Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pruneheap.c
4 : * heap page pruning and HOT-chain management code
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/heap/pruneheap.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include "access/heapam.h"
18 : #include "access/heapam_xlog.h"
19 : #include "access/htup_details.h"
20 : #include "access/multixact.h"
21 : #include "access/transam.h"
22 : #include "access/visibilitymapdefs.h"
23 : #include "access/xlog.h"
24 : #include "access/xloginsert.h"
25 : #include "commands/vacuum.h"
26 : #include "executor/instrument.h"
27 : #include "miscadmin.h"
28 : #include "pgstat.h"
29 : #include "storage/bufmgr.h"
30 : #include "utils/rel.h"
31 : #include "utils/snapmgr.h"
32 :
33 : /* Working data for heap_page_prune_and_freeze() and subroutines */
34 : typedef struct
35 : {
36 : /*-------------------------------------------------------
37 : * Arguments passed to heap_page_prune_and_freeze()
38 : *-------------------------------------------------------
39 : */
40 :
41 : /* tuple visibility test, initialized for the relation */
42 : GlobalVisState *vistest;
43 : /* whether or not dead items can be set LP_UNUSED during pruning */
44 : bool mark_unused_now;
45 : /* whether to attempt freezing tuples */
46 : bool attempt_freeze;
47 : struct VacuumCutoffs *cutoffs;
48 :
49 : /*-------------------------------------------------------
50 : * Fields describing what to do to the page
51 : *-------------------------------------------------------
52 : */
53 : TransactionId new_prune_xid; /* new prune hint value */
54 : TransactionId latest_xid_removed;
55 : int nredirected; /* numbers of entries in arrays below */
56 : int ndead;
57 : int nunused;
58 : int nfrozen;
59 : /* arrays that accumulate indexes of items to be changed */
60 : OffsetNumber redirected[MaxHeapTuplesPerPage * 2];
61 : OffsetNumber nowdead[MaxHeapTuplesPerPage];
62 : OffsetNumber nowunused[MaxHeapTuplesPerPage];
63 : HeapTupleFreeze frozen[MaxHeapTuplesPerPage];
64 :
65 : /*-------------------------------------------------------
66 : * Working state for HOT chain processing
67 : *-------------------------------------------------------
68 : */
69 :
70 : /*
71 : * 'root_items' contains offsets of all LP_REDIRECT line pointers and
72 : * normal non-HOT tuples. They can be stand-alone items or the first item
73 : * in a HOT chain. 'heaponly_items' contains heap-only tuples which can
74 : * only be removed as part of a HOT chain.
75 : */
76 : int nroot_items;
77 : OffsetNumber root_items[MaxHeapTuplesPerPage];
78 : int nheaponly_items;
79 : OffsetNumber heaponly_items[MaxHeapTuplesPerPage];
80 :
81 : /*
82 : * processed[offnum] is true if item at offnum has been processed.
83 : *
84 : * This needs to be MaxHeapTuplesPerPage + 1 long as FirstOffsetNumber is
85 : * 1. Otherwise every access would need to subtract 1.
86 : */
87 : bool processed[MaxHeapTuplesPerPage + 1];
88 :
89 : /*
90 : * Tuple visibility is only computed once for each tuple, for correctness
91 : * and efficiency reasons; see comment in heap_page_prune_and_freeze() for
92 : * details. This is of type int8[], instead of HTSV_Result[], so we can
93 : * use -1 to indicate no visibility has been computed, e.g. for LP_DEAD
94 : * items.
95 : *
96 : * This needs to be MaxHeapTuplesPerPage + 1 long as FirstOffsetNumber is
97 : * 1. Otherwise every access would need to subtract 1.
98 : */
99 : int8 htsv[MaxHeapTuplesPerPage + 1];
100 :
101 : /*
102 : * Freezing-related state.
103 : */
104 : HeapPageFreeze pagefrz;
105 :
106 : /*-------------------------------------------------------
107 : * Information about what was done
108 : *
109 : * These fields are not used by pruning itself for the most part, but are
110 : * used to collect information about what was pruned and what state the
111 : * page is in after pruning, for the benefit of the caller. They are
112 : * copied to the caller's PruneFreezeResult at the end.
113 : * -------------------------------------------------------
114 : */
115 :
116 : int ndeleted; /* Number of tuples deleted from the page */
117 :
118 : /* Number of live and recently dead tuples, after pruning */
119 : int live_tuples;
120 : int recently_dead_tuples;
121 :
122 : /* Whether or not the page makes rel truncation unsafe */
123 : bool hastup;
124 :
125 : /*
126 : * LP_DEAD items on the page after pruning. Includes existing LP_DEAD
127 : * items
128 : */
129 : int lpdead_items; /* number of items in the array */
130 : OffsetNumber *deadoffsets; /* points directly to presult->deadoffsets */
131 :
132 : /*
133 : * The snapshot conflict horizon used when freezing tuples. The final
134 : * snapshot conflict horizon for the record may be newer if pruning
135 : * removes newer transaction IDs.
136 : */
137 : TransactionId frz_conflict_horizon;
138 :
139 : /*
140 : * all_visible and all_frozen indicate if the all-visible and all-frozen
141 : * bits in the visibility map can be set for this page after pruning.
142 : *
143 : * visibility_cutoff_xid is the newest xmin of live tuples on the page.
144 : * The caller can use it as the conflict horizon, when setting the VM
145 : * bits. It is only valid if we froze some tuples, and all_frozen is
146 : * true.
147 : *
148 : * NOTE: all_visible and all_frozen initially don't include LP_DEAD items.
149 : * That's convenient for heap_page_prune_and_freeze() to use them to
150 : * decide whether to freeze the page or not. The all_visible and
151 : * all_frozen values returned to the caller are adjusted to include
152 : * LP_DEAD items after we determine whether to opportunistically freeze.
153 : */
154 : bool all_visible;
155 : bool all_frozen;
156 : TransactionId visibility_cutoff_xid;
157 : } PruneState;
158 :
159 : /* Local functions */
160 : static void prune_freeze_setup(PruneFreezeParams *params,
161 : TransactionId *new_relfrozen_xid,
162 : MultiXactId *new_relmin_mxid,
163 : const PruneFreezeResult *presult,
164 : PruneState *prstate);
165 : static void prune_freeze_plan(Oid reloid, Buffer buffer,
166 : PruneState *prstate,
167 : OffsetNumber *off_loc);
168 : static HTSV_Result heap_prune_satisfies_vacuum(PruneState *prstate,
169 : HeapTuple tup,
170 : Buffer buffer);
171 : static inline HTSV_Result htsv_get_valid_status(int status);
172 : static void heap_prune_chain(Page page, BlockNumber blockno, OffsetNumber maxoff,
173 : OffsetNumber rootoffnum, PruneState *prstate);
174 : static void heap_prune_record_prunable(PruneState *prstate, TransactionId xid);
175 : static void heap_prune_record_redirect(PruneState *prstate,
176 : OffsetNumber offnum, OffsetNumber rdoffnum,
177 : bool was_normal);
178 : static void heap_prune_record_dead(PruneState *prstate, OffsetNumber offnum,
179 : bool was_normal);
180 : static void heap_prune_record_dead_or_unused(PruneState *prstate, OffsetNumber offnum,
181 : bool was_normal);
182 : static void heap_prune_record_unused(PruneState *prstate, OffsetNumber offnum, bool was_normal);
183 :
184 : static void heap_prune_record_unchanged_lp_unused(Page page, PruneState *prstate, OffsetNumber offnum);
185 : static void heap_prune_record_unchanged_lp_normal(Page page, PruneState *prstate, OffsetNumber offnum);
186 : static void heap_prune_record_unchanged_lp_dead(Page page, PruneState *prstate, OffsetNumber offnum);
187 : static void heap_prune_record_unchanged_lp_redirect(PruneState *prstate, OffsetNumber offnum);
188 :
189 : static void page_verify_redirects(Page page);
190 :
191 : static bool heap_page_will_freeze(Relation relation, Buffer buffer,
192 : bool did_tuple_hint_fpi, bool do_prune, bool do_hint_prune,
193 : PruneState *prstate);
194 :
195 :
196 : /*
197 : * Optionally prune and repair fragmentation in the specified page.
198 : *
199 : * This is an opportunistic function. It will perform housekeeping
200 : * only if the page heuristically looks like a candidate for pruning and we
201 : * can acquire buffer cleanup lock without blocking.
202 : *
203 : * Note: this is called quite often. It's important that it fall out quickly
204 : * if there's not any use in pruning.
205 : *
206 : * Caller must have pin on the buffer, and must *not* have a lock on it.
207 : */
208 : void
209 32647530 : heap_page_prune_opt(Relation relation, Buffer buffer)
210 : {
211 32647530 : Page page = BufferGetPage(buffer);
212 : TransactionId prune_xid;
213 : GlobalVisState *vistest;
214 : Size minfree;
215 :
216 : /*
217 : * We can't write WAL in recovery mode, so there's no point trying to
218 : * clean the page. The primary will likely issue a cleaning WAL record
219 : * soon anyway, so this is no particular loss.
220 : */
221 32647530 : if (RecoveryInProgress())
222 419534 : return;
223 :
224 : /*
225 : * First check whether there's any chance there's something to prune,
226 : * determining the appropriate horizon is a waste if there's no prune_xid
227 : * (i.e. no updates/deletes left potentially dead tuples around).
228 : */
229 32227996 : prune_xid = ((PageHeader) page)->pd_prune_xid;
230 32227996 : if (!TransactionIdIsValid(prune_xid))
231 16739862 : return;
232 :
233 : /*
234 : * Check whether prune_xid indicates that there may be dead rows that can
235 : * be cleaned up.
236 : */
237 15488134 : vistest = GlobalVisTestFor(relation);
238 :
239 15488134 : if (!GlobalVisTestIsRemovableXid(vistest, prune_xid))
240 12991932 : return;
241 :
242 : /*
243 : * We prune when a previous UPDATE failed to find enough space on the page
244 : * for a new tuple version, or when free space falls below the relation's
245 : * fill-factor target (but not less than 10%).
246 : *
247 : * Checking free space here is questionable since we aren't holding any
248 : * lock on the buffer; in the worst case we could get a bogus answer. It's
249 : * unlikely to be *seriously* wrong, though, since reading either pd_lower
250 : * or pd_upper is probably atomic. Avoiding taking a lock seems more
251 : * important than sometimes getting a wrong answer in what is after all
252 : * just a heuristic estimate.
253 : */
254 2496202 : minfree = RelationGetTargetPageFreeSpace(relation,
255 : HEAP_DEFAULT_FILLFACTOR);
256 2496202 : minfree = Max(minfree, BLCKSZ / 10);
257 :
258 2496202 : if (PageIsFull(page) || PageGetHeapFreeSpace(page) < minfree)
259 : {
260 : /* OK, try to get exclusive buffer lock */
261 83836 : if (!ConditionalLockBufferForCleanup(buffer))
262 842 : return;
263 :
264 : /*
265 : * Now that we have buffer lock, get accurate information about the
266 : * page's free space, and recheck the heuristic about whether to
267 : * prune.
268 : */
269 82994 : if (PageIsFull(page) || PageGetHeapFreeSpace(page) < minfree)
270 : {
271 : OffsetNumber dummy_off_loc;
272 : PruneFreezeResult presult;
273 :
274 : /*
275 : * We don't pass the HEAP_PAGE_PRUNE_MARK_UNUSED_NOW option
276 : * regardless of whether or not the relation has indexes, since we
277 : * cannot safely determine that during on-access pruning with the
278 : * current implementation.
279 : */
280 82994 : PruneFreezeParams params = {
281 : .relation = relation,
282 : .buffer = buffer,
283 : .reason = PRUNE_ON_ACCESS,
284 : .options = 0,
285 : .vistest = vistest,
286 : .cutoffs = NULL,
287 : };
288 :
289 82994 : heap_page_prune_and_freeze(¶ms, &presult, &dummy_off_loc,
290 : NULL, NULL);
291 :
292 : /*
293 : * Report the number of tuples reclaimed to pgstats. This is
294 : * presult.ndeleted minus the number of newly-LP_DEAD-set items.
295 : *
296 : * We derive the number of dead tuples like this to avoid totally
297 : * forgetting about items that were set to LP_DEAD, since they
298 : * still need to be cleaned up by VACUUM. We only want to count
299 : * heap-only tuples that just became LP_UNUSED in our report,
300 : * which don't.
301 : *
302 : * VACUUM doesn't have to compensate in the same way when it
303 : * tracks ndeleted, since it will set the same LP_DEAD items to
304 : * LP_UNUSED separately.
