#include "postgres.h"
#include <signal.h>
#include <sys/time.h>
#include <unistd.h>
#include "access/heapam.h"
#include "access/htup_details.h"
#include "access/multixact.h"
#include "access/reloptions.h"
#include "access/tableam.h"
#include "access/transam.h"
#include "access/xact.h"
#include "catalog/dependency.h"
#include "catalog/namespace.h"
#include "catalog/pg_database.h"
#include "catalog/pg_namespace.h"
#include "commands/dbcommands.h"
#include "commands/vacuum.h"
#include "common/int.h"
#include "lib/ilist.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
#include "pgstat.h"
#include "postmaster/autovacuum.h"
#include "postmaster/interrupt.h"
#include "postmaster/postmaster.h"
#include "storage/bufmgr.h"
#include "storage/ipc.h"
#include "storage/latch.h"
#include "storage/lmgr.h"
#include "storage/pmsignal.h"
#include "storage/proc.h"
#include "storage/procsignal.h"
#include "storage/smgr.h"
#include "tcop/tcopprot.h"
#include "utils/fmgroids.h"
#include "utils/fmgrprotos.h"
#include "utils/guc_hooks.h"
#include "utils/lsyscache.h"
#include "utils/memutils.h"
#include "utils/ps_status.h"
#include "utils/rel.h"
#include "utils/snapmgr.h"
#include "utils/syscache.h"
#include "utils/timeout.h"
#include "utils/timestamp.h"

Include dependency graph for autovacuum.c:

Data Structures
struct	avl_dbase

struct	avw_dbase

struct	av_relation

struct	autovac_table

struct	WorkerInfoData

struct	AutoVacuumWorkItem

struct	AutoVacuumShmemStruct

Macros
#define	MIN_AUTOVAC_SLEEPTIME 100.0 /* milliseconds */

#define	MAX_AUTOVAC_SLEEPTIME 300 /* seconds */

#define	NUM_WORKITEMS 256

#define	MAX_AUTOVAC_ACTIV_LEN (NAMEDATALEN * 2 + 56)

Typedefs
typedef struct avl_dbase	avl_dbase

typedef struct avw_dbase	avw_dbase

typedef struct av_relation	av_relation

typedef struct autovac_table	autovac_table

typedef struct WorkerInfoData	WorkerInfoData

typedef struct WorkerInfoData *	WorkerInfo

typedef struct AutoVacuumWorkItem	AutoVacuumWorkItem

Enumerations
enum	AutoVacuumSignal { AutoVacForkFailed , AutoVacRebalance , AutoVacNumSignals }

Functions
static Oid	do_start_worker (void)

static void	HandleAutoVacLauncherInterrupts (void)

static void	AutoVacLauncherShutdown (void)

static void	launcher_determine_sleep (bool canlaunch, bool recursing, struct timeval *nap)

static void	rebuild_database_list (Oid newdb)

static int	db_comparator (const void a, const void b)

static void	launch_worker (TimestampTz now)

void	AutoVacWorkerFailed (void)

static void	avl_sigusr2_handler (SIGNAL_ARGS)

void	AutoVacWorkerMain (char *startup_data, size_t startup_data_len)

static void	FreeWorkerInfo (int code, Datum arg)

void	VacuumUpdateCosts (void)

void	AutoVacuumUpdateCostLimit (void)

static void	autovac_recalculate_workers_for_balance (void)

static List *	get_database_list (void)

static void	do_autovacuum (void)

static void	perform_work_item (AutoVacuumWorkItem *workitem)

static AutoVacOpts *	extract_autovac_opts (HeapTuple tup, TupleDesc pg_class_desc)

static autovac_table *	table_recheck_autovac (Oid relid, HTAB *table_toast_map, TupleDesc pg_class_desc, int effective_multixact_freeze_max_age)

static void	recheck_relation_needs_vacanalyze (Oid relid, AutoVacOpts avopts, Form_pg_class classForm, int effective_multixact_freeze_max_age, bool dovacuum, bool doanalyze, bool wraparound)

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)

static void	autovacuum_do_vac_analyze (autovac_table *tab, BufferAccessStrategy bstrategy)

static void	autovac_report_activity (autovac_table *tab)

static void	autovac_report_workitem (AutoVacuumWorkItem workitem, const char nspname, const char *relname)

bool	AutoVacuumingActive (void)

bool	AutoVacuumRequestWork (AutoVacuumWorkItemType type, Oid relationId, BlockNumber blkno)

void	autovac_init (void)

Size	AutoVacuumShmemSize (void)

void	AutoVacuumShmemInit (void)

bool	check_autovacuum_work_mem (int newval, void *extra, GucSource source)

Variables
bool	autovacuum_start_daemon = false

int	autovacuum_max_workers

int	autovacuum_work_mem = -1

int	autovacuum_naptime

int	autovacuum_vac_thresh

double	autovacuum_vac_scale

int	autovacuum_vac_ins_thresh

double	autovacuum_vac_ins_scale

int	autovacuum_anl_thresh

double	autovacuum_anl_scale

int	autovacuum_freeze_max_age

int	autovacuum_multixact_freeze_max_age

double	autovacuum_vac_cost_delay

int	autovacuum_vac_cost_limit

int	Log_autovacuum_min_duration = 600000

static double	av_storage_param_cost_delay = -1

static int	av_storage_param_cost_limit = -1

static volatile sig_atomic_t	got_SIGUSR2 = false

static TransactionId	recentXid

static MultiXactId	recentMulti

static int	default_freeze_min_age

static int	default_freeze_table_age

static int	default_multixact_freeze_min_age

static int	default_multixact_freeze_table_age

static MemoryContext	AutovacMemCxt

static AutoVacuumShmemStruct *	AutoVacuumShmem

static dlist_head	DatabaseList = DLIST_STATIC_INIT(DatabaseList)

static MemoryContext	DatabaseListCxt = NULL

static WorkerInfo	MyWorkerInfo = NULL

int	AutovacuumLauncherPid = 0

Macro Definition Documentation

◆ MAX_AUTOVAC_ACTIV_LEN

#define MAX_AUTOVAC_ACTIV_LEN (NAMEDATALEN * 2 + 56)

◆ MAX_AUTOVAC_SLEEPTIME

#define MAX_AUTOVAC_SLEEPTIME 300 /* seconds */

Definition at line 136 of file autovacuum.c.

◆ MIN_AUTOVAC_SLEEPTIME

#define MIN_AUTOVAC_SLEEPTIME 100.0 /* milliseconds */

Definition at line 135 of file autovacuum.c.

◆ NUM_WORKITEMS

#define NUM_WORKITEMS 256

Definition at line 268 of file autovacuum.c.

Typedef Documentation

◆ autovac_table

typedef struct autovac_table autovac_table

◆ AutoVacuumWorkItem

typedef struct AutoVacuumWorkItem AutoVacuumWorkItem

◆ av_relation

typedef struct av_relation av_relation

◆ avl_dbase

typedef struct avl_dbase avl_dbase

◆ avw_dbase

typedef struct avw_dbase avw_dbase

◆ WorkerInfo

typedef struct WorkerInfoData* WorkerInfo

Definition at line 238 of file autovacuum.c.

◆ WorkerInfoData

typedef struct WorkerInfoData WorkerInfoData

Enumeration Type Documentation

◆ AutoVacuumSignal

enum AutoVacuumSignal

Enumerator
AutoVacForkFailed
AutoVacRebalance
AutoVacNumSignals

Definition at line 245 of file autovacuum.c.

 {
     AutoVacForkFailed,          /* failed trying to start a worker */
     AutoVacRebalance,           /* rebalance the cost limits */
     AutoVacNumSignals,          /* must be last */
 }           AutoVacuumSignal;

Function Documentation

◆ autovac_init()

void autovac_init ( void )

Definition at line 3243 of file autovacuum.c.

 {
     if (autovacuum_start_daemon && !pgstat_track_counts)
         ereport(WARNING,
                 (errmsg("autovacuum not started because of misconfiguration"),
                  errhint("Enable the \"track_counts\" option.")));
 }

References autovacuum_start_daemon, ereport, errhint(), errmsg(), pgstat_track_counts, and WARNING.

Referenced by PostmasterMain().

◆ autovac_recalculate_workers_for_balance()

static void autovac_recalculate_workers_for_balance ( void )

static

Definition at line 1748 of file autovacuum.c.

 {
     dlist_iter  iter;
     int         orig_nworkers_for_balance;
     int         nworkers_for_balance = 0;
  
     Assert(LWLockHeldByMe(AutovacuumLock));
  
     orig_nworkers_for_balance =
         pg_atomic_read_u32(&AutoVacuumShmem->av_nworkersForBalance);
  
     dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
     {
         WorkerInfo  worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
  
         if (worker->wi_proc == NULL ||
             pg_atomic_unlocked_test_flag(&worker->wi_dobalance))
             continue;
  
         nworkers_for_balance++;
     }
  
     if (nworkers_for_balance != orig_nworkers_for_balance)
         pg_atomic_write_u32(&AutoVacuumShmem->av_nworkersForBalance,
                             nworkers_for_balance);
 }

References Assert, AutoVacuumShmem, AutoVacuumShmemStruct::av_nworkersForBalance, AutoVacuumShmemStruct::av_runningWorkers, dlist_iter::cur, dlist_container, dlist_foreach, LWLockHeldByMe(), pg_atomic_read_u32(), pg_atomic_unlocked_test_flag(), pg_atomic_write_u32(), WorkerInfoData::wi_dobalance, and WorkerInfoData::wi_proc.

Referenced by do_autovacuum().

◆ autovac_report_activity()

static void autovac_report_activity ( autovac_table * tab )

static

Definition at line 3114 of file autovacuum.c.

 {
 #define MAX_AUTOVAC_ACTIV_LEN (NAMEDATALEN * 2 + 56)
     char        activity[MAX_AUTOVAC_ACTIV_LEN];
     int         len;
  
     /* Report the command and possible options */
     if (tab->at_params.options & VACOPT_VACUUM)
         snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
                  "autovacuum: VACUUM%s",
                  tab->at_params.options & VACOPT_ANALYZE ? " ANALYZE" : "");
     else
         snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
                  "autovacuum: ANALYZE");
  
     /*
      * Report the qualified name of the relation.
      */
     len = strlen(activity);
  
     snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
              " %s.%s%s", tab->at_nspname, tab->at_relname,
              tab->at_params.is_wraparound ? " (to prevent wraparound)" : "");
  
     /* Set statement_timestamp() to current time for pg_stat_activity */
     SetCurrentStatementStartTimestamp();
  
     pgstat_report_activity(STATE_RUNNING, activity);
 }

References autovac_table::at_nspname, autovac_table::at_params, autovac_table::at_relname, VacuumParams::is_wraparound, len, MAX_AUTOVAC_ACTIV_LEN, VacuumParams::options, pgstat_report_activity(), SetCurrentStatementStartTimestamp(), snprintf, STATE_RUNNING, VACOPT_ANALYZE, and VACOPT_VACUUM.

Referenced by autovacuum_do_vac_analyze().

◆ autovac_report_workitem()

static void autovac_report_workitem	(	AutoVacuumWorkItem *	workitem,
		const char *	nspname,
		const char *	relname
	)

static

Definition at line 3149 of file autovacuum.c.

 {
     char        activity[MAX_AUTOVAC_ACTIV_LEN + 12 + 2];
     char        blk[12 + 2];
     int         len;
  
     switch (workitem->avw_type)
     {
         case AVW_BRINSummarizeRange:
             snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
                      "autovacuum: BRIN summarize");
             break;
     }
  
     /*
      * Report the qualified name of the relation, and the block number if any
      */
     len = strlen(activity);
  
     if (BlockNumberIsValid(workitem->avw_blockNumber))
         snprintf(blk, sizeof(blk), " %u", workitem->avw_blockNumber);
     else
         blk[0] = '\0';
  
     snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
              " %s.%s%s", nspname, relname, blk);
  
     /* Set statement_timestamp() to current time for pg_stat_activity */
     SetCurrentStatementStartTimestamp();
  
     pgstat_report_activity(STATE_RUNNING, activity);
 }

References AutoVacuumWorkItem::avw_blockNumber, AVW_BRINSummarizeRange, AutoVacuumWorkItem::avw_type, BlockNumberIsValid(), len, MAX_AUTOVAC_ACTIV_LEN, pgstat_report_activity(), relname, SetCurrentStatementStartTimestamp(), snprintf, and STATE_RUNNING.

Referenced by perform_work_item().

◆ AutoVacLauncherShutdown()

static void AutoVacLauncherShutdown ( void )

static

Definition at line 316 of file autovacuum.c.

