets distinct data values, and (b)
3761 : * we have a not-too-skewed data distribution. Otherwise the buckets will
3762 : * be nonuniformly occupied. If the other relation in the join has a key
3763 : * distribution similar to this one's, then the most-loaded buckets are
3764 : * exactly those that will be probed most often. Therefore, the "average"
3765 : * bucket size for costing purposes should really be taken as something close
3766 : * to the "worst case" bucket size. We try to estimate this by adjusting the
3767 : * fraction if there are too few distinct data values, and then scaling up
3768 : * by the ratio of the most common value's frequency to the average frequency.
3769 : *
3770 : * If no statistics are available, use a default estimate of 0.1. This will
3771 : * discourage use of a hash rather strongly if the inner relation is large,
3772 : * which is what we want. We do not want to hash unless we know that the
3773 : * inner rel is well-dispersed (or the alternatives seem much worse).
3774 : *
3775 : * The caller should also check that the mcv_freq is not so large that the
3776 : * most common value would by itself require an impractically large bucket.
3777 : * In a hash join, the executor can split buckets if they get too big, but
3778 : * obviously that doesn't help for a bucket that contains many duplicates of
3779 : * the same value.
3780 : */
3781 : void
3782 116646 : estimate_hash_bucket_stats(PlannerInfo *root, Node *hashkey, double nbuckets,
3783 : Selectivity *mcv_freq,
3784 : Selectivity *bucketsize_frac)
3785 : {
3786 : VariableStatData vardata;
3787 : double estfract,
3788 : ndistinct,
3789 : stanullfrac,
3790 : avgfreq;
3791 : bool isdefault;
3792 : AttStatsSlot sslot;
3793 :
3794 116646 : examine_variable(root, hashkey, 0, &vardata);
3795 :
3796 : /* Look up the frequency of the most common value, if available */
3797 116646 : *mcv_freq = 0.0;
3798 :
3799 116646 : if (HeapTupleIsValid(vardata.statsTuple))
3800 : {
3801 74204 : if (get_attstatsslot(&sslot, vardata.statsTuple,
3802 : STATISTIC_KIND_MCV, InvalidOid,
3803 : ATTSTATSSLOT_NUMBERS))
3804 : {
3805 : /*
3806 : * The first MCV stat is for the most common value.
3807 : */
3808 34014 : if (sslot.nnumbers > 0)
3809 34014 : *mcv_freq = sslot.numbers[0];
3810 34014 : free_attstatsslot(&sslot);
3811 : }
3812 : }
3813 :
3814 : /* Get number of distinct values */
3815 116646 : ndistinct = get_variable_numdistinct(&vardata, &isdefault);
3816 :
3817 : /*
3818 : * If ndistinct isn't real, punt. We normally return 0.1, but if the
3819 : * mcv_freq is known to be even higher than that, use it instead.
3820 : */
3821 116646 : if (isdefault)
3822 : {
3823 18492 : *bucketsize_frac = (Selectivity) Max(0.1, *mcv_freq);
3824 18492 : ReleaseVariableStats(vardata);
3825 18492 : return;
3826 : }
3827 :
3828 : /* Get fraction that are null */
3829 98154 : if (HeapTupleIsValid(vardata.statsTuple))
3830 : {
3831 : Form_pg_statistic stats;
3832 :
3833 74186 : stats = (Form_pg_statistic) GETSTRUCT(vardata.statsTuple);
3834 74186 : stanullfrac = stats->stanullfrac;
3835 : }
3836 : else
3837 23968 : stanullfrac = 0.0;
3838 :
3839 : /* Compute avg freq of all distinct data values in raw relation */
3840 98154 : avgfreq = (1.0 - stanullfrac) / ndistinct;
3841 :
3842 : /*
3843 : * Adjust ndistinct to account for restriction clauses. Observe we are
3844 : * assuming that the data distribution is affected uniformly by the
3845 : * restriction clauses!
3846 : *
3847 : * XXX Possibly better way, but much more expensive: multiply by
3848 : * selectivity of rel's restriction clauses that mention the target Var.
3849 : */
3850 98154 : if (vardata.rel && vardata.rel->tuples > 0)
3851 : {
3852 98140 : ndistinct *= vardata.rel->rows / vardata.rel->tuples;
3853 98140 : ndistinct = clamp_row_est(ndistinct);
3854 : }
3855 :
3856 : /*
3857 : * Initial estimate of bucketsize fraction is 1/nbuckets as long as the
3858 : * number of buckets is less than the expected number of distinct values;
3859 : * otherwise it is 1/ndistinct.
3860 : */
3861 98154 : if (ndistinct > nbuckets)
3862 74 : estfract = 1.0 / nbuckets;
3863 : else
3864 98080 : estfract = 1.0 / ndistinct;
3865 :
3866 : /*
3867 : * Adjust estimated bucketsize upward to account for skewed distribution.
3868 : */
3869 98154 : if (avgfreq > 0.0 && *mcv_freq > avgfreq)
3870 29520 : estfract *= *mcv_freq / avgfreq;
3871 :
3872 : /*
3873 : * Clamp bucketsize to sane range (the above adjustment could easily
3874 : * produce an out-of-range result). We set the lower bound a little above
3875 : * zero, since zero isn't a very sane result.
3876 : */
3877 98154 : if (estfract < 1.0e-6)
3878 0 : estfract = 1.0e-6;
3879 98154 : else if (estfract > 1.0)
3880 21416 : estfract = 1.0;
3881 :
3882 98154 : *bucketsize_frac = (Selectivity) estfract;
3883 :
3884 98154 : ReleaseVariableStats(vardata);
3885 : }
3886 :
3887 : /*
3888 : * estimate_hashagg_tablesize
3889 : * estimate the number of bytes that a hash aggregate hashtable will
3890 : * require based on the agg_costs, path width and number of groups.
3891 : *
3892 : * We return the result as "double" to forestall any possible overflow
3893 : * problem in the multiplication by dNumGroups.
3894 : *
3895 : * XXX this may be over-estimating the size now that hashagg knows to omit
3896 : * unneeded columns from the hashtable. Also for mixed-mode grouping sets,
3897 : * grouping columns not in the hashed set are counted here even though hashagg
3898 : * won't store them. Is this a problem?
3899 : */
3900 : double
3901 1978 : estimate_hashagg_tablesize(PlannerInfo *root, Path *path,
3902 : const AggClauseCosts *agg_costs, double dNumGroups)
3903 : {
3904 : Size hashentrysize;
3905 :
3906 1978 : hashentrysize = hash_agg_entry_size(list_length(root->aggtransinfos),
3907 1978 : path->pathtarget->width,
3908 : agg_costs->transitionSpace);
3909 :
3910 : /*
3911 : * Note that this disregards the effect of fill-factor and growth policy
3912 : * of the hash table. That's probably ok, given that the default
3913 : * fill-factor is relatively high. It'd be hard to meaningfully factor in
3914 : * "double-in-size" growth policies here.
3915 : */
3916 1978 : return hashentrysize * dNumGroups;
3917 : }
3918 :
3919 :
3920 : /*-------------------------------------------------------------------------
3921 : *
3922 : * Support routines
3923 : *
3924 : *-------------------------------------------------------------------------
3925 : */
3926 :
3927 : /*
3928 : * Find applicable ndistinct statistics for the given list of VarInfos (which
3929 : * must all belong to the given rel), and update *ndistinct to the estimate of
3930 : * the MVNDistinctItem that best matches. If a match it found, *varinfos is
3931 : * updated to remove the list of matched varinfos.
3932 : *
3933 : * Varinfos that aren't for simple Vars are ignored.
3934 : *
3935 : * Return true if we're able to find a match, false otherwise.
3936 : */
3937 : static bool
3938 1545130 : estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
3939 : List **varinfos, double *ndistinct)
3940 : {
3941 : ListCell *lc;
3942 : int nmatches_vars;
3943 : int nmatches_exprs;
3944 1545130 : Oid statOid = InvalidOid;
3945 : MVNDistinct *stats;
3946 1545130 : StatisticExtInfo *matched_info = NULL;
3947 : RangeTblEntry *rte;
3948 :
3949 : /* bail out immediately if the table has no extended statistics */
3950 1545130 : if (!rel->statlist)
3951 1529068 : return false;
3952 :
3953 : /* look for the ndistinct statistics object matching the most vars */
3954 16062 : nmatches_vars = 0; /* we require at least two matches */
3955 16062 : nmatches_exprs = 0;
3956 64092 : foreach(lc, rel->statlist)
3957 : {
3958 : ListCell *lc2;
3959 48030 : StatisticExtInfo *info = (StatisticExtInfo *) lfirst(lc);
3960 48030 : int nshared_vars = 0;
3961 48030 : int nshared_exprs = 0;
3962 :
3963 : /* skip statistics of other kinds */
3964 48030 : if (info->kind != STATS_EXT_NDISTINCT)
3965 23958 : continue;
3966 :
3967 : /*
3968 : * Determine how many expressions (and variables in non-matched
3969 : * expressions) match. We'll then use these numbers to pick the
3970 : * statistics object that best matches the clauses.
3971 : */
3972 71916 : foreach(lc2, *varinfos)
3973 : {
3974 : ListCell *lc3;
3975 47844 : GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc2);
3976 : AttrNumber attnum;
3977 :
3978 : Assert(varinfo->rel == rel);
3979 :
3980 : /* simple Var, search in statistics keys directly */
3981 47844 : if (IsA(varinfo->var, Var))
3982 : {
3983 39582 : attnum = ((Var *) varinfo->var)->varattno;
3984 :
3985 : /*
3986 : * Ignore system attributes - we don't support statistics on
3987 : * them, so can't match them (and it'd fail as the values are
3988 : * negative).
3989 : */
3990 39582 : if (!AttrNumberIsForUserDefinedAttr(attnum))
3991 168 : continue;
3992 :
3993 39414 : if (bms_is_member(attnum, info->keys))
3994 20640 : nshared_vars++;
3995 :
3996 39414 : continue;
3997 : }
3998 :
3999 : /* expression - see if it's in the statistics object */
4000 15324 : foreach(lc3, info->exprs)
4001 : {
4002 12132 : Node *expr = (Node *) lfirst(lc3);
4003 :
4004 12132 : if (equal(varinfo->var, expr))
4005 : {
4006 5070 : nshared_exprs++;
4007 5070 : break;
4008 : }
4009 : }
4010 : }
4011 :
4012 24072 : if (nshared_vars + nshared_exprs < 2)
4013 16212 : continue;
4014 :
4015 : /*
4016 : * Does this statistics object match more columns than the currently
4017 : * best object? If so, use this one instead.
4018 : *
4019 : * XXX This should break ties using name of the object, or something
4020 : * like that, to make the outcome stable.
4021 : */
4022 7860 : if ((nshared_exprs > nmatches_exprs) ||
4023 6090 : (((nshared_exprs == nmatches_exprs)) && (nshared_vars > nmatches_vars)))
4024 : {
4025 7074 : statOid = info->statOid;
4026 7074 : nmatches_vars = nshared_vars;
4027 7074 : nmatches_exprs = nshared_exprs;
4028 7074 : matched_info = info;
4029 : }
4030 : }
4031 :
4032 : /* No match? */
4033 16062 : if (statOid == InvalidOid)
4034 9204 : return false;
4035 :
4036 : Assert(nmatches_vars + nmatches_exprs > 1);
4037 :
4038 6858 : rte = planner_rt_fetch(rel->relid, root);
4039 6858 : stats = statext_ndistinct_load(statOid, rte->inh);
4040 :
4041 : /*
4042 : * If we have a match, search it for the specific item that matches (there
4043 : * must be one), and construct the output values.
4044 : */
4045 6858 : if (stats)
4046 : {
4047 : int i;
4048 6858 : List *newlist = NIL;
4049 6858 : MVNDistinctItem *item = NULL;
4050 : ListCell *lc2;
4051 6858 : Bitmapset *matched = NULL;
4052 : AttrNumber attnum_offset;
4053 :
4054 : /*
4055 : * How much we need to offset the attnums? If there are no
4056 : * expressions, no offset is needed. Otherwise offset enough to move
4057 : * the lowest one (which is equal to number of expressions) to 1.
4058 : */
4059 6858 : if (matched_info->exprs)
4060 1968 : attnum_offset = (list_length(matched_info->exprs) + 1);
4061 : else
4062 4890 : attnum_offset = 0;
4063 :
4064 : /* see what actually matched */
4065 25992 : foreach(lc2, *varinfos)
4066 : {
4067 : ListCell *lc3;
4068 : int idx;
4069 19134 : bool found = false;
4070 :
4071 19134 : GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc2);
4072 :
4073 : /*
4074 : * Process a simple Var expression, by matching it to keys
4075 : * directly. If there's a matching expression, we'll try matching
4076 : * it later.
4077 : */
4078 19134 : if (IsA(varinfo->var, Var))
4079 : {
4080 16260 : AttrNumber attnum = ((Var *) varinfo->var)->varattno;
4081 :
4082 : /*
4083 : * Ignore expressions on system attributes. Can't rely on the
4084 : * bms check for negative values.
4085 : */
4086 16260 : if (!AttrNumberIsForUserDefinedAttr(attnum))
4087 54 : continue;
4088 :
4089 : /* Is the variable covered by the statistics object? */
4090 16206 : if (!bms_is_member(attnum, matched_info->keys))
4091 3876 : continue;
4092 :
4093 12330 : attnum = attnum + attnum_offset;
4094 :
4095 : /* ensure sufficient offset */
4096 : Assert(AttrNumberIsForUserDefinedAttr(attnum));
4097 :
4098 12330 : matched = bms_add_member(matched, attnum);
4099 :
4100 12330 : found = true;
4101 : }
4102 :
4103 : /*
4104 : * XXX Maybe we should allow searching the expressions even if we
4105 : * found an attribute matching the expression? That would handle
4106 : * trivial expressions like "(a)" but it seems fairly useless.
4107 : */
4108 15204 : if (found)
4109 12330 : continue;
4110 :
4111 : /* expression - see if it's in the statistics object */
4112 2874 : idx = 0;
4113 5208 : foreach(lc3, matched_info->exprs)
4114 : {
4115 4560 : Node *expr = (Node *) lfirst(lc3);
4116 :
4117 4560 : if (equal(varinfo->var, expr))
4118 : {
4119 2226 : AttrNumber attnum = -(idx + 1);
4120 :
4121 2226 : attnum = attnum + attnum_offset;
4122 :
4123 : /* ensure sufficient offset */
4124 : Assert(AttrNumberIsForUserDefinedAttr(attnum));
4125 :
4126 2226 : matched = bms_add_member(matched, attnum);
4127 :
4128 : /* there should be just one matching expression */
4129 2226 : break;
4130 : }
4131 :
4132 2334 : idx++;
4133 : }
4134 : }
4135 :
4136 : /* Find the specific item that exactly matches the combination */
4137 13962 : for (i = 0; i < stats->nitems; i++)
4138 : {
4139 : int j;
4140 13962 : MVNDistinctItem *tmpitem = &stats->items[i];
4141 :
4142 13962 : if (tmpitem->nattributes != bms_num_members(matched))
4143 2520 : continue;
4144 :
4145 : /* assume it's the right item */
4146 11442 : item = tmpitem;
4147 :
4148 : /* check that all item attributes/expressions fit the match */
4149 28602 : for (j = 0; j < tmpitem->nattributes; j++)
4150 : {
4151 21744 : AttrNumber attnum = tmpitem->attributes[j];
4152 :
4153 : /*
4154 : * Thanks to how we constructed the matched bitmap above, we
4155 : * can just offset all attnums the same way.
4156 : */
4157 21744 : attnum = attnum + attnum_offset;
4158 :
4159 21744 : if (!bms_is_member(attnum, matched))
4160 : {
4161 : /* nah, it's not this item */
4162 4584 : item = NULL;
4163 4584 : break;
4164 : }
4165 : }
4166 :
4167 : /*
4168 : * If the item has all the matched attributes, we know it's the
4169 : * right one - there can't be a better one. matching more.
4170 : */
4171 11442 : if (item)
4172 6858 : break;
4173 : }
4174 :
4175 : /*
4176 : * Make sure we found an item. There has to be one, because ndistinct
4177 : * statistics includes all combinations of attributes.
4178 : */
4179 6858 : if (!item)
4180 0 : elog(ERROR, "corrupt MVNDistinct entry");
4181 :
4182 : /* Form the output varinfo list, keeping only unmatched ones */
4183 25992 : foreach(lc, *varinfos)
4184 : {
4185 19134 : GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc);
4186 : ListCell *lc3;
4187 19134 : bool found = false;
4188 :
4189 : /*
4190 : * Let's look at plain variables first, because it's the most
4191 : * common case and the check is quite cheap. We can simply get the
4192 : * attnum and check (with an offset) matched bitmap.
4193 : */
4194 19134 : if (IsA(varinfo->var, Var))
4195 : {
4196 16260 : AttrNumber attnum = ((Var *) varinfo->var)->varattno;
4197 :
4198 : /*
4199 : * If it's a system attribute, we're done. We don't support
4200 : * extended statistics on system attributes, so it's clearly
4201 : * not matched. Just keep the expression and continue.
4202 : */
4203 16260 : if (!AttrNumberIsForUserDefinedAttr(attnum))
4204 : {
4205 54 : newlist = lappend(newlist, varinfo);
4206 54 : continue;
4207 : }
4208 :
4209 : /* apply the same offset as above */
4210 16206 : attnum += attnum_offset;
4211 :
4212 : /* if it's not matched, keep the varinfo */
4213 16206 : if (!bms_is_member(attnum, matched))
4214 3876 : newlist = lappend(newlist, varinfo);
4215 :
4216 : /* The rest of the loop deals with complex expressions. */
4217 16206 : continue;
4218 : }
4219 :
4220 : /*
4221 : * Process complex expressions, not just simple Vars.
4222 : *
4223 : * First, we search for an exact match of an expression. If we
4224 : * find one, we can just discard the whole GroupExprInfo, with all
4225 : * the variables we extracted from it.
4226 : *
4227 : * Otherwise we inspect the individual vars, and try matching it
4228 : * to variables in the item.
4229 : */
4230 5208 : foreach(lc3, matched_info->exprs)
4231 : {
4232 4560 : Node *expr = (Node *) lfirst(lc3);
4233 :
4234 4560 : if (equal(varinfo->var, expr))
4235 : {
4236 2226 : found = true;
4237 2226 : break;
4238 : }
4239 : }
4240 :
4241 : /* found exact match, skip */
4242 2874 : if (found)
4243 2226 : continue;
4244 :
4245 648 : newlist = lappend(newlist, varinfo);
4246 : }
4247 :
4248 6858 : *varinfos = newlist;
4249 6858 : *ndistinct = item->ndistinct;
4250 6858 : return true;
4251 : }
4252 :
4253 0 : return false;
4254 : }
4255 :
4256 : /*
4257 : * convert_to_scalar
4258 : * Convert non-NULL values of the indicated types to the comparison
4259 : * scale needed by scalarineqsel().
4260 : * Returns "true" if successful.
4261 : *
4262 : * XXX this routine is a hack: ideally we should look up the conversion
4263 : * subroutines in pg_type.
4264 : *
4265 : * All numeric datatypes are simply converted to their equivalent
4266 : * "double" values. (NUMERIC values that are outside the range of "double"
4267 : * are clamped to +/- HUGE_VAL.)
4268 : *
4269 : * String datatypes are converted by convert_string_to_scalar(),
4270 : * which is explained below. The reason why this routine deals with
4271 : * three values at a time, not just one, is that we need it for strings.
4272 : *
4273 : * The bytea datatype is just enough different from strings that it has
4274 : * to be treated separately.
4275 : *
4276 : * The several datatypes representing absolute times are all converted
4277 : * to Timestamp, which is actually an int64, and then we promote that to
4278 : * a double. Note this will give correct results even for the "special"
4279 : * values of Timestamp, since those are chosen to compare correctly;
4280 : * see timestamp_cmp.
4281 : *
4282 : * The several datatypes representing relative times (intervals) are all
4283 : * converted to measurements expressed in seconds.
4284 : */
4285 : static bool
4286 64604 : convert_to_scalar(Datum value, Oid valuetypid, Oid collid, double *scaledvalue,
4287 : Datum lobound, Datum hibound, Oid boundstypid,
4288 : double *scaledlobound, double *scaledhibound)
4289 : {
4290 64604 : bool failure = false;
4291 :
4292 : /*
4293 : * Both the valuetypid and the boundstypid should exactly match the
4294 : * declared input type(s) of the operator we are invoked for. However,
4295 : * extensions might try to use scalarineqsel as estimator for operators
4296 : * with input type(s) we don't handle here; in such cases, we want to
4297 : * return false, not fail. In any case, we mustn't assume that valuetypid
4298 : * and boundstypid are identical.
4299 : *
4300 : * XXX The histogram we are interpolating between points of could belong
4301 : * to a column that's only binary-compatible with the declared type. In
4302 : * essence we are assuming that the semantics of binary-compatible types
4303 : * are enough alike that we can use a histogram generated with one type's
4304 : * operators to estimate selectivity for the other's. This is outright
4305 : * wrong in some cases --- in particular signed versus unsigned
4306 : * interpretation could trip us up. But it's useful enough in the
4307 : * majority of cases that we do it anyway. Should think about more
4308 : * rigorous ways to do it.
