PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
autovacuum.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * autovacuum.c
4  *
5  * PostgreSQL Integrated Autovacuum Daemon
6  *
7  * The autovacuum system is structured in two different kinds of processes: the
8  * autovacuum launcher and the autovacuum worker. The launcher is an
9  * always-running process, started by the postmaster when the autovacuum GUC
10  * parameter is set. The launcher schedules autovacuum workers to be started
11  * when appropriate. The workers are the processes which execute the actual
12  * vacuuming; they connect to a database as determined in the launcher, and
13  * once connected they examine the catalogs to select the tables to vacuum.
14  *
15  * The autovacuum launcher cannot start the worker processes by itself,
16  * because doing so would cause robustness issues (namely, failure to shut
17  * them down on exceptional conditions, and also, since the launcher is
18  * connected to shared memory and is thus subject to corruption there, it is
19  * not as robust as the postmaster). So it leaves that task to the postmaster.
20  *
21  * There is an autovacuum shared memory area, where the launcher stores
22  * information about the database it wants vacuumed. When it wants a new
23  * worker to start, it sets a flag in shared memory and sends a signal to the
24  * postmaster. Then postmaster knows nothing more than it must start a worker;
25  * so it forks a new child, which turns into a worker. This new process
26  * connects to shared memory, and there it can inspect the information that the
27  * launcher has set up.
28  *
29  * If the fork() call fails in the postmaster, it sets a flag in the shared
30  * memory area, and sends a signal to the launcher. The launcher, upon
31  * noticing the flag, can try starting the worker again by resending the
32  * signal. Note that the failure can only be transient (fork failure due to
33  * high load, memory pressure, too many processes, etc); more permanent
34  * problems, like failure to connect to a database, are detected later in the
35  * worker and dealt with just by having the worker exit normally. The launcher
36  * will launch a new worker again later, per schedule.
37  *
38  * When the worker is done vacuuming it sends SIGUSR2 to the launcher. The
39  * launcher then wakes up and is able to launch another worker, if the schedule
40  * is so tight that a new worker is needed immediately. At this time the
41  * launcher can also balance the settings for the various remaining workers'
42  * cost-based vacuum delay feature.
43  *
44  * Note that there can be more than one worker in a database concurrently.
45  * They will store the table they are currently vacuuming in shared memory, so
46  * that other workers avoid being blocked waiting for the vacuum lock for that
47  * table. They will also reload the pgstats data just before vacuuming each
48  * table, to avoid vacuuming a table that was just finished being vacuumed by
49  * another worker and thus is no longer noted in shared memory. However,
50  * there is a window (caused by pgstat delay) on which a worker may choose a
51  * table that was already vacuumed; this is a bug in the current design.
52  *
53  * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
54  * Portions Copyright (c) 1994, Regents of the University of California
55  *
56  *
57  * IDENTIFICATION
58  * src/backend/postmaster/autovacuum.c
59  *
60  *-------------------------------------------------------------------------
61  */
62 #include "postgres.h"
63 
64 #include <signal.h>
65 #include <sys/time.h>
66 #include <unistd.h>
67 
68 #include "access/heapam.h"
69 #include "access/htup_details.h"
70 #include "access/multixact.h"
71 #include "access/reloptions.h"
72 #include "access/transam.h"
73 #include "access/xact.h"
74 #include "catalog/dependency.h"
75 #include "catalog/namespace.h"
76 #include "catalog/pg_database.h"
77 #include "commands/dbcommands.h"
78 #include "commands/vacuum.h"
79 #include "lib/ilist.h"
80 #include "libpq/pqsignal.h"
81 #include "miscadmin.h"
82 #include "pgstat.h"
83 #include "postmaster/autovacuum.h"
85 #include "postmaster/postmaster.h"
86 #include "storage/bufmgr.h"
87 #include "storage/ipc.h"
88 #include "storage/latch.h"
89 #include "storage/lmgr.h"
90 #include "storage/pmsignal.h"
91 #include "storage/proc.h"
92 #include "storage/procsignal.h"
93 #include "storage/sinvaladt.h"
94 #include "storage/smgr.h"
95 #include "tcop/tcopprot.h"
96 #include "utils/dsa.h"
97 #include "utils/fmgroids.h"
98 #include "utils/fmgrprotos.h"
99 #include "utils/lsyscache.h"
100 #include "utils/memutils.h"
101 #include "utils/ps_status.h"
102 #include "utils/rel.h"
103 #include "utils/snapmgr.h"
104 #include "utils/syscache.h"
105 #include "utils/timeout.h"
106 #include "utils/timestamp.h"
107 #include "utils/tqual.h"
108 
109 
110 /*
111  * GUC parameters
112  */
123 
126 
128 
129 /* how long to keep pgstat data in the launcher, in milliseconds */
130 #define STATS_READ_DELAY 1000
131 
132 /* the minimum allowed time between two awakenings of the launcher */
133 #define MIN_AUTOVAC_SLEEPTIME 100.0 /* milliseconds */
134 #define MAX_AUTOVAC_SLEEPTIME 300 /* seconds */
135 
136 /* Flags to tell if we are in an autovacuum process */
137 static bool am_autovacuum_launcher = false;
138 static bool am_autovacuum_worker = false;
139 
140 /* Flags set by signal handlers */
141 static volatile sig_atomic_t got_SIGHUP = false;
142 static volatile sig_atomic_t got_SIGUSR2 = false;
143 static volatile sig_atomic_t got_SIGTERM = false;
144 
145 /* Comparison points for determining whether freeze_max_age is exceeded */
148 
149 /* Default freeze ages to use for autovacuum (varies by database) */
154 
155 /* Memory context for long-lived data */
157 
158 /* struct to keep track of databases in launcher */
159 typedef struct avl_dbase
160 {
161  Oid adl_datid; /* hash key -- must be first */
165 } avl_dbase;
166 
167 /* struct to keep track of databases in worker */
168 typedef struct avw_dbase
169 {
171  char *adw_name;
175 } avw_dbase;
176 
177 /* struct to keep track of tables to vacuum and/or analyze, in 1st pass */
178 typedef struct av_relation
179 {
180  Oid ar_toastrelid; /* hash key - must be first */
183  AutoVacOpts ar_reloptions; /* copy of AutoVacOpts from the main table's
184  * reloptions, or NULL if none */
185 } av_relation;
186 
187 /* struct to keep track of tables to vacuum and/or analyze, after rechecking */
188 typedef struct autovac_table
189 {
191  int at_vacoptions; /* bitmask of VacuumOption */
197  char *at_relname;
198  char *at_nspname;
199  char *at_datname;
200 } autovac_table;
201 
202 /*-------------
203  * This struct holds information about a single worker's whereabouts. We keep
204  * an array of these in shared memory, sized according to
205  * autovacuum_max_workers.
206  *
207  * wi_links entry into free list or running list
208  * wi_dboid OID of the database this worker is supposed to work on
209  * wi_tableoid OID of the table currently being vacuumed, if any
210  * wi_sharedrel flag indicating whether table is marked relisshared
211  * wi_proc pointer to PGPROC of the running worker, NULL if not started
212  * wi_launchtime Time at which this worker was launched
213  * wi_cost_* Vacuum cost-based delay parameters current in this worker
214  *
215  * All fields are protected by AutovacuumLock, except for wi_tableoid which is
216  * protected by AutovacuumScheduleLock (which is read-only for everyone except
217  * that worker itself).
218  *-------------
219  */
220 typedef struct WorkerInfoData
221 {
233 
234 typedef struct WorkerInfoData *WorkerInfo;
235 
236 /*
237  * Possible signals received by the launcher from remote processes. These are
238  * stored atomically in shared memory so that other processes can set them
239  * without locking.
240  */
241 typedef enum
242 {
243  AutoVacForkFailed, /* failed trying to start a worker */
244  AutoVacRebalance, /* rebalance the cost limits */
245  AutoVacNumSignals /* must be last */
247 
248 /*
249  * Autovacuum workitem array, stored in AutoVacuumShmem->av_workItems. This
250  * list is mostly protected by AutovacuumLock, except that if an item is
251  * marked 'active' other processes must not modify the work-identifying
252  * members.
253  */
254 typedef struct AutoVacuumWorkItem
255 {
257  bool avw_used; /* below data is valid */
258  bool avw_active; /* being processed */
263 
264 #define NUM_WORKITEMS 256
265 
266 /*-------------
267  * The main autovacuum shmem struct. On shared memory we store this main
268  * struct and the array of WorkerInfo structs. This struct keeps:
269  *
270  * av_signal set by other processes to indicate various conditions
271  * av_launcherpid the PID of the autovacuum launcher
272  * av_freeWorkers the WorkerInfo freelist
273  * av_runningWorkers the WorkerInfo non-free queue
274  * av_startingWorker pointer to WorkerInfo currently being started (cleared by
275  * the worker itself as soon as it's up and running)
276  * av_workItems work item array
277  *
278  * This struct is protected by AutovacuumLock, except for av_signal and parts
279  * of the worker list (see above).
280  *-------------
281  */
282 typedef struct
283 {
284  sig_atomic_t av_signal[AutoVacNumSignals];
288  WorkerInfo av_startingWorker;
291 
293 
294 /*
295  * the database list (of avl_dbase elements) in the launcher, and the context
296  * that contains it
297  */
300 
301 /* Pointer to my own WorkerInfo, valid on each worker */
302 static WorkerInfo MyWorkerInfo = NULL;
303 
304 /* PID of launcher, valid only in worker while shutting down */
306 
307 #ifdef EXEC_BACKEND
308 static pid_t avlauncher_forkexec(void);
309 static pid_t avworker_forkexec(void);
310 #endif
311 NON_EXEC_STATIC void AutoVacWorkerMain(int argc, char *argv[]) pg_attribute_noreturn();
312 NON_EXEC_STATIC void AutoVacLauncherMain(int argc, char *argv[]) pg_attribute_noreturn();
313 
314 static Oid do_start_worker(void);
315 static void launcher_determine_sleep(bool canlaunch, bool recursing,
316  struct timeval *nap);
317 static void launch_worker(TimestampTz now);
318 static List *get_database_list(void);
319 static void rebuild_database_list(Oid newdb);
320 static int db_comparator(const void *a, const void *b);
321 static void autovac_balance_cost(void);
322 
323 static void do_autovacuum(void);
324 static void FreeWorkerInfo(int code, Datum arg);
325 
326 static autovac_table *table_recheck_autovac(Oid relid, HTAB *table_toast_map,
327  TupleDesc pg_class_desc,
328  int effective_multixact_freeze_max_age);
329 static void relation_needs_vacanalyze(Oid relid, AutoVacOpts *relopts,
330  Form_pg_class classForm,
331  PgStat_StatTabEntry *tabentry,
332  int effective_multixact_freeze_max_age,
333  bool *dovacuum, bool *doanalyze, bool *wraparound);
334 
336  BufferAccessStrategy bstrategy);
338  TupleDesc pg_class_desc);
339 static PgStat_StatTabEntry *get_pgstat_tabentry_relid(Oid relid, bool isshared,
340  PgStat_StatDBEntry *shared,
341  PgStat_StatDBEntry *dbentry);
342 static void perform_work_item(AutoVacuumWorkItem *workitem);
343 static void autovac_report_activity(autovac_table *tab);
344 static void autovac_report_workitem(AutoVacuumWorkItem *workitem,
345  const char *nspname, const char *relname);
346 static void av_sighup_handler(SIGNAL_ARGS);
347 static void avl_sigusr2_handler(SIGNAL_ARGS);
348 static void avl_sigterm_handler(SIGNAL_ARGS);
349 static void autovac_refresh_stats(void);
350 
351 
352 
353 /********************************************************************
354  * AUTOVACUUM LAUNCHER CODE
355  ********************************************************************/
356 
357 #ifdef EXEC_BACKEND
358 /*
359  * forkexec routine for the autovacuum launcher process.
360  *
361  * Format up the arglist, then fork and exec.
362  */
363 static pid_t
364 avlauncher_forkexec(void)
365 {
366  char *av[10];
367  int ac = 0;
368 
369  av[ac++] = "postgres";
370  av[ac++] = "--forkavlauncher";
371  av[ac++] = NULL; /* filled in by postmaster_forkexec */
372  av[ac] = NULL;
373 
374  Assert(ac < lengthof(av));
375 
376  return postmaster_forkexec(ac, av);
377 }
378 
379 /*
380  * We need this set from the outside, before InitProcess is called
381  */
382 void
383 AutovacuumLauncherIAm(void)
384 {
385  am_autovacuum_launcher = true;
386 }
387 #endif
388 
389 /*
390  * Main entry point for autovacuum launcher process, to be called from the
391  * postmaster.
392  */
393 int
395 {
396  pid_t AutoVacPID;
397 
398 #ifdef EXEC_BACKEND
399  switch ((AutoVacPID = avlauncher_forkexec()))
400 #else
401  switch ((AutoVacPID = fork_process()))
402 #endif
403  {
404  case -1:
405  ereport(LOG,
406  (errmsg("could not fork autovacuum launcher process: %m")));
407  return 0;
408 
409 #ifndef EXEC_BACKEND
410  case 0:
411  /* in postmaster child ... */
413 
414  /* Close the postmaster's sockets */
415  ClosePostmasterPorts(false);
416 
417  AutoVacLauncherMain(0, NULL);
418  break;
419 #endif
420  default:
421  return (int) AutoVacPID;
422  }
423 
424  /* shouldn't get here */
425  return 0;
426 }
427 
428 /*
429  * Main loop for the autovacuum launcher process.
430  */
431 NON_EXEC_STATIC void
432 AutoVacLauncherMain(int argc, char *argv[])
433 {
434  sigjmp_buf local_sigjmp_buf;
435 
436  am_autovacuum_launcher = true;
437 
438  /* Identify myself via ps */
440 
441  ereport(DEBUG1,
442  (errmsg("autovacuum launcher started")));
443 
444  if (PostAuthDelay)
445  pg_usleep(PostAuthDelay * 1000000L);
446 
448 
449  /*
450  * Set up signal handlers. We operate on databases much like a regular
451  * backend, so we use the same signal handling. See equivalent code in
452  * tcop/postgres.c.
453  */
456  pqsignal(SIGTERM, avl_sigterm_handler);
457 
459  InitializeTimeouts(); /* establishes SIGALRM handler */
460 
466 
467  /* Early initialization */
468  BaseInit();
469 
470  /*
471  * Create a per-backend PGPROC struct in shared memory, except in the
472  * EXEC_BACKEND case where this was done in SubPostmasterMain. We must do
473  * this before we can use LWLocks (and in the EXEC_BACKEND case we already
474  * had to do some stuff with LWLocks).
475  */
476 #ifndef EXEC_BACKEND
477  InitProcess();
478 #endif
479 
480  InitPostgres(NULL, InvalidOid, NULL, InvalidOid, NULL);
481 
483 
484  /*
485  * Create a memory context that we will do all our work in. We do this so
486  * that we can reset the context during error recovery and thereby avoid
487  * possible memory leaks.
488  */
489  AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
490  "Autovacuum Launcher",
492  MemoryContextSwitchTo(AutovacMemCxt);
493 
494  /*
495  * If an exception is encountered, processing resumes here.
496  *
497  * This code is a stripped down version of PostgresMain error recovery.
498  */
499  if (sigsetjmp(local_sigjmp_buf, 1) != 0)
500  {
501  /* since not using PG_TRY, must reset error stack by hand */
502  error_context_stack = NULL;
503 
504  /* Prevents interrupts while cleaning up */
505  HOLD_INTERRUPTS();
506 
507  /* Forget any pending QueryCancel or timeout request */
508  disable_all_timeouts(false);
509  QueryCancelPending = false; /* second to avoid race condition */
510 
511  /* Report the error to the server log */
512  EmitErrorReport();
513 
514  /* Abort the current transaction in order to recover */
516 
517  /*
518  * Release any other resources, for the case where we were not in a
519  * transaction.
520  */
523  AbortBufferIO();
524  UnlockBuffers();
526  {
529  false, true);
530  /* we needn't bother with the other ResourceOwnerRelease phases */
531  }
532  AtEOXact_Buffers(false);
533  AtEOXact_SMgr();
534  AtEOXact_Files();
535  AtEOXact_HashTables(false);
536 
537  /*
538  * Now return to normal top-level context and clear ErrorContext for
539  * next time.
540  */
541  MemoryContextSwitchTo(AutovacMemCxt);
542  FlushErrorState();
543 
544  /* Flush any leaked data in the top-level context */
546 
547  /* don't leave dangling pointers to freed memory */
548  DatabaseListCxt = NULL;
549  dlist_init(&DatabaseList);
550 
551  /*
552  * Make sure pgstat also considers our stat data as gone. Note: we
553  * mustn't use autovac_refresh_stats here.
554  */
556 
557  /* Now we can allow interrupts again */
559 
560  /* if in shutdown mode, no need for anything further; just go away */
561  if (got_SIGTERM)
562  goto shutdown;
563 
564  /*
565  * Sleep at least 1 second after any error. We don't want to be
566  * filling the error logs as fast as we can.