 {
     sigjmp_buf  local_sigjmp_buf;
  
     Assert(startup_data_len == 0);
  
     /* Release postmaster's working memory context */
     if (PostmasterContext)
     {
         MemoryContextDelete(PostmasterContext);
         PostmasterContext = NULL;
     }
  
     MyBackendType = B_AUTOVAC_LAUNCHER;
     init_ps_display(NULL);
  
     ereport(DEBUG1,
             (errmsg_internal("autovacuum launcher started")));
  
     if (PostAuthDelay)
         pg_usleep(PostAuthDelay * 1000000L);
  
     SetProcessingMode(InitProcessing);
  
     /*
      * Set up signal handlers.  We operate on databases much like a regular
      * backend, so we use the same signal handling.  See equivalent code in
      * tcop/postgres.c.
      */
     pqsignal(SIGHUP, SignalHandlerForConfigReload);
     pqsignal(SIGINT, StatementCancelHandler);
     pqsignal(SIGTERM, SignalHandlerForShutdownRequest);
     /* SIGQUIT handler was already set up by InitPostmasterChild */
  
     InitializeTimeouts();       /* establishes SIGALRM handler */
  
     pqsignal(SIGPIPE, SIG_IGN);
     pqsignal(SIGUSR1, procsignal_sigusr1_handler);
     pqsignal(SIGUSR2, avl_sigusr2_handler);
     pqsignal(SIGFPE, FloatExceptionHandler);
     pqsignal(SIGCHLD, SIG_DFL);
  
     /*
      * Create a per-backend PGPROC struct in shared memory.  We must do this
      * before we can use LWLocks or access any shared memory.
      */
     InitProcess();
  
     /* Early initialization */
     BaseInit();
  
     InitPostgres(NULL, InvalidOid, NULL, InvalidOid, 0, NULL);
  
     SetProcessingMode(NormalProcessing);
  
     /*
      * Create a memory context that we will do all our work in.  We do this so
      * that we can reset the context during error recovery and thereby avoid
      * possible memory leaks.
      */
     AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
                                           "Autovacuum Launcher",
                                           ALLOCSET_DEFAULT_SIZES);
     MemoryContextSwitchTo(AutovacMemCxt);
  
     /*
      * If an exception is encountered, processing resumes here.
      *
      * This code is a stripped down version of PostgresMain error recovery.
      *
      * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
      * (to wit, BlockSig) will be restored when longjmp'ing to here.  Thus,
      * signals other than SIGQUIT will be blocked until we complete error
      * recovery.  It might seem that this policy makes the HOLD_INTERRUPTS()
      * call redundant, but it is not since InterruptPending might be set
      * already.
      */
     if (sigsetjmp(local_sigjmp_buf, 1) != 0)
     {
         /* since not using PG_TRY, must reset error stack by hand */
         error_context_stack = NULL;
  
         /* Prevents interrupts while cleaning up */
         HOLD_INTERRUPTS();
  
         /* Forget any pending QueryCancel or timeout request */
         disable_all_timeouts(false);
         QueryCancelPending = false; /* second to avoid race condition */
  
         /* Report the error to the server log */
         EmitErrorReport();
  
         /* Abort the current transaction in order to recover */
         AbortCurrentTransaction();
  
         /*
          * Release any other resources, for the case where we were not in a
          * transaction.
          */
         LWLockReleaseAll();
         pgstat_report_wait_end();
         UnlockBuffers();
         /* this is probably dead code, but let's be safe: */
         if (AuxProcessResourceOwner)
             ReleaseAuxProcessResources(false);
         AtEOXact_Buffers(false);
         AtEOXact_SMgr();
         AtEOXact_Files(false);
         AtEOXact_HashTables(false);
  
         /*
          * Now return to normal top-level context and clear ErrorContext for
          * next time.
          */
         MemoryContextSwitchTo(AutovacMemCxt);
         FlushErrorState();
  
         /* Flush any leaked data in the top-level context */
         MemoryContextReset(AutovacMemCxt);
  
         /* don't leave dangling pointers to freed memory */
         DatabaseListCxt = NULL;
         dlist_init(&DatabaseList);
  
         /* Now we can allow interrupts again */
         RESUME_INTERRUPTS();
  
         /* if in shutdown mode, no need for anything further; just go away */
         if (ShutdownRequestPending)
             AutoVacLauncherShutdown();
  
         /*
          * Sleep at least 1 second after any error.  We don't want to be
          * filling the error logs as fast as we can.
          */
         pg_usleep(1000000L);
     }
  
     /* We can now handle ereport(ERROR) */
     PG_exception_stack = &local_sigjmp_buf;
  
     /* must unblock signals before calling rebuild_database_list */
     sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
  
     /*
      * Set always-secure search path.  Launcher doesn't connect to a database,
      * so this has no effect.
      */
     SetConfigOption("search_path", "", PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force zero_damaged_pages OFF in the autovac process, even if it is set
      * in postgresql.conf.  We don't really want such a dangerous option being
      * applied non-interactively.
      */
     SetConfigOption("zero_damaged_pages", "false", PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force settable timeouts off to avoid letting these settings prevent
      * regular maintenance from being executed.
      */
     SetConfigOption("statement_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     SetConfigOption("transaction_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     SetConfigOption("lock_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     SetConfigOption("idle_in_transaction_session_timeout", "0",
                     PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force default_transaction_isolation to READ COMMITTED.  We don't want
      * to pay the overhead of serializable mode, nor add any risk of causing
      * deadlocks or delaying other transactions.
      */
     SetConfigOption("default_transaction_isolation", "read committed",
                     PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Even when system is configured to use a different fetch consistency,
      * for autovac we always want fresh stats.
      */
     SetConfigOption("stats_fetch_consistency", "none", PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * In emergency mode, just start a worker (unless shutdown was requested)
      * and go away.
      */
     if (!AutoVacuumingActive())
     {
         if (!ShutdownRequestPending)
             do_start_worker();
         proc_exit(0);           /* done */
     }
  
     AutoVacuumShmem->av_launcherpid = MyProcPid;
  
     /*
      * Create the initial database list.  The invariant we want this list to
      * keep is that it's ordered by decreasing next_time.  As soon as an entry
      * is updated to a higher time, it will be moved to the front (which is
      * correct because the only operation is to add autovacuum_naptime to the
      * entry, and time always increases).
      */
     rebuild_database_list(InvalidOid);
  
     /* loop until shutdown request */
     while (!ShutdownRequestPending)
     {
         struct timeval nap;
         TimestampTz current_time = 0;
         bool        can_launch;
  
         /*
          * This loop is a bit different from the normal use of WaitLatch,
          * because we'd like to sleep before the first launch of a child
          * process.  So it's WaitLatch, then ResetLatch, then check for
          * wakening conditions.
          */
  
         launcher_determine_sleep(!dlist_is_empty(&AutoVacuumShmem->av_freeWorkers),
                                  false, &nap);
  
         /*
          * Wait until naptime expires or we get some type of signal (all the
          * signal handlers will wake us by calling SetLatch).
          */
         (void) WaitLatch(MyLatch,
                          WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
                          (nap.tv_sec * 1000L) + (nap.tv_usec / 1000L),
                          WAIT_EVENT_AUTOVACUUM_MAIN);
  
         ResetLatch(MyLatch);
  
         HandleAutoVacLauncherInterrupts();
  
         /*
          * a worker finished, or postmaster signaled failure to start a worker
          */
         if (got_SIGUSR2)
         {
             got_SIGUSR2 = false;
  
             /* rebalance cost limits, if needed */
             if (AutoVacuumShmem->av_signal[AutoVacRebalance])
             {
                 LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
                 AutoVacuumShmem->av_signal[AutoVacRebalance] = false;
                 autovac_recalculate_workers_for_balance();
                 LWLockRelease(AutovacuumLock);
             }
  
             if (AutoVacuumShmem->av_signal[AutoVacForkFailed])
             {
                 /*
                  * If the postmaster failed to start a new worker, we sleep
                  * for a little while and resend the signal.  The new worker's
                  * state is still in memory, so this is sufficient.  After
                  * that, we restart the main loop.
                  *
                  * XXX should we put a limit to the number of times we retry?
                  * I don't think it makes much sense, because a future start
                  * of a worker will continue to fail in the same way.
                  */
                 AutoVacuumShmem->av_signal[AutoVacForkFailed] = false;
                 pg_usleep(1000000L);    /* 1s */
                 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER);
                 continue;
             }
         }
  
         /*
          * There are some conditions that we need to check before trying to
          * start a worker.  First, we need to make sure that there is a worker
          * slot available.  Second, we need to make sure that no other worker
          * failed while starting up.
          */
  
         current_time = GetCurrentTimestamp();
         LWLockAcquire(AutovacuumLock, LW_SHARED);
  
         can_launch = !dlist_is_empty(&AutoVacuumShmem->av_freeWorkers);
  
         if (AutoVacuumShmem->av_startingWorker != NULL)
         {
             int         waittime;
             WorkerInfo  worker = AutoVacuumShmem->av_startingWorker;
  
             /*
              * We can't launch another worker when another one is still
              * starting up (or failed while doing so), so just sleep for a bit
              * more; that worker will wake us up again as soon as it's ready.
              * We will only wait autovacuum_naptime seconds (up to a maximum
              * of 60 seconds) for this to happen however.  Note that failure
              * to connect to a particular database is not a problem here,
              * because the worker removes itself from the startingWorker
              * pointer before trying to connect.  Problems detected by the
              * postmaster (like fork() failure) are also reported and handled
              * differently.  The only problems that may cause this code to
              * fire are errors in the earlier sections of AutoVacWorkerMain,
              * before the worker removes the WorkerInfo from the
              * startingWorker pointer.
              */
             waittime = Min(autovacuum_naptime, 60) * 1000;
             if (TimestampDifferenceExceeds(worker->wi_launchtime, current_time,
                                            waittime))
             {
                 LWLockRelease(AutovacuumLock);
                 LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
  
                 /*
                  * No other process can put a worker in starting mode, so if
                  * startingWorker is still INVALID after exchanging our lock,
                  * we assume it's the same one we saw above (so we don't
                  * recheck the launch time).
                  */
                 if (AutoVacuumShmem->av_startingWorker != NULL)
                 {
                     worker = AutoVacuumShmem->av_startingWorker;
                     worker->wi_dboid = InvalidOid;
                     worker->wi_tableoid = InvalidOid;
                     worker->wi_sharedrel = false;
                     worker->wi_proc = NULL;
                     worker->wi_launchtime = 0;
                     dlist_push_head(&AutoVacuumShmem->av_freeWorkers,
                                     &worker->wi_links);
                     AutoVacuumShmem->av_startingWorker = NULL;
                     ereport(WARNING,
                             errmsg("autovacuum worker took too long to start; canceled"));
                 }
             }
             else
                 can_launch = false;
         }
         LWLockRelease(AutovacuumLock);  /* either shared or exclusive */
  
         /* if we can't do anything, just go back to sleep */
         if (!can_launch)
             continue;
  
         /* We're OK to start a new worker */
  
         if (dlist_is_empty(&DatabaseList))
         {
             /*
              * Special case when the list is empty: start a worker right away.
              * This covers the initial case, when no database is in pgstats
              * (thus the list is empty).  Note that the constraints in
              * launcher_determine_sleep keep us from starting workers too
              * quickly (at most once every autovacuum_naptime when the list is
              * empty).
              */
             launch_worker(current_time);
         }
         else
         {
             /*
              * because rebuild_database_list constructs a list with most
              * distant adl_next_worker first, we obtain our database from the
              * tail of the list.
              */
             avl_dbase  *avdb;
  
             avdb = dlist_tail_element(avl_dbase, adl_node, &DatabaseList);
  
             /*
              * launch a worker if next_worker is right now or it is in the
              * past
              */
             if (TimestampDifferenceExceeds(avdb->adl_next_worker,
                                            current_time, 0))
                 launch_worker(current_time);
         }
     }
  
     AutoVacLauncherShutdown();
 }

Referenced by HandleAutoVacLauncherInterrupts().

◆ autovacuum_do_vac_analyze()

static void autovacuum_do_vac_analyze	(	autovac_table *	tab,
		BufferAccessStrategy	bstrategy
	)

static

Definition at line 3078 of file autovacuum.c.

 {
     RangeVar   *rangevar;
     VacuumRelation *rel;
     List       *rel_list;
     MemoryContext vac_context;
  
     /* Let pgstat know what we're doing */
     autovac_report_activity(tab);
  
     /* Set up one VacuumRelation target, identified by OID, for vacuum() */
     rangevar = makeRangeVar(tab->at_nspname, tab->at_relname, -1);
     rel = makeVacuumRelation(rangevar, tab->at_relid, NIL);
     rel_list = list_make1(rel);
  
     vac_context = AllocSetContextCreate(CurrentMemoryContext,
                                         "Vacuum",
                                         ALLOCSET_DEFAULT_SIZES);
  
     vacuum(rel_list, &tab->at_params, bstrategy, vac_context, true);
  
     MemoryContextDelete(vac_context);
 }

References ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate, autovac_table::at_nspname, autovac_table::at_params, autovac_table::at_relid, autovac_table::at_relname, autovac_report_activity(), CurrentMemoryContext, list_make1, makeRangeVar(), makeVacuumRelation(), MemoryContextDelete(), NIL, and vacuum().

Referenced by do_autovacuum().

◆ AutoVacuumingActive()

bool AutoVacuumingActive ( void )

Definition at line 3189 of file autovacuum.c.

 {
     if (!autovacuum_start_daemon || !pgstat_track_counts)
         return false;
     return true;
 }

References autovacuum_start_daemon, and pgstat_track_counts.

Referenced by HandleAutoVacLauncherInterrupts(), process_pm_child_exit(), relation_needs_vacanalyze(), and ServerLoop().

◆ AutoVacuumRequestWork()

bool AutoVacuumRequestWork	(	AutoVacuumWorkItemType	type,
		Oid	relationId,
		BlockNumber	blkno
	)

Definition at line 3201 of file autovacuum.c.

 {
     int         i;
     bool        result = false;
  
     LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
  
     /*
      * Locate an unused work item and fill it with the given data.
      */
     for (i = 0; i < NUM_WORKITEMS; i++)
     {
         AutoVacuumWorkItem *workitem = &AutoVacuumShmem->av_workItems[i];
  
         if (workitem->avw_used)
             continue;
  
         workitem->avw_used = true;
         workitem->avw_active = false;
         workitem->avw_type = type;
         workitem->avw_database = MyDatabaseId;
         workitem->avw_relation = relationId;
         workitem->avw_blockNumber = blkno;
         result = true;
  
         /* done */
         break;
     }
  
     LWLockRelease(AutovacuumLock);
  
     return result;
 }

References AutoVacuumShmem, AutoVacuumShmemStruct::av_workItems, AutoVacuumWorkItem::avw_active, AutoVacuumWorkItem::avw_blockNumber, AutoVacuumWorkItem::avw_database, AutoVacuumWorkItem::avw_relation, AutoVacuumWorkItem::avw_type, AutoVacuumWorkItem::avw_used, i, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MyDatabaseId, NUM_WORKITEMS, and type.

Referenced by brininsert().

◆ AutoVacuumShmemInit()

void AutoVacuumShmemInit ( void )

Definition at line 3275 of file autovacuum.c.

 {
     bool        found;
  
     AutoVacuumShmem = (AutoVacuumShmemStruct *)
         ShmemInitStruct("AutoVacuum Data",
                         AutoVacuumShmemSize(),
                         &found);
  
     if (!IsUnderPostmaster)
     {
         WorkerInfo  worker;
         int         i;
  
         Assert(!found);
  
         AutoVacuumShmem->av_launcherpid = 0;
         dlist_init(&AutoVacuumShmem->av_freeWorkers);
         dlist_init(&AutoVacuumShmem->av_runningWorkers);
         AutoVacuumShmem->av_startingWorker = NULL;
         memset(AutoVacuumShmem->av_workItems, 0,
                sizeof(AutoVacuumWorkItem) * NUM_WORKITEMS);
  
         worker = (WorkerInfo) ((char *) AutoVacuumShmem +
                                MAXALIGN(sizeof(AutoVacuumShmemStruct)));
  
         /* initialize the WorkerInfo free list */
         for (i = 0; i < autovacuum_max_workers; i++)
         {
             dlist_push_head(&AutoVacuumShmem->av_freeWorkers,
                             &worker[i].wi_links);
             pg_atomic_init_flag(&worker[i].wi_dobalance);
         }
  
         pg_atomic_init_u32(&AutoVacuumShmem->av_nworkersForBalance, 0);
  
     }
     else
         Assert(found);
 }

References Assert, autovacuum_max_workers, AutoVacuumShmem, AutoVacuumShmemSize(), AutoVacuumShmemStruct::av_freeWorkers, AutoVacuumShmemStruct::av_launcherpid, AutoVacuumShmemStruct::av_nworkersForBalance, AutoVacuumShmemStruct::av_runningWorkers, AutoVacuumShmemStruct::av_startingWorker, AutoVacuumShmemStruct::av_workItems, dlist_init(), dlist_push_head(), i, IsUnderPostmaster, MAXALIGN, NUM_WORKITEMS, pg_atomic_init_flag(), pg_atomic_init_u32(), ShmemInitStruct(), and WorkerInfoData::wi_links.

