PostgreSQL Source Code  git master
ginget.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * ginget.c
4  * fetch tuples from a GIN scan.
5  *
6  *
7  * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
8  * Portions Copyright (c) 1994, Regents of the University of California
9  *
10  * IDENTIFICATION
11  * src/backend/access/gin/ginget.c
12  *-------------------------------------------------------------------------
13  */
14 
15 #include "postgres.h"
16 
17 #include "access/gin_private.h"
18 #include "access/relscan.h"
19 #include "common/pg_prng.h"
20 #include "miscadmin.h"
21 #include "storage/predicate.h"
22 #include "utils/datum.h"
23 #include "utils/memutils.h"
24 #include "utils/rel.h"
25 
26 /* GUC parameter */
28 
29 typedef struct pendingPosition
30 {
35  bool *hasMatchKey;
37 
38 
39 /*
40  * Goes to the next page if current offset is outside of bounds
41  */
42 static bool
44 {
45  Page page = BufferGetPage(stack->buffer);
46 
47  if (stack->off > PageGetMaxOffsetNumber(page))
48  {
49  /*
50  * We scanned the whole page, so we should take right page
51  */
52  if (GinPageRightMost(page))
53  return false; /* no more pages */
54 
55  stack->buffer = ginStepRight(stack->buffer, btree->index, GIN_SHARE);
56  stack->blkno = BufferGetBlockNumber(stack->buffer);
57  stack->off = FirstOffsetNumber;
58  PredicateLockPage(btree->index, stack->blkno, snapshot);
59  }
60 
61  return true;
62 }
63 
64 /*
65  * Scan all pages of a posting tree and save all its heap ItemPointers
66  * in scanEntry->matchBitmap
67  */
68 static void
70  BlockNumber rootPostingTree)
71 {
72  GinBtreeData btree;
73  GinBtreeStack *stack;
74  Buffer buffer;
75  Page page;
76 
77  /* Descend to the leftmost leaf page */
78  stack = ginScanBeginPostingTree(&btree, index, rootPostingTree);
79  buffer = stack->buffer;
80 
81  IncrBufferRefCount(buffer); /* prevent unpin in freeGinBtreeStack */
82 
83  freeGinBtreeStack(stack);
84 
85  /*
86  * Loop iterates through all leaf pages of posting tree
87  */
88  for (;;)
89  {
90  page = BufferGetPage(buffer);
91  if ((GinPageGetOpaque(page)->flags & GIN_DELETED) == 0)
92  {
93  int n = GinDataLeafPageGetItemsToTbm(page, scanEntry->matchBitmap);
94 
95  scanEntry->predictNumberResult += n;
96  }
97 
98  if (GinPageRightMost(page))
99  break; /* no more pages */
100 
101  buffer = ginStepRight(buffer, index, GIN_SHARE);
102  }
103 
104  UnlockReleaseBuffer(buffer);
105 }
106 
107 /*
108  * Collects TIDs into scanEntry->matchBitmap for all heap tuples that
109  * match the search entry. This supports three different match modes:
110  *
111  * 1. Partial-match support: scan from current point until the
112  * comparePartialFn says we're done.
113  * 2. SEARCH_MODE_ALL: scan from current point (which should be first
114  * key for the current attnum) until we hit null items or end of attnum
115  * 3. SEARCH_MODE_EVERYTHING: scan from current point (which should be first
116  * key for the current attnum) until we hit end of attnum
117  *
118  * Returns true if done, false if it's necessary to restart scan from scratch
119  */
120 static bool
122  GinScanEntry scanEntry, Snapshot snapshot)
123 {
125  Form_pg_attribute attr;
126 
127  /* Initialize empty bitmap result */
128  scanEntry->matchBitmap = tbm_create(work_mem * 1024L, NULL);
129 
130  /* Null query cannot partial-match anything */
131  if (scanEntry->isPartialMatch &&
132  scanEntry->queryCategory != GIN_CAT_NORM_KEY)
133  return true;
134 
135  /* Locate tupdesc entry for key column (for attbyval/attlen data) */
136  attnum = scanEntry->attnum;
137  attr = TupleDescAttr(btree->ginstate->origTupdesc, attnum - 1);
138 
139  /*
140  * Predicate lock entry leaf page, following pages will be locked by
141  * moveRightIfItNeeded()
142  */
143  PredicateLockPage(btree->index,
145  snapshot);
146 
147  for (;;)
148  {
149  Page page;
150  IndexTuple itup;
151  Datum idatum;
152  GinNullCategory icategory;
153 
154  /*
155  * stack->off points to the interested entry, buffer is already locked
156  */
157  if (moveRightIfItNeeded(btree, stack, snapshot) == false)
158  return true;
159 
160  page = BufferGetPage(stack->buffer);
161  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, stack->off));
162 
163  /*
164  * If tuple stores another attribute then stop scan
165  */
166  if (gintuple_get_attrnum(btree->ginstate, itup) != attnum)
167  return true;
168 
169  /* Safe to fetch attribute value */
170  idatum = gintuple_get_key(btree->ginstate, itup, &icategory);
171 
172  /*
173  * Check for appropriate scan stop conditions
174  */
175  if (scanEntry->isPartialMatch)
176  {
177  int32 cmp;
178 
179  /*
180  * In partial match, stop scan at any null (including
181  * placeholders); partial matches never match nulls
182  */
183  if (icategory != GIN_CAT_NORM_KEY)
184  return true;
185 
186  /*----------
187  * Check of partial match.
188  * case cmp == 0 => match
189  * case cmp > 0 => not match and finish scan
190  * case cmp < 0 => not match and continue scan
191  *----------
192  */
194  btree->ginstate->supportCollation[attnum - 1],
195  scanEntry->queryKey,
196  idatum,
197  UInt16GetDatum(scanEntry->strategy),
198  PointerGetDatum(scanEntry->extra_data)));
199 
200  if (cmp > 0)
201  return true;
202  else if (cmp < 0)
203  {
204  stack->off++;
205  continue;
206  }
207  }
208  else if (scanEntry->searchMode == GIN_SEARCH_MODE_ALL)
209  {
210  /*
211  * In ALL mode, we are not interested in null items, so we can
212  * stop if we get to a null-item placeholder (which will be the
213  * last entry for a given attnum). We do want to include NULL_KEY
214  * and EMPTY_ITEM entries, though.
215  */
216  if (icategory == GIN_CAT_NULL_ITEM)
217  return true;
218  }
219 
220  /*
221  * OK, we want to return the TIDs listed in this entry.
222  */
223  if (GinIsPostingTree(itup))
224  {
225  BlockNumber rootPostingTree = GinGetPostingTree(itup);
226 
227  /*
228  * We should unlock current page (but not unpin) during tree scan
229  * to prevent deadlock with vacuum processes.
230  *
231  * We save current entry value (idatum) to be able to re-find our
232  * tuple after re-locking
233  */
234  if (icategory == GIN_CAT_NORM_KEY)
235  idatum = datumCopy(idatum, attr->attbyval, attr->attlen);
236 
237  LockBuffer(stack->buffer, GIN_UNLOCK);
238 
239  /*
240  * Acquire predicate lock on the posting tree. We already hold a
241  * lock on the entry page, but insertions to the posting tree
242  * don't check for conflicts on that level.
243  */
244  PredicateLockPage(btree->index, rootPostingTree, snapshot);
245 
246  /* Collect all the TIDs in this entry's posting tree */
247  scanPostingTree(btree->index, scanEntry, rootPostingTree);
248 
249  /*
250  * We lock again the entry page and while it was unlocked insert
251  * might have occurred, so we need to re-find our position.
252  */
253  LockBuffer(stack->buffer, GIN_SHARE);
254  page = BufferGetPage(stack->buffer);
255  if (!GinPageIsLeaf(page))
256  {
257  /*
258  * Root page becomes non-leaf while we unlock it. We will
259  * start again, this situation doesn't occur often - root can
260  * became a non-leaf only once per life of index.
261  */
262  return false;
263  }
264 
265  /* Search forward to re-find idatum */
266  for (;;)
267  {
268  if (moveRightIfItNeeded(btree, stack, snapshot) == false)
269  ereport(ERROR,
270  (errcode(ERRCODE_INTERNAL_ERROR),
271  errmsg("failed to re-find tuple within index \"%s\"",
272  RelationGetRelationName(btree->index))));
273 
274  page = BufferGetPage(stack->buffer);
275  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, stack->off));
276 
277  if (gintuple_get_attrnum(btree->ginstate, itup) == attnum)
278  {
279  Datum newDatum;
280  GinNullCategory newCategory;
281 
282  newDatum = gintuple_get_key(btree->ginstate, itup,
283  &newCategory);
284 
285  if (ginCompareEntries(btree->ginstate, attnum,
286  newDatum, newCategory,
287  idatum, icategory) == 0)
288  break; /* Found! */
289  }
290 
291  stack->off++;
292  }
293 
294  if (icategory == GIN_CAT_NORM_KEY && !attr->attbyval)
295  pfree(DatumGetPointer(idatum));
296  }
297  else
298  {
299  ItemPointer ipd;
300  int nipd;
301 
302  ipd = ginReadTuple(btree->ginstate, scanEntry->attnum, itup, &nipd);
303  tbm_add_tuples(scanEntry->matchBitmap, ipd, nipd, false);
304  scanEntry->predictNumberResult += GinGetNPosting(itup);
305  pfree(ipd);
306  }
307 
308  /*
309  * Done with this entry, go to the next
310  */
311  stack->off++;
312  }
313 }
314 
315 /*
316  * Start* functions setup beginning state of searches: finds correct buffer and pins it.
