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) 2001-2008, PostgreSQL Global Development Group
16 * $PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.186 2008/12/17 09:15:03 heikki 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 int pgStatSock = -1;
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_deleted; /* tuples deleted in xact */
199 Oid t_id; /* table's OID */
200 bool t_shared; /* is it a shared catalog? */
201 } TwoPhasePgStatRecord;
204 * Info about current "snapshot" of stats file
206 static MemoryContext pgStatLocalContext = NULL;
207 static HTAB *pgStatDBHash = NULL;
208 static PgBackendStatus *localBackendStatusTable = NULL;
209 static int localNumBackends = 0;
212 * Cluster wide statistics, kept in the stats collector.
213 * Contains statistics that are not collected per database
216 static PgStat_GlobalStats globalStats;
218 /* Last time the collector successfully wrote the stats file */
219 static TimestampTz last_statwrite;
220 /* Latest statistics request time from backends */
221 static TimestampTz last_statrequest;
223 static volatile bool need_exit = false;
224 static volatile bool got_SIGHUP = false;
227 * Total time charged to functions so far in the current backend.
228 * We use this to help separate "self" and "other" time charges.
229 * (We assume this initializes to zero.)
231 static instr_time total_func_time;
235 * Local function forward declarations
239 static pid_t pgstat_forkexec(void);
242 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]);
243 static void pgstat_exit(SIGNAL_ARGS);
244 static void pgstat_beshutdown_hook(int code, Datum arg);
245 static void pgstat_sighup_handler(SIGNAL_ARGS);
247 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
248 static void pgstat_write_statsfile(bool permanent);
249 static HTAB *pgstat_read_statsfile(Oid onlydb, bool permanent);
250 static void backend_read_statsfile(void);
251 static void pgstat_read_current_status(void);
253 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
254 static void pgstat_send_funcstats(void);
255 static HTAB *pgstat_collect_oids(Oid catalogid);
257 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
259 static void pgstat_setup_memcxt(void);
261 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
262 static void pgstat_send(void *msg, int len);
264 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
265 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
266 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
267 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
268 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
269 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
270 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
271 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
272 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
273 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
274 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
277 /* ------------------------------------------------------------
278 * Public functions called from postmaster follow
279 * ------------------------------------------------------------
285 * Called from postmaster at startup. Create the resources required
286 * by the statistics collector process. If unable to do so, do not
287 * fail --- better to let the postmaster start with stats collection
294 ACCEPT_TYPE_ARG3 alen;
295 struct addrinfo *addrs = NULL,
305 #define TESTBYTEVAL ((char) 199)
308 * Create the UDP socket for sending and receiving statistic messages
310 hints.ai_flags = AI_PASSIVE;
311 hints.ai_family = PF_UNSPEC;
312 hints.ai_socktype = SOCK_DGRAM;
313 hints.ai_protocol = 0;
314 hints.ai_addrlen = 0;
315 hints.ai_addr = NULL;
316 hints.ai_canonname = NULL;
317 hints.ai_next = NULL;
318 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
322 (errmsg("could not resolve \"localhost\": %s",
323 gai_strerror(ret))));
328 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
329 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
330 * when kernel will reject IPv6). Worse, the failure may occur at the
331 * bind() or perhaps even connect() stage. So we must loop through the
332 * results till we find a working combination. We will generate LOG
333 * messages, but no error, for bogus combinations.
335 for (addr = addrs; addr; addr = addr->ai_next)
337 #ifdef HAVE_UNIX_SOCKETS
338 /* Ignore AF_UNIX sockets, if any are returned. */
339 if (addr->ai_family == AF_UNIX)
345 (errmsg("trying another address for the statistics collector")));
350 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) < 0)
353 (errcode_for_socket_access(),
354 errmsg("could not create socket for statistics collector: %m")));
359 * Bind it to a kernel assigned port on localhost and get the assigned
360 * port via getsockname().
362 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
365 (errcode_for_socket_access(),
366 errmsg("could not bind socket for statistics collector: %m")));
367 closesocket(pgStatSock);
372 alen = sizeof(pgStatAddr);
373 if (getsockname(pgStatSock, (struct sockaddr *) & pgStatAddr, &alen) < 0)
376 (errcode_for_socket_access(),
377 errmsg("could not get address of socket for statistics collector: %m")));
378 closesocket(pgStatSock);
384 * Connect the socket to its own address. This saves a few cycles by
385 * not having to respecify the target address on every send. This also
386 * provides a kernel-level check that only packets from this same
387 * address will be received.
389 if (connect(pgStatSock, (struct sockaddr *) & pgStatAddr, alen) < 0)
392 (errcode_for_socket_access(),
393 errmsg("could not connect socket for statistics collector: %m")));
394 closesocket(pgStatSock);
400 * Try to send and receive a one-byte test message on the socket. This
401 * is to catch situations where the socket can be created but will not
402 * actually pass data (for instance, because kernel packet filtering
405 test_byte = TESTBYTEVAL;
408 if (send(pgStatSock, &test_byte, 1, 0) != 1)
411 goto retry1; /* if interrupted, just retry */
413 (errcode_for_socket_access(),
414 errmsg("could not send test message on socket for statistics collector: %m")));
415 closesocket(pgStatSock);
421 * There could possibly be a little delay before the message can be
422 * received. We arbitrarily allow up to half a second before deciding
425 for (;;) /* need a loop to handle EINTR */
428 FD_SET(pgStatSock, &rset);
431 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
432 if (sel_res >= 0 || errno != EINTR)
438 (errcode_for_socket_access(),
439 errmsg("select() failed in statistics collector: %m")));
440 closesocket(pgStatSock);
444 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
447 * This is the case we actually think is likely, so take pains to
448 * give a specific message for it.
450 * errno will not be set meaningfully here, so don't use it.
453 (errcode(ERRCODE_CONNECTION_FAILURE),
454 errmsg("test message did not get through on socket for statistics collector")));
455 closesocket(pgStatSock);
460 test_byte++; /* just make sure variable is changed */
463 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
466 goto retry2; /* if interrupted, just retry */
468 (errcode_for_socket_access(),
469 errmsg("could not receive test message on socket for statistics collector: %m")));
470 closesocket(pgStatSock);
475 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
478 (errcode(ERRCODE_INTERNAL_ERROR),
479 errmsg("incorrect test message transmission on socket for statistics collector")));
480 closesocket(pgStatSock);
485 /* If we get here, we have a working socket */
489 /* Did we find a working address? */
490 if (!addr || pgStatSock < 0)
494 * Set the socket to non-blocking IO. This ensures that if the collector
495 * falls behind, statistics messages will be discarded; backends won't
496 * block waiting to send messages to the collector.
498 if (!pg_set_noblock(pgStatSock))
501 (errcode_for_socket_access(),
502 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
506 pg_freeaddrinfo_all(hints.ai_family, addrs);
512 (errmsg("disabling statistics collector for lack of working socket")));
515 pg_freeaddrinfo_all(hints.ai_family, addrs);
518 closesocket(pgStatSock);
522 * Adjust GUC variables to suppress useless activity, and for debugging
523 * purposes (seeing track_counts off is a clue that we failed here). We
524 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
525 * on from postgresql.conf without a restart.
527 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
531 * pgstat_reset_all() -
533 * Remove the stats file. This is currently used only if WAL
534 * recovery is needed after a crash.
537 pgstat_reset_all(void)
539 unlink(pgstat_stat_filename);
540 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
546 * pgstat_forkexec() -
548 * Format up the arglist for, then fork and exec, statistics collector process
551 pgstat_forkexec(void)
556 av[ac++] = "postgres";
557 av[ac++] = "--forkcol";
558 av[ac++] = NULL; /* filled in by postmaster_forkexec */
561 Assert(ac < lengthof(av));
563 return postmaster_forkexec(ac, av);
565 #endif /* EXEC_BACKEND */
571 * Called from postmaster at startup or after an existing collector
572 * died. Attempt to fire up a fresh statistics collector.
574 * Returns PID of child process, or 0 if fail.
576 * Note: if fail, we will be called again from the postmaster main loop.
585 * Check that the socket is there, else pgstat_init failed and we can do
592 * Do nothing if too soon since last collector start. This is a safety
593 * valve to protect against continuous respawn attempts if the collector
594 * is dying immediately at launch. Note that since we will be re-called
595 * from the postmaster main loop, we will get another chance later.
597 curtime = time(NULL);
598 if ((unsigned int) (curtime - last_pgstat_start_time) <
599 (unsigned int) PGSTAT_RESTART_INTERVAL)
601 last_pgstat_start_time = curtime;
604 * Okay, fork off the collector.
607 switch ((pgStatPid = pgstat_forkexec()))
609 switch ((pgStatPid = fork_process()))
614 (errmsg("could not fork statistics collector: %m")));
619 /* in postmaster child ... */
620 /* Close the postmaster's sockets */
621 ClosePostmasterPorts(false);
623 /* Lose the postmaster's on-exit routines */
626 /* Drop our connection to postmaster's shared memory, as well */
627 PGSharedMemoryDetach();
629 PgstatCollectorMain(0, NULL);
634 return (int) pgStatPid;
637 /* shouldn't get here */
642 allow_immediate_pgstat_restart(void)
644 last_pgstat_start_time = 0;
647 /* ------------------------------------------------------------
648 * Public functions used by backends follow
649 *------------------------------------------------------------
654 * pgstat_report_stat() -
656 * Called from tcop/postgres.c to send the so far collected per-table
657 * and function usage statistics to the collector. Note that this is
658 * called only when not within a transaction, so it is fair to use
659 * transaction stop time as an approximation of current time.
663 pgstat_report_stat(bool force)
665 /* we assume this inits to all zeroes: */
666 static const PgStat_TableCounts all_zeroes;
667 static TimestampTz last_report = 0;
670 PgStat_MsgTabstat regular_msg;
671 PgStat_MsgTabstat shared_msg;
675 /* Don't expend a clock check if nothing to do */
676 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0)
677 && !have_function_stats)
681 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
682 * msec since we last sent one, or the caller wants to force stats out.
684 now = GetCurrentTransactionStopTimestamp();
686 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
691 * Scan through the TabStatusArray struct(s) to find tables that actually
692 * have counts, and build messages to send. We have to separate shared
693 * relations from regular ones because the databaseid field in the message
694 * header has to depend on that.
696 regular_msg.m_databaseid = MyDatabaseId;
697 shared_msg.m_databaseid = InvalidOid;
698 regular_msg.m_nentries = 0;
699 shared_msg.m_nentries = 0;
701 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
703 for (i = 0; i < tsa->tsa_used; i++)
705 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
706 PgStat_MsgTabstat *this_msg;
707 PgStat_TableEntry *this_ent;
709 /* Shouldn't have any pending transaction-dependent counts */
710 Assert(entry->trans == NULL);
713 * Ignore entries that didn't accumulate any actual counts, such
714 * as indexes that were opened by the planner but not used.
716 if (memcmp(&entry->t_counts, &all_zeroes,
717 sizeof(PgStat_TableCounts)) == 0)
721 * OK, insert data into the appropriate message, and send if full.
