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/dbcommands.h"
144 : #include "commands/progress.h"
145 : #include "commands/vacuum.h"
146 : #include "common/int.h"
147 : #include "common/pg_prng.h"
148 : #include "executor/instrument.h"
149 : #include "miscadmin.h"
150 : #include "pgstat.h"
151 : #include "portability/instr_time.h"
152 : #include "postmaster/autovacuum.h"
153 : #include "storage/bufmgr.h"
154 : #include "storage/freespace.h"
155 : #include "storage/lmgr.h"
156 : #include "storage/read_stream.h"
157 : #include "utils/lsyscache.h"
158 : #include "utils/pg_rusage.h"
159 : #include "utils/timestamp.h"
160 :
161 :
162 : /*
163 : * Space/time tradeoff parameters: do these need to be user-tunable?
164 : *
165 : * To consider truncating the relation, we want there to be at least
166 : * REL_TRUNCATE_MINIMUM or (relsize / REL_TRUNCATE_FRACTION) (whichever
167 : * is less) potentially-freeable pages.
168 : */
169 : #define REL_TRUNCATE_MINIMUM 1000
170 : #define REL_TRUNCATE_FRACTION 16
171 :
172 : /*
173 : * Timing parameters for truncate locking heuristics.
174 : *
175 : * These were not exposed as user tunable GUC values because it didn't seem
176 : * that the potential for improvement was great enough to merit the cost of
177 : * supporting them.
178 : */
179 : #define VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL 20 /* ms */
180 : #define VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL 50 /* ms */
181 : #define VACUUM_TRUNCATE_LOCK_TIMEOUT 5000 /* ms */
182 :
183 : /*
184 : * Threshold that controls whether we bypass index vacuuming and heap
185 : * vacuuming as an optimization
186 : */
187 : #define BYPASS_THRESHOLD_PAGES 0.02 /* i.e. 2% of rel_pages */
188 :
189 : /*
190 : * Perform a failsafe check each time we scan another 4GB of pages.
191 : * (Note that this is deliberately kept to a power-of-two, usually 2^19.)
192 : */
193 : #define FAILSAFE_EVERY_PAGES \
194 : ((BlockNumber) (((uint64) 4 * 1024 * 1024 * 1024) / BLCKSZ))
195 :
196 : /*
197 : * When a table has no indexes, vacuum the FSM after every 8GB, approximately
198 : * (it won't be exact because we only vacuum FSM after processing a heap page
199 : * that has some removable tuples). When there are indexes, this is ignored,
200 : * and we vacuum FSM after each index/heap cleaning pass.
201 : */
202 : #define VACUUM_FSM_EVERY_PAGES \
203 : ((BlockNumber) (((uint64) 8 * 1024 * 1024 * 1024) / BLCKSZ))
204 :
205 : /*
206 : * Before we consider skipping a page that's marked as clean in
207 : * visibility map, we must've seen at least this many clean pages.
208 : */
209 : #define SKIP_PAGES_THRESHOLD ((BlockNumber) 32)
210 :
211 : /*
212 : * Size of the prefetch window for lazy vacuum backwards truncation scan.
213 : * Needs to be a power of 2.
214 : */
215 : #define PREFETCH_SIZE ((BlockNumber) 32)
216 :
217 : /*
218 : * Macro to check if we are in a parallel vacuum. If true, we are in the
219 : * parallel mode and the DSM segment is initialized.
220 : */
221 : #define ParallelVacuumIsActive(vacrel) ((vacrel)->pvs != NULL)
222 :
223 : /* Phases of vacuum during which we report error context. */
224 : typedef enum
225 : {
226 : VACUUM_ERRCB_PHASE_UNKNOWN,
227 : VACUUM_ERRCB_PHASE_SCAN_HEAP,
228 : VACUUM_ERRCB_PHASE_VACUUM_INDEX,
229 : VACUUM_ERRCB_PHASE_VACUUM_HEAP,
230 : VACUUM_ERRCB_PHASE_INDEX_CLEANUP,
231 : VACUUM_ERRCB_PHASE_TRUNCATE,
232 : } VacErrPhase;
233 :
234 : /*
235 : * An eager scan of a page that is set all-frozen in the VM is considered
236 : * "successful". To spread out freezing overhead across multiple normal
237 : * vacuums, we limit the number of successful eager page freezes. The maximum
238 : * number of eager page freezes is calculated as a ratio of the all-visible
239 : * but not all-frozen pages at the beginning of the vacuum.
240 : */
241 : #define MAX_EAGER_FREEZE_SUCCESS_RATE 0.2
242 :
243 : /*
244 : * On the assumption that different regions of the table tend to have
245 : * similarly aged data, once vacuum fails to freeze
246 : * vacuum_max_eager_freeze_failure_rate of the blocks in a region of size
247 : * EAGER_SCAN_REGION_SIZE, it suspends eager scanning until it has progressed
248 : * to another region of the table with potentially older data.
249 : */
250 : #define EAGER_SCAN_REGION_SIZE 4096
251 :
252 : /*
253 : * heap_vac_scan_next_block() sets these flags to communicate information
254 : * about the block it read to the caller.
255 : */
256 : #define VAC_BLK_WAS_EAGER_SCANNED (1 << 0)
257 : #define VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM (1 << 1)
258 :
259 : typedef struct LVRelState
260 : {
261 : /* Target heap relation and its indexes */
262 : Relation rel;
263 : Relation *indrels;
264 : int nindexes;
265 :
266 : /* Buffer access strategy and parallel vacuum state */
267 : BufferAccessStrategy bstrategy;
268 : ParallelVacuumState *pvs;
269 :
270 : /* Aggressive VACUUM? (must set relfrozenxid >= FreezeLimit) */
271 : bool aggressive;
272 : /* Use visibility map to skip? (disabled by DISABLE_PAGE_SKIPPING) */
273 : bool skipwithvm;
274 : /* Consider index vacuuming bypass optimization? */
275 : bool consider_bypass_optimization;
276 :
277 : /* Doing index vacuuming, index cleanup, rel truncation? */
278 : bool do_index_vacuuming;
279 : bool do_index_cleanup;
280 : bool do_rel_truncate;
281 :
282 : /* VACUUM operation's cutoffs for freezing and pruning */
283 : struct VacuumCutoffs cutoffs;
284 : GlobalVisState *vistest;
285 : /* Tracks oldest extant XID/MXID for setting relfrozenxid/relminmxid */
286 : TransactionId NewRelfrozenXid;
287 : MultiXactId NewRelminMxid;
288 : bool skippedallvis;
289 :
290 : /* Error reporting state */
291 : char *dbname;
292 : char *relnamespace;
293 : char *relname;
294 : char *indname; /* Current index name */
295 : BlockNumber blkno; /* used only for heap operations */
296 : OffsetNumber offnum; /* used only for heap operations */
297 : VacErrPhase phase;
298 : bool verbose; /* VACUUM VERBOSE? */
299 :
300 : /*
301 : * dead_items stores TIDs whose index tuples are deleted by index
302 : * vacuuming. Each TID points to an LP_DEAD line pointer from a heap page
303 : * that has been processed by lazy_scan_prune. Also needed by
304 : * lazy_vacuum_heap_rel, which marks the same LP_DEAD line pointers as
305 : * LP_UNUSED during second heap pass.
306 : *
307 : * Both dead_items and dead_items_info are allocated in shared memory in
308 : * parallel vacuum cases.
309 : */
310 : TidStore *dead_items; /* TIDs whose index tuples we'll delete */
311 : VacDeadItemsInfo *dead_items_info;
312 :
313 : BlockNumber rel_pages; /* total number of pages */
314 : BlockNumber scanned_pages; /* # pages examined (not skipped via VM) */
315 :
316 : /*
317 : * Count of all-visible blocks eagerly scanned (for logging only). This
318 : * does not include skippable blocks scanned due to SKIP_PAGES_THRESHOLD.
319 : */
320 : BlockNumber eager_scanned_pages;
321 :
322 : BlockNumber removed_pages; /* # pages removed by relation truncation */
323 : BlockNumber new_frozen_tuple_pages; /* # pages with newly frozen tuples */
324 :
325 : /* # pages newly set all-visible in the VM */
326 : BlockNumber vm_new_visible_pages;
327 :
328 : /*
329 : * # pages newly set all-visible and all-frozen in the VM. This is a
330 : * subset of vm_new_visible_pages. That is, vm_new_visible_pages includes
331 : * all pages set all-visible, but vm_new_visible_frozen_pages includes
332 : * only those which were also set all-frozen.
333 : */
334 : BlockNumber vm_new_visible_frozen_pages;
335 :
336 : /* # all-visible pages newly set all-frozen in the VM */
337 : BlockNumber vm_new_frozen_pages;
338 :
339 : BlockNumber lpdead_item_pages; /* # pages with LP_DEAD items */
340 : BlockNumber missed_dead_pages; /* # pages with missed dead tuples */
341 : BlockNumber nonempty_pages; /* actually, last nonempty page + 1 */
342 :
343 : /* Statistics output by us, for table */
344 : double new_rel_tuples; /* new estimated total # of tuples */
345 : double new_live_tuples; /* new estimated total # of live tuples */
346 : /* Statistics output by index AMs */
347 : IndexBulkDeleteResult **indstats;
348 :
349 : /* Instrumentation counters */
350 : int num_index_scans;
351 : /* Counters that follow are only for scanned_pages */
352 : int64 tuples_deleted; /* # deleted from table */
353 : int64 tuples_frozen; /* # newly frozen */
354 : int64 lpdead_items; /* # deleted from indexes */
355 : int64 live_tuples; /* # live tuples remaining */
356 : int64 recently_dead_tuples; /* # dead, but not yet removable */
357 : int64 missed_dead_tuples; /* # removable, but not removed */
358 :
359 : /* State maintained by heap_vac_scan_next_block() */
360 : BlockNumber current_block; /* last block returned */
361 : BlockNumber next_unskippable_block; /* next unskippable block */
362 : bool next_unskippable_allvis; /* its visibility status */
363 : bool next_unskippable_eager_scanned; /* if it was eagerly scanned */
364 : Buffer next_unskippable_vmbuffer; /* buffer containing its VM bit */
365 :
366 : /* State related to managing eager scanning of all-visible pages */
367 :
368 : /*
369 : * A normal vacuum that has failed to freeze too many eagerly scanned
370 : * blocks in a region suspends eager scanning.
371 : * next_eager_scan_region_start is the block number of the first block
372 : * eligible for resumed eager scanning.
373 : *
374 : * When eager scanning is permanently disabled, either initially
375 : * (including for aggressive vacuum) or due to hitting the success cap,
376 : * this is set to InvalidBlockNumber.
377 : */
378 : BlockNumber next_eager_scan_region_start;
379 :
380 : /*
381 : * The remaining number of blocks a normal vacuum will consider eager
382 : * scanning when it is successful. When eager scanning is enabled, this is
383 : * initialized to MAX_EAGER_FREEZE_SUCCESS_RATE of the total number of
384 : * all-visible but not all-frozen pages. For each eager freeze success,
385 : * this is decremented. Once it hits 0, eager scanning is permanently
386 : * disabled. It is initialized to 0 if eager scanning starts out disabled
387 : * (including for aggressive vacuum).
388 : */
389 : BlockNumber eager_scan_remaining_successes;
390 :
391 : /*
392 : * The maximum number of blocks which may be eagerly scanned and not
393 : * frozen before eager scanning is temporarily suspended. This is
394 : * configurable both globally, via the
395 : * vacuum_max_eager_freeze_failure_rate GUC, and per table, with a table
396 : * storage parameter of the same name. It is calculated as
397 : * vacuum_max_eager_freeze_failure_rate of EAGER_SCAN_REGION_SIZE blocks.
398 : * It is 0 when eager scanning is disabled.
399 : */
400 : BlockNumber eager_scan_max_fails_per_region;
401 :
402 : /*
403 : * The number of eagerly scanned blocks vacuum failed to freeze (due to
404 : * age) in the current eager scan region. Vacuum resets it to
405 : * eager_scan_max_fails_per_region each time it enters a new region of the
406 : * relation. If eager_scan_remaining_fails hits 0, eager scanning is
407 : * suspended until the next region. It is also 0 if eager scanning has
408 : * been permanently disabled.
409 : */
410 : BlockNumber eager_scan_remaining_fails;
411 : } LVRelState;
412 :
413 :
414 : /* Struct for saving and restoring vacuum error information. */
415 : typedef struct LVSavedErrInfo
416 : {
417 : BlockNumber blkno;
418 : OffsetNumber offnum;
419 : VacErrPhase phase;
420 : } LVSavedErrInfo;
421 :
422 :
423 : /* non-export function prototypes */
424 : static void lazy_scan_heap(LVRelState *vacrel);
425 : static void heap_vacuum_eager_scan_setup(LVRelState *vacrel,
426 : const VacuumParams params);
427 : static BlockNumber heap_vac_scan_next_block(ReadStream *stream,
428 : void *callback_private_data,
429 : void *per_buffer_data);
430 : static void find_next_unskippable_block(LVRelState *vacrel, bool *skipsallvis);
431 : static bool lazy_scan_new_or_empty(LVRelState *vacrel, Buffer buf,
432 : BlockNumber blkno, Page page,
433 : bool sharelock, Buffer vmbuffer);
434 : static int lazy_scan_prune(LVRelState *vacrel, Buffer buf,
435 : BlockNumber blkno, Page page,
436 : Buffer vmbuffer, bool all_visible_according_to_vm,
437 : bool *has_lpdead_items, bool *vm_page_frozen);
438 : static bool lazy_scan_noprune(LVRelState *vacrel, Buffer buf,
439 : BlockNumber blkno, Page page,
440 : bool *has_lpdead_items);
441 : static void lazy_vacuum(LVRelState *vacrel);
442 : static bool lazy_vacuum_all_indexes(LVRelState *vacrel);
443 : static void lazy_vacuum_heap_rel(LVRelState *vacrel);
444 : static void lazy_vacuum_heap_page(LVRelState *vacrel, BlockNumber blkno,
445 : Buffer buffer, OffsetNumber *deadoffsets,
446 : int num_offsets, Buffer vmbuffer);
447 : static bool lazy_check_wraparound_failsafe(LVRelState *vacrel);
448 : static void lazy_cleanup_all_indexes(LVRelState *vacrel);
449 : static IndexBulkDeleteResult *lazy_vacuum_one_index(Relation indrel,
450 : IndexBulkDeleteResult *istat,
451 : double reltuples,
452 : LVRelState *vacrel);
453 : static IndexBulkDeleteResult *lazy_cleanup_one_index(Relation indrel,
454 : IndexBulkDeleteResult *istat,
455 : double reltuples,
456 : bool estimated_count,
457 : LVRelState *vacrel);
458 : static bool should_attempt_truncation(LVRelState *vacrel);
459 : static void lazy_truncate_heap(LVRelState *vacrel);
460 : static BlockNumber count_nondeletable_pages(LVRelState *vacrel,
461 : bool *lock_waiter_detected);
462 : static void dead_items_alloc(LVRelState *vacrel, int nworkers);
463 : static void dead_items_add(LVRelState *vacrel, BlockNumber blkno, OffsetNumber *offsets,
464 : int num_offsets);
465 : static void dead_items_reset(LVRelState *vacrel);
466 : static void dead_items_cleanup(LVRelState *vacrel);
467 : static bool heap_page_is_all_visible(LVRelState *vacrel, Buffer buf,
468 : TransactionId *visibility_cutoff_xid, bool *all_frozen);
469 : static void update_relstats_all_indexes(LVRelState *vacrel);
470 : static void vacuum_error_callback(void *arg);
471 : static void update_vacuum_error_info(LVRelState *vacrel,
472 : LVSavedErrInfo *saved_vacrel,
473 : int phase, BlockNumber blkno,
474 : OffsetNumber offnum);
475 : static void restore_vacuum_error_info(LVRelState *vacrel,
476 : const LVSavedErrInfo *saved_vacrel);
477 :
478 :
479 :
480 : /*
481 : * Helper to set up the eager scanning state for vacuuming a single relation.
482 : * Initializes the eager scan management related members of the LVRelState.
483 : *
484 : * Caller provides whether or not an aggressive vacuum is required due to
485 : * vacuum options or for relfrozenxid/relminmxid advancement.
486 : */
487 : static void
488 245382 : heap_vacuum_eager_scan_setup(LVRelState *vacrel, const VacuumParams params)
489 : {
490 : uint32 randseed;
491 : BlockNumber allvisible;
492 : BlockNumber allfrozen;
493 : float first_region_ratio;
494 245382 : bool oldest_unfrozen_before_cutoff = false;
495 :
496 : /*
497 : * Initialize eager scan management fields to their disabled values.
498 : * Aggressive vacuums, normal vacuums of small tables, and normal vacuums
499 : * of tables without sufficiently old tuples disable eager scanning.
500 : */
501 245382 : vacrel->next_eager_scan_region_start = InvalidBlockNumber;
502 245382 : vacrel->eager_scan_max_fails_per_region = 0;
503 245382 : vacrel->eager_scan_remaining_fails = 0;
504 245382 : vacrel->eager_scan_remaining_successes = 0;
505 :
506 : /* If eager scanning is explicitly disabled, just return. */
507 245382 : if (params.max_eager_freeze_failure_rate == 0)
508 245382 : return;
509 :
510 : /*
511 : * The caller will have determined whether or not an aggressive vacuum is
512 : * required by either the vacuum parameters or the relative age of the
513 : * oldest unfrozen transaction IDs. An aggressive vacuum must scan every
514 : * all-visible page to safely advance the relfrozenxid and/or relminmxid,
515 : * so scans of all-visible pages are not considered eager.
516 : */
517 245382 : if (vacrel->aggressive)
518 234608 : return;
519 :
520 : /*
521 : * Aggressively vacuuming a small relation shouldn't take long, so it
522 : * isn't worth amortizing. We use two times the region size as the size
523 : * cutoff because the eager scan start block is a random spot somewhere in
524 : * the first region, making the second region the first to be eager
525 : * scanned normally.
526 : */
527 10774 : if (vacrel->rel_pages < 2 * EAGER_SCAN_REGION_SIZE)
528 10774 : return;
529 :
530 : /*
531 : * We only want to enable eager scanning if we are likely to be able to
532 : * freeze some of the pages in the relation.
533 : *
534 : * Tuples with XIDs older than OldestXmin or MXIDs older than OldestMxact
535 : * are technically freezable, but we won't freeze them unless the criteria
536 : * for opportunistic freezing is met. Only tuples with XIDs/MXIDs older
537 : * than the FreezeLimit/MultiXactCutoff are frozen in the common case.
538 : *
539 : * So, as a heuristic, we wait until the FreezeLimit has advanced past the
540 : * relfrozenxid or the MultiXactCutoff has advanced past the relminmxid to
541 : * enable eager scanning.
542 : */
543 0 : if (TransactionIdIsNormal(vacrel->cutoffs.relfrozenxid) &&
544 0 : TransactionIdPrecedes(vacrel->cutoffs.relfrozenxid,
545 : vacrel->cutoffs.FreezeLimit))
546 0 : oldest_unfrozen_before_cutoff = true;
547 :
548 0 : if (!oldest_unfrozen_before_cutoff &&
549 0 : MultiXactIdIsValid(vacrel->cutoffs.relminmxid) &&
550 0 : MultiXactIdPrecedes(vacrel->cutoffs.relminmxid,
551 : vacrel->cutoffs.MultiXactCutoff))
552 0 : oldest_unfrozen_before_cutoff = true;
553 :
554 0 : if (!oldest_unfrozen_before_cutoff)
555 0 : return;
556 :
557 : /* We have met the criteria to eagerly scan some pages. */
558 :
559 : /*
560 : * Our success cap is MAX_EAGER_FREEZE_SUCCESS_RATE of the number of
561 : * all-visible but not all-frozen blocks in the relation.
562 : */
563 0 : visibilitymap_count(vacrel->rel, &allvisible, &allfrozen);
564 :
565 0 : vacrel->eager_scan_remaining_successes =
566 0 : (BlockNumber) (MAX_EAGER_FREEZE_SUCCESS_RATE *
567 0 : (allvisible - allfrozen));
568 :
569 : /* If every all-visible page is frozen, eager scanning is disabled. */
570 0 : if (vacrel->eager_scan_remaining_successes == 0)
571 0 : return;
572 :
573 : /*
574 : * Now calculate the bounds of the first eager scan region. Its end block
575 : * will be a random spot somewhere in the first EAGER_SCAN_REGION_SIZE
576 : * blocks. This affects the bounds of all subsequent regions and avoids
577 : * eager scanning and failing to freeze the same blocks each vacuum of the
578 : * relation.
579 : */
580 0 : randseed = pg_prng_uint32(&pg_global_prng_state);
581 :
582 0 : vacrel->next_eager_scan_region_start = randseed % EAGER_SCAN_REGION_SIZE;
583 :
584 : Assert(params.max_eager_freeze_failure_rate > 0 &&
585 : params.max_eager_freeze_failure_rate <= 1);
586 :
587 0 : vacrel->eager_scan_max_fails_per_region =
588 0 : params.max_eager_freeze_failure_rate *
589 : EAGER_SCAN_REGION_SIZE;
590 :
591 : /*
592 : * The first region will be smaller than subsequent regions. As such,
593 : * adjust the eager freeze failures tolerated for this region.
