Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * vacuumparallel.c
4 : * Support routines for parallel vacuum execution.
5 : *
6 : * This file contains routines that are intended to support setting up, using,
7 : * and tearing down a ParallelVacuumState.
8 : *
9 : * In a parallel vacuum, we perform both index bulk deletion and index cleanup
10 : * with parallel worker processes. Individual indexes are processed by one
11 : * vacuum process. ParallelVacuumState contains shared information as well as
12 : * the memory space for storing dead items allocated in the DSA area. We
13 : * launch parallel worker processes at the start of parallel index
14 : * bulk-deletion and index cleanup and once all indexes are processed, the
15 : * parallel worker processes exit. Each time we process indexes in parallel,
16 : * the parallel context is re-initialized so that the same DSM can be used for
17 : * multiple passes of index bulk-deletion and index cleanup.
18 : *
19 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
20 : * Portions Copyright (c) 1994, Regents of the University of California
21 : *
22 : * IDENTIFICATION
23 : * src/backend/commands/vacuumparallel.c
24 : *
25 : *-------------------------------------------------------------------------
26 : */
27 : #include "postgres.h"
28 :
29 : #include "access/amapi.h"
30 : #include "access/table.h"
31 : #include "access/xact.h"
32 : #include "commands/progress.h"
33 : #include "commands/vacuum.h"
34 : #include "executor/instrument.h"
35 : #include "optimizer/paths.h"
36 : #include "pgstat.h"
37 : #include "storage/bufmgr.h"
38 : #include "tcop/tcopprot.h"
39 : #include "utils/lsyscache.h"
40 : #include "utils/rel.h"
41 :
42 : /*
43 : * DSM keys for parallel vacuum. Unlike other parallel execution code, since
44 : * we don't need to worry about DSM keys conflicting with plan_node_id we can
45 : * use small integers.
46 : */
47 : #define PARALLEL_VACUUM_KEY_SHARED 1
48 : #define PARALLEL_VACUUM_KEY_QUERY_TEXT 2
49 : #define PARALLEL_VACUUM_KEY_BUFFER_USAGE 3
50 : #define PARALLEL_VACUUM_KEY_WAL_USAGE 4
51 : #define PARALLEL_VACUUM_KEY_INDEX_STATS 5
52 :
53 : /*
54 : * Shared information among parallel workers. So this is allocated in the DSM
55 : * segment.
56 : */
57 : typedef struct PVShared
58 : {
59 : /*
60 : * Target table relid and log level (for messages about parallel workers
61 : * launched during VACUUM VERBOSE). These fields are not modified during
62 : * the parallel vacuum.
63 : */
64 : Oid relid;
65 : int elevel;
66 :
67 : /*
68 : * Fields for both index vacuum and cleanup.
69 : *
70 : * reltuples is the total number of input heap tuples. We set either old
71 : * live tuples in the index vacuum case or the new live tuples in the
72 : * index cleanup case.
73 : *
74 : * estimated_count is true if reltuples is an estimated value. (Note that
75 : * reltuples could be -1 in this case, indicating we have no idea.)
76 : */
77 : double reltuples;
78 : bool estimated_count;
79 :
80 : /*
81 : * In single process vacuum we could consume more memory during index
82 : * vacuuming or cleanup apart from the memory for heap scanning. In
83 : * parallel vacuum, since individual vacuum workers can consume memory
84 : * equal to maintenance_work_mem, the new maintenance_work_mem for each
85 : * worker is set such that the parallel operation doesn't consume more
86 : * memory than single process vacuum.
87 : */
88 : int maintenance_work_mem_worker;
89 :
90 : /*
91 : * The number of buffers each worker's Buffer Access Strategy ring should
92 : * contain.
93 : */
94 : int ring_nbuffers;
95 :
96 : /*
97 : * Shared vacuum cost balance. During parallel vacuum,
98 : * VacuumSharedCostBalance points to this value and it accumulates the
99 : * balance of each parallel vacuum worker.
100 : */
101 : pg_atomic_uint32 cost_balance;
102 :
103 : /*
104 : * Number of active parallel workers. This is used for computing the
105 : * minimum threshold of the vacuum cost balance before a worker sleeps for
106 : * cost-based delay.
107 : */
108 : pg_atomic_uint32 active_nworkers;
109 :
110 : /* Counter for vacuuming and cleanup */
111 : pg_atomic_uint32 idx;
112 :
113 : /* DSA handle where the TidStore lives */
114 : dsa_handle dead_items_dsa_handle;
115 :
116 : /* DSA pointer to the shared TidStore */
117 : dsa_pointer dead_items_handle;
118 :
119 : /* Statistics of shared dead items */
120 : VacDeadItemsInfo dead_items_info;
121 : } PVShared;
122 :
123 : /* Status used during parallel index vacuum or cleanup */
124 : typedef enum PVIndVacStatus
125 : {
126 : PARALLEL_INDVAC_STATUS_INITIAL = 0,
127 : PARALLEL_INDVAC_STATUS_NEED_BULKDELETE,
128 : PARALLEL_INDVAC_STATUS_NEED_CLEANUP,
129 : PARALLEL_INDVAC_STATUS_COMPLETED,
130 : } PVIndVacStatus;
131 :
132 : /*
133 : * Struct for index vacuum statistics of an index that is used for parallel vacuum.
134 : * This includes the status of parallel index vacuum as well as index statistics.
135 : */
136 : typedef struct PVIndStats
137 : {
138 : /*
139 : * The following two fields are set by leader process before executing
140 : * parallel index vacuum or parallel index cleanup. These fields are not
141 : * fixed for the entire VACUUM operation. They are only fixed for an
142 : * individual parallel index vacuum and cleanup.
143 : *
144 : * parallel_workers_can_process is true if both leader and worker can
145 : * process the index, otherwise only leader can process it.
146 : */
147 : PVIndVacStatus status;
148 : bool parallel_workers_can_process;
149 :
150 : /*
151 : * Individual worker or leader stores the result of index vacuum or
152 : * cleanup.
153 : */
154 : bool istat_updated; /* are the stats updated? */
155 : IndexBulkDeleteResult istat;
156 : } PVIndStats;
157 :
158 : /*
159 : * Struct for maintaining a parallel vacuum state. typedef appears in vacuum.h.
