Line data Source code
1 : /* ----------
2 : * pgstat_internal.h
3 : *
4 : * Definitions for the PostgreSQL cumulative statistics system that should
5 : * only be needed by files implementing statistics support (rather than ones
6 : * reporting / querying stats).
7 : *
8 : * Copyright (c) 2001-2026, PostgreSQL Global Development Group
9 : *
10 : * src/include/utils/pgstat_internal.h
11 : * ----------
12 : */
13 : #ifndef PGSTAT_INTERNAL_H
14 : #define PGSTAT_INTERNAL_H
15 :
16 :
17 : #include "common/hashfn_unstable.h"
18 : #include "lib/dshash.h"
19 : #include "lib/ilist.h"
20 : #include "pgstat.h"
21 : #include "storage/lwlock.h"
22 : #include "utils/dsa.h"
23 :
24 :
25 : /*
26 : * Types related to shared memory storage of statistics.
27 : *
28 : * Per-object statistics are stored in the "shared stats" hashtable. That
29 : * table's entries (PgStatShared_HashEntry) contain a pointer to the actual stats
30 : * data for the object (the size of the stats data varies depending on the
31 : * kind of stats). The table is keyed by PgStat_HashKey.
32 : *
33 : * Once a backend has a reference to a shared stats entry, it increments the
34 : * entry's refcount. Even after stats data is dropped (e.g., due to a DROP
35 : * TABLE), the entry itself can only be deleted once all references have been
36 : * released.
37 : *
38 : * These refcounts, in combination with a backend local hashtable
39 : * (pgStatEntryRefHash, with entries pointing to PgStat_EntryRef) in front of
40 : * the shared hash table, mean that most stats work can happen without
41 : * touching the shared hash table, reducing contention.
42 : *
43 : * Once there are pending stats updates for a table PgStat_EntryRef->pending
44 : * is allocated to contain a working space for as-of-yet-unapplied stats
45 : * updates. Once the stats are flushed, PgStat_EntryRef->pending is freed.
46 : *
47 : * Each stat kind in the shared hash table has a fixed member
48 : * PgStatShared_Common as the first element.
49 : */
50 :
51 : /*
52 : * Struct for shared statistics hash entry key.
53 : *
54 : * NB: We assume that this struct contains no padding. Also, 8 bytes
55 : * allocated for the object ID are good enough to ensure the uniqueness
56 : * of the hash key, hence the addition of new fields is not recommended.
57 : */
58 : typedef struct PgStat_HashKey
59 : {
60 : PgStat_Kind kind; /* statistics entry kind */
61 : Oid dboid; /* database ID. InvalidOid for shared objects. */
62 : uint64 objid; /* object ID (table, function, etc.), or
63 : * identifier. */
64 : } PgStat_HashKey;
65 :
66 : /*
67 : * Tracks if the stats file is being read, written or discarded, used in
68 : * combination with the finish callback.
69 : *
70 : * These states allow plugins that create auxiliary data files to determine
71 : * the current operation and perform any necessary file cleanup.
72 : */
73 : typedef enum PgStat_StatsFileOp
74 : {
75 : STATS_WRITE,
76 : STATS_READ,
77 : STATS_DISCARD,
78 : } PgStat_StatsFileOp;
79 :
80 : /*
81 : * PgStat_HashKey should not have any padding. Checking that the structure
82 : * size matches with the sum of each field is a check simple enough to
83 : * enforce this policy.
84 : */
85 : StaticAssertDecl((sizeof(PgStat_Kind) + sizeof(uint64) + sizeof(Oid)) ==
86 : sizeof(PgStat_HashKey),
87 : "PgStat_HashKey should have no padding");
88 :
89 : /*
90 : * Shared statistics hash entry. Doesn't itself contain any stats, but points
91 : * to them (with ->body). That allows the stats entries themselves to be of
92 : * variable size.
93 : */
94 : typedef struct PgStatShared_HashEntry
95 : {
96 : PgStat_HashKey key; /* hash key */
97 :
98 : /*
99 : * If dropped is set, backends need to release their references so that
100 : * the memory for the entry can be freed. No new references may be made
101 : * once marked as dropped.
102 : */
103 : bool dropped;
104 :
105 : /*
106 : * Refcount managing lifetime of the entry itself (as opposed to the
107 : * dshash entry pointing to it). The stats lifetime has to be separate
108 : * from the hash table entry lifetime because we allow backends to point
109 : * to a stats entry without holding a hash table lock (and some other
110 : * reasons).
111 : *
112 : * As long as the entry is not dropped, 1 is added to the refcount
113 : * representing that the entry should not be dropped. In addition each
114 : * backend that has a reference to the entry needs to increment the
115 : * refcount as long as it does.
116 : *
117 : * May only be incremented / decremented while holding at least a shared
118 : * lock on the dshash partition containing the entry. It needs to be an
119 : * atomic variable because multiple backends can increment the refcount
120 : * with just a shared lock.
121 : *
122 : * When the refcount reaches 0 the entry needs to be freed.
123 : */
124 : pg_atomic_uint32 refcount;
125 :
126 : /*
127 : * Counter tracking the number of times the entry has been reused.
