PostgreSQL Source Code  git master
nbtutils.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * nbtutils.c
4  * Utility code for Postgres btree implementation.
5  *
6  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  *
10  * IDENTIFICATION
11  * src/backend/access/nbtree/nbtutils.c
12  *
13  *-------------------------------------------------------------------------
14  */
15 
16 #include "postgres.h"
17 
18 #include <time.h>
19 
20 #include "access/nbtree.h"
21 #include "access/reloptions.h"
22 #include "access/relscan.h"
23 #include "commands/progress.h"
24 #include "miscadmin.h"
25 #include "utils/array.h"
26 #include "utils/datum.h"
27 #include "utils/lsyscache.h"
28 #include "utils/memutils.h"
29 #include "utils/rel.h"
30 
31 
32 typedef struct BTSortArrayContext
33 {
34  FmgrInfo flinfo;
35  Oid collation;
36  bool reverse;
37 } BTSortArrayContext;
38 
39 static Datum _bt_find_extreme_element(IndexScanDesc scan, ScanKey skey,
40  StrategyNumber strat,
41  Datum *elems, int nelems);
42 static int _bt_sort_array_elements(IndexScanDesc scan, ScanKey skey,
43  bool reverse,
44  Datum *elems, int nelems);
45 static int _bt_compare_array_elements(const void *a, const void *b, void *arg);
46 static bool _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op,
47  ScanKey leftarg, ScanKey rightarg,
48  bool *result);
49 static bool _bt_fix_scankey_strategy(ScanKey skey, int16 *indoption);
50 static void _bt_mark_scankey_required(ScanKey skey);
51 static bool _bt_check_rowcompare(ScanKey skey,
52  IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
53  ScanDirection dir, bool *continuescan);
54 static int _bt_keep_natts(Relation rel, IndexTuple lastleft,
55  IndexTuple firstright, BTScanInsert itup_key);
56 
57 
58 /*
59  * _bt_mkscankey
60  * Build an insertion scan key that contains comparison data from itup
61  * as well as comparator routines appropriate to the key datatypes.
62  *
63  * When itup is a non-pivot tuple, the returned insertion scan key is
64  * suitable for finding a place for it to go on the leaf level. Pivot
65  * tuples can be used to re-find leaf page with matching high key, but
66  * then caller needs to set scan key's pivotsearch field to true. This
67  * allows caller to search for a leaf page with a matching high key,
68  * which is usually to the left of the first leaf page a non-pivot match
69  * might appear on.
70  *
71  * The result is intended for use with _bt_compare() and _bt_truncate().
72  * Callers that don't need to fill out the insertion scankey arguments
73  * (e.g. they use an ad-hoc comparison routine, or only need a scankey
74  * for _bt_truncate()) can pass a NULL index tuple. The scankey will
75  * be initialized as if an "all truncated" pivot tuple was passed
76  * instead.
77  *
78  * Note that we may occasionally have to share lock the metapage to
79  * determine whether or not the keys in the index are expected to be
80  * unique (i.e. if this is a "heapkeyspace" index). We assume a
81  * heapkeyspace index when caller passes a NULL tuple, allowing index
82  * build callers to avoid accessing the non-existent metapage.
83  */
84 BTScanInsert
85 _bt_mkscankey(Relation rel, IndexTuple itup)
86 {
87  BTScanInsert key;
88  ScanKey skey;
89  TupleDesc itupdesc;
90  int indnkeyatts;
91  int16 *indoption;
92  int tupnatts;
93  int i;
94 
95  itupdesc = RelationGetDescr(rel);
96  indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
97  indoption = rel->rd_indoption;
98  tupnatts = itup ? BTreeTupleGetNAtts(itup, rel) : 0;
99 
100  Assert(tupnatts <= IndexRelationGetNumberOfAttributes(rel));
101 
102  /*
103  * We'll execute search using scan key constructed on key columns.
104  * Truncated attributes and non-key attributes are omitted from the final
105  * scan key.
106  */
107  key = palloc(offsetof(BTScanInsertData, scankeys) +
108  sizeof(ScanKeyData) * indnkeyatts);
109  key->heapkeyspace = itup == NULL || _bt_heapkeyspace(rel);
110  key->anynullkeys = false; /* initial assumption */
111  key->nextkey = false;
112  key->pivotsearch = false;
113  key->keysz = Min(indnkeyatts, tupnatts);
114  key->scantid = key->heapkeyspace && itup ?
115  BTreeTupleGetHeapTID(itup) : NULL;
116  skey = key->scankeys;
117  for (i = 0; i < indnkeyatts; i++)
118  {
119  FmgrInfo *procinfo;
120  Datum arg;
121  bool null;
122  int flags;
123 
124  /*
125  * We can use the cached (default) support procs since no cross-type
126  * comparison can be needed.
127  */
128  procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
129 
130  /*
131  * Key arguments built from truncated attributes (or when caller
132  * provides no tuple) are defensively represented as NULL values. They
133  * should never be used.
134  */
135  if (i < tupnatts)
136  arg = index_getattr(itup, i + 1, itupdesc, &null);
137  else
138  {
139  arg = (Datum) 0;
140  null = true;
141  }
142  flags = (null ? SK_ISNULL : 0) | (indoption[i] << SK_BT_INDOPTION_SHIFT);
143  ScanKeyEntryInitializeWithInfo(&skey[i],
144  flags,
145  (AttrNumber) (i + 1),
146  InvalidStrategy,
147  InvalidOid,
148  rel->rd_indcollation[i],
149  procinfo,
150  arg);
151  /* Record if any key attribute is NULL (or truncated) */
152  if (null)
153  key->anynullkeys = true;
154  }
155 
156  return key;
157 }
158 
159 /*
160  * free a retracement stack made by _bt_search.
161  */
162 void
163 _bt_freestack(BTStack stack)
164 {
165  BTStack ostack;
166 
167  while (stack != NULL)
168  {
169  ostack = stack;
170  stack = stack->bts_parent;
171  pfree(ostack);
172  }
173 }
174 
175 
176 /*
177  * _bt_preprocess_array_keys() -- Preprocess SK_SEARCHARRAY scan keys
178  *
179  * If there are any SK_SEARCHARRAY scan keys, deconstruct the array(s) and
180  * set up BTArrayKeyInfo info for each one that is an equality-type key.
181  * Prepare modified scan keys in so->arrayKeyData, which will hold the current
182  * array elements during each primitive indexscan operation. For inequality
183  * array keys, it's sufficient to find the extreme element value and replace
184  * the whole array with that scalar value.
185  *
186  * Note: the reason we need so->arrayKeyData, rather than just scribbling
187  * on scan->keyData, is that callers are permitted to call btrescan without
188  * supplying a new set of scankey data.
189  */
190 void
191 _bt_preprocess_array_keys(IndexScanDesc scan)
192 {
193  BTScanOpaque so = (BTScanOpaque) scan->opaque;
194  int numberOfKeys = scan->numberOfKeys;
195  int16 *indoption = scan->indexRelation->rd_indoption;
196  int numArrayKeys;
197  ScanKey cur;
198  int i;
199  MemoryContext oldContext;
200 
201  /* Quick check to see if there are any array keys */
202  numArrayKeys = 0;
203  for (i = 0; i < numberOfKeys; i++)
204  {
205  cur = &scan->keyData[i];
206  if (cur->sk_flags & SK_SEARCHARRAY)
207  {
208  numArrayKeys++;
209  Assert(!(cur->sk_flags & (SK_ROW_HEADER | SK_SEARCHNULL | SK_SEARCHNOTNULL)));
210  /* If any arrays are null as a whole, we can quit right now. */
211  if (cur->sk_flags & SK_ISNULL)
212  {
213  so->numArrayKeys = -1;
214  so->arrayKeyData = NULL;
215  return;
216  }
217  }
218  }
219 
220  /* Quit if nothing to do. */
221  if (numArrayKeys == 0)
222  {
223  so->numArrayKeys = 0;
224  so->arrayKeyData = NULL;
225  return;
226  }
227 
228  /*
229  * Make a scan-lifespan context to hold array-associated data, or reset it
230  * if we already have one from a previous rescan cycle.
231  */
232  if (so->arrayContext == NULL)
233  so->arrayContext = AllocSetContextCreate(CurrentMemoryContext,
234  "BTree array context",
235  ALLOCSET_SMALL_SIZES);
236  else
237  MemoryContextReset(so->arrayContext);
238 
239  oldContext = MemoryContextSwitchTo(so->arrayContext);
240 
241  /* Create modifiable copy of scan->keyData in the workspace context */
242  so->arrayKeyData = (ScanKey) palloc(scan->numberOfKeys * sizeof(ScanKeyData));
243  memcpy(so->arrayKeyData,
244  scan->keyData,
245  scan->numberOfKeys * sizeof(ScanKeyData));
246 
247  /* Allocate space for per-array data in the workspace context */
248  so->arrayKeys = (BTArrayKeyInfo *) palloc0(numArrayKeys * sizeof(BTArrayKeyInfo));
249 
250  /* Now process each array key */
251  numArrayKeys = 0;
252  for (i = 0; i < numberOfKeys; i++)
253  {
254  ArrayType *arrayval;
255  int16 elmlen;
256  bool elmbyval;
257  char elmalign;
258  int num_elems;
259  Datum *elem_values;
260  bool *elem_nulls;
261  int num_nonnulls;
262  int j;
263 
264  cur = &so->arrayKeyData[i];
265  if (!(cur->sk_flags & SK_SEARCHARRAY))
266  continue;
267 
268  /*
269  * First, deconstruct the array into elements. Anything allocated
270  * here (including a possibly detoasted array value) is in the
271  * workspace context.
272  */
273  arrayval = DatumGetArrayTypeP(cur->sk_argument);
274  /* We could cache this data, but not clear it's worth it */
275  get_typlenbyvalalign(ARR_ELEMTYPE(arrayval),
276  &elmlen, &elmbyval, &elmalign);
277  deconstruct_array(arrayval,
278  ARR_ELEMTYPE(arrayval),
279  elmlen, elmbyval, elmalign,
280  &elem_values, &elem_nulls, &num_elems);
281 
282  /*
283  * Compress out any null elements. We can ignore them since we assume
284  * all btree operators are strict.
285  */
286  num_nonnulls = 0;
287  for (j = 0; j < num_elems; j++)
288  {
289  if (!elem_nulls[j])
290  elem_values[num_nonnulls++] = elem_values[j];
291  }
292 
293  /* We could pfree(elem_nulls) now, but not worth the cycles */
294 
295  /* If there's no non-nulls, the scan qual is unsatisfiable */
296  if (num_nonnulls == 0)
297  {
298  numArrayKeys = -1;
299  break;
300  }
301 
302  /*
303  * If the comparison operator is not equality, then the array qual
304  * degenerates to a simple comparison against the smallest or largest
305  * non-null array element, as appropriate.
306  */
307  switch (cur->sk_strategy)
308  {
309  case BTLessStrategyNumber:
310  case BTLessEqualStrategyNumber:
311  cur->sk_argument =
312  _bt_find_extreme_element(scan, cur,
313  BTGreaterStrategyNumber,
314  elem_values, num_nonnulls);
315  continue;
316  case BTEqualStrategyNumber:
317  /* proceed with rest of loop */
318  break;
319  case BTGreaterEqualStrategyNumber:
320  case BTGreaterStrategyNumber:
321  cur->sk_argument =
322  _bt_find_extreme_element(scan, cur,
323  BTLessStrategyNumber,
324  elem_values, num_nonnulls);
325  continue;
326  default:
327  elog(ERROR, "unrecognized StrategyNumber: %d",
328  (int) cur->sk_strategy);
329  break;
330  }
331 
332  /*
333  * Sort the non-null elements and eliminate any duplicates. We must
334  * sort in the same ordering used by the index column, so that the
335  * successive primitive indexscans produce data in index order.
336  */
337  num_elems = _bt_sort_array_elements(scan, cur,
338  (indoption[cur->sk_attno - 1] & INDOPTION_DESC) != 0,
339  elem_values, num_nonnulls);
340 
341  /*
342  * And set up the BTArrayKeyInfo data.
343  */
344  so->arrayKeys[numArrayKeys].scan_key = i;
345  so->arrayKeys[numArrayKeys].num_elems = num_elems;
346  so->arrayKeys[numArrayKeys].elem_values = elem_values;
347  numArrayKeys++;
348  }
349 
350  so->numArrayKeys = numArrayKeys;
351 
352  MemoryContextSwitchTo(oldContext);
353 }
354 
355 /*
356  * _bt_find_extreme_element() -- get least or greatest array element
357  *
358  * scan and skey identify the index column, whose opfamily determines the
359  * comparison semantics. strat should be BTLessStrategyNumber to get the
360  * least element, or BTGreaterStrategyNumber to get the greatest.
361  */
362 static Datum
363 _bt_find_extreme_element(IndexScanDesc scan, ScanKey skey,
364  StrategyNumber strat,
365  Datum *elems, int nelems)
366 {
367  Relation rel = scan->indexRelation;
368  Oid elemtype,
369  cmp_op;
370  RegProcedure cmp_proc;
371  FmgrInfo flinfo;
372  Datum result;
373  int i;
374 
375  /*
376  * Determine the nominal datatype of the array elements. We have to
377  * support the convention that sk_subtype == InvalidOid means the opclass
378  * input type; this is a hack to simplify life for ScanKeyInit().
379  */
380  elemtype = skey->sk_subtype;
381  if (elemtype == InvalidOid)
382  elemtype = rel->rd_opcintype[skey->sk_attno - 1];
383 
384  /*
385  * Look up the appropriate comparison operator in the opfamily.
386  *
387  * Note: it's possible that this would fail, if the opfamily is
388  * incomplete, but it seems quite unlikely that an opfamily would omit
389  * non-cross-type comparison operators for any datatype that it supports
390  * at all.
391  */
392  cmp_op = get_opfamily_member(rel->rd_opfamily[skey->sk_attno - 1],
393  elemtype,
394  elemtype,
395  strat);
396  if (!OidIsValid(cmp_op))
397  elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
398  strat, elemtype, elemtype,
399  rel->rd_opfamily[skey->sk_attno - 1]);
400  cmp_proc = get_opcode(cmp_op);
401  if (!RegProcedureIsValid(cmp_proc))
402  elog(ERROR, "missing oprcode for operator %u", cmp_op);
403 
404  fmgr_info(cmp_proc, &flinfo);
405 
406  Assert(nelems > 0);
407  result = elems[0];
408  for (i = 1; i < nelems; i++)
409  {
410  if (DatumGetBool(FunctionCall2Coll(&flinfo,
411  skey->sk_collation,
412  elems[i],
413  result)))
414  result = elems[i];
415  }
416 
417  return result;
418 }
419 
420 /*
421  * _bt_sort_array_elements() -- sort and de-dup array elements
422  *
423  * The array elements are sorted in-place, and the new number of elements
424  * after duplicate removal is returned.