594 : */
595 0 : first_region_ratio = 1 - (float) vacrel->next_eager_scan_region_start /
596 : EAGER_SCAN_REGION_SIZE;
597 :
598 0 : vacrel->eager_scan_remaining_fails =
599 0 : vacrel->eager_scan_max_fails_per_region *
600 : first_region_ratio;
601 : }
602 :
603 : /*
604 : * heap_vacuum_rel() -- perform VACUUM for one heap relation
605 : *
606 : * This routine sets things up for and then calls lazy_scan_heap, where
607 : * almost all work actually takes place. Finalizes everything after call
608 : * returns by managing relation truncation and updating rel's pg_class
609 : * entry. (Also updates pg_class entries for any indexes that need it.)
610 : *
611 : * At entry, we have already established a transaction and opened
612 : * and locked the relation.
613 : */
614 : void
615 245382 : heap_vacuum_rel(Relation rel, const VacuumParams params,
616 : BufferAccessStrategy bstrategy)
617 : {
618 : LVRelState *vacrel;
619 : bool verbose,
620 : instrument,
621 : skipwithvm,
622 : frozenxid_updated,
623 : minmulti_updated;
624 : BlockNumber orig_rel_pages,
625 : new_rel_pages,
626 : new_rel_allvisible,
627 : new_rel_allfrozen;
628 : PGRUsage ru0;
629 245382 : TimestampTz starttime = 0;
630 245382 : PgStat_Counter startreadtime = 0,
631 245382 : startwritetime = 0;
632 245382 : WalUsage startwalusage = pgWalUsage;
633 245382 : BufferUsage startbufferusage = pgBufferUsage;
634 : ErrorContextCallback errcallback;
635 245382 : char **indnames = NULL;
636 :
637 245382 : verbose = (params.options & VACOPT_VERBOSE) != 0;
638 464576 : instrument = (verbose || (AmAutoVacuumWorkerProcess() &&
639 219194 : params.log_min_duration >= 0));
640 245382 : if (instrument)
641 : {
642 219214 : pg_rusage_init(&ru0);
643 219214 : if (track_io_timing)
644 : {
645 0 : startreadtime = pgStatBlockReadTime;
646 0 : startwritetime = pgStatBlockWriteTime;
647 : }
648 : }
649 :
650 : /* Used for instrumentation and stats report */
651 245382 : starttime = GetCurrentTimestamp();
652 :
653 245382 : pgstat_progress_start_command(PROGRESS_COMMAND_VACUUM,
654 : RelationGetRelid(rel));
655 :
656 : /*
657 : * Setup error traceback support for ereport() first. The idea is to set
658 : * up an error context callback to display additional information on any
659 : * error during a vacuum. During different phases of vacuum, we update
660 : * the state so that the error context callback always display current
661 : * information.
662 : *
663 : * Copy the names of heap rel into local memory for error reporting
664 : * purposes, too. It isn't always safe to assume that we can get the name
665 : * of each rel. It's convenient for code in lazy_scan_heap to always use
666 : * these temp copies.
667 : */
668 245382 : vacrel = (LVRelState *) palloc0(sizeof(LVRelState));
669 245382 : vacrel->dbname = get_database_name(MyDatabaseId);
670 245382 : vacrel->relnamespace = get_namespace_name(RelationGetNamespace(rel));
671 245382 : vacrel->relname = pstrdup(RelationGetRelationName(rel));
672 245382 : vacrel->indname = NULL;
673 245382 : vacrel->phase = VACUUM_ERRCB_PHASE_UNKNOWN;
674 245382 : vacrel->verbose = verbose;
675 245382 : errcallback.callback = vacuum_error_callback;
676 245382 : errcallback.arg = vacrel;
677 245382 : errcallback.previous = error_context_stack;
678 245382 : error_context_stack = &errcallback;
679 :
680 : /* Set up high level stuff about rel and its indexes */
681 245382 : vacrel->rel = rel;
682 245382 : vac_open_indexes(vacrel->rel, RowExclusiveLock, &vacrel->nindexes,
683 : &vacrel->indrels);
684 245382 : vacrel->bstrategy = bstrategy;
685 245382 : if (instrument && vacrel->nindexes > 0)
686 : {
687 : /* Copy index names used by instrumentation (not error reporting) */
688 209586 : indnames = palloc(sizeof(char *) * vacrel->nindexes);
689 540056 : for (int i = 0; i < vacrel->nindexes; i++)
690 330470 : indnames[i] = pstrdup(RelationGetRelationName(vacrel->indrels[i]));
691 : }
692 :
693 : /*
694 : * The index_cleanup param either disables index vacuuming and cleanup or
695 : * forces it to go ahead when we would otherwise apply the index bypass
696 : * optimization. The default is 'auto', which leaves the final decision
697 : * up to lazy_vacuum().
698 : *
699 : * The truncate param allows user to avoid attempting relation truncation,
700 : * though it can't force truncation to happen.
701 : */
702 : Assert(params.index_cleanup != VACOPTVALUE_UNSPECIFIED);
703 : Assert(params.truncate != VACOPTVALUE_UNSPECIFIED &&
704 : params.truncate != VACOPTVALUE_AUTO);
705 :
706 : /*
707 : * While VacuumFailSafeActive is reset to false before calling this, we
708 : * still need to reset it here due to recursive calls.
709 : */
710 245382 : VacuumFailsafeActive = false;
711 245382 : vacrel->consider_bypass_optimization = true;
712 245382 : vacrel->do_index_vacuuming = true;
713 245382 : vacrel->do_index_cleanup = true;
714 245382 : vacrel->do_rel_truncate = (params.truncate != VACOPTVALUE_DISABLED);
715 245382 : if (params.index_cleanup == VACOPTVALUE_DISABLED)
716 : {
717 : /* Force disable index vacuuming up-front */
718 260 : vacrel->do_index_vacuuming = false;
719 260 : vacrel->do_index_cleanup = false;
720 : }
721 245122 : else if (params.index_cleanup == VACOPTVALUE_ENABLED)
722 : {
723 : /* Force index vacuuming. Note that failsafe can still bypass. */
724 30 : vacrel->consider_bypass_optimization = false;
725 : }
726 : else
727 : {
728 : /* Default/auto, make all decisions dynamically */
729 : Assert(params.index_cleanup == VACOPTVALUE_AUTO);
730 : }
731 :
732 : /* Initialize page counters explicitly (be tidy) */
733 245382 : vacrel->scanned_pages = 0;
734 245382 : vacrel->eager_scanned_pages = 0;
735 245382 : vacrel->removed_pages = 0;
736 245382 : vacrel->new_frozen_tuple_pages = 0;
737 245382 : vacrel->lpdead_item_pages = 0;
738 245382 : vacrel->missed_dead_pages = 0;
739 245382 : vacrel->nonempty_pages = 0;
740 : /* dead_items_alloc allocates vacrel->dead_items later on */
741 :
742 : /* Allocate/initialize output statistics state */
743 245382 : vacrel->new_rel_tuples = 0;
744 245382 : vacrel->new_live_tuples = 0;
745 245382 : vacrel->indstats = (IndexBulkDeleteResult **)
746 245382 : palloc0(vacrel->nindexes * sizeof(IndexBulkDeleteResult *));
747 :
748 : /* Initialize remaining counters (be tidy) */
749 245382 : vacrel->num_index_scans = 0;
750 245382 : vacrel->tuples_deleted = 0;
751 245382 : vacrel->tuples_frozen = 0;
752 245382 : vacrel->lpdead_items = 0;
753 245382 : vacrel->live_tuples = 0;
754 245382 : vacrel->recently_dead_tuples = 0;
755 245382 : vacrel->missed_dead_tuples = 0;
756 :
757 245382 : vacrel->vm_new_visible_pages = 0;
758 245382 : vacrel->vm_new_visible_frozen_pages = 0;
759 245382 : vacrel->vm_new_frozen_pages = 0;
760 :
761 : /*
762 : * Get cutoffs that determine which deleted tuples are considered DEAD,
763 : * not just RECENTLY_DEAD, and which XIDs/MXIDs to freeze. Then determine
764 : * the extent of the blocks that we'll scan in lazy_scan_heap. It has to
765 : * happen in this order to ensure that the OldestXmin cutoff field works
766 : * as an upper bound on the XIDs stored in the pages we'll actually scan
767 : * (NewRelfrozenXid tracking must never be allowed to miss unfrozen XIDs).
768 : *
769 : * Next acquire vistest, a related cutoff that's used in pruning. We use
770 : * vistest in combination with OldestXmin to ensure that
771 : * heap_page_prune_and_freeze() always removes any deleted tuple whose
772 : * xmax is < OldestXmin. lazy_scan_prune must never become confused about
773 : * whether a tuple should be frozen or removed. (In the future we might
774 : * want to teach lazy_scan_prune to recompute vistest from time to time,
775 : * to increase the number of dead tuples it can prune away.)
776 : */
777 245382 : vacrel->aggressive = vacuum_get_cutoffs(rel, params, &vacrel->cutoffs);
778 245382 : vacrel->rel_pages = orig_rel_pages = RelationGetNumberOfBlocks(rel);
779 245382 : vacrel->vistest = GlobalVisTestFor(rel);
780 :
781 : /* Initialize state used to track oldest extant XID/MXID */
782 245382 : vacrel->NewRelfrozenXid = vacrel->cutoffs.OldestXmin;
783 245382 : vacrel->NewRelminMxid = vacrel->cutoffs.OldestMxact;
784 :
785 : /*
786 : * Initialize state related to tracking all-visible page skipping. This is
787 : * very important to determine whether or not it is safe to advance the
788 : * relfrozenxid/relminmxid.
789 : */
790 245382 : vacrel->skippedallvis = false;
791 245382 : skipwithvm = true;
792 245382 : if (params.options & VACOPT_DISABLE_PAGE_SKIPPING)
793 : {
794 : /*
795 : * Force aggressive mode, and disable skipping blocks using the
796 : * visibility map (even those set all-frozen)
797 : */
798 344 : vacrel->aggressive = true;
799 344 : skipwithvm = false;
800 : }
801 :
802 245382 : vacrel->skipwithvm = skipwithvm;
803 :
804 : /*
805 : * Set up eager scan tracking state. This must happen after determining
806 : * whether or not the vacuum must be aggressive, because only normal
807 : * vacuums use the eager scan algorithm.
808 : */
809 245382 : heap_vacuum_eager_scan_setup(vacrel, params);
810 :
811 245382 : if (verbose)
812 : {
813 20 : if (vacrel->aggressive)
814 2 : ereport(INFO,
815 : (errmsg("aggressively vacuuming \"%s.%s.%s\"",
816 : vacrel->dbname, vacrel->relnamespace,
817 : vacrel->relname)));
818 : else
819 18 : ereport(INFO,
820 : (errmsg("vacuuming \"%s.%s.%s\"",
821 : vacrel->dbname, vacrel->relnamespace,
822 : vacrel->relname)));
823 : }
824 :
825 : /*
826 : * Allocate dead_items memory using dead_items_alloc. This handles
827 : * parallel VACUUM initialization as part of allocating shared memory
828 : * space used for dead_items. (But do a failsafe precheck first, to
829 : * ensure that parallel VACUUM won't be attempted at all when relfrozenxid
830 : * is already dangerously old.)
831 : */
832 245382 : lazy_check_wraparound_failsafe(vacrel);
833 245382 : dead_items_alloc(vacrel, params.nworkers);
834 :
835 : /*
836 : * Call lazy_scan_heap to perform all required heap pruning, index
837 : * vacuuming, and heap vacuuming (plus related processing)
838 : */
839 245382 : lazy_scan_heap(vacrel);
840 :
841 : /*
842 : * Free resources managed by dead_items_alloc. This ends parallel mode in
843 : * passing when necessary.
844 : */
845 245382 : dead_items_cleanup(vacrel);
846 : Assert(!IsInParallelMode());
847 :
848 : /*
849 : * Update pg_class entries for each of rel's indexes where appropriate.
850 : *
851 : * Unlike the later update to rel's pg_class entry, this is not critical.
852 : * Maintains relpages/reltuples statistics used by the planner only.
853 : */
854 245382 : if (vacrel->do_index_cleanup)
855 177494 : update_relstats_all_indexes(vacrel);
856 :
857 : /* Done with rel's indexes */
858 245382 : vac_close_indexes(vacrel->nindexes, vacrel->indrels, NoLock);
859 :
860 : /* Optionally truncate rel */
861 245382 : if (should_attempt_truncation(vacrel))
862 330 : lazy_truncate_heap(vacrel);
863 :
864 : /* Pop the error context stack */
865 245382 : error_context_stack = errcallback.previous;
866 :
867 : /* Report that we are now doing final cleanup */
868 245382 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
869 : PROGRESS_VACUUM_PHASE_FINAL_CLEANUP);
870 :
871 : /*
872 : * Prepare to update rel's pg_class entry.
873 : *
874 : * Aggressive VACUUMs must always be able to advance relfrozenxid to a
875 : * value >= FreezeLimit, and relminmxid to a value >= MultiXactCutoff.
876 : * Non-aggressive VACUUMs may advance them by any amount, or not at all.
877 : */
878 : Assert(vacrel->NewRelfrozenXid == vacrel->cutoffs.OldestXmin ||
879 : TransactionIdPrecedesOrEquals(vacrel->aggressive ? vacrel->cutoffs.FreezeLimit :
880 : vacrel->cutoffs.relfrozenxid,
881 : vacrel->NewRelfrozenXid));
882 : Assert(vacrel->NewRelminMxid == vacrel->cutoffs.OldestMxact ||
883 : MultiXactIdPrecedesOrEquals(vacrel->aggressive ? vacrel->cutoffs.MultiXactCutoff :
884 : vacrel->cutoffs.relminmxid,
885 : vacrel->NewRelminMxid));
886 245382 : if (vacrel->skippedallvis)
887 : {
888 : /*
889 : * Must keep original relfrozenxid in a non-aggressive VACUUM that
890 : * chose to skip an all-visible page range. The state that tracks new
891 : * values will have missed unfrozen XIDs from the pages we skipped.
892 : */
893 : Assert(!vacrel->aggressive);
894 70 : vacrel->NewRelfrozenXid = InvalidTransactionId;
895 70 : vacrel->NewRelminMxid = InvalidMultiXactId;
896 : }
897 :
898 : /*
899 : * For safety, clamp relallvisible to be not more than what we're setting
900 : * pg_class.relpages to
901 : */
902 245382 : new_rel_pages = vacrel->rel_pages; /* After possible rel truncation */
903 245382 : visibilitymap_count(rel, &new_rel_allvisible, &new_rel_allfrozen);
904 245382 : if (new_rel_allvisible > new_rel_pages)
905 0 : new_rel_allvisible = new_rel_pages;
906 :
907 : /*
908 : * An all-frozen block _must_ be all-visible. As such, clamp the count of
909 : * all-frozen blocks to the count of all-visible blocks. This matches the
910 : * clamping of relallvisible above.
911 : */
912 245382 : if (new_rel_allfrozen > new_rel_allvisible)
913 0 : new_rel_allfrozen = new_rel_allvisible;
914 :
915 : /*
916 : * Now actually update rel's pg_class entry.
917 : *
918 : * In principle new_live_tuples could be -1 indicating that we (still)
919 : * don't know the tuple count. In practice that can't happen, since we
920 : * scan every page that isn't skipped using the visibility map.
921 : */
922 245382 : vac_update_relstats(rel, new_rel_pages, vacrel->new_live_tuples,
923 : new_rel_allvisible, new_rel_allfrozen,
924 245382 : vacrel->nindexes > 0,
925 : vacrel->NewRelfrozenXid, vacrel->NewRelminMxid,
926 : &frozenxid_updated, &minmulti_updated, false);
927 :
928 : /*
929 : * Report results to the cumulative stats system, too.
930 : *
931 : * Deliberately avoid telling the stats system about LP_DEAD items that
932 : * remain in the table due to VACUUM bypassing index and heap vacuuming.
933 : * ANALYZE will consider the remaining LP_DEAD items to be dead "tuples".
934 : * It seems like a good idea to err on the side of not vacuuming again too
935 : * soon in cases where the failsafe prevented significant amounts of heap
936 : * vacuuming.
937 : */
938 147364 : pgstat_report_vacuum(RelationGetRelid(rel),
939 245382 : rel->rd_rel->relisshared,
940 98018 : Max(vacrel->new_live_tuples, 0),
941 245382 : vacrel->recently_dead_tuples +
942 245382 : vacrel->missed_dead_tuples,
943 : starttime);
944 245382 : pgstat_progress_end_command();
945 :
946 245382 : if (instrument)
947 : {
948 219214 : TimestampTz endtime = GetCurrentTimestamp();
949 :
950 219496 : if (verbose || params.log_min_duration == 0 ||
951 282 : TimestampDifferenceExceeds(starttime, endtime,
952 282 : params.log_min_duration))
953 : {
954 : long secs_dur;
955 : int usecs_dur;
956 : WalUsage walusage;
957 : BufferUsage bufferusage;
958 : StringInfoData buf;
959 : char *msgfmt;
960 : int32 diff;
961 218932 : double read_rate = 0,
962 218932 : write_rate = 0;
963 : int64 total_blks_hit;
964 : int64 total_blks_read;
965 : int64 total_blks_dirtied;
966 :
967 218932 : TimestampDifference(starttime, endtime, &secs_dur, &usecs_dur);
968 218932 : memset(&walusage, 0, sizeof(WalUsage));
969 218932 : WalUsageAccumDiff(&walusage, &pgWalUsage, &startwalusage);
970 218932 : memset(&bufferusage, 0, sizeof(BufferUsage));
971 218932 : BufferUsageAccumDiff(&bufferusage, &pgBufferUsage, &startbufferusage);
972 :
973 218932 : total_blks_hit = bufferusage.shared_blks_hit +
974 218932 : bufferusage.local_blks_hit;
975 218932 : total_blks_read = bufferusage.shared_blks_read +
976 218932 : bufferusage.local_blks_read;
977 218932 : total_blks_dirtied = bufferusage.shared_blks_dirtied +
978 218932 : bufferusage.local_blks_dirtied;
979 :
980 218932 : initStringInfo(&buf);
981 218932 : if (verbose)
982 : {
983 : /*
984 : * Aggressiveness already reported earlier, in dedicated
985 : * VACUUM VERBOSE ereport
986 : */
987 : Assert(!params.is_wraparound);
988 20 : msgfmt = _("finished vacuuming \"%s.%s.%s\": index scans: %d\n");
989 : }
990 218912 : else if (params.is_wraparound)
991 : {
992 : /*
993 : * While it's possible for a VACUUM to be both is_wraparound
994 : * and !aggressive, that's just a corner-case -- is_wraparound
995 : * implies aggressive. Produce distinct output for the corner
996 : * case all the same, just in case.
