Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * slot.c
4 : * Replication slot management.
5 : *
6 : *
7 : * Copyright (c) 2012-2026, PostgreSQL Global Development Group
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/replication/slot.c
12 : *
13 : * NOTES
14 : *
15 : * Replication slots are used to keep state about replication streams
16 : * originating from this cluster. Their primary purpose is to prevent the
17 : * premature removal of WAL or of old tuple versions in a manner that would
18 : * interfere with replication; they are also useful for monitoring purposes.
19 : * Slots need to be permanent (to allow restarts), crash-safe, and allocatable
20 : * on standbys (to support cascading setups). The requirement that slots be
21 : * usable on standbys precludes storing them in the system catalogs.
22 : *
23 : * Each replication slot gets its own directory inside the directory
24 : * $PGDATA / PG_REPLSLOT_DIR. Inside that directory the state file will
25 : * contain the slot's own data. Additional data can be stored alongside that
26 : * file if required. While the server is running, the state data is also
27 : * cached in memory for efficiency.
28 : *
29 : * ReplicationSlotAllocationLock must be taken in exclusive mode to allocate
30 : * or free a slot. ReplicationSlotControlLock must be taken in shared mode
31 : * to iterate over the slots, and in exclusive mode to change the in_use flag
32 : * of a slot. The remaining data in each slot is protected by its mutex.
33 : *
34 : *-------------------------------------------------------------------------
35 : */
36 :
37 : #include "postgres.h"
38 :
39 : #include <unistd.h>
40 : #include <sys/stat.h>
41 :
42 : #include "access/transam.h"
43 : #include "access/xlog_internal.h"
44 : #include "access/xlogrecovery.h"
45 : #include "common/file_utils.h"
46 : #include "common/string.h"
47 : #include "miscadmin.h"
48 : #include "pgstat.h"
49 : #include "postmaster/interrupt.h"
50 : #include "replication/logicallauncher.h"
51 : #include "replication/slotsync.h"
52 : #include "replication/slot.h"
53 : #include "replication/walsender_private.h"
54 : #include "storage/fd.h"
55 : #include "storage/ipc.h"
56 : #include "storage/proc.h"
57 : #include "storage/procarray.h"
58 : #include "utils/builtins.h"
59 : #include "utils/guc_hooks.h"
60 : #include "utils/injection_point.h"
61 : #include "utils/varlena.h"
62 :
63 : /*
64 : * Replication slot on-disk data structure.
65 : */
66 : typedef struct ReplicationSlotOnDisk
67 : {
68 : /* first part of this struct needs to be version independent */
69 :
70 : /* data not covered by checksum */
71 : uint32 magic;
72 : pg_crc32c checksum;
73 :
74 : /* data covered by checksum */
75 : uint32 version;
76 : uint32 length;
77 :
78 : /*
79 : * The actual data in the slot that follows can differ based on the above
80 : * 'version'.
81 : */
82 :
83 : ReplicationSlotPersistentData slotdata;
84 : } ReplicationSlotOnDisk;
85 :
86 : /*
87 : * Struct for the configuration of synchronized_standby_slots.
88 : *
89 : * Note: this must be a flat representation that can be held in a single chunk
90 : * of guc_malloc'd memory, so that it can be stored as the "extra" data for the
91 : * synchronized_standby_slots GUC.
92 : */
93 : typedef struct
94 : {
95 : /* Number of slot names in the slot_names[] */
96 : int nslotnames;
97 :
98 : /*
99 : * slot_names contains 'nslotnames' consecutive null-terminated C strings.
100 : */
101 : char slot_names[FLEXIBLE_ARRAY_MEMBER];
102 : } SyncStandbySlotsConfigData;
103 :
104 : /*
105 : * Lookup table for slot invalidation causes.
106 : */
107 : typedef struct SlotInvalidationCauseMap
108 : {
109 : ReplicationSlotInvalidationCause cause;
110 : const char *cause_name;
111 : } SlotInvalidationCauseMap;
112 :
113 : static const SlotInvalidationCauseMap SlotInvalidationCauses[] = {
114 : {RS_INVAL_NONE, "none"},
115 : {RS_INVAL_WAL_REMOVED, "wal_removed"},
116 : {RS_INVAL_HORIZON, "rows_removed"},
117 : {RS_INVAL_WAL_LEVEL, "wal_level_insufficient"},
118 : {RS_INVAL_IDLE_TIMEOUT, "idle_timeout"},
119 : };
120 :
121 : /*
122 : * Ensure that the lookup table is up-to-date with the enums defined in
123 : * ReplicationSlotInvalidationCause.
124 : */
125 : StaticAssertDecl(lengthof(SlotInvalidationCauses) == (RS_INVAL_MAX_CAUSES + 1),
126 : "array length mismatch");
127 :
128 : /* size of version independent data */
129 : #define ReplicationSlotOnDiskConstantSize \
130 : offsetof(ReplicationSlotOnDisk, slotdata)
131 : /* size of the part of the slot not covered by the checksum */
132 : #define ReplicationSlotOnDiskNotChecksummedSize \
133 : offsetof(ReplicationSlotOnDisk, version)
134 : /* size of the part covered by the checksum */
135 : #define ReplicationSlotOnDiskChecksummedSize \
136 : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskNotChecksummedSize
137 : /* size of the slot data that is version dependent */
138 : #define ReplicationSlotOnDiskV2Size \
139 : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskConstantSize
140 :
141 : #define SLOT_MAGIC 0x1051CA1 /* format identifier */
142 : #define SLOT_VERSION 5 /* version for new files */
143 :
144 : /* Control array for replication slot management */
145 : ReplicationSlotCtlData *ReplicationSlotCtl = NULL;
146 :
147 : /* My backend's replication slot in the shared memory array */
148 : ReplicationSlot *MyReplicationSlot = NULL;
149 :
150 : /* GUC variables */
151 : int max_replication_slots = 10; /* the maximum number of replication
152 : * slots */
153 :
154 : /*
155 : * Invalidate replication slots that have remained idle longer than this
156 : * duration; '0' disables it.
157 : */
158 : int idle_replication_slot_timeout_secs = 0;
159 :
160 : /*
161 : * This GUC lists streaming replication standby server slot names that
162 : * logical WAL sender processes will wait for.
163 : */
164 : char *synchronized_standby_slots;
165 :
166 : /* This is the parsed and cached configuration for synchronized_standby_slots */
167 : static SyncStandbySlotsConfigData *synchronized_standby_slots_config;
168 :
169 : /*
170 : * Oldest LSN that has been confirmed to be flushed to the standbys
171 : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
172 : */
173 : static XLogRecPtr ss_oldest_flush_lsn = InvalidXLogRecPtr;
174 :
175 : static void ReplicationSlotShmemExit(int code, Datum arg);
176 : static bool IsSlotForConflictCheck(const char *name);
177 : static void ReplicationSlotDropPtr(ReplicationSlot *slot);
178 :
179 : /* internal persistency functions */
180 : static void RestoreSlotFromDisk(const char *name);
181 : static void CreateSlotOnDisk(ReplicationSlot *slot);
182 : static void SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel);
183 :
184 : /*
185 : * Report shared-memory space needed by ReplicationSlotsShmemInit.
186 : */
187 : Size
188 8778 : ReplicationSlotsShmemSize(void)
189 : {
190 8778 : Size size = 0;
191 :
192 8778 : if (max_replication_slots == 0)
193 4 : return size;
194 :
195 8774 : size = offsetof(ReplicationSlotCtlData, replication_slots);
196 8774 : size = add_size(size,
197 : mul_size(max_replication_slots, sizeof(ReplicationSlot)));
198 :
199 8774 : return size;
200 : }
201 :
202 : /*
203 : * Allocate and initialize shared memory for replication slots.
204 : */
205 : void
206 2272 : ReplicationSlotsShmemInit(void)
207 : {
208 : bool found;
209 :
210 2272 : if (max_replication_slots == 0)
211 2 : return;
212 :
213 2270 : ReplicationSlotCtl = (ReplicationSlotCtlData *)
214 2270 : ShmemInitStruct("ReplicationSlot Ctl", ReplicationSlotsShmemSize(),
215 : &found);
216 :
217 2270 : if (!found)
218 : {
219 : int i;
220 :
221 : /* First time through, so initialize */
222 4120 : MemSet(ReplicationSlotCtl, 0, ReplicationSlotsShmemSize());
223 :
224 24576 : for (i = 0; i < max_replication_slots; i++)
225 : {
226 22306 : ReplicationSlot *slot = &ReplicationSlotCtl->replication_slots[i];
227 :
228 : /* everything else is zeroed by the memset above */
229 22306 : SpinLockInit(&slot->mutex);
230 22306 : LWLockInitialize(&slot->io_in_progress_lock,
231 : LWTRANCHE_REPLICATION_SLOT_IO);
232 22306 : ConditionVariableInit(&slot->active_cv);
233 : }
234 : }
235 : }
236 :
237 : /*
238 : * Register the callback for replication slot cleanup and releasing.
239 : */
240 : void
241 45684 : ReplicationSlotInitialize(void)
242 : {
243 45684 : before_shmem_exit(ReplicationSlotShmemExit, 0);
244 45684 : }
245 :
246 : /*
247 : * Release and cleanup replication slots.
248 : */
249 : static void
250 45684 : ReplicationSlotShmemExit(int code, Datum arg)
251 : {
252 : /* Make sure active replication slots are released */
253 45684 : if (MyReplicationSlot != NULL)
254 516 : ReplicationSlotRelease();
255 :
256 : /* Also cleanup all the temporary slots. */
257 45684 : ReplicationSlotCleanup(false);
258 45684 : }
259 :
260 : /*
261 : * Check whether the passed slot name is valid and report errors at elevel.
262 : *
263 : * See comments for ReplicationSlotValidateNameInternal().
264 : */
265 : bool
266 1624 : ReplicationSlotValidateName(const char *name, bool allow_reserved_name,
267 : int elevel)
268 : {
269 : int err_code;
270 1624 : char *err_msg = NULL;
271 1624 : char *err_hint = NULL;
272 :
273 1624 : if (!ReplicationSlotValidateNameInternal(name, allow_reserved_name,
274 : &err_code, &err_msg, &err_hint))
275 : {
276 : /*
277 : * Use errmsg_internal() and errhint_internal() instead of errmsg()
278 : * and errhint(), since the messages from
279 : * ReplicationSlotValidateNameInternal() are already translated. This
280 : * avoids double translation.
281 : */
282 8 : ereport(elevel,
283 : errcode(err_code),
284 : errmsg_internal("%s", err_msg),
285 : (err_hint != NULL) ? errhint_internal("%s", err_hint) : 0);
286 :
287 0 : pfree(err_msg);
288 0 : if (err_hint != NULL)
289 0 : pfree(err_hint);
290 0 : return false;
291 : }
292 :
293 1616 : return true;
294 : }
295 :
296 : /*
297 : * Check whether the passed slot name is valid.
298 : *
299 : * An error will be reported for a reserved replication slot name if
300 : * allow_reserved_name is set to false.
301 : *
302 : * Slot names may consist out of [a-z0-9_]{1,NAMEDATALEN-1} which should allow
303 : * the name to be used as a directory name on every supported OS.
304 : *
305 : * Returns true if the slot name is valid. Otherwise, returns false and stores
306 : * the error code, error message, and optional hint in err_code, err_msg, and
307 : * err_hint, respectively. The caller is responsible for freeing err_msg and
308 : * err_hint, which are palloc'd.
309 : */
310 : bool
311 2062 : ReplicationSlotValidateNameInternal(const char *name, bool allow_reserved_name,
312 : int *err_code, char **err_msg, char **err_hint)
313 : {
314 : const char *cp;
315 :
316 2062 : if (strlen(name) == 0)
317 : {
318 6 : *err_code = ERRCODE_INVALID_NAME;
319 6 : *err_msg = psprintf(_("replication slot name \"%s\" is too short"), name);
320 6 : *err_hint = NULL;
321 6 : return false;
322 : }
323 :
324 2056 : if (strlen(name) >= NAMEDATALEN)
325 : {
326 0 : *err_code = ERRCODE_NAME_TOO_LONG;
327 0 : *err_msg = psprintf(_("replication slot name \"%s\" is too long"), name);
328 0 : *err_hint = NULL;
329 0 : return false;
330 : }
331 :
332 39760 : for (cp = name; *cp; cp++)
333 : {
334 37708 : if (!((*cp >= 'a' && *cp <= 'z')
335 17794 : || (*cp >= '0' && *cp <= '9')
336 3662 : || (*cp == '_')))
337 : {
338 4 : *err_code = ERRCODE_INVALID_NAME;
339 4 : *err_msg = psprintf(_("replication slot name \"%s\" contains invalid character"), name);
340 4 : *err_hint = psprintf(_("Replication slot names may only contain lower case letters, numbers, and the underscore character."));
341 4 : return false;
342 : }
343 : }
344 :
345 2052 : if (!allow_reserved_name && IsSlotForConflictCheck(name))
346 : {
347 2 : *err_code = ERRCODE_RESERVED_NAME;
348 2 : *err_msg = psprintf(_("replication slot name \"%s\" is reserved"), name);
349 2 : *err_hint = psprintf(_("The name \"%s\" is reserved for the conflict detection slot."),
350 : CONFLICT_DETECTION_SLOT);
351 2 : return false;
352 : }
353 :
354 2050 : return true;
355 : }
356 :
357 : /*
358 : * Return true if the replication slot name is "pg_conflict_detection".
359 : */
360 : static bool
361 4410 : IsSlotForConflictCheck(const char *name)
362 : {
363 4410 : return (strcmp(name, CONFLICT_DETECTION_SLOT) == 0);
364 : }
365 :
366 : /*
367 : * Create a new replication slot and mark it as used by this backend.
368 : *
369 : * name: Name of the slot
370 : * db_specific: logical decoding is db specific; if the slot is going to
371 : * be used for that pass true, otherwise false.
372 : * two_phase: If enabled, allows decoding of prepared transactions.
373 : * failover: If enabled, allows the slot to be synced to standbys so
374 : * that logical replication can be resumed after failover.
375 : * synced: True if the slot is synchronized from the primary server.
376 : */
377 : void
378 1358 : ReplicationSlotCreate(const char *name, bool db_specific,
379 : ReplicationSlotPersistency persistency,
380 : bool two_phase, bool failover, bool synced)
381 : {
382 1358 : ReplicationSlot *slot = NULL;
383 : int i;
384 :
385 : Assert(MyReplicationSlot == NULL);
386 :
387 : /*
388 : * The logical launcher or pg_upgrade may create or migrate an internal
389 : * slot, so using a reserved name is allowed in these cases.
390 : */
391 1358 : ReplicationSlotValidateName(name, IsBinaryUpgrade || IsLogicalLauncher(),
392 : ERROR);
393 :
394 1356 : if (failover)
395 : {
396 : /*
397 : * Do not allow users to create the failover enabled slots on the
398 : * standby as we do not support sync to the cascading standby.
399 : *
400 : * However, failover enabled slots can be created during slot
401 : * synchronization because we need to retain the same values as the
402 : * remote slot.
403 : */
404 52 : if (RecoveryInProgress() && !IsSyncingReplicationSlots())
405 0 : ereport(ERROR,
406 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
407 : errmsg("cannot enable failover for a replication slot created on the standby"));
408 :
409 : /*
410 : * Do not allow users to create failover enabled temporary slots,
411 : * because temporary slots will not be synced to the standby.
412 : *
413 : * However, failover enabled temporary slots can be created during
414 : * slot synchronization. See the comments atop slotsync.c for details.
415 : */
416 52 : if (persistency == RS_TEMPORARY && !IsSyncingReplicationSlots())
417 2 : ereport(ERROR,
418 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
419 : errmsg("cannot enable failover for a temporary replication slot"));
420 : }
421 :
422 : /*
423 : * If some other backend ran this code concurrently with us, we'd likely
424 : * both allocate the same slot, and that would be bad. We'd also be at
425 : * risk of missing a name collision. Also, we don't want to try to create
426 : * a new slot while somebody's busy cleaning up an old one, because we
427 : * might both be monkeying with the same directory.