128 : *
129 : * Set to 0 when the entry is created, and incremented by one each time
130 : * the shared entry is reinitialized with pgstat_reinit_entry().
131 : *
132 : * May only be incremented / decremented while holding at least a shared
133 : * lock on the dshash partition containing the entry. Like refcount, it
134 : * needs to be an atomic variable because multiple backends can increment
135 : * the generation with just a shared lock.
136 : */
137 : pg_atomic_uint32 generation;
138 :
139 : /*
140 : * Pointer to shared stats. The stats entry always starts with
141 : * PgStatShared_Common, embedded in a larger struct containing the
142 : * PgStat_Kind specific stats fields.
143 : */
144 : dsa_pointer body;
145 : } PgStatShared_HashEntry;
146 :
147 : /*
148 : * Common header struct for PgStatShared_*.
149 : */
150 : typedef struct PgStatShared_Common
151 : {
152 : uint32 magic; /* just a validity cross-check */
153 : /* lock protecting stats contents (i.e. data following the header) */
154 : LWLock lock;
155 : } PgStatShared_Common;
156 :
157 : /*
158 : * A backend local reference to a shared stats entry. As long as at least one
159 : * such reference exists, the shared stats entry will not be released.
160 : *
161 : * If there are pending stats update to the shared stats, these are stored in
162 : * ->pending.
163 : */
164 : typedef struct PgStat_EntryRef
165 : {
166 : /*
167 : * Pointer to the PgStatShared_HashEntry entry in the shared stats
168 : * hashtable.
169 : */
170 : PgStatShared_HashEntry *shared_entry;
171 :
172 : /*
173 : * Pointer to the stats data (i.e. PgStatShared_HashEntry->body), resolved
174 : * as a local pointer, to avoid repeated dsa_get_address() calls.
175 : */
176 : PgStatShared_Common *shared_stats;
177 :
178 : /*
179 : * Copy of PgStatShared_HashEntry->generation, keeping locally track of
180 : * the shared stats entry "generation" retrieved (number of times reused).
181 : */
182 : uint32 generation;
183 :
184 : /*
185 : * Pending statistics data that will need to be flushed to shared memory
186 : * stats eventually. Each stats kind utilizing pending data defines what
187 : * format its pending data has and needs to provide a
188 : * PgStat_KindInfo->flush_pending_cb callback to merge pending entries
189 : * into the shared stats hash table.
190 : */
191 : void *pending;
192 : dlist_node pending_node; /* membership in pgStatPending list */
193 : } PgStat_EntryRef;
194 :
195 :
196 : /*
197 : * Some stats changes are transactional. To maintain those, a stack of
198 : * PgStat_SubXactStatus entries is maintained, which contain data pertaining
199 : * to the current transaction and its active subtransactions.
200 : */
201 : typedef struct PgStat_SubXactStatus
202 : {
203 : int nest_level; /* subtransaction nest level */
204 :
205 : struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */
206 :
207 : /*
208 : * Statistics for transactionally dropped objects need to be
209 : * transactionally dropped as well. Collect the stats dropped in the
210 : * current (sub-)transaction and only execute the stats drop when we know
211 : * if the transaction commits/aborts. To handle replicas and crashes,
212 : * stats drops are included in commit / abort records.
213 : */
214 : dclist_head pending_drops;
215 :
216 : /*
217 : * Tuple insertion/deletion counts for an open transaction can't be
218 : * propagated into PgStat_TableStatus counters until we know if it is
219 : * going to commit or abort. Hence, we keep these counts in per-subxact
220 : * structs that live in TopTransactionContext. This data structure is
221 : * designed on the assumption that subxacts won't usually modify very many
222 : * tables.
223 : */
224 : PgStat_TableXactStatus *first; /* head of list for this subxact */
225 : } PgStat_SubXactStatus;
226 :
227 :
228 : /*
229 : * Metadata for a specific kind of statistics.
230 : */
231 : typedef struct PgStat_KindInfo
232 : {
233 : /*
234 : * Do a fixed number of stats objects exist for this kind of stats (e.g.
235 : * bgwriter stats) or not (e.g. tables).
236 : */
237 : bool fixed_amount:1;
238 :
239 : /*
240 : * Can stats of this kind be accessed from another database? Determines
241 : * whether a stats object gets included in stats snapshots.
242 : */
243 : bool accessed_across_databases:1;
244 :
245 : /* Should stats be written to the on-disk stats file? */
246 : bool write_to_file:1;
247 :
248 : /*
249 : * Should the number of entries be tracked? For variable-numbered stats,
250 : * to update its PgStat_ShmemControl.entry_counts.
251 : */
252 : bool track_entry_count:1;
253 :
254 : /*
255 : * The size of an entry in the shared stats hash table (pointed to by
256 : * PgStatShared_HashEntry->body). For fixed-numbered statistics, this is
257 : * the size of an entry in PgStat_ShmemControl->custom_data.
258 : */
259 : uint32 shared_size;
260 :
261 : /*
262 : * The offset of the statistics struct in the cached statistics snapshot
263 : * PgStat_Snapshot, for fixed-numbered statistics.
264 : */
265 : uint32 snapshot_ctl_off;
266 :
267 : /*
268 : * The offset of the statistics struct in the containing shared memory
269 : * control structure PgStat_ShmemControl, for fixed-numbered statistics.
