| 1 |
/* ------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* pgstat_database.c |
| 4 |
* Implementation of database statistics. |
| 5 |
* |
| 6 |
* This file contains the implementation of database statistics. It is kept |
| 7 |
* separate from pgstat.c to enforce the line between the statistics access / |
| 8 |
* storage implementation and the details about individual types of |
| 9 |
* statistics. |
| 10 |
* |
| 11 |
* Copyright (c) 2001-2025, PostgreSQL Global Development Group |
| 12 |
* |
| 13 |
* IDENTIFICATION |
| 14 |
* src/backend/utils/activity/pgstat_database.c |
| 15 |
* ------------------------------------------------------------------------- |
| 16 |
*/ |
| 17 |
|
| 18 |
#include "postgres.h" |
| 19 |
|
| 20 |
#include "storage/procsignal.h" |
| 21 |
#include "utils/pgstat_internal.h" |
| 22 |
#include "utils/timestamp.h" |
| 23 |
|
| 24 |
|
| 25 |
static bool pgstat_should_report_connstat(void); |
| 26 |
|
| 27 |
|
| 28 |
PgStat_Counter pgStatBlockReadTime = 0; |
| 29 |
PgStat_Counter pgStatBlockWriteTime = 0; |
| 30 |
PgStat_Counter pgStatActiveTime = 0; |
| 31 |
PgStat_Counter pgStatTransactionIdleTime = 0; |
| 32 |
SessionEndType pgStatSessionEndCause = DISCONNECT_NORMAL; |
| 33 |
|
| 34 |
|
| 35 |
static int pgStatXactCommit = 0; |
| 36 |
static int pgStatXactRollback = 0; |
| 37 |
static PgStat_Counter pgLastSessionReportTime = 0; |
| 38 |
|
| 39 |
|
| 40 |
/* |
| 41 |
* Remove entry for the database being dropped. |
| 42 |
*/ |
| 43 |
void |
| 44 |
pgstat_drop_database(Oid databaseid) |
| 45 |
{ |
| 46 |
pgstat_drop_transactional(PGSTAT_KIND_DATABASE, databaseid, InvalidOid); |
| 47 |
} |
| 48 |
|
| 49 |
/* |
| 50 |
* Called from autovacuum.c to report startup of an autovacuum process. |
| 51 |
* We are called before InitPostgres is done, so can't rely on MyDatabaseId; |
| 52 |
* the db OID must be passed in, instead. |
| 53 |
*/ |
| 54 |
void |
| 55 |
pgstat_report_autovac(Oid dboid) |
| 56 |
{ |
| 57 |
PgStat_EntryRef *entry_ref; |
| 58 |
PgStatShared_Database *dbentry; |
| 59 |
|
| 60 |
/* can't get here in single user mode */ |
| 61 |
Assert(IsUnderPostmaster); |
| 62 |
|
| 63 |
/* |
| 64 |
* End-of-vacuum is reported instantly. Report the start the same way for |
| 65 |
* consistency. Vacuum doesn't run frequently and is a long-lasting |
| 66 |
* operation so it doesn't matter if we get blocked here a little. |
| 67 |
*/ |
| 68 |
entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, |
| 69 |
dboid, InvalidOid, false); |
| 70 |
|
| 71 |
dbentry = (PgStatShared_Database *) entry_ref->shared_stats; |
| 72 |
dbentry->stats.last_autovac_time = GetCurrentTimestamp(); |
| 73 |
|
| 74 |
pgstat_unlock_entry(entry_ref); |
| 75 |
} |
| 76 |
|
| 77 |
/* |
| 78 |
* Report a Hot Standby recovery conflict. |
| 79 |
*/ |
| 80 |
void |
| 81 |
pgstat_report_recovery_conflict(int reason) |
| 82 |
{ |
| 83 |
PgStat_StatDBEntry *dbentry; |
| •84 |
