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) 2001-2019, 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/tableam.h"
40 #include "access/transam.h"
41 #include "access/twophase_rmgr.h"
42 #include "access/xact.h"
43 #include "catalog/pg_database.h"
44 #include "catalog/pg_proc.h"
45 #include "common/ip.h"
46 #include "libpq/libpq.h"
47 #include "libpq/pqsignal.h"
48 #include "mb/pg_wchar.h"
49 #include "miscadmin.h"
50 #include "pg_trace.h"
51 #include "postmaster/autovacuum.h"
52 #include "postmaster/fork_process.h"
53 #include "postmaster/postmaster.h"
54 #include "replication/walsender.h"
55 #include "storage/backendid.h"
56 #include "storage/dsm.h"
57 #include "storage/fd.h"
58 #include "storage/ipc.h"
59 #include "storage/latch.h"
60 #include "storage/lmgr.h"
61 #include "storage/pg_shmem.h"
62 #include "storage/procsignal.h"
63 #include "storage/sinvaladt.h"
64 #include "utils/ascii.h"
65 #include "utils/guc.h"
66 #include "utils/memutils.h"
67 #include "utils/ps_status.h"
68 #include "utils/rel.h"
69 #include "utils/snapmgr.h"
70 #include "utils/timestamp.h"
71
72
73 /* ----------
74 * Timer definitions.
75 * ----------
76 */
77 #define PGSTAT_STAT_INTERVAL 500 /* Minimum time between stats file
78 * updates; in milliseconds. */
79
80 #define PGSTAT_RETRY_DELAY 10 /* How long to wait between checks for a
81 * new file; in milliseconds. */
82
83 #define PGSTAT_MAX_WAIT_TIME 10000 /* Maximum time to wait for a stats
84 * file update; in milliseconds. */
85
86 #define PGSTAT_INQ_INTERVAL 640 /* How often to ping the collector for a
87 * new file; in milliseconds. */
88
89 #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a
90 * failed statistics collector; in
91 * seconds. */
92
93 #define PGSTAT_POLL_LOOP_COUNT (PGSTAT_MAX_WAIT_TIME / PGSTAT_RETRY_DELAY)
94 #define PGSTAT_INQ_LOOP_COUNT (PGSTAT_INQ_INTERVAL / PGSTAT_RETRY_DELAY)
95
96 /* Minimum receive buffer size for the collector's socket. */
97 #define PGSTAT_MIN_RCVBUF (100 * 1024)
98
99
100 /* ----------
101 * The initial size hints for the hash tables used in the collector.
102 * ----------
103 */
104 #define PGSTAT_DB_HASH_SIZE 16
105 #define PGSTAT_TAB_HASH_SIZE 512
106 #define PGSTAT_FUNCTION_HASH_SIZE 512
107
108
109 /* ----------
110 * Total number of backends including auxiliary
111 *
112 * We reserve a slot for each possible BackendId, plus one for each
113 * possible auxiliary process type. (This scheme assumes there is not
114 * more than one of any auxiliary process type at a time.) MaxBackends
115 * includes autovacuum workers and background workers as well.
116 * ----------
117 */
118 #define NumBackendStatSlots (MaxBackends + NUM_AUXPROCTYPES)
119
120
121 /* ----------
122 * GUC parameters
123 * ----------
124 */
125 bool pgstat_track_activities = false;
126 bool pgstat_track_counts = false;
127 int pgstat_track_functions = TRACK_FUNC_OFF;
128 int pgstat_track_activity_query_size = 1024;
129
130 /* ----------
131 * Built from GUC parameter
132 * ----------
133 */
134 char *pgstat_stat_directory = NULL;
135 char *pgstat_stat_filename = NULL;
136 char *pgstat_stat_tmpname = NULL;
137
138 /*
139 * BgWriter global statistics counters (unused in other processes).
140 * Stored directly in a stats message structure so it can be sent
141 * without needing to copy things around. We assume this inits to zeroes.
142 */
143 PgStat_MsgBgWriter BgWriterStats;
144
145 /* ----------
146 * Local data
147 * ----------
148 */
149 NON_EXEC_STATIC pgsocket pgStatSock = PGINVALID_SOCKET;
150
151 static struct sockaddr_storage pgStatAddr;
152
153 static time_t last_pgstat_start_time;
154
155 static bool pgStatRunningInCollector = false;
156
157 /*
158 * Structures in which backends store per-table info that's waiting to be
159 * sent to the collector.
160 *
161 * NOTE: once allocated, TabStatusArray structures are never moved or deleted
162 * for the life of the backend. Also, we zero out the t_id fields of the
163 * contained PgStat_TableStatus structs whenever they are not actively in use.
164 * This allows relcache pgstat_info pointers to be treated as long-lived data,
165 * avoiding repeated searches in pgstat_initstats() when a relation is
166 * repeatedly opened during a transaction.
167 */
168 #define TABSTAT_QUANTUM 100 /* we alloc this many at a time */
169
170 typedef struct TabStatusArray
171 {
172 struct TabStatusArray *tsa_next; /* link to next array, if any */
173 int tsa_used; /* # entries currently used */
174 PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */
175 } TabStatusArray;
176
177 static TabStatusArray *pgStatTabList = NULL;
178
179 /*
180 * pgStatTabHash entry: map from relation OID to PgStat_TableStatus pointer
181 */
182 typedef struct TabStatHashEntry
183 {
184 Oid t_id;
185 PgStat_TableStatus *tsa_entry;
186 } TabStatHashEntry;
187
188 /*
189 * Hash table for O(1) t_id -> tsa_entry lookup
190 */
191 static HTAB *pgStatTabHash = NULL;
192
193 /*
194 * Backends store per-function info that's waiting to be sent to the collector
195 * in this hash table (indexed by function OID).
196 */
197 static HTAB *pgStatFunctions = NULL;
198
199 /*
200 * Indicates if backend has some function stats that it hasn't yet
201 * sent to the collector.
202 */
203 static bool have_function_stats = false;
204
205 /*
206 * Tuple insertion/deletion counts for an open transaction can't be propagated
207 * into PgStat_TableStatus counters until we know if it is going to commit
208 * or abort. Hence, we keep these counts in per-subxact structs that live
209 * in TopTransactionContext. This data structure is designed on the assumption
210 * that subxacts won't usually modify very many tables.
211 */
212 typedef struct PgStat_SubXactStatus
213 {
214 int nest_level; /* subtransaction nest level */
215 struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
216 PgStat_TableXactStatus *first; /* head of list for this subxact */
217 } PgStat_SubXactStatus;
218
219 static PgStat_SubXactStatus *pgStatXactStack = NULL;
220
221 static int pgStatXactCommit = 0;
222 static int pgStatXactRollback = 0;
223 PgStat_Counter pgStatBlockReadTime = 0;
224 PgStat_Counter pgStatBlockWriteTime = 0;
225
226 /* Record that's written to 2PC state file when pgstat state is persisted */
227 typedef struct TwoPhasePgStatRecord
228 {
229 PgStat_Counter tuples_inserted; /* tuples inserted in xact */
230 PgStat_Counter tuples_updated; /* tuples updated in xact */
231 PgStat_Counter tuples_deleted; /* tuples deleted in xact */
232 PgStat_Counter inserted_pre_trunc; /* tuples inserted prior to truncate */
233 PgStat_Counter updated_pre_trunc; /* tuples updated prior to truncate */
234 PgStat_Counter deleted_pre_trunc; /* tuples deleted prior to truncate */
235 Oid t_id; /* table's OID */
236 bool t_shared; /* is it a shared catalog? */
237 bool t_truncated; /* was the relation truncated? */
238 } TwoPhasePgStatRecord;
239
240 /*
241 * Info about current "snapshot" of stats file
242 */
243 static MemoryContext pgStatLocalContext = NULL;
244 static HTAB *pgStatDBHash = NULL;
245
246 /* Status for backends including auxiliary */
247 static LocalPgBackendStatus *localBackendStatusTable = NULL;
248
249 /* Total number of backends including auxiliary */
250 static int localNumBackends = 0;
251
252 /*
253 * Cluster wide statistics, kept in the stats collector.
254 * Contains statistics that are not collected per database
255 * or per table.
256 */
257 static PgStat_ArchiverStats archiverStats;
258 static PgStat_GlobalStats globalStats;
259
260 /*
261 * List of OIDs of databases we need to write out. If an entry is InvalidOid,
262 * it means to write only the shared-catalog stats ("DB 0"); otherwise, we
263 * will write both that DB's data and the shared stats.
264 */
265 static List *pending_write_requests = NIL;
266
267 /* Signal handler flags */
268 static volatile bool need_exit = false;
269 static volatile bool got_SIGHUP = false;
270
271 /*
272 * Total time charged to functions so far in the current backend.
273 * We use this to help separate "self" and "other" time charges.
274 * (We assume this initializes to zero.)
275 */
276 static instr_time total_func_time;
277
278
279 /* ----------
280 * Local function forward declarations
281 * ----------
282 */
283 #ifdef EXEC_BACKEND
284 static pid_t pgstat_forkexec(void);
285 #endif
286
287 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
288 static void pgstat_exit(SIGNAL_ARGS);
289 static void pgstat_beshutdown_hook(int code, Datum arg);
290 static void pgstat_sighup_handler(SIGNAL_ARGS);
291
292 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
293 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
294 Oid tableoid, bool create);
295 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
296 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
297 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
298 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
299 static void backend_read_statsfile(void);
300 static void pgstat_read_current_status(void);
301
302 static bool pgstat_write_statsfile_needed(void);
303 static bool pgstat_db_requested(Oid databaseid);
304
305 static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg);
306 static void pgstat_send_funcstats(void);
307 static HTAB *pgstat_collect_oids(Oid catalogid, AttrNumber anum_oid);
308
309 static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared);
310
311 static void pgstat_setup_memcxt(void);
312
313 static const char *pgstat_get_wait_activity(WaitEventActivity w);
314 static const char *pgstat_get_wait_client(WaitEventClient w);
315 static const char *pgstat_get_wait_ipc(WaitEventIPC w);
316 static const char *pgstat_get_wait_timeout(WaitEventTimeout w);
317 static const char *pgstat_get_wait_io(WaitEventIO w);
318
319 static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype);
320 static void pgstat_send(void *msg, int len);
321
322 static void pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len);
323 static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len);
324 static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len);
325 static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len);
326 static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len);
327 static void pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len);
328 static void pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len);
329 static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len);
330 static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
331 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
332 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
333 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
334 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
335 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
336 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
337 static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
338 static void pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len);
339 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
340
341 /* ------------------------------------------------------------
342 * Public functions called from postmaster follow
343 * ------------------------------------------------------------
344 */
345
346 /* ----------
347 * pgstat_init() -
348 *
349 * Called from postmaster at startup. Create the resources required
350 * by the statistics collector process. If unable to do so, do not
351 * fail --- better to let the postmaster start with stats collection
352 * disabled.
353 * ----------
354 */
355 void
pgstat_init(void)356 pgstat_init(void)
357 {
358 ACCEPT_TYPE_ARG3 alen;
359 struct addrinfo *addrs = NULL,
360 *addr,
361 hints;
362 int ret;
363 fd_set rset;
364 struct timeval tv;
365 char test_byte;
366 int sel_res;
367 int tries = 0;
368
369 #define TESTBYTEVAL ((char) 199)
370
371 /*
372 * This static assertion verifies that we didn't mess up the calculations
373 * involved in selecting maximum payload sizes for our UDP messages.
374 * Because the only consequence of overrunning PGSTAT_MAX_MSG_SIZE would
375 * be silent performance loss from fragmentation, it seems worth having a
376 * compile-time cross-check that we didn't.
377 */
378 StaticAssertStmt(sizeof(PgStat_Msg) <= PGSTAT_MAX_MSG_SIZE,
379 "maximum stats message size exceeds PGSTAT_MAX_MSG_SIZE");
380
381 /*
382 * Create the UDP socket for sending and receiving statistic messages
383 */
384 hints.ai_flags = AI_PASSIVE;
385 hints.ai_family = AF_UNSPEC;
386 hints.ai_socktype = SOCK_DGRAM;
387 hints.ai_protocol = 0;
388 hints.ai_addrlen = 0;
389 hints.ai_addr = NULL;
390 hints.ai_canonname = NULL;
391 hints.ai_next = NULL;
392 ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);
393 if (ret || !addrs)
394 {
395 ereport(LOG,
396 (errmsg("could not resolve \"localhost\": %s",
397 gai_strerror(ret))));
398 goto startup_failed;
399 }
400
401 /*
402 * On some platforms, pg_getaddrinfo_all() may return multiple addresses
403 * only one of which will actually work (eg, both IPv6 and IPv4 addresses
404 * when kernel will reject IPv6). Worse, the failure may occur at the
405 * bind() or perhaps even connect() stage. So we must loop through the
406 * results till we find a working combination. We will generate LOG
407 * messages, but no error, for bogus combinations.
408 */
409 for (addr = addrs; addr; addr = addr->ai_next)
410 {
411 #ifdef HAVE_UNIX_SOCKETS
412 /* Ignore AF_UNIX sockets, if any are returned. */
413 if (addr->ai_family == AF_UNIX)
414 continue;
415 #endif
416
417 if (++tries > 1)
418 ereport(LOG,
419 (errmsg("trying another address for the statistics collector")));
420
421 /*
422 * Create the socket.
423 */
424 if ((pgStatSock = socket(addr->ai_family, SOCK_DGRAM, 0)) == PGINVALID_SOCKET)
425 {
426 ereport(LOG,
427 (errcode_for_socket_access(),
428 errmsg("could not create socket for statistics collector: %m")));
429 continue;
430 }
431
432 /*
433 * Bind it to a kernel assigned port on localhost and get the assigned
434 * port via getsockname().
435 */
436 if (bind(pgStatSock, addr->ai_addr, addr->ai_addrlen) < 0)
437 {
438 ereport(LOG,
439 (errcode_for_socket_access(),
440 errmsg("could not bind socket for statistics collector: %m")));
441 closesocket(pgStatSock);
442 pgStatSock = PGINVALID_SOCKET;
443 continue;
444 }
445
446 alen = sizeof(pgStatAddr);
447 if (getsockname(pgStatSock, (struct sockaddr *) &pgStatAddr, &alen) < 0)
448 {
449 ereport(LOG,
450 (errcode_for_socket_access(),
451 errmsg("could not get address of socket for statistics collector: %m")));
452 closesocket(pgStatSock);
453 pgStatSock = PGINVALID_SOCKET;
454 continue;
455 }
456
457 /*
458 * Connect the socket to its own address. This saves a few cycles by
459 * not having to respecify the target address on every send. This also
460 * provides a kernel-level check that only packets from this same
461 * address will be received.
462 */
463 if (connect(pgStatSock, (struct sockaddr *) &pgStatAddr, alen) < 0)
464 {
465 ereport(LOG,
466 (errcode_for_socket_access(),
467 errmsg("could not connect socket for statistics collector: %m")));
468 closesocket(pgStatSock);
469 pgStatSock = PGINVALID_SOCKET;
470 continue;
471 }
472
473 /*
474 * Try to send and receive a one-byte test message on the socket. This
475 * is to catch situations where the socket can be created but will not
476 * actually pass data (for instance, because kernel packet filtering
477 * rules prevent it).
478 */
479 test_byte = TESTBYTEVAL;
480
481 retry1:
482 if (send(pgStatSock, &test_byte, 1, 0) != 1)
483 {
484 if (errno == EINTR)
485 goto retry1; /* if interrupted, just retry */
486 ereport(LOG,
487 (errcode_for_socket_access(),
488 errmsg("could not send test message on socket for statistics collector: %m")));
489 closesocket(pgStatSock);
490 pgStatSock = PGINVALID_SOCKET;
491 continue;
492 }
493
494 /*
495 * There could possibly be a little delay before the message can be
496 * received. We arbitrarily allow up to half a second before deciding
497 * it's broken.
498 */
499 for (;;) /* need a loop to handle EINTR */
500 {
501 FD_ZERO(&rset);
502 FD_SET(pgStatSock, &rset);
503
504 tv.tv_sec = 0;
505 tv.tv_usec = 500000;
506 sel_res = select(pgStatSock + 1, &rset, NULL, NULL, &tv);
507 if (sel_res >= 0 || errno != EINTR)
508 break;
509 }
510 if (sel_res < 0)
511 {
512 ereport(LOG,
513 (errcode_for_socket_access(),
514 errmsg("select() failed in statistics collector: %m")));
515 closesocket(pgStatSock);
516 pgStatSock = PGINVALID_SOCKET;
517 continue;
518 }
519 if (sel_res == 0 || !FD_ISSET(pgStatSock, &rset))
520 {
521 /*
522 * This is the case we actually think is likely, so take pains to
523 * give a specific message for it.
524 *
525 * errno will not be set meaningfully here, so don't use it.
526 */
527 ereport(LOG,
528 (errcode(ERRCODE_CONNECTION_FAILURE),
529 errmsg("test message did not get through on socket for statistics collector")));
530 closesocket(pgStatSock);
531 pgStatSock = PGINVALID_SOCKET;
532 continue;
533 }
534
535 test_byte++; /* just make sure variable is changed */
536
537 retry2:
538 if (recv(pgStatSock, &test_byte, 1, 0) != 1)
539 {
540 if (errno == EINTR)
541 goto retry2; /* if interrupted, just retry */
542 ereport(LOG,
543 (errcode_for_socket_access(),
544 errmsg("could not receive test message on socket for statistics collector: %m")));
545 closesocket(pgStatSock);
546 pgStatSock = PGINVALID_SOCKET;
547 continue;
548 }
549
550 if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */
551 {
552 ereport(LOG,
553 (errcode(ERRCODE_INTERNAL_ERROR),
554 errmsg("incorrect test message transmission on socket for statistics collector")));
555 closesocket(pgStatSock);
556 pgStatSock = PGINVALID_SOCKET;
557 continue;
558 }
559
560 /* If we get here, we have a working socket */
561 break;
562 }
563
564 /* Did we find a working address? */
565 if (!addr || pgStatSock == PGINVALID_SOCKET)
566 goto startup_failed;
567
568 /*
569 * Set the socket to non-blocking IO. This ensures that if the collector
570 * falls behind, statistics messages will be discarded; backends won't
571 * block waiting to send messages to the collector.
572 */
573 if (!pg_set_noblock(pgStatSock))
574 {
575 ereport(LOG,
576 (errcode_for_socket_access(),
577 errmsg("could not set statistics collector socket to nonblocking mode: %m")));
578 goto startup_failed;
579 }
580
581 /*
582 * Try to ensure that the socket's receive buffer is at least
583 * PGSTAT_MIN_RCVBUF bytes, so that it won't easily overflow and lose
584 * data. Use of UDP protocol means that we are willing to lose data under
585 * heavy load, but we don't want it to happen just because of ridiculously
586 * small default buffer sizes (such as 8KB on older Windows versions).
587 */
588 {
589 int old_rcvbuf;
590 int new_rcvbuf;
591 ACCEPT_TYPE_ARG3 rcvbufsize = sizeof(old_rcvbuf);
592
593 if (getsockopt(pgStatSock, SOL_SOCKET, SO_RCVBUF,
594 (char *) &old_rcvbuf, &rcvbufsize) < 0)
595 {
596 elog(LOG, "getsockopt(SO_RCVBUF) failed: %m");
597 /* if we can't get existing size, always try to set it */
598 old_rcvbuf = 0;
599 }
600
601 new_rcvbuf = PGSTAT_MIN_RCVBUF;
602 if (old_rcvbuf < new_rcvbuf)
603 {
604 if (setsockopt(pgStatSock, SOL_SOCKET, SO_RCVBUF,
605 (char *) &new_rcvbuf, sizeof(new_rcvbuf)) < 0)
606 elog(LOG, "setsockopt(SO_RCVBUF) failed: %m");
607 }
608 }
609
610 pg_freeaddrinfo_all(hints.ai_family, addrs);
611
612 return;
613
614 startup_failed:
615 ereport(LOG,
616 (errmsg("disabling statistics collector for lack of working socket")));
617
618 if (addrs)
619 pg_freeaddrinfo_all(hints.ai_family, addrs);
620
621 if (pgStatSock != PGINVALID_SOCKET)
622 closesocket(pgStatSock);
623 pgStatSock = PGINVALID_SOCKET;
624
625 /*
626 * Adjust GUC variables to suppress useless activity, and for debugging
627 * purposes (seeing track_counts off is a clue that we failed here). We
628 * use PGC_S_OVERRIDE because there is no point in trying to turn it back
629 * on from postgresql.conf without a restart.
630 */
631 SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
632 }
633
634 /*
635 * subroutine for pgstat_reset_all
636 */
637 static void
pgstat_reset_remove_files(const char * directory)638 pgstat_reset_remove_files(const char *directory)
639 {
640 DIR *dir;
641 struct dirent *entry;
642 char fname[MAXPGPATH * 2];
643
644 dir = AllocateDir(directory);
645 while ((entry = ReadDir(dir, directory)) != NULL)
646 {
647 int nchars;
648 Oid tmp_oid;
649
650 /*
651 * Skip directory entries that don't match the file names we write.
652 * See get_dbstat_filename for the database-specific pattern.
653 */
654 if (strncmp(entry->d_name, "global.", 7) == 0)
655 nchars = 7;
656 else
657 {
658 nchars = 0;
659 (void) sscanf(entry->d_name, "db_%u.%n",
660 &tmp_oid, &nchars);
661 if (nchars <= 0)
662 continue;
663 /* %u allows leading whitespace, so reject that */
664 if (strchr("0123456789", entry->d_name[3]) == NULL)
665 continue;
666 }
667
668 if (strcmp(entry->d_name + nchars, "tmp") != 0 &&
669 strcmp(entry->d_name + nchars, "stat") != 0)
670 continue;
671
672 snprintf(fname, sizeof(fname), "%s/%s", directory,
673 entry->d_name);
674 unlink(fname);
675 }
676 FreeDir(dir);
677 }
678
679 /*
680 * pgstat_reset_all() -
681 *
682 * Remove the stats files. This is currently used only if WAL
683 * recovery is needed after a crash.
684 */
685 void
pgstat_reset_all(void)686 pgstat_reset_all(void)
687 {
688 pgstat_reset_remove_files(pgstat_stat_directory);
689 pgstat_reset_remove_files(PGSTAT_STAT_PERMANENT_DIRECTORY);
690 }
691
692 #ifdef EXEC_BACKEND
693
694 /*
695 * pgstat_forkexec() -
696 *
697 * Format up the arglist for, then fork and exec, statistics collector process
698 */
699 static pid_t
pgstat_forkexec(void)700 pgstat_forkexec(void)
701 {
702 char *av[10];
703 int ac = 0;
704
705 av[ac++] = "postgres";
706 av[ac++] = "--forkcol";
707 av[ac++] = NULL; /* filled in by postmaster_forkexec */
708
709 av[ac] = NULL;
710 Assert(ac < lengthof(av));
711
712 return postmaster_forkexec(ac, av);
713 }
714 #endif /* EXEC_BACKEND */
715
716
717 /*
718 * pgstat_start() -
719 *
720 * Called from postmaster at startup or after an existing collector
721 * died. Attempt to fire up a fresh statistics collector.
722 *
723 * Returns PID of child process, or 0 if fail.
724 *
725 * Note: if fail, we will be called again from the postmaster main loop.
726 */
727 int
pgstat_start(void)728 pgstat_start(void)
729 {
730 time_t curtime;
731 pid_t pgStatPid;
732
733 /*
734 * Check that the socket is there, else pgstat_init failed and we can do
735 * nothing useful.
736 */
737 if (pgStatSock == PGINVALID_SOCKET)
738 return 0;
739
740 /*
741 * Do nothing if too soon since last collector start. This is a safety
742 * valve to protect against continuous respawn attempts if the collector
743 * is dying immediately at launch. Note that since we will be re-called
744 * from the postmaster main loop, we will get another chance later.
745 */
746 curtime = time(NULL);
747 if ((unsigned int) (curtime - last_pgstat_start_time) <
748 (unsigned int) PGSTAT_RESTART_INTERVAL)
749 return 0;
750 last_pgstat_start_time = curtime;
751
752 /*
753 * Okay, fork off the collector.
754 */
755 #ifdef EXEC_BACKEND
756 switch ((pgStatPid = pgstat_forkexec()))
757 #else
758 switch ((pgStatPid = fork_process()))
759 #endif
760 {
761 case -1:
762 ereport(LOG,
763 (errmsg("could not fork statistics collector: %m")));
764 return 0;
765
766 #ifndef EXEC_BACKEND
767 case 0:
768 /* in postmaster child ... */
769 InitPostmasterChild();
770
771 /* Close the postmaster's sockets */
772 ClosePostmasterPorts(false);
773
774 /* Drop our connection to postmaster's shared memory, as well */
775 dsm_detach_all();
776 PGSharedMemoryDetach();
777
778 PgstatCollectorMain(0, NULL);
779 break;
780 #endif
781
782 default:
783 return (int) pgStatPid;
784 }
785
786 /* shouldn't get here */
787 return 0;
788 }
789
790 void
allow_immediate_pgstat_restart(void)791 allow_immediate_pgstat_restart(void)
792 {
793 last_pgstat_start_time = 0;
794 }
795
796 /* ------------------------------------------------------------
797 * Public functions used by backends follow
798 *------------------------------------------------------------
799 */
800
801
802 /* ----------
803 * pgstat_report_stat() -
804 *
805 * Must be called by processes that performs DML: tcop/postgres.c, logical
806 * receiver processes, SPI worker, etc. to send the so far collected
807 * per-table and function usage statistics to the collector. Note that this
808 * is called only when not within a transaction, so it is fair to use
809 * transaction stop time as an approximation of current time.
810 * ----------
811 */
812 void
pgstat_report_stat(bool force)813 pgstat_report_stat(bool force)
814 {
815 /* we assume this inits to all zeroes: */
816 static const PgStat_TableCounts all_zeroes;
817 static TimestampTz last_report = 0;
818
819 TimestampTz now;
820 PgStat_MsgTabstat regular_msg;
821 PgStat_MsgTabstat shared_msg;
822 TabStatusArray *tsa;
823 int i;
824
825 /* Don't expend a clock check if nothing to do */
826 if ((pgStatTabList == NULL || pgStatTabList->tsa_used == 0) &&
827 pgStatXactCommit == 0 && pgStatXactRollback == 0 &&
828 !have_function_stats)
829 return;
830
831 /*
832 * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL
833 * msec since we last sent one, or the caller wants to force stats out.
834 */
835 now = GetCurrentTransactionStopTimestamp();
836 if (!force &&
837 !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL))
838 return;
839 last_report = now;
840
841 /*
842 * Destroy pgStatTabHash before we start invalidating PgStat_TableEntry
843 * entries it points to. (Should we fail partway through the loop below,
844 * it's okay to have removed the hashtable already --- the only
845 * consequence is we'd get multiple entries for the same table in the
846 * pgStatTabList, and that's safe.)
847 */
848 if (pgStatTabHash)
849 hash_destroy(pgStatTabHash);
850 pgStatTabHash = NULL;
851
852 /*
853 * Scan through the TabStatusArray struct(s) to find tables that actually
854 * have counts, and build messages to send. We have to separate shared
855 * relations from regular ones because the databaseid field in the message
856 * header has to depend on that.
857 */
858 regular_msg.m_databaseid = MyDatabaseId;
859 shared_msg.m_databaseid = InvalidOid;
860 regular_msg.m_nentries = 0;
861 shared_msg.m_nentries = 0;
862
863 for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next)
864 {
865 for (i = 0; i < tsa->tsa_used; i++)
866 {
867 PgStat_TableStatus *entry = &tsa->tsa_entries[i];
868 PgStat_MsgTabstat *this_msg;
869 PgStat_TableEntry *this_ent;
870
871 /* Shouldn't have any pending transaction-dependent counts */
872 Assert(entry->trans == NULL);
873
874 /*
875 * Ignore entries that didn't accumulate any actual counts, such
876 * as indexes that were opened by the planner but not used.
877 */
878 if (memcmp(&entry->t_counts, &all_zeroes,
879 sizeof(PgStat_TableCounts)) == 0)
880 continue;
881
882 /*
883 * OK, insert data into the appropriate message, and send if full.
884 */
885 this_msg = entry->t_shared ? &shared_msg : ®ular_msg;
886 this_ent = &this_msg->m_entry[this_msg->m_nentries];
887 this_ent->t_id = entry->t_id;
888 memcpy(&this_ent->t_counts, &entry->t_counts,
889 sizeof(PgStat_TableCounts));
890 if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES)
891 {
892 pgstat_send_tabstat(this_msg);
893 this_msg->m_nentries = 0;
894 }
895 }
896 /* zero out TableStatus structs after use */
897 MemSet(tsa->tsa_entries, 0,
898 tsa->tsa_used * sizeof(PgStat_TableStatus));
899 tsa->tsa_used = 0;
900 }
901
902 /*
903 * Send partial messages. Make sure that any pending xact commit/abort
904 * gets counted, even if there are no table stats to send.
