PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
array_typanalyze.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * array_typanalyze.c
4  * Functions for gathering statistics from array columns
5  *
6  * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  *
10  * IDENTIFICATION
11  * src/backend/utils/adt/array_typanalyze.c
12  *
13  *-------------------------------------------------------------------------
14  */
15 #include "postgres.h"
16 
17 #include "access/tuptoaster.h"
18 #include "catalog/pg_collation.h"
19 #include "commands/vacuum.h"
20 #include "utils/array.h"
21 #include "utils/builtins.h"
22 #include "utils/datum.h"
23 #include "utils/lsyscache.h"
24 #include "utils/typcache.h"
25 
26 
27 /*
28  * To avoid consuming too much memory, IO and CPU load during analysis, and/or
29  * too much space in the resulting pg_statistic rows, we ignore arrays that
30  * are wider than ARRAY_WIDTH_THRESHOLD (after detoasting!). Note that this
31  * number is considerably more than the similar WIDTH_THRESHOLD limit used
32  * in analyze.c's standard typanalyze code.
33  */
34 #define ARRAY_WIDTH_THRESHOLD 0x10000
35 
36 /* Extra data for compute_array_stats function */
37 typedef struct
38 {
39  /* Information about array element type */
40  Oid type_id; /* element type's OID */
41  Oid eq_opr; /* default equality operator's OID */
42  bool typbyval; /* physical properties of element type */
44  char typalign;
45 
46  /*
47  * Lookup data for element type's comparison and hash functions (these are
48  * in the type's typcache entry, which we expect to remain valid over the
49  * lifespan of the ANALYZE run)
50  */
53 
54  /* Saved state from std_typanalyze() */
58 
59 /*
60  * While compute_array_stats is running, we keep a pointer to the extra data
61  * here for use by assorted subroutines. compute_array_stats doesn't
62  * currently need to be re-entrant, so avoiding this is not worth the extra
63  * notational cruft that would be needed.
64  */
66 
67 /* A hash table entry for the Lossy Counting algorithm */
68 typedef struct
69 {
70  Datum key; /* This is 'e' from the LC algorithm. */
71  int frequency; /* This is 'f'. */
72  int delta; /* And this is 'delta'. */
73  int last_container; /* For de-duplication of array elements. */
74 } TrackItem;
75 
76 /* A hash table entry for distinct-elements counts */
77 typedef struct
78 {
79  int count; /* Count of distinct elements in an array */
80  int frequency; /* Number of arrays seen with this count */
81 } DECountItem;
82 
83 static void compute_array_stats(VacAttrStats *stats,
84  AnalyzeAttrFetchFunc fetchfunc, int samplerows, double totalrows);
85 static void prune_element_hashtable(HTAB *elements_tab, int b_current);
86 static uint32 element_hash(const void *key, Size keysize);
87 static int element_match(const void *key1, const void *key2, Size keysize);
88 static int element_compare(const void *key1, const void *key2);
89 static int trackitem_compare_frequencies_desc(const void *e1, const void *e2);
90 static int trackitem_compare_element(const void *e1, const void *e2);
91 static int countitem_compare_count(const void *e1, const void *e2);
92 
93 
94 /*
95  * array_typanalyze -- typanalyze function for array columns
96  */
97 Datum
99 {
101  Oid element_typeid;
102  TypeCacheEntry *typentry;
103  ArrayAnalyzeExtraData *extra_data;
104 
105  /*
106  * Call the standard typanalyze function. It may fail to find needed
107  * operators, in which case we also can't do anything, so just fail.
108  */
109  if (!std_typanalyze(stats))
110  PG_RETURN_BOOL(false);
111 
112  /*
113  * Check attribute data type is a varlena array (or a domain over one).
114  */
115  element_typeid = get_base_element_type(stats->attrtypid);
116  if (!OidIsValid(element_typeid))
117  elog(ERROR, "array_typanalyze was invoked for non-array type %u",
118  stats->attrtypid);
119 
120  /*
121  * Gather information about the element type. If we fail to find
122  * something, return leaving the state from std_typanalyze() in place.
