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) 2001-2018, 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>
31 #ifdef HAVE_SYS_SELECT_H
32 #include <sys/select.h>
37 #include "access/heapam.h"
38 #include "access/htup_details.h"
39 #include "access/transam.h"
40 #include "access/twophase_rmgr.h"
41 #include "access/xact.h"
42 #include "catalog/pg_database.h"
43 #include "catalog/pg_proc.h"
44 #include "common/ip.h"
45 #include "libpq/libpq.h"
46 #include "libpq/pqsignal.h"
47 #include "mb/pg_wchar.h"
48 #include "miscadmin.h"
50 #include "postmaster/autovacuum.h"
51 #include "postmaster/fork_process.h"
52 #include "postmaster/postmaster.h"
53 #include "replication/walsender.h"
54 #include "storage/backendid.h"
55 #include "storage/dsm.h"
56 #include "storage/fd.h"
57 #include "storage/ipc.h"
58 #include "storage/latch.h"
59 #include "storage/lmgr.h"
60 #include "storage/pg_shmem.h"
61 #include "storage/procsignal.h"
62 #include "storage/sinvaladt.h"
63 #include "utils/ascii.h"
64 #include "utils/guc.h"
65 #include "utils/memutils.h"
66 #include "utils/ps_status.h"
67 #include "utils/rel.h"
68 #include "utils/snapmgr.h"
69 #include "utils/timestamp.h"
70 #include "utils/tqual.h"
77 #define PGSTAT_STAT_INTERVAL 500 /* Minimum time between stats file
78 * updates; in milliseconds. */
80 #define PGSTAT_RETRY_DELAY 10 /* How long to wait between checks for a
81 * new file; in milliseconds. */
83 #define PGSTAT_MAX_WAIT_TIME 10000 /* Maximum time to wait for a stats
84 * file update; in milliseconds. */
86 #define PGSTAT_INQ_INTERVAL 640 /* How often to ping the collector for a
87 * new file; in milliseconds. */
89 #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a
90 * failed statistics collector; in
93 #define PGSTAT_POLL_LOOP_COUNT (PGSTAT_MAX_WAIT_TIME / PGSTAT_RETRY_DELAY)
94 #define PGSTAT_INQ_LOOP_COUNT (PGSTAT_INQ_INTERVAL / PGSTAT_RETRY_DELAY)
96 /* Minimum receive buffer size for the collector's socket. */
97 #define PGSTAT_MIN_RCVBUF (100 * 1024)
101 * The initial size hints for the hash tables used in the collector.
104 #define PGSTAT_DB_HASH_SIZE 16
105 #define PGSTAT_TAB_HASH_SIZE 512
106 #define PGSTAT_FUNCTION_HASH_SIZE 512
110 * Total number of backends including auxiliary
112 * We reserve a slot for each possible BackendId, plus one for each
113 * possible auxiliary process type. (This scheme assumes there is not
114 * more than one of any auxiliary process type at a time.) MaxBackends
115 * includes autovacuum workers and background workers as well.
118 #define NumBackendStatSlots (MaxBackends + NUM_AUXPROCTYPES)
125 bool pgstat_track_activities = false;
126 bool pgstat_track_counts = false;
127 int pgstat_track_functions = TRACK_FUNC_OFF;
128 int pgstat_track_activity_query_size = 1024;
131 * Built from GUC parameter
134 char *pgstat_stat_directory = NULL;
135 char *pgstat_stat_filename = NULL;
136 char *pgstat_stat_tmpname = NULL;
139 * BgWriter global statistics counters (unused in other processes).
140 * Stored directly in a stats message structure so it can be sent
141 * without needing to copy things around. We assume this inits to zeroes.
143 PgStat_MsgBgWriter BgWriterStats;
149 NON_EXEC_STATIC pgsocket pgStatSock = PGINVALID_SOCKET;
151 static struct sockaddr_storage pgStatAddr;
153 static time_t last_pgstat_start_time;
155 static bool pgStatRunningInCollector = false;
158 * Structures in which backends store per-table info that's waiting to be
159 * sent to the collector.
161 * NOTE: once allocated, TabStatusArray structures are never moved or deleted
162 * for the life of the backend. Also, we zero out the t_id fields of the
163 * contained PgStat_TableStatus structs whenever they are not actively in use.
164 * This allows relcache pgstat_info pointers to be treated as long-lived data,
165 * avoiding repeated searches in pgstat_initstats() when a relation is
166 * repeatedly opened during a transaction.
168 #define TABSTAT_QUANTUM 100 /* we alloc this many at a time */
170 typedef struct TabStatusArray
172 struct TabStatusArray *tsa_next; /* link to next array, if any */
173 int tsa_used; /* # entries currently used */
174 PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */
177 static TabStatusArray *pgStatTabList = NULL;
180 * pgStatTabHash entry: map from relation OID to PgStat_TableStatus pointer
182 typedef struct TabStatHashEntry
185 PgStat_TableStatus *tsa_entry;
189 * Hash table for O(1) t_id -> tsa_entry lookup
191 static HTAB *pgStatTabHash = NULL;
194 * Backends store per-function info that's waiting to be sent to the collector
195 * in this hash table (indexed by function OID).
197 static HTAB *pgStatFunctions = NULL;
200 * Indicates if backend has some function stats that it hasn't yet
201 * sent to the collector.
203 static bool have_function_stats = false;
206 * Tuple insertion/deletion counts for an open transaction can't be propagated
207 * into PgStat_TableStatus counters until we know if it is going to commit
208 * or abort. Hence, we keep these counts in per-subxact structs that live
209 * in TopTransactionContext. This data structure is designed on the assumption
210 * that subxacts won't usually modify very many tables.
212 typedef struct PgStat_SubXactStatus
214 int nest_level; /* subtransaction nest level */
215 struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
216 PgStat_TableXactStatus *first; /* head of list for this subxact */
217 } PgStat_SubXactStatus;
219 static PgStat_SubXactStatus *pgStatXactStack = NULL;
221 static int pgStatXactCommit = 0;
222 static int pgStatXactRollback = 0;
223 PgStat_Counter pgStatBlockReadTime = 0;
224 PgStat_Counter pgStatBlockWriteTime = 0;
226 /* Record that's written to 2PC state file when pgstat state is persisted */
227 typedef struct TwoPhasePgStatRecord
229 PgStat_Counter tuples_inserted; /* tuples inserted in xact */
230 PgStat_Counter tuples_updated; /* tuples updated in xact */
231 PgStat_Counter tuples_deleted; /* tuples deleted in xact */
232 PgStat_Counter inserted_pre_trunc; /* tuples inserted prior to truncate */
233 PgStat_Counter updated_pre_trunc; /* tuples updated prior to truncate */
234 PgStat_Counter deleted_pre_trunc; /* tuples deleted prior to truncate */
235 Oid t_id; /* table's OID */
236 bool t_shared; /* is it a shared catalog? */
237 bool t_truncated; /* was the relation truncated? */
238 } TwoPhasePgStatRecord;
241 * Info about current "snapshot" of stats file
243 static MemoryContext pgStatLocalContext = NULL;
244 static HTAB *pgStatDBHash = NULL;
246 /* Status for backends including auxiliary */
247 static LocalPgBackendStatus *localBackendStatusTable = NULL;
249 /* Total number of backends including auxiliary */
250 static int localNumBackends = 0;
253 * Cluster wide statistics, kept in the stats collector.
254 * Contains statistics that are not collected per database
257 static PgStat_ArchiverStats archiverStats;
258 static PgStat_GlobalStats globalStats;
261 * List of OIDs of databases we need to write out. If an entry is InvalidOid,
262 * it means to write only the shared-catalog stats ("DB 0"); otherwise, we
263 * will write both that DB's data and the shared stats.
265 static List *pending_write_requests = NIL;
267 /* Signal handler flags */
268 static volatile bool need_exit = false;
269 static volatile bool got_SIGHUP = false;
272 * Total time charged to functions so far in the current backend.
273 * We use this to help separate "self" and "other" time charges.
274 * (We assume this initializes to zero.)
276 static instr_time total_func_time;
280 * Local function forward declarations
284 static pid_t pgstat_forkexec(void);
287 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
288 static void pgstat_exit(SIGNAL_ARGS);
289 static void pgstat_beshutdown_hook(int code, Datum arg);
290 static void pgstat_sighup_handler(SIGNAL_ARGS);
292 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
293 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
294 Oid tableoid, bool create);
295 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
296 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
297 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
298 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
299 static void backend_read_statsfile(void);
300 static void pgstat_read_current_status(void);
302 static bool pgstat_write_statsfile_needed(void);
303 static bool pgstat_db_requested(Oid databaseid);
305 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
306 static void pgstat_send_funcstats(void);
307 static HTAB *pgstat_collect_oids(Oid catalogid);
309 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
311 static void pgstat_setup_memcxt(void);
313 static const char *pgstat_get_wait_activity(WaitEventActivity w);
314 static const char *pgstat_get_wait_client(WaitEventClient w);
315 static const char *pgstat_get_wait_ipc(WaitEventIPC w);
316 static const char *pgstat_get_wait_timeout(WaitEventTimeout w);
317 static const char *pgstat_get_wait_io(WaitEventIO w);
319 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
320 static void pgstat_send(void *msg, int len);
322 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
323 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
324 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
325 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
326 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
327 static void pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len);
328 static void pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len);
329 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
330 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
331 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
332 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
333 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
334 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
335 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
336 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
337 static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
338 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
340 /* ------------------------------------------------------------
341 * Public functions called from postmaster follow
342 * ------------------------------------------------------------
348 * Called from postmaster at startup. Create the resources required
349 * by the statistics collector process. If unable to do so, do not
350 * fail --- better to let the postmaster start with stats collection
357 ACCEPT_TYPE_ARG3 alen;
358 struct addrinfo *addrs = NULL,
368 #define TESTBYTEVAL ((char) 199)
371 * This static assertion verifies that we didn't mess up the calculations
372 * involved in selecting maximum payload sizes for our UDP messages.
373 * Because the only consequence of overrunning PGSTAT_MAX_MSG_SIZE would
374 * be silent performance loss from fragmentation, it seems worth having a
375 * compile-time cross-check that we didn't.
377 StaticAssertStmt(sizeof(PgStat_Msg) <= PGSTAT_MAX_MSG_SIZE,
378 "maximum stats message size exceeds PGSTAT_MAX_MSG_SIZE");
381 * Create the UDP socket for sending and receiving statistic messages
383 hints.ai_flags = AI_PASSIVE;
384 hints.ai_family = AF_UNSPEC;
385 hints.ai_socktype = SOCK_DGRAM;
386 hints.ai_protocol = 0;
387 hints.ai_addrlen = 0;
388 hints.ai_addr = NULL;
389 hints.ai_canonname = NULL;
390 hints.ai_next = NULL;
391 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
395 (errmsg("could not resolve \"localhost\": %s",
396 gai_strerror(ret))));
401 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
402 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
403 * when kernel will reject IPv6). Worse, the failure may occur at the
404 * bind() or perhaps even connect() stage. So we must loop through the
405 * results till we find a working combination. We will generate LOG
406 * messages, but no error, for bogus combinations.
408 for (addr = addrs; addr; addr = addr->ai_next)
410 #ifdef HAVE_UNIX_SOCKETS
411 /* Ignore AF_UNIX sockets, if any are returned. */
412 if (addr->ai_family == AF_UNIX)
418 (errmsg("trying another address for the statistics collector")));
423 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) == PGINVALID_SOCKET)
426 (errcode_for_socket_access(),
427 errmsg("could not create socket for statistics collector: %m")));
432 * Bind it to a kernel assigned port on localhost and get the assigned
433 * port via getsockname().
435 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
438 (errcode_for_socket_access(),
439 errmsg("could not bind socket for statistics collector: %m")));
440 closesocket(pgStatSock);
441 pgStatSock = PGINVALID_SOCKET;
445 alen = sizeof(pgStatAddr);
446 if (getsockname(pgStatSock, (struct sockaddr *) &pgStatAddr, &alen) < 0)
449 (errcode_for_socket_access(),
450 errmsg("could not get address of socket for statistics collector: %m")));
451 closesocket(pgStatSock);
452 pgStatSock = PGINVALID_SOCKET;
457 * Connect the socket to its own address. This saves a few cycles by
458 * not having to respecify the target address on every send. This also
459 * provides a kernel-level check that only packets from this same
460 * address will be received.
462 if (connect(pgStatSock, (struct sockaddr *) &pgStatAddr, alen) < 0)
465 (errcode_for_socket_access(),
466 errmsg("could not connect socket for statistics collector: %m")));
467 closesocket(pgStatSock);
468 pgStatSock = PGINVALID_SOCKET;
473 * Try to send and receive a one-byte test message on the socket. This
474 * is to catch situations where the socket can be created but will not
475 * actually pass data (for instance, because kernel packet filtering
478 test_byte = TESTBYTEVAL;
481 if (send(pgStatSock, &test_byte, 1, 0) != 1)
484 goto retry1; /* if interrupted, just retry */
486 (errcode_for_socket_access(),
487 errmsg("could not send test message on socket for statistics collector: %m")));
488 closesocket(pgStatSock);
489 pgStatSock = PGINVALID_SOCKET;
494 * There could possibly be a little delay before the message can be
495 * received. We arbitrarily allow up to half a second before deciding
498 for (;;) /* need a loop to handle EINTR */
501 FD_SET(pgStatSock, &rset);
505 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
506 if (sel_res >= 0 || errno != EINTR)
512 (errcode_for_socket_access(),
513 errmsg("select() failed in statistics collector: %m")));
514 closesocket(pgStatSock);
515 pgStatSock = PGINVALID_SOCKET;
518 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
521 * This is the case we actually think is likely, so take pains to
522 * give a specific message for it.
524 * errno will not be set meaningfully here, so don't use it.
527 (errcode(ERRCODE_CONNECTION_FAILURE),
528 errmsg("test message did not get through on socket for statistics collector")));
529 closesocket(pgStatSock);
530 pgStatSock = PGINVALID_SOCKET;
534 test_byte++; /* just make sure variable is changed */
537 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
540 goto retry2; /* if interrupted, just retry */
542 (errcode_for_socket_access(),
543 errmsg("could not receive test message on socket for statistics collector: %m")));
544 closesocket(pgStatSock);
545 pgStatSock = PGINVALID_SOCKET;
549 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
552 (errcode(ERRCODE_INTERNAL_ERROR),
553 errmsg("incorrect test message transmission on socket for statistics collector")));
554 closesocket(pgStatSock);
555 pgStatSock = PGINVALID_SOCKET;
559 /* If we get here, we have a working socket */
563 /* Did we find a working address? */
564 if (!addr || pgStatSock == PGINVALID_SOCKET)
568 * Set the socket to non-blocking IO. This ensures that if the collector
569 * falls behind, statistics messages will be discarded; backends won't
570 * block waiting to send messages to the collector.
572 if (!pg_set_noblock(pgStatSock))
575 (errcode_for_socket_access(),
576 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
581 * Try to ensure that the socket's receive buffer is at least
582 * PGSTAT_MIN_RCVBUF bytes, so that it won't easily overflow and lose
583 * data. Use of UDP protocol means that we are willing to lose data under
584 * heavy load, but we don't want it to happen just because of ridiculously
585 * small default buffer sizes (such as 8KB on older Windows versions).
590 ACCEPT_TYPE_ARG3 rcvbufsize = sizeof(old_rcvbuf);
592 if (getsockopt(pgStatSock, SOL_SOCKET, SO_RCVBUF,
593 (char *) &old_rcvbuf, &rcvbufsize) < 0)
595 elog(LOG, "getsockopt(SO_RCVBUF) failed: %m");
596 /* if we can't get existing size, always try to set it */
600 new_rcvbuf = PGSTAT_MIN_RCVBUF;
601 if (old_rcvbuf < new_rcvbuf)
603 if (setsockopt(pgStatSock, SOL_SOCKET, SO_RCVBUF,
604 (char *) &new_rcvbuf, sizeof(new_rcvbuf)) < 0)
605 elog(LOG, "setsockopt(SO_RCVBUF) failed: %m");
609 pg_freeaddrinfo_all(hints.ai_family, addrs);
615 (errmsg("disabling statistics collector for lack of working socket")));
618 pg_freeaddrinfo_all(hints.ai_family, addrs);
620 if (pgStatSock != PGINVALID_SOCKET)
621 closesocket(pgStatSock);
622 pgStatSock = PGINVALID_SOCKET;
625 * Adjust GUC variables to suppress useless activity, and for debugging
626 * purposes (seeing track_counts off is a clue that we failed here). We
627 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
628 * on from postgresql.conf without a restart.
630 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
634 * subroutine for pgstat_reset_all
637 pgstat_reset_remove_files(const char *directory)
640 struct dirent *entry;
641 char fname[MAXPGPATH * 2];
643 dir = AllocateDir(directory);
644 while ((entry = ReadDir(dir, directory)) != NULL)
650 * Skip directory entries that don't match the file names we write.
651 * See get_dbstat_filename for the database-specific pattern.
653 if (strncmp(entry->d_name, "global.", 7) == 0)
658 (void) sscanf(entry->d_name, "db_%u.%n",
662 /* %u allows leading whitespace, so reject that */
663 if (strchr("0123456789", entry->d_name[3]) == NULL)
667 if (strcmp(entry->d_name + nchars, "tmp") != 0 &&
668 strcmp(entry->d_name + nchars, "stat") != 0)
671 snprintf(fname, sizeof(fname), "%s/%s", directory,
679 * pgstat_reset_all() -
681 * Remove the stats files. This is currently used only if WAL
682 * recovery is needed after a crash.
685 pgstat_reset_all(void)
687 pgstat_reset_remove_files(pgstat_stat_directory);
688 pgstat_reset_remove_files(PGSTAT_STAT_PERMANENT_DIRECTORY);
694 * pgstat_forkexec() -
696 * Format up the arglist for, then fork and exec, statistics collector process
699 pgstat_forkexec(void)
704 av[ac++] = "postgres";
705 av[ac++] = "--forkcol";
706 av[ac++] = NULL; /* filled in by postmaster_forkexec */
709 Assert(ac < lengthof(av));
711 return postmaster_forkexec(ac, av);
713 #endif /* EXEC_BACKEND */
719 * Called from postmaster at startup or after an existing collector
720 * died. Attempt to fire up a fresh statistics collector.
722 * Returns PID of child process, or 0 if fail.
724 * Note: if fail, we will be called again from the postmaster main loop.
733 * Check that the socket is there, else pgstat_init failed and we can do
736 if (pgStatSock == PGINVALID_SOCKET)
740 * Do nothing if too soon since last collector start. This is a safety
741 * valve to protect against continuous respawn attempts if the collector
742 * is dying immediately at launch. Note that since we will be re-called
743 * from the postmaster main loop, we will get another chance later.
745 curtime = time(NULL);
746 if ((unsigned int) (curtime - last_pgstat_start_time) <
747 (unsigned int) PGSTAT_RESTART_INTERVAL)
749 last_pgstat_start_time = curtime;
752 * Okay, fork off the collector.
755 switch ((pgStatPid = pgstat_forkexec()))
757 switch ((pgStatPid = fork_process()))
762 (errmsg("could not fork statistics collector: %m")));
767 /* in postmaster child ... */
768 InitPostmasterChild();
770 /* Close the postmaster's sockets */
771 ClosePostmasterPorts(false);
773 /* Drop our connection to postmaster's shared memory, as well */
775 PGSharedMemoryDetach();
777 PgstatCollectorMain(0, NULL);
782 return (int) pgStatPid;
785 /* shouldn't get here */
790 allow_immediate_pgstat_restart(void)
792 last_pgstat_start_time = 0;
795 /* ------------------------------------------------------------
796 * Public functions used by backends follow
797 *------------------------------------------------------------
802 * pgstat_report_stat() -
804 * Must be called by processes that performs DML: tcop/postgres.c, logical
805 * receiver processes, SPI worker, etc. to send the so far collected
806 * per-table and function usage statistics to the collector. Note that this
807 * is called only when not within a transaction, so it is fair to use
808 * transaction stop time as an approximation of current time.
812 pgstat_report_stat(bool force)
814 /* we assume this inits to all zeroes: */
815 static const PgStat_TableCounts all_zeroes;
816 static TimestampTz last_report = 0;
819 PgStat_MsgTabstat regular_msg;
820 PgStat_MsgTabstat shared_msg;
824 /* Don't expend a clock check if nothing to do */
825 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0) &&
826 pgStatXactCommit == 0 && pgStatXactRollback == 0 &&
827 !have_function_stats)
831 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
832 * msec since we last sent one, or the caller wants to force stats out.
