| 1 |
/* ---------- |
| 2 |
* pgstat.c |
| 3 |
* Infrastructure for the cumulative statistics system. |
| 4 |
* |
| 5 |
* The cumulative statistics system accumulates statistics for different kinds |
| 6 |
* of objects. Some kinds of statistics are collected for a fixed number of |
| 7 |
* objects (most commonly 1), e.g., checkpointer statistics. Other kinds of |
| 8 |
* statistics are collected for a varying number of objects |
| 9 |
* (e.g. relations). See PgStat_KindInfo for a list of currently handled |
| 10 |
* statistics. |
| 11 |
* |
| 12 |
* Statistics are loaded from the filesystem during startup (by the startup |
| 13 |
* process), unless preceded by a crash, in which case all stats are |
| 14 |
* discarded. They are written out by the checkpointer process just before |
| 15 |
* shutting down (if the stats kind allows it), except when shutting down in |
| 16 |
* immediate mode. |
| 17 |
* |
| 18 |
* Fixed-numbered stats are stored in plain (non-dynamic) shared memory. |
| 19 |
* |
| 20 |
* Statistics for variable-numbered objects are stored in dynamic shared |
| 21 |
* memory and can be found via a dshash hashtable. The statistics counters are |
| 22 |
* not part of the dshash entry (PgStatShared_HashEntry) directly, but are |
| 23 |
* separately allocated (PgStatShared_HashEntry->body). The separate |
| 24 |
* allocation allows different kinds of statistics to be stored in the same |
| 25 |
* hashtable without wasting space in PgStatShared_HashEntry. |
| 26 |
* |
| 27 |
* Variable-numbered stats are addressed by PgStat_HashKey while running. It |
| 28 |
* is not possible to have statistics for an object that cannot be addressed |
| 29 |
* that way at runtime. A wider identifier can be used when serializing to |
| 30 |
* disk (used for replication slot stats). |
| 31 |
* |
| 32 |
* To avoid contention on the shared hashtable, each backend has a |
| 33 |
* backend-local hashtable (pgStatEntryRefHash) in front of the shared |
| 34 |
* hashtable, containing references (PgStat_EntryRef) to shared hashtable |
| 35 |
* entries. The shared hashtable only needs to be accessed when no prior |
| 36 |
* reference is found in the local hashtable. Besides pointing to the |
| 37 |
* shared hashtable entry (PgStatShared_HashEntry) PgStat_EntryRef also |
| 38 |
* contains a pointer to the shared statistics data, as a process-local |
| 39 |
* address, to reduce access costs. |
| 40 |
* |
| 41 |
* The names for structs stored in shared memory are prefixed with |
| 42 |
* PgStatShared instead of PgStat. Each stats entry in shared memory is |
| 43 |
* protected by a dedicated lwlock. |
| 44 |
* |
| 45 |
* Most stats updates are first accumulated locally in each process as pending |
| 46 |
* entries, then later flushed to shared memory (just after commit, or by |
| 47 |
* idle-timeout). This practically eliminates contention on individual stats |
| 48 |
* entries. For most kinds of variable-numbered pending stats data is stored |
| 49 |
* in PgStat_EntryRef->pending. All entries with pending data are in the |
| 50 |
* pgStatPending list. Pending statistics updates are flushed out by |
| 51 |
* pgstat_report_stat(). |
| 52 |
* |
| 53 |
* It is possible for external modules to define custom statistics kinds, |
| 54 |
* that can use the same properties as any built-in stats kinds. Each custom |
| 55 |
* stats kind needs to assign a unique ID to ensure that it does not overlap |
| 56 |
* with other extensions. In order to reserve a unique stats kind ID, refer |
| 57 |
* to https://wiki.postgresql.org/wiki/CustomCumulativeStats. |
| 58 |
* |
| 59 |
* The behavior of different kinds of statistics is determined by the kind's |
| 60 |
* entry in pgstat_kind_builtin_infos for all the built-in statistics kinds |
| 61 |
* defined, and pgstat_kind_custom_infos for custom kinds registered at |
| 62 |
* startup by pgstat_register_kind(). See PgStat_KindInfo for details. |
| 63 |
* |
| 64 |
* The consistency of read accesses to statistics can be configured using the |
| 65 |
* stats_fetch_consistency GUC (see config.sgml and monitoring.sgml for the |
| 66 |
* settings). When using PGSTAT_FETCH_CONSISTENCY_CACHE or |
| 67 |
* PGSTAT_FETCH_CONSISTENCY_SNAPSHOT statistics are stored in |
| 68 |
* pgStatLocal.snapshot. |
| 69 |
* |
| 70 |
* To keep things manageable, stats handling is split across several |
| 71 |
* files. Infrastructure pieces are in: |
| 72 |
* - pgstat.c - this file, to tie it all together |
| 73 |
* - pgstat_shmem.c - nearly everything dealing with shared memory, including |
| 74 |
* the maintenance of hashtable entries |
| 75 |
* - pgstat_xact.c - transactional integration, including the transactional |
| 76 |
* creation and dropping of stats entries |
| 77 |
* |
| 78 |
* Each statistics kind is handled in a dedicated file: |
| 79 |
* - pgstat_archiver.c |
| 80 |
* - pgstat_backend.c |
| 81 |
* - pgstat_bgwriter.c |
| 82 |
* - pgstat_checkpointer.c |
| 83 |
* - pgstat_database.c |
| 84 |
* - pgstat_function.c |
| 85 |
* - pgstat_io.c |
| 86 |
* - pgstat_relation.c |
| 87 |
* - pgstat_replslot.c |
| 88 |
* - pgstat_slru.c |
| 89 |
* - pgstat_subscription.c |
| 90 |
* - pgstat_wal.c |
| 91 |
* |
| 92 |
* Whenever possible infrastructure files should not contain code related to |
| 93 |
* specific kinds of stats. |
| 94 |
* |
| 95 |
* |
| 96 |
* Copyright (c) 2001-2025, PostgreSQL Global Development Group |
| 97 |
* |
| 98 |
* IDENTIFICATION |
| 99 |
* src/backend/utils/activity/pgstat.c |
| 100 |
* ---------- |
| 101 |
*/ |
| 102 |
#include "postgres.h" |
| 103 |
|
| 104 |
#include <unistd.h> |
| 105 |
|
| 106 |
#include "access/xact.h" |
| 107 |
#include "lib/dshash.h" |
| 108 |
#include "pgstat.h" |
| 109 |
#include "storage/fd.h" |
| 110 |
#include "storage/ipc.h" |
| 111 |
#include "storage/lwlock.h" |
| 112 |
#include "utils/guc_hooks.h" |
| 113 |
#include "utils/memutils.h" |
| 114 |
#include "utils/pgstat_internal.h" |
| 115 |
#include "utils/timestamp.h" |
| 116 |
|
| 117 |
|
| 118 |
/* ---------- |
| 119 |
* Timer definitions. |
| 120 |
* |
| 121 |
* In milliseconds. |
| 122 |
* ---------- |
| 123 |
*/ |
| 124 |
|
| 125 |
/* minimum interval non-forced stats flushes.*/ |
| 126 |
#define PGSTAT_MIN_INTERVAL 1000 |
| 127 |
/* how long until to block flushing pending stats updates */ |
| 128 |
#define PGSTAT_MAX_INTERVAL 60000 |
| 129 |
/* when to call pgstat_report_stat() again, even when idle */ |
| 130 |
#define PGSTAT_IDLE_INTERVAL 10000 |
| 131 |
|
| 132 |
/* ---------- |
| 133 |
* Initial size hints for the hash tables used in statistics. |
| 134 |
* ---------- |
| 135 |
*/ |
| 136 |
|
| 137 |
#define PGSTAT_SNAPSHOT_HASH_SIZE 512 |
| 138 |
|
| 139 |
/* --------- |
| 140 |
* Identifiers in stats file. |
| 141 |
* --------- |
| 142 |
*/ |
| 143 |
#define PGSTAT_FILE_ENTRY_END 'E' /* end of file */ |
| 144 |
#define PGSTAT_FILE_ENTRY_FIXED 'F' /* fixed-numbered stats entry */ |
| 145 |
#define PGSTAT_FILE_ENTRY_NAME 'N' /* stats entry identified by name */ |
| 146 |
#define PGSTAT_FILE_ENTRY_HASH 'S' /* stats entry identified by |
| 147 |
* PgStat_HashKey */ |
| 148 |
|
| 149 |
/* hash table for statistics snapshots entry */ |
| 150 |
typedef struct PgStat_SnapshotEntry |
| 151 |
{ |
| 152 |
PgStat_HashKey key; |
| 153 |
char status; /* for simplehash use */ |
| 154 |
void *data; /* the stats data itself */ |
| 155 |
} PgStat_SnapshotEntry; |
| 156 |
|
| 157 |
|
| 158 |
/* ---------- |
| 159 |
* Backend-local Hash Table Definitions |
| 160 |
* ---------- |
| 161 |
*/ |
| 162 |
|
| 163 |
/* for stats snapshot entries */ |
| 164 |
#define SH_PREFIX pgstat_snapshot |
| 165 |
#define SH_ELEMENT_TYPE PgStat_SnapshotEntry |
| 166 |
#define SH_KEY_TYPE PgStat_HashKey |
| 167 |
#define SH_KEY key |
| 168 |
#define SH_HASH_KEY(tb, key) \ |
| 169 |
pgstat_hash_hash_key(&key, sizeof(PgStat_HashKey), NULL) |
| 170 |
#define SH_EQUAL(tb, a, b) \ |
| 171 |
pgstat_cmp_hash_key(&a, &b, sizeof(PgStat_HashKey), NULL) == 0 |
| 172 |
#define SH_SCOPE static inline |
| 173 |
#define SH_DEFINE |
| 174 |
#define SH_DECLARE |
| 175 |
#include "lib/simplehash.h" |
| 176 |
|
| 177 |
|
| 178 |
/* ---------- |
| 179 |
* Local function forward declarations |
| 180 |
* ---------- |
| 181 |
*/ |
| 182 |
|
| 183 |
static void pgstat_write_statsfile(void); |
| 184 |
static void pgstat_read_statsfile(void); |
| 185 |
|
| 186 |
static void pgstat_init_snapshot_fixed(void); |
| 187 |
|
| 188 |
static void pgstat_reset_after_failure(void); |
| 189 |
|
| 190 |
static bool pgstat_flush_pending_entries(bool nowait); |
| 191 |
|
| 192 |
static void pgstat_prep_snapshot(void); |
| 193 |
static void pgstat_build_snapshot(void); |
| 194 |
static void pgstat_build_snapshot_fixed(PgStat_Kind kind); |
| 195 |
|
| 196 |
static inline bool pgstat_is_kind_valid(PgStat_Kind kind); |
| 197 |
|
| 198 |
|
| 199 |
/* ---------- |
| 200 |
* GUC parameters |
| 201 |
* ---------- |
| 202 |
*/ |
| 203 |
|
| 204 |
bool pgstat_track_counts = false; |
| 205 |
int pgstat_fetch_consistency = PGSTAT_FETCH_CONSISTENCY_CACHE; |
| 206 |
|
| 207 |
|
| 208 |
/* ---------- |
| 209 |
* state shared with pgstat_*.c |
| 210 |
* ---------- |
| 211 |
*/ |
| 212 |
|
| 213 |
PgStat_LocalState pgStatLocal; |
| 214 |
|
| 215 |
/* |
| 216 |
* Track pending reports for fixed-numbered stats, used by |
| 217 |
* pgstat_report_stat(). |
| 218 |
*/ |
| 219 |
bool pgstat_report_fixed = false; |
| 220 |
|
| 221 |
/* ---------- |
| 222 |
