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_check_rowcompare(ScanKey skey,
66 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
67 : ScanDirection dir, bool forcenonrequired, bool *continuescan);
68 : static void _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
69 : int tupnatts, TupleDesc tupdesc);
70 : static int _bt_keep_natts(Relation rel, IndexTuple lastleft,
71 : IndexTuple firstright, BTScanInsert itup_key);
72 :
73 :
74 : /*
75 : * _bt_mkscankey
76 : * Build an insertion scan key that contains comparison data from itup
77 : * as well as comparator routines appropriate to the key datatypes.
78 : *
79 : * The result is intended for use with _bt_compare() and _bt_truncate().
80 : * Callers that don't need to fill out the insertion scankey arguments
81 : * (e.g. they use an ad-hoc comparison routine, or only need a scankey
82 : * for _bt_truncate()) can pass a NULL index tuple. The scankey will
83 : * be initialized as if an "all truncated" pivot tuple was passed
84 : * instead.
85 : *
86 : * Note that we may occasionally have to share lock the metapage to
87 : * determine whether or not the keys in the index are expected to be
88 : * unique (i.e. if this is a "heapkeyspace" index). We assume a
89 : * heapkeyspace index when caller passes a NULL tuple, allowing index
90 : * build callers to avoid accessing the non-existent metapage. We
91 : * also assume that the index is _not_ allequalimage when a NULL tuple
92 : * is passed; CREATE INDEX callers call _bt_allequalimage() to set the
93 : * field themselves.
94 : */
95 : BTScanInsert
96 11803526 : _bt_mkscankey(Relation rel, IndexTuple itup)
97 : {
98 : BTScanInsert key;
99 : ScanKey skey;
100 : TupleDesc itupdesc;
101 : int indnkeyatts;
102 : int16 *indoption;
103 : int tupnatts;
104 : int i;
105 :
106 11803526 : itupdesc = RelationGetDescr(rel);
107 11803526 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
108 11803526 : indoption = rel->rd_indoption;
109 11803526 : tupnatts = itup ? BTreeTupleGetNAtts(itup, rel) : 0;
110 :
111 : Assert(tupnatts <= IndexRelationGetNumberOfAttributes(rel));
112 :
113 : /*
114 : * We'll execute search using scan key constructed on key columns.
115 : * Truncated attributes and non-key attributes are omitted from the final
116 : * scan key.
117 : */
118 11803526 : key = palloc(offsetof(BTScanInsertData, scankeys) +
119 11803526 : sizeof(ScanKeyData) * indnkeyatts);
120 11803526 : if (itup)
121 11661386 : _bt_metaversion(rel, &key->heapkeyspace, &key->allequalimage);
122 : else
123 : {
124 : /* Utility statement callers can set these fields themselves */
125 142140 : key->heapkeyspace = true;
126 142140 : key->allequalimage = false;
127 : }
128 11803526 : key->anynullkeys = false; /* initial assumption */
129 11803526 : key->nextkey = false; /* usual case, required by btinsert */
130 11803526 : key->backward = false; /* usual case, required by btinsert */
131 11803526 : key->keysz = Min(indnkeyatts, tupnatts);
132 11803526 : key->scantid = key->heapkeyspace && itup ?
133 23607052 : BTreeTupleGetHeapTID(itup) : NULL;
134 11803526 : skey = key->scankeys;
135 31933760 : for (i = 0; i < indnkeyatts; i++)
136 : {
137 : FmgrInfo *procinfo;
138 : Datum arg;
139 : bool null;
140 : int flags;
141 :
142 : /*
143 : * We can use the cached (default) support procs since no cross-type
144 : * comparison can be needed.
145 : */
146 20130234 : procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
147 :
148 : /*
149 : * Key arguments built from truncated attributes (or when caller
150 : * provides no tuple) are defensively represented as NULL values. They
151 : * should never be used.
152 : */
153 20130234 : if (i < tupnatts)
154 19875248 : arg = index_getattr(itup, i + 1, itupdesc, &null);
155 : else
156 : {
157 254986 : arg = (Datum) 0;
158 254986 : null = true;
159 : }
160 20130234 : flags = (null ? SK_ISNULL : 0) | (indoption[i] << SK_BT_INDOPTION_SHIFT);
161 20130234 : ScanKeyEntryInitializeWithInfo(&skey[i],
162 : flags,
163 20130234 : (AttrNumber) (i + 1),
164 : InvalidStrategy,
165 : InvalidOid,
166 20130234 : rel->rd_indcollation[i],
167 : procinfo,
168 : arg);
169 : /* Record if any key attribute is NULL (or truncated) */
170 20130234 : if (null)
171 275634 : key->anynullkeys = true;
172 : }
173 :
174 : /*
175 : * In NULLS NOT DISTINCT mode, we pretend that there are no null keys, so
176 : * that full uniqueness check is done.
177 : */
178 11803526 : if (rel->rd_index->indnullsnotdistinct)
179 186 : key->anynullkeys = false;
180 :
181 11803526 : return key;
182 : }
183 :
184 : /*
185 : * free a retracement stack made by _bt_search.
186 : */
187 : void
188 22839208 : _bt_freestack(BTStack stack)
189 : {
190 : BTStack ostack;
191 :
192 41853040 : while (stack != NULL)
193 : {
194 19013832 : ostack = stack;
195 19013832 : stack = stack->bts_parent;
196 19013832 : pfree(ostack);
197 : }
198 22839208 : }
199 :
200 : /*
201 : * _bt_compare_array_skey() -- apply array comparison function
202 : *
203 : * Compares caller's tuple attribute value to a scan key/array element.
204 : * Helper function used during binary searches of SK_SEARCHARRAY arrays.
205 : *
206 : * This routine returns:
207 : * <0 if tupdatum < arrdatum;
208 : * 0 if tupdatum == arrdatum;
209 : * >0 if tupdatum > arrdatum.
210 : *
211 : * This is essentially the same interface as _bt_compare: both functions
212 : * compare the value that they're searching for to a binary search pivot.
213 : * However, unlike _bt_compare, this function's "tuple argument" comes first,
214 : * while its "array/scankey argument" comes second.
215 : */
216 : static inline int32
217 463468 : _bt_compare_array_skey(FmgrInfo *orderproc,
218 : Datum tupdatum, bool tupnull,
219 : Datum arrdatum, ScanKey cur)
220 : {
221 463468 : int32 result = 0;
222 :
223 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
224 : Assert(!(cur->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL)));
225 :
226 463468 : if (tupnull) /* NULL tupdatum */
227 : {
228 228 : if (cur->sk_flags & SK_ISNULL)
229 132 : result = 0; /* NULL "=" NULL */
230 96 : else if (cur->sk_flags & SK_BT_NULLS_FIRST)
231 0 : result = -1; /* NULL "<" NOT_NULL */
232 : else
233 96 : result = 1; /* NULL ">" NOT_NULL */
234 : }
235 463240 : else if (cur->sk_flags & SK_ISNULL) /* NOT_NULL tupdatum, NULL arrdatum */
236 : {
237 30540 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
238 54 : result = 1; /* NOT_NULL ">" NULL */
239 : else
240 30486 : result = -1; /* NOT_NULL "<" NULL */
241 : }
242 : else
243 : {
244 : /*
245 : * Like _bt_compare, we need to be careful of cross-type comparisons,
246 : * so the left value has to be the value that came from an index tuple
247 : */
248 432700 : result = DatumGetInt32(FunctionCall2Coll(orderproc, cur->sk_collation,
249 : tupdatum, arrdatum));
250 :
251 : /*
252 : * We flip the sign by following the obvious rule: flip whenever the
253 : * column is a DESC column.
254 : *
255 : * _bt_compare does it the wrong way around (flip when *ASC*) in order
256 : * to compensate for passing its orderproc arguments backwards. We
257 : * don't need to play these games because we find it natural to pass
258 : * tupdatum as the left value (and arrdatum as the right value).
259 : */
260 432700 : if (cur->sk_flags & SK_BT_DESC)
261 45498 : INVERT_COMPARE_RESULT(result);
262 : }
263 :
264 463468 : return result;
265 : }
266 :
267 : /*
268 : * _bt_binsrch_array_skey() -- Binary search for next matching array key
269 : *
270 : * Returns an index to the first array element >= caller's tupdatum argument.
271 : * This convention is more natural for forwards scan callers, but that can't
272 : * really matter to backwards scan callers. Both callers require handling for
273 : * the case where the match we return is < tupdatum, and symmetric handling
274 : * for the case where our best match is > tupdatum.
275 : *
276 : * Also sets *set_elem_result to the result _bt_compare_array_skey returned
277 : * when we used it to compare the matching array element to tupdatum/tupnull.
278 : *
279 : * cur_elem_trig indicates if array advancement was triggered by this array's
280 : * scan key, and that the array is for a required scan key. We can apply this
281 : * information to find the next matching array element in the current scan
282 : * direction using far fewer comparisons (fewer on average, compared to naive
283 : * binary search). This scheme takes advantage of an important property of
284 : * required arrays: required arrays always advance in lockstep with the index
285 : * scan's progress through the index's key space.
286 : */
287 : int
288 31190 : _bt_binsrch_array_skey(FmgrInfo *orderproc,
289 : bool cur_elem_trig, ScanDirection dir,
290 : Datum tupdatum, bool tupnull,
291 : BTArrayKeyInfo *array, ScanKey cur,
292 : int32 *set_elem_result)
293 : {
294 31190 : int low_elem = 0,
295 31190 : mid_elem = -1,
296 31190 : high_elem = array->num_elems - 1,
297 31190 : result = 0;
298 : Datum arrdatum;
299 :
300 : Assert(cur->sk_flags & SK_SEARCHARRAY);
301 : Assert(!(cur->sk_flags & SK_BT_SKIP));
302 : Assert(!(cur->sk_flags & SK_ISNULL)); /* SAOP arrays never have NULLs */
303 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
304 :
305 31190 : if (cur_elem_trig)
306 : {
307 : Assert(!ScanDirectionIsNoMovement(dir));
308 : Assert(cur->sk_flags & SK_BT_REQFWD);
309 :
310 : /*
311 : * When the scan key that triggered array advancement is a required
312 : * array scan key, it is now certain that the current array element
313 : * (plus all prior elements relative to the current scan direction)
314 : * cannot possibly be at or ahead of the corresponding tuple value.
315 : * (_bt_checkkeys must have called _bt_tuple_before_array_skeys, which
316 : * makes sure this is true as a condition of advancing the arrays.)
317 : *
318 : * This makes it safe to exclude array elements up to and including
319 : * the former-current array element from our search.
320 : *
321 : * Separately, when array advancement was triggered by a required scan
322 : * key, the array element immediately after the former-current element
323 : * is often either an exact tupdatum match, or a "close by" near-match
324 : * (a near-match tupdatum is one whose key space falls _between_ the
325 : * former-current and new-current array elements). We'll detect both
326 : * cases via an optimistic comparison of the new search lower bound
327 : * (or new search upper bound in the case of backwards scans).
328 : */
329 30872 : if (ScanDirectionIsForward(dir))
330 : {
331 30812 : low_elem = array->cur_elem + 1; /* old cur_elem exhausted */
332 :
333 : /* Compare prospective new cur_elem (also the new lower bound) */
334 30812 : if (high_elem >= low_elem)
335 : {
336 23002 : arrdatum = array->elem_values[low_elem];
337 23002 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
338 : arrdatum, cur);
339 :
340 23002 : if (result <= 0)
341 : {
342 : /* Optimistic comparison optimization worked out */
343 22916 : *set_elem_result = result;
344 22916 : return low_elem;
345 : }
346 86 : mid_elem = low_elem;
347 86 : low_elem++; /* this cur_elem exhausted, too */
348 : }
349 :
350 7896 : if (high_elem < low_elem)
351 : {
352 : /* Caller needs to perform "beyond end" array advancement */
353 7816 : *set_elem_result = 1;
354 7816 : return high_elem;
355 : }
356 : }
357 : else
358 : {
359 60 : high_elem = array->cur_elem - 1; /* old cur_elem exhausted */
360 :
361 : /* Compare prospective new cur_elem (also the new upper bound) */
362 60 : if (high_elem >= low_elem)
363 : {
364 42 : arrdatum = array->elem_values[high_elem];
365 42 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
366 : arrdatum, cur);
367 :
368 42 : if (result >= 0)
369 : {
370 : /* Optimistic comparison optimization worked out */
371 30 : *set_elem_result = result;
372 30 : return high_elem;
373 : }
374 12 : mid_elem = high_elem;
375 12 : high_elem--; /* this cur_elem exhausted, too */
376 : }
377 :
378 30 : if (high_elem < low_elem)
379 : {
380 : /* Caller needs to perform "beyond end" array advancement */
381 30 : *set_elem_result = -1;
382 30 : return low_elem;
383 : }
384 : }
385 : }
386 :
387 698 : while (high_elem > low_elem)
388 : {
389 438 : mid_elem = low_elem + ((high_elem - low_elem) / 2);
390 438 : arrdatum = array->elem_values[mid_elem];
391 :
392 438 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
393 : arrdatum, cur);
394 :
395 438 : if (result == 0)
396 : {
397 : /*
398 : * It's safe to quit as soon as we see an equal array element.
399 : * This often saves an extra comparison or two...
400 : */
401 138 : low_elem = mid_elem;
402 138 : break;
403 : }
404 :
405 300 : if (result > 0)
406 270 : low_elem = mid_elem + 1;
407 : else
408 30 : high_elem = mid_elem;
409 : }
410 :
411 : /*
412 : * ...but our caller also cares about how its searched-for tuple datum
413 : * compares to the low_elem datum. Must always set *set_elem_result with
414 : * the result of that comparison specifically.
415 : */
416 398 : if (low_elem != mid_elem)
417 242 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
418 242 : array->elem_values[low_elem], cur);
419 :
420 398 : *set_elem_result = result;
421 :
422 398 : return low_elem;
423 : }
424 :
425 : /*
426 : * _bt_binsrch_skiparray_skey() -- "Binary search" within a skip array
427 : *
428 : * Does not return an index into the array, since skip arrays don't really
429 : * contain elements (they generate their array elements procedurally instead).
430 : * Our interface matches that of _bt_binsrch_array_skey in every other way.
431 : *
432 : * Sets *set_elem_result just like _bt_binsrch_array_skey would with a true
433 : * array. The value 0 indicates that tupdatum/tupnull is within the range of
434 : * the skip array. We return -1 when tupdatum/tupnull is lower that any value
435 : * within the range of the array, and 1 when it is higher than every value.
436 : * Caller should pass *set_elem_result to _bt_skiparray_set_element to advance
437 : * the array.
438 : *
439 : * cur_elem_trig indicates if array advancement was triggered by this array's
440 : * scan key. We use this to optimize-away comparisons that are known by our
441 : * caller to be unnecessary from context, just like _bt_binsrch_array_skey.
442 : */
443 : static void
444 166548 : _bt_binsrch_skiparray_skey(bool cur_elem_trig, ScanDirection dir,
445 : Datum tupdatum, bool tupnull,
446 : BTArrayKeyInfo *array, ScanKey cur,
447 : int32 *set_elem_result)
448 : {
449 : Assert(cur->sk_flags & SK_BT_SKIP);
450 : Assert(cur->sk_flags & SK_SEARCHARRAY);
451 : Assert(cur->sk_flags & SK_BT_REQFWD);
452 : Assert(array->num_elems == -1);
453 : Assert(!ScanDirectionIsNoMovement(dir));
454 :
455 166548 : if (array->null_elem)
456 : {
457 : Assert(!array->low_compare && !array->high_compare);
458 :
459 138986 : *set_elem_result = 0;
460 138986 : return;
461 : }
462 :
463 27562 : if (tupnull) /* NULL tupdatum */
464 : {
465 24 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
466 0 : *set_elem_result = -1; /* NULL "<" NOT_NULL */
467 : else
468 24 : *set_elem_result = 1; /* NULL ">" NOT_NULL */
469 24 : return;
470 : }
471 :
472 : /*
473 : * Array inequalities determine whether tupdatum is within the range of
474 : * caller's skip array
475 : */
476 27538 : *set_elem_result = 0;
477 27538 : if (ScanDirectionIsForward(dir))
478 : {
479 : /*
480 : * Evaluate low_compare first (unless cur_elem_trig tells us that it
481 : * cannot possibly fail to be satisfied), then evaluate high_compare
482 : */
483 27490 : if (!cur_elem_trig && array->low_compare &&
484 772 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
485 772 : array->low_compare->sk_collation,
486 : tupdatum,
487 772 : array->low_compare->sk_argument)))
488 0 : *set_elem_result = -1;
489 27490 : else if (array->high_compare &&
490 11398 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
491 11398 : array->high_compare->sk_collation,
492 : tupdatum,
493 11398 : array->high_compare->sk_argument)))
494 6406 : *set_elem_result = 1;
495 : }
496 : else
497 : {
498 : /*
499 : * Evaluate high_compare first (unless cur_elem_trig tells us that it
500 : * cannot possibly fail to be satisfied), then evaluate low_compare
501 : */
502 48 : if (!cur_elem_trig && array->high_compare &&
503 6 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
504 6 : array->high_compare->sk_collation,
505 : tupdatum,
506 6 : array->high_compare->sk_argument)))
507 0 : *set_elem_result = 1;
508 48 : else if (array->low_compare &&
509 24 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
510 24 : array->low_compare->sk_collation,
511 : tupdatum,
512 24 : array->low_compare->sk_argument)))
513 0 : *set_elem_result = -1;
514 : }
515 :
516 : /*
517 : * Assert that any keys that were assumed to be satisfied already (due to
518 : * caller passing cur_elem_trig=true) really are satisfied as expected
519 : */
520 : #ifdef USE_ASSERT_CHECKING
521 : if (cur_elem_trig)
522 : {
523 : if (ScanDirectionIsForward(dir) && array->low_compare)
524 : Assert(DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
525 : array->low_compare->sk_collation,
526 : tupdatum,
527 : array->low_compare->sk_argument)));
528 :
529 : if (ScanDirectionIsBackward(dir) && array->high_compare)
530 : Assert(DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
531 : array->high_compare->sk_collation,
532 : tupdatum,
533 : array->high_compare->sk_argument)));
534 : }
535 : #endif
536 : }
537 :
538 : /*
539 : * _bt_skiparray_set_element() -- Set skip array scan key's sk_argument
540 : *
541 : * Caller passes set_elem_result returned by _bt_binsrch_skiparray_skey for
542 : * caller's tupdatum/tupnull.
543 : *
544 : * We copy tupdatum/tupnull into skey's sk_argument iff set_elem_result == 0.
545 : * Otherwise, we set skey to either the lowest or highest value that's within
546 : * the range of caller's skip array (whichever is the best available match to
547 : * tupdatum/tupnull that is still within the range of the skip array according
548 : * to _bt_binsrch_skiparray_skey/set_elem_result).
549 : */
550 : static void
551 154836 : _bt_skiparray_set_element(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
552 : int32 set_elem_result, Datum tupdatum, bool tupnull)
553 : {
554 : Assert(skey->sk_flags & SK_BT_SKIP);
555 : Assert(skey->sk_flags & SK_SEARCHARRAY);
556 :
557 154836 : if (set_elem_result)
558 : {
559 : /* tupdatum/tupnull is out of the range of the skip array */
560 : Assert(!array->null_elem);
561 :
562 642 : _bt_array_set_low_or_high(rel, skey, array, set_elem_result < 0);
563 642 : return;
564 : }
565 :
566 : /* Advance skip array to tupdatum (or tupnull) value */
567 154194 : if (unlikely(tupnull))
568 : {
569 36 : _bt_skiparray_set_isnull(rel, skey, array);
570 36 : return;
571 : }
572 :
573 : /* Free memory previously allocated for sk_argument if needed */
574 154158 : if (!array->attbyval && skey->sk_argument)
575 78148 : pfree(DatumGetPointer(skey->sk_argument));
576 :
577 : /* tupdatum becomes new sk_argument/new current element */
578 154158 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL |
579 : SK_BT_MINVAL | SK_BT_MAXVAL |
580 : SK_BT_NEXT | SK_BT_PRIOR);
581 154158 : skey->sk_argument = datumCopy(tupdatum, array->attbyval, array->attlen);
582 : }
583 :
584 : /*
585 : * _bt_skiparray_set_isnull() -- set skip array scan key to NULL
586 : */
587 : static void
588 48 : _bt_skiparray_set_isnull(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
589 : {
590 : Assert(skey->sk_flags & SK_BT_SKIP);
591 : Assert(skey->sk_flags & SK_SEARCHARRAY);
592 : Assert(array->null_elem && !array->low_compare && !array->high_compare);
593 :
594 : /* Free memory previously allocated for sk_argument if needed */
595 48 : if (!array->attbyval && skey->sk_argument)
596 6 : pfree(DatumGetPointer(skey->sk_argument));
597 :
598 : /* NULL becomes new sk_argument/new current element */
599 48 : skey->sk_argument = (Datum) 0;
600 48 : skey->sk_flags &= ~(SK_BT_MINVAL | SK_BT_MAXVAL |
601 : SK_BT_NEXT | SK_BT_PRIOR);
602 48 : skey->sk_flags |= (SK_SEARCHNULL | SK_ISNULL);
603 48 : }
604 :
605 : /*
606 : * _bt_start_array_keys() -- Initialize array keys at start of a scan
607 : *
608 : * Set up the cur_elem counters and fill in the first sk_argument value for
609 : * each array scankey.