270 : */
271 : uint32 shared_ctl_off;
272 :
273 : /*
274 : * The offset/size of statistics inside the shared stats entry. Used when
275 : * [de-]serializing statistics to / from disk respectively. Separate from
276 : * shared_size because [de-]serialization may not include in-memory state
277 : * like lwlocks.
278 : */
279 : uint32 shared_data_off;
280 : uint32 shared_data_len;
281 :
282 : /*
283 : * The size of the pending data for this kind. E.g. how large
284 : * PgStat_EntryRef->pending is. Used for allocations.
285 : *
286 : * 0 signals that an entry of this kind should never have a pending entry.
287 : */
288 : uint32 pending_size;
289 :
290 : /*
291 : * Perform custom actions when initializing a backend (standalone or under
292 : * postmaster). Optional.
293 : */
294 : void (*init_backend_cb) (void);
295 :
296 : /*
297 : * For variable-numbered stats: flush pending stats. Required if pending
298 : * data is used. See flush_static_cb when dealing with stats data that
299 : * that cannot use PgStat_EntryRef->pending.
300 : */
301 : bool (*flush_pending_cb) (PgStat_EntryRef *sr, bool nowait);
302 :
303 : /*
304 : * For variable-numbered stats: delete pending stats. Optional.
305 : */
306 : void (*delete_pending_cb) (PgStat_EntryRef *sr);
307 :
308 : /*
309 : * For variable-numbered stats: reset the reset timestamp. Optional.
310 : */
311 : void (*reset_timestamp_cb) (PgStatShared_Common *header, TimestampTz ts);
312 :
313 : /*
314 : * For variable-numbered stats. Optional.
315 : */
316 : void (*to_serialized_name) (const PgStat_HashKey *key,
317 : const PgStatShared_Common *header, NameData *name);
318 : bool (*from_serialized_name) (const NameData *name, PgStat_HashKey *key);
319 :
320 : /*
321 : * For variable-numbered stats: read or write additional data related to
322 : * an entry, in the stats file or optionally in a different file.
323 : * Optional.
324 : *
325 : * to_serialized_data: write auxiliary data for an entry.
326 : *
327 : * from_serialized_data: read auxiliary data for an entry. Returns true
328 : * on success, false on read error.
329 : *
330 : * "statfile" is a pointer to the on-disk stats file, named
331 : * PGSTAT_STAT_PERMANENT_FILENAME. "key" is the hash key of the entry
332 : * just written or read. "header" is a pointer to the stats data; it may
333 : * be modified only in from_serialized_data to reconstruct an entry.
334 : */
335 : void (*to_serialized_data) (const PgStat_HashKey *key,
336 : const PgStatShared_Common *header,
337 : FILE *statfile);
338 : bool (*from_serialized_data) (const PgStat_HashKey *key,
339 : PgStatShared_Common *header,
340 : FILE *statfile);
341 :
342 : /*
343 : * For fixed-numbered or variable-numbered statistics.
344 : *
345 : * Perform custom actions when done processing the on-disk stats file
346 : * after all the stats entries have been processed. Optional.
347 : *
348 : * "status" tracks the operation done for the on-disk stats file (read,
349 : * write, discard).
350 : */
351 : void (*finish) (PgStat_StatsFileOp status);
352 :
353 : /*
354 : * For fixed-numbered statistics: Initialize shared memory state.
355 : *
356 : * "stats" is the pointer to the allocated shared memory area.
357 : */
358 : void (*init_shmem_cb) (void *stats);
359 :
360 : /*
361 : * For fixed-numbered or variable-numbered statistics: Flush pending stats
362 : * entries, for stats kinds that do not use PgStat_EntryRef->pending.
363 : *
364 : * Returns true if some of the stats could not be flushed, due to lock
365 : * contention for example. Optional.
366 : *
367 : * "pgstat_report_fixed" needs to be set to trigger the flush of pending
368 : * stats.
369 : */
370 : bool (*flush_static_cb) (bool nowait);
371 :
372 : /*
373 : * For fixed-numbered statistics: Reset All.
374 : */
375 : void (*reset_all_cb) (TimestampTz ts);
376 :
377 : /*
378 : * For fixed-numbered statistics: Build snapshot for entry
379 : */
380 : void (*snapshot_cb) (void);
381 :
382 : /* name of the kind of stats */
383 : const char *const name;
384 : } PgStat_KindInfo;
385 :
386 :
387 : /*
388 : * List of SLRU names that we keep stats for. There is no central registry of
389 : * SLRUs, so we use this fixed list instead. The "other" entry is used for
390 : * all SLRUs without an explicit entry (e.g. SLRUs in extensions).
391 : *
392 : * This is only defined here so that SLRU_NUM_ELEMENTS is known for later type
393 : * definitions.
394 : */
395 : static const char *const slru_names[] = {
396 : "commit_timestamp",
397 : "multixact_member",
398 : "multixact_offset",
399 : "notify",
400 : "serializable",
401 : "subtransaction",
402 : "transaction",
403 : "other" /* has to be last */
404 : };
405 :
406 : #define SLRU_NUM_ELEMENTS lengthof(slru_names)
407 :
408 :
409 : /* ----------
410 : * Types and definitions for different kinds of fixed-amount stats.