4309 : */
4310 64604 : switch (valuetypid)
4311 : {
4312 : /*
4313 : * Built-in numeric types
4314 : */
4315 61076 : case BOOLOID:
4316 : case INT2OID:
4317 : case INT4OID:
4318 : case INT8OID:
4319 : case FLOAT4OID:
4320 : case FLOAT8OID:
4321 : case NUMERICOID:
4322 : case OIDOID:
4323 : case REGPROCOID:
4324 : case REGPROCEDUREOID:
4325 : case REGOPEROID:
4326 : case REGOPERATOROID:
4327 : case REGCLASSOID:
4328 : case REGTYPEOID:
4329 : case REGCONFIGOID:
4330 : case REGDICTIONARYOID:
4331 : case REGROLEOID:
4332 : case REGNAMESPACEOID:
4333 61076 : *scaledvalue = convert_numeric_to_scalar(value, valuetypid,
4334 : &failure);
4335 61076 : *scaledlobound = convert_numeric_to_scalar(lobound, boundstypid,
4336 : &failure);
4337 61076 : *scaledhibound = convert_numeric_to_scalar(hibound, boundstypid,
4338 : &failure);
4339 61076 : return !failure;
4340 :
4341 : /*
4342 : * Built-in string types
4343 : */
4344 3528 : case CHAROID:
4345 : case BPCHAROID:
4346 : case VARCHAROID:
4347 : case TEXTOID:
4348 : case NAMEOID:
4349 : {
4350 3528 : char *valstr = convert_string_datum(value, valuetypid,
4351 : collid, &failure);
4352 3528 : char *lostr = convert_string_datum(lobound, boundstypid,
4353 : collid, &failure);
4354 3528 : char *histr = convert_string_datum(hibound, boundstypid,
4355 : collid, &failure);
4356 :
4357 : /*
4358 : * Bail out if any of the values is not of string type. We
4359 : * might leak converted strings for the other value(s), but
4360 : * that's not worth troubling over.
4361 : */
4362 3528 : if (failure)
4363 0 : return false;
4364 :
4365 3528 : convert_string_to_scalar(valstr, scaledvalue,
4366 : lostr, scaledlobound,
4367 : histr, scaledhibound);
4368 3528 : pfree(valstr);
4369 3528 : pfree(lostr);
4370 3528 : pfree(histr);
4371 3528 : return true;
4372 : }
4373 :
4374 : /*
4375 : * Built-in bytea type
4376 : */
4377 0 : case BYTEAOID:
4378 : {
4379 : /* We only support bytea vs bytea comparison */
4380 0 : if (boundstypid != BYTEAOID)
4381 0 : return false;
4382 0 : convert_bytea_to_scalar(value, scaledvalue,
4383 : lobound, scaledlobound,
4384 : hibound, scaledhibound);
4385 0 : return true;
4386 : }
4387 :
4388 : /*
4389 : * Built-in time types
4390 : */
4391 0 : case TIMESTAMPOID:
4392 : case TIMESTAMPTZOID:
4393 : case DATEOID:
4394 : case INTERVALOID:
4395 : case TIMEOID:
4396 : case TIMETZOID:
4397 0 : *scaledvalue = convert_timevalue_to_scalar(value, valuetypid,
4398 : &failure);
4399 0 : *scaledlobound = convert_timevalue_to_scalar(lobound, boundstypid,
4400 : &failure);
4401 0 : *scaledhibound = convert_timevalue_to_scalar(hibound, boundstypid,
4402 : &failure);
4403 0 : return !failure;
4404 :
4405 : /*
4406 : * Built-in network types
4407 : */
4408 0 : case INETOID:
4409 : case CIDROID:
4410 : case MACADDROID:
4411 : case MACADDR8OID:
4412 0 : *scaledvalue = convert_network_to_scalar(value, valuetypid,
4413 : &failure);
4414 0 : *scaledlobound = convert_network_to_scalar(lobound, boundstypid,
4415 : &failure);
4416 0 : *scaledhibound = convert_network_to_scalar(hibound, boundstypid,
4417 : &failure);
4418 0 : return !failure;
4419 : }
4420 : /* Don't know how to convert */
4421 0 : *scaledvalue = *scaledlobound = *scaledhibound = 0;
4422 0 : return false;
4423 : }
4424 :
4425 : /*
4426 : * Do convert_to_scalar()'s work for any numeric data type.
4427 : *
4428 : * On failure (e.g., unsupported typid), set *failure to true;
4429 : * otherwise, that variable is not changed.
4430 : */
4431 : static double
4432 183228 : convert_numeric_to_scalar(Datum value, Oid typid, bool *failure)
4433 : {
4434 183228 : switch (typid)
4435 : {
4436 0 : case BOOLOID:
4437 0 : return (double) DatumGetBool(value);
4438 12 : case INT2OID:
4439 12 : return (double) DatumGetInt16(value);
4440 25326 : case INT4OID:
4441 25326 : return (double) DatumGetInt32(value);
4442 0 : case INT8OID:
4443 0 : return (double) DatumGetInt64(value);
4444 0 : case FLOAT4OID:
4445 0 : return (double) DatumGetFloat4(value);
4446 36 : case FLOAT8OID:
4447 36 : return (double) DatumGetFloat8(value);
4448 0 : case NUMERICOID:
4449 : /* Note: out-of-range values will be clamped to +-HUGE_VAL */
4450 0 : return (double)
4451 0 : DatumGetFloat8(DirectFunctionCall1(numeric_float8_no_overflow,
4452 : value));
4453 157854 : case OIDOID:
4454 : case REGPROCOID:
4455 : case REGPROCEDUREOID:
4456 : case REGOPEROID:
4457 : case REGOPERATOROID:
4458 : case REGCLASSOID:
4459 : case REGTYPEOID:
4460 : case REGCONFIGOID:
4461 : case REGDICTIONARYOID:
4462 : case REGROLEOID:
4463 : case REGNAMESPACEOID:
4464 : /* we can treat OIDs as integers... */
4465 157854 : return (double) DatumGetObjectId(value);
4466 : }
4467 :
4468 0 : *failure = true;
4469 0 : return 0;
4470 : }
4471 :
4472 : /*
4473 : * Do convert_to_scalar()'s work for any character-string data type.
4474 : *
4475 : * String datatypes are converted to a scale that ranges from 0 to 1,
4476 : * where we visualize the bytes of the string as fractional digits.
4477 : *
4478 : * We do not want the base to be 256, however, since that tends to
4479 : * generate inflated selectivity estimates; few databases will have
4480 : * occurrences of all 256 possible byte values at each position.
4481 : * Instead, use the smallest and largest byte values seen in the bounds
4482 : * as the estimated range for each byte, after some fudging to deal with
4483 : * the fact that we probably aren't going to see the full range that way.
4484 : *
4485 : * An additional refinement is that we discard any common prefix of the
4486 : * three strings before computing the scaled values. This allows us to
4487 : * "zoom in" when we encounter a narrow data range. An example is a phone
4488 : * number database where all the values begin with the same area code.
4489 : * (Actually, the bounds will be adjacent histogram-bin-boundary values,
4490 : * so this is more likely to happen than you might think.)
4491 : */
4492 : static void
4493 3528 : convert_string_to_scalar(char *value,
4494 : double *scaledvalue,
4495 : char *lobound,
4496 : double *scaledlobound,
4497 : char *hibound,
4498 : double *scaledhibound)
4499 : {
4500 : int rangelo,
4501 : rangehi;
4502 : char *sptr;
4503 :
4504 3528 : rangelo = rangehi = (unsigned char) hibound[0];
4505 46254 : for (sptr = lobound; *sptr; sptr++)
4506 : {
4507 42726 : if (rangelo > (unsigned char) *sptr)
4508 9210 : rangelo = (unsigned char) *sptr;
4509 42726 : if (rangehi < (unsigned char) *sptr)
4510 4704 : rangehi = (unsigned char) *sptr;
4511 : }
4512 47956 : for (sptr = hibound; *sptr; sptr++)
4513 : {
4514 44428 : if (rangelo > (unsigned char) *sptr)
4515 794 : rangelo = (unsigned char) *sptr;
4516 44428 : if (rangehi < (unsigned char) *sptr)
4517 2896 : rangehi = (unsigned char) *sptr;
4518 : }
4519 : /* If range includes any upper-case ASCII chars, make it include all */
4520 3528 : if (rangelo <= 'Z' && rangehi >= 'A')
4521 : {
4522 1006 : if (rangelo > 'A')
4523 90 : rangelo = 'A';
4524 1006 : if (rangehi < 'Z')
4525 480 : rangehi = 'Z';
4526 : }
4527 : /* Ditto lower-case */
4528 3528 : if (rangelo <= 'z' && rangehi >= 'a')
4529 : {
4530 3036 : if (rangelo > 'a')
4531 0 : rangelo = 'a';
4532 3036 : if (rangehi < 'z')
4533 2946 : rangehi = 'z';
4534 : }
4535 : /* Ditto digits */
4536 3528 : if (rangelo <= '9' && rangehi >= '0')
4537 : {
4538 448 : if (rangelo > '0')
4539 340 : rangelo = '0';
4540 448 : if (rangehi < '9')
4541 6 : rangehi = '9';
4542 : }
4543 :
4544 : /*
4545 : * If range includes less than 10 chars, assume we have not got enough
4546 : * data, and make it include regular ASCII set.
4547 : */
4548 3528 : if (rangehi - rangelo < 9)
4549 : {
4550 0 : rangelo = ' ';
4551 0 : rangehi = 127;
4552 : }
4553 :
4554 : /*
4555 : * Now strip any common prefix of the three strings.
4556 : */
4557 5276 : while (*lobound)
4558 : {
4559 5266 : if (*lobound != *hibound || *lobound != *value)
4560 : break;
4561 1748 : lobound++, hibound++, value++;
4562 : }
4563 :
4564 : /*
4565 : * Now we can do the conversions.
4566 : */
4567 3528 : *scaledvalue = convert_one_string_to_scalar(value, rangelo, rangehi);
4568 3528 : *scaledlobound = convert_one_string_to_scalar(lobound, rangelo, rangehi);
4569 3528 : *scaledhibound = convert_one_string_to_scalar(hibound, rangelo, rangehi);
4570 3528 : }
4571 :
4572 : static double
4573 10584 : convert_one_string_to_scalar(char *value, int rangelo, int rangehi)
4574 : {
4575 10584 : int slen = strlen(value);
4576 : double num,
4577 : denom,
4578 : base;
4579 :
4580 10584 : if (slen <= 0)
4581 10 : return 0.0; /* empty string has scalar value 0 */
4582 :
4583 : /*
4584 : * There seems little point in considering more than a dozen bytes from
4585 : * the string. Since base is at least 10, that will give us nominal
4586 : * resolution of at least 12 decimal digits, which is surely far more
4587 : * precision than this estimation technique has got anyway (especially in
4588 : * non-C locales). Also, even with the maximum possible base of 256, this
4589 : * ensures denom cannot grow larger than 256^13 = 2.03e31, which will not
4590 : * overflow on any known machine.
4591 : */
4592 10574 : if (slen > 12)
4593 3002 : slen = 12;
4594 :
4595 : /* Convert initial characters to fraction */
4596 10574 : base = rangehi - rangelo + 1;
4597 10574 : num = 0.0;
4598 10574 : denom = base;
4599 92046 : while (slen-- > 0)
4600 : {
4601 81472 : int ch = (unsigned char) *value++;
4602 :
4603 81472 : if (ch < rangelo)
4604 136 : ch = rangelo - 1;
4605 81336 : else if (ch > rangehi)
4606 0 : ch = rangehi + 1;
4607 81472 : num += ((double) (ch - rangelo)) / denom;
4608 81472 : denom *= base;
4609 : }
4610 :
4611 10574 : return num;
4612 : }
4613 :
4614 : /*
4615 : * Convert a string-type Datum into a palloc'd, null-terminated string.
4616 : *
4617 : * On failure (e.g., unsupported typid), set *failure to true;
4618 : * otherwise, that variable is not changed. (We'll return NULL on failure.)
4619 : *
4620 : * When using a non-C locale, we must pass the string through strxfrm()
4621 : * before continuing, so as to generate correct locale-specific results.
4622 : */
4623 : static char *
4624 10584 : convert_string_datum(Datum value, Oid typid, Oid collid, bool *failure)
4625 : {
4626 : char *val;
4627 :
4628 10584 : switch (typid)
4629 : {
4630 0 : case CHAROID:
4631 0 : val = (char *) palloc(2);
4632 0 : val[0] = DatumGetChar(value);
4633 0 : val[1] = '\0';
4634 0 : break;
4635 3128 : case BPCHAROID:
4636 : case VARCHAROID:
4637 : case TEXTOID:
4638 3128 : val = TextDatumGetCString(value);
4639 3128 : break;
4640 7456 : case NAMEOID:
4641 : {
4642 7456 : NameData *nm = (NameData *) DatumGetPointer(value);
4643 :
4644 7456 : val = pstrdup(NameStr(*nm));
4645 7456 : break;
4646 : }
4647 0 : default:
4648 0 : *failure = true;
4649 0 : return NULL;
4650 : }
4651 :
4652 10584 : if (!lc_collate_is_c(collid))
4653 : {
4654 : char *xfrmstr;
4655 : size_t xfrmlen;
4656 : size_t xfrmlen2 PG_USED_FOR_ASSERTS_ONLY;
4657 :
4658 : /*
4659 : * XXX: We could guess at a suitable output buffer size and only call
4660 : * strxfrm twice if our guess is too small.
4661 : *
4662 : * XXX: strxfrm doesn't support UTF-8 encoding on Win32, it can return
4663 : * bogus data or set an error. This is not really a problem unless it
4664 : * crashes since it will only give an estimation error and nothing
4665 : * fatal.
4666 : */
4667 108 : xfrmlen = strxfrm(NULL, val, 0);
4668 : #ifdef WIN32
4669 :
4670 : /*
4671 : * On Windows, strxfrm returns INT_MAX when an error occurs. Instead
4672 : * of trying to allocate this much memory (and fail), just return the
4673 : * original string unmodified as if we were in the C locale.
4674 : */
4675 : if (xfrmlen == INT_MAX)
4676 : return val;
4677 : #endif
4678 108 : xfrmstr = (char *) palloc(xfrmlen + 1);
4679 108 : xfrmlen2 = strxfrm(xfrmstr, val, xfrmlen + 1);
4680 :
4681 : /*
4682 : * Some systems (e.g., glibc) can return a smaller value from the
4683 : * second call than the first; thus the Assert must be <= not ==.
4684 : */
4685 : Assert(xfrmlen2 <= xfrmlen);
4686 108 : pfree(val);
4687 108 : val = xfrmstr;
4688 : }
4689 :
4690 10584 : return val;
4691 : }
4692 :
4693 : /*
4694 : * Do convert_to_scalar()'s work for any bytea data type.
4695 : *
4696 : * Very similar to convert_string_to_scalar except we can't assume
4697 : * null-termination and therefore pass explicit lengths around.
4698 : *
4699 : * Also, assumptions about likely "normal" ranges of characters have been
4700 : * removed - a data range of 0..255 is always used, for now. (Perhaps
4701 : * someday we will add information about actual byte data range to
4702 : * pg_statistic.)
4703 : */
4704 : static void
4705 0 : convert_bytea_to_scalar(Datum value,
4706 : double *scaledvalue,
4707 : Datum lobound,
4708 : double *scaledlobound,
4709 : Datum hibound,
4710 : double *scaledhibound)
4711 : {
4712 0 : bytea *valuep = DatumGetByteaPP(value);
4713 0 : bytea *loboundp = DatumGetByteaPP(lobound);
4714 0 : bytea *hiboundp = DatumGetByteaPP(hibound);
4715 : int rangelo,
4716 : rangehi,
4717 0 : valuelen = VARSIZE_ANY_EXHDR(valuep),
4718 0 : loboundlen = VARSIZE_ANY_EXHDR(loboundp),
4719 0 : hiboundlen = VARSIZE_ANY_EXHDR(hiboundp),
4720 : i,
4721 : minlen;
4722 0 : unsigned char *valstr = (unsigned char *) VARDATA_ANY(valuep);
4723 0 : unsigned char *lostr = (unsigned char *) VARDATA_ANY(loboundp);
4724 0 : unsigned char *histr = (unsigned char *) VARDATA_ANY(hiboundp);
4725 :
4726 : /*
4727 : * Assume bytea data is uniformly distributed across all byte values.
4728 : */
4729 0 : rangelo = 0;
4730 0 : rangehi = 255;
4731 :
4732 : /*
4733 : * Now strip any common prefix of the three strings.
4734 : */
4735 0 : minlen = Min(Min(valuelen, loboundlen), hiboundlen);
4736 0 : for (i = 0; i < minlen; i++)
4737 : {
4738 0 : if (*lostr != *histr || *lostr != *valstr)
4739 : break;
4740 0 : lostr++, histr++, valstr++;
4741 0 : loboundlen--, hiboundlen--, valuelen--;
4742 : }
4743 :
4744 : /*
4745 : * Now we can do the conversions.
4746 : */
4747 0 : *scaledvalue = convert_one_bytea_to_scalar(valstr, valuelen, rangelo, rangehi);
4748 0 : *scaledlobound = convert_one_bytea_to_scalar(lostr, loboundlen, rangelo, rangehi);
4749 0 : *scaledhibound = convert_one_bytea_to_scalar(histr, hiboundlen, rangelo, rangehi);
4750 0 : }
4751 :
4752 : static double
4753 0 : convert_one_bytea_to_scalar(unsigned char *value, int valuelen,
4754 : int rangelo, int rangehi)
4755 : {
4756 : double num,
4757 : denom,
4758 : base;
4759 :
4760 0 : if (valuelen <= 0)
4761 0 : return 0.0; /* empty string has scalar value 0 */
4762 :
4763 : /*
4764 : * Since base is 256, need not consider more than about 10 chars (even
4765 : * this many seems like overkill)
4766 : */
4767 0 : if (valuelen > 10)
4768 0 : valuelen = 10;
4769 :
4770 : /* Convert initial characters to fraction */
4771 0 : base = rangehi - rangelo + 1;
4772 0 : num = 0.0;
4773 0 : denom = base;
4774 0 : while (valuelen-- > 0)
4775 : {
4776 0 : int ch = *value++;
4777 :
4778 0 : if (ch < rangelo)
4779 0 : ch = rangelo - 1;
4780 0 : else if (ch > rangehi)
4781 0 : ch = rangehi + 1;
4782 0 : num += ((double) (ch - rangelo)) / denom;
4783 0 : denom *= base;
4784 : }
4785 :
4786 0 : return num;
4787 : }
4788 :
4789 : /*
4790 : * Do convert_to_scalar()'s work for any timevalue data type.
4791 : *
4792 : * On failure (e.g., unsupported typid), set *failure to true;
4793 : * otherwise, that variable is not changed.
4794 : */
4795 : static double
4796 0 : convert_timevalue_to_scalar(Datum value, Oid typid, bool *failure)
4797 : {
4798 0 : switch (typid)
4799 : {
4800 0 : case TIMESTAMPOID:
4801 0 : return DatumGetTimestamp(value);
4802 0 : case TIMESTAMPTZOID:
4803 0 : return DatumGetTimestampTz(value);
4804 0 : case DATEOID:
4805 0 : return date2timestamp_no_overflow(DatumGetDateADT(value));
4806 0 : case INTERVALOID:
4807 : {
4808 0 : Interval *interval = DatumGetIntervalP(value);
4809 :
4810 : /*
4811 : * Convert the month part of Interval to days using assumed
4812 : * average month length of 365.25/12.0 days. Not too
4813 : * accurate, but plenty good enough for our purposes.
4814 : */
4815 0 : return interval->time + interval->day * (double) USECS_PER_DAY +
4816 0 : interval->month * ((DAYS_PER_YEAR / (double) MONTHS_PER_YEAR) * USECS_PER_DAY);
4817 : }
4818 0 : case TIMEOID:
4819 0 : return DatumGetTimeADT(value);
4820 0 : case TIMETZOID:
4821 : {
4822 0 : TimeTzADT *timetz = DatumGetTimeTzADTP(value);
4823 :
4824 : /* use GMT-equivalent time */
4825 0 : return (double) (timetz->time + (timetz->zone * 1000000.0));
4826 : }
4827 : }
4828 :
4829 0 : *failure = true;
4830 0 : return 0;
4831 : }
4832 :
4833 :
4834 : /*
4835 : * get_restriction_variable
4836 : * Examine the args of a restriction clause to see if it's of the
4837 : * form (variable op pseudoconstant) or (pseudoconstant op variable),
4838 : * where "variable" could be either a Var or an expression in vars of a
4839 : * single relation. If so, extract information about the variable,
4840 : * and also indicate which side it was on and the other argument.
4841 : *
4842 : * Inputs:
4843 : * root: the planner info
4844 : * args: clause argument list
4845 : * varRelid: see specs for restriction selectivity functions
4846 : *
4847 : * Outputs: (these are valid only if true is returned)
4848 : * *vardata: gets information about variable (see examine_variable)
4849 : * *other: gets other clause argument, aggressively reduced to a constant
4850 : * *varonleft: set true if variable is on the left, false if on the right
4851 : *
4852 : * Returns true if a variable is identified, otherwise false.
4853 : *
4854 : * Note: if there are Vars on both sides of the clause, we must fail, because
4855 : * callers are expecting that the other side will act like a pseudoconstant.