160 : */
161 : struct ParallelVacuumState
162 : {
163 : /* NULL for worker processes */
164 : ParallelContext *pcxt;
165 :
166 : /* Parent Heap Relation */
167 : Relation heaprel;
168 :
169 : /* Target indexes */
170 : Relation *indrels;
171 : int nindexes;
172 :
173 : /* Shared information among parallel vacuum workers */
174 : PVShared *shared;
175 :
176 : /*
177 : * Shared index statistics among parallel vacuum workers. The array
178 : * element is allocated for every index, even those indexes where parallel
179 : * index vacuuming is unsafe or not worthwhile (e.g.,
180 : * will_parallel_vacuum[] is false). During parallel vacuum,
181 : * IndexBulkDeleteResult of each index is kept in DSM and is copied into
182 : * local memory at the end of parallel vacuum.
183 : */
184 : PVIndStats *indstats;
185 :
186 : /* Shared dead items space among parallel vacuum workers */
187 : TidStore *dead_items;
188 :
189 : /* Points to buffer usage area in DSM */
190 : BufferUsage *buffer_usage;
191 :
192 : /* Points to WAL usage area in DSM */
193 : WalUsage *wal_usage;
194 :
195 : /*
196 : * False if the index is totally unsuitable target for all parallel
197 : * processing. For example, the index could be <
198 : * min_parallel_index_scan_size cutoff.
199 : */
200 : bool *will_parallel_vacuum;
201 :
202 : /*
203 : * The number of indexes that support parallel index bulk-deletion and
204 : * parallel index cleanup respectively.
205 : */
206 : int nindexes_parallel_bulkdel;
207 : int nindexes_parallel_cleanup;
208 : int nindexes_parallel_condcleanup;
209 :
210 : /* Buffer access strategy used by leader process */
211 : BufferAccessStrategy bstrategy;
212 :
213 : /*
214 : * Error reporting state. The error callback is set only for workers
215 : * processes during parallel index vacuum.
216 : */
217 : char *relnamespace;
218 : char *relname;
219 : char *indname;
220 : PVIndVacStatus status;
221 : };
222 :
223 : static int parallel_vacuum_compute_workers(Relation *indrels, int nindexes, int nrequested,
224 : bool *will_parallel_vacuum);
225 : static void parallel_vacuum_process_all_indexes(ParallelVacuumState *pvs, int num_index_scans,
226 : bool vacuum);
227 : static void parallel_vacuum_process_safe_indexes(ParallelVacuumState *pvs);
228 : static void parallel_vacuum_process_unsafe_indexes(ParallelVacuumState *pvs);
229 : static void parallel_vacuum_process_one_index(ParallelVacuumState *pvs, Relation indrel,
230 : PVIndStats *indstats);
231 : static bool parallel_vacuum_index_is_parallel_safe(Relation indrel, int num_index_scans,
232 : bool vacuum);
233 : static void parallel_vacuum_error_callback(void *arg);
234 :
235 : /*
236 : * Try to enter parallel mode and create a parallel context. Then initialize
237 : * shared memory state.
238 : *
239 : * On success, return parallel vacuum state. Otherwise return NULL.
240 : */
241 : ParallelVacuumState *
242 7336 : parallel_vacuum_init(Relation rel, Relation *indrels, int nindexes,
243 : int nrequested_workers, int vac_work_mem,
244 : int elevel, BufferAccessStrategy bstrategy)
245 : {
246 : ParallelVacuumState *pvs;
247 : ParallelContext *pcxt;
248 : PVShared *shared;
249 : TidStore *dead_items;
250 : PVIndStats *indstats;
251 : BufferUsage *buffer_usage;
252 : WalUsage *wal_usage;
253 : bool *will_parallel_vacuum;
254 : Size est_indstats_len;
255 : Size est_shared_len;
256 7336 : int nindexes_mwm = 0;
257 7336 : int parallel_workers = 0;
258 : int querylen;
259 :
260 : /*
261 : * A parallel vacuum must be requested and there must be indexes on the
262 : * relation
263 : */
264 : Assert(nrequested_workers >= 0);
265 : Assert(nindexes > 0);
266 :
267 : /*
268 : * Compute the number of parallel vacuum workers to launch
269 : */
270 7336 : will_parallel_vacuum = (bool *) palloc0(sizeof(bool) * nindexes);
271 7336 : parallel_workers = parallel_vacuum_compute_workers(indrels, nindexes,
272 : nrequested_workers,
273 : will_parallel_vacuum);
274 7336 : if (parallel_workers <= 0)
275 : {
276 : /* Can't perform vacuum in parallel -- return NULL */
277 7318 : pfree(will_parallel_vacuum);
278 7318 : return NULL;
279 : }
280 :
281 18 : pvs = (ParallelVacuumState *) palloc0(sizeof(ParallelVacuumState));
282 18 : pvs->indrels = indrels;
283 18 : pvs->nindexes = nindexes;
284 18 : pvs->will_parallel_vacuum = will_parallel_vacuum;
285 18 : pvs->bstrategy = bstrategy;
286 18 : pvs->heaprel = rel;
287 :
288 18 : EnterParallelMode();
289 18 : pcxt = CreateParallelContext("postgres", "parallel_vacuum_main",
290 : parallel_workers);
291 : Assert(pcxt->nworkers > 0);
292 18 : pvs->pcxt = pcxt;
293 :
294 : /* Estimate size for index vacuum stats -- PARALLEL_VACUUM_KEY_INDEX_STATS */
295 18 : est_indstats_len = mul_size(sizeof(PVIndStats), nindexes);
296 18 : shm_toc_estimate_chunk(&pcxt->estimator, est_indstats_len);
297 18 : shm_toc_estimate_keys(&pcxt->estimator, 1);
298 :
299 : /* Estimate size for shared information -- PARALLEL_VACUUM_KEY_SHARED */
300 18 : est_shared_len = sizeof(PVShared);
301 18 : shm_toc_estimate_chunk(&pcxt->estimator, est_shared_len);
302 18 : shm_toc_estimate_keys(&pcxt->estimator, 1);
303 :
304 : /*
305 : * Estimate space for BufferUsage and WalUsage --
306 : * PARALLEL_VACUUM_KEY_BUFFER_USAGE and PARALLEL_VACUUM_KEY_WAL_USAGE.
