LCOV - code coverage report
Current view: top level - src/backend/postmaster - autovacuum.c (source / functions) Hit Total Coverage
Test: PostgreSQL 19devel Lines: 824 985 83.7 %
Date: 2025-10-15 14:18:20 Functions: 32 33 97.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /*-------------------------------------------------------------------------
       2             :  *
       3             :  * autovacuum.c
       4             :  *
       5             :  * PostgreSQL Integrated Autovacuum Daemon
       6             :  *
       7             :  * The autovacuum system is structured in two different kinds of processes: the
       8             :  * autovacuum launcher and the autovacuum worker.  The launcher is an
       9             :  * always-running process, started by the postmaster when the autovacuum GUC
      10             :  * parameter is set.  The launcher schedules autovacuum workers to be started
      11             :  * when appropriate.  The workers are the processes which execute the actual
      12             :  * vacuuming; they connect to a database as determined in the launcher, and
      13             :  * once connected they examine the catalogs to select the tables to vacuum.
      14             :  *
      15             :  * The autovacuum launcher cannot start the worker processes by itself,
      16             :  * because doing so would cause robustness issues (namely, failure to shut
      17             :  * them down on exceptional conditions, and also, since the launcher is
      18             :  * connected to shared memory and is thus subject to corruption there, it is
      19             :  * not as robust as the postmaster).  So it leaves that task to the postmaster.
      20             :  *
      21             :  * There is an autovacuum shared memory area, where the launcher stores
      22             :  * information about the database it wants vacuumed.  When it wants a new
      23             :  * worker to start, it sets a flag in shared memory and sends a signal to the
      24             :  * postmaster.  Then postmaster knows nothing more than it must start a worker;
      25             :  * so it forks a new child, which turns into a worker.  This new process
      26             :  * connects to shared memory, and there it can inspect the information that the
      27             :  * launcher has set up.
      28             :  *
      29             :  * If the fork() call fails in the postmaster, it sets a flag in the shared
      30             :  * memory area, and sends a signal to the launcher.  The launcher, upon
      31             :  * noticing the flag, can try starting the worker again by resending the
      32             :  * signal.  Note that the failure can only be transient (fork failure due to
      33             :  * high load, memory pressure, too many processes, etc); more permanent
      34             :  * problems, like failure to connect to a database, are detected later in the
      35             :  * worker and dealt with just by having the worker exit normally.  The launcher
      36             :  * will launch a new worker again later, per schedule.
      37             :  *
      38             :  * When the worker is done vacuuming it sends SIGUSR2 to the launcher.  The
      39             :  * launcher then wakes up and is able to launch another worker, if the schedule
      40             :  * is so tight that a new worker is needed immediately.  At this time the
      41             :  * launcher can also balance the settings for the various remaining workers'
      42             :  * cost-based vacuum delay feature.
      43             :  *
      44             :  * Note that there can be more than one worker in a database concurrently.
      45             :  * They will store the table they are currently vacuuming in shared memory, so
      46             :  * that other workers avoid being blocked waiting for the vacuum lock for that
      47             :  * table.  They will also fetch the last time the table was vacuumed from
      48             :  * pgstats just before vacuuming each table, to avoid vacuuming a table that
      49             :  * was just finished being vacuumed by another worker and thus is no longer
      50             :  * noted in shared memory.  However, there is a small window (due to not yet
      51             :  * holding the relation lock) during which a worker may choose a table that was
      52             :  * already vacuumed; this is a bug in the current design.
      53             :  *
      54             :  * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
      55             :  * Portions Copyright (c) 1994, Regents of the University of California
      56             :  *
      57             :  *
      58             :  * IDENTIFICATION
      59             :  *    src/backend/postmaster/autovacuum.c
      60             :  *
      61             :  *-------------------------------------------------------------------------
      62             :  */
      63             : #include "postgres.h"
      64             : 
      65             : #include <signal.h>
      66             : #include <sys/time.h>
      67             : #include <unistd.h>
      68             : 
      69             : #include "access/heapam.h"
      70             : #include "access/htup_details.h"
      71             : #include "access/multixact.h"
      72             : #include "access/reloptions.h"
      73             : #include "access/tableam.h"
      74             : #include "access/transam.h"
      75             : #include "access/xact.h"
      76             : #include "catalog/dependency.h"
      77             : #include "catalog/namespace.h"
      78             : #include "catalog/pg_database.h"
      79             : #include "catalog/pg_namespace.h"
      80             : #include "commands/vacuum.h"
      81             : #include "common/int.h"
      82             : #include "lib/ilist.h"
      83             : #include "libpq/pqsignal.h"
      84             : #include "miscadmin.h"
      85             : #include "nodes/makefuncs.h"
      86             : #include "pgstat.h"
      87             : #include "postmaster/autovacuum.h"
      88             : #include "postmaster/interrupt.h"
      89             : #include "postmaster/postmaster.h"
      90             : #include "storage/aio_subsys.h"
      91             : #include "storage/bufmgr.h"
      92             : #include "storage/ipc.h"
      93             : #include "storage/latch.h"
      94             : #include "storage/lmgr.h"
      95             : #include "storage/pmsignal.h"
      96             : #include "storage/proc.h"
      97             : #include "storage/procsignal.h"
      98             : #include "storage/smgr.h"
      99             : #include "tcop/tcopprot.h"
     100             : #include "utils/fmgroids.h"
     101             : #include "utils/fmgrprotos.h"
     102             : #include "utils/guc_hooks.h"
     103             : #include "utils/injection_point.h"
     104             : #include "utils/lsyscache.h"
     105             : #include "utils/memutils.h"
     106             : #include "utils/ps_status.h"
     107             : #include "utils/rel.h"
     108             : #include "utils/snapmgr.h"
     109             : #include "utils/syscache.h"
     110             : #include "utils/timeout.h"
     111             : #include "utils/timestamp.h"
     112             : 
     113             : 
     114             : /*
     115             :  * GUC parameters
     116             :  */
     117             : bool        autovacuum_start_daemon = false;
     118             : int         autovacuum_worker_slots;
     119             : int         autovacuum_max_workers;
     120             : int         autovacuum_work_mem = -1;
     121             : int         autovacuum_naptime;
     122             : int         autovacuum_vac_thresh;
     123             : int         autovacuum_vac_max_thresh;
     124             : double      autovacuum_vac_scale;
     125             : int         autovacuum_vac_ins_thresh;
     126             : double      autovacuum_vac_ins_scale;
     127             : int         autovacuum_anl_thresh;
     128             : double      autovacuum_anl_scale;
     129             : int         autovacuum_freeze_max_age;
     130             : int         autovacuum_multixact_freeze_max_age;
     131             : 
     132             : double      autovacuum_vac_cost_delay;
     133             : int         autovacuum_vac_cost_limit;
     134             : 
     135             : int         Log_autovacuum_min_duration = 600000;
     136             : int         Log_autoanalyze_min_duration = 600000;
     137             : 
     138             : /* the minimum allowed time between two awakenings of the launcher */
     139             : #define MIN_AUTOVAC_SLEEPTIME 100.0 /* milliseconds */
     140             : #define MAX_AUTOVAC_SLEEPTIME 300   /* seconds */
     141             : 
     142             : /*
     143             :  * Variables to save the cost-related storage parameters for the current
     144             :  * relation being vacuumed by this autovacuum worker. Using these, we can
     145             :  * ensure we don't overwrite the values of vacuum_cost_delay and
     146             :  * vacuum_cost_limit after reloading the configuration file. They are
     147             :  * initialized to "invalid" values to indicate that no cost-related storage
     148             :  * parameters were specified and will be set in do_autovacuum() after checking
     149             :  * the storage parameters in table_recheck_autovac().
     150             :  */
     151             : static double av_storage_param_cost_delay = -1;
     152             : static int  av_storage_param_cost_limit = -1;
     153             : 
     154             : /* Flags set by signal handlers */
     155             : static volatile sig_atomic_t got_SIGUSR2 = false;
     156             : 
     157             : /* Comparison points for determining whether freeze_max_age is exceeded */
     158             : static TransactionId recentXid;
     159             : static MultiXactId recentMulti;
     160             : 
     161             : /* Default freeze ages to use for autovacuum (varies by database) */
     162             : static int  default_freeze_min_age;
     163             : static int  default_freeze_table_age;
     164             : static int  default_multixact_freeze_min_age;
     165             : static int  default_multixact_freeze_table_age;
     166             : 
     167             : /* Memory context for long-lived data */
     168             : static MemoryContext AutovacMemCxt;
     169             : 
     170             : /* struct to keep track of databases in launcher */
     171             : typedef struct avl_dbase
     172             : {
     173             :     Oid         adl_datid;      /* hash key -- must be first */
     174             :     TimestampTz adl_next_worker;
     175             :     int         adl_score;
     176             :     dlist_node  adl_node;
     177             : } avl_dbase;
     178             : 
     179             : /* struct to keep track of databases in worker */
     180             : typedef struct avw_dbase
     181             : {
     182             :     Oid         adw_datid;
     183             :     char       *adw_name;
     184             :     TransactionId adw_frozenxid;
     185             :     MultiXactId adw_minmulti;
     186             :     PgStat_StatDBEntry *adw_entry;
     187             : } avw_dbase;
     188             : 
     189             : /* struct to keep track of tables to vacuum and/or analyze, in 1st pass */
     190             : typedef struct av_relation
     191             : {
     192             :     Oid         ar_toastrelid;  /* hash key - must be first */
     193             :     Oid         ar_relid;
     194             :     bool        ar_hasrelopts;
     195             :     AutoVacOpts ar_reloptions;  /* copy of AutoVacOpts from the main table's
     196             :                                  * reloptions, or NULL if none */
     197             : } av_relation;
     198             : 
     199             : /* struct to keep track of tables to vacuum and/or analyze, after rechecking */
     200             : typedef struct autovac_table
     201             : {
     202             :     Oid         at_relid;
     203             :     VacuumParams at_params;
     204             :     double      at_storage_param_vac_cost_delay;
     205             :     int         at_storage_param_vac_cost_limit;
     206             :     bool        at_dobalance;
     207             :     bool        at_sharedrel;
     208             :     char       *at_relname;
     209             :     char       *at_nspname;
     210             :     char       *at_datname;
     211             : } autovac_table;
     212             : 
     213             : /*-------------
     214             :  * This struct holds information about a single worker's whereabouts.  We keep
     215             :  * an array of these in shared memory, sized according to
     216             :  * autovacuum_worker_slots.
     217             :  *
     218             :  * wi_links     entry into free list or running list
     219             :  * wi_dboid     OID of the database this worker is supposed to work on
     220             :  * wi_tableoid  OID of the table currently being vacuumed, if any
     221             :  * wi_sharedrel flag indicating whether table is marked relisshared
     222             :  * wi_proc      pointer to PGPROC of the running worker, NULL if not started
     223             :  * wi_launchtime Time at which this worker was launched
     224             :  * wi_dobalance Whether this worker should be included in balance calculations
     225             :  *
     226             :  * All fields are protected by AutovacuumLock, except for wi_tableoid and
     227             :  * wi_sharedrel which are protected by AutovacuumScheduleLock (note these
     228             :  * two fields are read-only for everyone except that worker itself).
     229             :  *-------------
     230             :  */
     231             : typedef struct WorkerInfoData
     232             : {
     233             :     dlist_node  wi_links;
     234             :     Oid         wi_dboid;
     235             :     Oid         wi_tableoid;
     236             :     PGPROC     *wi_proc;
     237             :     TimestampTz wi_launchtime;
     238             :     pg_atomic_flag wi_dobalance;
     239             :     bool        wi_sharedrel;
     240             : } WorkerInfoData;
     241             : 
     242             : typedef struct WorkerInfoData *WorkerInfo;
     243             : 
     244             : /*
     245             :  * Possible signals received by the launcher from remote processes.  These are
     246             :  * stored atomically in shared memory so that other processes can set them
     247             :  * without locking.
     248             :  */
     249             : typedef enum
     250             : {
     251             :     AutoVacForkFailed,          /* failed trying to start a worker */
     252             :     AutoVacRebalance,           /* rebalance the cost limits */
     253             : }           AutoVacuumSignal;
     254             : 
     255             : #define AutoVacNumSignals (AutoVacRebalance + 1)
     256             : 
     257             : /*
     258             :  * Autovacuum workitem array, stored in AutoVacuumShmem->av_workItems.  This
     259             :  * list is mostly protected by AutovacuumLock, except that if an item is
     260             :  * marked 'active' other processes must not modify the work-identifying
     261             :  * members.
     262             :  */
     263             : typedef struct AutoVacuumWorkItem
     264             : {
     265             :     AutoVacuumWorkItemType avw_type;
     266             :     bool        avw_used;       /* below data is valid */
     267             :     bool        avw_active;     /* being processed */
     268             :     Oid         avw_database;
     269             :     Oid         avw_relation;
     270             :     BlockNumber avw_blockNumber;
     271             : } AutoVacuumWorkItem;
     272             : 
     273             : #define NUM_WORKITEMS   256
     274             : 
     275             : /*-------------
     276             :  * The main autovacuum shmem struct.  On shared memory we store this main
     277             :  * struct and the array of WorkerInfo structs.  This struct keeps:
     278             :  *
     279             :  * av_signal        set by other processes to indicate various conditions
     280             :  * av_launcherpid   the PID of the autovacuum launcher
     281             :  * av_freeWorkers   the WorkerInfo freelist
     282             :  * av_runningWorkers the WorkerInfo non-free queue
     283             :  * av_startingWorker pointer to WorkerInfo currently being started (cleared by
     284             :  *                  the worker itself as soon as it's up and running)
     285             :  * av_workItems     work item array
     286             :  * av_nworkersForBalance the number of autovacuum workers to use when
     287             :  *                  calculating the per worker cost limit
     288             :  *
     289             :  * This struct is protected by AutovacuumLock, except for av_signal and parts
     290             :  * of the worker list (see above).
     291             :  *-------------
     292             :  */
     293             : typedef struct
     294             : {
     295             :     sig_atomic_t av_signal[AutoVacNumSignals];
     296             :     pid_t       av_launcherpid;
     297             :     dclist_head av_freeWorkers;
     298             :     dlist_head  av_runningWorkers;
     299             :     WorkerInfo  av_startingWorker;
     300             :     AutoVacuumWorkItem av_workItems[NUM_WORKITEMS];
     301             :     pg_atomic_uint32 av_nworkersForBalance;
     302             : } AutoVacuumShmemStruct;
     303             : 
     304             : static AutoVacuumShmemStruct *AutoVacuumShmem;
     305             : 
     306             : /*
     307             :  * the database list (of avl_dbase elements) in the launcher, and the context
     308             :  * that contains it
     309             :  */
     310             : static dlist_head DatabaseList = DLIST_STATIC_INIT(DatabaseList);
     311             : static MemoryContext DatabaseListCxt = NULL;
     312             : 
     313             : /*
     314             :  * Dummy pointer to persuade Valgrind that we've not leaked the array of
     315             :  * avl_dbase structs.  Make it global to ensure the compiler doesn't
     316             :  * optimize it away.
     317             :  */
     318             : #ifdef USE_VALGRIND
     319             : extern avl_dbase *avl_dbase_array;
     320             : avl_dbase  *avl_dbase_array;
     321             : #endif
     322             : 
     323             : /* Pointer to my own WorkerInfo, valid on each worker */
     324             : static WorkerInfo MyWorkerInfo = NULL;
     325             : 
     326             : /* PID of launcher, valid only in worker while shutting down */
     327             : int         AutovacuumLauncherPid = 0;
     328             : 
     329             : static Oid  do_start_worker(void);
     330             : static void ProcessAutoVacLauncherInterrupts(void);
     331             : pg_noreturn static void AutoVacLauncherShutdown(void);
     332             : static void launcher_determine_sleep(bool canlaunch, bool recursing,
     333             :                                      struct timeval *nap);
     334             : static void launch_worker(TimestampTz now);
     335             : static List *get_database_list(void);
     336             : static void rebuild_database_list(Oid newdb);
     337             : static int  db_comparator(const void *a, const void *b);
     338             : static void autovac_recalculate_workers_for_balance(void);
     339             : 
     340             : static void do_autovacuum(void);
     341             : static void FreeWorkerInfo(int code, Datum arg);
     342             : 
     343             : static autovac_table *table_recheck_autovac(Oid relid, HTAB *table_toast_map,
     344             :                                             TupleDesc pg_class_desc,
     345             :                                             int effective_multixact_freeze_max_age);
     346             : static void recheck_relation_needs_vacanalyze(Oid relid, AutoVacOpts *avopts,
     347             :                                               Form_pg_class classForm,
     348             :                                               int effective_multixact_freeze_max_age,
     349             :                                               bool *dovacuum, bool *doanalyze, bool *wraparound);
     350             : static void relation_needs_vacanalyze(Oid relid, AutoVacOpts *relopts,
     351             :                                       Form_pg_class classForm,
     352             :                                       PgStat_StatTabEntry *tabentry,
     353             :                                       int effective_multixact_freeze_max_age,
     354             :                                       bool *dovacuum, bool *doanalyze, bool *wraparound);
     355             : 
     356             : static void autovacuum_do_vac_analyze(autovac_table *tab,
     357             :                                       BufferAccessStrategy bstrategy);
     358             : static AutoVacOpts *extract_autovac_opts(HeapTuple tup,
     359             :                                          TupleDesc pg_class_desc);
     360             : static void perform_work_item(AutoVacuumWorkItem *workitem);
     361             : static void autovac_report_activity(autovac_table *tab);
     362             : static void autovac_report_workitem(AutoVacuumWorkItem *workitem,
     363             :                                     const char *nspname, const char *relname);
     364             : static void avl_sigusr2_handler(SIGNAL_ARGS);
     365             : static bool av_worker_available(void);
     366             : static void check_av_worker_gucs(void);
     367             : 
     368             : 
     369             : 
     370             : /********************************************************************
     371             :  *                    AUTOVACUUM LAUNCHER CODE
     372             :  ********************************************************************/
     373             : 
     374             : /*
     375             :  * Main entry point for the autovacuum launcher process.
     376             :  */
     377             : void
     378         790 : AutoVacLauncherMain(const void *startup_data, size_t startup_data_len)
     379             : {
     380             :     sigjmp_buf  local_sigjmp_buf;
     381             : 
     382             :     Assert(startup_data_len == 0);
     383             : 
     384             :     /* Release postmaster's working memory context */
     385         790 :     if (PostmasterContext)
     386             :     {
     387         790 :         MemoryContextDelete(PostmasterContext);
     388         790 :         PostmasterContext = NULL;
     389             :     }
     390             : 
     391         790 :     MyBackendType = B_AUTOVAC_LAUNCHER;
     392         790 :     init_ps_display(NULL);
     393             : 
     394         790 :     ereport(DEBUG1,
     395             :             (errmsg_internal("autovacuum launcher started")));
     396             : 
     397         790 :     if (PostAuthDelay)
     398           0 :         pg_usleep(PostAuthDelay * 1000000L);
     399             : 
     400             :     Assert(GetProcessingMode() == InitProcessing);
     401             : 
     402             :     /*
     403             :      * Set up signal handlers.  We operate on databases much like a regular
     404             :      * backend, so we use the same signal handling.  See equivalent code in
     405             :      * tcop/postgres.c.
     406             :      */
     407         790 :     pqsignal(SIGHUP, SignalHandlerForConfigReload);
     408         790 :     pqsignal(SIGINT, StatementCancelHandler);
     409         790 :     pqsignal(SIGTERM, SignalHandlerForShutdownRequest);
     410             :     /* SIGQUIT handler was already set up by InitPostmasterChild */
     411             : 
     412         790 :     InitializeTimeouts();       /* establishes SIGALRM handler */
     413             : 
     414         790 :     pqsignal(SIGPIPE, SIG_IGN);
     415         790 :     pqsignal(SIGUSR1, procsignal_sigusr1_handler);
     416         790 :     pqsignal(SIGUSR2, avl_sigusr2_handler);
     417         790 :     pqsignal(SIGFPE, FloatExceptionHandler);
     418         790 :     pqsignal(SIGCHLD, SIG_DFL);
     419             : 
     420             :     /*
     421             :      * Create a per-backend PGPROC struct in shared memory.  We must do this
     422             :      * before we can use LWLocks or access any shared memory.
     423             :      */
     424         790 :     InitProcess();
     425             : 
     426             :     /* Early initialization */
     427         790 :     BaseInit();
     428             : 
     429         790 :     InitPostgres(NULL, InvalidOid, NULL, InvalidOid, 0, NULL);
     430             : 
     431         790 :     SetProcessingMode(NormalProcessing);
     432             : 
     433             :     /*
     434             :      * Create a memory context that we will do all our work in.  We do this so
     435             :      * that we can reset the context during error recovery and thereby avoid
     436             :      * possible memory leaks.