* Local data |
| 223 |
* |
| 224 |
* NB: There should be only variables related to stats infrastructure here, |
| 225 |
* not for specific kinds of stats. |
| 226 |
* ---------- |
| 227 |
*/ |
| 228 |
|
| 229 |
/* |
| 230 |
* Memory contexts containing the pgStatEntryRefHash table, the |
| 231 |
* pgStatSharedRef entries, and pending data respectively. Mostly to make it |
| 232 |
* easier to track / attribute memory usage. |
| 233 |
*/ |
| 234 |
|
| 235 |
static MemoryContext pgStatPendingContext = NULL; |
| 236 |
|
| 237 |
/* |
| 238 |
* Backend local list of PgStat_EntryRef with unflushed pending stats. |
| 239 |
* |
| 240 |
* Newly pending entries should only ever be added to the end of the list, |
| 241 |
* otherwise pgstat_flush_pending_entries() might not see them immediately. |
| 242 |
*/ |
| 243 |
static dlist_head pgStatPending = DLIST_STATIC_INIT(pgStatPending); |
| 244 |
|
| 245 |
|
| 246 |
/* |
| 247 |
* Force the next stats flush to happen regardless of |
| 248 |
* PGSTAT_MIN_INTERVAL. Useful in test scripts. |
| 249 |
*/ |
| 250 |
static bool pgStatForceNextFlush = false; |
| 251 |
|
| 252 |
/* |
| 253 |
* Force-clear existing snapshot before next use when stats_fetch_consistency |
| 254 |
* is changed. |
| 255 |
*/ |
| 256 |
static bool force_stats_snapshot_clear = false; |
| 257 |
|
| 258 |
|
| 259 |
/* |
| 260 |
* For assertions that check pgstat is not used before initialization / after |
| 261 |
* shutdown. |
| 262 |
*/ |
| 263 |
#ifdef USE_ASSERT_CHECKING |
| 264 |
static bool pgstat_is_initialized = false; |
| 265 |
static bool pgstat_is_shutdown = false; |
| 266 |
#endif |
| 267 |
|
| 268 |
|
| 269 |
/* |
| 270 |
* The different kinds of built-in statistics. |
| 271 |
* |
| 272 |
* If reasonably possible, handling specific to one kind of stats should go |
| 273 |
* through this abstraction, rather than making more of pgstat.c aware. |
| 274 |
* |
| 275 |
* See comments for struct PgStat_KindInfo for details about the individual |
| 276 |
* fields. |
| 277 |
* |
| 278 |
* XXX: It'd be nicer to define this outside of this file. But there doesn't |
| 279 |
* seem to be a great way of doing that, given the split across multiple |
| 280 |
* files. |
| 281 |
*/ |
| 282 |
static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] = { |
| 283 |
|
| 284 |
/* stats kinds for variable-numbered objects */ |
| 285 |
|
| 286 |
[PGSTAT_KIND_DATABASE] = { |
| 287 |
.name = "database", |
| 288 |
|
| 289 |
.fixed_amount = false, |
| 290 |
.write_to_file = true, |
| 291 |
/* so pg_stat_database entries can be seen in all databases */ |
| 292 |
.accessed_across_databases = true, |
| 293 |
|
| 294 |
.shared_size = sizeof(PgStatShared_Database), |
| 295 |
.shared_data_off = offsetof(PgStatShared_Database, stats), |
| 296 |
.shared_data_len = sizeof(((PgStatShared_Database *) 0)->stats), |
| 297 |
.pending_size = sizeof(PgStat_StatDBEntry), |
| 298 |
|
| 299 |
.flush_pending_cb = pgstat_database_flush_cb, |
| 300 |
.reset_timestamp_cb = pgstat_database_reset_timestamp_cb, |
| 301 |
}, |
| 302 |
|
| 303 |
[PGSTAT_KIND_RELATION] = { |
| 304 |
.name = "relation", |
| 305 |
|
| 306 |
.fixed_amount = false, |
| 307 |
.write_to_file = true, |
| 308 |
|
| 309 |
.shared_size = sizeof(PgStatShared_Relation), |
| 310 |
.shared_data_off = offsetof(PgStatShared_Relation, stats), |
| 311 |
.shared_data_len = sizeof(((PgStatShared_Relation *) 0)->stats), |
| 312 |
.pending_size = sizeof(PgStat_TableStatus), |
| 313 |
|
| 314 |
.flush_pending_cb = pgstat_relation_flush_cb, |
| 315 |
.delete_pending_cb = pgstat_relation_delete_pending_cb, |
| 316 |
}, |
| 317 |
|
| 318 |
[PGSTAT_KIND_FUNCTION] = { |
| 319 |
.name = "function", |
| 320 |
|
| 321 |
.fixed_amount = false, |
| 322 |
.write_to_file = true, |
| 323 |
|
| 324 |
.shared_size = sizeof(PgStatShared_Function), |
| 325 |
.shared_data_off = offsetof(PgStatShared_Function, stats), |
| 326 |
.shared_data_len = sizeof(((PgStatShared_Function *) 0)->stats), |
| 327 |
.pending_size = sizeof(PgStat_FunctionCounts), |
| 328 |
|
| 329 |
.flush_pending_cb = pgstat_function_flush_cb, |
| 330 |
}, |
| 331 |
|
| 332 |
[PGSTAT_KIND_REPLSLOT] = { |
| 333 |
.name = "replslot", |
| 334 |
|
| 335 |
.fixed_amount = false, |
| 336 |
.write_to_file = true, |
| 337 |
|
| 338 |
.accessed_across_databases = true, |
| 339 |
|
| 340 |
.shared_size = sizeof(PgStatShared_ReplSlot), |
| 341 |
.shared_data_off = offsetof(PgStatShared_ReplSlot, stats), |
| 342 |
.shared_data_len = sizeof(((PgStatShared_ReplSlot *) 0)->stats), |
| 343 |
|
| 344 |
.reset_timestamp_cb = pgstat_replslot_reset_timestamp_cb, |
| 345 |
.to_serialized_name = pgstat_replslot_to_serialized_name_cb, |
| 346 |
.from_serialized_name = pgstat_replslot_from_serialized_name_cb, |
| 347 |
}, |
| 348 |
|
| 349 |
[PGSTAT_KIND_SUBSCRIPTION] = { |
| 350 |
.name = "subscription", |
| 351 |
|
| 352 |
.fixed_amount = false, |
| 353 |
.write_to_file = true, |
| 354 |
/* so pg_stat_subscription_stats entries can be seen in all databases */ |
| 355 |
.accessed_across_databases = true, |
| 356 |
|
| 357 |
.shared_size = sizeof(PgStatShared_Subscription), |
| 358 |
.shared_data_off = offsetof(PgStatShared_Subscription, stats), |
| 359 |
.shared_data_len = sizeof(((PgStatShared_Subscription *) 0)->stats), |
| 360 |
.pending_size = sizeof(PgStat_BackendSubEntry), |
| 361 |
|
| 362 |
.flush_pending_cb = pgstat_subscription_flush_cb, |
| 363 |
.reset_timestamp_cb = pgstat_subscription_reset_timestamp_cb, |
| 364 |
}, |
| 365 |
|
| 366 |
[PGSTAT_KIND_BACKEND] = { |
| 367 |
.name = "backend", |
| 368 |
|
| 369 |
.fixed_amount = false, |
| 370 |
.write_to_file = false, |
| 371 |
|
| 372 |
.accessed_across_databases = true, |
| 373 |
|
| 374 |
.shared_size = sizeof(PgStatShared_Backend), |
| 375 |
.shared_data_off = offsetof(PgStatShared_Backend, stats), |
| 376 |
.shared_data_len = sizeof(((PgStatShared_Backend *) 0)->stats), |
| 377 |
|
| 378 |
.flush_static_cb = pgstat_backend_flush_cb, |
| 379 |
.reset_timestamp_cb = pgstat_backend_reset_timestamp_cb, |
| 380 |
}, |
| 381 |
|
| 382 |
/* stats for fixed-numbered (mostly 1) objects */ |
| 383 |
|
| 384 |
[PGSTAT_KIND_ARCHIVER] = { |
| 385 |
.name = "archiver", |
| 386 |
|
| 387 |
.fixed_amount = true, |
| 388 |
.write_to_file = true, |
| 389 |
|
| 390 |
.snapshot_ctl_off = offsetof(PgStat_Snapshot, archiver), |
| 391 |
.shared_ctl_off = offsetof(PgStat_ShmemControl, archiver), |
| 392 |
.shared_data_off = offsetof(PgStatShared_Archiver, stats), |
| 393 |
.shared_data_len = sizeof(((PgStatShared_Archiver *) 0)->stats), |
| 394 |
|
| 395 |
.init_shmem_cb = pgstat_archiver_init_shmem_cb, |
| 396 |
.reset_all_cb = pgstat_archiver_reset_all_cb, |
| 397 |
.snapshot_cb = pgstat_archiver_snapshot_cb, |
| 398 |
}, |
| 399 |
|
| 400 |
[PGSTAT_KIND_BGWRITER] = { |
| 401 |
.name = "bgwriter", |
| 402 |
|
| 403 |
.fixed_amount = true, |
| 404 |
.write_to_file = true, |
| 405 |
|
| 406 |
.snapshot_ctl_off = offsetof(PgStat_Snapshot, bgwriter), |
| 407 |
.shared_ctl_off = offsetof(PgStat_ShmemControl, bgwriter), |
| 408 |
.shared_data_off = offsetof(PgStatShared_BgWriter, stats), |
| 409 |
.shared_data_len = sizeof(((PgStatShared_BgWriter *) 0)->stats), |
| 410 |
|
| 411 |
.init_shmem_cb = pgstat_bgwriter_init_shmem_cb, |
| 412 |
.reset_all_cb = pgstat_bgwriter_reset_all_cb, |
| 413 |
.snapshot_cb = pgstat_bgwriter_snapshot_cb, |
| 414 |
}, |
| 415 |
|
| 416 |
[PGSTAT_KIND_CHECKPOINTER] = { |
| 417 |
.name = "checkpointer", |
| 418 |
|
| 419 |
.fixed_amount = true, |
| 420 |
.write_to_file = true, |
| 421 |
|
| 422 |
.snapshot_ctl_off = offsetof(PgStat_Snapshot, checkpointer), |
| 423 |
.shared_ctl_off = offsetof(PgStat_ShmemControl, checkpointer), |
| 424 |
.shared_data_off = offsetof(PgStatShared_Checkpointer, stats), |
| 425 |
.shared_data_len = sizeof(((PgStatShared_Checkpointer *) 0)->stats), |
| 426 |
|
| 427 |
.init_shmem_cb = pgstat_checkpointer_init_shmem_cb, |
| 428 |
.reset_all_cb = pgstat_checkpointer_reset_all_cb, |
| 429 |
.snapshot_cb = pgstat_checkpointer_snapshot_cb, |
| 430 |
}, |
| 431 |
|
| 432 |
[PGSTAT_KIND_IO] = { |
| 433 |
.name = "io", |
| 434 |
|
| 435 |
.fixed_amount = true, |
| 436 |
.write_to_file = true, |
| 437 |
|
| 438 |
.snapshot_ctl_off = offsetof(PgStat_Snapshot, io), |
| 439 |
.shared_ctl_off = offsetof(PgStat_ShmemControl, io), |
| 440 |
.shared_data_off = offsetof(PgStatShared_IO, stats), |
| 441 |
.shared_data_len = sizeof(((PgStatShared_IO *) 0)->stats), |
| 442 |
|
| 443 |
.flush_static_cb = pgstat_io_flush_cb, |
| 444 |
.init_shmem_cb = pgstat_io_init_shmem_cb, |
| 445 |
.reset_all_cb = pgstat_io_reset_all_cb, |
| 446 |
.snapshot_cb = pgstat_io_snapshot_cb, |
| 447 |
}, |
| 448 |
|
| 449 |
[PGSTAT_KIND_SLRU] = { |
| 450 |
.name = "slru", |
| 451 |
|
| 452 |
.fixed_amount = true, |
| 453 |
.write_to_file = true, |
| 454 |
|
| 455 |
.snapshot_ctl_off = offsetof(PgStat_Snapshot, slru), |
| 456 |
.shared_ctl_off = offsetof(PgStat_ShmemControl, slru), |
| 457 |
.shared_data_off = offsetof(PgStatShared_SLRU, stats), |
| 458 |
.shared_data_len = sizeof(((PgStatShared_SLRU *) 0)->stats), |
| 459 |
|
| 460 |
.flush_static_cb = pgstat_slru_flush_cb, |
| 461 |
.init_shmem_cb = pgstat_slru_init_shmem_cb, |
| 462 |
.reset_all_cb = pgstat_slru_reset_all_cb, |
| 463 |
.snapshot_cb = pgstat_slru_snapshot_cb, |
| 464 |
}, |
| 465 |
|
| 466 |
[PGSTAT_KIND_WAL] = { |
| 467 |
.name = "wal", |
| 468 |
|
| 469 |
.fixed_amount = true, |
