Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * vacuumlazy.c
4 : * Concurrent ("lazy") vacuuming.
5 : *
6 : * Heap relations are vacuumed in three main phases. In phase I, vacuum scans
7 : * relation pages, pruning and freezing tuples and saving dead tuples' TIDs in
8 : * a TID store. If that TID store fills up or vacuum finishes scanning the
9 : * relation, it progresses to phase II: index vacuuming. Index vacuuming
10 : * deletes the dead index entries referenced in the TID store. In phase III,
11 : * vacuum scans the blocks of the relation referred to by the TIDs in the TID
12 : * store and reaps the corresponding dead items, freeing that space for future
13 : * tuples.
14 : *
15 : * If there are no indexes or index scanning is disabled, phase II may be
16 : * skipped. If phase I identified very few dead index entries or if vacuum's
17 : * failsafe mechanism has triggered (to avoid transaction ID wraparound),
18 : * vacuum may skip phases II and III.
19 : *
20 : * If the TID store fills up in phase I, vacuum suspends phase I and proceeds
21 : * to phases II and III, cleaning up the dead tuples referenced in the current
22 : * TID store. This empties the TID store, allowing vacuum to resume phase I.
23 : *
24 : * In a way, the phases are more like states in a state machine, but they have
25 : * been referred to colloquially as phases for so long that they are referred
26 : * to as such here.
27 : *
28 : * Manually invoked VACUUMs may scan indexes during phase II in parallel. For
29 : * more information on this, see the comment at the top of vacuumparallel.c.
30 : *
31 : * In between phases, vacuum updates the freespace map (every
32 : * VACUUM_FSM_EVERY_PAGES).
33 : *
34 : * After completing all three phases, vacuum may truncate the relation if it
35 : * has emptied pages at the end. Finally, vacuum updates relation statistics
36 : * in pg_class and the cumulative statistics subsystem.
37 : *
38 : * Relation Scanning:
39 : *
40 : * Vacuum scans the heap relation, starting at the beginning and progressing
41 : * to the end, skipping pages as permitted by their visibility status, vacuum
42 : * options, and various other requirements.
43 : *
44 : * Vacuums are either aggressive or normal. Aggressive vacuums must scan every
45 : * unfrozen tuple in order to advance relfrozenxid and avoid transaction ID
46 : * wraparound. Normal vacuums may scan otherwise skippable pages for one of
47 : * two reasons:
48 : *
49 : * When page skipping is not disabled, a normal vacuum may scan pages that are
50 : * marked all-visible (and even all-frozen) in the visibility map if the range
51 : * of skippable pages is below SKIP_PAGES_THRESHOLD. This is primarily for the
52 : * benefit of kernel readahead (see comment in heap_vac_scan_next_block()).
53 : *
54 : * A normal vacuum may also scan skippable pages in an effort to freeze them
55 : * and decrease the backlog of all-visible but not all-frozen pages that have
56 : * to be processed by the next aggressive vacuum. These are referred to as
57 : * eagerly scanned pages. Pages scanned due to SKIP_PAGES_THRESHOLD do not
58 : * count as eagerly scanned pages.
59 : *
60 : * Eagerly scanned pages that are set all-frozen in the VM are successful
61 : * eager freezes and those not set all-frozen in the VM are failed eager
62 : * freezes.
63 : *
64 : * Because we want to amortize the overhead of freezing pages over multiple
65 : * vacuums, normal vacuums cap the number of successful eager freezes to
66 : * MAX_EAGER_FREEZE_SUCCESS_RATE of the number of all-visible but not
67 : * all-frozen pages at the beginning of the vacuum. Since eagerly frozen pages
68 : * may be unfrozen before the next aggressive vacuum, capping the number of
69 : * successful eager freezes also caps the downside of eager freezing:
70 : * potentially wasted work.
71 : *
72 : * Once the success cap has been hit, eager scanning is disabled for the
73 : * remainder of the vacuum of the relation.
74 : *
75 : * Success is capped globally because we don't want to limit our successes if
76 : * old data happens to be concentrated in a particular part of the table. This
77 : * is especially likely to happen for append-mostly workloads where the oldest
78 : * data is at the beginning of the unfrozen portion of the relation.
79 : *
80 : * On the assumption that different regions of the table are likely to contain
81 : * similarly aged data, normal vacuums use a localized eager freeze failure
82 : * cap. The failure count is reset for each region of the table -- comprised
83 : * of EAGER_SCAN_REGION_SIZE blocks. In each region, we tolerate
84 : * vacuum_max_eager_freeze_failure_rate of EAGER_SCAN_REGION_SIZE failures
85 : * before suspending eager scanning until the end of the region.
86 : * vacuum_max_eager_freeze_failure_rate is configurable both globally and per
87 : * table.
88 : *
89 : * Aggressive vacuums must examine every unfrozen tuple and thus are not
90 : * subject to any of the limits imposed by the eager scanning algorithm.
91 : *
92 : * Once vacuum has decided to scan a given block, it must read the block and
93 : * obtain a cleanup lock to prune tuples on the page. A non-aggressive vacuum
94 : * may choose to skip pruning and freezing if it cannot acquire a cleanup lock
95 : * on the buffer right away. In this case, it may miss cleaning up dead tuples
96 : * and their associated index entries (though it is free to reap any existing
97 : * dead items on the page).
98 : *
99 : * After pruning and freezing, pages that are newly all-visible and all-frozen
100 : * are marked as such in the visibility map.
101 : *
102 : * Dead TID Storage:
103 : *
104 : * The major space usage for vacuuming is storage for the dead tuple IDs that
105 : * are to be removed from indexes. We want to ensure we can vacuum even the
106 : * very largest relations with finite memory space usage. To do that, we set
107 : * upper bounds on the memory that can be used for keeping track of dead TIDs
108 : * at once.
109 : *
110 : * We are willing to use at most maintenance_work_mem (or perhaps
111 : * autovacuum_work_mem) memory space to keep track of dead TIDs. If the
112 : * TID store is full, we must call lazy_vacuum to vacuum indexes (and to vacuum
113 : * the pages that we've pruned). This frees up the memory space dedicated to
114 : * store dead TIDs.
115 : *
116 : * In practice VACUUM will often complete its initial pass over the target
117 : * heap relation without ever running out of space to store TIDs. This means
118 : * that there only needs to be one call to lazy_vacuum, after the initial pass
119 : * completes.
120 : *
121 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
122 : * Portions Copyright (c) 1994, Regents of the University of California
123 : *
124 : *
125 : * IDENTIFICATION
126 : * src/backend/access/heap/vacuumlazy.c
127 : *
128 : *-------------------------------------------------------------------------
129 : */
130 : #include "postgres.h"
131 :
132 : #include <math.h>
133 :
134 : #include "access/genam.h"
135 : #include "access/heapam.h"
136 : #include "access/htup_details.h"
137 : #include "access/multixact.h"
138 : #include "access/tidstore.h"
139 : #include "access/transam.h"
140 : #include "access/visibilitymap.h"
141 : #include "access/xloginsert.h"
142 : #include "catalog/storage.h"
143 : #include "commands/progress.h"
144 : #include "commands/vacuum.h"
145 : #include "common/int.h"
146 : #include "common/pg_prng.h"
147 : #include "executor/instrument.h"
148 : #include "miscadmin.h"
149 : #include "pgstat.h"
150 : #include "portability/instr_time.h"
151 : #include "postmaster/autovacuum.h"
152 : #include "storage/bufmgr.h"
153 : #include "storage/freespace.h"
154 : #include "storage/lmgr.h"
155 : #include "storage/read_stream.h"
156 : #include "utils/lsyscache.h"
157 : #include "utils/pg_rusage.h"
158 : #include "utils/timestamp.h"
159 :
160 :
161 : /*
162 : * Space/time tradeoff parameters: do these need to be user-tunable?
163 : *
164 : * To consider truncating the relation, we want there to be at least
165 : * REL_TRUNCATE_MINIMUM or (relsize / REL_TRUNCATE_FRACTION) (whichever
166 : * is less) potentially-freeable pages.
167 : */
168 : #define REL_TRUNCATE_MINIMUM 1000
169 : #define REL_TRUNCATE_FRACTION 16
170 :
171 : /*
172 : * Timing parameters for truncate locking heuristics.
173 : *
174 : * These were not exposed as user tunable GUC values because it didn't seem
175 : * that the potential for improvement was great enough to merit the cost of
176 : * supporting them.
177 : */
178 : #define VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL 20 /* ms */
179 : #define VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL 50 /* ms */
180 : #define VACUUM_TRUNCATE_LOCK_TIMEOUT 5000 /* ms */
181 :
182 : /*
183 : * Threshold that controls whether we bypass index vacuuming and heap
184 : * vacuuming as an optimization
185 : */
186 : #define BYPASS_THRESHOLD_PAGES 0.02 /* i.e. 2% of rel_pages */
187 :
188 : /*
189 : * Perform a failsafe check each time we scan another 4GB of pages.
190 : * (Note that this is deliberately kept to a power-of-two, usually 2^19.)
191 : */
192 : #define FAILSAFE_EVERY_PAGES \
193 : ((BlockNumber) (((uint64) 4 * 1024 * 1024 * 1024) / BLCKSZ))
194 :
195 : /*
196 : * When a table has no indexes, vacuum the FSM after every 8GB, approximately
197 : * (it won't be exact because we only vacuum FSM after processing a heap page
198 : * that has some removable tuples). When there are indexes, this is ignored,
199 : * and we vacuum FSM after each index/heap cleaning pass.
200 : */
201 : #define VACUUM_FSM_EVERY_PAGES \
202 : ((BlockNumber) (((uint64) 8 * 1024 * 1024 * 1024) / BLCKSZ))
203 :
204 : /*
205 : * Before we consider skipping a page that's marked as clean in
206 : * visibility map, we must've seen at least this many clean pages.
207 : */
208 : #define SKIP_PAGES_THRESHOLD ((BlockNumber) 32)
209 :
210 : /*
211 : * Size of the prefetch window for lazy vacuum backwards truncation scan.
212 : * Needs to be a power of 2.
213 : */
214 : #define PREFETCH_SIZE ((BlockNumber) 32)
215 :
216 : /*
217 : * Macro to check if we are in a parallel vacuum. If true, we are in the
218 : * parallel mode and the DSM segment is initialized.
219 : */
220 : #define ParallelVacuumIsActive(vacrel) ((vacrel)->pvs != NULL)
221 :
222 : /* Phases of vacuum during which we report error context. */
223 : typedef enum
224 : {
225 : VACUUM_ERRCB_PHASE_UNKNOWN,
226 : VACUUM_ERRCB_PHASE_SCAN_HEAP,
227 : VACUUM_ERRCB_PHASE_VACUUM_INDEX,
228 : VACUUM_ERRCB_PHASE_VACUUM_HEAP,
229 : VACUUM_ERRCB_PHASE_INDEX_CLEANUP,
230 : VACUUM_ERRCB_PHASE_TRUNCATE,
231 : } VacErrPhase;
232 :
233 : /*
234 : * An eager scan of a page that is set all-frozen in the VM is considered
235 : * "successful". To spread out freezing overhead across multiple normal
236 : * vacuums, we limit the number of successful eager page freezes. The maximum
237 : * number of eager page freezes is calculated as a ratio of the all-visible
238 : * but not all-frozen pages at the beginning of the vacuum.
239 : */
240 : #define MAX_EAGER_FREEZE_SUCCESS_RATE 0.2
241 :
242 : /*
243 : * On the assumption that different regions of the table tend to have
244 : * similarly aged data, once vacuum fails to freeze
245 : * vacuum_max_eager_freeze_failure_rate of the blocks in a region of size
246 : * EAGER_SCAN_REGION_SIZE, it suspends eager scanning until it has progressed
247 : * to another region of the table with potentially older data.
248 : */
249 : #define EAGER_SCAN_REGION_SIZE 4096
250 :
251 : /*
252 : * heap_vac_scan_next_block() sets these flags to communicate information
253 : * about the block it read to the caller.
254 : */
255 : #define VAC_BLK_WAS_EAGER_SCANNED (1 << 0)
256 : #define VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM (1 << 1)
257 :
258 : typedef struct LVRelState
259 : {
260 : /* Target heap relation and its indexes */
261 : Relation rel;
262 : Relation *indrels;
263 : int nindexes;
264 :
265 : /* Buffer access strategy and parallel vacuum state */
266 : BufferAccessStrategy bstrategy;
267 : ParallelVacuumState *pvs;
268 :
269 : /* Aggressive VACUUM? (must set relfrozenxid >= FreezeLimit) */
270 : bool aggressive;
271 : /* Use visibility map to skip? (disabled by DISABLE_PAGE_SKIPPING) */
272 : bool skipwithvm;
273 : /* Consider index vacuuming bypass optimization? */
274 : bool consider_bypass_optimization;
275 :
276 : /* Doing index vacuuming, index cleanup, rel truncation? */
277 : bool do_index_vacuuming;
278 : bool do_index_cleanup;
279 : bool do_rel_truncate;
280 :
281 : /* VACUUM operation's cutoffs for freezing and pruning */
282 : struct VacuumCutoffs cutoffs;
283 : GlobalVisState *vistest;
284 : /* Tracks oldest extant XID/MXID for setting relfrozenxid/relminmxid */
285 : TransactionId NewRelfrozenXid;
286 : MultiXactId NewRelminMxid;
287 : bool skippedallvis;
288 :
289 : /* Error reporting state */
290 : char *dbname;
291 : char *relnamespace;
292 : char *relname;
293 : char *indname; /* Current index name */
294 : BlockNumber blkno; /* used only for heap operations */
295 : OffsetNumber offnum; /* used only for heap operations */
296 : VacErrPhase phase;
297 : bool verbose; /* VACUUM VERBOSE? */
298 :
299 : /*
300 : * dead_items stores TIDs whose index tuples are deleted by index
301 : * vacuuming. Each TID points to an LP_DEAD line pointer from a heap page
302 : * that has been processed by lazy_scan_prune. Also needed by
303 : * lazy_vacuum_heap_rel, which marks the same LP_DEAD line pointers as
304 : * LP_UNUSED during second heap pass.
305 : *
306 : * Both dead_items and dead_items_info are allocated in shared memory in
307 : * parallel vacuum cases.
308 : */
309 : TidStore *dead_items; /* TIDs whose index tuples we'll delete */
310 : VacDeadItemsInfo *dead_items_info;
311 :
312 : BlockNumber rel_pages; /* total number of pages */
313 : BlockNumber scanned_pages; /* # pages examined (not skipped via VM) */
314 :
315 : /*
316 : * Count of all-visible blocks eagerly scanned (for logging only). This
317 : * does not include skippable blocks scanned due to SKIP_PAGES_THRESHOLD.
318 : */
319 : BlockNumber eager_scanned_pages;
320 :
321 : BlockNumber removed_pages; /* # pages removed by relation truncation */
322 : BlockNumber new_frozen_tuple_pages; /* # pages with newly frozen tuples */
323 :
324 : /* # pages newly set all-visible in the VM */
325 : BlockNumber vm_new_visible_pages;
326 :
327 : /*
328 : * # pages newly set all-visible and all-frozen in the VM. This is a
329 : * subset of vm_new_visible_pages. That is, vm_new_visible_pages includes
330 : * all pages set all-visible, but vm_new_visible_frozen_pages includes
331 : * only those which were also set all-frozen.
332 : */
333 : BlockNumber vm_new_visible_frozen_pages;
334 :
335 : /* # all-visible pages newly set all-frozen in the VM */
336 : BlockNumber vm_new_frozen_pages;
337 :
338 : BlockNumber lpdead_item_pages; /* # pages with LP_DEAD items */
339 : BlockNumber missed_dead_pages; /* # pages with missed dead tuples */
340 : BlockNumber nonempty_pages; /* actually, last nonempty page + 1 */
341 :
342 : /* Statistics output by us, for table */
343 : double new_rel_tuples; /* new estimated total # of tuples */
344 : double new_live_tuples; /* new estimated total # of live tuples */
345 : /* Statistics output by index AMs */
346 : IndexBulkDeleteResult **indstats;
347 :
348 : /* Instrumentation counters */
349 : int num_index_scans;
350 : /* Counters that follow are only for scanned_pages */
351 : int64 tuples_deleted; /* # deleted from table */
352 : int64 tuples_frozen; /* # newly frozen */
353 : int64 lpdead_items; /* # deleted from indexes */
354 : int64 live_tuples; /* # live tuples remaining */
355 : int64 recently_dead_tuples; /* # dead, but not yet removable */
356 : int64 missed_dead_tuples; /* # removable, but not removed */
357 :
358 : /* State maintained by heap_vac_scan_next_block() */
359 : BlockNumber current_block; /* last block returned */
360 : BlockNumber next_unskippable_block; /* next unskippable block */
361 : bool next_unskippable_allvis; /* its visibility status */
362 : bool next_unskippable_eager_scanned; /* if it was eagerly scanned */
363 : Buffer next_unskippable_vmbuffer; /* buffer containing its VM bit */
364 :
365 : /* State related to managing eager scanning of all-visible pages */
366 :
367 : /*
368 : * A normal vacuum that has failed to freeze too many eagerly scanned
369 : * blocks in a region suspends eager scanning.
370 : * next_eager_scan_region_start is the block number of the first block
371 : * eligible for resumed eager scanning.
372 : *
373 : * When eager scanning is permanently disabled, either initially
374 : * (including for aggressive vacuum) or due to hitting the success cap,
375 : * this is set to InvalidBlockNumber.
376 : */
377 : BlockNumber next_eager_scan_region_start;
378 :
379 : /*
380 : * The remaining number of blocks a normal vacuum will consider eager
381 : * scanning when it is successful. When eager scanning is enabled, this is
382 : * initialized to MAX_EAGER_FREEZE_SUCCESS_RATE of the total number of
383 : * all-visible but not all-frozen pages. For each eager freeze success,
384 : * this is decremented. Once it hits 0, eager scanning is permanently
385 : * disabled. It is initialized to 0 if eager scanning starts out disabled
386 : * (including for aggressive vacuum).
387 : */
388 : BlockNumber eager_scan_remaining_successes;
389 :
390 : /*
391 : * The maximum number of blocks which may be eagerly scanned and not
392 : * frozen before eager scanning is temporarily suspended. This is
393 : * configurable both globally, via the
394 : * vacuum_max_eager_freeze_failure_rate GUC, and per table, with a table
395 : * storage parameter of the same name. It is calculated as
396 : * vacuum_max_eager_freeze_failure_rate of EAGER_SCAN_REGION_SIZE blocks.
397 : * It is 0 when eager scanning is disabled.
398 : */
399 : BlockNumber eager_scan_max_fails_per_region;
400 :
401 : /*
402 : * The number of eagerly scanned blocks vacuum failed to freeze (due to
403 : * age) in the current eager scan region. Vacuum resets it to
404 : * eager_scan_max_fails_per_region each time it enters a new region of the
405 : * relation. If eager_scan_remaining_fails hits 0, eager scanning is
406 : * suspended until the next region. It is also 0 if eager scanning has
407 : * been permanently disabled.
408 : */
409 : BlockNumber eager_scan_remaining_fails;
410 : } LVRelState;
411 :
412 :
413 : /* Struct for saving and restoring vacuum error information. */
414 : typedef struct LVSavedErrInfo
415 : {
416 : BlockNumber blkno;
417 : OffsetNumber offnum;
418 : VacErrPhase phase;
419 : } LVSavedErrInfo;
420 :
421 :
422 : /* non-export function prototypes */
423 : static void lazy_scan_heap(LVRelState *vacrel);
424 : static void heap_vacuum_eager_scan_setup(LVRelState *vacrel,
425 : const VacuumParams params);
426 : static BlockNumber heap_vac_scan_next_block(ReadStream *stream,
427 : void *callback_private_data,
428 : void *per_buffer_data);
429 : static void find_next_unskippable_block(LVRelState *vacrel, bool *skipsallvis);
430 : static bool lazy_scan_new_or_empty(LVRelState *vacrel, Buffer buf,
431 : BlockNumber blkno, Page page,
432 : bool sharelock, Buffer vmbuffer);
433 : static int lazy_scan_prune(LVRelState *vacrel, Buffer buf,
434 : BlockNumber blkno, Page page,
435 : Buffer vmbuffer, bool all_visible_according_to_vm,
436 : bool *has_lpdead_items, bool *vm_page_frozen);
437 : static bool lazy_scan_noprune(LVRelState *vacrel, Buffer buf,
438 : BlockNumber blkno, Page page,
439 : bool *has_lpdead_items);
440 : static void lazy_vacuum(LVRelState *vacrel);
441 : static bool lazy_vacuum_all_indexes(LVRelState *vacrel);
442 : static void lazy_vacuum_heap_rel(LVRelState *vacrel);
443 : static void lazy_vacuum_heap_page(LVRelState *vacrel, BlockNumber blkno,
444 : Buffer buffer, OffsetNumber *deadoffsets,
445 : int num_offsets, Buffer vmbuffer);
446 : static bool lazy_check_wraparound_failsafe(LVRelState *vacrel);
447 : static void lazy_cleanup_all_indexes(LVRelState *vacrel);
448 : static IndexBulkDeleteResult *lazy_vacuum_one_index(Relation indrel,
449 : IndexBulkDeleteResult *istat,
450 : double reltuples,
451 : LVRelState *vacrel);
452 : static IndexBulkDeleteResult *lazy_cleanup_one_index(Relation indrel,
453 : IndexBulkDeleteResult *istat,
454 : double reltuples,
455 : bool estimated_count,
456 : LVRelState *vacrel);
457 : static bool should_attempt_truncation(LVRelState *vacrel);
458 : static void lazy_truncate_heap(LVRelState *vacrel);
459 : static BlockNumber count_nondeletable_pages(LVRelState *vacrel,
460 : bool *lock_waiter_detected);
461 : static void dead_items_alloc(LVRelState *vacrel, int nworkers);
462 : static void dead_items_add(LVRelState *vacrel, BlockNumber blkno, OffsetNumber *offsets,
463 : int num_offsets);
464 : static void dead_items_reset(LVRelState *vacrel);
465 : static void dead_items_cleanup(LVRelState *vacrel);
466 :
467 : #ifdef USE_ASSERT_CHECKING
468 : static bool heap_page_is_all_visible(Relation rel, Buffer buf,
469 : TransactionId OldestXmin,
470 : bool *all_frozen,
471 : TransactionId *visibility_cutoff_xid,
472 : OffsetNumber *logging_offnum);
473 : #endif
474 : static bool heap_page_would_be_all_visible(Relation rel, Buffer buf,
475 : TransactionId OldestXmin,
476 : OffsetNumber *deadoffsets,
477 : int ndeadoffsets,
478 : bool *all_frozen,
479 : TransactionId *visibility_cutoff_xid,
480 : OffsetNumber *logging_offnum);
481 : static void update_relstats_all_indexes(LVRelState *vacrel);
482 : static void vacuum_error_callback(void *arg);
483 : static void update_vacuum_error_info(LVRelState *vacrel,
484 : LVSavedErrInfo *saved_vacrel,
485 : int phase, BlockNumber blkno,
486 : OffsetNumber offnum);
487 : static void restore_vacuum_error_info(LVRelState *vacrel,
488 : const LVSavedErrInfo *saved_vacrel);
489 :
490 :
491 :
492 : /*
493 : * Helper to set up the eager scanning state for vacuuming a single relation.
