Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nbtutils.c
4 : * Utility code for Postgres btree implementation.
5 : *
6 : * Portions Copyright (c) 1996-2025, 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/datum.h"
26 : #include "utils/lsyscache.h"
27 : #include "utils/rel.h"
28 :
29 :
30 : #define LOOK_AHEAD_REQUIRED_RECHECKS 3
31 : #define LOOK_AHEAD_DEFAULT_DISTANCE 5
32 : #define NSKIPADVANCES_THRESHOLD 3
33 :
34 : static inline int32 _bt_compare_array_skey(FmgrInfo *orderproc,
35 : Datum tupdatum, bool tupnull,
36 : Datum arrdatum, ScanKey cur);
37 : static void _bt_binsrch_skiparray_skey(bool cur_elem_trig, ScanDirection dir,
38 : Datum tupdatum, bool tupnull,
39 : BTArrayKeyInfo *array, ScanKey cur,
40 : int32 *set_elem_result);
41 : static void _bt_skiparray_set_element(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
42 : int32 set_elem_result, Datum tupdatum, bool tupnull);
43 : static void _bt_skiparray_set_isnull(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
44 : static void _bt_array_set_low_or_high(Relation rel, ScanKey skey,
45 : BTArrayKeyInfo *array, bool low_not_high);
46 : static bool _bt_array_decrement(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
47 : static bool _bt_array_increment(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
48 : static bool _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
49 : bool *skip_array_set);
50 : static bool _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
51 : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
52 : bool readpagetup, int sktrig, bool *scanBehind);
53 : static bool _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
54 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
55 : int sktrig, bool sktrig_required);
56 : #ifdef USE_ASSERT_CHECKING
57 : static bool _bt_verify_keys_with_arraykeys(IndexScanDesc scan);
58 : #endif
59 : static bool _bt_oppodir_checkkeys(IndexScanDesc scan, ScanDirection dir,
60 : IndexTuple finaltup);
61 : static bool _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
62 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
63 : bool advancenonrequired, bool forcenonrequired,
64 : bool *continuescan, int *ikey);
65 : static bool _bt_rowcompare_cmpresult(ScanKey subkey, int cmpresult);
66 : static bool _bt_check_rowcompare(ScanKey skey,
67 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
68 : ScanDirection dir, bool forcenonrequired, bool *continuescan);
69 : static void _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
70 : int tupnatts, TupleDesc tupdesc);
71 : static int _bt_keep_natts(Relation rel, IndexTuple lastleft,
72 : IndexTuple firstright, BTScanInsert itup_key);
73 :
74 :
75 : /*
76 : * _bt_mkscankey
77 : * Build an insertion scan key that contains comparison data from itup
78 : * as well as comparator routines appropriate to the key datatypes.
79 : *
80 : * The result is intended for use with _bt_compare() and _bt_truncate().
81 : * Callers that don't need to fill out the insertion scankey arguments
82 : * (e.g. they use an ad-hoc comparison routine, or only need a scankey
83 : * for _bt_truncate()) can pass a NULL index tuple. The scankey will
84 : * be initialized as if an "all truncated" pivot tuple was passed
85 : * instead.
86 : *
87 : * Note that we may occasionally have to share lock the metapage to
88 : * determine whether or not the keys in the index are expected to be
89 : * unique (i.e. if this is a "heapkeyspace" index). We assume a
90 : * heapkeyspace index when caller passes a NULL tuple, allowing index
91 : * build callers to avoid accessing the non-existent metapage. We
92 : * also assume that the index is _not_ allequalimage when a NULL tuple
93 : * is passed; CREATE INDEX callers call _bt_allequalimage() to set the
94 : * field themselves.
95 : */
96 : BTScanInsert
97 11845518 : _bt_mkscankey(Relation rel, IndexTuple itup)
98 : {
99 : BTScanInsert key;
100 : ScanKey skey;
101 : TupleDesc itupdesc;
102 : int indnkeyatts;
103 : int16 *indoption;
104 : int tupnatts;
105 : int i;
106 :
107 11845518 : itupdesc = RelationGetDescr(rel);
108 11845518 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
109 11845518 : indoption = rel->rd_indoption;
110 11845518 : tupnatts = itup ? BTreeTupleGetNAtts(itup, rel) : 0;
111 :
112 : Assert(tupnatts <= IndexRelationGetNumberOfAttributes(rel));
113 :
114 : /*
115 : * We'll execute search using scan key constructed on key columns.
116 : * Truncated attributes and non-key attributes are omitted from the final
117 : * scan key.
118 : */
119 11845518 : key = palloc(offsetof(BTScanInsertData, scankeys) +
120 11845518 : sizeof(ScanKeyData) * indnkeyatts);
121 11845518 : if (itup)
122 11702482 : _bt_metaversion(rel, &key->heapkeyspace, &key->allequalimage);
123 : else
124 : {
125 : /* Utility statement callers can set these fields themselves */
126 143036 : key->heapkeyspace = true;
127 143036 : key->allequalimage = false;
128 : }
129 11845518 : key->anynullkeys = false; /* initial assumption */
130 11845518 : key->nextkey = false; /* usual case, required by btinsert */
131 11845518 : key->backward = false; /* usual case, required by btinsert */
132 11845518 : key->keysz = Min(indnkeyatts, tupnatts);
133 11845518 : key->scantid = key->heapkeyspace && itup ?
134 23691036 : BTreeTupleGetHeapTID(itup) : NULL;
135 11845518 : skey = key->scankeys;
136 32061892 : for (i = 0; i < indnkeyatts; i++)
137 : {
138 : FmgrInfo *procinfo;
139 : Datum arg;
140 : bool null;
141 : int flags;
142 :
143 : /*
144 : * We can use the cached (default) support procs since no cross-type
145 : * comparison can be needed.
146 : */
147 20216374 : procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
148 :
149 : /*
150 : * Key arguments built from truncated attributes (or when caller
151 : * provides no tuple) are defensively represented as NULL values. They
152 : * should never be used.
153 : */
154 20216374 : if (i < tupnatts)
155 19959812 : arg = index_getattr(itup, i + 1, itupdesc, &null);
156 : else
157 : {
158 256562 : arg = (Datum) 0;
159 256562 : null = true;
160 : }
161 20216374 : flags = (null ? SK_ISNULL : 0) | (indoption[i] << SK_BT_INDOPTION_SHIFT);
162 20216374 : ScanKeyEntryInitializeWithInfo(&skey[i],
163 : flags,
164 20216374 : (AttrNumber) (i + 1),
165 : InvalidStrategy,
166 : InvalidOid,
167 20216374 : rel->rd_indcollation[i],
168 : procinfo,
169 : arg);
170 : /* Record if any key attribute is NULL (or truncated) */
171 20216374 : if (null)
172 277210 : key->anynullkeys = true;
173 : }
174 :
175 : /*
176 : * In NULLS NOT DISTINCT mode, we pretend that there are no null keys, so
177 : * that full uniqueness check is done.
178 : */
179 11845518 : if (rel->rd_index->indnullsnotdistinct)
180 186 : key->anynullkeys = false;
181 :
182 11845518 : return key;
183 : }
184 :
185 : /*
186 : * free a retracement stack made by _bt_search.
187 : */
188 : void
189 23177586 : _bt_freestack(BTStack stack)
190 : {
191 : BTStack ostack;
192 :
193 42470708 : while (stack != NULL)
194 : {
195 19293122 : ostack = stack;
196 19293122 : stack = stack->bts_parent;
197 19293122 : pfree(ostack);
198 : }
199 23177586 : }
200 :
201 : /*
202 : * _bt_compare_array_skey() -- apply array comparison function
203 : *
204 : * Compares caller's tuple attribute value to a scan key/array element.
205 : * Helper function used during binary searches of SK_SEARCHARRAY arrays.
206 : *
207 : * This routine returns:
208 : * <0 if tupdatum < arrdatum;
209 : * 0 if tupdatum == arrdatum;
210 : * >0 if tupdatum > arrdatum.
211 : *
212 : * This is essentially the same interface as _bt_compare: both functions
213 : * compare the value that they're searching for to a binary search pivot.
214 : * However, unlike _bt_compare, this function's "tuple argument" comes first,
215 : * while its "array/scankey argument" comes second.
216 : */
217 : static inline int32
218 465446 : _bt_compare_array_skey(FmgrInfo *orderproc,
219 : Datum tupdatum, bool tupnull,
220 : Datum arrdatum, ScanKey cur)
221 : {
222 465446 : int32 result = 0;
223 :
224 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
225 : Assert(!(cur->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL)));
226 :
227 465446 : if (tupnull) /* NULL tupdatum */
228 : {
229 228 : if (cur->sk_flags & SK_ISNULL)
230 132 : result = 0; /* NULL "=" NULL */
231 96 : else if (cur->sk_flags & SK_BT_NULLS_FIRST)
232 0 : result = -1; /* NULL "<" NOT_NULL */
233 : else
234 96 : result = 1; /* NULL ">" NOT_NULL */
235 : }
236 465218 : else if (cur->sk_flags & SK_ISNULL) /* NOT_NULL tupdatum, NULL arrdatum */
237 : {
238 30540 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
239 54 : result = 1; /* NOT_NULL ">" NULL */
240 : else
241 30486 : result = -1; /* NOT_NULL "<" NULL */
242 : }
243 : else
244 : {
245 : /*
246 : * Like _bt_compare, we need to be careful of cross-type comparisons,
247 : * so the left value has to be the value that came from an index tuple
248 : */
249 434678 : result = DatumGetInt32(FunctionCall2Coll(orderproc, cur->sk_collation,
250 : tupdatum, arrdatum));
251 :
252 : /*
253 : * We flip the sign by following the obvious rule: flip whenever the
254 : * column is a DESC column.
255 : *
256 : * _bt_compare does it the wrong way around (flip when *ASC*) in order
257 : * to compensate for passing its orderproc arguments backwards. We
258 : * don't need to play these games because we find it natural to pass
259 : * tupdatum as the left value (and arrdatum as the right value).
260 : */
261 434678 : if (cur->sk_flags & SK_BT_DESC)
262 45498 : INVERT_COMPARE_RESULT(result);
263 : }
264 :
265 465446 : return result;
266 : }
267 :
268 : /*
269 : * _bt_binsrch_array_skey() -- Binary search for next matching array key
270 : *
271 : * Returns an index to the first array element >= caller's tupdatum argument.
272 : * This convention is more natural for forwards scan callers, but that can't
273 : * really matter to backwards scan callers. Both callers require handling for
274 : * the case where the match we return is < tupdatum, and symmetric handling
275 : * for the case where our best match is > tupdatum.
276 : *
277 : * Also sets *set_elem_result to the result _bt_compare_array_skey returned
278 : * when we used it to compare the matching array element to tupdatum/tupnull.
279 : *
280 : * cur_elem_trig indicates if array advancement was triggered by this array's
281 : * scan key, and that the array is for a required scan key. We can apply this
282 : * information to find the next matching array element in the current scan
283 : * direction using far fewer comparisons (fewer on average, compared to naive
284 : * binary search). This scheme takes advantage of an important property of
285 : * required arrays: required arrays always advance in lockstep with the index
286 : * scan's progress through the index's key space.
287 : */
288 : int
289 32136 : _bt_binsrch_array_skey(FmgrInfo *orderproc,
290 : bool cur_elem_trig, ScanDirection dir,
291 : Datum tupdatum, bool tupnull,
292 : BTArrayKeyInfo *array, ScanKey cur,
293 : int32 *set_elem_result)
294 : {
295 32136 : int low_elem = 0,
296 32136 : mid_elem = -1,
297 32136 : high_elem = array->num_elems - 1,
298 32136 : result = 0;
299 : Datum arrdatum;
300 :
301 : Assert(cur->sk_flags & SK_SEARCHARRAY);
302 : Assert(!(cur->sk_flags & SK_BT_SKIP));
303 : Assert(!(cur->sk_flags & SK_ISNULL)); /* SAOP arrays never have NULLs */
304 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
305 :
306 32136 : if (cur_elem_trig)
307 : {
308 : Assert(!ScanDirectionIsNoMovement(dir));
309 : Assert(cur->sk_flags & SK_BT_REQFWD);
310 :
311 : /*
312 : * When the scan key that triggered array advancement is a required
313 : * array scan key, it is now certain that the current array element
314 : * (plus all prior elements relative to the current scan direction)
315 : * cannot possibly be at or ahead of the corresponding tuple value.
316 : * (_bt_checkkeys must have called _bt_tuple_before_array_skeys, which
317 : * makes sure this is true as a condition of advancing the arrays.)
318 : *
319 : * This makes it safe to exclude array elements up to and including
320 : * the former-current array element from our search.
321 : *
322 : * Separately, when array advancement was triggered by a required scan
323 : * key, the array element immediately after the former-current element
324 : * is often either an exact tupdatum match, or a "close by" near-match
325 : * (a near-match tupdatum is one whose key space falls _between_ the
326 : * former-current and new-current array elements). We'll detect both
327 : * cases via an optimistic comparison of the new search lower bound
328 : * (or new search upper bound in the case of backwards scans).
329 : */
330 31818 : if (ScanDirectionIsForward(dir))
331 : {
332 31758 : low_elem = array->cur_elem + 1; /* old cur_elem exhausted */
333 :
334 : /* Compare prospective new cur_elem (also the new lower bound) */
335 31758 : if (high_elem >= low_elem)
336 : {
337 23692 : arrdatum = array->elem_values[low_elem];
338 23692 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
339 : arrdatum, cur);
340 :
341 23692 : if (result <= 0)
342 : {
343 : /* Optimistic comparison optimization worked out */
344 23606 : *set_elem_result = result;
345 23606 : return low_elem;
346 : }
347 86 : mid_elem = low_elem;
348 86 : low_elem++; /* this cur_elem exhausted, too */
349 : }
350 :
351 8152 : if (high_elem < low_elem)
352 : {
353 : /* Caller needs to perform "beyond end" array advancement */
354 8072 : *set_elem_result = 1;
355 8072 : return high_elem;
356 : }
357 : }
358 : else
359 : {
360 60 : high_elem = array->cur_elem - 1; /* old cur_elem exhausted */
361 :
362 : /* Compare prospective new cur_elem (also the new upper bound) */
363 60 : if (high_elem >= low_elem)
364 : {
365 42 : arrdatum = array->elem_values[high_elem];
366 42 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
367 : arrdatum, cur);
368 :
369 42 : if (result >= 0)
370 : {
371 : /* Optimistic comparison optimization worked out */
372 30 : *set_elem_result = result;
373 30 : return high_elem;
374 : }
375 12 : mid_elem = high_elem;
376 12 : high_elem--; /* this cur_elem exhausted, too */
377 : }
378 :
379 30 : if (high_elem < low_elem)
380 : {
381 : /* Caller needs to perform "beyond end" array advancement */
382 30 : *set_elem_result = -1;
383 30 : return low_elem;
384 : }
385 : }
386 : }
387 :
388 698 : while (high_elem > low_elem)
389 : {
390 438 : mid_elem = low_elem + ((high_elem - low_elem) / 2);
391 438 : arrdatum = array->elem_values[mid_elem];
392 :
393 438 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
394 : arrdatum, cur);
395 :
396 438 : if (result == 0)
397 : {
398 : /*
399 : * It's safe to quit as soon as we see an equal array element.
400 : * This often saves an extra comparison or two...
401 : */
402 138 : low_elem = mid_elem;
403 138 : break;
404 : }
405 :
406 300 : if (result > 0)
407 270 : low_elem = mid_elem + 1;
408 : else
409 30 : high_elem = mid_elem;
410 : }
411 :
412 : /*
413 : * ...but our caller also cares about how its searched-for tuple datum
414 : * compares to the low_elem datum. Must always set *set_elem_result with
415 : * the result of that comparison specifically.
416 : */
417 398 : if (low_elem != mid_elem)
418 242 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
419 242 : array->elem_values[low_elem], cur);
420 :
421 398 : *set_elem_result = result;
422 :
423 398 : return low_elem;
424 : }
425 :
426 : /*
427 : * _bt_binsrch_skiparray_skey() -- "Binary search" within a skip array
428 : *
429 : * Does not return an index into the array, since skip arrays don't really
430 : * contain elements (they generate their array elements procedurally instead).
431 : * Our interface matches that of _bt_binsrch_array_skey in every other way.
432 : *
433 : * Sets *set_elem_result just like _bt_binsrch_array_skey would with a true
434 : * array. The value 0 indicates that tupdatum/tupnull is within the range of
435 : * the skip array. We return -1 when tupdatum/tupnull is lower that any value
436 : * within the range of the array, and 1 when it is higher than every value.
437 : * Caller should pass *set_elem_result to _bt_skiparray_set_element to advance
438 : * the array.
439 : *
440 : * cur_elem_trig indicates if array advancement was triggered by this array's
441 : * scan key. We use this to optimize-away comparisons that are known by our
442 : * caller to be unnecessary from context, just like _bt_binsrch_array_skey.
443 : */
444 : static void
445 165336 : _bt_binsrch_skiparray_skey(bool cur_elem_trig, ScanDirection dir,
446 : Datum tupdatum, bool tupnull,
447 : BTArrayKeyInfo *array, ScanKey cur,
448 : int32 *set_elem_result)
449 : {
450 : Assert(cur->sk_flags & SK_BT_SKIP);
451 : Assert(cur->sk_flags & SK_SEARCHARRAY);
452 : Assert(cur->sk_flags & SK_BT_REQFWD);
453 : Assert(array->num_elems == -1);
454 : Assert(!ScanDirectionIsNoMovement(dir));
455 :
456 165336 : if (array->null_elem)
457 : {
458 : Assert(!array->low_compare && !array->high_compare);
459 :
460 138736 : *set_elem_result = 0;
461 138736 : return;
462 : }
463 :
464 26600 : if (tupnull) /* NULL tupdatum */
465 : {
466 24 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
467 0 : *set_elem_result = -1; /* NULL "<" NOT_NULL */
468 : else
469 24 : *set_elem_result = 1; /* NULL ">" NOT_NULL */
470 24 : return;
471 : }
472 :
473 : /*
474 : * Array inequalities determine whether tupdatum is within the range of
475 : * caller's skip array
476 : */
477 26576 : *set_elem_result = 0;
478 26576 : if (ScanDirectionIsForward(dir))
479 : {
480 : /*
481 : * Evaluate low_compare first (unless cur_elem_trig tells us that it
482 : * cannot possibly fail to be satisfied), then evaluate high_compare
483 : */
484 26528 : if (!cur_elem_trig && array->low_compare &&
485 734 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
486 734 : array->low_compare->sk_collation,
487 : tupdatum,
488 734 : array->low_compare->sk_argument)))
489 0 : *set_elem_result = -1;
490 26528 : else if (array->high_compare &&
491 10436 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
492 10436 : array->high_compare->sk_collation,
493 : tupdatum,
494 10436 : array->high_compare->sk_argument)))
495 6406 : *set_elem_result = 1;
496 : }
497 : else
498 : {
499 : /*
500 : * Evaluate high_compare first (unless cur_elem_trig tells us that it
501 : * cannot possibly fail to be satisfied), then evaluate low_compare
502 : */
503 48 : if (!cur_elem_trig && array->high_compare &&
504 6 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
505 6 : array->high_compare->sk_collation,
506 : tupdatum,
507 6 : array->high_compare->sk_argument)))
508 0 : *set_elem_result = 1;
509 48 : else if (array->low_compare &&
510 24 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
511 24 : array->low_compare->sk_collation,
512 : tupdatum,
513 24 : array->low_compare->sk_argument)))
514 0 : *set_elem_result = -1;
515 : }
516 :
517 : /*
518 : * Assert that any keys that were assumed to be satisfied already (due to
519 : * caller passing cur_elem_trig=true) really are satisfied as expected
520 : */
521 : #ifdef USE_ASSERT_CHECKING
522 : if (cur_elem_trig)
523 : {
524 : if (ScanDirectionIsForward(dir) && array->low_compare)
525 : Assert(DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
526 : array->low_compare->sk_collation,
527 : tupdatum,
528 : array->low_compare->sk_argument)));
529 :
530 : if (ScanDirectionIsBackward(dir) && array->high_compare)
531 : Assert(DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
532 : array->high_compare->sk_collation,
533 : tupdatum,
534 : array->high_compare->sk_argument)));
535 : }
536 : #endif
537 : }
538 :
539 : /*
540 : * _bt_skiparray_set_element() -- Set skip array scan key's sk_argument
541 : *
542 : * Caller passes set_elem_result returned by _bt_binsrch_skiparray_skey for
543 : * caller's tupdatum/tupnull.
544 : *
545 : * We copy tupdatum/tupnull into skey's sk_argument iff set_elem_result == 0.
546 : * Otherwise, we set skey to either the lowest or highest value that's within
547 : * the range of caller's skip array (whichever is the best available match to
548 : * tupdatum/tupnull that is still within the range of the skip array according
549 : * to _bt_binsrch_skiparray_skey/set_elem_result).
550 : */
551 : static void
552 153868 : _bt_skiparray_set_element(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
553 : int32 set_elem_result, Datum tupdatum, bool tupnull)
554 : {
555 : Assert(skey->sk_flags & SK_BT_SKIP);
556 : Assert(skey->sk_flags & SK_SEARCHARRAY);
557 :
558 153868 : if (set_elem_result)
559 : {
560 : /* tupdatum/tupnull is out of the range of the skip array */
561 : Assert(!array->null_elem);
562 :
563 642 : _bt_array_set_low_or_high(rel, skey, array, set_elem_result < 0);
564 642 : return;
565 : }
566 :
567 : /* Advance skip array to tupdatum (or tupnull) value */
568 153226 : if (unlikely(tupnull))
569 : {
570 36 : _bt_skiparray_set_isnull(rel, skey, array);
571 36 : return;
572 : }
573 :
574 : /* Free memory previously allocated for sk_argument if needed */
575 153190 : if (!array->attbyval && skey->sk_argument)
576 77422 : pfree(DatumGetPointer(skey->sk_argument));
577 :
578 : /* tupdatum becomes new sk_argument/new current element */
579 153190 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL |
580 : SK_BT_MINVAL | SK_BT_MAXVAL |
581 : SK_BT_NEXT | SK_BT_PRIOR);
582 153190 : skey->sk_argument = datumCopy(tupdatum, array->attbyval, array->attlen);
583 : }
584 :
585 : /*
586 : * _bt_skiparray_set_isnull() -- set skip array scan key to NULL
587 : */
588 : static void
589 48 : _bt_skiparray_set_isnull(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
590 : {
591 : Assert(skey->sk_flags & SK_BT_SKIP);
592 : Assert(skey->sk_flags & SK_SEARCHARRAY);
593 : Assert(array->null_elem && !array->low_compare && !array->high_compare);
594 :
595 : /* Free memory previously allocated for sk_argument if needed */
596 48 : if (!array->attbyval && skey->sk_argument)
597 6 : pfree(DatumGetPointer(skey->sk_argument));
598 :
599 : /* NULL becomes new sk_argument/new current element */
600 48 : skey->sk_argument = (Datum) 0;
601 48 : skey->sk_flags &= ~(SK_BT_MINVAL | SK_BT_MAXVAL |
602 : SK_BT_NEXT | SK_BT_PRIOR);
603 48 : skey->sk_flags |= (SK_SEARCHNULL | SK_ISNULL);
604 48 : }
605 :
606 : /*
607 : * _bt_start_array_keys() -- Initialize array keys at start of a scan
608 : *
609 : * Set up the cur_elem counters and fill in the first sk_argument value for
610 : * each array scankey.
