1 /* ----------
2 * pgstat.c
3 *
4 * All the statistics collector stuff hacked up in one big, ugly file.
5 *
6 * TODO: - Separate collector, postmaster and backend stuff
7 * into different files.
8 *
9 * - Add some automatic call for pgstat vacuuming.
10 *
11 * - Add a pgstat config column to pg_database, so this
12 * entire thing can be enabled/disabled on a per db basis.
13 *
14 * Copyright (c) 2001-2016, PostgreSQL Global Development Group
15 *
16 * src/backend/postmaster/pgstat.c
17 * ----------
18 */
19 #include "postgres.h"
20
21 #include <unistd.h>
22 #include <fcntl.h>
23 #include <sys/param.h>
24 #include <sys/time.h>
25 #include <sys/socket.h>
26 #include <netdb.h>
27 #include <netinet/in.h>
28 #include <arpa/inet.h>
29 #include <signal.h>
30 #include <time.h>
31 #ifdef HAVE_SYS_SELECT_H
32 #include <sys/select.h>
33 #endif
34
35 #include "pgstat.h"
36
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 "libpq/ip.h"
45 #include "libpq/libpq.h"
46 #include "libpq/pqsignal.h"
47 #include "mb/pg_wchar.h"
48 #include "miscadmin.h"
49 #include "pg_trace.h"
50 #include "postmaster/autovacuum.h"
51 #include "postmaster/fork_process.h"
52 #include "postmaster/postmaster.h"
53 #include "storage/backendid.h"
54 #include "storage/dsm.h"
55 #include "storage/fd.h"
56 #include "storage/ipc.h"
57 #include "storage/latch.h"
58 #include "storage/lmgr.h"
59 #include "storage/pg_shmem.h"
60 #include "storage/procsignal.h"
61 #include "storage/sinvaladt.h"
62 #include "utils/ascii.h"
63 #include "utils/guc.h"
64 #include "utils/memutils.h"
65 #include "utils/ps_status.h"
66 #include "utils/rel.h"
67 #include "utils/snapmgr.h"
68 #include "utils/timestamp.h"
69 #include "utils/tqual.h"
70
71
72 /* ----------
73 * Timer definitions.
74 * ----------
75 */
76 #define PGSTAT_STAT_INTERVAL 500 /* Minimum time between stats file
77 * updates; in milliseconds. */
78
79 #define PGSTAT_RETRY_DELAY 10 /* How long to wait between checks for
80 * a new file; in milliseconds. */
81
82 #define PGSTAT_MAX_WAIT_TIME 10000 /* Maximum time to wait for a stats
83 * file update; in milliseconds. */
84
85 #define PGSTAT_INQ_INTERVAL 640 /* How often to ping the collector for
86 * a new file; in milliseconds. */
87
88 #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a
89 * failed statistics collector; in
90 * seconds. */
91
92 #define PGSTAT_POLL_LOOP_COUNT (PGSTAT_MAX_WAIT_TIME / PGSTAT_RETRY_DELAY)
93 #define PGSTAT_INQ_LOOP_COUNT (PGSTAT_INQ_INTERVAL / PGSTAT_RETRY_DELAY)
94
95 /* Minimum receive buffer size for the collector's socket. */
96 #define PGSTAT_MIN_RCVBUF (100 * 1024)
97
98
99 /* ----------
100 * The initial size hints for the hash tables used in the collector.
101 * ----------
102 */
103 #define PGSTAT_DB_HASH_SIZE 16
104 #define PGSTAT_TAB_HASH_SIZE 512
105 #define PGSTAT_FUNCTION_HASH_SIZE 512
106
107
108 /* ----------
109 * GUC parameters
110 * ----------
111 */
112 bool pgstat_track_activities = false;
113 bool pgstat_track_counts = false;
114 int pgstat_track_functions = TRACK_FUNC_OFF;
115 int pgstat_track_activity_query_size = 1024;
116
117 /* ----------
118 * Built from GUC parameter
119 * ----------
120 */
121 char *pgstat_stat_directory = NULL;
122 char *pgstat_stat_filename = NULL;
123 char *pgstat_stat_tmpname = NULL;
124
125 /*
126 * BgWriter global statistics counters (unused in other processes).
127 * Stored directly in a stats message structure so it can be sent
128 * without needing to copy things around. We assume this inits to zeroes.
129 */
130 PgStat_MsgBgWriter BgWriterStats;
131
132 /* ----------
133 * Local data
134 * ----------
135 */
136 NON_EXEC_STATIC pgsocket pgStatSock = PGINVALID_SOCKET;
137
138 static struct sockaddr_storage pgStatAddr;
139
140 static time_t last_pgstat_start_time;
141
142 static bool pgStatRunningInCollector = false;
143
144 /*
145 * Structures in which backends store per-table info that's waiting to be
146 * sent to the collector.
147 *
148 * NOTE: once allocated, TabStatusArray structures are never moved or deleted
149 * for the life of the backend. Also, we zero out the t_id fields of the
150 * contained PgStat_TableStatus structs whenever they are not actively in use.
151 * This allows relcache pgstat_info pointers to be treated as long-lived data,
152 * avoiding repeated searches in pgstat_initstats() when a relation is
153 * repeatedly opened during a transaction.
154 */
155 #define TABSTAT_QUANTUM 100 /* we alloc this many at a time */
156
157 typedef struct TabStatusArray
158 {
159 struct TabStatusArray *tsa_next; /* link to next array, if any */
160 int tsa_used; /* # entries currently used */
161 PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */
162 } TabStatusArray;
163
164 static TabStatusArray *pgStatTabList = NULL;
165
166 /*
167 * Backends store per-function info that's waiting to be sent to the collector
168 * in this hash table (indexed by function OID).
169 */
170 static HTAB *pgStatFunctions = NULL;
171
172 /*
173 * Indicates if backend has some function stats that it hasn't yet
174 * sent to the collector.
175 */
176 static bool have_function_stats = false;
177
178 /*
179 * Tuple insertion/deletion counts for an open transaction can't be propagated
180 * into PgStat_TableStatus counters until we know if it is going to commit
181 * or abort. Hence, we keep these counts in per-subxact structs that live
182 * in TopTransactionContext. This data structure is designed on the assumption
183 * that subxacts won't usually modify very many tables.
184 */
185 typedef struct PgStat_SubXactStatus
186 {
187 int nest_level; /* subtransaction nest level */
188 struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
189 PgStat_TableXactStatus *first; /* head of list for this subxact */
190 } PgStat_SubXactStatus;
191
192 static PgStat_SubXactStatus *pgStatXactStack = NULL;
193
194 static int pgStatXactCommit = 0;
195 static int pgStatXactRollback = 0;
196 PgStat_Counter pgStatBlockReadTime = 0;
197 PgStat_Counter pgStatBlockWriteTime = 0;
198
199 /* Record that's written to 2PC state file when pgstat state is persisted */
200 typedef struct TwoPhasePgStatRecord
201 {
202 PgStat_Counter tuples_inserted; /* tuples inserted in xact */
203 PgStat_Counter tuples_updated; /* tuples updated in xact */
204 PgStat_Counter tuples_deleted; /* tuples deleted in xact */
205 PgStat_Counter inserted_pre_trunc; /* tuples inserted prior to truncate */
206 PgStat_Counter updated_pre_trunc; /* tuples updated prior to truncate */
207 PgStat_Counter deleted_pre_trunc; /* tuples deleted prior to truncate */
208 Oid t_id; /* table's OID */
209 bool t_shared; /* is it a shared catalog? */
210 bool t_truncated; /* was the relation truncated? */
211 } TwoPhasePgStatRecord;
212
213 /*
214 * Info about current "snapshot" of stats file
215 */
216 static MemoryContext pgStatLocalContext = NULL;
217 static HTAB *pgStatDBHash = NULL;
218 static LocalPgBackendStatus *localBackendStatusTable = NULL;
219 static int localNumBackends = 0;
220
221 /*
222 * Cluster wide statistics, kept in the stats collector.
223 * Contains statistics that are not collected per database
224 * or per table.
225 */
226 static PgStat_ArchiverStats archiverStats;
227 static PgStat_GlobalStats globalStats;
228
229 /*
230 * List of OIDs of databases we need to write out. If an entry is InvalidOid,
231 * it means to write only the shared-catalog stats ("DB 0"); otherwise, we
232 * will write both that DB's data and the shared stats.
233 */
234 static List *pending_write_requests = NIL;
235
236 /* Signal handler flags */
237 static volatile bool need_exit = false;
238 static volatile bool got_SIGHUP = false;
239
240 /*
241 * Total time charged to functions so far in the current backend.
242 * We use this to help separate "self" and "other" time charges.
243 * (We assume this initializes to zero.)
244 */
245 static instr_time total_func_time;
246
247
248 /* ----------
249 * Local function forward declarations
250 * ----------
251 */
252 #ifdef EXEC_BACKEND
253 static pid_t pgstat_forkexec(void);
254 #endif
255
256 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
257 static void pgstat_exit(SIGNAL_ARGS);
258 static void pgstat_beshutdown_hook(int code, Datum arg);
259 static void pgstat_sighup_handler(SIGNAL_ARGS);
260
261 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
262 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
263 Oid tableoid, bool create);
264 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
265 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
266 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
267 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
268 static void backend_read_statsfile(void);
269 static void pgstat_read_current_status(void);
270
271 static bool pgstat_write_statsfile_needed(void);
272 static bool pgstat_db_requested(Oid databaseid);
273
274 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
275 static void pgstat_send_funcstats(void);
276 static HTAB *pgstat_collect_oids(Oid catalogid);
277
278 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
279
280 static void pgstat_setup_memcxt(void);
281
282 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
283 static void pgstat_send(void *msg, int len);
284
285 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
286 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
287 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
288 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
289 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
290 static void pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len);
291 static void pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len);
292 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
293 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
294 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
295 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
296 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
297 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
298 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
299 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
300 static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
301 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
302
303 /* ------------------------------------------------------------
304 * Public functions called from postmaster follow
305 * ------------------------------------------------------------
306 */
307
308 /* ----------
309 * pgstat_init() -
310 *
311 * Called from postmaster at startup. Create the resources required
312 * by the statistics collector process. If unable to do so, do not
313 * fail --- better to let the postmaster start with stats collection
314 * disabled.
315 * ----------
316 */
317 void
pgstat_init(void)318 pgstat_init(void)
319 {
320 ACCEPT_TYPE_ARG3 alen;
321 struct addrinfo *addrs = NULL,
322 *addr,
323 hints;
324 int ret;
325 fd_set rset;
326 struct timeval tv;
327 char test_byte;
328 int sel_res;
329 int tries = 0;
330
331 #define TESTBYTEVAL ((char) 199)
332
333 /*
334 * This static assertion verifies that we didn't mess up the calculations
335 * involved in selecting maximum payload sizes for our UDP messages.
336 * Because the only consequence of overrunning PGSTAT_MAX_MSG_SIZE would
337 * be silent performance loss from fragmentation, it seems worth having a
338 * compile-time cross-check that we didn't.
339 */
340 StaticAssertStmt(sizeof(PgStat_Msg) <= PGSTAT_MAX_MSG_SIZE,
341 "maximum stats message size exceeds PGSTAT_MAX_MSG_SIZE");
342
343 /*
344 * Create the UDP socket for sending and receiving statistic messages
345 */
346 hints.ai_flags = AI_PASSIVE;
347 hints.ai_family = AF_UNSPEC;
348 hints.ai_socktype = SOCK_DGRAM;
349 hints.ai_protocol = 0;
350 hints.ai_addrlen = 0;
351 hints.ai_addr = NULL;
352 hints.ai_canonname = NULL;
353 hints.ai_next = NULL;
354 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
355 if (ret || !addrs)
356 {
357 ereport(LOG,
358 (errmsg("could not resolve \"localhost\": %s",
359 gai_strerror(ret))));
360 goto startup_failed;
361 }
362
363 /*
364 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
365 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
366 * when kernel will reject IPv6). Worse, the failure may occur at the
367 * bind() or perhaps even connect() stage. So we must loop through the
368 * results till we find a working combination. We will generate LOG
369 * messages, but no error, for bogus combinations.
370 */
371 for (addr = addrs; addr; addr = addr->ai_next)
372 {
373 #ifdef HAVE_UNIX_SOCKETS
374 /* Ignore AF_UNIX sockets, if any are returned. */
375 if (addr->ai_family == AF_UNIX)
376 continue;
377 #endif
378
379 if (++tries > 1)
380 ereport(LOG,
381 (errmsg("trying another address for the statistics collector")));
382
383 /*
384 * Create the socket.
385 */
386 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) == PGINVALID_SOCKET)
387 {
388 ereport(LOG,
389 (errcode_for_socket_access(),
390 errmsg("could not create socket for statistics collector: %m")));
391 continue;
392 }
393
394 /*
395 * Bind it to a kernel assigned port on localhost and get the assigned
396 * port via getsockname().
397 */
398 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
399 {
400 ereport(LOG,
401 (errcode_for_socket_access(),
402 errmsg("could not bind socket for statistics collector: %m")));
403 closesocket(pgStatSock);
404 pgStatSock = PGINVALID_SOCKET;
405 continue;
406 }
407
408 alen = sizeof(pgStatAddr);
409 if (getsockname(pgStatSock, (struct sockaddr *) & pgStatAddr, &alen) < 0)
410 {
411 ereport(LOG,
412 (errcode_for_socket_access(),
413 errmsg("could not get address of socket for statistics collector: %m")));
414 closesocket(pgStatSock);
415 pgStatSock = PGINVALID_SOCKET;
416 continue;
417 }
418
419 /*
420 * Connect the socket to its own address. This saves a few cycles by
421 * not having to respecify the target address on every send. This also
422 * provides a kernel-level check that only packets from this same
423 * address will be received.
424 */
425 if (connect(pgStatSock, (struct sockaddr *) & pgStatAddr, alen) < 0)
426 {
427 ereport(LOG,
428 (errcode_for_socket_access(),
429 errmsg("could not connect socket for statistics collector: %m")));
430 closesocket(pgStatSock);
431 pgStatSock = PGINVALID_SOCKET;
432 continue;
433 }
434
435 /*
436 * Try to send and receive a one-byte test message on the socket. This
437 * is to catch situations where the socket can be created but will not
438 * actually pass data (for instance, because kernel packet filtering
439 * rules prevent it).
440 */
441 test_byte = TESTBYTEVAL;
442
443 retry1:
444 if (send(pgStatSock, &test_byte, 1, 0) != 1)
445 {
446 if (errno == EINTR)
447 goto retry1; /* if interrupted, just retry */
448 ereport(LOG,
449 (errcode_for_socket_access(),
450 errmsg("could not send test message on socket for statistics collector: %m")));
451 closesocket(pgStatSock);
452 pgStatSock = PGINVALID_SOCKET;
453 continue;
454 }
455
456 /*
457 * There could possibly be a little delay before the message can be
458 * received. We arbitrarily allow up to half a second before deciding
459 * it's broken.
460 */
461 for (;;) /* need a loop to handle EINTR */
462 {
463 FD_ZERO(&rset);
464 FD_SET(pgStatSock, &rset);
465
466 tv.tv_sec = 0;
467 tv.tv_usec = 500000;
468 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
469 if (sel_res >= 0 || errno != EINTR)
470 break;
471 }
472 if (sel_res < 0)
473 {
474 ereport(LOG,
475 (errcode_for_socket_access(),
476 errmsg("select() failed in statistics collector: %m")));
477 closesocket(pgStatSock);
478 pgStatSock = PGINVALID_SOCKET;
479 continue;
480 }
481 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
482 {
483 /*
484 * This is the case we actually think is likely, so take pains to
485 * give a specific message for it.
486 *
487 * errno will not be set meaningfully here, so don't use it.
488 */
489 ereport(LOG,
490 (errcode(ERRCODE_CONNECTION_FAILURE),
491 errmsg("test message did not get through on socket for statistics collector")));
492 closesocket(pgStatSock);
493 pgStatSock = PGINVALID_SOCKET;
494 continue;
495 }
496
497 test_byte++; /* just make sure variable is changed */
498
499 retry2:
500 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
501 {
502 if (errno == EINTR)
503 goto retry2; /* if interrupted, just retry */
504 ereport(LOG,
505 (errcode_for_socket_access(),
506 errmsg("could not receive test message on socket for statistics collector: %m")));
507 closesocket(pgStatSock);
508 pgStatSock = PGINVALID_SOCKET;
509 continue;
510 }
511
512 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
513 {
514 ereport(LOG,
515 (errcode(ERRCODE_INTERNAL_ERROR),
516 errmsg("incorrect test message transmission on socket for statistics collector")));
517 closesocket(pgStatSock);
518 pgStatSock = PGINVALID_SOCKET;
519 continue;
520 }
521
522 /* If we get here, we have a working socket */
523 break;
524 }
525
526 /* Did we find a working address? */
527 if (!addr || pgStatSock == PGINVALID_SOCKET)
528 goto startup_failed;
529
530 /*
531 * Set the socket to non-blocking IO. This ensures that if the collector
532 * falls behind, statistics messages will be discarded; backends won't
533 * block waiting to send messages to the collector.
534 */
535 if (!pg_set_noblock(pgStatSock))
536 {
537 ereport(LOG,
538 (errcode_for_socket_access(),
539 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
540 goto startup_failed;
541 }
542
543 /*
544 * Try to ensure that the socket's receive buffer is at least
545 * PGSTAT_MIN_RCVBUF bytes, so that it won't easily overflow and lose
546 * data. Use of UDP protocol means that we are willing to lose data under
547 * heavy load, but we don't want it to happen just because of ridiculously
548 * small default buffer sizes (such as 8KB on older Windows versions).
549 */
550 {
551 int old_rcvbuf;
552 int new_rcvbuf;
553 ACCEPT_TYPE_ARG3 rcvbufsize = sizeof(old_rcvbuf);
554
555 if (getsockopt(pgStatSock, SOL_SOCKET, SO_RCVBUF,
556 (char *) &old_rcvbuf, &rcvbufsize) < 0)
557 {
558 elog(LOG, "getsockopt(SO_RCVBUF) failed: %m");
559 /* if we can't get existing size, always try to set it */
560 old_rcvbuf = 0;
561 }
562
563 new_rcvbuf = PGSTAT_MIN_RCVBUF;
564 if (old_rcvbuf < new_rcvbuf)
565 {
566 if (setsockopt(pgStatSock, SOL_SOCKET, SO_RCVBUF,
567 (char *) &new_rcvbuf, sizeof(new_rcvbuf)) < 0)
568 elog(LOG, "setsockopt(SO_RCVBUF) failed: %m");
569 }
570 }
571
572 pg_freeaddrinfo_all(hints.ai_family, addrs);
573
574 return;
575
576 startup_failed:
577 ereport(LOG,
578 (errmsg("disabling statistics collector for lack of working socket")));
579
580 if (addrs)
581 pg_freeaddrinfo_all(hints.ai_family, addrs);
582
583 if (pgStatSock != PGINVALID_SOCKET)
584 closesocket(pgStatSock);
585 pgStatSock = PGINVALID_SOCKET;
586
587 /*
588 * Adjust GUC variables to suppress useless activity, and for debugging
589 * purposes (seeing track_counts off is a clue that we failed here). We
590 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
591 * on from postgresql.conf without a restart.
592 */
593 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
594 }
595
596 /*
597 * subroutine for pgstat_reset_all
598 */
599 static void
pgstat_reset_remove_files(const char * directory)600 pgstat_reset_remove_files(const char *directory)
601 {
602 DIR *dir;
603 struct dirent *entry;
604 char fname[MAXPGPATH * 2];
605
606 dir = AllocateDir(directory);
607 while ((entry = ReadDir(dir, directory)) != NULL)
608 {
609 int nchars;
610 Oid tmp_oid;
611
612 /*
613 * Skip directory entries that don't match the file names we write.
614 * See get_dbstat_filename for the database-specific pattern.
615 */
616 if (strncmp(entry->d_name, "global.", 7) == 0)
617 nchars = 7;
618 else
619 {
620 nchars = 0;
621 (void) sscanf(entry->d_name, "db_%u.%n",
622 &tmp_oid, &nchars);
623 if (nchars <= 0)
624 continue;
625 /* %u allows leading whitespace, so reject that */
626 if (strchr("0123456789", entry->d_name[3]) == NULL)
627 continue;
628 }
629
630 if (strcmp(entry->d_name + nchars, "tmp") != 0 &&
631 strcmp(entry->d_name + nchars, "stat") != 0)
632 continue;
633
634 snprintf(fname, sizeof(fname), "%s/%s", directory,
635 entry->d_name);
636 unlink(fname);
637 }
638 FreeDir(dir);
639 }
640
641 /*
642 * pgstat_reset_all() -
643 *
644 * Remove the stats files. This is currently used only if WAL
645 * recovery is needed after a crash.
646 */
647 void
pgstat_reset_all(void)648 pgstat_reset_all(void)
649 {
650 pgstat_reset_remove_files(pgstat_stat_directory);
651 pgstat_reset_remove_files(PGSTAT_STAT_PERMANENT_DIRECTORY);
652 }
653
654 #ifdef EXEC_BACKEND
655
656 /*
657 * pgstat_forkexec() -
658 *
659 * Format up the arglist for, then fork and exec, statistics collector process
660 */
661 static pid_t
pgstat_forkexec(void)662 pgstat_forkexec(void)
663 {
664 char *av[10];
665 int ac = 0;
666
667 av[ac++] = "postgres";
668 av[ac++] = "--forkcol";
669 av[ac++] = NULL; /* filled in by postmaster_forkexec */
670
671 av[ac] = NULL;
672 Assert(ac < lengthof(av));
673
674 return postmaster_forkexec(ac, av);
675 }
676 #endif /* EXEC_BACKEND */
677
678
679 /*
680 * pgstat_start() -
681 *
682 * Called from postmaster at startup or after an existing collector
683 * died. Attempt to fire up a fresh statistics collector.
684 *
685 * Returns PID of child process, or 0 if fail.
686 *
687 * Note: if fail, we will be called again from the postmaster main loop.
688 */
689 int
pgstat_start(void)690 pgstat_start(void)
691 {
692 time_t curtime;
693 pid_t pgStatPid;
694
695 /*
696 * Check that the socket is there, else pgstat_init failed and we can do
697 * nothing useful.
698 */
699 if (pgStatSock == PGINVALID_SOCKET)
700 return 0;
701
702 /*
703 * Do nothing if too soon since last collector start. This is a safety
704 * valve to protect against continuous respawn attempts if the collector
705 * is dying immediately at launch. Note that since we will be re-called
706 * from the postmaster main loop, we will get another chance later.
707 */
708 curtime = time(NULL);
709 if ((unsigned int) (curtime - last_pgstat_start_time) <
710 (unsigned int) PGSTAT_RESTART_INTERVAL)
711 return 0;
712 last_pgstat_start_time = curtime;
713
714 /*
715 * Okay, fork off the collector.