123  */
124  typentry = lookup_type_cache(element_typeid,
128 
129  if (!OidIsValid(typentry->eq_opr) ||
130  !OidIsValid(typentry->cmp_proc_finfo.fn_oid) ||
131  !OidIsValid(typentry->hash_proc_finfo.fn_oid))
132  PG_RETURN_BOOL(true);
133 
134  /* Store our findings for use by compute_array_stats() */
135  extra_data = (ArrayAnalyzeExtraData *) palloc(sizeof(ArrayAnalyzeExtraData));
136  extra_data->type_id = typentry->type_id;
137  extra_data->eq_opr = typentry->eq_opr;
138  extra_data->typbyval = typentry->typbyval;
139  extra_data->typlen = typentry->typlen;
140  extra_data->typalign = typentry->typalign;
141  extra_data->cmp = &typentry->cmp_proc_finfo;
142  extra_data->hash = &typentry->hash_proc_finfo;
143 
144  /* Save old compute_stats and extra_data for scalar statistics ... */
145  extra_data->std_compute_stats = stats->compute_stats;
146  extra_data->std_extra_data = stats->extra_data;
147 
148  /* ... and replace with our info */
150  stats->extra_data = extra_data;
151 
152  /*
153  * Note we leave stats->minrows set as std_typanalyze set it. Should it
154  * be increased for array analysis purposes?
155  */
156 
157  PG_RETURN_BOOL(true);
158 }
159 
160 /*
161  * compute_array_stats() -- compute statistics for an array column
162  *
163  * This function computes statistics useful for determining selectivity of
164  * the array operators <@, &&, and @>. It is invoked by ANALYZE via the
165  * compute_stats hook after sample rows have been collected.
166  *
167  * We also invoke the standard compute_stats function, which will compute
168  * "scalar" statistics relevant to the btree-style array comparison operators.
169  * However, exact duplicates of an entire array may be rare despite many
170  * arrays sharing individual elements. This especially afflicts long arrays,
171  * which are also liable to lack all scalar statistics due to the low
172  * WIDTH_THRESHOLD used in analyze.c. So, in addition to the standard stats,
173  * we find the most common array elements and compute a histogram of distinct
174  * element counts.
175  *
176  * The algorithm used is Lossy Counting, as proposed in the paper "Approximate
177  * frequency counts over data streams" by G. S. Manku and R. Motwani, in
178  * Proceedings of the 28th International Conference on Very Large Data Bases,
179  * Hong Kong, China, August 2002, section 4.2. The paper is available at
180  * http://www.vldb.org/conf/2002/S10P03.pdf
181  *
182  * The Lossy Counting (aka LC) algorithm goes like this:
183  * Let s be the threshold frequency for an item (the minimum frequency we
184  * are interested in) and epsilon the error margin for the frequency. Let D
185  * be a set of triples (e, f, delta), where e is an element value, f is that
186  * element's frequency (actually, its current occurrence count) and delta is
187  * the maximum error in f. We start with D empty and process the elements in
188  * batches of size w. (The batch size is also known as "bucket size" and is
189  * equal to 1/epsilon.) Let the current batch number be b_current, starting
190  * with 1. For each element e we either increment its f count, if it's
191  * already in D, or insert a new triple into D with values (e, 1, b_current
192  * - 1). After processing each batch we prune D, by removing from it all
193  * elements with f + delta <= b_current. After the algorithm finishes we
194  * suppress all elements from D that do not satisfy f >= (s - epsilon) * N,
195  * where N is the total number of elements in the input. We emit the
196  * remaining elements with estimated frequency f/N. The LC paper proves
197  * that this algorithm finds all elements with true frequency at least s,
198  * and that no frequency is overestimated or is underestimated by more than
199  * epsilon. Furthermore, given reasonable assumptions about the input
200  * distribution, the required table size is no more than about 7 times w.
201  *
202  * In the absence of a principled basis for other particular values, we
203  * follow ts_typanalyze() and use parameters s = 0.07/K, epsilon = s/10.
204  * But we leave out the correction for stopwords, which do not apply to
205  * arrays. These parameters give bucket width w = K/0.007 and maximum
206  * expected hashtable size of about 1000 * K.
207  *
208  * Elements may repeat within an array. Since duplicates do not change the
209  * behavior of <@, && or @>, we want to count each element only once per
210  * array. Therefore, we store in the finished pg_statistic entry each
211  * element's frequency as the fraction of all non-null rows that contain it.
212  * We divide the raw counts by nonnull_cnt to get those figures.