317  */
318 static void
319 startScanEntry(GinState *ginstate, GinScanEntry entry, Snapshot snapshot)
320 {
321  GinBtreeData btreeEntry;
322  GinBtreeStack *stackEntry;
323  Page page;
324  bool needUnlock;
325 
326 restartScanEntry:
327  entry->buffer = InvalidBuffer;
328  ItemPointerSetMin(&entry->curItem);
329  entry->offset = InvalidOffsetNumber;
330  if (entry->list)
331  pfree(entry->list);
332  entry->list = NULL;
333  entry->nlist = 0;
334  entry->matchBitmap = NULL;
335  entry->matchResult = NULL;
336  entry->reduceResult = false;
337  entry->predictNumberResult = 0;
338 
339  /*
340  * we should find entry, and begin scan of posting tree or just store
341  * posting list in memory
342  */
343  ginPrepareEntryScan(&btreeEntry, entry->attnum,
344  entry->queryKey, entry->queryCategory,
345  ginstate);
346  stackEntry = ginFindLeafPage(&btreeEntry, true, false);
347  page = BufferGetPage(stackEntry->buffer);
348 
349  /* ginFindLeafPage() will have already checked snapshot age. */
350  needUnlock = true;
351 
352  entry->isFinished = true;
353 
354  if (entry->isPartialMatch ||
356  {
357  /*
358  * btreeEntry.findItem locates the first item >= given search key.
359  * (For GIN_CAT_EMPTY_QUERY, it will find the leftmost index item
360  * because of the way the GIN_CAT_EMPTY_QUERY category code is
361  * assigned.) We scan forward from there and collect all TIDs needed
362  * for the entry type.
363  */
364  btreeEntry.findItem(&btreeEntry, stackEntry);
365  if (collectMatchBitmap(&btreeEntry, stackEntry, entry, snapshot)
366  == false)
367  {
368  /*
369  * GIN tree was seriously restructured, so we will cleanup all
370  * found data and rescan. See comments near 'return false' in
371  * collectMatchBitmap()
372  */
373  if (entry->matchBitmap)
374  {
375  if (entry->matchIterator)
377  entry->matchIterator = NULL;
378  tbm_free(entry->matchBitmap);
379  entry->matchBitmap = NULL;
380  }
381  LockBuffer(stackEntry->buffer, GIN_UNLOCK);
382  freeGinBtreeStack(stackEntry);
383  goto restartScanEntry;
384  }
385 
386  if (entry->matchBitmap && !tbm_is_empty(entry->matchBitmap))
387  {
389  entry->isFinished = false;
390  }
391  }
392  else if (btreeEntry.findItem(&btreeEntry, stackEntry))
393  {
394  IndexTuple itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, stackEntry->off));
395 
396  if (GinIsPostingTree(itup))
397  {
398  BlockNumber rootPostingTree = GinGetPostingTree(itup);
399  GinBtreeStack *stack;
400  Page entrypage;
401  ItemPointerData minItem;
402 
403  /*
404  * This is an equality scan, so lock the root of the posting tree.
405  * It represents a lock on the exact key value, and covers all the
406  * items in the posting tree.
407  */
408  PredicateLockPage(ginstate->index, rootPostingTree, snapshot);
409 
410  /*
411  * We should unlock entry page before touching posting tree to
412  * prevent deadlocks with vacuum processes. Because entry is never
413  * deleted from page and posting tree is never reduced to the
414  * posting list, we can unlock page after getting BlockNumber of
415  * root of posting tree.
416  */
417  LockBuffer(stackEntry->buffer, GIN_UNLOCK);
418  needUnlock = false;
419 
420  stack = ginScanBeginPostingTree(&entry->btree, ginstate->index,
421  rootPostingTree);
422  entry->buffer = stack->buffer;
423 
424  /*
425  * We keep buffer pinned because we need to prevent deletion of
426  * page during scan. See GIN's vacuum implementation. RefCount is
427  * increased to keep buffer pinned after freeGinBtreeStack() call.
428  */
429  IncrBufferRefCount(entry->buffer);
430 
431  entrypage = BufferGetPage(entry->buffer);
432 
433  /*
434  * Load the first page into memory.
435  */
436  ItemPointerSetMin(&minItem);
437  entry->list = GinDataLeafPageGetItems(entrypage, &entry->nlist, minItem);
438 
439  entry->predictNumberResult = stack->predictNumber * entry->nlist;
440 
441  LockBuffer(entry->buffer, GIN_UNLOCK);
442  freeGinBtreeStack(stack);
443  entry->isFinished = false;
444  }
445  else
446  {
447  /*
448  * Lock the entry leaf page. This is more coarse-grained than
449  * necessary, because it will conflict with any insertions that
450  * land on the same leaf page, not only the exact key we searched
451  * for. But locking an individual tuple would require updating
452  * that lock whenever it moves because of insertions or vacuums,
453  * which seems too complicated.
454  */
455  PredicateLockPage(ginstate->index,
456  BufferGetBlockNumber(stackEntry->buffer),
457  snapshot);
458  if (GinGetNPosting(itup) > 0)
459  {
460  entry->list = ginReadTuple(ginstate, entry->attnum, itup,
461  &entry->nlist);
462  entry->predictNumberResult = entry->nlist;
463 
464  entry->isFinished = false;
465  }
466  }
467  }
468  else
469  {
470  /*
471  * No entry found. Predicate lock the leaf page, to lock the place
472  * where the entry would've been, had there been one.
473  */
474  PredicateLockPage(ginstate->index,
475  BufferGetBlockNumber(stackEntry->buffer), snapshot);
476  }
477 
478  if (needUnlock)
479  LockBuffer(stackEntry->buffer, GIN_UNLOCK);
480  freeGinBtreeStack(stackEntry);
481 }
482 
483 /*
484  * Comparison function for scan entry indexes. Sorts by predictNumberResult,
485  * least frequent items first.
486  */
487 static int
488 entryIndexByFrequencyCmp(const void *a1, const void *a2, void *arg)
489 {
490  const GinScanKey key = (const GinScanKey) arg;
491  int i1 = *(const int *) a1;
492  int i2 = *(const int *) a2;
493  uint32 n1 = key->scanEntry[i1]->predictNumberResult;
494  uint32 n2 = key->scanEntry[i2]->predictNumberResult;
495 
496  if (n1 < n2)
497  return -1;
498  else if (n1 == n2)
499  return 0;
500  else
501  return 1;
502 }
503 
504 static void
506 {
508  int i;
509  int j;
510  int *entryIndexes;
511 
512  ItemPointerSetMin(&key->curItem);
513  key->curItemMatches = false;
514  key->recheckCurItem = false;
515  key->isFinished = false;
516 
517  /*
518  * Divide the entries into two distinct sets: required and additional.
519  * Additional entries are not enough for a match alone, without any items
520  * from the required set, but are needed by the consistent function to
521  * decide if an item matches. When scanning, we can skip over items from
522  * additional entries that have no corresponding matches in any of the
523  * required entries. That speeds up queries like "frequent & rare"
524  * considerably, if the frequent term can be put in the additional set.
525  *
526  * There can be many legal ways to divide them entries into these two
527  * sets. A conservative division is to just put everything in the required
528  * set, but the more you can put in the additional set, the more you can
529  * skip during the scan. To maximize skipping, we try to put as many
530  * frequent items as possible into additional, and less frequent ones into
531  * required. To do that, sort the entries by frequency
532  * (predictNumberResult), and put entries into the required set in that
533  * order, until the consistent function says that none of the remaining
534  * entries can form a match, without any items from the required set. The
535  * rest go to the additional set.
536  *
537  * Exclude-only scan keys are known to have no required entries.
538  */
539  if (key->excludeOnly)
540  {
542 
543  key->nrequired = 0;
544  key->nadditional = key->nentries;
545  key->additionalEntries = palloc(key->nadditional * sizeof(GinScanEntry));
546  for (i = 0; i < key->nadditional; i++)
547  key->additionalEntries[i] = key->scanEntry[i];
548  }
549  else if (key->nentries > 1)
550  {
552 
553  entryIndexes = (int *) palloc(sizeof(int) * key->nentries);
554  for (i = 0; i < key->nentries; i++)
555  entryIndexes[i] = i;
556  qsort_arg(entryIndexes, key->nentries, sizeof(int),
558 
559  for (i = 0; i < key->nentries - 1; i++)
560  {
561  /* Pass all entries <= i as FALSE, and the rest as MAYBE */
562  for (j = 0; j <= i; j++)
563  key->entryRes[entryIndexes[j]] = GIN_FALSE;
564  for (j = i + 1; j < key->nentries; j++)
565  key->entryRes[entryIndexes[j]] = GIN_MAYBE;
566 
567  if (key->triConsistentFn(key) == GIN_FALSE)
568  break;
569  }
570  /* i is now the last required entry. */
571 
573 
574  key->nrequired = i + 1;
575  key->nadditional = key->nentries - key->nrequired;
576  key->requiredEntries = palloc(key->nrequired * sizeof(GinScanEntry));
577  key->additionalEntries = palloc(key->nadditional * sizeof(GinScanEntry));
578 
579  j = 0;
580  for (i = 0; i < key->nrequired; i++)
581  key->requiredEntries[i] = key->scanEntry[entryIndexes[j++]];
582  for (i = 0; i < key->nadditional; i++)
583  key->additionalEntries[i] = key->scanEntry[entryIndexes[j++]];
584 
585  /* clean up after consistentFn calls (also frees entryIndexes) */
587  }
588  else
589  {
591 
592  key->nrequired = 1;
593  key->nadditional = 0;
594  key->requiredEntries = palloc(1 * sizeof(GinScanEntry));
595  key->requiredEntries[0] = key->scanEntry[0];
596  }
597  MemoryContextSwitchTo(oldCtx);
598 }
599 
600 static void
602 {
603  GinScanOpaque so = (GinScanOpaque) scan->opaque;
604  GinState *ginstate = &so->ginstate;
605  uint32 i;
606 
607  for (i = 0; i < so->totalentries; i++)
608  startScanEntry(ginstate, so->entries[i], scan->xs_snapshot);
609 
610  if (GinFuzzySearchLimit > 0)
611  {
612  /*
613  * If all of keys more than threshold we will try to reduce result, we
614  * hope (and only hope, for intersection operation of array our
615  * supposition isn't true), that total result will not more than
616  * minimal predictNumberResult.