PgStat_EntryRef *entry_ref; |
| 85 |
PgStatShared_Database *sharedent; |
| 86 |
|
| 87 |
Assert(IsUnderPostmaster); |
| 88 |
if (!pgstat_track_counts) |
| 89 |
return; |
| 90 |
|
| •91 |
/* |
| 92 |
* Update the shared stats directly - recovery conflicts should never be |
| 93 |
* common enough for that to be a problem. |
| 94 |
*/ |
| 95 |
entry_ref = |
| 96 |
pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, MyDatabaseId, InvalidOid, false); |
| 97 |
|
| 98 |
sharedent = (PgStatShared_Database *) entry_ref->shared_stats; |
| 99 |
dbentry = &sharedent->stats; |
| 100 |
|
| 101 |
switch (reason) |
| 102 |
{ |
| 103 |
case PROCSIG_RECOVERY_CONFLICT_DATABASE: |
| 104 |
|
| 105 |
/* |
| 106 |
* Since we drop the information about the database as soon as it |
| 107 |
* replicates, there is no point in counting these conflicts. |
| 108 |
*/ |
| 109 |
break; |
| 110 |
case PROCSIG_RECOVERY_CONFLICT_TABLESPACE: |
| 111 |
dbentry->conflict_tablespace++; |
| 112 |
break; |
| 113 |
case PROCSIG_RECOVERY_CONFLICT_LOCK: |
| 114 |
dbentry->conflict_lock++; |
| 115 |
break; |
| 116 |
case PROCSIG_RECOVERY_CONFLICT_SNAPSHOT: |
| 117 |
dbentry->conflict_snapshot++; |
| 118 |
break; |
| 119 |
case PROCSIG_RECOVERY_CONFLICT_BUFFERPIN: |
| 120 |
dbentry->conflict_bufferpin++; |
| 121 |
break; |
| 122 |
case PROCSIG_RECOVERY_CONFLICT_LOGICALSLOT: |
| 123 |
dbentry->conflict_logicalslot++; |
| 124 |
break; |
| 125 |
case PROCSIG_RECOVERY_CONFLICT_STARTUP_DEADLOCK: |
| 126 |
dbentry->conflict_startup_deadlock++; |
| 127 |
break; |
| 128 |
} |
| •129 |
|
| 130 |
pgstat_unlock_entry(entry_ref); |
| 131 |
} |
| 132 |
|
| 133 |
/* |
| 134 |
* Report a detected deadlock. |
| 135 |
*/ |
| 136 |
void |
| 137 |
pgstat_report_deadlock(void) |
| 138 |
{ |
| 139 |
PgStat_StatDBEntry *dbent; |
| 140 |
|
| 141 |
if (!pgstat_track_counts) |
| 142 |
return; |
| 143 |
|
| 144 |
dbent = pgstat_prep_database_pending(MyDatabaseId); |
| 145 |
dbent->deadlocks++; |
| 146 |
} |
| 147 |
|
| 148 |
/* |
| 149 |
* Allow this backend to later report checksum failures for dboid, even if in |
| 150 |
* a critical section at the time of the report. |
| 151 |
* |
| 152 |
* Without this function having been called first, the backend might need to |
| 153 |
* allocate an EntryRef or might need to map in DSM segments. Neither should |
| 154 |
* happen in a critical section. |
| 155 |
*/ |
| 156 |
void |
| 157 |
pgstat_prepare_report_checksum_failure(Oid dboid) |
| 158 |
{ |
| 159 |
Assert(!CritSectionCount); |
| 160 |
|
| 161 |
/* |
| 162 |
* Just need to ensure this backend has an entry ref for the database. |
| 163 |
* That will allows us to report checksum failures without e.g. needing to |
| 164 |
* map in DSM segments. |
| 165 |
*/ |
| 166 |
pgstat_get_entry_ref(PGSTAT_KIND_DATABASE, dboid, InvalidOid, |
| 167 |
true, NULL); |
| 168 |
} |
| 169 |
|
| 170 |
/* |
| 171 |
* Report one or more checksum failures. |
| 172 |
* |