213  */
214 static void
216  int samplerows, double totalrows)
217 {
218  ArrayAnalyzeExtraData *extra_data;
219  int num_mcelem;
220  int null_cnt = 0;
221  int null_elem_cnt = 0;
222  int analyzed_rows = 0;
223 
224  /* This is D from the LC algorithm. */
225  HTAB *elements_tab;
226  HASHCTL elem_hash_ctl;
227  HASH_SEQ_STATUS scan_status;
228 
229  /* This is the current bucket number from the LC algorithm */
230  int b_current;
231 
232  /* This is 'w' from the LC algorithm */
233  int bucket_width;
234  int array_no;
235  int64 element_no;
236  TrackItem *item;
237  int slot_idx;
238  HTAB *count_tab;
239  HASHCTL count_hash_ctl;
240  DECountItem *count_item;
241 
242  extra_data = (ArrayAnalyzeExtraData *) stats->extra_data;
243 
244  /*
245  * Invoke analyze.c's standard analysis function to create scalar-style
246  * stats for the column. It will expect its own extra_data pointer, so
247  * temporarily install that.
248  */
249  stats->extra_data = extra_data->std_extra_data;
250  extra_data->std_compute_stats(stats, fetchfunc, samplerows, totalrows);
251  stats->extra_data = extra_data;
252 
253  /*
254  * Set up static pointer for use by subroutines. We wait till here in
255  * case std_compute_stats somehow recursively invokes us (probably not
256  * possible, but ...)
257  */
258  array_extra_data = extra_data;
259 
260  /*
261  * We want statistics_target * 10 elements in the MCELEM array. This
262  * multiplier is pretty arbitrary, but is meant to reflect the fact that
263  * the number of individual elements tracked in pg_statistic ought to be
264  * more than the number of values for a simple scalar column.
265  */
266  num_mcelem = stats->attr->attstattarget * 10;
267 
268  /*
269  * We set bucket width equal to num_mcelem / 0.007 as per the comment
270  * above.
271  */
272  bucket_width = num_mcelem * 1000 / 7;
273 
274  /*
275  * Create the hashtable. It will be in local memory, so we don't need to
276  * worry about overflowing the initial size. Also we don't need to pay any
277  * attention to locking and memory management.
278  */
279  MemSet(&elem_hash_ctl, 0, sizeof(elem_hash_ctl));
280  elem_hash_ctl.keysize = sizeof(Datum);
281  elem_hash_ctl.entrysize = sizeof(TrackItem);
282  elem_hash_ctl.hash = element_hash;
283  elem_hash_ctl.match = element_match;
284  elem_hash_ctl.hcxt = CurrentMemoryContext;
285  elements_tab = hash_create("Analyzed elements table",
286  num_mcelem,
287  &elem_hash_ctl,
289 
290  /* hashtable for array distinct elements counts */
291  MemSet(&count_hash_ctl, 0, sizeof(count_hash_ctl));
292  count_hash_ctl.keysize = sizeof(int);
293  count_hash_ctl.entrysize = sizeof(DECountItem);
294  count_hash_ctl.hcxt = CurrentMemoryContext;
295  count_tab = hash_create("Array distinct element count table",
296  64,
297  &count_hash_ctl,
299 
300  /* Initialize counters. */
301  b_current = 1;
302  element_no = 0;
303 
304  /* Loop over the arrays. */
305  for (array_no = 0; array_no < samplerows; array_no++)
306  {
307  Datum value;
308  bool isnull;
309  ArrayType *array;
310  int num_elems;
311  Datum *elem_values;
312  bool *elem_nulls;
313  bool null_present;
314  int j;
315  int64 prev_element_no = element_no;
316  int distinct_count;
317  bool count_item_found;
318 
320 
321  value = fetchfunc(stats, array_no, &isnull);
322  if (isnull)
323  {
324  /* array is null, just count that */
325  null_cnt++;
326  continue;
327  }
328 
329  /* Skip too-large values. */
331  continue;
332  else
333  analyzed_rows++;
334 
335  /*
336  * Now detoast the array if needed, and deconstruct into datums.
337  */
338  array = DatumGetArrayTypeP(value);
339 
340  Assert(ARR_ELEMTYPE(array) == extra_data->type_id);
341  deconstruct_array(array,
342  extra_data->type_id,
343  extra_data->typlen,
344  extra_data->typbyval,
345  extra_data->typalign,
346  &elem_values, &elem_nulls, &num_elems);
347 
348  /*
349  * We loop through the elements in the array and add them to our
350  * tracking hashtable.
