TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nbtree.c
4 : * Implementation of Lehman and Yao's btree management algorithm for
5 : * Postgres.
6 : *
7 : * NOTES
8 : * This file contains only the public interface routines.
9 : *
10 : *
11 : * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
12 : * Portions Copyright (c) 1994, Regents of the University of California
13 : *
14 : * IDENTIFICATION
15 : * src/backend/access/nbtree/nbtree.c
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 : #include "postgres.h"
20 :
21 : #include "access/nbtree.h"
22 : #include "access/nbtxlog.h"
23 : #include "access/relscan.h"
24 : #include "access/xlog.h"
25 : #include "access/xloginsert.h"
26 : #include "commands/progress.h"
27 : #include "commands/vacuum.h"
28 : #include "miscadmin.h"
29 : #include "nodes/execnodes.h"
30 : #include "pgstat.h"
31 : #include "postmaster/autovacuum.h"
32 : #include "storage/condition_variable.h"
33 : #include "storage/indexfsm.h"
34 : #include "storage/ipc.h"
35 : #include "storage/lmgr.h"
36 : #include "storage/smgr.h"
37 : #include "utils/builtins.h"
38 : #include "utils/index_selfuncs.h"
39 : #include "utils/memutils.h"
40 :
41 :
42 : /*
43 : * BTPARALLEL_NOT_INITIALIZED indicates that the scan has not started.
44 : *
45 : * BTPARALLEL_ADVANCING indicates that some process is advancing the scan to
46 : * a new page; others must wait.
47 : *
48 : * BTPARALLEL_IDLE indicates that no backend is currently advancing the scan
49 : * to a new page; some process can start doing that.
50 : *
51 : * BTPARALLEL_DONE indicates that the scan is complete (including error exit).
52 : * We reach this state once for every distinct combination of array keys.
53 : */
54 : typedef enum
55 : {
56 : BTPARALLEL_NOT_INITIALIZED,
57 : BTPARALLEL_ADVANCING,
58 : BTPARALLEL_IDLE,
59 : BTPARALLEL_DONE
60 : } BTPS_State;
61 :
62 : /*
63 : * BTParallelScanDescData contains btree specific shared information required
64 : * for parallel scan.
65 : */
66 : typedef struct BTParallelScanDescData
67 : {
68 : BlockNumber btps_scanPage; /* latest or next page to be scanned */
69 : BTPS_State btps_pageStatus; /* indicates whether next page is
70 : * available for scan. see above for
71 : * possible states of parallel scan. */
72 : int btps_arrayKeyCount; /* count indicating number of array scan
73 : * keys processed by parallel scan */
74 : slock_t btps_mutex; /* protects above variables */
75 : ConditionVariable btps_cv; /* used to synchronize parallel scan */
76 : } BTParallelScanDescData;
77 :
78 : typedef struct BTParallelScanDescData *BTParallelScanDesc;
79 :
80 :
81 : static void btvacuumscan(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
82 : IndexBulkDeleteCallback callback, void *callback_state,
83 : BTCycleId cycleid);
84 : static void btvacuumpage(BTVacState *vstate, BlockNumber scanblkno);
85 : static BTVacuumPosting btreevacuumposting(BTVacState *vstate,
86 : IndexTuple posting,
87 : OffsetNumber updatedoffset,
88 : int *nremaining);
89 :
90 :
91 : /*
92 : * Btree handler function: return IndexAmRoutine with access method parameters
93 : * and callbacks.
94 : */
95 : Datum
96 CBC 1158799 : bthandler(PG_FUNCTION_ARGS)
97 : {
98 1158799 : IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
99 :
100 1158799 : amroutine->amstrategies = BTMaxStrategyNumber;
101 1158799 : amroutine->amsupport = BTNProcs;
102 1158799 : amroutine->amoptsprocnum = BTOPTIONS_PROC;
103 1158799 : amroutine->amcanorder = true;
104 1158799 : amroutine->amcanorderbyop = false;
105 1158799 : amroutine->amcanbackward = true;
106 1158799 : amroutine->amcanunique = true;
107 1158799 : amroutine->amcanmulticol = true;
108 1158799 : amroutine->amoptionalkey = true;
109 1158799 : amroutine->amsearcharray = true;
110 1158799 : amroutine->amsearchnulls = true;
111 1158799 : amroutine->amstorage = false;
112 1158799 : amroutine->amclusterable = true;
113 1158799 : amroutine->ampredlocks = true;
114 1158799 : amroutine->amcanparallel = true;
115 1158799 : amroutine->amcaninclude = true;
116 1158799 : amroutine->amusemaintenanceworkmem = false;
117 GNC 1158799 : amroutine->amsummarizing = false;
118 CBC 1158799 : amroutine->amparallelvacuumoptions =
119 ECB : VACUUM_OPTION_PARALLEL_BULKDEL | VACUUM_OPTION_PARALLEL_COND_CLEANUP;
120 GIC 1158799 : amroutine->amkeytype = InvalidOid;
121 ECB :
122 GIC 1158799 : amroutine->ambuild = btbuild;
123 CBC 1158799 : amroutine->ambuildempty = btbuildempty;
124 1158799 : amroutine->aminsert = btinsert;
125 1158799 : amroutine->ambulkdelete = btbulkdelete;
126 1158799 : amroutine->amvacuumcleanup = btvacuumcleanup;
127 1158799 : amroutine->amcanreturn = btcanreturn;
128 1158799 : amroutine->amcostestimate = btcostestimate;
129 1158799 : amroutine->amoptions = btoptions;
130 1158799 : amroutine->amproperty = btproperty;
131 1158799 : amroutine->ambuildphasename = btbuildphasename;
132 1158799 : amroutine->amvalidate = btvalidate;
133 1158799 : amroutine->amadjustmembers = btadjustmembers;
134 1158799 : amroutine->ambeginscan = btbeginscan;
135 1158799 : amroutine->amrescan = btrescan;
136 1158799 : amroutine->amgettuple = btgettuple;
137 1158799 : amroutine->amgetbitmap = btgetbitmap;
138 1158799 : amroutine->amendscan = btendscan;
139 1158799 : amroutine->ammarkpos = btmarkpos;
140 1158799 : amroutine->amrestrpos = btrestrpos;
141 1158799 : amroutine->amestimateparallelscan = btestimateparallelscan;
142 1158799 : amroutine->aminitparallelscan = btinitparallelscan;
143 1158799 : amroutine->amparallelrescan = btparallelrescan;
144 ECB :
145 GIC 1158799 : PG_RETURN_POINTER(amroutine);
146 ECB : }
147 :
148 : /*
149 : * btbuildempty() -- build an empty btree index in the initialization fork
150 : */
151 : void
152 GIC 62 : btbuildempty(Relation index)
153 ECB : {
154 : Page metapage;
155 :
156 : /* Construct metapage. */
157 GNC 62 : metapage = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
158 CBC 62 : _bt_initmetapage(metapage, P_NONE, 0, _bt_allequalimage(index, false));
159 ECB :
160 : /*
161 : * Write the page and log it. It might seem that an immediate sync would
162 : * be sufficient to guarantee that the file exists on disk, but recovery
163 : * itself might remove it while replaying, for example, an
164 : * XLOG_DBASE_CREATE* or XLOG_TBLSPC_CREATE record. Therefore, we need
165 : * this even when wal_level=minimal.
166 : */
167 GIC 62 : PageSetChecksumInplace(metapage, BTREE_METAPAGE);
168 CBC 62 : smgrwrite(RelationGetSmgr(index), INIT_FORKNUM, BTREE_METAPAGE,
169 : metapage, true);
170 GNC 62 : log_newpage(&RelationGetSmgr(index)->smgr_rlocator.locator, INIT_FORKNUM,
171 ECB : BTREE_METAPAGE, metapage, true);
172 :
173 : /*
174 : * An immediate sync is required even if we xlog'd the page, because the
175 : * write did not go through shared_buffers and therefore a concurrent
176 : * checkpoint may have moved the redo pointer past our xlog record.