4856 : */
4857 : bool
4858 512858 : get_restriction_variable(PlannerInfo *root, List *args, int varRelid,
4859 : VariableStatData *vardata, Node **other,
4860 : bool *varonleft)
4861 : {
4862 : Node *left,
4863 : *right;
4864 : VariableStatData rdata;
4865 :
4866 : /* Fail if not a binary opclause (probably shouldn't happen) */
4867 512858 : if (list_length(args) != 2)
4868 0 : return false;
4869 :
4870 512858 : left = (Node *) linitial(args);
4871 512858 : right = (Node *) lsecond(args);
4872 :
4873 : /*
4874 : * Examine both sides. Note that when varRelid is nonzero, Vars of other
4875 : * relations will be treated as pseudoconstants.
4876 : */
4877 512858 : examine_variable(root, left, varRelid, vardata);
4878 512858 : examine_variable(root, right, varRelid, &rdata);
4879 :
4880 : /*
4881 : * If one side is a variable and the other not, we win.
4882 : */
4883 512858 : if (vardata->rel && rdata.rel == NULL)
4884 : {
4885 420362 : *varonleft = true;
4886 420362 : *other = estimate_expression_value(root, rdata.var);
4887 : /* Assume we need no ReleaseVariableStats(rdata) here */
4888 420362 : return true;
4889 : }
4890 :
4891 92496 : if (vardata->rel == NULL && rdata.rel)
4892 : {
4893 89552 : *varonleft = false;
4894 89552 : *other = estimate_expression_value(root, vardata->var);
4895 : /* Assume we need no ReleaseVariableStats(*vardata) here */
4896 89552 : *vardata = rdata;
4897 89552 : return true;
4898 : }
4899 :
4900 : /* Oops, clause has wrong structure (probably var op var) */
4901 2944 : ReleaseVariableStats(*vardata);
4902 2944 : ReleaseVariableStats(rdata);
4903 :
4904 2944 : return false;
4905 : }
4906 :
4907 : /*
4908 : * get_join_variables
4909 : * Apply examine_variable() to each side of a join clause.
4910 : * Also, attempt to identify whether the join clause has the same
4911 : * or reversed sense compared to the SpecialJoinInfo.
4912 : *
4913 : * We consider the join clause "normal" if it is "lhs_var OP rhs_var",
4914 : * or "reversed" if it is "rhs_var OP lhs_var". In complicated cases
4915 : * where we can't tell for sure, we default to assuming it's normal.
4916 : */
4917 : void
4918 166362 : get_join_variables(PlannerInfo *root, List *args, SpecialJoinInfo *sjinfo,
4919 : VariableStatData *vardata1, VariableStatData *vardata2,
4920 : bool *join_is_reversed)
4921 : {
4922 : Node *left,
4923 : *right;
4924 :
4925 166362 : if (list_length(args) != 2)
4926 0 : elog(ERROR, "join operator should take two arguments");
4927 :
4928 166362 : left = (Node *) linitial(args);
4929 166362 : right = (Node *) lsecond(args);
4930 :
4931 166362 : examine_variable(root, left, 0, vardata1);
4932 166362 : examine_variable(root, right, 0, vardata2);
4933 :
4934 332604 : if (vardata1->rel &&
4935 166242 : bms_is_subset(vardata1->rel->relids, sjinfo->syn_righthand))
4936 27528 : *join_is_reversed = true; /* var1 is on RHS */
4937 277590 : else if (vardata2->rel &&
4938 138756 : bms_is_subset(vardata2->rel->relids, sjinfo->syn_lefthand))
4939 120 : *join_is_reversed = true; /* var2 is on LHS */
4940 : else
4941 138714 : *join_is_reversed = false;
4942 166362 : }
4943 :
4944 : /* statext_expressions_load copies the tuple, so just pfree it. */
4945 : static void
4946 3354 : ReleaseDummy(HeapTuple tuple)
4947 : {
4948 3354 : pfree(tuple);
4949 3354 : }
4950 :
4951 : /*
4952 : * examine_variable
4953 : * Try to look up statistical data about an expression.
4954 : * Fill in a VariableStatData struct to describe the expression.
4955 : *
4956 : * Inputs:
4957 : * root: the planner info
4958 : * node: the expression tree to examine
4959 : * varRelid: see specs for restriction selectivity functions
4960 : *
4961 : * Outputs: *vardata is filled as follows:
4962 : * var: the input expression (with any binary relabeling stripped, if
4963 : * it is or contains a variable; but otherwise the type is preserved)
4964 : * rel: RelOptInfo for relation containing variable; NULL if expression
4965 : * contains no Vars (NOTE this could point to a RelOptInfo of a
4966 : * subquery, not one in the current query).
4967 : * statsTuple: the pg_statistic entry for the variable, if one exists;
4968 : * otherwise NULL.
4969 : * freefunc: pointer to a function to release statsTuple with.
4970 : * vartype: exposed type of the expression; this should always match
4971 : * the declared input type of the operator we are estimating for.
4972 : * atttype, atttypmod: actual type/typmod of the "var" expression. This is
4973 : * commonly the same as the exposed type of the variable argument,
4974 : * but can be different in binary-compatible-type cases.
4975 : * isunique: true if we were able to match the var to a unique index or a
4976 : * single-column DISTINCT clause, implying its values are unique for
4977 : * this query. (Caution: this should be trusted for statistical
4978 : * purposes only, since we do not check indimmediate nor verify that
4979 : * the exact same definition of equality applies.)
4980 : * acl_ok: true if current user has permission to read the column(s)
4981 : * underlying the pg_statistic entry. This is consulted by
4982 : * statistic_proc_security_check().
4983 : *
4984 : * Caller is responsible for doing ReleaseVariableStats() before exiting.
4985 : */
4986 : void
4987 3455660 : examine_variable(PlannerInfo *root, Node *node, int varRelid,
4988 : VariableStatData *vardata)
4989 : {
4990 : Node *basenode;
4991 : Relids varnos;
4992 : RelOptInfo *onerel;
4993 :
4994 : /* Make sure we don't return dangling pointers in vardata */
4995 24189620 : MemSet(vardata, 0, sizeof(VariableStatData));
4996 :
4997 : /* Save the exposed type of the expression */
4998 3455660 : vardata->vartype = exprType(node);
4999 :
5000 : /* Look inside any binary-compatible relabeling */
5001 :
5002 3455660 : if (IsA(node, RelabelType))
5003 57392 : basenode = (Node *) ((RelabelType *) node)->arg;
5004 : else
5005 3398268 : basenode = node;
5006 :
5007 : /* Fast path for a simple Var */
5008 :
5009 3455660 : if (IsA(basenode, Var) &&
5010 451078 : (varRelid == 0 || varRelid == ((Var *) basenode)->varno))
5011 : {
5012 2852186 : Var *var = (Var *) basenode;
5013 :
5014 : /* Set up result fields other than the stats tuple */
5015 2852186 : vardata->var = basenode; /* return Var without relabeling */
5016 2852186 : vardata->rel = find_base_rel(root, var->varno);
5017 2852186 : vardata->atttype = var->vartype;
5018 2852186 : vardata->atttypmod = var->vartypmod;
5019 2852186 : vardata->isunique = has_unique_index(vardata->rel, var->varattno);
5020 :
5021 : /* Try to locate some stats */
5022 2852186 : examine_simple_variable(root, var, vardata);
5023 :
5024 2852186 : return;
5025 : }
5026 :
5027 : /*
5028 : * Okay, it's a more complicated expression. Determine variable
5029 : * membership. Note that when varRelid isn't zero, only vars of that
5030 : * relation are considered "real" vars.
5031 : */
5032 603474 : varnos = pull_varnos(root, basenode);
5033 :
5034 603474 : onerel = NULL;
5035 :
5036 603474 : switch (bms_membership(varnos))
5037 : {
5038 304726 : case BMS_EMPTY_SET:
5039 : /* No Vars at all ... must be pseudo-constant clause */
5040 304726 : break;
5041 293670 : case BMS_SINGLETON:
5042 293670 : if (varRelid == 0 || bms_is_member(varRelid, varnos))
5043 : {
5044 139394 : onerel = find_base_rel(root,
5045 68506 : (varRelid ? varRelid : bms_singleton_member(varnos)));
5046 70888 : vardata->rel = onerel;
5047 70888 : node = basenode; /* strip any relabeling */
5048 : }
5049 : /* else treat it as a constant */
5050 293670 : break;
5051 5078 : case BMS_MULTIPLE:
5052 5078 : if (varRelid == 0)
5053 : {
5054 : /* treat it as a variable of a join relation */
5055 4852 : vardata->rel = find_join_rel(root, varnos);
5056 4852 : node = basenode; /* strip any relabeling */
5057 : }
5058 226 : else if (bms_is_member(varRelid, varnos))
5059 : {
5060 : /* ignore the vars belonging to other relations */
5061 52 : vardata->rel = find_base_rel(root, varRelid);
5062 52 : node = basenode; /* strip any relabeling */
5063 : /* note: no point in expressional-index search here */
5064 : }
5065 : /* else treat it as a constant */
5066 5078 : break;
5067 : }
5068 :
5069 603474 : bms_free(varnos);
5070 :
5071 603474 : vardata->var = node;
5072 603474 : vardata->atttype = exprType(node);
5073 603474 : vardata->atttypmod = exprTypmod(node);
5074 :
5075 603474 : if (onerel)
5076 : {
5077 : /*
5078 : * We have an expression in vars of a single relation. Try to match
5079 : * it to expressional index columns, in hopes of finding some
5080 : * statistics.
5081 : *
5082 : * Note that we consider all index columns including INCLUDE columns,
5083 : * since there could be stats for such columns. But the test for
5084 : * uniqueness needs to be warier.
5085 : *
5086 : * XXX it's conceivable that there are multiple matches with different
5087 : * index opfamilies; if so, we need to pick one that matches the
5088 : * operator we are estimating for. FIXME later.
5089 : */
5090 : ListCell *ilist;
5091 : ListCell *slist;
5092 :
5093 111630 : foreach(ilist, onerel->indexlist)
5094 : {
5095 43544 : IndexOptInfo *index = (IndexOptInfo *) lfirst(ilist);
5096 : ListCell *indexpr_item;
5097 : int pos;
5098 :
5099 43544 : indexpr_item = list_head(index->indexprs);
5100 43544 : if (indexpr_item == NULL)
5101 38888 : continue; /* no expressions here... */
5102 :
5103 6582 : for (pos = 0; pos < index->ncolumns; pos++)
5104 : {
5105 4728 : if (index->indexkeys[pos] == 0)
5106 : {
5107 : Node *indexkey;
5108 :
5109 4656 : if (indexpr_item == NULL)
5110 0 : elog(ERROR, "too few entries in indexprs list");
5111 4656 : indexkey = (Node *) lfirst(indexpr_item);
5112 4656 : if (indexkey && IsA(indexkey, RelabelType))
5113 0 : indexkey = (Node *) ((RelabelType *) indexkey)->arg;
5114 4656 : if (equal(node, indexkey))
5115 : {
5116 : /*
5117 : * Found a match ... is it a unique index? Tests here
5118 : * should match has_unique_index().
5119 : */
5120 3390 : if (index->unique &&
5121 390 : index->nkeycolumns == 1 &&
5122 390 : pos == 0 &&
5123 390 : (index->indpred == NIL || index->predOK))
5124 390 : vardata->isunique = true;
5125 :
5126 : /*
5127 : * Has it got stats? We only consider stats for
5128 : * non-partial indexes, since partial indexes probably
5129 : * don't reflect whole-relation statistics; the above
5130 : * check for uniqueness is the only info we take from
5131 : * a partial index.
5132 : *
5133 : * An index stats hook, however, must make its own
5134 : * decisions about what to do with partial indexes.
5135 : */
5136 3390 : if (get_index_stats_hook &&
5137 0 : (*get_index_stats_hook) (root, index->indexoid,
5138 0 : pos + 1, vardata))
5139 : {
5140 : /*
5141 : * The hook took control of acquiring a stats
5142 : * tuple. If it did supply a tuple, it'd better
5143 : * have supplied a freefunc.
5144 : */
5145 0 : if (HeapTupleIsValid(vardata->statsTuple) &&
5146 0 : !vardata->freefunc)
5147 0 : elog(ERROR, "no function provided to release variable stats with");
5148 : }
5149 3390 : else if (index->indpred == NIL)
5150 : {
5151 3390 : vardata->statsTuple =
5152 3390 : SearchSysCache3(STATRELATTINH,
5153 3390 : ObjectIdGetDatum(index->indexoid),
5154 3390 : Int16GetDatum(pos + 1),
5155 : BoolGetDatum(false));
5156 3390 : vardata->freefunc = ReleaseSysCache;
5157 :
5158 3390 : if (HeapTupleIsValid(vardata->statsTuple))
5159 : {
5160 : /* Get index's table for permission check */
5161 : RangeTblEntry *rte;
5162 : Oid userid;
5163 :
5164 2802 : rte = planner_rt_fetch(index->rel->relid, root);
5165 : Assert(rte->rtekind == RTE_RELATION);
5166 :
5167 : /*
5168 : * Use checkAsUser if it's set, in case we're
5169 : * accessing the table via a view.
5170 : */
5171 2802 : userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
5172 :
5173 : /*
5174 : * For simplicity, we insist on the whole
5175 : * table being selectable, rather than trying
5176 : * to identify which column(s) the index
5177 : * depends on. Also require all rows to be
5178 : * selectable --- there must be no
5179 : * securityQuals from security barrier views
5180 : * or RLS policies.
5181 : */
5182 2802 : vardata->acl_ok =
5183 5604 : rte->securityQuals == NIL &&
5184 2802 : (pg_class_aclcheck(rte->relid, userid,
5185 : ACL_SELECT) == ACLCHECK_OK);
5186 :
5187 : /*
5188 : * If the user doesn't have permissions to
5189 : * access an inheritance child relation, check
5190 : * the permissions of the table actually
5191 : * mentioned in the query, since most likely
5192 : * the user does have that permission. Note
5193 : * that whole-table select privilege on the
5194 : * parent doesn't quite guarantee that the
5195 : * user could read all columns of the child.
5196 : * But in practice it's unlikely that any
5197 : * interesting security violation could result
5198 : * from allowing access to the expression
5199 : * index's stats, so we allow it anyway. See
5200 : * similar code in examine_simple_variable()
5201 : * for additional comments.
5202 : */
5203 2802 : if (!vardata->acl_ok &&
5204 18 : root->append_rel_array != NULL)
5205 : {
5206 : AppendRelInfo *appinfo;
5207 12 : Index varno = index->rel->relid;
5208 :
5209 12 : appinfo = root->append_rel_array[varno];
5210 36 : while (appinfo &&
5211 24 : planner_rt_fetch(appinfo->parent_relid,
5212 24 : root)->rtekind == RTE_RELATION)
5213 : {
5214 24 : varno = appinfo->parent_relid;
5215 24 : appinfo = root->append_rel_array[varno];
5216 : }
5217 12 : if (varno != index->rel->relid)
5218 : {
5219 : /* Repeat access check on this rel */
5220 12 : rte = planner_rt_fetch(varno, root);
5221 : Assert(rte->rtekind == RTE_RELATION);
5222 :
5223 12 : userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
5224 :
5225 12 : vardata->acl_ok =
5226 24 : rte->securityQuals == NIL &&
5227 12 : (pg_class_aclcheck(rte->relid,
5228 : userid,
5229 : ACL_SELECT) == ACLCHECK_OK);
5230 : }
5231 : }
5232 : }
5233 : else
5234 : {
5235 : /* suppress leakproofness checks later */
5236 588 : vardata->acl_ok = true;
5237 : }
5238 : }
5239 3390 : if (vardata->statsTuple)
5240 2802 : break;
5241 : }
5242 1854 : indexpr_item = lnext(index->indexprs, indexpr_item);
5243 : }
5244 : }
5245 4656 : if (vardata->statsTuple)
5246 2802 : break;
5247 : }
5248 :
5249 : /*
5250 : * Search extended statistics for one with a matching expression.
5251 : * There might be multiple ones, so just grab the first one. In the
5252 : * future, we might consider the statistics target (and pick the most
5253 : * accurate statistics) and maybe some other parameters.
5254 : */
5255 84934 : foreach(slist, onerel->statlist)
5256 : {
5257 14886 : StatisticExtInfo *info = (StatisticExtInfo *) lfirst(slist);
5258 14886 : RangeTblEntry *rte = planner_rt_fetch(onerel->relid, root);
5259 : ListCell *expr_item;
5260 : int pos;
5261 :
5262 : /*
5263 : * Stop once we've found statistics for the expression (either
5264 : * from extended stats, or for an index in the preceding loop).
5265 : */
5266 14886 : if (vardata->statsTuple)
5267 840 : break;
5268 :
5269 : /* skip stats without per-expression stats */
5270 14046 : if (info->kind != STATS_EXT_EXPRESSIONS)
5271 8964 : continue;
5272 :
5273 5082 : pos = 0;
5274 10398 : foreach(expr_item, info->exprs)
5275 : {
5276 8670 : Node *expr = (Node *) lfirst(expr_item);
5277 :
5278 : Assert(expr);
5279 :
5280 : /* strip RelabelType before comparing it */
5281 8670 : if (expr && IsA(expr, RelabelType))
5282 0 : expr = (Node *) ((RelabelType *) expr)->arg;
5283 :
5284 : /* found a match, see if we can extract pg_statistic row */
5285 8670 : if (equal(node, expr))
5286 : {
5287 : Oid userid;
5288 :
5289 : /*
5290 : * XXX Not sure if we should cache the tuple somewhere.
5291 : * Now we just create a new copy every time.
5292 : */
5293 3354 : vardata->statsTuple =
5294 3354 : statext_expressions_load(info->statOid, rte->inh, pos);
5295 :
5296 3354 : vardata->freefunc = ReleaseDummy;
5297 :
5298 : /*
5299 : * Use checkAsUser if it's set, in case we're accessing
5300 : * the table via a view.
5301 : */
5302 3354 : userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
5303 :
5304 : /*
5305 : * For simplicity, we insist on the whole table being
5306 : * selectable, rather than trying to identify which
5307 : * column(s) the statistics object depends on. Also
5308 : * require all rows to be selectable --- there must be no
5309 : * securityQuals from security barrier views or RLS
5310 : * policies.
5311 : */
5312 3354 : vardata->acl_ok =
5313 6708 : rte->securityQuals == NIL &&
5314 3354 : (pg_class_aclcheck(rte->relid, userid,
5315 : ACL_SELECT) == ACLCHECK_OK);
5316 :
5317 : /*
5318 : * If the user doesn't have permissions to access an
5319 : * inheritance child relation, check the permissions of
5320 : * the table actually mentioned in the query, since most
5321 : * likely the user does have that permission. Note that
5322 : * whole-table select privilege on the parent doesn't
5323 : * quite guarantee that the user could read all columns of
5324 : * the child. But in practice it's unlikely that any
5325 : * interesting security violation could result from
5326 : * allowing access to the expression stats, so we allow it
5327 : * anyway. See similar code in examine_simple_variable()
5328 : * for additional comments.
5329 : */
5330 3354 : if (!vardata->acl_ok &&
5331 0 : root->append_rel_array != NULL)
5332 : {
5333 : AppendRelInfo *appinfo;
5334 0 : Index varno = onerel->relid;
5335 :
5336 0 : appinfo = root->append_rel_array[varno];
5337 0 : while (appinfo &&
5338 0 : planner_rt_fetch(appinfo->parent_relid,
5339 0 : root)->rtekind == RTE_RELATION)
5340 : {
5341 0 : varno = appinfo->parent_relid;
5342 0 : appinfo = root->append_rel_array[varno];
5343 : }
5344 0 : if (varno != onerel->relid)
5345 : {
5346 : /* Repeat access check on this rel */
5347 0 : rte = planner_rt_fetch(varno, root);
5348 : Assert(rte->rtekind == RTE_RELATION);
5349 :
5350 0 : userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
5351 :
5352 0 : vardata->acl_ok =
5353 0 : rte->securityQuals == NIL &&
5354 0 : (pg_class_aclcheck(rte->relid,
5355 : userid,
5356 : ACL_SELECT) == ACLCHECK_OK);
5357 : }
5358 : }
5359 :
5360 3354 : break;
5361 : }
5362 :
5363 5316 : pos++;
5364 : }
5365 : }
5366 : }
5367 : }
5368 :
5369 : /*
5370 : * examine_simple_variable
5371 : * Handle a simple Var for examine_variable
5372 : *
5373 : * This is split out as a subroutine so that we can recurse to deal with
5374 : * Vars referencing subqueries.
5375 : *
5376 : * We already filled in all the fields of *vardata except for the stats tuple.
5377 : */
5378 : static void
5379 2857616 : examine_simple_variable(PlannerInfo *root, Var *var,
5380 : VariableStatData *vardata)
5381 : {
5382 2857616 : RangeTblEntry *rte = root->simple_rte_array[var->varno];
5383 :
5384 : Assert(IsA(rte, RangeTblEntry));
5385 :
5386 2857616 : if (get_relation_stats_hook &&
5387 0 : (*get_relation_stats_hook) (root, rte, var->varattno, vardata))
5388 : {
5389 : /*
5390 : * The hook took control of acquiring a stats tuple. If it did supply
5391 : * a tuple, it'd better have supplied a freefunc.