997 : */
998 218874 : if (vacrel->aggressive)
999 218858 : msgfmt = _("automatic aggressive vacuum to prevent wraparound of table \"%s.%s.%s\": index scans: %d\n");
1000 : else
1001 16 : msgfmt = _("automatic vacuum to prevent wraparound of table \"%s.%s.%s\": index scans: %d\n");
1002 : }
1003 : else
1004 : {
1005 38 : if (vacrel->aggressive)
1006 30 : msgfmt = _("automatic aggressive vacuum of table \"%s.%s.%s\": index scans: %d\n");
1007 : else
1008 8 : msgfmt = _("automatic vacuum of table \"%s.%s.%s\": index scans: %d\n");
1009 : }
1010 218932 : appendStringInfo(&buf, msgfmt,
1011 : vacrel->dbname,
1012 : vacrel->relnamespace,
1013 : vacrel->relname,
1014 : vacrel->num_index_scans);
1015 305898 : appendStringInfo(&buf, _("pages: %u removed, %u remain, %u scanned (%.2f%% of total), %u eagerly scanned\n"),
1016 : vacrel->removed_pages,
1017 : new_rel_pages,
1018 : vacrel->scanned_pages,
1019 : orig_rel_pages == 0 ? 100.0 :
1020 86966 : 100.0 * vacrel->scanned_pages /
1021 : orig_rel_pages,
1022 : vacrel->eager_scanned_pages);
1023 218932 : appendStringInfo(&buf,
1024 218932 : _("tuples: %" PRId64 " removed, %" PRId64 " remain, %" PRId64 " are dead but not yet removable\n"),
1025 : vacrel->tuples_deleted,
1026 218932 : (int64) vacrel->new_rel_tuples,
1027 : vacrel->recently_dead_tuples);
1028 218932 : if (vacrel->missed_dead_tuples > 0)
1029 0 : appendStringInfo(&buf,
1030 0 : _("tuples missed: %" PRId64 " dead from %u pages not removed due to cleanup lock contention\n"),
1031 : vacrel->missed_dead_tuples,
1032 : vacrel->missed_dead_pages);
1033 218932 : diff = (int32) (ReadNextTransactionId() -
1034 218932 : vacrel->cutoffs.OldestXmin);
1035 218932 : appendStringInfo(&buf,
1036 218932 : _("removable cutoff: %u, which was %d XIDs old when operation ended\n"),
1037 : vacrel->cutoffs.OldestXmin, diff);
1038 218932 : if (frozenxid_updated)
1039 : {
1040 39704 : diff = (int32) (vacrel->NewRelfrozenXid -
1041 39704 : vacrel->cutoffs.relfrozenxid);
1042 39704 : appendStringInfo(&buf,
1043 39704 : _("new relfrozenxid: %u, which is %d XIDs ahead of previous value\n"),
1044 : vacrel->NewRelfrozenXid, diff);
1045 : }
1046 218932 : if (minmulti_updated)
1047 : {
1048 8 : diff = (int32) (vacrel->NewRelminMxid -
1049 8 : vacrel->cutoffs.relminmxid);
1050 8 : appendStringInfo(&buf,
1051 8 : _("new relminmxid: %u, which is %d MXIDs ahead of previous value\n"),
1052 : vacrel->NewRelminMxid, diff);
1053 : }
1054 305898 : appendStringInfo(&buf, _("frozen: %u pages from table (%.2f%% of total) had %" PRId64 " tuples frozen\n"),
1055 : vacrel->new_frozen_tuple_pages,
1056 : orig_rel_pages == 0 ? 100.0 :
1057 86966 : 100.0 * vacrel->new_frozen_tuple_pages /
1058 : orig_rel_pages,
1059 : vacrel->tuples_frozen);
1060 :
1061 218932 : appendStringInfo(&buf,
1062 218932 : _("visibility map: %u pages set all-visible, %u pages set all-frozen (%u were all-visible)\n"),
1063 : vacrel->vm_new_visible_pages,
1064 218932 : vacrel->vm_new_visible_frozen_pages +
1065 218932 : vacrel->vm_new_frozen_pages,
1066 : vacrel->vm_new_frozen_pages);
1067 218932 : if (vacrel->do_index_vacuuming)
1068 : {
1069 151522 : if (vacrel->nindexes == 0 || vacrel->num_index_scans == 0)
1070 151492 : appendStringInfoString(&buf, _("index scan not needed: "));
1071 : else
1072 30 : appendStringInfoString(&buf, _("index scan needed: "));
1073 :
1074 151522 : msgfmt = _("%u pages from table (%.2f%% of total) had %" PRId64 " dead item identifiers removed\n");
1075 : }
1076 : else
1077 : {
1078 67410 : if (!VacuumFailsafeActive)
1079 0 : appendStringInfoString(&buf, _("index scan bypassed: "));
1080 : else
1081 67410 : appendStringInfoString(&buf, _("index scan bypassed by failsafe: "));
1082 :
1083 67410 : msgfmt = _("%u pages from table (%.2f%% of total) have %" PRId64 " dead item identifiers\n");
1084 : }
1085 305898 : appendStringInfo(&buf, msgfmt,
1086 : vacrel->lpdead_item_pages,
1087 : orig_rel_pages == 0 ? 100.0 :
1088 86966 : 100.0 * vacrel->lpdead_item_pages / orig_rel_pages,
1089 : vacrel->lpdead_items);
1090 548896 : for (int i = 0; i < vacrel->nindexes; i++)
1091 : {
1092 329964 : IndexBulkDeleteResult *istat = vacrel->indstats[i];
1093 :
1094 329964 : if (!istat)
1095 329914 : continue;
1096 :
1097 50 : appendStringInfo(&buf,
1098 50 : _("index \"%s\": pages: %u in total, %u newly deleted, %u currently deleted, %u reusable\n"),
1099 50 : indnames[i],
1100 : istat->num_pages,
1101 : istat->pages_newly_deleted,
1102 : istat->pages_deleted,
1103 : istat->pages_free);
1104 : }
1105 218932 : if (track_cost_delay_timing)
1106 : {
1107 : /*
1108 : * We bypass the changecount mechanism because this value is
1109 : * only updated by the calling process. We also rely on the
1110 : * above call to pgstat_progress_end_command() to not clear
1111 : * the st_progress_param array.
1112 : */
1113 0 : appendStringInfo(&buf, _("delay time: %.3f ms\n"),
1114 0 : (double) MyBEEntry->st_progress_param[PROGRESS_VACUUM_DELAY_TIME] / 1000000.0);
1115 : }
1116 218932 : if (track_io_timing)
1117 : {
1118 0 : double read_ms = (double) (pgStatBlockReadTime - startreadtime) / 1000;
1119 0 : double write_ms = (double) (pgStatBlockWriteTime - startwritetime) / 1000;
1120 :
1121 0 : appendStringInfo(&buf, _("I/O timings: read: %.3f ms, write: %.3f ms\n"),
1122 : read_ms, write_ms);
1123 : }
1124 218932 : if (secs_dur > 0 || usecs_dur > 0)
1125 : {
1126 218932 : read_rate = (double) BLCKSZ * total_blks_read /
1127 218932 : (1024 * 1024) / (secs_dur + usecs_dur / 1000000.0);
1128 218932 : write_rate = (double) BLCKSZ * total_blks_dirtied /
1129 218932 : (1024 * 1024) / (secs_dur + usecs_dur / 1000000.0);
1130 : }
1131 218932 : appendStringInfo(&buf, _("avg read rate: %.3f MB/s, avg write rate: %.3f MB/s\n"),
1132 : read_rate, write_rate);
1133 218932 : appendStringInfo(&buf,
1134 218932 : _("buffer usage: %" PRId64 " hits, %" PRId64 " reads, %" PRId64 " dirtied\n"),
1135 : total_blks_hit,
1136 : total_blks_read,
1137 : total_blks_dirtied);
1138 218932 : appendStringInfo(&buf,
1139 218932 : _("WAL usage: %" PRId64 " records, %" PRId64 " full page images, %" PRIu64 " bytes, %" PRId64 " buffers full\n"),
1140 : walusage.wal_records,
1141 : walusage.wal_fpi,
1142 : walusage.wal_bytes,
1143 : walusage.wal_buffers_full);
1144 218932 : appendStringInfo(&buf, _("system usage: %s"), pg_rusage_show(&ru0));
1145 :
1146 218932 : ereport(verbose ? INFO : LOG,
1147 : (errmsg_internal("%s", buf.data)));
1148 218930 : pfree(buf.data);
1149 : }
1150 : }
1151 :
1152 : /* Cleanup index statistics and index names */
1153 613382 : for (int i = 0; i < vacrel->nindexes; i++)
1154 : {
1155 368002 : if (vacrel->indstats[i])
1156 2826 : pfree(vacrel->indstats[i]);
1157 :
1158 368002 : if (instrument)
1159 330468 : pfree(indnames[i]);
1160 : }
1161 245380 : }
1162 :
1163 : /*
1164 : * lazy_scan_heap() -- workhorse function for VACUUM
1165 : *
1166 : * This routine prunes each page in the heap, and considers the need to
1167 : * freeze remaining tuples with storage (not including pages that can be
1168 : * skipped using the visibility map). Also performs related maintenance
1169 : * of the FSM and visibility map. These steps all take place during an
1170 : * initial pass over the target heap relation.
1171 : *
1172 : * Also invokes lazy_vacuum_all_indexes to vacuum indexes, which largely
1173 : * consists of deleting index tuples that point to LP_DEAD items left in
1174 : * heap pages following pruning. Earlier initial pass over the heap will
1175 : * have collected the TIDs whose index tuples need to be removed.
1176 : *
1177 : * Finally, invokes lazy_vacuum_heap_rel to vacuum heap pages, which
1178 : * largely consists of marking LP_DEAD items (from vacrel->dead_items)
1179 : * as LP_UNUSED. This has to happen in a second, final pass over the
1180 : * heap, to preserve a basic invariant that all index AMs rely on: no
1181 : * extant index tuple can ever be allowed to contain a TID that points to
1182 : * an LP_UNUSED line pointer in the heap. We must disallow premature
1183 : * recycling of line pointers to avoid index scans that get confused
1184 : * about which TID points to which tuple immediately after recycling.
1185 : * (Actually, this isn't a concern when target heap relation happens to
1186 : * have no indexes, which allows us to safely apply the one-pass strategy
1187 : * as an optimization).
1188 : *
1189 : * In practice we often have enough space to fit all TIDs, and so won't
1190 : * need to call lazy_vacuum more than once, after our initial pass over
1191 : * the heap has totally finished. Otherwise things are slightly more
1192 : * complicated: our "initial pass" over the heap applies only to those
1193 : * pages that were pruned before we needed to call lazy_vacuum, and our
1194 : * "final pass" over the heap only vacuums these same heap pages.
1195 : * However, we process indexes in full every time lazy_vacuum is called,
1196 : * which makes index processing very inefficient when memory is in short
1197 : * supply.
1198 : */
1199 : static void
1200 245382 : lazy_scan_heap(LVRelState *vacrel)
1201 : {
1202 : ReadStream *stream;
1203 245382 : BlockNumber rel_pages = vacrel->rel_pages,
1204 245382 : blkno = 0,
1205 245382 : next_fsm_block_to_vacuum = 0;
1206 245382 : BlockNumber orig_eager_scan_success_limit =
1207 : vacrel->eager_scan_remaining_successes; /* for logging */
1208 245382 : Buffer vmbuffer = InvalidBuffer;
1209 245382 : const int initprog_index[] = {
1210 : PROGRESS_VACUUM_PHASE,
1211 : PROGRESS_VACUUM_TOTAL_HEAP_BLKS,
1212 : PROGRESS_VACUUM_MAX_DEAD_TUPLE_BYTES
1213 : };
1214 : int64 initprog_val[3];
1215 :
1216 : /* Report that we're scanning the heap, advertising total # of blocks */
1217 245382 : initprog_val[0] = PROGRESS_VACUUM_PHASE_SCAN_HEAP;
1218 245382 : initprog_val[1] = rel_pages;
1219 245382 : initprog_val[2] = vacrel->dead_items_info->max_bytes;
1220 245382 : pgstat_progress_update_multi_param(3, initprog_index, initprog_val);
1221 :
1222 : /* Initialize for the first heap_vac_scan_next_block() call */
1223 245382 : vacrel->current_block = InvalidBlockNumber;
1224 245382 : vacrel->next_unskippable_block = InvalidBlockNumber;
1225 245382 : vacrel->next_unskippable_allvis = false;
1226 245382 : vacrel->next_unskippable_eager_scanned = false;
1227 245382 : vacrel->next_unskippable_vmbuffer = InvalidBuffer;
1228 :
1229 : /*
1230 : * Set up the read stream for vacuum's first pass through the heap.
1231 : *
1232 : * This could be made safe for READ_STREAM_USE_BATCHING, but only with
1233 : * explicit work in heap_vac_scan_next_block.
1234 : */
1235 245382 : stream = read_stream_begin_relation(READ_STREAM_MAINTENANCE,
1236 : vacrel->bstrategy,
1237 : vacrel->rel,
1238 : MAIN_FORKNUM,
1239 : heap_vac_scan_next_block,
1240 : vacrel,
1241 : sizeof(uint8));
1242 :
1243 : while (true)
1244 881364 : {
1245 : Buffer buf;
1246 : Page page;
1247 1126746 : uint8 blk_info = 0;
1248 1126746 : int ndeleted = 0;
1249 : bool has_lpdead_items;
1250 1126746 : void *per_buffer_data = NULL;
1251 1126746 : bool vm_page_frozen = false;
1252 1126746 : bool got_cleanup_lock = false;
1253 :
1254 1126746 : vacuum_delay_point(false);
1255 :
1256 : /*
1257 : * Regularly check if wraparound failsafe should trigger.
1258 : *
1259 : * There is a similar check inside lazy_vacuum_all_indexes(), but
1260 : * relfrozenxid might start to look dangerously old before we reach
1261 : * that point. This check also provides failsafe coverage for the
1262 : * one-pass strategy, and the two-pass strategy with the index_cleanup
1263 : * param set to 'off'.
1264 : */
1265 1126746 : if (vacrel->scanned_pages > 0 &&
1266 881364 : vacrel->scanned_pages % FAILSAFE_EVERY_PAGES == 0)
1267 0 : lazy_check_wraparound_failsafe(vacrel);
1268 :
1269 : /*
1270 : * Consider if we definitely have enough space to process TIDs on page
1271 : * already. If we are close to overrunning the available space for
1272 : * dead_items TIDs, pause and do a cycle of vacuuming before we tackle
1273 : * this page. However, let's force at least one page-worth of tuples
1274 : * to be stored as to ensure we do at least some work when the memory
1275 : * configured is so low that we run out before storing anything.
1276 : */
1277 1126746 : if (vacrel->dead_items_info->num_items > 0 &&
1278 51976 : TidStoreMemoryUsage(vacrel->dead_items) > vacrel->dead_items_info->max_bytes)
1279 : {
1280 : /*
1281 : * Before beginning index vacuuming, we release any pin we may
1282 : * hold on the visibility map page. This isn't necessary for
1283 : * correctness, but we do it anyway to avoid holding the pin
1284 : * across a lengthy, unrelated operation.
1285 : */
1286 4 : if (BufferIsValid(vmbuffer))
1287 : {
1288 4 : ReleaseBuffer(vmbuffer);
1289 4 : vmbuffer = InvalidBuffer;
1290 : }
1291 :
1292 : /* Perform a round of index and heap vacuuming */
1293 4 : vacrel->consider_bypass_optimization = false;
1294 4 : lazy_vacuum(vacrel);
1295 :
1296 : /*
1297 : * Vacuum the Free Space Map to make newly-freed space visible on
1298 : * upper-level FSM pages. Note that blkno is the previously
1299 : * processed block.
1300 : */
1301 4 : FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
1302 : blkno + 1);
1303 4 : next_fsm_block_to_vacuum = blkno;
1304 :
1305 : /* Report that we are once again scanning the heap */
1306 4 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
1307 : PROGRESS_VACUUM_PHASE_SCAN_HEAP);
1308 : }
1309 :
1310 1126746 : buf = read_stream_next_buffer(stream, &per_buffer_data);
1311 :
1312 : /* The relation is exhausted. */
1313 1126746 : if (!BufferIsValid(buf))
1314 245382 : break;
1315 :
1316 881364 : blk_info = *((uint8 *) per_buffer_data);
1317 881364 : CheckBufferIsPinnedOnce(buf);
1318 881364 : page = BufferGetPage(buf);
1319 881364 : blkno = BufferGetBlockNumber(buf);
1320 :
1321 881364 : vacrel->scanned_pages++;
1322 881364 : if (blk_info & VAC_BLK_WAS_EAGER_SCANNED)
1323 0 : vacrel->eager_scanned_pages++;
1324 :
1325 : /* Report as block scanned, update error traceback information */
1326 881364 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_SCANNED, blkno);
1327 881364 : update_vacuum_error_info(vacrel, NULL, VACUUM_ERRCB_PHASE_SCAN_HEAP,
1328 : blkno, InvalidOffsetNumber);
1329 :
1330 : /*
1331 : * Pin the visibility map page in case we need to mark the page
1332 : * all-visible. In most cases this will be very cheap, because we'll
1333 : * already have the correct page pinned anyway.
1334 : */
1335 881364 : visibilitymap_pin(vacrel->rel, blkno, &vmbuffer);
1336 :
1337 : /*
1338 : * We need a buffer cleanup lock to prune HOT chains and defragment
1339 : * the page in lazy_scan_prune. But when it's not possible to acquire
1340 : * a cleanup lock right away, we may be able to settle for reduced
1341 : * processing using lazy_scan_noprune.
1342 : */
1343 881364 : got_cleanup_lock = ConditionalLockBufferForCleanup(buf);
1344 :
1345 881364 : if (!got_cleanup_lock)
1346 268 : LockBuffer(buf, BUFFER_LOCK_SHARE);
1347 :
1348 : /* Check for new or empty pages before lazy_scan_[no]prune call */
1349 881364 : if (lazy_scan_new_or_empty(vacrel, buf, blkno, page, !got_cleanup_lock,
1350 881364 : vmbuffer))
1351 : {
1352 : /* Processed as new/empty page (lock and pin released) */
1353 1960 : continue;
1354 : }
1355 :
1356 : /*
1357 : * If we didn't get the cleanup lock, we can still collect LP_DEAD
1358 : * items in the dead_items area for later vacuuming, count live and
1359 : * recently dead tuples for vacuum logging, and determine if this
1360 : * block could later be truncated. If we encounter any xid/mxids that
1361 : * require advancing the relfrozenxid/relminxid, we'll have to wait
1362 : * for a cleanup lock and call lazy_scan_prune().
1363 : */
1364 879404 : if (!got_cleanup_lock &&
1365 268 : !lazy_scan_noprune(vacrel, buf, blkno, page, &has_lpdead_items))
1366 : {
1367 : /*
1368 : * lazy_scan_noprune could not do all required processing. Wait
1369 : * for a cleanup lock, and call lazy_scan_prune in the usual way.
1370 : */
1371 : Assert(vacrel->aggressive);
1372 74 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
1373 74 : LockBufferForCleanup(buf);
1374 74 : got_cleanup_lock = true;
1375 : }
1376 :
1377 : /*
1378 : * If we have a cleanup lock, we must now prune, freeze, and count
1379 : * tuples. We may have acquired the cleanup lock originally, or we may
1380 : * have gone back and acquired it after lazy_scan_noprune() returned
1381 : * false. Either way, the page hasn't been processed yet.
1382 : *
1383 : * Like lazy_scan_noprune(), lazy_scan_prune() will count
1384 : * recently_dead_tuples and live tuples for vacuum logging, determine
1385 : * if the block can later be truncated, and accumulate the details of
1386 : * remaining LP_DEAD line pointers on the page into dead_items. These
1387 : * dead items include those pruned by lazy_scan_prune() as well as
1388 : * line pointers previously marked LP_DEAD.
1389 : */
1390 879404 : if (got_cleanup_lock)
1391 879210 : ndeleted = lazy_scan_prune(vacrel, buf, blkno, page,
1392 : vmbuffer,
1393 879210 : blk_info & VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM,
1394 : &has_lpdead_items, &vm_page_frozen);
1395 :
1396 : /*
1397 : * Count an eagerly scanned page as a failure or a success.
1398 : *
1399 : * Only lazy_scan_prune() freezes pages, so if we didn't get the
1400 : * cleanup lock, we won't have frozen the page. However, we only count
1401 : * pages that were too new to require freezing as eager freeze
1402 : * failures.
1403 : *
1404 : * We could gather more information from lazy_scan_noprune() about
1405 : * whether or not there were tuples with XIDs or MXIDs older than the
1406 : * FreezeLimit or MultiXactCutoff. However, for simplicity, we simply
1407 : * exclude pages skipped due to cleanup lock contention from eager
1408 : * freeze algorithm caps.
1409 : */
1410 879404 : if (got_cleanup_lock &&
1411 879210 : (blk_info & VAC_BLK_WAS_EAGER_SCANNED))
1412 : {
1413 : /* Aggressive vacuums do not eager scan. */
1414 : Assert(!vacrel->aggressive);
1415 :
1416 0 : if (vm_page_frozen)
1417 : {
1418 0 : if (vacrel->eager_scan_remaining_successes > 0)
1419 0 : vacrel->eager_scan_remaining_successes--;
1420 :
1421 0 : if (vacrel->eager_scan_remaining_successes == 0)
1422 : {
1423 : /*
1424 : * Report only once that we disabled eager scanning. We
1425 : * may eagerly read ahead blocks in excess of the success
1426 : * or failure caps before attempting to freeze them, so we
1427 : * could reach here even after disabling additional eager
1428 : * scanning.
1429 : */
1430 0 : if (vacrel->eager_scan_max_fails_per_region > 0)
1431 0 : ereport(vacrel->verbose ? INFO : DEBUG2,
1432 : (errmsg("disabling eager scanning after freezing %u eagerly scanned blocks of relation \"%s.%s.%s\"",
1433 : orig_eager_scan_success_limit,
1434 : vacrel->dbname, vacrel->relnamespace,
1435 : vacrel->relname)));
1436 :
1437 : /*
1438 : * If we hit our success cap, permanently disable eager
1439 : * scanning by setting the other eager scan management
1440 : * fields to their disabled values.
1441 : */
1442 0 : vacrel->eager_scan_remaining_fails = 0;
1443 0 : vacrel->next_eager_scan_region_start = InvalidBlockNumber;
1444 0 : vacrel->eager_scan_max_fails_per_region = 0;
1445 : }
1446 : }
1447 0 : else if (vacrel->eager_scan_remaining_fails > 0)
1448 0 : vacrel->eager_scan_remaining_fails--;
1449 : }
1450 :
1451 : /*
1452 : * Now drop the buffer lock and, potentially, update the FSM.