305 : */
306 82994 : if (presult.ndeleted > presult.nnewlpdead)
307 37774 : pgstat_update_heap_dead_tuples(relation,
308 37774 : presult.ndeleted - presult.nnewlpdead);
309 : }
310 :
311 : /* And release buffer lock */
312 82994 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
313 :
314 : /*
315 : * We avoid reuse of any free space created on the page by unrelated
316 : * UPDATEs/INSERTs by opting to not update the FSM at this point. The
317 : * free space should be reused by UPDATEs to *this* page.
318 : */
319 : }
320 : }
321 :
322 : /*
323 : * Helper for heap_page_prune_and_freeze() to initialize the PruneState using
324 : * the provided parameters.
325 : */
326 : static void
327 1549884 : prune_freeze_setup(PruneFreezeParams *params,
328 : TransactionId *new_relfrozen_xid,
329 : MultiXactId *new_relmin_mxid,
330 : const PruneFreezeResult *presult,
331 : PruneState *prstate)
332 : {
333 : /* Copy parameters to prstate */
334 1549884 : prstate->vistest = params->vistest;
335 1549884 : prstate->mark_unused_now =
336 1549884 : (params->options & HEAP_PAGE_PRUNE_MARK_UNUSED_NOW) != 0;
337 :
338 : /* cutoffs must be provided if we will attempt freezing */
339 : Assert(!(params->options & HEAP_PAGE_PRUNE_FREEZE) || params->cutoffs);
340 1549884 : prstate->attempt_freeze = (params->options & HEAP_PAGE_PRUNE_FREEZE) != 0;
341 1549884 : prstate->cutoffs = params->cutoffs;
342 :
343 : /*
344 : * Our strategy is to scan the page and make lists of items to change,
345 : * then apply the changes within a critical section. This keeps as much
346 : * logic as possible out of the critical section, and also ensures that
347 : * WAL replay will work the same as the normal case.
348 : *
349 : * First, initialize the new pd_prune_xid value to zero (indicating no
350 : * prunable tuples). If we find any tuples which may soon become
351 : * prunable, we will save the lowest relevant XID in new_prune_xid. Also
352 : * initialize the rest of our working state.
353 : */
354 1549884 : prstate->new_prune_xid = InvalidTransactionId;
355 1549884 : prstate->latest_xid_removed = InvalidTransactionId;
356 1549884 : prstate->nredirected = prstate->ndead = prstate->nunused = 0;
357 1549884 : prstate->nfrozen = 0;
358 1549884 : prstate->nroot_items = 0;
359 1549884 : prstate->nheaponly_items = 0;
360 :
361 : /* initialize page freezing working state */
362 1549884 : prstate->pagefrz.freeze_required = false;
363 1549884 : if (prstate->attempt_freeze)
364 : {
365 : Assert(new_relfrozen_xid && new_relmin_mxid);
366 1466890 : prstate->pagefrz.FreezePageRelfrozenXid = *new_relfrozen_xid;
367 1466890 : prstate->pagefrz.NoFreezePageRelfrozenXid = *new_relfrozen_xid;
368 1466890 : prstate->pagefrz.FreezePageRelminMxid = *new_relmin_mxid;
369 1466890 : prstate->pagefrz.NoFreezePageRelminMxid = *new_relmin_mxid;
370 : }
371 : else
372 : {
373 : Assert(!new_relfrozen_xid && !new_relmin_mxid);
374 82994 : prstate->pagefrz.FreezePageRelminMxid = InvalidMultiXactId;
375 82994 : prstate->pagefrz.NoFreezePageRelminMxid = InvalidMultiXactId;
376 82994 : prstate->pagefrz.FreezePageRelfrozenXid = InvalidTransactionId;
377 82994 : prstate->pagefrz.NoFreezePageRelfrozenXid = InvalidTransactionId;
378 : }
379 :
380 1549884 : prstate->ndeleted = 0;
381 1549884 : prstate->live_tuples = 0;
382 1549884 : prstate->recently_dead_tuples = 0;
383 1549884 : prstate->hastup = false;
384 1549884 : prstate->lpdead_items = 0;
385 1549884 : prstate->deadoffsets = (OffsetNumber *) presult->deadoffsets;
386 1549884 : prstate->frz_conflict_horizon = InvalidTransactionId;
387 :
388 : /*
389 : * Vacuum may update the VM after we're done. We can keep track of
390 : * whether the page will be all-visible and all-frozen after pruning and
391 : * freezing to help the caller to do that.
392 : *
393 : * Currently, only VACUUM sets the VM bits. To save the effort, only do
394 : * the bookkeeping if the caller needs it. Currently, that's tied to
395 : * HEAP_PAGE_PRUNE_FREEZE, but it could be a separate flag if you wanted
396 : * to update the VM bits without also freezing or freeze without also
397 : * setting the VM bits.
398 : *
399 : * In addition to telling the caller whether it can set the VM bit, we
400 : * also use 'all_visible' and 'all_frozen' for our own decision-making. If
401 : * the whole page would become frozen, we consider opportunistically
402 : * freezing tuples. We will not be able to freeze the whole page if there
403 : * are tuples present that are not visible to everyone or if there are
404 : * dead tuples which are not yet removable. However, dead tuples which
405 : * will be removed by the end of vacuuming should not preclude us from
406 : * opportunistically freezing. Because of that, we do not immediately
407 : * clear all_visible and all_frozen when we see LP_DEAD items. We fix
408 : * that after scanning the line pointers. We must correct all_visible and
409 : * all_frozen before we return them to the caller, so that the caller
410 : * doesn't set the VM bits incorrectly.
411 : */
412 1549884 : if (prstate->attempt_freeze)
413 : {
414 1466890 : prstate->all_visible = true;
415 1466890 : prstate->all_frozen = true;
416 : }
417 : else
418 : {
419 : /*
420 : * Initializing to false allows skipping the work to update them in
421 : * heap_prune_record_unchanged_lp_normal().
422 : */
423 82994 : prstate->all_visible = false;
424 82994 : prstate->all_frozen = false;
425 : }
426 :
427 : /*
428 : * The visibility cutoff xid is the newest xmin of live tuples on the
429 : * page. In the common case, this will be set as the conflict horizon the
430 : * caller can use for updating the VM. If, at the end of freezing and
431 : * pruning, the page is all-frozen, there is no possibility that any
432 : * running transaction on the standby does not see tuples on the page as
433 : * all-visible, so the conflict horizon remains InvalidTransactionId.
434 : */
435 1549884 : prstate->visibility_cutoff_xid = InvalidTransactionId;
436 1549884 : }
437 :
438 : /*
439 : * Helper for heap_page_prune_and_freeze(). Iterates over every tuple on the
440 : * page, examines its visibility information, and determines the appropriate
441 : * action for each tuple. All tuples are processed and classified during this
442 : * phase, but no modifications are made to the page until the later execution
443 : * stage.
444 : *
445 : * *off_loc is used for error callback and cleared before returning.
446 : */
447 : static void
448 1549884 : prune_freeze_plan(Oid reloid, Buffer buffer, PruneState *prstate,
449 : OffsetNumber *off_loc)
450 : {
451 1549884 : Page page = BufferGetPage(buffer);
452 1549884 : BlockNumber blockno = BufferGetBlockNumber(buffer);
453 1549884 : OffsetNumber maxoff = PageGetMaxOffsetNumber(page);
454 : OffsetNumber offnum;
455 : HeapTupleData tup;
456 :
457 1549884 : tup.t_tableOid = reloid;
458 :
459 : /*
460 : * Determine HTSV for all tuples, and queue them up for processing as HOT
461 : * chain roots or as heap-only items.
462 : *
463 : * Determining HTSV only once for each tuple is required for correctness,
464 : * to deal with cases where running HTSV twice could result in different
465 : * results. For example, RECENTLY_DEAD can turn to DEAD if another
466 : * checked item causes GlobalVisTestIsRemovableFullXid() to update the
467 : * horizon, or INSERT_IN_PROGRESS can change to DEAD if the inserting
468 : * transaction aborts.
469 : *
470 : * It's also good for performance. Most commonly tuples within a page are
471 : * stored at decreasing offsets (while the items are stored at increasing
472 : * offsets). When processing all tuples on a page this leads to reading
473 : * memory at decreasing offsets within a page, with a variable stride.
474 : * That's hard for CPU prefetchers to deal with. Processing the items in
475 : * reverse order (and thus the tuples in increasing order) increases
476 : * prefetching efficiency significantly / decreases the number of cache
477 : * misses.
478 : */
479 1549884 : for (offnum = maxoff;
480 71700234 : offnum >= FirstOffsetNumber;
481 70150350 : offnum = OffsetNumberPrev(offnum))
482 : {
483 70150350 : ItemId itemid = PageGetItemId(page, offnum);
484 : HeapTupleHeader htup;
485 :
486 : /*
487 : * Set the offset number so that we can display it along with any
488 : * error that occurred while processing this tuple.
489 : */
490 70150350 : *off_loc = offnum;
491 :
492 70150350 : prstate->processed[offnum] = false;
493 70150350 : prstate->htsv[offnum] = -1;
494 :
495 : /* Nothing to do if slot doesn't contain a tuple */
496 70150350 : if (!ItemIdIsUsed(itemid))
497 : {
498 643472 : heap_prune_record_unchanged_lp_unused(page, prstate, offnum);
499 643472 : continue;
500 : }
501 :
502 69506878 : if (ItemIdIsDead(itemid))
503 : {
504 : /*
505 : * If the caller set mark_unused_now true, we can set dead line
506 : * pointers LP_UNUSED now.
507 : */
508 1897418 : if (unlikely(prstate->mark_unused_now))
509 1302 : heap_prune_record_unused(prstate, offnum, false);
510 : else
511 1896116 : heap_prune_record_unchanged_lp_dead(page, prstate, offnum);
512 1897418 : continue;
513 : }
514 :
515 67609460 : if (ItemIdIsRedirected(itemid))
516 : {
517 : /* This is the start of a HOT chain */
518 645480 : prstate->root_items[prstate->nroot_items++] = offnum;
519 645480 : continue;
520 : }
521 :
522 : Assert(ItemIdIsNormal(itemid));
523 :
524 : /*
525 : * Get the tuple's visibility status and queue it up for processing.
526 : */
527 66963980 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
528 66963980 : tup.t_data = htup;
529 66963980 : tup.t_len = ItemIdGetLength(itemid);
530 66963980 : ItemPointerSet(&tup.t_self, blockno, offnum);
531 :
532 66963980 : prstate->htsv[offnum] = heap_prune_satisfies_vacuum(prstate, &tup,
533 : buffer);
534 :
535 66963980 : if (!HeapTupleHeaderIsHeapOnly(htup))
536 66081390 : prstate->root_items[prstate->nroot_items++] = offnum;
537 : else
538 882590 : prstate->heaponly_items[prstate->nheaponly_items++] = offnum;
539 : }
540 :
541 : /*
542 : * Process HOT chains.
543 : *
544 : * We added the items to the array starting from 'maxoff', so by
545 : * processing the array in reverse order, we process the items in
546 : * ascending offset number order. The order doesn't matter for
547 : * correctness, but some quick micro-benchmarking suggests that this is
548 : * faster. (Earlier PostgreSQL versions, which scanned all the items on
549 : * the page instead of using the root_items array, also did it in
550 : * ascending offset number order.)
551 : */
552 68276754 : for (int i = prstate->nroot_items - 1; i >= 0; i--)
553 : {
554 66726870 : offnum = prstate->root_items[i];
555 :
556 : /* Ignore items already processed as part of an earlier chain */
557 66726870 : if (prstate->processed[offnum])
558 0 : continue;
559 :
560 : /* see preceding loop */
561 66726870 : *off_loc = offnum;
562 :
563 : /* Process this item or chain of items */
564 66726870 : heap_prune_chain(page, blockno, maxoff, offnum, prstate);
565 : }
566 :
567 : /*
568 : * Process any heap-only tuples that were not already processed as part of
569 : * a HOT chain.
570 : */
571 2432474 : for (int i = prstate->nheaponly_items - 1; i >= 0; i--)
572 : {
573 882590 : offnum = prstate->heaponly_items[i];
574 :
575 882590 : if (prstate->processed[offnum])
576 855942 : continue;
577 :
578 : /* see preceding loop */
579 26648 : *off_loc = offnum;
580 :
581 : /*
582 : * If the tuple is DEAD and doesn't chain to anything else, mark it
583 : * unused. (If it does chain, we can only remove it as part of
584 : * pruning its chain.)