567  */
568  pg_usleep(1000000L);
569  }
570 
571  /* We can now handle ereport(ERROR) */
572  PG_exception_stack = &local_sigjmp_buf;
573 
574  /* must unblock signals before calling rebuild_database_list */
576 
577  /*
578  * Force zero_damaged_pages OFF in the autovac process, even if it is set
579  * in postgresql.conf. We don't really want such a dangerous option being
580  * applied non-interactively.
581  */
582  SetConfigOption("zero_damaged_pages", "false", PGC_SUSET, PGC_S_OVERRIDE);
583 
584  /*
585  * Force settable timeouts off to avoid letting these settings prevent
586  * regular maintenance from being executed.
587  */
588  SetConfigOption("statement_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
589  SetConfigOption("lock_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
590  SetConfigOption("idle_in_transaction_session_timeout", "0",
592 
593  /*
594  * Force default_transaction_isolation to READ COMMITTED. We don't want
595  * to pay the overhead of serializable mode, nor add any risk of causing
596  * deadlocks or delaying other transactions.
597  */
598  SetConfigOption("default_transaction_isolation", "read committed",
600 
601  /*
602  * In emergency mode, just start a worker (unless shutdown was requested)
603  * and go away.
604  */
605  if (!AutoVacuumingActive())
606  {
607  if (!got_SIGTERM)
608  do_start_worker();
609  proc_exit(0); /* done */
610  }
611 
612  AutoVacuumShmem->av_launcherpid = MyProcPid;
613 
614  /*
615  * Create the initial database list. The invariant we want this list to
616  * keep is that it's ordered by decreasing next_time. As soon as an entry
617  * is updated to a higher time, it will be moved to the front (which is
618  * correct because the only operation is to add autovacuum_naptime to the
619  * entry, and time always increases).
620  */
622 
623  /* loop until shutdown request */
624  while (!got_SIGTERM)
625  {
626  struct timeval nap;
627  TimestampTz current_time = 0;
628  bool can_launch;
629  int rc;
630 
631  /*
632  * This loop is a bit different from the normal use of WaitLatch,
633  * because we'd like to sleep before the first launch of a child
634  * process. So it's WaitLatch, then ResetLatch, then check for
635  * wakening conditions.
636  */
637 
639  false, &nap);
640 
641  /*
642  * Wait until naptime expires or we get some type of signal (all the
643  * signal handlers will wake us by calling SetLatch).
644  */
645  rc = WaitLatch(MyLatch,
647  (nap.tv_sec * 1000L) + (nap.tv_usec / 1000L),
649 
651 
652  /* Process sinval catchup interrupts that happened while sleeping */
654 
655  /*
656  * Emergency bailout if postmaster has died. This is to avoid the
657  * necessity for manual cleanup of all postmaster children.
658  */
659  if (rc & WL_POSTMASTER_DEATH)
660  proc_exit(1);
661 
662  /* the normal shutdown case */
663  if (got_SIGTERM)
664  break;
665 
666  if (got_SIGHUP)
667  {
668  got_SIGHUP = false;
670 
671  /* shutdown requested in config file? */
672  if (!AutoVacuumingActive())
673  break;
674 
675  /* rebalance in case the default cost parameters changed */
676  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
678  LWLockRelease(AutovacuumLock);
679 
680  /* rebuild the list in case the naptime changed */
682  }
683 
684  /*
685  * a worker finished, or postmaster signalled failure to start a
686  * worker
687  */
688  if (got_SIGUSR2)
689  {
690  got_SIGUSR2 = false;
691 
692  /* rebalance cost limits, if needed */
693  if (AutoVacuumShmem->av_signal[AutoVacRebalance])
694  {
695  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
696  AutoVacuumShmem->av_signal[AutoVacRebalance] = false;
698  LWLockRelease(AutovacuumLock);
699  }
700 
701  if (AutoVacuumShmem->av_signal[AutoVacForkFailed])
702  {
703  /*
704  * If the postmaster failed to start a new worker, we sleep
705  * for a little while and resend the signal. The new worker's
706  * state is still in memory, so this is sufficient. After
707  * that, we restart the main loop.
708  *
709  * XXX should we put a limit to the number of times we retry?
710  * I don't think it makes much sense, because a future start
711  * of a worker will continue to fail in the same way.
712  */
713  AutoVacuumShmem->av_signal[AutoVacForkFailed] = false;
714  pg_usleep(1000000L); /* 1s */
716  continue;
717  }
718  }
719 
720  /*
721  * There are some conditions that we need to check before trying to
722  * start a worker. First, we need to make sure that there is a worker
723  * slot available. Second, we need to make sure that no other worker
724  * failed while starting up.
725  */
726 
727  current_time = GetCurrentTimestamp();
728  LWLockAcquire(AutovacuumLock, LW_SHARED);
729 
730  can_launch = !dlist_is_empty(&AutoVacuumShmem->av_freeWorkers);
731 
732  if (AutoVacuumShmem->av_startingWorker != NULL)
733  {
734  int waittime;
735  WorkerInfo worker = AutoVacuumShmem->av_startingWorker;
736 
737  /*
738  * We can't launch another worker when another one is still
739  * starting up (or failed while doing so), so just sleep for a bit
740  * more; that worker will wake us up again as soon as it's ready.
741  * We will only wait autovacuum_naptime seconds (up to a maximum
742  * of 60 seconds) for this to happen however. Note that failure
743  * to connect to a particular database is not a problem here,
744  * because the worker removes itself from the startingWorker
745  * pointer before trying to connect. Problems detected by the
746  * postmaster (like fork() failure) are also reported and handled
747  * differently. The only problems that may cause this code to
748  * fire are errors in the earlier sections of AutoVacWorkerMain,
749  * before the worker removes the WorkerInfo from the
750  * startingWorker pointer.
751  */
752  waittime = Min(autovacuum_naptime, 60) * 1000;
753  if (TimestampDifferenceExceeds(worker->wi_launchtime, current_time,
754  waittime))
755  {
756  LWLockRelease(AutovacuumLock);
757  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
758 
759  /*
760  * No other process can put a worker in starting mode, so if
761  * startingWorker is still INVALID after exchanging our lock,
762  * we assume it's the same one we saw above (so we don't
763  * recheck the launch time).
764  */
765  if (AutoVacuumShmem->av_startingWorker != NULL)
766  {
767  worker = AutoVacuumShmem->av_startingWorker;
768  worker->wi_dboid = InvalidOid;
769  worker->wi_tableoid = InvalidOid;
770  worker->wi_sharedrel = false;
771  worker->wi_proc = NULL;
772  worker->wi_launchtime = 0;
773  dlist_push_head(&AutoVacuumShmem->av_freeWorkers,
774  &worker->wi_links);
775  AutoVacuumShmem->av_startingWorker = NULL;
776  elog(WARNING, "worker took too long to start; canceled");
777  }
778  }
779  else
780  can_launch = false;
781  }
782  LWLockRelease(AutovacuumLock); /* either shared or exclusive */
783 
784  /* if we can't do anything, just go back to sleep */
785  if (!can_launch)
786  continue;
787 
788  /* We're OK to start a new worker */
789 
790  if (dlist_is_empty(&DatabaseList))
791  {
792  /*
793  * Special case when the list is empty: start a worker right away.
794  * This covers the initial case, when no database is in pgstats
795  * (thus the list is empty). Note that the constraints in
796  * launcher_determine_sleep keep us from starting workers too
797  * quickly (at most once every autovacuum_naptime when the list is
798  * empty).
799  */
800  launch_worker(current_time);
801  }
802  else
803  {
804  /*
805  * because rebuild_database_list constructs a list with most
806  * distant adl_next_worker first, we obtain our database from the
807  * tail of the list.
808  */
809  avl_dbase *avdb;
810 
811  avdb = dlist_tail_element(avl_dbase, adl_node, &DatabaseList);
812 
813  /*
814  * launch a worker if next_worker is right now or it is in the
815  * past
816  */
818  current_time, 0))
819  launch_worker(current_time);
820  }
821  }
822 
823  /* Normal exit from the autovac launcher is here */
824 shutdown:
825  ereport(DEBUG1,
826  (errmsg("autovacuum launcher shutting down")));
827  AutoVacuumShmem->av_launcherpid = 0;
828 
829  proc_exit(0); /* done */
830 }
831 
832 /*
833  * Determine the time to sleep, based on the database list.
834  *
835  * The "canlaunch" parameter indicates whether we can start a worker right now,
836  * for example due to the workers being all busy. If this is false, we will
837  * cause a long sleep, which will be interrupted when a worker exits.
838  */
839 static void
840 launcher_determine_sleep(bool canlaunch, bool recursing, struct timeval *nap)
841 {
842  /*
843  * We sleep until the next scheduled vacuum. We trust that when the
844  * database list was built, care was taken so that no entries have times
845  * in the past; if the first entry has too close a next_worker value, or a
846  * time in the past, we will sleep a small nominal time.
847  */
848  if (!canlaunch)
849  {
850  nap->tv_sec = autovacuum_naptime;
851  nap->tv_usec = 0;
852  }
853  else if (!dlist_is_empty(&DatabaseList))
854  {
855  TimestampTz current_time = GetCurrentTimestamp();
856  TimestampTz next_wakeup;
857  avl_dbase *avdb;
858  long secs;
859  int usecs;
860 
861  avdb = dlist_tail_element(avl_dbase, adl_node, &DatabaseList);
862 
863  next_wakeup = avdb->adl_next_worker;
864  TimestampDifference(current_time, next_wakeup, &secs, &usecs);
865 
866  nap->tv_sec = secs;
867  nap->tv_usec = usecs;
868  }
869  else
870  {
871  /* list is empty, sleep for whole autovacuum_naptime seconds */
872  nap->tv_sec = autovacuum_naptime;
873  nap->tv_usec = 0;
874  }
875 
876  /*
877  * If the result is exactly zero, it means a database had an entry with
878  * time in the past. Rebuild the list so that the databases are evenly
879  * distributed again, and recalculate the time to sleep. This can happen
880  * if there are more tables needing vacuum than workers, and they all take
881  * longer to vacuum than autovacuum_naptime.
882  *
883  * We only recurse once. rebuild_database_list should always return times
884  * in the future, but it seems best not to trust too much on that.
885  */
886  if (nap->tv_sec == 0 && nap->tv_usec == 0 && !recursing)
887  {
889  launcher_determine_sleep(canlaunch, true, nap);
890  return;
891  }
892 
893  /* The smallest time we'll allow the launcher to sleep. */
894  if (nap->tv_sec <= 0 && nap->tv_usec <= MIN_AUTOVAC_SLEEPTIME * 1000)
895  {
896  nap->tv_sec = 0;
897  nap->tv_usec = MIN_AUTOVAC_SLEEPTIME * 1000;
898  }
899 
900  /*
901  * If the sleep time is too large, clamp it to an arbitrary maximum (plus
902  * any fractional seconds, for simplicity). This avoids an essentially
903  * infinite sleep in strange cases like the system clock going backwards a
904  * few years.
905  */
906  if (nap->tv_sec > MAX_AUTOVAC_SLEEPTIME)
907  nap->tv_sec = MAX_AUTOVAC_SLEEPTIME;
908 }
909 
910 /*
911  * Build an updated DatabaseList. It must only contain databases that appear
912  * in pgstats, and must be sorted by next_worker from highest to lowest,
913  * distributed regularly across the next autovacuum_naptime interval.
914  *
915  * Receives the Oid of the database that made this list be generated (we call
916  * this the "new" database, because when the database was already present on
917  * the list, we expect that this function is not called at all). The
918  * preexisting list, if any, will be used to preserve the order of the
919  * databases in the autovacuum_naptime period. The new database is put at the
920  * end of the interval. The actual values are not saved, which should not be
921  * much of a problem.
922  */
923 static void
925 {
926  List *dblist;
927  ListCell *cell;
928  MemoryContext newcxt;
929  MemoryContext oldcxt;
930  MemoryContext tmpcxt;
931  HASHCTL hctl;
932  int score;
933  int nelems;
934  HTAB *dbhash;
935  dlist_iter iter;
936 
937  /* use fresh stats */
939 
940  newcxt = AllocSetContextCreate(AutovacMemCxt,
941  "AV dblist",
943  tmpcxt = AllocSetContextCreate(newcxt,
944  "tmp AV dblist",
946  oldcxt = MemoryContextSwitchTo(tmpcxt);
947 
948  /*
949  * Implementing this is not as simple as it sounds, because we need to put
950  * the new database at the end of the list; next the databases that were
951  * already on the list, and finally (at the tail of the list) all the
952  * other databases that are not on the existing list.
953  *
954  * To do this, we build an empty hash table of scored databases. We will
955  * start with the lowest score (zero) for the new database, then
956  * increasing scores for the databases in the existing list, in order, and
957  * lastly increasing scores for all databases gotten via
958  * get_database_list() that are not already on the hash.
959  *
960  * Then we will put all the hash elements into an array, sort the array by
961  * score, and finally put the array elements into the new doubly linked
962  * list.
963  */
964  hctl.keysize = sizeof(Oid);
965  hctl.entrysize = sizeof(avl_dbase);
966  hctl.hcxt = tmpcxt;
967  dbhash = hash_create("db hash", 20, &hctl, /* magic number here FIXME */
969 
970  /* start by inserting the new database */
971  score = 0;
972  if (OidIsValid(newdb))
973  {
974  avl_dbase *db;
975  PgStat_StatDBEntry *entry;
976 
977  /* only consider this database if it has a pgstat entry */
978  entry = pgstat_fetch_stat_dbentry(newdb);
979  if (entry != NULL)
980  {
981  /* we assume it isn't found because the hash was just created */
982  db = hash_search(dbhash, &newdb, HASH_ENTER, NULL);
983 
984  /* hash_search already filled in the key */
985  db->adl_score = score++;
986  /* next_worker is filled in later */
987  }
988  }
989 
990  /* Now insert the databases from the existing list */
991  dlist_foreach(iter, &DatabaseList)
992  {
993  avl_dbase *avdb = dlist_container(avl_dbase, adl_node, iter.cur);
994  avl_dbase *db;
995  bool found;
996  PgStat_StatDBEntry *entry;
997 
998  /*
999  * skip databases with no stat entries -- in particular, this gets rid
1000  * of dropped databases
1001  */
1002  entry = pgstat_fetch_stat_dbentry(avdb->adl_datid);
1003  if (entry == NULL)
1004  continue;
1005 
1006  db = hash_search(dbhash, &(avdb->adl_datid), HASH_ENTER, &found);
1007 
1008  if (!found)
1009  {
1010  /* hash_search already filled in the key */
1011  db->adl_score = score++;
1012  /* next_worker is filled in later */
1013  }
1014  }
1015 
1016  /* finally, insert all qualifying databases not previously inserted */
1017  dblist = get_database_list();
1018  foreach(cell, dblist)
1019  {
1020  avw_dbase *avdb = lfirst(cell);
1021  avl_dbase *db;
1022  bool found;
1023  PgStat_StatDBEntry *entry;
1024 
1025  /* only consider databases with a pgstat entry */
1026  entry = pgstat_fetch_stat_dbentry(avdb->adw_datid);
1027  if (entry == NULL)
1028  continue;
1029 
1030  db = hash_search(dbhash, &(avdb->adw_datid), HASH_ENTER, &found);
1031  /* only update the score if the database was not already on the hash */
1032  if (!found)
1033  {
1034  /* hash_search already filled in the key */
1035  db->adl_score = score++;
1036  /* next_worker is filled in later */
1037  }
1038  }
1039  nelems = score;
1040 
1041  /* from here on, the allocated memory belongs to the new list */
1042  MemoryContextSwitchTo(newcxt);
1043  dlist_init(&DatabaseList);
1044 
1045  if (nelems > 0)
1046  {
1047  TimestampTz current_time;
1048  int millis_increment;
1049  avl_dbase *dbary;
1050  avl_dbase *db;
1051  HASH_SEQ_STATUS seq;
1052  int i;
1053 
1054  /* put all the hash elements into an array */
1055  dbary = palloc(nelems * sizeof(avl_dbase));
1056 
1057  i = 0;
1058  hash_seq_init(&seq, dbhash);
1059  while ((db = hash_seq_search(&seq)) != NULL)
1060  memcpy(&(dbary[i++]), db, sizeof(avl_dbase));
1061 
1062  /* sort the array */
1063  qsort(dbary, nelems, sizeof(avl_dbase), db_comparator);
1064 
1065  /*
1066  * Determine the time interval between databases in the schedule. If
1067  * we see that the configured naptime would take us to sleep times
1068  * lower than our min sleep time (which launcher_determine_sleep is
1069  * coded not to allow), silently use a larger naptime (but don't touch
1070  * the GUC variable).
