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) 2001-2012, PostgreSQL Global Development Group
16 * src/backend/postmaster/pgstat.c
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 "storage/procsignal.h"
61 #include "utils/ascii.h"
62 #include "utils/guc.h"
63 #include "utils/memutils.h"
64 #include "utils/ps_status.h"
65 #include "utils/rel.h"
66 #include "utils/timestamp.h"
67 #include "utils/tqual.h"
71 * Paths for the statistics files (relative to installation's $PGDATA).
74 #define PGSTAT_STAT_PERMANENT_FILENAME "global/pgstat.stat"
75 #define PGSTAT_STAT_PERMANENT_TMPFILE "global/pgstat.tmp"
81 #define PGSTAT_STAT_INTERVAL 500 /* Minimum time between stats file
82 * updates; in milliseconds. */
84 #define PGSTAT_RETRY_DELAY 10 /* How long to wait between checks for
85 * a new file; in milliseconds. */
87 #define PGSTAT_MAX_WAIT_TIME 10000 /* Maximum time to wait for a stats
88 * file update; in milliseconds. */
90 #define PGSTAT_INQ_INTERVAL 640 /* How often to ping the collector for
91 * a new file; in milliseconds. */
93 #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a
94 * failed statistics collector; in
97 #define PGSTAT_SELECT_TIMEOUT 2 /* How often to check for postmaster
98 * death; in seconds. */
100 #define PGSTAT_POLL_LOOP_COUNT (PGSTAT_MAX_WAIT_TIME / PGSTAT_RETRY_DELAY)
101 #define PGSTAT_INQ_LOOP_COUNT (PGSTAT_INQ_INTERVAL / PGSTAT_RETRY_DELAY)
105 * The initial size hints for the hash tables used in the collector.
108 #define PGSTAT_DB_HASH_SIZE 16
109 #define PGSTAT_TAB_HASH_SIZE 512
110 #define PGSTAT_FUNCTION_HASH_SIZE 512
117 bool pgstat_track_activities = false;
118 bool pgstat_track_counts = false;
119 int pgstat_track_functions = TRACK_FUNC_OFF;
120 int pgstat_track_activity_query_size = 1024;
123 * Built from GUC parameter
126 char *pgstat_stat_filename = NULL;
127 char *pgstat_stat_tmpname = NULL;
130 * BgWriter global statistics counters (unused in other processes).
131 * Stored directly in a stats message structure so it can be sent
132 * without needing to copy things around. We assume this inits to zeroes.
134 PgStat_MsgBgWriter BgWriterStats;
140 NON_EXEC_STATIC pgsocket pgStatSock = PGINVALID_SOCKET;
142 static struct sockaddr_storage pgStatAddr;
144 static time_t last_pgstat_start_time;
146 static bool pgStatRunningInCollector = false;
149 * Structures in which backends store per-table info that's waiting to be
150 * sent to the collector.
152 * NOTE: once allocated, TabStatusArray structures are never moved or deleted
153 * for the life of the backend. Also, we zero out the t_id fields of the
154 * contained PgStat_TableStatus structs whenever they are not actively in use.
155 * This allows relcache pgstat_info pointers to be treated as long-lived data,
156 * avoiding repeated searches in pgstat_initstats() when a relation is
157 * repeatedly opened during a transaction.
159 #define TABSTAT_QUANTUM 100 /* we alloc this many at a time */
161 typedef struct TabStatusArray
163 struct TabStatusArray *tsa_next; /* link to next array, if any */
164 int tsa_used; /* # entries currently used */
165 PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */
168 static TabStatusArray *pgStatTabList = NULL;
171 * Backends store per-function info that's waiting to be sent to the collector
172 * in this hash table (indexed by function OID).
174 static HTAB *pgStatFunctions = NULL;
177 * Indicates if backend has some function stats that it hasn't yet
178 * sent to the collector.
180 static bool have_function_stats = false;
183 * Tuple insertion/deletion counts for an open transaction can't be propagated
184 * into PgStat_TableStatus counters until we know if it is going to commit
185 * or abort. Hence, we keep these counts in per-subxact structs that live
186 * in TopTransactionContext. This data structure is designed on the assumption
187 * that subxacts won't usually modify very many tables.
189 typedef struct PgStat_SubXactStatus
191 int nest_level; /* subtransaction nest level */
192 struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
193 PgStat_TableXactStatus *first; /* head of list for this subxact */
194 } PgStat_SubXactStatus;
196 static PgStat_SubXactStatus *pgStatXactStack = NULL;
198 static int pgStatXactCommit = 0;
199 static int pgStatXactRollback = 0;
200 PgStat_Counter pgStatBlockTimeRead = 0;
201 PgStat_Counter pgStatBlockTimeWrite = 0;
203 /* Record that's written to 2PC state file when pgstat state is persisted */
204 typedef struct TwoPhasePgStatRecord
206 PgStat_Counter tuples_inserted; /* tuples inserted in xact */
207 PgStat_Counter tuples_updated; /* tuples updated in xact */
208 PgStat_Counter tuples_deleted; /* tuples deleted in xact */
209 Oid t_id; /* table's OID */
210 bool t_shared; /* is it a shared catalog? */
211 } TwoPhasePgStatRecord;
214 * Info about current "snapshot" of stats file
216 static MemoryContext pgStatLocalContext = NULL;
217 static HTAB *pgStatDBHash = NULL;
218 static PgBackendStatus *localBackendStatusTable = NULL;
219 static int localNumBackends = 0;
222 * Cluster wide statistics, kept in the stats collector.
223 * Contains statistics that are not collected per database
226 static PgStat_GlobalStats globalStats;
228 /* Last time the collector successfully wrote the stats file */
229 static TimestampTz last_statwrite;
231 /* Latest statistics request time from backends */
232 static TimestampTz last_statrequest;
234 static volatile bool need_exit = false;
235 static volatile bool got_SIGHUP = false;
238 * Total time charged to functions so far in the current backend.
239 * We use this to help separate "self" and "other" time charges.
240 * (We assume this initializes to zero.)
242 static instr_time total_func_time;
246 * Local function forward declarations
250 static pid_t pgstat_forkexec(void);
253 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]);
254 static void pgstat_exit(SIGNAL_ARGS);
255 static void pgstat_beshutdown_hook(int code, Datum arg);
256 static void pgstat_sighup_handler(SIGNAL_ARGS);
258 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
259 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
260 Oid tableoid, bool create);
261 static void pgstat_write_statsfile(bool permanent);
262 static HTAB *pgstat_read_statsfile(Oid onlydb, bool permanent);
263 static void backend_read_statsfile(void);
264 static void pgstat_read_current_status(void);
266 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
267 static void pgstat_send_funcstats(void);
268 static HTAB *pgstat_collect_oids(Oid catalogid);
270 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
272 static void pgstat_setup_memcxt(void);
274 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
275 static void pgstat_send(void *msg, int len);
277 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
278 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
279 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
280 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
281 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
282 static void pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len);
283 static void pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len);
284 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
285 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
286 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
287 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
288 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
289 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
290 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
291 static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
292 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
295 /* ------------------------------------------------------------
296 * Public functions called from postmaster follow
297 * ------------------------------------------------------------
303 * Called from postmaster at startup. Create the resources required
304 * by the statistics collector process. If unable to do so, do not
305 * fail --- better to let the postmaster start with stats collection
312 ACCEPT_TYPE_ARG3 alen;
313 struct addrinfo *addrs = NULL,
323 #define TESTBYTEVAL ((char) 199)
326 * Create the UDP socket for sending and receiving statistic messages
328 hints.ai_flags = AI_PASSIVE;
329 hints.ai_family = PF_UNSPEC;
330 hints.ai_socktype = SOCK_DGRAM;
331 hints.ai_protocol = 0;
332 hints.ai_addrlen = 0;
333 hints.ai_addr = NULL;
334 hints.ai_canonname = NULL;
335 hints.ai_next = NULL;
336 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
340 (errmsg("could not resolve \"localhost\": %s",
341 gai_strerror(ret))));
346 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
347 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
348 * when kernel will reject IPv6). Worse, the failure may occur at the
349 * bind() or perhaps even connect() stage. So we must loop through the
350 * results till we find a working combination. We will generate LOG
351 * messages, but no error, for bogus combinations.
353 for (addr = addrs; addr; addr = addr->ai_next)
355 #ifdef HAVE_UNIX_SOCKETS
356 /* Ignore AF_UNIX sockets, if any are returned. */
357 if (addr->ai_family == AF_UNIX)
363 (errmsg("trying another address for the statistics collector")));
368 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) == PGINVALID_SOCKET)
371 (errcode_for_socket_access(),
372 errmsg("could not create socket for statistics collector: %m")));
377 * Bind it to a kernel assigned port on localhost and get the assigned
378 * port via getsockname().
380 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
383 (errcode_for_socket_access(),
384 errmsg("could not bind socket for statistics collector: %m")));
385 closesocket(pgStatSock);
386 pgStatSock = PGINVALID_SOCKET;
390 alen = sizeof(pgStatAddr);
391 if (getsockname(pgStatSock, (struct sockaddr *) & pgStatAddr, &alen) < 0)
394 (errcode_for_socket_access(),
395 errmsg("could not get address of socket for statistics collector: %m")));
396 closesocket(pgStatSock);
397 pgStatSock = PGINVALID_SOCKET;
402 * Connect the socket to its own address. This saves a few cycles by
403 * not having to respecify the target address on every send. This also
404 * provides a kernel-level check that only packets from this same
405 * address will be received.
407 if (connect(pgStatSock, (struct sockaddr *) & pgStatAddr, alen) < 0)
410 (errcode_for_socket_access(),
411 errmsg("could not connect socket for statistics collector: %m")));
412 closesocket(pgStatSock);
413 pgStatSock = PGINVALID_SOCKET;
418 * Try to send and receive a one-byte test message on the socket. This
419 * is to catch situations where the socket can be created but will not
420 * actually pass data (for instance, because kernel packet filtering
423 test_byte = TESTBYTEVAL;
426 if (send(pgStatSock, &test_byte, 1, 0) != 1)
429 goto retry1; /* if interrupted, just retry */
431 (errcode_for_socket_access(),
432 errmsg("could not send test message on socket for statistics collector: %m")));
433 closesocket(pgStatSock);
434 pgStatSock = PGINVALID_SOCKET;
439 * There could possibly be a little delay before the message can be
440 * received. We arbitrarily allow up to half a second before deciding
443 for (;;) /* need a loop to handle EINTR */
446 FD_SET(pgStatSock, &rset);
450 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
451 if (sel_res >= 0 || errno != EINTR)
457 (errcode_for_socket_access(),
458 errmsg("select() failed in statistics collector: %m")));
459 closesocket(pgStatSock);
460 pgStatSock = PGINVALID_SOCKET;
463 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
466 * This is the case we actually think is likely, so take pains to
467 * give a specific message for it.
469 * errno will not be set meaningfully here, so don't use it.
472 (errcode(ERRCODE_CONNECTION_FAILURE),
473 errmsg("test message did not get through on socket for statistics collector")));
474 closesocket(pgStatSock);
475 pgStatSock = PGINVALID_SOCKET;
479 test_byte++; /* just make sure variable is changed */
482 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
485 goto retry2; /* if interrupted, just retry */
487 (errcode_for_socket_access(),
488 errmsg("could not receive test message on socket for statistics collector: %m")));
489 closesocket(pgStatSock);
490 pgStatSock = PGINVALID_SOCKET;
494 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
497 (errcode(ERRCODE_INTERNAL_ERROR),
498 errmsg("incorrect test message transmission on socket for statistics collector")));
499 closesocket(pgStatSock);
500 pgStatSock = PGINVALID_SOCKET;
504 /* If we get here, we have a working socket */
508 /* Did we find a working address? */
509 if (!addr || pgStatSock == PGINVALID_SOCKET)
513 * Set the socket to non-blocking IO. This ensures that if the collector
514 * falls behind, statistics messages will be discarded; backends won't
515 * block waiting to send messages to the collector.
517 if (!pg_set_noblock(pgStatSock))
520 (errcode_for_socket_access(),
521 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
525 pg_freeaddrinfo_all(hints.ai_family, addrs);
531 (errmsg("disabling statistics collector for lack of working socket")));
534 pg_freeaddrinfo_all(hints.ai_family, addrs);
536 if (pgStatSock != PGINVALID_SOCKET)
537 closesocket(pgStatSock);
538 pgStatSock = PGINVALID_SOCKET;
541 * Adjust GUC variables to suppress useless activity, and for debugging
542 * purposes (seeing track_counts off is a clue that we failed here). We
543 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
544 * on from postgresql.conf without a restart.
546 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
550 * pgstat_reset_all() -
552 * Remove the stats file. This is currently used only if WAL
553 * recovery is needed after a crash.
556 pgstat_reset_all(void)
558 unlink(pgstat_stat_filename);
559 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
565 * pgstat_forkexec() -
567 * Format up the arglist for, then fork and exec, statistics collector process
570 pgstat_forkexec(void)
575 av[ac++] = "postgres";
576 av[ac++] = "--forkcol";
577 av[ac++] = NULL; /* filled in by postmaster_forkexec */
580 Assert(ac < lengthof(av));
582 return postmaster_forkexec(ac, av);
584 #endif /* EXEC_BACKEND */
590 * Called from postmaster at startup or after an existing collector
591 * died. Attempt to fire up a fresh statistics collector.
593 * Returns PID of child process, or 0 if fail.
595 * Note: if fail, we will be called again from the postmaster main loop.
604 * Check that the socket is there, else pgstat_init failed and we can do
607 if (pgStatSock == PGINVALID_SOCKET)
611 * Do nothing if too soon since last collector start. This is a safety
612 * valve to protect against continuous respawn attempts if the collector
613 * is dying immediately at launch. Note that since we will be re-called
614 * from the postmaster main loop, we will get another chance later.