723 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
724 this_ent = &this_msg->m_entry[this_msg->m_nentries];
725 this_ent->t_id = entry->t_id;
726 memcpy(&this_ent->t_counts, &entry->t_counts,
727 sizeof(PgStat_TableCounts));
728 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
730 pgstat_send_tabstat(this_msg);
731 this_msg->m_nentries = 0;
734 /* zero out TableStatus structs after use */
735 MemSet(tsa->tsa_entries, 0,
736 tsa->tsa_used * sizeof(PgStat_TableStatus));
741 * Send partial messages. If force is true, make sure that any pending
742 * xact commit/abort gets counted, even if no table stats to send.
744 if (regular_msg.m_nentries > 0 ||
745 (force && (pgStatXactCommit > 0 || pgStatXactRollback > 0)))
746 pgstat_send_tabstat(®ular_msg);
747 if (shared_msg.m_nentries > 0)
748 pgstat_send_tabstat(&shared_msg);
750 /* Now, send function statistics */
751 pgstat_send_funcstats();
755 * Subroutine for pgstat_report_stat: finish and send a tabstat message
758 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
763 /* It's unlikely we'd get here with no socket, but maybe not impossible */
768 * Report accumulated xact commit/rollback whenever we send a normal
771 if (OidIsValid(tsmsg->m_databaseid))
773 tsmsg->m_xact_commit = pgStatXactCommit;
774 tsmsg->m_xact_rollback = pgStatXactRollback;
775 pgStatXactCommit = 0;
776 pgStatXactRollback = 0;
780 tsmsg->m_xact_commit = 0;
781 tsmsg->m_xact_rollback = 0;
784 n = tsmsg->m_nentries;
785 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
786 n * sizeof(PgStat_TableEntry);
788 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
789 pgstat_send(tsmsg, len);
793 * Subroutine for pgstat_report_stat: populate and send a function stat message
796 pgstat_send_funcstats(void)
798 /* we assume this inits to all zeroes: */
799 static const PgStat_FunctionCounts all_zeroes;
801 PgStat_MsgFuncstat msg;
802 PgStat_BackendFunctionEntry *entry;
803 HASH_SEQ_STATUS fstat;
805 if (pgStatFunctions == NULL)
808 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
809 msg.m_databaseid = MyDatabaseId;
812 hash_seq_init(&fstat, pgStatFunctions);
813 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
815 PgStat_FunctionEntry *m_ent;
817 /* Skip it if no counts accumulated since last time */
818 if (memcmp(&entry->f_counts, &all_zeroes,
819 sizeof(PgStat_FunctionCounts)) == 0)
822 /* need to convert format of time accumulators */
823 m_ent = &msg.m_entry[msg.m_nentries];
824 m_ent->f_id = entry->f_id;
825 m_ent->f_numcalls = entry->f_counts.f_numcalls;
826 m_ent->f_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time);
827 m_ent->f_time_self = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time_self);
829 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
831 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
832 msg.m_nentries * sizeof(PgStat_FunctionEntry));
836 /* reset the entry's counts */
837 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
840 if (msg.m_nentries > 0)
841 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
842 msg.m_nentries * sizeof(PgStat_FunctionEntry));
844 have_function_stats = false;
849 * pgstat_vacuum_stat() -
851 * Will tell the collector about objects he can get rid of.
855 pgstat_vacuum_stat(void)
858 PgStat_MsgTabpurge msg;
859 PgStat_MsgFuncpurge f_msg;
860 HASH_SEQ_STATUS hstat;
861 PgStat_StatDBEntry *dbentry;
862 PgStat_StatTabEntry *tabentry;
863 PgStat_StatFuncEntry *funcentry;
870 * If not done for this transaction, read the statistics collector stats
871 * file into some hash tables.
873 backend_read_statsfile();
876 * Read pg_database and make a list of OIDs of all existing databases
878 htab = pgstat_collect_oids(DatabaseRelationId);
881 * Search the database hash table for dead databases and tell the
882 * collector to drop them.
884 hash_seq_init(&hstat, pgStatDBHash);
885 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
887 Oid dbid = dbentry->databaseid;
889 CHECK_FOR_INTERRUPTS();
891 /* the DB entry for shared tables (with InvalidOid) is never dropped */
892 if (OidIsValid(dbid) &&
893 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
894 pgstat_drop_database(dbid);
901 * Lookup our own database entry; if not found, nothing more to do.
903 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
904 (void *) &MyDatabaseId,
906 if (dbentry == NULL || dbentry->tables == NULL)
910 * Similarly to above, make a list of all known relations in this DB.
912 htab = pgstat_collect_oids(RelationRelationId);
915 * Initialize our messages table counter to zero
920 * Check for all tables listed in stats hashtable if they still exist.
922 hash_seq_init(&hstat, dbentry->tables);
923 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
925 Oid tabid = tabentry->tableid;
927 CHECK_FOR_INTERRUPTS();
929 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
933 * Not there, so add this table's Oid to the message
935 msg.m_tableid[msg.m_nentries++] = tabid;
938 * If the message is full, send it out and reinitialize to empty
940 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
942 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
943 +msg.m_nentries * sizeof(Oid);
945 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
946 msg.m_databaseid = MyDatabaseId;
947 pgstat_send(&msg, len);
956 if (msg.m_nentries > 0)
958 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
959 +msg.m_nentries * sizeof(Oid);
961 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
962 msg.m_databaseid = MyDatabaseId;
963 pgstat_send(&msg, len);
970 * Now repeat the above steps for functions. However, we needn't bother
971 * in the common case where no function stats are being collected.
973 if (dbentry->functions != NULL &&
974 hash_get_num_entries(dbentry->functions) > 0)
976 htab = pgstat_collect_oids(ProcedureRelationId);
978 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
979 f_msg.m_databaseid = MyDatabaseId;
980 f_msg.m_nentries = 0;
982 hash_seq_init(&hstat, dbentry->functions);
983 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
985 Oid funcid = funcentry->functionid;
987 CHECK_FOR_INTERRUPTS();
989 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
993 * Not there, so add this function's Oid to the message
995 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
998 * If the message is full, send it out and reinitialize to empty
1000 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1002 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1003 +f_msg.m_nentries * sizeof(Oid);
1005 pgstat_send(&f_msg, len);
1007 f_msg.m_nentries = 0;
1014 if (f_msg.m_nentries > 0)
1016 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1017 +f_msg.m_nentries * sizeof(Oid);
1019 pgstat_send(&f_msg, len);
1028 * pgstat_collect_oids() -
1030 * Collect the OIDs of all objects listed in the specified system catalog
1031 * into a temporary hash table. Caller should hash_destroy the result
1032 * when done with it.
1036 pgstat_collect_oids(Oid catalogid)
1044 memset(&hash_ctl, 0, sizeof(hash_ctl));
1045 hash_ctl.keysize = sizeof(Oid);
1046 hash_ctl.entrysize = sizeof(Oid);
1047 hash_ctl.hash = oid_hash;
1048 htab = hash_create("Temporary table of OIDs",
1049 PGSTAT_TAB_HASH_SIZE,
1051 HASH_ELEM | HASH_FUNCTION);
1053 rel = heap_open(catalogid, AccessShareLock);
1054 scan = heap_beginscan(rel, SnapshotNow, 0, NULL);
1055 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1057 Oid thisoid = HeapTupleGetOid(tup);
1059 CHECK_FOR_INTERRUPTS();
1061 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1064 heap_close(rel, AccessShareLock);
1071 * pgstat_drop_database() -
1073 * Tell the collector that we just dropped a database.
1074 * (If the message gets lost, we will still clean the dead DB eventually
1075 * via future invocations of pgstat_vacuum_stat().)
1079 pgstat_drop_database(Oid databaseid)
1081 PgStat_MsgDropdb msg;
1086 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1087 msg.m_databaseid = databaseid;
1088 pgstat_send(&msg, sizeof(msg));
1093 * pgstat_drop_relation() -
1095 * Tell the collector that we just dropped a relation.
1096 * (If the message gets lost, we will still clean the dead entry eventually
1097 * via future invocations of pgstat_vacuum_stat().)
1099 * Currently not used for lack of any good place to call it; we rely
1100 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1105 pgstat_drop_relation(Oid relid)
1107 PgStat_MsgTabpurge msg;
1113 msg.m_tableid[0] = relid;
1116 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) +sizeof(Oid);
1118 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1119 msg.m_databaseid = MyDatabaseId;
1120 pgstat_send(&msg, len);
1122 #endif /* NOT_USED */
1126 * pgstat_reset_counters() -
1128 * Tell the statistics collector to reset counters for our database.
1132 pgstat_reset_counters(void)
1134 PgStat_MsgResetcounter msg;
1141 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1142 errmsg("must be superuser to reset statistics counters")));
1144 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1145 msg.m_databaseid = MyDatabaseId;
1146 pgstat_send(&msg, sizeof(msg));
1151 * pgstat_report_autovac() -
1153 * Called from autovacuum.c to report startup of an autovacuum process.
1154 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1155 * the db OID must be passed in, instead.
1159 pgstat_report_autovac(Oid dboid)
1161 PgStat_MsgAutovacStart msg;
1166 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1167 msg.m_databaseid = dboid;
1168 msg.m_start_time = GetCurrentTimestamp();
1170 pgstat_send(&msg, sizeof(msg));
1175 * pgstat_report_vacuum() -
1177 * Tell the collector about the table we just vacuumed.
1181 pgstat_report_vacuum(Oid tableoid, bool shared, bool scanned_all,
1182 bool analyze, PgStat_Counter tuples)
1184 PgStat_MsgVacuum msg;
1186 if (pgStatSock < 0 || !pgstat_track_counts)
1189 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1190 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1191 msg.m_tableoid = tableoid;
1192 msg.m_scanned_all = scanned_all;
1193 msg.m_analyze = analyze;
1194 msg.m_autovacuum = IsAutoVacuumWorkerProcess(); /* is this autovacuum? */
1195 msg.m_vacuumtime = GetCurrentTimestamp();
1196 msg.m_tuples = tuples;
1197 pgstat_send(&msg, sizeof(msg));
1201 * pgstat_report_analyze() -
1203 * Tell the collector about the table we just analyzed.
1207 pgstat_report_analyze(Relation rel, PgStat_Counter livetuples,
1208 PgStat_Counter deadtuples)
1210 PgStat_MsgAnalyze msg;
1212 if (pgStatSock < 0 || !pgstat_track_counts)
1216 * Unlike VACUUM, ANALYZE might be running inside a transaction that
1217 * has already inserted and/or deleted rows in the target table.
1218 * ANALYZE will have counted such rows as live or dead respectively.
1219 * Because we will report our counts of such rows at transaction end,
1220 * we should subtract off these counts from what we send to the collector
1221 * now, else they'll be double-counted after commit. (This approach also
1222 * ensures that the collector ends up with the right numbers if we abort
1223 * instead of committing.)