Referenced by CreateOrAttachShmemStructs().

◆ AutoVacuumShmemSize()

Size AutoVacuumShmemSize ( void )

Definition at line 3256 of file autovacuum.c.

 {
     Size        size;
  
     /*
      * Need the fixed struct and the array of WorkerInfoData.
      */
     size = sizeof(AutoVacuumShmemStruct);
     size = MAXALIGN(size);
     size = add_size(size, mul_size(autovacuum_max_workers,
                                    sizeof(WorkerInfoData)));
     return size;
 }

References add_size(), autovacuum_max_workers, MAXALIGN, mul_size(), and size.

Referenced by AutoVacuumShmemInit(), and CalculateShmemSize().

◆ AutoVacuumUpdateCostLimit()

void AutoVacuumUpdateCostLimit ( void )

Definition at line 1702 of file autovacuum.c.

 {
     if (!MyWorkerInfo)
         return;
  
     /*
      * note: in cost_limit, zero also means use value from elsewhere, because
      * zero is not a valid value.
      */
  
     if (av_storage_param_cost_limit > 0)
         vacuum_cost_limit = av_storage_param_cost_limit;
     else
     {
         int         nworkers_for_balance;
  
         if (autovacuum_vac_cost_limit > 0)
             vacuum_cost_limit = autovacuum_vac_cost_limit;
         else
             vacuum_cost_limit = VacuumCostLimit;
  
         /* Only balance limit if no cost-related storage parameters specified */
         if (pg_atomic_unlocked_test_flag(&MyWorkerInfo->wi_dobalance))
             return;
  
         Assert(vacuum_cost_limit > 0);
  
         nworkers_for_balance = pg_atomic_read_u32(&AutoVacuumShmem->av_nworkersForBalance);
  
         /* There is at least 1 autovac worker (this worker) */
         if (nworkers_for_balance <= 0)
             elog(ERROR, "nworkers_for_balance must be > 0");
  
         vacuum_cost_limit = Max(vacuum_cost_limit / nworkers_for_balance, 1);
     }
 }

References Assert, autovacuum_vac_cost_limit, AutoVacuumShmem, AutoVacuumShmemStruct::av_nworkersForBalance, av_storage_param_cost_limit, elog, ERROR, Max, MyWorkerInfo, pg_atomic_read_u32(), pg_atomic_unlocked_test_flag(), vacuum_cost_limit, VacuumCostLimit, and WorkerInfoData::wi_dobalance.

Referenced by vacuum_delay_point(), and VacuumUpdateCosts().

◆ AutoVacWorkerFailed()

void AutoVacWorkerFailed ( void )

Definition at line 1337 of file autovacuum.c.

 {
     AutoVacuumShmem->av_signal[AutoVacForkFailed] = true;
 }

References AutoVacForkFailed, AutoVacuumShmem, and AutoVacuumShmemStruct::av_signal.

Referenced by StartAutovacuumWorker().

◆ AutoVacWorkerMain()

void AutoVacWorkerMain	(	char *	startup_data,
		size_t	startup_data_len
	)

Definition at line 1359 of file autovacuum.c.

 {
     sigjmp_buf  local_sigjmp_buf;
     Oid         dbid;
  
     Assert(startup_data_len == 0);
  
     /* Release postmaster's working memory context */
     if (PostmasterContext)
     {
         MemoryContextDelete(PostmasterContext);
         PostmasterContext = NULL;
     }
  
     MyBackendType = B_AUTOVAC_WORKER;
     init_ps_display(NULL);
  
     SetProcessingMode(InitProcessing);
  
     /*
      * Set up signal handlers.  We operate on databases much like a regular
      * backend, so we use the same signal handling.  See equivalent code in
      * tcop/postgres.c.
      */
     pqsignal(SIGHUP, SignalHandlerForConfigReload);
  
     /*
      * SIGINT is used to signal canceling the current table's vacuum; SIGTERM
      * means abort and exit cleanly, and SIGQUIT means abandon ship.
      */
     pqsignal(SIGINT, StatementCancelHandler);
     pqsignal(SIGTERM, die);
     /* SIGQUIT handler was already set up by InitPostmasterChild */
  
     InitializeTimeouts();       /* establishes SIGALRM handler */
  
     pqsignal(SIGPIPE, SIG_IGN);
     pqsignal(SIGUSR1, procsignal_sigusr1_handler);
     pqsignal(SIGUSR2, SIG_IGN);
     pqsignal(SIGFPE, FloatExceptionHandler);
     pqsignal(SIGCHLD, SIG_DFL);
  
     /*
      * Create a per-backend PGPROC struct in shared memory.  We must do this
      * before we can use LWLocks or access any shared memory.
      */
     InitProcess();
  
     /* Early initialization */
     BaseInit();
  
     /*
      * If an exception is encountered, processing resumes here.
      *
      * Unlike most auxiliary processes, we don't attempt to continue
      * processing after an error; we just clean up and exit.  The autovac
      * launcher is responsible for spawning another worker later.
      *
      * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
      * (to wit, BlockSig) will be restored when longjmp'ing to here.  Thus,
      * signals other than SIGQUIT will be blocked until we exit.  It might
      * seem that this policy makes the HOLD_INTERRUPTS() call redundant, but
      * it is not since InterruptPending might be set already.
      */
     if (sigsetjmp(local_sigjmp_buf, 1) != 0)
     {
         /* since not using PG_TRY, must reset error stack by hand */
         error_context_stack = NULL;
  
         /* Prevents interrupts while cleaning up */
         HOLD_INTERRUPTS();
  
         /* Report the error to the server log */
         EmitErrorReport();
  
         /*
          * We can now go away.  Note that because we called InitProcess, a
          * callback was registered to do ProcKill, which will clean up
          * necessary state.
          */
         proc_exit(0);
     }
  
     /* We can now handle ereport(ERROR) */
     PG_exception_stack = &local_sigjmp_buf;
  
     sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
  
     /*
      * Set always-secure search path, so malicious users can't redirect user
      * code (e.g. pg_index.indexprs).  (That code runs in a
      * SECURITY_RESTRICTED_OPERATION sandbox, so malicious users could not
      * take control of the entire autovacuum worker in any case.)
      */
     SetConfigOption("search_path", "", PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force zero_damaged_pages OFF in the autovac process, even if it is set
      * in postgresql.conf.  We don't really want such a dangerous option being
      * applied non-interactively.
      */
     SetConfigOption("zero_damaged_pages", "false", PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force settable timeouts off to avoid letting these settings prevent
      * regular maintenance from being executed.
      */
     SetConfigOption("statement_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     SetConfigOption("transaction_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     SetConfigOption("lock_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     SetConfigOption("idle_in_transaction_session_timeout", "0",
                     PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force default_transaction_isolation to READ COMMITTED.  We don't want
      * to pay the overhead of serializable mode, nor add any risk of causing
      * deadlocks or delaying other transactions.
      */
     SetConfigOption("default_transaction_isolation", "read committed",
                     PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Force synchronous replication off to allow regular maintenance even if
      * we are waiting for standbys to connect. This is important to ensure we
      * aren't blocked from performing anti-wraparound tasks.
      */
     if (synchronous_commit > SYNCHRONOUS_COMMIT_LOCAL_FLUSH)
         SetConfigOption("synchronous_commit", "local",
                         PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Even when system is configured to use a different fetch consistency,
      * for autovac we always want fresh stats.
      */
     SetConfigOption("stats_fetch_consistency", "none", PGC_SUSET, PGC_S_OVERRIDE);
  
     /*
      * Get the info about the database we're going to work on.
      */
     LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
  
     /*
      * beware of startingWorker being INVALID; this should normally not
      * happen, but if a worker fails after forking and before this, the
      * launcher might have decided to remove it from the queue and start
      * again.
      */
     if (AutoVacuumShmem->av_startingWorker != NULL)
     {
         MyWorkerInfo = AutoVacuumShmem->av_startingWorker;
         dbid = MyWorkerInfo->wi_dboid;
         MyWorkerInfo->wi_proc = MyProc;
  
         /* insert into the running list */
         dlist_push_head(&AutoVacuumShmem->av_runningWorkers,
                         &MyWorkerInfo->wi_links);
  
         /*
          * remove from the "starting" pointer, so that the launcher can start
          * a new worker if required
          */
         AutoVacuumShmem->av_startingWorker = NULL;
         LWLockRelease(AutovacuumLock);
  
         on_shmem_exit(FreeWorkerInfo, 0);
  
         /* wake up the launcher */
         if (AutoVacuumShmem->av_launcherpid != 0)
             kill(AutoVacuumShmem->av_launcherpid, SIGUSR2);
     }
     else
     {
         /* no worker entry for me, go away */
         elog(WARNING, "autovacuum worker started without a worker entry");
         dbid = InvalidOid;
         LWLockRelease(AutovacuumLock);
     }
  
     if (OidIsValid(dbid))
     {
         char        dbname[NAMEDATALEN];
  
         /*
          * Report autovac startup to the cumulative stats system.  We
          * deliberately do this before InitPostgres, so that the
          * last_autovac_time will get updated even if the connection attempt
          * fails.  This is to prevent autovac from getting "stuck" repeatedly
          * selecting an unopenable database, rather than making any progress
          * on stuff it can connect to.
          */
         pgstat_report_autovac(dbid);
  
         /*
          * Connect to the selected database, specifying no particular user
          *
          * Note: if we have selected a just-deleted database (due to using
          * stale stats info), we'll fail and exit here.
          */
         InitPostgres(NULL, dbid, NULL, InvalidOid, 0, dbname);
         SetProcessingMode(NormalProcessing);
         set_ps_display(dbname);
         ereport(DEBUG1,
                 (errmsg_internal("autovacuum: processing database \"%s\"", dbname)));
  
         if (PostAuthDelay)
             pg_usleep(PostAuthDelay * 1000000L);
  
         /* And do an appropriate amount of work */
         recentXid = ReadNextTransactionId();
         recentMulti = ReadNextMultiXactId();
         do_autovacuum();
     }
  
     /*
      * The launcher will be notified of my death in ProcKill, *if* we managed
      * to get a worker slot at all
      */
  
     /* All done, go away */
     proc_exit(0);
 }

References Assert, AutoVacuumShmem, AutoVacuumShmemStruct::av_launcherpid, AutoVacuumShmemStruct::av_runningWorkers, AutoVacuumShmemStruct::av_startingWorker, B_AUTOVAC_WORKER, BaseInit(), dbname, DEBUG1, die, dlist_push_head(), do_autovacuum(), elog, EmitErrorReport(), ereport, errmsg_internal(), error_context_stack, FloatExceptionHandler(), FreeWorkerInfo(), HOLD_INTERRUPTS, init_ps_display(), InitializeTimeouts(), InitPostgres(), InitProcess(), InitProcessing, InvalidOid, kill, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MemoryContextDelete(), MyBackendType, MyProc, MyWorkerInfo, NAMEDATALEN, NormalProcessing, OidIsValid, on_shmem_exit(), PG_exception_stack, pg_usleep(), PGC_S_OVERRIDE, PGC_SUSET, pgstat_report_autovac(), PostAuthDelay, PostmasterContext, pqsignal(), proc_exit(), procsignal_sigusr1_handler(), ReadNextMultiXactId(), ReadNextTransactionId(), recentMulti, recentXid, set_ps_display(), SetConfigOption(), SetProcessingMode, SIG_DFL, SIG_IGN, SIGCHLD, SIGHUP, SignalHandlerForConfigReload(), SIGPIPE, SIGUSR1, SIGUSR2, StatementCancelHandler(), synchronous_commit, SYNCHRONOUS_COMMIT_LOCAL_FLUSH, UnBlockSig, WARNING, WorkerInfoData::wi_dboid, WorkerInfoData::wi_links, and WorkerInfoData::wi_proc.

◆ avl_sigusr2_handler()

static void avl_sigusr2_handler ( SIGNAL_ARGS )

static

Definition at line 1344 of file autovacuum.c.

 {
     got_SIGUSR2 = true;
     SetLatch(MyLatch);
 }

References got_SIGUSR2, MyLatch, and SetLatch().

◆ check_autovacuum_work_mem()

bool check_autovacuum_work_mem	(	int *	newval,
		void **	extra,
		GucSource	source
	)

Definition at line 3320 of file autovacuum.c.

 {
     /*
      * -1 indicates fallback.
      *
      * If we haven't yet changed the boot_val default of -1, just let it be.
      * Autovacuum will look to maintenance_work_mem instead.
      */
     if (*newval == -1)
         return true;
  
     /*
      * We clamp manually-set values to at least 1MB.  Since
      * maintenance_work_mem is always set to at least this value, do the same
      * here.
      */
     if (*newval < 1024)
         *newval = 1024;
  
     return true;
 }

References newval.

◆ db_comparator()

static int db_comparator	(	const void *	a,
		const void *	b
	)

static

Definition at line 1055 of file autovacuum.c.

 {
     return pg_cmp_s32(((const avl_dbase *) a)->adl_score,
                       ((const avl_dbase *) b)->adl_score);
 }

References a, b, and pg_cmp_s32().

Referenced by rebuild_database_list().

◆ do_autovacuum()

static void do_autovacuum ( void )

static

Definition at line 1873 of file autovacuum.c.

