4 * All the statistics collector stuff hacked up in one big, ugly file.
6 * TODO: - Separate collector, postmaster and backend stuff
7 * into different files.
9 * - Add some automatic call for pgstat vacuuming.
11 * - Add a pgstat config column to pg_database, so this
12 * entire thing can be enabled/disabled on a per db basis.
14 * Copyright (c) 2001-2010, PostgreSQL Global Development Group
16 * $PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.199 2010/01/28 14:25:41 mha Exp $
23 #include <sys/param.h>
25 #include <sys/socket.h>
27 #include <netinet/in.h>
28 #include <arpa/inet.h>
34 #ifdef HAVE_SYS_POLL_H
40 #include "access/heapam.h"
41 #include "access/transam.h"
42 #include "access/twophase_rmgr.h"
43 #include "access/xact.h"
44 #include "catalog/pg_database.h"
45 #include "catalog/pg_proc.h"
47 #include "libpq/libpq.h"
48 #include "libpq/pqsignal.h"
49 #include "mb/pg_wchar.h"
50 #include "miscadmin.h"
52 #include "postmaster/autovacuum.h"
53 #include "postmaster/fork_process.h"
54 #include "postmaster/postmaster.h"
55 #include "storage/backendid.h"
56 #include "storage/fd.h"
57 #include "storage/ipc.h"
58 #include "storage/pg_shmem.h"
59 #include "storage/pmsignal.h"
60 #include "utils/guc.h"
61 #include "utils/memutils.h"
62 #include "utils/ps_status.h"
63 #include "utils/rel.h"
64 #include "utils/tqual.h"
68 * Paths for the statistics files (relative to installation's $PGDATA).
71 #define PGSTAT_STAT_PERMANENT_FILENAME "global/pgstat.stat"
72 #define PGSTAT_STAT_PERMANENT_TMPFILE "global/pgstat.tmp"
78 #define PGSTAT_STAT_INTERVAL 500 /* Minimum time between stats file
79 * updates; in milliseconds. */
81 #define PGSTAT_RETRY_DELAY 10 /* How long to wait between statistics
82 * update requests; in milliseconds. */
84 #define PGSTAT_MAX_WAIT_TIME 5000 /* Maximum time to wait for a stats
85 * file update; in milliseconds. */
87 #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a
88 * failed statistics collector; in
91 #define PGSTAT_SELECT_TIMEOUT 2 /* How often to check for postmaster
92 * death; in seconds. */
94 #define PGSTAT_POLL_LOOP_COUNT (PGSTAT_MAX_WAIT_TIME / PGSTAT_RETRY_DELAY)
98 * The initial size hints for the hash tables used in the collector.
101 #define PGSTAT_DB_HASH_SIZE 16
102 #define PGSTAT_TAB_HASH_SIZE 512
103 #define PGSTAT_FUNCTION_HASH_SIZE 512
110 bool pgstat_track_activities = false;
111 bool pgstat_track_counts = false;
112 int pgstat_track_functions = TRACK_FUNC_OFF;
113 int pgstat_track_activity_query_size = 1024;
116 * Built from GUC parameter
119 char *pgstat_stat_filename = NULL;
120 char *pgstat_stat_tmpname = NULL;
123 * BgWriter global statistics counters (unused in other processes).
124 * Stored directly in a stats message structure so it can be sent
125 * without needing to copy things around. We assume this inits to zeroes.
127 PgStat_MsgBgWriter BgWriterStats;
133 NON_EXEC_STATIC pgsocket pgStatSock = PGINVALID_SOCKET;
135 static struct sockaddr_storage pgStatAddr;
137 static time_t last_pgstat_start_time;
139 static bool pgStatRunningInCollector = false;
142 * Structures in which backends store per-table info that's waiting to be
143 * sent to the collector.
145 * NOTE: once allocated, TabStatusArray structures are never moved or deleted
146 * for the life of the backend. Also, we zero out the t_id fields of the
147 * contained PgStat_TableStatus structs whenever they are not actively in use.
148 * This allows relcache pgstat_info pointers to be treated as long-lived data,
149 * avoiding repeated searches in pgstat_initstats() when a relation is
150 * repeatedly opened during a transaction.
152 #define TABSTAT_QUANTUM 100 /* we alloc this many at a time */
154 typedef struct TabStatusArray
156 struct TabStatusArray *tsa_next; /* link to next array, if any */
157 int tsa_used; /* # entries currently used */
158 PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */
161 static TabStatusArray *pgStatTabList = NULL;
164 * Backends store per-function info that's waiting to be sent to the collector
165 * in this hash table (indexed by function OID).
167 static HTAB *pgStatFunctions = NULL;
170 * Indicates if backend has some function stats that it hasn't yet
171 * sent to the collector.
173 static bool have_function_stats = false;
176 * Tuple insertion/deletion counts for an open transaction can't be propagated
177 * into PgStat_TableStatus counters until we know if it is going to commit
178 * or abort. Hence, we keep these counts in per-subxact structs that live
179 * in TopTransactionContext. This data structure is designed on the assumption
180 * that subxacts won't usually modify very many tables.
182 typedef struct PgStat_SubXactStatus
184 int nest_level; /* subtransaction nest level */
185 struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
186 PgStat_TableXactStatus *first; /* head of list for this subxact */
187 } PgStat_SubXactStatus;
189 static PgStat_SubXactStatus *pgStatXactStack = NULL;
191 static int pgStatXactCommit = 0;
192 static int pgStatXactRollback = 0;
194 /* Record that's written to 2PC state file when pgstat state is persisted */
195 typedef struct TwoPhasePgStatRecord
197 PgStat_Counter tuples_inserted; /* tuples inserted in xact */
198 PgStat_Counter tuples_updated; /* tuples updated in xact */
199 PgStat_Counter tuples_deleted; /* tuples deleted in xact */
200 Oid t_id; /* table's OID */
201 bool t_shared; /* is it a shared catalog? */
202 } TwoPhasePgStatRecord;
205 * Info about current "snapshot" of stats file
207 static MemoryContext pgStatLocalContext = NULL;
208 static HTAB *pgStatDBHash = NULL;
209 static PgBackendStatus *localBackendStatusTable = NULL;
210 static int localNumBackends = 0;
213 * Cluster wide statistics, kept in the stats collector.
214 * Contains statistics that are not collected per database
217 static PgStat_GlobalStats globalStats;
219 /* Last time the collector successfully wrote the stats file */
220 static TimestampTz last_statwrite;
222 /* Latest statistics request time from backends */
223 static TimestampTz last_statrequest;
225 static volatile bool need_exit = false;
226 static volatile bool got_SIGHUP = false;
229 * Total time charged to functions so far in the current backend.
230 * We use this to help separate "self" and "other" time charges.
231 * (We assume this initializes to zero.)
233 static instr_time total_func_time;
237 * Local function forward declarations
241 static pid_t pgstat_forkexec(void);
244 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]);
245 static void pgstat_exit(SIGNAL_ARGS);
246 static void pgstat_beshutdown_hook(int code, Datum arg);
247 static void pgstat_sighup_handler(SIGNAL_ARGS);
249 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
250 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
251 Oid tableoid, bool create);
252 static void pgstat_write_statsfile(bool permanent);
253 static HTAB *pgstat_read_statsfile(Oid onlydb, bool permanent);
254 static void backend_read_statsfile(void);
255 static void pgstat_read_current_status(void);
257 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
258 static void pgstat_send_funcstats(void);
259 static HTAB *pgstat_collect_oids(Oid catalogid);
261 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
263 static void pgstat_setup_memcxt(void);
265 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
266 static void pgstat_send(void *msg, int len);
268 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
269 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
270 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
271 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
272 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
273 static void pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len);
274 static void pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len);
275 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
276 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
277 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
278 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
279 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
280 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
283 /* ------------------------------------------------------------
284 * Public functions called from postmaster follow
285 * ------------------------------------------------------------
291 * Called from postmaster at startup. Create the resources required
292 * by the statistics collector process. If unable to do so, do not
293 * fail --- better to let the postmaster start with stats collection
300 ACCEPT_TYPE_ARG3 alen;
301 struct addrinfo *addrs = NULL,
311 #define TESTBYTEVAL ((char) 199)
314 * Create the UDP socket for sending and receiving statistic messages
316 hints.ai_flags = AI_PASSIVE;
317 hints.ai_family = PF_UNSPEC;
318 hints.ai_socktype = SOCK_DGRAM;
319 hints.ai_protocol = 0;
320 hints.ai_addrlen = 0;
321 hints.ai_addr = NULL;
322 hints.ai_canonname = NULL;
323 hints.ai_next = NULL;
324 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
328 (errmsg("could not resolve \"localhost\": %s",
329 gai_strerror(ret))));
334 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
335 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
336 * when kernel will reject IPv6). Worse, the failure may occur at the
337 * bind() or perhaps even connect() stage. So we must loop through the
338 * results till we find a working combination. We will generate LOG
339 * messages, but no error, for bogus combinations.
341 for (addr = addrs; addr; addr = addr->ai_next)
343 #ifdef HAVE_UNIX_SOCKETS
344 /* Ignore AF_UNIX sockets, if any are returned. */
345 if (addr->ai_family == AF_UNIX)
351 (errmsg("trying another address for the statistics collector")));
356 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) < 0)
359 (errcode_for_socket_access(),
360 errmsg("could not create socket for statistics collector: %m")));
365 * Bind it to a kernel assigned port on localhost and get the assigned
366 * port via getsockname().
368 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
371 (errcode_for_socket_access(),
372 errmsg("could not bind socket for statistics collector: %m")));
373 closesocket(pgStatSock);
374 pgStatSock = PGINVALID_SOCKET;
378 alen = sizeof(pgStatAddr);
379 if (getsockname(pgStatSock, (struct sockaddr *) & pgStatAddr, &alen) < 0)
382 (errcode_for_socket_access(),
383 errmsg("could not get address of socket for statistics collector: %m")));
384 closesocket(pgStatSock);
385 pgStatSock = PGINVALID_SOCKET;
390 * Connect the socket to its own address. This saves a few cycles by
391 * not having to respecify the target address on every send. This also
392 * provides a kernel-level check that only packets from this same
393 * address will be received.
395 if (connect(pgStatSock, (struct sockaddr *) & pgStatAddr, alen) < 0)
398 (errcode_for_socket_access(),
399 errmsg("could not connect socket for statistics collector: %m")));
400 closesocket(pgStatSock);
401 pgStatSock = PGINVALID_SOCKET;
406 * Try to send and receive a one-byte test message on the socket. This
407 * is to catch situations where the socket can be created but will not
408 * actually pass data (for instance, because kernel packet filtering
411 test_byte = TESTBYTEVAL;
414 if (send(pgStatSock, &test_byte, 1, 0) != 1)
417 goto retry1; /* if interrupted, just retry */
419 (errcode_for_socket_access(),
420 errmsg("could not send test message on socket for statistics collector: %m")));
421 closesocket(pgStatSock);
422 pgStatSock = PGINVALID_SOCKET;
427 * There could possibly be a little delay before the message can be
428 * received. We arbitrarily allow up to half a second before deciding
431 for (;;) /* need a loop to handle EINTR */
434 FD_SET (pgStatSock, &rset);
438 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
439 if (sel_res >= 0 || errno != EINTR)
445 (errcode_for_socket_access(),
446 errmsg("select() failed in statistics collector: %m")));
447 closesocket(pgStatSock);
448 pgStatSock = PGINVALID_SOCKET;
451 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
454 * This is the case we actually think is likely, so take pains to
455 * give a specific message for it.
457 * errno will not be set meaningfully here, so don't use it.
460 (errcode(ERRCODE_CONNECTION_FAILURE),
461 errmsg("test message did not get through on socket for statistics collector")));
462 closesocket(pgStatSock);
463 pgStatSock = PGINVALID_SOCKET;
467 test_byte++; /* just make sure variable is changed */
470 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
473 goto retry2; /* if interrupted, just retry */
475 (errcode_for_socket_access(),
476 errmsg("could not receive test message on socket for statistics collector: %m")));
477 closesocket(pgStatSock);
478 pgStatSock = PGINVALID_SOCKET;
482 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
485 (errcode(ERRCODE_INTERNAL_ERROR),
486 errmsg("incorrect test message transmission on socket for statistics collector")));
487 closesocket(pgStatSock);
488 pgStatSock = PGINVALID_SOCKET;
492 /* If we get here, we have a working socket */
496 /* Did we find a working address? */
497 if (!addr || pgStatSock < 0)
501 * Set the socket to non-blocking IO. This ensures that if the collector
502 * falls behind, statistics messages will be discarded; backends won't
503 * block waiting to send messages to the collector.