| 173 |
* To be allowed to report checksum failures in critical sections, we require |
| 174 |
* pgstat_prepare_report_checksum_failure() to have been called before this |
| 175 |
* function is called. |
| 176 |
*/ |
| 177 |
void |
| 178 |
pgstat_report_checksum_failures_in_db(Oid dboid, int failurecount) |
| 179 |
{ |
| 180 |
PgStat_EntryRef *entry_ref; |
| 181 |
PgStatShared_Database *sharedent; |
| 182 |
|
| 183 |
if (!pgstat_track_counts) |
| 184 |
return; |
| 185 |
|
| 186 |
/* |
| 187 |
* Update the shared stats directly - checksum failures should never be |
| 188 |
* common enough for that to be a problem. Note that we pass create=false |
| 189 |
* here, as we want to be sure to not require memory allocations, so this |
| 190 |
* can be called in critical sections. |
| 191 |
*/ |
| 192 |
entry_ref = pgstat_get_entry_ref(PGSTAT_KIND_DATABASE, dboid, InvalidOid, |
| 193 |
false, NULL); |
| 194 |
|
| 195 |
/* |
| 196 |
* Should always have been created by |
| 197 |
* pgstat_prepare_report_checksum_failure(). |
| 198 |
* |
| 199 |
* When not using assertions, we don't want to crash should something have |
| 200 |
* gone wrong, so just return. |
| 201 |
*/ |
| 202 |
Assert(entry_ref); |
| 203 |
if (!entry_ref) |
| 204 |
{ |
| 205 |
elog(WARNING, "could not report %d checksum failures for database %u", |
| 206 |
failurecount, dboid); |
| 207 |
return; |
| 208 |
} |
| 209 |
|
| 210 |
(void) pgstat_lock_entry(entry_ref, false); |
| 211 |
|
| 212 |
sharedent = (PgStatShared_Database *) entry_ref->shared_stats; |
| 213 |
sharedent->stats.checksum_failures += failurecount; |
| 214 |
sharedent->stats.last_checksum_failure = GetCurrentTimestamp(); |
| 215 |
|
| 216 |
pgstat_unlock_entry(entry_ref); |
| 217 |
} |
| 218 |
|
| 219 |
/* |
| 220 |
* Report creation of temporary file. |
| 221 |
*/ |
| 222 |
void |
| 223 |
pgstat_report_tempfile(size_t filesize) |
| 224 |
{ |
| 225 |
PgStat_StatDBEntry *dbent; |
| 226 |
|
| 227 |
if (!pgstat_track_counts) |
| 228 |
return; |
| 229 |
|
| 230 |
dbent = pgstat_prep_database_pending(MyDatabaseId); |
| 231 |
dbent->temp_bytes += filesize; |
| 232 |
dbent->temp_files++; |
| 233 |
} |
| 234 |
|
| 235 |
/* |
| 236 |
* Notify stats system of a new connection. |
| 237 |
*/ |
| 238 |
void |
| 239 |
pgstat_report_connect(Oid dboid) |
| 240 |
{ |
| 241 |
PgStat_StatDBEntry *dbentry; |
| 242 |
|
| 243 |
if (!pgstat_should_report_connstat()) |
| 244 |
return; |
| 245 |
|
| 246 |
pgLastSessionReportTime = MyStartTimestamp; |
| 247 |
|
| 248 |
dbentry = pgstat_prep_database_pending(dboid); |
| 249 |
dbentry->sessions++; |
| 250 |
} |
| 251 |
|
| 252 |
/* |
| 253 |
* Notify the stats system of a disconnect. |
| 254 |
*/ |
| 255 |
void |
| 256 |
pgstat_report_disconnect(Oid dboid) |
| 257 |
{ |
| 258 |
PgStat_StatDBEntry *dbentry; |
| 259 |
|
| 260 |
if (!pgstat_should_report_connstat()) |
| 261 |
return; |
| 262 |
|
| 263 |
dbentry = pgstat_prep_database_pending(dboid); |