428 : */
429 1354 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
430 :
431 : /*
432 : * Check for name collision, and identify an allocatable slot. We need to
433 : * hold ReplicationSlotControlLock in shared mode for this, so that nobody
434 : * else can change the in_use flags while we're looking at them.
435 : */
436 1354 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
437 13088 : for (i = 0; i < max_replication_slots; i++)
438 : {
439 11740 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
440 :
441 11740 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
442 6 : ereport(ERROR,
443 : (errcode(ERRCODE_DUPLICATE_OBJECT),
444 : errmsg("replication slot \"%s\" already exists", name)));
445 11734 : if (!s->in_use && slot == NULL)
446 1346 : slot = s;
447 : }
448 1348 : LWLockRelease(ReplicationSlotControlLock);
449 :
450 : /* If all slots are in use, we're out of luck. */
451 1348 : if (slot == NULL)
452 2 : ereport(ERROR,
453 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
454 : errmsg("all replication slots are in use"),
455 : errhint("Free one or increase \"max_replication_slots\".")));
456 :
457 : /*
458 : * Since this slot is not in use, nobody should be looking at any part of
459 : * it other than the in_use field unless they're trying to allocate it.
460 : * And since we hold ReplicationSlotAllocationLock, nobody except us can
461 : * be doing that. So it's safe to initialize the slot.
462 : */
463 : Assert(!slot->in_use);
464 : Assert(slot->active_pid == 0);
465 :
466 : /* first initialize persistent data */
467 1346 : memset(&slot->data, 0, sizeof(ReplicationSlotPersistentData));
468 1346 : namestrcpy(&slot->data.name, name);
469 1346 : slot->data.database = db_specific ? MyDatabaseId : InvalidOid;
470 1346 : slot->data.persistency = persistency;
471 1346 : slot->data.two_phase = two_phase;
472 1346 : slot->data.two_phase_at = InvalidXLogRecPtr;
473 1346 : slot->data.failover = failover;
474 1346 : slot->data.synced = synced;
475 :
476 : /* and then data only present in shared memory */
477 1346 : slot->just_dirtied = false;
478 1346 : slot->dirty = false;
479 1346 : slot->effective_xmin = InvalidTransactionId;
480 1346 : slot->effective_catalog_xmin = InvalidTransactionId;
481 1346 : slot->candidate_catalog_xmin = InvalidTransactionId;
482 1346 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
483 1346 : slot->candidate_restart_valid = InvalidXLogRecPtr;
484 1346 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
485 1346 : slot->last_saved_confirmed_flush = InvalidXLogRecPtr;
486 1346 : slot->last_saved_restart_lsn = InvalidXLogRecPtr;
487 1346 : slot->inactive_since = 0;
488 1346 : slot->slotsync_skip_reason = SS_SKIP_NONE;
489 :
490 : /*
491 : * Create the slot on disk. We haven't actually marked the slot allocated
492 : * yet, so no special cleanup is required if this errors out.
493 : */
494 1346 : CreateSlotOnDisk(slot);
495 :
496 : /*
497 : * We need to briefly prevent any other backend from iterating over the
498 : * slots while we flip the in_use flag. We also need to set the active
499 : * flag while holding the ControlLock as otherwise a concurrent
500 : * ReplicationSlotAcquire() could acquire the slot as well.
501 : */
502 1346 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
503 :
504 1346 : slot->in_use = true;
505 :
506 : /* We can now mark the slot active, and that makes it our slot. */
507 1346 : SpinLockAcquire(&slot->mutex);
508 : Assert(slot->active_pid == 0);
509 1346 : slot->active_pid = MyProcPid;
510 1346 : SpinLockRelease(&slot->mutex);
511 1346 : MyReplicationSlot = slot;
512 :
513 1346 : LWLockRelease(ReplicationSlotControlLock);
514 :
515 : /*
516 : * Create statistics entry for the new logical slot. We don't collect any
517 : * stats for physical slots, so no need to create an entry for the same.
518 : * See ReplicationSlotDropPtr for why we need to do this before releasing
519 : * ReplicationSlotAllocationLock.
520 : */
521 1346 : if (SlotIsLogical(slot))
522 968 : pgstat_create_replslot(slot);
523 :
524 : /*
525 : * Now that the slot has been marked as in_use and active, it's safe to
526 : * let somebody else try to allocate a slot.
527 : */
528 1346 : LWLockRelease(ReplicationSlotAllocationLock);
529 :
530 : /* Let everybody know we've modified this slot */
531 1346 : ConditionVariableBroadcast(&slot->active_cv);
532 1346 : }
533 :
534 : /*
535 : * Search for the named replication slot.
536 : *
537 : * Return the replication slot if found, otherwise NULL.
538 : */
539 : ReplicationSlot *
540 3910 : SearchNamedReplicationSlot(const char *name, bool need_lock)
541 : {
542 : int i;
543 3910 : ReplicationSlot *slot = NULL;
544 :
545 3910 : if (need_lock)
546 1140 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
547 :
548 15014 : for (i = 0; i < max_replication_slots; i++)
549 : {
550 14072 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
551 :
552 14072 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
553 : {
554 2968 : slot = s;
555 2968 : break;
556 : }
557 : }
558 :
559 3910 : if (need_lock)
560 1140 : LWLockRelease(ReplicationSlotControlLock);
561 :
562 3910 : return slot;
563 : }
564 :
565 : /*
566 : * Return the index of the replication slot in
567 : * ReplicationSlotCtl->replication_slots.
568 : *
569 : * This is mainly useful to have an efficient key for storing replication slot
570 : * stats.
571 : */
572 : int
573 16092 : ReplicationSlotIndex(ReplicationSlot *slot)
574 : {
575 : Assert(slot >= ReplicationSlotCtl->replication_slots &&
576 : slot < ReplicationSlotCtl->replication_slots + max_replication_slots);
577 :
578 16092 : return slot - ReplicationSlotCtl->replication_slots;
579 : }
580 :
581 : /*
582 : * If the slot at 'index' is unused, return false. Otherwise 'name' is set to
583 : * the slot's name and true is returned.
584 : *
585 : * This likely is only useful for pgstat_replslot.c during shutdown, in other
586 : * cases there are obvious TOCTOU issues.
587 : */
588 : bool
589 204 : ReplicationSlotName(int index, Name name)
590 : {
591 : ReplicationSlot *slot;
592 : bool found;
593 :
594 204 : slot = &ReplicationSlotCtl->replication_slots[index];
595 :
596 : /*
597 : * Ensure that the slot cannot be dropped while we copy the name. Don't
598 : * need the spinlock as the name of an existing slot cannot change.
599 : */
600 204 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
601 204 : found = slot->in_use;
602 204 : if (slot->in_use)
603 204 : namestrcpy(name, NameStr(slot->data.name));
604 204 : LWLockRelease(ReplicationSlotControlLock);
605 :
606 204 : return found;
607 : }
608 :
609 : /*
610 : * Find a previously created slot and mark it as used by this process.
611 : *
612 : * An error is raised if nowait is true and the slot is currently in use. If
613 : * nowait is false, we sleep until the slot is released by the owning process.
614 : *
615 : * An error is raised if error_if_invalid is true and the slot is found to
616 : * be invalid. It should always be set to true, except when we are temporarily
617 : * acquiring the slot and don't intend to change it.
618 : */
619 : void
620 2618 : ReplicationSlotAcquire(const char *name, bool nowait, bool error_if_invalid)
621 : {
622 : ReplicationSlot *s;
623 : int active_pid;
624 :
625 : Assert(name != NULL);
626 :
627 2618 : retry:
628 : Assert(MyReplicationSlot == NULL);
629 :
630 2618 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
631 :
632 : /* Check if the slot exists with the given name. */
633 2618 : s = SearchNamedReplicationSlot(name, false);
634 2618 : if (s == NULL || !s->in_use)
635 : {
636 18 : LWLockRelease(ReplicationSlotControlLock);
637 :
638 18 : ereport(ERROR,
639 : (errcode(ERRCODE_UNDEFINED_OBJECT),
640 : errmsg("replication slot \"%s\" does not exist",
641 : name)));
642 : }
643 :
644 : /*
645 : * Do not allow users to acquire the reserved slot. This scenario may
646 : * occur if the launcher that owns the slot has terminated unexpectedly
647 : * due to an error, and a backend process attempts to reuse the slot.
648 : */
649 2600 : if (!IsLogicalLauncher() && IsSlotForConflictCheck(name))
650 0 : ereport(ERROR,
651 : errcode(ERRCODE_UNDEFINED_OBJECT),
652 : errmsg("cannot acquire replication slot \"%s\"", name),
653 : errdetail("The slot is reserved for conflict detection and can only be acquired by logical replication launcher."));
654 :
655 : /*
656 : * This is the slot we want; check if it's active under some other
657 : * process. In single user mode, we don't need this check.
658 : */
659 2600 : if (IsUnderPostmaster)
660 : {
661 : /*
662 : * Get ready to sleep on the slot in case it is active. (We may end
663 : * up not sleeping, but we don't want to do this while holding the
664 : * spinlock.)
665 : */
666 2590 : if (!nowait)
667 558 : ConditionVariablePrepareToSleep(&s->active_cv);
668 :
669 : /*
670 : * It is important to reset the inactive_since under spinlock here to
671 : * avoid race conditions with slot invalidation. See comments related
672 : * to inactive_since in InvalidatePossiblyObsoleteSlot.
673 : */
674 2590 : SpinLockAcquire(&s->mutex);
675 2590 : if (s->active_pid == 0)
676 2300 : s->active_pid = MyProcPid;
677 2590 : active_pid = s->active_pid;
678 2590 : ReplicationSlotSetInactiveSince(s, 0, false);
679 2590 : SpinLockRelease(&s->mutex);
680 : }
681 : else
682 : {
683 10 : s->active_pid = active_pid = MyProcPid;
684 10 : ReplicationSlotSetInactiveSince(s, 0, true);
685 : }
686 2600 : LWLockRelease(ReplicationSlotControlLock);
687 :
688 : /*
689 : * If we found the slot but it's already active in another process, we
690 : * wait until the owning process signals us that it's been released, or
691 : * error out.
692 : */
693 2600 : if (active_pid != MyProcPid)
694 : {
695 0 : if (!nowait)
696 : {
697 : /* Wait here until we get signaled, and then restart */
698 0 : ConditionVariableSleep(&s->active_cv,
699 : WAIT_EVENT_REPLICATION_SLOT_DROP);
700 0 : ConditionVariableCancelSleep();
701 0 : goto retry;
702 : }
703 :
704 0 : ereport(ERROR,
705 : (errcode(ERRCODE_OBJECT_IN_USE),
706 : errmsg("replication slot \"%s\" is active for PID %d",
707 : NameStr(s->data.name), active_pid)));
708 : }
709 2600 : else if (!nowait)
710 558 : ConditionVariableCancelSleep(); /* no sleep needed after all */
711 :
712 : /* We made this slot active, so it's ours now. */
713 2600 : MyReplicationSlot = s;
714 :
715 : /*
716 : * We need to check for invalidation after making the slot ours to avoid
717 : * the possible race condition with the checkpointer that can otherwise
718 : * invalidate the slot immediately after the check.
719 : */
720 2600 : if (error_if_invalid && s->data.invalidated != RS_INVAL_NONE)
721 14 : ereport(ERROR,
722 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
723 : errmsg("can no longer access replication slot \"%s\"",
724 : NameStr(s->data.name)),
725 : errdetail("This replication slot has been invalidated due to \"%s\".",
726 : GetSlotInvalidationCauseName(s->data.invalidated)));
727 :
728 : /* Let everybody know we've modified this slot */
729 2586 : ConditionVariableBroadcast(&s->active_cv);
730 :
731 : /*
732 : * The call to pgstat_acquire_replslot() protects against stats for a
733 : * different slot, from before a restart or such, being present during
734 : * pgstat_report_replslot().
735 : */
736 2586 : if (SlotIsLogical(s))
737 2162 : pgstat_acquire_replslot(s);
738 :
739 :
740 2586 : if (am_walsender)
741 : {
742 1748 : ereport(log_replication_commands ? LOG : DEBUG1,
743 : SlotIsLogical(s)
744 : ? errmsg("acquired logical replication slot \"%s\"",
745 : NameStr(s->data.name))
746 : : errmsg("acquired physical replication slot \"%s\"",
747 : NameStr(s->data.name)));
748 : }
749 2586 : }
750 :
751 : /*
752 : * Release the replication slot that this backend considers to own.
753 : *
754 : * This or another backend can re-acquire the slot later.
755 : * Resources this slot requires will be preserved.
756 : */
757 : void
758 3138 : ReplicationSlotRelease(void)
759 : {
760 3138 : ReplicationSlot *slot = MyReplicationSlot;
761 3138 : char *slotname = NULL; /* keep compiler quiet */
762 : bool is_logical;
763 3138 : TimestampTz now = 0;
764 :
765 : Assert(slot != NULL && slot->active_pid != 0);
766 :
767 3138 : is_logical = SlotIsLogical(slot);
768 :
769 3138 : if (am_walsender)
770 2176 : slotname = pstrdup(NameStr(slot->data.name));
771 :
772 3138 : if (slot->data.persistency == RS_EPHEMERAL)
773 : {
774 : /*
775 : * Delete the slot. There is no !PANIC case where this is allowed to
776 : * fail, all that may happen is an incomplete cleanup of the on-disk
777 : * data.
778 : */
779 12 : ReplicationSlotDropAcquired();
780 :
781 : /*
782 : * Request to disable logical decoding, even though this slot may not
783 : * have been the last logical slot. The checkpointer will verify if
784 : * logical decoding should actually be disabled.
785 : */
786 12 : if (is_logical)
787 12 : RequestDisableLogicalDecoding();
788 : }
789 :
790 : /*
791 : * If slot needed to temporarily restrain both data and catalog xmin to
792 : * create the catalog snapshot, remove that temporary constraint.
793 : * Snapshots can only be exported while the initial snapshot is still
794 : * acquired.
795 : */
796 3138 : if (!TransactionIdIsValid(slot->data.xmin) &&
797 3078 : TransactionIdIsValid(slot->effective_xmin))
798 : {
799 404 : SpinLockAcquire(&slot->mutex);
800 404 : slot->effective_xmin = InvalidTransactionId;
801 404 : SpinLockRelease(&slot->mutex);
802 404 : ReplicationSlotsComputeRequiredXmin(false);
803 : }
804 :
805 : /*
806 : * Set the time since the slot has become inactive. We get the current
807 : * time beforehand to avoid system call while holding the spinlock.
808 : */
809 3138 : now = GetCurrentTimestamp();
810 :
811 3138 : if (slot->data.persistency == RS_PERSISTENT)
812 : {
813 : /*
814 : * Mark persistent slot inactive. We're not freeing it, just
815 : * disconnecting, but wake up others that may be waiting for it.
816 : */
817 2542 : SpinLockAcquire(&slot->mutex);
818 2542 : slot->active_pid = 0;
819 2542 : ReplicationSlotSetInactiveSince(slot, now, false);
820 2542 : SpinLockRelease(&slot->mutex);
821 2542 : ConditionVariableBroadcast(&slot->active_cv);
822 : }
823 : else
824 596 : ReplicationSlotSetInactiveSince(slot, now, true);
825 :
826 3138 : MyReplicationSlot = NULL;
827 :
828 : /* might not have been set when we've been a plain slot */
829 3138 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
830 3138 : MyProc->statusFlags &= ~PROC_IN_LOGICAL_DECODING;
831 3138 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
832 3138 : LWLockRelease(ProcArrayLock);
833 :
834 3138 : if (am_walsender)
835 : {
836 2176 : ereport(log_replication_commands ? LOG : DEBUG1,
837 : is_logical
838 : ? errmsg("released logical replication slot \"%s\"",
839 : slotname)
840 : : errmsg("released physical replication slot \"%s\"",
841 : slotname));
842 :
843 2176 : pfree(slotname);
844 : }
845 3138 : }
846 :
847 : /*
848 : * Cleanup temporary slots created in current session.