505 if (!pg_set_noblock(pgStatSock))
508 (errcode_for_socket_access(),
509 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
513 pg_freeaddrinfo_all(hints.ai_family, addrs);
519 (errmsg("disabling statistics collector for lack of working socket")));
522 pg_freeaddrinfo_all(hints.ai_family, addrs);
525 closesocket(pgStatSock);
526 pgStatSock = PGINVALID_SOCKET;
529 * Adjust GUC variables to suppress useless activity, and for debugging
530 * purposes (seeing track_counts off is a clue that we failed here). We
531 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
532 * on from postgresql.conf without a restart.
534 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
538 * pgstat_reset_all() -
540 * Remove the stats file. This is currently used only if WAL
541 * recovery is needed after a crash.
544 pgstat_reset_all(void)
546 unlink(pgstat_stat_filename);
547 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
553 * pgstat_forkexec() -
555 * Format up the arglist for, then fork and exec, statistics collector process
558 pgstat_forkexec(void)
563 av[ac++] = "postgres";
564 av[ac++] = "--forkcol";
565 av[ac++] = NULL; /* filled in by postmaster_forkexec */
568 Assert(ac < lengthof(av));
570 return postmaster_forkexec(ac, av);
572 #endif /* EXEC_BACKEND */
578 * Called from postmaster at startup or after an existing collector
579 * died. Attempt to fire up a fresh statistics collector.
581 * Returns PID of child process, or 0 if fail.
583 * Note: if fail, we will be called again from the postmaster main loop.
592 * Check that the socket is there, else pgstat_init failed and we can do
599 * Do nothing if too soon since last collector start. This is a safety
600 * valve to protect against continuous respawn attempts if the collector
601 * is dying immediately at launch. Note that since we will be re-called
602 * from the postmaster main loop, we will get another chance later.
604 curtime = time(NULL);
605 if ((unsigned int) (curtime - last_pgstat_start_time) <
606 (unsigned int) PGSTAT_RESTART_INTERVAL)
608 last_pgstat_start_time = curtime;
611 * Okay, fork off the collector.
614 switch ((pgStatPid = pgstat_forkexec()))
616 switch ((pgStatPid = fork_process()))
621 (errmsg("could not fork statistics collector: %m")));
626 /* in postmaster child ... */
627 /* Close the postmaster's sockets */
628 ClosePostmasterPorts(false);
630 /* Lose the postmaster's on-exit routines */
633 /* Drop our connection to postmaster's shared memory, as well */
634 PGSharedMemoryDetach();
636 PgstatCollectorMain(0, NULL);
641 return (int) pgStatPid;
644 /* shouldn't get here */
649 allow_immediate_pgstat_restart(void)
651 last_pgstat_start_time = 0;
654 /* ------------------------------------------------------------
655 * Public functions used by backends follow
656 *------------------------------------------------------------
661 * pgstat_report_stat() -
663 * Called from tcop/postgres.c to send the so far collected per-table
664 * and function usage statistics to the collector. Note that this is
665 * called only when not within a transaction, so it is fair to use
666 * transaction stop time as an approximation of current time.
670 pgstat_report_stat(bool force)
672 /* we assume this inits to all zeroes: */
673 static const PgStat_TableCounts all_zeroes;
674 static TimestampTz last_report = 0;
677 PgStat_MsgTabstat regular_msg;
678 PgStat_MsgTabstat shared_msg;
682 /* Don't expend a clock check if nothing to do */
683 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0)
684 && !have_function_stats)
688 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
689 * msec since we last sent one, or the caller wants to force stats out.
691 now = GetCurrentTransactionStopTimestamp();
693 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
698 * Scan through the TabStatusArray struct(s) to find tables that actually
699 * have counts, and build messages to send. We have to separate shared
700 * relations from regular ones because the databaseid field in the message
701 * header has to depend on that.
703 regular_msg.m_databaseid = MyDatabaseId;
704 shared_msg.m_databaseid = InvalidOid;
705 regular_msg.m_nentries = 0;
706 shared_msg.m_nentries = 0;
708 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
710 for (i = 0; i < tsa->tsa_used; i++)
712 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
713 PgStat_MsgTabstat *this_msg;
714 PgStat_TableEntry *this_ent;
716 /* Shouldn't have any pending transaction-dependent counts */
717 Assert(entry->trans == NULL);
720 * Ignore entries that didn't accumulate any actual counts, such
721 * as indexes that were opened by the planner but not used.
723 if (memcmp(&entry->t_counts, &all_zeroes,
724 sizeof(PgStat_TableCounts)) == 0)
728 * OK, insert data into the appropriate message, and send if full.
730 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
731 this_ent = &this_msg->m_entry[this_msg->m_nentries];
732 this_ent->t_id = entry->t_id;
733 memcpy(&this_ent->t_counts, &entry->t_counts,
734 sizeof(PgStat_TableCounts));
735 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
737 pgstat_send_tabstat(this_msg);
738 this_msg->m_nentries = 0;
741 /* zero out TableStatus structs after use */
742 MemSet(tsa->tsa_entries, 0,
743 tsa->tsa_used * sizeof(PgStat_TableStatus));
748 * Send partial messages. If force is true, make sure that any pending
749 * xact commit/abort gets counted, even if no table stats to send.
751 if (regular_msg.m_nentries > 0 ||
752 (force && (pgStatXactCommit > 0 || pgStatXactRollback > 0)))
753 pgstat_send_tabstat(®ular_msg);
754 if (shared_msg.m_nentries > 0)
755 pgstat_send_tabstat(&shared_msg);
757 /* Now, send function statistics */
758 pgstat_send_funcstats();
762 * Subroutine for pgstat_report_stat: finish and send a tabstat message
765 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
770 /* It's unlikely we'd get here with no socket, but maybe not impossible */
775 * Report accumulated xact commit/rollback whenever we send a normal
778 if (OidIsValid(tsmsg->m_databaseid))
780 tsmsg->m_xact_commit = pgStatXactCommit;
781 tsmsg->m_xact_rollback = pgStatXactRollback;
782 pgStatXactCommit = 0;
783 pgStatXactRollback = 0;
787 tsmsg->m_xact_commit = 0;
788 tsmsg->m_xact_rollback = 0;
791 n = tsmsg->m_nentries;
792 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
793 n * sizeof(PgStat_TableEntry);
795 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
796 pgstat_send(tsmsg, len);
800 * Subroutine for pgstat_report_stat: populate and send a function stat message
803 pgstat_send_funcstats(void)
805 /* we assume this inits to all zeroes: */
806 static const PgStat_FunctionCounts all_zeroes;
808 PgStat_MsgFuncstat msg;
809 PgStat_BackendFunctionEntry *entry;
810 HASH_SEQ_STATUS fstat;
812 if (pgStatFunctions == NULL)
815 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
816 msg.m_databaseid = MyDatabaseId;
819 hash_seq_init(&fstat, pgStatFunctions);
820 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
822 PgStat_FunctionEntry *m_ent;
824 /* Skip it if no counts accumulated since last time */
825 if (memcmp(&entry->f_counts, &all_zeroes,
826 sizeof(PgStat_FunctionCounts)) == 0)
829 /* need to convert format of time accumulators */
830 m_ent = &msg.m_entry[msg.m_nentries];
831 m_ent->f_id = entry->f_id;
832 m_ent->f_numcalls = entry->f_counts.f_numcalls;
833 m_ent->f_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time);
834 m_ent->f_time_self = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time_self);
836 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
838 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
839 msg.m_nentries * sizeof(PgStat_FunctionEntry));
843 /* reset the entry's counts */
844 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
847 if (msg.m_nentries > 0)
848 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
849 msg.m_nentries * sizeof(PgStat_FunctionEntry));
851 have_function_stats = false;
856 * pgstat_vacuum_stat() -
858 * Will tell the collector about objects he can get rid of.
862 pgstat_vacuum_stat(void)
865 PgStat_MsgTabpurge msg;
866 PgStat_MsgFuncpurge f_msg;
867 HASH_SEQ_STATUS hstat;
868 PgStat_StatDBEntry *dbentry;
869 PgStat_StatTabEntry *tabentry;
870 PgStat_StatFuncEntry *funcentry;
877 * If not done for this transaction, read the statistics collector stats
878 * file into some hash tables.
880 backend_read_statsfile();
883 * Read pg_database and make a list of OIDs of all existing databases
885 htab = pgstat_collect_oids(DatabaseRelationId);
888 * Search the database hash table for dead databases and tell the
889 * collector to drop them.
891 hash_seq_init(&hstat, pgStatDBHash);
892 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
894 Oid dbid = dbentry->databaseid;
896 CHECK_FOR_INTERRUPTS();
898 /* the DB entry for shared tables (with InvalidOid) is never dropped */
899 if (OidIsValid(dbid) &&
900 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
901 pgstat_drop_database(dbid);
908 * Lookup our own database entry; if not found, nothing more to do.
910 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
911 (void *) &MyDatabaseId,
913 if (dbentry == NULL || dbentry->tables == NULL)
917 * Similarly to above, make a list of all known relations in this DB.
919 htab = pgstat_collect_oids(RelationRelationId);
922 * Initialize our messages table counter to zero
927 * Check for all tables listed in stats hashtable if they still exist.
929 hash_seq_init(&hstat, dbentry->tables);
930 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
932 Oid tabid = tabentry->tableid;
934 CHECK_FOR_INTERRUPTS();
936 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
940 * Not there, so add this table's Oid to the message
942 msg.m_tableid[msg.m_nentries++] = tabid;
945 * If the message is full, send it out and reinitialize to empty
947 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
949 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
950 +msg.m_nentries * sizeof(Oid);
952 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
953 msg.m_databaseid = MyDatabaseId;
954 pgstat_send(&msg, len);
963 if (msg.m_nentries > 0)
965 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
966 +msg.m_nentries * sizeof(Oid);
968 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
969 msg.m_databaseid = MyDatabaseId;
970 pgstat_send(&msg, len);
977 * Now repeat the above steps for functions. However, we needn't bother
978 * in the common case where no function stats are being collected.
980 if (dbentry->functions != NULL &&
981 hash_get_num_entries(dbentry->functions) > 0)
983 htab = pgstat_collect_oids(ProcedureRelationId);
985 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
986 f_msg.m_databaseid = MyDatabaseId;
987 f_msg.m_nentries = 0;
989 hash_seq_init(&hstat, dbentry->functions);
990 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
992 Oid funcid = funcentry->functionid;
994 CHECK_FOR_INTERRUPTS();
996 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
1000 * Not there, so add this function's Oid to the message
1002 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
1005 * If the message is full, send it out and reinitialize to empty
1007 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1009 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1010 +f_msg.m_nentries * sizeof(Oid);
1012 pgstat_send(&f_msg, len);
1014 f_msg.m_nentries = 0;
1021 if (f_msg.m_nentries > 0)
1023 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1024 +f_msg.m_nentries * sizeof(Oid);
1026 pgstat_send(&f_msg, len);
1035 * pgstat_collect_oids() -
1037 * Collect the OIDs of all objects listed in the specified system catalog
1038 * into a temporary hash table. Caller should hash_destroy the result
1039 * when done with it. (However, we make the table in CurrentMemoryContext
1040 * so that it will be freed properly in event of an error.)
1044 pgstat_collect_oids(Oid catalogid)
1052 memset(&hash_ctl, 0, sizeof(hash_ctl));
1053 hash_ctl.keysize = sizeof(Oid);
1054 hash_ctl.entrysize = sizeof(Oid);
1055 hash_ctl.hash = oid_hash;
1056 hash_ctl.hcxt = CurrentMemoryContext;
1057 htab = hash_create("Temporary table of OIDs",
1058 PGSTAT_TAB_HASH_SIZE,
1060 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
1062 rel = heap_open(catalogid, AccessShareLock);
1063 scan = heap_beginscan(rel, SnapshotNow, 0, NULL);
1064 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1066 Oid thisoid = HeapTupleGetOid(tup);
1068 CHECK_FOR_INTERRUPTS();
1070 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1073 heap_close(rel, AccessShareLock);
1080 * pgstat_drop_database() -
1082 * Tell the collector that we just dropped a database.