5392 : */
5393 0 : if (HeapTupleIsValid(vardata->statsTuple) &&
5394 0 : !vardata->freefunc)
5395 0 : elog(ERROR, "no function provided to release variable stats with");
5396 : }
5397 2857616 : else if (rte->rtekind == RTE_RELATION)
5398 : {
5399 : /*
5400 : * Plain table or parent of an inheritance appendrel, so look up the
5401 : * column in pg_statistic
5402 : */
5403 2620440 : vardata->statsTuple = SearchSysCache3(STATRELATTINH,
5404 2620440 : ObjectIdGetDatum(rte->relid),
5405 2620440 : Int16GetDatum(var->varattno),
5406 2620440 : BoolGetDatum(rte->inh));
5407 2620440 : vardata->freefunc = ReleaseSysCache;
5408 :
5409 2620440 : if (HeapTupleIsValid(vardata->statsTuple))
5410 : {
5411 : Oid userid;
5412 :
5413 : /*
5414 : * Check if user has permission to read this column. We require
5415 : * all rows to be accessible, so there must be no securityQuals
5416 : * from security barrier views or RLS policies. Use checkAsUser
5417 : * if it's set, in case we're accessing the table via a view.
5418 : */
5419 1874054 : userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
5420 :
5421 1874054 : vardata->acl_ok =
5422 3748474 : rte->securityQuals == NIL &&
5423 1874012 : ((pg_class_aclcheck(rte->relid, userid,
5424 408 : ACL_SELECT) == ACLCHECK_OK) ||
5425 408 : (pg_attribute_aclcheck(rte->relid, var->varattno, userid,
5426 : ACL_SELECT) == ACLCHECK_OK));
5427 :
5428 : /*
5429 : * If the user doesn't have permissions to access an inheritance
5430 : * child relation or specifically this attribute, check the
5431 : * permissions of the table/column actually mentioned in the
5432 : * query, since most likely the user does have that permission
5433 : * (else the query will fail at runtime), and if the user can read
5434 : * the column there then he can get the values of the child table
5435 : * too. To do that, we must find out which of the root parent's
5436 : * attributes the child relation's attribute corresponds to.
5437 : */
5438 1874054 : if (!vardata->acl_ok && var->varattno > 0 &&
5439 84 : root->append_rel_array != NULL)
5440 : {
5441 : AppendRelInfo *appinfo;
5442 12 : Index varno = var->varno;
5443 12 : int varattno = var->varattno;
5444 12 : bool found = false;
5445 :
5446 12 : appinfo = root->append_rel_array[varno];
5447 :
5448 : /*
5449 : * Partitions are mapped to their immediate parent, not the
5450 : * root parent, so must be ready to walk up multiple
5451 : * AppendRelInfos. But stop if we hit a parent that is not
5452 : * RTE_RELATION --- that's a flattened UNION ALL subquery, not
5453 : * an inheritance parent.
5454 : */
5455 36 : while (appinfo &&
5456 24 : planner_rt_fetch(appinfo->parent_relid,
5457 24 : root)->rtekind == RTE_RELATION)
5458 : {
5459 : int parent_varattno;
5460 :
5461 24 : found = false;
5462 24 : if (varattno <= 0 || varattno > appinfo->num_child_cols)
5463 : break; /* safety check */
5464 24 : parent_varattno = appinfo->parent_colnos[varattno - 1];
5465 24 : if (parent_varattno == 0)
5466 0 : break; /* Var is local to child */
5467 :
5468 24 : varno = appinfo->parent_relid;
5469 24 : varattno = parent_varattno;
5470 24 : found = true;
5471 :
5472 : /* If the parent is itself a child, continue up. */
5473 24 : appinfo = root->append_rel_array[varno];
5474 : }
5475 :
5476 : /*
5477 : * In rare cases, the Var may be local to the child table, in
5478 : * which case, we've got to live with having no access to this
5479 : * column's stats.
5480 : */
5481 12 : if (!found)
5482 0 : return;
5483 :
5484 : /* Repeat the access check on this parent rel & column */
5485 12 : rte = planner_rt_fetch(varno, root);
5486 : Assert(rte->rtekind == RTE_RELATION);
5487 :
5488 12 : userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
5489 :
5490 12 : vardata->acl_ok =
5491 30 : rte->securityQuals == NIL &&
5492 12 : ((pg_class_aclcheck(rte->relid, userid,
5493 6 : ACL_SELECT) == ACLCHECK_OK) ||
5494 6 : (pg_attribute_aclcheck(rte->relid, varattno, userid,
5495 : ACL_SELECT) == ACLCHECK_OK));
5496 : }
5497 : }
5498 : else
5499 : {
5500 : /* suppress any possible leakproofness checks later */
5501 746386 : vardata->acl_ok = true;
5502 : }
5503 : }
5504 237176 : else if (rte->rtekind == RTE_SUBQUERY && !rte->inh)
5505 : {
5506 : /*
5507 : * Plain subquery (not one that was converted to an appendrel).
5508 : */
5509 14226 : Query *subquery = rte->subquery;
5510 : RelOptInfo *rel;
5511 : TargetEntry *ste;
5512 :
5513 : /*
5514 : * Punt if it's a whole-row var rather than a plain column reference.
5515 : */
5516 14226 : if (var->varattno == InvalidAttrNumber)
5517 0 : return;
5518 :
5519 : /*
5520 : * Punt if subquery uses set operations or GROUP BY, as these will
5521 : * mash underlying columns' stats beyond recognition. (Set ops are
5522 : * particularly nasty; if we forged ahead, we would return stats
5523 : * relevant to only the leftmost subselect...) DISTINCT is also
5524 : * problematic, but we check that later because there is a possibility
5525 : * of learning something even with it.
5526 : */
5527 14226 : if (subquery->setOperations ||
5528 11890 : subquery->groupClause ||
5529 10838 : subquery->groupingSets)
5530 3388 : return;
5531 :
5532 : /*
5533 : * OK, fetch RelOptInfo for subquery. Note that we don't change the
5534 : * rel returned in vardata, since caller expects it to be a rel of the
5535 : * caller's query level. Because we might already be recursing, we
5536 : * can't use that rel pointer either, but have to look up the Var's
5537 : * rel afresh.
5538 : */
5539 10838 : rel = find_base_rel(root, var->varno);
5540 :
5541 : /* If the subquery hasn't been planned yet, we have to punt */
5542 10838 : if (rel->subroot == NULL)
5543 0 : return;
5544 : Assert(IsA(rel->subroot, PlannerInfo));
5545 :
5546 : /*
5547 : * Switch our attention to the subquery as mangled by the planner. It
5548 : * was okay to look at the pre-planning version for the tests above,
5549 : * but now we need a Var that will refer to the subroot's live
5550 : * RelOptInfos. For instance, if any subquery pullup happened during
5551 : * planning, Vars in the targetlist might have gotten replaced, and we
5552 : * need to see the replacement expressions.
5553 : */
5554 10838 : subquery = rel->subroot->parse;
5555 : Assert(IsA(subquery, Query));
5556 :
5557 : /* Get the subquery output expression referenced by the upper Var */
5558 10838 : ste = get_tle_by_resno(subquery->targetList, var->varattno);
5559 10838 : if (ste == NULL || ste->resjunk)
5560 0 : elog(ERROR, "subquery %s does not have attribute %d",
5561 : rte->eref->aliasname, var->varattno);
5562 10838 : var = (Var *) ste->expr;
5563 :
5564 : /*
5565 : * If subquery uses DISTINCT, we can't make use of any stats for the
5566 : * variable ... but, if it's the only DISTINCT column, we are entitled
5567 : * to consider it unique. We do the test this way so that it works
5568 : * for cases involving DISTINCT ON.
5569 : */
5570 10838 : if (subquery->distinctClause)
5571 : {
5572 846 : if (list_length(subquery->distinctClause) == 1 &&
5573 120 : targetIsInSortList(ste, InvalidOid, subquery->distinctClause))
5574 120 : vardata->isunique = true;
5575 : /* cannot go further */
5576 726 : return;
5577 : }
5578 :
5579 : /*
5580 : * If the sub-query originated from a view with the security_barrier
5581 : * attribute, we must not look at the variable's statistics, though it
5582 : * seems all right to notice the existence of a DISTINCT clause. So
5583 : * stop here.
5584 : *
5585 : * This is probably a harsher restriction than necessary; it's
5586 : * certainly OK for the selectivity estimator (which is a C function,
5587 : * and therefore omnipotent anyway) to look at the statistics. But
5588 : * many selectivity estimators will happily *invoke the operator
5589 : * function* to try to work out a good estimate - and that's not OK.
5590 : * So for now, don't dig down for stats.
5591 : */
5592 10112 : if (rte->security_barrier)
5593 600 : return;
5594 :
5595 : /* Can only handle a simple Var of subquery's query level */
5596 9512 : if (var && IsA(var, Var) &&
5597 5430 : var->varlevelsup == 0)
5598 : {
5599 : /*
5600 : * OK, recurse into the subquery. Note that the original setting
5601 : * of vardata->isunique (which will surely be false) is left
5602 : * unchanged in this situation. That's what we want, since even
5603 : * if the underlying column is unique, the subquery may have
5604 : * joined to other tables in a way that creates duplicates.
5605 : */
5606 5430 : examine_simple_variable(rel->subroot, var, vardata);
5607 : }
5608 : }
5609 : else
5610 : {
5611 : /*
5612 : * Otherwise, the Var comes from a FUNCTION, VALUES, or CTE RTE. (We
5613 : * won't see RTE_JOIN here because join alias Vars have already been
5614 : * flattened.) There's not much we can do with function outputs, but
5615 : * maybe someday try to be smarter about VALUES and/or CTEs.
5616 : */
5617 : }
5618 : }
5619 :
5620 : /*
5621 : * Check whether it is permitted to call func_oid passing some of the
5622 : * pg_statistic data in vardata. We allow this either if the user has SELECT
5623 : * privileges on the table or column underlying the pg_statistic data or if
5624 : * the function is marked leak-proof.
5625 : */
5626 : bool
5627 834842 : statistic_proc_security_check(VariableStatData *vardata, Oid func_oid)
5628 : {
5629 834842 : if (vardata->acl_ok)
5630 834674 : return true;
5631 :
5632 168 : if (!OidIsValid(func_oid))
5633 0 : return false;
5634 :
5635 168 : if (get_func_leakproof(func_oid))
5636 0 : return true;
5637 :
5638 168 : ereport(DEBUG2,
5639 : (errmsg_internal("not using statistics because function \"%s\" is not leak-proof",
5640 : get_func_name(func_oid))));
5641 168 : return false;
5642 : }
5643 :
5644 : /*
5645 : * get_variable_numdistinct
5646 : * Estimate the number of distinct values of a variable.
5647 : *
5648 : * vardata: results of examine_variable
5649 : * *isdefault: set to true if the result is a default rather than based on
5650 : * anything meaningful.
5651 : *
5652 : * NB: be careful to produce a positive integral result, since callers may
5653 : * compare the result to exact integer counts, or might divide by it.
5654 : */
5655 : double
5656 2213520 : get_variable_numdistinct(VariableStatData *vardata, bool *isdefault)
5657 : {
5658 : double stadistinct;
5659 2213520 : double stanullfrac = 0.0;
5660 : double ntuples;
5661 :
5662 2213520 : *isdefault = false;
5663 :
5664 : /*
5665 : * Determine the stadistinct value to use. There are cases where we can
5666 : * get an estimate even without a pg_statistic entry, or can get a better
5667 : * value than is in pg_statistic. Grab stanullfrac too if we can find it
5668 : * (otherwise, assume no nulls, for lack of any better idea).
5669 : */
5670 2213520 : if (HeapTupleIsValid(vardata->statsTuple))
5671 : {
5672 : /* Use the pg_statistic entry */
5673 : Form_pg_statistic stats;
5674 :
5675 1576510 : stats = (Form_pg_statistic) GETSTRUCT(vardata->statsTuple);
5676 1576510 : stadistinct = stats->stadistinct;
5677 1576510 : stanullfrac = stats->stanullfrac;
5678 : }
5679 637010 : else if (vardata->vartype == BOOLOID)
5680 : {
5681 : /*
5682 : * Special-case boolean columns: presumably, two distinct values.
5683 : *
5684 : * Are there any other datatypes we should wire in special estimates
5685 : * for?
5686 : */
5687 210 : stadistinct = 2.0;
5688 : }
5689 636800 : else if (vardata->rel && vardata->rel->rtekind == RTE_VALUES)
5690 : {
5691 : /*
5692 : * If the Var represents a column of a VALUES RTE, assume it's unique.
5693 : * This could of course be very wrong, but it should tend to be true
5694 : * in well-written queries. We could consider examining the VALUES'
5695 : * contents to get some real statistics; but that only works if the
5696 : * entries are all constants, and it would be pretty expensive anyway.
5697 : */
5698 3792 : stadistinct = -1.0; /* unique (and all non null) */
5699 : }
5700 : else
5701 : {
5702 : /*
5703 : * We don't keep statistics for system columns, but in some cases we
5704 : * can infer distinctness anyway.
5705 : */
5706 633008 : if (vardata->var && IsA(vardata->var, Var))
5707 : {
5708 616992 : switch (((Var *) vardata->var)->varattno)
5709 : {
5710 1484 : case SelfItemPointerAttributeNumber:
5711 1484 : stadistinct = -1.0; /* unique (and all non null) */
5712 1484 : break;
5713 6336 : case TableOidAttributeNumber:
5714 6336 : stadistinct = 1.0; /* only 1 value */
5715 6336 : break;
5716 609172 : default:
5717 609172 : stadistinct = 0.0; /* means "unknown" */
5718 609172 : break;
5719 : }
5720 : }
5721 : else
5722 16016 : stadistinct = 0.0; /* means "unknown" */
5723 :
5724 : /*
5725 : * XXX consider using estimate_num_groups on expressions?
5726 : */
5727 : }
5728 :
5729 : /*
5730 : * If there is a unique index or DISTINCT clause for the variable, assume
5731 : * it is unique no matter what pg_statistic says; the statistics could be
5732 : * out of date, or we might have found a partial unique index that proves
5733 : * the var is unique for this query. However, we'd better still believe
5734 : * the null-fraction statistic.
5735 : */
5736 2213520 : if (vardata->isunique)
5737 700424 : stadistinct = -1.0 * (1.0 - stanullfrac);
5738 :
5739 : /*
5740 : * If we had an absolute estimate, use that.
5741 : */
5742 2213520 : if (stadistinct > 0.0)
5743 482258 : return clamp_row_est(stadistinct);
5744 :
5745 : /*
5746 : * Otherwise we need to get the relation size; punt if not available.
5747 : */
5748 1731262 : if (vardata->rel == NULL)
5749 : {
5750 198 : *isdefault = true;
5751 198 : return DEFAULT_NUM_DISTINCT;
5752 : }
5753 1731064 : ntuples = vardata->rel->tuples;
5754 1731064 : if (ntuples <= 0.0)
5755 : {
5756 83592 : *isdefault = true;
5757 83592 : return DEFAULT_NUM_DISTINCT;
5758 : }
5759 :
5760 : /*
5761 : * If we had a relative estimate, use that.
5762 : */
5763 1647472 : if (stadistinct < 0.0)
5764 1208724 : return clamp_row_est(-stadistinct * ntuples);
5765 :
5766 : /*
5767 : * With no data, estimate ndistinct = ntuples if the table is small, else
5768 : * use default. We use DEFAULT_NUM_DISTINCT as the cutoff for "small" so
5769 : * that the behavior isn't discontinuous.
5770 : */
5771 438748 : if (ntuples < DEFAULT_NUM_DISTINCT)
5772 255790 : return clamp_row_est(ntuples);
5773 :
5774 182958 : *isdefault = true;
5775 182958 : return DEFAULT_NUM_DISTINCT;
5776 : }
5777 :
5778 : /*
5779 : * get_variable_range
5780 : * Estimate the minimum and maximum value of the specified variable.
5781 : * If successful, store values in *min and *max, and return true.
5782 : * If no data available, return false.
5783 : *
5784 : * sortop is the "<" comparison operator to use. This should generally
5785 : * be "<" not ">", as only the former is likely to be found in pg_statistic.
5786 : * The collation must be specified too.
5787 : */
5788 : static bool
5789 143122 : get_variable_range(PlannerInfo *root, VariableStatData *vardata,
5790 : Oid sortop, Oid collation,
5791 : Datum *min, Datum *max)
5792 : {
5793 143122 : Datum tmin = 0;
5794 143122 : Datum tmax = 0;
5795 143122 : bool have_data = false;
5796 : int16 typLen;
5797 : bool typByVal;
5798 : Oid opfuncoid;
5799 : FmgrInfo opproc;
5800 : AttStatsSlot sslot;
5801 :
5802 : /*
5803 : * XXX It's very tempting to try to use the actual column min and max, if
5804 : * we can get them relatively-cheaply with an index probe. However, since
5805 : * this function is called many times during join planning, that could
5806 : * have unpleasant effects on planning speed. Need more investigation
5807 : * before enabling this.
5808 : */
5809 : #ifdef NOT_USED
5810 : if (get_actual_variable_range(root, vardata, sortop, collation, min, max))
5811 : return true;
5812 : #endif
5813 :
5814 143122 : if (!HeapTupleIsValid(vardata->statsTuple))
5815 : {
5816 : /* no stats available, so default result */
5817 39740 : return false;
5818 : }
5819 :
5820 : /*
5821 : * If we can't apply the sortop to the stats data, just fail. In
5822 : * principle, if there's a histogram and no MCVs, we could return the
5823 : * histogram endpoints without ever applying the sortop ... but it's
5824 : * probably not worth trying, because whatever the caller wants to do with
5825 : * the endpoints would likely fail the security check too.
5826 : */
5827 103382 : if (!statistic_proc_security_check(vardata,
5828 103382 : (opfuncoid = get_opcode(sortop))))
5829 0 : return false;
5830 :
5831 103382 : opproc.fn_oid = InvalidOid; /* mark this as not looked up yet */
5832 :
5833 103382 : get_typlenbyval(vardata->atttype, &typLen, &typByVal);
5834 :
5835 : /*
5836 : * If there is a histogram with the ordering we want, grab the first and
5837 : * last values.
5838 : */
5839 103382 : if (get_attstatsslot(&sslot, vardata->statsTuple,
5840 : STATISTIC_KIND_HISTOGRAM, sortop,
5841 : ATTSTATSSLOT_VALUES))
5842 : {
5843 80756 : if (sslot.stacoll == collation && sslot.nvalues > 0)
5844 : {
5845 80756 : tmin = datumCopy(sslot.values[0], typByVal, typLen);
5846 80756 : tmax = datumCopy(sslot.values[sslot.nvalues - 1], typByVal, typLen);
5847 80756 : have_data = true;
5848 : }
5849 80756 : free_attstatsslot(&sslot);
5850 : }
5851 :
5852 : /*
5853 : * Otherwise, if there is a histogram with some other ordering, scan it
5854 : * and get the min and max values according to the ordering we want. This
5855 : * of course may not find values that are really extremal according to our
5856 : * ordering, but it beats ignoring available data.
5857 : */
5858 126008 : if (!have_data &&
5859 22626 : get_attstatsslot(&sslot, vardata->statsTuple,
5860 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
5861 : ATTSTATSSLOT_VALUES))
5862 : {
5863 0 : get_stats_slot_range(&sslot, opfuncoid, &opproc,
5864 : collation, typLen, typByVal,
5865 : &tmin, &tmax, &have_data);
5866 0 : free_attstatsslot(&sslot);
5867 : }
5868 :
5869 : /*
5870 : * If we have most-common-values info, look for extreme MCVs. This is
5871 : * needed even if we also have a histogram, since the histogram excludes
5872 : * the MCVs. However, if we *only* have MCVs and no histogram, we should
5873 : * be pretty wary of deciding that that is a full representation of the
5874 : * data. Proceed only if the MCVs represent the whole table (to within
5875 : * roundoff error).
5876 : */
5877 103382 : if (get_attstatsslot(&sslot, vardata->statsTuple,
5878 : STATISTIC_KIND_MCV, InvalidOid,
5879 103382 : have_data ? ATTSTATSSLOT_VALUES :
5880 : (ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS)))
5881 : {
5882 47058 : bool use_mcvs = have_data;
5883 :
5884 47058 : if (!have_data)
5885 : {
5886 21822 : double sumcommon = 0.0;
5887 : double nullfrac;
5888 : int i;
5889 :
5890 116012 : for (i = 0; i < sslot.nnumbers; i++)
5891 94190 : sumcommon += sslot.numbers[i];
5892 21822 : nullfrac = ((Form_pg_statistic) GETSTRUCT(vardata->statsTuple))->stanullfrac;
5893 21822 : if (sumcommon + nullfrac > 0.99999)
5894 20950 : use_mcvs = true;
5895 : }
5896 :
5897 47058 : if (use_mcvs)
5898 46186 : get_stats_slot_range(&sslot, opfuncoid, &opproc,
5899 : collation, typLen, typByVal,
5900 : &tmin, &tmax, &have_data);
5901 47058 : free_attstatsslot(&sslot);
5902 : }
5903 :
5904 103382 : *min = tmin;
5905 103382 : *max = tmax;
5906 103382 : return have_data;
5907 : }
5908 :
5909 : /*
5910 : * get_stats_slot_range: scan sslot for min/max values
5911 : *
5912 : * Subroutine for get_variable_range: update min/max/have_data according
5913 : * to what we find in the statistics array.