     437             :      */
     438         790 :     AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
     439             :                                           "Autovacuum Launcher",
     440             :                                           ALLOCSET_DEFAULT_SIZES);
     441         790 :     MemoryContextSwitchTo(AutovacMemCxt);
     442             : 
     443             :     /*
     444             :      * If an exception is encountered, processing resumes here.
     445             :      *
     446             :      * This code is a stripped down version of PostgresMain error recovery.
     447             :      *
     448             :      * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
     449             :      * (to wit, BlockSig) will be restored when longjmp'ing to here.  Thus,
     450             :      * signals other than SIGQUIT will be blocked until we complete error
     451             :      * recovery.  It might seem that this policy makes the HOLD_INTERRUPTS()
     452             :      * call redundant, but it is not since InterruptPending might be set
     453             :      * already.
     454             :      */
     455         790 :     if (sigsetjmp(local_sigjmp_buf, 1) != 0)
     456             :     {
     457             :         /* since not using PG_TRY, must reset error stack by hand */
     458           0 :         error_context_stack = NULL;
     459             : 
     460             :         /* Prevents interrupts while cleaning up */
     461           0 :         HOLD_INTERRUPTS();
     462             : 
     463             :         /* Forget any pending QueryCancel or timeout request */
     464           0 :         disable_all_timeouts(false);
     465           0 :         QueryCancelPending = false; /* second to avoid race condition */
     466             : 
     467             :         /* Report the error to the server log */
     468           0 :         EmitErrorReport();
     469             : 
     470             :         /* Abort the current transaction in order to recover */
     471           0 :         AbortCurrentTransaction();
     472             : 
     473             :         /*
     474             :          * Release any other resources, for the case where we were not in a
     475             :          * transaction.
     476             :          */
     477           0 :         LWLockReleaseAll();
     478           0 :         pgstat_report_wait_end();
     479           0 :         pgaio_error_cleanup();
     480           0 :         UnlockBuffers();
     481             :         /* this is probably dead code, but let's be safe: */
     482           0 :         if (AuxProcessResourceOwner)
     483           0 :             ReleaseAuxProcessResources(false);
     484           0 :         AtEOXact_Buffers(false);
     485           0 :         AtEOXact_SMgr();
     486           0 :         AtEOXact_Files(false);
     487           0 :         AtEOXact_HashTables(false);
     488             : 
     489             :         /*
     490             :          * Now return to normal top-level context and clear ErrorContext for
     491             :          * next time.
     492             :          */
     493           0 :         MemoryContextSwitchTo(AutovacMemCxt);
     494           0 :         FlushErrorState();
     495             : 
     496             :         /* Flush any leaked data in the top-level context */
     497           0 :         MemoryContextReset(AutovacMemCxt);
     498             : 
     499             :         /* don't leave dangling pointers to freed memory */
     500           0 :         DatabaseListCxt = NULL;
     501           0 :         dlist_init(&DatabaseList);
     502             : 
     503             :         /* Now we can allow interrupts again */
     504           0 :         RESUME_INTERRUPTS();
     505             : 
     506             :         /* if in shutdown mode, no need for anything further; just go away */
     507           0 :         if (ShutdownRequestPending)
     508           0 :             AutoVacLauncherShutdown();
     509             : 
     510             :         /*
     511             :          * Sleep at least 1 second after any error.  We don't want to be
     512             :          * filling the error logs as fast as we can.
     513             :          */
     514           0 :         pg_usleep(1000000L);
     515             :     }
     516             : 
     517             :     /* We can now handle ereport(ERROR) */
     518         790 :     PG_exception_stack = &local_sigjmp_buf;
     519             : 
     520             :     /* must unblock signals before calling rebuild_database_list */
     521         790 :     sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
     522             : 
     523             :     /*
     524             :      * Set always-secure search path.  Launcher doesn't connect to a database,
     525             :      * so this has no effect.
     526             :      */
     527         790 :     SetConfigOption("search_path", "", PGC_SUSET, PGC_S_OVERRIDE);
     528             : 
     529             :     /*
     530             :      * Force zero_damaged_pages OFF in the autovac process, even if it is set
     531             :      * in postgresql.conf.  We don't really want such a dangerous option being
     532             :      * applied non-interactively.
     533             :      */
     534         790 :     SetConfigOption("zero_damaged_pages", "false", PGC_SUSET, PGC_S_OVERRIDE);
     535             : 
     536             :     /*
     537             :      * Force settable timeouts off to avoid letting these settings prevent
     538             :      * regular maintenance from being executed.
     539             :      */
     540         790 :     SetConfigOption("statement_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     541         790 :     SetConfigOption("transaction_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     542         790 :     SetConfigOption("lock_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
     543         790 :     SetConfigOption("idle_in_transaction_session_timeout", "0",
     544             :                     PGC_SUSET, PGC_S_OVERRIDE);
     545             : 
     546             :     /*
     547             :      * Force default_transaction_isolation to READ COMMITTED.  We don't want
     548             :      * to pay the overhead of serializable mode, nor add any risk of causing
     549             :      * deadlocks or delaying other transactions.
     550             :      */
     551         790 :     SetConfigOption("default_transaction_isolation", "read committed",
     552             :                     PGC_SUSET, PGC_S_OVERRIDE);
     553             : 
     554             :     /*
     555             :      * Even when system is configured to use a different fetch consistency,
     556             :      * for autovac we always want fresh stats.
     557             :      */
     558         790 :     SetConfigOption("stats_fetch_consistency", "none", PGC_SUSET, PGC_S_OVERRIDE);
     559             : 
     560             :     /*
     561             :      * In emergency mode, just start a worker (unless shutdown was requested)
     562             :      * and go away.
     563             :      */
     564         790 :     if (!AutoVacuumingActive())
     565             :     {
     566           0 :         if (!ShutdownRequestPending)
     567           0 :             do_start_worker();
     568           0 :         proc_exit(0);           /* done */
     569             :     }
     570             : 
     571         790 :     AutoVacuumShmem->av_launcherpid = MyProcPid;
     572             : 
     573             :     /*
     574             :      * Create the initial database list.  The invariant we want this list to
     575             :      * keep is that it's ordered by decreasing next_worker.  As soon as an
     576             :      * entry is updated to a higher time, it will be moved to the front (which
     577             :      * is correct because the only operation is to add autovacuum_naptime to
     578             :      * the entry, and time always increases).
     579             :      */
     580         790 :     rebuild_database_list(InvalidOid);
     581             : 
     582             :     /* loop until shutdown request */
     583       15442 :     while (!ShutdownRequestPending)
     584             :     {
     585             :         struct timeval nap;
     586       15436 :         TimestampTz current_time = 0;
     587             :         bool        can_launch;
     588             : 
     589             :         /*
     590             :          * This loop is a bit different from the normal use of WaitLatch,
     591             :          * because we'd like to sleep before the first launch of a child
     592             :          * process.  So it's WaitLatch, then ResetLatch, then check for
     593             :          * wakening conditions.
     594             :          */
     595             : 
     596       15436 :         launcher_determine_sleep(av_worker_available(), false, &nap);
     597             : 
     598             :         /*
     599             :          * Wait until naptime expires or we get some type of signal (all the
     600             :          * signal handlers will wake us by calling SetLatch).
     601             :          */
     602       15436 :         (void) WaitLatch(MyLatch,
     603             :                          WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
     604       15436 :                          (nap.tv_sec * 1000L) + (nap.tv_usec / 1000L),
     605             :                          WAIT_EVENT_AUTOVACUUM_MAIN);
     606             : 
     607       15430 :         ResetLatch(MyLatch);
     608             : 
     609       15430 :         ProcessAutoVacLauncherInterrupts();
     610             : 
     611             :         /*
     612             :          * a worker finished, or postmaster signaled failure to start a worker
     613             :          */
     614       14652 :         if (got_SIGUSR2)
     615             :         {
     616        8634 :             got_SIGUSR2 = false;
     617             : 
     618             :             /* rebalance cost limits, if needed */
     619        8634 :             if (AutoVacuumShmem->av_signal[AutoVacRebalance])
     620             :             {
     621        4314 :                 LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
     622        4314 :                 AutoVacuumShmem->av_signal[AutoVacRebalance] = false;
     623        4314 :                 autovac_recalculate_workers_for_balance();
     624        4314 :                 LWLockRelease(AutovacuumLock);
     625             :             }
     626             : 
     627        8634 :             if (AutoVacuumShmem->av_signal[AutoVacForkFailed])
     628             :             {
     629             :                 /*
     630             :                  * If the postmaster failed to start a new worker, we sleep
     631             :                  * for a little while and resend the signal.  The new worker's
     632             :                  * state is still in memory, so this is sufficient.  After
     633             :                  * that, we restart the main loop.
     634             :                  *
     635             :                  * XXX should we put a limit to the number of times we retry?
     636             :                  * I don't think it makes much sense, because a future start
     637             :                  * of a worker will continue to fail in the same way.
     638             :                  */
     639           0 :                 AutoVacuumShmem->av_signal[AutoVacForkFailed] = false;
     640           0 :                 pg_usleep(1000000L);    /* 1s */
     641           0 :                 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER);
     642        4656 :                 continue;
     643             :             }
     644             :         }
     645             : 
     646             :         /*
     647             :          * There are some conditions that we need to check before trying to
     648             :          * start a worker.  First, we need to make sure that there is a worker
     649             :          * slot available.  Second, we need to make sure that no other worker
     650             :          * failed while starting up.
     651             :          */
     652             : 
     653       14652 :         current_time = GetCurrentTimestamp();
     654       14652 :         LWLockAcquire(AutovacuumLock, LW_SHARED);
     655             : 
     656       14652 :         can_launch = av_worker_available();
     657             : 
     658       14652 :         if (AutoVacuumShmem->av_startingWorker != NULL)
     659             :         {
     660             :             int         waittime;
     661          96 :             WorkerInfo  worker = AutoVacuumShmem->av_startingWorker;
     662             : 
     663             :             /*
     664             :              * We can't launch another worker when another one is still
     665             :              * starting up (or failed while doing so), so just sleep for a bit
     666             :              * more; that worker will wake us up again as soon as it's ready.
     667             :              * We will only wait autovacuum_naptime seconds (up to a maximum
     668             :              * of 60 seconds) for this to happen however.  Note that failure
     669             :              * to connect to a particular database is not a problem here,
     670             :              * because the worker removes itself from the startingWorker
     671             :              * pointer before trying to connect.  Problems detected by the
     672             :              * postmaster (like fork() failure) are also reported and handled
     673             :              * differently.  The only problems that may cause this code to
     674             :              * fire are errors in the earlier sections of AutoVacWorkerMain,
     675             :              * before the worker removes the WorkerInfo from the
     676             :              * startingWorker pointer.
     677             :              */
     678          96 :             waittime = Min(autovacuum_naptime, 60) * 1000;
     679          96 :             if (TimestampDifferenceExceeds(worker->wi_launchtime, current_time,
     680             :                                            waittime))
     681             :             {
     682           0 :                 LWLockRelease(AutovacuumLock);
     683           0 :                 LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
     684             : 
     685             :                 /*
     686             :                  * No other process can put a worker in starting mode, so if
     687             :                  * startingWorker is still INVALID after exchanging our lock,
     688             :                  * we assume it's the same one we saw above (so we don't
     689             :                  * recheck the launch time).
     690             :                  */
     691           0 :                 if (AutoVacuumShmem->av_startingWorker != NULL)
     692             :                 {
     693           0 :                     worker = AutoVacuumShmem->av_startingWorker;
     694           0 :                     worker->wi_dboid = InvalidOid;
     695           0 :                     worker->wi_tableoid = InvalidOid;
     696           0 :                     worker->wi_sharedrel = false;
     697           0 :                     worker->wi_proc = NULL;
     698           0 :                     worker->wi_launchtime = 0;
     699           0 :                     dclist_push_head(&AutoVacuumShmem->av_freeWorkers,
     700             :                                      &worker->wi_links);
     701           0 :                     AutoVacuumShmem->av_startingWorker = NULL;
     702           0 :                     ereport(WARNING,
     703             :                             errmsg("autovacuum worker took too long to start; canceled"));
     704             :                 }
     705             :             }
     706             :             else
     707          96 :                 can_launch = false;
     708             :         }
     709       14652 :         LWLockRelease(AutovacuumLock);  /* either shared or exclusive */
     710             : 
     711             :         /* if we can't do anything, just go back to sleep */
     712       14652 :         if (!can_launch)
     713        4656 :             continue;
     714             : 
     715             :         /* We're OK to start a new worker */
     716             : 
     717        9996 :         if (dlist_is_empty(&DatabaseList))
     718             :         {
     719             :             /*
     720             :              * Special case when the list is empty: start a worker right away.
     721             :              * This covers the initial case, when no database is in pgstats
     722             :              * (thus the list is empty).  Note that the constraints in
     723             :              * launcher_determine_sleep keep us from starting workers too
     724             :              * quickly (at most once every autovacuum_naptime when the list is
     725             :              * empty).
     726             :              */
     727          12 :             launch_worker(current_time);
     728             :         }
     729             :         else
     730             :         {
     731             :             /*
     732             :              * because rebuild_database_list constructs a list with most
     733             :              * distant adl_next_worker first, we obtain our database from the
     734             :              * tail of the list.
     735             :              */
     736             :             avl_dbase  *avdb;
     737             : 
     738        9984 :             avdb = dlist_tail_element(avl_dbase, adl_node, &DatabaseList);
     739             : 
     740             :             /*
     741             :              * launch a worker if next_worker is right now or it is in the
     742             :              * past
     743             :              */
     744        9984 :             if (TimestampDifferenceExceeds(avdb->adl_next_worker,
     745             :                                            current_time, 0))
     746        4322 :                 launch_worker(current_time);
     747             :         }
     748             :     }
     749             : 
     750           6 :     AutoVacLauncherShutdown();
     751             : }
     752             : 
     753             : /*
     754             :  * Process any new interrupts.
     755             :  */
     756             : static void
     757       15430 : ProcessAutoVacLauncherInterrupts(void)
     758             : {
     759             :     /* the normal shutdown case */
     760       15430 :     if (ShutdownRequestPending)
     761         778 :         AutoVacLauncherShutdown();
     762             : 
     763       14652 :     if (ConfigReloadPending)
     764             :     {
     765          72 :         int         autovacuum_max_workers_prev = autovacuum_max_workers;
     766             : 
     767          72 :         ConfigReloadPending = false;
     768          72 :         ProcessConfigFile(PGC_SIGHUP);
     769             : 
     770             :         /* shutdown requested in config file? */
     771          72 :         if (!AutoVacuumingActive())
     772           0 :             AutoVacLauncherShutdown();
     773             : 
     774             :         /*
     775             :          * If autovacuum_max_workers changed, emit a WARNING if
     776             :          * autovacuum_worker_slots < autovacuum_max_workers.  If it didn't
     777             :          * change, skip this to avoid too many repeated log messages.
     778             :          */
     779          72 :         if (autovacuum_max_workers_prev != autovacuum_max_workers)
     780           0 :             check_av_worker_gucs();
     781             : 
     782             :         /* rebuild the list in case the naptime changed */
     783          72 :         rebuild_database_list(InvalidOid);
     784             :     }
     785             : 
     786             :     /* Process barrier events */
     787       14652 :     if (ProcSignalBarrierPending)
     788          60 :         ProcessProcSignalBarrier();
     789             : 
     790             :     /* Perform logging of memory contexts of this process */
     791       14652 :     if (LogMemoryContextPending)
     792           0 :         ProcessLogMemoryContextInterrupt();
     793             : 
     794             :     /* Process sinval catchup interrupts that happened while sleeping */
     795       14652 :     ProcessCatchupInterrupt();
     796       14652 : }
     797             : 
     798             : /*
     799             :  * Perform a normal exit from the autovac launcher.
     800             :  */
     801             : static void
     802         784 : AutoVacLauncherShutdown(void)
     803             : {
     804         784 :     ereport(DEBUG1,
     805             :             (errmsg_internal("autovacuum launcher shutting down")));
     806         784 :     AutoVacuumShmem->av_launcherpid = 0;
     807             : 
     808         784 :     proc_exit(0);               /* done */
     809             : }
     810             : 
     811             : /*
     812             :  * Determine the time to sleep, based on the database list.
     813             :  *
     814             :  * The "canlaunch" parameter indicates whether we can start a worker right now,
     815             :  * for example due to the workers being all busy.  If this is false, we will
     816             :  * cause a long sleep, which will be interrupted when a worker exits.
     817             :  */
     818             : static void
     819       15628 : launcher_determine_sleep(bool canlaunch, bool recursing, struct timeval *nap)
     820             : {
     821             :     /*
     822             :      * We sleep until the next scheduled vacuum.  We trust that when the
     823             :      * database list was built, care was taken so that no entries have times
     824             :      * in the past; if the first entry has too close a next_worker value, or a
     825             :      * time in the past, we will sleep a small nominal time.
     826             :      */
     827       15628 :     if (!canlaunch)
     828             :     {
     829        8582 :         nap->tv_sec = autovacuum_naptime;
     830        8582 :         nap->tv_usec = 0;
     831             :     }
     832        7046 :     else if (!dlist_is_empty(&DatabaseList))
     833             :     {
     834        6996 :         TimestampTz current_time = GetCurrentTimestamp();
     835             :         TimestampTz next_wakeup;
     836             :         avl_dbase  *avdb;
     837             :         long        secs;
     838             :         int         usecs;
     839             : 
     840        6996 :         avdb = dlist_tail_element(avl_dbase, adl_node, &DatabaseList);
     841             : 
     842        6996 :         next_wakeup = avdb->adl_next_worker;
     843        6996 :         TimestampDifference(current_time, next_wakeup, &secs, &usecs);
     844             : 
     845        6996 :         nap->tv_sec = secs;
     846        6996 :         nap->tv_usec = usecs;
     847             :     }
     848             :     else
     849             :     {
     850             :         /* list is empty, sleep for whole autovacuum_naptime seconds  */
     851          50 :         nap->tv_sec = autovacuum_naptime;
     852          50 :         nap->tv_usec = 0;
     853             :     }
     854             : 
     855             :     /*
     856             :      * If the result is exactly zero, it means a database had an entry with
     857             :      * time in the past.  Rebuild the list so that the databases are evenly
     858             :      * distributed again, and recalculate the time to sleep.  This can happen
     859             :      * if there are more tables needing vacuum than workers, and they all take
     860             :      * longer to vacuum than autovacuum_naptime.
     861             :      *
     862             :      * We only recurse once.  rebuild_database_list should always return times
     863             :      * in the future, but it seems best not to trust too much on that.
     864             :      */
     865       15628 :     if (nap->tv_sec == 0 && nap->tv_usec == 0 && !recursing)
     866             :     {
     867         192 :         rebuild_database_list(InvalidOid);
     868         192 :         launcher_determine_sleep(canlaunch, true, nap);
     869         192 :         return;
     870             :     }
     871             : 
     872             :     /* The smallest time we'll allow the launcher to sleep. */
     873       15436 :     if (nap->tv_sec <= 0 && nap->tv_usec <= MIN_AUTOVAC_SLEEPTIME * 1000)
     874             :     {
     875        1390 :         nap->tv_sec = 0;
     876        1390 :         nap->tv_usec = MIN_AUTOVAC_SLEEPTIME * 1000;
     877             :     }
     878             : 
     879             :     /*
     880             :      * If the sleep time is too large, clamp it to an arbitrary maximum (plus
     881             :      * any fractional seconds, for simplicity).  This avoids an essentially
     882             :      * infinite sleep in strange cases like the system clock going backwards a
     883             :      * few years.
     884             :      */
     885       15436 :     if (nap->tv_sec > MAX_AUTOVAC_SLEEPTIME)
     886          18 :         nap->tv_sec = MAX_AUTOVAC_SLEEPTIME;
     887             : }
     888             : 
     889             : /*
     890             :  * Build an updated DatabaseList.  It must only contain databases that appear
     891             :  * in pgstats, and must be sorted by next_worker from highest to lowest,
     892             :  * distributed regularly across the next autovacuum_naptime interval.
     893             :  *
     894             :  * Receives the Oid of the database that made this list be generated (we call
     895             :  * this the "new" database, because when the database was already present on
     896             :  * the list, we expect that this function is not called at all).  The
     897             :  * preexisting list, if any, will be used to preserve the order of the
     898             :  * databases in the autovacuum_naptime period.  The new database is put at the
     899             :  * end of the interval.  The actual values are not saved, which should not be
     900             :  * much of a problem.
