Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * mcv.c
4 : * POSTGRES multivariate MCV lists
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
8 : * Portions Copyright (c) 1994, Regents of the University of California
9 : *
10 : * IDENTIFICATION
11 : * src/backend/statistics/mcv.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <math.h>
18 :
19 : #include "access/htup_details.h"
20 : #include "catalog/pg_statistic_ext.h"
21 : #include "catalog/pg_statistic_ext_data.h"
22 : #include "fmgr.h"
23 : #include "funcapi.h"
24 : #include "nodes/nodeFuncs.h"
25 : #include "statistics/extended_stats_internal.h"
26 : #include "statistics/statistics.h"
27 : #include "utils/array.h"
28 : #include "utils/builtins.h"
29 : #include "utils/fmgrprotos.h"
30 : #include "utils/lsyscache.h"
31 : #include "utils/selfuncs.h"
32 : #include "utils/syscache.h"
33 : #include "utils/typcache.h"
34 :
35 : /*
36 : * Computes size of a serialized MCV item, depending on the number of
37 : * dimensions (columns) the statistic is defined on. The datum values are
38 : * stored in a separate array (deduplicated, to minimize the size), and
39 : * so the serialized items only store uint16 indexes into that array.
40 : *
41 : * Each serialized item stores (in this order):
42 : *
43 : * - indexes to values (ndim * sizeof(uint16))
44 : * - null flags (ndim * sizeof(bool))
45 : * - frequency (sizeof(double))
46 : * - base_frequency (sizeof(double))
47 : *
48 : * There is no alignment padding within an MCV item.
49 : * So in total each MCV item requires this many bytes:
50 : *
51 : * ndim * (sizeof(uint16) + sizeof(bool)) + 2 * sizeof(double)
52 : */
53 : #define ITEM_SIZE(ndims) \
54 : ((ndims) * (sizeof(uint16) + sizeof(bool)) + 2 * sizeof(double))
55 :
56 : /*
57 : * Used to compute size of serialized MCV list representation.
58 : */
59 : #define MinSizeOfMCVList \
60 : (VARHDRSZ + sizeof(uint32) * 3 + sizeof(AttrNumber))
61 :
62 : /*
63 : * Size of the serialized MCV list, excluding the space needed for
64 : * deduplicated per-dimension values. The macro is meant to be used
65 : * when it's not yet safe to access the serialized info about amount
66 : * of data for each column.
67 : */
68 : #define SizeOfMCVList(ndims,nitems) \
69 : ((MinSizeOfMCVList + sizeof(Oid) * (ndims)) + \
70 : ((ndims) * sizeof(DimensionInfo)) + \
71 : ((nitems) * ITEM_SIZE(ndims)))
72 :
73 : static MultiSortSupport build_mss(StatsBuildData *data);
74 :
75 : static SortItem *build_distinct_groups(int numrows, SortItem *items,
76 : MultiSortSupport mss, int *ndistinct);
77 :
78 : static SortItem **build_column_frequencies(SortItem *groups, int ngroups,
79 : MultiSortSupport mss, int *ncounts);
80 :
81 : static int count_distinct_groups(int numrows, SortItem *items,
82 : MultiSortSupport mss);
83 :
84 : /*
85 : * Compute new value for bitmap item, considering whether it's used for
86 : * clauses connected by AND/OR.
87 : */
88 : #define RESULT_MERGE(value, is_or, match) \
89 : ((is_or) ? ((value) || (match)) : ((value) && (match)))
90 :
91 : /*
92 : * When processing a list of clauses, the bitmap item may get set to a value
93 : * such that additional clauses can't change it. For example, when processing
94 : * a list of clauses connected to AND, as soon as the item gets set to 'false'
95 : * then it'll remain like that. Similarly clauses connected by OR and 'true'.
96 : *
97 : * Returns true when the value in the bitmap can't change no matter how the
98 : * remaining clauses are evaluated.
99 : */
100 : #define RESULT_IS_FINAL(value, is_or) ((is_or) ? (value) : (!(value)))
101 :
102 : /*
103 : * get_mincount_for_mcv_list
104 : * Determine the minimum number of times a value needs to appear in
105 : * the sample for it to be included in the MCV list.
106 : *
107 : * We want to keep only values that appear sufficiently often in the
108 : * sample that it is reasonable to extrapolate their sample frequencies to
109 : * the entire table. We do this by placing an upper bound on the relative
110 : * standard error of the sample frequency, so that any estimates the
111 : * planner generates from the MCV statistics can be expected to be
112 : * reasonably accurate.
113 : *
114 : * Since we are sampling without replacement, the sample frequency of a
115 : * particular value is described by a hypergeometric distribution. A
116 : * common rule of thumb when estimating errors in this situation is to
117 : * require at least 10 instances of the value in the sample, in which case
118 : * the distribution can be approximated by a normal distribution, and
119 : * standard error analysis techniques can be applied. Given a sample size
120 : * of n, a population size of N, and a sample frequency of p=cnt/n, the
121 : * standard error of the proportion p is given by
122 : * SE = sqrt(p*(1-p)/n) * sqrt((N-n)/(N-1))
123 : * where the second term is the finite population correction. To get
124 : * reasonably accurate planner estimates, we impose an upper bound on the
125 : * relative standard error of 20% -- i.e., SE/p < 0.2. This 20% relative
126 : * error bound is fairly arbitrary, but has been found empirically to work
127 : * well. Rearranging this formula gives a lower bound on the number of
128 : * instances of the value seen:
129 : * cnt > n*(N-n) / (N-n+0.04*n*(N-1))
130 : * This bound is at most 25, and approaches 0 as n approaches 0 or N. The
131 : * case where n approaches 0 cannot happen in practice, since the sample
132 : * size is at least 300. The case where n approaches N corresponds to
133 : * sampling the whole table, in which case it is reasonable to keep
134 : * the whole MCV list (have no lower bound), so it makes sense to apply
135 : * this formula for all inputs, even though the above derivation is
136 : * technically only valid when the right hand side is at least around 10.
137 : *
138 : * An alternative way to look at this formula is as follows -- assume that
139 : * the number of instances of the value seen scales up to the entire
140 : * table, so that the population count is K=N*cnt/n. Then the distribution
141 : * in the sample is a hypergeometric distribution parameterised by N, n
142 : * and K, and the bound above is mathematically equivalent to demanding
143 : * that the standard deviation of that distribution is less than 20% of
144 : * its mean. Thus the relative errors in any planner estimates produced
145 : * from the MCV statistics are likely to be not too large.
146 : */
147 : static double
148 218 : get_mincount_for_mcv_list(int samplerows, double totalrows)
149 : {
150 218 : double n = samplerows;
151 218 : double N = totalrows;
152 : double numer,
153 : denom;
154 :
155 218 : numer = n * (N - n);
156 218 : denom = N - n + 0.04 * n * (N - 1);
157 :
158 : /* Guard against division by zero (possible if n = N = 1) */
159 218 : if (denom == 0.0)
160 12 : return 0.0;
161 :
162 206 : return numer / denom;
163 : }
164 :
165 : /*
166 : * Builds MCV list from the set of sampled rows.
167 : *
168 : * The algorithm is quite simple:
169 : *
170 : * (1) sort the data (default collation, '<' for the data type)
171 : *
172 : * (2) count distinct groups, decide how many to keep
173 : *
174 : * (3) build the MCV list using the threshold determined in (2)
175 : *
176 : * (4) remove rows represented by the MCV from the sample
177 : *
178 : */
179 : MCVList *
180 218 : statext_mcv_build(StatsBuildData *data, double totalrows, int stattarget)
181 : {
182 : int i,
183 : numattrs,
184 : numrows,
185 : ngroups,
186 : nitems;
187 : double mincount;
188 : SortItem *items;
189 : SortItem *groups;
190 218 : MCVList *mcvlist = NULL;
191 : MultiSortSupport mss;
192 :
193 : /* comparator for all the columns */
194 218 : mss = build_mss(data);
195 :
196 : /* sort the rows */
197 218 : items = build_sorted_items(data, &nitems, mss,
198 : data->nattnums, data->attnums);
199 :
200 218 : if (!items)
201 0 : return NULL;
202 :
203 : /* for convenience */
204 218 : numattrs = data->nattnums;
205 218 : numrows = data->numrows;
206 :
207 : /* transform the sorted rows into groups (sorted by frequency) */
208 218 : groups = build_distinct_groups(nitems, items, mss, &ngroups);
209 :
210 : /*
211 : * The maximum number of MCV items to store, based on the statistics
212 : * target we computed for the statistics object (from the target set for
213 : * the object itself, attributes and the system default). In any case, we
214 : * can't keep more groups than we have available.
215 : */
216 218 : nitems = stattarget;
217 218 : if (nitems > ngroups)
218 126 : nitems = ngroups;
219 :
220 : /*
221 : * Decide how many items to keep in the MCV list. We can't use the same
222 : * algorithm as per-column MCV lists, because that only considers the
223 : * actual group frequency - but we're primarily interested in how the
224 : * actual frequency differs from the base frequency (product of simple
225 : * per-column frequencies, as if the columns were independent).
226 : *
227 : * Using the same algorithm might exclude items that are close to the
228 : * "average" frequency of the sample. But that does not say whether the
229 : * observed frequency is close to the base frequency or not. We also need
230 : * to consider unexpectedly uncommon items (again, compared to the base
231 : * frequency), and the single-column algorithm does not have to.
232 : *
233 : * We simply decide how many items to keep by computing the minimum count
234 : * using get_mincount_for_mcv_list() and then keep all items that seem to
235 : * be more common than that.
236 : */
237 218 : mincount = get_mincount_for_mcv_list(numrows, totalrows);
238 :
239 : /*
240 : * Walk the groups until we find the first group with a count below the
241 : * mincount threshold (the index of that group is the number of groups we
242 : * want to keep).
243 : */
244 10150 : for (i = 0; i < nitems; i++)
245 : {
246 9932 : if (groups[i].count < mincount)
247 : {
248 0 : nitems = i;
249 0 : break;
250 : }
251 : }
252 :
253 : /*
254 : * At this point, we know the number of items for the MCV list. There
255 : * might be none (for uniform distribution with many groups), and in that
256 : * case, there will be no MCV list. Otherwise, construct the MCV list.
257 : */
258 218 : if (nitems > 0)
259 : {
260 : int j;
261 : SortItem key;
262 : MultiSortSupport tmp;
263 :
264 : /* frequencies for values in each attribute */
265 : SortItem **freqs;
266 : int *nfreqs;
267 :
268 : /* used to search values */
269 218 : tmp = (MultiSortSupport) palloc(offsetof(MultiSortSupportData, ssup)
270 : + sizeof(SortSupportData));
271 :
272 : /* compute frequencies for values in each column */
273 218 : nfreqs = (int *) palloc0(sizeof(int) * numattrs);
274 218 : freqs = build_column_frequencies(groups, ngroups, mss, nfreqs);
275 :
276 : /*
277 : * Allocate the MCV list structure, set the global parameters.