834 now = GetCurrentTransactionStopTimestamp();
836 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
841 * Destroy pgStatTabHash before we start invalidating PgStat_TableEntry
842 * entries it points to. (Should we fail partway through the loop below,
843 * it's okay to have removed the hashtable already --- the only
844 * consequence is we'd get multiple entries for the same table in the
845 * pgStatTabList, and that's safe.)
848 hash_destroy(pgStatTabHash);
849 pgStatTabHash = NULL;
852 * Scan through the TabStatusArray struct(s) to find tables that actually
853 * have counts, and build messages to send. We have to separate shared
854 * relations from regular ones because the databaseid field in the message
855 * header has to depend on that.
857 regular_msg.m_databaseid = MyDatabaseId;
858 shared_msg.m_databaseid = InvalidOid;
859 regular_msg.m_nentries = 0;
860 shared_msg.m_nentries = 0;
862 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
864 for (i = 0; i < tsa->tsa_used; i++)
866 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
867 PgStat_MsgTabstat *this_msg;
868 PgStat_TableEntry *this_ent;
870 /* Shouldn't have any pending transaction-dependent counts */
871 Assert(entry->trans == NULL);
874 * Ignore entries that didn't accumulate any actual counts, such
875 * as indexes that were opened by the planner but not used.
877 if (memcmp(&entry->t_counts, &all_zeroes,
878 sizeof(PgStat_TableCounts)) == 0)
882 * OK, insert data into the appropriate message, and send if full.
884 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
885 this_ent = &this_msg->m_entry[this_msg->m_nentries];
886 this_ent->t_id = entry->t_id;
887 memcpy(&this_ent->t_counts, &entry->t_counts,
888 sizeof(PgStat_TableCounts));
889 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
891 pgstat_send_tabstat(this_msg);
892 this_msg->m_nentries = 0;
895 /* zero out TableStatus structs after use */
896 MemSet(tsa->tsa_entries, 0,
897 tsa->tsa_used * sizeof(PgStat_TableStatus));
902 * Send partial messages. Make sure that any pending xact commit/abort
903 * gets counted, even if there are no table stats to send.
905 if (regular_msg.m_nentries > 0 ||
906 pgStatXactCommit > 0 || pgStatXactRollback > 0)
907 pgstat_send_tabstat(®ular_msg);
908 if (shared_msg.m_nentries > 0)
909 pgstat_send_tabstat(&shared_msg);
911 /* Now, send function statistics */
912 pgstat_send_funcstats();
916 * Subroutine for pgstat_report_stat: finish and send a tabstat message
919 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
924 /* It's unlikely we'd get here with no socket, but maybe not impossible */
925 if (pgStatSock == PGINVALID_SOCKET)
929 * Report and reset accumulated xact commit/rollback and I/O timings
930 * whenever we send a normal tabstat message
932 if (OidIsValid(tsmsg->m_databaseid))
934 tsmsg->m_xact_commit = pgStatXactCommit;
935 tsmsg->m_xact_rollback = pgStatXactRollback;
936 tsmsg->m_block_read_time = pgStatBlockReadTime;
937 tsmsg->m_block_write_time = pgStatBlockWriteTime;
938 pgStatXactCommit = 0;
939 pgStatXactRollback = 0;
940 pgStatBlockReadTime = 0;
941 pgStatBlockWriteTime = 0;
945 tsmsg->m_xact_commit = 0;
946 tsmsg->m_xact_rollback = 0;
947 tsmsg->m_block_read_time = 0;
948 tsmsg->m_block_write_time = 0;
951 n = tsmsg->m_nentries;
952 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
953 n * sizeof(PgStat_TableEntry);
955 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
956 pgstat_send(tsmsg, len);
960 * Subroutine for pgstat_report_stat: populate and send a function stat message
963 pgstat_send_funcstats(void)
965 /* we assume this inits to all zeroes: */
966 static const PgStat_FunctionCounts all_zeroes;
968 PgStat_MsgFuncstat msg;
969 PgStat_BackendFunctionEntry *entry;
970 HASH_SEQ_STATUS fstat;
972 if (pgStatFunctions == NULL)
975 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
976 msg.m_databaseid = MyDatabaseId;
979 hash_seq_init(&fstat, pgStatFunctions);
980 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
982 PgStat_FunctionEntry *m_ent;
984 /* Skip it if no counts accumulated since last time */
985 if (memcmp(&entry->f_counts, &all_zeroes,
986 sizeof(PgStat_FunctionCounts)) == 0)
989 /* need to convert format of time accumulators */
990 m_ent = &msg.m_entry[msg.m_nentries];
991 m_ent->f_id = entry->f_id;
992 m_ent->f_numcalls = entry->f_counts.f_numcalls;
993 m_ent->f_total_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_total_time);
994 m_ent->f_self_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_self_time);
996 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
998 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
999 msg.m_nentries * sizeof(PgStat_FunctionEntry));
1003 /* reset the entry's counts */
1004 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
1007 if (msg.m_nentries > 0)
1008 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
1009 msg.m_nentries * sizeof(PgStat_FunctionEntry));
1011 have_function_stats = false;
1016 * pgstat_vacuum_stat() -
1018 * Will tell the collector about objects he can get rid of.
1022 pgstat_vacuum_stat(void)
1025 PgStat_MsgTabpurge msg;
1026 PgStat_MsgFuncpurge f_msg;
1027 HASH_SEQ_STATUS hstat;
1028 PgStat_StatDBEntry *dbentry;
1029 PgStat_StatTabEntry *tabentry;
1030 PgStat_StatFuncEntry *funcentry;
1033 if (pgStatSock == PGINVALID_SOCKET)
1037 * If not done for this transaction, read the statistics collector stats
1038 * file into some hash tables.
1040 backend_read_statsfile();
1043 * Read pg_database and make a list of OIDs of all existing databases
1045 htab = pgstat_collect_oids(DatabaseRelationId);
1048 * Search the database hash table for dead databases and tell the
1049 * collector to drop them.
1051 hash_seq_init(&hstat, pgStatDBHash);
1052 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
1054 Oid dbid = dbentry->databaseid;
1056 CHECK_FOR_INTERRUPTS();
1058 /* the DB entry for shared tables (with InvalidOid) is never dropped */
1059 if (OidIsValid(dbid) &&
1060 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
1061 pgstat_drop_database(dbid);
1068 * Lookup our own database entry; if not found, nothing more to do.
1070 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1071 (void *) &MyDatabaseId,
1073 if (dbentry == NULL || dbentry->tables == NULL)
1077 * Similarly to above, make a list of all known relations in this DB.
1079 htab = pgstat_collect_oids(RelationRelationId);
1082 * Initialize our messages table counter to zero
1087 * Check for all tables listed in stats hashtable if they still exist.
1089 hash_seq_init(&hstat, dbentry->tables);
1090 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
1092 Oid tabid = tabentry->tableid;
1094 CHECK_FOR_INTERRUPTS();
1096 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
1100 * Not there, so add this table's Oid to the message
1102 msg.m_tableid[msg.m_nentries++] = tabid;
1105 * If the message is full, send it out and reinitialize to empty
1107 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
1109 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
1110 + msg.m_nentries * sizeof(Oid);
1112 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1113 msg.m_databaseid = MyDatabaseId;
1114 pgstat_send(&msg, len);
1123 if (msg.m_nentries > 0)
1125 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
1126 + msg.m_nentries * sizeof(Oid);
1128 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1129 msg.m_databaseid = MyDatabaseId;
1130 pgstat_send(&msg, len);
1137 * Now repeat the above steps for functions. However, we needn't bother
1138 * in the common case where no function stats are being collected.
1140 if (dbentry->functions != NULL &&
1141 hash_get_num_entries(dbentry->functions) > 0)
1143 htab = pgstat_collect_oids(ProcedureRelationId);
1145 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
1146 f_msg.m_databaseid = MyDatabaseId;
1147 f_msg.m_nentries = 0;
1149 hash_seq_init(&hstat, dbentry->functions);
1150 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
1152 Oid funcid = funcentry->functionid;
1154 CHECK_FOR_INTERRUPTS();
1156 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
1160 * Not there, so add this function's Oid to the message
1162 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
1165 * If the message is full, send it out and reinitialize to empty
1167 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1169 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1170 + f_msg.m_nentries * sizeof(Oid);
1172 pgstat_send(&f_msg, len);
1174 f_msg.m_nentries = 0;
1181 if (f_msg.m_nentries > 0)
1183 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1184 + f_msg.m_nentries * sizeof(Oid);
1186 pgstat_send(&f_msg, len);
1195 * pgstat_collect_oids() -
1197 * Collect the OIDs of all objects listed in the specified system catalog
1198 * into a temporary hash table. Caller should hash_destroy the result
1199 * when done with it. (However, we make the table in CurrentMemoryContext
1200 * so that it will be freed properly in event of an error.)
1204 pgstat_collect_oids(Oid catalogid)
1213 memset(&hash_ctl, 0, sizeof(hash_ctl));
1214 hash_ctl.keysize = sizeof(Oid);
1215 hash_ctl.entrysize = sizeof(Oid);
1216 hash_ctl.hcxt = CurrentMemoryContext;
1217 htab = hash_create("Temporary table of OIDs",
1218 PGSTAT_TAB_HASH_SIZE,
1220 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1222 rel = heap_open(catalogid, AccessShareLock);
1223 snapshot = RegisterSnapshot(GetLatestSnapshot());
1224 scan = heap_beginscan(rel, snapshot, 0, NULL);
1225 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1227 Oid thisoid = HeapTupleGetOid(tup);
1229 CHECK_FOR_INTERRUPTS();
1231 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1234 UnregisterSnapshot(snapshot);
1235 heap_close(rel, AccessShareLock);
1242 * pgstat_drop_database() -
1244 * Tell the collector that we just dropped a database.
1245 * (If the message gets lost, we will still clean the dead DB eventually
1246 * via future invocations of pgstat_vacuum_stat().)
1250 pgstat_drop_database(Oid databaseid)
1252 PgStat_MsgDropdb msg;
1254 if (pgStatSock == PGINVALID_SOCKET)
1257 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1258 msg.m_databaseid = databaseid;
1259 pgstat_send(&msg, sizeof(msg));
1264 * pgstat_drop_relation() -
1266 * Tell the collector that we just dropped a relation.
1267 * (If the message gets lost, we will still clean the dead entry eventually
1268 * via future invocations of pgstat_vacuum_stat().)
1270 * Currently not used for lack of any good place to call it; we rely
1271 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1276 pgstat_drop_relation(Oid relid)
1278 PgStat_MsgTabpurge msg;
1281 if (pgStatSock == PGINVALID_SOCKET)
1284 msg.m_tableid[0] = relid;
1287 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) + sizeof(Oid);
1289 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1290 msg.m_databaseid = MyDatabaseId;
1291 pgstat_send(&msg, len);
1293 #endif /* NOT_USED */
1297 * pgstat_reset_counters() -
1299 * Tell the statistics collector to reset counters for our database.
1301 * Permission checking for this function is managed through the normal
1306 pgstat_reset_counters(void)
1308 PgStat_MsgResetcounter msg;
1310 if (pgStatSock == PGINVALID_SOCKET)
1313 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1314 msg.m_databaseid = MyDatabaseId;
1315 pgstat_send(&msg, sizeof(msg));
1319 * pgstat_reset_shared_counters() -
1321 * Tell the statistics collector to reset cluster-wide shared counters.
1323 * Permission checking for this function is managed through the normal
1328 pgstat_reset_shared_counters(const char *target)
1330 PgStat_MsgResetsharedcounter msg;
1332 if (pgStatSock == PGINVALID_SOCKET)
1335 if (strcmp(target, "archiver") == 0)
1336 msg.m_resettarget = RESET_ARCHIVER;
1337 else if (strcmp(target, "bgwriter") == 0)
1338 msg.m_resettarget = RESET_BGWRITER;
1341 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1342 errmsg("unrecognized reset target: \"%s\"", target),
1343 errhint("Target must be \"archiver\" or \"bgwriter\".")));
1345 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSHAREDCOUNTER);
1346 pgstat_send(&msg, sizeof(msg));
1350 * pgstat_reset_single_counter() -
1352 * Tell the statistics collector to reset a single counter.
1354 * Permission checking for this function is managed through the normal
1359 pgstat_reset_single_counter(Oid objoid, PgStat_Single_Reset_Type type)
1361 PgStat_MsgResetsinglecounter msg;
1363 if (pgStatSock == PGINVALID_SOCKET)
1366 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSINGLECOUNTER);
1367 msg.m_databaseid = MyDatabaseId;
1368 msg.m_resettype = type;
1369 msg.m_objectid = objoid;
1371 pgstat_send(&msg, sizeof(msg));
1375 * pgstat_report_autovac() -
1377 * Called from autovacuum.c to report startup of an autovacuum process.
1378 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1379 * the db OID must be passed in, instead.
1383 pgstat_report_autovac(Oid dboid)
1385 PgStat_MsgAutovacStart msg;
1387 if (pgStatSock == PGINVALID_SOCKET)
1390 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1391 msg.m_databaseid = dboid;
1392 msg.m_start_time = GetCurrentTimestamp();
1394 pgstat_send(&msg, sizeof(msg));
1399 * pgstat_report_vacuum() -
1401 * Tell the collector about the table we just vacuumed.
1405 pgstat_report_vacuum(Oid tableoid, bool shared,
1406 PgStat_Counter livetuples, PgStat_Counter deadtuples)
1408 PgStat_MsgVacuum msg;
1410 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1413 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1414 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1415 msg.m_tableoid = tableoid;
1416 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1417 msg.m_vacuumtime = GetCurrentTimestamp();
1418 msg.m_live_tuples = livetuples;
1419 msg.m_dead_tuples = deadtuples;
1420 pgstat_send(&msg, sizeof(msg));
1424 * pgstat_report_analyze() -
1426 * Tell the collector about the table we just analyzed.
1428 * Caller must provide new live- and dead-tuples estimates, as well as a
1429 * flag indicating whether to reset the changes_since_analyze counter.
1433 pgstat_report_analyze(Relation rel,
1434 PgStat_Counter livetuples, PgStat_Counter deadtuples,
1437 PgStat_MsgAnalyze msg;
1439 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1443 * Unlike VACUUM, ANALYZE might be running inside a transaction that has
1444 * already inserted and/or deleted rows in the target table. ANALYZE will
1445 * have counted such rows as live or dead respectively. Because we will
1446 * report our counts of such rows at transaction end, we should subtract
1447 * off these counts from what we send to the collector now, else they'll
1448 * be double-counted after commit. (This approach also ensures that the
1449 * collector ends up with the right numbers if we abort instead of
1452 if (rel->pgstat_info != NULL)
1454 PgStat_TableXactStatus *trans;
1456 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1458 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1459 deadtuples -= trans->tuples_updated + trans->tuples_deleted;
1461 /* count stuff inserted by already-aborted subxacts, too */
1462 deadtuples -= rel->pgstat_info->t_counts.t_delta_dead_tuples;
1463 /* Since ANALYZE's counts are estimates, we could have underflowed */
1464 livetuples = Max(livetuples, 0);
1465 deadtuples = Max(deadtuples, 0);
1468 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1469 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1470 msg.m_tableoid = RelationGetRelid(rel);
1471 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1472 msg.m_resetcounter = resetcounter;
1473 msg.m_analyzetime = GetCurrentTimestamp();
1474 msg.m_live_tuples = livetuples;
1475 msg.m_dead_tuples = deadtuples;
1476 pgstat_send(&msg, sizeof(msg));
1480 * pgstat_report_recovery_conflict() -
1482 * Tell the collector about a Hot Standby recovery conflict.
1486 pgstat_report_recovery_conflict(int reason)
1488 PgStat_MsgRecoveryConflict msg;
1490 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1493 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RECOVERYCONFLICT);
1494 msg.m_databaseid = MyDatabaseId;
1495 msg.m_reason = reason;
1496 pgstat_send(&msg, sizeof(msg));
1500 * pgstat_report_deadlock() -
1502 * Tell the collector about a deadlock detected.
1506 pgstat_report_deadlock(void)
1508 PgStat_MsgDeadlock msg;
1510 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1513 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DEADLOCK);
1514 msg.m_databaseid = MyDatabaseId;
1515 pgstat_send(&msg, sizeof(msg));
1519 * pgstat_report_tempfile() -
1521 * Tell the collector about a temporary file.
1525 pgstat_report_tempfile(size_t filesize)
1527 PgStat_MsgTempFile msg;
1529 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1532 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TEMPFILE);
1533 msg.m_databaseid = MyDatabaseId;
1534 msg.m_filesize = filesize;
1535 pgstat_send(&msg, sizeof(msg));
1542 * Send some junk data to the collector to increase traffic.
1548 PgStat_MsgDummy msg;
1550 if (pgStatSock == PGINVALID_SOCKET)
1553 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1554 pgstat_send(&msg, sizeof(msg));
1558 * pgstat_send_inquiry() -
1560 * Notify collector that we need fresh data.
1564 pgstat_send_inquiry(TimestampTz clock_time, TimestampTz cutoff_time, Oid databaseid)
1566 PgStat_MsgInquiry msg;
1568 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1569 msg.clock_time = clock_time;
1570 msg.cutoff_time = cutoff_time;
1571 msg.databaseid = databaseid;
1572 pgstat_send(&msg, sizeof(msg));
1577 * Initialize function call usage data.
1578 * Called by the executor before invoking a function.
1581 pgstat_init_function_usage(FunctionCallInfoData *fcinfo,
1582 PgStat_FunctionCallUsage *fcu)
1584 PgStat_BackendFunctionEntry *htabent;
1587 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1589 /* stats not wanted */
1594 if (!pgStatFunctions)
1596 /* First time through - initialize function stat table */
1599 memset(&hash_ctl, 0, sizeof(hash_ctl));
1600 hash_ctl.keysize = sizeof(Oid);
1601 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1602 pgStatFunctions = hash_create("Function stat entries",
1603 PGSTAT_FUNCTION_HASH_SIZE,
1605 HASH_ELEM | HASH_BLOBS);
1608 /* Get the stats entry for this function, create if necessary */
1609 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1610 HASH_ENTER, &found);
1612 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1614 fcu->fs = &htabent->f_counts;
1616 /* save stats for this function, later used to compensate for recursion */
1617 fcu->save_f_total_time = htabent->f_counts.f_total_time;
1619 /* save current backend-wide total time */
1620 fcu->save_total = total_func_time;
1622 /* get clock time as of function start */
1623 INSTR_TIME_SET_CURRENT(fcu->f_start);
1627 * find_funcstat_entry - find any existing PgStat_BackendFunctionEntry entry
1628 * for specified function
1630 * If no entry, return NULL, don't create a new one
1632 PgStat_BackendFunctionEntry *
1633 find_funcstat_entry(Oid func_id)
1635 if (pgStatFunctions == NULL)
1638 return (PgStat_BackendFunctionEntry *) hash_search(pgStatFunctions,
1644 * Calculate function call usage and update stat counters.
1645 * Called by the executor after invoking a function.
1647 * In the case of a set-returning function that runs in value-per-call mode,
1648 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1649 * calls for what the user considers a single call of the function. The
1650 * finalize flag should be TRUE on the last call.
1653 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1655 PgStat_FunctionCounts *fs = fcu->fs;
1657 instr_time f_others;
1660 /* stats not wanted? */
1664 /* total elapsed time in this function call */
1665 INSTR_TIME_SET_CURRENT(f_total);
1666 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1668 /* self usage: elapsed minus anything already charged to other calls */
1669 f_others = total_func_time;
1670 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1672 INSTR_TIME_SUBTRACT(f_self, f_others);
1674 /* update backend-wide total time */
1675 INSTR_TIME_ADD(total_func_time, f_self);
1678 * Compute the new f_total_time as the total elapsed time added to the
1679 * pre-call value of f_total_time. This is necessary to avoid
1680 * double-counting any time taken by recursive calls of myself. (We do
1681 * not need any similar kluge for self time, since that already excludes
1682 * any recursive calls.)
1684 INSTR_TIME_ADD(f_total, fcu->save_f_total_time);
1686 /* update counters in function stats table */
1689 fs->f_total_time = f_total;
1690 INSTR_TIME_ADD(fs->f_self_time, f_self);
1692 /* indicate that we have something to send */
1693 have_function_stats = true;
1698 * pgstat_initstats() -
1700 * Initialize a relcache entry to count access statistics.
1701 * Called whenever a relation is opened.
1703 * We assume that a relcache entry's pgstat_info field is zeroed by
1704 * relcache.c when the relcache entry is made; thereafter it is long-lived
1705 * data. We can avoid repeated searches of the TabStatus arrays when the
1706 * same relation is touched repeatedly within a transaction.