1453 : *
1454 : * Our goal is to update the freespace map the last time we touch the
1455 : * page. If we'll process a block in the second pass, we may free up
1456 : * additional space on the page, so it is better to update the FSM
1457 : * after the second pass. If the relation has no indexes, or if index
1458 : * vacuuming is disabled, there will be no second heap pass; if this
1459 : * particular page has no dead items, the second heap pass will not
1460 : * touch this page. So, in those cases, update the FSM now.
1461 : *
1462 : * Note: In corner cases, it's possible to miss updating the FSM
1463 : * entirely. If index vacuuming is currently enabled, we'll skip the
1464 : * FSM update now. But if failsafe mode is later activated, or there
1465 : * are so few dead tuples that index vacuuming is bypassed, there will
1466 : * also be no opportunity to update the FSM later, because we'll never
1467 : * revisit this page. Since updating the FSM is desirable but not
1468 : * absolutely required, that's OK.
1469 : */
1470 879404 : if (vacrel->nindexes == 0
1471 843816 : || !vacrel->do_index_vacuuming
1472 625568 : || !has_lpdead_items)
1473 851188 : {
1474 851188 : Size freespace = PageGetHeapFreeSpace(page);
1475 :
1476 851188 : UnlockReleaseBuffer(buf);
1477 851188 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
1478 :
1479 : /*
1480 : * Periodically perform FSM vacuuming to make newly-freed space
1481 : * visible on upper FSM pages. This is done after vacuuming if the
1482 : * table has indexes. There will only be newly-freed space if we
1483 : * held the cleanup lock and lazy_scan_prune() was called.
1484 : */
1485 851188 : if (got_cleanup_lock && vacrel->nindexes == 0 && ndeleted > 0 &&
1486 904 : blkno - next_fsm_block_to_vacuum >= VACUUM_FSM_EVERY_PAGES)
1487 : {
1488 0 : FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
1489 : blkno);
1490 0 : next_fsm_block_to_vacuum = blkno;
1491 : }
1492 : }
1493 : else
1494 28216 : UnlockReleaseBuffer(buf);
1495 : }
1496 :
1497 245382 : vacrel->blkno = InvalidBlockNumber;
1498 245382 : if (BufferIsValid(vmbuffer))
1499 98182 : ReleaseBuffer(vmbuffer);
1500 :
1501 : /*
1502 : * Report that everything is now scanned. We never skip scanning the last
1503 : * block in the relation, so we can pass rel_pages here.
1504 : */
1505 245382 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_SCANNED,
1506 : rel_pages);
1507 :
1508 : /* now we can compute the new value for pg_class.reltuples */
1509 490764 : vacrel->new_live_tuples = vac_estimate_reltuples(vacrel->rel, rel_pages,
1510 : vacrel->scanned_pages,
1511 245382 : vacrel->live_tuples);
1512 :
1513 : /*
1514 : * Also compute the total number of surviving heap entries. In the
1515 : * (unlikely) scenario that new_live_tuples is -1, take it as zero.
1516 : */
1517 245382 : vacrel->new_rel_tuples =
1518 245382 : Max(vacrel->new_live_tuples, 0) + vacrel->recently_dead_tuples +
1519 245382 : vacrel->missed_dead_tuples;
1520 :
1521 245382 : read_stream_end(stream);
1522 :
1523 : /*
1524 : * Do index vacuuming (call each index's ambulkdelete routine), then do
1525 : * related heap vacuuming
1526 : */
1527 245382 : if (vacrel->dead_items_info->num_items > 0)
1528 1336 : lazy_vacuum(vacrel);
1529 :
1530 : /*
1531 : * Vacuum the remainder of the Free Space Map. We must do this whether or
1532 : * not there were indexes, and whether or not we bypassed index vacuuming.
1533 : * We can pass rel_pages here because we never skip scanning the last
1534 : * block of the relation.
1535 : */
1536 245382 : if (rel_pages > next_fsm_block_to_vacuum)
1537 98182 : FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum, rel_pages);
1538 :
1539 : /* report all blocks vacuumed */
1540 245382 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_VACUUMED, rel_pages);
1541 :
1542 : /* Do final index cleanup (call each index's amvacuumcleanup routine) */
1543 245382 : if (vacrel->nindexes > 0 && vacrel->do_index_cleanup)
1544 168810 : lazy_cleanup_all_indexes(vacrel);
1545 245382 : }
1546 :
1547 : /*
1548 : * heap_vac_scan_next_block() -- read stream callback to get the next block
1549 : * for vacuum to process
1550 : *
1551 : * Every time lazy_scan_heap() needs a new block to process during its first
1552 : * phase, it invokes read_stream_next_buffer() with a stream set up to call
1553 : * heap_vac_scan_next_block() to get the next block.
1554 : *
1555 : * heap_vac_scan_next_block() uses the visibility map, vacuum options, and
1556 : * various thresholds to skip blocks which do not need to be processed and
1557 : * returns the next block to process or InvalidBlockNumber if there are no
1558 : * remaining blocks.
1559 : *
1560 : * The visibility status of the next block to process and whether or not it
1561 : * was eager scanned is set in the per_buffer_data.
1562 : *
1563 : * callback_private_data contains a reference to the LVRelState, passed to the
1564 : * read stream API during stream setup. The LVRelState is an in/out parameter
1565 : * here (locally named `vacrel`). Vacuum options and information about the
1566 : * relation are read from it. vacrel->skippedallvis is set if we skip a block
1567 : * that's all-visible but not all-frozen (to ensure that we don't update
1568 : * relfrozenxid in that case). vacrel also holds information about the next
1569 : * unskippable block -- as bookkeeping for this function.
1570 : */
1571 : static BlockNumber
1572 1126746 : heap_vac_scan_next_block(ReadStream *stream,
1573 : void *callback_private_data,
1574 : void *per_buffer_data)
1575 : {
1576 : BlockNumber next_block;
1577 1126746 : LVRelState *vacrel = callback_private_data;
1578 1126746 : uint8 blk_info = 0;
1579 :
1580 : /* relies on InvalidBlockNumber + 1 overflowing to 0 on first call */
1581 1126746 : next_block = vacrel->current_block + 1;
1582 :
1583 : /* Have we reached the end of the relation? */
1584 1126746 : if (next_block >= vacrel->rel_pages)
1585 : {
1586 245382 : if (BufferIsValid(vacrel->next_unskippable_vmbuffer))
1587 : {
1588 95240 : ReleaseBuffer(vacrel->next_unskippable_vmbuffer);
1589 95240 : vacrel->next_unskippable_vmbuffer = InvalidBuffer;
1590 : }
1591 245382 : return InvalidBlockNumber;
1592 : }
1593 :
1594 : /*
1595 : * We must be in one of the three following states:
1596 : */
1597 881364 : if (next_block > vacrel->next_unskippable_block ||
1598 393978 : vacrel->next_unskippable_block == InvalidBlockNumber)
1599 : {
1600 : /*
1601 : * 1. We have just processed an unskippable block (or we're at the
1602 : * beginning of the scan). Find the next unskippable block using the
1603 : * visibility map.
1604 : */
1605 : bool skipsallvis;
1606 :
1607 585568 : find_next_unskippable_block(vacrel, &skipsallvis);
1608 :
1609 : /*
1610 : * We now know the next block that we must process. It can be the
1611 : * next block after the one we just processed, or something further
1612 : * ahead. If it's further ahead, we can jump to it, but we choose to
1613 : * do so only if we can skip at least SKIP_PAGES_THRESHOLD consecutive
1614 : * pages. Since we're reading sequentially, the OS should be doing
1615 : * readahead for us, so there's no gain in skipping a page now and
1616 : * then. Skipping such a range might even discourage sequential
1617 : * detection.
1618 : *
1619 : * This test also enables more frequent relfrozenxid advancement
1620 : * during non-aggressive VACUUMs. If the range has any all-visible
1621 : * pages then skipping makes updating relfrozenxid unsafe, which is a
1622 : * real downside.
1623 : */
1624 585568 : if (vacrel->next_unskippable_block - next_block >= SKIP_PAGES_THRESHOLD)
1625 : {
1626 8770 : next_block = vacrel->next_unskippable_block;
1627 8770 : if (skipsallvis)
1628 76 : vacrel->skippedallvis = true;
1629 : }
1630 : }
1631 :
1632 : /* Now we must be in one of the two remaining states: */
1633 881364 : if (next_block < vacrel->next_unskippable_block)
1634 : {
1635 : /*
1636 : * 2. We are processing a range of blocks that we could have skipped
1637 : * but chose not to. We know that they are all-visible in the VM,
1638 : * otherwise they would've been unskippable.
1639 : */
1640 295796 : vacrel->current_block = next_block;
1641 295796 : blk_info |= VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM;
1642 295796 : *((uint8 *) per_buffer_data) = blk_info;
1643 295796 : return vacrel->current_block;
1644 : }
1645 : else
1646 : {
1647 : /*
1648 : * 3. We reached the next unskippable block. Process it. On next
1649 : * iteration, we will be back in state 1.
1650 : */
1651 : Assert(next_block == vacrel->next_unskippable_block);
1652 :
1653 585568 : vacrel->current_block = next_block;
1654 585568 : if (vacrel->next_unskippable_allvis)
1655 89866 : blk_info |= VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM;
1656 585568 : if (vacrel->next_unskippable_eager_scanned)
1657 0 : blk_info |= VAC_BLK_WAS_EAGER_SCANNED;
1658 585568 : *((uint8 *) per_buffer_data) = blk_info;
1659 585568 : return vacrel->current_block;
1660 : }
1661 : }
1662 :
1663 : /*
1664 : * Find the next unskippable block in a vacuum scan using the visibility map.
1665 : * The next unskippable block and its visibility information is updated in
1666 : * vacrel.
1667 : *
1668 : * Note: our opinion of which blocks can be skipped can go stale immediately.
1669 : * It's okay if caller "misses" a page whose all-visible or all-frozen marking
1670 : * was concurrently cleared, though. All that matters is that caller scan all
1671 : * pages whose tuples might contain XIDs < OldestXmin, or MXIDs < OldestMxact.
1672 : * (Actually, non-aggressive VACUUMs can choose to skip all-visible pages with
1673 : * older XIDs/MXIDs. The *skippedallvis flag will be set here when the choice
1674 : * to skip such a range is actually made, making everything safe.)
1675 : */
1676 : static void
1677 585568 : find_next_unskippable_block(LVRelState *vacrel, bool *skipsallvis)
1678 : {
1679 585568 : BlockNumber rel_pages = vacrel->rel_pages;
1680 585568 : BlockNumber next_unskippable_block = vacrel->next_unskippable_block + 1;
1681 585568 : Buffer next_unskippable_vmbuffer = vacrel->next_unskippable_vmbuffer;
1682 585568 : bool next_unskippable_eager_scanned = false;
1683 : bool next_unskippable_allvis;
1684 :
1685 585568 : *skipsallvis = false;
1686 :
1687 913434 : for (;; next_unskippable_block++)
1688 913434 : {
1689 1499002 : uint8 mapbits = visibilitymap_get_status(vacrel->rel,
1690 : next_unskippable_block,
1691 : &next_unskippable_vmbuffer);
1692 :
1693 1499002 : next_unskippable_allvis = (mapbits & VISIBILITYMAP_ALL_VISIBLE) != 0;
1694 :
1695 : /*
1696 : * At the start of each eager scan region, normal vacuums with eager
1697 : * scanning enabled reset the failure counter, allowing vacuum to
1698 : * resume eager scanning if it had been suspended in the previous
1699 : * region.
1700 : */
1701 1499002 : if (next_unskippable_block >= vacrel->next_eager_scan_region_start)
1702 : {
1703 0 : vacrel->eager_scan_remaining_fails =
1704 0 : vacrel->eager_scan_max_fails_per_region;
1705 0 : vacrel->next_eager_scan_region_start += EAGER_SCAN_REGION_SIZE;
1706 : }
1707 :
1708 : /*
1709 : * A block is unskippable if it is not all visible according to the
1710 : * visibility map.
1711 : */
1712 1499002 : if (!next_unskippable_allvis)
1713 : {
1714 : Assert((mapbits & VISIBILITYMAP_ALL_FROZEN) == 0);
1715 495702 : break;
1716 : }
1717 :
1718 : /*
1719 : * Caller must scan the last page to determine whether it has tuples
1720 : * (caller must have the opportunity to set vacrel->nonempty_pages).
1721 : * This rule avoids having lazy_truncate_heap() take access-exclusive
1722 : * lock on rel to attempt a truncation that fails anyway, just because
1723 : * there are tuples on the last page (it is likely that there will be
1724 : * tuples on other nearby pages as well, but those can be skipped).
1725 : *
1726 : * Implement this by always treating the last block as unsafe to skip.
1727 : */
1728 1003300 : if (next_unskippable_block == rel_pages - 1)
1729 89054 : break;
1730 :
1731 : /* DISABLE_PAGE_SKIPPING makes all skipping unsafe */
1732 914246 : if (!vacrel->skipwithvm)
1733 812 : break;
1734 :
1735 : /*
1736 : * All-frozen pages cannot contain XIDs < OldestXmin (XIDs that aren't
1737 : * already frozen by now), so this page can be skipped.
1738 : */
1739 913434 : if ((mapbits & VISIBILITYMAP_ALL_FROZEN) != 0)
1740 907898 : continue;
1741 :
1742 : /*
1743 : * Aggressive vacuums cannot skip any all-visible pages that are not
1744 : * also all-frozen.
1745 : */
1746 5536 : if (vacrel->aggressive)
1747 0 : break;
1748 :
1749 : /*
1750 : * Normal vacuums with eager scanning enabled only skip all-visible
1751 : * but not all-frozen pages if they have hit the failure limit for the
1752 : * current eager scan region.
1753 : */
1754 5536 : if (vacrel->eager_scan_remaining_fails > 0)
1755 : {
1756 0 : next_unskippable_eager_scanned = true;
1757 0 : break;
1758 : }
1759 :
1760 : /*
1761 : * All-visible blocks are safe to skip in a normal vacuum. But
1762 : * remember that the final range contains such a block for later.
1763 : */
1764 5536 : *skipsallvis = true;
1765 : }
1766 :
1767 : /* write the local variables back to vacrel */
1768 585568 : vacrel->next_unskippable_block = next_unskippable_block;
1769 585568 : vacrel->next_unskippable_allvis = next_unskippable_allvis;
1770 585568 : vacrel->next_unskippable_eager_scanned = next_unskippable_eager_scanned;
1771 585568 : vacrel->next_unskippable_vmbuffer = next_unskippable_vmbuffer;
1772 585568 : }
1773 :
1774 : /*
1775 : * lazy_scan_new_or_empty() -- lazy_scan_heap() new/empty page handling.
1776 : *
1777 : * Must call here to handle both new and empty pages before calling
1778 : * lazy_scan_prune or lazy_scan_noprune, since they're not prepared to deal
1779 : * with new or empty pages.
1780 : *
1781 : * It's necessary to consider new pages as a special case, since the rules for
1782 : * maintaining the visibility map and FSM with empty pages are a little
1783 : * different (though new pages can be truncated away during rel truncation).
1784 : *
1785 : * Empty pages are not really a special case -- they're just heap pages that
1786 : * have no allocated tuples (including even LP_UNUSED items). You might
1787 : * wonder why we need to handle them here all the same. It's only necessary
1788 : * because of a corner-case involving a hard crash during heap relation
1789 : * extension. If we ever make relation-extension crash safe, then it should
1790 : * no longer be necessary to deal with empty pages here (or new pages, for
1791 : * that matter).
1792 : *
1793 : * Caller must hold at least a shared lock. We might need to escalate the
1794 : * lock in that case, so the type of lock caller holds needs to be specified
1795 : * using 'sharelock' argument.
1796 : *
1797 : * Returns false in common case where caller should go on to call
1798 : * lazy_scan_prune (or lazy_scan_noprune). Otherwise returns true, indicating
1799 : * that lazy_scan_heap is done processing the page, releasing lock on caller's
1800 : * behalf.
1801 : *
1802 : * No vm_page_frozen output parameter (like that passed to lazy_scan_prune())
1803 : * is passed here because neither empty nor new pages can be eagerly frozen.
1804 : * New pages are never frozen. Empty pages are always set frozen in the VM at
1805 : * the same time that they are set all-visible, and we don't eagerly scan
1806 : * frozen pages.
1807 : */
1808 : static bool
1809 881364 : lazy_scan_new_or_empty(LVRelState *vacrel, Buffer buf, BlockNumber blkno,
1810 : Page page, bool sharelock, Buffer vmbuffer)
1811 : {
1812 : Size freespace;
1813 :
1814 881364 : if (PageIsNew(page))
1815 : {
1816 : /*
1817 : * All-zeroes pages can be left over if either a backend extends the
1818 : * relation by a single page, but crashes before the newly initialized
1819 : * page has been written out, or when bulk-extending the relation
1820 : * (which creates a number of empty pages at the tail end of the
1821 : * relation), and then enters them into the FSM.
1822 : *
1823 : * Note we do not enter the page into the visibilitymap. That has the
1824 : * downside that we repeatedly visit this page in subsequent vacuums,
1825 : * but otherwise we'll never discover the space on a promoted standby.
1826 : * The harm of repeated checking ought to normally not be too bad. The
1827 : * space usually should be used at some point, otherwise there
1828 : * wouldn't be any regular vacuums.
1829 : *
1830 : * Make sure these pages are in the FSM, to ensure they can be reused.
1831 : * Do that by testing if there's any space recorded for the page. If
1832 : * not, enter it. We do so after releasing the lock on the heap page,
1833 : * the FSM is approximate, after all.
1834 : */
1835 1906 : UnlockReleaseBuffer(buf);
1836 :
1837 1906 : if (GetRecordedFreeSpace(vacrel->rel, blkno) == 0)
1838 : {
1839 950 : freespace = BLCKSZ - SizeOfPageHeaderData;
1840 :
1841 950 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
1842 : }
1843 :
1844 1906 : return true;
1845 : }
1846 :
1847 879458 : if (PageIsEmpty(page))
1848 : {
1849 : /*
1850 : * It seems likely that caller will always be able to get a cleanup
1851 : * lock on an empty page. But don't take any chances -- escalate to
1852 : * an exclusive lock (still don't need a cleanup lock, though).
1853 : */
1854 54 : if (sharelock)
1855 : {
1856 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
1857 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
1858 :
1859 0 : if (!PageIsEmpty(page))
1860 : {
1861 : /* page isn't new or empty -- keep lock and pin for now */
1862 0 : return false;
1863 : }
1864 : }
1865 : else
1866 : {
1867 : /* Already have a full cleanup lock (which is more than enough) */
1868 : }
1869 :
1870 : /*
1871 : * Unlike new pages, empty pages are always set all-visible and
1872 : * all-frozen.
1873 : */
1874 54 : if (!PageIsAllVisible(page))
1875 : {
1876 0 : START_CRIT_SECTION();
1877 :
1878 : /* mark buffer dirty before writing a WAL record */
1879 0 : MarkBufferDirty(buf);
1880 :
1881 : /*
1882 : * It's possible that another backend has extended the heap,
1883 : * initialized the page, and then failed to WAL-log the page due
1884 : * to an ERROR. Since heap extension is not WAL-logged, recovery
1885 : * might try to replay our record setting the page all-visible and
1886 : * find that the page isn't initialized, which will cause a PANIC.
1887 : * To prevent that, check whether the page has been previously
1888 : * WAL-logged, and if not, do that now.
1889 : */
1890 0 : if (RelationNeedsWAL(vacrel->rel) &&
1891 0 : PageGetLSN(page) == InvalidXLogRecPtr)
1892 0 : log_newpage_buffer(buf, true);
1893 :
1894 0 : PageSetAllVisible(page);
1895 0 : visibilitymap_set(vacrel->rel, blkno, buf,
1896 : InvalidXLogRecPtr,
1897 : vmbuffer, InvalidTransactionId,
1898 : VISIBILITYMAP_ALL_VISIBLE |
1899 : VISIBILITYMAP_ALL_FROZEN);
1900 0 : END_CRIT_SECTION();
1901 :
1902 : /* Count the newly all-frozen pages for logging */
1903 0 : vacrel->vm_new_visible_pages++;
1904 0 : vacrel->vm_new_visible_frozen_pages++;
1905 : }
1906 :
1907 54 : freespace = PageGetHeapFreeSpace(page);
1908 54 : UnlockReleaseBuffer(buf);
1909 54 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
1910 54 : return true;
1911 : }
1912 :
1913 : /* page isn't new or empty -- keep lock and pin */
1914 879404 : return false;
1915 : }
1916 :
1917 : /* qsort comparator for sorting OffsetNumbers */
1918 : static int
1919 6261364 : cmpOffsetNumbers(const void *a, const void *b)
1920 : {
1921 6261364 : return pg_cmp_u16(*(const OffsetNumber *) a, *(const OffsetNumber *) b);
1922 : }
1923 :
1924 : /*
1925 : * lazy_scan_prune() -- lazy_scan_heap() pruning and freezing.
1926 : *
1927 : * Caller must hold pin and buffer cleanup lock on the buffer.