1225 if (rel->pgstat_info != NULL)
1227 PgStat_TableXactStatus *trans;
1229 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1231 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1232 deadtuples -= trans->tuples_deleted;
1234 /* count stuff inserted by already-aborted subxacts, too */
1235 deadtuples -= rel->pgstat_info->t_counts.t_new_dead_tuples;
1236 /* Since ANALYZE's counts are estimates, we could have underflowed */
1237 livetuples = Max(livetuples, 0);
1238 deadtuples = Max(deadtuples, 0);
1241 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1242 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1243 msg.m_tableoid = RelationGetRelid(rel);
1244 msg.m_autovacuum = IsAutoVacuumWorkerProcess(); /* is this autovacuum? */
1245 msg.m_analyzetime = GetCurrentTimestamp();
1246 msg.m_live_tuples = livetuples;
1247 msg.m_dead_tuples = deadtuples;
1248 pgstat_send(&msg, sizeof(msg));
1255 * Send some junk data to the collector to increase traffic.
1261 PgStat_MsgDummy msg;
1266 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1267 pgstat_send(&msg, sizeof(msg));
1271 * pgstat_send_inquiry() -
1273 * Notify collector that we need fresh data.
1274 * ts specifies the minimum acceptable timestamp for the stats file.
1278 pgstat_send_inquiry(TimestampTz ts)
1280 PgStat_MsgInquiry msg;
1282 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1283 msg.inquiry_time = ts;
1284 pgstat_send(&msg, sizeof(msg));
1289 * Initialize function call usage data.
1290 * Called by the executor before invoking a function.
1293 pgstat_init_function_usage(FunctionCallInfoData *fcinfo,
1294 PgStat_FunctionCallUsage *fcu)
1296 PgStat_BackendFunctionEntry *htabent;
1299 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1301 /* stats not wanted */
1306 if (!pgStatFunctions)
1308 /* First time through - initialize function stat table */
1311 memset(&hash_ctl, 0, sizeof(hash_ctl));
1312 hash_ctl.keysize = sizeof(Oid);
1313 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1314 hash_ctl.hash = oid_hash;
1315 pgStatFunctions = hash_create("Function stat entries",
1316 PGSTAT_FUNCTION_HASH_SIZE,
1318 HASH_ELEM | HASH_FUNCTION);
1321 /* Get the stats entry for this function, create if necessary */
1322 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1323 HASH_ENTER, &found);
1325 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1327 fcu->fs = &htabent->f_counts;
1329 /* save stats for this function, later used to compensate for recursion */
1330 fcu->save_f_time = htabent->f_counts.f_time;
1332 /* save current backend-wide total time */
1333 fcu->save_total = total_func_time;
1335 /* get clock time as of function start */
1336 INSTR_TIME_SET_CURRENT(fcu->f_start);
1340 * Calculate function call usage and update stat counters.
1341 * Called by the executor after invoking a function.
1343 * In the case of a set-returning function that runs in value-per-call mode,
1344 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1345 * calls for what the user considers a single call of the function. The
1346 * finalize flag should be TRUE on the last call.
1349 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1351 PgStat_FunctionCounts *fs = fcu->fs;
1353 instr_time f_others;
1356 /* stats not wanted? */
1360 /* total elapsed time in this function call */
1361 INSTR_TIME_SET_CURRENT(f_total);
1362 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1364 /* self usage: elapsed minus anything already charged to other calls */
1365 f_others = total_func_time;
1366 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1368 INSTR_TIME_SUBTRACT(f_self, f_others);
1370 /* update backend-wide total time */
1371 INSTR_TIME_ADD(total_func_time, f_self);
1374 * Compute the new total f_time as the total elapsed time added to the
1375 * pre-call value of f_time. This is necessary to avoid double-counting
1376 * any time taken by recursive calls of myself. (We do not need any
1377 * similar kluge for self time, since that already excludes any
1380 INSTR_TIME_ADD(f_total, fcu->save_f_time);
1382 /* update counters in function stats table */
1385 fs->f_time = f_total;
1386 INSTR_TIME_ADD(fs->f_time_self, f_self);
1388 /* indicate that we have something to send */
1389 have_function_stats = true;
1394 * pgstat_initstats() -
1396 * Initialize a relcache entry to count access statistics.
1397 * Called whenever a relation is opened.
1399 * We assume that a relcache entry's pgstat_info field is zeroed by
1400 * relcache.c when the relcache entry is made; thereafter it is long-lived
1401 * data. We can avoid repeated searches of the TabStatus arrays when the
1402 * same relation is touched repeatedly within a transaction.
1406 pgstat_initstats(Relation rel)
1408 Oid rel_id = rel->rd_id;
1409 char relkind = rel->rd_rel->relkind;
1411 /* We only count stats for things that have storage */
1412 if (!(relkind == RELKIND_RELATION ||
1413 relkind == RELKIND_INDEX ||
1414 relkind == RELKIND_TOASTVALUE))
1416 rel->pgstat_info = NULL;
1420 if (pgStatSock < 0 || !pgstat_track_counts)
1422 /* We're not counting at all */
1423 rel->pgstat_info = NULL;
1428 * If we already set up this relation in the current transaction, nothing
1431 if (rel->pgstat_info != NULL &&
1432 rel->pgstat_info->t_id == rel_id)
1435 /* Else find or make the PgStat_TableStatus entry, and update link */
1436 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1440 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1442 static PgStat_TableStatus *
1443 get_tabstat_entry(Oid rel_id, bool isshared)
1445 PgStat_TableStatus *entry;
1446 TabStatusArray *tsa;
1447 TabStatusArray *prev_tsa;
1451 * Search the already-used tabstat slots for this relation.
1454 for (tsa = pgStatTabList; tsa != NULL; prev_tsa = tsa, tsa = tsa->tsa_next)
1456 for (i = 0; i < tsa->tsa_used; i++)
1458 entry = &tsa->tsa_entries[i];
1459 if (entry->t_id == rel_id)
1463 if (tsa->tsa_used < TABSTAT_QUANTUM)
1466 * It must not be present, but we found a free slot instead. Fine,
1467 * let's use this one. We assume the entry was already zeroed,
1468 * either at creation or after last use.
1470 entry = &tsa->tsa_entries[tsa->tsa_used++];
1471 entry->t_id = rel_id;
1472 entry->t_shared = isshared;
1478 * We ran out of tabstat slots, so allocate more. Be sure they're zeroed.
1480 tsa = (TabStatusArray *) MemoryContextAllocZero(TopMemoryContext,
1481 sizeof(TabStatusArray));
1483 prev_tsa->tsa_next = tsa;
1485 pgStatTabList = tsa;
1488 * Use the first entry of the new TabStatusArray.
1490 entry = &tsa->tsa_entries[tsa->tsa_used++];
1491 entry->t_id = rel_id;
1492 entry->t_shared = isshared;
1497 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1499 static PgStat_SubXactStatus *
1500 get_tabstat_stack_level(int nest_level)
1502 PgStat_SubXactStatus *xact_state;
1504 xact_state = pgStatXactStack;
1505 if (xact_state == NULL || xact_state->nest_level != nest_level)
1507 xact_state = (PgStat_SubXactStatus *)
1508 MemoryContextAlloc(TopTransactionContext,
1509 sizeof(PgStat_SubXactStatus));
1510 xact_state->nest_level = nest_level;
1511 xact_state->prev = pgStatXactStack;
1512 xact_state->first = NULL;
1513 pgStatXactStack = xact_state;
1519 * add_tabstat_xact_level - add a new (sub)transaction state record
1522 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1524 PgStat_SubXactStatus *xact_state;
1525 PgStat_TableXactStatus *trans;
1528 * If this is the first rel to be modified at the current nest level, we
1529 * first have to push a transaction stack entry.
1531 xact_state = get_tabstat_stack_level(nest_level);
1533 /* Now make a per-table stack entry */
1534 trans = (PgStat_TableXactStatus *)
1535 MemoryContextAllocZero(TopTransactionContext,
1536 sizeof(PgStat_TableXactStatus));
1537 trans->nest_level = nest_level;
1538 trans->upper = pgstat_info->trans;
1539 trans->parent = pgstat_info;
1540 trans->next = xact_state->first;
1541 xact_state->first = trans;
1542 pgstat_info->trans = trans;
1546 * pgstat_count_heap_insert - count a tuple insertion
1549 pgstat_count_heap_insert(Relation rel)
1551 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1553 if (pgstat_track_counts && pgstat_info != NULL)
1555 int nest_level = GetCurrentTransactionNestLevel();
1557 /* t_tuples_inserted is nontransactional, so just advance it */
1558 pgstat_info->t_counts.t_tuples_inserted++;
1560 /* We have to log the transactional effect at the proper level */
1561 if (pgstat_info->trans == NULL ||
1562 pgstat_info->trans->nest_level != nest_level)
1563 add_tabstat_xact_level(pgstat_info, nest_level);
1565 pgstat_info->trans->tuples_inserted++;
1570 * pgstat_count_heap_update - count a tuple update
1573 pgstat_count_heap_update(Relation rel, bool hot)
1575 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1577 if (pgstat_track_counts && pgstat_info != NULL)
1579 int nest_level = GetCurrentTransactionNestLevel();
1581 /* t_tuples_updated is nontransactional, so just advance it */
1582 pgstat_info->t_counts.t_tuples_updated++;
1583 /* ditto for the hot_update counter */
1585 pgstat_info->t_counts.t_tuples_hot_updated++;
1587 /* We have to log the transactional effect at the proper level */
1588 if (pgstat_info->trans == NULL ||
1589 pgstat_info->trans->nest_level != nest_level)
1590 add_tabstat_xact_level(pgstat_info, nest_level);
1592 /* An UPDATE both inserts a new tuple and deletes the old */
1593 pgstat_info->trans->tuples_inserted++;
1594 pgstat_info->trans->tuples_deleted++;
1599 * pgstat_count_heap_delete - count a tuple deletion
1602 pgstat_count_heap_delete(Relation rel)
1604 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1606 if (pgstat_track_counts && pgstat_info != NULL)
1608 int nest_level = GetCurrentTransactionNestLevel();
1610 /* t_tuples_deleted is nontransactional, so just advance it */
1611 pgstat_info->t_counts.t_tuples_deleted++;
1613 /* We have to log the transactional effect at the proper level */
1614 if (pgstat_info->trans == NULL ||
1615 pgstat_info->trans->nest_level != nest_level)
1616 add_tabstat_xact_level(pgstat_info, nest_level);
1618 pgstat_info->trans->tuples_deleted++;
1623 * pgstat_update_heap_dead_tuples - update dead-tuples count
1625 * The semantics of this are that we are reporting the nontransactional
1626 * recovery of "delta" dead tuples; so t_new_dead_tuples decreases
1627 * rather than increasing, and the change goes straight into the per-table
1628 * counter, not into transactional state.
1631 pgstat_update_heap_dead_tuples(Relation rel, int delta)
1633 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1635 if (pgstat_track_counts && pgstat_info != NULL)
1636 pgstat_info->t_counts.t_new_dead_tuples -= delta;
1643 * Called from access/transam/xact.c at top-level transaction commit/abort.