616 curtime = time(NULL);
617 if ((unsigned int) (curtime - last_pgstat_start_time) <
618 (unsigned int) PGSTAT_RESTART_INTERVAL)
620 last_pgstat_start_time = curtime;
623 * Okay, fork off the collector.
626 switch ((pgStatPid = pgstat_forkexec()))
628 switch ((pgStatPid = fork_process()))
633 (errmsg("could not fork statistics collector: %m")));
638 /* in postmaster child ... */
639 /* Close the postmaster's sockets */
640 ClosePostmasterPorts(false);
642 /* Lose the postmaster's on-exit routines */
645 /* Drop our connection to postmaster's shared memory, as well */
646 PGSharedMemoryDetach();
648 PgstatCollectorMain(0, NULL);
653 return (int) pgStatPid;
656 /* shouldn't get here */
661 allow_immediate_pgstat_restart(void)
663 last_pgstat_start_time = 0;
666 /* ------------------------------------------------------------
667 * Public functions used by backends follow
668 *------------------------------------------------------------
673 * pgstat_report_stat() -
675 * Called from tcop/postgres.c to send the so far collected per-table
676 * and function usage statistics to the collector. Note that this is
677 * called only when not within a transaction, so it is fair to use
678 * transaction stop time as an approximation of current time.
682 pgstat_report_stat(bool force)
684 /* we assume this inits to all zeroes: */
685 static const PgStat_TableCounts all_zeroes;
686 static TimestampTz last_report = 0;
689 PgStat_MsgTabstat regular_msg;
690 PgStat_MsgTabstat shared_msg;
694 /* Don't expend a clock check if nothing to do */
695 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0)
696 && !have_function_stats)
700 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
701 * msec since we last sent one, or the caller wants to force stats out.
703 now = GetCurrentTransactionStopTimestamp();
705 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
710 * Scan through the TabStatusArray struct(s) to find tables that actually
711 * have counts, and build messages to send. We have to separate shared
712 * relations from regular ones because the databaseid field in the message
713 * header has to depend on that.
715 regular_msg.m_databaseid = MyDatabaseId;
716 shared_msg.m_databaseid = InvalidOid;
717 regular_msg.m_nentries = 0;
718 shared_msg.m_nentries = 0;
720 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
722 for (i = 0; i < tsa->tsa_used; i++)
724 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
725 PgStat_MsgTabstat *this_msg;
726 PgStat_TableEntry *this_ent;
728 /* Shouldn't have any pending transaction-dependent counts */
729 Assert(entry->trans == NULL);
732 * Ignore entries that didn't accumulate any actual counts, such
733 * as indexes that were opened by the planner but not used.
735 if (memcmp(&entry->t_counts, &all_zeroes,
736 sizeof(PgStat_TableCounts)) == 0)
740 * OK, insert data into the appropriate message, and send if full.
742 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
743 this_ent = &this_msg->m_entry[this_msg->m_nentries];
744 this_ent->t_id = entry->t_id;
745 memcpy(&this_ent->t_counts, &entry->t_counts,
746 sizeof(PgStat_TableCounts));
747 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
749 pgstat_send_tabstat(this_msg);
750 this_msg->m_nentries = 0;
753 /* zero out TableStatus structs after use */
754 MemSet(tsa->tsa_entries, 0,
755 tsa->tsa_used * sizeof(PgStat_TableStatus));
760 * Send partial messages. If force is true, make sure that any pending
761 * xact commit/abort gets counted, even if no table stats to send.
763 if (regular_msg.m_nentries > 0 ||
764 (force && (pgStatXactCommit > 0 || pgStatXactRollback > 0)))
765 pgstat_send_tabstat(®ular_msg);
766 if (shared_msg.m_nentries > 0)
767 pgstat_send_tabstat(&shared_msg);
769 /* Now, send function statistics */
770 pgstat_send_funcstats();
774 * Subroutine for pgstat_report_stat: finish and send a tabstat message
777 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
782 /* It's unlikely we'd get here with no socket, but maybe not impossible */
783 if (pgStatSock == PGINVALID_SOCKET)
787 * Report accumulated xact commit/rollback and I/O timings whenever we send
788 * a normal tabstat message
790 if (OidIsValid(tsmsg->m_databaseid))
792 tsmsg->m_xact_commit = pgStatXactCommit;
793 tsmsg->m_xact_rollback = pgStatXactRollback;
794 tsmsg->m_block_time_read = pgStatBlockTimeRead;
795 tsmsg->m_block_time_write = pgStatBlockTimeWrite;
796 pgStatXactCommit = 0;
797 pgStatXactRollback = 0;
801 tsmsg->m_xact_commit = 0;
802 tsmsg->m_xact_rollback = 0;
803 tsmsg->m_block_time_read = 0;
804 tsmsg->m_block_time_write = 0;
807 n = tsmsg->m_nentries;
808 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
809 n * sizeof(PgStat_TableEntry);
811 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
812 pgstat_send(tsmsg, len);
816 * Subroutine for pgstat_report_stat: populate and send a function stat message
819 pgstat_send_funcstats(void)
821 /* we assume this inits to all zeroes: */
822 static const PgStat_FunctionCounts all_zeroes;
824 PgStat_MsgFuncstat msg;
825 PgStat_BackendFunctionEntry *entry;
826 HASH_SEQ_STATUS fstat;
828 if (pgStatFunctions == NULL)
831 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
832 msg.m_databaseid = MyDatabaseId;
835 hash_seq_init(&fstat, pgStatFunctions);
836 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
838 PgStat_FunctionEntry *m_ent;
840 /* Skip it if no counts accumulated since last time */
841 if (memcmp(&entry->f_counts, &all_zeroes,
842 sizeof(PgStat_FunctionCounts)) == 0)
845 /* need to convert format of time accumulators */
846 m_ent = &msg.m_entry[msg.m_nentries];
847 m_ent->f_id = entry->f_id;
848 m_ent->f_numcalls = entry->f_counts.f_numcalls;
849 m_ent->f_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time);
850 m_ent->f_time_self = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_time_self);
852 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
854 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
855 msg.m_nentries * sizeof(PgStat_FunctionEntry));
859 /* reset the entry's counts */
860 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
863 if (msg.m_nentries > 0)
864 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
865 msg.m_nentries * sizeof(PgStat_FunctionEntry));
867 have_function_stats = false;
872 * pgstat_vacuum_stat() -
874 * Will tell the collector about objects he can get rid of.
878 pgstat_vacuum_stat(void)
881 PgStat_MsgTabpurge msg;
882 PgStat_MsgFuncpurge f_msg;
883 HASH_SEQ_STATUS hstat;
884 PgStat_StatDBEntry *dbentry;
885 PgStat_StatTabEntry *tabentry;
886 PgStat_StatFuncEntry *funcentry;
889 if (pgStatSock == PGINVALID_SOCKET)
893 * If not done for this transaction, read the statistics collector stats
894 * file into some hash tables.
896 backend_read_statsfile();
899 * Read pg_database and make a list of OIDs of all existing databases
901 htab = pgstat_collect_oids(DatabaseRelationId);
904 * Search the database hash table for dead databases and tell the
905 * collector to drop them.
907 hash_seq_init(&hstat, pgStatDBHash);
908 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
910 Oid dbid = dbentry->databaseid;
912 CHECK_FOR_INTERRUPTS();
914 /* the DB entry for shared tables (with InvalidOid) is never dropped */
915 if (OidIsValid(dbid) &&
916 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
917 pgstat_drop_database(dbid);
924 * Lookup our own database entry; if not found, nothing more to do.
926 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
927 (void *) &MyDatabaseId,
929 if (dbentry == NULL || dbentry->tables == NULL)
933 * Similarly to above, make a list of all known relations in this DB.
935 htab = pgstat_collect_oids(RelationRelationId);
938 * Initialize our messages table counter to zero
943 * Check for all tables listed in stats hashtable if they still exist.
945 hash_seq_init(&hstat, dbentry->tables);
946 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
948 Oid tabid = tabentry->tableid;
950 CHECK_FOR_INTERRUPTS();
952 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
956 * Not there, so add this table's Oid to the message
958 msg.m_tableid[msg.m_nentries++] = tabid;
961 * If the message is full, send it out and reinitialize to empty
963 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
965 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
966 +msg.m_nentries * sizeof(Oid);
968 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
969 msg.m_databaseid = MyDatabaseId;
970 pgstat_send(&msg, len);
979 if (msg.m_nentries > 0)
981 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
982 +msg.m_nentries * sizeof(Oid);
984 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
985 msg.m_databaseid = MyDatabaseId;
986 pgstat_send(&msg, len);
993 * Now repeat the above steps for functions. However, we needn't bother
994 * in the common case where no function stats are being collected.
996 if (dbentry->functions != NULL &&
997 hash_get_num_entries(dbentry->functions) > 0)
999 htab = pgstat_collect_oids(ProcedureRelationId);
1001 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
1002 f_msg.m_databaseid = MyDatabaseId;
1003 f_msg.m_nentries = 0;
1005 hash_seq_init(&hstat, dbentry->functions);
1006 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
1008 Oid funcid = funcentry->functionid;
1010 CHECK_FOR_INTERRUPTS();
1012 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
1016 * Not there, so add this function's Oid to the message
1018 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
1021 * If the message is full, send it out and reinitialize to empty
1023 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1025 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1026 +f_msg.m_nentries * sizeof(Oid);
1028 pgstat_send(&f_msg, len);
1030 f_msg.m_nentries = 0;
1037 if (f_msg.m_nentries > 0)
1039 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1040 +f_msg.m_nentries * sizeof(Oid);
1042 pgstat_send(&f_msg, len);
1051 * pgstat_collect_oids() -
1053 * Collect the OIDs of all objects listed in the specified system catalog
1054 * into a temporary hash table. Caller should hash_destroy the result
1055 * when done with it. (However, we make the table in CurrentMemoryContext
1056 * so that it will be freed properly in event of an error.)
1060 pgstat_collect_oids(Oid catalogid)
1068 memset(&hash_ctl, 0, sizeof(hash_ctl));
1069 hash_ctl.keysize = sizeof(Oid);
1070 hash_ctl.entrysize = sizeof(Oid);
1071 hash_ctl.hash = oid_hash;
1072 hash_ctl.hcxt = CurrentMemoryContext;
1073 htab = hash_create("Temporary table of OIDs",
1074 PGSTAT_TAB_HASH_SIZE,
1076 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
1078 rel = heap_open(catalogid, AccessShareLock);
1079 scan = heap_beginscan(rel, SnapshotNow, 0, NULL);
1080 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1082 Oid thisoid = HeapTupleGetOid(tup);
1084 CHECK_FOR_INTERRUPTS();
1086 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1089 heap_close(rel, AccessShareLock);
1096 * pgstat_drop_database() -
1098 * Tell the collector that we just dropped a database.
1099 * (If the message gets lost, we will still clean the dead DB eventually
1100 * via future invocations of pgstat_vacuum_stat().)
1104 pgstat_drop_database(Oid databaseid)
1106 PgStat_MsgDropdb msg;
1108 if (pgStatSock == PGINVALID_SOCKET)
1111 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1112 msg.m_databaseid = databaseid;
1113 pgstat_send(&msg, sizeof(msg));
1118 * pgstat_drop_relation() -
1120 * Tell the collector that we just dropped a relation.
1121 * (If the message gets lost, we will still clean the dead entry eventually
1122 * via future invocations of pgstat_vacuum_stat().)
1124 * Currently not used for lack of any good place to call it; we rely
1125 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1130 pgstat_drop_relation(Oid relid)
1132 PgStat_MsgTabpurge msg;
1135 if (pgStatSock == PGINVALID_SOCKET)
1138 msg.m_tableid[0] = relid;
1141 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) +sizeof(Oid);
1143 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1144 msg.m_databaseid = MyDatabaseId;
1145 pgstat_send(&msg, len);
1147 #endif /* NOT_USED */
1151 * pgstat_reset_counters() -
1153 * Tell the statistics collector to reset counters for our database.
1157 pgstat_reset_counters(void)
1159 PgStat_MsgResetcounter msg;
1161 if (pgStatSock == PGINVALID_SOCKET)
1166 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1167 errmsg("must be superuser to reset statistics counters")));
1169 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1170 msg.m_databaseid = MyDatabaseId;
1171 pgstat_send(&msg, sizeof(msg));
1175 * pgstat_reset_shared_counters() -
1177 * Tell the statistics collector to reset cluster-wide shared counters.
1181 pgstat_reset_shared_counters(const char *target)
1183 PgStat_MsgResetsharedcounter msg;
1185 if (pgStatSock == PGINVALID_SOCKET)
1190 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1191 errmsg("must be superuser to reset statistics counters")));
1193 if (strcmp(target, "bgwriter") == 0)
1194 msg.m_resettarget = RESET_BGWRITER;
1197 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1198 errmsg("unrecognized reset target: \"%s\"", target),
1199 errhint("Target must be \"bgwriter\".")));
1201 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSHAREDCOUNTER);
1202 pgstat_send(&msg, sizeof(msg));
1206 * pgstat_reset_single_counter() -
1208 * Tell the statistics collector to reset a single counter.
1212 pgstat_reset_single_counter(Oid objoid, PgStat_Single_Reset_Type type)
1214 PgStat_MsgResetsinglecounter msg;
1216 if (pgStatSock == PGINVALID_SOCKET)
1221 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1222 errmsg("must be superuser to reset statistics counters")));
1224 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSINGLECOUNTER);
1225 msg.m_databaseid = MyDatabaseId;
1226 msg.m_resettype = type;
1227 msg.m_objectid = objoid;
1229 pgstat_send(&msg, sizeof(msg));
1233 * pgstat_report_autovac() -
1235 * Called from autovacuum.c to report startup of an autovacuum process.
1236 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1237 * the db OID must be passed in, instead.