905 */
906 if (regular_msg.m_nentries > 0 ||
907 pgStatXactCommit > 0 || pgStatXactRollback > 0)
908 pgstat_send_tabstat(®ular_msg);
909 if (shared_msg.m_nentries > 0)
910 pgstat_send_tabstat(&shared_msg);
911
912 /* Now, send function statistics */
913 pgstat_send_funcstats();
914 }
915
916 /*
917 * Subroutine for pgstat_report_stat: finish and send a tabstat message
918 */
919 static void
pgstat_send_tabstat(PgStat_MsgTabstat * tsmsg)920 pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg)
921 {
922 int n;
923 int len;
924
925 /* It's unlikely we'd get here with no socket, but maybe not impossible */
926 if (pgStatSock == PGINVALID_SOCKET)
927 return;
928
929 /*
930 * Report and reset accumulated xact commit/rollback and I/O timings
931 * whenever we send a normal tabstat message
932 */
933 if (OidIsValid(tsmsg->m_databaseid))
934 {
935 tsmsg->m_xact_commit = pgStatXactCommit;
936 tsmsg->m_xact_rollback = pgStatXactRollback;
937 tsmsg->m_block_read_time = pgStatBlockReadTime;
938 tsmsg->m_block_write_time = pgStatBlockWriteTime;
939 pgStatXactCommit = 0;
940 pgStatXactRollback = 0;
941 pgStatBlockReadTime = 0;
942 pgStatBlockWriteTime = 0;
943 }
944 else
945 {
946 tsmsg->m_xact_commit = 0;
947 tsmsg->m_xact_rollback = 0;
948 tsmsg->m_block_read_time = 0;
949 tsmsg->m_block_write_time = 0;
950 }
951
952 n = tsmsg->m_nentries;
953 len = offsetof(PgStat_MsgTabstat, m_entry[0]) +
954 n * sizeof(PgStat_TableEntry);
955
956 pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT);
957 pgstat_send(tsmsg, len);
958 }
959
960 /*
961 * Subroutine for pgstat_report_stat: populate and send a function stat message
962 */
963 static void
pgstat_send_funcstats(void)964 pgstat_send_funcstats(void)
965 {
966 /* we assume this inits to all zeroes: */
967 static const PgStat_FunctionCounts all_zeroes;
968
969 PgStat_MsgFuncstat msg;
970 PgStat_BackendFunctionEntry *entry;
971 HASH_SEQ_STATUS fstat;
972
973 if (pgStatFunctions == NULL)
974 return;
975
976 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_FUNCSTAT);
977 msg.m_databaseid = MyDatabaseId;
978 msg.m_nentries = 0;
979
980 hash_seq_init(&fstat, pgStatFunctions);
981 while ((entry = (PgStat_BackendFunctionEntry *) hash_seq_search(&fstat)) != NULL)
982 {
983 PgStat_FunctionEntry *m_ent;
984
985 /* Skip it if no counts accumulated since last time */
986 if (memcmp(&entry->f_counts, &all_zeroes,
987 sizeof(PgStat_FunctionCounts)) == 0)
988 continue;
989
990 /* need to convert format of time accumulators */
991 m_ent = &msg.m_entry[msg.m_nentries];
992 m_ent->f_id = entry->f_id;
993 m_ent->f_numcalls = entry->f_counts.f_numcalls;
994 m_ent->f_total_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_total_time);
995 m_ent->f_self_time = INSTR_TIME_GET_MICROSEC(entry->f_counts.f_self_time);
996
997 if (++msg.m_nentries >= PGSTAT_NUM_FUNCENTRIES)
998 {
999 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
1000 msg.m_nentries * sizeof(PgStat_FunctionEntry));
1001 msg.m_nentries = 0;
1002 }
1003
1004 /* reset the entry's counts */
1005 MemSet(&entry->f_counts, 0, sizeof(PgStat_FunctionCounts));
1006 }
1007
1008 if (msg.m_nentries > 0)
1009 pgstat_send(&msg, offsetof(PgStat_MsgFuncstat, m_entry[0]) +
1010 msg.m_nentries * sizeof(PgStat_FunctionEntry));
1011
1012 have_function_stats = false;
1013 }
1014
1015
1016 /* ----------
1017 * pgstat_vacuum_stat() -
1018 *
1019 * Will tell the collector about objects he can get rid of.
1020 * ----------
1021 */
1022 void
pgstat_vacuum_stat(void)1023 pgstat_vacuum_stat(void)
1024 {
1025 HTAB *htab;
1026 PgStat_MsgTabpurge msg;
1027 PgStat_MsgFuncpurge f_msg;
1028 HASH_SEQ_STATUS hstat;
1029 PgStat_StatDBEntry *dbentry;
1030 PgStat_StatTabEntry *tabentry;
1031 PgStat_StatFuncEntry *funcentry;
1032 int len;
1033
1034 if (pgStatSock == PGINVALID_SOCKET)
1035 return;
1036
1037 /*
1038 * If not done for this transaction, read the statistics collector stats
1039 * file into some hash tables.
1040 */
1041 backend_read_statsfile();
1042
1043 /*
1044 * Read pg_database and make a list of OIDs of all existing databases
1045 */
1046 htab = pgstat_collect_oids(DatabaseRelationId, Anum_pg_database_oid);
1047
1048 /*
1049 * Search the database hash table for dead databases and tell the
1050 * collector to drop them.
1051 */
1052 hash_seq_init(&hstat, pgStatDBHash);
1053 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
1054 {
1055 Oid dbid = dbentry->databaseid;
1056
1057 CHECK_FOR_INTERRUPTS();
1058
1059 /* the DB entry for shared tables (with InvalidOid) is never dropped */
1060 if (OidIsValid(dbid) &&
1061 hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL)
1062 pgstat_drop_database(dbid);
1063 }
1064
1065 /* Clean up */
1066 hash_destroy(htab);
1067
1068 /*
1069 * Lookup our own database entry; if not found, nothing more to do.
1070 */
1071 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
1072 (void *) &MyDatabaseId,
1073 HASH_FIND, NULL);
1074 if (dbentry == NULL || dbentry->tables == NULL)
1075 return;
1076
1077 /*
1078 * Similarly to above, make a list of all known relations in this DB.
1079 */
1080 htab = pgstat_collect_oids(RelationRelationId, Anum_pg_class_oid);
1081
1082 /*
1083 * Initialize our messages table counter to zero
1084 */
1085 msg.m_nentries = 0;
1086
1087 /*
1088 * Check for all tables listed in stats hashtable if they still exist.
1089 */
1090 hash_seq_init(&hstat, dbentry->tables);
1091 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL)
1092 {
1093 Oid tabid = tabentry->tableid;
1094
1095 CHECK_FOR_INTERRUPTS();
1096
1097 if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL)
1098 continue;
1099
1100 /*
1101 * Not there, so add this table's Oid to the message
1102 */
1103 msg.m_tableid[msg.m_nentries++] = tabid;
1104
1105 /*
1106 * If the message is full, send it out and reinitialize to empty
1107 */
1108 if (msg.m_nentries >= PGSTAT_NUM_TABPURGE)
1109 {
1110 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
1111 + msg.m_nentries * sizeof(Oid);
1112
1113 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1114 msg.m_databaseid = MyDatabaseId;
1115 pgstat_send(&msg, len);
1116
1117 msg.m_nentries = 0;
1118 }
1119 }
1120
1121 /*
1122 * Send the rest
1123 */
1124 if (msg.m_nentries > 0)
1125 {
1126 len = offsetof(PgStat_MsgTabpurge, m_tableid[0])
1127 + msg.m_nentries * sizeof(Oid);
1128
1129 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1130 msg.m_databaseid = MyDatabaseId;
1131 pgstat_send(&msg, len);
1132 }
1133
1134 /* Clean up */
1135 hash_destroy(htab);
1136
1137 /*
1138 * Now repeat the above steps for functions. However, we needn't bother
1139 * in the common case where no function stats are being collected.
1140 */
1141 if (dbentry->functions != NULL &&
1142 hash_get_num_entries(dbentry->functions) > 0)
1143 {
1144 htab = pgstat_collect_oids(ProcedureRelationId, Anum_pg_proc_oid);
1145
1146 pgstat_setheader(&f_msg.m_hdr, PGSTAT_MTYPE_FUNCPURGE);
1147 f_msg.m_databaseid = MyDatabaseId;
1148 f_msg.m_nentries = 0;
1149
1150 hash_seq_init(&hstat, dbentry->functions);
1151 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&hstat)) != NULL)
1152 {
1153 Oid funcid = funcentry->functionid;
1154
1155 CHECK_FOR_INTERRUPTS();
1156
1157 if (hash_search(htab, (void *) &funcid, HASH_FIND, NULL) != NULL)
1158 continue;
1159
1160 /*
1161 * Not there, so add this function's Oid to the message
1162 */
1163 f_msg.m_functionid[f_msg.m_nentries++] = funcid;
1164
1165 /*
1166 * If the message is full, send it out and reinitialize to empty
1167 */
1168 if (f_msg.m_nentries >= PGSTAT_NUM_FUNCPURGE)
1169 {
1170 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1171 + f_msg.m_nentries * sizeof(Oid);
1172
1173 pgstat_send(&f_msg, len);
1174
1175 f_msg.m_nentries = 0;
1176 }
1177 }
1178
1179 /*
1180 * Send the rest
1181 */
1182 if (f_msg.m_nentries > 0)
1183 {
1184 len = offsetof(PgStat_MsgFuncpurge, m_functionid[0])
1185 + f_msg.m_nentries * sizeof(Oid);
1186
1187 pgstat_send(&f_msg, len);
1188 }
1189
1190 hash_destroy(htab);
1191 }
1192 }
1193
1194
1195 /* ----------
1196 * pgstat_collect_oids() -
1197 *
1198 * Collect the OIDs of all objects listed in the specified system catalog
1199 * into a temporary hash table. Caller should hash_destroy the result
1200 * when done with it. (However, we make the table in CurrentMemoryContext
1201 * so that it will be freed properly in event of an error.)
1202 * ----------
1203 */
1204 static HTAB *
pgstat_collect_oids(Oid catalogid,AttrNumber anum_oid)1205 pgstat_collect_oids(Oid catalogid, AttrNumber anum_oid)
1206 {
1207 HTAB *htab;
1208 HASHCTL hash_ctl;
1209 Relation rel;
1210 TableScanDesc scan;
1211 HeapTuple tup;
1212 Snapshot snapshot;
1213
1214 memset(&hash_ctl, 0, sizeof(hash_ctl));
1215 hash_ctl.keysize = sizeof(Oid);
1216 hash_ctl.entrysize = sizeof(Oid);
1217 hash_ctl.hcxt = CurrentMemoryContext;
1218 htab = hash_create("Temporary table of OIDs",
1219 PGSTAT_TAB_HASH_SIZE,
1220 &hash_ctl,
1221 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1222
1223 rel = table_open(catalogid, AccessShareLock);
1224 snapshot = RegisterSnapshot(GetLatestSnapshot());
1225 scan = table_beginscan(rel, snapshot, 0, NULL);
1226 while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL)
1227 {
1228 Oid thisoid;
1229 bool isnull;
1230
1231 thisoid = heap_getattr(tup, anum_oid, RelationGetDescr(rel), &isnull);
1232 Assert(!isnull);
1233
1234 CHECK_FOR_INTERRUPTS();
1235
1236 (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL);
1237 }
1238 table_endscan(scan);
1239 UnregisterSnapshot(snapshot);
1240 table_close(rel, AccessShareLock);
1241
1242 return htab;
1243 }
1244
1245
1246 /* ----------
1247 * pgstat_drop_database() -
1248 *
1249 * Tell the collector that we just dropped a database.
1250 * (If the message gets lost, we will still clean the dead DB eventually
1251 * via future invocations of pgstat_vacuum_stat().)
1252 * ----------
1253 */
1254 void
pgstat_drop_database(Oid databaseid)1255 pgstat_drop_database(Oid databaseid)
1256 {
1257 PgStat_MsgDropdb msg;
1258
1259 if (pgStatSock == PGINVALID_SOCKET)
1260 return;
1261
1262 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DROPDB);
1263 msg.m_databaseid = databaseid;
1264 pgstat_send(&msg, sizeof(msg));
1265 }
1266
1267
1268 /* ----------
1269 * pgstat_drop_relation() -
1270 *
1271 * Tell the collector that we just dropped a relation.
1272 * (If the message gets lost, we will still clean the dead entry eventually
1273 * via future invocations of pgstat_vacuum_stat().)
1274 *
1275 * Currently not used for lack of any good place to call it; we rely
1276 * entirely on pgstat_vacuum_stat() to clean out stats for dead rels.
1277 * ----------
1278 */
1279 #ifdef NOT_USED
1280 void
pgstat_drop_relation(Oid relid)1281 pgstat_drop_relation(Oid relid)
1282 {
1283 PgStat_MsgTabpurge msg;
1284 int len;
1285
1286 if (pgStatSock == PGINVALID_SOCKET)
1287 return;
1288
1289 msg.m_tableid[0] = relid;
1290 msg.m_nentries = 1;
1291
1292 len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) + sizeof(Oid);
1293
1294 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE);
1295 msg.m_databaseid = MyDatabaseId;
1296 pgstat_send(&msg, len);
1297 }
1298 #endif /* NOT_USED */
1299
1300
1301 /* ----------
1302 * pgstat_reset_counters() -
1303 *
1304 * Tell the statistics collector to reset counters for our database.
1305 *
1306 * Permission checking for this function is managed through the normal
1307 * GRANT system.
1308 * ----------
1309 */
1310 void
pgstat_reset_counters(void)1311 pgstat_reset_counters(void)
1312 {
1313 PgStat_MsgResetcounter msg;
1314
1315 if (pgStatSock == PGINVALID_SOCKET)
1316 return;
1317
1318 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETCOUNTER);
1319 msg.m_databaseid = MyDatabaseId;
1320 pgstat_send(&msg, sizeof(msg));
1321 }
1322
1323 /* ----------
1324 * pgstat_reset_shared_counters() -
1325 *
1326 * Tell the statistics collector to reset cluster-wide shared counters.
1327 *
1328 * Permission checking for this function is managed through the normal
1329 * GRANT system.
1330 * ----------
1331 */
1332 void
pgstat_reset_shared_counters(const char * target)1333 pgstat_reset_shared_counters(const char *target)
1334 {
1335 PgStat_MsgResetsharedcounter msg;
1336
1337 if (pgStatSock == PGINVALID_SOCKET)
1338 return;
1339
1340 if (strcmp(target, "archiver") == 0)
1341 msg.m_resettarget = RESET_ARCHIVER;
1342 else if (strcmp(target, "bgwriter") == 0)
1343 msg.m_resettarget = RESET_BGWRITER;
1344 else
1345 ereport(ERROR,
1346 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1347 errmsg("unrecognized reset target: \"%s\"", target),
1348 errhint("Target must be \"archiver\" or \"bgwriter\".")));
1349
1350 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSHAREDCOUNTER);
1351 pgstat_send(&msg, sizeof(msg));
1352 }
1353
1354 /* ----------
1355 * pgstat_reset_single_counter() -
1356 *
1357 * Tell the statistics collector to reset a single counter.
1358 *
1359 * Permission checking for this function is managed through the normal
1360 * GRANT system.
1361 * ----------
1362 */
1363 void
pgstat_reset_single_counter(Oid objoid,PgStat_Single_Reset_Type type)1364 pgstat_reset_single_counter(Oid objoid, PgStat_Single_Reset_Type type)
1365 {
1366 PgStat_MsgResetsinglecounter msg;
1367
1368 if (pgStatSock == PGINVALID_SOCKET)
1369 return;
1370
1371 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RESETSINGLECOUNTER);
1372 msg.m_databaseid = MyDatabaseId;
1373 msg.m_resettype = type;
1374 msg.m_objectid = objoid;
1375
1376 pgstat_send(&msg, sizeof(msg));
1377 }
1378
1379 /* ----------
1380 * pgstat_report_autovac() -
1381 *
1382 * Called from autovacuum.c to report startup of an autovacuum process.
1383 * We are called before InitPostgres is done, so can't rely on MyDatabaseId;
1384 * the db OID must be passed in, instead.
1385 * ----------
1386 */
1387 void
pgstat_report_autovac(Oid dboid)1388 pgstat_report_autovac(Oid dboid)
1389 {
1390 PgStat_MsgAutovacStart msg;
1391
1392 if (pgStatSock == PGINVALID_SOCKET)
1393 return;
1394
1395 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START);
1396 msg.m_databaseid = dboid;
1397 msg.m_start_time = GetCurrentTimestamp();
1398
1399 pgstat_send(&msg, sizeof(msg));
1400 }
1401
1402
1403 /* ---------
1404 * pgstat_report_vacuum() -
1405 *
1406 * Tell the collector about the table we just vacuumed.
1407 * ---------
1408 */
1409 void
pgstat_report_vacuum(Oid tableoid,bool shared,PgStat_Counter livetuples,PgStat_Counter deadtuples)1410 pgstat_report_vacuum(Oid tableoid, bool shared,
1411 PgStat_Counter livetuples, PgStat_Counter deadtuples)
1412 {
1413 PgStat_MsgVacuum msg;
1414
1415 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1416 return;
1417
1418 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
1419 msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
1420 msg.m_tableoid = tableoid;
1421 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1422 msg.m_vacuumtime = GetCurrentTimestamp();
1423 msg.m_live_tuples = livetuples;
1424 msg.m_dead_tuples = deadtuples;
1425 pgstat_send(&msg, sizeof(msg));
1426 }
1427
1428 /* --------
1429 * pgstat_report_analyze() -
1430 *
1431 * Tell the collector about the table we just analyzed.
1432 *
1433 * Caller must provide new live- and dead-tuples estimates, as well as a
1434 * flag indicating whether to reset the changes_since_analyze counter.
1435 * --------
1436 */
1437 void
pgstat_report_analyze(Relation rel,PgStat_Counter livetuples,PgStat_Counter deadtuples,bool resetcounter)1438 pgstat_report_analyze(Relation rel,
1439 PgStat_Counter livetuples, PgStat_Counter deadtuples,
1440 bool resetcounter)
1441 {
1442 PgStat_MsgAnalyze msg;
1443
1444 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1445 return;
1446
1447 /*
1448 * Unlike VACUUM, ANALYZE might be running inside a transaction that has
1449 * already inserted and/or deleted rows in the target table. ANALYZE will
1450 * have counted such rows as live or dead respectively. Because we will
1451 * report our counts of such rows at transaction end, we should subtract
1452 * off these counts from what we send to the collector now, else they'll
1453 * be double-counted after commit. (This approach also ensures that the
1454 * collector ends up with the right numbers if we abort instead of
1455 * committing.)
1456 */
1457 if (rel->pgstat_info != NULL)
1458 {
1459 PgStat_TableXactStatus *trans;
1460
1461 for (trans = rel->pgstat_info->trans; trans; trans = trans->upper)
1462 {
1463 livetuples -= trans->tuples_inserted - trans->tuples_deleted;
1464 deadtuples -= trans->tuples_updated + trans->tuples_deleted;
1465 }
1466 /* count stuff inserted by already-aborted subxacts, too */
1467 deadtuples -= rel->pgstat_info->t_counts.t_delta_dead_tuples;
1468 /* Since ANALYZE's counts are estimates, we could have underflowed */
1469 livetuples = Max(livetuples, 0);
1470 deadtuples = Max(deadtuples, 0);
1471 }
1472
1473 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE);
1474 msg.m_databaseid = rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
1475 msg.m_tableoid = RelationGetRelid(rel);
1476 msg.m_autovacuum = IsAutoVacuumWorkerProcess();
1477 msg.m_resetcounter = resetcounter;
1478 msg.m_analyzetime = GetCurrentTimestamp();
1479 msg.m_live_tuples = livetuples;
1480 msg.m_dead_tuples = deadtuples;
1481 pgstat_send(&msg, sizeof(msg));
1482 }
1483
1484 /* --------
1485 * pgstat_report_recovery_conflict() -
1486 *
1487 * Tell the collector about a Hot Standby recovery conflict.
1488 * --------
1489 */
1490 void
pgstat_report_recovery_conflict(int reason)1491 pgstat_report_recovery_conflict(int reason)
1492 {
1493 PgStat_MsgRecoveryConflict msg;
1494
1495 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1496 return;
1497
1498 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_RECOVERYCONFLICT);
1499 msg.m_databaseid = MyDatabaseId;
1500 msg.m_reason = reason;
1501 pgstat_send(&msg, sizeof(msg));
1502 }
1503
1504 /* --------
1505 * pgstat_report_deadlock() -
1506 *
1507 * Tell the collector about a deadlock detected.
1508 * --------
1509 */
1510 void
pgstat_report_deadlock(void)1511 pgstat_report_deadlock(void)
1512 {
1513 PgStat_MsgDeadlock msg;
1514
1515 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1516 return;
1517
1518 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DEADLOCK);
1519 msg.m_databaseid = MyDatabaseId;
1520 pgstat_send(&msg, sizeof(msg));
1521 }
1522
1523
1524
1525 /* --------
1526 * pgstat_report_checksum_failures_in_db() -
1527 *
1528 * Tell the collector about one or more checksum failures.
1529 * --------
1530 */
1531 void
pgstat_report_checksum_failures_in_db(Oid dboid,int failurecount)1532 pgstat_report_checksum_failures_in_db(Oid dboid, int failurecount)
1533 {
1534 PgStat_MsgChecksumFailure msg;
1535
1536 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1537 return;
1538
1539 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_CHECKSUMFAILURE);
1540 msg.m_databaseid = dboid;
1541 msg.m_failurecount = failurecount;
1542 msg.m_failure_time = GetCurrentTimestamp();
1543
1544 pgstat_send(&msg, sizeof(msg));
1545 }
1546
1547 /* --------
1548 * pgstat_report_checksum_failure() -
1549 *
1550 * Tell the collector about a checksum failure.
1551 * --------
1552 */
1553 void
pgstat_report_checksum_failure(void)1554 pgstat_report_checksum_failure(void)
1555 {
1556 pgstat_report_checksum_failures_in_db(MyDatabaseId, 1);
1557 }
1558
1559 /* --------
1560 * pgstat_report_tempfile() -
1561 *
1562 * Tell the collector about a temporary file.
1563 * --------
1564 */
1565 void
pgstat_report_tempfile(size_t filesize)1566 pgstat_report_tempfile(size_t filesize)
1567 {
1568 PgStat_MsgTempFile msg;
1569
1570 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1571 return;
1572
1573 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TEMPFILE);
1574 msg.m_databaseid = MyDatabaseId;
1575 msg.m_filesize = filesize;
1576 pgstat_send(&msg, sizeof(msg));
1577 }
1578
1579
1580 /* ----------
1581 * pgstat_ping() -
1582 *
1583 * Send some junk data to the collector to increase traffic.
1584 * ----------
1585 */
1586 void
pgstat_ping(void)1587 pgstat_ping(void)
1588 {
1589 PgStat_MsgDummy msg;
1590
1591 if (pgStatSock == PGINVALID_SOCKET)
1592 return;
1593
1594 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY);
1595 pgstat_send(&msg, sizeof(msg));
1596 }
1597
1598 /* ----------
1599 * pgstat_send_inquiry() -
1600 *
1601 * Notify collector that we need fresh data.
1602 * ----------
1603 */
1604 static void
pgstat_send_inquiry(TimestampTz clock_time,TimestampTz cutoff_time,Oid databaseid)1605 pgstat_send_inquiry(TimestampTz clock_time, TimestampTz cutoff_time, Oid databaseid)
1606 {
1607 PgStat_MsgInquiry msg;
1608
1609 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_INQUIRY);
1610 msg.clock_time = clock_time;
1611 msg.cutoff_time = cutoff_time;
1612 msg.databaseid = databaseid;
1613 pgstat_send(&msg, sizeof(msg));
1614 }
1615
1616
1617 /*
1618 * Initialize function call usage data.
1619 * Called by the executor before invoking a function.
1620 */
1621 void
pgstat_init_function_usage(FunctionCallInfo fcinfo,PgStat_FunctionCallUsage * fcu)1622 pgstat_init_function_usage(FunctionCallInfo fcinfo,
1623 PgStat_FunctionCallUsage *fcu)
1624 {
1625 PgStat_BackendFunctionEntry *htabent;
1626 bool found;
1627
1628 if (pgstat_track_functions <= fcinfo->flinfo->fn_stats)
1629 {
1630 /* stats not wanted */
1631 fcu->fs = NULL;
1632 return;
1633 }
1634
1635 if (!pgStatFunctions)
1636 {
1637 /* First time through - initialize function stat table */
1638 HASHCTL hash_ctl;
1639
1640 memset(&hash_ctl, 0, sizeof(hash_ctl));
1641 hash_ctl.keysize = sizeof(Oid);
1642 hash_ctl.entrysize = sizeof(PgStat_BackendFunctionEntry);
1643 pgStatFunctions = hash_create("Function stat entries",
1644 PGSTAT_FUNCTION_HASH_SIZE,
1645 &hash_ctl,
1646 HASH_ELEM | HASH_BLOBS);
1647 }
1648
1649 /* Get the stats entry for this function, create if necessary */
1650 htabent = hash_search(pgStatFunctions, &fcinfo->flinfo->fn_oid,
1651 HASH_ENTER, &found);
1652 if (!found)
1653 MemSet(&htabent->f_counts, 0, sizeof(PgStat_FunctionCounts));
1654
1655 fcu->fs = &htabent->f_counts;
1656
1657 /* save stats for this function, later used to compensate for recursion */
1658 fcu->save_f_total_time = htabent->f_counts.f_total_time;
1659
1660 /* save current backend-wide total time */
1661 fcu->save_total = total_func_time;
1662
1663 /* get clock time as of function start */
1664 INSTR_TIME_SET_CURRENT(fcu->f_start);
1665 }
1666
1667 /*
1668 * find_funcstat_entry - find any existing PgStat_BackendFunctionEntry entry
1669 * for specified function
1670 *
1671 * If no entry, return NULL, don't create a new one
1672 */
1673 PgStat_BackendFunctionEntry *
find_funcstat_entry(Oid func_id)1674 find_funcstat_entry(Oid func_id)
1675 {
1676 if (pgStatFunctions == NULL)
1677 return NULL;
1678
1679 return (PgStat_BackendFunctionEntry *) hash_search(pgStatFunctions,
1680 (void *) &func_id,
1681 HASH_FIND, NULL);
1682 }
1683
1684 /*
1685 * Calculate function call usage and update stat counters.
1686 * Called by the executor after invoking a function.
1687 *
1688 * In the case of a set-returning function that runs in value-per-call mode,
1689 * we will see multiple pgstat_init_function_usage/pgstat_end_function_usage
1690 * calls for what the user considers a single call of the function. The
1691 * finalize flag should be TRUE on the last call.
1692 */
1693 void
pgstat_end_function_usage(PgStat_FunctionCallUsage * fcu,bool finalize)1694 pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize)
1695 {
1696 PgStat_FunctionCounts *fs = fcu->fs;
1697 instr_time f_total;
1698 instr_time f_others;
1699 instr_time f_self;
1700
1701 /* stats not wanted? */
1702 if (fs == NULL)
1703 return;
1704
1705 /* total elapsed time in this function call */
1706 INSTR_TIME_SET_CURRENT(f_total);
1707 INSTR_TIME_SUBTRACT(f_total, fcu->f_start);
1708
1709 /* self usage: elapsed minus anything already charged to other calls */
1710 f_others = total_func_time;
1711 INSTR_TIME_SUBTRACT(f_others, fcu->save_total);
1712 f_self = f_total;
1713 INSTR_TIME_SUBTRACT(f_self, f_others);
1714
1715 /* update backend-wide total time */
1716 INSTR_TIME_ADD(total_func_time, f_self);
1717
1718 /*
1719 * Compute the new f_total_time as the total elapsed time added to the
1720 * pre-call value of f_total_time. This is necessary to avoid
1721 * double-counting any time taken by recursive calls of myself. (We do
1722 * not need any similar kluge for self time, since that already excludes
1723 * any recursive calls.)
1724 */
1725 INSTR_TIME_ADD(f_total, fcu->save_f_total_time);
1726
1727 /* update counters in function stats table */
1728 if (finalize)
1729 fs->f_numcalls++;
1730 fs->f_total_time = f_total;
1731 INSTR_TIME_ADD(fs->f_self_time, f_self);
1732
1733 /* indicate that we have something to send */
1734 have_function_stats = true;
1735 }
1736
1737
1738 /* ----------
1739 * pgstat_initstats() -
1740 *
1741 * Initialize a relcache entry to count access statistics.
1742 * Called whenever a relation is opened.
1743 *
1744 * We assume that a relcache entry's pgstat_info field is zeroed by
1745 * relcache.c when the relcache entry is made; thereafter it is long-lived
1746 * data. We can avoid repeated searches of the TabStatus arrays when the
1747 * same relation is touched repeatedly within a transaction.