1071  */
1072  millis_increment = 1000.0 * autovacuum_naptime / nelems;
1073  if (millis_increment <= MIN_AUTOVAC_SLEEPTIME)
1074  millis_increment = MIN_AUTOVAC_SLEEPTIME * 1.1;
1075 
1076  current_time = GetCurrentTimestamp();
1077 
1078  /*
1079  * move the elements from the array into the dllist, setting the
1080  * next_worker while walking the array
1081  */
1082  for (i = 0; i < nelems; i++)
1083  {
1084  avl_dbase *db = &(dbary[i]);
1085 
1086  current_time = TimestampTzPlusMilliseconds(current_time,
1087  millis_increment);
1088  db->adl_next_worker = current_time;
1089 
1090  /* later elements should go closer to the head of the list */
1091  dlist_push_head(&DatabaseList, &db->adl_node);
1092  }
1093  }
1094 
1095  /* all done, clean up memory */
1096  if (DatabaseListCxt != NULL)
1097  MemoryContextDelete(DatabaseListCxt);
1098  MemoryContextDelete(tmpcxt);
1099  DatabaseListCxt = newcxt;
1100  MemoryContextSwitchTo(oldcxt);
1101 }
1102 
1103 /* qsort comparator for avl_dbase, using adl_score */
1104 static int
1105 db_comparator(const void *a, const void *b)
1106 {
1107  if (((const avl_dbase *) a)->adl_score == ((const avl_dbase *) b)->adl_score)
1108  return 0;
1109  else
1110  return (((const avl_dbase *) a)->adl_score < ((const avl_dbase *) b)->adl_score) ? 1 : -1;
1111 }
1112 
1113 /*
1114  * do_start_worker
1115  *
1116  * Bare-bones procedure for starting an autovacuum worker from the launcher.
1117  * It determines what database to work on, sets up shared memory stuff and
1118  * signals postmaster to start the worker. It fails gracefully if invoked when
1119  * autovacuum_workers are already active.
1120  *
1121  * Return value is the OID of the database that the worker is going to process,
1122  * or InvalidOid if no worker was actually started.
1123  */
1124 static Oid
1126 {
1127  List *dblist;
1128  ListCell *cell;
1129  TransactionId xidForceLimit;
1130  MultiXactId multiForceLimit;
1131  bool for_xid_wrap;
1132  bool for_multi_wrap;
1133  avw_dbase *avdb;
1134  TimestampTz current_time;
1135  bool skipit = false;
1136  Oid retval = InvalidOid;
1137  MemoryContext tmpcxt,
1138  oldcxt;
1139 
1140  /* return quickly when there are no free workers */
1141  LWLockAcquire(AutovacuumLock, LW_SHARED);
1142  if (dlist_is_empty(&AutoVacuumShmem->av_freeWorkers))
1143  {
1144  LWLockRelease(AutovacuumLock);
1145  return InvalidOid;
1146  }
1147  LWLockRelease(AutovacuumLock);
1148 
1149  /*
1150  * Create and switch to a temporary context to avoid leaking the memory
1151  * allocated for the database list.
1152  */
1154  "Start worker tmp cxt",
1156  oldcxt = MemoryContextSwitchTo(tmpcxt);
1157 
1158  /* use fresh stats */
1160 
1161  /* Get a list of databases */
1162  dblist = get_database_list();
1163 
1164  /*
1165  * Determine the oldest datfrozenxid/relfrozenxid that we will allow to
1166  * pass without forcing a vacuum. (This limit can be tightened for
1167  * particular tables, but not loosened.)
1168  */
1170  xidForceLimit = recentXid - autovacuum_freeze_max_age;
1171  /* ensure it's a "normal" XID, else TransactionIdPrecedes misbehaves */
1172  /* this can cause the limit to go backwards by 3, but that's OK */
1173  if (xidForceLimit < FirstNormalTransactionId)
1174  xidForceLimit -= FirstNormalTransactionId;
1175 
1176  /* Also determine the oldest datminmxid we will consider. */
1178  multiForceLimit = recentMulti - MultiXactMemberFreezeThreshold();
1179  if (multiForceLimit < FirstMultiXactId)
1180  multiForceLimit -= FirstMultiXactId;
1181 
1182  /*
1183  * Choose a database to connect to. We pick the database that was least
1184  * recently auto-vacuumed, or one that needs vacuuming to prevent Xid
1185  * wraparound-related data loss. If any db at risk of Xid wraparound is
1186  * found, we pick the one with oldest datfrozenxid, independently of
1187  * autovacuum times; similarly we pick the one with the oldest datminmxid
1188  * if any is in MultiXactId wraparound. Note that those in Xid wraparound
1189  * danger are given more priority than those in multi wraparound danger.
1190  *
1191  * Note that a database with no stats entry is not considered, except for
1192  * Xid wraparound purposes. The theory is that if no one has ever
1193  * connected to it since the stats were last initialized, it doesn't need
1194  * vacuuming.
1195  *
1196  * XXX This could be improved if we had more info about whether it needs
1197  * vacuuming before connecting to it. Perhaps look through the pgstats
1198  * data for the database's tables? One idea is to keep track of the
1199  * number of new and dead tuples per database in pgstats. However it
1200  * isn't clear how to construct a metric that measures that and not cause
1201  * starvation for less busy databases.
1202  */
1203  avdb = NULL;
1204  for_xid_wrap = false;
1205  for_multi_wrap = false;
1206  current_time = GetCurrentTimestamp();
1207  foreach(cell, dblist)
1208  {
1209  avw_dbase *tmp = lfirst(cell);
1210  dlist_iter iter;
1211 
1212  /* Check to see if this one is at risk of wraparound */
1213  if (TransactionIdPrecedes(tmp->adw_frozenxid, xidForceLimit))
1214  {
1215  if (avdb == NULL ||
1217  avdb->adw_frozenxid))
1218  avdb = tmp;
1219  for_xid_wrap = true;
1220  continue;
1221  }
1222  else if (for_xid_wrap)
1223  continue; /* ignore not-at-risk DBs */
1224  else if (MultiXactIdPrecedes(tmp->adw_minmulti, multiForceLimit))
1225  {
1226  if (avdb == NULL ||
1228  avdb = tmp;
1229  for_multi_wrap = true;
1230  continue;
1231  }
1232  else if (for_multi_wrap)
1233  continue; /* ignore not-at-risk DBs */
1234 
1235  /* Find pgstat entry if any */
1237 
1238  /*
1239  * Skip a database with no pgstat entry; it means it hasn't seen any
1240  * activity.
1241  */
1242  if (!tmp->adw_entry)
1243  continue;
1244 
1245  /*
1246  * Also, skip a database that appears on the database list as having
1247  * been processed recently (less than autovacuum_naptime seconds ago).
1248  * We do this so that we don't select a database which we just
1249  * selected, but that pgstat hasn't gotten around to updating the last
1250  * autovacuum time yet.
1251  */
1252  skipit = false;
1253 
1254  dlist_reverse_foreach(iter, &DatabaseList)
1255  {
1256  avl_dbase *dbp = dlist_container(avl_dbase, adl_node, iter.cur);
1257 
1258  if (dbp->adl_datid == tmp->adw_datid)
1259  {
1260  /*
1261  * Skip this database if its next_worker value falls between
1262  * the current time and the current time plus naptime.
1263  */
1265  current_time, 0) &&
1266  !TimestampDifferenceExceeds(current_time,
1267  dbp->adl_next_worker,
1268  autovacuum_naptime * 1000))
1269  skipit = true;
1270 
1271  break;
1272  }
1273  }
1274  if (skipit)
1275  continue;
1276 
1277  /*
1278  * Remember the db with oldest autovac time. (If we are here, both
1279  * tmp->entry and db->entry must be non-null.)
1280  */
1281  if (avdb == NULL ||
1283  avdb = tmp;
1284  }
1285 
1286  /* Found a database -- process it */
1287  if (avdb != NULL)
1288  {
1289  WorkerInfo worker;
1290  dlist_node *wptr;
1291 
1292  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
1293 
1294  /*
1295  * Get a worker entry from the freelist. We checked above, so there
1296  * really should be a free slot.
1297  */
1298  wptr = dlist_pop_head_node(&AutoVacuumShmem->av_freeWorkers);
1299 
1300  worker = dlist_container(WorkerInfoData, wi_links, wptr);
1301  worker->wi_dboid = avdb->adw_datid;
1302  worker->wi_proc = NULL;
1303  worker->wi_launchtime = GetCurrentTimestamp();
1304 
1305  AutoVacuumShmem->av_startingWorker = worker;
1306 
1307  LWLockRelease(AutovacuumLock);
1308 
1310 
1311  retval = avdb->adw_datid;
1312  }
1313  else if (skipit)
1314  {
1315  /*
1316  * If we skipped all databases on the list, rebuild it, because it
1317  * probably contains a dropped database.
1318  */
1320  }
1321 
1322  MemoryContextSwitchTo(oldcxt);
1323  MemoryContextDelete(tmpcxt);
1324 
1325  return retval;
1326 }
1327 
1328 /*
1329  * launch_worker
1330  *
1331  * Wrapper for starting a worker from the launcher. Besides actually starting
1332  * it, update the database list to reflect the next time that another one will
1333  * need to be started on the selected database. The actual database choice is
1334  * left to do_start_worker.
1335  *
1336  * This routine is also expected to insert an entry into the database list if
1337  * the selected database was previously absent from the list.
1338  */
1339 static void
1341 {
1342  Oid dbid;
1343  dlist_iter iter;
1344 
1345  dbid = do_start_worker();
1346  if (OidIsValid(dbid))
1347  {
1348  bool found = false;
1349 
1350  /*
1351  * Walk the database list and update the corresponding entry. If the
1352  * database is not on the list, we'll recreate the list.
1353  */
1354  dlist_foreach(iter, &DatabaseList)
1355  {
1356  avl_dbase *avdb = dlist_container(avl_dbase, adl_node, iter.cur);
1357 
1358  if (avdb->adl_datid == dbid)
1359  {
1360  found = true;
1361 
1362  /*
1363  * add autovacuum_naptime seconds to the current time, and use
1364  * that as the new "next_worker" field for this database.
1365  */
1366  avdb->adl_next_worker =
1368 
1369  dlist_move_head(&DatabaseList, iter.cur);
1370  break;
1371  }
1372  }
1373 
1374  /*
1375  * If the database was not present in the database list, we rebuild
1376  * the list. It's possible that the database does not get into the
1377  * list anyway, for example if it's a database that doesn't have a
1378  * pgstat entry, but this is not a problem because we don't want to
1379  * schedule workers regularly into those in any case.
1380  */
1381  if (!found)
1382  rebuild_database_list(dbid);
1383  }
1384 }
1385 
1386 /*
1387  * Called from postmaster to signal a failure to fork a process to become
1388  * worker. The postmaster should kill(SIGUSR2) the launcher shortly
1389  * after calling this function.
1390  */
1391 void
1393 {
1394  AutoVacuumShmem->av_signal[AutoVacForkFailed] = true;
1395 }
1396 
1397 /* SIGHUP: set flag to re-read config file at next convenient time */
1398 static void
1400 {
1401  int save_errno = errno;
1402 
1403  got_SIGHUP = true;
1404  SetLatch(MyLatch);
1405 
1406  errno = save_errno;
1407 }
1408 
1409 /* SIGUSR2: a worker is up and running, or just finished, or failed to fork */
1410 static void
1412 {
1413  int save_errno = errno;
1414 
1415  got_SIGUSR2 = true;
1416  SetLatch(MyLatch);
1417 
1418  errno = save_errno;
1419 }
1420 
1421 /* SIGTERM: time to die */
1422 static void
1424 {
1425  int save_errno = errno;
1426 
1427  got_SIGTERM = true;
1428  SetLatch(MyLatch);
1429 
1430  errno = save_errno;
1431 }
1432 
1433 
1434 /********************************************************************
1435  * AUTOVACUUM WORKER CODE
1436  ********************************************************************/
1437 
1438 #ifdef EXEC_BACKEND
1439 /*
1440  * forkexec routines for the autovacuum worker.
1441  *
1442  * Format up the arglist, then fork and exec.
1443  */
1444 static pid_t
1445 avworker_forkexec(void)
1446 {
1447  char *av[10];
1448  int ac = 0;
1449 
1450  av[ac++] = "postgres";
1451  av[ac++] = "--forkavworker";
1452  av[ac++] = NULL; /* filled in by postmaster_forkexec */
1453  av[ac] = NULL;
1454 
1455  Assert(ac < lengthof(av));
1456 
1457  return postmaster_forkexec(ac, av);
1458 }
1459 
1460 /*
1461  * We need this set from the outside, before InitProcess is called
1462  */
1463 void
1464 AutovacuumWorkerIAm(void)
1465 {
1466  am_autovacuum_worker = true;
1467 }
1468 #endif
1469 
1470 /*
1471  * Main entry point for autovacuum worker process.
1472  *
1473  * This code is heavily based on pgarch.c, q.v.
1474  */
1475 int
1477 {
1478  pid_t worker_pid;
1479 
1480 #ifdef EXEC_BACKEND
1481  switch ((worker_pid = avworker_forkexec()))
1482 #else
1483  switch ((worker_pid = fork_process()))
1484 #endif
1485  {
1486  case -1:
1487  ereport(LOG,
1488  (errmsg("could not fork autovacuum worker process: %m")));
1489  return 0;
1490 
1491 #ifndef EXEC_BACKEND
1492  case 0:
1493  /* in postmaster child ... */
1495 
1496  /* Close the postmaster's sockets */
1497  ClosePostmasterPorts(false);
1498 
1499  AutoVacWorkerMain(0, NULL);
1500  break;
1501 #endif
1502  default:
1503  return (int) worker_pid;
1504  }
1505 
1506  /* shouldn't get here */
1507  return 0;
1508 }
1509 
1510 /*
1511  * AutoVacWorkerMain
1512  */
1513 NON_EXEC_STATIC void
1514 AutoVacWorkerMain(int argc, char *argv[])
1515 {
1516  sigjmp_buf local_sigjmp_buf;
1517  Oid dbid;
1518 
1519  am_autovacuum_worker = true;
1520 
1521  /* Identify myself via ps */
1523 
1525 
1526  /*
1527  * Set up signal handlers. We operate on databases much like a regular
1528  * backend, so we use the same signal handling. See equivalent code in
1529  * tcop/postgres.c.
1530  */
1532 
1533  /*
1534  * SIGINT is used to signal canceling the current table's vacuum; SIGTERM
1535  * means abort and exit cleanly, and SIGQUIT means abandon ship.
1536  */
1538  pqsignal(SIGTERM, die);
1540  InitializeTimeouts(); /* establishes SIGALRM handler */
1541 
1547 
1548  /* Early initialization */
1549  BaseInit();
1550 
1551  /*
1552  * Create a per-backend PGPROC struct in shared memory, except in the
1553  * EXEC_BACKEND case where this was done in SubPostmasterMain. We must do
1554  * this before we can use LWLocks (and in the EXEC_BACKEND case we already
1555  * had to do some stuff with LWLocks).
1556  */
1557 #ifndef EXEC_BACKEND
1558  InitProcess();
1559 #endif
1560 
1561  /*
1562  * If an exception is encountered, processing resumes here.
1563  *
1564  * See notes in postgres.c about the design of this coding.
1565  */
1566  if (sigsetjmp(local_sigjmp_buf, 1) != 0)
1567  {
1568  /* Prevents interrupts while cleaning up */
1569  HOLD_INTERRUPTS();
1570 
1571  /* Report the error to the server log */
1572  EmitErrorReport();
1573 
1574  /*
1575  * We can now go away. Note that because we called InitProcess, a
1576  * callback was registered to do ProcKill, which will clean up
1577  * necessary state.
1578  */
1579  proc_exit(0);
1580  }
1581 
1582  /* We can now handle ereport(ERROR) */
1583  PG_exception_stack = &local_sigjmp_buf;
1584 
1586 
1587  /*
1588  * Force zero_damaged_pages OFF in the autovac process, even if it is set
1589  * in postgresql.conf. We don't really want such a dangerous option being
1590  * applied non-interactively.
1591  */
1592  SetConfigOption("zero_damaged_pages", "false", PGC_SUSET, PGC_S_OVERRIDE);
1593 
1594  /*
1595  * Force settable timeouts off to avoid letting these settings prevent
1596  * regular maintenance from being executed.
1597  */
1598  SetConfigOption("statement_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
1599  SetConfigOption("lock_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
1600  SetConfigOption("idle_in_transaction_session_timeout", "0",
1602 
1603  /*
1604  * Force default_transaction_isolation to READ COMMITTED. We don't want
1605  * to pay the overhead of serializable mode, nor add any risk of causing
1606  * deadlocks or delaying other transactions.
1607  */
1608  SetConfigOption("default_transaction_isolation", "read committed",
1610 
1611  /*
1612  * Force synchronous replication off to allow regular maintenance even if
1613  * we are waiting for standbys to connect. This is important to ensure we
1614  * aren't blocked from performing anti-wraparound tasks.
1615  */
1617  SetConfigOption("synchronous_commit", "local",
1619 
1620  /*
1621  * Get the info about the database we're going to work on.
1622  */
1623  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
1624 
1625  /*
1626  * beware of startingWorker being INVALID; this should normally not
1627  * happen, but if a worker fails after forking and before this, the
1628  * launcher might have decided to remove it from the queue and start
1629  * again.