 {
     Relation    classRel;
     HeapTuple   tuple;
     TableScanDesc relScan;
     Form_pg_database dbForm;
     List       *table_oids = NIL;
     List       *orphan_oids = NIL;
     HASHCTL     ctl;
     HTAB       *table_toast_map;
     ListCell   *volatile cell;
     BufferAccessStrategy bstrategy;
     ScanKeyData key;
     TupleDesc   pg_class_desc;
     int         effective_multixact_freeze_max_age;
     bool        did_vacuum = false;
     bool        found_concurrent_worker = false;
     int         i;
  
     /*
      * StartTransactionCommand and CommitTransactionCommand will automatically
      * switch to other contexts.  We need this one to keep the list of
      * relations to vacuum/analyze across transactions.
      */
     AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
                                           "Autovacuum worker",
                                           ALLOCSET_DEFAULT_SIZES);
     MemoryContextSwitchTo(AutovacMemCxt);
  
     /* Start a transaction so our commands have one to play into. */
     StartTransactionCommand();
  
     /*
      * Compute the multixact age for which freezing is urgent.  This is
      * normally autovacuum_multixact_freeze_max_age, but may be less if we are
      * short of multixact member space.
      */
     effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
  
     /*
      * Find the pg_database entry and select the default freeze ages. We use
      * zero in template and nonconnectable databases, else the system-wide
      * default.
      */
     tuple = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(MyDatabaseId));
     if (!HeapTupleIsValid(tuple))
         elog(ERROR, "cache lookup failed for database %u", MyDatabaseId);
     dbForm = (Form_pg_database) GETSTRUCT(tuple);
  
     if (dbForm->datistemplate || !dbForm->datallowconn)
     {
         default_freeze_min_age = 0;
         default_freeze_table_age = 0;
         default_multixact_freeze_min_age = 0;
         default_multixact_freeze_table_age = 0;
     }
     else
     {
         default_freeze_min_age = vacuum_freeze_min_age;
         default_freeze_table_age = vacuum_freeze_table_age;
         default_multixact_freeze_min_age = vacuum_multixact_freeze_min_age;
         default_multixact_freeze_table_age = vacuum_multixact_freeze_table_age;
     }
  
     ReleaseSysCache(tuple);
  
     /* StartTransactionCommand changed elsewhere */
     MemoryContextSwitchTo(AutovacMemCxt);
  
     classRel = table_open(RelationRelationId, AccessShareLock);
  
     /* create a copy so we can use it after closing pg_class */
     pg_class_desc = CreateTupleDescCopy(RelationGetDescr(classRel));
  
     /* create hash table for toast <-> main relid mapping */
     ctl.keysize = sizeof(Oid);
     ctl.entrysize = sizeof(av_relation);
  
     table_toast_map = hash_create("TOAST to main relid map",
                                   100,
                                   &ctl,
                                   HASH_ELEM | HASH_BLOBS);
  
     /*
      * Scan pg_class to determine which tables to vacuum.
      *
      * We do this in two passes: on the first one we collect the list of plain
      * relations and materialized views, and on the second one we collect
      * TOAST tables. The reason for doing the second pass is that during it we
      * want to use the main relation's pg_class.reloptions entry if the TOAST
      * table does not have any, and we cannot obtain it unless we know
      * beforehand what's the main table OID.
      *
      * We need to check TOAST tables separately because in cases with short,
      * wide tables there might be proportionally much more activity in the
      * TOAST table than in its parent.
      */
     relScan = table_beginscan_catalog(classRel, 0, NULL);
  
     /*
      * On the first pass, we collect main tables to vacuum, and also the main
      * table relid to TOAST relid mapping.
      */
     while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
     {
         Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
         PgStat_StatTabEntry *tabentry;
         AutoVacOpts *relopts;
         Oid         relid;
         bool        dovacuum;
         bool        doanalyze;
         bool        wraparound;
  
         if (classForm->relkind != RELKIND_RELATION &&
             classForm->relkind != RELKIND_MATVIEW)
             continue;
  
         relid = classForm->oid;
  
         /*
          * Check if it is a temp table (presumably, of some other backend's).
          * We cannot safely process other backends' temp tables.
          */
         if (classForm->relpersistence == RELPERSISTENCE_TEMP)
         {
             /*
              * We just ignore it if the owning backend is still active and
              * using the temporary schema.  Also, for safety, ignore it if the
              * namespace doesn't exist or isn't a temp namespace after all.
              */
             if (checkTempNamespaceStatus(classForm->relnamespace) == TEMP_NAMESPACE_IDLE)
             {
                 /*
                  * The table seems to be orphaned -- although it might be that
                  * the owning backend has already deleted it and exited; our
                  * pg_class scan snapshot is not necessarily up-to-date
                  * anymore, so we could be looking at a committed-dead entry.
                  * Remember it so we can try to delete it later.
                  */
                 orphan_oids = lappend_oid(orphan_oids, relid);
             }
             continue;
         }
  
         /* Fetch reloptions and the pgstat entry for this table */
         relopts = extract_autovac_opts(tuple, pg_class_desc);
         tabentry = pgstat_fetch_stat_tabentry_ext(classForm->relisshared,
                                                   relid);
  
         /* Check if it needs vacuum or analyze */
         relation_needs_vacanalyze(relid, relopts, classForm, tabentry,
                                   effective_multixact_freeze_max_age,
                                   &dovacuum, &doanalyze, &wraparound);
  
         /* Relations that need work are added to table_oids */
         if (dovacuum || doanalyze)
             table_oids = lappend_oid(table_oids, relid);
  
         /*
          * Remember TOAST associations for the second pass.  Note: we must do
          * this whether or not the table is going to be vacuumed, because we
          * don't automatically vacuum toast tables along the parent table.
          */
         if (OidIsValid(classForm->reltoastrelid))
         {
             av_relation *hentry;
             bool        found;
  
             hentry = hash_search(table_toast_map,
                                  &classForm->reltoastrelid,
                                  HASH_ENTER, &found);
  
             if (!found)
             {
                 /* hash_search already filled in the key */
                 hentry->ar_relid = relid;
                 hentry->ar_hasrelopts = false;
                 if (relopts != NULL)
                 {
                     hentry->ar_hasrelopts = true;
                     memcpy(&hentry->ar_reloptions, relopts,
                            sizeof(AutoVacOpts));
                 }
             }
         }
     }
  
     table_endscan(relScan);
  
     /* second pass: check TOAST tables */
     ScanKeyInit(&key,
                 Anum_pg_class_relkind,
                 BTEqualStrategyNumber, F_CHAREQ,
                 CharGetDatum(RELKIND_TOASTVALUE));
  
     relScan = table_beginscan_catalog(classRel, 1, &key);
     while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
     {
         Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
         PgStat_StatTabEntry *tabentry;
         Oid         relid;
         AutoVacOpts *relopts = NULL;
         bool        dovacuum;
         bool        doanalyze;
         bool        wraparound;
  
         /*
          * We cannot safely process other backends' temp tables, so skip 'em.
          */
         if (classForm->relpersistence == RELPERSISTENCE_TEMP)
             continue;
  
         relid = classForm->oid;
  
         /*
          * fetch reloptions -- if this toast table does not have them, try the
          * main rel
          */
         relopts = extract_autovac_opts(tuple, pg_class_desc);
         if (relopts == NULL)
         {
             av_relation *hentry;
             bool        found;
  
             hentry = hash_search(table_toast_map, &relid, HASH_FIND, &found);
             if (found && hentry->ar_hasrelopts)
                 relopts = &hentry->ar_reloptions;
         }
  
         /* Fetch the pgstat entry for this table */
         tabentry = pgstat_fetch_stat_tabentry_ext(classForm->relisshared,
                                                   relid);
  
         relation_needs_vacanalyze(relid, relopts, classForm, tabentry,
                                   effective_multixact_freeze_max_age,
                                   &dovacuum, &doanalyze, &wraparound);
  
         /* ignore analyze for toast tables */
         if (dovacuum)
             table_oids = lappend_oid(table_oids, relid);
     }
  
     table_endscan(relScan);
     table_close(classRel, AccessShareLock);
  
     /*
      * Recheck orphan temporary tables, and if they still seem orphaned, drop
      * them.  We'll eat a transaction per dropped table, which might seem
      * excessive, but we should only need to do anything as a result of a
      * previous backend crash, so this should not happen often enough to
      * justify "optimizing".  Using separate transactions ensures that we
      * don't bloat the lock table if there are many temp tables to be dropped,
      * and it ensures that we don't lose work if a deletion attempt fails.
      */
     foreach(cell, orphan_oids)
     {
         Oid         relid = lfirst_oid(cell);
         Form_pg_class classForm;
         ObjectAddress object;
  
         /*
          * Check for user-requested abort.
          */
         CHECK_FOR_INTERRUPTS();
  
         /*
          * Try to lock the table.  If we can't get the lock immediately,
          * somebody else is using (or dropping) the table, so it's not our
          * concern anymore.  Having the lock prevents race conditions below.
          */
         if (!ConditionalLockRelationOid(relid, AccessExclusiveLock))
             continue;
  
         /*
          * Re-fetch the pg_class tuple and re-check whether it still seems to
          * be an orphaned temp table.  If it's not there or no longer the same
          * relation, ignore it.
          */
         tuple = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
         if (!HeapTupleIsValid(tuple))
         {
             /* be sure to drop useless lock so we don't bloat lock table */
             UnlockRelationOid(relid, AccessExclusiveLock);
             continue;
         }
         classForm = (Form_pg_class) GETSTRUCT(tuple);
  
         /*
          * Make all the same tests made in the loop above.  In event of OID
          * counter wraparound, the pg_class entry we have now might be
          * completely unrelated to the one we saw before.
          */
         if (!((classForm->relkind == RELKIND_RELATION ||
                classForm->relkind == RELKIND_MATVIEW) &&
               classForm->relpersistence == RELPERSISTENCE_TEMP))
         {
             UnlockRelationOid(relid, AccessExclusiveLock);
             continue;
         }
  
         if (checkTempNamespaceStatus(classForm->relnamespace) != TEMP_NAMESPACE_IDLE)
         {
             UnlockRelationOid(relid, AccessExclusiveLock);
             continue;
         }
  
         /*
          * Try to lock the temp namespace, too.  Even though we have lock on
          * the table itself, there's a risk of deadlock against an incoming
          * backend trying to clean out the temp namespace, in case this table
          * has dependencies (such as sequences) that the backend's
          * performDeletion call might visit in a different order.  If we can
          * get AccessShareLock on the namespace, that's sufficient to ensure
          * we're not running concurrently with RemoveTempRelations.  If we
          * can't, back off and let RemoveTempRelations do its thing.
          */
         if (!ConditionalLockDatabaseObject(NamespaceRelationId,
                                            classForm->relnamespace, 0,
                                            AccessShareLock))
         {
             UnlockRelationOid(relid, AccessExclusiveLock);
             continue;
         }
  
         /* OK, let's delete it */
         ereport(LOG,
                 (errmsg("autovacuum: dropping orphan temp table \"%s.%s.%s\"",
                         get_database_name(MyDatabaseId),
                         get_namespace_name(classForm->relnamespace),
                         NameStr(classForm->relname))));
  
         object.classId = RelationRelationId;
         object.objectId = relid;
         object.objectSubId = 0;
         performDeletion(&object, DROP_CASCADE,
                         PERFORM_DELETION_INTERNAL |
                         PERFORM_DELETION_QUIETLY |
                         PERFORM_DELETION_SKIP_EXTENSIONS);
  
         /*
          * To commit the deletion, end current transaction and start a new
          * one.  Note this also releases the locks we took.
          */
         CommitTransactionCommand();
         StartTransactionCommand();
  
         /* StartTransactionCommand changed current memory context */
         MemoryContextSwitchTo(AutovacMemCxt);
     }
  
     /*
      * Optionally, create a buffer access strategy object for VACUUM to use.
      * We use the same BufferAccessStrategy object for all tables VACUUMed by
      * this worker to prevent autovacuum from blowing out shared buffers.
      *
      * VacuumBufferUsageLimit being set to 0 results in
      * GetAccessStrategyWithSize returning NULL, effectively meaning we can
      * use up to all of shared buffers.
      *
      * If we later enter failsafe mode on any of the tables being vacuumed, we
      * will cease use of the BufferAccessStrategy only for that table.
      *
      * XXX should we consider adding code to adjust the size of this if
      * VacuumBufferUsageLimit changes?
      */
     bstrategy = GetAccessStrategyWithSize(BAS_VACUUM, VacuumBufferUsageLimit);
  
     /*
      * create a memory context to act as fake PortalContext, so that the
      * contexts created in the vacuum code are cleaned up for each table.
      */
     PortalContext = AllocSetContextCreate(AutovacMemCxt,
                                           "Autovacuum Portal",
                                           ALLOCSET_DEFAULT_SIZES);
  
     /*
      * Perform operations on collected tables.
      */
     foreach(cell, table_oids)
     {
         Oid         relid = lfirst_oid(cell);
         HeapTuple   classTup;
         autovac_table *tab;
         bool        isshared;
         bool        skipit;
         dlist_iter  iter;
  
         CHECK_FOR_INTERRUPTS();
  
         /*
          * Check for config changes before processing each collected table.
          */
         if (ConfigReloadPending)
         {
             ConfigReloadPending = false;
             ProcessConfigFile(PGC_SIGHUP);
  
             /*
              * You might be tempted to bail out if we see autovacuum is now
              * disabled.  Must resist that temptation -- this might be a
              * for-wraparound emergency worker, in which case that would be
              * entirely inappropriate.
              */
         }
  
         /*
          * Find out whether the table is shared or not.  (It's slightly
          * annoying to fetch the syscache entry just for this, but in typical
          * cases it adds little cost because table_recheck_autovac would
          * refetch the entry anyway.  We could buy that back by copying the
          * tuple here and passing it to table_recheck_autovac, but that
          * increases the odds of that function working with stale data.)
          */
         classTup = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
         if (!HeapTupleIsValid(classTup))
             continue;           /* somebody deleted the rel, forget it */
         isshared = ((Form_pg_class) GETSTRUCT(classTup))->relisshared;
         ReleaseSysCache(classTup);
  
         /*
          * Hold schedule lock from here until we've claimed the table.  We
          * also need the AutovacuumLock to walk the worker array, but that one
          * can just be a shared lock.
          */
         LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
         LWLockAcquire(AutovacuumLock, LW_SHARED);
  
         /*
          * Check whether the table is being vacuumed concurrently by another
          * worker.
          */
         skipit = false;
         dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
         {
             WorkerInfo  worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
  
             /* ignore myself */
             if (worker == MyWorkerInfo)
                 continue;
  