611 : */
612 : void
613 81434 : _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
614 : {
615 81434 : Relation rel = scan->indexRelation;
616 81434 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
617 :
618 : Assert(so->numArrayKeys);
619 : Assert(so->qual_ok);
620 :
621 163492 : for (int i = 0; i < so->numArrayKeys; i++)
622 : {
623 82058 : BTArrayKeyInfo *array = &so->arrayKeys[i];
624 82058 : ScanKey skey = &so->keyData[array->scan_key];
625 :
626 : Assert(skey->sk_flags & SK_SEARCHARRAY);
627 :
628 82058 : _bt_array_set_low_or_high(rel, skey, array,
629 : ScanDirectionIsForward(dir));
630 : }
631 81434 : so->scanBehind = so->oppositeDirCheck = false; /* reset */
632 81434 : }
633 :
634 : /*
635 : * _bt_array_set_low_or_high() -- Set array scan key to lowest/highest element
636 : *
637 : * Caller also passes associated scan key, which will have its argument set to
638 : * the lowest/highest array value in passing.
639 : */
640 : static void
641 93512 : _bt_array_set_low_or_high(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
642 : bool low_not_high)
643 : {
644 : Assert(skey->sk_flags & SK_SEARCHARRAY);
645 :
646 93512 : if (array->num_elems != -1)
647 : {
648 : /* set low or high element for SAOP array */
649 84682 : int set_elem = 0;
650 :
651 : Assert(!(skey->sk_flags & SK_BT_SKIP));
652 :
653 84682 : if (!low_not_high)
654 8304 : set_elem = array->num_elems - 1;
655 :
656 : /*
657 : * Just copy over array datum (only skip arrays require freeing and
658 : * allocating memory for sk_argument)
659 : */
660 84682 : array->cur_elem = set_elem;
661 84682 : skey->sk_argument = array->elem_values[set_elem];
662 :
663 84682 : return;
664 : }
665 :
666 : /* set low or high element for skip array */
667 : Assert(skey->sk_flags & SK_BT_SKIP);
668 : Assert(array->num_elems == -1);
669 :
670 : /* Free memory previously allocated for sk_argument if needed */
671 8830 : if (!array->attbyval && skey->sk_argument)
672 1650 : pfree(DatumGetPointer(skey->sk_argument));
673 :
674 : /* Reset flags */
675 8830 : skey->sk_argument = (Datum) 0;
676 8830 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL |
677 : SK_BT_MINVAL | SK_BT_MAXVAL |
678 : SK_BT_NEXT | SK_BT_PRIOR);
679 :
680 8830 : if (array->null_elem &&
681 7000 : (low_not_high == ((skey->sk_flags & SK_BT_NULLS_FIRST) != 0)))
682 : {
683 : /* Requested element (either lowest or highest) has the value NULL */
684 966 : skey->sk_flags |= (SK_SEARCHNULL | SK_ISNULL);
685 : }
686 7864 : else if (low_not_high)
687 : {
688 : /* Setting array to lowest element (according to low_compare) */
689 7144 : skey->sk_flags |= SK_BT_MINVAL;
690 : }
691 : else
692 : {
693 : /* Setting array to highest element (according to high_compare) */
694 720 : skey->sk_flags |= SK_BT_MAXVAL;
695 : }
696 : }
697 :
698 : /*
699 : * _bt_array_decrement() -- decrement array scan key's sk_argument
700 : *
701 : * Return value indicates whether caller's array was successfully decremented.
702 : * Cannot decrement an array whose current element is already the first one.
703 : */
704 : static bool
705 912 : _bt_array_decrement(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
706 : {
707 912 : bool uflow = false;
708 : Datum dec_sk_argument;
709 :
710 : Assert(skey->sk_flags & SK_SEARCHARRAY);
711 : Assert(!(skey->sk_flags & (SK_BT_MAXVAL | SK_BT_NEXT | SK_BT_PRIOR)));
712 :
713 : /* SAOP array? */
714 912 : if (array->num_elems != -1)
715 : {
716 : Assert(!(skey->sk_flags & (SK_BT_SKIP | SK_BT_MINVAL | SK_BT_MAXVAL)));
717 36 : if (array->cur_elem > 0)
718 : {
719 : /*
720 : * Just decrement current element, and assign its datum to skey
721 : * (only skip arrays need us to free existing sk_argument memory)
722 : */
723 6 : array->cur_elem--;
724 6 : skey->sk_argument = array->elem_values[array->cur_elem];
725 :
726 : /* Successfully decremented array */
727 6 : return true;
728 : }
729 :
730 : /* Cannot decrement to before first array element */
731 30 : return false;
732 : }
733 :
734 : /* Nope, this is a skip array */
735 : Assert(skey->sk_flags & SK_BT_SKIP);
736 :
737 : /*
738 : * The sentinel value that represents the minimum value within the range
739 : * of a skip array (often just -inf) is never decrementable
740 : */
741 876 : if (skey->sk_flags & SK_BT_MINVAL)
742 0 : return false;
743 :
744 : /*
745 : * When the current array element is NULL, and the lowest sorting value in
746 : * the index is also NULL, we cannot decrement before first array element
747 : */
748 876 : if ((skey->sk_flags & SK_ISNULL) && (skey->sk_flags & SK_BT_NULLS_FIRST))
749 0 : return false;
750 :
751 : /*
752 : * Opclasses without skip support "decrement" the scan key's current
753 : * element by setting the PRIOR flag. The true prior value is determined
754 : * by repositioning to the last index tuple < existing sk_argument/current
755 : * array element. Note that this works in the usual way when the scan key
756 : * is already marked ISNULL (i.e. when the current element is NULL).
757 : */
758 876 : if (!array->sksup)
759 : {
760 : /* Successfully "decremented" array */
761 12 : skey->sk_flags |= SK_BT_PRIOR;
762 12 : return true;
763 : }
764 :
765 : /*
766 : * Opclasses with skip support directly decrement sk_argument
767 : */
768 864 : if (skey->sk_flags & SK_ISNULL)
769 : {
770 : Assert(!(skey->sk_flags & SK_BT_NULLS_FIRST));
771 :
772 : /*
773 : * Existing sk_argument/array element is NULL (for an IS NULL qual).
774 : *
775 : * "Decrement" from NULL to the high_elem value provided by opclass
776 : * skip support routine.
777 : */
778 6 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL);
779 12 : skey->sk_argument = datumCopy(array->sksup->high_elem,
780 6 : array->attbyval, array->attlen);
781 6 : return true;
782 : }
783 :
784 : /*
785 : * Ask opclass support routine to provide decremented copy of existing
786 : * non-NULL sk_argument
787 : */
788 858 : dec_sk_argument = array->sksup->decrement(rel, skey->sk_argument, &uflow);
789 858 : if (unlikely(uflow))
790 : {
791 : /* dec_sk_argument has undefined value (so no pfree) */
792 0 : if (array->null_elem && (skey->sk_flags & SK_BT_NULLS_FIRST))
793 : {
794 0 : _bt_skiparray_set_isnull(rel, skey, array);
795 :
796 : /* Successfully "decremented" array to NULL */
797 0 : return true;
798 : }
799 :
800 : /* Cannot decrement to before first array element */
801 0 : return false;
802 : }
803 :
804 : /*
805 : * Successfully decremented sk_argument to a non-NULL value. Make sure
806 : * that the decremented value is still within the range of the array.
807 : */
808 858 : if (array->low_compare &&
809 12 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
810 12 : array->low_compare->sk_collation,
811 : dec_sk_argument,
812 12 : array->low_compare->sk_argument)))
813 : {
814 : /* Keep existing sk_argument after all */
815 6 : if (!array->attbyval)
816 0 : pfree(DatumGetPointer(dec_sk_argument));
817 :
818 : /* Cannot decrement to before first array element */
819 6 : return false;
820 : }
821 :
822 : /* Accept value returned by opclass decrement callback */
823 852 : if (!array->attbyval && skey->sk_argument)
824 0 : pfree(DatumGetPointer(skey->sk_argument));
825 852 : skey->sk_argument = dec_sk_argument;
826 :
827 : /* Successfully decremented array */
828 852 : return true;
829 : }
830 :
831 : /*
832 : * _bt_array_increment() -- increment array scan key's sk_argument
833 : *
834 : * Return value indicates whether caller's array was successfully incremented.
835 : * Cannot increment an array whose current element is already the final one.
836 : */
837 : static bool
838 31982 : _bt_array_increment(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
839 : {
840 31982 : bool oflow = false;
841 : Datum inc_sk_argument;
842 :
843 : Assert(skey->sk_flags & SK_SEARCHARRAY);
844 : Assert(!(skey->sk_flags & (SK_BT_MINVAL | SK_BT_NEXT | SK_BT_PRIOR)));
845 :
846 : /* SAOP array? */
847 31982 : if (array->num_elems != -1)
848 : {
849 : Assert(!(skey->sk_flags & (SK_BT_SKIP | SK_BT_MINVAL | SK_BT_MAXVAL)));
850 8328 : if (array->cur_elem < array->num_elems - 1)
851 : {
852 : /*
853 : * Just increment current element, and assign its datum to skey
854 : * (only skip arrays need us to free existing sk_argument memory)
855 : */
856 38 : array->cur_elem++;
857 38 : skey->sk_argument = array->elem_values[array->cur_elem];
858 :
859 : /* Successfully incremented array */
860 38 : return true;
861 : }
862 :
863 : /* Cannot increment past final array element */
864 8290 : return false;
865 : }
866 :
867 : /* Nope, this is a skip array */
868 : Assert(skey->sk_flags & SK_BT_SKIP);
869 :
870 : /*
871 : * The sentinel value that represents the maximum value within the range
872 : * of a skip array (often just +inf) is never incrementable
873 : */
874 23654 : if (skey->sk_flags & SK_BT_MAXVAL)
875 642 : return false;
876 :
877 : /*
878 : * When the current array element is NULL, and the highest sorting value
879 : * in the index is also NULL, we cannot increment past the final element
880 : */
881 23012 : if ((skey->sk_flags & SK_ISNULL) && !(skey->sk_flags & SK_BT_NULLS_FIRST))
882 438 : return false;
883 :
884 : /*
885 : * Opclasses without skip support "increment" the scan key's current
886 : * element by setting the NEXT flag. The true next value is determined by
887 : * repositioning to the first index tuple > existing sk_argument/current
888 : * array element. Note that this works in the usual way when the scan key
889 : * is already marked ISNULL (i.e. when the current element is NULL).
890 : */
891 22574 : if (!array->sksup)
892 : {
893 : /* Successfully "incremented" array */
894 15284 : skey->sk_flags |= SK_BT_NEXT;
895 15284 : return true;
896 : }
897 :
898 : /*
899 : * Opclasses with skip support directly increment sk_argument
900 : */
901 7290 : if (skey->sk_flags & SK_ISNULL)
902 : {
903 : Assert(skey->sk_flags & SK_BT_NULLS_FIRST);
904 :
905 : /*
906 : * Existing sk_argument/array element is NULL (for an IS NULL qual).
907 : *
908 : * "Increment" from NULL to the low_elem value provided by opclass
909 : * skip support routine.
910 : */
911 36 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL);
912 72 : skey->sk_argument = datumCopy(array->sksup->low_elem,
913 36 : array->attbyval, array->attlen);
914 36 : return true;
915 : }
916 :
917 : /*
918 : * Ask opclass support routine to provide incremented copy of existing
919 : * non-NULL sk_argument
920 : */
921 7254 : inc_sk_argument = array->sksup->increment(rel, skey->sk_argument, &oflow);
922 7254 : if (unlikely(oflow))
923 : {
924 : /* inc_sk_argument has undefined value (so no pfree) */
925 30 : if (array->null_elem && !(skey->sk_flags & SK_BT_NULLS_FIRST))
926 : {
927 12 : _bt_skiparray_set_isnull(rel, skey, array);
928 :
929 : /* Successfully "incremented" array to NULL */
930 12 : return true;
931 : }
932 :
933 : /* Cannot increment past final array element */
934 18 : return false;
935 : }
936 :
937 : /*
938 : * Successfully incremented sk_argument to a non-NULL value. Make sure
939 : * that the incremented value is still within the range of the array.
940 : */
941 7224 : if (array->high_compare &&
942 42 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
943 42 : array->high_compare->sk_collation,
944 : inc_sk_argument,
945 42 : array->high_compare->sk_argument)))
946 : {
947 : /* Keep existing sk_argument after all */
948 12 : if (!array->attbyval)
949 0 : pfree(DatumGetPointer(inc_sk_argument));
950 :
951 : /* Cannot increment past final array element */
952 12 : return false;
953 : }
954 :
955 : /* Accept value returned by opclass increment callback */
956 7212 : if (!array->attbyval && skey->sk_argument)
957 0 : pfree(DatumGetPointer(skey->sk_argument));
958 7212 : skey->sk_argument = inc_sk_argument;
959 :
960 : /* Successfully incremented array */
961 7212 : return true;
962 : }
963 :
964 : /*
965 : * _bt_advance_array_keys_increment() -- Advance to next set of array elements
966 : *
967 : * Advances the array keys by a single increment in the current scan
968 : * direction. When there are multiple array keys this can roll over from the
969 : * lowest order array to higher order arrays.
970 : *
971 : * Returns true if there is another set of values to consider, false if not.
972 : * On true result, the scankeys are initialized with the next set of values.
973 : * On false result, the scankeys stay the same, and the array keys are not
974 : * advanced (every array remains at its final element for scan direction).
975 : */
976 : static bool
977 31754 : _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
978 : bool *skip_array_set)
979 : {
980 31754 : Relation rel = scan->indexRelation;
981 31754 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
982 :
983 : /*
984 : * We must advance the last array key most quickly, since it will
985 : * correspond to the lowest-order index column among the available
986 : * qualifications
987 : */
988 41190 : for (int i = so->numArrayKeys - 1; i >= 0; i--)
989 : {
990 32894 : BTArrayKeyInfo *array = &so->arrayKeys[i];
991 32894 : ScanKey skey = &so->keyData[array->scan_key];
992 :
993 32894 : if (array->num_elems == -1)
994 24530 : *skip_array_set = true;
995 :
996 32894 : if (ScanDirectionIsForward(dir))
997 : {
998 31982 : if (_bt_array_increment(rel, skey, array))
999 22582 : return true;
1000 : }
1001 : else
1002 : {
1003 912 : if (_bt_array_decrement(rel, skey, array))
1004 876 : return true;
1005 : }
1006 :
1007 : /*
1008 : * Couldn't increment (or decrement) array. Handle array roll over.
1009 : *
1010 : * Start over at the array's lowest sorting value (or its highest
1011 : * value, for backward scans)...
1012 : */
1013 9436 : _bt_array_set_low_or_high(rel, skey, array,
1014 : ScanDirectionIsForward(dir));
1015 :
1016 : /* ...then increment (or decrement) next most significant array */
1017 : }
1018 :
1019 : /*
1020 : * The array keys are now exhausted.
1021 : *
1022 : * Restore the array keys to the state they were in immediately before we
1023 : * were called. This ensures that the arrays only ever ratchet in the
1024 : * current scan direction.
1025 : *
1026 : * Without this, scans could overlook matching tuples when the scan
1027 : * direction gets reversed just before btgettuple runs out of items to
1028 : * return, but just after _bt_readpage prepares all the items from the
1029 : * scan's final page in so->currPos. When we're on the final page it is
1030 : * typical for so->currPos to get invalidated once btgettuple finally
1031 : * returns false, which'll effectively invalidate the scan's array keys.
1032 : * That hasn't happened yet, though -- and in general it may never happen.
1033 : */
1034 8296 : _bt_start_array_keys(scan, -dir);
1035 :
1036 8296 : return false;
1037 : }
1038 :
1039 : /*
1040 : * _bt_tuple_before_array_skeys() -- too early to advance required arrays?
1041 : *
1042 : * We always compare the tuple using the current array keys (which we assume
1043 : * are already set in so->keyData[]). readpagetup indicates if tuple is the
1044 : * scan's current _bt_readpage-wise tuple.
1045 : *
1046 : * readpagetup callers must only call here when _bt_check_compare already set
1047 : * continuescan=false. We help these callers deal with _bt_check_compare's
1048 : * inability to distinguish between the < and > cases (it uses equality
1049 : * operator scan keys, whereas we use 3-way ORDER procs). These callers pass
1050 : * a _bt_check_compare-set sktrig value that indicates which scan key
1051 : * triggered the call (!readpagetup callers just pass us sktrig=0 instead).
1052 : * This information allows us to avoid wastefully checking earlier scan keys
1053 : * that were already deemed to have been satisfied inside _bt_check_compare.
1054 : *
1055 : * Returns false when caller's tuple is >= the current required equality scan
1056 : * keys (or <=, in the case of backwards scans). This happens to readpagetup
1057 : * callers when the scan has reached the point of needing its array keys
1058 : * advanced; caller will need to advance required and non-required arrays at
1059 : * scan key offsets >= sktrig, plus scan keys < sktrig iff sktrig rolls over.
1060 : * (When we return false to readpagetup callers, tuple can only be == current
1061 : * required equality scan keys when caller's sktrig indicates that the arrays
1062 : * need to be advanced due to an unsatisfied required inequality key trigger.)
1063 : *
1064 : * Returns true when caller passes a tuple that is < the current set of
1065 : * equality keys for the most significant non-equal required scan key/column
1066 : * (or > the keys, during backwards scans). This happens to readpagetup
1067 : * callers when tuple is still before the start of matches for the scan's
1068 : * required equality strategy scan keys. (sktrig can't have indicated that an
1069 : * inequality strategy scan key wasn't satisfied in _bt_check_compare when we
1070 : * return true. In fact, we automatically return false when passed such an
1071 : * inequality sktrig by readpagetup callers -- _bt_check_compare's initial
1072 : * continuescan=false doesn't really need to be confirmed here by us.)
1073 : *
1074 : * !readpagetup callers optionally pass us *scanBehind, which tracks whether
1075 : * any missing truncated attributes might have affected array advancement
1076 : * (compared to what would happen if it was shown the first non-pivot tuple on
1077 : * the page to the right of caller's finaltup/high key tuple instead). It's
1078 : * only possible that we'll set *scanBehind to true when caller passes us a
1079 : * pivot tuple (with truncated -inf attributes) that we return false for.
1080 : */
1081 : static bool
1082 324452 : _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
1083 : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
1084 : bool readpagetup, int sktrig, bool *scanBehind)
1085 : {
1086 324452 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1087 :
1088 : Assert(so->numArrayKeys);
1089 : Assert(so->numberOfKeys);
1090 : Assert(sktrig == 0 || readpagetup);
1091 : Assert(!readpagetup || scanBehind == NULL);
1092 :
1093 324452 : if (scanBehind)
1094 82994 : *scanBehind = false;
1095 :
1096 327650 : for (int ikey = sktrig; ikey < so->numberOfKeys; ikey++)
1097 : {
1098 326968 : ScanKey cur = so->keyData + ikey;
1099 : Datum tupdatum;
1100 : bool tupnull;
1101 : int32 result;
1102 :
1103 : /* readpagetup calls require one ORDER proc comparison (at most) */
1104 : Assert(!readpagetup || ikey == sktrig);
1105 :
1106 : /*
1107 : * Once we reach a non-required scan key, we're completely done.
1108 : *
1109 : * Note: we deliberately don't consider the scan direction here.
1110 : * _bt_advance_array_keys caller requires that we track *scanBehind
1111 : * without concern for scan direction.
1112 : */
1113 326968 : if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) == 0)
1114 : {
1115 : Assert(!readpagetup);
1116 : Assert(ikey > sktrig || ikey == 0);
1117 323770 : return false;
1118 : }
1119 :
1120 326968 : if (cur->sk_attno > tupnatts)
1121 : {
1122 : Assert(!readpagetup);
1123 :
1124 : /*
1125 : * When we reach a high key's truncated attribute, assume that the
1126 : * tuple attribute's value is >= the scan's equality constraint
1127 : * scan keys (but set *scanBehind to let interested callers know
1128 : * that a truncated attribute might have affected our answer).
1129 : */
1130 36 : if (scanBehind)
1131 36 : *scanBehind = true;
1132 :
1133 36 : return false;
1134 : }
1135 :
1136 : /*
1137 : * Deal with inequality strategy scan keys that _bt_check_compare set
1138 : * continuescan=false for
1139 : */
1140 326932 : if (cur->sk_strategy != BTEqualStrategyNumber)
1141 : {
1142 : /*
1143 : * When _bt_check_compare indicated that a required inequality
1144 : * scan key wasn't satisfied, there's no need to verify anything;
1145 : * caller always calls _bt_advance_array_keys with this sktrig.
1146 : */
1147 618 : if (readpagetup)
1148 348 : return false;
1149 :
1150 : /*
1151 : * Otherwise we can't give up, since we must check all required
1152 : * scan keys (required in either direction) in order to correctly
1153 : * track *scanBehind for caller
1154 : */
1155 270 : continue;
1156 : }
1157 :
1158 326314 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1159 :
1160 326314 : if (likely(!(cur->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL))))
1161 : {
1162 : /* Scankey has a valid/comparable sk_argument value */
1163 321136 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1164 : tupdatum, tupnull,
1165 : cur->sk_argument, cur);
1166 :
1167 321136 : if (result == 0)
1168 : {
1169 : /*
1170 : * Interpret result in a way that takes NEXT/PRIOR into
1171 : * account
1172 : */
1173 17154 : if (cur->sk_flags & SK_BT_NEXT)
1174 14196 : result = -1;
1175 2958 : else if (cur->sk_flags & SK_BT_PRIOR)
1176 30 : result = 1;
1177 :
1178 : Assert(result == 0 || (cur->sk_flags & SK_BT_SKIP));
1179 : }
1180 : }
1181 : else
1182 : {
1183 5178 : BTArrayKeyInfo *array = NULL;
1184 :
1185 : /*
1186 : * Current array element/array = scan key value is a sentinel
1187 : * value that represents the lowest (or highest) possible value
1188 : * that's still within the range of the array.