1241 pgstat_report_autovac(Oid dboid)
1243 PgStat_MsgAutovacStart msg;
1245 if (pgStatSock == PGINVALID_SOCKET)
1248 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1249 msg.m_databaseid = dboid;
1250 msg.m_start_time = GetCurrentTimestamp();
1252 pgstat_send(&msg, sizeof(msg));
1257 * pgstat_report_vacuum() -
1259 * Tell the collector about the table we just vacuumed.
1263 pgstat_report_vacuum(Oid tableoid, bool shared, PgStat_Counter tuples)
1265 PgStat_MsgVacuum msg;
1267 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1270 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1271 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1272 msg.m_tableoid = tableoid;
1273 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1274 msg.m_vacuumtime = GetCurrentTimestamp();
1275 msg.m_tuples = tuples;
1276 pgstat_send(&msg, sizeof(msg));
1280 * pgstat_report_analyze() -
1282 * Tell the collector about the table we just analyzed.
1286 pgstat_report_analyze(Relation rel,
1287 PgStat_Counter livetuples, PgStat_Counter deadtuples)
1289 PgStat_MsgAnalyze msg;
1291 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1295 * Unlike VACUUM, ANALYZE might be running inside a transaction that has
1296 * already inserted and/or deleted rows in the target table. ANALYZE will
1297 * have counted such rows as live or dead respectively. Because we will
1298 * report our counts of such rows at transaction end, we should subtract
1299 * off these counts from what we send to the collector now, else they'll
1300 * be double-counted after commit. (This approach also ensures that the
1301 * collector ends up with the right numbers if we abort instead of
1304 if (rel->pgstat_info != NULL)
1306 PgStat_TableXactStatus *trans;
1308 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1310 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1311 deadtuples -= trans->tuples_updated + trans->tuples_deleted;
1313 /* count stuff inserted by already-aborted subxacts, too */
1314 deadtuples -= rel->pgstat_info->t_counts.t_delta_dead_tuples;
1315 /* Since ANALYZE's counts are estimates, we could have underflowed */
1316 livetuples = Max(livetuples, 0);
1317 deadtuples = Max(deadtuples, 0);
1320 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1321 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1322 msg.m_tableoid = RelationGetRelid(rel);
1323 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1324 msg.m_analyzetime = GetCurrentTimestamp();
1325 msg.m_live_tuples = livetuples;
1326 msg.m_dead_tuples = deadtuples;
1327 pgstat_send(&msg, sizeof(msg));
1331 * pgstat_report_recovery_conflict() -
1333 * Tell the collector about a Hot Standby recovery conflict.
1337 pgstat_report_recovery_conflict(int reason)
1339 PgStat_MsgRecoveryConflict msg;
1341 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1344 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RECOVERYCONFLICT);
1345 msg.m_databaseid = MyDatabaseId;
1346 msg.m_reason = reason;
1347 pgstat_send(&msg, sizeof(msg));
1351 * pgstat_report_deadlock() -
1353 * Tell the collector about a deadlock detected.
1357 pgstat_report_deadlock(void)
1359 PgStat_MsgDeadlock msg;
1361 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1364 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DEADLOCK);
1365 msg.m_databaseid = MyDatabaseId;
1366 pgstat_send(&msg, sizeof(msg));
1370 * pgstat_report_tempfile() -
1372 * Tell the collector about a temporary file.
1376 pgstat_report_tempfile(size_t filesize)
1378 PgStat_MsgTempFile msg;
1380 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1383 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TEMPFILE);
1384 msg.m_databaseid = MyDatabaseId;
1385 msg.m_filesize = filesize;
1386 pgstat_send(&msg, sizeof(msg));
1393 * Send some junk data to the collector to increase traffic.
1399 PgStat_MsgDummy msg;
1401 if (pgStatSock == PGINVALID_SOCKET)
1404 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1405 pgstat_send(&msg, sizeof(msg));
1409 * pgstat_send_inquiry() -
1411 * Notify collector that we need fresh data.
1412 * ts specifies the minimum acceptable timestamp for the stats file.
1416 pgstat_send_inquiry(TimestampTz ts)
1418 PgStat_MsgInquiry msg;
1420 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1421 msg.inquiry_time = ts;
1422 pgstat_send(&msg, sizeof(msg));
1427 * Initialize function call usage data.
1428 * Called by the executor before invoking a function.
1431 pgstat_init_function_usage(FunctionCallInfoData *fcinfo,
1432 PgStat_FunctionCallUsage *fcu)
1434 PgStat_BackendFunctionEntry *htabent;
1437 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1439 /* stats not wanted */
1444 if (!pgStatFunctions)
1446 /* First time through - initialize function stat table */
1449 memset(&hash_ctl, 0, sizeof(hash_ctl));
1450 hash_ctl.keysize = sizeof(Oid);
1451 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1452 hash_ctl.hash = oid_hash;
1453 pgStatFunctions = hash_create("Function stat entries",
1454 PGSTAT_FUNCTION_HASH_SIZE,
1456 HASH_ELEM | HASH_FUNCTION);
1459 /* Get the stats entry for this function, create if necessary */
1460 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1461 HASH_ENTER, &found);
1463 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1465 fcu->fs = &htabent->f_counts;
1467 /* save stats for this function, later used to compensate for recursion */
1468 fcu->save_f_time = htabent->f_counts.f_time;
1470 /* save current backend-wide total time */
1471 fcu->save_total = total_func_time;
1473 /* get clock time as of function start */
1474 INSTR_TIME_SET_CURRENT(fcu->f_start);
1478 * find_funcstat_entry - find any existing PgStat_BackendFunctionEntry entry
1479 * for specified function
1481 * If no entry, return NULL, don't create a new one
1483 PgStat_BackendFunctionEntry *
1484 find_funcstat_entry(Oid func_id)
1486 if (pgStatFunctions == NULL)
1489 return (PgStat_BackendFunctionEntry *) hash_search(pgStatFunctions,
1495 * Calculate function call usage and update stat counters.
1496 * Called by the executor after invoking a function.
1498 * In the case of a set-returning function that runs in value-per-call mode,
1499 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1500 * calls for what the user considers a single call of the function. The
1501 * finalize flag should be TRUE on the last call.
1504 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1506 PgStat_FunctionCounts *fs = fcu->fs;
1508 instr_time f_others;
1511 /* stats not wanted? */
1515 /* total elapsed time in this function call */
1516 INSTR_TIME_SET_CURRENT(f_total);
1517 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1519 /* self usage: elapsed minus anything already charged to other calls */
1520 f_others = total_func_time;
1521 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1523 INSTR_TIME_SUBTRACT(f_self, f_others);
1525 /* update backend-wide total time */
1526 INSTR_TIME_ADD(total_func_time, f_self);
1529 * Compute the new total f_time as the total elapsed time added to the
1530 * pre-call value of f_time. This is necessary to avoid double-counting
1531 * any time taken by recursive calls of myself. (We do not need any
1532 * similar kluge for self time, since that already excludes any recursive
1535 INSTR_TIME_ADD(f_total, fcu->save_f_time);
1537 /* update counters in function stats table */
1540 fs->f_time = f_total;
1541 INSTR_TIME_ADD(fs->f_time_self, f_self);
1543 /* indicate that we have something to send */
1544 have_function_stats = true;
1549 * pgstat_initstats() -
1551 * Initialize a relcache entry to count access statistics.
1552 * Called whenever a relation is opened.
1554 * We assume that a relcache entry's pgstat_info field is zeroed by
1555 * relcache.c when the relcache entry is made; thereafter it is long-lived
1556 * data. We can avoid repeated searches of the TabStatus arrays when the
1557 * same relation is touched repeatedly within a transaction.
1561 pgstat_initstats(Relation rel)
1563 Oid rel_id = rel->rd_id;
1564 char relkind = rel->rd_rel->relkind;
1566 /* We only count stats for things that have storage */
1567 if (!(relkind == RELKIND_RELATION ||
1568 relkind == RELKIND_INDEX ||
1569 relkind == RELKIND_TOASTVALUE ||
1570 relkind == RELKIND_SEQUENCE))
1572 rel->pgstat_info = NULL;
1576 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1578 /* We're not counting at all */
1579 rel->pgstat_info = NULL;
1584 * If we already set up this relation in the current transaction, nothing
1587 if (rel->pgstat_info != NULL &&
1588 rel->pgstat_info->t_id == rel_id)
1591 /* Else find or make the PgStat_TableStatus entry, and update link */
1592 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1596 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1598 static PgStat_TableStatus *
1599 get_tabstat_entry(Oid rel_id, bool isshared)
1601 PgStat_TableStatus *entry;
1602 TabStatusArray *tsa;
1603 TabStatusArray *prev_tsa;
1607 * Search the already-used tabstat slots for this relation.
1610 for (tsa = pgStatTabList; tsa != NULL; prev_tsa = tsa, tsa = tsa->tsa_next)
1612 for (i = 0; i < tsa->tsa_used; i++)
1614 entry = &tsa->tsa_entries[i];
1615 if (entry->t_id == rel_id)
1619 if (tsa->tsa_used < TABSTAT_QUANTUM)
1622 * It must not be present, but we found a free slot instead. Fine,
1623 * let's use this one. We assume the entry was already zeroed,
1624 * either at creation or after last use.
1626 entry = &tsa->tsa_entries[tsa->tsa_used++];
1627 entry->t_id = rel_id;
1628 entry->t_shared = isshared;
1634 * We ran out of tabstat slots, so allocate more. Be sure they're zeroed.
1636 tsa = (TabStatusArray *) MemoryContextAllocZero(TopMemoryContext,
1637 sizeof(TabStatusArray));
1639 prev_tsa->tsa_next = tsa;
1641 pgStatTabList = tsa;
1644 * Use the first entry of the new TabStatusArray.
1646 entry = &tsa->tsa_entries[tsa->tsa_used++];
1647 entry->t_id = rel_id;
1648 entry->t_shared = isshared;
1653 * find_tabstat_entry - find any existing PgStat_TableStatus entry for rel
1655 * If no entry, return NULL, don't create a new one
1657 PgStat_TableStatus *
1658 find_tabstat_entry(Oid rel_id)
1660 PgStat_TableStatus *entry;
1661 TabStatusArray *tsa;
1664 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
1666 for (i = 0; i < tsa->tsa_used; i++)
1668 entry = &tsa->tsa_entries[i];
1669 if (entry->t_id == rel_id)
1679 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1681 static PgStat_SubXactStatus *
1682 get_tabstat_stack_level(int nest_level)
1684 PgStat_SubXactStatus *xact_state;
1686 xact_state = pgStatXactStack;
1687 if (xact_state == NULL || xact_state->nest_level != nest_level)
1689 xact_state = (PgStat_SubXactStatus *)
1690 MemoryContextAlloc(TopTransactionContext,
1691 sizeof(PgStat_SubXactStatus));
1692 xact_state->nest_level = nest_level;
1693 xact_state->prev = pgStatXactStack;
1694 xact_state->first = NULL;
1695 pgStatXactStack = xact_state;
1701 * add_tabstat_xact_level - add a new (sub)transaction state record
1704 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1706 PgStat_SubXactStatus *xact_state;
1707 PgStat_TableXactStatus *trans;
1710 * If this is the first rel to be modified at the current nest level, we
1711 * first have to push a transaction stack entry.
1713 xact_state = get_tabstat_stack_level(nest_level);
1715 /* Now make a per-table stack entry */
1716 trans = (PgStat_TableXactStatus *)
1717 MemoryContextAllocZero(TopTransactionContext,
1718 sizeof(PgStat_TableXactStatus));
1719 trans->nest_level = nest_level;
1720 trans->upper = pgstat_info->trans;
1721 trans->parent = pgstat_info;
1722 trans->next = xact_state->first;
1723 xact_state->first = trans;
1724 pgstat_info->trans = trans;
1728 * pgstat_count_heap_insert - count a tuple insertion of n tuples
1731 pgstat_count_heap_insert(Relation rel, int n)
1733 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1735 if (pgstat_info != NULL)
1737 /* We have to log the effect at the proper transactional level */
1738 int nest_level = GetCurrentTransactionNestLevel();
1740 if (pgstat_info->trans == NULL ||
1741 pgstat_info->trans->nest_level != nest_level)
1742 add_tabstat_xact_level(pgstat_info, nest_level);
1744 pgstat_info->trans->tuples_inserted += n;
1749 * pgstat_count_heap_update - count a tuple update
1752 pgstat_count_heap_update(Relation rel, bool hot)
1754 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1756 if (pgstat_info != NULL)
1758 /* We have to log the effect at the proper transactional level */
1759 int nest_level = GetCurrentTransactionNestLevel();
1761 if (pgstat_info->trans == NULL ||
1762 pgstat_info->trans->nest_level != nest_level)
1763 add_tabstat_xact_level(pgstat_info, nest_level);
1765 pgstat_info->trans->tuples_updated++;
1767 /* t_tuples_hot_updated is nontransactional, so just advance it */
1769 pgstat_info->t_counts.t_tuples_hot_updated++;
1774 * pgstat_count_heap_delete - count a tuple deletion
1777 pgstat_count_heap_delete(Relation rel)
1779 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1781 if (pgstat_info != NULL)
1783 /* We have to log the effect at the proper transactional level */
1784 int nest_level = GetCurrentTransactionNestLevel();
1786 if (pgstat_info->trans == NULL ||
1787 pgstat_info->trans->nest_level != nest_level)
1788 add_tabstat_xact_level(pgstat_info, nest_level);
1790 pgstat_info->trans->tuples_deleted++;
1795 * pgstat_update_heap_dead_tuples - update dead-tuples count
1797 * The semantics of this are that we are reporting the nontransactional
1798 * recovery of "delta" dead tuples; so t_delta_dead_tuples decreases
1799 * rather than increasing, and the change goes straight into the per-table
1800 * counter, not into transactional state.
