Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * slot.c
4 : * Replication slot management.
5 : *
6 : *
7 : * Copyright (c) 2012-2025, PostgreSQL Global Development Group
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/replication/slot.c
12 : *
13 : * NOTES
14 : *
15 : * Replication slots are used to keep state about replication streams
16 : * originating from this cluster. Their primary purpose is to prevent the
17 : * premature removal of WAL or of old tuple versions in a manner that would
18 : * interfere with replication; they are also useful for monitoring purposes.
19 : * Slots need to be permanent (to allow restarts), crash-safe, and allocatable
20 : * on standbys (to support cascading setups). The requirement that slots be
21 : * usable on standbys precludes storing them in the system catalogs.
22 : *
23 : * Each replication slot gets its own directory inside the directory
24 : * $PGDATA / PG_REPLSLOT_DIR. Inside that directory the state file will
25 : * contain the slot's own data. Additional data can be stored alongside that
26 : * file if required. While the server is running, the state data is also
27 : * cached in memory for efficiency.
28 : *
29 : * ReplicationSlotAllocationLock must be taken in exclusive mode to allocate
30 : * or free a slot. ReplicationSlotControlLock must be taken in shared mode
31 : * to iterate over the slots, and in exclusive mode to change the in_use flag
32 : * of a slot. The remaining data in each slot is protected by its mutex.
33 : *
34 : *-------------------------------------------------------------------------
35 : */
36 :
37 : #include "postgres.h"
38 :
39 : #include <unistd.h>
40 : #include <sys/stat.h>
41 :
42 : #include "access/transam.h"
43 : #include "access/xlog_internal.h"
44 : #include "access/xlogrecovery.h"
45 : #include "common/file_utils.h"
46 : #include "common/string.h"
47 : #include "miscadmin.h"
48 : #include "pgstat.h"
49 : #include "postmaster/interrupt.h"
50 : #include "replication/logicallauncher.h"
51 : #include "replication/slotsync.h"
52 : #include "replication/slot.h"
53 : #include "replication/walsender_private.h"
54 : #include "storage/fd.h"
55 : #include "storage/ipc.h"
56 : #include "storage/proc.h"
57 : #include "storage/procarray.h"
58 : #include "utils/builtins.h"
59 : #include "utils/guc_hooks.h"
60 : #include "utils/injection_point.h"
61 : #include "utils/varlena.h"
62 :
63 : /*
64 : * Replication slot on-disk data structure.
65 : */
66 : typedef struct ReplicationSlotOnDisk
67 : {
68 : /* first part of this struct needs to be version independent */
69 :
70 : /* data not covered by checksum */
71 : uint32 magic;
72 : pg_crc32c checksum;
73 :
74 : /* data covered by checksum */
75 : uint32 version;
76 : uint32 length;
77 :
78 : /*
79 : * The actual data in the slot that follows can differ based on the above
80 : * 'version'.
81 : */
82 :
83 : ReplicationSlotPersistentData slotdata;
84 : } ReplicationSlotOnDisk;
85 :
86 : /*
87 : * Struct for the configuration of synchronized_standby_slots.
88 : *
89 : * Note: this must be a flat representation that can be held in a single chunk
90 : * of guc_malloc'd memory, so that it can be stored as the "extra" data for the
91 : * synchronized_standby_slots GUC.
92 : */
93 : typedef struct
94 : {
95 : /* Number of slot names in the slot_names[] */
96 : int nslotnames;
97 :
98 : /*
99 : * slot_names contains 'nslotnames' consecutive null-terminated C strings.
100 : */
101 : char slot_names[FLEXIBLE_ARRAY_MEMBER];
102 : } SyncStandbySlotsConfigData;
103 :
104 : /*
105 : * Lookup table for slot invalidation causes.
106 : */
107 : typedef struct SlotInvalidationCauseMap
108 : {
109 : ReplicationSlotInvalidationCause cause;
110 : const char *cause_name;
111 : } SlotInvalidationCauseMap;
112 :
113 : static const SlotInvalidationCauseMap SlotInvalidationCauses[] = {
114 : {RS_INVAL_NONE, "none"},
115 : {RS_INVAL_WAL_REMOVED, "wal_removed"},
116 : {RS_INVAL_HORIZON, "rows_removed"},
117 : {RS_INVAL_WAL_LEVEL, "wal_level_insufficient"},
118 : {RS_INVAL_IDLE_TIMEOUT, "idle_timeout"},
119 : };
120 :
121 : /*
122 : * Ensure that the lookup table is up-to-date with the enums defined in
123 : * ReplicationSlotInvalidationCause.
124 : */
125 : StaticAssertDecl(lengthof(SlotInvalidationCauses) == (RS_INVAL_MAX_CAUSES + 1),
126 : "array length mismatch");
127 :
128 : /* size of version independent data */
129 : #define ReplicationSlotOnDiskConstantSize \
130 : offsetof(ReplicationSlotOnDisk, slotdata)
131 : /* size of the part of the slot not covered by the checksum */
132 : #define ReplicationSlotOnDiskNotChecksummedSize \
133 : offsetof(ReplicationSlotOnDisk, version)
134 : /* size of the part covered by the checksum */
135 : #define ReplicationSlotOnDiskChecksummedSize \
136 : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskNotChecksummedSize
137 : /* size of the slot data that is version dependent */
138 : #define ReplicationSlotOnDiskV2Size \
139 : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskConstantSize
140 :
141 : #define SLOT_MAGIC 0x1051CA1 /* format identifier */
142 : #define SLOT_VERSION 5 /* version for new files */
143 :
144 : /* Control array for replication slot management */
145 : ReplicationSlotCtlData *ReplicationSlotCtl = NULL;
146 :
147 : /* My backend's replication slot in the shared memory array */
148 : ReplicationSlot *MyReplicationSlot = NULL;
149 :
150 : /* GUC variables */
151 : int max_replication_slots = 10; /* the maximum number of replication
152 : * slots */
153 :
154 : /*
155 : * Invalidate replication slots that have remained idle longer than this
156 : * duration; '0' disables it.
157 : */
158 : int idle_replication_slot_timeout_secs = 0;
159 :
160 : /*
161 : * This GUC lists streaming replication standby server slot names that
162 : * logical WAL sender processes will wait for.
163 : */
164 : char *synchronized_standby_slots;
165 :
166 : /* This is the parsed and cached configuration for synchronized_standby_slots */
167 : static SyncStandbySlotsConfigData *synchronized_standby_slots_config;
168 :
169 : /*
170 : * Oldest LSN that has been confirmed to be flushed to the standbys
171 : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
172 : */
173 : static XLogRecPtr ss_oldest_flush_lsn = InvalidXLogRecPtr;
174 :
175 : static void ReplicationSlotShmemExit(int code, Datum arg);
176 : static bool IsSlotForConflictCheck(const char *name);
177 : static void ReplicationSlotDropPtr(ReplicationSlot *slot);
178 :
179 : /* internal persistency functions */
180 : static void RestoreSlotFromDisk(const char *name);
181 : static void CreateSlotOnDisk(ReplicationSlot *slot);
182 : static void SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel);
183 :
184 : /*
185 : * Report shared-memory space needed by ReplicationSlotsShmemInit.
186 : */
187 : Size
188 8298 : ReplicationSlotsShmemSize(void)
189 : {
190 8298 : Size size = 0;
191 :
192 8298 : if (max_replication_slots == 0)
193 4 : return size;
194 :
195 8294 : size = offsetof(ReplicationSlotCtlData, replication_slots);
196 8294 : size = add_size(size,
197 : mul_size(max_replication_slots, sizeof(ReplicationSlot)));
198 :
199 8294 : return size;
200 : }
201 :
202 : /*
203 : * Allocate and initialize shared memory for replication slots.
