4 * All the statistics collector stuff hacked up in one big, ugly file.
6 * TODO: - Separate collector, postmaster and backend stuff
7 * into different files.
9 * - Add some automatic call for pgstat vacuuming.
11 * - Add a pgstat config column to pg_database, so this
12 * entire thing can be enabled/disabled on a per db basis.
14 * Copyright (c) 2001-2010, PostgreSQL Global Development Group
16 * $PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.197 2010/01/10 14:16:07 mha Exp $
23 #include <sys/param.h>
25 #include <sys/socket.h>
27 #include <netinet/in.h>
28 #include <arpa/inet.h>
34 #ifdef HAVE_SYS_POLL_H
40 #include "access/heapam.h"
41 #include "access/transam.h"
42 #include "access/twophase_rmgr.h"
43 #include "access/xact.h"
44 #include "catalog/pg_database.h"
45 #include "catalog/pg_proc.h"
47 #include "libpq/libpq.h"
48 #include "libpq/pqsignal.h"
49 #include "mb/pg_wchar.h"
50 #include "miscadmin.h"
52 #include "postmaster/autovacuum.h"
53 #include "postmaster/fork_process.h"
54 #include "postmaster/postmaster.h"
55 #include "storage/backendid.h"
56 #include "storage/fd.h"
57 #include "storage/ipc.h"
58 #include "storage/pg_shmem.h"
59 #include "storage/pmsignal.h"
60 #include "utils/guc.h"
61 #include "utils/memutils.h"
62 #include "utils/ps_status.h"
63 #include "utils/rel.h"
64 #include "utils/tqual.h"
68 * Paths for the statistics files (relative to installation's $PGDATA).
71 #define PGSTAT_STAT_PERMANENT_FILENAME "global/pgstat.stat"
72 #define PGSTAT_STAT_PERMANENT_TMPFILE "global/pgstat.tmp"
78 #define PGSTAT_STAT_INTERVAL 500 /* Minimum time between stats file
79 * updates; in milliseconds. */
81 #define PGSTAT_RETRY_DELAY 10 /* How long to wait between statistics
82 * update requests; in milliseconds. */
84 #define PGSTAT_MAX_WAIT_TIME 5000 /* Maximum time to wait for a stats
85 * file update; in milliseconds. */
87 #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a
88 * failed statistics collector; in
91 #define PGSTAT_SELECT_TIMEOUT 2 /* How often to check for postmaster
92 * death; in seconds. */
94 #define PGSTAT_POLL_LOOP_COUNT (PGSTAT_MAX_WAIT_TIME / PGSTAT_RETRY_DELAY)
98 * The initial size hints for the hash tables used in the collector.
101 #define PGSTAT_DB_HASH_SIZE 16
102 #define PGSTAT_TAB_HASH_SIZE 512
103 #define PGSTAT_FUNCTION_HASH_SIZE 512
110 bool pgstat_track_activities = false;
111 bool pgstat_track_counts = false;
112 int pgstat_track_functions = TRACK_FUNC_OFF;
113 int pgstat_track_activity_query_size = 1024;
116 * Built from GUC parameter
119 char *pgstat_stat_filename = NULL;
120 char *pgstat_stat_tmpname = NULL;
123 * BgWriter global statistics counters (unused in other processes).
124 * Stored directly in a stats message structure so it can be sent
125 * without needing to copy things around. We assume this inits to zeroes.
127 PgStat_MsgBgWriter BgWriterStats;
133 NON_EXEC_STATIC pgsocket pgStatSock = PGINVALID_SOCKET;
135 static struct sockaddr_storage pgStatAddr;
137 static time_t last_pgstat_start_time;
139 static bool pgStatRunningInCollector = false;
142 * Structures in which backends store per-table info that's waiting to be
143 * sent to the collector.
145 * NOTE: once allocated, TabStatusArray structures are never moved or deleted
146 * for the life of the backend. Also, we zero out the t_id fields of the
147 * contained PgStat_TableStatus structs whenever they are not actively in use.
148 * This allows relcache pgstat_info pointers to be treated as long-lived data,
149 * avoiding repeated searches in pgstat_initstats() when a relation is
150 * repeatedly opened during a transaction.
152 #define TABSTAT_QUANTUM 100 /* we alloc this many at a time */
154 typedef struct TabStatusArray
156 struct TabStatusArray *tsa_next; /* link to next array, if any */
157 int tsa_used; /* # entries currently used */
158 PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */
161 static TabStatusArray *pgStatTabList = NULL;
164 * Backends store per-function info that's waiting to be sent to the collector
165 * in this hash table (indexed by function OID).
167 static HTAB *pgStatFunctions = NULL;
170 * Indicates if backend has some function stats that it hasn't yet
171 * sent to the collector.
173 static bool have_function_stats = false;
176 * Tuple insertion/deletion counts for an open transaction can't be propagated
177 * into PgStat_TableStatus counters until we know if it is going to commit
178 * or abort. Hence, we keep these counts in per-subxact structs that live
179 * in TopTransactionContext. This data structure is designed on the assumption
180 * that subxacts won't usually modify very many tables.
182 typedef struct PgStat_SubXactStatus
184 int nest_level; /* subtransaction nest level */
185 struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
186 PgStat_TableXactStatus *first; /* head of list for this subxact */
187 } PgStat_SubXactStatus;
189 static PgStat_SubXactStatus *pgStatXactStack = NULL;
191 static int pgStatXactCommit = 0;
192 static int pgStatXactRollback = 0;
194 /* Record that's written to 2PC state file when pgstat state is persisted */
195 typedef struct TwoPhasePgStatRecord
197 PgStat_Counter tuples_inserted; /* tuples inserted in xact */
198 PgStat_Counter tuples_updated; /* tuples updated in xact */
199 PgStat_Counter tuples_deleted; /* tuples deleted in xact */
200 Oid t_id; /* table's OID */
201 bool t_shared; /* is it a shared catalog? */
202 } TwoPhasePgStatRecord;
205 * Info about current "snapshot" of stats file
207 static MemoryContext pgStatLocalContext = NULL;
208 static HTAB *pgStatDBHash = NULL;
209 static PgBackendStatus *localBackendStatusTable = NULL;
210 static int localNumBackends = 0;
213 * Cluster wide statistics, kept in the stats collector.
214 * Contains statistics that are not collected per database
217 static PgStat_GlobalStats globalStats;
219 /* Last time the collector successfully wrote the stats file */
220 static TimestampTz last_statwrite;
222 /* Latest statistics request time from backends */
223 static TimestampTz last_statrequest;
225 static volatile bool need_exit = false;
226 static volatile bool got_SIGHUP = false;
229 * Total time charged to functions so far in the current backend.
230 * We use this to help separate "self" and "other" time charges.
231 * (We assume this initializes to zero.)
233 static instr_time total_func_time;
237 * Local function forward declarations
241 static pid_t pgstat_forkexec(void);
244 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]);
245 static void pgstat_exit(SIGNAL_ARGS);
246 static void pgstat_beshutdown_hook(int code, Datum arg);
247 static void pgstat_sighup_handler(SIGNAL_ARGS);
249 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
250 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
251 Oid tableoid, bool create);
252 static void pgstat_write_statsfile(bool permanent);
253 static HTAB *pgstat_read_statsfile(Oid onlydb, bool permanent);
254 static void backend_read_statsfile(void);
255 static void pgstat_read_current_status(void);
257 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
258 static void pgstat_send_funcstats(void);
259 static HTAB *pgstat_collect_oids(Oid catalogid);
261 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
263 static void pgstat_setup_memcxt(void);
265 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
266 static void pgstat_send(void *msg, int len);
268 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
269 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
270 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
271 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
272 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
273 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
274 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
275 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
276 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
277 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
278 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
281 /* ------------------------------------------------------------
282 * Public functions called from postmaster follow
283 * ------------------------------------------------------------
289 * Called from postmaster at startup. Create the resources required
290 * by the statistics collector process. If unable to do so, do not
291 * fail --- better to let the postmaster start with stats collection
298 ACCEPT_TYPE_ARG3 alen;
299 struct addrinfo *addrs = NULL,
309 #define TESTBYTEVAL ((char) 199)
312 * Create the UDP socket for sending and receiving statistic messages
314 hints.ai_flags = AI_PASSIVE;
315 hints.ai_family = PF_UNSPEC;
316 hints.ai_socktype = SOCK_DGRAM;
317 hints.ai_protocol = 0;
318 hints.ai_addrlen = 0;
319 hints.ai_addr = NULL;
320 hints.ai_canonname = NULL;
321 hints.ai_next = NULL;
322 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
326 (errmsg("could not resolve \"localhost\": %s",
327 gai_strerror(ret))));
332 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
333 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
334 * when kernel will reject IPv6). Worse, the failure may occur at the
335 * bind() or perhaps even connect() stage. So we must loop through the
336 * results till we find a working combination. We will generate LOG
337 * messages, but no error, for bogus combinations.
339 for (addr = addrs; addr; addr = addr->ai_next)
341 #ifdef HAVE_UNIX_SOCKETS
342 /* Ignore AF_UNIX sockets, if any are returned. */
343 if (addr->ai_family == AF_UNIX)
349 (errmsg("trying another address for the statistics collector")));
354 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) < 0)
357 (errcode_for_socket_access(),
358 errmsg("could not create socket for statistics collector: %m")));
363 * Bind it to a kernel assigned port on localhost and get the assigned
364 * port via getsockname().
366 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
369 (errcode_for_socket_access(),
370 errmsg("could not bind socket for statistics collector: %m")));
371 closesocket(pgStatSock);
372 pgStatSock = PGINVALID_SOCKET;
376 alen = sizeof(pgStatAddr);
377 if (getsockname(pgStatSock, (struct sockaddr *) & pgStatAddr, &alen) < 0)
380 (errcode_for_socket_access(),
381 errmsg("could not get address of socket for statistics collector: %m")));
382 closesocket(pgStatSock);
383 pgStatSock = PGINVALID_SOCKET;
388 * Connect the socket to its own address. This saves a few cycles by
389 * not having to respecify the target address on every send. This also
390 * provides a kernel-level check that only packets from this same
391 * address will be received.
393 if (connect(pgStatSock, (struct sockaddr *) & pgStatAddr, alen) < 0)
396 (errcode_for_socket_access(),
397 errmsg("could not connect socket for statistics collector: %m")));
398 closesocket(pgStatSock);
399 pgStatSock = PGINVALID_SOCKET;
404 * Try to send and receive a one-byte test message on the socket. This
405 * is to catch situations where the socket can be created but will not
406 * actually pass data (for instance, because kernel packet filtering
409 test_byte = TESTBYTEVAL;
412 if (send(pgStatSock, &test_byte, 1, 0) != 1)
415 goto retry1; /* if interrupted, just retry */
417 (errcode_for_socket_access(),
418 errmsg("could not send test message on socket for statistics collector: %m")));
419 closesocket(pgStatSock);
420 pgStatSock = PGINVALID_SOCKET;
425 * There could possibly be a little delay before the message can be
426 * received. We arbitrarily allow up to half a second before deciding
429 for (;;) /* need a loop to handle EINTR */
432 FD_SET (pgStatSock, &rset);
436 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
437 if (sel_res >= 0 || errno != EINTR)
443 (errcode_for_socket_access(),
444 errmsg("select() failed in statistics collector: %m")));
445 closesocket(pgStatSock);
446 pgStatSock = PGINVALID_SOCKET;
449 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
452 * This is the case we actually think is likely, so take pains to
453 * give a specific message for it.
455 * errno will not be set meaningfully here, so don't use it.