1748 * ----------
1749 */
1750 void
pgstat_initstats(Relation rel)1751 pgstat_initstats(Relation rel)
1752 {
1753 Oid rel_id = rel->rd_id;
1754 char relkind = rel->rd_rel->relkind;
1755
1756 /* We only count stats for things that have storage */
1757 if (!(relkind == RELKIND_RELATION ||
1758 relkind == RELKIND_MATVIEW ||
1759 relkind == RELKIND_INDEX ||
1760 relkind == RELKIND_TOASTVALUE ||
1761 relkind == RELKIND_SEQUENCE))
1762 {
1763 rel->pgstat_info = NULL;
1764 return;
1765 }
1766
1767 if (pgStatSock == PGINVALID_SOCKET || !pgstat_track_counts)
1768 {
1769 /* We're not counting at all */
1770 rel->pgstat_info = NULL;
1771 return;
1772 }
1773
1774 /*
1775 * If we already set up this relation in the current transaction, nothing
1776 * to do.
1777 */
1778 if (rel->pgstat_info != NULL &&
1779 rel->pgstat_info->t_id == rel_id)
1780 return;
1781
1782 /* Else find or make the PgStat_TableStatus entry, and update link */
1783 rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared);
1784 }
1785
1786 /*
1787 * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel
1788 */
1789 static PgStat_TableStatus *
get_tabstat_entry(Oid rel_id,bool isshared)1790 get_tabstat_entry(Oid rel_id, bool isshared)
1791 {
1792 TabStatHashEntry *hash_entry;
1793 PgStat_TableStatus *entry;
1794 TabStatusArray *tsa;
1795 bool found;
1796
1797 /*
1798 * Create hash table if we don't have it already.
1799 */
1800 if (pgStatTabHash == NULL)
1801 {
1802 HASHCTL ctl;
1803
1804 memset(&ctl, 0, sizeof(ctl));
1805 ctl.keysize = sizeof(Oid);
1806 ctl.entrysize = sizeof(TabStatHashEntry);
1807
1808 pgStatTabHash = hash_create("pgstat TabStatusArray lookup hash table",
1809 TABSTAT_QUANTUM,
1810 &ctl,
1811 HASH_ELEM | HASH_BLOBS);
1812 }
1813
1814 /*
1815 * Find an entry or create a new one.
1816 */
1817 hash_entry = hash_search(pgStatTabHash, &rel_id, HASH_ENTER, &found);
1818 if (!found)
1819 {
1820 /* initialize new entry with null pointer */
1821 hash_entry->tsa_entry = NULL;
1822 }
1823
1824 /*
1825 * If entry is already valid, we're done.
1826 */
1827 if (hash_entry->tsa_entry)
1828 return hash_entry->tsa_entry;
1829
1830 /*
1831 * Locate the first pgStatTabList entry with free space, making a new list
1832 * entry if needed. Note that we could get an OOM failure here, but if so
1833 * we have left the hashtable and the list in a consistent state.
1834 */
1835 if (pgStatTabList == NULL)
1836 {
1837 /* Set up first pgStatTabList entry */
1838 pgStatTabList = (TabStatusArray *)
1839 MemoryContextAllocZero(TopMemoryContext,
1840 sizeof(TabStatusArray));
1841 }
1842
1843 tsa = pgStatTabList;
1844 while (tsa->tsa_used >= TABSTAT_QUANTUM)
1845 {
1846 if (tsa->tsa_next == NULL)
1847 tsa->tsa_next = (TabStatusArray *)
1848 MemoryContextAllocZero(TopMemoryContext,
1849 sizeof(TabStatusArray));
1850 tsa = tsa->tsa_next;
1851 }
1852
1853 /*
1854 * Allocate a PgStat_TableStatus entry within this list entry. We assume
1855 * the entry was already zeroed, either at creation or after last use.
1856 */
1857 entry = &tsa->tsa_entries[tsa->tsa_used++];
1858 entry->t_id = rel_id;
1859 entry->t_shared = isshared;
1860
1861 /*
1862 * Now we can fill the entry in pgStatTabHash.
1863 */
1864 hash_entry->tsa_entry = entry;
1865
1866 return entry;
1867 }
1868
1869 /*
1870 * find_tabstat_entry - find any existing PgStat_TableStatus entry for rel
1871 *
1872 * If no entry, return NULL, don't create a new one
1873 *
1874 * Note: if we got an error in the most recent execution of pgstat_report_stat,
1875 * it's possible that an entry exists but there's no hashtable entry for it.
1876 * That's okay, we'll treat this case as "doesn't exist".
1877 */
1878 PgStat_TableStatus *
find_tabstat_entry(Oid rel_id)1879 find_tabstat_entry(Oid rel_id)
1880 {
1881 TabStatHashEntry *hash_entry;
1882
1883 /* If hashtable doesn't exist, there are no entries at all */
1884 if (!pgStatTabHash)
1885 return NULL;
1886
1887 hash_entry = hash_search(pgStatTabHash, &rel_id, HASH_FIND, NULL);
1888 if (!hash_entry)
1889 return NULL;
1890
1891 /* Note that this step could also return NULL, but that's correct */
1892 return hash_entry->tsa_entry;
1893 }
1894
1895 /*
1896 * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed
1897 */
1898 static PgStat_SubXactStatus *
get_tabstat_stack_level(int nest_level)1899 get_tabstat_stack_level(int nest_level)
1900 {
1901 PgStat_SubXactStatus *xact_state;
1902
1903 xact_state = pgStatXactStack;
1904 if (xact_state == NULL || xact_state->nest_level != nest_level)
1905 {
1906 xact_state = (PgStat_SubXactStatus *)
1907 MemoryContextAlloc(TopTransactionContext,
1908 sizeof(PgStat_SubXactStatus));
1909 xact_state->nest_level = nest_level;
1910 xact_state->prev = pgStatXactStack;
1911 xact_state->first = NULL;
1912 pgStatXactStack = xact_state;
1913 }
1914 return xact_state;
1915 }
1916
1917 /*
1918 * add_tabstat_xact_level - add a new (sub)transaction state record
1919 */
1920 static void
add_tabstat_xact_level(PgStat_TableStatus * pgstat_info,int nest_level)1921 add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level)
1922 {
1923 PgStat_SubXactStatus *xact_state;
1924 PgStat_TableXactStatus *trans;
1925
1926 /*
1927 * If this is the first rel to be modified at the current nest level, we
1928 * first have to push a transaction stack entry.
1929 */
1930 xact_state = get_tabstat_stack_level(nest_level);
1931
1932 /* Now make a per-table stack entry */
1933 trans = (PgStat_TableXactStatus *)
1934 MemoryContextAllocZero(TopTransactionContext,
1935 sizeof(PgStat_TableXactStatus));
1936 trans->nest_level = nest_level;
1937 trans->upper = pgstat_info->trans;
1938 trans->parent = pgstat_info;
1939 trans->next = xact_state->first;
1940 xact_state->first = trans;
1941 pgstat_info->trans = trans;
1942 }
1943
1944 /*
1945 * pgstat_count_heap_insert - count a tuple insertion of n tuples
1946 */
1947 void
pgstat_count_heap_insert(Relation rel,PgStat_Counter n)1948 pgstat_count_heap_insert(Relation rel, PgStat_Counter n)
1949 {
1950 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1951
1952 if (pgstat_info != NULL)
1953 {
1954 /* We have to log the effect at the proper transactional level */
1955 int nest_level = GetCurrentTransactionNestLevel();
1956
1957 if (pgstat_info->trans == NULL ||
1958 pgstat_info->trans->nest_level != nest_level)
1959 add_tabstat_xact_level(pgstat_info, nest_level);
1960
1961 pgstat_info->trans->tuples_inserted += n;
1962 }
1963 }
1964
1965 /*
1966 * pgstat_count_heap_update - count a tuple update
1967 */
1968 void
pgstat_count_heap_update(Relation rel,bool hot)1969 pgstat_count_heap_update(Relation rel, bool hot)
1970 {
1971 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1972
1973 if (pgstat_info != NULL)
1974 {
1975 /* We have to log the effect at the proper transactional level */
1976 int nest_level = GetCurrentTransactionNestLevel();
1977
1978 if (pgstat_info->trans == NULL ||
1979 pgstat_info->trans->nest_level != nest_level)
1980 add_tabstat_xact_level(pgstat_info, nest_level);
1981
1982 pgstat_info->trans->tuples_updated++;
1983
1984 /* t_tuples_hot_updated is nontransactional, so just advance it */
1985 if (hot)
1986 pgstat_info->t_counts.t_tuples_hot_updated++;
1987 }
1988 }
1989
1990 /*
1991 * pgstat_count_heap_delete - count a tuple deletion
1992 */
1993 void
pgstat_count_heap_delete(Relation rel)1994 pgstat_count_heap_delete(Relation rel)
1995 {
1996 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
1997
1998 if (pgstat_info != NULL)
1999 {
2000 /* We have to log the effect at the proper transactional level */
2001 int nest_level = GetCurrentTransactionNestLevel();
2002
2003 if (pgstat_info->trans == NULL ||
2004 pgstat_info->trans->nest_level != nest_level)
2005 add_tabstat_xact_level(pgstat_info, nest_level);
2006
2007 pgstat_info->trans->tuples_deleted++;
2008 }
2009 }
2010
2011 /*
2012 * pgstat_truncate_save_counters
2013 *
2014 * Whenever a table is truncated, we save its i/u/d counters so that they can
2015 * be cleared, and if the (sub)xact that executed the truncate later aborts,
2016 * the counters can be restored to the saved (pre-truncate) values. Note we do
2017 * this on the first truncate in any particular subxact level only.
2018 */
2019 static void
pgstat_truncate_save_counters(PgStat_TableXactStatus * trans)2020 pgstat_truncate_save_counters(PgStat_TableXactStatus *trans)
2021 {
2022 if (!trans->truncated)
2023 {
2024 trans->inserted_pre_trunc = trans->tuples_inserted;
2025 trans->updated_pre_trunc = trans->tuples_updated;
2026 trans->deleted_pre_trunc = trans->tuples_deleted;
2027 trans->truncated = true;
2028 }
2029 }
2030
2031 /*
2032 * pgstat_truncate_restore_counters - restore counters when a truncate aborts
2033 */
2034 static void
pgstat_truncate_restore_counters(PgStat_TableXactStatus * trans)2035 pgstat_truncate_restore_counters(PgStat_TableXactStatus *trans)
2036 {
2037 if (trans->truncated)
2038 {
2039 trans->tuples_inserted = trans->inserted_pre_trunc;
2040 trans->tuples_updated = trans->updated_pre_trunc;
2041 trans->tuples_deleted = trans->deleted_pre_trunc;
2042 }
2043 }
2044
2045 /*
2046 * pgstat_count_truncate - update tuple counters due to truncate
2047 */
2048 void
pgstat_count_truncate(Relation rel)2049 pgstat_count_truncate(Relation rel)
2050 {
2051 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
2052
2053 if (pgstat_info != NULL)
2054 {
2055 /* We have to log the effect at the proper transactional level */
2056 int nest_level = GetCurrentTransactionNestLevel();
2057
2058 if (pgstat_info->trans == NULL ||
2059 pgstat_info->trans->nest_level != nest_level)
2060 add_tabstat_xact_level(pgstat_info, nest_level);
2061
2062 pgstat_truncate_save_counters(pgstat_info->trans);
2063 pgstat_info->trans->tuples_inserted = 0;
2064 pgstat_info->trans->tuples_updated = 0;
2065 pgstat_info->trans->tuples_deleted = 0;
2066 }
2067 }
2068
2069 /*
2070 * pgstat_update_heap_dead_tuples - update dead-tuples count
2071 *
2072 * The semantics of this are that we are reporting the nontransactional
2073 * recovery of "delta" dead tuples; so t_delta_dead_tuples decreases
2074 * rather than increasing, and the change goes straight into the per-table
2075 * counter, not into transactional state.
2076 */
2077 void
pgstat_update_heap_dead_tuples(Relation rel,int delta)2078 pgstat_update_heap_dead_tuples(Relation rel, int delta)
2079 {
2080 PgStat_TableStatus *pgstat_info = rel->pgstat_info;
2081
2082 if (pgstat_info != NULL)
2083 pgstat_info->t_counts.t_delta_dead_tuples -= delta;
2084 }
2085
2086
2087 /* ----------
2088 * AtEOXact_PgStat
2089 *
2090 * Called from access/transam/xact.c at top-level transaction commit/abort.
2091 * ----------
2092 */
2093 void
AtEOXact_PgStat(bool isCommit,bool parallel)2094 AtEOXact_PgStat(bool isCommit, bool parallel)
2095 {
2096 PgStat_SubXactStatus *xact_state;
2097
2098 /* Don't count parallel worker transaction stats */
2099 if (!parallel)
2100 {
2101 /*
2102 * Count transaction commit or abort. (We use counters, not just
2103 * bools, in case the reporting message isn't sent right away.)
2104 */
2105 if (isCommit)
2106 pgStatXactCommit++;
2107 else
2108 pgStatXactRollback++;
2109 }
2110
2111 /*
2112 * Transfer transactional insert/update counts into the base tabstat
2113 * entries. We don't bother to free any of the transactional state, since
2114 * it's all in TopTransactionContext and will go away anyway.
2115 */
2116 xact_state = pgStatXactStack;
2117 if (xact_state != NULL)
2118 {
2119 PgStat_TableXactStatus *trans;
2120
2121 Assert(xact_state->nest_level == 1);
2122 Assert(xact_state->prev == NULL);
2123 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2124 {
2125 PgStat_TableStatus *tabstat;
2126
2127 Assert(trans->nest_level == 1);
2128 Assert(trans->upper == NULL);
2129 tabstat = trans->parent;
2130 Assert(tabstat->trans == trans);
2131 /* restore pre-truncate stats (if any) in case of aborted xact */
2132 if (!isCommit)
2133 pgstat_truncate_restore_counters(trans);
2134 /* count attempted actions regardless of commit/abort */
2135 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
2136 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
2137 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
2138 if (isCommit)
2139 {
2140 tabstat->t_counts.t_truncated = trans->truncated;
2141 if (trans->truncated)
2142 {
2143 /* forget live/dead stats seen by backend thus far */
2144 tabstat->t_counts.t_delta_live_tuples = 0;
2145 tabstat->t_counts.t_delta_dead_tuples = 0;
2146 }
2147 /* insert adds a live tuple, delete removes one */
2148 tabstat->t_counts.t_delta_live_tuples +=
2149 trans->tuples_inserted - trans->tuples_deleted;
2150 /* update and delete each create a dead tuple */
2151 tabstat->t_counts.t_delta_dead_tuples +=
2152 trans->tuples_updated + trans->tuples_deleted;
2153 /* insert, update, delete each count as one change event */
2154 tabstat->t_counts.t_changed_tuples +=
2155 trans->tuples_inserted + trans->tuples_updated +
2156 trans->tuples_deleted;
2157 }
2158 else
2159 {
2160 /* inserted tuples are dead, deleted tuples are unaffected */
2161 tabstat->t_counts.t_delta_dead_tuples +=
2162 trans->tuples_inserted + trans->tuples_updated;
2163 /* an aborted xact generates no changed_tuple events */
2164 }
2165 tabstat->trans = NULL;
2166 }
2167 }
2168 pgStatXactStack = NULL;
2169
2170 /* Make sure any stats snapshot is thrown away */
2171 pgstat_clear_snapshot();
2172 }
2173
2174 /* ----------
2175 * AtEOSubXact_PgStat
2176 *
2177 * Called from access/transam/xact.c at subtransaction commit/abort.
2178 * ----------
2179 */
2180 void
AtEOSubXact_PgStat(bool isCommit,int nestDepth)2181 AtEOSubXact_PgStat(bool isCommit, int nestDepth)
2182 {
2183 PgStat_SubXactStatus *xact_state;
2184
2185 /*
2186 * Transfer transactional insert/update counts into the next higher
2187 * subtransaction state.
2188 */
2189 xact_state = pgStatXactStack;
2190 if (xact_state != NULL &&
2191 xact_state->nest_level >= nestDepth)
2192 {
2193 PgStat_TableXactStatus *trans;
2194 PgStat_TableXactStatus *next_trans;
2195
2196 /* delink xact_state from stack immediately to simplify reuse case */
2197 pgStatXactStack = xact_state->prev;
2198
2199 for (trans = xact_state->first; trans != NULL; trans = next_trans)
2200 {
2201 PgStat_TableStatus *tabstat;
2202
2203 next_trans = trans->next;
2204 Assert(trans->nest_level == nestDepth);
2205 tabstat = trans->parent;
2206 Assert(tabstat->trans == trans);
2207 if (isCommit)
2208 {
2209 if (trans->upper && trans->upper->nest_level == nestDepth - 1)
2210 {
2211 if (trans->truncated)
2212 {
2213 /* propagate the truncate status one level up */
2214 pgstat_truncate_save_counters(trans->upper);
2215 /* replace upper xact stats with ours */
2216 trans->upper->tuples_inserted = trans->tuples_inserted;
2217 trans->upper->tuples_updated = trans->tuples_updated;
2218 trans->upper->tuples_deleted = trans->tuples_deleted;
2219 }
2220 else
2221 {
2222 trans->upper->tuples_inserted += trans->tuples_inserted;
2223 trans->upper->tuples_updated += trans->tuples_updated;
2224 trans->upper->tuples_deleted += trans->tuples_deleted;
2225 }
2226 tabstat->trans = trans->upper;
2227 pfree(trans);
2228 }
2229 else
2230 {
2231 /*
2232 * When there isn't an immediate parent state, we can just
2233 * reuse the record instead of going through a
2234 * palloc/pfree pushup (this works since it's all in
2235 * TopTransactionContext anyway). We have to re-link it
2236 * into the parent level, though, and that might mean
2237 * pushing a new entry into the pgStatXactStack.
2238 */
2239 PgStat_SubXactStatus *upper_xact_state;
2240
2241 upper_xact_state = get_tabstat_stack_level(nestDepth - 1);
2242 trans->next = upper_xact_state->first;
2243 upper_xact_state->first = trans;
2244 trans->nest_level = nestDepth - 1;
2245 }
2246 }
2247 else
2248 {
2249 /*
2250 * On abort, update top-level tabstat counts, then forget the
2251 * subtransaction
2252 */
2253
2254 /* first restore values obliterated by truncate */
2255 pgstat_truncate_restore_counters(trans);
2256 /* count attempted actions regardless of commit/abort */
2257 tabstat->t_counts.t_tuples_inserted += trans->tuples_inserted;
2258 tabstat->t_counts.t_tuples_updated += trans->tuples_updated;
2259 tabstat->t_counts.t_tuples_deleted += trans->tuples_deleted;
2260 /* inserted tuples are dead, deleted tuples are unaffected */
2261 tabstat->t_counts.t_delta_dead_tuples +=
2262 trans->tuples_inserted + trans->tuples_updated;
2263 tabstat->trans = trans->upper;
2264 pfree(trans);
2265 }
2266 }
2267 pfree(xact_state);
2268 }
2269 }
2270
2271
2272 /*
2273 * AtPrepare_PgStat
2274 * Save the transactional stats state at 2PC transaction prepare.
2275 *
2276 * In this phase we just generate 2PC records for all the pending
2277 * transaction-dependent stats work.
2278 */
2279 void
AtPrepare_PgStat(void)2280 AtPrepare_PgStat(void)
2281 {
2282 PgStat_SubXactStatus *xact_state;
2283
2284 xact_state = pgStatXactStack;
2285 if (xact_state != NULL)
2286 {
2287 PgStat_TableXactStatus *trans;
2288
2289 Assert(xact_state->nest_level == 1);
2290 Assert(xact_state->prev == NULL);
2291 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2292 {
2293 PgStat_TableStatus *tabstat;
2294 TwoPhasePgStatRecord record;
2295
2296 Assert(trans->nest_level == 1);
2297 Assert(trans->upper == NULL);
2298 tabstat = trans->parent;
2299 Assert(tabstat->trans == trans);
2300
2301 record.tuples_inserted = trans->tuples_inserted;
2302 record.tuples_updated = trans->tuples_updated;
2303 record.tuples_deleted = trans->tuples_deleted;
2304 record.inserted_pre_trunc = trans->inserted_pre_trunc;
2305 record.updated_pre_trunc = trans->updated_pre_trunc;
2306 record.deleted_pre_trunc = trans->deleted_pre_trunc;
2307 record.t_id = tabstat->t_id;
2308 record.t_shared = tabstat->t_shared;
2309 record.t_truncated = trans->truncated;
2310
2311 RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0,
2312 &record, sizeof(TwoPhasePgStatRecord));
2313 }
2314 }
2315 }
2316
2317 /*
2318 * PostPrepare_PgStat
2319 * Clean up after successful PREPARE.
2320 *
2321 * All we need do here is unlink the transaction stats state from the
2322 * nontransactional state. The nontransactional action counts will be
2323 * reported to the stats collector immediately, while the effects on live
2324 * and dead tuple counts are preserved in the 2PC state file.
2325 *
2326 * Note: AtEOXact_PgStat is not called during PREPARE.
2327 */
2328 void
PostPrepare_PgStat(void)2329 PostPrepare_PgStat(void)
2330 {
2331 PgStat_SubXactStatus *xact_state;
2332
2333 /*
2334 * We don't bother to free any of the transactional state, since it's all
2335 * in TopTransactionContext and will go away anyway.
2336 */
2337 xact_state = pgStatXactStack;
2338 if (xact_state != NULL)
2339 {
2340 PgStat_TableXactStatus *trans;
2341
2342 for (trans = xact_state->first; trans != NULL; trans = trans->next)
2343 {
2344 PgStat_TableStatus *tabstat;
2345
2346 tabstat = trans->parent;
2347 tabstat->trans = NULL;
2348 }
2349 }
2350 pgStatXactStack = NULL;
2351
2352 /* Make sure any stats snapshot is thrown away */
2353 pgstat_clear_snapshot();
2354 }
2355
2356 /*
2357 * 2PC processing routine for COMMIT PREPARED case.
2358 *
2359 * Load the saved counts into our local pgstats state.
2360 */
2361 void
pgstat_twophase_postcommit(TransactionId xid,uint16 info,void * recdata,uint32 len)2362 pgstat_twophase_postcommit(TransactionId xid, uint16 info,
2363 void *recdata, uint32 len)
2364 {
2365 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2366 PgStat_TableStatus *pgstat_info;
2367
2368 /* Find or create a tabstat entry for the rel */
2369 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2370
2371 /* Same math as in AtEOXact_PgStat, commit case */
2372 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2373 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2374 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2375 pgstat_info->t_counts.t_truncated = rec->t_truncated;
2376 if (rec->t_truncated)
2377 {
2378 /* forget live/dead stats seen by backend thus far */
2379 pgstat_info->t_counts.t_delta_live_tuples = 0;
2380 pgstat_info->t_counts.t_delta_dead_tuples = 0;
2381 }
2382 pgstat_info->t_counts.t_delta_live_tuples +=
2383 rec->tuples_inserted - rec->tuples_deleted;
2384 pgstat_info->t_counts.t_delta_dead_tuples +=
2385 rec->tuples_updated + rec->tuples_deleted;
2386 pgstat_info->t_counts.t_changed_tuples +=
2387 rec->tuples_inserted + rec->tuples_updated +
2388 rec->tuples_deleted;
2389 }
2390
2391 /*
2392 * 2PC processing routine for ROLLBACK PREPARED case.
2393 *
2394 * Load the saved counts into our local pgstats state, but treat them
2395 * as aborted.
2396 */
2397 void
pgstat_twophase_postabort(TransactionId xid,uint16 info,void * recdata,uint32 len)2398 pgstat_twophase_postabort(TransactionId xid, uint16 info,
2399 void *recdata, uint32 len)
2400 {
2401 TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata;
2402 PgStat_TableStatus *pgstat_info;
2403
2404 /* Find or create a tabstat entry for the rel */
2405 pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared);
2406
2407 /* Same math as in AtEOXact_PgStat, abort case */
2408 if (rec->t_truncated)
2409 {
2410 rec->tuples_inserted = rec->inserted_pre_trunc;
2411 rec->tuples_updated = rec->updated_pre_trunc;
2412 rec->tuples_deleted = rec->deleted_pre_trunc;
2413 }
2414 pgstat_info->t_counts.t_tuples_inserted += rec->tuples_inserted;
2415 pgstat_info->t_counts.t_tuples_updated += rec->tuples_updated;
2416 pgstat_info->t_counts.t_tuples_deleted += rec->tuples_deleted;
2417 pgstat_info->t_counts.t_delta_dead_tuples +=
2418 rec->tuples_inserted + rec->tuples_updated;
2419 }
2420
2421
2422 /* ----------
2423 * pgstat_fetch_stat_dbentry() -
2424 *
2425 * Support function for the SQL-callable pgstat* functions. Returns
2426 * the collected statistics for one database or NULL. NULL doesn't mean
2427 * that the database doesn't exist, it is just not yet known by the
2428 * collector, so the caller is better off to report ZERO instead.
2429 * ----------
2430 */
2431 PgStat_StatDBEntry *
pgstat_fetch_stat_dbentry(Oid dbid)2432 pgstat_fetch_stat_dbentry(Oid dbid)
2433 {
2434 /*
2435 * If not done for this transaction, read the statistics collector stats
2436 * file into some hash tables.
2437 */
2438 backend_read_statsfile();
2439
2440 /*
2441 * Lookup the requested database; return NULL if not found
2442 */
2443 return (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2444 (void *) &dbid,
2445 HASH_FIND, NULL);
2446 }
2447
2448
2449 /* ----------
2450 * pgstat_fetch_stat_tabentry() -
2451 *
2452 * Support function for the SQL-callable pgstat* functions. Returns
2453 * the collected statistics for one table or NULL. NULL doesn't mean
2454 * that the table doesn't exist, it is just not yet known by the
2455 * collector, so the caller is better off to report ZERO instead.
2456 * ----------
2457 */
2458 PgStat_StatTabEntry *
pgstat_fetch_stat_tabentry(Oid relid)2459 pgstat_fetch_stat_tabentry(Oid relid)
2460 {
2461 Oid dbid;
2462 PgStat_StatDBEntry *dbentry;
2463 PgStat_StatTabEntry *tabentry;
2464
2465 /*
2466 * If not done for this transaction, read the statistics collector stats
2467 * file into some hash tables.
2468 */
2469 backend_read_statsfile();
2470
2471 /*
2472 * Lookup our database, then look in its table hash table.
2473 */
2474 dbid = MyDatabaseId;
2475 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2476 (void *) &dbid,
2477 HASH_FIND, NULL);
2478 if (dbentry != NULL && dbentry->tables != NULL)
2479 {
2480 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2481 (void *) &relid,
2482 HASH_FIND, NULL);
2483 if (tabentry)
2484 return tabentry;
2485 }
2486
2487 /*
2488 * If we didn't find it, maybe it's a shared table.
2489 */
2490 dbid = InvalidOid;
2491 dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
2492 (void *) &dbid,
2493 HASH_FIND, NULL);
2494 if (dbentry != NULL && dbentry->tables != NULL)
2495 {
2496 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
2497 (void *) &relid,
2498 HASH_FIND, NULL);
2499 if (tabentry)
2500 return tabentry;
2501 }
2502
2503 return NULL;
2504 }
2505
2506
2507 /* ----------
2508 * pgstat_fetch_stat_funcentry() -
2509 *
2510 * Support function for the SQL-callable pgstat* functions. Returns
2511 * the collected statistics for one function or NULL.
2512 * ----------
2513 */
2514 PgStat_StatFuncEntry *
pgstat_fetch_stat_funcentry(Oid func_id)2515 pgstat_fetch_stat_funcentry(Oid func_id)
2516 {
2517 PgStat_StatDBEntry *dbentry;
2518 PgStat_StatFuncEntry *funcentry = NULL;
2519
2520 /* load the stats file if needed */
2521 backend_read_statsfile();
2522
2523 /* Lookup our database, then find the requested function. */
2524 dbentry = pgstat_fetch_stat_dbentry(MyDatabaseId);
2525 if (dbentry != NULL && dbentry->functions != NULL)
2526 {
2527 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
2528 (void *) &func_id,
2529 HASH_FIND, NULL);
2530 }
2531
2532 return funcentry;
2533 }
2534
2535
2536 /* ----------
2537 * pgstat_fetch_stat_beentry() -
2538 *
2539 * Support function for the SQL-callable pgstat* functions. Returns
2540 * our local copy of the current-activity entry for one backend.
2541 *
2542 * NB: caller is responsible for a check if the user is permitted to see
2543 * this info (especially the querystring).