1710 pgstat_initstats(Relation rel)
1712 Oid rel_id = rel->rd_id;
1713 char relkind = rel->rd_rel->relkind;
1715 /* We only count stats for things that have storage */
1716 if (!(relkind == RELKIND_RELATION ||
1717 relkind == RELKIND_MATVIEW ||
1718 relkind == RELKIND_INDEX ||
1719 relkind == RELKIND_TOASTVALUE ||
1720 relkind == RELKIND_SEQUENCE))
1722 rel->pgstat_info = NULL;
1726 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1728 /* We're not counting at all */
1729 rel->pgstat_info = NULL;
1734 * If we already set up this relation in the current transaction, nothing
1737 if (rel->pgstat_info != NULL &&
1738 rel->pgstat_info->t_id == rel_id)
1741 /* Else find or make the PgStat_TableStatus entry, and update link */
1742 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1746 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1748 static PgStat_TableStatus *
1749 get_tabstat_entry(Oid rel_id, bool isshared)
1751 TabStatHashEntry *hash_entry;
1752 PgStat_TableStatus *entry;
1753 TabStatusArray *tsa;
1757 * Create hash table if we don't have it already.
1759 if (pgStatTabHash == NULL)
1763 memset(&ctl, 0, sizeof(ctl));
1764 ctl.keysize = sizeof(Oid);
1765 ctl.entrysize = sizeof(TabStatHashEntry);
1767 pgStatTabHash = hash_create("pgstat TabStatusArray lookup hash table",
1770 HASH_ELEM | HASH_BLOBS);
1774 * Find an entry or create a new one.
1776 hash_entry = hash_search(pgStatTabHash, &rel_id, HASH_ENTER, &found);
1779 /* initialize new entry with null pointer */
1780 hash_entry->tsa_entry = NULL;
1784 * If entry is already valid, we're done.
1786 if (hash_entry->tsa_entry)
1787 return hash_entry->tsa_entry;
1790 * Locate the first pgStatTabList entry with free space, making a new list
1791 * entry if needed. Note that we could get an OOM failure here, but if so
1792 * we have left the hashtable and the list in a consistent state.
1794 if (pgStatTabList == NULL)
1796 /* Set up first pgStatTabList entry */
1797 pgStatTabList = (TabStatusArray *)
1798 MemoryContextAllocZero(TopMemoryContext,
1799 sizeof(TabStatusArray));
1802 tsa = pgStatTabList;
1803 while (tsa->tsa_used >= TABSTAT_QUANTUM)
1805 if (tsa->tsa_next == NULL)
1806 tsa->tsa_next = (TabStatusArray *)
1807 MemoryContextAllocZero(TopMemoryContext,
1808 sizeof(TabStatusArray));
1809 tsa = tsa->tsa_next;
1813 * Allocate a PgStat_TableStatus entry within this list entry. We assume
1814 * the entry was already zeroed, either at creation or after last use.
1816 entry = &tsa->tsa_entries[tsa->tsa_used++];
1817 entry->t_id = rel_id;
1818 entry->t_shared = isshared;
1821 * Now we can fill the entry in pgStatTabHash.
1823 hash_entry->tsa_entry = entry;
1829 * find_tabstat_entry - find any existing PgStat_TableStatus entry for rel
1831 * If no entry, return NULL, don't create a new one
1833 * Note: if we got an error in the most recent execution of pgstat_report_stat,
1834 * it's possible that an entry exists but there's no hashtable entry for it.
1835 * That's okay, we'll treat this case as "doesn't exist".
1837 PgStat_TableStatus *
1838 find_tabstat_entry(Oid rel_id)
1840 TabStatHashEntry *hash_entry;
1842 /* If hashtable doesn't exist, there are no entries at all */
1846 hash_entry = hash_search(pgStatTabHash, &rel_id, HASH_FIND, NULL);
1850 /* Note that this step could also return NULL, but that's correct */
1851 return hash_entry->tsa_entry;
1855 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1857 static PgStat_SubXactStatus *
1858 get_tabstat_stack_level(int nest_level)
1860 PgStat_SubXactStatus *xact_state;
1862 xact_state = pgStatXactStack;
1863 if (xact_state == NULL || xact_state->nest_level != nest_level)
1865 xact_state = (PgStat_SubXactStatus *)
1866 MemoryContextAlloc(TopTransactionContext,
1867 sizeof(PgStat_SubXactStatus));
1868 xact_state->nest_level = nest_level;
1869 xact_state->prev = pgStatXactStack;
1870 xact_state->first = NULL;
1871 pgStatXactStack = xact_state;
1877 * add_tabstat_xact_level - add a new (sub)transaction state record
1880 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1882 PgStat_SubXactStatus *xact_state;
1883 PgStat_TableXactStatus *trans;
1886 * If this is the first rel to be modified at the current nest level, we
1887 * first have to push a transaction stack entry.
1889 xact_state = get_tabstat_stack_level(nest_level);
1891 /* Now make a per-table stack entry */
1892 trans = (PgStat_TableXactStatus *)
1893 MemoryContextAllocZero(TopTransactionContext,
1894 sizeof(PgStat_TableXactStatus));
1895 trans->nest_level = nest_level;
1896 trans->upper = pgstat_info->trans;
1897 trans->parent = pgstat_info;
1898 trans->next = xact_state->first;
1899 xact_state->first = trans;
1900 pgstat_info->trans = trans;
1904 * pgstat_count_heap_insert - count a tuple insertion of n tuples
1907 pgstat_count_heap_insert(Relation rel, PgStat_Counter n)
1909 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1911 if (pgstat_info != NULL)
1913 /* We have to log the effect at the proper transactional level */
1914 int nest_level = GetCurrentTransactionNestLevel();
1916 if (pgstat_info->trans == NULL ||
1917 pgstat_info->trans->nest_level != nest_level)
1918 add_tabstat_xact_level(pgstat_info, nest_level);
1920 pgstat_info->trans->tuples_inserted += n;
1925 * pgstat_count_heap_update - count a tuple update
1928 pgstat_count_heap_update(Relation rel, bool hot)
1930 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1932 if (pgstat_info != NULL)
1934 /* We have to log the effect at the proper transactional level */
1935 int nest_level = GetCurrentTransactionNestLevel();
1937 if (pgstat_info->trans == NULL ||
1938 pgstat_info->trans->nest_level != nest_level)
1939 add_tabstat_xact_level(pgstat_info, nest_level);
1941 pgstat_info->trans->tuples_updated++;
1943 /* t_tuples_hot_updated is nontransactional, so just advance it */
1945 pgstat_info->t_counts.t_tuples_hot_updated++;
1950 * pgstat_count_heap_delete - count a tuple deletion
1953 pgstat_count_heap_delete(Relation rel)
1955 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1957 if (pgstat_info != NULL)
1959 /* We have to log the effect at the proper transactional level */
1960 int nest_level = GetCurrentTransactionNestLevel();
1962 if (pgstat_info->trans == NULL ||
1963 pgstat_info->trans->nest_level != nest_level)
1964 add_tabstat_xact_level(pgstat_info, nest_level);
1966 pgstat_info->trans->tuples_deleted++;
1971 * pgstat_truncate_save_counters
1973 * Whenever a table is truncated, we save its i/u/d counters so that they can
1974 * be cleared, and if the (sub)xact that executed the truncate later aborts,
1975 * the counters can be restored to the saved (pre-truncate) values. Note we do
1976 * this on the first truncate in any particular subxact level only.
1979 pgstat_truncate_save_counters(PgStat_TableXactStatus *trans)
1981 if (!trans->truncated)
1983 trans->inserted_pre_trunc = trans->tuples_inserted;
1984 trans->updated_pre_trunc = trans->tuples_updated;
1985 trans->deleted_pre_trunc = trans->tuples_deleted;
1986 trans->truncated = true;
1991 * pgstat_truncate_restore_counters - restore counters when a truncate aborts
1994 pgstat_truncate_restore_counters(PgStat_TableXactStatus *trans)
1996 if (trans->truncated)
1998 trans->tuples_inserted = trans->inserted_pre_trunc;
1999 trans->tuples_updated = trans->updated_pre_trunc;
2000 trans->tuples_deleted = trans->deleted_pre_trunc;
2005 * pgstat_count_truncate - update tuple counters due to truncate
2008 pgstat_count_truncate(Relation rel)
2010 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
2012 if (pgstat_info != NULL)
2014 /* We have to log the effect at the proper transactional level */
2015 int nest_level = GetCurrentTransactionNestLevel();
2017 if (pgstat_info->trans == NULL ||
2018 pgstat_info->trans->nest_level != nest_level)
2019 add_tabstat_xact_level(pgstat_info, nest_level);
2021 pgstat_truncate_save_counters(pgstat_info->trans);
2022 pgstat_info->trans->tuples_inserted = 0;
2023 pgstat_info->trans->tuples_updated = 0;
2024 pgstat_info->trans->tuples_deleted = 0;
2029 * pgstat_update_heap_dead_tuples - update dead-tuples count
2031 * The semantics of this are that we are reporting the nontransactional
2032 * recovery of "delta" dead tuples; so t_delta_dead_tuples decreases
2033 * rather than increasing, and the change goes straight into the per-table
2034 * counter, not into transactional state.
2037 pgstat_update_heap_dead_tuples(Relation rel, int delta)
2039 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
2041 if (pgstat_info != NULL)
2042 pgstat_info->t_counts.t_delta_dead_tuples -= delta;
2049 * Called from access/transam/xact.c at top-level transaction commit/abort.
2053 AtEOXact_PgStat(bool isCommit)
2055 PgStat_SubXactStatus *xact_state;
2058 * Count transaction commit or abort. (We use counters, not just bools,
2059 * in case the reporting message isn't sent right away.)
2064 pgStatXactRollback++;
2067 * Transfer transactional insert/update counts into the base tabstat
2068 * entries. We don't bother to free any of the transactional state, since
2069 * it's all in TopTransactionContext and will go away anyway.
2071 xact_state = pgStatXactStack;
2072 if (xact_state != NULL)
2074 PgStat_TableXactStatus *trans;
2076 Assert(xact_state->nest_level == 1);
2077 Assert(xact_state->prev == NULL);
2078 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2080 PgStat_TableStatus *tabstat;
2082 Assert(trans->nest_level == 1);
2083 Assert(trans->upper == NULL);
2084 tabstat = trans->parent;
2085 Assert(tabstat->trans == trans);
2086 /* restore pre-truncate stats (if any) in case of aborted xact */
2088 pgstat_truncate_restore_counters(trans);
2089 /* count attempted actions regardless of commit/abort */
2090 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
2091 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
2092 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
2095 tabstat->t_counts.t_truncated = trans->truncated;
2096 if (trans->truncated)
2098 /* forget live/dead stats seen by backend thus far */
2099 tabstat->t_counts.t_delta_live_tuples = 0;
2100 tabstat->t_counts.t_delta_dead_tuples = 0;
2102 /* insert adds a live tuple, delete removes one */
2103 tabstat->t_counts.t_delta_live_tuples +=
2104 trans->tuples_inserted - trans->tuples_deleted;
2105 /* update and delete each create a dead tuple */
2106 tabstat->t_counts.t_delta_dead_tuples +=
2107 trans->tuples_updated + trans->tuples_deleted;
2108 /* insert, update, delete each count as one change event */
2109 tabstat->t_counts.t_changed_tuples +=
2110 trans->tuples_inserted + trans->tuples_updated +
2111 trans->tuples_deleted;
2115 /* inserted tuples are dead, deleted tuples are unaffected */
2116 tabstat->t_counts.t_delta_dead_tuples +=
2117 trans->tuples_inserted + trans->tuples_updated;
2118 /* an aborted xact generates no changed_tuple events */
2120 tabstat->trans = NULL;
2123 pgStatXactStack = NULL;
2125 /* Make sure any stats snapshot is thrown away */
2126 pgstat_clear_snapshot();
2130 * AtEOSubXact_PgStat
2132 * Called from access/transam/xact.c at subtransaction commit/abort.
2136 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
2138 PgStat_SubXactStatus *xact_state;
2141 * Transfer transactional insert/update counts into the next higher
2142 * subtransaction state.
2144 xact_state = pgStatXactStack;
2145 if (xact_state != NULL &&
2146 xact_state->nest_level >= nestDepth)
2148 PgStat_TableXactStatus *trans;
2149 PgStat_TableXactStatus *next_trans;
2151 /* delink xact_state from stack immediately to simplify reuse case */
2152 pgStatXactStack = xact_state->prev;
2154 for (trans = xact_state->first; trans != NULL; trans = next_trans)
2156 PgStat_TableStatus *tabstat;
2158 next_trans = trans->next;
2159 Assert(trans->nest_level == nestDepth);
2160 tabstat = trans->parent;
2161 Assert(tabstat->trans == trans);
2164 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
2166 if (trans->truncated)
2168 /* propagate the truncate status one level up */
2169 pgstat_truncate_save_counters(trans->upper);
2170 /* replace upper xact stats with ours */
2171 trans->upper->tuples_inserted = trans->tuples_inserted;
2172 trans->upper->tuples_updated = trans->tuples_updated;
2173 trans->upper->tuples_deleted = trans->tuples_deleted;
2177 trans->upper->tuples_inserted += trans->tuples_inserted;
2178 trans->upper->tuples_updated += trans->tuples_updated;
2179 trans->upper->tuples_deleted += trans->tuples_deleted;
2181 tabstat->trans = trans->upper;
2187 * When there isn't an immediate parent state, we can just
2188 * reuse the record instead of going through a
2189 * palloc/pfree pushup (this works since it's all in
2190 * TopTransactionContext anyway). We have to re-link it
2191 * into the parent level, though, and that might mean
2192 * pushing a new entry into the pgStatXactStack.
2194 PgStat_SubXactStatus *upper_xact_state;
2196 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
2197 trans->next = upper_xact_state->first;
2198 upper_xact_state->first = trans;
2199 trans->nest_level = nestDepth - 1;
2205 * On abort, update top-level tabstat counts, then forget the
2209 /* first restore values obliterated by truncate */
2210 pgstat_truncate_restore_counters(trans);
2211 /* count attempted actions regardless of commit/abort */
2212 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
2213 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
2214 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
2215 /* inserted tuples are dead, deleted tuples are unaffected */
2216 tabstat->t_counts.t_delta_dead_tuples +=
2217 trans->tuples_inserted + trans->tuples_updated;
2218 tabstat->trans = trans->upper;
2229 * Save the transactional stats state at 2PC transaction prepare.
2231 * In this phase we just generate 2PC records for all the pending
2232 * transaction-dependent stats work.
2235 AtPrepare_PgStat(void)
2237 PgStat_SubXactStatus *xact_state;
2239 xact_state = pgStatXactStack;
2240 if (xact_state != NULL)
2242 PgStat_TableXactStatus *trans;
2244 Assert(xact_state->nest_level == 1);
2245 Assert(xact_state->prev == NULL);
2246 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2248 PgStat_TableStatus *tabstat;
2249 TwoPhasePgStatRecord record;
2251 Assert(trans->nest_level == 1);
2252 Assert(trans->upper == NULL);
2253 tabstat = trans->parent;
2254 Assert(tabstat->trans == trans);
2256 record.tuples_inserted = trans->tuples_inserted;
2257 record.tuples_updated = trans->tuples_updated;
2258 record.tuples_deleted = trans->tuples_deleted;
2259 record.inserted_pre_trunc = trans->inserted_pre_trunc;
2260 record.updated_pre_trunc = trans->updated_pre_trunc;
2261 record.deleted_pre_trunc = trans->deleted_pre_trunc;
2262 record.t_id = tabstat->t_id;
2263 record.t_shared = tabstat->t_shared;
2264 record.t_truncated = trans->truncated;
2266 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
2267 &record, sizeof(TwoPhasePgStatRecord));
2273 * PostPrepare_PgStat
2274 * Clean up after successful PREPARE.
2276 * All we need do here is unlink the transaction stats state from the
2277 * nontransactional state. The nontransactional action counts will be
2278 * reported to the stats collector immediately, while the effects on live
2279 * and dead tuple counts are preserved in the 2PC state file.
2281 * Note: AtEOXact_PgStat is not called during PREPARE.
2284 PostPrepare_PgStat(void)
2286 PgStat_SubXactStatus *xact_state;
2289 * We don't bother to free any of the transactional state, since it's all
2290 * in TopTransactionContext and will go away anyway.
2292 xact_state = pgStatXactStack;
2293 if (xact_state != NULL)
2295 PgStat_TableXactStatus *trans;
2297 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2299 PgStat_TableStatus *tabstat;
2301 tabstat = trans->parent;
2302 tabstat->trans = NULL;
2305 pgStatXactStack = NULL;
2307 /* Make sure any stats snapshot is thrown away */
2308 pgstat_clear_snapshot();
2312 * 2PC processing routine for COMMIT PREPARED case.
2314 * Load the saved counts into our local pgstats state.
2317 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
2318 void *recdata, uint32 len)
2320 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2321 PgStat_TableStatus *pgstat_info;
2323 /* Find or create a tabstat entry for the rel */
2324 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2326 /* Same math as in AtEOXact_PgStat, commit case */
2327 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2328 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2329 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2330 pgstat_info->t_counts.t_truncated = rec->t_truncated;
2331 if (rec->t_truncated)
2333 /* forget live/dead stats seen by backend thus far */
2334 pgstat_info->t_counts.t_delta_live_tuples = 0;
2335 pgstat_info->t_counts.t_delta_dead_tuples = 0;
2337 pgstat_info->t_counts.t_delta_live_tuples +=
2338 rec->tuples_inserted - rec->tuples_deleted;
2339 pgstat_info->t_counts.t_delta_dead_tuples +=
2340 rec->tuples_updated + rec->tuples_deleted;
2341 pgstat_info->t_counts.t_changed_tuples +=
2342 rec->tuples_inserted + rec->tuples_updated +
2343 rec->tuples_deleted;
2347 * 2PC processing routine for ROLLBACK PREPARED case.
2349 * Load the saved counts into our local pgstats state, but treat them
2353 pgstat_twophase_postabort(TransactionId xid, uint16 info,
2354 void *recdata, uint32 len)
2356 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2357 PgStat_TableStatus *pgstat_info;
2359 /* Find or create a tabstat entry for the rel */
2360 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2362 /* Same math as in AtEOXact_PgStat, abort case */
2363 if (rec->t_truncated)
2365 rec->tuples_inserted = rec->inserted_pre_trunc;
2366 rec->tuples_updated = rec->updated_pre_trunc;
2367 rec->tuples_deleted = rec->deleted_pre_trunc;
2369 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2370 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2371 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2372 pgstat_info->t_counts.t_delta_dead_tuples +=
2373 rec->tuples_inserted + rec->tuples_updated;
2378 * pgstat_fetch_stat_dbentry() -
2380 * Support function for the SQL-callable pgstat* functions. Returns
2381 * the collected statistics for one database or NULL. NULL doesn't mean
2382 * that the database doesn't exist, it is just not yet known by the
2383 * collector, so the caller is better off to report ZERO instead.
2386 PgStat_StatDBEntry *
2387 pgstat_fetch_stat_dbentry(Oid dbid)
2390 * If not done for this transaction, read the statistics collector stats
2391 * file into some hash tables.
2393 backend_read_statsfile();
2396 * Lookup the requested database; return NULL if not found
2398 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2405 * pgstat_fetch_stat_tabentry() -
2407 * Support function for the SQL-callable pgstat* functions. Returns
2408 * the collected statistics for one table or NULL. NULL doesn't mean
2409 * that the table doesn't exist, it is just not yet known by the
2410 * collector, so the caller is better off to report ZERO instead.
2413 PgStat_StatTabEntry *
2414 pgstat_fetch_stat_tabentry(Oid relid)
2417 PgStat_StatDBEntry *dbentry;
2418 PgStat_StatTabEntry *tabentry;
2421 * If not done for this transaction, read the statistics collector stats
2422 * file into some hash tables.
2424 backend_read_statsfile();
2427 * Lookup our database, then look in its table hash table.
2429 dbid = MyDatabaseId;
2430 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2433 if (dbentry != NULL && dbentry->tables != NULL)
2435 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2443 * If we didn't find it, maybe it's a shared table.
2446 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2449 if (dbentry != NULL && dbentry->tables != NULL)
2451 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2463 * pgstat_fetch_stat_funcentry() -
2465 * Support function for the SQL-callable pgstat* functions. Returns
2466 * the collected statistics for one function or NULL.