278 : */
279 218 : mcvlist = (MCVList *) palloc0(offsetof(MCVList, items) +
280 218 : sizeof(MCVItem) * nitems);
281 :
282 218 : mcvlist->magic = STATS_MCV_MAGIC;
283 218 : mcvlist->type = STATS_MCV_TYPE_BASIC;
284 218 : mcvlist->ndimensions = numattrs;
285 218 : mcvlist->nitems = nitems;
286 :
287 : /* store info about data type OIDs */
288 804 : for (i = 0; i < numattrs; i++)
289 586 : mcvlist->types[i] = data->stats[i]->attrtypid;
290 :
291 : /* Copy the first chunk of groups into the result. */
292 10150 : for (i = 0; i < nitems; i++)
293 : {
294 : /* just point to the proper place in the list */
295 9932 : MCVItem *item = &mcvlist->items[i];
296 :
297 9932 : item->values = (Datum *) palloc(sizeof(Datum) * numattrs);
298 9932 : item->isnull = (bool *) palloc(sizeof(bool) * numattrs);
299 :
300 : /* copy values for the group */
301 9932 : memcpy(item->values, groups[i].values, sizeof(Datum) * numattrs);
302 9932 : memcpy(item->isnull, groups[i].isnull, sizeof(bool) * numattrs);
303 :
304 : /* groups should be sorted by frequency in descending order */
305 : Assert((i == 0) || (groups[i - 1].count >= groups[i].count));
306 :
307 : /* group frequency */
308 9932 : item->frequency = (double) groups[i].count / numrows;
309 :
310 : /* base frequency, if the attributes were independent */
311 9932 : item->base_frequency = 1.0;
312 36870 : for (j = 0; j < numattrs; j++)
313 : {
314 : SortItem *freq;
315 :
316 : /* single dimension */
317 26938 : tmp->ndims = 1;
318 26938 : tmp->ssup[0] = mss->ssup[j];
319 :
320 : /* fill search key */
321 26938 : key.values = &groups[i].values[j];
322 26938 : key.isnull = &groups[i].isnull[j];
323 :
324 26938 : freq = (SortItem *) bsearch_arg(&key, freqs[j], nfreqs[j],
325 : sizeof(SortItem),
326 : multi_sort_compare, tmp);
327 :
328 26938 : item->base_frequency *= ((double) freq->count) / numrows;
329 : }
330 : }
331 :
332 218 : pfree(nfreqs);
333 218 : pfree(freqs);
334 : }
335 :
336 218 : pfree(items);
337 218 : pfree(groups);
338 :
339 218 : return mcvlist;
340 : }
341 :
342 : /*
343 : * build_mss
344 : * Build a MultiSortSupport for the given StatsBuildData.
345 : */
346 : static MultiSortSupport
347 218 : build_mss(StatsBuildData *data)
348 : {
349 : int i;
350 218 : int numattrs = data->nattnums;
351 :
352 : /* Sort by multiple columns (using array of SortSupport) */
353 218 : MultiSortSupport mss = multi_sort_init(numattrs);
354 :
355 : /* prepare the sort functions for all the attributes */
356 804 : for (i = 0; i < numattrs; i++)
357 : {
358 586 : VacAttrStats *colstat = data->stats[i];
359 : TypeCacheEntry *type;
360 :
361 586 : type = lookup_type_cache(colstat->attrtypid, TYPECACHE_LT_OPR);
362 586 : if (type->lt_opr == InvalidOid) /* shouldn't happen */
363 0 : elog(ERROR, "cache lookup failed for ordering operator for type %u",
364 : colstat->attrtypid);
365 :
366 586 : multi_sort_add_dimension(mss, i, type->lt_opr, colstat->attrcollid);
367 : }
368 :
369 218 : return mss;
370 : }
371 :
372 : /*
373 : * count_distinct_groups
374 : * Count distinct combinations of SortItems in the array.
375 : *
376 : * The array is assumed to be sorted according to the MultiSortSupport.
377 : */
378 : static int
379 218 : count_distinct_groups(int numrows, SortItem *items, MultiSortSupport mss)
380 : {
381 : int i;
382 : int ndistinct;
383 :
384 218 : ndistinct = 1;
385 483200 : for (i = 1; i < numrows; i++)
386 : {
387 : /* make sure the array really is sorted */
388 : Assert(multi_sort_compare(&items[i], &items[i - 1], mss) >= 0);
389 :
390 482982 : if (multi_sort_compare(&items[i], &items[i - 1], mss) != 0)
391 89862 : ndistinct += 1;
392 : }
393 :
394 218 : return ndistinct;
395 : }
396 :
397 : /*
398 : * compare_sort_item_count
399 : * Comparator for sorting items by count (frequencies) in descending
400 : * order.
401 : */
402 : static int
403 96972 : compare_sort_item_count(const void *a, const void *b, void *arg)
404 : {
405 96972 : SortItem *ia = (SortItem *) a;
406 96972 : SortItem *ib = (SortItem *) b;
407 :
408 96972 : if (ia->count == ib->count)
409 95994 : return 0;
410 978 : else if (ia->count > ib->count)
411 648 : return -1;
412 :
413 330 : return 1;
414 : }
415 :
416 : /*
417 : * build_distinct_groups
418 : * Build an array of SortItems for distinct groups and counts matching
419 : * items.
420 : *
421 : * The 'items' array is assumed to be sorted.
422 : */
423 : static SortItem *
424 218 : build_distinct_groups(int numrows, SortItem *items, MultiSortSupport mss,
425 : int *ndistinct)
426 : {
427 : int i,
428 : j;
429 218 : int ngroups = count_distinct_groups(numrows, items, mss);
430 :
431 218 : SortItem *groups = (SortItem *) palloc(ngroups * sizeof(SortItem));
432 :
433 218 : j = 0;
434 218 : groups[0] = items[0];
435 218 : groups[0].count = 1;
436 :
437 483200 : for (i = 1; i < numrows; i++)
438 : {
439 : /* Assume sorted in ascending order. */
440 : Assert(multi_sort_compare(&items[i], &items[i - 1], mss) >= 0);
441 :
442 : /* New distinct group detected. */
443 482982 : if (multi_sort_compare(&items[i], &items[i - 1], mss) != 0)
444 : {
445 89862 : groups[++j] = items[i];
446 89862 : groups[j].count = 0;
447 : }
448 :
449 482982 : groups[j].count++;
450 : }
451 :
452 : /* ensure we filled the expected number of distinct groups */
453 : Assert(j + 1 == ngroups);
454 :
455 : /* Sort the distinct groups by frequency (in descending order). */
456 218 : qsort_interruptible(groups, ngroups, sizeof(SortItem),
457 : compare_sort_item_count, NULL);
458 :
459 218 : *ndistinct = ngroups;
460 218 : return groups;
461 : }
462 :
463 : /* compare sort items (single dimension) */
464 : static int
465 1029600 : sort_item_compare(const void *a, const void *b, void *arg)
466 : {
467 1029600 : SortSupport ssup = (SortSupport) arg;
468 1029600 : SortItem *ia = (SortItem *) a;
469 1029600 : SortItem *ib = (SortItem *) b;
470 :
471 2059200 : return ApplySortComparator(ia->values[0], ia->isnull[0],
472 1029600 : ib->values[0], ib->isnull[0],
473 : ssup);
474 : }
475 :
476 : /*
477 : * build_column_frequencies
478 : * Compute frequencies of values in each column.
479 : *
480 : * This returns an array of SortItems for each attribute the MCV is built
481 : * on, with a frequency (number of occurrences) for each value. This is
482 : * then used to compute "base" frequency of MCV items.
483 : *
484 : * All the memory is allocated in a single chunk, so that a single pfree
485 : * is enough to release it. We do not allocate space for values/isnull
486 : * arrays in the SortItems, because we can simply point into the input
487 : * groups directly.
488 : */
489 : static SortItem **
490 218 : build_column_frequencies(SortItem *groups, int ngroups,
491 : MultiSortSupport mss, int *ncounts)
492 : {
493 : int i,
494 : dim;
495 : SortItem **result;
496 : char *ptr;
497 :
498 : Assert(groups);
499 : Assert(ncounts);
500 :
501 : /* allocate arrays for all columns as a single chunk */
502 218 : ptr = palloc(MAXALIGN(sizeof(SortItem *) * mss->ndims) +
503 218 : mss->ndims * MAXALIGN(sizeof(SortItem) * ngroups));
504 :
505 : /* initial array of pointers */
506 218 : result = (SortItem **) ptr;
507 218 : ptr += MAXALIGN(sizeof(SortItem *) * mss->ndims);
508 :
509 804 : for (dim = 0; dim < mss->ndims; dim++)
510 : {
511 586 : SortSupport ssup = &mss->ssup[dim];
512 :
513 : /* array of values for a single column */
514 586 : result[dim] = (SortItem *) ptr;
515 586 : ptr += MAXALIGN(sizeof(SortItem) * ngroups);
516 :
517 : /* extract data for the dimension */
518 251420 : for (i = 0; i < ngroups; i++)
519 : {
520 : /* point into the input groups */
521 250834 : result[dim][i].values = &groups[i].values[dim];
522 250834 : result[dim][i].isnull = &groups[i].isnull[dim];
523 250834 : result[dim][i].count = groups[i].count;
524 : }
525 :
526 : /* sort the values, deduplicate */
527 586 : qsort_interruptible(result[dim], ngroups, sizeof(SortItem),
528 : sort_item_compare, ssup);
529 :
530 : /*
531 : * Identify distinct values, compute frequency (there might be
532 : * multiple MCV items containing this value, so we need to sum counts
533 : * from all of them.
534 : */
535 586 : ncounts[dim] = 1;
536 250834 : for (i = 1; i < ngroups; i++)
537 : {
538 250248 : if (sort_item_compare(&result[dim][i - 1], &result[dim][i], ssup) == 0)
539 : {
540 148200 : result[dim][ncounts[dim] - 1].count += result[dim][i].count;
541 148200 : continue;
542 : }
543 :
544 102048 : result[dim][ncounts[dim]] = result[dim][i];
545 :
546 102048 : ncounts[dim]++;
547 : }
548 : }
549 :
550 218 : return result;
551 : }
552 :
553 : /*
554 : * statext_mcv_load
555 : * Load the MCV list for the indicated pg_statistic_ext_data tuple.
556 : */
557 : MCVList *
558 606 : statext_mcv_load(Oid mvoid, bool inh)
559 : {
560 : MCVList *result;
561 : bool isnull;
562 : Datum mcvlist;
563 606 : HeapTuple htup = SearchSysCache2(STATEXTDATASTXOID,
564 : ObjectIdGetDatum(mvoid), BoolGetDatum(inh));
565 :
566 606 : if (!HeapTupleIsValid(htup))
567 0 : elog(ERROR, "cache lookup failed for statistics object %u", mvoid);
568 :
569 606 : mcvlist = SysCacheGetAttr(STATEXTDATASTXOID, htup,
570 : Anum_pg_statistic_ext_data_stxdmcv, &isnull);
571 :
572 606 : if (isnull)
573 0 : elog(ERROR,
574 : "requested statistics kind \"%c\" is not yet built for statistics object %u",
575 : STATS_EXT_MCV, mvoid);
576 :
577 606 : result = statext_mcv_deserialize(DatumGetByteaP(mcvlist));
578 :
579 606 : ReleaseSysCache(htup);
580 :
581 606 : return result;
582 : }
583 :
584 :
585 : /*
586 : * statext_mcv_serialize
587 : * Serialize MCV list into a pg_mcv_list value.