1647 AtEOXact_PgStat(bool isCommit)
1649 PgStat_SubXactStatus *xact_state;
1652 * Count transaction commit or abort. (We use counters, not just bools,
1653 * in case the reporting message isn't sent right away.)
1658 pgStatXactRollback++;
1661 * Transfer transactional insert/update counts into the base tabstat
1662 * entries. We don't bother to free any of the transactional state, since
1663 * it's all in TopTransactionContext and will go away anyway.
1665 xact_state = pgStatXactStack;
1666 if (xact_state != NULL)
1668 PgStat_TableXactStatus *trans;
1670 Assert(xact_state->nest_level == 1);
1671 Assert(xact_state->prev == NULL);
1672 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1674 PgStat_TableStatus *tabstat;
1676 Assert(trans->nest_level == 1);
1677 Assert(trans->upper == NULL);
1678 tabstat = trans->parent;
1679 Assert(tabstat->trans == trans);
1682 tabstat->t_counts.t_new_live_tuples +=
1683 trans->tuples_inserted - trans->tuples_deleted;
1684 tabstat->t_counts.t_new_dead_tuples += trans->tuples_deleted;
1688 /* inserted tuples are dead, deleted tuples are unaffected */
1689 tabstat->t_counts.t_new_dead_tuples += trans->tuples_inserted;
1691 tabstat->trans = NULL;
1694 pgStatXactStack = NULL;
1696 /* Make sure any stats snapshot is thrown away */
1697 pgstat_clear_snapshot();
1701 * AtEOSubXact_PgStat
1703 * Called from access/transam/xact.c at subtransaction commit/abort.
1707 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
1709 PgStat_SubXactStatus *xact_state;
1712 * Transfer transactional insert/update counts into the next higher
1713 * subtransaction state.
1715 xact_state = pgStatXactStack;
1716 if (xact_state != NULL &&
1717 xact_state->nest_level >= nestDepth)
1719 PgStat_TableXactStatus *trans;
1720 PgStat_TableXactStatus *next_trans;
1722 /* delink xact_state from stack immediately to simplify reuse case */
1723 pgStatXactStack = xact_state->prev;
1725 for (trans = xact_state->first; trans != NULL; trans = next_trans)
1727 PgStat_TableStatus *tabstat;
1729 next_trans = trans->next;
1730 Assert(trans->nest_level == nestDepth);
1731 tabstat = trans->parent;
1732 Assert(tabstat->trans == trans);
1735 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
1737 trans->upper->tuples_inserted += trans->tuples_inserted;
1738 trans->upper->tuples_deleted += trans->tuples_deleted;
1739 tabstat->trans = trans->upper;
1745 * When there isn't an immediate parent state, we can just
1746 * reuse the record instead of going through a
1747 * palloc/pfree pushup (this works since it's all in
1748 * TopTransactionContext anyway). We have to re-link it
1749 * into the parent level, though, and that might mean
1750 * pushing a new entry into the pgStatXactStack.
1752 PgStat_SubXactStatus *upper_xact_state;
1754 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
1755 trans->next = upper_xact_state->first;
1756 upper_xact_state->first = trans;
1757 trans->nest_level = nestDepth - 1;
1763 * On abort, inserted tuples are dead (and can be bounced out
1764 * to the top-level tabstat), deleted tuples are unaffected
1766 tabstat->t_counts.t_new_dead_tuples += trans->tuples_inserted;
1767 tabstat->trans = trans->upper;
1778 * Save the transactional stats state at 2PC transaction prepare.
1780 * In this phase we just generate 2PC records for all the pending
1781 * transaction-dependent stats work.
1784 AtPrepare_PgStat(void)
1786 PgStat_SubXactStatus *xact_state;
1788 xact_state = pgStatXactStack;
1789 if (xact_state != NULL)
1791 PgStat_TableXactStatus *trans;
1793 Assert(xact_state->nest_level == 1);
1794 Assert(xact_state->prev == NULL);
1795 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1797 PgStat_TableStatus *tabstat;
1798 TwoPhasePgStatRecord record;
1800 Assert(trans->nest_level == 1);
1801 Assert(trans->upper == NULL);
1802 tabstat = trans->parent;
1803 Assert(tabstat->trans == trans);
1805 record.tuples_inserted = trans->tuples_inserted;
1806 record.tuples_deleted = trans->tuples_deleted;
1807 record.t_id = tabstat->t_id;
1808 record.t_shared = tabstat->t_shared;
1810 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
1811 &record, sizeof(TwoPhasePgStatRecord));
1817 * PostPrepare_PgStat
1818 * Clean up after successful PREPARE.
1820 * All we need do here is unlink the transaction stats state from the
1821 * nontransactional state. The nontransactional action counts will be
1822 * reported to the stats collector immediately, while the effects on live
1823 * and dead tuple counts are preserved in the 2PC state file.
1825 * Note: AtEOXact_PgStat is not called during PREPARE.
1828 PostPrepare_PgStat(void)
1830 PgStat_SubXactStatus *xact_state;
1833 * We don't bother to free any of the transactional state, since it's all
1834 * in TopTransactionContext and will go away anyway.
1836 xact_state = pgStatXactStack;
1837 if (xact_state != NULL)
1839 PgStat_TableXactStatus *trans;
1841 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1843 PgStat_TableStatus *tabstat;
1845 tabstat = trans->parent;
1846 tabstat->trans = NULL;
1849 pgStatXactStack = NULL;
1851 /* Make sure any stats snapshot is thrown away */
1852 pgstat_clear_snapshot();
1856 * 2PC processing routine for COMMIT PREPARED case.
1858 * Load the saved counts into our local pgstats state.
1861 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
1862 void *recdata, uint32 len)
1864 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
1865 PgStat_TableStatus *pgstat_info;
1867 /* Find or create a tabstat entry for the rel */
1868 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
1870 pgstat_info->t_counts.t_new_live_tuples +=
1871 rec->tuples_inserted - rec->tuples_deleted;
1872 pgstat_info->t_counts.t_new_dead_tuples += rec->tuples_deleted;
1876 * 2PC processing routine for ROLLBACK PREPARED case.
1878 * Load the saved counts into our local pgstats state, but treat them
1882 pgstat_twophase_postabort(TransactionId xid, uint16 info,
1883 void *recdata, uint32 len)
1885 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
1886 PgStat_TableStatus *pgstat_info;
1888 /* Find or create a tabstat entry for the rel */
1889 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
1891 /* inserted tuples are dead, deleted tuples are no-ops */
1892 pgstat_info->t_counts.t_new_dead_tuples += rec->tuples_inserted;
1897 * pgstat_fetch_stat_dbentry() -
1899 * Support function for the SQL-callable pgstat* functions. Returns
1900 * the collected statistics for one database or NULL. NULL doesn't mean
1901 * that the database doesn't exist, it is just not yet known by the
1902 * collector, so the caller is better off to report ZERO instead.
1905 PgStat_StatDBEntry *
1906 pgstat_fetch_stat_dbentry(Oid dbid)
1909 * If not done for this transaction, read the statistics collector stats
1910 * file into some hash tables.
1912 backend_read_statsfile();
1915 * Lookup the requested database; return NULL if not found
1917 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1924 * pgstat_fetch_stat_tabentry() -
1926 * Support function for the SQL-callable pgstat* functions. Returns
1927 * the collected statistics for one table or NULL. NULL doesn't mean
1928 * that the table doesn't exist, it is just not yet known by the
1929 * collector, so the caller is better off to report ZERO instead.
1932 PgStat_StatTabEntry *
1933 pgstat_fetch_stat_tabentry(Oid relid)
1936 PgStat_StatDBEntry *dbentry;
1937 PgStat_StatTabEntry *tabentry;
1940 * If not done for this transaction, read the statistics collector stats
1941 * file into some hash tables.
1943 backend_read_statsfile();
1946 * Lookup our database, then look in its table hash table.
1948 dbid = MyDatabaseId;
1949 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1952 if (dbentry != NULL && dbentry->tables != NULL)
1954 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
1962 * If we didn't find it, maybe it's a shared table.
1965 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1968 if (dbentry != NULL && dbentry->tables != NULL)
1970 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
1982 * pgstat_fetch_stat_funcentry() -
1984 * Support function for the SQL-callable pgstat* functions. Returns
1985 * the collected statistics for one function or NULL.
1988 PgStat_StatFuncEntry *
1989 pgstat_fetch_stat_funcentry(Oid func_id)
1991 PgStat_StatDBEntry *dbentry;
1992 PgStat_StatFuncEntry *funcentry = NULL;
1994 /* load the stats file if needed */
1995 backend_read_statsfile();
1997 /* Lookup our database, then find the requested function. */
1998 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
1999 if (dbentry != NULL && dbentry->functions != NULL)
2001 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2011 * pgstat_fetch_stat_beentry() -
2013 * Support function for the SQL-callable pgstat* functions. Returns
2014 * our local copy of the current-activity entry for one backend.
2016 * NB: caller is responsible for a check if the user is permitted to see
2017 * this info (especially the querystring).
2021 pgstat_fetch_stat_beentry(int beid)
2023 pgstat_read_current_status();
2025 if (beid < 1 || beid > localNumBackends)
2028 return &localBackendStatusTable[beid - 1];
2033 * pgstat_fetch_stat_numbackends() -
2035 * Support function for the SQL-callable pgstat* functions. Returns
2036 * the maximum current backend id.
2040 pgstat_fetch_stat_numbackends(void)
2042 pgstat_read_current_status();
2044 return localNumBackends;
2049 * pgstat_fetch_global() -
2051 * Support function for the SQL-callable pgstat* functions. Returns
2052 * a pointer to the global statistics struct.
2055 PgStat_GlobalStats *
2056 pgstat_fetch_global(void)
2058 backend_read_statsfile();
2060 return &globalStats;
2064 /* ------------------------------------------------------------
2065 * Functions for management of the shared-memory PgBackendStatus array
2066 * ------------------------------------------------------------
2069 static PgBackendStatus *BackendStatusArray = NULL;
2070 static PgBackendStatus *MyBEEntry = NULL;
2071 static char *BackendActivityBuffer = NULL;
2075 * Report shared-memory space needed by CreateSharedBackendStatus.
2078 BackendStatusShmemSize(void)
2082 size = add_size(mul_size(sizeof(PgBackendStatus), MaxBackends),
2083 mul_size(pgstat_track_activity_query_size, MaxBackends));
2088 * Initialize the shared status array and activity string buffer during
2089 * postmaster startup.
2092 CreateSharedBackendStatus(void)
2099 /* Create or attach to the shared array */
2100 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2101 BackendStatusArray = (PgBackendStatus *)
2102 ShmemInitStruct("Backend Status Array", size, &found);
2107 * We're the first - initialize.
2109 MemSet(BackendStatusArray, 0, size);
2112 /* Create or attach to the shared activity buffer */
2113 size = mul_size(pgstat_track_activity_query_size, MaxBackends);
2114 BackendActivityBuffer = (char*)
2115 ShmemInitStruct("Backend Activity Buffer", size, &found);
2119 MemSet(BackendActivityBuffer, 0, size);
2121 /* Initialize st_activity pointers. */
2122 buffer = BackendActivityBuffer;
2123 for (i = 0; i < MaxBackends; i++) {
2124 BackendStatusArray[i].st_activity = buffer;
2125 buffer += pgstat_track_activity_query_size;
2132 * pgstat_initialize() -
2134 * Initialize pgstats state, and set up our on-proc-exit hook.