5914 : */
5915 : static void
5916 46186 : get_stats_slot_range(AttStatsSlot *sslot, Oid opfuncoid, FmgrInfo *opproc,
5917 : Oid collation, int16 typLen, bool typByVal,
5918 : Datum *min, Datum *max, bool *p_have_data)
5919 : {
5920 46186 : Datum tmin = *min;
5921 46186 : Datum tmax = *max;
5922 46186 : bool have_data = *p_have_data;
5923 46186 : bool found_tmin = false;
5924 46186 : bool found_tmax = false;
5925 :
5926 : /* Look up the comparison function, if we didn't already do so */
5927 46186 : if (opproc->fn_oid != opfuncoid)
5928 46186 : fmgr_info(opfuncoid, opproc);
5929 :
5930 : /* Scan all the slot's values */
5931 1236130 : for (int i = 0; i < sslot->nvalues; i++)
5932 : {
5933 1189944 : if (!have_data)
5934 : {
5935 20950 : tmin = tmax = sslot->values[i];
5936 20950 : found_tmin = found_tmax = true;
5937 20950 : *p_have_data = have_data = true;
5938 20950 : continue;
5939 : }
5940 1168994 : if (DatumGetBool(FunctionCall2Coll(opproc,
5941 : collation,
5942 : sslot->values[i], tmin)))
5943 : {
5944 30838 : tmin = sslot->values[i];
5945 30838 : found_tmin = true;
5946 : }
5947 1168994 : if (DatumGetBool(FunctionCall2Coll(opproc,
5948 : collation,
5949 : tmax, sslot->values[i])))
5950 : {
5951 52744 : tmax = sslot->values[i];
5952 52744 : found_tmax = true;
5953 : }
5954 : }
5955 :
5956 : /*
5957 : * Copy the slot's values, if we found new extreme values.
5958 : */
5959 46186 : if (found_tmin)
5960 38654 : *min = datumCopy(tmin, typByVal, typLen);
5961 46186 : if (found_tmax)
5962 23070 : *max = datumCopy(tmax, typByVal, typLen);
5963 46186 : }
5964 :
5965 :
5966 : /*
5967 : * get_actual_variable_range
5968 : * Attempt to identify the current *actual* minimum and/or maximum
5969 : * of the specified variable, by looking for a suitable btree index
5970 : * and fetching its low and/or high values.
5971 : * If successful, store values in *min and *max, and return true.
5972 : * (Either pointer can be NULL if that endpoint isn't needed.)
5973 : * If no data available, return false.
5974 : *
5975 : * sortop is the "<" comparison operator to use.
5976 : * collation is the required collation.
5977 : */
5978 : static bool
5979 130226 : get_actual_variable_range(PlannerInfo *root, VariableStatData *vardata,
5980 : Oid sortop, Oid collation,
5981 : Datum *min, Datum *max)
5982 : {
5983 130226 : bool have_data = false;
5984 130226 : RelOptInfo *rel = vardata->rel;
5985 : RangeTblEntry *rte;
5986 : ListCell *lc;
5987 :
5988 : /* No hope if no relation or it doesn't have indexes */
5989 130226 : if (rel == NULL || rel->indexlist == NIL)
5990 13246 : return false;
5991 : /* If it has indexes it must be a plain relation */
5992 116980 : rte = root->simple_rte_array[rel->relid];
5993 : Assert(rte->rtekind == RTE_RELATION);
5994 :
5995 : /* Search through the indexes to see if any match our problem */
5996 248820 : foreach(lc, rel->indexlist)
5997 : {
5998 215062 : IndexOptInfo *index = (IndexOptInfo *) lfirst(lc);
5999 : ScanDirection indexscandir;
6000 :
6001 : /* Ignore non-btree indexes */
6002 215062 : if (index->relam != BTREE_AM_OID)
6003 0 : continue;
6004 :
6005 : /*
6006 : * Ignore partial indexes --- we only want stats that cover the entire
6007 : * relation.
6008 : */
6009 215062 : if (index->indpred != NIL)
6010 180 : continue;
6011 :
6012 : /*
6013 : * The index list might include hypothetical indexes inserted by a
6014 : * get_relation_info hook --- don't try to access them.
6015 : */
6016 214882 : if (index->hypothetical)
6017 0 : continue;
6018 :
6019 : /*
6020 : * The first index column must match the desired variable, sortop, and
6021 : * collation --- but we can use a descending-order index.
6022 : */
6023 214882 : if (collation != index->indexcollations[0])
6024 34776 : continue; /* test first 'cause it's cheapest */
6025 180106 : if (!match_index_to_operand(vardata->var, 0, index))
6026 96884 : continue;
6027 83222 : switch (get_op_opfamily_strategy(sortop, index->sortopfamily[0]))
6028 : {
6029 83222 : case BTLessStrategyNumber:
6030 83222 : if (index->reverse_sort[0])
6031 0 : indexscandir = BackwardScanDirection;
6032 : else
6033 83222 : indexscandir = ForwardScanDirection;
6034 83222 : break;
6035 0 : case BTGreaterStrategyNumber:
6036 0 : if (index->reverse_sort[0])
6037 0 : indexscandir = ForwardScanDirection;
6038 : else
6039 0 : indexscandir = BackwardScanDirection;
6040 0 : break;
6041 0 : default:
6042 : /* index doesn't match the sortop */
6043 0 : continue;
6044 : }
6045 :
6046 : /*
6047 : * Found a suitable index to extract data from. Set up some data that
6048 : * can be used by both invocations of get_actual_variable_endpoint.
6049 : */
6050 : {
6051 : MemoryContext tmpcontext;
6052 : MemoryContext oldcontext;
6053 : Relation heapRel;
6054 : Relation indexRel;
6055 : TupleTableSlot *slot;
6056 : int16 typLen;
6057 : bool typByVal;
6058 : ScanKeyData scankeys[1];
6059 :
6060 : /* Make sure any cruft gets recycled when we're done */
6061 83222 : tmpcontext = AllocSetContextCreate(CurrentMemoryContext,
6062 : "get_actual_variable_range workspace",
6063 : ALLOCSET_DEFAULT_SIZES);
6064 83222 : oldcontext = MemoryContextSwitchTo(tmpcontext);
6065 :
6066 : /*
6067 : * Open the table and index so we can read from them. We should
6068 : * already have some type of lock on each.
6069 : */
6070 83222 : heapRel = table_open(rte->relid, NoLock);
6071 83222 : indexRel = index_open(index->indexoid, NoLock);
6072 :
6073 : /* build some stuff needed for indexscan execution */
6074 83222 : slot = table_slot_create(heapRel, NULL);
6075 83222 : get_typlenbyval(vardata->atttype, &typLen, &typByVal);
6076 :
6077 : /* set up an IS NOT NULL scan key so that we ignore nulls */
6078 83222 : ScanKeyEntryInitialize(&scankeys[0],
6079 : SK_ISNULL | SK_SEARCHNOTNULL,
6080 : 1, /* index col to scan */
6081 : InvalidStrategy, /* no strategy */
6082 : InvalidOid, /* no strategy subtype */
6083 : InvalidOid, /* no collation */
6084 : InvalidOid, /* no reg proc for this */
6085 : (Datum) 0); /* constant */
6086 :
6087 : /* If min is requested ... */
6088 83222 : if (min)
6089 : {
6090 46982 : have_data = get_actual_variable_endpoint(heapRel,
6091 : indexRel,
6092 : indexscandir,
6093 : scankeys,
6094 : typLen,
6095 : typByVal,
6096 : slot,
6097 : oldcontext,
6098 : min);
6099 : }
6100 : else
6101 : {
6102 : /* If min not requested, assume index is nonempty */
6103 36240 : have_data = true;
6104 : }
6105 :
6106 : /* If max is requested, and we didn't find the index is empty */
6107 83222 : if (max && have_data)
6108 : {
6109 : /* scan in the opposite direction; all else is the same */
6110 37028 : have_data = get_actual_variable_endpoint(heapRel,
6111 : indexRel,
6112 37028 : -indexscandir,
6113 : scankeys,
6114 : typLen,
6115 : typByVal,
6116 : slot,
6117 : oldcontext,
6118 : max);
6119 : }
6120 :
6121 : /* Clean everything up */
6122 83222 : ExecDropSingleTupleTableSlot(slot);
6123 :
6124 83222 : index_close(indexRel, NoLock);
6125 83222 : table_close(heapRel, NoLock);
6126 :
6127 83222 : MemoryContextSwitchTo(oldcontext);
6128 83222 : MemoryContextDelete(tmpcontext);
6129 :
6130 : /* And we're done */
6131 83222 : break;
6132 : }
6133 : }
6134 :
6135 116980 : return have_data;
6136 : }
6137 :
6138 : /*
6139 : * Get one endpoint datum (min or max depending on indexscandir) from the
6140 : * specified index. Return true if successful, false if index is empty.
6141 : * On success, endpoint value is stored to *endpointDatum (and copied into
6142 : * outercontext).
6143 : *
6144 : * scankeys is a 1-element scankey array set up to reject nulls.
6145 : * typLen/typByVal describe the datatype of the index's first column.
6146 : * tableslot is a slot suitable to hold table tuples, in case we need
6147 : * to probe the heap.
6148 : * (We could compute these values locally, but that would mean computing them
6149 : * twice when get_actual_variable_range needs both the min and the max.)
6150 : */
6151 : static bool
6152 84010 : get_actual_variable_endpoint(Relation heapRel,
6153 : Relation indexRel,
6154 : ScanDirection indexscandir,
6155 : ScanKey scankeys,
6156 : int16 typLen,
6157 : bool typByVal,
6158 : TupleTableSlot *tableslot,
6159 : MemoryContext outercontext,
6160 : Datum *endpointDatum)
6161 : {
6162 84010 : bool have_data = false;
6163 : SnapshotData SnapshotNonVacuumable;
6164 : IndexScanDesc index_scan;
6165 84010 : Buffer vmbuffer = InvalidBuffer;
6166 : ItemPointer tid;
6167 : Datum values[INDEX_MAX_KEYS];
6168 : bool isnull[INDEX_MAX_KEYS];
6169 : MemoryContext oldcontext;
6170 :
6171 : /*
6172 : * We use the index-only-scan machinery for this. With mostly-static
6173 : * tables that's a win because it avoids a heap visit. It's also a win
6174 : * for dynamic data, but the reason is less obvious; read on for details.
6175 : *
6176 : * In principle, we should scan the index with our current active
6177 : * snapshot, which is the best approximation we've got to what the query
6178 : * will see when executed. But that won't be exact if a new snap is taken
6179 : * before running the query, and it can be very expensive if a lot of
6180 : * recently-dead or uncommitted rows exist at the beginning or end of the
6181 : * index (because we'll laboriously fetch each one and reject it).
6182 : * Instead, we use SnapshotNonVacuumable. That will accept recently-dead
6183 : * and uncommitted rows as well as normal visible rows. On the other
6184 : * hand, it will reject known-dead rows, and thus not give a bogus answer
6185 : * when the extreme value has been deleted (unless the deletion was quite
6186 : * recent); that case motivates not using SnapshotAny here.
6187 : *
6188 : * A crucial point here is that SnapshotNonVacuumable, with
6189 : * GlobalVisTestFor(heapRel) as horizon, yields the inverse of the
6190 : * condition that the indexscan will use to decide that index entries are
6191 : * killable (see heap_hot_search_buffer()). Therefore, if the snapshot
6192 : * rejects a tuple (or more precisely, all tuples of a HOT chain) and we
6193 : * have to continue scanning past it, we know that the indexscan will mark
6194 : * that index entry killed. That means that the next
6195 : * get_actual_variable_endpoint() call will not have to re-consider that
6196 : * index entry. In this way we avoid repetitive work when this function
6197 : * is used a lot during planning.
6198 : *
6199 : * But using SnapshotNonVacuumable creates a hazard of its own. In a
6200 : * recently-created index, some index entries may point at "broken" HOT
6201 : * chains in which not all the tuple versions contain data matching the
6202 : * index entry. The live tuple version(s) certainly do match the index,
6203 : * but SnapshotNonVacuumable can accept recently-dead tuple versions that
6204 : * don't match. Hence, if we took data from the selected heap tuple, we
6205 : * might get a bogus answer that's not close to the index extremal value,
6206 : * or could even be NULL. We avoid this hazard because we take the data
6207 : * from the index entry not the heap.
6208 : */
6209 84010 : InitNonVacuumableSnapshot(SnapshotNonVacuumable,
6210 : GlobalVisTestFor(heapRel));
6211 :
6212 84010 : index_scan = index_beginscan(heapRel, indexRel,
6213 : &SnapshotNonVacuumable,
6214 : 1, 0);
6215 : /* Set it up for index-only scan */
6216 84010 : index_scan->xs_want_itup = true;
6217 84010 : index_rescan(index_scan, scankeys, 1, NULL, 0);
6218 :
6219 : /* Fetch first/next tuple in specified direction */
6220 91204 : while ((tid = index_getnext_tid(index_scan, indexscandir)) != NULL)
6221 : {
6222 91204 : if (!VM_ALL_VISIBLE(heapRel,
6223 : ItemPointerGetBlockNumber(tid),
6224 : &vmbuffer))
6225 : {
6226 : /* Rats, we have to visit the heap to check visibility */
6227 71592 : if (!index_fetch_heap(index_scan, tableslot))
6228 7194 : continue; /* no visible tuple, try next index entry */
6229 :
6230 : /* We don't actually need the heap tuple for anything */
6231 64398 : ExecClearTuple(tableslot);
6232 :
6233 : /*
6234 : * We don't care whether there's more than one visible tuple in
6235 : * the HOT chain; if any are visible, that's good enough.
6236 : */
6237 : }
6238 :
6239 : /*
6240 : * We expect that btree will return data in IndexTuple not HeapTuple
6241 : * format. It's not lossy either.
6242 : */
6243 84010 : if (!index_scan->xs_itup)
6244 0 : elog(ERROR, "no data returned for index-only scan");
6245 84010 : if (index_scan->xs_recheck)
6246 0 : elog(ERROR, "unexpected recheck indication from btree");
6247 :
6248 : /* OK to deconstruct the index tuple */
6249 84010 : index_deform_tuple(index_scan->xs_itup,
6250 : index_scan->xs_itupdesc,
6251 : values, isnull);
6252 :
6253 : /* Shouldn't have got a null, but be careful */
6254 84010 : if (isnull[0])
6255 0 : elog(ERROR, "found unexpected null value in index \"%s\"",
6256 : RelationGetRelationName(indexRel));
6257 :
6258 : /* Copy the index column value out to caller's context */
6259 84010 : oldcontext = MemoryContextSwitchTo(outercontext);
6260 84010 : *endpointDatum = datumCopy(values[0], typByVal, typLen);
6261 84010 : MemoryContextSwitchTo(oldcontext);
6262 84010 : have_data = true;
6263 84010 : break;
6264 : }
6265 :
6266 84010 : if (vmbuffer != InvalidBuffer)
6267 67844 : ReleaseBuffer(vmbuffer);
6268 84010 : index_endscan(index_scan);
6269 :
6270 84010 : return have_data;
6271 : }
6272 :
6273 : /*
6274 : * find_join_input_rel
6275 : * Look up the input relation for a join.
6276 : *
6277 : * We assume that the input relation's RelOptInfo must have been constructed
6278 : * already.
6279 : */
6280 : static RelOptInfo *
6281 6838 : find_join_input_rel(PlannerInfo *root, Relids relids)
6282 : {
6283 6838 : RelOptInfo *rel = NULL;
6284 :
6285 6838 : switch (bms_membership(relids))
6286 : {
6287 0 : case BMS_EMPTY_SET:
6288 : /* should not happen */
6289 0 : break;
6290 6622 : case BMS_SINGLETON:
6291 6622 : rel = find_base_rel(root, bms_singleton_member(relids));
6292 6622 : break;
6293 216 : case BMS_MULTIPLE:
6294 216 : rel = find_join_rel(root, relids);
6295 216 : break;
6296 : }
6297 :
6298 6838 : if (rel == NULL)
6299 0 : elog(ERROR, "could not find RelOptInfo for given relids");
6300 :
6301 6838 : return rel;
6302 : }
6303 :
6304 :
6305 : /*-------------------------------------------------------------------------
6306 : *
6307 : * Index cost estimation functions
6308 : *
6309 : *-------------------------------------------------------------------------
6310 : */
6311 :
6312 : /*
6313 : * Extract the actual indexquals (as RestrictInfos) from an IndexClause list
6314 : */
6315 : List *
6316 507828 : get_quals_from_indexclauses(List *indexclauses)
6317 : {
6318 507828 : List *result = NIL;
6319 : ListCell *lc;
6320 :
6321 895250 : foreach(lc, indexclauses)
6322 : {
6323 387422 : IndexClause *iclause = lfirst_node(IndexClause, lc);
6324 : ListCell *lc2;
6325 :
6326 777372 : foreach(lc2, iclause->indexquals)
6327 : {
6328 389950 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc2);
6329 :
6330 389950 : result = lappend(result, rinfo);
6331 : }
6332 : }
6333 507828 : return result;
6334 : }
6335 :
6336 : /*
6337 : * Compute the total evaluation cost of the comparison operands in a list
6338 : * of index qual expressions. Since we know these will be evaluated just
6339 : * once per scan, there's no need to distinguish startup from per-row cost.
6340 : *
6341 : * This can be used either on the result of get_quals_from_indexclauses(),
6342 : * or directly on an indexorderbys list. In both cases, we expect that the
6343 : * index key expression is on the left side of binary clauses.
6344 : */
6345 : Cost
6346 1003458 : index_other_operands_eval_cost(PlannerInfo *root, List *indexquals)
6347 : {
6348 1003458 : Cost qual_arg_cost = 0;
6349 : ListCell *lc;
6350 :
6351 1393858 : foreach(lc, indexquals)
6352 : {
6353 390400 : Expr *clause = (Expr *) lfirst(lc);
6354 : Node *other_operand;
6355 : QualCost index_qual_cost;
6356 :
6357 : /*
6358 : * Index quals will have RestrictInfos, indexorderbys won't. Look
6359 : * through RestrictInfo if present.
6360 : */
6361 390400 : if (IsA(clause, RestrictInfo))
6362 389938 : clause = ((RestrictInfo *) clause)->clause;
6363 :
6364 390400 : if (IsA(clause, OpExpr))
6365 : {
6366 381150 : OpExpr *op = (OpExpr *) clause;
6367 :
6368 381150 : other_operand = (Node *) lsecond(op->args);
6369 : }
6370 9250 : else if (IsA(clause, RowCompareExpr))
6371 : {
6372 132 : RowCompareExpr *rc = (RowCompareExpr *) clause;
6373 :
6374 132 : other_operand = (Node *) rc->rargs;
6375 : }
6376 9118 : else if (IsA(clause, ScalarArrayOpExpr))
6377 : {
6378 6548 : ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
6379 :
6380 6548 : other_operand = (Node *) lsecond(saop->args);
6381 : }
6382 2570 : else if (IsA(clause, NullTest))
6383 : {
6384 2570 : other_operand = NULL;
6385 : }
6386 : else
6387 : {
6388 0 : elog(ERROR, "unsupported indexqual type: %d",
6389 : (int) nodeTag(clause));
6390 : other_operand = NULL; /* keep compiler quiet */
6391 : }
6392 :
6393 390400 : cost_qual_eval_node(&index_qual_cost, other_operand, root);
6394 390400 : qual_arg_cost += index_qual_cost.startup + index_qual_cost.per_tuple;
6395 : }
6396 1003458 : return qual_arg_cost;
6397 : }
6398 :
6399 : void
6400 495642 : genericcostestimate(PlannerInfo *root,
6401 : IndexPath *path,
6402 : double loop_count,
6403 : GenericCosts *costs)
6404 : {
6405 495642 : IndexOptInfo *index = path->indexinfo;
6406 495642 : List *indexQuals = get_quals_from_indexclauses(path->indexclauses);
6407 495642 : List *indexOrderBys = path->indexorderbys;
6408 : Cost indexStartupCost;
6409 : Cost indexTotalCost;
6410 : Selectivity indexSelectivity;
6411 : double indexCorrelation;
6412 : double numIndexPages;
6413 : double numIndexTuples;
6414 : double spc_random_page_cost;
6415 : double num_sa_scans;
6416 : double num_outer_scans;
6417 : double num_scans;
6418 : double qual_op_cost;
6419 : double qual_arg_cost;
6420 : List *selectivityQuals;
6421 : ListCell *l;
6422 :
6423 : /*
6424 : * If the index is partial, AND the index predicate with the explicitly
6425 : * given indexquals to produce a more accurate idea of the index
6426 : * selectivity.
6427 : */
6428 495642 : selectivityQuals = add_predicate_to_index_quals(index, indexQuals);
6429 :
6430 : /*
6431 : * Check for ScalarArrayOpExpr index quals, and estimate the number of
6432 : * index scans that will be performed.
6433 : */
6434 495642 : num_sa_scans = 1;
6435 873208 : foreach(l, indexQuals)
6436 : {
6437 377566 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(l);
6438 :
6439 377566 : if (IsA(rinfo->clause, ScalarArrayOpExpr))
6440 : {
6441 6542 : ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) rinfo->clause;
6442 6542 : int alength = estimate_array_length(lsecond(saop->args));
6443 :
6444 6542 : if (alength > 1)
6445 6446 : num_sa_scans *= alength;
6446 : }
6447 : }
6448 :
6449 : /* Estimate the fraction of main-table tuples that will be visited */
6450 495642 : indexSelectivity = clauselist_selectivity(root, selectivityQuals,
6451 495642 : index->rel->relid,
6452 : JOIN_INNER,
6453 : NULL);
6454 :
6455 : /*
6456 : * If caller didn't give us an estimate, estimate the number of index
6457 : * tuples that will be visited. We do it in this rather peculiar-looking
6458 : * way in order to get the right answer for partial indexes.