1189 : *
1190 : * Like _bt_first, we only see MINVAL keys during forwards scans
1191 : * (and similarly only see MAXVAL keys during backwards scans).
1192 : * Even if the scan's direction changes, we'll stop at some higher
1193 : * order key before we can ever reach any MAXVAL (or MINVAL) keys.
1194 : * (However, unlike _bt_first we _can_ get to keys marked either
1195 : * NEXT or PRIOR, regardless of the scan's current direction.)
1196 : */
1197 : Assert(ScanDirectionIsForward(dir) ?
1198 : !(cur->sk_flags & SK_BT_MAXVAL) :
1199 : !(cur->sk_flags & SK_BT_MINVAL));
1200 :
1201 : /*
1202 : * There are no valid sk_argument values in MINVAL/MAXVAL keys.
1203 : * Check if tupdatum is within the range of skip array instead.
1204 : */
1205 5694 : for (int arrayidx = 0; arrayidx < so->numArrayKeys; arrayidx++)
1206 : {
1207 5694 : array = &so->arrayKeys[arrayidx];
1208 5694 : if (array->scan_key == ikey)
1209 5178 : break;
1210 : }
1211 :
1212 5178 : _bt_binsrch_skiparray_skey(false, dir, tupdatum, tupnull,
1213 : array, cur, &result);
1214 :
1215 5178 : if (result == 0)
1216 : {
1217 : /*
1218 : * tupdatum satisfies both low_compare and high_compare, so
1219 : * it's time to advance the array keys.
1220 : *
1221 : * Note: It's possible that the skip array will "advance" from
1222 : * its MINVAL (or MAXVAL) representation to an alternative,
1223 : * logically equivalent representation of the same value: a
1224 : * representation where the = key gets a valid datum in its
1225 : * sk_argument. This is only possible when low_compare uses
1226 : * the >= strategy (or high_compare uses the <= strategy).
1227 : */
1228 5170 : return false;
1229 : }
1230 : }
1231 :
1232 : /*
1233 : * Does this comparison indicate that caller must _not_ advance the
1234 : * scan's arrays just yet?
1235 : */
1236 321144 : if ((ScanDirectionIsForward(dir) && result < 0) ||
1237 3252 : (ScanDirectionIsBackward(dir) && result > 0))
1238 61372 : return true;
1239 :
1240 : /*
1241 : * Does this comparison indicate that caller should now advance the
1242 : * scan's arrays? (Must be if we get here during a readpagetup call.)
1243 : */
1244 259772 : if (readpagetup || result != 0)
1245 : {
1246 : Assert(result != 0);
1247 256844 : return false;
1248 : }
1249 :
1250 : /*
1251 : * Inconclusive -- need to check later scan keys, too.
1252 : *
1253 : * This must be a finaltup precheck, or a call made from an assertion.
1254 : */
1255 : Assert(result == 0);
1256 : }
1257 :
1258 : Assert(!readpagetup);
1259 :
1260 682 : return false;
1261 : }
1262 :
1263 : /*
1264 : * _bt_start_prim_scan() -- start scheduled primitive index scan?
1265 : *
1266 : * Returns true if _bt_checkkeys scheduled another primitive index scan, just
1267 : * as the last one ended. Otherwise returns false, indicating that the array
1268 : * keys are now fully exhausted.
1269 : *
1270 : * Only call here during scans with one or more equality type array scan keys,
1271 : * after _bt_first or _bt_next return false.
1272 : */
1273 : bool
1274 88730 : _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir)
1275 : {
1276 88730 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1277 :
1278 : Assert(so->numArrayKeys);
1279 :
1280 88730 : so->scanBehind = so->oppositeDirCheck = false; /* reset */
1281 :
1282 : /*
1283 : * Array keys are advanced within _bt_checkkeys when the scan reaches the
1284 : * leaf level (more precisely, they're advanced when the scan reaches the
1285 : * end of each distinct set of array elements). This process avoids
1286 : * repeat access to leaf pages (across multiple primitive index scans) by
1287 : * advancing the scan's array keys when it allows the primitive index scan
1288 : * to find nearby matching tuples (or when it eliminates ranges of array
1289 : * key space that can't possibly be satisfied by any index tuple).
1290 : *
1291 : * _bt_checkkeys sets a simple flag variable to schedule another primitive
1292 : * index scan. The flag tells us what to do.
1293 : *
1294 : * We cannot rely on _bt_first always reaching _bt_checkkeys. There are
1295 : * various cases where that won't happen. For example, if the index is
1296 : * completely empty, then _bt_first won't call _bt_readpage/_bt_checkkeys.
1297 : * We also don't expect a call to _bt_checkkeys during searches for a
1298 : * non-existent value that happens to be lower/higher than any existing
1299 : * value in the index.
1300 : *
1301 : * We don't require special handling for these cases -- we don't need to
1302 : * be explicitly instructed to _not_ perform another primitive index scan.
1303 : * It's up to code under the control of _bt_first to always set the flag
1304 : * when another primitive index scan will be required.
1305 : *
1306 : * This works correctly, even with the tricky cases listed above, which
1307 : * all involve access to leaf pages "near the boundaries of the key space"
1308 : * (whether it's from a leftmost/rightmost page, or an imaginary empty
1309 : * leaf root page). If _bt_checkkeys cannot be reached by a primitive
1310 : * index scan for one set of array keys, then it also won't be reached for
1311 : * any later set ("later" in terms of the direction that we scan the index
1312 : * and advance the arrays). The array keys won't have advanced in these
1313 : * cases, but that's the correct behavior (even _bt_advance_array_keys
1314 : * won't always advance the arrays at the point they become "exhausted").
1315 : */
1316 88730 : if (so->needPrimScan)
1317 : {
1318 : /*
1319 : * Flag was set -- must call _bt_first again, which will reset the
1320 : * scan's needPrimScan flag
1321 : */
1322 17526 : return true;
1323 : }
1324 :
1325 : /* The top-level index scan ran out of tuples in this scan direction */
1326 71204 : if (scan->parallel_scan != NULL)
1327 30 : _bt_parallel_done(scan);
1328 :
1329 71204 : return false;
1330 : }
1331 :
1332 : /*
1333 : * _bt_advance_array_keys() -- Advance array elements using a tuple
1334 : *
1335 : * The scan always gets a new qual as a consequence of calling here (except
1336 : * when we determine that the top-level scan has run out of matching tuples).
1337 : * All later _bt_check_compare calls also use the same new qual that was first
1338 : * used here (at least until the next call here advances the keys once again).
1339 : * It's convenient to structure _bt_check_compare rechecks of caller's tuple
1340 : * (using the new qual) as one the steps of advancing the scan's array keys,
1341 : * so this function works as a wrapper around _bt_check_compare.
1342 : *
1343 : * Like _bt_check_compare, we'll set pstate.continuescan on behalf of the
1344 : * caller, and return a boolean indicating if caller's tuple satisfies the
1345 : * scan's new qual. But unlike _bt_check_compare, we set so->needPrimScan
1346 : * when we set continuescan=false, indicating if a new primitive index scan
1347 : * has been scheduled (otherwise, the top-level scan has run out of tuples in
1348 : * the current scan direction).
1349 : *
1350 : * Caller must use _bt_tuple_before_array_skeys to determine if the current
1351 : * place in the scan is >= the current array keys _before_ calling here.
1352 : * We're responsible for ensuring that caller's tuple is <= the newly advanced
1353 : * required array keys once we return. We try to find an exact match, but
1354 : * failing that we'll advance the array keys to whatever set of array elements
1355 : * comes next in the key space for the current scan direction. Required array
1356 : * keys "ratchet forwards" (or backwards). They can only advance as the scan
1357 : * itself advances through the index/key space.
1358 : *
1359 : * (The rules are the same for backwards scans, except that the operators are
1360 : * flipped: just replace the precondition's >= operator with a <=, and the
1361 : * postcondition's <= operator with a >=. In other words, just swap the
1362 : * precondition with the postcondition.)
1363 : *
1364 : * We also deal with "advancing" non-required arrays here (or arrays that are
1365 : * treated as non-required for the duration of a _bt_readpage call). Callers
1366 : * whose sktrig scan key is non-required specify sktrig_required=false. These
1367 : * calls are the only exception to the general rule about always advancing the
1368 : * required array keys (the scan may not even have a required array). These
1369 : * callers should just pass a NULL pstate (since there is never any question
1370 : * of stopping the scan). No call to _bt_tuple_before_array_skeys is required
1371 : * ahead of these calls (it's already clear that any required scan keys must
1372 : * be satisfied by caller's tuple).
1373 : *
1374 : * Note that we deal with non-array required equality strategy scan keys as
1375 : * degenerate single element arrays here. Obviously, they can never really
1376 : * advance in the way that real arrays can, but they must still affect how we
1377 : * advance real array scan keys (exactly like true array equality scan keys).
1378 : * We have to keep around a 3-way ORDER proc for these (using the "=" operator
1379 : * won't do), since in general whether the tuple is < or > _any_ unsatisfied
1380 : * required equality key influences how the scan's real arrays must advance.
1381 : *
1382 : * Note also that we may sometimes need to advance the array keys when the
1383 : * existing required array keys (and other required equality keys) are already
1384 : * an exact match for every corresponding value from caller's tuple. We must
1385 : * do this for inequalities that _bt_check_compare set continuescan=false for.
1386 : * They'll advance the array keys here, just like any other scan key that
1387 : * _bt_check_compare stops on. (This can even happen _after_ we advance the
1388 : * array keys, in which case we'll advance the array keys a second time. That
1389 : * way _bt_checkkeys caller always has its required arrays advance to the
1390 : * maximum possible extent that its tuple will allow.)
1391 : */
1392 : static bool
1393 199778 : _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
1394 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
1395 : int sktrig, bool sktrig_required)
1396 : {
1397 199778 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1398 199778 : Relation rel = scan->indexRelation;
1399 199778 : ScanDirection dir = so->currPos.dir;
1400 199778 : int arrayidx = 0;
1401 199778 : bool beyond_end_advance = false,
1402 199778 : skip_array_advanced = false,
1403 199778 : has_required_opposite_direction_only = false,
1404 199778 : all_required_satisfied = true,
1405 199778 : all_satisfied = true;
1406 :
1407 : Assert(!so->needPrimScan && !so->scanBehind && !so->oppositeDirCheck);
1408 : Assert(_bt_verify_keys_with_arraykeys(scan));
1409 :
1410 199778 : if (sktrig_required)
1411 : {
1412 : /*
1413 : * Precondition array state assertion
1414 : */
1415 : Assert(!_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc,
1416 : tupnatts, false, 0, NULL));
1417 :
1418 : /*
1419 : * Once we return we'll have a new set of required array keys, so
1420 : * reset state used by "look ahead" optimization
1421 : */
1422 191108 : pstate->rechecks = 0;
1423 191108 : pstate->targetdistance = 0;
1424 : }
1425 8670 : else if (sktrig < so->numberOfKeys - 1 &&
1426 8670 : !(so->keyData[so->numberOfKeys - 1].sk_flags & SK_SEARCHARRAY))
1427 : {
1428 8670 : int least_sign_ikey = so->numberOfKeys - 1;
1429 : bool continuescan;
1430 :
1431 : /*
1432 : * Optimization: perform a precheck of the least significant key
1433 : * during !sktrig_required calls when it isn't already our sktrig
1434 : * (provided the precheck key is not itself an array).
1435 : *
1436 : * When the precheck works out we'll avoid an expensive binary search
1437 : * of sktrig's array (plus any other arrays before least_sign_ikey).
1438 : */
1439 : Assert(so->keyData[sktrig].sk_flags & SK_SEARCHARRAY);
1440 8670 : if (!_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
1441 : false, &continuescan,
1442 : &least_sign_ikey))
1443 2382 : return false;
1444 : }
1445 :
1446 579152 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
1447 : {
1448 387522 : ScanKey cur = so->keyData + ikey;
1449 387522 : BTArrayKeyInfo *array = NULL;
1450 : Datum tupdatum;
1451 387522 : bool required = false,
1452 : tupnull;
1453 : int32 result;
1454 387522 : int set_elem = 0;
1455 :
1456 387522 : if (cur->sk_strategy == BTEqualStrategyNumber)
1457 : {
1458 : /* Manage array state */
1459 339400 : if (cur->sk_flags & SK_SEARCHARRAY)
1460 : {
1461 207698 : array = &so->arrayKeys[arrayidx++];
1462 : Assert(array->scan_key == ikey);
1463 : }
1464 : }
1465 : else
1466 : {
1467 : /*
1468 : * Are any inequalities required in the opposite direction only
1469 : * present here?
1470 : */
1471 48122 : if (((ScanDirectionIsForward(dir) &&
1472 48122 : (cur->sk_flags & (SK_BT_REQBKWD))) ||
1473 0 : (ScanDirectionIsBackward(dir) &&
1474 0 : (cur->sk_flags & (SK_BT_REQFWD)))))
1475 15578 : has_required_opposite_direction_only = true;
1476 : }
1477 :
1478 : /* Optimization: skip over known-satisfied scan keys */
1479 387522 : if (ikey < sktrig)
1480 75886 : continue;
1481 :
1482 371162 : if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
1483 : {
1484 371162 : required = true;
1485 :
1486 371162 : if (cur->sk_attno > tupnatts)
1487 : {
1488 : /* Set this just like _bt_tuple_before_array_skeys */
1489 : Assert(sktrig < ikey);
1490 2036 : so->scanBehind = true;
1491 : }
1492 : }
1493 :
1494 : /*
1495 : * Handle a required non-array scan key that the initial call to
1496 : * _bt_check_compare indicated triggered array advancement, if any.
1497 : *
1498 : * The non-array scan key's strategy will be <, <=, or = during a
1499 : * forwards scan (or any one of =, >=, or > during a backwards scan).
1500 : * It follows that the corresponding tuple attribute's value must now
1501 : * be either > or >= the scan key value (for backwards scans it must
1502 : * be either < or <= that value).
1503 : *
1504 : * If this is a required equality strategy scan key, this is just an
1505 : * optimization; _bt_tuple_before_array_skeys already confirmed that
1506 : * this scan key places us ahead of caller's tuple. There's no need
1507 : * to repeat that work now. (The same underlying principle also gets
1508 : * applied by the cur_elem_trig optimization used to speed up searches
1509 : * for the next array element.)
1510 : *
1511 : * If this is a required inequality strategy scan key, we _must_ rely
1512 : * on _bt_check_compare like this; we aren't capable of directly
1513 : * evaluating required inequality strategy scan keys here, on our own.
1514 : */
1515 371162 : if (ikey == sktrig && !array)
1516 : {
1517 : Assert(sktrig_required && required && all_required_satisfied);
1518 :
1519 : /* Use "beyond end" advancement. See below for an explanation. */
1520 7444 : beyond_end_advance = true;
1521 7444 : all_satisfied = all_required_satisfied = false;
1522 :
1523 7444 : continue;
1524 : }
1525 :
1526 : /*
1527 : * Nothing more for us to do with an inequality strategy scan key that
1528 : * wasn't the one that _bt_check_compare stopped on, though.
1529 : *
1530 : * Note: if our later call to _bt_check_compare (to recheck caller's
1531 : * tuple) sets continuescan=false due to finding this same inequality
1532 : * unsatisfied (possible when it's required in the scan direction),
1533 : * we'll deal with it via a recursive "second pass" call.
1534 : */
1535 363718 : else if (cur->sk_strategy != BTEqualStrategyNumber)
1536 47552 : continue;
1537 :
1538 : /*
1539 : * Nothing for us to do with an equality strategy scan key that isn't
1540 : * marked required, either -- unless it's a non-required array
1541 : */
1542 316166 : else if (!required && !array)
1543 0 : continue;
1544 :
1545 : /*
1546 : * Here we perform steps for all array scan keys after a required
1547 : * array scan key whose binary search triggered "beyond end of array
1548 : * element" array advancement due to encountering a tuple attribute
1549 : * value > the closest matching array key (or < for backwards scans).
1550 : */
1551 316166 : if (beyond_end_advance)
1552 : {
1553 1430 : if (array)
1554 596 : _bt_array_set_low_or_high(rel, cur, array,
1555 : ScanDirectionIsBackward(dir));
1556 :
1557 1430 : continue;
1558 : }
1559 :
1560 : /*
1561 : * Here we perform steps for all array scan keys after a required
1562 : * array scan key whose tuple attribute was < the closest matching
1563 : * array key when we dealt with it (or > for backwards scans).
1564 : *
1565 : * This earlier required array key already puts us ahead of caller's
1566 : * tuple in the key space (for the current scan direction). We must
1567 : * make sure that subsequent lower-order array keys do not put us too
1568 : * far ahead (ahead of tuples that have yet to be seen by our caller).
1569 : * For example, when a tuple "(a, b) = (42, 5)" advances the array
1570 : * keys on "a" from 40 to 45, we must also set "b" to whatever the
1571 : * first array element for "b" is. It would be wrong to allow "b" to
1572 : * be set based on the tuple value.
1573 : *
1574 : * Perform the same steps with truncated high key attributes. You can
1575 : * think of this as a "binary search" for the element closest to the
1576 : * value -inf. Again, the arrays must never get ahead of the scan.
1577 : */
1578 314736 : if (!all_required_satisfied || cur->sk_attno > tupnatts)
1579 : {
1580 3100 : if (array)
1581 780 : _bt_array_set_low_or_high(rel, cur, array,
1582 : ScanDirectionIsForward(dir));
1583 :
1584 3100 : continue;
1585 : }
1586 :
1587 : /*
1588 : * Search in scankey's array for the corresponding tuple attribute
1589 : * value from caller's tuple
1590 : */
1591 311636 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1592 :
1593 311636 : if (array)
1594 : {
1595 191740 : bool cur_elem_trig = (sktrig_required && ikey == sktrig);
1596 :
1597 : /*
1598 : * "Binary search" by checking if tupdatum/tupnull are within the
1599 : * range of the skip array
1600 : */
1601 191740 : if (array->num_elems == -1)
1602 159634 : _bt_binsrch_skiparray_skey(cur_elem_trig, dir,
1603 : tupdatum, tupnull, array, cur,
1604 : &result);
1605 :
1606 : /*
1607 : * Binary search for the closest match from the SAOP array
1608 : */
1609 : else
1610 32106 : set_elem = _bt_binsrch_array_skey(&so->orderProcs[ikey],
1611 : cur_elem_trig, dir,
1612 : tupdatum, tupnull, array, cur,
1613 : &result);
1614 : }
1615 : else
1616 : {
1617 : Assert(required);
1618 :
1619 : /*
1620 : * This is a required non-array equality strategy scan key, which
1621 : * we'll treat as a degenerate single element array.
1622 : *
1623 : * This scan key's imaginary "array" can't really advance, but it
1624 : * can still roll over like any other array. (Actually, this is
1625 : * no different to real single value arrays, which never advance
1626 : * without rolling over -- they can never truly advance, either.)
1627 : */
1628 119896 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1629 : tupdatum, tupnull,
1630 : cur->sk_argument, cur);
1631 : }
1632 :
1633 : /*
1634 : * Consider "beyond end of array element" array advancement.
1635 : *
1636 : * When the tuple attribute value is > the closest matching array key
1637 : * (or < in the backwards scan case), we need to ratchet this array
1638 : * forward (backward) by one increment, so that caller's tuple ends up
1639 : * being < final array value instead (or > final array value instead).
1640 : * This process has to work for all of the arrays, not just this one:
1641 : * it must "carry" to higher-order arrays when the set_elem that we
1642 : * just found happens to be the final one for the scan's direction.
1643 : * Incrementing (decrementing) set_elem itself isn't good enough.
1644 : *
1645 : * Our approach is to provisionally use set_elem as if it was an exact
1646 : * match now, then set each later/less significant array to whatever
1647 : * its final element is. Once outside the loop we'll then "increment
1648 : * this array's set_elem" by calling _bt_advance_array_keys_increment.
1649 : * That way the process rolls over to higher order arrays as needed.
1650 : *
1651 : * Under this scheme any required arrays only ever ratchet forwards
1652 : * (or backwards), and always do so to the maximum possible extent
1653 : * that we can know will be safe without seeing the scan's next tuple.
1654 : * We don't need any special handling for required scan keys that lack
1655 : * a real array to advance, nor for redundant scan keys that couldn't
1656 : * be eliminated by _bt_preprocess_keys. It won't matter if some of
1657 : * our "true" array scan keys (or even all of them) are non-required.
1658 : */
1659 311636 : if (sktrig_required && required &&
1660 305348 : ((ScanDirectionIsForward(dir) && result > 0) ||
1661 1716 : (ScanDirectionIsBackward(dir) && result < 0)))
1662 24310 : beyond_end_advance = true;
1663 :
1664 : Assert(all_required_satisfied && all_satisfied);
1665 311636 : if (result != 0)
1666 : {
1667 : /*
1668 : * Track whether caller's tuple satisfies our new post-advancement
1669 : * qual, for required scan keys, as well as for the entire set of
1670 : * interesting scan keys (all required scan keys plus non-required
1671 : * array scan keys are considered interesting.)
1672 : */
1673 142846 : all_satisfied = false;
1674 142846 : if (sktrig_required && required)
1675 137080 : all_required_satisfied = false;
1676 : else
1677 : {
1678 : /*
1679 : * There's no need to advance the arrays using the best
1680 : * available match for a non-required array. Give up now.
1681 : * (Though note that sktrig_required calls still have to do
1682 : * all the usual post-advancement steps, including the recheck
1683 : * call to _bt_check_compare.)
1684 : */
1685 : break;
1686 : }
1687 : }
1688 :
1689 : /* Advance array keys, even when we don't have an exact match */
1690 305870 : if (array)
1691 : {
1692 185974 : if (array->num_elems == -1)
1693 : {
1694 : /* Skip array's new element is tupdatum (or MINVAL/MAXVAL) */
1695 153868 : _bt_skiparray_set_element(rel, cur, array, result,
1696 : tupdatum, tupnull);
1697 153868 : skip_array_advanced = true;
1698 : }
1699 32106 : else if (array->cur_elem != set_elem)
1700 : {
1701 : /* SAOP array's new element is set_elem datum */
1702 23962 : array->cur_elem = set_elem;
1703 23962 : cur->sk_argument = array->elem_values[set_elem];
1704 : }
1705 : }
1706 : }
1707 :
1708 : /*
1709 : * Advance the array keys incrementally whenever "beyond end of array
1710 : * element" array advancement happens, so that advancement will carry to
1711 : * higher-order arrays (might exhaust all the scan's arrays instead, which
1712 : * ends the top-level scan).