             /* ignore workers in other databases (unless table is shared) */
             if (!worker->wi_sharedrel && worker->wi_dboid != MyDatabaseId)
                 continue;
  
             if (worker->wi_tableoid == relid)
             {
                 skipit = true;
                 found_concurrent_worker = true;
                 break;
             }
         }
         LWLockRelease(AutovacuumLock);
         if (skipit)
         {
             LWLockRelease(AutovacuumScheduleLock);
             continue;
         }
  
         /*
          * Store the table's OID in shared memory before releasing the
          * schedule lock, so that other workers don't try to vacuum it
          * concurrently.  (We claim it here so as not to hold
          * AutovacuumScheduleLock while rechecking the stats.)
          */
         MyWorkerInfo->wi_tableoid = relid;
         MyWorkerInfo->wi_sharedrel = isshared;
         LWLockRelease(AutovacuumScheduleLock);
  
         /*
          * Check whether pgstat data still says we need to vacuum this table.
          * It could have changed if something else processed the table while
          * we weren't looking. This doesn't entirely close the race condition,
          * but it is very small.
          */
         MemoryContextSwitchTo(AutovacMemCxt);
         tab = table_recheck_autovac(relid, table_toast_map, pg_class_desc,
                                     effective_multixact_freeze_max_age);
         if (tab == NULL)
         {
             /* someone else vacuumed the table, or it went away */
             LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
             MyWorkerInfo->wi_tableoid = InvalidOid;
             MyWorkerInfo->wi_sharedrel = false;
             LWLockRelease(AutovacuumScheduleLock);
             continue;
         }
  
         /*
          * Save the cost-related storage parameter values in global variables
          * for reference when updating vacuum_cost_delay and vacuum_cost_limit
          * during vacuuming this table.
          */
         av_storage_param_cost_delay = tab->at_storage_param_vac_cost_delay;
         av_storage_param_cost_limit = tab->at_storage_param_vac_cost_limit;
  
         /*
          * We only expect this worker to ever set the flag, so don't bother
          * checking the return value. We shouldn't have to retry.
          */
         if (tab->at_dobalance)
             pg_atomic_test_set_flag(&MyWorkerInfo->wi_dobalance);
         else
             pg_atomic_clear_flag(&MyWorkerInfo->wi_dobalance);
  
         LWLockAcquire(AutovacuumLock, LW_SHARED);
         autovac_recalculate_workers_for_balance();
         LWLockRelease(AutovacuumLock);
  
         /*
          * We wait until this point to update cost delay and cost limit
          * values, even though we reloaded the configuration file above, so
          * that we can take into account the cost-related storage parameters.
          */
         VacuumUpdateCosts();
  
  
         /* clean up memory before each iteration */
         MemoryContextReset(PortalContext);
  
         /*
          * Save the relation name for a possible error message, to avoid a
          * catalog lookup in case of an error.  If any of these return NULL,
          * then the relation has been dropped since last we checked; skip it.
          * Note: they must live in a long-lived memory context because we call
          * vacuum and analyze in different transactions.
          */
  
         tab->at_relname = get_rel_name(tab->at_relid);
         tab->at_nspname = get_namespace_name(get_rel_namespace(tab->at_relid));
         tab->at_datname = get_database_name(MyDatabaseId);
         if (!tab->at_relname || !tab->at_nspname || !tab->at_datname)
             goto deleted;
  
         /*
          * We will abort vacuuming the current table if something errors out,
          * and continue with the next one in schedule; in particular, this
          * happens if we are interrupted with SIGINT.
          */
         PG_TRY();
         {
             /* Use PortalContext for any per-table allocations */
             MemoryContextSwitchTo(PortalContext);
  
             /* have at it */
             autovacuum_do_vac_analyze(tab, bstrategy);
  
             /*
              * Clear a possible query-cancel signal, to avoid a late reaction
              * to an automatically-sent signal because of vacuuming the
              * current table (we're done with it, so it would make no sense to
              * cancel at this point.)
              */
             QueryCancelPending = false;
         }
         PG_CATCH();
         {
             /*
              * Abort the transaction, start a new one, and proceed with the
              * next table in our list.
              */
             HOLD_INTERRUPTS();
             if (tab->at_params.options & VACOPT_VACUUM)
                 errcontext("automatic vacuum of table \"%s.%s.%s\"",
                            tab->at_datname, tab->at_nspname, tab->at_relname);
             else
                 errcontext("automatic analyze of table \"%s.%s.%s\"",
                            tab->at_datname, tab->at_nspname, tab->at_relname);
             EmitErrorReport();
  
             /* this resets ProcGlobal->statusFlags[i] too */
             AbortOutOfAnyTransaction();
             FlushErrorState();
             MemoryContextReset(PortalContext);
  
             /* restart our transaction for the following operations */
             StartTransactionCommand();
             RESUME_INTERRUPTS();
         }
         PG_END_TRY();
  
         /* Make sure we're back in AutovacMemCxt */
         MemoryContextSwitchTo(AutovacMemCxt);
  
         did_vacuum = true;
  
         /* ProcGlobal->statusFlags[i] are reset at the next end of xact */
  
         /* be tidy */
 deleted:
         if (tab->at_datname != NULL)
             pfree(tab->at_datname);
         if (tab->at_nspname != NULL)
             pfree(tab->at_nspname);
         if (tab->at_relname != NULL)
             pfree(tab->at_relname);
         pfree(tab);
  
         /*
          * Remove my info from shared memory.  We set wi_dobalance on the
          * assumption that we are more likely than not to vacuum a table with
          * no cost-related storage parameters next, so we want to claim our
          * share of I/O as soon as possible to avoid thrashing the global
          * balance.
          */
         LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
         MyWorkerInfo->wi_tableoid = InvalidOid;
         MyWorkerInfo->wi_sharedrel = false;
         LWLockRelease(AutovacuumScheduleLock);
         pg_atomic_test_set_flag(&MyWorkerInfo->wi_dobalance);
     }
  
     /*
      * Perform additional work items, as requested by backends.
      */
     LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
     for (i = 0; i < NUM_WORKITEMS; i++)
     {
         AutoVacuumWorkItem *workitem = &AutoVacuumShmem->av_workItems[i];
  
         if (!workitem->avw_used)
             continue;
         if (workitem->avw_active)
             continue;
         if (workitem->avw_database != MyDatabaseId)
             continue;
  
         /* claim this one, and release lock while performing it */
         workitem->avw_active = true;
         LWLockRelease(AutovacuumLock);
  
         perform_work_item(workitem);
  
         /*
          * Check for config changes before acquiring lock for further jobs.
          */
         CHECK_FOR_INTERRUPTS();
         if (ConfigReloadPending)
         {
             ConfigReloadPending = false;
             ProcessConfigFile(PGC_SIGHUP);
             VacuumUpdateCosts();
         }
  
         LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
  
         /* and mark it done */
         workitem->avw_active = false;
         workitem->avw_used = false;
     }
     LWLockRelease(AutovacuumLock);
  
     /*
      * We leak table_toast_map here (among other things), but since we're
      * going away soon, it's not a problem.
      */
  
     /*
      * Update pg_database.datfrozenxid, and truncate pg_xact if possible. We
      * only need to do this once, not after each table.
      *
      * Even if we didn't vacuum anything, it may still be important to do
      * this, because one indirect effect of vac_update_datfrozenxid() is to
      * update TransamVariables->xidVacLimit.  That might need to be done even
      * if we haven't vacuumed anything, because relations with older
      * relfrozenxid values or other databases with older datfrozenxid values
      * might have been dropped, allowing xidVacLimit to advance.
      *
      * However, it's also important not to do this blindly in all cases,
      * because when autovacuum=off this will restart the autovacuum launcher.
      * If we're not careful, an infinite loop can result, where workers find
      * no work to do and restart the launcher, which starts another worker in
      * the same database that finds no work to do.  To prevent that, we skip
      * this if (1) we found no work to do and (2) we skipped at least one
      * table due to concurrent autovacuum activity.  In that case, the other
      * worker has already done it, or will do so when it finishes.
      */
     if (did_vacuum || !found_concurrent_worker)
         vac_update_datfrozenxid();
  
     /* Finally close out the last transaction. */
     CommitTransactionCommand();
 }

Referenced by AutoVacWorkerMain().

◆ do_start_worker()

static Oid do_start_worker ( void )

static

Definition at line 1073 of file autovacuum.c.

 {
     List       *dblist;
     ListCell   *cell;
     TransactionId xidForceLimit;
     MultiXactId multiForceLimit;
     bool        for_xid_wrap;
     bool        for_multi_wrap;
     avw_dbase  *avdb;
     TimestampTz current_time;
     bool        skipit = false;
     Oid         retval = InvalidOid;
     MemoryContext tmpcxt,
                 oldcxt;
  
     /* return quickly when there are no free workers */
     LWLockAcquire(AutovacuumLock, LW_SHARED);
     if (dlist_is_empty(&AutoVacuumShmem->av_freeWorkers))
     {
         LWLockRelease(AutovacuumLock);
         return InvalidOid;
     }
     LWLockRelease(AutovacuumLock);
  
     /*
      * Create and switch to a temporary context to avoid leaking the memory
      * allocated for the database list.
      */
     tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
                                    "Autovacuum start worker (tmp)",
                                    ALLOCSET_DEFAULT_SIZES);
     oldcxt = MemoryContextSwitchTo(tmpcxt);
  
     /* Get a list of databases */
     dblist = get_database_list();
  
     /*
      * Determine the oldest datfrozenxid/relfrozenxid that we will allow to
      * pass without forcing a vacuum.  (This limit can be tightened for
      * particular tables, but not loosened.)
      */
     recentXid = ReadNextTransactionId();
     xidForceLimit = recentXid - autovacuum_freeze_max_age;
     /* ensure it's a "normal" XID, else TransactionIdPrecedes misbehaves */
     /* this can cause the limit to go backwards by 3, but that's OK */
     if (xidForceLimit < FirstNormalTransactionId)
         xidForceLimit -= FirstNormalTransactionId;
  
     /* Also determine the oldest datminmxid we will consider. */
     recentMulti = ReadNextMultiXactId();
     multiForceLimit = recentMulti - MultiXactMemberFreezeThreshold();
     if (multiForceLimit < FirstMultiXactId)
         multiForceLimit -= FirstMultiXactId;
  
     /*
      * Choose a database to connect to.  We pick the database that was least
      * recently auto-vacuumed, or one that needs vacuuming to prevent Xid
      * wraparound-related data loss.  If any db at risk of Xid wraparound is
      * found, we pick the one with oldest datfrozenxid, independently of
      * autovacuum times; similarly we pick the one with the oldest datminmxid
      * if any is in MultiXactId wraparound.  Note that those in Xid wraparound
      * danger are given more priority than those in multi wraparound danger.
      *
      * Note that a database with no stats entry is not considered, except for
      * Xid wraparound purposes.  The theory is that if no one has ever
      * connected to it since the stats were last initialized, it doesn't need
      * vacuuming.
      *
      * XXX This could be improved if we had more info about whether it needs
      * vacuuming before connecting to it.  Perhaps look through the pgstats
      * data for the database's tables?  One idea is to keep track of the
      * number of new and dead tuples per database in pgstats.  However it
      * isn't clear how to construct a metric that measures that and not cause
      * starvation for less busy databases.
      */
     avdb = NULL;
     for_xid_wrap = false;
     for_multi_wrap = false;
     current_time = GetCurrentTimestamp();
     foreach(cell, dblist)
     {
         avw_dbase  *tmp = lfirst(cell);
         dlist_iter  iter;
  
         /* Check to see if this one is at risk of wraparound */
         if (TransactionIdPrecedes(tmp->adw_frozenxid, xidForceLimit))
         {
             if (avdb == NULL ||
                 TransactionIdPrecedes(tmp->adw_frozenxid,
                                       avdb->adw_frozenxid))
                 avdb = tmp;
             for_xid_wrap = true;
             continue;
         }
         else if (for_xid_wrap)
             continue;           /* ignore not-at-risk DBs */
         else if (MultiXactIdPrecedes(tmp->adw_minmulti, multiForceLimit))
         {
             if (avdb == NULL ||
                 MultiXactIdPrecedes(tmp->adw_minmulti, avdb->adw_minmulti))
                 avdb = tmp;
             for_multi_wrap = true;
             continue;
         }
         else if (for_multi_wrap)
             continue;           /* ignore not-at-risk DBs */
  
         /* Find pgstat entry if any */
         tmp->adw_entry = pgstat_fetch_stat_dbentry(tmp->adw_datid);
  
         /*
          * Skip a database with no pgstat entry; it means it hasn't seen any
          * activity.
          */
         if (!tmp->adw_entry)
             continue;
  
         /*
          * Also, skip a database that appears on the database list as having
          * been processed recently (less than autovacuum_naptime seconds ago).
          * We do this so that we don't select a database which we just
          * selected, but that pgstat hasn't gotten around to updating the last
          * autovacuum time yet.
          */
         skipit = false;
  
         dlist_reverse_foreach(iter, &DatabaseList)
         {
             avl_dbase  *dbp = dlist_container(avl_dbase, adl_node, iter.cur);
  
             if (dbp->adl_datid == tmp->adw_datid)
             {
                 /*
                  * Skip this database if its next_worker value falls between
                  * the current time and the current time plus naptime.
                  */
                 if (!TimestampDifferenceExceeds(dbp->adl_next_worker,
                                                 current_time, 0) &&
                     !TimestampDifferenceExceeds(current_time,
                                                 dbp->adl_next_worker,
                                                 autovacuum_naptime * 1000))
                     skipit = true;
  
                 break;
             }
         }
         if (skipit)
             continue;
  
         /*
          * Remember the db with oldest autovac time.  (If we are here, both
          * tmp->entry and db->entry must be non-null.)
          */
         if (avdb == NULL ||
             tmp->adw_entry->last_autovac_time < avdb->adw_entry->last_autovac_time)
             avdb = tmp;
     }
  
     /* Found a database -- process it */
     if (avdb != NULL)
     {
         WorkerInfo  worker;
         dlist_node *wptr;
  
         LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
  
         /*
          * Get a worker entry from the freelist.  We checked above, so there
          * really should be a free slot.
          */
         wptr = dlist_pop_head_node(&AutoVacuumShmem->av_freeWorkers);
  
         worker = dlist_container(WorkerInfoData, wi_links, wptr);
         worker->wi_dboid = avdb->adw_datid;
         worker->wi_proc = NULL;
         worker->wi_launchtime = GetCurrentTimestamp();
  
         AutoVacuumShmem->av_startingWorker = worker;
  
         LWLockRelease(AutovacuumLock);
  
         SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER);
  
         retval = avdb->adw_datid;
     }
     else if (skipit)
     {
         /*
          * If we skipped all databases on the list, rebuild it, because it
          * probably contains a dropped database.
          */
         rebuild_database_list(InvalidOid);
     }
  
     MemoryContextSwitchTo(oldcxt);
     MemoryContextDelete(tmpcxt);
  
     return retval;
 }

References avl_dbase::adl_datid, avl_dbase::adl_next_worker, avw_dbase::adw_datid, avw_dbase::adw_entry, avw_dbase::adw_frozenxid, avw_dbase::adw_minmulti, ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate, autovacuum_freeze_max_age, autovacuum_naptime, AutoVacuumShmem, AutoVacuumShmemStruct::av_freeWorkers, AutoVacuumShmemStruct::av_startingWorker, dlist_iter::cur, CurrentMemoryContext, DatabaseList, dblist, dlist_container, dlist_is_empty(), dlist_pop_head_node(), dlist_reverse_foreach, FirstMultiXactId, FirstNormalTransactionId, get_database_list(), GetCurrentTimestamp(), InvalidOid, PgStat_StatDBEntry::last_autovac_time, lfirst, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), MemoryContextDelete(), MemoryContextSwitchTo(), MultiXactIdPrecedes(), MultiXactMemberFreezeThreshold(), pgstat_fetch_stat_dbentry(), PMSIGNAL_START_AUTOVAC_WORKER, ReadNextMultiXactId(), ReadNextTransactionId(), rebuild_database_list(), recentMulti, recentXid, SendPostmasterSignal(), TimestampDifferenceExceeds(), TransactionIdPrecedes(), WorkerInfoData::wi_dboid, WorkerInfoData::wi_launchtime, and WorkerInfoData::wi_proc.