177 : */
178 GIC 62 : smgrimmedsync(RelationGetSmgr(index), INIT_FORKNUM);
179 CBC 62 : }
180 ECB :
181 : /*
182 : * btinsert() -- insert an index tuple into a btree.
183 : *
184 : * Descend the tree recursively, find the appropriate location for our
185 : * new tuple, and put it there.
186 : */
187 : bool
188 GIC 7167159 : btinsert(Relation rel, Datum *values, bool *isnull,
189 ECB : ItemPointer ht_ctid, Relation heapRel,
190 : IndexUniqueCheck checkUnique,
191 : bool indexUnchanged,
192 : IndexInfo *indexInfo)
193 : {
194 : bool result;
195 : IndexTuple itup;
196 :
197 : /* generate an index tuple */
198 GIC 7167159 : itup = index_form_tuple(RelationGetDescr(rel), values, isnull);
199 CBC 7167159 : itup->t_tid = *ht_ctid;
200 ECB :
201 GIC 7167159 : result = _bt_doinsert(rel, itup, checkUnique, indexUnchanged, heapRel);
202 ECB :
203 GIC 7166901 : pfree(itup);
204 ECB :
205 GIC 7166901 : return result;
206 ECB : }
207 :
208 : /*
209 : * btgettuple() -- Get the next tuple in the scan.
210 : */
211 : bool
212 GIC 21029716 : btgettuple(IndexScanDesc scan, ScanDirection dir)
213 ECB : {
214 GIC 21029716 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
215 ECB : bool res;
216 :
217 : /* btree indexes are never lossy */
218 GIC 21029716 : scan->xs_recheck = false;
219 ECB :
220 : /*
221 : * If we have any array keys, initialize them during first call for a
222 : * scan. We can't do this in btrescan because we don't know the scan
223 : * direction at that time.
224 : */
225 GIC 21029716 : if (so->numArrayKeys && !BTScanPosIsValid(so->currPos))
226 ECB : {
227 : /* punt if we have any unsatisfiable array keys */
228 GIC 88 : if (so->numArrayKeys < 0)
229 LBC 0 : return false;
230 EUB :
231 GIC 88 : _bt_start_array_keys(scan, dir);
232 ECB : }
233 :
234 : /* This loop handles advancing to the next array elements, if any */
235 : do
236 : {
237 : /*
238 : * If we've already initialized this scan, we can just advance it in
239 : * the appropriate direction. If we haven't done so yet, we call
240 : * _bt_first() to get the first item in the scan.
241 : */
242 GIC 21029962 : if (!BTScanPosIsValid(so->currPos))
243 CBC 9080657 : res = _bt_first(scan, dir);
244 ECB : else
245 : {
246 : /*
247 : * Check to see if we should kill the previously-fetched tuple.
248 : */
249 GIC 11949305 : if (scan->kill_prior_tuple)
250 ECB : {
251 : /*
252 : * Yes, remember it for later. (We'll deal with all such
253 : * tuples at once right before leaving the index page.) The
254 : * test for numKilled overrun is not just paranoia: if the
255 : * caller reverses direction in the indexscan then the same
256 : * item might get entered multiple times. It's not worth
257 : * trying to optimize that, so we don't detect it, but instead
258 : * just forget any excess entries.
259 : */
260 GIC 263130 : if (so->killedItems == NULL)
261 CBC 117431 : so->killedItems = (int *)
262 117431 : palloc(MaxTIDsPerBTreePage * sizeof(int));
263 263130 : if (so->numKilled < MaxTIDsPerBTreePage)
264 263130 : so->killedItems[so->numKilled++] = so->currPos.itemIndex;
265 ECB : }
266 :
267 : /*
268 : * Now continue the scan.
269 : */
270 GIC 11949305 : res = _bt_next(scan, dir);
271 ECB : }
272 :
273 : /* If we have a tuple, return it ... */
274 GIC 21029962 : if (res)
275 CBC 15809739 : break;
276 ECB : /* ... otherwise see if we have more array keys to deal with */
277 GIC 5220223 : } while (so->numArrayKeys && _bt_advance_array_keys(scan, dir));
278 ECB :
279 GIC 21029716 : return res;
280 ECB : }
281 :
282 : /*
283 : * btgetbitmap() -- gets all matching tuples, and adds them to a bitmap
284 : */
285 : int64
286 GIC 5496 : btgetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
287 ECB : {
288 GIC 5496 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
289 CBC 5496 : int64 ntids = 0;
290 ECB : ItemPointer heapTid;
291 :
292 : /*
293 : * If we have any array keys, initialize them.
294 : */
295 GIC 5496 : if (so->numArrayKeys)
296 ECB : {
297 : /* punt if we have any unsatisfiable array keys */
298 GIC 222 : if (so->numArrayKeys < 0)
299 LBC 0 : return ntids;
300 EUB :
301 GIC 222 : _bt_start_array_keys(scan, ForwardScanDirection);
302 ECB : }
303 :
304 : /* This loop handles advancing to the next array elements, if any */
305 : do
306 : {
307 : /* Fetch the first page & tuple */
308 GIC 6838 : if (_bt_first(scan, ForwardScanDirection))
309 ECB : {
310 : /* Save tuple ID, and continue scanning */
311 GIC 5466 : heapTid = &scan->xs_heaptid;
312 CBC 5466 : tbm_add_tuples(tbm, heapTid, 1, false);
313 5466 : ntids++;
314 ECB :
315 : for (;;)
316 : {
317 : /*
318 : * Advance to next tuple within page. This is the same as the
319 : * easy case in _bt_next().
320 : */
321 GIC 850887 : if (++so->currPos.itemIndex > so->currPos.lastItem)
322 ECB : {
323 : /* let _bt_next do the heavy lifting */
324 GIC 7583 : if (!_bt_next(scan, ForwardScanDirection))
325 CBC 5466 : break;
326 ECB : }
327 :
328 : /* Save tuple ID, and continue scanning */
329 GIC 845421 : heapTid = &so->currPos.items[so->currPos.itemIndex].heapTid;
330 CBC 845421 : tbm_add_tuples(tbm, heapTid, 1, false);
331 845421 : ntids++;
332 ECB : }
333 : }
334 : /* Now see if we have more array keys to deal with */
335 GIC 6838 : } while (so->numArrayKeys && _bt_advance_array_keys(scan, ForwardScanDirection));
336 ECB :
337 GIC 5496 : return ntids;
338 ECB : }
339 :
340 : /*
341 : * btbeginscan() -- start a scan on a btree index
342 : */
343 : IndexScanDesc
344 GIC 8901539 : btbeginscan(Relation rel, int nkeys, int norderbys)
345 ECB : {
346 : IndexScanDesc scan;
347 : BTScanOpaque so;
348 :
349 : /* no order by operators allowed */
350 GIC 8901539 : Assert(norderbys == 0);
351 ECB :
352 : /* get the scan */
353 GIC 8901539 : scan = RelationGetIndexScan(rel, nkeys, norderbys);
354 ECB :
355 : /* allocate private workspace */
356 GIC 8901539 : so = (BTScanOpaque) palloc(sizeof(BTScanOpaqueData));
357 CBC 8901539 : BTScanPosInvalidate(so->currPos);
358 8901539 : BTScanPosInvalidate(so->markPos);
359 8901539 : if (scan->numberOfKeys > 0)
360 8896655 : so->keyData = (ScanKey) palloc(scan->numberOfKeys * sizeof(ScanKeyData));
361 ECB : else
362 GIC 4884 : so->keyData = NULL;
363 ECB :
364 GIC 8901539 : so->arrayKeyData = NULL; /* assume no array keys for now */
365 CBC 8901539 : so->numArrayKeys = 0;
366 8901539 : so->arrayKeys = NULL;
367 8901539 : so->arrayContext = NULL;
368 ECB :
369 GIC 8901539 : so->killedItems = NULL; /* until needed */
370 CBC 8901539 : so->numKilled = 0;
371 ECB :
372 : /*
373 : * We don't know yet whether the scan will be index-only, so we do not
374 : * allocate the tuple workspace arrays until btrescan. However, we set up
375 : * scan->xs_itupdesc whether we'll need it or not, since that's so cheap.