716 */
717 #ifdef EXEC_BACKEND
718 switch ((pgStatPid = pgstat_forkexec()))
719 #else
720 switch ((pgStatPid = fork_process()))
721 #endif
722 {
723 case -1:
724 ereport(LOG,
725 (errmsg("could not fork statistics collector: %m")));
726 return 0;
727
728 #ifndef EXEC_BACKEND
729 case 0:
730 /* in postmaster child ... */
731 InitPostmasterChild();
732
733 /* Close the postmaster's sockets */
734 ClosePostmasterPorts(false);
735
736 /* Drop our connection to postmaster's shared memory, as well */
737 dsm_detach_all();
738 PGSharedMemoryDetach();
739
740 PgstatCollectorMain(0, NULL);
741 break;
742 #endif
743
744 default:
745 return (int) pgStatPid;
746 }
747
748 /* shouldn't get here */
749 return 0;
750 }
751
752 void
allow_immediate_pgstat_restart(void)753 allow_immediate_pgstat_restart(void)
754 {
755 last_pgstat_start_time = 0;
756 }
757
758 /* ------------------------------------------------------------
759 * Public functions used by backends follow
760 *------------------------------------------------------------
761 */
762
763
764 /* ----------
765 * pgstat_report_stat() -
766 *
767 * Called from tcop/postgres.c to send the so far collected per-table
768 * and function usage statistics to the collector. Note that this is
769 * called only when not within a transaction, so it is fair to use
770 * transaction stop time as an approximation of current time.
771 * ----------
772 */
773 void
pgstat_report_stat(bool force)774 pgstat_report_stat(bool force)
775 {
776 /* we assume this inits to all zeroes: */
777 static const PgStat_TableCounts all_zeroes;
778 static TimestampTz last_report = 0;
779
780 TimestampTz now;
781 PgStat_MsgTabstat regular_msg;
782 PgStat_MsgTabstat shared_msg;
783 TabStatusArray *tsa;
784 int i;
785
786 /* Don't expend a clock check if nothing to do */
787 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0) &&
788 pgStatXactCommit == 0 && pgStatXactRollback == 0 &&
789 !have_function_stats)
790 return;
791
792 /*
793 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
794 * msec since we last sent one, or the caller wants to force stats out.
795 */
796 now = GetCurrentTransactionStopTimestamp();
797 if (!force &&
798 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
799 return;
800 last_report = now;
801
802 /*
803 * Scan through the TabStatusArray struct(s) to find tables that actually
804 * have counts, and build messages to send. We have to separate shared
805 * relations from regular ones because the databaseid field in the message
806 * header has to depend on that.
807 */
808 regular_msg.m_databaseid = MyDatabaseId;
809 shared_msg.m_databaseid = InvalidOid;
810 regular_msg.m_nentries = 0;
811 shared_msg.m_nentries = 0;
812
813 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
814 {
815 for (i = 0; i < tsa->tsa_used; i++)
816 {
817 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
818 PgStat_MsgTabstat *this_msg;
819 PgStat_TableEntry *this_ent;
820
821 /* Shouldn't have any pending transaction-dependent counts */
822 Assert(entry->trans == NULL);
823
824 /*
825 * Ignore entries that didn't accumulate any actual counts, such
826 * as indexes that were opened by the planner but not used.
827 */
828 if (memcmp(&entry->t_counts, &all_zeroes,
829 sizeof(PgStat_TableCounts)) == 0)
830 continue;
831
832 /*
833 * OK, insert data into the appropriate message, and send if full.
834 */
835 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
836 this_ent = &this_msg->m_entry[this_msg->m_nentries];
837 this_ent->t_id = entry->t_id;
838 memcpy(&this_ent->t_counts, &entry->t_counts,
839 sizeof(PgStat_TableCounts));
840 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
841 {
842 pgstat_send_tabstat(this_msg);
843 this_msg->m_nentries = 0;
844 }
845 }
846 /* zero out TableStatus structs after use */
847 MemSet(tsa->tsa_entries, 0,
848 tsa->tsa_used * sizeof(PgStat_TableStatus));
849 tsa->tsa_used = 0;
850 }
851
852 /*
853 * Send partial messages. Make sure that any pending xact commit/abort
854 * gets counted, even if there are no table stats to send.
855 */
856 if (regular_msg.m_nentries > 0 ||
857 pgStatXactCommit > 0 || pgStatXactRollback > 0)
858 pgstat_send_tabstat(®ular_msg);
859 if (shared_msg.m_nentries > 0)
860 pgstat_send_tabstat(&shared_msg);
861
862 /* Now, send function statistics */
863 pgstat_send_funcstats();
864 }
865
866 /*
867 * Subroutine for pgstat_report_stat: finish and send a tabstat message
868 */
869 static void
pgstat_send_tabstat(PgStat_MsgTabstat * tsmsg)870 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
871 {
872 int n;
873 int len;
874
875 /* It's unlikely we'd get here with no socket, but maybe not impossible */
876 if (pgStatSock == PGINVALID_SOCKET)
877 return;
878
879 /*
880 * Report and reset accumulated xact commit/rollback and I/O timings
881 * whenever we send a normal tabstat message
882 */
883 if (OidIsValid(tsmsg->m_databaseid))
884 {
885 tsmsg->m_xact_commit = pgStatXactCommit;
886 tsmsg->m_xact_rollback = pgStatXactRollback;
887 tsmsg->m_block_read_time = pgStatBlockReadTime;
888 tsmsg->m_block_write_time = pgStatBlockWriteTime;
889 pgStatXactCommit = 0;
890 pgStatXactRollback = 0;
891 pgStatBlockReadTime = 0;
892 pgStatBlockWriteTime = 0;
893 }
894 else
895 {
896 tsmsg->m_xact_commit = 0;
897 tsmsg->m_xact_rollback = 0;
898 tsmsg->m_block_read_time = 0;
899 tsmsg->m_block_write_time = 0;
900 }
901
902 n = tsmsg->m_nentries;
903 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
904 n * sizeof(PgStat_TableEntry);
905
906 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
907 pgstat_send(tsmsg, len);
908 }
909
910 /*
911 * Subroutine for pgstat_report_stat: populate and send a function stat message
912 */
913 static void
pgstat_send_funcstats(void)914 pgstat_send_funcstats(void)
915 {
916 /* we assume this inits to all zeroes: */
917 static const PgStat_FunctionCounts all_zeroes;
918
919 PgStat_MsgFuncstat msg;
920 PgStat_BackendFunctionEntry *entry;
921 HASH_SEQ_STATUS fstat;
922
923 if (pgStatFunctions == NULL)
924 return;
925
926 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
927 msg.m_databaseid = MyDatabaseId;
928 msg.m_nentries = 0;
929
930 hash_seq_init(&fstat, pgStatFunctions);
931 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
932 {
933 PgStat_FunctionEntry *m_ent;
934
935 /* Skip it if no counts accumulated since last time */
936 if (memcmp(&entry->f_counts, &all_zeroes,
937 sizeof(PgStat_FunctionCounts)) == 0)
938 continue;
939
940 /* need to convert format of time accumulators */
941 m_ent = &msg.m_entry[msg.m_nentries];
942 m_ent->f_id = entry->f_id;
943 m_ent->f_numcalls = entry->f_counts.f_numcalls;
944 m_ent->f_total_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_total_time);
945 m_ent->f_self_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_self_time);
946
947 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
948 {
949 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
950 msg.m_nentries * sizeof(PgStat_FunctionEntry));
951 msg.m_nentries = 0;
952 }
953
954 /* reset the entry's counts */
955 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
956 }
957
958 if (msg.m_nentries > 0)
959 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
960 msg.m_nentries * sizeof(PgStat_FunctionEntry));
961
962 have_function_stats = false;
963 }
964
965
966 /* ----------
967 * pgstat_vacuum_stat() -
968 *
969 * Will tell the collector about objects he can get rid of.
970 * ----------
971 */
972 void
pgstat_vacuum_stat(void)973 pgstat_vacuum_stat(void)
974 {
975 HTAB *htab;
976 PgStat_MsgTabpurge msg;
977 PgStat_MsgFuncpurge f_msg;
978 HASH_SEQ_STATUS hstat;
979 PgStat_StatDBEntry *dbentry;
980 PgStat_StatTabEntry *tabentry;
981 PgStat_StatFuncEntry *funcentry;
982 int len;
983
984 if (pgStatSock == PGINVALID_SOCKET)
985 return;
986
987 /*
988 * If not done for this transaction, read the statistics collector stats
989 * file into some hash tables.
990 */
991 backend_read_statsfile();
992
993 /*
994 * Read pg_database and make a list of OIDs of all existing databases
995 */
996 htab = pgstat_collect_oids(DatabaseRelationId);
997
998 /*
999 * Search the database hash table for dead databases and tell the
1000 * collector to drop them.
1001 */
1002 hash_seq_init(&hstat, pgStatDBHash);
1003 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
1004 {
1005 Oid dbid = dbentry->databaseid;
1006
1007 CHECK_FOR_INTERRUPTS();
1008
1009 /* the DB entry for shared tables (with InvalidOid) is never dropped */
1010 if (OidIsValid(dbid) &&
1011 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
1012 pgstat_drop_database(dbid);
1013 }
1014
1015 /* Clean up */
1016 hash_destroy(htab);
1017
1018 /*
1019 * Lookup our own database entry; if not found, nothing more to do.
1020 */
1021 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1022 (void *) &MyDatabaseId,
1023 HASH_FIND, NULL);
1024 if (dbentry == NULL || dbentry->tables == NULL)
1025 return;
1026
1027 /*
1028 * Similarly to above, make a list of all known relations in this DB.
1029 */
1030 htab = pgstat_collect_oids(RelationRelationId);
1031
1032 /*
1033 * Initialize our messages table counter to zero
1034 */
1035 msg.m_nentries = 0;
1036
1037 /*
1038 * Check for all tables listed in stats hashtable if they still exist.
1039 */
1040 hash_seq_init(&hstat, dbentry->tables);
1041 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
1042 {
1043 Oid tabid = tabentry->tableid;
1044
1045 CHECK_FOR_INTERRUPTS();
1046
1047 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
1048 continue;
1049
1050 /*
1051 * Not there, so add this table's Oid to the message
1052 */
1053 msg.m_tableid[msg.m_nentries++] = tabid;
1054
1055 /*
1056 * If the message is full, send it out and reinitialize to empty
1057 */
1058 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
1059 {
1060 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
1061 +msg.m_nentries * sizeof(Oid);
1062
1063 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1064 msg.m_databaseid = MyDatabaseId;
1065 pgstat_send(&msg, len);
1066
1067 msg.m_nentries = 0;
1068 }
1069 }
1070
1071 /*
1072 * Send the rest
1073 */
1074 if (msg.m_nentries > 0)
1075 {
1076 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
1077 +msg.m_nentries * sizeof(Oid);
1078
1079 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1080 msg.m_databaseid = MyDatabaseId;
1081 pgstat_send(&msg, len);
1082 }
1083
1084 /* Clean up */
1085 hash_destroy(htab);
1086
1087 /*
1088 * Now repeat the above steps for functions. However, we needn't bother
1089 * in the common case where no function stats are being collected.
1090 */
1091 if (dbentry->functions != NULL &&
1092 hash_get_num_entries(dbentry->functions) > 0)
1093 {
1094 htab = pgstat_collect_oids(ProcedureRelationId);
1095
1096 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
1097 f_msg.m_databaseid = MyDatabaseId;
1098 f_msg.m_nentries = 0;
1099
1100 hash_seq_init(&hstat, dbentry->functions);
1101 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
1102 {
1103 Oid funcid = funcentry->functionid;
1104
1105 CHECK_FOR_INTERRUPTS();
1106
1107 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
1108 continue;
1109
1110 /*
1111 * Not there, so add this function's Oid to the message
1112 */
1113 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
1114
1115 /*
1116 * If the message is full, send it out and reinitialize to empty
1117 */
1118 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1119 {
1120 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1121 +f_msg.m_nentries * sizeof(Oid);
1122
1123 pgstat_send(&f_msg, len);
1124
1125 f_msg.m_nentries = 0;
1126 }
1127 }
1128
1129 /*
1130 * Send the rest
1131 */
1132 if (f_msg.m_nentries > 0)
1133 {
1134 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1135 +f_msg.m_nentries * sizeof(Oid);
1136
1137 pgstat_send(&f_msg, len);
1138 }
1139
1140 hash_destroy(htab);
1141 }
1142 }
1143
1144
1145 /* ----------
1146 * pgstat_collect_oids() -
1147 *
1148 * Collect the OIDs of all objects listed in the specified system catalog
1149 * into a temporary hash table. Caller should hash_destroy the result
1150 * when done with it. (However, we make the table in CurrentMemoryContext
1151 * so that it will be freed properly in event of an error.)
1152 * ----------
1153 */
1154 static HTAB *
pgstat_collect_oids(Oid catalogid)1155 pgstat_collect_oids(Oid catalogid)
1156 {
1157 HTAB *htab;
1158 HASHCTL hash_ctl;
1159 Relation rel;
1160 HeapScanDesc scan;
1161 HeapTuple tup;
1162 Snapshot snapshot;
1163
1164 memset(&hash_ctl, 0, sizeof(hash_ctl));
1165 hash_ctl.keysize = sizeof(Oid);
1166 hash_ctl.entrysize = sizeof(Oid);
1167 hash_ctl.hcxt = CurrentMemoryContext;
1168 htab = hash_create("Temporary table of OIDs",
1169 PGSTAT_TAB_HASH_SIZE,
1170 &hash_ctl,
1171 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1172
1173 rel = heap_open(catalogid, AccessShareLock);
1174 snapshot = RegisterSnapshot(GetLatestSnapshot());
1175 scan = heap_beginscan(rel, snapshot, 0, NULL);
1176 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1177 {
1178 Oid thisoid = HeapTupleGetOid(tup);
1179
1180 CHECK_FOR_INTERRUPTS();
1181
1182 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1183 }
1184 heap_endscan(scan);
1185 UnregisterSnapshot(snapshot);
1186 heap_close(rel, AccessShareLock);
1187
1188 return htab;
1189 }
1190
1191
1192 /* ----------
1193 * pgstat_drop_database() -
1194 *
1195 * Tell the collector that we just dropped a database.
1196 * (If the message gets lost, we will still clean the dead DB eventually
1197 * via future invocations of pgstat_vacuum_stat().)
1198 * ----------
1199 */
1200 void
pgstat_drop_database(Oid databaseid)1201 pgstat_drop_database(Oid databaseid)
1202 {
1203 PgStat_MsgDropdb msg;
1204
1205 if (pgStatSock == PGINVALID_SOCKET)
1206 return;
1207
1208 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1209 msg.m_databaseid = databaseid;
1210 pgstat_send(&msg, sizeof(msg));
1211 }
1212
1213
1214 /* ----------
1215 * pgstat_drop_relation() -
1216 *
1217 * Tell the collector that we just dropped a relation.
1218 * (If the message gets lost, we will still clean the dead entry eventually
1219 * via future invocations of pgstat_vacuum_stat().)
1220 *
1221 * Currently not used for lack of any good place to call it; we rely
1222 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1223 * ----------
1224 */
1225 #ifdef NOT_USED
1226 void
pgstat_drop_relation(Oid relid)1227 pgstat_drop_relation(Oid relid)
1228 {
1229 PgStat_MsgTabpurge msg;
1230 int len;
1231
1232 if (pgStatSock == PGINVALID_SOCKET)
1233 return;
1234
1235 msg.m_tableid[0] = relid;
1236 msg.m_nentries = 1;
1237
1238 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) +sizeof(Oid);
1239
1240 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1241 msg.m_databaseid = MyDatabaseId;
1242 pgstat_send(&msg, len);
1243 }
1244 #endif /* NOT_USED */
1245
1246
1247 /* ----------
1248 * pgstat_reset_counters() -
1249 *
1250 * Tell the statistics collector to reset counters for our database.
1251 *
1252 * Permission checking for this function is managed through the normal
1253 * GRANT system.
1254 * ----------
1255 */
1256 void
pgstat_reset_counters(void)1257 pgstat_reset_counters(void)
1258 {
1259 PgStat_MsgResetcounter msg;
1260
1261 if (pgStatSock == PGINVALID_SOCKET)
1262 return;
1263
1264 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1265 msg.m_databaseid = MyDatabaseId;
1266 pgstat_send(&msg, sizeof(msg));
1267 }
1268
1269 /* ----------
1270 * pgstat_reset_shared_counters() -
1271 *
1272 * Tell the statistics collector to reset cluster-wide shared counters.
1273 *
1274 * Permission checking for this function is managed through the normal
1275 * GRANT system.
1276 * ----------
1277 */
1278 void
pgstat_reset_shared_counters(const char * target)1279 pgstat_reset_shared_counters(const char *target)
1280 {
1281 PgStat_MsgResetsharedcounter msg;
1282
1283 if (pgStatSock == PGINVALID_SOCKET)
1284 return;
1285
1286 if (strcmp(target, "archiver") == 0)
1287 msg.m_resettarget = RESET_ARCHIVER;
1288 else if (strcmp(target, "bgwriter") == 0)
1289 msg.m_resettarget = RESET_BGWRITER;
1290 else
1291 ereport(ERROR,
1292 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1293 errmsg("unrecognized reset target: \"%s\"", target),
1294 errhint("Target must be \"archiver\" or \"bgwriter\".")));
1295
1296 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSHAREDCOUNTER);
1297 pgstat_send(&msg, sizeof(msg));
1298 }
1299
1300 /* ----------
1301 * pgstat_reset_single_counter() -
1302 *
1303 * Tell the statistics collector to reset a single counter.
1304 *
1305 * Permission checking for this function is managed through the normal
1306 * GRANT system.
1307 * ----------
1308 */
1309 void
pgstat_reset_single_counter(Oid objoid,PgStat_Single_Reset_Type type)1310 pgstat_reset_single_counter(Oid objoid, PgStat_Single_Reset_Type type)
1311 {
1312 PgStat_MsgResetsinglecounter msg;
1313
1314 if (pgStatSock == PGINVALID_SOCKET)
1315 return;
1316
1317 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSINGLECOUNTER);
1318 msg.m_databaseid = MyDatabaseId;
1319 msg.m_resettype = type;
1320 msg.m_objectid = objoid;
1321
1322 pgstat_send(&msg, sizeof(msg));
1323 }
1324
1325 /* ----------
1326 * pgstat_report_autovac() -
1327 *
1328 * Called from autovacuum.c to report startup of an autovacuum process.
1329 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1330 * the db OID must be passed in, instead.
1331 * ----------
1332 */
1333 void
pgstat_report_autovac(Oid dboid)1334 pgstat_report_autovac(Oid dboid)
1335 {
1336 PgStat_MsgAutovacStart msg;
1337
1338 if (pgStatSock == PGINVALID_SOCKET)
1339 return;
1340
1341 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1342 msg.m_databaseid = dboid;
1343 msg.m_start_time = GetCurrentTimestamp();
1344
1345 pgstat_send(&msg, sizeof(msg));
1346 }
1347
1348
1349 /* ---------
1350 * pgstat_report_vacuum() -
1351 *
1352 * Tell the collector about the table we just vacuumed.
1353 * ---------
1354 */
1355 void
pgstat_report_vacuum(Oid tableoid,bool shared,PgStat_Counter livetuples,PgStat_Counter deadtuples)1356 pgstat_report_vacuum(Oid tableoid, bool shared,
1357 PgStat_Counter livetuples, PgStat_Counter deadtuples)
1358 {
1359 PgStat_MsgVacuum msg;
1360
1361 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1362 return;
1363
1364 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1365 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1366 msg.m_tableoid = tableoid;
1367 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1368 msg.m_vacuumtime = GetCurrentTimestamp();
1369 msg.m_live_tuples = livetuples;
1370 msg.m_dead_tuples = deadtuples;
1371 pgstat_send(&msg, sizeof(msg));
1372 }
1373
1374 /* --------
1375 * pgstat_report_analyze() -
1376 *
1377 * Tell the collector about the table we just analyzed.
1378 *
1379 * Caller must provide new live- and dead-tuples estimates, as well as a
1380 * flag indicating whether to reset the changes_since_analyze counter.
1381 * --------
1382 */
1383 void
pgstat_report_analyze(Relation rel,PgStat_Counter livetuples,PgStat_Counter deadtuples,bool resetcounter)1384 pgstat_report_analyze(Relation rel,
1385 PgStat_Counter livetuples, PgStat_Counter deadtuples,
1386 bool resetcounter)
1387 {
1388 PgStat_MsgAnalyze msg;
1389
1390 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1391 return;
1392
1393 /*
1394 * Unlike VACUUM, ANALYZE might be running inside a transaction that has
1395 * already inserted and/or deleted rows in the target table. ANALYZE will
1396 * have counted such rows as live or dead respectively. Because we will
1397 * report our counts of such rows at transaction end, we should subtract
1398 * off these counts from what we send to the collector now, else they'll
1399 * be double-counted after commit. (This approach also ensures that the
1400 * collector ends up with the right numbers if we abort instead of
1401 * committing.)
1402 */
1403 if (rel->pgstat_info != NULL)
1404 {
1405 PgStat_TableXactStatus *trans;
1406
1407 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1408 {
1409 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1410 deadtuples -= trans->tuples_updated + trans->tuples_deleted;
1411 }
1412 /* count stuff inserted by already-aborted subxacts, too */
1413 deadtuples -= rel->pgstat_info->t_counts.t_delta_dead_tuples;
1414 /* Since ANALYZE's counts are estimates, we could have underflowed */
1415 livetuples = Max(livetuples, 0);
1416 deadtuples = Max(deadtuples, 0);
1417 }
1418
1419 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1420 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1421 msg.m_tableoid = RelationGetRelid(rel);
1422 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1423 msg.m_resetcounter = resetcounter;
1424 msg.m_analyzetime = GetCurrentTimestamp();
1425 msg.m_live_tuples = livetuples;
1426 msg.m_dead_tuples = deadtuples;
1427 pgstat_send(&msg, sizeof(msg));
1428 }
1429
1430 /* --------
1431 * pgstat_report_recovery_conflict() -
1432 *
1433 * Tell the collector about a Hot Standby recovery conflict.
1434 * --------
1435 */
1436 void
pgstat_report_recovery_conflict(int reason)1437 pgstat_report_recovery_conflict(int reason)
1438 {
1439 PgStat_MsgRecoveryConflict msg;
1440
1441 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1442 return;
1443
1444 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RECOVERYCONFLICT);
1445 msg.m_databaseid = MyDatabaseId;
1446 msg.m_reason = reason;
1447 pgstat_send(&msg, sizeof(msg));
1448 }
1449
1450 /* --------
1451 * pgstat_report_deadlock() -
1452 *
1453 * Tell the collector about a deadlock detected.
1454 * --------
1455 */
1456 void
pgstat_report_deadlock(void)1457 pgstat_report_deadlock(void)
1458 {
1459 PgStat_MsgDeadlock msg;
1460
1461 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1462 return;
1463
1464 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DEADLOCK);
1465 msg.m_databaseid = MyDatabaseId;
1466 pgstat_send(&msg, sizeof(msg));
1467 }
1468
1469 /* --------
1470 * pgstat_report_tempfile() -
1471 *
1472 * Tell the collector about a temporary file.
1473 * --------
1474 */
1475 void
pgstat_report_tempfile(size_t filesize)1476 pgstat_report_tempfile(size_t filesize)
1477 {
1478 PgStat_MsgTempFile msg;
1479
1480 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1481 return;
1482
1483 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TEMPFILE);
1484 msg.m_databaseid = MyDatabaseId;
1485 msg.m_filesize = filesize;
1486 pgstat_send(&msg, sizeof(msg));
1487 }
1488
1489
1490 /* ----------
1491 * pgstat_ping() -
1492 *
1493 * Send some junk data to the collector to increase traffic.