6459 : */
6460 495642 : numIndexTuples = costs->numIndexTuples;
6461 495642 : if (numIndexTuples <= 0.0)
6462 : {
6463 34482 : numIndexTuples = indexSelectivity * index->rel->tuples;
6464 :
6465 : /*
6466 : * The above calculation counts all the tuples visited across all
6467 : * scans induced by ScalarArrayOpExpr nodes. We want to consider the
6468 : * average per-indexscan number, so adjust. This is a handy place to
6469 : * round to integer, too. (If caller supplied tuple estimate, it's
6470 : * responsible for handling these considerations.)
6471 : */
6472 34482 : numIndexTuples = rint(numIndexTuples / num_sa_scans);
6473 : }
6474 :
6475 : /*
6476 : * We can bound the number of tuples by the index size in any case. Also,
6477 : * always estimate at least one tuple is touched, even when
6478 : * indexSelectivity estimate is tiny.
6479 : */
6480 495642 : if (numIndexTuples > index->tuples)
6481 4264 : numIndexTuples = index->tuples;
6482 495642 : if (numIndexTuples < 1.0)
6483 33698 : numIndexTuples = 1.0;
6484 :
6485 : /*
6486 : * Estimate the number of index pages that will be retrieved.
6487 : *
6488 : * We use the simplistic method of taking a pro-rata fraction of the total
6489 : * number of index pages. In effect, this counts only leaf pages and not
6490 : * any overhead such as index metapage or upper tree levels.
6491 : *
6492 : * In practice access to upper index levels is often nearly free because
6493 : * those tend to stay in cache under load; moreover, the cost involved is
6494 : * highly dependent on index type. We therefore ignore such costs here
6495 : * and leave it to the caller to add a suitable charge if needed.
6496 : */
6497 495642 : if (index->pages > 1 && index->tuples > 1)
6498 465278 : numIndexPages = ceil(numIndexTuples * index->pages / index->tuples);
6499 : else
6500 30364 : numIndexPages = 1.0;
6501 :
6502 : /* fetch estimated page cost for tablespace containing index */
6503 495642 : get_tablespace_page_costs(index->reltablespace,
6504 : &spc_random_page_cost,
6505 : NULL);
6506 :
6507 : /*
6508 : * Now compute the disk access costs.
6509 : *
6510 : * The above calculations are all per-index-scan. However, if we are in a
6511 : * nestloop inner scan, we can expect the scan to be repeated (with
6512 : * different search keys) for each row of the outer relation. Likewise,
6513 : * ScalarArrayOpExpr quals result in multiple index scans. This creates
6514 : * the potential for cache effects to reduce the number of disk page
6515 : * fetches needed. We want to estimate the average per-scan I/O cost in
6516 : * the presence of caching.
6517 : *
6518 : * We use the Mackert-Lohman formula (see costsize.c for details) to
6519 : * estimate the total number of page fetches that occur. While this
6520 : * wasn't what it was designed for, it seems a reasonable model anyway.
6521 : * Note that we are counting pages not tuples anymore, so we take N = T =
6522 : * index size, as if there were one "tuple" per page.
6523 : */
6524 495642 : num_outer_scans = loop_count;
6525 495642 : num_scans = num_sa_scans * num_outer_scans;
6526 :
6527 495642 : if (num_scans > 1)
6528 : {
6529 : double pages_fetched;
6530 :
6531 : /* total page fetches ignoring cache effects */
6532 57408 : pages_fetched = numIndexPages * num_scans;
6533 :
6534 : /* use Mackert and Lohman formula to adjust for cache effects */
6535 57408 : pages_fetched = index_pages_fetched(pages_fetched,
6536 : index->pages,
6537 57408 : (double) index->pages,
6538 : root);
6539 :
6540 : /*
6541 : * Now compute the total disk access cost, and then report a pro-rated
6542 : * share for each outer scan. (Don't pro-rate for ScalarArrayOpExpr,
6543 : * since that's internal to the indexscan.)
6544 : */
6545 57408 : indexTotalCost = (pages_fetched * spc_random_page_cost)
6546 : / num_outer_scans;
6547 : }
6548 : else
6549 : {
6550 : /*
6551 : * For a single index scan, we just charge spc_random_page_cost per
6552 : * page touched.
6553 : */
6554 438234 : indexTotalCost = numIndexPages * spc_random_page_cost;
6555 : }
6556 :
6557 : /*
6558 : * CPU cost: any complex expressions in the indexquals will need to be
6559 : * evaluated once at the start of the scan to reduce them to runtime keys
6560 : * to pass to the index AM (see nodeIndexscan.c). We model the per-tuple
6561 : * CPU costs as cpu_index_tuple_cost plus one cpu_operator_cost per
6562 : * indexqual operator. Because we have numIndexTuples as a per-scan
6563 : * number, we have to multiply by num_sa_scans to get the correct result
6564 : * for ScalarArrayOpExpr cases. Similarly add in costs for any index
6565 : * ORDER BY expressions.
6566 : *
6567 : * Note: this neglects the possible costs of rechecking lossy operators.
6568 : * Detecting that that might be needed seems more expensive than it's
6569 : * worth, though, considering all the other inaccuracies here ...
6570 : */
6571 495642 : qual_arg_cost = index_other_operands_eval_cost(root, indexQuals) +
6572 495642 : index_other_operands_eval_cost(root, indexOrderBys);
6573 495642 : qual_op_cost = cpu_operator_cost *
6574 495642 : (list_length(indexQuals) + list_length(indexOrderBys));
6575 :
6576 495642 : indexStartupCost = qual_arg_cost;
6577 495642 : indexTotalCost += qual_arg_cost;
6578 495642 : indexTotalCost += numIndexTuples * num_sa_scans * (cpu_index_tuple_cost + qual_op_cost);
6579 :
6580 : /*
6581 : * Generic assumption about index correlation: there isn't any.
6582 : */
6583 495642 : indexCorrelation = 0.0;
6584 :
6585 : /*
6586 : * Return everything to caller.
6587 : */
6588 495642 : costs->indexStartupCost = indexStartupCost;
6589 495642 : costs->indexTotalCost = indexTotalCost;
6590 495642 : costs->indexSelectivity = indexSelectivity;
6591 495642 : costs->indexCorrelation = indexCorrelation;
6592 495642 : costs->numIndexPages = numIndexPages;
6593 495642 : costs->numIndexTuples = numIndexTuples;
6594 495642 : costs->spc_random_page_cost = spc_random_page_cost;
6595 495642 : costs->num_sa_scans = num_sa_scans;
6596 495642 : }
6597 :
6598 : /*
6599 : * If the index is partial, add its predicate to the given qual list.
6600 : *
6601 : * ANDing the index predicate with the explicitly given indexquals produces
6602 : * a more accurate idea of the index's selectivity. However, we need to be
6603 : * careful not to insert redundant clauses, because clauselist_selectivity()
6604 : * is easily fooled into computing a too-low selectivity estimate. Our
6605 : * approach is to add only the predicate clause(s) that cannot be proven to
6606 : * be implied by the given indexquals. This successfully handles cases such
6607 : * as a qual "x = 42" used with a partial index "WHERE x >= 40 AND x < 50".
6608 : * There are many other cases where we won't detect redundancy, leading to a
6609 : * too-low selectivity estimate, which will bias the system in favor of using
6610 : * partial indexes where possible. That is not necessarily bad though.
6611 : *
6612 : * Note that indexQuals contains RestrictInfo nodes while the indpred
6613 : * does not, so the output list will be mixed. This is OK for both
6614 : * predicate_implied_by() and clauselist_selectivity(), but might be
6615 : * problematic if the result were passed to other things.
6616 : */
6617 : List *
6618 818576 : add_predicate_to_index_quals(IndexOptInfo *index, List *indexQuals)
6619 : {
6620 818576 : List *predExtraQuals = NIL;
6621 : ListCell *lc;
6622 :
6623 818576 : if (index->indpred == NIL)
6624 816700 : return indexQuals;
6625 :
6626 3764 : foreach(lc, index->indpred)
6627 : {
6628 1888 : Node *predQual = (Node *) lfirst(lc);
6629 1888 : List *oneQual = list_make1(predQual);
6630 :
6631 1888 : if (!predicate_implied_by(oneQual, indexQuals, false))
6632 1692 : predExtraQuals = list_concat(predExtraQuals, oneQual);
6633 : }
6634 1876 : return list_concat(predExtraQuals, indexQuals);
6635 : }
6636 :
6637 :
6638 : void
6639 489468 : btcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
6640 : Cost *indexStartupCost, Cost *indexTotalCost,
6641 : Selectivity *indexSelectivity, double *indexCorrelation,
6642 : double *indexPages)
6643 : {
6644 489468 : IndexOptInfo *index = path->indexinfo;
6645 : GenericCosts costs;
6646 : Oid relid;
6647 : AttrNumber colnum;
6648 : VariableStatData vardata;
6649 : double numIndexTuples;
6650 : Cost descentCost;
6651 : List *indexBoundQuals;
6652 : int indexcol;
6653 : bool eqQualHere;
6654 : bool found_saop;
6655 : bool found_is_null_op;
6656 : double num_sa_scans;
6657 : ListCell *lc;
6658 :
6659 : /*
6660 : * For a btree scan, only leading '=' quals plus inequality quals for the
6661 : * immediately next attribute contribute to index selectivity (these are
6662 : * the "boundary quals" that determine the starting and stopping points of
6663 : * the index scan). Additional quals can suppress visits to the heap, so
6664 : * it's OK to count them in indexSelectivity, but they should not count
6665 : * for estimating numIndexTuples. So we must examine the given indexquals
6666 : * to find out which ones count as boundary quals. We rely on the
6667 : * knowledge that they are given in index column order.
6668 : *
6669 : * For a RowCompareExpr, we consider only the first column, just as
6670 : * rowcomparesel() does.
6671 : *
6672 : * If there's a ScalarArrayOpExpr in the quals, we'll actually perform N
6673 : * index scans not one, but the ScalarArrayOpExpr's operator can be
6674 : * considered to act the same as it normally does.
6675 : */
6676 489468 : indexBoundQuals = NIL;
6677 489468 : indexcol = 0;
6678 489468 : eqQualHere = false;
6679 489468 : found_saop = false;
6680 489468 : found_is_null_op = false;
6681 489468 : num_sa_scans = 1;
6682 840288 : foreach(lc, path->indexclauses)
6683 : {
6684 367804 : IndexClause *iclause = lfirst_node(IndexClause, lc);
6685 : ListCell *lc2;
6686 :
6687 367804 : if (indexcol != iclause->indexcol)
6688 : {
6689 : /* Beginning of a new column's quals */
6690 58856 : if (!eqQualHere)
6691 16020 : break; /* done if no '=' qual for indexcol */
6692 42836 : eqQualHere = false;
6693 42836 : indexcol++;
6694 42836 : if (indexcol != iclause->indexcol)
6695 964 : break; /* no quals at all for indexcol */
6696 : }
6697 :
6698 : /* Examine each indexqual associated with this index clause */
6699 704024 : foreach(lc2, iclause->indexquals)
6700 : {
6701 353204 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc2);
6702 353204 : Expr *clause = rinfo->clause;
6703 353204 : Oid clause_op = InvalidOid;
6704 : int op_strategy;
6705 :
6706 353204 : if (IsA(clause, OpExpr))
6707 : {
6708 344730 : OpExpr *op = (OpExpr *) clause;
6709 :
6710 344730 : clause_op = op->opno;
6711 : }
6712 8474 : else if (IsA(clause, RowCompareExpr))
6713 : {
6714 132 : RowCompareExpr *rc = (RowCompareExpr *) clause;
6715 :
6716 132 : clause_op = linitial_oid(rc->opnos);
6717 : }
6718 8342 : else if (IsA(clause, ScalarArrayOpExpr))
6719 : {
6720 6344 : ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
6721 6344 : Node *other_operand = (Node *) lsecond(saop->args);
6722 6344 : int alength = estimate_array_length(other_operand);
6723 :
6724 6344 : clause_op = saop->opno;
6725 6344 : found_saop = true;
6726 : /* count number of SA scans induced by indexBoundQuals only */
6727 6344 : if (alength > 1)
6728 6248 : num_sa_scans *= alength;
6729 : }
6730 1998 : else if (IsA(clause, NullTest))
6731 : {
6732 1998 : NullTest *nt = (NullTest *) clause;
6733 :
6734 1998 : if (nt->nulltesttype == IS_NULL)
6735 : {
6736 186 : found_is_null_op = true;
6737 : /* IS NULL is like = for selectivity purposes */
6738 186 : eqQualHere = true;
6739 : }
6740 : }
6741 : else
6742 0 : elog(ERROR, "unsupported indexqual type: %d",
6743 : (int) nodeTag(clause));
6744 :
6745 : /* check for equality operator */
6746 353204 : if (OidIsValid(clause_op))
6747 : {
6748 351206 : op_strategy = get_op_opfamily_strategy(clause_op,
6749 351206 : index->opfamily[indexcol]);
6750 : Assert(op_strategy != 0); /* not a member of opfamily?? */
6751 351206 : if (op_strategy == BTEqualStrategyNumber)
6752 332048 : eqQualHere = true;
6753 : }
6754 :
6755 353204 : indexBoundQuals = lappend(indexBoundQuals, rinfo);
6756 : }
6757 : }
6758 :
6759 : /*
6760 : * If index is unique and we found an '=' clause for each column, we can
6761 : * just assume numIndexTuples = 1 and skip the expensive
6762 : * clauselist_selectivity calculations. However, a ScalarArrayOp or
6763 : * NullTest invalidates that theory, even though it sets eqQualHere.
6764 : */
6765 489468 : if (index->unique &&
6766 404478 : indexcol == index->nkeycolumns - 1 &&
6767 171778 : eqQualHere &&
6768 171778 : !found_saop &&
6769 168442 : !found_is_null_op)
6770 168376 : numIndexTuples = 1.0;
6771 : else
6772 : {
6773 : List *selectivityQuals;
6774 : Selectivity btreeSelectivity;
6775 :
6776 : /*
6777 : * If the index is partial, AND the index predicate with the
6778 : * index-bound quals to produce a more accurate idea of the number of
6779 : * rows covered by the bound conditions.
6780 : */
6781 321092 : selectivityQuals = add_predicate_to_index_quals(index, indexBoundQuals);
6782 :
6783 321092 : btreeSelectivity = clauselist_selectivity(root, selectivityQuals,
6784 321092 : index->rel->relid,
6785 : JOIN_INNER,
6786 : NULL);
6787 321092 : numIndexTuples = btreeSelectivity * index->rel->tuples;
6788 :
6789 : /*
6790 : * As in genericcostestimate(), we have to adjust for any
6791 : * ScalarArrayOpExpr quals included in indexBoundQuals, and then round
6792 : * to integer.
6793 : */
6794 321092 : numIndexTuples = rint(numIndexTuples / num_sa_scans);
6795 : }
6796 :
6797 : /*
6798 : * Now do generic index cost estimation.
6799 : */
6800 4405212 : MemSet(&costs, 0, sizeof(costs));
6801 489468 : costs.numIndexTuples = numIndexTuples;
6802 :
6803 489468 : genericcostestimate(root, path, loop_count, &costs);
6804 :
6805 : /*
6806 : * Add a CPU-cost component to represent the costs of initial btree
6807 : * descent. We don't charge any I/O cost for touching upper btree levels,
6808 : * since they tend to stay in cache, but we still have to do about log2(N)
6809 : * comparisons to descend a btree of N leaf tuples. We charge one
6810 : * cpu_operator_cost per comparison.
6811 : *
6812 : * If there are ScalarArrayOpExprs, charge this once per SA scan. The
6813 : * ones after the first one are not startup cost so far as the overall
6814 : * plan is concerned, so add them only to "total" cost.
6815 : */
6816 489468 : if (index->tuples > 1) /* avoid computing log(0) */
6817 : {
6818 465188 : descentCost = ceil(log(index->tuples) / log(2.0)) * cpu_operator_cost;
6819 465188 : costs.indexStartupCost += descentCost;
6820 465188 : costs.indexTotalCost += costs.num_sa_scans * descentCost;
6821 : }
6822 :
6823 : /*
6824 : * Even though we're not charging I/O cost for touching upper btree pages,
6825 : * it's still reasonable to charge some CPU cost per page descended
6826 : * through. Moreover, if we had no such charge at all, bloated indexes
6827 : * would appear to have the same search cost as unbloated ones, at least
6828 : * in cases where only a single leaf page is expected to be visited. This
6829 : * cost is somewhat arbitrarily set at 50x cpu_operator_cost per page
6830 : * touched. The number of such pages is btree tree height plus one (ie,
6831 : * we charge for the leaf page too). As above, charge once per SA scan.
6832 : */
6833 489468 : descentCost = (index->tree_height + 1) * 50.0 * cpu_operator_cost;
6834 489468 : costs.indexStartupCost += descentCost;
6835 489468 : costs.indexTotalCost += costs.num_sa_scans * descentCost;
6836 :
6837 : /*
6838 : * If we can get an estimate of the first column's ordering correlation C
6839 : * from pg_statistic, estimate the index correlation as C for a
6840 : * single-column index, or C * 0.75 for multiple columns. (The idea here
6841 : * is that multiple columns dilute the importance of the first column's
6842 : * ordering, but don't negate it entirely. Before 8.0 we divided the
6843 : * correlation by the number of columns, but that seems too strong.)
6844 : */
6845 3426276 : MemSet(&vardata, 0, sizeof(vardata));
6846 :
6847 489468 : if (index->indexkeys[0] != 0)
6848 : {
6849 : /* Simple variable --- look to stats for the underlying table */
6850 487504 : RangeTblEntry *rte = planner_rt_fetch(index->rel->relid, root);
6851 :
6852 : Assert(rte->rtekind == RTE_RELATION);
6853 487504 : relid = rte->relid;
6854 : Assert(relid != InvalidOid);
6855 487504 : colnum = index->indexkeys[0];
6856 :
6857 487504 : if (get_relation_stats_hook &&
6858 0 : (*get_relation_stats_hook) (root, rte, colnum, &vardata))
6859 : {
6860 : /*
6861 : * The hook took control of acquiring a stats tuple. If it did
6862 : * supply a tuple, it'd better have supplied a freefunc.
6863 : */
6864 0 : if (HeapTupleIsValid(vardata.statsTuple) &&
6865 0 : !vardata.freefunc)
6866 0 : elog(ERROR, "no function provided to release variable stats with");
6867 : }
6868 : else
6869 : {
6870 487504 : vardata.statsTuple = SearchSysCache3(STATRELATTINH,
6871 : ObjectIdGetDatum(relid),
6872 : Int16GetDatum(colnum),
6873 487504 : BoolGetDatum(rte->inh));
6874 487504 : vardata.freefunc = ReleaseSysCache;
6875 : }
6876 : }
6877 : else
6878 : {
6879 : /* Expression --- maybe there are stats for the index itself */
6880 1964 : relid = index->indexoid;
6881 1964 : colnum = 1;
6882 :
6883 1964 : if (get_index_stats_hook &&
6884 0 : (*get_index_stats_hook) (root, relid, colnum, &vardata))
6885 : {
6886 : /*
6887 : * The hook took control of acquiring a stats tuple. If it did
6888 : * supply a tuple, it'd better have supplied a freefunc.
6889 : */
6890 0 : if (HeapTupleIsValid(vardata.statsTuple) &&
6891 0 : !vardata.freefunc)
6892 0 : elog(ERROR, "no function provided to release variable stats with");
6893 : }
6894 : else
6895 : {
6896 1964 : vardata.statsTuple = SearchSysCache3(STATRELATTINH,
6897 : ObjectIdGetDatum(relid),
6898 : Int16GetDatum(colnum),
6899 : BoolGetDatum(false));
6900 1964 : vardata.freefunc = ReleaseSysCache;
6901 : }
6902 : }
6903 :
6904 489468 : if (HeapTupleIsValid(vardata.statsTuple))
6905 : {
6906 : Oid sortop;
6907 : AttStatsSlot sslot;
6908 :
6909 352050 : sortop = get_opfamily_member(index->opfamily[0],
6910 352050 : index->opcintype[0],
6911 352050 : index->opcintype[0],
6912 : BTLessStrategyNumber);
6913 704100 : if (OidIsValid(sortop) &&
6914 352050 : get_attstatsslot(&sslot, vardata.statsTuple,
6915 : STATISTIC_KIND_CORRELATION, sortop,
6916 : ATTSTATSSLOT_NUMBERS))
6917 : {
6918 : double varCorrelation;
6919 :
6920 : Assert(sslot.nnumbers == 1);
6921 348004 : varCorrelation = sslot.numbers[0];
6922 :
6923 348004 : if (index->reverse_sort[0])
6924 0 : varCorrelation = -varCorrelation;
6925 :
6926 348004 : if (index->nkeycolumns > 1)
6927 102470 : costs.indexCorrelation = varCorrelation * 0.75;
6928 : else
6929 245534 : costs.indexCorrelation = varCorrelation;
6930 :
6931 348004 : free_attstatsslot(&sslot);
6932 : }
6933 : }
6934 :
6935 489468 : ReleaseVariableStats(vardata);
6936 :
6937 489468 : *indexStartupCost = costs.indexStartupCost;
6938 489468 : *indexTotalCost = costs.indexTotalCost;
6939 489468 : *indexSelectivity = costs.indexSelectivity;
6940 489468 : *indexCorrelation = costs.indexCorrelation;
6941 489468 : *indexPages = costs.numIndexPages;
6942 489468 : }
6943 :
6944 : void
6945 394 : hashcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
6946 : Cost *indexStartupCost, Cost *indexTotalCost,
6947 : Selectivity *indexSelectivity, double *indexCorrelation,
6948 : double *indexPages)
6949 : {
6950 : GenericCosts costs;
6951 :
6952 3546 : MemSet(&costs, 0, sizeof(costs));
6953 :
6954 394 : genericcostestimate(root, path, loop_count, &costs);
6955 :
6956 : /*
6957 : * A hash index has no descent costs as such, since the index AM can go
6958 : * directly to the target bucket after computing the hash value. There
6959 : * are a couple of other hash-specific costs that we could conceivably add
6960 : * here, though:
6961 : *
6962 : * Ideally we'd charge spc_random_page_cost for each page in the target
6963 : * bucket, not just the numIndexPages pages that genericcostestimate
6964 : * thought we'd visit. However in most cases we don't know which bucket
6965 : * that will be. There's no point in considering the average bucket size
6966 : * because the hash AM makes sure that's always one page.
