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 : #include "utils/wait_event.h"
63 :
64 : /*
65 : * Replication slot on-disk data structure.
66 : */
67 : typedef struct ReplicationSlotOnDisk
68 : {
69 : /* first part of this struct needs to be version independent */
70 :
71 : /* data not covered by checksum */
72 : uint32 magic;
73 : pg_crc32c checksum;
74 :
75 : /* data covered by checksum */
76 : uint32 version;
77 : uint32 length;
78 :
79 : /*
80 : * The actual data in the slot that follows can differ based on the above
81 : * 'version'.
82 : */
83 :
84 : ReplicationSlotPersistentData slotdata;
85 : } ReplicationSlotOnDisk;
86 :
87 : /*
88 : * Struct for the configuration of synchronized_standby_slots.
89 : *
90 : * Note: this must be a flat representation that can be held in a single chunk
91 : * of guc_malloc'd memory, so that it can be stored as the "extra" data for the
92 : * synchronized_standby_slots GUC.
93 : */
94 : typedef struct
95 : {
96 : /* Number of slot names in the slot_names[] */
97 : int nslotnames;
98 :
99 : /*
100 : * slot_names contains 'nslotnames' consecutive null-terminated C strings.
101 : */
102 : char slot_names[FLEXIBLE_ARRAY_MEMBER];
103 : } SyncStandbySlotsConfigData;
104 :
105 : /*
106 : * Lookup table for slot invalidation causes.
107 : */
108 : typedef struct SlotInvalidationCauseMap
109 : {
110 : ReplicationSlotInvalidationCause cause;
111 : const char *cause_name;
112 : } SlotInvalidationCauseMap;
113 :
114 : static const SlotInvalidationCauseMap SlotInvalidationCauses[] = {
115 : {RS_INVAL_NONE, "none"},
116 : {RS_INVAL_WAL_REMOVED, "wal_removed"},
117 : {RS_INVAL_HORIZON, "rows_removed"},
118 : {RS_INVAL_WAL_LEVEL, "wal_level_insufficient"},
119 : {RS_INVAL_IDLE_TIMEOUT, "idle_timeout"},
120 : };
121 :
122 : /*
123 : * Ensure that the lookup table is up-to-date with the enums defined in
124 : * ReplicationSlotInvalidationCause.
125 : */
126 : StaticAssertDecl(lengthof(SlotInvalidationCauses) == (RS_INVAL_MAX_CAUSES + 1),
127 : "array length mismatch");
128 :
129 : /* size of version independent data */
130 : #define ReplicationSlotOnDiskConstantSize \
131 : offsetof(ReplicationSlotOnDisk, slotdata)
132 : /* size of the part of the slot not covered by the checksum */
133 : #define ReplicationSlotOnDiskNotChecksummedSize \
134 : offsetof(ReplicationSlotOnDisk, version)
135 : /* size of the part covered by the checksum */
136 : #define ReplicationSlotOnDiskChecksummedSize \
137 : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskNotChecksummedSize
138 : /* size of the slot data that is version dependent */
139 : #define ReplicationSlotOnDiskV2Size \
140 : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskConstantSize
141 :
142 : #define SLOT_MAGIC 0x1051CA1 /* format identifier */
143 : #define SLOT_VERSION 5 /* version for new files */
144 :
145 : /* Control array for replication slot management */
146 : ReplicationSlotCtlData *ReplicationSlotCtl = NULL;
147 :
148 : /* My backend's replication slot in the shared memory array */
149 : ReplicationSlot *MyReplicationSlot = NULL;
150 :
151 : /* GUC variables */
152 : int max_replication_slots = 10; /* the maximum number of replication
153 : * slots */
154 :
155 : /*
156 : * Invalidate replication slots that have remained idle longer than this
157 : * duration; '0' disables it.
158 : */
159 : int idle_replication_slot_timeout_secs = 0;
160 :
161 : /*
162 : * This GUC lists streaming replication standby server slot names that
163 : * logical WAL sender processes will wait for.
164 : */
165 : char *synchronized_standby_slots;
166 :
167 : /* This is the parsed and cached configuration for synchronized_standby_slots */
168 : static SyncStandbySlotsConfigData *synchronized_standby_slots_config;
169 :
170 : /*
171 : * Oldest LSN that has been confirmed to be flushed to the standbys
172 : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
173 : */
174 : static XLogRecPtr ss_oldest_flush_lsn = InvalidXLogRecPtr;
175 :
176 : static void ReplicationSlotShmemExit(int code, Datum arg);
177 : static bool IsSlotForConflictCheck(const char *name);
178 : static void ReplicationSlotDropPtr(ReplicationSlot *slot);
179 :
180 : /* internal persistency functions */
181 : static void RestoreSlotFromDisk(const char *name);
182 : static void CreateSlotOnDisk(ReplicationSlot *slot);
183 : static void SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel);
184 :
185 : /*
186 : * Report shared-memory space needed by ReplicationSlotsShmemInit.
187 : */
188 : Size
189 4565 : ReplicationSlotsShmemSize(void)
190 : {
191 4565 : Size size = 0;
192 :
193 4565 : if (max_replication_slots == 0)
194 2 : return size;
195 :
196 4563 : size = offsetof(ReplicationSlotCtlData, replication_slots);
197 4563 : size = add_size(size,
198 : mul_size(max_replication_slots, sizeof(ReplicationSlot)));
199 :
200 4563 : return size;
201 : }
202 :
203 : /*
204 : * Allocate and initialize shared memory for replication slots.
205 : */
206 : void
207 1180 : ReplicationSlotsShmemInit(void)
208 : {
209 : bool found;
210 :
211 1180 : if (max_replication_slots == 0)
212 1 : return;
213 :
214 1179 : ReplicationSlotCtl = (ReplicationSlotCtlData *)
215 1179 : ShmemInitStruct("ReplicationSlot Ctl", ReplicationSlotsShmemSize(),
216 : &found);
217 :
218 1179 : if (!found)
219 : {
220 : int i;
221 :
222 : /* First time through, so initialize */
223 2289 : MemSet(ReplicationSlotCtl, 0, ReplicationSlotsShmemSize());
224 :
225 12777 : for (i = 0; i < max_replication_slots; i++)
226 : {
227 11598 : ReplicationSlot *slot = &ReplicationSlotCtl->replication_slots[i];
228 :
229 : /* everything else is zeroed by the memset above */
230 11598 : slot->active_proc = INVALID_PROC_NUMBER;
231 11598 : SpinLockInit(&slot->mutex);
232 11598 : LWLockInitialize(&slot->io_in_progress_lock,
233 : LWTRANCHE_REPLICATION_SLOT_IO);
234 11598 : ConditionVariableInit(&slot->active_cv);
235 : }
236 : }
237 : }
238 :
239 : /*
240 : * Register the callback for replication slot cleanup and releasing.
241 : */
242 : void
243 24243 : ReplicationSlotInitialize(void)
244 : {
245 24243 : before_shmem_exit(ReplicationSlotShmemExit, 0);
246 24243 : }
247 :
248 : /*
249 : * Release and cleanup replication slots.
250 : */
251 : static void
252 24243 : ReplicationSlotShmemExit(int code, Datum arg)
253 : {
254 : /* Make sure active replication slots are released */
255 24243 : if (MyReplicationSlot != NULL)
256 298 : ReplicationSlotRelease();
257 :
258 : /* Also cleanup all the temporary slots. */
259 24243 : ReplicationSlotCleanup(false);
260 24243 : }
261 :
262 : /*
263 : * Check whether the passed slot name is valid and report errors at elevel.
264 : *
265 : * See comments for ReplicationSlotValidateNameInternal().
266 : */
267 : bool
268 850 : ReplicationSlotValidateName(const char *name, bool allow_reserved_name,
269 : int elevel)
270 : {
271 : int err_code;
272 850 : char *err_msg = NULL;
273 850 : char *err_hint = NULL;
274 :
275 850 : if (!ReplicationSlotValidateNameInternal(name, allow_reserved_name,
276 : &err_code, &err_msg, &err_hint))
277 : {
278 : /*
279 : * Use errmsg_internal() and errhint_internal() instead of errmsg()
280 : * and errhint(), since the messages from
281 : * ReplicationSlotValidateNameInternal() are already translated. This
282 : * avoids double translation.
283 : */
284 5 : ereport(elevel,
285 : errcode(err_code),
286 : errmsg_internal("%s", err_msg),
287 : (err_hint != NULL) ? errhint_internal("%s", err_hint) : 0);
288 :
289 0 : pfree(err_msg);
290 0 : if (err_hint != NULL)
291 0 : pfree(err_hint);
292 0 : return false;
293 : }
294 :
295 845 : return true;
296 : }
297 :
298 : /*
299 : * Check whether the passed slot name is valid.
300 : *
301 : * An error will be reported for a reserved replication slot name if
302 : * allow_reserved_name is set to false.
303 : *
304 : * Slot names may consist out of [a-z0-9_]{1,NAMEDATALEN-1} which should allow
305 : * the name to be used as a directory name on every supported OS.
306 : *
307 : * Returns true if the slot name is valid. Otherwise, returns false and stores
308 : * the error code, error message, and optional hint in err_code, err_msg, and
309 : * err_hint, respectively. The caller is responsible for freeing err_msg and
310 : * err_hint, which are palloc'd.
311 : */
312 : bool
313 1069 : ReplicationSlotValidateNameInternal(const char *name, bool allow_reserved_name,
314 : int *err_code, char **err_msg, char **err_hint)
315 : {
316 : const char *cp;
317 :
318 1069 : if (strlen(name) == 0)
319 : {
320 4 : *err_code = ERRCODE_INVALID_NAME;
321 4 : *err_msg = psprintf(_("replication slot name \"%s\" is too short"), name);
322 4 : *err_hint = NULL;
323 4 : return false;
324 : }
325 :
326 1065 : if (strlen(name) >= NAMEDATALEN)
327 : {
328 0 : *err_code = ERRCODE_NAME_TOO_LONG;
329 0 : *err_msg = psprintf(_("replication slot name \"%s\" is too long"), name);
330 0 : *err_hint = NULL;
331 0 : return false;
332 : }
333 :
334 20675 : for (cp = name; *cp; cp++)
335 : {
336 19612 : if (!((*cp >= 'a' && *cp <= 'z')
337 9382 : || (*cp >= '0' && *cp <= '9')
338 1908 : || (*cp == '_')))
339 : {
340 2 : *err_code = ERRCODE_INVALID_NAME;
341 2 : *err_msg = psprintf(_("replication slot name \"%s\" contains invalid character"), name);
342 2 : *err_hint = psprintf(_("Replication slot names may only contain lower case letters, numbers, and the underscore character."));
343 2 : return false;
344 : }
345 : }
346 :
347 1063 : if (!allow_reserved_name && IsSlotForConflictCheck(name))
348 : {
349 1 : *err_code = ERRCODE_RESERVED_NAME;
350 1 : *err_msg = psprintf(_("replication slot name \"%s\" is reserved"), name);
351 1 : *err_hint = psprintf(_("The name \"%s\" is reserved for the conflict detection slot."),
352 : CONFLICT_DETECTION_SLOT);
353 1 : return false;
354 : }
355 :
356 1062 : return true;
357 : }
358 :
359 : /*
360 : * Return true if the replication slot name is "pg_conflict_detection".
361 : */
362 : static bool
363 2324 : IsSlotForConflictCheck(const char *name)
364 : {
365 2324 : return (strcmp(name, CONFLICT_DETECTION_SLOT) == 0);
366 : }
367 :
368 : /*
369 : * Create a new replication slot and mark it as used by this backend.
370 : *
371 : * name: Name of the slot
372 : * db_specific: logical decoding is db specific; if the slot is going to
373 : * be used for that pass true, otherwise false.
374 : * two_phase: If enabled, allows decoding of prepared transactions.
375 : * failover: If enabled, allows the slot to be synced to standbys so
376 : * that logical replication can be resumed after failover.
377 : * synced: True if the slot is synchronized from the primary server.
378 : */
379 : void
380 707 : ReplicationSlotCreate(const char *name, bool db_specific,
381 : ReplicationSlotPersistency persistency,
382 : bool two_phase, bool failover, bool synced)
383 : {
384 707 : ReplicationSlot *slot = NULL;
385 : int i;
386 :
387 : Assert(MyReplicationSlot == NULL);
388 :
389 : /*
390 : * The logical launcher or pg_upgrade may create or migrate an internal
391 : * slot, so using a reserved name is allowed in these cases.
392 : */
393 707 : ReplicationSlotValidateName(name, IsBinaryUpgrade || IsLogicalLauncher(),
394 707 : ERROR);
395 :
396 706 : if (failover)
397 : {
398 : /*
399 : * Do not allow users to create the failover enabled slots on the
400 : * standby as we do not support sync to the cascading standby.
401 : *
402 : * However, failover enabled slots can be created during slot
403 : * synchronization because we need to retain the same values as the
404 : * remote slot.
405 : */
406 28 : if (RecoveryInProgress() && !IsSyncingReplicationSlots())
407 0 : ereport(ERROR,
408 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
409 : errmsg("cannot enable failover for a replication slot created on the standby"));
410 :
411 : /*
412 : * Do not allow users to create failover enabled temporary slots,
413 : * because temporary slots will not be synced to the standby.
414 : *
415 : * However, failover enabled temporary slots can be created during
416 : * slot synchronization. See the comments atop slotsync.c for details.
417 : */
418 28 : if (persistency == RS_TEMPORARY && !IsSyncingReplicationSlots())
419 1 : ereport(ERROR,
420 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
421 : errmsg("cannot enable failover for a temporary replication slot"));
422 : }
423 :
424 : /*
425 : * If some other backend ran this code concurrently with us, we'd likely
426 : * both allocate the same slot, and that would be bad. We'd also be at
427 : * risk of missing a name collision. Also, we don't want to try to create
428 : * a new slot while somebody's busy cleaning up an old one, because we
429 : * might both be monkeying with the same directory.
430 : */
431 705 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
432 :
433 : /*
434 : * Check for name collision, and identify an allocatable slot. We need to
435 : * hold ReplicationSlotControlLock in shared mode for this, so that nobody
436 : * else can change the in_use flags while we're looking at them.
437 : */
438 705 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
439 6847 : for (i = 0; i < max_replication_slots; i++)
440 : {
441 6145 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
442 :
443 6145 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
444 3 : ereport(ERROR,
445 : (errcode(ERRCODE_DUPLICATE_OBJECT),
446 : errmsg("replication slot \"%s\" already exists", name)));
447 6142 : if (!s->in_use && slot == NULL)
448 701 : slot = s;
449 : }
450 702 : LWLockRelease(ReplicationSlotControlLock);
451 :
452 : /* If all slots are in use, we're out of luck. */
453 702 : if (slot == NULL)
454 1 : ereport(ERROR,
455 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
456 : errmsg("all replication slots are in use"),
457 : errhint("Free one or increase \"max_replication_slots\".")));
458 :
459 : /*
460 : * Since this slot is not in use, nobody should be looking at any part of
461 : * it other than the in_use field unless they're trying to allocate it.
462 : * And since we hold ReplicationSlotAllocationLock, nobody except us can
463 : * be doing that. So it's safe to initialize the slot.