| 470 |
.write_to_file = true, |
| 471 |
|
| 472 |
.snapshot_ctl_off = offsetof(PgStat_Snapshot, wal), |
| 473 |
.shared_ctl_off = offsetof(PgStat_ShmemControl, wal), |
| 474 |
.shared_data_off = offsetof(PgStatShared_Wal, stats), |
| 475 |
.shared_data_len = sizeof(((PgStatShared_Wal *) 0)->stats), |
| 476 |
|
| 477 |
.init_backend_cb = pgstat_wal_init_backend_cb, |
| 478 |
.flush_static_cb = pgstat_wal_flush_cb, |
| 479 |
.init_shmem_cb = pgstat_wal_init_shmem_cb, |
| 480 |
.reset_all_cb = pgstat_wal_reset_all_cb, |
| 481 |
.snapshot_cb = pgstat_wal_snapshot_cb, |
| 482 |
}, |
| 483 |
}; |
| 484 |
|
| 485 |
/* |
| 486 |
* Information about custom statistics kinds. |
| 487 |
* |
| 488 |
* These are saved in a different array than the built-in kinds to save |
| 489 |
* in clarity with the initializations. |
| 490 |
* |
| 491 |
* Indexed by PGSTAT_KIND_CUSTOM_MIN, of size PGSTAT_KIND_CUSTOM_SIZE. |
| 492 |
*/ |
| 493 |
static const PgStat_KindInfo **pgstat_kind_custom_infos = NULL; |
| 494 |
|
| 495 |
/* ------------------------------------------------------------ |
| 496 |
* Functions managing the state of the stats system for all backends. |
| 497 |
* ------------------------------------------------------------ |
| 498 |
*/ |
| 499 |
|
| 500 |
/* |
| 501 |
* Read on-disk stats into memory at server start. |
| 502 |
* |
| 503 |
* Should only be called by the startup process or in single user mode. |
| 504 |
*/ |
| 505 |
void |
| 506 |
pgstat_restore_stats(void) |
| 507 |
{ |
| 508 |
pgstat_read_statsfile(); |
| 509 |
} |
| 510 |
|
| 511 |
/* |
| 512 |
* Remove the stats file. This is currently used only if WAL recovery is |
| 513 |
* needed after a crash. |
| 514 |
* |
| 515 |
* Should only be called by the startup process or in single user mode. |
| 516 |
*/ |
| 517 |
void |
| 518 |
pgstat_discard_stats(void) |
| 519 |
{ |
| 520 |
int ret; |
| 521 |
|
| 522 |
/* NB: this needs to be done even in single user mode */ |
| 523 |
|
| 524 |
ret = unlink(PGSTAT_STAT_PERMANENT_FILENAME); |
| 525 |
if (ret != 0) |
| 526 |
{ |
| 527 |
if (errno == ENOENT) |
| 528 |
elog(DEBUG2, |
| 529 |
"didn't need to unlink permanent stats file \"%s\" - didn't exist", |
| 530 |
PGSTAT_STAT_PERMANENT_FILENAME); |
| 531 |
else |
| 532 |
ereport(LOG, |
| 533 |
(errcode_for_file_access(), |
| 534 |
errmsg("could not unlink permanent statistics file \"%s\": %m", |
| 535 |
PGSTAT_STAT_PERMANENT_FILENAME))); |
| 536 |
} |
| 537 |
else |
| 538 |
{ |
| 539 |
ereport(DEBUG2, |
| 540 |
(errcode_for_file_access(), |
| 541 |
errmsg_internal("unlinked permanent statistics file \"%s\"", |
| 542 |
PGSTAT_STAT_PERMANENT_FILENAME))); |
| 543 |
} |
| 544 |
|
| 545 |
/* |
| 546 |
* Reset stats contents. This will set reset timestamps of fixed-numbered |
| 547 |
* stats to the current time (no variable stats exist). |
| 548 |
*/ |
| 549 |
pgstat_reset_after_failure(); |
| 550 |
} |
| 551 |
|
| 552 |
/* |
| 553 |
* pgstat_before_server_shutdown() needs to be called by exactly one process |
| 554 |
* during regular server shutdowns. Otherwise all stats will be lost. |
| 555 |
* |
| 556 |
* We currently only write out stats for proc_exit(0). We might want to change |
| 557 |
* that at some point... But right now pgstat_discard_stats() would be called |
| 558 |
* during the start after a disorderly shutdown, anyway. |
| 559 |
*/ |
| 560 |
void |
| 561 |
pgstat_before_server_shutdown(int code, Datum arg) |
| 562 |
{ |
| 563 |
Assert(pgStatLocal.shmem != NULL); |
| 564 |
Assert(!pgStatLocal.shmem->is_shutdown); |
| 565 |
|
| 566 |
/* |
| 567 |
* Stats should only be reported after pgstat_initialize() and before |
| 568 |
* pgstat_shutdown(). This is a convenient point to catch most violations |
| 569 |
* of this rule. |
| 570 |
*/ |
| 571 |
Assert(pgstat_is_initialized && !pgstat_is_shutdown); |
| 572 |
|
| 573 |
/* flush out our own pending changes before writing out */ |
| 574 |
pgstat_report_stat(true); |
| 575 |
|
| 576 |
/* |
| 577 |
* Only write out file during normal shutdown. Don't even signal that |
| 578 |
* we've shutdown during irregular shutdowns, because the shutdown |
| 579 |
* sequence isn't coordinated to ensure this backend shuts down last. |
| 580 |
*/ |
| 581 |
if (code == 0) |
| 582 |
{ |
| 583 |
pgStatLocal.shmem->is_shutdown = true; |
| 584 |
pgstat_write_statsfile(); |
| 585 |
} |
| 586 |
} |
| 587 |
|
| 588 |
|
| 589 |
/* ------------------------------------------------------------ |
| 590 |
* Backend initialization / shutdown functions |
| 591 |
* ------------------------------------------------------------ |
| 592 |
*/ |
| 593 |
|
| 594 |
/* |
| 595 |
* Shut down a single backend's statistics reporting at process exit. |
| 596 |
* |
| 597 |
* Flush out any remaining statistics counts. Without this, operations |
| 598 |
* triggered during backend exit (such as temp table deletions) won't be |
| 599 |
* counted. |
| 600 |
*/ |
| 601 |
static void |
| 602 |
pgstat_shutdown_hook(int code, Datum arg) |
| 603 |
{ |
| 604 |
Assert(!pgstat_is_shutdown); |
| 605 |
Assert(IsUnderPostmaster || !IsPostmasterEnvironment); |
| 606 |
|
| 607 |
/* |
| 608 |
* If we got as far as discovering our own database ID, we can flush out |
| 609 |
* what we did so far. Otherwise, we'd be reporting an invalid database |
| 610 |
* ID, so forget it. (This means that accesses to pg_database during |
| 611 |
* failed backend starts might never get counted.) |
| 612 |
*/ |
| 613 |
if (OidIsValid(MyDatabaseId)) |
| 614 |
pgstat_report_disconnect(MyDatabaseId); |
| 615 |
|
| 616 |
pgstat_report_stat(true); |
| 617 |
|
| 618 |
/* there shouldn't be any pending changes left */ |
| 619 |
Assert(dlist_is_empty(&pgStatPending)); |
| 620 |
dlist_init(&pgStatPending); |
| 621 |
|
| 622 |
/* drop the backend stats entry */ |
| 623 |
if (!pgstat_drop_entry_ext(PGSTAT_KIND_BACKEND, InvalidOid, MyProcNumber, false)) |
| 624 |
pgstat_request_entry_refs_gc(); |
| 625 |
|
| 626 |
pgstat_detach_shmem(); |
| 627 |
|
| 628 |
#ifdef USE_ASSERT_CHECKING |
| 629 |
pgstat_is_shutdown = true; |
| 630 |
#endif |
| 631 |
} |
| 632 |
|
| 633 |
/* |
| 634 |
* Initialize pgstats state, and set up our on-proc-exit hook. Called from |
| 635 |
* BaseInit(). |
| 636 |
* |
| 637 |
* NOTE: MyDatabaseId isn't set yet; so the shutdown hook has to be careful. |
| 638 |
*/ |
| 639 |
void |
| 640 |
pgstat_initialize(void) |
| 641 |
{ |
| 642 |
Assert(!pgstat_is_initialized); |
| 643 |
|
| 644 |
pgstat_attach_shmem(); |
| 645 |
|
| 646 |
pgstat_init_snapshot_fixed(); |
| 647 |
|
| 648 |
/* Backend initialization callbacks */ |
| 649 |
for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) |
| 650 |
{ |
| 651 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 652 |
|
| 653 |
if (kind_info == NULL || kind_info->init_backend_cb == NULL) |
| 654 |
continue; |
| 655 |
|
| 656 |
kind_info->init_backend_cb(); |
| 657 |
} |
| 658 |
|
| 659 |
/* Set up a process-exit hook to clean up */ |
| 660 |
before_shmem_exit(pgstat_shutdown_hook, 0); |
| 661 |
|
| 662 |
#ifdef USE_ASSERT_CHECKING |
| 663 |
pgstat_is_initialized = true; |
| 664 |
#endif |
| 665 |
} |
| 666 |
|
| 667 |
|
| 668 |
/* ------------------------------------------------------------ |
| 669 |
* Public functions used by backends follow |
| 670 |
* ------------------------------------------------------------ |
| 671 |
*/ |
| 672 |
|
| 673 |
/* |
| 674 |
* Must be called by processes that performs DML: tcop/postgres.c, logical |
| 675 |
* receiver processes, SPI worker, etc. to flush pending statistics updates to |
| 676 |
* shared memory. |
| 677 |
* |
| 678 |
* Unless called with 'force', pending stats updates are flushed happen once |
| 679 |
* per PGSTAT_MIN_INTERVAL (1000ms). When not forced, stats flushes do not |
| 680 |
* block on lock acquisition, except if stats updates have been pending for |
| 681 |
* longer than PGSTAT_MAX_INTERVAL (60000ms). |
| 682 |
* |
| 683 |
* Whenever pending stats updates remain at the end of pgstat_report_stat() a |
| 684 |
* suggested idle timeout is returned. Currently this is always |
| 685 |
* PGSTAT_IDLE_INTERVAL (10000ms). Callers can use the returned time to set up |
| 686 |
* a timeout after which to call pgstat_report_stat(true), but are not |
| 687 |
* required to do so. |
| 688 |
* |
| 689 |
* Note that this is called only when not within a transaction, so it is fair |
| 690 |
* to use transaction stop time as an approximation of current time. |
| 691 |
*/ |
| 692 |
long |
| 693 |
pgstat_report_stat(bool force) |
| 694 |
{ |
| 695 |
static TimestampTz pending_since = 0; |
| 696 |
static TimestampTz last_flush = 0; |
| 697 |
bool partial_flush; |
| 698 |
TimestampTz now; |
| 699 |
bool nowait; |
| 700 |
|
| 701 |
pgstat_assert_is_up(); |
| 702 |
Assert(!IsTransactionOrTransactionBlock()); |
| 703 |
|
| 704 |
/* "absorb" the forced flush even if there's nothing to flush */ |
| 705 |
if (pgStatForceNextFlush) |
| 706 |
{ |
| 707 |
force = true; |
| 708 |
pgStatForceNextFlush = false; |
| 709 |
} |
| 710 |
|
| 711 |
/* Don't expend a clock check if nothing to do */ |
| 712 |
if (dlist_is_empty(&pgStatPending) && |
| 713 |
!pgstat_report_fixed) |
| 714 |
{ |
| 715 |
return 0; |
| 716 |
} |
| 717 |
|
| 718 |
/* |
| 719 |
* There should never be stats to report once stats are shut down. Can't |
| 720 |
* assert that before the checks above, as there is an unconditional |
| 721 |
* pgstat_report_stat() call in pgstat_shutdown_hook() - which at least |