376 : */
377 GIC 8901539 : so->currTuples = so->markTuples = NULL;
378 ECB :
379 GIC 8901539 : scan->xs_itupdesc = RelationGetDescr(rel);
380 ECB :
381 GIC 8901539 : scan->opaque = so;
382 ECB :
383 GIC 8901539 : return scan;
384 ECB : }
385 :
386 : /*
387 : * btrescan() -- rescan an index relation
388 : */
389 : void
390 GIC 9087654 : btrescan(IndexScanDesc scan, ScanKey scankey, int nscankeys,
391 ECB : ScanKey orderbys, int norderbys)
392 : {
393 GIC 9087654 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
394 ECB :
395 : /* we aren't holding any read locks, but gotta drop the pins */
396 GIC 9087654 : if (BTScanPosIsValid(so->currPos))
397 ECB : {
398 : /* Before leaving current page, deal with any killed items */
399 GIC 24995 : if (so->numKilled > 0)
400 CBC 152 : _bt_killitems(scan);
401 24995 : BTScanPosUnpinIfPinned(so->currPos);
402 24995 : BTScanPosInvalidate(so->currPos);
403 ECB : }
404 :
405 GIC 9087654 : so->markItemIndex = -1;
406 CBC 9087654 : so->arrayKeyCount = 0;
407 9087654 : BTScanPosUnpinIfPinned(so->markPos);
408 9087654 : BTScanPosInvalidate(so->markPos);
409 ECB :
410 : /*
411 : * Allocate tuple workspace arrays, if needed for an index-only scan and
412 : * not already done in a previous rescan call. To save on palloc
413 : * overhead, both workspaces are allocated as one palloc block; only this
414 : * function and btendscan know that.
415 : *
416 : * NOTE: this data structure also makes it safe to return data from a
417 : * "name" column, even though btree name_ops uses an underlying storage
418 : * datatype of cstring. The risk there is that "name" is supposed to be
419 : * padded to NAMEDATALEN, but the actual index tuple is probably shorter.
420 : * However, since we only return data out of tuples sitting in the
421 : * currTuples array, a fetch of NAMEDATALEN bytes can at worst pull some
422 : * data out of the markTuples array --- running off the end of memory for
423 : * a SIGSEGV is not possible. Yeah, this is ugly as sin, but it beats
424 : * adding special-case treatment for name_ops elsewhere.
425 : */
426 GIC 9087654 : if (scan->xs_want_itup && so->currTuples == NULL)
427 ECB : {
428 GIC 50124 : so->currTuples = (char *) palloc(BLCKSZ * 2);
429 CBC 50124 : so->markTuples = so->currTuples + BLCKSZ;
430 ECB : }
431 :
432 : /*
433 : * Reset the scan keys
434 : */
435 GIC 9087654 : if (scankey && scan->numberOfKeys > 0)
436 CBC 9082744 : memmove(scan->keyData,
437 ECB : scankey,
438 GIC 9082744 : scan->numberOfKeys * sizeof(ScanKeyData));
439 CBC 9087654 : so->numberOfKeys = 0; /* until _bt_preprocess_keys sets it */
440 ECB :
441 : /* If any keys are SK_SEARCHARRAY type, set up array-key info */
442 GIC 9087654 : _bt_preprocess_array_keys(scan);
443 CBC 9087654 : }
444 ECB :
445 : /*
446 : * btendscan() -- close down a scan
447 : */
448 : void
449 GIC 8900856 : btendscan(IndexScanDesc scan)
450 ECB : {
451 GIC 8900856 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
452 ECB :
453 : /* we aren't holding any read locks, but gotta drop the pins */
454 GIC 8900856 : if (BTScanPosIsValid(so->currPos))
455 ECB : {
456 : /* Before leaving current page, deal with any killed items */
457 GIC 3835001 : if (so->numKilled > 0)
458 CBC 78401 : _bt_killitems(scan);
459 3835001 : BTScanPosUnpinIfPinned(so->currPos);
460 ECB : }
461 :
462 GIC 8900856 : so->markItemIndex = -1;
463 CBC 8900856 : BTScanPosUnpinIfPinned(so->markPos);
464 ECB :
465 : /* No need to invalidate positions, the RAM is about to be freed. */
466 :
467 : /* Release storage */
468 GIC 8900856 : if (so->keyData != NULL)
469 CBC 8896031 : pfree(so->keyData);
470 ECB : /* so->arrayKeyData and so->arrayKeys are in arrayContext */
471 GIC 8900856 : if (so->arrayContext != NULL)
472 CBC 314 : MemoryContextDelete(so->arrayContext);
473 8900856 : if (so->killedItems != NULL)
474 117406 : pfree(so->killedItems);
475 8900856 : if (so->currTuples != NULL)
476 50058 : pfree(so->currTuples);
477 ECB : /* so->markTuples should not be pfree'd, see btrescan */
478 GIC 8900856 : pfree(so);
479 CBC 8900856 : }
480 ECB :
481 : /*
482 : * btmarkpos() -- save current scan position
483 : */
484 : void
485 GIC 65029 : btmarkpos(IndexScanDesc scan)
486 ECB : {
487 GIC 65029 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
488 ECB :
489 : /* There may be an old mark with a pin (but no lock). */
490 GIC 65029 : BTScanPosUnpinIfPinned(so->markPos);
491 ECB :
492 : /*
493 : * Just record the current itemIndex. If we later step to next page
494 : * before releasing the marked position, _bt_steppage makes a full copy of
495 : * the currPos struct in markPos. If (as often happens) the mark is moved
496 : * before we leave the page, we don't have to do that work.
497 : */
498 GIC 65029 : if (BTScanPosIsValid(so->currPos))
499 CBC 65029 : so->markItemIndex = so->currPos.itemIndex;
500 ECB : else
501 : {
502 UIC 0 : BTScanPosInvalidate(so->markPos);
503 UBC 0 : so->markItemIndex = -1;
504 EUB : }
505 :
506 : /* Also record the current positions of any array keys */
507 GIC 65029 : if (so->numArrayKeys)
508 CBC 3 : _bt_mark_array_keys(scan);
509 65029 : }
510 ECB :
511 : /*
512 : * btrestrpos() -- restore scan to last saved position
513 : */
514 : void
515 GIC 27015 : btrestrpos(IndexScanDesc scan)
516 ECB : {
517 GIC 27015 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
518 ECB :
519 : /* Restore the marked positions of any array keys */
520 GIC 27015 : if (so->numArrayKeys)
521 CBC 3 : _bt_restore_array_keys(scan);
522 ECB :
523 GIC 27015 : if (so->markItemIndex >= 0)
524 ECB : {
525 : /*
526 : * The scan has never moved to a new page since the last mark. Just
527 : * restore the itemIndex.
528 : *
529 : * NB: In this case we can't count on anything in so->markPos to be
530 : * accurate.
531 : */
532 GIC 26961 : so->currPos.itemIndex = so->markItemIndex;
533 ECB : }
534 : else
535 : {
536 : /*
537 : * The scan moved to a new page after last mark or restore, and we are
538 : * now restoring to the marked page. We aren't holding any read
539 : * locks, but if we're still holding the pin for the current position,
540 : * we must drop it.