1083 * (If the message gets lost, we will still clean the dead DB eventually
1084 * via future invocations of pgstat_vacuum_stat().)
1088 pgstat_drop_database(Oid databaseid)
1090 PgStat_MsgDropdb msg;
1095 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1096 msg.m_databaseid = databaseid;
1097 pgstat_send(&msg, sizeof(msg));
1102 * pgstat_drop_relation() -
1104 * Tell the collector that we just dropped a relation.
1105 * (If the message gets lost, we will still clean the dead entry eventually
1106 * via future invocations of pgstat_vacuum_stat().)
1108 * Currently not used for lack of any good place to call it; we rely
1109 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1114 pgstat_drop_relation(Oid relid)
1116 PgStat_MsgTabpurge msg;
1122 msg.m_tableid[0] = relid;
1125 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) +sizeof(Oid);
1127 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1128 msg.m_databaseid = MyDatabaseId;
1129 pgstat_send(&msg, len);
1131 #endif /* NOT_USED */
1135 * pgstat_reset_counters() -
1137 * Tell the statistics collector to reset counters for our database.
1141 pgstat_reset_counters(void)
1143 PgStat_MsgResetcounter msg;
1150 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1151 errmsg("must be superuser to reset statistics counters")));
1153 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1154 msg.m_databaseid = MyDatabaseId;
1155 pgstat_send(&msg, sizeof(msg));
1159 * pgstat_reset_shared_counters() -
1161 * Tell the statistics collector to reset cluster-wide shared counters.
1165 pgstat_reset_shared_counters(const char *target)
1167 PgStat_MsgResetsharedcounter msg;
1174 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1175 errmsg("must be superuser to reset statistics counters")));
1177 if (strcmp(target, "bgwriter") == 0)
1178 msg.m_resettarget = RESET_BGWRITER;
1182 (errcode(ERRCODE_SYNTAX_ERROR),
1183 errmsg("unrecognized reset target: '%s'", target),
1184 errhint("allowed targets are 'bgwriter'.")));
1187 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSHAREDCOUNTER);
1188 pgstat_send(&msg, sizeof(msg));
1192 * pgstat_reset_single_counter() -
1194 * Tell the statistics collector to reset a single counter.
1197 void pgstat_reset_single_counter(Oid objoid, PgStat_Single_Reset_Type type)
1199 PgStat_MsgResetsinglecounter msg;
1206 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1207 errmsg("must be superuser to reset statistics counters")));
1209 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSINGLECOUNTER);
1210 msg.m_databaseid = MyDatabaseId;
1211 msg.m_resettype = type;
1212 msg.m_objectid = objoid;
1214 pgstat_send(&msg, sizeof(msg));
1218 * pgstat_report_autovac() -
1220 * Called from autovacuum.c to report startup of an autovacuum process.
1221 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1222 * the db OID must be passed in, instead.
1226 pgstat_report_autovac(Oid dboid)
1228 PgStat_MsgAutovacStart msg;
1233 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1234 msg.m_databaseid = dboid;
1235 msg.m_start_time = GetCurrentTimestamp();
1237 pgstat_send(&msg, sizeof(msg));
1242 * pgstat_report_vacuum() -
1244 * Tell the collector about the table we just vacuumed.
1248 pgstat_report_vacuum(Oid tableoid, bool shared, bool adopt_counts,
1249 PgStat_Counter tuples)
1251 PgStat_MsgVacuum msg;
1253 if (pgStatSock < 0 || !pgstat_track_counts)
1256 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1257 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1258 msg.m_tableoid = tableoid;
1259 msg.m_adopt_counts = adopt_counts;
1260 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1261 msg.m_vacuumtime = GetCurrentTimestamp();
1262 msg.m_tuples = tuples;
1263 pgstat_send(&msg, sizeof(msg));
1267 * pgstat_report_analyze() -
1269 * Tell the collector about the table we just analyzed.
1273 pgstat_report_analyze(Relation rel, bool adopt_counts,
1274 PgStat_Counter livetuples, PgStat_Counter deadtuples)
1276 PgStat_MsgAnalyze msg;
1278 if (pgStatSock < 0 || !pgstat_track_counts)
1282 * Unlike VACUUM, ANALYZE might be running inside a transaction that has
1283 * already inserted and/or deleted rows in the target table. ANALYZE will
1284 * have counted such rows as live or dead respectively. Because we will
1285 * report our counts of such rows at transaction end, we should subtract
1286 * off these counts from what we send to the collector now, else they'll
1287 * be double-counted after commit. (This approach also ensures that the
1288 * collector ends up with the right numbers if we abort instead of
1291 if (rel->pgstat_info != NULL)
1293 PgStat_TableXactStatus *trans;
1295 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1297 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1298 deadtuples -= trans->tuples_updated + trans->tuples_deleted;
1300 /* count stuff inserted by already-aborted subxacts, too */
1301 deadtuples -= rel->pgstat_info->t_counts.t_delta_dead_tuples;
1302 /* Since ANALYZE's counts are estimates, we could have underflowed */
1303 livetuples = Max(livetuples, 0);
1304 deadtuples = Max(deadtuples, 0);
1307 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1308 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1309 msg.m_tableoid = RelationGetRelid(rel);
1310 msg.m_adopt_counts = adopt_counts;
1311 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1312 msg.m_analyzetime = GetCurrentTimestamp();
1313 msg.m_live_tuples = livetuples;
1314 msg.m_dead_tuples = deadtuples;
1315 pgstat_send(&msg, sizeof(msg));
1322 * Send some junk data to the collector to increase traffic.
1328 PgStat_MsgDummy msg;
1333 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1334 pgstat_send(&msg, sizeof(msg));
1338 * pgstat_send_inquiry() -
1340 * Notify collector that we need fresh data.
1341 * ts specifies the minimum acceptable timestamp for the stats file.
1345 pgstat_send_inquiry(TimestampTz ts)
1347 PgStat_MsgInquiry msg;
1349 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1350 msg.inquiry_time = ts;
1351 pgstat_send(&msg, sizeof(msg));
1356 * Initialize function call usage data.
1357 * Called by the executor before invoking a function.
1360 pgstat_init_function_usage(FunctionCallInfoData *fcinfo,
1361 PgStat_FunctionCallUsage *fcu)
1363 PgStat_BackendFunctionEntry *htabent;
1366 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1368 /* stats not wanted */
1373 if (!pgStatFunctions)
1375 /* First time through - initialize function stat table */
1378 memset(&hash_ctl, 0, sizeof(hash_ctl));
1379 hash_ctl.keysize = sizeof(Oid);
1380 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1381 hash_ctl.hash = oid_hash;
1382 pgStatFunctions = hash_create("Function stat entries",
1383 PGSTAT_FUNCTION_HASH_SIZE,
1385 HASH_ELEM | HASH_FUNCTION);
1388 /* Get the stats entry for this function, create if necessary */
1389 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1390 HASH_ENTER, &found);
1392 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1394 fcu->fs = &htabent->f_counts;
1396 /* save stats for this function, later used to compensate for recursion */
1397 fcu->save_f_time = htabent->f_counts.f_time;
1399 /* save current backend-wide total time */
1400 fcu->save_total = total_func_time;
1402 /* get clock time as of function start */
1403 INSTR_TIME_SET_CURRENT(fcu->f_start);
1407 * Calculate function call usage and update stat counters.
1408 * Called by the executor after invoking a function.
1410 * In the case of a set-returning function that runs in value-per-call mode,
1411 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1412 * calls for what the user considers a single call of the function. The
1413 * finalize flag should be TRUE on the last call.
1416 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1418 PgStat_FunctionCounts *fs = fcu->fs;
1420 instr_time f_others;
1423 /* stats not wanted? */
1427 /* total elapsed time in this function call */
1428 INSTR_TIME_SET_CURRENT(f_total);
1429 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1431 /* self usage: elapsed minus anything already charged to other calls */
1432 f_others = total_func_time;
1433 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1435 INSTR_TIME_SUBTRACT(f_self, f_others);
1437 /* update backend-wide total time */
1438 INSTR_TIME_ADD(total_func_time, f_self);
1441 * Compute the new total f_time as the total elapsed time added to the
1442 * pre-call value of f_time. This is necessary to avoid double-counting
1443 * any time taken by recursive calls of myself. (We do not need any
1444 * similar kluge for self time, since that already excludes any recursive
1447 INSTR_TIME_ADD(f_total, fcu->save_f_time);
1449 /* update counters in function stats table */
1452 fs->f_time = f_total;
1453 INSTR_TIME_ADD(fs->f_time_self, f_self);
1455 /* indicate that we have something to send */
1456 have_function_stats = true;
1461 * pgstat_initstats() -
1463 * Initialize a relcache entry to count access statistics.
1464 * Called whenever a relation is opened.
1466 * We assume that a relcache entry's pgstat_info field is zeroed by
1467 * relcache.c when the relcache entry is made; thereafter it is long-lived
1468 * data. We can avoid repeated searches of the TabStatus arrays when the
1469 * same relation is touched repeatedly within a transaction.
1473 pgstat_initstats(Relation rel)
1475 Oid rel_id = rel->rd_id;
1476 char relkind = rel->rd_rel->relkind;
1478 /* We only count stats for things that have storage */
1479 if (!(relkind == RELKIND_RELATION ||
1480 relkind == RELKIND_INDEX ||
1481 relkind == RELKIND_TOASTVALUE ||
1482 relkind == RELKIND_SEQUENCE))
1484 rel->pgstat_info = NULL;
1488 if (pgStatSock < 0 || !pgstat_track_counts)
1490 /* We're not counting at all */
1491 rel->pgstat_info = NULL;
1496 * If we already set up this relation in the current transaction, nothing
1499 if (rel->pgstat_info != NULL &&
1500 rel->pgstat_info->t_id == rel_id)
1503 /* Else find or make the PgStat_TableStatus entry, and update link */
1504 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1508 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1510 static PgStat_TableStatus *
1511 get_tabstat_entry(Oid rel_id, bool isshared)
1513 PgStat_TableStatus *entry;
1514 TabStatusArray *tsa;
1515 TabStatusArray *prev_tsa;
1519 * Search the already-used tabstat slots for this relation.
1522 for (tsa = pgStatTabList; tsa != NULL; prev_tsa = tsa, tsa = tsa->tsa_next)
1524 for (i = 0; i < tsa->tsa_used; i++)
1526 entry = &tsa->tsa_entries[i];
1527 if (entry->t_id == rel_id)
1531 if (tsa->tsa_used < TABSTAT_QUANTUM)
1534 * It must not be present, but we found a free slot instead. Fine,
1535 * let's use this one. We assume the entry was already zeroed,
1536 * either at creation or after last use.
1538 entry = &tsa->tsa_entries[tsa->tsa_used++];
1539 entry->t_id = rel_id;
1540 entry->t_shared = isshared;
1546 * We ran out of tabstat slots, so allocate more. Be sure they're zeroed.
1548 tsa = (TabStatusArray *) MemoryContextAllocZero(TopMemoryContext,
1549 sizeof(TabStatusArray));
1551 prev_tsa->tsa_next = tsa;
1553 pgStatTabList = tsa;
1556 * Use the first entry of the new TabStatusArray.
1558 entry = &tsa->tsa_entries[tsa->tsa_used++];
1559 entry->t_id = rel_id;
1560 entry->t_shared = isshared;
1565 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1567 static PgStat_SubXactStatus *
1568 get_tabstat_stack_level(int nest_level)
1570 PgStat_SubXactStatus *xact_state;
1572 xact_state = pgStatXactStack;
1573 if (xact_state == NULL || xact_state->nest_level != nest_level)
1575 xact_state = (PgStat_SubXactStatus *)
1576 MemoryContextAlloc(TopTransactionContext,
1577 sizeof(PgStat_SubXactStatus));
1578 xact_state->nest_level = nest_level;
1579 xact_state->prev = pgStatXactStack;
1580 xact_state->first = NULL;
1581 pgStatXactStack = xact_state;
1587 * add_tabstat_xact_level - add a new (sub)transaction state record
1590 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1592 PgStat_SubXactStatus *xact_state;
1593 PgStat_TableXactStatus *trans;
1596 * If this is the first rel to be modified at the current nest level, we
1597 * first have to push a transaction stack entry.