464 : */
465 : Assert(!slot->in_use);
466 : Assert(slot->active_proc == INVALID_PROC_NUMBER);
467 :
468 : /* first initialize persistent data */
469 701 : memset(&slot->data, 0, sizeof(ReplicationSlotPersistentData));
470 701 : namestrcpy(&slot->data.name, name);
471 701 : slot->data.database = db_specific ? MyDatabaseId : InvalidOid;
472 701 : slot->data.persistency = persistency;
473 701 : slot->data.two_phase = two_phase;
474 701 : slot->data.two_phase_at = InvalidXLogRecPtr;
475 701 : slot->data.failover = failover;
476 701 : slot->data.synced = synced;
477 :
478 : /* and then data only present in shared memory */
479 701 : slot->just_dirtied = false;
480 701 : slot->dirty = false;
481 701 : slot->effective_xmin = InvalidTransactionId;
482 701 : slot->effective_catalog_xmin = InvalidTransactionId;
483 701 : slot->candidate_catalog_xmin = InvalidTransactionId;
484 701 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
485 701 : slot->candidate_restart_valid = InvalidXLogRecPtr;
486 701 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
487 701 : slot->last_saved_confirmed_flush = InvalidXLogRecPtr;
488 701 : slot->last_saved_restart_lsn = InvalidXLogRecPtr;
489 701 : slot->inactive_since = 0;
490 701 : slot->slotsync_skip_reason = SS_SKIP_NONE;
491 :
492 : /*
493 : * Create the slot on disk. We haven't actually marked the slot allocated
494 : * yet, so no special cleanup is required if this errors out.
495 : */
496 701 : CreateSlotOnDisk(slot);
497 :
498 : /*
499 : * We need to briefly prevent any other backend from iterating over the
500 : * slots while we flip the in_use flag. We also need to set the active
501 : * flag while holding the ControlLock as otherwise a concurrent
502 : * ReplicationSlotAcquire() could acquire the slot as well.
503 : */
504 701 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
505 :
506 701 : slot->in_use = true;
507 :
508 : /* We can now mark the slot active, and that makes it our slot. */
509 701 : SpinLockAcquire(&slot->mutex);
510 : Assert(slot->active_proc == INVALID_PROC_NUMBER);
511 701 : slot->active_proc = MyProcNumber;
512 701 : SpinLockRelease(&slot->mutex);
513 701 : MyReplicationSlot = slot;
514 :
515 701 : LWLockRelease(ReplicationSlotControlLock);
516 :
517 : /*
518 : * Create statistics entry for the new logical slot. We don't collect any
519 : * stats for physical slots, so no need to create an entry for the same.
520 : * See ReplicationSlotDropPtr for why we need to do this before releasing
521 : * ReplicationSlotAllocationLock.
522 : */
523 701 : if (SlotIsLogical(slot))
524 507 : pgstat_create_replslot(slot);
525 :
526 : /*
527 : * Now that the slot has been marked as in_use and active, it's safe to
528 : * let somebody else try to allocate a slot.
529 : */
530 701 : LWLockRelease(ReplicationSlotAllocationLock);
531 :
532 : /* Let everybody know we've modified this slot */
533 701 : ConditionVariableBroadcast(&slot->active_cv);
534 701 : }
535 :
536 : /*
537 : * Search for the named replication slot.
538 : *
539 : * Return the replication slot if found, otherwise NULL.
540 : */
541 : ReplicationSlot *
542 2088 : SearchNamedReplicationSlot(const char *name, bool need_lock)
543 : {
544 : int i;
545 2088 : ReplicationSlot *slot = NULL;
546 :
547 2088 : if (need_lock)
548 615 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
549 :
550 8106 : for (i = 0; i < max_replication_slots; i++)
551 : {
552 7598 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
553 :
554 7598 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
555 : {
556 1580 : slot = s;
557 1580 : break;
558 : }
559 : }
560 :
561 2088 : if (need_lock)
562 615 : LWLockRelease(ReplicationSlotControlLock);
563 :
564 2088 : return slot;
565 : }
566 :
567 : /*
568 : * Return the index of the replication slot in
569 : * ReplicationSlotCtl->replication_slots.
570 : *
571 : * This is mainly useful to have an efficient key for storing replication slot
572 : * stats.
573 : */
574 : int
575 8982 : ReplicationSlotIndex(ReplicationSlot *slot)
576 : {
577 : Assert(slot >= ReplicationSlotCtl->replication_slots &&
578 : slot < ReplicationSlotCtl->replication_slots + max_replication_slots);
579 :
580 8982 : return slot - ReplicationSlotCtl->replication_slots;
581 : }
582 :
583 : /*
584 : * If the slot at 'index' is unused, return false. Otherwise 'name' is set to
585 : * the slot's name and true is returned.
586 : *
587 : * This likely is only useful for pgstat_replslot.c during shutdown, in other
588 : * cases there are obvious TOCTOU issues.
589 : */
590 : bool
591 107 : ReplicationSlotName(int index, Name name)
592 : {
593 : ReplicationSlot *slot;
594 : bool found;
595 :
596 107 : slot = &ReplicationSlotCtl->replication_slots[index];
597 :
598 : /*
599 : * Ensure that the slot cannot be dropped while we copy the name. Don't
600 : * need the spinlock as the name of an existing slot cannot change.
601 : */
602 107 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
603 107 : found = slot->in_use;
604 107 : if (slot->in_use)
605 107 : namestrcpy(name, NameStr(slot->data.name));
606 107 : LWLockRelease(ReplicationSlotControlLock);
607 :
608 107 : return found;
609 : }
610 :
611 : /*
612 : * Find a previously created slot and mark it as used by this process.
613 : *
614 : * An error is raised if nowait is true and the slot is currently in use. If
615 : * nowait is false, we sleep until the slot is released by the owning process.
616 : *
617 : * An error is raised if error_if_invalid is true and the slot is found to
618 : * be invalid. It should always be set to true, except when we are temporarily
619 : * acquiring the slot and don't intend to change it.
620 : */
621 : void
622 1395 : ReplicationSlotAcquire(const char *name, bool nowait, bool error_if_invalid)
623 : {
624 : ReplicationSlot *s;
625 : ProcNumber active_proc;
626 : int active_pid;
627 :
628 : Assert(name != NULL);
629 :
630 1395 : retry:
631 : Assert(MyReplicationSlot == NULL);
632 :
633 1395 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
634 :
635 : /* Check if the slot exists with the given name. */
636 1395 : s = SearchNamedReplicationSlot(name, false);
637 1395 : if (s == NULL || !s->in_use)
638 : {
639 9 : LWLockRelease(ReplicationSlotControlLock);
640 :
641 9 : ereport(ERROR,
642 : (errcode(ERRCODE_UNDEFINED_OBJECT),
643 : errmsg("replication slot \"%s\" does not exist",
644 : name)));
645 : }
646 :
647 : /*
648 : * Do not allow users to acquire the reserved slot. This scenario may
649 : * occur if the launcher that owns the slot has terminated unexpectedly
650 : * due to an error, and a backend process attempts to reuse the slot.
651 : */
652 1386 : if (!IsLogicalLauncher() && IsSlotForConflictCheck(name))
653 0 : ereport(ERROR,
654 : errcode(ERRCODE_UNDEFINED_OBJECT),
655 : errmsg("cannot acquire replication slot \"%s\"", name),
656 : errdetail("The slot is reserved for conflict detection and can only be acquired by logical replication launcher."));
657 :
658 : /*
659 : * This is the slot we want; check if it's active under some other
660 : * process. In single user mode, we don't need this check.
661 : */
662 1386 : if (IsUnderPostmaster)
663 : {
664 : /*
665 : * Get ready to sleep on the slot in case it is active. (We may end
666 : * up not sleeping, but we don't want to do this while holding the
667 : * spinlock.)
668 : */
669 1381 : if (!nowait)
670 296 : ConditionVariablePrepareToSleep(&s->active_cv);
671 :
672 : /*
673 : * It is important to reset the inactive_since under spinlock here to
674 : * avoid race conditions with slot invalidation. See comments related
675 : * to inactive_since in InvalidatePossiblyObsoleteSlot.
676 : */
677 1381 : SpinLockAcquire(&s->mutex);
678 1381 : if (s->active_proc == INVALID_PROC_NUMBER)
679 1229 : s->active_proc = MyProcNumber;
680 1381 : active_proc = s->active_proc;
681 1381 : ReplicationSlotSetInactiveSince(s, 0, false);
682 1381 : SpinLockRelease(&s->mutex);
683 : }
684 : else
685 : {
686 5 : s->active_proc = active_proc = MyProcNumber;
687 5 : ReplicationSlotSetInactiveSince(s, 0, true);
688 : }
689 1386 : active_pid = GetPGProcByNumber(active_proc)->pid;
690 1386 : LWLockRelease(ReplicationSlotControlLock);
691 :
692 : /*
693 : * If we found the slot but it's already active in another process, we
694 : * wait until the owning process signals us that it's been released, or
695 : * error out.
696 : */
697 1386 : if (active_proc != MyProcNumber)
698 : {
699 0 : if (!nowait)
700 : {
701 : /* Wait here until we get signaled, and then restart */
702 0 : ConditionVariableSleep(&s->active_cv,
703 : WAIT_EVENT_REPLICATION_SLOT_DROP);
704 0 : ConditionVariableCancelSleep();
705 0 : goto retry;
706 : }
707 :
708 0 : ereport(ERROR,
709 : (errcode(ERRCODE_OBJECT_IN_USE),
710 : errmsg("replication slot \"%s\" is active for PID %d",
711 : NameStr(s->data.name), active_pid)));
712 : }
713 1386 : else if (!nowait)
714 296 : ConditionVariableCancelSleep(); /* no sleep needed after all */
715 :
716 : /* We made this slot active, so it's ours now. */
717 1386 : MyReplicationSlot = s;
718 :
719 : /*
720 : * We need to check for invalidation after making the slot ours to avoid
721 : * the possible race condition with the checkpointer that can otherwise
722 : * invalidate the slot immediately after the check.
723 : */
724 1386 : if (error_if_invalid && s->data.invalidated != RS_INVAL_NONE)
725 8 : ereport(ERROR,
726 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
727 : errmsg("can no longer access replication slot \"%s\"",
728 : NameStr(s->data.name)),
729 : errdetail("This replication slot has been invalidated due to \"%s\".",
730 : GetSlotInvalidationCauseName(s->data.invalidated)));
731 :
732 : /* Let everybody know we've modified this slot */
733 1378 : ConditionVariableBroadcast(&s->active_cv);
734 :
735 : /*
736 : * The call to pgstat_acquire_replslot() protects against stats for a
737 : * different slot, from before a restart or such, being present during
738 : * pgstat_report_replslot().
739 : */
740 1378 : if (SlotIsLogical(s))
741 1158 : pgstat_acquire_replslot(s);
742 :
743 :
744 1378 : if (am_walsender)
745 : {
746 949 : ereport(log_replication_commands ? LOG : DEBUG1,
747 : SlotIsLogical(s)
748 : ? errmsg("acquired logical replication slot \"%s\"",
749 : NameStr(s->data.name))
750 : : errmsg("acquired physical replication slot \"%s\"",
751 : NameStr(s->data.name)));
752 : }
753 1378 : }
754 :
755 : /*
756 : * Release the replication slot that this backend considers to own.
757 : *
758 : * This or another backend can re-acquire the slot later.
759 : * Resources this slot requires will be preserved.
760 : */
761 : void
762 1664 : ReplicationSlotRelease(void)
763 : {
764 1664 : ReplicationSlot *slot = MyReplicationSlot;
765 1664 : char *slotname = NULL; /* keep compiler quiet */
766 : bool is_logical;
767 1664 : TimestampTz now = 0;
768 :
769 : Assert(slot != NULL && slot->active_proc != INVALID_PROC_NUMBER);
770 :
771 1664 : is_logical = SlotIsLogical(slot);
772 :
773 1664 : if (am_walsender)
774 1168 : slotname = pstrdup(NameStr(slot->data.name));
775 :
776 1664 : if (slot->data.persistency == RS_EPHEMERAL)
777 : {
778 : /*
779 : * Delete the slot. There is no !PANIC case where this is allowed to
780 : * fail, all that may happen is an incomplete cleanup of the on-disk
781 : * data.
782 : */
783 6 : ReplicationSlotDropAcquired();
784 :
785 : /*
786 : * Request to disable logical decoding, even though this slot may not
787 : * have been the last logical slot. The checkpointer will verify if
788 : * logical decoding should actually be disabled.
789 : */
790 6 : if (is_logical)
791 6 : RequestDisableLogicalDecoding();
792 : }
793 :
794 : /*
795 : * If slot needed to temporarily restrain both data and catalog xmin to
796 : * create the catalog snapshot, remove that temporary constraint.
797 : * Snapshots can only be exported while the initial snapshot is still
798 : * acquired.
799 : */
800 1664 : if (!TransactionIdIsValid(slot->data.xmin) &&
801 1634 : TransactionIdIsValid(slot->effective_xmin))
802 : {
803 215 : SpinLockAcquire(&slot->mutex);
804 215 : slot->effective_xmin = InvalidTransactionId;
805 215 : SpinLockRelease(&slot->mutex);
806 215 : ReplicationSlotsComputeRequiredXmin(false);
807 : }
808 :
809 : /*
810 : * Set the time since the slot has become inactive. We get the current
811 : * time beforehand to avoid system call while holding the spinlock.
812 : */
813 1664 : now = GetCurrentTimestamp();
814 :
815 1664 : if (slot->data.persistency == RS_PERSISTENT)
816 : {
817 : /*
818 : * Mark persistent slot inactive. We're not freeing it, just
819 : * disconnecting, but wake up others that may be waiting for it.
820 : */
821 1354 : SpinLockAcquire(&slot->mutex);
822 1354 : slot->active_proc = INVALID_PROC_NUMBER;
823 1354 : ReplicationSlotSetInactiveSince(slot, now, false);
824 1354 : SpinLockRelease(&slot->mutex);
825 1354 : ConditionVariableBroadcast(&slot->active_cv);
826 : }
827 : else
828 310 : ReplicationSlotSetInactiveSince(slot, now, true);
829 :
830 1664 : MyReplicationSlot = NULL;
831 :
832 : /* might not have been set when we've been a plain slot */
833 1664 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
834 1664 : MyProc->statusFlags &= ~PROC_IN_LOGICAL_DECODING;
835 1664 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
836 1664 : LWLockRelease(ProcArrayLock);
837 :
838 1664 : if (am_walsender)
839 : {
840 1168 : ereport(log_replication_commands ? LOG : DEBUG1,
841 : is_logical
842 : ? errmsg("released logical replication slot \"%s\"",
843 : slotname)
844 : : errmsg("released physical replication slot \"%s\"",
845 : slotname));
846 :
847 1168 : pfree(slotname);
848 : }
849 1664 : }
850 :
851 : /*
852 : * Cleanup temporary slots created in current session.
853 : *
854 : * Cleanup only synced temporary slots if 'synced_only' is true, else
855 : * cleanup all temporary slots.
856 : *
857 : * If it drops the last logical slot in the cluster, requests to disable
858 : * logical decoding.
859 : */
860 : void
861 54632 : ReplicationSlotCleanup(bool synced_only)
862 : {
863 : int i;
864 : bool found_valid_logicalslot;
865 54632 : bool dropped_logical = false;
866 :
867 : Assert(MyReplicationSlot == NULL);
868 :
869 54784 : restart:
870 54784 : found_valid_logicalslot = false;
871 54784 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
872 595158 : for (i = 0; i < max_replication_slots; i++)
873 : {
874 540526 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
875 :
876 540526 : if (!s->in_use)
877 525370 : continue;
878 :
879 15156 : SpinLockAcquire(&s->mutex);
880 :
881 30312 : found_valid_logicalslot |=
882 15156 : (SlotIsLogical(s) && s->data.invalidated == RS_INVAL_NONE);
883 :
884 15156 : if ((s->active_proc == MyProcNumber &&
885 152 : (!synced_only || s->data.synced)))
886 : {
887 : Assert(s->data.persistency == RS_TEMPORARY);
888 152 : SpinLockRelease(&s->mutex);
889 152 : LWLockRelease(ReplicationSlotControlLock); /* avoid deadlock */
890 :
891 152 : if (SlotIsLogical(s))
892 10 : dropped_logical = true;
893 :
894 152 : ReplicationSlotDropPtr(s);
895 :
896 152 : ConditionVariableBroadcast(&s->active_cv);
897 152 : goto restart;
898 : }
899 : else
900 15004 : SpinLockRelease(&s->mutex);
901 : }
902 :
903 54632 : LWLockRelease(ReplicationSlotControlLock);
904 :
905 54632 : if (dropped_logical && !found_valid_logicalslot)
906 3 : RequestDisableLogicalDecoding();
907 54632 : }
908 :
909 : /*
910 : * Permanently drop replication slot identified by the passed in name.