2544 * ----------
2545 */
2546 PgBackendStatus *
pgstat_fetch_stat_beentry(int beid)2547 pgstat_fetch_stat_beentry(int beid)
2548 {
2549 pgstat_read_current_status();
2550
2551 if (beid < 1 || beid > localNumBackends)
2552 return NULL;
2553
2554 return &localBackendStatusTable[beid - 1].backendStatus;
2555 }
2556
2557
2558 /* ----------
2559 * pgstat_fetch_stat_local_beentry() -
2560 *
2561 * Like pgstat_fetch_stat_beentry() but with locally computed additions (like
2562 * xid and xmin values of the backend)
2563 *
2564 * NB: caller is responsible for a check if the user is permitted to see
2565 * this info (especially the querystring).
2566 * ----------
2567 */
2568 LocalPgBackendStatus *
pgstat_fetch_stat_local_beentry(int beid)2569 pgstat_fetch_stat_local_beentry(int beid)
2570 {
2571 pgstat_read_current_status();
2572
2573 if (beid < 1 || beid > localNumBackends)
2574 return NULL;
2575
2576 return &localBackendStatusTable[beid - 1];
2577 }
2578
2579
2580 /* ----------
2581 * pgstat_fetch_stat_numbackends() -
2582 *
2583 * Support function for the SQL-callable pgstat* functions. Returns
2584 * the maximum current backend id.
2585 * ----------
2586 */
2587 int
pgstat_fetch_stat_numbackends(void)2588 pgstat_fetch_stat_numbackends(void)
2589 {
2590 pgstat_read_current_status();
2591
2592 return localNumBackends;
2593 }
2594
2595 /*
2596 * ---------
2597 * pgstat_fetch_stat_archiver() -
2598 *
2599 * Support function for the SQL-callable pgstat* functions. Returns
2600 * a pointer to the archiver statistics struct.
2601 * ---------
2602 */
2603 PgStat_ArchiverStats *
pgstat_fetch_stat_archiver(void)2604 pgstat_fetch_stat_archiver(void)
2605 {
2606 backend_read_statsfile();
2607
2608 return &archiverStats;
2609 }
2610
2611
2612 /*
2613 * ---------
2614 * pgstat_fetch_global() -
2615 *
2616 * Support function for the SQL-callable pgstat* functions. Returns
2617 * a pointer to the global statistics struct.
2618 * ---------
2619 */
2620 PgStat_GlobalStats *
pgstat_fetch_global(void)2621 pgstat_fetch_global(void)
2622 {
2623 backend_read_statsfile();
2624
2625 return &globalStats;
2626 }
2627
2628
2629 /* ------------------------------------------------------------
2630 * Functions for management of the shared-memory PgBackendStatus array
2631 * ------------------------------------------------------------
2632 */
2633
2634 static PgBackendStatus *BackendStatusArray = NULL;
2635 static PgBackendStatus *MyBEEntry = NULL;
2636 static char *BackendAppnameBuffer = NULL;
2637 static char *BackendClientHostnameBuffer = NULL;
2638 static char *BackendActivityBuffer = NULL;
2639 static Size BackendActivityBufferSize = 0;
2640 #ifdef USE_SSL
2641 static PgBackendSSLStatus *BackendSslStatusBuffer = NULL;
2642 #endif
2643 #ifdef ENABLE_GSS
2644 static PgBackendGSSStatus *BackendGssStatusBuffer = NULL;
2645 #endif
2646
2647
2648 /*
2649 * Report shared-memory space needed by CreateSharedBackendStatus.
2650 */
2651 Size
BackendStatusShmemSize(void)2652 BackendStatusShmemSize(void)
2653 {
2654 Size size;
2655
2656 /* BackendStatusArray: */
2657 size = mul_size(sizeof(PgBackendStatus), NumBackendStatSlots);
2658 /* BackendAppnameBuffer: */
2659 size = add_size(size,
2660 mul_size(NAMEDATALEN, NumBackendStatSlots));
2661 /* BackendClientHostnameBuffer: */
2662 size = add_size(size,
2663 mul_size(NAMEDATALEN, NumBackendStatSlots));
2664 /* BackendActivityBuffer: */
2665 size = add_size(size,
2666 mul_size(pgstat_track_activity_query_size, NumBackendStatSlots));
2667 #ifdef USE_SSL
2668 /* BackendSslStatusBuffer: */
2669 size = add_size(size,
2670 mul_size(sizeof(PgBackendSSLStatus), NumBackendStatSlots));
2671 #endif
2672 #ifdef ENABLE_GSS
2673 /* BackendGssStatusBuffer: */
2674 size = add_size(size,
2675 mul_size(sizeof(PgBackendGSSStatus), NumBackendStatSlots));
2676 #endif
2677 return size;
2678 }
2679
2680 /*
2681 * Initialize the shared status array and several string buffers
2682 * during postmaster startup.
2683 */
2684 void
CreateSharedBackendStatus(void)2685 CreateSharedBackendStatus(void)
2686 {
2687 Size size;
2688 bool found;
2689 int i;
2690 char *buffer;
2691
2692 /* Create or attach to the shared array */
2693 size = mul_size(sizeof(PgBackendStatus), NumBackendStatSlots);
2694 BackendStatusArray = (PgBackendStatus *)
2695 ShmemInitStruct("Backend Status Array", size, &found);
2696
2697 if (!found)
2698 {
2699 /*
2700 * We're the first - initialize.
2701 */
2702 MemSet(BackendStatusArray, 0, size);
2703 }
2704
2705 /* Create or attach to the shared appname buffer */
2706 size = mul_size(NAMEDATALEN, NumBackendStatSlots);
2707 BackendAppnameBuffer = (char *)
2708 ShmemInitStruct("Backend Application Name Buffer", size, &found);
2709
2710 if (!found)
2711 {
2712 MemSet(BackendAppnameBuffer, 0, size);
2713
2714 /* Initialize st_appname pointers. */
2715 buffer = BackendAppnameBuffer;
2716 for (i = 0; i < NumBackendStatSlots; i++)
2717 {
2718 BackendStatusArray[i].st_appname = buffer;
2719 buffer += NAMEDATALEN;
2720 }
2721 }
2722
2723 /* Create or attach to the shared client hostname buffer */
2724 size = mul_size(NAMEDATALEN, NumBackendStatSlots);
2725 BackendClientHostnameBuffer = (char *)
2726 ShmemInitStruct("Backend Client Host Name Buffer", size, &found);
2727
2728 if (!found)
2729 {
2730 MemSet(BackendClientHostnameBuffer, 0, size);
2731
2732 /* Initialize st_clienthostname pointers. */
2733 buffer = BackendClientHostnameBuffer;
2734 for (i = 0; i < NumBackendStatSlots; i++)
2735 {
2736 BackendStatusArray[i].st_clienthostname = buffer;
2737 buffer += NAMEDATALEN;
2738 }
2739 }
2740
2741 /* Create or attach to the shared activity buffer */
2742 BackendActivityBufferSize = mul_size(pgstat_track_activity_query_size,
2743 NumBackendStatSlots);
2744 BackendActivityBuffer = (char *)
2745 ShmemInitStruct("Backend Activity Buffer",
2746 BackendActivityBufferSize,
2747 &found);
2748
2749 if (!found)
2750 {
2751 MemSet(BackendActivityBuffer, 0, BackendActivityBufferSize);
2752
2753 /* Initialize st_activity pointers. */
2754 buffer = BackendActivityBuffer;
2755 for (i = 0; i < NumBackendStatSlots; i++)
2756 {
2757 BackendStatusArray[i].st_activity_raw = buffer;
2758 buffer += pgstat_track_activity_query_size;
2759 }
2760 }
2761
2762 #ifdef USE_SSL
2763 /* Create or attach to the shared SSL status buffer */
2764 size = mul_size(sizeof(PgBackendSSLStatus), NumBackendStatSlots);
2765 BackendSslStatusBuffer = (PgBackendSSLStatus *)
2766 ShmemInitStruct("Backend SSL Status Buffer", size, &found);
2767
2768 if (!found)
2769 {
2770 PgBackendSSLStatus *ptr;
2771
2772 MemSet(BackendSslStatusBuffer, 0, size);
2773
2774 /* Initialize st_sslstatus pointers. */
2775 ptr = BackendSslStatusBuffer;
2776 for (i = 0; i < NumBackendStatSlots; i++)
2777 {
2778 BackendStatusArray[i].st_sslstatus = ptr;
2779 ptr++;
2780 }
2781 }
2782 #endif
2783
2784 #ifdef ENABLE_GSS
2785 /* Create or attach to the shared GSSAPI status buffer */
2786 size = mul_size(sizeof(PgBackendGSSStatus), NumBackendStatSlots);
2787 BackendGssStatusBuffer = (PgBackendGSSStatus *)
2788 ShmemInitStruct("Backend GSS Status Buffer", size, &found);
2789
2790 if (!found)
2791 {
2792 PgBackendGSSStatus *ptr;
2793
2794 MemSet(BackendGssStatusBuffer, 0, size);
2795
2796 /* Initialize st_gssstatus pointers. */
2797 ptr = BackendGssStatusBuffer;
2798 for (i = 0; i < NumBackendStatSlots; i++)
2799 {
2800 BackendStatusArray[i].st_gssstatus = ptr;
2801 ptr++;
2802 }
2803 }
2804 #endif
2805 }
2806
2807
2808 /* ----------
2809 * pgstat_initialize() -
2810 *
2811 * Initialize pgstats state, and set up our on-proc-exit hook.
2812 * Called from InitPostgres and AuxiliaryProcessMain. For auxiliary process,
2813 * MyBackendId is invalid. Otherwise, MyBackendId must be set,
2814 * but we must not have started any transaction yet (since the
2815 * exit hook must run after the last transaction exit).
2816 * NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful.
2817 * ----------
2818 */
2819 void
pgstat_initialize(void)2820 pgstat_initialize(void)
2821 {
2822 /* Initialize MyBEEntry */
2823 if (MyBackendId != InvalidBackendId)
2824 {
2825 Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends);
2826 MyBEEntry = &BackendStatusArray[MyBackendId - 1];
2827 }
2828 else
2829 {
2830 /* Must be an auxiliary process */
2831 Assert(MyAuxProcType != NotAnAuxProcess);
2832
2833 /*
2834 * Assign the MyBEEntry for an auxiliary process. Since it doesn't
2835 * have a BackendId, the slot is statically allocated based on the
2836 * auxiliary process type (MyAuxProcType). Backends use slots indexed
2837 * in the range from 1 to MaxBackends (inclusive), so we use
2838 * MaxBackends + AuxBackendType + 1 as the index of the slot for an
2839 * auxiliary process.
2840 */
2841 MyBEEntry = &BackendStatusArray[MaxBackends + MyAuxProcType];
2842 }
2843
2844 /* Set up a process-exit hook to clean up */
2845 on_shmem_exit(pgstat_beshutdown_hook, 0);
2846 }
2847
2848 /* ----------
2849 * pgstat_bestart() -
2850 *
2851 * Initialize this backend's entry in the PgBackendStatus array.
2852 * Called from InitPostgres.
2853 *
2854 * Apart from auxiliary processes, MyBackendId, MyDatabaseId,
2855 * session userid, and application_name must be set for a
2856 * backend (hence, this cannot be combined with pgstat_initialize).
2857 * Note also that we must be inside a transaction if this isn't an aux
2858 * process, as we may need to do encoding conversion on some strings.
2859 * ----------
2860 */
2861 void
pgstat_bestart(void)2862 pgstat_bestart(void)
2863 {
2864 volatile PgBackendStatus *vbeentry = MyBEEntry;
2865 PgBackendStatus lbeentry;
2866 #ifdef USE_SSL
2867 PgBackendSSLStatus lsslstatus;
2868 #endif
2869 #ifdef ENABLE_GSS
2870 PgBackendGSSStatus lgssstatus;
2871 #endif
2872
2873 /* pgstats state must be initialized from pgstat_initialize() */
2874 Assert(vbeentry != NULL);
2875
2876 /*
2877 * To minimize the time spent modifying the PgBackendStatus entry, and
2878 * avoid risk of errors inside the critical section, we first copy the
2879 * shared-memory struct to a local variable, then modify the data in the
2880 * local variable, then copy the local variable back to shared memory.
2881 * Only the last step has to be inside the critical section.
2882 *
2883 * Most of the data we copy from shared memory is just going to be
2884 * overwritten, but the struct's not so large that it's worth the
2885 * maintenance hassle to copy only the needful fields.
2886 */
2887 memcpy(&lbeentry,
2888 unvolatize(PgBackendStatus *, vbeentry),
2889 sizeof(PgBackendStatus));
2890
2891 /* These structs can just start from zeroes each time, though */
2892 #ifdef USE_SSL
2893 memset(&lsslstatus, 0, sizeof(lsslstatus));
2894 #endif
2895 #ifdef ENABLE_GSS
2896 memset(&lgssstatus, 0, sizeof(lgssstatus));
2897 #endif
2898
2899 /*
2900 * Now fill in all the fields of lbeentry, except for strings that are
2901 * out-of-line data. Those have to be handled separately, below.
2902 */
2903 lbeentry.st_procpid = MyProcPid;
2904
2905 if (MyBackendId != InvalidBackendId)
2906 {
2907 if (IsAutoVacuumLauncherProcess())
2908 {
2909 /* Autovacuum Launcher */
2910 lbeentry.st_backendType = B_AUTOVAC_LAUNCHER;
2911 }
2912 else if (IsAutoVacuumWorkerProcess())
2913 {
2914 /* Autovacuum Worker */
2915 lbeentry.st_backendType = B_AUTOVAC_WORKER;
2916 }
2917 else if (am_walsender)
2918 {
2919 /* Wal sender */
2920 lbeentry.st_backendType = B_WAL_SENDER;
2921 }
2922 else if (IsBackgroundWorker)
2923 {
2924 /* bgworker */
2925 lbeentry.st_backendType = B_BG_WORKER;
2926 }
2927 else
2928 {
2929 /* client-backend */
2930 lbeentry.st_backendType = B_BACKEND;
2931 }
2932 }
2933 else
2934 {
2935 /* Must be an auxiliary process */
2936 Assert(MyAuxProcType != NotAnAuxProcess);
2937 switch (MyAuxProcType)
2938 {
2939 case StartupProcess:
2940 lbeentry.st_backendType = B_STARTUP;
2941 break;
2942 case BgWriterProcess:
2943 lbeentry.st_backendType = B_BG_WRITER;
2944 break;
2945 case CheckpointerProcess:
2946 lbeentry.st_backendType = B_CHECKPOINTER;
2947 break;
2948 case WalWriterProcess:
2949 lbeentry.st_backendType = B_WAL_WRITER;
2950 break;
2951 case WalReceiverProcess:
2952 lbeentry.st_backendType = B_WAL_RECEIVER;
2953 break;
2954 default:
2955 elog(FATAL, "unrecognized process type: %d",
2956 (int) MyAuxProcType);
2957 }
2958 }
2959
2960 lbeentry.st_proc_start_timestamp = MyStartTimestamp;
2961 lbeentry.st_activity_start_timestamp = 0;
2962 lbeentry.st_state_start_timestamp = 0;
2963 lbeentry.st_xact_start_timestamp = 0;
2964 lbeentry.st_databaseid = MyDatabaseId;
2965
2966 /* We have userid for client-backends, wal-sender and bgworker processes */
2967 if (lbeentry.st_backendType == B_BACKEND
2968 || lbeentry.st_backendType == B_WAL_SENDER
2969 || lbeentry.st_backendType == B_BG_WORKER)
2970 lbeentry.st_userid = GetSessionUserId();
2971 else
2972 lbeentry.st_userid = InvalidOid;
2973
2974 /*
2975 * We may not have a MyProcPort (eg, if this is the autovacuum process).
2976 * If so, use all-zeroes client address, which is dealt with specially in
2977 * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port.
2978 */
2979 if (MyProcPort)
2980 memcpy(&lbeentry.st_clientaddr, &MyProcPort->raddr,
2981 sizeof(lbeentry.st_clientaddr));
2982 else
2983 MemSet(&lbeentry.st_clientaddr, 0, sizeof(lbeentry.st_clientaddr));
2984
2985 #ifdef USE_SSL
2986 if (MyProcPort && MyProcPort->ssl != NULL)
2987 {
2988 lbeentry.st_ssl = true;
2989 lsslstatus.ssl_bits = be_tls_get_cipher_bits(MyProcPort);
2990 lsslstatus.ssl_compression = be_tls_get_compression(MyProcPort);
2991 strlcpy(lsslstatus.ssl_version, be_tls_get_version(MyProcPort), NAMEDATALEN);
2992 strlcpy(lsslstatus.ssl_cipher, be_tls_get_cipher(MyProcPort), NAMEDATALEN);
2993 be_tls_get_peer_subject_name(MyProcPort, lsslstatus.ssl_client_dn, NAMEDATALEN);
2994 be_tls_get_peer_serial(MyProcPort, lsslstatus.ssl_client_serial, NAMEDATALEN);
2995 be_tls_get_peer_issuer_name(MyProcPort, lsslstatus.ssl_issuer_dn, NAMEDATALEN);
2996 }
2997 else
2998 {
2999 lbeentry.st_ssl = false;
3000 }
3001 #else
3002 lbeentry.st_ssl = false;
3003 #endif
3004
3005 #ifdef ENABLE_GSS
3006 if (MyProcPort && MyProcPort->gss != NULL)
3007 {
3008 const char *princ = be_gssapi_get_princ(MyProcPort);
3009
3010 lbeentry.st_gss = true;
3011 lgssstatus.gss_auth = be_gssapi_get_auth(MyProcPort);
3012 lgssstatus.gss_enc = be_gssapi_get_enc(MyProcPort);
3013 if (princ)
3014 strlcpy(lgssstatus.gss_princ, princ, NAMEDATALEN);
3015 }
3016 else
3017 {
3018 lbeentry.st_gss = false;
3019 }
3020 #else
3021 lbeentry.st_gss = false;
3022 #endif
3023
3024 lbeentry.st_state = STATE_UNDEFINED;
3025 lbeentry.st_progress_command = PROGRESS_COMMAND_INVALID;
3026 lbeentry.st_progress_command_target = InvalidOid;
3027
3028 /*
3029 * we don't zero st_progress_param here to save cycles; nobody should
3030 * examine it until st_progress_command has been set to something other
3031 * than PROGRESS_COMMAND_INVALID
3032 */
3033
3034 /*
3035 * We're ready to enter the critical section that fills the shared-memory
3036 * status entry. We follow the protocol of bumping st_changecount before
3037 * and after; and make sure it's even afterwards. We use a volatile
3038 * pointer here to ensure the compiler doesn't try to get cute.
3039 */
3040 PGSTAT_BEGIN_WRITE_ACTIVITY(vbeentry);
3041
3042 /* make sure we'll memcpy the same st_changecount back */
3043 lbeentry.st_changecount = vbeentry->st_changecount;
3044
3045 memcpy(unvolatize(PgBackendStatus *, vbeentry),
3046 &lbeentry,
3047 sizeof(PgBackendStatus));
3048
3049 /*
3050 * We can write the out-of-line strings and structs using the pointers
3051 * that are in lbeentry; this saves some de-volatilizing messiness.
3052 */
3053 lbeentry.st_appname[0] = '\0';
3054 if (MyProcPort && MyProcPort->remote_hostname)
3055 strlcpy(lbeentry.st_clienthostname, MyProcPort->remote_hostname,
3056 NAMEDATALEN);
3057 else
3058 lbeentry.st_clienthostname[0] = '\0';
3059 lbeentry.st_activity_raw[0] = '\0';
3060 /* Also make sure the last byte in each string area is always 0 */
3061 lbeentry.st_appname[NAMEDATALEN - 1] = '\0';
3062 lbeentry.st_clienthostname[NAMEDATALEN - 1] = '\0';
3063 lbeentry.st_activity_raw[pgstat_track_activity_query_size - 1] = '\0';
3064
3065 #ifdef USE_SSL
3066 memcpy(lbeentry.st_sslstatus, &lsslstatus, sizeof(PgBackendSSLStatus));
3067 #endif
3068 #ifdef ENABLE_GSS
3069 memcpy(lbeentry.st_gssstatus, &lgssstatus, sizeof(PgBackendGSSStatus));
3070 #endif
3071
3072 PGSTAT_END_WRITE_ACTIVITY(vbeentry);
3073
3074 /* Update app name to current GUC setting */
3075 if (application_name)
3076 pgstat_report_appname(application_name);
3077 }
3078
3079 /*
3080 * Shut down a single backend's statistics reporting at process exit.
3081 *
3082 * Flush any remaining statistics counts out to the collector.
3083 * Without this, operations triggered during backend exit (such as
3084 * temp table deletions) won't be counted.
3085 *
3086 * Lastly, clear out our entry in the PgBackendStatus array.
3087 */
3088 static void
pgstat_beshutdown_hook(int code,Datum arg)3089 pgstat_beshutdown_hook(int code, Datum arg)
3090 {
3091 volatile PgBackendStatus *beentry = MyBEEntry;
3092
3093 /*
3094 * If we got as far as discovering our own database ID, we can report what
3095 * we did to the collector. Otherwise, we'd be sending an invalid
3096 * database ID, so forget it. (This means that accesses to pg_database
3097 * during failed backend starts might never get counted.)
3098 */
3099 if (OidIsValid(MyDatabaseId))
3100 pgstat_report_stat(true);
3101
3102 /*
3103 * Clear my status entry, following the protocol of bumping st_changecount
3104 * before and after. We use a volatile pointer here to ensure the
3105 * compiler doesn't try to get cute.
3106 */
3107 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3108
3109 beentry->st_procpid = 0; /* mark invalid */
3110
3111 PGSTAT_END_WRITE_ACTIVITY(beentry);
3112 }
3113
3114
3115 /* ----------
3116 * pgstat_report_activity() -
3117 *
3118 * Called from tcop/postgres.c to report what the backend is actually doing
3119 * (but note cmd_str can be NULL for certain cases).
3120 *
3121 * All updates of the status entry follow the protocol of bumping
3122 * st_changecount before and after. We use a volatile pointer here to
3123 * ensure the compiler doesn't try to get cute.
3124 * ----------
3125 */
3126 void
pgstat_report_activity(BackendState state,const char * cmd_str)3127 pgstat_report_activity(BackendState state, const char *cmd_str)
3128 {
3129 volatile PgBackendStatus *beentry = MyBEEntry;
3130 TimestampTz start_timestamp;
3131 TimestampTz current_timestamp;
3132 int len = 0;
3133
3134 TRACE_POSTGRESQL_STATEMENT_STATUS(cmd_str);
3135
3136 if (!beentry)
3137 return;
3138
3139 if (!pgstat_track_activities)
3140 {
3141 if (beentry->st_state != STATE_DISABLED)
3142 {
3143 volatile PGPROC *proc = MyProc;
3144
3145 /*
3146 * track_activities is disabled, but we last reported a
3147 * non-disabled state. As our final update, change the state and
3148 * clear fields we will not be updating anymore.
3149 */
3150 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3151 beentry->st_state = STATE_DISABLED;
3152 beentry->st_state_start_timestamp = 0;
3153 beentry->st_activity_raw[0] = '\0';
3154 beentry->st_activity_start_timestamp = 0;
3155 /* st_xact_start_timestamp and wait_event_info are also disabled */
3156 beentry->st_xact_start_timestamp = 0;
3157 proc->wait_event_info = 0;
3158 PGSTAT_END_WRITE_ACTIVITY(beentry);
3159 }
3160 return;
3161 }
3162
3163 /*
3164 * To minimize the time spent modifying the entry, and avoid risk of
3165 * errors inside the critical section, fetch all the needed data first.
3166 */
3167 start_timestamp = GetCurrentStatementStartTimestamp();
3168 if (cmd_str != NULL)
3169 {
3170 /*
3171 * Compute length of to-be-stored string unaware of multi-byte
3172 * characters. For speed reasons that'll get corrected on read, rather
3173 * than computed every write.
3174 */
3175 len = Min(strlen(cmd_str), pgstat_track_activity_query_size - 1);
3176 }
3177 current_timestamp = GetCurrentTimestamp();
3178
3179 /*
3180 * Now update the status entry
3181 */
3182 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3183
3184 beentry->st_state = state;
3185 beentry->st_state_start_timestamp = current_timestamp;
3186
3187 if (cmd_str != NULL)
3188 {
3189 memcpy((char *) beentry->st_activity_raw, cmd_str, len);
3190 beentry->st_activity_raw[len] = '\0';
3191 beentry->st_activity_start_timestamp = start_timestamp;
3192 }
3193
3194 PGSTAT_END_WRITE_ACTIVITY(beentry);
3195 }
3196
3197 /*-----------
3198 * pgstat_progress_start_command() -
3199 *
3200 * Set st_progress_command (and st_progress_command_target) in own backend
3201 * entry. Also, zero-initialize st_progress_param array.
3202 *-----------
3203 */
3204 void
pgstat_progress_start_command(ProgressCommandType cmdtype,Oid relid)3205 pgstat_progress_start_command(ProgressCommandType cmdtype, Oid relid)
3206 {
3207 volatile PgBackendStatus *beentry = MyBEEntry;
3208
3209 if (!beentry || !pgstat_track_activities)
3210 return;
3211
3212 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3213 beentry->st_progress_command = cmdtype;
3214 beentry->st_progress_command_target = relid;
3215 MemSet(&beentry->st_progress_param, 0, sizeof(beentry->st_progress_param));
3216 PGSTAT_END_WRITE_ACTIVITY(beentry);
3217 }
3218
3219 /*-----------
3220 * pgstat_progress_update_param() -
3221 *
3222 * Update index'th member in st_progress_param[] of own backend entry.
3223 *-----------
3224 */
3225 void
pgstat_progress_update_param(int index,int64 val)3226 pgstat_progress_update_param(int index, int64 val)
3227 {
3228 volatile PgBackendStatus *beentry = MyBEEntry;
3229
3230 Assert(index >= 0 && index < PGSTAT_NUM_PROGRESS_PARAM);
3231
3232 if (!beentry || !pgstat_track_activities)
3233 return;
3234
3235 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3236 beentry->st_progress_param[index] = val;
3237 PGSTAT_END_WRITE_ACTIVITY(beentry);
3238 }
3239
3240 /*-----------
3241 * pgstat_progress_update_multi_param() -
3242 *
3243 * Update multiple members in st_progress_param[] of own backend entry.
3244 * This is atomic; readers won't see intermediate states.
3245 *-----------
3246 */
3247 void
pgstat_progress_update_multi_param(int nparam,const int * index,const int64 * val)3248 pgstat_progress_update_multi_param(int nparam, const int *index,
3249 const int64 *val)
3250 {
3251 volatile PgBackendStatus *beentry = MyBEEntry;
3252 int i;
3253
3254 if (!beentry || !pgstat_track_activities || nparam == 0)
3255 return;
3256
3257 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3258
3259 for (i = 0; i < nparam; ++i)
3260 {
3261 Assert(index[i] >= 0 && index[i] < PGSTAT_NUM_PROGRESS_PARAM);
3262
3263 beentry->st_progress_param[index[i]] = val[i];
3264 }
3265
3266 PGSTAT_END_WRITE_ACTIVITY(beentry);
3267 }
3268
3269 /*-----------
3270 * pgstat_progress_end_command() -
3271 *
3272 * Reset st_progress_command (and st_progress_command_target) in own backend
3273 * entry. This signals the end of the command.
3274 *-----------
3275 */
3276 void
pgstat_progress_end_command(void)3277 pgstat_progress_end_command(void)
3278 {
3279 volatile PgBackendStatus *beentry = MyBEEntry;
3280
3281 if (!beentry || !pgstat_track_activities)
3282 return;
3283
3284 if (beentry->st_progress_command == PROGRESS_COMMAND_INVALID)
3285 return;
3286
3287 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3288 beentry->st_progress_command = PROGRESS_COMMAND_INVALID;
3289 beentry->st_progress_command_target = InvalidOid;
3290 PGSTAT_END_WRITE_ACTIVITY(beentry);
3291 }
3292
3293 /* ----------
3294 * pgstat_report_appname() -
3295 *
3296 * Called to update our application name.
3297 * ----------
3298 */
3299 void
pgstat_report_appname(const char * appname)3300 pgstat_report_appname(const char *appname)
3301 {
3302 volatile PgBackendStatus *beentry = MyBEEntry;
3303 int len;
3304
3305 if (!beentry)
3306 return;
3307
3308 /* This should be unnecessary if GUC did its job, but be safe */
3309 len = pg_mbcliplen(appname, strlen(appname), NAMEDATALEN - 1);
3310
3311 /*
3312 * Update my status entry, following the protocol of bumping
3313 * st_changecount before and after. We use a volatile pointer here to
3314 * ensure the compiler doesn't try to get cute.
3315 */
3316 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3317
3318 memcpy((char *) beentry->st_appname, appname, len);
3319 beentry->st_appname[len] = '\0';
3320
3321 PGSTAT_END_WRITE_ACTIVITY(beentry);
3322 }
3323
3324 /*
3325 * Report current transaction start timestamp as the specified value.
3326 * Zero means there is no active transaction.