Referenced by launch_worker().

◆ extract_autovac_opts()

static AutoVacOpts* extract_autovac_opts	(	HeapTuple	tup,
		TupleDesc	pg_class_desc
	)

static

Definition at line 2674 of file autovacuum.c.

 {
     bytea      *relopts;
     AutoVacOpts *av;
  
     Assert(((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_RELATION ||
            ((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_MATVIEW ||
            ((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_TOASTVALUE);
  
     relopts = extractRelOptions(tup, pg_class_desc, NULL);
     if (relopts == NULL)
         return NULL;
  
     av = palloc(sizeof(AutoVacOpts));
     memcpy(av, &(((StdRdOptions *) relopts)->autovacuum), sizeof(AutoVacOpts));
     pfree(relopts);
  
     return av;
 }

References Assert, av, extractRelOptions(), GETSTRUCT, palloc(), and pfree().

Referenced by do_autovacuum(), and table_recheck_autovac().

◆ FreeWorkerInfo()

static void FreeWorkerInfo	(	int	code,
		Datum	arg
	)

static

Definition at line 1585 of file autovacuum.c.

 {
     if (MyWorkerInfo != NULL)
     {
         LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
  
         /*
          * Wake the launcher up so that he can launch a new worker immediately
          * if required.  We only save the launcher's PID in local memory here;
          * the actual signal will be sent when the PGPROC is recycled.  Note
          * that we always do this, so that the launcher can rebalance the cost
          * limit setting of the remaining workers.
          *
          * We somewhat ignore the risk that the launcher changes its PID
          * between us reading it and the actual kill; we expect ProcKill to be
          * called shortly after us, and we assume that PIDs are not reused too
          * quickly after a process exits.
          */
         AutovacuumLauncherPid = AutoVacuumShmem->av_launcherpid;
  
         dlist_delete(&MyWorkerInfo->wi_links);
         MyWorkerInfo->wi_dboid = InvalidOid;
         MyWorkerInfo->wi_tableoid = InvalidOid;
         MyWorkerInfo->wi_sharedrel = false;
         MyWorkerInfo->wi_proc = NULL;
         MyWorkerInfo->wi_launchtime = 0;
         pg_atomic_clear_flag(&MyWorkerInfo->wi_dobalance);
         dlist_push_head(&AutoVacuumShmem->av_freeWorkers,
                         &MyWorkerInfo->wi_links);
         /* not mine anymore */
         MyWorkerInfo = NULL;
  
         /*
          * now that we're inactive, cause a rebalancing of the surviving
          * workers
          */
         AutoVacuumShmem->av_signal[AutoVacRebalance] = true;
         LWLockRelease(AutovacuumLock);
     }
 }

References AutoVacRebalance, AutovacuumLauncherPid, AutoVacuumShmem, AutoVacuumShmemStruct::av_freeWorkers, AutoVacuumShmemStruct::av_launcherpid, AutoVacuumShmemStruct::av_signal, dlist_delete(), dlist_push_head(), InvalidOid, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MyWorkerInfo, pg_atomic_clear_flag(), WorkerInfoData::wi_dboid, WorkerInfoData::wi_dobalance, WorkerInfoData::wi_launchtime, WorkerInfoData::wi_links, WorkerInfoData::wi_proc, WorkerInfoData::wi_sharedrel, and WorkerInfoData::wi_tableoid.

Referenced by AutoVacWorkerMain().

◆ get_database_list()

static List* get_database_list ( void )

static

Definition at line 1788 of file autovacuum.c.

 {
     List       *dblist = NIL;
     Relation    rel;
     TableScanDesc scan;
     HeapTuple   tup;
     MemoryContext resultcxt;
  
     /* This is the context that we will allocate our output data in */
     resultcxt = CurrentMemoryContext;
  
     /*
      * Start a transaction so we can access pg_database, and get a snapshot.
      * We don't have a use for the snapshot itself, but we're interested in
      * the secondary effect that it sets RecentGlobalXmin.  (This is critical
      * for anything that reads heap pages, because HOT may decide to prune
      * them even if the process doesn't attempt to modify any tuples.)
      *
      * FIXME: This comment is inaccurate / the code buggy. A snapshot that is
      * not pushed/active does not reliably prevent HOT pruning (->xmin could
      * e.g. be cleared when cache invalidations are processed).
      */
     StartTransactionCommand();
     (void) GetTransactionSnapshot();
  
     rel = table_open(DatabaseRelationId, AccessShareLock);
     scan = table_beginscan_catalog(rel, 0, NULL);
  
     while (HeapTupleIsValid(tup = heap_getnext(scan, ForwardScanDirection)))
     {
         Form_pg_database pgdatabase = (Form_pg_database) GETSTRUCT(tup);
         avw_dbase  *avdb;
         MemoryContext oldcxt;
  
         /*
          * If database has partially been dropped, we can't, nor need to,
          * vacuum it.
          */
         if (database_is_invalid_form(pgdatabase))
         {
             elog(DEBUG2,
                  "autovacuum: skipping invalid database \"%s\"",
                  NameStr(pgdatabase->datname));
             continue;
         }
  
         /*
          * Allocate our results in the caller's context, not the
          * transaction's. We do this inside the loop, and restore the original
          * context at the end, so that leaky things like heap_getnext() are
          * not called in a potentially long-lived context.
          */
         oldcxt = MemoryContextSwitchTo(resultcxt);
  
         avdb = (avw_dbase *) palloc(sizeof(avw_dbase));
  
         avdb->adw_datid = pgdatabase->oid;
         avdb->adw_name = pstrdup(NameStr(pgdatabase->datname));
         avdb->adw_frozenxid = pgdatabase->datfrozenxid;
         avdb->adw_minmulti = pgdatabase->datminmxid;
         /* this gets set later: */
         avdb->adw_entry = NULL;
  
         dblist = lappend(dblist, avdb);
         MemoryContextSwitchTo(oldcxt);
     }
  
     table_endscan(scan);
     table_close(rel, AccessShareLock);
  
     CommitTransactionCommand();
  
     /* Be sure to restore caller's memory context */
     MemoryContextSwitchTo(resultcxt);
  
     return dblist;
 }

References AccessShareLock, avw_dbase::adw_datid, avw_dbase::adw_entry, avw_dbase::adw_frozenxid, avw_dbase::adw_minmulti, avw_dbase::adw_name, CommitTransactionCommand(), CurrentMemoryContext, database_is_invalid_form(), dblist, DEBUG2, elog, ForwardScanDirection, GETSTRUCT, GetTransactionSnapshot(), heap_getnext(), HeapTupleIsValid, lappend(), MemoryContextSwitchTo(), NameStr, NIL, palloc(), pstrdup(), StartTransactionCommand(), table_beginscan_catalog(), table_close(), table_endscan(), and table_open().

Referenced by do_start_worker(), and rebuild_database_list().

◆ HandleAutoVacLauncherInterrupts()

static void HandleAutoVacLauncherInterrupts ( void )

static

Definition at line 740 of file autovacuum.c.

 {
     /* the normal shutdown case */
     if (ShutdownRequestPending)
         AutoVacLauncherShutdown();
  
     if (ConfigReloadPending)
     {
         ConfigReloadPending = false;
         ProcessConfigFile(PGC_SIGHUP);
  
         /* shutdown requested in config file? */
         if (!AutoVacuumingActive())
             AutoVacLauncherShutdown();
  
         /* rebuild the list in case the naptime changed */
         rebuild_database_list(InvalidOid);
     }
  
     /* Process barrier events */
     if (ProcSignalBarrierPending)
         ProcessProcSignalBarrier();
  
     /* Perform logging of memory contexts of this process */
     if (LogMemoryContextPending)
         ProcessLogMemoryContextInterrupt();
  
     /* Process sinval catchup interrupts that happened while sleeping */
     ProcessCatchupInterrupt();
 }

References AutoVacLauncherShutdown(), AutoVacuumingActive(), ConfigReloadPending, InvalidOid, LogMemoryContextPending, PGC_SIGHUP, ProcessCatchupInterrupt(), ProcessConfigFile(), ProcessLogMemoryContextInterrupt(), ProcessProcSignalBarrier(), ProcSignalBarrierPending, rebuild_database_list(), and ShutdownRequestPending.

◆ launch_worker()

static void launch_worker ( TimestampTz now )

static

Definition at line 1285 of file autovacuum.c.

 {
     Oid         dbid;
     dlist_iter  iter;
  
     dbid = do_start_worker();
     if (OidIsValid(dbid))
     {
         bool        found = false;
  
         /*
          * Walk the database list and update the corresponding entry.  If the
          * database is not on the list, we'll recreate the list.
          */
         dlist_foreach(iter, &DatabaseList)
         {
             avl_dbase  *avdb = dlist_container(avl_dbase, adl_node, iter.cur);
  
             if (avdb->adl_datid == dbid)
             {
                 found = true;
  
                 /*
                  * add autovacuum_naptime seconds to the current time, and use
                  * that as the new "next_worker" field for this database.
                  */
                 avdb->adl_next_worker =
                     TimestampTzPlusMilliseconds(now, autovacuum_naptime * 1000);
  
                 dlist_move_head(&DatabaseList, iter.cur);
                 break;
             }
         }
  
         /*
          * If the database was not present in the database list, we rebuild
          * the list.  It's possible that the database does not get into the
          * list anyway, for example if it's a database that doesn't have a
          * pgstat entry, but this is not a problem because we don't want to
          * schedule workers regularly into those in any case.
          */
         if (!found)
             rebuild_database_list(dbid);
     }
 }

References avl_dbase::adl_datid, avl_dbase::adl_next_worker, autovacuum_naptime, dlist_iter::cur, DatabaseList, dlist_container, dlist_foreach, dlist_move_head(), do_start_worker(), now(), OidIsValid, rebuild_database_list(), and TimestampTzPlusMilliseconds.

◆ launcher_determine_sleep()

static void launcher_determine_sleep	(	bool	canlaunch,
		bool	recursing,
		struct timeval *	nap
	)

static

Definition at line 792 of file autovacuum.c.

 {
     /*
      * We sleep until the next scheduled vacuum.  We trust that when the
      * database list was built, care was taken so that no entries have times
      * in the past; if the first entry has too close a next_worker value, or a
      * time in the past, we will sleep a small nominal time.
      */
     if (!canlaunch)
     {
         nap->tv_sec = autovacuum_naptime;
         nap->tv_usec = 0;
     }
     else if (!dlist_is_empty(&DatabaseList))
     {
         TimestampTz current_time = GetCurrentTimestamp();
         TimestampTz next_wakeup;
         avl_dbase  *avdb;
         long        secs;
         int         usecs;
  
         avdb = dlist_tail_element(avl_dbase, adl_node, &DatabaseList);
  
         next_wakeup = avdb->adl_next_worker;
         TimestampDifference(current_time, next_wakeup, &secs, &usecs);
  
         nap->tv_sec = secs;
         nap->tv_usec = usecs;
     }
     else
     {
         /* list is empty, sleep for whole autovacuum_naptime seconds  */
         nap->tv_sec = autovacuum_naptime;
         nap->tv_usec = 0;
     }
  
     /*
      * If the result is exactly zero, it means a database had an entry with
      * time in the past.  Rebuild the list so that the databases are evenly
      * distributed again, and recalculate the time to sleep.  This can happen
      * if there are more tables needing vacuum than workers, and they all take
      * longer to vacuum than autovacuum_naptime.
      *
      * We only recurse once.  rebuild_database_list should always return times
      * in the future, but it seems best not to trust too much on that.
      */
     if (nap->tv_sec == 0 && nap->tv_usec == 0 && !recursing)
     {
         rebuild_database_list(InvalidOid);
         launcher_determine_sleep(canlaunch, true, nap);
         return;
     }
  
     /* The smallest time we'll allow the launcher to sleep. */
     if (nap->tv_sec <= 0 && nap->tv_usec <= MIN_AUTOVAC_SLEEPTIME * 1000)
     {
         nap->tv_sec = 0;
         nap->tv_usec = MIN_AUTOVAC_SLEEPTIME * 1000;
     }
  
     /*
      * If the sleep time is too large, clamp it to an arbitrary maximum (plus
      * any fractional seconds, for simplicity).  This avoids an essentially
      * infinite sleep in strange cases like the system clock going backwards a
      * few years.
      */
     if (nap->tv_sec > MAX_AUTOVAC_SLEEPTIME)
         nap->tv_sec = MAX_AUTOVAC_SLEEPTIME;
 }

References avl_dbase::adl_next_worker, autovacuum_naptime, DatabaseList, dlist_is_empty(), dlist_tail_element, GetCurrentTimestamp(), InvalidOid, MAX_AUTOVAC_SLEEPTIME, MIN_AUTOVAC_SLEEPTIME, rebuild_database_list(), and TimestampDifference().