458 (errcode(ERRCODE_CONNECTION_FAILURE),
459 errmsg("test message did not get through on socket for statistics collector")));
460 closesocket(pgStatSock);
461 pgStatSock = PGINVALID_SOCKET;
465 test_byte++; /* just make sure variable is changed */
468 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
471 goto retry2; /* if interrupted, just retry */
473 (errcode_for_socket_access(),
474 errmsg("could not receive test message on socket for statistics collector: %m")));
475 closesocket(pgStatSock);
476 pgStatSock = PGINVALID_SOCKET;
480 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
483 (errcode(ERRCODE_INTERNAL_ERROR),
484 errmsg("incorrect test message transmission on socket for statistics collector")));
485 closesocket(pgStatSock);
486 pgStatSock = PGINVALID_SOCKET;
490 /* If we get here, we have a working socket */
494 /* Did we find a working address? */
495 if (!addr || pgStatSock < 0)
499 * Set the socket to non-blocking IO. This ensures that if the collector
500 * falls behind, statistics messages will be discarded; backends won't
501 * block waiting to send messages to the collector.
503 if (!pg_set_noblock(pgStatSock))
506 (errcode_for_socket_access(),
507 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
511 pg_freeaddrinfo_all(hints.ai_family, addrs);
517 (errmsg("disabling statistics collector for lack of working socket")));
520 pg_freeaddrinfo_all(hints.ai_family, addrs);
523 closesocket(pgStatSock);
524 pgStatSock = PGINVALID_SOCKET;
527 * Adjust GUC variables to suppress useless activity, and for debugging
528 * purposes (seeing track_counts off is a clue that we failed here). We
529 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
530 * on from postgresql.conf without a restart.
532 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
536 * pgstat_reset_all() -
538 * Remove the stats file. This is currently used only if WAL
539 * recovery is needed after a crash.
542 pgstat_reset_all(void)
544 unlink(pgstat_stat_filename);
545 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
551 * pgstat_forkexec() -
553 * Format up the arglist for, then fork and exec, statistics collector process
556 pgstat_forkexec(void)
561 av[ac++] = "postgres";
562 av[ac++] = "--forkcol";
563 av[ac++] = NULL; /* filled in by postmaster_forkexec */
566 Assert(ac < lengthof(av));
568 return postmaster_forkexec(ac, av);
570 #endif /* EXEC_BACKEND */
576 * Called from postmaster at startup or after an existing collector
577 * died. Attempt to fire up a fresh statistics collector.
579 * Returns PID of child process, or 0 if fail.
581 * Note: if fail, we will be called again from the postmaster main loop.
590 * Check that the socket is there, else pgstat_init failed and we can do
597 * Do nothing if too soon since last collector start. This is a safety
598 * valve to protect against continuous respawn attempts if the collector
599 * is dying immediately at launch. Note that since we will be re-called
600 * from the postmaster main loop, we will get another chance later.
602 curtime = time(NULL);
603 if ((unsigned int) (curtime - last_pgstat_start_time) <
604 (unsigned int) PGSTAT_RESTART_INTERVAL)
606 last_pgstat_start_time = curtime;
609 * Okay, fork off the collector.
612 switch ((pgStatPid = pgstat_forkexec()))
614 switch ((pgStatPid = fork_process()))
619 (errmsg("could not fork statistics collector: %m")));
624 /* in postmaster child ... */
625 /* Close the postmaster's sockets */
626 ClosePostmasterPorts(false);
628 /* Lose the postmaster's on-exit routines */
631 /* Drop our connection to postmaster's shared memory, as well */
632 PGSharedMemoryDetach();
634 PgstatCollectorMain(0, NULL);
639 return (int) pgStatPid;
642 /* shouldn't get here */
647 allow_immediate_pgstat_restart(void)
649 last_pgstat_start_time = 0;
652 /* ------------------------------------------------------------
653 * Public functions used by backends follow
654 *------------------------------------------------------------
659 * pgstat_report_stat() -
661 * Called from tcop/postgres.c to send the so far collected per-table
662 * and function usage statistics to the collector. Note that this is
663 * called only when not within a transaction, so it is fair to use
664 * transaction stop time as an approximation of current time.
668 pgstat_report_stat(bool force)
670 /* we assume this inits to all zeroes: */
671 static const PgStat_TableCounts all_zeroes;
672 static TimestampTz last_report = 0;
675 PgStat_MsgTabstat regular_msg;
676 PgStat_MsgTabstat shared_msg;
680 /* Don't expend a clock check if nothing to do */
681 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0)
682 && !have_function_stats)
686 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
687 * msec since we last sent one, or the caller wants to force stats out.
689 now = GetCurrentTransactionStopTimestamp();
691 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
696 * Scan through the TabStatusArray struct(s) to find tables that actually
697 * have counts, and build messages to send. We have to separate shared
698 * relations from regular ones because the databaseid field in the message
699 * header has to depend on that.
701 regular_msg.m_databaseid = MyDatabaseId;
702 shared_msg.m_databaseid = InvalidOid;
703 regular_msg.m_nentries = 0;
704 shared_msg.m_nentries = 0;
706 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
708 for (i = 0; i < tsa->tsa_used; i++)
710 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
711 PgStat_MsgTabstat *this_msg;
712 PgStat_TableEntry *this_ent;
714 /* Shouldn't have any pending transaction-dependent counts */
715 Assert(entry->trans == NULL);
718 * Ignore entries that didn't accumulate any actual counts, such
719 * as indexes that were opened by the planner but not used.
721 if (memcmp(&entry->t_counts, &all_zeroes,
722 sizeof(PgStat_TableCounts)) == 0)
726 * OK, insert data into the appropriate message, and send if full.
728 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
729 this_ent = &this_msg->m_entry[this_msg->m_nentries];
730 this_ent->t_id = entry->t_id;
731 memcpy(&this_ent->t_counts, &entry->t_counts,
732 sizeof(PgStat_TableCounts));
733 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
735 pgstat_send_tabstat(this_msg);
736 this_msg->m_nentries = 0;
739 /* zero out TableStatus structs after use */
740 MemSet(tsa->tsa_entries, 0,
741 tsa->tsa_used * sizeof(PgStat_TableStatus));
746 * Send partial messages. If force is true, make sure that any pending
747 * xact commit/abort gets counted, even if no table stats to send.
749 if (regular_msg.m_nentries > 0 ||
750 (force && (pgStatXactCommit > 0 || pgStatXactRollback > 0)))
751 pgstat_send_tabstat(®ular_msg);
752 if (shared_msg.m_nentries > 0)
753 pgstat_send_tabstat(&shared_msg);
755 /* Now, send function statistics */
756 pgstat_send_funcstats();
760 * Subroutine for pgstat_report_stat: finish and send a tabstat message
763 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
768 /* It's unlikely we'd get here with no socket, but maybe not impossible */
773 * Report accumulated xact commit/rollback whenever we send a normal
776 if (OidIsValid(tsmsg->m_databaseid))
778 tsmsg->m_xact_commit = pgStatXactCommit;
779 tsmsg->m_xact_rollback = pgStatXactRollback;
780 pgStatXactCommit = 0;
781 pgStatXactRollback = 0;
785 tsmsg->m_xact_commit = 0;
786 tsmsg->m_xact_rollback = 0;
789 n = tsmsg->m_nentries;
790 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
791 n * sizeof(PgStat_TableEntry);
793 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
794 pgstat_send(tsmsg, len);
798 * Subroutine for pgstat_report_stat: populate and send a function stat message
801 pgstat_send_funcstats(void)
803 /* we assume this inits to all zeroes: */
804 static const PgStat_FunctionCounts all_zeroes;
806 PgStat_MsgFuncstat msg;
807 PgStat_BackendFunctionEntry *entry;
808 HASH_SEQ_STATUS fstat;
810 if (pgStatFunctions == NULL)
813 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
814 msg.m_databaseid = MyDatabaseId;
817 hash_seq_init(&fstat, pgStatFunctions);
818 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
820 PgStat_FunctionEntry *m_ent;
822 /* Skip it if no counts accumulated since last time */
823 if (memcmp(&entry->f_counts, &all_zeroes,
824 sizeof(PgStat_FunctionCounts)) == 0)
827 /* need to convert format of time accumulators */
828 m_ent = &msg.m_entry[msg.m_nentries];
829 m_ent->f_id = entry->f_id;
830 m_ent->f_numcalls = entry->f_counts.f_numcalls;
831 m_ent->f_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time);
832 m_ent->f_time_self = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time_self);
834 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
836 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
837 msg.m_nentries * sizeof(PgStat_FunctionEntry));
841 /* reset the entry's counts */
842 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
845 if (msg.m_nentries > 0)
846 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
847 msg.m_nentries * sizeof(PgStat_FunctionEntry));
849 have_function_stats = false;
854 * pgstat_vacuum_stat() -
856 * Will tell the collector about objects he can get rid of.
860 pgstat_vacuum_stat(void)
863 PgStat_MsgTabpurge msg;
864 PgStat_MsgFuncpurge f_msg;
865 HASH_SEQ_STATUS hstat;
866 PgStat_StatDBEntry *dbentry;
867 PgStat_StatTabEntry *tabentry;
868 PgStat_StatFuncEntry *funcentry;
875 * If not done for this transaction, read the statistics collector stats
876 * file into some hash tables.
878 backend_read_statsfile();
881 * Read pg_database and make a list of OIDs of all existing databases
883 htab = pgstat_collect_oids(DatabaseRelationId);
886 * Search the database hash table for dead databases and tell the
887 * collector to drop them.
889 hash_seq_init(&hstat, pgStatDBHash);
890 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
892 Oid dbid = dbentry->databaseid;
894 CHECK_FOR_INTERRUPTS();
896 /* the DB entry for shared tables (with InvalidOid) is never dropped */
897 if (OidIsValid(dbid) &&
898 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
899 pgstat_drop_database(dbid);
906 * Lookup our own database entry; if not found, nothing more to do.
908 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
909 (void *) &MyDatabaseId,
911 if (dbentry == NULL || dbentry->tables == NULL)
915 * Similarly to above, make a list of all known relations in this DB.
917 htab = pgstat_collect_oids(RelationRelationId);
920 * Initialize our messages table counter to zero
925 * Check for all tables listed in stats hashtable if they still exist.
927 hash_seq_init(&hstat, dbentry->tables);
928 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
930 Oid tabid = tabentry->tableid;
932 CHECK_FOR_INTERRUPTS();
934 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
938 * Not there, so add this table's Oid to the message
940 msg.m_tableid[msg.m_nentries++] = tabid;
943 * If the message is full, send it out and reinitialize to empty
945 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
947 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
948 +msg.m_nentries * sizeof(Oid);
950 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
951 msg.m_databaseid = MyDatabaseId;
952 pgstat_send(&msg, len);
961 if (msg.m_nentries > 0)
963 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
964 +msg.m_nentries * sizeof(Oid);
966 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
967 msg.m_databaseid = MyDatabaseId;
968 pgstat_send(&msg, len);
975 * Now repeat the above steps for functions. However, we needn't bother
976 * in the common case where no function stats are being collected.
978 if (dbentry->functions != NULL &&
979 hash_get_num_entries(dbentry->functions) > 0)
981 htab = pgstat_collect_oids(ProcedureRelationId);
983 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
984 f_msg.m_databaseid = MyDatabaseId;
985 f_msg.m_nentries = 0;
987 hash_seq_init(&hstat, dbentry->functions);
988 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
990 Oid funcid = funcentry->functionid;
992 CHECK_FOR_INTERRUPTS();
994 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
998 * Not there, so add this function's Oid to the message
1000 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
1003 * If the message is full, send it out and reinitialize to empty
1005 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1007 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1008 +f_msg.m_nentries * sizeof(Oid);
1010 pgstat_send(&f_msg, len);
1012 f_msg.m_nentries = 0;
1019 if (f_msg.m_nentries > 0)
1021 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1022 +f_msg.m_nentries * sizeof(Oid);
1024 pgstat_send(&f_msg, len);
1033 * pgstat_collect_oids() -
1035 * Collect the OIDs of all objects listed in the specified system catalog
1036 * into a temporary hash table. Caller should hash_destroy the result
1037 * when done with it. (However, we make the table in CurrentMemoryContext
1038 * so that it will be freed properly in event of an error.)