1928 : *
1929 : * vmbuffer is the buffer containing the VM block with visibility information
1930 : * for the heap block, blkno. all_visible_according_to_vm is the saved
1931 : * visibility status of the heap block looked up earlier by the caller. We
1932 : * won't rely entirely on this status, as it may be out of date.
1933 : *
1934 : * *has_lpdead_items is set to true or false depending on whether, upon return
1935 : * from this function, any LP_DEAD items are still present on the page.
1936 : *
1937 : * *vm_page_frozen is set to true if the page is newly set all-frozen in the
1938 : * VM. The caller currently only uses this for determining whether an eagerly
1939 : * scanned page was successfully set all-frozen.
1940 : *
1941 : * Returns the number of tuples deleted from the page during HOT pruning.
1942 : */
1943 : static int
1944 879210 : lazy_scan_prune(LVRelState *vacrel,
1945 : Buffer buf,
1946 : BlockNumber blkno,
1947 : Page page,
1948 : Buffer vmbuffer,
1949 : bool all_visible_according_to_vm,
1950 : bool *has_lpdead_items,
1951 : bool *vm_page_frozen)
1952 : {
1953 879210 : Relation rel = vacrel->rel;
1954 : PruneFreezeResult presult;
1955 879210 : int prune_options = 0;
1956 :
1957 : Assert(BufferGetBlockNumber(buf) == blkno);
1958 :
1959 : /*
1960 : * Prune all HOT-update chains and potentially freeze tuples on this page.
1961 : *
1962 : * If the relation has no indexes, we can immediately mark would-be dead
1963 : * items LP_UNUSED.
1964 : *
1965 : * The number of tuples removed from the page is returned in
1966 : * presult.ndeleted. It should not be confused with presult.lpdead_items;
1967 : * presult.lpdead_items's final value can be thought of as the number of
1968 : * tuples that were deleted from indexes.
1969 : *
1970 : * We will update the VM after collecting LP_DEAD items and freezing
1971 : * tuples. Pruning will have determined whether or not the page is
1972 : * all-visible.
1973 : */
1974 879210 : prune_options = HEAP_PAGE_PRUNE_FREEZE;
1975 879210 : if (vacrel->nindexes == 0)
1976 35588 : prune_options |= HEAP_PAGE_PRUNE_MARK_UNUSED_NOW;
1977 :
1978 879210 : heap_page_prune_and_freeze(rel, buf, vacrel->vistest, prune_options,
1979 : &vacrel->cutoffs, &presult, PRUNE_VACUUM_SCAN,
1980 : &vacrel->offnum,
1981 : &vacrel->NewRelfrozenXid, &vacrel->NewRelminMxid);
1982 :
1983 : Assert(MultiXactIdIsValid(vacrel->NewRelminMxid));
1984 : Assert(TransactionIdIsValid(vacrel->NewRelfrozenXid));
1985 :
1986 879210 : if (presult.nfrozen > 0)
1987 : {
1988 : /*
1989 : * We don't increment the new_frozen_tuple_pages instrumentation
1990 : * counter when nfrozen == 0, since it only counts pages with newly
1991 : * frozen tuples (don't confuse that with pages newly set all-frozen
1992 : * in VM).
1993 : */
1994 46910 : vacrel->new_frozen_tuple_pages++;
1995 : }
1996 :
1997 : /*
1998 : * VACUUM will call heap_page_is_all_visible() during the second pass over
1999 : * the heap to determine all_visible and all_frozen for the page -- this
2000 : * is a specialized version of the logic from this function. Now that
2001 : * we've finished pruning and freezing, make sure that we're in total
2002 : * agreement with heap_page_is_all_visible() using an assertion.
2003 : */
2004 : #ifdef USE_ASSERT_CHECKING
2005 : /* Note that all_frozen value does not matter when !all_visible */
2006 : if (presult.all_visible)
2007 : {
2008 : TransactionId debug_cutoff;
2009 : bool debug_all_frozen;
2010 :
2011 : Assert(presult.lpdead_items == 0);
2012 :
2013 : if (!heap_page_is_all_visible(vacrel, buf,
2014 : &debug_cutoff, &debug_all_frozen))
2015 : Assert(false);
2016 :
2017 : Assert(presult.all_frozen == debug_all_frozen);
2018 :
2019 : Assert(!TransactionIdIsValid(debug_cutoff) ||
2020 : debug_cutoff == presult.vm_conflict_horizon);
2021 : }
2022 : #endif
2023 :
2024 : /*
2025 : * Now save details of the LP_DEAD items from the page in vacrel
2026 : */
2027 879210 : if (presult.lpdead_items > 0)
2028 : {
2029 32874 : vacrel->lpdead_item_pages++;
2030 :
2031 : /*
2032 : * deadoffsets are collected incrementally in
2033 : * heap_page_prune_and_freeze() as each dead line pointer is recorded,
2034 : * with an indeterminate order, but dead_items_add requires them to be
2035 : * sorted.
2036 : */
2037 32874 : qsort(presult.deadoffsets, presult.lpdead_items, sizeof(OffsetNumber),
2038 : cmpOffsetNumbers);
2039 :
2040 32874 : dead_items_add(vacrel, blkno, presult.deadoffsets, presult.lpdead_items);
2041 : }
2042 :
2043 : /* Finally, add page-local counts to whole-VACUUM counts */
2044 879210 : vacrel->tuples_deleted += presult.ndeleted;
2045 879210 : vacrel->tuples_frozen += presult.nfrozen;
2046 879210 : vacrel->lpdead_items += presult.lpdead_items;
2047 879210 : vacrel->live_tuples += presult.live_tuples;
2048 879210 : vacrel->recently_dead_tuples += presult.recently_dead_tuples;
2049 :
2050 : /* Can't truncate this page */
2051 879210 : if (presult.hastup)
2052 863598 : vacrel->nonempty_pages = blkno + 1;
2053 :
2054 : /* Did we find LP_DEAD items? */
2055 879210 : *has_lpdead_items = (presult.lpdead_items > 0);
2056 :
2057 : Assert(!presult.all_visible || !(*has_lpdead_items));
2058 :
2059 : /*
2060 : * Handle setting visibility map bit based on information from the VM (as
2061 : * of last heap_vac_scan_next_block() call), and from all_visible and
2062 : * all_frozen variables
2063 : */
2064 879210 : if (!all_visible_according_to_vm && presult.all_visible)
2065 73470 : {
2066 : uint8 old_vmbits;
2067 73470 : uint8 flags = VISIBILITYMAP_ALL_VISIBLE;
2068 :
2069 73470 : if (presult.all_frozen)
2070 : {
2071 : Assert(!TransactionIdIsValid(presult.vm_conflict_horizon));
2072 54200 : flags |= VISIBILITYMAP_ALL_FROZEN;
2073 : }
2074 :
2075 : /*
2076 : * It should never be the case that the visibility map page is set
2077 : * while the page-level bit is clear, but the reverse is allowed (if
2078 : * checksums are not enabled). Regardless, set both bits so that we
2079 : * get back in sync.
2080 : *
2081 : * NB: If the heap page is all-visible but the VM bit is not set, we
2082 : * don't need to dirty the heap page. However, if checksums are
2083 : * enabled, we do need to make sure that the heap page is dirtied
2084 : * before passing it to visibilitymap_set(), because it may be logged.
2085 : * Given that this situation should only happen in rare cases after a
2086 : * crash, it is not worth optimizing.
2087 : */
2088 73470 : PageSetAllVisible(page);
2089 73470 : MarkBufferDirty(buf);
2090 73470 : old_vmbits = visibilitymap_set(vacrel->rel, blkno, buf,
2091 : InvalidXLogRecPtr,
2092 : vmbuffer, presult.vm_conflict_horizon,
2093 : flags);
2094 :
2095 : /*
2096 : * If the page wasn't already set all-visible and/or all-frozen in the
2097 : * VM, count it as newly set for logging.
2098 : */
2099 73470 : if ((old_vmbits & VISIBILITYMAP_ALL_VISIBLE) == 0)
2100 : {
2101 73470 : vacrel->vm_new_visible_pages++;
2102 73470 : if (presult.all_frozen)
2103 : {
2104 54200 : vacrel->vm_new_visible_frozen_pages++;
2105 54200 : *vm_page_frozen = true;
2106 : }
2107 : }
2108 0 : else if ((old_vmbits & VISIBILITYMAP_ALL_FROZEN) == 0 &&
2109 0 : presult.all_frozen)
2110 : {
2111 0 : vacrel->vm_new_frozen_pages++;
2112 0 : *vm_page_frozen = true;
2113 : }
2114 : }
2115 :
2116 : /*
2117 : * As of PostgreSQL 9.2, the visibility map bit should never be set if the
2118 : * page-level bit is clear. However, it's possible that the bit got
2119 : * cleared after heap_vac_scan_next_block() was called, so we must recheck
2120 : * with buffer lock before concluding that the VM is corrupt.
2121 : */
2122 805740 : else if (all_visible_according_to_vm && !PageIsAllVisible(page) &&
2123 0 : visibilitymap_get_status(vacrel->rel, blkno, &vmbuffer) != 0)
2124 : {
2125 0 : elog(WARNING, "page is not marked all-visible but visibility map bit is set in relation \"%s\" page %u",
2126 : vacrel->relname, blkno);
2127 0 : visibilitymap_clear(vacrel->rel, blkno, vmbuffer,
2128 : VISIBILITYMAP_VALID_BITS);
2129 : }
2130 :
2131 : /*
2132 : * It's possible for the value returned by
2133 : * GetOldestNonRemovableTransactionId() to move backwards, so it's not
2134 : * wrong for us to see tuples that appear to not be visible to everyone
2135 : * yet, while PD_ALL_VISIBLE is already set. The real safe xmin value
2136 : * never moves backwards, but GetOldestNonRemovableTransactionId() is
2137 : * conservative and sometimes returns a value that's unnecessarily small,
2138 : * so if we see that contradiction it just means that the tuples that we
2139 : * think are not visible to everyone yet actually are, and the
2140 : * PD_ALL_VISIBLE flag is correct.
2141 : *
2142 : * There should never be LP_DEAD items on a page with PD_ALL_VISIBLE set,
2143 : * however.
2144 : */
2145 805740 : else if (presult.lpdead_items > 0 && PageIsAllVisible(page))
2146 : {
2147 0 : elog(WARNING, "page containing LP_DEAD items is marked as all-visible in relation \"%s\" page %u",
2148 : vacrel->relname, blkno);
2149 0 : PageClearAllVisible(page);
2150 0 : MarkBufferDirty(buf);
2151 0 : visibilitymap_clear(vacrel->rel, blkno, vmbuffer,
2152 : VISIBILITYMAP_VALID_BITS);
2153 : }
2154 :
2155 : /*
2156 : * If the all-visible page is all-frozen but not marked as such yet, mark
2157 : * it as all-frozen. Note that all_frozen is only valid if all_visible is
2158 : * true, so we must check both all_visible and all_frozen.
2159 : */
2160 805740 : else if (all_visible_according_to_vm && presult.all_visible &&
2161 385428 : presult.all_frozen && !VM_ALL_FROZEN(vacrel->rel, blkno, &vmbuffer))
2162 : {
2163 : uint8 old_vmbits;
2164 :
2165 : /*
2166 : * Avoid relying on all_visible_according_to_vm as a proxy for the
2167 : * page-level PD_ALL_VISIBLE bit being set, since it might have become
2168 : * stale -- even when all_visible is set
2169 : */
2170 44 : if (!PageIsAllVisible(page))
2171 : {
2172 0 : PageSetAllVisible(page);
2173 0 : MarkBufferDirty(buf);
2174 : }
2175 :
2176 : /*
2177 : * Set the page all-frozen (and all-visible) in the VM.
2178 : *
2179 : * We can pass InvalidTransactionId as our cutoff_xid, since a
2180 : * snapshotConflictHorizon sufficient to make everything safe for REDO
2181 : * was logged when the page's tuples were frozen.
2182 : */
2183 : Assert(!TransactionIdIsValid(presult.vm_conflict_horizon));
2184 44 : old_vmbits = visibilitymap_set(vacrel->rel, blkno, buf,
2185 : InvalidXLogRecPtr,
2186 : vmbuffer, InvalidTransactionId,
2187 : VISIBILITYMAP_ALL_VISIBLE |
2188 : VISIBILITYMAP_ALL_FROZEN);
2189 :
2190 : /*
2191 : * The page was likely already set all-visible in the VM. However,
2192 : * there is a small chance that it was modified sometime between
2193 : * setting all_visible_according_to_vm and checking the visibility
2194 : * during pruning. Check the return value of old_vmbits anyway to
2195 : * ensure the visibility map counters used for logging are accurate.
2196 : */
2197 44 : if ((old_vmbits & VISIBILITYMAP_ALL_VISIBLE) == 0)
2198 : {
2199 0 : vacrel->vm_new_visible_pages++;
2200 0 : vacrel->vm_new_visible_frozen_pages++;
2201 0 : *vm_page_frozen = true;
2202 : }
2203 :
2204 : /*
2205 : * We already checked that the page was not set all-frozen in the VM
2206 : * above, so we don't need to test the value of old_vmbits.
2207 : */
2208 : else
2209 : {
2210 44 : vacrel->vm_new_frozen_pages++;
2211 44 : *vm_page_frozen = true;
2212 : }
2213 : }
2214 :
2215 879210 : return presult.ndeleted;
2216 : }
2217 :
2218 : /*
2219 : * lazy_scan_noprune() -- lazy_scan_prune() without pruning or freezing
2220 : *
2221 : * Caller need only hold a pin and share lock on the buffer, unlike
2222 : * lazy_scan_prune, which requires a full cleanup lock. While pruning isn't
2223 : * performed here, it's quite possible that an earlier opportunistic pruning
2224 : * operation left LP_DEAD items behind. We'll at least collect any such items
2225 : * in dead_items for removal from indexes.
2226 : *
2227 : * For aggressive VACUUM callers, we may return false to indicate that a full
2228 : * cleanup lock is required for processing by lazy_scan_prune. This is only
2229 : * necessary when the aggressive VACUUM needs to freeze some tuple XIDs from
2230 : * one or more tuples on the page. We always return true for non-aggressive
2231 : * callers.
2232 : *
2233 : * If this function returns true, *has_lpdead_items gets set to true or false
2234 : * depending on whether, upon return from this function, any LP_DEAD items are
2235 : * present on the page. If this function returns false, *has_lpdead_items
2236 : * is not updated.
2237 : */
2238 : static bool
2239 268 : lazy_scan_noprune(LVRelState *vacrel,
2240 : Buffer buf,
2241 : BlockNumber blkno,
2242 : Page page,
2243 : bool *has_lpdead_items)
2244 : {
2245 : OffsetNumber offnum,
2246 : maxoff;
2247 : int lpdead_items,
2248 : live_tuples,
2249 : recently_dead_tuples,
2250 : missed_dead_tuples;
2251 : bool hastup;
2252 : HeapTupleHeader tupleheader;
2253 268 : TransactionId NoFreezePageRelfrozenXid = vacrel->NewRelfrozenXid;
2254 268 : MultiXactId NoFreezePageRelminMxid = vacrel->NewRelminMxid;
2255 : OffsetNumber deadoffsets[MaxHeapTuplesPerPage];
2256 :
2257 : Assert(BufferGetBlockNumber(buf) == blkno);
2258 :
2259 268 : hastup = false; /* for now */
2260 :
2261 268 : lpdead_items = 0;
2262 268 : live_tuples = 0;
2263 268 : recently_dead_tuples = 0;
2264 268 : missed_dead_tuples = 0;
2265 :
2266 268 : maxoff = PageGetMaxOffsetNumber(page);
2267 268 : for (offnum = FirstOffsetNumber;
2268 8594 : offnum <= maxoff;
2269 8326 : offnum = OffsetNumberNext(offnum))
2270 : {
2271 : ItemId itemid;
2272 : HeapTupleData tuple;
2273 :
2274 8400 : vacrel->offnum = offnum;
2275 8400 : itemid = PageGetItemId(page, offnum);
2276 :
2277 8400 : if (!ItemIdIsUsed(itemid))
2278 2010 : continue;
2279 :
2280 6990 : if (ItemIdIsRedirected(itemid))
2281 : {
2282 600 : hastup = true;
2283 600 : continue;
2284 : }
2285 :
2286 6390 : if (ItemIdIsDead(itemid))
2287 : {
2288 : /*
2289 : * Deliberately don't set hastup=true here. See same point in
2290 : * lazy_scan_prune for an explanation.
2291 : */
2292 0 : deadoffsets[lpdead_items++] = offnum;
2293 0 : continue;
2294 : }
2295 :
2296 6390 : hastup = true; /* page prevents rel truncation */
2297 6390 : tupleheader = (HeapTupleHeader) PageGetItem(page, itemid);
2298 6390 : if (heap_tuple_should_freeze(tupleheader, &vacrel->cutoffs,
2299 : &NoFreezePageRelfrozenXid,
2300 : &NoFreezePageRelminMxid))
2301 : {
2302 : /* Tuple with XID < FreezeLimit (or MXID < MultiXactCutoff) */
2303 202 : if (vacrel->aggressive)
2304 : {
2305 : /*
2306 : * Aggressive VACUUMs must always be able to advance rel's
2307 : * relfrozenxid to a value >= FreezeLimit (and be able to
2308 : * advance rel's relminmxid to a value >= MultiXactCutoff).
2309 : * The ongoing aggressive VACUUM won't be able to do that
2310 : * unless it can freeze an XID (or MXID) from this tuple now.
2311 : *
2312 : * The only safe option is to have caller perform processing
2313 : * of this page using lazy_scan_prune. Caller might have to
2314 : * wait a while for a cleanup lock, but it can't be helped.
2315 : */
2316 74 : vacrel->offnum = InvalidOffsetNumber;
2317 74 : return false;
2318 : }
2319 :
2320 : /*
2321 : * Non-aggressive VACUUMs are under no obligation to advance
2322 : * relfrozenxid (even by one XID). We can be much laxer here.
2323 : *
2324 : * Currently we always just accept an older final relfrozenxid
2325 : * and/or relminmxid value. We never make caller wait or work a
2326 : * little harder, even when it likely makes sense to do so.
2327 : */
2328 : }
2329 :
2330 6316 : ItemPointerSet(&(tuple.t_self), blkno, offnum);
2331 6316 : tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid);
2332 6316 : tuple.t_len = ItemIdGetLength(itemid);
2333 6316 : tuple.t_tableOid = RelationGetRelid(vacrel->rel);
2334 :
2335 6316 : switch (HeapTupleSatisfiesVacuum(&tuple, vacrel->cutoffs.OldestXmin,
2336 : buf))
2337 : {
2338 6222 : case HEAPTUPLE_DELETE_IN_PROGRESS:
2339 : case HEAPTUPLE_LIVE:
2340 :
2341 : /*
2342 : * Count both cases as live, just like lazy_scan_prune
2343 : */
2344 6222 : live_tuples++;
2345 :
2346 6222 : break;
2347 82 : case HEAPTUPLE_DEAD:
2348 :
2349 : /*
2350 : * There is some useful work for pruning to do, that won't be
2351 : * done due to failure to get a cleanup lock.
2352 : */
2353 82 : missed_dead_tuples++;
2354 82 : break;
2355 4 : case HEAPTUPLE_RECENTLY_DEAD:
2356 :
2357 : /*
2358 : * Count in recently_dead_tuples, just like lazy_scan_prune
2359 : */
2360 4 : recently_dead_tuples++;
2361 4 : break;
2362 8 : case HEAPTUPLE_INSERT_IN_PROGRESS:
2363 :
2364 : /*
2365 : * Do not count these rows as live, just like lazy_scan_prune
2366 : */
2367 8 : break;
2368 0 : default:
2369 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
2370 : break;
2371 : }
2372 : }
2373 :
2374 194 : vacrel->offnum = InvalidOffsetNumber;
2375 :
2376 : /*
2377 : * By here we know for sure that caller can put off freezing and pruning
2378 : * this particular page until the next VACUUM. Remember its details now.
2379 : * (lazy_scan_prune expects a clean slate, so we have to do this last.)
2380 : */
2381 194 : vacrel->NewRelfrozenXid = NoFreezePageRelfrozenXid;
2382 194 : vacrel->NewRelminMxid = NoFreezePageRelminMxid;
2383 :
2384 : /* Save any LP_DEAD items found on the page in dead_items */
2385 194 : if (vacrel->nindexes == 0)
2386 : {
2387 : /* Using one-pass strategy (since table has no indexes) */
2388 0 : if (lpdead_items > 0)
2389 : {
2390 : /*
2391 : * Perfunctory handling for the corner case where a single pass
2392 : * strategy VACUUM cannot get a cleanup lock, and it turns out
2393 : * that there is one or more LP_DEAD items: just count the LP_DEAD
2394 : * items as missed_dead_tuples instead. (This is a bit dishonest,
2395 : * but it beats having to maintain specialized heap vacuuming code
2396 : * forever, for vanishingly little benefit.)