425  *
426  * scan and skey identify the index column, whose opfamily determines the
427  * comparison semantics. If reverse is true, we sort in descending order.
428  */
429 static int
430 _bt_sort_array_elements(IndexScanDesc scan, ScanKey skey,
431  bool reverse,
432  Datum *elems, int nelems)
433 {
434  Relation rel = scan->indexRelation;
435  Oid elemtype;
436  RegProcedure cmp_proc;
437  BTSortArrayContext cxt;
438  int last_non_dup;
439  int i;
440 
441  if (nelems <= 1)
442  return nelems; /* no work to do */
443 
444  /*
445  * Determine the nominal datatype of the array elements. We have to
446  * support the convention that sk_subtype == InvalidOid means the opclass
447  * input type; this is a hack to simplify life for ScanKeyInit().
448  */
449  elemtype = skey->sk_subtype;
450  if (elemtype == InvalidOid)
451  elemtype = rel->rd_opcintype[skey->sk_attno - 1];
452 
453  /*
454  * Look up the appropriate comparison function in the opfamily.
455  *
456  * Note: it's possible that this would fail, if the opfamily is
457  * incomplete, but it seems quite unlikely that an opfamily would omit
458  * non-cross-type support functions for any datatype that it supports at
459  * all.
460  */
461  cmp_proc = get_opfamily_proc(rel->rd_opfamily[skey->sk_attno - 1],
462  elemtype,
463  elemtype,
464  BTORDER_PROC);
465  if (!RegProcedureIsValid(cmp_proc))
466  elog(ERROR, "missing support function %d(%u,%u) in opfamily %u",
467  BTORDER_PROC, elemtype, elemtype,
468  rel->rd_opfamily[skey->sk_attno - 1]);
469 
470  /* Sort the array elements */
471  fmgr_info(cmp_proc, &cxt.flinfo);
472  cxt.collation = skey->sk_collation;
473  cxt.reverse = reverse;
474  qsort_arg((void *) elems, nelems, sizeof(Datum),
475  _bt_compare_array_elements, (void *) &cxt);
476 
477  /* Now scan the sorted elements and remove duplicates */
478  last_non_dup = 0;
479  for (i = 1; i < nelems; i++)
480  {
481  int32 compare;
482 
483  compare = DatumGetInt32(FunctionCall2Coll(&cxt.flinfo,
484  cxt.collation,
485  elems[last_non_dup],
486  elems[i]));
487  if (compare != 0)
488  elems[++last_non_dup] = elems[i];
489  }
490 
491  return last_non_dup + 1;
492 }
493 
494 /*
495  * qsort_arg comparator for sorting array elements
496  */
497 static int
498 _bt_compare_array_elements(const void *a, const void *b, void *arg)
499 {
500  Datum da = *((const Datum *) a);
501  Datum db = *((const Datum *) b);
502  BTSortArrayContext *cxt = (BTSortArrayContext *) arg;
503  int32 compare;
504 
505  compare = DatumGetInt32(FunctionCall2Coll(&cxt->flinfo,
506  cxt->collation,
507  da, db));
508  if (cxt->reverse)
509  INVERT_COMPARE_RESULT(compare);
510  return compare;
511 }
512 
513 /*
514  * _bt_start_array_keys() -- Initialize array keys at start of a scan
515  *
516  * Set up the cur_elem counters and fill in the first sk_argument value for
517  * each array scankey. We can't do this until we know the scan direction.
518  */
519 void
520 _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
521 {
522  BTScanOpaque so = (BTScanOpaque) scan->opaque;
523  int i;
524 
525  for (i = 0; i < so->numArrayKeys; i++)
526  {
527  BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
528  ScanKey skey = &so->arrayKeyData[curArrayKey->scan_key];
529 
530  Assert(curArrayKey->num_elems > 0);
531  if (ScanDirectionIsBackward(dir))
532  curArrayKey->cur_elem = curArrayKey->num_elems - 1;
533  else
534  curArrayKey->cur_elem = 0;
535  skey->sk_argument = curArrayKey->elem_values[curArrayKey->cur_elem];
536  }
537 }
538 
539 /*
540  * _bt_advance_array_keys() -- Advance to next set of array elements
541  *
542  * Returns true if there is another set of values to consider, false if not.
543  * On true result, the scankeys are initialized with the next set of values.
544  */
545 bool
546 _bt_advance_array_keys(IndexScanDesc scan, ScanDirection dir)
547 {
548  BTScanOpaque so = (BTScanOpaque) scan->opaque;
549  bool found = false;
550  int i;
551 
552  /*
553  * We must advance the last array key most quickly, since it will
554  * correspond to the lowest-order index column among the available
555  * qualifications. This is necessary to ensure correct ordering of output
556  * when there are multiple array keys.
557  */
558  for (i = so->numArrayKeys - 1; i >= 0; i--)
559  {
560  BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
561  ScanKey skey = &so->arrayKeyData[curArrayKey->scan_key];
562  int cur_elem = curArrayKey->cur_elem;
563  int num_elems = curArrayKey->num_elems;
564 
565  if (ScanDirectionIsBackward(dir))
566  {
567  if (--cur_elem < 0)
568  {
569  cur_elem = num_elems - 1;
570  found = false; /* need to advance next array key */
571  }
572  else
573  found = true;
574  }
575  else
576  {
577  if (++cur_elem >= num_elems)
578  {
579  cur_elem = 0;
580  found = false; /* need to advance next array key */
581  }
582  else
583  found = true;
584  }
585 
586  curArrayKey->cur_elem = cur_elem;
587  skey->sk_argument = curArrayKey->elem_values[cur_elem];
588  if (found)
589  break;
590  }
591 
592  /* advance parallel scan */
593  if (scan->parallel_scan != NULL)
594  _bt_parallel_advance_array_keys(scan);
595 
596  return found;
597 }
598 
599 /*
600  * _bt_mark_array_keys() -- Handle array keys during btmarkpos
601  *
602  * Save the current state of the array keys as the "mark" position.
603  */
604 void
605 _bt_mark_array_keys(IndexScanDesc scan)
606 {
607  BTScanOpaque so = (BTScanOpaque) scan->opaque;
608  int i;
609 
610  for (i = 0; i < so->numArrayKeys; i++)
611  {
612  BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
613 
614  curArrayKey->mark_elem = curArrayKey->cur_elem;
615  }
616 }
617 
618 /*
619  * _bt_restore_array_keys() -- Handle array keys during btrestrpos
620  *
621  * Restore the array keys to where they were when the mark was set.
622  */
623 void
624 _bt_restore_array_keys(IndexScanDesc scan)
625 {
626  BTScanOpaque so = (BTScanOpaque) scan->opaque;
627  bool changed = false;
628  int i;
629 
630  /* Restore each array key to its position when the mark was set */
631  for (i = 0; i < so->numArrayKeys; i++)
632  {
633  BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
634  ScanKey skey = &so->arrayKeyData[curArrayKey->scan_key];
635  int mark_elem = curArrayKey->mark_elem;
636 
637  if (curArrayKey->cur_elem != mark_elem)
638  {
639  curArrayKey->cur_elem = mark_elem;
640  skey->sk_argument = curArrayKey->elem_values[mark_elem];
641  changed = true;
642  }
643  }
644 
645  /*
646  * If we changed any keys, we must redo _bt_preprocess_keys. That might
647  * sound like overkill, but in cases with multiple keys per index column
648  * it seems necessary to do the full set of pushups.
649  */
650  if (changed)
651  {
652  _bt_preprocess_keys(scan);
653  /* The mark should have been set on a consistent set of keys... */
654  Assert(so->qual_ok);
655  }
656 }
657 
658 
659 /*
660  * _bt_preprocess_keys() -- Preprocess scan keys
661  *
662  * The given search-type keys (in scan->keyData[] or so->arrayKeyData[])
663  * are copied to so->keyData[] with possible transformation.
664  * scan->numberOfKeys is the number of input keys, so->numberOfKeys gets
665  * the number of output keys (possibly less, never greater).
666  *
667  * The output keys are marked with additional sk_flags bits beyond the
668  * system-standard bits supplied by the caller. The DESC and NULLS_FIRST
669  * indoption bits for the relevant index attribute are copied into the flags.
670  * Also, for a DESC column, we commute (flip) all the sk_strategy numbers
671  * so that the index sorts in the desired direction.
672  *
673  * One key purpose of this routine is to discover which scan keys must be
674  * satisfied to continue the scan. It also attempts to eliminate redundant
675  * keys and detect contradictory keys. (If the index opfamily provides
676  * incomplete sets of cross-type operators, we may fail to detect redundant
677  * or contradictory keys, but we can survive that.)
678  *
679  * The output keys must be sorted by index attribute. Presently we expect
680  * (but verify) that the input keys are already so sorted --- this is done
681  * by match_clauses_to_index() in indxpath.c. Some reordering of the keys
682  * within each attribute may be done as a byproduct of the processing here,
683  * but no other code depends on that.
684  *
685  * The output keys are marked with flags SK_BT_REQFWD and/or SK_BT_REQBKWD
686  * if they must be satisfied in order to continue the scan forward or backward
687  * respectively. _bt_checkkeys uses these flags. For example, if the quals
688  * are "x = 1 AND y < 4 AND z < 5", then _bt_checkkeys will reject a tuple
689  * (1,2,7), but we must continue the scan in case there are tuples (1,3,z).
690  * But once we reach tuples like (1,4,z) we can stop scanning because no
691  * later tuples could match. This is reflected by marking the x and y keys,
692  * but not the z key, with SK_BT_REQFWD. In general, the keys for leading
693  * attributes with "=" keys are marked both SK_BT_REQFWD and SK_BT_REQBKWD.
694  * For the first attribute without an "=" key, any "<" and "<=" keys are
695  * marked SK_BT_REQFWD while any ">" and ">=" keys are marked SK_BT_REQBKWD.
696  * This can be seen to be correct by considering the above example. Note
697  * in particular that if there are no keys for a given attribute, the keys for
698  * subsequent attributes can never be required; for instance "WHERE y = 4"
699  * requires a full-index scan.
700  *
701  * If possible, redundant keys are eliminated: we keep only the tightest
702  * >/>= bound and the tightest </<= bound, and if there's an = key then
703  * that's the only one returned. (So, we return either a single = key,
704  * or one or two boundary-condition keys for each attr.) However, if we
705  * cannot compare two keys for lack of a suitable cross-type operator,
706  * we cannot eliminate either. If there are two such keys of the same
707  * operator strategy, the second one is just pushed into the output array
708  * without further processing here. We may also emit both >/>= or both
709  * </<= keys if we can't compare them. The logic about required keys still
710  * works if we don't eliminate redundant keys.
711  *
712  * Note that one reason we need direction-sensitive required-key flags is
713  * precisely that we may not be able to eliminate redundant keys. Suppose
714  * we have "x > 4::int AND x > 10::bigint", and we are unable to determine
715  * which key is more restrictive for lack of a suitable cross-type operator.
716  * _bt_first will arbitrarily pick one of the keys to do the initial
717  * positioning with. If it picks x > 4, then the x > 10 condition will fail
718  * until we reach index entries > 10; but we can't stop the scan just because
719  * x > 10 is failing. On the other hand, if we are scanning backwards, then
720  * failure of either key is indeed enough to stop the scan. (In general, when
721  * inequality keys are present, the initial-positioning code only promises to
722  * position before the first possible match, not exactly at the first match,
723  * for a forward scan; or after the last match for a backward scan.)
724  *
725  * As a byproduct of this work, we can detect contradictory quals such
726  * as "x = 1 AND x > 2". If we see that, we return so->qual_ok = false,
727  * indicating the scan need not be run at all since no tuples can match.
728  * (In this case we do not bother completing the output key array!)
729  * Again, missing cross-type operators might cause us to fail to prove the
730  * quals contradictory when they really are, but the scan will work correctly.
731  *
732  * Row comparison keys are currently also treated without any smarts:
733  * we just transfer them into the preprocessed array without any
734  * editorialization. We can treat them the same as an ordinary inequality
735  * comparison on the row's first index column, for the purposes of the logic
736  * about required keys.
737  *
738  * Note: the reason we have to copy the preprocessed scan keys into private
739  * storage is that we are modifying the array based on comparisons of the
740  * key argument values, which could change on a rescan or after moving to
741  * new elements of array keys. Therefore we can't overwrite the source data.
742  */
743 void
744 _bt_preprocess_keys(IndexScanDesc scan)
745 {
746  BTScanOpaque so = (BTScanOpaque) scan->opaque;
747  int numberOfKeys = scan->numberOfKeys;
748  int16 *indoption = scan->indexRelation->rd_indoption;
749  int new_numberOfKeys;
750  int numberOfEqualCols;
751  ScanKey inkeys;
752  ScanKey outkeys;
753  ScanKey cur;
754  ScanKey xform[BTMaxStrategyNumber];
755  bool test_result;
756  int i,
757  j;
758  AttrNumber attno;
759 
760  /* initialize result variables */
761  so->qual_ok = true;
762  so->numberOfKeys = 0;
763 
764  if (numberOfKeys < 1)
765  return; /* done if qual-less scan */
766 
767  /*
768  * Read so->arrayKeyData if array keys are present, else scan->keyData
769  */
770  if (so->arrayKeyData != NULL)
771  inkeys = so->arrayKeyData;
772  else
773  inkeys = scan->keyData;
774 
775  outkeys = so->keyData;
776  cur = &inkeys[0];
777  /* we check that input keys are correctly ordered */
778  if (cur->sk_attno < 1)
779  elog(ERROR, "btree index keys must be ordered by attribute");
780 
781  /* We can short-circuit most of the work if there's just one key */
782  if (numberOfKeys == 1)
783  {
784  /* Apply indoption to scankey (might change sk_strategy!) */
785  if (!_bt_fix_scankey_strategy(cur, indoption))
786  so->qual_ok = false;
787  memcpy(outkeys, cur, sizeof(ScanKeyData));
788  so->numberOfKeys = 1;
789  /* We can mark the qual as required if it's for first index col */
790  if (cur->sk_attno == 1)
791  _bt_mark_scankey_required(outkeys);
792  return;
793  }
794 
795  /*
796  * Otherwise, do the full set of pushups.
797  */
798  new_numberOfKeys = 0;
799  numberOfEqualCols = 0;
800 
801  /*
802  * Initialize for processing of keys for attr 1.
803  *
804  * xform[i] points to the currently best scan key of strategy type i+1; it
805  * is NULL if we haven't yet found such a key for this attr.
806  */
807  attno = 1;
808  memset(xform, 0, sizeof(xform));
809 
810  /*
811  * Loop iterates from 0 to numberOfKeys inclusive; we use the last pass to
812  * handle after-last-key processing. Actual exit from the loop is at the
813  * "break" statement below.