204 : */
205 : void
206 2152 : ReplicationSlotsShmemInit(void)
207 : {
208 : bool found;
209 :
210 2152 : if (max_replication_slots == 0)
211 2 : return;
212 :
213 2150 : ReplicationSlotCtl = (ReplicationSlotCtlData *)
214 2150 : ShmemInitStruct("ReplicationSlot Ctl", ReplicationSlotsShmemSize(),
215 : &found);
216 :
217 2150 : if (!found)
218 : {
219 : int i;
220 :
221 : /* First time through, so initialize */
222 3950 : MemSet(ReplicationSlotCtl, 0, ReplicationSlotsShmemSize());
223 :
224 23256 : for (i = 0; i < max_replication_slots; i++)
225 : {
226 21106 : ReplicationSlot *slot = &ReplicationSlotCtl->replication_slots[i];
227 :
228 : /* everything else is zeroed by the memset above */
229 21106 : SpinLockInit(&slot->mutex);
230 21106 : LWLockInitialize(&slot->io_in_progress_lock,
231 : LWTRANCHE_REPLICATION_SLOT_IO);
232 21106 : ConditionVariableInit(&slot->active_cv);
233 : }
234 : }
235 : }
236 :
237 : /*
238 : * Register the callback for replication slot cleanup and releasing.
239 : */
240 : void
241 42654 : ReplicationSlotInitialize(void)
242 : {
243 42654 : before_shmem_exit(ReplicationSlotShmemExit, 0);
244 42654 : }
245 :
246 : /*
247 : * Release and cleanup replication slots.
248 : */
249 : static void
250 42654 : ReplicationSlotShmemExit(int code, Datum arg)
251 : {
252 : /* Make sure active replication slots are released */
253 42654 : if (MyReplicationSlot != NULL)
254 498 : ReplicationSlotRelease();
255 :
256 : /* Also cleanup all the temporary slots. */
257 42654 : ReplicationSlotCleanup(false);
258 42654 : }
259 :
260 : /*
261 : * Check whether the passed slot name is valid and report errors at elevel.
262 : *
263 : * An error will be reported for a reserved replication slot name if
264 : * allow_reserved_name is set to false.
265 : *
266 : * Slot names may consist out of [a-z0-9_]{1,NAMEDATALEN-1} which should allow
267 : * the name to be used as a directory name on every supported OS.
268 : *
269 : * Returns whether the directory name is valid or not if elevel < ERROR.
270 : */
271 : bool
272 1884 : ReplicationSlotValidateName(const char *name, bool allow_reserved_name,
273 : int elevel)
274 : {
275 : const char *cp;
276 :
277 1884 : if (strlen(name) == 0)
278 : {
279 6 : ereport(elevel,
280 : (errcode(ERRCODE_INVALID_NAME),
281 : errmsg("replication slot name \"%s\" is too short",
282 : name)));
283 0 : return false;
284 : }
285 :
286 1878 : if (strlen(name) >= NAMEDATALEN)
287 : {
288 0 : ereport(elevel,
289 : (errcode(ERRCODE_NAME_TOO_LONG),
290 : errmsg("replication slot name \"%s\" is too long",
291 : name)));
292 0 : return false;
293 : }
294 :
295 37076 : for (cp = name; *cp; cp++)
296 : {
297 35200 : if (!((*cp >= 'a' && *cp <= 'z')
298 17016 : || (*cp >= '0' && *cp <= '9')
299 3408 : || (*cp == '_')))
300 : {
301 2 : ereport(elevel,
302 : (errcode(ERRCODE_INVALID_NAME),
303 : errmsg("replication slot name \"%s\" contains invalid character",
304 : name),
305 : errhint("Replication slot names may only contain lower case letters, numbers, and the underscore character.")));
306 0 : return false;
307 : }
308 : }
309 :
310 1876 : if (!allow_reserved_name && IsSlotForConflictCheck(name))
311 : {
312 0 : ereport(elevel,
313 : errcode(ERRCODE_RESERVED_NAME),
314 : errmsg("replication slot name \"%s\" is reserved",
315 : name),
316 : errdetail("The name \"%s\" is reserved for the conflict detection slot.",
317 : CONFLICT_DETECTION_SLOT));
318 :
319 0 : return false;
320 : }
321 :
322 1876 : return true;
323 : }
324 :
325 : /*
326 : * Return true if the replication slot name is "pg_conflict_detection".
327 : */
328 : static bool
329 4228 : IsSlotForConflictCheck(const char *name)
330 : {
331 4228 : return (strcmp(name, CONFLICT_DETECTION_SLOT) == 0);
332 : }
333 :
334 : /*
335 : * Create a new replication slot and mark it as used by this backend.
336 : *
337 : * name: Name of the slot
338 : * db_specific: logical decoding is db specific; if the slot is going to
339 : * be used for that pass true, otherwise false.
340 : * two_phase: Allows decoding of prepared transactions. We allow this option
341 : * to be enabled only at the slot creation time. If we allow this option
342 : * to be changed during decoding then it is quite possible that we skip
343 : * prepare first time because this option was not enabled. Now next time
344 : * during getting changes, if the two_phase option is enabled it can skip
345 : * prepare because by that time start decoding point has been moved. So the
346 : * user will only get commit prepared.
347 : * failover: If enabled, allows the slot to be synced to standbys so
348 : * that logical replication can be resumed after failover.
349 : * synced: True if the slot is synchronized from the primary server.
350 : */
351 : void
352 1288 : ReplicationSlotCreate(const char *name, bool db_specific,
353 : ReplicationSlotPersistency persistency,
354 : bool two_phase, bool failover, bool synced)
355 : {
356 1288 : ReplicationSlot *slot = NULL;
357 : int i;
358 :
359 : Assert(MyReplicationSlot == NULL);
360 :
361 : /*
362 : * The logical launcher or pg_upgrade may create or migrate an internal
363 : * slot, so using a reserved name is allowed in these cases.
364 : */
365 1288 : ReplicationSlotValidateName(name, IsBinaryUpgrade || IsLogicalLauncher(),
366 : ERROR);
367 :
368 1286 : if (failover)
369 : {
370 : /*
371 : * Do not allow users to create the failover enabled slots on the
372 : * standby as we do not support sync to the cascading standby.
373 : *
374 : * However, failover enabled slots can be created during slot
375 : * synchronization because we need to retain the same values as the
376 : * remote slot.
377 : */
378 44 : if (RecoveryInProgress() && !IsSyncingReplicationSlots())
379 0 : ereport(ERROR,
380 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
381 : errmsg("cannot enable failover for a replication slot created on the standby"));
382 :
383 : /*
384 : * Do not allow users to create failover enabled temporary slots,
385 : * because temporary slots will not be synced to the standby.
386 : *
387 : * However, failover enabled temporary slots can be created during
388 : * slot synchronization. See the comments atop slotsync.c for details.
389 : */
390 44 : if (persistency == RS_TEMPORARY && !IsSyncingReplicationSlots())
391 2 : ereport(ERROR,
392 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
393 : errmsg("cannot enable failover for a temporary replication slot"));
394 : }
395 :
396 : /*
397 : * If some other backend ran this code concurrently with us, we'd likely
398 : * both allocate the same slot, and that would be bad. We'd also be at
399 : * risk of missing a name collision. Also, we don't want to try to create
400 : * a new slot while somebody's busy cleaning up an old one, because we
401 : * might both be monkeying with the same directory.
402 : */
403 1284 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
404 :
405 : /*
406 : * Check for name collision, and identify an allocatable slot. We need to
407 : * hold ReplicationSlotControlLock in shared mode for this, so that nobody
408 : * else can change the in_use flags while we're looking at them.
409 : */
410 1284 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
411 12318 : for (i = 0; i < max_replication_slots; i++)
412 : {
413 11040 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
414 :
415 11040 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
416 6 : ereport(ERROR,
417 : (errcode(ERRCODE_DUPLICATE_OBJECT),
418 : errmsg("replication slot \"%s\" already exists", name)));
419 11034 : if (!s->in_use && slot == NULL)
420 1276 : slot = s;
421 : }
422 1278 : LWLockRelease(ReplicationSlotControlLock);
423 :
424 : /* If all slots are in use, we're out of luck. */
425 1278 : if (slot == NULL)
426 2 : ereport(ERROR,
427 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
428 : errmsg("all replication slots are in use"),
429 : errhint("Free one or increase \"max_replication_slots\".")));
430 :
431 : /*
432 : * Since this slot is not in use, nobody should be looking at any part of
433 : * it other than the in_use field unless they're trying to allocate it.