351  */
352  null_present = false;
353  for (j = 0; j < num_elems; j++)
354  {
355  Datum elem_value;
356  bool found;
357 
358  /* No null element processing other than flag setting here */
359  if (elem_nulls[j])
360  {
361  null_present = true;
362  continue;
363  }
364 
365  /* Lookup current element in hashtable, adding it if new */
366  elem_value = elem_values[j];
367  item = (TrackItem *) hash_search(elements_tab,
368  (const void *) &elem_value,
369  HASH_ENTER, &found);
370 
371  if (found)
372  {
373  /* The element value is already on the tracking list */
374 
375  /*
376  * The operators we assist ignore duplicate array elements, so
377  * count a given distinct element only once per array.
378  */
379  if (item->last_container == array_no)
380  continue;
381 
382  item->frequency++;
383  item->last_container = array_no;
384  }
385  else
386  {
387  /* Initialize new tracking list element */
388 
389  /*
390  * If element type is pass-by-reference, we must copy it into
391  * palloc'd space, so that we can release the array below. (We
392  * do this so that the space needed for element values is
393  * limited by the size of the hashtable; if we kept all the
394  * array values around, it could be much more.)
395  */
396  item->key = datumCopy(elem_value,
397  extra_data->typbyval,
398  extra_data->typlen);
399 
400  item->frequency = 1;
401  item->delta = b_current - 1;
402  item->last_container = array_no;
403  }
404 
405  /* element_no is the number of elements processed (ie N) */
406  element_no++;
407 
408  /* We prune the D structure after processing each bucket */
409  if (element_no % bucket_width == 0)
410  {
411  prune_element_hashtable(elements_tab, b_current);
412  b_current++;
413  }
414  }
415 
416  /* Count null element presence once per array. */
417  if (null_present)
418  null_elem_cnt++;
419 
420  /* Update frequency of the particular array distinct element count. */
421  distinct_count = (int) (element_no - prev_element_no);
422  count_item = (DECountItem *) hash_search(count_tab, &distinct_count,
423  HASH_ENTER,
424  &count_item_found);
425 
426  if (count_item_found)
427  count_item->frequency++;
428  else
429  count_item->frequency = 1;
430 
431  /* Free memory allocated while detoasting. */
432  if (PointerGetDatum(array) != value)
433  pfree(array);
434  pfree(elem_values);
435  pfree(elem_nulls);
436  }
437 
438  /* Skip pg_statistic slots occupied by standard statistics */
439  slot_idx = 0;
440  while (slot_idx < STATISTIC_NUM_SLOTS && stats->stakind[slot_idx] != 0)
441  slot_idx++;
442  if (slot_idx > STATISTIC_NUM_SLOTS - 2)
443  elog(ERROR, "insufficient pg_statistic slots for array stats");
444 
445  /* We can only compute real stats if we found some non-null values. */
446  if (analyzed_rows > 0)
447  {
448  int nonnull_cnt = analyzed_rows;
449  int count_items_count;
450  int i;
451  TrackItem **sort_table;
452  int track_len;
453  int64 cutoff_freq;
454  int64 minfreq,
455  maxfreq;
456 
457  /*
458  * We assume the standard stats code already took care of setting
459  * stats_valid, stanullfrac, stawidth, stadistinct. We'd have to
460  * re-compute those values if we wanted to not store the standard
461  * stats.
462  */
463 
464  /*
465  * Construct an array of the interesting hashtable items, that is,
466  * those meeting the cutoff frequency (s - epsilon)*N. Also identify
467  * the minimum and maximum frequencies among these items.
468  *
469  * Since epsilon = s/10 and bucket_width = 1/epsilon, the cutoff
470  * frequency is 9*N / bucket_width.