1494 * ----------
1495 */
1496 void
pgstat_ping(void)1497 pgstat_ping(void)
1498 {
1499 PgStat_MsgDummy msg;
1500
1501 if (pgStatSock == PGINVALID_SOCKET)
1502 return;
1503
1504 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1505 pgstat_send(&msg, sizeof(msg));
1506 }
1507
1508 /* ----------
1509 * pgstat_send_inquiry() -
1510 *
1511 * Notify collector that we need fresh data.
1512 * ----------
1513 */
1514 static void
pgstat_send_inquiry(TimestampTz clock_time,TimestampTz cutoff_time,Oid databaseid)1515 pgstat_send_inquiry(TimestampTz clock_time, TimestampTz cutoff_time, Oid databaseid)
1516 {
1517 PgStat_MsgInquiry msg;
1518
1519 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1520 msg.clock_time = clock_time;
1521 msg.cutoff_time = cutoff_time;
1522 msg.databaseid = databaseid;
1523 pgstat_send(&msg, sizeof(msg));
1524 }
1525
1526
1527 /*
1528 * Initialize function call usage data.
1529 * Called by the executor before invoking a function.
1530 */
1531 void
pgstat_init_function_usage(FunctionCallInfoData * fcinfo,PgStat_FunctionCallUsage * fcu)1532 pgstat_init_function_usage(FunctionCallInfoData *fcinfo,
1533 PgStat_FunctionCallUsage *fcu)
1534 {
1535 PgStat_BackendFunctionEntry *htabent;
1536 bool found;
1537
1538 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1539 {
1540 /* stats not wanted */
1541 fcu->fs = NULL;
1542 return;
1543 }
1544
1545 if (!pgStatFunctions)
1546 {
1547 /* First time through - initialize function stat table */
1548 HASHCTL hash_ctl;
1549
1550 memset(&hash_ctl, 0, sizeof(hash_ctl));
1551 hash_ctl.keysize = sizeof(Oid);
1552 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1553 pgStatFunctions = hash_create("Function stat entries",
1554 PGSTAT_FUNCTION_HASH_SIZE,
1555 &hash_ctl,
1556 HASH_ELEM | HASH_BLOBS);
1557 }
1558
1559 /* Get the stats entry for this function, create if necessary */
1560 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1561 HASH_ENTER, &found);
1562 if (!found)
1563 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1564
1565 fcu->fs = &htabent->f_counts;
1566
1567 /* save stats for this function, later used to compensate for recursion */
1568 fcu->save_f_total_time = htabent->f_counts.f_total_time;
1569
1570 /* save current backend-wide total time */
1571 fcu->save_total = total_func_time;
1572
1573 /* get clock time as of function start */
1574 INSTR_TIME_SET_CURRENT(fcu->f_start);
1575 }
1576
1577 /*
1578 * find_funcstat_entry - find any existing PgStat_BackendFunctionEntry entry
1579 * for specified function
1580 *
1581 * If no entry, return NULL, don't create a new one
1582 */
1583 PgStat_BackendFunctionEntry *
find_funcstat_entry(Oid func_id)1584 find_funcstat_entry(Oid func_id)
1585 {
1586 if (pgStatFunctions == NULL)
1587 return NULL;
1588
1589 return (PgStat_BackendFunctionEntry *) hash_search(pgStatFunctions,
1590 (void *) &func_id,
1591 HASH_FIND, NULL);
1592 }
1593
1594 /*
1595 * Calculate function call usage and update stat counters.
1596 * Called by the executor after invoking a function.
1597 *
1598 * In the case of a set-returning function that runs in value-per-call mode,
1599 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1600 * calls for what the user considers a single call of the function. The
1601 * finalize flag should be TRUE on the last call.
1602 */
1603 void
pgstat_end_function_usage(PgStat_FunctionCallUsage * fcu,bool finalize)1604 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1605 {
1606 PgStat_FunctionCounts *fs = fcu->fs;
1607 instr_time f_total;
1608 instr_time f_others;
1609 instr_time f_self;
1610
1611 /* stats not wanted? */
1612 if (fs == NULL)
1613 return;
1614
1615 /* total elapsed time in this function call */
1616 INSTR_TIME_SET_CURRENT(f_total);
1617 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1618
1619 /* self usage: elapsed minus anything already charged to other calls */
1620 f_others = total_func_time;
1621 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1622 f_self = f_total;
1623 INSTR_TIME_SUBTRACT(f_self, f_others);
1624
1625 /* update backend-wide total time */
1626 INSTR_TIME_ADD(total_func_time, f_self);
1627
1628 /*
1629 * Compute the new f_total_time as the total elapsed time added to the
1630 * pre-call value of f_total_time. This is necessary to avoid
1631 * double-counting any time taken by recursive calls of myself. (We do
1632 * not need any similar kluge for self time, since that already excludes
1633 * any recursive calls.)
1634 */
1635 INSTR_TIME_ADD(f_total, fcu->save_f_total_time);
1636
1637 /* update counters in function stats table */
1638 if (finalize)
1639 fs->f_numcalls++;
1640 fs->f_total_time = f_total;
1641 INSTR_TIME_ADD(fs->f_self_time, f_self);
1642
1643 /* indicate that we have something to send */
1644 have_function_stats = true;
1645 }
1646
1647
1648 /* ----------
1649 * pgstat_initstats() -
1650 *
1651 * Initialize a relcache entry to count access statistics.
1652 * Called whenever a relation is opened.
1653 *
1654 * We assume that a relcache entry's pgstat_info field is zeroed by
1655 * relcache.c when the relcache entry is made; thereafter it is long-lived
1656 * data. We can avoid repeated searches of the TabStatus arrays when the
1657 * same relation is touched repeatedly within a transaction.
1658 * ----------
1659 */
1660 void
pgstat_initstats(Relation rel)1661 pgstat_initstats(Relation rel)
1662 {
1663 Oid rel_id = rel->rd_id;
1664 char relkind = rel->rd_rel->relkind;
1665
1666 /* We only count stats for things that have storage */
1667 if (!(relkind == RELKIND_RELATION ||
1668 relkind == RELKIND_MATVIEW ||
1669 relkind == RELKIND_INDEX ||
1670 relkind == RELKIND_TOASTVALUE ||
1671 relkind == RELKIND_SEQUENCE))
1672 {
1673 rel->pgstat_info = NULL;
1674 return;
1675 }
1676
1677 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1678 {
1679 /* We're not counting at all */
1680 rel->pgstat_info = NULL;
1681 return;
1682 }
1683
1684 /*
1685 * If we already set up this relation in the current transaction, nothing
1686 * to do.
1687 */
1688 if (rel->pgstat_info != NULL &&
1689 rel->pgstat_info->t_id == rel_id)
1690 return;
1691
1692 /* Else find or make the PgStat_TableStatus entry, and update link */
1693 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1694 }
1695
1696 /*
1697 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1698 */
1699 static PgStat_TableStatus *
get_tabstat_entry(Oid rel_id,bool isshared)1700 get_tabstat_entry(Oid rel_id, bool isshared)
1701 {
1702 PgStat_TableStatus *entry;
1703 TabStatusArray *tsa;
1704 TabStatusArray *prev_tsa;
1705 int i;
1706
1707 /*
1708 * Search the already-used tabstat slots for this relation.
1709 */
1710 prev_tsa = NULL;
1711 for (tsa = pgStatTabList; tsa != NULL; prev_tsa = tsa, tsa = tsa->tsa_next)
1712 {
1713 for (i = 0; i < tsa->tsa_used; i++)
1714 {
1715 entry = &tsa->tsa_entries[i];
1716 if (entry->t_id == rel_id)
1717 return entry;
1718 }
1719
1720 if (tsa->tsa_used < TABSTAT_QUANTUM)
1721 {
1722 /*
1723 * It must not be present, but we found a free slot instead. Fine,
1724 * let's use this one. We assume the entry was already zeroed,
1725 * either at creation or after last use.
1726 */
1727 entry = &tsa->tsa_entries[tsa->tsa_used++];
1728 entry->t_id = rel_id;
1729 entry->t_shared = isshared;
1730 return entry;
1731 }
1732 }
1733
1734 /*
1735 * We ran out of tabstat slots, so allocate more. Be sure they're zeroed.
1736 */
1737 tsa = (TabStatusArray *) MemoryContextAllocZero(TopMemoryContext,
1738 sizeof(TabStatusArray));
1739 if (prev_tsa)
1740 prev_tsa->tsa_next = tsa;
1741 else
1742 pgStatTabList = tsa;
1743
1744 /*
1745 * Use the first entry of the new TabStatusArray.
1746 */
1747 entry = &tsa->tsa_entries[tsa->tsa_used++];
1748 entry->t_id = rel_id;
1749 entry->t_shared = isshared;
1750 return entry;
1751 }
1752
1753 /*
1754 * find_tabstat_entry - find any existing PgStat_TableStatus entry for rel
1755 *
1756 * If no entry, return NULL, don't create a new one
1757 */
1758 PgStat_TableStatus *
find_tabstat_entry(Oid rel_id)1759 find_tabstat_entry(Oid rel_id)
1760 {
1761 PgStat_TableStatus *entry;
1762 TabStatusArray *tsa;
1763 int i;
1764
1765 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
1766 {
1767 for (i = 0; i < tsa->tsa_used; i++)
1768 {
1769 entry = &tsa->tsa_entries[i];
1770 if (entry->t_id == rel_id)
1771 return entry;
1772 }
1773 }
1774
1775 /* Not present */
1776 return NULL;
1777 }
1778
1779 /*
1780 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1781 */
1782 static PgStat_SubXactStatus *
get_tabstat_stack_level(int nest_level)1783 get_tabstat_stack_level(int nest_level)
1784 {
1785 PgStat_SubXactStatus *xact_state;
1786
1787 xact_state = pgStatXactStack;
1788 if (xact_state == NULL || xact_state->nest_level != nest_level)
1789 {
1790 xact_state = (PgStat_SubXactStatus *)
1791 MemoryContextAlloc(TopTransactionContext,
1792 sizeof(PgStat_SubXactStatus));
1793 xact_state->nest_level = nest_level;
1794 xact_state->prev = pgStatXactStack;
1795 xact_state->first = NULL;
1796 pgStatXactStack = xact_state;
1797 }
1798 return xact_state;
1799 }
1800
1801 /*
1802 * add_tabstat_xact_level - add a new (sub)transaction state record
1803 */
1804 static void
add_tabstat_xact_level(PgStat_TableStatus * pgstat_info,int nest_level)1805 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1806 {
1807 PgStat_SubXactStatus *xact_state;
1808 PgStat_TableXactStatus *trans;
1809
1810 /*
1811 * If this is the first rel to be modified at the current nest level, we
1812 * first have to push a transaction stack entry.
1813 */
1814 xact_state = get_tabstat_stack_level(nest_level);
1815
1816 /* Now make a per-table stack entry */
1817 trans = (PgStat_TableXactStatus *)
1818 MemoryContextAllocZero(TopTransactionContext,
1819 sizeof(PgStat_TableXactStatus));
1820 trans->nest_level = nest_level;
1821 trans->upper = pgstat_info->trans;
1822 trans->parent = pgstat_info;
1823 trans->next = xact_state->first;
1824 xact_state->first = trans;
1825 pgstat_info->trans = trans;
1826 }
1827
1828 /*
1829 * pgstat_count_heap_insert - count a tuple insertion of n tuples
1830 */
1831 void
pgstat_count_heap_insert(Relation rel,int n)1832 pgstat_count_heap_insert(Relation rel, int n)
1833 {
1834 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1835
1836 if (pgstat_info != NULL)
1837 {
1838 /* We have to log the effect at the proper transactional level */
1839 int nest_level = GetCurrentTransactionNestLevel();
1840
1841 if (pgstat_info->trans == NULL ||
1842 pgstat_info->trans->nest_level != nest_level)
1843 add_tabstat_xact_level(pgstat_info, nest_level);
1844
1845 pgstat_info->trans->tuples_inserted += n;
1846 }
1847 }
1848
1849 /*
1850 * pgstat_count_heap_update - count a tuple update
1851 */
1852 void
pgstat_count_heap_update(Relation rel,bool hot)1853 pgstat_count_heap_update(Relation rel, bool hot)
1854 {
1855 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1856
1857 if (pgstat_info != NULL)
1858 {
1859 /* We have to log the effect at the proper transactional level */
1860 int nest_level = GetCurrentTransactionNestLevel();
1861
1862 if (pgstat_info->trans == NULL ||
1863 pgstat_info->trans->nest_level != nest_level)
1864 add_tabstat_xact_level(pgstat_info, nest_level);
1865
1866 pgstat_info->trans->tuples_updated++;
1867
1868 /* t_tuples_hot_updated is nontransactional, so just advance it */
1869 if (hot)
1870 pgstat_info->t_counts.t_tuples_hot_updated++;
1871 }
1872 }
1873
1874 /*
1875 * pgstat_count_heap_delete - count a tuple deletion
1876 */
1877 void
pgstat_count_heap_delete(Relation rel)1878 pgstat_count_heap_delete(Relation rel)
1879 {
1880 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1881
1882 if (pgstat_info != NULL)
1883 {
1884 /* We have to log the effect at the proper transactional level */
1885 int nest_level = GetCurrentTransactionNestLevel();
1886
1887 if (pgstat_info->trans == NULL ||
1888 pgstat_info->trans->nest_level != nest_level)
1889 add_tabstat_xact_level(pgstat_info, nest_level);
1890
1891 pgstat_info->trans->tuples_deleted++;
1892 }
1893 }
1894
1895 /*
1896 * pgstat_truncate_save_counters
1897 *
1898 * Whenever a table is truncated, we save its i/u/d counters so that they can
1899 * be cleared, and if the (sub)xact that executed the truncate later aborts,
1900 * the counters can be restored to the saved (pre-truncate) values. Note we do
1901 * this on the first truncate in any particular subxact level only.
1902 */
1903 static void
pgstat_truncate_save_counters(PgStat_TableXactStatus * trans)1904 pgstat_truncate_save_counters(PgStat_TableXactStatus *trans)
1905 {
1906 if (!trans->truncated)
1907 {
1908 trans->inserted_pre_trunc = trans->tuples_inserted;
1909 trans->updated_pre_trunc = trans->tuples_updated;
1910 trans->deleted_pre_trunc = trans->tuples_deleted;
1911 trans->truncated = true;
1912 }
1913 }
1914
1915 /*
1916 * pgstat_truncate_restore_counters - restore counters when a truncate aborts
1917 */
1918 static void
pgstat_truncate_restore_counters(PgStat_TableXactStatus * trans)1919 pgstat_truncate_restore_counters(PgStat_TableXactStatus *trans)
1920 {
1921 if (trans->truncated)
1922 {
1923 trans->tuples_inserted = trans->inserted_pre_trunc;
1924 trans->tuples_updated = trans->updated_pre_trunc;
1925 trans->tuples_deleted = trans->deleted_pre_trunc;
1926 }
1927 }
1928
1929 /*
1930 * pgstat_count_truncate - update tuple counters due to truncate
1931 */
1932 void
pgstat_count_truncate(Relation rel)1933 pgstat_count_truncate(Relation rel)
1934 {
1935 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1936
1937 if (pgstat_info != NULL)
1938 {
1939 /* We have to log the effect at the proper transactional level */
1940 int nest_level = GetCurrentTransactionNestLevel();
1941
1942 if (pgstat_info->trans == NULL ||
1943 pgstat_info->trans->nest_level != nest_level)
1944 add_tabstat_xact_level(pgstat_info, nest_level);
1945
1946 pgstat_truncate_save_counters(pgstat_info->trans);
1947 pgstat_info->trans->tuples_inserted = 0;
1948 pgstat_info->trans->tuples_updated = 0;
1949 pgstat_info->trans->tuples_deleted = 0;
1950 }
1951 }
1952
1953 /*
1954 * pgstat_update_heap_dead_tuples - update dead-tuples count
1955 *
1956 * The semantics of this are that we are reporting the nontransactional
1957 * recovery of "delta" dead tuples; so t_delta_dead_tuples decreases
1958 * rather than increasing, and the change goes straight into the per-table
1959 * counter, not into transactional state.
1960 */
1961 void
pgstat_update_heap_dead_tuples(Relation rel,int delta)1962 pgstat_update_heap_dead_tuples(Relation rel, int delta)
1963 {
1964 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1965
1966 if (pgstat_info != NULL)
1967 pgstat_info->t_counts.t_delta_dead_tuples -= delta;
1968 }
1969
1970
1971 /* ----------
1972 * AtEOXact_PgStat
1973 *
1974 * Called from access/transam/xact.c at top-level transaction commit/abort.
1975 * ----------
1976 */
1977 void
AtEOXact_PgStat(bool isCommit,bool parallel)1978 AtEOXact_PgStat(bool isCommit, bool parallel)
1979 {
1980 PgStat_SubXactStatus *xact_state;
1981
1982 /* Don't count parallel worker transaction stats */
1983 if (!parallel)
1984 {
1985 /*
1986 * Count transaction commit or abort. (We use counters, not just
1987 * bools, in case the reporting message isn't sent right away.)
1988 */
1989 if (isCommit)
1990 pgStatXactCommit++;
1991 else
1992 pgStatXactRollback++;
1993 }
1994
1995 /*
1996 * Transfer transactional insert/update counts into the base tabstat
1997 * entries. We don't bother to free any of the transactional state, since
1998 * it's all in TopTransactionContext and will go away anyway.
1999 */
2000 xact_state = pgStatXactStack;
2001 if (xact_state != NULL)
2002 {
2003 PgStat_TableXactStatus *trans;
2004
2005 Assert(xact_state->nest_level == 1);
2006 Assert(xact_state->prev == NULL);
2007 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2008 {
2009 PgStat_TableStatus *tabstat;
2010
2011 Assert(trans->nest_level == 1);
2012 Assert(trans->upper == NULL);
2013 tabstat = trans->parent;
2014 Assert(tabstat->trans == trans);
2015 /* restore pre-truncate stats (if any) in case of aborted xact */
2016 if (!isCommit)
2017 pgstat_truncate_restore_counters(trans);
2018 /* count attempted actions regardless of commit/abort */
2019 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
2020 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
2021 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
2022 if (isCommit)
2023 {
2024 tabstat->t_counts.t_truncated = trans->truncated;
2025 if (trans->truncated)
2026 {
2027 /* forget live/dead stats seen by backend thus far */
2028 tabstat->t_counts.t_delta_live_tuples = 0;
2029 tabstat->t_counts.t_delta_dead_tuples = 0;
2030 }
2031 /* insert adds a live tuple, delete removes one */
2032 tabstat->t_counts.t_delta_live_tuples +=
2033 trans->tuples_inserted - trans->tuples_deleted;
2034 /* update and delete each create a dead tuple */
2035 tabstat->t_counts.t_delta_dead_tuples +=
2036 trans->tuples_updated + trans->tuples_deleted;
2037 /* insert, update, delete each count as one change event */
2038 tabstat->t_counts.t_changed_tuples +=
2039 trans->tuples_inserted + trans->tuples_updated +
2040 trans->tuples_deleted;
2041 }
2042 else
2043 {
2044 /* inserted tuples are dead, deleted tuples are unaffected */
2045 tabstat->t_counts.t_delta_dead_tuples +=
2046 trans->tuples_inserted + trans->tuples_updated;
2047 /* an aborted xact generates no changed_tuple events */
2048 }
2049 tabstat->trans = NULL;
2050 }
2051 }
2052 pgStatXactStack = NULL;
2053
2054 /* Make sure any stats snapshot is thrown away */
2055 pgstat_clear_snapshot();
2056 }
2057
2058 /* ----------
2059 * AtEOSubXact_PgStat
2060 *
2061 * Called from access/transam/xact.c at subtransaction commit/abort.
2062 * ----------
2063 */
2064 void
AtEOSubXact_PgStat(bool isCommit,int nestDepth)2065 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
2066 {
2067 PgStat_SubXactStatus *xact_state;
2068
2069 /*
2070 * Transfer transactional insert/update counts into the next higher
2071 * subtransaction state.
2072 */
2073 xact_state = pgStatXactStack;
2074 if (xact_state != NULL &&
2075 xact_state->nest_level >= nestDepth)
2076 {
2077 PgStat_TableXactStatus *trans;
2078 PgStat_TableXactStatus *next_trans;
2079
2080 /* delink xact_state from stack immediately to simplify reuse case */
2081 pgStatXactStack = xact_state->prev;
2082
2083 for (trans = xact_state->first; trans != NULL; trans = next_trans)
2084 {
2085 PgStat_TableStatus *tabstat;
2086
2087 next_trans = trans->next;
2088 Assert(trans->nest_level == nestDepth);
2089 tabstat = trans->parent;
2090 Assert(tabstat->trans == trans);
2091 if (isCommit)
2092 {
2093 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
2094 {
2095 if (trans->truncated)
2096 {
2097 /* propagate the truncate status one level up */
2098 pgstat_truncate_save_counters(trans->upper);
2099 /* replace upper xact stats with ours */
2100 trans->upper->tuples_inserted = trans->tuples_inserted;
2101 trans->upper->tuples_updated = trans->tuples_updated;
2102 trans->upper->tuples_deleted = trans->tuples_deleted;
2103 }
2104 else
2105 {
2106 trans->upper->tuples_inserted += trans->tuples_inserted;
2107 trans->upper->tuples_updated += trans->tuples_updated;
2108 trans->upper->tuples_deleted += trans->tuples_deleted;
2109 }
2110 tabstat->trans = trans->upper;
2111 pfree(trans);
2112 }
2113 else
2114 {
2115 /*
2116 * When there isn't an immediate parent state, we can just
2117 * reuse the record instead of going through a
2118 * palloc/pfree pushup (this works since it's all in
2119 * TopTransactionContext anyway). We have to re-link it
2120 * into the parent level, though, and that might mean
2121 * pushing a new entry into the pgStatXactStack.
2122 */
2123 PgStat_SubXactStatus *upper_xact_state;
2124
2125 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
2126 trans->next = upper_xact_state->first;
2127 upper_xact_state->first = trans;
2128 trans->nest_level = nestDepth - 1;
2129 }
2130 }
2131 else
2132 {
2133 /*
2134 * On abort, update top-level tabstat counts, then forget the
2135 * subtransaction
2136 */
2137
2138 /* first restore values obliterated by truncate */
2139 pgstat_truncate_restore_counters(trans);
2140 /* count attempted actions regardless of commit/abort */
2141 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
2142 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
2143 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
2144 /* inserted tuples are dead, deleted tuples are unaffected */
2145 tabstat->t_counts.t_delta_dead_tuples +=
2146 trans->tuples_inserted + trans->tuples_updated;
2147 tabstat->trans = trans->upper;
2148 pfree(trans);
2149 }
2150 }
2151 pfree(xact_state);
2152 }
2153 }
2154
2155
2156 /*
2157 * AtPrepare_PgStat
2158 * Save the transactional stats state at 2PC transaction prepare.
2159 *
2160 * In this phase we just generate 2PC records for all the pending
2161 * transaction-dependent stats work.