541 : */
542 GIC 54 : if (BTScanPosIsValid(so->currPos))
543 ECB : {
544 : /* Before leaving current page, deal with any killed items */
545 GIC 54 : if (so->numKilled > 0)
546 LBC 0 : _bt_killitems(scan);
547 GBC 54 : BTScanPosUnpinIfPinned(so->currPos);
548 ECB : }
549 :
550 GIC 54 : if (BTScanPosIsValid(so->markPos))
551 ECB : {
552 : /* bump pin on mark buffer for assignment to current buffer */
553 GIC 54 : if (BTScanPosIsPinned(so->markPos))
554 LBC 0 : IncrBufferRefCount(so->markPos.buf);
555 GBC 54 : memcpy(&so->currPos, &so->markPos,
556 ECB : offsetof(BTScanPosData, items[1]) +
557 GIC 54 : so->markPos.lastItem * sizeof(BTScanPosItem));
558 CBC 54 : if (so->currTuples)
559 LBC 0 : memcpy(so->currTuples, so->markTuples,
560 UBC 0 : so->markPos.nextTupleOffset);
561 EUB : }
562 : else
563 UIC 0 : BTScanPosInvalidate(so->currPos);
564 EUB : }
565 GIC 27015 : }
566 ECB :
567 : /*
568 : * btestimateparallelscan -- estimate storage for BTParallelScanDescData
569 : */
570 : Size
571 GIC 26 : btestimateparallelscan(void)
572 ECB : {
573 GIC 26 : return sizeof(BTParallelScanDescData);
574 ECB : }
575 :
576 : /*
577 : * btinitparallelscan -- initialize BTParallelScanDesc for parallel btree scan
578 : */
579 : void
580 GIC 26 : btinitparallelscan(void *target)
581 ECB : {
582 GIC 26 : BTParallelScanDesc bt_target = (BTParallelScanDesc) target;
583 ECB :
584 GIC 26 : SpinLockInit(&bt_target->btps_mutex);
585 CBC 26 : bt_target->btps_scanPage = InvalidBlockNumber;
586 26 : bt_target->btps_pageStatus = BTPARALLEL_NOT_INITIALIZED;
587 26 : bt_target->btps_arrayKeyCount = 0;
588 26 : ConditionVariableInit(&bt_target->btps_cv);
589 26 : }
590 ECB :
591 : /*
592 : * btparallelrescan() -- reset parallel scan
593 : */
594 : void
595 GIC 12 : btparallelrescan(IndexScanDesc scan)
596 ECB : {
597 : BTParallelScanDesc btscan;
598 GIC 12 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
599 ECB :
600 GIC 12 : Assert(parallel_scan);
601 ECB :
602 GIC 12 : btscan = (BTParallelScanDesc) OffsetToPointer((void *) parallel_scan,
603 ECB : parallel_scan->ps_offset);
604 :
605 : /*
606 : * In theory, we don't need to acquire the spinlock here, because there
607 : * shouldn't be any other workers running at this point, but we do so for
608 : * consistency.
609 : */
610 GIC 12 : SpinLockAcquire(&btscan->btps_mutex);
611 CBC 12 : btscan->btps_scanPage = InvalidBlockNumber;
612 12 : btscan->btps_pageStatus = BTPARALLEL_NOT_INITIALIZED;
613 12 : btscan->btps_arrayKeyCount = 0;
614 12 : SpinLockRelease(&btscan->btps_mutex);
615 12 : }
616 ECB :
617 : /*
618 : * _bt_parallel_seize() -- Begin the process of advancing the scan to a new
619 : * page. Other scans must wait until we call _bt_parallel_release()
620 : * or _bt_parallel_done().
621 : *
622 : * The return value is true if we successfully seized the scan and false
623 : * if we did not. The latter case occurs if no pages remain for the current
624 : * set of scankeys.
625 : *
626 : * If the return value is true, *pageno returns the next or current page
627 : * of the scan (depending on the scan direction). An invalid block number
628 : * means the scan hasn't yet started, and P_NONE means we've reached the end.
629 : * The first time a participating process reaches the last page, it will return
630 : * true and set *pageno to P_NONE; after that, further attempts to seize the
631 : * scan will return false.
632 : *
633 : * Callers should ignore the value of pageno if the return value is false.
634 : */
635 : bool
636 GIC 828 : _bt_parallel_seize(IndexScanDesc scan, BlockNumber *pageno)
637 ECB : {
638 GIC 828 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
639 ECB : BTPS_State pageStatus;
640 GIC 828 : bool exit_loop = false;
641 CBC 828 : bool status = true;
642 828 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
643 ECB : BTParallelScanDesc btscan;
644 :
645 GIC 828 : *pageno = P_NONE;
646 ECB :
647 GIC 828 : btscan = (BTParallelScanDesc) OffsetToPointer((void *) parallel_scan,
648 ECB : parallel_scan->ps_offset);
649 :
650 : while (1)
651 : {
652 GIC 838 : SpinLockAcquire(&btscan->btps_mutex);
653 CBC 838 : pageStatus = btscan->btps_pageStatus;
654 ECB :
655 GIC 838 : if (so->arrayKeyCount < btscan->btps_arrayKeyCount)
656 ECB : {
657 : /* Parallel scan has already advanced to a new set of scankeys. */
658 UIC 0 : status = false;
659 EUB : }
660 GIC 838 : else if (pageStatus == BTPARALLEL_DONE)
661 ECB : {
662 : /*
663 : * We're done with this set of scankeys. This may be the end, or
664 : * there could be more sets to try.
665 : */
666 GIC 146 : status = false;
667 ECB : }
668 GIC 692 : else if (pageStatus != BTPARALLEL_ADVANCING)
669 ECB : {
670 : /*
671 : * We have successfully seized control of the scan for the purpose
672 : * of advancing it to a new page!
673 : */
674 GIC 682 : btscan->btps_pageStatus = BTPARALLEL_ADVANCING;
675 CBC 682 : *pageno = btscan->btps_scanPage;
676 682 : exit_loop = true;
677 ECB : }
678 GIC 838 : SpinLockRelease(&btscan->btps_mutex);
679 CBC 838 : if (exit_loop || !status)
680 ECB : break;
681 GIC 10 : ConditionVariableSleep(&btscan->btps_cv, WAIT_EVENT_BTREE_PAGE);
682 ECB : }
683 GIC 828 : ConditionVariableCancelSleep();
684 ECB :
685 GIC 828 : return status;
686 ECB : }
687 :
688 : /*
689 : * _bt_parallel_release() -- Complete the process of advancing the scan to a
690 : * new page. We now have the new value btps_scanPage; some other backend
691 : * can now begin advancing the scan.
692 : */
693 : void
694 GIC 644 : _bt_parallel_release(IndexScanDesc scan, BlockNumber scan_page)
695 ECB : {
696 GIC 644 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
697 ECB : BTParallelScanDesc btscan;
698 :
699 GIC 644 : btscan = (BTParallelScanDesc) OffsetToPointer((void *) parallel_scan,
700 ECB : parallel_scan->ps_offset);
701 :
702 GIC 644 : SpinLockAcquire(&btscan->btps_mutex);
703 CBC 644 : btscan->btps_scanPage = scan_page;
704 644 : btscan->btps_pageStatus = BTPARALLEL_IDLE;
705 644 : SpinLockRelease(&btscan->btps_mutex);
706 644 : ConditionVariableSignal(&btscan->btps_cv);
707 644 : }
708 ECB :
709 : /*
710 : * _bt_parallel_done() -- Mark the parallel scan as complete.
711 : *
712 : * When there are no pages left to scan, this function should be called to
713 : * notify other workers. Otherwise, they might wait forever for the scan to
714 : * advance to the next page.
715 : */
716 : void
717 GIC 5227696 : _bt_parallel_done(IndexScanDesc scan)
718 ECB : {
719 GIC 5227696 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
720 CBC 5227696 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
721 ECB : BTParallelScanDesc btscan;
722 GIC 5227696 : bool status_changed = false;
723 ECB :
724 : /* Do nothing, for non-parallel scans */
725 GIC 5227696 : if (parallel_scan == NULL)
726 CBC 5227658 : return;
727 ECB :
728 GIC 38 : btscan = (BTParallelScanDesc) OffsetToPointer((void *) parallel_scan,
729 ECB : parallel_scan->ps_offset);
730 :
731 : /*
732 : * Mark the parallel scan as done for this combination of scan keys,
733 : * unless some other process already did so. See also
734 : * _bt_advance_array_keys.