     901             :  */
     902             : static void
     903        1084 : rebuild_database_list(Oid newdb)
     904             : {
     905             :     List       *dblist;
     906             :     ListCell   *cell;
     907             :     MemoryContext newcxt;
     908             :     MemoryContext oldcxt;
     909             :     MemoryContext tmpcxt;
     910             :     HASHCTL     hctl;
     911             :     int         score;
     912             :     int         nelems;
     913             :     HTAB       *dbhash;
     914             :     dlist_iter  iter;
     915             : 
     916        1084 :     newcxt = AllocSetContextCreate(AutovacMemCxt,
     917             :                                    "Autovacuum database list",
     918             :                                    ALLOCSET_DEFAULT_SIZES);
     919        1084 :     tmpcxt = AllocSetContextCreate(newcxt,
     920             :                                    "Autovacuum database list (tmp)",
     921             :                                    ALLOCSET_DEFAULT_SIZES);
     922        1084 :     oldcxt = MemoryContextSwitchTo(tmpcxt);
     923             : 
     924             :     /*
     925             :      * Implementing this is not as simple as it sounds, because we need to put
     926             :      * the new database at the end of the list; next the databases that were
     927             :      * already on the list, and finally (at the tail of the list) all the
     928             :      * other databases that are not on the existing list.
     929             :      *
     930             :      * To do this, we build an empty hash table of scored databases.  We will
     931             :      * start with the lowest score (zero) for the new database, then
     932             :      * increasing scores for the databases in the existing list, in order, and
     933             :      * lastly increasing scores for all databases gotten via
     934             :      * get_database_list() that are not already on the hash.
     935             :      *
     936             :      * Then we will put all the hash elements into an array, sort the array by
     937             :      * score, and finally put the array elements into the new doubly linked
     938             :      * list.
     939             :      */
     940        1084 :     hctl.keysize = sizeof(Oid);
     941        1084 :     hctl.entrysize = sizeof(avl_dbase);
     942        1084 :     hctl.hcxt = tmpcxt;
     943        1084 :     dbhash = hash_create("autovacuum db hash", 20, &hctl, /* magic number here
     944             :                                                              * FIXME */
     945             :                          HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
     946             : 
     947             :     /* start by inserting the new database */
     948        1084 :     score = 0;
     949        1084 :     if (OidIsValid(newdb))
     950             :     {
     951             :         avl_dbase  *db;
     952             :         PgStat_StatDBEntry *entry;
     953             : 
     954             :         /* only consider this database if it has a pgstat entry */
     955          30 :         entry = pgstat_fetch_stat_dbentry(newdb);
     956          30 :         if (entry != NULL)
     957             :         {
     958             :             /* we assume it isn't found because the hash was just created */
     959          24 :             db = hash_search(dbhash, &newdb, HASH_ENTER, NULL);
     960             : 
     961             :             /* hash_search already filled in the key */
     962          24 :             db->adl_score = score++;
     963             :             /* next_worker is filled in later */
     964             :         }
     965             :     }
     966             : 
     967             :     /* Now insert the databases from the existing list */
     968        1808 :     dlist_foreach(iter, &DatabaseList)
     969             :     {
     970         724 :         avl_dbase  *avdb = dlist_container(avl_dbase, adl_node, iter.cur);
     971             :         avl_dbase  *db;
     972             :         bool        found;
     973             :         PgStat_StatDBEntry *entry;
     974             : 
     975             :         /*
     976             :          * skip databases with no stat entries -- in particular, this gets rid
     977             :          * of dropped databases
     978             :          */
     979         724 :         entry = pgstat_fetch_stat_dbentry(avdb->adl_datid);
     980         724 :         if (entry == NULL)
     981           0 :             continue;
     982             : 
     983         724 :         db = hash_search(dbhash, &(avdb->adl_datid), HASH_ENTER, &found);
     984             : 
     985         724 :         if (!found)
     986             :         {
     987             :             /* hash_search already filled in the key */
     988         724 :             db->adl_score = score++;
     989             :             /* next_worker is filled in later */
     990             :         }
     991             :     }
     992             : 
     993             :     /* finally, insert all qualifying databases not previously inserted */
     994        1084 :     dblist = get_database_list();
     995        4656 :     foreach(cell, dblist)
     996             :     {
     997        3572 :         avw_dbase  *avdb = lfirst(cell);
     998             :         avl_dbase  *db;
     999             :         bool        found;
    1000             :         PgStat_StatDBEntry *entry;
    1001             : 
    1002             :         /* only consider databases with a pgstat entry */
    1003        3572 :         entry = pgstat_fetch_stat_dbentry(avdb->adw_datid);
    1004        3572 :         if (entry == NULL)
    1005        1704 :             continue;
    1006             : 
    1007        1868 :         db = hash_search(dbhash, &(avdb->adw_datid), HASH_ENTER, &found);
    1008             :         /* only update the score if the database was not already on the hash */
    1009        1868 :         if (!found)
    1010             :         {
    1011             :             /* hash_search already filled in the key */
    1012        1120 :             db->adl_score = score++;
    1013             :             /* next_worker is filled in later */
    1014             :         }
    1015             :     }
    1016        1084 :     nelems = score;
    1017             : 
    1018             :     /* from here on, the allocated memory belongs to the new list */
    1019        1084 :     MemoryContextSwitchTo(newcxt);
    1020        1084 :     dlist_init(&DatabaseList);
    1021             : 
    1022        1084 :     if (nelems > 0)
    1023             :     {
    1024             :         TimestampTz current_time;
    1025             :         int         millis_increment;
    1026             :         avl_dbase  *dbary;
    1027             :         avl_dbase  *db;
    1028             :         HASH_SEQ_STATUS seq;
    1029             :         int         i;
    1030             : 
    1031             :         /* put all the hash elements into an array */
    1032        1046 :         dbary = palloc(nelems * sizeof(avl_dbase));
    1033             :         /* keep Valgrind quiet */
    1034             : #ifdef USE_VALGRIND
    1035             :         avl_dbase_array = dbary;
    1036             : #endif
    1037             : 
    1038        1046 :         i = 0;
    1039        1046 :         hash_seq_init(&seq, dbhash);
    1040        2914 :         while ((db = hash_seq_search(&seq)) != NULL)
    1041        1868 :             memcpy(&(dbary[i++]), db, sizeof(avl_dbase));
    1042             : 
    1043             :         /* sort the array */
    1044        1046 :         qsort(dbary, nelems, sizeof(avl_dbase), db_comparator);
    1045             : 
    1046             :         /*
    1047             :          * Determine the time interval between databases in the schedule. If
    1048             :          * we see that the configured naptime would take us to sleep times
    1049             :          * lower than our min sleep time (which launcher_determine_sleep is
    1050             :          * coded not to allow), silently use a larger naptime (but don't touch
    1051             :          * the GUC variable).
    1052             :          */
    1053        1046 :         millis_increment = 1000.0 * autovacuum_naptime / nelems;
    1054        1046 :         if (millis_increment <= MIN_AUTOVAC_SLEEPTIME)
    1055           0 :             millis_increment = MIN_AUTOVAC_SLEEPTIME * 1.1;
    1056             : 
    1057        1046 :         current_time = GetCurrentTimestamp();
    1058             : 
    1059             :         /*
    1060             :          * move the elements from the array into the dlist, setting the
    1061             :          * next_worker while walking the array
    1062             :          */
    1063        2914 :         for (i = 0; i < nelems; i++)
    1064             :         {
    1065        1868 :             db = &(dbary[i]);
    1066             : 
    1067        1868 :             current_time = TimestampTzPlusMilliseconds(current_time,
    1068             :                                                        millis_increment);
    1069        1868 :             db->adl_next_worker = current_time;
    1070             : 
    1071             :             /* later elements should go closer to the head of the list */
    1072        1868 :             dlist_push_head(&DatabaseList, &db->adl_node);
    1073             :         }
    1074             :     }
    1075             : 
    1076             :     /* all done, clean up memory */
    1077        1084 :     if (DatabaseListCxt != NULL)
    1078         294 :         MemoryContextDelete(DatabaseListCxt);
    1079        1084 :     MemoryContextDelete(tmpcxt);
    1080        1084 :     DatabaseListCxt = newcxt;
    1081        1084 :     MemoryContextSwitchTo(oldcxt);
    1082        1084 : }
    1083             : 
    1084             : /* qsort comparator for avl_dbase, using adl_score */
    1085             : static int
    1086        1178 : db_comparator(const void *a, const void *b)
    1087             : {
    1088        2356 :     return pg_cmp_s32(((const avl_dbase *) a)->adl_score,
    1089        1178 :                       ((const avl_dbase *) b)->adl_score);
    1090             : }
    1091             : 
    1092             : /*
    1093             :  * do_start_worker
    1094             :  *
    1095             :  * Bare-bones procedure for starting an autovacuum worker from the launcher.
    1096             :  * It determines what database to work on, sets up shared memory stuff and
    1097             :  * signals postmaster to start the worker.  It fails gracefully if invoked when
    1098             :  * autovacuum_workers are already active.
    1099             :  *
    1100             :  * Return value is the OID of the database that the worker is going to process,
    1101             :  * or InvalidOid if no worker was actually started.
    1102             :  */
    1103             : static Oid
    1104        4334 : do_start_worker(void)
    1105             : {
    1106             :     List       *dblist;
    1107             :     ListCell   *cell;
    1108             :     TransactionId xidForceLimit;
    1109             :     MultiXactId multiForceLimit;
    1110             :     bool        for_xid_wrap;
    1111             :     bool        for_multi_wrap;
    1112             :     avw_dbase  *avdb;
    1113             :     TimestampTz current_time;
    1114        4334 :     bool        skipit = false;
    1115        4334 :     Oid         retval = InvalidOid;
    1116             :     MemoryContext tmpcxt,
    1117             :                 oldcxt;
    1118             : 
    1119             :     /* return quickly when there are no free workers */
    1120        4334 :     LWLockAcquire(AutovacuumLock, LW_SHARED);
    1121        4334 :     if (!av_worker_available())
    1122             :     {
    1123           0 :         LWLockRelease(AutovacuumLock);
    1124           0 :         return InvalidOid;
    1125             :     }
    1126        4334 :     LWLockRelease(AutovacuumLock);
    1127             : 
    1128             :     /*
    1129             :      * Create and switch to a temporary context to avoid leaking the memory
    1130             :      * allocated for the database list.
    1131             :      */
    1132        4334 :     tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
    1133             :                                    "Autovacuum start worker (tmp)",
    1134             :                                    ALLOCSET_DEFAULT_SIZES);
    1135        4334 :     oldcxt = MemoryContextSwitchTo(tmpcxt);
    1136             : 
    1137             :     /* Get a list of databases */
    1138        4334 :     dblist = get_database_list();
    1139             : 
    1140             :     /*
    1141             :      * Determine the oldest datfrozenxid/relfrozenxid that we will allow to
    1142             :      * pass without forcing a vacuum.  (This limit can be tightened for
    1143             :      * particular tables, but not loosened.)
    1144             :      */
    1145        4334 :     recentXid = ReadNextTransactionId();
    1146        4334 :     xidForceLimit = recentXid - autovacuum_freeze_max_age;
    1147             :     /* ensure it's a "normal" XID, else TransactionIdPrecedes misbehaves */
    1148             :     /* this can cause the limit to go backwards by 3, but that's OK */
    1149        4334 :     if (xidForceLimit < FirstNormalTransactionId)
    1150           0 :         xidForceLimit -= FirstNormalTransactionId;
    1151             : 
    1152             :     /* Also determine the oldest datminmxid we will consider. */
    1153        4334 :     recentMulti = ReadNextMultiXactId();
    1154        4334 :     multiForceLimit = recentMulti - MultiXactMemberFreezeThreshold();
    1155        4334 :     if (multiForceLimit < FirstMultiXactId)
    1156           0 :         multiForceLimit -= FirstMultiXactId;
    1157             : 
    1158             :     /*
    1159             :      * Choose a database to connect to.  We pick the database that was least
    1160             :      * recently auto-vacuumed, or one that needs vacuuming to prevent Xid
    1161             :      * wraparound-related data loss.  If any db at risk of Xid wraparound is
    1162             :      * found, we pick the one with oldest datfrozenxid, independently of
    1163             :      * autovacuum times; similarly we pick the one with the oldest datminmxid
    1164             :      * if any is in MultiXactId wraparound.  Note that those in Xid wraparound
    1165             :      * danger are given more priority than those in multi wraparound danger.
    1166             :      *
    1167             :      * Note that a database with no stats entry is not considered, except for
    1168             :      * Xid wraparound purposes.  The theory is that if no one has ever
    1169             :      * connected to it since the stats were last initialized, it doesn't need
    1170             :      * vacuuming.
    1171             :      *
    1172             :      * XXX This could be improved if we had more info about whether it needs
    1173             :      * vacuuming before connecting to it.  Perhaps look through the pgstats
    1174             :      * data for the database's tables?  One idea is to keep track of the
    1175             :      * number of new and dead tuples per database in pgstats.  However it
    1176             :      * isn't clear how to construct a metric that measures that and not cause
    1177             :      * starvation for less busy databases.
    1178             :      */
    1179        4334 :     avdb = NULL;
    1180        4334 :     for_xid_wrap = false;
    1181        4334 :     for_multi_wrap = false;
    1182        4334 :     current_time = GetCurrentTimestamp();
    1183       17442 :     foreach(cell, dblist)
    1184             :     {
    1185       13108 :         avw_dbase  *tmp = lfirst(cell);
    1186             :         dlist_iter  iter;
    1187             : 
    1188             :         /* Check to see if this one is at risk of wraparound */
    1189       13108 :         if (TransactionIdPrecedes(tmp->adw_frozenxid, xidForceLimit))
    1190             :         {
    1191        9916 :             if (avdb == NULL ||
    1192        3902 :                 TransactionIdPrecedes(tmp->adw_frozenxid,
    1193             :                                       avdb->adw_frozenxid))
    1194        2210 :                 avdb = tmp;
    1195        6014 :             for_xid_wrap = true;
    1196       10390 :             continue;
    1197             :         }
    1198        7094 :         else if (for_xid_wrap)
    1199         210 :             continue;           /* ignore not-at-risk DBs */
    1200        6884 :         else if (MultiXactIdPrecedes(tmp->adw_minmulti, multiForceLimit))
    1201             :         {
    1202           0 :             if (avdb == NULL ||
    1203           0 :                 MultiXactIdPrecedes(tmp->adw_minmulti, avdb->adw_minmulti))
    1204           0 :                 avdb = tmp;
    1205           0 :             for_multi_wrap = true;
    1206           0 :             continue;
    1207             :         }
    1208        6884 :         else if (for_multi_wrap)
    1209           0 :             continue;           /* ignore not-at-risk DBs */
    1210             : 
    1211             :         /* Find pgstat entry if any */
    1212        6884 :         tmp->adw_entry = pgstat_fetch_stat_dbentry(tmp->adw_datid);
    1213             : 
    1214             :         /*
    1215             :          * Skip a database with no pgstat entry; it means it hasn't seen any
    1216             :          * activity.
    1217             :          */
    1218        6884 :         if (!tmp->adw_entry)
    1219         178 :             continue;
    1220             : 
    1221             :         /*
    1222             :          * Also, skip a database that appears on the database list as having
    1223             :          * been processed recently (less than autovacuum_naptime seconds ago).
    1224             :          * We do this so that we don't select a database which we just
    1225             :          * selected, but that pgstat hasn't gotten around to updating the last
    1226             :          * autovacuum time yet.
    1227             :          */
    1228        6706 :         skipit = false;
    1229             : 
    1230       13546 :         dlist_reverse_foreach(iter, &DatabaseList)
    1231             :         {
    1232       13496 :             avl_dbase  *dbp = dlist_container(avl_dbase, adl_node, iter.cur);
    1233             : 
    1234       13496 :             if (dbp->adl_datid == tmp->adw_datid)
    1235             :             {
    1236             :                 /*
    1237             :                  * Skip this database if its next_worker value falls between
    1238             :                  * the current time and the current time plus naptime.
    1239             :                  */
    1240        6656 :                 if (!TimestampDifferenceExceeds(dbp->adl_next_worker,
    1241        3988 :                                                 current_time, 0) &&
    1242        3988 :                     !TimestampDifferenceExceeds(current_time,
    1243             :                                                 dbp->adl_next_worker,
    1244             :                                                 autovacuum_naptime * 1000))
    1245        3988 :                     skipit = true;
    1246             : 
    1247        6656 :                 break;
    1248             :             }
    1249             :         }
    1250        6706 :         if (skipit)
    1251        3988 :             continue;
    1252             : 
    1253             :         /*
    1254             :          * Remember the db with oldest autovac time.  (If we are here, both
    1255             :          * tmp->entry and db->entry must be non-null.)
    1256             :          */
    1257        2718 :         if (avdb == NULL ||
    1258         508 :             tmp->adw_entry->last_autovac_time < avdb->adw_entry->last_autovac_time)
    1259        2380 :             avdb = tmp;
    1260             :     }
    1261             : 
    1262             :     /* Found a database -- process it */
    1263        4334 :     if (avdb != NULL)
    1264             :     {
    1265             :         WorkerInfo  worker;
    1266             :         dlist_node *wptr;
    1267             : 
    1268        4322 :         LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
    1269             : 
    1270             :         /*
    1271             :          * Get a worker entry from the freelist.  We checked above, so there
    1272             :          * really should be a free slot.
    1273             :          */
    1274        4322 :         wptr = dclist_pop_head_node(&AutoVacuumShmem->av_freeWorkers);
    1275             : 
    1276        4322 :         worker = dlist_container(WorkerInfoData, wi_links, wptr);
    1277        4322 :         worker->wi_dboid = avdb->adw_datid;
    1278        4322 :         worker->wi_proc = NULL;
    1279        4322 :         worker->wi_launchtime = GetCurrentTimestamp();
    1280             : 
    1281        4322 :         AutoVacuumShmem->av_startingWorker = worker;
    1282             : 
    1283        4322 :         LWLockRelease(AutovacuumLock);
    1284             : 
    1285        4322 :         SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER);
    1286             : 
    1287        4322 :         retval = avdb->adw_datid;
    1288             :     }
    1289          12 :     else if (skipit)
    1290             :     {
    1291             :         /*
    1292             :          * If we skipped all databases on the list, rebuild it, because it
    1293             :          * probably contains a dropped database.
    1294             :          */
    1295           0 :         rebuild_database_list(InvalidOid);
    1296             :     }
    1297             : 
    1298        4334 :     MemoryContextSwitchTo(oldcxt);
    1299        4334 :     MemoryContextDelete(tmpcxt);
    1300             : 
    1301        4334 :     return retval;
    1302             : }
    1303             : 
    1304             : /*
    1305             :  * launch_worker
    1306             :  *
    1307             :  * Wrapper for starting a worker from the launcher.  Besides actually starting
    1308             :  * it, update the database list to reflect the next time that another one will
    1309             :  * need to be started on the selected database.  The actual database choice is
    1310             :  * left to do_start_worker.
    1311             :  *
    1312             :  * This routine is also expected to insert an entry into the database list if
    1313             :  * the selected database was previously absent from the list.
    1314             :  */
    1315             : static void
    1316        4334 : launch_worker(TimestampTz now)
    1317             : {
    1318             :     Oid         dbid;
    1319             :     dlist_iter  iter;
    1320             : 
    1321        4334 :     dbid = do_start_worker();
    1322        4334 :     if (OidIsValid(dbid))
    1323             :     {
    1324        4322 :         bool        found = false;
    1325             : 
    1326             :         /*
    1327             :          * Walk the database list and update the corresponding entry.  If the
    1328             :          * database is not on the list, we'll recreate the list.
    1329             :          */
    1330        9502 :         dlist_foreach(iter, &DatabaseList)
    1331             :         {
    1332        9472 :             avl_dbase  *avdb = dlist_container(avl_dbase, adl_node, iter.cur);
    1333             : 
    1334        9472 :             if (avdb->adl_datid == dbid)
    1335             :             {
    1336        4292 :                 found = true;
    1337             : 
    1338             :                 /*
    1339             :                  * add autovacuum_naptime seconds to the current time, and use
    1340             :                  * that as the new "next_worker" field for this database.
    1341             :                  */
    1342        4292 :                 avdb->adl_next_worker =
    1343        4292 :                     TimestampTzPlusMilliseconds(now, autovacuum_naptime * 1000);
    1344             : 
    1345        4292 :                 dlist_move_head(&DatabaseList, iter.cur);
    1346        4292 :                 break;
    1347             :             }
    1348             :         }
    1349             : 
    1350             :         /*
    1351             :          * If the database was not present in the database list, we rebuild
    1352             :          * the list.  It's possible that the database does not get into the
    1353             :          * list anyway, for example if it's a database that doesn't have a
    1354             :          * pgstat entry, but this is not a problem because we don't want to
    1355             :          * schedule workers regularly into those in any case.