588 : *
589 : * The MCV items may include values of various data types, and it's reasonable
590 : * to expect redundancy (values for a given attribute, repeated for multiple
591 : * MCV list items). So we deduplicate the values into arrays, and then replace
592 : * the values by indexes into those arrays.
593 : *
594 : * The overall structure of the serialized representation looks like this:
595 : *
596 : * +---------------+----------------+---------------------+-------+
597 : * | header fields | dimension info | deduplicated values | items |
598 : * +---------------+----------------+---------------------+-------+
599 : *
600 : * Where dimension info stores information about the type of the K-th
601 : * attribute (e.g. typlen, typbyval and length of deduplicated values).
602 : * Deduplicated values store deduplicated values for each attribute. And
603 : * items store the actual MCV list items, with values replaced by indexes into
604 : * the arrays.
605 : *
606 : * When serializing the items, we use uint16 indexes. The number of MCV items
607 : * is limited by the statistics target (which is capped to 10k at the moment).
608 : * We might increase this to 65k and still fit into uint16, so there's a bit of
609 : * slack. Furthermore, this limit is on the number of distinct values per column,
610 : * and we usually have few of those (and various combinations of them for the
611 : * those MCV list). So uint16 seems fine for now.
612 : *
613 : * We don't really expect the serialization to save as much space as for
614 : * histograms, as we are not doing any bucket splits (which is the source
615 : * of high redundancy in histograms).
616 : *
617 : * TODO: Consider packing boolean flags (NULL) for each item into a single char
618 : * (or a longer type) instead of using an array of bool items.
619 : */
620 : bytea *
621 218 : statext_mcv_serialize(MCVList *mcvlist, VacAttrStats **stats)
622 : {
623 : int i;
624 : int dim;
625 218 : int ndims = mcvlist->ndimensions;
626 :
627 : SortSupport ssup;
628 : DimensionInfo *info;
629 :
630 : Size total_length;
631 :
632 : /* serialized items (indexes into arrays, etc.) */
633 : bytea *raw;
634 : char *ptr;
635 : char *endptr PG_USED_FOR_ASSERTS_ONLY;
636 :
637 : /* values per dimension (and number of non-NULL values) */
638 218 : Datum **values = (Datum **) palloc0(sizeof(Datum *) * ndims);
639 218 : int *counts = (int *) palloc0(sizeof(int) * ndims);
640 :
641 : /*
642 : * We'll include some rudimentary information about the attribute types
643 : * (length, by-val flag), so that we don't have to look them up while
644 : * deserializing the MCV list (we already have the type OID in the
645 : * header). This is safe because when changing the type of the attribute
646 : * the statistics gets dropped automatically. We need to store the info
647 : * about the arrays of deduplicated values anyway.
648 : */
649 218 : info = (DimensionInfo *) palloc0(sizeof(DimensionInfo) * ndims);
650 :
651 : /* sort support data for all attributes included in the MCV list */
652 218 : ssup = (SortSupport) palloc0(sizeof(SortSupportData) * ndims);
653 :
654 : /* collect and deduplicate values for each dimension (attribute) */
655 804 : for (dim = 0; dim < ndims; dim++)
656 : {
657 : int ndistinct;
658 : TypeCacheEntry *typentry;
659 :
660 : /*
661 : * Lookup the LT operator (can't get it from stats extra_data, as we
662 : * don't know how to interpret that - scalar vs. array etc.).
663 : */
664 586 : typentry = lookup_type_cache(stats[dim]->attrtypid, TYPECACHE_LT_OPR);
665 :
666 : /* copy important info about the data type (length, by-value) */
667 586 : info[dim].typlen = stats[dim]->attrtype->typlen;
668 586 : info[dim].typbyval = stats[dim]->attrtype->typbyval;
669 :
670 : /* allocate space for values in the attribute and collect them */
671 586 : values[dim] = (Datum *) palloc0(sizeof(Datum) * mcvlist->nitems);
672 :
673 27524 : for (i = 0; i < mcvlist->nitems; i++)
674 : {
675 : /* skip NULL values - we don't need to deduplicate those */
676 26938 : if (mcvlist->items[i].isnull[dim])
677 66 : continue;
678 :
679 : /* append the value at the end */
680 26872 : values[dim][counts[dim]] = mcvlist->items[i].values[dim];
681 26872 : counts[dim] += 1;
682 : }
683 :
684 : /* if there are just NULL values in this dimension, we're done */
685 586 : if (counts[dim] == 0)
686 6 : continue;
687 :
688 : /* sort and deduplicate the data */
689 580 : ssup[dim].ssup_cxt = CurrentMemoryContext;
690 580 : ssup[dim].ssup_collation = stats[dim]->attrcollid;
691 580 : ssup[dim].ssup_nulls_first = false;
692 :
693 580 : PrepareSortSupportFromOrderingOp(typentry->lt_opr, &ssup[dim]);
694 :
695 580 : qsort_interruptible(values[dim], counts[dim], sizeof(Datum),
696 580 : compare_scalars_simple, &ssup[dim]);
697 :
698 : /*
699 : * Walk through the array and eliminate duplicate values, but keep the
700 : * ordering (so that we can do a binary search later). We know there's
701 : * at least one item as (counts[dim] != 0), so we can skip the first
702 : * element.
703 : */
704 580 : ndistinct = 1; /* number of distinct values */
705 26872 : for (i = 1; i < counts[dim]; i++)
706 : {
707 : /* expect sorted array */
708 : Assert(compare_datums_simple(values[dim][i - 1], values[dim][i], &ssup[dim]) <= 0);
709 :
710 : /* if the value is the same as the previous one, we can skip it */
711 26292 : if (!compare_datums_simple(values[dim][i - 1], values[dim][i], &ssup[dim]))
712 11160 : continue;
713 :
714 15132 : values[dim][ndistinct] = values[dim][i];
715 15132 : ndistinct += 1;
716 : }
717 :
718 : /* we must not exceed PG_UINT16_MAX, as we use uint16 indexes */
719 : Assert(ndistinct <= PG_UINT16_MAX);
720 :
721 : /*
722 : * Store additional info about the attribute - number of deduplicated
723 : * values, and also size of the serialized data. For fixed-length data
724 : * types this is trivial to compute, for varwidth types we need to
725 : * actually walk the array and sum the sizes.
726 : */
727 580 : info[dim].nvalues = ndistinct;
728 :
729 580 : if (info[dim].typbyval) /* by-value data types */
730 : {
731 436 : info[dim].nbytes = info[dim].nvalues * info[dim].typlen;
732 :
733 : /*
734 : * We copy the data into the MCV item during deserialization, so
735 : * we don't need to allocate any extra space.
736 : */
737 436 : info[dim].nbytes_aligned = 0;
738 : }
739 144 : else if (info[dim].typlen > 0) /* fixed-length by-ref */
740 : {
741 : /*
742 : * We don't care about alignment in the serialized data, so we
743 : * pack the data as much as possible. But we also track how much
744 : * data will be needed after deserialization, and in that case we
745 : * need to account for alignment of each item.
746 : *
747 : * Note: As the items are fixed-length, we could easily compute
748 : * this during deserialization, but we do it here anyway.
749 : */
750 24 : info[dim].nbytes = info[dim].nvalues * info[dim].typlen;
751 24 : info[dim].nbytes_aligned = info[dim].nvalues * MAXALIGN(info[dim].typlen);
752 : }
753 120 : else if (info[dim].typlen == -1) /* varlena */
754 : {
755 120 : info[dim].nbytes = 0;
756 120 : info[dim].nbytes_aligned = 0;
757 2910 : for (i = 0; i < info[dim].nvalues; i++)
758 : {
759 : Size len;
760 :
761 : /*
762 : * For varlena values, we detoast the values and store the
763 : * length and data separately. We don't bother with alignment
764 : * here, which means that during deserialization we need to
765 : * copy the fields and only access the copies.
766 : */
767 2790 : values[dim][i] = PointerGetDatum(PG_DETOAST_DATUM(values[dim][i]));
768 :
769 : /* serialized length (uint32 length + data) */
770 2790 : len = VARSIZE_ANY_EXHDR(values[dim][i]);
771 2790 : info[dim].nbytes += sizeof(uint32); /* length */
772 2790 : info[dim].nbytes += len; /* value (no header) */
773 :
774 : /*
775 : * During deserialization we'll build regular varlena values
776 : * with full headers, and we need to align them properly.
777 : */
778 2790 : info[dim].nbytes_aligned += MAXALIGN(VARHDRSZ + len);
779 : }
780 : }
781 0 : else if (info[dim].typlen == -2) /* cstring */
782 : {
783 0 : info[dim].nbytes = 0;
784 0 : info[dim].nbytes_aligned = 0;
785 0 : for (i = 0; i < info[dim].nvalues; i++)
786 : {
787 : Size len;
788 :
789 : /*
790 : * cstring is handled similar to varlena - first we store the
791 : * length as uint32 and then the data. We don't care about
792 : * alignment, which means that during deserialization we need
793 : * to copy the fields and only access the copies.
794 : */
795 :
796 : /* c-strings include terminator, so +1 byte */
797 0 : len = strlen(DatumGetCString(values[dim][i])) + 1;
798 0 : info[dim].nbytes += sizeof(uint32); /* length */
799 0 : info[dim].nbytes += len; /* value */
800 :
801 : /* space needed for properly aligned deserialized copies */
802 0 : info[dim].nbytes_aligned += MAXALIGN(len);
803 : }
804 : }
805 :
806 : /* we know (count>0) so there must be some data */
807 : Assert(info[dim].nbytes > 0);
808 : }
809 :
810 : /*
811 : * Now we can finally compute how much space we'll actually need for the
812 : * whole serialized MCV list (varlena header, MCV header, dimension info
813 : * for each attribute, deduplicated values and items).
814 : */
815 218 : total_length = (3 * sizeof(uint32)) /* magic + type + nitems */
816 : + sizeof(AttrNumber) /* ndimensions */
817 218 : + (ndims * sizeof(Oid)); /* attribute types */
818 :
819 : /* dimension info */
820 218 : total_length += ndims * sizeof(DimensionInfo);
821 :
822 : /* add space for the arrays of deduplicated values */
823 804 : for (i = 0; i < ndims; i++)
824 586 : total_length += info[i].nbytes;
825 :
826 : /*
827 : * And finally account for the items (those are fixed-length, thanks to
828 : * replacing values with uint16 indexes into the deduplicated arrays).
829 : */
830 218 : total_length += mcvlist->nitems * ITEM_SIZE(dim);
831 :
832 : /*
833 : * Allocate space for the whole serialized MCV list (we'll skip bytes, so
834 : * we set them to zero to make the result more compressible).