617  */
618  bool reduce = true;
619 
620  for (i = 0; i < so->totalentries; i++)
621  {
623  {
624  reduce = false;
625  break;
626  }
627  }
628  if (reduce)
629  {
630  for (i = 0; i < so->totalentries; i++)
631  {
633  so->entries[i]->reduceResult = true;
634  }
635  }
636  }
637 
638  /*
639  * Now that we have the estimates for the entry frequencies, finish
640  * initializing the scan keys.
641  */
642  for (i = 0; i < so->nkeys; i++)
643  startScanKey(ginstate, so, so->keys + i);
644 }
645 
646 /*
647  * Load the next batch of item pointers from a posting tree.
648  *
649  * Note that we copy the page into GinScanEntry->list array and unlock it, but
650  * keep it pinned to prevent interference with vacuum.
651  */
652 static void
654  ItemPointerData advancePast)
655 {
656  Page page;
657  int i;
658  bool stepright;
659 
660  if (!BufferIsValid(entry->buffer))
661  {
662  entry->isFinished = true;
663  return;
664  }
665 
666  /*
667  * We have two strategies for finding the correct page: step right from
668  * the current page, or descend the tree again from the root. If
669  * advancePast equals the current item, the next matching item should be
670  * on the next page, so we step right. Otherwise, descend from root.
671  */
672  if (ginCompareItemPointers(&entry->curItem, &advancePast) == 0)
673  {
674  stepright = true;
675  LockBuffer(entry->buffer, GIN_SHARE);
676  }
677  else
678  {
679  GinBtreeStack *stack;
680 
681  ReleaseBuffer(entry->buffer);
682 
683  /*
684  * Set the search key, and find the correct leaf page.
685  */
686  if (ItemPointerIsLossyPage(&advancePast))
687  {
688  ItemPointerSet(&entry->btree.itemptr,
689  GinItemPointerGetBlockNumber(&advancePast) + 1,
691  }
692  else
693  {
694  ItemPointerSet(&entry->btree.itemptr,
695  GinItemPointerGetBlockNumber(&advancePast),
697  }
698  entry->btree.fullScan = false;
699  stack = ginFindLeafPage(&entry->btree, true, false);
700 
701  /* we don't need the stack, just the buffer. */
702  entry->buffer = stack->buffer;
703  IncrBufferRefCount(entry->buffer);
704  freeGinBtreeStack(stack);
705  stepright = false;
706  }
707 
708  elog(DEBUG2, "entryLoadMoreItems, %u/%u, skip: %d",
709  GinItemPointerGetBlockNumber(&advancePast),
710  GinItemPointerGetOffsetNumber(&advancePast),
711  !stepright);
712 
713  page = BufferGetPage(entry->buffer);
714  for (;;)
715  {
716  entry->offset = InvalidOffsetNumber;
717  if (entry->list)
718  {
719  pfree(entry->list);
720  entry->list = NULL;
721  entry->nlist = 0;
722  }
723 
724  if (stepright)
725  {
726  /*
727  * We've processed all the entries on this page. If it was the
728  * last page in the tree, we're done.
729  */
730  if (GinPageRightMost(page))
731  {
732  UnlockReleaseBuffer(entry->buffer);
733  entry->buffer = InvalidBuffer;
734  entry->isFinished = true;
735  return;
736  }
737 
738  /*
739  * Step to next page, following the right link. then find the
740  * first ItemPointer greater than advancePast.
741  */
742  entry->buffer = ginStepRight(entry->buffer,
743  ginstate->index,
744  GIN_SHARE);
745  page = BufferGetPage(entry->buffer);
746  }
747  stepright = true;
748 
749  if (GinPageGetOpaque(page)->flags & GIN_DELETED)
750  continue; /* page was deleted by concurrent vacuum */
751 
752  /*
753  * The first item > advancePast might not be on this page, but
754  * somewhere to the right, if the page was split, or a non-match from
755  * another key in the query allowed us to skip some items from this
756  * entry. Keep following the right-links until we re-find the correct
757  * page.
758  */
759  if (!GinPageRightMost(page) &&
760  ginCompareItemPointers(&advancePast, GinDataPageGetRightBound(page)) >= 0)
761  {
762  /*
763  * the item we're looking is > the right bound of the page, so it
764  * can't be on this page.
765  */
766  continue;
767  }
768 
769  entry->list = GinDataLeafPageGetItems(page, &entry->nlist, advancePast);
770 
771  for (i = 0; i < entry->nlist; i++)
772  {
773  if (ginCompareItemPointers(&advancePast, &entry->list[i]) < 0)
774  {
775  entry->offset = i;
776 
777  if (GinPageRightMost(page))
778  {
779  /* after processing the copied items, we're done. */
780  UnlockReleaseBuffer(entry->buffer);
781  entry->buffer = InvalidBuffer;
782  }
783  else
784  LockBuffer(entry->buffer, GIN_UNLOCK);
785  return;
786  }
787  }
788  }
789 }
790 
791 #define gin_rand() pg_prng_double(&pg_global_prng_state)
792 #define dropItem(e) ( gin_rand() > ((double)GinFuzzySearchLimit)/((double)((e)->predictNumberResult)) )
793 
794 /*
795  * Sets entry->curItem to next heap item pointer > advancePast, for one entry
796  * of one scan key, or sets entry->isFinished to true if there are no more.
797  *
798  * Item pointers are returned in ascending order.
799  *
800  * Note: this can return a "lossy page" item pointer, indicating that the
801  * entry potentially matches all items on that heap page. However, it is
802  * not allowed to return both a lossy page pointer and exact (regular)
803  * item pointers for the same page. (Doing so would break the key-combination
804  * logic in keyGetItem and scanGetItem; see comment in scanGetItem.) In the
805  * current implementation this is guaranteed by the behavior of tidbitmaps.
806  */
807 static void
809  ItemPointerData advancePast)
810 {
811  Assert(!entry->isFinished);
812 
813  Assert(!ItemPointerIsValid(&entry->curItem) ||
814  ginCompareItemPointers(&entry->curItem, &advancePast) <= 0);
815 
816  if (entry->matchBitmap)
817  {
818  /* A bitmap result */
819  BlockNumber advancePastBlk = GinItemPointerGetBlockNumber(&advancePast);
820  OffsetNumber advancePastOff = GinItemPointerGetOffsetNumber(&advancePast);
821 
822  for (;;)
823  {
824  /*
825  * If we've exhausted all items on this block, move to next block
826  * in the bitmap.
827  */
828  while (entry->matchResult == NULL ||
829  (entry->matchResult->ntuples >= 0 &&
830  entry->offset >= entry->matchResult->ntuples) ||
831  entry->matchResult->blockno < advancePastBlk ||
832  (ItemPointerIsLossyPage(&advancePast) &&
833  entry->matchResult->blockno == advancePastBlk))
834  {
835  entry->matchResult = tbm_iterate(entry->matchIterator);
836 
837  if (entry->matchResult == NULL)
838  {
841  entry->matchIterator = NULL;
842  entry->isFinished = true;
843  break;
844  }
845 
846  /*
847  * Reset counter to the beginning of entry->matchResult. Note:
848  * entry->offset is still greater than matchResult->ntuples if
849  * matchResult is lossy. So, on next call we will get next
850  * result from TIDBitmap.
851  */
852  entry->offset = 0;
853  }
854  if (entry->isFinished)
855  break;
856 
857  /*
858  * We're now on the first page after advancePast which has any
859  * items on it. If it's a lossy result, return that.
860  */
861  if (entry->matchResult->ntuples < 0)
862  {
864  entry->matchResult->blockno);
865 
866  /*
867  * We might as well fall out of the loop; we could not
868  * estimate number of results on this page to support correct
869  * reducing of result even if it's enabled.
870  */
871  break;
872  }
873 
874  /*
875  * Not a lossy page. Skip over any offsets <= advancePast, and
876  * return that.
877  */
878  if (entry->matchResult->blockno == advancePastBlk)
879  {
880  /*
881  * First, do a quick check against the last offset on the
882  * page. If that's > advancePast, so are all the other
883  * offsets, so just go back to the top to get the next page.
884  */
885  if (entry->matchResult->offsets[entry->matchResult->ntuples - 1] <= advancePastOff)
886  {
887  entry->offset = entry->matchResult->ntuples;
888  continue;
889  }
890 
891  /* Otherwise scan to find the first item > advancePast */
892  while (entry->matchResult->offsets[entry->offset] <= advancePastOff)
893  entry->offset++;
894  }
895 
896  ItemPointerSet(&entry->curItem,
897  entry->matchResult->blockno,
898  entry->matchResult->offsets[entry->offset]);
899  entry->offset++;
900 
901  /* Done unless we need to reduce the result */
902  if (!entry->reduceResult || !dropItem(entry))
903  break;
904  }
905  }
906  else if (!BufferIsValid(entry->buffer))
907  {
908  /*
909  * A posting list from an entry tuple, or the last page of a posting
910  * tree.