1599 xact_state = get_tabstat_stack_level(nest_level);
1601 /* Now make a per-table stack entry */
1602 trans = (PgStat_TableXactStatus *)
1603 MemoryContextAllocZero(TopTransactionContext,
1604 sizeof(PgStat_TableXactStatus));
1605 trans->nest_level = nest_level;
1606 trans->upper = pgstat_info->trans;
1607 trans->parent = pgstat_info;
1608 trans->next = xact_state->first;
1609 xact_state->first = trans;
1610 pgstat_info->trans = trans;
1614 * pgstat_count_heap_insert - count a tuple insertion
1617 pgstat_count_heap_insert(Relation rel)
1619 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1621 if (pgstat_track_counts && pgstat_info != NULL)
1623 /* We have to log the effect at the proper transactional level */
1624 int nest_level = GetCurrentTransactionNestLevel();
1626 if (pgstat_info->trans == NULL ||
1627 pgstat_info->trans->nest_level != nest_level)
1628 add_tabstat_xact_level(pgstat_info, nest_level);
1630 pgstat_info->trans->tuples_inserted++;
1635 * pgstat_count_heap_update - count a tuple update
1638 pgstat_count_heap_update(Relation rel, bool hot)
1640 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1642 if (pgstat_track_counts && pgstat_info != NULL)
1644 /* We have to log the effect at the proper transactional level */
1645 int nest_level = GetCurrentTransactionNestLevel();
1647 if (pgstat_info->trans == NULL ||
1648 pgstat_info->trans->nest_level != nest_level)
1649 add_tabstat_xact_level(pgstat_info, nest_level);
1651 pgstat_info->trans->tuples_updated++;
1653 /* t_tuples_hot_updated is nontransactional, so just advance it */
1655 pgstat_info->t_counts.t_tuples_hot_updated++;
1660 * pgstat_count_heap_delete - count a tuple deletion
1663 pgstat_count_heap_delete(Relation rel)
1665 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1667 if (pgstat_track_counts && pgstat_info != NULL)
1669 /* We have to log the effect at the proper transactional level */
1670 int nest_level = GetCurrentTransactionNestLevel();
1672 if (pgstat_info->trans == NULL ||
1673 pgstat_info->trans->nest_level != nest_level)
1674 add_tabstat_xact_level(pgstat_info, nest_level);
1676 pgstat_info->trans->tuples_deleted++;
1681 * pgstat_update_heap_dead_tuples - update dead-tuples count
1683 * The semantics of this are that we are reporting the nontransactional
1684 * recovery of "delta" dead tuples; so t_delta_dead_tuples decreases
1685 * rather than increasing, and the change goes straight into the per-table
1686 * counter, not into transactional state.
1689 pgstat_update_heap_dead_tuples(Relation rel, int delta)
1691 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1693 if (pgstat_track_counts && pgstat_info != NULL)
1694 pgstat_info->t_counts.t_delta_dead_tuples -= delta;
1701 * Called from access/transam/xact.c at top-level transaction commit/abort.
1705 AtEOXact_PgStat(bool isCommit)
1707 PgStat_SubXactStatus *xact_state;
1710 * Count transaction commit or abort. (We use counters, not just bools,
1711 * in case the reporting message isn't sent right away.)
1716 pgStatXactRollback++;
1719 * Transfer transactional insert/update counts into the base tabstat
1720 * entries. We don't bother to free any of the transactional state, since
1721 * it's all in TopTransactionContext and will go away anyway.
1723 xact_state = pgStatXactStack;
1724 if (xact_state != NULL)
1726 PgStat_TableXactStatus *trans;
1728 Assert(xact_state->nest_level == 1);
1729 Assert(xact_state->prev == NULL);
1730 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1732 PgStat_TableStatus *tabstat;
1734 Assert(trans->nest_level == 1);
1735 Assert(trans->upper == NULL);
1736 tabstat = trans->parent;
1737 Assert(tabstat->trans == trans);
1738 /* count attempted actions regardless of commit/abort */
1739 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
1740 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
1741 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
1744 /* insert adds a live tuple, delete removes one */
1745 tabstat->t_counts.t_delta_live_tuples +=
1746 trans->tuples_inserted - trans->tuples_deleted;
1747 /* update and delete each create a dead tuple */
1748 tabstat->t_counts.t_delta_dead_tuples +=
1749 trans->tuples_updated + trans->tuples_deleted;
1750 /* insert, update, delete each count as one change event */
1751 tabstat->t_counts.t_changed_tuples +=
1752 trans->tuples_inserted + trans->tuples_updated +
1753 trans->tuples_deleted;
1757 /* inserted tuples are dead, deleted tuples are unaffected */
1758 tabstat->t_counts.t_delta_dead_tuples +=
1759 trans->tuples_inserted + trans->tuples_updated;
1760 /* an aborted xact generates no changed_tuple events */
1762 tabstat->trans = NULL;
1765 pgStatXactStack = NULL;
1767 /* Make sure any stats snapshot is thrown away */
1768 pgstat_clear_snapshot();
1772 * AtEOSubXact_PgStat
1774 * Called from access/transam/xact.c at subtransaction commit/abort.
1778 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
1780 PgStat_SubXactStatus *xact_state;
1783 * Transfer transactional insert/update counts into the next higher
1784 * subtransaction state.
1786 xact_state = pgStatXactStack;
1787 if (xact_state != NULL &&
1788 xact_state->nest_level >= nestDepth)
1790 PgStat_TableXactStatus *trans;
1791 PgStat_TableXactStatus *next_trans;
1793 /* delink xact_state from stack immediately to simplify reuse case */
1794 pgStatXactStack = xact_state->prev;
1796 for (trans = xact_state->first; trans != NULL; trans = next_trans)
1798 PgStat_TableStatus *tabstat;
1800 next_trans = trans->next;
1801 Assert(trans->nest_level == nestDepth);
1802 tabstat = trans->parent;
1803 Assert(tabstat->trans == trans);
1806 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
1808 trans->upper->tuples_inserted += trans->tuples_inserted;
1809 trans->upper->tuples_updated += trans->tuples_updated;
1810 trans->upper->tuples_deleted += trans->tuples_deleted;
1811 tabstat->trans = trans->upper;
1817 * When there isn't an immediate parent state, we can just
1818 * reuse the record instead of going through a
1819 * palloc/pfree pushup (this works since it's all in
1820 * TopTransactionContext anyway). We have to re-link it
1821 * into the parent level, though, and that might mean
1822 * pushing a new entry into the pgStatXactStack.
1824 PgStat_SubXactStatus *upper_xact_state;
1826 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
1827 trans->next = upper_xact_state->first;
1828 upper_xact_state->first = trans;
1829 trans->nest_level = nestDepth - 1;
1835 * On abort, update top-level tabstat counts, then forget
1836 * the subtransaction
1839 /* count attempted actions regardless of commit/abort */
1840 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
1841 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
1842 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
1843 /* inserted tuples are dead, deleted tuples are unaffected */
1844 tabstat->t_counts.t_delta_dead_tuples +=
1845 trans->tuples_inserted + trans->tuples_updated;
1846 tabstat->trans = trans->upper;
1857 * Save the transactional stats state at 2PC transaction prepare.
1859 * In this phase we just generate 2PC records for all the pending
1860 * transaction-dependent stats work.
1863 AtPrepare_PgStat(void)
1865 PgStat_SubXactStatus *xact_state;
1867 xact_state = pgStatXactStack;
1868 if (xact_state != NULL)
1870 PgStat_TableXactStatus *trans;
1872 Assert(xact_state->nest_level == 1);
1873 Assert(xact_state->prev == NULL);
1874 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1876 PgStat_TableStatus *tabstat;
1877 TwoPhasePgStatRecord record;
1879 Assert(trans->nest_level == 1);
1880 Assert(trans->upper == NULL);
1881 tabstat = trans->parent;
1882 Assert(tabstat->trans == trans);
1884 record.tuples_inserted = trans->tuples_inserted;
1885 record.tuples_updated = trans->tuples_updated;
1886 record.tuples_deleted = trans->tuples_deleted;
1887 record.t_id = tabstat->t_id;
1888 record.t_shared = tabstat->t_shared;
1890 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
1891 &record, sizeof(TwoPhasePgStatRecord));
1897 * PostPrepare_PgStat
1898 * Clean up after successful PREPARE.
1900 * All we need do here is unlink the transaction stats state from the
1901 * nontransactional state. The nontransactional action counts will be
1902 * reported to the stats collector immediately, while the effects on live
1903 * and dead tuple counts are preserved in the 2PC state file.
1905 * Note: AtEOXact_PgStat is not called during PREPARE.
1908 PostPrepare_PgStat(void)
1910 PgStat_SubXactStatus *xact_state;
1913 * We don't bother to free any of the transactional state, since it's all
1914 * in TopTransactionContext and will go away anyway.
1916 xact_state = pgStatXactStack;
1917 if (xact_state != NULL)
1919 PgStat_TableXactStatus *trans;
1921 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1923 PgStat_TableStatus *tabstat;
1925 tabstat = trans->parent;
1926 tabstat->trans = NULL;
1929 pgStatXactStack = NULL;
1931 /* Make sure any stats snapshot is thrown away */
1932 pgstat_clear_snapshot();
1936 * 2PC processing routine for COMMIT PREPARED case.
1938 * Load the saved counts into our local pgstats state.
1941 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
1942 void *recdata, uint32 len)
1944 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
1945 PgStat_TableStatus *pgstat_info;
1947 /* Find or create a tabstat entry for the rel */
1948 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
1950 /* Same math as in AtEOXact_PgStat, commit case */
1951 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
1952 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
1953 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
1954 pgstat_info->t_counts.t_delta_live_tuples +=
1955 rec->tuples_inserted - rec->tuples_deleted;
1956 pgstat_info->t_counts.t_delta_dead_tuples +=
1957 rec->tuples_updated + rec->tuples_deleted;
1958 pgstat_info->t_counts.t_changed_tuples +=
1959 rec->tuples_inserted + rec->tuples_updated +
1960 rec->tuples_deleted;
1964 * 2PC processing routine for ROLLBACK PREPARED case.
1966 * Load the saved counts into our local pgstats state, but treat them
1970 pgstat_twophase_postabort(TransactionId xid, uint16 info,
1971 void *recdata, uint32 len)
1973 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
1974 PgStat_TableStatus *pgstat_info;
1976 /* Find or create a tabstat entry for the rel */
1977 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
1979 /* Same math as in AtEOXact_PgStat, abort case */
1980 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
1981 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
1982 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
1983 pgstat_info->t_counts.t_delta_dead_tuples +=
1984 rec->tuples_inserted + rec->tuples_updated;
1989 * pgstat_fetch_stat_dbentry() -
1991 * Support function for the SQL-callable pgstat* functions. Returns
1992 * the collected statistics for one database or NULL. NULL doesn't mean
1993 * that the database doesn't exist, it is just not yet known by the
1994 * collector, so the caller is better off to report ZERO instead.
1997 PgStat_StatDBEntry *
1998 pgstat_fetch_stat_dbentry(Oid dbid)
2001 * If not done for this transaction, read the statistics collector stats
2002 * file into some hash tables.
2004 backend_read_statsfile();
2007 * Lookup the requested database; return NULL if not found
2009 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2016 * pgstat_fetch_stat_tabentry() -
2018 * Support function for the SQL-callable pgstat* functions. Returns
2019 * the collected statistics for one table or NULL. NULL doesn't mean
2020 * that the table doesn't exist, it is just not yet known by the
2021 * collector, so the caller is better off to report ZERO instead.
2024 PgStat_StatTabEntry *
2025 pgstat_fetch_stat_tabentry(Oid relid)
2028 PgStat_StatDBEntry *dbentry;
2029 PgStat_StatTabEntry *tabentry;
2032 * If not done for this transaction, read the statistics collector stats
2033 * file into some hash tables.
2035 backend_read_statsfile();
2038 * Lookup our database, then look in its table hash table.
2040 dbid = MyDatabaseId;
2041 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2044 if (dbentry != NULL && dbentry->tables != NULL)
2046 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2054 * If we didn't find it, maybe it's a shared table.
2057 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2060 if (dbentry != NULL && dbentry->tables != NULL)
2062 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2074 * pgstat_fetch_stat_funcentry() -
2076 * Support function for the SQL-callable pgstat* functions. Returns
2077 * the collected statistics for one function or NULL.