| 264 |
|
| 265 |
switch (pgStatSessionEndCause) |
| 266 |
{ |
| 267 |
case DISCONNECT_NOT_YET: |
| 268 |
case DISCONNECT_NORMAL: |
| 269 |
/* we don't collect these */ |
| 270 |
break; |
| 271 |
case DISCONNECT_CLIENT_EOF: |
| 272 |
dbentry->sessions_abandoned++; |
| 273 |
break; |
| 274 |
case DISCONNECT_FATAL: |
| 275 |
dbentry->sessions_fatal++; |
| 276 |
break; |
| 277 |
case DISCONNECT_KILLED: |
| 278 |
dbentry->sessions_killed++; |
| 279 |
break; |
| 280 |
} |
| 281 |
} |
| 282 |
|
| 283 |
/* |
| 284 |
* Support function for the SQL-callable pgstat* functions. Returns |
| 285 |
* the collected statistics for one database or NULL. NULL doesn't mean |
| 286 |
* that the database doesn't exist, just that there are no statistics, so the |
| 287 |
* caller is better off to report ZERO instead. |
| 288 |
*/ |
| 289 |
PgStat_StatDBEntry * |
| 290 |
pgstat_fetch_stat_dbentry(Oid dboid) |
| 291 |
{ |
| 292 |
return (PgStat_StatDBEntry *) |
| 293 |
pgstat_fetch_entry(PGSTAT_KIND_DATABASE, dboid, InvalidOid); |
| 294 |
} |
| 295 |
|
| 296 |
void |
| 297 |
AtEOXact_PgStat_Database(bool isCommit, bool parallel) |
| 298 |
{ |
| 299 |
/* Don't count parallel worker transaction stats */ |
| 300 |
if (!parallel) |
| 301 |
{ |
| 302 |
/* |
| 303 |
* Count transaction commit or abort. (We use counters, not just |
| 304 |
* bools, in case the reporting message isn't sent right away.) |
| 305 |
*/ |
| 306 |
if (isCommit) |
| 307 |
pgStatXactCommit++; |
| 308 |
else |
| 309 |
pgStatXactRollback++; |
| 310 |
} |
| 311 |
} |
| 312 |
|
| 313 |
/* |
| 314 |
* Notify the stats system about parallel worker information. |
| 315 |
*/ |
| 316 |
void |
| 317 |
pgstat_update_parallel_workers_stats(PgStat_Counter workers_to_launch, |
| 318 |
PgStat_Counter workers_launched) |
| 319 |
{ |
| 320 |
PgStat_StatDBEntry *dbentry; |
| 321 |
|
| 322 |
if (!OidIsValid(MyDatabaseId)) |
| 323 |
return; |
| 324 |
|
| 325 |
dbentry = pgstat_prep_database_pending(MyDatabaseId); |
| 326 |
dbentry->parallel_workers_to_launch += workers_to_launch; |
| 327 |
dbentry->parallel_workers_launched += workers_launched; |
| 328 |
} |
| 329 |
|
| 330 |
/* |
| 331 |
* Subroutine for pgstat_report_stat(): Handle xact commit/rollback and I/O |
| 332 |
* timings. |
| 333 |
*/ |
| 334 |
void |
| 335 |
pgstat_update_dbstats(TimestampTz ts) |
| 336 |
{ |
| 337 |
PgStat_StatDBEntry *dbentry; |
| 338 |
|
| 339 |
/* |
| 340 |
* If not connected to a database yet, don't attribute time to "shared |
| 341 |
* state" (InvalidOid is used to track stats for shared relations, etc.). |
| 342 |
*/ |
| 343 |
if (!OidIsValid(MyDatabaseId)) |
| 344 |
return; |
| 345 |
|
| 346 |
dbentry = pgstat_prep_database_pending(MyDatabaseId); |
| 347 |
|
| 348 |
/* |
| 349 |
* Accumulate xact commit/rollback and I/O timings to stats entry of the |
| 350 |
* current database. |
| 351 |
*/ |
| 352 |
dbentry->xact_commit += pgStatXactCommit; |