911  */
912  for (;;)
913  {
914  if (entry->offset >= entry->nlist)
915  {
917  entry->isFinished = true;
918  break;
919  }
920 
921  entry->curItem = entry->list[entry->offset++];
922 
923  /* If we're not past advancePast, keep scanning */
924  if (ginCompareItemPointers(&entry->curItem, &advancePast) <= 0)
925  continue;
926 
927  /* Done unless we need to reduce the result */
928  if (!entry->reduceResult || !dropItem(entry))
929  break;
930  }
931  }
932  else
933  {
934  /* A posting tree */
935  for (;;)
936  {
937  /* If we've processed the current batch, load more items */
938  while (entry->offset >= entry->nlist)
939  {
940  entryLoadMoreItems(ginstate, entry, advancePast);
941 
942  if (entry->isFinished)
943  {
945  return;
946  }
947  }
948 
949  entry->curItem = entry->list[entry->offset++];
950 
951  /* If we're not past advancePast, keep scanning */
952  if (ginCompareItemPointers(&entry->curItem, &advancePast) <= 0)
953  continue;
954 
955  /* Done unless we need to reduce the result */
956  if (!entry->reduceResult || !dropItem(entry))
957  break;
958 
959  /*
960  * Advance advancePast (so that entryLoadMoreItems will load the
961  * right data), and keep scanning
962  */
963  advancePast = entry->curItem;
964  }
965  }
966 }
967 
968 /*
969  * Identify the "current" item among the input entry streams for this scan key
970  * that is greater than advancePast, and test whether it passes the scan key
971  * qual condition.
972  *
973  * The current item is the smallest curItem among the inputs. key->curItem
974  * is set to that value. key->curItemMatches is set to indicate whether that
975  * TID passes the consistentFn test. If so, key->recheckCurItem is set true
976  * iff recheck is needed for this item pointer (including the case where the
977  * item pointer is a lossy page pointer).
978  *
979  * If all entry streams are exhausted, sets key->isFinished to true.
980  *
981  * Item pointers must be returned in ascending order.
982  *
983  * Note: this can return a "lossy page" item pointer, indicating that the
984  * key potentially matches all items on that heap page. However, it is
985  * not allowed to return both a lossy page pointer and exact (regular)
986  * item pointers for the same page. (Doing so would break the key-combination
987  * logic in scanGetItem.)
988  */
989 static void
991  ItemPointerData advancePast)
992 {
993  ItemPointerData minItem;
994  ItemPointerData curPageLossy;
995  uint32 i;
996  bool haveLossyEntry;
997  GinScanEntry entry;
999  MemoryContext oldCtx;
1000  bool allFinished;
1001 
1002  Assert(!key->isFinished);
1003 
1004  /*
1005  * We might have already tested this item; if so, no need to repeat work.
1006  * (Note: the ">" case can happen, if advancePast is exact but we
1007  * previously had to set curItem to a lossy-page pointer.)
1008  */
1009  if (ginCompareItemPointers(&key->curItem, &advancePast) > 0)
1010  return;
1011 
1012  /*
1013  * Find the minimum item > advancePast among the active entry streams.
1014  *
1015  * Note: a lossy-page entry is encoded by a ItemPointer with max value for
1016  * offset (0xffff), so that it will sort after any exact entries for the
1017  * same page. So we'll prefer to return exact pointers not lossy
1018  * pointers, which is good.
1019  */
1020  ItemPointerSetMax(&minItem);
1021  allFinished = true;
1022  for (i = 0; i < key->nrequired; i++)
1023  {
1024  entry = key->requiredEntries[i];
1025 
1026  if (entry->isFinished)
1027  continue;
1028 
1029  /*
1030  * Advance this stream if necessary.
1031  *
1032  * In particular, since entry->curItem was initialized with
1033  * ItemPointerSetMin, this ensures we fetch the first item for each
1034  * entry on the first call.
1035  */
1036  if (ginCompareItemPointers(&entry->curItem, &advancePast) <= 0)
1037  {
1038  entryGetItem(ginstate, entry, advancePast);
1039  if (entry->isFinished)
1040  continue;
1041  }
1042 
1043  allFinished = false;
1044  if (ginCompareItemPointers(&entry->curItem, &minItem) < 0)
1045  minItem = entry->curItem;
1046  }
1047 
1048  if (allFinished && !key->excludeOnly)
1049  {
1050  /* all entries are finished */
1051  key->isFinished = true;
1052  return;
1053  }
1054 
1055  if (!key->excludeOnly)
1056  {
1057  /*
1058  * For a normal scan key, we now know there are no matches < minItem.
1059  *
1060  * If minItem is lossy, it means that there were no exact items on the
1061  * page among requiredEntries, because lossy pointers sort after exact
1062  * items. However, there might be exact items for the same page among
1063  * additionalEntries, so we mustn't advance past them.
1064  */
1065  if (ItemPointerIsLossyPage(&minItem))
1066  {
1067  if (GinItemPointerGetBlockNumber(&advancePast) <
1068  GinItemPointerGetBlockNumber(&minItem))
1069  {
1070  ItemPointerSet(&advancePast,
1071  GinItemPointerGetBlockNumber(&minItem),
1073  }
1074  }
1075  else
1076  {
1077  Assert(GinItemPointerGetOffsetNumber(&minItem) > 0);
1078  ItemPointerSet(&advancePast,
1079  GinItemPointerGetBlockNumber(&minItem),
1081  }
1082  }
1083  else
1084  {
1085  /*
1086  * excludeOnly scan keys don't have any entries that are necessarily
1087  * present in matching items. So, we consider the item just after
1088  * advancePast.
1089  */
1090  Assert(key->nrequired == 0);
1091  ItemPointerSet(&minItem,
1092  GinItemPointerGetBlockNumber(&advancePast),
1094  }
1095 
1096  /*
1097  * We might not have loaded all the entry streams for this TID yet. We
1098  * could call the consistent function, passing MAYBE for those entries, to
1099  * see if it can decide if this TID matches based on the information we
1100  * have. But if the consistent-function is expensive, and cannot in fact
1101  * decide with partial information, that could be a big loss. So, load all
1102  * the additional entries, before calling the consistent function.
1103  */
1104  for (i = 0; i < key->nadditional; i++)
1105  {
1106  entry = key->additionalEntries[i];
1107 
1108  if (entry->isFinished)
1109  continue;
1110 
1111  if (ginCompareItemPointers(&entry->curItem, &advancePast) <= 0)
1112  {
1113  entryGetItem(ginstate, entry, advancePast);
1114  if (entry->isFinished)
1115  continue;
1116  }
1117 
1118  /*
1119  * Normally, none of the items in additionalEntries can have a curItem
1120  * larger than minItem. But if minItem is a lossy page, then there
1121  * might be exact items on the same page among additionalEntries.
1122  */
1123  if (ginCompareItemPointers(&entry->curItem, &minItem) < 0)
1124  {
1125  Assert(ItemPointerIsLossyPage(&minItem));
1126  minItem = entry->curItem;
1127  }
1128  }
1129 
1130  /*
1131  * Ok, we've advanced all the entries up to minItem now. Set key->curItem,
1132  * and perform consistentFn test.
1133  *
1134  * Lossy-page entries pose a problem, since we don't know the correct
1135  * entryRes state to pass to the consistentFn, and we also don't know what
1136  * its combining logic will be (could be AND, OR, or even NOT). If the
1137  * logic is OR then the consistentFn might succeed for all items in the
1138  * lossy page even when none of the other entries match.
1139  *
1140  * Our strategy is to call the tri-state consistent function, with the
1141  * lossy-page entries set to MAYBE, and all the other entries FALSE. If it
1142  * returns FALSE, none of the lossy items alone are enough for a match, so
1143  * we don't need to return a lossy-page pointer. Otherwise, return a
1144  * lossy-page pointer to indicate that the whole heap page must be
1145  * checked. (On subsequent calls, we'll do nothing until minItem is past
1146  * the page altogether, thus ensuring that we never return both regular
1147  * and lossy pointers for the same page.)
1148  *
1149  * An exception is that it doesn't matter what we pass for lossy pointers
1150  * in "hidden" entries, because the consistentFn's result can't depend on
1151  * them. We could pass them as MAYBE as well, but if we're using the
1152  * "shim" implementation of a tri-state consistent function (see
1153  * ginlogic.c), it's better to pass as few MAYBEs as possible. So pass
1154  * them as true.
1155  *
1156  * Note that only lossy-page entries pointing to the current item's page
1157  * should trigger this processing; we might have future lossy pages in the
1158  * entry array, but they aren't relevant yet.
1159  */
1160  key->curItem = minItem;
1161  ItemPointerSetLossyPage(&curPageLossy,
1162  GinItemPointerGetBlockNumber(&key->curItem));
1163  haveLossyEntry = false;
1164  for (i = 0; i < key->nentries; i++)
1165  {
1166  entry = key->scanEntry[i];
1167  if (entry->isFinished == false &&
1168  ginCompareItemPointers(&entry->curItem, &curPageLossy) == 0)
1169  {
1170  if (i < key->nuserentries)
1171  key->entryRes[i] = GIN_MAYBE;
1172  else
1173  key->entryRes[i] = GIN_TRUE;
1174  haveLossyEntry = true;
1175  }
1176  else
1177  key->entryRes[i] = GIN_FALSE;
1178  }
1179 
1180  /* prepare for calling consistentFn in temp context */
1181  oldCtx = MemoryContextSwitchTo(tempCtx);
1182 
1183  if (haveLossyEntry)
1184  {
1185  /* Have lossy-page entries, so see if whole page matches */
1186  res = key->triConsistentFn(key);
1187 
1188  if (res == GIN_TRUE || res == GIN_MAYBE)
1189  {
1190  /* Yes, so clean up ... */
1191  MemoryContextSwitchTo(oldCtx);
1192  MemoryContextReset(tempCtx);
1193 
1194  /* and return lossy pointer for whole page */
1195  key->curItem = curPageLossy;
1196  key->curItemMatches = true;
1197  key->recheckCurItem = true;
1198  return;
1199  }
1200  }
1201 
1202  /*
1203  * At this point we know that we don't need to return a lossy whole-page
1204  * pointer, but we might have matches for individual exact item pointers,
1205  * possibly in combination with a lossy pointer. Pass lossy pointers as
1206  * MAYBE to the ternary consistent function, to let it decide if this
1207  * tuple satisfies the overall key, even though we don't know if the lossy
1208  * entries match.