411 : *
412 : * Single-writer stats use the changecount mechanism to achieve low-overhead
413 : * writes - they're obviously more performance critical than reads. Check the
414 : * definition of struct PgBackendStatus for some explanation of the
415 : * changecount mechanism.
416 : *
417 : * Because the obvious implementation of resetting single-writer stats isn't
418 : * compatible with that (another backend needs to write), we don't scribble on
419 : * shared stats while resetting. Instead, just record the current counter
420 : * values in a copy of the stats data, which is protected by ->lock. See
421 : * pgstat_fetch_stat_(archiver|bgwriter|checkpointer) for the reader side.
422 : *
423 : * The only exception to that is the stat_reset_timestamp in these structs,
424 : * which is protected by ->lock, because it has to be written by another
425 : * backend while resetting.
426 : * ----------
427 : */
428 :
429 : typedef struct PgStatShared_Archiver
430 : {
431 : /* lock protects ->reset_offset as well as stats->stat_reset_timestamp */
432 : LWLock lock;
433 : uint32 changecount;
434 : PgStat_ArchiverStats stats;
435 : PgStat_ArchiverStats reset_offset;
436 : } PgStatShared_Archiver;
437 :
438 : typedef struct PgStatShared_BgWriter
439 : {
440 : /* lock protects ->reset_offset as well as stats->stat_reset_timestamp */
441 : LWLock lock;
442 : uint32 changecount;
443 : PgStat_BgWriterStats stats;
444 : PgStat_BgWriterStats reset_offset;
445 : } PgStatShared_BgWriter;
446 :
447 : typedef struct PgStatShared_Checkpointer
448 : {
449 : /* lock protects ->reset_offset as well as stats->stat_reset_timestamp */
450 : LWLock lock;
451 : uint32 changecount;
452 : PgStat_CheckpointerStats stats;
453 : PgStat_CheckpointerStats reset_offset;
454 : } PgStatShared_Checkpointer;
455 :
456 : /* Shared-memory ready PgStat_IO */
457 : typedef struct PgStatShared_IO
458 : {
459 : /*
460 : * locks[i] protects stats.stats[i]. locks[0] also protects
461 : * stats.stat_reset_timestamp.
462 : */
463 : LWLock locks[BACKEND_NUM_TYPES];
464 : PgStat_IO stats;
465 : } PgStatShared_IO;
466 :
467 : typedef struct PgStatShared_SLRU
468 : {
469 : /* lock protects ->stats */
470 : LWLock lock;
471 : PgStat_SLRUStats stats[SLRU_NUM_ELEMENTS];
472 : } PgStatShared_SLRU;
473 :
474 : typedef struct PgStatShared_Wal
475 : {
476 : /* lock protects ->stats */
477 : LWLock lock;
478 : PgStat_WalStats stats;
479 : } PgStatShared_Wal;
480 :
481 :
482 :
483 : /* ----------
484 : * Types and definitions for different kinds of variable-amount stats.
485 : *
486 : * Each struct has to start with PgStatShared_Common, containing information
487 : * common across the different types of stats. Kind-specific data follows.
488 : * ----------
489 : */
490 :
491 : typedef struct PgStatShared_Database
492 : {
493 : PgStatShared_Common header;
494 : PgStat_StatDBEntry stats;
495 : } PgStatShared_Database;
496 :
497 : typedef struct PgStatShared_Relation
498 : {
499 : PgStatShared_Common header;
500 : PgStat_StatTabEntry stats;
501 : } PgStatShared_Relation;
502 :
503 : typedef struct PgStatShared_Function
504 : {
505 : PgStatShared_Common header;
506 : PgStat_StatFuncEntry stats;
507 : } PgStatShared_Function;
508 :
509 : typedef struct PgStatShared_Subscription
510 : {
511 : PgStatShared_Common header;
512 : PgStat_StatSubEntry stats;
513 : } PgStatShared_Subscription;
514 :
515 : typedef struct PgStatShared_ReplSlot
516 : {
517 : PgStatShared_Common header;
518 : PgStat_StatReplSlotEntry stats;
519 : } PgStatShared_ReplSlot;
520 :
521 : typedef struct PgStatShared_Backend
522 : {
523 : PgStatShared_Common header;
524 : PgStat_Backend stats;
525 : } PgStatShared_Backend;
526 :
527 : /*
528 : * Central shared memory entry for the cumulative stats system.
529 : *
530 : * Fixed amount stats, the dynamic shared memory hash table for
531 : * non-fixed-amount stats, as well as remaining bits and pieces are all
532 : * reached from here.
533 : */
534 : typedef struct PgStat_ShmemControl
535 : {
536 : void *raw_dsa_area;
537 :
538 : /*
539 : * Stats for variable-numbered objects are kept in this shared hash table.
540 : * See comment above PgStat_Kind for details.