307 : *
308 : * If there are no extensions loaded that care, we could skip this. We
309 : * have no way of knowing whether anyone's looking at pgBufferUsage or
310 : * pgWalUsage, so do it unconditionally.
311 : */
312 18 : shm_toc_estimate_chunk(&pcxt->estimator,
313 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
314 18 : shm_toc_estimate_keys(&pcxt->estimator, 1);
315 18 : shm_toc_estimate_chunk(&pcxt->estimator,
316 : mul_size(sizeof(WalUsage), pcxt->nworkers));
317 18 : shm_toc_estimate_keys(&pcxt->estimator, 1);
318 :
319 : /* Finally, estimate PARALLEL_VACUUM_KEY_QUERY_TEXT space */
320 18 : if (debug_query_string)
321 : {
322 18 : querylen = strlen(debug_query_string);
323 18 : shm_toc_estimate_chunk(&pcxt->estimator, querylen + 1);
324 18 : shm_toc_estimate_keys(&pcxt->estimator, 1);
325 : }
326 : else
327 0 : querylen = 0; /* keep compiler quiet */
328 :
329 18 : InitializeParallelDSM(pcxt);
330 :
331 : /* Prepare index vacuum stats */
332 18 : indstats = (PVIndStats *) shm_toc_allocate(pcxt->toc, est_indstats_len);
333 558 : MemSet(indstats, 0, est_indstats_len);
334 108 : for (int i = 0; i < nindexes; i++)
335 : {
336 90 : Relation indrel = indrels[i];
337 90 : uint8 vacoptions = indrel->rd_indam->amparallelvacuumoptions;
338 :
339 : /*
340 : * Cleanup option should be either disabled, always performing in
341 : * parallel or conditionally performing in parallel.
342 : */
343 : Assert(((vacoptions & VACUUM_OPTION_PARALLEL_CLEANUP) == 0) ||
344 : ((vacoptions & VACUUM_OPTION_PARALLEL_COND_CLEANUP) == 0));
345 : Assert(vacoptions <= VACUUM_OPTION_MAX_VALID_VALUE);
346 :
347 90 : if (!will_parallel_vacuum[i])
348 6 : continue;
349 :
350 84 : if (indrel->rd_indam->amusemaintenanceworkmem)
351 12 : nindexes_mwm++;
352 :
353 : /*
354 : * Remember the number of indexes that support parallel operation for
355 : * each phase.
356 : */
357 84 : if ((vacoptions & VACUUM_OPTION_PARALLEL_BULKDEL) != 0)
358 72 : pvs->nindexes_parallel_bulkdel++;
359 84 : if ((vacoptions & VACUUM_OPTION_PARALLEL_CLEANUP) != 0)
360 24 : pvs->nindexes_parallel_cleanup++;
361 84 : if ((vacoptions & VACUUM_OPTION_PARALLEL_COND_CLEANUP) != 0)
362 48 : pvs->nindexes_parallel_condcleanup++;
363 : }
364 18 : shm_toc_insert(pcxt->toc, PARALLEL_VACUUM_KEY_INDEX_STATS, indstats);
365 18 : pvs->indstats = indstats;
366 :
367 : /* Prepare shared information */
368 18 : shared = (PVShared *) shm_toc_allocate(pcxt->toc, est_shared_len);
369 180 : MemSet(shared, 0, est_shared_len);
370 18 : shared->relid = RelationGetRelid(rel);
371 18 : shared->elevel = elevel;
372 18 : shared->maintenance_work_mem_worker =
373 : (nindexes_mwm > 0) ?
374 18 : maintenance_work_mem / Min(parallel_workers, nindexes_mwm) :
375 : maintenance_work_mem;
376 18 : shared->dead_items_info.max_bytes = vac_work_mem * 1024L;
377 :
378 : /* Prepare DSA space for dead items */
379 18 : dead_items = TidStoreCreateShared(shared->dead_items_info.max_bytes,
380 : LWTRANCHE_PARALLEL_VACUUM_DSA);
381 18 : pvs->dead_items = dead_items;
382 18 : shared->dead_items_handle = TidStoreGetHandle(dead_items);
383 18 : shared->dead_items_dsa_handle = dsa_get_handle(TidStoreGetDSA(dead_items));
384 :
385 : /* Use the same buffer size for all workers */
386 18 : shared->ring_nbuffers = GetAccessStrategyBufferCount(bstrategy);
387 :
388 18 : pg_atomic_init_u32(&(shared->cost_balance), 0);
389 18 : pg_atomic_init_u32(&(shared->active_nworkers), 0);
390 18 : pg_atomic_init_u32(&(shared->idx), 0);
391 :
392 18 : shm_toc_insert(pcxt->toc, PARALLEL_VACUUM_KEY_SHARED, shared);
393 18 : pvs->shared = shared;
394 :
395 : /*
396 : * Allocate space for each worker's BufferUsage and WalUsage; no need to
397 : * initialize
398 : */
399 18 : buffer_usage = shm_toc_allocate(pcxt->toc,
400 18 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
401 18 : shm_toc_insert(pcxt->toc, PARALLEL_VACUUM_KEY_BUFFER_USAGE, buffer_usage);
402 18 : pvs->buffer_usage = buffer_usage;
403 18 : wal_usage = shm_toc_allocate(pcxt->toc,
404 18 : mul_size(sizeof(WalUsage), pcxt->nworkers));
405 18 : shm_toc_insert(pcxt->toc, PARALLEL_VACUUM_KEY_WAL_USAGE, wal_usage);
406 18 : pvs->wal_usage = wal_usage;
407 :
408 : /* Store query string for workers */
409 18 : if (debug_query_string)
410 : {
411 : char *sharedquery;
412 :
413 18 : sharedquery = (char *) shm_toc_allocate(pcxt->toc, querylen + 1);
414 18 : memcpy(sharedquery, debug_query_string, querylen + 1);
415 18 : sharedquery[querylen] = '\0';
416 18 : shm_toc_insert(pcxt->toc,
417 : PARALLEL_VACUUM_KEY_QUERY_TEXT, sharedquery);
418 : }
419 :
420 : /* Success -- return parallel vacuum state */
421 18 : return pvs;
422 : }
423 :
424 : /*
425 : * Destroy the parallel context, and end parallel mode.