1209  *
1210  * Prepare entryRes array to be passed to consistentFn.
1211  */
1212  for (i = 0; i < key->nentries; i++)
1213  {
1214  entry = key->scanEntry[i];
1215  if (entry->isFinished)
1216  key->entryRes[i] = GIN_FALSE;
1217 #if 0
1218 
1219  /*
1220  * This case can't currently happen, because we loaded all the entries
1221  * for this item earlier.
1222  */
1223  else if (ginCompareItemPointers(&entry->curItem, &advancePast) <= 0)
1224  key->entryRes[i] = GIN_MAYBE;
1225 #endif
1226  else if (ginCompareItemPointers(&entry->curItem, &curPageLossy) == 0)
1227  key->entryRes[i] = GIN_MAYBE;
1228  else if (ginCompareItemPointers(&entry->curItem, &minItem) == 0)
1229  key->entryRes[i] = GIN_TRUE;
1230  else
1231  key->entryRes[i] = GIN_FALSE;
1232  }
1233 
1234  res = key->triConsistentFn(key);
1235 
1236  switch (res)
1237  {
1238  case GIN_TRUE:
1239  key->curItemMatches = true;
1240  /* triConsistentFn set recheckCurItem */
1241  break;
1242 
1243  case GIN_FALSE:
1244  key->curItemMatches = false;
1245  break;
1246 
1247  case GIN_MAYBE:
1248  key->curItemMatches = true;
1249  key->recheckCurItem = true;
1250  break;
1251 
1252  default:
1253 
1254  /*
1255  * the 'default' case shouldn't happen, but if the consistent
1256  * function returns something bogus, this is the safe result
1257  */
1258  key->curItemMatches = true;
1259  key->recheckCurItem = true;
1260  break;
1261  }
1262 
1263  /*
1264  * We have a tuple, and we know if it matches or not. If it's a non-match,
1265  * we could continue to find the next matching tuple, but let's break out
1266  * and give scanGetItem a chance to advance the other keys. They might be
1267  * able to skip past to a much higher TID, allowing us to save work.
1268  */
1269 
1270  /* clean up after consistentFn calls */
1271  MemoryContextSwitchTo(oldCtx);
1272  MemoryContextReset(tempCtx);
1273 }
1274 
1275 /*
1276  * Get next heap item pointer (after advancePast) from scan.
1277  * Returns true if anything found.
1278  * On success, *item and *recheck are set.
1279  *
1280  * Note: this is very nearly the same logic as in keyGetItem(), except
1281  * that we know the keys are to be combined with AND logic, whereas in
1282  * keyGetItem() the combination logic is known only to the consistentFn.
1283  */
1284 static bool
1286  ItemPointerData *item, bool *recheck)
1287 {
1288  GinScanOpaque so = (GinScanOpaque) scan->opaque;
1289  uint32 i;
1290  bool match;
1291 
1292  /*----------
1293  * Advance the scan keys in lock-step, until we find an item that matches
1294  * all the keys. If any key reports isFinished, meaning its subset of the
1295  * entries is exhausted, we can stop. Otherwise, set *item to the next
1296  * matching item.
1297  *
1298  * This logic works only if a keyGetItem stream can never contain both
1299  * exact and lossy pointers for the same page. Else we could have a
1300  * case like
1301  *
1302  * stream 1 stream 2
1303  * ... ...
1304  * 42/6 42/7
1305  * 50/1 42/0xffff
1306  * ... ...
1307  *
1308  * We would conclude that 42/6 is not a match and advance stream 1,
1309  * thus never detecting the match to the lossy pointer in stream 2.
1310  * (keyGetItem has a similar problem versus entryGetItem.)
1311  *----------
1312  */
1313  do
1314  {
1315  ItemPointerSetMin(item);
1316  match = true;
1317  for (i = 0; i < so->nkeys && match; i++)
1318  {
1319  GinScanKey key = so->keys + i;
1320 
1321  /*
1322  * If we're considering a lossy page, skip excludeOnly keys. They
1323  * can't exclude the whole page anyway.
1324  */
1325  if (ItemPointerIsLossyPage(item) && key->excludeOnly)
1326  {
1327  /*
1328  * ginNewScanKey() should never mark the first key as
1329  * excludeOnly.
1330  */
1331  Assert(i > 0);
1332  continue;
1333  }
1334 
1335  /* Fetch the next item for this key that is > advancePast. */
1336  keyGetItem(&so->ginstate, so->tempCtx, key, advancePast);
1337 
1338  if (key->isFinished)
1339  return false;
1340 
1341  /*
1342  * If it's not a match, we can immediately conclude that nothing
1343  * <= this item matches, without checking the rest of the keys.
1344  */
1345  if (!key->curItemMatches)
1346  {
1347  advancePast = key->curItem;
1348  match = false;
1349  break;
1350  }
1351 
1352  /*
1353  * It's a match. We can conclude that nothing < matches, so the
1354  * other key streams can skip to this item.
1355  *
1356  * Beware of lossy pointers, though; from a lossy pointer, we can
1357  * only conclude that nothing smaller than this *block* matches.
1358  */
1359  if (ItemPointerIsLossyPage(&key->curItem))
1360  {
1361  if (GinItemPointerGetBlockNumber(&advancePast) <
1362  GinItemPointerGetBlockNumber(&key->curItem))
1363  {
1364  ItemPointerSet(&advancePast,
1365  GinItemPointerGetBlockNumber(&key->curItem),
1367  }
1368  }
1369  else
1370  {
1371  Assert(GinItemPointerGetOffsetNumber(&key->curItem) > 0);
1372  ItemPointerSet(&advancePast,
1373  GinItemPointerGetBlockNumber(&key->curItem),
1375  }
1376 
1377  /*
1378  * If this is the first key, remember this location as a potential
1379  * match, and proceed to check the rest of the keys.
1380  *
1381  * Otherwise, check if this is the same item that we checked the
1382  * previous keys for (or a lossy pointer for the same page). If
1383  * not, loop back to check the previous keys for this item (we
1384  * will check this key again too, but keyGetItem returns quickly
1385  * for that)
1386  */
1387  if (i == 0)
1388  {
1389  *item = key->curItem;
1390  }
1391  else
1392  {
1393  if (ItemPointerIsLossyPage(&key->curItem) ||
1394  ItemPointerIsLossyPage(item))
1395  {
1397  match = (GinItemPointerGetBlockNumber(&key->curItem) ==
1399  }
1400  else
1401  {
1402  Assert(ginCompareItemPointers(&key->curItem, item) >= 0);
1403  match = (ginCompareItemPointers(&key->curItem, item) == 0);
1404  }
1405  }
1406  }
1407  } while (!match);
1408 
1409  Assert(!ItemPointerIsMin(item));
1410 
1411  /*
1412  * Now *item contains the first ItemPointer after previous result that
1413  * satisfied all the keys for that exact TID, or a lossy reference to the
1414  * same page.
1415  *
1416  * We must return recheck = true if any of the keys are marked recheck.
1417  */
1418  *recheck = false;
1419  for (i = 0; i < so->nkeys; i++)
1420  {
1421  GinScanKey key = so->keys + i;
1422 
1423  if (key->recheckCurItem)
1424  {
1425  *recheck = true;
1426  break;
1427  }
1428  }
1429 
1430  return true;
1431 }
1432 
1433 
1434 /*
1435  * Functions for scanning the pending list
1436  */
1437 
1438 
1439 /*
1440  * Get ItemPointer of next heap row to be checked from pending list.
1441  * Returns false if there are no more. On pages with several heap rows
1442  * it returns each row separately, on page with part of heap row returns
1443  * per page data. pos->firstOffset and pos->lastOffset are set to identify
1444  * the range of pending-list tuples belonging to this heap row.
1445  *
1446  * The pendingBuffer is presumed pinned and share-locked on entry, and is
1447  * pinned and share-locked on success exit. On failure exit it's released.