2397 : */
2398 0 : hastup = true;
2399 0 : missed_dead_tuples += lpdead_items;
2400 : }
2401 : }
2402 194 : else if (lpdead_items > 0)
2403 : {
2404 : /*
2405 : * Page has LP_DEAD items, and so any references/TIDs that remain in
2406 : * indexes will be deleted during index vacuuming (and then marked
2407 : * LP_UNUSED in the heap)
2408 : */
2409 0 : vacrel->lpdead_item_pages++;
2410 :
2411 0 : dead_items_add(vacrel, blkno, deadoffsets, lpdead_items);
2412 :
2413 0 : vacrel->lpdead_items += lpdead_items;
2414 : }
2415 :
2416 : /*
2417 : * Finally, add relevant page-local counts to whole-VACUUM counts
2418 : */
2419 194 : vacrel->live_tuples += live_tuples;
2420 194 : vacrel->recently_dead_tuples += recently_dead_tuples;
2421 194 : vacrel->missed_dead_tuples += missed_dead_tuples;
2422 194 : if (missed_dead_tuples > 0)
2423 6 : vacrel->missed_dead_pages++;
2424 :
2425 : /* Can't truncate this page */
2426 194 : if (hastup)
2427 194 : vacrel->nonempty_pages = blkno + 1;
2428 :
2429 : /* Did we find LP_DEAD items? */
2430 194 : *has_lpdead_items = (lpdead_items > 0);
2431 :
2432 : /* Caller won't need to call lazy_scan_prune with same page */
2433 194 : return true;
2434 : }
2435 :
2436 : /*
2437 : * Main entry point for index vacuuming and heap vacuuming.
2438 : *
2439 : * Removes items collected in dead_items from table's indexes, then marks the
2440 : * same items LP_UNUSED in the heap. See the comments above lazy_scan_heap
2441 : * for full details.
2442 : *
2443 : * Also empties dead_items, freeing up space for later TIDs.
2444 : *
2445 : * We may choose to bypass index vacuuming at this point, though only when the
2446 : * ongoing VACUUM operation will definitely only have one index scan/round of
2447 : * index vacuuming.
2448 : */
2449 : static void
2450 1340 : lazy_vacuum(LVRelState *vacrel)
2451 : {
2452 : bool bypass;
2453 :
2454 : /* Should not end up here with no indexes */
2455 : Assert(vacrel->nindexes > 0);
2456 : Assert(vacrel->lpdead_item_pages > 0);
2457 :
2458 1340 : if (!vacrel->do_index_vacuuming)
2459 : {
2460 : Assert(!vacrel->do_index_cleanup);
2461 26 : dead_items_reset(vacrel);
2462 26 : return;
2463 : }
2464 :
2465 : /*
2466 : * Consider bypassing index vacuuming (and heap vacuuming) entirely.
2467 : *
2468 : * We currently only do this in cases where the number of LP_DEAD items
2469 : * for the entire VACUUM operation is close to zero. This avoids sharp
2470 : * discontinuities in the duration and overhead of successive VACUUM
2471 : * operations that run against the same table with a fixed workload.
2472 : * Ideally, successive VACUUM operations will behave as if there are
2473 : * exactly zero LP_DEAD items in cases where there are close to zero.
2474 : *
2475 : * This is likely to be helpful with a table that is continually affected
2476 : * by UPDATEs that can mostly apply the HOT optimization, but occasionally
2477 : * have small aberrations that lead to just a few heap pages retaining
2478 : * only one or two LP_DEAD items. This is pretty common; even when the
2479 : * DBA goes out of their way to make UPDATEs use HOT, it is practically
2480 : * impossible to predict whether HOT will be applied in 100% of cases.
2481 : * It's far easier to ensure that 99%+ of all UPDATEs against a table use
2482 : * HOT through careful tuning.
2483 : */
2484 1314 : bypass = false;
2485 1314 : if (vacrel->consider_bypass_optimization && vacrel->rel_pages > 0)
2486 : {
2487 : BlockNumber threshold;
2488 :
2489 : Assert(vacrel->num_index_scans == 0);
2490 : Assert(vacrel->lpdead_items == vacrel->dead_items_info->num_items);
2491 : Assert(vacrel->do_index_vacuuming);
2492 : Assert(vacrel->do_index_cleanup);
2493 :
2494 : /*
2495 : * This crossover point at which we'll start to do index vacuuming is
2496 : * expressed as a percentage of the total number of heap pages in the
2497 : * table that are known to have at least one LP_DEAD item. This is
2498 : * much more important than the total number of LP_DEAD items, since
2499 : * it's a proxy for the number of heap pages whose visibility map bits
2500 : * cannot be set on account of bypassing index and heap vacuuming.
2501 : *
2502 : * We apply one further precautionary test: the space currently used
2503 : * to store the TIDs (TIDs that now all point to LP_DEAD items) must
2504 : * not exceed 32MB. This limits the risk that we will bypass index
2505 : * vacuuming again and again until eventually there is a VACUUM whose
2506 : * dead_items space is not CPU cache resident.
2507 : *
2508 : * We don't take any special steps to remember the LP_DEAD items (such
2509 : * as counting them in our final update to the stats system) when the
2510 : * optimization is applied. Though the accounting used in analyze.c's
2511 : * acquire_sample_rows() will recognize the same LP_DEAD items as dead
2512 : * rows in its own stats report, that's okay. The discrepancy should
2513 : * be negligible. If this optimization is ever expanded to cover more
2514 : * cases then this may need to be reconsidered.
2515 : */
2516 1294 : threshold = (double) vacrel->rel_pages * BYPASS_THRESHOLD_PAGES;
2517 1298 : bypass = (vacrel->lpdead_item_pages < threshold &&
2518 4 : TidStoreMemoryUsage(vacrel->dead_items) < 32 * 1024 * 1024);
2519 : }
2520 :
2521 1314 : if (bypass)
2522 : {
2523 : /*
2524 : * There are almost zero TIDs. Behave as if there were precisely
2525 : * zero: bypass index vacuuming, but do index cleanup.
2526 : *
2527 : * We expect that the ongoing VACUUM operation will finish very
2528 : * quickly, so there is no point in considering speeding up as a
2529 : * failsafe against wraparound failure. (Index cleanup is expected to
2530 : * finish very quickly in cases where there were no ambulkdelete()
2531 : * calls.)
2532 : */
2533 4 : vacrel->do_index_vacuuming = false;
2534 : }
2535 1310 : else if (lazy_vacuum_all_indexes(vacrel))
2536 : {
2537 : /*
2538 : * We successfully completed a round of index vacuuming. Do related
2539 : * heap vacuuming now.
2540 : */
2541 1310 : lazy_vacuum_heap_rel(vacrel);
2542 : }
2543 : else
2544 : {
2545 : /*
2546 : * Failsafe case.
2547 : *
2548 : * We attempted index vacuuming, but didn't finish a full round/full
2549 : * index scan. This happens when relfrozenxid or relminmxid is too
2550 : * far in the past.
2551 : *
2552 : * From this point on the VACUUM operation will do no further index
2553 : * vacuuming or heap vacuuming. This VACUUM operation won't end up
2554 : * back here again.
2555 : */
2556 : Assert(VacuumFailsafeActive);
2557 : }
2558 :
2559 : /*
2560 : * Forget the LP_DEAD items that we just vacuumed (or just decided to not
2561 : * vacuum)
2562 : */
2563 1314 : dead_items_reset(vacrel);
2564 : }
2565 :
2566 : /*
2567 : * lazy_vacuum_all_indexes() -- Main entry for index vacuuming
2568 : *
2569 : * Returns true in the common case when all indexes were successfully
2570 : * vacuumed. Returns false in rare cases where we determined that the ongoing
2571 : * VACUUM operation is at risk of taking too long to finish, leading to
2572 : * wraparound failure.
2573 : */
2574 : static bool
2575 1310 : lazy_vacuum_all_indexes(LVRelState *vacrel)
2576 : {
2577 1310 : bool allindexes = true;
2578 1310 : double old_live_tuples = vacrel->rel->rd_rel->reltuples;
2579 1310 : const int progress_start_index[] = {
2580 : PROGRESS_VACUUM_PHASE,
2581 : PROGRESS_VACUUM_INDEXES_TOTAL
2582 : };
2583 1310 : const int progress_end_index[] = {
2584 : PROGRESS_VACUUM_INDEXES_TOTAL,
2585 : PROGRESS_VACUUM_INDEXES_PROCESSED,
2586 : PROGRESS_VACUUM_NUM_INDEX_VACUUMS
2587 : };
2588 : int64 progress_start_val[2];
2589 : int64 progress_end_val[3];
2590 :
2591 : Assert(vacrel->nindexes > 0);
2592 : Assert(vacrel->do_index_vacuuming);
2593 : Assert(vacrel->do_index_cleanup);
2594 :
2595 : /* Precheck for XID wraparound emergencies */
2596 1310 : if (lazy_check_wraparound_failsafe(vacrel))
2597 : {
2598 : /* Wraparound emergency -- don't even start an index scan */
2599 0 : return false;
2600 : }
2601 :
2602 : /*
2603 : * Report that we are now vacuuming indexes and the number of indexes to
2604 : * vacuum.
2605 : */
2606 1310 : progress_start_val[0] = PROGRESS_VACUUM_PHASE_VACUUM_INDEX;
2607 1310 : progress_start_val[1] = vacrel->nindexes;
2608 1310 : pgstat_progress_update_multi_param(2, progress_start_index, progress_start_val);
2609 :
2610 1310 : if (!ParallelVacuumIsActive(vacrel))
2611 : {
2612 3836 : for (int idx = 0; idx < vacrel->nindexes; idx++)
2613 : {
2614 2536 : Relation indrel = vacrel->indrels[idx];
2615 2536 : IndexBulkDeleteResult *istat = vacrel->indstats[idx];
2616 :
2617 2536 : vacrel->indstats[idx] = lazy_vacuum_one_index(indrel, istat,
2618 : old_live_tuples,
2619 : vacrel);
2620 :
2621 : /* Report the number of indexes vacuumed */
2622 2536 : pgstat_progress_update_param(PROGRESS_VACUUM_INDEXES_PROCESSED,
2623 2536 : idx + 1);
2624 :
2625 2536 : if (lazy_check_wraparound_failsafe(vacrel))
2626 : {
2627 : /* Wraparound emergency -- end current index scan */
2628 0 : allindexes = false;
2629 0 : break;
2630 : }
2631 : }
2632 : }
2633 : else
2634 : {
2635 : /* Outsource everything to parallel variant */
2636 10 : parallel_vacuum_bulkdel_all_indexes(vacrel->pvs, old_live_tuples,
2637 : vacrel->num_index_scans);
2638 :
2639 : /*
2640 : * Do a postcheck to consider applying wraparound failsafe now. Note
2641 : * that parallel VACUUM only gets the precheck and this postcheck.
2642 : */
2643 10 : if (lazy_check_wraparound_failsafe(vacrel))
2644 0 : allindexes = false;
2645 : }
2646 :
2647 : /*
2648 : * We delete all LP_DEAD items from the first heap pass in all indexes on
2649 : * each call here (except calls where we choose to do the failsafe). This
2650 : * makes the next call to lazy_vacuum_heap_rel() safe (except in the event
2651 : * of the failsafe triggering, which prevents the next call from taking
2652 : * place).
2653 : */
2654 : Assert(vacrel->num_index_scans > 0 ||
2655 : vacrel->dead_items_info->num_items == vacrel->lpdead_items);
2656 : Assert(allindexes || VacuumFailsafeActive);
2657 :
2658 : /*
2659 : * Increase and report the number of index scans. Also, we reset
2660 : * PROGRESS_VACUUM_INDEXES_TOTAL and PROGRESS_VACUUM_INDEXES_PROCESSED.
2661 : *
2662 : * We deliberately include the case where we started a round of bulk
2663 : * deletes that we weren't able to finish due to the failsafe triggering.
2664 : */
2665 1310 : vacrel->num_index_scans++;
2666 1310 : progress_end_val[0] = 0;
2667 1310 : progress_end_val[1] = 0;
2668 1310 : progress_end_val[2] = vacrel->num_index_scans;
2669 1310 : pgstat_progress_update_multi_param(3, progress_end_index, progress_end_val);
2670 :
2671 1310 : return allindexes;
2672 : }
2673 :
2674 : /*
2675 : * Read stream callback for vacuum's third phase (second pass over the heap).
2676 : * Gets the next block from the TID store and returns it or InvalidBlockNumber
2677 : * if there are no further blocks to vacuum.
2678 : *
2679 : * NB: Assumed to be safe to use with READ_STREAM_USE_BATCHING.
2680 : */
2681 : static BlockNumber
2682 29522 : vacuum_reap_lp_read_stream_next(ReadStream *stream,
2683 : void *callback_private_data,
2684 : void *per_buffer_data)
2685 : {
2686 29522 : TidStoreIter *iter = callback_private_data;
2687 : TidStoreIterResult *iter_result;
2688 :
2689 29522 : iter_result = TidStoreIterateNext(iter);
2690 29522 : if (iter_result == NULL)
2691 1310 : return InvalidBlockNumber;
2692 :
2693 : /*
2694 : * Save the TidStoreIterResult for later, so we can extract the offsets.
2695 : * It is safe to copy the result, according to TidStoreIterateNext().
2696 : */
2697 28212 : memcpy(per_buffer_data, iter_result, sizeof(*iter_result));
2698 :
2699 28212 : return iter_result->blkno;
2700 : }
2701 :
2702 : /*
2703 : * lazy_vacuum_heap_rel() -- second pass over the heap for two pass strategy
2704 : *
2705 : * This routine marks LP_DEAD items in vacrel->dead_items as LP_UNUSED. Pages
2706 : * that never had lazy_scan_prune record LP_DEAD items are not visited at all.
2707 : *
2708 : * We may also be able to truncate the line pointer array of the heap pages we
2709 : * visit. If there is a contiguous group of LP_UNUSED items at the end of the
2710 : * array, it can be reclaimed as free space. These LP_UNUSED items usually
2711 : * start out as LP_DEAD items recorded by lazy_scan_prune (we set items from
2712 : * each page to LP_UNUSED, and then consider if it's possible to truncate the
2713 : * page's line pointer array).
2714 : *
2715 : * Note: the reason for doing this as a second pass is we cannot remove the
2716 : * tuples until we've removed their index entries, and we want to process
2717 : * index entry removal in batches as large as possible.
2718 : */
2719 : static void
2720 1310 : lazy_vacuum_heap_rel(LVRelState *vacrel)
2721 : {
2722 : ReadStream *stream;
2723 1310 : BlockNumber vacuumed_pages = 0;
2724 1310 : Buffer vmbuffer = InvalidBuffer;
2725 : LVSavedErrInfo saved_err_info;
2726 : TidStoreIter *iter;
2727 :
2728 : Assert(vacrel->do_index_vacuuming);
2729 : Assert(vacrel->do_index_cleanup);
2730 : Assert(vacrel->num_index_scans > 0);
2731 :
2732 : /* Report that we are now vacuuming the heap */
2733 1310 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
2734 : PROGRESS_VACUUM_PHASE_VACUUM_HEAP);
2735 :
2736 : /* Update error traceback information */
2737 1310 : update_vacuum_error_info(vacrel, &saved_err_info,
2738 : VACUUM_ERRCB_PHASE_VACUUM_HEAP,
2739 : InvalidBlockNumber, InvalidOffsetNumber);
2740 :
2741 1310 : iter = TidStoreBeginIterate(vacrel->dead_items);
2742 :
2743 : /*
2744 : * Set up the read stream for vacuum's second pass through the heap.
2745 : *
2746 : * It is safe to use batchmode, as vacuum_reap_lp_read_stream_next() does
2747 : * not need to wait for IO and does not perform locking. Once we support
2748 : * parallelism it should still be fine, as presumably the holder of locks
2749 : * would never be blocked by IO while holding the lock.
2750 : */
2751 1310 : stream = read_stream_begin_relation(READ_STREAM_MAINTENANCE |
2752 : READ_STREAM_USE_BATCHING,
2753 : vacrel->bstrategy,
2754 : vacrel->rel,
2755 : MAIN_FORKNUM,
2756 : vacuum_reap_lp_read_stream_next,
2757 : iter,
2758 : sizeof(TidStoreIterResult));
2759 :
2760 : while (true)
2761 28212 : {
2762 : BlockNumber blkno;
2763 : Buffer buf;
2764 : Page page;
2765 : TidStoreIterResult *iter_result;
2766 : Size freespace;
2767 : OffsetNumber offsets[MaxOffsetNumber];
2768 : int num_offsets;
2769 :
2770 29522 : vacuum_delay_point(false);
2771 :
2772 29522 : buf = read_stream_next_buffer(stream, (void **) &iter_result);
2773 :
2774 : /* The relation is exhausted */
2775 29522 : if (!BufferIsValid(buf))
2776 1310 : break;
2777 :
2778 28212 : vacrel->blkno = blkno = BufferGetBlockNumber(buf);
2779 :
2780 : Assert(iter_result);
2781 28212 : num_offsets = TidStoreGetBlockOffsets(iter_result, offsets, lengthof(offsets));
2782 : Assert(num_offsets <= lengthof(offsets));
2783 :
2784 : /*
2785 : * Pin the visibility map page in case we need to mark the page
2786 : * all-visible. In most cases this will be very cheap, because we'll
2787 : * already have the correct page pinned anyway.
2788 : */
2789 28212 : visibilitymap_pin(vacrel->rel, blkno, &vmbuffer);
2790 :
2791 : /* We need a non-cleanup exclusive lock to mark dead_items unused */
2792 28212 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
2793 28212 : lazy_vacuum_heap_page(vacrel, blkno, buf, offsets,
2794 : num_offsets, vmbuffer);
2795 :
2796 : /* Now that we've vacuumed the page, record its available space */
2797 28212 : page = BufferGetPage(buf);
2798 28212 : freespace = PageGetHeapFreeSpace(page);
2799 :
2800 28212 : UnlockReleaseBuffer(buf);
2801 28212 : RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
2802 28212 : vacuumed_pages++;
2803 : }
2804 :
2805 1310 : read_stream_end(stream);
2806 1310 : TidStoreEndIterate(iter);
2807 :
2808 1310 : vacrel->blkno = InvalidBlockNumber;
2809 1310 : if (BufferIsValid(vmbuffer))
2810 1310 : ReleaseBuffer(vmbuffer);
2811 :
2812 : /*
2813 : * We set all LP_DEAD items from the first heap pass to LP_UNUSED during
2814 : * the second heap pass. No more, no less.
2815 : */
2816 : Assert(vacrel->num_index_scans > 1 ||
2817 : (vacrel->dead_items_info->num_items == vacrel->lpdead_items &&
2818 : vacuumed_pages == vacrel->lpdead_item_pages));
2819 :
2820 1310 : ereport(DEBUG2,
2821 : (errmsg("table \"%s\": removed %" PRId64 " dead item identifiers in %u pages",
2822 : vacrel->relname, vacrel->dead_items_info->num_items,
2823 : vacuumed_pages)));
2824 :
2825 : /* Revert to the previous phase information for error traceback */
2826 1310 : restore_vacuum_error_info(vacrel, &saved_err_info);
2827 1310 : }
2828 :
2829 : /*
2830 : * lazy_vacuum_heap_page() -- free page's LP_DEAD items listed in the
2831 : * vacrel->dead_items store.
2832 : *
2833 : * Caller must have an exclusive buffer lock on the buffer (though a full
2834 : * cleanup lock is also acceptable). vmbuffer must be valid and already have
2835 : * a pin on blkno's visibility map page.
2836 : */
2837 : static void
2838 28212 : lazy_vacuum_heap_page(LVRelState *vacrel, BlockNumber blkno, Buffer buffer,
2839 : OffsetNumber *deadoffsets, int num_offsets,
2840 : Buffer vmbuffer)
2841 : {
2842 28212 : Page page = BufferGetPage(buffer);
2843 : OffsetNumber unused[MaxHeapTuplesPerPage];
2844 28212 : int nunused = 0;
2845 : TransactionId visibility_cutoff_xid;
2846 : bool all_frozen;
2847 : LVSavedErrInfo saved_err_info;
2848 :
2849 : Assert(vacrel->do_index_vacuuming);
2850 :
2851 28212 : pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_VACUUMED, blkno);
2852 :
2853 : /* Update error traceback information */
2854 28212 : update_vacuum_error_info(vacrel, &saved_err_info,
2855 : VACUUM_ERRCB_PHASE_VACUUM_HEAP, blkno,
2856 : InvalidOffsetNumber);
2857 :
2858 28212 : START_CRIT_SECTION();
2859 :
2860 1725764 : for (int i = 0; i < num_offsets; i++)
2861 : {
2862 : ItemId itemid;
2863 1697552 : OffsetNumber toff = deadoffsets[i];
2864 :
2865 1697552 : itemid = PageGetItemId(page, toff);
2866 :
2867 : Assert(ItemIdIsDead(itemid) && !ItemIdHasStorage(itemid));
2868 1697552 : ItemIdSetUnused(itemid);
2869 1697552 : unused[nunused++] = toff;
2870 : }
2871 :
2872 : Assert(nunused > 0);
2873 :
2874 : /* Attempt to truncate line pointer array now */
2875 28212 : PageTruncateLinePointerArray(page);
2876 :
2877 : /*
2878 : * Mark buffer dirty before we write WAL.