541 : */
542 : dshash_table_handle hash_handle; /* shared dbstat hash */
543 :
544 : /* Has the stats system already been shut down? Just a debugging check. */
545 : bool is_shutdown;
546 :
547 : /*
548 : * Whenever statistics for dropped objects could not be freed - because
549 : * backends still have references - the dropping backend calls
550 : * pgstat_request_entry_refs_gc() incrementing this counter. Eventually
551 : * that causes backends to run pgstat_gc_entry_refs(), allowing memory to
552 : * be reclaimed.
553 : */
554 : pg_atomic_uint64 gc_request_count;
555 :
556 : /*
557 : * Counters for the number of entries associated to a single stats kind
558 : * that uses variable-numbered objects stored in the shared hash table.
559 : * These counters can be enabled on a per-kind basis, when
560 : * track_entry_count is set. This counter is incremented each time a new
561 : * entry is created (not reused) in the shared hash table, and is
562 : * decremented each time an entry is freed from the shared hash table.
563 : */
564 : pg_atomic_uint64 entry_counts[PGSTAT_KIND_MAX];
565 :
566 : /*
567 : * Stats data for fixed-numbered objects.
568 : */
569 : PgStatShared_Archiver archiver;
570 : PgStatShared_BgWriter bgwriter;
571 : PgStatShared_Checkpointer checkpointer;
572 : PgStatShared_IO io;
573 : PgStatShared_SLRU slru;
574 : PgStatShared_Wal wal;
575 :
576 : /*
577 : * Custom stats data with fixed-numbered objects, indexed by (PgStat_Kind
578 : * - PGSTAT_KIND_CUSTOM_MIN).
579 : */
580 : void *custom_data[PGSTAT_KIND_CUSTOM_SIZE];
581 :
582 : } PgStat_ShmemControl;
583 :
584 :
585 : /*
586 : * Cached statistics snapshot
587 : */
588 : typedef struct PgStat_Snapshot
589 : {
590 : PgStat_FetchConsistency mode;
591 :
592 : /* time at which snapshot was taken */
593 : TimestampTz snapshot_timestamp;
594 :
595 : bool fixed_valid[PGSTAT_KIND_BUILTIN_SIZE];
596 :
597 : PgStat_ArchiverStats archiver;
598 :
599 : PgStat_BgWriterStats bgwriter;
600 :
601 : PgStat_CheckpointerStats checkpointer;
602 :
603 : PgStat_IO io;
604 :
605 : PgStat_SLRUStats slru[SLRU_NUM_ELEMENTS];
606 :
607 : PgStat_WalStats wal;
608 :
609 : /*
610 : * Data in snapshot for custom fixed-numbered statistics, indexed by
611 : * (PgStat_Kind - PGSTAT_KIND_CUSTOM_MIN). Each entry is allocated in
612 : * TopMemoryContext, for a size of PgStat_KindInfo->shared_data_len.
613 : */
614 : bool custom_valid[PGSTAT_KIND_CUSTOM_SIZE];
615 : void *custom_data[PGSTAT_KIND_CUSTOM_SIZE];
616 :
617 : /* to free snapshot in bulk */
618 : MemoryContext context;
619 : struct pgstat_snapshot_hash *stats;
620 : } PgStat_Snapshot;
621 :
622 :
623 : /*
624 : * Collection of backend-local stats state.
625 : */
626 : typedef struct PgStat_LocalState
627 : {
628 : PgStat_ShmemControl *shmem;
629 : dsa_area *dsa;
630 : dshash_table *shared_hash;
631 :
632 : /* the current statistics snapshot */
633 : PgStat_Snapshot snapshot;
634 : } PgStat_LocalState;
635 :
636 :
637 : /*
638 : * Inline functions defined further below.
639 : */
640 :
641 : static inline void pgstat_begin_changecount_write(uint32 *cc);
642 : static inline void pgstat_end_changecount_write(uint32 *cc);
643 : static inline uint32 pgstat_begin_changecount_read(uint32 *cc);
644 : static inline bool pgstat_end_changecount_read(uint32 *cc, uint32 cc_before);
645 :
646 : static inline void pgstat_copy_changecounted_stats(void *dst, void *src, size_t len,
647 : uint32 *cc);
648 :
649 : static inline int pgstat_cmp_hash_key(const void *a, const void *b, size_t size, void *arg);
650 : static inline uint32 pgstat_hash_hash_key(const void *d, size_t size, void *arg);
651 : static inline size_t pgstat_get_entry_len(PgStat_Kind kind);
652 : static inline void *pgstat_get_entry_data(PgStat_Kind kind, PgStatShared_Common *entry);
653 : static inline void *pgstat_get_custom_shmem_data(PgStat_Kind kind);
654 : static inline void *pgstat_get_custom_snapshot_data(PgStat_Kind kind);
655 :
656 :
657 : /*
658 : * Functions in pgstat.c
659 : */
660 :
661 : extern const PgStat_KindInfo *pgstat_get_kind_info(PgStat_Kind kind);
662 : extern void pgstat_register_kind(PgStat_Kind kind,
663 : const PgStat_KindInfo *kind_info);
664 :