1448  */
1449 static bool
1451 {
1452  OffsetNumber maxoff;
1453  Page page;
1454  IndexTuple itup;
1455 
1456  ItemPointerSetInvalid(&pos->item);
1457  for (;;)
1458  {
1459  page = BufferGetPage(pos->pendingBuffer);
1460 
1461  maxoff = PageGetMaxOffsetNumber(page);
1462  if (pos->firstOffset > maxoff)
1463  {
1464  BlockNumber blkno = GinPageGetOpaque(page)->rightlink;
1465 
1466  if (blkno == InvalidBlockNumber)
1467  {
1470 
1471  return false;
1472  }
1473  else
1474  {
1475  /*
1476  * Here we must prevent deletion of next page by insertcleanup
1477  * process, which may be trying to obtain exclusive lock on
1478  * current page. So, we lock next page before releasing the
1479  * current one
1480  */
1481  Buffer tmpbuf = ReadBuffer(scan->indexRelation, blkno);
1482 
1485 
1486  pos->pendingBuffer = tmpbuf;
1488  }
1489  }
1490  else
1491  {
1492  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, pos->firstOffset));
1493  pos->item = itup->t_tid;
1494  if (GinPageHasFullRow(page))
1495  {
1496  /*
1497  * find itempointer to the next row
1498  */
1499  for (pos->lastOffset = pos->firstOffset + 1; pos->lastOffset <= maxoff; pos->lastOffset++)
1500  {
1501  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, pos->lastOffset));
1502  if (!ItemPointerEquals(&pos->item, &itup->t_tid))
1503  break;
1504  }
1505  }
1506  else
1507  {
1508  /*
1509  * All itempointers are the same on this page
1510  */
1511  pos->lastOffset = maxoff + 1;
1512  }
1513 
1514  /*
1515  * Now pos->firstOffset points to the first tuple of current heap
1516  * row, pos->lastOffset points to the first tuple of next heap row
1517  * (or to the end of page)
1518  */
1519  break;
1520  }
1521  }
1522 
1523  return true;
1524 }
1525 
1526 /*
1527  * Scan pending-list page from current tuple (off) up till the first of:
1528  * - match is found (then returns true)
1529  * - no later match is possible
1530  * - tuple's attribute number is not equal to entry's attrnum
1531  * - reach end of page
1532  *
1533  * datum[]/category[]/datumExtracted[] arrays are used to cache the results
1534  * of gintuple_get_key() on the current page.
1535  */
1536 static bool
1538  OffsetNumber off, OffsetNumber maxoff,
1539  GinScanEntry entry,
1540  Datum *datum, GinNullCategory *category,
1541  bool *datumExtracted)
1542 {
1543  IndexTuple itup;
1544  int32 cmp;
1545 
1546  /* Partial match to a null is not possible */
1547  if (entry->queryCategory != GIN_CAT_NORM_KEY)
1548  return false;
1549 
1550  while (off < maxoff)
1551  {
1552  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, off));
1553 
1554  if (gintuple_get_attrnum(ginstate, itup) != entry->attnum)
1555  return false;
1556 
1557  if (datumExtracted[off - 1] == false)
1558  {
1559  datum[off - 1] = gintuple_get_key(ginstate, itup,
1560  &category[off - 1]);
1561  datumExtracted[off - 1] = true;
1562  }
1563 
1564  /* Once we hit nulls, no further match is possible */
1565  if (category[off - 1] != GIN_CAT_NORM_KEY)
1566  return false;
1567 
1568  /*----------
1569  * Check partial match.
1570  * case cmp == 0 => match
1571  * case cmp > 0 => not match and end scan (no later match possible)
1572  * case cmp < 0 => not match and continue scan
1573  *----------
1574  */
1575  cmp = DatumGetInt32(FunctionCall4Coll(&ginstate->comparePartialFn[entry->attnum - 1],
1576  ginstate->supportCollation[entry->attnum - 1],
1577  entry->queryKey,
1578  datum[off - 1],
1579  UInt16GetDatum(entry->strategy),
1580  PointerGetDatum(entry->extra_data)));
1581  if (cmp == 0)
1582  return true;
1583  else if (cmp > 0)
1584  return false;
1585 
1586  off++;
1587  }
1588 
1589  return false;
1590 }
1591 
1592 /*
1593  * Set up the entryRes array for each key by looking at
1594  * every entry for current heap row in pending list.
1595  *
1596  * Returns true if each scan key has at least one entryRes match.
1597  * This corresponds to the situations where the normal index search will
1598  * try to apply the key's consistentFn. (A tuple not meeting that requirement
1599  * cannot be returned by the normal search since no entry stream will
1600  * source its TID.)
1601  *
1602  * The pendingBuffer is presumed pinned and share-locked on entry.
1603  */
1604 static bool
1606 {
1607  GinScanOpaque so = (GinScanOpaque) scan->opaque;
1608  OffsetNumber attrnum;
1609  Page page;
1610  IndexTuple itup;
1611  int i,
1612  j;
1613 
1614  /*
1615  * Reset all entryRes and hasMatchKey flags
1616  */
1617  for (i = 0; i < so->nkeys; i++)
1618  {
1619  GinScanKey key = so->keys + i;
1620 
1621  memset(key->entryRes, GIN_FALSE, key->nentries);
1622  }
1623  memset(pos->hasMatchKey, false, so->nkeys);
1624 
1625  /*
1626  * Outer loop iterates over multiple pending-list pages when a single heap
1627  * row has entries spanning those pages.
1628  */
1629  for (;;)
1630  {
1631  Datum datum[BLCKSZ / sizeof(IndexTupleData)];
1632  GinNullCategory category[BLCKSZ / sizeof(IndexTupleData)];
1633  bool datumExtracted[BLCKSZ / sizeof(IndexTupleData)];
1634 
1635  Assert(pos->lastOffset > pos->firstOffset);
1636  memset(datumExtracted + pos->firstOffset - 1, 0,
1637  sizeof(bool) * (pos->lastOffset - pos->firstOffset));
1638 
1639  page = BufferGetPage(pos->pendingBuffer);
1640 
1641  for (i = 0; i < so->nkeys; i++)
1642  {
1643  GinScanKey key = so->keys + i;
1644 
1645  for (j = 0; j < key->nentries; j++)
1646  {
1647  GinScanEntry entry = key->scanEntry[j];
1648  OffsetNumber StopLow = pos->firstOffset,
1649  StopHigh = pos->lastOffset,
1650  StopMiddle;
1651 
1652  /* If already matched on earlier page, do no extra work */
1653  if (key->entryRes[j])
1654  continue;
1655 
1656  /*
1657  * Interesting tuples are from pos->firstOffset to
1658  * pos->lastOffset and they are ordered by (attnum, Datum) as
1659  * it's done in entry tree. So we can use binary search to
1660  * avoid linear scanning.
1661  */
1662  while (StopLow < StopHigh)
1663  {
1664  int res;
1665 
1666  StopMiddle = StopLow + ((StopHigh - StopLow) >> 1);
1667 
1668  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, StopMiddle));
1669 
1670  attrnum = gintuple_get_attrnum(&so->ginstate, itup);
1671 
1672  if (key->attnum < attrnum)
1673  {
1674  StopHigh = StopMiddle;
1675  continue;
1676  }
1677  if (key->attnum > attrnum)
1678  {
1679  StopLow = StopMiddle + 1;
1680  continue;
1681  }
1682 
1683  if (datumExtracted[StopMiddle - 1] == false)
1684  {
1685  datum[StopMiddle - 1] =
1686  gintuple_get_key(&so->ginstate, itup,
1687  &category[StopMiddle - 1]);
1688  datumExtracted[StopMiddle - 1] = true;
1689  }
1690 
1691  if (entry->queryCategory == GIN_CAT_EMPTY_QUERY)
1692  {
1693  /* special behavior depending on searchMode */
1694  if (entry->searchMode == GIN_SEARCH_MODE_ALL)
1695  {
1696  /* match anything except NULL_ITEM */
1697  if (category[StopMiddle - 1] == GIN_CAT_NULL_ITEM)
1698  res = -1;
1699  else
1700  res = 0;
1701  }
1702  else
1703  {
1704  /* match everything */
1705  res = 0;
1706  }
1707  }
1708  else
1709  {
1711  entry->attnum,
1712  entry->queryKey,
1713  entry->queryCategory,
1714  datum[StopMiddle - 1],
1715  category[StopMiddle - 1]);
1716  }
1717 
1718  if (res == 0)
1719  {
1720  /*
1721  * Found exact match (there can be only one, except in
1722  * EMPTY_QUERY mode).
1723  *
1724  * If doing partial match, scan forward from here to
1725  * end of page to check for matches.
1726  *
1727  * See comment above about tuple's ordering.
1728  */
1729  if (entry->isPartialMatch)
1730  key->entryRes[j] =
1732  page,
1733  StopMiddle,
1734  pos->lastOffset,
1735  entry,
1736  datum,
1737  category,
1738  datumExtracted);
1739  else
1740  key->entryRes[j] = true;
1741 
1742  /* done with binary search */
1743  break;
1744  }
1745  else if (res < 0)
1746  StopHigh = StopMiddle;
1747  else
1748  StopLow = StopMiddle + 1;
1749  }
1750 
1751  if (StopLow >= StopHigh && entry->isPartialMatch)
1752  {
1753  /*
1754  * No exact match on this page. If doing partial match,
1755  * scan from the first tuple greater than target value to
1756  * end of page. Note that since we don't remember whether
1757  * the comparePartialFn told us to stop early on a
1758  * previous page, we will uselessly apply comparePartialFn
1759  * to the first tuple on each subsequent page.
1760  */
1761  key->entryRes[j] =
1763  page,
1764  StopHigh,
1765  pos->lastOffset,
1766  entry,
1767  datum,
1768  category,
1769  datumExtracted);
1770  }
1771 
1772  pos->hasMatchKey[i] |= key->entryRes[j];
1773  }
1774  }
1775 
1776  /* Advance firstOffset over the scanned tuples */
1777  pos->firstOffset = pos->lastOffset;
1778 
1779  if (GinPageHasFullRow(page))
1780  {
1781  /*
1782  * We have examined all pending entries for the current heap row.
1783  * Break out of loop over pages.
1784  */
1785  break;
1786  }
1787  else
1788  {
1789  /*
1790  * Advance to next page of pending entries for the current heap
1791  * row. Complain if there isn't one.