849 : *
850 : * Cleanup only synced temporary slots if 'synced_only' is true, else
851 : * cleanup all temporary slots.
852 : *
853 : * If it drops the last logical slot in the cluster, requests to disable
854 : * logical decoding.
855 : */
856 : void
857 90982 : ReplicationSlotCleanup(bool synced_only)
858 : {
859 : int i;
860 : bool found_valid_logicalslot;
861 90982 : bool dropped_logical = false;
862 :
863 : Assert(MyReplicationSlot == NULL);
864 :
865 91276 : restart:
866 91276 : found_valid_logicalslot = false;
867 91276 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
868 989588 : for (i = 0; i < max_replication_slots; i++)
869 : {
870 898606 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
871 :
872 898606 : if (!s->in_use)
873 869212 : continue;
874 :
875 29394 : SpinLockAcquire(&s->mutex);
876 :
877 58788 : found_valid_logicalslot |=
878 29394 : (SlotIsLogical(s) && s->data.invalidated == RS_INVAL_NONE);
879 :
880 29394 : if ((s->active_pid == MyProcPid &&
881 294 : (!synced_only || s->data.synced)))
882 : {
883 : Assert(s->data.persistency == RS_TEMPORARY);
884 294 : SpinLockRelease(&s->mutex);
885 294 : LWLockRelease(ReplicationSlotControlLock); /* avoid deadlock */
886 :
887 294 : if (SlotIsLogical(s))
888 18 : dropped_logical = true;
889 :
890 294 : ReplicationSlotDropPtr(s);
891 :
892 294 : ConditionVariableBroadcast(&s->active_cv);
893 294 : goto restart;
894 : }
895 : else
896 29100 : SpinLockRelease(&s->mutex);
897 : }
898 :
899 90982 : LWLockRelease(ReplicationSlotControlLock);
900 :
901 90982 : if (dropped_logical && !found_valid_logicalslot)
902 4 : RequestDisableLogicalDecoding();
903 90982 : }
904 :
905 : /*
906 : * Permanently drop replication slot identified by the passed in name.
907 : */
908 : void
909 850 : ReplicationSlotDrop(const char *name, bool nowait)
910 : {
911 : bool is_logical;
912 :
913 : Assert(MyReplicationSlot == NULL);
914 :
915 850 : ReplicationSlotAcquire(name, nowait, false);
916 :
917 : /*
918 : * Do not allow users to drop the slots which are currently being synced
919 : * from the primary to the standby.
920 : */
921 836 : if (RecoveryInProgress() && MyReplicationSlot->data.synced)
922 2 : ereport(ERROR,
923 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
924 : errmsg("cannot drop replication slot \"%s\"", name),
925 : errdetail("This replication slot is being synchronized from the primary server."));
926 :
927 834 : is_logical = SlotIsLogical(MyReplicationSlot);
928 :
929 834 : ReplicationSlotDropAcquired();
930 :
931 834 : if (is_logical)
932 794 : RequestDisableLogicalDecoding();
933 834 : }
934 :
935 : /*
936 : * Change the definition of the slot identified by the specified name.
937 : *
938 : * Altering the two_phase property of a slot requires caution on the
939 : * client-side. Enabling it at any random point during decoding has the
940 : * risk that transactions prepared before this change may be skipped by
941 : * the decoder, leading to missing prepare records on the client. So, we
942 : * enable it for subscription related slots only once the initial tablesync
943 : * is finished. See comments atop worker.c. Disabling it is safe only when
944 : * there are no pending prepared transaction, otherwise, the changes of
945 : * already prepared transactions can be replicated again along with their
946 : * corresponding commit leading to duplicate data or errors.
947 : */
948 : void
949 14 : ReplicationSlotAlter(const char *name, const bool *failover,
950 : const bool *two_phase)
951 : {
952 14 : bool update_slot = false;
953 :
954 : Assert(MyReplicationSlot == NULL);
955 : Assert(failover || two_phase);
956 :
957 14 : ReplicationSlotAcquire(name, false, true);
958 :
959 12 : if (SlotIsPhysical(MyReplicationSlot))
960 0 : ereport(ERROR,
961 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
962 : errmsg("cannot use %s with a physical replication slot",
963 : "ALTER_REPLICATION_SLOT"));
964 :
965 12 : if (RecoveryInProgress())
966 : {
967 : /*
968 : * Do not allow users to alter the slots which are currently being
969 : * synced from the primary to the standby.
970 : */
971 2 : if (MyReplicationSlot->data.synced)
972 2 : ereport(ERROR,
973 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
974 : errmsg("cannot alter replication slot \"%s\"", name),
975 : errdetail("This replication slot is being synchronized from the primary server."));
976 :
977 : /*
978 : * Do not allow users to enable failover on the standby as we do not
979 : * support sync to the cascading standby.
980 : */
981 0 : if (failover && *failover)
982 0 : ereport(ERROR,
983 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
984 : errmsg("cannot enable failover for a replication slot"
985 : " on the standby"));
986 : }
987 :
988 10 : if (failover)
989 : {
990 : /*
991 : * Do not allow users to enable failover for temporary slots as we do
992 : * not support syncing temporary slots to the standby.
993 : */
994 8 : if (*failover && MyReplicationSlot->data.persistency == RS_TEMPORARY)
995 0 : ereport(ERROR,
996 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
997 : errmsg("cannot enable failover for a temporary replication slot"));
998 :
999 8 : if (MyReplicationSlot->data.failover != *failover)
1000 : {
1001 8 : SpinLockAcquire(&MyReplicationSlot->mutex);
1002 8 : MyReplicationSlot->data.failover = *failover;
1003 8 : SpinLockRelease(&MyReplicationSlot->mutex);
1004 :
1005 8 : update_slot = true;
1006 : }
1007 : }
1008 :
1009 10 : if (two_phase && MyReplicationSlot->data.two_phase != *two_phase)
1010 : {
1011 2 : SpinLockAcquire(&MyReplicationSlot->mutex);
1012 2 : MyReplicationSlot->data.two_phase = *two_phase;
1013 2 : SpinLockRelease(&MyReplicationSlot->mutex);
1014 :
1015 2 : update_slot = true;
1016 : }
1017 :
1018 10 : if (update_slot)
1019 : {
1020 10 : ReplicationSlotMarkDirty();
1021 10 : ReplicationSlotSave();
1022 : }
1023 :
1024 10 : ReplicationSlotRelease();
1025 10 : }
1026 :
1027 : /*
1028 : * Permanently drop the currently acquired replication slot.
1029 : */
1030 : void
1031 862 : ReplicationSlotDropAcquired(void)
1032 : {
1033 862 : ReplicationSlot *slot = MyReplicationSlot;
1034 :
1035 : Assert(MyReplicationSlot != NULL);
1036 :
1037 : /* slot isn't acquired anymore */
1038 862 : MyReplicationSlot = NULL;
1039 :
1040 862 : ReplicationSlotDropPtr(slot);
1041 862 : }
1042 :
1043 : /*
1044 : * Permanently drop the replication slot which will be released by the point
1045 : * this function returns.
1046 : */
1047 : static void
1048 1156 : ReplicationSlotDropPtr(ReplicationSlot *slot)
1049 : {
1050 : char path[MAXPGPATH];
1051 : char tmppath[MAXPGPATH];
1052 :
1053 : /*
1054 : * If some other backend ran this code concurrently with us, we might try
1055 : * to delete a slot with a certain name while someone else was trying to
1056 : * create a slot with the same name.
1057 : */
1058 1156 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
1059 :
1060 : /* Generate pathnames. */
1061 1156 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
1062 1156 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
1063 :
1064 : /*
1065 : * Rename the slot directory on disk, so that we'll no longer recognize
1066 : * this as a valid slot. Note that if this fails, we've got to mark the
1067 : * slot inactive before bailing out. If we're dropping an ephemeral or a
1068 : * temporary slot, we better never fail hard as the caller won't expect
1069 : * the slot to survive and this might get called during error handling.
1070 : */
1071 1156 : if (rename(path, tmppath) == 0)
1072 : {
1073 : /*
1074 : * We need to fsync() the directory we just renamed and its parent to
1075 : * make sure that our changes are on disk in a crash-safe fashion. If
1076 : * fsync() fails, we can't be sure whether the changes are on disk or
1077 : * not. For now, we handle that by panicking;
1078 : * StartupReplicationSlots() will try to straighten it out after
1079 : * restart.
1080 : */
1081 1156 : START_CRIT_SECTION();
1082 1156 : fsync_fname(tmppath, true);
1083 1156 : fsync_fname(PG_REPLSLOT_DIR, true);
1084 1156 : END_CRIT_SECTION();
1085 : }
1086 : else
1087 : {
1088 0 : bool fail_softly = slot->data.persistency != RS_PERSISTENT;
1089 :
1090 0 : SpinLockAcquire(&slot->mutex);
1091 0 : slot->active_pid = 0;
1092 0 : SpinLockRelease(&slot->mutex);
1093 :
1094 : /* wake up anyone waiting on this slot */
1095 0 : ConditionVariableBroadcast(&slot->active_cv);
1096 :
1097 0 : ereport(fail_softly ? WARNING : ERROR,
1098 : (errcode_for_file_access(),
1099 : errmsg("could not rename file \"%s\" to \"%s\": %m",
1100 : path, tmppath)));
1101 : }
1102 :
1103 : /*
1104 : * The slot is definitely gone. Lock out concurrent scans of the array
1105 : * long enough to kill it. It's OK to clear the active PID here without
1106 : * grabbing the mutex because nobody else can be scanning the array here,
1107 : * and nobody can be attached to this slot and thus access it without
1108 : * scanning the array.
1109 : *
1110 : * Also wake up processes waiting for it.
1111 : */
1112 1156 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
1113 1156 : slot->active_pid = 0;
1114 1156 : slot->in_use = false;
1115 1156 : LWLockRelease(ReplicationSlotControlLock);
1116 1156 : ConditionVariableBroadcast(&slot->active_cv);
1117 :
1118 : /*
1119 : * Slot is dead and doesn't prevent resource removal anymore, recompute
1120 : * limits.
1121 : */
1122 1156 : ReplicationSlotsComputeRequiredXmin(false);
1123 1156 : ReplicationSlotsComputeRequiredLSN();
1124 :
1125 : /*
1126 : * If removing the directory fails, the worst thing that will happen is
1127 : * that the user won't be able to create a new slot with the same name
1128 : * until the next server restart. We warn about it, but that's all.
1129 : */
1130 1156 : if (!rmtree(tmppath, true))
1131 0 : ereport(WARNING,
1132 : (errmsg("could not remove directory \"%s\"", tmppath)));
1133 :
1134 : /*
1135 : * Drop the statistics entry for the replication slot. Do this while
1136 : * holding ReplicationSlotAllocationLock so that we don't drop a
1137 : * statistics entry for another slot with the same name just created in
1138 : * another session.
1139 : */
1140 1156 : if (SlotIsLogical(slot))
1141 838 : pgstat_drop_replslot(slot);
1142 :
1143 : /*
1144 : * We release this at the very end, so that nobody starts trying to create
1145 : * a slot while we're still cleaning up the detritus of the old one.
1146 : */
1147 1156 : LWLockRelease(ReplicationSlotAllocationLock);
1148 1156 : }
1149 :
1150 : /*
1151 : * Serialize the currently acquired slot's state from memory to disk, thereby
1152 : * guaranteeing the current state will survive a crash.
1153 : */
1154 : void
1155 2798 : ReplicationSlotSave(void)
1156 : {
1157 : char path[MAXPGPATH];
1158 :
1159 : Assert(MyReplicationSlot != NULL);
1160 :
1161 2798 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(MyReplicationSlot->data.name));
1162 2798 : SaveSlotToPath(MyReplicationSlot, path, ERROR);
1163 2798 : }
1164 :
1165 : /*
1166 : * Signal that it would be useful if the currently acquired slot would be
1167 : * flushed out to disk.
1168 : *
1169 : * Note that the actual flush to disk can be delayed for a long time, if
1170 : * required for correctness explicitly do a ReplicationSlotSave().
1171 : */
1172 : void
1173 58410 : ReplicationSlotMarkDirty(void)
1174 : {
1175 58410 : ReplicationSlot *slot = MyReplicationSlot;
1176 :
1177 : Assert(MyReplicationSlot != NULL);
1178 :
1179 58410 : SpinLockAcquire(&slot->mutex);
1180 58410 : MyReplicationSlot->just_dirtied = true;
1181 58410 : MyReplicationSlot->dirty = true;
1182 58410 : SpinLockRelease(&slot->mutex);
1183 58410 : }
1184 :
1185 : /*
1186 : * Convert a slot that's marked as RS_EPHEMERAL or RS_TEMPORARY to a
1187 : * RS_PERSISTENT slot, guaranteeing it will be there after an eventual crash.
1188 : */
1189 : void
1190 938 : ReplicationSlotPersist(void)
1191 : {
1192 938 : ReplicationSlot *slot = MyReplicationSlot;
1193 :
1194 : Assert(slot != NULL);
1195 : Assert(slot->data.persistency != RS_PERSISTENT);
1196 :
1197 938 : SpinLockAcquire(&slot->mutex);
1198 938 : slot->data.persistency = RS_PERSISTENT;
1199 938 : SpinLockRelease(&slot->mutex);
1200 :
1201 938 : ReplicationSlotMarkDirty();
1202 938 : ReplicationSlotSave();
1203 938 : }
1204 :
1205 : /*
1206 : * Compute the oldest xmin across all slots and store it in the ProcArray.
1207 : *
1208 : * If already_locked is true, both the ReplicationSlotControlLock and the
1209 : * ProcArrayLock have already been acquired exclusively. It is crucial that the
1210 : * caller first acquires the ReplicationSlotControlLock, followed by the
1211 : * ProcArrayLock, to prevent any undetectable deadlocks since this function
1212 : * acquires them in that order.
1213 : */
1214 : void
1215 4886 : ReplicationSlotsComputeRequiredXmin(bool already_locked)
1216 : {
1217 : int i;
1218 4886 : TransactionId agg_xmin = InvalidTransactionId;
1219 4886 : TransactionId agg_catalog_xmin = InvalidTransactionId;
1220 :
1221 : Assert(ReplicationSlotCtl != NULL);
1222 : Assert(!already_locked ||
1223 : (LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_EXCLUSIVE) &&
1224 : LWLockHeldByMeInMode(ProcArrayLock, LW_EXCLUSIVE)));
1225 :
1226 : /*
1227 : * Hold the ReplicationSlotControlLock until after updating the slot xmin
1228 : * values, so no backend updates the initial xmin for newly created slot
1229 : * concurrently. A shared lock is used here to minimize lock contention,
1230 : * especially when many slots exist and advancements occur frequently.
1231 : * This is safe since an exclusive lock is taken during initial slot xmin
1232 : * update in slot creation.
1233 : *
1234 : * One might think that we can hold the ProcArrayLock exclusively and
1235 : * update the slot xmin values, but it could increase lock contention on
1236 : * the ProcArrayLock, which is not great since this function can be called
1237 : * at non-negligible frequency.
1238 : *
1239 : * Concurrent invocation of this function may cause the computed slot xmin
1240 : * to regress. However, this is harmless because tuples prior to the most
1241 : * recent xmin are no longer useful once advancement occurs (see
1242 : * LogicalConfirmReceivedLocation where the slot's xmin value is flushed
1243 : * before updating the effective_xmin). Thus, such regression merely
1244 : * prevents VACUUM from prematurely removing tuples without causing the
1245 : * early deletion of required data.