814  */
815  for (i = 0;; cur++, i++)
816  {
817  if (i < numberOfKeys)
818  {
819  /* Apply indoption to scankey (might change sk_strategy!) */
820  if (!_bt_fix_scankey_strategy(cur, indoption))
821  {
822  /* NULL can't be matched, so give up */
823  so->qual_ok = false;
824  return;
825  }
826  }
827 
828  /*
829  * If we are at the end of the keys for a particular attr, finish up
830  * processing and emit the cleaned-up keys.
831  */
832  if (i == numberOfKeys || cur->sk_attno != attno)
833  {
834  int priorNumberOfEqualCols = numberOfEqualCols;
835 
836  /* check input keys are correctly ordered */
837  if (i < numberOfKeys && cur->sk_attno < attno)
838  elog(ERROR, "btree index keys must be ordered by attribute");
839 
840  /*
841  * If = has been specified, all other keys can be eliminated as
842  * redundant. If we have a case like key = 1 AND key > 2, we can
843  * set qual_ok to false and abandon further processing.
844  *
845  * We also have to deal with the case of "key IS NULL", which is
846  * unsatisfiable in combination with any other index condition. By
847  * the time we get here, that's been classified as an equality
848  * check, and we've rejected any combination of it with a regular
849  * equality condition; but not with other types of conditions.
850  */
851  if (xform[BTEqualStrategyNumber - 1])
852  {
853  ScanKey eq = xform[BTEqualStrategyNumber - 1];
854 
855  for (j = BTMaxStrategyNumber; --j >= 0;)
856  {
857  ScanKey chk = xform[j];
858 
859  if (!chk || j == (BTEqualStrategyNumber - 1))
860  continue;
861 
862  if (eq->sk_flags & SK_SEARCHNULL)
863  {
864  /* IS NULL is contradictory to anything else */
865  so->qual_ok = false;
866  return;
867  }
868 
869  if (_bt_compare_scankey_args(scan, chk, eq, chk,
870  &test_result))
871  {
872  if (!test_result)
873  {
874  /* keys proven mutually contradictory */
875  so->qual_ok = false;
876  return;
877  }
878  /* else discard the redundant non-equality key */
879  xform[j] = NULL;
880  }
881  /* else, cannot determine redundancy, keep both keys */
882  }
883  /* track number of attrs for which we have "=" keys */
884  numberOfEqualCols++;
885  }
886 
887  /* try to keep only one of <, <= */
888  if (xform[BTLessStrategyNumber - 1]
889  && xform[BTLessEqualStrategyNumber - 1])
890  {
891  ScanKey lt = xform[BTLessStrategyNumber - 1];
892  ScanKey le = xform[BTLessEqualStrategyNumber - 1];
893 
894  if (_bt_compare_scankey_args(scan, le, lt, le,
895  &test_result))
896  {
897  if (test_result)
898  xform[BTLessEqualStrategyNumber - 1] = NULL;
899  else
900  xform[BTLessStrategyNumber - 1] = NULL;
901  }
902  }
903 
904  /* try to keep only one of >, >= */
905  if (xform[BTGreaterStrategyNumber - 1]
906  && xform[BTGreaterEqualStrategyNumber - 1])
907  {
908  ScanKey gt = xform[BTGreaterStrategyNumber - 1];
909  ScanKey ge = xform[BTGreaterEqualStrategyNumber - 1];
910 
911  if (_bt_compare_scankey_args(scan, ge, gt, ge,
912  &test_result))
913  {
914  if (test_result)
915  xform[BTGreaterEqualStrategyNumber - 1] = NULL;
916  else
917  xform[BTGreaterStrategyNumber - 1] = NULL;
918  }
919  }
920 
921  /*
922  * Emit the cleaned-up keys into the outkeys[] array, and then
923  * mark them if they are required. They are required (possibly
924  * only in one direction) if all attrs before this one had "=".
925  */
926  for (j = BTMaxStrategyNumber; --j >= 0;)
927  {
928  if (xform[j])
929  {
930  ScanKey outkey = &outkeys[new_numberOfKeys++];
931 
932  memcpy(outkey, xform[j], sizeof(ScanKeyData));
933  if (priorNumberOfEqualCols == attno - 1)
934  _bt_mark_scankey_required(outkey);
935  }
936  }
937 
938  /*
939  * Exit loop here if done.
940  */
941  if (i == numberOfKeys)
942  break;
943 
944  /* Re-initialize for new attno */
945  attno = cur->sk_attno;
946  memset(xform, 0, sizeof(xform));
947  }
948 
949  /* check strategy this key's operator corresponds to */
950  j = cur->sk_strategy - 1;
951 
952  /* if row comparison, push it directly to the output array */
953  if (cur->sk_flags & SK_ROW_HEADER)
954  {
955  ScanKey outkey = &outkeys[new_numberOfKeys++];
956 
957  memcpy(outkey, cur, sizeof(ScanKeyData));
958  if (numberOfEqualCols == attno - 1)
959  _bt_mark_scankey_required(outkey);
960 
961  /*
962  * We don't support RowCompare using equality; such a qual would
963  * mess up the numberOfEqualCols tracking.
964  */
965  Assert(j != (BTEqualStrategyNumber - 1));
966  continue;
967  }
968 
969  /* have we seen one of these before? */
970  if (xform[j] == NULL)
971  {
972  /* nope, so remember this scankey */
973  xform[j] = cur;
974  }
975  else
976  {
977  /* yup, keep only the more restrictive key */
978  if (_bt_compare_scankey_args(scan, cur, cur, xform[j],
979  &test_result))
980  {
981  if (test_result)
982  xform[j] = cur;
983  else if (j == (BTEqualStrategyNumber - 1))
984  {
985  /* key == a && key == b, but a != b */
986  so->qual_ok = false;
987  return;
988  }
989  /* else old key is more restrictive, keep it */
990  }
991  else
992  {
993  /*
994  * We can't determine which key is more restrictive. Keep the
995  * previous one in xform[j] and push this one directly to the
996  * output array.
997  */
998  ScanKey outkey = &outkeys[new_numberOfKeys++];
999 
1000  memcpy(outkey, cur, sizeof(ScanKeyData));
1001  if (numberOfEqualCols == attno - 1)
1002  _bt_mark_scankey_required(outkey);
1003  }
1004  }
1005  }
1006 
1007  so->numberOfKeys = new_numberOfKeys;
1008 }
1009 
1010 /*
1011  * Compare two scankey values using a specified operator.
1012  *
1013  * The test we want to perform is logically "leftarg op rightarg", where
1014  * leftarg and rightarg are the sk_argument values in those ScanKeys, and
1015  * the comparison operator is the one in the op ScanKey. However, in
1016  * cross-data-type situations we may need to look up the correct operator in
1017  * the index's opfamily: it is the one having amopstrategy = op->sk_strategy
1018  * and amoplefttype/amoprighttype equal to the two argument datatypes.
1019  *
1020  * If the opfamily doesn't supply a complete set of cross-type operators we
1021  * may not be able to make the comparison. If we can make the comparison
1022  * we store the operator result in *result and return true. We return false
1023  * if the comparison could not be made.
1024  *
1025  * Note: op always points at the same ScanKey as either leftarg or rightarg.
1026  * Since we don't scribble on the scankeys, this aliasing should cause no
1027  * trouble.
1028  *
1029  * Note: this routine needs to be insensitive to any DESC option applied
1030  * to the index column. For example, "x < 4" is a tighter constraint than
1031  * "x < 5" regardless of which way the index is sorted.
1032  */
1033 static bool
1034 _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op,
1035  ScanKey leftarg, ScanKey rightarg,
1036  bool *result)
1037 {
1038  Relation rel = scan->indexRelation;
1039  Oid lefttype,
1040  righttype,
1041  optype,
1042  opcintype,
1043  cmp_op;
1044  StrategyNumber strat;
1045 
1046  /*
1047  * First, deal with cases where one or both args are NULL. This should
1048  * only happen when the scankeys represent IS NULL/NOT NULL conditions.
1049  */
1050  if ((leftarg->sk_flags | rightarg->sk_flags) & SK_ISNULL)
1051  {
1052  bool leftnull,
1053  rightnull;
1054 
1055  if (leftarg->sk_flags & SK_ISNULL)
1056  {
1057  Assert(leftarg->sk_flags & (SK_SEARCHNULL | SK_SEARCHNOTNULL));
1058  leftnull = true;
1059  }
1060  else
1061  leftnull = false;
1062  if (rightarg->sk_flags & SK_ISNULL)
1063  {
1064  Assert(rightarg->sk_flags & (SK_SEARCHNULL | SK_SEARCHNOTNULL));
1065  rightnull = true;
1066  }
1067  else
1068  rightnull = false;
1069 
1070  /*
1071  * We treat NULL as either greater than or less than all other values.
1072  * Since true > false, the tests below work correctly for NULLS LAST
1073  * logic. If the index is NULLS FIRST, we need to flip the strategy.
1074  */
1075  strat = op->sk_strategy;
1076  if (op->sk_flags & SK_BT_NULLS_FIRST)
1077  strat = BTCommuteStrategyNumber(strat);
1078 
1079  switch (strat)
1080  {
1081  case BTLessStrategyNumber:
1082  *result = (leftnull < rightnull);
1083  break;
1084  case BTLessEqualStrategyNumber:
1085  *result = (leftnull <= rightnull);
1086  break;
1087  case BTEqualStrategyNumber:
1088  *result = (leftnull == rightnull);
1089  break;
1090  case BTGreaterEqualStrategyNumber:
1091  *result = (leftnull >= rightnull);
1092  break;
1093  case BTGreaterStrategyNumber:
1094  *result = (leftnull > rightnull);
1095  break;
1096  default:
1097  elog(ERROR, "unrecognized StrategyNumber: %d", (int) strat);
1098  *result = false; /* keep compiler quiet */
1099  break;
1100  }
1101  return true;
1102  }
1103 
1104  /*
1105  * The opfamily we need to worry about is identified by the index column.
1106  */
1107  Assert(leftarg->sk_attno == rightarg->sk_attno);
1108 
1109  opcintype = rel->rd_opcintype[leftarg->sk_attno - 1];
1110 
1111  /*
1112  * Determine the actual datatypes of the ScanKey arguments. We have to
1113  * support the convention that sk_subtype == InvalidOid means the opclass
1114  * input type; this is a hack to simplify life for ScanKeyInit().
1115  */
1116  lefttype = leftarg->sk_subtype;
1117  if (lefttype == InvalidOid)
1118  lefttype = opcintype;
1119  righttype = rightarg->sk_subtype;
1120  if (righttype == InvalidOid)
1121  righttype = opcintype;
1122  optype = op->sk_subtype;
1123  if (optype == InvalidOid)
1124  optype = opcintype;
1125 
1126  /*
1127  * If leftarg and rightarg match the types expected for the "op" scankey,
1128  * we can use its already-looked-up comparison function.
1129  */
1130  if (lefttype == opcintype && righttype == optype)
1131  {
1132  *result = DatumGetBool(FunctionCall2Coll(&op->sk_func,
1133  op->sk_collation,
1134  leftarg->sk_argument,
1135  rightarg->sk_argument));
1136  return true;
1137  }
1138 
1139  /*
1140  * Otherwise, we need to go to the syscache to find the appropriate
1141  * operator. (This cannot result in infinite recursion, since no
1142  * indexscan initiated by syscache lookup will use cross-data-type
1143  * operators.)
1144  *
1145  * If the sk_strategy was flipped by _bt_fix_scankey_strategy, we have to
1146  * un-flip it to get the correct opfamily member.
1147  */
1148  strat = op->sk_strategy;
1149  if (op->sk_flags & SK_BT_DESC)
1150  strat = BTCommuteStrategyNumber(strat);
1151 
1152  cmp_op = get_opfamily_member(rel->rd_opfamily[leftarg->sk_attno - 1],
1153  lefttype,
1154  righttype,
1155  strat);
1156  if (OidIsValid(cmp_op))
1157  {
1158  RegProcedure cmp_proc = get_opcode(cmp_op);
1159 
1160  if (RegProcedureIsValid(cmp_proc))
1161  {
1162  *result = DatumGetBool(OidFunctionCall2Coll(cmp_proc,
1163  op->sk_collation,
1164  leftarg->sk_argument,
1165  rightarg->sk_argument));
1166  return true;
1167  }
1168  }
1169 
1170  /* Can't make the comparison */
1171  *result = false; /* suppress compiler warnings */
1172  return false;
1173 }
1174 
1175 /*
1176  * Adjust a scankey's strategy and flags setting as needed for indoptions.
1177  *
1178  * We copy the appropriate indoption value into the scankey sk_flags
1179  * (shifting to avoid clobbering system-defined flag bits). Also, if
1180  * the DESC option is set, commute (flip) the operator strategy number.
1181  *
1182  * A secondary purpose is to check for IS NULL/NOT NULL scankeys and set up
1183  * the strategy field correctly for them.
1184  *
1185  * Lastly, for ordinary scankeys (not IS NULL/NOT NULL), we check for a
1186  * NULL comparison value. Since all btree operators are assumed strict,
1187  * a NULL means that the qual cannot be satisfied. We return true if the
1188  * comparison value isn't NULL, or false if the scan should be abandoned.
1189  *
1190  * This function is applied to the *input* scankey structure; therefore
1191  * on a rescan we will be looking at already-processed scankeys. Hence
1192  * we have to be careful not to re-commute the strategy if we already did it.
1193  * It's a bit ugly to modify the caller's copy of the scankey but in practice
1194  * there shouldn't be any problem, since the index's indoptions are certainly
1195  * not going to change while the scankey survives.
1196  */
1197 static bool
1198 _bt_fix_scankey_strategy(ScanKey skey, int16 *indoption)
1199 {
1200  int addflags;
1201 
1202  addflags = indoption[skey->sk_attno - 1] << SK_BT_INDOPTION_SHIFT;
1203 
1204  /*
1205  * We treat all btree operators as strict (even if they're not so marked
1206  * in pg_proc). This means that it is impossible for an operator condition
1207  * with a NULL comparison constant to succeed, and we can reject it right
1208  * away.
1209  *
1210  * However, we now also support "x IS NULL" clauses as search conditions,
1211  * so in that case keep going. The planner has not filled in any
1212  * particular strategy in this case, so set it to BTEqualStrategyNumber
1213  * --- we can treat IS NULL as an equality operator for purposes of search
1214  * strategy.
1215  *
1216  * Likewise, "x IS NOT NULL" is supported. We treat that as either "less
1217  * than NULL" in a NULLS LAST index, or "greater than NULL" in a NULLS
1218  * FIRST index.
1219  *
1220  * Note: someday we might have to fill in sk_collation from the index
1221  * column's collation. At the moment this is a non-issue because we'll
1222  * never actually call the comparison operator on a NULL.