| 722 |
* the process that ran pgstat_before_server_shutdown() will still call. |
| 723 |
*/ |
| 724 |
Assert(!pgStatLocal.shmem->is_shutdown); |
| 725 |
|
| 726 |
if (force) |
| 727 |
{ |
| 728 |
/* |
| 729 |
* Stats reports are forced either when it's been too long since stats |
| 730 |
* have been reported or in processes that force stats reporting to |
| 731 |
* happen at specific points (including shutdown). In the former case |
| 732 |
* the transaction stop time might be quite old, in the latter it |
| 733 |
* would never get cleared. |
| 734 |
*/ |
| 735 |
now = GetCurrentTimestamp(); |
| 736 |
} |
| 737 |
else |
| 738 |
{ |
| 739 |
now = GetCurrentTransactionStopTimestamp(); |
| 740 |
|
| 741 |
if (pending_since > 0 && |
| 742 |
TimestampDifferenceExceeds(pending_since, now, PGSTAT_MAX_INTERVAL)) |
| 743 |
{ |
| 744 |
/* don't keep pending updates longer than PGSTAT_MAX_INTERVAL */ |
| 745 |
force = true; |
| 746 |
} |
| 747 |
else if (last_flush > 0 && |
| 748 |
!TimestampDifferenceExceeds(last_flush, now, PGSTAT_MIN_INTERVAL)) |
| 749 |
{ |
| 750 |
/* don't flush too frequently */ |
| 751 |
if (pending_since == 0) |
| 752 |
pending_since = now; |
| 753 |
|
| 754 |
return PGSTAT_IDLE_INTERVAL; |
| 755 |
} |
| 756 |
} |
| 757 |
|
| 758 |
pgstat_update_dbstats(now); |
| 759 |
|
| 760 |
/* don't wait for lock acquisition when !force */ |
| 761 |
nowait = !force; |
| 762 |
|
| 763 |
partial_flush = false; |
| 764 |
|
| 765 |
/* flush of variable-numbered stats tracked in pending entries list */ |
| 766 |
partial_flush |= pgstat_flush_pending_entries(nowait); |
| 767 |
|
| 768 |
/* flush of other stats kinds */ |
| 769 |
if (pgstat_report_fixed) |
| 770 |
{ |
| 771 |
for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) |
| 772 |
{ |
| 773 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 774 |
|
| 775 |
if (!kind_info) |
| 776 |
continue; |
| 777 |
if (!kind_info->flush_static_cb) |
| 778 |
continue; |
| 779 |
|
| 780 |
partial_flush |= kind_info->flush_static_cb(nowait); |
| 781 |
} |
| 782 |
} |
| 783 |
|
| 784 |
last_flush = now; |
| 785 |
|
| 786 |
/* |
| 787 |
* If some of the pending stats could not be flushed due to lock |
| 788 |
* contention, let the caller know when to retry. |
| 789 |
*/ |
| 790 |
if (partial_flush) |
| 791 |
{ |
| 792 |
/* force should have prevented us from getting here */ |
| 793 |
Assert(!force); |
| 794 |
|
| 795 |
/* remember since when stats have been pending */ |
| 796 |
if (pending_since == 0) |
| 797 |
pending_since = now; |
| 798 |
|
| 799 |
return PGSTAT_IDLE_INTERVAL; |
| 800 |
} |
| 801 |
|
| 802 |
pending_since = 0; |
| 803 |
pgstat_report_fixed = false; |
| 804 |
|
| 805 |
return 0; |
| 806 |
} |
| 807 |
|
| 808 |
/* |
| 809 |
* Force locally pending stats to be flushed during the next |
| 810 |
* pgstat_report_stat() call. This is useful for writing tests. |
| 811 |
*/ |
| 812 |
void |
| 813 |
pgstat_force_next_flush(void) |
| 814 |
{ |
| 815 |
pgStatForceNextFlush = true; |
| 816 |
} |
| 817 |
|
| 818 |
/* |
| 819 |
* Only for use by pgstat_reset_counters() |
| 820 |
*/ |
| 821 |
static bool |
| 822 |
match_db_entries(PgStatShared_HashEntry *entry, Datum match_data) |
| 823 |
{ |
| 824 |
return entry->key.dboid == DatumGetObjectId(MyDatabaseId); |
| 825 |
} |
| 826 |
|
| 827 |
/* |
| 828 |
* Reset counters for our database. |
| 829 |
* |
| 830 |
* Permission checking for this function is managed through the normal |
| 831 |
* GRANT system. |
| 832 |
*/ |
| 833 |
void |
| 834 |
pgstat_reset_counters(void) |
| 835 |
{ |
| 836 |
TimestampTz ts = GetCurrentTimestamp(); |
| 837 |
|
| 838 |
pgstat_reset_matching_entries(match_db_entries, |
| 839 |
ObjectIdGetDatum(MyDatabaseId), |
| 840 |
ts); |
| 841 |
} |
| 842 |
|
| 843 |
/* |
| 844 |
* Reset a single variable-numbered entry. |
| 845 |
* |
| 846 |
* If the stats kind is within a database, also reset the database's |
| 847 |
* stat_reset_timestamp. |
| 848 |
* |
| 849 |
* Permission checking for this function is managed through the normal |
| 850 |
* GRANT system. |
| 851 |
*/ |
| 852 |
void |
| 853 |
pgstat_reset(PgStat_Kind kind, Oid dboid, uint64 objid) |
| 854 |
{ |
| 855 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 856 |
TimestampTz ts = GetCurrentTimestamp(); |
| 857 |
|
| 858 |
/* not needed atm, and doesn't make sense with the current signature */ |
| 859 |
Assert(!pgstat_get_kind_info(kind)->fixed_amount); |
| 860 |
|
| 861 |
/* reset the "single counter" */ |
| 862 |
pgstat_reset_entry(kind, dboid, objid, ts); |
| 863 |
|
| 864 |
if (!kind_info->accessed_across_databases) |
| 865 |
pgstat_reset_database_timestamp(dboid, ts); |
| 866 |
} |
| 867 |
|
| 868 |
/* |
| 869 |
* Reset stats for all entries of a kind. |
| 870 |
* |
| 871 |
* Permission checking for this function is managed through the normal |
| 872 |
* GRANT system. |
| 873 |
*/ |
| 874 |
void |
| 875 |
pgstat_reset_of_kind(PgStat_Kind kind) |
| 876 |
{ |
| 877 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 878 |
TimestampTz ts = GetCurrentTimestamp(); |
| 879 |
|
| 880 |
if (kind_info->fixed_amount) |
| 881 |
kind_info->reset_all_cb(ts); |
| 882 |
else |
| 883 |
pgstat_reset_entries_of_kind(kind, ts); |
| 884 |
} |
| 885 |
|
| 886 |
|
| 887 |
/* ------------------------------------------------------------ |
| 888 |
* Fetching of stats |
| 889 |
* ------------------------------------------------------------ |
| 890 |
*/ |
| 891 |
|
| 892 |
/* |
| 893 |
* Discard any data collected in the current transaction. Any subsequent |
| 894 |
* request will cause new snapshots to be read. |
| 895 |
* |
| 896 |
* This is also invoked during transaction commit or abort to discard |
| 897 |
* the no-longer-wanted snapshot. Updates of stats_fetch_consistency can |
| 898 |
* cause this routine to be called. |
| 899 |
*/ |
| 900 |
void |
| 901 |
pgstat_clear_snapshot(void) |
| 902 |
{ |
| 903 |
pgstat_assert_is_up(); |
| 904 |
|
| 905 |
memset(&pgStatLocal.snapshot.fixed_valid, 0, |
| 906 |
sizeof(pgStatLocal.snapshot.fixed_valid)); |
| 907 |
memset(&pgStatLocal.snapshot.custom_valid, 0, |
| 908 |
sizeof(pgStatLocal.snapshot.custom_valid)); |
| 909 |
pgStatLocal.snapshot.stats = NULL; |
| 910 |
pgStatLocal.snapshot.mode = PGSTAT_FETCH_CONSISTENCY_NONE; |
| 911 |
|
| 912 |
/* Release memory, if any was allocated */ |
| 913 |
if (pgStatLocal.snapshot.context) |
| 914 |
{ |
| 915 |
MemoryContextDelete(pgStatLocal.snapshot.context); |
| 916 |
|
| 917 |
/* Reset variables */ |
| 918 |
pgStatLocal.snapshot.context = NULL; |
| 919 |
} |
| 920 |
|
| 921 |
/* |
| 922 |
* Historically the backend_status.c facilities lived in this file, and |
| 923 |
* were reset with the same function. For now keep it that way, and |
| 924 |
* forward the reset request. |
| 925 |
*/ |
| 926 |
pgstat_clear_backend_activity_snapshot(); |
| 927 |
|
| 928 |
/* Reset this flag, as it may be possible that a cleanup was forced. */ |
| 929 |
force_stats_snapshot_clear = false; |
| 930 |
} |
| 931 |
|
| 932 |
void * |
| 933 |
pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, uint64 objid) |
| 934 |
{ |
| 935 |
PgStat_HashKey key; |
| 936 |
PgStat_EntryRef *entry_ref; |
| 937 |
void *stats_data; |
| 938 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 939 |
|
| 940 |
/* should be called from backends */ |
| 941 |
Assert(IsUnderPostmaster || !IsPostmasterEnvironment); |
| 942 |
Assert(!kind_info->fixed_amount); |
| 943 |
|
| 944 |
pgstat_prep_snapshot(); |
| 945 |
|
| 946 |
/* clear padding */ |
| 947 |
memset(&key, 0, sizeof(struct PgStat_HashKey)); |
| 948 |
|
| 949 |
key.kind = kind; |
| 950 |
key.dboid = dboid; |
| 951 |
key.objid = objid; |
| 952 |
|
| 953 |
/* if we need to build a full snapshot, do so */ |
| 954 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT) |
| 955 |
pgstat_build_snapshot(); |
| 956 |
|
| 957 |
/* if caching is desired, look up in cache */ |
| 958 |
if (pgstat_fetch_consistency > PGSTAT_FETCH_CONSISTENCY_NONE) |
| 959 |
{ |
| 960 |
PgStat_SnapshotEntry *entry = NULL; |
| 961 |
|
| 962 |
entry = pgstat_snapshot_lookup(pgStatLocal.snapshot.stats, key); |
| 963 |
|
| 964 |
if (entry) |
| 965 |
return entry->data; |
| 966 |
|
| 967 |
/* |
| 968 |
* If we built a full snapshot and the key is not in |
| 969 |
* pgStatLocal.snapshot.stats, there are no matching stats. |
| 970 |
*/ |
| 971 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT) |
| 972 |
return NULL; |
| 973 |
} |
| 974 |
|
| 975 |
pgStatLocal.snapshot.mode = pgstat_fetch_consistency; |
| 976 |
|
| 977 |
entry_ref = pgstat_get_entry_ref(kind, dboid, objid, false, NULL); |
| 978 |
|
| 979 |
if (entry_ref == NULL || entry_ref->shared_entry->dropped) |
| 980 |
{ |
| 981 |
/* create empty entry when using PGSTAT_FETCH_CONSISTENCY_CACHE */ |
| 982 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_CACHE) |
| 983 |
{ |
| 984 |
PgStat_SnapshotEntry *entry = NULL; |
| 985 |
bool found; |
| 986 |
|
| 987 |
entry = pgstat_snapshot_insert(pgStatLocal.snapshot.stats, key, &found); |
| 988 |
Assert(!found); |
| 989 |
entry->data = NULL; |
| 990 |
} |
| 991 |
return NULL; |
| 992 |
} |
| 993 |
|
| 994 |
/* |
| 995 |
* Allocate in caller's context for PGSTAT_FETCH_CONSISTENCY_NONE, |
| 996 |
* otherwise we could quickly end up with a fair bit of memory used due to |
| 997 |
* repeated accesses. |
| 998 |
*/ |
| 999 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_NONE) |
| 1000 |
stats_data = palloc(kind_info->shared_data_len); |