426 : *
427 : * Since writes are not allowed during parallel mode, copy the
428 : * updated index statistics from DSM into local memory and then later use that
429 : * to update the index statistics. One might think that we can exit from
430 : * parallel mode, update the index statistics and then destroy parallel
431 : * context, but that won't be safe (see ExitParallelMode).
432 : */
433 : void
434 18 : parallel_vacuum_end(ParallelVacuumState *pvs, IndexBulkDeleteResult **istats)
435 : {
436 : Assert(!IsParallelWorker());
437 :
438 : /* Copy the updated statistics */
439 108 : for (int i = 0; i < pvs->nindexes; i++)
440 : {
441 90 : PVIndStats *indstats = &(pvs->indstats[i]);
442 :
443 90 : if (indstats->istat_updated)
444 : {
445 74 : istats[i] = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
446 74 : memcpy(istats[i], &indstats->istat, sizeof(IndexBulkDeleteResult));
447 : }
448 : else
449 16 : istats[i] = NULL;
450 : }
451 :
452 18 : TidStoreDestroy(pvs->dead_items);
453 :
454 18 : DestroyParallelContext(pvs->pcxt);
455 18 : ExitParallelMode();
456 :
457 18 : pfree(pvs->will_parallel_vacuum);
458 18 : pfree(pvs);
459 18 : }
460 :
461 : /*
462 : * Returns the dead items space and dead items information.
463 : */
464 : TidStore *
465 18 : parallel_vacuum_get_dead_items(ParallelVacuumState *pvs, VacDeadItemsInfo **dead_items_info_p)
466 : {
467 18 : *dead_items_info_p = &(pvs->shared->dead_items_info);
468 18 : return pvs->dead_items;
469 : }
470 :
471 : /* Forget all items in dead_items */
472 : void
473 10 : parallel_vacuum_reset_dead_items(ParallelVacuumState *pvs)
474 : {
475 10 : TidStore *dead_items = pvs->dead_items;
476 10 : VacDeadItemsInfo *dead_items_info = &(pvs->shared->dead_items_info);
477 :
478 : /*
479 : * Free the current tidstore and return allocated DSA segments to the
480 : * operating system. Then we recreate the tidstore with the same max_bytes
481 : * limitation we just used.
482 : */
483 10 : TidStoreDestroy(dead_items);
484 10 : pvs->dead_items = TidStoreCreateShared(dead_items_info->max_bytes,
485 : LWTRANCHE_PARALLEL_VACUUM_DSA);
486 :
487 : /* Update the DSA pointer for dead_items to the new one */
488 10 : pvs->shared->dead_items_dsa_handle = dsa_get_handle(TidStoreGetDSA(dead_items));
489 10 : pvs->shared->dead_items_handle = TidStoreGetHandle(dead_items);
490 :
491 : /* Reset the counter */
492 10 : dead_items_info->num_items = 0;
493 10 : }
494 :
495 : /*
496 : * Do parallel index bulk-deletion with parallel workers.
497 : */
498 : void
499 10 : parallel_vacuum_bulkdel_all_indexes(ParallelVacuumState *pvs, long num_table_tuples,
500 : int num_index_scans)
501 : {
502 : Assert(!IsParallelWorker());
503 :
504 : /*
505 : * We can only provide an approximate value of num_heap_tuples, at least
506 : * for now.
507 : */
508 10 : pvs->shared->reltuples = num_table_tuples;
509 10 : pvs->shared->estimated_count = true;
510 :
511 10 : parallel_vacuum_process_all_indexes(pvs, num_index_scans, true);
512 10 : }
513 :
514 : /*
515 : * Do parallel index cleanup with parallel workers.
516 : */
517 : void
518 18 : parallel_vacuum_cleanup_all_indexes(ParallelVacuumState *pvs, long num_table_tuples,
519 : int num_index_scans, bool estimated_count)
520 : {
521 : Assert(!IsParallelWorker());
522 :
523 : /*
524 : * We can provide a better estimate of total number of surviving tuples
525 : * (we assume indexes are more interested in that than in the number of
526 : * nominally live tuples).
527 : */
528 18 : pvs->shared->reltuples = num_table_tuples;
529 18 : pvs->shared->estimated_count = estimated_count;
530 :
531 18 : parallel_vacuum_process_all_indexes(pvs, num_index_scans, false);
532 18 : }
533 :
534 : /*
535 : * Compute the number of parallel worker processes to request. Both index
536 : * vacuum and index cleanup can be executed with parallel workers.
537 : * The index is eligible for parallel vacuum iff its size is greater than
538 : * min_parallel_index_scan_size as invoking workers for very small indexes
539 : * can hurt performance.
540 : *
541 : * nrequested is the number of parallel workers that user requested. If
542 : * nrequested is 0, we compute the parallel degree based on nindexes, that is
543 : * the number of indexes that support parallel vacuum. This function also
544 : * sets will_parallel_vacuum to remember indexes that participate in parallel
545 : * vacuum.
546 : */
547 : static int
548 7336 : parallel_vacuum_compute_workers(Relation *indrels, int nindexes, int nrequested,
549 : bool *will_parallel_vacuum)
550 : {
551 7336 : int nindexes_parallel = 0;
552 7336 : int nindexes_parallel_bulkdel = 0;
553 7336 : int nindexes_parallel_cleanup = 0;
554 : int parallel_workers;
555 :
556 : /*
557 : * We don't allow performing parallel operation in standalone backend or
558 : * when parallelism is disabled.
559 : */
560 7336 : if (!IsUnderPostmaster || max_parallel_maintenance_workers == 0)
561 3724 : return 0;
562 :
563 : /*
564 : * Compute the number of indexes that can participate in parallel vacuum.