2162 */
2163 void
AtPrepare_PgStat(void)2164 AtPrepare_PgStat(void)
2165 {
2166 PgStat_SubXactStatus *xact_state;
2167
2168 xact_state = pgStatXactStack;
2169 if (xact_state != NULL)
2170 {
2171 PgStat_TableXactStatus *trans;
2172
2173 Assert(xact_state->nest_level == 1);
2174 Assert(xact_state->prev == NULL);
2175 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2176 {
2177 PgStat_TableStatus *tabstat;
2178 TwoPhasePgStatRecord record;
2179
2180 Assert(trans->nest_level == 1);
2181 Assert(trans->upper == NULL);
2182 tabstat = trans->parent;
2183 Assert(tabstat->trans == trans);
2184
2185 record.tuples_inserted = trans->tuples_inserted;
2186 record.tuples_updated = trans->tuples_updated;
2187 record.tuples_deleted = trans->tuples_deleted;
2188 record.inserted_pre_trunc = trans->inserted_pre_trunc;
2189 record.updated_pre_trunc = trans->updated_pre_trunc;
2190 record.deleted_pre_trunc = trans->deleted_pre_trunc;
2191 record.t_id = tabstat->t_id;
2192 record.t_shared = tabstat->t_shared;
2193 record.t_truncated = trans->truncated;
2194
2195 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
2196 &record, sizeof(TwoPhasePgStatRecord));
2197 }
2198 }
2199 }
2200
2201 /*
2202 * PostPrepare_PgStat
2203 * Clean up after successful PREPARE.
2204 *
2205 * All we need do here is unlink the transaction stats state from the
2206 * nontransactional state. The nontransactional action counts will be
2207 * reported to the stats collector immediately, while the effects on live
2208 * and dead tuple counts are preserved in the 2PC state file.
2209 *
2210 * Note: AtEOXact_PgStat is not called during PREPARE.
2211 */
2212 void
PostPrepare_PgStat(void)2213 PostPrepare_PgStat(void)
2214 {
2215 PgStat_SubXactStatus *xact_state;
2216
2217 /*
2218 * We don't bother to free any of the transactional state, since it's all
2219 * in TopTransactionContext and will go away anyway.
2220 */
2221 xact_state = pgStatXactStack;
2222 if (xact_state != NULL)
2223 {
2224 PgStat_TableXactStatus *trans;
2225
2226 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2227 {
2228 PgStat_TableStatus *tabstat;
2229
2230 tabstat = trans->parent;
2231 tabstat->trans = NULL;
2232 }
2233 }
2234 pgStatXactStack = NULL;
2235
2236 /* Make sure any stats snapshot is thrown away */
2237 pgstat_clear_snapshot();
2238 }
2239
2240 /*
2241 * 2PC processing routine for COMMIT PREPARED case.
2242 *
2243 * Load the saved counts into our local pgstats state.
2244 */
2245 void
pgstat_twophase_postcommit(TransactionId xid,uint16 info,void * recdata,uint32 len)2246 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
2247 void *recdata, uint32 len)
2248 {
2249 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2250 PgStat_TableStatus *pgstat_info;
2251
2252 /* Find or create a tabstat entry for the rel */
2253 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2254
2255 /* Same math as in AtEOXact_PgStat, commit case */
2256 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2257 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2258 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2259 pgstat_info->t_counts.t_truncated = rec->t_truncated;
2260 if (rec->t_truncated)
2261 {
2262 /* forget live/dead stats seen by backend thus far */
2263 pgstat_info->t_counts.t_delta_live_tuples = 0;
2264 pgstat_info->t_counts.t_delta_dead_tuples = 0;
2265 }
2266 pgstat_info->t_counts.t_delta_live_tuples +=
2267 rec->tuples_inserted - rec->tuples_deleted;
2268 pgstat_info->t_counts.t_delta_dead_tuples +=
2269 rec->tuples_updated + rec->tuples_deleted;
2270 pgstat_info->t_counts.t_changed_tuples +=
2271 rec->tuples_inserted + rec->tuples_updated +
2272 rec->tuples_deleted;
2273 }
2274
2275 /*
2276 * 2PC processing routine for ROLLBACK PREPARED case.
2277 *
2278 * Load the saved counts into our local pgstats state, but treat them
2279 * as aborted.
2280 */
2281 void
pgstat_twophase_postabort(TransactionId xid,uint16 info,void * recdata,uint32 len)2282 pgstat_twophase_postabort(TransactionId xid, uint16 info,
2283 void *recdata, uint32 len)
2284 {
2285 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2286 PgStat_TableStatus *pgstat_info;
2287
2288 /* Find or create a tabstat entry for the rel */
2289 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2290
2291 /* Same math as in AtEOXact_PgStat, abort case */
2292 if (rec->t_truncated)
2293 {
2294 rec->tuples_inserted = rec->inserted_pre_trunc;
2295 rec->tuples_updated = rec->updated_pre_trunc;
2296 rec->tuples_deleted = rec->deleted_pre_trunc;
2297 }
2298 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2299 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2300 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2301 pgstat_info->t_counts.t_delta_dead_tuples +=
2302 rec->tuples_inserted + rec->tuples_updated;
2303 }
2304
2305
2306 /* ----------
2307 * pgstat_fetch_stat_dbentry() -
2308 *
2309 * Support function for the SQL-callable pgstat* functions. Returns
2310 * the collected statistics for one database or NULL. NULL doesn't mean
2311 * that the database doesn't exist, it is just not yet known by the
2312 * collector, so the caller is better off to report ZERO instead.
2313 * ----------
2314 */
2315 PgStat_StatDBEntry *
pgstat_fetch_stat_dbentry(Oid dbid)2316 pgstat_fetch_stat_dbentry(Oid dbid)
2317 {
2318 /*
2319 * If not done for this transaction, read the statistics collector stats
2320 * file into some hash tables.
2321 */
2322 backend_read_statsfile();
2323
2324 /*
2325 * Lookup the requested database; return NULL if not found
2326 */
2327 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2328 (void *) &dbid,
2329 HASH_FIND, NULL);
2330 }
2331
2332
2333 /* ----------
2334 * pgstat_fetch_stat_tabentry() -
2335 *
2336 * Support function for the SQL-callable pgstat* functions. Returns
2337 * the collected statistics for one table or NULL. NULL doesn't mean
2338 * that the table doesn't exist, it is just not yet known by the
2339 * collector, so the caller is better off to report ZERO instead.
2340 * ----------
2341 */
2342 PgStat_StatTabEntry *
pgstat_fetch_stat_tabentry(Oid relid)2343 pgstat_fetch_stat_tabentry(Oid relid)
2344 {
2345 Oid dbid;
2346 PgStat_StatDBEntry *dbentry;
2347 PgStat_StatTabEntry *tabentry;
2348
2349 /*
2350 * If not done for this transaction, read the statistics collector stats
2351 * file into some hash tables.
2352 */
2353 backend_read_statsfile();
2354
2355 /*
2356 * Lookup our database, then look in its table hash table.
2357 */
2358 dbid = MyDatabaseId;
2359 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2360 (void *) &dbid,
2361 HASH_FIND, NULL);
2362 if (dbentry != NULL && dbentry->tables != NULL)
2363 {
2364 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2365 (void *) &relid,
2366 HASH_FIND, NULL);
2367 if (tabentry)
2368 return tabentry;
2369 }
2370
2371 /*
2372 * If we didn't find it, maybe it's a shared table.
2373 */
2374 dbid = InvalidOid;
2375 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2376 (void *) &dbid,
2377 HASH_FIND, NULL);
2378 if (dbentry != NULL && dbentry->tables != NULL)
2379 {
2380 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2381 (void *) &relid,
2382 HASH_FIND, NULL);
2383 if (tabentry)
2384 return tabentry;
2385 }
2386
2387 return NULL;
2388 }
2389
2390
2391 /* ----------
2392 * pgstat_fetch_stat_funcentry() -
2393 *
2394 * Support function for the SQL-callable pgstat* functions. Returns
2395 * the collected statistics for one function or NULL.
2396 * ----------
2397 */
2398 PgStat_StatFuncEntry *
pgstat_fetch_stat_funcentry(Oid func_id)2399 pgstat_fetch_stat_funcentry(Oid func_id)
2400 {
2401 PgStat_StatDBEntry *dbentry;
2402 PgStat_StatFuncEntry *funcentry = NULL;
2403
2404 /* load the stats file if needed */
2405 backend_read_statsfile();
2406
2407 /* Lookup our database, then find the requested function. */
2408 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
2409 if (dbentry != NULL && dbentry->functions != NULL)
2410 {
2411 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2412 (void *) &func_id,
2413 HASH_FIND, NULL);
2414 }
2415
2416 return funcentry;
2417 }
2418
2419
2420 /* ----------
2421 * pgstat_fetch_stat_beentry() -
2422 *
2423 * Support function for the SQL-callable pgstat* functions. Returns
2424 * our local copy of the current-activity entry for one backend.
2425 *
2426 * NB: caller is responsible for a check if the user is permitted to see
2427 * this info (especially the querystring).
2428 * ----------
2429 */
2430 PgBackendStatus *
pgstat_fetch_stat_beentry(int beid)2431 pgstat_fetch_stat_beentry(int beid)
2432 {
2433 pgstat_read_current_status();
2434
2435 if (beid < 1 || beid > localNumBackends)
2436 return NULL;
2437
2438 return &localBackendStatusTable[beid - 1].backendStatus;
2439 }
2440
2441
2442 /* ----------
2443 * pgstat_fetch_stat_local_beentry() -
2444 *
2445 * Like pgstat_fetch_stat_beentry() but with locally computed additions (like
2446 * xid and xmin values of the backend)
2447 *
2448 * NB: caller is responsible for a check if the user is permitted to see
2449 * this info (especially the querystring).
2450 * ----------
2451 */
2452 LocalPgBackendStatus *
pgstat_fetch_stat_local_beentry(int beid)2453 pgstat_fetch_stat_local_beentry(int beid)
2454 {
2455 pgstat_read_current_status();
2456
2457 if (beid < 1 || beid > localNumBackends)
2458 return NULL;
2459
2460 return &localBackendStatusTable[beid - 1];
2461 }
2462
2463
2464 /* ----------
2465 * pgstat_fetch_stat_numbackends() -
2466 *
2467 * Support function for the SQL-callable pgstat* functions. Returns
2468 * the maximum current backend id.
2469 * ----------
2470 */
2471 int
pgstat_fetch_stat_numbackends(void)2472 pgstat_fetch_stat_numbackends(void)
2473 {
2474 pgstat_read_current_status();
2475
2476 return localNumBackends;
2477 }
2478
2479 /*
2480 * ---------
2481 * pgstat_fetch_stat_archiver() -
2482 *
2483 * Support function for the SQL-callable pgstat* functions. Returns
2484 * a pointer to the archiver statistics struct.
2485 * ---------
2486 */
2487 PgStat_ArchiverStats *
pgstat_fetch_stat_archiver(void)2488 pgstat_fetch_stat_archiver(void)
2489 {
2490 backend_read_statsfile();
2491
2492 return &archiverStats;
2493 }
2494
2495
2496 /*
2497 * ---------
2498 * pgstat_fetch_global() -
2499 *
2500 * Support function for the SQL-callable pgstat* functions. Returns
2501 * a pointer to the global statistics struct.
2502 * ---------
2503 */
2504 PgStat_GlobalStats *
pgstat_fetch_global(void)2505 pgstat_fetch_global(void)
2506 {
2507 backend_read_statsfile();
2508
2509 return &globalStats;
2510 }
2511
2512
2513 /* ------------------------------------------------------------
2514 * Functions for management of the shared-memory PgBackendStatus array
2515 * ------------------------------------------------------------
2516 */
2517
2518 static PgBackendStatus *BackendStatusArray = NULL;
2519 static PgBackendStatus *MyBEEntry = NULL;
2520 static char *BackendAppnameBuffer = NULL;
2521 static char *BackendClientHostnameBuffer = NULL;
2522 static char *BackendActivityBuffer = NULL;
2523 static Size BackendActivityBufferSize = 0;
2524 #ifdef USE_SSL
2525 static PgBackendSSLStatus *BackendSslStatusBuffer = NULL;
2526 #endif
2527
2528
2529 /*
2530 * Report shared-memory space needed by CreateSharedBackendStatus.
2531 */
2532 Size
BackendStatusShmemSize(void)2533 BackendStatusShmemSize(void)
2534 {
2535 Size size;
2536
2537 /* BackendStatusArray: */
2538 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2539 /* BackendAppnameBuffer: */
2540 size = add_size(size,
2541 mul_size(NAMEDATALEN, MaxBackends));
2542 /* BackendClientHostnameBuffer: */
2543 size = add_size(size,
2544 mul_size(NAMEDATALEN, MaxBackends));
2545 /* BackendActivityBuffer: */
2546 size = add_size(size,
2547 mul_size(pgstat_track_activity_query_size, MaxBackends));
2548 #ifdef USE_SSL
2549 /* BackendSslStatusBuffer: */
2550 size = add_size(size,
2551 mul_size(sizeof(PgBackendSSLStatus), MaxBackends));
2552 #endif
2553 return size;
2554 }
2555
2556 /*
2557 * Initialize the shared status array and several string buffers
2558 * during postmaster startup.
2559 */
2560 void
CreateSharedBackendStatus(void)2561 CreateSharedBackendStatus(void)
2562 {
2563 Size size;
2564 bool found;
2565 int i;
2566 char *buffer;
2567
2568 /* Create or attach to the shared array */
2569 size = mul_size(sizeof(PgBackendStatus), MaxBackends);
2570 BackendStatusArray = (PgBackendStatus *)
2571 ShmemInitStruct("Backend Status Array", size, &found);
2572
2573 if (!found)
2574 {
2575 /*
2576 * We're the first - initialize.
2577 */
2578 MemSet(BackendStatusArray, 0, size);
2579 }
2580
2581 /* Create or attach to the shared appname buffer */
2582 size = mul_size(NAMEDATALEN, MaxBackends);
2583 BackendAppnameBuffer = (char *)
2584 ShmemInitStruct("Backend Application Name Buffer", size, &found);
2585
2586 if (!found)
2587 {
2588 MemSet(BackendAppnameBuffer, 0, size);
2589
2590 /* Initialize st_appname pointers. */
2591 buffer = BackendAppnameBuffer;
2592 for (i = 0; i < MaxBackends; i++)
2593 {
2594 BackendStatusArray[i].st_appname = buffer;
2595 buffer += NAMEDATALEN;
2596 }
2597 }
2598
2599 /* Create or attach to the shared client hostname buffer */
2600 size = mul_size(NAMEDATALEN, MaxBackends);
2601 BackendClientHostnameBuffer = (char *)
2602 ShmemInitStruct("Backend Client Host Name Buffer", size, &found);
2603
2604 if (!found)
2605 {
2606 MemSet(BackendClientHostnameBuffer, 0, size);
2607
2608 /* Initialize st_clienthostname pointers. */
2609 buffer = BackendClientHostnameBuffer;
2610 for (i = 0; i < MaxBackends; i++)
2611 {
2612 BackendStatusArray[i].st_clienthostname = buffer;
2613 buffer += NAMEDATALEN;
2614 }
2615 }
2616
2617 /* Create or attach to the shared activity buffer */
2618 BackendActivityBufferSize = mul_size(pgstat_track_activity_query_size,
2619 MaxBackends);
2620 BackendActivityBuffer = (char *)
2621 ShmemInitStruct("Backend Activity Buffer",
2622 BackendActivityBufferSize,
2623 &found);
2624
2625 if (!found)
2626 {
2627 MemSet(BackendActivityBuffer, 0, BackendActivityBufferSize);
2628
2629 /* Initialize st_activity pointers. */
2630 buffer = BackendActivityBuffer;
2631 for (i = 0; i < MaxBackends; i++)
2632 {
2633 BackendStatusArray[i].st_activity = buffer;
2634 buffer += pgstat_track_activity_query_size;
2635 }
2636 }
2637
2638 #ifdef USE_SSL
2639 /* Create or attach to the shared SSL status buffer */
2640 size = mul_size(sizeof(PgBackendSSLStatus), MaxBackends);
2641 BackendSslStatusBuffer = (PgBackendSSLStatus *)
2642 ShmemInitStruct("Backend SSL Status Buffer", size, &found);
2643
2644 if (!found)
2645 {
2646 PgBackendSSLStatus *ptr;
2647
2648 MemSet(BackendSslStatusBuffer, 0, size);
2649
2650 /* Initialize st_sslstatus pointers. */
2651 ptr = BackendSslStatusBuffer;
2652 for (i = 0; i < MaxBackends; i++)
2653 {
2654 BackendStatusArray[i].st_sslstatus = ptr;
2655 ptr++;
2656 }
2657 }
2658 #endif
2659 }
2660
2661
2662 /* ----------
2663 * pgstat_initialize() -
2664 *
2665 * Initialize pgstats state, and set up our on-proc-exit hook.
2666 * Called from InitPostgres. MyBackendId must be set,
2667 * but we must not have started any transaction yet (since the
2668 * exit hook must run after the last transaction exit).
2669 * NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful.
2670 * ----------
2671 */
2672 void
pgstat_initialize(void)2673 pgstat_initialize(void)
2674 {
2675 /* Initialize MyBEEntry */
2676 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2677 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2678
2679 /* Set up a process-exit hook to clean up */
2680 on_shmem_exit(pgstat_beshutdown_hook, 0);
2681 }
2682
2683 /* ----------
2684 * pgstat_bestart() -
2685 *
2686 * Initialize this backend's entry in the PgBackendStatus array.
2687 * Called from InitPostgres.
2688 * Apart from auxiliary processes, MyBackendId, MyDatabaseId,
2689 * session userid, and application_name must be set for a
2690 * backend (hence, this cannot be combined with pgstat_initialize).
2691 * Note also that we must be inside a transaction if this isn't an aux
2692 * process, as we may need to do encoding conversion on some strings.
2693 * ----------
2694 */
2695 void
pgstat_bestart(void)2696 pgstat_bestart(void)
2697 {
2698 volatile PgBackendStatus *vbeentry = MyBEEntry;
2699 PgBackendStatus lbeentry;
2700 #ifdef USE_SSL
2701 PgBackendSSLStatus lsslstatus;
2702 #endif
2703
2704 /* pgstats state must be initialized from pgstat_initialize() */
2705 Assert(vbeentry != NULL);
2706
2707 /*
2708 * To minimize the time spent modifying the PgBackendStatus entry, and
2709 * avoid risk of errors inside the critical section, we first copy the
2710 * shared-memory struct to a local variable, then modify the data in the
2711 * local variable, then copy the local variable back to shared memory.
2712 * Only the last step has to be inside the critical section.
2713 *
2714 * Most of the data we copy from shared memory is just going to be
2715 * overwritten, but the struct's not so large that it's worth the
2716 * maintenance hassle to copy only the needful fields.
2717 */
2718 memcpy(&lbeentry,
2719 (char *) vbeentry,
2720 sizeof(PgBackendStatus));
2721
2722 /* This struct can just start from zeroes each time, though */
2723 #ifdef USE_SSL
2724 memset(&lsslstatus, 0, sizeof(lsslstatus));
2725 #endif
2726
2727 /*
2728 * Now fill in all the fields of lbeentry, except for strings that are
2729 * out-of-line data. Those have to be handled separately, below.
2730 */
2731 lbeentry.st_procpid = MyProcPid;
2732
2733 /*
2734 * If we have a MyProcPort, use its session start time (for consistency,
2735 * and to save a kernel call).
2736 */
2737 if (MyProcPort)
2738 lbeentry.st_proc_start_timestamp = MyProcPort->SessionStartTime;
2739 else
2740 lbeentry.st_proc_start_timestamp = GetCurrentTimestamp();
2741
2742 lbeentry.st_activity_start_timestamp = 0;
2743 lbeentry.st_state_start_timestamp = 0;
2744 lbeentry.st_xact_start_timestamp = 0;
2745 lbeentry.st_databaseid = MyDatabaseId;
2746 lbeentry.st_userid = GetSessionUserId();
2747
2748 /*
2749 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2750 * If so, use all-zeroes client address, which is dealt with specially in
2751 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2752 */
2753 if (MyProcPort)
2754 memcpy(&lbeentry.st_clientaddr, &MyProcPort->raddr,
2755 sizeof(lbeentry.st_clientaddr));
2756 else
2757 MemSet(&lbeentry.st_clientaddr, 0, sizeof(lbeentry.st_clientaddr));
2758
2759 #ifdef USE_SSL
2760 if (MyProcPort && MyProcPort->ssl != NULL)
2761 {
2762 lbeentry.st_ssl = true;
2763 lsslstatus.ssl_bits = be_tls_get_cipher_bits(MyProcPort);
2764 lsslstatus.ssl_compression = be_tls_get_compression(MyProcPort);
2765 be_tls_get_version(MyProcPort, lsslstatus.ssl_version, NAMEDATALEN);
2766 be_tls_get_cipher(MyProcPort, lsslstatus.ssl_cipher, NAMEDATALEN);
2767 be_tls_get_peerdn_name(MyProcPort, lsslstatus.ssl_clientdn, NAMEDATALEN);
2768 }
2769 else
2770 {
2771 lbeentry.st_ssl = false;
2772 }
2773 #else
2774 lbeentry.st_ssl = false;
2775 #endif
2776
2777 lbeentry.st_state = STATE_UNDEFINED;
2778 lbeentry.st_progress_command = PROGRESS_COMMAND_INVALID;
2779 lbeentry.st_progress_command_target = InvalidOid;
2780
2781 /*
2782 * we don't zero st_progress_param here to save cycles; nobody should
2783 * examine it until st_progress_command has been set to something other
2784 * than PROGRESS_COMMAND_INVALID
2785 */
2786
2787 /*
2788 * We're ready to enter the critical section that fills the shared-memory
2789 * status entry. We follow the protocol of bumping st_changecount before
2790 * and after; and make sure it's even afterwards. We use a volatile
2791 * pointer here to ensure the compiler doesn't try to get cute.
2792 */
2793 PGSTAT_BEGIN_WRITE_ACTIVITY(vbeentry);
2794
2795 /* make sure we'll memcpy the same st_changecount back */
2796 lbeentry.st_changecount = vbeentry->st_changecount;
2797
2798 memcpy((char *) vbeentry,
2799 &lbeentry,
2800 sizeof(PgBackendStatus));
2801
2802 /*
2803 * We can write the out-of-line strings and structs using the pointers
2804 * that are in lbeentry; this saves some de-volatilizing messiness.
2805 */
2806 lbeentry.st_appname[0] = '\0';
2807 if (MyProcPort && MyProcPort->remote_hostname)
2808 strlcpy(lbeentry.st_clienthostname, MyProcPort->remote_hostname,
2809 NAMEDATALEN);
2810 else
2811 lbeentry.st_clienthostname[0] = '\0';
2812 lbeentry.st_activity[0] = '\0';
2813 /* Also make sure the last byte in each string area is always 0 */
2814 lbeentry.st_appname[NAMEDATALEN - 1] = '\0';
2815 lbeentry.st_clienthostname[NAMEDATALEN - 1] = '\0';
2816 lbeentry.st_activity[pgstat_track_activity_query_size - 1] = '\0';
2817
2818 #ifdef USE_SSL
2819 memcpy(lbeentry.st_sslstatus, &lsslstatus, sizeof(PgBackendSSLStatus));
2820 #endif
2821
2822 PGSTAT_END_WRITE_ACTIVITY(vbeentry);
2823
2824 /* Update app name to current GUC setting */
2825 if (application_name)
2826 pgstat_report_appname(application_name);
2827 }
2828
2829 /*
2830 * Shut down a single backend's statistics reporting at process exit.
2831 *
2832 * Flush any remaining statistics counts out to the collector.
2833 * Without this, operations triggered during backend exit (such as
2834 * temp table deletions) won't be counted.
2835 *
2836 * Lastly, clear out our entry in the PgBackendStatus array.