| 1001 |
else |
| 1002 |
stats_data = MemoryContextAlloc(pgStatLocal.snapshot.context, |
| 1003 |
kind_info->shared_data_len); |
| 1004 |
|
| 1005 |
(void) pgstat_lock_entry_shared(entry_ref, false); |
| 1006 |
memcpy(stats_data, |
| 1007 |
pgstat_get_entry_data(kind, entry_ref->shared_stats), |
| 1008 |
kind_info->shared_data_len); |
| 1009 |
pgstat_unlock_entry(entry_ref); |
| 1010 |
|
| 1011 |
if (pgstat_fetch_consistency > PGSTAT_FETCH_CONSISTENCY_NONE) |
| 1012 |
{ |
| 1013 |
PgStat_SnapshotEntry *entry = NULL; |
| 1014 |
bool found; |
| 1015 |
|
| 1016 |
entry = pgstat_snapshot_insert(pgStatLocal.snapshot.stats, key, &found); |
| 1017 |
entry->data = stats_data; |
| 1018 |
} |
| 1019 |
|
| 1020 |
return stats_data; |
| 1021 |
} |
| 1022 |
|
| 1023 |
/* |
| 1024 |
* If a stats snapshot has been taken, return the timestamp at which that was |
| 1025 |
* done, and set *have_snapshot to true. Otherwise *have_snapshot is set to |
| 1026 |
* false. |
| 1027 |
*/ |
| 1028 |
TimestampTz |
| 1029 |
pgstat_get_stat_snapshot_timestamp(bool *have_snapshot) |
| 1030 |
{ |
| 1031 |
if (force_stats_snapshot_clear) |
| 1032 |
pgstat_clear_snapshot(); |
| 1033 |
|
| 1034 |
if (pgStatLocal.snapshot.mode == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT) |
| 1035 |
{ |
| 1036 |
*have_snapshot = true; |
| 1037 |
return pgStatLocal.snapshot.snapshot_timestamp; |
| 1038 |
} |
| 1039 |
|
| 1040 |
*have_snapshot = false; |
| 1041 |
|
| 1042 |
return 0; |
| 1043 |
} |
| 1044 |
|
| 1045 |
bool |
| 1046 |
pgstat_have_entry(PgStat_Kind kind, Oid dboid, uint64 objid) |
| 1047 |
{ |
| 1048 |
/* fixed-numbered stats always exist */ |
| 1049 |
if (pgstat_get_kind_info(kind)->fixed_amount) |
| 1050 |
return true; |
| 1051 |
|
| 1052 |
return pgstat_get_entry_ref(kind, dboid, objid, false, NULL) != NULL; |
| 1053 |
} |
| 1054 |
|
| 1055 |
/* |
| 1056 |
* Ensure snapshot for fixed-numbered 'kind' exists. |
| 1057 |
* |
| 1058 |
* Typically used by the pgstat_fetch_* functions for a kind of stats, before |
| 1059 |
* massaging the data into the desired format. |
| 1060 |
*/ |
| 1061 |
void |
| 1062 |
pgstat_snapshot_fixed(PgStat_Kind kind) |
| 1063 |
{ |
| 1064 |
Assert(pgstat_is_kind_valid(kind)); |
| 1065 |
Assert(pgstat_get_kind_info(kind)->fixed_amount); |
| 1066 |
|
| 1067 |
if (force_stats_snapshot_clear) |
| 1068 |
pgstat_clear_snapshot(); |
| 1069 |
|
| 1070 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT) |
| 1071 |
pgstat_build_snapshot(); |
| 1072 |
else |
| 1073 |
pgstat_build_snapshot_fixed(kind); |
| 1074 |
|
| 1075 |
if (pgstat_is_kind_builtin(kind)) |
| 1076 |
Assert(pgStatLocal.snapshot.fixed_valid[kind]); |
| 1077 |
else if (pgstat_is_kind_custom(kind)) |
| 1078 |
Assert(pgStatLocal.snapshot.custom_valid[kind - PGSTAT_KIND_CUSTOM_MIN]); |
| 1079 |
} |
| 1080 |
|
| 1081 |
static void |
| 1082 |
pgstat_init_snapshot_fixed(void) |
| 1083 |
{ |
| 1084 |
/* |
| 1085 |
* Initialize fixed-numbered statistics data in snapshots, only for custom |
| 1086 |
* stats kinds. |
| 1087 |
*/ |
| 1088 |
for (PgStat_Kind kind = PGSTAT_KIND_CUSTOM_MIN; kind <= PGSTAT_KIND_CUSTOM_MAX; kind++) |
| 1089 |
{ |
| 1090 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 1091 |
|
| 1092 |
if (!kind_info || !kind_info->fixed_amount) |
| 1093 |
continue; |
| 1094 |
|
| 1095 |
pgStatLocal.snapshot.custom_data[kind - PGSTAT_KIND_CUSTOM_MIN] = |
| 1096 |
MemoryContextAlloc(TopMemoryContext, kind_info->shared_data_len); |
| 1097 |
} |
| 1098 |
} |
| 1099 |
|
| 1100 |
static void |
| 1101 |
pgstat_prep_snapshot(void) |
| 1102 |
{ |
| 1103 |
if (force_stats_snapshot_clear) |
| 1104 |
pgstat_clear_snapshot(); |
| 1105 |
|
| 1106 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_NONE || |
| 1107 |
pgStatLocal.snapshot.stats != NULL) |
| 1108 |
return; |
| 1109 |
|
| 1110 |
if (!pgStatLocal.snapshot.context) |
| 1111 |
pgStatLocal.snapshot.context = AllocSetContextCreate(TopMemoryContext, |
| 1112 |
"PgStat Snapshot", |
| 1113 |
ALLOCSET_SMALL_SIZES); |
| 1114 |
|
| 1115 |
pgStatLocal.snapshot.stats = |
| 1116 |
pgstat_snapshot_create(pgStatLocal.snapshot.context, |
| 1117 |
PGSTAT_SNAPSHOT_HASH_SIZE, |
| 1118 |
NULL); |
| 1119 |
} |
| 1120 |
|
| 1121 |
static void |
| 1122 |
pgstat_build_snapshot(void) |
| 1123 |
{ |
| 1124 |
dshash_seq_status hstat; |
| 1125 |
PgStatShared_HashEntry *p; |
| 1126 |
|
| 1127 |
/* should only be called when we need a snapshot */ |
| 1128 |
Assert(pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT); |
| 1129 |
|
| 1130 |
/* snapshot already built */ |
| 1131 |
if (pgStatLocal.snapshot.mode == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT) |
| 1132 |
return; |
| 1133 |
|
| 1134 |
pgstat_prep_snapshot(); |
| 1135 |
|
| 1136 |
Assert(pgStatLocal.snapshot.stats->members == 0); |
| 1137 |
|
| 1138 |
pgStatLocal.snapshot.snapshot_timestamp = GetCurrentTimestamp(); |
| 1139 |
|
| 1140 |
/* |
| 1141 |
* Snapshot all variable stats. |
| 1142 |
*/ |
| 1143 |
dshash_seq_init(&hstat, pgStatLocal.shared_hash, false); |
| 1144 |
while ((p = dshash_seq_next(&hstat)) != NULL) |
| 1145 |
{ |
| 1146 |
PgStat_Kind kind = p->key.kind; |
| 1147 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 1148 |
bool found; |
| 1149 |
PgStat_SnapshotEntry *entry; |
| 1150 |
PgStatShared_Common *stats_data; |
| 1151 |
|
| 1152 |
/* |
| 1153 |
* Check if the stats object should be included in the snapshot. |
| 1154 |
* Unless the stats kind can be accessed from all databases (e.g., |
| 1155 |
* database stats themselves), we only include stats for the current |
| 1156 |
* database or objects not associated with a database (e.g. shared |
| 1157 |
* relations). |
| 1158 |
*/ |
| 1159 |
if (p->key.dboid != MyDatabaseId && |
| 1160 |
p->key.dboid != InvalidOid && |
| 1161 |
!kind_info->accessed_across_databases) |
| 1162 |
continue; |
| 1163 |
|
| 1164 |
if (p->dropped) |
| 1165 |
continue; |
| 1166 |
|
| 1167 |
Assert(pg_atomic_read_u32(&p->refcount) > 0); |
| 1168 |
|
| 1169 |
stats_data = dsa_get_address(pgStatLocal.dsa, p->body); |
| 1170 |
Assert(stats_data); |
| 1171 |
|
| 1172 |
entry = pgstat_snapshot_insert(pgStatLocal.snapshot.stats, p->key, &found); |
| 1173 |
Assert(!found); |
| 1174 |
|
| 1175 |
entry->data = MemoryContextAlloc(pgStatLocal.snapshot.context, |
| 1176 |
pgstat_get_entry_len(kind)); |
| 1177 |
|
| 1178 |
/* |
| 1179 |
* Acquire the LWLock directly instead of using |
| 1180 |
* pg_stat_lock_entry_shared() which requires a reference. |
| 1181 |
*/ |
| 1182 |
LWLockAcquire(&stats_data->lock, LW_SHARED); |
| 1183 |
memcpy(entry->data, |
| 1184 |
pgstat_get_entry_data(kind, stats_data), |
| 1185 |
pgstat_get_entry_len(kind)); |
| 1186 |
LWLockRelease(&stats_data->lock); |
| 1187 |
} |
| 1188 |
dshash_seq_term(&hstat); |
| 1189 |
|
| 1190 |
/* |
| 1191 |
* Build snapshot of all fixed-numbered stats. |
| 1192 |
*/ |
| 1193 |
for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) |
| 1194 |
{ |
| 1195 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 1196 |
|
| 1197 |
if (!kind_info) |
| 1198 |
continue; |
| 1199 |
if (!kind_info->fixed_amount) |
| 1200 |
{ |
| 1201 |
Assert(kind_info->snapshot_cb == NULL); |
| 1202 |
continue; |
| 1203 |
} |
| 1204 |
|
| 1205 |
pgstat_build_snapshot_fixed(kind); |
| 1206 |
} |
| 1207 |
|
| 1208 |
pgStatLocal.snapshot.mode = PGSTAT_FETCH_CONSISTENCY_SNAPSHOT; |
| 1209 |
} |
| 1210 |
|
| 1211 |
static void |
| 1212 |
pgstat_build_snapshot_fixed(PgStat_Kind kind) |
| 1213 |
{ |
| 1214 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 1215 |
int idx; |
| 1216 |
bool *valid; |
| 1217 |
|
| 1218 |
/* Position in fixed_valid or custom_valid */ |
| 1219 |
if (pgstat_is_kind_builtin(kind)) |
| 1220 |
{ |
| 1221 |
idx = kind; |
| 1222 |
valid = pgStatLocal.snapshot.fixed_valid; |
| 1223 |
} |
| 1224 |
else |
| 1225 |
{ |
| 1226 |
idx = kind - PGSTAT_KIND_CUSTOM_MIN; |
| 1227 |
valid = pgStatLocal.snapshot.custom_valid; |
| 1228 |
} |
| 1229 |
|
| 1230 |
Assert(kind_info->fixed_amount); |
| 1231 |
Assert(kind_info->snapshot_cb != NULL); |
| 1232 |
|
| 1233 |
if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_NONE) |
| 1234 |
{ |
| 1235 |
/* rebuild every time */ |
| 1236 |
valid[idx] = false; |
| 1237 |
} |
| 1238 |
else if (valid[idx]) |
| 1239 |
{ |
| 1240 |
/* in snapshot mode we shouldn't get called again */ |
| 1241 |
Assert(pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_CACHE); |
| 1242 |
return; |
| 1243 |
} |
| 1244 |
|
| 1245 |
Assert(!valid[idx]); |
| 1246 |
|
| 1247 |
kind_info->snapshot_cb(); |
| 1248 |
|
| 1249 |
Assert(!valid[idx]); |
| 1250 |
valid[idx] = true; |
| 1251 |
} |
| 1252 |
|
| 1253 |
|
| 1254 |
/* ------------------------------------------------------------ |
| 1255 |
* Backend-local pending stats infrastructure |
| 1256 |
* ------------------------------------------------------------ |
| 1257 |
*/ |
| 1258 |
|
| 1259 |
/* |
| 1260 |
* Returns the appropriate PgStat_EntryRef, preparing it to receive pending |
| 1261 |
* stats if not already done. |
| 1262 |
* |
| 1263 |
* If created_entry is non-NULL, it'll be set to true if the entry is newly |
| 1264 |
* created, false otherwise. |
| 1265 |
*/ |
| 1266 |
PgStat_EntryRef * |
| 1267 |
pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, uint64 objid, bool *created_entry) |
| 1268 |
{ |
| 1269 |
PgStat_EntryRef *entry_ref; |
| 1270 |
|
| 1271 |
/* need to be able to flush out */ |
| 1272 |