1803 pgstat_update_heap_dead_tuples(Relation rel, int delta)
1805 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1807 if (pgstat_info != NULL)
1808 pgstat_info->t_counts.t_delta_dead_tuples -= delta;
1815 * Called from access/transam/xact.c at top-level transaction commit/abort.
1819 AtEOXact_PgStat(bool isCommit)
1821 PgStat_SubXactStatus *xact_state;
1824 * Count transaction commit or abort. (We use counters, not just bools,
1825 * in case the reporting message isn't sent right away.)
1830 pgStatXactRollback++;
1833 * Transfer transactional insert/update counts into the base tabstat
1834 * entries. We don't bother to free any of the transactional state, since
1835 * it's all in TopTransactionContext and will go away anyway.
1837 xact_state = pgStatXactStack;
1838 if (xact_state != NULL)
1840 PgStat_TableXactStatus *trans;
1842 Assert(xact_state->nest_level == 1);
1843 Assert(xact_state->prev == NULL);
1844 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1846 PgStat_TableStatus *tabstat;
1848 Assert(trans->nest_level == 1);
1849 Assert(trans->upper == NULL);
1850 tabstat = trans->parent;
1851 Assert(tabstat->trans == trans);
1852 /* count attempted actions regardless of commit/abort */
1853 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
1854 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
1855 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
1858 /* insert adds a live tuple, delete removes one */
1859 tabstat->t_counts.t_delta_live_tuples +=
1860 trans->tuples_inserted - trans->tuples_deleted;
1861 /* update and delete each create a dead tuple */
1862 tabstat->t_counts.t_delta_dead_tuples +=
1863 trans->tuples_updated + trans->tuples_deleted;
1864 /* insert, update, delete each count as one change event */
1865 tabstat->t_counts.t_changed_tuples +=
1866 trans->tuples_inserted + trans->tuples_updated +
1867 trans->tuples_deleted;
1871 /* inserted tuples are dead, deleted tuples are unaffected */
1872 tabstat->t_counts.t_delta_dead_tuples +=
1873 trans->tuples_inserted + trans->tuples_updated;
1874 /* an aborted xact generates no changed_tuple events */
1876 tabstat->trans = NULL;
1879 pgStatXactStack = NULL;
1881 /* Make sure any stats snapshot is thrown away */
1882 pgstat_clear_snapshot();
1886 * AtEOSubXact_PgStat
1888 * Called from access/transam/xact.c at subtransaction commit/abort.
1892 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
1894 PgStat_SubXactStatus *xact_state;
1897 * Transfer transactional insert/update counts into the next higher
1898 * subtransaction state.
1900 xact_state = pgStatXactStack;
1901 if (xact_state != NULL &&
1902 xact_state->nest_level >= nestDepth)
1904 PgStat_TableXactStatus *trans;
1905 PgStat_TableXactStatus *next_trans;
1907 /* delink xact_state from stack immediately to simplify reuse case */
1908 pgStatXactStack = xact_state->prev;
1910 for (trans = xact_state->first; trans != NULL; trans = next_trans)
1912 PgStat_TableStatus *tabstat;
1914 next_trans = trans->next;
1915 Assert(trans->nest_level == nestDepth);
1916 tabstat = trans->parent;
1917 Assert(tabstat->trans == trans);
1920 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
1922 trans->upper->tuples_inserted += trans->tuples_inserted;
1923 trans->upper->tuples_updated += trans->tuples_updated;
1924 trans->upper->tuples_deleted += trans->tuples_deleted;
1925 tabstat->trans = trans->upper;
1931 * When there isn't an immediate parent state, we can just
1932 * reuse the record instead of going through a
1933 * palloc/pfree pushup (this works since it's all in
1934 * TopTransactionContext anyway). We have to re-link it
1935 * into the parent level, though, and that might mean
1936 * pushing a new entry into the pgStatXactStack.
1938 PgStat_SubXactStatus *upper_xact_state;
1940 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
1941 trans->next = upper_xact_state->first;
1942 upper_xact_state->first = trans;
1943 trans->nest_level = nestDepth - 1;
1949 * On abort, update top-level tabstat counts, then forget the
1953 /* count attempted actions regardless of commit/abort */
1954 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
1955 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
1956 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
1957 /* inserted tuples are dead, deleted tuples are unaffected */
1958 tabstat->t_counts.t_delta_dead_tuples +=
1959 trans->tuples_inserted + trans->tuples_updated;
1960 tabstat->trans = trans->upper;
1971 * Save the transactional stats state at 2PC transaction prepare.
1973 * In this phase we just generate 2PC records for all the pending
1974 * transaction-dependent stats work.
1977 AtPrepare_PgStat(void)
1979 PgStat_SubXactStatus *xact_state;
1981 xact_state = pgStatXactStack;
1982 if (xact_state != NULL)
1984 PgStat_TableXactStatus *trans;
1986 Assert(xact_state->nest_level == 1);
1987 Assert(xact_state->prev == NULL);
1988 for (trans = xact_state->first; trans != NULL; trans = trans->next)
1990 PgStat_TableStatus *tabstat;
1991 TwoPhasePgStatRecord record;
1993 Assert(trans->nest_level == 1);
1994 Assert(trans->upper == NULL);
1995 tabstat = trans->parent;
1996 Assert(tabstat->trans == trans);
1998 record.tuples_inserted = trans->tuples_inserted;
1999 record.tuples_updated = trans->tuples_updated;
2000 record.tuples_deleted = trans->tuples_deleted;
2001 record.t_id = tabstat->t_id;
2002 record.t_shared = tabstat->t_shared;
2004 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
2005 &record, sizeof(TwoPhasePgStatRecord));
2011 * PostPrepare_PgStat
2012 * Clean up after successful PREPARE.
2014 * All we need do here is unlink the transaction stats state from the
2015 * nontransactional state. The nontransactional action counts will be
2016 * reported to the stats collector immediately, while the effects on live
2017 * and dead tuple counts are preserved in the 2PC state file.
2019 * Note: AtEOXact_PgStat is not called during PREPARE.
2022 PostPrepare_PgStat(void)
2024 PgStat_SubXactStatus *xact_state;
2027 * We don't bother to free any of the transactional state, since it's all
2028 * in TopTransactionContext and will go away anyway.
2030 xact_state = pgStatXactStack;
2031 if (xact_state != NULL)
2033 PgStat_TableXactStatus *trans;
2035 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2037 PgStat_TableStatus *tabstat;
2039 tabstat = trans->parent;
2040 tabstat->trans = NULL;
2043 pgStatXactStack = NULL;
2045 /* Make sure any stats snapshot is thrown away */
2046 pgstat_clear_snapshot();
2050 * 2PC processing routine for COMMIT PREPARED case.
2052 * Load the saved counts into our local pgstats state.
2055 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
2056 void *recdata, uint32 len)
2058 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2059 PgStat_TableStatus *pgstat_info;
2061 /* Find or create a tabstat entry for the rel */
2062 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2064 /* Same math as in AtEOXact_PgStat, commit case */
2065 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2066 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2067 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2068 pgstat_info->t_counts.t_delta_live_tuples +=
2069 rec->tuples_inserted - rec->tuples_deleted;
2070 pgstat_info->t_counts.t_delta_dead_tuples +=
2071 rec->tuples_updated + rec->tuples_deleted;
2072 pgstat_info->t_counts.t_changed_tuples +=
2073 rec->tuples_inserted + rec->tuples_updated +
2074 rec->tuples_deleted;
2078 * 2PC processing routine for ROLLBACK PREPARED case.
2080 * Load the saved counts into our local pgstats state, but treat them
2084 pgstat_twophase_postabort(TransactionId xid, uint16 info,
2085 void *recdata, uint32 len)
2087 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2088 PgStat_TableStatus *pgstat_info;
2090 /* Find or create a tabstat entry for the rel */
2091 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2093 /* Same math as in AtEOXact_PgStat, abort case */
2094 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2095 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2096 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2097 pgstat_info->t_counts.t_delta_dead_tuples +=
2098 rec->tuples_inserted + rec->tuples_updated;
2103 * pgstat_fetch_stat_dbentry() -
2105 * Support function for the SQL-callable pgstat* functions. Returns
2106 * the collected statistics for one database or NULL. NULL doesn't mean
2107 * that the database doesn't exist, it is just not yet known by the
2108 * collector, so the caller is better off to report ZERO instead.
2111 PgStat_StatDBEntry *
2112 pgstat_fetch_stat_dbentry(Oid dbid)
2115 * If not done for this transaction, read the statistics collector stats
2116 * file into some hash tables.
2118 backend_read_statsfile();
2121 * Lookup the requested database; return NULL if not found
2123 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2130 * pgstat_fetch_stat_tabentry() -
2132 * Support function for the SQL-callable pgstat* functions. Returns
2133 * the collected statistics for one table or NULL. NULL doesn't mean
2134 * that the table doesn't exist, it is just not yet known by the
2135 * collector, so the caller is better off to report ZERO instead.
2138 PgStat_StatTabEntry *
2139 pgstat_fetch_stat_tabentry(Oid relid)
2142 PgStat_StatDBEntry *dbentry;
2143 PgStat_StatTabEntry *tabentry;
2146 * If not done for this transaction, read the statistics collector stats
2147 * file into some hash tables.
2149 backend_read_statsfile();
2152 * Lookup our database, then look in its table hash table.
2154 dbid = MyDatabaseId;
2155 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2158 if (dbentry != NULL && dbentry->tables != NULL)
2160 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2168 * If we didn't find it, maybe it's a shared table.
2171 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2174 if (dbentry != NULL && dbentry->tables != NULL)
2176 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2188 * pgstat_fetch_stat_funcentry() -
2190 * Support function for the SQL-callable pgstat* functions. Returns
2191 * the collected statistics for one function or NULL.
2194 PgStat_StatFuncEntry *
2195 pgstat_fetch_stat_funcentry(Oid func_id)
2197 PgStat_StatDBEntry *dbentry;
2198 PgStat_StatFuncEntry *funcentry = NULL;
2200 /* load the stats file if needed */
2201 backend_read_statsfile();
2203 /* Lookup our database, then find the requested function. */
2204 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
2205 if (dbentry != NULL && dbentry->functions != NULL)
2207 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2217 * pgstat_fetch_stat_beentry() -
2219 * Support function for the SQL-callable pgstat* functions. Returns
2220 * our local copy of the current-activity entry for one backend.
2222 * NB: caller is responsible for a check if the user is permitted to see
2223 * this info (especially the querystring).
2227 pgstat_fetch_stat_beentry(int beid)
2229 pgstat_read_current_status();
2231 if (beid < 1 || beid > localNumBackends)
2234 return &localBackendStatusTable[beid - 1];
2239 * pgstat_fetch_stat_numbackends() -
2241 * Support function for the SQL-callable pgstat* functions. Returns
2242 * the maximum current backend id.
2246 pgstat_fetch_stat_numbackends(void)
2248 pgstat_read_current_status();
2250 return localNumBackends;
2255 * pgstat_fetch_global() -
2257 * Support function for the SQL-callable pgstat* functions. Returns
2258 * a pointer to the global statistics struct.
2261 PgStat_GlobalStats *
2262 pgstat_fetch_global(void)
2264 backend_read_statsfile();
2266 return &globalStats;
2270 /* ------------------------------------------------------------
2271 * Functions for management of the shared-memory PgBackendStatus array
2272 * ------------------------------------------------------------
2275 static PgBackendStatus *BackendStatusArray = NULL;
2276 static PgBackendStatus *MyBEEntry = NULL;
2277 static char *BackendClientHostnameBuffer = NULL;
2278 static char *BackendAppnameBuffer = NULL;
2279 static char *BackendActivityBuffer = NULL;
2280 static Size BackendActivityBufferSize = 0;
2284 * Report shared-memory space needed by CreateSharedBackendStatus.
2287 BackendStatusShmemSize(void)
2291 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2292 size = add_size(size,
2293 mul_size(NAMEDATALEN, MaxBackends));
2294 size = add_size(size,
2295 mul_size(pgstat_track_activity_query_size, MaxBackends));
2296 size = add_size(size,
2297 mul_size(NAMEDATALEN, MaxBackends));
2302 * Initialize the shared status array and several string buffers
2303 * during postmaster startup.
2306 CreateSharedBackendStatus(void)
2313 /* Create or attach to the shared array */
2314 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2315 BackendStatusArray = (PgBackendStatus *)
2316 ShmemInitStruct("Backend Status Array", size, &found);
2321 * We're the first - initialize.
2323 MemSet(BackendStatusArray, 0, size);
2326 /* Create or attach to the shared appname buffer */
2327 size = mul_size(NAMEDATALEN, MaxBackends);
2328 BackendAppnameBuffer = (char *)
2329 ShmemInitStruct("Backend Application Name Buffer", size, &found);
2333 MemSet(BackendAppnameBuffer, 0, size);
2335 /* Initialize st_appname pointers. */
2336 buffer = BackendAppnameBuffer;
2337 for (i = 0; i < MaxBackends; i++)
2339 BackendStatusArray[i].st_appname = buffer;
2340 buffer += NAMEDATALEN;
2344 /* Create or attach to the shared client hostname buffer */
2345 size = mul_size(NAMEDATALEN, MaxBackends);
2346 BackendClientHostnameBuffer = (char *)
2347 ShmemInitStruct("Backend Client Host Name Buffer", size, &found);
2351 MemSet(BackendClientHostnameBuffer, 0, size);
2353 /* Initialize st_clienthostname pointers. */
2354 buffer = BackendClientHostnameBuffer;
2355 for (i = 0; i < MaxBackends; i++)
2357 BackendStatusArray[i].st_clienthostname = buffer;
2358 buffer += NAMEDATALEN;
2362 /* Create or attach to the shared activity buffer */
2363 BackendActivityBufferSize = mul_size(pgstat_track_activity_query_size,
2365 BackendActivityBuffer = (char *)
2366 ShmemInitStruct("Backend Activity Buffer",
2367 BackendActivityBufferSize,
2372 MemSet(BackendActivityBuffer, 0, size);
2374 /* Initialize st_activity pointers. */
2375 buffer = BackendActivityBuffer;
2376 for (i = 0; i < MaxBackends; i++)
2378 BackendStatusArray[i].st_activity = buffer;
2379 buffer += pgstat_track_activity_query_size;
2386 * pgstat_initialize() -
2388 * Initialize pgstats state, and set up our on-proc-exit hook.