◆ perform_work_item()

static void perform_work_item ( AutoVacuumWorkItem * workitem )

static

Definition at line 2560 of file autovacuum.c.

 {
     char       *cur_datname = NULL;
     char       *cur_nspname = NULL;
     char       *cur_relname = NULL;
  
     /*
      * Note we do not store table info in MyWorkerInfo, since this is not
      * vacuuming proper.
      */
  
     /*
      * Save the relation name for a possible error message, to avoid a catalog
      * lookup in case of an error.  If any of these return NULL, then the
      * relation has been dropped since last we checked; skip it.
      */
     Assert(CurrentMemoryContext == AutovacMemCxt);
  
     cur_relname = get_rel_name(workitem->avw_relation);
     cur_nspname = get_namespace_name(get_rel_namespace(workitem->avw_relation));
     cur_datname = get_database_name(MyDatabaseId);
     if (!cur_relname || !cur_nspname || !cur_datname)
         goto deleted2;
  
     autovac_report_workitem(workitem, cur_nspname, cur_relname);
  
     /* clean up memory before each work item */
     MemoryContextReset(PortalContext);
  
     /*
      * We will abort the current work item if something errors out, and
      * continue with the next one; in particular, this happens if we are
      * interrupted with SIGINT.  Note that this means that the work item list
      * can be lossy.
      */
     PG_TRY();
     {
         /* Use PortalContext for any per-work-item allocations */
         MemoryContextSwitchTo(PortalContext);
  
         /*
          * Have at it.  Functions called here are responsible for any required
          * user switch and sandbox.
          */
         switch (workitem->avw_type)
         {
             case AVW_BRINSummarizeRange:
                 DirectFunctionCall2(brin_summarize_range,
                                     ObjectIdGetDatum(workitem->avw_relation),
                                     Int64GetDatum((int64) workitem->avw_blockNumber));
                 break;
             default:
                 elog(WARNING, "unrecognized work item found: type %d",
                      workitem->avw_type);
                 break;
         }
  
         /*
          * Clear a possible query-cancel signal, to avoid a late reaction to
          * an automatically-sent signal because of vacuuming the current table
          * (we're done with it, so it would make no sense to cancel at this
          * point.)
          */
         QueryCancelPending = false;
     }
     PG_CATCH();
     {
         /*
          * Abort the transaction, start a new one, and proceed with the next
          * table in our list.
          */
         HOLD_INTERRUPTS();
         errcontext("processing work entry for relation \"%s.%s.%s\"",
                    cur_datname, cur_nspname, cur_relname);
         EmitErrorReport();
  
         /* this resets ProcGlobal->statusFlags[i] too */
         AbortOutOfAnyTransaction();
         FlushErrorState();
         MemoryContextReset(PortalContext);
  
         /* restart our transaction for the following operations */
         StartTransactionCommand();
         RESUME_INTERRUPTS();
     }
     PG_END_TRY();
  
     /* Make sure we're back in AutovacMemCxt */
     MemoryContextSwitchTo(AutovacMemCxt);
  
     /* We intentionally do not set did_vacuum here */
  
     /* be tidy */
 deleted2:
     if (cur_datname)
         pfree(cur_datname);
     if (cur_nspname)
         pfree(cur_nspname);
     if (cur_relname)
         pfree(cur_relname);
 }

References AbortOutOfAnyTransaction(), Assert, autovac_report_workitem(), AutovacMemCxt, AutoVacuumWorkItem::avw_blockNumber, AVW_BRINSummarizeRange, AutoVacuumWorkItem::avw_relation, AutoVacuumWorkItem::avw_type, brin_summarize_range(), CurrentMemoryContext, DirectFunctionCall2, elog, EmitErrorReport(), errcontext, FlushErrorState(), get_database_name(), get_namespace_name(), get_rel_name(), get_rel_namespace(), HOLD_INTERRUPTS, Int64GetDatum(), MemoryContextReset(), MemoryContextSwitchTo(), MyDatabaseId, ObjectIdGetDatum(), pfree(), PG_CATCH, PG_END_TRY, PG_TRY, PortalContext, QueryCancelPending, RESUME_INTERRUPTS, StartTransactionCommand(), and WARNING.

Referenced by do_autovacuum().

◆ rebuild_database_list()

static void rebuild_database_list ( Oid newdb )

static

Definition at line 876 of file autovacuum.c.

 {
     List       *dblist;
     ListCell   *cell;
     MemoryContext newcxt;
     MemoryContext oldcxt;
     MemoryContext tmpcxt;
     HASHCTL     hctl;
     int         score;
     int         nelems;
     HTAB       *dbhash;
     dlist_iter  iter;
  
     newcxt = AllocSetContextCreate(AutovacMemCxt,
                                    "Autovacuum database list",
                                    ALLOCSET_DEFAULT_SIZES);
     tmpcxt = AllocSetContextCreate(newcxt,
                                    "Autovacuum database list (tmp)",
                                    ALLOCSET_DEFAULT_SIZES);
     oldcxt = MemoryContextSwitchTo(tmpcxt);
  
     /*
      * Implementing this is not as simple as it sounds, because we need to put
      * the new database at the end of the list; next the databases that were
      * already on the list, and finally (at the tail of the list) all the
      * other databases that are not on the existing list.
      *
      * To do this, we build an empty hash table of scored databases.  We will
      * start with the lowest score (zero) for the new database, then
      * increasing scores for the databases in the existing list, in order, and
      * lastly increasing scores for all databases gotten via
      * get_database_list() that are not already on the hash.
      *
      * Then we will put all the hash elements into an array, sort the array by
      * score, and finally put the array elements into the new doubly linked
      * list.
      */
     hctl.keysize = sizeof(Oid);
     hctl.entrysize = sizeof(avl_dbase);
     hctl.hcxt = tmpcxt;
     dbhash = hash_create("autovacuum db hash", 20, &hctl,   /* magic number here
                                                              * FIXME */
                          HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
  
     /* start by inserting the new database */
     score = 0;
     if (OidIsValid(newdb))
     {
         avl_dbase  *db;
         PgStat_StatDBEntry *entry;
  
         /* only consider this database if it has a pgstat entry */
         entry = pgstat_fetch_stat_dbentry(newdb);
         if (entry != NULL)
         {
             /* we assume it isn't found because the hash was just created */
             db = hash_search(dbhash, &newdb, HASH_ENTER, NULL);
  
             /* hash_search already filled in the key */
             db->adl_score = score++;
             /* next_worker is filled in later */
         }
     }
  
     /* Now insert the databases from the existing list */
     dlist_foreach(iter, &DatabaseList)
     {
         avl_dbase  *avdb = dlist_container(avl_dbase, adl_node, iter.cur);
         avl_dbase  *db;
         bool        found;
         PgStat_StatDBEntry *entry;
  
         /*
          * skip databases with no stat entries -- in particular, this gets rid
          * of dropped databases
          */
         entry = pgstat_fetch_stat_dbentry(avdb->adl_datid);
         if (entry == NULL)
             continue;
  
         db = hash_search(dbhash, &(avdb->adl_datid), HASH_ENTER, &found);
  
         if (!found)
         {
             /* hash_search already filled in the key */
             db->adl_score = score++;
             /* next_worker is filled in later */
         }
     }
  
     /* finally, insert all qualifying databases not previously inserted */
     dblist = get_database_list();
     foreach(cell, dblist)
     {
         avw_dbase  *avdb = lfirst(cell);
         avl_dbase  *db;
         bool        found;
         PgStat_StatDBEntry *entry;
  
         /* only consider databases with a pgstat entry */
         entry = pgstat_fetch_stat_dbentry(avdb->adw_datid);
         if (entry == NULL)
             continue;
  
         db = hash_search(dbhash, &(avdb->adw_datid), HASH_ENTER, &found);
         /* only update the score if the database was not already on the hash */
         if (!found)
         {
             /* hash_search already filled in the key */
             db->adl_score = score++;
             /* next_worker is filled in later */
         }
     }
     nelems = score;
  
     /* from here on, the allocated memory belongs to the new list */
     MemoryContextSwitchTo(newcxt);
     dlist_init(&DatabaseList);
  
     if (nelems > 0)
     {
         TimestampTz current_time;
         int         millis_increment;
         avl_dbase  *dbary;
         avl_dbase  *db;
         HASH_SEQ_STATUS seq;
         int         i;
  
         /* put all the hash elements into an array */
         dbary = palloc(nelems * sizeof(avl_dbase));
  
         i = 0;
         hash_seq_init(&seq, dbhash);
         while ((db = hash_seq_search(&seq)) != NULL)
             memcpy(&(dbary[i++]), db, sizeof(avl_dbase));
  
         /* sort the array */
         qsort(dbary, nelems, sizeof(avl_dbase), db_comparator);
  
         /*
          * Determine the time interval between databases in the schedule. If
          * we see that the configured naptime would take us to sleep times
          * lower than our min sleep time (which launcher_determine_sleep is
          * coded not to allow), silently use a larger naptime (but don't touch
          * the GUC variable).
          */
         millis_increment = 1000.0 * autovacuum_naptime / nelems;
         if (millis_increment <= MIN_AUTOVAC_SLEEPTIME)
             millis_increment = MIN_AUTOVAC_SLEEPTIME * 1.1;
  
         current_time = GetCurrentTimestamp();
  
         /*
          * move the elements from the array into the dlist, setting the
          * next_worker while walking the array
          */
         for (i = 0; i < nelems; i++)
         {
             db = &(dbary[i]);
  
             current_time = TimestampTzPlusMilliseconds(current_time,
                                                        millis_increment);
             db->adl_next_worker = current_time;
  
             /* later elements should go closer to the head of the list */
             dlist_push_head(&DatabaseList, &db->adl_node);
         }
     }
  
     /* all done, clean up memory */
     if (DatabaseListCxt != NULL)
         MemoryContextDelete(DatabaseListCxt);
     MemoryContextDelete(tmpcxt);
     DatabaseListCxt = newcxt;
     MemoryContextSwitchTo(oldcxt);
 }

References avl_dbase::adl_datid, avl_dbase::adl_next_worker, avl_dbase::adl_node, avl_dbase::adl_score, avw_dbase::adw_datid, ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate, AutovacMemCxt, autovacuum_naptime, dlist_iter::cur, DatabaseList, DatabaseListCxt, db_comparator(), dblist, dlist_container, dlist_foreach, dlist_init(), dlist_push_head(), HASHCTL::entrysize, get_database_list(), GetCurrentTimestamp(), HASH_BLOBS, HASH_CONTEXT, hash_create(), HASH_ELEM, HASH_ENTER, hash_search(), hash_seq_init(), hash_seq_search(), HASHCTL::hcxt, i, HASHCTL::keysize, lfirst, MemoryContextDelete(), MemoryContextSwitchTo(), MIN_AUTOVAC_SLEEPTIME, OidIsValid, palloc(), pgstat_fetch_stat_dbentry(), qsort, and TimestampTzPlusMilliseconds.

Referenced by do_start_worker(), HandleAutoVacLauncherInterrupts(), launch_worker(), and launcher_determine_sleep().

◆ recheck_relation_needs_vacanalyze()

static void recheck_relation_needs_vacanalyze	(	Oid	relid,
		AutoVacOpts *	avopts,
		Form_pg_class	classForm,
		int	effective_multixact_freeze_max_age,
		bool *	dovacuum,
		bool *	doanalyze,
		bool *	wraparound
	)

static

Definition at line 2844 of file autovacuum.c.

 {
     PgStat_StatTabEntry *tabentry;
  
     /* fetch the pgstat table entry */
     tabentry = pgstat_fetch_stat_tabentry_ext(classForm->relisshared,
                                               relid);
  
     relation_needs_vacanalyze(relid, avopts, classForm, tabentry,
                               effective_multixact_freeze_max_age,
                               dovacuum, doanalyze, wraparound);
  
     /* ignore ANALYZE for toast tables */
     if (classForm->relkind == RELKIND_TOASTVALUE)
         *doanalyze = false;
 }

References pgstat_fetch_stat_tabentry_ext(), and relation_needs_vacanalyze().

Referenced by table_recheck_autovac().

◆ relation_needs_vacanalyze()

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
	)

static

Definition at line 2905 of file autovacuum.c.

 {
     bool        force_vacuum;
     bool        av_enabled;
     float4      reltuples;      /* pg_class.reltuples */
  
     /* constants from reloptions or GUC variables */
     int         vac_base_thresh,
                 vac_ins_base_thresh,
                 anl_base_thresh;
     float4      vac_scale_factor,
                 vac_ins_scale_factor,
                 anl_scale_factor;
  
     /* thresholds calculated from above constants */
     float4      vacthresh,
                 vacinsthresh,
                 anlthresh;
  
     /* number of vacuum (resp. analyze) tuples at this time */
     float4      vactuples,
                 instuples,
                 anltuples;
  
     /* freeze parameters */
     int         freeze_max_age;
     int         multixact_freeze_max_age;
     TransactionId xidForceLimit;
     MultiXactId multiForceLimit;
  
     Assert(classForm != NULL);
     Assert(OidIsValid(relid));
  
     /*
      * Determine vacuum/analyze equation parameters.  We have two possible
      * sources: the passed reloptions (which could be a main table or a toast
      * table), or the autovacuum GUC variables.
      */
  
     /* -1 in autovac setting means use plain vacuum_scale_factor */
     vac_scale_factor = (relopts && relopts->vacuum_scale_factor >= 0)
         ? relopts->vacuum_scale_factor
         : autovacuum_vac_scale;
  
     vac_base_thresh = (relopts && relopts->vacuum_threshold >= 0)
         ? relopts->vacuum_threshold
         : autovacuum_vac_thresh;
  
     vac_ins_scale_factor = (relopts && relopts->vacuum_ins_scale_factor >= 0)
         ? relopts->vacuum_ins_scale_factor
         : autovacuum_vac_ins_scale;
  