2469 PgStat_StatFuncEntry *
2470 pgstat_fetch_stat_funcentry(Oid func_id)
2472 PgStat_StatDBEntry *dbentry;
2473 PgStat_StatFuncEntry *funcentry = NULL;
2475 /* load the stats file if needed */
2476 backend_read_statsfile();
2478 /* Lookup our database, then find the requested function. */
2479 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
2480 if (dbentry != NULL && dbentry->functions != NULL)
2482 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2492 * pgstat_fetch_stat_beentry() -
2494 * Support function for the SQL-callable pgstat* functions. Returns
2495 * our local copy of the current-activity entry for one backend.
2497 * NB: caller is responsible for a check if the user is permitted to see
2498 * this info (especially the querystring).
2502 pgstat_fetch_stat_beentry(int beid)
2504 pgstat_read_current_status();
2506 if (beid < 1 || beid > localNumBackends)
2509 return &localBackendStatusTable[beid - 1].backendStatus;
2514 * pgstat_fetch_stat_local_beentry() -
2516 * Like pgstat_fetch_stat_beentry() but with locally computed additions (like
2517 * xid and xmin values of the backend)
2519 * NB: caller is responsible for a check if the user is permitted to see
2520 * this info (especially the querystring).
2523 LocalPgBackendStatus *
2524 pgstat_fetch_stat_local_beentry(int beid)
2526 pgstat_read_current_status();
2528 if (beid < 1 || beid > localNumBackends)
2531 return &localBackendStatusTable[beid - 1];
2536 * pgstat_fetch_stat_numbackends() -
2538 * Support function for the SQL-callable pgstat* functions. Returns
2539 * the maximum current backend id.
2543 pgstat_fetch_stat_numbackends(void)
2545 pgstat_read_current_status();
2547 return localNumBackends;
2552 * pgstat_fetch_stat_archiver() -
2554 * Support function for the SQL-callable pgstat* functions. Returns
2555 * a pointer to the archiver statistics struct.
2558 PgStat_ArchiverStats *
2559 pgstat_fetch_stat_archiver(void)
2561 backend_read_statsfile();
2563 return &archiverStats;
2569 * pgstat_fetch_global() -
2571 * Support function for the SQL-callable pgstat* functions. Returns
2572 * a pointer to the global statistics struct.
2575 PgStat_GlobalStats *
2576 pgstat_fetch_global(void)
2578 backend_read_statsfile();
2580 return &globalStats;
2584 /* ------------------------------------------------------------
2585 * Functions for management of the shared-memory PgBackendStatus array
2586 * ------------------------------------------------------------
2589 static PgBackendStatus *BackendStatusArray = NULL;
2590 static PgBackendStatus *MyBEEntry = NULL;
2591 static char *BackendAppnameBuffer = NULL;
2592 static char *BackendClientHostnameBuffer = NULL;
2593 static char *BackendActivityBuffer = NULL;
2594 static Size BackendActivityBufferSize = 0;
2596 static PgBackendSSLStatus *BackendSslStatusBuffer = NULL;
2601 * Report shared-memory space needed by CreateSharedBackendStatus.
2604 BackendStatusShmemSize(void)
2608 /* BackendStatusArray: */
2609 size = mul_size(sizeof(PgBackendStatus), NumBackendStatSlots);
2610 /* BackendAppnameBuffer: */
2611 size = add_size(size,
2612 mul_size(NAMEDATALEN, NumBackendStatSlots));
2613 /* BackendClientHostnameBuffer: */
2614 size = add_size(size,
2615 mul_size(NAMEDATALEN, NumBackendStatSlots));
2616 /* BackendActivityBuffer: */
2617 size = add_size(size,
2618 mul_size(pgstat_track_activity_query_size, NumBackendStatSlots));
2620 /* BackendSslStatusBuffer: */
2621 size = add_size(size,
2622 mul_size(sizeof(PgBackendSSLStatus), NumBackendStatSlots));
2628 * Initialize the shared status array and several string buffers
2629 * during postmaster startup.
2632 CreateSharedBackendStatus(void)
2639 /* Create or attach to the shared array */
2640 size = mul_size(sizeof(PgBackendStatus), NumBackendStatSlots);
2641 BackendStatusArray = (PgBackendStatus *)
2642 ShmemInitStruct("Backend Status Array", size, &found);
2647 * We're the first - initialize.
2649 MemSet(BackendStatusArray, 0, size);
2652 /* Create or attach to the shared appname buffer */
2653 size = mul_size(NAMEDATALEN, MaxBackends);
2654 BackendAppnameBuffer = (char *)
2655 ShmemInitStruct("Backend Application Name Buffer", size, &found);
2659 MemSet(BackendAppnameBuffer, 0, size);
2661 /* Initialize st_appname pointers. */
2662 buffer = BackendAppnameBuffer;
2663 for (i = 0; i < NumBackendStatSlots; i++)
2665 BackendStatusArray[i].st_appname = buffer;
2666 buffer += NAMEDATALEN;
2670 /* Create or attach to the shared client hostname buffer */
2671 size = mul_size(NAMEDATALEN, MaxBackends);
2672 BackendClientHostnameBuffer = (char *)
2673 ShmemInitStruct("Backend Client Host Name Buffer", size, &found);
2677 MemSet(BackendClientHostnameBuffer, 0, size);
2679 /* Initialize st_clienthostname pointers. */
2680 buffer = BackendClientHostnameBuffer;
2681 for (i = 0; i < NumBackendStatSlots; i++)
2683 BackendStatusArray[i].st_clienthostname = buffer;
2684 buffer += NAMEDATALEN;
2688 /* Create or attach to the shared activity buffer */
2689 BackendActivityBufferSize = mul_size(pgstat_track_activity_query_size,
2690 NumBackendStatSlots);
2691 BackendActivityBuffer = (char *)
2692 ShmemInitStruct("Backend Activity Buffer",
2693 BackendActivityBufferSize,
2698 MemSet(BackendActivityBuffer, 0, size);
2700 /* Initialize st_activity pointers. */
2701 buffer = BackendActivityBuffer;
2702 for (i = 0; i < NumBackendStatSlots; i++)
2704 BackendStatusArray[i].st_activity_raw = buffer;
2705 buffer += pgstat_track_activity_query_size;
2710 /* Create or attach to the shared SSL status buffer */
2711 size = mul_size(sizeof(PgBackendSSLStatus), NumBackendStatSlots);
2712 BackendSslStatusBuffer = (PgBackendSSLStatus *)
2713 ShmemInitStruct("Backend SSL Status Buffer", size, &found);
2717 PgBackendSSLStatus *ptr;
2719 MemSet(BackendSslStatusBuffer, 0, size);
2721 /* Initialize st_sslstatus pointers. */
2722 ptr = BackendSslStatusBuffer;
2723 for (i = 0; i < NumBackendStatSlots; i++)
2725 BackendStatusArray[i].st_sslstatus = ptr;
2734 * pgstat_initialize() -
2736 * Initialize pgstats state, and set up our on-proc-exit hook.
2737 * Called from InitPostgres and AuxiliaryProcessMain. For auxiliary process,
2738 * MyBackendId is invalid. Otherwise, MyBackendId must be set,
2739 * but we must not have started any transaction yet (since the
2740 * exit hook must run after the last transaction exit).
2741 * NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful.
2745 pgstat_initialize(void)
2747 /* Initialize MyBEEntry */
2748 if (MyBackendId != InvalidBackendId)
2750 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2751 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2755 /* Must be an auxiliary process */
2756 Assert(MyAuxProcType != NotAnAuxProcess);
2759 * Assign the MyBEEntry for an auxiliary process. Since it doesn't
2760 * have a BackendId, the slot is statically allocated based on the
2761 * auxiliary process type (MyAuxProcType). Backends use slots indexed
2762 * in the range from 1 to MaxBackends (inclusive), so we use
2763 * MaxBackends + AuxBackendType + 1 as the index of the slot for an
2764 * auxiliary process.
2766 MyBEEntry = &BackendStatusArray[MaxBackends + MyAuxProcType];
2769 /* Set up a process-exit hook to clean up */
2770 on_shmem_exit(pgstat_beshutdown_hook, 0);
2774 * pgstat_bestart() -
2776 * Initialize this backend's entry in the PgBackendStatus array.
2777 * Called from InitPostgres.
2779 * Apart from auxiliary processes, MyBackendId, MyDatabaseId,
2780 * session userid, and application_name must be set for a
2781 * backend (hence, this cannot be combined with pgstat_initialize).
2785 pgstat_bestart(void)
2787 TimestampTz proc_start_timestamp;
2788 SockAddr clientaddr;
2789 volatile PgBackendStatus *beentry;
2792 * To minimize the time spent modifying the PgBackendStatus entry, fetch
2793 * all the needed data first.
2795 * If we have a MyProcPort, use its session start time (for consistency,
2796 * and to save a kernel call).
2799 proc_start_timestamp = MyProcPort->SessionStartTime;
2801 proc_start_timestamp = GetCurrentTimestamp();
2804 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2805 * If so, use all-zeroes client address, which is dealt with specially in
2806 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2809 memcpy(&clientaddr, &MyProcPort->raddr, sizeof(clientaddr));
2811 MemSet(&clientaddr, 0, sizeof(clientaddr));
2814 * Initialize my status entry, following the protocol of bumping
2815 * st_changecount before and after; and make sure it's even afterwards. We
2816 * use a volatile pointer here to ensure the compiler doesn't try to get
2819 beentry = MyBEEntry;
2821 /* pgstats state must be initialized from pgstat_initialize() */
2822 Assert(beentry != NULL);
2824 if (MyBackendId != InvalidBackendId)
2826 if (IsAutoVacuumLauncherProcess())
2828 /* Autovacuum Launcher */
2829 beentry->st_backendType = B_AUTOVAC_LAUNCHER;
2831 else if (IsAutoVacuumWorkerProcess())
2833 /* Autovacuum Worker */
2834 beentry->st_backendType = B_AUTOVAC_WORKER;
2836 else if (am_walsender)
2839 beentry->st_backendType = B_WAL_SENDER;
2841 else if (IsBackgroundWorker)
2844 beentry->st_backendType = B_BG_WORKER;
2848 /* client-backend */
2849 beentry->st_backendType = B_BACKEND;
2854 /* Must be an auxiliary process */
2855 Assert(MyAuxProcType != NotAnAuxProcess);
2856 switch (MyAuxProcType)
2858 case StartupProcess:
2859 beentry->st_backendType = B_STARTUP;
2861 case BgWriterProcess:
2862 beentry->st_backendType = B_BG_WRITER;
2864 case CheckpointerProcess:
2865 beentry->st_backendType = B_CHECKPOINTER;
2867 case WalWriterProcess:
2868 beentry->st_backendType = B_WAL_WRITER;
2870 case WalReceiverProcess:
2871 beentry->st_backendType = B_WAL_RECEIVER;
2874 elog(FATAL, "unrecognized process type: %d",
2875 (int) MyAuxProcType);
2882 pgstat_increment_changecount_before(beentry);
2883 } while ((beentry->st_changecount & 1) == 0);
2885 beentry->st_procpid = MyProcPid;
2886 beentry->st_proc_start_timestamp = proc_start_timestamp;
2887 beentry->st_activity_start_timestamp = 0;
2888 beentry->st_state_start_timestamp = 0;
2889 beentry->st_xact_start_timestamp = 0;
2890 beentry->st_databaseid = MyDatabaseId;
2892 /* We have userid for client-backends, wal-sender and bgworker processes */
2893 if (beentry->st_backendType == B_BACKEND
2894 || beentry->st_backendType == B_WAL_SENDER
2895 || beentry->st_backendType == B_BG_WORKER)
2896 beentry->st_userid = GetSessionUserId();
2898 beentry->st_userid = InvalidOid;
2900 beentry->st_clientaddr = clientaddr;
2901 if (MyProcPort && MyProcPort->remote_hostname)
2902 strlcpy(beentry->st_clienthostname, MyProcPort->remote_hostname,
2905 beentry->st_clienthostname[0] = '\0';
2907 if (MyProcPort && MyProcPort->ssl != NULL)
2909 beentry->st_ssl = true;
2910 beentry->st_sslstatus->ssl_bits = be_tls_get_cipher_bits(MyProcPort);
2911 beentry->st_sslstatus->ssl_compression = be_tls_get_compression(MyProcPort);
2912 strlcpy(beentry->st_sslstatus->ssl_version, be_tls_get_version(MyProcPort), NAMEDATALEN);
2913 strlcpy(beentry->st_sslstatus->ssl_cipher, be_tls_get_cipher(MyProcPort), NAMEDATALEN);
2914 be_tls_get_peerdn_name(MyProcPort, beentry->st_sslstatus->ssl_clientdn, NAMEDATALEN);
2918 beentry->st_ssl = false;
2921 beentry->st_ssl = false;
2923 beentry->st_state = STATE_UNDEFINED;
2924 beentry->st_appname[0] = '\0';
2925 beentry->st_activity_raw[0] = '\0';
2926 /* Also make sure the last byte in each string area is always 0 */
2927 beentry->st_clienthostname[NAMEDATALEN - 1] = '\0';
2928 beentry->st_appname[NAMEDATALEN - 1] = '\0';
2929 beentry->st_activity_raw[pgstat_track_activity_query_size - 1] = '\0';
2930 beentry->st_progress_command = PROGRESS_COMMAND_INVALID;
2931 beentry->st_progress_command_target = InvalidOid;
2934 * we don't zero st_progress_param here to save cycles; nobody should
2935 * examine it until st_progress_command has been set to something other
2936 * than PROGRESS_COMMAND_INVALID
2939 pgstat_increment_changecount_after(beentry);
2941 /* Update app name to current GUC setting */
2942 if (application_name)
2943 pgstat_report_appname(application_name);
2947 * Shut down a single backend's statistics reporting at process exit.
2949 * Flush any remaining statistics counts out to the collector.
2950 * Without this, operations triggered during backend exit (such as
2951 * temp table deletions) won't be counted.
2953 * Lastly, clear out our entry in the PgBackendStatus array.
2956 pgstat_beshutdown_hook(int code, Datum arg)
2958 volatile PgBackendStatus *beentry = MyBEEntry;
2961 * If we got as far as discovering our own database ID, we can report what
2962 * we did to the collector. Otherwise, we'd be sending an invalid
2963 * database ID, so forget it. (This means that accesses to pg_database
2964 * during failed backend starts might never get counted.)
2966 if (OidIsValid(MyDatabaseId))
2967 pgstat_report_stat(true);
2970 * Clear my status entry, following the protocol of bumping st_changecount
2971 * before and after. We use a volatile pointer here to ensure the
2972 * compiler doesn't try to get cute.
2974 pgstat_increment_changecount_before(beentry);
2976 beentry->st_procpid = 0; /* mark invalid */
2978 pgstat_increment_changecount_after(beentry);
2983 * pgstat_report_activity() -
2985 * Called from tcop/postgres.c to report what the backend is actually doing
2986 * (but note cmd_str can be NULL for certain cases).
2988 * All updates of the status entry follow the protocol of bumping
2989 * st_changecount before and after. We use a volatile pointer here to
2990 * ensure the compiler doesn't try to get cute.
2994 pgstat_report_activity(BackendState state, const char *cmd_str)
2996 volatile PgBackendStatus *beentry = MyBEEntry;
2997 TimestampTz start_timestamp;
2998 TimestampTz current_timestamp;
3001 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
3006 if (!pgstat_track_activities)
3008 if (beentry->st_state != STATE_DISABLED)
3010 volatile PGPROC *proc = MyProc;
3013 * track_activities is disabled, but we last reported a
3014 * non-disabled state. As our final update, change the state and
3015 * clear fields we will not be updating anymore.
3017 pgstat_increment_changecount_before(beentry);
3018 beentry->st_state = STATE_DISABLED;
3019 beentry->st_state_start_timestamp = 0;
3020 beentry->st_activity_raw[0] = '\0';
3021 beentry->st_activity_start_timestamp = 0;
3022 /* st_xact_start_timestamp and wait_event_info are also disabled */
3023 beentry->st_xact_start_timestamp = 0;
3024 proc->wait_event_info = 0;
3025 pgstat_increment_changecount_after(beentry);
3031 * To minimize the time spent modifying the entry, fetch all the needed
3034 start_timestamp = GetCurrentStatementStartTimestamp();
3035 if (cmd_str != NULL)
3038 * Compute length of to-be-stored string unaware of multi-byte
3039 * characters. For speed reasons that'll get corrected on read, rather
3040 * than computed every write.
3042 len = Min(strlen(cmd_str), pgstat_track_activity_query_size - 1);
3044 current_timestamp = GetCurrentTimestamp();
3047 * Now update the status entry
3049 pgstat_increment_changecount_before(beentry);
3051 beentry->st_state = state;
3052 beentry->st_state_start_timestamp = current_timestamp;
3054 if (cmd_str != NULL)
3056 memcpy((char *) beentry->st_activity_raw, cmd_str, len);
3057 beentry->st_activity_raw[len] = '\0';
3058 beentry->st_activity_start_timestamp = start_timestamp;
3061 pgstat_increment_changecount_after(beentry);
3065 * pgstat_progress_start_command() -
3067 * Set st_progress_command (and st_progress_command_target) in own backend
3068 * entry. Also, zero-initialize st_progress_param array.
3072 pgstat_progress_start_command(ProgressCommandType cmdtype, Oid relid)
3074 volatile PgBackendStatus *beentry = MyBEEntry;
3076 if (!beentry || !pgstat_track_activities)
3079 pgstat_increment_changecount_before(beentry);
3080 beentry->st_progress_command = cmdtype;
3081 beentry->st_progress_command_target = relid;
3082 MemSet(&beentry->st_progress_param, 0, sizeof(beentry->st_progress_param));
3083 pgstat_increment_changecount_after(beentry);
3087 * pgstat_progress_update_param() -
3089 * Update index'th member in st_progress_param[] of own backend entry.
3093 pgstat_progress_update_param(int index, int64 val)
3095 volatile PgBackendStatus *beentry = MyBEEntry;
3097 Assert(index >= 0 && index < PGSTAT_NUM_PROGRESS_PARAM);
3099 if (!beentry || !pgstat_track_activities)
3102 pgstat_increment_changecount_before(beentry);
3103 beentry->st_progress_param[index] = val;
3104 pgstat_increment_changecount_after(beentry);
3108 * pgstat_progress_update_multi_param() -
3110 * Update multiple members in st_progress_param[] of own backend entry.
3111 * This is atomic; readers won't see intermediate states.
3115 pgstat_progress_update_multi_param(int nparam, const int *index,
3118 volatile PgBackendStatus *beentry = MyBEEntry;
3121 if (!beentry || !pgstat_track_activities || nparam == 0)
3124 pgstat_increment_changecount_before(beentry);
3126 for (i = 0; i < nparam; ++i)
3128 Assert(index[i] >= 0 && index[i] < PGSTAT_NUM_PROGRESS_PARAM);
3130 beentry->st_progress_param[index[i]] = val[i];
3133 pgstat_increment_changecount_after(beentry);
3137 * pgstat_progress_end_command() -
3139 * Reset st_progress_command (and st_progress_command_target) in own backend
3140 * entry. This signals the end of the command.
3144 pgstat_progress_end_command(void)
3146 volatile PgBackendStatus *beentry = MyBEEntry;
3150 if (!pgstat_track_activities
3151 && beentry->st_progress_command == PROGRESS_COMMAND_INVALID)
3154 pgstat_increment_changecount_before(beentry);
3155 beentry->st_progress_command = PROGRESS_COMMAND_INVALID;
3156 beentry->st_progress_command_target = InvalidOid;
3157 pgstat_increment_changecount_after(beentry);
3161 * pgstat_report_appname() -
3163 * Called to update our application name.
3167 pgstat_report_appname(const char *appname)
3169 volatile PgBackendStatus *beentry = MyBEEntry;
3175 /* This should be unnecessary if GUC did its job, but be safe */
3176 len = pg_mbcliplen(appname, strlen(appname), NAMEDATALEN - 1);
3179 * Update my status entry, following the protocol of bumping
3180 * st_changecount before and after. We use a volatile pointer here to
3181 * ensure the compiler doesn't try to get cute.
3183 pgstat_increment_changecount_before(beentry);
3185 memcpy((char *) beentry->st_appname, appname, len);
3186 beentry->st_appname[len] = '\0';
3188 pgstat_increment_changecount_after(beentry);
3192 * Report current transaction start timestamp as the specified value.
3193 * Zero means there is no active transaction.
3196 pgstat_report_xact_timestamp(TimestampTz tstamp)
3198 volatile PgBackendStatus *beentry = MyBEEntry;
3200 if (!pgstat_track_activities || !beentry)
3204 * Update my status entry, following the protocol of bumping
3205 * st_changecount before and after. We use a volatile pointer here to
3206 * ensure the compiler doesn't try to get cute.