6967 : *
6968 : * Likewise, we could consider charging some CPU for each index tuple in
6969 : * the bucket, if we knew how many there were. But the per-tuple cost is
6970 : * just a hash value comparison, not a general datatype-dependent
6971 : * comparison, so any such charge ought to be quite a bit less than
6972 : * cpu_operator_cost; which makes it probably not worth worrying about.
6973 : *
6974 : * A bigger issue is that chance hash-value collisions will result in
6975 : * wasted probes into the heap. We don't currently attempt to model this
6976 : * cost on the grounds that it's rare, but maybe it's not rare enough.
6977 : * (Any fix for this ought to consider the generic lossy-operator problem,
6978 : * though; it's not entirely hash-specific.)
6979 : */
6980 :
6981 394 : *indexStartupCost = costs.indexStartupCost;
6982 394 : *indexTotalCost = costs.indexTotalCost;
6983 394 : *indexSelectivity = costs.indexSelectivity;
6984 394 : *indexCorrelation = costs.indexCorrelation;
6985 394 : *indexPages = costs.numIndexPages;
6986 394 : }
6987 :
6988 : void
6989 3190 : gistcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
6990 : Cost *indexStartupCost, Cost *indexTotalCost,
6991 : Selectivity *indexSelectivity, double *indexCorrelation,
6992 : double *indexPages)
6993 : {
6994 3190 : IndexOptInfo *index = path->indexinfo;
6995 : GenericCosts costs;
6996 : Cost descentCost;
6997 :
6998 28710 : MemSet(&costs, 0, sizeof(costs));
6999 :
7000 3190 : genericcostestimate(root, path, loop_count, &costs);
7001 :
7002 : /*
7003 : * We model index descent costs similarly to those for btree, but to do
7004 : * that we first need an idea of the tree height. We somewhat arbitrarily
7005 : * assume that the fanout is 100, meaning the tree height is at most
7006 : * log100(index->pages).
7007 : *
7008 : * Although this computation isn't really expensive enough to require
7009 : * caching, we might as well use index->tree_height to cache it.
7010 : */
7011 3190 : if (index->tree_height < 0) /* unknown? */
7012 : {
7013 3178 : if (index->pages > 1) /* avoid computing log(0) */
7014 2682 : index->tree_height = (int) (log(index->pages) / log(100.0));
7015 : else
7016 496 : index->tree_height = 0;
7017 : }
7018 :
7019 : /*
7020 : * Add a CPU-cost component to represent the costs of initial descent. We
7021 : * just use log(N) here not log2(N) since the branching factor isn't
7022 : * necessarily two anyway. As for btree, charge once per SA scan.
7023 : */
7024 3190 : if (index->tuples > 1) /* avoid computing log(0) */
7025 : {
7026 3190 : descentCost = ceil(log(index->tuples)) * cpu_operator_cost;
7027 3190 : costs.indexStartupCost += descentCost;
7028 3190 : costs.indexTotalCost += costs.num_sa_scans * descentCost;
7029 : }
7030 :
7031 : /*
7032 : * Likewise add a per-page charge, calculated the same as for btrees.
7033 : */
7034 3190 : descentCost = (index->tree_height + 1) * 50.0 * cpu_operator_cost;
7035 3190 : costs.indexStartupCost += descentCost;
7036 3190 : costs.indexTotalCost += costs.num_sa_scans * descentCost;
7037 :
7038 3190 : *indexStartupCost = costs.indexStartupCost;
7039 3190 : *indexTotalCost = costs.indexTotalCost;
7040 3190 : *indexSelectivity = costs.indexSelectivity;
7041 3190 : *indexCorrelation = costs.indexCorrelation;
7042 3190 : *indexPages = costs.numIndexPages;
7043 3190 : }
7044 :
7045 : void
7046 1778 : spgcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
7047 : Cost *indexStartupCost, Cost *indexTotalCost,
7048 : Selectivity *indexSelectivity, double *indexCorrelation,
7049 : double *indexPages)
7050 : {
7051 1778 : IndexOptInfo *index = path->indexinfo;
7052 : GenericCosts costs;
7053 : Cost descentCost;
7054 :
7055 16002 : MemSet(&costs, 0, sizeof(costs));
7056 :
7057 1778 : genericcostestimate(root, path, loop_count, &costs);
7058 :
7059 : /*
7060 : * We model index descent costs similarly to those for btree, but to do
7061 : * that we first need an idea of the tree height. We somewhat arbitrarily
7062 : * assume that the fanout is 100, meaning the tree height is at most
7063 : * log100(index->pages).
7064 : *
7065 : * Although this computation isn't really expensive enough to require
7066 : * caching, we might as well use index->tree_height to cache it.
7067 : */
7068 1778 : if (index->tree_height < 0) /* unknown? */
7069 : {
7070 1772 : if (index->pages > 1) /* avoid computing log(0) */
7071 1772 : index->tree_height = (int) (log(index->pages) / log(100.0));
7072 : else
7073 0 : index->tree_height = 0;
7074 : }
7075 :
7076 : /*
7077 : * Add a CPU-cost component to represent the costs of initial descent. We
7078 : * just use log(N) here not log2(N) since the branching factor isn't
7079 : * necessarily two anyway. As for btree, charge once per SA scan.
7080 : */
7081 1778 : if (index->tuples > 1) /* avoid computing log(0) */
7082 : {
7083 1778 : descentCost = ceil(log(index->tuples)) * cpu_operator_cost;
7084 1778 : costs.indexStartupCost += descentCost;
7085 1778 : costs.indexTotalCost += costs.num_sa_scans * descentCost;
7086 : }
7087 :
7088 : /*
7089 : * Likewise add a per-page charge, calculated the same as for btrees.
7090 : */
7091 1778 : descentCost = (index->tree_height + 1) * 50.0 * cpu_operator_cost;
7092 1778 : costs.indexStartupCost += descentCost;
7093 1778 : costs.indexTotalCost += costs.num_sa_scans * descentCost;
7094 :
7095 1778 : *indexStartupCost = costs.indexStartupCost;
7096 1778 : *indexTotalCost = costs.indexTotalCost;
7097 1778 : *indexSelectivity = costs.indexSelectivity;
7098 1778 : *indexCorrelation = costs.indexCorrelation;
7099 1778 : *indexPages = costs.numIndexPages;
7100 1778 : }
7101 :
7102 :
7103 : /*
7104 : * Support routines for gincostestimate
7105 : */
7106 :
7107 : typedef struct
7108 : {
7109 : bool attHasFullScan[INDEX_MAX_KEYS];
7110 : bool attHasNormalScan[INDEX_MAX_KEYS];
7111 : double partialEntries;
7112 : double exactEntries;
7113 : double searchEntries;
7114 : double arrayScans;
7115 : } GinQualCounts;
7116 :
7117 : /*
7118 : * Estimate the number of index terms that need to be searched for while
7119 : * testing the given GIN query, and increment the counts in *counts
7120 : * appropriately. If the query is unsatisfiable, return false.
7121 : */
7122 : static bool
7123 2046 : gincost_pattern(IndexOptInfo *index, int indexcol,
7124 : Oid clause_op, Datum query,
7125 : GinQualCounts *counts)
7126 : {
7127 : FmgrInfo flinfo;
7128 : Oid extractProcOid;
7129 : Oid collation;
7130 : int strategy_op;
7131 : Oid lefttype,
7132 : righttype;
7133 2046 : int32 nentries = 0;
7134 2046 : bool *partial_matches = NULL;
7135 2046 : Pointer *extra_data = NULL;
7136 2046 : bool *nullFlags = NULL;
7137 2046 : int32 searchMode = GIN_SEARCH_MODE_DEFAULT;
7138 : int32 i;
7139 :
7140 : Assert(indexcol < index->nkeycolumns);
7141 :
7142 : /*
7143 : * Get the operator's strategy number and declared input data types within
7144 : * the index opfamily. (We don't need the latter, but we use
7145 : * get_op_opfamily_properties because it will throw error if it fails to
7146 : * find a matching pg_amop entry.)
7147 : */
7148 2046 : get_op_opfamily_properties(clause_op, index->opfamily[indexcol], false,
7149 : &strategy_op, &lefttype, &righttype);
7150 :
7151 : /*
7152 : * GIN always uses the "default" support functions, which are those with
7153 : * lefttype == righttype == the opclass' opcintype (see
7154 : * IndexSupportInitialize in relcache.c).
7155 : */
7156 2046 : extractProcOid = get_opfamily_proc(index->opfamily[indexcol],
7157 2046 : index->opcintype[indexcol],
7158 2046 : index->opcintype[indexcol],
7159 : GIN_EXTRACTQUERY_PROC);
7160 :
7161 2046 : if (!OidIsValid(extractProcOid))
7162 : {
7163 : /* should not happen; throw same error as index_getprocinfo */
7164 0 : elog(ERROR, "missing support function %d for attribute %d of index \"%s\"",
7165 : GIN_EXTRACTQUERY_PROC, indexcol + 1,
7166 : get_rel_name(index->indexoid));
7167 : }
7168 :
7169 : /*
7170 : * Choose collation to pass to extractProc (should match initGinState).
7171 : */
7172 2046 : if (OidIsValid(index->indexcollations[indexcol]))
7173 362 : collation = index->indexcollations[indexcol];
7174 : else
7175 1684 : collation = DEFAULT_COLLATION_OID;
7176 :
7177 2046 : fmgr_info(extractProcOid, &flinfo);
7178 :
7179 2046 : set_fn_opclass_options(&flinfo, index->opclassoptions[indexcol]);
7180 :
7181 2046 : FunctionCall7Coll(&flinfo,
7182 : collation,
7183 : query,
7184 : PointerGetDatum(&nentries),
7185 : UInt16GetDatum(strategy_op),
7186 : PointerGetDatum(&partial_matches),
7187 : PointerGetDatum(&extra_data),
7188 : PointerGetDatum(&nullFlags),
7189 : PointerGetDatum(&searchMode));
7190 :
7191 2046 : if (nentries <= 0 && searchMode == GIN_SEARCH_MODE_DEFAULT)
7192 : {
7193 : /* No match is possible */
7194 12 : return false;
7195 : }
7196 :
7197 5756 : for (i = 0; i < nentries; i++)
7198 : {
7199 : /*
7200 : * For partial match we haven't any information to estimate number of
7201 : * matched entries in index, so, we just estimate it as 100
7202 : */
7203 3722 : if (partial_matches && partial_matches[i])
7204 316 : counts->partialEntries += 100;
7205 : else
7206 3406 : counts->exactEntries++;
7207 :
7208 3722 : counts->searchEntries++;
7209 : }
7210 :
7211 2034 : if (searchMode == GIN_SEARCH_MODE_DEFAULT)
7212 : {
7213 1564 : counts->attHasNormalScan[indexcol] = true;
7214 : }
7215 470 : else if (searchMode == GIN_SEARCH_MODE_INCLUDE_EMPTY)
7216 : {
7217 : /* Treat "include empty" like an exact-match item */
7218 44 : counts->attHasNormalScan[indexcol] = true;
7219 44 : counts->exactEntries++;
7220 44 : counts->searchEntries++;
7221 : }
7222 : else
7223 : {
7224 : /* It's GIN_SEARCH_MODE_ALL */
7225 426 : counts->attHasFullScan[indexcol] = true;
7226 : }
7227 :
7228 2034 : return true;
7229 : }
7230 :
7231 : /*
7232 : * Estimate the number of index terms that need to be searched for while
7233 : * testing the given GIN index clause, and increment the counts in *counts
7234 : * appropriately. If the query is unsatisfiable, return false.
7235 : */
7236 : static bool
7237 2034 : gincost_opexpr(PlannerInfo *root,
7238 : IndexOptInfo *index,
7239 : int indexcol,
7240 : OpExpr *clause,
7241 : GinQualCounts *counts)
7242 : {
7243 2034 : Oid clause_op = clause->opno;
7244 2034 : Node *operand = (Node *) lsecond(clause->args);
7245 :
7246 : /* aggressively reduce to a constant, and look through relabeling */
7247 2034 : operand = estimate_expression_value(root, operand);
7248 :
7249 2034 : if (IsA(operand, RelabelType))
7250 0 : operand = (Node *) ((RelabelType *) operand)->arg;
7251 :
7252 : /*
7253 : * It's impossible to call extractQuery method for unknown operand. So
7254 : * unless operand is a Const we can't do much; just assume there will be
7255 : * one ordinary search entry from the operand at runtime.
7256 : */
7257 2034 : if (!IsA(operand, Const))
7258 : {
7259 0 : counts->exactEntries++;
7260 0 : counts->searchEntries++;
7261 0 : return true;
7262 : }
7263 :
7264 : /* If Const is null, there can be no matches */
7265 2034 : if (((Const *) operand)->constisnull)
7266 0 : return false;
7267 :
7268 : /* Otherwise, apply extractQuery and get the actual term counts */
7269 2034 : return gincost_pattern(index, indexcol, clause_op,
7270 : ((Const *) operand)->constvalue,
7271 : counts);
7272 : }
7273 :
7274 : /*
7275 : * Estimate the number of index terms that need to be searched for while
7276 : * testing the given GIN index clause, and increment the counts in *counts
7277 : * appropriately. If the query is unsatisfiable, return false.
7278 : *
7279 : * A ScalarArrayOpExpr will give rise to N separate indexscans at runtime,
7280 : * each of which involves one value from the RHS array, plus all the
7281 : * non-array quals (if any). To model this, we average the counts across
7282 : * the RHS elements, and add the averages to the counts in *counts (which
7283 : * correspond to per-indexscan costs). We also multiply counts->arrayScans
7284 : * by N, causing gincostestimate to scale up its estimates accordingly.
7285 : */
7286 : static bool
7287 6 : gincost_scalararrayopexpr(PlannerInfo *root,
7288 : IndexOptInfo *index,
7289 : int indexcol,
7290 : ScalarArrayOpExpr *clause,
7291 : double numIndexEntries,
7292 : GinQualCounts *counts)
7293 : {
7294 6 : Oid clause_op = clause->opno;
7295 6 : Node *rightop = (Node *) lsecond(clause->args);
7296 : ArrayType *arrayval;
7297 : int16 elmlen;
7298 : bool elmbyval;
7299 : char elmalign;
7300 : int numElems;
7301 : Datum *elemValues;
7302 : bool *elemNulls;
7303 : GinQualCounts arraycounts;
7304 6 : int numPossible = 0;
7305 : int i;
7306 :
7307 : Assert(clause->useOr);
7308 :
7309 : /* aggressively reduce to a constant, and look through relabeling */
7310 6 : rightop = estimate_expression_value(root, rightop);
7311 :
7312 6 : if (IsA(rightop, RelabelType))
7313 0 : rightop = (Node *) ((RelabelType *) rightop)->arg;
7314 :
7315 : /*
7316 : * It's impossible to call extractQuery method for unknown operand. So
7317 : * unless operand is a Const we can't do much; just assume there will be
7318 : * one ordinary search entry from each array entry at runtime, and fall
7319 : * back on a probably-bad estimate of the number of array entries.
7320 : */
7321 6 : if (!IsA(rightop, Const))
7322 : {
7323 0 : counts->exactEntries++;
7324 0 : counts->searchEntries++;
7325 0 : counts->arrayScans *= estimate_array_length(rightop);
7326 0 : return true;
7327 : }
7328 :
7329 : /* If Const is null, there can be no matches */
7330 6 : if (((Const *) rightop)->constisnull)
7331 0 : return false;
7332 :
7333 : /* Otherwise, extract the array elements and iterate over them */
7334 6 : arrayval = DatumGetArrayTypeP(((Const *) rightop)->constvalue);
7335 6 : get_typlenbyvalalign(ARR_ELEMTYPE(arrayval),
7336 : &elmlen, &elmbyval, &elmalign);
7337 6 : deconstruct_array(arrayval,
7338 : ARR_ELEMTYPE(arrayval),
7339 : elmlen, elmbyval, elmalign,
7340 : &elemValues, &elemNulls, &numElems);
7341 :
7342 6 : memset(&arraycounts, 0, sizeof(arraycounts));
7343 :
7344 18 : for (i = 0; i < numElems; i++)
7345 : {
7346 : GinQualCounts elemcounts;
7347 :
7348 : /* NULL can't match anything, so ignore, as the executor will */
7349 12 : if (elemNulls[i])
7350 0 : continue;
7351 :
7352 : /* Otherwise, apply extractQuery and get the actual term counts */
7353 12 : memset(&elemcounts, 0, sizeof(elemcounts));
7354 :
7355 12 : if (gincost_pattern(index, indexcol, clause_op, elemValues[i],
7356 : &elemcounts))
7357 : {
7358 : /* We ignore array elements that are unsatisfiable patterns */
7359 12 : numPossible++;
7360 :
7361 12 : if (elemcounts.attHasFullScan[indexcol] &&
7362 0 : !elemcounts.attHasNormalScan[indexcol])
7363 : {
7364 : /*
7365 : * Full index scan will be required. We treat this as if
7366 : * every key in the index had been listed in the query; is
7367 : * that reasonable?
7368 : */
7369 0 : elemcounts.partialEntries = 0;
7370 0 : elemcounts.exactEntries = numIndexEntries;
7371 0 : elemcounts.searchEntries = numIndexEntries;
7372 : }
7373 12 : arraycounts.partialEntries += elemcounts.partialEntries;
7374 12 : arraycounts.exactEntries += elemcounts.exactEntries;
7375 12 : arraycounts.searchEntries += elemcounts.searchEntries;
7376 : }
7377 : }
7378 :
7379 6 : if (numPossible == 0)
7380 : {
7381 : /* No satisfiable patterns in the array */
7382 0 : return false;
7383 : }
7384 :
7385 : /*
7386 : * Now add the averages to the global counts. This will give us an
7387 : * estimate of the average number of terms searched for in each indexscan,
7388 : * including contributions from both array and non-array quals.
7389 : */
7390 6 : counts->partialEntries += arraycounts.partialEntries / numPossible;
7391 6 : counts->exactEntries += arraycounts.exactEntries / numPossible;
7392 6 : counts->searchEntries += arraycounts.searchEntries / numPossible;
7393 :
7394 6 : counts->arrayScans *= numPossible;
7395 :
7396 6 : return true;
7397 : }
7398 :
7399 : /*
7400 : * GIN has search behavior completely different from other index types
7401 : */
7402 : void
7403 1842 : gincostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
7404 : Cost *indexStartupCost, Cost *indexTotalCost,
7405 : Selectivity *indexSelectivity, double *indexCorrelation,
7406 : double *indexPages)
7407 : {
7408 1842 : IndexOptInfo *index = path->indexinfo;
7409 1842 : List *indexQuals = get_quals_from_indexclauses(path->indexclauses);
7410 : List *selectivityQuals;
7411 1842 : double numPages = index->pages,
7412 1842 : numTuples = index->tuples;
7413 : double numEntryPages,
7414 : numDataPages,
7415 : numPendingPages,
7416 : numEntries;
7417 : GinQualCounts counts;
7418 : bool matchPossible;
7419 : bool fullIndexScan;
7420 : double partialScale;
7421 : double entryPagesFetched,
7422 : dataPagesFetched,
7423 : dataPagesFetchedBySel;
7424 : double qual_op_cost,
7425 : qual_arg_cost,
7426 : spc_random_page_cost,
7427 : outer_scans;
7428 : Relation indexRel;
7429 : GinStatsData ginStats;
7430 : ListCell *lc;
7431 : int i;
7432 :
7433 : /*
7434 : * Obtain statistical information from the meta page, if possible. Else
7435 : * set ginStats to zeroes, and we'll cope below.
7436 : */
7437 1842 : if (!index->hypothetical)
7438 : {
7439 : /* Lock should have already been obtained in plancat.c */
7440 1842 : indexRel = index_open(index->indexoid, NoLock);
7441 1842 : ginGetStats(indexRel, &ginStats);
7442 1842 : index_close(indexRel, NoLock);
7443 : }
7444 : else
7445 : {
7446 0 : memset(&ginStats, 0, sizeof(ginStats));
7447 : }
7448 :
7449 : /*
7450 : * Assuming we got valid (nonzero) stats at all, nPendingPages can be
7451 : * trusted, but the other fields are data as of the last VACUUM. We can
7452 : * scale them up to account for growth since then, but that method only
7453 : * goes so far; in the worst case, the stats might be for a completely
7454 : * empty index, and scaling them will produce pretty bogus numbers.