2389 * Called from InitPostgres. MyBackendId must be set,
2390 * but we must not have started any transaction yet (since the
2391 * exit hook must run after the last transaction exit).
2392 * NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful.
2396 pgstat_initialize(void)
2398 /* Initialize MyBEEntry */
2399 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2400 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2402 /* Set up a process-exit hook to clean up */
2403 on_shmem_exit(pgstat_beshutdown_hook, 0);
2407 * pgstat_bestart() -
2409 * Initialize this backend's entry in the PgBackendStatus array.
2410 * Called from InitPostgres.
2411 * MyDatabaseId, session userid, and application_name must be set
2412 * (hence, this cannot be combined with pgstat_initialize).
2416 pgstat_bestart(void)
2418 TimestampTz proc_start_timestamp;
2420 SockAddr clientaddr;
2421 volatile PgBackendStatus *beentry;
2424 * To minimize the time spent modifying the PgBackendStatus entry, fetch
2425 * all the needed data first.
2427 * If we have a MyProcPort, use its session start time (for consistency,
2428 * and to save a kernel call).
2431 proc_start_timestamp = MyProcPort->SessionStartTime;
2433 proc_start_timestamp = GetCurrentTimestamp();
2434 userid = GetSessionUserId();
2437 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2438 * If so, use all-zeroes client address, which is dealt with specially in
2439 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2442 memcpy(&clientaddr, &MyProcPort->raddr, sizeof(clientaddr));
2444 MemSet(&clientaddr, 0, sizeof(clientaddr));
2447 * Initialize my status entry, following the protocol of bumping
2448 * st_changecount before and after; and make sure it's even afterwards. We
2449 * use a volatile pointer here to ensure the compiler doesn't try to get
2452 beentry = MyBEEntry;
2455 beentry->st_changecount++;
2456 } while ((beentry->st_changecount & 1) == 0);
2458 beentry->st_procpid = MyProcPid;
2459 beentry->st_proc_start_timestamp = proc_start_timestamp;
2460 beentry->st_activity_start_timestamp = 0;
2461 beentry->st_state_start_timestamp = 0;
2462 beentry->st_xact_start_timestamp = 0;
2463 beentry->st_databaseid = MyDatabaseId;
2464 beentry->st_userid = userid;
2465 beentry->st_clientaddr = clientaddr;
2466 beentry->st_clienthostname[0] = '\0';
2467 beentry->st_waiting = false;
2468 beentry->st_state = STATE_UNDEFINED;
2469 beentry->st_appname[0] = '\0';
2470 beentry->st_activity[0] = '\0';
2471 /* Also make sure the last byte in each string area is always 0 */
2472 beentry->st_clienthostname[NAMEDATALEN - 1] = '\0';
2473 beentry->st_appname[NAMEDATALEN - 1] = '\0';
2474 beentry->st_activity[pgstat_track_activity_query_size - 1] = '\0';
2476 beentry->st_changecount++;
2477 Assert((beentry->st_changecount & 1) == 0);
2479 if (MyProcPort && MyProcPort->remote_hostname)
2480 strlcpy(beentry->st_clienthostname, MyProcPort->remote_hostname, NAMEDATALEN);
2482 /* Update app name to current GUC setting */
2483 if (application_name)
2484 pgstat_report_appname(application_name);
2488 * Shut down a single backend's statistics reporting at process exit.
2490 * Flush any remaining statistics counts out to the collector.
2491 * Without this, operations triggered during backend exit (such as
2492 * temp table deletions) won't be counted.
2494 * Lastly, clear out our entry in the PgBackendStatus array.
2497 pgstat_beshutdown_hook(int code, Datum arg)
2499 volatile PgBackendStatus *beentry = MyBEEntry;
2502 * If we got as far as discovering our own database ID, we can report what
2503 * we did to the collector. Otherwise, we'd be sending an invalid
2504 * database ID, so forget it. (This means that accesses to pg_database
2505 * during failed backend starts might never get counted.)
2507 if (OidIsValid(MyDatabaseId))
2508 pgstat_report_stat(true);
2511 * Clear my status entry, following the protocol of bumping st_changecount
2512 * before and after. We use a volatile pointer here to ensure the
2513 * compiler doesn't try to get cute.
2515 beentry->st_changecount++;
2517 beentry->st_procpid = 0; /* mark invalid */
2519 beentry->st_changecount++;
2520 Assert((beentry->st_changecount & 1) == 0);
2525 * pgstat_report_activity() -
2527 * Called from tcop/postgres.c to report what the backend is actually doing
2528 * (usually "<IDLE>" or the start of the query to be executed).
2530 * All updates of the status entry follow the protocol of bumping
2531 * st_changecount before and after. We use a volatile pointer here to
2532 * ensure the compiler doesn't try to get cute.
2536 pgstat_report_activity(BackendState state, const char *cmd_str)
2538 volatile PgBackendStatus *beentry = MyBEEntry;
2539 TimestampTz start_timestamp;
2540 TimestampTz current_timestamp;
2543 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
2549 * To minimize the time spent modifying the entry, fetch all the needed
2552 current_timestamp = GetCurrentTimestamp();
2554 if (!pgstat_track_activities && beentry->st_state != STATE_DISABLED)
2557 * Track activities is disabled, but we have a non-disabled state set.
2558 * That means the status changed - so as our last update, tell the
2559 * collector that we disabled it and will no longer update.
2561 beentry->st_changecount++;
2562 beentry->st_state = STATE_DISABLED;
2563 beentry->st_state_start_timestamp = current_timestamp;
2564 beentry->st_changecount++;
2565 Assert((beentry->st_changecount & 1) == 0);
2570 * Fetch more data before we start modifying the entry
2572 start_timestamp = GetCurrentStatementStartTimestamp();
2573 if (cmd_str != NULL)
2575 len = pg_mbcliplen(cmd_str, strlen(cmd_str),
2576 pgstat_track_activity_query_size - 1);
2580 * Now update the status entry
2582 beentry->st_changecount++;
2584 beentry->st_state = state;
2585 beentry->st_state_start_timestamp = current_timestamp;
2587 if (cmd_str != NULL)
2589 memcpy((char *) beentry->st_activity, cmd_str, len);
2590 beentry->st_activity[len] = '\0';
2591 beentry->st_activity_start_timestamp = start_timestamp;
2594 beentry->st_changecount++;
2595 Assert((beentry->st_changecount & 1) == 0);
2599 * pgstat_report_appname() -
2601 * Called to update our application name.
2605 pgstat_report_appname(const char *appname)
2607 volatile PgBackendStatus *beentry = MyBEEntry;
2613 /* This should be unnecessary if GUC did its job, but be safe */
2614 len = pg_mbcliplen(appname, strlen(appname), NAMEDATALEN - 1);
2617 * Update my status entry, following the protocol of bumping
2618 * st_changecount before and after. We use a volatile pointer here to
2619 * ensure the compiler doesn't try to get cute.
2621 beentry->st_changecount++;
2623 memcpy((char *) beentry->st_appname, appname, len);
2624 beentry->st_appname[len] = '\0';
2626 beentry->st_changecount++;
2627 Assert((beentry->st_changecount & 1) == 0);
2631 * Report current transaction start timestamp as the specified value.
2632 * Zero means there is no active transaction.
2635 pgstat_report_xact_timestamp(TimestampTz tstamp)
2637 volatile PgBackendStatus *beentry = MyBEEntry;
2639 if (!pgstat_track_activities || !beentry)
2643 * Update my status entry, following the protocol of bumping
2644 * st_changecount before and after. We use a volatile pointer here to
2645 * ensure the compiler doesn't try to get cute.
2647 beentry->st_changecount++;
2648 beentry->st_xact_start_timestamp = tstamp;
2649 beentry->st_changecount++;
2650 Assert((beentry->st_changecount & 1) == 0);
2654 * pgstat_report_waiting() -
2656 * Called from lock manager to report beginning or end of a lock wait.
2658 * NB: this *must* be able to survive being called before MyBEEntry has been
2663 pgstat_report_waiting(bool waiting)
2665 volatile PgBackendStatus *beentry = MyBEEntry;
2667 if (!pgstat_track_activities || !beentry)
2671 * Since this is a single-byte field in a struct that only this process
2672 * may modify, there seems no need to bother with the st_changecount
2673 * protocol. The update must appear atomic in any case.
2675 beentry->st_waiting = waiting;
2680 * pgstat_read_current_status() -
2682 * Copy the current contents of the PgBackendStatus array to local memory,
2683 * if not already done in this transaction.
2687 pgstat_read_current_status(void)
2689 volatile PgBackendStatus *beentry;
2690 PgBackendStatus *localtable;
2691 PgBackendStatus *localentry;
2696 Assert(!pgStatRunningInCollector);
2697 if (localBackendStatusTable)
2698 return; /* already done */
2700 pgstat_setup_memcxt();
2702 localtable = (PgBackendStatus *)
2703 MemoryContextAlloc(pgStatLocalContext,
2704 sizeof(PgBackendStatus) * MaxBackends);
2705 localappname = (char *)
2706 MemoryContextAlloc(pgStatLocalContext,
2707 NAMEDATALEN * MaxBackends);
2708 localactivity = (char *)
2709 MemoryContextAlloc(pgStatLocalContext,
2710 pgstat_track_activity_query_size * MaxBackends);
2711 localNumBackends = 0;
2713 beentry = BackendStatusArray;
2714 localentry = localtable;
2715 for (i = 1; i <= MaxBackends; i++)
2718 * Follow the protocol of retrying if st_changecount changes while we
2719 * copy the entry, or if it's odd. (The check for odd is needed to
2720 * cover the case where we are able to completely copy the entry while
2721 * the source backend is between increment steps.) We use a volatile
2722 * pointer here to ensure the compiler doesn't try to get cute.
2726 int save_changecount = beentry->st_changecount;
2728 localentry->st_procpid = beentry->st_procpid;
2729 if (localentry->st_procpid > 0)
2731 memcpy(localentry, (char *) beentry, sizeof(PgBackendStatus));
2734 * strcpy is safe even if the string is modified concurrently,
2735 * because there's always a \0 at the end of the buffer.
2737 strcpy(localappname, (char *) beentry->st_appname);
2738 localentry->st_appname = localappname;
2739 strcpy(localactivity, (char *) beentry->st_activity);
2740 localentry->st_activity = localactivity;
2743 if (save_changecount == beentry->st_changecount &&
2744 (save_changecount & 1) == 0)
2747 /* Make sure we can break out of loop if stuck... */
2748 CHECK_FOR_INTERRUPTS();
2752 /* Only valid entries get included into the local array */
2753 if (localentry->st_procpid > 0)
2756 localappname += NAMEDATALEN;
2757 localactivity += pgstat_track_activity_query_size;
2762 /* Set the pointer only after completion of a valid table */
2763 localBackendStatusTable = localtable;
2768 * pgstat_get_backend_current_activity() -
2770 * Return a string representing the current activity of the backend with
2771 * the specified PID. This looks directly at the BackendStatusArray,
2772 * and so will provide current information regardless of the age of our
2773 * transaction's snapshot of the status array.
2775 * It is the caller's responsibility to invoke this only for backends whose
2776 * state is expected to remain stable while the result is in use. The
2777 * only current use is in deadlock reporting, where we can expect that
2778 * the target backend is blocked on a lock. (There are corner cases
2779 * where the target's wait could get aborted while we are looking at it,
2780 * but the very worst consequence is to return a pointer to a string
2781 * that's been changed, so we won't worry too much.)
2783 * Note: return strings for special cases match pg_stat_get_backend_activity.
2787 pgstat_get_backend_current_activity(int pid, bool checkUser)
2789 PgBackendStatus *beentry;
2792 beentry = BackendStatusArray;
2793 for (i = 1; i <= MaxBackends; i++)
2796 * Although we expect the target backend's entry to be stable, that
2797 * doesn't imply that anyone else's is. To avoid identifying the
2798 * wrong backend, while we check for a match to the desired PID we
2799 * must follow the protocol of retrying if st_changecount changes
2800 * while we examine the entry, or if it's odd. (This might be
2801 * unnecessary, since fetching or storing an int is almost certainly
2802 * atomic, but let's play it safe.) We use a volatile pointer here to
2803 * ensure the compiler doesn't try to get cute.
2805 volatile PgBackendStatus *vbeentry = beentry;
2810 int save_changecount = vbeentry->st_changecount;
2812 found = (vbeentry->st_procpid == pid);
2814 if (save_changecount == vbeentry->st_changecount &&
2815 (save_changecount & 1) == 0)
2818 /* Make sure we can break out of loop if stuck... */
2819 CHECK_FOR_INTERRUPTS();
2824 /* Now it is safe to use the non-volatile pointer */
2825 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
2826 return "<insufficient privilege>";
2827 else if (*(beentry->st_activity) == '\0')
2828 return "<command string not enabled>";
2830 return beentry->st_activity;
2836 /* If we get here, caller is in error ... */
2837 return "<backend information not available>";
2841 * pgstat_get_crashed_backend_activity() -
2843 * Return a string representing the current activity of the backend with
2844 * the specified PID. Like the function above, but reads shared memory with
2845 * the expectation that it may be corrupt. On success, copy the string
2846 * into the "buffer" argument and return that pointer. On failure,
2849 * This function is only intended to be used by the postmaster to report the
2850 * query that crashed a backend. In particular, no attempt is made to
2851 * follow the correct concurrency protocol when accessing the
2852 * BackendStatusArray. But that's OK, in the worst case we'll return a
2853 * corrupted message. We also must take care not to trip on ereport(ERROR).