1246 : */
1247 4886 : if (!already_locked)
1248 3916 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1249 :
1250 49626 : for (i = 0; i < max_replication_slots; i++)
1251 : {
1252 44740 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1253 : TransactionId effective_xmin;
1254 : TransactionId effective_catalog_xmin;
1255 : bool invalidated;
1256 :
1257 44740 : if (!s->in_use)
1258 40066 : continue;
1259 :
1260 4674 : SpinLockAcquire(&s->mutex);
1261 4674 : effective_xmin = s->effective_xmin;
1262 4674 : effective_catalog_xmin = s->effective_catalog_xmin;
1263 4674 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1264 4674 : SpinLockRelease(&s->mutex);
1265 :
1266 : /* invalidated slots need not apply */
1267 4674 : if (invalidated)
1268 52 : continue;
1269 :
1270 : /* check the data xmin */
1271 4622 : if (TransactionIdIsValid(effective_xmin) &&
1272 36 : (!TransactionIdIsValid(agg_xmin) ||
1273 36 : TransactionIdPrecedes(effective_xmin, agg_xmin)))
1274 722 : agg_xmin = effective_xmin;
1275 :
1276 : /* check the catalog xmin */
1277 4622 : if (TransactionIdIsValid(effective_catalog_xmin) &&
1278 1926 : (!TransactionIdIsValid(agg_catalog_xmin) ||
1279 1926 : TransactionIdPrecedes(effective_catalog_xmin, agg_catalog_xmin)))
1280 2420 : agg_catalog_xmin = effective_catalog_xmin;
1281 : }
1282 :
1283 4886 : ProcArraySetReplicationSlotXmin(agg_xmin, agg_catalog_xmin, already_locked);
1284 :
1285 4886 : if (!already_locked)
1286 3916 : LWLockRelease(ReplicationSlotControlLock);
1287 4886 : }
1288 :
1289 : /*
1290 : * Compute the oldest restart LSN across all slots and inform xlog module.
1291 : *
1292 : * Note: while max_slot_wal_keep_size is theoretically relevant for this
1293 : * purpose, we don't try to account for that, because this module doesn't
1294 : * know what to compare against.
1295 : */
1296 : void
1297 59812 : ReplicationSlotsComputeRequiredLSN(void)
1298 : {
1299 : int i;
1300 59812 : XLogRecPtr min_required = InvalidXLogRecPtr;
1301 :
1302 : Assert(ReplicationSlotCtl != NULL);
1303 :
1304 59812 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1305 649652 : for (i = 0; i < max_replication_slots; i++)
1306 : {
1307 589840 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1308 : XLogRecPtr restart_lsn;
1309 : XLogRecPtr last_saved_restart_lsn;
1310 : bool invalidated;
1311 : ReplicationSlotPersistency persistency;
1312 :
1313 589840 : if (!s->in_use)
1314 529780 : continue;
1315 :
1316 60060 : SpinLockAcquire(&s->mutex);
1317 60060 : persistency = s->data.persistency;
1318 60060 : restart_lsn = s->data.restart_lsn;
1319 60060 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1320 60060 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1321 60060 : SpinLockRelease(&s->mutex);
1322 :
1323 : /* invalidated slots need not apply */
1324 60060 : if (invalidated)
1325 54 : continue;
1326 :
1327 : /*
1328 : * For persistent slot use last_saved_restart_lsn to compute the
1329 : * oldest LSN for removal of WAL segments. The segments between
1330 : * last_saved_restart_lsn and restart_lsn might be needed by a
1331 : * persistent slot in the case of database crash. Non-persistent
1332 : * slots can't survive the database crash, so we don't care about
1333 : * last_saved_restart_lsn for them.
1334 : */
1335 60006 : if (persistency == RS_PERSISTENT)
1336 : {
1337 58324 : if (XLogRecPtrIsValid(last_saved_restart_lsn) &&
1338 : restart_lsn > last_saved_restart_lsn)
1339 : {
1340 53690 : restart_lsn = last_saved_restart_lsn;
1341 : }
1342 : }
1343 :
1344 60006 : if (XLogRecPtrIsValid(restart_lsn) &&
1345 2356 : (!XLogRecPtrIsValid(min_required) ||
1346 : restart_lsn < min_required))
1347 57798 : min_required = restart_lsn;
1348 : }
1349 59812 : LWLockRelease(ReplicationSlotControlLock);
1350 :
1351 59812 : XLogSetReplicationSlotMinimumLSN(min_required);
1352 59812 : }
1353 :
1354 : /*
1355 : * Compute the oldest WAL LSN required by *logical* decoding slots..
1356 : *
1357 : * Returns InvalidXLogRecPtr if logical decoding is disabled or no logical
1358 : * slots exist.
1359 : *
1360 : * NB: this returns a value >= ReplicationSlotsComputeRequiredLSN(), since it
1361 : * ignores physical replication slots.
1362 : *
1363 : * The results aren't required frequently, so we don't maintain a precomputed
1364 : * value like we do for ComputeRequiredLSN() and ComputeRequiredXmin().
1365 : */
1366 : XLogRecPtr
1367 7096 : ReplicationSlotsComputeLogicalRestartLSN(void)
1368 : {
1369 7096 : XLogRecPtr result = InvalidXLogRecPtr;
1370 : int i;
1371 :
1372 7096 : if (max_replication_slots <= 0)
1373 4 : return InvalidXLogRecPtr;
1374 :
1375 7092 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1376 :
1377 76920 : for (i = 0; i < max_replication_slots; i++)
1378 : {
1379 : ReplicationSlot *s;
1380 : XLogRecPtr restart_lsn;
1381 : XLogRecPtr last_saved_restart_lsn;
1382 : bool invalidated;
1383 : ReplicationSlotPersistency persistency;
1384 :
1385 69828 : s = &ReplicationSlotCtl->replication_slots[i];
1386 :
1387 : /* cannot change while ReplicationSlotCtlLock is held */
1388 69828 : if (!s->in_use)
1389 68272 : continue;
1390 :
1391 : /* we're only interested in logical slots */
1392 1556 : if (!SlotIsLogical(s))
1393 1080 : continue;
1394 :
1395 : /* read once, it's ok if it increases while we're checking */
1396 476 : SpinLockAcquire(&s->mutex);
1397 476 : persistency = s->data.persistency;
1398 476 : restart_lsn = s->data.restart_lsn;
1399 476 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1400 476 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1401 476 : SpinLockRelease(&s->mutex);
1402 :
1403 : /* invalidated slots need not apply */
1404 476 : if (invalidated)
1405 20 : continue;
1406 :
1407 : /*
1408 : * For persistent slot use last_saved_restart_lsn to compute the
1409 : * oldest LSN for removal of WAL segments. The segments between
1410 : * last_saved_restart_lsn and restart_lsn might be needed by a
1411 : * persistent slot in the case of database crash. Non-persistent
1412 : * slots can't survive the database crash, so we don't care about
1413 : * last_saved_restart_lsn for them.
1414 : */
1415 456 : if (persistency == RS_PERSISTENT)
1416 : {
1417 452 : if (XLogRecPtrIsValid(last_saved_restart_lsn) &&
1418 : restart_lsn > last_saved_restart_lsn)
1419 : {
1420 0 : restart_lsn = last_saved_restart_lsn;
1421 : }
1422 : }
1423 :
1424 456 : if (!XLogRecPtrIsValid(restart_lsn))
1425 0 : continue;
1426 :
1427 456 : if (!XLogRecPtrIsValid(result) ||
1428 : restart_lsn < result)
1429 360 : result = restart_lsn;
1430 : }
1431 :
1432 7092 : LWLockRelease(ReplicationSlotControlLock);
1433 :
1434 7092 : return result;
1435 : }
1436 :
1437 : /*
1438 : * ReplicationSlotsCountDBSlots -- count the number of slots that refer to the
1439 : * passed database oid.
1440 : *
1441 : * Returns true if there are any slots referencing the database. *nslots will
1442 : * be set to the absolute number of slots in the database, *nactive to ones
1443 : * currently active.
1444 : */
1445 : bool
1446 98 : ReplicationSlotsCountDBSlots(Oid dboid, int *nslots, int *nactive)
1447 : {
1448 : int i;
1449 :
1450 98 : *nslots = *nactive = 0;
1451 :
1452 98 : if (max_replication_slots <= 0)
1453 0 : return false;
1454 :
1455 98 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1456 1020 : for (i = 0; i < max_replication_slots; i++)
1457 : {
1458 : ReplicationSlot *s;
1459 :
1460 922 : s = &ReplicationSlotCtl->replication_slots[i];
1461 :
1462 : /* cannot change while ReplicationSlotCtlLock is held */
1463 922 : if (!s->in_use)
1464 880 : continue;
1465 :
1466 : /* only logical slots are database specific, skip */
1467 42 : if (!SlotIsLogical(s))
1468 20 : continue;
1469 :
1470 : /* not our database, skip */
1471 22 : if (s->data.database != dboid)
1472 16 : continue;
1473 :
1474 : /* NB: intentionally counting invalidated slots */
1475 :
1476 : /* count slots with spinlock held */
1477 6 : SpinLockAcquire(&s->mutex);
1478 6 : (*nslots)++;
1479 6 : if (s->active_pid != 0)
1480 2 : (*nactive)++;
1481 6 : SpinLockRelease(&s->mutex);
1482 : }
1483 98 : LWLockRelease(ReplicationSlotControlLock);
1484 :
1485 98 : if (*nslots > 0)
1486 6 : return true;
1487 92 : return false;
1488 : }
1489 :
1490 : /*
1491 : * ReplicationSlotsDropDBSlots -- Drop all db-specific slots relating to the
1492 : * passed database oid. The caller should hold an exclusive lock on the
1493 : * pg_database oid for the database to prevent creation of new slots on the db
1494 : * or replay from existing slots.
1495 : *
1496 : * Another session that concurrently acquires an existing slot on the target DB
1497 : * (most likely to drop it) may cause this function to ERROR. If that happens
1498 : * it may have dropped some but not all slots.
1499 : *
1500 : * This routine isn't as efficient as it could be - but we don't drop
1501 : * databases often, especially databases with lots of slots.
1502 : *
1503 : * If it drops the last logical slot in the cluster, it requests to disable
1504 : * logical decoding.
1505 : */
1506 : void
1507 124 : ReplicationSlotsDropDBSlots(Oid dboid)
1508 : {
1509 : int i;
1510 : bool found_valid_logicalslot;
1511 124 : bool dropped = false;
1512 :
1513 124 : if (max_replication_slots <= 0)
1514 0 : return;
1515 :
1516 124 : restart:
1517 134 : found_valid_logicalslot = false;
1518 134 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1519 1300 : for (i = 0; i < max_replication_slots; i++)
1520 : {
1521 : ReplicationSlot *s;
1522 : char *slotname;
1523 : int active_pid;
1524 :
1525 1176 : s = &ReplicationSlotCtl->replication_slots[i];
1526 :
1527 : /* cannot change while ReplicationSlotCtlLock is held */
1528 1176 : if (!s->in_use)
1529 1116 : continue;
1530 :
1531 : /* only logical slots are database specific, skip */
1532 60 : if (!SlotIsLogical(s))
1533 22 : continue;
1534 :
1535 : /*
1536 : * Check logical slots on other databases too so we can disable
1537 : * logical decoding only if no slots in the cluster.
1538 : */
1539 38 : SpinLockAcquire(&s->mutex);
1540 38 : found_valid_logicalslot |= (s->data.invalidated == RS_INVAL_NONE);
1541 38 : SpinLockRelease(&s->mutex);
1542 :
1543 : /* not our database, skip */
1544 38 : if (s->data.database != dboid)
1545 28 : continue;
1546 :
1547 : /* NB: intentionally including invalidated slots to drop */
1548 :
1549 : /* acquire slot, so ReplicationSlotDropAcquired can be reused */
1550 10 : SpinLockAcquire(&s->mutex);
1551 : /* can't change while ReplicationSlotControlLock is held */
1552 10 : slotname = NameStr(s->data.name);
1553 10 : active_pid = s->active_pid;
1554 10 : if (active_pid == 0)
1555 : {
1556 10 : MyReplicationSlot = s;
1557 10 : s->active_pid = MyProcPid;
1558 : }
1559 10 : SpinLockRelease(&s->mutex);
1560 :
1561 : /*
1562 : * Even though we hold an exclusive lock on the database object a
1563 : * logical slot for that DB can still be active, e.g. if it's
1564 : * concurrently being dropped by a backend connected to another DB.
1565 : *
1566 : * That's fairly unlikely in practice, so we'll just bail out.
1567 : *
1568 : * The slot sync worker holds a shared lock on the database before
1569 : * operating on synced logical slots to avoid conflict with the drop
1570 : * happening here. The persistent synced slots are thus safe but there
1571 : * is a possibility that the slot sync worker has created a temporary
1572 : * slot (which stays active even on release) and we are trying to drop
1573 : * that here. In practice, the chances of hitting this scenario are
1574 : * less as during slot synchronization, the temporary slot is
1575 : * immediately converted to persistent and thus is safe due to the
1576 : * shared lock taken on the database. So, we'll just bail out in such
1577 : * a case.
1578 : *
1579 : * XXX: We can consider shutting down the slot sync worker before
1580 : * trying to drop synced temporary slots here.
1581 : */
1582 10 : if (active_pid)
1583 0 : ereport(ERROR,
1584 : (errcode(ERRCODE_OBJECT_IN_USE),
1585 : errmsg("replication slot \"%s\" is active for PID %d",
1586 : slotname, active_pid)));
1587 :
1588 : /*
1589 : * To avoid duplicating ReplicationSlotDropAcquired() and to avoid
1590 : * holding ReplicationSlotControlLock over filesystem operations,
1591 : * release ReplicationSlotControlLock and use
1592 : * ReplicationSlotDropAcquired.
1593 : *
1594 : * As that means the set of slots could change, restart scan from the
1595 : * beginning each time we release the lock.
1596 : */
1597 10 : LWLockRelease(ReplicationSlotControlLock);
1598 10 : ReplicationSlotDropAcquired();
1599 10 : dropped = true;
1600 10 : goto restart;
1601 : }
1602 124 : LWLockRelease(ReplicationSlotControlLock);
1603 :
1604 124 : if (dropped && !found_valid_logicalslot)
1605 0 : RequestDisableLogicalDecoding();
1606 : }
1607 :
1608 : /*
1609 : * Returns true if there is at least one in-use valid logical replication slot.
1610 : */
1611 : bool
1612 930 : CheckLogicalSlotExists(void)
1613 : {
1614 930 : bool found = false;
1615 :
1616 930 : if (max_replication_slots <= 0)
1617 2 : return false;
1618 :
1619 928 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1620 10040 : for (int i = 0; i < max_replication_slots; i++)
1621 : {
1622 : ReplicationSlot *s;
1623 : bool invalidated;
1624 :
1625 9124 : s = &ReplicationSlotCtl->replication_slots[i];
1626 :
1627 : /* cannot change while ReplicationSlotCtlLock is held */
1628 9124 : if (!s->in_use)
1629 9068 : continue;
1630 :
1631 56 : if (SlotIsPhysical(s))
1632 38 : continue;
1633 :
1634 18 : SpinLockAcquire(&s->mutex);
1635 18 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1636 18 : SpinLockRelease(&s->mutex);
1637 :
1638 18 : if (invalidated)
1639 6 : continue;
1640 :
1641 12 : found = true;
1642 12 : break;
1643 : }
1644 928 : LWLockRelease(ReplicationSlotControlLock);
1645 :
1646 928 : return found;
1647 : }
1648 :
1649 : /*
1650 : * Check whether the server's configuration supports using replication
1651 : * slots.
1652 : */
1653 : void
1654 3574 : CheckSlotRequirements(void)
1655 : {
1656 : /*
1657 : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
1658 : * needs the same check.
1659 : */
1660 :
1661 3574 : if (max_replication_slots == 0)
1662 0 : ereport(ERROR,
1663 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1664 : errmsg("replication slots can only be used if \"max_replication_slots\" > 0")));
1665 :
1666 3574 : if (wal_level < WAL_LEVEL_REPLICA)
1667 0 : ereport(ERROR,
1668 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1669 : errmsg("replication slots can only be used if \"wal_level\" >= \"replica\"")));
1670 3574 : }
1671 :
1672 : /*
1673 : * Check whether the user has privilege to use replication slots.