610 : */
611 : void
612 81198 : _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
613 : {
614 81198 : Relation rel = scan->indexRelation;
615 81198 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
616 :
617 : Assert(so->numArrayKeys);
618 : Assert(so->qual_ok);
619 :
620 163008 : for (int i = 0; i < so->numArrayKeys; i++)
621 : {
622 81810 : BTArrayKeyInfo *array = &so->arrayKeys[i];
623 81810 : ScanKey skey = &so->keyData[array->scan_key];
624 :
625 : Assert(skey->sk_flags & SK_SEARCHARRAY);
626 :
627 81810 : _bt_array_set_low_or_high(rel, skey, array,
628 : ScanDirectionIsForward(dir));
629 : }
630 81198 : so->scanBehind = so->oppositeDirCheck = false; /* reset */
631 81198 : }
632 :
633 : /*
634 : * _bt_array_set_low_or_high() -- Set array scan key to lowest/highest element
635 : *
636 : * Caller also passes associated scan key, which will have its argument set to
637 : * the lowest/highest array value in passing.
638 : */
639 : static void
640 92972 : _bt_array_set_low_or_high(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
641 : bool low_not_high)
642 : {
643 : Assert(skey->sk_flags & SK_SEARCHARRAY);
644 :
645 92972 : if (array->num_elems != -1)
646 : {
647 : /* set low or high element for SAOP array */
648 83920 : int set_elem = 0;
649 :
650 : Assert(!(skey->sk_flags & SK_BT_SKIP));
651 :
652 83920 : if (!low_not_high)
653 8048 : set_elem = array->num_elems - 1;
654 :
655 : /*
656 : * Just copy over array datum (only skip arrays require freeing and
657 : * allocating memory for sk_argument)
658 : */
659 83920 : array->cur_elem = set_elem;
660 83920 : skey->sk_argument = array->elem_values[set_elem];
661 :
662 83920 : return;
663 : }
664 :
665 : /* set low or high element for skip array */
666 : Assert(skey->sk_flags & SK_BT_SKIP);
667 : Assert(array->num_elems == -1);
668 :
669 : /* Free memory previously allocated for sk_argument if needed */
670 9052 : if (!array->attbyval && skey->sk_argument)
671 1858 : pfree(DatumGetPointer(skey->sk_argument));
672 :
673 : /* Reset flags */
674 9052 : skey->sk_argument = (Datum) 0;
675 9052 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL |
676 : SK_BT_MINVAL | SK_BT_MAXVAL |
677 : SK_BT_NEXT | SK_BT_PRIOR);
678 :
679 9052 : if (array->null_elem &&
680 7194 : (low_not_high == ((skey->sk_flags & SK_BT_NULLS_FIRST) != 0)))
681 : {
682 : /* Requested element (either lowest or highest) has the value NULL */
683 954 : skey->sk_flags |= (SK_SEARCHNULL | SK_ISNULL);
684 : }
685 8098 : else if (low_not_high)
686 : {
687 : /* Setting array to lowest element (according to low_compare) */
688 7378 : skey->sk_flags |= SK_BT_MINVAL;
689 : }
690 : else
691 : {
692 : /* Setting array to highest element (according to high_compare) */
693 720 : skey->sk_flags |= SK_BT_MAXVAL;
694 : }
695 : }
696 :
697 : /*
698 : * _bt_array_decrement() -- decrement array scan key's sk_argument
699 : *
700 : * Return value indicates whether caller's array was successfully decremented.
701 : * Cannot decrement an array whose current element is already the first one.
702 : */
703 : static bool
704 912 : _bt_array_decrement(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
705 : {
706 912 : bool uflow = false;
707 : Datum dec_sk_argument;
708 :
709 : Assert(skey->sk_flags & SK_SEARCHARRAY);
710 : Assert(!(skey->sk_flags & (SK_BT_MAXVAL | SK_BT_NEXT | SK_BT_PRIOR)));
711 :
712 : /* SAOP array? */
713 912 : if (array->num_elems != -1)
714 : {
715 : Assert(!(skey->sk_flags & (SK_BT_SKIP | SK_BT_MINVAL | SK_BT_MAXVAL)));
716 36 : if (array->cur_elem > 0)
717 : {
718 : /*
719 : * Just decrement current element, and assign its datum to skey
720 : * (only skip arrays need us to free existing sk_argument memory)
721 : */
722 6 : array->cur_elem--;
723 6 : skey->sk_argument = array->elem_values[array->cur_elem];
724 :
725 : /* Successfully decremented array */
726 6 : return true;
727 : }
728 :
729 : /* Cannot decrement to before first array element */
730 30 : return false;
731 : }
732 :
733 : /* Nope, this is a skip array */
734 : Assert(skey->sk_flags & SK_BT_SKIP);
735 :
736 : /*
737 : * The sentinel value that represents the minimum value within the range
738 : * of a skip array (often just -inf) is never decrementable
739 : */
740 876 : if (skey->sk_flags & SK_BT_MINVAL)
741 0 : return false;
742 :
743 : /*
744 : * When the current array element is NULL, and the lowest sorting value in
745 : * the index is also NULL, we cannot decrement before first array element
746 : */
747 876 : if ((skey->sk_flags & SK_ISNULL) && (skey->sk_flags & SK_BT_NULLS_FIRST))
748 0 : return false;
749 :
750 : /*
751 : * Opclasses without skip support "decrement" the scan key's current
752 : * element by setting the PRIOR flag. The true prior value is determined
753 : * by repositioning to the last index tuple < existing sk_argument/current
754 : * array element. Note that this works in the usual way when the scan key
755 : * is already marked ISNULL (i.e. when the current element is NULL).
756 : */
757 876 : if (!array->sksup)
758 : {
759 : /* Successfully "decremented" array */
760 12 : skey->sk_flags |= SK_BT_PRIOR;
761 12 : return true;
762 : }
763 :
764 : /*
765 : * Opclasses with skip support directly decrement sk_argument
766 : */
767 864 : if (skey->sk_flags & SK_ISNULL)
768 : {
769 : Assert(!(skey->sk_flags & SK_BT_NULLS_FIRST));
770 :
771 : /*
772 : * Existing sk_argument/array element is NULL (for an IS NULL qual).
773 : *
774 : * "Decrement" from NULL to the high_elem value provided by opclass
775 : * skip support routine.
776 : */
777 6 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL);
778 12 : skey->sk_argument = datumCopy(array->sksup->high_elem,
779 6 : array->attbyval, array->attlen);
780 6 : return true;
781 : }
782 :
783 : /*
784 : * Ask opclass support routine to provide decremented copy of existing
785 : * non-NULL sk_argument
786 : */
787 858 : dec_sk_argument = array->sksup->decrement(rel, skey->sk_argument, &uflow);
788 858 : if (unlikely(uflow))
789 : {
790 : /* dec_sk_argument has undefined value (so no pfree) */
791 0 : if (array->null_elem && (skey->sk_flags & SK_BT_NULLS_FIRST))
792 : {
793 0 : _bt_skiparray_set_isnull(rel, skey, array);
794 :
795 : /* Successfully "decremented" array to NULL */
796 0 : return true;
797 : }
798 :
799 : /* Cannot decrement to before first array element */
800 0 : return false;
801 : }
802 :
803 : /*
804 : * Successfully decremented sk_argument to a non-NULL value. Make sure
805 : * that the decremented value is still within the range of the array.
806 : */
807 858 : if (array->low_compare &&
808 12 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
809 12 : array->low_compare->sk_collation,
810 : dec_sk_argument,
811 12 : array->low_compare->sk_argument)))
812 : {
813 : /* Keep existing sk_argument after all */
814 6 : if (!array->attbyval)
815 0 : pfree(DatumGetPointer(dec_sk_argument));
816 :
817 : /* Cannot decrement to before first array element */
818 6 : return false;
819 : }
820 :
821 : /* Accept value returned by opclass decrement callback */
822 852 : if (!array->attbyval && skey->sk_argument)
823 0 : pfree(DatumGetPointer(skey->sk_argument));
824 852 : skey->sk_argument = dec_sk_argument;
825 :
826 : /* Successfully decremented array */
827 852 : return true;
828 : }
829 :
830 : /*
831 : * _bt_array_increment() -- increment array scan key's sk_argument
832 : *
833 : * Return value indicates whether caller's array was successfully incremented.
834 : * Cannot increment an array whose current element is already the final one.
835 : */
836 : static bool
837 32292 : _bt_array_increment(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
838 : {
839 32292 : bool oflow = false;
840 : Datum inc_sk_argument;
841 :
842 : Assert(skey->sk_flags & SK_SEARCHARRAY);
843 : Assert(!(skey->sk_flags & (SK_BT_MINVAL | SK_BT_NEXT | SK_BT_PRIOR)));
844 :
845 : /* SAOP array? */
846 32292 : if (array->num_elems != -1)
847 : {
848 : Assert(!(skey->sk_flags & (SK_BT_SKIP | SK_BT_MINVAL | SK_BT_MAXVAL)));
849 8072 : if (array->cur_elem < array->num_elems - 1)
850 : {
851 : /*
852 : * Just increment current element, and assign its datum to skey
853 : * (only skip arrays need us to free existing sk_argument memory)
854 : */
855 38 : array->cur_elem++;
856 38 : skey->sk_argument = array->elem_values[array->cur_elem];
857 :
858 : /* Successfully incremented array */
859 38 : return true;
860 : }
861 :
862 : /* Cannot increment past final array element */
863 8034 : return false;
864 : }
865 :
866 : /* Nope, this is a skip array */
867 : Assert(skey->sk_flags & SK_BT_SKIP);
868 :
869 : /*
870 : * The sentinel value that represents the maximum value within the range
871 : * of a skip array (often just +inf) is never incrementable
872 : */
873 24220 : if (skey->sk_flags & SK_BT_MAXVAL)
874 642 : return false;
875 :
876 : /*
877 : * When the current array element is NULL, and the highest sorting value
878 : * in the index is also NULL, we cannot increment past the final element
879 : */
880 23578 : if ((skey->sk_flags & SK_ISNULL) && !(skey->sk_flags & SK_BT_NULLS_FIRST))
881 432 : return false;
882 :
883 : /*
884 : * Opclasses without skip support "increment" the scan key's current
885 : * element by setting the NEXT flag. The true next value is determined by
886 : * repositioning to the first index tuple > existing sk_argument/current
887 : * array element. Note that this works in the usual way when the scan key
888 : * is already marked ISNULL (i.e. when the current element is NULL).
889 : */
890 23146 : if (!array->sksup)
891 : {
892 : /* Successfully "incremented" array */
893 15864 : skey->sk_flags |= SK_BT_NEXT;
894 15864 : return true;
895 : }
896 :
897 : /*
898 : * Opclasses with skip support directly increment sk_argument
899 : */
900 7282 : if (skey->sk_flags & SK_ISNULL)
901 : {
902 : Assert(skey->sk_flags & SK_BT_NULLS_FIRST);
903 :
904 : /*
905 : * Existing sk_argument/array element is NULL (for an IS NULL qual).
906 : *
907 : * "Increment" from NULL to the low_elem value provided by opclass
908 : * skip support routine.
909 : */
910 36 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL);
911 72 : skey->sk_argument = datumCopy(array->sksup->low_elem,
912 36 : array->attbyval, array->attlen);
913 36 : return true;
914 : }
915 :
916 : /*
917 : * Ask opclass support routine to provide incremented copy of existing
918 : * non-NULL sk_argument
919 : */
920 7246 : inc_sk_argument = array->sksup->increment(rel, skey->sk_argument, &oflow);
921 7246 : if (unlikely(oflow))
922 : {
923 : /* inc_sk_argument has undefined value (so no pfree) */
924 30 : if (array->null_elem && !(skey->sk_flags & SK_BT_NULLS_FIRST))
925 : {
926 12 : _bt_skiparray_set_isnull(rel, skey, array);
927 :
928 : /* Successfully "incremented" array to NULL */
929 12 : return true;
930 : }
931 :
932 : /* Cannot increment past final array element */
933 18 : return false;
934 : }
935 :
936 : /*
937 : * Successfully incremented sk_argument to a non-NULL value. Make sure
938 : * that the incremented value is still within the range of the array.
939 : */
940 7216 : if (array->high_compare &&
941 42 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
942 42 : array->high_compare->sk_collation,
943 : inc_sk_argument,
944 42 : array->high_compare->sk_argument)))
945 : {
946 : /* Keep existing sk_argument after all */
947 12 : if (!array->attbyval)
948 0 : pfree(DatumGetPointer(inc_sk_argument));
949 :
950 : /* Cannot increment past final array element */
951 12 : return false;
952 : }
953 :
954 : /* Accept value returned by opclass increment callback */
955 7204 : if (!array->attbyval && skey->sk_argument)
956 0 : pfree(DatumGetPointer(skey->sk_argument));
957 7204 : skey->sk_argument = inc_sk_argument;
958 :
959 : /* Successfully incremented array */
960 7204 : return true;
961 : }
962 :
963 : /*
964 : * _bt_advance_array_keys_increment() -- Advance to next set of array elements
965 : *
966 : * Advances the array keys by a single increment in the current scan
967 : * direction. When there are multiple array keys this can roll over from the
968 : * lowest order array to higher order arrays.
969 : *
970 : * Returns true if there is another set of values to consider, false if not.
971 : * On true result, the scankeys are initialized with the next set of values.
972 : * On false result, the scankeys stay the same, and the array keys are not
973 : * advanced (every array remains at its final element for scan direction).
974 : */
975 : static bool
976 32070 : _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
977 : bool *skip_array_set)
978 : {
979 32070 : Relation rel = scan->indexRelation;
980 32070 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
981 :
982 : /*
983 : * We must advance the last array key most quickly, since it will
984 : * correspond to the lowest-order index column among the available
985 : * qualifications
986 : */
987 41244 : for (int i = so->numArrayKeys - 1; i >= 0; i--)
988 : {
989 33204 : BTArrayKeyInfo *array = &so->arrayKeys[i];
990 33204 : ScanKey skey = &so->keyData[array->scan_key];
991 :
992 33204 : if (array->num_elems == -1)
993 25096 : *skip_array_set = true;
994 :
995 33204 : if (ScanDirectionIsForward(dir))
996 : {
997 32292 : if (_bt_array_increment(rel, skey, array))
998 23154 : return true;
999 : }
1000 : else
1001 : {
1002 912 : if (_bt_array_decrement(rel, skey, array))
1003 876 : return true;
1004 : }
1005 :
1006 : /*
1007 : * Couldn't increment (or decrement) array. Handle array roll over.
1008 : *
1009 : * Start over at the array's lowest sorting value (or its highest
1010 : * value, for backward scans)...
1011 : */
1012 9174 : _bt_array_set_low_or_high(rel, skey, array,
1013 : ScanDirectionIsForward(dir));
1014 :
1015 : /* ...then increment (or decrement) next most significant array */
1016 : }
1017 :
1018 : /*
1019 : * The array keys are now exhausted.
1020 : *
1021 : * Restore the array keys to the state they were in immediately before we
1022 : * were called. This ensures that the arrays only ever ratchet in the
1023 : * current scan direction.
1024 : *
1025 : * Without this, scans could overlook matching tuples when the scan
1026 : * direction gets reversed just before btgettuple runs out of items to
1027 : * return, but just after _bt_readpage prepares all the items from the
1028 : * scan's final page in so->currPos. When we're on the final page it is
1029 : * typical for so->currPos to get invalidated once btgettuple finally
1030 : * returns false, which'll effectively invalidate the scan's array keys.
1031 : * That hasn't happened yet, though -- and in general it may never happen.
1032 : */
1033 8040 : _bt_start_array_keys(scan, -dir);
1034 :
1035 8040 : return false;
1036 : }
1037 :
1038 : /*
1039 : * _bt_tuple_before_array_skeys() -- too early to advance required arrays?
1040 : *
1041 : * We always compare the tuple using the current array keys (which we assume
1042 : * are already set in so->keyData[]). readpagetup indicates if tuple is the
1043 : * scan's current _bt_readpage-wise tuple.
1044 : *
1045 : * readpagetup callers must only call here when _bt_check_compare already set
1046 : * continuescan=false. We help these callers deal with _bt_check_compare's
1047 : * inability to distinguish between the < and > cases (it uses equality
1048 : * operator scan keys, whereas we use 3-way ORDER procs). These callers pass
1049 : * a _bt_check_compare-set sktrig value that indicates which scan key
1050 : * triggered the call (!readpagetup callers just pass us sktrig=0 instead).
1051 : * This information allows us to avoid wastefully checking earlier scan keys
1052 : * that were already deemed to have been satisfied inside _bt_check_compare.
1053 : *
1054 : * Returns false when caller's tuple is >= the current required equality scan
1055 : * keys (or <=, in the case of backwards scans). This happens to readpagetup
1056 : * callers when the scan has reached the point of needing its array keys
1057 : * advanced; caller will need to advance required and non-required arrays at
1058 : * scan key offsets >= sktrig, plus scan keys < sktrig iff sktrig rolls over.
1059 : * (When we return false to readpagetup callers, tuple can only be == current
1060 : * required equality scan keys when caller's sktrig indicates that the arrays
1061 : * need to be advanced due to an unsatisfied required inequality key trigger.)
1062 : *
1063 : * Returns true when caller passes a tuple that is < the current set of
1064 : * equality keys for the most significant non-equal required scan key/column
1065 : * (or > the keys, during backwards scans). This happens to readpagetup
1066 : * callers when tuple is still before the start of matches for the scan's
1067 : * required equality strategy scan keys. (sktrig can't have indicated that an
1068 : * inequality strategy scan key wasn't satisfied in _bt_check_compare when we
1069 : * return true. In fact, we automatically return false when passed such an
1070 : * inequality sktrig by readpagetup callers -- _bt_check_compare's initial
1071 : * continuescan=false doesn't really need to be confirmed here by us.)
1072 : *
1073 : * !readpagetup callers optionally pass us *scanBehind, which tracks whether
1074 : * any missing truncated attributes might have affected array advancement
1075 : * (compared to what would happen if it was shown the first non-pivot tuple on
1076 : * the page to the right of caller's finaltup/high key tuple instead). It's
1077 : * only possible that we'll set *scanBehind to true when caller passes us a
1078 : * pivot tuple (with truncated -inf attributes) that we return false for.
1079 : */
1080 : static bool
1081 322780 : _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
1082 : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
1083 : bool readpagetup, int sktrig, bool *scanBehind)
1084 : {
1085 322780 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1086 :
1087 : Assert(so->numArrayKeys);
1088 : Assert(so->numberOfKeys);
1089 : Assert(sktrig == 0 || readpagetup);
1090 : Assert(!readpagetup || scanBehind == NULL);
1091 :
1092 322780 : if (scanBehind)
1093 83250 : *scanBehind = false;
1094 :
1095 325656 : for (int ikey = sktrig; ikey < so->numberOfKeys; ikey++)
1096 : {
1097 325222 : ScanKey cur = so->keyData + ikey;
1098 : Datum tupdatum;
1099 : bool tupnull;
1100 : int32 result;
1101 :
1102 : /* readpagetup calls require one ORDER proc comparison (at most) */
1103 : Assert(!readpagetup || ikey == sktrig);
1104 :
1105 : /*
1106 : * Once we reach a non-required scan key, we're completely done.
1107 : *
1108 : * Note: we deliberately don't consider the scan direction here.
1109 : * _bt_advance_array_keys caller requires that we track *scanBehind
1110 : * without concern for scan direction.
1111 : */
1112 325222 : if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) == 0)
1113 : {
1114 : Assert(!readpagetup);
1115 : Assert(ikey > sktrig || ikey == 0);
1116 322346 : return false;
1117 : }
1118 :
1119 325222 : if (cur->sk_attno > tupnatts)
1120 : {
1121 : Assert(!readpagetup);
1122 :
1123 : /*
1124 : * When we reach a high key's truncated attribute, assume that the
1125 : * tuple attribute's value is >= the scan's equality constraint
1126 : * scan keys (but set *scanBehind to let interested callers know
1127 : * that a truncated attribute might have affected our answer).
1128 : */
1129 32 : if (scanBehind)
1130 32 : *scanBehind = true;
1131 :
1132 32 : return false;
1133 : }
1134 :
1135 : /*
1136 : * Deal with inequality strategy scan keys that _bt_check_compare set
1137 : * continuescan=false for
1138 : */
1139 325190 : if (cur->sk_strategy != BTEqualStrategyNumber)
1140 : {
1141 : /*
1142 : * When _bt_check_compare indicated that a required inequality
1143 : * scan key wasn't satisfied, there's no need to verify anything;
1144 : * caller always calls _bt_advance_array_keys with this sktrig.
1145 : */
1146 628 : if (readpagetup)
1147 348 : return false;
1148 :
1149 : /*
1150 : * Otherwise we can't give up, since we must check all required
1151 : * scan keys (required in either direction) in order to correctly
1152 : * track *scanBehind for caller
1153 : */
1154 280 : continue;
1155 : }
1156 :
1157 324562 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1158 :
1159 324562 : if (likely(!(cur->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL))))
1160 : {
1161 : /* Scankey has a valid/comparable sk_argument value */
1162 319136 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1163 : tupdatum, tupnull,
1164 : cur->sk_argument, cur);
1165 :
1166 319136 : if (result == 0)
1167 : {
1168 : /*
1169 : * Interpret result in a way that takes NEXT/PRIOR into
1170 : * account
1171 : */
1172 16884 : if (cur->sk_flags & SK_BT_NEXT)
1173 14258 : result = -1;
1174 2626 : else if (cur->sk_flags & SK_BT_PRIOR)
1175 30 : result = 1;
1176 :
1177 : Assert(result == 0 || (cur->sk_flags & SK_BT_SKIP));
1178 : }
1179 : }
1180 : else
1181 : {
1182 5426 : BTArrayKeyInfo *array = NULL;
1183 :
1184 : /*
1185 : * Current array element/array = scan key value is a sentinel
1186 : * value that represents the lowest (or highest) possible value
1187 : * that's still within the range of the array.
1188 : *
1189 : * Like _bt_first, we only see MINVAL keys during forwards scans
1190 : * (and similarly only see MAXVAL keys during backwards scans).
1191 : * Even if the scan's direction changes, we'll stop at some higher
1192 : * order key before we can ever reach any MAXVAL (or MINVAL) keys.
1193 : * (However, unlike _bt_first we _can_ get to keys marked either
1194 : * NEXT or PRIOR, regardless of the scan's current direction.)