2857 pgstat_get_crashed_backend_activity(int pid, char *buffer, int buflen)
2859 volatile PgBackendStatus *beentry;
2862 beentry = BackendStatusArray;
2865 * We probably shouldn't get here before shared memory has been set up,
2868 if (beentry == NULL || BackendActivityBuffer == NULL)
2871 for (i = 1; i <= MaxBackends; i++)
2873 if (beentry->st_procpid == pid)
2875 /* Read pointer just once, so it can't change after validation */
2876 const char *activity = beentry->st_activity;
2877 const char *activity_last;
2880 * We mustn't access activity string before we verify that it
2881 * falls within the BackendActivityBuffer. To make sure that the
2882 * entire string including its ending is contained within the
2883 * buffer, subtract one activity length from the buffer size.
2885 activity_last = BackendActivityBuffer + BackendActivityBufferSize
2886 - pgstat_track_activity_query_size;
2888 if (activity < BackendActivityBuffer ||
2889 activity > activity_last)
2892 /* If no string available, no point in a report */
2893 if (activity[0] == '\0')
2897 * Copy only ASCII-safe characters so we don't run into encoding
2898 * problems when reporting the message; and be sure not to run
2899 * off the end of memory.
2901 ascii_safe_strlcpy(buffer, activity,
2902 Min(buflen, pgstat_track_activity_query_size));
2915 /* ------------------------------------------------------------
2916 * Local support functions follow
2917 * ------------------------------------------------------------
2922 * pgstat_setheader() -
2924 * Set common header fields in a statistics message
2928 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
2930 hdr->m_type = mtype;
2937 * Send out one statistics message to the collector
2941 pgstat_send(void *msg, int len)
2945 if (pgStatSock == PGINVALID_SOCKET)
2948 ((PgStat_MsgHdr *) msg)->m_size = len;
2950 /* We'll retry after EINTR, but ignore all other failures */
2953 rc = send(pgStatSock, msg, len, 0);
2954 } while (rc < 0 && errno == EINTR);
2956 #ifdef USE_ASSERT_CHECKING
2957 /* In debug builds, log send failures ... */
2959 elog(LOG, "could not send to statistics collector: %m");
2964 * pgstat_send_bgwriter() -
2966 * Send bgwriter statistics to the collector
2970 pgstat_send_bgwriter(void)
2972 /* We assume this initializes to zeroes */
2973 static const PgStat_MsgBgWriter all_zeroes;
2976 * This function can be called even if nothing at all has happened. In
2977 * this case, avoid sending a completely empty message to the stats
2980 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
2984 * Prepare and send the message
2986 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
2987 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
2990 * Clear out the statistics buffer, so it can be re-used.
2992 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
2997 * PgstatCollectorMain() -
2999 * Start up the statistics collector process. This is the body of the
3000 * postmaster child process.
3002 * The argc/argv parameters are valid only in EXEC_BACKEND case.
3005 NON_EXEC_STATIC void
3006 PgstatCollectorMain(int argc, char *argv[])
3013 struct pollfd input_fd;
3015 struct timeval sel_timeout;
3020 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
3022 MyProcPid = getpid(); /* reset MyProcPid */
3024 MyStartTime = time(NULL); /* record Start Time for logging */
3027 * If possible, make this process a group leader, so that the postmaster
3028 * can signal any child processes too. (pgstat probably never has any
3029 * child processes, but for consistency we make all postmaster child
3030 * processes do this.)
3034 elog(FATAL, "setsid() failed: %m");
3038 * Ignore all signals usually bound to some action in the postmaster,
3041 pqsignal(SIGHUP, pgstat_sighup_handler);
3042 pqsignal(SIGINT, SIG_IGN);
3043 pqsignal(SIGTERM, SIG_IGN);
3044 pqsignal(SIGQUIT, pgstat_exit);
3045 pqsignal(SIGALRM, SIG_IGN);
3046 pqsignal(SIGPIPE, SIG_IGN);
3047 pqsignal(SIGUSR1, SIG_IGN);
3048 pqsignal(SIGUSR2, SIG_IGN);
3049 pqsignal(SIGCHLD, SIG_DFL);
3050 pqsignal(SIGTTIN, SIG_DFL);
3051 pqsignal(SIGTTOU, SIG_DFL);
3052 pqsignal(SIGCONT, SIG_DFL);
3053 pqsignal(SIGWINCH, SIG_DFL);
3054 PG_SETMASK(&UnBlockSig);
3057 * Identify myself via ps
3059 init_ps_display("stats collector process", "", "", "");
3062 * Arrange to write the initial status file right away
3064 last_statrequest = GetCurrentTimestamp();
3065 last_statwrite = last_statrequest - 1;
3068 * Read in an existing statistics stats file or initialize the stats to
3071 pgStatRunningInCollector = true;
3072 pgStatDBHash = pgstat_read_statsfile(InvalidOid, true);
3075 * Setup the descriptor set for select(2). Since only one bit in the set
3076 * ever changes, we need not repeat FD_ZERO each time.
3078 #if !defined(HAVE_POLL) && !defined(WIN32)
3083 * Loop to process messages until we get SIGQUIT or detect ungraceful
3084 * death of our parent postmaster.
3086 * For performance reasons, we don't want to do a PostmasterIsAlive() test
3087 * after every message; instead, do it only when select()/poll() is
3088 * interrupted by timeout. In essence, we'll stay alive as long as
3089 * backends keep sending us stuff often, even if the postmaster is gone.
3096 * Quit if we get SIGQUIT from the postmaster.
3102 * Reload configuration if we got SIGHUP from the postmaster.
3106 ProcessConfigFile(PGC_SIGHUP);
3111 * Write the stats file if a new request has arrived that is not
3112 * satisfied by existing file.
3114 if (last_statwrite < last_statrequest)
3115 pgstat_write_statsfile(false);
3118 * Wait for a message to arrive; but not for more than
3119 * PGSTAT_SELECT_TIMEOUT seconds. (This determines how quickly we will
3120 * shut down after an ungraceful postmaster termination; so it needn't
3121 * be very fast. However, on some systems SIGQUIT won't interrupt the
3122 * poll/select call, so this also limits speed of response to SIGQUIT,
3123 * which is more important.)
3125 * We use poll(2) if available, otherwise select(2). Win32 has its own
3130 input_fd.fd = pgStatSock;
3131 input_fd.events = POLLIN | POLLERR;
3132 input_fd.revents = 0;
3134 if (poll(&input_fd, 1, PGSTAT_SELECT_TIMEOUT * 1000) < 0)
3139 (errcode_for_socket_access(),
3140 errmsg("poll() failed in statistics collector: %m")));
3143 got_data = (input_fd.revents != 0);
3144 #else /* !HAVE_POLL */
3146 FD_SET(pgStatSock, &rfds);
3149 * timeout struct is modified by select() on some operating systems,
3150 * so re-fill it each time.
3152 sel_timeout.tv_sec = PGSTAT_SELECT_TIMEOUT;
3153 sel_timeout.tv_usec = 0;
3155 if (select(pgStatSock + 1, &rfds, NULL, NULL, &sel_timeout) < 0)
3160 (errcode_for_socket_access(),
3161 errmsg("select() failed in statistics collector: %m")));
3164 got_data = FD_ISSET(pgStatSock, &rfds);
3165 #endif /* HAVE_POLL */
3167 got_data = pgwin32_waitforsinglesocket(pgStatSock, FD_READ,
3168 PGSTAT_SELECT_TIMEOUT * 1000);
3172 * If there is a message on the socket, read it and check for
3177 len = recv(pgStatSock, (char *) &msg,
3178 sizeof(PgStat_Msg), 0);
3184 (errcode_for_socket_access(),
3185 errmsg("could not read statistics message: %m")));
3189 * We ignore messages that are smaller than our common header
3191 if (len < sizeof(PgStat_MsgHdr))
3195 * The received length must match the length in the header
3197 if (msg.msg_hdr.m_size != len)
3201 * O.K. - we accept this message. Process it.
3203 switch (msg.msg_hdr.m_type)
3205 case PGSTAT_MTYPE_DUMMY:
3208 case PGSTAT_MTYPE_INQUIRY:
3209 pgstat_recv_inquiry((PgStat_MsgInquiry *) &msg, len);
3212 case PGSTAT_MTYPE_TABSTAT:
3213 pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len);
3216 case PGSTAT_MTYPE_TABPURGE:
3217 pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len);
3220 case PGSTAT_MTYPE_DROPDB:
3221 pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len);
3224 case PGSTAT_MTYPE_RESETCOUNTER:
3225 pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg,
3229 case PGSTAT_MTYPE_RESETSHAREDCOUNTER:
3230 pgstat_recv_resetsharedcounter(
3231 (PgStat_MsgResetsharedcounter *) &msg,
3235 case PGSTAT_MTYPE_RESETSINGLECOUNTER:
3236 pgstat_recv_resetsinglecounter(
3237 (PgStat_MsgResetsinglecounter *) &msg,
3241 case PGSTAT_MTYPE_AUTOVAC_START:
3242 pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
3245 case PGSTAT_MTYPE_VACUUM:
3246 pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
3249 case PGSTAT_MTYPE_ANALYZE:
3250 pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len);
3253 case PGSTAT_MTYPE_BGWRITER:
3254 pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len);
3257 case PGSTAT_MTYPE_FUNCSTAT:
3258 pgstat_recv_funcstat((PgStat_MsgFuncstat *) &msg, len);
3261 case PGSTAT_MTYPE_FUNCPURGE:
3262 pgstat_recv_funcpurge((PgStat_MsgFuncpurge *) &msg, len);
3265 case PGSTAT_MTYPE_RECOVERYCONFLICT:
3266 pgstat_recv_recoveryconflict((PgStat_MsgRecoveryConflict *) &msg, len);
3269 case PGSTAT_MTYPE_DEADLOCK:
3270 pgstat_recv_deadlock((PgStat_MsgDeadlock *) &msg, len);
3273 case PGSTAT_MTYPE_TEMPFILE:
3274 pgstat_recv_tempfile((PgStat_MsgTempFile *) &msg, len);
3284 * We can only get here if the select/poll timeout elapsed. Check
3285 * for postmaster death.
3287 if (!PostmasterIsAlive())
3290 } /* end of message-processing loop */
3293 * Save the final stats to reuse at next startup.
3295 pgstat_write_statsfile(true);
3301 /* SIGQUIT signal handler for collector process */
3303 pgstat_exit(SIGNAL_ARGS)
3308 /* SIGHUP handler for collector process */
3310 pgstat_sighup_handler(SIGNAL_ARGS)
3317 * Lookup the hash table entry for the specified database. If no hash
3318 * table entry exists, initialize it, if the create parameter is true.
3319 * Else, return NULL.
3321 static PgStat_StatDBEntry *
3322 pgstat_get_db_entry(Oid databaseid, bool create)
3324 PgStat_StatDBEntry *result;
3326 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3328 /* Lookup or create the hash table entry for this database */
3329 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
3333 if (!create && !found)
3336 /* If not found, initialize the new one. */
3341 result->tables = NULL;
3342 result->functions = NULL;
3343 result->n_xact_commit = 0;
3344 result->n_xact_rollback = 0;
3345 result->n_blocks_fetched = 0;
3346 result->n_blocks_hit = 0;
3347 result->n_tuples_returned = 0;
3348 result->n_tuples_fetched = 0;
3349 result->n_tuples_inserted = 0;
3350 result->n_tuples_updated = 0;
3351 result->n_tuples_deleted = 0;
3352 result->last_autovac_time = 0;
3353 result->n_conflict_tablespace = 0;
3354 result->n_conflict_lock = 0;
3355 result->n_conflict_snapshot = 0;
3356 result->n_conflict_bufferpin = 0;
3357 result->n_conflict_startup_deadlock = 0;
3358 result->n_temp_files = 0;
3359 result->n_temp_bytes = 0;
3360 result->n_deadlocks = 0;
3361 result->n_block_time_read = 0;
3362 result->n_block_time_write = 0;
3364 result->stat_reset_timestamp = GetCurrentTimestamp();
3366 memset(&hash_ctl, 0, sizeof(hash_ctl));
3367 hash_ctl.keysize = sizeof(Oid);
3368 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3369 hash_ctl.hash = oid_hash;
3370 result->tables = hash_create("Per-database table",
3371 PGSTAT_TAB_HASH_SIZE,
3373 HASH_ELEM | HASH_FUNCTION);
3375 hash_ctl.keysize = sizeof(Oid);
3376 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3377 hash_ctl.hash = oid_hash;
3378 result->functions = hash_create("Per-database function",
3379 PGSTAT_FUNCTION_HASH_SIZE,
3381 HASH_ELEM | HASH_FUNCTION);
3389 * Lookup the hash table entry for the specified table. If no hash
3390 * table entry exists, initialize it, if the create parameter is true.
3391 * Else, return NULL.
3393 static PgStat_StatTabEntry *
3394 pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
3396 PgStat_StatTabEntry *result;
3398 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3400 /* Lookup or create the hash table entry for this table */
3401 result = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3405 if (!create && !found)
3408 /* If not found, initialize the new one. */
3411 result->numscans = 0;
3412 result->tuples_returned = 0;
3413 result->tuples_fetched = 0;
3414 result->tuples_inserted = 0;
3415 result->tuples_updated = 0;
3416 result->tuples_deleted = 0;
3417 result->tuples_hot_updated = 0;
3418 result->n_live_tuples = 0;
3419 result->n_dead_tuples = 0;
3420 result->changes_since_analyze = 0;
3421 result->blocks_fetched = 0;
3422 result->blocks_hit = 0;
3423 result->vacuum_timestamp = 0;
3424 result->vacuum_count = 0;
3425 result->autovac_vacuum_timestamp = 0;
3426 result->autovac_vacuum_count = 0;
3427 result->analyze_timestamp = 0;
3428 result->analyze_count = 0;
3429 result->autovac_analyze_timestamp = 0;
3430 result->autovac_analyze_count = 0;
3438 * pgstat_write_statsfile() -
3441 * If writing to the permanent file (happens when the collector is
3442 * shutting down only), remove the temporary file so that backends
3443 * starting up under a new postmaster can't read the old data before
3444 * the new collector is ready.