1674 : */
1675 : void
1676 1166 : CheckSlotPermissions(void)
1677 : {
1678 1166 : if (!has_rolreplication(GetUserId()))
1679 10 : ereport(ERROR,
1680 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1681 : errmsg("permission denied to use replication slots"),
1682 : errdetail("Only roles with the %s attribute may use replication slots.",
1683 : "REPLICATION")));
1684 1156 : }
1685 :
1686 : /*
1687 : * Reserve WAL for the currently active slot.
1688 : *
1689 : * Compute and set restart_lsn in a manner that's appropriate for the type of
1690 : * the slot and concurrency safe.
1691 : */
1692 : void
1693 1248 : ReplicationSlotReserveWal(void)
1694 : {
1695 1248 : ReplicationSlot *slot = MyReplicationSlot;
1696 : XLogSegNo segno;
1697 : XLogRecPtr restart_lsn;
1698 :
1699 : Assert(slot != NULL);
1700 : Assert(!XLogRecPtrIsValid(slot->data.restart_lsn));
1701 : Assert(!XLogRecPtrIsValid(slot->last_saved_restart_lsn));
1702 :
1703 : /*
1704 : * The replication slot mechanism is used to prevent the removal of
1705 : * required WAL.
1706 : *
1707 : * Acquire an exclusive lock to prevent the checkpoint process from
1708 : * concurrently computing the minimum slot LSN (see
1709 : * CheckPointReplicationSlots). This ensures that the WAL reserved for
1710 : * replication cannot be removed during a checkpoint.
1711 : *
1712 : * The mechanism is reliable because if WAL reservation occurs first, the
1713 : * checkpoint must wait for the restart_lsn update before determining the
1714 : * minimum non-removable LSN. On the other hand, if the checkpoint happens
1715 : * first, subsequent WAL reservations will select positions at or beyond
1716 : * the redo pointer of that checkpoint.
1717 : */
1718 1248 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
1719 :
1720 : /*
1721 : * For logical slots log a standby snapshot and start logical decoding at
1722 : * exactly that position. That allows the slot to start up more quickly.
1723 : * But on a standby we cannot do WAL writes, so just use the replay
1724 : * pointer; effectively, an attempt to create a logical slot on standby
1725 : * will cause it to wait for an xl_running_xact record to be logged
1726 : * independently on the primary, so that a snapshot can be built using the
1727 : * record.
1728 : *
1729 : * None of this is needed (or indeed helpful) for physical slots as
1730 : * they'll start replay at the last logged checkpoint anyway. Instead,
1731 : * return the location of the last redo LSN, where a base backup has to
1732 : * start replay at.
1733 : */
1734 1248 : if (SlotIsPhysical(slot))
1735 312 : restart_lsn = GetRedoRecPtr();
1736 936 : else if (RecoveryInProgress())
1737 52 : restart_lsn = GetXLogReplayRecPtr(NULL);
1738 : else
1739 884 : restart_lsn = GetXLogInsertRecPtr();
1740 :
1741 1248 : SpinLockAcquire(&slot->mutex);
1742 1248 : slot->data.restart_lsn = restart_lsn;
1743 1248 : SpinLockRelease(&slot->mutex);
1744 :
1745 : /* prevent WAL removal as fast as possible */
1746 1248 : ReplicationSlotsComputeRequiredLSN();
1747 :
1748 : /* Checkpoint shouldn't remove the required WAL. */
1749 1248 : XLByteToSeg(slot->data.restart_lsn, segno, wal_segment_size);
1750 1248 : if (XLogGetLastRemovedSegno() >= segno)
1751 0 : elog(ERROR, "WAL required by replication slot %s has been removed concurrently",
1752 : NameStr(slot->data.name));
1753 :
1754 1248 : LWLockRelease(ReplicationSlotAllocationLock);
1755 :
1756 1248 : if (!RecoveryInProgress() && SlotIsLogical(slot))
1757 : {
1758 : XLogRecPtr flushptr;
1759 :
1760 : /* make sure we have enough information to start */
1761 884 : flushptr = LogStandbySnapshot();
1762 :
1763 : /* and make sure it's fsynced to disk */
1764 884 : XLogFlush(flushptr);
1765 : }
1766 1248 : }
1767 :
1768 : /*
1769 : * Report that replication slot needs to be invalidated
1770 : */
1771 : static void
1772 46 : ReportSlotInvalidation(ReplicationSlotInvalidationCause cause,
1773 : bool terminating,
1774 : int pid,
1775 : NameData slotname,
1776 : XLogRecPtr restart_lsn,
1777 : XLogRecPtr oldestLSN,
1778 : TransactionId snapshotConflictHorizon,
1779 : long slot_idle_seconds)
1780 : {
1781 : StringInfoData err_detail;
1782 : StringInfoData err_hint;
1783 :
1784 46 : initStringInfo(&err_detail);
1785 46 : initStringInfo(&err_hint);
1786 :
1787 46 : switch (cause)
1788 : {
1789 14 : case RS_INVAL_WAL_REMOVED:
1790 : {
1791 14 : uint64 ex = oldestLSN - restart_lsn;
1792 :
1793 14 : appendStringInfo(&err_detail,
1794 14 : ngettext("The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " byte.",
1795 : "The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " bytes.",
1796 : ex),
1797 14 : LSN_FORMAT_ARGS(restart_lsn),
1798 : ex);
1799 : /* translator: %s is a GUC variable name */
1800 14 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1801 : "max_slot_wal_keep_size");
1802 14 : break;
1803 : }
1804 24 : case RS_INVAL_HORIZON:
1805 24 : appendStringInfo(&err_detail, _("The slot conflicted with xid horizon %u."),
1806 : snapshotConflictHorizon);
1807 24 : break;
1808 :
1809 8 : case RS_INVAL_WAL_LEVEL:
1810 8 : appendStringInfoString(&err_detail, _("Logical decoding on standby requires the primary server to either set \"wal_level\" >= \"logical\" or have at least one logical slot when \"wal_level\" = \"replica\"."));
1811 8 : break;
1812 :
1813 0 : case RS_INVAL_IDLE_TIMEOUT:
1814 : {
1815 : /* translator: %s is a GUC variable name */
1816 0 : appendStringInfo(&err_detail, _("The slot's idle time of %lds exceeds the configured \"%s\" duration of %ds."),
1817 : slot_idle_seconds, "idle_replication_slot_timeout",
1818 : idle_replication_slot_timeout_secs);
1819 : /* translator: %s is a GUC variable name */
1820 0 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1821 : "idle_replication_slot_timeout");
1822 0 : break;
1823 : }
1824 : case RS_INVAL_NONE:
1825 : pg_unreachable();
1826 : }
1827 :
1828 46 : ereport(LOG,
1829 : terminating ?
1830 : errmsg("terminating process %d to release replication slot \"%s\"",
1831 : pid, NameStr(slotname)) :
1832 : errmsg("invalidating obsolete replication slot \"%s\"",
1833 : NameStr(slotname)),
1834 : errdetail_internal("%s", err_detail.data),
1835 : err_hint.len ? errhint("%s", err_hint.data) : 0);
1836 :
1837 46 : pfree(err_detail.data);
1838 46 : pfree(err_hint.data);
1839 46 : }
1840 :
1841 : /*
1842 : * Can we invalidate an idle replication slot?
1843 : *
1844 : * Idle timeout invalidation is allowed only when:
1845 : *
1846 : * 1. Idle timeout is set
1847 : * 2. Slot has reserved WAL
1848 : * 3. Slot is inactive
1849 : * 4. The slot is not being synced from the primary while the server is in
1850 : * recovery. This is because synced slots are always considered to be
1851 : * inactive because they don't perform logical decoding to produce changes.
1852 : */
1853 : static inline bool
1854 736 : CanInvalidateIdleSlot(ReplicationSlot *s)
1855 : {
1856 736 : return (idle_replication_slot_timeout_secs != 0 &&
1857 0 : XLogRecPtrIsValid(s->data.restart_lsn) &&
1858 736 : s->inactive_since > 0 &&
1859 0 : !(RecoveryInProgress() && s->data.synced));
1860 : }
1861 :
1862 : /*
1863 : * DetermineSlotInvalidationCause - Determine the cause for which a slot
1864 : * becomes invalid among the given possible causes.
1865 : *
1866 : * This function sequentially checks all possible invalidation causes and
1867 : * returns the first one for which the slot is eligible for invalidation.
1868 : */
1869 : static ReplicationSlotInvalidationCause
1870 802 : DetermineSlotInvalidationCause(uint32 possible_causes, ReplicationSlot *s,
1871 : XLogRecPtr oldestLSN, Oid dboid,
1872 : TransactionId snapshotConflictHorizon,
1873 : TimestampTz *inactive_since, TimestampTz now)
1874 : {
1875 : Assert(possible_causes != RS_INVAL_NONE);
1876 :
1877 802 : if (possible_causes & RS_INVAL_WAL_REMOVED)
1878 : {
1879 750 : XLogRecPtr restart_lsn = s->data.restart_lsn;
1880 :
1881 750 : if (XLogRecPtrIsValid(restart_lsn) &&
1882 : restart_lsn < oldestLSN)
1883 14 : return RS_INVAL_WAL_REMOVED;
1884 : }
1885 :
1886 788 : if (possible_causes & RS_INVAL_HORIZON)
1887 : {
1888 : /* invalid DB oid signals a shared relation */
1889 44 : if (SlotIsLogical(s) &&
1890 34 : (dboid == InvalidOid || dboid == s->data.database))
1891 : {
1892 44 : TransactionId effective_xmin = s->effective_xmin;
1893 44 : TransactionId catalog_effective_xmin = s->effective_catalog_xmin;
1894 :
1895 44 : if (TransactionIdIsValid(effective_xmin) &&
1896 0 : TransactionIdPrecedesOrEquals(effective_xmin,
1897 : snapshotConflictHorizon))
1898 0 : return RS_INVAL_HORIZON;
1899 88 : else if (TransactionIdIsValid(catalog_effective_xmin) &&
1900 44 : TransactionIdPrecedesOrEquals(catalog_effective_xmin,
1901 : snapshotConflictHorizon))
1902 24 : return RS_INVAL_HORIZON;
1903 : }
1904 : }
1905 :
1906 764 : if (possible_causes & RS_INVAL_WAL_LEVEL)
1907 : {
1908 8 : if (SlotIsLogical(s))
1909 8 : return RS_INVAL_WAL_LEVEL;
1910 : }
1911 :
1912 756 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1913 : {
1914 : Assert(now > 0);
1915 :
1916 736 : if (CanInvalidateIdleSlot(s))
1917 : {
1918 : /*
1919 : * Simulate the invalidation due to idle_timeout to test the
1920 : * timeout behavior promptly, without waiting for it to trigger
1921 : * naturally.
1922 : */
1923 : #ifdef USE_INJECTION_POINTS
1924 0 : if (IS_INJECTION_POINT_ATTACHED("slot-timeout-inval"))
1925 : {
1926 0 : *inactive_since = 0; /* since the beginning of time */
1927 0 : return RS_INVAL_IDLE_TIMEOUT;
1928 : }
1929 : #endif
1930 :
1931 : /*
1932 : * Check if the slot needs to be invalidated due to
1933 : * idle_replication_slot_timeout GUC.
1934 : */
1935 0 : if (TimestampDifferenceExceedsSeconds(s->inactive_since, now,
1936 : idle_replication_slot_timeout_secs))
1937 : {
1938 0 : *inactive_since = s->inactive_since;
1939 0 : return RS_INVAL_IDLE_TIMEOUT;
1940 : }
1941 : }
1942 : }
1943 :
1944 756 : return RS_INVAL_NONE;
1945 : }
1946 :
1947 : /*
1948 : * Helper for InvalidateObsoleteReplicationSlots
1949 : *
1950 : * Acquires the given slot and mark it invalid, if necessary and possible.
1951 : *
1952 : * Returns true if the slot was invalidated.
1953 : *
1954 : * Set *released_lock_out if ReplicationSlotControlLock was released in the
1955 : * interim (and in that case we're not holding the lock at return, otherwise
1956 : * we are).
1957 : *
1958 : * This is inherently racy, because we release the LWLock
1959 : * for syscalls, so caller must restart if we return true.
1960 : */
1961 : static bool
1962 880 : InvalidatePossiblyObsoleteSlot(uint32 possible_causes,
1963 : ReplicationSlot *s,
1964 : XLogRecPtr oldestLSN,
1965 : Oid dboid, TransactionId snapshotConflictHorizon,
1966 : bool *released_lock_out)
1967 : {
1968 880 : int last_signaled_pid = 0;
1969 880 : bool released_lock = false;
1970 880 : bool invalidated = false;
1971 880 : TimestampTz inactive_since = 0;
1972 :
1973 : for (;;)
1974 14 : {
1975 : XLogRecPtr restart_lsn;
1976 : NameData slotname;
1977 894 : int active_pid = 0;
1978 894 : ReplicationSlotInvalidationCause invalidation_cause = RS_INVAL_NONE;
1979 894 : TimestampTz now = 0;
1980 894 : long slot_idle_secs = 0;
1981 :
1982 : Assert(LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_SHARED));
1983 :
1984 894 : if (!s->in_use)
1985 : {
1986 0 : if (released_lock)
1987 0 : LWLockRelease(ReplicationSlotControlLock);
1988 0 : break;
1989 : }
1990 :
1991 894 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1992 : {
1993 : /*
1994 : * Assign the current time here to avoid system call overhead
1995 : * while holding the spinlock in subsequent code.
1996 : */
1997 778 : now = GetCurrentTimestamp();
1998 : }
1999 :
2000 : /*
2001 : * Check if the slot needs to be invalidated. If it needs to be
2002 : * invalidated, and is not currently acquired, acquire it and mark it
2003 : * as having been invalidated. We do this with the spinlock held to
2004 : * avoid race conditions -- for example the restart_lsn could move
2005 : * forward, or the slot could be dropped.
2006 : */
2007 894 : SpinLockAcquire(&s->mutex);
2008 :
2009 894 : restart_lsn = s->data.restart_lsn;
2010 :
2011 : /* we do nothing if the slot is already invalid */
2012 894 : if (s->data.invalidated == RS_INVAL_NONE)
2013 802 : invalidation_cause = DetermineSlotInvalidationCause(possible_causes,
2014 : s, oldestLSN,
2015 : dboid,
2016 : snapshotConflictHorizon,
2017 : &inactive_since,
2018 : now);
2019 :
2020 : /* if there's no invalidation, we're done */
2021 894 : if (invalidation_cause == RS_INVAL_NONE)
2022 : {
2023 848 : SpinLockRelease(&s->mutex);
2024 848 : if (released_lock)
2025 0 : LWLockRelease(ReplicationSlotControlLock);
2026 848 : break;
2027 : }
2028 :
2029 46 : slotname = s->data.name;
2030 46 : active_pid = s->active_pid;
2031 :
2032 : /*
2033 : * If the slot can be acquired, do so and mark it invalidated
2034 : * immediately. Otherwise we'll signal the owning process, below, and
2035 : * retry.
2036 : *
2037 : * Note: Unlike other slot attributes, slot's inactive_since can't be
2038 : * changed until the acquired slot is released or the owning process
2039 : * is terminated. So, the inactive slot can only be invalidated
2040 : * immediately without being terminated.
2041 : */
2042 46 : if (active_pid == 0)
2043 : {
2044 32 : MyReplicationSlot = s;
2045 32 : s->active_pid = MyProcPid;
2046 32 : s->data.invalidated = invalidation_cause;
2047 :
2048 : /*
2049 : * XXX: We should consider not overwriting restart_lsn and instead
2050 : * just rely on .invalidated.
2051 : */
2052 32 : if (invalidation_cause == RS_INVAL_WAL_REMOVED)
2053 : {
2054 10 : s->data.restart_lsn = InvalidXLogRecPtr;
2055 10 : s->last_saved_restart_lsn = InvalidXLogRecPtr;
2056 : }
2057 :
2058 : /* Let caller know */
2059 32 : invalidated = true;
2060 : }
2061 :
2062 46 : SpinLockRelease(&s->mutex);
2063 :
2064 : /*
2065 : * Calculate the idle time duration of the slot if slot is marked
2066 : * invalidated with RS_INVAL_IDLE_TIMEOUT.