1195 : */
1196 : Assert(ScanDirectionIsForward(dir) ?
1197 : !(cur->sk_flags & SK_BT_MAXVAL) :
1198 : !(cur->sk_flags & SK_BT_MINVAL));
1199 :
1200 : /*
1201 : * There are no valid sk_argument values in MINVAL/MAXVAL keys.
1202 : * Check if tupdatum is within the range of skip array instead.
1203 : */
1204 5912 : for (int arrayidx = 0; arrayidx < so->numArrayKeys; arrayidx++)
1205 : {
1206 5912 : array = &so->arrayKeys[arrayidx];
1207 5912 : if (array->scan_key == ikey)
1208 5426 : break;
1209 : }
1210 :
1211 5426 : _bt_binsrch_skiparray_skey(false, dir, tupdatum, tupnull,
1212 : array, cur, &result);
1213 :
1214 5426 : if (result == 0)
1215 : {
1216 : /*
1217 : * tupdatum satisfies both low_compare and high_compare, so
1218 : * it's time to advance the array keys.
1219 : *
1220 : * Note: It's possible that the skip array will "advance" from
1221 : * its MINVAL (or MAXVAL) representation to an alternative,
1222 : * logically equivalent representation of the same value: a
1223 : * representation where the = key gets a valid datum in its
1224 : * sk_argument. This is only possible when low_compare uses
1225 : * the >= strategy (or high_compare uses the <= strategy).
1226 : */
1227 5418 : return false;
1228 : }
1229 : }
1230 :
1231 : /*
1232 : * Does this comparison indicate that caller must _not_ advance the
1233 : * scan's arrays just yet?
1234 : */
1235 319144 : if ((ScanDirectionIsForward(dir) && result < 0) ||
1236 3252 : (ScanDirectionIsBackward(dir) && result > 0))
1237 59612 : return true;
1238 :
1239 : /*
1240 : * Does this comparison indicate that caller should now advance the
1241 : * scan's arrays? (Must be if we get here during a readpagetup call.)
1242 : */
1243 259532 : if (readpagetup || result != 0)
1244 : {
1245 : Assert(result != 0);
1246 256936 : return false;
1247 : }
1248 :
1249 : /*
1250 : * Inconclusive -- need to check later scan keys, too.
1251 : *
1252 : * This must be a finaltup precheck, or a call made from an assertion.
1253 : */
1254 : Assert(result == 0);
1255 : }
1256 :
1257 : Assert(!readpagetup);
1258 :
1259 434 : return false;
1260 : }
1261 :
1262 : /*
1263 : * _bt_start_prim_scan() -- start scheduled primitive index scan?
1264 : *
1265 : * Returns true if _bt_checkkeys scheduled another primitive index scan, just
1266 : * as the last one ended. Otherwise returns false, indicating that the array
1267 : * keys are now fully exhausted.
1268 : *
1269 : * Only call here during scans with one or more equality type array scan keys,
1270 : * after _bt_first or _bt_next return false.
1271 : */
1272 : bool
1273 88498 : _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir)
1274 : {
1275 88498 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1276 :
1277 : Assert(so->numArrayKeys);
1278 :
1279 88498 : so->scanBehind = so->oppositeDirCheck = false; /* reset */
1280 :
1281 : /*
1282 : * Array keys are advanced within _bt_checkkeys when the scan reaches the
1283 : * leaf level (more precisely, they're advanced when the scan reaches the
1284 : * end of each distinct set of array elements). This process avoids
1285 : * repeat access to leaf pages (across multiple primitive index scans) by
1286 : * advancing the scan's array keys when it allows the primitive index scan
1287 : * to find nearby matching tuples (or when it eliminates ranges of array
1288 : * key space that can't possibly be satisfied by any index tuple).
1289 : *
1290 : * _bt_checkkeys sets a simple flag variable to schedule another primitive
1291 : * index scan. The flag tells us what to do.
1292 : *
1293 : * We cannot rely on _bt_first always reaching _bt_checkkeys. There are
1294 : * various cases where that won't happen. For example, if the index is
1295 : * completely empty, then _bt_first won't call _bt_readpage/_bt_checkkeys.
1296 : * We also don't expect a call to _bt_checkkeys during searches for a
1297 : * non-existent value that happens to be lower/higher than any existing
1298 : * value in the index.
1299 : *
1300 : * We don't require special handling for these cases -- we don't need to
1301 : * be explicitly instructed to _not_ perform another primitive index scan.
1302 : * It's up to code under the control of _bt_first to always set the flag
1303 : * when another primitive index scan will be required.
1304 : *
1305 : * This works correctly, even with the tricky cases listed above, which
1306 : * all involve access to leaf pages "near the boundaries of the key space"
1307 : * (whether it's from a leftmost/rightmost page, or an imaginary empty
1308 : * leaf root page). If _bt_checkkeys cannot be reached by a primitive
1309 : * index scan for one set of array keys, then it also won't be reached for
1310 : * any later set ("later" in terms of the direction that we scan the index
1311 : * and advance the arrays). The array keys won't have advanced in these
1312 : * cases, but that's the correct behavior (even _bt_advance_array_keys
1313 : * won't always advance the arrays at the point they become "exhausted").
1314 : */
1315 88498 : if (so->needPrimScan)
1316 : {
1317 : /*
1318 : * Flag was set -- must call _bt_first again, which will reset the
1319 : * scan's needPrimScan flag
1320 : */
1321 17512 : return true;
1322 : }
1323 :
1324 : /* The top-level index scan ran out of tuples in this scan direction */
1325 70986 : if (scan->parallel_scan != NULL)
1326 30 : _bt_parallel_done(scan);
1327 :
1328 70986 : return false;
1329 : }
1330 :
1331 : /*
1332 : * _bt_advance_array_keys() -- Advance array elements using a tuple
1333 : *
1334 : * The scan always gets a new qual as a consequence of calling here (except
1335 : * when we determine that the top-level scan has run out of matching tuples).
1336 : * All later _bt_check_compare calls also use the same new qual that was first
1337 : * used here (at least until the next call here advances the keys once again).
1338 : * It's convenient to structure _bt_check_compare rechecks of caller's tuple
1339 : * (using the new qual) as one the steps of advancing the scan's array keys,
1340 : * so this function works as a wrapper around _bt_check_compare.
1341 : *
1342 : * Like _bt_check_compare, we'll set pstate.continuescan on behalf of the
1343 : * caller, and return a boolean indicating if caller's tuple satisfies the
1344 : * scan's new qual. But unlike _bt_check_compare, we set so->needPrimScan
1345 : * when we set continuescan=false, indicating if a new primitive index scan
1346 : * has been scheduled (otherwise, the top-level scan has run out of tuples in
1347 : * the current scan direction).
1348 : *
1349 : * Caller must use _bt_tuple_before_array_skeys to determine if the current
1350 : * place in the scan is >= the current array keys _before_ calling here.
1351 : * We're responsible for ensuring that caller's tuple is <= the newly advanced
1352 : * required array keys once we return. We try to find an exact match, but
1353 : * failing that we'll advance the array keys to whatever set of array elements
1354 : * comes next in the key space for the current scan direction. Required array
1355 : * keys "ratchet forwards" (or backwards). They can only advance as the scan
1356 : * itself advances through the index/key space.
1357 : *
1358 : * (The rules are the same for backwards scans, except that the operators are
1359 : * flipped: just replace the precondition's >= operator with a <=, and the
1360 : * postcondition's <= operator with a >=. In other words, just swap the
1361 : * precondition with the postcondition.)
1362 : *
1363 : * We also deal with "advancing" non-required arrays here (or arrays that are
1364 : * treated as non-required for the duration of a _bt_readpage call). Callers
1365 : * whose sktrig scan key is non-required specify sktrig_required=false. These
1366 : * calls are the only exception to the general rule about always advancing the
1367 : * required array keys (the scan may not even have a required array). These
1368 : * callers should just pass a NULL pstate (since there is never any question
1369 : * of stopping the scan). No call to _bt_tuple_before_array_skeys is required
1370 : * ahead of these calls (it's already clear that any required scan keys must
1371 : * be satisfied by caller's tuple).
1372 : *
1373 : * Note that we deal with non-array required equality strategy scan keys as
1374 : * degenerate single element arrays here. Obviously, they can never really
1375 : * advance in the way that real arrays can, but they must still affect how we
1376 : * advance real array scan keys (exactly like true array equality scan keys).
1377 : * We have to keep around a 3-way ORDER proc for these (using the "=" operator
1378 : * won't do), since in general whether the tuple is < or > _any_ unsatisfied
1379 : * required equality key influences how the scan's real arrays must advance.
1380 : *
1381 : * Note also that we may sometimes need to advance the array keys when the
1382 : * existing required array keys (and other required equality keys) are already
1383 : * an exact match for every corresponding value from caller's tuple. We must
1384 : * do this for inequalities that _bt_check_compare set continuescan=false for.
1385 : * They'll advance the array keys here, just like any other scan key that
1386 : * _bt_check_compare stops on. (This can even happen _after_ we advance the
1387 : * array keys, in which case we'll advance the array keys a second time. That
1388 : * way _bt_checkkeys caller always has its required arrays advance to the
1389 : * maximum possible extent that its tuple will allow.)
1390 : */
1391 : static bool
1392 200026 : _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
1393 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
1394 : int sktrig, bool sktrig_required)
1395 : {
1396 200026 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1397 200026 : Relation rel = scan->indexRelation;
1398 200026 : ScanDirection dir = so->currPos.dir;
1399 200026 : int arrayidx = 0;
1400 200026 : bool beyond_end_advance = false,
1401 200026 : skip_array_advanced = false,
1402 200026 : has_required_opposite_direction_only = false,
1403 200026 : all_required_satisfied = true,
1404 200026 : all_satisfied = true;
1405 :
1406 : Assert(!so->needPrimScan && !so->scanBehind && !so->oppositeDirCheck);
1407 : Assert(_bt_verify_keys_with_arraykeys(scan));
1408 :
1409 200026 : if (sktrig_required)
1410 : {
1411 : /*
1412 : * Precondition array state assertion
1413 : */
1414 : Assert(!_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc,
1415 : tupnatts, false, 0, NULL));
1416 :
1417 : /*
1418 : * Once we return we'll have a new set of required array keys, so
1419 : * reset state used by "look ahead" optimization
1420 : */
1421 191142 : pstate->rechecks = 0;
1422 191142 : pstate->targetdistance = 0;
1423 : }
1424 8884 : else if (sktrig < so->numberOfKeys - 1 &&
1425 8884 : !(so->keyData[so->numberOfKeys - 1].sk_flags & SK_SEARCHARRAY))
1426 : {
1427 8884 : int least_sign_ikey = so->numberOfKeys - 1;
1428 : bool continuescan;
1429 :
1430 : /*
1431 : * Optimization: perform a precheck of the least significant key
1432 : * during !sktrig_required calls when it isn't already our sktrig
1433 : * (provided the precheck key is not itself an array).
1434 : *
1435 : * When the precheck works out we'll avoid an expensive binary search
1436 : * of sktrig's array (plus any other arrays before least_sign_ikey).
1437 : */
1438 : Assert(so->keyData[sktrig].sk_flags & SK_SEARCHARRAY);
1439 8884 : if (!_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
1440 : false, &continuescan,
1441 : &least_sign_ikey))
1442 2582 : return false;
1443 : }
1444 :
1445 580056 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
1446 : {
1447 388378 : ScanKey cur = so->keyData + ikey;
1448 388378 : BTArrayKeyInfo *array = NULL;
1449 : Datum tupdatum;
1450 388378 : bool required = false,
1451 388378 : required_opposite_direction_only = false,
1452 : tupnull;
1453 : int32 result;
1454 388378 : int set_elem = 0;
1455 :
1456 388378 : if (cur->sk_strategy == BTEqualStrategyNumber)
1457 : {
1458 : /* Manage array state */
1459 340254 : if (cur->sk_flags & SK_SEARCHARRAY)
1460 : {
1461 207638 : 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 48124 : if (((ScanDirectionIsForward(dir) &&
1472 48124 : (cur->sk_flags & (SK_BT_REQBKWD))) ||
1473 0 : (ScanDirectionIsBackward(dir) &&
1474 0 : (cur->sk_flags & (SK_BT_REQFWD)))))
1475 15580 : has_required_opposite_direction_only =
1476 15580 : required_opposite_direction_only = true;
1477 : }
1478 :
1479 : /* Optimization: skip over known-satisfied scan keys */
1480 388378 : if (ikey < sktrig)
1481 76008 : continue;
1482 :
1483 372070 : if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
1484 : {
1485 372070 : required = true;
1486 :
1487 372070 : if (cur->sk_attno > tupnatts)
1488 : {
1489 : /* Set this just like _bt_tuple_before_array_skeys */
1490 : Assert(sktrig < ikey);
1491 2296 : so->scanBehind = true;
1492 : }
1493 : }
1494 :
1495 : /*
1496 : * Handle a required non-array scan key that the initial call to
1497 : * _bt_check_compare indicated triggered array advancement, if any.
1498 : *
1499 : * The non-array scan key's strategy will be <, <=, or = during a
1500 : * forwards scan (or any one of =, >=, or > during a backwards scan).
1501 : * It follows that the corresponding tuple attribute's value must now
1502 : * be either > or >= the scan key value (for backwards scans it must
1503 : * be either < or <= that value).
1504 : *
1505 : * If this is a required equality strategy scan key, this is just an
1506 : * optimization; _bt_tuple_before_array_skeys already confirmed that
1507 : * this scan key places us ahead of caller's tuple. There's no need
1508 : * to repeat that work now. (The same underlying principle also gets
1509 : * applied by the cur_elem_trig optimization used to speed up searches
1510 : * for the next array element.)
1511 : *
1512 : * If this is a required inequality strategy scan key, we _must_ rely
1513 : * on _bt_check_compare like this; we aren't capable of directly
1514 : * evaluating required inequality strategy scan keys here, on our own.
1515 : */
1516 372070 : if (ikey == sktrig && !array)
1517 : {
1518 : Assert(sktrig_required && required && all_required_satisfied);
1519 :
1520 : /* Use "beyond end" advancement. See below for an explanation. */
1521 7422 : beyond_end_advance = true;
1522 7422 : all_satisfied = all_required_satisfied = false;
1523 :
1524 7422 : continue;
1525 : }
1526 :
1527 : /*
1528 : * Nothing more for us to do with an inequality strategy scan key that
1529 : * wasn't the one that _bt_check_compare stopped on, though.
1530 : *
1531 : * Note: if our later call to _bt_check_compare (to recheck caller's
1532 : * tuple) sets continuescan=false due to finding this same inequality
1533 : * unsatisfied (possible when it's required in the scan direction),
1534 : * we'll deal with it via a recursive "second pass" call.
1535 : */
1536 364648 : else if (cur->sk_strategy != BTEqualStrategyNumber)
1537 47554 : continue;
1538 :
1539 : /*
1540 : * Nothing for us to do with an equality strategy scan key that isn't
1541 : * marked required, either -- unless it's a non-required array
1542 : */
1543 317094 : else if (!required && !array)
1544 0 : continue;
1545 :
1546 : /*
1547 : * Here we perform steps for all array scan keys after a required
1548 : * array scan key whose binary search triggered "beyond end of array
1549 : * element" array advancement due to encountering a tuple attribute
1550 : * value > the closest matching array key (or < for backwards scans).
1551 : */
1552 317094 : if (beyond_end_advance)
1553 : {
1554 1418 : if (array)
1555 590 : _bt_array_set_low_or_high(rel, cur, array,
1556 : ScanDirectionIsBackward(dir));
1557 :
1558 1418 : continue;
1559 : }
1560 :
1561 : /*
1562 : * Here we perform steps for all array scan keys after a required
1563 : * array scan key whose tuple attribute was < the closest matching
1564 : * array key when we dealt with it (or > for backwards scans).
1565 : *
1566 : * This earlier required array key already puts us ahead of caller's
1567 : * tuple in the key space (for the current scan direction). We must
1568 : * make sure that subsequent lower-order array keys do not put us too
1569 : * far ahead (ahead of tuples that have yet to be seen by our caller).
1570 : * For example, when a tuple "(a, b) = (42, 5)" advances the array
1571 : * keys on "a" from 40 to 45, we must also set "b" to whatever the
1572 : * first array element for "b" is. It would be wrong to allow "b" to
1573 : * be set based on the tuple value.
1574 : *
1575 : * Perform the same steps with truncated high key attributes. You can
1576 : * think of this as a "binary search" for the element closest to the
1577 : * value -inf. Again, the arrays must never get ahead of the scan.
1578 : */
1579 315676 : if (!all_required_satisfied || cur->sk_attno > tupnatts)
1580 : {
1581 3306 : if (array)
1582 756 : _bt_array_set_low_or_high(rel, cur, array,
1583 : ScanDirectionIsForward(dir));
1584 :
1585 3306 : continue;
1586 : }
1587 :
1588 : /*
1589 : * Search in scankey's array for the corresponding tuple attribute
1590 : * value from caller's tuple
1591 : */
1592 312370 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1593 :
1594 312370 : if (array)
1595 : {
1596 191762 : bool cur_elem_trig = (sktrig_required && ikey == sktrig);
1597 :
1598 : /*
1599 : * "Binary search" by checking if tupdatum/tupnull are within the
1600 : * range of the skip array
1601 : */
1602 191762 : if (array->num_elems == -1)
1603 160602 : _bt_binsrch_skiparray_skey(cur_elem_trig, dir,
1604 : tupdatum, tupnull, array, cur,
1605 : &result);
1606 :
1607 : /*
1608 : * Binary search for the closest match from the SAOP array
1609 : */
1610 : else
1611 31160 : set_elem = _bt_binsrch_array_skey(&so->orderProcs[ikey],
1612 : cur_elem_trig, dir,
1613 : tupdatum, tupnull, array, cur,
1614 : &result);
1615 : }
1616 : else
1617 : {
1618 : Assert(required);
1619 :
1620 : /*
1621 : * This is a required non-array equality strategy scan key, which
1622 : * we'll treat as a degenerate single element array.
1623 : *
1624 : * This scan key's imaginary "array" can't really advance, but it
1625 : * can still roll over like any other array. (Actually, this is
1626 : * no different to real single value arrays, which never advance
1627 : * without rolling over -- they can never truly advance, either.)
1628 : */
1629 120608 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1630 : tupdatum, tupnull,
1631 : cur->sk_argument, cur);
1632 : }
1633 :
1634 : /*
1635 : * Consider "beyond end of array element" array advancement.
1636 : *
1637 : * When the tuple attribute value is > the closest matching array key
1638 : * (or < in the backwards scan case), we need to ratchet this array
1639 : * forward (backward) by one increment, so that caller's tuple ends up
1640 : * being < final array value instead (or > final array value instead).
1641 : * This process has to work for all of the arrays, not just this one:
1642 : * it must "carry" to higher-order arrays when the set_elem that we
1643 : * just found happens to be the final one for the scan's direction.
1644 : * Incrementing (decrementing) set_elem itself isn't good enough.
1645 : *
1646 : * Our approach is to provisionally use set_elem as if it was an exact
1647 : * match now, then set each later/less significant array to whatever
1648 : * its final element is. Once outside the loop we'll then "increment
1649 : * this array's set_elem" by calling _bt_advance_array_keys_increment.
1650 : * That way the process rolls over to higher order arrays as needed.
1651 : *
1652 : * Under this scheme any required arrays only ever ratchet forwards
1653 : * (or backwards), and always do so to the maximum possible extent
1654 : * that we can know will be safe without seeing the scan's next tuple.
1655 : * We don't need any special handling for required scan keys that lack
1656 : * a real array to advance, nor for redundant scan keys that couldn't
1657 : * be eliminated by _bt_preprocess_keys. It won't matter if some of
1658 : * our "true" array scan keys (or even all of them) are non-required.
1659 : */
1660 312370 : if (sktrig_required && required &&
1661 306054 : ((ScanDirectionIsForward(dir) && result > 0) ||
1662 1716 : (ScanDirectionIsBackward(dir) && result < 0)))
1663 24648 : beyond_end_advance = true;
1664 :
1665 : Assert(all_required_satisfied && all_satisfied);
1666 312370 : if (result != 0)
1667 : {
1668 : /*
1669 : * Track whether caller's tuple satisfies our new post-advancement
1670 : * qual, for required scan keys, as well as for the entire set of
1671 : * interesting scan keys (all required scan keys plus non-required
1672 : * array scan keys are considered interesting.)
1673 : */
1674 142878 : all_satisfied = false;
1675 142878 : if (sktrig_required && required)
1676 137112 : all_required_satisfied = false;
1677 : else
1678 : {
1679 : /*
1680 : * There's no need to advance the arrays using the best
1681 : * available match for a non-required array. Give up now.
1682 : * (Though note that sktrig_required calls still have to do
1683 : * all the usual post-advancement steps, including the recheck
1684 : * call to _bt_check_compare.)
1685 : */
1686 : break;
1687 : }
1688 : }
1689 :
1690 : /* Advance array keys, even when we don't have an exact match */
1691 306604 : if (array)
1692 : {
1693 185996 : if (array->num_elems == -1)
1694 : {
1695 : /* Skip array's new element is tupdatum (or MINVAL/MAXVAL) */
1696 154836 : _bt_skiparray_set_element(rel, cur, array, result,
1697 : tupdatum, tupnull);
1698 154836 : skip_array_advanced = true;
1699 : }
1700 31160 : else if (array->cur_elem != set_elem)
1701 : {
1702 : /* SAOP array's new element is set_elem datum */
1703 23272 : array->cur_elem = set_elem;
1704 23272 : cur->sk_argument = array->elem_values[set_elem];
1705 : }
1706 : }
1707 : }
1708 :
1709 : /*
1710 : * Advance the array keys incrementally whenever "beyond end of array
1711 : * element" array advancement happens, so that advancement will carry to
1712 : * higher-order arrays (might exhaust all the scan's arrays instead, which
1713 : * ends the top-level scan).