3327 */
3328 void
pgstat_report_xact_timestamp(TimestampTz tstamp)3329 pgstat_report_xact_timestamp(TimestampTz tstamp)
3330 {
3331 volatile PgBackendStatus *beentry = MyBEEntry;
3332
3333 if (!pgstat_track_activities || !beentry)
3334 return;
3335
3336 /*
3337 * Update my status entry, following the protocol of bumping
3338 * st_changecount before and after. We use a volatile pointer here to
3339 * ensure the compiler doesn't try to get cute.
3340 */
3341 PGSTAT_BEGIN_WRITE_ACTIVITY(beentry);
3342
3343 beentry->st_xact_start_timestamp = tstamp;
3344
3345 PGSTAT_END_WRITE_ACTIVITY(beentry);
3346 }
3347
3348 /* ----------
3349 * pgstat_read_current_status() -
3350 *
3351 * Copy the current contents of the PgBackendStatus array to local memory,
3352 * if not already done in this transaction.
3353 * ----------
3354 */
3355 static void
pgstat_read_current_status(void)3356 pgstat_read_current_status(void)
3357 {
3358 volatile PgBackendStatus *beentry;
3359 LocalPgBackendStatus *localtable;
3360 LocalPgBackendStatus *localentry;
3361 char *localappname,
3362 *localclienthostname,
3363 *localactivity;
3364 #ifdef USE_SSL
3365 PgBackendSSLStatus *localsslstatus;
3366 #endif
3367 #ifdef ENABLE_GSS
3368 PgBackendGSSStatus *localgssstatus;
3369 #endif
3370 int i;
3371
3372 Assert(!pgStatRunningInCollector);
3373 if (localBackendStatusTable)
3374 return; /* already done */
3375
3376 pgstat_setup_memcxt();
3377
3378 /*
3379 * Allocate storage for local copy of state data. We can presume that
3380 * none of these requests overflow size_t, because we already calculated
3381 * the same values using mul_size during shmem setup. However, with
3382 * probably-silly values of pgstat_track_activity_query_size and
3383 * max_connections, the localactivity buffer could exceed 1GB, so use
3384 * "huge" allocation for that one.
3385 */
3386 localtable = (LocalPgBackendStatus *)
3387 MemoryContextAlloc(pgStatLocalContext,
3388 sizeof(LocalPgBackendStatus) * NumBackendStatSlots);
3389 localappname = (char *)
3390 MemoryContextAlloc(pgStatLocalContext,
3391 NAMEDATALEN * NumBackendStatSlots);
3392 localclienthostname = (char *)
3393 MemoryContextAlloc(pgStatLocalContext,
3394 NAMEDATALEN * NumBackendStatSlots);
3395 localactivity = (char *)
3396 MemoryContextAllocHuge(pgStatLocalContext,
3397 pgstat_track_activity_query_size * NumBackendStatSlots);
3398 #ifdef USE_SSL
3399 localsslstatus = (PgBackendSSLStatus *)
3400 MemoryContextAlloc(pgStatLocalContext,
3401 sizeof(PgBackendSSLStatus) * NumBackendStatSlots);
3402 #endif
3403 #ifdef ENABLE_GSS
3404 localgssstatus = (PgBackendGSSStatus *)
3405 MemoryContextAlloc(pgStatLocalContext,
3406 sizeof(PgBackendGSSStatus) * NumBackendStatSlots);
3407 #endif
3408
3409 localNumBackends = 0;
3410
3411 beentry = BackendStatusArray;
3412 localentry = localtable;
3413 for (i = 1; i <= NumBackendStatSlots; i++)
3414 {
3415 /*
3416 * Follow the protocol of retrying if st_changecount changes while we
3417 * copy the entry, or if it's odd. (The check for odd is needed to
3418 * cover the case where we are able to completely copy the entry while
3419 * the source backend is between increment steps.) We use a volatile
3420 * pointer here to ensure the compiler doesn't try to get cute.
3421 */
3422 for (;;)
3423 {
3424 int before_changecount;
3425 int after_changecount;
3426
3427 pgstat_begin_read_activity(beentry, before_changecount);
3428
3429 localentry->backendStatus.st_procpid = beentry->st_procpid;
3430 /* Skip all the data-copying work if entry is not in use */
3431 if (localentry->backendStatus.st_procpid > 0)
3432 {
3433 memcpy(&localentry->backendStatus, unvolatize(PgBackendStatus *, beentry), sizeof(PgBackendStatus));
3434
3435 /*
3436 * For each PgBackendStatus field that is a pointer, copy the
3437 * pointed-to data, then adjust the local copy of the pointer
3438 * field to point at the local copy of the data.
3439 *
3440 * strcpy is safe even if the string is modified concurrently,
3441 * because there's always a \0 at the end of the buffer.
3442 */
3443 strcpy(localappname, (char *) beentry->st_appname);
3444 localentry->backendStatus.st_appname = localappname;
3445 strcpy(localclienthostname, (char *) beentry->st_clienthostname);
3446 localentry->backendStatus.st_clienthostname = localclienthostname;
3447 strcpy(localactivity, (char *) beentry->st_activity_raw);
3448 localentry->backendStatus.st_activity_raw = localactivity;
3449 #ifdef USE_SSL
3450 if (beentry->st_ssl)
3451 {
3452 memcpy(localsslstatus, beentry->st_sslstatus, sizeof(PgBackendSSLStatus));
3453 localentry->backendStatus.st_sslstatus = localsslstatus;
3454 }
3455 #endif
3456 #ifdef ENABLE_GSS
3457 if (beentry->st_gss)
3458 {
3459 memcpy(localgssstatus, beentry->st_gssstatus, sizeof(PgBackendGSSStatus));
3460 localentry->backendStatus.st_gssstatus = localgssstatus;
3461 }
3462 #endif
3463 }
3464
3465 pgstat_end_read_activity(beentry, after_changecount);
3466
3467 if (pgstat_read_activity_complete(before_changecount,
3468 after_changecount))
3469 break;
3470
3471 /* Make sure we can break out of loop if stuck... */
3472 CHECK_FOR_INTERRUPTS();
3473 }
3474
3475 beentry++;
3476 /* Only valid entries get included into the local array */
3477 if (localentry->backendStatus.st_procpid > 0)
3478 {
3479 BackendIdGetTransactionIds(i,
3480 &localentry->backend_xid,
3481 &localentry->backend_xmin);
3482
3483 localentry++;
3484 localappname += NAMEDATALEN;
3485 localclienthostname += NAMEDATALEN;
3486 localactivity += pgstat_track_activity_query_size;
3487 #ifdef USE_SSL
3488 localsslstatus++;
3489 #endif
3490 #ifdef ENABLE_GSS
3491 localgssstatus++;
3492 #endif
3493 localNumBackends++;
3494 }
3495 }
3496
3497 /* Set the pointer only after completion of a valid table */
3498 localBackendStatusTable = localtable;
3499 }
3500
3501 /* ----------
3502 * pgstat_get_wait_event_type() -
3503 *
3504 * Return a string representing the current wait event type, backend is
3505 * waiting on.
3506 */
3507 const char *
pgstat_get_wait_event_type(uint32 wait_event_info)3508 pgstat_get_wait_event_type(uint32 wait_event_info)
3509 {
3510 uint32 classId;
3511 const char *event_type;
3512
3513 /* report process as not waiting. */
3514 if (wait_event_info == 0)
3515 return NULL;
3516
3517 classId = wait_event_info & 0xFF000000;
3518
3519 switch (classId)
3520 {
3521 case PG_WAIT_LWLOCK:
3522 event_type = "LWLock";
3523 break;
3524 case PG_WAIT_LOCK:
3525 event_type = "Lock";
3526 break;
3527 case PG_WAIT_BUFFER_PIN:
3528 event_type = "BufferPin";
3529 break;
3530 case PG_WAIT_ACTIVITY:
3531 event_type = "Activity";
3532 break;
3533 case PG_WAIT_CLIENT:
3534 event_type = "Client";
3535 break;
3536 case PG_WAIT_EXTENSION:
3537 event_type = "Extension";
3538 break;
3539 case PG_WAIT_IPC:
3540 event_type = "IPC";
3541 break;
3542 case PG_WAIT_TIMEOUT:
3543 event_type = "Timeout";
3544 break;
3545 case PG_WAIT_IO:
3546 event_type = "IO";
3547 break;
3548 default:
3549 event_type = "???";
3550 break;
3551 }
3552
3553 return event_type;
3554 }
3555
3556 /* ----------
3557 * pgstat_get_wait_event() -
3558 *
3559 * Return a string representing the current wait event, backend is
3560 * waiting on.
3561 */
3562 const char *
pgstat_get_wait_event(uint32 wait_event_info)3563 pgstat_get_wait_event(uint32 wait_event_info)
3564 {
3565 uint32 classId;
3566 uint16 eventId;
3567 const char *event_name;
3568
3569 /* report process as not waiting. */
3570 if (wait_event_info == 0)
3571 return NULL;
3572
3573 classId = wait_event_info & 0xFF000000;
3574 eventId = wait_event_info & 0x0000FFFF;
3575
3576 switch (classId)
3577 {
3578 case PG_WAIT_LWLOCK:
3579 event_name = GetLWLockIdentifier(classId, eventId);
3580 break;
3581 case PG_WAIT_LOCK:
3582 event_name = GetLockNameFromTagType(eventId);
3583 break;
3584 case PG_WAIT_BUFFER_PIN:
3585 event_name = "BufferPin";
3586 break;
3587 case PG_WAIT_ACTIVITY:
3588 {
3589 WaitEventActivity w = (WaitEventActivity) wait_event_info;
3590
3591 event_name = pgstat_get_wait_activity(w);
3592 break;
3593 }
3594 case PG_WAIT_CLIENT:
3595 {
3596 WaitEventClient w = (WaitEventClient) wait_event_info;
3597
3598 event_name = pgstat_get_wait_client(w);
3599 break;
3600 }
3601 case PG_WAIT_EXTENSION:
3602 event_name = "Extension";
3603 break;
3604 case PG_WAIT_IPC:
3605 {
3606 WaitEventIPC w = (WaitEventIPC) wait_event_info;
3607
3608 event_name = pgstat_get_wait_ipc(w);
3609 break;
3610 }
3611 case PG_WAIT_TIMEOUT:
3612 {
3613 WaitEventTimeout w = (WaitEventTimeout) wait_event_info;
3614
3615 event_name = pgstat_get_wait_timeout(w);
3616 break;
3617 }
3618 case PG_WAIT_IO:
3619 {
3620 WaitEventIO w = (WaitEventIO) wait_event_info;
3621
3622 event_name = pgstat_get_wait_io(w);
3623 break;
3624 }
3625 default:
3626 event_name = "unknown wait event";
3627 break;
3628 }
3629
3630 return event_name;
3631 }
3632
3633 /* ----------
3634 * pgstat_get_wait_activity() -
3635 *
3636 * Convert WaitEventActivity to string.
3637 * ----------
3638 */
3639 static const char *
pgstat_get_wait_activity(WaitEventActivity w)3640 pgstat_get_wait_activity(WaitEventActivity w)
3641 {
3642 const char *event_name = "unknown wait event";
3643
3644 switch (w)
3645 {
3646 case WAIT_EVENT_ARCHIVER_MAIN:
3647 event_name = "ArchiverMain";
3648 break;
3649 case WAIT_EVENT_AUTOVACUUM_MAIN:
3650 event_name = "AutoVacuumMain";
3651 break;
3652 case WAIT_EVENT_BGWRITER_HIBERNATE:
3653 event_name = "BgWriterHibernate";
3654 break;
3655 case WAIT_EVENT_BGWRITER_MAIN:
3656 event_name = "BgWriterMain";
3657 break;
3658 case WAIT_EVENT_CHECKPOINTER_MAIN:
3659 event_name = "CheckpointerMain";
3660 break;
3661 case WAIT_EVENT_LOGICAL_APPLY_MAIN:
3662 event_name = "LogicalApplyMain";
3663 break;
3664 case WAIT_EVENT_LOGICAL_LAUNCHER_MAIN:
3665 event_name = "LogicalLauncherMain";
3666 break;
3667 case WAIT_EVENT_PGSTAT_MAIN:
3668 event_name = "PgStatMain";
3669 break;
3670 case WAIT_EVENT_RECOVERY_WAL_ALL:
3671 event_name = "RecoveryWalAll";
3672 break;
3673 case WAIT_EVENT_RECOVERY_WAL_STREAM:
3674 event_name = "RecoveryWalStream";
3675 break;
3676 case WAIT_EVENT_SYSLOGGER_MAIN:
3677 event_name = "SysLoggerMain";
3678 break;
3679 case WAIT_EVENT_WAL_RECEIVER_MAIN:
3680 event_name = "WalReceiverMain";
3681 break;
3682 case WAIT_EVENT_WAL_SENDER_MAIN:
3683 event_name = "WalSenderMain";
3684 break;
3685 case WAIT_EVENT_WAL_WRITER_MAIN:
3686 event_name = "WalWriterMain";
3687 break;
3688 /* no default case, so that compiler will warn */
3689 }
3690
3691 return event_name;
3692 }
3693
3694 /* ----------
3695 * pgstat_get_wait_client() -
3696 *
3697 * Convert WaitEventClient to string.
3698 * ----------
3699 */
3700 static const char *
pgstat_get_wait_client(WaitEventClient w)3701 pgstat_get_wait_client(WaitEventClient w)
3702 {
3703 const char *event_name = "unknown wait event";
3704
3705 switch (w)
3706 {
3707 case WAIT_EVENT_CLIENT_READ:
3708 event_name = "ClientRead";
3709 break;
3710 case WAIT_EVENT_CLIENT_WRITE:
3711 event_name = "ClientWrite";
3712 break;
3713 case WAIT_EVENT_GSS_OPEN_SERVER:
3714 event_name = "GSSOpenServer";
3715 break;
3716 case WAIT_EVENT_LIBPQWALRECEIVER_CONNECT:
3717 event_name = "LibPQWalReceiverConnect";
3718 break;
3719 case WAIT_EVENT_LIBPQWALRECEIVER_RECEIVE:
3720 event_name = "LibPQWalReceiverReceive";
3721 break;
3722 case WAIT_EVENT_SSL_OPEN_SERVER:
3723 event_name = "SSLOpenServer";
3724 break;
3725 case WAIT_EVENT_WAL_RECEIVER_WAIT_START:
3726 event_name = "WalReceiverWaitStart";
3727 break;
3728 case WAIT_EVENT_WAL_SENDER_WAIT_WAL:
3729 event_name = "WalSenderWaitForWAL";
3730 break;
3731 case WAIT_EVENT_WAL_SENDER_WRITE_DATA:
3732 event_name = "WalSenderWriteData";
3733 break;
3734 /* no default case, so that compiler will warn */
3735 }
3736
3737 return event_name;
3738 }
3739
3740 /* ----------
3741 * pgstat_get_wait_ipc() -
3742 *
3743 * Convert WaitEventIPC to string.
3744 * ----------
3745 */
3746 static const char *
pgstat_get_wait_ipc(WaitEventIPC w)3747 pgstat_get_wait_ipc(WaitEventIPC w)
3748 {
3749 const char *event_name = "unknown wait event";
3750
3751 switch (w)
3752 {
3753 case WAIT_EVENT_BGWORKER_SHUTDOWN:
3754 event_name = "BgWorkerShutdown";
3755 break;
3756 case WAIT_EVENT_BGWORKER_STARTUP:
3757 event_name = "BgWorkerStartup";
3758 break;
3759 case WAIT_EVENT_BTREE_PAGE:
3760 event_name = "BtreePage";
3761 break;
3762 case WAIT_EVENT_CHECKPOINT_DONE:
3763 event_name = "CheckpointDone";
3764 break;
3765 case WAIT_EVENT_CHECKPOINT_START:
3766 event_name = "CheckpointStart";
3767 break;
3768 case WAIT_EVENT_CLOG_GROUP_UPDATE:
3769 event_name = "ClogGroupUpdate";
3770 break;
3771 case WAIT_EVENT_EXECUTE_GATHER:
3772 event_name = "ExecuteGather";
3773 break;
3774 case WAIT_EVENT_HASH_BATCH_ALLOCATING:
3775 event_name = "Hash/Batch/Allocating";
3776 break;
3777 case WAIT_EVENT_HASH_BATCH_ELECTING:
3778 event_name = "Hash/Batch/Electing";
3779 break;
3780 case WAIT_EVENT_HASH_BATCH_LOADING:
3781 event_name = "Hash/Batch/Loading";
3782 break;
3783 case WAIT_EVENT_HASH_BUILD_ALLOCATING:
3784 event_name = "Hash/Build/Allocating";
3785 break;
3786 case WAIT_EVENT_HASH_BUILD_ELECTING:
3787 event_name = "Hash/Build/Electing";
3788 break;
3789 case WAIT_EVENT_HASH_BUILD_HASHING_INNER:
3790 event_name = "Hash/Build/HashingInner";
3791 break;
3792 case WAIT_EVENT_HASH_BUILD_HASHING_OUTER:
3793 event_name = "Hash/Build/HashingOuter";
3794 break;
3795 case WAIT_EVENT_HASH_GROW_BATCHES_ALLOCATING:
3796 event_name = "Hash/GrowBatches/Allocating";
3797 break;
3798 case WAIT_EVENT_HASH_GROW_BATCHES_DECIDING:
3799 event_name = "Hash/GrowBatches/Deciding";
3800 break;
3801 case WAIT_EVENT_HASH_GROW_BATCHES_ELECTING:
3802 event_name = "Hash/GrowBatches/Electing";
3803 break;
3804 case WAIT_EVENT_HASH_GROW_BATCHES_FINISHING:
3805 event_name = "Hash/GrowBatches/Finishing";
3806 break;
3807 case WAIT_EVENT_HASH_GROW_BATCHES_REPARTITIONING:
3808 event_name = "Hash/GrowBatches/Repartitioning";
3809 break;
3810 case WAIT_EVENT_HASH_GROW_BUCKETS_ALLOCATING:
3811 event_name = "Hash/GrowBuckets/Allocating";
3812 break;
3813 case WAIT_EVENT_HASH_GROW_BUCKETS_ELECTING:
3814 event_name = "Hash/GrowBuckets/Electing";
3815 break;
3816 case WAIT_EVENT_HASH_GROW_BUCKETS_REINSERTING:
3817 event_name = "Hash/GrowBuckets/Reinserting";
3818 break;
3819 case WAIT_EVENT_LOGICAL_SYNC_DATA:
3820 event_name = "LogicalSyncData";
3821 break;
3822 case WAIT_EVENT_LOGICAL_SYNC_STATE_CHANGE:
3823 event_name = "LogicalSyncStateChange";
3824 break;
3825 case WAIT_EVENT_MQ_INTERNAL:
3826 event_name = "MessageQueueInternal";
3827 break;
3828 case WAIT_EVENT_MQ_PUT_MESSAGE:
3829 event_name = "MessageQueuePutMessage";
3830 break;
3831 case WAIT_EVENT_MQ_RECEIVE:
3832 event_name = "MessageQueueReceive";
3833 break;
3834 case WAIT_EVENT_MQ_SEND:
3835 event_name = "MessageQueueSend";
3836 break;
3837 case WAIT_EVENT_PARALLEL_BITMAP_SCAN:
3838 event_name = "ParallelBitmapScan";
3839 break;
3840 case WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN:
3841 event_name = "ParallelCreateIndexScan";
3842 break;
3843 case WAIT_EVENT_PARALLEL_FINISH:
3844 event_name = "ParallelFinish";
3845 break;
3846 case WAIT_EVENT_PROCARRAY_GROUP_UPDATE:
3847 event_name = "ProcArrayGroupUpdate";
3848 break;
3849 case WAIT_EVENT_PROMOTE:
3850 event_name = "Promote";
3851 break;
3852 case WAIT_EVENT_REPLICATION_ORIGIN_DROP:
3853 event_name = "ReplicationOriginDrop";
3854 break;
3855 case WAIT_EVENT_REPLICATION_SLOT_DROP:
3856 event_name = "ReplicationSlotDrop";
3857 break;
3858 case WAIT_EVENT_SAFE_SNAPSHOT:
3859 event_name = "SafeSnapshot";
3860 break;
3861 case WAIT_EVENT_SYNC_REP:
3862 event_name = "SyncRep";
3863 break;
3864 /* no default case, so that compiler will warn */
3865 }
3866
3867 return event_name;
3868 }
3869
3870 /* ----------
3871 * pgstat_get_wait_timeout() -
3872 *
3873 * Convert WaitEventTimeout to string.
3874 * ----------
3875 */
3876 static const char *
pgstat_get_wait_timeout(WaitEventTimeout w)3877 pgstat_get_wait_timeout(WaitEventTimeout w)
3878 {
3879 const char *event_name = "unknown wait event";
3880
3881 switch (w)
3882 {
3883 case WAIT_EVENT_BASE_BACKUP_THROTTLE:
3884 event_name = "BaseBackupThrottle";
3885 break;
3886 case WAIT_EVENT_PG_SLEEP:
3887 event_name = "PgSleep";
3888 break;
3889 case WAIT_EVENT_RECOVERY_APPLY_DELAY:
3890 event_name = "RecoveryApplyDelay";
3891 break;
3892 /* no default case, so that compiler will warn */
3893 }
3894
3895 return event_name;
3896 }
3897
3898 /* ----------
3899 * pgstat_get_wait_io() -
3900 *
3901 * Convert WaitEventIO to string.
3902 * ----------
3903 */
3904 static const char *
pgstat_get_wait_io(WaitEventIO w)3905 pgstat_get_wait_io(WaitEventIO w)
3906 {
3907 const char *event_name = "unknown wait event";
3908
3909 switch (w)
3910 {
3911 case WAIT_EVENT_BUFFILE_READ:
3912 event_name = "BufFileRead";
3913 break;
3914 case WAIT_EVENT_BUFFILE_WRITE:
3915 event_name = "BufFileWrite";
3916 break;
3917 case WAIT_EVENT_CONTROL_FILE_READ:
3918 event_name = "ControlFileRead";
3919 break;
3920 case WAIT_EVENT_CONTROL_FILE_SYNC:
3921 event_name = "ControlFileSync";
3922 break;
3923 case WAIT_EVENT_CONTROL_FILE_SYNC_UPDATE:
3924 event_name = "ControlFileSyncUpdate";
3925 break;
3926 case WAIT_EVENT_CONTROL_FILE_WRITE:
3927 event_name = "ControlFileWrite";
3928 break;
3929 case WAIT_EVENT_CONTROL_FILE_WRITE_UPDATE:
3930 event_name = "ControlFileWriteUpdate";
3931 break;
3932 case WAIT_EVENT_COPY_FILE_READ:
3933 event_name = "CopyFileRead";
3934 break;
3935 case WAIT_EVENT_COPY_FILE_WRITE:
3936 event_name = "CopyFileWrite";
3937 break;
3938 case WAIT_EVENT_DATA_FILE_EXTEND:
3939 event_name = "DataFileExtend";
3940 break;
3941 case WAIT_EVENT_DATA_FILE_FLUSH:
3942 event_name = "DataFileFlush";
3943 break;
3944 case WAIT_EVENT_DATA_FILE_IMMEDIATE_SYNC:
3945 event_name = "DataFileImmediateSync";
3946 break;
3947 case WAIT_EVENT_DATA_FILE_PREFETCH:
3948 event_name = "DataFilePrefetch";
3949 break;
3950 case WAIT_EVENT_DATA_FILE_READ:
3951 event_name = "DataFileRead";
3952 break;
3953 case WAIT_EVENT_DATA_FILE_SYNC:
3954 event_name = "DataFileSync";
3955 break;
3956 case WAIT_EVENT_DATA_FILE_TRUNCATE:
3957 event_name = "DataFileTruncate";
3958 break;
3959 case WAIT_EVENT_DATA_FILE_WRITE:
3960 event_name = "DataFileWrite";
3961 break;
3962 case WAIT_EVENT_DSM_FILL_ZERO_WRITE:
3963 event_name = "DSMFillZeroWrite";
3964 break;
3965 case WAIT_EVENT_LOCK_FILE_ADDTODATADIR_READ:
3966 event_name = "LockFileAddToDataDirRead";
3967 break;
3968 case WAIT_EVENT_LOCK_FILE_ADDTODATADIR_SYNC:
3969 event_name = "LockFileAddToDataDirSync";
3970 break;
3971 case WAIT_EVENT_LOCK_FILE_ADDTODATADIR_WRITE:
3972 event_name = "LockFileAddToDataDirWrite";
3973 break;
3974 case WAIT_EVENT_LOCK_FILE_CREATE_READ:
3975 event_name = "LockFileCreateRead";
3976 break;
3977 case WAIT_EVENT_LOCK_FILE_CREATE_SYNC:
3978 event_name = "LockFileCreateSync";
3979 break;
3980 case WAIT_EVENT_LOCK_FILE_CREATE_WRITE:
3981 event_name = "LockFileCreateWrite";
3982 break;
3983 case WAIT_EVENT_LOCK_FILE_RECHECKDATADIR_READ:
3984 event_name = "LockFileReCheckDataDirRead";
3985 break;
3986 case WAIT_EVENT_LOGICAL_REWRITE_CHECKPOINT_SYNC:
3987 event_name = "LogicalRewriteCheckpointSync";
3988 break;
3989 case WAIT_EVENT_LOGICAL_REWRITE_MAPPING_SYNC:
3990 event_name = "LogicalRewriteMappingSync";
3991 break;
3992 case WAIT_EVENT_LOGICAL_REWRITE_MAPPING_WRITE:
3993 event_name = "LogicalRewriteMappingWrite";
3994 break;
3995 case WAIT_EVENT_LOGICAL_REWRITE_SYNC:
3996 event_name = "LogicalRewriteSync";
3997 break;
3998 case WAIT_EVENT_LOGICAL_REWRITE_TRUNCATE:
3999 event_name = "LogicalRewriteTruncate";
4000 break;
4001 case WAIT_EVENT_LOGICAL_REWRITE_WRITE:
4002 event_name = "LogicalRewriteWrite";
4003 break;
4004 case WAIT_EVENT_RELATION_MAP_READ:
4005 event_name = "RelationMapRead";
4006 break;
4007 case WAIT_EVENT_RELATION_MAP_SYNC:
4008 event_name = "RelationMapSync";
4009 break;
4010 case WAIT_EVENT_RELATION_MAP_WRITE:
4011 event_name = "RelationMapWrite";
4012 break;
4013 case WAIT_EVENT_REORDER_BUFFER_READ:
4014 event_name = "ReorderBufferRead";
4015 break;
4016 case WAIT_EVENT_REORDER_BUFFER_WRITE:
4017 event_name = "ReorderBufferWrite";
4018 break;
4019 case WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ:
4020 event_name = "ReorderLogicalMappingRead";
4021 break;
4022 case WAIT_EVENT_REPLICATION_SLOT_READ:
4023 event_name = "ReplicationSlotRead";
4024 break;
4025 case WAIT_EVENT_REPLICATION_SLOT_RESTORE_SYNC:
4026 event_name = "ReplicationSlotRestoreSync";
4027 break;
4028 case WAIT_EVENT_REPLICATION_SLOT_SYNC:
4029 event_name = "ReplicationSlotSync";
4030 break;
4031 case WAIT_EVENT_REPLICATION_SLOT_WRITE:
4032 event_name = "ReplicationSlotWrite";
4033 break;
4034 case WAIT_EVENT_SLRU_FLUSH_SYNC:
4035 event_name = "SLRUFlushSync";
4036 break;
4037 case WAIT_EVENT_SLRU_READ:
4038 event_name = "SLRURead";
4039 break;
4040 case WAIT_EVENT_SLRU_SYNC:
4041 event_name = "SLRUSync";
4042 break;
4043 case WAIT_EVENT_SLRU_WRITE:
4044 event_name = "SLRUWrite";
4045 break;
4046 case WAIT_EVENT_SNAPBUILD_READ:
4047 event_name = "SnapbuildRead";
4048 break;
4049 case WAIT_EVENT_SNAPBUILD_SYNC:
4050 event_name = "SnapbuildSync";
4051 break;
4052 case WAIT_EVENT_SNAPBUILD_WRITE:
4053 event_name = "SnapbuildWrite";
4054 break;
4055 case WAIT_EVENT_TIMELINE_HISTORY_FILE_SYNC:
4056 event_name = "TimelineHistoryFileSync";
4057 break;
4058 case WAIT_EVENT_TIMELINE_HISTORY_FILE_WRITE:
4059 event_name = "TimelineHistoryFileWrite";
4060 break;
4061 case WAIT_EVENT_TIMELINE_HISTORY_READ:
4062 event_name = "TimelineHistoryRead";
4063 break;
4064 case WAIT_EVENT_TIMELINE_HISTORY_SYNC:
4065 event_name = "TimelineHistorySync";
4066 break;
4067 case WAIT_EVENT_TIMELINE_HISTORY_WRITE:
4068 event_name = "TimelineHistoryWrite";
4069 break;
4070 case WAIT_EVENT_TWOPHASE_FILE_READ:
4071 event_name = "TwophaseFileRead";
4072 break;
4073 case WAIT_EVENT_TWOPHASE_FILE_SYNC:
4074 event_name = "TwophaseFileSync";
4075 break;
4076 case WAIT_EVENT_TWOPHASE_FILE_WRITE:
4077 event_name = "TwophaseFileWrite";
4078 break;
4079 case WAIT_EVENT_WALSENDER_TIMELINE_HISTORY_READ:
4080 event_name = "WALSenderTimelineHistoryRead";
4081 break;
4082 case WAIT_EVENT_WAL_BOOTSTRAP_SYNC:
4083 event_name = "WALBootstrapSync";
4084 break;
4085 case WAIT_EVENT_WAL_BOOTSTRAP_WRITE:
4086 event_name = "WALBootstrapWrite";
4087 break;
4088 case WAIT_EVENT_WAL_COPY_READ:
4089 event_name = "WALCopyRead";
4090 break;
4091 case WAIT_EVENT_WAL_COPY_SYNC:
4092 event_name = "WALCopySync";
4093 break;
4094 case WAIT_EVENT_WAL_COPY_WRITE:
4095 event_name = "WALCopyWrite";
4096 break;
4097 case WAIT_EVENT_WAL_INIT_SYNC:
4098 event_name = "WALInitSync";
4099 break;
4100 case WAIT_EVENT_WAL_INIT_WRITE:
4101 event_name = "WALInitWrite";
4102 break;
4103 case WAIT_EVENT_WAL_READ:
4104 event_name = "WALRead";
4105 break;
4106 case WAIT_EVENT_WAL_SYNC:
4107 event_name = "WALSync";
4108 break;
4109 case WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN:
4110 event_name = "WALSyncMethodAssign";
4111 break;
4112 case WAIT_EVENT_WAL_WRITE:
4113 event_name = "WALWrite";
4114 break;
4115
4116 /* no default case, so that compiler will warn */
4117 }
4118
4119 return event_name;
4120 }
4121
4122
4123 /* ----------
4124 * pgstat_get_backend_current_activity() -
4125 *
4126 * Return a string representing the current activity of the backend with
4127 * the specified PID. This looks directly at the BackendStatusArray,
4128 * and so will provide current information regardless of the age of our
4129 * transaction's snapshot of the status array.