565 : */
566 11800 : for (int i = 0; i < nindexes; i++)
567 : {
568 8188 : Relation indrel = indrels[i];
569 8188 : uint8 vacoptions = indrel->rd_indam->amparallelvacuumoptions;
570 :
571 : /* Skip index that is not a suitable target for parallel index vacuum */
572 8188 : if (vacoptions == VACUUM_OPTION_NO_PARALLEL ||
573 8188 : RelationGetNumberOfBlocks(indrel) < min_parallel_index_scan_size)
574 8092 : continue;
575 :
576 96 : will_parallel_vacuum[i] = true;
577 :
578 96 : if ((vacoptions & VACUUM_OPTION_PARALLEL_BULKDEL) != 0)
579 84 : nindexes_parallel_bulkdel++;
580 96 : if (((vacoptions & VACUUM_OPTION_PARALLEL_CLEANUP) != 0) ||
581 72 : ((vacoptions & VACUUM_OPTION_PARALLEL_COND_CLEANUP) != 0))
582 84 : nindexes_parallel_cleanup++;
583 : }
584 :
585 3612 : nindexes_parallel = Max(nindexes_parallel_bulkdel,
586 : nindexes_parallel_cleanup);
587 :
588 : /* The leader process takes one index */
589 3612 : nindexes_parallel--;
590 :
591 : /* No index supports parallel vacuum */
592 3612 : if (nindexes_parallel <= 0)
593 3594 : return 0;
594 :
595 : /* Compute the parallel degree */
596 18 : parallel_workers = (nrequested > 0) ?
597 18 : Min(nrequested, nindexes_parallel) : nindexes_parallel;
598 :
599 : /* Cap by max_parallel_maintenance_workers */
600 18 : parallel_workers = Min(parallel_workers, max_parallel_maintenance_workers);
601 :
602 18 : return parallel_workers;
603 : }
604 :
605 : /*
606 : * Perform index vacuum or index cleanup with parallel workers. This function
607 : * must be used by the parallel vacuum leader process.
608 : */
609 : static void
610 28 : parallel_vacuum_process_all_indexes(ParallelVacuumState *pvs, int num_index_scans,
611 : bool vacuum)
612 : {
613 : int nworkers;
614 : PVIndVacStatus new_status;
615 :
616 : Assert(!IsParallelWorker());
617 :
618 28 : if (vacuum)
619 : {
620 10 : new_status = PARALLEL_INDVAC_STATUS_NEED_BULKDELETE;
621 :
622 : /* Determine the number of parallel workers to launch */
623 10 : nworkers = pvs->nindexes_parallel_bulkdel;
624 : }
625 : else
626 : {
627 18 : new_status = PARALLEL_INDVAC_STATUS_NEED_CLEANUP;
628 :
629 : /* Determine the number of parallel workers to launch */
630 18 : nworkers = pvs->nindexes_parallel_cleanup;
631 :
632 : /* Add conditionally parallel-aware indexes if in the first time call */
633 18 : if (num_index_scans == 0)
634 8 : nworkers += pvs->nindexes_parallel_condcleanup;
635 : }
636 :
637 : /* The leader process will participate */
638 28 : nworkers--;
639 :
640 : /*
641 : * It is possible that parallel context is initialized with fewer workers
642 : * than the number of indexes that need a separate worker in the current
643 : * phase, so we need to consider it. See
644 : * parallel_vacuum_compute_workers().
645 : */
646 28 : nworkers = Min(nworkers, pvs->pcxt->nworkers);
647 :
648 : /*
649 : * Set index vacuum status and mark whether parallel vacuum worker can
650 : * process it.
651 : */
652 160 : for (int i = 0; i < pvs->nindexes; i++)
653 : {
654 132 : PVIndStats *indstats = &(pvs->indstats[i]);
655 :
656 : Assert(indstats->status == PARALLEL_INDVAC_STATUS_INITIAL);
657 132 : indstats->status = new_status;
658 132 : indstats->parallel_workers_can_process =
659 252 : (pvs->will_parallel_vacuum[i] &&
660 120 : parallel_vacuum_index_is_parallel_safe(pvs->indrels[i],
661 : num_index_scans,
662 : vacuum));
663 : }
664 :
665 : /* Reset the parallel index processing and progress counters */
666 28 : pg_atomic_write_u32(&(pvs->shared->idx), 0);
667 :
668 : /* Setup the shared cost-based vacuum delay and launch workers */
669 28 : if (nworkers > 0)
670 : {
671 : /* Reinitialize parallel context to relaunch parallel workers */
672 22 : if (num_index_scans > 0)
673 4 : ReinitializeParallelDSM(pvs->pcxt);
674 :
675 : /*
676 : * Set up shared cost balance and the number of active workers for
677 : * vacuum delay. We need to do this before launching workers as
678 : * otherwise, they might not see the updated values for these
679 : * parameters.
680 : */
681 22 : pg_atomic_write_u32(&(pvs->shared->cost_balance), VacuumCostBalance);
682 22 : pg_atomic_write_u32(&(pvs->shared->active_nworkers), 0);
683 :
684 : /*
685 : * The number of workers can vary between bulkdelete and cleanup
686 : * phase.
687 : */
688 22 : ReinitializeParallelWorkers(pvs->pcxt, nworkers);
689 :
690 22 : LaunchParallelWorkers(pvs->pcxt);
691 :
692 22 : if (pvs->pcxt->nworkers_launched > 0)
693 : {
694 : /*
695 : * Reset the local cost values for leader backend as we have
696 : * already accumulated the remaining balance of heap.
697 : */
698 22 : VacuumCostBalance = 0;
699 22 : VacuumCostBalanceLocal = 0;
700 :
701 : /* Enable shared cost balance for leader backend */
702 22 : VacuumSharedCostBalance = &(pvs->shared->cost_balance);
703 22 : VacuumActiveNWorkers = &(pvs->shared->active_nworkers);
704 : }
705 :
706 22 : if (vacuum)
707 10 : ereport(pvs->shared->elevel,
708 : (errmsg(ngettext("launched %d parallel vacuum worker for index vacuuming (planned: %d)",
709 : "launched %d parallel vacuum workers for index vacuuming (planned: %d)",
710 : pvs->pcxt->nworkers_launched),
711 : pvs->pcxt->nworkers_launched, nworkers)));
712 : else
713 12 : ereport(pvs->shared->elevel,
714 : (errmsg(ngettext("launched %d parallel vacuum worker for index cleanup (planned: %d)",
715 : "launched %d parallel vacuum workers for index cleanup (planned: %d)",
716 : pvs->pcxt->nworkers_launched),
717 : pvs->pcxt->nworkers_launched, nworkers)));
718 : }
719 :
720 : /* Vacuum the indexes that can be processed by only leader process */
721 28 : parallel_vacuum_process_unsafe_indexes(pvs);
722 :
723 : /*
724 : * Join as a parallel worker. The leader vacuums alone processes all
725 : * parallel-safe indexes in the case where no workers are launched.