2135 * Called from InitPostgres. MyBackendId must be set,
2136 * but we must not have started any transaction yet (since the
2137 * exit hook must run after the last transaction exit).
2141 pgstat_initialize(void)
2143 /* Initialize MyBEEntry */
2144 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2145 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2147 /* Set up a process-exit hook to clean up */
2148 on_shmem_exit(pgstat_beshutdown_hook, 0);
2152 * pgstat_bestart() -
2154 * Initialize this backend's entry in the PgBackendStatus array.
2155 * Called from InitPostgres. MyDatabaseId and session userid must be set
2156 * (hence, this cannot be combined with pgstat_initialize).
2160 pgstat_bestart(void)
2162 TimestampTz proc_start_timestamp;
2164 SockAddr clientaddr;
2165 volatile PgBackendStatus *beentry;
2168 * To minimize the time spent modifying the PgBackendStatus entry, fetch
2169 * all the needed data first.
2171 * If we have a MyProcPort, use its session start time (for consistency,
2172 * and to save a kernel call).
2175 proc_start_timestamp = MyProcPort->SessionStartTime;
2177 proc_start_timestamp = GetCurrentTimestamp();
2178 userid = GetSessionUserId();
2181 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2182 * If so, use all-zeroes client address, which is dealt with specially in
2183 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2186 memcpy(&clientaddr, &MyProcPort->raddr, sizeof(clientaddr));
2188 MemSet(&clientaddr, 0, sizeof(clientaddr));
2191 * Initialize my status entry, following the protocol of bumping
2192 * st_changecount before and after; and make sure it's even afterwards. We
2193 * use a volatile pointer here to ensure the compiler doesn't try to get
2196 beentry = MyBEEntry;
2199 beentry->st_changecount++;
2200 } while ((beentry->st_changecount & 1) == 0);
2202 beentry->st_procpid = MyProcPid;
2203 beentry->st_proc_start_timestamp = proc_start_timestamp;
2204 beentry->st_activity_start_timestamp = 0;
2205 beentry->st_xact_start_timestamp = 0;
2206 beentry->st_databaseid = MyDatabaseId;
2207 beentry->st_userid = userid;
2208 beentry->st_clientaddr = clientaddr;
2209 beentry->st_waiting = false;
2210 beentry->st_activity[0] = '\0';
2211 /* Also make sure the last byte in the string area is always 0 */
2212 beentry->st_activity[pgstat_track_activity_query_size - 1] = '\0';
2214 beentry->st_changecount++;
2215 Assert((beentry->st_changecount & 1) == 0);
2219 * Shut down a single backend's statistics reporting at process exit.
2221 * Flush any remaining statistics counts out to the collector.
2222 * Without this, operations triggered during backend exit (such as
2223 * temp table deletions) won't be counted.
2225 * Lastly, clear out our entry in the PgBackendStatus array.
2228 pgstat_beshutdown_hook(int code, Datum arg)
2230 volatile PgBackendStatus *beentry = MyBEEntry;
2232 pgstat_report_stat(true);
2235 * Clear my status entry, following the protocol of bumping st_changecount
2236 * before and after. We use a volatile pointer here to ensure the
2237 * compiler doesn't try to get cute.
2239 beentry->st_changecount++;
2241 beentry->st_procpid = 0; /* mark invalid */
2243 beentry->st_changecount++;
2244 Assert((beentry->st_changecount & 1) == 0);
2249 * pgstat_report_activity() -
2251 * Called from tcop/postgres.c to report what the backend is actually doing
2252 * (usually "<IDLE>" or the start of the query to be executed).
2256 pgstat_report_activity(const char *cmd_str)
2258 volatile PgBackendStatus *beentry = MyBEEntry;
2259 TimestampTz start_timestamp;
2262 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
2264 if (!pgstat_track_activities || !beentry)
2268 * To minimize the time spent modifying the entry, fetch all the needed
2271 start_timestamp = GetCurrentStatementStartTimestamp();
2273 len = strlen(cmd_str);
2274 len = pg_mbcliplen(cmd_str, len, pgstat_track_activity_query_size - 1);
2277 * Update my status entry, following the protocol of bumping
2278 * st_changecount before and after. We use a volatile pointer here to
2279 * ensure the compiler doesn't try to get cute.
2281 beentry->st_changecount++;
2283 beentry->st_activity_start_timestamp = start_timestamp;
2284 memcpy((char *) beentry->st_activity, cmd_str, len);
2285 beentry->st_activity[len] = '\0';
2287 beentry->st_changecount++;
2288 Assert((beentry->st_changecount & 1) == 0);
2292 * Report current transaction start timestamp as the specified value.
2293 * Zero means there is no active transaction.
2296 pgstat_report_xact_timestamp(TimestampTz tstamp)
2298 volatile PgBackendStatus *beentry = MyBEEntry;
2300 if (!pgstat_track_activities || !beentry)
2304 * Update my status entry, following the protocol of bumping
2305 * st_changecount before and after. We use a volatile pointer here to
2306 * ensure the compiler doesn't try to get cute.
2308 beentry->st_changecount++;
2309 beentry->st_xact_start_timestamp = tstamp;
2310 beentry->st_changecount++;
2311 Assert((beentry->st_changecount & 1) == 0);
2315 * pgstat_report_waiting() -
2317 * Called from lock manager to report beginning or end of a lock wait.
2319 * NB: this *must* be able to survive being called before MyBEEntry has been
2324 pgstat_report_waiting(bool waiting)
2326 volatile PgBackendStatus *beentry = MyBEEntry;
2328 if (!pgstat_track_activities || !beentry)
2332 * Since this is a single-byte field in a struct that only this process
2333 * may modify, there seems no need to bother with the st_changecount
2334 * protocol. The update must appear atomic in any case.
2336 beentry->st_waiting = waiting;
2341 * pgstat_read_current_status() -
2343 * Copy the current contents of the PgBackendStatus array to local memory,
2344 * if not already done in this transaction.
2348 pgstat_read_current_status(void)
2350 volatile PgBackendStatus *beentry;
2351 PgBackendStatus *localtable;
2352 PgBackendStatus *localentry;
2353 char *localactivity;
2356 Assert(!pgStatRunningInCollector);
2357 if (localBackendStatusTable)
2358 return; /* already done */
2360 pgstat_setup_memcxt();
2362 localtable = (PgBackendStatus *)
2363 MemoryContextAlloc(pgStatLocalContext,
2364 sizeof(PgBackendStatus) * MaxBackends);
2365 localactivity = (char *)
2366 MemoryContextAlloc(pgStatLocalContext,
2367 pgstat_track_activity_query_size * MaxBackends);
2368 localNumBackends = 0;
2370 beentry = BackendStatusArray;
2371 localentry = localtable;
2372 for (i = 1; i <= MaxBackends; i++)
2375 * Follow the protocol of retrying if st_changecount changes while we
2376 * copy the entry, or if it's odd. (The check for odd is needed to
2377 * cover the case where we are able to completely copy the entry while
2378 * the source backend is between increment steps.) We use a volatile
2379 * pointer here to ensure the compiler doesn't try to get cute.
2383 int save_changecount = beentry->st_changecount;
2385 localentry->st_procpid = beentry->st_procpid;
2386 if (localentry->st_procpid > 0)
2388 memcpy(localentry, (char *) beentry, sizeof(PgBackendStatus));
2390 * strcpy is safe even if the string is modified concurrently,
2391 * because there's always a \0 at the end of the buffer.
2393 strcpy(localactivity, (char *) beentry->st_activity);
2394 localentry->st_activity = localactivity;
2397 if (save_changecount == beentry->st_changecount &&
2398 (save_changecount & 1) == 0)
2401 /* Make sure we can break out of loop if stuck... */
2402 CHECK_FOR_INTERRUPTS();
2406 /* Only valid entries get included into the local array */
2407 if (localentry->st_procpid > 0)
2410 localactivity += pgstat_track_activity_query_size;
2415 /* Set the pointer only after completion of a valid table */
2416 localBackendStatusTable = localtable;
2421 * pgstat_get_backend_current_activity() -
2423 * Return a string representing the current activity of the backend with
2424 * the specified PID. This looks directly at the BackendStatusArray,
2425 * and so will provide current information regardless of the age of our
2426 * transaction's snapshot of the status array.
2428 * It is the caller's responsibility to invoke this only for backends whose
2429 * state is expected to remain stable while the result is in use. The
2430 * only current use is in deadlock reporting, where we can expect that
2431 * the target backend is blocked on a lock. (There are corner cases
2432 * where the target's wait could get aborted while we are looking at it,
2433 * but the very worst consequence is to return a pointer to a string
2434 * that's been changed, so we won't worry too much.)
2436 * Note: return strings for special cases match pg_stat_get_backend_activity.
2440 pgstat_get_backend_current_activity(int pid, bool checkUser)
2442 PgBackendStatus *beentry;
2445 beentry = BackendStatusArray;
2446 for (i = 1; i <= MaxBackends; i++)
2449 * Although we expect the target backend's entry to be stable, that
2450 * doesn't imply that anyone else's is. To avoid identifying the
2451 * wrong backend, while we check for a match to the desired PID we
2452 * must follow the protocol of retrying if st_changecount changes
2453 * while we examine the entry, or if it's odd. (This might be
2454 * unnecessary, since fetching or storing an int is almost certainly
2455 * atomic, but let's play it safe.) We use a volatile pointer here
2456 * to ensure the compiler doesn't try to get cute.
2458 volatile PgBackendStatus *vbeentry = beentry;
2463 int save_changecount = vbeentry->st_changecount;
2465 found = (vbeentry->st_procpid == pid);
2467 if (save_changecount == vbeentry->st_changecount &&
2468 (save_changecount & 1) == 0)
2471 /* Make sure we can break out of loop if stuck... */
2472 CHECK_FOR_INTERRUPTS();
2477 /* Now it is safe to use the non-volatile pointer */
2478 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
2479 return "<insufficient privilege>";
2480 else if (*(beentry->st_activity) == '\0')
2481 return "<command string not enabled>";
2483 return beentry->st_activity;
2489 /* If we get here, caller is in error ... */
2490 return "<backend information not available>";
2494 /* ------------------------------------------------------------
2495 * Local support functions follow
2496 * ------------------------------------------------------------
2501 * pgstat_setheader() -
2503 * Set common header fields in a statistics message
2507 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
2509 hdr->m_type = mtype;
2516 * Send out one statistics message to the collector
2520 pgstat_send(void *msg, int len)
2527 ((PgStat_MsgHdr *) msg)->m_size = len;
2529 /* We'll retry after EINTR, but ignore all other failures */
2532 rc = send(pgStatSock, msg, len, 0);
2533 } while (rc < 0 && errno == EINTR);
2535 #ifdef USE_ASSERT_CHECKING
2536 /* In debug builds, log send failures ... */
2538 elog(LOG, "could not send to statistics collector: %m");
2543 * pgstat_send_bgwriter() -
2545 * Send bgwriter statistics to the collector
2549 pgstat_send_bgwriter(void)
2551 /* We assume this initializes to zeroes */
2552 static const PgStat_MsgBgWriter all_zeroes;
2555 * This function can be called even if nothing at all has happened. In
2556 * this case, avoid sending a completely empty message to the stats
2559 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
2563 * Prepare and send the message
2565 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
2566 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
2569 * Clear out the statistics buffer, so it can be re-used.