665 : #ifdef USE_ASSERT_CHECKING
666 : extern void pgstat_assert_is_up(void);
667 : #else
668 : #define pgstat_assert_is_up() ((void)true)
669 : #endif
670 :
671 : extern void pgstat_delete_pending_entry(PgStat_EntryRef *entry_ref);
672 : extern PgStat_EntryRef *pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid,
673 : uint64 objid,
674 : bool *created_entry);
675 : extern PgStat_EntryRef *pgstat_fetch_pending_entry(PgStat_Kind kind,
676 : Oid dboid, uint64 objid);
677 :
678 : extern void *pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, uint64 objid);
679 : extern void pgstat_snapshot_fixed(PgStat_Kind kind);
680 :
681 :
682 : /*
683 : * Functions in pgstat_archiver.c
684 : */
685 :
686 : extern void pgstat_archiver_init_shmem_cb(void *stats);
687 : extern void pgstat_archiver_reset_all_cb(TimestampTz ts);
688 : extern void pgstat_archiver_snapshot_cb(void);
689 :
690 : /*
691 : * Functions in pgstat_backend.c
692 : */
693 :
694 : /* flags for pgstat_flush_backend() */
695 : #define PGSTAT_BACKEND_FLUSH_IO (1 << 0) /* Flush I/O statistics */
696 : #define PGSTAT_BACKEND_FLUSH_WAL (1 << 1) /* Flush WAL statistics */
697 : #define PGSTAT_BACKEND_FLUSH_ALL (PGSTAT_BACKEND_FLUSH_IO | PGSTAT_BACKEND_FLUSH_WAL)
698 :
699 : extern bool pgstat_flush_backend(bool nowait, bits32 flags);
700 : extern bool pgstat_backend_flush_cb(bool nowait);
701 : extern void pgstat_backend_reset_timestamp_cb(PgStatShared_Common *header,
702 : TimestampTz ts);
703 :
704 : /*
705 : * Functions in pgstat_bgwriter.c
706 : */
707 :
708 : extern void pgstat_bgwriter_init_shmem_cb(void *stats);
709 : extern void pgstat_bgwriter_reset_all_cb(TimestampTz ts);
710 : extern void pgstat_bgwriter_snapshot_cb(void);
711 :
712 :
713 : /*
714 : * Functions in pgstat_checkpointer.c
715 : */
716 :
717 : extern void pgstat_checkpointer_init_shmem_cb(void *stats);
718 : extern void pgstat_checkpointer_reset_all_cb(TimestampTz ts);
719 : extern void pgstat_checkpointer_snapshot_cb(void);
720 :
721 :
722 : /*
723 : * Functions in pgstat_database.c
724 : */
725 :
726 : extern void pgstat_report_disconnect(Oid dboid);
727 : extern void pgstat_update_dbstats(TimestampTz ts);
728 : extern void AtEOXact_PgStat_Database(bool isCommit, bool parallel);
729 :
730 : extern PgStat_StatDBEntry *pgstat_prep_database_pending(Oid dboid);
731 : extern void pgstat_reset_database_timestamp(Oid dboid, TimestampTz ts);
732 : extern bool pgstat_database_flush_cb(PgStat_EntryRef *entry_ref, bool nowait);
733 : extern void pgstat_database_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts);
734 :
735 :
736 : /*
737 : * Functions in pgstat_function.c
738 : */
739 :
740 : extern bool pgstat_function_flush_cb(PgStat_EntryRef *entry_ref, bool nowait);
741 : extern void pgstat_function_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts);
742 :
743 :
744 : /*
745 : * Functions in pgstat_io.c
746 : */
747 :
748 : extern void pgstat_flush_io(bool nowait);
749 :
750 : extern bool pgstat_io_flush_cb(bool nowait);
751 : extern void pgstat_io_init_shmem_cb(void *stats);
752 : extern void pgstat_io_reset_all_cb(TimestampTz ts);
753 : extern void pgstat_io_snapshot_cb(void);
754 :
755 :
756 : /*
757 : * Functions in pgstat_relation.c
758 : */
759 :
760 : extern void AtEOXact_PgStat_Relations(PgStat_SubXactStatus *xact_state, bool isCommit);
761 : extern void AtEOSubXact_PgStat_Relations(PgStat_SubXactStatus *xact_state, bool isCommit, int nestDepth);
762 : extern void AtPrepare_PgStat_Relations(PgStat_SubXactStatus *xact_state);
763 : extern void PostPrepare_PgStat_Relations(PgStat_SubXactStatus *xact_state);
764 :
765 : extern bool pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait);
766 : extern void pgstat_relation_delete_pending_cb(PgStat_EntryRef *entry_ref);
767 : extern void pgstat_relation_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts);
768 :
769 :
770 : /*
771 : * Functions in pgstat_replslot.c
772 : */
773 :
774 : extern void pgstat_replslot_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts);
775 : extern void pgstat_replslot_to_serialized_name_cb(const PgStat_HashKey *key, const PgStatShared_Common *header, NameData *name);