    1356             :          */
    1357        4322 :         if (!found)
    1358          30 :             rebuild_database_list(dbid);
    1359             :     }
    1360        4334 : }
    1361             : 
    1362             : /*
    1363             :  * Called from postmaster to signal a failure to fork a process to become
    1364             :  * worker.  The postmaster should kill(SIGUSR2) the launcher shortly
    1365             :  * after calling this function.
    1366             :  */
    1367             : void
    1368           0 : AutoVacWorkerFailed(void)
    1369             : {
    1370           0 :     AutoVacuumShmem->av_signal[AutoVacForkFailed] = true;
    1371           0 : }
    1372             : 
    1373             : /* SIGUSR2: a worker is up and running, or just finished, or failed to fork */
    1374             : static void
    1375        8636 : avl_sigusr2_handler(SIGNAL_ARGS)
    1376             : {
    1377        8636 :     got_SIGUSR2 = true;
    1378        8636 :     SetLatch(MyLatch);
    1379        8636 : }
    1380             : 
    1381             : 
    1382             : /********************************************************************
    1383             :  *                    AUTOVACUUM WORKER CODE
    1384             :  ********************************************************************/
    1385             : 
    1386             : /*
    1387             :  * Main entry point for autovacuum worker processes.
    1388             :  */
    1389             : void
    1390        4324 : AutoVacWorkerMain(const void *startup_data, size_t startup_data_len)
    1391             : {
    1392             :     sigjmp_buf  local_sigjmp_buf;
    1393             :     Oid         dbid;
    1394             : 
    1395             :     Assert(startup_data_len == 0);
    1396             : 
    1397             :     /* Release postmaster's working memory context */
    1398        4324 :     if (PostmasterContext)
    1399             :     {
    1400        4324 :         MemoryContextDelete(PostmasterContext);
    1401        4324 :         PostmasterContext = NULL;
    1402             :     }
    1403             : 
    1404        4324 :     MyBackendType = B_AUTOVAC_WORKER;
    1405        4324 :     init_ps_display(NULL);
    1406             : 
    1407             :     Assert(GetProcessingMode() == InitProcessing);
    1408             : 
    1409             :     /*
    1410             :      * Set up signal handlers.  We operate on databases much like a regular
    1411             :      * backend, so we use the same signal handling.  See equivalent code in
    1412             :      * tcop/postgres.c.
    1413             :      */
    1414        4324 :     pqsignal(SIGHUP, SignalHandlerForConfigReload);
    1415             : 
    1416             :     /*
    1417             :      * SIGINT is used to signal canceling the current table's vacuum; SIGTERM
    1418             :      * means abort and exit cleanly, and SIGQUIT means abandon ship.
    1419             :      */
    1420        4324 :     pqsignal(SIGINT, StatementCancelHandler);
    1421        4324 :     pqsignal(SIGTERM, die);
    1422             :     /* SIGQUIT handler was already set up by InitPostmasterChild */
    1423             : 
    1424        4324 :     InitializeTimeouts();       /* establishes SIGALRM handler */
    1425             : 
    1426        4324 :     pqsignal(SIGPIPE, SIG_IGN);
    1427        4324 :     pqsignal(SIGUSR1, procsignal_sigusr1_handler);
    1428        4324 :     pqsignal(SIGUSR2, SIG_IGN);
    1429        4324 :     pqsignal(SIGFPE, FloatExceptionHandler);
    1430        4324 :     pqsignal(SIGCHLD, SIG_DFL);
    1431             : 
    1432             :     /*
    1433             :      * Create a per-backend PGPROC struct in shared memory.  We must do this
    1434             :      * before we can use LWLocks or access any shared memory.
    1435             :      */
    1436        4324 :     InitProcess();
    1437             : 
    1438             :     /* Early initialization */
    1439        4324 :     BaseInit();
    1440             : 
    1441             :     /*
    1442             :      * If an exception is encountered, processing resumes here.
    1443             :      *
    1444             :      * Unlike most auxiliary processes, we don't attempt to continue
    1445             :      * processing after an error; we just clean up and exit.  The autovac
    1446             :      * launcher is responsible for spawning another worker later.
    1447             :      *
    1448             :      * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
    1449             :      * (to wit, BlockSig) will be restored when longjmp'ing to here.  Thus,
    1450             :      * signals other than SIGQUIT will be blocked until we exit.  It might
    1451             :      * seem that this policy makes the HOLD_INTERRUPTS() call redundant, but
    1452             :      * it is not since InterruptPending might be set already.
    1453             :      */
    1454        4324 :     if (sigsetjmp(local_sigjmp_buf, 1) != 0)
    1455             :     {
    1456             :         /* since not using PG_TRY, must reset error stack by hand */
    1457           0 :         error_context_stack = NULL;
    1458             : 
    1459             :         /* Prevents interrupts while cleaning up */
    1460           0 :         HOLD_INTERRUPTS();
    1461             : 
    1462             :         /* Report the error to the server log */
    1463           0 :         EmitErrorReport();
    1464             : 
    1465             :         /*
    1466             :          * We can now go away.  Note that because we called InitProcess, a
    1467             :          * callback was registered to do ProcKill, which will clean up
    1468             :          * necessary state.
    1469             :          */
    1470           0 :         proc_exit(0);
    1471             :     }
    1472             : 
    1473             :     /* We can now handle ereport(ERROR) */
    1474        4324 :     PG_exception_stack = &local_sigjmp_buf;
    1475             : 
    1476        4324 :     sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
    1477             : 
    1478             :     /*
    1479             :      * Set always-secure search path, so malicious users can't redirect user
    1480             :      * code (e.g. pg_index.indexprs).  (That code runs in a
    1481             :      * SECURITY_RESTRICTED_OPERATION sandbox, so malicious users could not
    1482             :      * take control of the entire autovacuum worker in any case.)
    1483             :      */
    1484        4324 :     SetConfigOption("search_path", "", PGC_SUSET, PGC_S_OVERRIDE);
    1485             : 
    1486             :     /*
    1487             :      * Force zero_damaged_pages OFF in the autovac process, even if it is set
    1488             :      * in postgresql.conf.  We don't really want such a dangerous option being
    1489             :      * applied non-interactively.
    1490             :      */
    1491        4324 :     SetConfigOption("zero_damaged_pages", "false", PGC_SUSET, PGC_S_OVERRIDE);
    1492             : 
    1493             :     /*
    1494             :      * Force settable timeouts off to avoid letting these settings prevent
    1495             :      * regular maintenance from being executed.
    1496             :      */
    1497        4324 :     SetConfigOption("statement_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
    1498        4324 :     SetConfigOption("transaction_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
    1499        4324 :     SetConfigOption("lock_timeout", "0", PGC_SUSET, PGC_S_OVERRIDE);
    1500        4324 :     SetConfigOption("idle_in_transaction_session_timeout", "0",
    1501             :                     PGC_SUSET, PGC_S_OVERRIDE);
    1502             : 
    1503             :     /*
    1504             :      * Force default_transaction_isolation to READ COMMITTED.  We don't want
    1505             :      * to pay the overhead of serializable mode, nor add any risk of causing
    1506             :      * deadlocks or delaying other transactions.
    1507             :      */
    1508        4324 :     SetConfigOption("default_transaction_isolation", "read committed",
    1509             :                     PGC_SUSET, PGC_S_OVERRIDE);
    1510             : 
    1511             :     /*
    1512             :      * Force synchronous replication off to allow regular maintenance even if
    1513             :      * we are waiting for standbys to connect. This is important to ensure we
    1514             :      * aren't blocked from performing anti-wraparound tasks.
    1515             :      */
    1516        4324 :     if (synchronous_commit > SYNCHRONOUS_COMMIT_LOCAL_FLUSH)
    1517        4324 :         SetConfigOption("synchronous_commit", "local",
    1518             :                         PGC_SUSET, PGC_S_OVERRIDE);
    1519             : 
    1520             :     /*
    1521             :      * Even when system is configured to use a different fetch consistency,
    1522             :      * for autovac we always want fresh stats.
    1523             :      */
    1524        4324 :     SetConfigOption("stats_fetch_consistency", "none", PGC_SUSET, PGC_S_OVERRIDE);
    1525             : 
    1526             :     /*
    1527             :      * Get the info about the database we're going to work on.
    1528             :      */
    1529        4324 :     LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
    1530             : 
    1531             :     /*
    1532             :      * beware of startingWorker being INVALID; this should normally not
    1533             :      * happen, but if a worker fails after forking and before this, the
    1534             :      * launcher might have decided to remove it from the queue and start
    1535             :      * again.
    1536             :      */
    1537        4324 :     if (AutoVacuumShmem->av_startingWorker != NULL)
    1538             :     {
    1539        4324 :         MyWorkerInfo = AutoVacuumShmem->av_startingWorker;
    1540        4324 :         dbid = MyWorkerInfo->wi_dboid;
    1541        4324 :         MyWorkerInfo->wi_proc = MyProc;
    1542             : 
    1543             :         /* insert into the running list */
    1544        4324 :         dlist_push_head(&AutoVacuumShmem->av_runningWorkers,
    1545        4324 :                         &MyWorkerInfo->wi_links);
    1546             : 
    1547             :         /*
    1548             :          * remove from the "starting" pointer, so that the launcher can start
    1549             :          * a new worker if required
    1550             :          */
    1551        4324 :         AutoVacuumShmem->av_startingWorker = NULL;
    1552        4324 :         LWLockRelease(AutovacuumLock);
    1553             : 
    1554        4324 :         on_shmem_exit(FreeWorkerInfo, 0);
    1555             : 
    1556             :         /* wake up the launcher */
    1557        4324 :         if (AutoVacuumShmem->av_launcherpid != 0)
    1558        4324 :             kill(AutoVacuumShmem->av_launcherpid, SIGUSR2);
    1559             :     }
    1560             :     else
    1561             :     {
    1562             :         /* no worker entry for me, go away */
    1563           0 :         elog(WARNING, "autovacuum worker started without a worker entry");
    1564           0 :         dbid = InvalidOid;
    1565           0 :         LWLockRelease(AutovacuumLock);
    1566             :     }
    1567             : 
    1568        4324 :     if (OidIsValid(dbid))
    1569             :     {
    1570             :         char        dbname[NAMEDATALEN];
    1571             : 
    1572             :         /*
    1573             :          * Report autovac startup to the cumulative stats system.  We
    1574             :          * deliberately do this before InitPostgres, so that the
    1575             :          * last_autovac_time will get updated even if the connection attempt
    1576             :          * fails.  This is to prevent autovac from getting "stuck" repeatedly
    1577             :          * selecting an unopenable database, rather than making any progress
    1578             :          * on stuff it can connect to.
    1579             :          */
    1580        4324 :         pgstat_report_autovac(dbid);
    1581             : 
    1582             :         /*
    1583             :          * Connect to the selected database, specifying no particular user,
    1584             :          * and ignoring datallowconn.  Collect the database's name for
    1585             :          * display.
    1586             :          *
    1587             :          * Note: if we have selected a just-deleted database (due to using
    1588             :          * stale stats info), we'll fail and exit here.
    1589             :          */
    1590        4324 :         InitPostgres(NULL, dbid, NULL, InvalidOid,
    1591             :                      INIT_PG_OVERRIDE_ALLOW_CONNS,
    1592             :                      dbname);
    1593        4324 :         SetProcessingMode(NormalProcessing);
    1594        4324 :         set_ps_display(dbname);
    1595        4324 :         ereport(DEBUG1,
    1596             :                 (errmsg_internal("autovacuum: processing database \"%s\"", dbname)));
    1597             : 
    1598        4324 :         if (PostAuthDelay)
    1599           0 :             pg_usleep(PostAuthDelay * 1000000L);
    1600             : 
    1601             :         /* And do an appropriate amount of work */
    1602        4324 :         recentXid = ReadNextTransactionId();
    1603        4324 :         recentMulti = ReadNextMultiXactId();
    1604        4324 :         do_autovacuum();
    1605             :     }
    1606             : 
    1607             :     /*
    1608             :      * The launcher will be notified of my death in ProcKill, *if* we managed
    1609             :      * to get a worker slot at all
    1610             :      */
    1611             : 
    1612             :     /* All done, go away */
    1613        4318 :     proc_exit(0);
    1614             : }
    1615             : 
    1616             : /*
    1617             :  * Return a WorkerInfo to the free list
    1618             :  */
    1619             : static void
    1620        4324 : FreeWorkerInfo(int code, Datum arg)
    1621             : {
    1622        4324 :     if (MyWorkerInfo != NULL)
    1623             :     {
    1624        4324 :         LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
    1625             : 
    1626             :         /*
    1627             :          * Wake the launcher up so that he can launch a new worker immediately
    1628             :          * if required.  We only save the launcher's PID in local memory here;
    1629             :          * the actual signal will be sent when the PGPROC is recycled.  Note
    1630             :          * that we always do this, so that the launcher can rebalance the cost
    1631             :          * limit setting of the remaining workers.
    1632             :          *
    1633             :          * We somewhat ignore the risk that the launcher changes its PID
    1634             :          * between us reading it and the actual kill; we expect ProcKill to be
    1635             :          * called shortly after us, and we assume that PIDs are not reused too
    1636             :          * quickly after a process exits.
    1637             :          */
    1638        4324 :         AutovacuumLauncherPid = AutoVacuumShmem->av_launcherpid;
    1639             : 
    1640        4324 :         dlist_delete(&MyWorkerInfo->wi_links);
    1641        4324 :         MyWorkerInfo->wi_dboid = InvalidOid;
    1642        4324 :         MyWorkerInfo->wi_tableoid = InvalidOid;
    1643        4324 :         MyWorkerInfo->wi_sharedrel = false;
    1644        4324 :         MyWorkerInfo->wi_proc = NULL;
    1645        4324 :         MyWorkerInfo->wi_launchtime = 0;
    1646        4324 :         pg_atomic_clear_flag(&MyWorkerInfo->wi_dobalance);
    1647        4324 :         dclist_push_head(&AutoVacuumShmem->av_freeWorkers,
    1648        4324 :                          &MyWorkerInfo->wi_links);
    1649             :         /* not mine anymore */
    1650        4324 :         MyWorkerInfo = NULL;
    1651             : 
    1652             :         /*
    1653             :          * now that we're inactive, cause a rebalancing of the surviving
    1654             :          * workers
    1655             :          */
    1656        4324 :         AutoVacuumShmem->av_signal[AutoVacRebalance] = true;
    1657        4324 :         LWLockRelease(AutovacuumLock);
    1658             :     }
    1659        4324 : }
    1660             : 
    1661             : /*
    1662             :  * Update vacuum cost-based delay-related parameters for autovacuum workers and
    1663             :  * backends executing VACUUM or ANALYZE using the value of relevant GUCs and
    1664             :  * global state. This must be called during setup for vacuum and after every
    1665             :  * config reload to ensure up-to-date values.
    1666             :  */
    1667             : void
    1668      471276 : VacuumUpdateCosts(void)
    1669             : {
    1670      471276 :     if (MyWorkerInfo)
    1671             :     {
    1672      457564 :         if (av_storage_param_cost_delay >= 0)
    1673           0 :             vacuum_cost_delay = av_storage_param_cost_delay;
    1674      457564 :         else if (autovacuum_vac_cost_delay >= 0)
    1675      457564 :             vacuum_cost_delay = autovacuum_vac_cost_delay;
    1676             :         else
    1677             :             /* fall back to VacuumCostDelay */
    1678           0 :             vacuum_cost_delay = VacuumCostDelay;
    1679             : 
    1680      457564 :         AutoVacuumUpdateCostLimit();
    1681             :     }
    1682             :     else
    1683             :     {
    1684             :         /* Must be explicit VACUUM or ANALYZE */
    1685       13712 :         vacuum_cost_delay = VacuumCostDelay;
    1686       13712 :         vacuum_cost_limit = VacuumCostLimit;
    1687             :     }
    1688             : 
    1689             :     /*
    1690             :      * If configuration changes are allowed to impact VacuumCostActive, make
    1691             :      * sure it is updated.
    1692             :      */
    1693      471276 :     if (VacuumFailsafeActive)
    1694             :         Assert(!VacuumCostActive);
    1695      471276 :     else if (vacuum_cost_delay > 0)
    1696      457564 :         VacuumCostActive = true;
    1697             :     else
    1698             :     {
    1699       13712 :         VacuumCostActive = false;
    1700       13712 :         VacuumCostBalance = 0;
    1701             :     }
    1702             : 
    1703             :     /*
    1704             :      * Since the cost logging requires a lock, avoid rendering the log message
    1705             :      * in case we are using a message level where the log wouldn't be emitted.
    1706             :      */
    1707      471276 :     if (MyWorkerInfo && message_level_is_interesting(DEBUG2))
    1708             :     {
    1709             :         Oid         dboid,
    1710             :                     tableoid;
    1711             : 
    1712             :         Assert(!LWLockHeldByMe(AutovacuumLock));
    1713             : 
    1714           0 :         LWLockAcquire(AutovacuumLock, LW_SHARED);
    1715           0 :         dboid = MyWorkerInfo->wi_dboid;
    1716           0 :         tableoid = MyWorkerInfo->wi_tableoid;
    1717           0 :         LWLockRelease(AutovacuumLock);
    1718             : 
    1719           0 :         elog(DEBUG2,
    1720             :              "Autovacuum VacuumUpdateCosts(db=%u, rel=%u, dobalance=%s, cost_limit=%d, cost_delay=%g active=%s failsafe=%s)",
    1721             :              dboid, tableoid, pg_atomic_unlocked_test_flag(&MyWorkerInfo->wi_dobalance) ? "no" : "yes",
    1722             :              vacuum_cost_limit, vacuum_cost_delay,
    1723             :              vacuum_cost_delay > 0 ? "yes" : "no",
    1724             :              VacuumFailsafeActive ? "yes" : "no");
    1725             :     }
    1726      471276 : }
    1727             : 
    1728             : /*
    1729             :  * Update vacuum_cost_limit with the correct value for an autovacuum worker,
    1730             :  * given the value of other relevant cost limit parameters and the number of
    1731             :  * workers across which the limit must be balanced. Autovacuum workers must
    1732             :  * call this regularly in case av_nworkersForBalance has been updated by
    1733             :  * another worker or by the autovacuum launcher. They must also call it after a
    1734             :  * config reload.
    1735             :  */
    1736             : void
    1737      465818 : AutoVacuumUpdateCostLimit(void)
    1738             : {
    1739      465818 :     if (!MyWorkerInfo)
    1740           0 :         return;
    1741             : 
    1742             :     /*
    1743             :      * note: in cost_limit, zero also means use value from elsewhere, because
    1744             :      * zero is not a valid value.
    1745             :      */
    1746             : 
    1747      465818 :     if (av_storage_param_cost_limit > 0)
    1748           0 :         vacuum_cost_limit = av_storage_param_cost_limit;
    1749             :     else
    1750             :     {
    1751             :         int         nworkers_for_balance;
    1752             : 
    1753      465818 :         if (autovacuum_vac_cost_limit > 0)
    1754           0 :             vacuum_cost_limit = autovacuum_vac_cost_limit;
    1755             :         else
    1756      465818 :             vacuum_cost_limit = VacuumCostLimit;
    1757             : 
    1758             :         /* Only balance limit if no cost-related storage parameters specified */
    1759      465818 :         if (pg_atomic_unlocked_test_flag(&MyWorkerInfo->wi_dobalance))
    1760           0 :             return;
    1761             : 
    1762             :         Assert(vacuum_cost_limit > 0);
    1763             : 
    1764      465818 :         nworkers_for_balance = pg_atomic_read_u32(&AutoVacuumShmem->av_nworkersForBalance);
    1765             : 
    1766             :         /* There is at least 1 autovac worker (this worker) */
    1767      465818 :         if (nworkers_for_balance <= 0)
    1768           0 :             elog(ERROR, "nworkers_for_balance must be > 0");
    1769             : 
    1770      465818 :         vacuum_cost_limit = Max(vacuum_cost_limit / nworkers_for_balance, 1);
    1771             :     }
    1772             : }
    1773             : 
    1774             : /*
    1775             :  * autovac_recalculate_workers_for_balance
    1776             :  *      Recalculate the number of workers to consider, given cost-related
    1777             :  *      storage parameters and the current number of active workers.
    1778             :  *
    1779             :  * Caller must hold the AutovacuumLock in at least shared mode to access
    1780             :  * worker->wi_proc.