735 : */
736 GIC 38 : SpinLockAcquire(&btscan->btps_mutex);
737 CBC 38 : if (so->arrayKeyCount >= btscan->btps_arrayKeyCount &&
738 38 : btscan->btps_pageStatus != BTPARALLEL_DONE)
739 ECB : {
740 GIC 38 : btscan->btps_pageStatus = BTPARALLEL_DONE;
741 CBC 38 : status_changed = true;
742 ECB : }
743 GIC 38 : SpinLockRelease(&btscan->btps_mutex);
744 ECB :
745 : /* wake up all the workers associated with this parallel scan */
746 GIC 38 : if (status_changed)
747 CBC 38 : ConditionVariableBroadcast(&btscan->btps_cv);
748 ECB : }
749 :
750 : /*
751 : * _bt_parallel_advance_array_keys() -- Advances the parallel scan for array
752 : * keys.
753 : *
754 : * Updates the count of array keys processed for both local and parallel
755 : * scans.
756 : */
757 : void
758 UIC 0 : _bt_parallel_advance_array_keys(IndexScanDesc scan)
759 EUB : {
760 UIC 0 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
761 UBC 0 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
762 EUB : BTParallelScanDesc btscan;
763 :
764 UIC 0 : btscan = (BTParallelScanDesc) OffsetToPointer((void *) parallel_scan,
765 EUB : parallel_scan->ps_offset);
766 :
767 UIC 0 : so->arrayKeyCount++;
768 UBC 0 : SpinLockAcquire(&btscan->btps_mutex);
769 0 : if (btscan->btps_pageStatus == BTPARALLEL_DONE)
770 EUB : {
771 UIC 0 : btscan->btps_scanPage = InvalidBlockNumber;
772 UBC 0 : btscan->btps_pageStatus = BTPARALLEL_NOT_INITIALIZED;
773 0 : btscan->btps_arrayKeyCount++;
774 EUB : }
775 UIC 0 : SpinLockRelease(&btscan->btps_mutex);
776 UBC 0 : }
777 EUB :
778 : /*
779 : * Bulk deletion of all index entries pointing to a set of heap tuples.
780 : * The set of target tuples is specified via a callback routine that tells
781 : * whether any given heap tuple (identified by ItemPointer) is being deleted.
782 : *
783 : * Result: a palloc'd struct containing statistical info for VACUUM displays.
784 : */
785 : IndexBulkDeleteResult *
786 GIC 4011 : btbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
787 ECB : IndexBulkDeleteCallback callback, void *callback_state)
788 : {
789 GIC 4011 : Relation rel = info->index;
790 ECB : BTCycleId cycleid;
791 :
792 : /* allocate stats if first time through, else re-use existing struct */
793 GIC 4011 : if (stats == NULL)
794 CBC 4011 : stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
795 ECB :
796 : /* Establish the vacuum cycle ID to use for this scan */
797 : /* The ENSURE stuff ensures we clean up shared memory on failure */
798 GIC 4011 : PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel));
799 ECB : {
800 GIC 4011 : cycleid = _bt_start_vacuum(rel);
801 ECB :
802 GIC 4011 : btvacuumscan(info, stats, callback, callback_state, cycleid);
803 ECB : }
804 GIC 4011 : PG_END_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel));
805 CBC 4011 : _bt_end_vacuum(rel);
806 ECB :
807 GIC 4011 : return stats;
808 ECB : }
809 :
810 : /*
811 : * Post-VACUUM cleanup.
812 : *
813 : * Result: a palloc'd struct containing statistical info for VACUUM displays.
814 : */
815 : IndexBulkDeleteResult *
816 GIC 94456 : btvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
817 ECB : {
818 : BlockNumber num_delpages;
819 :
820 : /* No-op in ANALYZE ONLY mode */
821 GIC 94456 : if (info->analyze_only)
822 CBC 39995 : return stats;
823 ECB :
824 : /*
825 : * If btbulkdelete was called, we need not do anything (we just maintain
826 : * the information used within _bt_vacuum_needs_cleanup() by calling
827 : * _bt_set_cleanup_info() below).
828 : *
829 : * If btbulkdelete was _not_ called, then we have a choice to make: we
830 : * must decide whether or not a btvacuumscan() call is needed now (i.e.
831 : * whether the ongoing VACUUM operation can entirely avoid a physical scan
832 : * of the index). A call to _bt_vacuum_needs_cleanup() decides it for us
833 : * now.
834 : */
835 GIC 54461 : if (stats == NULL)
836 ECB : {
837 : /* Check if VACUUM operation can entirely avoid btvacuumscan() call */
838 GNC 50713 : if (!_bt_vacuum_needs_cleanup(info->index, info->heaprel))
839 CBC 50708 : return NULL;
840 ECB :
841 : /*
842 : * Since we aren't going to actually delete any leaf items, there's no
843 : * need to go through all the vacuum-cycle-ID pushups here.
844 : *
845 : * Posting list tuples are a source of inaccuracy for cleanup-only
846 : * scans. btvacuumscan() will assume that the number of index tuples
847 : * from each page can be used as num_index_tuples, even though
848 : * num_index_tuples is supposed to represent the number of TIDs in the
849 : * index. This naive approach can underestimate the number of tuples
850 : * in the index significantly.
851 : *
852 : * We handle the problem by making num_index_tuples an estimate in
853 : * cleanup-only case.
854 : */
855 GIC 5 : stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
856 CBC 5 : btvacuumscan(info, stats, NULL, NULL, 0);
857 5 : stats->estimated_count = true;
858 ECB : }
859 :
860 : /*
861 : * Maintain num_delpages value in metapage for _bt_vacuum_needs_cleanup().
862 : *
863 : * num_delpages is the number of deleted pages now in the index that were
864 : * not safe to place in the FSM to be recycled just yet. num_delpages is
865 : * greater than 0 only when _bt_pagedel() actually deleted pages during
866 : * our call to btvacuumscan(). Even then, _bt_pendingfsm_finalize() must
867 : * have failed to place any newly deleted pages in the FSM just moments
868 : * ago. (Actually, there are edge cases where recycling of the current
869 : * VACUUM's newly deleted pages does not even become safe by the time the
870 : * next VACUUM comes around. See nbtree/README.)
871 : */
872 GIC 3753 : Assert(stats->pages_deleted >= stats->pages_free);
873 CBC 3753 : num_delpages = stats->pages_deleted - stats->pages_free;
874 GNC 3753 : _bt_set_cleanup_info(info->index, info->heaprel, num_delpages);
875 ECB :
876 : /*
877 : * It's quite possible for us to be fooled by concurrent page splits into
878 : * double-counting some index tuples, so disbelieve any total that exceeds
879 : * the underlying heap's count ... if we know that accurately. Otherwise
880 : * this might just make matters worse.
881 : */
882 GIC 3753 : if (!info->estimated_count)
883 ECB : {
884 GIC 3731 : if (stats->num_index_tuples > info->num_heap_tuples)
885 CBC 7 : stats->num_index_tuples = info->num_heap_tuples;
886 ECB : }
887 :
888 GIC 3753 : return stats;
889 ECB : }
890 :
891 : /*
892 : * btvacuumscan --- scan the index for VACUUMing purposes
893 : *
894 : * This combines the functions of looking for leaf tuples that are deletable
895 : * according to the vacuum callback, looking for empty pages that can be
896 : * deleted, and looking for old deleted pages that can be recycled. Both
897 : * btbulkdelete and btvacuumcleanup invoke this (the latter only if no
898 : * btbulkdelete call occurred and _bt_vacuum_needs_cleanup returned true).
899 : *
900 : * The caller is responsible for initially allocating/zeroing a stats struct
901 : * and for obtaining a vacuum cycle ID if necessary.