1714 : */
1715 197444 : if (beyond_end_advance &&
1716 32070 : !_bt_advance_array_keys_increment(scan, dir, &skip_array_advanced))
1717 8040 : goto end_toplevel_scan;
1718 :
1719 : Assert(_bt_verify_keys_with_arraykeys(scan));
1720 :
1721 : /*
1722 : * Maintain a page-level count of the number of times the scan's array
1723 : * keys advanced in a way that affected at least one skip array
1724 : */
1725 189404 : if (sktrig_required && skip_array_advanced)
1726 160830 : pstate->nskipadvances++;
1727 :
1728 : /*
1729 : * Does tuple now satisfy our new qual? Recheck with _bt_check_compare.
1730 : *
1731 : * Calls triggered by an unsatisfied required scan key, whose tuple now
1732 : * satisfies all required scan keys, but not all nonrequired array keys,
1733 : * will still require a recheck call to _bt_check_compare. They'll still
1734 : * need its "second pass" handling of required inequality scan keys.
1735 : * (Might have missed a still-unsatisfied required inequality scan key
1736 : * that caller didn't detect as the sktrig scan key during its initial
1737 : * _bt_check_compare call that used the old/original qual.)
1738 : *
1739 : * Calls triggered by an unsatisfied nonrequired array scan key never need
1740 : * "second pass" handling of required inequalities (nor any other handling
1741 : * of any required scan key). All that matters is whether caller's tuple
1742 : * satisfies the new qual, so it's safe to just skip the _bt_check_compare
1743 : * recheck when we've already determined that it can only return 'false'.
1744 : *
1745 : * Note: In practice most scan keys are marked required by preprocessing,
1746 : * if necessary by generating a preceding skip array. We nevertheless
1747 : * often handle array keys marked required as if they were nonrequired.
1748 : * This behavior is requested by our _bt_check_compare caller, though only
1749 : * when it is passed "forcenonrequired=true" by _bt_checkkeys.
1750 : */
1751 189404 : if ((sktrig_required && all_required_satisfied) ||
1752 142796 : (!sktrig_required && all_satisfied))
1753 : {
1754 47144 : int nsktrig = sktrig + 1;
1755 : bool continuescan;
1756 :
1757 : Assert(all_required_satisfied);
1758 :
1759 : /* Recheck _bt_check_compare on behalf of caller */
1760 47144 : if (_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
1761 47144 : !sktrig_required, &continuescan,
1762 47144 : &nsktrig) &&
1763 39482 : !so->scanBehind)
1764 : {
1765 : /* This tuple satisfies the new qual */
1766 : Assert(all_satisfied && continuescan);
1767 :
1768 37284 : if (pstate)
1769 36748 : pstate->continuescan = true;
1770 :
1771 37506 : return true;
1772 : }
1773 :
1774 : /*
1775 : * Consider "second pass" handling of required inequalities.
1776 : *
1777 : * It's possible that our _bt_check_compare call indicated that the
1778 : * scan should end due to some unsatisfied inequality that wasn't
1779 : * initially recognized as such by us. Handle this by calling
1780 : * ourselves recursively, this time indicating that the trigger is the
1781 : * inequality that we missed first time around (and using a set of
1782 : * required array/equality keys that are now exact matches for tuple).
1783 : *
1784 : * We make a strong, general guarantee that every _bt_checkkeys call
1785 : * here will advance the array keys to the maximum possible extent
1786 : * that we can know to be safe based on caller's tuple alone. If we
1787 : * didn't perform this step, then that guarantee wouldn't quite hold.
1788 : */
1789 9860 : if (unlikely(!continuescan))
1790 : {
1791 : bool satisfied PG_USED_FOR_ASSERTS_ONLY;
1792 :
1793 : Assert(sktrig_required);
1794 : Assert(so->keyData[nsktrig].sk_strategy != BTEqualStrategyNumber);
1795 :
1796 : /*
1797 : * The tuple must use "beyond end" advancement during the
1798 : * recursive call, so we cannot possibly end up back here when
1799 : * recursing. We'll consume a small, fixed amount of stack space.
1800 : */
1801 : Assert(!beyond_end_advance);
1802 :
1803 : /* Advance the array keys a second time using same tuple */
1804 222 : satisfied = _bt_advance_array_keys(scan, pstate, tuple, tupnatts,
1805 : tupdesc, nsktrig, true);
1806 :
1807 : /* This tuple doesn't satisfy the inequality */
1808 : Assert(!satisfied);
1809 222 : return false;
1810 : }
1811 :
1812 : /*
1813 : * Some non-required scan key (from new qual) still not satisfied.
1814 : *
1815 : * All scan keys required in the current scan direction must still be
1816 : * satisfied, though, so we can trust all_required_satisfied below.
1817 : */
1818 : }
1819 :
1820 : /*
1821 : * When we were called just to deal with "advancing" non-required arrays,
1822 : * this is as far as we can go (cannot stop the scan for these callers)
1823 : */
1824 151898 : if (!sktrig_required)
1825 : {
1826 : /* Caller's tuple doesn't match any qual */
1827 5766 : return false;
1828 : }
1829 :
1830 : /*
1831 : * Postcondition array state assertion (for still-unsatisfied tuples).
1832 : *
1833 : * By here we have established that the scan's required arrays (scan must
1834 : * have at least one required array) advanced, without becoming exhausted.
1835 : *
1836 : * Caller's tuple is now < the newly advanced array keys (or > when this
1837 : * is a backwards scan), except in the case where we only got this far due
1838 : * to an unsatisfied non-required scan key. Verify that with an assert.
1839 : *
1840 : * Note: we don't just quit at this point when all required scan keys were
1841 : * found to be satisfied because we need to consider edge-cases involving
1842 : * scan keys required in the opposite direction only; those aren't tracked
1843 : * by all_required_satisfied.
1844 : */
1845 : Assert(_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts,
1846 : false, 0, NULL) ==
1847 : !all_required_satisfied);
1848 :
1849 : /*
1850 : * We generally permit primitive index scans to continue onto the next
1851 : * sibling page when the page's finaltup satisfies all required scan keys
1852 : * at the point where we're between pages.
1853 : *
1854 : * If caller's tuple is also the page's finaltup, and we see that required
1855 : * scan keys still aren't satisfied, start a new primitive index scan.
1856 : */
1857 146132 : if (!all_required_satisfied && pstate->finaltup == tuple)
1858 528 : goto new_prim_scan;
1859 :
1860 : /*
1861 : * Proactively check finaltup (don't wait until finaltup is reached by the
1862 : * scan) when it might well turn out to not be satisfied later on.
1863 : *
1864 : * Note: if so->scanBehind hasn't already been set for finaltup by us,
1865 : * it'll be set during this call to _bt_tuple_before_array_skeys. Either
1866 : * way, it'll be set correctly (for the whole page) after this point.
1867 : */
1868 226274 : if (!all_required_satisfied && pstate->finaltup &&
1869 161340 : _bt_tuple_before_array_skeys(scan, dir, pstate->finaltup, tupdesc,
1870 161340 : BTreeTupleGetNAtts(pstate->finaltup, rel),
1871 : false, 0, &so->scanBehind))
1872 17436 : goto new_prim_scan;
1873 :
1874 : /*
1875 : * When we encounter a truncated finaltup high key attribute, we're
1876 : * optimistic about the chances of its corresponding required scan key
1877 : * being satisfied when we go on to recheck it against tuples from this
1878 : * page's right sibling leaf page. We consider truncated attributes to be
1879 : * satisfied by required scan keys, which allows the primitive index scan
1880 : * to continue to the next leaf page. We must set so->scanBehind to true
1881 : * to remember that the last page's finaltup had "satisfied" required scan
1882 : * keys for one or more truncated attribute values (scan keys required in
1883 : * _either_ scan direction).
1884 : *
1885 : * There is a chance that _bt_readpage (which checks so->scanBehind) will
1886 : * find that even the sibling leaf page's finaltup is < the new array
1887 : * keys. When that happens, our optimistic policy will have incurred a
1888 : * single extra leaf page access that could have been avoided.
1889 : *
1890 : * A pessimistic policy would give backward scans a gratuitous advantage
1891 : * over forward scans. We'd punish forward scans for applying more
1892 : * accurate information from the high key, rather than just using the
1893 : * final non-pivot tuple as finaltup, in the style of backward scans.
1894 : * Being pessimistic would also give some scans with non-required arrays a
1895 : * perverse advantage over similar scans that use required arrays instead.
1896 : *
1897 : * This is similar to our scan-level heuristics, below. They also set
1898 : * scanBehind to speculatively continue the primscan onto the next page.
1899 : */
1900 128168 : if (so->scanBehind)
1901 : {
1902 : /* Truncated high key -- _bt_scanbehind_checkkeys recheck scheduled */
1903 : }
1904 :
1905 : /*
1906 : * Handle inequalities marked required in the opposite scan direction.
1907 : * They can also signal that we should start a new primitive index scan.
1908 : *
1909 : * It's possible that the scan is now positioned where "matching" tuples
1910 : * begin, and that caller's tuple satisfies all scan keys required in the
1911 : * current scan direction. But if caller's tuple still doesn't satisfy
1912 : * other scan keys that are required in the opposite scan direction only
1913 : * (e.g., a required >= strategy scan key when scan direction is forward),
1914 : * it's still possible that there are many leaf pages before the page that
1915 : * _bt_first could skip straight to. Groveling through all those pages
1916 : * will always give correct answers, but it can be very inefficient. We
1917 : * must avoid needlessly scanning extra pages.
1918 : *
1919 : * Separately, it's possible that _bt_check_compare set continuescan=false
1920 : * for a scan key that's required in the opposite direction only. This is
1921 : * a special case, that happens only when _bt_check_compare sees that the
1922 : * inequality encountered a NULL value. This signals the end of non-NULL
1923 : * values in the current scan direction, which is reason enough to end the
1924 : * (primitive) scan. If this happens at the start of a large group of
1925 : * NULL values, then we shouldn't expect to be called again until after
1926 : * the scan has already read indefinitely-many leaf pages full of tuples
1927 : * with NULL suffix values. (_bt_first is expected to skip over the group
1928 : * of NULLs by applying a similar "deduce NOT NULL" rule of its own, which
1929 : * involves consing up an explicit SK_SEARCHNOTNULL key.)
1930 : *
1931 : * Apply a test against finaltup to detect and recover from the problem:
1932 : * if even finaltup doesn't satisfy such an inequality, we just skip by
1933 : * starting a new primitive index scan. When we skip, we know for sure
1934 : * that all of the tuples on the current page following caller's tuple are
1935 : * also before the _bt_first-wise start of tuples for our new qual. That
1936 : * at least suggests many more skippable pages beyond the current page.
1937 : * (when so->scanBehind and so->oppositeDirCheck are set, this'll happen
1938 : * when we test the next page's finaltup/high key instead.)
1939 : */
1940 125938 : else if (has_required_opposite_direction_only && pstate->finaltup &&
1941 4292 : unlikely(!_bt_oppodir_checkkeys(scan, dir, pstate->finaltup)))
1942 2 : goto new_prim_scan;
1943 :
1944 125936 : continue_scan:
1945 :
1946 : /*
1947 : * Stick with the ongoing primitive index scan for now.
1948 : *
1949 : * It's possible that later tuples will also turn out to have values that
1950 : * are still < the now-current array keys (or > the current array keys).
1951 : * Our caller will handle this by performing what amounts to a linear
1952 : * search of the page, implemented by calling _bt_check_compare and then
1953 : * _bt_tuple_before_array_skeys for each tuple.
1954 : *
1955 : * This approach has various advantages over a binary search of the page.
1956 : * Repeated binary searches of the page (one binary search for every array
1957 : * advancement) won't outperform a continuous linear search. While there
1958 : * are workloads that a naive linear search won't handle well, our caller
1959 : * has a "look ahead" fallback mechanism to deal with that problem.
1960 : */
1961 129028 : pstate->continuescan = true; /* Override _bt_check_compare */
1962 129028 : so->needPrimScan = false; /* _bt_readpage has more tuples to check */
1963 :
1964 129028 : if (so->scanBehind)
1965 : {
1966 : /*
1967 : * Remember if recheck needs to call _bt_oppodir_checkkeys for next
1968 : * page's finaltup (see above comments about "Handle inequalities
1969 : * marked required in the opposite scan direction" for why).
1970 : */
1971 3092 : so->oppositeDirCheck = has_required_opposite_direction_only;
1972 :
1973 : /*
1974 : * skip by setting "look ahead" mechanism's offnum for forwards scans
1975 : * (backwards scans check scanBehind flag directly instead)
1976 : */
1977 3092 : if (ScanDirectionIsForward(dir))
1978 3074 : pstate->skip = pstate->maxoff + 1;
1979 : }
1980 :
1981 : /* Caller's tuple doesn't match the new qual */
1982 129028 : return false;
1983 :
1984 17966 : new_prim_scan:
1985 :
1986 : Assert(pstate->finaltup); /* not on rightmost/leftmost page */
1987 :
1988 : /*
1989 : * Looks like another primitive index scan is required. But consider
1990 : * continuing the current primscan based on scan-level heuristics.
1991 : *
1992 : * Continue the ongoing primitive scan (and schedule a recheck for when
1993 : * the scan arrives on the next sibling leaf page) when it has already
1994 : * read at least one leaf page before the one we're reading now. This
1995 : * makes primscan scheduling more efficient when scanning subsets of an
1996 : * index with many distinct attribute values matching many array elements.
1997 : * It encourages fewer, larger primitive scans where that makes sense.
1998 : * This will in turn encourage _bt_readpage to apply the pstate.startikey
1999 : * optimization more often.
2000 : *
2001 : * Also continue the ongoing primitive index scan when it is still on the
2002 : * first page if there have been more than NSKIPADVANCES_THRESHOLD calls
2003 : * here that each advanced at least one of the scan's skip arrays
2004 : * (deliberately ignore advancements that only affected SAOP arrays here).
2005 : * A page that cycles through this many skip array elements is quite
2006 : * likely to neighbor similar pages, that we'll also need to read.
2007 : *
2008 : * Note: These heuristics aren't as aggressive as you might think. We're
2009 : * conservative about allowing a primitive scan to step from the first
2010 : * leaf page it reads to the page's sibling page (we only allow it on
2011 : * first pages whose finaltup strongly suggests that it'll work out, as
2012 : * well as first pages that have a large number of skip array advances).
2013 : * Clearing this first page finaltup hurdle is a strong signal in itself.
2014 : *
2015 : * Note: The NSKIPADVANCES_THRESHOLD heuristic exists only to avoid
2016 : * pathological cases. Specifically, cases where a skip scan should just
2017 : * behave like a traditional full index scan, but ends up "skipping" again
2018 : * and again, descending to the prior leaf page's direct sibling leaf page
2019 : * each time. This misbehavior would otherwise be possible during scans
2020 : * that never quite manage to "clear the first page finaltup hurdle".
2021 : */
2022 17966 : if (!pstate->firstpage || pstate->nskipadvances > NSKIPADVANCES_THRESHOLD)
2023 : {
2024 : /* Schedule a recheck once on the next (or previous) page */
2025 862 : so->scanBehind = true;
2026 :
2027 : /* Continue the current primitive scan after all */
2028 862 : goto continue_scan;
2029 : }
2030 :
2031 : /*
2032 : * End this primitive index scan, but schedule another.
2033 : *
2034 : * Note: We make a soft assumption that the current scan direction will
2035 : * also be used within _bt_next, when it is asked to step off this page.
2036 : * It is up to _bt_next to cancel this scheduled primitive index scan
2037 : * whenever it steps to a page in the direction opposite currPos.dir.
2038 : */
2039 17104 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2040 17104 : so->needPrimScan = true; /* ...but call _bt_first again */
2041 :
2042 17104 : if (scan->parallel_scan)
2043 36 : _bt_parallel_primscan_schedule(scan, so->currPos.currPage);
2044 :
2045 : /* Caller's tuple doesn't match the new qual */
2046 17104 : return false;
2047 :
2048 8040 : end_toplevel_scan:
2049 :
2050 : /*
2051 : * End the current primitive index scan, but don't schedule another.
2052 : *
2053 : * This ends the entire top-level scan in the current scan direction.
2054 : *
2055 : * Note: The scan's arrays (including any non-required arrays) are now in
2056 : * their final positions for the current scan direction. If the scan
2057 : * direction happens to change, then the arrays will already be in their
2058 : * first positions for what will then be the current scan direction.
2059 : */
2060 8040 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2061 8040 : so->needPrimScan = false; /* ...and don't call _bt_first again */
2062 :
2063 : /* Caller's tuple doesn't match any qual */
2064 8040 : return false;
2065 : }
2066 :
2067 : #ifdef USE_ASSERT_CHECKING
2068 : /*
2069 : * Verify that the scan's "so->keyData[]" scan keys are in agreement with
2070 : * its array key state
2071 : */
2072 : static bool
2073 : _bt_verify_keys_with_arraykeys(IndexScanDesc scan)
2074 : {
2075 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2076 : int last_sk_attno = InvalidAttrNumber,
2077 : arrayidx = 0;
2078 : bool nonrequiredseen = false;
2079 :
2080 : if (!so->qual_ok)
2081 : return false;
2082 :
2083 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
2084 : {
2085 : ScanKey cur = so->keyData + ikey;
2086 : BTArrayKeyInfo *array;
2087 :
2088 : if (cur->sk_strategy != BTEqualStrategyNumber ||
2089 : !(cur->sk_flags & SK_SEARCHARRAY))
2090 : continue;
2091 :
2092 : array = &so->arrayKeys[arrayidx++];
2093 : if (array->scan_key != ikey)
2094 : return false;
2095 :
2096 : if (array->num_elems == 0 || array->num_elems < -1)
2097 : return false;
2098 :
2099 : if (array->num_elems != -1 &&
2100 : cur->sk_argument != array->elem_values[array->cur_elem])
2101 : return false;
2102 : if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
2103 : {
2104 : if (last_sk_attno > cur->sk_attno)
2105 : return false;
2106 : if (nonrequiredseen)
2107 : return false;
2108 : }
2109 : else
2110 : nonrequiredseen = true;
2111 :
2112 : last_sk_attno = cur->sk_attno;
2113 : }
2114 :
2115 : if (arrayidx != so->numArrayKeys)
2116 : return false;
2117 :
2118 : return true;
2119 : }
2120 : #endif
2121 :
2122 : /*
2123 : * Test whether an indextuple satisfies all the scankey conditions.
2124 : *
2125 : * Return true if so, false if not. If the tuple fails to pass the qual,
2126 : * we also determine whether there's any need to continue the scan beyond
2127 : * this tuple, and set pstate.continuescan accordingly. See comments for
2128 : * _bt_preprocess_keys() about how this is done.
2129 : *
2130 : * Forward scan callers can pass a high key tuple in the hopes of having
2131 : * us set *continuescan to false, and avoiding an unnecessary visit to
2132 : * the page to the right.
2133 : *
2134 : * Advances the scan's array keys when necessary for arrayKeys=true callers.
2135 : * Scans without any array keys must always pass arrayKeys=false.
2136 : *
2137 : * Also stops and starts primitive index scans for arrayKeys=true callers.
2138 : * Scans with array keys are required to set up page state that helps us with
2139 : * this. The page's finaltup tuple (the page high key for a forward scan, or
2140 : * the page's first non-pivot tuple for a backward scan) must be set in
2141 : * pstate.finaltup ahead of the first call here for the page. Set this to
2142 : * NULL for rightmost page (or the leftmost page for backwards scans).
2143 : *
2144 : * scan: index scan descriptor (containing a search-type scankey)
2145 : * pstate: page level input and output parameters
2146 : * arrayKeys: should we advance the scan's array keys if necessary?
2147 : * tuple: index tuple to test
2148 : * tupnatts: number of attributes in tupnatts (high key may be truncated)
2149 : */
2150 : bool
2151 61379848 : _bt_checkkeys(IndexScanDesc scan, BTReadPageState *pstate, bool arrayKeys,
2152 : IndexTuple tuple, int tupnatts)
2153 : {
2154 61379848 : TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
2155 61379848 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2156 61379848 : ScanDirection dir = so->currPos.dir;
2157 61379848 : int ikey = pstate->startikey;
2158 : bool res;
2159 :
2160 : Assert(BTreeTupleGetNAtts(tuple, scan->indexRelation) == tupnatts);
2161 : Assert(!so->needPrimScan && !so->scanBehind && !so->oppositeDirCheck);
2162 : Assert(arrayKeys || so->numArrayKeys == 0);
2163 :
2164 61379848 : res = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, arrayKeys,
2165 61379848 : pstate->forcenonrequired, &pstate->continuescan,
2166 : &ikey);
2167 :
2168 : /*
2169 : * If _bt_check_compare relied on the pstate.startikey optimization, call
2170 : * again (in assert-enabled builds) to verify it didn't affect our answer.
2171 : *
2172 : * Note: we can't do this when !pstate.forcenonrequired, since any arrays
2173 : * before pstate.startikey won't have advanced on this page at all.
2174 : */
2175 : Assert(!pstate->forcenonrequired || arrayKeys);
2176 : #ifdef USE_ASSERT_CHECKING
2177 : if (pstate->startikey > 0 && !pstate->forcenonrequired)
2178 : {
2179 : bool dres,
2180 : dcontinuescan;
2181 : int dikey = 0;
2182 :
2183 : /* Pass arrayKeys=false to avoid array side-effects */
2184 : dres = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
2185 : pstate->forcenonrequired, &dcontinuescan,
2186 : &dikey);
2187 : Assert(res == dres);
2188 : Assert(pstate->continuescan == dcontinuescan);
2189 :
2190 : /*
2191 : * Should also get the same ikey result. We need a slightly weaker
2192 : * assertion during arrayKeys calls, since they might be using an
2193 : * array that couldn't be marked required during preprocessing.