| 353 |
dbentry->xact_rollback += pgStatXactRollback; |
| 354 |
dbentry->blk_read_time += pgStatBlockReadTime; |
| 355 |
dbentry->blk_write_time += pgStatBlockWriteTime; |
| 356 |
|
| 357 |
if (pgstat_should_report_connstat()) |
| 358 |
{ |
| 359 |
long secs; |
| 360 |
int usecs; |
| 361 |
|
| 362 |
/* |
| 363 |
* pgLastSessionReportTime is initialized to MyStartTimestamp by |
| 364 |
* pgstat_report_connect(). |
| 365 |
*/ |
| 366 |
TimestampDifference(pgLastSessionReportTime, ts, &secs, &usecs); |
| 367 |
pgLastSessionReportTime = ts; |
| 368 |
dbentry->session_time += (PgStat_Counter) secs * 1000000 + usecs; |
| 369 |
dbentry->active_time += pgStatActiveTime; |
| 370 |
dbentry->idle_in_transaction_time += pgStatTransactionIdleTime; |
| 371 |
} |
| 372 |
|
| 373 |
pgStatXactCommit = 0; |
| 374 |
pgStatXactRollback = 0; |
| 375 |
pgStatBlockReadTime = 0; |
| 376 |
pgStatBlockWriteTime = 0; |
| 377 |
pgStatActiveTime = 0; |
| 378 |
pgStatTransactionIdleTime = 0; |
| 379 |
} |
| 380 |
|
| 381 |
/* |
| 382 |
* We report session statistics only for normal backend processes. Parallel |
| 383 |
* workers run in parallel, so they don't contribute to session times, even |
| 384 |
* though they use CPU time. Walsender processes could be considered here, |
| 385 |
* but they have different session characteristics from normal backends (for |
| 386 |
* example, they are always "active"), so they would skew session statistics. |
| 387 |
*/ |
| 388 |
static bool |
| 389 |
pgstat_should_report_connstat(void) |
| 390 |
{ |
| 391 |
return MyBackendType == B_BACKEND; |
| 392 |
} |
| 393 |
|
| 394 |
/* |
| 395 |
* Find or create a local PgStat_StatDBEntry entry for dboid. |
| 396 |
*/ |
| 397 |
PgStat_StatDBEntry * |
| 398 |
pgstat_prep_database_pending(Oid dboid) |
| 399 |
{ |
| 400 |
PgStat_EntryRef *entry_ref; |
| 401 |
|
| 402 |
/* |
| 403 |
* This should not report stats on database objects before having |
| 404 |
* connected to a database. |
| 405 |
*/ |
| 406 |
Assert(!OidIsValid(dboid) || OidIsValid(MyDatabaseId)); |
| 407 |
|
| 408 |
entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_DATABASE, dboid, InvalidOid, |
| 409 |
NULL); |
| 410 |
|
| 411 |
return entry_ref->pending; |
| 412 |
} |
| 413 |
|
| 414 |
/* |
| 415 |
* Reset the database's reset timestamp, without resetting the contents of the |
| 416 |
* database stats. |
| 417 |
*/ |
| 418 |
void |
| 419 |
pgstat_reset_database_timestamp(Oid dboid, TimestampTz ts) |
| 420 |
{ |
| 421 |
PgStat_EntryRef *dbref; |
| 422 |
PgStatShared_Database *dbentry; |
| 423 |
|
| 424 |
dbref = pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, dboid, InvalidOid, |
| 425 |
false); |
| 426 |
|
| 427 |
dbentry = (PgStatShared_Database *) dbref->shared_stats; |
| 428 |
dbentry->stats.stat_reset_timestamp = ts; |
| 429 |
|
| 430 |
pgstat_unlock_entry(dbref); |
| 431 |
} |
| 432 |
|
| 433 |
/* |
| 434 |
* Flush out pending stats for the entry |