585 : *
586 : * We need this primarily to handle aborted HOT updates, that is,
587 : * XMIN_INVALID heap-only tuples. Those might not be linked to by any
588 : * chain, since the parent tuple might be re-updated before any
589 : * pruning occurs. So we have to be able to reap them separately from
590 : * chain-pruning. (Note that HeapTupleHeaderIsHotUpdated will never
591 : * return true for an XMIN_INVALID tuple, so this code will work even
592 : * when there were sequential updates within the aborted transaction.)
593 : */
594 26648 : if (prstate->htsv[offnum] == HEAPTUPLE_DEAD)
595 : {
596 4140 : ItemId itemid = PageGetItemId(page, offnum);
597 4140 : HeapTupleHeader htup = (HeapTupleHeader) PageGetItem(page, itemid);
598 :
599 4140 : if (likely(!HeapTupleHeaderIsHotUpdated(htup)))
600 : {
601 4140 : HeapTupleHeaderAdvanceConflictHorizon(htup,
602 : &prstate->latest_xid_removed);
603 4140 : heap_prune_record_unused(prstate, offnum, true);
604 : }
605 : else
606 : {
607 : /*
608 : * This tuple should've been processed and removed as part of
609 : * a HOT chain, so something's wrong. To preserve evidence,
610 : * we don't dare to remove it. We cannot leave behind a DEAD
611 : * tuple either, because that will cause VACUUM to error out.
612 : * Throwing an error with a distinct error message seems like
613 : * the least bad option.
614 : */
615 0 : elog(ERROR, "dead heap-only tuple (%u, %d) is not linked to from any HOT chain",
616 : blockno, offnum);
617 : }
618 : }
619 : else
620 22508 : heap_prune_record_unchanged_lp_normal(page, prstate, offnum);
621 : }
622 :
623 : /* We should now have processed every tuple exactly once */
624 : #ifdef USE_ASSERT_CHECKING
625 : for (offnum = FirstOffsetNumber;
626 : offnum <= maxoff;
627 : offnum = OffsetNumberNext(offnum))
628 : {
629 : *off_loc = offnum;
630 :
631 : Assert(prstate->processed[offnum]);
632 : }
633 : #endif
634 :
635 : /* Clear the offset information once we have processed the given page. */
636 1549884 : *off_loc = InvalidOffsetNumber;
637 1549884 : }
638 :
639 : /*
640 : * Decide whether to proceed with freezing according to the freeze plans
641 : * prepared for the given heap buffer. If freezing is chosen, this function
642 : * performs several pre-freeze checks.
643 : *
644 : * The values of do_prune, do_hint_prune, and did_tuple_hint_fpi must be
645 : * determined before calling this function.
646 : *
647 : * prstate is both an input and output parameter.
648 : *
649 : * Returns true if we should apply the freeze plans and freeze tuples on the
650 : * page, and false otherwise.
651 : */
652 : static bool
653 1549884 : heap_page_will_freeze(Relation relation, Buffer buffer,
654 : bool did_tuple_hint_fpi,
655 : bool do_prune,
656 : bool do_hint_prune,
657 : PruneState *prstate)
658 : {
659 1549884 : bool do_freeze = false;
660 :
661 : /*
662 : * If the caller specified we should not attempt to freeze any tuples,
663 : * validate that everything is in the right state and return.
664 : */
665 1549884 : if (!prstate->attempt_freeze)
666 : {
667 : Assert(!prstate->all_frozen && prstate->nfrozen == 0);
668 : Assert(prstate->lpdead_items == 0 || !prstate->all_visible);
669 82994 : return false;
670 : }
671 :
672 1466890 : if (prstate->pagefrz.freeze_required)
673 : {
674 : /*
675 : * heap_prepare_freeze_tuple indicated that at least one XID/MXID from
676 : * before FreezeLimit/MultiXactCutoff is present. Must freeze to
677 : * advance relfrozenxid/relminmxid.
678 : */
679 41652 : do_freeze = true;
680 : }
681 : else
682 : {
683 : /*
684 : * Opportunistically freeze the page if we are generating an FPI
685 : * anyway and if doing so means that we can set the page all-frozen
686 : * afterwards (might not happen until VACUUM's final heap pass).
687 : *
688 : * XXX: Previously, we knew if pruning emitted an FPI by checking
689 : * pgWalUsage.wal_fpi before and after pruning. Once the freeze and
690 : * prune records were combined, this heuristic couldn't be used
691 : * anymore. The opportunistic freeze heuristic must be improved;
692 : * however, for now, try to approximate the old logic.
693 : */
694 1425238 : if (prstate->all_frozen && prstate->nfrozen > 0)
695 : {
696 : Assert(prstate->all_visible);
697 :
698 : /*
699 : * Freezing would make the page all-frozen. Have already emitted
700 : * an FPI or will do so anyway?
701 : */
702 29498 : if (RelationNeedsWAL(relation))
703 : {
704 26188 : if (did_tuple_hint_fpi)
705 3218 : do_freeze = true;
706 22970 : else if (do_prune)
707 : {
708 2070 : if (XLogCheckBufferNeedsBackup(buffer))
709 1160 : do_freeze = true;
710 : }
711 20900 : else if (do_hint_prune)
712 : {
713 16 : if (XLogHintBitIsNeeded() && XLogCheckBufferNeedsBackup(buffer))
714 10 : do_freeze = true;
715 : }
716 : }
717 : }
718 : }
719 :
720 1466890 : if (do_freeze)
721 : {
722 : /*
723 : * Validate the tuples we will be freezing before entering the
724 : * critical section.
725 : */
726 46040 : heap_pre_freeze_checks(buffer, prstate->frozen, prstate->nfrozen);
727 :
728 : /*
729 : * Calculate what the snapshot conflict horizon should be for a record
730 : * freezing tuples. We can use the visibility_cutoff_xid as our cutoff
731 : * for conflicts when the whole page is eligible to become all-frozen
732 : * in the VM once we're done with it. Otherwise, we generate a
733 : * conservative cutoff by stepping back from OldestXmin.
734 : */
735 46040 : if (prstate->all_frozen)
736 41492 : prstate->frz_conflict_horizon = prstate->visibility_cutoff_xid;
737 : else
738 : {
739 : /* Avoids false conflicts when hot_standby_feedback in use */
740 4548 : prstate->frz_conflict_horizon = prstate->cutoffs->OldestXmin;
741 4548 : TransactionIdRetreat(prstate->frz_conflict_horizon);
742 : }
743 : }
744 1420850 : else if (prstate->nfrozen > 0)
745 : {
746 : /*
747 : * The page contained some tuples that were not already frozen, and we
748 : * chose not to freeze them now. The page won't be all-frozen then.
749 : */
750 : Assert(!prstate->pagefrz.freeze_required);
751 :
752 26658 : prstate->all_frozen = false;
753 26658 : prstate->nfrozen = 0; /* avoid miscounts in instrumentation */
754 : }
755 : else
756 : {
757 : /*
758 : * We have no freeze plans to execute. The page might already be
759 : * all-frozen (perhaps only following pruning), though. Such pages
760 : * can be marked all-frozen in the VM by our caller, even though none
761 : * of its tuples were newly frozen here.
762 : */
763 : }
764 :
765 1466890 : return do_freeze;
766 : }
767 :
768 :
769 : /*
770 : * Prune and repair fragmentation and potentially freeze tuples on the
771 : * specified page.
772 : *
773 : * Caller must have pin and buffer cleanup lock on the page. Note that we
774 : * don't update the FSM information for page on caller's behalf. Caller might
775 : * also need to account for a reduction in the length of the line pointer
776 : * array following array truncation by us.
777 : *
778 : * params contains the input parameters used to control freezing and pruning
779 : * behavior. See the definition of PruneFreezeParams for more on what each
780 : * parameter does.
781 : *
782 : * If the HEAP_PAGE_PRUNE_FREEZE option is set in params, we will freeze
783 : * tuples if it's required in order to advance relfrozenxid / relminmxid, or
784 : * if it's considered advantageous for overall system performance to do so
785 : * now. The 'params.cutoffs', 'presult', 'new_relfrozen_xid' and
786 : * 'new_relmin_mxid' arguments are required when freezing. When
787 : * HEAP_PAGE_PRUNE_FREEZE option is passed, we also set presult->all_visible
788 : * and presult->all_frozen after determining whether or not to
789 : * opportunistically freeze, to indicate if the VM bits can be set. They are
790 : * always set to false when the HEAP_PAGE_PRUNE_FREEZE option is not passed,
791 : * because at the moment only callers that also freeze need that information.
792 : *
793 : * presult contains output parameters needed by callers, such as the number of
794 : * tuples removed and the offsets of dead items on the page after pruning.
795 : * heap_page_prune_and_freeze() is responsible for initializing it. Required
796 : * by all callers.
797 : *
798 : * off_loc is the offset location required by the caller to use in error
799 : * callback.
800 : *
801 : * new_relfrozen_xid and new_relmin_mxid must be provided by the caller if the
802 : * HEAP_PAGE_PRUNE_FREEZE option is set in params. On entry, they contain the
803 : * oldest XID and multi-XID seen on the relation so far. They will be updated
804 : * with the oldest values present on the page after pruning. After processing
805 : * the whole relation, VACUUM can use these values as the new
806 : * relfrozenxid/relminmxid for the relation.
807 : */
808 : void
809 1549884 : heap_page_prune_and_freeze(PruneFreezeParams *params,
810 : PruneFreezeResult *presult,
811 : OffsetNumber *off_loc,
812 : TransactionId *new_relfrozen_xid,
813 : MultiXactId *new_relmin_mxid)
814 : {
815 1549884 : Buffer buffer = params->buffer;
816 1549884 : Page page = BufferGetPage(buffer);
817 : PruneState prstate;
818 : bool do_freeze;
819 : bool do_prune;
820 : bool do_hint_prune;
821 : bool did_tuple_hint_fpi;
822 1549884 : int64 fpi_before = pgWalUsage.wal_fpi;
823 :
824 : /* Initialize prstate */
825 1549884 : prune_freeze_setup(params,
826 : new_relfrozen_xid, new_relmin_mxid,
827 : presult, &prstate);
828 :
829 : /*
830 : * Examine all line pointers and tuple visibility information to determine
831 : * which line pointers should change state and which tuples may be frozen.
832 : * Prepare queue of state changes to later be executed in a critical
833 : * section.
834 : */
835 1549884 : prune_freeze_plan(RelationGetRelid(params->relation),
836 : buffer, &prstate, off_loc);
837 :
838 : /*
839 : * If checksums are enabled, calling heap_prune_satisfies_vacuum() while
840 : * checking tuple visibility information in prune_freeze_plan() may have
841 : * caused an FPI to be emitted.
842 : */
843 1549884 : did_tuple_hint_fpi = fpi_before != pgWalUsage.wal_fpi;
844 :
845 4616382 : do_prune = prstate.nredirected > 0 ||
846 2995982 : prstate.ndead > 0 ||
847 1446098 : prstate.nunused > 0;
848 :
849 : /*
850 : * Even if we don't prune anything, if we found a new value for the
851 : * pd_prune_xid field or the page was marked full, we will update the hint
852 : * bit.
853 : */
854 2995336 : do_hint_prune = ((PageHeader) page)->pd_prune_xid != prstate.new_prune_xid ||
855 1445452 : PageIsFull(page);
856 :
857 : /*
858 : * Decide if we want to go ahead with freezing according to the freeze
859 : * plans we prepared, or not.
860 : */
861 1549884 : do_freeze = heap_page_will_freeze(params->relation, buffer,
862 : did_tuple_hint_fpi,
863 : do_prune,
864 : do_hint_prune,
865 : &prstate);
866 :
867 : /*
868 : * While scanning the line pointers, we did not clear
869 : * all_visible/all_frozen when encountering LP_DEAD items because we
870 : * wanted the decision whether or not to freeze the page to be unaffected
871 : * by the short-term presence of LP_DEAD items. These LP_DEAD items are
872 : * effectively assumed to be LP_UNUSED items in the making. It doesn't
873 : * matter which vacuum heap pass (initial pass or final pass) ends up
874 : * setting the page all-frozen, as long as the ongoing VACUUM does it.
875 : *
876 : * Now that we finished determining whether or not to freeze the page,
877 : * update all_visible and all_frozen so that they reflect the true state
878 : * of the page for setting PD_ALL_VISIBLE and VM bits.