776 : extern bool pgstat_replslot_from_serialized_name_cb(const NameData *name, PgStat_HashKey *key);
777 :
778 :
779 : /*
780 : * Functions in pgstat_shmem.c
781 : */
782 :
783 : extern void pgstat_attach_shmem(void);
784 : extern void pgstat_detach_shmem(void);
785 :
786 : extern PgStat_EntryRef *pgstat_get_entry_ref(PgStat_Kind kind, Oid dboid, uint64 objid,
787 : bool create, bool *created_entry);
788 : extern bool pgstat_lock_entry(PgStat_EntryRef *entry_ref, bool nowait);
789 : extern bool pgstat_lock_entry_shared(PgStat_EntryRef *entry_ref, bool nowait);
790 : extern void pgstat_unlock_entry(PgStat_EntryRef *entry_ref);
791 : extern bool pgstat_drop_entry(PgStat_Kind kind, Oid dboid, uint64 objid);
792 : extern void pgstat_drop_all_entries(void);
793 : extern void pgstat_drop_matching_entries(bool (*do_drop) (PgStatShared_HashEntry *, Datum),
794 : Datum match_data);
795 : extern PgStat_EntryRef *pgstat_get_entry_ref_locked(PgStat_Kind kind, Oid dboid, uint64 objid,
796 : bool nowait);
797 : extern void pgstat_reset_entry(PgStat_Kind kind, Oid dboid, uint64 objid, TimestampTz ts);
798 : extern void pgstat_reset_entries_of_kind(PgStat_Kind kind, TimestampTz ts);
799 : extern void pgstat_reset_matching_entries(bool (*do_reset) (PgStatShared_HashEntry *, Datum),
800 : Datum match_data,
801 : TimestampTz ts);
802 :
803 : extern void pgstat_request_entry_refs_gc(void);
804 : extern PgStatShared_Common *pgstat_init_entry(PgStat_Kind kind,
805 : PgStatShared_HashEntry *shhashent);
806 :
807 :
808 : /*
809 : * Functions in pgstat_slru.c
810 : */
811 :
812 : extern bool pgstat_slru_flush_cb(bool nowait);
813 : extern void pgstat_slru_init_shmem_cb(void *stats);
814 : extern void pgstat_slru_reset_all_cb(TimestampTz ts);
815 : extern void pgstat_slru_snapshot_cb(void);
816 :
817 :
818 : /*
819 : * Functions in pgstat_wal.c
820 : */
821 :
822 : extern void pgstat_wal_init_backend_cb(void);
823 : extern bool pgstat_wal_flush_cb(bool nowait);
824 : extern void pgstat_wal_init_shmem_cb(void *stats);
825 : extern void pgstat_wal_reset_all_cb(TimestampTz ts);
826 : extern void pgstat_wal_snapshot_cb(void);
827 :
828 :
829 : /*
830 : * Functions in pgstat_subscription.c
831 : */
832 :
833 : extern bool pgstat_subscription_flush_cb(PgStat_EntryRef *entry_ref, bool nowait);
834 : extern void pgstat_subscription_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts);
835 :
836 :
837 : /*
838 : * Functions in pgstat_xact.c
839 : */
840 :
841 : extern PgStat_SubXactStatus *pgstat_get_xact_stack_level(int nest_level);
842 : extern void pgstat_drop_transactional(PgStat_Kind kind, Oid dboid, uint64 objid);
843 : extern void pgstat_create_transactional(PgStat_Kind kind, Oid dboid, uint64 objid);
844 :
845 :
846 : /*
847 : * Variables in pgstat.c
848 : */
849 :
850 : /*
851 : * Track if *any* pending fixed-numbered statistics should be flushed to
852 : * shared memory.
853 : *
854 : * This flag can be switched to true by fixed-numbered statistics to let
855 : * pgstat_report_stat() know if it needs to go through one round of
856 : * reports, calling flush_static_cb for each fixed-numbered statistics
857 : * kind. When this flag is not set, pgstat_report_stat() is able to do
858 : * a fast exit, knowing that there are no pending fixed-numbered statistics.
859 : *
860 : * Statistics callbacks should never reset this flag; pgstat_report_stat()
861 : * is in charge of doing that.
862 : */
863 : extern PGDLLIMPORT bool pgstat_report_fixed;
864 :
865 : /* Backend-local stats state */
866 : extern PGDLLIMPORT PgStat_LocalState pgStatLocal;
867 :
868 : /* Helper functions for reading and writing of on-disk stats file */
869 : extern void pgstat_write_chunk(FILE *fpout, void *ptr, size_t len);
870 : extern bool pgstat_read_chunk(FILE *fpin, void *ptr, size_t len);
871 : #define pgstat_read_chunk_s(fpin, ptr) pgstat_read_chunk(fpin, ptr, sizeof(*ptr))
872 : #define pgstat_write_chunk_s(fpout, ptr) pgstat_write_chunk(fpout, ptr, sizeof(*ptr))
873 :
874 : /*
875 : * Implementation of inline functions declared above.
876 : */
877 :
878 : /*
879 : * Helpers for changecount manipulation. See comments around struct
880 : * PgBackendStatus for details.