434 : * And since we hold ReplicationSlotAllocationLock, nobody except us can
435 : * be doing that. So it's safe to initialize the slot.
436 : */
437 : Assert(!slot->in_use);
438 : Assert(slot->active_pid == 0);
439 :
440 : /* first initialize persistent data */
441 1276 : memset(&slot->data, 0, sizeof(ReplicationSlotPersistentData));
442 1276 : namestrcpy(&slot->data.name, name);
443 1276 : slot->data.database = db_specific ? MyDatabaseId : InvalidOid;
444 1276 : slot->data.persistency = persistency;
445 1276 : slot->data.two_phase = two_phase;
446 1276 : slot->data.two_phase_at = InvalidXLogRecPtr;
447 1276 : slot->data.failover = failover;
448 1276 : slot->data.synced = synced;
449 :
450 : /* and then data only present in shared memory */
451 1276 : slot->just_dirtied = false;
452 1276 : slot->dirty = false;
453 1276 : slot->effective_xmin = InvalidTransactionId;
454 1276 : slot->effective_catalog_xmin = InvalidTransactionId;
455 1276 : slot->candidate_catalog_xmin = InvalidTransactionId;
456 1276 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
457 1276 : slot->candidate_restart_valid = InvalidXLogRecPtr;
458 1276 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
459 1276 : slot->last_saved_confirmed_flush = InvalidXLogRecPtr;
460 1276 : slot->last_saved_restart_lsn = InvalidXLogRecPtr;
461 1276 : slot->inactive_since = 0;
462 :
463 : /*
464 : * Create the slot on disk. We haven't actually marked the slot allocated
465 : * yet, so no special cleanup is required if this errors out.
466 : */
467 1276 : CreateSlotOnDisk(slot);
468 :
469 : /*
470 : * We need to briefly prevent any other backend from iterating over the
471 : * slots while we flip the in_use flag. We also need to set the active
472 : * flag while holding the ControlLock as otherwise a concurrent
473 : * ReplicationSlotAcquire() could acquire the slot as well.
474 : */
475 1276 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
476 :
477 1276 : slot->in_use = true;
478 :
479 : /* We can now mark the slot active, and that makes it our slot. */
480 1276 : SpinLockAcquire(&slot->mutex);
481 : Assert(slot->active_pid == 0);
482 1276 : slot->active_pid = MyProcPid;
483 1276 : SpinLockRelease(&slot->mutex);
484 1276 : MyReplicationSlot = slot;
485 :
486 1276 : LWLockRelease(ReplicationSlotControlLock);
487 :
488 : /*
489 : * Create statistics entry for the new logical slot. We don't collect any
490 : * stats for physical slots, so no need to create an entry for the same.
491 : * See ReplicationSlotDropPtr for why we need to do this before releasing
492 : * ReplicationSlotAllocationLock.
493 : */
494 1276 : if (SlotIsLogical(slot))
495 922 : pgstat_create_replslot(slot);
496 :
497 : /*
498 : * Now that the slot has been marked as in_use and active, it's safe to
499 : * let somebody else try to allocate a slot.
500 : */
501 1276 : LWLockRelease(ReplicationSlotAllocationLock);
502 :
503 : /* Let everybody know we've modified this slot */
504 1276 : ConditionVariableBroadcast(&slot->active_cv);
505 1276 : }
506 :
507 : /*
508 : * Search for the named replication slot.
509 : *
510 : * Return the replication slot if found, otherwise NULL.
511 : */
512 : ReplicationSlot *
513 3716 : SearchNamedReplicationSlot(const char *name, bool need_lock)
514 : {
515 : int i;
516 3716 : ReplicationSlot *slot = NULL;
517 :
518 3716 : if (need_lock)
519 1028 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
520 :
521 13946 : for (i = 0; i < max_replication_slots; i++)
522 : {
523 13082 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
524 :
525 13082 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
526 : {
527 2852 : slot = s;
528 2852 : break;
529 : }
530 : }
531 :
532 3716 : if (need_lock)
533 1028 : LWLockRelease(ReplicationSlotControlLock);
534 :
535 3716 : return slot;
536 : }
537 :
538 : /*
539 : * Return the index of the replication slot in
540 : * ReplicationSlotCtl->replication_slots.
541 : *
542 : * This is mainly useful to have an efficient key for storing replication slot
543 : * stats.
544 : */
545 : int
546 14724 : ReplicationSlotIndex(ReplicationSlot *slot)
547 : {
548 : Assert(slot >= ReplicationSlotCtl->replication_slots &&
549 : slot < ReplicationSlotCtl->replication_slots + max_replication_slots);
550 :
551 14724 : return slot - ReplicationSlotCtl->replication_slots;
552 : }
553 :
554 : /*
555 : * If the slot at 'index' is unused, return false. Otherwise 'name' is set to
556 : * the slot's name and true is returned.
557 : *
558 : * This likely is only useful for pgstat_replslot.c during shutdown, in other
559 : * cases there are obvious TOCTOU issues.
560 : */
561 : bool
562 158 : ReplicationSlotName(int index, Name name)
563 : {
564 : ReplicationSlot *slot;
565 : bool found;
566 :
567 158 : slot = &ReplicationSlotCtl->replication_slots[index];
568 :
569 : /*
570 : * Ensure that the slot cannot be dropped while we copy the name. Don't
571 : * need the spinlock as the name of an existing slot cannot change.
572 : */
573 158 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
574 158 : found = slot->in_use;
575 158 : if (slot->in_use)
576 158 : namestrcpy(name, NameStr(slot->data.name));
577 158 : LWLockRelease(ReplicationSlotControlLock);
578 :
579 158 : return found;
580 : }
581 :
582 : /*
583 : * Find a previously created slot and mark it as used by this process.
584 : *
585 : * An error is raised if nowait is true and the slot is currently in use. If
586 : * nowait is false, we sleep until the slot is released by the owning process.
587 : *
588 : * An error is raised if error_if_invalid is true and the slot is found to
589 : * be invalid. It should always be set to true, except when we are temporarily
590 : * acquiring the slot and don't intend to change it.
591 : */
592 : void
593 2542 : ReplicationSlotAcquire(const char *name, bool nowait, bool error_if_invalid)
594 : {
595 : ReplicationSlot *s;
596 : int active_pid;
597 :
598 : Assert(name != NULL);
599 :
600 2542 : retry:
601 : Assert(MyReplicationSlot == NULL);
602 :
603 2542 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
604 :
605 : /* Check if the slot exits with the given name. */
606 2542 : s = SearchNamedReplicationSlot(name, false);
607 2542 : if (s == NULL || !s->in_use)
608 : {
609 18 : LWLockRelease(ReplicationSlotControlLock);
610 :
611 18 : ereport(ERROR,
612 : (errcode(ERRCODE_UNDEFINED_OBJECT),
613 : errmsg("replication slot \"%s\" does not exist",
614 : name)));
615 : }
616 :
617 : /*
618 : * Do not allow users to acquire the reserved slot. This scenario may
619 : * occur if the launcher that owns the slot has terminated unexpectedly
620 : * due to an error, and a backend process attempts to reuse the slot.
621 : */
622 2524 : if (!IsLogicalLauncher() && IsSlotForConflictCheck(name))
623 0 : ereport(ERROR,
624 : errcode(ERRCODE_UNDEFINED_OBJECT),
625 : errmsg("cannot acquire replication slot \"%s\"", name),
626 : errdetail("The slot is reserved for conflict detection and can only be acquired by logical replication launcher."));
627 :
628 : /*
629 : * This is the slot we want; check if it's active under some other
630 : * process. In single user mode, we don't need this check.
631 : */
632 2524 : if (IsUnderPostmaster)
633 : {
634 : /*
635 : * Get ready to sleep on the slot in case it is active. (We may end
636 : * up not sleeping, but we don't want to do this while holding the
637 : * spinlock.)