1042 pgstat_collect_oids(Oid catalogid)
1050 memset(&hash_ctl, 0, sizeof(hash_ctl));
1051 hash_ctl.keysize = sizeof(Oid);
1052 hash_ctl.entrysize = sizeof(Oid);
1053 hash_ctl.hash = oid_hash;
1054 hash_ctl.hcxt = CurrentMemoryContext;
1055 htab = hash_create("Temporary table of OIDs",
1056 PGSTAT_TAB_HASH_SIZE,
1058 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
1060 rel = heap_open(catalogid, AccessShareLock);
1061 scan = heap_beginscan(rel, SnapshotNow, 0, NULL);
1062 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1064 Oid thisoid = HeapTupleGetOid(tup);
1066 CHECK_FOR_INTERRUPTS();
1068 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1071 heap_close(rel, AccessShareLock);
1078 * pgstat_drop_database() -
1080 * Tell the collector that we just dropped a database.
1081 * (If the message gets lost, we will still clean the dead DB eventually
1082 * via future invocations of pgstat_vacuum_stat().)
1086 pgstat_drop_database(Oid databaseid)
1088 PgStat_MsgDropdb msg;
1093 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1094 msg.m_databaseid = databaseid;
1095 pgstat_send(&msg, sizeof(msg));
1100 * pgstat_drop_relation() -
1102 * Tell the collector that we just dropped a relation.
1103 * (If the message gets lost, we will still clean the dead entry eventually
1104 * via future invocations of pgstat_vacuum_stat().)
1106 * Currently not used for lack of any good place to call it; we rely
1107 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1112 pgstat_drop_relation(Oid relid)
1114 PgStat_MsgTabpurge msg;
1120 msg.m_tableid[0] = relid;
1123 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) +sizeof(Oid);
1125 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1126 msg.m_databaseid = MyDatabaseId;
1127 pgstat_send(&msg, len);
1129 #endif /* NOT_USED */
1133 * pgstat_reset_counters() -
1135 * Tell the statistics collector to reset counters for our database.
1139 pgstat_reset_counters(void)
1141 PgStat_MsgResetcounter msg;
1148 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1149 errmsg("must be superuser to reset statistics counters")));
1151 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1152 msg.m_databaseid = MyDatabaseId;
1153 pgstat_send(&msg, sizeof(msg));
1158 * pgstat_report_autovac() -
1160 * Called from autovacuum.c to report startup of an autovacuum process.
1161 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1162 * the db OID must be passed in, instead.
1166 pgstat_report_autovac(Oid dboid)
1168 PgStat_MsgAutovacStart msg;
1173 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1174 msg.m_databaseid = dboid;
1175 msg.m_start_time = GetCurrentTimestamp();
1177 pgstat_send(&msg, sizeof(msg));
1182 * pgstat_report_vacuum() -
1184 * Tell the collector about the table we just vacuumed.
1188 pgstat_report_vacuum(Oid tableoid, bool shared, bool adopt_counts,
1189 PgStat_Counter tuples)
1191 PgStat_MsgVacuum msg;
1193 if (pgStatSock < 0 || !pgstat_track_counts)
1196 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1197 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1198 msg.m_tableoid = tableoid;
1199 msg.m_adopt_counts = adopt_counts;
1200 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1201 msg.m_vacuumtime = GetCurrentTimestamp();
1202 msg.m_tuples = tuples;
1203 pgstat_send(&msg, sizeof(msg));
1207 * pgstat_report_analyze() -
1209 * Tell the collector about the table we just analyzed.
1213 pgstat_report_analyze(Relation rel, bool adopt_counts,
1214 PgStat_Counter livetuples, PgStat_Counter deadtuples)
1216 PgStat_MsgAnalyze msg;
1218 if (pgStatSock < 0 || !pgstat_track_counts)
1222 * Unlike VACUUM, ANALYZE might be running inside a transaction that has
1223 * already inserted and/or deleted rows in the target table. ANALYZE will
1224 * have counted such rows as live or dead respectively. Because we will
1225 * report our counts of such rows at transaction end, we should subtract
1226 * off these counts from what we send to the collector now, else they'll
1227 * be double-counted after commit. (This approach also ensures that the
1228 * collector ends up with the right numbers if we abort instead of
1231 if (rel->pgstat_info != NULL)
1233 PgStat_TableXactStatus *trans;
1235 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1237 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1238 deadtuples -= trans->tuples_updated + trans->tuples_deleted;
1240 /* count stuff inserted by already-aborted subxacts, too */
1241 deadtuples -= rel->pgstat_info->t_counts.t_delta_dead_tuples;
1242 /* Since ANALYZE's counts are estimates, we could have underflowed */
1243 livetuples = Max(livetuples, 0);
1244 deadtuples = Max(deadtuples, 0);
1247 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1248 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1249 msg.m_tableoid = RelationGetRelid(rel);
1250 msg.m_adopt_counts = adopt_counts;
1251 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1252 msg.m_analyzetime = GetCurrentTimestamp();
1253 msg.m_live_tuples = livetuples;
1254 msg.m_dead_tuples = deadtuples;
1255 pgstat_send(&msg, sizeof(msg));
1262 * Send some junk data to the collector to increase traffic.
1268 PgStat_MsgDummy msg;
1273 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1274 pgstat_send(&msg, sizeof(msg));
1278 * pgstat_send_inquiry() -
1280 * Notify collector that we need fresh data.
1281 * ts specifies the minimum acceptable timestamp for the stats file.
1285 pgstat_send_inquiry(TimestampTz ts)
1287 PgStat_MsgInquiry msg;
1289 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1290 msg.inquiry_time = ts;
1291 pgstat_send(&msg, sizeof(msg));
1296 * Initialize function call usage data.
1297 * Called by the executor before invoking a function.
1300 pgstat_init_function_usage(FunctionCallInfoData *fcinfo,
1301 PgStat_FunctionCallUsage *fcu)
1303 PgStat_BackendFunctionEntry *htabent;
1306 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1308 /* stats not wanted */
1313 if (!pgStatFunctions)
1315 /* First time through - initialize function stat table */
1318 memset(&hash_ctl, 0, sizeof(hash_ctl));
1319 hash_ctl.keysize = sizeof(Oid);
1320 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1321 hash_ctl.hash = oid_hash;
1322 pgStatFunctions = hash_create("Function stat entries",
1323 PGSTAT_FUNCTION_HASH_SIZE,
1325 HASH_ELEM | HASH_FUNCTION);
1328 /* Get the stats entry for this function, create if necessary */
1329 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1330 HASH_ENTER, &found);
1332 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1334 fcu->fs = &htabent->f_counts;
1336 /* save stats for this function, later used to compensate for recursion */
1337 fcu->save_f_time = htabent->f_counts.f_time;
1339 /* save current backend-wide total time */
1340 fcu->save_total = total_func_time;
1342 /* get clock time as of function start */
1343 INSTR_TIME_SET_CURRENT(fcu->f_start);
1347 * Calculate function call usage and update stat counters.
1348 * Called by the executor after invoking a function.
1350 * In the case of a set-returning function that runs in value-per-call mode,
1351 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1352 * calls for what the user considers a single call of the function. The
1353 * finalize flag should be TRUE on the last call.
1356 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1358 PgStat_FunctionCounts *fs = fcu->fs;
1360 instr_time f_others;
1363 /* stats not wanted? */
1367 /* total elapsed time in this function call */
1368 INSTR_TIME_SET_CURRENT(f_total);
1369 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1371 /* self usage: elapsed minus anything already charged to other calls */
1372 f_others = total_func_time;
1373 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1375 INSTR_TIME_SUBTRACT(f_self, f_others);
1377 /* update backend-wide total time */
1378 INSTR_TIME_ADD(total_func_time, f_self);
1381 * Compute the new total f_time as the total elapsed time added to the
1382 * pre-call value of f_time. This is necessary to avoid double-counting
1383 * any time taken by recursive calls of myself. (We do not need any
1384 * similar kluge for self time, since that already excludes any recursive
1387 INSTR_TIME_ADD(f_total, fcu->save_f_time);
1389 /* update counters in function stats table */
1392 fs->f_time = f_total;
1393 INSTR_TIME_ADD(fs->f_time_self, f_self);
1395 /* indicate that we have something to send */
1396 have_function_stats = true;
1401 * pgstat_initstats() -
1403 * Initialize a relcache entry to count access statistics.
1404 * Called whenever a relation is opened.
1406 * We assume that a relcache entry's pgstat_info field is zeroed by
1407 * relcache.c when the relcache entry is made; thereafter it is long-lived
1408 * data. We can avoid repeated searches of the TabStatus arrays when the
1409 * same relation is touched repeatedly within a transaction.
1413 pgstat_initstats(Relation rel)
1415 Oid rel_id = rel->rd_id;
1416 char relkind = rel->rd_rel->relkind;
1418 /* We only count stats for things that have storage */
1419 if (!(relkind == RELKIND_RELATION ||
1420 relkind == RELKIND_INDEX ||
1421 relkind == RELKIND_TOASTVALUE ||
1422 relkind == RELKIND_SEQUENCE))
1424 rel->pgstat_info = NULL;
1428 if (pgStatSock < 0 || !pgstat_track_counts)
1430 /* We're not counting at all */
1431 rel->pgstat_info = NULL;
1436 * If we already set up this relation in the current transaction, nothing
1439 if (rel->pgstat_info != NULL &&
1440 rel->pgstat_info->t_id == rel_id)
1443 /* Else find or make the PgStat_TableStatus entry, and update link */
1444 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1448 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1450 static PgStat_TableStatus *
1451 get_tabstat_entry(Oid rel_id, bool isshared)
1453 PgStat_TableStatus *entry;
1454 TabStatusArray *tsa;
1455 TabStatusArray *prev_tsa;
1459 * Search the already-used tabstat slots for this relation.
1462 for (tsa = pgStatTabList; tsa != NULL; prev_tsa = tsa, tsa = tsa->tsa_next)
1464 for (i = 0; i < tsa->tsa_used; i++)
1466 entry = &tsa->tsa_entries[i];
1467 if (entry->t_id == rel_id)
1471 if (tsa->tsa_used < TABSTAT_QUANTUM)
1474 * It must not be present, but we found a free slot instead. Fine,
1475 * let's use this one. We assume the entry was already zeroed,
1476 * either at creation or after last use.
1478 entry = &tsa->tsa_entries[tsa->tsa_used++];
1479 entry->t_id = rel_id;
1480 entry->t_shared = isshared;
1486 * We ran out of tabstat slots, so allocate more. Be sure they're zeroed.
1488 tsa = (TabStatusArray *) MemoryContextAllocZero(TopMemoryContext,
1489 sizeof(TabStatusArray));
1491 prev_tsa->tsa_next = tsa;
1493 pgStatTabList = tsa;
1496 * Use the first entry of the new TabStatusArray.
1498 entry = &tsa->tsa_entries[tsa->tsa_used++];
1499 entry->t_id = rel_id;
1500 entry->t_shared = isshared;
1505 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1507 static PgStat_SubXactStatus *
1508 get_tabstat_stack_level(int nest_level)
1510 PgStat_SubXactStatus *xact_state;
1512 xact_state = pgStatXactStack;
1513 if (xact_state == NULL || xact_state->nest_level != nest_level)
1515 xact_state = (PgStat_SubXactStatus *)
1516 MemoryContextAlloc(TopTransactionContext,
1517 sizeof(PgStat_SubXactStatus));
1518 xact_state->nest_level = nest_level;
1519 xact_state->prev = pgStatXactStack;
1520 xact_state->first = NULL;
1521 pgStatXactStack = xact_state;
1527 * add_tabstat_xact_level - add a new (sub)transaction state record
1530 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1532 PgStat_SubXactStatus *xact_state;
1533 PgStat_TableXactStatus *trans;
1536 * If this is the first rel to be modified at the current nest level, we
1537 * first have to push a transaction stack entry.