881 : */
882 :
883 : static inline void
884 24310 : pgstat_begin_changecount_write(uint32 *cc)
885 : {
886 : Assert((*cc & 1) == 0);
887 :
888 24310 : START_CRIT_SECTION();
889 24310 : (*cc)++;
890 24310 : pg_write_barrier();
891 24310 : }
892 :
893 : static inline void
894 24310 : pgstat_end_changecount_write(uint32 *cc)
895 : {
896 : Assert((*cc & 1) == 1);
897 :
898 24310 : pg_write_barrier();
899 :
900 24310 : (*cc)++;
901 :
902 24310 : END_CRIT_SECTION();
903 24310 : }
904 :
905 : static inline uint32
906 5838 : pgstat_begin_changecount_read(uint32 *cc)
907 : {
908 5838 : uint32 before_cc = *cc;
909 :
910 5838 : CHECK_FOR_INTERRUPTS();
911 :
912 5838 : pg_read_barrier();
913 :
914 5838 : return before_cc;
915 : }
916 :
917 : /*
918 : * Returns true if the read succeeded, false if it needs to be repeated.
919 : */
920 : static inline bool
921 5838 : pgstat_end_changecount_read(uint32 *cc, uint32 before_cc)
922 : {
923 : uint32 after_cc;
924 :
925 5838 : pg_read_barrier();
926 :
927 5838 : after_cc = *cc;
928 :
929 : /* was a write in progress when we started? */
930 5838 : if (before_cc & 1)
931 0 : return false;
932 :
933 : /* did writes start and complete while we read? */
934 5838 : return before_cc == after_cc;
935 : }
936 :
937 :
938 : /*
939 : * helper function for PgStat_KindInfo->snapshot_cb
940 : * PgStat_KindInfo->reset_all_cb callbacks.
941 : *
942 : * Copies out the specified memory area following change-count protocol.
943 : */
944 : static inline void
945 5838 : pgstat_copy_changecounted_stats(void *dst, void *src, size_t len,
946 : uint32 *cc)
947 : {
948 : uint32 cc_before;
949 :
950 : do
951 : {
952 5838 : cc_before = pgstat_begin_changecount_read(cc);
953 :
954 5838 : memcpy(dst, src, len);
955 : }
956 5838 : while (!pgstat_end_changecount_read(cc, cc_before));
957 5838 : }
958 :
959 : /* helpers for dshash / simplehash hashtables */
960 : static inline int
961 13353202 : pgstat_cmp_hash_key(const void *a, const void *b, size_t size, void *arg)
962 : {
963 : Assert(size == sizeof(PgStat_HashKey) && arg == NULL);
964 13353202 : return memcmp(a, b, sizeof(PgStat_HashKey));
965 : }
966 :
967 : static inline uint32
968 16090552 : pgstat_hash_hash_key(const void *d, size_t size, void *arg)
969 : {
970 16090552 : const char *key = (const char *) d;
971 :
972 : Assert(size == sizeof(PgStat_HashKey) && arg == NULL);
973 16090552 : return fasthash32(key, size, 0);
974 : }
975 :
976 : /*
977 : * The length of the data portion of a shared memory stats entry (i.e. without
978 : * transient data such as refcounts, lwlocks, ...).
979 : */
980 : static inline size_t
981 925110 : pgstat_get_entry_len(PgStat_Kind kind)
982 : {
983 925110 : return pgstat_get_kind_info(kind)->shared_data_len;
984 : }
985 :
986 : /*
987 : * Returns a pointer to the data portion of a shared memory stats entry.
988 : */
989 : static inline void *
990 1362968 : pgstat_get_entry_data(PgStat_Kind kind, PgStatShared_Common *entry)
991 : {
992 1362968 : size_t off = pgstat_get_kind_info(kind)->shared_data_off;
993 :
994 : Assert(off != 0 && off < PG_UINT32_MAX);
995 :
996 1362968 : return ((char *) (entry)) + off;
997 : }
998 :
999 : /*
1000 : * Returns the number of entries counted for a stats kind.
1001 : */
1002 : static inline uint64
1003 : pgstat_get_entry_count(PgStat_Kind kind)
1004 : {
1005 : Assert(pgstat_get_kind_info(kind)->track_entry_count);
1006 :
1007 : return pg_atomic_read_u64(&pgStatLocal.shmem->entry_counts[kind - 1]);
1008 : }
1009 :
1010 : /*
1011 : * Returns a pointer to the shared memory area of custom stats for
1012 : * fixed-numbered statistics.
1013 : */
1014 : static inline void *
1015 20 : pgstat_get_custom_shmem_data(PgStat_Kind kind)
1016 : {
1017 20 : int idx = kind - PGSTAT_KIND_CUSTOM_MIN;
1018 :
1019 : Assert(pgstat_is_kind_custom(kind));
1020 : Assert(pgstat_get_kind_info(kind)->fixed_amount);
1021 :
1022 20 : return pgStatLocal.shmem->custom_data[idx];
1023 : }
1024 :
1025 : /*
1026 : * Returns a pointer to the portion of custom data for fixed-numbered
1027 : * statistics in the current snapshot.
1028 : */
1029 : static inline void *
1030 10 : pgstat_get_custom_snapshot_data(PgStat_Kind kind)
1031 : {
1032 10 : int idx = kind - PGSTAT_KIND_CUSTOM_MIN;
1033 :
1034 : Assert(pgstat_is_kind_custom(kind));
1035 : Assert(pgstat_get_kind_info(kind)->fixed_amount);
1036 :
1037 10 : return pgStatLocal.snapshot.custom_data[idx];
1038 : }
1039 :
1040 : #endif /* PGSTAT_INTERNAL_H */
|