2080 PgStat_StatFuncEntry *
2081 pgstat_fetch_stat_funcentry(Oid func_id)
2083 PgStat_StatDBEntry *dbentry;
2084 PgStat_StatFuncEntry *funcentry = NULL;
2086 /* load the stats file if needed */
2087 backend_read_statsfile();
2089 /* Lookup our database, then find the requested function. */
2090 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
2091 if (dbentry != NULL && dbentry->functions != NULL)
2093 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2103 * pgstat_fetch_stat_beentry() -
2105 * Support function for the SQL-callable pgstat* functions. Returns
2106 * our local copy of the current-activity entry for one backend.
2108 * NB: caller is responsible for a check if the user is permitted to see
2109 * this info (especially the querystring).
2113 pgstat_fetch_stat_beentry(int beid)
2115 pgstat_read_current_status();
2117 if (beid < 1 || beid > localNumBackends)
2120 return &localBackendStatusTable[beid - 1];
2125 * pgstat_fetch_stat_numbackends() -
2127 * Support function for the SQL-callable pgstat* functions. Returns
2128 * the maximum current backend id.
2132 pgstat_fetch_stat_numbackends(void)
2134 pgstat_read_current_status();
2136 return localNumBackends;
2141 * pgstat_fetch_global() -
2143 * Support function for the SQL-callable pgstat* functions. Returns
2144 * a pointer to the global statistics struct.
2147 PgStat_GlobalStats *
2148 pgstat_fetch_global(void)
2150 backend_read_statsfile();
2152 return &globalStats;
2156 /* ------------------------------------------------------------
2157 * Functions for management of the shared-memory PgBackendStatus array
2158 * ------------------------------------------------------------
2161 static PgBackendStatus *BackendStatusArray = NULL;
2162 static PgBackendStatus *MyBEEntry = NULL;
2163 static char *BackendAppnameBuffer = NULL;
2164 static char *BackendActivityBuffer = NULL;
2168 * Report shared-memory space needed by CreateSharedBackendStatus.
2171 BackendStatusShmemSize(void)
2175 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2176 size = add_size(size,
2177 mul_size(NAMEDATALEN, MaxBackends));
2178 size = add_size(size,
2179 mul_size(pgstat_track_activity_query_size, MaxBackends));
2184 * Initialize the shared status array and activity/appname string buffers
2185 * during postmaster startup.
2188 CreateSharedBackendStatus(void)
2195 /* Create or attach to the shared array */
2196 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2197 BackendStatusArray = (PgBackendStatus *)
2198 ShmemInitStruct("Backend Status Array", size, &found);
2203 * We're the first - initialize.
2205 MemSet(BackendStatusArray, 0, size);
2208 /* Create or attach to the shared appname buffer */
2209 size = mul_size(NAMEDATALEN, MaxBackends);
2210 BackendAppnameBuffer = (char *)
2211 ShmemInitStruct("Backend Application Name Buffer", size, &found);
2215 MemSet(BackendAppnameBuffer, 0, size);
2217 /* Initialize st_appname pointers. */
2218 buffer = BackendAppnameBuffer;
2219 for (i = 0; i < MaxBackends; i++)
2221 BackendStatusArray[i].st_appname = buffer;
2222 buffer += NAMEDATALEN;
2226 /* Create or attach to the shared activity buffer */
2227 size = mul_size(pgstat_track_activity_query_size, MaxBackends);
2228 BackendActivityBuffer = (char *)
2229 ShmemInitStruct("Backend Activity Buffer", size, &found);
2233 MemSet(BackendActivityBuffer, 0, size);
2235 /* Initialize st_activity pointers. */
2236 buffer = BackendActivityBuffer;
2237 for (i = 0; i < MaxBackends; i++)
2239 BackendStatusArray[i].st_activity = buffer;
2240 buffer += pgstat_track_activity_query_size;
2247 * pgstat_initialize() -
2249 * Initialize pgstats state, and set up our on-proc-exit hook.
2250 * Called from InitPostgres. MyBackendId must be set,
2251 * but we must not have started any transaction yet (since the
2252 * exit hook must run after the last transaction exit).
2253 * NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful.
2257 pgstat_initialize(void)
2259 /* Initialize MyBEEntry */
2260 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2261 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2263 /* Set up a process-exit hook to clean up */
2264 on_shmem_exit(pgstat_beshutdown_hook, 0);
2268 * pgstat_bestart() -
2270 * Initialize this backend's entry in the PgBackendStatus array.
2271 * Called from InitPostgres.
2272 * MyDatabaseId, session userid, and application_name must be set
2273 * (hence, this cannot be combined with pgstat_initialize).
2277 pgstat_bestart(void)
2279 TimestampTz proc_start_timestamp;
2281 SockAddr clientaddr;
2282 volatile PgBackendStatus *beentry;
2285 * To minimize the time spent modifying the PgBackendStatus entry, fetch
2286 * all the needed data first.
2288 * If we have a MyProcPort, use its session start time (for consistency,
2289 * and to save a kernel call).
2292 proc_start_timestamp = MyProcPort->SessionStartTime;
2294 proc_start_timestamp = GetCurrentTimestamp();
2295 userid = GetSessionUserId();
2298 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2299 * If so, use all-zeroes client address, which is dealt with specially in
2300 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2303 memcpy(&clientaddr, &MyProcPort->raddr, sizeof(clientaddr));
2305 MemSet(&clientaddr, 0, sizeof(clientaddr));
2308 * Initialize my status entry, following the protocol of bumping
2309 * st_changecount before and after; and make sure it's even afterwards. We
2310 * use a volatile pointer here to ensure the compiler doesn't try to get
2313 beentry = MyBEEntry;
2316 beentry->st_changecount++;
2317 } while ((beentry->st_changecount & 1) == 0);
2319 beentry->st_procpid = MyProcPid;
2320 beentry->st_proc_start_timestamp = proc_start_timestamp;
2321 beentry->st_activity_start_timestamp = 0;
2322 beentry->st_xact_start_timestamp = 0;
2323 beentry->st_databaseid = MyDatabaseId;
2324 beentry->st_userid = userid;
2325 beentry->st_clientaddr = clientaddr;
2326 beentry->st_waiting = false;
2327 beentry->st_appname[0] = '\0';
2328 beentry->st_activity[0] = '\0';
2329 /* Also make sure the last byte in each string area is always 0 */
2330 beentry->st_appname[NAMEDATALEN - 1] = '\0';
2331 beentry->st_activity[pgstat_track_activity_query_size - 1] = '\0';
2333 beentry->st_changecount++;
2334 Assert((beentry->st_changecount & 1) == 0);
2336 /* Update app name to current GUC setting */
2337 if (application_name)
2338 pgstat_report_appname(application_name);
2342 * Shut down a single backend's statistics reporting at process exit.
2344 * Flush any remaining statistics counts out to the collector.
2345 * Without this, operations triggered during backend exit (such as
2346 * temp table deletions) won't be counted.
2348 * Lastly, clear out our entry in the PgBackendStatus array.
2351 pgstat_beshutdown_hook(int code, Datum arg)
2353 volatile PgBackendStatus *beentry = MyBEEntry;
2356 * If we got as far as discovering our own database ID, we can report
2357 * what we did to the collector. Otherwise, we'd be sending an invalid
2358 * database ID, so forget it. (This means that accesses to pg_database
2359 * during failed backend starts might never get counted.)
2361 if (OidIsValid(MyDatabaseId))
2362 pgstat_report_stat(true);
2365 * Clear my status entry, following the protocol of bumping st_changecount
2366 * before and after. We use a volatile pointer here to ensure the
2367 * compiler doesn't try to get cute.
2369 beentry->st_changecount++;
2371 beentry->st_procpid = 0; /* mark invalid */
2373 beentry->st_changecount++;
2374 Assert((beentry->st_changecount & 1) == 0);
2379 * pgstat_report_activity() -
2381 * Called from tcop/postgres.c to report what the backend is actually doing
2382 * (usually "<IDLE>" or the start of the query to be executed).
2386 pgstat_report_activity(const char *cmd_str)
2388 volatile PgBackendStatus *beentry = MyBEEntry;
2389 TimestampTz start_timestamp;
2392 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
2394 if (!pgstat_track_activities || !beentry)
2398 * To minimize the time spent modifying the entry, fetch all the needed
2401 start_timestamp = GetCurrentStatementStartTimestamp();
2403 len = strlen(cmd_str);
2404 len = pg_mbcliplen(cmd_str, len, pgstat_track_activity_query_size - 1);
2407 * Update my status entry, following the protocol of bumping
2408 * st_changecount before and after. We use a volatile pointer here to
2409 * ensure the compiler doesn't try to get cute.
2411 beentry->st_changecount++;
2413 beentry->st_activity_start_timestamp = start_timestamp;
2414 memcpy((char *) beentry->st_activity, cmd_str, len);
2415 beentry->st_activity[len] = '\0';
2417 beentry->st_changecount++;
2418 Assert((beentry->st_changecount & 1) == 0);
2422 * pgstat_report_appname() -
2424 * Called to update our application name.
2428 pgstat_report_appname(const char *appname)
2430 volatile PgBackendStatus *beentry = MyBEEntry;
2436 /* This should be unnecessary if GUC did its job, but be safe */
2437 len = pg_mbcliplen(appname, strlen(appname), NAMEDATALEN - 1);
2440 * Update my status entry, following the protocol of bumping
2441 * st_changecount before and after. We use a volatile pointer here to
2442 * ensure the compiler doesn't try to get cute.
2444 beentry->st_changecount++;
2446 memcpy((char *) beentry->st_appname, appname, len);
2447 beentry->st_appname[len] = '\0';
2449 beentry->st_changecount++;
2450 Assert((beentry->st_changecount & 1) == 0);
2454 * Report current transaction start timestamp as the specified value.
2455 * Zero means there is no active transaction.
2458 pgstat_report_xact_timestamp(TimestampTz tstamp)
2460 volatile PgBackendStatus *beentry = MyBEEntry;
2462 if (!pgstat_track_activities || !beentry)
2466 * Update my status entry, following the protocol of bumping
2467 * st_changecount before and after. We use a volatile pointer here to
2468 * ensure the compiler doesn't try to get cute.
2470 beentry->st_changecount++;
2471 beentry->st_xact_start_timestamp = tstamp;
2472 beentry->st_changecount++;
2473 Assert((beentry->st_changecount & 1) == 0);
2477 * pgstat_report_waiting() -
2479 * Called from lock manager to report beginning or end of a lock wait.
2481 * NB: this *must* be able to survive being called before MyBEEntry has been
2486 pgstat_report_waiting(bool waiting)
2488 volatile PgBackendStatus *beentry = MyBEEntry;
2490 if (!pgstat_track_activities || !beentry)
2494 * Since this is a single-byte field in a struct that only this process
2495 * may modify, there seems no need to bother with the st_changecount
2496 * protocol. The update must appear atomic in any case.
2498 beentry->st_waiting = waiting;
2503 * pgstat_read_current_status() -
2505 * Copy the current contents of the PgBackendStatus array to local memory,
2506 * if not already done in this transaction.
2510 pgstat_read_current_status(void)
2512 volatile PgBackendStatus *beentry;
2513 PgBackendStatus *localtable;
2514 PgBackendStatus *localentry;
2519 Assert(!pgStatRunningInCollector);
2520 if (localBackendStatusTable)
2521 return; /* already done */
2523 pgstat_setup_memcxt();
2525 localtable = (PgBackendStatus *)
2526 MemoryContextAlloc(pgStatLocalContext,
2527 sizeof(PgBackendStatus) * MaxBackends);
2528 localappname = (char *)
2529 MemoryContextAlloc(pgStatLocalContext,
2530 NAMEDATALEN * MaxBackends);
2531 localactivity = (char *)
2532 MemoryContextAlloc(pgStatLocalContext,
2533 pgstat_track_activity_query_size * MaxBackends);
2534 localNumBackends = 0;
2536 beentry = BackendStatusArray;
2537 localentry = localtable;
2538 for (i = 1; i <= MaxBackends; i++)
2541 * Follow the protocol of retrying if st_changecount changes while we
2542 * copy the entry, or if it's odd. (The check for odd is needed to
2543 * cover the case where we are able to completely copy the entry while
2544 * the source backend is between increment steps.) We use a volatile
2545 * pointer here to ensure the compiler doesn't try to get cute.