1539 xact_state = get_tabstat_stack_level(nest_level);
1541 /* Now make a per-table stack entry */
1542 trans = (PgStat_TableXactStatus *)
1543 MemoryContextAllocZero(TopTransactionContext,
1544 sizeof(PgStat_TableXactStatus));
1545 trans->nest_level = nest_level;
1546 trans->upper = pgstat_info->trans;
1547 trans->parent = pgstat_info;
1548 trans->next = xact_state->first;
1549 xact_state->first = trans;
1550 pgstat_info->trans = trans;
1554 * pgstat_count_heap_insert - count a tuple insertion
1557 pgstat_count_heap_insert(Relation rel)
1559 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1561 if (pgstat_track_counts && pgstat_info != NULL)
1563 /* We have to log the effect at the proper transactional level */
1564 int nest_level = GetCurrentTransactionNestLevel();
1566 if (pgstat_info->trans == NULL ||
1567 pgstat_info->trans->nest_level != nest_level)
1568 add_tabstat_xact_level(pgstat_info, nest_level);
1570 pgstat_info->trans->tuples_inserted++;
1575 * pgstat_count_heap_update - count a tuple update
1578 pgstat_count_heap_update(Relation rel, bool hot)
1580 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1582 if (pgstat_track_counts && pgstat_info != NULL)
1584 /* We have to log the effect at the proper transactional level */
1585 int nest_level = GetCurrentTransactionNestLevel();
1587 if (pgstat_info->trans == NULL ||
1588 pgstat_info->trans->nest_level != nest_level)
1589 add_tabstat_xact_level(pgstat_info, nest_level);
1591 pgstat_info->trans->tuples_updated++;
1593 /* t_tuples_hot_updated is nontransactional, so just advance it */
1595 pgstat_info->t_counts.t_tuples_hot_updated++;
1600 * pgstat_count_heap_delete - count a tuple deletion
1603 pgstat_count_heap_delete(Relation rel)
1605 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1607 if (pgstat_track_counts && pgstat_info != NULL)
1609 /* We have to log the effect at the proper transactional level */
1610 int nest_level = GetCurrentTransactionNestLevel();
1612 if (pgstat_info->trans == NULL ||
1613 pgstat_info->trans->nest_level != nest_level)
1614 add_tabstat_xact_level(pgstat_info, nest_level);
1616 pgstat_info->trans->tuples_deleted++;
1621 * pgstat_update_heap_dead_tuples - update dead-tuples count
1623 * The semantics of this are that we are reporting the nontransactional
1624 * recovery of "delta" dead tuples; so t_delta_dead_tuples decreases
1625 * rather than increasing, and the change goes straight into the per-table
1626 * counter, not into transactional state.
1629 pgstat_update_heap_dead_tuples(Relation rel, int delta)
1631 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1633 if (pgstat_track_counts && pgstat_info != NULL)
1634 pgstat_info->t_counts.t_delta_dead_tuples -= delta;
1641 * Called from access/transam/xact.c at top-level transaction commit/abort.
1645 AtEOXact_PgStat(bool isCommit)
1647 PgStat_SubXactStatus *xact_state;
1650 * Count transaction commit or abort. (We use counters, not just bools,
1651 * in case the reporting message isn't sent right away.)
1656 pgStatXactRollback++;
1659 * Transfer transactional insert/update counts into the base tabstat
1660 * entries. We don't bother to free any of the transactional state, since
1661 * it's all in TopTransactionContext and will go away anyway.
1663 xact_state = pgStatXactStack;
1664 if (xact_state != NULL)
1666 PgStat_TableXactStatus *trans;
1668 Assert(xact_state->nest_level == 1);
1669 Assert(xact_state->prev == NULL);
1670 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1672 PgStat_TableStatus *tabstat;
1674 Assert(trans->nest_level == 1);
1675 Assert(trans->upper == NULL);
1676 tabstat = trans->parent;
1677 Assert(tabstat->trans == trans);
1678 /* count attempted actions regardless of commit/abort */
1679 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
1680 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
1681 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
1684 /* insert adds a live tuple, delete removes one */
1685 tabstat->t_counts.t_delta_live_tuples +=
1686 trans->tuples_inserted - trans->tuples_deleted;
1687 /* update and delete each create a dead tuple */
1688 tabstat->t_counts.t_delta_dead_tuples +=
1689 trans->tuples_updated + trans->tuples_deleted;
1690 /* insert, update, delete each count as one change event */
1691 tabstat->t_counts.t_changed_tuples +=
1692 trans->tuples_inserted + trans->tuples_updated +
1693 trans->tuples_deleted;
1697 /* inserted tuples are dead, deleted tuples are unaffected */
1698 tabstat->t_counts.t_delta_dead_tuples +=
1699 trans->tuples_inserted + trans->tuples_updated;
1700 /* an aborted xact generates no changed_tuple events */
1702 tabstat->trans = NULL;
1705 pgStatXactStack = NULL;
1707 /* Make sure any stats snapshot is thrown away */
1708 pgstat_clear_snapshot();
1712 * AtEOSubXact_PgStat
1714 * Called from access/transam/xact.c at subtransaction commit/abort.
1718 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
1720 PgStat_SubXactStatus *xact_state;
1723 * Transfer transactional insert/update counts into the next higher
1724 * subtransaction state.
1726 xact_state = pgStatXactStack;
1727 if (xact_state != NULL &&
1728 xact_state->nest_level >= nestDepth)
1730 PgStat_TableXactStatus *trans;
1731 PgStat_TableXactStatus *next_trans;
1733 /* delink xact_state from stack immediately to simplify reuse case */
1734 pgStatXactStack = xact_state->prev;
1736 for (trans = xact_state->first; trans != NULL; trans = next_trans)
1738 PgStat_TableStatus *tabstat;
1740 next_trans = trans->next;
1741 Assert(trans->nest_level == nestDepth);
1742 tabstat = trans->parent;
1743 Assert(tabstat->trans == trans);
1746 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
1748 trans->upper->tuples_inserted += trans->tuples_inserted;
1749 trans->upper->tuples_updated += trans->tuples_updated;
1750 trans->upper->tuples_deleted += trans->tuples_deleted;
1751 tabstat->trans = trans->upper;
1757 * When there isn't an immediate parent state, we can just
1758 * reuse the record instead of going through a
1759 * palloc/pfree pushup (this works since it's all in
1760 * TopTransactionContext anyway). We have to re-link it
1761 * into the parent level, though, and that might mean
1762 * pushing a new entry into the pgStatXactStack.
1764 PgStat_SubXactStatus *upper_xact_state;
1766 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
1767 trans->next = upper_xact_state->first;
1768 upper_xact_state->first = trans;
1769 trans->nest_level = nestDepth - 1;
1775 * On abort, update top-level tabstat counts, then forget
1776 * the subtransaction
1779 /* count attempted actions regardless of commit/abort */
1780 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
1781 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
1782 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
1783 /* inserted tuples are dead, deleted tuples are unaffected */
1784 tabstat->t_counts.t_delta_dead_tuples +=
1785 trans->tuples_inserted + trans->tuples_updated;
1786 tabstat->trans = trans->upper;
1797 * Save the transactional stats state at 2PC transaction prepare.
1799 * In this phase we just generate 2PC records for all the pending
1800 * transaction-dependent stats work.
1803 AtPrepare_PgStat(void)
1805 PgStat_SubXactStatus *xact_state;
1807 xact_state = pgStatXactStack;
1808 if (xact_state != NULL)
1810 PgStat_TableXactStatus *trans;
1812 Assert(xact_state->nest_level == 1);
1813 Assert(xact_state->prev == NULL);
1814 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1816 PgStat_TableStatus *tabstat;
1817 TwoPhasePgStatRecord record;
1819 Assert(trans->nest_level == 1);
1820 Assert(trans->upper == NULL);
1821 tabstat = trans->parent;
1822 Assert(tabstat->trans == trans);
1824 record.tuples_inserted = trans->tuples_inserted;
1825 record.tuples_updated = trans->tuples_updated;
1826 record.tuples_deleted = trans->tuples_deleted;
1827 record.t_id = tabstat->t_id;
1828 record.t_shared = tabstat->t_shared;
1830 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
1831 &record, sizeof(TwoPhasePgStatRecord));
1837 * PostPrepare_PgStat
1838 * Clean up after successful PREPARE.
1840 * All we need do here is unlink the transaction stats state from the
1841 * nontransactional state. The nontransactional action counts will be
1842 * reported to the stats collector immediately, while the effects on live
1843 * and dead tuple counts are preserved in the 2PC state file.
1845 * Note: AtEOXact_PgStat is not called during PREPARE.
1848 PostPrepare_PgStat(void)
1850 PgStat_SubXactStatus *xact_state;
1853 * We don't bother to free any of the transactional state, since it's all
1854 * in TopTransactionContext and will go away anyway.
1856 xact_state = pgStatXactStack;
1857 if (xact_state != NULL)
1859 PgStat_TableXactStatus *trans;
1861 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1863 PgStat_TableStatus *tabstat;
1865 tabstat = trans->parent;
1866 tabstat->trans = NULL;
1869 pgStatXactStack = NULL;
1871 /* Make sure any stats snapshot is thrown away */
1872 pgstat_clear_snapshot();
1876 * 2PC processing routine for COMMIT PREPARED case.
1878 * Load the saved counts into our local pgstats state.
1881 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
1882 void *recdata, uint32 len)
1884 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
1885 PgStat_TableStatus *pgstat_info;
1887 /* Find or create a tabstat entry for the rel */
1888 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
1890 /* Same math as in AtEOXact_PgStat, commit case */
1891 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
1892 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
1893 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
1894 pgstat_info->t_counts.t_delta_live_tuples +=
1895 rec->tuples_inserted - rec->tuples_deleted;
1896 pgstat_info->t_counts.t_delta_dead_tuples +=
1897 rec->tuples_updated + rec->tuples_deleted;
1898 pgstat_info->t_counts.t_changed_tuples +=
1899 rec->tuples_inserted + rec->tuples_updated +
1900 rec->tuples_deleted;
1904 * 2PC processing routine for ROLLBACK PREPARED case.
1906 * Load the saved counts into our local pgstats state, but treat them
1910 pgstat_twophase_postabort(TransactionId xid, uint16 info,
1911 void *recdata, uint32 len)
1913 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
1914 PgStat_TableStatus *pgstat_info;
1916 /* Find or create a tabstat entry for the rel */
1917 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
1919 /* Same math as in AtEOXact_PgStat, abort case */
1920 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
1921 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
1922 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
1923 pgstat_info->t_counts.t_delta_dead_tuples +=
1924 rec->tuples_inserted + rec->tuples_updated;
1929 * pgstat_fetch_stat_dbentry() -
1931 * Support function for the SQL-callable pgstat* functions. Returns
1932 * the collected statistics for one database or NULL. NULL doesn't mean
1933 * that the database doesn't exist, it is just not yet known by the
1934 * collector, so the caller is better off to report ZERO instead.
1937 PgStat_StatDBEntry *
1938 pgstat_fetch_stat_dbentry(Oid dbid)
1941 * If not done for this transaction, read the statistics collector stats
1942 * file into some hash tables.
1944 backend_read_statsfile();
1947 * Lookup the requested database; return NULL if not found
1949 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1956 * pgstat_fetch_stat_tabentry() -
1958 * Support function for the SQL-callable pgstat* functions. Returns
1959 * the collected statistics for one table or NULL. NULL doesn't mean
1960 * that the table doesn't exist, it is just not yet known by the
1961 * collector, so the caller is better off to report ZERO instead.
1964 PgStat_StatTabEntry *
1965 pgstat_fetch_stat_tabentry(Oid relid)
1968 PgStat_StatDBEntry *dbentry;
1969 PgStat_StatTabEntry *tabentry;
1972 * If not done for this transaction, read the statistics collector stats
1973 * file into some hash tables.
1975 backend_read_statsfile();
1978 * Lookup our database, then look in its table hash table.
1980 dbid = MyDatabaseId;
1981 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1984 if (dbentry != NULL && dbentry->tables != NULL)
1986 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
1994 * If we didn't find it, maybe it's a shared table.
1997 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2000 if (dbentry != NULL && dbentry->tables != NULL)
2002 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2014 * pgstat_fetch_stat_funcentry() -
2016 * Support function for the SQL-callable pgstat* functions. Returns
2017 * the collected statistics for one function or NULL.