835 : */
836 218 : raw = (bytea *) palloc0(VARHDRSZ + total_length);
837 218 : SET_VARSIZE(raw, VARHDRSZ + total_length);
838 :
839 218 : ptr = VARDATA(raw);
840 218 : endptr = ptr + total_length;
841 :
842 : /* copy the MCV list header fields, one by one */
843 218 : memcpy(ptr, &mcvlist->magic, sizeof(uint32));
844 218 : ptr += sizeof(uint32);
845 :
846 218 : memcpy(ptr, &mcvlist->type, sizeof(uint32));
847 218 : ptr += sizeof(uint32);
848 :
849 218 : memcpy(ptr, &mcvlist->nitems, sizeof(uint32));
850 218 : ptr += sizeof(uint32);
851 :
852 218 : memcpy(ptr, &mcvlist->ndimensions, sizeof(AttrNumber));
853 218 : ptr += sizeof(AttrNumber);
854 :
855 218 : memcpy(ptr, mcvlist->types, sizeof(Oid) * ndims);
856 218 : ptr += (sizeof(Oid) * ndims);
857 :
858 : /* store information about the attributes (data amounts, ...) */
859 218 : memcpy(ptr, info, sizeof(DimensionInfo) * ndims);
860 218 : ptr += sizeof(DimensionInfo) * ndims;
861 :
862 : /* Copy the deduplicated values for all attributes to the output. */
863 804 : for (dim = 0; dim < ndims; dim++)
864 : {
865 : /* remember the starting point for Asserts later */
866 586 : char *start PG_USED_FOR_ASSERTS_ONLY = ptr;
867 :
868 16298 : for (i = 0; i < info[dim].nvalues; i++)
869 : {
870 15712 : Datum value = values[dim][i];
871 :
872 15712 : if (info[dim].typbyval) /* passed by value */
873 : {
874 : Datum tmp;
875 :
876 : /*
877 : * For byval types, we need to copy just the significant bytes
878 : * - we can't use memcpy directly, as that assumes
879 : * little-endian behavior. store_att_byval does almost what
880 : * we need, but it requires a properly aligned buffer - the
881 : * output buffer does not guarantee that. So we simply use a
882 : * local Datum variable (which guarantees proper alignment),
883 : * and then copy the value from it.
884 : */
885 11812 : store_att_byval(&tmp, value, info[dim].typlen);
886 :
887 11812 : memcpy(ptr, &tmp, info[dim].typlen);
888 11812 : ptr += info[dim].typlen;
889 : }
890 3900 : else if (info[dim].typlen > 0) /* passed by reference */
891 : {
892 : /* no special alignment needed, treated as char array */
893 1110 : memcpy(ptr, DatumGetPointer(value), info[dim].typlen);
894 1110 : ptr += info[dim].typlen;
895 : }
896 2790 : else if (info[dim].typlen == -1) /* varlena */
897 : {
898 2790 : uint32 len = VARSIZE_ANY_EXHDR(DatumGetPointer(value));
899 :
900 : /* copy the length */
901 2790 : memcpy(ptr, &len, sizeof(uint32));
902 2790 : ptr += sizeof(uint32);
903 :
904 : /* data from the varlena value (without the header) */
905 2790 : memcpy(ptr, VARDATA_ANY(DatumGetPointer(value)), len);
906 2790 : ptr += len;
907 : }
908 0 : else if (info[dim].typlen == -2) /* cstring */
909 : {
910 0 : uint32 len = (uint32) strlen(DatumGetCString(value)) + 1;
911 :
912 : /* copy the length */
913 0 : memcpy(ptr, &len, sizeof(uint32));
914 0 : ptr += sizeof(uint32);
915 :
916 : /* value */
917 0 : memcpy(ptr, DatumGetCString(value), len);
918 0 : ptr += len;
919 : }
920 :
921 : /* no underflows or overflows */
922 : Assert((ptr > start) && ((ptr - start) <= info[dim].nbytes));
923 : }
924 :
925 : /* we should get exactly nbytes of data for this dimension */
926 : Assert((ptr - start) == info[dim].nbytes);
927 : }
928 :
929 : /* Serialize the items, with uint16 indexes instead of the values. */
930 10150 : for (i = 0; i < mcvlist->nitems; i++)
931 : {
932 9932 : MCVItem *mcvitem = &mcvlist->items[i];
933 :
934 : /* don't write beyond the allocated space */
935 : Assert(ptr <= (endptr - ITEM_SIZE(dim)));
936 :
937 : /* copy NULL and frequency flags into the serialized MCV */
938 9932 : memcpy(ptr, mcvitem->isnull, sizeof(bool) * ndims);
939 9932 : ptr += sizeof(bool) * ndims;
940 :
941 9932 : memcpy(ptr, &mcvitem->frequency, sizeof(double));
942 9932 : ptr += sizeof(double);
943 :
944 9932 : memcpy(ptr, &mcvitem->base_frequency, sizeof(double));
945 9932 : ptr += sizeof(double);
946 :
947 : /* store the indexes last */
948 36870 : for (dim = 0; dim < ndims; dim++)
949 : {
950 26938 : uint16 index = 0;
951 : Datum *value;
952 :
953 : /* do the lookup only for non-NULL values */
954 26938 : if (!mcvitem->isnull[dim])
955 : {
956 26872 : value = (Datum *) bsearch_arg(&mcvitem->values[dim], values[dim],
957 26872 : info[dim].nvalues, sizeof(Datum),
958 26872 : compare_scalars_simple, &ssup[dim]);
959 :
960 : Assert(value != NULL); /* serialization or deduplication
961 : * error */
962 :
963 : /* compute index within the deduplicated array */
964 26872 : index = (uint16) (value - values[dim]);
965 :
966 : /* check the index is within expected bounds */
967 : Assert(index < info[dim].nvalues);
968 : }
969 :
970 : /* copy the index into the serialized MCV */
971 26938 : memcpy(ptr, &index, sizeof(uint16));
972 26938 : ptr += sizeof(uint16);
973 : }
974 :
975 : /* make sure we don't overflow the allocated value */
976 : Assert(ptr <= endptr);
977 : }
978 :
979 : /* at this point we expect to match the total_length exactly */
980 : Assert(ptr == endptr);
981 :
982 218 : pfree(values);
983 218 : pfree(counts);
984 :
985 218 : return raw;
986 : }
987 :
988 : /*
989 : * statext_mcv_deserialize
990 : * Reads serialized MCV list into MCVList structure.
991 : *
992 : * All the memory needed by the MCV list is allocated as a single chunk, so
993 : * it's possible to simply pfree() it at once.
994 : */
995 : MCVList *
996 630 : statext_mcv_deserialize(bytea *data)
997 : {
998 : int dim,
999 : i;
1000 : Size expected_size;
1001 : MCVList *mcvlist;
1002 : char *raw;
1003 : char *ptr;
1004 : char *endptr PG_USED_FOR_ASSERTS_ONLY;
1005 :
1006 : int ndims,
1007 : nitems;
1008 630 : DimensionInfo *info = NULL;
1009 :
1010 : /* local allocation buffer (used only for deserialization) */
1011 630 : Datum **map = NULL;
1012 :
1013 : /* MCV list */
1014 : Size mcvlen;
1015 :
1016 : /* buffer used for the result */
1017 : Size datalen;
1018 : char *dataptr;
1019 : char *valuesptr;
1020 : char *isnullptr;
1021 :
1022 630 : if (data == NULL)
1023 0 : return NULL;
1024 :
1025 : /*
1026 : * We can't possibly deserialize a MCV list if there's not even a complete
1027 : * header. We need an explicit formula here, because we serialize the
1028 : * header fields one by one, so we need to ignore struct alignment.
1029 : */
1030 630 : if (VARSIZE_ANY(data) < MinSizeOfMCVList)
1031 0 : elog(ERROR, "invalid MCV size %zu (expected at least %zu)",
1032 : VARSIZE_ANY(data), MinSizeOfMCVList);
1033 :
1034 : /* read the MCV list header */
1035 630 : mcvlist = (MCVList *) palloc0(offsetof(MCVList, items));
1036 :
1037 : /* pointer to the data part (skip the varlena header) */
1038 630 : raw = (char *) data;
1039 630 : ptr = VARDATA_ANY(raw);
1040 630 : endptr = (char *) raw + VARSIZE_ANY(data);
1041 :
1042 : /* get the header and perform further sanity checks */
1043 630 : memcpy(&mcvlist->magic, ptr, sizeof(uint32));
1044 630 : ptr += sizeof(uint32);
1045 :
1046 630 : memcpy(&mcvlist->type, ptr, sizeof(uint32));
1047 630 : ptr += sizeof(uint32);
1048 :
1049 630 : memcpy(&mcvlist->nitems, ptr, sizeof(uint32));
1050 630 : ptr += sizeof(uint32);
1051 :
1052 630 : memcpy(&mcvlist->ndimensions, ptr, sizeof(AttrNumber));
1053 630 : ptr += sizeof(AttrNumber);
1054 :
1055 630 : if (mcvlist->magic != STATS_MCV_MAGIC)
1056 0 : elog(ERROR, "invalid MCV magic %u (expected %u)",
1057 : mcvlist->magic, STATS_MCV_MAGIC);
1058 :
1059 630 : if (mcvlist->type != STATS_MCV_TYPE_BASIC)
1060 0 : elog(ERROR, "invalid MCV type %u (expected %u)",
1061 : mcvlist->type, STATS_MCV_TYPE_BASIC);
1062 :
1063 630 : if (mcvlist->ndimensions == 0)
1064 0 : elog(ERROR, "invalid zero-length dimension array in MCVList");
1065 630 : else if ((mcvlist->ndimensions > STATS_MAX_DIMENSIONS) ||
1066 630 : (mcvlist->ndimensions < 0))
1067 0 : elog(ERROR, "invalid length (%d) dimension array in MCVList",
1068 : mcvlist->ndimensions);
1069 :
1070 630 : if (mcvlist->nitems == 0)
1071 0 : elog(ERROR, "invalid zero-length item array in MCVList");
1072 630 : else if (mcvlist->nitems > STATS_MCVLIST_MAX_ITEMS)
1073 0 : elog(ERROR, "invalid length (%u) item array in MCVList",
1074 : mcvlist->nitems);
1075 :
1076 630 : nitems = mcvlist->nitems;
1077 630 : ndims = mcvlist->ndimensions;
1078 :
1079 : /*
1080 : * Check amount of data including DimensionInfo for all dimensions and
1081 : * also the serialized items (including uint16 indexes). Also, walk
1082 : * through the dimension information and add it to the sum.
1083 : */
1084 630 : expected_size = SizeOfMCVList(ndims, nitems);
1085 :
1086 : /*
1087 : * Check that we have at least the dimension and info records, along with
1088 : * the items. We don't know the size of the serialized values yet. We need
1089 : * to do this check first, before accessing the dimension info.