902 : */
903 : static void
904 GIC 4016 : btvacuumscan(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
905 ECB : IndexBulkDeleteCallback callback, void *callback_state,
906 : BTCycleId cycleid)
907 : {
908 GIC 4016 : Relation rel = info->index;
909 ECB : BTVacState vstate;
910 : BlockNumber num_pages;
911 : BlockNumber scanblkno;
912 : bool needLock;
913 :
914 : /*
915 : * Reset fields that track information about the entire index now. This
916 : * avoids double-counting in the case where a single VACUUM command
917 : * requires multiple scans of the index.
918 : *
919 : * Avoid resetting the tuples_removed and pages_newly_deleted fields here,
920 : * since they track information about the VACUUM command, and so must last
921 : * across each call to btvacuumscan().
922 : *
923 : * (Note that pages_free is treated as state about the whole index, not
924 : * the current VACUUM. This is appropriate because RecordFreeIndexPage()
925 : * calls are idempotent, and get repeated for the same deleted pages in
926 : * some scenarios. The point for us is to track the number of recyclable
927 : * pages in the index at the end of the VACUUM command.)
928 : */
929 GIC 4016 : stats->num_pages = 0;
930 CBC 4016 : stats->num_index_tuples = 0;
931 4016 : stats->pages_deleted = 0;
932 4016 : stats->pages_free = 0;
933 ECB :
934 : /* Set up info to pass down to btvacuumpage */
935 GIC 4016 : vstate.info = info;
936 CBC 4016 : vstate.stats = stats;
937 4016 : vstate.callback = callback;
938 4016 : vstate.callback_state = callback_state;
939 4016 : vstate.cycleid = cycleid;
940 ECB :
941 : /* Create a temporary memory context to run _bt_pagedel in */
942 GIC 4016 : vstate.pagedelcontext = AllocSetContextCreate(CurrentMemoryContext,
943 ECB : "_bt_pagedel",
944 : ALLOCSET_DEFAULT_SIZES);
945 :
946 : /* Initialize vstate fields used by _bt_pendingfsm_finalize */
947 GIC 4016 : vstate.bufsize = 0;
948 CBC 4016 : vstate.maxbufsize = 0;
949 4016 : vstate.pendingpages = NULL;
950 4016 : vstate.npendingpages = 0;
951 ECB : /* Consider applying _bt_pendingfsm_finalize optimization */
952 GIC 4016 : _bt_pendingfsm_init(rel, &vstate, (callback == NULL));
953 ECB :
954 : /*
955 : * The outer loop iterates over all index pages except the metapage, in
956 : * physical order (we hope the kernel will cooperate in providing
957 : * read-ahead for speed). It is critical that we visit all leaf pages,
958 : * including ones added after we start the scan, else we might fail to
959 : * delete some deletable tuples. Hence, we must repeatedly check the
960 : * relation length. We must acquire the relation-extension lock while
961 : * doing so to avoid a race condition: if someone else is extending the
962 : * relation, there is a window where bufmgr/smgr have created a new
963 : * all-zero page but it hasn't yet been write-locked by _bt_getbuf(). If
964 : * we manage to scan such a page here, we'll improperly assume it can be
965 : * recycled. Taking the lock synchronizes things enough to prevent a
966 : * problem: either num_pages won't include the new page, or _bt_getbuf
967 : * already has write lock on the buffer and it will be fully initialized
968 : * before we can examine it. Also, we need not worry if a page is added
969 : * immediately after we look; the page splitting code already has
970 : * write-lock on the left page before it adds a right page, so we must
971 : * already have processed any tuples due to be moved into such a page.
972 : *
973 : * XXX: Now that new pages are locked with RBM_ZERO_AND_LOCK, I don't
974 : * think the use of the extension lock is still required.
975 : *
976 : * We can skip locking for new or temp relations, however, since no one
977 : * else could be accessing them.
978 : */
979 GIC 4016 : needLock = !RELATION_IS_LOCAL(rel);
980 :
981 4016 : scanblkno = BTREE_METAPAGE + 1;
982 : for (;;)
983 ECB : {
984 : /* Get the current relation length */
985 CBC 7935 : if (needLock)
986 GIC 7933 : LockRelationForExtension(rel, ExclusiveLock);
987 7935 : num_pages = RelationGetNumberOfBlocks(rel);
988 7935 : if (needLock)
989 CBC 7933 : UnlockRelationForExtension(rel, ExclusiveLock);
990 ECB :
991 CBC 7935 : if (info->report_progress)
992 429 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
993 ECB : num_pages);
994 :
995 : /* Quit if we've scanned the whole relation */
996 CBC 7935 : if (scanblkno >= num_pages)
997 GIC 4016 : break;
998 : /* Iterate over pages, then loop back to recheck length */
999 36840 : for (; scanblkno < num_pages; scanblkno++)
1000 ECB : {
1001 CBC 32921 : btvacuumpage(&vstate, scanblkno);
1002 GIC 32921 : if (info->report_progress)
1003 CBC 211 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1004 : scanblkno);
1005 ECB : }
1006 : }
1007 :
1008 : /* Set statistics num_pages field to final size of index */
1009 GIC 4016 : stats->num_pages = num_pages;
1010 :
1011 4016 : MemoryContextDelete(vstate.pagedelcontext);
1012 :
1013 ECB : /*
1014 : * If there were any calls to _bt_pagedel() during scan of the index then
1015 : * see if any of the resulting pages can be placed in the FSM now. When
1016 : * it's not safe we'll have to leave it up to a future VACUUM operation.
1017 : *
1018 : * Finally, if we placed any pages in the FSM (either just now or during
1019 : * the scan), forcibly update the upper-level FSM pages to ensure that
1020 : * searchers can find them.
1021 : */
1022 GIC 4016 : _bt_pendingfsm_finalize(rel, &vstate);
1023 4016 : if (stats->pages_free > 0)
1024 18 : IndexFreeSpaceMapVacuum(rel);
1025 4016 : }
1026 ECB :
1027 : /*
1028 : * btvacuumpage --- VACUUM one page
1029 : *
1030 : * This processes a single page for btvacuumscan(). In some cases we must
1031 : * backtrack to re-examine and VACUUM pages that were the scanblkno during
1032 : * a previous call here. This is how we handle page splits (that happened
1033 : * after our cycleid was acquired) whose right half page happened to reuse
1034 : * a block that we might have processed at some point before it was
1035 : * recycled (i.e. before the page split).
1036 : */
1037 : static void
1038 GIC 32921 : btvacuumpage(BTVacState *vstate, BlockNumber scanblkno)
1039 : {
1040 32921 : IndexVacuumInfo *info = vstate->info;
1041 32921 : IndexBulkDeleteResult *stats = vstate->stats;
1042 CBC 32921 : IndexBulkDeleteCallback callback = vstate->callback;
1043 GIC 32921 : void *callback_state = vstate->callback_state;
1044 CBC 32921 : Relation rel = info->index;
1045 GNC 32921 : Relation heaprel = info->heaprel;
1046 ECB : bool attempt_pagedel;
1047 : BlockNumber blkno,
1048 : backtrack_to;
1049 : Buffer buf;
1050 : Page page;
1051 : BTPageOpaque opaque;
1052 :
1053 GIC 32921 : blkno = scanblkno;
1054 :
1055 32921 : backtrack:
1056 :
1057 32921 : attempt_pagedel = false;
1058 CBC 32921 : backtrack_to = P_NONE;
1059 :
1060 ECB : /* call vacuum_delay_point while not holding any buffer lock */
1061 GIC 32921 : vacuum_delay_point();
1062 ECB :
1063 : /*
1064 : * We can't use _bt_getbuf() here because it always applies
1065 : * _bt_checkpage(), which will barf on an all-zero page. We want to
1066 : * recycle all-zero pages, not fail. Also, we want to use a nondefault
1067 : * buffer access strategy.