1630  */
1631  if (AutoVacuumShmem->av_startingWorker != NULL)
1632  {
1633  MyWorkerInfo = AutoVacuumShmem->av_startingWorker;
1634  dbid = MyWorkerInfo->wi_dboid;
1635  MyWorkerInfo->wi_proc = MyProc;
1636 
1637  /* insert into the running list */
1638  dlist_push_head(&AutoVacuumShmem->av_runningWorkers,
1639  &MyWorkerInfo->wi_links);
1640 
1641  /*
1642  * remove from the "starting" pointer, so that the launcher can start
1643  * a new worker if required
1644  */
1645  AutoVacuumShmem->av_startingWorker = NULL;
1646  LWLockRelease(AutovacuumLock);
1647 
1649 
1650  /* wake up the launcher */
1651  if (AutoVacuumShmem->av_launcherpid != 0)
1652  kill(AutoVacuumShmem->av_launcherpid, SIGUSR2);
1653  }
1654  else
1655  {
1656  /* no worker entry for me, go away */
1657  elog(WARNING, "autovacuum worker started without a worker entry");
1658  dbid = InvalidOid;
1659  LWLockRelease(AutovacuumLock);
1660  }
1661 
1662  if (OidIsValid(dbid))
1663  {
1664  char dbname[NAMEDATALEN];
1665 
1666  /*
1667  * Report autovac startup to the stats collector. We deliberately do
1668  * this before InitPostgres, so that the last_autovac_time will get
1669  * updated even if the connection attempt fails. This is to prevent
1670  * autovac from getting "stuck" repeatedly selecting an unopenable
1671  * database, rather than making any progress on stuff it can connect
1672  * to.
1673  */
1674  pgstat_report_autovac(dbid);
1675 
1676  /*
1677  * Connect to the selected database
1678  *
1679  * Note: if we have selected a just-deleted database (due to using
1680  * stale stats info), we'll fail and exit here.
1681  */
1682  InitPostgres(NULL, dbid, NULL, InvalidOid, dbname);
1684  set_ps_display(dbname, false);
1685  ereport(DEBUG1,
1686  (errmsg("autovacuum: processing database \"%s\"", dbname)));
1687 
1688  if (PostAuthDelay)
1689  pg_usleep(PostAuthDelay * 1000000L);
1690 
1691  /* And do an appropriate amount of work */
1694  do_autovacuum();
1695  }
1696 
1697  /*
1698  * The launcher will be notified of my death in ProcKill, *if* we managed
1699  * to get a worker slot at all
1700  */
1701 
1702  /* All done, go away */
1703  proc_exit(0);
1704 }
1705 
1706 /*
1707  * Return a WorkerInfo to the free list
1708  */
1709 static void
1711 {
1712  if (MyWorkerInfo != NULL)
1713  {
1714  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
1715 
1716  /*
1717  * Wake the launcher up so that he can launch a new worker immediately
1718  * if required. We only save the launcher's PID in local memory here;
1719  * the actual signal will be sent when the PGPROC is recycled. Note
1720  * that we always do this, so that the launcher can rebalance the cost
1721  * limit setting of the remaining workers.
1722  *
1723  * We somewhat ignore the risk that the launcher changes its PID
1724  * between us reading it and the actual kill; we expect ProcKill to be
1725  * called shortly after us, and we assume that PIDs are not reused too
1726  * quickly after a process exits.
1727  */
1728  AutovacuumLauncherPid = AutoVacuumShmem->av_launcherpid;
1729 
1730  dlist_delete(&MyWorkerInfo->wi_links);
1731  MyWorkerInfo->wi_dboid = InvalidOid;
1732  MyWorkerInfo->wi_tableoid = InvalidOid;
1733  MyWorkerInfo->wi_sharedrel = false;
1734  MyWorkerInfo->wi_proc = NULL;
1735  MyWorkerInfo->wi_launchtime = 0;
1736  MyWorkerInfo->wi_dobalance = false;
1737  MyWorkerInfo->wi_cost_delay = 0;
1738  MyWorkerInfo->wi_cost_limit = 0;
1739  MyWorkerInfo->wi_cost_limit_base = 0;
1740  dlist_push_head(&AutoVacuumShmem->av_freeWorkers,
1741  &MyWorkerInfo->wi_links);
1742  /* not mine anymore */
1743  MyWorkerInfo = NULL;
1744 
1745  /*
1746  * now that we're inactive, cause a rebalancing of the surviving
1747  * workers
1748  */
1749  AutoVacuumShmem->av_signal[AutoVacRebalance] = true;
1750  LWLockRelease(AutovacuumLock);
1751  }
1752 }
1753 
1754 /*
1755  * Update the cost-based delay parameters, so that multiple workers consume
1756  * each a fraction of the total available I/O.
1757  */
1758 void
1760 {
1761  if (MyWorkerInfo)
1762  {
1763  VacuumCostDelay = MyWorkerInfo->wi_cost_delay;
1764  VacuumCostLimit = MyWorkerInfo->wi_cost_limit;
1765  }
1766 }
1767 
1768 /*
1769  * autovac_balance_cost
1770  * Recalculate the cost limit setting for each active worker.
1771  *
1772  * Caller must hold the AutovacuumLock in exclusive mode.
1773  */
1774 static void
1776 {
1777  /*
1778  * The idea here is that we ration out I/O equally. The amount of I/O
1779  * that a worker can consume is determined by cost_limit/cost_delay, so we
1780  * try to equalize those ratios rather than the raw limit settings.
1781  *
1782  * note: in cost_limit, zero also means use value from elsewhere, because
1783  * zero is not a valid value.
1784  */
1785  int vac_cost_limit = (autovacuum_vac_cost_limit > 0 ?
1787  int vac_cost_delay = (autovacuum_vac_cost_delay >= 0 ?
1789  double cost_total;
1790  double cost_avail;
1791  dlist_iter iter;
1792 
1793  /* not set? nothing to do */
1794  if (vac_cost_limit <= 0 || vac_cost_delay <= 0)
1795  return;
1796 
1797  /* calculate the total base cost limit of participating active workers */
1798  cost_total = 0.0;
1799  dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
1800  {
1801  WorkerInfo worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
1802 
1803  if (worker->wi_proc != NULL &&
1804  worker->wi_dobalance &&
1805  worker->wi_cost_limit_base > 0 && worker->wi_cost_delay > 0)
1806  cost_total +=
1807  (double) worker->wi_cost_limit_base / worker->wi_cost_delay;
1808  }
1809 
1810  /* there are no cost limits -- nothing to do */
1811  if (cost_total <= 0)
1812  return;
1813 
1814  /*
1815  * Adjust cost limit of each active worker to balance the total of cost
1816  * limit to autovacuum_vacuum_cost_limit.
1817  */
1818  cost_avail = (double) vac_cost_limit / vac_cost_delay;
1819  dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
1820  {
1821  WorkerInfo worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
1822 
1823  if (worker->wi_proc != NULL &&
1824  worker->wi_dobalance &&
1825  worker->wi_cost_limit_base > 0 && worker->wi_cost_delay > 0)
1826  {
1827  int limit = (int)
1828  (cost_avail * worker->wi_cost_limit_base / cost_total);
1829 
1830  /*
1831  * We put a lower bound of 1 on the cost_limit, to avoid division-
1832  * by-zero in the vacuum code. Also, in case of roundoff trouble
1833  * in these calculations, let's be sure we don't ever set
1834  * cost_limit to more than the base value.
1835  */
1836  worker->wi_cost_limit = Max(Min(limit,
1837  worker->wi_cost_limit_base),
1838  1);
1839  }
1840 
1841  if (worker->wi_proc != NULL)
1842  elog(DEBUG2, "autovac_balance_cost(pid=%u db=%u, rel=%u, dobalance=%s cost_limit=%d, cost_limit_base=%d, cost_delay=%d)",
1843  worker->wi_proc->pid, worker->wi_dboid, worker->wi_tableoid,
1844  worker->wi_dobalance ? "yes" : "no",
1845  worker->wi_cost_limit, worker->wi_cost_limit_base,
1846  worker->wi_cost_delay);
1847  }
1848 }
1849 
1850 /*
1851  * get_database_list
1852  * Return a list of all databases found in pg_database.
1853  *
1854  * The list and associated data is allocated in the caller's memory context,
1855  * which is in charge of ensuring that it's properly cleaned up afterwards.
1856  *
1857  * Note: this is the only function in which the autovacuum launcher uses a
1858  * transaction. Although we aren't attached to any particular database and
1859  * therefore can't access most catalogs, we do have enough infrastructure
1860  * to do a seqscan on pg_database.
1861  */
1862 static List *
1864 {
1865  List *dblist = NIL;
1866  Relation rel;
1867  HeapScanDesc scan;
1868  HeapTuple tup;
1869  MemoryContext resultcxt;
1870 
1871  /* This is the context that we will allocate our output data in */
1872  resultcxt = CurrentMemoryContext;
1873 
1874  /*
1875  * Start a transaction so we can access pg_database, and get a snapshot.
1876  * We don't have a use for the snapshot itself, but we're interested in
1877  * the secondary effect that it sets RecentGlobalXmin. (This is critical
1878  * for anything that reads heap pages, because HOT may decide to prune
1879  * them even if the process doesn't attempt to modify any tuples.)
1880  */
1882  (void) GetTransactionSnapshot();
1883 
1885  scan = heap_beginscan_catalog(rel, 0, NULL);
1886 
1887  while (HeapTupleIsValid(tup = heap_getnext(scan, ForwardScanDirection)))
1888  {
1889  Form_pg_database pgdatabase = (Form_pg_database) GETSTRUCT(tup);
1890  avw_dbase *avdb;
1891  MemoryContext oldcxt;
1892 
1893  /*
1894  * Allocate our results in the caller's context, not the
1895  * transaction's. We do this inside the loop, and restore the original
1896  * context at the end, so that leaky things like heap_getnext() are
1897  * not called in a potentially long-lived context.
1898  */
1899  oldcxt = MemoryContextSwitchTo(resultcxt);
1900 
1901  avdb = (avw_dbase *) palloc(sizeof(avw_dbase));
1902 
1903  avdb->adw_datid = HeapTupleGetOid(tup);
1904  avdb->adw_name = pstrdup(NameStr(pgdatabase->datname));
1905  avdb->adw_frozenxid = pgdatabase->datfrozenxid;
1906  avdb->adw_minmulti = pgdatabase->datminmxid;
1907  /* this gets set later: */
1908  avdb->adw_entry = NULL;
1909 
1910  dblist = lappend(dblist, avdb);
1911  MemoryContextSwitchTo(oldcxt);
1912  }
1913 
1914  heap_endscan(scan);
1916 
1918 
1919  return dblist;
1920 }
1921 
1922 /*
1923  * Process a database table-by-table
1924  *
1925  * Note that CHECK_FOR_INTERRUPTS is supposed to be used in certain spots in
1926  * order not to ignore shutdown commands for too long.
1927  */
1928 static void
1930 {
1931  Relation classRel;
1932  HeapTuple tuple;
1933  HeapScanDesc relScan;
1934  Form_pg_database dbForm;
1935  List *table_oids = NIL;
1936  List *orphan_oids = NIL;
1937  HASHCTL ctl;
1938  HTAB *table_toast_map;
1939  ListCell *volatile cell;
1940  PgStat_StatDBEntry *shared;
1941  PgStat_StatDBEntry *dbentry;
1942  BufferAccessStrategy bstrategy;
1943  ScanKeyData key;
1944  TupleDesc pg_class_desc;
1945  int effective_multixact_freeze_max_age;
1946  bool did_vacuum = false;
1947  bool found_concurrent_worker = false;
1948  int i;
1949 
1950  /*
1951  * StartTransactionCommand and CommitTransactionCommand will automatically
1952  * switch to other contexts. We need this one to keep the list of
1953  * relations to vacuum/analyze across transactions.
1954  */
1955  AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
1956  "AV worker",
1958  MemoryContextSwitchTo(AutovacMemCxt);
1959 
1960  /*
1961  * may be NULL if we couldn't find an entry (only happens if we are
1962  * forcing a vacuum for anti-wrap purposes).
1963  */
1965 
1966  /* Start a transaction so our commands have one to play into. */
1968 
1969  /*
1970  * Clean up any dead statistics collector entries for this DB. We always
1971  * want to do this exactly once per DB-processing cycle, even if we find
1972  * nothing worth vacuuming in the database.
1973  */
1975 
1976  /*
1977  * Compute the multixact age for which freezing is urgent. This is
1978  * normally autovacuum_multixact_freeze_max_age, but may be less if we are
1979  * short of multixact member space.
1980  */
1981  effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
1982 
1983  /*
1984  * Find the pg_database entry and select the default freeze ages. We use
1985  * zero in template and nonconnectable databases, else the system-wide
1986  * default.
1987  */
1989  if (!HeapTupleIsValid(tuple))
1990  elog(ERROR, "cache lookup failed for database %u", MyDatabaseId);
1991  dbForm = (Form_pg_database) GETSTRUCT(tuple);
1992 
1993  if (dbForm->datistemplate || !dbForm->datallowconn)
1994  {
1999  }
2000  else
2001  {
2006  }
2007 
2008  ReleaseSysCache(tuple);
2009 
2010  /* StartTransactionCommand changed elsewhere */
2011  MemoryContextSwitchTo(AutovacMemCxt);
2012 
2013  /* The database hash where pgstat keeps shared relations */
2015 
2017 
2018  /* create a copy so we can use it after closing pg_class */
2019  pg_class_desc = CreateTupleDescCopy(RelationGetDescr(classRel));
2020 
2021  /* create hash table for toast <-> main relid mapping */
2022  MemSet(&ctl, 0, sizeof(ctl));
2023  ctl.keysize = sizeof(Oid);
2024  ctl.entrysize = sizeof(av_relation);
2025 
2026  table_toast_map = hash_create("TOAST to main relid map",
2027  100,
2028  &ctl,
2029  HASH_ELEM | HASH_BLOBS);
2030 
2031  /*
2032  * Scan pg_class to determine which tables to vacuum.
2033  *
2034  * We do this in two passes: on the first one we collect the list of plain
2035  * relations and materialized views, and on the second one we collect
2036  * TOAST tables. The reason for doing the second pass is that during it we
2037  * want to use the main relation's pg_class.reloptions entry if the TOAST
2038  * table does not have any, and we cannot obtain it unless we know
2039  * beforehand what's the main table OID.
2040  *
2041  * We need to check TOAST tables separately because in cases with short,
2042  * wide tables there might be proportionally much more activity in the
2043  * TOAST table than in its parent.
2044  */
2045  relScan = heap_beginscan_catalog(classRel, 0, NULL);
2046 
2047  /*
2048  * On the first pass, we collect main tables to vacuum, and also the main
2049  * table relid to TOAST relid mapping.
2050  */
2051  while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
2052  {
2053  Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
2054  PgStat_StatTabEntry *tabentry;
2055  AutoVacOpts *relopts;
2056  Oid relid;
2057  bool dovacuum;
2058  bool doanalyze;
2059  bool wraparound;
2060 
2061  if (classForm->relkind != RELKIND_RELATION &&
2062  classForm->relkind != RELKIND_MATVIEW)
2063  continue;
2064 
2065  relid = HeapTupleGetOid(tuple);
2066 
2067  /*
2068  * Check if it is a temp table (presumably, of some other backend's).
2069  * We cannot safely process other backends' temp tables.
2070  */
2071  if (classForm->relpersistence == RELPERSISTENCE_TEMP)
2072  {
2073  int backendID;
2074 
2075  backendID = GetTempNamespaceBackendId(classForm->relnamespace);
2076 
2077  /* We just ignore it if the owning backend is still active */
2078  if (backendID != InvalidBackendId &&
2079  (backendID == MyBackendId ||
2080  BackendIdGetProc(backendID) == NULL))
2081  {
2082  /*
2083  * The table seems to be orphaned -- although it might be that
2084  * the owning backend has already deleted it and exited; our
2085  * pg_class scan snapshot is not necessarily up-to-date
2086  * anymore, so we could be looking at a committed-dead entry.
2087  * Remember it so we can try to delete it later.
2088  */
2089  orphan_oids = lappend_oid(orphan_oids, relid);
2090  }
2091  continue;
2092  }
2093 
2094  /* Fetch reloptions and the pgstat entry for this table */
2095  relopts = extract_autovac_opts(tuple, pg_class_desc);
2096  tabentry = get_pgstat_tabentry_relid(relid, classForm->relisshared,
2097  shared, dbentry);
2098 
2099  /* Check if it needs vacuum or analyze */
2100  relation_needs_vacanalyze(relid, relopts, classForm, tabentry,
2101  effective_multixact_freeze_max_age,
2102  &dovacuum, &doanalyze, &wraparound);
2103 
2104  /* Relations that need work are added to table_oids */
2105  if (dovacuum || doanalyze)
2106  table_oids = lappend_oid(table_oids, relid);
2107 
2108  /*
2109  * Remember TOAST associations for the second pass. Note: we must do
2110  * this whether or not the table is going to be vacuumed, because we
2111  * don't automatically vacuum toast tables along the parent table.