7455 : * Somewhat arbitrarily, set the cutoff for doing scaling at 4X growth; if
7456 : * it's grown more than that, fall back to estimating things only from the
7457 : * assumed-accurate index size. But we'll trust nPendingPages in any case
7458 : * so long as it's not clearly insane, ie, more than the index size.
7459 : */
7460 1842 : if (ginStats.nPendingPages < numPages)
7461 1842 : numPendingPages = ginStats.nPendingPages;
7462 : else
7463 0 : numPendingPages = 0;
7464 :
7465 1842 : if (numPages > 0 && ginStats.nTotalPages <= numPages &&
7466 1842 : ginStats.nTotalPages > numPages / 4 &&
7467 1800 : ginStats.nEntryPages > 0 && ginStats.nEntries > 0)
7468 1548 : {
7469 : /*
7470 : * OK, the stats seem close enough to sane to be trusted. But we
7471 : * still need to scale them by the ratio numPages / nTotalPages to
7472 : * account for growth since the last VACUUM.
7473 : */
7474 1548 : double scale = numPages / ginStats.nTotalPages;
7475 :
7476 1548 : numEntryPages = ceil(ginStats.nEntryPages * scale);
7477 1548 : numDataPages = ceil(ginStats.nDataPages * scale);
7478 1548 : numEntries = ceil(ginStats.nEntries * scale);
7479 : /* ensure we didn't round up too much */
7480 1548 : numEntryPages = Min(numEntryPages, numPages - numPendingPages);
7481 1548 : numDataPages = Min(numDataPages,
7482 : numPages - numPendingPages - numEntryPages);
7483 : }
7484 : else
7485 : {
7486 : /*
7487 : * We might get here because it's a hypothetical index, or an index
7488 : * created pre-9.1 and never vacuumed since upgrading (in which case
7489 : * its stats would read as zeroes), or just because it's grown too
7490 : * much since the last VACUUM for us to put our faith in scaling.
7491 : *
7492 : * Invent some plausible internal statistics based on the index page
7493 : * count (and clamp that to at least 10 pages, just in case). We
7494 : * estimate that 90% of the index is entry pages, and the rest is data
7495 : * pages. Estimate 100 entries per entry page; this is rather bogus
7496 : * since it'll depend on the size of the keys, but it's more robust
7497 : * than trying to predict the number of entries per heap tuple.
7498 : */
7499 294 : numPages = Max(numPages, 10);
7500 294 : numEntryPages = floor((numPages - numPendingPages) * 0.90);
7501 294 : numDataPages = numPages - numPendingPages - numEntryPages;
7502 294 : numEntries = floor(numEntryPages * 100);
7503 : }
7504 :
7505 : /* In an empty index, numEntries could be zero. Avoid divide-by-zero */
7506 1842 : if (numEntries < 1)
7507 0 : numEntries = 1;
7508 :
7509 : /*
7510 : * If the index is partial, AND the index predicate with the index-bound
7511 : * quals to produce a more accurate idea of the number of rows covered by
7512 : * the bound conditions.
7513 : */
7514 1842 : selectivityQuals = add_predicate_to_index_quals(index, indexQuals);
7515 :
7516 : /* Estimate the fraction of main-table tuples that will be visited */
7517 3684 : *indexSelectivity = clauselist_selectivity(root, selectivityQuals,
7518 1842 : index->rel->relid,
7519 : JOIN_INNER,
7520 : NULL);
7521 :
7522 : /* fetch estimated page cost for tablespace containing index */
7523 1842 : get_tablespace_page_costs(index->reltablespace,
7524 : &spc_random_page_cost,
7525 : NULL);
7526 :
7527 : /*
7528 : * Generic assumption about index correlation: there isn't any.
7529 : */
7530 1842 : *indexCorrelation = 0.0;
7531 :
7532 : /*
7533 : * Examine quals to estimate number of search entries & partial matches
7534 : */
7535 1842 : memset(&counts, 0, sizeof(counts));
7536 1842 : counts.arrayScans = 1;
7537 1842 : matchPossible = true;
7538 :
7539 3882 : foreach(lc, path->indexclauses)
7540 : {
7541 2040 : IndexClause *iclause = lfirst_node(IndexClause, lc);
7542 : ListCell *lc2;
7543 :
7544 4068 : foreach(lc2, iclause->indexquals)
7545 : {
7546 2040 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc2);
7547 2040 : Expr *clause = rinfo->clause;
7548 :
7549 2040 : if (IsA(clause, OpExpr))
7550 : {
7551 2034 : matchPossible = gincost_opexpr(root,
7552 : index,
7553 2034 : iclause->indexcol,
7554 : (OpExpr *) clause,
7555 : &counts);
7556 2034 : if (!matchPossible)
7557 12 : break;
7558 : }
7559 6 : else if (IsA(clause, ScalarArrayOpExpr))
7560 : {
7561 6 : matchPossible = gincost_scalararrayopexpr(root,
7562 : index,
7563 6 : iclause->indexcol,
7564 : (ScalarArrayOpExpr *) clause,
7565 : numEntries,
7566 : &counts);
7567 6 : if (!matchPossible)
7568 0 : break;
7569 : }
7570 : else
7571 : {
7572 : /* shouldn't be anything else for a GIN index */
7573 0 : elog(ERROR, "unsupported GIN indexqual type: %d",
7574 : (int) nodeTag(clause));
7575 : }
7576 : }
7577 : }
7578 :
7579 : /* Fall out if there were any provably-unsatisfiable quals */
7580 1842 : if (!matchPossible)
7581 : {
7582 12 : *indexStartupCost = 0;
7583 12 : *indexTotalCost = 0;
7584 12 : *indexSelectivity = 0;
7585 12 : return;
7586 : }
7587 :
7588 : /*
7589 : * If attribute has a full scan and at the same time doesn't have normal
7590 : * scan, then we'll have to scan all non-null entries of that attribute.
7591 : * Currently, we don't have per-attribute statistics for GIN. Thus, we
7592 : * must assume the whole GIN index has to be scanned in this case.
7593 : */
7594 1830 : fullIndexScan = false;
7595 3552 : for (i = 0; i < index->nkeycolumns; i++)
7596 : {
7597 2058 : if (counts.attHasFullScan[i] && !counts.attHasNormalScan[i])
7598 : {
7599 336 : fullIndexScan = true;
7600 336 : break;
7601 : }
7602 : }
7603 :
7604 1830 : if (fullIndexScan || indexQuals == NIL)
7605 : {
7606 : /*
7607 : * Full index scan will be required. We treat this as if every key in
7608 : * the index had been listed in the query; is that reasonable?
7609 : */
7610 336 : counts.partialEntries = 0;
7611 336 : counts.exactEntries = numEntries;
7612 336 : counts.searchEntries = numEntries;
7613 : }
7614 :
7615 : /* Will we have more than one iteration of a nestloop scan? */
7616 1830 : outer_scans = loop_count;
7617 :
7618 : /*
7619 : * Compute cost to begin scan, first of all, pay attention to pending
7620 : * list.
7621 : */
7622 1830 : entryPagesFetched = numPendingPages;
7623 :
7624 : /*
7625 : * Estimate number of entry pages read. We need to do
7626 : * counts.searchEntries searches. Use a power function as it should be,
7627 : * but tuples on leaf pages usually is much greater. Here we include all
7628 : * searches in entry tree, including search of first entry in partial
7629 : * match algorithm
7630 : */
7631 1830 : entryPagesFetched += ceil(counts.searchEntries * rint(pow(numEntryPages, 0.15)));
7632 :
7633 : /*
7634 : * Add an estimate of entry pages read by partial match algorithm. It's a
7635 : * scan over leaf pages in entry tree. We haven't any useful stats here,
7636 : * so estimate it as proportion. Because counts.partialEntries is really
7637 : * pretty bogus (see code above), it's possible that it is more than
7638 : * numEntries; clamp the proportion to ensure sanity.
7639 : */
7640 1830 : partialScale = counts.partialEntries / numEntries;
7641 1830 : partialScale = Min(partialScale, 1.0);
7642 :
7643 1830 : entryPagesFetched += ceil(numEntryPages * partialScale);
7644 :
7645 : /*
7646 : * Partial match algorithm reads all data pages before doing actual scan,
7647 : * so it's a startup cost. Again, we haven't any useful stats here, so
7648 : * estimate it as proportion.
7649 : */
7650 1830 : dataPagesFetched = ceil(numDataPages * partialScale);
7651 :
7652 : /*
7653 : * Calculate cache effects if more than one scan due to nestloops or array
7654 : * quals. The result is pro-rated per nestloop scan, but the array qual
7655 : * factor shouldn't be pro-rated (compare genericcostestimate).
7656 : */
7657 1830 : if (outer_scans > 1 || counts.arrayScans > 1)
7658 : {
7659 6 : entryPagesFetched *= outer_scans * counts.arrayScans;
7660 6 : entryPagesFetched = index_pages_fetched(entryPagesFetched,
7661 : (BlockNumber) numEntryPages,
7662 : numEntryPages, root);
7663 6 : entryPagesFetched /= outer_scans;
7664 6 : dataPagesFetched *= outer_scans * counts.arrayScans;
7665 6 : dataPagesFetched = index_pages_fetched(dataPagesFetched,
7666 : (BlockNumber) numDataPages,
7667 : numDataPages, root);
7668 6 : dataPagesFetched /= outer_scans;
7669 : }
7670 :
7671 : /*
7672 : * Here we use random page cost because logically-close pages could be far
7673 : * apart on disk.
7674 : */
7675 1830 : *indexStartupCost = (entryPagesFetched + dataPagesFetched) * spc_random_page_cost;
7676 :
7677 : /*
7678 : * Now compute the number of data pages fetched during the scan.
7679 : *
7680 : * We assume every entry to have the same number of items, and that there
7681 : * is no overlap between them. (XXX: tsvector and array opclasses collect
7682 : * statistics on the frequency of individual keys; it would be nice to use
7683 : * those here.)
7684 : */
7685 1830 : dataPagesFetched = ceil(numDataPages * counts.exactEntries / numEntries);
7686 :
7687 : /*
7688 : * If there is a lot of overlap among the entries, in particular if one of
7689 : * the entries is very frequent, the above calculation can grossly
7690 : * under-estimate. As a simple cross-check, calculate a lower bound based
7691 : * on the overall selectivity of the quals. At a minimum, we must read
7692 : * one item pointer for each matching entry.
7693 : *
7694 : * The width of each item pointer varies, based on the level of
7695 : * compression. We don't have statistics on that, but an average of
7696 : * around 3 bytes per item is fairly typical.
7697 : */
7698 1830 : dataPagesFetchedBySel = ceil(*indexSelectivity *
7699 1830 : (numTuples / (BLCKSZ / 3)));
7700 1830 : if (dataPagesFetchedBySel > dataPagesFetched)
7701 1462 : dataPagesFetched = dataPagesFetchedBySel;
7702 :
7703 : /* Account for cache effects, the same as above */
7704 1830 : if (outer_scans > 1 || counts.arrayScans > 1)
7705 : {
7706 6 : dataPagesFetched *= outer_scans * counts.arrayScans;
7707 6 : dataPagesFetched = index_pages_fetched(dataPagesFetched,
7708 : (BlockNumber) numDataPages,
7709 : numDataPages, root);
7710 6 : dataPagesFetched /= outer_scans;
7711 : }
7712 :
7713 : /* And apply random_page_cost as the cost per page */
7714 1830 : *indexTotalCost = *indexStartupCost +
7715 1830 : dataPagesFetched * spc_random_page_cost;
7716 :
7717 : /*
7718 : * Add on index qual eval costs, much as in genericcostestimate. But we
7719 : * can disregard indexorderbys, since GIN doesn't support those.
7720 : */
7721 1830 : qual_arg_cost = index_other_operands_eval_cost(root, indexQuals);
7722 1830 : qual_op_cost = cpu_operator_cost * list_length(indexQuals);
7723 :
7724 1830 : *indexStartupCost += qual_arg_cost;
7725 1830 : *indexTotalCost += qual_arg_cost;
7726 1830 : *indexTotalCost += (numTuples * *indexSelectivity) * (cpu_index_tuple_cost + qual_op_cost);
7727 1830 : *indexPages = dataPagesFetched;
7728 : }
7729 :
7730 : /*
7731 : * BRIN has search behavior completely different from other index types
7732 : */
7733 : void
7734 10344 : brincostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
7735 : Cost *indexStartupCost, Cost *indexTotalCost,
7736 : Selectivity *indexSelectivity, double *indexCorrelation,
7737 : double *indexPages)
7738 : {
7739 10344 : IndexOptInfo *index = path->indexinfo;
7740 10344 : List *indexQuals = get_quals_from_indexclauses(path->indexclauses);
7741 10344 : double numPages = index->pages;
7742 10344 : RelOptInfo *baserel = index->rel;
7743 10344 : RangeTblEntry *rte = planner_rt_fetch(baserel->relid, root);
7744 : Cost spc_seq_page_cost;
7745 : Cost spc_random_page_cost;
7746 : double qual_arg_cost;
7747 : double qualSelectivity;
7748 : BrinStatsData statsData;
7749 : double indexRanges;
7750 : double minimalRanges;
7751 : double estimatedRanges;
7752 : double selec;
7753 : Relation indexRel;
7754 : ListCell *l;
7755 : VariableStatData vardata;
7756 :
7757 : Assert(rte->rtekind == RTE_RELATION);
7758 :
7759 : /* fetch estimated page cost for the tablespace containing the index */
7760 10344 : get_tablespace_page_costs(index->reltablespace,
7761 : &spc_random_page_cost,
7762 : &spc_seq_page_cost);
7763 :
7764 : /*
7765 : * Obtain some data from the index itself, if possible. Otherwise invent
7766 : * some plausible internal statistics based on the relation page count.
7767 : */
7768 10344 : if (!index->hypothetical)
7769 : {
7770 : /*
7771 : * A lock should have already been obtained on the index in plancat.c.
7772 : */
7773 10344 : indexRel = index_open(index->indexoid, NoLock);
7774 10344 : brinGetStats(indexRel, &statsData);
7775 10344 : index_close(indexRel, NoLock);
7776 :
7777 : /* work out the actual number of ranges in the index */
7778 10344 : indexRanges = Max(ceil((double) baserel->pages /
7779 : statsData.pagesPerRange), 1.0);
7780 : }
7781 : else
7782 : {
7783 : /*
7784 : * Assume default number of pages per range, and estimate the number
7785 : * of ranges based on that.
7786 : */
7787 0 : indexRanges = Max(ceil((double) baserel->pages /
7788 : BRIN_DEFAULT_PAGES_PER_RANGE), 1.0);
7789 :
7790 0 : statsData.pagesPerRange = BRIN_DEFAULT_PAGES_PER_RANGE;
7791 0 : statsData.revmapNumPages = (indexRanges / REVMAP_PAGE_MAXITEMS) + 1;
7792 : }
7793 :
7794 : /*
7795 : * Compute index correlation
7796 : *
7797 : * Because we can use all index quals equally when scanning, we can use
7798 : * the largest correlation (in absolute value) among columns used by the
7799 : * query. Start at zero, the worst possible case. If we cannot find any
7800 : * correlation statistics, we will keep it as 0.
7801 : */
7802 10344 : *indexCorrelation = 0;
7803 :
7804 20688 : foreach(l, path->indexclauses)
7805 : {
7806 10344 : IndexClause *iclause = lfirst_node(IndexClause, l);
7807 10344 : AttrNumber attnum = index->indexkeys[iclause->indexcol];
7808 :
7809 : /* attempt to lookup stats in relation for this index column */
7810 10344 : if (attnum != 0)
7811 : {
7812 : /* Simple variable -- look to stats for the underlying table */
7813 10344 : if (get_relation_stats_hook &&
7814 0 : (*get_relation_stats_hook) (root, rte, attnum, &vardata))
7815 : {
7816 : /*
7817 : * The hook took control of acquiring a stats tuple. If it
7818 : * did supply a tuple, it'd better have supplied a freefunc.
7819 : */
7820 0 : if (HeapTupleIsValid(vardata.statsTuple) && !vardata.freefunc)
7821 0 : elog(ERROR,
7822 : "no function provided to release variable stats with");
7823 : }
7824 : else
7825 : {
7826 10344 : vardata.statsTuple =
7827 10344 : SearchSysCache3(STATRELATTINH,
7828 10344 : ObjectIdGetDatum(rte->relid),
7829 : Int16GetDatum(attnum),
7830 : BoolGetDatum(false));
7831 10344 : vardata.freefunc = ReleaseSysCache;
7832 : }
7833 : }
7834 : else
7835 : {
7836 : /*
7837 : * Looks like we've found an expression column in the index. Let's
7838 : * see if there's any stats for it.
7839 : */
7840 :
7841 : /* get the attnum from the 0-based index. */
7842 0 : attnum = iclause->indexcol + 1;
7843 :
7844 0 : if (get_index_stats_hook &&
7845 0 : (*get_index_stats_hook) (root, index->indexoid, attnum, &vardata))
7846 : {
7847 : /*
7848 : * The hook took control of acquiring a stats tuple. If it
7849 : * did supply a tuple, it'd better have supplied a freefunc.
7850 : */
7851 0 : if (HeapTupleIsValid(vardata.statsTuple) &&
7852 0 : !vardata.freefunc)
7853 0 : elog(ERROR, "no function provided to release variable stats with");
7854 : }
7855 : else
7856 : {
7857 0 : vardata.statsTuple = SearchSysCache3(STATRELATTINH,
7858 0 : ObjectIdGetDatum(index->indexoid),
7859 : Int16GetDatum(attnum),
7860 : BoolGetDatum(false));
7861 0 : vardata.freefunc = ReleaseSysCache;
7862 : }
7863 : }
7864 :
7865 10344 : if (HeapTupleIsValid(vardata.statsTuple))
7866 : {
7867 : AttStatsSlot sslot;
7868 :
7869 36 : if (get_attstatsslot(&sslot, vardata.statsTuple,
7870 : STATISTIC_KIND_CORRELATION, InvalidOid,
7871 : ATTSTATSSLOT_NUMBERS))
7872 : {
7873 36 : double varCorrelation = 0.0;
7874 :
7875 36 : if (sslot.nnumbers > 0)
7876 36 : varCorrelation = Abs(sslot.numbers[0]);
7877 :
7878 36 : if (varCorrelation > *indexCorrelation)
7879 36 : *indexCorrelation = varCorrelation;
7880 :
7881 36 : free_attstatsslot(&sslot);
7882 : }
7883 : }
7884 :
7885 10344 : ReleaseVariableStats(vardata);
7886 : }
7887 :
7888 10344 : qualSelectivity = clauselist_selectivity(root, indexQuals,
7889 10344 : baserel->relid,
7890 : JOIN_INNER, NULL);
7891 :
7892 : /*
7893 : * Now calculate the minimum possible ranges we could match with if all of
7894 : * the rows were in the perfect order in the table's heap.
7895 : */
7896 10344 : minimalRanges = ceil(indexRanges * qualSelectivity);
7897 :
7898 : /*
7899 : * Now estimate the number of ranges that we'll touch by using the
7900 : * indexCorrelation from the stats. Careful not to divide by zero (note
7901 : * we're using the absolute value of the correlation).
7902 : */
7903 10344 : if (*indexCorrelation < 1.0e-10)
7904 10308 : estimatedRanges = indexRanges;
7905 : else
7906 36 : estimatedRanges = Min(minimalRanges / *indexCorrelation, indexRanges);
7907 :
7908 : /* we expect to visit this portion of the table */
7909 10344 : selec = estimatedRanges / indexRanges;
7910 :
7911 10344 : CLAMP_PROBABILITY(selec);
7912 :
7913 10344 : *indexSelectivity = selec;
7914 :
7915 : /*
7916 : * Compute the index qual costs, much as in genericcostestimate, to add to
7917 : * the index costs. We can disregard indexorderbys, since BRIN doesn't
7918 : * support those.
7919 : */
7920 10344 : qual_arg_cost = index_other_operands_eval_cost(root, indexQuals);
7921 :
7922 : /*
7923 : * Compute the startup cost as the cost to read the whole revmap
7924 : * sequentially, including the cost to execute the index quals.
7925 : */
7926 10344 : *indexStartupCost =
7927 10344 : spc_seq_page_cost * statsData.revmapNumPages * loop_count;
7928 10344 : *indexStartupCost += qual_arg_cost;
7929 :
7930 : /*
7931 : * To read a BRIN index there might be a bit of back and forth over
7932 : * regular pages, as revmap might point to them out of sequential order;
7933 : * calculate the total cost as reading the whole index in random order.
7934 : */
7935 10344 : *indexTotalCost = *indexStartupCost +
7936 10344 : spc_random_page_cost * (numPages - statsData.revmapNumPages) * loop_count;
7937 :
7938 : /*
7939 : * Charge a small amount per range tuple which we expect to match to. This
7940 : * is meant to reflect the costs of manipulating the bitmap. The BRIN scan
7941 : * will set a bit for each page in the range when we find a matching
7942 : * range, so we must multiply the charge by the number of pages in the
7943 : * range.
7944 : */
7945 10344 : *indexTotalCost += 0.1 * cpu_operator_cost * estimatedRanges *
7946 10344 : statsData.pagesPerRange;
7947 :
7948 10344 : *indexPages = index->pages;
7949 10344 : }
|