494 : * Initializes the eager scan management related members of the LVRelState.
495 : *
496 : * Caller provides whether or not an aggressive vacuum is required due to
497 : * vacuum options or for relfrozenxid/relminmxid advancement.
498 : */
499 : static void
500 250582 : heap_vacuum_eager_scan_setup(LVRelState *vacrel, const VacuumParams params)
501 : {
502 : uint32 randseed;
503 : BlockNumber allvisible;
504 : BlockNumber allfrozen;
505 : float first_region_ratio;
506 250582 : bool oldest_unfrozen_before_cutoff = false;
507 :
508 : /*
509 : * Initialize eager scan management fields to their disabled values.
510 : * Aggressive vacuums, normal vacuums of small tables, and normal vacuums
511 : * of tables without sufficiently old tuples disable eager scanning.
512 : */
513 250582 : vacrel->next_eager_scan_region_start = InvalidBlockNumber;
514 250582 : vacrel->eager_scan_max_fails_per_region = 0;
515 250582 : vacrel->eager_scan_remaining_fails = 0;
516 250582 : vacrel->eager_scan_remaining_successes = 0;
517 :
518 : /* If eager scanning is explicitly disabled, just return. */
519 250582 : if (params.max_eager_freeze_failure_rate == 0)
520 250582 : return;
521 :
522 : /*
523 : * The caller will have determined whether or not an aggressive vacuum is
524 : * required by either the vacuum parameters or the relative age of the
525 : * oldest unfrozen transaction IDs. An aggressive vacuum must scan every
526 : * all-visible page to safely advance the relfrozenxid and/or relminmxid,
527 : * so scans of all-visible pages are not considered eager.
528 : */
529 250582 : if (vacrel->aggressive)
530 239710 : return;
531 :
532 : /*
533 : * Aggressively vacuuming a small relation shouldn't take long, so it
534 : * isn't worth amortizing. We use two times the region size as the size
535 : * cutoff because the eager scan start block is a random spot somewhere in
536 : * the first region, making the second region the first to be eager
537 : * scanned normally.
538 : */
539 10872 : if (vacrel->rel_pages < 2 * EAGER_SCAN_REGION_SIZE)
540 10872 : return;
541 :
542 : /*
543 : * We only want to enable eager scanning if we are likely to be able to
544 : * freeze some of the pages in the relation.
545 : *
546 : * Tuples with XIDs older than OldestXmin or MXIDs older than OldestMxact
547 : * are technically freezable, but we won't freeze them unless the criteria
548 : * for opportunistic freezing is met. Only tuples with XIDs/MXIDs older
549 : * than the FreezeLimit/MultiXactCutoff are frozen in the common case.
550 : *
551 : * So, as a heuristic, we wait until the FreezeLimit has advanced past the
552 : * relfrozenxid or the MultiXactCutoff has advanced past the relminmxid to
553 : * enable eager scanning.
554 : */
555 0 : if (TransactionIdIsNormal(vacrel->cutoffs.relfrozenxid) &&
556 0 : TransactionIdPrecedes(vacrel->cutoffs.relfrozenxid,
557 : vacrel->cutoffs.FreezeLimit))
558 0 : oldest_unfrozen_before_cutoff = true;
559 :
560 0 : if (!oldest_unfrozen_before_cutoff &&
561 0 : MultiXactIdIsValid(vacrel->cutoffs.relminmxid) &&
562 0 : MultiXactIdPrecedes(vacrel->cutoffs.relminmxid,
563 : vacrel->cutoffs.MultiXactCutoff))
564 0 : oldest_unfrozen_before_cutoff = true;
565 :
566 0 : if (!oldest_unfrozen_before_cutoff)
567 0 : return;
568 :
569 : /* We have met the criteria to eagerly scan some pages. */
570 :
571 : /*
572 : * Our success cap is MAX_EAGER_FREEZE_SUCCESS_RATE of the number of
573 : * all-visible but not all-frozen blocks in the relation.
574 : */
575 0 : visibilitymap_count(vacrel->rel, &allvisible, &allfrozen);
576 :
577 0 : vacrel->eager_scan_remaining_successes =
578 0 : (BlockNumber) (MAX_EAGER_FREEZE_SUCCESS_RATE *
579 0 : (allvisible - allfrozen));
580 :
581 : /* If every all-visible page is frozen, eager scanning is disabled. */
582 0 : if (vacrel->eager_scan_remaining_successes == 0)
583 0 : return;
584 :
585 : /*
586 : * Now calculate the bounds of the first eager scan region. Its end block
587 : * will be a random spot somewhere in the first EAGER_SCAN_REGION_SIZE
588 : * blocks. This affects the bounds of all subsequent regions and avoids
589 : * eager scanning and failing to freeze the same blocks each vacuum of the
590 : * relation.
591 : */
592 0 : randseed = pg_prng_uint32(&pg_global_prng_state);
593 :
594 0 : vacrel->next_eager_scan_region_start = randseed % EAGER_SCAN_REGION_SIZE;
595 :
596 : Assert(params.max_eager_freeze_failure_rate > 0 &&
597 : params.max_eager_freeze_failure_rate <= 1);
598 :
599 0 : vacrel->eager_scan_max_fails_per_region =
600 0 : params.max_eager_freeze_failure_rate *
601 : EAGER_SCAN_REGION_SIZE;
602 :
603 : /*
604 : * The first region will be smaller than subsequent regions. As such,
605 : * adjust the eager freeze failures tolerated for this region.
606 : */
607 0 : first_region_ratio = 1 - (float) vacrel->next_eager_scan_region_start /
608 : EAGER_SCAN_REGION_SIZE;
609 :
610 0 : vacrel->eager_scan_remaining_fails =
611 0 : vacrel->eager_scan_max_fails_per_region *
612 : first_region_ratio;
613 : }
614 :
615 : /*
616 : * heap_vacuum_rel() -- perform VACUUM for one heap relation
617 : *
618 : * This routine sets things up for and then calls lazy_scan_heap, where
619 : * almost all work actually takes place. Finalizes everything after call
620 : * returns by managing relation truncation and updating rel's pg_class
621 : * entry. (Also updates pg_class entries for any indexes that need it.)
622 : *
623 : * At entry, we have already established a transaction and opened
624 : * and locked the relation.
625 : */
626 : void
627 250582 : heap_vacuum_rel(Relation rel, const VacuumParams params,
628 : BufferAccessStrategy bstrategy)
629 : {
630 : LVRelState *vacrel;
631 : bool verbose,
632 : instrument,
633 : skipwithvm,
634 : frozenxid_updated,
635 : minmulti_updated;
636 : BlockNumber orig_rel_pages,
637 : new_rel_pages,
638 : new_rel_allvisible,
639 : new_rel_allfrozen;
640 : PGRUsage ru0;
641 250582 : TimestampTz starttime = 0;
642 250582 : PgStat_Counter startreadtime = 0,
643 250582 : startwritetime = 0;
644 250582 : WalUsage startwalusage = pgWalUsage;
645 250582 : BufferUsage startbufferusage = pgBufferUsage;
646 : ErrorContextCallback errcallback;
647 250582 : char **indnames = NULL;
648 :
649 250582 : verbose = (params.options & VACOPT_VERBOSE) != 0;
650 474738 : instrument = (verbose || (AmAutoVacuumWorkerProcess() &&
651 224156 : params.log_vacuum_min_duration >= 0));
652 250582 : if (instrument)
653 : {
654 224180 : pg_rusage_init(&ru0);
655 224180 : if (track_io_timing)
656 : {
657 0 : startreadtime = pgStatBlockReadTime;
658 0 : startwritetime = pgStatBlockWriteTime;
659 : }
660 : }
661 :
662 : /* Used for instrumentation and stats report */
663 250582 : starttime = GetCurrentTimestamp();
664 :
665 250582 : pgstat_progress_start_command(PROGRESS_COMMAND_VACUUM,
666 : RelationGetRelid(rel));
667 :
668 : /*
669 : * Setup error traceback support for ereport() first. The idea is to set
670 : * up an error context callback to display additional information on any
671 : * error during a vacuum. During different phases of vacuum, we update
672 : * the state so that the error context callback always display current
673 : * information.
674 : *
675 : * Copy the names of heap rel into local memory for error reporting
676 : * purposes, too. It isn't always safe to assume that we can get the name
677 : * of each rel. It's convenient for code in lazy_scan_heap to always use
678 : * these temp copies.
679 : */
680 250582 : vacrel = (LVRelState *) palloc0(sizeof(LVRelState));
681 250582 : vacrel->dbname = get_database_name(MyDatabaseId);
682 250582 : vacrel->relnamespace = get_namespace_name(RelationGetNamespace(rel));
683 250582 : vacrel->relname = pstrdup(RelationGetRelationName(rel));
684 250582 : vacrel->indname = NULL;
685 250582 : vacrel->phase = VACUUM_ERRCB_PHASE_UNKNOWN;
686 250582 : vacrel->verbose = verbose;
687 250582 : errcallback.callback = vacuum_error_callback;
688 250582 : errcallback.arg = vacrel;
689 250582 : errcallback.previous = error_context_stack;
690 250582 : error_context_stack = &errcallback;
691 :
692 : /* Set up high level stuff about rel and its indexes */
693 250582 : vacrel->rel = rel;
694 250582 : vac_open_indexes(vacrel->rel, RowExclusiveLock, &vacrel->nindexes,
695 : &vacrel->indrels);
696 250582 : vacrel->bstrategy = bstrategy;
697 250582 : if (instrument && vacrel->nindexes > 0)
698 : {
699 : /* Copy index names used by instrumentation (not error reporting) */
700 214434 : indnames = palloc(sizeof(char *) * vacrel->nindexes);
701 551566 : for (int i = 0; i < vacrel->nindexes; i++)
702 337132 : indnames[i] = pstrdup(RelationGetRelationName(vacrel->indrels[i]));
703 : }
704 :
705 : /*
706 : * The index_cleanup param either disables index vacuuming and cleanup or
707 : * forces it to go ahead when we would otherwise apply the index bypass
708 : * optimization. The default is 'auto', which leaves the final decision
709 : * up to lazy_vacuum().
710 : *
711 : * The truncate param allows user to avoid attempting relation truncation,
712 : * though it can't force truncation to happen.
713 : */
714 : Assert(params.index_cleanup != VACOPTVALUE_UNSPECIFIED);
715 : Assert(params.truncate != VACOPTVALUE_UNSPECIFIED &&
716 : params.truncate != VACOPTVALUE_AUTO);
717 :
718 : /*
719 : * While VacuumFailSafeActive is reset to false before calling this, we
720 : * still need to reset it here due to recursive calls.
721 : */
722 250582 : VacuumFailsafeActive = false;
723 250582 : vacrel->consider_bypass_optimization = true;
724 250582 : vacrel->do_index_vacuuming = true;
725 250582 : vacrel->do_index_cleanup = true;
726 250582 : vacrel->do_rel_truncate = (params.truncate != VACOPTVALUE_DISABLED);
727 250582 : if (params.index_cleanup == VACOPTVALUE_DISABLED)
728 : {
729 : /* Force disable index vacuuming up-front */
730 260 : vacrel->do_index_vacuuming = false;
731 260 : vacrel->do_index_cleanup = false;
732 : }
733 250322 : else if (params.index_cleanup == VACOPTVALUE_ENABLED)
734 : {
735 : /* Force index vacuuming. Note that failsafe can still bypass. */
736 30 : vacrel->consider_bypass_optimization = false;
737 : }
738 : else
739 : {
740 : /* Default/auto, make all decisions dynamically */
741 : Assert(params.index_cleanup == VACOPTVALUE_AUTO);
742 : }
743 :
744 : /* Initialize page counters explicitly (be tidy) */
745 250582 : vacrel->scanned_pages = 0;
746 250582 : vacrel->eager_scanned_pages = 0;
747 250582 : vacrel->removed_pages = 0;
748 250582 : vacrel->new_frozen_tuple_pages = 0;
749 250582 : vacrel->lpdead_item_pages = 0;
750 250582 : vacrel->missed_dead_pages = 0;
751 250582 : vacrel->nonempty_pages = 0;
752 : /* dead_items_alloc allocates vacrel->dead_items later on */
753 :
754 : /* Allocate/initialize output statistics state */
755 250582 : vacrel->new_rel_tuples = 0;
756 250582 : vacrel->new_live_tuples = 0;
757 250582 : vacrel->indstats = (IndexBulkDeleteResult **)
758 250582 : palloc0(vacrel->nindexes * sizeof(IndexBulkDeleteResult *));
759 :
760 : /* Initialize remaining counters (be tidy) */
761 250582 : vacrel->num_index_scans = 0;
762 250582 : vacrel->tuples_deleted = 0;
763 250582 : vacrel->tuples_frozen = 0;
764 250582 : vacrel->lpdead_items = 0;
765 250582 : vacrel->live_tuples = 0;
766 250582 : vacrel->recently_dead_tuples = 0;
767 250582 : vacrel->missed_dead_tuples = 0;
768 :
769 250582 : vacrel->vm_new_visible_pages = 0;
770 250582 : vacrel->vm_new_visible_frozen_pages = 0;
771 250582 : vacrel->vm_new_frozen_pages = 0;
772 :
773 : /*
774 : * Get cutoffs that determine which deleted tuples are considered DEAD,
775 : * not just RECENTLY_DEAD, and which XIDs/MXIDs to freeze. Then determine
776 : * the extent of the blocks that we'll scan in lazy_scan_heap. It has to
777 : * happen in this order to ensure that the OldestXmin cutoff field works
778 : * as an upper bound on the XIDs stored in the pages we'll actually scan
779 : * (NewRelfrozenXid tracking must never be allowed to miss unfrozen XIDs).
780 : *
781 : * Next acquire vistest, a related cutoff that's used in pruning. We use
782 : * vistest in combination with OldestXmin to ensure that
783 : * heap_page_prune_and_freeze() always removes any deleted tuple whose
784 : * xmax is < OldestXmin. lazy_scan_prune must never become confused about
785 : * whether a tuple should be frozen or removed. (In the future we might
786 : * want to teach lazy_scan_prune to recompute vistest from time to time,
787 : * to increase the number of dead tuples it can prune away.)
788 : */
789 250582 : vacrel->aggressive = vacuum_get_cutoffs(rel, params, &vacrel->cutoffs);
790 250582 : vacrel->rel_pages = orig_rel_pages = RelationGetNumberOfBlocks(rel);
791 250582 : vacrel->vistest = GlobalVisTestFor(rel);
792 :
793 : /* Initialize state used to track oldest extant XID/MXID */
794 250582 : vacrel->NewRelfrozenXid = vacrel->cutoffs.OldestXmin;
795 250582 : vacrel->NewRelminMxid = vacrel->cutoffs.OldestMxact;
796 :
797 : /*
798 : * Initialize state related to tracking all-visible page skipping. This is
799 : * very important to determine whether or not it is safe to advance the
800 : * relfrozenxid/relminmxid.
801 : */
802 250582 : vacrel->skippedallvis = false;
803 250582 : skipwithvm = true;
804 250582 : if (params.options & VACOPT_DISABLE_PAGE_SKIPPING)
805 : {
806 : /*
807 : * Force aggressive mode, and disable skipping blocks using the
808 : * visibility map (even those set all-frozen)
809 : */
810 344 : vacrel->aggressive = true;
811 344 : skipwithvm = false;
812 : }
813 :
814 250582 : vacrel->skipwithvm = skipwithvm;
815 :
816 : /*
817 : * Set up eager scan tracking state. This must happen after determining
818 : * whether or not the vacuum must be aggressive, because only normal
819 : * vacuums use the eager scan algorithm.
820 : */
821 250582 : heap_vacuum_eager_scan_setup(vacrel, params);
822 :
823 250582 : if (verbose)
824 : {
825 24 : if (vacrel->aggressive)
826 2 : ereport(INFO,
827 : (errmsg("aggressively vacuuming \"%s.%s.%s\"",
828 : vacrel->dbname, vacrel->relnamespace,
829 : vacrel->relname)));
830 : else
831 22 : ereport(INFO,
832 : (errmsg("vacuuming \"%s.%s.%s\"",
833 : vacrel->dbname, vacrel->relnamespace,
834 : vacrel->relname)));
835 : }
836 :
837 : /*
838 : * Allocate dead_items memory using dead_items_alloc. This handles
839 : * parallel VACUUM initialization as part of allocating shared memory
840 : * space used for dead_items. (But do a failsafe precheck first, to
841 : * ensure that parallel VACUUM won't be attempted at all when relfrozenxid
842 : * is already dangerously old.)
843 : */
844 250582 : lazy_check_wraparound_failsafe(vacrel);
845 250582 : dead_items_alloc(vacrel, params.nworkers);
846 :
847 : /*
848 : * Call lazy_scan_heap to perform all required heap pruning, index
849 : * vacuuming, and heap vacuuming (plus related processing)
850 : */
851 250582 : lazy_scan_heap(vacrel);
852 :
853 : /*
854 : * Free resources managed by dead_items_alloc. This ends parallel mode in
855 : * passing when necessary.
856 : */
857 250582 : dead_items_cleanup(vacrel);
858 : Assert(!IsInParallelMode());
859 :
860 : /*
861 : * Update pg_class entries for each of rel's indexes where appropriate.
862 : *
863 : * Unlike the later update to rel's pg_class entry, this is not critical.
864 : * Maintains relpages/reltuples statistics used by the planner only.
865 : */
866 250582 : if (vacrel->do_index_cleanup)
867 167842 : update_relstats_all_indexes(vacrel);
868 :
869 : /* Done with rel's indexes */
870 250582 : vac_close_indexes(vacrel->nindexes, vacrel->indrels, NoLock);
871 :
872 : /* Optionally truncate rel */
873 250582 : if (should_attempt_truncation(vacrel))
874 286 : lazy_truncate_heap(vacrel);
875 :
876 : /* Pop the error context stack */
877 250582 : error_context_stack = errcallback.previous;
878 :
879 : /* Report that we are now doing final cleanup */
880 250582 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
881 : PROGRESS_VACUUM_PHASE_FINAL_CLEANUP);
882 :
883 : /*
884 : * Prepare to update rel's pg_class entry.
885 : *
886 : * Aggressive VACUUMs must always be able to advance relfrozenxid to a
887 : * value >= FreezeLimit, and relminmxid to a value >= MultiXactCutoff.
888 : * Non-aggressive VACUUMs may advance them by any amount, or not at all.
889 : */
890 : Assert(vacrel->NewRelfrozenXid == vacrel->cutoffs.OldestXmin ||
891 : TransactionIdPrecedesOrEquals(vacrel->aggressive ? vacrel->cutoffs.FreezeLimit :
892 : vacrel->cutoffs.relfrozenxid,
893 : vacrel->NewRelfrozenXid));
894 : Assert(vacrel->NewRelminMxid == vacrel->cutoffs.OldestMxact ||
895 : MultiXactIdPrecedesOrEquals(vacrel->aggressive ? vacrel->cutoffs.MultiXactCutoff :
896 : vacrel->cutoffs.relminmxid,
897 : vacrel->NewRelminMxid));
898 250582 : if (vacrel->skippedallvis)
899 : {
900 : /*
901 : * Must keep original relfrozenxid in a non-aggressive VACUUM that
902 : * chose to skip an all-visible page range. The state that tracks new
903 : * values will have missed unfrozen XIDs from the pages we skipped.
904 : */
905 : Assert(!vacrel->aggressive);
906 56 : vacrel->NewRelfrozenXid = InvalidTransactionId;
907 56 : vacrel->NewRelminMxid = InvalidMultiXactId;
908 : }
909 :
910 : /*
911 : * For safety, clamp relallvisible to be not more than what we're setting
912 : * pg_class.relpages to
913 : */
914 250582 : new_rel_pages = vacrel->rel_pages; /* After possible rel truncation */
915 250582 : visibilitymap_count(rel, &new_rel_allvisible, &new_rel_allfrozen);
916 250582 : if (new_rel_allvisible > new_rel_pages)
917 0 : new_rel_allvisible = new_rel_pages;
918 :
919 : /*
920 : * An all-frozen block _must_ be all-visible. As such, clamp the count of
921 : * all-frozen blocks to the count of all-visible blocks. This matches the
922 : * clamping of relallvisible above.
923 : */
924 250582 : if (new_rel_allfrozen > new_rel_allvisible)
925 0 : new_rel_allfrozen = new_rel_allvisible;
926 :
927 : /*
928 : * Now actually update rel's pg_class entry.
929 : *
930 : * In principle new_live_tuples could be -1 indicating that we (still)
931 : * don't know the tuple count. In practice that can't happen, since we
932 : * scan every page that isn't skipped using the visibility map.
933 : */
934 250582 : vac_update_relstats(rel, new_rel_pages, vacrel->new_live_tuples,
935 : new_rel_allvisible, new_rel_allfrozen,
936 250582 : vacrel->nindexes > 0,
937 : vacrel->NewRelfrozenXid, vacrel->NewRelminMxid,
938 : &frozenxid_updated, &minmulti_updated, false);
939 :
940 : /*
941 : * Report results to the cumulative stats system, too.
942 : *
943 : * Deliberately avoid telling the stats system about LP_DEAD items that
944 : * remain in the table due to VACUUM bypassing index and heap vacuuming.
945 : * ANALYZE will consider the remaining LP_DEAD items to be dead "tuples".
946 : * It seems like a good idea to err on the side of not vacuuming again too
947 : * soon in cases where the failsafe prevented significant amounts of heap
948 : * vacuuming.