2112  */
2113  if (OidIsValid(classForm->reltoastrelid))
2114  {
2115  av_relation *hentry;
2116  bool found;
2117 
2118  hentry = hash_search(table_toast_map,
2119  &classForm->reltoastrelid,
2120  HASH_ENTER, &found);
2121 
2122  if (!found)
2123  {
2124  /* hash_search already filled in the key */
2125  hentry->ar_relid = relid;
2126  hentry->ar_hasrelopts = false;
2127  if (relopts != NULL)
2128  {
2129  hentry->ar_hasrelopts = true;
2130  memcpy(&hentry->ar_reloptions, relopts,
2131  sizeof(AutoVacOpts));
2132  }
2133  }
2134  }
2135  }
2136 
2137  heap_endscan(relScan);
2138 
2139  /* second pass: check TOAST tables */
2140  ScanKeyInit(&key,
2142  BTEqualStrategyNumber, F_CHAREQ,
2144 
2145  relScan = heap_beginscan_catalog(classRel, 1, &key);
2146  while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
2147  {
2148  Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
2149  PgStat_StatTabEntry *tabentry;
2150  Oid relid;
2151  AutoVacOpts *relopts = NULL;
2152  bool dovacuum;
2153  bool doanalyze;
2154  bool wraparound;
2155 
2156  /*
2157  * We cannot safely process other backends' temp tables, so skip 'em.
2158  */
2159  if (classForm->relpersistence == RELPERSISTENCE_TEMP)
2160  continue;
2161 
2162  relid = HeapTupleGetOid(tuple);
2163 
2164  /*
2165  * fetch reloptions -- if this toast table does not have them, try the
2166  * main rel
2167  */
2168  relopts = extract_autovac_opts(tuple, pg_class_desc);
2169  if (relopts == NULL)
2170  {
2171  av_relation *hentry;
2172  bool found;
2173 
2174  hentry = hash_search(table_toast_map, &relid, HASH_FIND, &found);
2175  if (found && hentry->ar_hasrelopts)
2176  relopts = &hentry->ar_reloptions;
2177  }
2178 
2179  /* Fetch the pgstat entry for this table */
2180  tabentry = get_pgstat_tabentry_relid(relid, classForm->relisshared,
2181  shared, dbentry);
2182 
2183  relation_needs_vacanalyze(relid, relopts, classForm, tabentry,
2184  effective_multixact_freeze_max_age,
2185  &dovacuum, &doanalyze, &wraparound);
2186 
2187  /* ignore analyze for toast tables */
2188  if (dovacuum)
2189  table_oids = lappend_oid(table_oids, relid);
2190  }
2191 
2192  heap_endscan(relScan);
2193  heap_close(classRel, AccessShareLock);
2194 
2195  /*
2196  * Recheck orphan temporary tables, and if they still seem orphaned, drop
2197  * them. We'll eat a transaction per dropped table, which might seem
2198  * excessive, but we should only need to do anything as a result of a
2199  * previous backend crash, so this should not happen often enough to
2200  * justify "optimizing". Using separate transactions ensures that we
2201  * don't bloat the lock table if there are many temp tables to be dropped,
2202  * and it ensures that we don't lose work if a deletion attempt fails.
2203  */
2204  foreach(cell, orphan_oids)
2205  {
2206  Oid relid = lfirst_oid(cell);
2207  Form_pg_class classForm;
2208  int backendID;
2209  ObjectAddress object;
2210 
2211  /*
2212  * Check for user-requested abort.
2213  */
2215 
2216  /*
2217  * Try to lock the table. If we can't get the lock immediately,
2218  * somebody else is using (or dropping) the table, so it's not our
2219  * concern anymore. Having the lock prevents race conditions below.
2220  */
2222  continue;
2223 
2224  /*
2225  * Re-fetch the pg_class tuple and re-check whether it still seems to
2226  * be an orphaned temp table. If it's not there or no longer the same
2227  * relation, ignore it.
2228  */
2229  tuple = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
2230  if (!HeapTupleIsValid(tuple))
2231  {
2232  /* be sure to drop useless lock so we don't bloat lock table */
2234  continue;
2235  }
2236  classForm = (Form_pg_class) GETSTRUCT(tuple);
2237 
2238  /*
2239  * Make all the same tests made in the loop above. In event of OID
2240  * counter wraparound, the pg_class entry we have now might be
2241  * completely unrelated to the one we saw before.
2242  */
2243  if (!((classForm->relkind == RELKIND_RELATION ||
2244  classForm->relkind == RELKIND_MATVIEW) &&
2245  classForm->relpersistence == RELPERSISTENCE_TEMP))
2246  {
2248  continue;
2249  }
2250  backendID = GetTempNamespaceBackendId(classForm->relnamespace);
2251  if (!(backendID != InvalidBackendId &&
2252  (backendID == MyBackendId ||
2253  BackendIdGetProc(backendID) == NULL)))
2254  {
2256  continue;
2257  }
2258 
2259  /* OK, let's delete it */
2260  ereport(LOG,
2261  (errmsg("autovacuum: dropping orphan temp table \"%s.%s.%s\"",
2263  get_namespace_name(classForm->relnamespace),
2264  NameStr(classForm->relname))));
2265 
2266  object.classId = RelationRelationId;
2267  object.objectId = relid;
2268  object.objectSubId = 0;
2269  performDeletion(&object, DROP_CASCADE,
2273 
2274  /*
2275  * To commit the deletion, end current transaction and start a new
2276  * one. Note this also releases the lock we took.
2277  */
2280 
2281  /* StartTransactionCommand changed current memory context */
2282  MemoryContextSwitchTo(AutovacMemCxt);
2283  }
2284 
2285  /*
2286  * Create a buffer access strategy object for VACUUM to use. We want to
2287  * use the same one across all the vacuum operations we perform, since the
2288  * point is for VACUUM not to blow out the shared cache.
2289  */
2290  bstrategy = GetAccessStrategy(BAS_VACUUM);
2291 
2292  /*
2293  * create a memory context to act as fake PortalContext, so that the
2294  * contexts created in the vacuum code are cleaned up for each table.
2295  */
2296  PortalContext = AllocSetContextCreate(AutovacMemCxt,
2297  "Autovacuum Portal",
2299 
2300  /*
2301  * Perform operations on collected tables.
2302  */
2303  foreach(cell, table_oids)
2304  {
2305  Oid relid = lfirst_oid(cell);
2306  autovac_table *tab;
2307  bool skipit;
2308  int stdVacuumCostDelay;
2309  int stdVacuumCostLimit;
2310  dlist_iter iter;
2311 
2313 
2314  /*
2315  * Check for config changes before processing each collected table.
2316  */
2317  if (got_SIGHUP)
2318  {
2319  got_SIGHUP = false;
2321 
2322  /*
2323  * You might be tempted to bail out if we see autovacuum is now
2324  * disabled. Must resist that temptation -- this might be a
2325  * for-wraparound emergency worker, in which case that would be
2326  * entirely inappropriate.
2327  */
2328  }
2329 
2330  /*
2331  * hold schedule lock from here until we're sure that this table still
2332  * needs vacuuming. We also need the AutovacuumLock to walk the
2333  * worker array, but we'll let go of that one quickly.
2334  */
2335  LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
2336  LWLockAcquire(AutovacuumLock, LW_SHARED);
2337 
2338  /*
2339  * Check whether the table is being vacuumed concurrently by another
2340  * worker.
2341  */
2342  skipit = false;
2343  dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
2344  {
2345  WorkerInfo worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
2346 
2347  /* ignore myself */
2348  if (worker == MyWorkerInfo)
2349  continue;
2350 
2351  /* ignore workers in other databases (unless table is shared) */
2352  if (!worker->wi_sharedrel && worker->wi_dboid != MyDatabaseId)
2353  continue;
2354 
2355  if (worker->wi_tableoid == relid)
2356  {
2357  skipit = true;
2358  found_concurrent_worker = true;
2359  break;
2360  }
2361  }
2362  LWLockRelease(AutovacuumLock);
2363  if (skipit)
2364  {
2365  LWLockRelease(AutovacuumScheduleLock);
2366  continue;
2367  }
2368 
2369  /*
2370  * Check whether pgstat data still says we need to vacuum this table.
2371  * It could have changed if something else processed the table while
2372  * we weren't looking.
2373  *
2374  * Note: we have a special case in pgstat code to ensure that the
2375  * stats we read are as up-to-date as possible, to avoid the problem
2376  * that somebody just finished vacuuming this table. The window to
2377  * the race condition is not closed but it is very small.
2378  */
2379  MemoryContextSwitchTo(AutovacMemCxt);
2380  tab = table_recheck_autovac(relid, table_toast_map, pg_class_desc,
2381  effective_multixact_freeze_max_age);
2382  if (tab == NULL)
2383  {
2384  /* someone else vacuumed the table, or it went away */
2385  LWLockRelease(AutovacuumScheduleLock);
2386  continue;
2387  }
2388 
2389  /*
2390  * Ok, good to go. Store the table in shared memory before releasing
2391  * the lock so that other workers don't vacuum it concurrently.
2392  */
2393  MyWorkerInfo->wi_tableoid = relid;
2394  MyWorkerInfo->wi_sharedrel = tab->at_sharedrel;
2395  LWLockRelease(AutovacuumScheduleLock);
2396 
2397  /*
2398  * Remember the prevailing values of the vacuum cost GUCs. We have to
2399  * restore these at the bottom of the loop, else we'll compute wrong
2400  * values in the next iteration of autovac_balance_cost().
2401  */
2402  stdVacuumCostDelay = VacuumCostDelay;
2403  stdVacuumCostLimit = VacuumCostLimit;
2404 
2405  /* Must hold AutovacuumLock while mucking with cost balance info */
2406  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
2407 
2408  /* advertise my cost delay parameters for the balancing algorithm */
2409  MyWorkerInfo->wi_dobalance = tab->at_dobalance;
2410  MyWorkerInfo->wi_cost_delay = tab->at_vacuum_cost_delay;
2411  MyWorkerInfo->wi_cost_limit = tab->at_vacuum_cost_limit;
2412  MyWorkerInfo->wi_cost_limit_base = tab->at_vacuum_cost_limit;
2413 
2414  /* do a balance */
2416 
2417  /* set the active cost parameters from the result of that */
2419 
2420  /* done */
2421  LWLockRelease(AutovacuumLock);
2422 
2423  /* clean up memory before each iteration */
2425 
2426  /*
2427  * Save the relation name for a possible error message, to avoid a
2428  * catalog lookup in case of an error. If any of these return NULL,
2429  * then the relation has been dropped since last we checked; skip it.
2430  * Note: they must live in a long-lived memory context because we call
2431  * vacuum and analyze in different transactions.
2432  */
2433 
2434  tab->at_relname = get_rel_name(tab->at_relid);
2437  if (!tab->at_relname || !tab->at_nspname || !tab->at_datname)
2438  goto deleted;
2439 
2440  /*
2441  * We will abort vacuuming the current table if something errors out,
2442  * and continue with the next one in schedule; in particular, this
2443  * happens if we are interrupted with SIGINT.
2444  */
2445  PG_TRY();
2446  {
2447  /* Use PortalContext for any per-table allocations */
2449 
2450  /* have at it */
2451  autovacuum_do_vac_analyze(tab, bstrategy);
2452 
2453  /*
2454  * Clear a possible query-cancel signal, to avoid a late reaction
2455  * to an automatically-sent signal because of vacuuming the
2456  * current table (we're done with it, so it would make no sense to
2457  * cancel at this point.)
2458  */
2459  QueryCancelPending = false;
2460  }
2461  PG_CATCH();
2462  {
2463  /*
2464  * Abort the transaction, start a new one, and proceed with the
2465  * next table in our list.
2466  */
2467  HOLD_INTERRUPTS();
2468  if (tab->at_vacoptions & VACOPT_VACUUM)
2469  errcontext("automatic vacuum of table \"%s.%s.%s\"",
2470  tab->at_datname, tab->at_nspname, tab->at_relname);
2471  else
2472  errcontext("automatic analyze of table \"%s.%s.%s\"",
2473  tab->at_datname, tab->at_nspname, tab->at_relname);
2474  EmitErrorReport();
2475 
2476  /* this resets the PGXACT flags too */
2478  FlushErrorState();
2480 
2481  /* restart our transaction for the following operations */
2484  }
2485  PG_END_TRY();
2486 
2487  /* Make sure we're back in AutovacMemCxt */
2488  MemoryContextSwitchTo(AutovacMemCxt);
2489 
2490  did_vacuum = true;
2491 
2492  /* the PGXACT flags are reset at the next end of transaction */
2493 
2494  /* be tidy */
2495 deleted:
2496  if (tab->at_datname != NULL)
2497  pfree(tab->at_datname);
2498  if (tab->at_nspname != NULL)
2499  pfree(tab->at_nspname);
2500  if (tab->at_relname != NULL)
2501  pfree(tab->at_relname);
2502  pfree(tab);
2503 
2504  /*
2505  * Remove my info from shared memory. We could, but intentionally
2506  * don't, clear wi_cost_limit and friends --- this is on the
2507  * assumption that we probably have more to do with similar cost
2508  * settings, so we don't want to give up our share of I/O for a very
2509  * short interval and thereby thrash the global balance.
2510  */
2511  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
2512  MyWorkerInfo->wi_tableoid = InvalidOid;
2513  MyWorkerInfo->wi_sharedrel = false;
2514  LWLockRelease(AutovacuumLock);
2515 
2516  /* restore vacuum cost GUCs for the next iteration */
2517  VacuumCostDelay = stdVacuumCostDelay;
2518  VacuumCostLimit = stdVacuumCostLimit;
2519  }
2520 
2521  /*
2522  * Perform additional work items, as requested by backends.
2523  */
2524  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
2525  for (i = 0; i < NUM_WORKITEMS; i++)
2526  {
2527  AutoVacuumWorkItem *workitem = &AutoVacuumShmem->av_workItems[i];
2528 
2529  if (!workitem->avw_used)
2530  continue;
2531  if (workitem->avw_active)
2532  continue;
2533 
2534  /* claim this one, and release lock while performing it */
2535  workitem->avw_active = true;
2536  LWLockRelease(AutovacuumLock);
2537 
2538  perform_work_item(workitem);
2539 
2540  /*
2541  * Check for config changes before acquiring lock for further jobs.
2542  */
2544  if (got_SIGHUP)
2545  {
2546  got_SIGHUP = false;
2548  }
2549 
2550  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
2551 
2552  /* and mark it done */
2553  workitem->avw_active = false;
2554  workitem->avw_used = false;
2555  }
2556  LWLockRelease(AutovacuumLock);
2557 
2558  /*
2559  * We leak table_toast_map here (among other things), but since we're
2560  * going away soon, it's not a problem.
2561  */
2562 
2563  /*
2564  * Update pg_database.datfrozenxid, and truncate pg_xact if possible. We
2565  * only need to do this once, not after each table.
2566  *
2567  * Even if we didn't vacuum anything, it may still be important to do
2568  * this, because one indirect effect of vac_update_datfrozenxid() is to
2569  * update ShmemVariableCache->xidVacLimit. That might need to be done
2570  * even if we haven't vacuumed anything, because relations with older
2571  * relfrozenxid values or other databases with older datfrozenxid values
2572  * might have been dropped, allowing xidVacLimit to advance.
2573  *
2574  * However, it's also important not to do this blindly in all cases,
2575  * because when autovacuum=off this will restart the autovacuum launcher.
2576  * If we're not careful, an infinite loop can result, where workers find
2577  * no work to do and restart the launcher, which starts another worker in
2578  * the same database that finds no work to do. To prevent that, we skip
2579  * this if (1) we found no work to do and (2) we skipped at least one
2580  * table due to concurrent autovacuum activity. In that case, the other
2581  * worker has already done it, or will do so when it finishes.
2582  */
2583  if (did_vacuum || !found_concurrent_worker)
2585 
2586  /* Finally close out the last transaction. */
2588 }
2589 
2590 /*
2591  * Execute a previously registered work item.
2592  */
2593 static void
2595 {
2596  char *cur_datname = NULL;
2597  char *cur_nspname = NULL;
2598  char *cur_relname = NULL;
2599 
2600  /*
2601  * Note we do not store table info in MyWorkerInfo, since this is not
2602  * vacuuming proper.
2603  */
2604 
2605  /*
2606  * Save the relation name for a possible error message, to avoid a catalog
2607  * lookup in case of an error. If any of these return NULL, then the
2608  * relation has been dropped since last we checked; skip it. Note: they
2609  * must live in a long-lived memory context because we call vacuum and
2610  * analyze in different transactions.
2611  */
2612  Assert(CurrentMemoryContext == AutovacMemCxt);
2613 
2614  cur_relname = get_rel_name(workitem->avw_relation);
2615  cur_nspname = get_namespace_name(get_rel_namespace(workitem->avw_relation));
2616  cur_datname = get_database_name(MyDatabaseId);
2617  if (!cur_relname || !cur_nspname || !cur_datname)
2618  goto deleted2;
2619 
2620  autovac_report_workitem(workitem, cur_nspname, cur_datname);
2621 
2622  /* clean up memory before each work item */
2624 
2625  /*
2626  * We will abort the current work item if something errors out, and
2627  * continue with the next one; in particular, this happens if we are
2628  * interrupted with SIGINT. Note that this means that the work item list
2629  * can be lossy.