471  */
472  cutoff_freq = 9 * element_no / bucket_width;
473 
474  i = hash_get_num_entries(elements_tab); /* surely enough space */
475  sort_table = (TrackItem **) palloc(sizeof(TrackItem *) * i);
476 
477  hash_seq_init(&scan_status, elements_tab);
478  track_len = 0;
479  minfreq = element_no;
480  maxfreq = 0;
481  while ((item = (TrackItem *) hash_seq_search(&scan_status)) != NULL)
482  {
483  if (item->frequency > cutoff_freq)
484  {
485  sort_table[track_len++] = item;
486  minfreq = Min(minfreq, item->frequency);
487  maxfreq = Max(maxfreq, item->frequency);
488  }
489  }
490  Assert(track_len <= i);
491 
492  /* emit some statistics for debug purposes */
493  elog(DEBUG3, "compute_array_stats: target # mces = %d, "
494  "bucket width = %d, "
495  "# elements = " INT64_FORMAT ", hashtable size = %d, "
496  "usable entries = %d",
497  num_mcelem, bucket_width, element_no, i, track_len);
498 
499  /*
500  * If we obtained more elements than we really want, get rid of those
501  * with least frequencies. The easiest way is to qsort the array into
502  * descending frequency order and truncate the array.
503  */
504  if (num_mcelem < track_len)
505  {
506  qsort(sort_table, track_len, sizeof(TrackItem *),
508  /* reset minfreq to the smallest frequency we're keeping */
509  minfreq = sort_table[num_mcelem - 1]->frequency;
510  }
511  else
512  num_mcelem = track_len;
513 
514  /* Generate MCELEM slot entry */
515  if (num_mcelem > 0)
516  {
517  MemoryContext old_context;
518  Datum *mcelem_values;
519  float4 *mcelem_freqs;
520 
521  /*
522  * We want to store statistics sorted on the element value using
523  * the element type's default comparison function. This permits
524  * fast binary searches in selectivity estimation functions.
525  */
526  qsort(sort_table, num_mcelem, sizeof(TrackItem *),
528 
529  /* Must copy the target values into anl_context */
530  old_context = MemoryContextSwitchTo(stats->anl_context);
531 
532  /*
533  * We sorted statistics on the element value, but we want to be
534  * able to find the minimal and maximal frequencies without going
535  * through all the values. We also want the frequency of null
536  * elements. Store these three values at the end of mcelem_freqs.
537  */
538  mcelem_values = (Datum *) palloc(num_mcelem * sizeof(Datum));
539  mcelem_freqs = (float4 *) palloc((num_mcelem + 3) * sizeof(float4));
540 
541  /*
542  * See comments above about use of nonnull_cnt as the divisor for
543  * the final frequency estimates.
544  */
545  for (i = 0; i < num_mcelem; i++)
546  {
547  TrackItem *item = sort_table[i];
548 
549  mcelem_values[i] = datumCopy(item->key,
550  extra_data->typbyval,
551  extra_data->typlen);
552  mcelem_freqs[i] = (double) item->frequency /
553  (double) nonnull_cnt;
554  }
555  mcelem_freqs[i++] = (double) minfreq / (double) nonnull_cnt;
556  mcelem_freqs[i++] = (double) maxfreq / (double) nonnull_cnt;
557  mcelem_freqs[i++] = (double) null_elem_cnt / (double) nonnull_cnt;
558 
559  MemoryContextSwitchTo(old_context);
560 
561  stats->stakind[slot_idx] = STATISTIC_KIND_MCELEM;
562  stats->staop[slot_idx] = extra_data->eq_opr;
563  stats->stanumbers[slot_idx] = mcelem_freqs;
564  /* See above comment about extra stanumber entries */
565  stats->numnumbers[slot_idx] = num_mcelem + 3;
566  stats->stavalues[slot_idx] = mcelem_values;
567  stats->numvalues[slot_idx] = num_mcelem;
568  /* We are storing values of element type */
569  stats->statypid[slot_idx] = extra_data->type_id;
570  stats->statyplen[slot_idx] = extra_data->typlen;
571  stats->statypbyval[slot_idx] = extra_data->typbyval;
572  stats->statypalign[slot_idx] = extra_data->typalign;
573  slot_idx++;
574  }
575 
576  /* Generate DECHIST slot entry */
577  count_items_count = hash_get_num_entries(count_tab);
578  if (count_items_count > 0)
579  {
580  int num_hist = stats->attr->attstattarget;
581  DECountItem **sorted_count_items;
582  int j;
583  int delta;
584  int64 frac;
585  float4 *hist;
586 
587  /* num_hist must be at least 2 for the loop below to work */
588  num_hist = Max(num_hist, 2);
589 
590  /*
591  * Create an array of DECountItem pointers, and sort them into
592  * increasing count order.