726 : */
727 28 : parallel_vacuum_process_safe_indexes(pvs);
728 :
729 : /*
730 : * Next, accumulate buffer and WAL usage. (This must wait for the workers
731 : * to finish, or we might get incomplete data.)
732 : */
733 28 : if (nworkers > 0)
734 : {
735 : /* Wait for all vacuum workers to finish */
736 22 : WaitForParallelWorkersToFinish(pvs->pcxt);
737 :
738 56 : for (int i = 0; i < pvs->pcxt->nworkers_launched; i++)
739 34 : InstrAccumParallelQuery(&pvs->buffer_usage[i], &pvs->wal_usage[i]);
740 : }
741 :
742 : /*
743 : * Reset all index status back to initial (while checking that we have
744 : * vacuumed all indexes).
745 : */
746 160 : for (int i = 0; i < pvs->nindexes; i++)
747 : {
748 132 : PVIndStats *indstats = &(pvs->indstats[i]);
749 :
750 132 : if (indstats->status != PARALLEL_INDVAC_STATUS_COMPLETED)
751 0 : elog(ERROR, "parallel index vacuum on index \"%s\" is not completed",
752 : RelationGetRelationName(pvs->indrels[i]));
753 :
754 132 : indstats->status = PARALLEL_INDVAC_STATUS_INITIAL;
755 : }
756 :
757 : /*
758 : * Carry the shared balance value to heap scan and disable shared costing
759 : */
760 28 : if (VacuumSharedCostBalance)
761 : {
762 22 : VacuumCostBalance = pg_atomic_read_u32(VacuumSharedCostBalance);
763 22 : VacuumSharedCostBalance = NULL;
764 22 : VacuumActiveNWorkers = NULL;
765 : }
766 28 : }
767 :
768 : /*
769 : * Index vacuum/cleanup routine used by the leader process and parallel
770 : * vacuum worker processes to vacuum the indexes in parallel.
771 : */
772 : static void
773 62 : parallel_vacuum_process_safe_indexes(ParallelVacuumState *pvs)
774 : {
775 : /*
776 : * Increment the active worker count if we are able to launch any worker.
777 : */
778 62 : if (VacuumActiveNWorkers)
779 56 : pg_atomic_add_fetch_u32(VacuumActiveNWorkers, 1);
780 :
781 : /* Loop until all indexes are vacuumed */
782 : for (;;)
783 132 : {
784 : int idx;
785 : PVIndStats *indstats;
786 :
787 : /* Get an index number to process */
788 194 : idx = pg_atomic_fetch_add_u32(&(pvs->shared->idx), 1);
789 :
790 : /* Done for all indexes? */
791 194 : if (idx >= pvs->nindexes)
792 62 : break;
793 :
794 132 : indstats = &(pvs->indstats[idx]);
795 :
796 : /*
797 : * Skip vacuuming index that is unsafe for workers or has an
798 : * unsuitable target for parallel index vacuum (this is vacuumed in
799 : * parallel_vacuum_process_unsafe_indexes() by the leader).
800 : */
801 132 : if (!indstats->parallel_workers_can_process)
802 52 : continue;
803 :
804 : /* Do vacuum or cleanup of the index */
805 80 : parallel_vacuum_process_one_index(pvs, pvs->indrels[idx], indstats);
806 : }
807 :
808 : /*
809 : * We have completed the index vacuum so decrement the active worker
810 : * count.
811 : */
812 62 : if (VacuumActiveNWorkers)
813 56 : pg_atomic_sub_fetch_u32(VacuumActiveNWorkers, 1);
814 62 : }
815 :
816 : /*
817 : * Perform parallel vacuuming of indexes in leader process.
818 : *
819 : * Handles index vacuuming (or index cleanup) for indexes that are not
820 : * parallel safe. It's possible that this will vary for a given index, based
821 : * on details like whether we're performing index cleanup right now.
822 : *
823 : * Also performs vacuuming of smaller indexes that fell under the size cutoff
824 : * enforced by parallel_vacuum_compute_workers().
825 : */
826 : static void
827 28 : parallel_vacuum_process_unsafe_indexes(ParallelVacuumState *pvs)
828 : {
829 : Assert(!IsParallelWorker());
830 :
831 : /*
832 : * Increment the active worker count if we are able to launch any worker.
833 : */
834 28 : if (VacuumActiveNWorkers)
835 22 : pg_atomic_add_fetch_u32(VacuumActiveNWorkers, 1);
836 :
837 160 : for (int i = 0; i < pvs->nindexes; i++)
838 : {
839 132 : PVIndStats *indstats = &(pvs->indstats[i]);
840 :
841 : /* Skip, indexes that are safe for workers */
842 132 : if (indstats->parallel_workers_can_process)
843 80 : continue;
844 :
845 : /* Do vacuum or cleanup of the index */
846 52 : parallel_vacuum_process_one_index(pvs, pvs->indrels[i], indstats);
847 : }
848 :
849 : /*
850 : * We have completed the index vacuum so decrement the active worker
851 : * count.
852 : */
853 28 : if (VacuumActiveNWorkers)
854 22 : pg_atomic_sub_fetch_u32(VacuumActiveNWorkers, 1);
855 28 : }
856 :
857 : /*
858 : * Vacuum or cleanup index either by leader process or by one of the worker
859 : * process. After vacuuming the index this function copies the index
860 : * statistics returned from ambulkdelete and amvacuumcleanup to the DSM
861 : * segment.