1223  */
1224  if (skey->sk_flags & SK_ISNULL)
1225  {
1226  /* SK_ISNULL shouldn't be set in a row header scankey */
1227  Assert(!(skey->sk_flags & SK_ROW_HEADER));
1228 
1229  /* Set indoption flags in scankey (might be done already) */
1230  skey->sk_flags |= addflags;
1231 
1232  /* Set correct strategy for IS NULL or NOT NULL search */
1233  if (skey->sk_flags & SK_SEARCHNULL)
1234  {
1235  skey->sk_strategy = BTEqualStrategyNumber;
1236  skey->sk_subtype = InvalidOid;
1237  skey->sk_collation = InvalidOid;
1238  }
1239  else if (skey->sk_flags & SK_SEARCHNOTNULL)
1240  {
1241  if (skey->sk_flags & SK_BT_NULLS_FIRST)
1242  skey->sk_strategy = BTGreaterStrategyNumber;
1243  else
1244  skey->sk_strategy = BTLessStrategyNumber;
1245  skey->sk_subtype = InvalidOid;
1246  skey->sk_collation = InvalidOid;
1247  }
1248  else
1249  {
1250  /* regular qual, so it cannot be satisfied */
1251  return false;
1252  }
1253 
1254  /* Needn't do the rest */
1255  return true;
1256  }
1257 
1258  /* Adjust strategy for DESC, if we didn't already */
1259  if ((addflags & SK_BT_DESC) && !(skey->sk_flags & SK_BT_DESC))
1260  skey->sk_strategy = BTCommuteStrategyNumber(skey->sk_strategy);
1261  skey->sk_flags |= addflags;
1262 
1263  /* If it's a row header, fix row member flags and strategies similarly */
1264  if (skey->sk_flags & SK_ROW_HEADER)
1265  {
1266  ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
1267 
1268  for (;;)
1269  {
1270  Assert(subkey->sk_flags & SK_ROW_MEMBER);
1271  addflags = indoption[subkey->sk_attno - 1] << SK_BT_INDOPTION_SHIFT;
1272  if ((addflags & SK_BT_DESC) && !(subkey->sk_flags & SK_BT_DESC))
1273  subkey->sk_strategy = BTCommuteStrategyNumber(subkey->sk_strategy);
1274  subkey->sk_flags |= addflags;
1275  if (subkey->sk_flags & SK_ROW_END)
1276  break;
1277  subkey++;
1278  }
1279  }
1280 
1281  return true;
1282 }
1283 
1284 /*
1285  * Mark a scankey as "required to continue the scan".
1286  *
1287  * Depending on the operator type, the key may be required for both scan
1288  * directions or just one. Also, if the key is a row comparison header,
1289  * we have to mark its first subsidiary ScanKey as required. (Subsequent
1290  * subsidiary ScanKeys are normally for lower-order columns, and thus
1291  * cannot be required, since they're after the first non-equality scankey.)
1292  *
1293  * Note: when we set required-key flag bits in a subsidiary scankey, we are
1294  * scribbling on a data structure belonging to the index AM's caller, not on
1295  * our private copy. This should be OK because the marking will not change
1296  * from scan to scan within a query, and so we'd just re-mark the same way
1297  * anyway on a rescan. Something to keep an eye on though.
1298  */
1299 static void
1300 _bt_mark_scankey_required(ScanKey skey)
1301 {
1302  int addflags;
1303 
1304  switch (skey->sk_strategy)
1305  {
1306  case BTLessStrategyNumber:
1307  case BTLessEqualStrategyNumber:
1308  addflags = SK_BT_REQFWD;
1309  break;
1310  case BTEqualStrategyNumber:
1311  addflags = SK_BT_REQFWD | SK_BT_REQBKWD;
1312  break;
1313  case BTGreaterEqualStrategyNumber:
1314  case BTGreaterStrategyNumber:
1315  addflags = SK_BT_REQBKWD;
1316  break;
1317  default:
1318  elog(ERROR, "unrecognized StrategyNumber: %d",
1319  (int) skey->sk_strategy);
1320  addflags = 0; /* keep compiler quiet */
1321  break;
1322  }
1323 
1324  skey->sk_flags |= addflags;
1325 
1326  if (skey->sk_flags & SK_ROW_HEADER)
1327  {
1328  ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
1329 
1330  /* First subkey should be same column/operator as the header */
1331  Assert(subkey->sk_flags & SK_ROW_MEMBER);
1332  Assert(subkey->sk_attno == skey->sk_attno);
1333  Assert(subkey->sk_strategy == skey->sk_strategy);
1334  subkey->sk_flags |= addflags;
1335  }
1336 }
1337 
1338 /*
1339  * Test whether an indextuple satisfies all the scankey conditions.
1340  *
1341  * Return true if so, false if not. If the tuple fails to pass the qual,
1342  * we also determine whether there's any need to continue the scan beyond
1343  * this tuple, and set *continuescan accordingly. See comments for
1344  * _bt_preprocess_keys(), above, about how this is done.
1345  *
1346  * Forward scan callers can pass a high key tuple in the hopes of having
1347  * us set *continuescan to false, and avoiding an unnecessary visit to
1348  * the page to the right.
1349  *
1350  * scan: index scan descriptor (containing a search-type scankey)
1351  * tuple: index tuple to test
1352  * tupnatts: number of attributes in tupnatts (high key may be truncated)
1353  * dir: direction we are scanning in
1354  * continuescan: output parameter (will be set correctly in all cases)
1355  */
1356 bool
1357 _bt_checkkeys(IndexScanDesc scan, IndexTuple tuple, int tupnatts,
1358  ScanDirection dir, bool *continuescan)
1359 {
1360  TupleDesc tupdesc;
1361  BTScanOpaque so;
1362  int keysz;
1363  int ikey;
1364  ScanKey key;
1365 
1366  Assert(BTreeTupleGetNAtts(tuple, scan->indexRelation) == tupnatts);
1367 
1368  *continuescan = true; /* default assumption */
1369 
1370  tupdesc = RelationGetDescr(scan->indexRelation);
1371  so = (BTScanOpaque) scan->opaque;
1372  keysz = so->numberOfKeys;
1373 
1374  for (key = so->keyData, ikey = 0; ikey < keysz; key++, ikey++)
1375  {
1376  Datum datum;
1377  bool isNull;
1378  Datum test;
1379 
1380  if (key->sk_attno > tupnatts)
1381  {
1382  /*
1383  * This attribute is truncated (must be high key). The value for
1384  * this attribute in the first non-pivot tuple on the page to the
1385  * right could be any possible value. Assume that truncated
1386  * attribute passes the qual.
1387  */
1388  Assert(ScanDirectionIsForward(dir));
1389  continue;
1390  }
1391 
1392  /* row-comparison keys need special processing */
1393  if (key->sk_flags & SK_ROW_HEADER)
1394  {
1395  if (_bt_check_rowcompare(key, tuple, tupnatts, tupdesc, dir,
1396  continuescan))
1397  continue;
1398  return false;
1399  }
1400 
1401  datum = index_getattr(tuple,
1402  key->sk_attno,
1403  tupdesc,
1404  &isNull);
1405 
1406  if (key->sk_flags & SK_ISNULL)
1407  {
1408  /* Handle IS NULL/NOT NULL tests */
1409  if (key->sk_flags & SK_SEARCHNULL)
1410  {
1411  if (isNull)
1412  continue; /* tuple satisfies this qual */
1413  }
1414  else
1415  {
1416  Assert(key->sk_flags & SK_SEARCHNOTNULL);
1417  if (!isNull)
1418  continue; /* tuple satisfies this qual */
1419  }
1420 
1421  /*
1422  * Tuple fails this qual. If it's a required qual for the current
1423  * scan direction, then we can conclude no further tuples will
1424  * pass, either.
1425  */
1426  if ((key->sk_flags & SK_BT_REQFWD) &&
1427  ScanDirectionIsForward(dir))
1428  *continuescan = false;
1429  else if ((key->sk_flags & SK_BT_REQBKWD) &&
1430  ScanDirectionIsBackward(dir))
1431  *continuescan = false;
1432 
1433  /*
1434  * In any case, this indextuple doesn't match the qual.
1435  */
1436  return false;
1437  }
1438 
1439  if (isNull)
1440  {
1441  if (key->sk_flags & SK_BT_NULLS_FIRST)
1442  {
1443  /*
1444  * Since NULLs are sorted before non-NULLs, we know we have
1445  * reached the lower limit of the range of values for this
1446  * index attr. On a backward scan, we can stop if this qual
1447  * is one of the "must match" subset. We can stop regardless
1448  * of whether the qual is > or <, so long as it's required,
1449  * because it's not possible for any future tuples to pass. On
1450  * a forward scan, however, we must keep going, because we may
1451  * have initially positioned to the start of the index.
1452  */
1453  if ((key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
1454  ScanDirectionIsBackward(dir))
1455  *continuescan = false;
1456  }
1457  else
1458  {
1459  /*
1460  * Since NULLs are sorted after non-NULLs, we know we have
1461  * reached the upper limit of the range of values for this
1462  * index attr. On a forward scan, we can stop if this qual is
1463  * one of the "must match" subset. We can stop regardless of
1464  * whether the qual is > or <, so long as it's required,
1465  * because it's not possible for any future tuples to pass. On
1466  * a backward scan, however, we must keep going, because we
1467  * may have initially positioned to the end of the index.
1468  */
1469  if ((key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
1470  ScanDirectionIsForward(dir))
1471  *continuescan = false;
1472  }
1473 
1474  /*
1475  * In any case, this indextuple doesn't match the qual.
1476  */
1477  return false;
1478  }
1479 
1480  test = FunctionCall2Coll(&key->sk_func, key->sk_collation,
1481  datum, key->sk_argument);
1482 
1483  if (!DatumGetBool(test))
1484  {
1485  /*
1486  * Tuple fails this qual. If it's a required qual for the current
1487  * scan direction, then we can conclude no further tuples will
1488  * pass, either.
1489  *
1490  * Note: because we stop the scan as soon as any required equality
1491  * qual fails, it is critical that equality quals be used for the
1492  * initial positioning in _bt_first() when they are available. See
1493  * comments in _bt_first().
1494  */
1495  if ((key->sk_flags & SK_BT_REQFWD) &&
1496  ScanDirectionIsForward(dir))
1497  *continuescan = false;
1498  else if ((key->sk_flags & SK_BT_REQBKWD) &&
1499  ScanDirectionIsBackward(dir))
1500  *continuescan = false;
1501 
1502  /*
1503  * In any case, this indextuple doesn't match the qual.
1504  */
1505  return false;
1506  }
1507  }
1508 
1509  /* If we get here, the tuple passes all index quals. */
1510  return true;
1511 }
1512 
1513 /*
1514  * Test whether an indextuple satisfies a row-comparison scan condition.
1515  *
1516  * Return true if so, false if not. If not, also clear *continuescan if
1517  * it's not possible for any future tuples in the current scan direction
1518  * to pass the qual.
1519  *
1520  * This is a subroutine for _bt_checkkeys, which see for more info.
1521  */
1522 static bool
1523 _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
1524  TupleDesc tupdesc, ScanDirection dir, bool *continuescan)
1525 {
1526  ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
1527  int32 cmpresult = 0;
1528  bool result;
1529 
1530  /* First subkey should be same as the header says */
1531  Assert(subkey->sk_attno == skey->sk_attno);
1532 
1533  /* Loop over columns of the row condition */
1534  for (;;)
1535  {
1536  Datum datum;
1537  bool isNull;
1538 
1539  Assert(subkey->sk_flags & SK_ROW_MEMBER);
1540 
1541  if (subkey->sk_attno > tupnatts)
1542  {
1543  /*
1544  * This attribute is truncated (must be high key). The value for
1545  * this attribute in the first non-pivot tuple on the page to the
1546  * right could be any possible value. Assume that truncated
1547  * attribute passes the qual.
1548  */
1549  Assert(ScanDirectionIsForward(dir));
1550  cmpresult = 0;
1551  if (subkey->sk_flags & SK_ROW_END)
1552  break;
1553  subkey++;
1554  continue;
1555  }
1556 
1557  datum = index_getattr(tuple,
1558  subkey->sk_attno,
1559  tupdesc,
1560  &isNull);
1561 
1562  if (isNull)
1563  {
1564  if (subkey->sk_flags & SK_BT_NULLS_FIRST)
1565  {
1566  /*
1567  * Since NULLs are sorted before non-NULLs, we know we have
1568  * reached the lower limit of the range of values for this
1569  * index attr. On a backward scan, we can stop if this qual
1570  * is one of the "must match" subset. We can stop regardless
1571  * of whether the qual is > or <, so long as it's required,
1572  * because it's not possible for any future tuples to pass. On
1573  * a forward scan, however, we must keep going, because we may
1574  * have initially positioned to the start of the index.
1575  */
1576  if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
1577  ScanDirectionIsBackward(dir))
1578  *continuescan = false;
1579  }
1580  else
1581  {
1582  /*
1583  * Since NULLs are sorted after non-NULLs, we know we have
1584  * reached the upper limit of the range of values for this
1585  * index attr. On a forward scan, we can stop if this qual is
1586  * one of the "must match" subset. We can stop regardless of
1587  * whether the qual is > or <, so long as it's required,
1588  * because it's not possible for any future tuples to pass. On
1589  * a backward scan, however, we must keep going, because we
1590  * may have initially positioned to the end of the index.
1591  */
1592  if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
1593  ScanDirectionIsForward(dir))
1594  *continuescan = false;
1595  }
1596 
1597  /*
1598  * In any case, this indextuple doesn't match the qual.
1599  */
1600  return false;
1601  }
1602 
1603  if (subkey->sk_flags & SK_ISNULL)
1604  {
1605  /*
1606  * Unlike the simple-scankey case, this isn't a disallowed case.
1607  * But it can never match. If all the earlier row comparison
1608  * columns are required for the scan direction, we can stop the
1609  * scan, because there can't be another tuple that will succeed.
1610  */
1611  if (subkey != (ScanKey) DatumGetPointer(skey->sk_argument))
1612  subkey--;
1613  if ((subkey->sk_flags & SK_BT_REQFWD) &&
1614  ScanDirectionIsForward(dir))
1615  *continuescan = false;
1616  else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
1617  ScanDirectionIsBackward(dir))
1618  *continuescan = false;
1619  return false;
1620  }
1621 
1622  /* Perform the test --- three-way comparison not bool operator */
1623  cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
1624  subkey->sk_collation,
1625  datum,
1626  subkey->sk_argument));
1627 
1628  if (subkey->sk_flags & SK_BT_DESC)
1629  INVERT_COMPARE_RESULT(cmpresult);
1630 
1631  /* Done comparing if unequal, else advance to next column */
1632  if (cmpresult != 0)
1633  break;
1634 
1635  if (subkey->sk_flags & SK_ROW_END)
1636  break;
1637  subkey++;
1638  }
1639 
1640  /*
1641  * At this point cmpresult indicates the overall result of the row
1642  * comparison, and subkey points to the deciding column (or the last
1643  * column if the result is "=").