1713 : */
1714 197396 : if (beyond_end_advance &&
1715 31754 : !_bt_advance_array_keys_increment(scan, dir, &skip_array_advanced))
1716 8296 : goto end_toplevel_scan;
1717 :
1718 : Assert(_bt_verify_keys_with_arraykeys(scan));
1719 :
1720 : /*
1721 : * Maintain a page-level count of the number of times the scan's array
1722 : * keys advanced in a way that affected at least one skip array
1723 : */
1724 189100 : if (sktrig_required && skip_array_advanced)
1725 159898 : pstate->nskipadvances++;
1726 :
1727 : /*
1728 : * Does tuple now satisfy our new qual? Recheck with _bt_check_compare.
1729 : *
1730 : * Calls triggered by an unsatisfied required scan key, whose tuple now
1731 : * satisfies all required scan keys, but not all nonrequired array keys,
1732 : * will still require a recheck call to _bt_check_compare. They'll still
1733 : * need its "second pass" handling of required inequality scan keys.
1734 : * (Might have missed a still-unsatisfied required inequality scan key
1735 : * that caller didn't detect as the sktrig scan key during its initial
1736 : * _bt_check_compare call that used the old/original qual.)
1737 : *
1738 : * Calls triggered by an unsatisfied nonrequired array scan key never need
1739 : * "second pass" handling of required inequalities (nor any other handling
1740 : * of any required scan key). All that matters is whether caller's tuple
1741 : * satisfies the new qual, so it's safe to just skip the _bt_check_compare
1742 : * recheck when we've already determined that it can only return 'false'.
1743 : *
1744 : * Note: In practice most scan keys are marked required by preprocessing,
1745 : * if necessary by generating a preceding skip array. We nevertheless
1746 : * often handle array keys marked required as if they were nonrequired.
1747 : * This behavior is requested by our _bt_check_compare caller, though only
1748 : * when it is passed "forcenonrequired=true" by _bt_checkkeys.
1749 : */
1750 189100 : if ((sktrig_required && all_required_satisfied) ||
1751 142516 : (!sktrig_required && all_satisfied))
1752 : {
1753 47106 : int nsktrig = sktrig + 1;
1754 : bool continuescan;
1755 :
1756 : Assert(all_required_satisfied);
1757 :
1758 : /* Recheck _bt_check_compare on behalf of caller */
1759 47106 : if (_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
1760 47106 : !sktrig_required, &continuescan,
1761 47106 : &nsktrig) &&
1762 39442 : !so->scanBehind)
1763 : {
1764 : /* This tuple satisfies the new qual */
1765 : Assert(all_satisfied && continuescan);
1766 :
1767 37504 : if (pstate)
1768 36982 : pstate->continuescan = true;
1769 :
1770 37726 : return true;
1771 : }
1772 :
1773 : /*
1774 : * Consider "second pass" handling of required inequalities.
1775 : *
1776 : * It's possible that our _bt_check_compare call indicated that the
1777 : * scan should end due to some unsatisfied inequality that wasn't
1778 : * initially recognized as such by us. Handle this by calling
1779 : * ourselves recursively, this time indicating that the trigger is the
1780 : * inequality that we missed first time around (and using a set of
1781 : * required array/equality keys that are now exact matches for tuple).
1782 : *
1783 : * We make a strong, general guarantee that every _bt_checkkeys call
1784 : * here will advance the array keys to the maximum possible extent
1785 : * that we can know to be safe based on caller's tuple alone. If we
1786 : * didn't perform this step, then that guarantee wouldn't quite hold.
1787 : */
1788 9602 : if (unlikely(!continuescan))
1789 : {
1790 : bool satisfied PG_USED_FOR_ASSERTS_ONLY;
1791 :
1792 : Assert(sktrig_required);
1793 : Assert(so->keyData[nsktrig].sk_strategy != BTEqualStrategyNumber);
1794 :
1795 : /*
1796 : * The tuple must use "beyond end" advancement during the
1797 : * recursive call, so we cannot possibly end up back here when
1798 : * recursing. We'll consume a small, fixed amount of stack space.
1799 : */
1800 : Assert(!beyond_end_advance);
1801 :
1802 : /* Advance the array keys a second time using same tuple */
1803 222 : satisfied = _bt_advance_array_keys(scan, pstate, tuple, tupnatts,
1804 : tupdesc, nsktrig, true);
1805 :
1806 : /* This tuple doesn't satisfy the inequality */
1807 : Assert(!satisfied);
1808 222 : return false;
1809 : }
1810 :
1811 : /*
1812 : * Some non-required scan key (from new qual) still not satisfied.
1813 : *
1814 : * All scan keys required in the current scan direction must still be
1815 : * satisfied, though, so we can trust all_required_satisfied below.
1816 : */
1817 : }
1818 :
1819 : /*
1820 : * When we were called just to deal with "advancing" non-required arrays,
1821 : * this is as far as we can go (cannot stop the scan for these callers)
1822 : */
1823 151374 : if (!sktrig_required)
1824 : {
1825 : /* Caller's tuple doesn't match any qual */
1826 5766 : return false;
1827 : }
1828 :
1829 : /*
1830 : * Postcondition array state assertion (for still-unsatisfied tuples).
1831 : *
1832 : * By here we have established that the scan's required arrays (scan must
1833 : * have at least one required array) advanced, without becoming exhausted.
1834 : *
1835 : * Caller's tuple is now < the newly advanced array keys (or > when this
1836 : * is a backwards scan), except in the case where we only got this far due
1837 : * to an unsatisfied non-required scan key. Verify that with an assert.
1838 : *
1839 : * Note: we don't just quit at this point when all required scan keys were
1840 : * found to be satisfied because we need to consider edge-cases involving
1841 : * scan keys required in the opposite direction only; those aren't tracked
1842 : * by all_required_satisfied.
1843 : */
1844 : Assert(_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts,
1845 : false, 0, NULL) ==
1846 : !all_required_satisfied);
1847 :
1848 : /*
1849 : * We generally permit primitive index scans to continue onto the next
1850 : * sibling page when the page's finaltup satisfies all required scan keys
1851 : * at the point where we're between pages.
1852 : *
1853 : * If caller's tuple is also the page's finaltup, and we see that required
1854 : * scan keys still aren't satisfied, start a new primitive index scan.
1855 : */
1856 145608 : if (!all_required_satisfied && pstate->finaltup == tuple)
1857 510 : goto new_prim_scan;
1858 :
1859 : /*
1860 : * Proactively check finaltup (don't wait until finaltup is reached by the
1861 : * scan) when it might well turn out to not be satisfied later on.
1862 : *
1863 : * Note: if so->scanBehind hasn't already been set for finaltup by us,
1864 : * it'll be set during this call to _bt_tuple_before_array_skeys. Either
1865 : * way, it'll be set correctly (for the whole page) after this point.
1866 : */
1867 225754 : if (!all_required_satisfied && pstate->finaltup &&
1868 161312 : _bt_tuple_before_array_skeys(scan, dir, pstate->finaltup, tupdesc,
1869 161312 : BTreeTupleGetNAtts(pstate->finaltup, rel),
1870 : false, 0, &so->scanBehind))
1871 17468 : goto new_prim_scan;
1872 :
1873 : /*
1874 : * When we encounter a truncated finaltup high key attribute, we're
1875 : * optimistic about the chances of its corresponding required scan key
1876 : * being satisfied when we go on to recheck it against tuples from this
1877 : * page's right sibling leaf page. We consider truncated attributes to be
1878 : * satisfied by required scan keys, which allows the primitive index scan
1879 : * to continue to the next leaf page. We must set so->scanBehind to true
1880 : * to remember that the last page's finaltup had "satisfied" required scan
1881 : * keys for one or more truncated attribute values (scan keys required in
1882 : * _either_ scan direction).
1883 : *
1884 : * There is a chance that _bt_readpage (which checks so->scanBehind) will
1885 : * find that even the sibling leaf page's finaltup is < the new array
1886 : * keys. When that happens, our optimistic policy will have incurred a
1887 : * single extra leaf page access that could have been avoided.
1888 : *
1889 : * A pessimistic policy would give backward scans a gratuitous advantage
1890 : * over forward scans. We'd punish forward scans for applying more
1891 : * accurate information from the high key, rather than just using the
1892 : * final non-pivot tuple as finaltup, in the style of backward scans.
1893 : * Being pessimistic would also give some scans with non-required arrays a
1894 : * perverse advantage over similar scans that use required arrays instead.
1895 : *
1896 : * This is similar to our scan-level heuristics, below. They also set
1897 : * scanBehind to speculatively continue the primscan onto the next page.
1898 : */
1899 127630 : if (so->scanBehind)
1900 : {
1901 : /* Truncated high key -- _bt_scanbehind_checkkeys recheck scheduled */
1902 : }
1903 :
1904 : /*
1905 : * Handle inequalities marked required in the opposite scan direction.
1906 : * They can also signal that we should start a new primitive index scan.
1907 : *
1908 : * It's possible that the scan is now positioned where "matching" tuples
1909 : * begin, and that caller's tuple satisfies all scan keys required in the
1910 : * current scan direction. But if caller's tuple still doesn't satisfy
1911 : * other scan keys that are required in the opposite scan direction only
1912 : * (e.g., a required >= strategy scan key when scan direction is forward),
1913 : * it's still possible that there are many leaf pages before the page that
1914 : * _bt_first could skip straight to. Groveling through all those pages
1915 : * will always give correct answers, but it can be very inefficient. We
1916 : * must avoid needlessly scanning extra pages.
1917 : *
1918 : * Separately, it's possible that _bt_check_compare set continuescan=false
1919 : * for a scan key that's required in the opposite direction only. This is
1920 : * a special case, that happens only when _bt_check_compare sees that the
1921 : * inequality encountered a NULL value. This signals the end of non-NULL
1922 : * values in the current scan direction, which is reason enough to end the
1923 : * (primitive) scan. If this happens at the start of a large group of
1924 : * NULL values, then we shouldn't expect to be called again until after
1925 : * the scan has already read indefinitely-many leaf pages full of tuples
1926 : * with NULL suffix values. (_bt_first is expected to skip over the group
1927 : * of NULLs by applying a similar "deduce NOT NULL" rule of its own, which
1928 : * involves consing up an explicit SK_SEARCHNOTNULL key.)
1929 : *
1930 : * Apply a test against finaltup to detect and recover from the problem:
1931 : * if even finaltup doesn't satisfy such an inequality, we just skip by
1932 : * starting a new primitive index scan. When we skip, we know for sure
1933 : * that all of the tuples on the current page following caller's tuple are
1934 : * also before the _bt_first-wise start of tuples for our new qual. That
1935 : * at least suggests many more skippable pages beyond the current page.
1936 : * (when so->scanBehind and so->oppositeDirCheck are set, this'll happen
1937 : * when we test the next page's finaltup/high key instead.)
1938 : */
1939 125656 : else if (has_required_opposite_direction_only && pstate->finaltup &&
1940 4292 : unlikely(!_bt_oppodir_checkkeys(scan, dir, pstate->finaltup)))
1941 2 : goto new_prim_scan;
1942 :
1943 125654 : continue_scan:
1944 :
1945 : /*
1946 : * Stick with the ongoing primitive index scan for now.
1947 : *
1948 : * It's possible that later tuples will also turn out to have values that
1949 : * are still < the now-current array keys (or > the current array keys).
1950 : * Our caller will handle this by performing what amounts to a linear
1951 : * search of the page, implemented by calling _bt_check_compare and then
1952 : * _bt_tuple_before_array_skeys for each tuple.
1953 : *
1954 : * This approach has various advantages over a binary search of the page.
1955 : * Repeated binary searches of the page (one binary search for every array
1956 : * advancement) won't outperform a continuous linear search. While there
1957 : * are workloads that a naive linear search won't handle well, our caller
1958 : * has a "look ahead" fallback mechanism to deal with that problem.
1959 : */
1960 128492 : pstate->continuescan = true; /* Override _bt_check_compare */
1961 128492 : so->needPrimScan = false; /* _bt_readpage has more tuples to check */
1962 :
1963 128492 : if (so->scanBehind)
1964 : {
1965 : /*
1966 : * Remember if recheck needs to call _bt_oppodir_checkkeys for next
1967 : * page's finaltup (see above comments about "Handle inequalities
1968 : * marked required in the opposite scan direction" for why).
1969 : */
1970 2838 : so->oppositeDirCheck = has_required_opposite_direction_only;
1971 :
1972 : /*
1973 : * skip by setting "look ahead" mechanism's offnum for forwards scans
1974 : * (backwards scans check scanBehind flag directly instead)
1975 : */
1976 2838 : if (ScanDirectionIsForward(dir))
1977 2820 : pstate->skip = pstate->maxoff + 1;
1978 : }
1979 :
1980 : /* Caller's tuple doesn't match the new qual */
1981 128492 : return false;
1982 :
1983 17980 : new_prim_scan:
1984 :
1985 : Assert(pstate->finaltup); /* not on rightmost/leftmost page */
1986 :
1987 : /*
1988 : * Looks like another primitive index scan is required. But consider
1989 : * continuing the current primscan based on scan-level heuristics.
1990 : *
1991 : * Continue the ongoing primitive scan (and schedule a recheck for when
1992 : * the scan arrives on the next sibling leaf page) when it has already
1993 : * read at least one leaf page before the one we're reading now. This
1994 : * makes primscan scheduling more efficient when scanning subsets of an
1995 : * index with many distinct attribute values matching many array elements.
1996 : * It encourages fewer, larger primitive scans where that makes sense.
1997 : * This will in turn encourage _bt_readpage to apply the pstate.startikey
1998 : * optimization more often.
1999 : *
2000 : * Also continue the ongoing primitive index scan when it is still on the
2001 : * first page if there have been more than NSKIPADVANCES_THRESHOLD calls
2002 : * here that each advanced at least one of the scan's skip arrays
2003 : * (deliberately ignore advancements that only affected SAOP arrays here).
2004 : * A page that cycles through this many skip array elements is quite
2005 : * likely to neighbor similar pages, that we'll also need to read.
2006 : *
2007 : * Note: These heuristics aren't as aggressive as you might think. We're
2008 : * conservative about allowing a primitive scan to step from the first
2009 : * leaf page it reads to the page's sibling page (we only allow it on
2010 : * first pages whose finaltup strongly suggests that it'll work out, as
2011 : * well as first pages that have a large number of skip array advances).
2012 : * Clearing this first page finaltup hurdle is a strong signal in itself.
2013 : *
2014 : * Note: The NSKIPADVANCES_THRESHOLD heuristic exists only to avoid
2015 : * pathological cases. Specifically, cases where a skip scan should just
2016 : * behave like a traditional full index scan, but ends up "skipping" again
2017 : * and again, descending to the prior leaf page's direct sibling leaf page
2018 : * each time. This misbehavior would otherwise be possible during scans
2019 : * that never quite manage to "clear the first page finaltup hurdle".
2020 : */
2021 17980 : if (!pstate->firstpage || pstate->nskipadvances > NSKIPADVANCES_THRESHOLD)
2022 : {
2023 : /* Schedule a recheck once on the next (or previous) page */
2024 864 : so->scanBehind = true;
2025 :
2026 : /* Continue the current primitive scan after all */
2027 864 : goto continue_scan;
2028 : }
2029 :
2030 : /*
2031 : * End this primitive index scan, but schedule another.
2032 : *
2033 : * Note: We make a soft assumption that the current scan direction will
2034 : * also be used within _bt_next, when it is asked to step off this page.
2035 : * It is up to _bt_next to cancel this scheduled primitive index scan
2036 : * whenever it steps to a page in the direction opposite currPos.dir.
2037 : */
2038 17116 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2039 17116 : so->needPrimScan = true; /* ...but call _bt_first again */
2040 :
2041 17116 : if (scan->parallel_scan)
2042 36 : _bt_parallel_primscan_schedule(scan, so->currPos.currPage);
2043 :
2044 : /* Caller's tuple doesn't match the new qual */
2045 17116 : return false;
2046 :
2047 8296 : end_toplevel_scan:
2048 :
2049 : /*
2050 : * End the current primitive index scan, but don't schedule another.
2051 : *
2052 : * This ends the entire top-level scan in the current scan direction.
2053 : *
2054 : * Note: The scan's arrays (including any non-required arrays) are now in
2055 : * their final positions for the current scan direction. If the scan
2056 : * direction happens to change, then the arrays will already be in their
2057 : * first positions for what will then be the current scan direction.
2058 : */
2059 8296 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2060 8296 : so->needPrimScan = false; /* ...and don't call _bt_first again */
2061 :
2062 : /* Caller's tuple doesn't match any qual */
2063 8296 : return false;
2064 : }
2065 :
2066 : #ifdef USE_ASSERT_CHECKING
2067 : /*
2068 : * Verify that the scan's "so->keyData[]" scan keys are in agreement with
2069 : * its array key state
2070 : */
2071 : static bool
2072 : _bt_verify_keys_with_arraykeys(IndexScanDesc scan)
2073 : {
2074 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2075 : int last_sk_attno = InvalidAttrNumber,
2076 : arrayidx = 0;
2077 : bool nonrequiredseen = false;
2078 :
2079 : if (!so->qual_ok)
2080 : return false;
2081 :
2082 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
2083 : {
2084 : ScanKey cur = so->keyData + ikey;
2085 : BTArrayKeyInfo *array;
2086 :
2087 : if (cur->sk_strategy != BTEqualStrategyNumber ||
2088 : !(cur->sk_flags & SK_SEARCHARRAY))
2089 : continue;
2090 :
2091 : array = &so->arrayKeys[arrayidx++];
2092 : if (array->scan_key != ikey)
2093 : return false;
2094 :
2095 : if (array->num_elems == 0 || array->num_elems < -1)
2096 : return false;
2097 :
2098 : if (array->num_elems != -1 &&
2099 : cur->sk_argument != array->elem_values[array->cur_elem])
2100 : return false;
2101 : if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
2102 : {
2103 : if (last_sk_attno > cur->sk_attno)
2104 : return false;
2105 : if (nonrequiredseen)
2106 : return false;
2107 : }
2108 : else
2109 : nonrequiredseen = true;
2110 :
2111 : last_sk_attno = cur->sk_attno;
2112 : }
2113 :
2114 : if (arrayidx != so->numArrayKeys)
2115 : return false;
2116 :
2117 : return true;
2118 : }
2119 : #endif
2120 :
2121 : /*
2122 : * Test whether an indextuple satisfies all the scankey conditions.
2123 : *
2124 : * Return true if so, false if not. If the tuple fails to pass the qual,
2125 : * we also determine whether there's any need to continue the scan beyond
2126 : * this tuple, and set pstate.continuescan accordingly. See comments for
2127 : * _bt_preprocess_keys() about how this is done.
2128 : *
2129 : * Forward scan callers can pass a high key tuple in the hopes of having
2130 : * us set *continuescan to false, and avoiding an unnecessary visit to
2131 : * the page to the right.
2132 : *
2133 : * Advances the scan's array keys when necessary for arrayKeys=true callers.
2134 : * Scans without any array keys must always pass arrayKeys=false.
2135 : *
2136 : * Also stops and starts primitive index scans for arrayKeys=true callers.
2137 : * Scans with array keys are required to set up page state that helps us with
2138 : * this. The page's finaltup tuple (the page high key for a forward scan, or
2139 : * the page's first non-pivot tuple for a backward scan) must be set in
2140 : * pstate.finaltup ahead of the first call here for the page. Set this to
2141 : * NULL for rightmost page (or the leftmost page for backwards scans).
2142 : *
2143 : * scan: index scan descriptor (containing a search-type scankey)
2144 : * pstate: page level input and output parameters
2145 : * arrayKeys: should we advance the scan's array keys if necessary?
2146 : * tuple: index tuple to test
2147 : * tupnatts: number of attributes in tupnatts (high key may be truncated)
2148 : */
2149 : bool
2150 62258566 : _bt_checkkeys(IndexScanDesc scan, BTReadPageState *pstate, bool arrayKeys,
2151 : IndexTuple tuple, int tupnatts)
2152 : {
2153 62258566 : TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
2154 62258566 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2155 62258566 : ScanDirection dir = so->currPos.dir;
2156 62258566 : int ikey = pstate->startikey;
2157 : bool res;
2158 :
2159 : Assert(BTreeTupleGetNAtts(tuple, scan->indexRelation) == tupnatts);
2160 : Assert(!so->needPrimScan && !so->scanBehind && !so->oppositeDirCheck);
2161 : Assert(arrayKeys || so->numArrayKeys == 0);
2162 :
2163 62258566 : res = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, arrayKeys,
2164 62258566 : pstate->forcenonrequired, &pstate->continuescan,
2165 : &ikey);
2166 :
2167 : /*
2168 : * If _bt_check_compare relied on the pstate.startikey optimization, call
2169 : * again (in assert-enabled builds) to verify it didn't affect our answer.
2170 : *
2171 : * Note: we can't do this when !pstate.forcenonrequired, since any arrays
2172 : * before pstate.startikey won't have advanced on this page at all.
2173 : */
2174 : Assert(!pstate->forcenonrequired || arrayKeys);
2175 : #ifdef USE_ASSERT_CHECKING
2176 : if (pstate->startikey > 0 && !pstate->forcenonrequired)
2177 : {
2178 : bool dres,
2179 : dcontinuescan;
2180 : int dikey = 0;
2181 :
2182 : /* Pass arrayKeys=false to avoid array side-effects */
2183 : dres = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
2184 : pstate->forcenonrequired, &dcontinuescan,
2185 : &dikey);
2186 : Assert(res == dres);
2187 : Assert(pstate->continuescan == dcontinuescan);
2188 :
2189 : /*
2190 : * Should also get the same ikey result. We need a slightly weaker
2191 : * assertion during arrayKeys calls, since they might be using an
2192 : * array that couldn't be marked required during preprocessing.
2193 : */
2194 : Assert(arrayKeys || ikey == dikey);
2195 : Assert(ikey <= dikey);
2196 : }
2197 : #endif
2198 :
2199 : /*
2200 : * Only one _bt_check_compare call is required in the common case where
2201 : * there are no equality strategy array scan keys. Otherwise we can only
2202 : * accept _bt_check_compare's answer unreservedly when it didn't set
2203 : * pstate.continuescan=false.
2204 : */
2205 62258566 : if (!arrayKeys || pstate->continuescan)
2206 62027786 : return res;
2207 :
2208 : /*
2209 : * _bt_check_compare call set continuescan=false in the presence of
2210 : * equality type array keys. This could mean that the tuple is just past
2211 : * the end of matches for the current array keys.
2212 : *
2213 : * It's also possible that the scan is still _before_ the _start_ of
2214 : * tuples matching the current set of array keys. Check for that first.