879 : */
880 1549884 : if (prstate.lpdead_items > 0)
881 105056 : prstate.all_visible = prstate.all_frozen = false;
882 :
883 : Assert(!prstate.all_frozen || prstate.all_visible);
884 :
885 : /* Any error while applying the changes is critical */
886 1549884 : START_CRIT_SECTION();
887 :
888 1549884 : if (do_hint_prune)
889 : {
890 : /*
891 : * Update the page's pd_prune_xid field to either zero, or the lowest
892 : * XID of any soon-prunable tuple.
893 : */
894 104516 : ((PageHeader) page)->pd_prune_xid = prstate.new_prune_xid;
895 :
896 : /*
897 : * Also clear the "page is full" flag, since there's no point in
898 : * repeating the prune/defrag process until something else happens to
899 : * the page.
900 : */
901 104516 : PageClearFull(page);
902 :
903 : /*
904 : * If that's all we had to do to the page, this is a non-WAL-logged
905 : * hint. If we are going to freeze or prune the page, we will mark
906 : * the buffer dirty below.
907 : */
908 104516 : if (!do_freeze && !do_prune)
909 360 : MarkBufferDirtyHint(buffer, true);
910 : }
911 :
912 1549884 : if (do_prune || do_freeze)
913 : {
914 : /* Apply the planned item changes and repair page fragmentation. */
915 146834 : if (do_prune)
916 : {
917 104554 : heap_page_prune_execute(buffer, false,
918 : prstate.redirected, prstate.nredirected,
919 : prstate.nowdead, prstate.ndead,
920 : prstate.nowunused, prstate.nunused);
921 : }
922 :
923 146834 : if (do_freeze)
924 46040 : heap_freeze_prepared_tuples(buffer, prstate.frozen, prstate.nfrozen);
925 :
926 146834 : MarkBufferDirty(buffer);
927 :
928 : /*
929 : * Emit a WAL XLOG_HEAP2_PRUNE* record showing what we did
930 : */
931 146834 : if (RelationNeedsWAL(params->relation))
932 : {
933 : /*
934 : * The snapshotConflictHorizon for the whole record should be the
935 : * most conservative of all the horizons calculated for any of the
936 : * possible modifications. If this record will prune tuples, any
937 : * transactions on the standby older than the youngest xmax of the
938 : * most recently removed tuple this record will prune will
939 : * conflict. If this record will freeze tuples, any transactions
940 : * on the standby with xids older than the youngest tuple this
941 : * record will freeze will conflict.
942 : */
943 : TransactionId conflict_xid;
944 :
945 145080 : if (TransactionIdFollows(prstate.frz_conflict_horizon,
946 : prstate.latest_xid_removed))
947 42934 : conflict_xid = prstate.frz_conflict_horizon;
948 : else
949 102146 : conflict_xid = prstate.latest_xid_removed;
950 :
951 145080 : log_heap_prune_and_freeze(params->relation, buffer,
952 : InvalidBuffer, /* vmbuffer */
953 : 0, /* vmflags */
954 : conflict_xid,
955 : true, params->reason,
956 : prstate.frozen, prstate.nfrozen,
957 : prstate.redirected, prstate.nredirected,
958 : prstate.nowdead, prstate.ndead,
959 : prstate.nowunused, prstate.nunused);
960 : }
961 : }
962 :
963 1549884 : END_CRIT_SECTION();
964 :
965 : /* Copy information back for caller */
966 1549884 : presult->ndeleted = prstate.ndeleted;
967 1549884 : presult->nnewlpdead = prstate.ndead;
968 1549884 : presult->nfrozen = prstate.nfrozen;
969 1549884 : presult->live_tuples = prstate.live_tuples;
970 1549884 : presult->recently_dead_tuples = prstate.recently_dead_tuples;
971 1549884 : presult->all_visible = prstate.all_visible;
972 1549884 : presult->all_frozen = prstate.all_frozen;
973 1549884 : presult->hastup = prstate.hastup;
974 :
975 : /*
976 : * For callers planning to update the visibility map, the conflict horizon
977 : * for that record must be the newest xmin on the page. However, if the
978 : * page is completely frozen, there can be no conflict and the
979 : * vm_conflict_horizon should remain InvalidTransactionId. This includes
980 : * the case that we just froze all the tuples; the prune-freeze record
981 : * included the conflict XID already so the caller doesn't need it.
982 : */
983 1549884 : if (presult->all_frozen)
984 443516 : presult->vm_conflict_horizon = InvalidTransactionId;
985 : else
986 1106368 : presult->vm_conflict_horizon = prstate.visibility_cutoff_xid;
987 :
988 1549884 : presult->lpdead_items = prstate.lpdead_items;
989 : /* the presult->deadoffsets array was already filled in */
990 :
991 1549884 : if (prstate.attempt_freeze)
992 : {
993 1466890 : if (presult->nfrozen > 0)
994 : {
995 46040 : *new_relfrozen_xid = prstate.pagefrz.FreezePageRelfrozenXid;
996 46040 : *new_relmin_mxid = prstate.pagefrz.FreezePageRelminMxid;
997 : }
998 : else
999 : {
1000 1420850 : *new_relfrozen_xid = prstate.pagefrz.NoFreezePageRelfrozenXid;
1001 1420850 : *new_relmin_mxid = prstate.pagefrz.NoFreezePageRelminMxid;
1002 : }
1003 : }
1004 1549884 : }
1005 :
1006 :
1007 : /*
1008 : * Perform visibility checks for heap pruning.
1009 : */
1010 : static HTSV_Result
1011 66963980 : heap_prune_satisfies_vacuum(PruneState *prstate, HeapTuple tup, Buffer buffer)
1012 : {
1013 : HTSV_Result res;
1014 : TransactionId dead_after;
1015 :
1016 66963980 : res = HeapTupleSatisfiesVacuumHorizon(tup, buffer, &dead_after);
1017 :
1018 66963980 : if (res != HEAPTUPLE_RECENTLY_DEAD)
1019 63690284 : return res;
1020 :
1021 : /*
1022 : * For VACUUM, we must be sure to prune tuples with xmax older than
1023 : * OldestXmin -- a visibility cutoff determined at the beginning of
1024 : * vacuuming the relation. OldestXmin is used for freezing determination
1025 : * and we cannot freeze dead tuples' xmaxes.
1026 : */
1027 3273696 : if (prstate->cutoffs &&
1028 1834656 : TransactionIdIsValid(prstate->cutoffs->OldestXmin) &&
1029 1834656 : NormalTransactionIdPrecedes(dead_after, prstate->cutoffs->OldestXmin))
1030 1329296 : return HEAPTUPLE_DEAD;
1031 :
1032 : /*
1033 : * Determine whether or not the tuple is considered dead when compared
1034 : * with the provided GlobalVisState. On-access pruning does not provide
1035 : * VacuumCutoffs. And for vacuum, even if the tuple's xmax is not older
1036 : * than OldestXmin, GlobalVisTestIsRemovableXid() could find the row dead
1037 : * if the GlobalVisState has been updated since the beginning of vacuuming
1038 : * the relation.
1039 : */
1040 1944400 : if (GlobalVisTestIsRemovableXid(prstate->vistest, dead_after))
1041 1382516 : return HEAPTUPLE_DEAD;
1042 :
1043 561884 : return res;
1044 : }
1045 :
1046 :
1047 : /*
1048 : * Pruning calculates tuple visibility once and saves the results in an array
1049 : * of int8. See PruneState.htsv for details. This helper function is meant
1050 : * to guard against examining visibility status array members which have not
1051 : * yet been computed.
1052 : */
1053 : static inline HTSV_Result
1054 66937332 : htsv_get_valid_status(int status)
1055 : {
1056 : Assert(status >= HEAPTUPLE_DEAD &&
1057 : status <= HEAPTUPLE_DELETE_IN_PROGRESS);
1058 66937332 : return (HTSV_Result) status;
1059 : }
1060 :
1061 : /*
1062 : * Prune specified line pointer or a HOT chain originating at line pointer.
1063 : *
1064 : * Tuple visibility information is provided in prstate->htsv.
1065 : *
1066 : * If the item is an index-referenced tuple (i.e. not a heap-only tuple),
1067 : * the HOT chain is pruned by removing all DEAD tuples at the start of the HOT
1068 : * chain. We also prune any RECENTLY_DEAD tuples preceding a DEAD tuple.
1069 : * This is OK because a RECENTLY_DEAD tuple preceding a DEAD tuple is really
1070 : * DEAD, our visibility test is just too coarse to detect it.
1071 : *
1072 : * Pruning must never leave behind a DEAD tuple that still has tuple storage.
1073 : * VACUUM isn't prepared to deal with that case.
1074 : *
1075 : * The root line pointer is redirected to the tuple immediately after the
1076 : * latest DEAD tuple. If all tuples in the chain are DEAD, the root line
1077 : * pointer is marked LP_DEAD. (This includes the case of a DEAD simple
1078 : * tuple, which we treat as a chain of length 1.)
1079 : *
1080 : * We don't actually change the page here. We just add entries to the arrays in
1081 : * prstate showing the changes to be made. Items to be redirected are added
1082 : * to the redirected[] array (two entries per redirection); items to be set to
1083 : * LP_DEAD state are added to nowdead[]; and items to be set to LP_UNUSED
1084 : * state are added to nowunused[]. We perform bookkeeping of live tuples,
1085 : * visibility etc. based on what the page will look like after the changes
1086 : * applied. All that bookkeeping is performed in the heap_prune_record_*()
1087 : * subroutines. The division of labor is that heap_prune_chain() decides the
1088 : * fate of each tuple, ie. whether it's going to be removed, redirected or
1089 : * left unchanged, and the heap_prune_record_*() subroutines update PruneState
1090 : * based on that outcome.
1091 : */
1092 : static void
1093 66726870 : heap_prune_chain(Page page, BlockNumber blockno, OffsetNumber maxoff,
1094 : OffsetNumber rootoffnum, PruneState *prstate)
1095 : {
1096 66726870 : TransactionId priorXmax = InvalidTransactionId;
1097 : ItemId rootlp;
1098 : OffsetNumber offnum;
1099 : OffsetNumber chainitems[MaxHeapTuplesPerPage];
1100 :
1101 : /*
1102 : * After traversing the HOT chain, ndeadchain is the index in chainitems
1103 : * of the first live successor after the last dead item.
1104 : */
1105 66726870 : int ndeadchain = 0,
1106 66726870 : nchain = 0;
1107 :
1108 66726870 : rootlp = PageGetItemId(page, rootoffnum);
1109 :
1110 : /* Start from the root tuple */
1111 66726870 : offnum = rootoffnum;
1112 :
1113 : /* while not end of the chain */
1114 : for (;;)
1115 855942 : {
1116 : HeapTupleHeader htup;
1117 : ItemId lp;
1118 :
1119 : /* Sanity check (pure paranoia) */
1120 67582812 : if (offnum < FirstOffsetNumber)
1121 0 : break;
1122 :
1123 : /*
1124 : * An offset past the end of page's line pointer array is possible
1125 : * when the array was truncated (original item must have been unused)
1126 : */
1127 67582812 : if (offnum > maxoff)
1128 0 : break;
1129 :
1130 : /* If item is already processed, stop --- it must not be same chain */
1131 67582812 : if (prstate->processed[offnum])
1132 0 : break;
1133 :
1134 67582812 : lp = PageGetItemId(page, offnum);
1135 :
1136 : /*
1137 : * Unused item obviously isn't part of the chain. Likewise, a dead
1138 : * line pointer can't be part of the chain. Both of those cases were
1139 : * already marked as processed.
1140 : */
1141 : Assert(ItemIdIsUsed(lp));
1142 : Assert(!ItemIdIsDead(lp));
1143 :
1144 : /*
1145 : * If we are looking at the redirected root line pointer, jump to the
1146 : * first normal tuple in the chain. If we find a redirect somewhere
1147 : * else, stop --- it must not be same chain.
1148 : */
1149 67582812 : if (ItemIdIsRedirected(lp))
1150 : {
1151 645480 : if (nchain > 0)
1152 0 : break; /* not at start of chain */
1153 645480 : chainitems[nchain++] = offnum;
1154 645480 : offnum = ItemIdGetRedirect(rootlp);
1155 645480 : continue;
1156 : }
1157 :
1158 : Assert(ItemIdIsNormal(lp));
1159 :
1160 66937332 : htup = (HeapTupleHeader) PageGetItem(page, lp);
1161 :
1162 : /*
1163 : * Check the tuple XMIN against prior XMAX, if any
1164 : */
1165 67147794 : if (TransactionIdIsValid(priorXmax) &&
1166 210462 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
1167 0 : break;
1168 :
1169 : /*
1170 : * OK, this tuple is indeed a member of the chain.