    1781             :  */
    1782             : static void
    1783      233096 : autovac_recalculate_workers_for_balance(void)
    1784             : {
    1785             :     dlist_iter  iter;
    1786             :     int         orig_nworkers_for_balance;
    1787      233096 :     int         nworkers_for_balance = 0;
    1788             : 
    1789             :     Assert(LWLockHeldByMe(AutovacuumLock));
    1790             : 
    1791      233096 :     orig_nworkers_for_balance =
    1792      233096 :         pg_atomic_read_u32(&AutoVacuumShmem->av_nworkersForBalance);
    1793             : 
    1794      727530 :     dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
    1795             :     {
    1796      494434 :         WorkerInfo  worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
    1797             : 
    1798      988868 :         if (worker->wi_proc == NULL ||
    1799      494434 :             pg_atomic_unlocked_test_flag(&worker->wi_dobalance))
    1800       10272 :             continue;
    1801             : 
    1802      484162 :         nworkers_for_balance++;
    1803             :     }
    1804             : 
    1805      233096 :     if (nworkers_for_balance != orig_nworkers_for_balance)
    1806        4518 :         pg_atomic_write_u32(&AutoVacuumShmem->av_nworkersForBalance,
    1807             :                             nworkers_for_balance);
    1808      233096 : }
    1809             : 
    1810             : /*
    1811             :  * get_database_list
    1812             :  *      Return a list of all databases found in pg_database.
    1813             :  *
    1814             :  * The list and associated data is allocated in the caller's memory context,
    1815             :  * which is in charge of ensuring that it's properly cleaned up afterwards.
    1816             :  *
    1817             :  * Note: this is the only function in which the autovacuum launcher uses a
    1818             :  * transaction.  Although we aren't attached to any particular database and
    1819             :  * therefore can't access most catalogs, we do have enough infrastructure
    1820             :  * to do a seqscan on pg_database.
    1821             :  */
    1822             : static List *
    1823        5418 : get_database_list(void)
    1824             : {
    1825        5418 :     List       *dblist = NIL;
    1826             :     Relation    rel;
    1827             :     TableScanDesc scan;
    1828             :     HeapTuple   tup;
    1829             :     MemoryContext resultcxt;
    1830             : 
    1831             :     /* This is the context that we will allocate our output data in */
    1832        5418 :     resultcxt = CurrentMemoryContext;
    1833             : 
    1834             :     /*
    1835             :      * Start a transaction so we can access pg_database.
    1836             :      */
    1837        5418 :     StartTransactionCommand();
    1838             : 
    1839        5418 :     rel = table_open(DatabaseRelationId, AccessShareLock);
    1840        5418 :     scan = table_beginscan_catalog(rel, 0, NULL);
    1841             : 
    1842       22102 :     while (HeapTupleIsValid(tup = heap_getnext(scan, ForwardScanDirection)))
    1843             :     {
    1844       16684 :         Form_pg_database pgdatabase = (Form_pg_database) GETSTRUCT(tup);
    1845             :         avw_dbase  *avdb;
    1846             :         MemoryContext oldcxt;
    1847             : 
    1848             :         /*
    1849             :          * If database has partially been dropped, we can't, nor need to,
    1850             :          * vacuum it.
    1851             :          */
    1852       16684 :         if (database_is_invalid_form(pgdatabase))
    1853             :         {
    1854           4 :             elog(DEBUG2,
    1855             :                  "autovacuum: skipping invalid database \"%s\"",
    1856             :                  NameStr(pgdatabase->datname));
    1857           4 :             continue;
    1858             :         }
    1859             : 
    1860             :         /*
    1861             :          * Allocate our results in the caller's context, not the
    1862             :          * transaction's. We do this inside the loop, and restore the original
    1863             :          * context at the end, so that leaky things like heap_getnext() are
    1864             :          * not called in a potentially long-lived context.
    1865             :          */
    1866       16680 :         oldcxt = MemoryContextSwitchTo(resultcxt);
    1867             : 
    1868       16680 :         avdb = (avw_dbase *) palloc(sizeof(avw_dbase));
    1869             : 
    1870       16680 :         avdb->adw_datid = pgdatabase->oid;
    1871       16680 :         avdb->adw_name = pstrdup(NameStr(pgdatabase->datname));
    1872       16680 :         avdb->adw_frozenxid = pgdatabase->datfrozenxid;
    1873       16680 :         avdb->adw_minmulti = pgdatabase->datminmxid;
    1874             :         /* this gets set later: */
    1875       16680 :         avdb->adw_entry = NULL;
    1876             : 
    1877       16680 :         dblist = lappend(dblist, avdb);
    1878       16680 :         MemoryContextSwitchTo(oldcxt);
    1879             :     }
    1880             : 
    1881        5418 :     table_endscan(scan);
    1882        5418 :     table_close(rel, AccessShareLock);
    1883             : 
    1884        5418 :     CommitTransactionCommand();
    1885             : 
    1886             :     /* Be sure to restore caller's memory context */
    1887        5418 :     MemoryContextSwitchTo(resultcxt);
    1888             : 
    1889        5418 :     return dblist;
    1890             : }
    1891             : 
    1892             : /*
    1893             :  * Process a database table-by-table
    1894             :  *
    1895             :  * Note that CHECK_FOR_INTERRUPTS is supposed to be used in certain spots in
    1896             :  * order not to ignore shutdown commands for too long.
    1897             :  */
    1898             : static void
    1899        4324 : do_autovacuum(void)
    1900             : {
    1901             :     Relation    classRel;
    1902             :     HeapTuple   tuple;
    1903             :     TableScanDesc relScan;
    1904             :     Form_pg_database dbForm;
    1905        4324 :     List       *table_oids = NIL;
    1906        4324 :     List       *orphan_oids = NIL;
    1907             :     HASHCTL     ctl;
    1908             :     HTAB       *table_toast_map;
    1909             :     ListCell   *volatile cell;
    1910             :     BufferAccessStrategy bstrategy;
    1911             :     ScanKeyData key;
    1912             :     TupleDesc   pg_class_desc;
    1913             :     int         effective_multixact_freeze_max_age;
    1914        4324 :     bool        did_vacuum = false;
    1915        4324 :     bool        found_concurrent_worker = false;
    1916             :     int         i;
    1917             : 
    1918             :     /*
    1919             :      * StartTransactionCommand and CommitTransactionCommand will automatically
    1920             :      * switch to other contexts.  We need this one to keep the list of
    1921             :      * relations to vacuum/analyze across transactions.
    1922             :      */
    1923        4324 :     AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
    1924             :                                           "Autovacuum worker",
    1925             :                                           ALLOCSET_DEFAULT_SIZES);
    1926        4324 :     MemoryContextSwitchTo(AutovacMemCxt);
    1927             : 
    1928             :     /* Start a transaction so our commands have one to play into. */
    1929        4324 :     StartTransactionCommand();
    1930             : 
    1931             :     /*
    1932             :      * This injection point is put in a transaction block to work with a wait
    1933             :      * that uses a condition variable.
    1934             :      */
    1935        4324 :     INJECTION_POINT("autovacuum-worker-start", NULL);
    1936             : 
    1937             :     /*
    1938             :      * Compute the multixact age for which freezing is urgent.  This is
    1939             :      * normally autovacuum_multixact_freeze_max_age, but may be less if we are
    1940             :      * short of multixact member space.
    1941             :      */
    1942        4322 :     effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
    1943             : 
    1944             :     /*
    1945             :      * Find the pg_database entry and select the default freeze ages. We use
    1946             :      * zero in template and nonconnectable databases, else the system-wide
    1947             :      * default.
    1948             :      */
    1949        4322 :     tuple = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(MyDatabaseId));
    1950        4322 :     if (!HeapTupleIsValid(tuple))
    1951           0 :         elog(ERROR, "cache lookup failed for database %u", MyDatabaseId);
    1952        4322 :     dbForm = (Form_pg_database) GETSTRUCT(tuple);
    1953             : 
    1954        4322 :     if (dbForm->datistemplate || !dbForm->datallowconn)
    1955             :     {
    1956        1654 :         default_freeze_min_age = 0;
    1957        1654 :         default_freeze_table_age = 0;
    1958        1654 :         default_multixact_freeze_min_age = 0;
    1959        1654 :         default_multixact_freeze_table_age = 0;
    1960             :     }
    1961             :     else
    1962             :     {
    1963        2668 :         default_freeze_min_age = vacuum_freeze_min_age;
    1964        2668 :         default_freeze_table_age = vacuum_freeze_table_age;
    1965        2668 :         default_multixact_freeze_min_age = vacuum_multixact_freeze_min_age;
    1966        2668 :         default_multixact_freeze_table_age = vacuum_multixact_freeze_table_age;
    1967             :     }
    1968             : 
    1969        4322 :     ReleaseSysCache(tuple);
    1970             : 
    1971             :     /* StartTransactionCommand changed elsewhere */
    1972        4322 :     MemoryContextSwitchTo(AutovacMemCxt);
    1973             : 
    1974        4322 :     classRel = table_open(RelationRelationId, AccessShareLock);
    1975             : 
    1976             :     /* create a copy so we can use it after closing pg_class */
    1977        4322 :     pg_class_desc = CreateTupleDescCopy(RelationGetDescr(classRel));
    1978             : 
    1979             :     /* create hash table for toast <-> main relid mapping */
    1980        4322 :     ctl.keysize = sizeof(Oid);
    1981        4322 :     ctl.entrysize = sizeof(av_relation);
    1982             : 
    1983        4322 :     table_toast_map = hash_create("TOAST to main relid map",
    1984             :                                   100,
    1985             :                                   &ctl,
    1986             :                                   HASH_ELEM | HASH_BLOBS);
    1987             : 
    1988             :     /*
    1989             :      * Scan pg_class to determine which tables to vacuum.
    1990             :      *
    1991             :      * We do this in two passes: on the first one we collect the list of plain
    1992             :      * relations and materialized views, and on the second one we collect
    1993             :      * TOAST tables. The reason for doing the second pass is that during it we
    1994             :      * want to use the main relation's pg_class.reloptions entry if the TOAST
    1995             :      * table does not have any, and we cannot obtain it unless we know
    1996             :      * beforehand what's the main table OID.
    1997             :      *
    1998             :      * We need to check TOAST tables separately because in cases with short,
    1999             :      * wide tables there might be proportionally much more activity in the
    2000             :      * TOAST table than in its parent.
    2001             :      */
    2002        4322 :     relScan = table_beginscan_catalog(classRel, 0, NULL);
    2003             : 
    2004             :     /*
    2005             :      * On the first pass, we collect main tables to vacuum, and also the main
    2006             :      * table relid to TOAST relid mapping.
    2007             :      */
    2008     1830606 :     while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
    2009             :     {
    2010     1826284 :         Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
    2011             :         PgStat_StatTabEntry *tabentry;
    2012             :         AutoVacOpts *relopts;
    2013             :         Oid         relid;
    2014             :         bool        dovacuum;
    2015             :         bool        doanalyze;
    2016             :         bool        wraparound;
    2017             : 
    2018     1826284 :         if (classForm->relkind != RELKIND_RELATION &&
    2019     1524722 :             classForm->relkind != RELKIND_MATVIEW)
    2020     1524658 :             continue;
    2021             : 
    2022      301642 :         relid = classForm->oid;
    2023             : 
    2024             :         /*
    2025             :          * Check if it is a temp table (presumably, of some other backend's).
    2026             :          * We cannot safely process other backends' temp tables.
    2027             :          */
    2028      301642 :         if (classForm->relpersistence == RELPERSISTENCE_TEMP)
    2029             :         {
    2030             :             /*
    2031             :              * We just ignore it if the owning backend is still active and
    2032             :              * using the temporary schema.  Also, for safety, ignore it if the
    2033             :              * namespace doesn't exist or isn't a temp namespace after all.
    2034             :              */
    2035          16 :             if (checkTempNamespaceStatus(classForm->relnamespace) == TEMP_NAMESPACE_IDLE)
    2036             :             {
    2037             :                 /*
    2038             :                  * The table seems to be orphaned -- although it might be that
    2039             :                  * the owning backend has already deleted it and exited; our
    2040             :                  * pg_class scan snapshot is not necessarily up-to-date
    2041             :                  * anymore, so we could be looking at a committed-dead entry.
    2042             :                  * Remember it so we can try to delete it later.
    2043             :                  */
    2044           0 :                 orphan_oids = lappend_oid(orphan_oids, relid);
    2045             :             }
    2046          16 :             continue;
    2047             :         }
    2048             : 
    2049             :         /* Fetch reloptions and the pgstat entry for this table */
    2050      301626 :         relopts = extract_autovac_opts(tuple, pg_class_desc);
    2051      301626 :         tabentry = pgstat_fetch_stat_tabentry_ext(classForm->relisshared,
    2052             :                                                   relid);
    2053             : 
    2054             :         /* Check if it needs vacuum or analyze */
    2055      301626 :         relation_needs_vacanalyze(relid, relopts, classForm, tabentry,
    2056             :                                   effective_multixact_freeze_max_age,
    2057             :                                   &dovacuum, &doanalyze, &wraparound);
    2058             : 
    2059             :         /* Relations that need work are added to table_oids */
    2060      301626 :         if (dovacuum || doanalyze)
    2061      145856 :             table_oids = lappend_oid(table_oids, relid);
    2062             : 
    2063             :         /*
    2064             :          * Remember TOAST associations for the second pass.  Note: we must do
    2065             :          * this whether or not the table is going to be vacuumed, because we
    2066             :          * don't automatically vacuum toast tables along the parent table.
    2067             :          */
    2068      301626 :         if (OidIsValid(classForm->reltoastrelid))
    2069             :         {
    2070             :             av_relation *hentry;
    2071             :             bool        found;
    2072             : 
    2073      347572 :             hentry = hash_search(table_toast_map,
    2074      173786 :                                  &classForm->reltoastrelid,
    2075             :                                  HASH_ENTER, &found);
    2076             : 
    2077      173786 :             if (!found)
    2078             :             {
    2079             :                 /* hash_search already filled in the key */
    2080      173786 :                 hentry->ar_relid = relid;
    2081      173786 :                 hentry->ar_hasrelopts = false;
    2082      173786 :                 if (relopts != NULL)
    2083             :                 {
    2084        3762 :                     hentry->ar_hasrelopts = true;
    2085        3762 :                     memcpy(&hentry->ar_reloptions, relopts,
    2086             :                            sizeof(AutoVacOpts));
    2087             :                 }
    2088             :             }
    2089             :         }
    2090             : 
    2091             :         /* Release stuff to avoid per-relation leakage */
    2092      301626 :         if (relopts)
    2093        3830 :             pfree(relopts);
    2094      301626 :         if (tabentry)
    2095      297258 :             pfree(tabentry);
    2096             :     }
    2097             : 
    2098        4322 :     table_endscan(relScan);
    2099             : 
    2100             :     /* second pass: check TOAST tables */
    2101        4322 :     ScanKeyInit(&key,
    2102             :                 Anum_pg_class_relkind,
    2103             :                 BTEqualStrategyNumber, F_CHAREQ,
    2104             :                 CharGetDatum(RELKIND_TOASTVALUE));
    2105             : 
    2106        4322 :     relScan = table_beginscan_catalog(classRel, 1, &key);
    2107      178114 :     while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
    2108             :     {
    2109      173792 :         Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
    2110             :         PgStat_StatTabEntry *tabentry;
    2111             :         Oid         relid;
    2112             :         AutoVacOpts *relopts;
    2113      173792 :         bool        free_relopts = false;
    2114             :         bool        dovacuum;
    2115             :         bool        doanalyze;
    2116             :         bool        wraparound;
    2117             : 
    2118             :         /*
    2119             :          * We cannot safely process other backends' temp tables, so skip 'em.
    2120             :          */
    2121      173792 :         if (classForm->relpersistence == RELPERSISTENCE_TEMP)
    2122           6 :             continue;
    2123             : 
    2124      173786 :         relid = classForm->oid;
    2125             : 
    2126             :         /*
    2127             :          * fetch reloptions -- if this toast table does not have them, try the
    2128             :          * main rel
    2129             :          */
    2130      173786 :         relopts = extract_autovac_opts(tuple, pg_class_desc);
    2131      173786 :         if (relopts)
    2132           2 :             free_relopts = true;
    2133             :         else
    2134             :         {
    2135             :             av_relation *hentry;
    2136             :             bool        found;
    2137             : 
    2138      173784 :             hentry = hash_search(table_toast_map, &relid, HASH_FIND, &found);
    2139      173784 :             if (found && hentry->ar_hasrelopts)
    2140        3760 :                 relopts = &hentry->ar_reloptions;
    2141             :         }
    2142             : 
    2143             :         /* Fetch the pgstat entry for this table */
    2144      173786 :         tabentry = pgstat_fetch_stat_tabentry_ext(classForm->relisshared,
    2145             :                                                   relid);
    2146             : 
    2147      173786 :         relation_needs_vacanalyze(relid, relopts, classForm, tabentry,
    2148             :                                   effective_multixact_freeze_max_age,
    2149             :                                   &dovacuum, &doanalyze, &wraparound);
    2150             : 
    2151             :         /* ignore analyze for toast tables */
    2152      173786 :         if (dovacuum)
    2153       84414 :             table_oids = lappend_oid(table_oids, relid);
    2154             : 
    2155             :         /* Release stuff to avoid leakage */
    2156      173786 :         if (free_relopts)
    2157           2 :             pfree(relopts);
    2158      173786 :         if (tabentry)
    2159      170482 :             pfree(tabentry);
    2160             :     }
    2161             : 
    2162        4322 :     table_endscan(relScan);
    2163        4322 :     table_close(classRel, AccessShareLock);
    2164             : 
    2165             :     /*
    2166             :      * Recheck orphan temporary tables, and if they still seem orphaned, drop
    2167             :      * them.  We'll eat a transaction per dropped table, which might seem
    2168             :      * excessive, but we should only need to do anything as a result of a
    2169             :      * previous backend crash, so this should not happen often enough to
    2170             :      * justify "optimizing".  Using separate transactions ensures that we
    2171             :      * don't bloat the lock table if there are many temp tables to be dropped,
    2172             :      * and it ensures that we don't lose work if a deletion attempt fails.
    2173             :      */
    2174        4322 :     foreach(cell, orphan_oids)
    2175             :     {
    2176           0 :         Oid         relid = lfirst_oid(cell);
    2177             :         Form_pg_class classForm;
    2178             :         ObjectAddress object;
    2179             : 
    2180             :         /*
    2181             :          * Check for user-requested abort.
    2182             :          */
    2183           0 :         CHECK_FOR_INTERRUPTS();
    2184             : 
    2185             :         /*
    2186             :          * Try to lock the table.  If we can't get the lock immediately,
    2187             :          * somebody else is using (or dropping) the table, so it's not our
    2188             :          * concern anymore.  Having the lock prevents race conditions below.
    2189             :          */
    2190           0 :         if (!ConditionalLockRelationOid(relid, AccessExclusiveLock))
    2191           0 :             continue;
    2192             : 
    2193             :         /*
    2194             :          * Re-fetch the pg_class tuple and re-check whether it still seems to
    2195             :          * be an orphaned temp table.  If it's not there or no longer the same
    2196             :          * relation, ignore it.
    2197             :          */
    2198           0 :         tuple = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
    2199           0 :         if (!HeapTupleIsValid(tuple))
    2200             :         {
    2201             :             /* be sure to drop useless lock so we don't bloat lock table */
    2202           0 :             UnlockRelationOid(relid, AccessExclusiveLock);
    2203           0 :             continue;
    2204             :         }
    2205           0 :         classForm = (Form_pg_class) GETSTRUCT(tuple);
    2206             : 
    2207             :         /*
    2208             :          * Make all the same tests made in the loop above.  In event of OID
    2209             :          * counter wraparound, the pg_class entry we have now might be
    2210             :          * completely unrelated to the one we saw before.
    2211             :          */
    2212           0 :         if (!((classForm->relkind == RELKIND_RELATION ||
    2213           0 :                classForm->relkind == RELKIND_MATVIEW) &&
    2214           0 :               classForm->relpersistence == RELPERSISTENCE_TEMP))
    2215             :         {
    2216           0 :             UnlockRelationOid(relid, AccessExclusiveLock);
    2217           0 :             continue;
    2218             :         }
    2219             : 
    2220           0 :         if (checkTempNamespaceStatus(classForm->relnamespace) != TEMP_NAMESPACE_IDLE)
    2221             :         {
    2222           0 :             UnlockRelationOid(relid, AccessExclusiveLock);
    2223           0 :             continue;
    2224             :         }
    2225             : 
    2226             :         /*
    2227             :          * Try to lock the temp namespace, too.  Even though we have lock on
    2228             :          * the table itself, there's a risk of deadlock against an incoming
    2229             :          * backend trying to clean out the temp namespace, in case this table
    2230             :          * has dependencies (such as sequences) that the backend's
    2231             :          * performDeletion call might visit in a different order.  If we can
    2232             :          * get AccessShareLock on the namespace, that's sufficient to ensure
    2233             :          * we're not running concurrently with RemoveTempRelations.  If we
    2234             :          * can't, back off and let RemoveTempRelations do its thing.