3448 pgstat_write_statsfile(bool permanent)
3450 HASH_SEQ_STATUS hstat;
3451 HASH_SEQ_STATUS tstat;
3452 HASH_SEQ_STATUS fstat;
3453 PgStat_StatDBEntry *dbentry;
3454 PgStat_StatTabEntry *tabentry;
3455 PgStat_StatFuncEntry *funcentry;
3458 const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
3459 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3463 * Open the statistics temp file to write out the current values.
3465 fpout = AllocateFile(tmpfile, PG_BINARY_W);
3469 (errcode_for_file_access(),
3470 errmsg("could not open temporary statistics file \"%s\": %m",
3476 * Set the timestamp of the stats file.
3478 globalStats.stats_timestamp = GetCurrentTimestamp();
3481 * Write the file header --- currently just a format ID.
3483 format_id = PGSTAT_FILE_FORMAT_ID;
3484 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
3485 (void) rc; /* we'll check for error with ferror */
3488 * Write global stats struct
3490 rc = fwrite(&globalStats, sizeof(globalStats), 1, fpout);
3491 (void) rc; /* we'll check for error with ferror */
3494 * Walk through the database table.
3496 hash_seq_init(&hstat, pgStatDBHash);
3497 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
3500 * Write out the DB entry including the number of live backends. We
3501 * don't write the tables or functions pointers, since they're of no
3502 * use to any other process.
3505 rc = fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
3506 (void) rc; /* we'll check for error with ferror */
3509 * Walk through the database's access stats per table.
3511 hash_seq_init(&tstat, dbentry->tables);
3512 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
3515 rc = fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
3516 (void) rc; /* we'll check for error with ferror */
3520 * Walk through the database's function stats table.
3522 hash_seq_init(&fstat, dbentry->functions);
3523 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
3526 rc = fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
3527 (void) rc; /* we'll check for error with ferror */
3531 * Mark the end of this DB
3537 * No more output to be done. Close the temp file and replace the old
3538 * pgstat.stat with it. The ferror() check replaces testing for error
3539 * after each individual fputc or fwrite above.
3546 (errcode_for_file_access(),
3547 errmsg("could not write temporary statistics file \"%s\": %m",
3552 else if (FreeFile(fpout) < 0)
3555 (errcode_for_file_access(),
3556 errmsg("could not close temporary statistics file \"%s\": %m",
3560 else if (rename(tmpfile, statfile) < 0)
3563 (errcode_for_file_access(),
3564 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
3565 tmpfile, statfile)));
3571 * Successful write, so update last_statwrite.
3573 last_statwrite = globalStats.stats_timestamp;
3576 * If there is clock skew between backends and the collector, we could
3577 * receive a stats request time that's in the future. If so, complain
3578 * and reset last_statrequest. Resetting ensures that no inquiry
3579 * message can cause more than one stats file write to occur.
3581 if (last_statrequest > last_statwrite)
3586 /* Copy because timestamptz_to_str returns a static buffer */
3587 reqtime = pstrdup(timestamptz_to_str(last_statrequest));
3588 mytime = pstrdup(timestamptz_to_str(last_statwrite));
3589 elog(LOG, "last_statrequest %s is later than collector's time %s",
3594 last_statrequest = last_statwrite;
3599 unlink(pgstat_stat_filename);
3604 * pgstat_read_statsfile() -
3606 * Reads in an existing statistics collector file and initializes the
3607 * databases' hash table (whose entries point to the tables' hash tables).
3611 pgstat_read_statsfile(Oid onlydb, bool permanent)
3613 PgStat_StatDBEntry *dbentry;
3614 PgStat_StatDBEntry dbbuf;
3615 PgStat_StatTabEntry *tabentry;
3616 PgStat_StatTabEntry tabbuf;
3617 PgStat_StatFuncEntry funcbuf;
3618 PgStat_StatFuncEntry *funcentry;
3621 HTAB *tabhash = NULL;
3622 HTAB *funchash = NULL;
3626 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3629 * The tables will live in pgStatLocalContext.
3631 pgstat_setup_memcxt();
3634 * Create the DB hashtable
3636 memset(&hash_ctl, 0, sizeof(hash_ctl));
3637 hash_ctl.keysize = sizeof(Oid);
3638 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
3639 hash_ctl.hash = oid_hash;
3640 hash_ctl.hcxt = pgStatLocalContext;
3641 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
3642 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3645 * Clear out global statistics so they start from zero in case we can't
3646 * load an existing statsfile.
3648 memset(&globalStats, 0, sizeof(globalStats));
3651 * Set the current timestamp (will be kept only in case we can't load an
3652 * existing statsfile.
3654 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
3657 * Try to open the status file. If it doesn't exist, the backends simply
3658 * return zero for anything and the collector simply starts from scratch
3659 * with empty counters.
3661 * ENOENT is a possibility if the stats collector is not running or has
3662 * not yet written the stats file the first time. Any other failure
3663 * condition is suspicious.
3665 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3667 if (errno != ENOENT)
3668 ereport(pgStatRunningInCollector ? LOG : WARNING,
3669 (errcode_for_file_access(),
3670 errmsg("could not open statistics file \"%s\": %m",
3676 * Verify it's of the expected format.
3678 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3679 || format_id != PGSTAT_FILE_FORMAT_ID)
3681 ereport(pgStatRunningInCollector ? LOG : WARNING,
3682 (errmsg("corrupted statistics file \"%s\"", statfile)));
3687 * Read global stats struct
3689 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
3691 ereport(pgStatRunningInCollector ? LOG : WARNING,
3692 (errmsg("corrupted statistics file \"%s\"", statfile)));
3697 * We found an existing collector stats file. Read it and put all the
3698 * hashtable entries into place.
3702 switch (fgetc(fpin))
3705 * 'D' A PgStat_StatDBEntry struct describing a database
3706 * follows. Subsequently, zero to many 'T' and 'F' entries
3707 * will follow until a 'd' is encountered.
3710 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
3711 fpin) != offsetof(PgStat_StatDBEntry, tables))
3713 ereport(pgStatRunningInCollector ? LOG : WARNING,
3714 (errmsg("corrupted statistics file \"%s\"",
3720 * Add to the DB hash
3722 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
3723 (void *) &dbbuf.databaseid,
3728 ereport(pgStatRunningInCollector ? LOG : WARNING,
3729 (errmsg("corrupted statistics file \"%s\"",
3734 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
3735 dbentry->tables = NULL;
3736 dbentry->functions = NULL;
3739 * Don't collect tables if not the requested DB (or the
3740 * shared-table info)
3742 if (onlydb != InvalidOid)
3744 if (dbbuf.databaseid != onlydb &&
3745 dbbuf.databaseid != InvalidOid)
3749 memset(&hash_ctl, 0, sizeof(hash_ctl));
3750 hash_ctl.keysize = sizeof(Oid);
3751 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3752 hash_ctl.hash = oid_hash;
3753 hash_ctl.hcxt = pgStatLocalContext;
3754 dbentry->tables = hash_create("Per-database table",
3755 PGSTAT_TAB_HASH_SIZE,
3757 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3759 hash_ctl.keysize = sizeof(Oid);
3760 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3761 hash_ctl.hash = oid_hash;
3762 hash_ctl.hcxt = pgStatLocalContext;
3763 dbentry->functions = hash_create("Per-database function",
3764 PGSTAT_FUNCTION_HASH_SIZE,
3766 HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
3769 * Arrange that following records add entries to this
3770 * database's hash tables.
3772 tabhash = dbentry->tables;
3773 funchash = dbentry->functions;
3777 * 'd' End of this database.
3785 * 'T' A PgStat_StatTabEntry follows.
3788 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
3789 fpin) != sizeof(PgStat_StatTabEntry))
3791 ereport(pgStatRunningInCollector ? LOG : WARNING,
3792 (errmsg("corrupted statistics file \"%s\"",
3798 * Skip if table belongs to a not requested database.
3800 if (tabhash == NULL)
3803 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
3804 (void *) &tabbuf.tableid,
3805 HASH_ENTER, &found);
3809 ereport(pgStatRunningInCollector ? LOG : WARNING,
3810 (errmsg("corrupted statistics file \"%s\"",
3815 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
3819 * 'F' A PgStat_StatFuncEntry follows.
3822 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
3823 fpin) != sizeof(PgStat_StatFuncEntry))
3825 ereport(pgStatRunningInCollector ? LOG : WARNING,
3826 (errmsg("corrupted statistics file \"%s\"",
3832 * Skip if function belongs to a not requested database.
3834 if (funchash == NULL)
3837 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
3838 (void *) &funcbuf.functionid,
3839 HASH_ENTER, &found);
3843 ereport(pgStatRunningInCollector ? LOG : WARNING,
3844 (errmsg("corrupted statistics file \"%s\"",
3849 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
3853 * 'E' The EOF marker of a complete stats file.
3859 ereport(pgStatRunningInCollector ? LOG : WARNING,
3860 (errmsg("corrupted statistics file \"%s\"",
3870 unlink(PGSTAT_STAT_PERMANENT_FILENAME);
3876 * pgstat_read_statsfile_timestamp() -
3878 * Attempt to fetch the timestamp of an existing stats file.
3879 * Returns TRUE if successful (timestamp is stored at *ts).
3883 pgstat_read_statsfile_timestamp(bool permanent, TimestampTz *ts)
3885 PgStat_GlobalStats myGlobalStats;
3888 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3891 * Try to open the status file. As above, anything but ENOENT is worthy
3892 * of complaining about.
3894 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
3896 if (errno != ENOENT)
3897 ereport(pgStatRunningInCollector ? LOG : WARNING,
3898 (errcode_for_file_access(),
3899 errmsg("could not open statistics file \"%s\": %m",
3905 * Verify it's of the expected format.
3907 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id)
3908 || format_id != PGSTAT_FILE_FORMAT_ID)
3910 ereport(pgStatRunningInCollector ? LOG : WARNING,
3911 (errmsg("corrupted statistics file \"%s\"", statfile)));
3917 * Read global stats struct
3919 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats), fpin) != sizeof(myGlobalStats))
3921 ereport(pgStatRunningInCollector ? LOG : WARNING,
3922 (errmsg("corrupted statistics file \"%s\"", statfile)));
3927 *ts = myGlobalStats.stats_timestamp;
3934 * If not already done, read the statistics collector stats file into
3935 * some hash tables. The results will be kept until pgstat_clear_snapshot()
3936 * is called (typically, at end of transaction).
3939 backend_read_statsfile(void)
3945 /* already read it? */
3948 Assert(!pgStatRunningInCollector);
3951 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL msec
3952 * before now. This indirectly ensures that the collector needn't write
3953 * the file more often than PGSTAT_STAT_INTERVAL. In an autovacuum
3954 * worker, however, we want a lower delay to avoid using stale data, so we
3955 * use PGSTAT_RETRY_DELAY (since the number of worker is low, this
3956 * shouldn't be a problem).
3958 * Note that we don't recompute min_ts after sleeping; so we might end up
3959 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In practice
3960 * that shouldn't happen, though, as long as the sleep time is less than
3961 * PGSTAT_STAT_INTERVAL; and we don't want to lie to the collector about
3962 * what our cutoff time really is.
3964 cur_ts = GetCurrentTimestamp();
3965 if (IsAutoVacuumWorkerProcess())
3966 min_ts = TimestampTzPlusMilliseconds(cur_ts, -PGSTAT_RETRY_DELAY);
3968 min_ts = TimestampTzPlusMilliseconds(cur_ts, -PGSTAT_STAT_INTERVAL);
3971 * Loop until fresh enough stats file is available or we ran out of time.
3972 * The stats inquiry message is sent repeatedly in case collector drops
3973 * it; but not every single time, as that just swamps the collector.
3975 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
3977 TimestampTz file_ts = 0;
3979 CHECK_FOR_INTERRUPTS();
3981 if (pgstat_read_statsfile_timestamp(false, &file_ts) &&
3985 /* Not there or too old, so kick the collector and wait a bit */
3986 if ((count % PGSTAT_INQ_LOOP_COUNT) == 0)
3987 pgstat_send_inquiry(min_ts);
3989 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
3992 if (count >= PGSTAT_POLL_LOOP_COUNT)
3993 elog(WARNING, "pgstat wait timeout");
3995 /* Autovacuum launcher wants stats about all databases */
3996 if (IsAutoVacuumLauncherProcess())
3997 pgStatDBHash = pgstat_read_statsfile(InvalidOid, false);
3999 pgStatDBHash = pgstat_read_statsfile(MyDatabaseId, false);
4004 * pgstat_setup_memcxt() -
4006 * Create pgStatLocalContext, if not already done.
4010 pgstat_setup_memcxt(void)
4012 if (!pgStatLocalContext)
4013 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
4014 "Statistics snapshot",
4015 ALLOCSET_SMALL_MINSIZE,
4016 ALLOCSET_SMALL_INITSIZE,
4017 ALLOCSET_SMALL_MAXSIZE);
4022 * pgstat_clear_snapshot() -
4024 * Discard any data collected in the current transaction. Any subsequent
4025 * request will cause new snapshots to be read.
4027 * This is also invoked during transaction commit or abort to discard
4028 * the no-longer-wanted snapshot.