2067 : */
2068 46 : if (invalidation_cause == RS_INVAL_IDLE_TIMEOUT)
2069 : {
2070 : int slot_idle_usecs;
2071 :
2072 0 : TimestampDifference(inactive_since, now, &slot_idle_secs,
2073 : &slot_idle_usecs);
2074 : }
2075 :
2076 46 : if (active_pid != 0)
2077 : {
2078 : /*
2079 : * Prepare the sleep on the slot's condition variable before
2080 : * releasing the lock, to close a possible race condition if the
2081 : * slot is released before the sleep below.
2082 : */
2083 14 : ConditionVariablePrepareToSleep(&s->active_cv);
2084 :
2085 14 : LWLockRelease(ReplicationSlotControlLock);
2086 14 : released_lock = true;
2087 :
2088 : /*
2089 : * Signal to terminate the process that owns the slot, if we
2090 : * haven't already signalled it. (Avoidance of repeated
2091 : * signalling is the only reason for there to be a loop in this
2092 : * routine; otherwise we could rely on caller's restart loop.)
2093 : *
2094 : * There is the race condition that other process may own the slot
2095 : * after its current owner process is terminated and before this
2096 : * process owns it. To handle that, we signal only if the PID of
2097 : * the owning process has changed from the previous time. (This
2098 : * logic assumes that the same PID is not reused very quickly.)
2099 : */
2100 14 : if (last_signaled_pid != active_pid)
2101 : {
2102 14 : ReportSlotInvalidation(invalidation_cause, true, active_pid,
2103 : slotname, restart_lsn,
2104 : oldestLSN, snapshotConflictHorizon,
2105 : slot_idle_secs);
2106 :
2107 14 : if (MyBackendType == B_STARTUP)
2108 10 : (void) SendProcSignal(active_pid,
2109 : PROCSIG_RECOVERY_CONFLICT_LOGICALSLOT,
2110 : INVALID_PROC_NUMBER);
2111 : else
2112 4 : (void) kill(active_pid, SIGTERM);
2113 :
2114 14 : last_signaled_pid = active_pid;
2115 : }
2116 :
2117 : /* Wait until the slot is released. */
2118 14 : ConditionVariableSleep(&s->active_cv,
2119 : WAIT_EVENT_REPLICATION_SLOT_DROP);
2120 :
2121 : /*
2122 : * Re-acquire lock and start over; we expect to invalidate the
2123 : * slot next time (unless another process acquires the slot in the
2124 : * meantime).
2125 : *
2126 : * Note: It is possible for a slot to advance its restart_lsn or
2127 : * xmin values sufficiently between when we release the mutex and
2128 : * when we recheck, moving from a conflicting state to a non
2129 : * conflicting state. This is intentional and safe: if the slot
2130 : * has caught up while we're busy here, the resources we were
2131 : * concerned about (WAL segments or tuples) have not yet been
2132 : * removed, and there's no reason to invalidate the slot.
2133 : */
2134 14 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2135 14 : continue;
2136 : }
2137 : else
2138 : {
2139 : /*
2140 : * We hold the slot now and have already invalidated it; flush it
2141 : * to ensure that state persists.
2142 : *
2143 : * Don't want to hold ReplicationSlotControlLock across file
2144 : * system operations, so release it now but be sure to tell caller
2145 : * to restart from scratch.
2146 : */
2147 32 : LWLockRelease(ReplicationSlotControlLock);
2148 32 : released_lock = true;
2149 :
2150 : /* Make sure the invalidated state persists across server restart */
2151 32 : ReplicationSlotMarkDirty();
2152 32 : ReplicationSlotSave();
2153 32 : ReplicationSlotRelease();
2154 :
2155 32 : ReportSlotInvalidation(invalidation_cause, false, active_pid,
2156 : slotname, restart_lsn,
2157 : oldestLSN, snapshotConflictHorizon,
2158 : slot_idle_secs);
2159 :
2160 : /* done with this slot for now */
2161 32 : break;
2162 : }
2163 : }
2164 :
2165 : Assert(released_lock == !LWLockHeldByMe(ReplicationSlotControlLock));
2166 :
2167 880 : *released_lock_out = released_lock;
2168 880 : return invalidated;
2169 : }
2170 :
2171 : /*
2172 : * Invalidate slots that require resources about to be removed.
2173 : *
2174 : * Returns true when any slot have got invalidated.
2175 : *
2176 : * Whether a slot needs to be invalidated depends on the invalidation cause.
2177 : * A slot is invalidated if it:
2178 : * - RS_INVAL_WAL_REMOVED: requires a LSN older than the given segment
2179 : * - RS_INVAL_HORIZON: requires a snapshot <= the given horizon in the given
2180 : * db; dboid may be InvalidOid for shared relations
2181 : * - RS_INVAL_WAL_LEVEL: is a logical slot and effective_wal_level is not
2182 : * logical.
2183 : * - RS_INVAL_IDLE_TIMEOUT: has been idle longer than the configured
2184 : * "idle_replication_slot_timeout" duration.
2185 : *
2186 : * Note: This function attempts to invalidate the slot for multiple possible
2187 : * causes in a single pass, minimizing redundant iterations. The "cause"
2188 : * parameter can be a MASK representing one or more of the defined causes.
2189 : *
2190 : * If it invalidates the last logical slot in the cluster, it requests to
2191 : * disable logical decoding.
2192 : *
2193 : * NB - this runs as part of checkpoint, so avoid raising errors if possible.
2194 : */
2195 : bool
2196 3598 : InvalidateObsoleteReplicationSlots(uint32 possible_causes,
2197 : XLogSegNo oldestSegno, Oid dboid,
2198 : TransactionId snapshotConflictHorizon)
2199 : {
2200 : XLogRecPtr oldestLSN;
2201 3598 : bool invalidated = false;
2202 3598 : bool invalidated_logical = false;
2203 : bool found_valid_logicalslot;
2204 :
2205 : Assert(!(possible_causes & RS_INVAL_HORIZON) || TransactionIdIsValid(snapshotConflictHorizon));
2206 : Assert(!(possible_causes & RS_INVAL_WAL_REMOVED) || oldestSegno > 0);
2207 : Assert(possible_causes != RS_INVAL_NONE);
2208 :
2209 3598 : if (max_replication_slots == 0)
2210 2 : return invalidated;
2211 :
2212 3596 : XLogSegNoOffsetToRecPtr(oldestSegno, 0, wal_segment_size, oldestLSN);
2213 :
2214 3628 : restart:
2215 3628 : found_valid_logicalslot = false;
2216 3628 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2217 38882 : for (int i = 0; i < max_replication_slots; i++)
2218 : {
2219 35286 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2220 35286 : bool released_lock = false;
2221 :
2222 35286 : if (!s->in_use)
2223 34406 : continue;
2224 :
2225 : /* Prevent invalidation of logical slots during binary upgrade */
2226 894 : if (SlotIsLogical(s) && IsBinaryUpgrade)
2227 : {
2228 14 : SpinLockAcquire(&s->mutex);
2229 14 : found_valid_logicalslot |= (s->data.invalidated == RS_INVAL_NONE);
2230 14 : SpinLockRelease(&s->mutex);
2231 :
2232 14 : continue;
2233 : }
2234 :
2235 880 : if (InvalidatePossiblyObsoleteSlot(possible_causes, s, oldestLSN,
2236 : dboid, snapshotConflictHorizon,
2237 : &released_lock))
2238 : {
2239 : Assert(released_lock);
2240 :
2241 : /* Remember we have invalidated a physical or logical slot */
2242 32 : invalidated = true;
2243 :
2244 : /*
2245 : * Additionally, remember we have invalidated a logical slot as we
2246 : * can request disabling logical decoding later.
2247 : */
2248 32 : if (SlotIsLogical(s))
2249 26 : invalidated_logical = true;
2250 : }
2251 : else
2252 : {
2253 : /*
2254 : * We need to check if the slot is invalidated here since
2255 : * InvalidatePossiblyObsoleteSlot() returns false also if the slot
2256 : * is already invalidated.
2257 : */
2258 848 : SpinLockAcquire(&s->mutex);
2259 1696 : found_valid_logicalslot |=
2260 848 : (SlotIsLogical(s) && (s->data.invalidated == RS_INVAL_NONE));
2261 848 : SpinLockRelease(&s->mutex);
2262 : }
2263 :
2264 : /* if the lock was released, start from scratch */
2265 880 : if (released_lock)
2266 32 : goto restart;
2267 : }
2268 3596 : LWLockRelease(ReplicationSlotControlLock);
2269 :
2270 : /*
2271 : * If any slots have been invalidated, recalculate the resource limits.
2272 : */
2273 3596 : if (invalidated)
2274 : {
2275 22 : ReplicationSlotsComputeRequiredXmin(false);
2276 22 : ReplicationSlotsComputeRequiredLSN();
2277 : }
2278 :
2279 : /*
2280 : * Request the checkpointer to disable logical decoding if no valid
2281 : * logical slots remain. If called by the checkpointer during a
2282 : * checkpoint, only the request is initiated; actual deactivation is
2283 : * deferred until after the checkpoint completes.
2284 : */
2285 3596 : if (invalidated_logical && !found_valid_logicalslot)
2286 16 : RequestDisableLogicalDecoding();
2287 :
2288 3596 : return invalidated;
2289 : }
2290 :
2291 : /*
2292 : * Flush all replication slots to disk.
2293 : *
2294 : * It is convenient to flush dirty replication slots at the time of checkpoint.
2295 : * Additionally, in case of a shutdown checkpoint, we also identify the slots
2296 : * for which the confirmed_flush LSN has been updated since the last time it
2297 : * was saved and flush them.
2298 : */
2299 : void
2300 3548 : CheckPointReplicationSlots(bool is_shutdown)
2301 : {
2302 : int i;
2303 3548 : bool last_saved_restart_lsn_updated = false;
2304 :
2305 3548 : elog(DEBUG1, "performing replication slot checkpoint");
2306 :
2307 : /*
2308 : * Prevent any slot from being created/dropped while we're active. As we
2309 : * explicitly do *not* want to block iterating over replication_slots or
2310 : * acquiring a slot we cannot take the control lock - but that's OK,
2311 : * because holding ReplicationSlotAllocationLock is strictly stronger, and
2312 : * enough to guarantee that nobody can change the in_use bits on us.
2313 : *
2314 : * Additionally, acquiring the Allocation lock is necessary to serialize
2315 : * the slot flush process with concurrent slot WAL reservation. This
2316 : * ensures that the WAL position being reserved is either flushed to disk
2317 : * or is beyond or equal to the redo pointer of the current checkpoint
2318 : * (See ReplicationSlotReserveWal for details).
2319 : */
2320 3548 : LWLockAcquire(ReplicationSlotAllocationLock, LW_SHARED);
2321 :
2322 38462 : for (i = 0; i < max_replication_slots; i++)
2323 : {
2324 34914 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2325 : char path[MAXPGPATH];
2326 :
2327 34914 : if (!s->in_use)
2328 34136 : continue;
2329 :
2330 : /* save the slot to disk, locking is handled in SaveSlotToPath() */
2331 778 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(s->data.name));
2332 :
2333 : /*
2334 : * Slot's data is not flushed each time the confirmed_flush LSN is
2335 : * updated as that could lead to frequent writes. However, we decide
2336 : * to force a flush of all logical slot's data at the time of shutdown
2337 : * if the confirmed_flush LSN is changed since we last flushed it to
2338 : * disk. This helps in avoiding an unnecessary retreat of the
2339 : * confirmed_flush LSN after restart.
2340 : */
2341 778 : if (is_shutdown && SlotIsLogical(s))
2342 : {
2343 168 : SpinLockAcquire(&s->mutex);
2344 :
2345 168 : if (s->data.invalidated == RS_INVAL_NONE &&
2346 166 : s->data.confirmed_flush > s->last_saved_confirmed_flush)
2347 : {
2348 72 : s->just_dirtied = true;
2349 72 : s->dirty = true;
2350 : }
2351 168 : SpinLockRelease(&s->mutex);
2352 : }
2353 :
2354 : /*
2355 : * Track if we're going to update slot's last_saved_restart_lsn. We
2356 : * need this to know if we need to recompute the required LSN.
2357 : */
2358 778 : if (s->last_saved_restart_lsn != s->data.restart_lsn)
2359 392 : last_saved_restart_lsn_updated = true;
2360 :
2361 778 : SaveSlotToPath(s, path, LOG);
2362 : }
2363 3548 : LWLockRelease(ReplicationSlotAllocationLock);
2364 :
2365 : /*
2366 : * Recompute the required LSN if SaveSlotToPath() updated
2367 : * last_saved_restart_lsn for any slot.
2368 : */
2369 3548 : if (last_saved_restart_lsn_updated)
2370 392 : ReplicationSlotsComputeRequiredLSN();
2371 3548 : }
2372 :
2373 : /*
2374 : * Load all replication slots from disk into memory at server startup. This
2375 : * needs to be run before we start crash recovery.
2376 : */
2377 : void
2378 1982 : StartupReplicationSlots(void)
2379 : {
2380 : DIR *replication_dir;
2381 : struct dirent *replication_de;
2382 :
2383 1982 : elog(DEBUG1, "starting up replication slots");
2384 :
2385 : /* restore all slots by iterating over all on-disk entries */
2386 1982 : replication_dir = AllocateDir(PG_REPLSLOT_DIR);
2387 6168 : while ((replication_de = ReadDir(replication_dir, PG_REPLSLOT_DIR)) != NULL)
2388 : {
2389 : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2390 : PGFileType de_type;
2391 :
2392 4190 : if (strcmp(replication_de->d_name, ".") == 0 ||
2393 2212 : strcmp(replication_de->d_name, "..") == 0)
2394 3956 : continue;
2395 :
2396 234 : snprintf(path, sizeof(path), "%s/%s", PG_REPLSLOT_DIR, replication_de->d_name);
2397 234 : de_type = get_dirent_type(path, replication_de, false, DEBUG1);
2398 :
2399 : /* we're only creating directories here, skip if it's not our's */
2400 234 : if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_DIR)
2401 0 : continue;
2402 :
2403 : /* we crashed while a slot was being setup or deleted, clean up */
2404 234 : if (pg_str_endswith(replication_de->d_name, ".tmp"))
2405 : {
2406 0 : if (!rmtree(path, true))
2407 : {
2408 0 : ereport(WARNING,
2409 : (errmsg("could not remove directory \"%s\"",
2410 : path)));
2411 0 : continue;
2412 : }
2413 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2414 0 : continue;
2415 : }
2416 :
2417 : /* looks like a slot in a normal state, restore */
2418 234 : RestoreSlotFromDisk(replication_de->d_name);
2419 : }
2420 1978 : FreeDir(replication_dir);
2421 :
2422 : /* currently no slots exist, we're done. */
2423 1978 : if (max_replication_slots <= 0)
2424 2 : return;
2425 :
2426 : /* Now that we have recovered all the data, compute replication xmin */
2427 1976 : ReplicationSlotsComputeRequiredXmin(false);
2428 1976 : ReplicationSlotsComputeRequiredLSN();
2429 : }
2430 :
2431 : /* ----
2432 : * Manipulation of on-disk state of replication slots
2433 : *
2434 : * NB: none of the routines below should take any notice whether a slot is the
2435 : * current one or not, that's all handled a layer above.
2436 : * ----
2437 : */
2438 : static void
2439 1346 : CreateSlotOnDisk(ReplicationSlot *slot)
2440 : {
2441 : char tmppath[MAXPGPATH];
2442 : char path[MAXPGPATH];
2443 : struct stat st;
2444 :
2445 : /*
2446 : * No need to take out the io_in_progress_lock, nobody else can see this
2447 : * slot yet, so nobody else will write. We're reusing SaveSlotToPath which
2448 : * takes out the lock, if we'd take the lock here, we'd deadlock.