2571 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
2576 * PgstatCollectorMain() -
2578 * Start up the statistics collector process. This is the body of the
2579 * postmaster child process.
2581 * The argc/argv parameters are valid only in EXEC_BACKEND case.
2584 NON_EXEC_STATIC void
2585 PgstatCollectorMain(int argc, char *argv[])
2592 struct pollfd input_fd;
2594 struct timeval sel_timeout;
2599 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2601 MyProcPid = getpid(); /* reset MyProcPid */
2603 MyStartTime = time(NULL); /* record Start Time for logging */
2606 * If possible, make this process a group leader, so that the postmaster
2607 * can signal any child processes too. (pgstat probably never has any
2608 * child processes, but for consistency we make all postmaster child
2609 * processes do this.)
2613 elog(FATAL, "setsid() failed: %m");
2617 * Ignore all signals usually bound to some action in the postmaster,
2620 pqsignal(SIGHUP, pgstat_sighup_handler);
2621 pqsignal(SIGINT, SIG_IGN);
2622 pqsignal(SIGTERM, SIG_IGN);
2623 pqsignal(SIGQUIT, pgstat_exit);
2624 pqsignal(SIGALRM, SIG_IGN);
2625 pqsignal(SIGPIPE, SIG_IGN);
2626 pqsignal(SIGUSR1, SIG_IGN);
2627 pqsignal(SIGUSR2, SIG_IGN);
2628 pqsignal(SIGCHLD, SIG_DFL);
2629 pqsignal(SIGTTIN, SIG_DFL);
2630 pqsignal(SIGTTOU, SIG_DFL);
2631 pqsignal(SIGCONT, SIG_DFL);
2632 pqsignal(SIGWINCH, SIG_DFL);
2633 PG_SETMASK(&UnBlockSig);
2636 * Identify myself via ps
2638 init_ps_display("stats collector process", "", "", "");
2641 * Arrange to write the initial status file right away
2643 last_statrequest = GetCurrentTimestamp();
2644 last_statwrite = last_statrequest - 1;
2647 * Read in an existing statistics stats file or initialize the stats to
2650 pgStatRunningInCollector = true;
2651 pgStatDBHash = pgstat_read_statsfile(InvalidOid, true);
2654 * Setup the descriptor set for select(2). Since only one bit in the set
2655 * ever changes, we need not repeat FD_ZERO each time.
2657 #if !defined(HAVE_POLL) && !defined(WIN32)
2662 * Loop to process messages until we get SIGQUIT or detect ungraceful
2663 * death of our parent postmaster.
2665 * For performance reasons, we don't want to do a PostmasterIsAlive() test
2666 * after every message; instead, do it only when select()/poll() is
2667 * interrupted by timeout. In essence, we'll stay alive as long as
2668 * backends keep sending us stuff often, even if the postmaster is gone.
2675 * Quit if we get SIGQUIT from the postmaster.
2681 * Reload configuration if we got SIGHUP from the postmaster.
2685 ProcessConfigFile(PGC_SIGHUP);
2690 * Write the stats file if a new request has arrived that is not
2691 * satisfied by existing file.
2693 if (last_statwrite < last_statrequest)
2694 pgstat_write_statsfile(false);
2697 * Wait for a message to arrive; but not for more than
2698 * PGSTAT_SELECT_TIMEOUT seconds. (This determines how quickly we will
2699 * shut down after an ungraceful postmaster termination; so it needn't
2700 * be very fast. However, on some systems SIGQUIT won't interrupt the
2701 * poll/select call, so this also limits speed of response to SIGQUIT,
2702 * which is more important.)
2704 * We use poll(2) if available, otherwise select(2). Win32 has its own
2709 input_fd.fd = pgStatSock;
2710 input_fd.events = POLLIN | POLLERR;
2711 input_fd.revents = 0;
2713 if (poll(&input_fd, 1, PGSTAT_SELECT_TIMEOUT * 1000) < 0)
2718 (errcode_for_socket_access(),
2719 errmsg("poll() failed in statistics collector: %m")));
2722 got_data = (input_fd.revents != 0);
2723 #else /* !HAVE_POLL */
2725 FD_SET(pgStatSock, &rfds);
2728 * timeout struct is modified by select() on some operating systems,
2729 * so re-fill it each time.
2731 sel_timeout.tv_sec = PGSTAT_SELECT_TIMEOUT;
2732 sel_timeout.tv_usec = 0;
2734 if (select(pgStatSock + 1, &rfds, NULL, NULL, &sel_timeout) < 0)
2739 (errcode_for_socket_access(),
2740 errmsg("select() failed in statistics collector: %m")));
2743 got_data = FD_ISSET(pgStatSock, &rfds);
2744 #endif /* HAVE_POLL */
2746 got_data = pgwin32_waitforsinglesocket(pgStatSock, FD_READ,
2747 PGSTAT_SELECT_TIMEOUT * 1000);
2751 * If there is a message on the socket, read it and check for
2756 len = recv(pgStatSock, (char *) &msg,
2757 sizeof(PgStat_Msg), 0);
2763 (errcode_for_socket_access(),
2764 errmsg("could not read statistics message: %m")));
2768 * We ignore messages that are smaller than our common header
2770 if (len < sizeof(PgStat_MsgHdr))
2774 * The received length must match the length in the header
2776 if (msg.msg_hdr.m_size != len)
2780 * O.K. - we accept this message. Process it.
2782 switch (msg.msg_hdr.m_type)
2784 case PGSTAT_MTYPE_DUMMY:
2787 case PGSTAT_MTYPE_INQUIRY:
2788 pgstat_recv_inquiry((PgStat_MsgInquiry *) &msg, len);
2791 case PGSTAT_MTYPE_TABSTAT:
2792 pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len);
2795 case PGSTAT_MTYPE_TABPURGE:
2796 pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len);
2799 case PGSTAT_MTYPE_DROPDB:
2800 pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len);
2803 case PGSTAT_MTYPE_RESETCOUNTER:
2804 pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg,
2808 case PGSTAT_MTYPE_AUTOVAC_START:
2809 pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
2812 case PGSTAT_MTYPE_VACUUM:
2813 pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
2816 case PGSTAT_MTYPE_ANALYZE:
2817 pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len);
2820 case PGSTAT_MTYPE_BGWRITER:
2821 pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len);
2824 case PGSTAT_MTYPE_FUNCSTAT:
2825 pgstat_recv_funcstat((PgStat_MsgFuncstat *) &msg, len);
2828 case PGSTAT_MTYPE_FUNCPURGE:
2829 pgstat_recv_funcpurge((PgStat_MsgFuncpurge *) &msg, len);
2839 * We can only get here if the select/poll timeout elapsed. Check
2840 * for postmaster death.
2842 if (!PostmasterIsAlive(true))
2845 } /* end of message-processing loop */
2848 * Save the final stats to reuse at next startup.
2850 pgstat_write_statsfile(true);
2856 /* SIGQUIT signal handler for collector process */
2858 pgstat_exit(SIGNAL_ARGS)
2863 /* SIGHUP handler for collector process */
2865 pgstat_sighup_handler(SIGNAL_ARGS)
2872 * Lookup the hash table entry for the specified database. If no hash
2873 * table entry exists, initialize it, if the create parameter is true.
2874 * Else, return NULL.
2876 static PgStat_StatDBEntry *
2877 pgstat_get_db_entry(Oid databaseid, bool create)
2879 PgStat_StatDBEntry *result;
2881 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
2883 /* Lookup or create the hash table entry for this database */
2884 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2888 if (!create && !found)
2891 /* If not found, initialize the new one. */
2896 result->tables = NULL;
2897 result->functions = NULL;
2898 result->n_xact_commit = 0;
2899 result->n_xact_rollback = 0;
2900 result->n_blocks_fetched = 0;
2901 result->n_blocks_hit = 0;
2902 result->n_tuples_returned = 0;
2903 result->n_tuples_fetched = 0;
2904 result->n_tuples_inserted = 0;
2905 result->n_tuples_updated = 0;
2906 result->n_tuples_deleted = 0;
2907 result->last_autovac_time = 0;
2909 memset(&hash_ctl, 0, sizeof(hash_ctl));
2910 hash_ctl.keysize = sizeof(Oid);
2911 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
2912 hash_ctl.hash = oid_hash;
2913 result->tables = hash_create("Per-database table",
2914 PGSTAT_TAB_HASH_SIZE,
2916 HASH_ELEM | HASH_FUNCTION);
2918 hash_ctl.keysize = sizeof(Oid);
2919 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
2920 hash_ctl.hash = oid_hash;
2921 result->functions = hash_create("Per-database function",
2922 PGSTAT_FUNCTION_HASH_SIZE,
2924 HASH_ELEM | HASH_FUNCTION);
2932 * pgstat_write_statsfile() -
2935 * If writing to the permanent file (happens when the collector is
2936 * shutting down only), remove the temporary file so that backends
2937 * starting up under a new postmaster can't read the old data before
2938 * the new collector is ready.
2942 pgstat_write_statsfile(bool permanent)
2944 HASH_SEQ_STATUS hstat;
2945 HASH_SEQ_STATUS tstat;
2946 HASH_SEQ_STATUS fstat;
2947 PgStat_StatDBEntry *dbentry;
2948 PgStat_StatTabEntry *tabentry;
2949 PgStat_StatFuncEntry *funcentry;
2952 const char *tmpfile = permanent?PGSTAT_STAT_PERMANENT_TMPFILE:pgstat_stat_tmpname;
2953 const char *statfile = permanent?PGSTAT_STAT_PERMANENT_FILENAME:pgstat_stat_filename;
2956 * Open the statistics temp file to write out the current values.
2958 fpout = AllocateFile(tmpfile, PG_BINARY_W);
2962 (errcode_for_file_access(),
2963 errmsg("could not open temporary statistics file \"%s\": %m",
2969 * Set the timestamp of the stats file.
2971 globalStats.stats_timestamp = GetCurrentTimestamp();
2974 * Write the file header --- currently just a format ID.
2976 format_id = PGSTAT_FILE_FORMAT_ID;
2977 fwrite(&format_id, sizeof(format_id), 1, fpout);
2980 * Write global stats struct
2982 fwrite(&globalStats, sizeof(globalStats), 1, fpout);
2985 * Walk through the database table.
2987 hash_seq_init(&hstat, pgStatDBHash);
2988 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
2991 * Write out the DB entry including the number of live backends. We
2992 * don't write the tables or functions pointers, since they're of
2993 * no use to any other process.