949 : */
950 150384 : pgstat_report_vacuum(RelationGetRelid(rel),
951 250582 : rel->rd_rel->relisshared,
952 100198 : Max(vacrel->new_live_tuples, 0),
953 250582 : vacrel->recently_dead_tuples +
954 250582 : vacrel->missed_dead_tuples,
955 : starttime);
956 250582 : pgstat_progress_end_command();
957 :
958 250582 : if (instrument)
959 : {
960 224180 : TimestampTz endtime = GetCurrentTimestamp();
961 :
962 224382 : if (verbose || params.log_vacuum_min_duration == 0 ||
963 202 : TimestampDifferenceExceeds(starttime, endtime,
964 202 : params.log_vacuum_min_duration))
965 : {
966 : long secs_dur;
967 : int usecs_dur;
968 : WalUsage walusage;
969 : BufferUsage bufferusage;
970 : StringInfoData buf;
971 : char *msgfmt;
972 : int32 diff;
973 223978 : double read_rate = 0,
974 223978 : write_rate = 0;
975 : int64 total_blks_hit;
976 : int64 total_blks_read;
977 : int64 total_blks_dirtied;
978 :
979 223978 : TimestampDifference(starttime, endtime, &secs_dur, &usecs_dur);
980 223978 : memset(&walusage, 0, sizeof(WalUsage));
981 223978 : WalUsageAccumDiff(&walusage, &pgWalUsage, &startwalusage);
982 223978 : memset(&bufferusage, 0, sizeof(BufferUsage));
983 223978 : BufferUsageAccumDiff(&bufferusage, &pgBufferUsage, &startbufferusage);
984 :
985 223978 : total_blks_hit = bufferusage.shared_blks_hit +
986 223978 : bufferusage.local_blks_hit;
987 223978 : total_blks_read = bufferusage.shared_blks_read +
988 223978 : bufferusage.local_blks_read;
989 223978 : total_blks_dirtied = bufferusage.shared_blks_dirtied +
990 223978 : bufferusage.local_blks_dirtied;
991 :
992 223978 : initStringInfo(&buf);
993 223978 : if (verbose)
994 : {
995 : /*
996 : * Aggressiveness already reported earlier, in dedicated
997 : * VACUUM VERBOSE ereport
998 : */
999 : Assert(!params.is_wraparound);
1000 24 : msgfmt = _("finished vacuuming \"%s.%s.%s\": index scans: %d\n");
1001 : }
1002 223954 : else if (params.is_wraparound)
1003 : {
1004 : /*
1005 : * While it's possible for a VACUUM to be both is_wraparound
1006 : * and !aggressive, that's just a corner-case -- is_wraparound
1007 : * implies aggressive. Produce distinct output for the corner
1008 : * case all the same, just in case.
1009 : */
1010 223916 : if (vacrel->aggressive)
1011 223910 : msgfmt = _("automatic aggressive vacuum to prevent wraparound of table \"%s.%s.%s\": index scans: %d\n");
1012 : else
1013 6 : msgfmt = _("automatic vacuum to prevent wraparound of table \"%s.%s.%s\": index scans: %d\n");
1014 : }
1015 : else
1016 : {
1017 38 : if (vacrel->aggressive)
1018 32 : msgfmt = _("automatic aggressive vacuum of table \"%s.%s.%s\": index scans: %d\n");
1019 : else
1020 6 : msgfmt = _("automatic vacuum of table \"%s.%s.%s\": index scans: %d\n");
1021 : }
1022 223978 : appendStringInfo(&buf, msgfmt,
1023 : vacrel->dbname,
1024 : vacrel->relnamespace,
1025 : vacrel->relname,
1026 : vacrel->num_index_scans);
1027 313118 : appendStringInfo(&buf, _("pages: %u removed, %u remain, %u scanned (%.2f%% of total), %u eagerly scanned\n"),
1028 : vacrel->removed_pages,
1029 : new_rel_pages,
1030 : vacrel->scanned_pages,
1031 : orig_rel_pages == 0 ? 100.0 :
1032 89140 : 100.0 * vacrel->scanned_pages /
1033 : orig_rel_pages,
1034 : vacrel->eager_scanned_pages);
1035 223978 : appendStringInfo(&buf,
1036 223978 : _("tuples: %" PRId64 " removed, %" PRId64 " remain, %" PRId64 " are dead but not yet removable\n"),
1037 : vacrel->tuples_deleted,
1038 223978 : (int64) vacrel->new_rel_tuples,
1039 : vacrel->recently_dead_tuples);
1040 223978 : if (vacrel->missed_dead_tuples > 0)
1041 0 : appendStringInfo(&buf,
1042 0 : _("tuples missed: %" PRId64 " dead from %u pages not removed due to cleanup lock contention\n"),
1043 : vacrel->missed_dead_tuples,
1044 : vacrel->missed_dead_pages);
1045 223978 : diff = (int32) (ReadNextTransactionId() -
1046 223978 : vacrel->cutoffs.OldestXmin);
1047 223978 : appendStringInfo(&buf,
1048 223978 : _("removable cutoff: %u, which was %d XIDs old when operation ended\n"),
1049 : vacrel->cutoffs.OldestXmin, diff);
1050 223978 : if (frozenxid_updated)
1051 : {
1052 35186 : diff = (int32) (vacrel->NewRelfrozenXid -
1053 35186 : vacrel->cutoffs.relfrozenxid);
1054 35186 : appendStringInfo(&buf,
1055 35186 : _("new relfrozenxid: %u, which is %d XIDs ahead of previous value\n"),
1056 : vacrel->NewRelfrozenXid, diff);
1057 : }
1058 223978 : if (minmulti_updated)
1059 : {
1060 8 : diff = (int32) (vacrel->NewRelminMxid -
1061 8 : vacrel->cutoffs.relminmxid);
1062 8 : appendStringInfo(&buf,
1063 8 : _("new relminmxid: %u, which is %d MXIDs ahead of previous value\n"),
1064 : vacrel->NewRelminMxid, diff);
1065 : }
1066 313118 : appendStringInfo(&buf, _("frozen: %u pages from table (%.2f%% of total) had %" PRId64 " tuples frozen\n"),
1067 : vacrel->new_frozen_tuple_pages,
1068 : orig_rel_pages == 0 ? 100.0 :
1069 89140 : 100.0 * vacrel->new_frozen_tuple_pages /
1070 : orig_rel_pages,
1071 : vacrel->tuples_frozen);
1072 :
1073 223978 : appendStringInfo(&buf,
1074 223978 : _("visibility map: %u pages set all-visible, %u pages set all-frozen (%u were all-visible)\n"),
1075 : vacrel->vm_new_visible_pages,
1076 223978 : vacrel->vm_new_visible_frozen_pages +
1077 223978 : vacrel->vm_new_frozen_pages,
1078 : vacrel->vm_new_frozen_pages);
1079 223978 : if (vacrel->do_index_vacuuming)
1080 : {
1081 141556 : if (vacrel->nindexes == 0 || vacrel->num_index_scans == 0)
1082 141526 : appendStringInfoString(&buf, _("index scan not needed: "));
1083 : else
1084 30 : appendStringInfoString(&buf, _("index scan needed: "));
1085 :
1086 141556 : msgfmt = _("%u pages from table (%.2f%% of total) had %" PRId64 " dead item identifiers removed\n");
1087 : }
1088 : else
1089 : {
1090 82422 : if (!VacuumFailsafeActive)
1091 0 : appendStringInfoString(&buf, _("index scan bypassed: "));
1092 : else
1093 82422 : appendStringInfoString(&buf, _("index scan bypassed by failsafe: "));
1094 :
1095 82422 : msgfmt = _("%u pages from table (%.2f%% of total) have %" PRId64 " dead item identifiers\n");
1096 : }
1097 313118 : appendStringInfo(&buf, msgfmt,
1098 : vacrel->lpdead_item_pages,
1099 : orig_rel_pages == 0 ? 100.0 :
1100 89140 : 100.0 * vacrel->lpdead_item_pages / orig_rel_pages,
1101 : vacrel->lpdead_items);
1102 560748 : for (int i = 0; i < vacrel->nindexes; i++)
1103 : {
1104 336770 : IndexBulkDeleteResult *istat = vacrel->indstats[i];
1105 :
1106 336770 : if (!istat)
1107 336720 : continue;
1108 :
1109 50 : appendStringInfo(&buf,
1110 50 : _("index \"%s\": pages: %u in total, %u newly deleted, %u currently deleted, %u reusable\n"),
1111 50 : indnames[i],
1112 : istat->num_pages,
1113 : istat->pages_newly_deleted,
1114 : istat->pages_deleted,
1115 : istat->pages_free);
1116 : }
1117 223978 : if (track_cost_delay_timing)
1118 : {
1119 : /*
1120 : * We bypass the changecount mechanism because this value is
1121 : * only updated by the calling process. We also rely on the
1122 : * above call to pgstat_progress_end_command() to not clear
1123 : * the st_progress_param array.
1124 : */
1125 0 : appendStringInfo(&buf, _("delay time: %.3f ms\n"),
1126 0 : (double) MyBEEntry->st_progress_param[PROGRESS_VACUUM_DELAY_TIME] / 1000000.0);
1127 : }
1128 223978 : if (track_io_timing)
1129 : {
1130 0 : double read_ms = (double) (pgStatBlockReadTime - startreadtime) / 1000;
1131 0 : double write_ms = (double) (pgStatBlockWriteTime - startwritetime) / 1000;
1132 :
1133 0 : appendStringInfo(&buf, _("I/O timings: read: %.3f ms, write: %.3f ms\n"),
1134 : read_ms, write_ms);
1135 : }
1136 223978 : if (secs_dur > 0 || usecs_dur > 0)
1137 : {
1138 223978 : read_rate = (double) BLCKSZ * total_blks_read /
1139 223978 : (1024 * 1024) / (secs_dur + usecs_dur / 1000000.0);
1140 223978 : write_rate = (double) BLCKSZ * total_blks_dirtied /
1141 223978 : (1024 * 1024) / (secs_dur + usecs_dur / 1000000.0);
1142 : }
1143 223978 : appendStringInfo(&buf, _("avg read rate: %.3f MB/s, avg write rate: %.3f MB/s\n"),
1144 : read_rate, write_rate);
1145 223978 : appendStringInfo(&buf,
1146 223978 : _("buffer usage: %" PRId64 " hits, %" PRId64 " reads, %" PRId64 " dirtied\n"),
1147 : total_blks_hit,
1148 : total_blks_read,
1149 : total_blks_dirtied);
1150 223978 : appendStringInfo(&buf,
1151 223978 : _("WAL usage: %" PRId64 " records, %" PRId64 " full page images, %" PRIu64 " bytes, %" PRIu64 " full page image bytes, %" PRId64 " buffers full\n"),
1152 : walusage.wal_records,
1153 : walusage.wal_fpi,
1154 : walusage.wal_bytes,
1155 : walusage.wal_fpi_bytes,
1156 : walusage.wal_buffers_full);
1157 223978 : appendStringInfo(&buf, _("system usage: %s"), pg_rusage_show(&ru0));
1158 :
1159 223978 : ereport(verbose ? INFO : LOG,
1160 : (errmsg_internal("%s", buf.data)));
1161 223978 : pfree(buf.data);
1162 : }
1163 : }
1164 :
1165 : /* Cleanup index statistics and index names */
1166 625572 : for (int i = 0; i < vacrel->nindexes; i++)
1167 : {
1168 374990 : if (vacrel->indstats[i])
1169 2694 : pfree(vacrel->indstats[i]);
1170 :
1171 374990 : if (instrument)
1172 337132 : pfree(indnames[i]);
1173 : }
1174 250582 : }
1175 :
1176 : /*
1177 : * lazy_scan_heap() -- workhorse function for VACUUM
1178 : *
1179 : * This routine prunes each page in the heap, and considers the need to
1180 : * freeze remaining tuples with storage (not including pages that can be
1181 : * skipped using the visibility map). Also performs related maintenance
1182 : * of the FSM and visibility map. These steps all take place during an
1183 : * initial pass over the target heap relation.
1184 : *
1185 : * Also invokes lazy_vacuum_all_indexes to vacuum indexes, which largely
1186 : * consists of deleting index tuples that point to LP_DEAD items left in
1187 : * heap pages following pruning. Earlier initial pass over the heap will
1188 : * have collected the TIDs whose index tuples need to be removed.
1189 : *
1190 : * Finally, invokes lazy_vacuum_heap_rel to vacuum heap pages, which
1191 : * largely consists of marking LP_DEAD items (from vacrel->dead_items)
1192 : * as LP_UNUSED. This has to happen in a second, final pass over the
1193 : * heap, to preserve a basic invariant that all index AMs rely on: no
1194 : * extant index tuple can ever be allowed to contain a TID that points to
1195 : * an LP_UNUSED line pointer in the heap. We must disallow premature
1196 : * recycling of line pointers to avoid index scans that get confused
1197 : * about which TID points to which tuple immediately after recycling.
1198 : * (Actually, this isn't a concern when target heap relation happens to
1199 : * have no indexes, which allows us to safely apply the one-pass strategy
1200 : * as an optimization).
1201 : *
1202 : * In practice we often have enough space to fit all TIDs, and so won't
1203 : * need to call lazy_vacuum more than once, after our initial pass over
1204 : * the heap has totally finished. Otherwise things are slightly more
1205 : * complicated: our "initial pass" over the heap applies only to those
1206 : * pages that were pruned before we needed to call lazy_vacuum, and our
1207 : * "final pass" over the heap only vacuums these same heap pages.
1208 : * However, we process indexes in full every time lazy_vacuum is called,
1209 : * which makes index processing very inefficient when memory is in short
1210 : * supply.
1211 : */
1212 : static void
1213 250582 : lazy_scan_heap(LVRelState *vacrel)
1214 : {
1215 : ReadStream *stream;
1216 250582 : BlockNumber rel_pages = vacrel->rel_pages,
1217 250582 : blkno = 0,
1218 250582 : next_fsm_block_to_vacuum = 0;
1219 250582 : BlockNumber orig_eager_scan_success_limit =
1220 : vacrel->eager_scan_remaining_successes; /* for logging */
1221 250582 : Buffer vmbuffer = InvalidBuffer;
1222 250582 : const int initprog_index[] = {
1223 : PROGRESS_VACUUM_PHASE,
1224 : PROGRESS_VACUUM_TOTAL_HEAP_BLKS,
1225 : PROGRESS_VACUUM_MAX_DEAD_TUPLE_BYTES
1226 : };
1227 : int64 initprog_val[3];
1228 :
1229 : /* Report that we're scanning the heap, advertising total # of blocks */
1230 250582 : initprog_val[0] = PROGRESS_VACUUM_PHASE_SCAN_HEAP;
1231 250582 : initprog_val[1] = rel_pages;
1232 250582 : initprog_val[2] = vacrel->dead_items_info->max_bytes;
1233 250582 : pgstat_progress_update_multi_param(3, initprog_index, initprog_val);
1234 :
1235 : /* Initialize for the first heap_vac_scan_next_block() call */
1236 250582 : vacrel->current_block = InvalidBlockNumber;
1237 250582 : vacrel->next_unskippable_block = InvalidBlockNumber;
1238 250582 : vacrel->next_unskippable_allvis = false;
1239 250582 : vacrel->next_unskippable_eager_scanned = false;
1240 250582 : vacrel->next_unskippable_vmbuffer = InvalidBuffer;
1241 :
1242 : /*
1243 : * Set up the read stream for vacuum's first pass through the heap.
1244 : *
1245 : * This could be made safe for READ_STREAM_USE_BATCHING, but only with
1246 : * explicit work in heap_vac_scan_next_block.
1247 : */
1248 250582 : stream = read_stream_begin_relation(READ_STREAM_MAINTENANCE,
1249 : vacrel->bstrategy,
1250 : vacrel->rel,
1251 : MAIN_FORKNUM,
1252 : heap_vac_scan_next_block,
1253 : vacrel,
1254 : sizeof(uint8));
1255 :
1256 : while (true)
1257 1176678 : {
1258 : Buffer buf;
1259 : Page page;
1260 1427260 : uint8 blk_info = 0;
1261 1427260 : int ndeleted = 0;
1262 : bool has_lpdead_items;
1263 1427260 : void *per_buffer_data = NULL;
1264 1427260 : bool vm_page_frozen = false;
1265 1427260 : bool got_cleanup_lock = false;
1266 :
1267 1427260 : vacuum_delay_point(false);
1268 :
1269 : /*
1270 : * Regularly check if wraparound failsafe should trigger.
1271 : *
1272 : * There is a similar check inside lazy_vacuum_all_indexes(), but
1273 : * relfrozenxid might start to look dangerously old before we reach
1274 : * that point. This check also provides failsafe coverage for the
1275 : * one-pass strategy, and the two-pass strategy with the index_cleanup
1276 : * param set to 'off'.
1277 : */
1278 1427260 : if (vacrel->scanned_pages > 0 &&
1279 1176678 : vacrel->scanned_pages % FAILSAFE_EVERY_PAGES == 0)
1280 0 : lazy_check_wraparound_failsafe(vacrel);
1281 :
1282 : /*
1283 : * Consider if we definitely have enough space to process TIDs on page
1284 : * already. If we are close to overrunning the available space for
1285 : * dead_items TIDs, pause and do a cycle of vacuuming before we tackle
1286 : * this page. However, let's force at least one page-worth of tuples
1287 : * to be stored as to ensure we do at least some work when the memory
1288 : * configured is so low that we run out before storing anything.
1289 : */
1290 1427260 : if (vacrel->dead_items_info->num_items > 0 &&
1291 50966 : TidStoreMemoryUsage(vacrel->dead_items) > vacrel->dead_items_info->max_bytes)
1292 : {
1293 : /*
1294 : * Before beginning index vacuuming, we release any pin we may
1295 : * hold on the visibility map page. This isn't necessary for
1296 : * correctness, but we do it anyway to avoid holding the pin
1297 : * across a lengthy, unrelated operation.
1298 : */
1299 14 : if (BufferIsValid(vmbuffer))
1300 : {
1301 14 : ReleaseBuffer(vmbuffer);
1302 14 : vmbuffer = InvalidBuffer;
1303 : }
1304 :
1305 : /* Perform a round of index and heap vacuuming */
1306 14 : vacrel->consider_bypass_optimization = false;
1307 14 : lazy_vacuum(vacrel);
1308 :
1309 : /*
1310 : * Vacuum the Free Space Map to make newly-freed space visible on
1311 : * upper-level FSM pages. Note that blkno is the previously
1312 : * processed block.
1313 : */
1314 14 : FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
1315 : blkno + 1);
1316 14 : next_fsm_block_to_vacuum = blkno;
1317 :
1318 : /* Report that we are once again scanning the heap */
1319 14 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
1320 : PROGRESS_VACUUM_PHASE_SCAN_HEAP);
1321 : }
1322 :
1323 1427260 : buf = read_stream_next_buffer(stream, &per_buffer_data);
1324 :
1325 : /* The relation is exhausted. */
1326 1427260 : if (!BufferIsValid(buf))
1327 250582 : break;
1328 :
1329 1176678 : blk_info = *((uint8 *) per_buffer_data);
1330 1176678 : CheckBufferIsPinnedOnce(buf);
1331 1176678 : page = BufferGetPage(buf);
1332 1176678 : blkno = BufferGetBlockNumber(buf);
1333 :
1334 1176678 : vacrel->scanned_pages++;
1335 1176678 : if (blk_info & VAC_BLK_WAS_EAGER_SCANNED)
1336 0 : vacrel->eager_scanned_pages++;
1337 :
1338 : /* Report as block scanned, update error traceback information */
1339 1176678 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_SCANNED, blkno);
1340 1176678 : update_vacuum_error_info(vacrel, NULL, VACUUM_ERRCB_PHASE_SCAN_HEAP,
1341 : blkno, InvalidOffsetNumber);
1342 :
1343 : /*
1344 : * Pin the visibility map page in case we need to mark the page
1345 : * all-visible. In most cases this will be very cheap, because we'll
1346 : * already have the correct page pinned anyway.
1347 : */
1348 1176678 : visibilitymap_pin(vacrel->rel, blkno, &vmbuffer);
1349 :
1350 : /*
1351 : * We need a buffer cleanup lock to prune HOT chains and defragment
1352 : * the page in lazy_scan_prune. But when it's not possible to acquire
1353 : * a cleanup lock right away, we may be able to settle for reduced
1354 : * processing using lazy_scan_noprune.
1355 : */
1356 1176678 : got_cleanup_lock = ConditionalLockBufferForCleanup(buf);
1357 :
1358 1176678 : if (!got_cleanup_lock)
1359 362 : LockBuffer(buf, BUFFER_LOCK_SHARE);
1360 :
1361 : /* Check for new or empty pages before lazy_scan_[no]prune call */
1362 1176678 : if (lazy_scan_new_or_empty(vacrel, buf, blkno, page, !got_cleanup_lock,
1363 1176678 : vmbuffer))
1364 : {
1365 : /* Processed as new/empty page (lock and pin released) */
1366 1224 : continue;
1367 : }
1368 :
1369 : /*
1370 : * If we didn't get the cleanup lock, we can still collect LP_DEAD
1371 : * items in the dead_items area for later vacuuming, count live and
1372 : * recently dead tuples for vacuum logging, and determine if this
1373 : * block could later be truncated. If we encounter any xid/mxids that
1374 : * require advancing the relfrozenxid/relminxid, we'll have to wait
1375 : * for a cleanup lock and call lazy_scan_prune().
1376 : */
1377 1175454 : if (!got_cleanup_lock &&
1378 362 : !lazy_scan_noprune(vacrel, buf, blkno, page, &has_lpdead_items))
1379 : {
1380 : /*
1381 : * lazy_scan_noprune could not do all required processing. Wait
1382 : * for a cleanup lock, and call lazy_scan_prune in the usual way.
1383 : */
1384 : Assert(vacrel->aggressive);
1385 126 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
1386 126 : LockBufferForCleanup(buf);
1387 126 : got_cleanup_lock = true;
1388 : }
1389 :
1390 : /*
1391 : * If we have a cleanup lock, we must now prune, freeze, and count
1392 : * tuples. We may have acquired the cleanup lock originally, or we may
1393 : * have gone back and acquired it after lazy_scan_noprune() returned
1394 : * false. Either way, the page hasn't been processed yet.