    2235             :          */
    2236           0 :         if (!ConditionalLockDatabaseObject(NamespaceRelationId,
    2237             :                                            classForm->relnamespace, 0,
    2238             :                                            AccessShareLock))
    2239             :         {
    2240           0 :             UnlockRelationOid(relid, AccessExclusiveLock);
    2241           0 :             continue;
    2242             :         }
    2243             : 
    2244             :         /* OK, let's delete it */
    2245           0 :         ereport(LOG,
    2246             :                 (errmsg("autovacuum: dropping orphan temp table \"%s.%s.%s\"",
    2247             :                         get_database_name(MyDatabaseId),
    2248             :                         get_namespace_name(classForm->relnamespace),
    2249             :                         NameStr(classForm->relname))));
    2250             : 
    2251             :         /*
    2252             :          * Deletion might involve TOAST table access, so ensure we have a
    2253             :          * valid snapshot.
    2254             :          */
    2255           0 :         PushActiveSnapshot(GetTransactionSnapshot());
    2256             : 
    2257           0 :         object.classId = RelationRelationId;
    2258           0 :         object.objectId = relid;
    2259           0 :         object.objectSubId = 0;
    2260           0 :         performDeletion(&object, DROP_CASCADE,
    2261             :                         PERFORM_DELETION_INTERNAL |
    2262             :                         PERFORM_DELETION_QUIETLY |
    2263             :                         PERFORM_DELETION_SKIP_EXTENSIONS);
    2264             : 
    2265             :         /*
    2266             :          * To commit the deletion, end current transaction and start a new
    2267             :          * one.  Note this also releases the locks we took.
    2268             :          */
    2269           0 :         PopActiveSnapshot();
    2270           0 :         CommitTransactionCommand();
    2271           0 :         StartTransactionCommand();
    2272             : 
    2273             :         /* StartTransactionCommand changed current memory context */
    2274           0 :         MemoryContextSwitchTo(AutovacMemCxt);
    2275             :     }
    2276             : 
    2277             :     /*
    2278             :      * Optionally, create a buffer access strategy object for VACUUM to use.
    2279             :      * We use the same BufferAccessStrategy object for all tables VACUUMed by
    2280             :      * this worker to prevent autovacuum from blowing out shared buffers.
    2281             :      *
    2282             :      * VacuumBufferUsageLimit being set to 0 results in
    2283             :      * GetAccessStrategyWithSize returning NULL, effectively meaning we can
    2284             :      * use up to all of shared buffers.
    2285             :      *
    2286             :      * If we later enter failsafe mode on any of the tables being vacuumed, we
    2287             :      * will cease use of the BufferAccessStrategy only for that table.
    2288             :      *
    2289             :      * XXX should we consider adding code to adjust the size of this if
    2290             :      * VacuumBufferUsageLimit changes?
    2291             :      */
    2292        4322 :     bstrategy = GetAccessStrategyWithSize(BAS_VACUUM, VacuumBufferUsageLimit);
    2293             : 
    2294             :     /*
    2295             :      * create a memory context to act as fake PortalContext, so that the
    2296             :      * contexts created in the vacuum code are cleaned up for each table.
    2297             :      */
    2298        4322 :     PortalContext = AllocSetContextCreate(AutovacMemCxt,
    2299             :                                           "Autovacuum Portal",
    2300             :                                           ALLOCSET_DEFAULT_SIZES);
    2301             : 
    2302             :     /*
    2303             :      * Perform operations on collected tables.
    2304             :      */
    2305      234462 :     foreach(cell, table_oids)
    2306             :     {
    2307      230144 :         Oid         relid = lfirst_oid(cell);
    2308             :         HeapTuple   classTup;
    2309             :         autovac_table *tab;
    2310             :         bool        isshared;
    2311             :         bool        skipit;
    2312             :         dlist_iter  iter;
    2313             : 
    2314      230144 :         CHECK_FOR_INTERRUPTS();
    2315             : 
    2316             :         /*
    2317             :          * Check for config changes before processing each collected table.
    2318             :          */
    2319      230144 :         if (ConfigReloadPending)
    2320             :         {
    2321           0 :             ConfigReloadPending = false;
    2322           0 :             ProcessConfigFile(PGC_SIGHUP);
    2323             : 
    2324             :             /*
    2325             :              * You might be tempted to bail out if we see autovacuum is now
    2326             :              * disabled.  Must resist that temptation -- this might be a
    2327             :              * for-wraparound emergency worker, in which case that would be
    2328             :              * entirely inappropriate.
    2329             :              */
    2330             :         }
    2331             : 
    2332             :         /*
    2333             :          * Find out whether the table is shared or not.  (It's slightly
    2334             :          * annoying to fetch the syscache entry just for this, but in typical
    2335             :          * cases it adds little cost because table_recheck_autovac would
    2336             :          * refetch the entry anyway.  We could buy that back by copying the
    2337             :          * tuple here and passing it to table_recheck_autovac, but that
    2338             :          * increases the odds of that function working with stale data.)
    2339             :          */
    2340      230144 :         classTup = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
    2341      230144 :         if (!HeapTupleIsValid(classTup))
    2342        1362 :             continue;           /* somebody deleted the rel, forget it */
    2343      230144 :         isshared = ((Form_pg_class) GETSTRUCT(classTup))->relisshared;
    2344      230144 :         ReleaseSysCache(classTup);
    2345             : 
    2346             :         /*
    2347             :          * Hold schedule lock from here until we've claimed the table.  We
    2348             :          * also need the AutovacuumLock to walk the worker array, but that one
    2349             :          * can just be a shared lock.
    2350             :          */
    2351      230144 :         LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
    2352      230144 :         LWLockAcquire(AutovacuumLock, LW_SHARED);
    2353             : 
    2354             :         /*
    2355             :          * Check whether the table is being vacuumed concurrently by another
    2356             :          * worker.
    2357             :          */
    2358      230144 :         skipit = false;
    2359      723638 :         dlist_foreach(iter, &AutoVacuumShmem->av_runningWorkers)
    2360             :         {
    2361      494608 :             WorkerInfo  worker = dlist_container(WorkerInfoData, wi_links, iter.cur);
    2362             : 
    2363             :             /* ignore myself */
    2364      494608 :             if (worker == MyWorkerInfo)
    2365      229636 :                 continue;
    2366             : 
    2367             :             /* ignore workers in other databases (unless table is shared) */
    2368      264972 :             if (!worker->wi_sharedrel && worker->wi_dboid != MyDatabaseId)
    2369          92 :                 continue;
    2370             : 
    2371      264880 :             if (worker->wi_tableoid == relid)
    2372             :             {
    2373        1114 :                 skipit = true;
    2374        1114 :                 found_concurrent_worker = true;
    2375        1114 :                 break;
    2376             :             }
    2377             :         }
    2378      230144 :         LWLockRelease(AutovacuumLock);
    2379      230144 :         if (skipit)
    2380             :         {
    2381        1114 :             LWLockRelease(AutovacuumScheduleLock);
    2382        1114 :             continue;
    2383             :         }
    2384             : 
    2385             :         /*
    2386             :          * Store the table's OID in shared memory before releasing the
    2387             :          * schedule lock, so that other workers don't try to vacuum it
    2388             :          * concurrently.  (We claim it here so as not to hold
    2389             :          * AutovacuumScheduleLock while rechecking the stats.)
    2390             :          */
    2391      229030 :         MyWorkerInfo->wi_tableoid = relid;
    2392      229030 :         MyWorkerInfo->wi_sharedrel = isshared;
    2393      229030 :         LWLockRelease(AutovacuumScheduleLock);
    2394             : 
    2395             :         /*
    2396             :          * Check whether pgstat data still says we need to vacuum this table.
    2397             :          * It could have changed if something else processed the table while
    2398             :          * we weren't looking. This doesn't entirely close the race condition,
    2399             :          * but it is very small.
    2400             :          */
    2401      229030 :         MemoryContextSwitchTo(AutovacMemCxt);
    2402      229030 :         tab = table_recheck_autovac(relid, table_toast_map, pg_class_desc,
    2403             :                                     effective_multixact_freeze_max_age);
    2404      229030 :         if (tab == NULL)
    2405             :         {
    2406             :             /* someone else vacuumed the table, or it went away */
    2407         248 :             LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
    2408         248 :             MyWorkerInfo->wi_tableoid = InvalidOid;
    2409         248 :             MyWorkerInfo->wi_sharedrel = false;
    2410         248 :             LWLockRelease(AutovacuumScheduleLock);
    2411         248 :             continue;
    2412             :         }
    2413             : 
    2414             :         /*
    2415             :          * Save the cost-related storage parameter values in global variables
    2416             :          * for reference when updating vacuum_cost_delay and vacuum_cost_limit
    2417             :          * during vacuuming this table.
    2418             :          */
    2419      228782 :         av_storage_param_cost_delay = tab->at_storage_param_vac_cost_delay;
    2420      228782 :         av_storage_param_cost_limit = tab->at_storage_param_vac_cost_limit;
    2421             : 
    2422             :         /*
    2423             :          * We only expect this worker to ever set the flag, so don't bother
    2424             :          * checking the return value. We shouldn't have to retry.
    2425             :          */
    2426      228782 :         if (tab->at_dobalance)
    2427      228782 :             pg_atomic_test_set_flag(&MyWorkerInfo->wi_dobalance);
    2428             :         else
    2429           0 :             pg_atomic_clear_flag(&MyWorkerInfo->wi_dobalance);
    2430             : 
    2431      228782 :         LWLockAcquire(AutovacuumLock, LW_SHARED);
    2432      228782 :         autovac_recalculate_workers_for_balance();
    2433      228782 :         LWLockRelease(AutovacuumLock);
    2434             : 
    2435             :         /*
    2436             :          * We wait until this point to update cost delay and cost limit
    2437             :          * values, even though we reloaded the configuration file above, so
    2438             :          * that we can take into account the cost-related storage parameters.
    2439             :          */
    2440      228782 :         VacuumUpdateCosts();
    2441             : 
    2442             : 
    2443             :         /* clean up memory before each iteration */
    2444      228782 :         MemoryContextReset(PortalContext);
    2445             : 
    2446             :         /*
    2447             :          * Save the relation name for a possible error message, to avoid a
    2448             :          * catalog lookup in case of an error.  If any of these return NULL,
    2449             :          * then the relation has been dropped since last we checked; skip it.
    2450             :          * Note: they must live in a long-lived memory context because we call
    2451             :          * vacuum and analyze in different transactions.
    2452             :          */
    2453             : 
    2454      228782 :         tab->at_relname = get_rel_name(tab->at_relid);
    2455      228782 :         tab->at_nspname = get_namespace_name(get_rel_namespace(tab->at_relid));
    2456      228782 :         tab->at_datname = get_database_name(MyDatabaseId);
    2457      228782 :         if (!tab->at_relname || !tab->at_nspname || !tab->at_datname)
    2458           0 :             goto deleted;
    2459             : 
    2460             :         /*
    2461             :          * We will abort vacuuming the current table if something errors out,
    2462             :          * and continue with the next one in schedule; in particular, this
    2463             :          * happens if we are interrupted with SIGINT.
    2464             :          */
    2465      228782 :         PG_TRY();
    2466             :         {
    2467             :             /* Use PortalContext for any per-table allocations */
    2468      228782 :             MemoryContextSwitchTo(PortalContext);
    2469             : 
    2470             :             /* have at it */
    2471      228782 :             autovacuum_do_vac_analyze(tab, bstrategy);
    2472             : 
    2473             :             /*
    2474             :              * Clear a possible query-cancel signal, to avoid a late reaction
    2475             :              * to an automatically-sent signal because of vacuuming the
    2476             :              * current table (we're done with it, so it would make no sense to
    2477             :              * cancel at this point.)
    2478             :              */
    2479      228778 :             QueryCancelPending = false;
    2480             :         }
    2481           0 :         PG_CATCH();
    2482             :         {
    2483             :             /*
    2484             :              * Abort the transaction, start a new one, and proceed with the
    2485             :              * next table in our list.
    2486             :              */
    2487           0 :             HOLD_INTERRUPTS();
    2488           0 :             if (tab->at_params.options & VACOPT_VACUUM)
    2489           0 :                 errcontext("automatic vacuum of table \"%s.%s.%s\"",
    2490             :                            tab->at_datname, tab->at_nspname, tab->at_relname);
    2491             :             else
    2492           0 :                 errcontext("automatic analyze of table \"%s.%s.%s\"",
    2493             :                            tab->at_datname, tab->at_nspname, tab->at_relname);
    2494           0 :             EmitErrorReport();
    2495             : 
    2496             :             /* this resets ProcGlobal->statusFlags[i] too */
    2497           0 :             AbortOutOfAnyTransaction();
    2498           0 :             FlushErrorState();
    2499           0 :             MemoryContextReset(PortalContext);
    2500             : 
    2501             :             /* restart our transaction for the following operations */
    2502           0 :             StartTransactionCommand();
    2503           0 :             RESUME_INTERRUPTS();
    2504             :         }
    2505      228778 :         PG_END_TRY();
    2506             : 
    2507             :         /* Make sure we're back in AutovacMemCxt */
    2508      228778 :         MemoryContextSwitchTo(AutovacMemCxt);
    2509             : 
    2510      228778 :         did_vacuum = true;
    2511             : 
    2512             :         /* ProcGlobal->statusFlags[i] are reset at the next end of xact */
    2513             : 
    2514             :         /* be tidy */
    2515      228778 : deleted:
    2516      228778 :         if (tab->at_datname != NULL)
    2517      228778 :             pfree(tab->at_datname);
    2518      228778 :         if (tab->at_nspname != NULL)
    2519      228778 :             pfree(tab->at_nspname);
    2520      228778 :         if (tab->at_relname != NULL)
    2521      228778 :             pfree(tab->at_relname);
    2522      228778 :         pfree(tab);
    2523             : 
    2524             :         /*
    2525             :          * Remove my info from shared memory.  We set wi_dobalance on the
    2526             :          * assumption that we are more likely than not to vacuum a table with
    2527             :          * no cost-related storage parameters next, so we want to claim our
    2528             :          * share of I/O as soon as possible to avoid thrashing the global
    2529             :          * balance.
    2530             :          */
    2531      228778 :         LWLockAcquire(AutovacuumScheduleLock, LW_EXCLUSIVE);
    2532      228778 :         MyWorkerInfo->wi_tableoid = InvalidOid;
    2533      228778 :         MyWorkerInfo->wi_sharedrel = false;
    2534      228778 :         LWLockRelease(AutovacuumScheduleLock);
    2535      228778 :         pg_atomic_test_set_flag(&MyWorkerInfo->wi_dobalance);
    2536             :     }
    2537             : 
    2538        4318 :     list_free(table_oids);
    2539             : 
    2540             :     /*
    2541             :      * Perform additional work items, as requested by backends.
    2542             :      */
    2543        4318 :     LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
    2544     1109726 :     for (i = 0; i < NUM_WORKITEMS; i++)
    2545             :     {
    2546     1105408 :         AutoVacuumWorkItem *workitem = &AutoVacuumShmem->av_workItems[i];
    2547             : 
    2548     1105408 :         if (!workitem->avw_used)
    2549     1105402 :             continue;
    2550           6 :         if (workitem->avw_active)
    2551           0 :             continue;
    2552           6 :         if (workitem->avw_database != MyDatabaseId)
    2553           0 :             continue;
    2554             : 
    2555             :         /* claim this one, and release lock while performing it */
    2556           6 :         workitem->avw_active = true;
    2557           6 :         LWLockRelease(AutovacuumLock);
    2558             : 
    2559           6 :         perform_work_item(workitem);
    2560             : 
    2561             :         /*
    2562             :          * Check for config changes before acquiring lock for further jobs.
    2563             :          */
    2564           6 :         CHECK_FOR_INTERRUPTS();
    2565           6 :         if (ConfigReloadPending)
    2566             :         {
    2567           0 :             ConfigReloadPending = false;
    2568           0 :             ProcessConfigFile(PGC_SIGHUP);
    2569           0 :             VacuumUpdateCosts();
    2570             :         }
    2571             : 
    2572           6 :         LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
    2573             : 
    2574             :         /* and mark it done */
    2575           6 :         workitem->avw_active = false;
    2576           6 :         workitem->avw_used = false;
    2577             :     }
    2578        4318 :     LWLockRelease(AutovacuumLock);
    2579             : 
    2580             :     /*
    2581             :      * We leak table_toast_map here (among other things), but since we're
    2582             :      * going away soon, it's not a problem normally.  But when using Valgrind,
    2583             :      * release some stuff to reduce complaints about leaked storage.
    2584             :      */
    2585             : #ifdef USE_VALGRIND
    2586             :     hash_destroy(table_toast_map);
    2587             :     FreeTupleDesc(pg_class_desc);
    2588             :     if (bstrategy)
    2589             :         pfree(bstrategy);
    2590             : #endif
    2591             : 
    2592             :     /* Run the rest in xact context, mainly to avoid Valgrind leak warnings */
    2593        4318 :     MemoryContextSwitchTo(TopTransactionContext);
    2594             : 
    2595             :     /*
    2596             :      * Update pg_database.datfrozenxid, and truncate pg_xact if possible. We
    2597             :      * only need to do this once, not after each table.
    2598             :      *
    2599             :      * Even if we didn't vacuum anything, it may still be important to do
    2600             :      * this, because one indirect effect of vac_update_datfrozenxid() is to
    2601             :      * update TransamVariables->xidVacLimit.  That might need to be done even
    2602             :      * if we haven't vacuumed anything, because relations with older
    2603             :      * relfrozenxid values or other databases with older datfrozenxid values
    2604             :      * might have been dropped, allowing xidVacLimit to advance.
    2605             :      *
    2606             :      * However, it's also important not to do this blindly in all cases,
    2607             :      * because when autovacuum=off this will restart the autovacuum launcher.
    2608             :      * If we're not careful, an infinite loop can result, where workers find
    2609             :      * no work to do and restart the launcher, which starts another worker in
    2610             :      * the same database that finds no work to do.  To prevent that, we skip
    2611             :      * this if (1) we found no work to do and (2) we skipped at least one
    2612             :      * table due to concurrent autovacuum activity.  In that case, the other
    2613             :      * worker has already done it, or will do so when it finishes.
    2614             :      */
    2615        4318 :     if (did_vacuum || !found_concurrent_worker)
    2616        4318 :         vac_update_datfrozenxid();
    2617             : 
    2618             :     /* Finally close out the last transaction. */
    2619        4318 :     CommitTransactionCommand();
    2620        4318 : }
    2621             : 
    2622             : /*
    2623             :  * Execute a previously registered work item.
    2624             :  */
    2625             : static void
    2626           6 : perform_work_item(AutoVacuumWorkItem *workitem)
    2627             : {
    2628           6 :     char       *cur_datname = NULL;
    2629           6 :     char       *cur_nspname = NULL;
    2630           6 :     char       *cur_relname = NULL;
    2631             : 
    2632             :     /*
    2633             :      * Note we do not store table info in MyWorkerInfo, since this is not
    2634             :      * vacuuming proper.
    2635             :      */
    2636             : 
    2637             :     /*
    2638             :      * Save the relation name for a possible error message, to avoid a catalog
    2639             :      * lookup in case of an error.  If any of these return NULL, then the
    2640             :      * relation has been dropped since last we checked; skip it.
    2641             :      */
    2642             :     Assert(CurrentMemoryContext == AutovacMemCxt);
    2643             : 
    2644           6 :     cur_relname = get_rel_name(workitem->avw_relation);
    2645           6 :     cur_nspname = get_namespace_name(get_rel_namespace(workitem->avw_relation));
    2646           6 :     cur_datname = get_database_name(MyDatabaseId);
    2647           6 :     if (!cur_relname || !cur_nspname || !cur_datname)
    2648           0 :         goto deleted2;
    2649             : 
    2650           6 :     autovac_report_workitem(workitem, cur_nspname, cur_relname);
    2651             : 
    2652             :     /* clean up memory before each work item */
    2653           6 :     MemoryContextReset(PortalContext);
    2654             : 
    2655             :     /*
    2656             :      * We will abort the current work item if something errors out, and
    2657             :      * continue with the next one; in particular, this happens if we are
    2658             :      * interrupted with SIGINT.  Note that this means that the work item list
    2659             :      * can be lossy.