2549 int save_changecount = beentry->st_changecount;
2551 localentry->st_procpid = beentry->st_procpid;
2552 if (localentry->st_procpid > 0)
2554 memcpy(localentry, (char *) beentry, sizeof(PgBackendStatus));
2557 * strcpy is safe even if the string is modified concurrently,
2558 * because there's always a \0 at the end of the buffer.
2560 strcpy(localappname, (char *) beentry->st_appname);
2561 localentry->st_appname = localappname;
2562 strcpy(localactivity, (char *) beentry->st_activity);
2563 localentry->st_activity = localactivity;
2566 if (save_changecount == beentry->st_changecount &&
2567 (save_changecount & 1) == 0)
2570 /* Make sure we can break out of loop if stuck... */
2571 CHECK_FOR_INTERRUPTS();
2575 /* Only valid entries get included into the local array */
2576 if (localentry->st_procpid > 0)
2579 localappname += NAMEDATALEN;
2580 localactivity += pgstat_track_activity_query_size;
2585 /* Set the pointer only after completion of a valid table */
2586 localBackendStatusTable = localtable;
2591 * pgstat_get_backend_current_activity() -
2593 * Return a string representing the current activity of the backend with
2594 * the specified PID. This looks directly at the BackendStatusArray,
2595 * and so will provide current information regardless of the age of our
2596 * transaction's snapshot of the status array.
2598 * It is the caller's responsibility to invoke this only for backends whose
2599 * state is expected to remain stable while the result is in use. The
2600 * only current use is in deadlock reporting, where we can expect that
2601 * the target backend is blocked on a lock. (There are corner cases
2602 * where the target's wait could get aborted while we are looking at it,
2603 * but the very worst consequence is to return a pointer to a string
2604 * that's been changed, so we won't worry too much.)
2606 * Note: return strings for special cases match pg_stat_get_backend_activity.
2610 pgstat_get_backend_current_activity(int pid, bool checkUser)
2612 PgBackendStatus *beentry;
2615 beentry = BackendStatusArray;
2616 for (i = 1; i <= MaxBackends; i++)
2619 * Although we expect the target backend's entry to be stable, that
2620 * doesn't imply that anyone else's is. To avoid identifying the
2621 * wrong backend, while we check for a match to the desired PID we
2622 * must follow the protocol of retrying if st_changecount changes
2623 * while we examine the entry, or if it's odd. (This might be
2624 * unnecessary, since fetching or storing an int is almost certainly
2625 * atomic, but let's play it safe.) We use a volatile pointer here to
2626 * ensure the compiler doesn't try to get cute.
2628 volatile PgBackendStatus *vbeentry = beentry;
2633 int save_changecount = vbeentry->st_changecount;
2635 found = (vbeentry->st_procpid == pid);
2637 if (save_changecount == vbeentry->st_changecount &&
2638 (save_changecount & 1) == 0)
2641 /* Make sure we can break out of loop if stuck... */
2642 CHECK_FOR_INTERRUPTS();
2647 /* Now it is safe to use the non-volatile pointer */
2648 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
2649 return "<insufficient privilege>";
2650 else if (*(beentry->st_activity) == '\0')
2651 return "<command string not enabled>";
2653 return beentry->st_activity;
2659 /* If we get here, caller is in error ... */
2660 return "<backend information not available>";
2664 /* ------------------------------------------------------------
2665 * Local support functions follow
2666 * ------------------------------------------------------------
2671 * pgstat_setheader() -
2673 * Set common header fields in a statistics message
2677 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
2679 hdr->m_type = mtype;
2686 * Send out one statistics message to the collector
2690 pgstat_send(void *msg, int len)
2697 ((PgStat_MsgHdr *) msg)->m_size = len;
2699 /* We'll retry after EINTR, but ignore all other failures */
2702 rc = send(pgStatSock, msg, len, 0);
2703 } while (rc < 0 && errno == EINTR);
2705 #ifdef USE_ASSERT_CHECKING
2706 /* In debug builds, log send failures ... */
2708 elog(LOG, "could not send to statistics collector: %m");
2713 * pgstat_send_bgwriter() -
2715 * Send bgwriter statistics to the collector
2719 pgstat_send_bgwriter(void)
2721 /* We assume this initializes to zeroes */
2722 static const PgStat_MsgBgWriter all_zeroes;
2725 * This function can be called even if nothing at all has happened. In
2726 * this case, avoid sending a completely empty message to the stats
2729 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
2733 * Prepare and send the message
2735 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
2736 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
2739 * Clear out the statistics buffer, so it can be re-used.
2741 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
2746 * PgstatCollectorMain() -
2748 * Start up the statistics collector process. This is the body of the
2749 * postmaster child process.
2751 * The argc/argv parameters are valid only in EXEC_BACKEND case.
2754 NON_EXEC_STATIC void
2755 PgstatCollectorMain(int argc, char *argv[])
2762 struct pollfd input_fd;
2764 struct timeval sel_timeout;
2769 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2771 MyProcPid = getpid(); /* reset MyProcPid */
2773 MyStartTime = time(NULL); /* record Start Time for logging */
2776 * If possible, make this process a group leader, so that the postmaster
2777 * can signal any child processes too. (pgstat probably never has any
2778 * child processes, but for consistency we make all postmaster child
2779 * processes do this.)
2783 elog(FATAL, "setsid() failed: %m");
2787 * Ignore all signals usually bound to some action in the postmaster,
2790 pqsignal(SIGHUP, pgstat_sighup_handler);
2791 pqsignal(SIGINT, SIG_IGN);
2792 pqsignal(SIGTERM, SIG_IGN);
2793 pqsignal(SIGQUIT, pgstat_exit);
2794 pqsignal(SIGALRM, SIG_IGN);
2795 pqsignal(SIGPIPE, SIG_IGN);
2796 pqsignal(SIGUSR1, SIG_IGN);
2797 pqsignal(SIGUSR2, SIG_IGN);
2798 pqsignal(SIGCHLD, SIG_DFL);
2799 pqsignal(SIGTTIN, SIG_DFL);
2800 pqsignal(SIGTTOU, SIG_DFL);
2801 pqsignal(SIGCONT, SIG_DFL);
2802 pqsignal(SIGWINCH, SIG_DFL);
2803 PG_SETMASK(&UnBlockSig);
2806 * Identify myself via ps
2808 init_ps_display("stats collector process", "", "", "");
2811 * Arrange to write the initial status file right away
2813 last_statrequest = GetCurrentTimestamp();
2814 last_statwrite = last_statrequest - 1;
2817 * Read in an existing statistics stats file or initialize the stats to
2820 pgStatRunningInCollector = true;
2821 pgStatDBHash = pgstat_read_statsfile(InvalidOid, true);
2824 * Setup the descriptor set for select(2). Since only one bit in the set
2825 * ever changes, we need not repeat FD_ZERO each time.
2827 #if !defined(HAVE_POLL) && !defined(WIN32)
2832 * Loop to process messages until we get SIGQUIT or detect ungraceful
2833 * death of our parent postmaster.
2835 * For performance reasons, we don't want to do a PostmasterIsAlive() test
2836 * after every message; instead, do it only when select()/poll() is
2837 * interrupted by timeout. In essence, we'll stay alive as long as
2838 * backends keep sending us stuff often, even if the postmaster is gone.
2845 * Quit if we get SIGQUIT from the postmaster.
2851 * Reload configuration if we got SIGHUP from the postmaster.
2855 ProcessConfigFile(PGC_SIGHUP);
2860 * Write the stats file if a new request has arrived that is not
2861 * satisfied by existing file.
2863 if (last_statwrite < last_statrequest)
2864 pgstat_write_statsfile(false);
2867 * Wait for a message to arrive; but not for more than
2868 * PGSTAT_SELECT_TIMEOUT seconds. (This determines how quickly we will
2869 * shut down after an ungraceful postmaster termination; so it needn't
2870 * be very fast. However, on some systems SIGQUIT won't interrupt the
2871 * poll/select call, so this also limits speed of response to SIGQUIT,
2872 * which is more important.)
2874 * We use poll(2) if available, otherwise select(2). Win32 has its own
2879 input_fd.fd = pgStatSock;
2880 input_fd.events = POLLIN | POLLERR;
2881 input_fd.revents = 0;
2883 if (poll(&input_fd, 1, PGSTAT_SELECT_TIMEOUT * 1000) < 0)
2888 (errcode_for_socket_access(),
2889 errmsg("poll() failed in statistics collector: %m")));
2892 got_data = (input_fd.revents != 0);
2893 #else /* !HAVE_POLL */
2895 FD_SET (pgStatSock, &rfds);
2898 * timeout struct is modified by select() on some operating systems,
2899 * so re-fill it each time.
2901 sel_timeout.tv_sec = PGSTAT_SELECT_TIMEOUT;
2902 sel_timeout.tv_usec = 0;
2904 if (select(pgStatSock + 1, &rfds, NULL, NULL, &sel_timeout) < 0)
2909 (errcode_for_socket_access(),
2910 errmsg("select() failed in statistics collector: %m")));
2913 got_data = FD_ISSET(pgStatSock, &rfds);
2914 #endif /* HAVE_POLL */
2916 got_data = pgwin32_waitforsinglesocket(pgStatSock, FD_READ,
2917 PGSTAT_SELECT_TIMEOUT * 1000);
2921 * If there is a message on the socket, read it and check for
2926 len = recv(pgStatSock, (char *) &msg,
2927 sizeof(PgStat_Msg), 0);
2933 (errcode_for_socket_access(),
2934 errmsg("could not read statistics message: %m")));
2938 * We ignore messages that are smaller than our common header
2940 if (len < sizeof(PgStat_MsgHdr))
2944 * The received length must match the length in the header
2946 if (msg.msg_hdr.m_size != len)
2950 * O.K. - we accept this message. Process it.
2952 switch (msg.msg_hdr.m_type)
2954 case PGSTAT_MTYPE_DUMMY:
2957 case PGSTAT_MTYPE_INQUIRY:
2958 pgstat_recv_inquiry((PgStat_MsgInquiry *) &msg, len);
2961 case PGSTAT_MTYPE_TABSTAT:
2962 pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len);
2965 case PGSTAT_MTYPE_TABPURGE:
2966 pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len);
2969 case PGSTAT_MTYPE_DROPDB:
2970 pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len);
2973 case PGSTAT_MTYPE_RESETCOUNTER:
2974 pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg,
2978 case PGSTAT_MTYPE_RESETSHAREDCOUNTER:
2979 pgstat_recv_resetsharedcounter(
2980 (PgStat_MsgResetsharedcounter *) &msg,
2984 case PGSTAT_MTYPE_RESETSINGLECOUNTER:
2985 pgstat_recv_resetsinglecounter(
2986 (PgStat_MsgResetsinglecounter *) &msg,
2990 case PGSTAT_MTYPE_AUTOVAC_START:
2991 pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
2994 case PGSTAT_MTYPE_VACUUM:
2995 pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
2998 case PGSTAT_MTYPE_ANALYZE:
2999 pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len);
3002 case PGSTAT_MTYPE_BGWRITER:
3003 pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len);
3006 case PGSTAT_MTYPE_FUNCSTAT:
3007 pgstat_recv_funcstat((PgStat_MsgFuncstat *) &msg, len);
3010 case PGSTAT_MTYPE_FUNCPURGE:
3011 pgstat_recv_funcpurge((PgStat_MsgFuncpurge *) &msg, len);
3021 * We can only get here if the select/poll timeout elapsed. Check
3022 * for postmaster death.
3024 if (!PostmasterIsAlive(true))
3027 } /* end of message-processing loop */
3030 * Save the final stats to reuse at next startup.
3032 pgstat_write_statsfile(true);
3038 /* SIGQUIT signal handler for collector process */
3040 pgstat_exit(SIGNAL_ARGS)
3045 /* SIGHUP handler for collector process */
3047 pgstat_sighup_handler(SIGNAL_ARGS)
3054 * Lookup the hash table entry for the specified database. If no hash
3055 * table entry exists, initialize it, if the create parameter is true.
3056 * Else, return NULL.