1090 : */
1091 630 : if (VARSIZE_ANY(data) < expected_size)
1092 0 : elog(ERROR, "invalid MCV size %zu (expected %zu)",
1093 : VARSIZE_ANY(data), expected_size);
1094 :
1095 : /* Now copy the array of type Oids. */
1096 630 : memcpy(mcvlist->types, ptr, sizeof(Oid) * ndims);
1097 630 : ptr += (sizeof(Oid) * ndims);
1098 :
1099 : /* Now it's safe to access the dimension info. */
1100 630 : info = palloc(ndims * sizeof(DimensionInfo));
1101 :
1102 630 : memcpy(info, ptr, ndims * sizeof(DimensionInfo));
1103 630 : ptr += (ndims * sizeof(DimensionInfo));
1104 :
1105 : /* account for the value arrays */
1106 2532 : for (dim = 0; dim < ndims; dim++)
1107 : {
1108 : /*
1109 : * XXX I wonder if we can/should rely on asserts here. Maybe those
1110 : * checks should be done every time?
1111 : */
1112 : Assert(info[dim].nvalues >= 0);
1113 : Assert(info[dim].nbytes >= 0);
1114 :
1115 1902 : expected_size += info[dim].nbytes;
1116 : }
1117 :
1118 : /*
1119 : * Now we know the total expected MCV size, including all the pieces
1120 : * (header, dimension info. items and deduplicated data). So do the final
1121 : * check on size.
1122 : */
1123 630 : if (VARSIZE_ANY(data) != expected_size)
1124 0 : elog(ERROR, "invalid MCV size %zu (expected %zu)",
1125 : VARSIZE_ANY(data), expected_size);
1126 :
1127 : /*
1128 : * We need an array of Datum values for each dimension, so that we can
1129 : * easily translate the uint16 indexes later. We also need a top-level
1130 : * array of pointers to those per-dimension arrays.
1131 : *
1132 : * While allocating the arrays for dimensions, compute how much space we
1133 : * need for a copy of the by-ref data, as we can't simply point to the
1134 : * original values (it might go away).
1135 : */
1136 630 : datalen = 0; /* space for by-ref data */
1137 630 : map = (Datum **) palloc(ndims * sizeof(Datum *));
1138 :
1139 2532 : for (dim = 0; dim < ndims; dim++)
1140 : {
1141 1902 : map[dim] = (Datum *) palloc(sizeof(Datum) * info[dim].nvalues);
1142 :
1143 : /* space needed for a copy of data for by-ref types */
1144 1902 : datalen += info[dim].nbytes_aligned;
1145 : }
1146 :
1147 : /*
1148 : * Now resize the MCV list so that the allocation includes all the data.
1149 : *
1150 : * Allocate space for a copy of the data, as we can't simply reference the
1151 : * serialized data - it's not aligned properly, and it may disappear while
1152 : * we're still using the MCV list, e.g. due to catcache release.
1153 : *
1154 : * We do care about alignment here, because we will allocate all the
1155 : * pieces at once, but then use pointers to different parts.
1156 : */
1157 630 : mcvlen = MAXALIGN(offsetof(MCVList, items) + (sizeof(MCVItem) * nitems));
1158 :
1159 : /* arrays of values and isnull flags for all MCV items */
1160 630 : mcvlen += nitems * MAXALIGN(sizeof(Datum) * ndims);
1161 630 : mcvlen += nitems * MAXALIGN(sizeof(bool) * ndims);
1162 :
1163 : /* we don't quite need to align this, but it makes some asserts easier */
1164 630 : mcvlen += MAXALIGN(datalen);
1165 :
1166 : /* now resize the deserialized MCV list, and compute pointers to parts */
1167 630 : mcvlist = repalloc(mcvlist, mcvlen);
1168 :
1169 : /* pointer to the beginning of values/isnull arrays */
1170 630 : valuesptr = (char *) mcvlist
1171 630 : + MAXALIGN(offsetof(MCVList, items) + (sizeof(MCVItem) * nitems));
1172 :
1173 630 : isnullptr = valuesptr + (nitems * MAXALIGN(sizeof(Datum) * ndims));
1174 :
1175 630 : dataptr = isnullptr + (nitems * MAXALIGN(sizeof(bool) * ndims));
1176 :
1177 : /*
1178 : * Build mapping (index => value) for translating the serialized data into
1179 : * the in-memory representation.
1180 : */
1181 2532 : for (dim = 0; dim < ndims; dim++)
1182 : {
1183 : /* remember start position in the input array */
1184 1902 : char *start PG_USED_FOR_ASSERTS_ONLY = ptr;
1185 :
1186 1902 : if (info[dim].typbyval)
1187 : {
1188 : /* for by-val types we simply copy data into the mapping */
1189 56946 : for (i = 0; i < info[dim].nvalues; i++)
1190 : {
1191 55614 : Datum v = 0;
1192 :
1193 55614 : memcpy(&v, ptr, info[dim].typlen);
1194 55614 : ptr += info[dim].typlen;
1195 :
1196 55614 : map[dim][i] = fetch_att(&v, true, info[dim].typlen);
1197 :
1198 : /* no under/overflow of input array */
1199 : Assert(ptr <= (start + info[dim].nbytes));
1200 : }
1201 : }
1202 : else
1203 : {
1204 : /* for by-ref types we need to also make a copy of the data */
1205 :
1206 : /* passed by reference, but fixed length (name, tid, ...) */
1207 570 : if (info[dim].typlen > 0)
1208 : {
1209 2202 : for (i = 0; i < info[dim].nvalues; i++)
1210 : {
1211 2160 : memcpy(dataptr, ptr, info[dim].typlen);
1212 2160 : ptr += info[dim].typlen;
1213 :
1214 : /* just point into the array */
1215 2160 : map[dim][i] = PointerGetDatum(dataptr);
1216 2160 : dataptr += MAXALIGN(info[dim].typlen);
1217 : }
1218 : }
1219 528 : else if (info[dim].typlen == -1)
1220 : {
1221 : /* varlena */
1222 15024 : for (i = 0; i < info[dim].nvalues; i++)
1223 : {
1224 : uint32 len;
1225 :
1226 : /* read the uint32 length */
1227 14496 : memcpy(&len, ptr, sizeof(uint32));
1228 14496 : ptr += sizeof(uint32);
1229 :
1230 : /* the length is data-only */
1231 14496 : SET_VARSIZE(dataptr, len + VARHDRSZ);
1232 14496 : memcpy(VARDATA(dataptr), ptr, len);
1233 14496 : ptr += len;
1234 :
1235 : /* just point into the array */
1236 14496 : map[dim][i] = PointerGetDatum(dataptr);
1237 :
1238 : /* skip to place of the next deserialized value */
1239 14496 : dataptr += MAXALIGN(len + VARHDRSZ);
1240 : }
1241 : }
1242 0 : else if (info[dim].typlen == -2)
1243 : {
1244 : /* cstring */
1245 0 : for (i = 0; i < info[dim].nvalues; i++)
1246 : {
1247 : uint32 len;
1248 :
1249 0 : memcpy(&len, ptr, sizeof(uint32));
1250 0 : ptr += sizeof(uint32);
1251 :
1252 0 : memcpy(dataptr, ptr, len);
1253 0 : ptr += len;
1254 :
1255 : /* just point into the array */
1256 0 : map[dim][i] = PointerGetDatum(dataptr);
1257 0 : dataptr += MAXALIGN(len);
1258 : }
1259 : }
1260 :
1261 : /* no under/overflow of input array */
1262 : Assert(ptr <= (start + info[dim].nbytes));
1263 :
1264 : /* no overflow of the output mcv value */
1265 : Assert(dataptr <= ((char *) mcvlist + mcvlen));
1266 : }
1267 :
1268 : /* check we consumed input data for this dimension exactly */
1269 : Assert(ptr == (start + info[dim].nbytes));
1270 : }
1271 :
1272 : /* we should have also filled the MCV list exactly */
1273 : Assert(dataptr == ((char *) mcvlist + mcvlen));
1274 :
1275 : /* deserialize the MCV items and translate the indexes to Datums */
1276 43212 : for (i = 0; i < nitems; i++)
1277 : {
1278 42582 : MCVItem *item = &mcvlist->items[i];
1279 :
1280 42582 : item->values = (Datum *) valuesptr;
1281 42582 : valuesptr += MAXALIGN(sizeof(Datum) * ndims);
1282 :
1283 42582 : item->isnull = (bool *) isnullptr;
1284 42582 : isnullptr += MAXALIGN(sizeof(bool) * ndims);
1285 :
1286 42582 : memcpy(item->isnull, ptr, sizeof(bool) * ndims);
1287 42582 : ptr += sizeof(bool) * ndims;
1288 :
1289 42582 : memcpy(&item->frequency, ptr, sizeof(double));
1290 42582 : ptr += sizeof(double);
1291 :
1292 42582 : memcpy(&item->base_frequency, ptr, sizeof(double));
1293 42582 : ptr += sizeof(double);
1294 :
1295 : /* finally translate the indexes (for non-NULL only) */
1296 171696 : for (dim = 0; dim < ndims; dim++)
1297 : {
1298 : uint16 index;
1299 :
1300 129114 : memcpy(&index, ptr, sizeof(uint16));
1301 129114 : ptr += sizeof(uint16);
1302 :
1303 129114 : if (item->isnull[dim])
1304 306 : continue;
1305 :
1306 128808 : item->values[dim] = map[dim][index];
1307 : }
1308 :
1309 : /* check we're not overflowing the input */
1310 : Assert(ptr <= endptr);
1311 : }
1312 :
1313 : /* check that we processed all the data */
1314 : Assert(ptr == endptr);
1315 :
1316 : /* release the buffers used for mapping */
1317 2532 : for (dim = 0; dim < ndims; dim++)
1318 1902 : pfree(map[dim]);
1319 :
1320 630 : pfree(map);
1321 :
1322 630 : return mcvlist;
1323 : }
1324 :
1325 : /*
1326 : * SRF with details about buckets of a histogram:
1327 : *
1328 : * - item ID (0...nitems)
1329 : * - values (string array)
1330 : * - nulls only (boolean array)
1331 : * - frequency (double precision)
1332 : * - base_frequency (double precision)
1333 : *
1334 : * The input is the OID of the statistics, and there are no rows returned if
1335 : * the statistics contains no histogram.