1171 : */
1172 66937332 : chainitems[nchain++] = offnum;
1173 :
1174 66937332 : switch (htsv_get_valid_status(prstate->htsv[offnum]))
1175 : {
1176 2794982 : case HEAPTUPLE_DEAD:
1177 :
1178 : /* Remember the last DEAD tuple seen */
1179 2794982 : ndeadchain = nchain;
1180 2794982 : HeapTupleHeaderAdvanceConflictHorizon(htup,
1181 : &prstate->latest_xid_removed);
1182 : /* Advance to next chain member */
1183 2794982 : break;
1184 :
1185 561884 : case HEAPTUPLE_RECENTLY_DEAD:
1186 :
1187 : /*
1188 : * We don't need to advance the conflict horizon for
1189 : * RECENTLY_DEAD tuples, even if we are removing them. This
1190 : * is because we only remove RECENTLY_DEAD tuples if they
1191 : * precede a DEAD tuple, and the DEAD tuple must have been
1192 : * inserted by a newer transaction than the RECENTLY_DEAD
1193 : * tuple by virtue of being later in the chain. We will have
1194 : * advanced the conflict horizon for the DEAD tuple.
1195 : */
1196 :
1197 : /*
1198 : * Advance past RECENTLY_DEAD tuples just in case there's a
1199 : * DEAD one after them. We have to make sure that we don't
1200 : * miss any DEAD tuples, since DEAD tuples that still have
1201 : * tuple storage after pruning will confuse VACUUM.
1202 : */
1203 561884 : break;
1204 :
1205 63580466 : case HEAPTUPLE_DELETE_IN_PROGRESS:
1206 : case HEAPTUPLE_LIVE:
1207 : case HEAPTUPLE_INSERT_IN_PROGRESS:
1208 63580466 : goto process_chain;
1209 :
1210 0 : default:
1211 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1212 : goto process_chain;
1213 : }
1214 :
1215 : /*
1216 : * If the tuple is not HOT-updated, then we are at the end of this
1217 : * HOT-update chain.
1218 : */
1219 3356866 : if (!HeapTupleHeaderIsHotUpdated(htup))
1220 3146404 : goto process_chain;
1221 :
1222 : /* HOT implies it can't have moved to different partition */
1223 : Assert(!HeapTupleHeaderIndicatesMovedPartitions(htup));
1224 :
1225 : /*
1226 : * Advance to next chain member.
1227 : */
1228 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blockno);
1229 210462 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
1230 210462 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
1231 : }
1232 :
1233 0 : if (ItemIdIsRedirected(rootlp) && nchain < 2)
1234 : {
1235 : /*
1236 : * We found a redirect item that doesn't point to a valid follow-on
1237 : * item. This can happen if the loop in heap_page_prune_and_freeze()
1238 : * caused us to visit the dead successor of a redirect item before
1239 : * visiting the redirect item. We can clean up by setting the
1240 : * redirect item to LP_DEAD state or LP_UNUSED if the caller
1241 : * indicated.
1242 : */
1243 0 : heap_prune_record_dead_or_unused(prstate, rootoffnum, false);
1244 0 : return;
1245 : }
1246 :
1247 0 : process_chain:
1248 :
1249 66726870 : if (ndeadchain == 0)
1250 : {
1251 : /*
1252 : * No DEAD tuple was found, so the chain is entirely composed of
1253 : * normal, unchanged tuples. Leave it alone.
1254 : */
1255 64000828 : int i = 0;
1256 :
1257 64000828 : if (ItemIdIsRedirected(rootlp))
1258 : {
1259 612122 : heap_prune_record_unchanged_lp_redirect(prstate, rootoffnum);
1260 612122 : i++;
1261 : }
1262 128009880 : for (; i < nchain; i++)
1263 64009052 : heap_prune_record_unchanged_lp_normal(page, prstate, chainitems[i]);
1264 : }
1265 2726042 : else if (ndeadchain == nchain)
1266 : {
1267 : /*
1268 : * The entire chain is dead. Mark the root line pointer LP_DEAD, and
1269 : * fully remove the other tuples in the chain.
1270 : */
1271 2596382 : heap_prune_record_dead_or_unused(prstate, rootoffnum, ItemIdIsNormal(rootlp));
1272 2661950 : for (int i = 1; i < nchain; i++)
1273 65568 : heap_prune_record_unused(prstate, chainitems[i], true);
1274 : }
1275 : else
1276 : {
1277 : /*
1278 : * We found a DEAD tuple in the chain. Redirect the root line pointer
1279 : * to the first non-DEAD tuple, and mark as unused each intermediate
1280 : * item that we are able to remove from the chain.
1281 : */
1282 129660 : heap_prune_record_redirect(prstate, rootoffnum, chainitems[ndeadchain],
1283 129660 : ItemIdIsNormal(rootlp));
1284 166390 : for (int i = 1; i < ndeadchain; i++)
1285 36730 : heap_prune_record_unused(prstate, chainitems[i], true);
1286 :
1287 : /* the rest of tuples in the chain are normal, unchanged tuples */
1288 262958 : for (int i = ndeadchain; i < nchain; i++)
1289 133298 : heap_prune_record_unchanged_lp_normal(page, prstate, chainitems[i]);
1290 : }
1291 : }
1292 :
1293 : /* Record lowest soon-prunable XID */
1294 : static void
1295 24596638 : heap_prune_record_prunable(PruneState *prstate, TransactionId xid)
1296 : {
1297 : /*
1298 : * This should exactly match the PageSetPrunable macro. We can't store
1299 : * directly into the page header yet, so we update working state.
1300 : */
1301 : Assert(TransactionIdIsNormal(xid));
1302 48216796 : if (!TransactionIdIsValid(prstate->new_prune_xid) ||
1303 23620158 : TransactionIdPrecedes(xid, prstate->new_prune_xid))
1304 978454 : prstate->new_prune_xid = xid;
1305 24596638 : }
1306 :
1307 : /* Record line pointer to be redirected */
1308 : static void
1309 129660 : heap_prune_record_redirect(PruneState *prstate,
1310 : OffsetNumber offnum, OffsetNumber rdoffnum,
1311 : bool was_normal)
1312 : {
1313 : Assert(!prstate->processed[offnum]);
1314 129660 : prstate->processed[offnum] = true;
1315 :
1316 : /*
1317 : * Do not mark the redirect target here. It needs to be counted
1318 : * separately as an unchanged tuple.
1319 : */
1320 :
1321 : Assert(prstate->nredirected < MaxHeapTuplesPerPage);
1322 129660 : prstate->redirected[prstate->nredirected * 2] = offnum;
1323 129660 : prstate->redirected[prstate->nredirected * 2 + 1] = rdoffnum;
1324 :
1325 129660 : prstate->nredirected++;
1326 :
1327 : /*
1328 : * If the root entry had been a normal tuple, we are deleting it, so count
1329 : * it in the result. But changing a redirect (even to DEAD state) doesn't
1330 : * count.
1331 : */
1332 129660 : if (was_normal)
1333 114394 : prstate->ndeleted++;
1334 :
1335 129660 : prstate->hastup = true;
1336 129660 : }
1337 :
1338 : /* Record line pointer to be marked dead */
1339 : static void
1340 2527550 : heap_prune_record_dead(PruneState *prstate, OffsetNumber offnum,
1341 : bool was_normal)
1342 : {
1343 : Assert(!prstate->processed[offnum]);
1344 2527550 : prstate->processed[offnum] = true;
1345 :
1346 : Assert(prstate->ndead < MaxHeapTuplesPerPage);
1347 2527550 : prstate->nowdead[prstate->ndead] = offnum;
1348 2527550 : prstate->ndead++;
1349 :
1350 : /*
1351 : * Deliberately delay unsetting all_visible and all_frozen until later
1352 : * during pruning. Removable dead tuples shouldn't preclude freezing the
1353 : * page.
1354 : */
1355 :
1356 : /* Record the dead offset for vacuum */
1357 2527550 : prstate->deadoffsets[prstate->lpdead_items++] = offnum;
1358 :
1359 : /*
1360 : * If the root entry had been a normal tuple, we are deleting it, so count
1361 : * it in the result. But changing a redirect (even to DEAD state) doesn't
1362 : * count.
1363 : */
1364 2527550 : if (was_normal)
1365 2509458 : prstate->ndeleted++;
1366 2527550 : }
1367 :
1368 : /*
1369 : * Depending on whether or not the caller set mark_unused_now to true, record that a
1370 : * line pointer should be marked LP_DEAD or LP_UNUSED. There are other cases in
1371 : * which we will mark line pointers LP_UNUSED, but we will not mark line
1372 : * pointers LP_DEAD if mark_unused_now is true.
1373 : */
1374 : static void
1375 2596382 : heap_prune_record_dead_or_unused(PruneState *prstate, OffsetNumber offnum,
1376 : bool was_normal)
1377 : {
1378 : /*
1379 : * If the caller set mark_unused_now to true, we can remove dead tuples
1380 : * during pruning instead of marking their line pointers dead. Set this
1381 : * tuple's line pointer LP_UNUSED. We hint that this option is less
1382 : * likely.
1383 : */
1384 2596382 : if (unlikely(prstate->mark_unused_now))
1385 68832 : heap_prune_record_unused(prstate, offnum, was_normal);
1386 : else
1387 2527550 : heap_prune_record_dead(prstate, offnum, was_normal);
1388 2596382 : }
1389 :
1390 : /* Record line pointer to be marked unused */
1391 : static void
1392 176572 : heap_prune_record_unused(PruneState *prstate, OffsetNumber offnum, bool was_normal)
1393 : {
1394 : Assert(!prstate->processed[offnum]);
1395 176572 : prstate->processed[offnum] = true;
1396 :
1397 : Assert(prstate->nunused < MaxHeapTuplesPerPage);
1398 176572 : prstate->nowunused[prstate->nunused] = offnum;
1399 176572 : prstate->nunused++;
1400 :
1401 : /*
1402 : * If the root entry had been a normal tuple, we are deleting it, so count
1403 : * it in the result. But changing a redirect (even to DEAD state) doesn't
1404 : * count.
1405 : */
1406 176572 : if (was_normal)
1407 175270 : prstate->ndeleted++;
1408 176572 : }
1409 :
1410 : /*
1411 : * Record an unused line pointer that is left unchanged.
1412 : */
1413 : static void
1414 643472 : heap_prune_record_unchanged_lp_unused(Page page, PruneState *prstate, OffsetNumber offnum)
1415 : {
1416 : Assert(!prstate->processed[offnum]);
1417 643472 : prstate->processed[offnum] = true;
1418 643472 : }
1419 :
1420 : /*
1421 : * Record line pointer that is left unchanged. We consider freezing it, and
1422 : * update bookkeeping of tuple counts and page visibility.
1423 : */
1424 : static void
1425 64164858 : heap_prune_record_unchanged_lp_normal(Page page, PruneState *prstate, OffsetNumber offnum)
1426 : {
1427 : HeapTupleHeader htup;
1428 :
1429 : Assert(!prstate->processed[offnum]);
1430 64164858 : prstate->processed[offnum] = true;
1431 :
1432 64164858 : prstate->hastup = true; /* the page is not empty */
1433 :
1434 : /*
1435 : * The criteria for counting a tuple as live in this block need to match
1436 : * what analyze.c's acquire_sample_rows() does, otherwise VACUUM and
1437 : * ANALYZE may produce wildly different reltuples values, e.g. when there
1438 : * are many recently-dead tuples.
1439 : *
1440 : * The logic here is a bit simpler than acquire_sample_rows(), as VACUUM
1441 : * can't run inside a transaction block, which makes some cases impossible
1442 : * (e.g. in-progress insert from the same transaction).
1443 : *
1444 : * HEAPTUPLE_DEAD are handled by the other heap_prune_record_*()
1445 : * subroutines. They don't count dead items like acquire_sample_rows()
1446 : * does, because we assume that all dead items will become LP_UNUSED
1447 : * before VACUUM finishes. This difference is only superficial. VACUUM
1448 : * effectively agrees with ANALYZE about DEAD items, in the end. VACUUM
1449 : * won't remember LP_DEAD items, but only because they're not supposed to
1450 : * be left behind when it is done. (Cases where we bypass index vacuuming
1451 : * will violate this optimistic assumption, but the overall impact of that
1452 : * should be negligible.)