3058 static PgStat_StatDBEntry *
3059 pgstat_get_db_entry(Oid databaseid, bool create)
3061 PgStat_StatDBEntry *result;
3063 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3065 /* Lookup or create the hash table entry for this database */
3066 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
3070 if (!create && !found)
3073 /* If not found, initialize the new one. */
3078 result->tables = NULL;
3079 result->functions = NULL;
3080 result->n_xact_commit = 0;
3081 result->n_xact_rollback = 0;
3082 result->n_blocks_fetched = 0;
3083 result->n_blocks_hit = 0;
3084 result->n_tuples_returned = 0;
3085 result->n_tuples_fetched = 0;
3086 result->n_tuples_inserted = 0;
3087 result->n_tuples_updated = 0;
3088 result->n_tuples_deleted = 0;
3089 result->last_autovac_time = 0;
3091 memset(&hash_ctl, 0, sizeof(hash_ctl));
3092 hash_ctl.keysize = sizeof(Oid);
3093 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3094 hash_ctl.hash = oid_hash;
3095 result->tables = hash_create("Per-database table",
3096 PGSTAT_TAB_HASH_SIZE,
3098 HASH_ELEM | HASH_FUNCTION);
3100 hash_ctl.keysize = sizeof(Oid);
3101 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3102 hash_ctl.hash = oid_hash;
3103 result->functions = hash_create("Per-database function",
3104 PGSTAT_FUNCTION_HASH_SIZE,
3106 HASH_ELEM | HASH_FUNCTION);
3114 * Lookup the hash table entry for the specified table. If no hash
3115 * table entry exists, initialize it, if the create parameter is true.
3116 * Else, return NULL.
3118 static PgStat_StatTabEntry *
3119 pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
3121 PgStat_StatTabEntry *result;
3123 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3125 /* Lookup or create the hash table entry for this table */
3126 result = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3130 if (!create && !found)
3133 /* If not found, initialize the new one. */
3136 result->numscans = 0;
3137 result->tuples_returned = 0;
3138 result->tuples_fetched = 0;
3139 result->tuples_inserted = 0;
3140 result->tuples_updated = 0;
3141 result->tuples_deleted = 0;
3142 result->tuples_hot_updated = 0;
3143 result->n_live_tuples = 0;
3144 result->n_dead_tuples = 0;
3145 result->changes_since_analyze = 0;
3146 result->blocks_fetched = 0;
3147 result->blocks_hit = 0;
3149 result->vacuum_timestamp = 0;
3150 result->autovac_vacuum_timestamp = 0;
3151 result->analyze_timestamp = 0;
3152 result->autovac_analyze_timestamp = 0;
3160 * pgstat_write_statsfile() -
3163 * If writing to the permanent file (happens when the collector is
3164 * shutting down only), remove the temporary file so that backends
3165 * starting up under a new postmaster can't read the old data before
3166 * the new collector is ready.
3170 pgstat_write_statsfile(bool permanent)
3172 HASH_SEQ_STATUS hstat;
3173 HASH_SEQ_STATUS tstat;
3174 HASH_SEQ_STATUS fstat;
3175 PgStat_StatDBEntry *dbentry;
3176 PgStat_StatTabEntry *tabentry;
3177 PgStat_StatFuncEntry *funcentry;
3180 const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
3181 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3184 * Open the statistics temp file to write out the current values.
3186 fpout = AllocateFile(tmpfile, PG_BINARY_W);
3190 (errcode_for_file_access(),
3191 errmsg("could not open temporary statistics file \"%s\": %m",
3197 * Set the timestamp of the stats file.
3199 globalStats.stats_timestamp = GetCurrentTimestamp();
3202 * Write the file header --- currently just a format ID.
3204 format_id = PGSTAT_FILE_FORMAT_ID;
3205 fwrite(&format_id, sizeof(format_id), 1, fpout);
3208 * Write global stats struct
3210 fwrite(&globalStats, sizeof(globalStats), 1, fpout);
3213 * Walk through the database table.
3215 hash_seq_init(&hstat, pgStatDBHash);
3216 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
3219 * Write out the DB entry including the number of live backends. We
3220 * don't write the tables or functions pointers, since they're of no
3221 * use to any other process.
3224 fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
3227 * Walk through the database's access stats per table.
3229 hash_seq_init(&tstat, dbentry->tables);
3230 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
3233 fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
3237 * Walk through the database's function stats table.
3239 hash_seq_init(&fstat, dbentry->functions);
3240 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
3243 fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
3247 * Mark the end of this DB
3253 * No more output to be done. Close the temp file and replace the old
3254 * pgstat.stat with it. The ferror() check replaces testing for error
3255 * after each individual fputc or fwrite above.
3262 (errcode_for_file_access(),
3263 errmsg("could not write temporary statistics file \"%s\": %m",
3268 else if (FreeFile(fpout) < 0)
3271 (errcode_for_file_access(),
3272 errmsg("could not close temporary statistics file \"%s\": %m",
3276 else if (rename(tmpfile, statfile) < 0)
3279 (errcode_for_file_access(),
3280 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
3281 tmpfile, statfile)));
3287 * Successful write, so update last_statwrite.
3289 last_statwrite = globalStats.stats_timestamp;
3292 * It's not entirely clear whether there could be clock skew between
3293 * backends and the collector; but just in case someone manages to
3294 * send us a stats request time that's far in the future, reset it.
3295 * This ensures that no inquiry message can cause more than one stats
3296 * file write to occur.
3298 last_statrequest = last_statwrite;
3302 unlink(pgstat_stat_filename);
3307 * pgstat_read_statsfile() -
3309 * Reads in an existing statistics collector file and initializes the
3310 * databases' hash table (whose entries point to the tables' hash tables).
3314 pgstat_read_statsfile(Oid onlydb, bool permanent)
3316 PgStat_StatDBEntry *dbentry;
3317 PgStat_StatDBEntry dbbuf;
3318 PgStat_StatTabEntry *tabentry;
3319 PgStat_StatTabEntry tabbuf;
3320 PgStat_StatFuncEntry funcbuf;
3321 PgStat_StatFuncEntry *funcentry;
3324 HTAB *tabhash = NULL;
3325 HTAB *funchash = NULL;
3329 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3332 * The tables will live in pgStatLocalContext.
3334 pgstat_setup_memcxt();
3337 * Create the DB hashtable
3339 memset(&hash_ctl, 0, sizeof(hash_ctl));
3340 hash_ctl.keysize = sizeof(Oid);
3341 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
3342 hash_ctl.hash = oid_hash;
3343 hash_ctl.hcxt = pgStatLocalContext;
3344 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
3345 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3348 * Clear out global statistics so they start from zero in case we can't
3349 * load an existing statsfile.
3351 memset(&globalStats, 0, sizeof(globalStats));
3354 * Try to open the status file. If it doesn't exist, the backends simply
3355 * return zero for anything and the collector simply starts from scratch
3356 * with empty counters.
3358 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3362 * Verify it's of the expected format.
3364 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3365 || format_id != PGSTAT_FILE_FORMAT_ID)
3367 ereport(pgStatRunningInCollector ? LOG : WARNING,
3368 (errmsg("corrupted pgstat.stat file")));
3373 * Read global stats struct
3375 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
3377 ereport(pgStatRunningInCollector ? LOG : WARNING,
3378 (errmsg("corrupted pgstat.stat file")));
3383 * We found an existing collector stats file. Read it and put all the
3384 * hashtable entries into place.
3388 switch (fgetc(fpin))
3391 * 'D' A PgStat_StatDBEntry struct describing a database
3392 * follows. Subsequently, zero to many 'T' and 'F' entries
3393 * will follow until a 'd' is encountered.
3396 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
3397 fpin) != offsetof(PgStat_StatDBEntry, tables))
3399 ereport(pgStatRunningInCollector ? LOG : WARNING,
3400 (errmsg("corrupted pgstat.stat file")));
3405 * Add to the DB hash
3407 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
3408 (void *) &dbbuf.databaseid,
3413 ereport(pgStatRunningInCollector ? LOG : WARNING,
3414 (errmsg("corrupted pgstat.stat file")));
3418 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
3419 dbentry->tables = NULL;
3420 dbentry->functions = NULL;
3423 * Don't collect tables if not the requested DB (or the
3424 * shared-table info)
3426 if (onlydb != InvalidOid)
3428 if (dbbuf.databaseid != onlydb &&
3429 dbbuf.databaseid != InvalidOid)
3433 memset(&hash_ctl, 0, sizeof(hash_ctl));
3434 hash_ctl.keysize = sizeof(Oid);
3435 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3436 hash_ctl.hash = oid_hash;
3437 hash_ctl.hcxt = pgStatLocalContext;
3438 dbentry->tables = hash_create("Per-database table",
3439 PGSTAT_TAB_HASH_SIZE,
3441 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3443 hash_ctl.keysize = sizeof(Oid);
3444 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3445 hash_ctl.hash = oid_hash;
3446 hash_ctl.hcxt = pgStatLocalContext;
3447 dbentry->functions = hash_create("Per-database function",
3448 PGSTAT_FUNCTION_HASH_SIZE,
3450 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3453 * Arrange that following records add entries to this
3454 * database's hash tables.
3456 tabhash = dbentry->tables;
3457 funchash = dbentry->functions;
3461 * 'd' End of this database.
3469 * 'T' A PgStat_StatTabEntry follows.
3472 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
3473 fpin) != sizeof(PgStat_StatTabEntry))
3475 ereport(pgStatRunningInCollector ? LOG : WARNING,
3476 (errmsg("corrupted pgstat.stat file")));
3481 * Skip if table belongs to a not requested database.
3483 if (tabhash == NULL)
3486 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
3487 (void *) &tabbuf.tableid,
3488 HASH_ENTER, &found);
3492 ereport(pgStatRunningInCollector ? LOG : WARNING,
3493 (errmsg("corrupted pgstat.stat file")));
3497 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
3501 * 'F' A PgStat_StatFuncEntry follows.
3504 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
3505 fpin) != sizeof(PgStat_StatFuncEntry))
3507 ereport(pgStatRunningInCollector ? LOG : WARNING,
3508 (errmsg("corrupted pgstat.stat file")));
3513 * Skip if function belongs to a not requested database.
3515 if (funchash == NULL)
3518 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
3519 (void *) &funcbuf.functionid,
3520 HASH_ENTER, &found);
3524 ereport(pgStatRunningInCollector ? LOG : WARNING,
3525 (errmsg("corrupted pgstat.stat file")));
3529 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
3533 * 'E' The EOF marker of a complete stats file.
3539 ereport(pgStatRunningInCollector ? LOG : WARNING,
3540 (errmsg("corrupted pgstat.stat file")));
3549 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
3555 * pgstat_read_statsfile_timestamp() -
3557 * Attempt to fetch the timestamp of an existing stats file.
3558 * Returns TRUE if successful (timestamp is stored at *ts).
3562 pgstat_read_statsfile_timestamp(bool permanent, TimestampTz *ts)
3564 PgStat_GlobalStats myGlobalStats;
3567 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3570 * Try to open the status file.
3572 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3576 * Verify it's of the expected format.
3578 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3579 || format_id != PGSTAT_FILE_FORMAT_ID)
3586 * Read global stats struct
3588 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats), fpin) != sizeof(myGlobalStats))
3594 *ts = myGlobalStats.stats_timestamp;
3601 * If not already done, read the statistics collector stats file into
3602 * some hash tables. The results will be kept until pgstat_clear_snapshot()
3603 * is called (typically, at end of transaction).
3606 backend_read_statsfile(void)
3611 /* already read it? */
3614 Assert(!pgStatRunningInCollector);
3617 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL msec
3618 * before now. This indirectly ensures that the collector needn't write
3619 * the file more often than PGSTAT_STAT_INTERVAL. In an autovacuum
3620 * worker, however, we want a lower delay to avoid using stale data, so we
3621 * use PGSTAT_RETRY_DELAY (since the number of worker is low, this
3622 * shouldn't be a problem).
3624 * Note that we don't recompute min_ts after sleeping; so we might end up
3625 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In practice
3626 * that shouldn't happen, though, as long as the sleep time is less than
3627 * PGSTAT_STAT_INTERVAL; and we don't want to lie to the collector about
3628 * what our cutoff time really is.
3630 if (IsAutoVacuumWorkerProcess())
3631 min_ts = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
3632 -PGSTAT_RETRY_DELAY);
3634 min_ts = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
3635 -PGSTAT_STAT_INTERVAL);
3638 * Loop until fresh enough stats file is available or we ran out of time.