638 : */
639 2524 : if (!nowait)
640 538 : ConditionVariablePrepareToSleep(&s->active_cv);
641 :
642 : /*
643 : * It is important to reset the inactive_since under spinlock here to
644 : * avoid race conditions with slot invalidation. See comments related
645 : * to inactive_since in InvalidatePossiblyObsoleteSlot.
646 : */
647 2524 : SpinLockAcquire(&s->mutex);
648 2524 : if (s->active_pid == 0)
649 2248 : s->active_pid = MyProcPid;
650 2524 : active_pid = s->active_pid;
651 2524 : ReplicationSlotSetInactiveSince(s, 0, false);
652 2524 : SpinLockRelease(&s->mutex);
653 : }
654 : else
655 : {
656 0 : active_pid = MyProcPid;
657 0 : ReplicationSlotSetInactiveSince(s, 0, true);
658 : }
659 2524 : LWLockRelease(ReplicationSlotControlLock);
660 :
661 : /*
662 : * If we found the slot but it's already active in another process, we
663 : * wait until the owning process signals us that it's been released, or
664 : * error out.
665 : */
666 2524 : if (active_pid != MyProcPid)
667 : {
668 0 : if (!nowait)
669 : {
670 : /* Wait here until we get signaled, and then restart */
671 0 : ConditionVariableSleep(&s->active_cv,
672 : WAIT_EVENT_REPLICATION_SLOT_DROP);
673 0 : ConditionVariableCancelSleep();
674 0 : goto retry;
675 : }
676 :
677 0 : ereport(ERROR,
678 : (errcode(ERRCODE_OBJECT_IN_USE),
679 : errmsg("replication slot \"%s\" is active for PID %d",
680 : NameStr(s->data.name), active_pid)));
681 : }
682 2524 : else if (!nowait)
683 538 : ConditionVariableCancelSleep(); /* no sleep needed after all */
684 :
685 : /* We made this slot active, so it's ours now. */
686 2524 : MyReplicationSlot = s;
687 :
688 : /*
689 : * We need to check for invalidation after making the slot ours to avoid
690 : * the possible race condition with the checkpointer that can otherwise
691 : * invalidate the slot immediately after the check.
692 : */
693 2524 : if (error_if_invalid && s->data.invalidated != RS_INVAL_NONE)
694 14 : ereport(ERROR,
695 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
696 : errmsg("can no longer access replication slot \"%s\"",
697 : NameStr(s->data.name)),
698 : errdetail("This replication slot has been invalidated due to \"%s\".",
699 : GetSlotInvalidationCauseName(s->data.invalidated)));
700 :
701 : /* Let everybody know we've modified this slot */
702 2510 : ConditionVariableBroadcast(&s->active_cv);
703 :
704 : /*
705 : * The call to pgstat_acquire_replslot() protects against stats for a
706 : * different slot, from before a restart or such, being present during
707 : * pgstat_report_replslot().
708 : */
709 2510 : if (SlotIsLogical(s))
710 2104 : pgstat_acquire_replslot(s);
711 :
712 :
713 2510 : if (am_walsender)
714 : {
715 1700 : ereport(log_replication_commands ? LOG : DEBUG1,
716 : SlotIsLogical(s)
717 : ? errmsg("acquired logical replication slot \"%s\"",
718 : NameStr(s->data.name))
719 : : errmsg("acquired physical replication slot \"%s\"",
720 : NameStr(s->data.name)));
721 : }
722 2510 : }
723 :
724 : /*
725 : * Release the replication slot that this backend considers to own.
726 : *
727 : * This or another backend can re-acquire the slot later.
728 : * Resources this slot requires will be preserved.
729 : */
730 : void
731 3034 : ReplicationSlotRelease(void)
732 : {
733 3034 : ReplicationSlot *slot = MyReplicationSlot;
734 3034 : char *slotname = NULL; /* keep compiler quiet */
735 3034 : bool is_logical = false; /* keep compiler quiet */
736 3034 : TimestampTz now = 0;
737 :
738 : Assert(slot != NULL && slot->active_pid != 0);
739 :
740 3034 : if (am_walsender)
741 : {
742 2112 : slotname = pstrdup(NameStr(slot->data.name));
743 2112 : is_logical = SlotIsLogical(slot);
744 : }
745 :
746 3034 : if (slot->data.persistency == RS_EPHEMERAL)
747 : {
748 : /*
749 : * Delete the slot. There is no !PANIC case where this is allowed to
750 : * fail, all that may happen is an incomplete cleanup of the on-disk
751 : * data.
752 : */
753 10 : ReplicationSlotDropAcquired();
754 : }
755 :
756 : /*
757 : * If slot needed to temporarily restrain both data and catalog xmin to
758 : * create the catalog snapshot, remove that temporary constraint.
759 : * Snapshots can only be exported while the initial snapshot is still
760 : * acquired.
761 : */
762 3034 : if (!TransactionIdIsValid(slot->data.xmin) &&
763 2978 : TransactionIdIsValid(slot->effective_xmin))
764 : {
765 390 : SpinLockAcquire(&slot->mutex);
766 390 : slot->effective_xmin = InvalidTransactionId;
767 390 : SpinLockRelease(&slot->mutex);
768 390 : ReplicationSlotsComputeRequiredXmin(false);
769 : }
770 :
771 : /*
772 : * Set the time since the slot has become inactive. We get the current
773 : * time beforehand to avoid system call while holding the spinlock.
774 : */
775 3034 : now = GetCurrentTimestamp();
776 :
777 3034 : if (slot->data.persistency == RS_PERSISTENT)
778 : {
779 : /*
780 : * Mark persistent slot inactive. We're not freeing it, just
781 : * disconnecting, but wake up others that may be waiting for it.
782 : */
783 2468 : SpinLockAcquire(&slot->mutex);
784 2468 : slot->active_pid = 0;
785 2468 : ReplicationSlotSetInactiveSince(slot, now, false);
786 2468 : SpinLockRelease(&slot->mutex);
787 2468 : ConditionVariableBroadcast(&slot->active_cv);
788 : }
789 : else
790 566 : ReplicationSlotSetInactiveSince(slot, now, true);
791 :
792 3034 : MyReplicationSlot = NULL;
793 :
794 : /* might not have been set when we've been a plain slot */
795 3034 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
796 3034 : MyProc->statusFlags &= ~PROC_IN_LOGICAL_DECODING;
797 3034 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
798 3034 : LWLockRelease(ProcArrayLock);
799 :
800 3034 : if (am_walsender)
801 : {
802 2112 : ereport(log_replication_commands ? LOG : DEBUG1,
803 : is_logical
804 : ? errmsg("released logical replication slot \"%s\"",
805 : slotname)
806 : : errmsg("released physical replication slot \"%s\"",
807 : slotname));
808 :
809 2112 : pfree(slotname);
810 : }
811 3034 : }
812 :
813 : /*
814 : * Cleanup temporary slots created in current session.
815 : *
816 : * Cleanup only synced temporary slots if 'synced_only' is true, else
817 : * cleanup all temporary slots.
818 : */
819 : void
820 86124 : ReplicationSlotCleanup(bool synced_only)
821 : {
822 : int i;
823 :
824 : Assert(MyReplicationSlot == NULL);
825 :
826 86404 : restart:
827 86404 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
828 936190 : for (i = 0; i < max_replication_slots; i++)
829 : {
830 850066 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
831 :
832 850066 : if (!s->in_use)
833 822718 : continue;
834 :
835 27348 : SpinLockAcquire(&s->mutex);
836 27348 : if ((s->active_pid == MyProcPid &&
837 280 : (!synced_only || s->data.synced)))
838 : {
839 : Assert(s->data.persistency == RS_TEMPORARY);
840 280 : SpinLockRelease(&s->mutex);
841 280 : LWLockRelease(ReplicationSlotControlLock); /* avoid deadlock */
842 :
843 280 : ReplicationSlotDropPtr(s);
844 :
845 280 : ConditionVariableBroadcast(&s->active_cv);
846 280 : goto restart;
847 : }
848 : else
849 27068 : SpinLockRelease(&s->mutex);
850 : }
851 :
852 86124 : LWLockRelease(ReplicationSlotControlLock);
853 86124 : }
854 :
855 : /*
856 : * Permanently drop replication slot identified by the passed in name.