3208 pgstat_increment_changecount_before(beentry);
3209 beentry->st_xact_start_timestamp = tstamp;
3210 pgstat_increment_changecount_after(beentry);
3214 * pgstat_read_current_status() -
3216 * Copy the current contents of the PgBackendStatus array to local memory,
3217 * if not already done in this transaction.
3221 pgstat_read_current_status(void)
3223 volatile PgBackendStatus *beentry;
3224 LocalPgBackendStatus *localtable;
3225 LocalPgBackendStatus *localentry;
3227 *localclienthostname,
3230 PgBackendSSLStatus *localsslstatus;
3234 Assert(!pgStatRunningInCollector);
3235 if (localBackendStatusTable)
3236 return; /* already done */
3238 pgstat_setup_memcxt();
3240 localtable = (LocalPgBackendStatus *)
3241 MemoryContextAlloc(pgStatLocalContext,
3242 sizeof(LocalPgBackendStatus) * NumBackendStatSlots);
3243 localappname = (char *)
3244 MemoryContextAlloc(pgStatLocalContext,
3245 NAMEDATALEN * NumBackendStatSlots);
3246 localclienthostname = (char *)
3247 MemoryContextAlloc(pgStatLocalContext,
3248 NAMEDATALEN * NumBackendStatSlots);
3249 localactivity = (char *)
3250 MemoryContextAlloc(pgStatLocalContext,
3251 pgstat_track_activity_query_size * NumBackendStatSlots);
3253 localsslstatus = (PgBackendSSLStatus *)
3254 MemoryContextAlloc(pgStatLocalContext,
3255 sizeof(PgBackendSSLStatus) * NumBackendStatSlots);
3258 localNumBackends = 0;
3260 beentry = BackendStatusArray;
3261 localentry = localtable;
3262 for (i = 1; i <= NumBackendStatSlots; i++)
3265 * Follow the protocol of retrying if st_changecount changes while we
3266 * copy the entry, or if it's odd. (The check for odd is needed to
3267 * cover the case where we are able to completely copy the entry while
3268 * the source backend is between increment steps.) We use a volatile
3269 * pointer here to ensure the compiler doesn't try to get cute.
3273 int before_changecount;
3274 int after_changecount;
3276 pgstat_save_changecount_before(beentry, before_changecount);
3278 localentry->backendStatus.st_procpid = beentry->st_procpid;
3279 if (localentry->backendStatus.st_procpid > 0)
3281 memcpy(&localentry->backendStatus, (char *) beentry, sizeof(PgBackendStatus));
3284 * strcpy is safe even if the string is modified concurrently,
3285 * because there's always a \0 at the end of the buffer.
3287 strcpy(localappname, (char *) beentry->st_appname);
3288 localentry->backendStatus.st_appname = localappname;
3289 strcpy(localclienthostname, (char *) beentry->st_clienthostname);
3290 localentry->backendStatus.st_clienthostname = localclienthostname;
3291 strcpy(localactivity, (char *) beentry->st_activity_raw);
3292 localentry->backendStatus.st_activity_raw = localactivity;
3293 localentry->backendStatus.st_ssl = beentry->st_ssl;
3295 if (beentry->st_ssl)
3297 memcpy(localsslstatus, beentry->st_sslstatus, sizeof(PgBackendSSLStatus));
3298 localentry->backendStatus.st_sslstatus = localsslstatus;
3303 pgstat_save_changecount_after(beentry, after_changecount);
3304 if (before_changecount == after_changecount &&
3305 (before_changecount & 1) == 0)
3308 /* Make sure we can break out of loop if stuck... */
3309 CHECK_FOR_INTERRUPTS();
3313 /* Only valid entries get included into the local array */
3314 if (localentry->backendStatus.st_procpid > 0)
3316 BackendIdGetTransactionIds(i,
3317 &localentry->backend_xid,
3318 &localentry->backend_xmin);
3321 localappname += NAMEDATALEN;
3322 localclienthostname += NAMEDATALEN;
3323 localactivity += pgstat_track_activity_query_size;
3331 /* Set the pointer only after completion of a valid table */
3332 localBackendStatusTable = localtable;
3336 * pgstat_get_wait_event_type() -
3338 * Return a string representing the current wait event type, backend is
3342 pgstat_get_wait_event_type(uint32 wait_event_info)
3345 const char *event_type;
3347 /* report process as not waiting. */
3348 if (wait_event_info == 0)
3351 classId = wait_event_info & 0xFF000000;
3355 case PG_WAIT_LWLOCK:
3356 event_type = "LWLock";
3359 event_type = "Lock";
3361 case PG_WAIT_BUFFER_PIN:
3362 event_type = "BufferPin";
3364 case PG_WAIT_ACTIVITY:
3365 event_type = "Activity";
3367 case PG_WAIT_CLIENT:
3368 event_type = "Client";
3370 case PG_WAIT_EXTENSION:
3371 event_type = "Extension";
3376 case PG_WAIT_TIMEOUT:
3377 event_type = "Timeout";
3391 * pgstat_get_wait_event() -
3393 * Return a string representing the current wait event, backend is
3397 pgstat_get_wait_event(uint32 wait_event_info)
3401 const char *event_name;
3403 /* report process as not waiting. */
3404 if (wait_event_info == 0)
3407 classId = wait_event_info & 0xFF000000;
3408 eventId = wait_event_info & 0x0000FFFF;
3412 case PG_WAIT_LWLOCK:
3413 event_name = GetLWLockIdentifier(classId, eventId);
3416 event_name = GetLockNameFromTagType(eventId);
3418 case PG_WAIT_BUFFER_PIN:
3419 event_name = "BufferPin";
3421 case PG_WAIT_ACTIVITY:
3423 WaitEventActivity w = (WaitEventActivity) wait_event_info;
3425 event_name = pgstat_get_wait_activity(w);
3428 case PG_WAIT_CLIENT:
3430 WaitEventClient w = (WaitEventClient) wait_event_info;
3432 event_name = pgstat_get_wait_client(w);
3435 case PG_WAIT_EXTENSION:
3436 event_name = "Extension";
3440 WaitEventIPC w = (WaitEventIPC) wait_event_info;
3442 event_name = pgstat_get_wait_ipc(w);
3445 case PG_WAIT_TIMEOUT:
3447 WaitEventTimeout w = (WaitEventTimeout) wait_event_info;
3449 event_name = pgstat_get_wait_timeout(w);
3454 WaitEventIO w = (WaitEventIO) wait_event_info;
3456 event_name = pgstat_get_wait_io(w);
3460 event_name = "unknown wait event";
3468 * pgstat_get_wait_activity() -
3470 * Convert WaitEventActivity to string.
3474 pgstat_get_wait_activity(WaitEventActivity w)
3476 const char *event_name = "unknown wait event";
3480 case WAIT_EVENT_ARCHIVER_MAIN:
3481 event_name = "ArchiverMain";
3483 case WAIT_EVENT_AUTOVACUUM_MAIN:
3484 event_name = "AutoVacuumMain";
3486 case WAIT_EVENT_BGWRITER_HIBERNATE:
3487 event_name = "BgWriterHibernate";
3489 case WAIT_EVENT_BGWRITER_MAIN:
3490 event_name = "BgWriterMain";
3492 case WAIT_EVENT_CHECKPOINTER_MAIN:
3493 event_name = "CheckpointerMain";
3495 case WAIT_EVENT_LOGICAL_LAUNCHER_MAIN:
3496 event_name = "LogicalLauncherMain";
3498 case WAIT_EVENT_LOGICAL_APPLY_MAIN:
3499 event_name = "LogicalApplyMain";
3501 case WAIT_EVENT_PGSTAT_MAIN:
3502 event_name = "PgStatMain";
3504 case WAIT_EVENT_RECOVERY_WAL_ALL:
3505 event_name = "RecoveryWalAll";
3507 case WAIT_EVENT_RECOVERY_WAL_STREAM:
3508 event_name = "RecoveryWalStream";
3510 case WAIT_EVENT_SYSLOGGER_MAIN:
3511 event_name = "SysLoggerMain";
3513 case WAIT_EVENT_WAL_RECEIVER_MAIN:
3514 event_name = "WalReceiverMain";
3516 case WAIT_EVENT_WAL_SENDER_MAIN:
3517 event_name = "WalSenderMain";
3519 case WAIT_EVENT_WAL_WRITER_MAIN:
3520 event_name = "WalWriterMain";
3522 /* no default case, so that compiler will warn */
3529 * pgstat_get_wait_client() -
3531 * Convert WaitEventClient to string.
3535 pgstat_get_wait_client(WaitEventClient w)
3537 const char *event_name = "unknown wait event";
3541 case WAIT_EVENT_CLIENT_READ:
3542 event_name = "ClientRead";
3544 case WAIT_EVENT_CLIENT_WRITE:
3545 event_name = "ClientWrite";
3547 case WAIT_EVENT_LIBPQWALRECEIVER_CONNECT:
3548 event_name = "LibPQWalReceiverConnect";
3550 case WAIT_EVENT_LIBPQWALRECEIVER_RECEIVE:
3551 event_name = "LibPQWalReceiverReceive";
3553 case WAIT_EVENT_SSL_OPEN_SERVER:
3554 event_name = "SSLOpenServer";
3556 case WAIT_EVENT_WAL_RECEIVER_WAIT_START:
3557 event_name = "WalReceiverWaitStart";
3559 case WAIT_EVENT_WAL_SENDER_WAIT_WAL:
3560 event_name = "WalSenderWaitForWAL";
3562 case WAIT_EVENT_WAL_SENDER_WRITE_DATA:
3563 event_name = "WalSenderWriteData";
3565 /* no default case, so that compiler will warn */
3572 * pgstat_get_wait_ipc() -
3574 * Convert WaitEventIPC to string.
3578 pgstat_get_wait_ipc(WaitEventIPC w)
3580 const char *event_name = "unknown wait event";
3584 case WAIT_EVENT_BGWORKER_SHUTDOWN:
3585 event_name = "BgWorkerShutdown";
3587 case WAIT_EVENT_BGWORKER_STARTUP:
3588 event_name = "BgWorkerStartup";
3590 case WAIT_EVENT_BTREE_PAGE:
3591 event_name = "BtreePage";
3593 case WAIT_EVENT_EXECUTE_GATHER:
3594 event_name = "ExecuteGather";
3596 case WAIT_EVENT_HASH_BATCH_ALLOCATING:
3597 event_name = "Hash/Batch/Allocating";
3599 case WAIT_EVENT_HASH_BATCH_ELECTING:
3600 event_name = "Hash/Batch/Electing";
3602 case WAIT_EVENT_HASH_BATCH_LOADING:
3603 event_name = "Hash/Batch/Loading";
3605 case WAIT_EVENT_HASH_BUILD_ALLOCATING:
3606 event_name = "Hash/Build/Allocating";
3608 case WAIT_EVENT_HASH_BUILD_ELECTING:
3609 event_name = "Hash/Build/Electing";
3611 case WAIT_EVENT_HASH_BUILD_HASHING_INNER:
3612 event_name = "Hash/Build/HashingInner";
3614 case WAIT_EVENT_HASH_BUILD_HASHING_OUTER:
3615 event_name = "Hash/Build/HashingOuter";
3617 case WAIT_EVENT_HASH_GROW_BATCHES_ALLOCATING:
3618 event_name = "Hash/GrowBatches/Allocating";
3620 case WAIT_EVENT_HASH_GROW_BATCHES_DECIDING:
3621 event_name = "Hash/GrowBatches/Deciding";
3623 case WAIT_EVENT_HASH_GROW_BATCHES_ELECTING:
3624 event_name = "Hash/GrowBatches/Electing";
3626 case WAIT_EVENT_HASH_GROW_BATCHES_FINISHING:
3627 event_name = "Hash/GrowBatches/Finishing";
3629 case WAIT_EVENT_HASH_GROW_BATCHES_REPARTITIONING:
3630 event_name = "Hash/GrowBatches/Repartitioning";
3632 case WAIT_EVENT_HASH_GROW_BUCKETS_ALLOCATING:
3633 event_name = "Hash/GrowBuckets/Allocating";
3635 case WAIT_EVENT_HASH_GROW_BUCKETS_ELECTING:
3636 event_name = "Hash/GrowBuckets/Electing";
3638 case WAIT_EVENT_HASH_GROW_BUCKETS_REINSERTING:
3639 event_name = "Hash/GrowBuckets/Reinserting";
3641 case WAIT_EVENT_LOGICAL_SYNC_DATA:
3642 event_name = "LogicalSyncData";
3644 case WAIT_EVENT_LOGICAL_SYNC_STATE_CHANGE:
3645 event_name = "LogicalSyncStateChange";
3647 case WAIT_EVENT_MQ_INTERNAL:
3648 event_name = "MessageQueueInternal";
3650 case WAIT_EVENT_MQ_PUT_MESSAGE:
3651 event_name = "MessageQueuePutMessage";
3653 case WAIT_EVENT_MQ_RECEIVE:
3654 event_name = "MessageQueueReceive";
3656 case WAIT_EVENT_MQ_SEND:
3657 event_name = "MessageQueueSend";
3659 case WAIT_EVENT_PARALLEL_FINISH:
3660 event_name = "ParallelFinish";
3662 case WAIT_EVENT_PARALLEL_BITMAP_SCAN:
3663 event_name = "ParallelBitmapScan";
3665 case WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN:
3666 event_name = "ParallelCreateIndexScan";
3668 case WAIT_EVENT_PROCARRAY_GROUP_UPDATE:
3669 event_name = "ProcArrayGroupUpdate";
3671 case WAIT_EVENT_CLOG_GROUP_UPDATE:
3672 event_name = "ClogGroupUpdate";
3674 case WAIT_EVENT_REPLICATION_ORIGIN_DROP:
3675 event_name = "ReplicationOriginDrop";
3677 case WAIT_EVENT_REPLICATION_SLOT_DROP:
3678 event_name = "ReplicationSlotDrop";
3680 case WAIT_EVENT_SAFE_SNAPSHOT:
3681 event_name = "SafeSnapshot";
3683 case WAIT_EVENT_SYNC_REP:
3684 event_name = "SyncRep";
3686 /* no default case, so that compiler will warn */
3693 * pgstat_get_wait_timeout() -
3695 * Convert WaitEventTimeout to string.
3699 pgstat_get_wait_timeout(WaitEventTimeout w)
3701 const char *event_name = "unknown wait event";
3705 case WAIT_EVENT_BASE_BACKUP_THROTTLE:
3706 event_name = "BaseBackupThrottle";
3708 case WAIT_EVENT_PG_SLEEP:
3709 event_name = "PgSleep";
3711 case WAIT_EVENT_RECOVERY_APPLY_DELAY:
3712 event_name = "RecoveryApplyDelay";
3714 /* no default case, so that compiler will warn */
3721 * pgstat_get_wait_io() -
3723 * Convert WaitEventIO to string.
3727 pgstat_get_wait_io(WaitEventIO w)
3729 const char *event_name = "unknown wait event";
3733 case WAIT_EVENT_BUFFILE_READ:
3734 event_name = "BufFileRead";
3736 case WAIT_EVENT_BUFFILE_WRITE:
3737 event_name = "BufFileWrite";
3739 case WAIT_EVENT_CONTROL_FILE_READ:
3740 event_name = "ControlFileRead";
3742 case WAIT_EVENT_CONTROL_FILE_SYNC:
3743 event_name = "ControlFileSync";
3745 case WAIT_EVENT_CONTROL_FILE_SYNC_UPDATE:
3746 event_name = "ControlFileSyncUpdate";
3748 case WAIT_EVENT_CONTROL_FILE_WRITE:
3749 event_name = "ControlFileWrite";
3751 case WAIT_EVENT_CONTROL_FILE_WRITE_UPDATE:
3752 event_name = "ControlFileWriteUpdate";
3754 case WAIT_EVENT_COPY_FILE_READ:
3755 event_name = "CopyFileRead";
3757 case WAIT_EVENT_COPY_FILE_WRITE:
3758 event_name = "CopyFileWrite";
3760 case WAIT_EVENT_DATA_FILE_EXTEND:
3761 event_name = "DataFileExtend";
3763 case WAIT_EVENT_DATA_FILE_FLUSH:
3764 event_name = "DataFileFlush";
3766 case WAIT_EVENT_DATA_FILE_IMMEDIATE_SYNC:
3767 event_name = "DataFileImmediateSync";
3769 case WAIT_EVENT_DATA_FILE_PREFETCH:
3770 event_name = "DataFilePrefetch";
3772 case WAIT_EVENT_DATA_FILE_READ:
3773 event_name = "DataFileRead";
3775 case WAIT_EVENT_DATA_FILE_SYNC:
3776 event_name = "DataFileSync";
3778 case WAIT_EVENT_DATA_FILE_TRUNCATE:
3779 event_name = "DataFileTruncate";
3781 case WAIT_EVENT_DATA_FILE_WRITE:
3782 event_name = "DataFileWrite";
3784 case WAIT_EVENT_DSM_FILL_ZERO_WRITE:
3785 event_name = "DSMFillZeroWrite";
3787 case WAIT_EVENT_LOCK_FILE_ADDTODATADIR_READ:
3788 event_name = "LockFileAddToDataDirRead";
3790 case WAIT_EVENT_LOCK_FILE_ADDTODATADIR_SYNC:
3791 event_name = "LockFileAddToDataDirSync";
3793 case WAIT_EVENT_LOCK_FILE_ADDTODATADIR_WRITE:
3794 event_name = "LockFileAddToDataDirWrite";
3796 case WAIT_EVENT_LOCK_FILE_CREATE_READ:
3797 event_name = "LockFileCreateRead";
3799 case WAIT_EVENT_LOCK_FILE_CREATE_SYNC:
3800 event_name = "LockFileCreateSync";
3802 case WAIT_EVENT_LOCK_FILE_CREATE_WRITE:
3803 event_name = "LockFileCreateWRITE";
3805 case WAIT_EVENT_LOCK_FILE_RECHECKDATADIR_READ:
3806 event_name = "LockFileReCheckDataDirRead";
3808 case WAIT_EVENT_LOGICAL_REWRITE_CHECKPOINT_SYNC:
3809 event_name = "LogicalRewriteCheckpointSync";
3811 case WAIT_EVENT_LOGICAL_REWRITE_MAPPING_SYNC:
3812 event_name = "LogicalRewriteMappingSync";
3814 case WAIT_EVENT_LOGICAL_REWRITE_MAPPING_WRITE:
3815 event_name = "LogicalRewriteMappingWrite";
3817 case WAIT_EVENT_LOGICAL_REWRITE_SYNC:
3818 event_name = "LogicalRewriteSync";
3820 case WAIT_EVENT_LOGICAL_REWRITE_TRUNCATE:
3821 event_name = "LogicalRewriteTruncate";
3823 case WAIT_EVENT_LOGICAL_REWRITE_WRITE:
3824 event_name = "LogicalRewriteWrite";
3826 case WAIT_EVENT_RELATION_MAP_READ:
3827 event_name = "RelationMapRead";
3829 case WAIT_EVENT_RELATION_MAP_SYNC:
3830 event_name = "RelationMapSync";
3832 case WAIT_EVENT_RELATION_MAP_WRITE:
3833 event_name = "RelationMapWrite";
3835 case WAIT_EVENT_REORDER_BUFFER_READ:
3836 event_name = "ReorderBufferRead";
3838 case WAIT_EVENT_REORDER_BUFFER_WRITE:
3839 event_name = "ReorderBufferWrite";
3841 case WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ:
3842 event_name = "ReorderLogicalMappingRead";
3844 case WAIT_EVENT_REPLICATION_SLOT_READ:
3845 event_name = "ReplicationSlotRead";
3847 case WAIT_EVENT_REPLICATION_SLOT_RESTORE_SYNC:
3848 event_name = "ReplicationSlotRestoreSync";
3850 case WAIT_EVENT_REPLICATION_SLOT_SYNC:
3851 event_name = "ReplicationSlotSync";
3853 case WAIT_EVENT_REPLICATION_SLOT_WRITE:
3854 event_name = "ReplicationSlotWrite";
3856 case WAIT_EVENT_SLRU_FLUSH_SYNC:
3857 event_name = "SLRUFlushSync";
3859 case WAIT_EVENT_SLRU_READ:
3860 event_name = "SLRURead";
3862 case WAIT_EVENT_SLRU_SYNC:
3863 event_name = "SLRUSync";
3865 case WAIT_EVENT_SLRU_WRITE:
3866 event_name = "SLRUWrite";
3868 case WAIT_EVENT_SNAPBUILD_READ:
3869 event_name = "SnapbuildRead";
3871 case WAIT_EVENT_SNAPBUILD_SYNC:
3872 event_name = "SnapbuildSync";
3874 case WAIT_EVENT_SNAPBUILD_WRITE:
3875 event_name = "SnapbuildWrite";
3877 case WAIT_EVENT_TIMELINE_HISTORY_FILE_SYNC:
3878 event_name = "TimelineHistoryFileSync";
3880 case WAIT_EVENT_TIMELINE_HISTORY_FILE_WRITE:
3881 event_name = "TimelineHistoryFileWrite";
3883 case WAIT_EVENT_TIMELINE_HISTORY_READ:
3884 event_name = "TimelineHistoryRead";
3886 case WAIT_EVENT_TIMELINE_HISTORY_SYNC:
3887 event_name = "TimelineHistorySync";
3889 case WAIT_EVENT_TIMELINE_HISTORY_WRITE:
3890 event_name = "TimelineHistoryWrite";
3892 case WAIT_EVENT_TWOPHASE_FILE_READ:
3893 event_name = "TwophaseFileRead";
3895 case WAIT_EVENT_TWOPHASE_FILE_SYNC:
3896 event_name = "TwophaseFileSync";
3898 case WAIT_EVENT_TWOPHASE_FILE_WRITE:
3899 event_name = "TwophaseFileWrite";
3901 case WAIT_EVENT_WALSENDER_TIMELINE_HISTORY_READ:
3902 event_name = "WALSenderTimelineHistoryRead";
3904 case WAIT_EVENT_WAL_BOOTSTRAP_SYNC:
3905 event_name = "WALBootstrapSync";
3907 case WAIT_EVENT_WAL_BOOTSTRAP_WRITE:
3908 event_name = "WALBootstrapWrite";
3910 case WAIT_EVENT_WAL_COPY_READ:
3911 event_name = "WALCopyRead";
3913 case WAIT_EVENT_WAL_COPY_SYNC:
3914 event_name = "WALCopySync";
3916 case WAIT_EVENT_WAL_COPY_WRITE:
3917 event_name = "WALCopyWrite";
3919 case WAIT_EVENT_WAL_INIT_SYNC:
3920 event_name = "WALInitSync";
3922 case WAIT_EVENT_WAL_INIT_WRITE:
3923 event_name = "WALInitWrite";
3925 case WAIT_EVENT_WAL_READ:
3926 event_name = "WALRead";
3928 case WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN:
3929 event_name = "WALSyncMethodAssign";
3931 case WAIT_EVENT_WAL_WRITE:
3932 event_name = "WALWrite";
3935 /* no default case, so that compiler will warn */
3943 * pgstat_get_backend_current_activity() -
3945 * Return a string representing the current activity of the backend with
3946 * the specified PID. This looks directly at the BackendStatusArray,
3947 * and so will provide current information regardless of the age of our
3948 * transaction's snapshot of the status array.