1453 : */
1454 64164858 : htup = (HeapTupleHeader) PageGetItem(page, PageGetItemId(page, offnum));
1455 :
1456 64164858 : switch (prstate->htsv[offnum])
1457 : {
1458 39443586 : case HEAPTUPLE_LIVE:
1459 :
1460 : /*
1461 : * Count it as live. Not only is this natural, but it's also what
1462 : * acquire_sample_rows() does.
1463 : */
1464 39443586 : prstate->live_tuples++;
1465 :
1466 : /*
1467 : * Is the tuple definitely visible to all transactions?
1468 : *
1469 : * NB: Like with per-tuple hint bits, we can't set the
1470 : * PD_ALL_VISIBLE flag if the inserter committed asynchronously.
1471 : * See SetHintBits for more info. Check that the tuple is hinted
1472 : * xmin-committed because of that.
1473 : */
1474 39443586 : if (prstate->all_visible)
1475 : {
1476 : TransactionId xmin;
1477 :
1478 26737832 : if (!HeapTupleHeaderXminCommitted(htup))
1479 : {
1480 314 : prstate->all_visible = false;
1481 314 : prstate->all_frozen = false;
1482 314 : break;
1483 : }
1484 :
1485 : /*
1486 : * The inserter definitely committed. But is it old enough
1487 : * that everyone sees it as committed? A FrozenTransactionId
1488 : * is seen as committed to everyone. Otherwise, we check if
1489 : * there is a snapshot that considers this xid to still be
1490 : * running, and if so, we don't consider the page all-visible.
1491 : */
1492 26737518 : xmin = HeapTupleHeaderGetXmin(htup);
1493 :
1494 : /*
1495 : * For now always use prstate->cutoffs for this test, because
1496 : * we only update 'all_visible' and 'all_frozen' when freezing
1497 : * is requested. We could use GlobalVisTestIsRemovableXid
1498 : * instead, if a non-freezing caller wanted to set the VM bit.
1499 : */
1500 : Assert(prstate->cutoffs);
1501 26737518 : if (!TransactionIdPrecedes(xmin, prstate->cutoffs->OldestXmin))
1502 : {
1503 6458 : prstate->all_visible = false;
1504 6458 : prstate->all_frozen = false;
1505 6458 : break;
1506 : }
1507 :
1508 : /* Track newest xmin on page. */
1509 26731060 : if (TransactionIdFollows(xmin, prstate->visibility_cutoff_xid) &&
1510 : TransactionIdIsNormal(xmin))
1511 214720 : prstate->visibility_cutoff_xid = xmin;
1512 : }
1513 39436814 : break;
1514 :
1515 561884 : case HEAPTUPLE_RECENTLY_DEAD:
1516 561884 : prstate->recently_dead_tuples++;
1517 561884 : prstate->all_visible = false;
1518 561884 : prstate->all_frozen = false;
1519 :
1520 : /*
1521 : * This tuple will soon become DEAD. Update the hint field so
1522 : * that the page is reconsidered for pruning in future.
1523 : */
1524 561884 : heap_prune_record_prunable(prstate,
1525 : HeapTupleHeaderGetUpdateXid(htup));
1526 561884 : break;
1527 :
1528 124634 : case HEAPTUPLE_INSERT_IN_PROGRESS:
1529 :
1530 : /*
1531 : * We do not count these rows as live, because we expect the
1532 : * inserting transaction to update the counters at commit, and we
1533 : * assume that will happen only after we report our results. This
1534 : * assumption is a bit shaky, but it is what acquire_sample_rows()
1535 : * does, so be consistent.
1536 : */
1537 124634 : prstate->all_visible = false;
1538 124634 : prstate->all_frozen = false;
1539 :
1540 : /*
1541 : * If we wanted to optimize for aborts, we might consider marking
1542 : * the page prunable when we see INSERT_IN_PROGRESS. But we
1543 : * don't. See related decisions about when to mark the page
1544 : * prunable in heapam.c.
1545 : */
1546 124634 : break;
1547 :
1548 24034754 : case HEAPTUPLE_DELETE_IN_PROGRESS:
1549 :
1550 : /*
1551 : * This an expected case during concurrent vacuum. Count such
1552 : * rows as live. As above, we assume the deleting transaction
1553 : * will commit and update the counters after we report.
1554 : */
1555 24034754 : prstate->live_tuples++;
1556 24034754 : prstate->all_visible = false;
1557 24034754 : prstate->all_frozen = false;
1558 :
1559 : /*
1560 : * This tuple may soon become DEAD. Update the hint field so that
1561 : * the page is reconsidered for pruning in future.
1562 : */
1563 24034754 : heap_prune_record_prunable(prstate,
1564 : HeapTupleHeaderGetUpdateXid(htup));
1565 24034754 : break;
1566 :
1567 0 : default:
1568 :
1569 : /*
1570 : * DEAD tuples should've been passed to heap_prune_record_dead()
1571 : * or heap_prune_record_unused() instead.
1572 : */
1573 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result %d",
1574 : prstate->htsv[offnum]);
1575 : break;
1576 : }
1577 :
1578 : /* Consider freezing any normal tuples which will not be removed */
1579 64164858 : if (prstate->attempt_freeze)
1580 : {
1581 : bool totally_frozen;
1582 :
1583 61190002 : if ((heap_prepare_freeze_tuple(htup,
1584 61190002 : prstate->cutoffs,
1585 : &prstate->pagefrz,
1586 61190002 : &prstate->frozen[prstate->nfrozen],
1587 : &totally_frozen)))
1588 : {
1589 : /* Save prepared freeze plan for later */
1590 4405200 : prstate->frozen[prstate->nfrozen++].offset = offnum;
1591 : }
1592 :
1593 : /*
1594 : * If any tuple isn't either totally frozen already or eligible to
1595 : * become totally frozen (according to its freeze plan), then the page
1596 : * definitely cannot be set all-frozen in the visibility map later on.
1597 : */
1598 61190002 : if (!totally_frozen)
1599 25157736 : prstate->all_frozen = false;
1600 : }
1601 64164858 : }
1602 :
1603 :
1604 : /*
1605 : * Record line pointer that was already LP_DEAD and is left unchanged.
1606 : */
1607 : static void
1608 1896116 : heap_prune_record_unchanged_lp_dead(Page page, PruneState *prstate, OffsetNumber offnum)
1609 : {
1610 : Assert(!prstate->processed[offnum]);
1611 1896116 : prstate->processed[offnum] = true;
1612 :
1613 : /*
1614 : * Deliberately don't set hastup for LP_DEAD items. We make the soft
1615 : * assumption that any LP_DEAD items encountered here will become
1616 : * LP_UNUSED later on, before count_nondeletable_pages is reached. If we
1617 : * don't make this assumption then rel truncation will only happen every
1618 : * other VACUUM, at most. Besides, VACUUM must treat
1619 : * hastup/nonempty_pages as provisional no matter how LP_DEAD items are
1620 : * handled (handled here, or handled later on).
1621 : *
1622 : * Similarly, don't unset all_visible and all_frozen until later, at the
1623 : * end of heap_page_prune_and_freeze(). This will allow us to attempt to
1624 : * freeze the page after pruning. As long as we unset it before updating
1625 : * the visibility map, this will be correct.
1626 : */
1627 :
1628 : /* Record the dead offset for vacuum */
1629 1896116 : prstate->deadoffsets[prstate->lpdead_items++] = offnum;
1630 1896116 : }
1631 :
1632 : /*
1633 : * Record LP_REDIRECT that is left unchanged.
1634 : */
1635 : static void
1636 612122 : heap_prune_record_unchanged_lp_redirect(PruneState *prstate, OffsetNumber offnum)
1637 : {
1638 : /*
1639 : * A redirect line pointer doesn't count as a live tuple.
1640 : *
1641 : * If we leave a redirect line pointer in place, there will be another
1642 : * tuple on the page that it points to. We will do the bookkeeping for
1643 : * that separately. So we have nothing to do here, except remember that
1644 : * we processed this item.
1645 : */
1646 : Assert(!prstate->processed[offnum]);
1647 612122 : prstate->processed[offnum] = true;
1648 612122 : }
1649 :
1650 : /*
1651 : * Perform the actual page changes needed by heap_page_prune_and_freeze().
1652 : *
1653 : * If 'lp_truncate_only' is set, we are merely marking LP_DEAD line pointers
1654 : * as unused, not redirecting or removing anything else. The
1655 : * PageRepairFragmentation() call is skipped in that case.
1656 : *
1657 : * If 'lp_truncate_only' is not set, the caller must hold a cleanup lock on
1658 : * the buffer. If it is set, an ordinary exclusive lock suffices.
1659 : */
1660 : void
1661 123166 : heap_page_prune_execute(Buffer buffer, bool lp_truncate_only,
1662 : OffsetNumber *redirected, int nredirected,
1663 : OffsetNumber *nowdead, int ndead,
1664 : OffsetNumber *nowunused, int nunused)
1665 : {
1666 123166 : Page page = BufferGetPage(buffer);
1667 : OffsetNumber *offnum;
1668 : HeapTupleHeader htup PG_USED_FOR_ASSERTS_ONLY;
1669 :
1670 : /* Shouldn't be called unless there's something to do */
1671 : Assert(nredirected > 0 || ndead > 0 || nunused > 0);
1672 :
1673 : /* If 'lp_truncate_only', we can only remove already-dead line pointers */
1674 : Assert(!lp_truncate_only || (nredirected == 0 && ndead == 0));
1675 :
1676 : /* Update all redirected line pointers */
1677 123166 : offnum = redirected;
1678 290016 : for (int i = 0; i < nredirected; i++)
1679 : {
1680 166850 : OffsetNumber fromoff = *offnum++;
1681 166850 : OffsetNumber tooff = *offnum++;
1682 166850 : ItemId fromlp = PageGetItemId(page, fromoff);
1683 : ItemId tolp PG_USED_FOR_ASSERTS_ONLY;
1684 :
1685 : #ifdef USE_ASSERT_CHECKING
1686 :
1687 : /*
1688 : * Any existing item that we set as an LP_REDIRECT (any 'from' item)
1689 : * must be the first item from a HOT chain. If the item has tuple
1690 : * storage then it can't be a heap-only tuple. Otherwise we are just
1691 : * maintaining an existing LP_REDIRECT from an existing HOT chain that
1692 : * has been pruned at least once before now.
1693 : */
1694 : if (!ItemIdIsRedirected(fromlp))
1695 : {
1696 : Assert(ItemIdHasStorage(fromlp) && ItemIdIsNormal(fromlp));
1697 :
1698 : htup = (HeapTupleHeader) PageGetItem(page, fromlp);
1699 : Assert(!HeapTupleHeaderIsHeapOnly(htup));
1700 : }
1701 : else
1702 : {
1703 : /* We shouldn't need to redundantly set the redirect */
1704 : Assert(ItemIdGetRedirect(fromlp) != tooff);
1705 : }
1706 :
1707 : /*
1708 : * The item that we're about to set as an LP_REDIRECT (the 'from'
1709 : * item) will point to an existing item (the 'to' item) that is
1710 : * already a heap-only tuple. There can be at most one LP_REDIRECT
1711 : * item per HOT chain.
1712 : *
1713 : * We need to keep around an LP_REDIRECT item (after original
1714 : * non-heap-only root tuple gets pruned away) so that it's always
1715 : * possible for VACUUM to easily figure out what TID to delete from
1716 : * indexes when an entire HOT chain becomes dead. A heap-only tuple
1717 : * can never become LP_DEAD; an LP_REDIRECT item or a regular heap
1718 : * tuple can.
1719 : *
1720 : * This check may miss problems, e.g. the target of a redirect could
1721 : * be marked as unused subsequently. The page_verify_redirects() check
1722 : * below will catch such problems.
1723 : */
1724 : tolp = PageGetItemId(page, tooff);
1725 : Assert(ItemIdHasStorage(tolp) && ItemIdIsNormal(tolp));
1726 : htup = (HeapTupleHeader) PageGetItem(page, tolp);
1727 : Assert(HeapTupleHeaderIsHeapOnly(htup));
1728 : #endif
1729 :
1730 166850 : ItemIdSetRedirect(fromlp, tooff);
1731 : }
1732 :
1733 : /* Update all now-dead line pointers */
1734 123166 : offnum = nowdead;
1735 3182976 : for (int i = 0; i < ndead; i++)
1736 : {
1737 3059810 : OffsetNumber off = *offnum++;
1738 3059810 : ItemId lp = PageGetItemId(page, off);
1739 :
1740 : #ifdef USE_ASSERT_CHECKING
1741 :
1742 : /*
1743 : * An LP_DEAD line pointer must be left behind when the original item
1744 : * (which is dead to everybody) could still be referenced by a TID in
1745 : * an index. This should never be necessary with any individual
1746 : * heap-only tuple item, though. (It's not clear how much of a problem
1747 : * that would be, but there is no reason to allow it.)