    2660             :      */
    2661           6 :     PG_TRY();
    2662             :     {
    2663             :         /* Use PortalContext for any per-work-item allocations */
    2664           6 :         MemoryContextSwitchTo(PortalContext);
    2665             : 
    2666             :         /*
    2667             :          * Have at it.  Functions called here are responsible for any required
    2668             :          * user switch and sandbox.
    2669             :          */
    2670           6 :         switch (workitem->avw_type)
    2671             :         {
    2672           6 :             case AVW_BRINSummarizeRange:
    2673           6 :                 DirectFunctionCall2(brin_summarize_range,
    2674             :                                     ObjectIdGetDatum(workitem->avw_relation),
    2675             :                                     Int64GetDatum((int64) workitem->avw_blockNumber));
    2676           6 :                 break;
    2677           0 :             default:
    2678           0 :                 elog(WARNING, "unrecognized work item found: type %d",
    2679             :                      workitem->avw_type);
    2680           0 :                 break;
    2681             :         }
    2682             : 
    2683             :         /*
    2684             :          * Clear a possible query-cancel signal, to avoid a late reaction to
    2685             :          * an automatically-sent signal because of vacuuming the current table
    2686             :          * (we're done with it, so it would make no sense to cancel at this
    2687             :          * point.)
    2688             :          */
    2689           6 :         QueryCancelPending = false;
    2690             :     }
    2691           0 :     PG_CATCH();
    2692             :     {
    2693             :         /*
    2694             :          * Abort the transaction, start a new one, and proceed with the next
    2695             :          * table in our list.
    2696             :          */
    2697           0 :         HOLD_INTERRUPTS();
    2698           0 :         errcontext("processing work entry for relation \"%s.%s.%s\"",
    2699             :                    cur_datname, cur_nspname, cur_relname);
    2700           0 :         EmitErrorReport();
    2701             : 
    2702             :         /* this resets ProcGlobal->statusFlags[i] too */
    2703           0 :         AbortOutOfAnyTransaction();
    2704           0 :         FlushErrorState();
    2705           0 :         MemoryContextReset(PortalContext);
    2706             : 
    2707             :         /* restart our transaction for the following operations */
    2708           0 :         StartTransactionCommand();
    2709           0 :         RESUME_INTERRUPTS();
    2710             :     }
    2711           6 :     PG_END_TRY();
    2712             : 
    2713             :     /* Make sure we're back in AutovacMemCxt */
    2714           6 :     MemoryContextSwitchTo(AutovacMemCxt);
    2715             : 
    2716             :     /* We intentionally do not set did_vacuum here */
    2717             : 
    2718             :     /* be tidy */
    2719           6 : deleted2:
    2720           6 :     if (cur_datname)
    2721           6 :         pfree(cur_datname);
    2722           6 :     if (cur_nspname)
    2723           6 :         pfree(cur_nspname);
    2724           6 :     if (cur_relname)
    2725           6 :         pfree(cur_relname);
    2726           6 : }
    2727             : 
    2728             : /*
    2729             :  * extract_autovac_opts
    2730             :  *
    2731             :  * Given a relation's pg_class tuple, return a palloc'd copy of the
    2732             :  * AutoVacOpts portion of reloptions, if set; otherwise, return NULL.
    2733             :  *
    2734             :  * Note: callers do not have a relation lock on the table at this point,
    2735             :  * so the table could have been dropped, and its catalog rows gone, after
    2736             :  * we acquired the pg_class row.  If pg_class had a TOAST table, this would
    2737             :  * be a risk; fortunately, it doesn't.
    2738             :  */
    2739             : static AutoVacOpts *
    2740      704442 : extract_autovac_opts(HeapTuple tup, TupleDesc pg_class_desc)
    2741             : {
    2742             :     bytea      *relopts;
    2743             :     AutoVacOpts *av;
    2744             : 
    2745             :     Assert(((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_RELATION ||
    2746             :            ((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_MATVIEW ||
    2747             :            ((Form_pg_class) GETSTRUCT(tup))->relkind == RELKIND_TOASTVALUE);
    2748             : 
    2749      704442 :     relopts = extractRelOptions(tup, pg_class_desc, NULL);
    2750      704442 :     if (relopts == NULL)
    2751      697776 :         return NULL;
    2752             : 
    2753        6666 :     av = palloc(sizeof(AutoVacOpts));
    2754        6666 :     memcpy(av, &(((StdRdOptions *) relopts)->autovacuum), sizeof(AutoVacOpts));
    2755        6666 :     pfree(relopts);
    2756             : 
    2757        6666 :     return av;
    2758             : }
    2759             : 
    2760             : 
    2761             : /*
    2762             :  * table_recheck_autovac
    2763             :  *
    2764             :  * Recheck whether a table still needs vacuum or analyze.  Return value is a
    2765             :  * valid autovac_table pointer if it does, NULL otherwise.
    2766             :  *
    2767             :  * Note that the returned autovac_table does not have the name fields set.
    2768             :  */
    2769             : static autovac_table *
    2770      229030 : table_recheck_autovac(Oid relid, HTAB *table_toast_map,
    2771             :                       TupleDesc pg_class_desc,
    2772             :                       int effective_multixact_freeze_max_age)
    2773             : {
    2774             :     Form_pg_class classForm;
    2775             :     HeapTuple   classTup;
    2776             :     bool        dovacuum;
    2777             :     bool        doanalyze;
    2778      229030 :     autovac_table *tab = NULL;
    2779             :     bool        wraparound;
    2780             :     AutoVacOpts *avopts;
    2781      229030 :     bool        free_avopts = false;
    2782             : 
    2783             :     /* fetch the relation's relcache entry */
    2784      229030 :     classTup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
    2785      229030 :     if (!HeapTupleIsValid(classTup))
    2786           0 :         return NULL;
    2787      229030 :     classForm = (Form_pg_class) GETSTRUCT(classTup);
    2788             : 
    2789             :     /*
    2790             :      * Get the applicable reloptions.  If it is a TOAST table, try to get the
    2791             :      * main table reloptions if the toast table itself doesn't have.
    2792             :      */
    2793      229030 :     avopts = extract_autovac_opts(classTup, pg_class_desc);
    2794      229030 :     if (avopts)
    2795        2834 :         free_avopts = true;
    2796      226196 :     else if (classForm->relkind == RELKIND_TOASTVALUE &&
    2797             :              table_toast_map != NULL)
    2798             :     {
    2799             :         av_relation *hentry;
    2800             :         bool        found;
    2801             : 
    2802       83992 :         hentry = hash_search(table_toast_map, &relid, HASH_FIND, &found);
    2803       83992 :         if (found && hentry->ar_hasrelopts)
    2804        2816 :             avopts = &hentry->ar_reloptions;
    2805             :     }
    2806             : 
    2807      229030 :     recheck_relation_needs_vacanalyze(relid, avopts, classForm,
    2808             :                                       effective_multixact_freeze_max_age,
    2809             :                                       &dovacuum, &doanalyze, &wraparound);
    2810             : 
    2811             :     /* OK, it needs something done */
    2812      229030 :     if (doanalyze || dovacuum)
    2813             :     {
    2814             :         int         freeze_min_age;
    2815             :         int         freeze_table_age;
    2816             :         int         multixact_freeze_min_age;
    2817             :         int         multixact_freeze_table_age;
    2818             :         int         log_vacuum_min_duration;
    2819             :         int         log_analyze_min_duration;
    2820             : 
    2821             :         /*
    2822             :          * Calculate the vacuum cost parameters and the freeze ages.  If there
    2823             :          * are options set in pg_class.reloptions, use them; in the case of a
    2824             :          * toast table, try the main table too.  Otherwise use the GUC
    2825             :          * defaults, autovacuum's own first and plain vacuum second.
    2826             :          */
    2827             : 
    2828             :         /* -1 in autovac setting means use log_autovacuum_min_duration */
    2829        5646 :         log_vacuum_min_duration = (avopts && avopts->log_vacuum_min_duration >= 0)
    2830             :             ? avopts->log_vacuum_min_duration
    2831      234428 :             : Log_autovacuum_min_duration;
    2832             : 
    2833             :         /* -1 in autovac setting means use log_autoanalyze_min_duration */
    2834        5646 :         log_analyze_min_duration = (avopts && avopts->log_analyze_min_duration >= 0)
    2835             :             ? avopts->log_analyze_min_duration
    2836      234428 :             : Log_autoanalyze_min_duration;
    2837             : 
    2838             :         /* these do not have autovacuum-specific settings */
    2839        5646 :         freeze_min_age = (avopts && avopts->freeze_min_age >= 0)
    2840             :             ? avopts->freeze_min_age
    2841      234428 :             : default_freeze_min_age;
    2842             : 
    2843        5646 :         freeze_table_age = (avopts && avopts->freeze_table_age >= 0)
    2844             :             ? avopts->freeze_table_age
    2845      234428 :             : default_freeze_table_age;
    2846             : 
    2847      234428 :         multixact_freeze_min_age = (avopts &&
    2848        5646 :                                     avopts->multixact_freeze_min_age >= 0)
    2849             :             ? avopts->multixact_freeze_min_age
    2850      234428 :             : default_multixact_freeze_min_age;
    2851             : 
    2852      234428 :         multixact_freeze_table_age = (avopts &&
    2853        5646 :                                       avopts->multixact_freeze_table_age >= 0)
    2854             :             ? avopts->multixact_freeze_table_age
    2855      234428 :             : default_multixact_freeze_table_age;
    2856             : 
    2857      228782 :         tab = palloc(sizeof(autovac_table));
    2858      228782 :         tab->at_relid = relid;
    2859      228782 :         tab->at_sharedrel = classForm->relisshared;
    2860             : 
    2861             :         /*
    2862             :          * Select VACUUM options.  Note we don't say VACOPT_PROCESS_TOAST, so
    2863             :          * that vacuum() skips toast relations.  Also note we tell vacuum() to
    2864             :          * skip vac_update_datfrozenxid(); we'll do that separately.
    2865             :          */
    2866      228782 :         tab->at_params.options =
    2867      228782 :             (dovacuum ? (VACOPT_VACUUM |
    2868             :                          VACOPT_PROCESS_MAIN |
    2869      228782 :                          VACOPT_SKIP_DATABASE_STATS) : 0) |
    2870      228782 :             (doanalyze ? VACOPT_ANALYZE : 0) |
    2871      228782 :             (!wraparound ? VACOPT_SKIP_LOCKED : 0);
    2872             : 
    2873             :         /*
    2874             :          * index_cleanup and truncate are unspecified at first in autovacuum.
    2875             :          * They will be filled in with usable values using their reloptions
    2876             :          * (or reloption defaults) later.
    2877             :          */
    2878      228782 :         tab->at_params.index_cleanup = VACOPTVALUE_UNSPECIFIED;
    2879      228782 :         tab->at_params.truncate = VACOPTVALUE_UNSPECIFIED;
    2880             :         /* As of now, we don't support parallel vacuum for autovacuum */
    2881      228782 :         tab->at_params.nworkers = -1;
    2882      228782 :         tab->at_params.freeze_min_age = freeze_min_age;
    2883      228782 :         tab->at_params.freeze_table_age = freeze_table_age;
    2884      228782 :         tab->at_params.multixact_freeze_min_age = multixact_freeze_min_age;
    2885      228782 :         tab->at_params.multixact_freeze_table_age = multixact_freeze_table_age;
    2886      228782 :         tab->at_params.is_wraparound = wraparound;
    2887      228782 :         tab->at_params.log_vacuum_min_duration = log_vacuum_min_duration;
    2888      228782 :         tab->at_params.log_analyze_min_duration = log_analyze_min_duration;
    2889      228782 :         tab->at_params.toast_parent = InvalidOid;
    2890             : 
    2891             :         /*
    2892             :          * Later, in vacuum_rel(), we check reloptions for any
    2893             :          * vacuum_max_eager_freeze_failure_rate override.
    2894             :          */
    2895      228782 :         tab->at_params.max_eager_freeze_failure_rate = vacuum_max_eager_freeze_failure_rate;
    2896      228782 :         tab->at_storage_param_vac_cost_limit = avopts ?
    2897      228782 :             avopts->vacuum_cost_limit : 0;
    2898      228782 :         tab->at_storage_param_vac_cost_delay = avopts ?
    2899      228782 :             avopts->vacuum_cost_delay : -1;
    2900      228782 :         tab->at_relname = NULL;
    2901      228782 :         tab->at_nspname = NULL;
    2902      228782 :         tab->at_datname = NULL;
    2903             : 
    2904             :         /*
    2905             :          * If any of the cost delay parameters has been set individually for
    2906             :          * this table, disable the balancing algorithm.
    2907             :          */
    2908      228782 :         tab->at_dobalance =
    2909      234428 :             !(avopts && (avopts->vacuum_cost_limit > 0 ||
    2910        5646 :                          avopts->vacuum_cost_delay >= 0));
    2911             :     }
    2912             : 
    2913      229030 :     if (free_avopts)
    2914        2834 :         pfree(avopts);
    2915      229030 :     heap_freetuple(classTup);
    2916      229030 :     return tab;
    2917             : }
    2918             : 
    2919             : /*
    2920             :  * recheck_relation_needs_vacanalyze
    2921             :  *
    2922             :  * Subroutine for table_recheck_autovac.
    2923             :  *
    2924             :  * Fetch the pgstat of a relation and recheck whether a relation
    2925             :  * needs to be vacuumed or analyzed.
    2926             :  */
    2927             : static void
    2928      229030 : recheck_relation_needs_vacanalyze(Oid relid,
    2929             :                                   AutoVacOpts *avopts,
    2930             :                                   Form_pg_class classForm,
    2931             :                                   int effective_multixact_freeze_max_age,
    2932             :                                   bool *dovacuum,
    2933             :                                   bool *doanalyze,
    2934             :                                   bool *wraparound)
    2935             : {
    2936             :     PgStat_StatTabEntry *tabentry;
    2937             : 
    2938             :     /* fetch the pgstat table entry */
    2939      229030 :     tabentry = pgstat_fetch_stat_tabentry_ext(classForm->relisshared,
    2940             :                                               relid);
    2941             : 
    2942      229030 :     relation_needs_vacanalyze(relid, avopts, classForm, tabentry,
    2943             :                               effective_multixact_freeze_max_age,
    2944             :                               dovacuum, doanalyze, wraparound);
    2945             : 
    2946             :     /* Release tabentry to avoid leakage */
    2947      229030 :     if (tabentry)
    2948      228110 :         pfree(tabentry);
    2949             : 
    2950             :     /* ignore ANALYZE for toast tables */
    2951      229030 :     if (classForm->relkind == RELKIND_TOASTVALUE)
    2952       83992 :         *doanalyze = false;
    2953      229030 : }
    2954             : 
    2955             : /*
    2956             :  * relation_needs_vacanalyze
    2957             :  *
    2958             :  * Check whether a relation needs to be vacuumed or analyzed; return each into
    2959             :  * "dovacuum" and "doanalyze", respectively.  Also return whether the vacuum is
    2960             :  * being forced because of Xid or multixact wraparound.
    2961             :  *
    2962             :  * relopts is a pointer to the AutoVacOpts options (either for itself in the
    2963             :  * case of a plain table, or for either itself or its parent table in the case
    2964             :  * of a TOAST table), NULL if none; tabentry is the pgstats entry, which can be
    2965             :  * NULL.
    2966             :  *
    2967             :  * A table needs to be vacuumed if the number of dead tuples exceeds a
    2968             :  * threshold.  This threshold is calculated as
    2969             :  *
    2970             :  * threshold = vac_base_thresh + vac_scale_factor * reltuples
    2971             :  * if (threshold > vac_max_thresh)
    2972             :  *     threshold = vac_max_thresh;
    2973             :  *
    2974             :  * For analyze, the analysis done is that the number of tuples inserted,
    2975             :  * deleted and updated since the last analyze exceeds a threshold calculated
    2976             :  * in the same fashion as above.  Note that the cumulative stats system stores
    2977             :  * the number of tuples (both live and dead) that there were as of the last
    2978             :  * analyze.  This is asymmetric to the VACUUM case.
    2979             :  *
    2980             :  * We also force vacuum if the table's relfrozenxid is more than freeze_max_age
    2981             :  * transactions back, and if its relminmxid is more than
    2982             :  * multixact_freeze_max_age multixacts back.
    2983             :  *
    2984             :  * A table whose autovacuum_enabled option is false is
    2985             :  * automatically skipped (unless we have to vacuum it due to freeze_max_age).
    2986             :  * Thus autovacuum can be disabled for specific tables. Also, when the cumulative
    2987             :  * stats system does not have data about a table, it will be skipped.
    2988             :  *
    2989             :  * A table whose vac_base_thresh value is < 0 takes the base value from the
    2990             :  * autovacuum_vacuum_threshold GUC variable.  Similarly, a vac_scale_factor
    2991             :  * value < 0 is substituted with the value of
    2992             :  * autovacuum_vacuum_scale_factor GUC variable.  Ditto for analyze.
    2993             :  */
    2994             : static void
    2995      704442 : relation_needs_vacanalyze(Oid relid,
    2996             :                           AutoVacOpts *relopts,
    2997             :                           Form_pg_class classForm,
    2998             :                           PgStat_StatTabEntry *tabentry,
    2999             :                           int effective_multixact_freeze_max_age,
    3000             :  /* output params below */
    3001             :                           bool *dovacuum,
    3002             :                           bool *doanalyze,
    3003             :                           bool *wraparound)
    3004             : {
    3005             :     bool        force_vacuum;
    3006             :     bool        av_enabled;
    3007             : 
    3008             :     /* constants from reloptions or GUC variables */
    3009             :     int         vac_base_thresh,
    3010             :                 vac_max_thresh,
    3011             :                 vac_ins_base_thresh,
    3012             :                 anl_base_thresh;
    3013             :     float4      vac_scale_factor,
    3014             :                 vac_ins_scale_factor,
    3015             :                 anl_scale_factor;
    3016             : 
    3017             :     /* thresholds calculated from above constants */
    3018             :     float4      vacthresh,
    3019             :                 vacinsthresh,
    3020             :                 anlthresh;
    3021             : 
    3022             :     /* number of vacuum (resp. analyze) tuples at this time */
    3023             :     float4      vactuples,
    3024             :                 instuples,
    3025             :                 anltuples;
    3026             : 
    3027             :     /* freeze parameters */
    3028             :     int         freeze_max_age;
    3029             :     int         multixact_freeze_max_age;
    3030             :     TransactionId xidForceLimit;
    3031             :     TransactionId relfrozenxid;
    3032             :     MultiXactId multiForceLimit;
    3033             : 
    3034             :     Assert(classForm != NULL);
    3035             :     Assert(OidIsValid(relid));
    3036             : 
    3037             :     /*
    3038             :      * Determine vacuum/analyze equation parameters.  We have two possible
    3039             :      * sources: the passed reloptions (which could be a main table or a toast
    3040             :      * table), or the autovacuum GUC variables.