1792  */
1793  ItemPointerData item = pos->item;
1794 
1795  if (scanGetCandidate(scan, pos) == false ||
1796  !ItemPointerEquals(&pos->item, &item))
1797  elog(ERROR, "could not find additional pending pages for same heap tuple");
1798  }
1799  }
1800 
1801  /*
1802  * All scan keys except excludeOnly require at least one entry to match.
1803  * excludeOnly keys are an exception, because their implied
1804  * GIN_CAT_EMPTY_QUERY scanEntry always matches. So return "true" if all
1805  * non-excludeOnly scan keys have at least one match.
1806  */
1807  for (i = 0; i < so->nkeys; i++)
1808  {
1809  if (pos->hasMatchKey[i] == false && !so->keys[i].excludeOnly)
1810  return false;
1811  }
1812 
1813  return true;
1814 }
1815 
1816 /*
1817  * Collect all matched rows from pending list into bitmap.
1818  */
1819 static void
1820 scanPendingInsert(IndexScanDesc scan, TIDBitmap *tbm, int64 *ntids)
1821 {
1822  GinScanOpaque so = (GinScanOpaque) scan->opaque;
1823  MemoryContext oldCtx;
1824  bool recheck,
1825  match;
1826  int i;
1827  pendingPosition pos;
1828  Buffer metabuffer = ReadBuffer(scan->indexRelation, GIN_METAPAGE_BLKNO);
1829  Page page;
1830  BlockNumber blkno;
1831 
1832  *ntids = 0;
1833 
1834  /*
1835  * Acquire predicate lock on the metapage, to conflict with any fastupdate
1836  * insertions.
1837  */
1839 
1840  LockBuffer(metabuffer, GIN_SHARE);
1841  page = BufferGetPage(metabuffer);
1842  blkno = GinPageGetMeta(page)->head;
1843 
1844  /*
1845  * fetch head of list before unlocking metapage. head page must be pinned
1846  * to prevent deletion by vacuum process
1847  */
1848  if (blkno == InvalidBlockNumber)
1849  {
1850  /* No pending list, so proceed with normal scan */
1851  UnlockReleaseBuffer(metabuffer);
1852  return;
1853  }
1854 
1855  pos.pendingBuffer = ReadBuffer(scan->indexRelation, blkno);
1856  LockBuffer(pos.pendingBuffer, GIN_SHARE);
1857  pos.firstOffset = FirstOffsetNumber;
1858  UnlockReleaseBuffer(metabuffer);
1859  pos.hasMatchKey = palloc(sizeof(bool) * so->nkeys);
1860 
1861  /*
1862  * loop for each heap row. scanGetCandidate returns full row or row's
1863  * tuples from first page.
1864  */
1865  while (scanGetCandidate(scan, &pos))
1866  {
1867  /*
1868  * Check entries in tuple and set up entryRes array.
1869  *
1870  * If pending tuples belonging to the current heap row are spread
1871  * across several pages, collectMatchesForHeapRow will read all of
1872  * those pages.
1873  */
1874  if (!collectMatchesForHeapRow(scan, &pos))
1875  continue;
1876 
1877  /*
1878  * Matching of entries of one row is finished, so check row using
1879  * consistent functions.
1880  */
1881  oldCtx = MemoryContextSwitchTo(so->tempCtx);
1882  recheck = false;
1883  match = true;
1884 
1885  for (i = 0; i < so->nkeys; i++)
1886  {
1887  GinScanKey key = so->keys + i;
1888 
1889  if (!key->boolConsistentFn(key))
1890  {
1891  match = false;
1892  break;
1893  }
1894  recheck |= key->recheckCurItem;
1895  }
1896 
1897  MemoryContextSwitchTo(oldCtx);
1899 
1900  if (match)
1901  {
1902  tbm_add_tuples(tbm, &pos.item, 1, recheck);
1903  (*ntids)++;
1904  }
1905  }
1906 
1907  pfree(pos.hasMatchKey);
1908 }
1909 
1910 
1911 #define GinIsVoidRes(s) ( ((GinScanOpaque) scan->opaque)->isVoidRes )
1912 
1913 int64
1915 {
1916  GinScanOpaque so = (GinScanOpaque) scan->opaque;
1917  int64 ntids;
1918  ItemPointerData iptr;
1919  bool recheck;
1920 
1921  /*
1922  * Set up the scan keys, and check for unsatisfiable query.
1923  */
1924  ginFreeScanKeys(so); /* there should be no keys yet, but just to be
1925  * sure */
1926  ginNewScanKey(scan);
1927 
1928  if (GinIsVoidRes(scan))
1929  return 0;
1930 
1931  ntids = 0;
1932 
1933  /*
1934  * First, scan the pending list and collect any matching entries into the
1935  * bitmap. After we scan a pending item, some other backend could post it
1936  * into the main index, and so we might visit it a second time during the
1937  * main scan. This is okay because we'll just re-set the same bit in the
1938  * bitmap. (The possibility of duplicate visits is a major reason why GIN
1939  * can't support the amgettuple API, however.) Note that it would not do
1940  * to scan the main index before the pending list, since concurrent
1941  * cleanup could then make us miss entries entirely.
1942  */
1943  scanPendingInsert(scan, tbm, &ntids);
1944 
1945  /*
1946  * Now scan the main index.
1947  */
1948  startScan(scan);
1949 
1950  ItemPointerSetMin(&iptr);
1951 
1952  for (;;)
1953  {
1955 
1956  if (!scanGetItem(scan, iptr, &iptr, &recheck))
1957  break;
1958 
1959  if (ItemPointerIsLossyPage(&iptr))
1961  else
1962  tbm_add_tuples(tbm, &iptr, 1, recheck);
1963  ntids++;
1964  }
1965 
1966  return ntids;
1967 }
uint32 BlockNumber
Definition: block.h:31
#define InvalidBlockNumber
Definition: block.h:33
int Buffer
Definition: buf.h:23
#define InvalidBuffer
Definition: buf.h:25
void IncrBufferRefCount(Buffer buffer)
Definition: bufmgr.c:4882
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:3667
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:4850
void UnlockReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:4867
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:5085
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:745
static Page BufferGetPage(Buffer buffer)
Definition: bufmgr.h:408
static bool BufferIsValid(Buffer bufnum)
Definition: bufmgr.h:359
Pointer Page
Definition: bufpage.h:78
static Item PageGetItem(Page page, ItemId itemId)
Definition: bufpage.h:351
static ItemId PageGetItemId(Page page, OffsetNumber offsetNumber)
Definition: bufpage.h:240
static OffsetNumber PageGetMaxOffsetNumber(Page page)
Definition: bufpage.h:369
unsigned int uint32
Definition: c.h:506
signed int int32
Definition: c.h:494
#define Assert(condition)
Definition: c.h:858
Datum datumCopy(Datum value, bool typByVal, int typLen)
Definition: datum.c:132
int errcode(int sqlerrcode)
Definition: elog.c:859
int errmsg(const char *fmt,...)
Definition: elog.c:1072
#define DEBUG2
Definition: elog.h:29
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:224
#define ereport(elevel,...)