2020 PgStat_StatFuncEntry *
2021 pgstat_fetch_stat_funcentry(Oid func_id)
2023 PgStat_StatDBEntry *dbentry;
2024 PgStat_StatFuncEntry *funcentry = NULL;
2026 /* load the stats file if needed */
2027 backend_read_statsfile();
2029 /* Lookup our database, then find the requested function. */
2030 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
2031 if (dbentry != NULL && dbentry->functions != NULL)
2033 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2043 * pgstat_fetch_stat_beentry() -
2045 * Support function for the SQL-callable pgstat* functions. Returns
2046 * our local copy of the current-activity entry for one backend.
2048 * NB: caller is responsible for a check if the user is permitted to see
2049 * this info (especially the querystring).
2053 pgstat_fetch_stat_beentry(int beid)
2055 pgstat_read_current_status();
2057 if (beid < 1 || beid > localNumBackends)
2060 return &localBackendStatusTable[beid - 1];
2065 * pgstat_fetch_stat_numbackends() -
2067 * Support function for the SQL-callable pgstat* functions. Returns
2068 * the maximum current backend id.
2072 pgstat_fetch_stat_numbackends(void)
2074 pgstat_read_current_status();
2076 return localNumBackends;
2081 * pgstat_fetch_global() -
2083 * Support function for the SQL-callable pgstat* functions. Returns
2084 * a pointer to the global statistics struct.
2087 PgStat_GlobalStats *
2088 pgstat_fetch_global(void)
2090 backend_read_statsfile();
2092 return &globalStats;
2096 /* ------------------------------------------------------------
2097 * Functions for management of the shared-memory PgBackendStatus array
2098 * ------------------------------------------------------------
2101 static PgBackendStatus *BackendStatusArray = NULL;
2102 static PgBackendStatus *MyBEEntry = NULL;
2103 static char *BackendAppnameBuffer = NULL;
2104 static char *BackendActivityBuffer = NULL;
2108 * Report shared-memory space needed by CreateSharedBackendStatus.
2111 BackendStatusShmemSize(void)
2115 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2116 size = add_size(size,
2117 mul_size(NAMEDATALEN, MaxBackends));
2118 size = add_size(size,
2119 mul_size(pgstat_track_activity_query_size, MaxBackends));
2124 * Initialize the shared status array and activity/appname string buffers
2125 * during postmaster startup.
2128 CreateSharedBackendStatus(void)
2135 /* Create or attach to the shared array */
2136 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2137 BackendStatusArray = (PgBackendStatus *)
2138 ShmemInitStruct("Backend Status Array", size, &found);
2143 * We're the first - initialize.
2145 MemSet(BackendStatusArray, 0, size);
2148 /* Create or attach to the shared appname buffer */
2149 size = mul_size(NAMEDATALEN, MaxBackends);
2150 BackendAppnameBuffer = (char *)
2151 ShmemInitStruct("Backend Application Name Buffer", size, &found);
2155 MemSet(BackendAppnameBuffer, 0, size);
2157 /* Initialize st_appname pointers. */
2158 buffer = BackendAppnameBuffer;
2159 for (i = 0; i < MaxBackends; i++)
2161 BackendStatusArray[i].st_appname = buffer;
2162 buffer += NAMEDATALEN;
2166 /* Create or attach to the shared activity buffer */
2167 size = mul_size(pgstat_track_activity_query_size, MaxBackends);
2168 BackendActivityBuffer = (char *)
2169 ShmemInitStruct("Backend Activity Buffer", size, &found);
2173 MemSet(BackendActivityBuffer, 0, size);
2175 /* Initialize st_activity pointers. */
2176 buffer = BackendActivityBuffer;
2177 for (i = 0; i < MaxBackends; i++)
2179 BackendStatusArray[i].st_activity = buffer;
2180 buffer += pgstat_track_activity_query_size;
2187 * pgstat_initialize() -
2189 * Initialize pgstats state, and set up our on-proc-exit hook.
2190 * Called from InitPostgres. MyBackendId must be set,
2191 * but we must not have started any transaction yet (since the
2192 * exit hook must run after the last transaction exit).
2193 * NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful.
2197 pgstat_initialize(void)
2199 /* Initialize MyBEEntry */
2200 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2201 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2203 /* Set up a process-exit hook to clean up */
2204 on_shmem_exit(pgstat_beshutdown_hook, 0);
2208 * pgstat_bestart() -
2210 * Initialize this backend's entry in the PgBackendStatus array.
2211 * Called from InitPostgres.
2212 * MyDatabaseId, session userid, and application_name must be set
2213 * (hence, this cannot be combined with pgstat_initialize).
2217 pgstat_bestart(void)
2219 TimestampTz proc_start_timestamp;
2221 SockAddr clientaddr;
2222 volatile PgBackendStatus *beentry;
2225 * To minimize the time spent modifying the PgBackendStatus entry, fetch
2226 * all the needed data first.
2228 * If we have a MyProcPort, use its session start time (for consistency,
2229 * and to save a kernel call).
2232 proc_start_timestamp = MyProcPort->SessionStartTime;
2234 proc_start_timestamp = GetCurrentTimestamp();
2235 userid = GetSessionUserId();
2238 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2239 * If so, use all-zeroes client address, which is dealt with specially in
2240 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2243 memcpy(&clientaddr, &MyProcPort->raddr, sizeof(clientaddr));
2245 MemSet(&clientaddr, 0, sizeof(clientaddr));
2248 * Initialize my status entry, following the protocol of bumping
2249 * st_changecount before and after; and make sure it's even afterwards. We
2250 * use a volatile pointer here to ensure the compiler doesn't try to get
2253 beentry = MyBEEntry;
2256 beentry->st_changecount++;
2257 } while ((beentry->st_changecount & 1) == 0);
2259 beentry->st_procpid = MyProcPid;
2260 beentry->st_proc_start_timestamp = proc_start_timestamp;
2261 beentry->st_activity_start_timestamp = 0;
2262 beentry->st_xact_start_timestamp = 0;
2263 beentry->st_databaseid = MyDatabaseId;
2264 beentry->st_userid = userid;
2265 beentry->st_clientaddr = clientaddr;
2266 beentry->st_waiting = false;
2267 beentry->st_appname[0] = '\0';
2268 beentry->st_activity[0] = '\0';
2269 /* Also make sure the last byte in each string area is always 0 */
2270 beentry->st_appname[NAMEDATALEN - 1] = '\0';
2271 beentry->st_activity[pgstat_track_activity_query_size - 1] = '\0';
2273 beentry->st_changecount++;
2274 Assert((beentry->st_changecount & 1) == 0);
2276 /* Update app name to current GUC setting */
2277 if (application_name)
2278 pgstat_report_appname(application_name);
2282 * Shut down a single backend's statistics reporting at process exit.
2284 * Flush any remaining statistics counts out to the collector.
2285 * Without this, operations triggered during backend exit (such as
2286 * temp table deletions) won't be counted.
2288 * Lastly, clear out our entry in the PgBackendStatus array.
2291 pgstat_beshutdown_hook(int code, Datum arg)
2293 volatile PgBackendStatus *beentry = MyBEEntry;
2296 * If we got as far as discovering our own database ID, we can report
2297 * what we did to the collector. Otherwise, we'd be sending an invalid
2298 * database ID, so forget it. (This means that accesses to pg_database
2299 * during failed backend starts might never get counted.)
2301 if (OidIsValid(MyDatabaseId))
2302 pgstat_report_stat(true);
2305 * Clear my status entry, following the protocol of bumping st_changecount
2306 * before and after. We use a volatile pointer here to ensure the
2307 * compiler doesn't try to get cute.
2309 beentry->st_changecount++;
2311 beentry->st_procpid = 0; /* mark invalid */
2313 beentry->st_changecount++;
2314 Assert((beentry->st_changecount & 1) == 0);
2319 * pgstat_report_activity() -
2321 * Called from tcop/postgres.c to report what the backend is actually doing
2322 * (usually "<IDLE>" or the start of the query to be executed).
2326 pgstat_report_activity(const char *cmd_str)
2328 volatile PgBackendStatus *beentry = MyBEEntry;
2329 TimestampTz start_timestamp;
2332 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
2334 if (!pgstat_track_activities || !beentry)
2338 * To minimize the time spent modifying the entry, fetch all the needed
2341 start_timestamp = GetCurrentStatementStartTimestamp();
2343 len = strlen(cmd_str);
2344 len = pg_mbcliplen(cmd_str, len, pgstat_track_activity_query_size - 1);
2347 * Update my status entry, following the protocol of bumping
2348 * st_changecount before and after. We use a volatile pointer here to
2349 * ensure the compiler doesn't try to get cute.
2351 beentry->st_changecount++;
2353 beentry->st_activity_start_timestamp = start_timestamp;
2354 memcpy((char *) beentry->st_activity, cmd_str, len);
2355 beentry->st_activity[len] = '\0';
2357 beentry->st_changecount++;
2358 Assert((beentry->st_changecount & 1) == 0);
2362 * pgstat_report_appname() -
2364 * Called to update our application name.
2368 pgstat_report_appname(const char *appname)
2370 volatile PgBackendStatus *beentry = MyBEEntry;
2376 /* This should be unnecessary if GUC did its job, but be safe */
2377 len = pg_mbcliplen(appname, strlen(appname), NAMEDATALEN - 1);
2380 * Update my status entry, following the protocol of bumping
2381 * st_changecount before and after. We use a volatile pointer here to
2382 * ensure the compiler doesn't try to get cute.
2384 beentry->st_changecount++;
2386 memcpy((char *) beentry->st_appname, appname, len);
2387 beentry->st_appname[len] = '\0';
2389 beentry->st_changecount++;
2390 Assert((beentry->st_changecount & 1) == 0);
2394 * Report current transaction start timestamp as the specified value.
2395 * Zero means there is no active transaction.
2398 pgstat_report_xact_timestamp(TimestampTz tstamp)
2400 volatile PgBackendStatus *beentry = MyBEEntry;
2402 if (!pgstat_track_activities || !beentry)
2406 * Update my status entry, following the protocol of bumping
2407 * st_changecount before and after. We use a volatile pointer here to
2408 * ensure the compiler doesn't try to get cute.
2410 beentry->st_changecount++;
2411 beentry->st_xact_start_timestamp = tstamp;
2412 beentry->st_changecount++;
2413 Assert((beentry->st_changecount & 1) == 0);
2417 * pgstat_report_waiting() -
2419 * Called from lock manager to report beginning or end of a lock wait.
2421 * NB: this *must* be able to survive being called before MyBEEntry has been
2426 pgstat_report_waiting(bool waiting)
2428 volatile PgBackendStatus *beentry = MyBEEntry;
2430 if (!pgstat_track_activities || !beentry)
2434 * Since this is a single-byte field in a struct that only this process
2435 * may modify, there seems no need to bother with the st_changecount
2436 * protocol. The update must appear atomic in any case.
2438 beentry->st_waiting = waiting;
2443 * pgstat_read_current_status() -
2445 * Copy the current contents of the PgBackendStatus array to local memory,
2446 * if not already done in this transaction.
2450 pgstat_read_current_status(void)
2452 volatile PgBackendStatus *beentry;
2453 PgBackendStatus *localtable;
2454 PgBackendStatus *localentry;
2459 Assert(!pgStatRunningInCollector);
2460 if (localBackendStatusTable)
2461 return; /* already done */
2463 pgstat_setup_memcxt();
2465 localtable = (PgBackendStatus *)
2466 MemoryContextAlloc(pgStatLocalContext,
2467 sizeof(PgBackendStatus) * MaxBackends);
2468 localappname = (char *)
2469 MemoryContextAlloc(pgStatLocalContext,
2470 NAMEDATALEN * MaxBackends);
2471 localactivity = (char *)
2472 MemoryContextAlloc(pgStatLocalContext,
2473 pgstat_track_activity_query_size * MaxBackends);
2474 localNumBackends = 0;
2476 beentry = BackendStatusArray;
2477 localentry = localtable;
2478 for (i = 1; i <= MaxBackends; i++)
2481 * Follow the protocol of retrying if st_changecount changes while we
2482 * copy the entry, or if it's odd. (The check for odd is needed to
2483 * cover the case where we are able to completely copy the entry while
2484 * the source backend is between increment steps.) We use a volatile
2485 * pointer here to ensure the compiler doesn't try to get cute.