857 : */
858 : void
859 804 : ReplicationSlotDrop(const char *name, bool nowait)
860 : {
861 : Assert(MyReplicationSlot == NULL);
862 :
863 804 : ReplicationSlotAcquire(name, nowait, false);
864 :
865 : /*
866 : * Do not allow users to drop the slots which are currently being synced
867 : * from the primary to the standby.
868 : */
869 790 : if (RecoveryInProgress() && MyReplicationSlot->data.synced)
870 2 : ereport(ERROR,
871 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
872 : errmsg("cannot drop replication slot \"%s\"", name),
873 : errdetail("This replication slot is being synchronized from the primary server."));
874 :
875 788 : ReplicationSlotDropAcquired();
876 788 : }
877 :
878 : /*
879 : * Change the definition of the slot identified by the specified name.
880 : */
881 : void
882 12 : ReplicationSlotAlter(const char *name, const bool *failover,
883 : const bool *two_phase)
884 : {
885 12 : bool update_slot = false;
886 :
887 : Assert(MyReplicationSlot == NULL);
888 : Assert(failover || two_phase);
889 :
890 12 : ReplicationSlotAcquire(name, false, true);
891 :
892 10 : if (SlotIsPhysical(MyReplicationSlot))
893 0 : ereport(ERROR,
894 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
895 : errmsg("cannot use %s with a physical replication slot",
896 : "ALTER_REPLICATION_SLOT"));
897 :
898 10 : if (RecoveryInProgress())
899 : {
900 : /*
901 : * Do not allow users to alter the slots which are currently being
902 : * synced from the primary to the standby.
903 : */
904 2 : if (MyReplicationSlot->data.synced)
905 2 : ereport(ERROR,
906 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
907 : errmsg("cannot alter replication slot \"%s\"", name),
908 : errdetail("This replication slot is being synchronized from the primary server."));
909 :
910 : /*
911 : * Do not allow users to enable failover on the standby as we do not
912 : * support sync to the cascading standby.
913 : */
914 0 : if (failover && *failover)
915 0 : ereport(ERROR,
916 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
917 : errmsg("cannot enable failover for a replication slot"
918 : " on the standby"));
919 : }
920 :
921 8 : if (failover)
922 : {
923 : /*
924 : * Do not allow users to enable failover for temporary slots as we do
925 : * not support syncing temporary slots to the standby.
926 : */
927 6 : if (*failover && MyReplicationSlot->data.persistency == RS_TEMPORARY)
928 0 : ereport(ERROR,
929 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
930 : errmsg("cannot enable failover for a temporary replication slot"));
931 :
932 6 : if (MyReplicationSlot->data.failover != *failover)
933 : {
934 6 : SpinLockAcquire(&MyReplicationSlot->mutex);
935 6 : MyReplicationSlot->data.failover = *failover;
936 6 : SpinLockRelease(&MyReplicationSlot->mutex);
937 :
938 6 : update_slot = true;
939 : }
940 : }
941 :
942 8 : if (two_phase && MyReplicationSlot->data.two_phase != *two_phase)
943 : {
944 2 : SpinLockAcquire(&MyReplicationSlot->mutex);
945 2 : MyReplicationSlot->data.two_phase = *two_phase;
946 2 : SpinLockRelease(&MyReplicationSlot->mutex);
947 :
948 2 : update_slot = true;
949 : }
950 :
951 8 : if (update_slot)
952 : {
953 8 : ReplicationSlotMarkDirty();
954 8 : ReplicationSlotSave();
955 : }
956 :
957 8 : ReplicationSlotRelease();
958 8 : }
959 :
960 : /*
961 : * Permanently drop the currently acquired replication slot.
962 : */
963 : void
964 814 : ReplicationSlotDropAcquired(void)
965 : {
966 814 : ReplicationSlot *slot = MyReplicationSlot;
967 :
968 : Assert(MyReplicationSlot != NULL);
969 :
970 : /* slot isn't acquired anymore */
971 814 : MyReplicationSlot = NULL;
972 :
973 814 : ReplicationSlotDropPtr(slot);
974 814 : }
975 :
976 : /*
977 : * Permanently drop the replication slot which will be released by the point
978 : * this function returns.
979 : */
980 : static void
981 1094 : ReplicationSlotDropPtr(ReplicationSlot *slot)
982 : {
983 : char path[MAXPGPATH];
984 : char tmppath[MAXPGPATH];
985 :
986 : /*
987 : * If some other backend ran this code concurrently with us, we might try
988 : * to delete a slot with a certain name while someone else was trying to
989 : * create a slot with the same name.
990 : */
991 1094 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
992 :
993 : /* Generate pathnames. */
994 1094 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
995 1094 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
996 :
997 : /*
998 : * Rename the slot directory on disk, so that we'll no longer recognize
999 : * this as a valid slot. Note that if this fails, we've got to mark the
1000 : * slot inactive before bailing out. If we're dropping an ephemeral or a
1001 : * temporary slot, we better never fail hard as the caller won't expect
1002 : * the slot to survive and this might get called during error handling.
1003 : */
1004 1094 : if (rename(path, tmppath) == 0)
1005 : {
1006 : /*
1007 : * We need to fsync() the directory we just renamed and its parent to
1008 : * make sure that our changes are on disk in a crash-safe fashion. If
1009 : * fsync() fails, we can't be sure whether the changes are on disk or
1010 : * not. For now, we handle that by panicking;
1011 : * StartupReplicationSlots() will try to straighten it out after
1012 : * restart.
1013 : */
1014 1094 : START_CRIT_SECTION();
1015 1094 : fsync_fname(tmppath, true);
1016 1094 : fsync_fname(PG_REPLSLOT_DIR, true);
1017 1094 : END_CRIT_SECTION();
1018 : }
1019 : else
1020 : {
1021 0 : bool fail_softly = slot->data.persistency != RS_PERSISTENT;
1022 :
1023 0 : SpinLockAcquire(&slot->mutex);
1024 0 : slot->active_pid = 0;
1025 0 : SpinLockRelease(&slot->mutex);
1026 :
1027 : /* wake up anyone waiting on this slot */
1028 0 : ConditionVariableBroadcast(&slot->active_cv);
1029 :
1030 0 : ereport(fail_softly ? WARNING : ERROR,
1031 : (errcode_for_file_access(),
1032 : errmsg("could not rename file \"%s\" to \"%s\": %m",
1033 : path, tmppath)));
1034 : }
1035 :
1036 : /*
1037 : * The slot is definitely gone. Lock out concurrent scans of the array
1038 : * long enough to kill it. It's OK to clear the active PID here without
1039 : * grabbing the mutex because nobody else can be scanning the array here,
1040 : * and nobody can be attached to this slot and thus access it without
1041 : * scanning the array.
1042 : *
1043 : * Also wake up processes waiting for it.
1044 : */
1045 1094 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
1046 1094 : slot->active_pid = 0;
1047 1094 : slot->in_use = false;
1048 1094 : LWLockRelease(ReplicationSlotControlLock);
1049 1094 : ConditionVariableBroadcast(&slot->active_cv);
1050 :
1051 : /*
1052 : * Slot is dead and doesn't prevent resource removal anymore, recompute
1053 : * limits.
1054 : */
1055 1094 : ReplicationSlotsComputeRequiredXmin(false);
1056 1094 : ReplicationSlotsComputeRequiredLSN();
1057 :
1058 : /*
1059 : * If removing the directory fails, the worst thing that will happen is
1060 : * that the user won't be able to create a new slot with the same name
1061 : * until the next server restart. We warn about it, but that's all.