2449 : */
2450 :
2451 1346 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2452 1346 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2453 :
2454 : /*
2455 : * It's just barely possible that some previous effort to create or drop a
2456 : * slot with this name left a temp directory lying around. If that seems
2457 : * to be the case, try to remove it. If the rmtree() fails, we'll error
2458 : * out at the MakePGDirectory() below, so we don't bother checking
2459 : * success.
2460 : */
2461 1346 : if (stat(tmppath, &st) == 0 && S_ISDIR(st.st_mode))
2462 0 : rmtree(tmppath, true);
2463 :
2464 : /* Create and fsync the temporary slot directory. */
2465 1346 : if (MakePGDirectory(tmppath) < 0)
2466 0 : ereport(ERROR,
2467 : (errcode_for_file_access(),
2468 : errmsg("could not create directory \"%s\": %m",
2469 : tmppath)));
2470 1346 : fsync_fname(tmppath, true);
2471 :
2472 : /* Write the actual state file. */
2473 1346 : slot->dirty = true; /* signal that we really need to write */
2474 1346 : SaveSlotToPath(slot, tmppath, ERROR);
2475 :
2476 : /* Rename the directory into place. */
2477 1346 : if (rename(tmppath, path) != 0)
2478 0 : ereport(ERROR,
2479 : (errcode_for_file_access(),
2480 : errmsg("could not rename file \"%s\" to \"%s\": %m",
2481 : tmppath, path)));
2482 :
2483 : /*
2484 : * If we'd now fail - really unlikely - we wouldn't know whether this slot
2485 : * would persist after an OS crash or not - so, force a restart. The
2486 : * restart would try to fsync this again till it works.
2487 : */
2488 1346 : START_CRIT_SECTION();
2489 :
2490 1346 : fsync_fname(path, true);
2491 1346 : fsync_fname(PG_REPLSLOT_DIR, true);
2492 :
2493 1346 : END_CRIT_SECTION();
2494 1346 : }
2495 :
2496 : /*
2497 : * Shared functionality between saving and creating a replication slot.
2498 : */
2499 : static void
2500 4922 : SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
2501 : {
2502 : char tmppath[MAXPGPATH];
2503 : char path[MAXPGPATH];
2504 : int fd;
2505 : ReplicationSlotOnDisk cp;
2506 : bool was_dirty;
2507 :
2508 : /* first check whether there's something to write out */
2509 4922 : SpinLockAcquire(&slot->mutex);
2510 4922 : was_dirty = slot->dirty;
2511 4922 : slot->just_dirtied = false;
2512 4922 : SpinLockRelease(&slot->mutex);
2513 :
2514 : /* and don't do anything if there's nothing to write */
2515 4922 : if (!was_dirty)
2516 292 : return;
2517 :
2518 4630 : LWLockAcquire(&slot->io_in_progress_lock, LW_EXCLUSIVE);
2519 :
2520 : /* silence valgrind :( */
2521 4630 : memset(&cp, 0, sizeof(ReplicationSlotOnDisk));
2522 :
2523 4630 : sprintf(tmppath, "%s/state.tmp", dir);
2524 4630 : sprintf(path, "%s/state", dir);
2525 :
2526 4630 : fd = OpenTransientFile(tmppath, O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
2527 4630 : if (fd < 0)
2528 : {
2529 : /*
2530 : * If not an ERROR, then release the lock before returning. In case
2531 : * of an ERROR, the error recovery path automatically releases the
2532 : * lock, but no harm in explicitly releasing even in that case. Note
2533 : * that LWLockRelease() could affect errno.
2534 : */
2535 0 : int save_errno = errno;
2536 :
2537 0 : LWLockRelease(&slot->io_in_progress_lock);
2538 0 : errno = save_errno;
2539 0 : ereport(elevel,
2540 : (errcode_for_file_access(),
2541 : errmsg("could not create file \"%s\": %m",
2542 : tmppath)));
2543 0 : return;
2544 : }
2545 :
2546 4630 : cp.magic = SLOT_MAGIC;
2547 4630 : INIT_CRC32C(cp.checksum);
2548 4630 : cp.version = SLOT_VERSION;
2549 4630 : cp.length = ReplicationSlotOnDiskV2Size;
2550 :
2551 4630 : SpinLockAcquire(&slot->mutex);
2552 :
2553 4630 : memcpy(&cp.slotdata, &slot->data, sizeof(ReplicationSlotPersistentData));
2554 :
2555 4630 : SpinLockRelease(&slot->mutex);
2556 :
2557 4630 : COMP_CRC32C(cp.checksum,
2558 : (char *) (&cp) + ReplicationSlotOnDiskNotChecksummedSize,
2559 : ReplicationSlotOnDiskChecksummedSize);
2560 4630 : FIN_CRC32C(cp.checksum);
2561 :
2562 4630 : errno = 0;
2563 4630 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_WRITE);
2564 4630 : if ((write(fd, &cp, sizeof(cp))) != sizeof(cp))
2565 : {
2566 0 : int save_errno = errno;
2567 :
2568 0 : pgstat_report_wait_end();
2569 0 : CloseTransientFile(fd);
2570 0 : unlink(tmppath);
2571 0 : LWLockRelease(&slot->io_in_progress_lock);
2572 :
2573 : /* if write didn't set errno, assume problem is no disk space */
2574 0 : errno = save_errno ? save_errno : ENOSPC;
2575 0 : ereport(elevel,
2576 : (errcode_for_file_access(),
2577 : errmsg("could not write to file \"%s\": %m",
2578 : tmppath)));
2579 0 : return;
2580 : }
2581 4630 : pgstat_report_wait_end();
2582 :
2583 : /* fsync the temporary file */
2584 4630 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_SYNC);
2585 4630 : if (pg_fsync(fd) != 0)
2586 : {
2587 0 : int save_errno = errno;
2588 :
2589 0 : pgstat_report_wait_end();
2590 0 : CloseTransientFile(fd);
2591 0 : unlink(tmppath);
2592 0 : LWLockRelease(&slot->io_in_progress_lock);
2593 :
2594 0 : errno = save_errno;
2595 0 : ereport(elevel,
2596 : (errcode_for_file_access(),
2597 : errmsg("could not fsync file \"%s\": %m",
2598 : tmppath)));
2599 0 : return;
2600 : }
2601 4630 : pgstat_report_wait_end();
2602 :
2603 4630 : if (CloseTransientFile(fd) != 0)
2604 : {
2605 0 : int save_errno = errno;
2606 :
2607 0 : unlink(tmppath);
2608 0 : LWLockRelease(&slot->io_in_progress_lock);
2609 :
2610 0 : errno = save_errno;
2611 0 : ereport(elevel,
2612 : (errcode_for_file_access(),
2613 : errmsg("could not close file \"%s\": %m",
2614 : tmppath)));
2615 0 : return;
2616 : }
2617 :
2618 : /* rename to permanent file, fsync file and directory */
2619 4630 : if (rename(tmppath, path) != 0)
2620 : {
2621 0 : int save_errno = errno;
2622 :
2623 0 : unlink(tmppath);
2624 0 : LWLockRelease(&slot->io_in_progress_lock);
2625 :
2626 0 : errno = save_errno;
2627 0 : ereport(elevel,
2628 : (errcode_for_file_access(),
2629 : errmsg("could not rename file \"%s\" to \"%s\": %m",
2630 : tmppath, path)));
2631 0 : return;
2632 : }
2633 :
2634 : /*
2635 : * Check CreateSlotOnDisk() for the reasoning of using a critical section.
2636 : */
2637 4630 : START_CRIT_SECTION();
2638 :
2639 4630 : fsync_fname(path, false);
2640 4630 : fsync_fname(dir, true);
2641 4630 : fsync_fname(PG_REPLSLOT_DIR, true);
2642 :
2643 4630 : END_CRIT_SECTION();
2644 :
2645 : /*
2646 : * Successfully wrote, unset dirty bit, unless somebody dirtied again
2647 : * already and remember the confirmed_flush LSN value.
2648 : */
2649 4630 : SpinLockAcquire(&slot->mutex);
2650 4630 : if (!slot->just_dirtied)
2651 4608 : slot->dirty = false;
2652 4630 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2653 4630 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2654 4630 : SpinLockRelease(&slot->mutex);
2655 :
2656 4630 : LWLockRelease(&slot->io_in_progress_lock);
2657 : }
2658 :
2659 : /*
2660 : * Load a single slot from disk into memory.
2661 : */
2662 : static void
2663 234 : RestoreSlotFromDisk(const char *name)
2664 : {
2665 : ReplicationSlotOnDisk cp;
2666 : int i;
2667 : char slotdir[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2668 : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR) + 10];
2669 : int fd;
2670 234 : bool restored = false;
2671 : int readBytes;
2672 : pg_crc32c checksum;
2673 234 : TimestampTz now = 0;
2674 :
2675 : /* no need to lock here, no concurrent access allowed yet */
2676 :
2677 : /* delete temp file if it exists */
2678 234 : sprintf(slotdir, "%s/%s", PG_REPLSLOT_DIR, name);
2679 234 : sprintf(path, "%s/state.tmp", slotdir);
2680 234 : if (unlink(path) < 0 && errno != ENOENT)
2681 0 : ereport(PANIC,
2682 : (errcode_for_file_access(),
2683 : errmsg("could not remove file \"%s\": %m", path)));
2684 :
2685 234 : sprintf(path, "%s/state", slotdir);
2686 :
2687 234 : elog(DEBUG1, "restoring replication slot from \"%s\"", path);
2688 :
2689 : /* on some operating systems fsyncing a file requires O_RDWR */
2690 234 : fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
2691 :
2692 : /*
2693 : * We do not need to handle this as we are rename()ing the directory into
2694 : * place only after we fsync()ed the state file.
2695 : */
2696 234 : if (fd < 0)
2697 0 : ereport(PANIC,
2698 : (errcode_for_file_access(),
2699 : errmsg("could not open file \"%s\": %m", path)));
2700 :
2701 : /*
2702 : * Sync state file before we're reading from it. We might have crashed
2703 : * while it wasn't synced yet and we shouldn't continue on that basis.
2704 : */
2705 234 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_RESTORE_SYNC);
2706 234 : if (pg_fsync(fd) != 0)
2707 0 : ereport(PANIC,
2708 : (errcode_for_file_access(),
2709 : errmsg("could not fsync file \"%s\": %m",
2710 : path)));
2711 234 : pgstat_report_wait_end();
2712 :
2713 : /* Also sync the parent directory */
2714 234 : START_CRIT_SECTION();
2715 234 : fsync_fname(slotdir, true);
2716 234 : END_CRIT_SECTION();
2717 :
2718 : /* read part of statefile that's guaranteed to be version independent */
2719 234 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2720 234 : readBytes = read(fd, &cp, ReplicationSlotOnDiskConstantSize);
2721 234 : pgstat_report_wait_end();
2722 234 : if (readBytes != ReplicationSlotOnDiskConstantSize)
2723 : {
2724 0 : if (readBytes < 0)
2725 0 : ereport(PANIC,
2726 : (errcode_for_file_access(),
2727 : errmsg("could not read file \"%s\": %m", path)));
2728 : else
2729 0 : ereport(PANIC,
2730 : (errcode(ERRCODE_DATA_CORRUPTED),
2731 : errmsg("could not read file \"%s\": read %d of %zu",
2732 : path, readBytes,
2733 : (Size) ReplicationSlotOnDiskConstantSize)));
2734 : }
2735 :
2736 : /* verify magic */
2737 234 : if (cp.magic != SLOT_MAGIC)
2738 0 : ereport(PANIC,
2739 : (errcode(ERRCODE_DATA_CORRUPTED),
2740 : errmsg("replication slot file \"%s\" has wrong magic number: %u instead of %u",
2741 : path, cp.magic, SLOT_MAGIC)));
2742 :
2743 : /* verify version */
2744 234 : if (cp.version != SLOT_VERSION)
2745 0 : ereport(PANIC,
2746 : (errcode(ERRCODE_DATA_CORRUPTED),
2747 : errmsg("replication slot file \"%s\" has unsupported version %u",
2748 : path, cp.version)));
2749 :
2750 : /* boundary check on length */
2751 234 : if (cp.length != ReplicationSlotOnDiskV2Size)
2752 0 : ereport(PANIC,
2753 : (errcode(ERRCODE_DATA_CORRUPTED),
2754 : errmsg("replication slot file \"%s\" has corrupted length %u",
2755 : path, cp.length)));
2756 :
2757 : /* Now that we know the size, read the entire file */
2758 234 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2759 468 : readBytes = read(fd,
2760 : (char *) &cp + ReplicationSlotOnDiskConstantSize,
2761 234 : cp.length);
2762 234 : pgstat_report_wait_end();
2763 234 : if (readBytes != cp.length)
2764 : {
2765 0 : if (readBytes < 0)
2766 0 : ereport(PANIC,
2767 : (errcode_for_file_access(),
2768 : errmsg("could not read file \"%s\": %m", path)));
2769 : else
2770 0 : ereport(PANIC,
2771 : (errcode(ERRCODE_DATA_CORRUPTED),
2772 : errmsg("could not read file \"%s\": read %d of %zu",
2773 : path, readBytes, (Size) cp.length)));
2774 : }
2775 :
2776 234 : if (CloseTransientFile(fd) != 0)
2777 0 : ereport(PANIC,
2778 : (errcode_for_file_access(),
2779 : errmsg("could not close file \"%s\": %m", path)));
2780 :
2781 : /* now verify the CRC */
2782 234 : INIT_CRC32C(checksum);
2783 234 : COMP_CRC32C(checksum,
2784 : (char *) &cp + ReplicationSlotOnDiskNotChecksummedSize,
2785 : ReplicationSlotOnDiskChecksummedSize);
2786 234 : FIN_CRC32C(checksum);
2787 :
2788 234 : if (!EQ_CRC32C(checksum, cp.checksum))
2789 0 : ereport(PANIC,
2790 : (errmsg("checksum mismatch for replication slot file \"%s\": is %u, should be %u",
2791 : path, checksum, cp.checksum)));
2792 :
2793 : /*
2794 : * If we crashed with an ephemeral slot active, don't restore but delete
2795 : * it.
2796 : */
2797 234 : if (cp.slotdata.persistency != RS_PERSISTENT)
2798 : {
2799 0 : if (!rmtree(slotdir, true))
2800 : {
2801 0 : ereport(WARNING,
2802 : (errmsg("could not remove directory \"%s\"",
2803 : slotdir)));
2804 : }
2805 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2806 0 : return;
2807 : }
2808 :
2809 : /*
2810 : * Verify that requirements for the specific slot type are met. That's
2811 : * important because if these aren't met we're not guaranteed to retain
2812 : * all the necessary resources for the slot.
2813 : *
2814 : * NB: We have to do so *after* the above checks for ephemeral slots,
2815 : * because otherwise a slot that shouldn't exist anymore could prevent
2816 : * restarts.
2817 : *
2818 : * NB: Changing the requirements here also requires adapting
2819 : * CheckSlotRequirements() and CheckLogicalDecodingRequirements().
2820 : */
2821 234 : if (cp.slotdata.database != InvalidOid)
2822 : {
2823 158 : if (wal_level < WAL_LEVEL_REPLICA)
2824 2 : ereport(FATAL,
2825 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2826 : errmsg("logical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2827 : NameStr(cp.slotdata.name)),
2828 : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2829 :
2830 : /*
2831 : * In standby mode, the hot standby must be enabled. This check is
2832 : * necessary to ensure logical slots are invalidated when they become
2833 : * incompatible due to insufficient wal_level. Otherwise, if the
2834 : * primary reduces effective_wal_level < logical while hot standby is
2835 : * disabled, primary disable logical decoding while hot standby is
2836 : * disabled, logical slots would remain valid even after promotion.