| 435 |
* |
| 436 |
* If nowait is true and the lock could not be immediately acquired, returns |
| 437 |
* false without flushing the entry. Otherwise returns true. |
| 438 |
*/ |
| 439 |
bool |
| 440 |
pgstat_database_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) |
| 441 |
{ |
| 442 |
PgStatShared_Database *sharedent; |
| 443 |
PgStat_StatDBEntry *pendingent; |
| 444 |
|
| 445 |
pendingent = (PgStat_StatDBEntry *) entry_ref->pending; |
| 446 |
sharedent = (PgStatShared_Database *) entry_ref->shared_stats; |
| 447 |
|
| 448 |
if (!pgstat_lock_entry(entry_ref, nowait)) |
| 449 |
return false; |
| 450 |
|
| 451 |
#define PGSTAT_ACCUM_DBCOUNT(item) \ |
| 452 |
(sharedent)->stats.item += (pendingent)->item |
| 453 |
|
| 454 |
PGSTAT_ACCUM_DBCOUNT(xact_commit); |
| 455 |
PGSTAT_ACCUM_DBCOUNT(xact_rollback); |
| 456 |
PGSTAT_ACCUM_DBCOUNT(blocks_fetched); |
| 457 |
PGSTAT_ACCUM_DBCOUNT(blocks_hit); |
| 458 |
|
| 459 |
PGSTAT_ACCUM_DBCOUNT(tuples_returned); |
| 460 |
PGSTAT_ACCUM_DBCOUNT(tuples_fetched); |
| 461 |
PGSTAT_ACCUM_DBCOUNT(tuples_inserted); |
| 462 |
PGSTAT_ACCUM_DBCOUNT(tuples_updated); |
| 463 |
PGSTAT_ACCUM_DBCOUNT(tuples_deleted); |
| 464 |
|
| 465 |
/* last_autovac_time is reported immediately */ |
| 466 |
Assert(pendingent->last_autovac_time == 0); |
| 467 |
|
| 468 |
PGSTAT_ACCUM_DBCOUNT(conflict_tablespace); |
| 469 |
PGSTAT_ACCUM_DBCOUNT(conflict_lock); |
| 470 |
PGSTAT_ACCUM_DBCOUNT(conflict_snapshot); |
| 471 |
PGSTAT_ACCUM_DBCOUNT(conflict_logicalslot); |
| 472 |
PGSTAT_ACCUM_DBCOUNT(conflict_bufferpin); |
| 473 |
PGSTAT_ACCUM_DBCOUNT(conflict_startup_deadlock); |
| 474 |
|
| 475 |
PGSTAT_ACCUM_DBCOUNT(temp_bytes); |
| 476 |
PGSTAT_ACCUM_DBCOUNT(temp_files); |
| 477 |
PGSTAT_ACCUM_DBCOUNT(deadlocks); |
| 478 |
|
| 479 |
/* checksum failures are reported immediately */ |
| 480 |
Assert(pendingent->checksum_failures == 0); |
| 481 |
Assert(pendingent->last_checksum_failure == 0); |
| 482 |
|
| 483 |
PGSTAT_ACCUM_DBCOUNT(blk_read_time); |
| 484 |
PGSTAT_ACCUM_DBCOUNT(blk_write_time); |
| 485 |
|
| 486 |
PGSTAT_ACCUM_DBCOUNT(sessions); |
| 487 |
PGSTAT_ACCUM_DBCOUNT(session_time); |
| 488 |
PGSTAT_ACCUM_DBCOUNT(active_time); |
| 489 |
PGSTAT_ACCUM_DBCOUNT(idle_in_transaction_time); |
| 490 |
PGSTAT_ACCUM_DBCOUNT(sessions_abandoned); |
| 491 |
PGSTAT_ACCUM_DBCOUNT(sessions_fatal); |
| 492 |
PGSTAT_ACCUM_DBCOUNT(sessions_killed); |
| 493 |
PGSTAT_ACCUM_DBCOUNT(parallel_workers_to_launch); |
| 494 |
PGSTAT_ACCUM_DBCOUNT(parallel_workers_launched); |
| 495 |
#undef PGSTAT_ACCUM_DBCOUNT |
| 496 |
|
| 497 |
pgstat_unlock_entry(entry_ref); |
| 498 |
|
| 499 |
memset(pendingent, 0, sizeof(*pendingent)); |
| 500 |
|
| 501 |
return true; |
| 502 |
} |
| 503 |
|
| 504 |
void |
| 505 |
pgstat_database_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts) |
| 506 |
{ |
| 507 |
((PgStatShared_Database *) header)->stats.stat_reset_timestamp = ts; |
| 508 |
} |