Definition: elog.h:149
Datum FunctionCall4Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2, Datum arg3, Datum arg4)
Definition: fmgr.c:1196
#define GIN_SEARCH_MODE_ALL
Definition: gin.h:36
#define GIN_FALSE
Definition: gin.h:63
char GinTernaryValue
Definition: gin.h:58
#define GIN_MAYBE
Definition: gin.h:65
#define GIN_TRUE
Definition: gin.h:64
GinScanOpaqueData * GinScanOpaque
Definition: gin_private.h:386
static int ginCompareItemPointers(ItemPointer a, ItemPointer b)
Definition: gin_private.h:488
#define GIN_UNLOCK
Definition: gin_private.h:49
struct GinScanKeyData * GinScanKey
Definition: gin_private.h:264
#define GIN_SHARE
Definition: gin_private.h:50
#define ItemPointerSetLossyPage(p, b)
Definition: ginblock.h:173
#define GinIsPostingTree(itup)
Definition: ginblock.h:231
#define GinItemPointerGetOffsetNumber(pointer)
Definition: ginblock.h:146
#define GIN_METAPAGE_BLKNO
Definition: ginblock.h:51
#define GinPageHasFullRow(page)
Definition: ginblock.h:119
#define GinPageGetOpaque(page)
Definition: ginblock.h:110
#define ItemPointerIsMin(p)
Definition: ginblock.h:168
#define GIN_CAT_NORM_KEY
Definition: ginblock.h:208
#define GIN_DELETED
Definition: ginblock.h:43
#define GinGetNPosting(itup)
Definition: ginblock.h:228
#define ItemPointerSetMax(p)
Definition: ginblock.h:171
#define GinDataPageGetRightBound(page)
Definition: ginblock.h:288
#define GinGetPostingTree(itup)
Definition: ginblock.h:233
#define ItemPointerIsLossyPage(p)
Definition: ginblock.h:175
signed char GinNullCategory
Definition: ginblock.h:206
#define GinPageRightMost(page)
Definition: ginblock.h:129
#define GIN_CAT_NULL_ITEM
Definition: ginblock.h:211
#define ItemPointerSetMin(p)
Definition: ginblock.h:166
#define GinPageGetMeta(p)
Definition: ginblock.h:104
#define GIN_CAT_EMPTY_QUERY
Definition: ginblock.h:212
#define GinItemPointerGetBlockNumber(pointer)
Definition: ginblock.h:143
#define GinPageIsLeaf(page)
Definition: ginblock.h:112
void freeGinBtreeStack(GinBtreeStack *stack)
Definition: ginbtree.c:198
Buffer ginStepRight(Buffer buffer, Relation index, int lockmode)
Definition: ginbtree.c:177
GinBtreeStack * ginFindLeafPage(GinBtree btree, bool searchMode, bool rootConflictCheck)
Definition: ginbtree.c:83
GinBtreeStack * ginScanBeginPostingTree(GinBtree btree, Relation index, BlockNumber rootBlkno)
Definition: gindatapage.c:1936
int GinDataLeafPageGetItemsToTbm(Page page, TIDBitmap *tbm)
Definition: gindatapage.c:182
ItemPointer GinDataLeafPageGetItems(Page page, int *nitems, ItemPointerData advancePast)
Definition: gindatapage.c:135
ItemPointer ginReadTuple(GinState *ginstate, OffsetNumber attnum, IndexTuple itup, int *nitems)
Definition: ginentrypage.c:162
void ginPrepareEntryScan(GinBtree btree, OffsetNumber attnum, Datum key, GinNullCategory category, GinState *ginstate)
Definition: ginentrypage.c:747
int64 gingetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
Definition: ginget.c:1914
static void scanPendingInsert(IndexScanDesc scan, TIDBitmap *tbm, int64 *ntids)
Definition: ginget.c:1820
static void startScan(IndexScanDesc scan)
Definition: ginget.c:601
struct pendingPosition pendingPosition
static void startScanEntry(GinState *ginstate, GinScanEntry entry, Snapshot snapshot)
Definition: ginget.c:319
static void startScanKey(GinState *ginstate, GinScanOpaque so, GinScanKey key)
Definition: ginget.c:505
#define GinIsVoidRes(s)
Definition: ginget.c:1911
static bool scanGetItem(IndexScanDesc scan, ItemPointerData advancePast, ItemPointerData *item, bool *recheck)
Definition: ginget.c:1285
int GinFuzzySearchLimit
Definition: ginget.c:27
static bool matchPartialInPendingList(GinState *ginstate, Page page, OffsetNumber off, OffsetNumber maxoff, GinScanEntry entry, Datum *datum, GinNullCategory *category, bool *datumExtracted)
Definition: ginget.c:1537
static bool scanGetCandidate(IndexScanDesc scan, pendingPosition *pos)
Definition: ginget.c:1450
static void scanPostingTree(Relation index, GinScanEntry scanEntry, BlockNumber rootPostingTree)
Definition: ginget.c:69
static bool moveRightIfItNeeded(GinBtreeData *btree, GinBtreeStack *stack, Snapshot snapshot)
Definition: ginget.c:43
static int entryIndexByFrequencyCmp(const void *a1, const void *a2, void *arg)
Definition: ginget.c:488
#define dropItem(e)
Definition: ginget.c:792
static bool collectMatchBitmap(GinBtreeData *btree, GinBtreeStack *stack, GinScanEntry scanEntry, Snapshot snapshot)
Definition: ginget.c:121
static bool collectMatchesForHeapRow(IndexScanDesc scan, pendingPosition *pos)
Definition: ginget.c:1605
static void keyGetItem(GinState *ginstate, MemoryContext tempCtx, GinScanKey key, ItemPointerData advancePast)
Definition: ginget.c:990
static void entryGetItem(GinState *ginstate, GinScanEntry entry, ItemPointerData advancePast)
Definition: ginget.c:808
static void entryLoadMoreItems(GinState *ginstate, GinScanEntry entry, ItemPointerData advancePast)
Definition: ginget.c:653
void ginFreeScanKeys(GinScanOpaque so)
Definition: ginscan.c:233
void ginNewScanKey(IndexScanDesc scan)
Definition: ginscan.c:263
OffsetNumber gintuple_get_attrnum(GinState *ginstate, IndexTuple tuple)
Definition: ginutil.c:226
Datum gintuple_get_key(GinState *ginstate, IndexTuple tuple, GinNullCategory *category)
Definition: ginutil.c:259
int ginCompareEntries(GinState *ginstate, OffsetNumber attnum, Datum a, GinNullCategory categorya, Datum b, GinNullCategory categoryb)
Definition: ginutil.c:388
int work_mem
Definition: globals.c:128
for(;;)
static const FormData_pg_attribute a1
Definition: heap.c:142
static const FormData_pg_attribute a2
Definition: heap.c:156
int j
Definition: isn.c:74
int i
Definition: isn.c:73
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:77
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:35
static void ItemPointerSet(ItemPointerData *pointer, BlockNumber blockNumber, OffsetNumber offNum)
Definition: itemptr.h:135
static void ItemPointerSetInvalid(ItemPointerData *pointer)
Definition: itemptr.h:184
static BlockNumber ItemPointerGetBlockNumber(const ItemPointerData *pointer)
Definition: itemptr.h:103
static bool ItemPointerIsValid(const ItemPointerData *pointer)
Definition: itemptr.h:83
IndexTupleData * IndexTuple
Definition: itup.h:53
struct IndexTupleData IndexTupleData
void MemoryContextReset(MemoryContext context)
Definition: mcxt.c:383
void pfree(void *pointer)
Definition: mcxt.c:1520
MemoryContext CurrentMemoryContext
Definition: mcxt.c:143
void * palloc(Size size)
Definition: mcxt.c:1316
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:122
#define InvalidOffsetNumber
Definition: off.h:26
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
uint16 OffsetNumber
Definition: off.h:24
#define FirstOffsetNumber
Definition: off.h:27
#define OffsetNumberPrev(offsetNumber)
Definition: off.h:54
static void reduce(void)
Definition: parse.c:260
int16 attnum
Definition: pg_attribute.h:74
FormData_pg_attribute * Form_pg_attribute
Definition: pg_attribute.h:209
void * arg
void qsort_arg(void *base, size_t nel, size_t elsize, qsort_arg_comparator cmp, void *arg)
static Datum PointerGetDatum(const void *X)
Definition: postgres.h:322
uintptr_t Datum
Definition: postgres.h:64
static Datum UInt16GetDatum(uint16 X)
Definition: postgres.h:192
static Pointer DatumGetPointer(Datum X)
Definition: postgres.h:312
static int32 DatumGetInt32(Datum X)
Definition: postgres.h:202
void PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
Definition: predicate.c:2584
MemoryContextSwitchTo(old_ctx)
static int cmp(const chr *x, const chr *y, size_t len)
Definition: regc_locale.c:743
#define RelationGetRelationName(relation)
Definition: rel.h:539
GinState * ginstate
Definition: gin_private.h:169
bool(* findItem)(GinBtree, GinBtreeStack *)
Definition: gin_private.h:156
ItemPointerData itemptr
Definition: gin_private.h:179
Relation index
Definition: gin_private.h:167
OffsetNumber off
Definition: gin_private.h:133
uint32 predictNumber
Definition: gin_private.h:136
BlockNumber blkno
Definition: gin_private.h:131
ItemPointerData curItem
Definition: gin_private.h:351
TBMIterateResult * matchResult
Definition: gin_private.h:356
GinBtreeData btree
Definition: gin_private.h:366
TIDBitmap * matchBitmap
Definition: gin_private.h:354
ItemPointerData * list
Definition: gin_private.h:359
GinNullCategory queryCategory
Definition: gin_private.h:340
TBMIterator * matchIterator
Definition: gin_private.h:355
StrategyNumber strategy
Definition: gin_private.h:343
Pointer extra_data
Definition: gin_private.h:342
OffsetNumber offset
Definition: gin_private.h:361
uint32 predictNumberResult
Definition: gin_private.h:365
OffsetNumber attnum
Definition: gin_private.h:345
GinScanKey keys
Definition: gin_private.h:374
MemoryContext keyCtx
Definition: gin_private.h:381
GinScanEntry * entries
Definition: gin_private.h:377
MemoryContext tempCtx
Definition: gin_private.h:371
FmgrInfo comparePartialFn[INDEX_MAX_KEYS]
Definition: gin_private.h:84
TupleDesc origTupdesc
Definition: gin_private.h:73
Relation index
Definition: gin_private.h:59
Oid supportCollation[INDEX_MAX_KEYS]
Definition: gin_private.h:88
Relation indexRelation
Definition: relscan.h:118
struct SnapshotData * xs_snapshot
Definition: relscan.h:119
ItemPointerData t_tid
Definition: itup.h:37
OffsetNumber offsets[FLEXIBLE_ARRAY_MEMBER]
Definition: tidbitmap.h:46
BlockNumber blockno
Definition: tidbitmap.h:42
Definition: type.h:95
Buffer pendingBuffer
Definition: ginget.c:31
OffsetNumber lastOffset
Definition: ginget.c:33
OffsetNumber firstOffset
Definition: ginget.c:32
bool * hasMatchKey
Definition: ginget.c:35
ItemPointerData item
Definition: ginget.c:34
void tbm_free(TIDBitmap *tbm)
Definition: tidbitmap.c:322
TBMIterator * tbm_begin_iterate(TIDBitmap *tbm)
Definition: tidbitmap.c:689
void tbm_add_tuples(TIDBitmap *tbm, const ItemPointer tids, int ntids, bool recheck)
Definition: tidbitmap.c:377
bool tbm_is_empty(const TIDBitmap *tbm)
Definition: tidbitmap.c:670
void tbm_end_iterate(TBMIterator *iterator)
Definition: tidbitmap.c:1146
void tbm_add_page(TIDBitmap *tbm, BlockNumber pageno)
Definition: tidbitmap.c:443
TIDBitmap * tbm_create(long maxbytes, dsa_area *dsa)
Definition: tidbitmap.c:266
TBMIterateResult * tbm_iterate(TBMIterator *iterator)
Definition: tidbitmap.c:971
#define TupleDescAttr(tupdesc, i)
Definition: tupdesc.h:92
static StringInfoData tmpbuf
Definition: walsender.c:170