2630  */
2631  PG_TRY();
2632  {
2633  /* Use PortalContext for any per-work-item allocations */
2635 
2636  /* have at it */
2637  switch (workitem->avw_type)
2638  {
2641  ObjectIdGetDatum(workitem->avw_relation),
2642  Int64GetDatum((int64) workitem->avw_blockNumber));
2643  break;
2644  default:
2645  elog(WARNING, "unrecognized work item found: type %d",
2646  workitem->avw_type);
2647  break;
2648  }
2649 
2650  /*
2651  * Clear a possible query-cancel signal, to avoid a late reaction to
2652  * an automatically-sent signal because of vacuuming the current table
2653  * (we're done with it, so it would make no sense to cancel at this
2654  * point.)
2655  */
2656  QueryCancelPending = false;
2657  }
2658  PG_CATCH();
2659  {
2660  /*
2661  * Abort the transaction, start a new one, and proceed with the next
2662  * table in our list.
2663  */
2664  HOLD_INTERRUPTS();
2665  errcontext("processing work entry for relation \"%s.%s.%s\"",
2666  cur_datname, cur_nspname, cur_relname);
2667  EmitErrorReport();
2668 
2669  /* this resets the PGXACT flags too */
2671  FlushErrorState();
2673 
2674  /* restart our transaction for the following operations */
2677  }
2678  PG_END_TRY();
2679 
2680  /* Make sure we're back in AutovacMemCxt */
2681  MemoryContextSwitchTo(AutovacMemCxt);
2682 
2683  /* We intentionally do not set did_vacuum here */
2684 
2685  /* be tidy */
2686 deleted2:
2687  if (cur_datname)
2688  pfree(cur_datname);
2689  if (cur_nspname)
2690  pfree(cur_nspname);
2691  if (cur_relname)
2692  pfree(cur_relname);
2693 }
2694 
2695 /*
2696  * extract_autovac_opts
2697  *
2698  * Given a relation's pg_class tuple, return the AutoVacOpts portion of
2699  * reloptions, if set; otherwise, return NULL.
2700  */
2701 static AutoVacOpts *
2703 {
2704  bytea *relopts;
2705  AutoVacOpts *av;
2706 
2707  Assert(((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_RELATION ||
2708  ((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_MATVIEW ||
2709  ((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_TOASTVALUE);
2710 
2711  relopts = extractRelOptions(tup, pg_class_desc, NULL);
2712  if (relopts == NULL)
2713  return NULL;
2714 
2715  av = palloc(sizeof(AutoVacOpts));
2716  memcpy(av, &(((StdRdOptions *) relopts)->autovacuum), sizeof(AutoVacOpts));
2717  pfree(relopts);
2718 
2719  return av;
2720 }
2721 
2722 /*
2723  * get_pgstat_tabentry_relid
2724  *
2725  * Fetch the pgstat entry of a table, either local to a database or shared.
2726  */
2727 static PgStat_StatTabEntry *
2728 get_pgstat_tabentry_relid(Oid relid, bool isshared, PgStat_StatDBEntry *shared,
2729  PgStat_StatDBEntry *dbentry)
2730 {
2731  PgStat_StatTabEntry *tabentry = NULL;
2732 
2733  if (isshared)
2734  {
2735  if (PointerIsValid(shared))
2736  tabentry = hash_search(shared->tables, &relid,
2737  HASH_FIND, NULL);
2738  }
2739  else if (PointerIsValid(dbentry))
2740  tabentry = hash_search(dbentry->tables, &relid,
2741  HASH_FIND, NULL);
2742 
2743  return tabentry;
2744 }
2745 
2746 /*
2747  * table_recheck_autovac
2748  *
2749  * Recheck whether a table still needs vacuum or analyze. Return value is a
2750  * valid autovac_table pointer if it does, NULL otherwise.
2751  *
2752  * Note that the returned autovac_table does not have the name fields set.
2753  */
2754 static autovac_table *
2755 table_recheck_autovac(Oid relid, HTAB *table_toast_map,
2756  TupleDesc pg_class_desc,
2757  int effective_multixact_freeze_max_age)
2758 {
2759  Form_pg_class classForm;
2760  HeapTuple classTup;
2761  bool dovacuum;
2762  bool doanalyze;
2763  autovac_table *tab = NULL;
2764  PgStat_StatTabEntry *tabentry;
2765  PgStat_StatDBEntry *shared;
2766  PgStat_StatDBEntry *dbentry;
2767  bool wraparound;
2768  AutoVacOpts *avopts;
2769 
2770  /* use fresh stats */
2772 
2775 
2776  /* fetch the relation's relcache entry */
2777  classTup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
2778  if (!HeapTupleIsValid(classTup))
2779  return NULL;
2780  classForm = (Form_pg_class) GETSTRUCT(classTup);
2781 
2782  /*
2783  * Get the applicable reloptions. If it is a TOAST table, try to get the
2784  * main table reloptions if the toast table itself doesn't have.
2785  */
2786  avopts = extract_autovac_opts(classTup, pg_class_desc);
2787  if (classForm->relkind == RELKIND_TOASTVALUE &&
2788  avopts == NULL && table_toast_map != NULL)
2789  {
2790  av_relation *hentry;
2791  bool found;
2792 
2793  hentry = hash_search(table_toast_map, &relid, HASH_FIND, &found);
2794  if (found && hentry->ar_hasrelopts)
2795  avopts = &hentry->ar_reloptions;
2796  }
2797 
2798  /* fetch the pgstat table entry */
2799  tabentry = get_pgstat_tabentry_relid(relid, classForm->relisshared,
2800  shared, dbentry);
2801 
2802  relation_needs_vacanalyze(relid, avopts, classForm, tabentry,
2803  effective_multixact_freeze_max_age,
2804  &dovacuum, &doanalyze, &wraparound);
2805 
2806  /* ignore ANALYZE for toast tables */
2807  if (classForm->relkind == RELKIND_TOASTVALUE)
2808  doanalyze = false;
2809 
2810  /* OK, it needs something done */
2811  if (doanalyze || dovacuum)
2812  {
2813  int freeze_min_age;
2814  int freeze_table_age;
2815  int multixact_freeze_min_age;
2816  int multixact_freeze_table_age;
2817  int vac_cost_limit;
2818  int vac_cost_delay;
2819  int log_min_duration;
2820 
2821  /*
2822  * Calculate the vacuum cost parameters and the freeze ages. If there
2823  * are options set in pg_class.reloptions, use them; in the case of a
2824  * toast table, try the main table too. Otherwise use the GUC
2825  * defaults, autovacuum's own first and plain vacuum second.
2826  */
2827 
2828  /* -1 in autovac setting means use plain vacuum_cost_delay */
2829  vac_cost_delay = (avopts && avopts->vacuum_cost_delay >= 0)
2830  ? avopts->vacuum_cost_delay
2831  : (autovacuum_vac_cost_delay >= 0)
2833  : VacuumCostDelay;
2834 
2835  /* 0 or -1 in autovac setting means use plain vacuum_cost_limit */
2836  vac_cost_limit = (avopts && avopts->vacuum_cost_limit > 0)
2837  ? avopts->vacuum_cost_limit
2840  : VacuumCostLimit;
2841 
2842  /* -1 in autovac setting means use log_autovacuum_min_duration */
2843  log_min_duration = (avopts && avopts->log_min_duration >= 0)
2844  ? avopts->log_min_duration
2846 
2847  /* these do not have autovacuum-specific settings */
2848  freeze_min_age = (avopts && avopts->freeze_min_age >= 0)
2849  ? avopts->freeze_min_age
2851 
2852  freeze_table_age = (avopts && avopts->freeze_table_age >= 0)
2853  ? avopts->freeze_table_age
2855 
2856  multixact_freeze_min_age = (avopts &&
2857  avopts->multixact_freeze_min_age >= 0)
2858  ? avopts->multixact_freeze_min_age
2860 
2861  multixact_freeze_table_age = (avopts &&
2862  avopts->multixact_freeze_table_age >= 0)
2863  ? avopts->multixact_freeze_table_age
2865 
2866  tab = palloc(sizeof(autovac_table));
2867  tab->at_relid = relid;
2868  tab->at_sharedrel = classForm->relisshared;
2870  (dovacuum ? VACOPT_VACUUM : 0) |
2871  (doanalyze ? VACOPT_ANALYZE : 0) |
2872  (!wraparound ? VACOPT_NOWAIT : 0);
2873  tab->at_params.freeze_min_age = freeze_min_age;
2874  tab->at_params.freeze_table_age = freeze_table_age;
2875  tab->at_params.multixact_freeze_min_age = multixact_freeze_min_age;
2876  tab->at_params.multixact_freeze_table_age = multixact_freeze_table_age;
2877  tab->at_params.is_wraparound = wraparound;
2878  tab->at_params.log_min_duration = log_min_duration;
2879  tab->at_vacuum_cost_limit = vac_cost_limit;
2880  tab->at_vacuum_cost_delay = vac_cost_delay;
2881  tab->at_relname = NULL;
2882  tab->at_nspname = NULL;
2883  tab->at_datname = NULL;
2884 
2885  /*
2886  * If any of the cost delay parameters has been set individually for
2887  * this table, disable the balancing algorithm.
2888  */
2889  tab->at_dobalance =
2890  !(avopts && (avopts->vacuum_cost_limit > 0 ||
2891  avopts->vacuum_cost_delay > 0));
2892  }
2893 
2894  heap_freetuple(classTup);
2895 
2896  return tab;
2897 }
2898 
2899 /*
2900  * relation_needs_vacanalyze
2901  *
2902  * Check whether a relation needs to be vacuumed or analyzed; return each into
2903  * "dovacuum" and "doanalyze", respectively. Also return whether the vacuum is
2904  * being forced because of Xid or multixact wraparound.
2905  *
2906  * relopts is a pointer to the AutoVacOpts options (either for itself in the
2907  * case of a plain table, or for either itself or its parent table in the case
2908  * of a TOAST table), NULL if none; tabentry is the pgstats entry, which can be
2909  * NULL.
2910  *
2911  * A table needs to be vacuumed if the number of dead tuples exceeds a
2912  * threshold. This threshold is calculated as
2913  *
2914  * threshold = vac_base_thresh + vac_scale_factor * reltuples
2915  *
2916  * For analyze, the analysis done is that the number of tuples inserted,
2917  * deleted and updated since the last analyze exceeds a threshold calculated
2918  * in the same fashion as above. Note that the collector actually stores
2919  * the number of tuples (both live and dead) that there were as of the last
2920  * analyze. This is asymmetric to the VACUUM case.
2921  *
2922  * We also force vacuum if the table's relfrozenxid is more than freeze_max_age
2923  * transactions back, and if its relminmxid is more than
2924  * multixact_freeze_max_age multixacts back.
2925  *
2926  * A table whose autovacuum_enabled option is false is
2927  * automatically skipped (unless we have to vacuum it due to freeze_max_age).
2928  * Thus autovacuum can be disabled for specific tables. Also, when the stats
2929  * collector does not have data about a table, it will be skipped.
2930  *
2931  * A table whose vac_base_thresh value is < 0 takes the base value from the
2932  * autovacuum_vacuum_threshold GUC variable. Similarly, a vac_scale_factor
2933  * value < 0 is substituted with the value of
2934  * autovacuum_vacuum_scale_factor GUC variable. Ditto for analyze.
2935  */
2936 static void
2938  AutoVacOpts *relopts,
2939  Form_pg_class classForm,
2940  PgStat_StatTabEntry *tabentry,
2941  int effective_multixact_freeze_max_age,
2942  /* output params below */
2943  bool *dovacuum,
2944  bool *doanalyze,
2945  bool *wraparound)
2946 {
2947  bool force_vacuum;
2948  bool av_enabled;
2949  float4 reltuples; /* pg_class.reltuples */
2950 
2951  /* constants from reloptions or GUC variables */
2952  int vac_base_thresh,
2953  anl_base_thresh;
2954  float4 vac_scale_factor,
2955  anl_scale_factor;
2956 
2957  /* thresholds calculated from above constants */
2958  float4 vacthresh,
2959  anlthresh;
2960 
2961  /* number of vacuum (resp. analyze) tuples at this time */
2962  float4 vactuples,
2963  anltuples;
2964 
2965  /* freeze parameters */
2966  int freeze_max_age;
2967  int multixact_freeze_max_age;
2968  TransactionId xidForceLimit;
2969  MultiXactId multiForceLimit;
2970 
2971  AssertArg(classForm != NULL);
2972  AssertArg(OidIsValid(relid));
2973 
2974  /*
2975  * Determine vacuum/analyze equation parameters. We have two possible
2976  * sources: the passed reloptions (which could be a main table or a toast
2977  * table), or the autovacuum GUC variables.
2978  */
2979 
2980  /* -1 in autovac setting means use plain vacuum_cost_delay */
2981  vac_scale_factor = (relopts && relopts->vacuum_scale_factor >= 0)
2982  ? relopts->vacuum_scale_factor
2984 
2985  vac_base_thresh = (relopts && relopts->vacuum_threshold >= 0)
2986  ? relopts->vacuum_threshold
2988 
2989  anl_scale_factor = (relopts && relopts->analyze_scale_factor >= 0)
2990  ? relopts->analyze_scale_factor
2992 
2993  anl_base_thresh = (relopts && relopts->analyze_threshold >= 0)
2994  ? relopts->analyze_threshold
2996 
2997  freeze_max_age = (relopts && relopts->freeze_max_age >= 0)
3000 
3001  multixact_freeze_max_age = (relopts && relopts->multixact_freeze_max_age >= 0)
3002  ? Min(relopts->multixact_freeze_max_age, effective_multixact_freeze_max_age)
3003  : effective_multixact_freeze_max_age;
3004 
3005  av_enabled = (relopts ? relopts->enabled : true);
3006 
3007  /* Force vacuum if table is at risk of wraparound */
3008  xidForceLimit = recentXid - freeze_max_age;
3009  if (xidForceLimit < FirstNormalTransactionId)
3010  xidForceLimit -= FirstNormalTransactionId;
3011  force_vacuum = (TransactionIdIsNormal(classForm->relfrozenxid) &&
3012  TransactionIdPrecedes(classForm->relfrozenxid,
3013  xidForceLimit));
3014  if (!force_vacuum)
3015  {
3016  multiForceLimit = recentMulti - multixact_freeze_max_age;
3017  if (multiForceLimit < FirstMultiXactId)
3018  multiForceLimit -= FirstMultiXactId;
3019  force_vacuum = MultiXactIdPrecedes(classForm->relminmxid,
3020  multiForceLimit);
3021  }
3022  *wraparound = force_vacuum;
3023 
3024  /* User disabled it in pg_class.reloptions? (But ignore if at risk) */
3025  if (!av_enabled && !force_vacuum)
3026  {
3027  *doanalyze = false;
3028  *dovacuum = false;
3029  return;
3030  }
3031 
3032  /*
3033  * If we found the table in the stats hash, and autovacuum is currently
3034  * enabled, make a threshold-based decision whether to vacuum and/or
3035  * analyze. If autovacuum is currently disabled, we must be here for
3036  * anti-wraparound vacuuming only, so don't vacuum (or analyze) anything
3037  * that's not being forced.
3038  */
3039  if (PointerIsValid(tabentry) && AutoVacuumingActive())
3040  {
3041  reltuples = classForm->reltuples;
3042  vactuples = tabentry->n_dead_tuples;
3043  anltuples = tabentry->changes_since_analyze;
3044 
3045  vacthresh = (float4) vac_base_thresh + vac_scale_factor * reltuples;
3046  anlthresh = (float4) anl_base_thresh + anl_scale_factor * reltuples;
3047 
3048  /*
3049  * Note that we don't need to take special consideration for stat
3050  * reset, because if that happens, the last vacuum and analyze counts
3051  * will be reset too.
3052  */
3053  elog(DEBUG3, "%s: vac: %.0f (threshold %.0f), anl: %.0f (threshold %.0f)",
3054  NameStr(classForm->relname),
3055  vactuples, vacthresh, anltuples, anlthresh);
3056 
3057  /* Determine if this table needs vacuum or analyze. */
3058  *dovacuum = force_vacuum || (vactuples > vacthresh);
3059  *doanalyze = (anltuples > anlthresh);
3060  }
3061  else
3062  {
3063  /*
3064  * Skip a table not found in stat hash, unless we have to force vacuum
3065  * for anti-wrap purposes. If it's not acted upon, there's no need to
3066  * vacuum it.