1644  */
1645  switch (subkey->sk_strategy)
1646  {
1647  /* EQ and NE cases aren't allowed here */
1648  case BTLessStrategyNumber:
1649  result = (cmpresult < 0);
1650  break;
1651  case BTLessEqualStrategyNumber:
1652  result = (cmpresult <= 0);
1653  break;
1654  case BTGreaterEqualStrategyNumber:
1655  result = (cmpresult >= 0);
1656  break;
1657  case BTGreaterStrategyNumber:
1658  result = (cmpresult > 0);
1659  break;
1660  default:
1661  elog(ERROR, "unrecognized RowCompareType: %d",
1662  (int) subkey->sk_strategy);
1663  result = 0; /* keep compiler quiet */
1664  break;
1665  }
1666 
1667  if (!result)
1668  {
1669  /*
1670  * Tuple fails this qual. If it's a required qual for the current
1671  * scan direction, then we can conclude no further tuples will pass,
1672  * either. Note we have to look at the deciding column, not
1673  * necessarily the first or last column of the row condition.
1674  */
1675  if ((subkey->sk_flags & SK_BT_REQFWD) &&
1676  ScanDirectionIsForward(dir))
1677  *continuescan = false;
1678  else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
1679  ScanDirectionIsBackward(dir))
1680  *continuescan = false;
1681  }
1682 
1683  return result;
1684 }
1685 
1686 /*
1687  * _bt_killitems - set LP_DEAD state for items an indexscan caller has
1688  * told us were killed
1689  *
1690  * scan->opaque, referenced locally through so, contains information about the
1691  * current page and killed tuples thereon (generally, this should only be
1692  * called if so->numKilled > 0).
1693  *
1694  * The caller does not have a lock on the page and may or may not have the
1695  * page pinned in a buffer. Note that read-lock is sufficient for setting
1696  * LP_DEAD status (which is only a hint).
1697  *
1698  * We match items by heap TID before assuming they are the right ones to
1699  * delete. We cope with cases where items have moved right due to insertions.
1700  * If an item has moved off the current page due to a split, we'll fail to
1701  * find it and do nothing (this is not an error case --- we assume the item
1702  * will eventually get marked in a future indexscan).
1703  *
1704  * Note that if we hold a pin on the target page continuously from initially
1705  * reading the items until applying this function, VACUUM cannot have deleted
1706  * any items from the page, and so there is no need to search left from the
1707  * recorded offset. (This observation also guarantees that the item is still
1708  * the right one to delete, which might otherwise be questionable since heap
1709  * TIDs can get recycled.) This holds true even if the page has been modified
1710  * by inserts and page splits, so there is no need to consult the LSN.
1711  *
1712  * If the pin was released after reading the page, then we re-read it. If it
1713  * has been modified since we read it (as determined by the LSN), we dare not
1714  * flag any entries because it is possible that the old entry was vacuumed
1715  * away and the TID was re-used by a completely different heap tuple.
1716  */
1717 void
1718 _bt_killitems(IndexScanDesc scan)
1719 {
1720  BTScanOpaque so = (BTScanOpaque) scan->opaque;
1721  Page page;
1722  BTPageOpaque opaque;
1723  OffsetNumber minoff;
1724  OffsetNumber maxoff;
1725  int i;
1726  int numKilled = so->numKilled;
1727  bool killedsomething = false;
1728 
1729  Assert(BTScanPosIsValid(so->currPos));
1730 
1731  /*
1732  * Always reset the scan state, so we don't look for same items on other
1733  * pages.
1734  */
1735  so->numKilled = 0;
1736 
1737  if (BTScanPosIsPinned(so->currPos))
1738  {
1739  /*
1740  * We have held the pin on this page since we read the index tuples,
1741  * so all we need to do is lock it. The pin will have prevented
1742  * re-use of any TID on the page, so there is no need to check the
1743  * LSN.
1744  */
1745  LockBuffer(so->currPos.buf, BT_READ);
1746 
1747  page = BufferGetPage(so->currPos.buf);
1748  }
1749  else
1750  {
1751  Buffer buf;
1752 
1753  /* Attempt to re-read the buffer, getting pin and lock. */
1754  buf = _bt_getbuf(scan->indexRelation, so->currPos.currPage, BT_READ);
1755 
1756  /* It might not exist anymore; in which case we can't hint it. */
1757  if (!BufferIsValid(buf))
1758  return;
1759 
1760  page = BufferGetPage(buf);
1761  if (BufferGetLSNAtomic(buf) == so->currPos.lsn)
1762  so->currPos.buf = buf;
1763  else
1764  {
1765  /* Modified while not pinned means hinting is not safe. */
1766  _bt_relbuf(scan->indexRelation, buf);
1767  return;
1768  }
1769  }
1770 
1771  opaque = (BTPageOpaque) PageGetSpecialPointer(page);
1772  minoff = P_FIRSTDATAKEY(opaque);
1773  maxoff = PageGetMaxOffsetNumber(page);
1774 
1775  for (i = 0; i < numKilled; i++)
1776  {
1777  int itemIndex = so->killedItems[i];
1778  BTScanPosItem *kitem = &so->currPos.items[itemIndex];
1779  OffsetNumber offnum = kitem->indexOffset;
1780 
1781  Assert(itemIndex >= so->currPos.firstItem &&
1782  itemIndex <= so->currPos.lastItem);
1783  if (offnum < minoff)
1784  continue; /* pure paranoia */
1785  while (offnum <= maxoff)
1786  {
1787  ItemId iid = PageGetItemId(page, offnum);
1788  IndexTuple ituple = (IndexTuple) PageGetItem(page, iid);
1789 
1790  if (ItemPointerEquals(&ituple->t_tid, &kitem->heapTid))
1791  {
1792  /* found the item */
1793  ItemIdMarkDead(iid);
1794  killedsomething = true;
1795  break; /* out of inner search loop */
1796  }
1797  offnum = OffsetNumberNext(offnum);
1798  }
1799  }
1800 
1801  /*
1802  * Since this can be redone later if needed, mark as dirty hint.
1803  *
1804  * Whenever we mark anything LP_DEAD, we also set the page's
1805  * BTP_HAS_GARBAGE flag, which is likewise just a hint.
1806  */
1807  if (killedsomething)
1808  {
1809  opaque->btpo_flags |= BTP_HAS_GARBAGE;
1810  MarkBufferDirtyHint(so->currPos.buf, true);
1811  }
1812 
1813  LockBuffer(so->currPos.buf, BUFFER_LOCK_UNLOCK);
1814 }
1815 
1816 
1817 /*
1818  * The following routines manage a shared-memory area in which we track
1819  * assignment of "vacuum cycle IDs" to currently-active btree vacuuming
1820  * operations. There is a single counter which increments each time we
1821  * start a vacuum to assign it a cycle ID. Since multiple vacuums could
1822  * be active concurrently, we have to track the cycle ID for each active
1823  * vacuum; this requires at most MaxBackends entries (usually far fewer).
1824  * We assume at most one vacuum can be active for a given index.
1825  *
1826  * Access to the shared memory area is controlled by BtreeVacuumLock.
1827  * In principle we could use a separate lmgr locktag for each index,
1828  * but a single LWLock is much cheaper, and given the short time that
1829  * the lock is ever held, the concurrency hit should be minimal.
1830  */
1831 
1832 typedef struct BTOneVacInfo
1833 {
1834  LockRelId relid; /* global identifier of an index */
1835  BTCycleId cycleid; /* cycle ID for its active VACUUM */
1836 } BTOneVacInfo;
1837 
1838 typedef struct BTVacInfo
1839 {
1840  BTCycleId cycle_ctr; /* cycle ID most recently assigned */
1841  int num_vacuums; /* number of currently active VACUUMs */
1842  int max_vacuums; /* allocated length of vacuums[] array */
1843  BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER];
1844 } BTVacInfo;
1845 
1846 static BTVacInfo *btvacinfo;
1847 
1848 
1849 /*
1850  * _bt_vacuum_cycleid --- get the active vacuum cycle ID for an index,
1851  * or zero if there is no active VACUUM
1852  *
1853  * Note: for correct interlocking, the caller must already hold pin and
1854  * exclusive lock on each buffer it will store the cycle ID into. This
1855  * ensures that even if a VACUUM starts immediately afterwards, it cannot
1856  * process those pages until the page split is complete.
1857  */
1858 BTCycleId
1859 _bt_vacuum_cycleid(Relation rel)
1860 {
1861  BTCycleId result = 0;
1862  int i;
1863 
1864  /* Share lock is enough since this is a read-only operation */
1865  LWLockAcquire(BtreeVacuumLock, LW_SHARED);
1866 
1867  for (i = 0; i < btvacinfo->num_vacuums; i++)
1868  {
1869  BTOneVacInfo *vac = &btvacinfo->vacuums[i];
1870 
1871  if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
1872  vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
1873  {
1874  result = vac->cycleid;
1875  break;
1876  }
1877  }
1878 
1879  LWLockRelease(BtreeVacuumLock);
1880  return result;
1881 }
1882 
1883 /*
1884  * _bt_start_vacuum --- assign a cycle ID to a just-starting VACUUM operation
1885  *
1886  * Note: the caller must guarantee that it will eventually call
1887  * _bt_end_vacuum, else we'll permanently leak an array slot. To ensure
1888  * that this happens even in elog(FATAL) scenarios, the appropriate coding
1889  * is not just a PG_TRY, but
1890  * PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel))
1891  */
1892 BTCycleId
1893 _bt_start_vacuum(Relation rel)
1894 {
1895  BTCycleId result;
1896  int i;
1897  BTOneVacInfo *vac;
1898 
1899  LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
1900 
1901  /*
1902  * Assign the next cycle ID, being careful to avoid zero as well as the
1903  * reserved high values.
1904  */
1905  result = ++(btvacinfo->cycle_ctr);
1906  if (result == 0 || result > MAX_BT_CYCLE_ID)
1907  result = btvacinfo->cycle_ctr = 1;
1908 
1909  /* Let's just make sure there's no entry already for this index */
1910  for (i = 0; i < btvacinfo->num_vacuums; i++)
1911  {
1912  vac = &btvacinfo->vacuums[i];
1913  if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
1914  vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
1915  {
1916  /*
1917  * Unlike most places in the backend, we have to explicitly
1918  * release our LWLock before throwing an error. This is because
1919  * we expect _bt_end_vacuum() to be called before transaction
1920  * abort cleanup can run to release LWLocks.
1921  */
1922  LWLockRelease(BtreeVacuumLock);
1923  elog(ERROR, "multiple active vacuums for index \"%s\"",
1924  RelationGetRelationName(rel));
1925  }
1926  }
1927 
1928  /* OK, add an entry */
1929  if (btvacinfo->num_vacuums >= btvacinfo->max_vacuums)
1930  {
1931  LWLockRelease(BtreeVacuumLock);
1932  elog(ERROR, "out of btvacinfo slots");
1933  }
1934  vac = &btvacinfo->vacuums[btvacinfo->num_vacuums];
1935  vac->relid = rel->rd_lockInfo.lockRelId;
1936  vac->cycleid = result;
1937  btvacinfo->num_vacuums++;
1938 
1939  LWLockRelease(BtreeVacuumLock);
1940  return result;
1941 }
1942 
1943 /*
1944  * _bt_end_vacuum --- mark a btree VACUUM operation as done
1945  *
1946  * Note: this is deliberately coded not to complain if no entry is found;
1947  * this allows the caller to put PG_TRY around the start_vacuum operation.
1948  */
1949 void
1950 _bt_end_vacuum(Relation rel)
1951 {
1952  int i;
1953 
1954  LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
1955 
1956  /* Find the array entry */
1957  for (i = 0; i < btvacinfo->num_vacuums; i++)
1958  {
1959  BTOneVacInfo *vac = &btvacinfo->vacuums[i];
1960 
1961  if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
1962  vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
1963  {
1964  /* Remove it by shifting down the last entry */
1965  *vac = btvacinfo->vacuums[btvacinfo->num_vacuums - 1];
1966  btvacinfo->num_vacuums--;
1967  break;
1968  }
1969  }
1970 
1971  LWLockRelease(BtreeVacuumLock);
1972 }
1973 
1974 /*
1975  * _bt_end_vacuum wrapped as an on_shmem_exit callback function
1976  */
1977 void
1978 _bt_end_vacuum_callback(int code, Datum arg)
1979 {
1980  _bt_end_vacuum((Relation) DatumGetPointer(arg));
1981 }
1982 
1983 /*
1984  * BTreeShmemSize --- report amount of shared memory space needed
1985  */
1986 Size
1987 BTreeShmemSize(void)
1988 {
1989  Size size;
1990 
1991  size = offsetof(BTVacInfo, vacuums);
1992  size = add_size(size, mul_size(MaxBackends, sizeof(BTOneVacInfo)));
1993  return size;
1994 }
1995 
1996 /*
1997  * BTreeShmemInit --- initialize this module's shared memory
1998  */
1999 void
2000 BTreeShmemInit(void)
2001 {
2002  bool found;
2003 
2004  btvacinfo = (BTVacInfo *) ShmemInitStruct("BTree Vacuum State",
2005  BTreeShmemSize(),
2006  &found);
2007 
2008  if (!IsUnderPostmaster)
2009  {
2010  /* Initialize shared memory area */
2011  Assert(!found);
2012 
2013  /*
2014  * It doesn't really matter what the cycle counter starts at, but
2015  * having it always start the same doesn't seem good. Seed with
2016  * low-order bits of time() instead.
2017  */
2018  btvacinfo->cycle_ctr = (BTCycleId) time(NULL);
2019 
2020  btvacinfo->num_vacuums = 0;
2021  btvacinfo->max_vacuums = MaxBackends;
2022  }
2023  else
2024  Assert(found);
2025 }
2026 
2027 bytea *
2028 btoptions(Datum reloptions, bool validate)
2029 {
2030  return default_reloptions(reloptions, validate, RELOPT_KIND_BTREE);
2031 }
2032 
2033 /*
2034  * btproperty() -- Check boolean properties of indexes.
2035  *
2036  * This is optional, but handling AMPROP_RETURNABLE here saves opening the rel
2037  * to call btcanreturn.
2038  */
2039 bool
2040 btproperty(Oid index_oid, int attno,
2041  IndexAMProperty prop, const char *propname,
2042  bool *res, bool *isnull)
2043 {
2044  switch (prop)
2045  {
2046  case AMPROP_RETURNABLE:
2047  /* answer only for columns, not AM or whole index */
2048  if (attno == 0)
2049  return false;
2050  /* otherwise, btree can always return data */
2051  *res = true;
2052  return true;
2053 
2054  default:
2055  return false; /* punt to generic code */
2056  }
2057 }
2058 
2059 /*
2060  * btbuildphasename() -- Return name of index build phase.