2879 : */
2880 28212 : MarkBufferDirty(buffer);
2881 :
2882 : /* XLOG stuff */
2883 28212 : if (RelationNeedsWAL(vacrel->rel))
2884 : {
2885 26516 : log_heap_prune_and_freeze(vacrel->rel, buffer,
2886 : InvalidTransactionId,
2887 : false, /* no cleanup lock required */
2888 : PRUNE_VACUUM_CLEANUP,
2889 : NULL, 0, /* frozen */
2890 : NULL, 0, /* redirected */
2891 : NULL, 0, /* dead */
2892 : unused, nunused);
2893 : }
2894 :
2895 : /*
2896 : * End critical section, so we safely can do visibility tests (which
2897 : * possibly need to perform IO and allocate memory!). If we crash now the
2898 : * page (including the corresponding vm bit) might not be marked all
2899 : * visible, but that's fine. A later vacuum will fix that.
2900 : */
2901 28212 : END_CRIT_SECTION();
2902 :
2903 : /*
2904 : * Now that we have removed the LP_DEAD items from the page, once again
2905 : * check if the page has become all-visible. The page is already marked
2906 : * dirty, exclusively locked, and, if needed, a full page image has been
2907 : * emitted.
2908 : */
2909 : Assert(!PageIsAllVisible(page));
2910 28212 : if (heap_page_is_all_visible(vacrel, buffer, &visibility_cutoff_xid,
2911 : &all_frozen))
2912 : {
2913 27554 : uint8 flags = VISIBILITYMAP_ALL_VISIBLE;
2914 :
2915 27554 : if (all_frozen)
2916 : {
2917 : Assert(!TransactionIdIsValid(visibility_cutoff_xid));
2918 21964 : flags |= VISIBILITYMAP_ALL_FROZEN;
2919 : }
2920 :
2921 27554 : PageSetAllVisible(page);
2922 27554 : visibilitymap_set(vacrel->rel, blkno, buffer,
2923 : InvalidXLogRecPtr,
2924 : vmbuffer, visibility_cutoff_xid,
2925 : flags);
2926 :
2927 : /* Count the newly set VM page for logging */
2928 27554 : vacrel->vm_new_visible_pages++;
2929 27554 : if (all_frozen)
2930 21964 : vacrel->vm_new_visible_frozen_pages++;
2931 : }
2932 :
2933 : /* Revert to the previous phase information for error traceback */
2934 28212 : restore_vacuum_error_info(vacrel, &saved_err_info);
2935 28212 : }
2936 :
2937 : /*
2938 : * Trigger the failsafe to avoid wraparound failure when vacrel table has a
2939 : * relfrozenxid and/or relminmxid that is dangerously far in the past.
2940 : * Triggering the failsafe makes the ongoing VACUUM bypass any further index
2941 : * vacuuming and heap vacuuming. Truncating the heap is also bypassed.
2942 : *
2943 : * Any remaining work (work that VACUUM cannot just bypass) is typically sped
2944 : * up when the failsafe triggers. VACUUM stops applying any cost-based delay
2945 : * that it started out with.
2946 : *
2947 : * Returns true when failsafe has been triggered.
2948 : */
2949 : static bool
2950 249238 : lazy_check_wraparound_failsafe(LVRelState *vacrel)
2951 : {
2952 : /* Don't warn more than once per VACUUM */
2953 249238 : if (VacuumFailsafeActive)
2954 0 : return true;
2955 :
2956 249238 : if (unlikely(vacuum_xid_failsafe_check(&vacrel->cutoffs)))
2957 : {
2958 67628 : const int progress_index[] = {
2959 : PROGRESS_VACUUM_INDEXES_TOTAL,
2960 : PROGRESS_VACUUM_INDEXES_PROCESSED
2961 : };
2962 67628 : int64 progress_val[2] = {0, 0};
2963 :
2964 67628 : VacuumFailsafeActive = true;
2965 :
2966 : /*
2967 : * Abandon use of a buffer access strategy to allow use of all of
2968 : * shared buffers. We assume the caller who allocated the memory for
2969 : * the BufferAccessStrategy will free it.
2970 : */
2971 67628 : vacrel->bstrategy = NULL;
2972 :
2973 : /* Disable index vacuuming, index cleanup, and heap rel truncation */
2974 67628 : vacrel->do_index_vacuuming = false;
2975 67628 : vacrel->do_index_cleanup = false;
2976 67628 : vacrel->do_rel_truncate = false;
2977 :
2978 : /* Reset the progress counters */
2979 67628 : pgstat_progress_update_multi_param(2, progress_index, progress_val);
2980 :
2981 67628 : ereport(WARNING,
2982 : (errmsg("bypassing nonessential maintenance of table \"%s.%s.%s\" as a failsafe after %d index scans",
2983 : vacrel->dbname, vacrel->relnamespace, vacrel->relname,
2984 : vacrel->num_index_scans),
2985 : errdetail("The table's relfrozenxid or relminmxid is too far in the past."),
2986 : errhint("Consider increasing configuration parameter \"maintenance_work_mem\" or \"autovacuum_work_mem\".\n"
2987 : "You might also need to consider other ways for VACUUM to keep up with the allocation of transaction IDs.")));
2988 :
2989 : /* Stop applying cost limits from this point on */
2990 67628 : VacuumCostActive = false;
2991 67628 : VacuumCostBalance = 0;
2992 :
2993 67628 : return true;
2994 : }
2995 :
2996 181610 : return false;
2997 : }
2998 :
2999 : /*
3000 : * lazy_cleanup_all_indexes() -- cleanup all indexes of relation.
3001 : */
3002 : static void
3003 168810 : lazy_cleanup_all_indexes(LVRelState *vacrel)
3004 : {
3005 168810 : double reltuples = vacrel->new_rel_tuples;
3006 168810 : bool estimated_count = vacrel->scanned_pages < vacrel->rel_pages;
3007 168810 : const int progress_start_index[] = {
3008 : PROGRESS_VACUUM_PHASE,
3009 : PROGRESS_VACUUM_INDEXES_TOTAL
3010 : };
3011 168810 : const int progress_end_index[] = {
3012 : PROGRESS_VACUUM_INDEXES_TOTAL,
3013 : PROGRESS_VACUUM_INDEXES_PROCESSED
3014 : };
3015 : int64 progress_start_val[2];
3016 168810 : int64 progress_end_val[2] = {0, 0};
3017 :
3018 : Assert(vacrel->do_index_cleanup);
3019 : Assert(vacrel->nindexes > 0);
3020 :
3021 : /*
3022 : * Report that we are now cleaning up indexes and the number of indexes to
3023 : * cleanup.
3024 : */
3025 168810 : progress_start_val[0] = PROGRESS_VACUUM_PHASE_INDEX_CLEANUP;
3026 168810 : progress_start_val[1] = vacrel->nindexes;
3027 168810 : pgstat_progress_update_multi_param(2, progress_start_index, progress_start_val);
3028 :
3029 168810 : if (!ParallelVacuumIsActive(vacrel))
3030 : {
3031 434648 : for (int idx = 0; idx < vacrel->nindexes; idx++)
3032 : {
3033 265872 : Relation indrel = vacrel->indrels[idx];
3034 265872 : IndexBulkDeleteResult *istat = vacrel->indstats[idx];
3035 :
3036 531744 : vacrel->indstats[idx] =
3037 265872 : lazy_cleanup_one_index(indrel, istat, reltuples,
3038 : estimated_count, vacrel);
3039 :
3040 : /* Report the number of indexes cleaned up */
3041 265872 : pgstat_progress_update_param(PROGRESS_VACUUM_INDEXES_PROCESSED,
3042 265872 : idx + 1);
3043 : }
3044 : }
3045 : else
3046 : {
3047 : /* Outsource everything to parallel variant */
3048 34 : parallel_vacuum_cleanup_all_indexes(vacrel->pvs, reltuples,
3049 : vacrel->num_index_scans,
3050 : estimated_count);
3051 : }
3052 :
3053 : /* Reset the progress counters */
3054 168810 : pgstat_progress_update_multi_param(2, progress_end_index, progress_end_val);
3055 168810 : }
3056 :
3057 : /*
3058 : * lazy_vacuum_one_index() -- vacuum index relation.
3059 : *
3060 : * Delete all the index tuples containing a TID collected in
3061 : * vacrel->dead_items. Also update running statistics. Exact
3062 : * details depend on index AM's ambulkdelete routine.
3063 : *
3064 : * reltuples is the number of heap tuples to be passed to the
3065 : * bulkdelete callback. It's always assumed to be estimated.
3066 : * See indexam.sgml for more info.
3067 : *
3068 : * Returns bulk delete stats derived from input stats
3069 : */
3070 : static IndexBulkDeleteResult *
3071 2536 : lazy_vacuum_one_index(Relation indrel, IndexBulkDeleteResult *istat,
3072 : double reltuples, LVRelState *vacrel)
3073 : {
3074 : IndexVacuumInfo ivinfo;
3075 : LVSavedErrInfo saved_err_info;
3076 :
3077 2536 : ivinfo.index = indrel;
3078 2536 : ivinfo.heaprel = vacrel->rel;
3079 2536 : ivinfo.analyze_only = false;
3080 2536 : ivinfo.report_progress = false;
3081 2536 : ivinfo.estimated_count = true;
3082 2536 : ivinfo.message_level = DEBUG2;
3083 2536 : ivinfo.num_heap_tuples = reltuples;
3084 2536 : ivinfo.strategy = vacrel->bstrategy;
3085 :
3086 : /*
3087 : * Update error traceback information.
3088 : *
3089 : * The index name is saved during this phase and restored immediately
3090 : * after this phase. See vacuum_error_callback.
3091 : */
3092 : Assert(vacrel->indname == NULL);
3093 2536 : vacrel->indname = pstrdup(RelationGetRelationName(indrel));
3094 2536 : update_vacuum_error_info(vacrel, &saved_err_info,
3095 : VACUUM_ERRCB_PHASE_VACUUM_INDEX,
3096 : InvalidBlockNumber, InvalidOffsetNumber);
3097 :
3098 : /* Do bulk deletion */
3099 2536 : istat = vac_bulkdel_one_index(&ivinfo, istat, vacrel->dead_items,
3100 : vacrel->dead_items_info);
3101 :
3102 : /* Revert to the previous phase information for error traceback */
3103 2536 : restore_vacuum_error_info(vacrel, &saved_err_info);
3104 2536 : pfree(vacrel->indname);
3105 2536 : vacrel->indname = NULL;
3106 :
3107 2536 : return istat;
3108 : }
3109 :
3110 : /*
3111 : * lazy_cleanup_one_index() -- do post-vacuum cleanup for index relation.
3112 : *
3113 : * Calls index AM's amvacuumcleanup routine. reltuples is the number
3114 : * of heap tuples and estimated_count is true if reltuples is an
3115 : * estimated value. See indexam.sgml for more info.
3116 : *
3117 : * Returns bulk delete stats derived from input stats
3118 : */
3119 : static IndexBulkDeleteResult *
3120 265872 : lazy_cleanup_one_index(Relation indrel, IndexBulkDeleteResult *istat,
3121 : double reltuples, bool estimated_count,
3122 : LVRelState *vacrel)
3123 : {
3124 : IndexVacuumInfo ivinfo;
3125 : LVSavedErrInfo saved_err_info;
3126 :
3127 265872 : ivinfo.index = indrel;
3128 265872 : ivinfo.heaprel = vacrel->rel;
3129 265872 : ivinfo.analyze_only = false;
3130 265872 : ivinfo.report_progress = false;
3131 265872 : ivinfo.estimated_count = estimated_count;
3132 265872 : ivinfo.message_level = DEBUG2;
3133 :
3134 265872 : ivinfo.num_heap_tuples = reltuples;
3135 265872 : ivinfo.strategy = vacrel->bstrategy;
3136 :
3137 : /*
3138 : * Update error traceback information.
3139 : *
3140 : * The index name is saved during this phase and restored immediately
3141 : * after this phase. See vacuum_error_callback.
3142 : */
3143 : Assert(vacrel->indname == NULL);
3144 265872 : vacrel->indname = pstrdup(RelationGetRelationName(indrel));
3145 265872 : update_vacuum_error_info(vacrel, &saved_err_info,
3146 : VACUUM_ERRCB_PHASE_INDEX_CLEANUP,
3147 : InvalidBlockNumber, InvalidOffsetNumber);
3148 :
3149 265872 : istat = vac_cleanup_one_index(&ivinfo, istat);
3150 :
3151 : /* Revert to the previous phase information for error traceback */
3152 265872 : restore_vacuum_error_info(vacrel, &saved_err_info);
3153 265872 : pfree(vacrel->indname);
3154 265872 : vacrel->indname = NULL;
3155 :
3156 265872 : return istat;
3157 : }
3158 :
3159 : /*
3160 : * should_attempt_truncation - should we attempt to truncate the heap?
3161 : *
3162 : * Don't even think about it unless we have a shot at releasing a goodly
3163 : * number of pages. Otherwise, the time taken isn't worth it, mainly because
3164 : * an AccessExclusive lock must be replayed on any hot standby, where it can
3165 : * be particularly disruptive.
3166 : *
3167 : * Also don't attempt it if wraparound failsafe is in effect. The entire
3168 : * system might be refusing to allocate new XIDs at this point. The system
3169 : * definitely won't return to normal unless and until VACUUM actually advances
3170 : * the oldest relfrozenxid -- which hasn't happened for target rel just yet.
3171 : * If lazy_truncate_heap attempted to acquire an AccessExclusiveLock to
3172 : * truncate the table under these circumstances, an XID exhaustion error might
3173 : * make it impossible for VACUUM to fix the underlying XID exhaustion problem.
3174 : * There is very little chance of truncation working out when the failsafe is
3175 : * in effect in any case. lazy_scan_prune makes the optimistic assumption
3176 : * that any LP_DEAD items it encounters will always be LP_UNUSED by the time
3177 : * we're called.
3178 : */
3179 : static bool
3180 245382 : should_attempt_truncation(LVRelState *vacrel)
3181 : {
3182 : BlockNumber possibly_freeable;
3183 :
3184 245382 : if (!vacrel->do_rel_truncate || VacuumFailsafeActive)
3185 67918 : return false;
3186 :
3187 177464 : possibly_freeable = vacrel->rel_pages - vacrel->nonempty_pages;
3188 177464 : if (possibly_freeable > 0 &&
3189 358 : (possibly_freeable >= REL_TRUNCATE_MINIMUM ||
3190 358 : possibly_freeable >= vacrel->rel_pages / REL_TRUNCATE_FRACTION))
3191 330 : return true;
3192 :
3193 177134 : return false;
3194 : }
3195 :
3196 : /*
3197 : * lazy_truncate_heap - try to truncate off any empty pages at the end
3198 : */
3199 : static void
3200 330 : lazy_truncate_heap(LVRelState *vacrel)
3201 : {
3202 330 : BlockNumber orig_rel_pages = vacrel->rel_pages;
3203 : BlockNumber new_rel_pages;
3204 : bool lock_waiter_detected;
3205 : int lock_retry;
3206 :
3207 : /* Report that we are now truncating */
3208 330 : pgstat_progress_update_param(PROGRESS_VACUUM_PHASE,
3209 : PROGRESS_VACUUM_PHASE_TRUNCATE);
3210 :
3211 : /* Update error traceback information one last time */
3212 330 : update_vacuum_error_info(vacrel, NULL, VACUUM_ERRCB_PHASE_TRUNCATE,
3213 : vacrel->nonempty_pages, InvalidOffsetNumber);
3214 :
3215 : /*
3216 : * Loop until no more truncating can be done.
3217 : */
3218 : do
3219 : {
3220 : /*
3221 : * We need full exclusive lock on the relation in order to do
3222 : * truncation. If we can't get it, give up rather than waiting --- we
3223 : * don't want to block other backends, and we don't want to deadlock
3224 : * (which is quite possible considering we already hold a lower-grade
3225 : * lock).
3226 : */
3227 330 : lock_waiter_detected = false;
3228 330 : lock_retry = 0;
3229 : while (true)
3230 : {
3231 734 : if (ConditionalLockRelation(vacrel->rel, AccessExclusiveLock))
3232 326 : break;
3233 :
3234 : /*
3235 : * Check for interrupts while trying to (re-)acquire the exclusive
3236 : * lock.
3237 : */
3238 408 : CHECK_FOR_INTERRUPTS();
3239 :
3240 408 : if (++lock_retry > (VACUUM_TRUNCATE_LOCK_TIMEOUT /
3241 : VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL))
3242 : {
3243 : /*
3244 : * We failed to establish the lock in the specified number of
3245 : * retries. This means we give up truncating.
3246 : */
3247 4 : ereport(vacrel->verbose ? INFO : DEBUG2,
3248 : (errmsg("\"%s\": stopping truncate due to conflicting lock request",
3249 : vacrel->relname)));
3250 6 : return;
3251 : }
3252 :
3253 404 : (void) WaitLatch(MyLatch,
3254 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
3255 : VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL,
3256 : WAIT_EVENT_VACUUM_TRUNCATE);
3257 404 : ResetLatch(MyLatch);
3258 : }
3259 :
3260 : /*
3261 : * Now that we have exclusive lock, look to see if the rel has grown
3262 : * whilst we were vacuuming with non-exclusive lock. If so, give up;
3263 : * the newly added pages presumably contain non-deletable tuples.
3264 : */
3265 326 : new_rel_pages = RelationGetNumberOfBlocks(vacrel->rel);
3266 326 : if (new_rel_pages != orig_rel_pages)
3267 : {
3268 : /*
3269 : * Note: we intentionally don't update vacrel->rel_pages with the
3270 : * new rel size here. If we did, it would amount to assuming that
3271 : * the new pages are empty, which is unlikely. Leaving the numbers
3272 : * alone amounts to assuming that the new pages have the same
3273 : * tuple density as existing ones, which is less unlikely.
3274 : */
3275 0 : UnlockRelation(vacrel->rel, AccessExclusiveLock);
3276 0 : return;
3277 : }
3278 :
3279 : /*
3280 : * Scan backwards from the end to verify that the end pages actually
3281 : * contain no tuples. This is *necessary*, not optional, because
3282 : * other backends could have added tuples to these pages whilst we
3283 : * were vacuuming.
3284 : */
3285 326 : new_rel_pages = count_nondeletable_pages(vacrel, &lock_waiter_detected);
3286 326 : vacrel->blkno = new_rel_pages;
3287 :
3288 326 : if (new_rel_pages >= orig_rel_pages)
3289 : {
3290 : /* can't do anything after all */
3291 2 : UnlockRelation(vacrel->rel, AccessExclusiveLock);
3292 2 : return;
3293 : }
3294 :
3295 : /*
3296 : * Okay to truncate.
3297 : */
3298 324 : RelationTruncate(vacrel->rel, new_rel_pages);
3299 :
3300 : /*
3301 : * We can release the exclusive lock as soon as we have truncated.
3302 : * Other backends can't safely access the relation until they have
3303 : * processed the smgr invalidation that smgrtruncate sent out ... but
3304 : * that should happen as part of standard invalidation processing once
3305 : * they acquire lock on the relation.
3306 : */
3307 324 : UnlockRelation(vacrel->rel, AccessExclusiveLock);
3308 :
3309 : /*
3310 : * Update statistics. Here, it *is* correct to adjust rel_pages
3311 : * without also touching reltuples, since the tuple count wasn't
3312 : * changed by the truncation.
3313 : */
3314 324 : vacrel->removed_pages += orig_rel_pages - new_rel_pages;
3315 324 : vacrel->rel_pages = new_rel_pages;
3316 :
3317 324 : ereport(vacrel->verbose ? INFO : DEBUG2,
3318 : (errmsg("table \"%s\": truncated %u to %u pages",
3319 : vacrel->relname,
3320 : orig_rel_pages, new_rel_pages)));
3321 324 : orig_rel_pages = new_rel_pages;
3322 324 : } while (new_rel_pages > vacrel->nonempty_pages && lock_waiter_detected);
3323 : }
3324 :
3325 : /*
3326 : * Rescan end pages to verify that they are (still) empty of tuples.
3327 : *
3328 : * Returns number of nondeletable pages (last nonempty page + 1).
3329 : */
3330 : static BlockNumber
3331 326 : count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
3332 : {
3333 : BlockNumber blkno;
3334 : BlockNumber prefetchedUntil;
3335 : instr_time starttime;
3336 :
3337 : /* Initialize the starttime if we check for conflicting lock requests */
3338 326 : INSTR_TIME_SET_CURRENT(starttime);
3339 :
3340 : /*
3341 : * Start checking blocks at what we believe relation end to be and move
3342 : * backwards. (Strange coding of loop control is needed because blkno is
3343 : * unsigned.) To make the scan faster, we prefetch a few blocks at a time
3344 : * in forward direction, so that OS-level readahead can kick in.