2215 : */
2216 : Assert(!pstate->forcenonrequired);
2217 230780 : if (_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts, true,
2218 : ikey, NULL))
2219 : {
2220 : /* Override _bt_check_compare, continue primitive scan */
2221 39894 : pstate->continuescan = true;
2222 :
2223 : /*
2224 : * We will end up here repeatedly given a group of tuples > the
2225 : * previous array keys and < the now-current keys (for a backwards
2226 : * scan it's just the same, though the operators swap positions).
2227 : *
2228 : * We must avoid allowing this linear search process to scan very many
2229 : * tuples from well before the start of tuples matching the current
2230 : * array keys (or from well before the point where we'll once again
2231 : * have to advance the scan's array keys).
2232 : *
2233 : * We keep the overhead under control by speculatively "looking ahead"
2234 : * to later still-unscanned items from this same leaf page. We'll
2235 : * only attempt this once the number of tuples that the linear search
2236 : * process has examined starts to get out of hand.
2237 : */
2238 39894 : pstate->rechecks++;
2239 39894 : if (pstate->rechecks >= LOOK_AHEAD_REQUIRED_RECHECKS)
2240 : {
2241 : /* See if we should skip ahead within the current leaf page */
2242 11028 : _bt_checkkeys_look_ahead(scan, pstate, tupnatts, tupdesc);
2243 :
2244 : /*
2245 : * Might have set pstate.skip to a later page offset. When that
2246 : * happens then _bt_readpage caller will inexpensively skip ahead
2247 : * to a later tuple from the same page (the one just after the
2248 : * tuple we successfully "looked ahead" to).
2249 : */
2250 : }
2251 :
2252 : /* This indextuple doesn't match the current qual, in any case */
2253 39894 : return false;
2254 : }
2255 :
2256 : /*
2257 : * Caller's tuple is >= the current set of array keys and other equality
2258 : * constraint scan keys (or <= if this is a backwards scan). It's now
2259 : * clear that we _must_ advance any required array keys in lockstep with
2260 : * the scan.
2261 : */
2262 190886 : return _bt_advance_array_keys(scan, pstate, tuple, tupnatts, tupdesc,
2263 : ikey, true);
2264 : }
2265 :
2266 : /*
2267 : * Test whether caller's finaltup tuple is still before the start of matches
2268 : * for the current array keys.
2269 : *
2270 : * Called at the start of reading a page during a scan with array keys, though
2271 : * only when the so->scanBehind flag was set on the scan's prior page.
2272 : *
2273 : * Returns false if the tuple is still before the start of matches. When that
2274 : * happens, caller should cut its losses and start a new primitive index scan.
2275 : * Otherwise returns true.
2276 : */
2277 : bool
2278 2338 : _bt_scanbehind_checkkeys(IndexScanDesc scan, ScanDirection dir,
2279 : IndexTuple finaltup)
2280 : {
2281 2338 : Relation rel = scan->indexRelation;
2282 2338 : TupleDesc tupdesc = RelationGetDescr(rel);
2283 2338 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2284 2338 : int nfinaltupatts = BTreeTupleGetNAtts(finaltup, rel);
2285 : bool scanBehind;
2286 :
2287 : Assert(so->numArrayKeys);
2288 :
2289 2338 : if (_bt_tuple_before_array_skeys(scan, dir, finaltup, tupdesc,
2290 : nfinaltupatts, false, 0, &scanBehind))
2291 410 : return false;
2292 :
2293 : /*
2294 : * If scanBehind was set, all of the untruncated attribute values from
2295 : * finaltup that correspond to an array match the array's current element,
2296 : * but there are other keys associated with truncated suffix attributes.
2297 : * Array advancement must have incremented the scan's arrays on the
2298 : * previous page, resulting in a set of array keys that happen to be an
2299 : * exact match for the current page high key's untruncated prefix values.
2300 : *
2301 : * This page definitely doesn't contain tuples that the scan will need to
2302 : * return. The next page may or may not contain relevant tuples. Handle
2303 : * this by cutting our losses and starting a new primscan.
2304 : */
2305 1928 : if (scanBehind)
2306 0 : return false;
2307 :
2308 1928 : if (!so->oppositeDirCheck)
2309 1806 : return true;
2310 :
2311 122 : return _bt_oppodir_checkkeys(scan, dir, finaltup);
2312 : }
2313 :
2314 : /*
2315 : * Test whether an indextuple fails to satisfy an inequality required in the
2316 : * opposite direction only.
2317 : *
2318 : * Caller's finaltup tuple is the page high key (for forwards scans), or the
2319 : * first non-pivot tuple (for backwards scans). Called during scans with
2320 : * required array keys and required opposite-direction inequalities.
2321 : *
2322 : * Returns false if an inequality scan key required in the opposite direction
2323 : * only isn't satisfied (and any earlier required scan keys are satisfied).
2324 : * Otherwise returns true.
2325 : *
2326 : * An unsatisfied inequality required in the opposite direction only might
2327 : * well enable skipping over many leaf pages, provided another _bt_first call
2328 : * takes place. This type of unsatisfied inequality won't usually cause
2329 : * _bt_checkkeys to stop the scan to consider array advancement/starting a new
2330 : * primitive index scan.
2331 : */
2332 : static bool
2333 4414 : _bt_oppodir_checkkeys(IndexScanDesc scan, ScanDirection dir,
2334 : IndexTuple finaltup)
2335 : {
2336 4414 : Relation rel = scan->indexRelation;
2337 4414 : TupleDesc tupdesc = RelationGetDescr(rel);
2338 4414 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2339 4414 : int nfinaltupatts = BTreeTupleGetNAtts(finaltup, rel);
2340 : bool continuescan;
2341 4414 : ScanDirection flipped = -dir;
2342 4414 : int ikey = 0;
2343 :
2344 : Assert(so->numArrayKeys);
2345 :
2346 4414 : _bt_check_compare(scan, flipped, finaltup, nfinaltupatts, tupdesc, false,
2347 : false, &continuescan,
2348 : &ikey);
2349 :
2350 4414 : if (!continuescan && so->keyData[ikey].sk_strategy != BTEqualStrategyNumber)
2351 2 : return false;
2352 :
2353 4412 : return true;
2354 : }
2355 :
2356 : /*
2357 : * Determines an offset to the first scan key (an so->keyData[]-wise offset)
2358 : * that is _not_ guaranteed to be satisfied by every tuple from pstate.page,
2359 : * which is set in pstate.startikey for _bt_checkkeys calls for the page.
2360 : * This allows caller to save cycles on comparisons of a prefix of keys while
2361 : * reading pstate.page.
2362 : *
2363 : * Also determines if later calls to _bt_checkkeys (for pstate.page) should be
2364 : * forced to treat all required scan keys >= pstate.startikey as nonrequired
2365 : * (that is, if they're to be treated as if any SK_BT_REQFWD/SK_BT_REQBKWD
2366 : * markings that were set by preprocessing were not set at all, for the
2367 : * duration of _bt_checkkeys calls prior to the call for pstate.finaltup).
2368 : * This is indicated to caller by setting pstate.forcenonrequired.
2369 : *
2370 : * Call here at the start of reading a leaf page beyond the first one for the
2371 : * primitive index scan. We consider all non-pivot tuples, so it doesn't make
2372 : * sense to call here when only a subset of those tuples can ever be read.
2373 : * This is also a good idea on performance grounds; not calling here when on
2374 : * the first page (first for the current primitive scan) avoids wasting cycles
2375 : * during selective point queries. They typically don't stand to gain as much
2376 : * when we can set pstate.startikey, and are likely to notice the overhead of
2377 : * calling here. (Also, allowing pstate.forcenonrequired to be set on a
2378 : * primscan's first page would mislead _bt_advance_array_keys, which expects
2379 : * pstate.nskipadvances to be representative of every first page's key space.)
2380 : *
2381 : * Caller must call _bt_start_array_keys and reset startikey/forcenonrequired
2382 : * ahead of the finaltup _bt_checkkeys call when we set forcenonrequired=true.
2383 : * This will give _bt_checkkeys the opportunity to call _bt_advance_array_keys
2384 : * with sktrig_required=true, restoring the invariant that the scan's required
2385 : * arrays always track the scan's progress through the index's key space.
2386 : * Caller won't need to do this on the rightmost/leftmost page in the index
2387 : * (where pstate.finaltup isn't ever set), since forcenonrequired will never
2388 : * be set here in the first place.
2389 : */
2390 : void
2391 29956 : _bt_set_startikey(IndexScanDesc scan, BTReadPageState *pstate)
2392 : {
2393 29956 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2394 29956 : Relation rel = scan->indexRelation;
2395 29956 : TupleDesc tupdesc = RelationGetDescr(rel);
2396 : ItemId iid;
2397 : IndexTuple firsttup,
2398 : lasttup;
2399 29956 : int startikey = 0,
2400 29956 : arrayidx = 0,
2401 : firstchangingattnum;
2402 29956 : bool start_past_saop_eq = false;
2403 :
2404 : Assert(!so->scanBehind);
2405 : Assert(pstate->minoff < pstate->maxoff);
2406 : Assert(!pstate->firstpage);
2407 : Assert(pstate->startikey == 0);
2408 : Assert(!so->numArrayKeys || pstate->finaltup ||
2409 : P_RIGHTMOST(BTPageGetOpaque(pstate->page)) ||
2410 : P_LEFTMOST(BTPageGetOpaque(pstate->page)));
2411 :
2412 29956 : if (so->numberOfKeys == 0)
2413 12094 : return;
2414 :
2415 : /* minoff is an offset to the lowest non-pivot tuple on the page */
2416 17862 : iid = PageGetItemId(pstate->page, pstate->minoff);
2417 17862 : firsttup = (IndexTuple) PageGetItem(pstate->page, iid);
2418 :
2419 : /* maxoff is an offset to the highest non-pivot tuple on the page */
2420 17862 : iid = PageGetItemId(pstate->page, pstate->maxoff);
2421 17862 : lasttup = (IndexTuple) PageGetItem(pstate->page, iid);
2422 :
2423 : /* Determine the first attribute whose values change on caller's page */
2424 17862 : firstchangingattnum = _bt_keep_natts_fast(rel, firsttup, lasttup);
2425 :
2426 26740 : for (; startikey < so->numberOfKeys; startikey++)
2427 : {
2428 20430 : ScanKey key = so->keyData + startikey;
2429 : BTArrayKeyInfo *array;
2430 : Datum firstdatum,
2431 : lastdatum;
2432 : bool firstnull,
2433 : lastnull;
2434 : int32 result;
2435 :
2436 : /*
2437 : * Determine if it's safe to set pstate.startikey to an offset to a
2438 : * key that comes after this key, by examining this key
2439 : */
2440 20430 : if (key->sk_flags & SK_ROW_HEADER)
2441 : {
2442 : /* RowCompare inequality (header key) */
2443 0 : ScanKey subkey = (ScanKey) DatumGetPointer(key->sk_argument);
2444 0 : bool satisfied = false;
2445 :
2446 : for (;;)
2447 0 : {
2448 : int cmpresult;
2449 0 : bool firstsatisfies = false;
2450 :
2451 0 : if (subkey->sk_attno > firstchangingattnum) /* >, not >= */
2452 0 : break; /* unsafe, preceding attr has multiple
2453 : * distinct values */
2454 :
2455 0 : if (subkey->sk_flags & SK_ISNULL)
2456 0 : break; /* unsafe, unsatisfiable NULL subkey arg */
2457 :
2458 0 : firstdatum = index_getattr(firsttup, subkey->sk_attno,
2459 : tupdesc, &firstnull);
2460 0 : lastdatum = index_getattr(lasttup, subkey->sk_attno,
2461 : tupdesc, &lastnull);
2462 :
2463 0 : if (firstnull || lastnull)
2464 : break; /* unsafe, NULL value won't satisfy subkey */
2465 :
2466 : /*
2467 : * Compare the first tuple's datum for this row compare member
2468 : */
2469 0 : cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
2470 : subkey->sk_collation,
2471 : firstdatum,
2472 : subkey->sk_argument));
2473 0 : if (subkey->sk_flags & SK_BT_DESC)
2474 0 : INVERT_COMPARE_RESULT(cmpresult);
2475 :
2476 0 : if (cmpresult != 0 || (subkey->sk_flags & SK_ROW_END))
2477 : {
2478 0 : firstsatisfies = _bt_rowcompare_cmpresult(subkey,
2479 : cmpresult);
2480 0 : if (!firstsatisfies)
2481 : {
2482 : /* Unsafe, firstdatum does not satisfy subkey */
2483 0 : break;
2484 : }
2485 : }
2486 :
2487 : /*
2488 : * Compare the last tuple's datum for this row compare member
2489 : */
2490 0 : cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
2491 : subkey->sk_collation,
2492 : lastdatum,
2493 : subkey->sk_argument));
2494 0 : if (subkey->sk_flags & SK_BT_DESC)
2495 0 : INVERT_COMPARE_RESULT(cmpresult);
2496 :
2497 0 : if (cmpresult != 0 || (subkey->sk_flags & SK_ROW_END))
2498 : {
2499 0 : if (!firstsatisfies)
2500 : {
2501 : /*
2502 : * It's only safe to set startikey beyond the row
2503 : * compare header key when both firsttup and lasttup
2504 : * satisfy the key as a whole based on the same
2505 : * deciding subkey/attribute. That can't happen now.
2506 : */
2507 0 : break; /* unsafe */
2508 : }
2509 :
2510 0 : satisfied = _bt_rowcompare_cmpresult(subkey, cmpresult);
2511 0 : break; /* safe iff 'satisfied' is true */
2512 : }
2513 :
2514 : /* Move on to next row member/subkey */
2515 0 : if (subkey->sk_flags & SK_ROW_END)
2516 0 : break; /* defensive */
2517 0 : subkey++;
2518 :
2519 : /*
2520 : * We deliberately don't check if the next subkey has the same
2521 : * strategy as this iteration's subkey (which happens when
2522 : * subkeys for both ASC and DESC columns are used together),
2523 : * nor if any subkey is marked required. This is safe because
2524 : * in general all prior index attributes must have only one
2525 : * distinct value (across all of the tuples on the page) in
2526 : * order for us to even consider any subkey's attribute.
2527 : */
2528 : }
2529 :
2530 0 : if (satisfied)
2531 : {
2532 : /* Safe, row compare satisfied by every tuple on page */
2533 8630 : continue;
2534 : }
2535 :
2536 11552 : break; /* unsafe */
2537 : }
2538 20430 : if (key->sk_strategy != BTEqualStrategyNumber)
2539 : {
2540 : /*
2541 : * Scalar inequality key.
2542 : *
2543 : * It's definitely safe for _bt_checkkeys to avoid assessing this
2544 : * inequality when the page's first and last non-pivot tuples both
2545 : * satisfy the inequality (since the same must also be true of all
2546 : * the tuples in between these two).
2547 : *
2548 : * Unlike the "=" case, it doesn't matter if this attribute has
2549 : * more than one distinct value (though it _is_ necessary for any
2550 : * and all _prior_ attributes to contain no more than one distinct
2551 : * value amongst all of the tuples from pstate.page).
2552 : */
2553 4638 : if (key->sk_attno > firstchangingattnum) /* >, not >= */
2554 356 : break; /* unsafe, preceding attr has multiple
2555 : * distinct values */
2556 :
2557 4282 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc, &firstnull);
2558 4282 : lastdatum = index_getattr(lasttup, key->sk_attno, tupdesc, &lastnull);
2559 :
2560 4282 : if (key->sk_flags & SK_ISNULL)
2561 : {
2562 : /* IS NOT NULL key */
2563 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
2564 :
2565 110 : if (firstnull || lastnull)
2566 : break; /* unsafe */
2567 :
2568 : /* Safe, IS NOT NULL key satisfied by every tuple */
2569 110 : continue;
2570 : }
2571 :
2572 : /* Test firsttup */
2573 4172 : if (firstnull ||
2574 4172 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2575 : key->sk_collation, firstdatum,
2576 : key->sk_argument)))
2577 : break; /* unsafe */
2578 :
2579 : /* Test lasttup */
2580 4172 : if (lastnull ||
2581 4172 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2582 : key->sk_collation, lastdatum,
2583 : key->sk_argument)))
2584 : break; /* unsafe */
2585 :
2586 : /* Safe, scalar inequality satisfied by every tuple */
2587 4080 : continue;
2588 : }
2589 :
2590 : /* Some = key (could be a scalar = key, could be an array = key) */
2591 : Assert(key->sk_strategy == BTEqualStrategyNumber);
2592 :
2593 15792 : if (!(key->sk_flags & SK_SEARCHARRAY))
2594 : {
2595 : /*
2596 : * Scalar = key (possibly an IS NULL key).
2597 : *
2598 : * It is unsafe to set pstate.startikey to an ikey beyond this
2599 : * key, unless the = key is satisfied by every possible tuple on
2600 : * the page (possible only when attribute has just one distinct
2601 : * value among all tuples on the page).
2602 : */
2603 12556 : if (key->sk_attno >= firstchangingattnum)
2604 10224 : break; /* unsafe, multiple distinct attr values */
2605 :
2606 2332 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc,
2607 : &firstnull);
2608 2332 : if (key->sk_flags & SK_ISNULL)
2609 : {
2610 : /* IS NULL key */
2611 : Assert(key->sk_flags & SK_SEARCHNULL);
2612 :
2613 0 : if (!firstnull)
2614 0 : break; /* unsafe */
2615 :
2616 : /* Safe, IS NULL key satisfied by every tuple */
2617 0 : continue;
2618 : }
2619 2332 : if (firstnull ||
2620 2332 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2621 : key->sk_collation, firstdatum,
2622 : key->sk_argument)))
2623 : break; /* unsafe */
2624 :
2625 : /* Safe, scalar = key satisfied by every tuple */
2626 2332 : continue;
2627 : }
2628 :
2629 : /* = array key (could be a SAOP array, could be a skip array) */
2630 3236 : array = &so->arrayKeys[arrayidx++];
2631 : Assert(array->scan_key == startikey);
2632 3236 : if (array->num_elems != -1)
2633 : {
2634 : /*
2635 : * SAOP array = key.
2636 : *
2637 : * Handle this like we handle scalar = keys (though binary search
2638 : * for a matching element, to avoid relying on key's sk_argument).
2639 : */
2640 818 : if (key->sk_attno >= firstchangingattnum)
2641 818 : break; /* unsafe, multiple distinct attr values */
2642 :
2643 0 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc,
2644 : &firstnull);
2645 0 : _bt_binsrch_array_skey(&so->orderProcs[startikey],
2646 : false, NoMovementScanDirection,
2647 : firstdatum, firstnull, array, key,
2648 : &result);
2649 0 : if (result != 0)
2650 0 : break; /* unsafe */
2651 :
2652 : /* Safe, SAOP = key satisfied by every tuple */
2653 0 : start_past_saop_eq = true;
2654 0 : continue;
2655 : }
2656 :
2657 : /*
2658 : * Skip array = key
2659 : */
2660 : Assert(key->sk_flags & SK_BT_SKIP);
2661 2418 : if (array->null_elem)
2662 : {
2663 : /*
2664 : * Non-range skip array = key.
2665 : *
2666 : * Safe, non-range skip array "satisfied" by every tuple on page
2667 : * (safe even when "key->sk_attno > firstchangingattnum").
2668 : */
2669 2108 : continue;
2670 : }
2671 :
2672 : /*
2673 : * Range skip array = key.
2674 : *
2675 : * Handle this like we handle scalar inequality keys (but avoid using
2676 : * key's sk_argument directly, as in the SAOP array case).
2677 : */
2678 310 : if (key->sk_attno > firstchangingattnum) /* >, not >= */
2679 48 : break; /* unsafe, preceding attr has multiple
2680 : * distinct values */
2681 :
2682 262 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc, &firstnull);
2683 262 : lastdatum = index_getattr(lasttup, key->sk_attno, tupdesc, &lastnull);
2684 :
2685 : /* Test firsttup */
2686 262 : _bt_binsrch_skiparray_skey(false, ForwardScanDirection,
2687 : firstdatum, firstnull, array, key,
2688 : &result);
2689 262 : if (result != 0)
2690 0 : break; /* unsafe */
2691 :
2692 : /* Test lasttup */
2693 262 : _bt_binsrch_skiparray_skey(false, ForwardScanDirection,
2694 : lastdatum, lastnull, array, key,
2695 : &result);
2696 262 : if (result != 0)
2697 14 : break; /* unsafe */
2698 :
2699 : /* Safe, range skip array satisfied by every tuple on page */
2700 : }
2701 :
2702 : /*
2703 : * Use of forcenonrequired is typically undesirable, since it'll force
2704 : * _bt_readpage caller to read every tuple on the page -- even though, in
2705 : * general, it might well be possible to end the scan on an earlier tuple.
2706 : * However, caller must use forcenonrequired when start_past_saop_eq=true,
2707 : * since the usual required array behavior might fail to roll over to the
2708 : * SAOP array.
2709 : *
2710 : * We always prefer forcenonrequired=true during scans with skip arrays
2711 : * (except on the first page of each primitive index scan), though -- even
2712 : * when "startikey == 0". That way, _bt_advance_array_keys's low-order
2713 : * key precheck optimization can always be used (unless on the first page
2714 : * of the scan). It seems slightly preferable to check more tuples when
2715 : * that allows us to do significantly less skip array maintenance.
2716 : */
2717 17862 : pstate->forcenonrequired = (start_past_saop_eq || so->skipScan);
2718 17862 : pstate->startikey = startikey;
2719 :
2720 : /*
2721 : * _bt_readpage caller is required to call _bt_checkkeys against page's
2722 : * finaltup with forcenonrequired=false whenever we initially set
2723 : * forcenonrequired=true. That way the scan's arrays will reliably track
2724 : * its progress through the index's key space.
2725 : *
2726 : * We don't expect this when _bt_readpage caller has no finaltup due to
2727 : * its page being the rightmost (or the leftmost, during backwards scans).
2728 : * When we see that _bt_readpage has no finaltup, back out of everything.
2729 : */
2730 : Assert(!pstate->forcenonrequired || so->numArrayKeys);
2731 17862 : if (pstate->forcenonrequired && !pstate->finaltup)
2732 : {
2733 458 : pstate->forcenonrequired = false;
2734 458 : pstate->startikey = 0;
2735 : }
2736 : }
2737 :
2738 : /*
2739 : * Test whether an indextuple satisfies current scan condition.
2740 : *
2741 : * Return true if so, false if not. If not, also sets *continuescan to false
2742 : * when it's also not possible for any later tuples to pass the current qual
2743 : * (with the scan's current set of array keys, in the current scan direction),
2744 : * in addition to setting *ikey to the so->keyData[] subscript/offset for the
2745 : * unsatisfied scan key (needed when caller must consider advancing the scan's
2746 : * array keys).