Assert(pgstat_get_kind_info(kind)->flush_pending_cb != NULL); |
| 1273 |
|
| 1274 |
if (unlikely(!pgStatPendingContext)) |
| 1275 |
{ |
| 1276 |
pgStatPendingContext = |
| 1277 |
AllocSetContextCreate(TopMemoryContext, |
| 1278 |
"PgStat Pending", |
| 1279 |
ALLOCSET_SMALL_SIZES); |
| 1280 |
} |
| 1281 |
|
| 1282 |
entry_ref = pgstat_get_entry_ref(kind, dboid, objid, |
| 1283 |
true, created_entry); |
| 1284 |
|
| 1285 |
if (entry_ref->pending == NULL) |
| 1286 |
{ |
| 1287 |
size_t entrysize = pgstat_get_kind_info(kind)->pending_size; |
| 1288 |
|
| 1289 |
Assert(entrysize != (size_t) -1); |
| 1290 |
|
| 1291 |
entry_ref->pending = MemoryContextAllocZero(pgStatPendingContext, entrysize); |
| 1292 |
dlist_push_tail(&pgStatPending, &entry_ref->pending_node); |
| 1293 |
} |
| 1294 |
|
| 1295 |
return entry_ref; |
| 1296 |
} |
| 1297 |
|
| 1298 |
/* |
| 1299 |
* Return an existing stats entry, or NULL. |
| 1300 |
* |
| 1301 |
* This should only be used for helper function for pgstatfuncs.c - outside of |
| 1302 |
* that it shouldn't be needed. |
| 1303 |
*/ |
| 1304 |
PgStat_EntryRef * |
| 1305 |
pgstat_fetch_pending_entry(PgStat_Kind kind, Oid dboid, uint64 objid) |
| 1306 |
{ |
| 1307 |
PgStat_EntryRef *entry_ref; |
| 1308 |
|
| 1309 |
entry_ref = pgstat_get_entry_ref(kind, dboid, objid, false, NULL); |
| 1310 |
|
| 1311 |
if (entry_ref == NULL || entry_ref->pending == NULL) |
| 1312 |
return NULL; |
| 1313 |
|
| 1314 |
return entry_ref; |
| 1315 |
} |
| 1316 |
|
| 1317 |
void |
| 1318 |
pgstat_delete_pending_entry(PgStat_EntryRef *entry_ref) |
| 1319 |
{ |
| 1320 |
PgStat_Kind kind = entry_ref->shared_entry->key.kind; |
| 1321 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 1322 |
void *pending_data = entry_ref->pending; |
| 1323 |
|
| 1324 |
Assert(pending_data != NULL); |
| 1325 |
/* !fixed_amount stats should be handled explicitly */ |
| 1326 |
Assert(!pgstat_get_kind_info(kind)->fixed_amount); |
| 1327 |
|
| 1328 |
if (kind_info->delete_pending_cb) |
| 1329 |
kind_info->delete_pending_cb(entry_ref); |
| 1330 |
|
| 1331 |
pfree(pending_data); |
| 1332 |
entry_ref->pending = NULL; |
| 1333 |
|
| 1334 |
dlist_delete(&entry_ref->pending_node); |
| 1335 |
} |
| 1336 |
|
| 1337 |
/* |
| 1338 |
* Flush out pending variable-numbered stats. |
| 1339 |
*/ |
| 1340 |
static bool |
| 1341 |
pgstat_flush_pending_entries(bool nowait) |
| 1342 |
{ |
| 1343 |
bool have_pending = false; |
| 1344 |
dlist_node *cur = NULL; |
| 1345 |
|
| 1346 |
/* |
| 1347 |
* Need to be a bit careful iterating over the list of pending entries. |
| 1348 |
* Processing a pending entry may queue further pending entries to the end |
| 1349 |
* of the list that we want to process, so a simple iteration won't do. |
| 1350 |
* Further complicating matters is that we want to delete the current |
| 1351 |
* entry in each iteration from the list if we flushed successfully. |
| 1352 |
* |
| 1353 |
* So we just keep track of the next pointer in each loop iteration. |
| 1354 |
*/ |
| 1355 |
if (!dlist_is_empty(&pgStatPending)) |
| 1356 |
cur = dlist_head_node(&pgStatPending); |
| 1357 |
|
| 1358 |
while (cur) |
| 1359 |
{ |
| 1360 |
PgStat_EntryRef *entry_ref = |
| 1361 |
dlist_container(PgStat_EntryRef, pending_node, cur); |
| 1362 |
PgStat_HashKey key = entry_ref->shared_entry->key; |
| 1363 |
PgStat_Kind kind = key.kind; |
| 1364 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 1365 |
bool did_flush; |
| 1366 |
dlist_node *next; |
| 1367 |
|
| 1368 |
Assert(!kind_info->fixed_amount); |
| 1369 |
Assert(kind_info->flush_pending_cb != NULL); |
| 1370 |
|
| 1371 |
/* flush the stats, if possible */ |
| 1372 |
did_flush = kind_info->flush_pending_cb(entry_ref, nowait); |
| 1373 |
|
| 1374 |
Assert(did_flush || nowait); |
| 1375 |
|
| 1376 |
/* determine next entry, before deleting the pending entry */ |
| 1377 |
if (dlist_has_next(&pgStatPending, cur)) |
| 1378 |
next = dlist_next_node(&pgStatPending, cur); |
| 1379 |
else |
| 1380 |
next = NULL; |
| 1381 |
|
| 1382 |
/* if successfully flushed, remove entry */ |
| 1383 |
if (did_flush) |
| 1384 |
pgstat_delete_pending_entry(entry_ref); |
| 1385 |
else |
| 1386 |
have_pending = true; |
| 1387 |
|
| 1388 |
cur = next; |
| 1389 |
} |
| 1390 |
|
| 1391 |
Assert(dlist_is_empty(&pgStatPending) == !have_pending); |
| 1392 |
|
| 1393 |
return have_pending; |
| 1394 |
} |
| 1395 |
|
| 1396 |
|
| 1397 |
/* ------------------------------------------------------------ |
| 1398 |
* Helper / infrastructure functions |
| 1399 |
* ------------------------------------------------------------ |
| 1400 |
*/ |
| 1401 |
|
| 1402 |
PgStat_Kind |
| 1403 |
pgstat_get_kind_from_str(char *kind_str) |
| 1404 |
{ |
| 1405 |
for (PgStat_Kind kind = PGSTAT_KIND_BUILTIN_MIN; kind <= PGSTAT_KIND_BUILTIN_MAX; kind++) |
| 1406 |
{ |
| 1407 |
if (pg_strcasecmp(kind_str, pgstat_kind_builtin_infos[kind].name) == 0) |
| 1408 |
return kind; |
| 1409 |
} |
| 1410 |
|
| 1411 |
/* Check the custom set of cumulative stats */ |
| 1412 |
if (pgstat_kind_custom_infos) |
| 1413 |
{ |
| 1414 |
for (PgStat_Kind kind = PGSTAT_KIND_CUSTOM_MIN; kind <= PGSTAT_KIND_CUSTOM_MAX; kind++) |
| 1415 |
{ |
| 1416 |
uint32 idx = kind - PGSTAT_KIND_CUSTOM_MIN; |
| 1417 |
|
| 1418 |
if (pgstat_kind_custom_infos[idx] && |
| 1419 |
pg_strcasecmp(kind_str, pgstat_kind_custom_infos[idx]->name) == 0) |
| 1420 |
return kind; |
| 1421 |
} |
| 1422 |
} |
| 1423 |
|
| 1424 |
ereport(ERROR, |
| 1425 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 1426 |
errmsg("invalid statistics kind: \"%s\"", kind_str))); |
| 1427 |
return PGSTAT_KIND_INVALID; /* avoid compiler warnings */ |
| 1428 |
} |
| 1429 |
|
| 1430 |
static inline bool |
| 1431 |
pgstat_is_kind_valid(PgStat_Kind kind) |
| 1432 |
{ |
| 1433 |
return pgstat_is_kind_builtin(kind) || pgstat_is_kind_custom(kind); |
| 1434 |
} |
| 1435 |
|
| 1436 |
const PgStat_KindInfo * |
| 1437 |
pgstat_get_kind_info(PgStat_Kind kind) |
| 1438 |
{ |
| 1439 |
if (pgstat_is_kind_builtin(kind)) |
| 1440 |
return &pgstat_kind_builtin_infos[kind]; |
| 1441 |
|
| 1442 |
if (pgstat_is_kind_custom(kind)) |
| 1443 |
{ |
| 1444 |
uint32 idx = kind - PGSTAT_KIND_CUSTOM_MIN; |
| 1445 |
|
| 1446 |
if (pgstat_kind_custom_infos == NULL || |
| 1447 |
pgstat_kind_custom_infos[idx] == NULL) |
| 1448 |
return NULL; |
| 1449 |
return pgstat_kind_custom_infos[idx]; |
| 1450 |
} |
| 1451 |
|
| 1452 |
return NULL; |
| 1453 |
} |
| 1454 |
|
| 1455 |
/* |
| 1456 |
* Register a new stats kind. |
| 1457 |
* |
| 1458 |
* PgStat_Kinds must be globally unique across all extensions. Refer |
| 1459 |
* to https://wiki.postgresql.org/wiki/CustomCumulativeStats to reserve a |
| 1460 |
* unique ID for your extension, to avoid conflicts with other extension |
| 1461 |
* developers. During development, use PGSTAT_KIND_EXPERIMENTAL to avoid |
| 1462 |
* needlessly reserving a new ID. |
| 1463 |
*/ |
| 1464 |
void |
| 1465 |
pgstat_register_kind(PgStat_Kind kind, const PgStat_KindInfo *kind_info) |
| 1466 |
{ |
| 1467 |
uint32 idx = kind - PGSTAT_KIND_CUSTOM_MIN; |
| 1468 |
|
| 1469 |
if (kind_info->name == NULL || strlen(kind_info->name) == 0) |
| 1470 |
ereport(ERROR, |
| 1471 |
(errmsg("custom cumulative statistics name is invalid"), |
| 1472 |
errhint("Provide a non-empty name for the custom cumulative statistics."))); |
| 1473 |
|
| 1474 |
if (!pgstat_is_kind_custom(kind)) |
| 1475 |
ereport(ERROR, (errmsg("custom cumulative statistics ID %u is out of range", kind), |
| 1476 |
errhint("Provide a custom cumulative statistics ID between %u and %u.", |
| 1477 |
PGSTAT_KIND_CUSTOM_MIN, PGSTAT_KIND_CUSTOM_MAX))); |
| 1478 |
|
| 1479 |
if (!process_shared_preload_libraries_in_progress) |
| 1480 |
ereport(ERROR, |
| 1481 |
(errmsg("failed to register custom cumulative statistics \"%s\" with ID %u", kind_info->name, kind), |
| 1482 |
errdetail("Custom cumulative statistics must be registered while initializing modules in \"shared_preload_libraries\"."))); |
| 1483 |
|
| 1484 |
/* |
| 1485 |
* Check some data for fixed-numbered stats. |
| 1486 |
*/ |
| 1487 |
if (kind_info->fixed_amount) |
| 1488 |
{ |
| 1489 |
if (kind_info->shared_size == 0) |
| 1490 |
ereport(ERROR, |
| 1491 |
(errmsg("custom cumulative statistics property is invalid"), |
| 1492 |
errhint("Custom cumulative statistics require a shared memory size for fixed-numbered objects."))); |
| 1493 |
} |
| 1494 |
|
| 1495 |
/* |
| 1496 |
* If pgstat_kind_custom_infos is not available yet, allocate it. |
| 1497 |
*/ |
| 1498 |
if (pgstat_kind_custom_infos == NULL) |
| 1499 |
{ |
| 1500 |
pgstat_kind_custom_infos = (const PgStat_KindInfo **) |
| 1501 |
MemoryContextAllocZero(TopMemoryContext, |
| 1502 |
sizeof(PgStat_KindInfo *) * PGSTAT_KIND_CUSTOM_SIZE); |
| 1503 |
} |
| 1504 |
|
| 1505 |
if (pgstat_kind_custom_infos[idx] != NULL && |
| 1506 |
pgstat_kind_custom_infos[idx]->name != NULL) |
| 1507 |
ereport(ERROR, |
| 1508 |
(errmsg("failed to register custom cumulative statistics \"%s\" with ID %u", kind_info->name, kind), |
| 1509 |
errdetail("Custom cumulative statistics \"%s\" already registered with the same ID.", |
| 1510 |
pgstat_kind_custom_infos[idx]->name))); |
| 1511 |
|
| 1512 |
/* check for existing custom stats with the same name */ |
| 1513 |
for (PgStat_Kind existing_kind = PGSTAT_KIND_CUSTOM_MIN; existing_kind <= PGSTAT_KIND_CUSTOM_MAX; existing_kind++) |
| 1514 |
{ |
| 1515 |
uint32 existing_idx = existing_kind - PGSTAT_KIND_CUSTOM_MIN; |