1068 : */
1069 GIC 32921 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL,
1070 : info->strategy);
1071 32921 : _bt_lockbuf(rel, buf, BT_READ);
1072 32921 : page = BufferGetPage(buf);
1073 32921 : opaque = NULL;
1074 CBC 32921 : if (!PageIsNew(page))
1075 : {
1076 32921 : _bt_checkpage(rel, buf);
1077 32921 : opaque = BTPageGetOpaque(page);
1078 ECB : }
1079 :
1080 GIC 32921 : Assert(blkno <= scanblkno);
1081 CBC 32921 : if (blkno != scanblkno)
1082 ECB : {
1083 : /*
1084 : * We're backtracking.
1085 : *
1086 : * We followed a right link to a sibling leaf page (a page that
1087 : * happens to be from a block located before scanblkno). The only
1088 : * case we want to do anything with is a live leaf page having the
1089 : * current vacuum cycle ID.
1090 : *
1091 : * The page had better be in a state that's consistent with what we
1092 : * expect. Check for conditions that imply corruption in passing. It
1093 : * can't be half-dead because only an interrupted VACUUM process can
1094 : * leave pages in that state, so we'd definitely have dealt with it
1095 : * back when the page was the scanblkno page (half-dead pages are
1096 : * always marked fully deleted by _bt_pagedel()). This assumes that
1097 : * there can be only one vacuum process running at a time.
1098 : */
1099 UIC 0 : if (!opaque || !P_ISLEAF(opaque) || P_ISHALFDEAD(opaque))
1100 : {
1101 0 : Assert(false);
1102 : ereport(LOG,
1103 : (errcode(ERRCODE_INDEX_CORRUPTED),
1104 EUB : errmsg_internal("right sibling %u of scanblkno %u unexpectedly in an inconsistent state in index \"%s\"",
1105 : blkno, scanblkno, RelationGetRelationName(rel))));
1106 : _bt_relbuf(rel, buf);
1107 : return;
1108 : }
1109 :
1110 : /*
1111 : * We may have already processed the page in an earlier call, when the
1112 : * page was scanblkno. This happens when the leaf page split occurred
1113 : * after the scan began, but before the right sibling page became the
1114 : * scanblkno.
1115 : *
1116 : * Page may also have been deleted by current btvacuumpage() call,
1117 : * since _bt_pagedel() sometimes deletes the right sibling page of
1118 : * scanblkno in passing (it does so after we decided where to
1119 : * backtrack to). We don't need to process this page as a deleted
1120 : * page a second time now (in fact, it would be wrong to count it as a
1121 : * deleted page in the bulk delete statistics a second time).
1122 : */
1123 UIC 0 : if (opaque->btpo_cycleid != vstate->cycleid || P_ISDELETED(opaque))
1124 : {
1125 : /* Done with current scanblkno (and all lower split pages) */
1126 0 : _bt_relbuf(rel, buf);
1127 0 : return;
1128 EUB : }
1129 : }
1130 :
1131 GNC 32921 : if (!opaque || BTPageIsRecyclable(page, heaprel))
1132 EUB : {
1133 : /* Okay to recycle this page (which could be leaf or internal) */
1134 GIC 120 : RecordFreeIndexPage(rel, blkno);
1135 120 : stats->pages_deleted++;
1136 CBC 120 : stats->pages_free++;
1137 : }
1138 GIC 32801 : else if (P_ISDELETED(opaque))
1139 ECB : {
1140 : /*
1141 : * Already deleted page (which could be leaf or internal). Can't
1142 : * recycle yet.
1143 : */
1144 GIC 97 : stats->pages_deleted++;
1145 : }
1146 32704 : else if (P_ISHALFDEAD(opaque))
1147 : {
1148 : /* Half-dead leaf page (from interrupted VACUUM) -- finish deleting */
1149 LBC 0 : attempt_pagedel = true;
1150 :
1151 ECB : /*
1152 : * _bt_pagedel() will increment both pages_newly_deleted and
1153 : * pages_deleted stats in all cases (barring corruption)
1154 EUB : */
1155 : }
1156 GIC 32704 : else if (P_ISLEAF(opaque))
1157 : {
1158 : OffsetNumber deletable[MaxIndexTuplesPerPage];
1159 : int ndeletable;
1160 : BTVacuumPosting updatable[MaxIndexTuplesPerPage];
1161 ECB : int nupdatable;
1162 : OffsetNumber offnum,
1163 : minoff,
1164 : maxoff;
1165 : int nhtidsdead,
1166 : nhtidslive;
1167 :
1168 : /*
1169 : * Trade in the initial read lock for a full cleanup lock on this
1170 : * page. We must get such a lock on every leaf page over the course
1171 : * of the vacuum scan, whether or not it actually contains any
1172 : * deletable tuples --- see nbtree/README.
1173 : */
1174 GIC 30039 : _bt_upgradelockbufcleanup(rel, buf);
1175 :
1176 : /*
1177 : * Check whether we need to backtrack to earlier pages. What we are
1178 : * concerned about is a page split that happened since we started the
1179 ECB : * vacuum scan. If the split moved tuples on the right half of the
1180 : * split (i.e. the tuples that sort high) to a block that we already
1181 : * passed over, then we might have missed the tuples. We need to
1182 : * backtrack now. (Must do this before possibly clearing btpo_cycleid
1183 : * or deleting scanblkno page below!)
1184 : */
1185 GIC 30039 : if (vstate->cycleid != 0 &&
1186 29981 : opaque->btpo_cycleid == vstate->cycleid &&
1187 UIC 0 : !(opaque->btpo_flags & BTP_SPLIT_END) &&
1188 0 : !P_RIGHTMOST(opaque) &&
1189 0 : opaque->btpo_next < scanblkno)
1190 LBC 0 : backtrack_to = opaque->btpo_next;
1191 ECB :
1192 GBC 30039 : ndeletable = 0;
1193 30039 : nupdatable = 0;
1194 30039 : minoff = P_FIRSTDATAKEY(opaque);
1195 30039 : maxoff = PageGetMaxOffsetNumber(page);
1196 GIC 30039 : nhtidsdead = 0;
1197 CBC 30039 : nhtidslive = 0;
1198 30039 : if (callback)
1199 ECB : {
1200 : /* btbulkdelete callback tells us what to delete (or update) */
1201 CBC 29981 : for (offnum = minoff;
1202 5948136 : offnum <= maxoff;
1203 5918155 : offnum = OffsetNumberNext(offnum))
1204 : {
1205 : IndexTuple itup;
1206 ECB :
1207 CBC 5918155 : itup = (IndexTuple) PageGetItem(page,
1208 ECB : PageGetItemId(page, offnum));
1209 :
1210 GIC 5918155 : Assert(!BTreeTupleIsPivot(itup));
1211 5918155 : if (!BTreeTupleIsPosting(itup))
1212 ECB : {
1213 : /* Regular tuple, standard table TID representation */
1214 GIC 5713458 : if (callback(&itup->t_tid, callback_state))
1215 ECB : {
1216 CBC 712323 : deletable[ndeletable++] = offnum;
1217 GIC 712323 : nhtidsdead++;
1218 : }
1219 ECB : else
1220 GIC 5001135 : nhtidslive++;
1221 ECB : }
1222 : else
1223 : {
1224 : BTVacuumPosting vacposting;
1225 : int nremaining;
1226 :
1227 : /* Posting list tuple */
1228 GIC 204697 : vacposting = btreevacuumposting(vstate, itup, offnum,
1229 : &nremaining);
1230 204697 : if (vacposting == NULL)
1231 : {
1232 : /*
1233 ECB : * All table TIDs from the posting tuple remain, so no
1234 : * delete or update required
1235 : */
1236 GIC 118403 : Assert(nremaining == BTreeTupleGetNPosting(itup));
1237 : }
1238 86294 : else if (nremaining > 0)
1239 : {
1240 :
1241 ECB : /*
1242 : * Store metadata about posting list tuple in
1243 : * updatable array for entire page. Existing tuple
1244 : * will be updated during the later call to
1245 : * _bt_delitems_vacuum().