2747 : *
2748 : * This is a subroutine for _bt_checkkeys. We provisionally assume that
2749 : * reaching the end of the current set of required keys (in particular the
2750 : * current required array keys) ends the ongoing (primitive) index scan.
2751 : * Callers without array keys should just end the scan right away when they
2752 : * find that continuescan has been set to false here by us. Things are more
2753 : * complicated for callers with array keys.
2754 : *
2755 : * Callers with array keys must first consider advancing the arrays when
2756 : * continuescan has been set to false here by us. They must then consider if
2757 : * it really does make sense to end the current (primitive) index scan, in
2758 : * light of everything that is known at that point. (In general when we set
2759 : * continuescan=false for these callers it must be treated as provisional.)
2760 : *
2761 : * We deal with advancing unsatisfied non-required arrays directly, though.
2762 : * This is safe, since by definition non-required keys can't end the scan.
2763 : * This is just how we determine if non-required arrays are just unsatisfied
2764 : * by the current array key, or if they're truly unsatisfied (that is, if
2765 : * they're unsatisfied by every possible array key).
2766 : *
2767 : * Pass advancenonrequired=false to avoid all array related side effects.
2768 : * This allows _bt_advance_array_keys caller to avoid infinite recursion.
2769 : *
2770 : * Pass forcenonrequired=true to instruct us to treat all keys as nonrequired.
2771 : * This is used to make it safe to temporarily stop properly maintaining the
2772 : * scan's required arrays. _bt_checkkeys caller (_bt_readpage, actually)
2773 : * determines a prefix of keys that must satisfy every possible corresponding
2774 : * index attribute value from its page, which is passed to us via *ikey arg
2775 : * (this is the first key that might be unsatisfied by tuples on the page).
2776 : * Obviously, we won't maintain any array keys from before *ikey, so it's
2777 : * quite possible for such arrays to "fall behind" the index's keyspace.
2778 : * Caller will need to "catch up" by passing forcenonrequired=true (alongside
2779 : * an *ikey=0) once the page's finaltup is reached.
2780 : *
2781 : * Note: it's safe to pass an *ikey > 0 with forcenonrequired=false, but only
2782 : * when caller determines that it won't affect array maintenance.
2783 : */
2784 : static bool
2785 62318756 : _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
2786 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
2787 : bool advancenonrequired, bool forcenonrequired,
2788 : bool *continuescan, int *ikey)
2789 : {
2790 62318756 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2791 :
2792 62318756 : *continuescan = true; /* default assumption */
2793 :
2794 119924108 : for (; *ikey < so->numberOfKeys; (*ikey)++)
2795 : {
2796 70949324 : ScanKey key = so->keyData + *ikey;
2797 : Datum datum;
2798 : bool isNull;
2799 70949324 : bool requiredSameDir = false,
2800 70949324 : requiredOppositeDirOnly = false;
2801 :
2802 : /*
2803 : * Check if the key is required in the current scan direction, in the
2804 : * opposite scan direction _only_, or in neither direction (except
2805 : * when we're forced to treat all scan keys as nonrequired)
2806 : */
2807 70949324 : if (forcenonrequired)
2808 : {
2809 : /* treating scan's keys as non-required */
2810 : }
2811 70544780 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
2812 14994768 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
2813 55574752 : requiredSameDir = true;
2814 14970028 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsBackward(dir)) ||
2815 6055734 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsForward(dir)))
2816 14970028 : requiredOppositeDirOnly = true;
2817 :
2818 70949324 : if (key->sk_attno > tupnatts)
2819 : {
2820 : /*
2821 : * This attribute is truncated (must be high key). The value for
2822 : * this attribute in the first non-pivot tuple on the page to the
2823 : * right could be any possible value. Assume that truncated
2824 : * attribute passes the qual.
2825 : */
2826 : Assert(BTreeTupleIsPivot(tuple));
2827 19327060 : continue;
2828 : }
2829 :
2830 : /*
2831 : * A skip array scan key uses one of several sentinel values. We just
2832 : * fall back on _bt_tuple_before_array_skeys when we see such a value.
2833 : */
2834 70947290 : if (key->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL |
2835 : SK_BT_NEXT | SK_BT_PRIOR))
2836 : {
2837 : Assert(key->sk_flags & SK_SEARCHARRAY);
2838 : Assert(key->sk_flags & SK_BT_SKIP);
2839 : Assert(requiredSameDir || forcenonrequired);
2840 :
2841 : /*
2842 : * Cannot fall back on _bt_tuple_before_array_skeys when we're
2843 : * treating the scan's keys as nonrequired, though. Just handle
2844 : * this like any other non-required equality-type array key.
2845 : */
2846 36344 : if (forcenonrequired)
2847 13343972 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2848 : tupdesc, *ikey, false);
2849 :
2850 34316 : *continuescan = false;
2851 34316 : return false;
2852 : }
2853 :
2854 : /* row-comparison keys need special processing */
2855 70910946 : if (key->sk_flags & SK_ROW_HEADER)
2856 : {
2857 2454 : if (_bt_check_rowcompare(key, tuple, tupnatts, tupdesc, dir,
2858 : forcenonrequired, continuescan))
2859 2388 : continue;
2860 66 : return false;
2861 : }
2862 :
2863 70908492 : datum = index_getattr(tuple,
2864 70908492 : key->sk_attno,
2865 : tupdesc,
2866 : &isNull);
2867 :
2868 70908492 : if (key->sk_flags & SK_ISNULL)
2869 : {
2870 : /* Handle IS NULL/NOT NULL tests */
2871 19340410 : if (key->sk_flags & SK_SEARCHNULL)
2872 : {
2873 18128 : if (isNull)
2874 428 : continue; /* tuple satisfies this qual */
2875 : }
2876 : else
2877 : {
2878 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
2879 : Assert(!(key->sk_flags & SK_BT_SKIP));
2880 19322282 : if (!isNull)
2881 19322210 : continue; /* tuple satisfies this qual */
2882 : }
2883 :
2884 : /*
2885 : * Tuple fails this qual. If it's a required qual for the current
2886 : * scan direction, then we can conclude no further tuples will
2887 : * pass, either.
2888 : */
2889 17772 : if (requiredSameDir)
2890 204 : *continuescan = false;
2891 17568 : else if (unlikely(key->sk_flags & SK_BT_SKIP))
2892 : {
2893 : /*
2894 : * If we're treating scan keys as nonrequired, and encounter a
2895 : * skip array scan key whose current element is NULL, then it
2896 : * must be a non-range skip array. It must be satisfied, so
2897 : * there's no need to call _bt_advance_array_keys to check.
2898 : */
2899 : Assert(forcenonrequired && *ikey > 0);
2900 0 : continue;
2901 : }
2902 :
2903 : /*
2904 : * This indextuple doesn't match the qual.
2905 : */
2906 17772 : return false;
2907 : }
2908 :
2909 51568082 : if (isNull)
2910 : {
2911 : /*
2912 : * Scalar scan key isn't satisfied by NULL tuple value.
2913 : *
2914 : * If we're treating scan keys as nonrequired, and key is for a
2915 : * skip array, then we must attempt to advance the array to NULL
2916 : * (if we're successful then the tuple might match the qual).
2917 : */
2918 228 : if (unlikely(forcenonrequired && key->sk_flags & SK_BT_SKIP))
2919 0 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2920 : tupdesc, *ikey, false);
2921 :
2922 228 : if (key->sk_flags & SK_BT_NULLS_FIRST)
2923 : {
2924 : /*
2925 : * Since NULLs are sorted before non-NULLs, we know we have
2926 : * reached the lower limit of the range of values for this
2927 : * index attr. On a backward scan, we can stop if this qual
2928 : * is one of the "must match" subset. We can stop regardless
2929 : * of whether the qual is > or <, so long as it's required,
2930 : * because it's not possible for any future tuples to pass. On
2931 : * a forward scan, however, we must keep going, because we may
2932 : * have initially positioned to the start of the index.
2933 : * (_bt_advance_array_keys also relies on this behavior during
2934 : * forward scans.)
2935 : */
2936 0 : if ((requiredSameDir || requiredOppositeDirOnly) &&
2937 : ScanDirectionIsBackward(dir))
2938 0 : *continuescan = false;
2939 : }
2940 : else
2941 : {
2942 : /*
2943 : * Since NULLs are sorted after non-NULLs, we know we have
2944 : * reached the upper limit of the range of values for this
2945 : * index attr. On a forward scan, we can stop if this qual is
2946 : * one of the "must match" subset. We can stop regardless of
2947 : * whether the qual is > or <, so long as it's required,
2948 : * because it's not possible for any future tuples to pass. On
2949 : * a backward scan, however, we must keep going, because we
2950 : * may have initially positioned to the end of the index.
2951 : * (_bt_advance_array_keys also relies on this behavior during
2952 : * backward scans.)
2953 : */
2954 228 : if ((requiredSameDir || requiredOppositeDirOnly) &&
2955 : ScanDirectionIsForward(dir))
2956 222 : *continuescan = false;
2957 : }
2958 :
2959 : /*
2960 : * This indextuple doesn't match the qual.
2961 : */
2962 228 : return false;
2963 : }
2964 :
2965 51567854 : if (!DatumGetBool(FunctionCall2Coll(&key->sk_func, key->sk_collation,
2966 : datum, key->sk_argument)))
2967 : {
2968 : /*
2969 : * Tuple fails this qual. If it's a required qual for the current
2970 : * scan direction, then we can conclude no further tuples will
2971 : * pass, either.
2972 : *
2973 : * Note: because we stop the scan as soon as any required equality
2974 : * qual fails, it is critical that equality quals be used for the
2975 : * initial positioning in _bt_first() when they are available. See
2976 : * comments in _bt_first().
2977 : */
2978 13289562 : if (requiredSameDir)
2979 12934614 : *continuescan = false;
2980 :
2981 : /*
2982 : * If this is a non-required equality-type array key, the tuple
2983 : * needs to be checked against every possible array key. Handle
2984 : * this by "advancing" the scan key's array to a matching value
2985 : * (if we're successful then the tuple might match the qual).
2986 : */
2987 354948 : else if (advancenonrequired &&
2988 347506 : key->sk_strategy == BTEqualStrategyNumber &&
2989 271726 : (key->sk_flags & SK_SEARCHARRAY))
2990 6642 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2991 : tupdesc, *ikey, false);
2992 :
2993 : /*
2994 : * This indextuple doesn't match the qual.
2995 : */
2996 13282920 : return false;
2997 : }
2998 : }
2999 :
3000 : /* If we get here, the tuple passes all index quals. */
3001 48974784 : return true;
3002 : }
3003 :
3004 : /*
3005 : * Call here when a row compare member returns a non-zero result, or with the
3006 : * result for the final ROW_END row compare member (no matter the cmpresult).
3007 : *
3008 : * cmpresult indicates the overall result of the row comparison (must already
3009 : * be commuted for DESC subkeys), and subkey is the deciding row member.
3010 : */
3011 : static bool
3012 2388 : _bt_rowcompare_cmpresult(ScanKey subkey, int cmpresult)
3013 : {
3014 : bool satisfied;
3015 :
3016 2388 : switch (subkey->sk_strategy)
3017 : {
3018 186 : case BTLessStrategyNumber:
3019 186 : satisfied = (cmpresult < 0);
3020 186 : break;
3021 1584 : case BTLessEqualStrategyNumber:
3022 1584 : satisfied = (cmpresult <= 0);
3023 1584 : break;
3024 246 : case BTGreaterEqualStrategyNumber:
3025 246 : satisfied = (cmpresult >= 0);
3026 246 : break;
3027 372 : case BTGreaterStrategyNumber:
3028 372 : satisfied = (cmpresult > 0);
3029 372 : break;
3030 0 : default:
3031 : /* EQ and NE cases aren't allowed here */
3032 0 : elog(ERROR, "unexpected strategy number %d", subkey->sk_strategy);
3033 : satisfied = false; /* keep compiler quiet */
3034 : break;
3035 : }
3036 :
3037 2388 : return satisfied;
3038 : }
3039 :
3040 : /*
3041 : * Test whether an indextuple satisfies a row-comparison scan condition.
3042 : *
3043 : * Return true if so, false if not. If not, also clear *continuescan if
3044 : * it's not possible for any future tuples in the current scan direction
3045 : * to pass the qual.
3046 : *
3047 : * This is a subroutine for _bt_checkkeys/_bt_check_compare.
3048 : */
3049 : static bool
3050 2454 : _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
3051 : TupleDesc tupdesc, ScanDirection dir,
3052 : bool forcenonrequired, bool *continuescan)
3053 : {
3054 2454 : ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
3055 2454 : int32 cmpresult = 0;
3056 : bool result;
3057 :
3058 : /* First subkey should be same as the header says */
3059 : Assert(subkey->sk_attno == skey->sk_attno);
3060 :
3061 : /* Loop over columns of the row condition */
3062 : for (;;)
3063 240 : {
3064 : Datum datum;
3065 : bool isNull;
3066 :
3067 : Assert(subkey->sk_flags & SK_ROW_MEMBER);
3068 :
3069 : /* When a NULL row member is compared, the row never matches */
3070 2694 : if (subkey->sk_flags & SK_ISNULL)
3071 : {
3072 : /*
3073 : * Unlike the simple-scankey case, this isn't a disallowed case
3074 : * (except when it's the first row element that has the NULL arg).
3075 : * But it can never match. If all the earlier row comparison
3076 : * columns are required for the scan direction, we can stop the
3077 : * scan, because there can't be another tuple that will succeed.
3078 : */
3079 : Assert(subkey != (ScanKey) DatumGetPointer(skey->sk_argument));
3080 12 : subkey--;
3081 12 : if (forcenonrequired)
3082 : {
3083 : /* treating scan's keys as non-required */
3084 : }
3085 12 : else if ((subkey->sk_flags & SK_BT_REQFWD) &&
3086 : ScanDirectionIsForward(dir))
3087 6 : *continuescan = false;
3088 6 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
3089 : ScanDirectionIsBackward(dir))
3090 6 : *continuescan = false;
3091 66 : return false;
3092 : }
3093 :
3094 2682 : if (subkey->sk_attno > tupnatts)
3095 : {
3096 : /*
3097 : * This attribute is truncated (must be high key). The value for
3098 : * this attribute in the first non-pivot tuple on the page to the
3099 : * right could be any possible value. Assume that truncated
3100 : * attribute passes the qual.
3101 : */
3102 : Assert(BTreeTupleIsPivot(tuple));
3103 6 : return true;
3104 : }
3105 :
3106 2676 : datum = index_getattr(tuple,
3107 2676 : subkey->sk_attno,
3108 : tupdesc,
3109 : &isNull);
3110 :
3111 2676 : if (isNull)
3112 : {
3113 : int reqflags;
3114 :
3115 48 : if (forcenonrequired)
3116 : {
3117 : /* treating scan's keys as non-required */
3118 : }
3119 48 : else if (subkey->sk_flags & SK_BT_NULLS_FIRST)
3120 : {
3121 : /*
3122 : * Since NULLs are sorted before non-NULLs, we know we have
3123 : * reached the lower limit of the range of values for this
3124 : * index attr. On a backward scan, we can stop if this qual
3125 : * is one of the "must match" subset. However, on a forwards
3126 : * scan, we must keep going, because we may have initially
3127 : * positioned to the start of the index.
3128 : *
3129 : * All required NULLS FIRST > row members can use NULL tuple
3130 : * values to end backwards scans, just like with other values.
3131 : * A qual "WHERE (a, b, c) > (9, 42, 'foo')" can terminate a
3132 : * backwards scan upon reaching the index's rightmost "a = 9"
3133 : * tuple whose "b" column contains a NULL (if not sooner).
3134 : * Since "b" is NULLS FIRST, we can treat its NULLs as "<" 42.
3135 : */
3136 0 : reqflags = SK_BT_REQBKWD;
3137 :
3138 : /*
3139 : * When a most significant required NULLS FIRST < row compare
3140 : * member sees NULL tuple values during a backwards scan, it
3141 : * signals the end of matches for the whole row compare/scan.
3142 : * A qual "WHERE (a, b, c) < (9, 42, 'foo')" will terminate a
3143 : * backwards scan upon reaching the rightmost tuple whose "a"
3144 : * column has a NULL. The "a" NULL value is "<" 9, and yet
3145 : * our < row compare will still end the scan. (This isn't
3146 : * safe with later/lower-order row members. Notice that it
3147 : * can only happen with an "a" NULL some time after the scan
3148 : * completely stops needing to use its "b" and "c" members.)
3149 : */
3150 0 : if (subkey == (ScanKey) DatumGetPointer(skey->sk_argument))
3151 0 : reqflags |= SK_BT_REQFWD; /* safe, first row member */
3152 :
3153 0 : if ((subkey->sk_flags & reqflags) &&
3154 : ScanDirectionIsBackward(dir))
3155 0 : *continuescan = false;
3156 : }
3157 : else
3158 : {
3159 : /*
3160 : * Since NULLs are sorted after non-NULLs, we know we have
3161 : * reached the upper limit of the range of values for this
3162 : * index attr. On a forward scan, we can stop if this qual is
3163 : * one of the "must match" subset. However, on a backward
3164 : * scan, we must keep going, because we may have initially
3165 : * positioned to the end of the index.
3166 : *
3167 : * All required NULLS LAST < row members can use NULL tuple
3168 : * values to end forwards scans, just like with other values.
3169 : * A qual "WHERE (a, b, c) < (9, 42, 'foo')" can terminate a
3170 : * forwards scan upon reaching the index's leftmost "a = 9"
3171 : * tuple whose "b" column contains a NULL (if not sooner).
3172 : * Since "b" is NULLS LAST, we can treat its NULLs as ">" 42.
3173 : */
3174 48 : reqflags = SK_BT_REQFWD;
3175 :
3176 : /*
3177 : * When a most significant required NULLS LAST > row compare
3178 : * member sees NULL tuple values during a forwards scan, it
3179 : * signals the end of matches for the whole row compare/scan.
3180 : * A qual "WHERE (a, b, c) > (9, 42, 'foo')" will terminate a
3181 : * forwards scan upon reaching the leftmost tuple whose "a"
3182 : * column has a NULL. The "a" NULL value is ">" 9, and yet
3183 : * our > row compare will end the scan. (This isn't safe with
3184 : * later/lower-order row members. Notice that it can only
3185 : * happen with an "a" NULL some time after the scan completely
3186 : * stops needing to use its "b" and "c" members.)
3187 : */
3188 48 : if (subkey == (ScanKey) DatumGetPointer(skey->sk_argument))
3189 0 : reqflags |= SK_BT_REQBKWD; /* safe, first row member */
3190 :
3191 48 : if ((subkey->sk_flags & reqflags) &&
3192 : ScanDirectionIsForward(dir))
3193 0 : *continuescan = false;
3194 : }
3195 :
3196 : /*
3197 : * In any case, this indextuple doesn't match the qual.
3198 : */
3199 48 : return false;
3200 : }
3201 :
3202 : /* Perform the test --- three-way comparison not bool operator */
3203 2628 : cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
3204 : subkey->sk_collation,
3205 : datum,
3206 : subkey->sk_argument));
3207 :
3208 2628 : if (subkey->sk_flags & SK_BT_DESC)
3209 0 : INVERT_COMPARE_RESULT(cmpresult);
3210 :
3211 : /* Done comparing if unequal, else advance to next column */
3212 2628 : if (cmpresult != 0)
3213 2388 : break;
3214 :
3215 240 : if (subkey->sk_flags & SK_ROW_END)
3216 0 : break;
3217 240 : subkey++;
3218 : }
3219 :
3220 : /* Final subkey/column determines if row compare is satisfied */
3221 2388 : result = _bt_rowcompare_cmpresult(subkey, cmpresult);
3222 :
3223 2388 : if (!result && !forcenonrequired)
3224 : {
3225 : /*
3226 : * Tuple fails this qual. If it's a required qual for the current
3227 : * scan direction, then we can conclude no further tuples will pass,
3228 : * either. Note we have to look at the deciding column, not
3229 : * necessarily the first or last column of the row condition.
3230 : */
3231 6 : if ((subkey->sk_flags & SK_BT_REQFWD) &&
3232 : ScanDirectionIsForward(dir))
3233 6 : *continuescan = false;
3234 0 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
3235 : ScanDirectionIsBackward(dir))
3236 0 : *continuescan = false;
3237 : }
3238 :
3239 2388 : return result;
3240 : }
3241 :
3242 : /*
3243 : * Determine if a scan with array keys should skip over uninteresting tuples.
3244 : *
3245 : * This is a subroutine for _bt_checkkeys. Called when _bt_readpage's linear
3246 : * search process (started after it finishes reading an initial group of
3247 : * matching tuples, used to locate the start of the next group of tuples
3248 : * matching the next set of required array keys) has already scanned an
3249 : * excessive number of tuples whose key space is "between arrays".
3250 : *
3251 : * When we perform look ahead successfully, we'll sets pstate.skip, which
3252 : * instructs _bt_readpage to skip ahead to that tuple next (could be past the
3253 : * end of the scan's leaf page). Pages where the optimization is effective
3254 : * will generally still need to skip several times. Each call here performs
3255 : * only a single "look ahead" comparison of a later tuple, whose distance from
3256 : * the current tuple's offset number is determined by applying heuristics.
3257 : */
3258 : static void
3259 11028 : _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
3260 : int tupnatts, TupleDesc tupdesc)
3261 : {
3262 11028 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3263 11028 : ScanDirection dir = so->currPos.dir;
3264 : OffsetNumber aheadoffnum;
3265 : IndexTuple ahead;
3266 :
3267 : Assert(!pstate->forcenonrequired);
3268 :
3269 : /* Avoid looking ahead when comparing the page high key */
3270 11028 : if (pstate->offnum < pstate->minoff)
3271 0 : return;
3272 :
3273 : /*
3274 : * Don't look ahead when there aren't enough tuples remaining on the page
3275 : * (in the current scan direction) for it to be worth our while
3276 : */
3277 11028 : if (ScanDirectionIsForward(dir) &&
3278 10950 : pstate->offnum >= pstate->maxoff - LOOK_AHEAD_DEFAULT_DISTANCE)
3279 326 : return;
3280 10702 : else if (ScanDirectionIsBackward(dir) &&
3281 78 : pstate->offnum <= pstate->minoff + LOOK_AHEAD_DEFAULT_DISTANCE)
3282 24 : return;
3283 :
3284 : /*
3285 : * The look ahead distance starts small, and ramps up as each call here
3286 : * allows _bt_readpage to skip over more tuples
3287 : */
3288 10678 : if (!pstate->targetdistance)
3289 6368 : pstate->targetdistance = LOOK_AHEAD_DEFAULT_DISTANCE;
3290 4310 : else if (pstate->targetdistance < MaxIndexTuplesPerPage / 2)
3291 4310 : pstate->targetdistance *= 2;
3292 :
3293 : /* Don't read past the end (or before the start) of the page, though */
3294 10678 : if (ScanDirectionIsForward(dir))
3295 10624 : aheadoffnum = Min((int) pstate->maxoff,
3296 : (int) pstate->offnum + pstate->targetdistance);
3297 : else
3298 54 : aheadoffnum = Max((int) pstate->minoff,
3299 : (int) pstate->offnum - pstate->targetdistance);
3300 :
3301 10678 : ahead = (IndexTuple) PageGetItem(pstate->page,
3302 10678 : PageGetItemId(pstate->page, aheadoffnum));
3303 10678 : if (_bt_tuple_before_array_skeys(scan, dir, ahead, tupdesc, tupnatts,
3304 : false, 0, NULL))
3305 : {
3306 : /*
3307 : * Success -- instruct _bt_readpage to skip ahead to very next tuple
3308 : * after the one we determined was still before the current array keys
3309 : */
3310 3600 : if (ScanDirectionIsForward(dir))
3311 3564 : pstate->skip = aheadoffnum + 1;
3312 : else
3313 36 : pstate->skip = aheadoffnum - 1;
3314 : }
3315 : else
3316 : {
3317 : /*
3318 : * Failure -- "ahead" tuple is too far ahead (we were too aggressive).