2194 : */
2195 : Assert(arrayKeys || ikey == dikey);
2196 : Assert(ikey <= dikey);
2197 : }
2198 : #endif
2199 :
2200 : /*
2201 : * Only one _bt_check_compare call is required in the common case where
2202 : * there are no equality strategy array scan keys. Otherwise we can only
2203 : * accept _bt_check_compare's answer unreservedly when it didn't set
2204 : * pstate.continuescan=false.
2205 : */
2206 61379848 : if (!arrayKeys || pstate->continuescan)
2207 61150420 : return res;
2208 :
2209 : /*
2210 : * _bt_check_compare call set continuescan=false in the presence of
2211 : * equality type array keys. This could mean that the tuple is just past
2212 : * the end of matches for the current array keys.
2213 : *
2214 : * It's also possible that the scan is still _before_ the _start_ of
2215 : * tuples matching the current set of array keys. Check for that first.
2216 : */
2217 : Assert(!pstate->forcenonrequired);
2218 229428 : if (_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts, true,
2219 : ikey, NULL))
2220 : {
2221 : /* Override _bt_check_compare, continue primitive scan */
2222 38508 : pstate->continuescan = true;
2223 :
2224 : /*
2225 : * We will end up here repeatedly given a group of tuples > the
2226 : * previous array keys and < the now-current keys (for a backwards
2227 : * scan it's just the same, though the operators swap positions).
2228 : *
2229 : * We must avoid allowing this linear search process to scan very many
2230 : * tuples from well before the start of tuples matching the current
2231 : * array keys (or from well before the point where we'll once again
2232 : * have to advance the scan's array keys).
2233 : *
2234 : * We keep the overhead under control by speculatively "looking ahead"
2235 : * to later still-unscanned items from this same leaf page. We'll
2236 : * only attempt this once the number of tuples that the linear search
2237 : * process has examined starts to get out of hand.
2238 : */
2239 38508 : pstate->rechecks++;
2240 38508 : if (pstate->rechecks >= LOOK_AHEAD_REQUIRED_RECHECKS)
2241 : {
2242 : /* See if we should skip ahead within the current leaf page */
2243 10624 : _bt_checkkeys_look_ahead(scan, pstate, tupnatts, tupdesc);
2244 :
2245 : /*
2246 : * Might have set pstate.skip to a later page offset. When that
2247 : * happens then _bt_readpage caller will inexpensively skip ahead
2248 : * to a later tuple from the same page (the one just after the
2249 : * tuple we successfully "looked ahead" to).
2250 : */
2251 : }
2252 :
2253 : /* This indextuple doesn't match the current qual, in any case */
2254 38508 : return false;
2255 : }
2256 :
2257 : /*
2258 : * Caller's tuple is >= the current set of array keys and other equality
2259 : * constraint scan keys (or <= if this is a backwards scan). It's now
2260 : * clear that we _must_ advance any required array keys in lockstep with
2261 : * the scan.
2262 : */
2263 190920 : return _bt_advance_array_keys(scan, pstate, tuple, tupnatts, tupdesc,
2264 : ikey, true);
2265 : }
2266 :
2267 : /*
2268 : * Test whether caller's finaltup tuple is still before the start of matches
2269 : * for the current array keys.
2270 : *
2271 : * Called at the start of reading a page during a scan with array keys, though
2272 : * only when the so->scanBehind flag was set on the scan's prior page.
2273 : *
2274 : * Returns false if the tuple is still before the start of matches. When that
2275 : * happens, caller should cut its losses and start a new primitive index scan.
2276 : * Otherwise returns true.
2277 : */
2278 : bool
2279 2580 : _bt_scanbehind_checkkeys(IndexScanDesc scan, ScanDirection dir,
2280 : IndexTuple finaltup)
2281 : {
2282 2580 : Relation rel = scan->indexRelation;
2283 2580 : TupleDesc tupdesc = RelationGetDescr(rel);
2284 2580 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2285 2580 : int nfinaltupatts = BTreeTupleGetNAtts(finaltup, rel);
2286 : bool scanBehind;
2287 :
2288 : Assert(so->numArrayKeys);
2289 :
2290 2580 : if (_bt_tuple_before_array_skeys(scan, dir, finaltup, tupdesc,
2291 : nfinaltupatts, false, 0, &scanBehind))
2292 408 : return false;
2293 :
2294 : /*
2295 : * If scanBehind was set, all of the untruncated attribute values from
2296 : * finaltup that correspond to an array match the array's current element,
2297 : * but there are other keys associated with truncated suffix attributes.
2298 : * Array advancement must have incremented the scan's arrays on the
2299 : * previous page, resulting in a set of array keys that happen to be an
2300 : * exact match for the current page high key's untruncated prefix values.
2301 : *
2302 : * This page definitely doesn't contain tuples that the scan will need to
2303 : * return. The next page may or may not contain relevant tuples. Handle
2304 : * this by cutting our losses and starting a new primscan.
2305 : */
2306 2172 : if (scanBehind)
2307 0 : return false;
2308 :
2309 2172 : if (!so->oppositeDirCheck)
2310 2044 : return true;
2311 :
2312 128 : return _bt_oppodir_checkkeys(scan, dir, finaltup);
2313 : }
2314 :
2315 : /*
2316 : * Test whether an indextuple fails to satisfy an inequality required in the
2317 : * opposite direction only.
2318 : *
2319 : * Caller's finaltup tuple is the page high key (for forwards scans), or the
2320 : * first non-pivot tuple (for backwards scans). Called during scans with
2321 : * required array keys and required opposite-direction inequalities.
2322 : *
2323 : * Returns false if an inequality scan key required in the opposite direction
2324 : * only isn't satisfied (and any earlier required scan keys are satisfied).
2325 : * Otherwise returns true.
2326 : *
2327 : * An unsatisfied inequality required in the opposite direction only might
2328 : * well enable skipping over many leaf pages, provided another _bt_first call
2329 : * takes place. This type of unsatisfied inequality won't usually cause
2330 : * _bt_checkkeys to stop the scan to consider array advancement/starting a new
2331 : * primitive index scan.
2332 : */
2333 : static bool
2334 4420 : _bt_oppodir_checkkeys(IndexScanDesc scan, ScanDirection dir,
2335 : IndexTuple finaltup)
2336 : {
2337 4420 : Relation rel = scan->indexRelation;
2338 4420 : TupleDesc tupdesc = RelationGetDescr(rel);
2339 4420 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2340 4420 : int nfinaltupatts = BTreeTupleGetNAtts(finaltup, rel);
2341 : bool continuescan;
2342 4420 : ScanDirection flipped = -dir;
2343 4420 : int ikey = 0;
2344 :
2345 : Assert(so->numArrayKeys);
2346 :
2347 4420 : _bt_check_compare(scan, flipped, finaltup, nfinaltupatts, tupdesc, false,
2348 : false, &continuescan,
2349 : &ikey);
2350 :
2351 4420 : if (!continuescan && so->keyData[ikey].sk_strategy != BTEqualStrategyNumber)
2352 2 : return false;
2353 :
2354 4418 : return true;
2355 : }
2356 :
2357 : /*
2358 : * Determines an offset to the first scan key (an so->keyData[]-wise offset)
2359 : * that is _not_ guaranteed to be satisfied by every tuple from pstate.page,
2360 : * which is set in pstate.startikey for _bt_checkkeys calls for the page.
2361 : * This allows caller to save cycles on comparisons of a prefix of keys while
2362 : * reading pstate.page.
2363 : *
2364 : * Also determines if later calls to _bt_checkkeys (for pstate.page) should be
2365 : * forced to treat all required scan keys >= pstate.startikey as nonrequired
2366 : * (that is, if they're to be treated as if any SK_BT_REQFWD/SK_BT_REQBKWD
2367 : * markings that were set by preprocessing were not set at all, for the
2368 : * duration of _bt_checkkeys calls prior to the call for pstate.finaltup).
2369 : * This is indicated to caller by setting pstate.forcenonrequired.
2370 : *
2371 : * Call here at the start of reading a leaf page beyond the first one for the
2372 : * primitive index scan. We consider all non-pivot tuples, so it doesn't make
2373 : * sense to call here when only a subset of those tuples can ever be read.
2374 : * This is also a good idea on performance grounds; not calling here when on
2375 : * the first page (first for the current primitive scan) avoids wasting cycles
2376 : * during selective point queries. They typically don't stand to gain as much
2377 : * when we can set pstate.startikey, and are likely to notice the overhead of
2378 : * calling here. (Also, allowing pstate.forcenonrequired to be set on a
2379 : * primscan's first page would mislead _bt_advance_array_keys, which expects
2380 : * pstate.nskipadvances to be representative of every first page's key space.)
2381 : *
2382 : * Caller must call _bt_start_array_keys and reset startikey/forcenonrequired
2383 : * ahead of the finaltup _bt_checkkeys call when we set forcenonrequired=true.
2384 : * This will give _bt_checkkeys the opportunity to call _bt_advance_array_keys
2385 : * with sktrig_required=true, restoring the invariant that the scan's required
2386 : * arrays always track the scan's progress through the index's key space.
2387 : * Caller won't need to do this on the rightmost/leftmost page in the index
2388 : * (where pstate.finaltup isn't ever set), since forcenonrequired will never
2389 : * be set here in the first place.
2390 : */
2391 : void
2392 30184 : _bt_set_startikey(IndexScanDesc scan, BTReadPageState *pstate)
2393 : {
2394 30184 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2395 30184 : Relation rel = scan->indexRelation;
2396 30184 : TupleDesc tupdesc = RelationGetDescr(rel);
2397 : ItemId iid;
2398 : IndexTuple firsttup,
2399 : lasttup;
2400 30184 : int startikey = 0,
2401 30184 : arrayidx = 0,
2402 : firstchangingattnum;
2403 30184 : bool start_past_saop_eq = false;
2404 :
2405 : Assert(!so->scanBehind);
2406 : Assert(pstate->minoff < pstate->maxoff);
2407 : Assert(!pstate->firstpage);
2408 : Assert(pstate->startikey == 0);
2409 : Assert(!so->numArrayKeys || pstate->finaltup ||
2410 : P_RIGHTMOST(BTPageGetOpaque(pstate->page)) ||
2411 : P_LEFTMOST(BTPageGetOpaque(pstate->page)));
2412 :
2413 30184 : if (so->numberOfKeys == 0)
2414 12440 : return;
2415 :
2416 : /* minoff is an offset to the lowest non-pivot tuple on the page */
2417 17744 : iid = PageGetItemId(pstate->page, pstate->minoff);
2418 17744 : firsttup = (IndexTuple) PageGetItem(pstate->page, iid);
2419 :
2420 : /* maxoff is an offset to the highest non-pivot tuple on the page */
2421 17744 : iid = PageGetItemId(pstate->page, pstate->maxoff);
2422 17744 : lasttup = (IndexTuple) PageGetItem(pstate->page, iid);
2423 :
2424 : /* Determine the first attribute whose values change on caller's page */
2425 17744 : firstchangingattnum = _bt_keep_natts_fast(rel, firsttup, lasttup);
2426 :
2427 26808 : for (; startikey < so->numberOfKeys; startikey++)
2428 : {
2429 20572 : ScanKey key = so->keyData + startikey;
2430 : BTArrayKeyInfo *array;
2431 : Datum firstdatum,
2432 : lastdatum;
2433 : bool firstnull,
2434 : lastnull;
2435 : int32 result;
2436 :
2437 : /*
2438 : * Determine if it's safe to set pstate.startikey to an offset to a
2439 : * key that comes after this key, by examining this key
2440 : */
2441 20572 : if (!(key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
2442 : {
2443 : /* Scan key isn't marked required (corner case) */
2444 11508 : break; /* unsafe */
2445 : }
2446 20572 : if (key->sk_flags & SK_ROW_HEADER)
2447 : {
2448 : /* RowCompare inequalities currently aren't supported */
2449 0 : break; /* "unsafe" */
2450 : }
2451 20572 : if (key->sk_strategy != BTEqualStrategyNumber)
2452 : {
2453 : /*
2454 : * Scalar inequality key.
2455 : *
2456 : * It's definitely safe for _bt_checkkeys to avoid assessing this
2457 : * inequality when the page's first and last non-pivot tuples both
2458 : * satisfy the inequality (since the same must also be true of all
2459 : * the tuples in between these two).
2460 : *
2461 : * Unlike the "=" case, it doesn't matter if this attribute has
2462 : * more than one distinct value (though it _is_ necessary for any
2463 : * and all _prior_ attributes to contain no more than one distinct
2464 : * value amongst all of the tuples from pstate.page).
2465 : */
2466 4600 : if (key->sk_attno > firstchangingattnum) /* >, not >= */
2467 362 : break; /* unsafe, preceding attr has multiple
2468 : * distinct values */
2469 :
2470 4238 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc, &firstnull);
2471 4238 : lastdatum = index_getattr(lasttup, key->sk_attno, tupdesc, &lastnull);
2472 :
2473 4238 : if (key->sk_flags & SK_ISNULL)
2474 : {
2475 : /* IS NOT NULL key */
2476 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
2477 :
2478 44 : if (firstnull || lastnull)
2479 : break; /* unsafe */
2480 :
2481 : /* Safe, IS NOT NULL key satisfied by every tuple */
2482 8818 : continue;
2483 : }
2484 :
2485 : /* Test firsttup */
2486 4194 : if (firstnull ||
2487 4194 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2488 : key->sk_collation, firstdatum,
2489 : key->sk_argument)))
2490 : break; /* unsafe */
2491 :
2492 : /* Test lasttup */
2493 4194 : if (lastnull ||
2494 4194 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2495 : key->sk_collation, lastdatum,
2496 : key->sk_argument)))
2497 : break; /* unsafe */
2498 :
2499 : /* Safe, scalar inequality satisfied by every tuple */
2500 4086 : continue;
2501 : }
2502 :
2503 : /* Some = key (could be a scalar = key, could be an array = key) */
2504 : Assert(key->sk_strategy == BTEqualStrategyNumber);
2505 :
2506 15972 : if (!(key->sk_flags & SK_SEARCHARRAY))
2507 : {
2508 : /*
2509 : * Scalar = key (possibly an IS NULL key).
2510 : *
2511 : * It is unsafe to set pstate.startikey to an ikey beyond this
2512 : * key, unless the = key is satisfied by every possible tuple on
2513 : * the page (possible only when attribute has just one distinct
2514 : * value among all tuples on the page).
2515 : */
2516 12722 : if (key->sk_attno >= firstchangingattnum)
2517 10396 : break; /* unsafe, multiple distinct attr values */
2518 :
2519 2326 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc,
2520 : &firstnull);
2521 2326 : if (key->sk_flags & SK_ISNULL)
2522 : {
2523 : /* IS NULL key */
2524 : Assert(key->sk_flags & SK_SEARCHNULL);
2525 :
2526 0 : if (!firstnull)
2527 0 : break; /* unsafe */
2528 :
2529 : /* Safe, IS NULL key satisfied by every tuple */
2530 0 : continue;
2531 : }
2532 2326 : if (firstnull ||
2533 2326 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2534 : key->sk_collation, firstdatum,
2535 : key->sk_argument)))
2536 : break; /* unsafe */
2537 :
2538 : /* Safe, scalar = key satisfied by every tuple */
2539 2326 : continue;
2540 : }
2541 :
2542 : /* = array key (could be a SAOP array, could be a skip array) */
2543 3250 : array = &so->arrayKeys[arrayidx++];
2544 : Assert(array->scan_key == startikey);
2545 3250 : if (array->num_elems != -1)
2546 : {
2547 : /*
2548 : * SAOP array = key.
2549 : *
2550 : * Handle this like we handle scalar = keys (though binary search
2551 : * for a matching element, to avoid relying on key's sk_argument).
2552 : */
2553 580 : if (key->sk_attno >= firstchangingattnum)
2554 580 : break; /* unsafe, multiple distinct attr values */
2555 :
2556 0 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc,
2557 : &firstnull);
2558 0 : _bt_binsrch_array_skey(&so->orderProcs[startikey],
2559 : false, NoMovementScanDirection,
2560 : firstdatum, firstnull, array, key,
2561 : &result);
2562 0 : if (result != 0)
2563 0 : break; /* unsafe */
2564 :
2565 : /* Safe, SAOP = key satisfied by every tuple */
2566 0 : start_past_saop_eq = true;
2567 0 : continue;
2568 : }
2569 :
2570 : /*
2571 : * Skip array = key
2572 : */
2573 : Assert(key->sk_flags & SK_BT_SKIP);
2574 2670 : if (array->null_elem)
2575 : {
2576 : /*
2577 : * Non-range skip array = key.
2578 : *
2579 : * Safe, non-range skip array "satisfied" by every tuple on page
2580 : * (safe even when "key->sk_attno > firstchangingattnum").
2581 : */
2582 2362 : continue;
2583 : }
2584 :
2585 : /*
2586 : * Range skip array = key.
2587 : *
2588 : * Handle this like we handle scalar inequality keys (but avoid using
2589 : * key's sk_argument directly, as in the SAOP array case).
2590 : */
2591 308 : if (key->sk_attno > firstchangingattnum) /* >, not >= */
2592 48 : break; /* unsafe, preceding attr has multiple
2593 : * distinct values */
2594 :
2595 260 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc, &firstnull);
2596 260 : lastdatum = index_getattr(lasttup, key->sk_attno, tupdesc, &lastnull);
2597 :
2598 : /* Test firsttup */
2599 260 : _bt_binsrch_skiparray_skey(false, ForwardScanDirection,
2600 : firstdatum, firstnull, array, key,
2601 : &result);
2602 260 : if (result != 0)
2603 0 : break; /* unsafe */
2604 :
2605 : /* Test lasttup */
2606 260 : _bt_binsrch_skiparray_skey(false, ForwardScanDirection,
2607 : lastdatum, lastnull, array, key,
2608 : &result);
2609 260 : if (result != 0)
2610 14 : break; /* unsafe */
2611 :
2612 : /* Safe, range skip array satisfied by every tuple on page */
2613 : }
2614 :
2615 : /*
2616 : * Use of forcenonrequired is typically undesirable, since it'll force
2617 : * _bt_readpage caller to read every tuple on the page -- even though, in
2618 : * general, it might well be possible to end the scan on an earlier tuple.
2619 : * However, caller must use forcenonrequired when start_past_saop_eq=true,
2620 : * since the usual required array behavior might fail to roll over to the
2621 : * SAOP array.
2622 : *
2623 : * We always prefer forcenonrequired=true during scans with skip arrays
2624 : * (except on the first page of each primitive index scan), though -- even
2625 : * when "startikey == 0". That way, _bt_advance_array_keys's low-order
2626 : * key precheck optimization can always be used (unless on the first page
2627 : * of the scan). It seems slightly preferable to check more tuples when
2628 : * that allows us to do significantly less skip array maintenance.
2629 : */
2630 17744 : pstate->forcenonrequired = (start_past_saop_eq || so->skipScan);
2631 17744 : pstate->startikey = startikey;
2632 :
2633 : /*
2634 : * _bt_readpage caller is required to call _bt_checkkeys against page's
2635 : * finaltup with forcenonrequired=false whenever we initially set
2636 : * forcenonrequired=true. That way the scan's arrays will reliably track
2637 : * its progress through the index's key space.
2638 : *
2639 : * We don't expect this when _bt_readpage caller has no finaltup due to
2640 : * its page being the rightmost (or the leftmost, during backwards scans).
2641 : * When we see that _bt_readpage has no finaltup, back out of everything.
2642 : */
2643 : Assert(!pstate->forcenonrequired || so->numArrayKeys);
2644 17744 : if (pstate->forcenonrequired && !pstate->finaltup)
2645 : {
2646 470 : pstate->forcenonrequired = false;
2647 470 : pstate->startikey = 0;
2648 : }
2649 : }
2650 :
2651 : /*
2652 : * Test whether an indextuple satisfies current scan condition.
2653 : *
2654 : * Return true if so, false if not. If not, also sets *continuescan to false
2655 : * when it's also not possible for any later tuples to pass the current qual
2656 : * (with the scan's current set of array keys, in the current scan direction),
2657 : * in addition to setting *ikey to the so->keyData[] subscript/offset for the
2658 : * unsatisfied scan key (needed when caller must consider advancing the scan's
2659 : * array keys).
2660 : *
2661 : * This is a subroutine for _bt_checkkeys. We provisionally assume that
2662 : * reaching the end of the current set of required keys (in particular the
2663 : * current required array keys) ends the ongoing (primitive) index scan.
2664 : * Callers without array keys should just end the scan right away when they
2665 : * find that continuescan has been set to false here by us. Things are more
2666 : * complicated for callers with array keys.
2667 : *
2668 : * Callers with array keys must first consider advancing the arrays when
2669 : * continuescan has been set to false here by us. They must then consider if
2670 : * it really does make sense to end the current (primitive) index scan, in
2671 : * light of everything that is known at that point. (In general when we set
2672 : * continuescan=false for these callers it must be treated as provisional.)
2673 : *
2674 : * We deal with advancing unsatisfied non-required arrays directly, though.
2675 : * This is safe, since by definition non-required keys can't end the scan.
2676 : * This is just how we determine if non-required arrays are just unsatisfied
2677 : * by the current array key, or if they're truly unsatisfied (that is, if
2678 : * they're unsatisfied by every possible array key).
2679 : *
2680 : * Pass advancenonrequired=false to avoid all array related side effects.
2681 : * This allows _bt_advance_array_keys caller to avoid infinite recursion.
2682 : *
2683 : * Pass forcenonrequired=true to instruct us to treat all keys as nonrequired.
2684 : * This is used to make it safe to temporarily stop properly maintaining the
2685 : * scan's required arrays. _bt_checkkeys caller (_bt_readpage, actually)
2686 : * determines a prefix of keys that must satisfy every possible corresponding
2687 : * index attribute value from its page, which is passed to us via *ikey arg
2688 : * (this is the first key that might be unsatisfied by tuples on the page).
2689 : * Obviously, we won't maintain any array keys from before *ikey, so it's
2690 : * quite possible for such arrays to "fall behind" the index's keyspace.