1748 : */
1749 : if (ItemIdHasStorage(lp))
1750 : {
1751 : Assert(ItemIdIsNormal(lp));
1752 : htup = (HeapTupleHeader) PageGetItem(page, lp);
1753 : Assert(!HeapTupleHeaderIsHeapOnly(htup));
1754 : }
1755 : else
1756 : {
1757 : /* Whole HOT chain becomes dead */
1758 : Assert(ItemIdIsRedirected(lp));
1759 : }
1760 : #endif
1761 :
1762 3059810 : ItemIdSetDead(lp);
1763 : }
1764 :
1765 : /* Update all now-unused line pointers */
1766 123166 : offnum = nowunused;
1767 685482 : for (int i = 0; i < nunused; i++)
1768 : {
1769 562316 : OffsetNumber off = *offnum++;
1770 562316 : ItemId lp = PageGetItemId(page, off);
1771 :
1772 : #ifdef USE_ASSERT_CHECKING
1773 :
1774 : if (lp_truncate_only)
1775 : {
1776 : /* Setting LP_DEAD to LP_UNUSED in vacuum's second pass */
1777 : Assert(ItemIdIsDead(lp) && !ItemIdHasStorage(lp));
1778 : }
1779 : else
1780 : {
1781 : /*
1782 : * When heap_page_prune_and_freeze() was called, mark_unused_now
1783 : * may have been passed as true, which allows would-be LP_DEAD
1784 : * items to be made LP_UNUSED instead. This is only possible if
1785 : * the relation has no indexes. If there are any dead items, then
1786 : * mark_unused_now was not true and every item being marked
1787 : * LP_UNUSED must refer to a heap-only tuple.
1788 : */
1789 : if (ndead > 0)
1790 : {
1791 : Assert(ItemIdHasStorage(lp) && ItemIdIsNormal(lp));
1792 : htup = (HeapTupleHeader) PageGetItem(page, lp);
1793 : Assert(HeapTupleHeaderIsHeapOnly(htup));
1794 : }
1795 : else
1796 : Assert(ItemIdIsUsed(lp));
1797 : }
1798 :
1799 : #endif
1800 :
1801 562316 : ItemIdSetUnused(lp);
1802 : }
1803 :
1804 123166 : if (lp_truncate_only)
1805 3908 : PageTruncateLinePointerArray(page);
1806 : else
1807 : {
1808 : /*
1809 : * Finally, repair any fragmentation, and update the page's hint bit
1810 : * about whether it has free pointers.
1811 : */
1812 119258 : PageRepairFragmentation(page);
1813 :
1814 : /*
1815 : * Now that the page has been modified, assert that redirect items
1816 : * still point to valid targets.
1817 : */
1818 119258 : page_verify_redirects(page);
1819 : }
1820 123166 : }
1821 :
1822 :
1823 : /*
1824 : * If built with assertions, verify that all LP_REDIRECT items point to a
1825 : * valid item.
1826 : *
1827 : * One way that bugs related to HOT pruning show is redirect items pointing to
1828 : * removed tuples. It's not trivial to reliably check that marking an item
1829 : * unused will not orphan a redirect item during heap_prune_chain() /
1830 : * heap_page_prune_execute(), so we additionally check the whole page after
1831 : * pruning. Without this check such bugs would typically only cause asserts
1832 : * later, potentially well after the corruption has been introduced.
1833 : *
1834 : * Also check comments in heap_page_prune_execute()'s redirection loop.
1835 : */
1836 : static void
1837 119258 : page_verify_redirects(Page page)
1838 : {
1839 : #ifdef USE_ASSERT_CHECKING
1840 : OffsetNumber offnum;
1841 : OffsetNumber maxoff;
1842 :
1843 : maxoff = PageGetMaxOffsetNumber(page);
1844 : for (offnum = FirstOffsetNumber;
1845 : offnum <= maxoff;
1846 : offnum = OffsetNumberNext(offnum))
1847 : {
1848 : ItemId itemid = PageGetItemId(page, offnum);
1849 : OffsetNumber targoff;
1850 : ItemId targitem;
1851 : HeapTupleHeader htup;
1852 :
1853 : if (!ItemIdIsRedirected(itemid))
1854 : continue;
1855 :
1856 : targoff = ItemIdGetRedirect(itemid);
1857 : targitem = PageGetItemId(page, targoff);
1858 :
1859 : Assert(ItemIdIsUsed(targitem));
1860 : Assert(ItemIdIsNormal(targitem));
1861 : Assert(ItemIdHasStorage(targitem));
1862 : htup = (HeapTupleHeader) PageGetItem(page, targitem);
1863 : Assert(HeapTupleHeaderIsHeapOnly(htup));
1864 : }
1865 : #endif
1866 119258 : }
1867 :
1868 :
1869 : /*
1870 : * For all items in this page, find their respective root line pointers.
1871 : * If item k is part of a HOT-chain with root at item j, then we set
1872 : * root_offsets[k - 1] = j.
1873 : *
1874 : * The passed-in root_offsets array must have MaxHeapTuplesPerPage entries.
1875 : * Unused entries are filled with InvalidOffsetNumber (zero).
1876 : *
1877 : * The function must be called with at least share lock on the buffer, to
1878 : * prevent concurrent prune operations.
1879 : *
1880 : * Note: The information collected here is valid only as long as the caller
1881 : * holds a pin on the buffer. Once pin is released, a tuple might be pruned
1882 : * and reused by a completely unrelated tuple.
1883 : */
1884 : void
1885 224390 : heap_get_root_tuples(Page page, OffsetNumber *root_offsets)
1886 : {
1887 : OffsetNumber offnum,
1888 : maxoff;
1889 :
1890 224390 : MemSet(root_offsets, InvalidOffsetNumber,
1891 : MaxHeapTuplesPerPage * sizeof(OffsetNumber));
1892 :
1893 224390 : maxoff = PageGetMaxOffsetNumber(page);
1894 18078406 : for (offnum = FirstOffsetNumber; offnum <= maxoff; offnum = OffsetNumberNext(offnum))
1895 : {
1896 17854016 : ItemId lp = PageGetItemId(page, offnum);
1897 : HeapTupleHeader htup;
1898 : OffsetNumber nextoffnum;
1899 : TransactionId priorXmax;
1900 :
1901 : /* skip unused and dead items */
1902 17854016 : if (!ItemIdIsUsed(lp) || ItemIdIsDead(lp))
1903 22102 : continue;
1904 :
1905 17831914 : if (ItemIdIsNormal(lp))
1906 : {
1907 17823388 : htup = (HeapTupleHeader) PageGetItem(page, lp);
1908 :
1909 : /*
1910 : * Check if this tuple is part of a HOT-chain rooted at some other
1911 : * tuple. If so, skip it for now; we'll process it when we find
1912 : * its root.
1913 : */
1914 17823388 : if (HeapTupleHeaderIsHeapOnly(htup))
1915 9018 : continue;
1916 :
1917 : /*
1918 : * This is either a plain tuple or the root of a HOT-chain.
1919 : * Remember it in the mapping.
1920 : */
1921 17814370 : root_offsets[offnum - 1] = offnum;
1922 :
1923 : /* If it's not the start of a HOT-chain, we're done with it */
1924 17814370 : if (!HeapTupleHeaderIsHotUpdated(htup))
1925 17813976 : continue;
1926 :
1927 : /* Set up to scan the HOT-chain */
1928 394 : nextoffnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
1929 394 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
1930 : }
1931 : else
1932 : {
1933 : /* Must be a redirect item. We do not set its root_offsets entry */
1934 : Assert(ItemIdIsRedirected(lp));
1935 : /* Set up to scan the HOT-chain */
1936 8526 : nextoffnum = ItemIdGetRedirect(lp);
1937 8526 : priorXmax = InvalidTransactionId;
1938 : }
1939 :
1940 : /*
1941 : * Now follow the HOT-chain and collect other tuples in the chain.
1942 : *
1943 : * Note: Even though this is a nested loop, the complexity of the
1944 : * function is O(N) because a tuple in the page should be visited not
1945 : * more than twice, once in the outer loop and once in HOT-chain
1946 : * chases.
1947 : */
1948 : for (;;)
1949 : {
1950 : /* Sanity check (pure paranoia) */
1951 9012 : if (offnum < FirstOffsetNumber)
1952 0 : break;
1953 :
1954 : /*
1955 : * An offset past the end of page's line pointer array is possible
1956 : * when the array was truncated
1957 : */
1958 9012 : if (offnum > maxoff)
1959 0 : break;
1960 :
1961 9012 : lp = PageGetItemId(page, nextoffnum);
1962 :
1963 : /* Check for broken chains */
1964 9012 : if (!ItemIdIsNormal(lp))
1965 0 : break;
1966 :
1967 9012 : htup = (HeapTupleHeader) PageGetItem(page, lp);
1968 :
1969 9498 : if (TransactionIdIsValid(priorXmax) &&
1970 486 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(htup)))
1971 0 : break;
1972 :
1973 : /* Remember the root line pointer for this item */
1974 9012 : root_offsets[nextoffnum - 1] = offnum;
1975 :
1976 : /* Advance to next chain member, if any */
1977 9012 : if (!HeapTupleHeaderIsHotUpdated(htup))
1978 8920 : break;
1979 :
1980 : /* HOT implies it can't have moved to different partition */
1981 : Assert(!HeapTupleHeaderIndicatesMovedPartitions(htup));
1982 :
1983 92 : nextoffnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
1984 92 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
1985 : }
1986 : }
1987 224390 : }
1988 :
1989 :
1990 : /*
1991 : * Compare fields that describe actions required to freeze tuple with caller's
1992 : * open plan. If everything matches then the frz tuple plan is equivalent to
1993 : * caller's plan.
1994 : */
1995 : static inline bool
1996 1901208 : heap_log_freeze_eq(xlhp_freeze_plan *plan, HeapTupleFreeze *frz)
1997 : {
1998 1901208 : if (plan->xmax == frz->xmax &&
1999 1898582 : plan->t_infomask2 == frz->t_infomask2 &&
2000 1896986 : plan->t_infomask == frz->t_infomask &&
2001 1892006 : plan->frzflags == frz->frzflags)
2002 1892006 : return true;
2003 :
2004 : /* Caller must call heap_log_freeze_new_plan again for frz */
2005 9202 : return false;
2006 : }
2007 :
2008 : /*
2009 : * Comparator used to deduplicate the freeze plans used in WAL records.
2010 : */
2011 : static int
2012 2634380 : heap_log_freeze_cmp(const void *arg1, const void *arg2)
2013 : {
2014 2634380 : HeapTupleFreeze *frz1 = (HeapTupleFreeze *) arg1;
2015 2634380 : HeapTupleFreeze *frz2 = (HeapTupleFreeze *) arg2;
2016 :
2017 2634380 : if (frz1->xmax < frz2->xmax)
2018 26452 : return -1;
2019 2607928 : else if (frz1->xmax > frz2->xmax)
2020 28600 : return 1;
2021 :
2022 2579328 : if (frz1->t_infomask2 < frz2->t_infomask2)
2023 8826 : return -1;
2024 2570502 : else if (frz1->t_infomask2 > frz2->t_infomask2)
2025 8602 : return 1;
2026 :
2027 2561900 : if (frz1->t_infomask < frz2->t_infomask)
2028 22248 : return -1;
2029 2539652 : else if (frz1->t_infomask > frz2->t_infomask)
2030 30592 : return 1;
2031 :
2032 2509060 : if (frz1->frzflags < frz2->frzflags)
2033 0 : return -1;
2034 2509060 : else if (frz1->frzflags > frz2->frzflags)
2035 0 : return 1;
2036 :
2037 : /*
2038 : * heap_log_freeze_eq would consider these tuple-wise plans to be equal.
2039 : * (So the tuples will share a single canonical freeze plan.)
2040 : *
2041 : * We tiebreak on page offset number to keep each freeze plan's page
2042 : * offset number array individually sorted. (Unnecessary, but be tidy.)
2043 : */
2044 2509060 : if (frz1->offset < frz2->offset)
2045 2117876 : return -1;
2046 391184 : else if (frz1->offset > frz2->offset)
2047 391184 : return 1;
2048 :
2049 : Assert(false);
2050 0 : return 0;
2051 : }
2052 :
2053 : /*
2054 : * Start new plan initialized using tuple-level actions. At least one tuple
2055 : * will have steps required to freeze described by caller's plan during REDO.