1336 : */
1337 : Datum
1338 78 : pg_stats_ext_mcvlist_items(PG_FUNCTION_ARGS)
1339 : {
1340 : FuncCallContext *funcctx;
1341 :
1342 : /* stuff done only on the first call of the function */
1343 78 : if (SRF_IS_FIRSTCALL())
1344 : {
1345 : MemoryContext oldcontext;
1346 : MCVList *mcvlist;
1347 : TupleDesc tupdesc;
1348 :
1349 : /* create a function context for cross-call persistence */
1350 24 : funcctx = SRF_FIRSTCALL_INIT();
1351 :
1352 : /* switch to memory context appropriate for multiple function calls */
1353 24 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1354 :
1355 24 : mcvlist = statext_mcv_deserialize(PG_GETARG_BYTEA_P(0));
1356 :
1357 24 : funcctx->user_fctx = mcvlist;
1358 :
1359 : /* total number of tuples to be returned */
1360 24 : funcctx->max_calls = 0;
1361 24 : if (funcctx->user_fctx != NULL)
1362 24 : funcctx->max_calls = mcvlist->nitems;
1363 :
1364 : /* Build a tuple descriptor for our result type */
1365 24 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1366 0 : ereport(ERROR,
1367 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1368 : errmsg("function returning record called in context "
1369 : "that cannot accept type record")));
1370 24 : tupdesc = BlessTupleDesc(tupdesc);
1371 :
1372 : /*
1373 : * generate attribute metadata needed later to produce tuples from raw
1374 : * C strings
1375 : */
1376 24 : funcctx->attinmeta = TupleDescGetAttInMetadata(tupdesc);
1377 :
1378 24 : MemoryContextSwitchTo(oldcontext);
1379 : }
1380 :
1381 : /* stuff done on every call of the function */
1382 78 : funcctx = SRF_PERCALL_SETUP();
1383 :
1384 78 : if (funcctx->call_cntr < funcctx->max_calls) /* do when there is more
1385 : * left to send */
1386 : {
1387 : Datum values[5];
1388 : bool nulls[5];
1389 : HeapTuple tuple;
1390 : Datum result;
1391 54 : ArrayBuildState *astate_values = NULL;
1392 54 : ArrayBuildState *astate_nulls = NULL;
1393 :
1394 : int i;
1395 : MCVList *mcvlist;
1396 : MCVItem *item;
1397 :
1398 54 : mcvlist = (MCVList *) funcctx->user_fctx;
1399 :
1400 : Assert(funcctx->call_cntr < mcvlist->nitems);
1401 :
1402 54 : item = &mcvlist->items[funcctx->call_cntr];
1403 :
1404 180 : for (i = 0; i < mcvlist->ndimensions; i++)
1405 : {
1406 :
1407 126 : astate_nulls = accumArrayResult(astate_nulls,
1408 126 : BoolGetDatum(item->isnull[i]),
1409 : false,
1410 : BOOLOID,
1411 : CurrentMemoryContext);
1412 :
1413 126 : if (!item->isnull[i])
1414 : {
1415 : bool isvarlena;
1416 : Oid outfunc;
1417 : FmgrInfo fmgrinfo;
1418 : Datum val;
1419 : text *txt;
1420 :
1421 : /* lookup output func for the type */
1422 102 : getTypeOutputInfo(mcvlist->types[i], &outfunc, &isvarlena);
1423 102 : fmgr_info(outfunc, &fmgrinfo);
1424 :
1425 102 : val = FunctionCall1(&fmgrinfo, item->values[i]);
1426 102 : txt = cstring_to_text(DatumGetPointer(val));
1427 :
1428 102 : astate_values = accumArrayResult(astate_values,
1429 : PointerGetDatum(txt),
1430 : false,
1431 : TEXTOID,
1432 : CurrentMemoryContext);
1433 : }
1434 : else
1435 24 : astate_values = accumArrayResult(astate_values,
1436 : (Datum) 0,
1437 : true,
1438 : TEXTOID,
1439 : CurrentMemoryContext);
1440 : }
1441 :
1442 54 : values[0] = Int32GetDatum(funcctx->call_cntr);
1443 54 : values[1] = makeArrayResult(astate_values, CurrentMemoryContext);
1444 54 : values[2] = makeArrayResult(astate_nulls, CurrentMemoryContext);
1445 54 : values[3] = Float8GetDatum(item->frequency);
1446 54 : values[4] = Float8GetDatum(item->base_frequency);
1447 :
1448 : /* no NULLs in the tuple */
1449 54 : memset(nulls, 0, sizeof(nulls));
1450 :
1451 : /* build a tuple */
1452 54 : tuple = heap_form_tuple(funcctx->attinmeta->tupdesc, values, nulls);
1453 :
1454 : /* make the tuple into a datum */
1455 54 : result = HeapTupleGetDatum(tuple);
1456 :
1457 54 : SRF_RETURN_NEXT(funcctx, result);
1458 : }
1459 : else /* do when there is no more left */
1460 : {
1461 24 : SRF_RETURN_DONE(funcctx);
1462 : }
1463 : }
1464 :
1465 : /*
1466 : * pg_mcv_list_in - input routine for type pg_mcv_list.
1467 : *
1468 : * pg_mcv_list is real enough to be a table column, but it has no operations
1469 : * of its own, and disallows input too
1470 : */
1471 : Datum
1472 0 : pg_mcv_list_in(PG_FUNCTION_ARGS)
1473 : {
1474 : /*
1475 : * pg_mcv_list stores the data in binary form and parsing text input is
1476 : * not needed, so disallow this.
1477 : */
1478 0 : ereport(ERROR,
1479 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1480 : errmsg("cannot accept a value of type %s", "pg_mcv_list")));
1481 :
1482 : PG_RETURN_VOID(); /* keep compiler quiet */
1483 : }
1484 :
1485 :
1486 : /*
1487 : * pg_mcv_list_out - output routine for type pg_mcv_list.
1488 : *
1489 : * MCV lists are serialized into a bytea value, so we simply call byteaout()
1490 : * to serialize the value into text. But it'd be nice to serialize that into
1491 : * a meaningful representation (e.g. for inspection by people).
1492 : *
1493 : * XXX This should probably return something meaningful, similar to what
1494 : * pg_dependencies_out does. Not sure how to deal with the deduplicated
1495 : * values, though - do we want to expand that or not?
1496 : */
1497 : Datum
1498 12 : pg_mcv_list_out(PG_FUNCTION_ARGS)
1499 : {
1500 12 : return byteaout(fcinfo);
1501 : }
1502 :
1503 : /*
1504 : * pg_mcv_list_recv - binary input routine for type pg_mcv_list.
1505 : */
1506 : Datum
1507 0 : pg_mcv_list_recv(PG_FUNCTION_ARGS)
1508 : {
1509 0 : ereport(ERROR,
1510 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1511 : errmsg("cannot accept a value of type %s", "pg_mcv_list")));
1512 :
1513 : PG_RETURN_VOID(); /* keep compiler quiet */
1514 : }
1515 :
1516 : /*
1517 : * pg_mcv_list_send - binary output routine for type pg_mcv_list.
1518 : *
1519 : * MCV lists are serialized in a bytea value (although the type is named
1520 : * differently), so let's just send that.
1521 : */
1522 : Datum
1523 0 : pg_mcv_list_send(PG_FUNCTION_ARGS)
1524 : {
1525 0 : return byteasend(fcinfo);
1526 : }
1527 :
1528 : /*
1529 : * match the attribute/expression to a dimension of the statistic
1530 : *
1531 : * Returns the zero-based index of the matching statistics dimension.
1532 : * Optionally determines the collation.
1533 : */
1534 : static int
1535 1410 : mcv_match_expression(Node *expr, Bitmapset *keys, List *exprs, Oid *collid)
1536 : {
1537 : int idx;
1538 :
1539 1410 : if (IsA(expr, Var))
1540 : {
1541 : /* simple Var, so just lookup using varattno */
1542 1164 : Var *var = (Var *) expr;
1543 :
1544 1164 : if (collid)
1545 1098 : *collid = var->varcollid;
1546 :
1547 1164 : idx = bms_member_index(keys, var->varattno);
1548 :
1549 1164 : if (idx < 0)
1550 0 : elog(ERROR, "variable not found in statistics object");
1551 : }
1552 : else
1553 : {
1554 : /* expression - lookup in stats expressions */
1555 : ListCell *lc;
1556 :
1557 246 : if (collid)
1558 240 : *collid = exprCollation(expr);
1559 :
1560 : /* expressions are stored after the simple columns */
1561 246 : idx = bms_num_members(keys);
1562 456 : foreach(lc, exprs)
1563 : {
1564 456 : Node *stat_expr = (Node *) lfirst(lc);
1565 :
1566 456 : if (equal(expr, stat_expr))
1567 246 : break;
1568 :
1569 210 : idx++;
1570 : }
1571 :
1572 246 : if (lc == NULL)
1573 0 : elog(ERROR, "expression not found in statistics object");
1574 : }
1575 :
1576 1410 : return idx;
1577 : }
1578 :
1579 : /*
1580 : * mcv_get_match_bitmap
1581 : * Evaluate clauses using the MCV list, and update the match bitmap.
1582 : *
1583 : * A match bitmap keeps match/mismatch status for each MCV item, and we
1584 : * update it based on additional clauses. We also use it to skip items
1585 : * that can't possibly match (e.g. item marked as "mismatch" can't change
1586 : * to "match" when evaluating AND clause list).
1587 : *
1588 : * The function also returns a flag indicating whether there was an
1589 : * equality condition for all attributes, the minimum frequency in the MCV
1590 : * list, and a total MCV frequency (sum of frequencies for all items).
1591 : *
1592 : * XXX Currently the match bitmap uses a bool for each MCV item, which is
1593 : * somewhat wasteful as we could do with just a single bit, thus reducing
1594 : * the size to ~1/8. It would also allow us to combine bitmaps simply using
1595 : * & and |, which should be faster than min/max. The bitmaps are fairly
1596 : * small, though (thanks to the cap on the MCV list size).
1597 : */
1598 : static bool *
1599 846 : mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
1600 : Bitmapset *keys, List *exprs,
1601 : MCVList *mcvlist, bool is_or)
1602 : {
1603 : ListCell *l;
1604 : bool *matches;
1605 :
1606 : /* The bitmap may be partially built. */
1607 : Assert(clauses != NIL);
1608 : Assert(mcvlist != NULL);
1609 : Assert(mcvlist->nitems > 0);
1610 : Assert(mcvlist->nitems <= STATS_MCVLIST_MAX_ITEMS);
1611 :
1612 846 : matches = palloc(sizeof(bool) * mcvlist->nitems);
1613 846 : memset(matches, !is_or, sizeof(bool) * mcvlist->nitems);
1614 :
1615 : /*
1616 : * Loop through the list of clauses, and for each of them evaluate all the
1617 : * MCV items not yet eliminated by the preceding clauses.
1618 : */
1619 2430 : foreach(l, clauses)
1620 : {
1621 1584 : Node *clause = (Node *) lfirst(l);
1622 :
1623 : /* if it's a RestrictInfo, then extract the clause */
1624 1584 : if (IsA(clause, RestrictInfo))
1625 1470 : clause = (Node *) ((RestrictInfo *) clause)->clause;
1626 :
1627 : /*
1628 : * Handle the various types of clauses - OpClause, NullTest and
1629 : * AND/OR/NOT
1630 : */
1631 1584 : if (is_opclause(clause))
1632 : {
1633 1062 : OpExpr *expr = (OpExpr *) clause;
1634 : FmgrInfo opproc;
1635 :
1636 : /* valid only after examine_opclause_args returns true */
1637 : Node *clause_expr;
1638 : Const *cst;
1639 : bool expronleft;
1640 : int idx;
1641 : Oid collid;
1642 :
1643 1062 : fmgr_info(get_opcode(expr->opno), &opproc);
1644 :
1645 : /* extract the var/expr and const from the expression */
1646 1062 : if (!examine_opclause_args(expr->args, &clause_expr, &cst, &expronleft))
1647 0 : elog(ERROR, "incompatible clause");
1648 :
1649 : /* match the attribute/expression to a dimension of the statistic */
1650 1062 : idx = mcv_match_expression(clause_expr, keys, exprs, &collid);
1651 :
1652 : /*
1653 : * Walk through the MCV items and evaluate the current clause. We
1654 : * can skip items that were already ruled out, and terminate if
1655 : * there are no remaining MCV items that might possibly match.