3950 * It is the caller's responsibility to invoke this only for backends whose
3951 * state is expected to remain stable while the result is in use. The
3952 * only current use is in deadlock reporting, where we can expect that
3953 * the target backend is blocked on a lock. (There are corner cases
3954 * where the target's wait could get aborted while we are looking at it,
3955 * but the very worst consequence is to return a pointer to a string
3956 * that's been changed, so we won't worry too much.)
3958 * Note: return strings for special cases match pg_stat_get_backend_activity.
3962 pgstat_get_backend_current_activity(int pid, bool checkUser)
3964 PgBackendStatus *beentry;
3967 beentry = BackendStatusArray;
3968 for (i = 1; i <= MaxBackends; i++)
3971 * Although we expect the target backend's entry to be stable, that
3972 * doesn't imply that anyone else's is. To avoid identifying the
3973 * wrong backend, while we check for a match to the desired PID we
3974 * must follow the protocol of retrying if st_changecount changes
3975 * while we examine the entry, or if it's odd. (This might be
3976 * unnecessary, since fetching or storing an int is almost certainly
3977 * atomic, but let's play it safe.) We use a volatile pointer here to
3978 * ensure the compiler doesn't try to get cute.
3980 volatile PgBackendStatus *vbeentry = beentry;
3985 int before_changecount;
3986 int after_changecount;
3988 pgstat_save_changecount_before(vbeentry, before_changecount);
3990 found = (vbeentry->st_procpid == pid);
3992 pgstat_save_changecount_after(vbeentry, after_changecount);
3994 if (before_changecount == after_changecount &&
3995 (before_changecount & 1) == 0)
3998 /* Make sure we can break out of loop if stuck... */
3999 CHECK_FOR_INTERRUPTS();
4004 /* Now it is safe to use the non-volatile pointer */
4005 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
4006 return "<insufficient privilege>";
4007 else if (*(beentry->st_activity_raw) == '\0')
4008 return "<command string not enabled>";
4011 /* this'll leak a bit of memory, but that seems acceptable */
4012 return pgstat_clip_activity(beentry->st_activity_raw);
4019 /* If we get here, caller is in error ... */
4020 return "<backend information not available>";
4024 * pgstat_get_crashed_backend_activity() -
4026 * Return a string representing the current activity of the backend with
4027 * the specified PID. Like the function above, but reads shared memory with
4028 * the expectation that it may be corrupt. On success, copy the string
4029 * into the "buffer" argument and return that pointer. On failure,
4032 * This function is only intended to be used by the postmaster to report the
4033 * query that crashed a backend. In particular, no attempt is made to
4034 * follow the correct concurrency protocol when accessing the
4035 * BackendStatusArray. But that's OK, in the worst case we'll return a
4036 * corrupted message. We also must take care not to trip on ereport(ERROR).
4040 pgstat_get_crashed_backend_activity(int pid, char *buffer, int buflen)
4042 volatile PgBackendStatus *beentry;
4045 beentry = BackendStatusArray;
4048 * We probably shouldn't get here before shared memory has been set up,
4051 if (beentry == NULL || BackendActivityBuffer == NULL)
4054 for (i = 1; i <= MaxBackends; i++)
4056 if (beentry->st_procpid == pid)
4058 /* Read pointer just once, so it can't change after validation */
4059 const char *activity = beentry->st_activity_raw;
4060 const char *activity_last;
4063 * We mustn't access activity string before we verify that it
4064 * falls within the BackendActivityBuffer. To make sure that the
4065 * entire string including its ending is contained within the
4066 * buffer, subtract one activity length from the buffer size.
4068 activity_last = BackendActivityBuffer + BackendActivityBufferSize
4069 - pgstat_track_activity_query_size;
4071 if (activity < BackendActivityBuffer ||
4072 activity > activity_last)
4075 /* If no string available, no point in a report */
4076 if (activity[0] == '\0')
4080 * Copy only ASCII-safe characters so we don't run into encoding
4081 * problems when reporting the message; and be sure not to run off
4082 * the end of memory. As only ASCII characters are reported, it
4083 * doesn't seem necessary to perform multibyte aware clipping.
4085 ascii_safe_strlcpy(buffer, activity,
4086 Min(buflen, pgstat_track_activity_query_size));
4099 pgstat_get_backend_desc(BackendType backendType)
4101 const char *backendDesc = "unknown process type";
4103 switch (backendType)
4105 case B_AUTOVAC_LAUNCHER:
4106 backendDesc = "autovacuum launcher";
4108 case B_AUTOVAC_WORKER:
4109 backendDesc = "autovacuum worker";
4112 backendDesc = "client backend";
4115 backendDesc = "background worker";
4118 backendDesc = "background writer";
4120 case B_CHECKPOINTER:
4121 backendDesc = "checkpointer";
4124 backendDesc = "startup";
4126 case B_WAL_RECEIVER:
4127 backendDesc = "walreceiver";
4130 backendDesc = "walsender";
4133 backendDesc = "walwriter";
4140 /* ------------------------------------------------------------
4141 * Local support functions follow
4142 * ------------------------------------------------------------
4147 * pgstat_setheader() -
4149 * Set common header fields in a statistics message
4153 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
4155 hdr->m_type = mtype;
4162 * Send out one statistics message to the collector
4166 pgstat_send(void *msg, int len)
4170 if (pgStatSock == PGINVALID_SOCKET)
4173 ((PgStat_MsgHdr *) msg)->m_size = len;
4175 /* We'll retry after EINTR, but ignore all other failures */
4178 rc = send(pgStatSock, msg, len, 0);
4179 } while (rc < 0 && errno == EINTR);
4181 #ifdef USE_ASSERT_CHECKING
4182 /* In debug builds, log send failures ... */
4184 elog(LOG, "could not send to statistics collector: %m");
4189 * pgstat_send_archiver() -
4191 * Tell the collector about the WAL file that we successfully
4192 * archived or failed to archive.
4196 pgstat_send_archiver(const char *xlog, bool failed)
4198 PgStat_MsgArchiver msg;
4201 * Prepare and send the message
4203 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ARCHIVER);
4204 msg.m_failed = failed;
4205 StrNCpy(msg.m_xlog, xlog, sizeof(msg.m_xlog));
4206 msg.m_timestamp = GetCurrentTimestamp();
4207 pgstat_send(&msg, sizeof(msg));
4211 * pgstat_send_bgwriter() -
4213 * Send bgwriter statistics to the collector
4217 pgstat_send_bgwriter(void)
4219 /* We assume this initializes to zeroes */
4220 static const PgStat_MsgBgWriter all_zeroes;
4223 * This function can be called even if nothing at all has happened. In
4224 * this case, avoid sending a completely empty message to the stats
4227 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
4231 * Prepare and send the message
4233 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
4234 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
4237 * Clear out the statistics buffer, so it can be re-used.
4239 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
4244 * PgstatCollectorMain() -
4246 * Start up the statistics collector process. This is the body of the
4247 * postmaster child process.
4249 * The argc/argv parameters are valid only in EXEC_BACKEND case.
4252 NON_EXEC_STATIC void
4253 PgstatCollectorMain(int argc, char *argv[])
4260 * Ignore all signals usually bound to some action in the postmaster,
4261 * except SIGHUP and SIGQUIT. Note we don't need a SIGUSR1 handler to
4262 * support latch operations, because we only use a local latch.
4264 pqsignal(SIGHUP, pgstat_sighup_handler);
4265 pqsignal(SIGINT, SIG_IGN);
4266 pqsignal(SIGTERM, SIG_IGN);
4267 pqsignal(SIGQUIT, pgstat_exit);
4268 pqsignal(SIGALRM, SIG_IGN);
4269 pqsignal(SIGPIPE, SIG_IGN);
4270 pqsignal(SIGUSR1, SIG_IGN);
4271 pqsignal(SIGUSR2, SIG_IGN);
4272 pqsignal(SIGCHLD, SIG_DFL);
4273 pqsignal(SIGTTIN, SIG_DFL);
4274 pqsignal(SIGTTOU, SIG_DFL);
4275 pqsignal(SIGCONT, SIG_DFL);
4276 pqsignal(SIGWINCH, SIG_DFL);
4277 PG_SETMASK(&UnBlockSig);
4280 * Identify myself via ps
4282 init_ps_display("stats collector", "", "", "");
4285 * Read in existing stats files or initialize the stats to zero.
4287 pgStatRunningInCollector = true;
4288 pgStatDBHash = pgstat_read_statsfiles(InvalidOid, true, true);
4291 * Loop to process messages until we get SIGQUIT or detect ungraceful
4292 * death of our parent postmaster.
4294 * For performance reasons, we don't want to do ResetLatch/WaitLatch after
4295 * every message; instead, do that only after a recv() fails to obtain a
4296 * message. (This effectively means that if backends are sending us stuff
4297 * like mad, we won't notice postmaster death until things slack off a
4298 * bit; which seems fine.) To do that, we have an inner loop that
4299 * iterates as long as recv() succeeds. We do recognize got_SIGHUP inside
4300 * the inner loop, which means that such interrupts will get serviced but
4301 * the latch won't get cleared until next time there is a break in the
4306 /* Clear any already-pending wakeups */
4307 ResetLatch(MyLatch);
4310 * Quit if we get SIGQUIT from the postmaster.
4316 * Inner loop iterates as long as we keep getting messages, or until
4317 * need_exit becomes set.
4322 * Reload configuration if we got SIGHUP from the postmaster.
4327 ProcessConfigFile(PGC_SIGHUP);
4331 * Write the stats file(s) if a new request has arrived that is
4332 * not satisfied by existing file(s).
4334 if (pgstat_write_statsfile_needed())
4335 pgstat_write_statsfiles(false, false);
4338 * Try to receive and process a message. This will not block,
4339 * since the socket is set to non-blocking mode.
4341 * XXX On Windows, we have to force pgwin32_recv to cooperate,
4342 * despite the previous use of pg_set_noblock() on the socket.
4343 * This is extremely broken and should be fixed someday.
4346 pgwin32_noblock = 1;
4349 len = recv(pgStatSock, (char *) &msg,
4350 sizeof(PgStat_Msg), 0);
4353 pgwin32_noblock = 0;
4358 if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
4359 break; /* out of inner loop */
4361 (errcode_for_socket_access(),
4362 errmsg("could not read statistics message: %m")));
4366 * We ignore messages that are smaller than our common header
4368 if (len < sizeof(PgStat_MsgHdr))
4372 * The received length must match the length in the header
4374 if (msg.msg_hdr.m_size != len)
4378 * O.K. - we accept this message. Process it.
4380 switch (msg.msg_hdr.m_type)
4382 case PGSTAT_MTYPE_DUMMY:
4385 case PGSTAT_MTYPE_INQUIRY:
4386 pgstat_recv_inquiry((PgStat_MsgInquiry *) &msg, len);
4389 case PGSTAT_MTYPE_TABSTAT:
4390 pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len);
4393 case PGSTAT_MTYPE_TABPURGE:
4394 pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len);
4397 case PGSTAT_MTYPE_DROPDB:
4398 pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len);
4401 case PGSTAT_MTYPE_RESETCOUNTER:
4402 pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg,
4406 case PGSTAT_MTYPE_RESETSHAREDCOUNTER:
4407 pgstat_recv_resetsharedcounter(
4408 (PgStat_MsgResetsharedcounter *) &msg,
4412 case PGSTAT_MTYPE_RESETSINGLECOUNTER:
4413 pgstat_recv_resetsinglecounter(
4414 (PgStat_MsgResetsinglecounter *) &msg,
4418 case PGSTAT_MTYPE_AUTOVAC_START:
4419 pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
4422 case PGSTAT_MTYPE_VACUUM:
4423 pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
4426 case PGSTAT_MTYPE_ANALYZE:
4427 pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len);
4430 case PGSTAT_MTYPE_ARCHIVER:
4431 pgstat_recv_archiver((PgStat_MsgArchiver *) &msg, len);
4434 case PGSTAT_MTYPE_BGWRITER:
4435 pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len);
4438 case PGSTAT_MTYPE_FUNCSTAT:
4439 pgstat_recv_funcstat((PgStat_MsgFuncstat *) &msg, len);
4442 case PGSTAT_MTYPE_FUNCPURGE:
4443 pgstat_recv_funcpurge((PgStat_MsgFuncpurge *) &msg, len);
4446 case PGSTAT_MTYPE_RECOVERYCONFLICT:
4447 pgstat_recv_recoveryconflict((PgStat_MsgRecoveryConflict *) &msg, len);
4450 case PGSTAT_MTYPE_DEADLOCK:
4451 pgstat_recv_deadlock((PgStat_MsgDeadlock *) &msg, len);
4454 case PGSTAT_MTYPE_TEMPFILE:
4455 pgstat_recv_tempfile((PgStat_MsgTempFile *) &msg, len);
4461 } /* end of inner message-processing loop */
4463 /* Sleep until there's something to do */
4465 wr = WaitLatchOrSocket(MyLatch,
4466 WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE,
4468 WAIT_EVENT_PGSTAT_MAIN);
4472 * Windows, at least in its Windows Server 2003 R2 incarnation,
4473 * sometimes loses FD_READ events. Waking up and retrying the recv()
4474 * fixes that, so don't sleep indefinitely. This is a crock of the
4475 * first water, but until somebody wants to debug exactly what's
4476 * happening there, this is the best we can do. The two-second
4477 * timeout matches our pre-9.2 behavior, and needs to be short enough
4478 * to not provoke "using stale statistics" complaints from
4479 * backend_read_statsfile.
4481 wr = WaitLatchOrSocket(MyLatch,
4482 WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE | WL_TIMEOUT,
4484 2 * 1000L /* msec */ ,
4485 WAIT_EVENT_PGSTAT_MAIN);
4489 * Emergency bailout if postmaster has died. This is to avoid the
4490 * necessity for manual cleanup of all postmaster children.
4492 if (wr & WL_POSTMASTER_DEATH)
4494 } /* end of outer loop */
4497 * Save the final stats to reuse at next startup.
4499 pgstat_write_statsfiles(true, true);
4505 /* SIGQUIT signal handler for collector process */
4507 pgstat_exit(SIGNAL_ARGS)
4509 int save_errno = errno;
4517 /* SIGHUP handler for collector process */
4519 pgstat_sighup_handler(SIGNAL_ARGS)
4521 int save_errno = errno;
4530 * Subroutine to clear stats in a database entry
4532 * Tables and functions hashes are initialized to empty.
4535 reset_dbentry_counters(PgStat_StatDBEntry *dbentry)
4539 dbentry->n_xact_commit = 0;
4540 dbentry->n_xact_rollback = 0;
4541 dbentry->n_blocks_fetched = 0;
4542 dbentry->n_blocks_hit = 0;
4543 dbentry->n_tuples_returned = 0;
4544 dbentry->n_tuples_fetched = 0;
4545 dbentry->n_tuples_inserted = 0;
4546 dbentry->n_tuples_updated = 0;
4547 dbentry->n_tuples_deleted = 0;
4548 dbentry->last_autovac_time = 0;
4549 dbentry->n_conflict_tablespace = 0;
4550 dbentry->n_conflict_lock = 0;
4551 dbentry->n_conflict_snapshot = 0;
4552 dbentry->n_conflict_bufferpin = 0;
4553 dbentry->n_conflict_startup_deadlock = 0;
4554 dbentry->n_temp_files = 0;
4555 dbentry->n_temp_bytes = 0;
4556 dbentry->n_deadlocks = 0;
4557 dbentry->n_block_read_time = 0;
4558 dbentry->n_block_write_time = 0;
4560 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
4561 dbentry->stats_timestamp = 0;
4563 memset(&hash_ctl, 0, sizeof(hash_ctl));
4564 hash_ctl.keysize = sizeof(Oid);
4565 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
4566 dbentry->tables = hash_create("Per-database table",
4567 PGSTAT_TAB_HASH_SIZE,
4569 HASH_ELEM | HASH_BLOBS);
4571 hash_ctl.keysize = sizeof(Oid);
4572 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
4573 dbentry->functions = hash_create("Per-database function",
4574 PGSTAT_FUNCTION_HASH_SIZE,
4576 HASH_ELEM | HASH_BLOBS);
4580 * Lookup the hash table entry for the specified database. If no hash
4581 * table entry exists, initialize it, if the create parameter is true.
4582 * Else, return NULL.
4584 static PgStat_StatDBEntry *
4585 pgstat_get_db_entry(Oid databaseid, bool create)
4587 PgStat_StatDBEntry *result;
4589 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
4591 /* Lookup or create the hash table entry for this database */
4592 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
4596 if (!create && !found)
4600 * If not found, initialize the new one. This creates empty hash tables
4601 * for tables and functions, too.
4604 reset_dbentry_counters(result);
4611 * Lookup the hash table entry for the specified table. If no hash
4612 * table entry exists, initialize it, if the create parameter is true.
4613 * Else, return NULL.
4615 static PgStat_StatTabEntry *
4616 pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
4618 PgStat_StatTabEntry *result;
4620 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
4622 /* Lookup or create the hash table entry for this table */
4623 result = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
4627 if (!create && !found)
4630 /* If not found, initialize the new one. */
4633 result->numscans = 0;
4634 result->tuples_returned = 0;
4635 result->tuples_fetched = 0;
4636 result->tuples_inserted = 0;
4637 result->tuples_updated = 0;
4638 result->tuples_deleted = 0;
4639 result->tuples_hot_updated = 0;
4640 result->n_live_tuples = 0;
4641 result->n_dead_tuples = 0;
4642 result->changes_since_analyze = 0;
4643 result->blocks_fetched = 0;
4644 result->blocks_hit = 0;
4645 result->vacuum_timestamp = 0;
4646 result->vacuum_count = 0;
4647 result->autovac_vacuum_timestamp = 0;
4648 result->autovac_vacuum_count = 0;
4649 result->analyze_timestamp = 0;
4650 result->analyze_count = 0;
4651 result->autovac_analyze_timestamp = 0;
4652 result->autovac_analyze_count = 0;
4660 * pgstat_write_statsfiles() -
4661 * Write the global statistics file, as well as requested DB files.