911 : */
912 : void
913 444 : ReplicationSlotDrop(const char *name, bool nowait)
914 : {
915 : bool is_logical;
916 :
917 : Assert(MyReplicationSlot == NULL);
918 :
919 444 : ReplicationSlotAcquire(name, nowait, false);
920 :
921 : /*
922 : * Do not allow users to drop the slots which are currently being synced
923 : * from the primary to the standby.
924 : */
925 437 : if (RecoveryInProgress() && MyReplicationSlot->data.synced)
926 1 : ereport(ERROR,
927 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
928 : errmsg("cannot drop replication slot \"%s\"", name),
929 : errdetail("This replication slot is being synchronized from the primary server."));
930 :
931 436 : is_logical = SlotIsLogical(MyReplicationSlot);
932 :
933 436 : ReplicationSlotDropAcquired();
934 :
935 436 : if (is_logical)
936 416 : RequestDisableLogicalDecoding();
937 436 : }
938 :
939 : /*
940 : * Change the definition of the slot identified by the specified name.
941 : *
942 : * Altering the two_phase property of a slot requires caution on the
943 : * client-side. Enabling it at any random point during decoding has the
944 : * risk that transactions prepared before this change may be skipped by
945 : * the decoder, leading to missing prepare records on the client. So, we
946 : * enable it for subscription related slots only once the initial tablesync
947 : * is finished. See comments atop worker.c. Disabling it is safe only when
948 : * there are no pending prepared transaction, otherwise, the changes of
949 : * already prepared transactions can be replicated again along with their
950 : * corresponding commit leading to duplicate data or errors.
951 : */
952 : void
953 7 : ReplicationSlotAlter(const char *name, const bool *failover,
954 : const bool *two_phase)
955 : {
956 7 : bool update_slot = false;
957 :
958 : Assert(MyReplicationSlot == NULL);
959 : Assert(failover || two_phase);
960 :
961 7 : ReplicationSlotAcquire(name, false, true);
962 :
963 6 : if (SlotIsPhysical(MyReplicationSlot))
964 0 : ereport(ERROR,
965 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
966 : errmsg("cannot use %s with a physical replication slot",
967 : "ALTER_REPLICATION_SLOT"));
968 :
969 6 : if (RecoveryInProgress())
970 : {
971 : /*
972 : * Do not allow users to alter the slots which are currently being
973 : * synced from the primary to the standby.
974 : */
975 1 : if (MyReplicationSlot->data.synced)
976 1 : ereport(ERROR,
977 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
978 : errmsg("cannot alter replication slot \"%s\"", name),
979 : errdetail("This replication slot is being synchronized from the primary server."));
980 :
981 : /*
982 : * Do not allow users to enable failover on the standby as we do not
983 : * support sync to the cascading standby.
984 : */
985 0 : if (failover && *failover)
986 0 : ereport(ERROR,
987 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
988 : errmsg("cannot enable failover for a replication slot"
989 : " on the standby"));
990 : }
991 :
992 5 : if (failover)
993 : {
994 : /*
995 : * Do not allow users to enable failover for temporary slots as we do
996 : * not support syncing temporary slots to the standby.
997 : */
998 4 : if (*failover && MyReplicationSlot->data.persistency == RS_TEMPORARY)
999 0 : ereport(ERROR,
1000 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1001 : errmsg("cannot enable failover for a temporary replication slot"));
1002 :
1003 4 : if (MyReplicationSlot->data.failover != *failover)
1004 : {
1005 4 : SpinLockAcquire(&MyReplicationSlot->mutex);
1006 4 : MyReplicationSlot->data.failover = *failover;
1007 4 : SpinLockRelease(&MyReplicationSlot->mutex);
1008 :
1009 4 : update_slot = true;
1010 : }
1011 : }
1012 :
1013 5 : if (two_phase && MyReplicationSlot->data.two_phase != *two_phase)
1014 : {
1015 1 : SpinLockAcquire(&MyReplicationSlot->mutex);
1016 1 : MyReplicationSlot->data.two_phase = *two_phase;
1017 1 : SpinLockRelease(&MyReplicationSlot->mutex);
1018 :
1019 1 : update_slot = true;
1020 : }
1021 :
1022 5 : if (update_slot)
1023 : {
1024 5 : ReplicationSlotMarkDirty();
1025 5 : ReplicationSlotSave();
1026 : }
1027 :
1028 5 : ReplicationSlotRelease();
1029 5 : }
1030 :
1031 : /*
1032 : * Permanently drop the currently acquired replication slot.
1033 : */
1034 : void
1035 450 : ReplicationSlotDropAcquired(void)
1036 : {
1037 450 : ReplicationSlot *slot = MyReplicationSlot;
1038 :
1039 : Assert(MyReplicationSlot != NULL);
1040 :
1041 : /* slot isn't acquired anymore */
1042 450 : MyReplicationSlot = NULL;
1043 :
1044 450 : ReplicationSlotDropPtr(slot);
1045 450 : }
1046 :
1047 : /*
1048 : * Permanently drop the replication slot which will be released by the point
1049 : * this function returns.
1050 : */
1051 : static void
1052 602 : ReplicationSlotDropPtr(ReplicationSlot *slot)
1053 : {
1054 : char path[MAXPGPATH];
1055 : char tmppath[MAXPGPATH];
1056 :
1057 : /*
1058 : * If some other backend ran this code concurrently with us, we might try
1059 : * to delete a slot with a certain name while someone else was trying to
1060 : * create a slot with the same name.
1061 : */
1062 602 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
1063 :
1064 : /* Generate pathnames. */
1065 602 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
1066 602 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
1067 :
1068 : /*
1069 : * Rename the slot directory on disk, so that we'll no longer recognize
1070 : * this as a valid slot. Note that if this fails, we've got to mark the
1071 : * slot inactive before bailing out. If we're dropping an ephemeral or a
1072 : * temporary slot, we better never fail hard as the caller won't expect
1073 : * the slot to survive and this might get called during error handling.
1074 : */
1075 602 : if (rename(path, tmppath) == 0)
1076 : {
1077 : /*
1078 : * We need to fsync() the directory we just renamed and its parent to
1079 : * make sure that our changes are on disk in a crash-safe fashion. If
1080 : * fsync() fails, we can't be sure whether the changes are on disk or
1081 : * not. For now, we handle that by panicking;
1082 : * StartupReplicationSlots() will try to straighten it out after
1083 : * restart.
1084 : */
1085 602 : START_CRIT_SECTION();
1086 602 : fsync_fname(tmppath, true);
1087 602 : fsync_fname(PG_REPLSLOT_DIR, true);
1088 602 : END_CRIT_SECTION();
1089 : }
1090 : else
1091 : {
1092 0 : bool fail_softly = slot->data.persistency != RS_PERSISTENT;
1093 :
1094 0 : SpinLockAcquire(&slot->mutex);
1095 0 : slot->active_proc = INVALID_PROC_NUMBER;
1096 0 : SpinLockRelease(&slot->mutex);
1097 :
1098 : /* wake up anyone waiting on this slot */
1099 0 : ConditionVariableBroadcast(&slot->active_cv);
1100 :
1101 0 : ereport(fail_softly ? WARNING : ERROR,
1102 : (errcode_for_file_access(),
1103 : errmsg("could not rename file \"%s\" to \"%s\": %m",
1104 : path, tmppath)));
1105 : }
1106 :
1107 : /*
1108 : * The slot is definitely gone. Lock out concurrent scans of the array
1109 : * long enough to kill it. It's OK to clear the active PID here without
1110 : * grabbing the mutex because nobody else can be scanning the array here,
1111 : * and nobody can be attached to this slot and thus access it without
1112 : * scanning the array.
1113 : *
1114 : * Also wake up processes waiting for it.
1115 : */
1116 602 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
1117 602 : slot->active_proc = INVALID_PROC_NUMBER;
1118 602 : slot->in_use = false;
1119 602 : LWLockRelease(ReplicationSlotControlLock);
1120 602 : ConditionVariableBroadcast(&slot->active_cv);
1121 :
1122 : /*
1123 : * Slot is dead and doesn't prevent resource removal anymore, recompute
1124 : * limits.
1125 : */
1126 602 : ReplicationSlotsComputeRequiredXmin(false);
1127 602 : ReplicationSlotsComputeRequiredLSN();
1128 :
1129 : /*
1130 : * If removing the directory fails, the worst thing that will happen is
1131 : * that the user won't be able to create a new slot with the same name
1132 : * until the next server restart. We warn about it, but that's all.
1133 : */
1134 602 : if (!rmtree(tmppath, true))
1135 0 : ereport(WARNING,
1136 : (errmsg("could not remove directory \"%s\"", tmppath)));
1137 :
1138 : /*
1139 : * Drop the statistics entry for the replication slot. Do this while
1140 : * holding ReplicationSlotAllocationLock so that we don't drop a
1141 : * statistics entry for another slot with the same name just created in
1142 : * another session.
1143 : */
1144 602 : if (SlotIsLogical(slot))
1145 439 : pgstat_drop_replslot(slot);
1146 :
1147 : /*
1148 : * We release this at the very end, so that nobody starts trying to create
1149 : * a slot while we're still cleaning up the detritus of the old one.
1150 : */
1151 602 : LWLockRelease(ReplicationSlotAllocationLock);
1152 602 : }
1153 :
1154 : /*
1155 : * Serialize the currently acquired slot's state from memory to disk, thereby
1156 : * guaranteeing the current state will survive a crash.
1157 : */
1158 : void
1159 1480 : ReplicationSlotSave(void)
1160 : {
1161 : char path[MAXPGPATH];
1162 :
1163 : Assert(MyReplicationSlot != NULL);
1164 :
1165 1480 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(MyReplicationSlot->data.name));
1166 1480 : SaveSlotToPath(MyReplicationSlot, path, ERROR);
1167 1480 : }
1168 :
1169 : /*
1170 : * Signal that it would be useful if the currently acquired slot would be
1171 : * flushed out to disk.
1172 : *
1173 : * Note that the actual flush to disk can be delayed for a long time, if
1174 : * required for correctness explicitly do a ReplicationSlotSave().
1175 : */
1176 : void
1177 39525 : ReplicationSlotMarkDirty(void)
1178 : {
1179 39525 : ReplicationSlot *slot = MyReplicationSlot;
1180 :
1181 : Assert(MyReplicationSlot != NULL);
1182 :
1183 39525 : SpinLockAcquire(&slot->mutex);
1184 39525 : MyReplicationSlot->just_dirtied = true;
1185 39525 : MyReplicationSlot->dirty = true;
1186 39525 : SpinLockRelease(&slot->mutex);
1187 39525 : }
1188 :
1189 : /*
1190 : * Convert a slot that's marked as RS_EPHEMERAL or RS_TEMPORARY to a
1191 : * RS_PERSISTENT slot, guaranteeing it will be there after an eventual crash.
1192 : */
1193 : void
1194 491 : ReplicationSlotPersist(void)
1195 : {
1196 491 : ReplicationSlot *slot = MyReplicationSlot;
1197 :
1198 : Assert(slot != NULL);
1199 : Assert(slot->data.persistency != RS_PERSISTENT);
1200 :
1201 491 : SpinLockAcquire(&slot->mutex);
1202 491 : slot->data.persistency = RS_PERSISTENT;
1203 491 : SpinLockRelease(&slot->mutex);
1204 :
1205 491 : ReplicationSlotMarkDirty();
1206 491 : ReplicationSlotSave();
1207 491 : }
1208 :
1209 : /*
1210 : * Compute the oldest xmin across all slots and store it in the ProcArray.
1211 : *
1212 : * If already_locked is true, both the ReplicationSlotControlLock and the
1213 : * ProcArrayLock have already been acquired exclusively. It is crucial that the
1214 : * caller first acquires the ReplicationSlotControlLock, followed by the
1215 : * ProcArrayLock, to prevent any undetectable deadlocks since this function
1216 : * acquires them in that order.
1217 : */
1218 : void
1219 2555 : ReplicationSlotsComputeRequiredXmin(bool already_locked)
1220 : {
1221 : int i;
1222 2555 : TransactionId agg_xmin = InvalidTransactionId;
1223 2555 : TransactionId agg_catalog_xmin = InvalidTransactionId;
1224 :
1225 : Assert(ReplicationSlotCtl != NULL);
1226 : Assert(!already_locked ||
1227 : (LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_EXCLUSIVE) &&
1228 : LWLockHeldByMeInMode(ProcArrayLock, LW_EXCLUSIVE)));
1229 :
1230 : /*
1231 : * Hold the ReplicationSlotControlLock until after updating the slot xmin
1232 : * values, so no backend updates the initial xmin for newly created slot
1233 : * concurrently. A shared lock is used here to minimize lock contention,
1234 : * especially when many slots exist and advancements occur frequently.
1235 : * This is safe since an exclusive lock is taken during initial slot xmin
1236 : * update in slot creation.
1237 : *
1238 : * One might think that we can hold the ProcArrayLock exclusively and
1239 : * update the slot xmin values, but it could increase lock contention on
1240 : * the ProcArrayLock, which is not great since this function can be called
1241 : * at non-negligible frequency.
1242 : *
1243 : * Concurrent invocation of this function may cause the computed slot xmin
1244 : * to regress. However, this is harmless because tuples prior to the most
1245 : * recent xmin are no longer useful once advancement occurs (see
1246 : * LogicalConfirmReceivedLocation where the slot's xmin value is flushed
1247 : * before updating the effective_xmin). Thus, such regression merely
1248 : * prevents VACUUM from prematurely removing tuples without causing the
1249 : * early deletion of required data.
1250 : */
1251 2555 : if (!already_locked)
1252 2047 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1253 :
1254 26039 : for (i = 0; i < max_replication_slots; i++)
1255 : {
1256 23484 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1257 : TransactionId effective_xmin;
1258 : TransactionId effective_catalog_xmin;
1259 : bool invalidated;
1260 :
1261 23484 : if (!s->in_use)
1262 21013 : continue;
1263 :
1264 2471 : SpinLockAcquire(&s->mutex);
1265 2471 : effective_xmin = s->effective_xmin;
1266 2471 : effective_catalog_xmin = s->effective_catalog_xmin;
1267 2471 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1268 2471 : SpinLockRelease(&s->mutex);
1269 :
1270 : /* invalidated slots need not apply */
1271 2471 : if (invalidated)
1272 26 : continue;
1273 :
1274 : /* check the data xmin */
1275 2445 : if (TransactionIdIsValid(effective_xmin) &&
1276 5 : (!TransactionIdIsValid(agg_xmin) ||
1277 5 : TransactionIdPrecedes(effective_xmin, agg_xmin)))
1278 344 : agg_xmin = effective_xmin;
1279 :
1280 : /* check the catalog xmin */
1281 2445 : if (TransactionIdIsValid(effective_catalog_xmin) &&
1282 1022 : (!TransactionIdIsValid(agg_catalog_xmin) ||
1283 1022 : TransactionIdPrecedes(effective_catalog_xmin, agg_catalog_xmin)))
1284 1284 : agg_catalog_xmin = effective_catalog_xmin;
1285 : }
1286 :
1287 2555 : ProcArraySetReplicationSlotXmin(agg_xmin, agg_catalog_xmin, already_locked);
1288 :
1289 2555 : if (!already_locked)
1290 2047 : LWLockRelease(ReplicationSlotControlLock);
1291 2555 : }
1292 :
1293 : /*
1294 : * Compute the oldest restart LSN across all slots and inform xlog module.