1656 : */
1657 77418 : for (int i = 0; i < mcvlist->nitems; i++)
1658 : {
1659 76356 : bool match = true;
1660 76356 : MCVItem *item = &mcvlist->items[i];
1661 :
1662 : Assert(idx >= 0);
1663 :
1664 : /*
1665 : * When the MCV item or the Const value is NULL we can treat
1666 : * this as a mismatch. We must not call the operator because
1667 : * of strictness.
1668 : */
1669 76356 : if (item->isnull[idx] || cst->constisnull)
1670 : {
1671 48 : matches[i] = RESULT_MERGE(matches[i], is_or, false);
1672 48 : continue;
1673 : }
1674 :
1675 : /*
1676 : * Skip MCV items that can't change result in the bitmap. Once
1677 : * the value gets false for AND-lists, or true for OR-lists,
1678 : * we don't need to look at more clauses.
1679 : */
1680 76308 : if (RESULT_IS_FINAL(matches[i], is_or))
1681 30522 : continue;
1682 :
1683 : /*
1684 : * First check whether the constant is below the lower
1685 : * boundary (in that case we can skip the bucket, because
1686 : * there's no overlap).
1687 : *
1688 : * We don't store collations used to build the statistics, but
1689 : * we can use the collation for the attribute itself, as
1690 : * stored in varcollid. We do reset the statistics after a
1691 : * type change (including collation change), so this is OK.
1692 : * For expressions, we use the collation extracted from the
1693 : * expression itself.
1694 : */
1695 45786 : if (expronleft)
1696 43170 : match = DatumGetBool(FunctionCall2Coll(&opproc,
1697 : collid,
1698 43170 : item->values[idx],
1699 43170 : cst->constvalue));
1700 : else
1701 2616 : match = DatumGetBool(FunctionCall2Coll(&opproc,
1702 : collid,
1703 2616 : cst->constvalue,
1704 2616 : item->values[idx]));
1705 :
1706 : /* update the match bitmap with the result */
1707 45786 : matches[i] = RESULT_MERGE(matches[i], is_or, match);
1708 : }
1709 : }
1710 522 : else if (IsA(clause, ScalarArrayOpExpr))
1711 : {
1712 276 : ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) clause;
1713 : FmgrInfo opproc;
1714 :
1715 : /* valid only after examine_opclause_args returns true */
1716 : Node *clause_expr;
1717 : Const *cst;
1718 : bool expronleft;
1719 : Oid collid;
1720 : int idx;
1721 :
1722 : /* array evaluation */
1723 : ArrayType *arrayval;
1724 : int16 elmlen;
1725 : bool elmbyval;
1726 : char elmalign;
1727 : int num_elems;
1728 : Datum *elem_values;
1729 : bool *elem_nulls;
1730 :
1731 276 : fmgr_info(get_opcode(expr->opno), &opproc);
1732 :
1733 : /* extract the var/expr and const from the expression */
1734 276 : if (!examine_opclause_args(expr->args, &clause_expr, &cst, &expronleft))
1735 0 : elog(ERROR, "incompatible clause");
1736 :
1737 : /* We expect Var on left */
1738 276 : if (!expronleft)
1739 0 : elog(ERROR, "incompatible clause");
1740 :
1741 : /*
1742 : * Deconstruct the array constant, unless it's NULL (we'll cover
1743 : * that case below)
1744 : */
1745 276 : if (!cst->constisnull)
1746 : {
1747 276 : arrayval = DatumGetArrayTypeP(cst->constvalue);
1748 276 : get_typlenbyvalalign(ARR_ELEMTYPE(arrayval),
1749 : &elmlen, &elmbyval, &elmalign);
1750 276 : deconstruct_array(arrayval,
1751 : ARR_ELEMTYPE(arrayval),
1752 : elmlen, elmbyval, elmalign,
1753 : &elem_values, &elem_nulls, &num_elems);
1754 : }
1755 :
1756 : /* match the attribute/expression to a dimension of the statistic */
1757 276 : idx = mcv_match_expression(clause_expr, keys, exprs, &collid);
1758 :
1759 : /*
1760 : * Walk through the MCV items and evaluate the current clause. We
1761 : * can skip items that were already ruled out, and terminate if
1762 : * there are no remaining MCV items that might possibly match.
1763 : */
1764 22968 : for (int i = 0; i < mcvlist->nitems; i++)
1765 : {
1766 : int j;
1767 22692 : bool match = !expr->useOr;
1768 22692 : MCVItem *item = &mcvlist->items[i];
1769 :
1770 : /*
1771 : * When the MCV item or the Const value is NULL we can treat
1772 : * this as a mismatch. We must not call the operator because
1773 : * of strictness.
1774 : */
1775 22692 : if (item->isnull[idx] || cst->constisnull)
1776 : {
1777 18 : matches[i] = RESULT_MERGE(matches[i], is_or, false);
1778 18 : continue;
1779 : }
1780 :
1781 : /*
1782 : * Skip MCV items that can't change result in the bitmap. Once
1783 : * the value gets false for AND-lists, or true for OR-lists,
1784 : * we don't need to look at more clauses.
1785 : */
1786 22674 : if (RESULT_IS_FINAL(matches[i], is_or))
1787 11544 : continue;
1788 :
1789 35754 : for (j = 0; j < num_elems; j++)
1790 : {
1791 30978 : Datum elem_value = elem_values[j];
1792 30978 : bool elem_isnull = elem_nulls[j];
1793 : bool elem_match;
1794 :
1795 : /* NULL values always evaluate as not matching. */
1796 30978 : if (elem_isnull)
1797 : {
1798 2112 : match = RESULT_MERGE(match, expr->useOr, false);
1799 2112 : continue;
1800 : }
1801 :
1802 : /*
1803 : * Stop evaluating the array elements once we reach a
1804 : * matching value that can't change - ALL() is the same as
1805 : * AND-list, ANY() is the same as OR-list.
1806 : */
1807 28866 : if (RESULT_IS_FINAL(match, expr->useOr))
1808 6354 : break;
1809 :
1810 22512 : elem_match = DatumGetBool(FunctionCall2Coll(&opproc,
1811 : collid,
1812 22512 : item->values[idx],
1813 : elem_value));
1814 :
1815 22512 : match = RESULT_MERGE(match, expr->useOr, elem_match);
1816 : }
1817 :
1818 : /* update the match bitmap with the result */
1819 11130 : matches[i] = RESULT_MERGE(matches[i], is_or, match);
1820 : }
1821 : }
1822 246 : else if (IsA(clause, NullTest))
1823 : {
1824 66 : NullTest *expr = (NullTest *) clause;
1825 66 : Node *clause_expr = (Node *) (expr->arg);
1826 :
1827 : /* match the attribute/expression to a dimension of the statistic */
1828 66 : int idx = mcv_match_expression(clause_expr, keys, exprs, NULL);
1829 :
1830 : /*
1831 : * Walk through the MCV items and evaluate the current clause. We
1832 : * can skip items that were already ruled out, and terminate if
1833 : * there are no remaining MCV items that might possibly match.
1834 : */
1835 6078 : for (int i = 0; i < mcvlist->nitems; i++)
1836 : {
1837 6012 : bool match = false; /* assume mismatch */
1838 6012 : MCVItem *item = &mcvlist->items[i];
1839 :
1840 : /* if the clause mismatches the MCV item, update the bitmap */
1841 6012 : switch (expr->nulltesttype)
1842 : {
1843 4212 : case IS_NULL:
1844 4212 : match = (item->isnull[idx]) ? true : match;
1845 4212 : break;
1846 :
1847 1800 : case IS_NOT_NULL:
1848 1800 : match = (!item->isnull[idx]) ? true : match;
1849 1800 : break;
1850 : }
1851 :
1852 : /* now, update the match bitmap, depending on OR/AND type */
1853 6012 : matches[i] = RESULT_MERGE(matches[i], is_or, match);
1854 : }
1855 : }
1856 180 : else if (is_orclause(clause) || is_andclause(clause))
1857 66 : {
1858 : /* AND/OR clause, with all subclauses being compatible */
1859 :
1860 : int i;
1861 66 : BoolExpr *bool_clause = ((BoolExpr *) clause);
1862 66 : List *bool_clauses = bool_clause->args;
1863 :
1864 : /* match/mismatch bitmap for each MCV item */
1865 66 : bool *bool_matches = NULL;
1866 :
1867 : Assert(bool_clauses != NIL);
1868 : Assert(list_length(bool_clauses) >= 2);
1869 :
1870 : /* build the match bitmap for the OR-clauses */
1871 66 : bool_matches = mcv_get_match_bitmap(root, bool_clauses, keys, exprs,
1872 66 : mcvlist, is_orclause(clause));
1873 :
1874 : /*
1875 : * Merge the bitmap produced by mcv_get_match_bitmap into the
1876 : * current one. We need to consider if we're evaluating AND or OR
1877 : * condition when merging the results.
1878 : */
1879 4362 : for (i = 0; i < mcvlist->nitems; i++)
1880 4296 : matches[i] = RESULT_MERGE(matches[i], is_or, bool_matches[i]);
1881 :
1882 66 : pfree(bool_matches);
1883 : }
1884 114 : else if (is_notclause(clause))
1885 : {
1886 : /* NOT clause, with all subclauses compatible */
1887 :
1888 : int i;
1889 30 : BoolExpr *not_clause = ((BoolExpr *) clause);
1890 30 : List *not_args = not_clause->args;
1891 :
1892 : /* match/mismatch bitmap for each MCV item */
1893 30 : bool *not_matches = NULL;
1894 :
1895 : Assert(not_args != NIL);
1896 : Assert(list_length(not_args) == 1);
1897 :
1898 : /* build the match bitmap for the NOT-clause */
1899 30 : not_matches = mcv_get_match_bitmap(root, not_args, keys, exprs,
1900 : mcvlist, false);
1901 :
1902 : /*
1903 : * Merge the bitmap produced by mcv_get_match_bitmap into the
1904 : * current one. We're handling a NOT clause, so invert the result
1905 : * before merging it into the global bitmap.
1906 : */
1907 150 : for (i = 0; i < mcvlist->nitems; i++)
1908 120 : matches[i] = RESULT_MERGE(matches[i], is_or, !not_matches[i]);
1909 :
1910 30 : pfree(not_matches);
1911 : }
1912 84 : else if (IsA(clause, Var))
1913 : {
1914 : /* Var (has to be a boolean Var, possibly from below NOT) */
1915 :
1916 78 : Var *var = (Var *) (clause);
1917 :
1918 : /* match the attribute to a dimension of the statistic */
1919 78 : int idx = bms_member_index(keys, var->varattno);
1920 :
1921 : Assert(var->vartype == BOOLOID);
1922 :
1923 : /*
1924 : * Walk through the MCV items and evaluate the current clause. We
1925 : * can skip items that were already ruled out, and terminate if
1926 : * there are no remaining MCV items that might possibly match.