1395 : *
1396 : * Like lazy_scan_noprune(), lazy_scan_prune() will count
1397 : * recently_dead_tuples and live tuples for vacuum logging, determine
1398 : * if the block can later be truncated, and accumulate the details of
1399 : * remaining LP_DEAD line pointers on the page into dead_items. These
1400 : * dead items include those pruned by lazy_scan_prune() as well as
1401 : * line pointers previously marked LP_DEAD.
1402 : */
1403 1175454 : if (got_cleanup_lock)
1404 1175218 : ndeleted = lazy_scan_prune(vacrel, buf, blkno, page,
1405 : vmbuffer,
1406 1175218 : blk_info & VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM,
1407 : &has_lpdead_items, &vm_page_frozen);
1408 :
1409 : /*
1410 : * Count an eagerly scanned page as a failure or a success.
1411 : *
1412 : * Only lazy_scan_prune() freezes pages, so if we didn't get the
1413 : * cleanup lock, we won't have frozen the page. However, we only count
1414 : * pages that were too new to require freezing as eager freeze
1415 : * failures.
1416 : *
1417 : * We could gather more information from lazy_scan_noprune() about
1418 : * whether or not there were tuples with XIDs or MXIDs older than the
1419 : * FreezeLimit or MultiXactCutoff. However, for simplicity, we simply
1420 : * exclude pages skipped due to cleanup lock contention from eager
1421 : * freeze algorithm caps.
1422 : */
1423 1175454 : if (got_cleanup_lock &&
1424 1175218 : (blk_info & VAC_BLK_WAS_EAGER_SCANNED))
1425 : {
1426 : /* Aggressive vacuums do not eager scan. */
1427 : Assert(!vacrel->aggressive);
1428 :
1429 0 : if (vm_page_frozen)
1430 : {
1431 0 : if (vacrel->eager_scan_remaining_successes > 0)
1432 0 : vacrel->eager_scan_remaining_successes--;
1433 :
1434 0 : if (vacrel->eager_scan_remaining_successes == 0)
1435 : {
1436 : /*
1437 : * Report only once that we disabled eager scanning. We
1438 : * may eagerly read ahead blocks in excess of the success
1439 : * or failure caps before attempting to freeze them, so we
1440 : * could reach here even after disabling additional eager
1441 : * scanning.
1442 : */
1443 0 : if (vacrel->eager_scan_max_fails_per_region > 0)
1444 0 : ereport(vacrel->verbose ? INFO : DEBUG2,
1445 : (errmsg("disabling eager scanning after freezing %u eagerly scanned blocks of relation \"%s.%s.%s\"",
1446 : orig_eager_scan_success_limit,
1447 : vacrel->dbname, vacrel->relnamespace,
1448 : vacrel->relname)));
1449 :
1450 : /*
1451 : * If we hit our success cap, permanently disable eager
1452 : * scanning by setting the other eager scan management
1453 : * fields to their disabled values.
1454 : */
1455 0 : vacrel->eager_scan_remaining_fails = 0;
1456 0 : vacrel->next_eager_scan_region_start = InvalidBlockNumber;
1457 0 : vacrel->eager_scan_max_fails_per_region = 0;
1458 : }
1459 : }
1460 0 : else if (vacrel->eager_scan_remaining_fails > 0)
1461 0 : vacrel->eager_scan_remaining_fails--;
1462 : }
1463 :
1464 : /*
1465 : * Now drop the buffer lock and, potentially, update the FSM.
1466 : *
1467 : * Our goal is to update the freespace map the last time we touch the
1468 : * page. If we'll process a block in the second pass, we may free up
1469 : * additional space on the page, so it is better to update the FSM
1470 : * after the second pass. If the relation has no indexes, or if index
1471 : * vacuuming is disabled, there will be no second heap pass; if this
1472 : * particular page has no dead items, the second heap pass will not
1473 : * touch this page. So, in those cases, update the FSM now.
1474 : *
1475 : * Note: In corner cases, it's possible to miss updating the FSM
1476 : * entirely. If index vacuuming is currently enabled, we'll skip the
1477 : * FSM update now. But if failsafe mode is later activated, or there
1478 : * are so few dead tuples that index vacuuming is bypassed, there will
1479 : * also be no opportunity to update the FSM later, because we'll never
1480 : * revisit this page. Since updating the FSM is desirable but not
1481 : * absolutely required, that's OK.
1482 : */
1483 1175454 : if (vacrel->nindexes == 0
1484 1139526 : || !vacrel->do_index_vacuuming
1485 775356 : || !has_lpdead_items)
1486 1147166 : {
1487 1147166 : Size freespace = PageGetHeapFreeSpace(page);
1488 :
1489 1147166 : UnlockReleaseBuffer(buf);
1490 1147166 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
1491 :
1492 : /*
1493 : * Periodically perform FSM vacuuming to make newly-freed space
1494 : * visible on upper FSM pages. This is done after vacuuming if the
1495 : * table has indexes. There will only be newly-freed space if we
1496 : * held the cleanup lock and lazy_scan_prune() was called.
1497 : */
1498 1147166 : if (got_cleanup_lock && vacrel->nindexes == 0 && ndeleted > 0 &&
1499 910 : blkno - next_fsm_block_to_vacuum >= VACUUM_FSM_EVERY_PAGES)
1500 : {
1501 0 : FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
1502 : blkno);
1503 0 : next_fsm_block_to_vacuum = blkno;
1504 : }
1505 : }
1506 : else
1507 28288 : UnlockReleaseBuffer(buf);
1508 : }
1509 :
1510 250582 : vacrel->blkno = InvalidBlockNumber;
1511 250582 : if (BufferIsValid(vmbuffer))
1512 100358 : ReleaseBuffer(vmbuffer);
1513 :
1514 : /*
1515 : * Report that everything is now scanned. We never skip scanning the last
1516 : * block in the relation, so we can pass rel_pages here.
1517 : */
1518 250582 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_SCANNED,
1519 : rel_pages);
1520 :
1521 : /* now we can compute the new value for pg_class.reltuples */
1522 501164 : vacrel->new_live_tuples = vac_estimate_reltuples(vacrel->rel, rel_pages,
1523 : vacrel->scanned_pages,
1524 250582 : vacrel->live_tuples);
1525 :
1526 : /*
1527 : * Also compute the total number of surviving heap entries. In the
1528 : * (unlikely) scenario that new_live_tuples is -1, take it as zero.
1529 : */
1530 250582 : vacrel->new_rel_tuples =
1531 250582 : Max(vacrel->new_live_tuples, 0) + vacrel->recently_dead_tuples +
1532 250582 : vacrel->missed_dead_tuples;
1533 :
1534 250582 : read_stream_end(stream);
1535 :
1536 : /*
1537 : * Do index vacuuming (call each index's ambulkdelete routine), then do
1538 : * related heap vacuuming
1539 : */
1540 250582 : if (vacrel->dead_items_info->num_items > 0)
1541 1272 : lazy_vacuum(vacrel);
1542 :
1543 : /*
1544 : * Vacuum the remainder of the Free Space Map. We must do this whether or
1545 : * not there were indexes, and whether or not we bypassed index vacuuming.
1546 : * We can pass rel_pages here because we never skip scanning the last
1547 : * block of the relation.
1548 : */
1549 250582 : if (rel_pages > next_fsm_block_to_vacuum)
1550 100360 : FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum, rel_pages);
1551 :
1552 : /* report all blocks vacuumed */
1553 250582 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_VACUUMED, rel_pages);
1554 :
1555 : /* Do final index cleanup (call each index's amvacuumcleanup routine) */
1556 250582 : if (vacrel->nindexes > 0 && vacrel->do_index_cleanup)
1557 159612 : lazy_cleanup_all_indexes(vacrel);
1558 250582 : }
1559 :
1560 : /*
1561 : * heap_vac_scan_next_block() -- read stream callback to get the next block
1562 : * for vacuum to process
1563 : *
1564 : * Every time lazy_scan_heap() needs a new block to process during its first
1565 : * phase, it invokes read_stream_next_buffer() with a stream set up to call
1566 : * heap_vac_scan_next_block() to get the next block.
1567 : *
1568 : * heap_vac_scan_next_block() uses the visibility map, vacuum options, and
1569 : * various thresholds to skip blocks which do not need to be processed and
1570 : * returns the next block to process or InvalidBlockNumber if there are no
1571 : * remaining blocks.
1572 : *
1573 : * The visibility status of the next block to process and whether or not it
1574 : * was eager scanned is set in the per_buffer_data.
1575 : *
1576 : * callback_private_data contains a reference to the LVRelState, passed to the
1577 : * read stream API during stream setup. The LVRelState is an in/out parameter
1578 : * here (locally named `vacrel`). Vacuum options and information about the
1579 : * relation are read from it. vacrel->skippedallvis is set if we skip a block
1580 : * that's all-visible but not all-frozen (to ensure that we don't update
1581 : * relfrozenxid in that case). vacrel also holds information about the next
1582 : * unskippable block -- as bookkeeping for this function.
1583 : */
1584 : static BlockNumber
1585 1427260 : heap_vac_scan_next_block(ReadStream *stream,
1586 : void *callback_private_data,
1587 : void *per_buffer_data)
1588 : {
1589 : BlockNumber next_block;
1590 1427260 : LVRelState *vacrel = callback_private_data;
1591 1427260 : uint8 blk_info = 0;
1592 :
1593 : /* relies on InvalidBlockNumber + 1 overflowing to 0 on first call */
1594 1427260 : next_block = vacrel->current_block + 1;
1595 :
1596 : /* Have we reached the end of the relation? */
1597 1427260 : if (next_block >= vacrel->rel_pages)
1598 : {
1599 250582 : if (BufferIsValid(vacrel->next_unskippable_vmbuffer))
1600 : {
1601 97450 : ReleaseBuffer(vacrel->next_unskippable_vmbuffer);
1602 97450 : vacrel->next_unskippable_vmbuffer = InvalidBuffer;
1603 : }
1604 250582 : return InvalidBlockNumber;
1605 : }
1606 :
1607 : /*
1608 : * We must be in one of the three following states:
1609 : */
1610 1176678 : if (next_block > vacrel->next_unskippable_block ||
1611 404308 : vacrel->next_unskippable_block == InvalidBlockNumber)
1612 : {
1613 : /*
1614 : * 1. We have just processed an unskippable block (or we're at the
1615 : * beginning of the scan). Find the next unskippable block using the
1616 : * visibility map.
1617 : */
1618 : bool skipsallvis;
1619 :
1620 872730 : find_next_unskippable_block(vacrel, &skipsallvis);
1621 :
1622 : /*
1623 : * We now know the next block that we must process. It can be the
1624 : * next block after the one we just processed, or something further
1625 : * ahead. If it's further ahead, we can jump to it, but we choose to
1626 : * do so only if we can skip at least SKIP_PAGES_THRESHOLD consecutive
1627 : * pages. Since we're reading sequentially, the OS should be doing
1628 : * readahead for us, so there's no gain in skipping a page now and
1629 : * then. Skipping such a range might even discourage sequential
1630 : * detection.
1631 : *
1632 : * This test also enables more frequent relfrozenxid advancement
1633 : * during non-aggressive VACUUMs. If the range has any all-visible
1634 : * pages then skipping makes updating relfrozenxid unsafe, which is a
1635 : * real downside.
1636 : */
1637 872730 : if (vacrel->next_unskippable_block - next_block >= SKIP_PAGES_THRESHOLD)
1638 : {
1639 9000 : next_block = vacrel->next_unskippable_block;
1640 9000 : if (skipsallvis)
1641 58 : vacrel->skippedallvis = true;
1642 : }
1643 : }
1644 :
1645 : /* Now we must be in one of the two remaining states: */
1646 1176678 : if (next_block < vacrel->next_unskippable_block)
1647 : {
1648 : /*
1649 : * 2. We are processing a range of blocks that we could have skipped
1650 : * but chose not to. We know that they are all-visible in the VM,
1651 : * otherwise they would've been unskippable.
1652 : */
1653 303948 : vacrel->current_block = next_block;
1654 303948 : blk_info |= VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM;
1655 303948 : *((uint8 *) per_buffer_data) = blk_info;
1656 303948 : return vacrel->current_block;
1657 : }
1658 : else
1659 : {
1660 : /*
1661 : * 3. We reached the next unskippable block. Process it. On next
1662 : * iteration, we will be back in state 1.
1663 : */
1664 : Assert(next_block == vacrel->next_unskippable_block);
1665 :
1666 872730 : vacrel->current_block = next_block;
1667 872730 : if (vacrel->next_unskippable_allvis)
1668 91496 : blk_info |= VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM;
1669 872730 : if (vacrel->next_unskippable_eager_scanned)
1670 0 : blk_info |= VAC_BLK_WAS_EAGER_SCANNED;
1671 872730 : *((uint8 *) per_buffer_data) = blk_info;
1672 872730 : return vacrel->current_block;
1673 : }
1674 : }
1675 :
1676 : /*
1677 : * Find the next unskippable block in a vacuum scan using the visibility map.
1678 : * The next unskippable block and its visibility information is updated in
1679 : * vacrel.
1680 : *
1681 : * Note: our opinion of which blocks can be skipped can go stale immediately.
1682 : * It's okay if caller "misses" a page whose all-visible or all-frozen marking
1683 : * was concurrently cleared, though. All that matters is that caller scan all
1684 : * pages whose tuples might contain XIDs < OldestXmin, or MXIDs < OldestMxact.
1685 : * (Actually, non-aggressive VACUUMs can choose to skip all-visible pages with
1686 : * older XIDs/MXIDs. The *skippedallvis flag will be set here when the choice
1687 : * to skip such a range is actually made, making everything safe.)
1688 : */
1689 : static void
1690 872730 : find_next_unskippable_block(LVRelState *vacrel, bool *skipsallvis)
1691 : {
1692 872730 : BlockNumber rel_pages = vacrel->rel_pages;
1693 872730 : BlockNumber next_unskippable_block = vacrel->next_unskippable_block + 1;
1694 872730 : Buffer next_unskippable_vmbuffer = vacrel->next_unskippable_vmbuffer;
1695 872730 : bool next_unskippable_eager_scanned = false;
1696 : bool next_unskippable_allvis;
1697 :
1698 872730 : *skipsallvis = false;
1699 :
1700 967068 : for (;; next_unskippable_block++)
1701 967068 : {
1702 1839798 : uint8 mapbits = visibilitymap_get_status(vacrel->rel,
1703 : next_unskippable_block,
1704 : &next_unskippable_vmbuffer);
1705 :
1706 1839798 : next_unskippable_allvis = (mapbits & VISIBILITYMAP_ALL_VISIBLE) != 0;
1707 :
1708 : /*
1709 : * At the start of each eager scan region, normal vacuums with eager
1710 : * scanning enabled reset the failure counter, allowing vacuum to
1711 : * resume eager scanning if it had been suspended in the previous
1712 : * region.
1713 : */
1714 1839798 : if (next_unskippable_block >= vacrel->next_eager_scan_region_start)
1715 : {
1716 0 : vacrel->eager_scan_remaining_fails =
1717 0 : vacrel->eager_scan_max_fails_per_region;
1718 0 : vacrel->next_eager_scan_region_start += EAGER_SCAN_REGION_SIZE;
1719 : }
1720 :
1721 : /*
1722 : * A block is unskippable if it is not all visible according to the
1723 : * visibility map.
1724 : */
1725 1839798 : if (!next_unskippable_allvis)
1726 : {
1727 : Assert((mapbits & VISIBILITYMAP_ALL_FROZEN) == 0);
1728 781234 : break;
1729 : }
1730 :
1731 : /*
1732 : * Caller must scan the last page to determine whether it has tuples
1733 : * (caller must have the opportunity to set vacrel->nonempty_pages).
1734 : * This rule avoids having lazy_truncate_heap() take access-exclusive
1735 : * lock on rel to attempt a truncation that fails anyway, just because
1736 : * there are tuples on the last page (it is likely that there will be
1737 : * tuples on other nearby pages as well, but those can be skipped).
1738 : *
1739 : * Implement this by always treating the last block as unsafe to skip.
1740 : */
1741 1058564 : if (next_unskippable_block == rel_pages - 1)
1742 90684 : break;
1743 :
1744 : /* DISABLE_PAGE_SKIPPING makes all skipping unsafe */
1745 967880 : if (!vacrel->skipwithvm)
1746 812 : break;
1747 :
1748 : /*
1749 : * All-frozen pages cannot contain XIDs < OldestXmin (XIDs that aren't
1750 : * already frozen by now), so this page can be skipped.
1751 : */
1752 967068 : if ((mapbits & VISIBILITYMAP_ALL_FROZEN) != 0)
1753 961796 : continue;
1754 :
1755 : /*
1756 : * Aggressive vacuums cannot skip any all-visible pages that are not
1757 : * also all-frozen.
1758 : */
1759 5272 : if (vacrel->aggressive)
1760 0 : break;
1761 :
1762 : /*
1763 : * Normal vacuums with eager scanning enabled only skip all-visible
1764 : * but not all-frozen pages if they have hit the failure limit for the
1765 : * current eager scan region.
1766 : */
1767 5272 : if (vacrel->eager_scan_remaining_fails > 0)
1768 : {
1769 0 : next_unskippable_eager_scanned = true;
1770 0 : break;
1771 : }
1772 :
1773 : /*
1774 : * All-visible blocks are safe to skip in a normal vacuum. But
1775 : * remember that the final range contains such a block for later.
1776 : */
1777 5272 : *skipsallvis = true;
1778 : }
1779 :
1780 : /* write the local variables back to vacrel */
1781 872730 : vacrel->next_unskippable_block = next_unskippable_block;
1782 872730 : vacrel->next_unskippable_allvis = next_unskippable_allvis;
1783 872730 : vacrel->next_unskippable_eager_scanned = next_unskippable_eager_scanned;
1784 872730 : vacrel->next_unskippable_vmbuffer = next_unskippable_vmbuffer;
1785 872730 : }
1786 :
1787 : /*
1788 : * lazy_scan_new_or_empty() -- lazy_scan_heap() new/empty page handling.
1789 : *
1790 : * Must call here to handle both new and empty pages before calling
1791 : * lazy_scan_prune or lazy_scan_noprune, since they're not prepared to deal
1792 : * with new or empty pages.
1793 : *
1794 : * It's necessary to consider new pages as a special case, since the rules for
1795 : * maintaining the visibility map and FSM with empty pages are a little
1796 : * different (though new pages can be truncated away during rel truncation).
1797 : *
1798 : * Empty pages are not really a special case -- they're just heap pages that
1799 : * have no allocated tuples (including even LP_UNUSED items). You might
1800 : * wonder why we need to handle them here all the same. It's only necessary
1801 : * because of a corner-case involving a hard crash during heap relation
1802 : * extension. If we ever make relation-extension crash safe, then it should
1803 : * no longer be necessary to deal with empty pages here (or new pages, for
1804 : * that matter).
1805 : *
1806 : * Caller must hold at least a shared lock. We might need to escalate the
1807 : * lock in that case, so the type of lock caller holds needs to be specified
1808 : * using 'sharelock' argument.
1809 : *
1810 : * Returns false in common case where caller should go on to call
1811 : * lazy_scan_prune (or lazy_scan_noprune). Otherwise returns true, indicating
1812 : * that lazy_scan_heap is done processing the page, releasing lock on caller's
1813 : * behalf.
1814 : *
1815 : * No vm_page_frozen output parameter (like that passed to lazy_scan_prune())
1816 : * is passed here because neither empty nor new pages can be eagerly frozen.
1817 : * New pages are never frozen. Empty pages are always set frozen in the VM at
1818 : * the same time that they are set all-visible, and we don't eagerly scan
1819 : * frozen pages.
1820 : */
1821 : static bool
1822 1176678 : lazy_scan_new_or_empty(LVRelState *vacrel, Buffer buf, BlockNumber blkno,
1823 : Page page, bool sharelock, Buffer vmbuffer)
1824 : {
1825 : Size freespace;
1826 :
1827 1176678 : if (PageIsNew(page))
1828 : {
1829 : /*
1830 : * All-zeroes pages can be left over if either a backend extends the
1831 : * relation by a single page, but crashes before the newly initialized
1832 : * page has been written out, or when bulk-extending the relation
1833 : * (which creates a number of empty pages at the tail end of the
1834 : * relation), and then enters them into the FSM.
1835 : *
1836 : * Note we do not enter the page into the visibilitymap. That has the
1837 : * downside that we repeatedly visit this page in subsequent vacuums,
1838 : * but otherwise we'll never discover the space on a promoted standby.
1839 : * The harm of repeated checking ought to normally not be too bad. The
1840 : * space usually should be used at some point, otherwise there
1841 : * wouldn't be any regular vacuums.
1842 : *
1843 : * Make sure these pages are in the FSM, to ensure they can be reused.
1844 : * Do that by testing if there's any space recorded for the page. If
1845 : * not, enter it. We do so after releasing the lock on the heap page,
1846 : * the FSM is approximate, after all.
1847 : */
1848 1170 : UnlockReleaseBuffer(buf);
1849 :
1850 1170 : if (GetRecordedFreeSpace(vacrel->rel, blkno) == 0)
1851 : {
1852 858 : freespace = BLCKSZ - SizeOfPageHeaderData;
1853 :
1854 858 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
1855 : }
1856 :
1857 1170 : return true;
1858 : }
1859 :
1860 1175508 : if (PageIsEmpty(page))
1861 : {
1862 : /*
1863 : * It seems likely that caller will always be able to get a cleanup
1864 : * lock on an empty page. But don't take any chances -- escalate to
1865 : * an exclusive lock (still don't need a cleanup lock, though).
1866 : */
1867 54 : if (sharelock)
1868 : {
1869 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
1870 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
1871 :
1872 0 : if (!PageIsEmpty(page))
1873 : {
1874 : /* page isn't new or empty -- keep lock and pin for now */
1875 0 : return false;
1876 : }
1877 : }
1878 : else
1879 : {
1880 : /* Already have a full cleanup lock (which is more than enough) */
1881 : }
1882 :
1883 : /*
1884 : * Unlike new pages, empty pages are always set all-visible and
1885 : * all-frozen.
1886 : */
1887 54 : if (!PageIsAllVisible(page))
1888 : {
1889 0 : START_CRIT_SECTION();
1890 :
1891 : /* mark buffer dirty before writing a WAL record */
1892 0 : MarkBufferDirty(buf);
1893 :
1894 : /*
1895 : * It's possible that another backend has extended the heap,
1896 : * initialized the page, and then failed to WAL-log the page due
1897 : * to an ERROR. Since heap extension is not WAL-logged, recovery
1898 : * might try to replay our record setting the page all-visible and
1899 : * find that the page isn't initialized, which will cause a PANIC.