1295 : *
1296 : * Note: while max_slot_wal_keep_size is theoretically relevant for this
1297 : * purpose, we don't try to account for that, because this module doesn't
1298 : * know what to compare against.
1299 : */
1300 : void
1301 40238 : ReplicationSlotsComputeRequiredLSN(void)
1302 : {
1303 : int i;
1304 40238 : XLogRecPtr min_required = InvalidXLogRecPtr;
1305 :
1306 : Assert(ReplicationSlotCtl != NULL);
1307 :
1308 40238 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1309 438475 : for (i = 0; i < max_replication_slots; i++)
1310 : {
1311 398237 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1312 : XLogRecPtr restart_lsn;
1313 : XLogRecPtr last_saved_restart_lsn;
1314 : bool invalidated;
1315 : ReplicationSlotPersistency persistency;
1316 :
1317 398237 : if (!s->in_use)
1318 357811 : continue;
1319 :
1320 40426 : SpinLockAcquire(&s->mutex);
1321 40426 : persistency = s->data.persistency;
1322 40426 : restart_lsn = s->data.restart_lsn;
1323 40426 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1324 40426 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1325 40426 : SpinLockRelease(&s->mutex);
1326 :
1327 : /* invalidated slots need not apply */
1328 40426 : if (invalidated)
1329 27 : continue;
1330 :
1331 : /*
1332 : * For persistent slot use last_saved_restart_lsn to compute the
1333 : * oldest LSN for removal of WAL segments. The segments between
1334 : * last_saved_restart_lsn and restart_lsn might be needed by a
1335 : * persistent slot in the case of database crash. Non-persistent
1336 : * slots can't survive the database crash, so we don't care about
1337 : * last_saved_restart_lsn for them.
1338 : */
1339 40399 : if (persistency == RS_PERSISTENT)
1340 : {
1341 39463 : if (XLogRecPtrIsValid(last_saved_restart_lsn) &&
1342 : restart_lsn > last_saved_restart_lsn)
1343 : {
1344 37063 : restart_lsn = last_saved_restart_lsn;
1345 : }
1346 : }
1347 :
1348 40399 : if (XLogRecPtrIsValid(restart_lsn) &&
1349 1281 : (!XLogRecPtrIsValid(min_required) ||
1350 : restart_lsn < min_required))
1351 39238 : min_required = restart_lsn;
1352 : }
1353 40238 : LWLockRelease(ReplicationSlotControlLock);
1354 :
1355 40238 : XLogSetReplicationSlotMinimumLSN(min_required);
1356 40238 : }
1357 :
1358 : /*
1359 : * Compute the oldest WAL LSN required by *logical* decoding slots..
1360 : *
1361 : * Returns InvalidXLogRecPtr if logical decoding is disabled or no logical
1362 : * slots exist.
1363 : *
1364 : * NB: this returns a value >= ReplicationSlotsComputeRequiredLSN(), since it
1365 : * ignores physical replication slots.
1366 : *
1367 : * The results aren't required frequently, so we don't maintain a precomputed
1368 : * value like we do for ComputeRequiredLSN() and ComputeRequiredXmin().
1369 : */
1370 : XLogRecPtr
1371 3670 : ReplicationSlotsComputeLogicalRestartLSN(void)
1372 : {
1373 3670 : XLogRecPtr result = InvalidXLogRecPtr;
1374 : int i;
1375 :
1376 3670 : if (max_replication_slots <= 0)
1377 2 : return InvalidXLogRecPtr;
1378 :
1379 3668 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1380 :
1381 39818 : for (i = 0; i < max_replication_slots; i++)
1382 : {
1383 : ReplicationSlot *s;
1384 : XLogRecPtr restart_lsn;
1385 : XLogRecPtr last_saved_restart_lsn;
1386 : bool invalidated;
1387 : ReplicationSlotPersistency persistency;
1388 :
1389 36150 : s = &ReplicationSlotCtl->replication_slots[i];
1390 :
1391 : /* cannot change while ReplicationSlotCtlLock is held */
1392 36150 : if (!s->in_use)
1393 35356 : continue;
1394 :
1395 : /* we're only interested in logical slots */
1396 794 : if (!SlotIsLogical(s))
1397 546 : continue;
1398 :
1399 : /* read once, it's ok if it increases while we're checking */
1400 248 : SpinLockAcquire(&s->mutex);
1401 248 : persistency = s->data.persistency;
1402 248 : restart_lsn = s->data.restart_lsn;
1403 248 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1404 248 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1405 248 : SpinLockRelease(&s->mutex);
1406 :
1407 : /* invalidated slots need not apply */
1408 248 : if (invalidated)
1409 10 : continue;
1410 :
1411 : /*
1412 : * For persistent slot use last_saved_restart_lsn to compute the
1413 : * oldest LSN for removal of WAL segments. The segments between
1414 : * last_saved_restart_lsn and restart_lsn might be needed by a
1415 : * persistent slot in the case of database crash. Non-persistent
1416 : * slots can't survive the database crash, so we don't care about
1417 : * last_saved_restart_lsn for them.
1418 : */
1419 238 : if (persistency == RS_PERSISTENT)
1420 : {
1421 236 : if (XLogRecPtrIsValid(last_saved_restart_lsn) &&
1422 : restart_lsn > last_saved_restart_lsn)
1423 : {
1424 0 : restart_lsn = last_saved_restart_lsn;
1425 : }
1426 : }
1427 :
1428 238 : if (!XLogRecPtrIsValid(restart_lsn))
1429 0 : continue;
1430 :
1431 238 : if (!XLogRecPtrIsValid(result) ||
1432 : restart_lsn < result)
1433 188 : result = restart_lsn;
1434 : }
1435 :
1436 3668 : LWLockRelease(ReplicationSlotControlLock);
1437 :
1438 3668 : return result;
1439 : }
1440 :
1441 : /*
1442 : * ReplicationSlotsCountDBSlots -- count the number of slots that refer to the
1443 : * passed database oid.
1444 : *
1445 : * Returns true if there are any slots referencing the database. *nslots will
1446 : * be set to the absolute number of slots in the database, *nactive to ones
1447 : * currently active.
1448 : */
1449 : bool
1450 53 : ReplicationSlotsCountDBSlots(Oid dboid, int *nslots, int *nactive)
1451 : {
1452 : int i;
1453 :
1454 53 : *nslots = *nactive = 0;
1455 :
1456 53 : if (max_replication_slots <= 0)
1457 0 : return false;
1458 :
1459 53 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1460 556 : for (i = 0; i < max_replication_slots; i++)
1461 : {
1462 : ReplicationSlot *s;
1463 :
1464 503 : s = &ReplicationSlotCtl->replication_slots[i];
1465 :
1466 : /* cannot change while ReplicationSlotCtlLock is held */
1467 503 : if (!s->in_use)
1468 482 : continue;
1469 :
1470 : /* only logical slots are database specific, skip */
1471 21 : if (!SlotIsLogical(s))
1472 10 : continue;
1473 :
1474 : /* not our database, skip */
1475 11 : if (s->data.database != dboid)
1476 8 : continue;
1477 :
1478 : /* NB: intentionally counting invalidated slots */
1479 :
1480 : /* count slots with spinlock held */
1481 3 : SpinLockAcquire(&s->mutex);
1482 3 : (*nslots)++;
1483 3 : if (s->active_proc != INVALID_PROC_NUMBER)
1484 1 : (*nactive)++;
1485 3 : SpinLockRelease(&s->mutex);
1486 : }
1487 53 : LWLockRelease(ReplicationSlotControlLock);
1488 :
1489 53 : if (*nslots > 0)
1490 3 : return true;
1491 50 : return false;
1492 : }
1493 :
1494 : /*
1495 : * ReplicationSlotsDropDBSlots -- Drop all db-specific slots relating to the
1496 : * passed database oid. The caller should hold an exclusive lock on the
1497 : * pg_database oid for the database to prevent creation of new slots on the db
1498 : * or replay from existing slots.
1499 : *
1500 : * Another session that concurrently acquires an existing slot on the target DB
1501 : * (most likely to drop it) may cause this function to ERROR. If that happens
1502 : * it may have dropped some but not all slots.
1503 : *
1504 : * This routine isn't as efficient as it could be - but we don't drop
1505 : * databases often, especially databases with lots of slots.
1506 : *
1507 : * If it drops the last logical slot in the cluster, it requests to disable
1508 : * logical decoding.
1509 : */
1510 : void
1511 66 : ReplicationSlotsDropDBSlots(Oid dboid)
1512 : {
1513 : int i;
1514 : bool found_valid_logicalslot;
1515 66 : bool dropped = false;
1516 :
1517 66 : if (max_replication_slots <= 0)
1518 0 : return;
1519 :
1520 66 : restart:
1521 71 : found_valid_logicalslot = false;
1522 71 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1523 696 : for (i = 0; i < max_replication_slots; i++)
1524 : {
1525 : ReplicationSlot *s;
1526 : char *slotname;
1527 : ProcNumber active_proc;
1528 :
1529 630 : s = &ReplicationSlotCtl->replication_slots[i];
1530 :
1531 : /* cannot change while ReplicationSlotCtlLock is held */
1532 630 : if (!s->in_use)
1533 600 : continue;
1534 :
1535 : /* only logical slots are database specific, skip */
1536 30 : if (!SlotIsLogical(s))
1537 11 : continue;
1538 :
1539 : /*
1540 : * Check logical slots on other databases too so we can disable
1541 : * logical decoding only if no slots in the cluster.
1542 : */
1543 19 : SpinLockAcquire(&s->mutex);
1544 19 : found_valid_logicalslot |= (s->data.invalidated == RS_INVAL_NONE);
1545 19 : SpinLockRelease(&s->mutex);
1546 :
1547 : /* not our database, skip */
1548 19 : if (s->data.database != dboid)
1549 14 : continue;
1550 :
1551 : /* NB: intentionally including invalidated slots to drop */
1552 :
1553 : /* acquire slot, so ReplicationSlotDropAcquired can be reused */
1554 5 : SpinLockAcquire(&s->mutex);
1555 : /* can't change while ReplicationSlotControlLock is held */
1556 5 : slotname = NameStr(s->data.name);
1557 5 : active_proc = s->active_proc;
1558 5 : if (active_proc == INVALID_PROC_NUMBER)
1559 : {
1560 5 : MyReplicationSlot = s;
1561 5 : s->active_proc = MyProcNumber;
1562 : }
1563 5 : SpinLockRelease(&s->mutex);
1564 :
1565 : /*
1566 : * Even though we hold an exclusive lock on the database object a
1567 : * logical slot for that DB can still be active, e.g. if it's
1568 : * concurrently being dropped by a backend connected to another DB.
1569 : *
1570 : * That's fairly unlikely in practice, so we'll just bail out.
1571 : *
1572 : * The slot sync worker holds a shared lock on the database before
1573 : * operating on synced logical slots to avoid conflict with the drop
1574 : * happening here. The persistent synced slots are thus safe but there
1575 : * is a possibility that the slot sync worker has created a temporary
1576 : * slot (which stays active even on release) and we are trying to drop
1577 : * that here. In practice, the chances of hitting this scenario are
1578 : * less as during slot synchronization, the temporary slot is
1579 : * immediately converted to persistent and thus is safe due to the
1580 : * shared lock taken on the database. So, we'll just bail out in such
1581 : * a case.
1582 : *
1583 : * XXX: We can consider shutting down the slot sync worker before
1584 : * trying to drop synced temporary slots here.
1585 : */
1586 5 : if (active_proc != INVALID_PROC_NUMBER)
1587 0 : ereport(ERROR,
1588 : (errcode(ERRCODE_OBJECT_IN_USE),
1589 : errmsg("replication slot \"%s\" is active for PID %d",
1590 : slotname, GetPGProcByNumber(active_proc)->pid)));
1591 :
1592 : /*
1593 : * To avoid duplicating ReplicationSlotDropAcquired() and to avoid
1594 : * holding ReplicationSlotControlLock over filesystem operations,
1595 : * release ReplicationSlotControlLock and use
1596 : * ReplicationSlotDropAcquired.
1597 : *
1598 : * As that means the set of slots could change, restart scan from the
1599 : * beginning each time we release the lock.
1600 : */
1601 5 : LWLockRelease(ReplicationSlotControlLock);
1602 5 : ReplicationSlotDropAcquired();
1603 5 : dropped = true;
1604 5 : goto restart;
1605 : }
1606 66 : LWLockRelease(ReplicationSlotControlLock);
1607 :
1608 66 : if (dropped && !found_valid_logicalslot)
1609 0 : RequestDisableLogicalDecoding();
1610 : }
1611 :
1612 : /*
1613 : * Returns true if there is at least one in-use valid logical replication slot.
1614 : */
1615 : bool
1616 471 : CheckLogicalSlotExists(void)
1617 : {
1618 471 : bool found = false;
1619 :
1620 471 : if (max_replication_slots <= 0)
1621 1 : return false;
1622 :
1623 470 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1624 5086 : for (int i = 0; i < max_replication_slots; i++)
1625 : {
1626 : ReplicationSlot *s;
1627 : bool invalidated;
1628 :
1629 4622 : s = &ReplicationSlotCtl->replication_slots[i];
1630 :
1631 : /* cannot change while ReplicationSlotCtlLock is held */
1632 4622 : if (!s->in_use)
1633 4594 : continue;
1634 :
1635 28 : if (SlotIsPhysical(s))
1636 19 : continue;
1637 :
1638 9 : SpinLockAcquire(&s->mutex);
1639 9 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1640 9 : SpinLockRelease(&s->mutex);
1641 :
1642 9 : if (invalidated)
1643 3 : continue;
1644 :
1645 6 : found = true;
1646 6 : break;
1647 : }
1648 470 : LWLockRelease(ReplicationSlotControlLock);
1649 :
1650 470 : return found;
1651 : }
1652 :
1653 : /*
1654 : * Check whether the server's configuration supports using replication
1655 : * slots.
1656 : */
1657 : void
1658 1886 : CheckSlotRequirements(void)
1659 : {
1660 : /*
1661 : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
1662 : * needs the same check.
1663 : */
1664 :
1665 1886 : if (max_replication_slots == 0)
1666 0 : ereport(ERROR,
1667 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1668 : errmsg("replication slots can only be used if \"max_replication_slots\" > 0")));
1669 :
1670 1886 : if (wal_level < WAL_LEVEL_REPLICA)
1671 0 : ereport(ERROR,
1672 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1673 : errmsg("replication slots can only be used if \"wal_level\" >= \"replica\"")));
1674 1886 : }
1675 :
1676 : /*
1677 : * Check whether the user has privilege to use replication slots.
1678 : */
1679 : void
1680 596 : CheckSlotPermissions(void)
1681 : {
1682 596 : if (!has_rolreplication(GetUserId()))
1683 5 : ereport(ERROR,
1684 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1685 : errmsg("permission denied to use replication slots"),
1686 : errdetail("Only roles with the %s attribute may use replication slots.",
1687 : "REPLICATION")));
1688 591 : }
1689 :
1690 : /*
1691 : * Reserve WAL for the currently active slot.
1692 : *
1693 : * Compute and set restart_lsn in a manner that's appropriate for the type of
1694 : * the slot and concurrency safe.
1695 : */
1696 : void
1697 650 : ReplicationSlotReserveWal(void)
1698 : {
1699 650 : ReplicationSlot *slot = MyReplicationSlot;
1700 : XLogSegNo segno;
1701 : XLogRecPtr restart_lsn;
1702 :
1703 : Assert(slot != NULL);
1704 : Assert(!XLogRecPtrIsValid(slot->data.restart_lsn));
1705 : Assert(!XLogRecPtrIsValid(slot->last_saved_restart_lsn));
1706 :
1707 : /*
1708 : * The replication slot mechanism is used to prevent the removal of
1709 : * required WAL.