2837 : */
2838 156 : if (StandbyMode && !EnableHotStandby)
2839 2 : ereport(FATAL,
2840 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2841 : errmsg("logical replication slot \"%s\" exists on the standby, but \"hot_standby\" = \"off\"",
2842 : NameStr(cp.slotdata.name)),
2843 : errhint("Change \"hot_standby\" to be \"on\".")));
2844 : }
2845 76 : else if (wal_level < WAL_LEVEL_REPLICA)
2846 0 : ereport(FATAL,
2847 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2848 : errmsg("physical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2849 : NameStr(cp.slotdata.name)),
2850 : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2851 :
2852 : /* nothing can be active yet, don't lock anything */
2853 328 : for (i = 0; i < max_replication_slots; i++)
2854 : {
2855 : ReplicationSlot *slot;
2856 :
2857 328 : slot = &ReplicationSlotCtl->replication_slots[i];
2858 :
2859 328 : if (slot->in_use)
2860 98 : continue;
2861 :
2862 : /* restore the entire set of persistent data */
2863 230 : memcpy(&slot->data, &cp.slotdata,
2864 : sizeof(ReplicationSlotPersistentData));
2865 :
2866 : /* initialize in memory state */
2867 230 : slot->effective_xmin = cp.slotdata.xmin;
2868 230 : slot->effective_catalog_xmin = cp.slotdata.catalog_xmin;
2869 230 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2870 230 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2871 :
2872 230 : slot->candidate_catalog_xmin = InvalidTransactionId;
2873 230 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
2874 230 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
2875 230 : slot->candidate_restart_valid = InvalidXLogRecPtr;
2876 :
2877 230 : slot->in_use = true;
2878 230 : slot->active_pid = 0;
2879 :
2880 : /*
2881 : * Set the time since the slot has become inactive after loading the
2882 : * slot from the disk into memory. Whoever acquires the slot i.e.
2883 : * makes the slot active will reset it. Use the same inactive_since
2884 : * time for all the slots.
2885 : */
2886 230 : if (now == 0)
2887 230 : now = GetCurrentTimestamp();
2888 :
2889 230 : ReplicationSlotSetInactiveSince(slot, now, false);
2890 :
2891 230 : restored = true;
2892 230 : break;
2893 : }
2894 :
2895 230 : if (!restored)
2896 0 : ereport(FATAL,
2897 : (errmsg("too many replication slots active before shutdown"),
2898 : errhint("Increase \"max_replication_slots\" and try again.")));
2899 : }
2900 :
2901 : /*
2902 : * Maps an invalidation reason for a replication slot to
2903 : * ReplicationSlotInvalidationCause.
2904 : */
2905 : ReplicationSlotInvalidationCause
2906 0 : GetSlotInvalidationCause(const char *cause_name)
2907 : {
2908 : Assert(cause_name);
2909 :
2910 : /* Search lookup table for the cause having this name */
2911 0 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2912 : {
2913 0 : if (strcmp(SlotInvalidationCauses[i].cause_name, cause_name) == 0)
2914 0 : return SlotInvalidationCauses[i].cause;
2915 : }
2916 :
2917 : Assert(false);
2918 0 : return RS_INVAL_NONE; /* to keep compiler quiet */
2919 : }
2920 :
2921 : /*
2922 : * Maps a ReplicationSlotInvalidationCause to the invalidation
2923 : * reason for a replication slot.
2924 : */
2925 : const char *
2926 88 : GetSlotInvalidationCauseName(ReplicationSlotInvalidationCause cause)
2927 : {
2928 : /* Search lookup table for the name of this cause */
2929 282 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2930 : {
2931 282 : if (SlotInvalidationCauses[i].cause == cause)
2932 88 : return SlotInvalidationCauses[i].cause_name;
2933 : }
2934 :
2935 : Assert(false);
2936 0 : return "none"; /* to keep compiler quiet */
2937 : }
2938 :
2939 : /*
2940 : * A helper function to validate slots specified in GUC synchronized_standby_slots.
2941 : *
2942 : * The rawname will be parsed, and the result will be saved into *elemlist.
2943 : */
2944 : static bool
2945 38 : validate_sync_standby_slots(char *rawname, List **elemlist)
2946 : {
2947 : /* Verify syntax and parse string into a list of identifiers */
2948 38 : if (!SplitIdentifierString(rawname, ',', elemlist))
2949 : {
2950 0 : GUC_check_errdetail("List syntax is invalid.");
2951 0 : return false;
2952 : }
2953 :
2954 : /* Iterate the list to validate each slot name */
2955 106 : foreach_ptr(char, name, *elemlist)
2956 : {
2957 : int err_code;
2958 38 : char *err_msg = NULL;
2959 38 : char *err_hint = NULL;
2960 :
2961 38 : if (!ReplicationSlotValidateNameInternal(name, false, &err_code,
2962 : &err_msg, &err_hint))
2963 : {
2964 4 : GUC_check_errcode(err_code);
2965 4 : GUC_check_errdetail("%s", err_msg);
2966 4 : if (err_hint != NULL)
2967 4 : GUC_check_errhint("%s", err_hint);
2968 4 : return false;
2969 : }
2970 : }
2971 :
2972 34 : return true;
2973 : }
2974 :
2975 : /*
2976 : * GUC check_hook for synchronized_standby_slots
2977 : */
2978 : bool
2979 2410 : check_synchronized_standby_slots(char **newval, void **extra, GucSource source)
2980 : {
2981 : char *rawname;
2982 : char *ptr;
2983 : List *elemlist;
2984 : int size;
2985 : bool ok;
2986 : SyncStandbySlotsConfigData *config;
2987 :
2988 2410 : if ((*newval)[0] == '\0')
2989 2372 : return true;
2990 :
2991 : /* Need a modifiable copy of the GUC string */
2992 38 : rawname = pstrdup(*newval);
2993 :
2994 : /* Now verify if the specified slots exist and have correct type */
2995 38 : ok = validate_sync_standby_slots(rawname, &elemlist);
2996 :
2997 38 : if (!ok || elemlist == NIL)
2998 : {
2999 4 : pfree(rawname);
3000 4 : list_free(elemlist);
3001 4 : return ok;
3002 : }
3003 :
3004 : /* Compute the size required for the SyncStandbySlotsConfigData struct */
3005 34 : size = offsetof(SyncStandbySlotsConfigData, slot_names);
3006 102 : foreach_ptr(char, slot_name, elemlist)
3007 34 : size += strlen(slot_name) + 1;
3008 :
3009 : /* GUC extra value must be guc_malloc'd, not palloc'd */
3010 34 : config = (SyncStandbySlotsConfigData *) guc_malloc(LOG, size);
3011 34 : if (!config)
3012 0 : return false;
3013 :
3014 : /* Transform the data into SyncStandbySlotsConfigData */
3015 34 : config->nslotnames = list_length(elemlist);
3016 :
3017 34 : ptr = config->slot_names;
3018 102 : foreach_ptr(char, slot_name, elemlist)
3019 : {
3020 34 : strcpy(ptr, slot_name);
3021 34 : ptr += strlen(slot_name) + 1;
3022 : }
3023 :
3024 34 : *extra = config;
3025 :
3026 34 : pfree(rawname);
3027 34 : list_free(elemlist);
3028 34 : return true;
3029 : }
3030 :
3031 : /*
3032 : * GUC assign_hook for synchronized_standby_slots
3033 : */
3034 : void
3035 2404 : assign_synchronized_standby_slots(const char *newval, void *extra)
3036 : {
3037 : /*
3038 : * The standby slots may have changed, so we must recompute the oldest
3039 : * LSN.
3040 : */
3041 2404 : ss_oldest_flush_lsn = InvalidXLogRecPtr;
3042 :
3043 2404 : synchronized_standby_slots_config = (SyncStandbySlotsConfigData *) extra;
3044 2404 : }
3045 :
3046 : /*
3047 : * Check if the passed slot_name is specified in the synchronized_standby_slots GUC.
3048 : */
3049 : bool
3050 54702 : SlotExistsInSyncStandbySlots(const char *slot_name)
3051 : {
3052 : const char *standby_slot_name;
3053 :
3054 : /* Return false if there is no value in synchronized_standby_slots */
3055 54702 : if (synchronized_standby_slots_config == NULL)
3056 54676 : return false;
3057 :
3058 : /*
3059 : * XXX: We are not expecting this list to be long so a linear search
3060 : * shouldn't hurt but if that turns out not to be true then we can cache
3061 : * this information for each WalSender as well.
3062 : */
3063 26 : standby_slot_name = synchronized_standby_slots_config->slot_names;
3064 38 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
3065 : {
3066 26 : if (strcmp(standby_slot_name, slot_name) == 0)
3067 14 : return true;
3068 :
3069 12 : standby_slot_name += strlen(standby_slot_name) + 1;
3070 : }
3071 :
3072 12 : return false;
3073 : }
3074 :
3075 : /*
3076 : * Return true if the slots specified in synchronized_standby_slots have caught up to
3077 : * the given WAL location, false otherwise.
3078 : *
3079 : * The elevel parameter specifies the error level used for logging messages
3080 : * related to slots that do not exist, are invalidated, or are inactive.
3081 : */
3082 : bool
3083 1344 : StandbySlotsHaveCaughtup(XLogRecPtr wait_for_lsn, int elevel)
3084 : {
3085 : const char *name;
3086 1344 : int caught_up_slot_num = 0;
3087 1344 : XLogRecPtr min_restart_lsn = InvalidXLogRecPtr;
3088 :
3089 : /*
3090 : * Don't need to wait for the standbys to catch up if there is no value in
3091 : * synchronized_standby_slots.
3092 : */
3093 1344 : if (synchronized_standby_slots_config == NULL)
3094 1304 : return true;
3095 :
3096 : /*
3097 : * Don't need to wait for the standbys to catch up if we are on a standby
3098 : * server, since we do not support syncing slots to cascading standbys.
3099 : */
3100 40 : if (RecoveryInProgress())
3101 0 : return true;
3102 :
3103 : /*
3104 : * Don't need to wait for the standbys to catch up if they are already
3105 : * beyond the specified WAL location.
3106 : */
3107 40 : if (XLogRecPtrIsValid(ss_oldest_flush_lsn) &&
3108 22 : ss_oldest_flush_lsn >= wait_for_lsn)
3109 12 : return true;
3110 :
3111 : /*
3112 : * To prevent concurrent slot dropping and creation while filtering the
3113 : * slots, take the ReplicationSlotControlLock outside of the loop.
3114 : */
3115 28 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
3116 :
3117 28 : name = synchronized_standby_slots_config->slot_names;
3118 40 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
3119 : {
3120 : XLogRecPtr restart_lsn;
3121 : bool invalidated;
3122 : bool inactive;
3123 : ReplicationSlot *slot;
3124 :
3125 28 : slot = SearchNamedReplicationSlot(name, false);
3126 :
3127 : /*
3128 : * If a slot name provided in synchronized_standby_slots does not
3129 : * exist, report a message and exit the loop.
3130 : */
3131 28 : if (!slot)
3132 : {
3133 0 : ereport(elevel,
3134 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3135 : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not exist",
3136 : name, "synchronized_standby_slots"),
3137 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3138 : name),
3139 : errhint("Create the replication slot \"%s\" or amend parameter \"%s\".",
3140 : name, "synchronized_standby_slots"));
3141 0 : break;
3142 : }
3143 :
3144 : /* Same as above: if a slot is not physical, exit the loop. */
3145 28 : if (SlotIsLogical(slot))
3146 : {
3147 0 : ereport(elevel,
3148 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3149 : errmsg("cannot specify logical replication slot \"%s\" in parameter \"%s\"",
3150 : name, "synchronized_standby_slots"),
3151 : errdetail("Logical replication is waiting for correction on replication slot \"%s\".",
3152 : name),
3153 : errhint("Remove the logical replication slot \"%s\" from parameter \"%s\".",
3154 : name, "synchronized_standby_slots"));
3155 0 : break;
3156 : }
3157 :
3158 28 : SpinLockAcquire(&slot->mutex);
3159 28 : restart_lsn = slot->data.restart_lsn;
3160 28 : invalidated = slot->data.invalidated != RS_INVAL_NONE;
3161 28 : inactive = slot->active_pid == 0;
3162 28 : SpinLockRelease(&slot->mutex);
3163 :
3164 28 : if (invalidated)
3165 : {
3166 : /* Specified physical slot has been invalidated */
3167 0 : ereport(elevel,
3168 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3169 : errmsg("physical replication slot \"%s\" specified in parameter \"%s\" has been invalidated",
3170 : name, "synchronized_standby_slots"),
3171 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3172 : name),
3173 : errhint("Drop and recreate the replication slot \"%s\", or amend parameter \"%s\".",
3174 : name, "synchronized_standby_slots"));
3175 0 : break;
3176 : }
3177 :
3178 28 : if (!XLogRecPtrIsValid(restart_lsn) || restart_lsn < wait_for_lsn)
3179 : {
3180 : /* Log a message if no active_pid for this physical slot */
3181 16 : if (inactive)
3182 12 : ereport(elevel,
3183 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3184 : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not have active_pid",
3185 : name, "synchronized_standby_slots"),
3186 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3187 : name),
3188 : errhint("Start the standby associated with the replication slot \"%s\", or amend parameter \"%s\".",
3189 : name, "synchronized_standby_slots"));
3190 :
3191 : /* Continue if the current slot hasn't caught up. */
3192 16 : break;
3193 : }
3194 :
3195 : Assert(restart_lsn >= wait_for_lsn);
3196 :
3197 12 : if (!XLogRecPtrIsValid(min_restart_lsn) ||
3198 : min_restart_lsn > restart_lsn)
3199 12 : min_restart_lsn = restart_lsn;
3200 :
3201 12 : caught_up_slot_num++;
3202 :
3203 12 : name += strlen(name) + 1;
3204 : }
3205 :
3206 28 : LWLockRelease(ReplicationSlotControlLock);
3207 :
3208 : /*
3209 : * Return false if not all the standbys have caught up to the specified
3210 : * WAL location.
3211 : */
3212 28 : if (caught_up_slot_num != synchronized_standby_slots_config->nslotnames)
3213 16 : return false;
3214 :
3215 : /* The ss_oldest_flush_lsn must not retreat. */
3216 : Assert(!XLogRecPtrIsValid(ss_oldest_flush_lsn) ||
3217 : min_restart_lsn >= ss_oldest_flush_lsn);
3218 :
3219 12 : ss_oldest_flush_lsn = min_restart_lsn;
3220 :
3221 12 : return true;
3222 : }
3223 :
3224 : /*
3225 : * Wait for physical standbys to confirm receiving the given lsn.
3226 : *
3227 : * Used by logical decoding SQL functions. It waits for physical standbys
3228 : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
3229 : */
3230 : void
3231 448 : WaitForStandbyConfirmation(XLogRecPtr wait_for_lsn)
3232 : {
3233 : /*
3234 : * Don't need to wait for the standby to catch up if the current acquired
3235 : * slot is not a logical failover slot, or there is no value in
3236 : * synchronized_standby_slots.
3237 : */
3238 448 : if (!MyReplicationSlot->data.failover || !synchronized_standby_slots_config)
3239 446 : return;
3240 :
3241 2 : ConditionVariablePrepareToSleep(&WalSndCtl->wal_confirm_rcv_cv);
3242 :
3243 : for (;;)
3244 : {
3245 4 : CHECK_FOR_INTERRUPTS();
3246 :
3247 4 : if (ConfigReloadPending)
3248 : {
3249 2 : ConfigReloadPending = false;
3250 2 : ProcessConfigFile(PGC_SIGHUP);
3251 : }
3252 :
3253 : /* Exit if done waiting for every slot. */
3254 4 : if (StandbySlotsHaveCaughtup(wait_for_lsn, WARNING))
3255 2 : break;
3256 :
3257 : /*
3258 : * Wait for the slots in the synchronized_standby_slots to catch up,
3259 : * but use a timeout (1s) so we can also check if the
3260 : * synchronized_standby_slots has been changed.
3261 : */
3262 2 : ConditionVariableTimedSleep(&WalSndCtl->wal_confirm_rcv_cv, 1000,
3263 : WAIT_EVENT_WAIT_FOR_STANDBY_CONFIRMATION);
3264 : }
3265 :
3266 2 : ConditionVariableCancelSleep();
3267 : }
|