    3041             :      */
    3042             : 
    3043             :     /* -1 in autovac setting means use plain vacuum_scale_factor */
    3044       13242 :     vac_scale_factor = (relopts && relopts->vacuum_scale_factor >= 0)
    3045           0 :         ? relopts->vacuum_scale_factor
    3046      717684 :         : autovacuum_vac_scale;
    3047             : 
    3048       13242 :     vac_base_thresh = (relopts && relopts->vacuum_threshold >= 0)
    3049             :         ? relopts->vacuum_threshold
    3050      717684 :         : autovacuum_vac_thresh;
    3051             : 
    3052             :     /* -1 is used to disable max threshold */
    3053       13242 :     vac_max_thresh = (relopts && relopts->vacuum_max_threshold >= -1)
    3054             :         ? relopts->vacuum_max_threshold
    3055      717684 :         : autovacuum_vac_max_thresh;
    3056             : 
    3057       13242 :     vac_ins_scale_factor = (relopts && relopts->vacuum_ins_scale_factor >= 0)
    3058           0 :         ? relopts->vacuum_ins_scale_factor
    3059      717684 :         : autovacuum_vac_ins_scale;
    3060             : 
    3061             :     /* -1 is used to disable insert vacuums */
    3062       13242 :     vac_ins_base_thresh = (relopts && relopts->vacuum_ins_threshold >= -1)
    3063             :         ? relopts->vacuum_ins_threshold
    3064      717684 :         : autovacuum_vac_ins_thresh;
    3065             : 
    3066       13242 :     anl_scale_factor = (relopts && relopts->analyze_scale_factor >= 0)
    3067           0 :         ? relopts->analyze_scale_factor
    3068      717684 :         : autovacuum_anl_scale;
    3069             : 
    3070       13242 :     anl_base_thresh = (relopts && relopts->analyze_threshold >= 0)
    3071             :         ? relopts->analyze_threshold
    3072      717684 :         : autovacuum_anl_thresh;
    3073             : 
    3074       13242 :     freeze_max_age = (relopts && relopts->freeze_max_age >= 0)
    3075           0 :         ? Min(relopts->freeze_max_age, autovacuum_freeze_max_age)
    3076      717684 :         : autovacuum_freeze_max_age;
    3077             : 
    3078       13242 :     multixact_freeze_max_age = (relopts && relopts->multixact_freeze_max_age >= 0)
    3079           0 :         ? Min(relopts->multixact_freeze_max_age, effective_multixact_freeze_max_age)
    3080      717684 :         : effective_multixact_freeze_max_age;
    3081             : 
    3082      704442 :     av_enabled = (relopts ? relopts->enabled : true);
    3083             : 
    3084             :     /* Force vacuum if table is at risk of wraparound */
    3085      704442 :     xidForceLimit = recentXid - freeze_max_age;
    3086      704442 :     if (xidForceLimit < FirstNormalTransactionId)
    3087           0 :         xidForceLimit -= FirstNormalTransactionId;
    3088      704442 :     relfrozenxid = classForm->relfrozenxid;
    3089     1408884 :     force_vacuum = (TransactionIdIsNormal(relfrozenxid) &&
    3090      704442 :                     TransactionIdPrecedes(relfrozenxid, xidForceLimit));
    3091      704442 :     if (!force_vacuum)
    3092             :     {
    3093      246634 :         MultiXactId relminmxid = classForm->relminmxid;
    3094             : 
    3095      246634 :         multiForceLimit = recentMulti - multixact_freeze_max_age;
    3096      246634 :         if (multiForceLimit < FirstMultiXactId)
    3097           0 :             multiForceLimit -= FirstMultiXactId;
    3098      493268 :         force_vacuum = MultiXactIdIsValid(relminmxid) &&
    3099      246634 :             MultiXactIdPrecedes(relminmxid, multiForceLimit);
    3100             :     }
    3101      704442 :     *wraparound = force_vacuum;
    3102             : 
    3103             :     /* User disabled it in pg_class.reloptions?  (But ignore if at risk) */
    3104      704442 :     if (!av_enabled && !force_vacuum)
    3105             :     {
    3106        1872 :         *doanalyze = false;
    3107        1872 :         *dovacuum = false;
    3108        1872 :         return;
    3109             :     }
    3110             : 
    3111             :     /*
    3112             :      * If we found stats for the table, and autovacuum is currently enabled,
    3113             :      * make a threshold-based decision whether to vacuum and/or analyze.  If
    3114             :      * autovacuum is currently disabled, we must be here for anti-wraparound
    3115             :      * vacuuming only, so don't vacuum (or analyze) anything that's not being
    3116             :      * forced.
    3117             :      */
    3118      702570 :     if (tabentry && AutoVacuumingActive())
    3119      693978 :     {
    3120      693978 :         float4      pcnt_unfrozen = 1;
    3121      693978 :         float4      reltuples = classForm->reltuples;
    3122      693978 :         int32       relpages = classForm->relpages;
    3123      693978 :         int32       relallfrozen = classForm->relallfrozen;
    3124             : 
    3125      693978 :         vactuples = tabentry->dead_tuples;
    3126      693978 :         instuples = tabentry->ins_since_vacuum;
    3127      693978 :         anltuples = tabentry->mod_since_analyze;
    3128             : 
    3129             :         /* If the table hasn't yet been vacuumed, take reltuples as zero */
    3130      693978 :         if (reltuples < 0)
    3131        2674 :             reltuples = 0;
    3132             : 
    3133             :         /*
    3134             :          * If we have data for relallfrozen, calculate the unfrozen percentage
    3135             :          * of the table to modify insert scale factor. This helps us decide
    3136             :          * whether or not to vacuum an insert-heavy table based on the number
    3137             :          * of inserts to the more "active" part of the table.
    3138             :          */
    3139      693978 :         if (relpages > 0 && relallfrozen > 0)
    3140             :         {
    3141             :             /*
    3142             :              * It could be the stats were updated manually and relallfrozen >
    3143             :              * relpages. Clamp relallfrozen to relpages to avoid nonsensical
    3144             :              * calculations.
    3145             :              */
    3146      269450 :             relallfrozen = Min(relallfrozen, relpages);
    3147      269450 :             pcnt_unfrozen = 1 - ((float4) relallfrozen / relpages);
    3148             :         }
    3149             : 
    3150      693978 :         vacthresh = (float4) vac_base_thresh + vac_scale_factor * reltuples;
    3151      693978 :         if (vac_max_thresh >= 0 && vacthresh > (float4) vac_max_thresh)
    3152           0 :             vacthresh = (float4) vac_max_thresh;
    3153             : 
    3154      693978 :         vacinsthresh = (float4) vac_ins_base_thresh +
    3155      693978 :             vac_ins_scale_factor * reltuples * pcnt_unfrozen;
    3156      693978 :         anlthresh = (float4) anl_base_thresh + anl_scale_factor * reltuples;
    3157             : 
    3158             :         /*
    3159             :          * Note that we don't need to take special consideration for stat
    3160             :          * reset, because if that happens, the last vacuum and analyze counts
    3161             :          * will be reset too.
    3162             :          */
    3163      693978 :         if (vac_ins_base_thresh >= 0)
    3164      693978 :             elog(DEBUG3, "%s: vac: %.0f (threshold %.0f), ins: %.0f (threshold %.0f), anl: %.0f (threshold %.0f)",
    3165             :                  NameStr(classForm->relname),
    3166             :                  vactuples, vacthresh, instuples, vacinsthresh, anltuples, anlthresh);
    3167             :         else
    3168           0 :             elog(DEBUG3, "%s: vac: %.0f (threshold %.0f), ins: (disabled), anl: %.0f (threshold %.0f)",
    3169             :                  NameStr(classForm->relname),
    3170             :                  vactuples, vacthresh, anltuples, anlthresh);
    3171             : 
    3172             :         /* Determine if this table needs vacuum or analyze. */
    3173      931658 :         *dovacuum = force_vacuum || (vactuples > vacthresh) ||
    3174      237680 :             (vac_ins_base_thresh >= 0 && instuples > vacinsthresh);
    3175      693978 :         *doanalyze = (anltuples > anlthresh);
    3176             :     }
    3177             :     else
    3178             :     {
    3179             :         /*
    3180             :          * Skip a table not found in stat hash, unless we have to force vacuum
    3181             :          * for anti-wrap purposes.  If it's not acted upon, there's no need to
    3182             :          * vacuum it.
    3183             :          */
    3184        8592 :         *dovacuum = force_vacuum;
    3185        8592 :         *doanalyze = false;
    3186             :     }
    3187             : 
    3188             :     /* ANALYZE refuses to work with pg_statistic */
    3189      702570 :     if (relid == StatisticRelationId)
    3190        6418 :         *doanalyze = false;
    3191             : }
    3192             : 
    3193             : /*
    3194             :  * autovacuum_do_vac_analyze
    3195             :  *      Vacuum and/or analyze the specified table
    3196             :  *
    3197             :  * We expect the caller to have switched into a memory context that won't
    3198             :  * disappear at transaction commit.
    3199             :  */
    3200             : static void
    3201      228782 : autovacuum_do_vac_analyze(autovac_table *tab, BufferAccessStrategy bstrategy)
    3202             : {
    3203             :     RangeVar   *rangevar;
    3204             :     VacuumRelation *rel;
    3205             :     List       *rel_list;
    3206             :     MemoryContext vac_context;
    3207             :     MemoryContext old_context;
    3208             : 
    3209             :     /* Let pgstat know what we're doing */
    3210      228782 :     autovac_report_activity(tab);
    3211             : 
    3212             :     /* Create a context that vacuum() can use as cross-transaction storage */
    3213      228782 :     vac_context = AllocSetContextCreate(CurrentMemoryContext,
    3214             :                                         "Vacuum",
    3215             :                                         ALLOCSET_DEFAULT_SIZES);
    3216             : 
    3217             :     /* Set up one VacuumRelation target, identified by OID, for vacuum() */
    3218      228782 :     old_context = MemoryContextSwitchTo(vac_context);
    3219      228782 :     rangevar = makeRangeVar(tab->at_nspname, tab->at_relname, -1);
    3220      228782 :     rel = makeVacuumRelation(rangevar, tab->at_relid, NIL);
    3221      228782 :     rel_list = list_make1(rel);
    3222      228782 :     MemoryContextSwitchTo(old_context);
    3223             : 
    3224      228782 :     vacuum(rel_list, tab->at_params, bstrategy, vac_context, true);
    3225             : 
    3226      228778 :     MemoryContextDelete(vac_context);
    3227      228778 : }
    3228             : 
    3229             : /*
    3230             :  * autovac_report_activity
    3231             :  *      Report to pgstat what autovacuum is doing
    3232             :  *
    3233             :  * We send a SQL string corresponding to what the user would see if the
    3234             :  * equivalent command was to be issued manually.
    3235             :  *
    3236             :  * Note we assume that we are going to report the next command as soon as we're
    3237             :  * done with the current one, and exit right after the last one, so we don't
    3238             :  * bother to report "<IDLE>" or some such.
    3239             :  */
    3240             : static void
    3241      228782 : autovac_report_activity(autovac_table *tab)
    3242             : {
    3243             : #define MAX_AUTOVAC_ACTIV_LEN (NAMEDATALEN * 2 + 56)
    3244             :     char        activity[MAX_AUTOVAC_ACTIV_LEN];
    3245             :     int         len;
    3246             : 
    3247             :     /* Report the command and possible options */
    3248      228782 :     if (tab->at_params.options & VACOPT_VACUUM)
    3249      228500 :         snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
    3250             :                  "autovacuum: VACUUM%s",
    3251      228500 :                  tab->at_params.options & VACOPT_ANALYZE ? " ANALYZE" : "");
    3252             :     else
    3253         282 :         snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
    3254             :                  "autovacuum: ANALYZE");
    3255             : 
    3256             :     /*
    3257             :      * Report the qualified name of the relation.
    3258             :      */
    3259      228782 :     len = strlen(activity);
    3260             : 
    3261      228782 :     snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
    3262             :              " %s.%s%s", tab->at_nspname, tab->at_relname,
    3263      228782 :              tab->at_params.is_wraparound ? " (to prevent wraparound)" : "");
    3264             : 
    3265             :     /* Set statement_timestamp() to current time for pg_stat_activity */
    3266      228782 :     SetCurrentStatementStartTimestamp();
    3267             : 
    3268      228782 :     pgstat_report_activity(STATE_RUNNING, activity);
    3269      228782 : }
    3270             : 
    3271             : /*
    3272             :  * autovac_report_workitem
    3273             :  *      Report to pgstat that autovacuum is processing a work item
    3274             :  */
    3275             : static void
    3276           6 : autovac_report_workitem(AutoVacuumWorkItem *workitem,
    3277             :                         const char *nspname, const char *relname)
    3278             : {
    3279             :     char        activity[MAX_AUTOVAC_ACTIV_LEN + 12 + 2];
    3280             :     char        blk[12 + 2];
    3281             :     int         len;
    3282             : 
    3283           6 :     switch (workitem->avw_type)
    3284             :     {
    3285           6 :         case AVW_BRINSummarizeRange:
    3286           6 :             snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
    3287             :                      "autovacuum: BRIN summarize");
    3288           6 :             break;
    3289             :     }
    3290             : 
    3291             :     /*
    3292             :      * Report the qualified name of the relation, and the block number if any
    3293             :      */
    3294           6 :     len = strlen(activity);
    3295             : 
    3296           6 :     if (BlockNumberIsValid(workitem->avw_blockNumber))
    3297           6 :         snprintf(blk, sizeof(blk), " %u", workitem->avw_blockNumber);
    3298             :     else
    3299           0 :         blk[0] = '\0';
    3300             : 
    3301           6 :     snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
    3302             :              " %s.%s%s", nspname, relname, blk);
    3303             : 
    3304             :     /* Set statement_timestamp() to current time for pg_stat_activity */
    3305           6 :     SetCurrentStatementStartTimestamp();
    3306             : 
    3307           6 :     pgstat_report_activity(STATE_RUNNING, activity);
    3308           6 : }
    3309             : 
    3310             : /*
    3311             :  * AutoVacuumingActive
    3312             :  *      Check GUC vars and report whether the autovacuum process should be
    3313             :  *      running.
    3314             :  */
    3315             : bool
    3316      769348 : AutoVacuumingActive(void)
    3317             : {
    3318      769348 :     if (!autovacuum_start_daemon || !pgstat_track_counts)
    3319        9942 :         return false;
    3320      759406 :     return true;
    3321             : }
    3322             : 
    3323             : /*
    3324             :  * Request one work item to the next autovacuum run processing our database.
    3325             :  * Return false if the request can't be recorded.
    3326             :  */
    3327             : bool
    3328           6 : AutoVacuumRequestWork(AutoVacuumWorkItemType type, Oid relationId,
    3329             :                       BlockNumber blkno)
    3330             : {
    3331             :     int         i;
    3332           6 :     bool        result = false;
    3333             : 
    3334           6 :     LWLockAcquire(AutovacuumLock, LW_EXCLUSIVE);
    3335             : 
    3336             :     /*
    3337             :      * Locate an unused work item and fill it with the given data.
    3338             :      */
    3339          12 :     for (i = 0; i < NUM_WORKITEMS; i++)
    3340             :     {
    3341          12 :         AutoVacuumWorkItem *workitem = &AutoVacuumShmem->av_workItems[i];
    3342             : 
    3343          12 :         if (workitem->avw_used)
    3344           6 :             continue;
    3345             : 
    3346           6 :         workitem->avw_used = true;
    3347           6 :         workitem->avw_active = false;
    3348           6 :         workitem->avw_type = type;
    3349           6 :         workitem->avw_database = MyDatabaseId;
    3350           6 :         workitem->avw_relation = relationId;
    3351           6 :         workitem->avw_blockNumber = blkno;
    3352           6 :         result = true;
    3353             : 
    3354             :         /* done */
    3355           6 :         break;
    3356             :     }
    3357             : 
    3358           6 :     LWLockRelease(AutovacuumLock);
    3359             : 
    3360           6 :     return result;
    3361             : }
    3362             : 
    3363             : /*
    3364             :  * autovac_init
    3365             :  *      This is called at postmaster initialization.
    3366             :  *
    3367             :  * All we do here is annoy the user if he got it wrong.
    3368             :  */
    3369             : void
    3370        1734 : autovac_init(void)
    3371             : {
    3372        1734 :     if (!autovacuum_start_daemon)
    3373         242 :         return;
    3374        1492 :     else if (!pgstat_track_counts)
    3375           0 :         ereport(WARNING,
    3376             :                 (errmsg("autovacuum not started because of misconfiguration"),
    3377             :                  errhint("Enable the \"track_counts\" option.")));
    3378             :     else
    3379        1492 :         check_av_worker_gucs();
    3380             : }
    3381             : 
    3382             : /*
    3383             :  * AutoVacuumShmemSize
    3384             :  *      Compute space needed for autovacuum-related shared memory
    3385             :  */
    3386             : Size
    3387        6246 : AutoVacuumShmemSize(void)
    3388             : {
    3389             :     Size        size;
    3390             : 
    3391             :     /*
    3392             :      * Need the fixed struct and the array of WorkerInfoData.
    3393             :      */
    3394        6246 :     size = sizeof(AutoVacuumShmemStruct);
    3395        6246 :     size = MAXALIGN(size);
    3396        6246 :     size = add_size(size, mul_size(autovacuum_worker_slots,
    3397             :                                    sizeof(WorkerInfoData)));
    3398        6246 :     return size;
    3399             : }
    3400             : 
    3401             : /*
    3402             :  * AutoVacuumShmemInit
    3403             :  *      Allocate and initialize autovacuum-related shared memory
    3404             :  */
    3405             : void
    3406        2182 : AutoVacuumShmemInit(void)
    3407             : {
    3408             :     bool        found;
    3409             : 
    3410        2182 :     AutoVacuumShmem = (AutoVacuumShmemStruct *)
    3411        2182 :         ShmemInitStruct("AutoVacuum Data",
    3412             :                         AutoVacuumShmemSize(),
    3413             :                         &found);
    3414             : 
    3415        2182 :     if (!IsUnderPostmaster)
    3416             :     {
    3417             :         WorkerInfo  worker;
    3418             :         int         i;
    3419             : 
    3420             :         Assert(!found);
    3421             : 
    3422        2182 :         AutoVacuumShmem->av_launcherpid = 0;
    3423        2182 :         dclist_init(&AutoVacuumShmem->av_freeWorkers);
    3424        2182 :         dlist_init(&AutoVacuumShmem->av_runningWorkers);
    3425        2182 :         AutoVacuumShmem->av_startingWorker = NULL;
    3426        2182 :         memset(AutoVacuumShmem->av_workItems, 0,
    3427             :                sizeof(AutoVacuumWorkItem) * NUM_WORKITEMS);
    3428             : 
    3429        2182 :         worker = (WorkerInfo) ((char *) AutoVacuumShmem +
    3430             :                                MAXALIGN(sizeof(AutoVacuumShmemStruct)));
    3431             : 
    3432             :         /* initialize the WorkerInfo free list */
    3433       26096 :         for (i = 0; i < autovacuum_worker_slots; i++)
    3434             :         {
    3435       23914 :             dclist_push_head(&AutoVacuumShmem->av_freeWorkers,
    3436       23914 :                              &worker[i].wi_links);
    3437       23914 :             pg_atomic_init_flag(&worker[i].wi_dobalance);
    3438             :         }
    3439             : 
    3440        2182 :         pg_atomic_init_u32(&AutoVacuumShmem->av_nworkersForBalance, 0);
    3441             : 
    3442             :     }
    3443             :     else
    3444             :         Assert(found);
    3445        2182 : }
    3446             : 
    3447             : /*
    3448             :  * GUC check_hook for autovacuum_work_mem
    3449             :  */
    3450             : bool
    3451        2262 : check_autovacuum_work_mem(int *newval, void **extra, GucSource source)
    3452             : {
    3453             :     /*
    3454             :      * -1 indicates fallback.
    3455             :      *
    3456             :      * If we haven't yet changed the boot_val default of -1, just let it be.
    3457             :      * Autovacuum will look to maintenance_work_mem instead.
    3458             :      */
    3459        2262 :     if (*newval == -1)
    3460        2258 :         return true;
    3461             : 
    3462             :     /*
    3463             :      * We clamp manually-set values to at least 64kB.  Since
    3464             :      * maintenance_work_mem is always set to at least this value, do the same
    3465             :      * here.
    3466             :      */
    3467           4 :     if (*newval < 64)
    3468           4 :         *newval = 64;
    3469             : 
    3470           4 :     return true;
    3471             : }
    3472             : 
    3473             : /*
    3474             :  * Returns whether there is a free autovacuum worker slot available.
    3475             :  */
    3476             : static bool
    3477       34422 : av_worker_available(void)
    3478             : {
    3479             :     int         free_slots;
    3480             :     int         reserved_slots;
    3481             : 
    3482       34422 :     free_slots = dclist_count(&AutoVacuumShmem->av_freeWorkers);
    3483             : 
    3484       34422 :     reserved_slots = autovacuum_worker_slots - autovacuum_max_workers;
    3485       34422 :     reserved_slots = Max(0, reserved_slots);
    3486             : 
    3487       34422 :     return free_slots > reserved_slots;
    3488             : }
    3489             : 
    3490             : /*
    3491             :  * Emits a WARNING if autovacuum_worker_slots < autovacuum_max_workers.
    3492             :  */
    3493             : static void
    3494        1492 : check_av_worker_gucs(void)
    3495             : {
    3496        1492 :     if (autovacuum_worker_slots < autovacuum_max_workers)
    3497           0 :         ereport(WARNING,
    3498             :                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
    3499             :                  errmsg("\"autovacuum_max_workers\" (%d) should be less than or equal to \"autovacuum_worker_slots\" (%d)",
    3500             :                         autovacuum_max_workers, autovacuum_worker_slots),
    3501             :                  errdetail("The server will only start up to \"autovacuum_worker_slots\" (%d) autovacuum workers at a given time.",
    3502             :                            autovacuum_worker_slots)));
    3503        1492 : }

Generated by: LCOV version 1.16