2061  */
2062 char *
2063 btbuildphasename(int64 phasenum)
2064 {
2065  switch (phasenum)
2066  {
2067  case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
2068  return "initializing";
2069  case PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN:
2070  return "scanning table";
2071  case PROGRESS_BTREE_PHASE_PERFORMSORT_1:
2072  return "sorting live tuples";
2073  case PROGRESS_BTREE_PHASE_PERFORMSORT_2:
2074  return "sorting dead tuples";
2075  case PROGRESS_BTREE_PHASE_LEAF_LOAD:
2076  return "loading tuples in tree";
2077  default:
2078  return NULL;
2079  }
2080 }
2081 
2082 /*
2083  * _bt_truncate() -- create tuple without unneeded suffix attributes.
2084  *
2085  * Returns truncated pivot index tuple allocated in caller's memory context,
2086  * with key attributes copied from caller's firstright argument. If rel is
2087  * an INCLUDE index, non-key attributes will definitely be truncated away,
2088  * since they're not part of the key space. More aggressive suffix
2089  * truncation can take place when it's clear that the returned tuple does not
2090  * need one or more suffix key attributes. We only need to keep firstright
2091  * attributes up to and including the first non-lastleft-equal attribute.
2092  * Caller's insertion scankey is used to compare the tuples; the scankey's
2093  * argument values are not considered here.
2094  *
2095  * Note that returned tuple's t_tid offset will hold the number of attributes
2096  * present, so the original item pointer offset is not represented. Caller
2097  * should only change truncated tuple's downlink. Note also that truncated
2098  * key attributes are treated as containing "minus infinity" values by
2099  * _bt_compare().
2100  *
2101  * In the worst case (when a heap TID must be appended to distinguish lastleft
2102  * from firstright), the size of the returned tuple is the size of firstright
2103  * plus the size of an additional MAXALIGN()'d item pointer. This guarantee
2104  * is important, since callers need to stay under the 1/3 of a page
2105  * restriction on tuple size. If this routine is ever taught to truncate
2106  * within an attribute/datum, it will need to avoid returning an enlarged
2107  * tuple to caller when truncation + TOAST compression ends up enlarging the
2108  * final datum.
2109  */
2110 IndexTuple
2111 _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright,
2112  BTScanInsert itup_key)
2113 {
2114  TupleDesc itupdesc = RelationGetDescr(rel);
2115  int16 natts = IndexRelationGetNumberOfAttributes(rel);
2116  int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
2117  int keepnatts;
2118  IndexTuple pivot;
2119  ItemPointer pivotheaptid;
2120  Size newsize;
2121 
2122  /*
2123  * We should only ever truncate leaf index tuples. It's never okay to
2124  * truncate a second time.
2125  */
2126  Assert(BTreeTupleGetNAtts(lastleft, rel) == natts);
2127  Assert(BTreeTupleGetNAtts(firstright, rel) == natts);
2128 
2129  /* Determine how many attributes must be kept in truncated tuple */
2130  keepnatts = _bt_keep_natts(rel, lastleft, firstright, itup_key);
2131 
2132 #ifdef DEBUG_NO_TRUNCATE
2133  /* Force truncation to be ineffective for testing purposes */
2134  keepnatts = nkeyatts + 1;
2135 #endif
2136 
2137  if (keepnatts <= natts)
2138  {
2139  IndexTuple tidpivot;
2140 
2141  pivot = index_truncate_tuple(itupdesc, firstright, keepnatts);
2142 
2143  /*
2144  * If there is a distinguishing key attribute within new pivot tuple,
2145  * there is no need to add an explicit heap TID attribute
2146  */
2147  if (keepnatts <= nkeyatts)
2148  {
2149  BTreeTupleSetNAtts(pivot, keepnatts);
2150  return pivot;
2151  }
2152 
2153  /*
2154  * Only truncation of non-key attributes was possible, since key
2155  * attributes are all equal. It's necessary to add a heap TID
2156  * attribute to the new pivot tuple.
2157  */
2158  Assert(natts != nkeyatts);
2159  newsize = IndexTupleSize(pivot) + MAXALIGN(sizeof(ItemPointerData));
2160  tidpivot = palloc0(newsize);
2161  memcpy(tidpivot, pivot, IndexTupleSize(pivot));
2162  /* cannot leak memory here */
2163  pfree(pivot);
2164  pivot = tidpivot;
2165  }
2166  else
2167  {
2168  /*
2169  * No truncation was possible, since key attributes are all equal.
2170  * It's necessary to add a heap TID attribute to the new pivot tuple.
2171  */
2172  Assert(natts == nkeyatts);
2173  newsize = IndexTupleSize(firstright) + MAXALIGN(sizeof(ItemPointerData));
2174  pivot = palloc0(newsize);
2175  memcpy(pivot, firstright, IndexTupleSize(firstright));
2176  }
2177 
2178  /*
2179  * We have to use heap TID as a unique-ifier in the new pivot tuple, since
2180  * no non-TID key attribute in the right item readily distinguishes the
2181  * right side of the split from the left side. Use enlarged space that
2182  * holds a copy of first right tuple; place a heap TID value within the
2183  * extra space that remains at the end.
2184  *
2185  * nbtree conceptualizes this case as an inability to truncate away any
2186  * key attribute. We must use an alternative representation of heap TID
2187  * within pivots because heap TID is only treated as an attribute within
2188  * nbtree (e.g., there is no pg_attribute entry).
2189  */
2190  Assert(itup_key->heapkeyspace);
2191  pivot->t_info &= ~INDEX_SIZE_MASK;
2192  pivot->t_info |= newsize;
2193 
2194  /*
2195  * Lehman & Yao use lastleft as the leaf high key in all cases, but don't
2196  * consider suffix truncation. It seems like a good idea to follow that
2197  * example in cases where no truncation takes place -- use lastleft's heap
2198  * TID. (This is also the closest value to negative infinity that's
2199  * legally usable.)
2200  */
2201  pivotheaptid = (ItemPointer) ((char *) pivot + newsize -
2202  sizeof(ItemPointerData));
2203  ItemPointerCopy(&lastleft->t_tid, pivotheaptid);
2204 
2205  /*
2206  * Lehman and Yao require that the downlink to the right page, which is to
2207  * be inserted into the parent page in the second phase of a page split be
2208  * a strict lower bound on items on the right page, and a non-strict upper
2209  * bound for items on the left page. Assert that heap TIDs follow these
2210  * invariants, since a heap TID value is apparently needed as a
2211  * tiebreaker.
2212  */
2213 #ifndef DEBUG_NO_TRUNCATE
2214  Assert(ItemPointerCompare(&lastleft->t_tid, &firstright->t_tid) < 0);
2215  Assert(ItemPointerCompare(pivotheaptid, &lastleft->t_tid) >= 0);
2216  Assert(ItemPointerCompare(pivotheaptid, &firstright->t_tid) < 0);
2217 #else
2218 
2219  /*
2220  * Those invariants aren't guaranteed to hold for lastleft + firstright
2221  * heap TID attribute values when they're considered here only because
2222  * DEBUG_NO_TRUNCATE is defined (a heap TID is probably not actually
2223  * needed as a tiebreaker). DEBUG_NO_TRUNCATE must therefore use a heap
2224  * TID value that always works as a strict lower bound for items to the
2225  * right. In particular, it must avoid using firstright's leading key
2226  * attribute values along with lastleft's heap TID value when lastleft's
2227  * TID happens to be greater than firstright's TID.
2228  */
2229  ItemPointerCopy(&firstright->t_tid, pivotheaptid);
2230 
2231  /*
2232  * Pivot heap TID should never be fully equal to firstright. Note that
2233  * the pivot heap TID will still end up equal to lastleft's heap TID when
2234  * that's the only usable value.
2235  */
2236  ItemPointerSetOffsetNumber(pivotheaptid,
2237  OffsetNumberPrev(ItemPointerGetOffsetNumber(pivotheaptid)));
2238  Assert(ItemPointerCompare(pivotheaptid, &firstright->t_tid) < 0);
2239 #endif
2240 
2241  BTreeTupleSetNAtts(pivot, nkeyatts);
2242  BTreeTupleSetAltHeapTID(pivot);
2243 
2244  return pivot;
2245 }
2246 
2247 /*
2248  * _bt_keep_natts - how many key attributes to keep when truncating.
2249  *
2250  * Caller provides two tuples that enclose a split point. Caller's insertion
2251  * scankey is used to compare the tuples; the scankey's argument values are
2252  * not considered here.
2253  *
2254  * This can return a number of attributes that is one greater than the
2255  * number of key attributes for the index relation. This indicates that the
2256  * caller must use a heap TID as a unique-ifier in new pivot tuple.
2257  */
2258 static int
2259 _bt_keep_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright,
2260  BTScanInsert itup_key)
2261 {
2262  int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
2263  TupleDesc itupdesc = RelationGetDescr(rel);
2264  int keepnatts;
2265  ScanKey scankey;
2266 
2267  /*
2268  * Be consistent about the representation of BTREE_VERSION 2/3 tuples
2269  * across Postgres versions; don't allow new pivot tuples to have
2270  * truncated key attributes there. _bt_compare() treats truncated key
2271  * attributes as having the value minus infinity, which would break
2272  * searches within !heapkeyspace indexes.
2273  */
2274  if (!itup_key->heapkeyspace)
2275  {
2276  Assert(nkeyatts != IndexRelationGetNumberOfAttributes(rel));
2277  return nkeyatts;
2278  }
2279 
2280  scankey = itup_key->scankeys;
2281  keepnatts = 1;
2282  for (int attnum = 1; attnum <= nkeyatts; attnum++, scankey++)
2283  {
2284  Datum datum1,
2285  datum2;
2286  bool isNull1,
2287  isNull2;
2288 
2289  datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
2290  datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
2291 
2292  if (isNull1 != isNull2)
2293  break;
2294 
2295  if (!isNull1 &&
2296  DatumGetInt32(FunctionCall2Coll(&scankey->sk_func,
2297  scankey->sk_collation,
2298  datum1,
2299  datum2)) != 0)
2300  break;
2301 
2302  keepnatts++;
2303  }
2304 
2305  return keepnatts;
2306 }
2307 
2308 /*
2309  * _bt_keep_natts_fast - fast bitwise variant of _bt_keep_natts.
2310  *
2311  * This is exported so that a candidate split point can have its effect on
2312  * suffix truncation inexpensively evaluated ahead of time when finding a
2313  * split location. A naive bitwise approach to datum comparisons is used to
2314  * save cycles.
2315  *
2316  * The approach taken here usually provides the same answer as _bt_keep_natts
2317  * will (for the same pair of tuples from a heapkeyspace index), since the
2318  * majority of btree opclasses can never indicate that two datums are equal
2319  * unless they're bitwise equal (once detoasted). Similarly, result may
2320  * differ from the _bt_keep_natts result when either tuple has TOASTed datums,
2321  * though this is barely possible in practice.
2322  *
2323  * These issues must be acceptable to callers, typically because they're only
2324  * concerned about making suffix truncation as effective as possible without
2325  * leaving excessive amounts of free space on either side of page split.
2326  * Callers can rely on the fact that attributes considered equal here are
2327  * definitely also equal according to _bt_keep_natts.
2328  */
2329 int
2330 _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
2331 {
2332  TupleDesc itupdesc = RelationGetDescr(rel);
2333  int keysz = IndexRelationGetNumberOfKeyAttributes(rel);
2334  int keepnatts;
2335 
2336  keepnatts = 1;
2337  for (int attnum = 1; attnum <= keysz; attnum++)
2338  {
2339  Datum datum1,
2340  datum2;
2341  bool isNull1,
2342  isNull2;
2343  Form_pg_attribute att;
2344 
2345  datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
2346  datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
2347  att = TupleDescAttr(itupdesc, attnum - 1);
2348 
2349  if (isNull1 != isNull2)
2350  break;
2351 
2352  if (!isNull1 &&
2353  !datumIsEqual(datum1, datum2, att->attbyval, att->attlen))
2354  break;
2355 
2356  keepnatts++;
2357  }
2358 
2359  return keepnatts;
2360 }
2361 
2362 /*
2363  * _bt_check_natts() -- Verify tuple has expected number of attributes.
2364  *
2365  * Returns value indicating if the expected number of attributes were found
2366  * for a particular offset on page. This can be used as a general purpose
2367  * sanity check.
2368  *
2369  * Testing a tuple directly with BTreeTupleGetNAtts() should generally be
2370  * preferred to calling here. That's usually more convenient, and is always
2371  * more explicit. Call here instead when offnum's tuple may be a negative
2372  * infinity tuple that uses the pre-v11 on-disk representation, or when a low
2373  * context check is appropriate. This routine is as strict as possible about
2374  * what is expected on each version of btree.
2375  */
2376 bool
2377 _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
2378 {
2379  int16 natts = IndexRelationGetNumberOfAttributes(rel);
2380  int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
2381  BTPageOpaque opaque = (BTPageOpaque) PageGetSpecialPointer(page);
2382  IndexTuple itup;
2383  int tupnatts;
2384 
2385  /*
2386  * We cannot reliably test a deleted or half-dead page, since they have
2387  * dummy high keys
2388  */
2389  if (P_IGNORE(opaque))
2390  return true;
2391 
2392  Assert(offnum >= FirstOffsetNumber &&
2393  offnum <= PageGetMaxOffsetNumber(page));
2394 
2395  /*
2396  * Mask allocated for number of keys in index tuple must be able to fit
2397  * maximum possible number of index attributes
2398  */
2399  StaticAssertStmt(BT_N_KEYS_OFFSET_MASK >= INDEX_MAX_KEYS,
2400  "BT_N_KEYS_OFFSET_MASK can't fit INDEX_MAX_KEYS");
2401 
2402  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
2403  tupnatts = BTreeTupleGetNAtts(itup, rel);
2404 
2405  if (P_ISLEAF(opaque))
2406  {
2407  if (offnum >= P_FIRSTDATAKEY(opaque))
2408  {
2409  /*
2410  * Non-pivot tuples currently never use alternative heap TID
2411  * representation -- even those within heapkeyspace indexes
2412  */
2413  if ((itup->t_info & INDEX_ALT_TID_MASK) != 0)
2414  return false;
2415 
2416  /*
2417  * Leaf tuples that are not the page high key (non-pivot tuples)
2418  * should never be truncated. (Note that tupnatts must have been
2419  * inferred, rather than coming from an explicit on-disk
2420  * representation.)
2421  */
2422  return tupnatts == natts;
2423  }
2424  else
2425  {
2426  /*
2427  * Rightmost page doesn't contain a page high key, so tuple was
2428  * checked above as ordinary leaf tuple
2429  */
2430  Assert(!P_RIGHTMOST(opaque));
2431 
2432  /*
2433  * !heapkeyspace high key tuple contains only key attributes. Note
2434  * that tupnatts will only have been explicitly represented in
2435  * !heapkeyspace indexes that happen to have non-key attributes.