4663 * 'permanent' specifies writing to the permanent files not temporary ones.
4664 * When true (happens only when the collector is shutting down), also remove
4665 * the temporary files so that backends starting up under a new postmaster
4666 * can't read old data before the new collector is ready.
4668 * When 'allDbs' is false, only the requested databases (listed in
4669 * pending_write_requests) will be written; otherwise, all databases
4674 pgstat_write_statsfiles(bool permanent, bool allDbs)
4676 HASH_SEQ_STATUS hstat;
4677 PgStat_StatDBEntry *dbentry;
4680 const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
4681 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
4684 elog(DEBUG2, "writing stats file \"%s\"", statfile);
4687 * Open the statistics temp file to write out the current values.
4689 fpout = AllocateFile(tmpfile, PG_BINARY_W);
4693 (errcode_for_file_access(),
4694 errmsg("could not open temporary statistics file \"%s\": %m",
4700 * Set the timestamp of the stats file.
4702 globalStats.stats_timestamp = GetCurrentTimestamp();
4705 * Write the file header --- currently just a format ID.
4707 format_id = PGSTAT_FILE_FORMAT_ID;
4708 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
4709 (void) rc; /* we'll check for error with ferror */
4712 * Write global stats struct
4714 rc = fwrite(&globalStats, sizeof(globalStats), 1, fpout);
4715 (void) rc; /* we'll check for error with ferror */
4718 * Write archiver stats struct
4720 rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
4721 (void) rc; /* we'll check for error with ferror */
4724 * Walk through the database table.
4726 hash_seq_init(&hstat, pgStatDBHash);
4727 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
4730 * Write out the table and function stats for this DB into the
4731 * appropriate per-DB stat file, if required.
4733 if (allDbs || pgstat_db_requested(dbentry->databaseid))
4735 /* Make DB's timestamp consistent with the global stats */
4736 dbentry->stats_timestamp = globalStats.stats_timestamp;
4738 pgstat_write_db_statsfile(dbentry, permanent);
4742 * Write out the DB entry. We don't write the tables or functions
4743 * pointers, since they're of no use to any other process.
4746 rc = fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
4747 (void) rc; /* we'll check for error with ferror */
4751 * No more output to be done. Close the temp file and replace the old
4752 * pgstat.stat with it. The ferror() check replaces testing for error
4753 * after each individual fputc or fwrite above.
4760 (errcode_for_file_access(),
4761 errmsg("could not write temporary statistics file \"%s\": %m",
4766 else if (FreeFile(fpout) < 0)
4769 (errcode_for_file_access(),
4770 errmsg("could not close temporary statistics file \"%s\": %m",
4774 else if (rename(tmpfile, statfile) < 0)
4777 (errcode_for_file_access(),
4778 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
4779 tmpfile, statfile)));
4784 unlink(pgstat_stat_filename);
4787 * Now throw away the list of requests. Note that requests sent after we
4788 * started the write are still waiting on the network socket.
4790 list_free(pending_write_requests);
4791 pending_write_requests = NIL;
4795 * return the filename for a DB stat file; filename is the output buffer,
4799 get_dbstat_filename(bool permanent, bool tempname, Oid databaseid,
4800 char *filename, int len)
4804 /* NB -- pgstat_reset_remove_files knows about the pattern this uses */
4805 printed = snprintf(filename, len, "%s/db_%u.%s",
4806 permanent ? PGSTAT_STAT_PERMANENT_DIRECTORY :
4807 pgstat_stat_directory,
4809 tempname ? "tmp" : "stat");
4811 elog(ERROR, "overlength pgstat path");
4815 * pgstat_write_db_statsfile() -
4816 * Write the stat file for a single database.
4818 * If writing to the permanent file (happens when the collector is
4819 * shutting down only), remove the temporary file so that backends
4820 * starting up under a new postmaster can't read the old data before
4821 * the new collector is ready.
4825 pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
4827 HASH_SEQ_STATUS tstat;
4828 HASH_SEQ_STATUS fstat;
4829 PgStat_StatTabEntry *tabentry;
4830 PgStat_StatFuncEntry *funcentry;
4833 Oid dbid = dbentry->databaseid;
4835 char tmpfile[MAXPGPATH];
4836 char statfile[MAXPGPATH];
4838 get_dbstat_filename(permanent, true, dbid, tmpfile, MAXPGPATH);
4839 get_dbstat_filename(permanent, false, dbid, statfile, MAXPGPATH);
4841 elog(DEBUG2, "writing stats file \"%s\"", statfile);
4844 * Open the statistics temp file to write out the current values.
4846 fpout = AllocateFile(tmpfile, PG_BINARY_W);
4850 (errcode_for_file_access(),
4851 errmsg("could not open temporary statistics file \"%s\": %m",
4857 * Write the file header --- currently just a format ID.
4859 format_id = PGSTAT_FILE_FORMAT_ID;
4860 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
4861 (void) rc; /* we'll check for error with ferror */
4864 * Walk through the database's access stats per table.
4866 hash_seq_init(&tstat, dbentry->tables);
4867 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
4870 rc = fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
4871 (void) rc; /* we'll check for error with ferror */
4875 * Walk through the database's function stats table.
4877 hash_seq_init(&fstat, dbentry->functions);
4878 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
4881 rc = fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
4882 (void) rc; /* we'll check for error with ferror */
4886 * No more output to be done. Close the temp file and replace the old
4887 * pgstat.stat with it. The ferror() check replaces testing for error
4888 * after each individual fputc or fwrite above.
4895 (errcode_for_file_access(),
4896 errmsg("could not write temporary statistics file \"%s\": %m",
4901 else if (FreeFile(fpout) < 0)
4904 (errcode_for_file_access(),
4905 errmsg("could not close temporary statistics file \"%s\": %m",
4909 else if (rename(tmpfile, statfile) < 0)
4912 (errcode_for_file_access(),
4913 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
4914 tmpfile, statfile)));
4920 get_dbstat_filename(false, false, dbid, statfile, MAXPGPATH);
4922 elog(DEBUG2, "removing temporary stats file \"%s\"", statfile);
4928 * pgstat_read_statsfiles() -
4930 * Reads in some existing statistics collector files and returns the
4931 * databases hash table that is the top level of the data.
4933 * If 'onlydb' is not InvalidOid, it means we only want data for that DB
4934 * plus the shared catalogs ("DB 0"). We'll still populate the DB hash
4935 * table for all databases, but we don't bother even creating table/function
4936 * hash tables for other databases.
4938 * 'permanent' specifies reading from the permanent files not temporary ones.
4939 * When true (happens only when the collector is starting up), remove the
4940 * files after reading; the in-memory status is now authoritative, and the
4941 * files would be out of date in case somebody else reads them.
4943 * If a 'deep' read is requested, table/function stats are read, otherwise
4944 * the table/function hash tables remain empty.
4948 pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
4950 PgStat_StatDBEntry *dbentry;
4951 PgStat_StatDBEntry dbbuf;
4957 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
4960 * The tables will live in pgStatLocalContext.
4962 pgstat_setup_memcxt();
4965 * Create the DB hashtable
4967 memset(&hash_ctl, 0, sizeof(hash_ctl));
4968 hash_ctl.keysize = sizeof(Oid);
4969 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
4970 hash_ctl.hcxt = pgStatLocalContext;
4971 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
4972 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4975 * Clear out global and archiver statistics so they start from zero in
4976 * case we can't load an existing statsfile.
4978 memset(&globalStats, 0, sizeof(globalStats));
4979 memset(&archiverStats, 0, sizeof(archiverStats));
4982 * Set the current timestamp (will be kept only in case we can't load an
4983 * existing statsfile).
4985 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
4986 archiverStats.stat_reset_timestamp = globalStats.stat_reset_timestamp;
4989 * Try to open the stats file. If it doesn't exist, the backends simply
4990 * return zero for anything and the collector simply starts from scratch
4991 * with empty counters.
4993 * ENOENT is a possibility if the stats collector is not running or has
4994 * not yet written the stats file the first time. Any other failure
4995 * condition is suspicious.
4997 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
4999 if (errno != ENOENT)
5000 ereport(pgStatRunningInCollector ? LOG : WARNING,
5001 (errcode_for_file_access(),
5002 errmsg("could not open statistics file \"%s\": %m",
5008 * Verify it's of the expected format.
5010 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
5011 format_id != PGSTAT_FILE_FORMAT_ID)
5013 ereport(pgStatRunningInCollector ? LOG : WARNING,
5014 (errmsg("corrupted statistics file \"%s\"", statfile)));
5019 * Read global stats struct
5021 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
5023 ereport(pgStatRunningInCollector ? LOG : WARNING,
5024 (errmsg("corrupted statistics file \"%s\"", statfile)));
5025 memset(&globalStats, 0, sizeof(globalStats));
5030 * In the collector, disregard the timestamp we read from the permanent
5031 * stats file; we should be willing to write a temp stats file immediately
5032 * upon the first request from any backend. This only matters if the old
5033 * file's timestamp is less than PGSTAT_STAT_INTERVAL ago, but that's not
5034 * an unusual scenario.
5036 if (pgStatRunningInCollector)
5037 globalStats.stats_timestamp = 0;
5040 * Read archiver stats struct
5042 if (fread(&archiverStats, 1, sizeof(archiverStats), fpin) != sizeof(archiverStats))
5044 ereport(pgStatRunningInCollector ? LOG : WARNING,
5045 (errmsg("corrupted statistics file \"%s\"", statfile)));
5046 memset(&archiverStats, 0, sizeof(archiverStats));
5051 * We found an existing collector stats file. Read it and put all the
5052 * hashtable entries into place.
5056 switch (fgetc(fpin))
5059 * 'D' A PgStat_StatDBEntry struct describing a database
5063 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
5064 fpin) != offsetof(PgStat_StatDBEntry, tables))
5066 ereport(pgStatRunningInCollector ? LOG : WARNING,
5067 (errmsg("corrupted statistics file \"%s\"",
5073 * Add to the DB hash
5075 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
5076 (void *) &dbbuf.databaseid,
5081 ereport(pgStatRunningInCollector ? LOG : WARNING,
5082 (errmsg("corrupted statistics file \"%s\"",
5087 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
5088 dbentry->tables = NULL;
5089 dbentry->functions = NULL;
5092 * In the collector, disregard the timestamp we read from the
5093 * permanent stats file; we should be willing to write a temp
5094 * stats file immediately upon the first request from any
5097 if (pgStatRunningInCollector)
5098 dbentry->stats_timestamp = 0;
5101 * Don't create tables/functions hashtables for uninteresting
5104 if (onlydb != InvalidOid)
5106 if (dbbuf.databaseid != onlydb &&
5107 dbbuf.databaseid != InvalidOid)
5111 memset(&hash_ctl, 0, sizeof(hash_ctl));
5112 hash_ctl.keysize = sizeof(Oid);
5113 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
5114 hash_ctl.hcxt = pgStatLocalContext;
5115 dbentry->tables = hash_create("Per-database table",
5116 PGSTAT_TAB_HASH_SIZE,
5118 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
5120 hash_ctl.keysize = sizeof(Oid);
5121 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
5122 hash_ctl.hcxt = pgStatLocalContext;
5123 dbentry->functions = hash_create("Per-database function",
5124 PGSTAT_FUNCTION_HASH_SIZE,
5126 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
5129 * If requested, read the data from the database-specific
5130 * file. Otherwise we just leave the hashtables empty.
5133 pgstat_read_db_statsfile(dbentry->databaseid,
5144 ereport(pgStatRunningInCollector ? LOG : WARNING,
5145 (errmsg("corrupted statistics file \"%s\"",
5154 /* If requested to read the permanent file, also get rid of it. */
5157 elog(DEBUG2, "removing permanent stats file \"%s\"", statfile);
5166 * pgstat_read_db_statsfile() -
5168 * Reads in the existing statistics collector file for the given database,
5169 * filling the passed-in tables and functions hash tables.
5171 * As in pgstat_read_statsfiles, if the permanent file is requested, it is
5172 * removed after reading.
5174 * Note: this code has the ability to skip storing per-table or per-function
5175 * data, if NULL is passed for the corresponding hashtable. That's not used
5176 * at the moment though.
5180 pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash,
5183 PgStat_StatTabEntry *tabentry;
5184 PgStat_StatTabEntry tabbuf;
5185 PgStat_StatFuncEntry funcbuf;
5186 PgStat_StatFuncEntry *funcentry;
5190 char statfile[MAXPGPATH];
5192 get_dbstat_filename(permanent, false, databaseid, statfile, MAXPGPATH);
5195 * Try to open the stats file. If it doesn't exist, the backends simply
5196 * return zero for anything and the collector simply starts from scratch
5197 * with empty counters.
5199 * ENOENT is a possibility if the stats collector is not running or has
5200 * not yet written the stats file the first time. Any other failure
5201 * condition is suspicious.
5203 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
5205 if (errno != ENOENT)
5206 ereport(pgStatRunningInCollector ? LOG : WARNING,
5207 (errcode_for_file_access(),
5208 errmsg("could not open statistics file \"%s\": %m",
5214 * Verify it's of the expected format.
5216 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
5217 format_id != PGSTAT_FILE_FORMAT_ID)
5219 ereport(pgStatRunningInCollector ? LOG : WARNING,
5220 (errmsg("corrupted statistics file \"%s\"", statfile)));
5225 * We found an existing collector stats file. Read it and put all the
5226 * hashtable entries into place.
5230 switch (fgetc(fpin))
5233 * 'T' A PgStat_StatTabEntry follows.
5236 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
5237 fpin) != sizeof(PgStat_StatTabEntry))
5239 ereport(pgStatRunningInCollector ? LOG : WARNING,
5240 (errmsg("corrupted statistics file \"%s\"",
5246 * Skip if table data not wanted.
5248 if (tabhash == NULL)
5251 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
5252 (void *) &tabbuf.tableid,
5253 HASH_ENTER, &found);
5257 ereport(pgStatRunningInCollector ? LOG : WARNING,
5258 (errmsg("corrupted statistics file \"%s\"",
5263 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
5267 * 'F' A PgStat_StatFuncEntry follows.
5270 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
5271 fpin) != sizeof(PgStat_StatFuncEntry))
5273 ereport(pgStatRunningInCollector ? LOG : WARNING,
5274 (errmsg("corrupted statistics file \"%s\"",
5280 * Skip if function data not wanted.
5282 if (funchash == NULL)
5285 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
5286 (void *) &funcbuf.functionid,
5287 HASH_ENTER, &found);
5291 ereport(pgStatRunningInCollector ? LOG : WARNING,
5292 (errmsg("corrupted statistics file \"%s\"",
5297 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
5301 * 'E' The EOF marker of a complete stats file.
5307 ereport(pgStatRunningInCollector ? LOG : WARNING,
5308 (errmsg("corrupted statistics file \"%s\"",
5319 elog(DEBUG2, "removing permanent stats file \"%s\"", statfile);
5325 * pgstat_read_db_statsfile_timestamp() -
5327 * Attempt to determine the timestamp of the last db statfile write.
5328 * Returns true if successful; the timestamp is stored in *ts.
5330 * This needs to be careful about handling databases for which no stats file
5331 * exists, such as databases without a stat entry or those not yet written:
5333 * - if there's a database entry in the global file, return the corresponding
5334 * stats_timestamp value.
5336 * - if there's no db stat entry (e.g. for a new or inactive database),
5337 * there's no stats_timestamp value, but also nothing to write so we return
5338 * the timestamp of the global statfile.
5342 pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
5345 PgStat_StatDBEntry dbentry;
5346 PgStat_GlobalStats myGlobalStats;
5347 PgStat_ArchiverStats myArchiverStats;
5350 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
5353 * Try to open the stats file. As above, anything but ENOENT is worthy of
5354 * complaining about.
5356 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
5358 if (errno != ENOENT)
5359 ereport(pgStatRunningInCollector ? LOG : WARNING,
5360 (errcode_for_file_access(),
5361 errmsg("could not open statistics file \"%s\": %m",
5367 * Verify it's of the expected format.
5369 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
5370 format_id != PGSTAT_FILE_FORMAT_ID)
5372 ereport(pgStatRunningInCollector ? LOG : WARNING,
5373 (errmsg("corrupted statistics file \"%s\"", statfile)));
5379 * Read global stats struct
5381 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats),
5382 fpin) != sizeof(myGlobalStats))
5384 ereport(pgStatRunningInCollector ? LOG : WARNING,
5385 (errmsg("corrupted statistics file \"%s\"", statfile)));
5391 * Read archiver stats struct
5393 if (fread(&myArchiverStats, 1, sizeof(myArchiverStats),
5394 fpin) != sizeof(myArchiverStats))
5396 ereport(pgStatRunningInCollector ? LOG : WARNING,
5397 (errmsg("corrupted statistics file \"%s\"", statfile)));
5402 /* By default, we're going to return the timestamp of the global file. */
5403 *ts = myGlobalStats.stats_timestamp;
5406 * We found an existing collector stats file. Read it and look for a
5407 * record for the requested database. If found, use its timestamp.
5411 switch (fgetc(fpin))
5414 * 'D' A PgStat_StatDBEntry struct describing a database
5418 if (fread(&dbentry, 1, offsetof(PgStat_StatDBEntry, tables),
5419 fpin) != offsetof(PgStat_StatDBEntry, tables))
5421 ereport(pgStatRunningInCollector ? LOG : WARNING,
5422 (errmsg("corrupted statistics file \"%s\"",
5428 * If this is the DB we're looking for, save its timestamp and
5431 if (dbentry.databaseid == databaseid)
5433 *ts = dbentry.stats_timestamp;
5443 ereport(pgStatRunningInCollector ? LOG : WARNING,
5444 (errmsg("corrupted statistics file \"%s\"",
5456 * If not already done, read the statistics collector stats file into
5457 * some hash tables. The results will be kept until pgstat_clear_snapshot()
5458 * is called (typically, at end of transaction).
5461 backend_read_statsfile(void)
5463 TimestampTz min_ts = 0;
5464 TimestampTz ref_ts = 0;
5468 /* already read it? */
5471 Assert(!pgStatRunningInCollector);
5474 * In a normal backend, we check staleness of the data for our own DB, and
5475 * so we send MyDatabaseId in inquiry messages. In the autovac launcher,
5476 * check staleness of the shared-catalog data, and send InvalidOid in
5477 * inquiry messages so as not to force writing unnecessary data.
5479 if (IsAutoVacuumLauncherProcess())
5480 inquiry_db = InvalidOid;
5482 inquiry_db = MyDatabaseId;
5485 * Loop until fresh enough stats file is available or we ran out of time.
5486 * The stats inquiry message is sent repeatedly in case collector drops
5487 * it; but not every single time, as that just swamps the collector.
5489 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
5492 TimestampTz file_ts = 0;
5495 CHECK_FOR_INTERRUPTS();
5497 ok = pgstat_read_db_statsfile_timestamp(inquiry_db, false, &file_ts);
5499 cur_ts = GetCurrentTimestamp();
5500 /* Calculate min acceptable timestamp, if we didn't already */
5501 if (count == 0 || cur_ts < ref_ts)
5504 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL
5505 * msec before now. This indirectly ensures that the collector
5506 * needn't write the file more often than PGSTAT_STAT_INTERVAL. In
5507 * an autovacuum worker, however, we want a lower delay to avoid
5508 * using stale data, so we use PGSTAT_RETRY_DELAY (since the
5509 * number of workers is low, this shouldn't be a problem).
5511 * We don't recompute min_ts after sleeping, except in the
5512 * unlikely case that cur_ts went backwards. So we might end up
5513 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In
5514 * practice that shouldn't happen, though, as long as the sleep
5515 * time is less than PGSTAT_STAT_INTERVAL; and we don't want to
5516 * tell the collector that our cutoff time is less than what we'd
5520 if (IsAutoVacuumWorkerProcess())
5521 min_ts = TimestampTzPlusMilliseconds(ref_ts,
5522 -PGSTAT_RETRY_DELAY);
5524 min_ts = TimestampTzPlusMilliseconds(ref_ts,
5525 -PGSTAT_STAT_INTERVAL);
5529 * If the file timestamp is actually newer than cur_ts, we must have
5530 * had a clock glitch (system time went backwards) or there is clock
5531 * skew between our processor and the stats collector's processor.
5532 * Accept the file, but send an inquiry message anyway to make
5533 * pgstat_recv_inquiry do a sanity check on the collector's time.