2489 int save_changecount = beentry->st_changecount;
2491 localentry->st_procpid = beentry->st_procpid;
2492 if (localentry->st_procpid > 0)
2494 memcpy(localentry, (char *) beentry, sizeof(PgBackendStatus));
2497 * strcpy is safe even if the string is modified concurrently,
2498 * because there's always a \0 at the end of the buffer.
2500 strcpy(localappname, (char *) beentry->st_appname);
2501 localentry->st_appname = localappname;
2502 strcpy(localactivity, (char *) beentry->st_activity);
2503 localentry->st_activity = localactivity;
2506 if (save_changecount == beentry->st_changecount &&
2507 (save_changecount & 1) == 0)
2510 /* Make sure we can break out of loop if stuck... */
2511 CHECK_FOR_INTERRUPTS();
2515 /* Only valid entries get included into the local array */
2516 if (localentry->st_procpid > 0)
2519 localappname += NAMEDATALEN;
2520 localactivity += pgstat_track_activity_query_size;
2525 /* Set the pointer only after completion of a valid table */
2526 localBackendStatusTable = localtable;
2531 * pgstat_get_backend_current_activity() -
2533 * Return a string representing the current activity of the backend with
2534 * the specified PID. This looks directly at the BackendStatusArray,
2535 * and so will provide current information regardless of the age of our
2536 * transaction's snapshot of the status array.
2538 * It is the caller's responsibility to invoke this only for backends whose
2539 * state is expected to remain stable while the result is in use. The
2540 * only current use is in deadlock reporting, where we can expect that
2541 * the target backend is blocked on a lock. (There are corner cases
2542 * where the target's wait could get aborted while we are looking at it,
2543 * but the very worst consequence is to return a pointer to a string
2544 * that's been changed, so we won't worry too much.)
2546 * Note: return strings for special cases match pg_stat_get_backend_activity.
2550 pgstat_get_backend_current_activity(int pid, bool checkUser)
2552 PgBackendStatus *beentry;
2555 beentry = BackendStatusArray;
2556 for (i = 1; i <= MaxBackends; i++)
2559 * Although we expect the target backend's entry to be stable, that
2560 * doesn't imply that anyone else's is. To avoid identifying the
2561 * wrong backend, while we check for a match to the desired PID we
2562 * must follow the protocol of retrying if st_changecount changes
2563 * while we examine the entry, or if it's odd. (This might be
2564 * unnecessary, since fetching or storing an int is almost certainly
2565 * atomic, but let's play it safe.) We use a volatile pointer here to
2566 * ensure the compiler doesn't try to get cute.
2568 volatile PgBackendStatus *vbeentry = beentry;
2573 int save_changecount = vbeentry->st_changecount;
2575 found = (vbeentry->st_procpid == pid);
2577 if (save_changecount == vbeentry->st_changecount &&
2578 (save_changecount & 1) == 0)
2581 /* Make sure we can break out of loop if stuck... */
2582 CHECK_FOR_INTERRUPTS();
2587 /* Now it is safe to use the non-volatile pointer */
2588 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
2589 return "<insufficient privilege>";
2590 else if (*(beentry->st_activity) == '\0')
2591 return "<command string not enabled>";
2593 return beentry->st_activity;
2599 /* If we get here, caller is in error ... */
2600 return "<backend information not available>";
2604 /* ------------------------------------------------------------
2605 * Local support functions follow
2606 * ------------------------------------------------------------
2611 * pgstat_setheader() -
2613 * Set common header fields in a statistics message
2617 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
2619 hdr->m_type = mtype;
2626 * Send out one statistics message to the collector
2630 pgstat_send(void *msg, int len)
2637 ((PgStat_MsgHdr *) msg)->m_size = len;
2639 /* We'll retry after EINTR, but ignore all other failures */
2642 rc = send(pgStatSock, msg, len, 0);
2643 } while (rc < 0 && errno == EINTR);
2645 #ifdef USE_ASSERT_CHECKING
2646 /* In debug builds, log send failures ... */
2648 elog(LOG, "could not send to statistics collector: %m");
2653 * pgstat_send_bgwriter() -
2655 * Send bgwriter statistics to the collector
2659 pgstat_send_bgwriter(void)
2661 /* We assume this initializes to zeroes */
2662 static const PgStat_MsgBgWriter all_zeroes;
2665 * This function can be called even if nothing at all has happened. In
2666 * this case, avoid sending a completely empty message to the stats
2669 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
2673 * Prepare and send the message
2675 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
2676 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
2679 * Clear out the statistics buffer, so it can be re-used.
2681 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
2686 * PgstatCollectorMain() -
2688 * Start up the statistics collector process. This is the body of the
2689 * postmaster child process.
2691 * The argc/argv parameters are valid only in EXEC_BACKEND case.
2694 NON_EXEC_STATIC void
2695 PgstatCollectorMain(int argc, char *argv[])
2702 struct pollfd input_fd;
2704 struct timeval sel_timeout;
2709 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2711 MyProcPid = getpid(); /* reset MyProcPid */
2713 MyStartTime = time(NULL); /* record Start Time for logging */
2716 * If possible, make this process a group leader, so that the postmaster
2717 * can signal any child processes too. (pgstat probably never has any
2718 * child processes, but for consistency we make all postmaster child
2719 * processes do this.)
2723 elog(FATAL, "setsid() failed: %m");
2727 * Ignore all signals usually bound to some action in the postmaster,
2730 pqsignal(SIGHUP, pgstat_sighup_handler);
2731 pqsignal(SIGINT, SIG_IGN);
2732 pqsignal(SIGTERM, SIG_IGN);
2733 pqsignal(SIGQUIT, pgstat_exit);
2734 pqsignal(SIGALRM, SIG_IGN);
2735 pqsignal(SIGPIPE, SIG_IGN);
2736 pqsignal(SIGUSR1, SIG_IGN);
2737 pqsignal(SIGUSR2, SIG_IGN);
2738 pqsignal(SIGCHLD, SIG_DFL);
2739 pqsignal(SIGTTIN, SIG_DFL);
2740 pqsignal(SIGTTOU, SIG_DFL);
2741 pqsignal(SIGCONT, SIG_DFL);
2742 pqsignal(SIGWINCH, SIG_DFL);
2743 PG_SETMASK(&UnBlockSig);
2746 * Identify myself via ps
2748 init_ps_display("stats collector process", "", "", "");
2751 * Arrange to write the initial status file right away
2753 last_statrequest = GetCurrentTimestamp();
2754 last_statwrite = last_statrequest - 1;
2757 * Read in an existing statistics stats file or initialize the stats to
2760 pgStatRunningInCollector = true;
2761 pgStatDBHash = pgstat_read_statsfile(InvalidOid, true);
2764 * Setup the descriptor set for select(2). Since only one bit in the set
2765 * ever changes, we need not repeat FD_ZERO each time.
2767 #if !defined(HAVE_POLL) && !defined(WIN32)
2772 * Loop to process messages until we get SIGQUIT or detect ungraceful
2773 * death of our parent postmaster.
2775 * For performance reasons, we don't want to do a PostmasterIsAlive() test
2776 * after every message; instead, do it only when select()/poll() is
2777 * interrupted by timeout. In essence, we'll stay alive as long as
2778 * backends keep sending us stuff often, even if the postmaster is gone.
2785 * Quit if we get SIGQUIT from the postmaster.
2791 * Reload configuration if we got SIGHUP from the postmaster.
2795 ProcessConfigFile(PGC_SIGHUP);
2800 * Write the stats file if a new request has arrived that is not
2801 * satisfied by existing file.
2803 if (last_statwrite < last_statrequest)
2804 pgstat_write_statsfile(false);
2807 * Wait for a message to arrive; but not for more than
2808 * PGSTAT_SELECT_TIMEOUT seconds. (This determines how quickly we will
2809 * shut down after an ungraceful postmaster termination; so it needn't
2810 * be very fast. However, on some systems SIGQUIT won't interrupt the
2811 * poll/select call, so this also limits speed of response to SIGQUIT,
2812 * which is more important.)
2814 * We use poll(2) if available, otherwise select(2). Win32 has its own
2819 input_fd.fd = pgStatSock;
2820 input_fd.events = POLLIN | POLLERR;
2821 input_fd.revents = 0;
2823 if (poll(&input_fd, 1, PGSTAT_SELECT_TIMEOUT * 1000) < 0)
2828 (errcode_for_socket_access(),
2829 errmsg("poll() failed in statistics collector: %m")));
2832 got_data = (input_fd.revents != 0);
2833 #else /* !HAVE_POLL */
2835 FD_SET (pgStatSock, &rfds);
2838 * timeout struct is modified by select() on some operating systems,
2839 * so re-fill it each time.
2841 sel_timeout.tv_sec = PGSTAT_SELECT_TIMEOUT;
2842 sel_timeout.tv_usec = 0;
2844 if (select(pgStatSock + 1, &rfds, NULL, NULL, &sel_timeout) < 0)
2849 (errcode_for_socket_access(),
2850 errmsg("select() failed in statistics collector: %m")));
2853 got_data = FD_ISSET(pgStatSock, &rfds);
2854 #endif /* HAVE_POLL */
2856 got_data = pgwin32_waitforsinglesocket(pgStatSock, FD_READ,
2857 PGSTAT_SELECT_TIMEOUT * 1000);
2861 * If there is a message on the socket, read it and check for
2866 len = recv(pgStatSock, (char *) &msg,
2867 sizeof(PgStat_Msg), 0);
2873 (errcode_for_socket_access(),
2874 errmsg("could not read statistics message: %m")));
2878 * We ignore messages that are smaller than our common header
2880 if (len < sizeof(PgStat_MsgHdr))
2884 * The received length must match the length in the header
2886 if (msg.msg_hdr.m_size != len)
2890 * O.K. - we accept this message. Process it.
2892 switch (msg.msg_hdr.m_type)
2894 case PGSTAT_MTYPE_DUMMY:
2897 case PGSTAT_MTYPE_INQUIRY:
2898 pgstat_recv_inquiry((PgStat_MsgInquiry *) &msg, len);
2901 case PGSTAT_MTYPE_TABSTAT:
2902 pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len);
2905 case PGSTAT_MTYPE_TABPURGE:
2906 pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len);
2909 case PGSTAT_MTYPE_DROPDB:
2910 pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len);
2913 case PGSTAT_MTYPE_RESETCOUNTER:
2914 pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg,
2918 case PGSTAT_MTYPE_AUTOVAC_START:
2919 pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
2922 case PGSTAT_MTYPE_VACUUM:
2923 pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
2926 case PGSTAT_MTYPE_ANALYZE:
2927 pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len);
2930 case PGSTAT_MTYPE_BGWRITER:
2931 pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len);
2934 case PGSTAT_MTYPE_FUNCSTAT:
2935 pgstat_recv_funcstat((PgStat_MsgFuncstat *) &msg, len);
2938 case PGSTAT_MTYPE_FUNCPURGE:
2939 pgstat_recv_funcpurge((PgStat_MsgFuncpurge *) &msg, len);
2949 * We can only get here if the select/poll timeout elapsed. Check
2950 * for postmaster death.
2952 if (!PostmasterIsAlive(true))
2955 } /* end of message-processing loop */
2958 * Save the final stats to reuse at next startup.
2960 pgstat_write_statsfile(true);
2966 /* SIGQUIT signal handler for collector process */
2968 pgstat_exit(SIGNAL_ARGS)
2973 /* SIGHUP handler for collector process */
2975 pgstat_sighup_handler(SIGNAL_ARGS)
2982 * Lookup the hash table entry for the specified database. If no hash
2983 * table entry exists, initialize it, if the create parameter is true.
2984 * Else, return NULL.