2837 */
2838 static void
pgstat_beshutdown_hook(int code,Datum arg)2839 pgstat_beshutdown_hook(int code, Datum arg)
2840 {
2841 volatile PgBackendStatus *beentry = MyBEEntry;
2842
2843 /*
2844 * If we got as far as discovering our own database ID, we can report what
2845 * we did to the collector. Otherwise, we'd be sending an invalid
2846 * database ID, so forget it. (This means that accesses to pg_database
2847 * during failed backend starts might never get counted.)
2848 */
2849 if (OidIsValid(MyDatabaseId))
2850 pgstat_report_stat(true);
2851
2852 /*
2853 * Clear my status entry, following the protocol of bumping st_changecount
2854 * before and after. We use a volatile pointer here to ensure the
2855 * compiler doesn't try to get cute.
2856 */
2857 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
2858
2859 beentry->st_procpid = 0; /* mark invalid */
2860
2861 PGSTAT_END_WRITE_ACTIVITY(beentry);
2862 }
2863
2864
2865 /* ----------
2866 * pgstat_report_activity() -
2867 *
2868 * Called from tcop/postgres.c to report what the backend is actually doing
2869 * (but note cmd_str can be NULL for certain cases).
2870 *
2871 * All updates of the status entry follow the protocol of bumping
2872 * st_changecount before and after. We use a volatile pointer here to
2873 * ensure the compiler doesn't try to get cute.
2874 * ----------
2875 */
2876 void
pgstat_report_activity(BackendState state,const char * cmd_str)2877 pgstat_report_activity(BackendState state, const char *cmd_str)
2878 {
2879 volatile PgBackendStatus *beentry = MyBEEntry;
2880 TimestampTz start_timestamp;
2881 TimestampTz current_timestamp;
2882 int len = 0;
2883
2884 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
2885
2886 if (!beentry)
2887 return;
2888
2889 if (!pgstat_track_activities)
2890 {
2891 if (beentry->st_state != STATE_DISABLED)
2892 {
2893 volatile PGPROC *proc = MyProc;
2894
2895 /*
2896 * track_activities is disabled, but we last reported a
2897 * non-disabled state. As our final update, change the state and
2898 * clear fields we will not be updating anymore.
2899 */
2900 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
2901 beentry->st_state = STATE_DISABLED;
2902 beentry->st_state_start_timestamp = 0;
2903 beentry->st_activity[0] = '\0';
2904 beentry->st_activity_start_timestamp = 0;
2905 /* st_xact_start_timestamp and wait_event_info are also disabled */
2906 beentry->st_xact_start_timestamp = 0;
2907 proc->wait_event_info = 0;
2908 PGSTAT_END_WRITE_ACTIVITY(beentry);
2909 }
2910 return;
2911 }
2912
2913 /*
2914 * To minimize the time spent modifying the entry, and avoid risk of
2915 * errors inside the critical section, fetch all the needed data first.
2916 */
2917 start_timestamp = GetCurrentStatementStartTimestamp();
2918 if (cmd_str != NULL)
2919 {
2920 len = pg_mbcliplen(cmd_str, strlen(cmd_str),
2921 pgstat_track_activity_query_size - 1);
2922 }
2923 current_timestamp = GetCurrentTimestamp();
2924
2925 /*
2926 * Now update the status entry
2927 */
2928 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
2929
2930 beentry->st_state = state;
2931 beentry->st_state_start_timestamp = current_timestamp;
2932
2933 if (cmd_str != NULL)
2934 {
2935 memcpy((char *) beentry->st_activity, cmd_str, len);
2936 beentry->st_activity[len] = '\0';
2937 beentry->st_activity_start_timestamp = start_timestamp;
2938 }
2939
2940 PGSTAT_END_WRITE_ACTIVITY(beentry);
2941 }
2942
2943 /*-----------
2944 * pgstat_progress_start_command() -
2945 *
2946 * Set st_progress_command (and st_progress_command_target) in own backend
2947 * entry. Also, zero-initialize st_progress_param array.
2948 *-----------
2949 */
2950 void
pgstat_progress_start_command(ProgressCommandType cmdtype,Oid relid)2951 pgstat_progress_start_command(ProgressCommandType cmdtype, Oid relid)
2952 {
2953 volatile PgBackendStatus *beentry = MyBEEntry;
2954
2955 if (!beentry || !pgstat_track_activities)
2956 return;
2957
2958 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
2959 beentry->st_progress_command = cmdtype;
2960 beentry->st_progress_command_target = relid;
2961 MemSet(&beentry->st_progress_param, 0, sizeof(beentry->st_progress_param));
2962 PGSTAT_END_WRITE_ACTIVITY(beentry);
2963 }
2964
2965 /*-----------
2966 * pgstat_progress_update_param() -
2967 *
2968 * Update index'th member in st_progress_param[] of own backend entry.
2969 *-----------
2970 */
2971 void
pgstat_progress_update_param(int index,int64 val)2972 pgstat_progress_update_param(int index, int64 val)
2973 {
2974 volatile PgBackendStatus *beentry = MyBEEntry;
2975
2976 Assert(index >= 0 && index < PGSTAT_NUM_PROGRESS_PARAM);
2977
2978 if (!beentry || !pgstat_track_activities)
2979 return;
2980
2981 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
2982 beentry->st_progress_param[index] = val;
2983 PGSTAT_END_WRITE_ACTIVITY(beentry);
2984 }
2985
2986 /*-----------
2987 * pgstat_progress_update_multi_param() -
2988 *
2989 * Update multiple members in st_progress_param[] of own backend entry.
2990 * This is atomic; readers won't see intermediate states.
2991 *-----------
2992 */
2993 void
pgstat_progress_update_multi_param(int nparam,const int * index,const int64 * val)2994 pgstat_progress_update_multi_param(int nparam, const int *index,
2995 const int64 *val)
2996 {
2997 volatile PgBackendStatus *beentry = MyBEEntry;
2998 int i;
2999
3000 if (!beentry || !pgstat_track_activities || nparam == 0)
3001 return;
3002
3003 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3004
3005 for (i = 0; i < nparam; ++i)
3006 {
3007 Assert(index[i] >= 0 && index[i] < PGSTAT_NUM_PROGRESS_PARAM);
3008
3009 beentry->st_progress_param[index[i]] = val[i];
3010 }
3011
3012 PGSTAT_END_WRITE_ACTIVITY(beentry);
3013 }
3014
3015 /*-----------
3016 * pgstat_progress_end_command() -
3017 *
3018 * Reset st_progress_command (and st_progress_command_target) in own backend
3019 * entry. This signals the end of the command.
3020 *-----------
3021 */
3022 void
pgstat_progress_end_command(void)3023 pgstat_progress_end_command(void)
3024 {
3025 volatile PgBackendStatus *beentry = MyBEEntry;
3026
3027 if (!beentry || !pgstat_track_activities)
3028 return;
3029
3030 if (beentry->st_progress_command == PROGRESS_COMMAND_INVALID)
3031 return;
3032
3033 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3034 beentry->st_progress_command = PROGRESS_COMMAND_INVALID;
3035 beentry->st_progress_command_target = InvalidOid;
3036 PGSTAT_END_WRITE_ACTIVITY(beentry);
3037 }
3038
3039 /* ----------
3040 * pgstat_report_appname() -
3041 *
3042 * Called to update our application name.
3043 * ----------
3044 */
3045 void
pgstat_report_appname(const char * appname)3046 pgstat_report_appname(const char *appname)
3047 {
3048 volatile PgBackendStatus *beentry = MyBEEntry;
3049 int len;
3050
3051 if (!beentry)
3052 return;
3053
3054 /* This should be unnecessary if GUC did its job, but be safe */
3055 len = pg_mbcliplen(appname, strlen(appname), NAMEDATALEN - 1);
3056
3057 /*
3058 * Update my status entry, following the protocol of bumping
3059 * st_changecount before and after. We use a volatile pointer here to
3060 * ensure the compiler doesn't try to get cute.
3061 */
3062 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3063
3064 memcpy((char *) beentry->st_appname, appname, len);
3065 beentry->st_appname[len] = '\0';
3066
3067 PGSTAT_END_WRITE_ACTIVITY(beentry);
3068 }
3069
3070 /*
3071 * Report current transaction start timestamp as the specified value.
3072 * Zero means there is no active transaction.
3073 */
3074 void
pgstat_report_xact_timestamp(TimestampTz tstamp)3075 pgstat_report_xact_timestamp(TimestampTz tstamp)
3076 {
3077 volatile PgBackendStatus *beentry = MyBEEntry;
3078
3079 if (!pgstat_track_activities || !beentry)
3080 return;
3081
3082 /*
3083 * Update my status entry, following the protocol of bumping
3084 * st_changecount before and after. We use a volatile pointer here to
3085 * ensure the compiler doesn't try to get cute.
3086 */
3087 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3088
3089 beentry->st_xact_start_timestamp = tstamp;
3090
3091 PGSTAT_END_WRITE_ACTIVITY(beentry);
3092 }
3093
3094 /* ----------
3095 * pgstat_read_current_status() -
3096 *
3097 * Copy the current contents of the PgBackendStatus array to local memory,
3098 * if not already done in this transaction.
3099 * ----------
3100 */
3101 static void
pgstat_read_current_status(void)3102 pgstat_read_current_status(void)
3103 {
3104 volatile PgBackendStatus *beentry;
3105 LocalPgBackendStatus *localtable;
3106 LocalPgBackendStatus *localentry;
3107 char *localappname,
3108 *localclienthostname,
3109 *localactivity;
3110 #ifdef USE_SSL
3111 PgBackendSSLStatus *localsslstatus;
3112 #endif
3113 int i;
3114
3115 Assert(!pgStatRunningInCollector);
3116 if (localBackendStatusTable)
3117 return; /* already done */
3118
3119 pgstat_setup_memcxt();
3120
3121 localtable = (LocalPgBackendStatus *)
3122 MemoryContextAlloc(pgStatLocalContext,
3123 sizeof(LocalPgBackendStatus) * MaxBackends);
3124 localappname = (char *)
3125 MemoryContextAlloc(pgStatLocalContext,
3126 NAMEDATALEN * MaxBackends);
3127 localclienthostname = (char *)
3128 MemoryContextAlloc(pgStatLocalContext,
3129 NAMEDATALEN * MaxBackends);
3130 localactivity = (char *)
3131 MemoryContextAlloc(pgStatLocalContext,
3132 pgstat_track_activity_query_size * MaxBackends);
3133 #ifdef USE_SSL
3134 localsslstatus = (PgBackendSSLStatus *)
3135 MemoryContextAlloc(pgStatLocalContext,
3136 sizeof(PgBackendSSLStatus) * MaxBackends);
3137 #endif
3138
3139 localNumBackends = 0;
3140
3141 beentry = BackendStatusArray;
3142 localentry = localtable;
3143 for (i = 1; i <= MaxBackends; i++)
3144 {
3145 /*
3146 * Follow the protocol of retrying if st_changecount changes while we
3147 * copy the entry, or if it's odd. (The check for odd is needed to
3148 * cover the case where we are able to completely copy the entry while
3149 * the source backend is between increment steps.) We use a volatile
3150 * pointer here to ensure the compiler doesn't try to get cute.
3151 */
3152 for (;;)
3153 {
3154 int before_changecount;
3155 int after_changecount;
3156
3157 pgstat_begin_read_activity(beentry, before_changecount);
3158
3159 localentry->backendStatus.st_procpid = beentry->st_procpid;
3160 /* Skip all the data-copying work if entry is not in use */
3161 if (localentry->backendStatus.st_procpid > 0)
3162 {
3163 memcpy(&localentry->backendStatus, (char *) beentry, sizeof(PgBackendStatus));
3164
3165 /*
3166 * For each PgBackendStatus field that is a pointer, copy the
3167 * pointed-to data, then adjust the local copy of the pointer
3168 * field to point at the local copy of the data.
3169 *
3170 * strcpy is safe even if the string is modified concurrently,
3171 * because there's always a \0 at the end of the buffer.
3172 */
3173 strcpy(localappname, (char *) beentry->st_appname);
3174 localentry->backendStatus.st_appname = localappname;
3175 strcpy(localclienthostname, (char *) beentry->st_clienthostname);
3176 localentry->backendStatus.st_clienthostname = localclienthostname;
3177 strcpy(localactivity, (char *) beentry->st_activity);
3178 localentry->backendStatus.st_activity = localactivity;
3179 #ifdef USE_SSL
3180 if (beentry->st_ssl)
3181 {
3182 memcpy(localsslstatus, beentry->st_sslstatus, sizeof(PgBackendSSLStatus));
3183 localentry->backendStatus.st_sslstatus = localsslstatus;
3184 }
3185 #endif
3186 }
3187
3188 pgstat_end_read_activity(beentry, after_changecount);
3189
3190 if (pgstat_read_activity_complete(before_changecount,
3191 after_changecount))
3192 break;
3193
3194 /* Make sure we can break out of loop if stuck... */
3195 CHECK_FOR_INTERRUPTS();
3196 }
3197
3198 beentry++;
3199 /* Only valid entries get included into the local array */
3200 if (localentry->backendStatus.st_procpid > 0)
3201 {
3202 BackendIdGetTransactionIds(i,
3203 &localentry->backend_xid,
3204 &localentry->backend_xmin);
3205
3206 localentry++;
3207 localappname += NAMEDATALEN;
3208 localclienthostname += NAMEDATALEN;
3209 localactivity += pgstat_track_activity_query_size;
3210 #ifdef USE_SSL
3211 localsslstatus++;
3212 #endif
3213 localNumBackends++;
3214 }
3215 }
3216
3217 /* Set the pointer only after completion of a valid table */
3218 localBackendStatusTable = localtable;
3219 }
3220
3221 /* ----------
3222 * pgstat_get_wait_event_type() -
3223 *
3224 * Return a string representing the current wait event type, backend is
3225 * waiting on.
3226 */
3227 const char *
pgstat_get_wait_event_type(uint32 wait_event_info)3228 pgstat_get_wait_event_type(uint32 wait_event_info)
3229 {
3230 uint8 classId;
3231 const char *event_type;
3232
3233 /* report process as not waiting. */
3234 if (wait_event_info == 0)
3235 return NULL;
3236
3237 wait_event_info = wait_event_info >> 24;
3238 classId = wait_event_info & 0XFF;
3239
3240 switch (classId)
3241 {
3242 case WAIT_LWLOCK_NAMED:
3243 event_type = "LWLockNamed";
3244 break;
3245 case WAIT_LWLOCK_TRANCHE:
3246 event_type = "LWLockTranche";
3247 break;
3248 case WAIT_LOCK:
3249 event_type = "Lock";
3250 break;
3251 case WAIT_BUFFER_PIN:
3252 event_type = "BufferPin";
3253 break;
3254 default:
3255 event_type = "???";
3256 break;
3257 }
3258
3259 return event_type;
3260 }
3261
3262 /* ----------
3263 * pgstat_get_wait_event() -
3264 *
3265 * Return a string representing the current wait event, backend is
3266 * waiting on.
3267 */
3268 const char *
pgstat_get_wait_event(uint32 wait_event_info)3269 pgstat_get_wait_event(uint32 wait_event_info)
3270 {
3271 uint8 classId;
3272 uint16 eventId;
3273 const char *event_name;
3274
3275 /* report process as not waiting. */
3276 if (wait_event_info == 0)
3277 return NULL;
3278
3279 eventId = wait_event_info & ((1 << 24) - 1);
3280 wait_event_info = wait_event_info >> 24;
3281 classId = wait_event_info & 0XFF;
3282
3283 switch (classId)
3284 {
3285 case WAIT_LWLOCK_NAMED:
3286 case WAIT_LWLOCK_TRANCHE:
3287 event_name = GetLWLockIdentifier(classId, eventId);
3288 break;
3289 case WAIT_LOCK:
3290 event_name = GetLockNameFromTagType(eventId);
3291 break;
3292 case WAIT_BUFFER_PIN:
3293 event_name = "BufferPin";
3294 break;
3295 default:
3296 event_name = "unknown wait event";
3297 break;
3298 }
3299
3300 return event_name;
3301 }
3302
3303 /* ----------
3304 * pgstat_get_backend_current_activity() -
3305 *
3306 * Return a string representing the current activity of the backend with
3307 * the specified PID. This looks directly at the BackendStatusArray,
3308 * and so will provide current information regardless of the age of our
3309 * transaction's snapshot of the status array.
3310 *
3311 * It is the caller's responsibility to invoke this only for backends whose
3312 * state is expected to remain stable while the result is in use. The
3313 * only current use is in deadlock reporting, where we can expect that
3314 * the target backend is blocked on a lock. (There are corner cases
3315 * where the target's wait could get aborted while we are looking at it,
3316 * but the very worst consequence is to return a pointer to a string
3317 * that's been changed, so we won't worry too much.)
3318 *
3319 * Note: return strings for special cases match pg_stat_get_backend_activity.
3320 * ----------
3321 */
3322 const char *
pgstat_get_backend_current_activity(int pid,bool checkUser)3323 pgstat_get_backend_current_activity(int pid, bool checkUser)
3324 {
3325 PgBackendStatus *beentry;
3326 int i;
3327
3328 beentry = BackendStatusArray;
3329 for (i = 1; i <= MaxBackends; i++)
3330 {
3331 /*
3332 * Although we expect the target backend's entry to be stable, that
3333 * doesn't imply that anyone else's is. To avoid identifying the
3334 * wrong backend, while we check for a match to the desired PID we
3335 * must follow the protocol of retrying if st_changecount changes
3336 * while we examine the entry, or if it's odd. (This might be
3337 * unnecessary, since fetching or storing an int is almost certainly
3338 * atomic, but let's play it safe.) We use a volatile pointer here to
3339 * ensure the compiler doesn't try to get cute.
3340 */
3341 volatile PgBackendStatus *vbeentry = beentry;
3342 bool found;
3343
3344 for (;;)
3345 {
3346 int before_changecount;
3347 int after_changecount;
3348
3349 pgstat_begin_read_activity(vbeentry, before_changecount);
3350
3351 found = (vbeentry->st_procpid == pid);
3352
3353 pgstat_end_read_activity(vbeentry, after_changecount);
3354
3355 if (pgstat_read_activity_complete(before_changecount,
3356 after_changecount))
3357 break;
3358
3359 /* Make sure we can break out of loop if stuck... */
3360 CHECK_FOR_INTERRUPTS();
3361 }
3362
3363 if (found)
3364 {
3365 /* Now it is safe to use the non-volatile pointer */
3366 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
3367 return "<insufficient privilege>";
3368 else if (*(beentry->st_activity) == '\0')
3369 return "<command string not enabled>";
3370 else
3371 return beentry->st_activity;
3372 }
3373
3374 beentry++;
3375 }
3376
3377 /* If we get here, caller is in error ... */
3378 return "<backend information not available>";
3379 }
3380
3381 /* ----------
3382 * pgstat_get_crashed_backend_activity() -
3383 *
3384 * Return a string representing the current activity of the backend with
3385 * the specified PID. Like the function above, but reads shared memory with
3386 * the expectation that it may be corrupt. On success, copy the string
3387 * into the "buffer" argument and return that pointer. On failure,
3388 * return NULL.
3389 *
3390 * This function is only intended to be used by the postmaster to report the
3391 * query that crashed a backend. In particular, no attempt is made to
3392 * follow the correct concurrency protocol when accessing the
3393 * BackendStatusArray. But that's OK, in the worst case we'll return a
3394 * corrupted message. We also must take care not to trip on ereport(ERROR).
3395 * ----------
3396 */
3397 const char *
pgstat_get_crashed_backend_activity(int pid,char * buffer,int buflen)3398 pgstat_get_crashed_backend_activity(int pid, char *buffer, int buflen)
3399 {
3400 volatile PgBackendStatus *beentry;
3401 int i;
3402
3403 beentry = BackendStatusArray;
3404
3405 /*
3406 * We probably shouldn't get here before shared memory has been set up,
3407 * but be safe.
3408 */
3409 if (beentry == NULL || BackendActivityBuffer == NULL)
3410 return NULL;
3411
3412 for (i = 1; i <= MaxBackends; i++)
3413 {
3414 if (beentry->st_procpid == pid)
3415 {
3416 /* Read pointer just once, so it can't change after validation */
3417 const char *activity = beentry->st_activity;
3418 const char *activity_last;
3419
3420 /*
3421 * We mustn't access activity string before we verify that it
3422 * falls within the BackendActivityBuffer. To make sure that the
3423 * entire string including its ending is contained within the
3424 * buffer, subtract one activity length from the buffer size.
3425 */
3426 activity_last = BackendActivityBuffer + BackendActivityBufferSize
3427 - pgstat_track_activity_query_size;
3428
3429 if (activity < BackendActivityBuffer ||
3430 activity > activity_last)
3431 return NULL;
3432
3433 /* If no string available, no point in a report */
3434 if (activity[0] == '\0')
3435 return NULL;
3436
3437 /*
3438 * Copy only ASCII-safe characters so we don't run into encoding
3439 * problems when reporting the message; and be sure not to run off
3440 * the end of memory.
3441 */
3442 ascii_safe_strlcpy(buffer, activity,
3443 Min(buflen, pgstat_track_activity_query_size));
3444
3445 return buffer;
3446 }
3447
3448 beentry++;
3449 }
3450
3451 /* PID not found */
3452 return NULL;
3453 }
3454
3455
3456 /* ------------------------------------------------------------
3457 * Local support functions follow
3458 * ------------------------------------------------------------
3459 */
3460
3461
3462 /* ----------
3463 * pgstat_setheader() -
3464 *
3465 * Set common header fields in a statistics message
3466 * ----------
3467 */
3468 static void
pgstat_setheader(PgStat_MsgHdr * hdr,StatMsgType mtype)3469 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
3470 {
3471 hdr->m_type = mtype;
3472 }
3473
3474
3475 /* ----------
3476 * pgstat_send() -
3477 *
3478 * Send out one statistics message to the collector
3479 * ----------
3480 */
3481 static void
pgstat_send(void * msg,int len)3482 pgstat_send(void *msg, int len)
3483 {
3484 int rc;
3485
3486 if (pgStatSock == PGINVALID_SOCKET)
3487 return;
3488
3489 ((PgStat_MsgHdr *) msg)->m_size = len;
3490
3491 /* We'll retry after EINTR, but ignore all other failures */
3492 do
3493 {
3494 rc = send(pgStatSock, msg, len, 0);
3495 } while (rc < 0 && errno == EINTR);
3496
3497 #ifdef USE_ASSERT_CHECKING
3498 /* In debug builds, log send failures ... */
3499 if (rc < 0)
3500 elog(LOG, "could not send to statistics collector: %m");
3501 #endif
3502 }
3503
3504 /* ----------
3505 * pgstat_send_archiver() -
3506 *
3507 * Tell the collector about the WAL file that we successfully
3508 * archived or failed to archive.
3509 * ----------
3510 */
3511 void
pgstat_send_archiver(const char * xlog,bool failed)3512 pgstat_send_archiver(const char *xlog, bool failed)
3513 {
3514 PgStat_MsgArchiver msg;
3515
3516 /*
3517 * Prepare and send the message
3518 */
3519 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ARCHIVER);
3520 msg.m_failed = failed;
3521 StrNCpy(msg.m_xlog, xlog, sizeof(msg.m_xlog));
3522 msg.m_timestamp = GetCurrentTimestamp();
3523 pgstat_send(&msg, sizeof(msg));
3524 }
3525
3526 /* ----------
3527 * pgstat_send_bgwriter() -
3528 *
3529 * Send bgwriter statistics to the collector
3530 * ----------
3531 */
3532 void
pgstat_send_bgwriter(void)3533 pgstat_send_bgwriter(void)
3534 {
3535 /* We assume this initializes to zeroes */
3536 static const PgStat_MsgBgWriter all_zeroes;
3537
3538 /*
3539 * This function can be called even if nothing at all has happened. In
3540 * this case, avoid sending a completely empty message to the stats
3541 * collector.