1710 : *
1711 : * Acquire an exclusive lock to prevent the checkpoint process from
1712 : * concurrently computing the minimum slot LSN (see
1713 : * CheckPointReplicationSlots). This ensures that the WAL reserved for
1714 : * replication cannot be removed during a checkpoint.
1715 : *
1716 : * The mechanism is reliable because if WAL reservation occurs first, the
1717 : * checkpoint must wait for the restart_lsn update before determining the
1718 : * minimum non-removable LSN. On the other hand, if the checkpoint happens
1719 : * first, subsequent WAL reservations will select positions at or beyond
1720 : * the redo pointer of that checkpoint.
1721 : */
1722 650 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
1723 :
1724 : /*
1725 : * For logical slots log a standby snapshot and start logical decoding at
1726 : * exactly that position. That allows the slot to start up more quickly.
1727 : * But on a standby we cannot do WAL writes, so just use the replay
1728 : * pointer; effectively, an attempt to create a logical slot on standby
1729 : * will cause it to wait for an xl_running_xact record to be logged
1730 : * independently on the primary, so that a snapshot can be built using the
1731 : * record.
1732 : *
1733 : * None of this is needed (or indeed helpful) for physical slots as
1734 : * they'll start replay at the last logged checkpoint anyway. Instead,
1735 : * return the location of the last redo LSN, where a base backup has to
1736 : * start replay at.
1737 : */
1738 650 : if (SlotIsPhysical(slot))
1739 160 : restart_lsn = GetRedoRecPtr();
1740 490 : else if (RecoveryInProgress())
1741 26 : restart_lsn = GetXLogReplayRecPtr(NULL);
1742 : else
1743 464 : restart_lsn = GetXLogInsertRecPtr();
1744 :
1745 650 : SpinLockAcquire(&slot->mutex);
1746 650 : slot->data.restart_lsn = restart_lsn;
1747 650 : SpinLockRelease(&slot->mutex);
1748 :
1749 : /* prevent WAL removal as fast as possible */
1750 650 : ReplicationSlotsComputeRequiredLSN();
1751 :
1752 : /* Checkpoint shouldn't remove the required WAL. */
1753 650 : XLByteToSeg(slot->data.restart_lsn, segno, wal_segment_size);
1754 650 : if (XLogGetLastRemovedSegno() >= segno)
1755 0 : elog(ERROR, "WAL required by replication slot %s has been removed concurrently",
1756 : NameStr(slot->data.name));
1757 :
1758 650 : LWLockRelease(ReplicationSlotAllocationLock);
1759 :
1760 650 : if (!RecoveryInProgress() && SlotIsLogical(slot))
1761 : {
1762 : XLogRecPtr flushptr;
1763 :
1764 : /* make sure we have enough information to start */
1765 464 : flushptr = LogStandbySnapshot();
1766 :
1767 : /* and make sure it's fsynced to disk */
1768 464 : XLogFlush(flushptr);
1769 : }
1770 650 : }
1771 :
1772 : /*
1773 : * Report that replication slot needs to be invalidated
1774 : */
1775 : static void
1776 23 : ReportSlotInvalidation(ReplicationSlotInvalidationCause cause,
1777 : bool terminating,
1778 : int pid,
1779 : NameData slotname,
1780 : XLogRecPtr restart_lsn,
1781 : XLogRecPtr oldestLSN,
1782 : TransactionId snapshotConflictHorizon,
1783 : long slot_idle_seconds)
1784 : {
1785 : StringInfoData err_detail;
1786 : StringInfoData err_hint;
1787 :
1788 23 : initStringInfo(&err_detail);
1789 23 : initStringInfo(&err_hint);
1790 :
1791 23 : switch (cause)
1792 : {
1793 7 : case RS_INVAL_WAL_REMOVED:
1794 : {
1795 7 : uint64 ex = oldestLSN - restart_lsn;
1796 :
1797 7 : appendStringInfo(&err_detail,
1798 7 : ngettext("The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " byte.",
1799 : "The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " bytes.",
1800 : ex),
1801 7 : LSN_FORMAT_ARGS(restart_lsn),
1802 : ex);
1803 : /* translator: %s is a GUC variable name */
1804 7 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1805 : "max_slot_wal_keep_size");
1806 7 : break;
1807 : }
1808 12 : case RS_INVAL_HORIZON:
1809 12 : appendStringInfo(&err_detail, _("The slot conflicted with xid horizon %u."),
1810 : snapshotConflictHorizon);
1811 12 : break;
1812 :
1813 4 : case RS_INVAL_WAL_LEVEL:
1814 4 : 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\"."));
1815 4 : break;
1816 :
1817 0 : case RS_INVAL_IDLE_TIMEOUT:
1818 : {
1819 : /* translator: %s is a GUC variable name */
1820 0 : appendStringInfo(&err_detail, _("The slot's idle time of %lds exceeds the configured \"%s\" duration of %ds."),
1821 : slot_idle_seconds, "idle_replication_slot_timeout",
1822 : idle_replication_slot_timeout_secs);
1823 : /* translator: %s is a GUC variable name */
1824 0 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1825 : "idle_replication_slot_timeout");
1826 0 : break;
1827 : }
1828 : case RS_INVAL_NONE:
1829 : pg_unreachable();
1830 : }
1831 :
1832 23 : ereport(LOG,
1833 : terminating ?
1834 : errmsg("terminating process %d to release replication slot \"%s\"",
1835 : pid, NameStr(slotname)) :
1836 : errmsg("invalidating obsolete replication slot \"%s\"",
1837 : NameStr(slotname)),
1838 : errdetail_internal("%s", err_detail.data),
1839 : err_hint.len ? errhint("%s", err_hint.data) : 0);
1840 :
1841 23 : pfree(err_detail.data);
1842 23 : pfree(err_hint.data);
1843 23 : }
1844 :
1845 : /*
1846 : * Can we invalidate an idle replication slot?
1847 : *
1848 : * Idle timeout invalidation is allowed only when:
1849 : *
1850 : * 1. Idle timeout is set
1851 : * 2. Slot has reserved WAL
1852 : * 3. Slot is inactive
1853 : * 4. The slot is not being synced from the primary while the server is in
1854 : * recovery. This is because synced slots are always considered to be
1855 : * inactive because they don't perform logical decoding to produce changes.
1856 : */
1857 : static inline bool
1858 374 : CanInvalidateIdleSlot(ReplicationSlot *s)
1859 : {
1860 374 : return (idle_replication_slot_timeout_secs != 0 &&
1861 0 : XLogRecPtrIsValid(s->data.restart_lsn) &&
1862 374 : s->inactive_since > 0 &&
1863 0 : !(RecoveryInProgress() && s->data.synced));
1864 : }
1865 :
1866 : /*
1867 : * DetermineSlotInvalidationCause - Determine the cause for which a slot
1868 : * becomes invalid among the given possible causes.
1869 : *
1870 : * This function sequentially checks all possible invalidation causes and
1871 : * returns the first one for which the slot is eligible for invalidation.
1872 : */
1873 : static ReplicationSlotInvalidationCause
1874 409 : DetermineSlotInvalidationCause(uint32 possible_causes, ReplicationSlot *s,
1875 : XLogRecPtr oldestLSN, Oid dboid,
1876 : TransactionId snapshotConflictHorizon,
1877 : TimestampTz *inactive_since, TimestampTz now)
1878 : {
1879 : Assert(possible_causes != RS_INVAL_NONE);
1880 :
1881 409 : if (possible_causes & RS_INVAL_WAL_REMOVED)
1882 : {
1883 381 : XLogRecPtr restart_lsn = s->data.restart_lsn;
1884 :
1885 381 : if (XLogRecPtrIsValid(restart_lsn) &&
1886 : restart_lsn < oldestLSN)
1887 7 : return RS_INVAL_WAL_REMOVED;
1888 : }
1889 :
1890 402 : if (possible_causes & RS_INVAL_HORIZON)
1891 : {
1892 : /* invalid DB oid signals a shared relation */
1893 24 : if (SlotIsLogical(s) &&
1894 19 : (dboid == InvalidOid || dboid == s->data.database))
1895 : {
1896 24 : TransactionId effective_xmin = s->effective_xmin;
1897 24 : TransactionId catalog_effective_xmin = s->effective_catalog_xmin;
1898 :
1899 24 : if (TransactionIdIsValid(effective_xmin) &&
1900 0 : TransactionIdPrecedesOrEquals(effective_xmin,
1901 : snapshotConflictHorizon))
1902 0 : return RS_INVAL_HORIZON;
1903 48 : else if (TransactionIdIsValid(catalog_effective_xmin) &&
1904 24 : TransactionIdPrecedesOrEquals(catalog_effective_xmin,
1905 : snapshotConflictHorizon))
1906 12 : return RS_INVAL_HORIZON;
1907 : }
1908 : }
1909 :
1910 390 : if (possible_causes & RS_INVAL_WAL_LEVEL)
1911 : {
1912 4 : if (SlotIsLogical(s))
1913 4 : return RS_INVAL_WAL_LEVEL;
1914 : }
1915 :
1916 386 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1917 : {
1918 : Assert(now > 0);
1919 :
1920 374 : if (CanInvalidateIdleSlot(s))
1921 : {
1922 : /*
1923 : * Simulate the invalidation due to idle_timeout to test the
1924 : * timeout behavior promptly, without waiting for it to trigger
1925 : * naturally.
1926 : */
1927 : #ifdef USE_INJECTION_POINTS
1928 0 : if (IS_INJECTION_POINT_ATTACHED("slot-timeout-inval"))
1929 : {
1930 0 : *inactive_since = 0; /* since the beginning of time */
1931 0 : return RS_INVAL_IDLE_TIMEOUT;
1932 : }
1933 : #endif
1934 :
1935 : /*
1936 : * Check if the slot needs to be invalidated due to
1937 : * idle_replication_slot_timeout GUC.
1938 : */
1939 0 : if (TimestampDifferenceExceedsSeconds(s->inactive_since, now,
1940 : idle_replication_slot_timeout_secs))
1941 : {
1942 0 : *inactive_since = s->inactive_since;
1943 0 : return RS_INVAL_IDLE_TIMEOUT;
1944 : }
1945 : }
1946 : }
1947 :
1948 386 : return RS_INVAL_NONE;
1949 : }
1950 :
1951 : /*
1952 : * Helper for InvalidateObsoleteReplicationSlots
1953 : *
1954 : * Acquires the given slot and mark it invalid, if necessary and possible.
1955 : *
1956 : * Returns true if the slot was invalidated.
1957 : *
1958 : * Set *released_lock_out if ReplicationSlotControlLock was released in the
1959 : * interim (and in that case we're not holding the lock at return, otherwise
1960 : * we are).
1961 : *
1962 : * This is inherently racy, because we release the LWLock
1963 : * for syscalls, so caller must restart if we return true.
1964 : */
1965 : static bool
1966 448 : InvalidatePossiblyObsoleteSlot(uint32 possible_causes,
1967 : ReplicationSlot *s,
1968 : XLogRecPtr oldestLSN,
1969 : Oid dboid, TransactionId snapshotConflictHorizon,
1970 : bool *released_lock_out)
1971 : {
1972 448 : int last_signaled_pid = 0;
1973 448 : bool released_lock = false;
1974 448 : bool invalidated = false;
1975 448 : TimestampTz inactive_since = 0;
1976 :
1977 : for (;;)
1978 7 : {
1979 : XLogRecPtr restart_lsn;
1980 : NameData slotname;
1981 : ProcNumber active_proc;
1982 455 : int active_pid = 0;
1983 455 : ReplicationSlotInvalidationCause invalidation_cause = RS_INVAL_NONE;
1984 455 : TimestampTz now = 0;
1985 455 : long slot_idle_secs = 0;
1986 :
1987 : Assert(LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_SHARED));
1988 :
1989 455 : if (!s->in_use)
1990 : {
1991 0 : if (released_lock)
1992 0 : LWLockRelease(ReplicationSlotControlLock);
1993 0 : break;
1994 : }
1995 :
1996 455 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1997 : {
1998 : /*
1999 : * Assign the current time here to avoid system call overhead
2000 : * while holding the spinlock in subsequent code.
2001 : */
2002 395 : now = GetCurrentTimestamp();
2003 : }
2004 :
2005 : /*
2006 : * Check if the slot needs to be invalidated. If it needs to be
2007 : * invalidated, and is not currently acquired, acquire it and mark it
2008 : * as having been invalidated. We do this with the spinlock held to
2009 : * avoid race conditions -- for example the restart_lsn could move
2010 : * forward, or the slot could be dropped.
2011 : */
2012 455 : SpinLockAcquire(&s->mutex);
2013 :
2014 455 : restart_lsn = s->data.restart_lsn;
2015 :
2016 : /* we do nothing if the slot is already invalid */
2017 455 : if (s->data.invalidated == RS_INVAL_NONE)
2018 409 : invalidation_cause = DetermineSlotInvalidationCause(possible_causes,
2019 : s, oldestLSN,
2020 : dboid,
2021 : snapshotConflictHorizon,
2022 : &inactive_since,
2023 : now);
2024 :
2025 : /* if there's no invalidation, we're done */
2026 455 : if (invalidation_cause == RS_INVAL_NONE)
2027 : {
2028 432 : SpinLockRelease(&s->mutex);
2029 432 : if (released_lock)
2030 0 : LWLockRelease(ReplicationSlotControlLock);
2031 432 : break;
2032 : }
2033 :
2034 23 : slotname = s->data.name;
2035 23 : active_proc = s->active_proc;
2036 :
2037 : /*
2038 : * If the slot can be acquired, do so and mark it invalidated
2039 : * immediately. Otherwise we'll signal the owning process, below, and
2040 : * retry.
2041 : *
2042 : * Note: Unlike other slot attributes, slot's inactive_since can't be
2043 : * changed until the acquired slot is released or the owning process
2044 : * is terminated. So, the inactive slot can only be invalidated
2045 : * immediately without being terminated.
2046 : */
2047 23 : if (active_proc == INVALID_PROC_NUMBER)
2048 : {
2049 16 : MyReplicationSlot = s;
2050 16 : s->active_proc = MyProcNumber;
2051 16 : s->data.invalidated = invalidation_cause;
2052 :
2053 : /*
2054 : * XXX: We should consider not overwriting restart_lsn and instead
2055 : * just rely on .invalidated.
2056 : */
2057 16 : if (invalidation_cause == RS_INVAL_WAL_REMOVED)
2058 : {
2059 5 : s->data.restart_lsn = InvalidXLogRecPtr;
2060 5 : s->last_saved_restart_lsn = InvalidXLogRecPtr;
2061 : }
2062 :
2063 : /* Let caller know */
2064 16 : invalidated = true;
2065 : }
2066 : else
2067 : {
2068 7 : active_pid = GetPGProcByNumber(active_proc)->pid;
2069 : Assert(active_pid != 0);
2070 : }
2071 :
2072 23 : SpinLockRelease(&s->mutex);
2073 :
2074 : /*
2075 : * Calculate the idle time duration of the slot if slot is marked
2076 : * invalidated with RS_INVAL_IDLE_TIMEOUT.
2077 : */
2078 23 : if (invalidation_cause == RS_INVAL_IDLE_TIMEOUT)
2079 : {
2080 : int slot_idle_usecs;
2081 :
2082 0 : TimestampDifference(inactive_since, now, &slot_idle_secs,
2083 : &slot_idle_usecs);
2084 : }
2085 :
2086 23 : if (active_proc != INVALID_PROC_NUMBER)
2087 : {
2088 : /*
2089 : * Prepare the sleep on the slot's condition variable before
2090 : * releasing the lock, to close a possible race condition if the
2091 : * slot is released before the sleep below.