1062 : */
1063 1094 : if (!rmtree(tmppath, true))
1064 0 : ereport(WARNING,
1065 : (errmsg("could not remove directory \"%s\"", tmppath)));
1066 :
1067 : /*
1068 : * Drop the statistics entry for the replication slot. Do this while
1069 : * holding ReplicationSlotAllocationLock so that we don't drop a
1070 : * statistics entry for another slot with the same name just created in
1071 : * another session.
1072 : */
1073 1094 : if (SlotIsLogical(slot))
1074 796 : pgstat_drop_replslot(slot);
1075 :
1076 : /*
1077 : * We release this at the very end, so that nobody starts trying to create
1078 : * a slot while we're still cleaning up the detritus of the old one.
1079 : */
1080 1094 : LWLockRelease(ReplicationSlotAllocationLock);
1081 1094 : }
1082 :
1083 : /*
1084 : * Serialize the currently acquired slot's state from memory to disk, thereby
1085 : * guaranteeing the current state will survive a crash.
1086 : */
1087 : void
1088 2672 : ReplicationSlotSave(void)
1089 : {
1090 : char path[MAXPGPATH];
1091 :
1092 : Assert(MyReplicationSlot != NULL);
1093 :
1094 2672 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(MyReplicationSlot->data.name));
1095 2672 : SaveSlotToPath(MyReplicationSlot, path, ERROR);
1096 2672 : }
1097 :
1098 : /*
1099 : * Signal that it would be useful if the currently acquired slot would be
1100 : * flushed out to disk.
1101 : *
1102 : * Note that the actual flush to disk can be delayed for a long time, if
1103 : * required for correctness explicitly do a ReplicationSlotSave().
1104 : */
1105 : void
1106 54430 : ReplicationSlotMarkDirty(void)
1107 : {
1108 54430 : ReplicationSlot *slot = MyReplicationSlot;
1109 :
1110 : Assert(MyReplicationSlot != NULL);
1111 :
1112 54430 : SpinLockAcquire(&slot->mutex);
1113 54430 : MyReplicationSlot->just_dirtied = true;
1114 54430 : MyReplicationSlot->dirty = true;
1115 54430 : SpinLockRelease(&slot->mutex);
1116 54430 : }
1117 :
1118 : /*
1119 : * Convert a slot that's marked as RS_EPHEMERAL or RS_TEMPORARY to a
1120 : * RS_PERSISTENT slot, guaranteeing it will be there after an eventual crash.
1121 : */
1122 : void
1123 896 : ReplicationSlotPersist(void)
1124 : {
1125 896 : ReplicationSlot *slot = MyReplicationSlot;
1126 :
1127 : Assert(slot != NULL);
1128 : Assert(slot->data.persistency != RS_PERSISTENT);
1129 :
1130 896 : SpinLockAcquire(&slot->mutex);
1131 896 : slot->data.persistency = RS_PERSISTENT;
1132 896 : SpinLockRelease(&slot->mutex);
1133 :
1134 896 : ReplicationSlotMarkDirty();
1135 896 : ReplicationSlotSave();
1136 896 : }
1137 :
1138 : /*
1139 : * Compute the oldest xmin across all slots and store it in the ProcArray.
1140 : *
1141 : * If already_locked is true, ProcArrayLock has already been acquired
1142 : * exclusively.
1143 : */
1144 : void
1145 4602 : ReplicationSlotsComputeRequiredXmin(bool already_locked)
1146 : {
1147 : int i;
1148 4602 : TransactionId agg_xmin = InvalidTransactionId;
1149 4602 : TransactionId agg_catalog_xmin = InvalidTransactionId;
1150 :
1151 : Assert(ReplicationSlotCtl != NULL);
1152 :
1153 4602 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1154 :
1155 46492 : for (i = 0; i < max_replication_slots; i++)
1156 : {
1157 41890 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1158 : TransactionId effective_xmin;
1159 : TransactionId effective_catalog_xmin;
1160 : bool invalidated;
1161 :
1162 41890 : if (!s->in_use)
1163 37614 : continue;
1164 :
1165 4276 : SpinLockAcquire(&s->mutex);
1166 4276 : effective_xmin = s->effective_xmin;
1167 4276 : effective_catalog_xmin = s->effective_catalog_xmin;
1168 4276 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1169 4276 : SpinLockRelease(&s->mutex);
1170 :
1171 : /* invalidated slots need not apply */
1172 4276 : if (invalidated)
1173 44 : continue;
1174 :
1175 : /* check the data xmin */
1176 4232 : if (TransactionIdIsValid(effective_xmin) &&
1177 14 : (!TransactionIdIsValid(agg_xmin) ||
1178 14 : TransactionIdPrecedes(effective_xmin, agg_xmin)))
1179 588 : agg_xmin = effective_xmin;
1180 :
1181 : /* check the catalog xmin */
1182 4232 : if (TransactionIdIsValid(effective_catalog_xmin) &&
1183 1776 : (!TransactionIdIsValid(agg_catalog_xmin) ||
1184 1776 : TransactionIdPrecedes(effective_catalog_xmin, agg_catalog_xmin)))
1185 2244 : agg_catalog_xmin = effective_catalog_xmin;
1186 : }
1187 :
1188 4602 : LWLockRelease(ReplicationSlotControlLock);
1189 :
1190 4602 : ProcArraySetReplicationSlotXmin(agg_xmin, agg_catalog_xmin, already_locked);
1191 4602 : }
1192 :
1193 : /*
1194 : * Compute the oldest restart LSN across all slots and inform xlog module.
1195 : *
1196 : * Note: while max_slot_wal_keep_size is theoretically relevant for this
1197 : * purpose, we don't try to account for that, because this module doesn't
1198 : * know what to compare against.
1199 : */
1200 : void
1201 55742 : ReplicationSlotsComputeRequiredLSN(void)
1202 : {
1203 : int i;
1204 55742 : XLogRecPtr min_required = InvalidXLogRecPtr;
1205 :
1206 : Assert(ReplicationSlotCtl != NULL);
1207 :
1208 55742 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1209 604060 : for (i = 0; i < max_replication_slots; i++)
1210 : {
1211 548318 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1212 : XLogRecPtr restart_lsn;
1213 : XLogRecPtr last_saved_restart_lsn;
1214 : bool invalidated;
1215 : ReplicationSlotPersistency persistency;
1216 :
1217 548318 : if (!s->in_use)
1218 492194 : continue;
1219 :
1220 56124 : SpinLockAcquire(&s->mutex);
1221 56124 : persistency = s->data.persistency;
1222 56124 : restart_lsn = s->data.restart_lsn;
1223 56124 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1224 56124 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1225 56124 : SpinLockRelease(&s->mutex);
1226 :
1227 : /* invalidated slots need not apply */
1228 56124 : if (invalidated)
1229 46 : continue;
1230 :
1231 : /*
1232 : * For persistent slot use last_saved_restart_lsn to compute the
1233 : * oldest LSN for removal of WAL segments. The segments between
1234 : * last_saved_restart_lsn and restart_lsn might be needed by a
1235 : * persistent slot in the case of database crash. Non-persistent
1236 : * slots can't survive the database crash, so we don't care about
1237 : * last_saved_restart_lsn for them.
1238 : */
1239 56078 : if (persistency == RS_PERSISTENT)
1240 : {
1241 54440 : if (last_saved_restart_lsn != InvalidXLogRecPtr &&
1242 : restart_lsn > last_saved_restart_lsn)
1243 : {
1244 49756 : restart_lsn = last_saved_restart_lsn;
1245 : }
1246 : }
1247 :
1248 56078 : if (restart_lsn != InvalidXLogRecPtr &&
1249 2326 : (min_required == InvalidXLogRecPtr ||
1250 : restart_lsn < min_required))
1251 53884 : min_required = restart_lsn;
1252 : }
1253 55742 : LWLockRelease(ReplicationSlotControlLock);
1254 :
1255 55742 : XLogSetReplicationSlotMinimumLSN(min_required);
1256 55742 : }
1257 :
1258 : /*
1259 : * Compute the oldest WAL LSN required by *logical* decoding slots..