4130 *
4131 * It is the caller's responsibility to invoke this only for backends whose
4132 * state is expected to remain stable while the result is in use. The
4133 * only current use is in deadlock reporting, where we can expect that
4134 * the target backend is blocked on a lock. (There are corner cases
4135 * where the target's wait could get aborted while we are looking at it,
4136 * but the very worst consequence is to return a pointer to a string
4137 * that's been changed, so we won't worry too much.)
4138 *
4139 * Note: return strings for special cases match pg_stat_get_backend_activity.
4140 * ----------
4141 */
4142 const char *
pgstat_get_backend_current_activity(int pid,bool checkUser)4143 pgstat_get_backend_current_activity(int pid, bool checkUser)
4144 {
4145 PgBackendStatus *beentry;
4146 int i;
4147
4148 beentry = BackendStatusArray;
4149 for (i = 1; i <= MaxBackends; i++)
4150 {
4151 /*
4152 * Although we expect the target backend's entry to be stable, that
4153 * doesn't imply that anyone else's is. To avoid identifying the
4154 * wrong backend, while we check for a match to the desired PID we
4155 * must follow the protocol of retrying if st_changecount changes
4156 * while we examine the entry, or if it's odd. (This might be
4157 * unnecessary, since fetching or storing an int is almost certainly
4158 * atomic, but let's play it safe.) We use a volatile pointer here to
4159 * ensure the compiler doesn't try to get cute.
4160 */
4161 volatile PgBackendStatus *vbeentry = beentry;
4162 bool found;
4163
4164 for (;;)
4165 {
4166 int before_changecount;
4167 int after_changecount;
4168
4169 pgstat_begin_read_activity(vbeentry, before_changecount);
4170
4171 found = (vbeentry->st_procpid == pid);
4172
4173 pgstat_end_read_activity(vbeentry, after_changecount);
4174
4175 if (pgstat_read_activity_complete(before_changecount,
4176 after_changecount))
4177 break;
4178
4179 /* Make sure we can break out of loop if stuck... */
4180 CHECK_FOR_INTERRUPTS();
4181 }
4182
4183 if (found)
4184 {
4185 /* Now it is safe to use the non-volatile pointer */
4186 if (checkUser && !superuser() && beentry->st_userid != GetUserId())
4187 return "<insufficient privilege>";
4188 else if (*(beentry->st_activity_raw) == '\0')
4189 return "<command string not enabled>";
4190 else
4191 {
4192 /* this'll leak a bit of memory, but that seems acceptable */
4193 return pgstat_clip_activity(beentry->st_activity_raw);
4194 }
4195 }
4196
4197 beentry++;
4198 }
4199
4200 /* If we get here, caller is in error ... */
4201 return "<backend information not available>";
4202 }
4203
4204 /* ----------
4205 * pgstat_get_crashed_backend_activity() -
4206 *
4207 * Return a string representing the current activity of the backend with
4208 * the specified PID. Like the function above, but reads shared memory with
4209 * the expectation that it may be corrupt. On success, copy the string
4210 * into the "buffer" argument and return that pointer. On failure,
4211 * return NULL.
4212 *
4213 * This function is only intended to be used by the postmaster to report the
4214 * query that crashed a backend. In particular, no attempt is made to
4215 * follow the correct concurrency protocol when accessing the
4216 * BackendStatusArray. But that's OK, in the worst case we'll return a
4217 * corrupted message. We also must take care not to trip on ereport(ERROR).
4218 * ----------
4219 */
4220 const char *
pgstat_get_crashed_backend_activity(int pid,char * buffer,int buflen)4221 pgstat_get_crashed_backend_activity(int pid, char *buffer, int buflen)
4222 {
4223 volatile PgBackendStatus *beentry;
4224 int i;
4225
4226 beentry = BackendStatusArray;
4227
4228 /*
4229 * We probably shouldn't get here before shared memory has been set up,
4230 * but be safe.
4231 */
4232 if (beentry == NULL || BackendActivityBuffer == NULL)
4233 return NULL;
4234
4235 for (i = 1; i <= MaxBackends; i++)
4236 {
4237 if (beentry->st_procpid == pid)
4238 {
4239 /* Read pointer just once, so it can't change after validation */
4240 const char *activity = beentry->st_activity_raw;
4241 const char *activity_last;
4242
4243 /*
4244 * We mustn't access activity string before we verify that it
4245 * falls within the BackendActivityBuffer. To make sure that the
4246 * entire string including its ending is contained within the
4247 * buffer, subtract one activity length from the buffer size.
4248 */
4249 activity_last = BackendActivityBuffer + BackendActivityBufferSize
4250 - pgstat_track_activity_query_size;
4251
4252 if (activity < BackendActivityBuffer ||
4253 activity > activity_last)
4254 return NULL;
4255
4256 /* If no string available, no point in a report */
4257 if (activity[0] == '\0')
4258 return NULL;
4259
4260 /*
4261 * Copy only ASCII-safe characters so we don't run into encoding
4262 * problems when reporting the message; and be sure not to run off
4263 * the end of memory. As only ASCII characters are reported, it
4264 * doesn't seem necessary to perform multibyte aware clipping.
4265 */
4266 ascii_safe_strlcpy(buffer, activity,
4267 Min(buflen, pgstat_track_activity_query_size));
4268
4269 return buffer;
4270 }
4271
4272 beentry++;
4273 }
4274
4275 /* PID not found */
4276 return NULL;
4277 }
4278
4279 const char *
pgstat_get_backend_desc(BackendType backendType)4280 pgstat_get_backend_desc(BackendType backendType)
4281 {
4282 const char *backendDesc = "unknown process type";
4283
4284 switch (backendType)
4285 {
4286 case B_AUTOVAC_LAUNCHER:
4287 backendDesc = "autovacuum launcher";
4288 break;
4289 case B_AUTOVAC_WORKER:
4290 backendDesc = "autovacuum worker";
4291 break;
4292 case B_BACKEND:
4293 backendDesc = "client backend";
4294 break;
4295 case B_BG_WORKER:
4296 backendDesc = "background worker";
4297 break;
4298 case B_BG_WRITER:
4299 backendDesc = "background writer";
4300 break;
4301 case B_CHECKPOINTER:
4302 backendDesc = "checkpointer";
4303 break;
4304 case B_STARTUP:
4305 backendDesc = "startup";
4306 break;
4307 case B_WAL_RECEIVER:
4308 backendDesc = "walreceiver";
4309 break;
4310 case B_WAL_SENDER:
4311 backendDesc = "walsender";
4312 break;
4313 case B_WAL_WRITER:
4314 backendDesc = "walwriter";
4315 break;
4316 }
4317
4318 return backendDesc;
4319 }
4320
4321 /* ------------------------------------------------------------
4322 * Local support functions follow
4323 * ------------------------------------------------------------
4324 */
4325
4326
4327 /* ----------
4328 * pgstat_setheader() -
4329 *
4330 * Set common header fields in a statistics message
4331 * ----------
4332 */
4333 static void
pgstat_setheader(PgStat_MsgHdr * hdr,StatMsgType mtype)4334 pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype)
4335 {
4336 hdr->m_type = mtype;
4337 }
4338
4339
4340 /* ----------
4341 * pgstat_send() -
4342 *
4343 * Send out one statistics message to the collector
4344 * ----------
4345 */
4346 static void
pgstat_send(void * msg,int len)4347 pgstat_send(void *msg, int len)
4348 {
4349 int rc;
4350
4351 if (pgStatSock == PGINVALID_SOCKET)
4352 return;
4353
4354 ((PgStat_MsgHdr *) msg)->m_size = len;
4355
4356 /* We'll retry after EINTR, but ignore all other failures */
4357 do
4358 {
4359 rc = send(pgStatSock, msg, len, 0);
4360 } while (rc < 0 && errno == EINTR);
4361
4362 #ifdef USE_ASSERT_CHECKING
4363 /* In debug builds, log send failures ... */
4364 if (rc < 0)
4365 elog(LOG, "could not send to statistics collector: %m");
4366 #endif
4367 }
4368
4369 /* ----------
4370 * pgstat_send_archiver() -
4371 *
4372 * Tell the collector about the WAL file that we successfully
4373 * archived or failed to archive.
4374 * ----------
4375 */
4376 void
pgstat_send_archiver(const char * xlog,bool failed)4377 pgstat_send_archiver(const char *xlog, bool failed)
4378 {
4379 PgStat_MsgArchiver msg;
4380
4381 /*
4382 * Prepare and send the message
4383 */
4384 pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ARCHIVER);
4385 msg.m_failed = failed;
4386 StrNCpy(msg.m_xlog, xlog, sizeof(msg.m_xlog));
4387 msg.m_timestamp = GetCurrentTimestamp();
4388 pgstat_send(&msg, sizeof(msg));
4389 }
4390
4391 /* ----------
4392 * pgstat_send_bgwriter() -
4393 *
4394 * Send bgwriter statistics to the collector
4395 * ----------
4396 */
4397 void
pgstat_send_bgwriter(void)4398 pgstat_send_bgwriter(void)
4399 {
4400 /* We assume this initializes to zeroes */
4401 static const PgStat_MsgBgWriter all_zeroes;
4402
4403 /*
4404 * This function can be called even if nothing at all has happened. In
4405 * this case, avoid sending a completely empty message to the stats
4406 * collector.
4407 */
4408 if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0)
4409 return;
4410
4411 /*
4412 * Prepare and send the message
4413 */
4414 pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER);
4415 pgstat_send(&BgWriterStats, sizeof(BgWriterStats));
4416
4417 /*
4418 * Clear out the statistics buffer, so it can be re-used.
4419 */
4420 MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
4421 }
4422
4423
4424 /* ----------
4425 * PgstatCollectorMain() -
4426 *
4427 * Start up the statistics collector process. This is the body of the
4428 * postmaster child process.
4429 *
4430 * The argc/argv parameters are valid only in EXEC_BACKEND case.
4431 * ----------
4432 */
4433 NON_EXEC_STATIC void
PgstatCollectorMain(int argc,char * argv[])4434 PgstatCollectorMain(int argc, char *argv[])
4435 {
4436 int len;
4437 PgStat_Msg msg;
4438 int wr;
4439
4440 /*
4441 * Ignore all signals usually bound to some action in the postmaster,
4442 * except SIGHUP and SIGQUIT. Note we don't need a SIGUSR1 handler to
4443 * support latch operations, because we only use a local latch.
4444 */
4445 pqsignal(SIGHUP, pgstat_sighup_handler);
4446 pqsignal(SIGINT, SIG_IGN);
4447 pqsignal(SIGTERM, SIG_IGN);
4448 pqsignal(SIGQUIT, pgstat_exit);
4449 pqsignal(SIGALRM, SIG_IGN);
4450 pqsignal(SIGPIPE, SIG_IGN);
4451 pqsignal(SIGUSR1, SIG_IGN);
4452 pqsignal(SIGUSR2, SIG_IGN);
4453 /* Reset some signals that are accepted by postmaster but not here */
4454 pqsignal(SIGCHLD, SIG_DFL);
4455 PG_SETMASK(&UnBlockSig);
4456
4457 /*
4458 * Identify myself via ps
4459 */
4460 init_ps_display("stats collector", "", "", "");
4461
4462 /*
4463 * Read in existing stats files or initialize the stats to zero.
4464 */
4465 pgStatRunningInCollector = true;
4466 pgStatDBHash = pgstat_read_statsfiles(InvalidOid, true, true);
4467
4468 /*
4469 * Loop to process messages until we get SIGQUIT or detect ungraceful
4470 * death of our parent postmaster.
4471 *
4472 * For performance reasons, we don't want to do ResetLatch/WaitLatch after
4473 * every message; instead, do that only after a recv() fails to obtain a
4474 * message. (This effectively means that if backends are sending us stuff
4475 * like mad, we won't notice postmaster death until things slack off a
4476 * bit; which seems fine.) To do that, we have an inner loop that
4477 * iterates as long as recv() succeeds. We do recognize got_SIGHUP inside
4478 * the inner loop, which means that such interrupts will get serviced but
4479 * the latch won't get cleared until next time there is a break in the
4480 * action.
4481 */
4482 for (;;)
4483 {
4484 /* Clear any already-pending wakeups */
4485 ResetLatch(MyLatch);
4486
4487 /*
4488 * Quit if we get SIGQUIT from the postmaster.
4489 */
4490 if (need_exit)
4491 break;
4492
4493 /*
4494 * Inner loop iterates as long as we keep getting messages, or until
4495 * need_exit becomes set.
4496 */
4497 while (!need_exit)
4498 {
4499 /*
4500 * Reload configuration if we got SIGHUP from the postmaster.
4501 */
4502 if (got_SIGHUP)
4503 {
4504 got_SIGHUP = false;
4505 ProcessConfigFile(PGC_SIGHUP);
4506 }
4507
4508 /*
4509 * Write the stats file(s) if a new request has arrived that is
4510 * not satisfied by existing file(s).
4511 */
4512 if (pgstat_write_statsfile_needed())
4513 pgstat_write_statsfiles(false, false);
4514
4515 /*
4516 * Try to receive and process a message. This will not block,
4517 * since the socket is set to non-blocking mode.
4518 *
4519 * XXX On Windows, we have to force pgwin32_recv to cooperate,
4520 * despite the previous use of pg_set_noblock() on the socket.
4521 * This is extremely broken and should be fixed someday.
4522 */
4523 #ifdef WIN32
4524 pgwin32_noblock = 1;
4525 #endif
4526
4527 len = recv(pgStatSock, (char *) &msg,
4528 sizeof(PgStat_Msg), 0);
4529
4530 #ifdef WIN32
4531 pgwin32_noblock = 0;
4532 #endif
4533
4534 if (len < 0)
4535 {
4536 if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
4537 break; /* out of inner loop */
4538 ereport(ERROR,
4539 (errcode_for_socket_access(),
4540 errmsg("could not read statistics message: %m")));
4541 }
4542
4543 /*
4544 * We ignore messages that are smaller than our common header
4545 */
4546 if (len < sizeof(PgStat_MsgHdr))
4547 continue;
4548
4549 /*
4550 * The received length must match the length in the header
4551 */
4552 if (msg.msg_hdr.m_size != len)
4553 continue;
4554
4555 /*
4556 * O.K. - we accept this message. Process it.
4557 */
4558 switch (msg.msg_hdr.m_type)
4559 {
4560 case PGSTAT_MTYPE_DUMMY:
4561 break;
4562
4563 case PGSTAT_MTYPE_INQUIRY:
4564 pgstat_recv_inquiry(&msg.msg_inquiry, len);
4565 break;
4566
4567 case PGSTAT_MTYPE_TABSTAT:
4568 pgstat_recv_tabstat(&msg.msg_tabstat, len);
4569 break;
4570
4571 case PGSTAT_MTYPE_TABPURGE:
4572 pgstat_recv_tabpurge(&msg.msg_tabpurge, len);
4573 break;
4574
4575 case PGSTAT_MTYPE_DROPDB:
4576 pgstat_recv_dropdb(&msg.msg_dropdb, len);
4577 break;
4578
4579 case PGSTAT_MTYPE_RESETCOUNTER:
4580 pgstat_recv_resetcounter(&msg.msg_resetcounter, len);
4581 break;
4582
4583 case PGSTAT_MTYPE_RESETSHAREDCOUNTER:
4584 pgstat_recv_resetsharedcounter(
4585 &msg.msg_resetsharedcounter,
4586 len);
4587 break;
4588
4589 case PGSTAT_MTYPE_RESETSINGLECOUNTER:
4590 pgstat_recv_resetsinglecounter(
4591 &msg.msg_resetsinglecounter,
4592 len);
4593 break;
4594
4595 case PGSTAT_MTYPE_AUTOVAC_START:
4596 pgstat_recv_autovac(&msg.msg_autovacuum_start, len);
4597 break;
4598
4599 case PGSTAT_MTYPE_VACUUM:
4600 pgstat_recv_vacuum(&msg.msg_vacuum, len);
4601 break;
4602
4603 case PGSTAT_MTYPE_ANALYZE:
4604 pgstat_recv_analyze(&msg.msg_analyze, len);
4605 break;
4606
4607 case PGSTAT_MTYPE_ARCHIVER:
4608 pgstat_recv_archiver(&msg.msg_archiver, len);
4609 break;
4610
4611 case PGSTAT_MTYPE_BGWRITER:
4612 pgstat_recv_bgwriter(&msg.msg_bgwriter, len);
4613 break;
4614
4615 case PGSTAT_MTYPE_FUNCSTAT:
4616 pgstat_recv_funcstat(&msg.msg_funcstat, len);
4617 break;
4618
4619 case PGSTAT_MTYPE_FUNCPURGE:
4620 pgstat_recv_funcpurge(&msg.msg_funcpurge, len);
4621 break;
4622
4623 case PGSTAT_MTYPE_RECOVERYCONFLICT:
4624 pgstat_recv_recoveryconflict(
4625 &msg.msg_recoveryconflict,
4626 len);
4627 break;
4628
4629 case PGSTAT_MTYPE_DEADLOCK:
4630 pgstat_recv_deadlock(&msg.msg_deadlock, len);
4631 break;
4632
4633 case PGSTAT_MTYPE_TEMPFILE:
4634 pgstat_recv_tempfile(&msg.msg_tempfile, len);
4635 break;
4636
4637 case PGSTAT_MTYPE_CHECKSUMFAILURE:
4638 pgstat_recv_checksum_failure(
4639 &msg.msg_checksumfailure,
4640 len);
4641 break;
4642
4643 default:
4644 break;
4645 }
4646 } /* end of inner message-processing loop */
4647
4648 /* Sleep until there's something to do */
4649 #ifndef WIN32
4650 wr = WaitLatchOrSocket(MyLatch,
4651 WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE,
4652 pgStatSock, -1L,
4653 WAIT_EVENT_PGSTAT_MAIN);
4654 #else
4655
4656 /*
4657 * Windows, at least in its Windows Server 2003 R2 incarnation,
4658 * sometimes loses FD_READ events. Waking up and retrying the recv()
4659 * fixes that, so don't sleep indefinitely. This is a crock of the
4660 * first water, but until somebody wants to debug exactly what's
4661 * happening there, this is the best we can do. The two-second
4662 * timeout matches our pre-9.2 behavior, and needs to be short enough
4663 * to not provoke "using stale statistics" complaints from
4664 * backend_read_statsfile.
4665 */
4666 wr = WaitLatchOrSocket(MyLatch,
4667 WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE | WL_TIMEOUT,
4668 pgStatSock,
4669 2 * 1000L /* msec */ ,
4670 WAIT_EVENT_PGSTAT_MAIN);
4671 #endif
4672
4673 /*
4674 * Emergency bailout if postmaster has died. This is to avoid the
4675 * necessity for manual cleanup of all postmaster children.
4676 */
4677 if (wr & WL_POSTMASTER_DEATH)
4678 break;
4679 } /* end of outer loop */
4680
4681 /*
4682 * Save the final stats to reuse at next startup.
4683 */
4684 pgstat_write_statsfiles(true, true);
4685
4686 exit(0);
4687 }
4688
4689
4690 /* SIGQUIT signal handler for collector process */
4691 static void
pgstat_exit(SIGNAL_ARGS)4692 pgstat_exit(SIGNAL_ARGS)
4693 {
4694 int save_errno = errno;
4695
4696 need_exit = true;
4697 SetLatch(MyLatch);
4698
4699 errno = save_errno;
4700 }
4701
4702 /* SIGHUP handler for collector process */
4703 static void
pgstat_sighup_handler(SIGNAL_ARGS)4704 pgstat_sighup_handler(SIGNAL_ARGS)
4705 {
4706 int save_errno = errno;
4707
4708 got_SIGHUP = true;
4709 SetLatch(MyLatch);
4710
4711 errno = save_errno;
4712 }
4713
4714 /*
4715 * Subroutine to clear stats in a database entry
4716 *
4717 * Tables and functions hashes are initialized to empty.
4718 */
4719 static void
reset_dbentry_counters(PgStat_StatDBEntry * dbentry)4720 reset_dbentry_counters(PgStat_StatDBEntry *dbentry)
4721 {
4722 HASHCTL hash_ctl;
4723
4724 dbentry->n_xact_commit = 0;
4725 dbentry->n_xact_rollback = 0;
4726 dbentry->n_blocks_fetched = 0;
4727 dbentry->n_blocks_hit = 0;
4728 dbentry->n_tuples_returned = 0;
4729 dbentry->n_tuples_fetched = 0;
4730 dbentry->n_tuples_inserted = 0;
4731 dbentry->n_tuples_updated = 0;
4732 dbentry->n_tuples_deleted = 0;
4733 dbentry->last_autovac_time = 0;
4734 dbentry->n_conflict_tablespace = 0;
4735 dbentry->n_conflict_lock = 0;
4736 dbentry->n_conflict_snapshot = 0;
4737 dbentry->n_conflict_bufferpin = 0;
4738 dbentry->n_conflict_startup_deadlock = 0;
4739 dbentry->n_temp_files = 0;
4740 dbentry->n_temp_bytes = 0;
4741 dbentry->n_deadlocks = 0;
4742 dbentry->n_checksum_failures = 0;
4743 dbentry->last_checksum_failure = 0;
4744 dbentry->n_block_read_time = 0;
4745 dbentry->n_block_write_time = 0;
4746
4747 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
4748 dbentry->stats_timestamp = 0;
4749
4750 memset(&hash_ctl, 0, sizeof(hash_ctl));
4751 hash_ctl.keysize = sizeof(Oid);
4752 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
4753 dbentry->tables = hash_create("Per-database table",
4754 PGSTAT_TAB_HASH_SIZE,
4755 &hash_ctl,
4756 HASH_ELEM | HASH_BLOBS);
4757
4758 hash_ctl.keysize = sizeof(Oid);
4759 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
4760 dbentry->functions = hash_create("Per-database function",
4761 PGSTAT_FUNCTION_HASH_SIZE,
4762 &hash_ctl,
4763 HASH_ELEM | HASH_BLOBS);
4764 }
4765
4766 /*
4767 * Lookup the hash table entry for the specified database. If no hash
4768 * table entry exists, initialize it, if the create parameter is true.
4769 * Else, return NULL.
4770 */
4771 static PgStat_StatDBEntry *
pgstat_get_db_entry(Oid databaseid,bool create)4772 pgstat_get_db_entry(Oid databaseid, bool create)
4773 {
4774 PgStat_StatDBEntry *result;
4775 bool found;
4776 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
4777
4778 /* Lookup or create the hash table entry for this database */
4779 result = (PgStat_StatDBEntry *) hash_search(pgStatDBHash,
4780 &databaseid,
4781 action, &found);
4782
4783 if (!create && !found)
4784 return NULL;
4785
4786 /*
4787 * If not found, initialize the new one. This creates empty hash tables
4788 * for tables and functions, too.
4789 */
4790 if (!found)
4791 reset_dbentry_counters(result);
4792
4793 return result;
4794 }
4795
4796
4797 /*
4798 * Lookup the hash table entry for the specified table. If no hash
4799 * table entry exists, initialize it, if the create parameter is true.
4800 * Else, return NULL.
4801 */
4802 static PgStat_StatTabEntry *
pgstat_get_tab_entry(PgStat_StatDBEntry * dbentry,Oid tableoid,bool create)4803 pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
4804 {
4805 PgStat_StatTabEntry *result;
4806 bool found;
4807 HASHACTION action = (create ? HASH_ENTER : HASH_FIND);
4808
4809 /* Lookup or create the hash table entry for this table */
4810 result = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
4811 &tableoid,
4812 action, &found);
4813
4814 if (!create && !found)
4815 return NULL;
4816
4817 /* If not found, initialize the new one. */
4818 if (!found)
4819 {
4820 result->numscans = 0;
4821 result->tuples_returned = 0;
4822 result->tuples_fetched = 0;
4823 result->tuples_inserted = 0;
4824 result->tuples_updated = 0;
4825 result->tuples_deleted = 0;
4826 result->tuples_hot_updated = 0;
4827 result->n_live_tuples = 0;
4828 result->n_dead_tuples = 0;
4829 result->changes_since_analyze = 0;
4830 result->blocks_fetched = 0;
4831 result->blocks_hit = 0;
4832 result->vacuum_timestamp = 0;
4833 result->vacuum_count = 0;
4834 result->autovac_vacuum_timestamp = 0;
4835 result->autovac_vacuum_count = 0;
4836 result->analyze_timestamp = 0;
4837 result->analyze_count = 0;
4838 result->autovac_analyze_timestamp = 0;
4839 result->autovac_analyze_count = 0;
4840 }
4841
4842 return result;
4843 }
4844
4845
4846 /* ----------
4847 * pgstat_write_statsfiles() -
4848 * Write the global statistics file, as well as requested DB files.
4849 *
4850 * 'permanent' specifies writing to the permanent files not temporary ones.
4851 * When true (happens only when the collector is shutting down), also remove
4852 * the temporary files so that backends starting up under a new postmaster
4853 * can't read old data before the new collector is ready.
4854 *
4855 * When 'allDbs' is false, only the requested databases (listed in
4856 * pending_write_requests) will be written; otherwise, all databases
4857 * will be written.
4858 * ----------
4859 */
4860 static void
pgstat_write_statsfiles(bool permanent,bool allDbs)4861 pgstat_write_statsfiles(bool permanent, bool allDbs)
4862 {
4863 HASH_SEQ_STATUS hstat;
4864 PgStat_StatDBEntry *dbentry;
4865 FILE *fpout;
4866 int32 format_id;
4867 const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
4868 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
4869 int rc;
4870
4871 elog(DEBUG2, "writing stats file \"%s\"", statfile);
4872
4873 /*
4874 * Open the statistics temp file to write out the current values.
4875 */
4876 fpout = AllocateFile(tmpfile, PG_BINARY_W);
4877 if (fpout == NULL)
4878 {
4879 ereport(LOG,
4880 (errcode_for_file_access(),
4881 errmsg("could not open temporary statistics file \"%s\": %m",
4882 tmpfile)));
4883 return;
4884 }
4885
4886 /*
4887 * Set the timestamp of the stats file.
4888 */
4889 globalStats.stats_timestamp = GetCurrentTimestamp();
4890
4891 /*
4892 * Write the file header --- currently just a format ID.
4893 */
4894 format_id = PGSTAT_FILE_FORMAT_ID;
4895 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
4896 (void) rc; /* we'll check for error with ferror */
4897
4898 /*
4899 * Write global stats struct
4900 */
4901 rc = fwrite(&globalStats, sizeof(globalStats), 1, fpout);
4902 (void) rc; /* we'll check for error with ferror */
4903
4904 /*
4905 * Write archiver stats struct
4906 */
4907 rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
4908 (void) rc; /* we'll check for error with ferror */
4909
4910 /*
4911 * Walk through the database table.