1927 : */
1928 378 : for (int i = 0; i < mcvlist->nitems; i++)
1929 : {
1930 300 : MCVItem *item = &mcvlist->items[i];
1931 300 : bool match = false;
1932 :
1933 : /* if the item is NULL, it's a mismatch */
1934 300 : if (!item->isnull[idx] && DatumGetBool(item->values[idx]))
1935 150 : match = true;
1936 :
1937 : /* update the result bitmap */
1938 300 : matches[i] = RESULT_MERGE(matches[i], is_or, match);
1939 : }
1940 : }
1941 : else
1942 : {
1943 : /* Otherwise, it must be a bare boolean-returning expression */
1944 : int idx;
1945 :
1946 : /* match the expression to a dimension of the statistic */
1947 6 : idx = mcv_match_expression(clause, keys, exprs, NULL);
1948 :
1949 : /*
1950 : * Walk through the MCV items and evaluate the current clause. We
1951 : * can skip items that were already ruled out, and terminate if
1952 : * there are no remaining MCV items that might possibly match.
1953 : */
1954 222 : for (int i = 0; i < mcvlist->nitems; i++)
1955 : {
1956 : bool match;
1957 216 : MCVItem *item = &mcvlist->items[i];
1958 :
1959 : /* "match" just means it's bool TRUE */
1960 216 : match = !item->isnull[idx] && DatumGetBool(item->values[idx]);
1961 :
1962 : /* now, update the match bitmap, depending on OR/AND type */
1963 216 : matches[i] = RESULT_MERGE(matches[i], is_or, match);
1964 : }
1965 : }
1966 : }
1967 :
1968 846 : return matches;
1969 : }
1970 :
1971 :
1972 : /*
1973 : * mcv_combine_selectivities
1974 : * Combine per-column and multi-column MCV selectivity estimates.
1975 : *
1976 : * simple_sel is a "simple" selectivity estimate (produced without using any
1977 : * extended statistics, essentially assuming independence of columns/clauses).
1978 : *
1979 : * mcv_sel and mcv_basesel are sums of the frequencies and base frequencies of
1980 : * all matching MCV items. The difference (mcv_sel - mcv_basesel) is then
1981 : * essentially interpreted as a correction to be added to simple_sel, as
1982 : * described below.
1983 : *
1984 : * mcv_totalsel is the sum of the frequencies of all MCV items (not just the
1985 : * matching ones). This is used as an upper bound on the portion of the
1986 : * selectivity estimates not covered by the MCV statistics.
1987 : *
1988 : * Note: While simple and base selectivities are defined in a quite similar
1989 : * way, the values are computed differently and are not therefore equal. The
1990 : * simple selectivity is computed as a product of per-clause estimates, while
1991 : * the base selectivity is computed by adding up base frequencies of matching
1992 : * items of the multi-column MCV list. So the values may differ for two main
1993 : * reasons - (a) the MCV list may not cover 100% of the data and (b) some of
1994 : * the MCV items did not match the estimated clauses.
1995 : *
1996 : * As both (a) and (b) reduce the base selectivity value, it generally holds
1997 : * that (simple_sel >= mcv_basesel). If the MCV list covers all the data, the
1998 : * values may be equal.
1999 : *
2000 : * So, other_sel = (simple_sel - mcv_basesel) is an estimate for the part not
2001 : * covered by the MCV list, and (mcv_sel - mcv_basesel) may be seen as a
2002 : * correction for the part covered by the MCV list. Those two statements are
2003 : * actually equivalent.
2004 : */
2005 : Selectivity
2006 798 : mcv_combine_selectivities(Selectivity simple_sel,
2007 : Selectivity mcv_sel,
2008 : Selectivity mcv_basesel,
2009 : Selectivity mcv_totalsel)
2010 : {
2011 : Selectivity other_sel;
2012 : Selectivity sel;
2013 :
2014 : /* estimated selectivity of values not covered by MCV matches */
2015 798 : other_sel = simple_sel - mcv_basesel;
2016 798 : CLAMP_PROBABILITY(other_sel);
2017 :
2018 : /* this non-MCV selectivity cannot exceed 1 - mcv_totalsel */
2019 798 : if (other_sel > 1.0 - mcv_totalsel)
2020 444 : other_sel = 1.0 - mcv_totalsel;
2021 :
2022 : /* overall selectivity is the sum of the MCV and non-MCV parts */
2023 798 : sel = mcv_sel + other_sel;
2024 798 : CLAMP_PROBABILITY(sel);
2025 :
2026 798 : return sel;
2027 : }
2028 :
2029 :
2030 : /*
2031 : * mcv_clauselist_selectivity
2032 : * Use MCV statistics to estimate the selectivity of an implicitly-ANDed
2033 : * list of clauses.
2034 : *
2035 : * This determines which MCV items match every clause in the list and returns
2036 : * the sum of the frequencies of those items.
2037 : *
2038 : * In addition, it returns the sum of the base frequencies of each of those
2039 : * items (that is the sum of the selectivities that each item would have if
2040 : * the columns were independent of one another), and the total selectivity of
2041 : * all the MCV items (not just the matching ones). These are expected to be
2042 : * used together with a "simple" selectivity estimate (one based only on
2043 : * per-column statistics) to produce an overall selectivity estimate that
2044 : * makes use of both per-column and multi-column statistics --- see
2045 : * mcv_combine_selectivities().
2046 : */
2047 : Selectivity
2048 510 : mcv_clauselist_selectivity(PlannerInfo *root, StatisticExtInfo *stat,
2049 : List *clauses, int varRelid,
2050 : JoinType jointype, SpecialJoinInfo *sjinfo,
2051 : RelOptInfo *rel,
2052 : Selectivity *basesel, Selectivity *totalsel)
2053 : {
2054 : int i;
2055 : MCVList *mcv;
2056 510 : Selectivity s = 0.0;
2057 510 : RangeTblEntry *rte = root->simple_rte_array[rel->relid];
2058 :
2059 : /* match/mismatch bitmap for each MCV item */
2060 510 : bool *matches = NULL;
2061 :
2062 : /* load the MCV list stored in the statistics object */
2063 510 : mcv = statext_mcv_load(stat->statOid, rte->inh);
2064 :
2065 : /* build a match bitmap for the clauses */
2066 510 : matches = mcv_get_match_bitmap(root, clauses, stat->keys, stat->exprs,
2067 : mcv, false);
2068 :
2069 : /* sum frequencies for all the matching MCV items */
2070 510 : *basesel = 0.0;
2071 510 : *totalsel = 0.0;
2072 37830 : for (i = 0; i < mcv->nitems; i++)
2073 : {
2074 37320 : *totalsel += mcv->items[i].frequency;
2075 :
2076 37320 : if (matches[i] != false)
2077 : {
2078 666 : *basesel += mcv->items[i].base_frequency;
2079 666 : s += mcv->items[i].frequency;
2080 : }
2081 : }
2082 :
2083 510 : return s;
2084 : }
2085 :
2086 :
2087 : /*
2088 : * mcv_clause_selectivity_or
2089 : * Use MCV statistics to estimate the selectivity of a clause that
2090 : * appears in an ORed list of clauses.
2091 : *
2092 : * As with mcv_clauselist_selectivity() this determines which MCV items match
2093 : * the clause and returns both the sum of the frequencies and the sum of the
2094 : * base frequencies of those items, as well as the sum of the frequencies of
2095 : * all MCV items (not just the matching ones) so that this information can be
2096 : * used by mcv_combine_selectivities() to produce a selectivity estimate that
2097 : * makes use of both per-column and multi-column statistics.
2098 : *
2099 : * Additionally, we return information to help compute the overall selectivity
2100 : * of the ORed list of clauses assumed to contain this clause. This function
2101 : * is intended to be called for each clause in the ORed list of clauses,
2102 : * allowing the overall selectivity to be computed using the following
2103 : * algorithm:
2104 : *
2105 : * Suppose P[n] = P(C[1] OR C[2] OR ... OR C[n]) is the combined selectivity
2106 : * of the first n clauses in the list. Then the combined selectivity taking
2107 : * into account the next clause C[n+1] can be written as
2108 : *
2109 : * P[n+1] = P[n] + P(C[n+1]) - P((C[1] OR ... OR C[n]) AND C[n+1])
2110 : *
2111 : * The final term above represents the overlap between the clauses examined so
2112 : * far and the (n+1)'th clause. To estimate its selectivity, we track the
2113 : * match bitmap for the ORed list of clauses examined so far and examine its
2114 : * intersection with the match bitmap for the (n+1)'th clause.
2115 : *
2116 : * We then also return the sums of the MCV item frequencies and base
2117 : * frequencies for the match bitmap intersection corresponding to the overlap
2118 : * term above, so that they can be combined with a simple selectivity estimate
2119 : * for that term.
2120 : *
2121 : * The parameter "or_matches" is an in/out parameter tracking the match bitmap
2122 : * for the clauses examined so far. The caller is expected to set it to NULL
2123 : * the first time it calls this function.
2124 : */
2125 : Selectivity
2126 240 : mcv_clause_selectivity_or(PlannerInfo *root, StatisticExtInfo *stat,
2127 : MCVList *mcv, Node *clause, bool **or_matches,
2128 : Selectivity *basesel, Selectivity *overlap_mcvsel,
2129 : Selectivity *overlap_basesel, Selectivity *totalsel)
2130 : {
2131 240 : Selectivity s = 0.0;
2132 : bool *new_matches;
2133 : int i;
2134 :
2135 : /* build the OR-matches bitmap, if not built already */
2136 240 : if (*or_matches == NULL)
2137 96 : *or_matches = palloc0(sizeof(bool) * mcv->nitems);
2138 :
2139 : /* build the match bitmap for the new clause */
2140 240 : new_matches = mcv_get_match_bitmap(root, list_make1(clause), stat->keys,
2141 : stat->exprs, mcv, false);
2142 :
2143 : /*
2144 : * Sum the frequencies for all the MCV items matching this clause and also
2145 : * those matching the overlap between this clause and any of the preceding
2146 : * clauses as described above.
2147 : */
2148 240 : *basesel = 0.0;
2149 240 : *overlap_mcvsel = 0.0;
2150 240 : *overlap_basesel = 0.0;
2151 240 : *totalsel = 0.0;
2152 15516 : for (i = 0; i < mcv->nitems; i++)
2153 : {
2154 15276 : *totalsel += mcv->items[i].frequency;
2155 :
2156 15276 : if (new_matches[i])
2157 : {
2158 336 : s += mcv->items[i].frequency;
2159 336 : *basesel += mcv->items[i].base_frequency;
2160 :
2161 336 : if ((*or_matches)[i])
2162 : {
2163 144 : *overlap_mcvsel += mcv->items[i].frequency;
2164 144 : *overlap_basesel += mcv->items[i].base_frequency;
2165 : }
2166 : }
2167 :
2168 : /* update the OR-matches bitmap for the next clause */
2169 15276 : (*or_matches)[i] = (*or_matches)[i] || new_matches[i];
2170 : }
2171 :
2172 240 : pfree(new_matches);
2173 :
2174 240 : return s;
2175 : }
|