2092 : */
2093 7 : ConditionVariablePrepareToSleep(&s->active_cv);
2094 :
2095 7 : LWLockRelease(ReplicationSlotControlLock);
2096 7 : released_lock = true;
2097 :
2098 : /*
2099 : * Signal to terminate the process that owns the slot, if we
2100 : * haven't already signalled it. (Avoidance of repeated
2101 : * signalling is the only reason for there to be a loop in this
2102 : * routine; otherwise we could rely on caller's restart loop.)
2103 : *
2104 : * There is the race condition that other process may own the slot
2105 : * after its current owner process is terminated and before this
2106 : * process owns it. To handle that, we signal only if the PID of
2107 : * the owning process has changed from the previous time. (This
2108 : * logic assumes that the same PID is not reused very quickly.)
2109 : */
2110 7 : if (last_signaled_pid != active_pid)
2111 : {
2112 7 : ReportSlotInvalidation(invalidation_cause, true, active_pid,
2113 : slotname, restart_lsn,
2114 : oldestLSN, snapshotConflictHorizon,
2115 : slot_idle_secs);
2116 :
2117 7 : if (MyBackendType == B_STARTUP)
2118 5 : (void) SignalRecoveryConflict(GetPGProcByNumber(active_proc),
2119 : active_pid,
2120 : RECOVERY_CONFLICT_LOGICALSLOT);
2121 : else
2122 2 : (void) kill(active_pid, SIGTERM);
2123 :
2124 7 : last_signaled_pid = active_pid;
2125 : }
2126 :
2127 : /* Wait until the slot is released. */
2128 7 : ConditionVariableSleep(&s->active_cv,
2129 : WAIT_EVENT_REPLICATION_SLOT_DROP);
2130 :
2131 : /*
2132 : * Re-acquire lock and start over; we expect to invalidate the
2133 : * slot next time (unless another process acquires the slot in the
2134 : * meantime).
2135 : *
2136 : * Note: It is possible for a slot to advance its restart_lsn or
2137 : * xmin values sufficiently between when we release the mutex and
2138 : * when we recheck, moving from a conflicting state to a non
2139 : * conflicting state. This is intentional and safe: if the slot
2140 : * has caught up while we're busy here, the resources we were
2141 : * concerned about (WAL segments or tuples) have not yet been
2142 : * removed, and there's no reason to invalidate the slot.
2143 : */
2144 7 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2145 7 : continue;
2146 : }
2147 : else
2148 : {
2149 : /*
2150 : * We hold the slot now and have already invalidated it; flush it
2151 : * to ensure that state persists.
2152 : *
2153 : * Don't want to hold ReplicationSlotControlLock across file
2154 : * system operations, so release it now but be sure to tell caller
2155 : * to restart from scratch.
2156 : */
2157 16 : LWLockRelease(ReplicationSlotControlLock);
2158 16 : released_lock = true;
2159 :
2160 : /* Make sure the invalidated state persists across server restart */
2161 16 : ReplicationSlotMarkDirty();
2162 16 : ReplicationSlotSave();
2163 16 : ReplicationSlotRelease();
2164 :
2165 16 : ReportSlotInvalidation(invalidation_cause, false, active_pid,
2166 : slotname, restart_lsn,
2167 : oldestLSN, snapshotConflictHorizon,
2168 : slot_idle_secs);
2169 :
2170 : /* done with this slot for now */
2171 16 : break;
2172 : }
2173 : }
2174 :
2175 : Assert(released_lock == !LWLockHeldByMe(ReplicationSlotControlLock));
2176 :
2177 448 : *released_lock_out = released_lock;
2178 448 : return invalidated;
2179 : }
2180 :
2181 : /*
2182 : * Invalidate slots that require resources about to be removed.
2183 : *
2184 : * Returns true when any slot have got invalidated.
2185 : *
2186 : * Whether a slot needs to be invalidated depends on the invalidation cause.
2187 : * A slot is invalidated if it:
2188 : * - RS_INVAL_WAL_REMOVED: requires a LSN older than the given segment
2189 : * - RS_INVAL_HORIZON: requires a snapshot <= the given horizon in the given
2190 : * db; dboid may be InvalidOid for shared relations
2191 : * - RS_INVAL_WAL_LEVEL: is a logical slot and effective_wal_level is not
2192 : * logical.
2193 : * - RS_INVAL_IDLE_TIMEOUT: has been idle longer than the configured
2194 : * "idle_replication_slot_timeout" duration.
2195 : *
2196 : * Note: This function attempts to invalidate the slot for multiple possible
2197 : * causes in a single pass, minimizing redundant iterations. The "cause"
2198 : * parameter can be a MASK representing one or more of the defined causes.
2199 : *
2200 : * If it invalidates the last logical slot in the cluster, it requests to
2201 : * disable logical decoding.
2202 : *
2203 : * NB - this runs as part of checkpoint, so avoid raising errors if possible.
2204 : */
2205 : bool
2206 1861 : InvalidateObsoleteReplicationSlots(uint32 possible_causes,
2207 : XLogSegNo oldestSegno, Oid dboid,
2208 : TransactionId snapshotConflictHorizon)
2209 : {
2210 : XLogRecPtr oldestLSN;
2211 1861 : bool invalidated = false;
2212 1861 : bool invalidated_logical = false;
2213 : bool found_valid_logicalslot;
2214 :
2215 : Assert(!(possible_causes & RS_INVAL_HORIZON) || TransactionIdIsValid(snapshotConflictHorizon));
2216 : Assert(!(possible_causes & RS_INVAL_WAL_REMOVED) || oldestSegno > 0);
2217 : Assert(possible_causes != RS_INVAL_NONE);
2218 :
2219 1861 : if (max_replication_slots == 0)
2220 1 : return invalidated;
2221 :
2222 1860 : XLogSegNoOffsetToRecPtr(oldestSegno, 0, wal_segment_size, oldestLSN);
2223 :
2224 1876 : restart:
2225 1876 : found_valid_logicalslot = false;
2226 1876 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2227 20126 : for (int i = 0; i < max_replication_slots; i++)
2228 : {
2229 18266 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2230 18266 : bool released_lock = false;
2231 :
2232 18266 : if (!s->in_use)
2233 17818 : continue;
2234 :
2235 : /* Prevent invalidation of logical slots during binary upgrade */
2236 457 : if (SlotIsLogical(s) && IsBinaryUpgrade)
2237 : {
2238 9 : SpinLockAcquire(&s->mutex);
2239 9 : found_valid_logicalslot |= (s->data.invalidated == RS_INVAL_NONE);
2240 9 : SpinLockRelease(&s->mutex);
2241 :
2242 9 : continue;
2243 : }
2244 :
2245 448 : if (InvalidatePossiblyObsoleteSlot(possible_causes, s, oldestLSN,
2246 : dboid, snapshotConflictHorizon,
2247 : &released_lock))
2248 : {
2249 : Assert(released_lock);
2250 :
2251 : /* Remember we have invalidated a physical or logical slot */
2252 16 : invalidated = true;
2253 :
2254 : /*
2255 : * Additionally, remember we have invalidated a logical slot as we
2256 : * can request disabling logical decoding later.
2257 : */
2258 16 : if (SlotIsLogical(s))
2259 13 : invalidated_logical = true;
2260 : }
2261 : else
2262 : {
2263 : /*
2264 : * We need to check if the slot is invalidated here since
2265 : * InvalidatePossiblyObsoleteSlot() returns false also if the slot
2266 : * is already invalidated.
2267 : */
2268 432 : SpinLockAcquire(&s->mutex);
2269 864 : found_valid_logicalslot |=
2270 432 : (SlotIsLogical(s) && (s->data.invalidated == RS_INVAL_NONE));
2271 432 : SpinLockRelease(&s->mutex);
2272 : }
2273 :
2274 : /* if the lock was released, start from scratch */
2275 448 : if (released_lock)
2276 16 : goto restart;
2277 : }
2278 1860 : LWLockRelease(ReplicationSlotControlLock);
2279 :
2280 : /*
2281 : * If any slots have been invalidated, recalculate the resource limits.
2282 : */
2283 1860 : if (invalidated)
2284 : {
2285 11 : ReplicationSlotsComputeRequiredXmin(false);
2286 11 : ReplicationSlotsComputeRequiredLSN();
2287 : }
2288 :
2289 : /*
2290 : * Request the checkpointer to disable logical decoding if no valid
2291 : * logical slots remain. If called by the checkpointer during a
2292 : * checkpoint, only the request is initiated; actual deactivation is
2293 : * deferred until after the checkpoint completes.
2294 : */
2295 1860 : if (invalidated_logical && !found_valid_logicalslot)
2296 8 : RequestDisableLogicalDecoding();
2297 :
2298 1860 : return invalidated;
2299 : }
2300 :
2301 : /*
2302 : * Flush all replication slots to disk.
2303 : *
2304 : * It is convenient to flush dirty replication slots at the time of checkpoint.
2305 : * Additionally, in case of a shutdown checkpoint, we also identify the slots
2306 : * for which the confirmed_flush LSN has been updated since the last time it
2307 : * was saved and flush them.
2308 : */
2309 : void
2310 1835 : CheckPointReplicationSlots(bool is_shutdown)
2311 : {
2312 : int i;
2313 1835 : bool last_saved_restart_lsn_updated = false;
2314 :
2315 1835 : elog(DEBUG1, "performing replication slot checkpoint");
2316 :
2317 : /*
2318 : * Prevent any slot from being created/dropped while we're active. As we
2319 : * explicitly do *not* want to block iterating over replication_slots or
2320 : * acquiring a slot we cannot take the control lock - but that's OK,
2321 : * because holding ReplicationSlotAllocationLock is strictly stronger, and
2322 : * enough to guarantee that nobody can change the in_use bits on us.
2323 : *
2324 : * Additionally, acquiring the Allocation lock is necessary to serialize
2325 : * the slot flush process with concurrent slot WAL reservation. This
2326 : * ensures that the WAL position being reserved is either flushed to disk
2327 : * or is beyond or equal to the redo pointer of the current checkpoint
2328 : * (See ReplicationSlotReserveWal for details).
2329 : */
2330 1835 : LWLockAcquire(ReplicationSlotAllocationLock, LW_SHARED);
2331 :
2332 19910 : for (i = 0; i < max_replication_slots; i++)
2333 : {
2334 18075 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2335 : char path[MAXPGPATH];
2336 :
2337 18075 : if (!s->in_use)
2338 17678 : continue;
2339 :
2340 : /* save the slot to disk, locking is handled in SaveSlotToPath() */
2341 397 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(s->data.name));
2342 :
2343 : /*
2344 : * Slot's data is not flushed each time the confirmed_flush LSN is
2345 : * updated as that could lead to frequent writes. However, we decide
2346 : * to force a flush of all logical slot's data at the time of shutdown
2347 : * if the confirmed_flush LSN is changed since we last flushed it to
2348 : * disk. This helps in avoiding an unnecessary retreat of the
2349 : * confirmed_flush LSN after restart.
2350 : */
2351 397 : if (is_shutdown && SlotIsLogical(s))
2352 : {
2353 89 : SpinLockAcquire(&s->mutex);
2354 :
2355 89 : if (s->data.invalidated == RS_INVAL_NONE &&
2356 88 : s->data.confirmed_flush > s->last_saved_confirmed_flush)
2357 : {
2358 40 : s->just_dirtied = true;
2359 40 : s->dirty = true;
2360 : }
2361 89 : SpinLockRelease(&s->mutex);
2362 : }
2363 :
2364 : /*
2365 : * Track if we're going to update slot's last_saved_restart_lsn. We
2366 : * need this to know if we need to recompute the required LSN.
2367 : */
2368 397 : if (s->last_saved_restart_lsn != s->data.restart_lsn)
2369 207 : last_saved_restart_lsn_updated = true;
2370 :
2371 397 : SaveSlotToPath(s, path, LOG);
2372 : }
2373 1835 : LWLockRelease(ReplicationSlotAllocationLock);
2374 :
2375 : /*
2376 : * Recompute the required LSN if SaveSlotToPath() updated
2377 : * last_saved_restart_lsn for any slot.
2378 : */
2379 1835 : if (last_saved_restart_lsn_updated)
2380 207 : ReplicationSlotsComputeRequiredLSN();
2381 1835 : }
2382 :
2383 : /*
2384 : * Load all replication slots from disk into memory at server startup. This
2385 : * needs to be run before we start crash recovery.
2386 : */
2387 : void
2388 1033 : StartupReplicationSlots(void)
2389 : {
2390 : DIR *replication_dir;
2391 : struct dirent *replication_de;
2392 :
2393 1033 : elog(DEBUG1, "starting up replication slots");
2394 :
2395 : /* restore all slots by iterating over all on-disk entries */
2396 1033 : replication_dir = AllocateDir(PG_REPLSLOT_DIR);
2397 3214 : while ((replication_de = ReadDir(replication_dir, PG_REPLSLOT_DIR)) != NULL)
2398 : {
2399 : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2400 : PGFileType de_type;
2401 :
2402 2183 : if (strcmp(replication_de->d_name, ".") == 0 ||
2403 1152 : strcmp(replication_de->d_name, "..") == 0)
2404 2062 : continue;
2405 :
2406 121 : snprintf(path, sizeof(path), "%s/%s", PG_REPLSLOT_DIR, replication_de->d_name);
2407 121 : de_type = get_dirent_type(path, replication_de, false, DEBUG1);
2408 :
2409 : /* we're only creating directories here, skip if it's not our's */
2410 121 : if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_DIR)
2411 0 : continue;
2412 :
2413 : /* we crashed while a slot was being setup or deleted, clean up */
2414 121 : if (pg_str_endswith(replication_de->d_name, ".tmp"))
2415 : {
2416 0 : if (!rmtree(path, true))
2417 : {
2418 0 : ereport(WARNING,
2419 : (errmsg("could not remove directory \"%s\"",
2420 : path)));
2421 0 : continue;
2422 : }
2423 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2424 0 : continue;
2425 : }
2426 :
2427 : /* looks like a slot in a normal state, restore */
2428 121 : RestoreSlotFromDisk(replication_de->d_name);
2429 : }
2430 1031 : FreeDir(replication_dir);
2431 :
2432 : /* currently no slots exist, we're done. */
2433 1031 : if (max_replication_slots <= 0)
2434 1 : return;
2435 :
2436 : /* Now that we have recovered all the data, compute replication xmin */
2437 1030 : ReplicationSlotsComputeRequiredXmin(false);
2438 1030 : ReplicationSlotsComputeRequiredLSN();
2439 : }
2440 :
2441 : /* ----
2442 : * Manipulation of on-disk state of replication slots
2443 : *
2444 : * NB: none of the routines below should take any notice whether a slot is the
2445 : * current one or not, that's all handled a layer above.
2446 : * ----
2447 : */
2448 : static void
2449 701 : CreateSlotOnDisk(ReplicationSlot *slot)
2450 : {
2451 : char tmppath[MAXPGPATH];
2452 : char path[MAXPGPATH];
2453 : struct stat st;
2454 :
2455 : /*
2456 : * No need to take out the io_in_progress_lock, nobody else can see this
2457 : * slot yet, so nobody else will write. We're reusing SaveSlotToPath which
2458 : * takes out the lock, if we'd take the lock here, we'd deadlock.
2459 : */
2460 :
2461 701 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2462 701 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2463 :
2464 : /*
2465 : * It's just barely possible that some previous effort to create or drop a
2466 : * slot with this name left a temp directory lying around. If that seems
2467 : * to be the case, try to remove it. If the rmtree() fails, we'll error
2468 : * out at the MakePGDirectory() below, so we don't bother checking
2469 : * success.