593  */
594  sorted_count_items = (DECountItem **)
595  palloc(sizeof(DECountItem *) * count_items_count);
596  hash_seq_init(&scan_status, count_tab);
597  j = 0;
598  while ((count_item = (DECountItem *) hash_seq_search(&scan_status)) != NULL)
599  {
600  sorted_count_items[j++] = count_item;
601  }
602  qsort(sorted_count_items, count_items_count,
604 
605  /*
606  * Prepare to fill stanumbers with the histogram, followed by the
607  * average count. This array must be stored in anl_context.
608  */
609  hist = (float4 *)
611  sizeof(float4) * (num_hist + 1));
612  hist[num_hist] = (double) element_no / (double) nonnull_cnt;
613 
614  /*----------
615  * Construct the histogram of distinct-element counts (DECs).
616  *
617  * The object of this loop is to copy the min and max DECs to
618  * hist[0] and hist[num_hist - 1], along with evenly-spaced DECs
619  * in between (where "evenly-spaced" is with reference to the
620  * whole input population of arrays). If we had a complete sorted
621  * array of DECs, one per analyzed row, the i'th hist value would
622  * come from DECs[i * (analyzed_rows - 1) / (num_hist - 1)]
623  * (compare the histogram-making loop in compute_scalar_stats()).
624  * But instead of that we have the sorted_count_items[] array,
625  * which holds unique DEC values with their frequencies (that is,
626  * a run-length-compressed version of the full array). So we
627  * control advancing through sorted_count_items[] with the
628  * variable "frac", which is defined as (x - y) * (num_hist - 1),
629  * where x is the index in the notional DECs array corresponding
630  * to the start of the next sorted_count_items[] element's run,
631  * and y is the index in DECs from which we should take the next
632  * histogram value. We have to advance whenever x <= y, that is
633  * frac <= 0. The x component is the sum of the frequencies seen
634  * so far (up through the current sorted_count_items[] element),
635  * and of course y * (num_hist - 1) = i * (analyzed_rows - 1),
636  * per the subscript calculation above. (The subscript calculation
637  * implies dropping any fractional part of y; in this formulation
638  * that's handled by not advancing until frac reaches 1.)
639  *
640  * Even though frac has a bounded range, it could overflow int32
641  * when working with very large statistics targets, so we do that
642  * math in int64.
643  *----------
644  */
645  delta = analyzed_rows - 1;
646  j = 0; /* current index in sorted_count_items */
647  /* Initialize frac for sorted_count_items[0]; y is initially 0 */
648  frac = (int64) sorted_count_items[0]->frequency * (num_hist - 1);
649  for (i = 0; i < num_hist; i++)
650  {
651  while (frac <= 0)
652  {
653  /* Advance, and update x component of frac */
654  j++;
655  frac += (int64) sorted_count_items[j]->frequency * (num_hist - 1);
656  }
657  hist[i] = sorted_count_items[j]->count;
658  frac -= delta; /* update y for upcoming i increment */
659  }
660  Assert(j == count_items_count - 1);
661 
662  stats->stakind[slot_idx] = STATISTIC_KIND_DECHIST;
663  stats->staop[slot_idx] = extra_data->eq_opr;
664  stats->stanumbers[slot_idx] = hist;
665  stats->numnumbers[slot_idx] = num_hist + 1;
666  slot_idx++;
667  }
668  }
669 
670  /*
671  * We don't need to bother cleaning up any of our temporary palloc's. The
672  * hashtable should also go away, as it used a child memory context.
673  */
674 }
675 
676 /*
677  * A function to prune the D structure from the Lossy Counting algorithm.
678  * Consult compute_tsvector_stats() for wider explanation.
679  */
680 static void
681 prune_element_hashtable(HTAB *elements_tab, int b_current)
682 {
683  HASH_SEQ_STATUS scan_status;
684  TrackItem *item;
685 
686  hash_seq_init(&scan_status, elements_tab);
687  while ((item = (TrackItem *) hash_seq_search(&scan_status)) != NULL)
688  {
689  if (item->frequency + item->delta <= b_current)
690  {
691  Datum value = item->key;
692 
693  if (hash_search(elements_tab, (const void *) &item->key,
694  HASH_REMOVE, NULL) == NULL)
695  elog(ERROR, "hash table corrupted");
696  /* We should free memory if element is not passed by value */
697  if (!array_extra_data->typbyval)
698  pfree(DatumGetPointer(value));
699  }
700  }
701 }
702 
703 /*
704  * Hash function for elements.