2996 fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
2999 * Walk through the database's access stats per table.
3001 hash_seq_init(&tstat, dbentry->tables);
3002 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
3005 fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
3009 * Walk through the database's function stats table.
3011 hash_seq_init(&fstat, dbentry->functions);
3012 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
3015 fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
3019 * Mark the end of this DB
3025 * No more output to be done. Close the temp file and replace the old
3026 * pgstat.stat with it. The ferror() check replaces testing for error
3027 * after each individual fputc or fwrite above.
3034 (errcode_for_file_access(),
3035 errmsg("could not write temporary statistics file \"%s\": %m",
3040 else if (FreeFile(fpout) < 0)
3043 (errcode_for_file_access(),
3044 errmsg("could not close temporary statistics file \"%s\": %m",
3048 else if (rename(tmpfile, statfile) < 0)
3051 (errcode_for_file_access(),
3052 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
3053 tmpfile, statfile)));
3059 * Successful write, so update last_statwrite.
3061 last_statwrite = globalStats.stats_timestamp;
3064 * It's not entirely clear whether there could be clock skew between
3065 * backends and the collector; but just in case someone manages to
3066 * send us a stats request time that's far in the future, reset it.
3067 * This ensures that no inquiry message can cause more than one stats
3068 * file write to occur.
3070 last_statrequest = last_statwrite;
3074 unlink(pgstat_stat_filename);
3079 * pgstat_read_statsfile() -
3081 * Reads in an existing statistics collector file and initializes the
3082 * databases' hash table (whose entries point to the tables' hash tables).
3086 pgstat_read_statsfile(Oid onlydb, bool permanent)
3088 PgStat_StatDBEntry *dbentry;
3089 PgStat_StatDBEntry dbbuf;
3090 PgStat_StatTabEntry *tabentry;
3091 PgStat_StatTabEntry tabbuf;
3092 PgStat_StatFuncEntry funcbuf;
3093 PgStat_StatFuncEntry *funcentry;
3096 HTAB *tabhash = NULL;
3097 HTAB *funchash = NULL;
3101 const char *statfile = permanent?PGSTAT_STAT_PERMANENT_FILENAME:pgstat_stat_filename;
3104 * The tables will live in pgStatLocalContext.
3106 pgstat_setup_memcxt();
3109 * Create the DB hashtable
3111 memset(&hash_ctl, 0, sizeof(hash_ctl));
3112 hash_ctl.keysize = sizeof(Oid);
3113 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
3114 hash_ctl.hash = oid_hash;
3115 hash_ctl.hcxt = pgStatLocalContext;
3116 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
3117 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3120 * Clear out global statistics so they start from zero in case we can't
3121 * load an existing statsfile.
3123 memset(&globalStats, 0, sizeof(globalStats));
3126 * Try to open the status file. If it doesn't exist, the backends simply
3127 * return zero for anything and the collector simply starts from scratch
3128 * with empty counters.
3130 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3134 * Verify it's of the expected format.
3136 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3137 || format_id != PGSTAT_FILE_FORMAT_ID)
3139 ereport(pgStatRunningInCollector ? LOG : WARNING,
3140 (errmsg("corrupted pgstat.stat file")));
3145 * Read global stats struct
3147 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
3149 ereport(pgStatRunningInCollector ? LOG : WARNING,
3150 (errmsg("corrupted pgstat.stat file")));
3155 * We found an existing collector stats file. Read it and put all the
3156 * hashtable entries into place.
3160 switch (fgetc(fpin))
3163 * 'D' A PgStat_StatDBEntry struct describing a database
3164 * follows. Subsequently, zero to many 'T' and 'F' entries
3165 * will follow until a 'd' is encountered.
3168 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
3169 fpin) != offsetof(PgStat_StatDBEntry, tables))
3171 ereport(pgStatRunningInCollector ? LOG : WARNING,
3172 (errmsg("corrupted pgstat.stat file")));
3177 * Add to the DB hash
3179 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
3180 (void *) &dbbuf.databaseid,
3185 ereport(pgStatRunningInCollector ? LOG : WARNING,
3186 (errmsg("corrupted pgstat.stat file")));
3190 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
3191 dbentry->tables = NULL;
3192 dbentry->functions = NULL;
3195 * Don't collect tables if not the requested DB (or the
3196 * shared-table info)
3198 if (onlydb != InvalidOid)
3200 if (dbbuf.databaseid != onlydb &&
3201 dbbuf.databaseid != InvalidOid)
3205 memset(&hash_ctl, 0, sizeof(hash_ctl));
3206 hash_ctl.keysize = sizeof(Oid);
3207 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3208 hash_ctl.hash = oid_hash;
3209 hash_ctl.hcxt = pgStatLocalContext;
3210 dbentry->tables = hash_create("Per-database table",
3211 PGSTAT_TAB_HASH_SIZE,
3213 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3215 hash_ctl.keysize = sizeof(Oid);
3216 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3217 hash_ctl.hash = oid_hash;
3218 hash_ctl.hcxt = pgStatLocalContext;
3219 dbentry->functions = hash_create("Per-database function",
3220 PGSTAT_FUNCTION_HASH_SIZE,
3222 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3224 * Arrange that following records add entries to this
3225 * database's hash tables.
3227 tabhash = dbentry->tables;
3228 funchash = dbentry->functions;
3232 * 'd' End of this database.
3240 * 'T' A PgStat_StatTabEntry follows.
3243 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
3244 fpin) != sizeof(PgStat_StatTabEntry))
3246 ereport(pgStatRunningInCollector ? LOG : WARNING,
3247 (errmsg("corrupted pgstat.stat file")));
3252 * Skip if table belongs to a not requested database.
3254 if (tabhash == NULL)
3257 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
3258 (void *) &tabbuf.tableid,
3259 HASH_ENTER, &found);
3263 ereport(pgStatRunningInCollector ? LOG : WARNING,
3264 (errmsg("corrupted pgstat.stat file")));
3268 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
3272 * 'F' A PgStat_StatFuncEntry follows.
3275 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
3276 fpin) != sizeof(PgStat_StatFuncEntry))
3278 ereport(pgStatRunningInCollector ? LOG : WARNING,
3279 (errmsg("corrupted pgstat.stat file")));
3284 * Skip if function belongs to a not requested database.
3286 if (funchash == NULL)
3289 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
3290 (void *) &funcbuf.functionid,
3291 HASH_ENTER, &found);
3295 ereport(pgStatRunningInCollector ? LOG : WARNING,
3296 (errmsg("corrupted pgstat.stat file")));
3300 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
3304 * 'E' The EOF marker of a complete stats file.
3310 ereport(pgStatRunningInCollector ? LOG : WARNING,
3311 (errmsg("corrupted pgstat.stat file")));
3320 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
3326 * pgstat_read_statsfile_timestamp() -
3328 * Attempt to fetch the timestamp of an existing stats file.
3329 * Returns TRUE if successful (timestamp is stored at *ts).
3333 pgstat_read_statsfile_timestamp(bool permanent, TimestampTz *ts)
3335 PgStat_GlobalStats myGlobalStats;
3338 const char *statfile = permanent?PGSTAT_STAT_PERMANENT_FILENAME:pgstat_stat_filename;
3341 * Try to open the status file.
3343 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3347 * Verify it's of the expected format.
3349 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3350 || format_id != PGSTAT_FILE_FORMAT_ID)
3357 * Read global stats struct
3359 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats), fpin) != sizeof(myGlobalStats))
3365 *ts = myGlobalStats.stats_timestamp;
3372 * If not already done, read the statistics collector stats file into
3373 * some hash tables. The results will be kept until pgstat_clear_snapshot()
3374 * is called (typically, at end of transaction).
3377 backend_read_statsfile(void)
3382 /* already read it? */
3385 Assert(!pgStatRunningInCollector);
3388 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL msec
3389 * before now. This indirectly ensures that the collector needn't write
3390 * the file more often than PGSTAT_STAT_INTERVAL. In an autovacuum
3391 * worker, however, we want a lower delay to avoid using stale data, so we
3392 * use PGSTAT_RETRY_DELAY (since the number of worker is low, this
3393 * shouldn't be a problem).
3395 * Note that we don't recompute min_ts after sleeping; so we might end up
3396 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In practice
3397 * that shouldn't happen, though, as long as the sleep time is less than
3398 * PGSTAT_STAT_INTERVAL; and we don't want to lie to the collector about
3399 * what our cutoff time really is.
3401 if (IsAutoVacuumWorkerProcess())
3402 min_ts = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
3403 -PGSTAT_RETRY_DELAY);
3405 min_ts = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
3406 -PGSTAT_STAT_INTERVAL);
3409 * Loop until fresh enough stats file is available or we ran out of time.
3410 * The stats inquiry message is sent repeatedly in case collector drops it.
3412 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
3414 TimestampTz file_ts = 0;
3416 CHECK_FOR_INTERRUPTS();
3418 if (pgstat_read_statsfile_timestamp(false, &file_ts) &&
3422 /* Not there or too old, so kick the collector and wait a bit */
3423 pgstat_send_inquiry(min_ts);
3424 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
3427 if (count >= PGSTAT_POLL_LOOP_COUNT)
3428 elog(WARNING, "pgstat wait timeout");
3430 /* Autovacuum launcher wants stats about all databases */
3431 if (IsAutoVacuumLauncherProcess())
3432 pgStatDBHash = pgstat_read_statsfile(InvalidOid, false);
3434 pgStatDBHash = pgstat_read_statsfile(MyDatabaseId, false);
3439 * pgstat_setup_memcxt() -
3441 * Create pgStatLocalContext, if not already done.
3445 pgstat_setup_memcxt(void)
3447 if (!pgStatLocalContext)
3448 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
3449 "Statistics snapshot",
3450 ALLOCSET_SMALL_MINSIZE,
3451 ALLOCSET_SMALL_INITSIZE,
3452 ALLOCSET_SMALL_MAXSIZE);
3457 * pgstat_clear_snapshot() -
3459 * Discard any data collected in the current transaction. Any subsequent
3460 * request will cause new snapshots to be read.
3462 * This is also invoked during transaction commit or abort to discard
3463 * the no-longer-wanted snapshot.
3467 pgstat_clear_snapshot(void)
3469 /* Release memory, if any was allocated */
3470 if (pgStatLocalContext)
3471 MemoryContextDelete(pgStatLocalContext);
3473 /* Reset variables */
3474 pgStatLocalContext = NULL;
3475 pgStatDBHash = NULL;
3476 localBackendStatusTable = NULL;
3477 localNumBackends = 0;
3482 * pgstat_recv_inquiry() -
3484 * Process stat inquiry requests.
3488 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
3490 if (msg->inquiry_time > last_statrequest)
3491 last_statrequest = msg->inquiry_time;
3496 * pgstat_recv_tabstat() -
3498 * Count what the backend has done.
3502 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
3504 PgStat_TableEntry *tabmsg = &(msg->m_entry[0]);
3505 PgStat_StatDBEntry *dbentry;
3506 PgStat_StatTabEntry *tabentry;
3510 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3513 * Update database-wide stats.