3542 */
3543 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
3544 return;
3545
3546 /*
3547 * Prepare and send the message
3548 */
3549 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
3550 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
3551
3552 /*
3553 * Clear out the statistics buffer, so it can be re-used.
3554 */
3555 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
3556 }
3557
3558
3559 /* ----------
3560 * PgstatCollectorMain() -
3561 *
3562 * Start up the statistics collector process. This is the body of the
3563 * postmaster child process.
3564 *
3565 * The argc/argv parameters are valid only in EXEC_BACKEND case.
3566 * ----------
3567 */
3568 NON_EXEC_STATIC void
PgstatCollectorMain(int argc,char * argv[])3569 PgstatCollectorMain(int argc, char *argv[])
3570 {
3571 int len;
3572 PgStat_Msg msg;
3573 int wr;
3574
3575 /*
3576 * Ignore all signals usually bound to some action in the postmaster,
3577 * except SIGHUP and SIGQUIT. Note we don't need a SIGUSR1 handler to
3578 * support latch operations, because we only use a local latch.
3579 */
3580 pqsignal(SIGHUP, pgstat_sighup_handler);
3581 pqsignal(SIGINT, SIG_IGN);
3582 pqsignal(SIGTERM, SIG_IGN);
3583 pqsignal(SIGQUIT, pgstat_exit);
3584 pqsignal(SIGALRM, SIG_IGN);
3585 pqsignal(SIGPIPE, SIG_IGN);
3586 pqsignal(SIGUSR1, SIG_IGN);
3587 pqsignal(SIGUSR2, SIG_IGN);
3588 pqsignal(SIGCHLD, SIG_DFL);
3589 pqsignal(SIGTTIN, SIG_DFL);
3590 pqsignal(SIGTTOU, SIG_DFL);
3591 pqsignal(SIGCONT, SIG_DFL);
3592 pqsignal(SIGWINCH, SIG_DFL);
3593 PG_SETMASK(&UnBlockSig);
3594
3595 /*
3596 * Identify myself via ps
3597 */
3598 init_ps_display("stats collector process", "", "", "");
3599
3600 /*
3601 * Read in existing stats files or initialize the stats to zero.
3602 */
3603 pgStatRunningInCollector = true;
3604 pgStatDBHash = pgstat_read_statsfiles(InvalidOid, true, true);
3605
3606 /*
3607 * Loop to process messages until we get SIGQUIT or detect ungraceful
3608 * death of our parent postmaster.
3609 *
3610 * For performance reasons, we don't want to do ResetLatch/WaitLatch after
3611 * every message; instead, do that only after a recv() fails to obtain a
3612 * message. (This effectively means that if backends are sending us stuff
3613 * like mad, we won't notice postmaster death until things slack off a
3614 * bit; which seems fine.) To do that, we have an inner loop that
3615 * iterates as long as recv() succeeds. We do recognize got_SIGHUP inside
3616 * the inner loop, which means that such interrupts will get serviced but
3617 * the latch won't get cleared until next time there is a break in the
3618 * action.
3619 */
3620 for (;;)
3621 {
3622 /* Clear any already-pending wakeups */
3623 ResetLatch(MyLatch);
3624
3625 /*
3626 * Quit if we get SIGQUIT from the postmaster.
3627 */
3628 if (need_exit)
3629 break;
3630
3631 /*
3632 * Inner loop iterates as long as we keep getting messages, or until
3633 * need_exit becomes set.
3634 */
3635 while (!need_exit)
3636 {
3637 /*
3638 * Reload configuration if we got SIGHUP from the postmaster.
3639 */
3640 if (got_SIGHUP)
3641 {
3642 got_SIGHUP = false;
3643 ProcessConfigFile(PGC_SIGHUP);
3644 }
3645
3646 /*
3647 * Write the stats file(s) if a new request has arrived that is
3648 * not satisfied by existing file(s).
3649 */
3650 if (pgstat_write_statsfile_needed())
3651 pgstat_write_statsfiles(false, false);
3652
3653 /*
3654 * Try to receive and process a message. This will not block,
3655 * since the socket is set to non-blocking mode.
3656 *
3657 * XXX On Windows, we have to force pgwin32_recv to cooperate,
3658 * despite the previous use of pg_set_noblock() on the socket.
3659 * This is extremely broken and should be fixed someday.
3660 */
3661 #ifdef WIN32
3662 pgwin32_noblock = 1;
3663 #endif
3664
3665 len = recv(pgStatSock, (char *) &msg,
3666 sizeof(PgStat_Msg), 0);
3667
3668 #ifdef WIN32
3669 pgwin32_noblock = 0;
3670 #endif
3671
3672 if (len < 0)
3673 {
3674 if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
3675 break; /* out of inner loop */
3676 ereport(ERROR,
3677 (errcode_for_socket_access(),
3678 errmsg("could not read statistics message: %m")));
3679 }
3680
3681 /*
3682 * We ignore messages that are smaller than our common header
3683 */
3684 if (len < sizeof(PgStat_MsgHdr))
3685 continue;
3686
3687 /*
3688 * The received length must match the length in the header
3689 */
3690 if (msg.msg_hdr.m_size != len)
3691 continue;
3692
3693 /*
3694 * O.K. - we accept this message. Process it.
3695 */
3696 switch (msg.msg_hdr.m_type)
3697 {
3698 case PGSTAT_MTYPE_DUMMY:
3699 break;
3700
3701 case PGSTAT_MTYPE_INQUIRY:
3702 pgstat_recv_inquiry((PgStat_MsgInquiry *) &msg, len);
3703 break;
3704
3705 case PGSTAT_MTYPE_TABSTAT:
3706 pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len);
3707 break;
3708
3709 case PGSTAT_MTYPE_TABPURGE:
3710 pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len);
3711 break;
3712
3713 case PGSTAT_MTYPE_DROPDB:
3714 pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len);
3715 break;
3716
3717 case PGSTAT_MTYPE_RESETCOUNTER:
3718 pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg,
3719 len);
3720 break;
3721
3722 case PGSTAT_MTYPE_RESETSHAREDCOUNTER:
3723 pgstat_recv_resetsharedcounter(
3724 (PgStat_MsgResetsharedcounter *) &msg,
3725 len);
3726 break;
3727
3728 case PGSTAT_MTYPE_RESETSINGLECOUNTER:
3729 pgstat_recv_resetsinglecounter(
3730 (PgStat_MsgResetsinglecounter *) &msg,
3731 len);
3732 break;
3733
3734 case PGSTAT_MTYPE_AUTOVAC_START:
3735 pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
3736 break;
3737
3738 case PGSTAT_MTYPE_VACUUM:
3739 pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
3740 break;
3741
3742 case PGSTAT_MTYPE_ANALYZE:
3743 pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len);
3744 break;
3745
3746 case PGSTAT_MTYPE_ARCHIVER:
3747 pgstat_recv_archiver((PgStat_MsgArchiver *) &msg, len);
3748 break;
3749
3750 case PGSTAT_MTYPE_BGWRITER:
3751 pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len);
3752 break;
3753
3754 case PGSTAT_MTYPE_FUNCSTAT:
3755 pgstat_recv_funcstat((PgStat_MsgFuncstat *) &msg, len);
3756 break;
3757
3758 case PGSTAT_MTYPE_FUNCPURGE:
3759 pgstat_recv_funcpurge((PgStat_MsgFuncpurge *) &msg, len);
3760 break;
3761
3762 case PGSTAT_MTYPE_RECOVERYCONFLICT:
3763 pgstat_recv_recoveryconflict((PgStat_MsgRecoveryConflict *) &msg, len);
3764 break;
3765
3766 case PGSTAT_MTYPE_DEADLOCK:
3767 pgstat_recv_deadlock((PgStat_MsgDeadlock *) &msg, len);
3768 break;
3769
3770 case PGSTAT_MTYPE_TEMPFILE:
3771 pgstat_recv_tempfile((PgStat_MsgTempFile *) &msg, len);
3772 break;
3773
3774 default:
3775 break;
3776 }
3777 } /* end of inner message-processing loop */
3778
3779 /* Sleep until there's something to do */
3780 #ifndef WIN32
3781 wr = WaitLatchOrSocket(MyLatch,
3782 WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE,
3783 pgStatSock,
3784 -1L);
3785 #else
3786
3787 /*
3788 * Windows, at least in its Windows Server 2003 R2 incarnation,
3789 * sometimes loses FD_READ events. Waking up and retrying the recv()
3790 * fixes that, so don't sleep indefinitely. This is a crock of the
3791 * first water, but until somebody wants to debug exactly what's
3792 * happening there, this is the best we can do. The two-second
3793 * timeout matches our pre-9.2 behavior, and needs to be short enough
3794 * to not provoke "using stale statistics" complaints from
3795 * backend_read_statsfile.
3796 */
3797 wr = WaitLatchOrSocket(MyLatch,
3798 WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE | WL_TIMEOUT,
3799 pgStatSock,
3800 2 * 1000L /* msec */ );
3801 #endif
3802
3803 /*
3804 * Emergency bailout if postmaster has died. This is to avoid the
3805 * necessity for manual cleanup of all postmaster children.
3806 */
3807 if (wr & WL_POSTMASTER_DEATH)
3808 break;
3809 } /* end of outer loop */
3810
3811 /*
3812 * Save the final stats to reuse at next startup.
3813 */
3814 pgstat_write_statsfiles(true, true);
3815
3816 exit(0);
3817 }
3818
3819
3820 /* SIGQUIT signal handler for collector process */
3821 static void
pgstat_exit(SIGNAL_ARGS)3822 pgstat_exit(SIGNAL_ARGS)
3823 {
3824 int save_errno = errno;
3825
3826 need_exit = true;
3827 SetLatch(MyLatch);
3828
3829 errno = save_errno;
3830 }
3831
3832 /* SIGHUP handler for collector process */
3833 static void
pgstat_sighup_handler(SIGNAL_ARGS)3834 pgstat_sighup_handler(SIGNAL_ARGS)
3835 {
3836 int save_errno = errno;
3837
3838 got_SIGHUP = true;
3839 SetLatch(MyLatch);
3840
3841 errno = save_errno;
3842 }
3843
3844 /*
3845 * Subroutine to clear stats in a database entry
3846 *
3847 * Tables and functions hashes are initialized to empty.
3848 */
3849 static void
reset_dbentry_counters(PgStat_StatDBEntry * dbentry)3850 reset_dbentry_counters(PgStat_StatDBEntry *dbentry)
3851 {
3852 HASHCTL hash_ctl;
3853
3854 dbentry->n_xact_commit = 0;
3855 dbentry->n_xact_rollback = 0;
3856 dbentry->n_blocks_fetched = 0;
3857 dbentry->n_blocks_hit = 0;
3858 dbentry->n_tuples_returned = 0;
3859 dbentry->n_tuples_fetched = 0;
3860 dbentry->n_tuples_inserted = 0;
3861 dbentry->n_tuples_updated = 0;
3862 dbentry->n_tuples_deleted = 0;
3863 dbentry->last_autovac_time = 0;
3864 dbentry->n_conflict_tablespace = 0;
3865 dbentry->n_conflict_lock = 0;
3866 dbentry->n_conflict_snapshot = 0;
3867 dbentry->n_conflict_bufferpin = 0;
3868 dbentry->n_conflict_startup_deadlock = 0;
3869 dbentry->n_temp_files = 0;
3870 dbentry->n_temp_bytes = 0;
3871 dbentry->n_deadlocks = 0;
3872 dbentry->n_block_read_time = 0;
3873 dbentry->n_block_write_time = 0;
3874
3875 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
3876 dbentry->stats_timestamp = 0;
3877
3878 memset(&hash_ctl, 0, sizeof(hash_ctl));
3879 hash_ctl.keysize = sizeof(Oid);
3880 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
3881 dbentry->tables = hash_create("Per-database table",
3882 PGSTAT_TAB_HASH_SIZE,
3883 &hash_ctl,
3884 HASH_ELEM | HASH_BLOBS);
3885
3886 hash_ctl.keysize = sizeof(Oid);
3887 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
3888 dbentry->functions = hash_create("Per-database function",
3889 PGSTAT_FUNCTION_HASH_SIZE,
3890 &hash_ctl,
3891 HASH_ELEM | HASH_BLOBS);
3892 }
3893
3894 /*
3895 * Lookup the hash table entry for the specified database. If no hash
3896 * table entry exists, initialize it, if the create parameter is true.
3897 * Else, return NULL.
3898 */
3899 static PgStat_StatDBEntry *
pgstat_get_db_entry(Oid databaseid,bool create)3900 pgstat_get_db_entry(Oid databaseid, bool create)
3901 {
3902 PgStat_StatDBEntry *result;
3903 bool found;
3904 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3905
3906 /* Lookup or create the hash table entry for this database */
3907 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
3908 &databaseid,
3909 action, &found);
3910
3911 if (!create && !found)
3912 return NULL;
3913
3914 /*
3915 * If not found, initialize the new one. This creates empty hash tables
3916 * for tables and functions, too.
3917 */
3918 if (!found)
3919 reset_dbentry_counters(result);
3920
3921 return result;
3922 }
3923
3924
3925 /*
3926 * Lookup the hash table entry for the specified table. If no hash
3927 * table entry exists, initialize it, if the create parameter is true.
3928 * Else, return NULL.
3929 */
3930 static PgStat_StatTabEntry *
pgstat_get_tab_entry(PgStat_StatDBEntry * dbentry,Oid tableoid,bool create)3931 pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
3932 {
3933 PgStat_StatTabEntry *result;
3934 bool found;
3935 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
3936
3937 /* Lookup or create the hash table entry for this table */
3938 result = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
3939 &tableoid,
3940 action, &found);
3941
3942 if (!create && !found)
3943 return NULL;
3944
3945 /* If not found, initialize the new one. */
3946 if (!found)
3947 {
3948 result->numscans = 0;
3949 result->tuples_returned = 0;
3950 result->tuples_fetched = 0;
3951 result->tuples_inserted = 0;
3952 result->tuples_updated = 0;
3953 result->tuples_deleted = 0;
3954 result->tuples_hot_updated = 0;
3955 result->n_live_tuples = 0;
3956 result->n_dead_tuples = 0;
3957 result->changes_since_analyze = 0;
3958 result->blocks_fetched = 0;
3959 result->blocks_hit = 0;
3960 result->vacuum_timestamp = 0;
3961 result->vacuum_count = 0;
3962 result->autovac_vacuum_timestamp = 0;
3963 result->autovac_vacuum_count = 0;
3964 result->analyze_timestamp = 0;
3965 result->analyze_count = 0;
3966 result->autovac_analyze_timestamp = 0;
3967 result->autovac_analyze_count = 0;
3968 }
3969
3970 return result;
3971 }
3972
3973
3974 /* ----------
3975 * pgstat_write_statsfiles() -
3976 * Write the global statistics file, as well as requested DB files.
3977 *
3978 * 'permanent' specifies writing to the permanent files not temporary ones.
3979 * When true (happens only when the collector is shutting down), also remove
3980 * the temporary files so that backends starting up under a new postmaster
3981 * can't read old data before the new collector is ready.
3982 *
3983 * When 'allDbs' is false, only the requested databases (listed in
3984 * pending_write_requests) will be written; otherwise, all databases
3985 * will be written.
3986 * ----------
3987 */
3988 static void
pgstat_write_statsfiles(bool permanent,bool allDbs)3989 pgstat_write_statsfiles(bool permanent, bool allDbs)
3990 {
3991 HASH_SEQ_STATUS hstat;
3992 PgStat_StatDBEntry *dbentry;
3993 FILE *fpout;
3994 int32 format_id;
3995 const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
3996 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
3997 int rc;
3998
3999 elog(DEBUG2, "writing stats file \"%s\"", statfile);
4000
4001 /*
4002 * Open the statistics temp file to write out the current values.
4003 */
4004 fpout = AllocateFile(tmpfile, PG_BINARY_W);
4005 if (fpout == NULL)
4006 {
4007 ereport(LOG,
4008 (errcode_for_file_access(),
4009 errmsg("could not open temporary statistics file \"%s\": %m",
4010 tmpfile)));
4011 return;
4012 }
4013
4014 /*
4015 * Set the timestamp of the stats file.
4016 */
4017 globalStats.stats_timestamp = GetCurrentTimestamp();
4018
4019 /*
4020 * Write the file header --- currently just a format ID.
4021 */
4022 format_id = PGSTAT_FILE_FORMAT_ID;
4023 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
4024 (void) rc; /* we'll check for error with ferror */
4025
4026 /*
4027 * Write global stats struct
4028 */
4029 rc = fwrite(&globalStats, sizeof(globalStats), 1, fpout);
4030 (void) rc; /* we'll check for error with ferror */
4031
4032 /*
4033 * Write archiver stats struct
4034 */
4035 rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
4036 (void) rc; /* we'll check for error with ferror */
4037
4038 /*
4039 * Walk through the database table.
4040 */
4041 hash_seq_init(&hstat, pgStatDBHash);
4042 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
4043 {
4044 /*
4045 * Write out the table and function stats for this DB into the
4046 * appropriate per-DB stat file, if required.
4047 */
4048 if (allDbs || pgstat_db_requested(dbentry->databaseid))
4049 {
4050 /* Make DB's timestamp consistent with the global stats */
4051 dbentry->stats_timestamp = globalStats.stats_timestamp;
4052
4053 pgstat_write_db_statsfile(dbentry, permanent);
4054 }
4055
4056 /*
4057 * Write out the DB entry. We don't write the tables or functions
4058 * pointers, since they're of no use to any other process.
4059 */
4060 fputc('D', fpout);
4061 rc = fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
4062 (void) rc; /* we'll check for error with ferror */
4063 }
4064
4065 /*
4066 * No more output to be done. Close the temp file and replace the old
4067 * pgstat.stat with it. The ferror() check replaces testing for error
4068 * after each individual fputc or fwrite above.
4069 */
4070 fputc('E', fpout);
4071
4072 if (ferror(fpout))
4073 {
4074 ereport(LOG,
4075 (errcode_for_file_access(),
4076 errmsg("could not write temporary statistics file \"%s\": %m",
4077 tmpfile)));
4078 FreeFile(fpout);
4079 unlink(tmpfile);
4080 }
4081 else if (FreeFile(fpout) < 0)
4082 {
4083 ereport(LOG,
4084 (errcode_for_file_access(),
4085 errmsg("could not close temporary statistics file \"%s\": %m",
4086 tmpfile)));
4087 unlink(tmpfile);
4088 }
4089 else if (rename(tmpfile, statfile) < 0)
4090 {
4091 ereport(LOG,
4092 (errcode_for_file_access(),
4093 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
4094 tmpfile, statfile)));
4095 unlink(tmpfile);
4096 }
4097
4098 if (permanent)
4099 unlink(pgstat_stat_filename);
4100
4101 /*
4102 * Now throw away the list of requests. Note that requests sent after we
4103 * started the write are still waiting on the network socket.
4104 */
4105 list_free(pending_write_requests);
4106 pending_write_requests = NIL;
4107 }
4108
4109 /*
4110 * return the filename for a DB stat file; filename is the output buffer,
4111 * of length len.
4112 */
4113 static void
get_dbstat_filename(bool permanent,bool tempname,Oid databaseid,char * filename,int len)4114 get_dbstat_filename(bool permanent, bool tempname, Oid databaseid,
4115 char *filename, int len)
4116 {
4117 int printed;
4118
4119 /* NB -- pgstat_reset_remove_files knows about the pattern this uses */
4120 printed = snprintf(filename, len, "%s/db_%u.%s",
4121 permanent ? PGSTAT_STAT_PERMANENT_DIRECTORY :
4122 pgstat_stat_directory,
4123 databaseid,
4124 tempname ? "tmp" : "stat");
4125 if (printed >= len)
4126 elog(ERROR, "overlength pgstat path");
4127 }
4128
4129 /* ----------
4130 * pgstat_write_db_statsfile() -
4131 * Write the stat file for a single database.
4132 *
4133 * If writing to the permanent file (happens when the collector is
4134 * shutting down only), remove the temporary file so that backends
4135 * starting up under a new postmaster can't read the old data before
4136 * the new collector is ready.
4137 * ----------
4138 */
4139 static void
pgstat_write_db_statsfile(PgStat_StatDBEntry * dbentry,bool permanent)4140 pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
4141 {
4142 HASH_SEQ_STATUS tstat;
4143 HASH_SEQ_STATUS fstat;
4144 PgStat_StatTabEntry *tabentry;
4145 PgStat_StatFuncEntry *funcentry;
4146 FILE *fpout;
4147 int32 format_id;
4148 Oid dbid = dbentry->databaseid;
4149 int rc;
4150 char tmpfile[MAXPGPATH];
4151 char statfile[MAXPGPATH];
4152
4153 get_dbstat_filename(permanent, true, dbid, tmpfile, MAXPGPATH);
4154 get_dbstat_filename(permanent, false, dbid, statfile, MAXPGPATH);
4155
4156 elog(DEBUG2, "writing stats file \"%s\"", statfile);
4157
4158 /*
4159 * Open the statistics temp file to write out the current values.
4160 */
4161 fpout = AllocateFile(tmpfile, PG_BINARY_W);
4162 if (fpout == NULL)
4163 {
4164 ereport(LOG,
4165 (errcode_for_file_access(),
4166 errmsg("could not open temporary statistics file \"%s\": %m",
4167 tmpfile)));
4168 return;
4169 }
4170
4171 /*
4172 * Write the file header --- currently just a format ID.
4173 */
4174 format_id = PGSTAT_FILE_FORMAT_ID;
4175 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
4176 (void) rc; /* we'll check for error with ferror */
4177
4178 /*
4179 * Walk through the database's access stats per table.
4180 */
4181 hash_seq_init(&tstat, dbentry->tables);
4182 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
4183 {
4184 fputc('T', fpout);
4185 rc = fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
4186 (void) rc; /* we'll check for error with ferror */
4187 }
4188
4189 /*
4190 * Walk through the database's function stats table.
4191 */
4192 hash_seq_init(&fstat, dbentry->functions);
4193 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
4194 {
4195 fputc('F', fpout);
4196 rc = fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
4197 (void) rc; /* we'll check for error with ferror */
4198 }
4199
4200 /*
4201 * No more output to be done. Close the temp file and replace the old
4202 * pgstat.stat with it. The ferror() check replaces testing for error
4203 * after each individual fputc or fwrite above.