     /* -1 is used to disable insert vacuums */
     vac_ins_base_thresh = (relopts && relopts->vacuum_ins_threshold >= -1)
         ? relopts->vacuum_ins_threshold
         : autovacuum_vac_ins_thresh;
  
     anl_scale_factor = (relopts && relopts->analyze_scale_factor >= 0)
         ? relopts->analyze_scale_factor
         : autovacuum_anl_scale;
  
     anl_base_thresh = (relopts && relopts->analyze_threshold >= 0)
         ? relopts->analyze_threshold
         : autovacuum_anl_thresh;
  
     freeze_max_age = (relopts && relopts->freeze_max_age >= 0)
         ? Min(relopts->freeze_max_age, autovacuum_freeze_max_age)
         : autovacuum_freeze_max_age;
  
     multixact_freeze_max_age = (relopts && relopts->multixact_freeze_max_age >= 0)
         ? Min(relopts->multixact_freeze_max_age, effective_multixact_freeze_max_age)
         : effective_multixact_freeze_max_age;
  
     av_enabled = (relopts ? relopts->enabled : true);
  
     /* Force vacuum if table is at risk of wraparound */
     xidForceLimit = recentXid - freeze_max_age;
     if (xidForceLimit < FirstNormalTransactionId)
         xidForceLimit -= FirstNormalTransactionId;
     force_vacuum = (TransactionIdIsNormal(classForm->relfrozenxid) &&
                     TransactionIdPrecedes(classForm->relfrozenxid,
                                           xidForceLimit));
     if (!force_vacuum)
     {
         multiForceLimit = recentMulti - multixact_freeze_max_age;
         if (multiForceLimit < FirstMultiXactId)
             multiForceLimit -= FirstMultiXactId;
         force_vacuum = MultiXactIdIsValid(classForm->relminmxid) &&
             MultiXactIdPrecedes(classForm->relminmxid, multiForceLimit);
     }
     *wraparound = force_vacuum;
  
     /* User disabled it in pg_class.reloptions?  (But ignore if at risk) */
     if (!av_enabled && !force_vacuum)
     {
         *doanalyze = false;
         *dovacuum = false;
         return;
     }
  
     /*
      * If we found stats for the table, and autovacuum is currently enabled,
      * make a threshold-based decision whether to vacuum and/or analyze.  If
      * autovacuum is currently disabled, we must be here for anti-wraparound
      * vacuuming only, so don't vacuum (or analyze) anything that's not being
      * forced.
      */
     if (PointerIsValid(tabentry) && AutoVacuumingActive())
     {
         reltuples = classForm->reltuples;
         vactuples = tabentry->dead_tuples;
         instuples = tabentry->ins_since_vacuum;
         anltuples = tabentry->mod_since_analyze;
  
         /* If the table hasn't yet been vacuumed, take reltuples as zero */
         if (reltuples < 0)
             reltuples = 0;
  
         vacthresh = (float4) vac_base_thresh + vac_scale_factor * reltuples;
         vacinsthresh = (float4) vac_ins_base_thresh + vac_ins_scale_factor * reltuples;
         anlthresh = (float4) anl_base_thresh + anl_scale_factor * reltuples;
  
         /*
          * Note that we don't need to take special consideration for stat
          * reset, because if that happens, the last vacuum and analyze counts
          * will be reset too.
          */
         if (vac_ins_base_thresh >= 0)
             elog(DEBUG3, "%s: vac: %.0f (threshold %.0f), ins: %.0f (threshold %.0f), anl: %.0f (threshold %.0f)",
                  NameStr(classForm->relname),
                  vactuples, vacthresh, instuples, vacinsthresh, anltuples, anlthresh);
         else
             elog(DEBUG3, "%s: vac: %.0f (threshold %.0f), ins: (disabled), anl: %.0f (threshold %.0f)",
                  NameStr(classForm->relname),
                  vactuples, vacthresh, anltuples, anlthresh);
  
         /* Determine if this table needs vacuum or analyze. */
         *dovacuum = force_vacuum || (vactuples > vacthresh) ||
             (vac_ins_base_thresh >= 0 && instuples > vacinsthresh);
         *doanalyze = (anltuples > anlthresh);
     }
     else
     {
         /*
          * Skip a table not found in stat hash, unless we have to force vacuum
          * for anti-wrap purposes.  If it's not acted upon, there's no need to
          * vacuum it.
          */
         *dovacuum = force_vacuum;
         *doanalyze = false;
     }
  
     /* ANALYZE refuses to work with pg_statistic */
     if (relid == StatisticRelationId)
         *doanalyze = false;
 }

References AutoVacOpts::analyze_scale_factor, AutoVacOpts::analyze_threshold, Assert, autovacuum_anl_scale, autovacuum_anl_thresh, autovacuum_freeze_max_age, autovacuum_vac_ins_scale, autovacuum_vac_ins_thresh, autovacuum_vac_scale, autovacuum_vac_thresh, AutoVacuumingActive(), PgStat_StatTabEntry::dead_tuples, DEBUG3, elog, AutoVacOpts::enabled, FirstMultiXactId, FirstNormalTransactionId, AutoVacOpts::freeze_max_age, PgStat_StatTabEntry::ins_since_vacuum, Min, PgStat_StatTabEntry::mod_since_analyze, AutoVacOpts::multixact_freeze_max_age, MultiXactIdIsValid, MultiXactIdPrecedes(), NameStr, OidIsValid, PointerIsValid, recentMulti, recentXid, TransactionIdIsNormal, TransactionIdPrecedes(), AutoVacOpts::vacuum_ins_scale_factor, AutoVacOpts::vacuum_ins_threshold, AutoVacOpts::vacuum_scale_factor, and AutoVacOpts::vacuum_threshold.

Referenced by do_autovacuum(), and recheck_relation_needs_vacanalyze().

◆ table_recheck_autovac()

static autovac_table* table_recheck_autovac	(	Oid	relid,
		HTAB *	table_toast_map,
		TupleDesc	pg_class_desc,
		int	effective_multixact_freeze_max_age
	)

static

Definition at line 2704 of file autovacuum.c.

 {
     Form_pg_class classForm;
     HeapTuple   classTup;
     bool        dovacuum;
     bool        doanalyze;
     autovac_table *tab = NULL;
     bool        wraparound;
     AutoVacOpts *avopts;
  
     /* fetch the relation's relcache entry */
     classTup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
     if (!HeapTupleIsValid(classTup))
         return NULL;
     classForm = (Form_pg_class) GETSTRUCT(classTup);
  
     /*
      * Get the applicable reloptions.  If it is a TOAST table, try to get the
      * main table reloptions if the toast table itself doesn't have.
      */
     avopts = extract_autovac_opts(classTup, pg_class_desc);
     if (classForm->relkind == RELKIND_TOASTVALUE &&
         avopts == NULL && table_toast_map != NULL)
     {
         av_relation *hentry;
         bool        found;
  
         hentry = hash_search(table_toast_map, &relid, HASH_FIND, &found);
         if (found && hentry->ar_hasrelopts)
             avopts = &hentry->ar_reloptions;
     }
  
     recheck_relation_needs_vacanalyze(relid, avopts, classForm,
                                       effective_multixact_freeze_max_age,
                                       &dovacuum, &doanalyze, &wraparound);
  
     /* OK, it needs something done */
     if (doanalyze || dovacuum)
     {
         int         freeze_min_age;
         int         freeze_table_age;
         int         multixact_freeze_min_age;
         int         multixact_freeze_table_age;
         int         log_min_duration;
  
         /*
          * Calculate the vacuum cost parameters and the freeze ages.  If there
          * are options set in pg_class.reloptions, use them; in the case of a
          * toast table, try the main table too.  Otherwise use the GUC
          * defaults, autovacuum's own first and plain vacuum second.
          */
  
         /* -1 in autovac setting means use log_autovacuum_min_duration */
         log_min_duration = (avopts && avopts->log_min_duration >= 0)
             ? avopts->log_min_duration
             : Log_autovacuum_min_duration;
  
         /* these do not have autovacuum-specific settings */
         freeze_min_age = (avopts && avopts->freeze_min_age >= 0)
             ? avopts->freeze_min_age
             : default_freeze_min_age;
  
         freeze_table_age = (avopts && avopts->freeze_table_age >= 0)
             ? avopts->freeze_table_age
             : default_freeze_table_age;
  
         multixact_freeze_min_age = (avopts &&
                                     avopts->multixact_freeze_min_age >= 0)
             ? avopts->multixact_freeze_min_age
             : default_multixact_freeze_min_age;
  
         multixact_freeze_table_age = (avopts &&
                                       avopts->multixact_freeze_table_age >= 0)
             ? avopts->multixact_freeze_table_age
             : default_multixact_freeze_table_age;
  
         tab = palloc(sizeof(autovac_table));
         tab->at_relid = relid;
         tab->at_sharedrel = classForm->relisshared;
  
         /*
          * Select VACUUM options.  Note we don't say VACOPT_PROCESS_TOAST, so
          * that vacuum() skips toast relations.  Also note we tell vacuum() to
          * skip vac_update_datfrozenxid(); we'll do that separately.
          */
         tab->at_params.options =
             (dovacuum ? (VACOPT_VACUUM |
                          VACOPT_PROCESS_MAIN |
                          VACOPT_SKIP_DATABASE_STATS) : 0) |
             (doanalyze ? VACOPT_ANALYZE : 0) |
             (!wraparound ? VACOPT_SKIP_LOCKED : 0);
  
         /*
          * index_cleanup and truncate are unspecified at first in autovacuum.
          * They will be filled in with usable values using their reloptions
          * (or reloption defaults) later.
          */
         tab->at_params.index_cleanup = VACOPTVALUE_UNSPECIFIED;
         tab->at_params.truncate = VACOPTVALUE_UNSPECIFIED;
         /* As of now, we don't support parallel vacuum for autovacuum */
         tab->at_params.nworkers = -1;
         tab->at_params.freeze_min_age = freeze_min_age;
         tab->at_params.freeze_table_age = freeze_table_age;
         tab->at_params.multixact_freeze_min_age = multixact_freeze_min_age;
         tab->at_params.multixact_freeze_table_age = multixact_freeze_table_age;
         tab->at_params.is_wraparound = wraparound;
         tab->at_params.log_min_duration = log_min_duration;
         tab->at_params.toast_parent = InvalidOid;
         tab->at_storage_param_vac_cost_limit = avopts ?
             avopts->vacuum_cost_limit : 0;
         tab->at_storage_param_vac_cost_delay = avopts ?
             avopts->vacuum_cost_delay : -1;
         tab->at_relname = NULL;
         tab->at_nspname = NULL;
         tab->at_datname = NULL;
  
         /*
          * If any of the cost delay parameters has been set individually for
          * this table, disable the balancing algorithm.
          */
         tab->at_dobalance =
             !(avopts && (avopts->vacuum_cost_limit > 0 ||
                          avopts->vacuum_cost_delay >= 0));
     }
  
     heap_freetuple(classTup);
     return tab;
 }

References av_relation::ar_hasrelopts, av_relation::ar_reloptions, autovac_table::at_datname, autovac_table::at_dobalance, autovac_table::at_nspname, autovac_table::at_params, autovac_table::at_relid, autovac_table::at_relname, autovac_table::at_sharedrel, autovac_table::at_storage_param_vac_cost_delay, autovac_table::at_storage_param_vac_cost_limit, default_freeze_min_age, default_freeze_table_age, default_multixact_freeze_min_age, default_multixact_freeze_table_age, extract_autovac_opts(), VacuumParams::freeze_min_age, AutoVacOpts::freeze_min_age, VacuumParams::freeze_table_age, AutoVacOpts::freeze_table_age, GETSTRUCT, HASH_FIND, hash_search(), heap_freetuple(), HeapTupleIsValid, VacuumParams::index_cleanup, InvalidOid, VacuumParams::is_wraparound, Log_autovacuum_min_duration, VacuumParams::log_min_duration, AutoVacOpts::log_min_duration, VacuumParams::multixact_freeze_min_age, AutoVacOpts::multixact_freeze_min_age, VacuumParams::multixact_freeze_table_age, AutoVacOpts::multixact_freeze_table_age, VacuumParams::nworkers, ObjectIdGetDatum(), VacuumParams::options, palloc(), recheck_relation_needs_vacanalyze(), SearchSysCacheCopy1, VacuumParams::toast_parent, VacuumParams::truncate, VACOPT_ANALYZE, VACOPT_PROCESS_MAIN, VACOPT_SKIP_DATABASE_STATS, VACOPT_SKIP_LOCKED, VACOPT_VACUUM, VACOPTVALUE_UNSPECIFIED, AutoVacOpts::vacuum_cost_delay, and AutoVacOpts::vacuum_cost_limit.

Referenced by do_autovacuum().

◆ VacuumUpdateCosts()

void VacuumUpdateCosts ( void )

Definition at line 1633 of file autovacuum.c.

 {
     if (MyWorkerInfo)
     {
         if (av_storage_param_cost_delay >= 0)
             vacuum_cost_delay = av_storage_param_cost_delay;
         else if (autovacuum_vac_cost_delay >= 0)
             vacuum_cost_delay = autovacuum_vac_cost_delay;
         else
             /* fall back to VacuumCostDelay */
             vacuum_cost_delay = VacuumCostDelay;
  
         AutoVacuumUpdateCostLimit();
     }
     else
     {
         /* Must be explicit VACUUM or ANALYZE */
         vacuum_cost_delay = VacuumCostDelay;
         vacuum_cost_limit = VacuumCostLimit;
     }
  
     /*
      * If configuration changes are allowed to impact VacuumCostActive, make
      * sure it is updated.
      */
     if (VacuumFailsafeActive)
         Assert(!VacuumCostActive);
     else if (vacuum_cost_delay > 0)
         VacuumCostActive = true;
     else
     {
         VacuumCostActive = false;
         VacuumCostBalance = 0;
     }
  
     /*
      * Since the cost logging requires a lock, avoid rendering the log message
      * in case we are using a message level where the log wouldn't be emitted.
      */
     if (MyWorkerInfo && message_level_is_interesting(DEBUG2))
     {
         Oid         dboid,
                     tableoid;
  
         Assert(!LWLockHeldByMe(AutovacuumLock));
  
         LWLockAcquire(AutovacuumLock, LW_SHARED);
         dboid = MyWorkerInfo->wi_dboid;
         tableoid = MyWorkerInfo->wi_tableoid;
         LWLockRelease(AutovacuumLock);
  
         elog(DEBUG2,
              "Autovacuum VacuumUpdateCosts(db=%u, rel=%u, dobalance=%s, cost_limit=%d, cost_delay=%g active=%s failsafe=%s)",
              dboid, tableoid, pg_atomic_unlocked_test_flag(&MyWorkerInfo->wi_dobalance) ? "no" : "yes",
              vacuum_cost_limit, vacuum_cost_delay,
              vacuum_cost_delay > 0 ? "yes" : "no",
              VacuumFailsafeActive ? "yes" : "no");
     }
 }