3319 : *
3320 : * Reset the number of rechecks, and aggressively reduce the target
3321 : * distance (we're much more aggressive here than we were when the
3322 : * distance was initially ramped up).
3323 : */
3324 7078 : pstate->rechecks = 0;
3325 7078 : pstate->targetdistance = Max(pstate->targetdistance / 8, 1);
3326 : }
3327 : }
3328 :
3329 : /*
3330 : * _bt_killitems - set LP_DEAD state for items an indexscan caller has
3331 : * told us were killed
3332 : *
3333 : * scan->opaque, referenced locally through so, contains information about the
3334 : * current page and killed tuples thereon (generally, this should only be
3335 : * called if so->numKilled > 0).
3336 : *
3337 : * Caller should not have a lock on the so->currPos page, but must hold a
3338 : * buffer pin when !so->dropPin. When we return, it still won't be locked.
3339 : * It'll continue to hold whatever pins were held before calling here.
3340 : *
3341 : * We match items by heap TID before assuming they are the right ones to set
3342 : * LP_DEAD. If the scan is one that holds a buffer pin on the target page
3343 : * continuously from initially reading the items until applying this function
3344 : * (if it is a !so->dropPin scan), VACUUM cannot have deleted any items on the
3345 : * page, so the page's TIDs can't have been recycled by now. There's no risk
3346 : * that we'll confuse a new index tuple that happens to use a recycled TID
3347 : * with a now-removed tuple with the same TID (that used to be on this same
3348 : * page). We can't rely on that during scans that drop buffer pins eagerly
3349 : * (so->dropPin scans), though, so we must condition setting LP_DEAD bits on
3350 : * the page LSN having not changed since back when _bt_readpage saw the page.
3351 : * We totally give up on setting LP_DEAD bits when the page LSN changed.
3352 : *
3353 : * We give up much less often during !so->dropPin scans, but it still happens.
3354 : * We cope with cases where items have moved right due to insertions. If an
3355 : * item has moved off the current page due to a split, we'll fail to find it
3356 : * and just give up on it.
3357 : */
3358 : void
3359 171882 : _bt_killitems(IndexScanDesc scan)
3360 : {
3361 171882 : Relation rel = scan->indexRelation;
3362 171882 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3363 : Page page;
3364 : BTPageOpaque opaque;
3365 : OffsetNumber minoff;
3366 : OffsetNumber maxoff;
3367 171882 : int numKilled = so->numKilled;
3368 171882 : bool killedsomething = false;
3369 : Buffer buf;
3370 :
3371 : Assert(numKilled > 0);
3372 : Assert(BTScanPosIsValid(so->currPos));
3373 : Assert(scan->heapRelation != NULL); /* can't be a bitmap index scan */
3374 :
3375 : /* Always invalidate so->killedItems[] before leaving so->currPos */
3376 171882 : so->numKilled = 0;
3377 :
3378 171882 : if (!so->dropPin)
3379 : {
3380 : /*
3381 : * We have held the pin on this page since we read the index tuples,
3382 : * so all we need to do is lock it. The pin will have prevented
3383 : * concurrent VACUUMs from recycling any of the TIDs on the page.
3384 : */
3385 : Assert(BTScanPosIsPinned(so->currPos));
3386 39064 : buf = so->currPos.buf;
3387 39064 : _bt_lockbuf(rel, buf, BT_READ);
3388 : }
3389 : else
3390 : {
3391 : XLogRecPtr latestlsn;
3392 :
3393 : Assert(!BTScanPosIsPinned(so->currPos));
3394 : Assert(RelationNeedsWAL(rel));
3395 132818 : buf = _bt_getbuf(rel, so->currPos.currPage, BT_READ);
3396 :
3397 132818 : latestlsn = BufferGetLSNAtomic(buf);
3398 : Assert(so->currPos.lsn <= latestlsn);
3399 132818 : if (so->currPos.lsn != latestlsn)
3400 : {
3401 : /* Modified, give up on hinting */
3402 114 : _bt_relbuf(rel, buf);
3403 114 : return;
3404 : }
3405 :
3406 : /* Unmodified, hinting is safe */
3407 : }
3408 :
3409 171768 : page = BufferGetPage(buf);
3410 171768 : opaque = BTPageGetOpaque(page);
3411 171768 : minoff = P_FIRSTDATAKEY(opaque);
3412 171768 : maxoff = PageGetMaxOffsetNumber(page);
3413 :
3414 592082 : for (int i = 0; i < numKilled; i++)
3415 : {
3416 420314 : int itemIndex = so->killedItems[i];
3417 420314 : BTScanPosItem *kitem = &so->currPos.items[itemIndex];
3418 420314 : OffsetNumber offnum = kitem->indexOffset;
3419 :
3420 : Assert(itemIndex >= so->currPos.firstItem &&
3421 : itemIndex <= so->currPos.lastItem);
3422 420314 : if (offnum < minoff)
3423 0 : continue; /* pure paranoia */
3424 9090686 : while (offnum <= maxoff)
3425 : {
3426 9022486 : ItemId iid = PageGetItemId(page, offnum);
3427 9022486 : IndexTuple ituple = (IndexTuple) PageGetItem(page, iid);
3428 9022486 : bool killtuple = false;
3429 :
3430 9022486 : if (BTreeTupleIsPosting(ituple))
3431 : {
3432 2888778 : int pi = i + 1;
3433 2888778 : int nposting = BTreeTupleGetNPosting(ituple);
3434 : int j;
3435 :
3436 : /*
3437 : * We rely on the convention that heap TIDs in the scanpos
3438 : * items array are stored in ascending heap TID order for a
3439 : * group of TIDs that originally came from a posting list
3440 : * tuple. This convention even applies during backwards
3441 : * scans, where returning the TIDs in descending order might
3442 : * seem more natural. This is about effectiveness, not
3443 : * correctness.
3444 : *
3445 : * Note that the page may have been modified in almost any way
3446 : * since we first read it (in the !so->dropPin case), so it's
3447 : * possible that this posting list tuple wasn't a posting list
3448 : * tuple when we first encountered its heap TIDs.
3449 : */
3450 2958044 : for (j = 0; j < nposting; j++)
3451 : {
3452 2955652 : ItemPointer item = BTreeTupleGetPostingN(ituple, j);
3453 :
3454 2955652 : if (!ItemPointerEquals(item, &kitem->heapTid))
3455 2886386 : break; /* out of posting list loop */
3456 :
3457 : /*
3458 : * kitem must have matching offnum when heap TIDs match,
3459 : * though only in the common case where the page can't
3460 : * have been concurrently modified
3461 : */
3462 : Assert(kitem->indexOffset == offnum || !so->dropPin);
3463 :
3464 : /*
3465 : * Read-ahead to later kitems here.
3466 : *
3467 : * We rely on the assumption that not advancing kitem here
3468 : * will prevent us from considering the posting list tuple
3469 : * fully dead by not matching its next heap TID in next
3470 : * loop iteration.
3471 : *
3472 : * If, on the other hand, this is the final heap TID in
3473 : * the posting list tuple, then tuple gets killed
3474 : * regardless (i.e. we handle the case where the last
3475 : * kitem is also the last heap TID in the last index tuple
3476 : * correctly -- posting tuple still gets killed).
3477 : */
3478 69266 : if (pi < numKilled)
3479 33906 : kitem = &so->currPos.items[so->killedItems[pi++]];
3480 : }
3481 :
3482 : /*
3483 : * Don't bother advancing the outermost loop's int iterator to
3484 : * avoid processing killed items that relate to the same
3485 : * offnum/posting list tuple. This micro-optimization hardly
3486 : * seems worth it. (Further iterations of the outermost loop
3487 : * will fail to match on this same posting list's first heap
3488 : * TID instead, so we'll advance to the next offnum/index
3489 : * tuple pretty quickly.)
3490 : */
3491 2888778 : if (j == nposting)
3492 2392 : killtuple = true;
3493 : }
3494 6133708 : else if (ItemPointerEquals(&ituple->t_tid, &kitem->heapTid))
3495 350366 : killtuple = true;
3496 :
3497 : /*
3498 : * Mark index item as dead, if it isn't already. Since this
3499 : * happens while holding a buffer lock possibly in shared mode,
3500 : * it's possible that multiple processes attempt to do this
3501 : * simultaneously, leading to multiple full-page images being sent
3502 : * to WAL (if wal_log_hints or data checksums are enabled), which
3503 : * is undesirable.
3504 : */
3505 9022486 : if (killtuple && !ItemIdIsDead(iid))
3506 : {
3507 : /* found the item/all posting list items */
3508 352114 : ItemIdMarkDead(iid);
3509 352114 : killedsomething = true;
3510 352114 : break; /* out of inner search loop */
3511 : }
3512 8670372 : offnum = OffsetNumberNext(offnum);
3513 : }
3514 : }
3515 :
3516 : /*
3517 : * Since this can be redone later if needed, mark as dirty hint.
3518 : *
3519 : * Whenever we mark anything LP_DEAD, we also set the page's
3520 : * BTP_HAS_GARBAGE flag, which is likewise just a hint. (Note that we
3521 : * only rely on the page-level flag in !heapkeyspace indexes.)
3522 : */
3523 171768 : if (killedsomething)
3524 : {
3525 134796 : opaque->btpo_flags |= BTP_HAS_GARBAGE;
3526 134796 : MarkBufferDirtyHint(buf, true);
3527 : }
3528 :
3529 171768 : if (!so->dropPin)
3530 39064 : _bt_unlockbuf(rel, buf);
3531 : else
3532 132704 : _bt_relbuf(rel, buf);
3533 : }
3534 :
3535 :
3536 : /*
3537 : * The following routines manage a shared-memory area in which we track
3538 : * assignment of "vacuum cycle IDs" to currently-active btree vacuuming
3539 : * operations. There is a single counter which increments each time we
3540 : * start a vacuum to assign it a cycle ID. Since multiple vacuums could
3541 : * be active concurrently, we have to track the cycle ID for each active
3542 : * vacuum; this requires at most MaxBackends entries (usually far fewer).
3543 : * We assume at most one vacuum can be active for a given index.
3544 : *
3545 : * Access to the shared memory area is controlled by BtreeVacuumLock.
3546 : * In principle we could use a separate lmgr locktag for each index,
3547 : * but a single LWLock is much cheaper, and given the short time that
3548 : * the lock is ever held, the concurrency hit should be minimal.
3549 : */
3550 :
3551 : typedef struct BTOneVacInfo
3552 : {
3553 : LockRelId relid; /* global identifier of an index */
3554 : BTCycleId cycleid; /* cycle ID for its active VACUUM */
3555 : } BTOneVacInfo;
3556 :
3557 : typedef struct BTVacInfo
3558 : {
3559 : BTCycleId cycle_ctr; /* cycle ID most recently assigned */
3560 : int num_vacuums; /* number of currently active VACUUMs */
3561 : int max_vacuums; /* allocated length of vacuums[] array */
3562 : BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER];
3563 : } BTVacInfo;
3564 :
3565 : static BTVacInfo *btvacinfo;
3566 :
3567 :
3568 : /*
3569 : * _bt_vacuum_cycleid --- get the active vacuum cycle ID for an index,
3570 : * or zero if there is no active VACUUM
3571 : *
3572 : * Note: for correct interlocking, the caller must already hold pin and
3573 : * exclusive lock on each buffer it will store the cycle ID into. This
3574 : * ensures that even if a VACUUM starts immediately afterwards, it cannot
3575 : * process those pages until the page split is complete.
3576 : */
3577 : BTCycleId
3578 22342 : _bt_vacuum_cycleid(Relation rel)
3579 : {
3580 22342 : BTCycleId result = 0;
3581 : int i;
3582 :
3583 : /* Share lock is enough since this is a read-only operation */
3584 22342 : LWLockAcquire(BtreeVacuumLock, LW_SHARED);
3585 :
3586 22388 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3587 : {
3588 48 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
3589 :
3590 48 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3591 2 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3592 : {
3593 2 : result = vac->cycleid;
3594 2 : break;
3595 : }
3596 : }
3597 :
3598 22342 : LWLockRelease(BtreeVacuumLock);
3599 22342 : return result;
3600 : }
3601 :
3602 : /*
3603 : * _bt_start_vacuum --- assign a cycle ID to a just-starting VACUUM operation
3604 : *
3605 : * Note: the caller must guarantee that it will eventually call
3606 : * _bt_end_vacuum, else we'll permanently leak an array slot. To ensure
3607 : * that this happens even in elog(FATAL) scenarios, the appropriate coding
3608 : * is not just a PG_TRY, but
3609 : * PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel))
3610 : */
3611 : BTCycleId
3612 2956 : _bt_start_vacuum(Relation rel)
3613 : {
3614 : BTCycleId result;
3615 : int i;
3616 : BTOneVacInfo *vac;
3617 :
3618 2956 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
3619 :
3620 : /*
3621 : * Assign the next cycle ID, being careful to avoid zero as well as the
3622 : * reserved high values.
3623 : */
3624 2956 : result = ++(btvacinfo->cycle_ctr);
3625 2956 : if (result == 0 || result > MAX_BT_CYCLE_ID)
3626 0 : result = btvacinfo->cycle_ctr = 1;
3627 :
3628 : /* Let's just make sure there's no entry already for this index */
3629 2956 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3630 : {
3631 0 : vac = &btvacinfo->vacuums[i];
3632 0 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3633 0 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3634 : {
3635 : /*
3636 : * Unlike most places in the backend, we have to explicitly
3637 : * release our LWLock before throwing an error. This is because
3638 : * we expect _bt_end_vacuum() to be called before transaction
3639 : * abort cleanup can run to release LWLocks.
3640 : */
3641 0 : LWLockRelease(BtreeVacuumLock);
3642 0 : elog(ERROR, "multiple active vacuums for index \"%s\"",
3643 : RelationGetRelationName(rel));
3644 : }
3645 : }
3646 :
3647 : /* OK, add an entry */
3648 2956 : if (btvacinfo->num_vacuums >= btvacinfo->max_vacuums)
3649 : {
3650 0 : LWLockRelease(BtreeVacuumLock);
3651 0 : elog(ERROR, "out of btvacinfo slots");
3652 : }
3653 2956 : vac = &btvacinfo->vacuums[btvacinfo->num_vacuums];
3654 2956 : vac->relid = rel->rd_lockInfo.lockRelId;
3655 2956 : vac->cycleid = result;
3656 2956 : btvacinfo->num_vacuums++;
3657 :
3658 2956 : LWLockRelease(BtreeVacuumLock);
3659 2956 : return result;
3660 : }
3661 :
3662 : /*
3663 : * _bt_end_vacuum --- mark a btree VACUUM operation as done
3664 : *
3665 : * Note: this is deliberately coded not to complain if no entry is found;
3666 : * this allows the caller to put PG_TRY around the start_vacuum operation.
3667 : */
3668 : void
3669 2956 : _bt_end_vacuum(Relation rel)
3670 : {
3671 : int i;
3672 :
3673 2956 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
3674 :
3675 : /* Find the array entry */
3676 2956 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3677 : {
3678 2956 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
3679 :
3680 2956 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3681 2956 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3682 : {
3683 : /* Remove it by shifting down the last entry */
3684 2956 : *vac = btvacinfo->vacuums[btvacinfo->num_vacuums - 1];
3685 2956 : btvacinfo->num_vacuums--;
3686 2956 : break;
3687 : }
3688 : }
3689 :
3690 2956 : LWLockRelease(BtreeVacuumLock);
3691 2956 : }
3692 :
3693 : /*
3694 : * _bt_end_vacuum wrapped as an on_shmem_exit callback function
3695 : */
3696 : void
3697 0 : _bt_end_vacuum_callback(int code, Datum arg)
3698 : {
3699 0 : _bt_end_vacuum((Relation) DatumGetPointer(arg));
3700 0 : }
3701 :
3702 : /*
3703 : * BTreeShmemSize --- report amount of shared memory space needed
3704 : */
3705 : Size
3706 6240 : BTreeShmemSize(void)
3707 : {
3708 : Size size;
3709 :
3710 6240 : size = offsetof(BTVacInfo, vacuums);
3711 6240 : size = add_size(size, mul_size(MaxBackends, sizeof(BTOneVacInfo)));
3712 6240 : return size;
3713 : }
3714 :
3715 : /*
3716 : * BTreeShmemInit --- initialize this module's shared memory
3717 : */
3718 : void
3719 2180 : BTreeShmemInit(void)
3720 : {
3721 : bool found;
3722 :
3723 2180 : btvacinfo = (BTVacInfo *) ShmemInitStruct("BTree Vacuum State",
3724 : BTreeShmemSize(),
3725 : &found);
3726 :
3727 2180 : if (!IsUnderPostmaster)
3728 : {
3729 : /* Initialize shared memory area */
3730 : Assert(!found);
3731 :
3732 : /*
3733 : * It doesn't really matter what the cycle counter starts at, but
3734 : * having it always start the same doesn't seem good. Seed with
3735 : * low-order bits of time() instead.
3736 : */
3737 2180 : btvacinfo->cycle_ctr = (BTCycleId) time(NULL);
3738 :
3739 2180 : btvacinfo->num_vacuums = 0;
3740 2180 : btvacinfo->max_vacuums = MaxBackends;
3741 : }
3742 : else
3743 : Assert(found);
3744 2180 : }
3745 :
3746 : bytea *
3747 322 : btoptions(Datum reloptions, bool validate)
3748 : {
3749 : static const relopt_parse_elt tab[] = {
3750 : {"fillfactor", RELOPT_TYPE_INT, offsetof(BTOptions, fillfactor)},
3751 : {"vacuum_cleanup_index_scale_factor", RELOPT_TYPE_REAL,
3752 : offsetof(BTOptions, vacuum_cleanup_index_scale_factor)},
3753 : {"deduplicate_items", RELOPT_TYPE_BOOL,
3754 : offsetof(BTOptions, deduplicate_items)}
3755 : };
3756 :
3757 322 : return (bytea *) build_reloptions(reloptions, validate,
3758 : RELOPT_KIND_BTREE,
3759 : sizeof(BTOptions),
3760 : tab, lengthof(tab));
3761 : }
3762 :
3763 : /*
3764 : * btproperty() -- Check boolean properties of indexes.
3765 : *
3766 : * This is optional, but handling AMPROP_RETURNABLE here saves opening the rel
3767 : * to call btcanreturn.
3768 : */
3769 : bool
3770 756 : btproperty(Oid index_oid, int attno,
3771 : IndexAMProperty prop, const char *propname,
3772 : bool *res, bool *isnull)
3773 : {
3774 756 : switch (prop)
3775 : {
3776 42 : case AMPROP_RETURNABLE:
3777 : /* answer only for columns, not AM or whole index */
3778 42 : if (attno == 0)
3779 12 : return false;
3780 : /* otherwise, btree can always return data */
3781 30 : *res = true;
3782 30 : return true;
3783 :
3784 714 : default:
3785 714 : return false; /* punt to generic code */
3786 : }
3787 : }
3788 :
3789 : /*
3790 : * btbuildphasename() -- Return name of index build phase.
3791 : */
3792 : char *
3793 0 : btbuildphasename(int64 phasenum)
3794 : {
3795 0 : switch (phasenum)
3796 : {
3797 0 : case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
3798 0 : return "initializing";
3799 0 : case PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN:
3800 0 : return "scanning table";
3801 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_1:
3802 0 : return "sorting live tuples";
3803 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_2:
3804 0 : return "sorting dead tuples";
3805 0 : case PROGRESS_BTREE_PHASE_LEAF_LOAD:
3806 0 : return "loading tuples in tree";
3807 0 : default:
3808 0 : return NULL;
3809 : }
3810 : }
3811 :
3812 : /*
3813 : * _bt_truncate() -- create tuple without unneeded suffix attributes.
3814 : *
3815 : * Returns truncated pivot index tuple allocated in caller's memory context,
3816 : * with key attributes copied from caller's firstright argument. If rel is
3817 : * an INCLUDE index, non-key attributes will definitely be truncated away,
3818 : * since they're not part of the key space. More aggressive suffix
3819 : * truncation can take place when it's clear that the returned tuple does not
3820 : * need one or more suffix key attributes. We only need to keep firstright
3821 : * attributes up to and including the first non-lastleft-equal attribute.
3822 : * Caller's insertion scankey is used to compare the tuples; the scankey's
3823 : * argument values are not considered here.
3824 : *
3825 : * Note that returned tuple's t_tid offset will hold the number of attributes
3826 : * present, so the original item pointer offset is not represented. Caller
3827 : * should only change truncated tuple's downlink. Note also that truncated
3828 : * key attributes are treated as containing "minus infinity" values by
3829 : * _bt_compare().
3830 : *
3831 : * In the worst case (when a heap TID must be appended to distinguish lastleft
3832 : * from firstright), the size of the returned tuple is the size of firstright
3833 : * plus the size of an additional MAXALIGN()'d item pointer. This guarantee
3834 : * is important, since callers need to stay under the 1/3 of a page
3835 : * restriction on tuple size. If this routine is ever taught to truncate
3836 : * within an attribute/datum, it will need to avoid returning an enlarged
3837 : * tuple to caller when truncation + TOAST compression ends up enlarging the
3838 : * final datum.