1900 : * To prevent that, check whether the page has been previously
1901 : * WAL-logged, and if not, do that now.
1902 : */
1903 0 : if (RelationNeedsWAL(vacrel->rel) &&
1904 0 : !XLogRecPtrIsValid(PageGetLSN(page)))
1905 0 : log_newpage_buffer(buf, true);
1906 :
1907 0 : PageSetAllVisible(page);
1908 0 : visibilitymap_set(vacrel->rel, blkno, buf,
1909 : InvalidXLogRecPtr,
1910 : vmbuffer, InvalidTransactionId,
1911 : VISIBILITYMAP_ALL_VISIBLE |
1912 : VISIBILITYMAP_ALL_FROZEN);
1913 0 : END_CRIT_SECTION();
1914 :
1915 : /* Count the newly all-frozen pages for logging */
1916 0 : vacrel->vm_new_visible_pages++;
1917 0 : vacrel->vm_new_visible_frozen_pages++;
1918 : }
1919 :
1920 54 : freespace = PageGetHeapFreeSpace(page);
1921 54 : UnlockReleaseBuffer(buf);
1922 54 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
1923 54 : return true;
1924 : }
1925 :
1926 : /* page isn't new or empty -- keep lock and pin */
1927 1175454 : return false;
1928 : }
1929 :
1930 : /* qsort comparator for sorting OffsetNumbers */
1931 : static int
1932 6761292 : cmpOffsetNumbers(const void *a, const void *b)
1933 : {
1934 6761292 : return pg_cmp_u16(*(const OffsetNumber *) a, *(const OffsetNumber *) b);
1935 : }
1936 :
1937 : /*
1938 : * lazy_scan_prune() -- lazy_scan_heap() pruning and freezing.
1939 : *
1940 : * Caller must hold pin and buffer cleanup lock on the buffer.
1941 : *
1942 : * vmbuffer is the buffer containing the VM block with visibility information
1943 : * for the heap block, blkno. all_visible_according_to_vm is the saved
1944 : * visibility status of the heap block looked up earlier by the caller. We
1945 : * won't rely entirely on this status, as it may be out of date.
1946 : *
1947 : * *has_lpdead_items is set to true or false depending on whether, upon return
1948 : * from this function, any LP_DEAD items are still present on the page.
1949 : *
1950 : * *vm_page_frozen is set to true if the page is newly set all-frozen in the
1951 : * VM. The caller currently only uses this for determining whether an eagerly
1952 : * scanned page was successfully set all-frozen.
1953 : *
1954 : * Returns the number of tuples deleted from the page during HOT pruning.
1955 : */
1956 : static int
1957 1175218 : lazy_scan_prune(LVRelState *vacrel,
1958 : Buffer buf,
1959 : BlockNumber blkno,
1960 : Page page,
1961 : Buffer vmbuffer,
1962 : bool all_visible_according_to_vm,
1963 : bool *has_lpdead_items,
1964 : bool *vm_page_frozen)
1965 : {
1966 1175218 : Relation rel = vacrel->rel;
1967 : PruneFreezeResult presult;
1968 1175218 : PruneFreezeParams params = {
1969 : .relation = rel,
1970 : .buffer = buf,
1971 : .reason = PRUNE_VACUUM_SCAN,
1972 : .options = HEAP_PAGE_PRUNE_FREEZE,
1973 1175218 : .vistest = vacrel->vistest,
1974 1175218 : .cutoffs = &vacrel->cutoffs,
1975 : };
1976 :
1977 : Assert(BufferGetBlockNumber(buf) == blkno);
1978 :
1979 : /*
1980 : * Prune all HOT-update chains and potentially freeze tuples on this page.
1981 : *
1982 : * If the relation has no indexes, we can immediately mark would-be dead
1983 : * items LP_UNUSED.
1984 : *
1985 : * The number of tuples removed from the page is returned in
1986 : * presult.ndeleted. It should not be confused with presult.lpdead_items;
1987 : * presult.lpdead_items's final value can be thought of as the number of
1988 : * tuples that were deleted from indexes.
1989 : *
1990 : * We will update the VM after collecting LP_DEAD items and freezing
1991 : * tuples. Pruning will have determined whether or not the page is
1992 : * all-visible.
1993 : */
1994 1175218 : if (vacrel->nindexes == 0)
1995 35928 : params.options |= HEAP_PAGE_PRUNE_MARK_UNUSED_NOW;
1996 :
1997 1175218 : heap_page_prune_and_freeze(¶ms,
1998 : &presult,
1999 : &vacrel->offnum,
2000 : &vacrel->NewRelfrozenXid, &vacrel->NewRelminMxid);
2001 :
2002 : Assert(MultiXactIdIsValid(vacrel->NewRelminMxid));
2003 : Assert(TransactionIdIsValid(vacrel->NewRelfrozenXid));
2004 :
2005 1175218 : if (presult.nfrozen > 0)
2006 : {
2007 : /*
2008 : * We don't increment the new_frozen_tuple_pages instrumentation
2009 : * counter when nfrozen == 0, since it only counts pages with newly
2010 : * frozen tuples (don't confuse that with pages newly set all-frozen
2011 : * in VM).
2012 : */
2013 45806 : vacrel->new_frozen_tuple_pages++;
2014 : }
2015 :
2016 : /*
2017 : * VACUUM will call heap_page_is_all_visible() during the second pass over
2018 : * the heap to determine all_visible and all_frozen for the page -- this
2019 : * is a specialized version of the logic from this function. Now that
2020 : * we've finished pruning and freezing, make sure that we're in total
2021 : * agreement with heap_page_is_all_visible() using an assertion.
2022 : */
2023 : #ifdef USE_ASSERT_CHECKING
2024 : if (presult.all_visible)
2025 : {
2026 : TransactionId debug_cutoff;
2027 : bool debug_all_frozen;
2028 :
2029 : Assert(presult.lpdead_items == 0);
2030 :
2031 : if (!heap_page_is_all_visible(vacrel->rel, buf,
2032 : vacrel->cutoffs.OldestXmin, &debug_all_frozen,
2033 : &debug_cutoff, &vacrel->offnum))
2034 : Assert(false);
2035 :
2036 : Assert(presult.all_frozen == debug_all_frozen);
2037 :
2038 : Assert(!TransactionIdIsValid(debug_cutoff) ||
2039 : debug_cutoff == presult.vm_conflict_horizon);
2040 : }
2041 : #endif
2042 :
2043 : /*
2044 : * Now save details of the LP_DEAD items from the page in vacrel
2045 : */
2046 1175218 : if (presult.lpdead_items > 0)
2047 : {
2048 32860 : vacrel->lpdead_item_pages++;
2049 :
2050 : /*
2051 : * deadoffsets are collected incrementally in
2052 : * heap_page_prune_and_freeze() as each dead line pointer is recorded,
2053 : * with an indeterminate order, but dead_items_add requires them to be
2054 : * sorted.
2055 : */
2056 32860 : qsort(presult.deadoffsets, presult.lpdead_items, sizeof(OffsetNumber),
2057 : cmpOffsetNumbers);
2058 :
2059 32860 : dead_items_add(vacrel, blkno, presult.deadoffsets, presult.lpdead_items);
2060 : }
2061 :
2062 : /* Finally, add page-local counts to whole-VACUUM counts */
2063 1175218 : vacrel->tuples_deleted += presult.ndeleted;
2064 1175218 : vacrel->tuples_frozen += presult.nfrozen;
2065 1175218 : vacrel->lpdead_items += presult.lpdead_items;
2066 1175218 : vacrel->live_tuples += presult.live_tuples;
2067 1175218 : vacrel->recently_dead_tuples += presult.recently_dead_tuples;
2068 :
2069 : /* Can't truncate this page */
2070 1175218 : if (presult.hastup)
2071 1158600 : vacrel->nonempty_pages = blkno + 1;
2072 :
2073 : /* Did we find LP_DEAD items? */
2074 1175218 : *has_lpdead_items = (presult.lpdead_items > 0);
2075 :
2076 : Assert(!presult.all_visible || !(*has_lpdead_items));
2077 : Assert(!presult.all_frozen || presult.all_visible);
2078 :
2079 : /*
2080 : * Handle setting visibility map bit based on information from the VM (as
2081 : * of last heap_vac_scan_next_block() call), and from all_visible and
2082 : * all_frozen variables
2083 : */
2084 1175218 : if (!all_visible_according_to_vm && presult.all_visible)
2085 71280 : {
2086 : uint8 old_vmbits;
2087 71280 : uint8 flags = VISIBILITYMAP_ALL_VISIBLE;
2088 :
2089 71280 : if (presult.all_frozen)
2090 : {
2091 : Assert(!TransactionIdIsValid(presult.vm_conflict_horizon));
2092 53252 : flags |= VISIBILITYMAP_ALL_FROZEN;
2093 : }
2094 :
2095 : /*
2096 : * It should never be the case that the visibility map page is set
2097 : * while the page-level bit is clear, but the reverse is allowed (if
2098 : * checksums are not enabled). Regardless, set both bits so that we
2099 : * get back in sync.
2100 : *
2101 : * NB: If the heap page is all-visible but the VM bit is not set, we
2102 : * don't need to dirty the heap page. However, if checksums are
2103 : * enabled, we do need to make sure that the heap page is dirtied
2104 : * before passing it to visibilitymap_set(), because it may be logged.
2105 : * Given that this situation should only happen in rare cases after a
2106 : * crash, it is not worth optimizing.
2107 : */
2108 71280 : PageSetAllVisible(page);
2109 71280 : MarkBufferDirty(buf);
2110 71280 : old_vmbits = visibilitymap_set(vacrel->rel, blkno, buf,
2111 : InvalidXLogRecPtr,
2112 : vmbuffer, presult.vm_conflict_horizon,
2113 : flags);
2114 :
2115 : /*
2116 : * If the page wasn't already set all-visible and/or all-frozen in the
2117 : * VM, count it as newly set for logging.
2118 : */
2119 71280 : if ((old_vmbits & VISIBILITYMAP_ALL_VISIBLE) == 0)
2120 : {
2121 71280 : vacrel->vm_new_visible_pages++;
2122 71280 : if (presult.all_frozen)
2123 : {
2124 53252 : vacrel->vm_new_visible_frozen_pages++;
2125 53252 : *vm_page_frozen = true;
2126 : }
2127 : }
2128 0 : else if ((old_vmbits & VISIBILITYMAP_ALL_FROZEN) == 0 &&
2129 0 : presult.all_frozen)
2130 : {
2131 0 : vacrel->vm_new_frozen_pages++;
2132 0 : *vm_page_frozen = true;
2133 : }
2134 : }
2135 :
2136 : /*
2137 : * As of PostgreSQL 9.2, the visibility map bit should never be set if the
2138 : * page-level bit is clear. However, it's possible that the bit got
2139 : * cleared after heap_vac_scan_next_block() was called, so we must recheck
2140 : * with buffer lock before concluding that the VM is corrupt.
2141 : */
2142 1103938 : else if (all_visible_according_to_vm && !PageIsAllVisible(page) &&
2143 0 : visibilitymap_get_status(vacrel->rel, blkno, &vmbuffer) != 0)
2144 : {
2145 0 : ereport(WARNING,
2146 : (errcode(ERRCODE_DATA_CORRUPTED),
2147 : errmsg("page is not marked all-visible but visibility map bit is set in relation \"%s\" page %u",
2148 : vacrel->relname, blkno)));
2149 :
2150 0 : visibilitymap_clear(vacrel->rel, blkno, vmbuffer,
2151 : VISIBILITYMAP_VALID_BITS);
2152 : }
2153 :
2154 : /*
2155 : * It's possible for the value returned by
2156 : * GetOldestNonRemovableTransactionId() to move backwards, so it's not
2157 : * wrong for us to see tuples that appear to not be visible to everyone
2158 : * yet, while PD_ALL_VISIBLE is already set. The real safe xmin value
2159 : * never moves backwards, but GetOldestNonRemovableTransactionId() is
2160 : * conservative and sometimes returns a value that's unnecessarily small,
2161 : * so if we see that contradiction it just means that the tuples that we
2162 : * think are not visible to everyone yet actually are, and the
2163 : * PD_ALL_VISIBLE flag is correct.
2164 : *
2165 : * There should never be LP_DEAD items on a page with PD_ALL_VISIBLE set,
2166 : * however.
2167 : */
2168 1103938 : else if (presult.lpdead_items > 0 && PageIsAllVisible(page))
2169 : {
2170 0 : ereport(WARNING,
2171 : (errcode(ERRCODE_DATA_CORRUPTED),
2172 : errmsg("page containing LP_DEAD items is marked as all-visible in relation \"%s\" page %u",
2173 : vacrel->relname, blkno)));
2174 :
2175 0 : PageClearAllVisible(page);
2176 0 : MarkBufferDirty(buf);
2177 0 : visibilitymap_clear(vacrel->rel, blkno, vmbuffer,
2178 : VISIBILITYMAP_VALID_BITS);
2179 : }
2180 :
2181 : /*
2182 : * If the all-visible page is all-frozen but not marked as such yet, mark
2183 : * it as all-frozen.
2184 : */
2185 1103938 : else if (all_visible_according_to_vm && presult.all_frozen &&
2186 394028 : !VM_ALL_FROZEN(vacrel->rel, blkno, &vmbuffer))
2187 : {
2188 : uint8 old_vmbits;
2189 :
2190 : /*
2191 : * Avoid relying on all_visible_according_to_vm as a proxy for the
2192 : * page-level PD_ALL_VISIBLE bit being set, since it might have become
2193 : * stale -- even when all_visible is set
2194 : */
2195 42 : if (!PageIsAllVisible(page))
2196 : {
2197 0 : PageSetAllVisible(page);
2198 0 : MarkBufferDirty(buf);
2199 : }
2200 :
2201 : /*
2202 : * Set the page all-frozen (and all-visible) in the VM.
2203 : *
2204 : * We can pass InvalidTransactionId as our cutoff_xid, since a
2205 : * snapshotConflictHorizon sufficient to make everything safe for REDO
2206 : * was logged when the page's tuples were frozen.
2207 : */
2208 : Assert(!TransactionIdIsValid(presult.vm_conflict_horizon));
2209 42 : old_vmbits = visibilitymap_set(vacrel->rel, blkno, buf,
2210 : InvalidXLogRecPtr,
2211 : vmbuffer, InvalidTransactionId,
2212 : VISIBILITYMAP_ALL_VISIBLE |
2213 : VISIBILITYMAP_ALL_FROZEN);
2214 :
2215 : /*
2216 : * The page was likely already set all-visible in the VM. However,
2217 : * there is a small chance that it was modified sometime between
2218 : * setting all_visible_according_to_vm and checking the visibility
2219 : * during pruning. Check the return value of old_vmbits anyway to
2220 : * ensure the visibility map counters used for logging are accurate.
2221 : */
2222 42 : if ((old_vmbits & VISIBILITYMAP_ALL_VISIBLE) == 0)
2223 : {
2224 0 : vacrel->vm_new_visible_pages++;
2225 0 : vacrel->vm_new_visible_frozen_pages++;
2226 0 : *vm_page_frozen = true;
2227 : }
2228 :
2229 : /*
2230 : * We already checked that the page was not set all-frozen in the VM
2231 : * above, so we don't need to test the value of old_vmbits.
2232 : */
2233 : else
2234 : {
2235 42 : vacrel->vm_new_frozen_pages++;
2236 42 : *vm_page_frozen = true;
2237 : }
2238 : }
2239 :
2240 1175218 : return presult.ndeleted;
2241 : }
2242 :
2243 : /*
2244 : * lazy_scan_noprune() -- lazy_scan_prune() without pruning or freezing
2245 : *
2246 : * Caller need only hold a pin and share lock on the buffer, unlike
2247 : * lazy_scan_prune, which requires a full cleanup lock. While pruning isn't
2248 : * performed here, it's quite possible that an earlier opportunistic pruning
2249 : * operation left LP_DEAD items behind. We'll at least collect any such items
2250 : * in dead_items for removal from indexes.
2251 : *
2252 : * For aggressive VACUUM callers, we may return false to indicate that a full
2253 : * cleanup lock is required for processing by lazy_scan_prune. This is only
2254 : * necessary when the aggressive VACUUM needs to freeze some tuple XIDs from
2255 : * one or more tuples on the page. We always return true for non-aggressive
2256 : * callers.
2257 : *
2258 : * If this function returns true, *has_lpdead_items gets set to true or false
2259 : * depending on whether, upon return from this function, any LP_DEAD items are
2260 : * present on the page. If this function returns false, *has_lpdead_items
2261 : * is not updated.
2262 : */
2263 : static bool
2264 362 : lazy_scan_noprune(LVRelState *vacrel,
2265 : Buffer buf,
2266 : BlockNumber blkno,
2267 : Page page,
2268 : bool *has_lpdead_items)
2269 : {
2270 : OffsetNumber offnum,
2271 : maxoff;
2272 : int lpdead_items,
2273 : live_tuples,
2274 : recently_dead_tuples,
2275 : missed_dead_tuples;
2276 : bool hastup;
2277 : HeapTupleHeader tupleheader;
2278 362 : TransactionId NoFreezePageRelfrozenXid = vacrel->NewRelfrozenXid;
2279 362 : MultiXactId NoFreezePageRelminMxid = vacrel->NewRelminMxid;
2280 : OffsetNumber deadoffsets[MaxHeapTuplesPerPage];
2281 :
2282 : Assert(BufferGetBlockNumber(buf) == blkno);
2283 :
2284 362 : hastup = false; /* for now */
2285 :
2286 362 : lpdead_items = 0;
2287 362 : live_tuples = 0;
2288 362 : recently_dead_tuples = 0;
2289 362 : missed_dead_tuples = 0;
2290 :
2291 362 : maxoff = PageGetMaxOffsetNumber(page);
2292 362 : for (offnum = FirstOffsetNumber;
2293 10634 : offnum <= maxoff;
2294 10272 : offnum = OffsetNumberNext(offnum))
2295 : {
2296 : ItemId itemid;
2297 : HeapTupleData tuple;
2298 :
2299 10398 : vacrel->offnum = offnum;
2300 10398 : itemid = PageGetItemId(page, offnum);
2301 :
2302 10398 : if (!ItemIdIsUsed(itemid))
2303 2666 : continue;
2304 :
2305 8968 : if (ItemIdIsRedirected(itemid))
2306 : {
2307 1010 : hastup = true;
2308 1010 : continue;
2309 : }
2310 :
2311 7958 : if (ItemIdIsDead(itemid))
2312 : {
2313 : /*
2314 : * Deliberately don't set hastup=true here. See same point in
2315 : * lazy_scan_prune for an explanation.
2316 : */
2317 226 : deadoffsets[lpdead_items++] = offnum;
2318 226 : continue;
2319 : }
2320 :
2321 7732 : hastup = true; /* page prevents rel truncation */
2322 7732 : tupleheader = (HeapTupleHeader) PageGetItem(page, itemid);
2323 7732 : if (heap_tuple_should_freeze(tupleheader, &vacrel->cutoffs,
2324 : &NoFreezePageRelfrozenXid,
2325 : &NoFreezePageRelminMxid))
2326 : {
2327 : /* Tuple with XID < FreezeLimit (or MXID < MultiXactCutoff) */
2328 254 : if (vacrel->aggressive)
2329 : {
2330 : /*
2331 : * Aggressive VACUUMs must always be able to advance rel's
2332 : * relfrozenxid to a value >= FreezeLimit (and be able to
2333 : * advance rel's relminmxid to a value >= MultiXactCutoff).
2334 : * The ongoing aggressive VACUUM won't be able to do that
2335 : * unless it can freeze an XID (or MXID) from this tuple now.
2336 : *
2337 : * The only safe option is to have caller perform processing
2338 : * of this page using lazy_scan_prune. Caller might have to
2339 : * wait a while for a cleanup lock, but it can't be helped.
2340 : */
2341 126 : vacrel->offnum = InvalidOffsetNumber;
2342 126 : return false;
2343 : }
2344 :
2345 : /*
2346 : * Non-aggressive VACUUMs are under no obligation to advance
2347 : * relfrozenxid (even by one XID). We can be much laxer here.
2348 : *
2349 : * Currently we always just accept an older final relfrozenxid
2350 : * and/or relminmxid value. We never make caller wait or work a
2351 : * little harder, even when it likely makes sense to do so.
2352 : */
2353 : }
2354 :
2355 7606 : ItemPointerSet(&(tuple.t_self), blkno, offnum);
2356 7606 : tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid);
2357 7606 : tuple.t_len = ItemIdGetLength(itemid);
2358 7606 : tuple.t_tableOid = RelationGetRelid(vacrel->rel);
2359 :
2360 7606 : switch (HeapTupleSatisfiesVacuum(&tuple, vacrel->cutoffs.OldestXmin,
2361 : buf))
2362 : {
2363 7598 : case HEAPTUPLE_DELETE_IN_PROGRESS:
2364 : case HEAPTUPLE_LIVE:
2365 :
2366 : /*
2367 : * Count both cases as live, just like lazy_scan_prune
2368 : */
2369 7598 : live_tuples++;
2370 :
2371 7598 : break;
2372 4 : case HEAPTUPLE_DEAD:
2373 :
2374 : /*
2375 : * There is some useful work for pruning to do, that won't be
2376 : * done due to failure to get a cleanup lock.
2377 : */
2378 4 : missed_dead_tuples++;
2379 4 : break;
2380 4 : case HEAPTUPLE_RECENTLY_DEAD:
2381 :
2382 : /*
2383 : * Count in recently_dead_tuples, just like lazy_scan_prune
2384 : */
2385 4 : recently_dead_tuples++;
2386 4 : break;
2387 0 : case HEAPTUPLE_INSERT_IN_PROGRESS:
2388 :
2389 : /*
2390 : * Do not count these rows as live, just like lazy_scan_prune
2391 : */
2392 0 : break;
2393 0 : default:
2394 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
2395 : break;
2396 : }
2397 : }
2398 :
2399 236 : vacrel->offnum = InvalidOffsetNumber;
2400 :
2401 : /*
2402 : * By here we know for sure that caller can put off freezing and pruning
2403 : * this particular page until the next VACUUM. Remember its details now.
2404 : * (lazy_scan_prune expects a clean slate, so we have to do this last.)