2691 : * Caller will need to "catch up" by passing forcenonrequired=true (alongside
2692 : * an *ikey=0) once the page's finaltup is reached.
2693 : *
2694 : * Note: it's safe to pass an *ikey > 0 with forcenonrequired=false, but only
2695 : * when caller determines that it won't affect array maintenance.
2696 : */
2697 : static bool
2698 61440296 : _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
2699 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
2700 : bool advancenonrequired, bool forcenonrequired,
2701 : bool *continuescan, int *ikey)
2702 : {
2703 61440296 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2704 :
2705 61440296 : *continuescan = true; /* default assumption */
2706 :
2707 118003814 : for (; *ikey < so->numberOfKeys; (*ikey)++)
2708 : {
2709 69635488 : ScanKey key = so->keyData + *ikey;
2710 : Datum datum;
2711 : bool isNull;
2712 69635488 : bool requiredSameDir = false,
2713 69635488 : requiredOppositeDirOnly = false;
2714 :
2715 : /*
2716 : * Check if the key is required in the current scan direction, in the
2717 : * opposite scan direction _only_, or in neither direction (except
2718 : * when we're forced to treat all scan keys as nonrequired)
2719 : */
2720 69635488 : if (forcenonrequired)
2721 : {
2722 : /* treating scan's keys as non-required */
2723 : }
2724 69232466 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
2725 14817706 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
2726 54439466 : requiredSameDir = true;
2727 14793000 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsBackward(dir)) ||
2728 5962264 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsForward(dir)))
2729 14793000 : requiredOppositeDirOnly = true;
2730 :
2731 69635488 : if (key->sk_attno > tupnatts)
2732 : {
2733 : /*
2734 : * This attribute is truncated (must be high key). The value for
2735 : * this attribute in the first non-pivot tuple on the page to the
2736 : * right could be any possible value. Assume that truncated
2737 : * attribute passes the qual.
2738 : */
2739 : Assert(BTreeTupleIsPivot(tuple));
2740 19064280 : continue;
2741 : }
2742 :
2743 : /*
2744 : * A skip array scan key uses one of several sentinel values. We just
2745 : * fall back on _bt_tuple_before_array_skeys when we see such a value.
2746 : */
2747 69633194 : if (key->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL |
2748 : SK_BT_NEXT | SK_BT_PRIOR))
2749 : {
2750 : Assert(key->sk_flags & SK_SEARCHARRAY);
2751 : Assert(key->sk_flags & SK_BT_SKIP);
2752 : Assert(requiredSameDir || forcenonrequired);
2753 :
2754 : /*
2755 : * Cannot fall back on _bt_tuple_before_array_skeys when we're
2756 : * treating the scan's keys as nonrequired, though. Just handle
2757 : * this like any other non-required equality-type array key.
2758 : */
2759 37234 : if (forcenonrequired)
2760 13071970 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2761 : tupdesc, *ikey, false);
2762 :
2763 35206 : *continuescan = false;
2764 35206 : return false;
2765 : }
2766 :
2767 : /* row-comparison keys need special processing */
2768 69595960 : if (key->sk_flags & SK_ROW_HEADER)
2769 : {
2770 2454 : if (_bt_check_rowcompare(key, tuple, tupnatts, tupdesc, dir,
2771 : forcenonrequired, continuescan))
2772 2388 : continue;
2773 66 : return false;
2774 : }
2775 :
2776 69593506 : datum = index_getattr(tuple,
2777 69593506 : key->sk_attno,
2778 : tupdesc,
2779 : &isNull);
2780 :
2781 69593506 : if (key->sk_flags & SK_ISNULL)
2782 : {
2783 : /* Handle IS NULL/NOT NULL tests */
2784 19077370 : if (key->sk_flags & SK_SEARCHNULL)
2785 : {
2786 18128 : if (isNull)
2787 428 : continue; /* tuple satisfies this qual */
2788 : }
2789 : else
2790 : {
2791 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
2792 : Assert(!(key->sk_flags & SK_BT_SKIP));
2793 19059242 : if (!isNull)
2794 19059170 : continue; /* tuple satisfies this qual */
2795 : }
2796 :
2797 : /*
2798 : * Tuple fails this qual. If it's a required qual for the current
2799 : * scan direction, then we can conclude no further tuples will
2800 : * pass, either.
2801 : */
2802 17772 : if (requiredSameDir)
2803 204 : *continuescan = false;
2804 17568 : else if (unlikely(key->sk_flags & SK_BT_SKIP))
2805 : {
2806 : /*
2807 : * If we're treating scan keys as nonrequired, and encounter a
2808 : * skip array scan key whose current element is NULL, then it
2809 : * must be a non-range skip array. It must be satisfied, so
2810 : * there's no need to call _bt_advance_array_keys to check.
2811 : */
2812 : Assert(forcenonrequired && *ikey > 0);
2813 0 : continue;
2814 : }
2815 :
2816 : /*
2817 : * This indextuple doesn't match the qual.
2818 : */
2819 17772 : return false;
2820 : }
2821 :
2822 50516136 : if (isNull)
2823 : {
2824 : /*
2825 : * Scalar scan key isn't satisfied by NULL tuple value.
2826 : *
2827 : * If we're treating scan keys as nonrequired, and key is for a
2828 : * skip array, then we must attempt to advance the array to NULL
2829 : * (if we're successful then the tuple might match the qual).
2830 : */
2831 228 : if (unlikely(forcenonrequired && key->sk_flags & SK_BT_SKIP))
2832 0 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2833 : tupdesc, *ikey, false);
2834 :
2835 228 : if (key->sk_flags & SK_BT_NULLS_FIRST)
2836 : {
2837 : /*
2838 : * Since NULLs are sorted before non-NULLs, we know we have
2839 : * reached the lower limit of the range of values for this
2840 : * index attr. On a backward scan, we can stop if this qual
2841 : * is one of the "must match" subset. We can stop regardless
2842 : * of whether the qual is > or <, so long as it's required,
2843 : * because it's not possible for any future tuples to pass. On
2844 : * a forward scan, however, we must keep going, because we may
2845 : * have initially positioned to the start of the index.
2846 : * (_bt_advance_array_keys also relies on this behavior during
2847 : * forward scans.)
2848 : */
2849 0 : if ((requiredSameDir || requiredOppositeDirOnly) &&
2850 : ScanDirectionIsBackward(dir))
2851 0 : *continuescan = false;
2852 : }
2853 : else
2854 : {
2855 : /*
2856 : * Since NULLs are sorted after non-NULLs, we know we have
2857 : * reached the upper limit of the range of values for this
2858 : * index attr. On a forward scan, we can stop if this qual is
2859 : * one of the "must match" subset. We can stop regardless of
2860 : * whether the qual is > or <, so long as it's required,
2861 : * because it's not possible for any future tuples to pass. On
2862 : * a backward scan, however, we must keep going, because we
2863 : * may have initially positioned to the end of the index.
2864 : * (_bt_advance_array_keys also relies on this behavior during
2865 : * backward scans.)
2866 : */
2867 228 : if ((requiredSameDir || requiredOppositeDirOnly) &&
2868 : ScanDirectionIsForward(dir))
2869 222 : *continuescan = false;
2870 : }
2871 :
2872 : /*
2873 : * This indextuple doesn't match the qual.
2874 : */
2875 228 : return false;
2876 : }
2877 :
2878 50515908 : if (!DatumGetBool(FunctionCall2Coll(&key->sk_func, key->sk_collation,
2879 : datum, key->sk_argument)))
2880 : {
2881 : /*
2882 : * Tuple fails this qual. If it's a required qual for the current
2883 : * scan direction, then we can conclude no further tuples will
2884 : * pass, either.
2885 : *
2886 : * Note: because we stop the scan as soon as any required equality
2887 : * qual fails, it is critical that equality quals be used for the
2888 : * initial positioning in _bt_first() when they are available. See
2889 : * comments in _bt_first().
2890 : */
2891 13016670 : if (requiredSameDir)
2892 12664326 : *continuescan = false;
2893 :
2894 : /*
2895 : * If this is a non-required equality-type array key, the tuple
2896 : * needs to be checked against every possible array key. Handle
2897 : * this by "advancing" the scan key's array to a matching value
2898 : * (if we're successful then the tuple might match the qual).
2899 : */
2900 352344 : else if (advancenonrequired &&
2901 344904 : key->sk_strategy == BTEqualStrategyNumber &&
2902 269110 : (key->sk_flags & SK_SEARCHARRAY))
2903 6856 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2904 : tupdesc, *ikey, false);
2905 :
2906 : /*
2907 : * This indextuple doesn't match the qual.
2908 : */
2909 13009814 : return false;
2910 : }
2911 : }
2912 :
2913 : /* If we get here, the tuple passes all index quals. */
2914 48368326 : return true;
2915 : }
2916 :
2917 : /*
2918 : * Test whether an indextuple satisfies a row-comparison scan condition.
2919 : *
2920 : * Return true if so, false if not. If not, also clear *continuescan if
2921 : * it's not possible for any future tuples in the current scan direction
2922 : * to pass the qual.
2923 : *
2924 : * This is a subroutine for _bt_checkkeys/_bt_check_compare.
2925 : */
2926 : static bool
2927 2454 : _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
2928 : TupleDesc tupdesc, ScanDirection dir,
2929 : bool forcenonrequired, bool *continuescan)
2930 : {
2931 2454 : ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
2932 2454 : int32 cmpresult = 0;
2933 : bool result;
2934 :
2935 : /* First subkey should be same as the header says */
2936 : Assert(subkey->sk_attno == skey->sk_attno);
2937 :
2938 : /* Loop over columns of the row condition */
2939 : for (;;)
2940 240 : {
2941 : Datum datum;
2942 : bool isNull;
2943 :
2944 : Assert(subkey->sk_flags & SK_ROW_MEMBER);
2945 :
2946 : /* When a NULL row member is compared, the row never matches */
2947 2694 : if (subkey->sk_flags & SK_ISNULL)
2948 : {
2949 : /*
2950 : * Unlike the simple-scankey case, this isn't a disallowed case
2951 : * (except when it's the first row element that has the NULL arg).
2952 : * But it can never match. If all the earlier row comparison
2953 : * columns are required for the scan direction, we can stop the
2954 : * scan, because there can't be another tuple that will succeed.
2955 : */
2956 : Assert(subkey != (ScanKey) DatumGetPointer(skey->sk_argument));
2957 12 : subkey--;
2958 12 : if (forcenonrequired)
2959 : {
2960 : /* treating scan's keys as non-required */
2961 : }
2962 12 : else if ((subkey->sk_flags & SK_BT_REQFWD) &&
2963 : ScanDirectionIsForward(dir))
2964 6 : *continuescan = false;
2965 6 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
2966 : ScanDirectionIsBackward(dir))
2967 6 : *continuescan = false;
2968 66 : return false;
2969 : }
2970 :
2971 2682 : if (subkey->sk_attno > tupnatts)
2972 : {
2973 : /*
2974 : * This attribute is truncated (must be high key). The value for
2975 : * this attribute in the first non-pivot tuple on the page to the
2976 : * right could be any possible value. Assume that truncated
2977 : * attribute passes the qual.
2978 : */
2979 : Assert(BTreeTupleIsPivot(tuple));
2980 6 : return true;
2981 : }
2982 :
2983 2676 : datum = index_getattr(tuple,
2984 2676 : subkey->sk_attno,
2985 : tupdesc,
2986 : &isNull);
2987 :
2988 2676 : if (isNull)
2989 : {
2990 : int reqflags;
2991 :
2992 48 : if (forcenonrequired)
2993 : {
2994 : /* treating scan's keys as non-required */
2995 : }
2996 48 : else if (subkey->sk_flags & SK_BT_NULLS_FIRST)
2997 : {
2998 : /*
2999 : * Since NULLs are sorted before non-NULLs, we know we have
3000 : * reached the lower limit of the range of values for this
3001 : * index attr. On a backward scan, we can stop if this qual
3002 : * is one of the "must match" subset. However, on a forwards
3003 : * scan, we must keep going, because we may have initially
3004 : * positioned to the start of the index.
3005 : *
3006 : * All required NULLS FIRST > row members can use NULL tuple
3007 : * values to end backwards scans, just like with other values.
3008 : * A qual "WHERE (a, b, c) > (9, 42, 'foo')" can terminate a
3009 : * backwards scan upon reaching the index's rightmost "a = 9"
3010 : * tuple whose "b" column contains a NULL (if not sooner).
3011 : * Since "b" is NULLS FIRST, we can treat its NULLs as "<" 42.
3012 : */
3013 0 : reqflags = SK_BT_REQBKWD;
3014 :
3015 : /*
3016 : * When a most significant required NULLS FIRST < row compare
3017 : * member sees NULL tuple values during a backwards scan, it
3018 : * signals the end of matches for the whole row compare/scan.
3019 : * A qual "WHERE (a, b, c) < (9, 42, 'foo')" will terminate a
3020 : * backwards scan upon reaching the rightmost tuple whose "a"
3021 : * column has a NULL. The "a" NULL value is "<" 9, and yet
3022 : * our < row compare will still end the scan. (This isn't
3023 : * safe with later/lower-order row members. Notice that it
3024 : * can only happen with an "a" NULL some time after the scan
3025 : * completely stops needing to use its "b" and "c" members.)
3026 : */
3027 0 : if (subkey == (ScanKey) DatumGetPointer(skey->sk_argument))
3028 0 : reqflags |= SK_BT_REQFWD; /* safe, first row member */
3029 :
3030 0 : if ((subkey->sk_flags & reqflags) &&
3031 : ScanDirectionIsBackward(dir))
3032 0 : *continuescan = false;
3033 : }
3034 : else
3035 : {
3036 : /*
3037 : * Since NULLs are sorted after non-NULLs, we know we have
3038 : * reached the upper limit of the range of values for this
3039 : * index attr. On a forward scan, we can stop if this qual is
3040 : * one of the "must match" subset. However, on a backward
3041 : * scan, we must keep going, because we may have initially
3042 : * positioned to the end of the index.
3043 : *
3044 : * All required NULLS LAST < row members can use NULL tuple
3045 : * values to end forwards scans, just like with other values.
3046 : * A qual "WHERE (a, b, c) < (9, 42, 'foo')" can terminate a
3047 : * forwards scan upon reaching the index's leftmost "a = 9"
3048 : * tuple whose "b" column contains a NULL (if not sooner).
3049 : * Since "b" is NULLS LAST, we can treat its NULLs as ">" 42.
3050 : */
3051 48 : reqflags = SK_BT_REQFWD;
3052 :
3053 : /*
3054 : * When a most significant required NULLS LAST > row compare
3055 : * member sees NULL tuple values during a forwards scan, it
3056 : * signals the end of matches for the whole row compare/scan.
3057 : * A qual "WHERE (a, b, c) > (9, 42, 'foo')" will terminate a
3058 : * forwards scan upon reaching the leftmost tuple whose "a"
3059 : * column has a NULL. The "a" NULL value is ">" 9, and yet
3060 : * our > row compare will end the scan. (This isn't safe with
3061 : * later/lower-order row members. Notice that it can only
3062 : * happen with an "a" NULL some time after the scan completely
3063 : * stops needing to use its "b" and "c" members.)
3064 : */
3065 48 : if (subkey == (ScanKey) DatumGetPointer(skey->sk_argument))
3066 0 : reqflags |= SK_BT_REQBKWD; /* safe, first row member */
3067 :
3068 48 : if ((subkey->sk_flags & reqflags) &&
3069 : ScanDirectionIsForward(dir))
3070 0 : *continuescan = false;
3071 : }
3072 :
3073 : /*
3074 : * In any case, this indextuple doesn't match the qual.
3075 : */
3076 48 : return false;
3077 : }
3078 :
3079 : /* Perform the test --- three-way comparison not bool operator */
3080 2628 : cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
3081 : subkey->sk_collation,
3082 : datum,
3083 : subkey->sk_argument));
3084 :
3085 2628 : if (subkey->sk_flags & SK_BT_DESC)
3086 0 : INVERT_COMPARE_RESULT(cmpresult);
3087 :
3088 : /* Done comparing if unequal, else advance to next column */
3089 2628 : if (cmpresult != 0)
3090 2388 : break;
3091 :
3092 240 : if (subkey->sk_flags & SK_ROW_END)
3093 0 : break;
3094 240 : subkey++;
3095 : }
3096 :
3097 : /*
3098 : * At this point cmpresult indicates the overall result of the row
3099 : * comparison, and subkey points to the deciding column (or the last
3100 : * column if the result is "=").
3101 : */
3102 2388 : switch (subkey->sk_strategy)
3103 : {
3104 : /* EQ and NE cases aren't allowed here */
3105 186 : case BTLessStrategyNumber:
3106 186 : result = (cmpresult < 0);
3107 186 : break;
3108 1584 : case BTLessEqualStrategyNumber:
3109 1584 : result = (cmpresult <= 0);
3110 1584 : break;
3111 246 : case BTGreaterEqualStrategyNumber:
3112 246 : result = (cmpresult >= 0);
3113 246 : break;
3114 372 : case BTGreaterStrategyNumber:
3115 372 : result = (cmpresult > 0);
3116 372 : break;
3117 0 : default:
3118 0 : elog(ERROR, "unexpected strategy number %d", subkey->sk_strategy);
3119 : result = 0; /* keep compiler quiet */
3120 : break;
3121 : }
3122 :
3123 2388 : if (!result && !forcenonrequired)
3124 : {
3125 : /*
3126 : * Tuple fails this qual. If it's a required qual for the current
3127 : * scan direction, then we can conclude no further tuples will pass,
3128 : * either. Note we have to look at the deciding column, not
3129 : * necessarily the first or last column of the row condition.
3130 : */
3131 6 : if ((subkey->sk_flags & SK_BT_REQFWD) &&
3132 : ScanDirectionIsForward(dir))
3133 6 : *continuescan = false;
3134 0 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
3135 : ScanDirectionIsBackward(dir))
3136 0 : *continuescan = false;
3137 : }
3138 :
3139 2388 : return result;
3140 : }
3141 :
3142 : /*
3143 : * Determine if a scan with array keys should skip over uninteresting tuples.
3144 : *
3145 : * This is a subroutine for _bt_checkkeys. Called when _bt_readpage's linear
3146 : * search process (started after it finishes reading an initial group of
3147 : * matching tuples, used to locate the start of the next group of tuples
3148 : * matching the next set of required array keys) has already scanned an
3149 : * excessive number of tuples whose key space is "between arrays".
3150 : *
3151 : * When we perform look ahead successfully, we'll sets pstate.skip, which
3152 : * instructs _bt_readpage to skip ahead to that tuple next (could be past the
3153 : * end of the scan's leaf page). Pages where the optimization is effective
3154 : * will generally still need to skip several times. Each call here performs
3155 : * only a single "look ahead" comparison of a later tuple, whose distance from
3156 : * the current tuple's offset number is determined by applying heuristics.
3157 : */
3158 : static void
3159 10624 : _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
3160 : int tupnatts, TupleDesc tupdesc)
3161 : {
3162 10624 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3163 10624 : ScanDirection dir = so->currPos.dir;
3164 : OffsetNumber aheadoffnum;
3165 : IndexTuple ahead;
3166 :
3167 : Assert(!pstate->forcenonrequired);
3168 :
3169 : /* Avoid looking ahead when comparing the page high key */
3170 10624 : if (pstate->offnum < pstate->minoff)
3171 0 : return;
3172 :
3173 : /*
3174 : * Don't look ahead when there aren't enough tuples remaining on the page
3175 : * (in the current scan direction) for it to be worth our while
3176 : */
3177 10624 : if (ScanDirectionIsForward(dir) &&
3178 10546 : pstate->offnum >= pstate->maxoff - LOOK_AHEAD_DEFAULT_DISTANCE)
3179 498 : return;
3180 10126 : else if (ScanDirectionIsBackward(dir) &&
3181 78 : pstate->offnum <= pstate->minoff + LOOK_AHEAD_DEFAULT_DISTANCE)
3182 24 : return;
3183 :
3184 : /*
3185 : * The look ahead distance starts small, and ramps up as each call here
3186 : * allows _bt_readpage to skip over more tuples
3187 : */
3188 10102 : if (!pstate->targetdistance)
3189 6220 : pstate->targetdistance = LOOK_AHEAD_DEFAULT_DISTANCE;
3190 3882 : else if (pstate->targetdistance < MaxIndexTuplesPerPage / 2)
3191 3882 : pstate->targetdistance *= 2;
3192 :
3193 : /* Don't read past the end (or before the start) of the page, though */
3194 10102 : if (ScanDirectionIsForward(dir))
3195 10048 : aheadoffnum = Min((int) pstate->maxoff,
3196 : (int) pstate->offnum + pstate->targetdistance);
3197 : else
3198 54 : aheadoffnum = Max((int) pstate->minoff,
3199 : (int) pstate->offnum - pstate->targetdistance);
3200 :
3201 10102 : ahead = (IndexTuple) PageGetItem(pstate->page,
3202 10102 : PageGetItemId(pstate->page, aheadoffnum));
3203 10102 : if (_bt_tuple_before_array_skeys(scan, dir, ahead, tupdesc, tupnatts,
3204 : false, 0, NULL))
3205 : {
3206 : /*
3207 : * Success -- instruct _bt_readpage to skip ahead to very next tuple
3208 : * after the one we determined was still before the current array keys
3209 : */
3210 3260 : if (ScanDirectionIsForward(dir))
3211 3224 : pstate->skip = aheadoffnum + 1;
3212 : else
3213 36 : pstate->skip = aheadoffnum - 1;
3214 : }
3215 : else
3216 : {
3217 : /*
3218 : * Failure -- "ahead" tuple is too far ahead (we were too aggressive).
3219 : *
3220 : * Reset the number of rechecks, and aggressively reduce the target
3221 : * distance (we're much more aggressive here than we were when the
3222 : * distance was initially ramped up).