3345 : */
3346 326 : blkno = vacrel->rel_pages;
3347 : StaticAssertStmt((PREFETCH_SIZE & (PREFETCH_SIZE - 1)) == 0,
3348 : "prefetch size must be power of 2");
3349 326 : prefetchedUntil = InvalidBlockNumber;
3350 5356 : while (blkno > vacrel->nonempty_pages)
3351 : {
3352 : Buffer buf;
3353 : Page page;
3354 : OffsetNumber offnum,
3355 : maxoff;
3356 : bool hastup;
3357 :
3358 : /*
3359 : * Check if another process requests a lock on our relation. We are
3360 : * holding an AccessExclusiveLock here, so they will be waiting. We
3361 : * only do this once per VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL, and we
3362 : * only check if that interval has elapsed once every 32 blocks to
3363 : * keep the number of system calls and actual shared lock table
3364 : * lookups to a minimum.
3365 : */
3366 5040 : if ((blkno % 32) == 0)
3367 : {
3368 : instr_time currenttime;
3369 : instr_time elapsed;
3370 :
3371 164 : INSTR_TIME_SET_CURRENT(currenttime);
3372 164 : elapsed = currenttime;
3373 164 : INSTR_TIME_SUBTRACT(elapsed, starttime);
3374 164 : if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
3375 : >= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
3376 : {
3377 0 : if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
3378 : {
3379 0 : ereport(vacrel->verbose ? INFO : DEBUG2,
3380 : (errmsg("table \"%s\": suspending truncate due to conflicting lock request",
3381 : vacrel->relname)));
3382 :
3383 0 : *lock_waiter_detected = true;
3384 0 : return blkno;
3385 : }
3386 0 : starttime = currenttime;
3387 : }
3388 : }
3389 :
3390 : /*
3391 : * We don't insert a vacuum delay point here, because we have an
3392 : * exclusive lock on the table which we want to hold for as short a
3393 : * time as possible. We still need to check for interrupts however.
3394 : */
3395 5040 : CHECK_FOR_INTERRUPTS();
3396 :
3397 5040 : blkno--;
3398 :
3399 : /* If we haven't prefetched this lot yet, do so now. */
3400 5040 : if (prefetchedUntil > blkno)
3401 : {
3402 : BlockNumber prefetchStart;
3403 : BlockNumber pblkno;
3404 :
3405 442 : prefetchStart = blkno & ~(PREFETCH_SIZE - 1);
3406 7388 : for (pblkno = prefetchStart; pblkno <= blkno; pblkno++)
3407 : {
3408 6946 : PrefetchBuffer(vacrel->rel, MAIN_FORKNUM, pblkno);
3409 6946 : CHECK_FOR_INTERRUPTS();
3410 : }
3411 442 : prefetchedUntil = prefetchStart;
3412 : }
3413 :
3414 5040 : buf = ReadBufferExtended(vacrel->rel, MAIN_FORKNUM, blkno, RBM_NORMAL,
3415 : vacrel->bstrategy);
3416 :
3417 : /* In this phase we only need shared access to the buffer */
3418 5040 : LockBuffer(buf, BUFFER_LOCK_SHARE);
3419 :
3420 5040 : page = BufferGetPage(buf);
3421 :
3422 5040 : if (PageIsNew(page) || PageIsEmpty(page))
3423 : {
3424 2384 : UnlockReleaseBuffer(buf);
3425 2384 : continue;
3426 : }
3427 :
3428 2656 : hastup = false;
3429 2656 : maxoff = PageGetMaxOffsetNumber(page);
3430 2656 : for (offnum = FirstOffsetNumber;
3431 5302 : offnum <= maxoff;
3432 2646 : offnum = OffsetNumberNext(offnum))
3433 : {
3434 : ItemId itemid;
3435 :
3436 2656 : itemid = PageGetItemId(page, offnum);
3437 :
3438 : /*
3439 : * Note: any non-unused item should be taken as a reason to keep
3440 : * this page. Even an LP_DEAD item makes truncation unsafe, since
3441 : * we must not have cleaned out its index entries.
3442 : */
3443 2656 : if (ItemIdIsUsed(itemid))
3444 : {
3445 10 : hastup = true;
3446 10 : break; /* can stop scanning */
3447 : }
3448 : } /* scan along page */
3449 :
3450 2656 : UnlockReleaseBuffer(buf);
3451 :
3452 : /* Done scanning if we found a tuple here */
3453 2656 : if (hastup)
3454 10 : return blkno + 1;
3455 : }
3456 :
3457 : /*
3458 : * If we fall out of the loop, all the previously-thought-to-be-empty
3459 : * pages still are; we need not bother to look at the last known-nonempty
3460 : * page.
3461 : */
3462 316 : return vacrel->nonempty_pages;
3463 : }
3464 :
3465 : /*
3466 : * Allocate dead_items and dead_items_info (either using palloc, or in dynamic
3467 : * shared memory). Sets both in vacrel for caller.
3468 : *
3469 : * Also handles parallel initialization as part of allocating dead_items in
3470 : * DSM when required.
3471 : */
3472 : static void
3473 245382 : dead_items_alloc(LVRelState *vacrel, int nworkers)
3474 : {
3475 : VacDeadItemsInfo *dead_items_info;
3476 709958 : int vac_work_mem = AmAutoVacuumWorkerProcess() &&
3477 219194 : autovacuum_work_mem != -1 ?
3478 464576 : autovacuum_work_mem : maintenance_work_mem;
3479 :
3480 : /*
3481 : * Initialize state for a parallel vacuum. As of now, only one worker can
3482 : * be used for an index, so we invoke parallelism only if there are at
3483 : * least two indexes on a table.
3484 : */
3485 245382 : if (nworkers >= 0 && vacrel->nindexes > 1 && vacrel->do_index_vacuuming)
3486 : {
3487 : /*
3488 : * Since parallel workers cannot access data in temporary tables, we
3489 : * can't perform parallel vacuum on them.
3490 : */
3491 10394 : if (RelationUsesLocalBuffers(vacrel->rel))
3492 : {
3493 : /*
3494 : * Give warning only if the user explicitly tries to perform a
3495 : * parallel vacuum on the temporary table.
3496 : */
3497 6 : if (nworkers > 0)
3498 6 : ereport(WARNING,
3499 : (errmsg("disabling parallel option of vacuum on \"%s\" --- cannot vacuum temporary tables in parallel",
3500 : vacrel->relname)));
3501 : }
3502 : else
3503 10388 : vacrel->pvs = parallel_vacuum_init(vacrel->rel, vacrel->indrels,
3504 : vacrel->nindexes, nworkers,
3505 : vac_work_mem,
3506 10388 : vacrel->verbose ? INFO : DEBUG2,
3507 : vacrel->bstrategy);
3508 :
3509 : /*
3510 : * If parallel mode started, dead_items and dead_items_info spaces are
3511 : * allocated in DSM.
3512 : */
3513 10394 : if (ParallelVacuumIsActive(vacrel))
3514 : {
3515 34 : vacrel->dead_items = parallel_vacuum_get_dead_items(vacrel->pvs,
3516 : &vacrel->dead_items_info);
3517 34 : return;
3518 : }
3519 : }
3520 :
3521 : /*
3522 : * Serial VACUUM case. Allocate both dead_items and dead_items_info
3523 : * locally.
3524 : */
3525 :
3526 245348 : dead_items_info = (VacDeadItemsInfo *) palloc(sizeof(VacDeadItemsInfo));
3527 245348 : dead_items_info->max_bytes = vac_work_mem * (Size) 1024;
3528 245348 : dead_items_info->num_items = 0;
3529 245348 : vacrel->dead_items_info = dead_items_info;
3530 :
3531 245348 : vacrel->dead_items = TidStoreCreateLocal(dead_items_info->max_bytes, true);
3532 : }
3533 :
3534 : /*
3535 : * Add the given block number and offset numbers to dead_items.
3536 : */
3537 : static void
3538 32874 : dead_items_add(LVRelState *vacrel, BlockNumber blkno, OffsetNumber *offsets,
3539 : int num_offsets)
3540 : {
3541 32874 : const int prog_index[2] = {
3542 : PROGRESS_VACUUM_NUM_DEAD_ITEM_IDS,
3543 : PROGRESS_VACUUM_DEAD_TUPLE_BYTES
3544 : };
3545 : int64 prog_val[2];
3546 :
3547 32874 : TidStoreSetBlockOffsets(vacrel->dead_items, blkno, offsets, num_offsets);
3548 32874 : vacrel->dead_items_info->num_items += num_offsets;
3549 :
3550 : /* update the progress information */
3551 32874 : prog_val[0] = vacrel->dead_items_info->num_items;
3552 32874 : prog_val[1] = TidStoreMemoryUsage(vacrel->dead_items);
3553 32874 : pgstat_progress_update_multi_param(2, prog_index, prog_val);
3554 32874 : }
3555 :
3556 : /*
3557 : * Forget all collected dead items.
3558 : */
3559 : static void
3560 1340 : dead_items_reset(LVRelState *vacrel)
3561 : {
3562 1340 : if (ParallelVacuumIsActive(vacrel))
3563 : {
3564 10 : parallel_vacuum_reset_dead_items(vacrel->pvs);
3565 10 : return;
3566 : }
3567 :
3568 : /* Recreate the tidstore with the same max_bytes limitation */
3569 1330 : TidStoreDestroy(vacrel->dead_items);
3570 1330 : vacrel->dead_items = TidStoreCreateLocal(vacrel->dead_items_info->max_bytes, true);
3571 :
3572 : /* Reset the counter */
3573 1330 : vacrel->dead_items_info->num_items = 0;
3574 : }
3575 :
3576 : /*
3577 : * Perform cleanup for resources allocated in dead_items_alloc
3578 : */
3579 : static void
3580 245382 : dead_items_cleanup(LVRelState *vacrel)
3581 : {
3582 245382 : if (!ParallelVacuumIsActive(vacrel))
3583 : {
3584 : /* Don't bother with pfree here */
3585 245348 : return;
3586 : }
3587 :
3588 : /* End parallel mode */
3589 34 : parallel_vacuum_end(vacrel->pvs, vacrel->indstats);
3590 34 : vacrel->pvs = NULL;
3591 : }
3592 :
3593 : /*
3594 : * Check if every tuple in the given page is visible to all current and future
3595 : * transactions. Also return the visibility_cutoff_xid which is the highest
3596 : * xmin amongst the visible tuples. Set *all_frozen to true if every tuple
3597 : * on this page is frozen.
3598 : *
3599 : * This is a stripped down version of lazy_scan_prune(). If you change
3600 : * anything here, make sure that everything stays in sync. Note that an
3601 : * assertion calls us to verify that everybody still agrees. Be sure to avoid
3602 : * introducing new side-effects here.
3603 : */
3604 : static bool
3605 28212 : heap_page_is_all_visible(LVRelState *vacrel, Buffer buf,
3606 : TransactionId *visibility_cutoff_xid,
3607 : bool *all_frozen)
3608 : {
3609 28212 : Page page = BufferGetPage(buf);
3610 28212 : BlockNumber blockno = BufferGetBlockNumber(buf);
3611 : OffsetNumber offnum,
3612 : maxoff;
3613 28212 : bool all_visible = true;
3614 :
3615 28212 : *visibility_cutoff_xid = InvalidTransactionId;
3616 28212 : *all_frozen = true;
3617 :
3618 28212 : maxoff = PageGetMaxOffsetNumber(page);
3619 28212 : for (offnum = FirstOffsetNumber;
3620 1374904 : offnum <= maxoff && all_visible;
3621 1346692 : offnum = OffsetNumberNext(offnum))
3622 : {
3623 : ItemId itemid;
3624 : HeapTupleData tuple;
3625 :
3626 : /*
3627 : * Set the offset number so that we can display it along with any
3628 : * error that occurred while processing this tuple.
3629 : */
3630 1346698 : vacrel->offnum = offnum;
3631 1346698 : itemid = PageGetItemId(page, offnum);
3632 :
3633 : /* Unused or redirect line pointers are of no interest */
3634 1346698 : if (!ItemIdIsUsed(itemid) || ItemIdIsRedirected(itemid))
3635 403070 : continue;
3636 :
3637 943628 : ItemPointerSet(&(tuple.t_self), blockno, offnum);
3638 :
3639 : /*
3640 : * Dead line pointers can have index pointers pointing to them. So
3641 : * they can't be treated as visible
3642 : */
3643 943628 : if (ItemIdIsDead(itemid))
3644 : {
3645 6 : all_visible = false;
3646 6 : *all_frozen = false;
3647 6 : break;
3648 : }
3649 :
3650 : Assert(ItemIdIsNormal(itemid));
3651 :
3652 943622 : tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid);
3653 943622 : tuple.t_len = ItemIdGetLength(itemid);
3654 943622 : tuple.t_tableOid = RelationGetRelid(vacrel->rel);
3655 :
3656 943622 : switch (HeapTupleSatisfiesVacuum(&tuple, vacrel->cutoffs.OldestXmin,
3657 : buf))
3658 : {
3659 943162 : case HEAPTUPLE_LIVE:
3660 : {
3661 : TransactionId xmin;
3662 :
3663 : /* Check comments in lazy_scan_prune. */
3664 943162 : if (!HeapTupleHeaderXminCommitted(tuple.t_data))
3665 : {
3666 0 : all_visible = false;
3667 0 : *all_frozen = false;
3668 0 : break;
3669 : }
3670 :
3671 : /*
3672 : * The inserter definitely committed. But is it old enough
3673 : * that everyone sees it as committed?
3674 : */
3675 943162 : xmin = HeapTupleHeaderGetXmin(tuple.t_data);
3676 943162 : if (!TransactionIdPrecedes(xmin,
3677 : vacrel->cutoffs.OldestXmin))
3678 : {
3679 192 : all_visible = false;
3680 192 : *all_frozen = false;
3681 192 : break;
3682 : }
3683 :
3684 : /* Track newest xmin on page. */
3685 942970 : if (TransactionIdFollows(xmin, *visibility_cutoff_xid) &&
3686 : TransactionIdIsNormal(xmin))
3687 21266 : *visibility_cutoff_xid = xmin;
3688 :
3689 : /* Check whether this tuple is already frozen or not */
3690 1249634 : if (all_visible && *all_frozen &&
3691 306664 : heap_tuple_needs_eventual_freeze(tuple.t_data))
3692 5746 : *all_frozen = false;
3693 : }
3694 942970 : break;
3695 :
3696 460 : case HEAPTUPLE_DEAD:
3697 : case HEAPTUPLE_RECENTLY_DEAD:
3698 : case HEAPTUPLE_INSERT_IN_PROGRESS:
3699 : case HEAPTUPLE_DELETE_IN_PROGRESS:
3700 : {
3701 460 : all_visible = false;
3702 460 : *all_frozen = false;
3703 460 : break;
3704 : }
3705 0 : default:
3706 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
3707 : break;
3708 : }
3709 : } /* scan along page */
3710 :
3711 : /* Clear the offset information once we have processed the given page. */
3712 28212 : vacrel->offnum = InvalidOffsetNumber;
3713 :
3714 28212 : return all_visible;
3715 : }
3716 :
3717 : /*
3718 : * Update index statistics in pg_class if the statistics are accurate.
3719 : */
3720 : static void
3721 177494 : update_relstats_all_indexes(LVRelState *vacrel)
3722 : {
3723 177494 : Relation *indrels = vacrel->indrels;
3724 177494 : int nindexes = vacrel->nindexes;
3725 177494 : IndexBulkDeleteResult **indstats = vacrel->indstats;
3726 :
3727 : Assert(vacrel->do_index_cleanup);
3728 :
3729 443488 : for (int idx = 0; idx < nindexes; idx++)
3730 : {
3731 265994 : Relation indrel = indrels[idx];
3732 265994 : IndexBulkDeleteResult *istat = indstats[idx];
3733 :
3734 265994 : if (istat == NULL || istat->estimated_count)
3735 263178 : continue;
3736 :
3737 : /* Update index statistics */
3738 2816 : vac_update_relstats(indrel,
3739 : istat->num_pages,
3740 : istat->num_index_tuples,
3741 : 0, 0,
3742 : false,
3743 : InvalidTransactionId,
3744 : InvalidMultiXactId,
3745 : NULL, NULL, false);
3746 : }
3747 177494 : }
3748 :
3749 : /*
3750 : * Error context callback for errors occurring during vacuum. The error
3751 : * context messages for index phases should match the messages set in parallel
3752 : * vacuum. If you change this function for those phases, change
3753 : * parallel_vacuum_error_callback() as well.
3754 : */
3755 : static void
3756 247832 : vacuum_error_callback(void *arg)
3757 : {
3758 247832 : LVRelState *errinfo = arg;
3759 :
3760 247832 : switch (errinfo->phase)
3761 : {
3762 0 : case VACUUM_ERRCB_PHASE_SCAN_HEAP:
3763 0 : if (BlockNumberIsValid(errinfo->blkno))
3764 : {
3765 0 : if (OffsetNumberIsValid(errinfo->offnum))
3766 0 : errcontext("while scanning block %u offset %u of relation \"%s.%s\"",
3767 0 : errinfo->blkno, errinfo->offnum, errinfo->relnamespace, errinfo->relname);
3768 : else
3769 0 : errcontext("while scanning block %u of relation \"%s.%s\"",
3770 : errinfo->blkno, errinfo->relnamespace, errinfo->relname);
3771 : }
3772 : else
3773 0 : errcontext("while scanning relation \"%s.%s\"",
3774 : errinfo->relnamespace, errinfo->relname);
3775 0 : break;
3776 :
3777 0 : case VACUUM_ERRCB_PHASE_VACUUM_HEAP:
3778 0 : if (BlockNumberIsValid(errinfo->blkno))
3779 : {
3780 0 : if (OffsetNumberIsValid(errinfo->offnum))
3781 0 : errcontext("while vacuuming block %u offset %u of relation \"%s.%s\"",
3782 0 : errinfo->blkno, errinfo->offnum, errinfo->relnamespace, errinfo->relname);
3783 : else
3784 0 : errcontext("while vacuuming block %u of relation \"%s.%s\"",
3785 : errinfo->blkno, errinfo->relnamespace, errinfo->relname);
3786 : }
3787 : else
3788 0 : errcontext("while vacuuming relation \"%s.%s\"",
3789 : errinfo->relnamespace, errinfo->relname);
3790 0 : break;
3791 :
3792 0 : case VACUUM_ERRCB_PHASE_VACUUM_INDEX:
3793 0 : errcontext("while vacuuming index \"%s\" of relation \"%s.%s\"",
3794 : errinfo->indname, errinfo->relnamespace, errinfo->relname);
3795 0 : break;
3796 :
3797 0 : case VACUUM_ERRCB_PHASE_INDEX_CLEANUP:
3798 0 : errcontext("while cleaning up index \"%s\" of relation \"%s.%s\"",
3799 : errinfo->indname, errinfo->relnamespace, errinfo->relname);
3800 0 : break;
3801 :
3802 6 : case VACUUM_ERRCB_PHASE_TRUNCATE:
3803 6 : if (BlockNumberIsValid(errinfo->blkno))
3804 6 : errcontext("while truncating relation \"%s.%s\" to %u blocks",
3805 : errinfo->relnamespace, errinfo->relname, errinfo->blkno);
3806 6 : break;
3807 :
3808 247826 : case VACUUM_ERRCB_PHASE_UNKNOWN:
3809 : default:
3810 247826 : return; /* do nothing; the errinfo may not be
3811 : * initialized */
3812 : }
3813 : }
3814 :
3815 : /*
3816 : * Updates the information required for vacuum error callback. This also saves
3817 : * the current information which can be later restored via restore_vacuum_error_info.
3818 : */
3819 : static void
3820 1179624 : update_vacuum_error_info(LVRelState *vacrel, LVSavedErrInfo *saved_vacrel,
3821 : int phase, BlockNumber blkno, OffsetNumber offnum)
3822 : {
3823 1179624 : if (saved_vacrel)
3824 : {
3825 297930 : saved_vacrel->offnum = vacrel->offnum;
3826 297930 : saved_vacrel->blkno = vacrel->blkno;
3827 297930 : saved_vacrel->phase = vacrel->phase;
3828 : }
3829 :
3830 1179624 : vacrel->blkno = blkno;
3831 1179624 : vacrel->offnum = offnum;
3832 1179624 : vacrel->phase = phase;
3833 1179624 : }
3834 :
3835 : /*
3836 : * Restores the vacuum information saved via a prior call to update_vacuum_error_info.
3837 : */
3838 : static void
3839 297930 : restore_vacuum_error_info(LVRelState *vacrel,
3840 : const LVSavedErrInfo *saved_vacrel)
3841 : {
3842 297930 : vacrel->blkno = saved_vacrel->blkno;
3843 297930 : vacrel->offnum = saved_vacrel->offnum;
3844 297930 : vacrel->phase = saved_vacrel->phase;
3845 297930 : }
|