1260 : *
1261 : * Returns InvalidXLogRecPtr if logical decoding is disabled or no logical
1262 : * slots exist.
1263 : *
1264 : * NB: this returns a value >= ReplicationSlotsComputeRequiredLSN(), since it
1265 : * ignores physical replication slots.
1266 : *
1267 : * The results aren't required frequently, so we don't maintain a precomputed
1268 : * value like we do for ComputeRequiredLSN() and ComputeRequiredXmin().
1269 : */
1270 : XLogRecPtr
1271 6776 : ReplicationSlotsComputeLogicalRestartLSN(void)
1272 : {
1273 6776 : XLogRecPtr result = InvalidXLogRecPtr;
1274 : int i;
1275 :
1276 6776 : if (max_replication_slots <= 0)
1277 4 : return InvalidXLogRecPtr;
1278 :
1279 6772 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1280 :
1281 73400 : for (i = 0; i < max_replication_slots; i++)
1282 : {
1283 : ReplicationSlot *s;
1284 : XLogRecPtr restart_lsn;
1285 : XLogRecPtr last_saved_restart_lsn;
1286 : bool invalidated;
1287 : ReplicationSlotPersistency persistency;
1288 :
1289 66628 : s = &ReplicationSlotCtl->replication_slots[i];
1290 :
1291 : /* cannot change while ReplicationSlotCtlLock is held */
1292 66628 : if (!s->in_use)
1293 65276 : continue;
1294 :
1295 : /* we're only interested in logical slots */
1296 1352 : if (!SlotIsLogical(s))
1297 1000 : continue;
1298 :
1299 : /* read once, it's ok if it increases while we're checking */
1300 352 : SpinLockAcquire(&s->mutex);
1301 352 : persistency = s->data.persistency;
1302 352 : restart_lsn = s->data.restart_lsn;
1303 352 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1304 352 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1305 352 : SpinLockRelease(&s->mutex);
1306 :
1307 : /* invalidated slots need not apply */
1308 352 : if (invalidated)
1309 8 : continue;
1310 :
1311 : /*
1312 : * For persistent slot use last_saved_restart_lsn to compute the
1313 : * oldest LSN for removal of WAL segments. The segments between
1314 : * last_saved_restart_lsn and restart_lsn might be needed by a
1315 : * persistent slot in the case of database crash. Non-persistent
1316 : * slots can't survive the database crash, so we don't care about
1317 : * last_saved_restart_lsn for them.
1318 : */
1319 344 : if (persistency == RS_PERSISTENT)
1320 : {
1321 340 : if (last_saved_restart_lsn != InvalidXLogRecPtr &&
1322 : restart_lsn > last_saved_restart_lsn)
1323 : {
1324 0 : restart_lsn = last_saved_restart_lsn;
1325 : }
1326 : }
1327 :
1328 344 : if (restart_lsn == InvalidXLogRecPtr)
1329 0 : continue;
1330 :
1331 344 : if (result == InvalidXLogRecPtr ||
1332 : restart_lsn < result)
1333 276 : result = restart_lsn;
1334 : }
1335 :
1336 6772 : LWLockRelease(ReplicationSlotControlLock);
1337 :
1338 6772 : return result;
1339 : }
1340 :
1341 : /*
1342 : * ReplicationSlotsCountDBSlots -- count the number of slots that refer to the
1343 : * passed database oid.
1344 : *
1345 : * Returns true if there are any slots referencing the database. *nslots will
1346 : * be set to the absolute number of slots in the database, *nactive to ones
1347 : * currently active.
1348 : */
1349 : bool
1350 90 : ReplicationSlotsCountDBSlots(Oid dboid, int *nslots, int *nactive)
1351 : {
1352 : int i;
1353 :
1354 90 : *nslots = *nactive = 0;
1355 :
1356 90 : if (max_replication_slots <= 0)
1357 0 : return false;
1358 :
1359 90 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1360 932 : for (i = 0; i < max_replication_slots; i++)
1361 : {
1362 : ReplicationSlot *s;
1363 :
1364 842 : s = &ReplicationSlotCtl->replication_slots[i];
1365 :
1366 : /* cannot change while ReplicationSlotCtlLock is held */
1367 842 : if (!s->in_use)
1368 806 : continue;
1369 :
1370 : /* only logical slots are database specific, skip */
1371 36 : if (!SlotIsLogical(s))
1372 20 : continue;
1373 :
1374 : /* not our database, skip */
1375 16 : if (s->data.database != dboid)
1376 10 : continue;
1377 :
1378 : /* NB: intentionally counting invalidated slots */
1379 :
1380 : /* count slots with spinlock held */
1381 6 : SpinLockAcquire(&s->mutex);
1382 6 : (*nslots)++;
1383 6 : if (s->active_pid != 0)
1384 2 : (*nactive)++;
1385 6 : SpinLockRelease(&s->mutex);
1386 : }
1387 90 : LWLockRelease(ReplicationSlotControlLock);
1388 :
1389 90 : if (*nslots > 0)
1390 6 : return true;
1391 84 : return false;
1392 : }
1393 :
1394 : /*
1395 : * ReplicationSlotsDropDBSlots -- Drop all db-specific slots relating to the
1396 : * passed database oid. The caller should hold an exclusive lock on the
1397 : * pg_database oid for the database to prevent creation of new slots on the db
1398 : * or replay from existing slots.
1399 : *
1400 : * Another session that concurrently acquires an existing slot on the target DB
1401 : * (most likely to drop it) may cause this function to ERROR. If that happens
1402 : * it may have dropped some but not all slots.
1403 : *
1404 : * This routine isn't as efficient as it could be - but we don't drop
1405 : * databases often, especially databases with lots of slots.
1406 : */
1407 : void
1408 114 : ReplicationSlotsDropDBSlots(Oid dboid)
1409 : {
1410 : int i;
1411 :
1412 114 : if (max_replication_slots <= 0)
1413 0 : return;
1414 :
1415 114 : restart:
1416 124 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1417 1190 : for (i = 0; i < max_replication_slots; i++)
1418 : {
1419 : ReplicationSlot *s;
1420 : char *slotname;
1421 : int active_pid;
1422 :
1423 1076 : s = &ReplicationSlotCtl->replication_slots[i];
1424 :
1425 : /* cannot change while ReplicationSlotCtlLock is held */
1426 1076 : if (!s->in_use)
1427 1022 : continue;
1428 :
1429 : /* only logical slots are database specific, skip */
1430 54 : if (!SlotIsLogical(s))
1431 22 : continue;
1432 :
1433 : /* not our database, skip */
1434 32 : if (s->data.database != dboid)
1435 22 : continue;
1436 :
1437 : /* NB: intentionally including invalidated slots */
1438 :
1439 : /* acquire slot, so ReplicationSlotDropAcquired can be reused */
1440 10 : SpinLockAcquire(&s->mutex);
1441 : /* can't change while ReplicationSlotControlLock is held */
1442 10 : slotname = NameStr(s->data.name);
1443 10 : active_pid = s->active_pid;
1444 10 : if (active_pid == 0)
1445 : {
1446 10 : MyReplicationSlot = s;
1447 10 : s->active_pid = MyProcPid;
1448 : }
1449 10 : SpinLockRelease(&s->mutex);
1450 :
1451 : /*
1452 : * Even though we hold an exclusive lock on the database object a
1453 : * logical slot for that DB can still be active, e.g. if it's
1454 : * concurrently being dropped by a backend connected to another DB.
1455 : *
1456 : * That's fairly unlikely in practice, so we'll just bail out.
1457 : *
1458 : * The slot sync worker holds a shared lock on the database before
1459 : * operating on synced logical slots to avoid conflict with the drop
1460 : * happening here. The persistent synced slots are thus safe but there
1461 : * is a possibility that the slot sync worker has created a temporary
1462 : * slot (which stays active even on release) and we are trying to drop
1463 : * that here. In practice, the chances of hitting this scenario are
1464 : * less as during slot synchronization, the temporary slot is
1465 : * immediately converted to persistent and thus is safe due to the
1466 : * shared lock taken on the database. So, we'll just bail out in such
1467 : * a case.