3223 : */
3224 6842 : pstate->rechecks = 0;
3225 6842 : pstate->targetdistance = Max(pstate->targetdistance / 8, 1);
3226 : }
3227 : }
3228 :
3229 : /*
3230 : * _bt_killitems - set LP_DEAD state for items an indexscan caller has
3231 : * told us were killed
3232 : *
3233 : * scan->opaque, referenced locally through so, contains information about the
3234 : * current page and killed tuples thereon (generally, this should only be
3235 : * called if so->numKilled > 0).
3236 : *
3237 : * Caller should not have a lock on the so->currPos page, but must hold a
3238 : * buffer pin when !so->dropPin. When we return, it still won't be locked.
3239 : * It'll continue to hold whatever pins were held before calling here.
3240 : *
3241 : * We match items by heap TID before assuming they are the right ones to set
3242 : * LP_DEAD. If the scan is one that holds a buffer pin on the target page
3243 : * continuously from initially reading the items until applying this function
3244 : * (if it is a !so->dropPin scan), VACUUM cannot have deleted any items on the
3245 : * page, so the page's TIDs can't have been recycled by now. There's no risk
3246 : * that we'll confuse a new index tuple that happens to use a recycled TID
3247 : * with a now-removed tuple with the same TID (that used to be on this same
3248 : * page). We can't rely on that during scans that drop buffer pins eagerly
3249 : * (so->dropPin scans), though, so we must condition setting LP_DEAD bits on
3250 : * the page LSN having not changed since back when _bt_readpage saw the page.
3251 : * We totally give up on setting LP_DEAD bits when the page LSN changed.
3252 : *
3253 : * We give up much less often during !so->dropPin scans, but it still happens.
3254 : * We cope with cases where items have moved right due to insertions. If an
3255 : * item has moved off the current page due to a split, we'll fail to find it
3256 : * and just give up on it.
3257 : */
3258 : void
3259 174098 : _bt_killitems(IndexScanDesc scan)
3260 : {
3261 174098 : Relation rel = scan->indexRelation;
3262 174098 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3263 : Page page;
3264 : BTPageOpaque opaque;
3265 : OffsetNumber minoff;
3266 : OffsetNumber maxoff;
3267 174098 : int numKilled = so->numKilled;
3268 174098 : bool killedsomething = false;
3269 : Buffer buf;
3270 :
3271 : Assert(numKilled > 0);
3272 : Assert(BTScanPosIsValid(so->currPos));
3273 : Assert(scan->heapRelation != NULL); /* can't be a bitmap index scan */
3274 :
3275 : /* Always invalidate so->killedItems[] before leaving so->currPos */
3276 174098 : so->numKilled = 0;
3277 :
3278 174098 : if (!so->dropPin)
3279 : {
3280 : /*
3281 : * We have held the pin on this page since we read the index tuples,
3282 : * so all we need to do is lock it. The pin will have prevented
3283 : * concurrent VACUUMs from recycling any of the TIDs on the page.
3284 : */
3285 : Assert(BTScanPosIsPinned(so->currPos));
3286 39108 : buf = so->currPos.buf;
3287 39108 : _bt_lockbuf(rel, buf, BT_READ);
3288 : }
3289 : else
3290 : {
3291 : XLogRecPtr latestlsn;
3292 :
3293 : Assert(!BTScanPosIsPinned(so->currPos));
3294 : Assert(RelationNeedsWAL(rel));
3295 134990 : buf = _bt_getbuf(rel, so->currPos.currPage, BT_READ);
3296 :
3297 134990 : latestlsn = BufferGetLSNAtomic(buf);
3298 : Assert(!XLogRecPtrIsInvalid(so->currPos.lsn));
3299 : Assert(so->currPos.lsn <= latestlsn);
3300 134990 : if (so->currPos.lsn != latestlsn)
3301 : {
3302 : /* Modified, give up on hinting */
3303 126 : _bt_relbuf(rel, buf);
3304 126 : return;
3305 : }
3306 :
3307 : /* Unmodified, hinting is safe */
3308 : }
3309 :
3310 173972 : page = BufferGetPage(buf);
3311 173972 : opaque = BTPageGetOpaque(page);
3312 173972 : minoff = P_FIRSTDATAKEY(opaque);
3313 173972 : maxoff = PageGetMaxOffsetNumber(page);
3314 :
3315 594942 : for (int i = 0; i < numKilled; i++)
3316 : {
3317 420970 : int itemIndex = so->killedItems[i];
3318 420970 : BTScanPosItem *kitem = &so->currPos.items[itemIndex];
3319 420970 : OffsetNumber offnum = kitem->indexOffset;
3320 :
3321 : Assert(itemIndex >= so->currPos.firstItem &&
3322 : itemIndex <= so->currPos.lastItem);
3323 420970 : if (offnum < minoff)
3324 0 : continue; /* pure paranoia */
3325 8987190 : while (offnum <= maxoff)
3326 : {
3327 8923514 : ItemId iid = PageGetItemId(page, offnum);
3328 8923514 : IndexTuple ituple = (IndexTuple) PageGetItem(page, iid);
3329 8923514 : bool killtuple = false;
3330 :
3331 8923514 : if (BTreeTupleIsPosting(ituple))
3332 : {
3333 2478392 : int pi = i + 1;
3334 2478392 : int nposting = BTreeTupleGetNPosting(ituple);
3335 : int j;
3336 :
3337 : /*
3338 : * We rely on the convention that heap TIDs in the scanpos
3339 : * items array are stored in ascending heap TID order for a
3340 : * group of TIDs that originally came from a posting list
3341 : * tuple. This convention even applies during backwards
3342 : * scans, where returning the TIDs in descending order might
3343 : * seem more natural. This is about effectiveness, not
3344 : * correctness.
3345 : *
3346 : * Note that the page may have been modified in almost any way
3347 : * since we first read it (in the !so->dropPin case), so it's
3348 : * possible that this posting list tuple wasn't a posting list
3349 : * tuple when we first encountered its heap TIDs.
3350 : */
3351 2540336 : for (j = 0; j < nposting; j++)
3352 : {
3353 2538586 : ItemPointer item = BTreeTupleGetPostingN(ituple, j);
3354 :
3355 2538586 : if (!ItemPointerEquals(item, &kitem->heapTid))
3356 2476642 : break; /* out of posting list loop */
3357 :
3358 : /*
3359 : * kitem must have matching offnum when heap TIDs match,
3360 : * though only in the common case where the page can't
3361 : * have been concurrently modified
3362 : */
3363 : Assert(kitem->indexOffset == offnum || !so->dropPin);
3364 :
3365 : /*
3366 : * Read-ahead to later kitems here.
3367 : *
3368 : * We rely on the assumption that not advancing kitem here
3369 : * will prevent us from considering the posting list tuple
3370 : * fully dead by not matching its next heap TID in next
3371 : * loop iteration.
3372 : *
3373 : * If, on the other hand, this is the final heap TID in
3374 : * the posting list tuple, then tuple gets killed
3375 : * regardless (i.e. we handle the case where the last
3376 : * kitem is also the last heap TID in the last index tuple
3377 : * correctly -- posting tuple still gets killed).
3378 : */
3379 61944 : if (pi < numKilled)
3380 25238 : kitem = &so->currPos.items[so->killedItems[pi++]];
3381 : }
3382 :
3383 : /*
3384 : * Don't bother advancing the outermost loop's int iterator to
3385 : * avoid processing killed items that relate to the same
3386 : * offnum/posting list tuple. This micro-optimization hardly
3387 : * seems worth it. (Further iterations of the outermost loop
3388 : * will fail to match on this same posting list's first heap
3389 : * TID instead, so we'll advance to the next offnum/index
3390 : * tuple pretty quickly.)
3391 : */
3392 2478392 : if (j == nposting)
3393 1750 : killtuple = true;
3394 : }
3395 6445122 : else if (ItemPointerEquals(&ituple->t_tid, &kitem->heapTid))
3396 356334 : killtuple = true;
3397 :
3398 : /*
3399 : * Mark index item as dead, if it isn't already. Since this
3400 : * happens while holding a buffer lock possibly in shared mode,
3401 : * it's possible that multiple processes attempt to do this
3402 : * simultaneously, leading to multiple full-page images being sent
3403 : * to WAL (if wal_log_hints or data checksums are enabled), which
3404 : * is undesirable.
3405 : */
3406 8923514 : if (killtuple && !ItemIdIsDead(iid))
3407 : {
3408 : /* found the item/all posting list items */
3409 357294 : ItemIdMarkDead(iid);
3410 357294 : killedsomething = true;
3411 357294 : break; /* out of inner search loop */
3412 : }
3413 8566220 : offnum = OffsetNumberNext(offnum);
3414 : }
3415 : }
3416 :
3417 : /*
3418 : * Since this can be redone later if needed, mark as dirty hint.
3419 : *
3420 : * Whenever we mark anything LP_DEAD, we also set the page's
3421 : * BTP_HAS_GARBAGE flag, which is likewise just a hint. (Note that we
3422 : * only rely on the page-level flag in !heapkeyspace indexes.)
3423 : */
3424 173972 : if (killedsomething)
3425 : {
3426 135764 : opaque->btpo_flags |= BTP_HAS_GARBAGE;
3427 135764 : MarkBufferDirtyHint(buf, true);
3428 : }
3429 :
3430 173972 : if (!so->dropPin)
3431 39108 : _bt_unlockbuf(rel, buf);
3432 : else
3433 134864 : _bt_relbuf(rel, buf);
3434 : }
3435 :
3436 :
3437 : /*
3438 : * The following routines manage a shared-memory area in which we track
3439 : * assignment of "vacuum cycle IDs" to currently-active btree vacuuming
3440 : * operations. There is a single counter which increments each time we
3441 : * start a vacuum to assign it a cycle ID. Since multiple vacuums could
3442 : * be active concurrently, we have to track the cycle ID for each active
3443 : * vacuum; this requires at most MaxBackends entries (usually far fewer).
3444 : * We assume at most one vacuum can be active for a given index.
3445 : *
3446 : * Access to the shared memory area is controlled by BtreeVacuumLock.
3447 : * In principle we could use a separate lmgr locktag for each index,
3448 : * but a single LWLock is much cheaper, and given the short time that
3449 : * the lock is ever held, the concurrency hit should be minimal.
3450 : */
3451 :
3452 : typedef struct BTOneVacInfo
3453 : {
3454 : LockRelId relid; /* global identifier of an index */
3455 : BTCycleId cycleid; /* cycle ID for its active VACUUM */
3456 : } BTOneVacInfo;
3457 :
3458 : typedef struct BTVacInfo
3459 : {
3460 : BTCycleId cycle_ctr; /* cycle ID most recently assigned */
3461 : int num_vacuums; /* number of currently active VACUUMs */
3462 : int max_vacuums; /* allocated length of vacuums[] array */
3463 : BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER];
3464 : } BTVacInfo;
3465 :
3466 : static BTVacInfo *btvacinfo;
3467 :
3468 :
3469 : /*
3470 : * _bt_vacuum_cycleid --- get the active vacuum cycle ID for an index,
3471 : * or zero if there is no active VACUUM
3472 : *
3473 : * Note: for correct interlocking, the caller must already hold pin and
3474 : * exclusive lock on each buffer it will store the cycle ID into. This
3475 : * ensures that even if a VACUUM starts immediately afterwards, it cannot
3476 : * process those pages until the page split is complete.
3477 : */
3478 : BTCycleId
3479 22378 : _bt_vacuum_cycleid(Relation rel)
3480 : {
3481 22378 : BTCycleId result = 0;
3482 : int i;
3483 :
3484 : /* Share lock is enough since this is a read-only operation */
3485 22378 : LWLockAcquire(BtreeVacuumLock, LW_SHARED);
3486 :
3487 22386 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3488 : {
3489 8 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
3490 :
3491 8 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3492 0 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3493 : {
3494 0 : result = vac->cycleid;
3495 0 : break;
3496 : }
3497 : }
3498 :
3499 22378 : LWLockRelease(BtreeVacuumLock);
3500 22378 : return result;
3501 : }
3502 :
3503 : /*
3504 : * _bt_start_vacuum --- assign a cycle ID to a just-starting VACUUM operation
3505 : *
3506 : * Note: the caller must guarantee that it will eventually call
3507 : * _bt_end_vacuum, else we'll permanently leak an array slot. To ensure
3508 : * that this happens even in elog(FATAL) scenarios, the appropriate coding
3509 : * is not just a PG_TRY, but
3510 : * PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel))
3511 : */
3512 : BTCycleId
3513 3052 : _bt_start_vacuum(Relation rel)
3514 : {
3515 : BTCycleId result;
3516 : int i;
3517 : BTOneVacInfo *vac;
3518 :
3519 3052 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
3520 :
3521 : /*
3522 : * Assign the next cycle ID, being careful to avoid zero as well as the
3523 : * reserved high values.
3524 : */
3525 3052 : result = ++(btvacinfo->cycle_ctr);
3526 3052 : if (result == 0 || result > MAX_BT_CYCLE_ID)
3527 0 : result = btvacinfo->cycle_ctr = 1;
3528 :
3529 : /* Let's just make sure there's no entry already for this index */
3530 3056 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3531 : {
3532 4 : vac = &btvacinfo->vacuums[i];
3533 4 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3534 0 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3535 : {
3536 : /*
3537 : * Unlike most places in the backend, we have to explicitly
3538 : * release our LWLock before throwing an error. This is because
3539 : * we expect _bt_end_vacuum() to be called before transaction
3540 : * abort cleanup can run to release LWLocks.
3541 : */
3542 0 : LWLockRelease(BtreeVacuumLock);
3543 0 : elog(ERROR, "multiple active vacuums for index \"%s\"",
3544 : RelationGetRelationName(rel));
3545 : }
3546 : }
3547 :
3548 : /* OK, add an entry */
3549 3052 : if (btvacinfo->num_vacuums >= btvacinfo->max_vacuums)
3550 : {
3551 0 : LWLockRelease(BtreeVacuumLock);
3552 0 : elog(ERROR, "out of btvacinfo slots");
3553 : }
3554 3052 : vac = &btvacinfo->vacuums[btvacinfo->num_vacuums];
3555 3052 : vac->relid = rel->rd_lockInfo.lockRelId;
3556 3052 : vac->cycleid = result;
3557 3052 : btvacinfo->num_vacuums++;
3558 :
3559 3052 : LWLockRelease(BtreeVacuumLock);
3560 3052 : return result;
3561 : }
3562 :
3563 : /*
3564 : * _bt_end_vacuum --- mark a btree VACUUM operation as done
3565 : *
3566 : * Note: this is deliberately coded not to complain if no entry is found;
3567 : * this allows the caller to put PG_TRY around the start_vacuum operation.
3568 : */
3569 : void
3570 3052 : _bt_end_vacuum(Relation rel)
3571 : {
3572 : int i;
3573 :
3574 3052 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
3575 :
3576 : /* Find the array entry */
3577 3054 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3578 : {
3579 3054 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
3580 :
3581 3054 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3582 3052 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3583 : {
3584 : /* Remove it by shifting down the last entry */
3585 3052 : *vac = btvacinfo->vacuums[btvacinfo->num_vacuums - 1];
3586 3052 : btvacinfo->num_vacuums--;
3587 3052 : break;
3588 : }
3589 : }
3590 :
3591 3052 : LWLockRelease(BtreeVacuumLock);
3592 3052 : }
3593 :
3594 : /*
3595 : * _bt_end_vacuum wrapped as an on_shmem_exit callback function
3596 : */
3597 : void
3598 0 : _bt_end_vacuum_callback(int code, Datum arg)
3599 : {
3600 0 : _bt_end_vacuum((Relation) DatumGetPointer(arg));
3601 0 : }
3602 :
3603 : /*
3604 : * BTreeShmemSize --- report amount of shared memory space needed
3605 : */
3606 : Size
3607 6156 : BTreeShmemSize(void)
3608 : {
3609 : Size size;
3610 :
3611 6156 : size = offsetof(BTVacInfo, vacuums);
3612 6156 : size = add_size(size, mul_size(MaxBackends, sizeof(BTOneVacInfo)));
3613 6156 : return size;
3614 : }
3615 :
3616 : /*
3617 : * BTreeShmemInit --- initialize this module's shared memory
3618 : */
3619 : void
3620 2152 : BTreeShmemInit(void)
3621 : {
3622 : bool found;
3623 :
3624 2152 : btvacinfo = (BTVacInfo *) ShmemInitStruct("BTree Vacuum State",
3625 : BTreeShmemSize(),
3626 : &found);
3627 :
3628 2152 : if (!IsUnderPostmaster)
3629 : {
3630 : /* Initialize shared memory area */
3631 : Assert(!found);
3632 :
3633 : /*
3634 : * It doesn't really matter what the cycle counter starts at, but
3635 : * having it always start the same doesn't seem good. Seed with
3636 : * low-order bits of time() instead.
3637 : */
3638 2152 : btvacinfo->cycle_ctr = (BTCycleId) time(NULL);
3639 :
3640 2152 : btvacinfo->num_vacuums = 0;
3641 2152 : btvacinfo->max_vacuums = MaxBackends;
3642 : }
3643 : else
3644 : Assert(found);
3645 2152 : }
3646 :
3647 : bytea *
3648 326 : btoptions(Datum reloptions, bool validate)
3649 : {
3650 : static const relopt_parse_elt tab[] = {
3651 : {"fillfactor", RELOPT_TYPE_INT, offsetof(BTOptions, fillfactor)},
3652 : {"vacuum_cleanup_index_scale_factor", RELOPT_TYPE_REAL,
3653 : offsetof(BTOptions, vacuum_cleanup_index_scale_factor)},
3654 : {"deduplicate_items", RELOPT_TYPE_BOOL,
3655 : offsetof(BTOptions, deduplicate_items)}
3656 : };
3657 :
3658 326 : return (bytea *) build_reloptions(reloptions, validate,
3659 : RELOPT_KIND_BTREE,
3660 : sizeof(BTOptions),
3661 : tab, lengthof(tab));
3662 : }
3663 :
3664 : /*
3665 : * btproperty() -- Check boolean properties of indexes.
3666 : *
3667 : * This is optional, but handling AMPROP_RETURNABLE here saves opening the rel
3668 : * to call btcanreturn.
3669 : */
3670 : bool
3671 756 : btproperty(Oid index_oid, int attno,
3672 : IndexAMProperty prop, const char *propname,
3673 : bool *res, bool *isnull)
3674 : {
3675 756 : switch (prop)
3676 : {
3677 42 : case AMPROP_RETURNABLE:
3678 : /* answer only for columns, not AM or whole index */
3679 42 : if (attno == 0)
3680 12 : return false;
3681 : /* otherwise, btree can always return data */
3682 30 : *res = true;
3683 30 : return true;
3684 :
3685 714 : default:
3686 714 : return false; /* punt to generic code */
3687 : }
3688 : }
3689 :
3690 : /*
3691 : * btbuildphasename() -- Return name of index build phase.
3692 : */
3693 : char *
3694 0 : btbuildphasename(int64 phasenum)
3695 : {
3696 0 : switch (phasenum)
3697 : {
3698 0 : case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
3699 0 : return "initializing";
3700 0 : case PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN:
3701 0 : return "scanning table";
3702 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_1:
3703 0 : return "sorting live tuples";
3704 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_2:
3705 0 : return "sorting dead tuples";
3706 0 : case PROGRESS_BTREE_PHASE_LEAF_LOAD:
3707 0 : return "loading tuples in tree";
3708 0 : default:
3709 0 : return NULL;
3710 : }
3711 : }
3712 :
3713 : /*
3714 : * _bt_truncate() -- create tuple without unneeded suffix attributes.
3715 : *
3716 : * Returns truncated pivot index tuple allocated in caller's memory context,
3717 : * with key attributes copied from caller's firstright argument. If rel is
3718 : * an INCLUDE index, non-key attributes will definitely be truncated away,
3719 : * since they're not part of the key space. More aggressive suffix
3720 : * truncation can take place when it's clear that the returned tuple does not
3721 : * need one or more suffix key attributes. We only need to keep firstright
3722 : * attributes up to and including the first non-lastleft-equal attribute.
3723 : * Caller's insertion scankey is used to compare the tuples; the scankey's
3724 : * argument values are not considered here.
3725 : *
3726 : * Note that returned tuple's t_tid offset will hold the number of attributes
3727 : * present, so the original item pointer offset is not represented. Caller
3728 : * should only change truncated tuple's downlink. Note also that truncated
3729 : * key attributes are treated as containing "minus infinity" values by
3730 : * _bt_compare().
3731 : *
3732 : * In the worst case (when a heap TID must be appended to distinguish lastleft
3733 : * from firstright), the size of the returned tuple is the size of firstright
3734 : * plus the size of an additional MAXALIGN()'d item pointer. This guarantee
3735 : * is important, since callers need to stay under the 1/3 of a page
3736 : * restriction on tuple size. If this routine is ever taught to truncate
3737 : * within an attribute/datum, it will need to avoid returning an enlarged
3738 : * tuple to caller when truncation + TOAST compression ends up enlarging the
3739 : * final datum.
3740 : */
3741 : IndexTuple
3742 62040 : _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright,
3743 : BTScanInsert itup_key)
3744 : {
3745 62040 : TupleDesc itupdesc = RelationGetDescr(rel);
3746 62040 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
3747 : int keepnatts;
3748 : IndexTuple pivot;
3749 : IndexTuple tidpivot;
3750 : ItemPointer pivotheaptid;
3751 : Size newsize;
3752 :
3753 : /*
3754 : * We should only ever truncate non-pivot tuples from leaf pages. It's
3755 : * never okay to truncate when splitting an internal page.