4204 */
4205 fputc('E', fpout);
4206
4207 if (ferror(fpout))
4208 {
4209 ereport(LOG,
4210 (errcode_for_file_access(),
4211 errmsg("could not write temporary statistics file \"%s\": %m",
4212 tmpfile)));
4213 FreeFile(fpout);
4214 unlink(tmpfile);
4215 }
4216 else if (FreeFile(fpout) < 0)
4217 {
4218 ereport(LOG,
4219 (errcode_for_file_access(),
4220 errmsg("could not close temporary statistics file \"%s\": %m",
4221 tmpfile)));
4222 unlink(tmpfile);
4223 }
4224 else if (rename(tmpfile, statfile) < 0)
4225 {
4226 ereport(LOG,
4227 (errcode_for_file_access(),
4228 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
4229 tmpfile, statfile)));
4230 unlink(tmpfile);
4231 }
4232
4233 if (permanent)
4234 {
4235 get_dbstat_filename(false, false, dbid, statfile, MAXPGPATH);
4236
4237 elog(DEBUG2, "removing temporary stats file \"%s\"", statfile);
4238 unlink(statfile);
4239 }
4240 }
4241
4242 /* ----------
4243 * pgstat_read_statsfiles() -
4244 *
4245 * Reads in some existing statistics collector files and returns the
4246 * databases hash table that is the top level of the data.
4247 *
4248 * If 'onlydb' is not InvalidOid, it means we only want data for that DB
4249 * plus the shared catalogs ("DB 0"). We'll still populate the DB hash
4250 * table for all databases, but we don't bother even creating table/function
4251 * hash tables for other databases.
4252 *
4253 * 'permanent' specifies reading from the permanent files not temporary ones.
4254 * When true (happens only when the collector is starting up), remove the
4255 * files after reading; the in-memory status is now authoritative, and the
4256 * files would be out of date in case somebody else reads them.
4257 *
4258 * If a 'deep' read is requested, table/function stats are read, otherwise
4259 * the table/function hash tables remain empty.
4260 * ----------
4261 */
4262 static HTAB *
pgstat_read_statsfiles(Oid onlydb,bool permanent,bool deep)4263 pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
4264 {
4265 PgStat_StatDBEntry *dbentry;
4266 PgStat_StatDBEntry dbbuf;
4267 HASHCTL hash_ctl;
4268 HTAB *dbhash;
4269 FILE *fpin;
4270 int32 format_id;
4271 bool found;
4272 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
4273
4274 /*
4275 * The tables will live in pgStatLocalContext.
4276 */
4277 pgstat_setup_memcxt();
4278
4279 /*
4280 * Create the DB hashtable
4281 */
4282 memset(&hash_ctl, 0, sizeof(hash_ctl));
4283 hash_ctl.keysize = sizeof(Oid);
4284 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
4285 hash_ctl.hcxt = pgStatLocalContext;
4286 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
4287 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4288
4289 /*
4290 * Clear out global and archiver statistics so they start from zero in
4291 * case we can't load an existing statsfile.
4292 */
4293 memset(&globalStats, 0, sizeof(globalStats));
4294 memset(&archiverStats, 0, sizeof(archiverStats));
4295
4296 /*
4297 * Set the current timestamp (will be kept only in case we can't load an
4298 * existing statsfile).
4299 */
4300 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
4301 archiverStats.stat_reset_timestamp = globalStats.stat_reset_timestamp;
4302
4303 /*
4304 * Try to open the stats file. If it doesn't exist, the backends simply
4305 * return zero for anything and the collector simply starts from scratch
4306 * with empty counters.
4307 *
4308 * ENOENT is a possibility if the stats collector is not running or has
4309 * not yet written the stats file the first time. Any other failure
4310 * condition is suspicious.
4311 */
4312 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
4313 {
4314 if (errno != ENOENT)
4315 ereport(pgStatRunningInCollector ? LOG : WARNING,
4316 (errcode_for_file_access(),
4317 errmsg("could not open statistics file \"%s\": %m",
4318 statfile)));
4319 return dbhash;
4320 }
4321
4322 /*
4323 * Verify it's of the expected format.
4324 */
4325 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
4326 format_id != PGSTAT_FILE_FORMAT_ID)
4327 {
4328 ereport(pgStatRunningInCollector ? LOG : WARNING,
4329 (errmsg("corrupted statistics file \"%s\"", statfile)));
4330 goto done;
4331 }
4332
4333 /*
4334 * Read global stats struct
4335 */
4336 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
4337 {
4338 ereport(pgStatRunningInCollector ? LOG : WARNING,
4339 (errmsg("corrupted statistics file \"%s\"", statfile)));
4340 memset(&globalStats, 0, sizeof(globalStats));
4341 goto done;
4342 }
4343
4344 /*
4345 * In the collector, disregard the timestamp we read from the permanent
4346 * stats file; we should be willing to write a temp stats file immediately
4347 * upon the first request from any backend. This only matters if the old
4348 * file's timestamp is less than PGSTAT_STAT_INTERVAL ago, but that's not
4349 * an unusual scenario.
4350 */
4351 if (pgStatRunningInCollector)
4352 globalStats.stats_timestamp = 0;
4353
4354 /*
4355 * Read archiver stats struct
4356 */
4357 if (fread(&archiverStats, 1, sizeof(archiverStats), fpin) != sizeof(archiverStats))
4358 {
4359 ereport(pgStatRunningInCollector ? LOG : WARNING,
4360 (errmsg("corrupted statistics file \"%s\"", statfile)));
4361 memset(&archiverStats, 0, sizeof(archiverStats));
4362 goto done;
4363 }
4364
4365 /*
4366 * We found an existing collector stats file. Read it and put all the
4367 * hashtable entries into place.
4368 */
4369 for (;;)
4370 {
4371 switch (fgetc(fpin))
4372 {
4373 /*
4374 * 'D' A PgStat_StatDBEntry struct describing a database
4375 * follows.
4376 */
4377 case 'D':
4378 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
4379 fpin) != offsetof(PgStat_StatDBEntry, tables))
4380 {
4381 ereport(pgStatRunningInCollector ? LOG : WARNING,
4382 (errmsg("corrupted statistics file \"%s\"",
4383 statfile)));
4384 goto done;
4385 }
4386
4387 /*
4388 * Add to the DB hash
4389 */
4390 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
4391 (void *) &dbbuf.databaseid,
4392 HASH_ENTER,
4393 &found);
4394 if (found)
4395 {
4396 ereport(pgStatRunningInCollector ? LOG : WARNING,
4397 (errmsg("corrupted statistics file \"%s\"",
4398 statfile)));
4399 goto done;
4400 }
4401
4402 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
4403 dbentry->tables = NULL;
4404 dbentry->functions = NULL;
4405
4406 /*
4407 * In the collector, disregard the timestamp we read from the
4408 * permanent stats file; we should be willing to write a temp
4409 * stats file immediately upon the first request from any
4410 * backend.
4411 */
4412 if (pgStatRunningInCollector)
4413 dbentry->stats_timestamp = 0;
4414
4415 /*
4416 * Don't create tables/functions hashtables for uninteresting
4417 * databases.
4418 */
4419 if (onlydb != InvalidOid)
4420 {
4421 if (dbbuf.databaseid != onlydb &&
4422 dbbuf.databaseid != InvalidOid)
4423 break;
4424 }
4425
4426 memset(&hash_ctl, 0, sizeof(hash_ctl));
4427 hash_ctl.keysize = sizeof(Oid);
4428 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
4429 hash_ctl.hcxt = pgStatLocalContext;
4430 dbentry->tables = hash_create("Per-database table",
4431 PGSTAT_TAB_HASH_SIZE,
4432 &hash_ctl,
4433 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4434
4435 hash_ctl.keysize = sizeof(Oid);
4436 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
4437 hash_ctl.hcxt = pgStatLocalContext;
4438 dbentry->functions = hash_create("Per-database function",
4439 PGSTAT_FUNCTION_HASH_SIZE,
4440 &hash_ctl,
4441 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4442
4443 /*
4444 * If requested, read the data from the database-specific
4445 * file. Otherwise we just leave the hashtables empty.
4446 */
4447 if (deep)
4448 pgstat_read_db_statsfile(dbentry->databaseid,
4449 dbentry->tables,
4450 dbentry->functions,
4451 permanent);
4452
4453 break;
4454
4455 case 'E':
4456 goto done;
4457
4458 default:
4459 ereport(pgStatRunningInCollector ? LOG : WARNING,
4460 (errmsg("corrupted statistics file \"%s\"",
4461 statfile)));
4462 goto done;
4463 }
4464 }
4465
4466 done:
4467 FreeFile(fpin);
4468
4469 /* If requested to read the permanent file, also get rid of it. */
4470 if (permanent)
4471 {
4472 elog(DEBUG2, "removing permanent stats file \"%s\"", statfile);
4473 unlink(statfile);
4474 }
4475
4476 return dbhash;
4477 }
4478
4479
4480 /* ----------
4481 * pgstat_read_db_statsfile() -
4482 *
4483 * Reads in the existing statistics collector file for the given database,
4484 * filling the passed-in tables and functions hash tables.
4485 *
4486 * As in pgstat_read_statsfiles, if the permanent file is requested, it is
4487 * removed after reading.
4488 *
4489 * Note: this code has the ability to skip storing per-table or per-function
4490 * data, if NULL is passed for the corresponding hashtable. That's not used
4491 * at the moment though.
4492 * ----------
4493 */
4494 static void
pgstat_read_db_statsfile(Oid databaseid,HTAB * tabhash,HTAB * funchash,bool permanent)4495 pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash,
4496 bool permanent)
4497 {
4498 PgStat_StatTabEntry *tabentry;
4499 PgStat_StatTabEntry tabbuf;
4500 PgStat_StatFuncEntry funcbuf;
4501 PgStat_StatFuncEntry *funcentry;
4502 FILE *fpin;
4503 int32 format_id;
4504 bool found;
4505 char statfile[MAXPGPATH];
4506
4507 get_dbstat_filename(permanent, false, databaseid, statfile, MAXPGPATH);
4508
4509 /*
4510 * Try to open the stats file. If it doesn't exist, the backends simply
4511 * return zero for anything and the collector simply starts from scratch
4512 * with empty counters.
4513 *
4514 * ENOENT is a possibility if the stats collector is not running or has
4515 * not yet written the stats file the first time. Any other failure
4516 * condition is suspicious.
4517 */
4518 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
4519 {
4520 if (errno != ENOENT)
4521 ereport(pgStatRunningInCollector ? LOG : WARNING,
4522 (errcode_for_file_access(),
4523 errmsg("could not open statistics file \"%s\": %m",
4524 statfile)));
4525 return;
4526 }
4527
4528 /*
4529 * Verify it's of the expected format.
4530 */
4531 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
4532 format_id != PGSTAT_FILE_FORMAT_ID)
4533 {
4534 ereport(pgStatRunningInCollector ? LOG : WARNING,
4535 (errmsg("corrupted statistics file \"%s\"", statfile)));
4536 goto done;
4537 }
4538
4539 /*
4540 * We found an existing collector stats file. Read it and put all the
4541 * hashtable entries into place.
4542 */
4543 for (;;)
4544 {
4545 switch (fgetc(fpin))
4546 {
4547 /*
4548 * 'T' A PgStat_StatTabEntry follows.
4549 */
4550 case 'T':
4551 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
4552 fpin) != sizeof(PgStat_StatTabEntry))
4553 {
4554 ereport(pgStatRunningInCollector ? LOG : WARNING,
4555 (errmsg("corrupted statistics file \"%s\"",
4556 statfile)));
4557 goto done;
4558 }
4559
4560 /*
4561 * Skip if table data not wanted.
4562 */
4563 if (tabhash == NULL)
4564 break;
4565
4566 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
4567 (void *) &tabbuf.tableid,
4568 HASH_ENTER, &found);
4569
4570 if (found)
4571 {
4572 ereport(pgStatRunningInCollector ? LOG : WARNING,
4573 (errmsg("corrupted statistics file \"%s\"",
4574 statfile)));
4575 goto done;
4576 }
4577
4578 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
4579 break;
4580
4581 /*
4582 * 'F' A PgStat_StatFuncEntry follows.
4583 */
4584 case 'F':
4585 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
4586 fpin) != sizeof(PgStat_StatFuncEntry))
4587 {
4588 ereport(pgStatRunningInCollector ? LOG : WARNING,
4589 (errmsg("corrupted statistics file \"%s\"",
4590 statfile)));
4591 goto done;
4592 }
4593
4594 /*
4595 * Skip if function data not wanted.
4596 */
4597 if (funchash == NULL)
4598 break;
4599
4600 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
4601 (void *) &funcbuf.functionid,
4602 HASH_ENTER, &found);
4603
4604 if (found)
4605 {
4606 ereport(pgStatRunningInCollector ? LOG : WARNING,
4607 (errmsg("corrupted statistics file \"%s\"",
4608 statfile)));
4609 goto done;
4610 }
4611
4612 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
4613 break;
4614
4615 /*
4616 * 'E' The EOF marker of a complete stats file.
4617 */
4618 case 'E':
4619 goto done;
4620
4621 default:
4622 ereport(pgStatRunningInCollector ? LOG : WARNING,
4623 (errmsg("corrupted statistics file \"%s\"",
4624 statfile)));
4625 goto done;
4626 }
4627 }
4628
4629 done:
4630 FreeFile(fpin);
4631
4632 if (permanent)
4633 {
4634 elog(DEBUG2, "removing permanent stats file \"%s\"", statfile);
4635 unlink(statfile);
4636 }
4637 }
4638
4639 /* ----------
4640 * pgstat_read_db_statsfile_timestamp() -
4641 *
4642 * Attempt to determine the timestamp of the last db statfile write.
4643 * Returns TRUE if successful; the timestamp is stored in *ts.
4644 *
4645 * This needs to be careful about handling databases for which no stats file
4646 * exists, such as databases without a stat entry or those not yet written:
4647 *
4648 * - if there's a database entry in the global file, return the corresponding
4649 * stats_timestamp value.
4650 *
4651 * - if there's no db stat entry (e.g. for a new or inactive database),
4652 * there's no stats_timestamp value, but also nothing to write so we return
4653 * the timestamp of the global statfile.
4654 * ----------
4655 */
4656 static bool
pgstat_read_db_statsfile_timestamp(Oid databaseid,bool permanent,TimestampTz * ts)4657 pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
4658 TimestampTz *ts)
4659 {
4660 PgStat_StatDBEntry dbentry;
4661 PgStat_GlobalStats myGlobalStats;
4662 PgStat_ArchiverStats myArchiverStats;
4663 FILE *fpin;
4664 int32 format_id;
4665 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
4666
4667 /*
4668 * Try to open the stats file. As above, anything but ENOENT is worthy of
4669 * complaining about.
4670 */
4671 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
4672 {
4673 if (errno != ENOENT)
4674 ereport(pgStatRunningInCollector ? LOG : WARNING,
4675 (errcode_for_file_access(),
4676 errmsg("could not open statistics file \"%s\": %m",
4677 statfile)));
4678 return false;
4679 }
4680
4681 /*
4682 * Verify it's of the expected format.
4683 */
4684 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
4685 format_id != PGSTAT_FILE_FORMAT_ID)
4686 {
4687 ereport(pgStatRunningInCollector ? LOG : WARNING,
4688 (errmsg("corrupted statistics file \"%s\"", statfile)));
4689 FreeFile(fpin);
4690 return false;
4691 }
4692
4693 /*
4694 * Read global stats struct
4695 */
4696 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats),
4697 fpin) != sizeof(myGlobalStats))
4698 {
4699 ereport(pgStatRunningInCollector ? LOG : WARNING,
4700 (errmsg("corrupted statistics file \"%s\"", statfile)));
4701 FreeFile(fpin);
4702 return false;
4703 }
4704
4705 /*
4706 * Read archiver stats struct
4707 */
4708 if (fread(&myArchiverStats, 1, sizeof(myArchiverStats),
4709 fpin) != sizeof(myArchiverStats))
4710 {
4711 ereport(pgStatRunningInCollector ? LOG : WARNING,
4712 (errmsg("corrupted statistics file \"%s\"", statfile)));
4713 FreeFile(fpin);
4714 return false;
4715 }
4716
4717 /* By default, we're going to return the timestamp of the global file. */
4718 *ts = myGlobalStats.stats_timestamp;
4719
4720 /*
4721 * We found an existing collector stats file. Read it and look for a
4722 * record for the requested database. If found, use its timestamp.
4723 */
4724 for (;;)
4725 {
4726 switch (fgetc(fpin))
4727 {
4728 /*
4729 * 'D' A PgStat_StatDBEntry struct describing a database
4730 * follows.
4731 */
4732 case 'D':
4733 if (fread(&dbentry, 1, offsetof(PgStat_StatDBEntry, tables),
4734 fpin) != offsetof(PgStat_StatDBEntry, tables))
4735 {
4736 ereport(pgStatRunningInCollector ? LOG : WARNING,
4737 (errmsg("corrupted statistics file \"%s\"",
4738 statfile)));
4739 goto done;
4740 }
4741
4742 /*
4743 * If this is the DB we're looking for, save its timestamp and
4744 * we're done.
4745 */
4746 if (dbentry.databaseid == databaseid)
4747 {
4748 *ts = dbentry.stats_timestamp;
4749 goto done;
4750 }
4751
4752 break;
4753
4754 case 'E':
4755 goto done;
4756
4757 default:
4758 ereport(pgStatRunningInCollector ? LOG : WARNING,
4759 (errmsg("corrupted statistics file \"%s\"",
4760 statfile)));
4761 goto done;
4762 }
4763 }
4764
4765 done:
4766 FreeFile(fpin);
4767 return true;
4768 }
4769
4770 /*
4771 * If not already done, read the statistics collector stats file into
4772 * some hash tables. The results will be kept until pgstat_clear_snapshot()
4773 * is called (typically, at end of transaction).
4774 */
4775 static void
backend_read_statsfile(void)4776 backend_read_statsfile(void)
4777 {
4778 TimestampTz min_ts = 0;
4779 TimestampTz ref_ts = 0;
4780 Oid inquiry_db;
4781 int count;
4782
4783 /* already read it? */
4784 if (pgStatDBHash)
4785 return;
4786 Assert(!pgStatRunningInCollector);
4787
4788 /*
4789 * In a normal backend, we check staleness of the data for our own DB, and
4790 * so we send MyDatabaseId in inquiry messages. In the autovac launcher,
4791 * check staleness of the shared-catalog data, and send InvalidOid in
4792 * inquiry messages so as not to force writing unnecessary data.
4793 */
4794 if (IsAutoVacuumLauncherProcess())
4795 inquiry_db = InvalidOid;
4796 else
4797 inquiry_db = MyDatabaseId;
4798
4799 /*
4800 * Loop until fresh enough stats file is available or we ran out of time.
4801 * The stats inquiry message is sent repeatedly in case collector drops
4802 * it; but not every single time, as that just swamps the collector.
4803 */
4804 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
4805 {
4806 bool ok;
4807 TimestampTz file_ts = 0;
4808 TimestampTz cur_ts;
4809
4810 CHECK_FOR_INTERRUPTS();
4811
4812 ok = pgstat_read_db_statsfile_timestamp(inquiry_db, false, &file_ts);
4813
4814 cur_ts = GetCurrentTimestamp();
4815 /* Calculate min acceptable timestamp, if we didn't already */
4816 if (count == 0 || cur_ts < ref_ts)
4817 {
4818 /*
4819 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL
4820 * msec before now. This indirectly ensures that the collector
4821 * needn't write the file more often than PGSTAT_STAT_INTERVAL. In
4822 * an autovacuum worker, however, we want a lower delay to avoid
4823 * using stale data, so we use PGSTAT_RETRY_DELAY (since the
4824 * number of workers is low, this shouldn't be a problem).
4825 *
4826 * We don't recompute min_ts after sleeping, except in the
4827 * unlikely case that cur_ts went backwards. So we might end up
4828 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In
4829 * practice that shouldn't happen, though, as long as the sleep
4830 * time is less than PGSTAT_STAT_INTERVAL; and we don't want to
4831 * tell the collector that our cutoff time is less than what we'd
4832 * actually accept.
4833 */
4834 ref_ts = cur_ts;
4835 if (IsAutoVacuumWorkerProcess())
4836 min_ts = TimestampTzPlusMilliseconds(ref_ts,
4837 -PGSTAT_RETRY_DELAY);
4838 else
4839 min_ts = TimestampTzPlusMilliseconds(ref_ts,
4840 -PGSTAT_STAT_INTERVAL);
4841 }
4842
4843 /*
4844 * If the file timestamp is actually newer than cur_ts, we must have
4845 * had a clock glitch (system time went backwards) or there is clock
4846 * skew between our processor and the stats collector's processor.
4847 * Accept the file, but send an inquiry message anyway to make
4848 * pgstat_recv_inquiry do a sanity check on the collector's time.
4849 */
4850 if (ok && file_ts > cur_ts)
4851 {
4852 /*
4853 * A small amount of clock skew between processors isn't terribly
4854 * surprising, but a large difference is worth logging. We
4855 * arbitrarily define "large" as 1000 msec.
4856 */
4857 if (file_ts >= TimestampTzPlusMilliseconds(cur_ts, 1000))
4858 {
4859 char *filetime;
4860 char *mytime;
4861
4862 /* Copy because timestamptz_to_str returns a static buffer */
4863 filetime = pstrdup(timestamptz_to_str(file_ts));
4864 mytime = pstrdup(timestamptz_to_str(cur_ts));
4865 elog(LOG, "stats collector's time %s is later than backend local time %s",
4866 filetime, mytime);
4867 pfree(filetime);
4868 pfree(mytime);
4869 }
4870
4871 pgstat_send_inquiry(cur_ts, min_ts, inquiry_db);
4872 break;
4873 }
4874
4875 /* Normal acceptance case: file is not older than cutoff time */
4876 if (ok && file_ts >= min_ts)
4877 break;
4878
4879 /* Not there or too old, so kick the collector and wait a bit */
4880 if ((count % PGSTAT_INQ_LOOP_COUNT) == 0)
4881 pgstat_send_inquiry(cur_ts, min_ts, inquiry_db);
4882
4883 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
4884 }
4885
4886 if (count >= PGSTAT_POLL_LOOP_COUNT)
4887 ereport(LOG,
4888 (errmsg("using stale statistics instead of current ones "
4889 "because stats collector is not responding")));
4890
4891 /*
4892 * Autovacuum launcher wants stats about all databases, but a shallow read
4893 * is sufficient. Regular backends want a deep read for just the tables
4894 * they can see (MyDatabaseId + shared catalogs).
4895 */
4896 if (IsAutoVacuumLauncherProcess())
4897 pgStatDBHash = pgstat_read_statsfiles(InvalidOid, false, false);
4898 else
4899 pgStatDBHash = pgstat_read_statsfiles(MyDatabaseId, false, true);
4900 }
4901
4902
4903 /* ----------
4904 * pgstat_setup_memcxt() -
4905 *
4906 * Create pgStatLocalContext, if not already done.
4907 * ----------
4908 */
4909 static void
pgstat_setup_memcxt(void)4910 pgstat_setup_memcxt(void)
4911 {
4912 if (!pgStatLocalContext)
4913 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
4914 "Statistics snapshot",
4915 ALLOCSET_SMALL_SIZES);
4916 }
4917
4918
4919 /* ----------
4920 * pgstat_clear_snapshot() -
4921 *
4922 * Discard any data collected in the current transaction. Any subsequent
4923 * request will cause new snapshots to be read.
4924 *
4925 * This is also invoked during transaction commit or abort to discard
4926 * the no-longer-wanted snapshot.
4927 * ----------
4928 */
4929 void
pgstat_clear_snapshot(void)4930 pgstat_clear_snapshot(void)
4931 {
4932 /* Release memory, if any was allocated */
4933 if (pgStatLocalContext)
4934 MemoryContextDelete(pgStatLocalContext);
4935
4936 /* Reset variables */
4937 pgStatLocalContext = NULL;
4938 pgStatDBHash = NULL;
4939 localBackendStatusTable = NULL;
4940 localNumBackends = 0;
4941 }
4942
4943
4944 /* ----------
4945 * pgstat_recv_inquiry() -
4946 *
4947 * Process stat inquiry requests.