705  *
706  * We use the element type's default hash opclass, and the default collation
707  * if the type is collation-sensitive.
708  */
709 static uint32
710 element_hash(const void *key, Size keysize)
711 {
712  Datum d = *((const Datum *) key);
713  Datum h;
714 
715  h = FunctionCall1Coll(array_extra_data->hash, DEFAULT_COLLATION_OID, d);
716  return DatumGetUInt32(h);
717 }
718 
719 /*
720  * Matching function for elements, to be used in hashtable lookups.
721  */
722 static int
723 element_match(const void *key1, const void *key2, Size keysize)
724 {
725  /* The keysize parameter is superfluous here */
726  return element_compare(key1, key2);
727 }
728 
729 /*
730  * Comparison function for elements.
731  *
732  * We use the element type's default btree opclass, and the default collation
733  * if the type is collation-sensitive.
734  *
735  * XXX consider using SortSupport infrastructure
736  */
737 static int
738 element_compare(const void *key1, const void *key2)
739 {
740  Datum d1 = *((const Datum *) key1);
741  Datum d2 = *((const Datum *) key2);
742  Datum c;
743 
744  c = FunctionCall2Coll(array_extra_data->cmp, DEFAULT_COLLATION_OID, d1, d2);
745  return DatumGetInt32(c);
746 }
747 
748 /*
749  * qsort() comparator for sorting TrackItems by frequencies (descending sort)
750  */
751 static int
752 trackitem_compare_frequencies_desc(const void *e1, const void *e2)
753 {
754  const TrackItem *const *t1 = (const TrackItem *const *) e1;
755  const TrackItem *const *t2 = (const TrackItem *const *) e2;
756 
757  return (*t2)->frequency - (*t1)->frequency;
758 }
759 
760 /*
761  * qsort() comparator for sorting TrackItems by element values
762  */
763 static int
764 trackitem_compare_element(const void *e1, const void *e2)
765 {
766  const TrackItem *const *t1 = (const TrackItem *const *) e1;
767  const TrackItem *const *t2 = (const TrackItem *const *) e2;
768 
769  return element_compare(&(*t1)->key, &(*t2)->key);
770 }
771 
772 /*
773  * qsort() comparator for sorting DECountItems by count
774  */
775 static int
776 countitem_compare_count(const void *e1, const void *e2)
777 {
778  const DECountItem *const *t1 = (const DECountItem *const *) e1;
779  const DECountItem *const *t2 = (const DECountItem *const *) e2;
780 
781  if ((*t1)->count < (*t2)->count)
782  return -1;
783  else if ((*t1)->count == (*t2)->count)
784  return 0;
785  else
786  return 1;
787 }
signed short int16
Definition: c.h:245
#define DatumGetUInt32(X)
Definition: postgres.h:492
Definition: fmgr.h:56
static void prune_element_hashtable(HTAB *elements_tab, int b_current)
#define HASH_CONTEXT
Definition: hsearch.h:93
#define HASH_ELEM
Definition: hsearch.h:87
MemoryContext hcxt
Definition: hsearch.h:78
#define DatumGetInt32(X)
Definition: postgres.h:478
#define DEBUG3
Definition: elog.h:23
#define PointerGetDatum(X)
Definition: postgres.h:562
#define Min(x, y)
Definition: c.h:812
#define TYPECACHE_HASH_PROC_FINFO
Definition: typcache.h:121
Datum * stavalues[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:108
static void compute_array_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc, int samplerows, double totalrows)
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
Size entrysize
Definition: hsearch.h:73
#define TYPECACHE_EQ_OPR
Definition: typcache.h:114
#define MemSet(start, val, len)
Definition: c.h:863
long hash_get_num_entries(HTAB *hashp)
Definition: dynahash.c:1331
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:241
bool std_typanalyze(VacAttrStats *stats)
Definition: analyze.c:1739
bool statypbyval[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:118
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:902
static int countitem_compare_count(const void *e1, const void *e2)
Datum FunctionCall2Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2)
Definition: fmgr.c:1042
unsigned int Oid
Definition: postgres_ext.h:31
int16 typlen
Definition: typcache.h:37
bool typbyval
Definition: typcache.h:38
#define OidIsValid(objectId)
Definition: c.h:532