2056 : */
2057 : static inline void
2058 55238 : heap_log_freeze_new_plan(xlhp_freeze_plan *plan, HeapTupleFreeze *frz)
2059 : {
2060 55238 : plan->xmax = frz->xmax;
2061 55238 : plan->t_infomask2 = frz->t_infomask2;
2062 55238 : plan->t_infomask = frz->t_infomask;
2063 55238 : plan->frzflags = frz->frzflags;
2064 55238 : plan->ntuples = 1; /* for now */
2065 55238 : }
2066 :
2067 : /*
2068 : * Deduplicate tuple-based freeze plans so that each distinct set of
2069 : * processing steps is only stored once in the WAL record.
2070 : * Called during original execution of freezing (for logged relations).
2071 : *
2072 : * Return value is number of plans set in *plans_out for caller. Also writes
2073 : * an array of offset numbers into *offsets_out output argument for caller
2074 : * (actually there is one array per freeze plan, but that's not of immediate
2075 : * concern to our caller).
2076 : */
2077 : static int
2078 46036 : heap_log_freeze_plan(HeapTupleFreeze *tuples, int ntuples,
2079 : xlhp_freeze_plan *plans_out,
2080 : OffsetNumber *offsets_out)
2081 : {
2082 46036 : int nplans = 0;
2083 :
2084 : /* Sort tuple-based freeze plans in the order required to deduplicate */
2085 46036 : qsort(tuples, ntuples, sizeof(HeapTupleFreeze), heap_log_freeze_cmp);
2086 :
2087 1993280 : for (int i = 0; i < ntuples; i++)
2088 : {
2089 1947244 : HeapTupleFreeze *frz = tuples + i;
2090 :
2091 1947244 : if (i == 0)
2092 : {
2093 : /* New canonical freeze plan starting with first tup */
2094 46036 : heap_log_freeze_new_plan(plans_out, frz);
2095 46036 : nplans++;
2096 : }
2097 1901208 : else if (heap_log_freeze_eq(plans_out, frz))
2098 : {
2099 : /* tup matches open canonical plan -- include tup in it */
2100 : Assert(offsets_out[i - 1] < frz->offset);
2101 1892006 : plans_out->ntuples++;
2102 : }
2103 : else
2104 : {
2105 : /* Tup doesn't match current plan -- done with it now */
2106 9202 : plans_out++;
2107 :
2108 : /* New canonical freeze plan starting with this tup */
2109 9202 : heap_log_freeze_new_plan(plans_out, frz);
2110 9202 : nplans++;
2111 : }
2112 :
2113 : /*
2114 : * Save page offset number in dedicated buffer in passing.
2115 : *
2116 : * REDO routine relies on the record's offset numbers array grouping
2117 : * offset numbers by freeze plan. The sort order within each grouping
2118 : * is ascending offset number order, just to keep things tidy.
2119 : */
2120 1947244 : offsets_out[i] = frz->offset;
2121 : }
2122 :
2123 : Assert(nplans > 0 && nplans <= ntuples);
2124 :
2125 46036 : return nplans;
2126 : }
2127 :
2128 : /*
2129 : * Write an XLOG_HEAP2_PRUNE* WAL record
2130 : *
2131 : * This is used for several different page maintenance operations:
2132 : *
2133 : * - Page pruning, in VACUUM's 1st pass or on access: Some items are
2134 : * redirected, some marked dead, and some removed altogether.
2135 : *
2136 : * - Freezing: Items are marked as 'frozen'.
2137 : *
2138 : * - Vacuum, 2nd pass: Items that are already LP_DEAD are marked as unused.
2139 : *
2140 : * They have enough commonalities that we use a single WAL record for them
2141 : * all.
2142 : *
2143 : * If replaying the record requires a cleanup lock, pass cleanup_lock = true.
2144 : * Replaying 'redirected' or 'dead' items always requires a cleanup lock, but
2145 : * replaying 'unused' items depends on whether they were all previously marked
2146 : * as dead.
2147 : *
2148 : * If the VM is being updated, vmflags will contain the bits to set. In this
2149 : * case, vmbuffer should already have been updated and marked dirty and should
2150 : * still be pinned and locked.
2151 : *
2152 : * Note: This function scribbles on the 'frozen' array.
2153 : *
2154 : * Note: This is called in a critical section, so careful what you do here.
2155 : */
2156 : void
2157 169710 : log_heap_prune_and_freeze(Relation relation, Buffer buffer,
2158 : Buffer vmbuffer, uint8 vmflags,
2159 : TransactionId conflict_xid,
2160 : bool cleanup_lock,
2161 : PruneReason reason,
2162 : HeapTupleFreeze *frozen, int nfrozen,
2163 : OffsetNumber *redirected, int nredirected,
2164 : OffsetNumber *dead, int ndead,
2165 : OffsetNumber *unused, int nunused)
2166 : {
2167 : xl_heap_prune xlrec;
2168 : XLogRecPtr recptr;
2169 : uint8 info;
2170 : uint8 regbuf_flags_heap;
2171 :
2172 : /* The following local variables hold data registered in the WAL record: */
2173 : xlhp_freeze_plan plans[MaxHeapTuplesPerPage];
2174 : xlhp_freeze_plans freeze_plans;
2175 : xlhp_prune_items redirect_items;
2176 : xlhp_prune_items dead_items;
2177 : xlhp_prune_items unused_items;
2178 : OffsetNumber frz_offsets[MaxHeapTuplesPerPage];
2179 169710 : bool do_prune = nredirected > 0 || ndead > 0 || nunused > 0;
2180 169710 : bool do_set_vm = vmflags & VISIBILITYMAP_VALID_BITS;
2181 :
2182 : Assert((vmflags & VISIBILITYMAP_VALID_BITS) == vmflags);
2183 :
2184 169710 : xlrec.flags = 0;
2185 169710 : regbuf_flags_heap = REGBUF_STANDARD;
2186 :
2187 : /*
2188 : * We can avoid an FPI of the heap page if the only modification we are
2189 : * making to it is to set PD_ALL_VISIBLE and checksums/wal_log_hints are
2190 : * disabled. Note that if we explicitly skip an FPI, we must not stamp the
2191 : * heap page with this record's LSN. Recovery skips records <= the stamped
2192 : * LSN, so this could lead to skipping an earlier FPI needed to repair a
2193 : * torn page.
2194 : */
2195 169710 : if (!do_prune &&
2196 0 : nfrozen == 0 &&
2197 0 : (!do_set_vm || !XLogHintBitIsNeeded()))
2198 0 : regbuf_flags_heap |= REGBUF_NO_IMAGE;
2199 :
2200 : /*
2201 : * Prepare data for the buffer. The arrays are not actually in the
2202 : * buffer, but we pretend that they are. When XLogInsert stores a full
2203 : * page image, the arrays can be omitted.
2204 : */
2205 169710 : XLogBeginInsert();
2206 169710 : XLogRegisterBuffer(0, buffer, regbuf_flags_heap);
2207 :
2208 169710 : if (do_set_vm)
2209 24376 : XLogRegisterBuffer(1, vmbuffer, 0);
2210 :
2211 169710 : if (nfrozen > 0)
2212 : {
2213 : int nplans;
2214 :
2215 46036 : xlrec.flags |= XLHP_HAS_FREEZE_PLANS;
2216 :
2217 : /*
2218 : * Prepare deduplicated representation for use in the WAL record. This
2219 : * destructively sorts frozen tuples array in-place.
2220 : */
2221 46036 : nplans = heap_log_freeze_plan(frozen, nfrozen, plans, frz_offsets);
2222 :
2223 46036 : freeze_plans.nplans = nplans;
2224 46036 : XLogRegisterBufData(0, &freeze_plans,
2225 : offsetof(xlhp_freeze_plans, plans));
2226 46036 : XLogRegisterBufData(0, plans,
2227 : sizeof(xlhp_freeze_plan) * nplans);
2228 : }
2229 169710 : if (nredirected > 0)
2230 : {
2231 33262 : xlrec.flags |= XLHP_HAS_REDIRECTIONS;
2232 :
2233 33262 : redirect_items.ntargets = nredirected;
2234 33262 : XLogRegisterBufData(0, &redirect_items,
2235 : offsetof(xlhp_prune_items, data));
2236 33262 : XLogRegisterBufData(0, redirected,
2237 : sizeof(OffsetNumber[2]) * nredirected);
2238 : }
2239 169710 : if (ndead > 0)
2240 : {
2241 76838 : xlrec.flags |= XLHP_HAS_DEAD_ITEMS;
2242 :
2243 76838 : dead_items.ntargets = ndead;
2244 76838 : XLogRegisterBufData(0, &dead_items,
2245 : offsetof(xlhp_prune_items, data));
2246 76838 : XLogRegisterBufData(0, dead,
2247 : sizeof(OffsetNumber) * ndead);
2248 : }
2249 169710 : if (nunused > 0)
2250 : {
2251 46658 : xlrec.flags |= XLHP_HAS_NOW_UNUSED_ITEMS;
2252 :
2253 46658 : unused_items.ntargets = nunused;
2254 46658 : XLogRegisterBufData(0, &unused_items,
2255 : offsetof(xlhp_prune_items, data));
2256 46658 : XLogRegisterBufData(0, unused,
2257 : sizeof(OffsetNumber) * nunused);
2258 : }
2259 169710 : if (nfrozen > 0)
2260 46036 : XLogRegisterBufData(0, frz_offsets,
2261 : sizeof(OffsetNumber) * nfrozen);
2262 :
2263 : /*
2264 : * Prepare the main xl_heap_prune record. We already set the XLHP_HAS_*
2265 : * flag above.
2266 : */
2267 169710 : if (vmflags & VISIBILITYMAP_ALL_VISIBLE)
2268 : {
2269 24376 : xlrec.flags |= XLHP_VM_ALL_VISIBLE;
2270 24376 : if (vmflags & VISIBILITYMAP_ALL_FROZEN)
2271 19452 : xlrec.flags |= XLHP_VM_ALL_FROZEN;
2272 : }
2273 169710 : if (RelationIsAccessibleInLogicalDecoding(relation))
2274 1288 : xlrec.flags |= XLHP_IS_CATALOG_REL;
2275 169710 : if (TransactionIdIsValid(conflict_xid))
2276 140302 : xlrec.flags |= XLHP_HAS_CONFLICT_HORIZON;
2277 169710 : if (cleanup_lock)
2278 145080 : xlrec.flags |= XLHP_CLEANUP_LOCK;
2279 : else
2280 : {
2281 : Assert(nredirected == 0 && ndead == 0);
2282 : /* also, any items in 'unused' must've been LP_DEAD previously */
2283 : }
2284 169710 : XLogRegisterData(&xlrec, SizeOfHeapPrune);
2285 169710 : if (TransactionIdIsValid(conflict_xid))
2286 140302 : XLogRegisterData(&conflict_xid, sizeof(TransactionId));
2287 :
2288 169710 : switch (reason)
2289 : {
2290 82492 : case PRUNE_ON_ACCESS:
2291 82492 : info = XLOG_HEAP2_PRUNE_ON_ACCESS;
2292 82492 : break;
2293 62588 : case PRUNE_VACUUM_SCAN:
2294 62588 : info = XLOG_HEAP2_PRUNE_VACUUM_SCAN;
2295 62588 : break;
2296 24630 : case PRUNE_VACUUM_CLEANUP:
2297 24630 : info = XLOG_HEAP2_PRUNE_VACUUM_CLEANUP;
2298 24630 : break;
2299 0 : default:
2300 0 : elog(ERROR, "unrecognized prune reason: %d", (int) reason);
2301 : break;
2302 : }
2303 169710 : recptr = XLogInsert(RM_HEAP2_ID, info);
2304 :
2305 169710 : if (do_set_vm)
2306 : {
2307 : Assert(BufferIsDirty(vmbuffer));
2308 24376 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2309 : }
2310 :
2311 : /*
2312 : * See comment at the top of the function about regbuf_flags_heap for
2313 : * details on when we can advance the page LSN.
2314 : */
2315 169710 : if (do_prune || nfrozen > 0 || (do_set_vm && XLogHintBitIsNeeded()))
2316 : {
2317 : Assert(BufferIsDirty(buffer));
2318 169710 : PageSetLSN(BufferGetPage(buffer), recptr);
2319 : }
2320 169710 : }
|