3756 : */
3757 : Assert(!BTreeTupleIsPivot(lastleft) && !BTreeTupleIsPivot(firstright));
3758 :
3759 : /* Determine how many attributes must be kept in truncated tuple */
3760 62040 : keepnatts = _bt_keep_natts(rel, lastleft, firstright, itup_key);
3761 :
3762 : #ifdef DEBUG_NO_TRUNCATE
3763 : /* Force truncation to be ineffective for testing purposes */
3764 : keepnatts = nkeyatts + 1;
3765 : #endif
3766 :
3767 62040 : pivot = index_truncate_tuple(itupdesc, firstright,
3768 : Min(keepnatts, nkeyatts));
3769 :
3770 62040 : if (BTreeTupleIsPosting(pivot))
3771 : {
3772 : /*
3773 : * index_truncate_tuple() just returns a straight copy of firstright
3774 : * when it has no attributes to truncate. When that happens, we may
3775 : * need to truncate away a posting list here instead.
3776 : */
3777 : Assert(keepnatts == nkeyatts || keepnatts == nkeyatts + 1);
3778 : Assert(IndexRelationGetNumberOfAttributes(rel) == nkeyatts);
3779 1252 : pivot->t_info &= ~INDEX_SIZE_MASK;
3780 1252 : pivot->t_info |= MAXALIGN(BTreeTupleGetPostingOffset(firstright));
3781 : }
3782 :
3783 : /*
3784 : * If there is a distinguishing key attribute within pivot tuple, we're
3785 : * done
3786 : */
3787 62040 : if (keepnatts <= nkeyatts)
3788 : {
3789 60936 : BTreeTupleSetNAtts(pivot, keepnatts, false);
3790 60936 : return pivot;
3791 : }
3792 :
3793 : /*
3794 : * We have to store a heap TID in the new pivot tuple, since no non-TID
3795 : * key attribute value in firstright distinguishes the right side of the
3796 : * split from the left side. nbtree conceptualizes this case as an
3797 : * inability to truncate away any key attributes, since heap TID is
3798 : * treated as just another key attribute (despite lacking a pg_attribute
3799 : * entry).
3800 : *
3801 : * Use enlarged space that holds a copy of pivot. We need the extra space
3802 : * to store a heap TID at the end (using the special pivot tuple
3803 : * representation). Note that the original pivot already has firstright's
3804 : * possible posting list/non-key attribute values removed at this point.
3805 : */
3806 1104 : newsize = MAXALIGN(IndexTupleSize(pivot)) + MAXALIGN(sizeof(ItemPointerData));
3807 1104 : tidpivot = palloc0(newsize);
3808 1104 : memcpy(tidpivot, pivot, MAXALIGN(IndexTupleSize(pivot)));
3809 : /* Cannot leak memory here */
3810 1104 : pfree(pivot);
3811 :
3812 : /*
3813 : * Store all of firstright's key attribute values plus a tiebreaker heap
3814 : * TID value in enlarged pivot tuple
3815 : */
3816 1104 : tidpivot->t_info &= ~INDEX_SIZE_MASK;
3817 1104 : tidpivot->t_info |= newsize;
3818 1104 : BTreeTupleSetNAtts(tidpivot, nkeyatts, true);
3819 1104 : pivotheaptid = BTreeTupleGetHeapTID(tidpivot);
3820 :
3821 : /*
3822 : * Lehman & Yao use lastleft as the leaf high key in all cases, but don't
3823 : * consider suffix truncation. It seems like a good idea to follow that
3824 : * example in cases where no truncation takes place -- use lastleft's heap
3825 : * TID. (This is also the closest value to negative infinity that's
3826 : * legally usable.)
3827 : */
3828 1104 : ItemPointerCopy(BTreeTupleGetMaxHeapTID(lastleft), pivotheaptid);
3829 :
3830 : /*
3831 : * We're done. Assert() that heap TID invariants hold before returning.
3832 : *
3833 : * Lehman and Yao require that the downlink to the right page, which is to
3834 : * be inserted into the parent page in the second phase of a page split be
3835 : * a strict lower bound on items on the right page, and a non-strict upper
3836 : * bound for items on the left page. Assert that heap TIDs follow these
3837 : * invariants, since a heap TID value is apparently needed as a
3838 : * tiebreaker.
3839 : */
3840 : #ifndef DEBUG_NO_TRUNCATE
3841 : Assert(ItemPointerCompare(BTreeTupleGetMaxHeapTID(lastleft),
3842 : BTreeTupleGetHeapTID(firstright)) < 0);
3843 : Assert(ItemPointerCompare(pivotheaptid,
3844 : BTreeTupleGetHeapTID(lastleft)) >= 0);
3845 : Assert(ItemPointerCompare(pivotheaptid,
3846 : BTreeTupleGetHeapTID(firstright)) < 0);
3847 : #else
3848 :
3849 : /*
3850 : * Those invariants aren't guaranteed to hold for lastleft + firstright
3851 : * heap TID attribute values when they're considered here only because
3852 : * DEBUG_NO_TRUNCATE is defined (a heap TID is probably not actually
3853 : * needed as a tiebreaker). DEBUG_NO_TRUNCATE must therefore use a heap
3854 : * TID value that always works as a strict lower bound for items to the
3855 : * right. In particular, it must avoid using firstright's leading key
3856 : * attribute values along with lastleft's heap TID value when lastleft's
3857 : * TID happens to be greater than firstright's TID.
3858 : */
3859 : ItemPointerCopy(BTreeTupleGetHeapTID(firstright), pivotheaptid);
3860 :
3861 : /*
3862 : * Pivot heap TID should never be fully equal to firstright. Note that
3863 : * the pivot heap TID will still end up equal to lastleft's heap TID when
3864 : * that's the only usable value.
3865 : */
3866 : ItemPointerSetOffsetNumber(pivotheaptid,
3867 : OffsetNumberPrev(ItemPointerGetOffsetNumber(pivotheaptid)));
3868 : Assert(ItemPointerCompare(pivotheaptid,
3869 : BTreeTupleGetHeapTID(firstright)) < 0);
3870 : #endif
3871 :
3872 1104 : return tidpivot;
3873 : }
3874 :
3875 : /*
3876 : * _bt_keep_natts - how many key attributes to keep when truncating.
3877 : *
3878 : * Caller provides two tuples that enclose a split point. Caller's insertion
3879 : * scankey is used to compare the tuples; the scankey's argument values are
3880 : * not considered here.
3881 : *
3882 : * This can return a number of attributes that is one greater than the
3883 : * number of key attributes for the index relation. This indicates that the
3884 : * caller must use a heap TID as a unique-ifier in new pivot tuple.
3885 : */
3886 : static int
3887 62040 : _bt_keep_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright,
3888 : BTScanInsert itup_key)
3889 : {
3890 62040 : int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
3891 62040 : TupleDesc itupdesc = RelationGetDescr(rel);
3892 : int keepnatts;
3893 : ScanKey scankey;
3894 :
3895 : /*
3896 : * _bt_compare() treats truncated key attributes as having the value minus
3897 : * infinity, which would break searches within !heapkeyspace indexes. We
3898 : * must still truncate away non-key attribute values, though.
3899 : */
3900 62040 : if (!itup_key->heapkeyspace)
3901 0 : return nkeyatts;
3902 :
3903 62040 : scankey = itup_key->scankeys;
3904 62040 : keepnatts = 1;
3905 75020 : for (int attnum = 1; attnum <= nkeyatts; attnum++, scankey++)
3906 : {
3907 : Datum datum1,
3908 : datum2;
3909 : bool isNull1,
3910 : isNull2;
3911 :
3912 73916 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
3913 73916 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
3914 :
3915 73916 : if (isNull1 != isNull2)
3916 60936 : break;
3917 :
3918 147802 : if (!isNull1 &&
3919 73886 : DatumGetInt32(FunctionCall2Coll(&scankey->sk_func,
3920 : scankey->sk_collation,
3921 : datum1,
3922 : datum2)) != 0)
3923 60936 : break;
3924 :
3925 12980 : keepnatts++;
3926 : }
3927 :
3928 : /*
3929 : * Assert that _bt_keep_natts_fast() agrees with us in passing. This is
3930 : * expected in an allequalimage index.
3931 : */
3932 : Assert(!itup_key->allequalimage ||
3933 : keepnatts == _bt_keep_natts_fast(rel, lastleft, firstright));
3934 :
3935 62040 : return keepnatts;
3936 : }
3937 :
3938 : /*
3939 : * _bt_keep_natts_fast - fast bitwise variant of _bt_keep_natts.
3940 : *
3941 : * This is exported so that a candidate split point can have its effect on
3942 : * suffix truncation inexpensively evaluated ahead of time when finding a
3943 : * split location. A naive bitwise approach to datum comparisons is used to
3944 : * save cycles.
3945 : *
3946 : * The approach taken here usually provides the same answer as _bt_keep_natts
3947 : * will (for the same pair of tuples from a heapkeyspace index), since the
3948 : * majority of btree opclasses can never indicate that two datums are equal
3949 : * unless they're bitwise equal after detoasting. When an index only has
3950 : * "equal image" columns, routine is guaranteed to give the same result as
3951 : * _bt_keep_natts would.
3952 : *
3953 : * Callers can rely on the fact that attributes considered equal here are
3954 : * definitely also equal according to _bt_keep_natts, even when the index uses
3955 : * an opclass or collation that is not "allequalimage"/deduplication-safe.
3956 : * This weaker guarantee is good enough for nbtsplitloc.c caller, since false
3957 : * negatives generally only have the effect of making leaf page splits use a
3958 : * more balanced split point.
3959 : */
3960 : int
3961 13305710 : _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
3962 : {
3963 13305710 : TupleDesc itupdesc = RelationGetDescr(rel);
3964 13305710 : int keysz = IndexRelationGetNumberOfKeyAttributes(rel);
3965 : int keepnatts;
3966 :
3967 13305710 : keepnatts = 1;
3968 22370616 : for (int attnum = 1; attnum <= keysz; attnum++)
3969 : {
3970 : Datum datum1,
3971 : datum2;
3972 : bool isNull1,
3973 : isNull2;
3974 : CompactAttribute *att;
3975 :
3976 20003046 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
3977 20003046 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
3978 20003046 : att = TupleDescCompactAttr(itupdesc, attnum - 1);
3979 :
3980 20003046 : if (isNull1 != isNull2)
3981 10938140 : break;
3982 :
3983 20002842 : if (!isNull1 &&
3984 19955786 : !datum_image_eq(datum1, datum2, att->attbyval, att->attlen))
3985 10937936 : break;
3986 :
3987 9064906 : keepnatts++;
3988 : }
3989 :
3990 13305710 : return keepnatts;
3991 : }
3992 :
3993 : /*
3994 : * _bt_check_natts() -- Verify tuple has expected number of attributes.
3995 : *
3996 : * Returns value indicating if the expected number of attributes were found
3997 : * for a particular offset on page. This can be used as a general purpose
3998 : * sanity check.
3999 : *
4000 : * Testing a tuple directly with BTreeTupleGetNAtts() should generally be
4001 : * preferred to calling here. That's usually more convenient, and is always
4002 : * more explicit. Call here instead when offnum's tuple may be a negative
4003 : * infinity tuple that uses the pre-v11 on-disk representation, or when a low
4004 : * context check is appropriate. This routine is as strict as possible about
4005 : * what is expected on each version of btree.
4006 : */
4007 : bool
4008 4057866 : _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
4009 : {
4010 4057866 : int16 natts = IndexRelationGetNumberOfAttributes(rel);
4011 4057866 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
4012 4057866 : BTPageOpaque opaque = BTPageGetOpaque(page);
4013 : IndexTuple itup;
4014 : int tupnatts;
4015 :
4016 : /*
4017 : * We cannot reliably test a deleted or half-dead page, since they have
4018 : * dummy high keys
4019 : */
4020 4057866 : if (P_IGNORE(opaque))
4021 0 : return true;
4022 :
4023 : Assert(offnum >= FirstOffsetNumber &&
4024 : offnum <= PageGetMaxOffsetNumber(page));
4025 :
4026 4057866 : itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
4027 4057866 : tupnatts = BTreeTupleGetNAtts(itup, rel);
4028 :
4029 : /* !heapkeyspace indexes do not support deduplication */
4030 4057866 : if (!heapkeyspace && BTreeTupleIsPosting(itup))
4031 0 : return false;
4032 :
4033 : /* Posting list tuples should never have "pivot heap TID" bit set */
4034 4057866 : if (BTreeTupleIsPosting(itup) &&
4035 21906 : (ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) &
4036 : BT_PIVOT_HEAP_TID_ATTR) != 0)
4037 0 : return false;
4038 :
4039 : /* INCLUDE indexes do not support deduplication */
4040 4057866 : if (natts != nkeyatts && BTreeTupleIsPosting(itup))
4041 0 : return false;
4042 :
4043 4057866 : if (P_ISLEAF(opaque))
4044 : {
4045 4043514 : if (offnum >= P_FIRSTDATAKEY(opaque))
4046 : {
4047 : /*
4048 : * Non-pivot tuple should never be explicitly marked as a pivot
4049 : * tuple
4050 : */
4051 4030282 : if (BTreeTupleIsPivot(itup))
4052 0 : return false;
4053 :
4054 : /*
4055 : * Leaf tuples that are not the page high key (non-pivot tuples)
4056 : * should never be truncated. (Note that tupnatts must have been
4057 : * inferred, even with a posting list tuple, because only pivot
4058 : * tuples store tupnatts directly.)
4059 : */
4060 4030282 : return tupnatts == natts;
4061 : }
4062 : else
4063 : {
4064 : /*
4065 : * Rightmost page doesn't contain a page high key, so tuple was
4066 : * checked above as ordinary leaf tuple
4067 : */
4068 : Assert(!P_RIGHTMOST(opaque));
4069 :
4070 : /*
4071 : * !heapkeyspace high key tuple contains only key attributes. Note
4072 : * that tupnatts will only have been explicitly represented in
4073 : * !heapkeyspace indexes that happen to have non-key attributes.
4074 : */
4075 13232 : if (!heapkeyspace)
4076 0 : return tupnatts == nkeyatts;
4077 :
4078 : /* Use generic heapkeyspace pivot tuple handling */
4079 : }
4080 : }
4081 : else /* !P_ISLEAF(opaque) */
4082 : {
4083 14352 : if (offnum == P_FIRSTDATAKEY(opaque))
4084 : {
4085 : /*
4086 : * The first tuple on any internal page (possibly the first after
4087 : * its high key) is its negative infinity tuple. Negative
4088 : * infinity tuples are always truncated to zero attributes. They
4089 : * are a particular kind of pivot tuple.
4090 : */
4091 1114 : if (heapkeyspace)
4092 1114 : return tupnatts == 0;
4093 :
4094 : /*
4095 : * The number of attributes won't be explicitly represented if the
4096 : * negative infinity tuple was generated during a page split that
4097 : * occurred with a version of Postgres before v11. There must be
4098 : * a problem when there is an explicit representation that is
4099 : * non-zero, or when there is no explicit representation and the
4100 : * tuple is evidently not a pre-pg_upgrade tuple.
4101 : *
4102 : * Prior to v11, downlinks always had P_HIKEY as their offset.
4103 : * Accept that as an alternative indication of a valid
4104 : * !heapkeyspace negative infinity tuple.
4105 : */
4106 0 : return tupnatts == 0 ||
4107 0 : ItemPointerGetOffsetNumber(&(itup->t_tid)) == P_HIKEY;
4108 : }
4109 : else
4110 : {
4111 : /*
4112 : * !heapkeyspace downlink tuple with separator key contains only
4113 : * key attributes. Note that tupnatts will only have been
4114 : * explicitly represented in !heapkeyspace indexes that happen to
4115 : * have non-key attributes.
4116 : */
4117 13238 : if (!heapkeyspace)
4118 0 : return tupnatts == nkeyatts;
4119 :
4120 : /* Use generic heapkeyspace pivot tuple handling */
4121 : }
4122 : }
4123 :
4124 : /* Handle heapkeyspace pivot tuples (excluding minus infinity items) */
4125 : Assert(heapkeyspace);
4126 :
4127 : /*
4128 : * Explicit representation of the number of attributes is mandatory with
4129 : * heapkeyspace index pivot tuples, regardless of whether or not there are
4130 : * non-key attributes.
4131 : */
4132 26470 : if (!BTreeTupleIsPivot(itup))
4133 0 : return false;
4134 :
4135 : /* Pivot tuple should not use posting list representation (redundant) */
4136 26470 : if (BTreeTupleIsPosting(itup))
4137 0 : return false;
4138 :
4139 : /*
4140 : * Heap TID is a tiebreaker key attribute, so it cannot be untruncated
4141 : * when any other key attribute is truncated
4142 : */
4143 26470 : if (BTreeTupleGetHeapTID(itup) != NULL && tupnatts != nkeyatts)
4144 0 : return false;
4145 :
4146 : /*
4147 : * Pivot tuple must have at least one untruncated key attribute (minus
4148 : * infinity pivot tuples are the only exception). Pivot tuples can never
4149 : * represent that there is a value present for a key attribute that
4150 : * exceeds pg_index.indnkeyatts for the index.
4151 : */
4152 26470 : return tupnatts > 0 && tupnatts <= nkeyatts;
4153 : }
4154 :
4155 : /*
4156 : *
4157 : * _bt_check_third_page() -- check whether tuple fits on a btree page at all.
4158 : *
4159 : * We actually need to be able to fit three items on every page, so restrict
4160 : * any one item to 1/3 the per-page available space. Note that itemsz should
4161 : * not include the ItemId overhead.
4162 : *
4163 : * It might be useful to apply TOAST methods rather than throw an error here.
4164 : * Using out of line storage would break assumptions made by suffix truncation
4165 : * and by contrib/amcheck, though.
4166 : */
4167 : void
4168 264 : _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace,
4169 : Page page, IndexTuple newtup)
4170 : {
4171 : Size itemsz;
4172 : BTPageOpaque opaque;
4173 :
4174 264 : itemsz = MAXALIGN(IndexTupleSize(newtup));
4175 :
4176 : /* Double check item size against limit */
4177 264 : if (itemsz <= BTMaxItemSize)
4178 0 : return;
4179 :
4180 : /*
4181 : * Tuple is probably too large to fit on page, but it's possible that the
4182 : * index uses version 2 or version 3, or that page is an internal page, in
4183 : * which case a slightly higher limit applies.
4184 : */
4185 264 : if (!needheaptidspace && itemsz <= BTMaxItemSizeNoHeapTid)
4186 264 : return;
4187 :
4188 : /*
4189 : * Internal page insertions cannot fail here, because that would mean that
4190 : * an earlier leaf level insertion that should have failed didn't
4191 : */
4192 0 : opaque = BTPageGetOpaque(page);
4193 0 : if (!P_ISLEAF(opaque))
4194 0 : elog(ERROR, "cannot insert oversized tuple of size %zu on internal page of index \"%s\"",
4195 : itemsz, RelationGetRelationName(rel));
4196 :
4197 0 : ereport(ERROR,
4198 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
4199 : errmsg("index row size %zu exceeds btree version %u maximum %zu for index \"%s\"",
4200 : itemsz,
4201 : needheaptidspace ? BTREE_VERSION : BTREE_NOVAC_VERSION,
4202 : needheaptidspace ? BTMaxItemSize : BTMaxItemSizeNoHeapTid,
4203 : RelationGetRelationName(rel)),
4204 : errdetail("Index row references tuple (%u,%u) in relation \"%s\".",
4205 : ItemPointerGetBlockNumber(BTreeTupleGetHeapTID(newtup)),
4206 : ItemPointerGetOffsetNumber(BTreeTupleGetHeapTID(newtup)),
4207 : RelationGetRelationName(heap)),
4208 : errhint("Values larger than 1/3 of a buffer page cannot be indexed.\n"
4209 : "Consider a function index of an MD5 hash of the value, "
4210 : "or use full text indexing."),
4211 : errtableconstraint(heap, RelationGetRelationName(rel))));
4212 : }
4213 :
4214 : /*
4215 : * Are all attributes in rel "equality is image equality" attributes?
4216 : *
4217 : * We use each attribute's BTEQUALIMAGE_PROC opclass procedure. If any
4218 : * opclass either lacks a BTEQUALIMAGE_PROC procedure or returns false, we
4219 : * return false; otherwise we return true.
4220 : *
4221 : * Returned boolean value is stored in index metapage during index builds.
4222 : * Deduplication can only be used when we return true.
4223 : */
4224 : bool
4225 58208 : _bt_allequalimage(Relation rel, bool debugmessage)
4226 : {
4227 58208 : bool allequalimage = true;
4228 :
4229 : /* INCLUDE indexes can never support deduplication */
4230 58208 : if (IndexRelationGetNumberOfAttributes(rel) !=
4231 58208 : IndexRelationGetNumberOfKeyAttributes(rel))
4232 272 : return false;
4233 :
4234 152912 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(rel); i++)
4235 : {
4236 95494 : Oid opfamily = rel->rd_opfamily[i];
4237 95494 : Oid opcintype = rel->rd_opcintype[i];
4238 95494 : Oid collation = rel->rd_indcollation[i];
4239 : Oid equalimageproc;
4240 :
4241 95494 : equalimageproc = get_opfamily_proc(opfamily, opcintype, opcintype,
4242 : BTEQUALIMAGE_PROC);
4243 :
4244 : /*
4245 : * If there is no BTEQUALIMAGE_PROC then deduplication is assumed to
4246 : * be unsafe. Otherwise, actually call proc and see what it says.
4247 : */
4248 95494 : if (!OidIsValid(equalimageproc) ||
4249 95020 : !DatumGetBool(OidFunctionCall1Coll(equalimageproc, collation,
4250 : ObjectIdGetDatum(opcintype))))
4251 : {
4252 518 : allequalimage = false;
4253 518 : break;
4254 : }
4255 : }
4256 :
4257 57936 : if (debugmessage)
4258 : {
4259 49924 : if (allequalimage)
4260 49406 : elog(DEBUG1, "index \"%s\" can safely use deduplication",
4261 : RelationGetRelationName(rel));
4262 : else
4263 518 : elog(DEBUG1, "index \"%s\" cannot use deduplication",
4264 : RelationGetRelationName(rel));
4265 : }
4266 :
4267 57936 : return allequalimage;
4268 : }
|