862 : */
863 : static void
864 132 : parallel_vacuum_process_one_index(ParallelVacuumState *pvs, Relation indrel,
865 : PVIndStats *indstats)
866 : {
867 132 : IndexBulkDeleteResult *istat = NULL;
868 : IndexBulkDeleteResult *istat_res;
869 : IndexVacuumInfo ivinfo;
870 :
871 : /*
872 : * Update the pointer to the corresponding bulk-deletion result if someone
873 : * has already updated it
874 : */
875 132 : if (indstats->istat_updated)
876 42 : istat = &(indstats->istat);
877 :
878 132 : ivinfo.index = indrel;
879 132 : ivinfo.heaprel = pvs->heaprel;
880 132 : ivinfo.analyze_only = false;
881 132 : ivinfo.report_progress = false;
882 132 : ivinfo.message_level = DEBUG2;
883 132 : ivinfo.estimated_count = pvs->shared->estimated_count;
884 132 : ivinfo.num_heap_tuples = pvs->shared->reltuples;
885 132 : ivinfo.strategy = pvs->bstrategy;
886 :
887 : /* Update error traceback information */
888 132 : pvs->indname = pstrdup(RelationGetRelationName(indrel));
889 132 : pvs->status = indstats->status;
890 :
891 132 : switch (indstats->status)
892 : {
893 42 : case PARALLEL_INDVAC_STATUS_NEED_BULKDELETE:
894 42 : istat_res = vac_bulkdel_one_index(&ivinfo, istat, pvs->dead_items,
895 42 : &pvs->shared->dead_items_info);
896 42 : break;
897 90 : case PARALLEL_INDVAC_STATUS_NEED_CLEANUP:
898 90 : istat_res = vac_cleanup_one_index(&ivinfo, istat);
899 90 : break;
900 0 : default:
901 0 : elog(ERROR, "unexpected parallel vacuum index status %d for index \"%s\"",
902 : indstats->status,
903 : RelationGetRelationName(indrel));
904 : }
905 :
906 : /*
907 : * Copy the index bulk-deletion result returned from ambulkdelete and
908 : * amvacuumcleanup to the DSM segment if it's the first cycle because they
909 : * allocate locally and it's possible that an index will be vacuumed by a
910 : * different vacuum process the next cycle. Copying the result normally
911 : * happens only the first time an index is vacuumed. For any additional
912 : * vacuum pass, we directly point to the result on the DSM segment and
913 : * pass it to vacuum index APIs so that workers can update it directly.
914 : *
915 : * Since all vacuum workers write the bulk-deletion result at different
916 : * slots we can write them without locking.
917 : */
918 132 : if (!indstats->istat_updated && istat_res != NULL)
919 : {
920 74 : memcpy(&(indstats->istat), istat_res, sizeof(IndexBulkDeleteResult));
921 74 : indstats->istat_updated = true;
922 :
923 : /* Free the locally-allocated bulk-deletion result */
924 74 : pfree(istat_res);
925 : }
926 :
927 : /*
928 : * Update the status to completed. No need to lock here since each worker
929 : * touches different indexes.
930 : */
931 132 : indstats->status = PARALLEL_INDVAC_STATUS_COMPLETED;
932 :
933 : /* Reset error traceback information */
934 132 : pvs->status = PARALLEL_INDVAC_STATUS_COMPLETED;
935 132 : pfree(pvs->indname);
936 132 : pvs->indname = NULL;
937 :
938 : /*
939 : * Call the parallel variant of pgstat_progress_incr_param so workers can
940 : * report progress of index vacuum to the leader.
941 : */
942 132 : pgstat_progress_parallel_incr_param(PROGRESS_VACUUM_INDEXES_PROCESSED, 1);
943 132 : }
944 :
945 : /*
946 : * Returns false, if the given index can't participate in the next execution of
947 : * parallel index vacuum or parallel index cleanup.
948 : */
949 : static bool
950 120 : parallel_vacuum_index_is_parallel_safe(Relation indrel, int num_index_scans,
951 : bool vacuum)
952 : {
953 : uint8 vacoptions;
954 :
955 120 : vacoptions = indrel->rd_indam->amparallelvacuumoptions;
956 :
957 : /* In parallel vacuum case, check if it supports parallel bulk-deletion */
958 120 : if (vacuum)
959 36 : return ((vacoptions & VACUUM_OPTION_PARALLEL_BULKDEL) != 0);
960 :
961 : /* Not safe, if the index does not support parallel cleanup */
962 84 : if (((vacoptions & VACUUM_OPTION_PARALLEL_CLEANUP) == 0) &&
963 60 : ((vacoptions & VACUUM_OPTION_PARALLEL_COND_CLEANUP) == 0))
964 12 : return false;
965 :
966 : /*
967 : * Not safe, if the index supports parallel cleanup conditionally, but we
968 : * have already processed the index (for bulkdelete). We do this to avoid
969 : * the need to invoke workers when parallel index cleanup doesn't need to
970 : * scan the index. See the comments for option
971 : * VACUUM_OPTION_PARALLEL_COND_CLEANUP to know when indexes support
972 : * parallel cleanup conditionally.
973 : */
974 72 : if (num_index_scans > 0 &&
975 32 : ((vacoptions & VACUUM_OPTION_PARALLEL_COND_CLEANUP) != 0))
976 24 : return false;
977 :
978 48 : return true;
979 : }
980 :
981 : /*
982 : * Perform work within a launched parallel process.
983 : *
984 : * Since parallel vacuum workers perform only index vacuum or index cleanup,
985 : * we don't need to report progress information.
986 : */
987 : void
988 34 : parallel_vacuum_main(dsm_segment *seg, shm_toc *toc)
989 : {
990 : ParallelVacuumState pvs;
991 : Relation rel;
992 : Relation *indrels;
993 : PVIndStats *indstats;
994 : PVShared *shared;
995 : TidStore *dead_items;
996 : BufferUsage *buffer_usage;
997 : WalUsage *wal_usage;
998 : int nindexes;
999 : char *sharedquery;
1000 : ErrorContextCallback errcallback;
1001 :
1002 : /*
1003 : * A parallel vacuum worker must have only PROC_IN_VACUUM flag since we
1004 : * don't support parallel vacuum for autovacuum as of now.