5535 if (ok && file_ts > cur_ts)
5538 * A small amount of clock skew between processors isn't terribly
5539 * surprising, but a large difference is worth logging. We
5540 * arbitrarily define "large" as 1000 msec.
5542 if (file_ts >= TimestampTzPlusMilliseconds(cur_ts, 1000))
5547 /* Copy because timestamptz_to_str returns a static buffer */
5548 filetime = pstrdup(timestamptz_to_str(file_ts));
5549 mytime = pstrdup(timestamptz_to_str(cur_ts));
5550 elog(LOG, "stats collector's time %s is later than backend local time %s",
5556 pgstat_send_inquiry(cur_ts, min_ts, inquiry_db);
5560 /* Normal acceptance case: file is not older than cutoff time */
5561 if (ok && file_ts >= min_ts)
5564 /* Not there or too old, so kick the collector and wait a bit */
5565 if ((count % PGSTAT_INQ_LOOP_COUNT) == 0)
5566 pgstat_send_inquiry(cur_ts, min_ts, inquiry_db);
5568 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
5571 if (count >= PGSTAT_POLL_LOOP_COUNT)
5573 (errmsg("using stale statistics instead of current ones "
5574 "because stats collector is not responding")));
5577 * Autovacuum launcher wants stats about all databases, but a shallow read
5578 * is sufficient. Regular backends want a deep read for just the tables
5579 * they can see (MyDatabaseId + shared catalogs).
5581 if (IsAutoVacuumLauncherProcess())
5582 pgStatDBHash = pgstat_read_statsfiles(InvalidOid, false, false);
5584 pgStatDBHash = pgstat_read_statsfiles(MyDatabaseId, false, true);
5589 * pgstat_setup_memcxt() -
5591 * Create pgStatLocalContext, if not already done.
5595 pgstat_setup_memcxt(void)
5597 if (!pgStatLocalContext)
5598 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
5599 "Statistics snapshot",
5600 ALLOCSET_SMALL_SIZES);
5605 * pgstat_clear_snapshot() -
5607 * Discard any data collected in the current transaction. Any subsequent
5608 * request will cause new snapshots to be read.
5610 * This is also invoked during transaction commit or abort to discard
5611 * the no-longer-wanted snapshot.
5615 pgstat_clear_snapshot(void)
5617 /* Release memory, if any was allocated */
5618 if (pgStatLocalContext)
5619 MemoryContextDelete(pgStatLocalContext);
5621 /* Reset variables */
5622 pgStatLocalContext = NULL;
5623 pgStatDBHash = NULL;
5624 localBackendStatusTable = NULL;
5625 localNumBackends = 0;
5630 * pgstat_recv_inquiry() -
5632 * Process stat inquiry requests.
5636 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
5638 PgStat_StatDBEntry *dbentry;
5640 elog(DEBUG2, "received inquiry for database %u", msg->databaseid);
5643 * If there's already a write request for this DB, there's nothing to do.
5645 * Note that if a request is found, we return early and skip the below
5646 * check for clock skew. This is okay, since the only way for a DB
5647 * request to be present in the list is that we have been here since the
5648 * last write round. It seems sufficient to check for clock skew once per
5651 if (list_member_oid(pending_write_requests, msg->databaseid))
5655 * Check to see if we last wrote this database at a time >= the requested
5656 * cutoff time. If so, this is a stale request that was generated before
5657 * we updated the DB file, and we don't need to do so again.
5659 * If the requestor's local clock time is older than stats_timestamp, we
5660 * should suspect a clock glitch, ie system time going backwards; though
5661 * the more likely explanation is just delayed message receipt. It is
5662 * worth expending a GetCurrentTimestamp call to be sure, since a large
5663 * retreat in the system clock reading could otherwise cause us to neglect
5664 * to update the stats file for a long time.
5666 dbentry = pgstat_get_db_entry(msg->databaseid, false);
5667 if (dbentry == NULL)
5670 * We have no data for this DB. Enter a write request anyway so that
5671 * the global stats will get updated. This is needed to prevent
5672 * backend_read_statsfile from waiting for data that we cannot supply,
5673 * in the case of a new DB that nobody has yet reported any stats for.
5674 * See the behavior of pgstat_read_db_statsfile_timestamp.
5677 else if (msg->clock_time < dbentry->stats_timestamp)
5679 TimestampTz cur_ts = GetCurrentTimestamp();
5681 if (cur_ts < dbentry->stats_timestamp)
5684 * Sure enough, time went backwards. Force a new stats file write
5685 * to get back in sync; but first, log a complaint.
5690 /* Copy because timestamptz_to_str returns a static buffer */
5691 writetime = pstrdup(timestamptz_to_str(dbentry->stats_timestamp));
5692 mytime = pstrdup(timestamptz_to_str(cur_ts));
5694 "stats_timestamp %s is later than collector's time %s for database %u",
5695 writetime, mytime, dbentry->databaseid);
5702 * Nope, it's just an old request. Assuming msg's clock_time is
5703 * >= its cutoff_time, it must be stale, so we can ignore it.
5708 else if (msg->cutoff_time <= dbentry->stats_timestamp)
5710 /* Stale request, ignore it */
5715 * We need to write this DB, so create a request.
5717 pending_write_requests = lappend_oid(pending_write_requests,
5723 * pgstat_recv_tabstat() -
5725 * Count what the backend has done.
5729 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
5731 PgStat_StatDBEntry *dbentry;
5732 PgStat_StatTabEntry *tabentry;
5736 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5739 * Update database-wide stats.
5741 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
5742 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
5743 dbentry->n_block_read_time += msg->m_block_read_time;
5744 dbentry->n_block_write_time += msg->m_block_write_time;
5747 * Process all table entries in the message.
5749 for (i = 0; i < msg->m_nentries; i++)
5751 PgStat_TableEntry *tabmsg = &(msg->m_entry[i]);
5753 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
5754 (void *) &(tabmsg->t_id),
5755 HASH_ENTER, &found);
5760 * If it's a new table entry, initialize counters to the values we
5763 tabentry->numscans = tabmsg->t_counts.t_numscans;
5764 tabentry->tuples_returned = tabmsg->t_counts.t_tuples_returned;
5765 tabentry->tuples_fetched = tabmsg->t_counts.t_tuples_fetched;
5766 tabentry->tuples_inserted = tabmsg->t_counts.t_tuples_inserted;
5767 tabentry->tuples_updated = tabmsg->t_counts.t_tuples_updated;
5768 tabentry->tuples_deleted = tabmsg->t_counts.t_tuples_deleted;
5769 tabentry->tuples_hot_updated = tabmsg->t_counts.t_tuples_hot_updated;
5770 tabentry->n_live_tuples = tabmsg->t_counts.t_delta_live_tuples;
5771 tabentry->n_dead_tuples = tabmsg->t_counts.t_delta_dead_tuples;
5772 tabentry->changes_since_analyze = tabmsg->t_counts.t_changed_tuples;
5773 tabentry->blocks_fetched = tabmsg->t_counts.t_blocks_fetched;
5774 tabentry->blocks_hit = tabmsg->t_counts.t_blocks_hit;
5776 tabentry->vacuum_timestamp = 0;
5777 tabentry->vacuum_count = 0;
5778 tabentry->autovac_vacuum_timestamp = 0;
5779 tabentry->autovac_vacuum_count = 0;
5780 tabentry->analyze_timestamp = 0;
5781 tabentry->analyze_count = 0;
5782 tabentry->autovac_analyze_timestamp = 0;
5783 tabentry->autovac_analyze_count = 0;
5788 * Otherwise add the values to the existing entry.
5790 tabentry->numscans += tabmsg->t_counts.t_numscans;
5791 tabentry->tuples_returned += tabmsg->t_counts.t_tuples_returned;
5792 tabentry->tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
5793 tabentry->tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
5794 tabentry->tuples_updated += tabmsg->t_counts.t_tuples_updated;
5795 tabentry->tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
5796 tabentry->tuples_hot_updated += tabmsg->t_counts.t_tuples_hot_updated;
5797 /* If table was truncated, first reset the live/dead counters */
5798 if (tabmsg->t_counts.t_truncated)
5800 tabentry->n_live_tuples = 0;
5801 tabentry->n_dead_tuples = 0;
5803 tabentry->n_live_tuples += tabmsg->t_counts.t_delta_live_tuples;
5804 tabentry->n_dead_tuples += tabmsg->t_counts.t_delta_dead_tuples;
5805 tabentry->changes_since_analyze += tabmsg->t_counts.t_changed_tuples;
5806 tabentry->blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
5807 tabentry->blocks_hit += tabmsg->t_counts.t_blocks_hit;
5810 /* Clamp n_live_tuples in case of negative delta_live_tuples */
5811 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
5812 /* Likewise for n_dead_tuples */
5813 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
5816 * Add per-table stats to the per-database entry, too.
5818 dbentry->n_tuples_returned += tabmsg->t_counts.t_tuples_returned;
5819 dbentry->n_tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
5820 dbentry->n_tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
5821 dbentry->n_tuples_updated += tabmsg->t_counts.t_tuples_updated;
5822 dbentry->n_tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
5823 dbentry->n_blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
5824 dbentry->n_blocks_hit += tabmsg->t_counts.t_blocks_hit;
5830 * pgstat_recv_tabpurge() -
5832 * Arrange for dead table removal.
5836 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
5838 PgStat_StatDBEntry *dbentry;
5841 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5844 * No need to purge if we don't even know the database.
5846 if (!dbentry || !dbentry->tables)
5850 * Process all table entries in the message.
5852 for (i = 0; i < msg->m_nentries; i++)
5854 /* Remove from hashtable if present; we don't care if it's not. */
5855 (void) hash_search(dbentry->tables,
5856 (void *) &(msg->m_tableid[i]),
5863 * pgstat_recv_dropdb() -
5865 * Arrange for dead database removal
5869 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
5871 Oid dbid = msg->m_databaseid;
5872 PgStat_StatDBEntry *dbentry;
5875 * Lookup the database in the hashtable.
5877 dbentry = pgstat_get_db_entry(dbid, false);
5880 * If found, remove it (along with the db statfile).
5884 char statfile[MAXPGPATH];
5886 get_dbstat_filename(false, false, dbid, statfile, MAXPGPATH);
5888 elog(DEBUG2, "removing stats file \"%s\"", statfile);
5891 if (dbentry->tables != NULL)
5892 hash_destroy(dbentry->tables);
5893 if (dbentry->functions != NULL)
5894 hash_destroy(dbentry->functions);
5896 if (hash_search(pgStatDBHash,
5898 HASH_REMOVE, NULL) == NULL)
5900 (errmsg("database hash table corrupted during cleanup --- abort")));
5906 * pgstat_recv_resetcounter() -
5908 * Reset the statistics for the specified database.
5912 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
5914 PgStat_StatDBEntry *dbentry;
5917 * Lookup the database in the hashtable. Nothing to do if not there.
5919 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5925 * We simply throw away all the database's table entries by recreating a
5926 * new hash table for them.
5928 if (dbentry->tables != NULL)
5929 hash_destroy(dbentry->tables);
5930 if (dbentry->functions != NULL)
5931 hash_destroy(dbentry->functions);
5933 dbentry->tables = NULL;
5934 dbentry->functions = NULL;
5937 * Reset database-level stats, too. This creates empty hash tables for
5938 * tables and functions.
5940 reset_dbentry_counters(dbentry);
5944 * pgstat_recv_resetshared() -
5946 * Reset some shared statistics of the cluster.
5950 pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
5952 if (msg->m_resettarget == RESET_BGWRITER)
5954 /* Reset the global background writer statistics for the cluster. */
5955 memset(&globalStats, 0, sizeof(globalStats));
5956 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
5958 else if (msg->m_resettarget == RESET_ARCHIVER)
5960 /* Reset the archiver statistics for the cluster. */
5961 memset(&archiverStats, 0, sizeof(archiverStats));
5962 archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
5966 * Presumably the sender of this message validated the target, don't
5967 * complain here if it's not valid
5972 * pgstat_recv_resetsinglecounter() -
5974 * Reset a statistics for a single object
5978 pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len)
5980 PgStat_StatDBEntry *dbentry;
5982 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5987 /* Set the reset timestamp for the whole database */
5988 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
5990 /* Remove object if it exists, ignore it if not */
5991 if (msg->m_resettype == RESET_TABLE)
5992 (void) hash_search(dbentry->tables, (void *) &(msg->m_objectid),
5994 else if (msg->m_resettype == RESET_FUNCTION)
5995 (void) hash_search(dbentry->functions, (void *) &(msg->m_objectid),
6000 * pgstat_recv_autovac() -
6002 * Process an autovacuum signalling message.
6006 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
6008 PgStat_StatDBEntry *dbentry;
6011 * Store the last autovacuum time in the database's hashtable entry.
6013 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6015 dbentry->last_autovac_time = msg->m_start_time;
6019 * pgstat_recv_vacuum() -
6021 * Process a VACUUM message.
6025 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
6027 PgStat_StatDBEntry *dbentry;
6028 PgStat_StatTabEntry *tabentry;
6031 * Store the data in the table's hashtable entry.
6033 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6035 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
6037 tabentry->n_live_tuples = msg->m_live_tuples;
6038 tabentry->n_dead_tuples = msg->m_dead_tuples;
6040 if (msg->m_autovacuum)
6042 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
6043 tabentry->autovac_vacuum_count++;
6047 tabentry->vacuum_timestamp = msg->m_vacuumtime;
6048 tabentry->vacuum_count++;
6053 * pgstat_recv_analyze() -
6055 * Process an ANALYZE message.
6059 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
6061 PgStat_StatDBEntry *dbentry;
6062 PgStat_StatTabEntry *tabentry;
6065 * Store the data in the table's hashtable entry.
6067 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6069 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
6071 tabentry->n_live_tuples = msg->m_live_tuples;
6072 tabentry->n_dead_tuples = msg->m_dead_tuples;
6075 * If commanded, reset changes_since_analyze to zero. This forgets any
6076 * changes that were committed while the ANALYZE was in progress, but we
6077 * have no good way to estimate how many of those there were.
6079 if (msg->m_resetcounter)
6080 tabentry->changes_since_analyze = 0;
6082 if (msg->m_autovacuum)
6084 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
6085 tabentry->autovac_analyze_count++;
6089 tabentry->analyze_timestamp = msg->m_analyzetime;
6090 tabentry->analyze_count++;
6096 * pgstat_recv_archiver() -
6098 * Process a ARCHIVER message.
6102 pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len)
6106 /* Failed archival attempt */
6107 ++archiverStats.failed_count;
6108 memcpy(archiverStats.last_failed_wal, msg->m_xlog,
6109 sizeof(archiverStats.last_failed_wal));
6110 archiverStats.last_failed_timestamp = msg->m_timestamp;
6114 /* Successful archival operation */
6115 ++archiverStats.archived_count;
6116 memcpy(archiverStats.last_archived_wal, msg->m_xlog,
6117 sizeof(archiverStats.last_archived_wal));
6118 archiverStats.last_archived_timestamp = msg->m_timestamp;
6123 * pgstat_recv_bgwriter() -
6125 * Process a BGWRITER message.
6129 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
6131 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
6132 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
6133 globalStats.checkpoint_write_time += msg->m_checkpoint_write_time;
6134 globalStats.checkpoint_sync_time += msg->m_checkpoint_sync_time;
6135 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
6136 globalStats.buf_written_clean += msg->m_buf_written_clean;
6137 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
6138 globalStats.buf_written_backend += msg->m_buf_written_backend;
6139 globalStats.buf_fsync_backend += msg->m_buf_fsync_backend;
6140 globalStats.buf_alloc += msg->m_buf_alloc;
6144 * pgstat_recv_recoveryconflict() -
6146 * Process a RECOVERYCONFLICT message.
6150 pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len)
6152 PgStat_StatDBEntry *dbentry;
6154 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6156 switch (msg->m_reason)
6158 case PROCSIG_RECOVERY_CONFLICT_DATABASE:
6161 * Since we drop the information about the database as soon as it
6162 * replicates, there is no point in counting these conflicts.
6165 case PROCSIG_RECOVERY_CONFLICT_TABLESPACE:
6166 dbentry->n_conflict_tablespace++;
6168 case PROCSIG_RECOVERY_CONFLICT_LOCK:
6169 dbentry->n_conflict_lock++;
6171 case PROCSIG_RECOVERY_CONFLICT_SNAPSHOT:
6172 dbentry->n_conflict_snapshot++;
6174 case PROCSIG_RECOVERY_CONFLICT_BUFFERPIN:
6175 dbentry->n_conflict_bufferpin++;
6177 case PROCSIG_RECOVERY_CONFLICT_STARTUP_DEADLOCK:
6178 dbentry->n_conflict_startup_deadlock++;
6184 * pgstat_recv_deadlock() -
6186 * Process a DEADLOCK message.
6190 pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len)
6192 PgStat_StatDBEntry *dbentry;
6194 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6196 dbentry->n_deadlocks++;
6200 * pgstat_recv_tempfile() -
6202 * Process a TEMPFILE message.
6206 pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len)
6208 PgStat_StatDBEntry *dbentry;
6210 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6212 dbentry->n_temp_bytes += msg->m_filesize;
6213 dbentry->n_temp_files += 1;
6217 * pgstat_recv_funcstat() -
6219 * Count what the backend has done.
6223 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
6225 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
6226 PgStat_StatDBEntry *dbentry;
6227 PgStat_StatFuncEntry *funcentry;
6231 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6234 * Process all function entries in the message.
6236 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
6238 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
6239 (void *) &(funcmsg->f_id),
6240 HASH_ENTER, &found);
6245 * If it's a new function entry, initialize counters to the values
6248 funcentry->f_numcalls = funcmsg->f_numcalls;
6249 funcentry->f_total_time = funcmsg->f_total_time;
6250 funcentry->f_self_time = funcmsg->f_self_time;
6255 * Otherwise add the values to the existing entry.
6257 funcentry->f_numcalls += funcmsg->f_numcalls;
6258 funcentry->f_total_time += funcmsg->f_total_time;
6259 funcentry->f_self_time += funcmsg->f_self_time;
6265 * pgstat_recv_funcpurge() -
6267 * Arrange for dead function removal.
6271 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
6273 PgStat_StatDBEntry *dbentry;
6276 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
6279 * No need to purge if we don't even know the database.
6281 if (!dbentry || !dbentry->functions)
6285 * Process all function entries in the message.
6287 for (i = 0; i < msg->m_nentries; i++)
6289 /* Remove from hashtable if present; we don't care if it's not. */
6290 (void) hash_search(dbentry->functions,
6291 (void *) &(msg->m_functionid[i]),
6297 * pgstat_write_statsfile_needed() -
6299 * Do we need to write out any stats files?
6303 pgstat_write_statsfile_needed(void)
6305 if (pending_write_requests != NIL)
6308 /* Everything was written recently */
6313 * pgstat_db_requested() -
6315 * Checks whether stats for a particular DB need to be written to a file.
6319 pgstat_db_requested(Oid databaseid)
6322 * If any requests are outstanding at all, we should write the stats for
6323 * shared catalogs (the "database" with OID 0). This ensures that
6324 * backends will see up-to-date stats for shared catalogs, even though
6325 * they send inquiry messages mentioning only their own DB.
6327 if (databaseid == InvalidOid && pending_write_requests != NIL)
6330 /* Search to see if there's an open request to write this database. */
6331 if (list_member_oid(pending_write_requests, databaseid))
6338 * Convert a potentially unsafely truncated activity string (see
6339 * PgBackendStatus.st_activity_raw's documentation) into a correctly truncated
6342 * The returned string is allocated in the caller's memory context and may be
6346 pgstat_clip_activity(const char *raw_activity)
6353 * Some callers, like pgstat_get_backend_current_activity(), do not
6354 * guarantee that the buffer isn't concurrently modified. We try to take
6355 * care that the buffer is always terminated by a NUL byte regardless, but
6356 * let's still be paranoid about the string's length. In those cases the
6357 * underlying buffer is guaranteed to be pgstat_track_activity_query_size
6360 activity = pnstrdup(raw_activity, pgstat_track_activity_query_size - 1);
6362 /* now double-guaranteed to be NUL terminated */
6363 rawlen = strlen(activity);
6366 * All supported server-encodings make it possible to determine the length
6367 * of a multi-byte character from its first byte (this is not the case for
6368 * client encodings, see GB18030). As st_activity is always stored using
6369 * server encoding, this allows us to perform multi-byte aware truncation,
6370 * even if the string earlier was truncated in the middle of a multi-byte
6373 cliplen = pg_mbcliplen(activity, rawlen,
6374 pgstat_track_activity_query_size - 1);
6376 activity[cliplen] = '\0';