| 1516 |
|
| 1517 |
if (pgstat_kind_custom_infos[existing_idx] == NULL) |
| 1518 |
continue; |
| 1519 |
if (!pg_strcasecmp(pgstat_kind_custom_infos[existing_idx]->name, kind_info->name)) |
| 1520 |
ereport(ERROR, |
| 1521 |
(errmsg("failed to register custom cumulative statistics \"%s\" with ID %u", kind_info->name, kind), |
| 1522 |
errdetail("Existing cumulative statistics with ID %u has the same name.", existing_kind))); |
| 1523 |
} |
| 1524 |
|
| 1525 |
/* Register it */ |
| 1526 |
pgstat_kind_custom_infos[idx] = kind_info; |
| 1527 |
ereport(LOG, |
| 1528 |
(errmsg("registered custom cumulative statistics \"%s\" with ID %u", |
| 1529 |
kind_info->name, kind))); |
| 1530 |
} |
| 1531 |
|
| 1532 |
/* |
| 1533 |
* Stats should only be reported after pgstat_initialize() and before |
| 1534 |
* pgstat_shutdown(). This check is put in a few central places to catch |
| 1535 |
* violations of this rule more easily. |
| 1536 |
*/ |
| 1537 |
#ifdef USE_ASSERT_CHECKING |
| 1538 |
void |
| 1539 |
pgstat_assert_is_up(void) |
| 1540 |
{ |
| 1541 |
Assert(pgstat_is_initialized && !pgstat_is_shutdown); |
| 1542 |
} |
| 1543 |
#endif |
| 1544 |
|
| 1545 |
|
| 1546 |
/* ------------------------------------------------------------ |
| 1547 |
* reading and writing of on-disk stats file |
| 1548 |
* ------------------------------------------------------------ |
| 1549 |
*/ |
| 1550 |
|
| 1551 |
/* helpers for pgstat_write_statsfile() */ |
| 1552 |
static void |
| 1553 |
write_chunk(FILE *fpout, void *ptr, size_t len) |
| 1554 |
{ |
| 1555 |
int rc; |
| 1556 |
|
| 1557 |
rc = fwrite(ptr, len, 1, fpout); |
| 1558 |
|
| 1559 |
/* we'll check for errors with ferror once at the end */ |
| 1560 |
(void) rc; |
| 1561 |
} |
| 1562 |
|
| 1563 |
#define write_chunk_s(fpout, ptr) write_chunk(fpout, ptr, sizeof(*ptr)) |
| 1564 |
|
| 1565 |
/* |
| 1566 |
* This function is called in the last process that is accessing the shared |
| 1567 |
* stats so locking is not required. |
| 1568 |
*/ |
| 1569 |
static void |
| 1570 |
pgstat_write_statsfile(void) |
| 1571 |
{ |
| 1572 |
FILE *fpout; |
| 1573 |
int32 format_id; |
| 1574 |
const char *tmpfile = PGSTAT_STAT_PERMANENT_TMPFILE; |
| 1575 |
const char *statfile = PGSTAT_STAT_PERMANENT_FILENAME; |
| 1576 |
dshash_seq_status hstat; |
| 1577 |
PgStatShared_HashEntry *ps; |
| 1578 |
|
| 1579 |
pgstat_assert_is_up(); |
| 1580 |
|
| 1581 |
/* should be called only by the checkpointer or single user mode */ |
| 1582 |
Assert(!IsUnderPostmaster || MyBackendType == B_CHECKPOINTER); |
| 1583 |
|
| 1584 |
/* we're shutting down, so it's ok to just override this */ |
| 1585 |
pgstat_fetch_consistency = PGSTAT_FETCH_CONSISTENCY_NONE; |
| 1586 |
|
| 1587 |
elog(DEBUG2, "writing stats file \"%s\"", statfile); |
| 1588 |
|
| 1589 |
/* |
| 1590 |
* Open the statistics temp file to write out the current values. |
| 1591 |
*/ |
| 1592 |
fpout = AllocateFile(tmpfile, PG_BINARY_W); |
| 1593 |
if (fpout == NULL) |
| 1594 |
{ |
| 1595 |
ereport(LOG, |
| 1596 |
(errcode_for_file_access(), |
| 1597 |
errmsg("could not open temporary statistics file \"%s\": %m", |
| 1598 |
tmpfile))); |
| 1599 |
return; |
| 1600 |
} |
| 1601 |
|
| 1602 |
/* |
| 1603 |
* Write the file header --- currently just a format ID. |
| 1604 |
*/ |
| 1605 |
format_id = PGSTAT_FILE_FORMAT_ID; |
| 1606 |
write_chunk_s(fpout, &format_id); |
| 1607 |
|
| 1608 |
/* Write various stats structs for fixed number of objects */ |
| 1609 |
for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) |
| 1610 |
{ |
| 1611 |
char *ptr; |
| 1612 |
const PgStat_KindInfo *info = pgstat_get_kind_info(kind); |
| 1613 |
|
| 1614 |
if (!info || !info->fixed_amount) |
| 1615 |
continue; |
| 1616 |
|
| 1617 |
if (pgstat_is_kind_builtin(kind)) |
| 1618 |
Assert(info->snapshot_ctl_off != 0); |
| 1619 |
|
| 1620 |
/* skip if no need to write to file */ |
| 1621 |
if (!info->write_to_file) |
| 1622 |
continue; |
| 1623 |
|
| 1624 |
pgstat_build_snapshot_fixed(kind); |
| 1625 |
if (pgstat_is_kind_builtin(kind)) |
| 1626 |
ptr = ((char *) &pgStatLocal.snapshot) + info->snapshot_ctl_off; |
| 1627 |
else |
| 1628 |
ptr = pgStatLocal.snapshot.custom_data[kind - PGSTAT_KIND_CUSTOM_MIN]; |
| 1629 |
|
| 1630 |
fputc(PGSTAT_FILE_ENTRY_FIXED, fpout); |
| 1631 |
write_chunk_s(fpout, &kind); |
| 1632 |
write_chunk(fpout, ptr, info->shared_data_len); |
| 1633 |
} |
| 1634 |
|
| 1635 |
/* |
| 1636 |
* Walk through the stats entries |
| 1637 |
*/ |
| 1638 |
dshash_seq_init(&hstat, pgStatLocal.shared_hash, false); |
| 1639 |
while ((ps = dshash_seq_next(&hstat)) != NULL) |
| 1640 |
{ |
| 1641 |
PgStatShared_Common *shstats; |
| 1642 |
const PgStat_KindInfo *kind_info = NULL; |
| 1643 |
|
| 1644 |
CHECK_FOR_INTERRUPTS(); |
| 1645 |
|
| 1646 |
/* |
| 1647 |
* We should not see any "dropped" entries when writing the stats |
| 1648 |
* file, as all backends and auxiliary processes should have cleaned |
| 1649 |
* up their references before they terminated. |
| 1650 |
* |
| 1651 |
* However, since we are already shutting down, it is not worth |
| 1652 |
* crashing the server over any potential cleanup issues, so we simply |
| 1653 |
* skip such entries if encountered. |
| 1654 |
*/ |
| 1655 |
Assert(!ps->dropped); |
| 1656 |
if (ps->dropped) |
| 1657 |
continue; |
| 1658 |
|
| 1659 |
/* |
| 1660 |
* This discards data related to custom stats kinds that are unknown |
| 1661 |
* to this process. |
| 1662 |
*/ |
| 1663 |
if (!pgstat_is_kind_valid(ps->key.kind)) |
| 1664 |
{ |
| 1665 |
elog(WARNING, "found unknown stats entry %u/%u/%" PRIu64, |
| 1666 |
ps->key.kind, ps->key.dboid, |
| 1667 |
ps->key.objid); |
| 1668 |
continue; |
| 1669 |
} |
| 1670 |
|
| 1671 |
shstats = (PgStatShared_Common *) dsa_get_address(pgStatLocal.dsa, ps->body); |
| 1672 |
|
| 1673 |
kind_info = pgstat_get_kind_info(ps->key.kind); |
| 1674 |
|
| 1675 |
/* if not dropped the valid-entry refcount should exist */ |
| 1676 |
Assert(pg_atomic_read_u32(&ps->refcount) > 0); |
| 1677 |
|
| 1678 |
/* skip if no need to write to file */ |
| 1679 |
if (!kind_info->write_to_file) |
| 1680 |
continue; |
| 1681 |
|
| 1682 |
if (!kind_info->to_serialized_name) |
| 1683 |
{ |
| 1684 |
/* normal stats entry, identified by PgStat_HashKey */ |
| 1685 |
fputc(PGSTAT_FILE_ENTRY_HASH, fpout); |
| 1686 |
write_chunk_s(fpout, &ps->key); |
| 1687 |
} |
| 1688 |
else |
| 1689 |
{ |
| 1690 |
/* stats entry identified by name on disk (e.g. slots) */ |
| 1691 |
NameData name; |
| 1692 |
|
| 1693 |
kind_info->to_serialized_name(&ps->key, shstats, &name); |
| 1694 |
|
| 1695 |
fputc(PGSTAT_FILE_ENTRY_NAME, fpout); |
| 1696 |
write_chunk_s(fpout, &ps->key.kind); |
| 1697 |
write_chunk_s(fpout, &name); |
| 1698 |
} |
| 1699 |
|
| 1700 |
/* Write except the header part of the entry */ |
| 1701 |
write_chunk(fpout, |
| 1702 |
pgstat_get_entry_data(ps->key.kind, shstats), |
| 1703 |
pgstat_get_entry_len(ps->key.kind)); |
| 1704 |
} |
| 1705 |
dshash_seq_term(&hstat); |
| 1706 |
|
| 1707 |
/* |
| 1708 |
* No more output to be done. Close the temp file and replace the old |
| 1709 |
* pgstat.stat with it. The ferror() check replaces testing for error |
| 1710 |
* after each individual fputc or fwrite (in write_chunk()) above. |
| 1711 |
*/ |
| 1712 |
fputc(PGSTAT_FILE_ENTRY_END, fpout); |
| 1713 |
|
| 1714 |
if (ferror(fpout)) |
| 1715 |
{ |
| 1716 |
ereport(LOG, |
| 1717 |
(errcode_for_file_access(), |
| 1718 |
errmsg("could not write temporary statistics file \"%s\": %m", |
| 1719 |
tmpfile))); |
| 1720 |
FreeFile(fpout); |
| 1721 |
unlink(tmpfile); |
| 1722 |
} |
| 1723 |
else if (FreeFile(fpout) < 0) |
| 1724 |
{ |
| 1725 |
ereport(LOG, |
| 1726 |
(errcode_for_file_access(), |
| 1727 |
errmsg("could not close temporary statistics file \"%s\": %m", |
| 1728 |
tmpfile))); |
| 1729 |
unlink(tmpfile); |
| 1730 |
} |
| 1731 |
else if (durable_rename(tmpfile, statfile, LOG) < 0) |
| 1732 |
{ |
| 1733 |
/* durable_rename already emitted log message */ |
| 1734 |
unlink(tmpfile); |
| 1735 |
} |
| 1736 |
} |
| 1737 |
|
| 1738 |
/* helpers for pgstat_read_statsfile() */ |
| 1739 |
static bool |
| 1740 |
read_chunk(FILE *fpin, void *ptr, size_t len) |
| 1741 |
{ |
| 1742 |
return fread(ptr, 1, len, fpin) == len; |
| 1743 |
} |
| 1744 |
|
| 1745 |
#define read_chunk_s(fpin, ptr) read_chunk(fpin, ptr, sizeof(*ptr)) |
| 1746 |
|
| 1747 |
/* |
| 1748 |
* Reads in existing statistics file into memory. |
| 1749 |
* |
| 1750 |
* This function is called in the only process that is accessing the shared |
| 1751 |
* stats so locking is not required. |
| 1752 |
*/ |
| 1753 |
static void |
| 1754 |
pgstat_read_statsfile(void) |
| 1755 |
{ |
| 1756 |
FILE *fpin; |
| 1757 |
int32 format_id; |
| 1758 |
bool found; |
| 1759 |
const char *statfile = PGSTAT_STAT_PERMANENT_FILENAME; |
| 1760 |
PgStat_ShmemControl *shmem = pgStatLocal.shmem; |
| 1761 |
|
| 1762 |
/* shouldn't be called from postmaster */ |
| 1763 |
Assert(IsUnderPostmaster || !IsPostmasterEnvironment); |
| 1764 |
|
| 1765 |
elog(DEBUG2, "reading stats file \"%s\"", statfile); |
| 1766 |
|
| 1767 |
/* |
| 1768 |
* Try to open the stats file. If it doesn't exist, the backends simply |
| 1769 |
* returns zero for anything and statistics simply starts from scratch |
| 1770 |
* with empty counters. |
| 1771 |
* |
| 1772 |
* ENOENT is a possibility if stats collection was previously disabled or |
| 1773 |
* has not yet written the stats file for the first time. Any other |