1246 : */
1247 GIC 74159 : Assert(nremaining < BTreeTupleGetNPosting(itup));
1248 74159 : updatable[nupdatable++] = vacposting;
1249 74159 : nhtidsdead += BTreeTupleGetNPosting(itup) - nremaining;
1250 : }
1251 : else
1252 ECB : {
1253 : /*
1254 : * All table TIDs from the posting list must be
1255 : * deleted. We'll delete the index tuple completely
1256 : * (no update required).
1257 : */
1258 GIC 12135 : Assert(nremaining == 0);
1259 12135 : deletable[ndeletable++] = offnum;
1260 12135 : nhtidsdead += BTreeTupleGetNPosting(itup);
1261 12135 : pfree(vacposting);
1262 : }
1263 ECB :
1264 CBC 204697 : nhtidslive += nremaining;
1265 ECB : }
1266 : }
1267 : }
1268 :
1269 : /*
1270 : * Apply any needed deletes or updates. We issue just one
1271 : * _bt_delitems_vacuum() call per page, so as to minimize WAL traffic.
1272 : */
1273 GIC 30039 : if (ndeletable > 0 || nupdatable > 0)
1274 : {
1275 16990 : Assert(nhtidsdead >= ndeletable + nupdatable);
1276 16990 : _bt_delitems_vacuum(rel, buf, deletable, ndeletable, updatable,
1277 : nupdatable);
1278 ECB :
1279 GIC 16990 : stats->tuples_removed += nhtidsdead;
1280 ECB : /* must recompute maxoff */
1281 CBC 16990 : maxoff = PageGetMaxOffsetNumber(page);
1282 :
1283 : /* can't leak memory here */
1284 91149 : for (int i = 0; i < nupdatable; i++)
1285 GIC 74159 : pfree(updatable[i]);
1286 ECB : }
1287 : else
1288 : {
1289 : /*
1290 : * If the leaf page has been split during this vacuum cycle, it
1291 : * seems worth expending a write to clear btpo_cycleid even if we
1292 : * don't have any deletions to do. (If we do, _bt_delitems_vacuum
1293 : * takes care of this.) This ensures we won't process the page
1294 : * again.
1295 : *
1296 : * We treat this like a hint-bit update because there's no need to
1297 : * WAL-log it.
1298 : */
1299 GIC 13049 : Assert(nhtidsdead == 0);
1300 13049 : if (vstate->cycleid != 0 &&
1301 12991 : opaque->btpo_cycleid == vstate->cycleid)
1302 : {
1303 UIC 0 : opaque->btpo_cycleid = 0;
1304 LBC 0 : MarkBufferDirtyHint(buf, true);
1305 ECB : }
1306 : }
1307 :
1308 EUB : /*
1309 : * If the leaf page is now empty, try to delete it; else count the
1310 : * live tuples (live table TIDs in posting lists are counted as
1311 : * separate live tuples). We don't delete when backtracking, though,
1312 : * since that would require teaching _bt_pagedel() about backtracking
1313 : * (doesn't seem worth adding more complexity to deal with that).
1314 : *
1315 : * We don't count the number of live TIDs during cleanup-only calls to
1316 : * btvacuumscan (i.e. when callback is not set). We count the number
1317 : * of index tuples directly instead. This avoids the expense of
1318 : * directly examining all of the tuples on each page. VACUUM will
1319 : * treat num_index_tuples as an estimate in cleanup-only case, so it
1320 : * doesn't matter that this underestimates num_index_tuples
1321 : * significantly in some cases.
1322 : */
1323 GIC 30039 : if (minoff > maxoff)
1324 2730 : attempt_pagedel = (blkno == scanblkno);
1325 27309 : else if (callback)
1326 27255 : stats->num_index_tuples += nhtidslive;
1327 : else
1328 CBC 54 : stats->num_index_tuples += maxoff - minoff + 1;
1329 ECB :
1330 CBC 30039 : Assert(!attempt_pagedel || nhtidslive == 0);
1331 ECB : }
1332 :
1333 CBC 32921 : if (attempt_pagedel)
1334 : {
1335 ECB : MemoryContext oldcontext;
1336 :
1337 : /* Run pagedel in a temp context to avoid memory leakage */
1338 CBC 2730 : MemoryContextReset(vstate->pagedelcontext);
1339 GIC 2730 : oldcontext = MemoryContextSwitchTo(vstate->pagedelcontext);
1340 :
1341 : /*
1342 : * _bt_pagedel maintains the bulk delete stats on our behalf;
1343 ECB : * pages_newly_deleted and pages_deleted are likely to be incremented
1344 : * during call
1345 : */
1346 GIC 2730 : Assert(blkno == scanblkno);
1347 2730 : _bt_pagedel(rel, buf, vstate);
1348 :
1349 2730 : MemoryContextSwitchTo(oldcontext);
1350 : /* pagedel released buffer, so we shouldn't */
1351 ECB : }
1352 : else
1353 GIC 30191 : _bt_relbuf(rel, buf);
1354 ECB :
1355 GIC 32921 : if (backtrack_to != P_NONE)
1356 : {
1357 UIC 0 : blkno = backtrack_to;
1358 LBC 0 : goto backtrack;
1359 : }
1360 ECB : }
1361 :
1362 EUB : /*
1363 : * btreevacuumposting --- determine TIDs still needed in posting list
1364 : *
1365 : * Returns metadata describing how to build replacement tuple without the TIDs
1366 : * that VACUUM needs to delete. Returned value is NULL in the common case
1367 : * where no changes are needed to caller's posting list tuple (we avoid
1368 : * allocating memory here as an optimization).
1369 : *
1370 : * The number of TIDs that should remain in the posting list tuple is set for
1371 : * caller in *nremaining.
1372 : */
1373 : static BTVacuumPosting
1374 GIC 204697 : btreevacuumposting(BTVacState *vstate, IndexTuple posting,
1375 : OffsetNumber updatedoffset, int *nremaining)
1376 : {
1377 204697 : int live = 0;
1378 204697 : int nitem = BTreeTupleGetNPosting(posting);
1379 CBC 204697 : ItemPointer items = BTreeTupleGetPosting(posting);
1380 GIC 204697 : BTVacuumPosting vacposting = NULL;
1381 :
1382 CBC 1120832 : for (int i = 0; i < nitem; i++)
1383 ECB : {
1384 CBC 916135 : if (!vstate->callback(items + i, vstate->callback_state))
1385 ECB : {
1386 : /* Live table TID */
1387 CBC 744283 : live++;
1388 : }
1389 171852 : else if (vacposting == NULL)
1390 : {
1391 : /*
1392 ECB : * First dead table TID encountered.
1393 : *
1394 : * It's now clear that we need to delete one or more dead table
1395 : * TIDs, so start maintaining metadata describing how to update
1396 : * existing posting list tuple.
1397 : */
1398 GIC 86294 : vacposting = palloc(offsetof(BTVacuumPostingData, deletetids) +
1399 : nitem * sizeof(uint16));
1400 :
1401 86294 : vacposting->itup = posting;
1402 86294 : vacposting->updatedoffset = updatedoffset;
1403 CBC 86294 : vacposting->ndeletedtids = 0;
1404 GIC 86294 : vacposting->deletetids[vacposting->ndeletedtids++] = i;
1405 : }
1406 ECB : else
1407 : {
1408 : /* Second or subsequent dead table TID */
1409 CBC 85558 : vacposting->deletetids[vacposting->ndeletedtids++] = i;
1410 : }
1411 : }
1412 :
1413 GIC 204697 : *nremaining = live;
1414 CBC 204697 : return vacposting;
1415 : }
1416 :
1417 : /*
1418 ECB : * btcanreturn() -- Check whether btree indexes support index-only scans.
1419 : *
1420 : * btrees always do, so this is trivial.
1421 : */
1422 : bool
1423 GIC 398723 : btcanreturn(Relation index, int attno)
1424 : {
1425 398723 : return true;
1426 : }
|