4948 * ----------
4949 */
4950 static void
pgstat_recv_inquiry(PgStat_MsgInquiry * msg,int len)4951 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
4952 {
4953 PgStat_StatDBEntry *dbentry;
4954
4955 elog(DEBUG2, "received inquiry for database %u", msg->databaseid);
4956
4957 /*
4958 * If there's already a write request for this DB, there's nothing to do.
4959 *
4960 * Note that if a request is found, we return early and skip the below
4961 * check for clock skew. This is okay, since the only way for a DB
4962 * request to be present in the list is that we have been here since the
4963 * last write round. It seems sufficient to check for clock skew once per
4964 * write round.
4965 */
4966 if (list_member_oid(pending_write_requests, msg->databaseid))
4967 return;
4968
4969 /*
4970 * Check to see if we last wrote this database at a time >= the requested
4971 * cutoff time. If so, this is a stale request that was generated before
4972 * we updated the DB file, and we don't need to do so again.
4973 *
4974 * If the requestor's local clock time is older than stats_timestamp, we
4975 * should suspect a clock glitch, ie system time going backwards; though
4976 * the more likely explanation is just delayed message receipt. It is
4977 * worth expending a GetCurrentTimestamp call to be sure, since a large
4978 * retreat in the system clock reading could otherwise cause us to neglect
4979 * to update the stats file for a long time.
4980 */
4981 dbentry = pgstat_get_db_entry(msg->databaseid, false);
4982 if (dbentry == NULL)
4983 {
4984 /*
4985 * We have no data for this DB. Enter a write request anyway so that
4986 * the global stats will get updated. This is needed to prevent
4987 * backend_read_statsfile from waiting for data that we cannot supply,
4988 * in the case of a new DB that nobody has yet reported any stats for.
4989 * See the behavior of pgstat_read_db_statsfile_timestamp.
4990 */
4991 }
4992 else if (msg->clock_time < dbentry->stats_timestamp)
4993 {
4994 TimestampTz cur_ts = GetCurrentTimestamp();
4995
4996 if (cur_ts < dbentry->stats_timestamp)
4997 {
4998 /*
4999 * Sure enough, time went backwards. Force a new stats file write
5000 * to get back in sync; but first, log a complaint.
5001 */
5002 char *writetime;
5003 char *mytime;
5004
5005 /* Copy because timestamptz_to_str returns a static buffer */
5006 writetime = pstrdup(timestamptz_to_str(dbentry->stats_timestamp));
5007 mytime = pstrdup(timestamptz_to_str(cur_ts));
5008 elog(LOG,
5009 "stats_timestamp %s is later than collector's time %s for database %u",
5010 writetime, mytime, dbentry->databaseid);
5011 pfree(writetime);
5012 pfree(mytime);
5013 }
5014 else
5015 {
5016 /*
5017 * Nope, it's just an old request. Assuming msg's clock_time is
5018 * >= its cutoff_time, it must be stale, so we can ignore it.
5019 */
5020 return;
5021 }
5022 }
5023 else if (msg->cutoff_time <= dbentry->stats_timestamp)
5024 {
5025 /* Stale request, ignore it */
5026 return;
5027 }
5028
5029 /*
5030 * We need to write this DB, so create a request.
5031 */
5032 pending_write_requests = lappend_oid(pending_write_requests,
5033 msg->databaseid);
5034 }
5035
5036
5037 /* ----------
5038 * pgstat_recv_tabstat() -
5039 *
5040 * Count what the backend has done.
5041 * ----------
5042 */
5043 static void
pgstat_recv_tabstat(PgStat_MsgTabstat * msg,int len)5044 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
5045 {
5046 PgStat_StatDBEntry *dbentry;
5047 PgStat_StatTabEntry *tabentry;
5048 int i;
5049 bool found;
5050
5051 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5052
5053 /*
5054 * Update database-wide stats.
5055 */
5056 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
5057 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
5058 dbentry->n_block_read_time += msg->m_block_read_time;
5059 dbentry->n_block_write_time += msg->m_block_write_time;
5060
5061 /*
5062 * Process all table entries in the message.
5063 */
5064 for (i = 0; i < msg->m_nentries; i++)
5065 {
5066 PgStat_TableEntry *tabmsg = &(msg->m_entry[i]);
5067
5068 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
5069 (void *) &(tabmsg->t_id),
5070 HASH_ENTER, &found);
5071
5072 if (!found)
5073 {
5074 /*
5075 * If it's a new table entry, initialize counters to the values we
5076 * just got.
5077 */
5078 tabentry->numscans = tabmsg->t_counts.t_numscans;
5079 tabentry->tuples_returned = tabmsg->t_counts.t_tuples_returned;
5080 tabentry->tuples_fetched = tabmsg->t_counts.t_tuples_fetched;
5081 tabentry->tuples_inserted = tabmsg->t_counts.t_tuples_inserted;
5082 tabentry->tuples_updated = tabmsg->t_counts.t_tuples_updated;
5083 tabentry->tuples_deleted = tabmsg->t_counts.t_tuples_deleted;
5084 tabentry->tuples_hot_updated = tabmsg->t_counts.t_tuples_hot_updated;
5085 tabentry->n_live_tuples = tabmsg->t_counts.t_delta_live_tuples;
5086 tabentry->n_dead_tuples = tabmsg->t_counts.t_delta_dead_tuples;
5087 tabentry->changes_since_analyze = tabmsg->t_counts.t_changed_tuples;
5088 tabentry->blocks_fetched = tabmsg->t_counts.t_blocks_fetched;
5089 tabentry->blocks_hit = tabmsg->t_counts.t_blocks_hit;
5090
5091 tabentry->vacuum_timestamp = 0;
5092 tabentry->vacuum_count = 0;
5093 tabentry->autovac_vacuum_timestamp = 0;
5094 tabentry->autovac_vacuum_count = 0;
5095 tabentry->analyze_timestamp = 0;
5096 tabentry->analyze_count = 0;
5097 tabentry->autovac_analyze_timestamp = 0;
5098 tabentry->autovac_analyze_count = 0;
5099 }
5100 else
5101 {
5102 /*
5103 * Otherwise add the values to the existing entry.
5104 */
5105 tabentry->numscans += tabmsg->t_counts.t_numscans;
5106 tabentry->tuples_returned += tabmsg->t_counts.t_tuples_returned;
5107 tabentry->tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
5108 tabentry->tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
5109 tabentry->tuples_updated += tabmsg->t_counts.t_tuples_updated;
5110 tabentry->tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
5111 tabentry->tuples_hot_updated += tabmsg->t_counts.t_tuples_hot_updated;
5112 /* If table was truncated, first reset the live/dead counters */
5113 if (tabmsg->t_counts.t_truncated)
5114 {
5115 tabentry->n_live_tuples = 0;
5116 tabentry->n_dead_tuples = 0;
5117 }
5118 tabentry->n_live_tuples += tabmsg->t_counts.t_delta_live_tuples;
5119 tabentry->n_dead_tuples += tabmsg->t_counts.t_delta_dead_tuples;
5120 tabentry->changes_since_analyze += tabmsg->t_counts.t_changed_tuples;
5121 tabentry->blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
5122 tabentry->blocks_hit += tabmsg->t_counts.t_blocks_hit;
5123 }
5124
5125 /* Clamp n_live_tuples in case of negative delta_live_tuples */
5126 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
5127 /* Likewise for n_dead_tuples */
5128 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
5129
5130 /*
5131 * Add per-table stats to the per-database entry, too.
5132 */
5133 dbentry->n_tuples_returned += tabmsg->t_counts.t_tuples_returned;
5134 dbentry->n_tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
5135 dbentry->n_tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
5136 dbentry->n_tuples_updated += tabmsg->t_counts.t_tuples_updated;
5137 dbentry->n_tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
5138 dbentry->n_blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
5139 dbentry->n_blocks_hit += tabmsg->t_counts.t_blocks_hit;
5140 }
5141 }
5142
5143
5144 /* ----------
5145 * pgstat_recv_tabpurge() -
5146 *
5147 * Arrange for dead table removal.
5148 * ----------
5149 */
5150 static void
pgstat_recv_tabpurge(PgStat_MsgTabpurge * msg,int len)5151 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
5152 {
5153 PgStat_StatDBEntry *dbentry;
5154 int i;
5155
5156 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5157
5158 /*
5159 * No need to purge if we don't even know the database.
5160 */
5161 if (!dbentry || !dbentry->tables)
5162 return;
5163
5164 /*
5165 * Process all table entries in the message.
5166 */
5167 for (i = 0; i < msg->m_nentries; i++)
5168 {
5169 /* Remove from hashtable if present; we don't care if it's not. */
5170 (void) hash_search(dbentry->tables,
5171 (void *) &(msg->m_tableid[i]),
5172 HASH_REMOVE, NULL);
5173 }
5174 }
5175
5176
5177 /* ----------
5178 * pgstat_recv_dropdb() -
5179 *
5180 * Arrange for dead database removal
5181 * ----------
5182 */
5183 static void
pgstat_recv_dropdb(PgStat_MsgDropdb * msg,int len)5184 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
5185 {
5186 Oid dbid = msg->m_databaseid;
5187 PgStat_StatDBEntry *dbentry;
5188
5189 /*
5190 * Lookup the database in the hashtable.
5191 */
5192 dbentry = pgstat_get_db_entry(dbid, false);
5193
5194 /*
5195 * If found, remove it (along with the db statfile).
5196 */
5197 if (dbentry)
5198 {
5199 char statfile[MAXPGPATH];
5200
5201 get_dbstat_filename(false, false, dbid, statfile, MAXPGPATH);
5202
5203 elog(DEBUG2, "removing stats file \"%s\"", statfile);
5204 unlink(statfile);
5205
5206 if (dbentry->tables != NULL)
5207 hash_destroy(dbentry->tables);
5208 if (dbentry->functions != NULL)
5209 hash_destroy(dbentry->functions);
5210
5211 if (hash_search(pgStatDBHash,
5212 (void *) &dbid,
5213 HASH_REMOVE, NULL) == NULL)
5214 ereport(ERROR,
5215 (errmsg("database hash table corrupted during cleanup --- abort")));
5216 }
5217 }
5218
5219
5220 /* ----------
5221 * pgstat_recv_resetcounter() -
5222 *
5223 * Reset the statistics for the specified database.
5224 * ----------
5225 */
5226 static void
pgstat_recv_resetcounter(PgStat_MsgResetcounter * msg,int len)5227 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
5228 {
5229 PgStat_StatDBEntry *dbentry;
5230
5231 /*
5232 * Lookup the database in the hashtable. Nothing to do if not there.
5233 */
5234 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5235
5236 if (!dbentry)
5237 return;
5238
5239 /*
5240 * We simply throw away all the database's table entries by recreating a
5241 * new hash table for them.
5242 */
5243 if (dbentry->tables != NULL)
5244 hash_destroy(dbentry->tables);
5245 if (dbentry->functions != NULL)
5246 hash_destroy(dbentry->functions);
5247
5248 dbentry->tables = NULL;
5249 dbentry->functions = NULL;
5250
5251 /*
5252 * Reset database-level stats, too. This creates empty hash tables for
5253 * tables and functions.
5254 */
5255 reset_dbentry_counters(dbentry);
5256 }
5257
5258 /* ----------
5259 * pgstat_recv_resetshared() -
5260 *
5261 * Reset some shared statistics of the cluster.
5262 * ----------
5263 */
5264 static void
pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter * msg,int len)5265 pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
5266 {
5267 if (msg->m_resettarget == RESET_BGWRITER)
5268 {
5269 /* Reset the global background writer statistics for the cluster. */
5270 memset(&globalStats, 0, sizeof(globalStats));
5271 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
5272 }
5273 else if (msg->m_resettarget == RESET_ARCHIVER)
5274 {
5275 /* Reset the archiver statistics for the cluster. */
5276 memset(&archiverStats, 0, sizeof(archiverStats));
5277 archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
5278 }
5279
5280 /*
5281 * Presumably the sender of this message validated the target, don't
5282 * complain here if it's not valid
5283 */
5284 }
5285
5286 /* ----------
5287 * pgstat_recv_resetsinglecounter() -
5288 *
5289 * Reset a statistics for a single object
5290 * ----------
5291 */
5292 static void
pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter * msg,int len)5293 pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len)
5294 {
5295 PgStat_StatDBEntry *dbentry;
5296
5297 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5298
5299 if (!dbentry)
5300 return;
5301
5302 /* Set the reset timestamp for the whole database */
5303 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
5304
5305 /* Remove object if it exists, ignore it if not */
5306 if (msg->m_resettype == RESET_TABLE)
5307 (void) hash_search(dbentry->tables, (void *) &(msg->m_objectid),
5308 HASH_REMOVE, NULL);
5309 else if (msg->m_resettype == RESET_FUNCTION)
5310 (void) hash_search(dbentry->functions, (void *) &(msg->m_objectid),
5311 HASH_REMOVE, NULL);
5312 }
5313
5314 /* ----------
5315 * pgstat_recv_autovac() -
5316 *
5317 * Process an autovacuum signalling message.
5318 * ----------
5319 */
5320 static void
pgstat_recv_autovac(PgStat_MsgAutovacStart * msg,int len)5321 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
5322 {
5323 PgStat_StatDBEntry *dbentry;
5324
5325 /*
5326 * Store the last autovacuum time in the database's hashtable entry.
5327 */
5328 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5329
5330 dbentry->last_autovac_time = msg->m_start_time;
5331 }
5332
5333 /* ----------
5334 * pgstat_recv_vacuum() -
5335 *
5336 * Process a VACUUM message.
5337 * ----------
5338 */
5339 static void
pgstat_recv_vacuum(PgStat_MsgVacuum * msg,int len)5340 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
5341 {
5342 PgStat_StatDBEntry *dbentry;
5343 PgStat_StatTabEntry *tabentry;
5344
5345 /*
5346 * Store the data in the table's hashtable entry.
5347 */
5348 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5349
5350 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
5351
5352 tabentry->n_live_tuples = msg->m_live_tuples;
5353 tabentry->n_dead_tuples = msg->m_dead_tuples;
5354
5355 if (msg->m_autovacuum)
5356 {
5357 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
5358 tabentry->autovac_vacuum_count++;
5359 }
5360 else
5361 {
5362 tabentry->vacuum_timestamp = msg->m_vacuumtime;
5363 tabentry->vacuum_count++;
5364 }
5365 }
5366
5367 /* ----------
5368 * pgstat_recv_analyze() -
5369 *
5370 * Process an ANALYZE message.
5371 * ----------
5372 */
5373 static void
pgstat_recv_analyze(PgStat_MsgAnalyze * msg,int len)5374 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
5375 {
5376 PgStat_StatDBEntry *dbentry;
5377 PgStat_StatTabEntry *tabentry;
5378
5379 /*
5380 * Store the data in the table's hashtable entry.
5381 */
5382 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5383
5384 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
5385
5386 tabentry->n_live_tuples = msg->m_live_tuples;
5387 tabentry->n_dead_tuples = msg->m_dead_tuples;
5388
5389 /*
5390 * If commanded, reset changes_since_analyze to zero. This forgets any
5391 * changes that were committed while the ANALYZE was in progress, but we
5392 * have no good way to estimate how many of those there were.
5393 */
5394 if (msg->m_resetcounter)
5395 tabentry->changes_since_analyze = 0;
5396
5397 if (msg->m_autovacuum)
5398 {
5399 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
5400 tabentry->autovac_analyze_count++;
5401 }
5402 else
5403 {
5404 tabentry->analyze_timestamp = msg->m_analyzetime;
5405 tabentry->analyze_count++;
5406 }
5407 }
5408
5409
5410 /* ----------
5411 * pgstat_recv_archiver() -
5412 *
5413 * Process a ARCHIVER message.
5414 * ----------
5415 */
5416 static void
pgstat_recv_archiver(PgStat_MsgArchiver * msg,int len)5417 pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len)
5418 {
5419 if (msg->m_failed)
5420 {
5421 /* Failed archival attempt */
5422 ++archiverStats.failed_count;
5423 memcpy(archiverStats.last_failed_wal, msg->m_xlog,
5424 sizeof(archiverStats.last_failed_wal));
5425 archiverStats.last_failed_timestamp = msg->m_timestamp;
5426 }
5427 else
5428 {
5429 /* Successful archival operation */
5430 ++archiverStats.archived_count;
5431 memcpy(archiverStats.last_archived_wal, msg->m_xlog,
5432 sizeof(archiverStats.last_archived_wal));
5433 archiverStats.last_archived_timestamp = msg->m_timestamp;
5434 }
5435 }
5436
5437 /* ----------
5438 * pgstat_recv_bgwriter() -
5439 *
5440 * Process a BGWRITER message.
5441 * ----------
5442 */
5443 static void
pgstat_recv_bgwriter(PgStat_MsgBgWriter * msg,int len)5444 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
5445 {
5446 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
5447 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
5448 globalStats.checkpoint_write_time += msg->m_checkpoint_write_time;
5449 globalStats.checkpoint_sync_time += msg->m_checkpoint_sync_time;
5450 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
5451 globalStats.buf_written_clean += msg->m_buf_written_clean;
5452 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
5453 globalStats.buf_written_backend += msg->m_buf_written_backend;
5454 globalStats.buf_fsync_backend += msg->m_buf_fsync_backend;
5455 globalStats.buf_alloc += msg->m_buf_alloc;
5456 }
5457
5458 /* ----------
5459 * pgstat_recv_recoveryconflict() -
5460 *
5461 * Process a RECOVERYCONFLICT message.
5462 * ----------
5463 */
5464 static void
pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict * msg,int len)5465 pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len)
5466 {
5467 PgStat_StatDBEntry *dbentry;
5468
5469 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5470
5471 switch (msg->m_reason)
5472 {
5473 case PROCSIG_RECOVERY_CONFLICT_DATABASE:
5474
5475 /*
5476 * Since we drop the information about the database as soon as it
5477 * replicates, there is no point in counting these conflicts.
5478 */
5479 break;
5480 case PROCSIG_RECOVERY_CONFLICT_TABLESPACE:
5481 dbentry->n_conflict_tablespace++;
5482 break;
5483 case PROCSIG_RECOVERY_CONFLICT_LOCK:
5484 dbentry->n_conflict_lock++;
5485 break;
5486 case PROCSIG_RECOVERY_CONFLICT_SNAPSHOT:
5487 dbentry->n_conflict_snapshot++;
5488 break;
5489 case PROCSIG_RECOVERY_CONFLICT_BUFFERPIN:
5490 dbentry->n_conflict_bufferpin++;
5491 break;
5492 case PROCSIG_RECOVERY_CONFLICT_STARTUP_DEADLOCK:
5493 dbentry->n_conflict_startup_deadlock++;
5494 break;
5495 }
5496 }
5497
5498 /* ----------
5499 * pgstat_recv_deadlock() -
5500 *
5501 * Process a DEADLOCK message.
5502 * ----------
5503 */
5504 static void
pgstat_recv_deadlock(PgStat_MsgDeadlock * msg,int len)5505 pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len)
5506 {
5507 PgStat_StatDBEntry *dbentry;
5508
5509 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5510
5511 dbentry->n_deadlocks++;
5512 }
5513
5514 /* ----------
5515 * pgstat_recv_tempfile() -
5516 *
5517 * Process a TEMPFILE message.
5518 * ----------
5519 */
5520 static void
pgstat_recv_tempfile(PgStat_MsgTempFile * msg,int len)5521 pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len)
5522 {
5523 PgStat_StatDBEntry *dbentry;
5524
5525 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5526
5527 dbentry->n_temp_bytes += msg->m_filesize;
5528 dbentry->n_temp_files += 1;
5529 }
5530
5531 /* ----------
5532 * pgstat_recv_funcstat() -
5533 *
5534 * Count what the backend has done.
5535 * ----------
5536 */
5537 static void
pgstat_recv_funcstat(PgStat_MsgFuncstat * msg,int len)5538 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
5539 {
5540 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
5541 PgStat_StatDBEntry *dbentry;
5542 PgStat_StatFuncEntry *funcentry;
5543 int i;
5544 bool found;
5545
5546 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5547
5548 /*
5549 * Process all function entries in the message.
5550 */
5551 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
5552 {
5553 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
5554 (void *) &(funcmsg->f_id),
5555 HASH_ENTER, &found);
5556
5557 if (!found)
5558 {
5559 /*
5560 * If it's a new function entry, initialize counters to the values
5561 * we just got.
5562 */
5563 funcentry->f_numcalls = funcmsg->f_numcalls;
5564 funcentry->f_total_time = funcmsg->f_total_time;
5565 funcentry->f_self_time = funcmsg->f_self_time;
5566 }
5567 else
5568 {
5569 /*
5570 * Otherwise add the values to the existing entry.
5571 */
5572 funcentry->f_numcalls += funcmsg->f_numcalls;
5573 funcentry->f_total_time += funcmsg->f_total_time;
5574 funcentry->f_self_time += funcmsg->f_self_time;
5575 }
5576 }
5577 }
5578
5579 /* ----------
5580 * pgstat_recv_funcpurge() -
5581 *
5582 * Arrange for dead function removal.
5583 * ----------
5584 */
5585 static void
pgstat_recv_funcpurge(PgStat_MsgFuncpurge * msg,int len)5586 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
5587 {
5588 PgStat_StatDBEntry *dbentry;
5589 int i;
5590
5591 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
5592
5593 /*
5594 * No need to purge if we don't even know the database.
5595 */
5596 if (!dbentry || !dbentry->functions)
5597 return;
5598
5599 /*
5600 * Process all function entries in the message.
5601 */
5602 for (i = 0; i < msg->m_nentries; i++)
5603 {
5604 /* Remove from hashtable if present; we don't care if it's not. */
5605 (void) hash_search(dbentry->functions,
5606 (void *) &(msg->m_functionid[i]),
5607 HASH_REMOVE, NULL);
5608 }
5609 }
5610
5611 /* ----------
5612 * pgstat_write_statsfile_needed() -
5613 *
5614 * Do we need to write out any stats files?
5615 * ----------
5616 */
5617 static bool
pgstat_write_statsfile_needed(void)5618 pgstat_write_statsfile_needed(void)
5619 {
5620 if (pending_write_requests != NIL)
5621 return true;
5622
5623 /* Everything was written recently */
5624 return false;
5625 }
5626
5627 /* ----------
5628 * pgstat_db_requested() -
5629 *
5630 * Checks whether stats for a particular DB need to be written to a file.
5631 * ----------
5632 */
5633 static bool
pgstat_db_requested(Oid databaseid)5634 pgstat_db_requested(Oid databaseid)
5635 {
5636 /*
5637 * If any requests are outstanding at all, we should write the stats for
5638 * shared catalogs (the "database" with OID 0). This ensures that
5639 * backends will see up-to-date stats for shared catalogs, even though
5640 * they send inquiry messages mentioning only their own DB.
5641 */
5642 if (databaseid == InvalidOid && pending_write_requests != NIL)
5643 return true;
5644
5645 /* Search to see if there's an open request to write this database. */
5646 if (list_member_oid(pending_write_requests, databaseid))
5647 return true;
5648
5649 return false;
5650 }
5651