2986 static PgStat_StatDBEntry *
2987 pgstat_get_db_entry(Oid databaseid, bool create)
2989 PgStat_StatDBEntry *result;
2991 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
2993 /* Lookup or create the hash table entry for this database */
2994 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2998 if (!create && !found)
3001 /* If not found, initialize the new one. */
3006 result->tables = NULL;
3007 result->functions = NULL;
3008 result->n_xact_commit = 0;
3009 result->n_xact_rollback = 0;
3010 result->n_blocks_fetched = 0;
3011 result->n_blocks_hit = 0;
3012 result->n_tuples_returned = 0;
3013 result->n_tuples_fetched = 0;
3014 result->n_tuples_inserted = 0;
3015 result->n_tuples_updated = 0;
3016 result->n_tuples_deleted = 0;
3017 result->last_autovac_time = 0;
3019 memset(&hash_ctl, 0, sizeof(hash_ctl));
3020 hash_ctl.keysize = sizeof(Oid);
3021 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3022 hash_ctl.hash = oid_hash;
3023 result->tables = hash_create("Per-database table",
3024 PGSTAT_TAB_HASH_SIZE,
3026 HASH_ELEM | HASH_FUNCTION);
3028 hash_ctl.keysize = sizeof(Oid);
3029 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3030 hash_ctl.hash = oid_hash;
3031 result->functions = hash_create("Per-database function",
3032 PGSTAT_FUNCTION_HASH_SIZE,
3034 HASH_ELEM | HASH_FUNCTION);
3042 * Lookup the hash table entry for the specified table. If no hash
3043 * table entry exists, initialize it, if the create parameter is true.
3044 * Else, return NULL.
3046 static PgStat_StatTabEntry *
3047 pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
3049 PgStat_StatTabEntry *result;
3051 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3053 /* Lookup or create the hash table entry for this table */
3054 result = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3058 if (!create && !found)
3061 /* If not found, initialize the new one. */
3064 result->numscans = 0;
3065 result->tuples_returned = 0;
3066 result->tuples_fetched = 0;
3067 result->tuples_inserted = 0;
3068 result->tuples_updated = 0;
3069 result->tuples_deleted = 0;
3070 result->tuples_hot_updated = 0;
3071 result->n_live_tuples = 0;
3072 result->n_dead_tuples = 0;
3073 result->changes_since_analyze = 0;
3074 result->blocks_fetched = 0;
3075 result->blocks_hit = 0;
3077 result->vacuum_timestamp = 0;
3078 result->autovac_vacuum_timestamp = 0;
3079 result->analyze_timestamp = 0;
3080 result->autovac_analyze_timestamp = 0;
3088 * pgstat_write_statsfile() -
3091 * If writing to the permanent file (happens when the collector is
3092 * shutting down only), remove the temporary file so that backends
3093 * starting up under a new postmaster can't read the old data before
3094 * the new collector is ready.
3098 pgstat_write_statsfile(bool permanent)
3100 HASH_SEQ_STATUS hstat;
3101 HASH_SEQ_STATUS tstat;
3102 HASH_SEQ_STATUS fstat;
3103 PgStat_StatDBEntry *dbentry;
3104 PgStat_StatTabEntry *tabentry;
3105 PgStat_StatFuncEntry *funcentry;
3108 const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
3109 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3112 * Open the statistics temp file to write out the current values.
3114 fpout = AllocateFile(tmpfile, PG_BINARY_W);
3118 (errcode_for_file_access(),
3119 errmsg("could not open temporary statistics file \"%s\": %m",
3125 * Set the timestamp of the stats file.
3127 globalStats.stats_timestamp = GetCurrentTimestamp();
3130 * Write the file header --- currently just a format ID.
3132 format_id = PGSTAT_FILE_FORMAT_ID;
3133 fwrite(&format_id, sizeof(format_id), 1, fpout);
3136 * Write global stats struct
3138 fwrite(&globalStats, sizeof(globalStats), 1, fpout);
3141 * Walk through the database table.
3143 hash_seq_init(&hstat, pgStatDBHash);
3144 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
3147 * Write out the DB entry including the number of live backends. We
3148 * don't write the tables or functions pointers, since they're of no
3149 * use to any other process.
3152 fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
3155 * Walk through the database's access stats per table.
3157 hash_seq_init(&tstat, dbentry->tables);
3158 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
3161 fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
3165 * Walk through the database's function stats table.
3167 hash_seq_init(&fstat, dbentry->functions);
3168 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
3171 fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
3175 * Mark the end of this DB
3181 * No more output to be done. Close the temp file and replace the old
3182 * pgstat.stat with it. The ferror() check replaces testing for error
3183 * after each individual fputc or fwrite above.
3190 (errcode_for_file_access(),
3191 errmsg("could not write temporary statistics file \"%s\": %m",
3196 else if (FreeFile(fpout) < 0)
3199 (errcode_for_file_access(),
3200 errmsg("could not close temporary statistics file \"%s\": %m",
3204 else if (rename(tmpfile, statfile) < 0)
3207 (errcode_for_file_access(),
3208 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
3209 tmpfile, statfile)));
3215 * Successful write, so update last_statwrite.
3217 last_statwrite = globalStats.stats_timestamp;
3220 * It's not entirely clear whether there could be clock skew between
3221 * backends and the collector; but just in case someone manages to
3222 * send us a stats request time that's far in the future, reset it.
3223 * This ensures that no inquiry message can cause more than one stats
3224 * file write to occur.
3226 last_statrequest = last_statwrite;
3230 unlink(pgstat_stat_filename);
3235 * pgstat_read_statsfile() -
3237 * Reads in an existing statistics collector file and initializes the
3238 * databases' hash table (whose entries point to the tables' hash tables).
3242 pgstat_read_statsfile(Oid onlydb, bool permanent)
3244 PgStat_StatDBEntry *dbentry;
3245 PgStat_StatDBEntry dbbuf;
3246 PgStat_StatTabEntry *tabentry;
3247 PgStat_StatTabEntry tabbuf;
3248 PgStat_StatFuncEntry funcbuf;
3249 PgStat_StatFuncEntry *funcentry;
3252 HTAB *tabhash = NULL;
3253 HTAB *funchash = NULL;
3257 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3260 * The tables will live in pgStatLocalContext.
3262 pgstat_setup_memcxt();
3265 * Create the DB hashtable
3267 memset(&hash_ctl, 0, sizeof(hash_ctl));
3268 hash_ctl.keysize = sizeof(Oid);
3269 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
3270 hash_ctl.hash = oid_hash;
3271 hash_ctl.hcxt = pgStatLocalContext;
3272 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
3273 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3276 * Clear out global statistics so they start from zero in case we can't
3277 * load an existing statsfile.
3279 memset(&globalStats, 0, sizeof(globalStats));
3282 * Try to open the status file. If it doesn't exist, the backends simply
3283 * return zero for anything and the collector simply starts from scratch
3284 * with empty counters.
3286 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3290 * Verify it's of the expected format.
3292 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3293 || format_id != PGSTAT_FILE_FORMAT_ID)
3295 ereport(pgStatRunningInCollector ? LOG : WARNING,
3296 (errmsg("corrupted pgstat.stat file")));
3301 * Read global stats struct
3303 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
3305 ereport(pgStatRunningInCollector ? LOG : WARNING,
3306 (errmsg("corrupted pgstat.stat file")));
3311 * We found an existing collector stats file. Read it and put all the
3312 * hashtable entries into place.
3316 switch (fgetc(fpin))
3319 * 'D' A PgStat_StatDBEntry struct describing a database
3320 * follows. Subsequently, zero to many 'T' and 'F' entries
3321 * will follow until a 'd' is encountered.
3324 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
3325 fpin) != offsetof(PgStat_StatDBEntry, tables))
3327 ereport(pgStatRunningInCollector ? LOG : WARNING,
3328 (errmsg("corrupted pgstat.stat file")));
3333 * Add to the DB hash
3335 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
3336 (void *) &dbbuf.databaseid,
3341 ereport(pgStatRunningInCollector ? LOG : WARNING,
3342 (errmsg("corrupted pgstat.stat file")));
3346 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
3347 dbentry->tables = NULL;
3348 dbentry->functions = NULL;
3351 * Don't collect tables if not the requested DB (or the
3352 * shared-table info)
3354 if (onlydb != InvalidOid)
3356 if (dbbuf.databaseid != onlydb &&
3357 dbbuf.databaseid != InvalidOid)
3361 memset(&hash_ctl, 0, sizeof(hash_ctl));
3362 hash_ctl.keysize = sizeof(Oid);
3363 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3364 hash_ctl.hash = oid_hash;
3365 hash_ctl.hcxt = pgStatLocalContext;
3366 dbentry->tables = hash_create("Per-database table",
3367 PGSTAT_TAB_HASH_SIZE,
3369 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3371 hash_ctl.keysize = sizeof(Oid);
3372 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3373 hash_ctl.hash = oid_hash;
3374 hash_ctl.hcxt = pgStatLocalContext;
3375 dbentry->functions = hash_create("Per-database function",
3376 PGSTAT_FUNCTION_HASH_SIZE,
3378 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3381 * Arrange that following records add entries to this
3382 * database's hash tables.
3384 tabhash = dbentry->tables;
3385 funchash = dbentry->functions;
3389 * 'd' End of this database.
3397 * 'T' A PgStat_StatTabEntry follows.
3400 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
3401 fpin) != sizeof(PgStat_StatTabEntry))
3403 ereport(pgStatRunningInCollector ? LOG : WARNING,
3404 (errmsg("corrupted pgstat.stat file")));
3409 * Skip if table belongs to a not requested database.
3411 if (tabhash == NULL)
3414 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
3415 (void *) &tabbuf.tableid,
3416 HASH_ENTER, &found);
3420 ereport(pgStatRunningInCollector ? LOG : WARNING,
3421 (errmsg("corrupted pgstat.stat file")));
3425 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
3429 * 'F' A PgStat_StatFuncEntry follows.
3432 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
3433 fpin) != sizeof(PgStat_StatFuncEntry))
3435 ereport(pgStatRunningInCollector ? LOG : WARNING,
3436 (errmsg("corrupted pgstat.stat file")));
3441 * Skip if function belongs to a not requested database.
3443 if (funchash == NULL)
3446 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
3447 (void *) &funcbuf.functionid,
3448 HASH_ENTER, &found);
3452 ereport(pgStatRunningInCollector ? LOG : WARNING,
3453 (errmsg("corrupted pgstat.stat file")));
3457 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
3461 * 'E' The EOF marker of a complete stats file.
3467 ereport(pgStatRunningInCollector ? LOG : WARNING,
3468 (errmsg("corrupted pgstat.stat file")));
3477 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
3483 * pgstat_read_statsfile_timestamp() -
3485 * Attempt to fetch the timestamp of an existing stats file.
3486 * Returns TRUE if successful (timestamp is stored at *ts).
3490 pgstat_read_statsfile_timestamp(bool permanent, TimestampTz *ts)
3492 PgStat_GlobalStats myGlobalStats;
3495 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3498 * Try to open the status file.
3500 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3504 * Verify it's of the expected format.
3506 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3507 || format_id != PGSTAT_FILE_FORMAT_ID)
3514 * Read global stats struct
3516 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats), fpin) != sizeof(myGlobalStats))
3522 *ts = myGlobalStats.stats_timestamp;
3529 * If not already done, read the statistics collector stats file into
3530 * some hash tables. The results will be kept until pgstat_clear_snapshot()
3531 * is called (typically, at end of transaction).
3534 backend_read_statsfile(void)
3539 /* already read it? */
3542 Assert(!pgStatRunningInCollector);
3545 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL msec
3546 * before now. This indirectly ensures that the collector needn't write
3547 * the file more often than PGSTAT_STAT_INTERVAL. In an autovacuum
3548 * worker, however, we want a lower delay to avoid using stale data, so we
3549 * use PGSTAT_RETRY_DELAY (since the number of worker is low, this
3550 * shouldn't be a problem).
3552 * Note that we don't recompute min_ts after sleeping; so we might end up
3553 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In practice
3554 * that shouldn't happen, though, as long as the sleep time is less than
3555 * PGSTAT_STAT_INTERVAL; and we don't want to lie to the collector about
3556 * what our cutoff time really is.