3839 : */
3840 : IndexTuple
3841 62122 : _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright,
3842 : BTScanInsert itup_key)
3843 : {
3844 62122 : TupleDesc itupdesc = RelationGetDescr(rel);
3845 62122 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
3846 : int keepnatts;
3847 : IndexTuple pivot;
3848 : IndexTuple tidpivot;
3849 : ItemPointer pivotheaptid;
3850 : Size newsize;
3851 :
3852 : /*
3853 : * We should only ever truncate non-pivot tuples from leaf pages. It's
3854 : * never okay to truncate when splitting an internal page.
3855 : */
3856 : Assert(!BTreeTupleIsPivot(lastleft) && !BTreeTupleIsPivot(firstright));
3857 :
3858 : /* Determine how many attributes must be kept in truncated tuple */
3859 62122 : keepnatts = _bt_keep_natts(rel, lastleft, firstright, itup_key);
3860 :
3861 : #ifdef DEBUG_NO_TRUNCATE
3862 : /* Force truncation to be ineffective for testing purposes */
3863 : keepnatts = nkeyatts + 1;
3864 : #endif
3865 :
3866 62122 : pivot = index_truncate_tuple(itupdesc, firstright,
3867 : Min(keepnatts, nkeyatts));
3868 :
3869 62122 : if (BTreeTupleIsPosting(pivot))
3870 : {
3871 : /*
3872 : * index_truncate_tuple() just returns a straight copy of firstright
3873 : * when it has no attributes to truncate. When that happens, we may
3874 : * need to truncate away a posting list here instead.
3875 : */
3876 : Assert(keepnatts == nkeyatts || keepnatts == nkeyatts + 1);
3877 : Assert(IndexRelationGetNumberOfAttributes(rel) == nkeyatts);
3878 1242 : pivot->t_info &= ~INDEX_SIZE_MASK;
3879 1242 : pivot->t_info |= MAXALIGN(BTreeTupleGetPostingOffset(firstright));
3880 : }
3881 :
3882 : /*
3883 : * If there is a distinguishing key attribute within pivot tuple, we're
3884 : * done
3885 : */
3886 62122 : if (keepnatts <= nkeyatts)
3887 : {
3888 61038 : BTreeTupleSetNAtts(pivot, keepnatts, false);
3889 61038 : return pivot;
3890 : }
3891 :
3892 : /*
3893 : * We have to store a heap TID in the new pivot tuple, since no non-TID
3894 : * key attribute value in firstright distinguishes the right side of the
3895 : * split from the left side. nbtree conceptualizes this case as an
3896 : * inability to truncate away any key attributes, since heap TID is
3897 : * treated as just another key attribute (despite lacking a pg_attribute
3898 : * entry).
3899 : *
3900 : * Use enlarged space that holds a copy of pivot. We need the extra space
3901 : * to store a heap TID at the end (using the special pivot tuple
3902 : * representation). Note that the original pivot already has firstright's
3903 : * possible posting list/non-key attribute values removed at this point.
3904 : */
3905 1084 : newsize = MAXALIGN(IndexTupleSize(pivot)) + MAXALIGN(sizeof(ItemPointerData));
3906 1084 : tidpivot = palloc0(newsize);
3907 1084 : memcpy(tidpivot, pivot, MAXALIGN(IndexTupleSize(pivot)));
3908 : /* Cannot leak memory here */
3909 1084 : pfree(pivot);
3910 :
3911 : /*
3912 : * Store all of firstright's key attribute values plus a tiebreaker heap
3913 : * TID value in enlarged pivot tuple
3914 : */
3915 1084 : tidpivot->t_info &= ~INDEX_SIZE_MASK;
3916 1084 : tidpivot->t_info |= newsize;
3917 1084 : BTreeTupleSetNAtts(tidpivot, nkeyatts, true);
3918 1084 : pivotheaptid = BTreeTupleGetHeapTID(tidpivot);
3919 :
3920 : /*
3921 : * Lehman & Yao use lastleft as the leaf high key in all cases, but don't
3922 : * consider suffix truncation. It seems like a good idea to follow that
3923 : * example in cases where no truncation takes place -- use lastleft's heap
3924 : * TID. (This is also the closest value to negative infinity that's
3925 : * legally usable.)
3926 : */
3927 1084 : ItemPointerCopy(BTreeTupleGetMaxHeapTID(lastleft), pivotheaptid);
3928 :
3929 : /*
3930 : * We're done. Assert() that heap TID invariants hold before returning.
3931 : *
3932 : * Lehman and Yao require that the downlink to the right page, which is to
3933 : * be inserted into the parent page in the second phase of a page split be
3934 : * a strict lower bound on items on the right page, and a non-strict upper
3935 : * bound for items on the left page. Assert that heap TIDs follow these
3936 : * invariants, since a heap TID value is apparently needed as a
3937 : * tiebreaker.
3938 : */
3939 : #ifndef DEBUG_NO_TRUNCATE
3940 : Assert(ItemPointerCompare(BTreeTupleGetMaxHeapTID(lastleft),
3941 : BTreeTupleGetHeapTID(firstright)) < 0);
3942 : Assert(ItemPointerCompare(pivotheaptid,
3943 : BTreeTupleGetHeapTID(lastleft)) >= 0);
3944 : Assert(ItemPointerCompare(pivotheaptid,
3945 : BTreeTupleGetHeapTID(firstright)) < 0);
3946 : #else
3947 :
3948 : /*
3949 : * Those invariants aren't guaranteed to hold for lastleft + firstright
3950 : * heap TID attribute values when they're considered here only because
3951 : * DEBUG_NO_TRUNCATE is defined (a heap TID is probably not actually
3952 : * needed as a tiebreaker). DEBUG_NO_TRUNCATE must therefore use a heap
3953 : * TID value that always works as a strict lower bound for items to the
3954 : * right. In particular, it must avoid using firstright's leading key
3955 : * attribute values along with lastleft's heap TID value when lastleft's
3956 : * TID happens to be greater than firstright's TID.
3957 : */
3958 : ItemPointerCopy(BTreeTupleGetHeapTID(firstright), pivotheaptid);
3959 :
3960 : /*
3961 : * Pivot heap TID should never be fully equal to firstright. Note that
3962 : * the pivot heap TID will still end up equal to lastleft's heap TID when
3963 : * that's the only usable value.
3964 : */
3965 : ItemPointerSetOffsetNumber(pivotheaptid,
3966 : OffsetNumberPrev(ItemPointerGetOffsetNumber(pivotheaptid)));
3967 : Assert(ItemPointerCompare(pivotheaptid,
3968 : BTreeTupleGetHeapTID(firstright)) < 0);
3969 : #endif
3970 :
3971 1084 : return tidpivot;
3972 : }
3973 :
3974 : /*
3975 : * _bt_keep_natts - how many key attributes to keep when truncating.
3976 : *
3977 : * Caller provides two tuples that enclose a split point. Caller's insertion
3978 : * scankey is used to compare the tuples; the scankey's argument values are
3979 : * not considered here.
3980 : *
3981 : * This can return a number of attributes that is one greater than the
3982 : * number of key attributes for the index relation. This indicates that the
3983 : * caller must use a heap TID as a unique-ifier in new pivot tuple.
3984 : */
3985 : static int
3986 62122 : _bt_keep_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright,
3987 : BTScanInsert itup_key)
3988 : {
3989 62122 : int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
3990 62122 : TupleDesc itupdesc = RelationGetDescr(rel);
3991 : int keepnatts;
3992 : ScanKey scankey;
3993 :
3994 : /*
3995 : * _bt_compare() treats truncated key attributes as having the value minus
3996 : * infinity, which would break searches within !heapkeyspace indexes. We
3997 : * must still truncate away non-key attribute values, though.
3998 : */
3999 62122 : if (!itup_key->heapkeyspace)
4000 0 : return nkeyatts;
4001 :
4002 62122 : scankey = itup_key->scankeys;
4003 62122 : keepnatts = 1;
4004 74732 : for (int attnum = 1; attnum <= nkeyatts; attnum++, scankey++)
4005 : {
4006 : Datum datum1,
4007 : datum2;
4008 : bool isNull1,
4009 : isNull2;
4010 :
4011 73648 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
4012 73648 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
4013 :
4014 73648 : if (isNull1 != isNull2)
4015 61038 : break;
4016 :
4017 147266 : if (!isNull1 &&
4018 73618 : DatumGetInt32(FunctionCall2Coll(&scankey->sk_func,
4019 : scankey->sk_collation,
4020 : datum1,
4021 : datum2)) != 0)
4022 61038 : break;
4023 :
4024 12610 : keepnatts++;
4025 : }
4026 :
4027 : /*
4028 : * Assert that _bt_keep_natts_fast() agrees with us in passing. This is
4029 : * expected in an allequalimage index.
4030 : */
4031 : Assert(!itup_key->allequalimage ||
4032 : keepnatts == _bt_keep_natts_fast(rel, lastleft, firstright));
4033 :
4034 62122 : return keepnatts;
4035 : }
4036 :
4037 : /*
4038 : * _bt_keep_natts_fast - fast bitwise variant of _bt_keep_natts.
4039 : *
4040 : * This is exported so that a candidate split point can have its effect on
4041 : * suffix truncation inexpensively evaluated ahead of time when finding a
4042 : * split location. A naive bitwise approach to datum comparisons is used to
4043 : * save cycles.
4044 : *
4045 : * The approach taken here usually provides the same answer as _bt_keep_natts
4046 : * will (for the same pair of tuples from a heapkeyspace index), since the
4047 : * majority of btree opclasses can never indicate that two datums are equal
4048 : * unless they're bitwise equal after detoasting. When an index only has
4049 : * "equal image" columns, routine is guaranteed to give the same result as
4050 : * _bt_keep_natts would.
4051 : *
4052 : * Callers can rely on the fact that attributes considered equal here are
4053 : * definitely also equal according to _bt_keep_natts, even when the index uses
4054 : * an opclass or collation that is not "allequalimage"/deduplication-safe.
4055 : * This weaker guarantee is good enough for nbtsplitloc.c caller, since false
4056 : * negatives generally only have the effect of making leaf page splits use a
4057 : * more balanced split point.
4058 : */
4059 : int
4060 13457560 : _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
4061 : {
4062 13457560 : TupleDesc itupdesc = RelationGetDescr(rel);
4063 13457560 : int keysz = IndexRelationGetNumberOfKeyAttributes(rel);
4064 : int keepnatts;
4065 :
4066 13457560 : keepnatts = 1;
4067 22577500 : for (int attnum = 1; attnum <= keysz; attnum++)
4068 : {
4069 : Datum datum1,
4070 : datum2;
4071 : bool isNull1,
4072 : isNull2;
4073 : CompactAttribute *att;
4074 :
4075 20202312 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
4076 20202312 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
4077 20202312 : att = TupleDescCompactAttr(itupdesc, attnum - 1);
4078 :
4079 20202312 : if (isNull1 != isNull2)
4080 11082372 : break;
4081 :
4082 20202108 : if (!isNull1 &&
4083 20155052 : !datum_image_eq(datum1, datum2, att->attbyval, att->attlen))
4084 11082168 : break;
4085 :
4086 9119940 : keepnatts++;
4087 : }
4088 :
4089 13457560 : return keepnatts;
4090 : }
4091 :
4092 : /*
4093 : * _bt_check_natts() -- Verify tuple has expected number of attributes.
4094 : *
4095 : * Returns value indicating if the expected number of attributes were found
4096 : * for a particular offset on page. This can be used as a general purpose
4097 : * sanity check.
4098 : *
4099 : * Testing a tuple directly with BTreeTupleGetNAtts() should generally be
4100 : * preferred to calling here. That's usually more convenient, and is always
4101 : * more explicit. Call here instead when offnum's tuple may be a negative
4102 : * infinity tuple that uses the pre-v11 on-disk representation, or when a low
4103 : * context check is appropriate. This routine is as strict as possible about
4104 : * what is expected on each version of btree.
4105 : */
4106 : bool
4107 4061496 : _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
4108 : {
4109 4061496 : int16 natts = IndexRelationGetNumberOfAttributes(rel);
4110 4061496 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
4111 4061496 : BTPageOpaque opaque = BTPageGetOpaque(page);
4112 : IndexTuple itup;
4113 : int tupnatts;
4114 :
4115 : /*
4116 : * We cannot reliably test a deleted or half-dead page, since they have
4117 : * dummy high keys
4118 : */
4119 4061496 : if (P_IGNORE(opaque))
4120 0 : return true;
4121 :
4122 : Assert(offnum >= FirstOffsetNumber &&
4123 : offnum <= PageGetMaxOffsetNumber(page));
4124 :
4125 4061496 : itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
4126 4061496 : tupnatts = BTreeTupleGetNAtts(itup, rel);
4127 :
4128 : /* !heapkeyspace indexes do not support deduplication */
4129 4061496 : if (!heapkeyspace && BTreeTupleIsPosting(itup))
4130 0 : return false;
4131 :
4132 : /* Posting list tuples should never have "pivot heap TID" bit set */
4133 4061496 : if (BTreeTupleIsPosting(itup) &&
4134 22114 : (ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) &
4135 : BT_PIVOT_HEAP_TID_ATTR) != 0)
4136 0 : return false;
4137 :
4138 : /* INCLUDE indexes do not support deduplication */
4139 4061496 : if (natts != nkeyatts && BTreeTupleIsPosting(itup))
4140 0 : return false;
4141 :
4142 4061496 : if (P_ISLEAF(opaque))
4143 : {
4144 4047118 : if (offnum >= P_FIRSTDATAKEY(opaque))
4145 : {
4146 : /*
4147 : * Non-pivot tuple should never be explicitly marked as a pivot
4148 : * tuple
4149 : */
4150 4033860 : if (BTreeTupleIsPivot(itup))
4151 0 : return false;
4152 :
4153 : /*
4154 : * Leaf tuples that are not the page high key (non-pivot tuples)
4155 : * should never be truncated. (Note that tupnatts must have been
4156 : * inferred, even with a posting list tuple, because only pivot
4157 : * tuples store tupnatts directly.)
4158 : */
4159 4033860 : return tupnatts == natts;
4160 : }
4161 : else
4162 : {
4163 : /*
4164 : * Rightmost page doesn't contain a page high key, so tuple was
4165 : * checked above as ordinary leaf tuple
4166 : */
4167 : Assert(!P_RIGHTMOST(opaque));
4168 :
4169 : /*
4170 : * !heapkeyspace high key tuple contains only key attributes. Note
4171 : * that tupnatts will only have been explicitly represented in
4172 : * !heapkeyspace indexes that happen to have non-key attributes.
4173 : */
4174 13258 : if (!heapkeyspace)
4175 0 : return tupnatts == nkeyatts;
4176 :
4177 : /* Use generic heapkeyspace pivot tuple handling */
4178 : }
4179 : }
4180 : else /* !P_ISLEAF(opaque) */
4181 : {
4182 14378 : if (offnum == P_FIRSTDATAKEY(opaque))
4183 : {
4184 : /*
4185 : * The first tuple on any internal page (possibly the first after
4186 : * its high key) is its negative infinity tuple. Negative
4187 : * infinity tuples are always truncated to zero attributes. They
4188 : * are a particular kind of pivot tuple.
4189 : */
4190 1114 : if (heapkeyspace)
4191 1114 : return tupnatts == 0;
4192 :
4193 : /*
4194 : * The number of attributes won't be explicitly represented if the
4195 : * negative infinity tuple was generated during a page split that
4196 : * occurred with a version of Postgres before v11. There must be
4197 : * a problem when there is an explicit representation that is
4198 : * non-zero, or when there is no explicit representation and the
4199 : * tuple is evidently not a pre-pg_upgrade tuple.
4200 : *
4201 : * Prior to v11, downlinks always had P_HIKEY as their offset.
4202 : * Accept that as an alternative indication of a valid
4203 : * !heapkeyspace negative infinity tuple.
4204 : */
4205 0 : return tupnatts == 0 ||
4206 0 : ItemPointerGetOffsetNumber(&(itup->t_tid)) == P_HIKEY;
4207 : }
4208 : else
4209 : {
4210 : /*
4211 : * !heapkeyspace downlink tuple with separator key contains only
4212 : * key attributes. Note that tupnatts will only have been
4213 : * explicitly represented in !heapkeyspace indexes that happen to
4214 : * have non-key attributes.
4215 : */
4216 13264 : if (!heapkeyspace)
4217 0 : return tupnatts == nkeyatts;
4218 :
4219 : /* Use generic heapkeyspace pivot tuple handling */
4220 : }
4221 : }
4222 :
4223 : /* Handle heapkeyspace pivot tuples (excluding minus infinity items) */
4224 : Assert(heapkeyspace);
4225 :
4226 : /*
4227 : * Explicit representation of the number of attributes is mandatory with
4228 : * heapkeyspace index pivot tuples, regardless of whether or not there are
4229 : * non-key attributes.
4230 : */
4231 26522 : if (!BTreeTupleIsPivot(itup))
4232 0 : return false;
4233 :
4234 : /* Pivot tuple should not use posting list representation (redundant) */
4235 26522 : if (BTreeTupleIsPosting(itup))
4236 0 : return false;
4237 :
4238 : /*
4239 : * Heap TID is a tiebreaker key attribute, so it cannot be untruncated
4240 : * when any other key attribute is truncated
4241 : */
4242 26522 : if (BTreeTupleGetHeapTID(itup) != NULL && tupnatts != nkeyatts)
4243 0 : return false;
4244 :
4245 : /*
4246 : * Pivot tuple must have at least one untruncated key attribute (minus
4247 : * infinity pivot tuples are the only exception). Pivot tuples can never
4248 : * represent that there is a value present for a key attribute that
4249 : * exceeds pg_index.indnkeyatts for the index.
4250 : */
4251 26522 : return tupnatts > 0 && tupnatts <= nkeyatts;
4252 : }
4253 :
4254 : /*
4255 : *
4256 : * _bt_check_third_page() -- check whether tuple fits on a btree page at all.
4257 : *
4258 : * We actually need to be able to fit three items on every page, so restrict
4259 : * any one item to 1/3 the per-page available space. Note that itemsz should
4260 : * not include the ItemId overhead.
4261 : *
4262 : * It might be useful to apply TOAST methods rather than throw an error here.
4263 : * Using out of line storage would break assumptions made by suffix truncation
4264 : * and by contrib/amcheck, though.
4265 : */
4266 : void
4267 264 : _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace,
4268 : Page page, IndexTuple newtup)
4269 : {
4270 : Size itemsz;
4271 : BTPageOpaque opaque;
4272 :
4273 264 : itemsz = MAXALIGN(IndexTupleSize(newtup));
4274 :
4275 : /* Double check item size against limit */
4276 264 : if (itemsz <= BTMaxItemSize)
4277 0 : return;
4278 :
4279 : /*
4280 : * Tuple is probably too large to fit on page, but it's possible that the
4281 : * index uses version 2 or version 3, or that page is an internal page, in
4282 : * which case a slightly higher limit applies.
4283 : */
4284 264 : if (!needheaptidspace && itemsz <= BTMaxItemSizeNoHeapTid)
4285 264 : return;
4286 :
4287 : /*
4288 : * Internal page insertions cannot fail here, because that would mean that
4289 : * an earlier leaf level insertion that should have failed didn't
4290 : */
4291 0 : opaque = BTPageGetOpaque(page);
4292 0 : if (!P_ISLEAF(opaque))
4293 0 : elog(ERROR, "cannot insert oversized tuple of size %zu on internal page of index \"%s\"",
4294 : itemsz, RelationGetRelationName(rel));
4295 :
4296 0 : ereport(ERROR,
4297 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
4298 : errmsg("index row size %zu exceeds btree version %u maximum %zu for index \"%s\"",
4299 : itemsz,
4300 : needheaptidspace ? BTREE_VERSION : BTREE_NOVAC_VERSION,
4301 : needheaptidspace ? BTMaxItemSize : BTMaxItemSizeNoHeapTid,
4302 : RelationGetRelationName(rel)),
4303 : errdetail("Index row references tuple (%u,%u) in relation \"%s\".",
4304 : ItemPointerGetBlockNumber(BTreeTupleGetHeapTID(newtup)),
4305 : ItemPointerGetOffsetNumber(BTreeTupleGetHeapTID(newtup)),
4306 : RelationGetRelationName(heap)),
4307 : errhint("Values larger than 1/3 of a buffer page cannot be indexed.\n"
4308 : "Consider a function index of an MD5 hash of the value, "
4309 : "or use full text indexing."),
4310 : errtableconstraint(heap, RelationGetRelationName(rel))));
4311 : }
4312 :
4313 : /*
4314 : * Are all attributes in rel "equality is image equality" attributes?
4315 : *
4316 : * We use each attribute's BTEQUALIMAGE_PROC opclass procedure. If any
4317 : * opclass either lacks a BTEQUALIMAGE_PROC procedure or returns false, we
4318 : * return false; otherwise we return true.
4319 : *
4320 : * Returned boolean value is stored in index metapage during index builds.
4321 : * Deduplication can only be used when we return true.
4322 : */
4323 : bool
4324 58528 : _bt_allequalimage(Relation rel, bool debugmessage)
4325 : {
4326 58528 : bool allequalimage = true;
4327 :
4328 : /* INCLUDE indexes can never support deduplication */
4329 58528 : if (IndexRelationGetNumberOfAttributes(rel) !=
4330 58528 : IndexRelationGetNumberOfKeyAttributes(rel))
4331 272 : return false;
4332 :
4333 153716 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(rel); i++)
4334 : {
4335 95978 : Oid opfamily = rel->rd_opfamily[i];
4336 95978 : Oid opcintype = rel->rd_opcintype[i];
4337 95978 : Oid collation = rel->rd_indcollation[i];
4338 : Oid equalimageproc;
4339 :
4340 95978 : equalimageproc = get_opfamily_proc(opfamily, opcintype, opcintype,
4341 : BTEQUALIMAGE_PROC);
4342 :
4343 : /*
4344 : * If there is no BTEQUALIMAGE_PROC then deduplication is assumed to
4345 : * be unsafe. Otherwise, actually call proc and see what it says.
4346 : */
4347 95978 : if (!OidIsValid(equalimageproc) ||
4348 95504 : !DatumGetBool(OidFunctionCall1Coll(equalimageproc, collation,
4349 : ObjectIdGetDatum(opcintype))))
4350 : {
4351 518 : allequalimage = false;
4352 518 : break;
4353 : }
4354 : }
4355 :
4356 58256 : if (debugmessage)
4357 : {
4358 50244 : if (allequalimage)
4359 49726 : elog(DEBUG1, "index \"%s\" can safely use deduplication",
4360 : RelationGetRelationName(rel));
4361 : else
4362 518 : elog(DEBUG1, "index \"%s\" cannot use deduplication",
4363 : RelationGetRelationName(rel));
4364 : }
4365 :
4366 58256 : return allequalimage;
4367 : }
|