2405 : */
2406 236 : vacrel->NewRelfrozenXid = NoFreezePageRelfrozenXid;
2407 236 : vacrel->NewRelminMxid = NoFreezePageRelminMxid;
2408 :
2409 : /* Save any LP_DEAD items found on the page in dead_items */
2410 236 : if (vacrel->nindexes == 0)
2411 : {
2412 : /* Using one-pass strategy (since table has no indexes) */
2413 0 : if (lpdead_items > 0)
2414 : {
2415 : /*
2416 : * Perfunctory handling for the corner case where a single pass
2417 : * strategy VACUUM cannot get a cleanup lock, and it turns out
2418 : * that there is one or more LP_DEAD items: just count the LP_DEAD
2419 : * items as missed_dead_tuples instead. (This is a bit dishonest,
2420 : * but it beats having to maintain specialized heap vacuuming code
2421 : * forever, for vanishingly little benefit.)
2422 : */
2423 0 : hastup = true;
2424 0 : missed_dead_tuples += lpdead_items;
2425 : }
2426 : }
2427 236 : else if (lpdead_items > 0)
2428 : {
2429 : /*
2430 : * Page has LP_DEAD items, and so any references/TIDs that remain in
2431 : * indexes will be deleted during index vacuuming (and then marked
2432 : * LP_UNUSED in the heap)
2433 : */
2434 8 : vacrel->lpdead_item_pages++;
2435 :
2436 8 : dead_items_add(vacrel, blkno, deadoffsets, lpdead_items);
2437 :
2438 8 : vacrel->lpdead_items += lpdead_items;
2439 : }
2440 :
2441 : /*
2442 : * Finally, add relevant page-local counts to whole-VACUUM counts
2443 : */
2444 236 : vacrel->live_tuples += live_tuples;
2445 236 : vacrel->recently_dead_tuples += recently_dead_tuples;
2446 236 : vacrel->missed_dead_tuples += missed_dead_tuples;
2447 236 : if (missed_dead_tuples > 0)
2448 4 : vacrel->missed_dead_pages++;
2449 :
2450 : /* Can't truncate this page */
2451 236 : if (hastup)
2452 234 : vacrel->nonempty_pages = blkno + 1;
2453 :
2454 : /* Did we find LP_DEAD items? */
2455 236 : *has_lpdead_items = (lpdead_items > 0);
2456 :
2457 : /* Caller won't need to call lazy_scan_prune with same page */
2458 236 : return true;
2459 : }
2460 :
2461 : /*
2462 : * Main entry point for index vacuuming and heap vacuuming.
2463 : *
2464 : * Removes items collected in dead_items from table's indexes, then marks the
2465 : * same items LP_UNUSED in the heap. See the comments above lazy_scan_heap
2466 : * for full details.
2467 : *
2468 : * Also empties dead_items, freeing up space for later TIDs.
2469 : *
2470 : * We may choose to bypass index vacuuming at this point, though only when the
2471 : * ongoing VACUUM operation will definitely only have one index scan/round of
2472 : * index vacuuming.
2473 : */
2474 : static void
2475 1286 : lazy_vacuum(LVRelState *vacrel)
2476 : {
2477 : bool bypass;
2478 :
2479 : /* Should not end up here with no indexes */
2480 : Assert(vacrel->nindexes > 0);
2481 : Assert(vacrel->lpdead_item_pages > 0);
2482 :
2483 1286 : if (!vacrel->do_index_vacuuming)
2484 : {
2485 : Assert(!vacrel->do_index_cleanup);
2486 16 : dead_items_reset(vacrel);
2487 16 : return;
2488 : }
2489 :
2490 : /*
2491 : * Consider bypassing index vacuuming (and heap vacuuming) entirely.
2492 : *
2493 : * We currently only do this in cases where the number of LP_DEAD items
2494 : * for the entire VACUUM operation is close to zero. This avoids sharp
2495 : * discontinuities in the duration and overhead of successive VACUUM
2496 : * operations that run against the same table with a fixed workload.
2497 : * Ideally, successive VACUUM operations will behave as if there are
2498 : * exactly zero LP_DEAD items in cases where there are close to zero.
2499 : *
2500 : * This is likely to be helpful with a table that is continually affected
2501 : * by UPDATEs that can mostly apply the HOT optimization, but occasionally
2502 : * have small aberrations that lead to just a few heap pages retaining
2503 : * only one or two LP_DEAD items. This is pretty common; even when the
2504 : * DBA goes out of their way to make UPDATEs use HOT, it is practically
2505 : * impossible to predict whether HOT will be applied in 100% of cases.
2506 : * It's far easier to ensure that 99%+ of all UPDATEs against a table use
2507 : * HOT through careful tuning.
2508 : */
2509 1270 : bypass = false;
2510 1270 : if (vacrel->consider_bypass_optimization && vacrel->rel_pages > 0)
2511 : {
2512 : BlockNumber threshold;
2513 :
2514 : Assert(vacrel->num_index_scans == 0);
2515 : Assert(vacrel->lpdead_items == vacrel->dead_items_info->num_items);
2516 : Assert(vacrel->do_index_vacuuming);
2517 : Assert(vacrel->do_index_cleanup);
2518 :
2519 : /*
2520 : * This crossover point at which we'll start to do index vacuuming is
2521 : * expressed as a percentage of the total number of heap pages in the
2522 : * table that are known to have at least one LP_DEAD item. This is
2523 : * much more important than the total number of LP_DEAD items, since
2524 : * it's a proxy for the number of heap pages whose visibility map bits
2525 : * cannot be set on account of bypassing index and heap vacuuming.
2526 : *
2527 : * We apply one further precautionary test: the space currently used
2528 : * to store the TIDs (TIDs that now all point to LP_DEAD items) must
2529 : * not exceed 32MB. This limits the risk that we will bypass index
2530 : * vacuuming again and again until eventually there is a VACUUM whose
2531 : * dead_items space is not CPU cache resident.
2532 : *
2533 : * We don't take any special steps to remember the LP_DEAD items (such
2534 : * as counting them in our final update to the stats system) when the
2535 : * optimization is applied. Though the accounting used in analyze.c's
2536 : * acquire_sample_rows() will recognize the same LP_DEAD items as dead
2537 : * rows in its own stats report, that's okay. The discrepancy should
2538 : * be negligible. If this optimization is ever expanded to cover more
2539 : * cases then this may need to be reconsidered.
2540 : */
2541 1236 : threshold = (double) vacrel->rel_pages * BYPASS_THRESHOLD_PAGES;
2542 1238 : bypass = (vacrel->lpdead_item_pages < threshold &&
2543 2 : TidStoreMemoryUsage(vacrel->dead_items) < 32 * 1024 * 1024);
2544 : }
2545 :
2546 1270 : if (bypass)
2547 : {
2548 : /*
2549 : * There are almost zero TIDs. Behave as if there were precisely
2550 : * zero: bypass index vacuuming, but do index cleanup.
2551 : *
2552 : * We expect that the ongoing VACUUM operation will finish very
2553 : * quickly, so there is no point in considering speeding up as a
2554 : * failsafe against wraparound failure. (Index cleanup is expected to
2555 : * finish very quickly in cases where there were no ambulkdelete()
2556 : * calls.)
2557 : */
2558 2 : vacrel->do_index_vacuuming = false;
2559 : }
2560 1268 : else if (lazy_vacuum_all_indexes(vacrel))
2561 : {
2562 : /*
2563 : * We successfully completed a round of index vacuuming. Do related
2564 : * heap vacuuming now.
2565 : */
2566 1268 : lazy_vacuum_heap_rel(vacrel);
2567 : }
2568 : else
2569 : {
2570 : /*
2571 : * Failsafe case.
2572 : *
2573 : * We attempted index vacuuming, but didn't finish a full round/full
2574 : * index scan. This happens when relfrozenxid or relminmxid is too
2575 : * far in the past.
2576 : *
2577 : * From this point on the VACUUM operation will do no further index
2578 : * vacuuming or heap vacuuming. This VACUUM operation won't end up
2579 : * back here again.
2580 : */
2581 : Assert(VacuumFailsafeActive);
2582 : }
2583 :
2584 : /*
2585 : * Forget the LP_DEAD items that we just vacuumed (or just decided to not
2586 : * vacuum)
2587 : */
2588 1270 : dead_items_reset(vacrel);
2589 : }
2590 :
2591 : /*
2592 : * lazy_vacuum_all_indexes() -- Main entry for index vacuuming
2593 : *
2594 : * Returns true in the common case when all indexes were successfully
2595 : * vacuumed. Returns false in rare cases where we determined that the ongoing
2596 : * VACUUM operation is at risk of taking too long to finish, leading to
2597 : * wraparound failure.
2598 : */
2599 : static bool
2600 1268 : lazy_vacuum_all_indexes(LVRelState *vacrel)
2601 : {
2602 1268 : bool allindexes = true;
2603 1268 : double old_live_tuples = vacrel->rel->rd_rel->reltuples;
2604 1268 : const int progress_start_index[] = {
2605 : PROGRESS_VACUUM_PHASE,
2606 : PROGRESS_VACUUM_INDEXES_TOTAL
2607 : };
2608 1268 : const int progress_end_index[] = {
2609 : PROGRESS_VACUUM_INDEXES_TOTAL,
2610 : PROGRESS_VACUUM_INDEXES_PROCESSED,
2611 : PROGRESS_VACUUM_NUM_INDEX_VACUUMS
2612 : };
2613 : int64 progress_start_val[2];
2614 : int64 progress_end_val[3];
2615 :
2616 : Assert(vacrel->nindexes > 0);
2617 : Assert(vacrel->do_index_vacuuming);
2618 : Assert(vacrel->do_index_cleanup);
2619 :
2620 : /* Precheck for XID wraparound emergencies */
2621 1268 : if (lazy_check_wraparound_failsafe(vacrel))
2622 : {
2623 : /* Wraparound emergency -- don't even start an index scan */
2624 0 : return false;
2625 : }
2626 :
2627 : /*
2628 : * Report that we are now vacuuming indexes and the number of indexes to
2629 : * vacuum.
2630 : */
2631 1268 : progress_start_val[0] = PROGRESS_VACUUM_PHASE_VACUUM_INDEX;
2632 1268 : progress_start_val[1] = vacrel->nindexes;
2633 1268 : pgstat_progress_update_multi_param(2, progress_start_index, progress_start_val);
2634 :
2635 1268 : if (!ParallelVacuumIsActive(vacrel))
2636 : {
2637 3644 : for (int idx = 0; idx < vacrel->nindexes; idx++)
2638 : {
2639 2400 : Relation indrel = vacrel->indrels[idx];
2640 2400 : IndexBulkDeleteResult *istat = vacrel->indstats[idx];
2641 :
2642 2400 : vacrel->indstats[idx] = lazy_vacuum_one_index(indrel, istat,
2643 : old_live_tuples,
2644 : vacrel);
2645 :
2646 : /* Report the number of indexes vacuumed */
2647 2400 : pgstat_progress_update_param(PROGRESS_VACUUM_INDEXES_PROCESSED,
2648 2400 : idx + 1);
2649 :
2650 2400 : if (lazy_check_wraparound_failsafe(vacrel))
2651 : {
2652 : /* Wraparound emergency -- end current index scan */
2653 0 : allindexes = false;
2654 0 : break;
2655 : }
2656 : }
2657 : }
2658 : else
2659 : {
2660 : /* Outsource everything to parallel variant */
2661 24 : parallel_vacuum_bulkdel_all_indexes(vacrel->pvs, old_live_tuples,
2662 : vacrel->num_index_scans);
2663 :
2664 : /*
2665 : * Do a postcheck to consider applying wraparound failsafe now. Note
2666 : * that parallel VACUUM only gets the precheck and this postcheck.
2667 : */
2668 24 : if (lazy_check_wraparound_failsafe(vacrel))
2669 0 : allindexes = false;
2670 : }
2671 :
2672 : /*
2673 : * We delete all LP_DEAD items from the first heap pass in all indexes on
2674 : * each call here (except calls where we choose to do the failsafe). This
2675 : * makes the next call to lazy_vacuum_heap_rel() safe (except in the event
2676 : * of the failsafe triggering, which prevents the next call from taking
2677 : * place).
2678 : */
2679 : Assert(vacrel->num_index_scans > 0 ||
2680 : vacrel->dead_items_info->num_items == vacrel->lpdead_items);
2681 : Assert(allindexes || VacuumFailsafeActive);
2682 :
2683 : /*
2684 : * Increase and report the number of index scans. Also, we reset
2685 : * PROGRESS_VACUUM_INDEXES_TOTAL and PROGRESS_VACUUM_INDEXES_PROCESSED.
2686 : *
2687 : * We deliberately include the case where we started a round of bulk
2688 : * deletes that we weren't able to finish due to the failsafe triggering.
2689 : */
2690 1268 : vacrel->num_index_scans++;
2691 1268 : progress_end_val[0] = 0;
2692 1268 : progress_end_val[1] = 0;
2693 1268 : progress_end_val[2] = vacrel->num_index_scans;
2694 1268 : pgstat_progress_update_multi_param(3, progress_end_index, progress_end_val);
2695 :
2696 1268 : return allindexes;
2697 : }
2698 :
2699 : /*
2700 : * Read stream callback for vacuum's third phase (second pass over the heap).
2701 : * Gets the next block from the TID store and returns it or InvalidBlockNumber
2702 : * if there are no further blocks to vacuum.
2703 : *
2704 : * NB: Assumed to be safe to use with READ_STREAM_USE_BATCHING.
2705 : */
2706 : static BlockNumber
2707 29554 : vacuum_reap_lp_read_stream_next(ReadStream *stream,
2708 : void *callback_private_data,
2709 : void *per_buffer_data)
2710 : {
2711 29554 : TidStoreIter *iter = callback_private_data;
2712 : TidStoreIterResult *iter_result;
2713 :
2714 29554 : iter_result = TidStoreIterateNext(iter);
2715 29554 : if (iter_result == NULL)
2716 1268 : return InvalidBlockNumber;
2717 :
2718 : /*
2719 : * Save the TidStoreIterResult for later, so we can extract the offsets.
2720 : * It is safe to copy the result, according to TidStoreIterateNext().
2721 : */
2722 28286 : memcpy(per_buffer_data, iter_result, sizeof(*iter_result));
2723 :
2724 28286 : return iter_result->blkno;
2725 : }
2726 :
2727 : /*
2728 : * lazy_vacuum_heap_rel() -- second pass over the heap for two pass strategy
2729 : *
2730 : * This routine marks LP_DEAD items in vacrel->dead_items as LP_UNUSED. Pages
2731 : * that never had lazy_scan_prune record LP_DEAD items are not visited at all.
2732 : *
2733 : * We may also be able to truncate the line pointer array of the heap pages we
2734 : * visit. If there is a contiguous group of LP_UNUSED items at the end of the
2735 : * array, it can be reclaimed as free space. These LP_UNUSED items usually
2736 : * start out as LP_DEAD items recorded by lazy_scan_prune (we set items from
2737 : * each page to LP_UNUSED, and then consider if it's possible to truncate the
2738 : * page's line pointer array).
2739 : *
2740 : * Note: the reason for doing this as a second pass is we cannot remove the
2741 : * tuples until we've removed their index entries, and we want to process
2742 : * index entry removal in batches as large as possible.
2743 : */
2744 : static void
2745 1268 : lazy_vacuum_heap_rel(LVRelState *vacrel)
2746 : {
2747 : ReadStream *stream;
2748 1268 : BlockNumber vacuumed_pages = 0;
2749 1268 : Buffer vmbuffer = InvalidBuffer;
2750 : LVSavedErrInfo saved_err_info;
2751 : TidStoreIter *iter;
2752 :
2753 : Assert(vacrel->do_index_vacuuming);
2754 : Assert(vacrel->do_index_cleanup);
2755 : Assert(vacrel->num_index_scans > 0);
2756 :
2757 : /* Report that we are now vacuuming the heap */
2758 1268 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
2759 : PROGRESS_VACUUM_PHASE_VACUUM_HEAP);
2760 :
2761 : /* Update error traceback information */
2762 1268 : update_vacuum_error_info(vacrel, &saved_err_info,
2763 : VACUUM_ERRCB_PHASE_VACUUM_HEAP,
2764 : InvalidBlockNumber, InvalidOffsetNumber);
2765 :
2766 1268 : iter = TidStoreBeginIterate(vacrel->dead_items);
2767 :
2768 : /*
2769 : * Set up the read stream for vacuum's second pass through the heap.
2770 : *
2771 : * It is safe to use batchmode, as vacuum_reap_lp_read_stream_next() does
2772 : * not need to wait for IO and does not perform locking. Once we support
2773 : * parallelism it should still be fine, as presumably the holder of locks
2774 : * would never be blocked by IO while holding the lock.
2775 : */
2776 1268 : stream = read_stream_begin_relation(READ_STREAM_MAINTENANCE |
2777 : READ_STREAM_USE_BATCHING,
2778 : vacrel->bstrategy,
2779 : vacrel->rel,
2780 : MAIN_FORKNUM,
2781 : vacuum_reap_lp_read_stream_next,
2782 : iter,
2783 : sizeof(TidStoreIterResult));
2784 :
2785 : while (true)
2786 28286 : {
2787 : BlockNumber blkno;
2788 : Buffer buf;
2789 : Page page;
2790 : TidStoreIterResult *iter_result;
2791 : Size freespace;
2792 : OffsetNumber offsets[MaxOffsetNumber];
2793 : int num_offsets;
2794 :
2795 29554 : vacuum_delay_point(false);
2796 :
2797 29554 : buf = read_stream_next_buffer(stream, (void **) &iter_result);
2798 :
2799 : /* The relation is exhausted */
2800 29554 : if (!BufferIsValid(buf))
2801 1268 : break;
2802 :
2803 28286 : vacrel->blkno = blkno = BufferGetBlockNumber(buf);
2804 :
2805 : Assert(iter_result);
2806 28286 : num_offsets = TidStoreGetBlockOffsets(iter_result, offsets, lengthof(offsets));
2807 : Assert(num_offsets <= lengthof(offsets));
2808 :
2809 : /*
2810 : * Pin the visibility map page in case we need to mark the page
2811 : * all-visible. In most cases this will be very cheap, because we'll
2812 : * already have the correct page pinned anyway.
2813 : */
2814 28286 : visibilitymap_pin(vacrel->rel, blkno, &vmbuffer);
2815 :
2816 : /* We need a non-cleanup exclusive lock to mark dead_items unused */
2817 28286 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
2818 28286 : lazy_vacuum_heap_page(vacrel, blkno, buf, offsets,
2819 : num_offsets, vmbuffer);
2820 :
2821 : /* Now that we've vacuumed the page, record its available space */
2822 28286 : page = BufferGetPage(buf);
2823 28286 : freespace = PageGetHeapFreeSpace(page);
2824 :
2825 28286 : UnlockReleaseBuffer(buf);
2826 28286 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
2827 28286 : vacuumed_pages++;
2828 : }
2829 :
2830 1268 : read_stream_end(stream);
2831 1268 : TidStoreEndIterate(iter);
2832 :
2833 1268 : vacrel->blkno = InvalidBlockNumber;
2834 1268 : if (BufferIsValid(vmbuffer))
2835 1268 : ReleaseBuffer(vmbuffer);
2836 :
2837 : /*
2838 : * We set all LP_DEAD items from the first heap pass to LP_UNUSED during
2839 : * the second heap pass. No more, no less.
2840 : */
2841 : Assert(vacrel->num_index_scans > 1 ||
2842 : (vacrel->dead_items_info->num_items == vacrel->lpdead_items &&
2843 : vacuumed_pages == vacrel->lpdead_item_pages));
2844 :
2845 1268 : ereport(DEBUG2,
2846 : (errmsg("table \"%s\": removed %" PRId64 " dead item identifiers in %u pages",
2847 : vacrel->relname, vacrel->dead_items_info->num_items,
2848 : vacuumed_pages)));
2849 :
2850 : /* Revert to the previous phase information for error traceback */
2851 1268 : restore_vacuum_error_info(vacrel, &saved_err_info);
2852 1268 : }
2853 :
2854 : /*
2855 : * lazy_vacuum_heap_page() -- free page's LP_DEAD items listed in the
2856 : * vacrel->dead_items store.
2857 : *
2858 : * Caller must have an exclusive buffer lock on the buffer (though a full
2859 : * cleanup lock is also acceptable). vmbuffer must be valid and already have
2860 : * a pin on blkno's visibility map page.
2861 : */
2862 : static void
2863 28286 : lazy_vacuum_heap_page(LVRelState *vacrel, BlockNumber blkno, Buffer buffer,
2864 : OffsetNumber *deadoffsets, int num_offsets,
2865 : Buffer vmbuffer)
2866 : {
2867 28286 : Page page = BufferGetPage(buffer);
2868 : OffsetNumber unused[MaxHeapTuplesPerPage];
2869 28286 : int nunused = 0;
2870 : TransactionId visibility_cutoff_xid;
2871 28286 : TransactionId conflict_xid = InvalidTransactionId;
2872 : bool all_frozen;
2873 : LVSavedErrInfo saved_err_info;
2874 28286 : uint8 vmflags = 0;
2875 :
2876 : Assert(vacrel->do_index_vacuuming);
2877 :
2878 28286 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_VACUUMED, blkno);
2879 :
2880 : /* Update error traceback information */
2881 28286 : update_vacuum_error_info(vacrel, &saved_err_info,
2882 : VACUUM_ERRCB_PHASE_VACUUM_HEAP, blkno,
2883 : InvalidOffsetNumber);
2884 :
2885 : /*
2886 : * Before marking dead items unused, check whether the page will become
2887 : * all-visible once that change is applied. This lets us reap the tuples
2888 : * and mark the page all-visible within the same critical section,
2889 : * enabling both changes to be emitted in a single WAL record. Since the
2890 : * visibility checks may perform I/O and allocate memory, they must be
2891 : * done outside the critical section.
2892 : */
2893 28286 : if (heap_page_would_be_all_visible(vacrel->rel, buffer,
2894 : vacrel->cutoffs.OldestXmin,
2895 : deadoffsets, num_offsets,
2896 : &all_frozen, &visibility_cutoff_xid,
2897 : &vacrel->offnum))
2898 : {
2899 28142 : vmflags |= VISIBILITYMAP_ALL_VISIBLE;
2900 28142 : if (all_frozen)
2901 : {
2902 22472 : vmflags |= VISIBILITYMAP_ALL_FROZEN;
2903 : Assert(!TransactionIdIsValid(visibility_cutoff_xid));
2904 : }
2905 :
2906 : /*
2907 : * Take the lock on the vmbuffer before entering a critical section.