2436  */
2437  if (!heapkeyspace)
2438  return tupnatts == nkeyatts;
2439 
2440  /* Use generic heapkeyspace pivot tuple handling */
2441  }
2442  }
2443  else /* !P_ISLEAF(opaque) */
2444  {
2445  if (offnum == P_FIRSTDATAKEY(opaque))
2446  {
2447  /*
2448  * The first tuple on any internal page (possibly the first after
2449  * its high key) is its negative infinity tuple. Negative
2450  * infinity tuples are always truncated to zero attributes. They
2451  * are a particular kind of pivot tuple.
2452  */
2453  if (heapkeyspace)
2454  return tupnatts == 0;
2455 
2456  /*
2457  * The number of attributes won't be explicitly represented if the
2458  * negative infinity tuple was generated during a page split that
2459  * occurred with a version of Postgres before v11. There must be
2460  * a problem when there is an explicit representation that is
2461  * non-zero, or when there is no explicit representation and the
2462  * tuple is evidently not a pre-pg_upgrade tuple.
2463  *
2464  * Prior to v11, downlinks always had P_HIKEY as their offset. Use
2465  * that to decide if the tuple is a pre-v11 tuple.
2466  */
2467  return tupnatts == 0 ||
2468  ((itup->t_info & INDEX_ALT_TID_MASK) == 0 &&
2469  ItemPointerGetOffsetNumber(&(itup->t_tid)) == P_HIKEY);
2470  }
2471  else
2472  {
2473  /*
2474  * !heapkeyspace downlink tuple with separator key contains only
2475  * key attributes. Note that tupnatts will only have been
2476  * explicitly represented in !heapkeyspace indexes that happen to
2477  * have non-key attributes.
2478  */
2479  if (!heapkeyspace)
2480  return tupnatts == nkeyatts;
2481 
2482  /* Use generic heapkeyspace pivot tuple handling */
2483  }
2484 
2485  }
2486 
2487  /* Handle heapkeyspace pivot tuples (excluding minus infinity items) */
2488  Assert(heapkeyspace);
2489 
2490  /*
2491  * Explicit representation of the number of attributes is mandatory with
2492  * heapkeyspace index pivot tuples, regardless of whether or not there are
2493  * non-key attributes.
2494  */
2495  if ((itup->t_info & INDEX_ALT_TID_MASK) == 0)
2496  return false;
2497 
2498  /*
2499  * Heap TID is a tiebreaker key attribute, so it cannot be untruncated
2500  * when any other key attribute is truncated
2501  */
2502  if (BTreeTupleGetHeapTID(itup) != NULL && tupnatts != nkeyatts)
2503  return false;
2504 
2505  /*
2506  * Pivot tuple must have at least one untruncated key attribute (minus
2507  * infinity pivot tuples are the only exception). Pivot tuples can never
2508  * represent that there is a value present for a key attribute that
2509  * exceeds pg_index.indnkeyatts for the index.
2510  */
2511  return tupnatts > 0 && tupnatts <= nkeyatts;
2512 }
2513 
2514 /*
2515  *
2516  * _bt_check_third_page() -- check whether tuple fits on a btree page at all.
2517  *
2518  * We actually need to be able to fit three items on every page, so restrict
2519  * any one item to 1/3 the per-page available space. Note that itemsz should
2520  * not include the ItemId overhead.
2521  *
2522  * It might be useful to apply TOAST methods rather than throw an error here.
2523  * Using out of line storage would break assumptions made by suffix truncation
2524  * and by contrib/amcheck, though.
2525  */
2526 void
2527 _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace,
2528  Page page, IndexTuple newtup)
2529 {
2530  Size itemsz;
2531  BTPageOpaque opaque;
2532 
2533  itemsz = MAXALIGN(IndexTupleSize(newtup));
2534 
2535  /* Double check item size against limit */
2536  if (itemsz <= BTMaxItemSize(page))
2537  return;
2538 
2539  /*
2540  * Tuple is probably too large to fit on page, but it's possible that the
2541  * index uses version 2 or version 3, or that page is an internal page, in
2542  * which case a slightly higher limit applies.
2543  */
2544  if (!needheaptidspace && itemsz <= BTMaxItemSizeNoHeapTid(page))
2545  return;
2546 
2547  /*
2548  * Internal page insertions cannot fail here, because that would mean that
2549  * an earlier leaf level insertion that should have failed didn't
2550  */
2551  opaque = (BTPageOpaque) PageGetSpecialPointer(page);
2552  if (!P_ISLEAF(opaque))
2553  elog(ERROR, "cannot insert oversized tuple of size %zu on internal page of index \"%s\"",
2554  itemsz, RelationGetRelationName(rel));
2555 
2556  ereport(ERROR,
2557  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
2558  errmsg("index row size %zu exceeds btree version %u maximum %zu for index \"%s\"",
2559  itemsz,
2560  needheaptidspace ? BTREE_VERSION : BTREE_NOVAC_VERSION,
2561  needheaptidspace ? BTMaxItemSize(page) :
2562  BTMaxItemSizeNoHeapTid(page),
2563  RelationGetRelationName(rel)),
2564  errdetail("Index row references tuple (%u,%u) in relation \"%s\".",
2565  ItemPointerGetBlockNumber(&newtup->t_tid),
2566  ItemPointerGetOffsetNumber(&newtup->t_tid),
2567  RelationGetRelationName(heap)),
2568  errhint("Values larger than 1/3 of a buffer page cannot be indexed.\n"
2569  "Consider a function index of an MD5 hash of the value, "
2570  "or use full text indexing."),
2571  errtableconstraint(heap, RelationGetRelationName(rel))));
2572 }
SK_BT_INDOPTION_SHIFT
#define SK_BT_INDOPTION_SHIFT
Definition: nbtree.h:679
_bt_keep_natts
static int _bt_keep_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright, BTScanInsert itup_key)
Definition: nbtutils.c:2259
ItemPointerCompare
int32 ItemPointerCompare(ItemPointer arg1, ItemPointer arg2)
Definition: itemptr.c:52
int16
signed short int16
Definition: c.h:345
InvalidStrategy
#define InvalidStrategy
Definition: stratnum.h:24
ScanKeyData::sk_subtype
Oid sk_subtype
Definition: skey.h:69
BUFFER_LOCK_UNLOCK
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
FmgrInfo
Definition: fmgr.h:56
_bt_mark_scankey_required
static void _bt_mark_scankey_required(ScanKey skey)
Definition: nbtutils.c:1300
SK_ROW_MEMBER
#define SK_ROW_MEMBER
Definition: skey.h:118
_bt_freestack
void _bt_freestack(BTStack stack)
Definition: nbtutils.c:163
LockInfoData::lockRelId
LockRelId lockRelId
Definition: rel.h:43
_bt_vacuum_cycleid
BTCycleId _bt_vacuum_cycleid(Relation rel)
Definition: nbtutils.c:1859
_bt_advance_array_keys
bool _bt_advance_array_keys(IndexScanDesc scan, ScanDirection dir)
Definition: nbtutils.c:546
AllocSetContextCreate
#define AllocSetContextCreate
Definition: memutils.h:169
PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN
#define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN
Definition: nbtree.h:688
BTORDER_PROC
#define BTORDER_PROC
Definition: nbtree.h:392
RelationData::rd_indoption
int16 * rd_indoption
Definition: rel.h:162
BTreeShmemInit
void BTreeShmemInit(void)
Definition: nbtutils.c:2000
errhint
int errhint(const char *fmt,...)
Definition: elog.c:974
btvacinfo
static BTVacInfo * btvacinfo
Definition: nbtutils.c:1846
BTGreaterStrategyNumber
#define BTGreaterStrategyNumber
Definition: stratnum.h:33
IndexAMProperty
IndexAMProperty
Definition: amapi.h:34
index_getprocinfo
FmgrInfo * index_getprocinfo(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:794
test
static void test(void)
Definition: preproc-autoprep.c:26
P_IGNORE
#define P_IGNORE(opaque)
Definition: nbtree.h:194
ScanDirectionIsForward
#define ScanDirectionIsForward(direction)
Definition: sdir.h:55
BTScanPosData::items
BTScanPosItem items[MaxIndexTuplesPerPage]
Definition: nbtree.h:581
DatumGetInt32
#define DatumGetInt32(X)
Definition: postgres.h:472
ScanKeyEntryInitializeWithInfo
void ScanKeyEntryInitializeWithInfo(ScanKey entry, int flags, AttrNumber attributeNumber, StrategyNumber strategy, Oid subtype, Oid collation, FmgrInfo *finfo, Datum argument)
Definition: scankey.c:101
BTArrayKeyInfo::mark_elem
int mark_elem
Definition: nbtree.h:623
MarkBufferDirtyHint
void MarkBufferDirtyHint(Buffer buffer, bool buffer_std)
Definition: bufmgr.c:3423
RelationGetDescr
#define RelationGetDescr(relation)
Definition: rel.h:442
ItemIdMarkDead
#define ItemIdMarkDead(itemId)
Definition: itemid.h:179
_bt_mkscankey
BTScanInsert _bt_mkscankey(Relation rel, IndexTuple itup)
Definition: nbtutils.c:85
BTScanInsertData::scantid
ItemPointer scantid
Definition: nbtree.h:475
_bt_start_array_keys
void _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
Definition: nbtutils.c:520
_bt_getbuf
Buffer _bt_getbuf(Relation rel, BlockNumber blkno, int access)
Definition: nbtpage.c:757
BTScanOpaqueData::arrayContext
MemoryContext arrayContext
Definition: nbtree.h:642
_bt_sort_array_elements
static int _bt_sort_array_elements(IndexScanDesc scan, ScanKey skey, bool reverse, Datum *elems, int nelems)
Definition: nbtutils.c:430
PROGRESS_BTREE_PHASE_PERFORMSORT_1
#define PROGRESS_BTREE_PHASE_PERFORMSORT_1
Definition: nbtree.h:689
btbuildphasename
char * btbuildphasename(int64 phasenum)
Definition: nbtutils.c:2063
get_typlenbyvalalign
void get_typlenbyvalalign(Oid typid, int16 *typlen, bool *typbyval, char *typalign)
Definition: lsyscache.c:2049
BTREE_VERSION
#define BTREE_VERSION
Definition: nbtree.h:133
BTVacInfo
struct BTVacInfo BTVacInfo
TupleDescAttr
#define TupleDescAttr(tupdesc, i)
Definition: tupdesc.h:92
BTSortArrayContext
struct BTSortArrayContext BTSortArrayContext
P_FIRSTDATAKEY
#define P_FIRSTDATAKEY(opaque)
Definition: nbtree.h:219
ALLOCSET_SMALL_SIZES
#define ALLOCSET_SMALL_SIZES
Definition: memutils.h:201
IndexTupleData::t_tid
ItemPointerData t_tid
Definition: itup.h:37
RegProcedure
regproc RegProcedure
Definition: c.h:505
BTOneVacInfo::relid
LockRelId relid
Definition: nbtutils.c:1834
Min
#define Min(x, y)
Definition: c.h:904
BTOneVacInfo
struct BTOneVacInfo BTOneVacInfo
LockRelId::dbId
Oid dbId
Definition: rel.h:38
datumIsEqual
bool datumIsEqual(Datum value1, Datum value2, bool typByVal, int typLen)
Definition: datum.c:221
BTreeTupleGetHeapTID
#define BTreeTupleGetHeapTID(itup)
Definition: nbtree.h:346
MemoryContextSwitchTo
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
BTMaxItemSizeNoHeapTid
#define BTMaxItemSizeNoHeapTid(page)
Definition: nbtree.h:153
BTSortArrayContext::flinfo
FmgrInfo flinfo
Definition: nbtutils.c:34
BTScanPosData::currPage
BlockNumber currPage
Definition: nbtree.h:552
INDEX_SIZE_MASK
#define INDEX_SIZE_MASK
Definition: itup.h:65
BTVacInfo::cycle_ctr
BTCycleId cycle_ctr
Definition: nbtutils.c:1840
cur
struct cursor * cur
Definition: ecpg.c:28
StrategyNumber
uint16 StrategyNumber
Definition: stratnum.h:22
errcode
int errcode(int sqlerrcode)
Definition: elog.c:570
BTVacInfo::vacuums
BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER]
Definition: nbtutils.c:1843
BTScanPosIsValid
#define BTScanPosIsValid(scanpos)
Definition: nbtree.h:603
BTScanPosItem::indexOffset
OffsetNumber indexOffset
Definition: nbtree.h:543
MemoryContextReset
void MemoryContextReset(MemoryContext context)
Definition: mcxt.c:136
_bt_end_vacuum_callback
void _bt_end_vacuum_callback(int code, Datum arg)
Definition: nbtutils.c:1978
FunctionCall2Coll
Datum FunctionCall2Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2)
Definition: fmgr.c:1150
Oid
unsigned int Oid
Definition: postgres_ext.h:31
BTArrayKeyInfo::elem_values
Datum * elem_values
Definition: nbtree.h:625
BTreeTupleGetNAtts
#define BTreeTupleGetNAtts(itup, rel)
Definition: nbtree.h:327
OidIsValid
#define OidIsValid(objectId)
Definition: c.h:638
PageGetMaxOffsetNumber
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:357
BTreeTupleSetNAtts
#define BTreeTupleSetNAtts(itup, n)
Definition: nbtree.h:336
SK_SEARCHARRAY
#define SK_SEARCHARRAY
Definition: skey.h:120
BTPageOpaque
BTPageOpaqueData * BTPageOpaque
Definition: nbtree.h:68
int32
signed int int32
Definition: c.h:346
BTreeShmemSize
Size BTreeShmemSize(void)
Definition: nbtutils.c:1987
BTLessEqualStrategyNumber
#define BTLessEqualStrategyNumber
Definition: stratnum.h:30
IndexScanDescData::indexRelation
Relation indexRelation
Definition: relscan.h:103
OffsetNumber
uint16 OffsetNumber
Definition: off.h:24
ItemPointer
ItemPointerData * ItemPointer
Definition: itemptr.h:49
LWLockRelease
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
SK_ROW_END
#define SK_ROW_END
Definition: skey.h:119
BTCommuteStrategyNumber
#define BTCommuteStrategyNumber(strat)
Definition: nbtree.h:374
_bt_truncate
IndexTuple _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright, BTScanInsert itup_key)
Definition: nbtutils.c:2111
ScanDirectionIsBackward
#define ScanDirectionIsBackward(direction)
Definition: sdir.h:41
_bt_restore_array_keys
void _bt_restore_array_keys(IndexScanDesc scan)
Definition: nbtutils.c:624
StaticAssertStmt
#define StaticAssertStmt(condition, errmessage)
Definition: c.h:842
errtableconstraint
int errtableconstraint(Relation rel, const char *conname)
Definition: relcache.c:5222
BT_READ
#define BT_READ
Definition: nbtree.h:402
_bt_find_extreme_element
static Datum _bt_find_extreme_element(IndexScanDesc scan, ScanKey skey, StrategyNumber strat, Datum *elems, int nelems)
Definition: nbtutils.c:363
_bt_killitems
void _bt_killitems(IndexScanDesc scan)
Definition: nbtutils.c:1718
pfree
void pfree(void *pointer)
Definition: mcxt.c:1031
compare
static int compare(const void *arg1, const void *arg2)
Definition: geqo_pool.c:145
RelationData::rd_indcollation
Oid * rd_indcollation
Definition: rel.h:168
ERROR
#define ERROR
Definition: elog.h:43
BTCycleId
uint16 BTCycleId
Definition: nbtree.h:27
LockRelId
Definition: rel.h:35
INDEX_ALT_TID_MASK
#define INDEX_ALT_TID_MASK
Definition: nbtree.h:293
BTScanOpaque
BTScanOpaqueData * BTScanOpaque
Definition: nbtree.h:670
ShmemInitStruct
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:372
fmgr_info
void fmgr_info(Oid functionId, FmgrInfo *finfo)
Definition: fmgr.c:124
ScanKeyData::sk_strategy
StrategyNumber sk_strategy
Definition: skey.h:68
PROGRESS_BTREE_PHASE_PERFORMSORT_2
#define PROGRESS_BTREE_PHASE_PERFORMSORT_2
Definition: nbtree.h:690
MaxBackends
int MaxBackends
Definition: globals.c:135
BTArrayKeyInfo::cur_elem
int cur_elem
Definition: nbtree.h:622
BTScanOpaqueData::numArrayKeys
int numArrayKeys
Definition: nbtree.h:637
_bt_end_vacuum
void _bt_end_vacuum(Relation rel)
Definition: nbtutils.c:1950
BufferGetLSNAtomic
XLogRecPtr BufferGetLSNAtomic(Buffer buffer)
Definition: bufmgr.c:2876
RelationData::rd_lockInfo
LockInfoData rd_lockInfo
Definition: rel.h:86
buf
static char * buf
Definition: pg_test_fsync.c:68
ScanKey
ScanKeyData * ScanKey
Definition: skey.h:75
BTScanPosIsPinned
#define BTScanPosIsPinned(scanpos)
Definition: nbtree.h:586
IsUnderPostmaster
bool IsUnderPostmaster
Definition: globals.c:109
RegProcedureIsValid
#define RegProcedureIsValid(p)
Definition: c.h:640
FirstOffsetNumber
#define FirstOffsetNumber
Definition: off.h:27
IndexTuple
IndexTupleData * IndexTuple
Definition: itup.h:53
errdetail
int errdetail(const char *fmt,...)