2470 : */
2471 701 : if (stat(tmppath, &st) == 0 && S_ISDIR(st.st_mode))
2472 0 : rmtree(tmppath, true);
2473 :
2474 : /* Create and fsync the temporary slot directory. */
2475 701 : if (MakePGDirectory(tmppath) < 0)
2476 0 : ereport(ERROR,
2477 : (errcode_for_file_access(),
2478 : errmsg("could not create directory \"%s\": %m",
2479 : tmppath)));
2480 701 : fsync_fname(tmppath, true);
2481 :
2482 : /* Write the actual state file. */
2483 701 : slot->dirty = true; /* signal that we really need to write */
2484 701 : SaveSlotToPath(slot, tmppath, ERROR);
2485 :
2486 : /* Rename the directory into place. */
2487 701 : if (rename(tmppath, path) != 0)
2488 0 : ereport(ERROR,
2489 : (errcode_for_file_access(),
2490 : errmsg("could not rename file \"%s\" to \"%s\": %m",
2491 : tmppath, path)));
2492 :
2493 : /*
2494 : * If we'd now fail - really unlikely - we wouldn't know whether this slot
2495 : * would persist after an OS crash or not - so, force a restart. The
2496 : * restart would try to fsync this again till it works.
2497 : */
2498 701 : START_CRIT_SECTION();
2499 :
2500 701 : fsync_fname(path, true);
2501 701 : fsync_fname(PG_REPLSLOT_DIR, true);
2502 :
2503 701 : END_CRIT_SECTION();
2504 701 : }
2505 :
2506 : /*
2507 : * Shared functionality between saving and creating a replication slot.
2508 : */
2509 : static void
2510 2578 : SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
2511 : {
2512 : char tmppath[MAXPGPATH];
2513 : char path[MAXPGPATH];
2514 : int fd;
2515 : ReplicationSlotOnDisk cp;
2516 : bool was_dirty;
2517 :
2518 : /* first check whether there's something to write out */
2519 2578 : SpinLockAcquire(&slot->mutex);
2520 2578 : was_dirty = slot->dirty;
2521 2578 : slot->just_dirtied = false;
2522 2578 : SpinLockRelease(&slot->mutex);
2523 :
2524 : /* and don't do anything if there's nothing to write */
2525 2578 : if (!was_dirty)
2526 139 : return;
2527 :
2528 2439 : LWLockAcquire(&slot->io_in_progress_lock, LW_EXCLUSIVE);
2529 :
2530 : /* silence valgrind :( */
2531 2439 : memset(&cp, 0, sizeof(ReplicationSlotOnDisk));
2532 :
2533 2439 : sprintf(tmppath, "%s/state.tmp", dir);
2534 2439 : sprintf(path, "%s/state", dir);
2535 :
2536 2439 : fd = OpenTransientFile(tmppath, O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
2537 2439 : if (fd < 0)
2538 : {
2539 : /*
2540 : * If not an ERROR, then release the lock before returning. In case
2541 : * of an ERROR, the error recovery path automatically releases the
2542 : * lock, but no harm in explicitly releasing even in that case. Note
2543 : * that LWLockRelease() could affect errno.
2544 : */
2545 0 : int save_errno = errno;
2546 :
2547 0 : LWLockRelease(&slot->io_in_progress_lock);
2548 0 : errno = save_errno;
2549 0 : ereport(elevel,
2550 : (errcode_for_file_access(),
2551 : errmsg("could not create file \"%s\": %m",
2552 : tmppath)));
2553 0 : return;
2554 : }
2555 :
2556 2439 : cp.magic = SLOT_MAGIC;
2557 2439 : INIT_CRC32C(cp.checksum);
2558 2439 : cp.version = SLOT_VERSION;
2559 2439 : cp.length = ReplicationSlotOnDiskV2Size;
2560 :
2561 2439 : SpinLockAcquire(&slot->mutex);
2562 :
2563 2439 : memcpy(&cp.slotdata, &slot->data, sizeof(ReplicationSlotPersistentData));
2564 :
2565 2439 : SpinLockRelease(&slot->mutex);
2566 :
2567 2439 : COMP_CRC32C(cp.checksum,
2568 : (char *) (&cp) + ReplicationSlotOnDiskNotChecksummedSize,
2569 : ReplicationSlotOnDiskChecksummedSize);
2570 2439 : FIN_CRC32C(cp.checksum);
2571 :
2572 2439 : errno = 0;
2573 2439 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_WRITE);
2574 2439 : if ((write(fd, &cp, sizeof(cp))) != sizeof(cp))
2575 : {
2576 0 : int save_errno = errno;
2577 :
2578 0 : pgstat_report_wait_end();
2579 0 : CloseTransientFile(fd);
2580 0 : unlink(tmppath);
2581 0 : LWLockRelease(&slot->io_in_progress_lock);
2582 :
2583 : /* if write didn't set errno, assume problem is no disk space */
2584 0 : errno = save_errno ? save_errno : ENOSPC;
2585 0 : ereport(elevel,
2586 : (errcode_for_file_access(),
2587 : errmsg("could not write to file \"%s\": %m",
2588 : tmppath)));
2589 0 : return;
2590 : }
2591 2439 : pgstat_report_wait_end();
2592 :
2593 : /* fsync the temporary file */
2594 2439 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_SYNC);
2595 2439 : if (pg_fsync(fd) != 0)
2596 : {
2597 0 : int save_errno = errno;
2598 :
2599 0 : pgstat_report_wait_end();
2600 0 : CloseTransientFile(fd);
2601 0 : unlink(tmppath);
2602 0 : LWLockRelease(&slot->io_in_progress_lock);
2603 :
2604 0 : errno = save_errno;
2605 0 : ereport(elevel,
2606 : (errcode_for_file_access(),
2607 : errmsg("could not fsync file \"%s\": %m",
2608 : tmppath)));
2609 0 : return;
2610 : }
2611 2439 : pgstat_report_wait_end();
2612 :
2613 2439 : if (CloseTransientFile(fd) != 0)
2614 : {
2615 0 : int save_errno = errno;
2616 :
2617 0 : unlink(tmppath);
2618 0 : LWLockRelease(&slot->io_in_progress_lock);
2619 :
2620 0 : errno = save_errno;
2621 0 : ereport(elevel,
2622 : (errcode_for_file_access(),
2623 : errmsg("could not close file \"%s\": %m",
2624 : tmppath)));
2625 0 : return;
2626 : }
2627 :
2628 : /* rename to permanent file, fsync file and directory */
2629 2439 : if (rename(tmppath, path) != 0)
2630 : {
2631 0 : int save_errno = errno;
2632 :
2633 0 : unlink(tmppath);
2634 0 : LWLockRelease(&slot->io_in_progress_lock);
2635 :
2636 0 : errno = save_errno;
2637 0 : ereport(elevel,
2638 : (errcode_for_file_access(),
2639 : errmsg("could not rename file \"%s\" to \"%s\": %m",
2640 : tmppath, path)));
2641 0 : return;
2642 : }
2643 :
2644 : /*
2645 : * Check CreateSlotOnDisk() for the reasoning of using a critical section.
2646 : */
2647 2439 : START_CRIT_SECTION();
2648 :
2649 2439 : fsync_fname(path, false);
2650 2439 : fsync_fname(dir, true);
2651 2439 : fsync_fname(PG_REPLSLOT_DIR, true);
2652 :
2653 2439 : END_CRIT_SECTION();
2654 :
2655 : /*
2656 : * Successfully wrote, unset dirty bit, unless somebody dirtied again
2657 : * already and remember the confirmed_flush LSN value.
2658 : */
2659 2439 : SpinLockAcquire(&slot->mutex);
2660 2439 : if (!slot->just_dirtied)
2661 2416 : slot->dirty = false;
2662 2439 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2663 2439 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2664 2439 : SpinLockRelease(&slot->mutex);
2665 :
2666 2439 : LWLockRelease(&slot->io_in_progress_lock);
2667 : }
2668 :
2669 : /*
2670 : * Load a single slot from disk into memory.
2671 : */
2672 : static void
2673 121 : RestoreSlotFromDisk(const char *name)
2674 : {
2675 : ReplicationSlotOnDisk cp;
2676 : int i;
2677 : char slotdir[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2678 : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR) + 10];
2679 : int fd;
2680 121 : bool restored = false;
2681 : int readBytes;
2682 : pg_crc32c checksum;
2683 121 : TimestampTz now = 0;
2684 :
2685 : /* no need to lock here, no concurrent access allowed yet */
2686 :
2687 : /* delete temp file if it exists */
2688 121 : sprintf(slotdir, "%s/%s", PG_REPLSLOT_DIR, name);
2689 121 : sprintf(path, "%s/state.tmp", slotdir);
2690 121 : if (unlink(path) < 0 && errno != ENOENT)
2691 0 : ereport(PANIC,
2692 : (errcode_for_file_access(),
2693 : errmsg("could not remove file \"%s\": %m", path)));
2694 :
2695 121 : sprintf(path, "%s/state", slotdir);
2696 :
2697 121 : elog(DEBUG1, "restoring replication slot from \"%s\"", path);
2698 :
2699 : /* on some operating systems fsyncing a file requires O_RDWR */
2700 121 : fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
2701 :
2702 : /*
2703 : * We do not need to handle this as we are rename()ing the directory into
2704 : * place only after we fsync()ed the state file.
2705 : */
2706 121 : if (fd < 0)
2707 0 : ereport(PANIC,
2708 : (errcode_for_file_access(),
2709 : errmsg("could not open file \"%s\": %m", path)));
2710 :
2711 : /*
2712 : * Sync state file before we're reading from it. We might have crashed
2713 : * while it wasn't synced yet and we shouldn't continue on that basis.
2714 : */
2715 121 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_RESTORE_SYNC);
2716 121 : if (pg_fsync(fd) != 0)
2717 0 : ereport(PANIC,
2718 : (errcode_for_file_access(),
2719 : errmsg("could not fsync file \"%s\": %m",
2720 : path)));
2721 121 : pgstat_report_wait_end();
2722 :
2723 : /* Also sync the parent directory */
2724 121 : START_CRIT_SECTION();
2725 121 : fsync_fname(slotdir, true);
2726 121 : END_CRIT_SECTION();
2727 :
2728 : /* read part of statefile that's guaranteed to be version independent */
2729 121 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2730 121 : readBytes = read(fd, &cp, ReplicationSlotOnDiskConstantSize);
2731 121 : pgstat_report_wait_end();
2732 121 : if (readBytes != ReplicationSlotOnDiskConstantSize)
2733 : {
2734 0 : if (readBytes < 0)
2735 0 : ereport(PANIC,
2736 : (errcode_for_file_access(),
2737 : errmsg("could not read file \"%s\": %m", path)));
2738 : else
2739 0 : ereport(PANIC,
2740 : (errcode(ERRCODE_DATA_CORRUPTED),
2741 : errmsg("could not read file \"%s\": read %d of %zu",
2742 : path, readBytes,
2743 : (Size) ReplicationSlotOnDiskConstantSize)));
2744 : }
2745 :
2746 : /* verify magic */
2747 121 : if (cp.magic != SLOT_MAGIC)
2748 0 : ereport(PANIC,
2749 : (errcode(ERRCODE_DATA_CORRUPTED),
2750 : errmsg("replication slot file \"%s\" has wrong magic number: %u instead of %u",
2751 : path, cp.magic, SLOT_MAGIC)));
2752 :
2753 : /* verify version */
2754 121 : if (cp.version != SLOT_VERSION)
2755 0 : ereport(PANIC,
2756 : (errcode(ERRCODE_DATA_CORRUPTED),
2757 : errmsg("replication slot file \"%s\" has unsupported version %u",
2758 : path, cp.version)));
2759 :
2760 : /* boundary check on length */
2761 121 : if (cp.length != ReplicationSlotOnDiskV2Size)
2762 0 : ereport(PANIC,
2763 : (errcode(ERRCODE_DATA_CORRUPTED),
2764 : errmsg("replication slot file \"%s\" has corrupted length %u",
2765 : path, cp.length)));
2766 :
2767 : /* Now that we know the size, read the entire file */
2768 121 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2769 242 : readBytes = read(fd,
2770 : (char *) &cp + ReplicationSlotOnDiskConstantSize,
2771 121 : cp.length);
2772 121 : pgstat_report_wait_end();
2773 121 : if (readBytes != cp.length)
2774 : {
2775 0 : if (readBytes < 0)
2776 0 : ereport(PANIC,
2777 : (errcode_for_file_access(),
2778 : errmsg("could not read file \"%s\": %m", path)));
2779 : else
2780 0 : ereport(PANIC,
2781 : (errcode(ERRCODE_DATA_CORRUPTED),
2782 : errmsg("could not read file \"%s\": read %d of %zu",
2783 : path, readBytes, (Size) cp.length)));
2784 : }
2785 :
2786 121 : if (CloseTransientFile(fd) != 0)
2787 0 : ereport(PANIC,
2788 : (errcode_for_file_access(),
2789 : errmsg("could not close file \"%s\": %m", path)));
2790 :
2791 : /* now verify the CRC */
2792 121 : INIT_CRC32C(checksum);
2793 121 : COMP_CRC32C(checksum,
2794 : (char *) &cp + ReplicationSlotOnDiskNotChecksummedSize,
2795 : ReplicationSlotOnDiskChecksummedSize);
2796 121 : FIN_CRC32C(checksum);
2797 :
2798 121 : if (!EQ_CRC32C(checksum, cp.checksum))
2799 0 : ereport(PANIC,
2800 : (errmsg("checksum mismatch for replication slot file \"%s\": is %u, should be %u",
2801 : path, checksum, cp.checksum)));
2802 :
2803 : /*
2804 : * If we crashed with an ephemeral slot active, don't restore but delete
2805 : * it.
2806 : */
2807 121 : if (cp.slotdata.persistency != RS_PERSISTENT)
2808 : {
2809 0 : if (!rmtree(slotdir, true))
2810 : {
2811 0 : ereport(WARNING,
2812 : (errmsg("could not remove directory \"%s\"",
2813 : slotdir)));
2814 : }
2815 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2816 0 : return;
2817 : }
2818 :
2819 : /*
2820 : * Verify that requirements for the specific slot type are met. That's
2821 : * important because if these aren't met we're not guaranteed to retain
2822 : * all the necessary resources for the slot.
2823 : *
2824 : * NB: We have to do so *after* the above checks for ephemeral slots,
2825 : * because otherwise a slot that shouldn't exist anymore could prevent
2826 : * restarts.
2827 : *
2828 : * NB: Changing the requirements here also requires adapting
2829 : * CheckSlotRequirements() and CheckLogicalDecodingRequirements().
2830 : */
2831 121 : if (cp.slotdata.database != InvalidOid)
2832 : {
2833 83 : if (wal_level < WAL_LEVEL_REPLICA)
2834 1 : ereport(FATAL,
2835 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2836 : errmsg("logical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2837 : NameStr(cp.slotdata.name)),
2838 : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2839 :
2840 : /*
2841 : * In standby mode, the hot standby must be enabled. This check is
2842 : * necessary to ensure logical slots are invalidated when they become
2843 : * incompatible due to insufficient wal_level. Otherwise, if the
2844 : * primary reduces effective_wal_level < logical while hot standby is
2845 : * disabled, primary disable logical decoding while hot standby is
2846 : * disabled, logical slots would remain valid even after promotion.