1005 : */
1006 : Assert(MyProc->statusFlags == PROC_IN_VACUUM);
1007 :
1008 34 : elog(DEBUG1, "starting parallel vacuum worker");
1009 :
1010 34 : shared = (PVShared *) shm_toc_lookup(toc, PARALLEL_VACUUM_KEY_SHARED, false);
1011 :
1012 : /* Set debug_query_string for individual workers */
1013 34 : sharedquery = shm_toc_lookup(toc, PARALLEL_VACUUM_KEY_QUERY_TEXT, true);
1014 34 : debug_query_string = sharedquery;
1015 34 : pgstat_report_activity(STATE_RUNNING, debug_query_string);
1016 :
1017 : /*
1018 : * Open table. The lock mode is the same as the leader process. It's
1019 : * okay because the lock mode does not conflict among the parallel
1020 : * workers.
1021 : */
1022 34 : rel = table_open(shared->relid, ShareUpdateExclusiveLock);
1023 :
1024 : /*
1025 : * Open all indexes. indrels are sorted in order by OID, which should be
1026 : * matched to the leader's one.
1027 : */
1028 34 : vac_open_indexes(rel, RowExclusiveLock, &nindexes, &indrels);
1029 : Assert(nindexes > 0);
1030 :
1031 34 : if (shared->maintenance_work_mem_worker > 0)
1032 34 : maintenance_work_mem = shared->maintenance_work_mem_worker;
1033 :
1034 : /* Set index statistics */
1035 34 : indstats = (PVIndStats *) shm_toc_lookup(toc,
1036 : PARALLEL_VACUUM_KEY_INDEX_STATS,
1037 : false);
1038 :
1039 : /* Find dead_items in shared memory */
1040 34 : dead_items = TidStoreAttach(shared->dead_items_dsa_handle,
1041 : shared->dead_items_handle);
1042 :
1043 : /* Set cost-based vacuum delay */
1044 34 : VacuumUpdateCosts();
1045 34 : VacuumCostBalance = 0;
1046 34 : VacuumPageHit = 0;
1047 34 : VacuumPageMiss = 0;
1048 34 : VacuumPageDirty = 0;
1049 34 : VacuumCostBalanceLocal = 0;
1050 34 : VacuumSharedCostBalance = &(shared->cost_balance);
1051 34 : VacuumActiveNWorkers = &(shared->active_nworkers);
1052 :
1053 : /* Set parallel vacuum state */
1054 34 : pvs.indrels = indrels;
1055 34 : pvs.nindexes = nindexes;
1056 34 : pvs.indstats = indstats;
1057 34 : pvs.shared = shared;
1058 34 : pvs.dead_items = dead_items;
1059 34 : pvs.relnamespace = get_namespace_name(RelationGetNamespace(rel));
1060 34 : pvs.relname = pstrdup(RelationGetRelationName(rel));
1061 34 : pvs.heaprel = rel;
1062 :
1063 : /* These fields will be filled during index vacuum or cleanup */
1064 34 : pvs.indname = NULL;
1065 34 : pvs.status = PARALLEL_INDVAC_STATUS_INITIAL;
1066 :
1067 : /* Each parallel VACUUM worker gets its own access strategy. */
1068 68 : pvs.bstrategy = GetAccessStrategyWithSize(BAS_VACUUM,
1069 34 : shared->ring_nbuffers * (BLCKSZ / 1024));
1070 :
1071 : /* Setup error traceback support for ereport() */
1072 34 : errcallback.callback = parallel_vacuum_error_callback;
1073 34 : errcallback.arg = &pvs;
1074 34 : errcallback.previous = error_context_stack;
1075 34 : error_context_stack = &errcallback;
1076 :
1077 : /* Prepare to track buffer usage during parallel execution */
1078 34 : InstrStartParallelQuery();
1079 :
1080 : /* Process indexes to perform vacuum/cleanup */
1081 34 : parallel_vacuum_process_safe_indexes(&pvs);
1082 :
1083 : /* Report buffer/WAL usage during parallel execution */
1084 34 : buffer_usage = shm_toc_lookup(toc, PARALLEL_VACUUM_KEY_BUFFER_USAGE, false);
1085 34 : wal_usage = shm_toc_lookup(toc, PARALLEL_VACUUM_KEY_WAL_USAGE, false);
1086 34 : InstrEndParallelQuery(&buffer_usage[ParallelWorkerNumber],
1087 34 : &wal_usage[ParallelWorkerNumber]);
1088 :
1089 34 : TidStoreDetach(dead_items);
1090 :
1091 : /* Pop the error context stack */
1092 34 : error_context_stack = errcallback.previous;
1093 :
1094 34 : vac_close_indexes(nindexes, indrels, RowExclusiveLock);
1095 34 : table_close(rel, ShareUpdateExclusiveLock);
1096 34 : FreeAccessStrategy(pvs.bstrategy);
1097 34 : }
1098 :
1099 : /*
1100 : * Error context callback for errors occurring during parallel index vacuum.
1101 : * The error context messages should match the messages set in the lazy vacuum
1102 : * error context. If you change this function, change vacuum_error_callback()
1103 : * as well.
1104 : */
1105 : static void
1106 0 : parallel_vacuum_error_callback(void *arg)
1107 : {
1108 0 : ParallelVacuumState *errinfo = arg;
1109 :
1110 0 : switch (errinfo->status)
1111 : {
1112 0 : case PARALLEL_INDVAC_STATUS_NEED_BULKDELETE:
1113 0 : errcontext("while vacuuming index \"%s\" of relation \"%s.%s\"",
1114 : errinfo->indname,
1115 : errinfo->relnamespace,
1116 : errinfo->relname);
1117 0 : break;
1118 0 : case PARALLEL_INDVAC_STATUS_NEED_CLEANUP:
1119 0 : errcontext("while cleaning up index \"%s\" of relation \"%s.%s\"",
1120 : errinfo->indname,
1121 : errinfo->relnamespace,
1122 : errinfo->relname);
1123 0 : break;
1124 0 : case PARALLEL_INDVAC_STATUS_INITIAL:
1125 : case PARALLEL_INDVAC_STATUS_COMPLETED:
1126 : default:
1127 0 : return;
1128 : }
1129 : }
|