4912 */
4913 hash_seq_init(&hstat, pgStatDBHash);
4914 while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL)
4915 {
4916 /*
4917 * Write out the table and function stats for this DB into the
4918 * appropriate per-DB stat file, if required.
4919 */
4920 if (allDbs || pgstat_db_requested(dbentry->databaseid))
4921 {
4922 /* Make DB's timestamp consistent with the global stats */
4923 dbentry->stats_timestamp = globalStats.stats_timestamp;
4924
4925 pgstat_write_db_statsfile(dbentry, permanent);
4926 }
4927
4928 /*
4929 * Write out the DB entry. We don't write the tables or functions
4930 * pointers, since they're of no use to any other process.
4931 */
4932 fputc('D', fpout);
4933 rc = fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout);
4934 (void) rc; /* we'll check for error with ferror */
4935 }
4936
4937 /*
4938 * No more output to be done. Close the temp file and replace the old
4939 * pgstat.stat with it. The ferror() check replaces testing for error
4940 * after each individual fputc or fwrite above.
4941 */
4942 fputc('E', fpout);
4943
4944 if (ferror(fpout))
4945 {
4946 ereport(LOG,
4947 (errcode_for_file_access(),
4948 errmsg("could not write temporary statistics file \"%s\": %m",
4949 tmpfile)));
4950 FreeFile(fpout);
4951 unlink(tmpfile);
4952 }
4953 else if (FreeFile(fpout) < 0)
4954 {
4955 ereport(LOG,
4956 (errcode_for_file_access(),
4957 errmsg("could not close temporary statistics file \"%s\": %m",
4958 tmpfile)));
4959 unlink(tmpfile);
4960 }
4961 else if (rename(tmpfile, statfile) < 0)
4962 {
4963 ereport(LOG,
4964 (errcode_for_file_access(),
4965 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
4966 tmpfile, statfile)));
4967 unlink(tmpfile);
4968 }
4969
4970 if (permanent)
4971 unlink(pgstat_stat_filename);
4972
4973 /*
4974 * Now throw away the list of requests. Note that requests sent after we
4975 * started the write are still waiting on the network socket.
4976 */
4977 list_free(pending_write_requests);
4978 pending_write_requests = NIL;
4979 }
4980
4981 /*
4982 * return the filename for a DB stat file; filename is the output buffer,
4983 * of length len.
4984 */
4985 static void
get_dbstat_filename(bool permanent,bool tempname,Oid databaseid,char * filename,int len)4986 get_dbstat_filename(bool permanent, bool tempname, Oid databaseid,
4987 char *filename, int len)
4988 {
4989 int printed;
4990
4991 /* NB -- pgstat_reset_remove_files knows about the pattern this uses */
4992 printed = snprintf(filename, len, "%s/db_%u.%s",
4993 permanent ? PGSTAT_STAT_PERMANENT_DIRECTORY :
4994 pgstat_stat_directory,
4995 databaseid,
4996 tempname ? "tmp" : "stat");
4997 if (printed >= len)
4998 elog(ERROR, "overlength pgstat path");
4999 }
5000
5001 /* ----------
5002 * pgstat_write_db_statsfile() -
5003 * Write the stat file for a single database.
5004 *
5005 * If writing to the permanent file (happens when the collector is
5006 * shutting down only), remove the temporary file so that backends
5007 * starting up under a new postmaster can't read the old data before
5008 * the new collector is ready.
5009 * ----------
5010 */
5011 static void
pgstat_write_db_statsfile(PgStat_StatDBEntry * dbentry,bool permanent)5012 pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
5013 {
5014 HASH_SEQ_STATUS tstat;
5015 HASH_SEQ_STATUS fstat;
5016 PgStat_StatTabEntry *tabentry;
5017 PgStat_StatFuncEntry *funcentry;
5018 FILE *fpout;
5019 int32 format_id;
5020 Oid dbid = dbentry->databaseid;
5021 int rc;
5022 char tmpfile[MAXPGPATH];
5023 char statfile[MAXPGPATH];
5024
5025 get_dbstat_filename(permanent, true, dbid, tmpfile, MAXPGPATH);
5026 get_dbstat_filename(permanent, false, dbid, statfile, MAXPGPATH);
5027
5028 elog(DEBUG2, "writing stats file \"%s\"", statfile);
5029
5030 /*
5031 * Open the statistics temp file to write out the current values.
5032 */
5033 fpout = AllocateFile(tmpfile, PG_BINARY_W);
5034 if (fpout == NULL)
5035 {
5036 ereport(LOG,
5037 (errcode_for_file_access(),
5038 errmsg("could not open temporary statistics file \"%s\": %m",
5039 tmpfile)));
5040 return;
5041 }
5042
5043 /*
5044 * Write the file header --- currently just a format ID.
5045 */
5046 format_id = PGSTAT_FILE_FORMAT_ID;
5047 rc = fwrite(&format_id, sizeof(format_id), 1, fpout);
5048 (void) rc; /* we'll check for error with ferror */
5049
5050 /*
5051 * Walk through the database's access stats per table.
5052 */
5053 hash_seq_init(&tstat, dbentry->tables);
5054 while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL)
5055 {
5056 fputc('T', fpout);
5057 rc = fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout);
5058 (void) rc; /* we'll check for error with ferror */
5059 }
5060
5061 /*
5062 * Walk through the database's function stats table.
5063 */
5064 hash_seq_init(&fstat, dbentry->functions);
5065 while ((funcentry = (PgStat_StatFuncEntry *) hash_seq_search(&fstat)) != NULL)
5066 {
5067 fputc('F', fpout);
5068 rc = fwrite(funcentry, sizeof(PgStat_StatFuncEntry), 1, fpout);
5069 (void) rc; /* we'll check for error with ferror */
5070 }
5071
5072 /*
5073 * No more output to be done. Close the temp file and replace the old
5074 * pgstat.stat with it. The ferror() check replaces testing for error
5075 * after each individual fputc or fwrite above.
5076 */
5077 fputc('E', fpout);
5078
5079 if (ferror(fpout))
5080 {
5081 ereport(LOG,
5082 (errcode_for_file_access(),
5083 errmsg("could not write temporary statistics file \"%s\": %m",
5084 tmpfile)));
5085 FreeFile(fpout);
5086 unlink(tmpfile);
5087 }
5088 else if (FreeFile(fpout) < 0)
5089 {
5090 ereport(LOG,
5091 (errcode_for_file_access(),
5092 errmsg("could not close temporary statistics file \"%s\": %m",
5093 tmpfile)));
5094 unlink(tmpfile);
5095 }
5096 else if (rename(tmpfile, statfile) < 0)
5097 {
5098 ereport(LOG,
5099 (errcode_for_file_access(),
5100 errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m",
5101 tmpfile, statfile)));
5102 unlink(tmpfile);
5103 }
5104
5105 if (permanent)
5106 {
5107 get_dbstat_filename(false, false, dbid, statfile, MAXPGPATH);
5108
5109 elog(DEBUG2, "removing temporary stats file \"%s\"", statfile);
5110 unlink(statfile);
5111 }
5112 }
5113
5114 /* ----------
5115 * pgstat_read_statsfiles() -
5116 *
5117 * Reads in some existing statistics collector files and returns the
5118 * databases hash table that is the top level of the data.
5119 *
5120 * If 'onlydb' is not InvalidOid, it means we only want data for that DB
5121 * plus the shared catalogs ("DB 0"). We'll still populate the DB hash
5122 * table for all databases, but we don't bother even creating table/function
5123 * hash tables for other databases.
5124 *
5125 * 'permanent' specifies reading from the permanent files not temporary ones.
5126 * When true (happens only when the collector is starting up), remove the
5127 * files after reading; the in-memory status is now authoritative, and the
5128 * files would be out of date in case somebody else reads them.
5129 *
5130 * If a 'deep' read is requested, table/function stats are read, otherwise
5131 * the table/function hash tables remain empty.
5132 * ----------
5133 */
5134 static HTAB *
pgstat_read_statsfiles(Oid onlydb,bool permanent,bool deep)5135 pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
5136 {
5137 PgStat_StatDBEntry *dbentry;
5138 PgStat_StatDBEntry dbbuf;
5139 HASHCTL hash_ctl;
5140 HTAB *dbhash;
5141 FILE *fpin;
5142 int32 format_id;
5143 bool found;
5144 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
5145
5146 /*
5147 * The tables will live in pgStatLocalContext.
5148 */
5149 pgstat_setup_memcxt();
5150
5151 /*
5152 * Create the DB hashtable
5153 */
5154 memset(&hash_ctl, 0, sizeof(hash_ctl));
5155 hash_ctl.keysize = sizeof(Oid);
5156 hash_ctl.entrysize = sizeof(PgStat_StatDBEntry);
5157 hash_ctl.hcxt = pgStatLocalContext;
5158 dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl,
5159 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
5160
5161 /*
5162 * Clear out global and archiver statistics so they start from zero in
5163 * case we can't load an existing statsfile.
5164 */
5165 memset(&globalStats, 0, sizeof(globalStats));
5166 memset(&archiverStats, 0, sizeof(archiverStats));
5167
5168 /*
5169 * Set the current timestamp (will be kept only in case we can't load an
5170 * existing statsfile).
5171 */
5172 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
5173 archiverStats.stat_reset_timestamp = globalStats.stat_reset_timestamp;
5174
5175 /*
5176 * Try to open the stats file. If it doesn't exist, the backends simply
5177 * return zero for anything and the collector simply starts from scratch
5178 * with empty counters.
5179 *
5180 * ENOENT is a possibility if the stats collector is not running or has
5181 * not yet written the stats file the first time. Any other failure
5182 * condition is suspicious.
5183 */
5184 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
5185 {
5186 if (errno != ENOENT)
5187 ereport(pgStatRunningInCollector ? LOG : WARNING,
5188 (errcode_for_file_access(),
5189 errmsg("could not open statistics file \"%s\": %m",
5190 statfile)));
5191 return dbhash;
5192 }
5193
5194 /*
5195 * Verify it's of the expected format.
5196 */
5197 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
5198 format_id != PGSTAT_FILE_FORMAT_ID)
5199 {
5200 ereport(pgStatRunningInCollector ? LOG : WARNING,
5201 (errmsg("corrupted statistics file \"%s\"", statfile)));
5202 goto done;
5203 }
5204
5205 /*
5206 * Read global stats struct
5207 */
5208 if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats))
5209 {
5210 ereport(pgStatRunningInCollector ? LOG : WARNING,
5211 (errmsg("corrupted statistics file \"%s\"", statfile)));
5212 memset(&globalStats, 0, sizeof(globalStats));
5213 goto done;
5214 }
5215
5216 /*
5217 * In the collector, disregard the timestamp we read from the permanent
5218 * stats file; we should be willing to write a temp stats file immediately
5219 * upon the first request from any backend. This only matters if the old
5220 * file's timestamp is less than PGSTAT_STAT_INTERVAL ago, but that's not
5221 * an unusual scenario.
5222 */
5223 if (pgStatRunningInCollector)
5224 globalStats.stats_timestamp = 0;
5225
5226 /*
5227 * Read archiver stats struct
5228 */
5229 if (fread(&archiverStats, 1, sizeof(archiverStats), fpin) != sizeof(archiverStats))
5230 {
5231 ereport(pgStatRunningInCollector ? LOG : WARNING,
5232 (errmsg("corrupted statistics file \"%s\"", statfile)));
5233 memset(&archiverStats, 0, sizeof(archiverStats));
5234 goto done;
5235 }
5236
5237 /*
5238 * We found an existing collector stats file. Read it and put all the
5239 * hashtable entries into place.
5240 */
5241 for (;;)
5242 {
5243 switch (fgetc(fpin))
5244 {
5245 /*
5246 * 'D' A PgStat_StatDBEntry struct describing a database
5247 * follows.
5248 */
5249 case 'D':
5250 if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables),
5251 fpin) != offsetof(PgStat_StatDBEntry, tables))
5252 {
5253 ereport(pgStatRunningInCollector ? LOG : WARNING,
5254 (errmsg("corrupted statistics file \"%s\"",
5255 statfile)));
5256 goto done;
5257 }
5258
5259 /*
5260 * Add to the DB hash
5261 */
5262 dbentry = (PgStat_StatDBEntry *) hash_search(dbhash,
5263 (void *) &dbbuf.databaseid,
5264 HASH_ENTER,
5265 &found);
5266 if (found)
5267 {
5268 ereport(pgStatRunningInCollector ? LOG : WARNING,
5269 (errmsg("corrupted statistics file \"%s\"",
5270 statfile)));
5271 goto done;
5272 }
5273
5274 memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry));
5275 dbentry->tables = NULL;
5276 dbentry->functions = NULL;
5277
5278 /*
5279 * In the collector, disregard the timestamp we read from the
5280 * permanent stats file; we should be willing to write a temp
5281 * stats file immediately upon the first request from any
5282 * backend.
5283 */
5284 if (pgStatRunningInCollector)
5285 dbentry->stats_timestamp = 0;
5286
5287 /*
5288 * Don't create tables/functions hashtables for uninteresting
5289 * databases.
5290 */
5291 if (onlydb != InvalidOid)
5292 {
5293 if (dbbuf.databaseid != onlydb &&
5294 dbbuf.databaseid != InvalidOid)
5295 break;
5296 }
5297
5298 memset(&hash_ctl, 0, sizeof(hash_ctl));
5299 hash_ctl.keysize = sizeof(Oid);
5300 hash_ctl.entrysize = sizeof(PgStat_StatTabEntry);
5301 hash_ctl.hcxt = pgStatLocalContext;
5302 dbentry->tables = hash_create("Per-database table",
5303 PGSTAT_TAB_HASH_SIZE,
5304 &hash_ctl,
5305 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
5306
5307 hash_ctl.keysize = sizeof(Oid);
5308 hash_ctl.entrysize = sizeof(PgStat_StatFuncEntry);
5309 hash_ctl.hcxt = pgStatLocalContext;
5310 dbentry->functions = hash_create("Per-database function",
5311 PGSTAT_FUNCTION_HASH_SIZE,
5312 &hash_ctl,
5313 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
5314
5315 /*
5316 * If requested, read the data from the database-specific
5317 * file. Otherwise we just leave the hashtables empty.
5318 */
5319 if (deep)
5320 pgstat_read_db_statsfile(dbentry->databaseid,
5321 dbentry->tables,
5322 dbentry->functions,
5323 permanent);
5324
5325 break;
5326
5327 case 'E':
5328 goto done;
5329
5330 default:
5331 ereport(pgStatRunningInCollector ? LOG : WARNING,
5332 (errmsg("corrupted statistics file \"%s\"",
5333 statfile)));
5334 goto done;
5335 }
5336 }
5337
5338 done:
5339 FreeFile(fpin);
5340
5341 /* If requested to read the permanent file, also get rid of it. */
5342 if (permanent)
5343 {
5344 elog(DEBUG2, "removing permanent stats file \"%s\"", statfile);
5345 unlink(statfile);
5346 }
5347
5348 return dbhash;
5349 }
5350
5351
5352 /* ----------
5353 * pgstat_read_db_statsfile() -
5354 *
5355 * Reads in the existing statistics collector file for the given database,
5356 * filling the passed-in tables and functions hash tables.
5357 *
5358 * As in pgstat_read_statsfiles, if the permanent file is requested, it is
5359 * removed after reading.
5360 *
5361 * Note: this code has the ability to skip storing per-table or per-function
5362 * data, if NULL is passed for the corresponding hashtable. That's not used
5363 * at the moment though.
5364 * ----------
5365 */
5366 static void
pgstat_read_db_statsfile(Oid databaseid,HTAB * tabhash,HTAB * funchash,bool permanent)5367 pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash,
5368 bool permanent)
5369 {
5370 PgStat_StatTabEntry *tabentry;
5371 PgStat_StatTabEntry tabbuf;
5372 PgStat_StatFuncEntry funcbuf;
5373 PgStat_StatFuncEntry *funcentry;
5374 FILE *fpin;
5375 int32 format_id;
5376 bool found;
5377 char statfile[MAXPGPATH];
5378
5379 get_dbstat_filename(permanent, false, databaseid, statfile, MAXPGPATH);
5380
5381 /*
5382 * Try to open the stats file. If it doesn't exist, the backends simply
5383 * return zero for anything and the collector simply starts from scratch
5384 * with empty counters.
5385 *
5386 * ENOENT is a possibility if the stats collector is not running or has
5387 * not yet written the stats file the first time. Any other failure
5388 * condition is suspicious.
5389 */
5390 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
5391 {
5392 if (errno != ENOENT)
5393 ereport(pgStatRunningInCollector ? LOG : WARNING,
5394 (errcode_for_file_access(),
5395 errmsg("could not open statistics file \"%s\": %m",
5396 statfile)));
5397 return;
5398 }
5399
5400 /*
5401 * Verify it's of the expected format.
5402 */
5403 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
5404 format_id != PGSTAT_FILE_FORMAT_ID)
5405 {
5406 ereport(pgStatRunningInCollector ? LOG : WARNING,
5407 (errmsg("corrupted statistics file \"%s\"", statfile)));
5408 goto done;
5409 }
5410
5411 /*
5412 * We found an existing collector stats file. Read it and put all the
5413 * hashtable entries into place.
5414 */
5415 for (;;)
5416 {
5417 switch (fgetc(fpin))
5418 {
5419 /*
5420 * 'T' A PgStat_StatTabEntry follows.
5421 */
5422 case 'T':
5423 if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry),
5424 fpin) != sizeof(PgStat_StatTabEntry))
5425 {
5426 ereport(pgStatRunningInCollector ? LOG : WARNING,
5427 (errmsg("corrupted statistics file \"%s\"",
5428 statfile)));
5429 goto done;
5430 }
5431
5432 /*
5433 * Skip if table data not wanted.
5434 */
5435 if (tabhash == NULL)
5436 break;
5437
5438 tabentry = (PgStat_StatTabEntry *) hash_search(tabhash,
5439 (void *) &tabbuf.tableid,
5440 HASH_ENTER, &found);
5441
5442 if (found)
5443 {
5444 ereport(pgStatRunningInCollector ? LOG : WARNING,
5445 (errmsg("corrupted statistics file \"%s\"",
5446 statfile)));
5447 goto done;
5448 }
5449
5450 memcpy(tabentry, &tabbuf, sizeof(tabbuf));
5451 break;
5452
5453 /*
5454 * 'F' A PgStat_StatFuncEntry follows.
5455 */
5456 case 'F':
5457 if (fread(&funcbuf, 1, sizeof(PgStat_StatFuncEntry),
5458 fpin) != sizeof(PgStat_StatFuncEntry))
5459 {
5460 ereport(pgStatRunningInCollector ? LOG : WARNING,
5461 (errmsg("corrupted statistics file \"%s\"",
5462 statfile)));
5463 goto done;
5464 }
5465
5466 /*
5467 * Skip if function data not wanted.
5468 */
5469 if (funchash == NULL)
5470 break;
5471
5472 funcentry = (PgStat_StatFuncEntry *) hash_search(funchash,
5473 (void *) &funcbuf.functionid,
5474 HASH_ENTER, &found);
5475
5476 if (found)
5477 {
5478 ereport(pgStatRunningInCollector ? LOG : WARNING,
5479 (errmsg("corrupted statistics file \"%s\"",
5480 statfile)));
5481 goto done;
5482 }
5483
5484 memcpy(funcentry, &funcbuf, sizeof(funcbuf));
5485 break;
5486
5487 /*
5488 * 'E' The EOF marker of a complete stats file.
5489 */
5490 case 'E':
5491 goto done;
5492
5493 default:
5494 ereport(pgStatRunningInCollector ? LOG : WARNING,
5495 (errmsg("corrupted statistics file \"%s\"",
5496 statfile)));
5497 goto done;
5498 }
5499 }
5500
5501 done:
5502 FreeFile(fpin);
5503
5504 if (permanent)
5505 {
5506 elog(DEBUG2, "removing permanent stats file \"%s\"", statfile);
5507 unlink(statfile);
5508 }
5509 }
5510
5511 /* ----------
5512 * pgstat_read_db_statsfile_timestamp() -
5513 *
5514 * Attempt to determine the timestamp of the last db statfile write.
5515 * Returns true if successful; the timestamp is stored in *ts.
5516 *
5517 * This needs to be careful about handling databases for which no stats file
5518 * exists, such as databases without a stat entry or those not yet written:
5519 *
5520 * - if there's a database entry in the global file, return the corresponding
5521 * stats_timestamp value.
5522 *
5523 * - if there's no db stat entry (e.g. for a new or inactive database),
5524 * there's no stats_timestamp value, but also nothing to write so we return
5525 * the timestamp of the global statfile.
5526 * ----------
5527 */
5528 static bool
pgstat_read_db_statsfile_timestamp(Oid databaseid,bool permanent,TimestampTz * ts)5529 pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
5530 TimestampTz *ts)
5531 {
5532 PgStat_StatDBEntry dbentry;
5533 PgStat_GlobalStats myGlobalStats;
5534 PgStat_ArchiverStats myArchiverStats;
5535 FILE *fpin;
5536 int32 format_id;
5537 const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
5538
5539 /*
5540 * Try to open the stats file. As above, anything but ENOENT is worthy of
5541 * complaining about.
5542 */
5543 if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL)
5544 {
5545 if (errno != ENOENT)
5546 ereport(pgStatRunningInCollector ? LOG : WARNING,
5547 (errcode_for_file_access(),
5548 errmsg("could not open statistics file \"%s\": %m",
5549 statfile)));
5550 return false;
5551 }
5552
5553 /*
5554 * Verify it's of the expected format.
5555 */
5556 if (fread(&format_id, 1, sizeof(format_id), fpin) != sizeof(format_id) ||
5557 format_id != PGSTAT_FILE_FORMAT_ID)
5558 {
5559 ereport(pgStatRunningInCollector ? LOG : WARNING,
5560 (errmsg("corrupted statistics file \"%s\"", statfile)));
5561 FreeFile(fpin);
5562 return false;
5563 }
5564
5565 /*
5566 * Read global stats struct
5567 */
5568 if (fread(&myGlobalStats, 1, sizeof(myGlobalStats),
5569 fpin) != sizeof(myGlobalStats))
5570 {
5571 ereport(pgStatRunningInCollector ? LOG : WARNING,
5572 (errmsg("corrupted statistics file \"%s\"", statfile)));
5573 FreeFile(fpin);
5574 return false;
5575 }
5576
5577 /*
5578 * Read archiver stats struct
5579 */
5580 if (fread(&myArchiverStats, 1, sizeof(myArchiverStats),
5581 fpin) != sizeof(myArchiverStats))
5582 {
5583 ereport(pgStatRunningInCollector ? LOG : WARNING,
5584 (errmsg("corrupted statistics file \"%s\"", statfile)));
5585 FreeFile(fpin);
5586 return false;
5587 }
5588
5589 /* By default, we're going to return the timestamp of the global file. */
5590 *ts = myGlobalStats.stats_timestamp;
5591
5592 /*
5593 * We found an existing collector stats file. Read it and look for a
5594 * record for the requested database. If found, use its timestamp.
5595 */
5596 for (;;)
5597 {
5598 switch (fgetc(fpin))
5599 {
5600 /*
5601 * 'D' A PgStat_StatDBEntry struct describing a database
5602 * follows.
5603 */
5604 case 'D':
5605 if (fread(&dbentry, 1, offsetof(PgStat_StatDBEntry, tables),
5606 fpin) != offsetof(PgStat_StatDBEntry, tables))
5607 {
5608 ereport(pgStatRunningInCollector ? LOG : WARNING,
5609 (errmsg("corrupted statistics file \"%s\"",
5610 statfile)));
5611 goto done;
5612 }
5613
5614 /*
5615 * If this is the DB we're looking for, save its timestamp and
5616 * we're done.
5617 */
5618 if (dbentry.databaseid == databaseid)
5619 {
5620 *ts = dbentry.stats_timestamp;
5621 goto done;
5622 }
5623
5624 break;
5625
5626 case 'E':
5627 goto done;
5628
5629 default:
5630 ereport(pgStatRunningInCollector ? LOG : WARNING,
5631 (errmsg("corrupted statistics file \"%s\"",
5632 statfile)));
5633 goto done;
5634 }
5635 }
5636
5637 done:
5638 FreeFile(fpin);
5639 return true;
5640 }
5641
5642 /*
5643 * If not already done, read the statistics collector stats file into
5644 * some hash tables. The results will be kept until pgstat_clear_snapshot()
5645 * is called (typically, at end of transaction).
5646 */
5647 static void
backend_read_statsfile(void)5648 backend_read_statsfile(void)
5649 {
5650 TimestampTz min_ts = 0;
5651 TimestampTz ref_ts = 0;
5652 Oid inquiry_db;
5653 int count;
5654
5655 /* already read it? */
5656 if (pgStatDBHash)
5657 return;
5658 Assert(!pgStatRunningInCollector);
5659
5660 /*
5661 * In a normal backend, we check staleness of the data for our own DB, and
5662 * so we send MyDatabaseId in inquiry messages. In the autovac launcher,
5663 * check staleness of the shared-catalog data, and send InvalidOid in
5664 * inquiry messages so as not to force writing unnecessary data.
5665 */
5666 if (IsAutoVacuumLauncherProcess())
5667 inquiry_db = InvalidOid;
5668 else
5669 inquiry_db = MyDatabaseId;
5670
5671 /*
5672 * Loop until fresh enough stats file is available or we ran out of time.
5673 * The stats inquiry message is sent repeatedly in case collector drops
5674 * it; but not every single time, as that just swamps the collector.
5675 */
5676 for (count = 0; count < PGSTAT_POLL_LOOP_COUNT; count++)
5677 {
5678 bool ok;
5679 TimestampTz file_ts = 0;
5680 TimestampTz cur_ts;
5681
5682 CHECK_FOR_INTERRUPTS();
5683
5684 ok = pgstat_read_db_statsfile_timestamp(inquiry_db, false, &file_ts);
5685
5686 cur_ts = GetCurrentTimestamp();
5687 /* Calculate min acceptable timestamp, if we didn't already */
5688 if (count == 0 || cur_ts < ref_ts)
5689 {
5690 /*
5691 * We set the minimum acceptable timestamp to PGSTAT_STAT_INTERVAL
5692 * msec before now. This indirectly ensures that the collector
5693 * needn't write the file more often than PGSTAT_STAT_INTERVAL. In
5694 * an autovacuum worker, however, we want a lower delay to avoid
5695 * using stale data, so we use PGSTAT_RETRY_DELAY (since the
5696 * number of workers is low, this shouldn't be a problem).
5697 *
5698 * We don't recompute min_ts after sleeping, except in the
5699 * unlikely case that cur_ts went backwards. So we might end up
5700 * accepting a file a bit older than PGSTAT_STAT_INTERVAL. In
5701 * practice that shouldn't happen, though, as long as the sleep
5702 * time is less than PGSTAT_STAT_INTERVAL; and we don't want to
5703 * tell the collector that our cutoff time is less than what we'd
5704 * actually accept.
5705 */
5706 ref_ts = cur_ts;
5707 if (IsAutoVacuumWorkerProcess())
5708 min_ts = TimestampTzPlusMilliseconds(ref_ts,
5709 -PGSTAT_RETRY_DELAY);
5710 else
5711 min_ts = TimestampTzPlusMilliseconds(ref_ts,
5712 -PGSTAT_STAT_INTERVAL);
5713 }
5714
5715 /*
5716 * If the file timestamp is actually newer than cur_ts, we must have
5717 * had a clock glitch (system time went backwards) or there is clock
5718 * skew between our processor and the stats collector's processor.
5719 * Accept the file, but send an inquiry message anyway to make
5720 * pgstat_recv_inquiry do a sanity check on the collector's time.
5721 */
5722 if (ok && file_ts > cur_ts)
5723 {
5724 /*
5725 * A small amount of clock skew between processors isn't terribly
5726 * surprising, but a large difference is worth logging. We
5727 * arbitrarily define "large" as 1000 msec.
5728 */
5729 if (file_ts >= TimestampTzPlusMilliseconds(cur_ts, 1000))
5730 {
5731 char *filetime;
5732 char *mytime;
5733
5734 /* Copy because timestamptz_to_str returns a static buffer */
5735 filetime = pstrdup(timestamptz_to_str(file_ts));
5736 mytime = pstrdup(timestamptz_to_str(cur_ts));
5737 elog(LOG, "stats collector's time %s is later than backend local time %s",
5738 filetime, mytime);
5739 pfree(filetime);
5740 pfree(mytime);
5741 }
5742
5743 pgstat_send_inquiry(cur_ts, min_ts, inquiry_db);
5744 break;
5745 }
5746
5747 /* Normal acceptance case: file is not older than cutoff time */
5748 if (ok && file_ts >= min_ts)
5749 break;
5750
5751 /* Not there or too old, so kick the collector and wait a bit */
5752 if ((count % PGSTAT_INQ_LOOP_COUNT) == 0)
5753 pgstat_send_inquiry(cur_ts, min_ts, inquiry_db);
5754
5755 pg_usleep(PGSTAT_RETRY_DELAY * 1000L);
5756 }
5757
5758 if (count >= PGSTAT_POLL_LOOP_COUNT)
5759 ereport(LOG,
5760 (errmsg("using stale statistics instead of current ones "
5761 "because stats collector is not responding")));
5762
5763 /*
5764 * Autovacuum launcher wants stats about all databases, but a shallow read
5765 * is sufficient. Regular backends want a deep read for just the tables
5766 * they can see (MyDatabaseId + shared catalogs).