Definition: elog.c:860
ScanKeyData::sk_func
FmgrInfo sk_func
Definition: skey.h:71
ScanDirection
ScanDirection
Definition: sdir.h:22
get_opfamily_member
Oid get_opfamily_member(Oid opfamily, Oid lefttype, Oid righttype, int16 strategy)
Definition: lsyscache.c:163
DatumGetBool
#define DatumGetBool(X)
Definition: postgres.h:393
_bt_mark_array_keys
void _bt_mark_array_keys(IndexScanDesc scan)
Definition: nbtutils.c:605
IndexRelationGetNumberOfAttributes
#define IndexRelationGetNumberOfAttributes(relation)
Definition: rel.h:428
index_truncate_tuple
IndexTuple index_truncate_tuple(TupleDesc sourceDescriptor, IndexTuple source, int leavenatts)
Definition: indextuple.c:538
RelationGetRelationName
#define RelationGetRelationName(relation)
Definition: rel.h:450
BTOneVacInfo::cycleid
BTCycleId cycleid
Definition: nbtutils.c:1835
Form_pg_attribute
FormData_pg_attribute * Form_pg_attribute
Definition: pg_attribute.h:200
SK_SEARCHNOTNULL
#define SK_SEARCHNOTNULL
Definition: skey.h:122
RelationData::rd_opfamily
Oid * rd_opfamily
Definition: rel.h:158
BTScanPosData::firstItem
int firstItem
Definition: nbtree.h:577
CurrentMemoryContext
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
IndexRelationGetNumberOfKeyAttributes
#define IndexRelationGetNumberOfKeyAttributes(relation)
Definition: rel.h:435
BTREE_NOVAC_VERSION
#define BTREE_NOVAC_VERSION
Definition: nbtree.h:135
SK_ISNULL
#define SK_ISNULL
Definition: skey.h:115
BufferGetPage
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
BTScanOpaqueData::arrayKeyData
ScanKey arrayKeyData
Definition: nbtree.h:636
OidFunctionCall2Coll
Datum OidFunctionCall2Coll(Oid functionId, Oid collation, Datum arg1, Datum arg2)
Definition: fmgr.c:1422
ereport
#define ereport(elevel, rest)
Definition: elog.h:141
BTScanInsertData::pivotsearch
bool pivotsearch
Definition: nbtree.h:474
SK_ROW_HEADER
#define SK_ROW_HEADER
Definition: skey.h:117
_bt_compare_array_elements
static int _bt_compare_array_elements(const void *a, const void *b, void *arg)
Definition: nbtutils.c:498
_bt_fix_scankey_strategy
static bool _bt_fix_scankey_strategy(ScanKey skey, int16 *indoption)
Definition: nbtutils.c:1198
MAX_BT_CYCLE_ID
#define MAX_BT_CYCLE_ID
Definition: nbtree.h:87
_bt_preprocess_keys
void _bt_preprocess_keys(IndexScanDesc scan)
Definition: nbtutils.c:744
BTVacInfo::max_vacuums
int max_vacuums
Definition: nbtutils.c:1842
qsort_arg
void qsort_arg(void *base, size_t nel, size_t elsize, qsort_arg_comparator cmp, void *arg)
Definition: qsort_arg.c:113
PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE
#define PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE
Definition: progress.h:87
PageGetItemId
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
SK_BT_NULLS_FIRST
#define SK_BT_NULLS_FIRST
Definition: nbtree.h:681
BTScanInsertData::anynullkeys
bool anynullkeys
Definition: nbtree.h:472
mul_size
Size mul_size(Size s1, Size s2)
Definition: shmem.c:492
IndexScanDescData::parallel_scan
struct ParallelIndexScanDescData * parallel_scan
Definition: relscan.h:151
palloc0
void * palloc0(Size size)
Definition: mcxt.c:955
Datum
uintptr_t Datum
Definition: postgres.h:367
BTArrayKeyInfo::num_elems
int num_elems
Definition: nbtree.h:624
add_size
Size add_size(Size s1, Size s2)
Definition: shmem.c:475
_bt_check_rowcompare
static bool _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts, TupleDesc tupdesc, ScanDirection dir, bool *continuescan)
Definition: nbtutils.c:1523
LockBuffer
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3590
SK_BT_REQFWD
#define SK_BT_REQFWD
Definition: nbtree.h:677
_bt_heapkeyspace
bool _bt_heapkeyspace(Relation rel)
Definition: nbtpage.c:636
_bt_checkkeys
bool _bt_checkkeys(IndexScanDesc scan, IndexTuple tuple, int tupnatts, ScanDirection dir, bool *continuescan)
Definition: nbtutils.c:1357
InvalidOid
#define InvalidOid
Definition: postgres_ext.h:36
get_opcode
RegProcedure get_opcode(Oid opno)
Definition: lsyscache.c:1092
attnum
int16 attnum
Definition: pg_attribute.h:79
_bt_relbuf
void _bt_relbuf(Relation rel, Buffer buf)
Definition: nbtpage.c:912
_bt_check_natts
bool _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
Definition: nbtutils.c:2377
BTScanOpaqueData::killedItems
int * killedItems
Definition: nbtree.h:645
ScanKeyData::sk_flags
int sk_flags
Definition: skey.h:66
Assert
#define Assert(condition)
Definition: c.h:732
SK_BT_DESC
#define SK_BT_DESC
Definition: nbtree.h:680
get_opfamily_proc
Oid get_opfamily_proc(Oid opfamily, Oid lefttype, Oid righttype, int16 procnum)
Definition: lsyscache.c:744
btproperty
bool btproperty(Oid index_oid, int attno, IndexAMProperty prop, const char *propname, bool *res, bool *isnull)
Definition: nbtutils.c:2040
BTScanInsertData::heapkeyspace
bool heapkeyspace
Definition: nbtree.h:471
ItemPointerData
struct ItemPointerData ItemPointerData
INDEX_MAX_KEYS
#define INDEX_MAX_KEYS
Definition: pg_config_manual.h:52
OffsetNumberNext
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
Size
size_t Size
Definition: c.h:466
PageGetSpecialPointer
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
BT_N_KEYS_OFFSET_MASK
#define BT_N_KEYS_OFFSET_MASK
Definition: nbtree.h:297
OffsetNumberPrev
#define OffsetNumberPrev(offsetNumber)
Definition: off.h:54
BTScanOpaqueData::numberOfKeys
int numberOfKeys
Definition: nbtree.h:632
LWLockAcquire
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
MAXALIGN
#define MAXALIGN(LEN)
Definition: c.h:685
BufferIsValid
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
index_getattr
#define index_getattr(tup, attnum, tupleDesc, isnull)
Definition: itup.h:100
ItemPointerGetOffsetNumber
#define ItemPointerGetOffsetNumber(pointer)
Definition: itemptr.h:117
BTStackData::bts_parent
struct BTStackData * bts_parent
Definition: nbtree.h:419
BTScanInsertData::scankeys
ScanKeyData scankeys[INDEX_MAX_KEYS]
Definition: nbtree.h:477
ItemPointerEquals
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:29
_bt_check_third_page
void _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace, Page page, IndexTuple newtup)
Definition: nbtutils.c:2527
BTreeTupleSetAltHeapTID
#define BTreeTupleSetAltHeapTID(itup)
Definition: nbtree.h:360
PROGRESS_BTREE_PHASE_LEAF_LOAD
#define PROGRESS_BTREE_PHASE_LEAF_LOAD
Definition: nbtree.h:691
BTScanPosItem::heapTid
ItemPointerData heapTid
Definition: nbtree.h:542
DatumGetPointer
#define DatumGetPointer(X)
Definition: postgres.h:549
IndexScanDescData::keyData
struct ScanKeyData * keyData
Definition: relscan.h:107
ItemPointerSetOffsetNumber
#define ItemPointerSetOffsetNumber(pointer, offsetNumber)
Definition: itemptr.h:148
BTMaxItemSize
#define BTMaxItemSize(page)
Definition: nbtree.h:147
P_HIKEY
#define P_HIKEY
Definition: nbtree.h:217
_bt_parallel_advance_array_keys
void _bt_parallel_advance_array_keys(IndexScanDesc scan)
Definition: nbtree.c:763
deconstruct_array
void deconstruct_array(ArrayType *array, Oid elmtype, int elmlen, bool elmbyval, char elmalign, Datum **elemsp, bool **nullsp, int *nelemsp)
Definition: arrayfuncs.c:3461
default_reloptions
bytea * default_reloptions(Datum reloptions, bool validate, relopt_kind kind)
Definition: reloptions.c:1369
BTVacInfo::num_vacuums
int num_vacuums
Definition: nbtutils.c:1841
BTScanOpaqueData::currPos
BTScanPosData currPos
Definition: nbtree.h:666
_bt_keep_natts_fast
int _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
Definition: nbtutils.c:2330
palloc
void * palloc(Size size)
Definition: mcxt.c:924
errmsg
int errmsg(const char *fmt,...)
Definition: elog.c:784
BTScanOpaqueData::keyData
ScanKey keyData
Definition: nbtree.h:633
ScanKeyData::sk_collation
Oid sk_collation
Definition: skey.h:70
elog
#define elog(elevel,...)
Definition: elog.h:226
BTScanOpaqueData::arrayKeys
BTArrayKeyInfo * arrayKeys
Definition: nbtree.h:641
i
int i
Definition: preproc-comment.c:23
_bt_preprocess_array_keys
void _bt_preprocess_array_keys(IndexScanDesc scan)
Definition: nbtutils.c:191
arg
void * arg
Definition: pg_backup_utils.c:27
varlena
Definition: c.h:549
btoptions
bytea * btoptions(Datum reloptions, bool validate)
Definition: nbtutils.c:2028
BTMaxStrategyNumber
#define BTMaxStrategyNumber
Definition: stratnum.h:35
_bt_compare_scankey_args
static bool _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op, ScanKey leftarg, ScanKey rightarg, bool *result)
Definition: nbtutils.c:1034
ItemPointerGetBlockNumber
#define ItemPointerGetBlockNumber(pointer)
Definition: itemptr.h:98
BTScanPosData::lsn
XLogRecPtr lsn
Definition: nbtree.h:551
BTScanPosData::buf
Buffer buf
Definition: nbtree.h:549
IndexTupleData::t_info
unsigned short t_info
Definition: itup.h:49
_bt_start_vacuum
BTCycleId _bt_start_vacuum(Relation rel)
Definition: nbtutils.c:1893
RelationData::rd_opcintype
Oid * rd_opcintype
Definition: rel.h:159
BTLessStrategyNumber
#define BTLessStrategyNumber
Definition: stratnum.h:29
SK_BT_REQBKWD
#define SK_BT_REQBKWD
Definition: nbtree.h:678
Buffer
int Buffer
Definition: buf.h:23
P_RIGHTMOST
#define P_RIGHTMOST(opaque)
Definition: nbtree.h:188
ScanKeyData::sk_argument
Datum sk_argument
Definition: skey.h:72
ARR_ELEMTYPE
#define ARR_ELEMTYPE(a)
Definition: array.h:280
AttrNumber
int16 AttrNumber
Definition: attnum.h:21
BTArrayKeyInfo::scan_key
int scan_key
Definition: nbtree.h:621
INVERT_COMPARE_RESULT
#define INVERT_COMPARE_RESULT(var)
Definition: c.h:1044
SK_SEARCHNULL
#define SK_SEARCHNULL
Definition: skey.h:121
BTEqualStrategyNumber
#define BTEqualStrategyNumber
Definition: stratnum.h:31
offsetof
#define offsetof(type, field)
Definition: c.h:655
BTGreaterEqualStrategyNumber
#define BTGreaterEqualStrategyNumber
Definition: stratnum.h:32
BTP_HAS_GARBAGE
#define BTP_HAS_GARBAGE
Definition: nbtree.h:77
PageGetItem
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
ScanKeyData::sk_attno
AttrNumber sk_attno
Definition: skey.h:67
Page
Pointer Page
Definition: bufpage.h:78
IndexTupleSize
#define IndexTupleSize(itup)
Definition: itup.h:71
ItemPointerCopy
#define ItemPointerCopy(fromPointer, toPointer)
Definition: itemptr.h:161
LockRelId::relId
Oid relId
Definition: rel.h:37
P_ISLEAF
#define P_ISLEAF(opaque)
Definition: nbtree.h:189
DatumGetArrayTypeP
#define DatumGetArrayTypeP(X)
Definition: array.h:249