2908 : * The heap page lock must also be held while updating the VM to
2909 : * ensure consistency.
2910 : */
2911 28142 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2912 : }
2913 :
2914 28286 : START_CRIT_SECTION();
2915 :
2916 1739924 : for (int i = 0; i < num_offsets; i++)
2917 : {
2918 : ItemId itemid;
2919 1711638 : OffsetNumber toff = deadoffsets[i];
2920 :
2921 1711638 : itemid = PageGetItemId(page, toff);
2922 :
2923 : Assert(ItemIdIsDead(itemid) && !ItemIdHasStorage(itemid));
2924 1711638 : ItemIdSetUnused(itemid);
2925 1711638 : unused[nunused++] = toff;
2926 : }
2927 :
2928 : Assert(nunused > 0);
2929 :
2930 : /* Attempt to truncate line pointer array now */
2931 28286 : PageTruncateLinePointerArray(page);
2932 :
2933 28286 : if ((vmflags & VISIBILITYMAP_VALID_BITS) != 0)
2934 : {
2935 : /*
2936 : * The page is guaranteed to have had dead line pointers, so we always
2937 : * set PD_ALL_VISIBLE.
2938 : */
2939 28142 : PageSetAllVisible(page);
2940 28142 : visibilitymap_set_vmbits(blkno,
2941 : vmbuffer, vmflags,
2942 28142 : vacrel->rel->rd_locator);
2943 28142 : conflict_xid = visibility_cutoff_xid;
2944 : }
2945 :
2946 : /*
2947 : * Mark buffer dirty before we write WAL.
2948 : */
2949 28286 : MarkBufferDirty(buffer);
2950 :
2951 : /* XLOG stuff */
2952 28286 : if (RelationNeedsWAL(vacrel->rel))
2953 : {
2954 26586 : log_heap_prune_and_freeze(vacrel->rel, buffer,
2955 : vmflags != 0 ? vmbuffer : InvalidBuffer,
2956 : vmflags,
2957 : conflict_xid,
2958 : false, /* no cleanup lock required */
2959 : PRUNE_VACUUM_CLEANUP,
2960 : NULL, 0, /* frozen */
2961 : NULL, 0, /* redirected */
2962 : NULL, 0, /* dead */
2963 : unused, nunused);
2964 : }
2965 :
2966 28286 : END_CRIT_SECTION();
2967 :
2968 28286 : if ((vmflags & VISIBILITYMAP_ALL_VISIBLE) != 0)
2969 : {
2970 : /* Count the newly set VM page for logging */
2971 28142 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2972 28142 : vacrel->vm_new_visible_pages++;
2973 28142 : if (all_frozen)
2974 22472 : vacrel->vm_new_visible_frozen_pages++;
2975 : }
2976 :
2977 : /* Revert to the previous phase information for error traceback */
2978 28286 : restore_vacuum_error_info(vacrel, &saved_err_info);
2979 28286 : }
2980 :
2981 : /*
2982 : * Trigger the failsafe to avoid wraparound failure when vacrel table has a
2983 : * relfrozenxid and/or relminmxid that is dangerously far in the past.
2984 : * Triggering the failsafe makes the ongoing VACUUM bypass any further index
2985 : * vacuuming and heap vacuuming. Truncating the heap is also bypassed.
2986 : *
2987 : * Any remaining work (work that VACUUM cannot just bypass) is typically sped
2988 : * up when the failsafe triggers. VACUUM stops applying any cost-based delay
2989 : * that it started out with.
2990 : *
2991 : * Returns true when failsafe has been triggered.
2992 : */
2993 : static bool
2994 254274 : lazy_check_wraparound_failsafe(LVRelState *vacrel)
2995 : {
2996 : /* Don't warn more than once per VACUUM */
2997 254274 : if (VacuumFailsafeActive)
2998 0 : return true;
2999 :
3000 254274 : if (unlikely(vacuum_xid_failsafe_check(&vacrel->cutoffs)))
3001 : {
3002 82480 : const int progress_index[] = {
3003 : PROGRESS_VACUUM_INDEXES_TOTAL,
3004 : PROGRESS_VACUUM_INDEXES_PROCESSED
3005 : };
3006 82480 : int64 progress_val[2] = {0, 0};
3007 :
3008 82480 : VacuumFailsafeActive = true;
3009 :
3010 : /*
3011 : * Abandon use of a buffer access strategy to allow use of all of
3012 : * shared buffers. We assume the caller who allocated the memory for
3013 : * the BufferAccessStrategy will free it.
3014 : */
3015 82480 : vacrel->bstrategy = NULL;
3016 :
3017 : /* Disable index vacuuming, index cleanup, and heap rel truncation */
3018 82480 : vacrel->do_index_vacuuming = false;
3019 82480 : vacrel->do_index_cleanup = false;
3020 82480 : vacrel->do_rel_truncate = false;
3021 :
3022 : /* Reset the progress counters */
3023 82480 : pgstat_progress_update_multi_param(2, progress_index, progress_val);
3024 :
3025 82480 : ereport(WARNING,
3026 : (errmsg("bypassing nonessential maintenance of table \"%s.%s.%s\" as a failsafe after %d index scans",
3027 : vacrel->dbname, vacrel->relnamespace, vacrel->relname,
3028 : vacrel->num_index_scans),
3029 : errdetail("The table's relfrozenxid or relminmxid is too far in the past."),
3030 : errhint("Consider increasing configuration parameter \"maintenance_work_mem\" or \"autovacuum_work_mem\".\n"
3031 : "You might also need to consider other ways for VACUUM to keep up with the allocation of transaction IDs.")));
3032 :
3033 : /* Stop applying cost limits from this point on */
3034 82480 : VacuumCostActive = false;
3035 82480 : VacuumCostBalance = 0;
3036 :
3037 82480 : return true;
3038 : }
3039 :
3040 171794 : return false;
3041 : }
3042 :
3043 : /*
3044 : * lazy_cleanup_all_indexes() -- cleanup all indexes of relation.
3045 : */
3046 : static void
3047 159612 : lazy_cleanup_all_indexes(LVRelState *vacrel)
3048 : {
3049 159612 : double reltuples = vacrel->new_rel_tuples;
3050 159612 : bool estimated_count = vacrel->scanned_pages < vacrel->rel_pages;
3051 159612 : const int progress_start_index[] = {
3052 : PROGRESS_VACUUM_PHASE,
3053 : PROGRESS_VACUUM_INDEXES_TOTAL
3054 : };
3055 159612 : const int progress_end_index[] = {
3056 : PROGRESS_VACUUM_INDEXES_TOTAL,
3057 : PROGRESS_VACUUM_INDEXES_PROCESSED
3058 : };
3059 : int64 progress_start_val[2];
3060 159612 : int64 progress_end_val[2] = {0, 0};
3061 :
3062 : Assert(vacrel->do_index_cleanup);
3063 : Assert(vacrel->nindexes > 0);
3064 :
3065 : /*
3066 : * Report that we are now cleaning up indexes and the number of indexes to
3067 : * cleanup.
3068 : */
3069 159612 : progress_start_val[0] = PROGRESS_VACUUM_PHASE_INDEX_CLEANUP;
3070 159612 : progress_start_val[1] = vacrel->nindexes;
3071 159612 : pgstat_progress_update_multi_param(2, progress_start_index, progress_start_val);
3072 :
3073 159612 : if (!ParallelVacuumIsActive(vacrel))
3074 : {
3075 410226 : for (int idx = 0; idx < vacrel->nindexes; idx++)
3076 : {
3077 250648 : Relation indrel = vacrel->indrels[idx];
3078 250648 : IndexBulkDeleteResult *istat = vacrel->indstats[idx];
3079 :
3080 501296 : vacrel->indstats[idx] =
3081 250648 : lazy_cleanup_one_index(indrel, istat, reltuples,
3082 : estimated_count, vacrel);
3083 :
3084 : /* Report the number of indexes cleaned up */
3085 250648 : pgstat_progress_update_param(PROGRESS_VACUUM_INDEXES_PROCESSED,
3086 250648 : idx + 1);
3087 : }
3088 : }
3089 : else
3090 : {
3091 : /* Outsource everything to parallel variant */
3092 34 : parallel_vacuum_cleanup_all_indexes(vacrel->pvs, reltuples,
3093 : vacrel->num_index_scans,
3094 : estimated_count);
3095 : }
3096 :
3097 : /* Reset the progress counters */
3098 159612 : pgstat_progress_update_multi_param(2, progress_end_index, progress_end_val);
3099 159612 : }
3100 :
3101 : /*
3102 : * lazy_vacuum_one_index() -- vacuum index relation.
3103 : *
3104 : * Delete all the index tuples containing a TID collected in
3105 : * vacrel->dead_items. Also update running statistics. Exact
3106 : * details depend on index AM's ambulkdelete routine.
3107 : *
3108 : * reltuples is the number of heap tuples to be passed to the
3109 : * bulkdelete callback. It's always assumed to be estimated.
3110 : * See indexam.sgml for more info.
3111 : *
3112 : * Returns bulk delete stats derived from input stats
3113 : */
3114 : static IndexBulkDeleteResult *
3115 2400 : lazy_vacuum_one_index(Relation indrel, IndexBulkDeleteResult *istat,
3116 : double reltuples, LVRelState *vacrel)
3117 : {
3118 : IndexVacuumInfo ivinfo;
3119 : LVSavedErrInfo saved_err_info;
3120 :
3121 2400 : ivinfo.index = indrel;
3122 2400 : ivinfo.heaprel = vacrel->rel;
3123 2400 : ivinfo.analyze_only = false;
3124 2400 : ivinfo.report_progress = false;
3125 2400 : ivinfo.estimated_count = true;
3126 2400 : ivinfo.message_level = DEBUG2;
3127 2400 : ivinfo.num_heap_tuples = reltuples;
3128 2400 : ivinfo.strategy = vacrel->bstrategy;
3129 :
3130 : /*
3131 : * Update error traceback information.
3132 : *
3133 : * The index name is saved during this phase and restored immediately
3134 : * after this phase. See vacuum_error_callback.
3135 : */
3136 : Assert(vacrel->indname == NULL);
3137 2400 : vacrel->indname = pstrdup(RelationGetRelationName(indrel));
3138 2400 : update_vacuum_error_info(vacrel, &saved_err_info,
3139 : VACUUM_ERRCB_PHASE_VACUUM_INDEX,
3140 : InvalidBlockNumber, InvalidOffsetNumber);
3141 :
3142 : /* Do bulk deletion */
3143 2400 : istat = vac_bulkdel_one_index(&ivinfo, istat, vacrel->dead_items,
3144 : vacrel->dead_items_info);
3145 :
3146 : /* Revert to the previous phase information for error traceback */
3147 2400 : restore_vacuum_error_info(vacrel, &saved_err_info);
3148 2400 : pfree(vacrel->indname);
3149 2400 : vacrel->indname = NULL;
3150 :
3151 2400 : return istat;
3152 : }
3153 :
3154 : /*
3155 : * lazy_cleanup_one_index() -- do post-vacuum cleanup for index relation.
3156 : *
3157 : * Calls index AM's amvacuumcleanup routine. reltuples is the number
3158 : * of heap tuples and estimated_count is true if reltuples is an
3159 : * estimated value. See indexam.sgml for more info.
3160 : *
3161 : * Returns bulk delete stats derived from input stats
3162 : */
3163 : static IndexBulkDeleteResult *
3164 250648 : lazy_cleanup_one_index(Relation indrel, IndexBulkDeleteResult *istat,
3165 : double reltuples, bool estimated_count,
3166 : LVRelState *vacrel)
3167 : {
3168 : IndexVacuumInfo ivinfo;
3169 : LVSavedErrInfo saved_err_info;
3170 :
3171 250648 : ivinfo.index = indrel;
3172 250648 : ivinfo.heaprel = vacrel->rel;
3173 250648 : ivinfo.analyze_only = false;
3174 250648 : ivinfo.report_progress = false;
3175 250648 : ivinfo.estimated_count = estimated_count;
3176 250648 : ivinfo.message_level = DEBUG2;
3177 :
3178 250648 : ivinfo.num_heap_tuples = reltuples;
3179 250648 : ivinfo.strategy = vacrel->bstrategy;
3180 :
3181 : /*
3182 : * Update error traceback information.
3183 : *
3184 : * The index name is saved during this phase and restored immediately
3185 : * after this phase. See vacuum_error_callback.
3186 : */
3187 : Assert(vacrel->indname == NULL);
3188 250648 : vacrel->indname = pstrdup(RelationGetRelationName(indrel));
3189 250648 : update_vacuum_error_info(vacrel, &saved_err_info,
3190 : VACUUM_ERRCB_PHASE_INDEX_CLEANUP,
3191 : InvalidBlockNumber, InvalidOffsetNumber);
3192 :
3193 250648 : istat = vac_cleanup_one_index(&ivinfo, istat);
3194 :
3195 : /* Revert to the previous phase information for error traceback */
3196 250648 : restore_vacuum_error_info(vacrel, &saved_err_info);
3197 250648 : pfree(vacrel->indname);
3198 250648 : vacrel->indname = NULL;
3199 :
3200 250648 : return istat;
3201 : }
3202 :
3203 : /*
3204 : * should_attempt_truncation - should we attempt to truncate the heap?
3205 : *
3206 : * Don't even think about it unless we have a shot at releasing a goodly
3207 : * number of pages. Otherwise, the time taken isn't worth it, mainly because
3208 : * an AccessExclusive lock must be replayed on any hot standby, where it can
3209 : * be particularly disruptive.
3210 : *
3211 : * Also don't attempt it if wraparound failsafe is in effect. The entire
3212 : * system might be refusing to allocate new XIDs at this point. The system
3213 : * definitely won't return to normal unless and until VACUUM actually advances
3214 : * the oldest relfrozenxid -- which hasn't happened for target rel just yet.
3215 : * If lazy_truncate_heap attempted to acquire an AccessExclusiveLock to
3216 : * truncate the table under these circumstances, an XID exhaustion error might
3217 : * make it impossible for VACUUM to fix the underlying XID exhaustion problem.
3218 : * There is very little chance of truncation working out when the failsafe is
3219 : * in effect in any case. lazy_scan_prune makes the optimistic assumption
3220 : * that any LP_DEAD items it encounters will always be LP_UNUSED by the time
3221 : * we're called.
3222 : */
3223 : static bool
3224 250582 : should_attempt_truncation(LVRelState *vacrel)
3225 : {
3226 : BlockNumber possibly_freeable;
3227 :
3228 250582 : if (!vacrel->do_rel_truncate || VacuumFailsafeActive)
3229 82770 : return false;
3230 :
3231 167812 : possibly_freeable = vacrel->rel_pages - vacrel->nonempty_pages;
3232 167812 : if (possibly_freeable > 0 &&
3233 314 : (possibly_freeable >= REL_TRUNCATE_MINIMUM ||
3234 314 : possibly_freeable >= vacrel->rel_pages / REL_TRUNCATE_FRACTION))
3235 286 : return true;
3236 :
3237 167526 : return false;
3238 : }
3239 :
3240 : /*
3241 : * lazy_truncate_heap - try to truncate off any empty pages at the end
3242 : */
3243 : static void
3244 286 : lazy_truncate_heap(LVRelState *vacrel)
3245 : {
3246 286 : BlockNumber orig_rel_pages = vacrel->rel_pages;
3247 : BlockNumber new_rel_pages;
3248 : bool lock_waiter_detected;
3249 : int lock_retry;
3250 :
3251 : /* Report that we are now truncating */
3252 286 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
3253 : PROGRESS_VACUUM_PHASE_TRUNCATE);
3254 :
3255 : /* Update error traceback information one last time */
3256 286 : update_vacuum_error_info(vacrel, NULL, VACUUM_ERRCB_PHASE_TRUNCATE,
3257 : vacrel->nonempty_pages, InvalidOffsetNumber);
3258 :
3259 : /*
3260 : * Loop until no more truncating can be done.
3261 : */
3262 : do
3263 : {
3264 : /*
3265 : * We need full exclusive lock on the relation in order to do
3266 : * truncation. If we can't get it, give up rather than waiting --- we
3267 : * don't want to block other backends, and we don't want to deadlock
3268 : * (which is quite possible considering we already hold a lower-grade
3269 : * lock).
3270 : */
3271 286 : lock_waiter_detected = false;
3272 286 : lock_retry = 0;
3273 : while (true)
3274 : {
3275 686 : if (ConditionalLockRelation(vacrel->rel, AccessExclusiveLock))
3276 282 : break;
3277 :
3278 : /*
3279 : * Check for interrupts while trying to (re-)acquire the exclusive
3280 : * lock.
3281 : */
3282 404 : CHECK_FOR_INTERRUPTS();
3283 :
3284 404 : if (++lock_retry > (VACUUM_TRUNCATE_LOCK_TIMEOUT /
3285 : VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL))
3286 : {
3287 : /*
3288 : * We failed to establish the lock in the specified number of
3289 : * retries. This means we give up truncating.
3290 : */
3291 4 : ereport(vacrel->verbose ? INFO : DEBUG2,
3292 : (errmsg("\"%s\": stopping truncate due to conflicting lock request",
3293 : vacrel->relname)));
3294 6 : return;
3295 : }
3296 :
3297 400 : (void) WaitLatch(MyLatch,
3298 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
3299 : VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL,
3300 : WAIT_EVENT_VACUUM_TRUNCATE);
3301 400 : ResetLatch(MyLatch);
3302 : }
3303 :
3304 : /*
3305 : * Now that we have exclusive lock, look to see if the rel has grown
3306 : * whilst we were vacuuming with non-exclusive lock. If so, give up;
3307 : * the newly added pages presumably contain non-deletable tuples.
3308 : */
3309 282 : new_rel_pages = RelationGetNumberOfBlocks(vacrel->rel);
3310 282 : if (new_rel_pages != orig_rel_pages)
3311 : {
3312 : /*
3313 : * Note: we intentionally don't update vacrel->rel_pages with the
3314 : * new rel size here. If we did, it would amount to assuming that
3315 : * the new pages are empty, which is unlikely. Leaving the numbers
3316 : * alone amounts to assuming that the new pages have the same
3317 : * tuple density as existing ones, which is less unlikely.
3318 : */
3319 0 : UnlockRelation(vacrel->rel, AccessExclusiveLock);
3320 0 : return;
3321 : }
3322 :
3323 : /*
3324 : * Scan backwards from the end to verify that the end pages actually
3325 : * contain no tuples. This is *necessary*, not optional, because
3326 : * other backends could have added tuples to these pages whilst we
3327 : * were vacuuming.
3328 : */
3329 282 : new_rel_pages = count_nondeletable_pages(vacrel, &lock_waiter_detected);
3330 282 : vacrel->blkno = new_rel_pages;
3331 :
3332 282 : if (new_rel_pages >= orig_rel_pages)
3333 : {
3334 : /* can't do anything after all */
3335 2 : UnlockRelation(vacrel->rel, AccessExclusiveLock);
3336 2 : return;
3337 : }
3338 :
3339 : /*
3340 : * Okay to truncate.
3341 : */
3342 280 : RelationTruncate(vacrel->rel, new_rel_pages);
3343 :
3344 : /*
3345 : * We can release the exclusive lock as soon as we have truncated.
3346 : * Other backends can't safely access the relation until they have
3347 : * processed the smgr invalidation that smgrtruncate sent out ... but
3348 : * that should happen as part of standard invalidation processing once
3349 : * they acquire lock on the relation.
3350 : */
3351 280 : UnlockRelation(vacrel->rel, AccessExclusiveLock);
3352 :
3353 : /*
3354 : * Update statistics. Here, it *is* correct to adjust rel_pages
3355 : * without also touching reltuples, since the tuple count wasn't
3356 : * changed by the truncation.
3357 : */
3358 280 : vacrel->removed_pages += orig_rel_pages - new_rel_pages;
3359 280 : vacrel->rel_pages = new_rel_pages;
3360 :
3361 280 : ereport(vacrel->verbose ? INFO : DEBUG2,
3362 : (errmsg("table \"%s\": truncated %u to %u pages",
3363 : vacrel->relname,
3364 : orig_rel_pages, new_rel_pages)));
3365 280 : orig_rel_pages = new_rel_pages;
3366 280 : } while (new_rel_pages > vacrel->nonempty_pages && lock_waiter_detected);
3367 : }
3368 :
3369 : /*
3370 : * Rescan end pages to verify that they are (still) empty of tuples.
3371 : *
3372 : * Returns number of nondeletable pages (last nonempty page + 1).
3373 : */
3374 : static BlockNumber
3375 282 : count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
3376 : {
3377 : BlockNumber blkno;
3378 : BlockNumber prefetchedUntil;
3379 : instr_time starttime;
3380 :
3381 : /* Initialize the starttime if we check for conflicting lock requests */
3382 282 : INSTR_TIME_SET_CURRENT(starttime);
3383 :
3384 : /*
3385 : * Start checking blocks at what we believe relation end to be and move
3386 : * backwards. (Strange coding of loop control is needed because blkno is
3387 : * unsigned.) To make the scan faster, we prefetch a few blocks at a time
3388 : * in forward direction, so that OS-level readahead can kick in.
3389 : */
3390 282 : blkno = vacrel->rel_pages;
3391 : StaticAssertStmt((PREFETCH_SIZE & (PREFETCH_SIZE - 1)) == 0,
3392 : "prefetch size must be power of 2");
3393 282 : prefetchedUntil = InvalidBlockNumber;
3394 4404 : while (blkno > vacrel->nonempty_pages)
3395 : {
3396 : Buffer buf;
3397 : Page page;
3398 : OffsetNumber offnum,
3399 : maxoff;
3400 : bool hastup;
3401 :
3402 : /*
3403 : * Check if another process requests a lock on our relation. We are
3404 : * holding an AccessExclusiveLock here, so they will be waiting. We
3405 : * only do this once per VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL, and we
3406 : * only check if that interval has elapsed once every 32 blocks to
3407 : * keep the number of system calls and actual shared lock table
3408 : * lookups to a minimum.