3067  */
3068  *dovacuum = force_vacuum;
3069  *doanalyze = false;
3070  }
3071 
3072  /* ANALYZE refuses to work with pg_statistic */
3073  if (relid == StatisticRelationId)
3074  *doanalyze = false;
3075 }
3076 
3077 /*
3078  * autovacuum_do_vac_analyze
3079  * Vacuum and/or analyze the specified table
3080  */
3081 static void
3083 {
3084  RangeVar rangevar;
3085 
3086  /* Set up command parameters --- use local variables instead of palloc */
3087  MemSet(&rangevar, 0, sizeof(rangevar));
3088 
3089  rangevar.schemaname = tab->at_nspname;
3090  rangevar.relname = tab->at_relname;
3091  rangevar.location = -1;
3092 
3093  /* Let pgstat know what we're doing */
3095 
3096  vacuum(tab->at_vacoptions, &rangevar, tab->at_relid, &tab->at_params, NIL,
3097  bstrategy, true);
3098 }
3099 
3100 /*
3101  * autovac_report_activity
3102  * Report to pgstat what autovacuum is doing
3103  *
3104  * We send a SQL string corresponding to what the user would see if the
3105  * equivalent command was to be issued manually.
3106  *
3107  * Note we assume that we are going to report the next command as soon as we're
3108  * done with the current one, and exit right after the last one, so we don't
3109  * bother to report "<IDLE>" or some such.
3110  */
3111 static void
3113 {
3114 #define MAX_AUTOVAC_ACTIV_LEN (NAMEDATALEN * 2 + 56)
3115  char activity[MAX_AUTOVAC_ACTIV_LEN];
3116  int len;
3117 
3118  /* Report the command and possible options */
3119  if (tab->at_vacoptions & VACOPT_VACUUM)
3120  snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
3121  "autovacuum: VACUUM%s",
3122  tab->at_vacoptions & VACOPT_ANALYZE ? " ANALYZE" : "");
3123  else
3124  snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
3125  "autovacuum: ANALYZE");
3126 
3127  /*
3128  * Report the qualified name of the relation.
3129  */
3130  len = strlen(activity);
3131 
3132  snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
3133  " %s.%s%s", tab->at_nspname, tab->at_relname,
3134  tab->at_params.is_wraparound ? " (to prevent wraparound)" : "");
3135 
3136  /* Set statement_timestamp() to current time for pg_stat_activity */
3138 
3140 }
3141 
3142 /*
3143  * autovac_report_workitem
3144  * Report to pgstat that autovacuum is processing a work item
3145  */
3146 static void
3148  const char *nspname, const char *relname)
3149 {
3150  char activity[MAX_AUTOVAC_ACTIV_LEN + 12 + 2];
3151  char blk[12 + 2];
3152  int len;
3153 
3154  switch (workitem->avw_type)
3155  {
3157  snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
3158  "autovacuum: BRIN summarize");
3159  break;
3160  }
3161 
3162  /*
3163  * Report the qualified name of the relation, and the block number if any
3164  */
3165  len = strlen(activity);
3166 
3167  if (BlockNumberIsValid(workitem->avw_blockNumber))
3168  snprintf(blk, sizeof(blk), " %u", workitem->avw_blockNumber);
3169  else
3170  blk[0] = '\0';
3171 
3172  snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
3173  " %s.%s%s", nspname, relname, blk);
3174 
3175  /* Set statement_timestamp() to current time for pg_stat_activity */
3177 
3179 }
3180 
3181 /*
3182  * AutoVacuumingActive
3183  * Check GUC vars and report whether the autovacuum process should be
3184  * running.
3185  */
3186 bool
3188 {
3190  return false;
3191  return true;
3192 }
3193 
3194 /*
3195  * Request one work item to the next autovacuum run processing our database.
3196  */
3197 void
3199  BlockNumber blkno)
3200 {
3201  int i;
3202 
3203  LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
3204 
3205  /*
3206  * Locate an unused work item and fill it with the given data.
3207  */
3208  for (i = 0; i < NUM_WORKITEMS; i++)
3209  {
3210  AutoVacuumWorkItem *workitem = &AutoVacuumShmem->av_workItems[i];
3211 
3212  if (workitem->avw_used)
3213  continue;
3214 
3215  workitem->avw_used = true;
3216  workitem->avw_active = false;
3217  workitem->avw_type = type;
3218  workitem->avw_database = MyDatabaseId;
3219  workitem->avw_relation = relationId;
3220  workitem->avw_blockNumber = blkno;
3221 
3222  /* done */
3223  break;
3224  }
3225 
3226  LWLockRelease(AutovacuumLock);
3227 }
3228 
3229 /*
3230  * autovac_init
3231  * This is called at postmaster initialization.
3232  *
3233  * All we do here is annoy the user if he got it wrong.
3234  */
3235 void
3237 {
3239  ereport(WARNING,
3240  (errmsg("autovacuum not started because of misconfiguration"),
3241  errhint("Enable the \"track_counts\" option.")));
3242 }
3243 
3244 /*
3245  * IsAutoVacuum functions
3246  * Return whether this is either a launcher autovacuum process or a worker
3247  * process.
3248  */
3249 bool
3251 {
3252  return am_autovacuum_launcher;
3253 }
3254 
3255 bool
3257 {
3258  return am_autovacuum_worker;
3259 }
3260 
3261 
3262 /*
3263  * AutoVacuumShmemSize
3264  * Compute space needed for autovacuum-related shared memory
3265  */
3266 Size
3268 {
3269  Size size;
3270 
3271  /*
3272  * Need the fixed struct and the array of WorkerInfoData.
3273  */
3274  size = sizeof(AutoVacuumShmemStruct);
3275  size = MAXALIGN(size);
3277  sizeof(WorkerInfoData)));
3278  return size;
3279 }
3280 
3281 /*
3282  * AutoVacuumShmemInit
3283  * Allocate and initialize autovacuum-related shared memory
3284  */
3285 void
3287 {
3288  bool found;
3289 
3290  AutoVacuumShmem = (AutoVacuumShmemStruct *)
3291  ShmemInitStruct("AutoVacuum Data",
3293  &found);
3294 
3295  if (!IsUnderPostmaster)
3296  {
3297  WorkerInfo worker;
3298  int i;
3299 
3300  Assert(!found);
3301 
3302  AutoVacuumShmem->av_launcherpid = 0;
3303  dlist_init(&AutoVacuumShmem->av_freeWorkers);
3304  dlist_init(&AutoVacuumShmem->av_runningWorkers);
3305  AutoVacuumShmem->av_startingWorker = NULL;
3306  memset(AutoVacuumShmem->av_workItems, 0,
3307  sizeof(AutoVacuumWorkItem) * NUM_WORKITEMS);
3308 
3309  worker = (WorkerInfo) ((char *) AutoVacuumShmem +
3310  MAXALIGN(sizeof(AutoVacuumShmemStruct)));
3311 
3312  /* initialize the WorkerInfo free list */
3313  for (i = 0; i < autovacuum_max_workers; i++)
3314  dlist_push_head(&AutoVacuumShmem->av_freeWorkers,
3315  &worker[i].wi_links);
3316  }
3317  else
3318  Assert(found);
3319 }
3320 
3321 /*
3322  * autovac_refresh_stats
3323  * Refresh pgstats data for an autovacuum process
3324  *
3325  * Cause the next pgstats read operation to obtain fresh data, but throttle
3326  * such refreshing in the autovacuum launcher. This is mostly to avoid
3327  * rereading the pgstats files too many times in quick succession when there
3328  * are many databases.
3329  *
3330  * Note: we avoid throttling in the autovac worker, as it would be
3331  * counterproductive in the recheck logic.
3332  */
3333 static void
3335 {
3337  {
3338  static TimestampTz last_read = 0;
3339  TimestampTz current_time;
3340 
3341  current_time = GetCurrentTimestamp();
3342 
3343  if (!TimestampDifferenceExceeds(last_read, current_time,
3345  return;
3346 
3347  last_read = current_time;
3348  }
3349 
3351 }
AutoVacuumWorkItemType avw_type
Definition: autovacuum.c:256
int vacuum_cost_delay
Definition: rel.h:263
BufferAccessStrategy GetAccessStrategy(BufferAccessStrategyType btype)
Definition: freelist.c:542
int autovacuum_work_mem
Definition: autovacuum.c:115
void ProcessCatchupInterrupt(void)
Definition: sinval.c:177
static bool am_autovacuum_worker
Definition: autovacuum.c:138
void InitializeTimeouts(void)
Definition: timeout.c:340
#define NIL
Definition: pg_list.h:69
bool ConditionalLockRelationOid(Oid relid, LOCKMODE lockmode)
Definition: lmgr.c:138
int multixact_freeze_table_age
Definition: vacuum.h:142
static int default_freeze_table_age
Definition: autovacuum.c:151
int autovacuum_naptime
Definition: autovacuum.c:116
void AbortCurrentTransaction(void)
Definition: xact.c:2996
#define SIGUSR1
Definition: win32.h:202
double autovacuum_anl_scale
Definition: autovacuum.c:120
TupleDesc CreateTupleDescCopy(TupleDesc tupdesc)
Definition: tupdesc.c:102
TimestampTz adl_next_worker
Definition: autovacuum.c:162
struct avw_dbase avw_dbase
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:200
MultiXactId adw_minmulti
Definition: autovacuum.c:173
#define DEBUG1
Definition: elog.h:25
AutoVacuumWorkItemType
Definition: autovacuum.h:23
int MyProcPid
Definition: globals.c:39
int errhint(const char *fmt,...)
Definition: elog.c:987
BackendId MyBackendId
Definition: globals.c:73
#define NUM_WORKITEMS
Definition: autovacuum.c:264
#define GETSTRUCT(TUP)
Definition: htup_details.h:656
#define MAX_AUTOVAC_SLEEPTIME
Definition: autovacuum.c:134
static void autovac_balance_cost(void)
Definition: autovacuum.c:1775
WorkerInfo av_startingWorker
Definition: autovacuum.c:288
sig_atomic_t av_signal[AutoVacNumSignals]
Definition: autovacuum.c:284
int vacuum_multixact_freeze_table_age
Definition: vacuum.c:61
void heap_endscan(HeapScanDesc scan)
Definition: heapam.c:1565
int at_vacuum_cost_delay
Definition: autovacuum.c:193
static bool am_autovacuum_launcher
Definition: autovacuum.c:137
#define HASH_CONTEXT
Definition: hsearch.h:93
#define HASH_ELEM
Definition: hsearch.h:87
#define WL_TIMEOUT
Definition: latch.h:127
void ProcessConfigFile(GucContext context)
bool autovacuum_start_daemon
Definition: autovacuum.c:113
dlist_head av_freeWorkers
Definition: autovacuum.c:286
uint32 TransactionId
Definition: c.h:391
int waittime
Definition: pg_standby.c:43
void vac_update_datfrozenxid(void)
Definition: vacuum.c:922
static int default_multixact_freeze_min_age
Definition: autovacuum.c:152
void InitPostmasterChild(void)
Definition: miscinit.c:176
MemoryContext hcxt
Definition: hsearch.h:78
NON_EXEC_STATIC void AutoVacLauncherMain(int argc, char *argv[]) pg_attribute_noreturn()
Definition: autovacuum.c:432
#define RelationGetDescr(relation)
Definition: rel.h:428
static void dlist_push_head(dlist_head *head, dlist_node *node)
Definition: ilist.h:300
#define DEBUG3
Definition: elog.h:23
void UnlockRelationOid(Oid relid, LOCKMODE lockmode)
Definition: lmgr.c:182
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1570
float8 vacuum_scale_factor
Definition: rel.h:272
void pgstat_report_activity(BackendState state, const char *cmd_str)
Definition: pgstat.c:2994
static void do_autovacuum(void)
Definition: autovacuum.c:1929
FormData_pg_database * Form_pg_database
Definition: pg_database.h:57
PGPROC * MyProc
Definition: proc.c:67
int64 TimestampTz
Definition: timestamp.h:39
struct WorkerInfoData * WorkerInfo
Definition: autovacuum.c:234
pid_t fork_process(void)
Definition: fork_process.c:31
int freeze_min_age
Definition: rel.h:265
#define dlist_foreach(iter, lhead)
Definition: ilist.h:507
void AtEOXact_Buffers(bool isCommit)
Definition: bufmgr.c:2415
ResourceOwner CurrentResourceOwner
Definition: resowner.c:138
int adl_score
Definition: autovacuum.c:163
char * pstrdup(const char *in)
Definition: mcxt.c:1076
#define DatabaseRelationId
Definition: pg_database.h:29
void CommitTransactionCommand(void)
Definition: xact.c:2756
const char * pgstat_get_backend_desc(BackendType backendType)
Definition: pgstat.c:4044
#define RelationRelationId
Definition: pg_class.h:29
TimestampTz wi_launchtime
Definition: autovacuum.c:226
static MemoryContext AutovacMemCxt
Definition: autovacuum.c:156
#define Min(x, y)
Definition: c.h:795
Oid get_rel_namespace(Oid relid)
Definition: lsyscache.c:1750
#define RELKIND_MATVIEW
Definition: pg_class.h:165
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
static MultiXactId recentMulti
Definition: autovacuum.c:147
#define AccessShareLock
Definition: lockdefs.h:36
struct WorkerInfoData WorkerInfoData
void set_ps_display(const char *activity, bool force)
Definition: ps_status.c:326
Size entrysize
Definition: hsearch.h:73
struct avl_dbase avl_dbase
int autovacuum_multixact_freeze_max_age
Definition: autovacuum.c:122
void proc_exit(int code)
Definition: ipc.c:99
static void autovac_refresh_stats(void)
Definition: autovacuum.c:3334
#define MemSet(start, val, len)
Definition: c.h:846
static volatile sig_atomic_t got_SIGHUP
Definition: autovacuum.c:141
PgStat_StatDBEntry * pgstat_fetch_stat_dbentry(Oid dbid)
Definition: pgstat.c:2387
#define SetProcessingMode(mode)
Definition: miscadmin.h:374
void BaseInit(void)
Definition: postinit.c:518
int snprintf(char *str, size_t count, const char *fmt,...) pg_attribute_printf(3
dlist_node wi_links
Definition: autovacuum.c:222
uint32 BlockNumber
Definition: block.h:31
int GetTempNamespaceBackendId(Oid namespaceId)
Definition: namespace.c:3195
WalTimeSample last_read[NUM_SYNC_REP_WAIT_MODE]
Definition: walsender.c:217
#define MIN_AUTOVAC_SLEEPTIME
Definition: autovacuum.c:133
#define heap_close(r, l)
Definition: heapam.h:97
#define lengthof(array)
Definition: c.h:556
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:902
char * at_datname
Definition: autovacuum.c:199
void ResetLatch(volatile Latch *latch)
Definition: latch.c:497
#define LOG
Definition: elog.h:26
#define STATS_READ_DELAY
Definition: autovacuum.c:130
void heap_freetuple(HeapTuple htup)
Definition: heaptuple.c:1373
unsigned int Oid
Definition: postgres_ext.h:31
NON_EXEC_STATIC void AutoVacWorkerMain(int argc, char *argv[]) pg_attribute_noreturn()
Definition: autovacuum.c:1514
volatile bool QueryCancelPending
Definition: globals.c:30
List * lappend_oid(List *list, Oid datum)
Definition: list.c:164
bool TimestampDifferenceExceeds(TimestampTz start_time, TimestampTz stop_time, int msec)
Definition: timestamp.c:1649
Snapshot GetTransactionSnapshot(void)
Definition: snapmgr.c:304
void ClosePostmasterPorts(bool am_syslogger)
Definition: postmaster.c:2452
#define OidIsValid(objectId)
Definition: c.h:532
void AbortOutOfAnyTransaction(void)
Definition: xact.c:4355
int freeze_table_age
Definition: vacuum.h:139
#define SIGQUIT
Definition: win32.h:189
int autovacuum_vac_cost_limit
Definition: autovacuum.c:125
void FlushErrorState(void)
Definition: elog.c:1587
#define PG_SETMASK(mask)
Definition: pqsignal.h:19
void StatementCancelHandler(SIGNAL_ARGS)
Definition: postgres.c:2675
#define SearchSysCache1(cacheId, key1)
Definition: syscache.h:159
void autovac_init(void)
Definition: autovacuum.c:3236
Oid ar_toastrelid
Definition: autovacuum.c:180
AutoVacuumSignal
Definition: autovacuum.c:241
MemoryContext PortalContext
Definition: mcxt.c:52
char * schemaname
Definition: primnodes.h:67
bool at_sharedrel
Definition: autovacuum.c:196
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1721
Datum brin_summarize_range(PG_FUNCTION_ARGS)
Definition: brin.c:853
#define RESUME_INTERRUPTS()
Definition: miscadmin.h:117
ErrorContextCallback * error_context_stack
Definition: elog.c:88
int location
Definition: primnodes.h:73
#define NAMEDATALEN
#define Anum_pg_class_relkind
Definition: pg_class.h:118
char * relname
Definition: primnodes.h:68
int freeze_table_age
Definition: rel.h:267
int at_vacuum_cost_limit
Definition: autovacuum.c:194
void pg_usleep(long microsec)
Definition: signal.c:53
Definition: dynahash.c:208