2847 : */
2848 82 : if (StandbyMode && !EnableHotStandby)
2849 1 : ereport(FATAL,
2850 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2851 : errmsg("logical replication slot \"%s\" exists on the standby, but \"hot_standby\" = \"off\"",
2852 : NameStr(cp.slotdata.name)),
2853 : errhint("Change \"hot_standby\" to be \"on\".")));
2854 : }
2855 38 : else if (wal_level < WAL_LEVEL_REPLICA)
2856 0 : ereport(FATAL,
2857 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2858 : errmsg("physical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2859 : NameStr(cp.slotdata.name)),
2860 : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2861 :
2862 : /* nothing can be active yet, don't lock anything */
2863 176 : for (i = 0; i < max_replication_slots; i++)
2864 : {
2865 : ReplicationSlot *slot;
2866 :
2867 176 : slot = &ReplicationSlotCtl->replication_slots[i];
2868 :
2869 176 : if (slot->in_use)
2870 57 : continue;
2871 :
2872 : /* restore the entire set of persistent data */
2873 119 : memcpy(&slot->data, &cp.slotdata,
2874 : sizeof(ReplicationSlotPersistentData));
2875 :
2876 : /* initialize in memory state */
2877 119 : slot->effective_xmin = cp.slotdata.xmin;
2878 119 : slot->effective_catalog_xmin = cp.slotdata.catalog_xmin;
2879 119 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2880 119 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2881 :
2882 119 : slot->candidate_catalog_xmin = InvalidTransactionId;
2883 119 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
2884 119 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
2885 119 : slot->candidate_restart_valid = InvalidXLogRecPtr;
2886 :
2887 119 : slot->in_use = true;
2888 119 : slot->active_proc = INVALID_PROC_NUMBER;
2889 :
2890 : /*
2891 : * Set the time since the slot has become inactive after loading the
2892 : * slot from the disk into memory. Whoever acquires the slot i.e.
2893 : * makes the slot active will reset it. Use the same inactive_since
2894 : * time for all the slots.
2895 : */
2896 119 : if (now == 0)
2897 119 : now = GetCurrentTimestamp();
2898 :
2899 119 : ReplicationSlotSetInactiveSince(slot, now, false);
2900 :
2901 119 : restored = true;
2902 119 : break;
2903 : }
2904 :
2905 119 : if (!restored)
2906 0 : ereport(FATAL,
2907 : (errmsg("too many replication slots active before shutdown"),
2908 : errhint("Increase \"max_replication_slots\" and try again.")));
2909 : }
2910 :
2911 : /*
2912 : * Maps an invalidation reason for a replication slot to
2913 : * ReplicationSlotInvalidationCause.
2914 : */
2915 : ReplicationSlotInvalidationCause
2916 0 : GetSlotInvalidationCause(const char *cause_name)
2917 : {
2918 : Assert(cause_name);
2919 :
2920 : /* Search lookup table for the cause having this name */
2921 0 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2922 : {
2923 0 : if (strcmp(SlotInvalidationCauses[i].cause_name, cause_name) == 0)
2924 0 : return SlotInvalidationCauses[i].cause;
2925 : }
2926 :
2927 : Assert(false);
2928 0 : return RS_INVAL_NONE; /* to keep compiler quiet */
2929 : }
2930 :
2931 : /*
2932 : * Maps a ReplicationSlotInvalidationCause to the invalidation
2933 : * reason for a replication slot.
2934 : */
2935 : const char *
2936 45 : GetSlotInvalidationCauseName(ReplicationSlotInvalidationCause cause)
2937 : {
2938 : /* Search lookup table for the name of this cause */
2939 146 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2940 : {
2941 146 : if (SlotInvalidationCauses[i].cause == cause)
2942 45 : return SlotInvalidationCauses[i].cause_name;
2943 : }
2944 :
2945 : Assert(false);
2946 0 : return "none"; /* to keep compiler quiet */
2947 : }
2948 :
2949 : /*
2950 : * A helper function to validate slots specified in GUC synchronized_standby_slots.
2951 : *
2952 : * The rawname will be parsed, and the result will be saved into *elemlist.
2953 : */
2954 : static bool
2955 19 : validate_sync_standby_slots(char *rawname, List **elemlist)
2956 : {
2957 : /* Verify syntax and parse string into a list of identifiers */
2958 19 : if (!SplitIdentifierString(rawname, ',', elemlist))
2959 : {
2960 0 : GUC_check_errdetail("List syntax is invalid.");
2961 0 : return false;
2962 : }
2963 :
2964 : /* Iterate the list to validate each slot name */
2965 53 : foreach_ptr(char, name, *elemlist)
2966 : {
2967 : int err_code;
2968 19 : char *err_msg = NULL;
2969 19 : char *err_hint = NULL;
2970 :
2971 19 : if (!ReplicationSlotValidateNameInternal(name, false, &err_code,
2972 : &err_msg, &err_hint))
2973 : {
2974 2 : GUC_check_errcode(err_code);
2975 2 : GUC_check_errdetail("%s", err_msg);
2976 2 : if (err_hint != NULL)
2977 2 : GUC_check_errhint("%s", err_hint);
2978 2 : return false;
2979 : }
2980 : }
2981 :
2982 17 : return true;
2983 : }
2984 :
2985 : /*
2986 : * GUC check_hook for synchronized_standby_slots
2987 : */
2988 : bool
2989 1254 : check_synchronized_standby_slots(char **newval, void **extra, GucSource source)
2990 : {
2991 : char *rawname;
2992 : char *ptr;
2993 : List *elemlist;
2994 : int size;
2995 : bool ok;
2996 : SyncStandbySlotsConfigData *config;
2997 :
2998 1254 : if ((*newval)[0] == '\0')
2999 1235 : return true;
3000 :
3001 : /* Need a modifiable copy of the GUC string */
3002 19 : rawname = pstrdup(*newval);
3003 :
3004 : /* Now verify if the specified slots exist and have correct type */
3005 19 : ok = validate_sync_standby_slots(rawname, &elemlist);
3006 :
3007 19 : if (!ok || elemlist == NIL)
3008 : {
3009 2 : pfree(rawname);
3010 2 : list_free(elemlist);
3011 2 : return ok;
3012 : }
3013 :
3014 : /* Compute the size required for the SyncStandbySlotsConfigData struct */
3015 17 : size = offsetof(SyncStandbySlotsConfigData, slot_names);
3016 51 : foreach_ptr(char, slot_name, elemlist)
3017 17 : size += strlen(slot_name) + 1;
3018 :
3019 : /* GUC extra value must be guc_malloc'd, not palloc'd */
3020 17 : config = (SyncStandbySlotsConfigData *) guc_malloc(LOG, size);
3021 17 : if (!config)
3022 0 : return false;
3023 :
3024 : /* Transform the data into SyncStandbySlotsConfigData */
3025 17 : config->nslotnames = list_length(elemlist);
3026 :
3027 17 : ptr = config->slot_names;
3028 51 : foreach_ptr(char, slot_name, elemlist)
3029 : {
3030 17 : strcpy(ptr, slot_name);
3031 17 : ptr += strlen(slot_name) + 1;
3032 : }
3033 :
3034 17 : *extra = config;
3035 :
3036 17 : pfree(rawname);
3037 17 : list_free(elemlist);
3038 17 : return true;
3039 : }
3040 :
3041 : /*
3042 : * GUC assign_hook for synchronized_standby_slots
3043 : */
3044 : void
3045 1251 : assign_synchronized_standby_slots(const char *newval, void *extra)
3046 : {
3047 : /*
3048 : * The standby slots may have changed, so we must recompute the oldest
3049 : * LSN.
3050 : */
3051 1251 : ss_oldest_flush_lsn = InvalidXLogRecPtr;
3052 :
3053 1251 : synchronized_standby_slots_config = (SyncStandbySlotsConfigData *) extra;
3054 1251 : }
3055 :
3056 : /*
3057 : * Check if the passed slot_name is specified in the synchronized_standby_slots GUC.
3058 : */
3059 : bool
3060 37567 : SlotExistsInSyncStandbySlots(const char *slot_name)
3061 : {
3062 : const char *standby_slot_name;
3063 :
3064 : /* Return false if there is no value in synchronized_standby_slots */
3065 37567 : if (synchronized_standby_slots_config == NULL)
3066 37554 : return false;
3067 :
3068 : /*
3069 : * XXX: We are not expecting this list to be long so a linear search
3070 : * shouldn't hurt but if that turns out not to be true then we can cache
3071 : * this information for each WalSender as well.
3072 : */
3073 13 : standby_slot_name = synchronized_standby_slots_config->slot_names;
3074 21 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
3075 : {
3076 13 : if (strcmp(standby_slot_name, slot_name) == 0)
3077 5 : return true;
3078 :
3079 8 : standby_slot_name += strlen(standby_slot_name) + 1;
3080 : }
3081 :
3082 8 : return false;
3083 : }
3084 :
3085 : /*
3086 : * Return true if the slots specified in synchronized_standby_slots have caught up to
3087 : * the given WAL location, false otherwise.
3088 : *
3089 : * The elevel parameter specifies the error level used for logging messages
3090 : * related to slots that do not exist, are invalidated, or are inactive.
3091 : */
3092 : bool
3093 690 : StandbySlotsHaveCaughtup(XLogRecPtr wait_for_lsn, int elevel)
3094 : {
3095 : const char *name;
3096 690 : int caught_up_slot_num = 0;
3097 690 : XLogRecPtr min_restart_lsn = InvalidXLogRecPtr;
3098 :
3099 : /*
3100 : * Don't need to wait for the standbys to catch up if there is no value in
3101 : * synchronized_standby_slots.
3102 : */
3103 690 : if (synchronized_standby_slots_config == NULL)
3104 667 : return true;
3105 :
3106 : /*
3107 : * Don't need to wait for the standbys to catch up if we are on a standby
3108 : * server, since we do not support syncing slots to cascading standbys.
3109 : */
3110 23 : if (RecoveryInProgress())
3111 0 : return true;
3112 :
3113 : /*
3114 : * Don't need to wait for the standbys to catch up if they are already
3115 : * beyond the specified WAL location.
3116 : */
3117 23 : if (XLogRecPtrIsValid(ss_oldest_flush_lsn) &&
3118 12 : ss_oldest_flush_lsn >= wait_for_lsn)
3119 7 : return true;
3120 :
3121 : /*
3122 : * To prevent concurrent slot dropping and creation while filtering the
3123 : * slots, take the ReplicationSlotControlLock outside of the loop.
3124 : */
3125 16 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
3126 :
3127 16 : name = synchronized_standby_slots_config->slot_names;
3128 22 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
3129 : {
3130 : XLogRecPtr restart_lsn;
3131 : bool invalidated;
3132 : bool inactive;
3133 : ReplicationSlot *slot;
3134 :
3135 16 : slot = SearchNamedReplicationSlot(name, false);
3136 :
3137 : /*
3138 : * If a slot name provided in synchronized_standby_slots does not
3139 : * exist, report a message and exit the loop.
3140 : */
3141 16 : if (!slot)
3142 : {
3143 0 : ereport(elevel,
3144 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3145 : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not exist",
3146 : name, "synchronized_standby_slots"),
3147 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3148 : name),
3149 : errhint("Create the replication slot \"%s\" or amend parameter \"%s\".",
3150 : name, "synchronized_standby_slots"));
3151 0 : break;
3152 : }
3153 :
3154 : /* Same as above: if a slot is not physical, exit the loop. */
3155 16 : if (SlotIsLogical(slot))
3156 : {
3157 0 : ereport(elevel,
3158 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3159 : errmsg("cannot specify logical replication slot \"%s\" in parameter \"%s\"",
3160 : name, "synchronized_standby_slots"),
3161 : errdetail("Logical replication is waiting for correction on replication slot \"%s\".",
3162 : name),
3163 : errhint("Remove the logical replication slot \"%s\" from parameter \"%s\".",
3164 : name, "synchronized_standby_slots"));
3165 0 : break;
3166 : }
3167 :
3168 16 : SpinLockAcquire(&slot->mutex);
3169 16 : restart_lsn = slot->data.restart_lsn;
3170 16 : invalidated = slot->data.invalidated != RS_INVAL_NONE;
3171 16 : inactive = slot->active_proc == INVALID_PROC_NUMBER;
3172 16 : SpinLockRelease(&slot->mutex);
3173 :
3174 16 : if (invalidated)
3175 : {
3176 : /* Specified physical slot has been invalidated */
3177 0 : ereport(elevel,
3178 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3179 : errmsg("physical replication slot \"%s\" specified in parameter \"%s\" has been invalidated",
3180 : name, "synchronized_standby_slots"),
3181 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3182 : name),
3183 : errhint("Drop and recreate the replication slot \"%s\", or amend parameter \"%s\".",
3184 : name, "synchronized_standby_slots"));
3185 0 : break;
3186 : }
3187 :
3188 16 : if (!XLogRecPtrIsValid(restart_lsn) || restart_lsn < wait_for_lsn)
3189 : {
3190 : /* Log a message if no active_pid for this physical slot */
3191 10 : if (inactive)
3192 9 : ereport(elevel,
3193 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3194 : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not have active_pid",
3195 : name, "synchronized_standby_slots"),
3196 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3197 : name),
3198 : errhint("Start the standby associated with the replication slot \"%s\", or amend parameter \"%s\".",
3199 : name, "synchronized_standby_slots"));
3200 :
3201 : /* Continue if the current slot hasn't caught up. */
3202 10 : break;
3203 : }
3204 :
3205 : Assert(restart_lsn >= wait_for_lsn);
3206 :
3207 6 : if (!XLogRecPtrIsValid(min_restart_lsn) ||
3208 : min_restart_lsn > restart_lsn)
3209 6 : min_restart_lsn = restart_lsn;
3210 :
3211 6 : caught_up_slot_num++;
3212 :
3213 6 : name += strlen(name) + 1;
3214 : }
3215 :
3216 16 : LWLockRelease(ReplicationSlotControlLock);
3217 :
3218 : /*
3219 : * Return false if not all the standbys have caught up to the specified
3220 : * WAL location.
3221 : */
3222 16 : if (caught_up_slot_num != synchronized_standby_slots_config->nslotnames)
3223 10 : return false;
3224 :
3225 : /* The ss_oldest_flush_lsn must not retreat. */
3226 : Assert(!XLogRecPtrIsValid(ss_oldest_flush_lsn) ||
3227 : min_restart_lsn >= ss_oldest_flush_lsn);
3228 :
3229 6 : ss_oldest_flush_lsn = min_restart_lsn;
3230 :
3231 6 : return true;
3232 : }
3233 :
3234 : /*
3235 : * Wait for physical standbys to confirm receiving the given lsn.
3236 : *
3237 : * Used by logical decoding SQL functions. It waits for physical standbys
3238 : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
3239 : */
3240 : void
3241 231 : WaitForStandbyConfirmation(XLogRecPtr wait_for_lsn)
3242 : {
3243 : /*
3244 : * Don't need to wait for the standby to catch up if the current acquired
3245 : * slot is not a logical failover slot, or there is no value in
3246 : * synchronized_standby_slots.
3247 : */
3248 231 : if (!MyReplicationSlot->data.failover || !synchronized_standby_slots_config)
3249 230 : return;
3250 :
3251 1 : ConditionVariablePrepareToSleep(&WalSndCtl->wal_confirm_rcv_cv);
3252 :
3253 : for (;;)
3254 : {
3255 2 : CHECK_FOR_INTERRUPTS();
3256 :
3257 2 : if (ConfigReloadPending)
3258 : {
3259 1 : ConfigReloadPending = false;
3260 1 : ProcessConfigFile(PGC_SIGHUP);
3261 : }
3262 :
3263 : /* Exit if done waiting for every slot. */
3264 2 : if (StandbySlotsHaveCaughtup(wait_for_lsn, WARNING))
3265 1 : break;
3266 :
3267 : /*
3268 : * Wait for the slots in the synchronized_standby_slots to catch up,
3269 : * but use a timeout (1s) so we can also check if the
3270 : * synchronized_standby_slots has been changed.
3271 : */
3272 1 : ConditionVariableTimedSleep(&WalSndCtl->wal_confirm_rcv_cv, 1000,
3273 : WAIT_EVENT_WAIT_FOR_STANDBY_CONFIRMATION);
3274 : }
3275 :
3276 1 : ConditionVariableCancelSleep();
3277 : }
|