3639 * The stats inquiry message is sent repeatedly in case collector drops
3642 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
3644 TimestampTz file_ts = 0;
3646 CHECK_FOR_INTERRUPTS();
3648 if (pgstat_read_statsfile_timestamp(false, &file_ts) &&
3652 /* Not there or too old, so kick the collector and wait a bit */
3653 pgstat_send_inquiry(min_ts);
3654 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
3657 if (count >= PGSTAT_POLL_LOOP_COUNT)
3658 elog(WARNING, "pgstat wait timeout");
3660 /* Autovacuum launcher wants stats about all databases */
3661 if (IsAutoVacuumLauncherProcess())
3662 pgStatDBHash = pgstat_read_statsfile(InvalidOid, false);
3664 pgStatDBHash = pgstat_read_statsfile(MyDatabaseId, false);
3669 * pgstat_setup_memcxt() -
3671 * Create pgStatLocalContext, if not already done.
3675 pgstat_setup_memcxt(void)
3677 if (!pgStatLocalContext)
3678 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
3679 "Statistics snapshot",
3680 ALLOCSET_SMALL_MINSIZE,
3681 ALLOCSET_SMALL_INITSIZE,
3682 ALLOCSET_SMALL_MAXSIZE);
3687 * pgstat_clear_snapshot() -
3689 * Discard any data collected in the current transaction. Any subsequent
3690 * request will cause new snapshots to be read.
3692 * This is also invoked during transaction commit or abort to discard
3693 * the no-longer-wanted snapshot.
3697 pgstat_clear_snapshot(void)
3699 /* Release memory, if any was allocated */
3700 if (pgStatLocalContext)
3701 MemoryContextDelete(pgStatLocalContext);
3703 /* Reset variables */
3704 pgStatLocalContext = NULL;
3705 pgStatDBHash = NULL;
3706 localBackendStatusTable = NULL;
3707 localNumBackends = 0;
3712 * pgstat_recv_inquiry() -
3714 * Process stat inquiry requests.
3718 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
3720 if (msg->inquiry_time > last_statrequest)
3721 last_statrequest = msg->inquiry_time;
3726 * pgstat_recv_tabstat() -
3728 * Count what the backend has done.
3732 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
3734 PgStat_StatDBEntry *dbentry;
3735 PgStat_StatTabEntry *tabentry;
3739 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3742 * Update database-wide stats.
3744 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
3745 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
3748 * Process all table entries in the message.
3750 for (i = 0; i < msg->m_nentries; i++)
3752 PgStat_TableEntry *tabmsg = &(msg->m_entry[i]);
3754 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3755 (void *) &(tabmsg->t_id),
3756 HASH_ENTER, &found);
3761 * If it's a new table entry, initialize counters to the values we
3764 tabentry->numscans = tabmsg->t_counts.t_numscans;
3765 tabentry->tuples_returned = tabmsg->t_counts.t_tuples_returned;
3766 tabentry->tuples_fetched = tabmsg->t_counts.t_tuples_fetched;
3767 tabentry->tuples_inserted = tabmsg->t_counts.t_tuples_inserted;
3768 tabentry->tuples_updated = tabmsg->t_counts.t_tuples_updated;
3769 tabentry->tuples_deleted = tabmsg->t_counts.t_tuples_deleted;
3770 tabentry->tuples_hot_updated = tabmsg->t_counts.t_tuples_hot_updated;
3771 tabentry->n_live_tuples = tabmsg->t_counts.t_delta_live_tuples;
3772 tabentry->n_dead_tuples = tabmsg->t_counts.t_delta_dead_tuples;
3773 tabentry->changes_since_analyze = tabmsg->t_counts.t_changed_tuples;
3774 tabentry->blocks_fetched = tabmsg->t_counts.t_blocks_fetched;
3775 tabentry->blocks_hit = tabmsg->t_counts.t_blocks_hit;
3777 tabentry->vacuum_timestamp = 0;
3778 tabentry->autovac_vacuum_timestamp = 0;
3779 tabentry->analyze_timestamp = 0;
3780 tabentry->autovac_analyze_timestamp = 0;
3785 * Otherwise add the values to the existing entry.
3787 tabentry->numscans += tabmsg->t_counts.t_numscans;
3788 tabentry->tuples_returned += tabmsg->t_counts.t_tuples_returned;
3789 tabentry->tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
3790 tabentry->tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
3791 tabentry->tuples_updated += tabmsg->t_counts.t_tuples_updated;
3792 tabentry->tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
3793 tabentry->tuples_hot_updated += tabmsg->t_counts.t_tuples_hot_updated;
3794 tabentry->n_live_tuples += tabmsg->t_counts.t_delta_live_tuples;
3795 tabentry->n_dead_tuples += tabmsg->t_counts.t_delta_dead_tuples;
3796 tabentry->changes_since_analyze += tabmsg->t_counts.t_changed_tuples;
3797 tabentry->blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
3798 tabentry->blocks_hit += tabmsg->t_counts.t_blocks_hit;
3801 /* Clamp n_live_tuples in case of negative delta_live_tuples */
3802 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
3803 /* Likewise for n_dead_tuples */
3804 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
3807 * Add per-table stats to the per-database entry, too.
3809 dbentry->n_tuples_returned += tabmsg->t_counts.t_tuples_returned;
3810 dbentry->n_tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
3811 dbentry->n_tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
3812 dbentry->n_tuples_updated += tabmsg->t_counts.t_tuples_updated;
3813 dbentry->n_tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
3814 dbentry->n_blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
3815 dbentry->n_blocks_hit += tabmsg->t_counts.t_blocks_hit;
3821 * pgstat_recv_tabpurge() -
3823 * Arrange for dead table removal.
3827 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
3829 PgStat_StatDBEntry *dbentry;
3832 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3835 * No need to purge if we don't even know the database.
3837 if (!dbentry || !dbentry->tables)
3841 * Process all table entries in the message.
3843 for (i = 0; i < msg->m_nentries; i++)
3845 /* Remove from hashtable if present; we don't care if it's not. */
3846 (void) hash_search(dbentry->tables,
3847 (void *) &(msg->m_tableid[i]),
3854 * pgstat_recv_dropdb() -
3856 * Arrange for dead database removal
3860 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
3862 PgStat_StatDBEntry *dbentry;
3865 * Lookup the database in the hashtable.
3867 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3870 * If found, remove it.
3874 if (dbentry->tables != NULL)
3875 hash_destroy(dbentry->tables);
3876 if (dbentry->functions != NULL)
3877 hash_destroy(dbentry->functions);
3879 if (hash_search(pgStatDBHash,
3880 (void *) &(dbentry->databaseid),
3881 HASH_REMOVE, NULL) == NULL)
3883 (errmsg("database hash table corrupted "
3884 "during cleanup --- abort")));
3890 * pgstat_recv_resetcounter() -
3892 * Reset the statistics for the specified database.
3896 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
3899 PgStat_StatDBEntry *dbentry;
3902 * Lookup the database in the hashtable. Nothing to do if not there.
3904 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3910 * We simply throw away all the database's table entries by recreating a
3911 * new hash table for them.
3913 if (dbentry->tables != NULL)
3914 hash_destroy(dbentry->tables);
3915 if (dbentry->functions != NULL)
3916 hash_destroy(dbentry->functions);
3918 dbentry->tables = NULL;
3919 dbentry->functions = NULL;
3920 dbentry->n_xact_commit = 0;
3921 dbentry->n_xact_rollback = 0;
3922 dbentry->n_blocks_fetched = 0;
3923 dbentry->n_blocks_hit = 0;
3925 memset(&hash_ctl, 0, sizeof(hash_ctl));
3926 hash_ctl.keysize = sizeof(Oid);
3927 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3928 hash_ctl.hash = oid_hash;
3929 dbentry->tables = hash_create("Per-database table",
3930 PGSTAT_TAB_HASH_SIZE,
3932 HASH_ELEM | HASH_FUNCTION);
3934 hash_ctl.keysize = sizeof(Oid);
3935 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3936 hash_ctl.hash = oid_hash;
3937 dbentry->functions = hash_create("Per-database function",
3938 PGSTAT_FUNCTION_HASH_SIZE,
3940 HASH_ELEM | HASH_FUNCTION);
3944 * pgstat_recv_resetshared() -
3946 * Reset some shared statistics of the cluster.
3950 pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
3952 if (msg->m_resettarget==RESET_BGWRITER)
3954 /* Reset the global background writer statistics for the cluster. */
3955 memset(&globalStats, 0, sizeof(globalStats));
3959 * Presumably the sender of this message validated the target, don't
3960 * complain here if it's not valid
3965 * pgstat_recv_resetsinglecounter() -
3967 * Reset a statistics for a single object
3971 pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len)
3973 PgStat_StatDBEntry *dbentry;
3975 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3981 /* Remove object if it exists, ignore it if not */
3982 if (msg->m_resettype == RESET_TABLE)
3983 (void) hash_search(dbentry->tables, (void *) &(msg->m_objectid), HASH_REMOVE, NULL);
3984 else if (msg->m_resettype == RESET_FUNCTION)
3985 (void) hash_search(dbentry->functions, (void *)&(msg->m_objectid), HASH_REMOVE, NULL);
3989 * pgstat_recv_autovac() -
3991 * Process an autovacuum signalling message.
3995 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
3997 PgStat_StatDBEntry *dbentry;
4000 * Store the last autovacuum time in the database's hashtable entry.
4002 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4004 dbentry->last_autovac_time = msg->m_start_time;
4008 * pgstat_recv_vacuum() -
4010 * Process a VACUUM message.
4014 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
4016 PgStat_StatDBEntry *dbentry;
4017 PgStat_StatTabEntry *tabentry;
4020 * Store the data in the table's hashtable entry.
4022 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4024 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
4026 if (msg->m_adopt_counts)
4027 tabentry->n_live_tuples = msg->m_tuples;
4028 /* Resetting dead_tuples to 0 is an approximation ... */
4029 tabentry->n_dead_tuples = 0;
4031 if (msg->m_autovacuum)
4032 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
4034 tabentry->vacuum_timestamp = msg->m_vacuumtime;
4038 * pgstat_recv_analyze() -
4040 * Process an ANALYZE message.
4044 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
4046 PgStat_StatDBEntry *dbentry;
4047 PgStat_StatTabEntry *tabentry;
4050 * Store the data in the table's hashtable entry.
4052 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4054 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
4056 if (msg->m_adopt_counts)
4058 tabentry->n_live_tuples = msg->m_live_tuples;
4059 tabentry->n_dead_tuples = msg->m_dead_tuples;
4063 * We reset changes_since_analyze to zero, forgetting any changes that
4064 * occurred while the ANALYZE was in progress.
4066 tabentry->changes_since_analyze = 0;
4068 if (msg->m_autovacuum)
4069 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
4071 tabentry->analyze_timestamp = msg->m_analyzetime;
4076 * pgstat_recv_bgwriter() -
4078 * Process a BGWRITER message.
4082 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
4084 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
4085 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
4086 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
4087 globalStats.buf_written_clean += msg->m_buf_written_clean;
4088 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
4089 globalStats.buf_written_backend += msg->m_buf_written_backend;
4090 globalStats.buf_alloc += msg->m_buf_alloc;
4094 * pgstat_recv_funcstat() -
4096 * Count what the backend has done.
4100 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
4102 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
4103 PgStat_StatDBEntry *dbentry;
4104 PgStat_StatFuncEntry *funcentry;
4108 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4111 * Process all function entries in the message.
4113 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
4115 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
4116 (void *) &(funcmsg->f_id),
4117 HASH_ENTER, &found);
4122 * If it's a new function entry, initialize counters to the values
4125 funcentry->f_numcalls = funcmsg->f_numcalls;
4126 funcentry->f_time = funcmsg->f_time;
4127 funcentry->f_time_self = funcmsg->f_time_self;
4132 * Otherwise add the values to the existing entry.
4134 funcentry->f_numcalls += funcmsg->f_numcalls;
4135 funcentry->f_time += funcmsg->f_time;
4136 funcentry->f_time_self += funcmsg->f_time_self;
4142 * pgstat_recv_funcpurge() -
4144 * Arrange for dead function removal.
4148 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
4150 PgStat_StatDBEntry *dbentry;
4153 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4156 * No need to purge if we don't even know the database.
4158 if (!dbentry || !dbentry->functions)
4162 * Process all function entries in the message.
4164 for (i = 0; i < msg->m_nentries; i++)
4166 /* Remove from hashtable if present; we don't care if it's not. */
4167 (void) hash_search(dbentry->functions,
4168 (void *) &(msg->m_functionid[i]),