3515 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
3516 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
3519 * Process all table entries in the message.
3521 for (i = 0; i < msg->m_nentries; i++)
3523 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3524 (void *) &(tabmsg[i].t_id),
3525 HASH_ENTER, &found);
3530 * If it's a new table entry, initialize counters to the values we
3533 tabentry->numscans = tabmsg[i].t_counts.t_numscans;
3534 tabentry->tuples_returned = tabmsg[i].t_counts.t_tuples_returned;
3535 tabentry->tuples_fetched = tabmsg[i].t_counts.t_tuples_fetched;
3536 tabentry->tuples_inserted = tabmsg[i].t_counts.t_tuples_inserted;
3537 tabentry->tuples_updated = tabmsg[i].t_counts.t_tuples_updated;
3538 tabentry->tuples_deleted = tabmsg[i].t_counts.t_tuples_deleted;
3539 tabentry->tuples_hot_updated = tabmsg[i].t_counts.t_tuples_hot_updated;
3540 tabentry->n_live_tuples = tabmsg[i].t_counts.t_new_live_tuples;
3541 tabentry->n_dead_tuples = tabmsg[i].t_counts.t_new_dead_tuples;
3542 tabentry->blocks_fetched = tabmsg[i].t_counts.t_blocks_fetched;
3543 tabentry->blocks_hit = tabmsg[i].t_counts.t_blocks_hit;
3545 tabentry->last_anl_tuples = 0;
3546 tabentry->vacuum_timestamp = 0;
3547 tabentry->autovac_vacuum_timestamp = 0;
3548 tabentry->analyze_timestamp = 0;
3549 tabentry->autovac_analyze_timestamp = 0;
3554 * Otherwise add the values to the existing entry.
3556 tabentry->numscans += tabmsg[i].t_counts.t_numscans;
3557 tabentry->tuples_returned += tabmsg[i].t_counts.t_tuples_returned;
3558 tabentry->tuples_fetched += tabmsg[i].t_counts.t_tuples_fetched;
3559 tabentry->tuples_inserted += tabmsg[i].t_counts.t_tuples_inserted;
3560 tabentry->tuples_updated += tabmsg[i].t_counts.t_tuples_updated;
3561 tabentry->tuples_deleted += tabmsg[i].t_counts.t_tuples_deleted;
3562 tabentry->tuples_hot_updated += tabmsg[i].t_counts.t_tuples_hot_updated;
3563 tabentry->n_live_tuples += tabmsg[i].t_counts.t_new_live_tuples;
3564 tabentry->n_dead_tuples += tabmsg[i].t_counts.t_new_dead_tuples;
3565 tabentry->blocks_fetched += tabmsg[i].t_counts.t_blocks_fetched;
3566 tabentry->blocks_hit += tabmsg[i].t_counts.t_blocks_hit;
3569 /* Clamp n_live_tuples in case of negative new_live_tuples */
3570 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
3571 /* Likewise for n_dead_tuples */
3572 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
3575 * Add per-table stats to the per-database entry, too.
3577 dbentry->n_tuples_returned += tabmsg[i].t_counts.t_tuples_returned;
3578 dbentry->n_tuples_fetched += tabmsg[i].t_counts.t_tuples_fetched;
3579 dbentry->n_tuples_inserted += tabmsg[i].t_counts.t_tuples_inserted;
3580 dbentry->n_tuples_updated += tabmsg[i].t_counts.t_tuples_updated;
3581 dbentry->n_tuples_deleted += tabmsg[i].t_counts.t_tuples_deleted;
3582 dbentry->n_blocks_fetched += tabmsg[i].t_counts.t_blocks_fetched;
3583 dbentry->n_blocks_hit += tabmsg[i].t_counts.t_blocks_hit;
3589 * pgstat_recv_tabpurge() -
3591 * Arrange for dead table removal.
3595 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
3597 PgStat_StatDBEntry *dbentry;
3600 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3603 * No need to purge if we don't even know the database.
3605 if (!dbentry || !dbentry->tables)
3609 * Process all table entries in the message.
3611 for (i = 0; i < msg->m_nentries; i++)
3613 /* Remove from hashtable if present; we don't care if it's not. */
3614 (void) hash_search(dbentry->tables,
3615 (void *) &(msg->m_tableid[i]),
3622 * pgstat_recv_dropdb() -
3624 * Arrange for dead database removal
3628 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
3630 PgStat_StatDBEntry *dbentry;
3633 * Lookup the database in the hashtable.
3635 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3638 * If found, remove it.
3642 if (dbentry->tables != NULL)
3643 hash_destroy(dbentry->tables);
3644 if (dbentry->functions != NULL)
3645 hash_destroy(dbentry->functions);
3647 if (hash_search(pgStatDBHash,
3648 (void *) &(dbentry->databaseid),
3649 HASH_REMOVE, NULL) == NULL)
3651 (errmsg("database hash table corrupted "
3652 "during cleanup --- abort")));
3658 * pgstat_recv_resetcounter() -
3660 * Reset the statistics for the specified database.
3664 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
3667 PgStat_StatDBEntry *dbentry;
3670 * Lookup the database in the hashtable. Nothing to do if not there.
3672 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3678 * We simply throw away all the database's table entries by recreating a
3679 * new hash table for them.
3681 if (dbentry->tables != NULL)
3682 hash_destroy(dbentry->tables);
3683 if (dbentry->functions != NULL)
3684 hash_destroy(dbentry->functions);
3686 dbentry->tables = NULL;
3687 dbentry->functions = NULL;
3688 dbentry->n_xact_commit = 0;
3689 dbentry->n_xact_rollback = 0;
3690 dbentry->n_blocks_fetched = 0;
3691 dbentry->n_blocks_hit = 0;
3693 memset(&hash_ctl, 0, sizeof(hash_ctl));
3694 hash_ctl.keysize = sizeof(Oid);
3695 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3696 hash_ctl.hash = oid_hash;
3697 dbentry->tables = hash_create("Per-database table",
3698 PGSTAT_TAB_HASH_SIZE,
3700 HASH_ELEM | HASH_FUNCTION);
3702 hash_ctl.keysize = sizeof(Oid);
3703 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3704 hash_ctl.hash = oid_hash;
3705 dbentry->functions = hash_create("Per-database function",
3706 PGSTAT_FUNCTION_HASH_SIZE,
3708 HASH_ELEM | HASH_FUNCTION);
3712 * pgstat_recv_autovac() -
3714 * Process an autovacuum signalling message.
3718 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
3720 PgStat_StatDBEntry *dbentry;
3723 * Lookup the database in the hashtable. Don't create the entry if it
3724 * doesn't exist, because autovacuum may be processing a template
3725 * database. If this isn't the case, the database is most likely to have
3726 * an entry already. (If it doesn't, not much harm is done anyway --
3727 * it'll get created as soon as somebody actually uses the database.)
3729 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3730 if (dbentry == NULL)
3734 * Store the last autovacuum time in the database entry.
3736 dbentry->last_autovac_time = msg->m_start_time;
3740 * pgstat_recv_vacuum() -
3742 * Process a VACUUM message.
3746 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
3748 PgStat_StatDBEntry *dbentry;
3749 PgStat_StatTabEntry *tabentry;
3752 * Don't create either the database or table entry if it doesn't already
3753 * exist. This avoids bloating the stats with entries for stuff that is
3754 * only touched by vacuum and not by live operations.
3756 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3757 if (dbentry == NULL)
3760 tabentry = hash_search(dbentry->tables, &(msg->m_tableoid),
3762 if (tabentry == NULL)
3765 if (msg->m_autovacuum)
3766 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
3768 tabentry->vacuum_timestamp = msg->m_vacuumtime;
3769 if (msg->m_scanned_all)
3770 tabentry->n_live_tuples = msg->m_tuples;
3771 /* Resetting dead_tuples to 0 is an approximation ... */
3772 tabentry->n_dead_tuples = 0;
3775 if (msg->m_scanned_all)
3776 tabentry->last_anl_tuples = msg->m_tuples;
3777 if (msg->m_autovacuum)
3778 tabentry->autovac_analyze_timestamp = msg->m_vacuumtime;
3780 tabentry->analyze_timestamp = msg->m_vacuumtime;
3784 /* last_anl_tuples must never exceed n_live_tuples+n_dead_tuples */
3785 tabentry->last_anl_tuples = Min(tabentry->last_anl_tuples,
3786 tabentry->n_live_tuples);
3791 * pgstat_recv_analyze() -
3793 * Process an ANALYZE message.
3797 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
3799 PgStat_StatDBEntry *dbentry;
3800 PgStat_StatTabEntry *tabentry;
3803 * Don't create either the database or table entry if it doesn't already
3804 * exist. This avoids bloating the stats with entries for stuff that is
3805 * only touched by analyze and not by live operations.
3807 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3808 if (dbentry == NULL)
3811 tabentry = hash_search(dbentry->tables, &(msg->m_tableoid),
3813 if (tabentry == NULL)
3816 if (msg->m_autovacuum)
3817 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
3819 tabentry->analyze_timestamp = msg->m_analyzetime;
3820 tabentry->n_live_tuples = msg->m_live_tuples;
3821 tabentry->n_dead_tuples = msg->m_dead_tuples;
3822 tabentry->last_anl_tuples = msg->m_live_tuples + msg->m_dead_tuples;
3827 * pgstat_recv_bgwriter() -
3829 * Process a BGWRITER message.
3833 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
3835 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
3836 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
3837 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
3838 globalStats.buf_written_clean += msg->m_buf_written_clean;
3839 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
3840 globalStats.buf_written_backend += msg->m_buf_written_backend;
3841 globalStats.buf_alloc += msg->m_buf_alloc;
3845 * pgstat_recv_funcstat() -
3847 * Count what the backend has done.
3851 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
3853 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
3854 PgStat_StatDBEntry *dbentry;
3855 PgStat_StatFuncEntry *funcentry;
3859 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3862 * Process all function entries in the message.
3864 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
3866 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
3867 (void *) &(funcmsg->f_id),
3868 HASH_ENTER, &found);
3873 * If it's a new function entry, initialize counters to the values
3876 funcentry->f_numcalls = funcmsg->f_numcalls;
3877 funcentry->f_time = funcmsg->f_time;
3878 funcentry->f_time_self = funcmsg->f_time_self;
3883 * Otherwise add the values to the existing entry.
3885 funcentry->f_numcalls += funcmsg->f_numcalls;
3886 funcentry->f_time += funcmsg->f_time;
3887 funcentry->f_time_self += funcmsg->f_time_self;
3893 * pgstat_recv_funcpurge() -
3895 * Arrange for dead function removal.
3899 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
3901 PgStat_StatDBEntry *dbentry;
3904 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3907 * No need to purge if we don't even know the database.
3909 if (!dbentry || !dbentry->functions)
3913 * Process all function entries in the message.
3915 for (i = 0; i < msg->m_nentries; i++)
3917 /* Remove from hashtable if present; we don't care if it's not. */
3918 (void) hash_search(dbentry->functions,
3919 (void *) &(msg->m_functionid[i]),