4032 pgstat_clear_snapshot(void)
4034 /* Release memory, if any was allocated */
4035 if (pgStatLocalContext)
4036 MemoryContextDelete(pgStatLocalContext);
4038 /* Reset variables */
4039 pgStatLocalContext = NULL;
4040 pgStatDBHash = NULL;
4041 localBackendStatusTable = NULL;
4042 localNumBackends = 0;
4047 * pgstat_recv_inquiry() -
4049 * Process stat inquiry requests.
4053 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
4055 if (msg->inquiry_time > last_statrequest)
4056 last_statrequest = msg->inquiry_time;
4061 * pgstat_recv_tabstat() -
4063 * Count what the backend has done.
4067 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
4069 PgStat_StatDBEntry *dbentry;
4070 PgStat_StatTabEntry *tabentry;
4074 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4077 * Update database-wide stats.
4079 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
4080 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
4081 dbentry->n_block_time_read += msg->m_block_time_read;
4082 dbentry->n_block_time_write += msg->m_block_time_write;
4085 * Process all table entries in the message.
4087 for (i = 0; i < msg->m_nentries; i++)
4089 PgStat_TableEntry *tabmsg = &(msg->m_entry[i]);
4091 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
4092 (void *) &(tabmsg->t_id),
4093 HASH_ENTER, &found);
4098 * If it's a new table entry, initialize counters to the values we
4101 tabentry->numscans = tabmsg->t_counts.t_numscans;
4102 tabentry->tuples_returned = tabmsg->t_counts.t_tuples_returned;
4103 tabentry->tuples_fetched = tabmsg->t_counts.t_tuples_fetched;
4104 tabentry->tuples_inserted = tabmsg->t_counts.t_tuples_inserted;
4105 tabentry->tuples_updated = tabmsg->t_counts.t_tuples_updated;
4106 tabentry->tuples_deleted = tabmsg->t_counts.t_tuples_deleted;
4107 tabentry->tuples_hot_updated = tabmsg->t_counts.t_tuples_hot_updated;
4108 tabentry->n_live_tuples = tabmsg->t_counts.t_delta_live_tuples;
4109 tabentry->n_dead_tuples = tabmsg->t_counts.t_delta_dead_tuples;
4110 tabentry->changes_since_analyze = tabmsg->t_counts.t_changed_tuples;
4111 tabentry->blocks_fetched = tabmsg->t_counts.t_blocks_fetched;
4112 tabentry->blocks_hit = tabmsg->t_counts.t_blocks_hit;
4114 tabentry->vacuum_timestamp = 0;
4115 tabentry->vacuum_count = 0;
4116 tabentry->autovac_vacuum_timestamp = 0;
4117 tabentry->autovac_vacuum_count = 0;
4118 tabentry->analyze_timestamp = 0;
4119 tabentry->analyze_count = 0;
4120 tabentry->autovac_analyze_timestamp = 0;
4121 tabentry->autovac_analyze_count = 0;
4126 * Otherwise add the values to the existing entry.
4128 tabentry->numscans += tabmsg->t_counts.t_numscans;
4129 tabentry->tuples_returned += tabmsg->t_counts.t_tuples_returned;
4130 tabentry->tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
4131 tabentry->tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
4132 tabentry->tuples_updated += tabmsg->t_counts.t_tuples_updated;
4133 tabentry->tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
4134 tabentry->tuples_hot_updated += tabmsg->t_counts.t_tuples_hot_updated;
4135 tabentry->n_live_tuples += tabmsg->t_counts.t_delta_live_tuples;
4136 tabentry->n_dead_tuples += tabmsg->t_counts.t_delta_dead_tuples;
4137 tabentry->changes_since_analyze += tabmsg->t_counts.t_changed_tuples;
4138 tabentry->blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
4139 tabentry->blocks_hit += tabmsg->t_counts.t_blocks_hit;
4142 /* Clamp n_live_tuples in case of negative delta_live_tuples */
4143 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
4144 /* Likewise for n_dead_tuples */
4145 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
4148 * Add per-table stats to the per-database entry, too.
4150 dbentry->n_tuples_returned += tabmsg->t_counts.t_tuples_returned;
4151 dbentry->n_tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
4152 dbentry->n_tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
4153 dbentry->n_tuples_updated += tabmsg->t_counts.t_tuples_updated;
4154 dbentry->n_tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
4155 dbentry->n_blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
4156 dbentry->n_blocks_hit += tabmsg->t_counts.t_blocks_hit;
4162 * pgstat_recv_tabpurge() -
4164 * Arrange for dead table removal.
4168 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
4170 PgStat_StatDBEntry *dbentry;
4173 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4176 * No need to purge if we don't even know the database.
4178 if (!dbentry || !dbentry->tables)
4182 * Process all table entries in the message.
4184 for (i = 0; i < msg->m_nentries; i++)
4186 /* Remove from hashtable if present; we don't care if it's not. */
4187 (void) hash_search(dbentry->tables,
4188 (void *) &(msg->m_tableid[i]),
4195 * pgstat_recv_dropdb() -
4197 * Arrange for dead database removal
4201 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
4203 PgStat_StatDBEntry *dbentry;
4206 * Lookup the database in the hashtable.
4208 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4211 * If found, remove it.
4215 if (dbentry->tables != NULL)
4216 hash_destroy(dbentry->tables);
4217 if (dbentry->functions != NULL)
4218 hash_destroy(dbentry->functions);
4220 if (hash_search(pgStatDBHash,
4221 (void *) &(dbentry->databaseid),
4222 HASH_REMOVE, NULL) == NULL)
4224 (errmsg("database hash table corrupted "
4225 "during cleanup --- abort")));
4231 * pgstat_recv_resetcounter() -
4233 * Reset the statistics for the specified database.
4237 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
4240 PgStat_StatDBEntry *dbentry;
4243 * Lookup the database in the hashtable. Nothing to do if not there.
4245 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4251 * We simply throw away all the database's table entries by recreating a
4252 * new hash table for them.
4254 if (dbentry->tables != NULL)
4255 hash_destroy(dbentry->tables);
4256 if (dbentry->functions != NULL)
4257 hash_destroy(dbentry->functions);
4259 dbentry->tables = NULL;
4260 dbentry->functions = NULL;
4263 * Reset database-level stats too. This should match the initialization
4264 * code in pgstat_get_db_entry().
4266 dbentry->n_xact_commit = 0;
4267 dbentry->n_xact_rollback = 0;
4268 dbentry->n_blocks_fetched = 0;
4269 dbentry->n_blocks_hit = 0;
4270 dbentry->n_tuples_returned = 0;
4271 dbentry->n_tuples_fetched = 0;
4272 dbentry->n_tuples_inserted = 0;
4273 dbentry->n_tuples_updated = 0;
4274 dbentry->n_tuples_deleted = 0;
4275 dbentry->last_autovac_time = 0;
4276 dbentry->n_temp_bytes = 0;
4277 dbentry->n_temp_files = 0;
4278 dbentry->n_deadlocks = 0;
4279 dbentry->n_block_time_read = 0;
4280 dbentry->n_block_time_write = 0;
4282 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
4284 memset(&hash_ctl, 0, sizeof(hash_ctl));
4285 hash_ctl.keysize = sizeof(Oid);
4286 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
4287 hash_ctl.hash = oid_hash;
4288 dbentry->tables = hash_create("Per-database table",
4289 PGSTAT_TAB_HASH_SIZE,
4291 HASH_ELEM | HASH_FUNCTION);
4293 hash_ctl.keysize = sizeof(Oid);
4294 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
4295 hash_ctl.hash = oid_hash;
4296 dbentry->functions = hash_create("Per-database function",
4297 PGSTAT_FUNCTION_HASH_SIZE,
4299 HASH_ELEM | HASH_FUNCTION);
4303 * pgstat_recv_resetshared() -
4305 * Reset some shared statistics of the cluster.
4309 pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
4311 if (msg->m_resettarget == RESET_BGWRITER)
4313 /* Reset the global background writer statistics for the cluster. */
4314 memset(&globalStats, 0, sizeof(globalStats));
4315 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
4319 * Presumably the sender of this message validated the target, don't
4320 * complain here if it's not valid
4325 * pgstat_recv_resetsinglecounter() -
4327 * Reset a statistics for a single object
4331 pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len)
4333 PgStat_StatDBEntry *dbentry;
4335 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4340 /* Set the reset timestamp for the whole database */
4341 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
4343 /* Remove object if it exists, ignore it if not */
4344 if (msg->m_resettype == RESET_TABLE)
4345 (void) hash_search(dbentry->tables, (void *) &(msg->m_objectid),
4347 else if (msg->m_resettype == RESET_FUNCTION)
4348 (void) hash_search(dbentry->functions, (void *) &(msg->m_objectid),
4353 * pgstat_recv_autovac() -
4355 * Process an autovacuum signalling message.
4359 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
4361 PgStat_StatDBEntry *dbentry;
4364 * Store the last autovacuum time in the database's hashtable entry.
4366 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4368 dbentry->last_autovac_time = msg->m_start_time;
4372 * pgstat_recv_vacuum() -
4374 * Process a VACUUM message.
4378 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
4380 PgStat_StatDBEntry *dbentry;
4381 PgStat_StatTabEntry *tabentry;
4384 * Store the data in the table's hashtable entry.
4386 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4388 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
4390 tabentry->n_live_tuples = msg->m_tuples;
4391 /* Resetting dead_tuples to 0 is an approximation ... */
4392 tabentry->n_dead_tuples = 0;
4394 if (msg->m_autovacuum)
4396 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
4397 tabentry->autovac_vacuum_count++;
4401 tabentry->vacuum_timestamp = msg->m_vacuumtime;
4402 tabentry->vacuum_count++;
4407 * pgstat_recv_analyze() -
4409 * Process an ANALYZE message.
4413 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
4415 PgStat_StatDBEntry *dbentry;
4416 PgStat_StatTabEntry *tabentry;
4419 * Store the data in the table's hashtable entry.
4421 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4423 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
4425 tabentry->n_live_tuples = msg->m_live_tuples;
4426 tabentry->n_dead_tuples = msg->m_dead_tuples;
4429 * We reset changes_since_analyze to zero, forgetting any changes that
4430 * occurred while the ANALYZE was in progress.
4432 tabentry->changes_since_analyze = 0;
4434 if (msg->m_autovacuum)
4436 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
4437 tabentry->autovac_analyze_count++;
4441 tabentry->analyze_timestamp = msg->m_analyzetime;
4442 tabentry->analyze_count++;
4448 * pgstat_recv_bgwriter() -
4450 * Process a BGWRITER message.
4454 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
4456 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
4457 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
4458 globalStats.checkpoint_write_time += msg->m_checkpoint_write_time;
4459 globalStats.checkpoint_sync_time += msg->m_checkpoint_sync_time;
4460 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
4461 globalStats.buf_written_clean += msg->m_buf_written_clean;
4462 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
4463 globalStats.buf_written_backend += msg->m_buf_written_backend;
4464 globalStats.buf_fsync_backend += msg->m_buf_fsync_backend;
4465 globalStats.buf_alloc += msg->m_buf_alloc;
4469 * pgstat_recv_recoveryconflict() -
4471 * Process a RECOVERYCONFLICT message.
4475 pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len)
4477 PgStat_StatDBEntry *dbentry;
4479 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4481 switch (msg->m_reason)
4483 case PROCSIG_RECOVERY_CONFLICT_DATABASE:
4486 * Since we drop the information about the database as soon as it
4487 * replicates, there is no point in counting these conflicts.
4490 case PROCSIG_RECOVERY_CONFLICT_TABLESPACE:
4491 dbentry->n_conflict_tablespace++;
4493 case PROCSIG_RECOVERY_CONFLICT_LOCK:
4494 dbentry->n_conflict_lock++;
4496 case PROCSIG_RECOVERY_CONFLICT_SNAPSHOT:
4497 dbentry->n_conflict_snapshot++;
4499 case PROCSIG_RECOVERY_CONFLICT_BUFFERPIN:
4500 dbentry->n_conflict_bufferpin++;
4502 case PROCSIG_RECOVERY_CONFLICT_STARTUP_DEADLOCK:
4503 dbentry->n_conflict_startup_deadlock++;
4509 * pgstat_recv_deadlock() -
4511 * Process a DEADLOCK message.
4515 pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len)
4517 PgStat_StatDBEntry *dbentry;
4519 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4521 dbentry->n_deadlocks++;
4525 * pgstat_recv_tempfile() -
4527 * Process a TEMPFILE message.
4531 pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len)
4533 PgStat_StatDBEntry *dbentry;
4535 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4537 dbentry->n_temp_bytes += msg->m_filesize;
4538 dbentry->n_temp_files += 1;
4542 * pgstat_recv_funcstat() -
4544 * Count what the backend has done.
4548 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
4550 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
4551 PgStat_StatDBEntry *dbentry;
4552 PgStat_StatFuncEntry *funcentry;
4556 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
4559 * Process all function entries in the message.
4561 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
4563 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
4564 (void *) &(funcmsg->f_id),
4565 HASH_ENTER, &found);
4570 * If it's a new function entry, initialize counters to the values
4573 funcentry->f_numcalls = funcmsg->f_numcalls;
4574 funcentry->f_time = funcmsg->f_time;
4575 funcentry->f_time_self = funcmsg->f_time_self;
4580 * Otherwise add the values to the existing entry.
4582 funcentry->f_numcalls += funcmsg->f_numcalls;
4583 funcentry->f_time += funcmsg->f_time;
4584 funcentry->f_time_self += funcmsg->f_time_self;
4590 * pgstat_recv_funcpurge() -
4592 * Arrange for dead function removal.
4596 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
4598 PgStat_StatDBEntry *dbentry;
4601 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
4604 * No need to purge if we don't even know the database.
4606 if (!dbentry || !dbentry->functions)
4610 * Process all function entries in the message.
4612 for (i = 0; i < msg->m_nentries; i++)
4614 /* Remove from hashtable if present; we don't care if it's not. */
4615 (void) hash_search(dbentry->functions,
4616 (void *) &(msg->m_functionid[i]),