| 1774 |
* failure condition is suspicious. |
| 1775 |
*/ |
| 1776 |
if ((fpin = AllocateFile(statfile, PG_BINARY_R)) == NULL) |
| 1777 |
{ |
| 1778 |
if (errno != ENOENT) |
| 1779 |
ereport(LOG, |
| 1780 |
(errcode_for_file_access(), |
| 1781 |
errmsg("could not open statistics file \"%s\": %m", |
| 1782 |
statfile))); |
| 1783 |
pgstat_reset_after_failure(); |
| 1784 |
return; |
| 1785 |
} |
| 1786 |
|
| 1787 |
/* |
| 1788 |
* Verify it's of the expected format. |
| 1789 |
*/ |
| 1790 |
if (!read_chunk_s(fpin, &format_id)) |
| 1791 |
{ |
| 1792 |
elog(WARNING, "could not read format ID"); |
| 1793 |
goto error; |
| 1794 |
} |
| 1795 |
|
| 1796 |
if (format_id != PGSTAT_FILE_FORMAT_ID) |
| 1797 |
{ |
| 1798 |
elog(WARNING, "found incorrect format ID %d (expected %d)", |
| 1799 |
format_id, PGSTAT_FILE_FORMAT_ID); |
| 1800 |
goto error; |
| 1801 |
} |
| 1802 |
|
| 1803 |
/* |
| 1804 |
* We found an existing statistics file. Read it and put all the stats |
| 1805 |
* data into place. |
| 1806 |
*/ |
| 1807 |
for (;;) |
| 1808 |
{ |
| 1809 |
int t = fgetc(fpin); |
| 1810 |
|
| 1811 |
switch (t) |
| 1812 |
{ |
| 1813 |
case PGSTAT_FILE_ENTRY_FIXED: |
| 1814 |
{ |
| 1815 |
PgStat_Kind kind; |
| 1816 |
const PgStat_KindInfo *info; |
| 1817 |
char *ptr; |
| 1818 |
|
| 1819 |
/* entry for fixed-numbered stats */ |
| 1820 |
if (!read_chunk_s(fpin, &kind)) |
| 1821 |
{ |
| 1822 |
elog(WARNING, "could not read stats kind for entry of type %c", t); |
| 1823 |
goto error; |
| 1824 |
} |
| 1825 |
|
| 1826 |
if (!pgstat_is_kind_valid(kind)) |
| 1827 |
{ |
| 1828 |
elog(WARNING, "invalid stats kind %u for entry of type %c", |
| 1829 |
kind, t); |
| 1830 |
goto error; |
| 1831 |
} |
| 1832 |
|
| 1833 |
info = pgstat_get_kind_info(kind); |
| 1834 |
if (!info) |
| 1835 |
{ |
| 1836 |
elog(WARNING, "could not find information of kind %u for entry of type %c", |
| 1837 |
kind, t); |
| 1838 |
goto error; |
| 1839 |
} |
| 1840 |
|
| 1841 |
if (!info->fixed_amount) |
| 1842 |
{ |
| 1843 |
elog(WARNING, "invalid fixed_amount in stats kind %u for entry of type %c", |
| 1844 |
kind, t); |
| 1845 |
goto error; |
| 1846 |
} |
| 1847 |
|
| 1848 |
/* Load back stats into shared memory */ |
| 1849 |
if (pgstat_is_kind_builtin(kind)) |
| 1850 |
ptr = ((char *) shmem) + info->shared_ctl_off + |
| 1851 |
info->shared_data_off; |
| 1852 |
else |
| 1853 |
{ |
| 1854 |
int idx = kind - PGSTAT_KIND_CUSTOM_MIN; |
| 1855 |
|
| 1856 |
ptr = ((char *) shmem->custom_data[idx]) + |
| 1857 |
info->shared_data_off; |
| 1858 |
} |
| 1859 |
|
| 1860 |
if (!read_chunk(fpin, ptr, info->shared_data_len)) |
| 1861 |
{ |
| 1862 |
elog(WARNING, "could not read data of stats kind %u for entry of type %c with size %u", |
| 1863 |
kind, t, info->shared_data_len); |
| 1864 |
goto error; |
| 1865 |
} |
| 1866 |
|
| 1867 |
break; |
| 1868 |
} |
| 1869 |
case PGSTAT_FILE_ENTRY_HASH: |
| 1870 |
case PGSTAT_FILE_ENTRY_NAME: |
| 1871 |
{ |
| 1872 |
PgStat_HashKey key; |
| 1873 |
PgStatShared_HashEntry *p; |
| 1874 |
PgStatShared_Common *header; |
| 1875 |
|
| 1876 |
CHECK_FOR_INTERRUPTS(); |
| 1877 |
|
| 1878 |
if (t == PGSTAT_FILE_ENTRY_HASH) |
| 1879 |
{ |
| 1880 |
/* normal stats entry, identified by PgStat_HashKey */ |
| 1881 |
if (!read_chunk_s(fpin, &key)) |
| 1882 |
{ |
| 1883 |
elog(WARNING, "could not read key for entry of type %c", t); |
| 1884 |
goto error; |
| 1885 |
} |
| 1886 |
|
| 1887 |
if (!pgstat_is_kind_valid(key.kind)) |
| 1888 |
{ |
| 1889 |
elog(WARNING, "invalid stats kind for entry %u/%u/%" PRIu64 " of type %c", |
| 1890 |
key.kind, key.dboid, |
| 1891 |
key.objid, t); |
| 1892 |
goto error; |
| 1893 |
} |
| 1894 |
|
| 1895 |
if (!pgstat_get_kind_info(key.kind)) |
| 1896 |
{ |
| 1897 |
elog(WARNING, "could not find information of kind for entry %u/%u/%" PRIu64 " of type %c", |
| 1898 |
key.kind, key.dboid, |
| 1899 |
key.objid, t); |
| 1900 |
goto error; |
| 1901 |
} |
| 1902 |
} |
| 1903 |
else |
| 1904 |
{ |
| 1905 |
/* stats entry identified by name on disk (e.g. slots) */ |
| 1906 |
const PgStat_KindInfo *kind_info = NULL; |
| 1907 |
PgStat_Kind kind; |
| 1908 |
NameData name; |
| 1909 |
|
| 1910 |
if (!read_chunk_s(fpin, &kind)) |
| 1911 |
{ |
| 1912 |
elog(WARNING, "could not read stats kind for entry of type %c", t); |
| 1913 |
goto error; |
| 1914 |
} |
| 1915 |
if (!read_chunk_s(fpin, &name)) |
| 1916 |
{ |
| 1917 |
elog(WARNING, "could not read name of stats kind %u for entry of type %c", |
| 1918 |
kind, t); |
| 1919 |
goto error; |
| 1920 |
} |
| 1921 |
if (!pgstat_is_kind_valid(kind)) |
| 1922 |
{ |
| 1923 |
elog(WARNING, "invalid stats kind %u for entry of type %c", |
| 1924 |
kind, t); |
| 1925 |
goto error; |
| 1926 |
} |
| 1927 |
|
| 1928 |
kind_info = pgstat_get_kind_info(kind); |
| 1929 |
if (!kind_info) |
| 1930 |
{ |
| 1931 |
elog(WARNING, "could not find information of kind %u for entry of type %c", |
| 1932 |
kind, t); |
| 1933 |
goto error; |
| 1934 |
} |
| 1935 |
|
| 1936 |
if (!kind_info->from_serialized_name) |
| 1937 |
{ |
| 1938 |
elog(WARNING, "invalid from_serialized_name in stats kind %u for entry of type %c", |
| 1939 |
kind, t); |
| 1940 |
goto error; |
| 1941 |
} |
| 1942 |
|
| 1943 |
if (!kind_info->from_serialized_name(&name, &key)) |
| 1944 |
{ |
| 1945 |
/* skip over data for entry we don't care about */ |
| 1946 |
if (fseek(fpin, pgstat_get_entry_len(kind), SEEK_CUR) != 0) |
| 1947 |
{ |
| 1948 |
elog(WARNING, "could not seek \"%s\" of stats kind %u for entry of type %c", |
| 1949 |
NameStr(name), kind, t); |
| 1950 |
goto error; |
| 1951 |
} |
| 1952 |
|
| 1953 |
continue; |
| 1954 |
} |
| 1955 |
|
| 1956 |
Assert(key.kind == kind); |
| 1957 |
} |
| 1958 |
|
| 1959 |
/* |
| 1960 |
* This intentionally doesn't use pgstat_get_entry_ref() - |
| 1961 |
* putting all stats into checkpointer's |
| 1962 |
* pgStatEntryRefHash would be wasted effort and memory. |
| 1963 |
*/ |
| 1964 |
p = dshash_find_or_insert(pgStatLocal.shared_hash, &key, &found); |
| 1965 |
|
| 1966 |
/* don't allow duplicate entries */ |
| 1967 |
if (found) |
| 1968 |
{ |
| 1969 |
dshash_release_lock(pgStatLocal.shared_hash, p); |
| 1970 |
elog(WARNING, "found duplicate stats entry %u/%u/%" PRIu64 " of type %c", |
| 1971 |
key.kind, key.dboid, |
| 1972 |
key.objid, t); |
| 1973 |
goto error; |
| 1974 |
} |
| 1975 |
|
| 1976 |
header = pgstat_init_entry(key.kind, p); |
| 1977 |
dshash_release_lock(pgStatLocal.shared_hash, p); |
| 1978 |
if (header == NULL) |
| 1979 |
{ |
| 1980 |
/* |
| 1981 |
* It would be tempting to switch this ERROR to a |
| 1982 |
* WARNING, but it would mean that all the statistics |
| 1983 |
* are discarded when the environment fails on OOM. |
| 1984 |
*/ |
| 1985 |
elog(ERROR, "could not allocate entry %u/%u/%" PRIu64 " of type %c", |
| 1986 |
key.kind, key.dboid, |
| 1987 |
key.objid, t); |
| 1988 |
} |
| 1989 |
|
| 1990 |
if (!read_chunk(fpin, |
| 1991 |
pgstat_get_entry_data(key.kind, header), |
| 1992 |
pgstat_get_entry_len(key.kind))) |
| 1993 |
{ |
| 1994 |
elog(WARNING, "could not read data for entry %u/%u/%" PRIu64 " of type %c", |
| 1995 |
key.kind, key.dboid, |
| 1996 |
key.objid, t); |
| 1997 |
goto error; |
| 1998 |
} |
| 1999 |
|
| 2000 |
break; |
| 2001 |
} |
| 2002 |
case PGSTAT_FILE_ENTRY_END: |
| 2003 |
|
| 2004 |
/* |
| 2005 |
* check that PGSTAT_FILE_ENTRY_END actually signals end of |
| 2006 |
* file |
| 2007 |
*/ |
| 2008 |
if (fgetc(fpin) != EOF) |
| 2009 |
{ |
| 2010 |
elog(WARNING, "could not read end-of-file"); |
| 2011 |
goto error; |
| 2012 |
} |
| 2013 |
|
| 2014 |
goto done; |
| 2015 |
|
| 2016 |
default: |
| 2017 |
elog(WARNING, "could not read entry of type %c", t); |
| 2018 |
goto error; |
| 2019 |
} |
| 2020 |
} |
| 2021 |
|
| 2022 |
done: |
| 2023 |
FreeFile(fpin); |
| 2024 |
|
| 2025 |
elog(DEBUG2, "removing permanent stats file \"%s\"", statfile); |
| 2026 |
unlink(statfile); |
| 2027 |
|
| 2028 |
return; |
| 2029 |
|
| 2030 |
error: |
| 2031 |
ereport(LOG, |
| 2032 |
(errmsg("corrupted statistics file \"%s\"", statfile))); |
| 2033 |
|
| 2034 |
pgstat_reset_after_failure(); |
| 2035 |
|
| 2036 |
goto done; |
| 2037 |
} |
| 2038 |
|
| 2039 |
/* |
| 2040 |
* Helper to reset / drop stats after a crash or after restoring stats from |
| 2041 |
* disk failed, potentially after already loading parts. |
| 2042 |
*/ |
| 2043 |
static void |
| 2044 |
pgstat_reset_after_failure(void) |
| 2045 |
{ |
| 2046 |
TimestampTz ts = GetCurrentTimestamp(); |
| 2047 |
|
| 2048 |
/* reset fixed-numbered stats */ |
| 2049 |
for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) |
| 2050 |
{ |
| 2051 |
const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); |
| 2052 |
|
| 2053 |
if (!kind_info || !kind_info->fixed_amount) |
| 2054 |
continue; |
| 2055 |
|
| 2056 |
kind_info->reset_all_cb(ts); |
| 2057 |
} |
| 2058 |
|
| 2059 |
/* and drop variable-numbered ones */ |
| 2060 |
pgstat_drop_all_entries(); |
| 2061 |
} |
| 2062 |
|
| 2063 |
/* |
| 2064 |
* GUC assign_hook for stats_fetch_consistency. |
| 2065 |
*/ |
| 2066 |
void |
| 2067 |
assign_stats_fetch_consistency(int newval, void *extra) |
| 2068 |
{ |
| 2069 |
/* |
| 2070 |
* Changing this value in a transaction may cause snapshot state |
| 2071 |
* inconsistencies, so force a clear of the current snapshot on the next |
| 2072 |
* snapshot build attempt. |
| 2073 |
*/ |
| 2074 |
if (pgstat_fetch_consistency != newval) |
| 2075 |
force_stats_snapshot_clear = true; |
| 2076 |
} |