3558 if (IsAutoVacuumWorkerProcess())
3559 min_ts = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
3560 -PGSTAT_RETRY_DELAY);
3562 min_ts = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
3563 -PGSTAT_STAT_INTERVAL);
3566 * Loop until fresh enough stats file is available or we ran out of time.
3567 * The stats inquiry message is sent repeatedly in case collector drops
3570 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
3572 TimestampTz file_ts = 0;
3574 CHECK_FOR_INTERRUPTS();
3576 if (pgstat_read_statsfile_timestamp(false, &file_ts) &&
3580 /* Not there or too old, so kick the collector and wait a bit */
3581 pgstat_send_inquiry(min_ts);
3582 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
3585 if (count >= PGSTAT_POLL_LOOP_COUNT)
3586 elog(WARNING, "pgstat wait timeout");
3588 /* Autovacuum launcher wants stats about all databases */
3589 if (IsAutoVacuumLauncherProcess())
3590 pgStatDBHash = pgstat_read_statsfile(InvalidOid, false);
3592 pgStatDBHash = pgstat_read_statsfile(MyDatabaseId, false);
3597 * pgstat_setup_memcxt() -
3599 * Create pgStatLocalContext, if not already done.
3603 pgstat_setup_memcxt(void)
3605 if (!pgStatLocalContext)
3606 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
3607 "Statistics snapshot",
3608 ALLOCSET_SMALL_MINSIZE,
3609 ALLOCSET_SMALL_INITSIZE,
3610 ALLOCSET_SMALL_MAXSIZE);
3615 * pgstat_clear_snapshot() -
3617 * Discard any data collected in the current transaction. Any subsequent
3618 * request will cause new snapshots to be read.
3620 * This is also invoked during transaction commit or abort to discard
3621 * the no-longer-wanted snapshot.
3625 pgstat_clear_snapshot(void)
3627 /* Release memory, if any was allocated */
3628 if (pgStatLocalContext)
3629 MemoryContextDelete(pgStatLocalContext);
3631 /* Reset variables */
3632 pgStatLocalContext = NULL;
3633 pgStatDBHash = NULL;
3634 localBackendStatusTable = NULL;
3635 localNumBackends = 0;
3640 * pgstat_recv_inquiry() -
3642 * Process stat inquiry requests.
3646 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
3648 if (msg->inquiry_time > last_statrequest)
3649 last_statrequest = msg->inquiry_time;
3654 * pgstat_recv_tabstat() -
3656 * Count what the backend has done.
3660 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
3662 PgStat_StatDBEntry *dbentry;
3663 PgStat_StatTabEntry *tabentry;
3667 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3670 * Update database-wide stats.
3672 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
3673 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
3676 * Process all table entries in the message.
3678 for (i = 0; i < msg->m_nentries; i++)
3680 PgStat_TableEntry *tabmsg = &(msg->m_entry[i]);
3682 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3683 (void *) &(tabmsg->t_id),
3684 HASH_ENTER, &found);
3689 * If it's a new table entry, initialize counters to the values we
3692 tabentry->numscans = tabmsg->t_counts.t_numscans;
3693 tabentry->tuples_returned = tabmsg->t_counts.t_tuples_returned;
3694 tabentry->tuples_fetched = tabmsg->t_counts.t_tuples_fetched;
3695 tabentry->tuples_inserted = tabmsg->t_counts.t_tuples_inserted;
3696 tabentry->tuples_updated = tabmsg->t_counts.t_tuples_updated;
3697 tabentry->tuples_deleted = tabmsg->t_counts.t_tuples_deleted;
3698 tabentry->tuples_hot_updated = tabmsg->t_counts.t_tuples_hot_updated;
3699 tabentry->n_live_tuples = tabmsg->t_counts.t_delta_live_tuples;
3700 tabentry->n_dead_tuples = tabmsg->t_counts.t_delta_dead_tuples;
3701 tabentry->changes_since_analyze = tabmsg->t_counts.t_changed_tuples;
3702 tabentry->blocks_fetched = tabmsg->t_counts.t_blocks_fetched;
3703 tabentry->blocks_hit = tabmsg->t_counts.t_blocks_hit;
3705 tabentry->vacuum_timestamp = 0;
3706 tabentry->autovac_vacuum_timestamp = 0;
3707 tabentry->analyze_timestamp = 0;
3708 tabentry->autovac_analyze_timestamp = 0;
3713 * Otherwise add the values to the existing entry.
3715 tabentry->numscans += tabmsg->t_counts.t_numscans;
3716 tabentry->tuples_returned += tabmsg->t_counts.t_tuples_returned;
3717 tabentry->tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
3718 tabentry->tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
3719 tabentry->tuples_updated += tabmsg->t_counts.t_tuples_updated;
3720 tabentry->tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
3721 tabentry->tuples_hot_updated += tabmsg->t_counts.t_tuples_hot_updated;
3722 tabentry->n_live_tuples += tabmsg->t_counts.t_delta_live_tuples;
3723 tabentry->n_dead_tuples += tabmsg->t_counts.t_delta_dead_tuples;
3724 tabentry->changes_since_analyze += tabmsg->t_counts.t_changed_tuples;
3725 tabentry->blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
3726 tabentry->blocks_hit += tabmsg->t_counts.t_blocks_hit;
3729 /* Clamp n_live_tuples in case of negative delta_live_tuples */
3730 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
3731 /* Likewise for n_dead_tuples */
3732 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
3735 * Add per-table stats to the per-database entry, too.
3737 dbentry->n_tuples_returned += tabmsg->t_counts.t_tuples_returned;
3738 dbentry->n_tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
3739 dbentry->n_tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
3740 dbentry->n_tuples_updated += tabmsg->t_counts.t_tuples_updated;
3741 dbentry->n_tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
3742 dbentry->n_blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
3743 dbentry->n_blocks_hit += tabmsg->t_counts.t_blocks_hit;
3749 * pgstat_recv_tabpurge() -
3751 * Arrange for dead table removal.
3755 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
3757 PgStat_StatDBEntry *dbentry;
3760 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3763 * No need to purge if we don't even know the database.
3765 if (!dbentry || !dbentry->tables)
3769 * Process all table entries in the message.
3771 for (i = 0; i < msg->m_nentries; i++)
3773 /* Remove from hashtable if present; we don't care if it's not. */
3774 (void) hash_search(dbentry->tables,
3775 (void *) &(msg->m_tableid[i]),
3782 * pgstat_recv_dropdb() -
3784 * Arrange for dead database removal
3788 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
3790 PgStat_StatDBEntry *dbentry;
3793 * Lookup the database in the hashtable.
3795 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3798 * If found, remove it.
3802 if (dbentry->tables != NULL)
3803 hash_destroy(dbentry->tables);
3804 if (dbentry->functions != NULL)
3805 hash_destroy(dbentry->functions);
3807 if (hash_search(pgStatDBHash,
3808 (void *) &(dbentry->databaseid),
3809 HASH_REMOVE, NULL) == NULL)
3811 (errmsg("database hash table corrupted "
3812 "during cleanup --- abort")));
3818 * pgstat_recv_resetcounter() -
3820 * Reset the statistics for the specified database.
3824 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
3827 PgStat_StatDBEntry *dbentry;
3830 * Lookup the database in the hashtable. Nothing to do if not there.
3832 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
3838 * We simply throw away all the database's table entries by recreating a
3839 * new hash table for them.
3841 if (dbentry->tables != NULL)
3842 hash_destroy(dbentry->tables);
3843 if (dbentry->functions != NULL)
3844 hash_destroy(dbentry->functions);
3846 dbentry->tables = NULL;
3847 dbentry->functions = NULL;
3848 dbentry->n_xact_commit = 0;
3849 dbentry->n_xact_rollback = 0;
3850 dbentry->n_blocks_fetched = 0;
3851 dbentry->n_blocks_hit = 0;
3853 memset(&hash_ctl, 0, sizeof(hash_ctl));
3854 hash_ctl.keysize = sizeof(Oid);
3855 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3856 hash_ctl.hash = oid_hash;
3857 dbentry->tables = hash_create("Per-database table",
3858 PGSTAT_TAB_HASH_SIZE,
3860 HASH_ELEM | HASH_FUNCTION);
3862 hash_ctl.keysize = sizeof(Oid);
3863 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3864 hash_ctl.hash = oid_hash;
3865 dbentry->functions = hash_create("Per-database function",
3866 PGSTAT_FUNCTION_HASH_SIZE,
3868 HASH_ELEM | HASH_FUNCTION);
3872 * pgstat_recv_autovac() -
3874 * Process an autovacuum signalling message.
3878 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
3880 PgStat_StatDBEntry *dbentry;
3883 * Store the last autovacuum time in the database's hashtable entry.
3885 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3887 dbentry->last_autovac_time = msg->m_start_time;
3891 * pgstat_recv_vacuum() -
3893 * Process a VACUUM message.
3897 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
3899 PgStat_StatDBEntry *dbentry;
3900 PgStat_StatTabEntry *tabentry;
3903 * Store the data in the table's hashtable entry.
3905 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3907 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
3909 if (msg->m_adopt_counts)
3910 tabentry->n_live_tuples = msg->m_tuples;
3911 /* Resetting dead_tuples to 0 is an approximation ... */
3912 tabentry->n_dead_tuples = 0;
3914 if (msg->m_autovacuum)
3915 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
3917 tabentry->vacuum_timestamp = msg->m_vacuumtime;
3921 * pgstat_recv_analyze() -
3923 * Process an ANALYZE message.
3927 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
3929 PgStat_StatDBEntry *dbentry;
3930 PgStat_StatTabEntry *tabentry;
3933 * Store the data in the table's hashtable entry.
3935 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3937 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
3939 if (msg->m_adopt_counts)
3941 tabentry->n_live_tuples = msg->m_live_tuples;
3942 tabentry->n_dead_tuples = msg->m_dead_tuples;
3946 * We reset changes_since_analyze to zero, forgetting any changes that
3947 * occurred while the ANALYZE was in progress.
3949 tabentry->changes_since_analyze = 0;
3951 if (msg->m_autovacuum)
3952 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
3954 tabentry->analyze_timestamp = msg->m_analyzetime;
3959 * pgstat_recv_bgwriter() -
3961 * Process a BGWRITER message.
3965 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
3967 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
3968 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
3969 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
3970 globalStats.buf_written_clean += msg->m_buf_written_clean;
3971 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
3972 globalStats.buf_written_backend += msg->m_buf_written_backend;
3973 globalStats.buf_alloc += msg->m_buf_alloc;
3977 * pgstat_recv_funcstat() -
3979 * Count what the backend has done.
3983 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
3985 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
3986 PgStat_StatDBEntry *dbentry;
3987 PgStat_StatFuncEntry *funcentry;
3991 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
3994 * Process all function entries in the message.
3996 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
3998 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
3999 (void *) &(funcmsg->f_id),
4000 HASH_ENTER, &found);
4005 * If it's a new function entry, initialize counters to the values
4008 funcentry->f_numcalls = funcmsg->f_numcalls;
4009 funcentry->f_time = funcmsg->f_time;
4010 funcentry->f_time_self = funcmsg->f_time_self;
4015 * Otherwise add the values to the existing entry.
4017 funcentry->f_numcalls += funcmsg->f_numcalls;
4018 funcentry->f_time += funcmsg->f_time;
4019 funcentry->f_time_self += funcmsg->f_time_self;
4025 * pgstat_recv_funcpurge() -
4027 * Arrange for dead function removal.
4031 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
4033 PgStat_StatDBEntry *dbentry;
4036 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4039 * No need to purge if we don't even know the database.
4041 if (!dbentry || !dbentry->functions)
4045 * Process all function entries in the message.
4047 for (i = 0; i < msg->m_nentries; i++)
4049 /* Remove from hashtable if present; we don't care if it's not. */
4050 (void) hash_search(dbentry->functions,
4051 (void *) &(msg->m_functionid[i]),