3409 : */
3410 4130 : if ((blkno % 32) == 0)
3411 : {
3412 : instr_time currenttime;
3413 : instr_time elapsed;
3414 :
3415 130 : INSTR_TIME_SET_CURRENT(currenttime);
3416 130 : elapsed = currenttime;
3417 130 : INSTR_TIME_SUBTRACT(elapsed, starttime);
3418 130 : if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
3419 : >= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
3420 : {
3421 0 : if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
3422 : {
3423 0 : ereport(vacrel->verbose ? INFO : DEBUG2,
3424 : (errmsg("table \"%s\": suspending truncate due to conflicting lock request",
3425 : vacrel->relname)));
3426 :
3427 0 : *lock_waiter_detected = true;
3428 0 : return blkno;
3429 : }
3430 0 : starttime = currenttime;
3431 : }
3432 : }
3433 :
3434 : /*
3435 : * We don't insert a vacuum delay point here, because we have an
3436 : * exclusive lock on the table which we want to hold for as short a
3437 : * time as possible. We still need to check for interrupts however.
3438 : */
3439 4130 : CHECK_FOR_INTERRUPTS();
3440 :
3441 4130 : blkno--;
3442 :
3443 : /* If we haven't prefetched this lot yet, do so now. */
3444 4130 : if (prefetchedUntil > blkno)
3445 : {
3446 : BlockNumber prefetchStart;
3447 : BlockNumber pblkno;
3448 :
3449 376 : prefetchStart = blkno & ~(PREFETCH_SIZE - 1);
3450 6044 : for (pblkno = prefetchStart; pblkno <= blkno; pblkno++)
3451 : {
3452 5668 : PrefetchBuffer(vacrel->rel, MAIN_FORKNUM, pblkno);
3453 5668 : CHECK_FOR_INTERRUPTS();
3454 : }
3455 376 : prefetchedUntil = prefetchStart;
3456 : }
3457 :
3458 4130 : buf = ReadBufferExtended(vacrel->rel, MAIN_FORKNUM, blkno, RBM_NORMAL,
3459 : vacrel->bstrategy);
3460 :
3461 : /* In this phase we only need shared access to the buffer */
3462 4130 : LockBuffer(buf, BUFFER_LOCK_SHARE);
3463 :
3464 4130 : page = BufferGetPage(buf);
3465 :
3466 4130 : if (PageIsNew(page) || PageIsEmpty(page))
3467 : {
3468 1646 : UnlockReleaseBuffer(buf);
3469 1646 : continue;
3470 : }
3471 :
3472 2484 : hastup = false;
3473 2484 : maxoff = PageGetMaxOffsetNumber(page);
3474 2484 : for (offnum = FirstOffsetNumber;
3475 5250 : offnum <= maxoff;
3476 2766 : offnum = OffsetNumberNext(offnum))
3477 : {
3478 : ItemId itemid;
3479 :
3480 2774 : itemid = PageGetItemId(page, offnum);
3481 :
3482 : /*
3483 : * Note: any non-unused item should be taken as a reason to keep
3484 : * this page. Even an LP_DEAD item makes truncation unsafe, since
3485 : * we must not have cleaned out its index entries.
3486 : */
3487 2774 : if (ItemIdIsUsed(itemid))
3488 : {
3489 8 : hastup = true;
3490 8 : break; /* can stop scanning */
3491 : }
3492 : } /* scan along page */
3493 :
3494 2484 : UnlockReleaseBuffer(buf);
3495 :
3496 : /* Done scanning if we found a tuple here */
3497 2484 : if (hastup)
3498 8 : return blkno + 1;
3499 : }
3500 :
3501 : /*
3502 : * If we fall out of the loop, all the previously-thought-to-be-empty
3503 : * pages still are; we need not bother to look at the last known-nonempty
3504 : * page.
3505 : */
3506 274 : return vacrel->nonempty_pages;
3507 : }
3508 :
3509 : /*
3510 : * Allocate dead_items and dead_items_info (either using palloc, or in dynamic
3511 : * shared memory). Sets both in vacrel for caller.
3512 : *
3513 : * Also handles parallel initialization as part of allocating dead_items in
3514 : * DSM when required.
3515 : */
3516 : static void
3517 250582 : dead_items_alloc(LVRelState *vacrel, int nworkers)
3518 : {
3519 : VacDeadItemsInfo *dead_items_info;
3520 725320 : int vac_work_mem = AmAutoVacuumWorkerProcess() &&
3521 224156 : autovacuum_work_mem != -1 ?
3522 474738 : autovacuum_work_mem : maintenance_work_mem;
3523 :
3524 : /*
3525 : * Initialize state for a parallel vacuum. As of now, only one worker can
3526 : * be used for an index, so we invoke parallelism only if there are at
3527 : * least two indexes on a table.
3528 : */
3529 250582 : if (nworkers >= 0 && vacrel->nindexes > 1 && vacrel->do_index_vacuuming)
3530 : {
3531 : /*
3532 : * Since parallel workers cannot access data in temporary tables, we
3533 : * can't perform parallel vacuum on them.
3534 : */
3535 10584 : if (RelationUsesLocalBuffers(vacrel->rel))
3536 : {
3537 : /*
3538 : * Give warning only if the user explicitly tries to perform a
3539 : * parallel vacuum on the temporary table.
3540 : */
3541 6 : if (nworkers > 0)
3542 6 : ereport(WARNING,
3543 : (errmsg("disabling parallel option of vacuum on \"%s\" --- cannot vacuum temporary tables in parallel",
3544 : vacrel->relname)));
3545 : }
3546 : else
3547 10578 : vacrel->pvs = parallel_vacuum_init(vacrel->rel, vacrel->indrels,
3548 : vacrel->nindexes, nworkers,
3549 : vac_work_mem,
3550 10578 : vacrel->verbose ? INFO : DEBUG2,
3551 : vacrel->bstrategy);
3552 :
3553 : /*
3554 : * If parallel mode started, dead_items and dead_items_info spaces are
3555 : * allocated in DSM.
3556 : */
3557 10584 : if (ParallelVacuumIsActive(vacrel))
3558 : {
3559 34 : vacrel->dead_items = parallel_vacuum_get_dead_items(vacrel->pvs,
3560 : &vacrel->dead_items_info);
3561 34 : return;
3562 : }
3563 : }
3564 :
3565 : /*
3566 : * Serial VACUUM case. Allocate both dead_items and dead_items_info
3567 : * locally.
3568 : */
3569 :
3570 250548 : dead_items_info = (VacDeadItemsInfo *) palloc(sizeof(VacDeadItemsInfo));
3571 250548 : dead_items_info->max_bytes = vac_work_mem * (Size) 1024;
3572 250548 : dead_items_info->num_items = 0;
3573 250548 : vacrel->dead_items_info = dead_items_info;
3574 :
3575 250548 : vacrel->dead_items = TidStoreCreateLocal(dead_items_info->max_bytes, true);
3576 : }
3577 :
3578 : /*
3579 : * Add the given block number and offset numbers to dead_items.
3580 : */
3581 : static void
3582 32868 : dead_items_add(LVRelState *vacrel, BlockNumber blkno, OffsetNumber *offsets,
3583 : int num_offsets)
3584 : {
3585 32868 : const int prog_index[2] = {
3586 : PROGRESS_VACUUM_NUM_DEAD_ITEM_IDS,
3587 : PROGRESS_VACUUM_DEAD_TUPLE_BYTES
3588 : };
3589 : int64 prog_val[2];
3590 :
3591 32868 : TidStoreSetBlockOffsets(vacrel->dead_items, blkno, offsets, num_offsets);
3592 32868 : vacrel->dead_items_info->num_items += num_offsets;
3593 :
3594 : /* update the progress information */
3595 32868 : prog_val[0] = vacrel->dead_items_info->num_items;
3596 32868 : prog_val[1] = TidStoreMemoryUsage(vacrel->dead_items);
3597 32868 : pgstat_progress_update_multi_param(2, prog_index, prog_val);
3598 32868 : }
3599 :
3600 : /*
3601 : * Forget all collected dead items.
3602 : */
3603 : static void
3604 1286 : dead_items_reset(LVRelState *vacrel)
3605 : {
3606 1286 : if (ParallelVacuumIsActive(vacrel))
3607 : {
3608 24 : parallel_vacuum_reset_dead_items(vacrel->pvs);
3609 24 : vacrel->dead_items = parallel_vacuum_get_dead_items(vacrel->pvs,
3610 : &vacrel->dead_items_info);
3611 24 : return;
3612 : }
3613 :
3614 : /* Recreate the tidstore with the same max_bytes limitation */
3615 1262 : TidStoreDestroy(vacrel->dead_items);
3616 1262 : vacrel->dead_items = TidStoreCreateLocal(vacrel->dead_items_info->max_bytes, true);
3617 :
3618 : /* Reset the counter */
3619 1262 : vacrel->dead_items_info->num_items = 0;
3620 : }
3621 :
3622 : /*
3623 : * Perform cleanup for resources allocated in dead_items_alloc
3624 : */
3625 : static void
3626 250582 : dead_items_cleanup(LVRelState *vacrel)
3627 : {
3628 250582 : if (!ParallelVacuumIsActive(vacrel))
3629 : {
3630 : /* Don't bother with pfree here */
3631 250548 : return;
3632 : }
3633 :
3634 : /* End parallel mode */
3635 34 : parallel_vacuum_end(vacrel->pvs, vacrel->indstats);
3636 34 : vacrel->pvs = NULL;
3637 : }
3638 :
3639 : #ifdef USE_ASSERT_CHECKING
3640 :
3641 : /*
3642 : * Wrapper for heap_page_would_be_all_visible() which can be used for callers
3643 : * that expect no LP_DEAD on the page. Currently assert-only, but there is no
3644 : * reason not to use it outside of asserts.
3645 : */
3646 : static bool
3647 : heap_page_is_all_visible(Relation rel, Buffer buf,
3648 : TransactionId OldestXmin,
3649 : bool *all_frozen,
3650 : TransactionId *visibility_cutoff_xid,
3651 : OffsetNumber *logging_offnum)
3652 : {
3653 :
3654 : return heap_page_would_be_all_visible(rel, buf,
3655 : OldestXmin,
3656 : NULL, 0,
3657 : all_frozen,
3658 : visibility_cutoff_xid,
3659 : logging_offnum);
3660 : }
3661 : #endif
3662 :
3663 : /*
3664 : * Check whether the heap page in buf is all-visible except for the dead
3665 : * tuples referenced in the deadoffsets array.
3666 : *
3667 : * Vacuum uses this to check if a page would become all-visible after reaping
3668 : * known dead tuples. This function does not remove the dead items.
3669 : *
3670 : * This cannot be called in a critical section, as the visibility checks may
3671 : * perform IO and allocate memory.
3672 : *
3673 : * Returns true if the page is all-visible other than the provided
3674 : * deadoffsets and false otherwise.
3675 : *
3676 : * OldestXmin is used to determine visibility.
3677 : *
3678 : * Output parameters:
3679 : *
3680 : * - *all_frozen: true if every tuple on the page is frozen
3681 : * - *visibility_cutoff_xid: newest xmin; valid only if page is all-visible
3682 : * - *logging_offnum: OffsetNumber of current tuple being processed;
3683 : * used by vacuum's error callback system.
3684 : *
3685 : * Callers looking to verify that the page is already all-visible can call
3686 : * heap_page_is_all_visible().
3687 : *
3688 : * This logic is closely related to heap_prune_record_unchanged_lp_normal().
3689 : * If you modify this function, ensure consistency with that code. An
3690 : * assertion cross-checks that both remain in agreement. Do not introduce new
3691 : * side-effects.
3692 : */
3693 : static bool
3694 28286 : heap_page_would_be_all_visible(Relation rel, Buffer buf,
3695 : TransactionId OldestXmin,
3696 : OffsetNumber *deadoffsets,
3697 : int ndeadoffsets,
3698 : bool *all_frozen,
3699 : TransactionId *visibility_cutoff_xid,
3700 : OffsetNumber *logging_offnum)
3701 : {
3702 28286 : Page page = BufferGetPage(buf);
3703 28286 : BlockNumber blockno = BufferGetBlockNumber(buf);
3704 : OffsetNumber offnum,
3705 : maxoff;
3706 28286 : bool all_visible = true;
3707 28286 : int matched_dead_count = 0;
3708 :
3709 28286 : *visibility_cutoff_xid = InvalidTransactionId;
3710 28286 : *all_frozen = true;
3711 :
3712 : Assert(ndeadoffsets == 0 || deadoffsets);
3713 :
3714 : #ifdef USE_ASSERT_CHECKING
3715 : /* Confirm input deadoffsets[] is strictly sorted */
3716 : if (ndeadoffsets > 1)
3717 : {
3718 : for (int i = 1; i < ndeadoffsets; i++)
3719 : Assert(deadoffsets[i - 1] < deadoffsets[i]);
3720 : }
3721 : #endif
3722 :
3723 28286 : maxoff = PageGetMaxOffsetNumber(page);
3724 28286 : for (offnum = FirstOffsetNumber;
3725 2711170 : offnum <= maxoff && all_visible;
3726 2682884 : offnum = OffsetNumberNext(offnum))
3727 : {
3728 : ItemId itemid;
3729 : HeapTupleData tuple;
3730 :
3731 : /*
3732 : * Set the offset number so that we can display it along with any
3733 : * error that occurred while processing this tuple.
3734 : */
3735 2682886 : *logging_offnum = offnum;
3736 2682886 : itemid = PageGetItemId(page, offnum);
3737 :
3738 : /* Unused or redirect line pointers are of no interest */
3739 2682886 : if (!ItemIdIsUsed(itemid) || ItemIdIsRedirected(itemid))
3740 1776012 : continue;
3741 :
3742 2614160 : ItemPointerSet(&(tuple.t_self), blockno, offnum);
3743 :
3744 : /*
3745 : * Dead line pointers can have index pointers pointing to them. So
3746 : * they can't be treated as visible
3747 : */
3748 2614160 : if (ItemIdIsDead(itemid))
3749 : {
3750 1707288 : if (!deadoffsets ||
3751 1707288 : matched_dead_count >= ndeadoffsets ||
3752 1707288 : deadoffsets[matched_dead_count] != offnum)
3753 : {
3754 2 : *all_frozen = all_visible = false;
3755 2 : break;
3756 : }
3757 1707286 : matched_dead_count++;
3758 1707286 : continue;
3759 : }
3760 :
3761 : Assert(ItemIdIsNormal(itemid));
3762 :
3763 906872 : tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid);
3764 906872 : tuple.t_len = ItemIdGetLength(itemid);
3765 906872 : tuple.t_tableOid = RelationGetRelid(rel);
3766 :
3767 : /* Visibility checks may do IO or allocate memory */
3768 : Assert(CritSectionCount == 0);
3769 906872 : switch (HeapTupleSatisfiesVacuum(&tuple, OldestXmin, buf))
3770 : {
3771 906776 : case HEAPTUPLE_LIVE:
3772 : {
3773 : TransactionId xmin;
3774 :
3775 : /* Check comments in lazy_scan_prune. */
3776 906776 : if (!HeapTupleHeaderXminCommitted(tuple.t_data))
3777 : {
3778 0 : all_visible = false;
3779 0 : *all_frozen = false;
3780 0 : break;
3781 : }
3782 :
3783 : /*
3784 : * The inserter definitely committed. But is it old enough
3785 : * that everyone sees it as committed?
3786 : */
3787 906776 : xmin = HeapTupleHeaderGetXmin(tuple.t_data);
3788 906776 : if (!TransactionIdPrecedes(xmin, OldestXmin))
3789 : {
3790 46 : all_visible = false;
3791 46 : *all_frozen = false;
3792 46 : break;
3793 : }
3794 :
3795 : /* Track newest xmin on page. */
3796 906730 : if (TransactionIdFollows(xmin, *visibility_cutoff_xid) &&
3797 : TransactionIdIsNormal(xmin))
3798 21768 : *visibility_cutoff_xid = xmin;
3799 :
3800 : /* Check whether this tuple is already frozen or not */
3801 1176230 : if (all_visible && *all_frozen &&
3802 269500 : heap_tuple_needs_eventual_freeze(tuple.t_data))
3803 5706 : *all_frozen = false;
3804 : }
3805 906730 : break;
3806 :
3807 96 : case HEAPTUPLE_DEAD:
3808 : case HEAPTUPLE_RECENTLY_DEAD:
3809 : case HEAPTUPLE_INSERT_IN_PROGRESS:
3810 : case HEAPTUPLE_DELETE_IN_PROGRESS:
3811 : {
3812 96 : all_visible = false;
3813 96 : *all_frozen = false;
3814 96 : break;
3815 : }
3816 0 : default:
3817 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
3818 : break;
3819 : }
3820 : } /* scan along page */
3821 :
3822 : /* Clear the offset information once we have processed the given page. */
3823 28286 : *logging_offnum = InvalidOffsetNumber;
3824 :
3825 28286 : return all_visible;
3826 : }
3827 :
3828 : /*
3829 : * Update index statistics in pg_class if the statistics are accurate.
3830 : */
3831 : static void
3832 167842 : update_relstats_all_indexes(LVRelState *vacrel)
3833 : {
3834 167842 : Relation *indrels = vacrel->indrels;
3835 167842 : int nindexes = vacrel->nindexes;
3836 167842 : IndexBulkDeleteResult **indstats = vacrel->indstats;
3837 :
3838 : Assert(vacrel->do_index_cleanup);
3839 :
3840 418612 : for (int idx = 0; idx < nindexes; idx++)
3841 : {
3842 250770 : Relation indrel = indrels[idx];
3843 250770 : IndexBulkDeleteResult *istat = indstats[idx];
3844 :
3845 250770 : if (istat == NULL || istat->estimated_count)
3846 248088 : continue;
3847 :
3848 : /* Update index statistics */
3849 2682 : vac_update_relstats(indrel,
3850 : istat->num_pages,
3851 : istat->num_index_tuples,
3852 : 0, 0,
3853 : false,
3854 : InvalidTransactionId,
3855 : InvalidMultiXactId,
3856 : NULL, NULL, false);
3857 : }
3858 167842 : }
3859 :
3860 : /*
3861 : * Error context callback for errors occurring during vacuum. The error
3862 : * context messages for index phases should match the messages set in parallel
3863 : * vacuum. If you change this function for those phases, change
3864 : * parallel_vacuum_error_callback() as well.
3865 : */
3866 : static void
3867 271900 : vacuum_error_callback(void *arg)
3868 : {
3869 271900 : LVRelState *errinfo = arg;
3870 :
3871 271900 : switch (errinfo->phase)
3872 : {
3873 0 : case VACUUM_ERRCB_PHASE_SCAN_HEAP:
3874 0 : if (BlockNumberIsValid(errinfo->blkno))
3875 : {
3876 0 : if (OffsetNumberIsValid(errinfo->offnum))
3877 0 : errcontext("while scanning block %u offset %u of relation \"%s.%s\"",
3878 0 : errinfo->blkno, errinfo->offnum, errinfo->relnamespace, errinfo->relname);
3879 : else
3880 0 : errcontext("while scanning block %u of relation \"%s.%s\"",
3881 : errinfo->blkno, errinfo->relnamespace, errinfo->relname);
3882 : }
3883 : else
3884 0 : errcontext("while scanning relation \"%s.%s\"",
3885 : errinfo->relnamespace, errinfo->relname);
3886 0 : break;
3887 :
3888 0 : case VACUUM_ERRCB_PHASE_VACUUM_HEAP:
3889 0 : if (BlockNumberIsValid(errinfo->blkno))
3890 : {
3891 0 : if (OffsetNumberIsValid(errinfo->offnum))
3892 0 : errcontext("while vacuuming block %u offset %u of relation \"%s.%s\"",
3893 0 : errinfo->blkno, errinfo->offnum, errinfo->relnamespace, errinfo->relname);
3894 : else
3895 0 : errcontext("while vacuuming block %u of relation \"%s.%s\"",
3896 : errinfo->blkno, errinfo->relnamespace, errinfo->relname);
3897 : }
3898 : else
3899 0 : errcontext("while vacuuming relation \"%s.%s\"",
3900 : errinfo->relnamespace, errinfo->relname);
3901 0 : break;
3902 :
3903 0 : case VACUUM_ERRCB_PHASE_VACUUM_INDEX:
3904 0 : errcontext("while vacuuming index \"%s\" of relation \"%s.%s\"",
3905 : errinfo->indname, errinfo->relnamespace, errinfo->relname);
3906 0 : break;
3907 :
3908 0 : case VACUUM_ERRCB_PHASE_INDEX_CLEANUP:
3909 0 : errcontext("while cleaning up index \"%s\" of relation \"%s.%s\"",
3910 : errinfo->indname, errinfo->relnamespace, errinfo->relname);
3911 0 : break;
3912 :
3913 6 : case VACUUM_ERRCB_PHASE_TRUNCATE:
3914 6 : if (BlockNumberIsValid(errinfo->blkno))
3915 6 : errcontext("while truncating relation \"%s.%s\" to %u blocks",
3916 : errinfo->relnamespace, errinfo->relname, errinfo->blkno);
3917 6 : break;
3918 :
3919 271894 : case VACUUM_ERRCB_PHASE_UNKNOWN:
3920 : default:
3921 271894 : return; /* do nothing; the errinfo may not be
3922 : * initialized */
3923 : }
3924 : }
3925 :
3926 : /*
3927 : * Updates the information required for vacuum error callback. This also saves
3928 : * the current information which can be later restored via restore_vacuum_error_info.
3929 : */
3930 : static void
3931 1459566 : update_vacuum_error_info(LVRelState *vacrel, LVSavedErrInfo *saved_vacrel,
3932 : int phase, BlockNumber blkno, OffsetNumber offnum)
3933 : {
3934 1459566 : if (saved_vacrel)
3935 : {
3936 282602 : saved_vacrel->offnum = vacrel->offnum;
3937 282602 : saved_vacrel->blkno = vacrel->blkno;
3938 282602 : saved_vacrel->phase = vacrel->phase;
3939 : }
3940 :
3941 1459566 : vacrel->blkno = blkno;
3942 1459566 : vacrel->offnum = offnum;
3943 1459566 : vacrel->phase = phase;
3944 1459566 : }
3945 :
3946 : /*
3947 : * Restores the vacuum information saved via a prior call to update_vacuum_error_info.
3948 : */
3949 : static void
3950 282602 : restore_vacuum_error_info(LVRelState *vacrel,
3951 : const LVSavedErrInfo *saved_vacrel)
3952 : {
3953 282602 : vacrel->blkno = saved_vacrel->blkno;
3954 282602 : vacrel->offnum = saved_vacrel->offnum;
3955 282602 : vacrel->phase = saved_vacrel->phase;
3956 282602 : }
|