char * at_relname
Definition: autovacuum.c:197
#define dlist_container(type, membername, ptr)
Definition: ilist.h:477
void AtEOXact_SMgr(void)
Definition: smgr.c:798
bool pgstat_track_counts
Definition: pgstat.c:126
int WaitLatch(volatile Latch *latch, int wakeEvents, long timeout, uint32 wait_event_info)
Definition: latch.c:336
void pfree(void *pointer)
Definition: mcxt.c:949
#define SIG_IGN
Definition: win32.h:185
#define dlist_tail_element(type, membername, lhead)
Definition: ilist.h:496
void disable_all_timeouts(bool keep_indicators)
Definition: timeout.c:596
#define FirstNormalTransactionId
Definition: transam.h:34
bool AutoVacuumingActive(void)
Definition: autovacuum.c:3187
#define ObjectIdGetDatum(X)
Definition: postgres.h:513
#define ERROR
Definition: elog.h:43
int AutovacuumLauncherPid
Definition: autovacuum.c:305
struct AutoVacuumWorkItem AutoVacuumWorkItem
int VacuumCostLimit
Definition: globals.c:132
int autovacuum_freeze_max_age
Definition: autovacuum.c:121
int StartAutoVacWorker(void)
Definition: autovacuum.c:1476
int freeze_min_age
Definition: vacuum.h:138
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:372
void AtEOXact_Files(void)
Definition: fd.c:2649
int vacuum_multixact_freeze_min_age
Definition: vacuum.c:60
Definition: guc.h:75
static List * get_database_list(void)
Definition: autovacuum.c:1863
static volatile sig_atomic_t got_SIGTERM
Definition: autovacuum.c:143
Oid adl_datid
Definition: autovacuum.c:161
PgStat_Counter n_dead_tuples
Definition: pgstat.h:630
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:165
bool is_wraparound
Definition: vacuum.h:144
char * get_database_name(Oid dbid)
Definition: dbcommands.c:2056
AutoVacuumWorkItem av_workItems[NUM_WORKITEMS]
Definition: autovacuum.c:289
#define DEBUG2
Definition: elog.h:24
void InitProcess(void)
Definition: proc.c:288
_stringlist * dblist
Definition: pg_regress.c:72
TimestampTz last_autovac_time
Definition: pgstat.h:587
char * get_namespace_name(Oid nspid)
Definition: lsyscache.c:3033
int analyze_threshold
Definition: rel.h:262
void on_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:348
void SetConfigOption(const char *name, const char *value, GucContext context, GucSource source)
Definition: guc.c:6682
int autovacuum_vac_thresh
Definition: autovacuum.c:117
static AutoVacOpts * extract_autovac_opts(HeapTuple tup, TupleDesc pg_class_desc)
Definition: autovacuum.c:2702
bool IsUnderPostmaster
Definition: globals.c:101
void performDeletion(const ObjectAddress *object, DropBehavior behavior, int flags)
Definition: dependency.c:303
static void autovac_report_workitem(AutoVacuumWorkItem *workitem, const char *nspname, const char *relname)
Definition: autovacuum.c:3147
static PgStat_StatTabEntry * get_pgstat_tabentry_relid(Oid relid, bool isshared, PgStat_StatDBEntry *shared, PgStat_StatDBEntry *dbentry)
Definition: autovacuum.c:2728
TransactionId ReadNewTransactionId(void)
Definition: varsup.c:250
static void perform_work_item(AutoVacuumWorkItem *workitem)
Definition: autovacuum.c:2594
sigset_t UnBlockSig
Definition: pqsignal.c:22
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1244
#define FirstMultiXactId
Definition: multixact.h:24
MemoryContext CurrentMemoryContext
Definition: mcxt.c:37
static WorkerInfo MyWorkerInfo
Definition: autovacuum.c:302
HeapScanDesc heap_beginscan_catalog(Relation relation, int nkeys, ScanKey key)
Definition: heapam.c:1405
static void relation_needs_vacanalyze(Oid relid, AutoVacOpts *relopts, Form_pg_class classForm, PgStat_StatTabEntry *tabentry, int effective_multixact_freeze_max_age, bool *dovacuum, bool *doanalyze, bool *wraparound)
Definition: autovacuum.c:2937
static void dlist_delete(dlist_node *node)
Definition: ilist.h:358
Datum Int64GetDatum(int64 X)
Definition: fmgr.c:1791
int autovacuum_vac_cost_delay
Definition: autovacuum.c:124
static dlist_head DatabaseList
Definition: autovacuum.c:298
bool IsAutoVacuumWorkerProcess(void)
Definition: autovacuum.c:3256
int synchronous_commit
Definition: xact.c:83
#define ereport(elevel, rest)
Definition: elog.h:122
void pgstat_vacuum_stat(void)
Definition: pgstat.c:1022
int MultiXactMemberFreezeThreshold(void)
Definition: multixact.c:2817
int wi_cost_limit_base
Definition: autovacuum.c:231
char * adw_name
Definition: autovacuum.c:171
#define AssertArg(condition)
Definition: c.h:666
MemoryContext TopMemoryContext
Definition: mcxt.c:43
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
Definition: guc.h:72
List * lappend(List *list, void *datum)
Definition: list.c:128
static void avl_sigusr2_handler(SIGNAL_ARGS)
Definition: autovacuum.c:1411
static void launcher_determine_sleep(bool canlaunch, bool recursing, struct timeval *nap)
Definition: autovacuum.c:840
TransactionId adw_frozenxid
Definition: autovacuum.c:172
#define DLIST_STATIC_INIT(name)
Definition: ilist.h:248
#define WARNING
Definition: elog.h:40
static int db_comparator(const void *a, const void *b)
Definition: autovacuum.c:1105
float float4
Definition: c.h:374
void UnlockBuffers(void)
Definition: bufmgr.c:3518
#define MemoryContextResetAndDeleteChildren(ctx)
Definition: memutils.h:67
static void autovac_report_activity(autovac_table *tab)
Definition: autovacuum.c:3112
struct @18::@19 av[32]
#define HASH_BLOBS
Definition: hsearch.h:88
char * at_nspname
Definition: autovacuum.c:198
int multixact_freeze_table_age
Definition: rel.h:270
Size mul_size(Size s1, Size s2)
Definition: shmem.c:492
#define InvalidBackendId
Definition: backendid.h:23
#define WL_POSTMASTER_DEATH
Definition: latch.h:128
int vacuum_threshold
Definition: rel.h:261
MemoryContext AllocSetContextCreate(MemoryContext parent, const char *name, Size minContextSize, Size initBlockSize, Size maxBlockSize)
Definition: aset.c:322
#define RELKIND_TOASTVALUE
Definition: pg_class.h:163
PgStat_StatDBEntry * adw_entry
Definition: autovacuum.c:174
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
uintptr_t Datum
Definition: postgres.h:372
void ReleaseSysCache(HeapTuple tuple)
Definition: syscache.c:1117
#define StatisticRelationId
Definition: pg_statistic.h:29
Size add_size(Size s1, Size s2)
Definition: shmem.c:475
Oid adw_datid
Definition: autovacuum.c:170
Oid MyDatabaseId
Definition: globals.c:77
int PostAuthDelay
Definition: postgres.c:99
HeapTuple heap_getnext(HeapScanDesc scan, ScanDirection direction)
Definition: heapam.c:1808
Relation heap_open(Oid relationId, LOCKMODE lockmode)
Definition: heapam.c:1290
struct autovac_table autovac_table
#define MAX_AUTOVAC_ACTIV_LEN
Size AutoVacuumShmemSize(void)
Definition: autovacuum.c:3267
BlockNumber avw_blockNumber
Definition: autovacuum.c:261
Size keysize
Definition: hsearch.h:72
static TransactionId recentXid
Definition: autovacuum.c:146
dlist_node * cur
Definition: ilist.h:161
void EmitErrorReport(void)
Definition: elog.c:1446
PgStat_Counter changes_since_analyze
Definition: pgstat.h:631
#define SIGPIPE
Definition: win32.h:193
bytea * extractRelOptions(HeapTuple tuple, TupleDesc tupdesc, amoptions_function amoptions)
Definition: reloptions.c:956
#define SIGHUP
Definition: win32.h:188
int autovacuum_max_workers
Definition: autovacuum.c:114
#define InvalidOid
Definition: postgres_ext.h:36
#define TimestampTzPlusMilliseconds(tz, ms)
Definition: timestamp.h:56
static void dlist_init(dlist_head *head)
Definition: ilist.h:278
double autovacuum_vac_scale
Definition: autovacuum.c:118
void vacuum(int options, RangeVar *relation, Oid relid, VacuumParams *params, List *va_cols, BufferAccessStrategy bstrategy, bool isTopLevel)
Definition: vacuum.c:149
#define SIG_DFL
Definition: win32.h:183
int VacuumCostDelay
Definition: globals.c:133
VacuumParams at_params
Definition: autovacuum.c:192
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
pqsigfunc pqsignal(int signum, pqsigfunc handler)
Definition: signal.c:168
static volatile sig_atomic_t got_SIGUSR2
Definition: autovacuum.c:142
float8 analyze_scale_factor
Definition: rel.h:273
TransactionId MultiXactId
Definition: c.h:401
bool IsAutoVacuumLauncherProcess(void)
Definition: autovacuum.c:3250
#define PG_CATCH()
Definition: elog.h:293
#define Max(x, y)
Definition: c.h:789
int freeze_max_age
Definition: rel.h:266
int vacuum_cost_limit
Definition: rel.h:264
void SetLatch(volatile Latch *latch)
Definition: latch.c:414
#define HeapTupleIsValid(tuple)
Definition: htup.h:77
#define SIGNAL_ARGS
Definition: c.h:1068
#define Assert(condition)
Definition: c.h:664
#define lfirst(lc)
Definition: pg_list.h:106
#define PERFORM_DELETION_QUIETLY
Definition: dependency.h:176
void ResourceOwnerRelease(ResourceOwner owner, ResourceReleasePhase phase, bool isCommit, bool isTopLevel)
Definition: resowner.c:471
bool enabled
Definition: rel.h:260
int multixact_freeze_min_age
Definition: rel.h:268
void StartTransactionCommand(void)
Definition: xact.c:2685
static bool dlist_is_empty(dlist_head *head)
Definition: ilist.h:289
void pgstat_clear_snapshot(void)
Definition: pgstat.c:5560
size_t Size
Definition: c.h:350
char * dbname
Definition: streamutil.c:45
static void FreeWorkerInfo(int code, Datum arg)
Definition: autovacuum.c:1710
static AutoVacuumShmemStruct * AutoVacuumShmem
Definition: autovacuum.c:292
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1117
int vacuum_freeze_min_age
Definition: vacuum.c:58
void AbortBufferIO(void)
Definition: bufmgr.c:3974
#define MAXALIGN(LEN)
Definition: c.h:576
int log_min_duration
Definition: vacuum.h:145
static void av_sighup_handler(SIGNAL_ARGS)
Definition: autovacuum.c:1399
bool MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
Definition: multixact.c:3140
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1385
void AutoVacWorkerFailed(void)
Definition: autovacuum.c:1392
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1375
int vacuum_freeze_table_age
Definition: vacuum.c:59
int log_min_duration
Definition: rel.h:271
sigjmp_buf * PG_exception_stack
Definition: elog.c:90
#define CharGetDatum(X)
Definition: postgres.h:422
dlist_node adl_node
Definition: autovacuum.c:164
#define pg_attribute_noreturn()
Definition: c.h:642
static void dlist_move_head(dlist_head *head, dlist_node *node)
Definition: ilist.h:385
static void rebuild_database_list(Oid newdb)
Definition: autovacuum.c:924
int Log_autovacuum_min_duration
Definition: autovacuum.c:127
static pid_t AutoVacPID
Definition: postmaster.c:253
FormData_pg_class * Form_pg_class
Definition: pg_class.h:95
#define SearchSysCacheCopy1(cacheId, key1)
Definition: syscache.h:168
AutoVacOpts ar_reloptions
Definition: autovacuum.c:183
#define AccessExclusiveLock
Definition: lockdefs.h:45
static void autovacuum_do_vac_analyze(autovac_table *tab, BufferAccessStrategy bstrategy)
Definition: autovacuum.c:3082
void SetCurrentStatementStartTimestamp(void)
Definition: xact.c:742
void InitPostgres(const char *in_dbname, Oid dboid, const char *username, Oid useroid, char *out_dbname)
Definition: postinit.c:559
void * palloc(Size size)
Definition: mcxt.c:848
int errmsg(const char *fmt,...)
Definition: elog.c:797
void die(SIGNAL_ARGS)
Definition: postgres.c:2644
static dlist_node * dlist_pop_head_node(dlist_head *head)
Definition: ilist.h:368
#define HOLD_INTERRUPTS()
Definition: miscadmin.h:115
static MemoryContext DatabaseListCxt
Definition: autovacuum.c:299
int i
void FloatExceptionHandler(SIGNAL_ARGS)
Definition: postgres.c:2696
#define errcontext
Definition: elog.h:164
#define NameStr(name)
Definition: c.h:493
struct av_relation av_relation
void ScanKeyInit(ScanKey entry, AttrNumber attributeNumber, StrategyNumber strategy, RegProcedure procedure, Datum argument)
Definition: scankey.c:76
int multixact_freeze_max_age
Definition: rel.h:269
void * arg
void AutoVacuumUpdateDelay(void)
Definition: autovacuum.c:1759
struct Latch * MyLatch
Definition: globals.c:52
void AutoVacuumRequestWork(AutoVacuumWorkItemType type, Oid relationId, BlockNumber blkno)
Definition: autovacuum.c:3198
Definition: c.h:433
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:98
PGPROC * wi_proc
Definition: autovacuum.c:225
static void launch_worker(TimestampTz now)
Definition: autovacuum.c:1340
void SendPostmasterSignal(PMSignalReason reason)
Definition: pmsignal.c:113
#define elog
Definition: elog.h:219
int StartAutoVacLauncher(void)
Definition: autovacuum.c:394
void LWLockReleaseAll(void)
Definition: lwlock.c:1820
#define HeapTupleGetOid(tuple)
Definition: htup_details.h:695
#define qsort(a, b, c, d)
Definition: port.h:443
void procsignal_sigusr1_handler(SIGNAL_ARGS)
Definition: procsignal.c:261
void AtEOXact_HashTables(bool isCommit)
Definition: dynahash.c:1830
#define RELPERSISTENCE_TEMP
Definition: pg_class.h:172
void AutoVacuumShmemInit(void)
Definition: autovacuum.c:3286
dlist_head av_runningWorkers
Definition: autovacuum.c:287
#define SIGCHLD
Definition: win32.h:198
void TimestampDifference(TimestampTz start_time, TimestampTz stop_time, long *secs, int *microsecs)
Definition: timestamp.c:1624
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
#define PG_TRY()
Definition: elog.h:284
#define PERFORM_DELETION_SKIP_EXTENSIONS
Definition: dependency.h:178
#define RELKIND_RELATION
Definition: pg_class.h:160
Definition: proc.h:95
bool ar_hasrelopts
Definition: autovacuum.c:182
Definition: pg_list.h:45
char * get_rel_name(Oid relid)
Definition: lsyscache.c:1726
#define PointerIsValid(pointer)
Definition: c.h:520
int pid
Definition: proc.h:109
static int default_multixact_freeze_table_age
Definition: autovacuum.c:153
#define WL_LATCH_SET
Definition: latch.h:124
void pgstat_report_autovac(Oid dboid)
Definition: pgstat.c:1383
void quickdie(SIGNAL_ARGS)
Definition: postgres.c:2583
static Oid do_start_worker(void)
Definition: autovacuum.c:1125
int multixact_freeze_min_age
Definition: vacuum.h:140
#define dlist_reverse_foreach(iter, lhead)
Definition: ilist.h:538
static int default_freeze_min_age
Definition: autovacuum.c:150
Datum now(PG_FUNCTION_ARGS)
Definition: timestamp.c:1534
#define DirectFunctionCall2(func, arg1, arg2)
Definition: fmgr.h:587
#define PG_END_TRY()
Definition: elog.h:300
#define BTEqualStrategyNumber
Definition: stratnum.h:31
#define SIGUSR2
Definition: win32.h:203
bool at_dobalance
Definition: autovacuum.c:195
#define lfirst_oid(lc)
Definition: pg_list.h:108
static autovac_table * table_recheck_autovac(Oid relid, HTAB *table_toast_map, TupleDesc pg_class_desc, int effective_multixact_freeze_max_age)
Definition: autovacuum.c:2755
static void avl_sigterm_handler(SIGNAL_ARGS)
Definition: autovacuum.c:1423
#define PERFORM_DELETION_INTERNAL
Definition: dependency.h:174
void init_ps_display(const char *username, const char *dbname, const char *host_info, const char *initial_str)
Definition: ps_status.c:244
int autovacuum_anl_thresh
Definition: autovacuum.c:119
PGPROC * BackendIdGetProc(int backendID)
Definition: sinvaladt.c:377
MultiXactId ReadNextMultiXactId(void)
Definition: multixact.c:721
#define NON_EXEC_STATIC
Definition: c.h:1102