char statypalign[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:119
Form_pg_attribute attr
Definition: vacuum.h:81
static int element_match(const void *key1, const void *key2, Size keysize)
FmgrInfo cmp_proc_finfo
Definition: typcache.h:71
Definition: dynahash.c:208
void pfree(void *pointer)
Definition: mcxt.c:949
Size toast_raw_datum_size(Datum value)
Definition: tuptoaster.c:353
Oid attrtypid
Definition: vacuum.h:82
#define ERROR
Definition: elog.h:43
#define STATISTIC_KIND_DECHIST
Definition: pg_statistic.h:270
static struct @121 value
char * c
AnalyzeAttrComputeStatsFunc std_compute_stats
#define DEFAULT_COLLATION_OID
Definition: pg_collation.h:75
int numnumbers[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:105
LexemeHashKey key
Definition: ts_typanalyze.c:33
unsigned int uint32
Definition: c.h:258
FmgrInfo hash_proc_finfo
Definition: typcache.h:72
static ArrayAnalyzeExtraData * array_extra_data
MemoryContext CurrentMemoryContext
Definition: mcxt.c:37
Datum datumCopy(Datum value, bool typByVal, int typLen)
Definition: datum.c:128
Datum array_typanalyze(PG_FUNCTION_ARGS)
static int trackitem_compare_element(const void *e1, const void *e2)
Oid staop[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:104
float float4
Definition: c.h:374
#define HASH_BLOBS
Definition: hsearch.h:88
#define PG_RETURN_BOOL(x)
Definition: fmgr.h:319
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
uintptr_t Datum
Definition: postgres.h:372
int16 stakind[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:103
Datum FunctionCall1Coll(FmgrInfo *flinfo, Oid collation, Datum arg1)
Definition: fmgr.c:1022
Size keysize
Definition: hsearch.h:72
Datum(* AnalyzeAttrFetchFunc)(VacAttrStatsP stats, int rownum, bool *isNull)
Definition: vacuum.h:61
HashCompareFunc match
Definition: hsearch.h:75
Oid statypid[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:116
TypeCacheEntry * lookup_type_cache(Oid type_id, int flags)
Definition: typcache.c:305
Oid fn_oid
Definition: fmgr.h:59
#define STATISTIC_NUM_SLOTS
Definition: pg_statistic.h:121
#define Max(x, y)
Definition: c.h:806
#define ARRAY_WIDTH_THRESHOLD
#define Assert(condition)
Definition: c.h:681
float4 * stanumbers[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:106
#define HASH_COMPARE
Definition: hsearch.h:90
size_t Size
Definition: c.h:350
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1385
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1375
#define INT64_FORMAT
Definition: c.h:300
MemoryContext anl_context
Definition: vacuum.h:85
#define DatumGetPointer(X)
Definition: postgres.h:555
static uint32 element_hash(const void *key, Size keysize)
void deconstruct_array(ArrayType *array, Oid elmtype, int elmlen, bool elmbyval, char elmalign, Datum **elemsp, bool **nullsp, int *nelemsp)
Definition: arrayfuncs.c:3449
Oid get_base_element_type(Oid typid)
Definition: lsyscache.c:2557
void(* AnalyzeAttrComputeStatsFunc)(VacAttrStatsP stats, AnalyzeAttrFetchFunc fetchfunc, int samplerows, double totalrows)
Definition: vacuum.h:64
int numvalues[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:107
char typalign
Definition: typcache.h:39
void * palloc(Size size)
Definition: mcxt.c:848
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:706
int16 statyplen[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:117
#define STATISTIC_KIND_MCELEM
Definition: pg_statistic.h:257
int i
AnalyzeAttrComputeStatsFunc compute_stats
Definition: vacuum.h:91
#define PG_FUNCTION_ARGS
Definition: fmgr.h:158
#define TYPECACHE_CMP_PROC_FINFO
Definition: typcache.h:120
void * extra_data
Definition: vacuum.h:93
#define elog
Definition: elog.h:219
#define qsort(a, b, c, d)
Definition: port.h:447
void vacuum_delay_point(void)
Definition: vacuum.c:1658
static int element_compare(const void *key1, const void *key2)
#define ARR_ELEMTYPE(a)
Definition: array.h:277
HashValueFunc hash
Definition: hsearch.h:74
#define HASH_FUNCTION
Definition: hsearch.h:89
#define DatumGetArrayTypeP(X)
Definition: array.h:246
static int trackitem_compare_frequencies_desc(const void *e1, const void *e2)