5767 */
5768 if (IsAutoVacuumLauncherProcess())
5769 pgStatDBHash = pgstat_read_statsfiles(InvalidOid, false, false);
5770 else
5771 pgStatDBHash = pgstat_read_statsfiles(MyDatabaseId, false, true);
5772 }
5773
5774
5775 /* ----------
5776 * pgstat_setup_memcxt() -
5777 *
5778 * Create pgStatLocalContext, if not already done.
5779 * ----------
5780 */
5781 static void
pgstat_setup_memcxt(void)5782 pgstat_setup_memcxt(void)
5783 {
5784 if (!pgStatLocalContext)
5785 pgStatLocalContext = AllocSetContextCreate(TopMemoryContext,
5786 "Statistics snapshot",
5787 ALLOCSET_SMALL_SIZES);
5788 }
5789
5790
5791 /* ----------
5792 * pgstat_clear_snapshot() -
5793 *
5794 * Discard any data collected in the current transaction. Any subsequent
5795 * request will cause new snapshots to be read.
5796 *
5797 * This is also invoked during transaction commit or abort to discard
5798 * the no-longer-wanted snapshot.
5799 * ----------
5800 */
5801 void
pgstat_clear_snapshot(void)5802 pgstat_clear_snapshot(void)
5803 {
5804 /* Release memory, if any was allocated */
5805 if (pgStatLocalContext)
5806 MemoryContextDelete(pgStatLocalContext);
5807
5808 /* Reset variables */
5809 pgStatLocalContext = NULL;
5810 pgStatDBHash = NULL;
5811 localBackendStatusTable = NULL;
5812 localNumBackends = 0;
5813 }
5814
5815
5816 /* ----------
5817 * pgstat_recv_inquiry() -
5818 *
5819 * Process stat inquiry requests.
5820 * ----------
5821 */
5822 static void
pgstat_recv_inquiry(PgStat_MsgInquiry * msg,int len)5823 pgstat_recv_inquiry(PgStat_MsgInquiry *msg, int len)
5824 {
5825 PgStat_StatDBEntry *dbentry;
5826
5827 elog(DEBUG2, "received inquiry for database %u", msg->databaseid);
5828
5829 /*
5830 * If there's already a write request for this DB, there's nothing to do.
5831 *
5832 * Note that if a request is found, we return early and skip the below
5833 * check for clock skew. This is okay, since the only way for a DB
5834 * request to be present in the list is that we have been here since the
5835 * last write round. It seems sufficient to check for clock skew once per
5836 * write round.
5837 */
5838 if (list_member_oid(pending_write_requests, msg->databaseid))
5839 return;
5840
5841 /*
5842 * Check to see if we last wrote this database at a time >= the requested
5843 * cutoff time. If so, this is a stale request that was generated before
5844 * we updated the DB file, and we don't need to do so again.
5845 *
5846 * If the requestor's local clock time is older than stats_timestamp, we
5847 * should suspect a clock glitch, ie system time going backwards; though
5848 * the more likely explanation is just delayed message receipt. It is
5849 * worth expending a GetCurrentTimestamp call to be sure, since a large
5850 * retreat in the system clock reading could otherwise cause us to neglect
5851 * to update the stats file for a long time.
5852 */
5853 dbentry = pgstat_get_db_entry(msg->databaseid, false);
5854 if (dbentry == NULL)
5855 {
5856 /*
5857 * We have no data for this DB. Enter a write request anyway so that
5858 * the global stats will get updated. This is needed to prevent
5859 * backend_read_statsfile from waiting for data that we cannot supply,
5860 * in the case of a new DB that nobody has yet reported any stats for.
5861 * See the behavior of pgstat_read_db_statsfile_timestamp.
5862 */
5863 }
5864 else if (msg->clock_time < dbentry->stats_timestamp)
5865 {
5866 TimestampTz cur_ts = GetCurrentTimestamp();
5867
5868 if (cur_ts < dbentry->stats_timestamp)
5869 {
5870 /*
5871 * Sure enough, time went backwards. Force a new stats file write
5872 * to get back in sync; but first, log a complaint.
5873 */
5874 char *writetime;
5875 char *mytime;
5876
5877 /* Copy because timestamptz_to_str returns a static buffer */
5878 writetime = pstrdup(timestamptz_to_str(dbentry->stats_timestamp));
5879 mytime = pstrdup(timestamptz_to_str(cur_ts));
5880 elog(LOG,
5881 "stats_timestamp %s is later than collector's time %s for database %u",
5882 writetime, mytime, dbentry->databaseid);
5883 pfree(writetime);
5884 pfree(mytime);
5885 }
5886 else
5887 {
5888 /*
5889 * Nope, it's just an old request. Assuming msg's clock_time is
5890 * >= its cutoff_time, it must be stale, so we can ignore it.
5891 */
5892 return;
5893 }
5894 }
5895 else if (msg->cutoff_time <= dbentry->stats_timestamp)
5896 {
5897 /* Stale request, ignore it */
5898 return;
5899 }
5900
5901 /*
5902 * We need to write this DB, so create a request.
5903 */
5904 pending_write_requests = lappend_oid(pending_write_requests,
5905 msg->databaseid);
5906 }
5907
5908
5909 /* ----------
5910 * pgstat_recv_tabstat() -
5911 *
5912 * Count what the backend has done.
5913 * ----------
5914 */
5915 static void
pgstat_recv_tabstat(PgStat_MsgTabstat * msg,int len)5916 pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len)
5917 {
5918 PgStat_StatDBEntry *dbentry;
5919 PgStat_StatTabEntry *tabentry;
5920 int i;
5921 bool found;
5922
5923 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
5924
5925 /*
5926 * Update database-wide stats.
5927 */
5928 dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit);
5929 dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback);
5930 dbentry->n_block_read_time += msg->m_block_read_time;
5931 dbentry->n_block_write_time += msg->m_block_write_time;
5932
5933 /*
5934 * Process all table entries in the message.
5935 */
5936 for (i = 0; i < msg->m_nentries; i++)
5937 {
5938 PgStat_TableEntry *tabmsg = &(msg->m_entry[i]);
5939
5940 tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables,
5941 (void *) &(tabmsg->t_id),
5942 HASH_ENTER, &found);
5943
5944 if (!found)
5945 {
5946 /*
5947 * If it's a new table entry, initialize counters to the values we
5948 * just got.
5949 */
5950 tabentry->numscans = tabmsg->t_counts.t_numscans;
5951 tabentry->tuples_returned = tabmsg->t_counts.t_tuples_returned;
5952 tabentry->tuples_fetched = tabmsg->t_counts.t_tuples_fetched;
5953 tabentry->tuples_inserted = tabmsg->t_counts.t_tuples_inserted;
5954 tabentry->tuples_updated = tabmsg->t_counts.t_tuples_updated;
5955 tabentry->tuples_deleted = tabmsg->t_counts.t_tuples_deleted;
5956 tabentry->tuples_hot_updated = tabmsg->t_counts.t_tuples_hot_updated;
5957 tabentry->n_live_tuples = tabmsg->t_counts.t_delta_live_tuples;
5958 tabentry->n_dead_tuples = tabmsg->t_counts.t_delta_dead_tuples;
5959 tabentry->changes_since_analyze = tabmsg->t_counts.t_changed_tuples;
5960 tabentry->blocks_fetched = tabmsg->t_counts.t_blocks_fetched;
5961 tabentry->blocks_hit = tabmsg->t_counts.t_blocks_hit;
5962
5963 tabentry->vacuum_timestamp = 0;
5964 tabentry->vacuum_count = 0;
5965 tabentry->autovac_vacuum_timestamp = 0;
5966 tabentry->autovac_vacuum_count = 0;
5967 tabentry->analyze_timestamp = 0;
5968 tabentry->analyze_count = 0;
5969 tabentry->autovac_analyze_timestamp = 0;
5970 tabentry->autovac_analyze_count = 0;
5971 }
5972 else
5973 {
5974 /*
5975 * Otherwise add the values to the existing entry.
5976 */
5977 tabentry->numscans += tabmsg->t_counts.t_numscans;
5978 tabentry->tuples_returned += tabmsg->t_counts.t_tuples_returned;
5979 tabentry->tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
5980 tabentry->tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
5981 tabentry->tuples_updated += tabmsg->t_counts.t_tuples_updated;
5982 tabentry->tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
5983 tabentry->tuples_hot_updated += tabmsg->t_counts.t_tuples_hot_updated;
5984 /* If table was truncated, first reset the live/dead counters */
5985 if (tabmsg->t_counts.t_truncated)
5986 {
5987 tabentry->n_live_tuples = 0;
5988 tabentry->n_dead_tuples = 0;
5989 }
5990 tabentry->n_live_tuples += tabmsg->t_counts.t_delta_live_tuples;
5991 tabentry->n_dead_tuples += tabmsg->t_counts.t_delta_dead_tuples;
5992 tabentry->changes_since_analyze += tabmsg->t_counts.t_changed_tuples;
5993 tabentry->blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
5994 tabentry->blocks_hit += tabmsg->t_counts.t_blocks_hit;
5995 }
5996
5997 /* Clamp n_live_tuples in case of negative delta_live_tuples */
5998 tabentry->n_live_tuples = Max(tabentry->n_live_tuples, 0);
5999 /* Likewise for n_dead_tuples */
6000 tabentry->n_dead_tuples = Max(tabentry->n_dead_tuples, 0);
6001
6002 /*
6003 * Add per-table stats to the per-database entry, too.
6004 */
6005 dbentry->n_tuples_returned += tabmsg->t_counts.t_tuples_returned;
6006 dbentry->n_tuples_fetched += tabmsg->t_counts.t_tuples_fetched;
6007 dbentry->n_tuples_inserted += tabmsg->t_counts.t_tuples_inserted;
6008 dbentry->n_tuples_updated += tabmsg->t_counts.t_tuples_updated;
6009 dbentry->n_tuples_deleted += tabmsg->t_counts.t_tuples_deleted;
6010 dbentry->n_blocks_fetched += tabmsg->t_counts.t_blocks_fetched;
6011 dbentry->n_blocks_hit += tabmsg->t_counts.t_blocks_hit;
6012 }
6013 }
6014
6015
6016 /* ----------
6017 * pgstat_recv_tabpurge() -
6018 *
6019 * Arrange for dead table removal.
6020 * ----------
6021 */
6022 static void
pgstat_recv_tabpurge(PgStat_MsgTabpurge * msg,int len)6023 pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len)
6024 {
6025 PgStat_StatDBEntry *dbentry;
6026 int i;
6027
6028 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
6029
6030 /*
6031 * No need to purge if we don't even know the database.
6032 */
6033 if (!dbentry || !dbentry->tables)
6034 return;
6035
6036 /*
6037 * Process all table entries in the message.
6038 */
6039 for (i = 0; i < msg->m_nentries; i++)
6040 {
6041 /* Remove from hashtable if present; we don't care if it's not. */
6042 (void) hash_search(dbentry->tables,
6043 (void *) &(msg->m_tableid[i]),
6044 HASH_REMOVE, NULL);
6045 }
6046 }
6047
6048
6049 /* ----------
6050 * pgstat_recv_dropdb() -
6051 *
6052 * Arrange for dead database removal
6053 * ----------
6054 */
6055 static void
pgstat_recv_dropdb(PgStat_MsgDropdb * msg,int len)6056 pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len)
6057 {
6058 Oid dbid = msg->m_databaseid;
6059 PgStat_StatDBEntry *dbentry;
6060
6061 /*
6062 * Lookup the database in the hashtable.
6063 */
6064 dbentry = pgstat_get_db_entry(dbid, false);
6065
6066 /*
6067 * If found, remove it (along with the db statfile).
6068 */
6069 if (dbentry)
6070 {
6071 char statfile[MAXPGPATH];
6072
6073 get_dbstat_filename(false, false, dbid, statfile, MAXPGPATH);
6074
6075 elog(DEBUG2, "removing stats file \"%s\"", statfile);
6076 unlink(statfile);
6077
6078 if (dbentry->tables != NULL)
6079 hash_destroy(dbentry->tables);
6080 if (dbentry->functions != NULL)
6081 hash_destroy(dbentry->functions);
6082
6083 if (hash_search(pgStatDBHash,
6084 (void *) &dbid,
6085 HASH_REMOVE, NULL) == NULL)
6086 ereport(ERROR,
6087 (errmsg("database hash table corrupted during cleanup --- abort")));
6088 }
6089 }
6090
6091
6092 /* ----------
6093 * pgstat_recv_resetcounter() -
6094 *
6095 * Reset the statistics for the specified database.
6096 * ----------
6097 */
6098 static void
pgstat_recv_resetcounter(PgStat_MsgResetcounter * msg,int len)6099 pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len)
6100 {
6101 PgStat_StatDBEntry *dbentry;
6102
6103 /*
6104 * Lookup the database in the hashtable. Nothing to do if not there.
6105 */
6106 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
6107
6108 if (!dbentry)
6109 return;
6110
6111 /*
6112 * We simply throw away all the database's table entries by recreating a
6113 * new hash table for them.
6114 */
6115 if (dbentry->tables != NULL)
6116 hash_destroy(dbentry->tables);
6117 if (dbentry->functions != NULL)
6118 hash_destroy(dbentry->functions);
6119
6120 dbentry->tables = NULL;
6121 dbentry->functions = NULL;
6122
6123 /*
6124 * Reset database-level stats, too. This creates empty hash tables for
6125 * tables and functions.
6126 */
6127 reset_dbentry_counters(dbentry);
6128 }
6129
6130 /* ----------
6131 * pgstat_recv_resetshared() -
6132 *
6133 * Reset some shared statistics of the cluster.
6134 * ----------
6135 */
6136 static void
pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter * msg,int len)6137 pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
6138 {
6139 if (msg->m_resettarget == RESET_BGWRITER)
6140 {
6141 /* Reset the global background writer statistics for the cluster. */
6142 memset(&globalStats, 0, sizeof(globalStats));
6143 globalStats.stat_reset_timestamp = GetCurrentTimestamp();
6144 }
6145 else if (msg->m_resettarget == RESET_ARCHIVER)
6146 {
6147 /* Reset the archiver statistics for the cluster. */
6148 memset(&archiverStats, 0, sizeof(archiverStats));
6149 archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
6150 }
6151
6152 /*
6153 * Presumably the sender of this message validated the target, don't
6154 * complain here if it's not valid
6155 */
6156 }
6157
6158 /* ----------
6159 * pgstat_recv_resetsinglecounter() -
6160 *
6161 * Reset a statistics for a single object
6162 * ----------
6163 */
6164 static void
pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter * msg,int len)6165 pgstat_recv_resetsinglecounter(PgStat_MsgResetsinglecounter *msg, int len)
6166 {
6167 PgStat_StatDBEntry *dbentry;
6168
6169 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
6170
6171 if (!dbentry)
6172 return;
6173
6174 /* Set the reset timestamp for the whole database */
6175 dbentry->stat_reset_timestamp = GetCurrentTimestamp();
6176
6177 /* Remove object if it exists, ignore it if not */
6178 if (msg->m_resettype == RESET_TABLE)
6179 (void) hash_search(dbentry->tables, (void *) &(msg->m_objectid),
6180 HASH_REMOVE, NULL);
6181 else if (msg->m_resettype == RESET_FUNCTION)
6182 (void) hash_search(dbentry->functions, (void *) &(msg->m_objectid),
6183 HASH_REMOVE, NULL);
6184 }
6185
6186 /* ----------
6187 * pgstat_recv_autovac() -
6188 *
6189 * Process an autovacuum signalling message.
6190 * ----------
6191 */
6192 static void
pgstat_recv_autovac(PgStat_MsgAutovacStart * msg,int len)6193 pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len)
6194 {
6195 PgStat_StatDBEntry *dbentry;
6196
6197 /*
6198 * Store the last autovacuum time in the database's hashtable entry.
6199 */
6200 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6201
6202 dbentry->last_autovac_time = msg->m_start_time;
6203 }
6204
6205 /* ----------
6206 * pgstat_recv_vacuum() -
6207 *
6208 * Process a VACUUM message.
6209 * ----------
6210 */
6211 static void
pgstat_recv_vacuum(PgStat_MsgVacuum * msg,int len)6212 pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
6213 {
6214 PgStat_StatDBEntry *dbentry;
6215 PgStat_StatTabEntry *tabentry;
6216
6217 /*
6218 * Store the data in the table's hashtable entry.
6219 */
6220 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6221
6222 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
6223
6224 tabentry->n_live_tuples = msg->m_live_tuples;
6225 tabentry->n_dead_tuples = msg->m_dead_tuples;
6226
6227 if (msg->m_autovacuum)
6228 {
6229 tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
6230 tabentry->autovac_vacuum_count++;
6231 }
6232 else
6233 {
6234 tabentry->vacuum_timestamp = msg->m_vacuumtime;
6235 tabentry->vacuum_count++;
6236 }
6237 }
6238
6239 /* ----------
6240 * pgstat_recv_analyze() -
6241 *
6242 * Process an ANALYZE message.
6243 * ----------
6244 */
6245 static void
pgstat_recv_analyze(PgStat_MsgAnalyze * msg,int len)6246 pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len)
6247 {
6248 PgStat_StatDBEntry *dbentry;
6249 PgStat_StatTabEntry *tabentry;
6250
6251 /*
6252 * Store the data in the table's hashtable entry.
6253 */
6254 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6255
6256 tabentry = pgstat_get_tab_entry(dbentry, msg->m_tableoid, true);
6257
6258 tabentry->n_live_tuples = msg->m_live_tuples;
6259 tabentry->n_dead_tuples = msg->m_dead_tuples;
6260
6261 /*
6262 * If commanded, reset changes_since_analyze to zero. This forgets any
6263 * changes that were committed while the ANALYZE was in progress, but we
6264 * have no good way to estimate how many of those there were.
6265 */
6266 if (msg->m_resetcounter)
6267 tabentry->changes_since_analyze = 0;
6268
6269 if (msg->m_autovacuum)
6270 {
6271 tabentry->autovac_analyze_timestamp = msg->m_analyzetime;
6272 tabentry->autovac_analyze_count++;
6273 }
6274 else
6275 {
6276 tabentry->analyze_timestamp = msg->m_analyzetime;
6277 tabentry->analyze_count++;
6278 }
6279 }
6280
6281
6282 /* ----------
6283 * pgstat_recv_archiver() -
6284 *
6285 * Process a ARCHIVER message.
6286 * ----------
6287 */
6288 static void
pgstat_recv_archiver(PgStat_MsgArchiver * msg,int len)6289 pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len)
6290 {
6291 if (msg->m_failed)
6292 {
6293 /* Failed archival attempt */
6294 ++archiverStats.failed_count;
6295 memcpy(archiverStats.last_failed_wal, msg->m_xlog,
6296 sizeof(archiverStats.last_failed_wal));
6297 archiverStats.last_failed_timestamp = msg->m_timestamp;
6298 }
6299 else
6300 {
6301 /* Successful archival operation */
6302 ++archiverStats.archived_count;
6303 memcpy(archiverStats.last_archived_wal, msg->m_xlog,
6304 sizeof(archiverStats.last_archived_wal));
6305 archiverStats.last_archived_timestamp = msg->m_timestamp;
6306 }
6307 }
6308
6309 /* ----------
6310 * pgstat_recv_bgwriter() -
6311 *
6312 * Process a BGWRITER message.
6313 * ----------
6314 */
6315 static void
pgstat_recv_bgwriter(PgStat_MsgBgWriter * msg,int len)6316 pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
6317 {
6318 globalStats.timed_checkpoints += msg->m_timed_checkpoints;
6319 globalStats.requested_checkpoints += msg->m_requested_checkpoints;
6320 globalStats.checkpoint_write_time += msg->m_checkpoint_write_time;
6321 globalStats.checkpoint_sync_time += msg->m_checkpoint_sync_time;
6322 globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints;
6323 globalStats.buf_written_clean += msg->m_buf_written_clean;
6324 globalStats.maxwritten_clean += msg->m_maxwritten_clean;
6325 globalStats.buf_written_backend += msg->m_buf_written_backend;
6326 globalStats.buf_fsync_backend += msg->m_buf_fsync_backend;
6327 globalStats.buf_alloc += msg->m_buf_alloc;
6328 }
6329
6330 /* ----------
6331 * pgstat_recv_recoveryconflict() -
6332 *
6333 * Process a RECOVERYCONFLICT message.
6334 * ----------
6335 */
6336 static void
pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict * msg,int len)6337 pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len)
6338 {
6339 PgStat_StatDBEntry *dbentry;
6340
6341 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6342
6343 switch (msg->m_reason)
6344 {
6345 case PROCSIG_RECOVERY_CONFLICT_DATABASE:
6346
6347 /*
6348 * Since we drop the information about the database as soon as it
6349 * replicates, there is no point in counting these conflicts.
6350 */
6351 break;
6352 case PROCSIG_RECOVERY_CONFLICT_TABLESPACE:
6353 dbentry->n_conflict_tablespace++;
6354 break;
6355 case PROCSIG_RECOVERY_CONFLICT_LOCK:
6356 dbentry->n_conflict_lock++;
6357 break;
6358 case PROCSIG_RECOVERY_CONFLICT_SNAPSHOT:
6359 dbentry->n_conflict_snapshot++;
6360 break;
6361 case PROCSIG_RECOVERY_CONFLICT_BUFFERPIN:
6362 dbentry->n_conflict_bufferpin++;
6363 break;
6364 case PROCSIG_RECOVERY_CONFLICT_STARTUP_DEADLOCK:
6365 dbentry->n_conflict_startup_deadlock++;
6366 break;
6367 }
6368 }
6369
6370 /* ----------
6371 * pgstat_recv_deadlock() -
6372 *
6373 * Process a DEADLOCK message.
6374 * ----------
6375 */
6376 static void
pgstat_recv_deadlock(PgStat_MsgDeadlock * msg,int len)6377 pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len)
6378 {
6379 PgStat_StatDBEntry *dbentry;
6380
6381 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6382
6383 dbentry->n_deadlocks++;
6384 }
6385
6386 /* ----------
6387 * pgstat_recv_checksum_failure() -
6388 *
6389 * Process a CHECKSUMFAILURE message.
6390 * ----------
6391 */
6392 static void
pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure * msg,int len)6393 pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len)
6394 {
6395 PgStat_StatDBEntry *dbentry;
6396
6397 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6398
6399 dbentry->n_checksum_failures += msg->m_failurecount;
6400 dbentry->last_checksum_failure = msg->m_failure_time;
6401 }
6402
6403 /* ----------
6404 * pgstat_recv_tempfile() -
6405 *
6406 * Process a TEMPFILE message.
6407 * ----------
6408 */
6409 static void
pgstat_recv_tempfile(PgStat_MsgTempFile * msg,int len)6410 pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len)
6411 {
6412 PgStat_StatDBEntry *dbentry;
6413
6414 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6415
6416 dbentry->n_temp_bytes += msg->m_filesize;
6417 dbentry->n_temp_files += 1;
6418 }
6419
6420 /* ----------
6421 * pgstat_recv_funcstat() -
6422 *
6423 * Count what the backend has done.
6424 * ----------
6425 */
6426 static void
pgstat_recv_funcstat(PgStat_MsgFuncstat * msg,int len)6427 pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len)
6428 {
6429 PgStat_FunctionEntry *funcmsg = &(msg->m_entry[0]);
6430 PgStat_StatDBEntry *dbentry;
6431 PgStat_StatFuncEntry *funcentry;
6432 int i;
6433 bool found;
6434
6435 dbentry = pgstat_get_db_entry(msg->m_databaseid, true);
6436
6437 /*
6438 * Process all function entries in the message.
6439 */
6440 for (i = 0; i < msg->m_nentries; i++, funcmsg++)
6441 {
6442 funcentry = (PgStat_StatFuncEntry *) hash_search(dbentry->functions,
6443 (void *) &(funcmsg->f_id),
6444 HASH_ENTER, &found);
6445
6446 if (!found)
6447 {
6448 /*
6449 * If it's a new function entry, initialize counters to the values
6450 * we just got.
6451 */
6452 funcentry->f_numcalls = funcmsg->f_numcalls;
6453 funcentry->f_total_time = funcmsg->f_total_time;
6454 funcentry->f_self_time = funcmsg->f_self_time;
6455 }
6456 else
6457 {
6458 /*
6459 * Otherwise add the values to the existing entry.
6460 */
6461 funcentry->f_numcalls += funcmsg->f_numcalls;
6462 funcentry->f_total_time += funcmsg->f_total_time;
6463 funcentry->f_self_time += funcmsg->f_self_time;
6464 }
6465 }
6466 }
6467
6468 /* ----------
6469 * pgstat_recv_funcpurge() -
6470 *
6471 * Arrange for dead function removal.
6472 * ----------
6473 */
6474 static void
pgstat_recv_funcpurge(PgStat_MsgFuncpurge * msg,int len)6475 pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len)
6476 {
6477 PgStat_StatDBEntry *dbentry;
6478 int i;
6479
6480 dbentry = pgstat_get_db_entry(msg->m_databaseid, false);
6481
6482 /*
6483 * No need to purge if we don't even know the database.
6484 */
6485 if (!dbentry || !dbentry->functions)
6486 return;
6487
6488 /*
6489 * Process all function entries in the message.
6490 */
6491 for (i = 0; i < msg->m_nentries; i++)
6492 {
6493 /* Remove from hashtable if present; we don't care if it's not. */
6494 (void) hash_search(dbentry->functions,
6495 (void *) &(msg->m_functionid[i]),
6496 HASH_REMOVE, NULL);
6497 }
6498 }
6499
6500 /* ----------
6501 * pgstat_write_statsfile_needed() -
6502 *
6503 * Do we need to write out any stats files?
6504 * ----------
6505 */
6506 static bool
pgstat_write_statsfile_needed(void)6507 pgstat_write_statsfile_needed(void)
6508 {
6509 if (pending_write_requests != NIL)
6510 return true;
6511
6512 /* Everything was written recently */
6513 return false;
6514 }
6515
6516 /* ----------
6517 * pgstat_db_requested() -
6518 *
6519 * Checks whether stats for a particular DB need to be written to a file.
6520 * ----------
6521 */
6522 static bool
pgstat_db_requested(Oid databaseid)6523 pgstat_db_requested(Oid databaseid)
6524 {
6525 /*
6526 * If any requests are outstanding at all, we should write the stats for
6527 * shared catalogs (the "database" with OID 0). This ensures that
6528 * backends will see up-to-date stats for shared catalogs, even though
6529 * they send inquiry messages mentioning only their own DB.
6530 */
6531 if (databaseid == InvalidOid && pending_write_requests != NIL)
6532 return true;
6533
6534 /* Search to see if there's an open request to write this database. */
6535 if (list_member_oid(pending_write_requests, databaseid))
6536 return true;
6537
6538 return false;
6539 }
6540
6541 /*
6542 * Convert a potentially unsafely truncated activity string (see
6543 * PgBackendStatus.st_activity_raw's documentation) into a correctly truncated
6544 * one.
6545 *
6546 * The returned string is allocated in the caller's memory context and may be
6547 * freed.
6548 */
6549 char *
pgstat_clip_activity(const char * raw_activity)6550 pgstat_clip_activity(const char *raw_activity)
6551 {
6552 char *activity;
6553 int rawlen;
6554 int cliplen;
6555
6556 /*
6557 * Some callers, like pgstat_get_backend_current_activity(), do not
6558 * guarantee that the buffer isn't concurrently modified. We try to take
6559 * care that the buffer is always terminated by a NUL byte regardless, but
6560 * let's still be paranoid about the string's length. In those cases the
6561 * underlying buffer is guaranteed to be pgstat_track_activity_query_size
6562 * large.
6563 */
6564 activity = pnstrdup(raw_activity, pgstat_track_activity_query_size - 1);
6565
6566 /* now double-guaranteed to be NUL terminated */
6567 rawlen = strlen(activity);
6568
6569 /*
6570 * All supported server-encodings make it possible to determine the length
6571 * of a multi-byte character from its first byte (this is not the case for
6572 * client encodings, see GB18030). As st_activity is always stored using
6573 * server encoding, this allows us to perform multi-byte aware truncation,
6574 * even if the string earlier was truncated in the middle of a multi-byte
6575 * character.
6576 */
6577 cliplen = pg_mbcliplen(activity, rawlen,
6578 pgstat_track_activity_query_size - 1);
6579
6580 activity[cliplen] = '\0';
6581
6582 return activity;
6583 }
6584