1468 : *
1469 : * XXX: We can consider shutting down the slot sync worker before
1470 : * trying to drop synced temporary slots here.
1471 : */
1472 10 : if (active_pid)
1473 0 : ereport(ERROR,
1474 : (errcode(ERRCODE_OBJECT_IN_USE),
1475 : errmsg("replication slot \"%s\" is active for PID %d",
1476 : slotname, active_pid)));
1477 :
1478 : /*
1479 : * To avoid duplicating ReplicationSlotDropAcquired() and to avoid
1480 : * holding ReplicationSlotControlLock over filesystem operations,
1481 : * release ReplicationSlotControlLock and use
1482 : * ReplicationSlotDropAcquired.
1483 : *
1484 : * As that means the set of slots could change, restart scan from the
1485 : * beginning each time we release the lock.
1486 : */
1487 10 : LWLockRelease(ReplicationSlotControlLock);
1488 10 : ReplicationSlotDropAcquired();
1489 10 : goto restart;
1490 : }
1491 114 : LWLockRelease(ReplicationSlotControlLock);
1492 : }
1493 :
1494 :
1495 : /*
1496 : * Check whether the server's configuration supports using replication
1497 : * slots.
1498 : */
1499 : void
1500 3422 : CheckSlotRequirements(void)
1501 : {
1502 : /*
1503 : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
1504 : * needs the same check.
1505 : */
1506 :
1507 3422 : if (max_replication_slots == 0)
1508 0 : ereport(ERROR,
1509 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1510 : errmsg("replication slots can only be used if \"max_replication_slots\" > 0")));
1511 :
1512 3422 : if (wal_level < WAL_LEVEL_REPLICA)
1513 0 : ereport(ERROR,
1514 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1515 : errmsg("replication slots can only be used if \"wal_level\" >= \"replica\"")));
1516 3422 : }
1517 :
1518 : /*
1519 : * Check whether the user has privilege to use replication slots.
1520 : */
1521 : void
1522 1096 : CheckSlotPermissions(void)
1523 : {
1524 1096 : if (!has_rolreplication(GetUserId()))
1525 10 : ereport(ERROR,
1526 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1527 : errmsg("permission denied to use replication slots"),
1528 : errdetail("Only roles with the %s attribute may use replication slots.",
1529 : "REPLICATION")));
1530 1086 : }
1531 :
1532 : /*
1533 : * Reserve WAL for the currently active slot.
1534 : *
1535 : * Compute and set restart_lsn in a manner that's appropriate for the type of
1536 : * the slot and concurrency safe.
1537 : */
1538 : void
1539 1194 : ReplicationSlotReserveWal(void)
1540 : {
1541 1194 : ReplicationSlot *slot = MyReplicationSlot;
1542 :
1543 : Assert(slot != NULL);
1544 : Assert(slot->data.restart_lsn == InvalidXLogRecPtr);
1545 : Assert(slot->last_saved_restart_lsn == InvalidXLogRecPtr);
1546 :
1547 : /*
1548 : * The replication slot mechanism is used to prevent removal of required
1549 : * WAL. As there is no interlock between this routine and checkpoints, WAL
1550 : * segments could concurrently be removed when a now stale return value of
1551 : * ReplicationSlotsComputeRequiredLSN() is used. In the unlikely case that
1552 : * this happens we'll just retry.
1553 : */
1554 : while (true)
1555 0 : {
1556 : XLogSegNo segno;
1557 : XLogRecPtr restart_lsn;
1558 :
1559 : /*
1560 : * For logical slots log a standby snapshot and start logical decoding
1561 : * at exactly that position. That allows the slot to start up more
1562 : * quickly. But on a standby we cannot do WAL writes, so just use the
1563 : * replay pointer; effectively, an attempt to create a logical slot on
1564 : * standby will cause it to wait for an xl_running_xact record to be
1565 : * logged independently on the primary, so that a snapshot can be
1566 : * built using the record.
1567 : *
1568 : * None of this is needed (or indeed helpful) for physical slots as
1569 : * they'll start replay at the last logged checkpoint anyway. Instead
1570 : * return the location of the last redo LSN. While that slightly
1571 : * increases the chance that we have to retry, it's where a base
1572 : * backup has to start replay at.
1573 : */
1574 1194 : if (SlotIsPhysical(slot))
1575 298 : restart_lsn = GetRedoRecPtr();
1576 896 : else if (RecoveryInProgress())
1577 46 : restart_lsn = GetXLogReplayRecPtr(NULL);
1578 : else
1579 850 : restart_lsn = GetXLogInsertRecPtr();
1580 :
1581 1194 : SpinLockAcquire(&slot->mutex);
1582 1194 : slot->data.restart_lsn = restart_lsn;
1583 1194 : SpinLockRelease(&slot->mutex);
1584 :
1585 : /* prevent WAL removal as fast as possible */
1586 1194 : ReplicationSlotsComputeRequiredLSN();
1587 :
1588 : /*
1589 : * If all required WAL is still there, great, otherwise retry. The
1590 : * slot should prevent further removal of WAL, unless there's a
1591 : * concurrent ReplicationSlotsComputeRequiredLSN() after we've written
1592 : * the new restart_lsn above, so normally we should never need to loop
1593 : * more than twice.
1594 : */
1595 1194 : XLByteToSeg(slot->data.restart_lsn, segno, wal_segment_size);
1596 1194 : if (XLogGetLastRemovedSegno() < segno)
1597 1194 : break;
1598 : }
1599 :
1600 1194 : if (!RecoveryInProgress() && SlotIsLogical(slot))
1601 : {
1602 : XLogRecPtr flushptr;
1603 :
1604 : /* make sure we have enough information to start */
1605 850 : flushptr = LogStandbySnapshot();
1606 :
1607 : /* and make sure it's fsynced to disk */
1608 850 : XLogFlush(flushptr);
1609 : }
1610 1194 : }
1611 :
1612 : /*
1613 : * Report that replication slot needs to be invalidated
1614 : */
1615 : static void
1616 42 : ReportSlotInvalidation(ReplicationSlotInvalidationCause cause,
1617 : bool terminating,
1618 : int pid,
1619 : NameData slotname,
1620 : XLogRecPtr restart_lsn,
1621 : XLogRecPtr oldestLSN,
1622 : TransactionId snapshotConflictHorizon,
1623 : long slot_idle_seconds)
1624 : {
1625 : StringInfoData err_detail;
1626 : StringInfoData err_hint;
1627 :
1628 42 : initStringInfo(&err_detail);
1629 42 : initStringInfo(&err_hint);
1630 :
1631 42 : switch (cause)
1632 : {
1633 12 : case RS_INVAL_WAL_REMOVED:
1634 : {
1635 12 : uint64 ex = oldestLSN - restart_lsn;
1636 :
1637 12 : appendStringInfo(&err_detail,
1638 12 : ngettext("The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " byte.",
1639 : "The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " bytes.",
1640 : ex),
1641 12 : LSN_FORMAT_ARGS(restart_lsn),
1642 : ex);
1643 : /* translator: %s is a GUC variable name */
1644 12 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1645 : "max_slot_wal_keep_size");
1646 12 : break;
1647 : }
1648 24 : case RS_INVAL_HORIZON:
1649 24 : appendStringInfo(&err_detail, _("The slot conflicted with xid horizon %u."),
1650 : snapshotConflictHorizon);
1651 24 : break;
1652 :
1653 6 : case RS_INVAL_WAL_LEVEL:
1654 6 : appendStringInfoString(&err_detail, _("Logical decoding on standby requires \"wal_level\" >= \"logical\" on the primary server."));
1655 6 : break;
1656 :
1657 0 : case RS_INVAL_IDLE_TIMEOUT:
1658 : {
1659 : /* translator: %s is a GUC variable name */
1660 0 : appendStringInfo(&err_detail, _("The slot's idle time of %lds exceeds the configured \"%s\" duration of %ds."),
1661 : slot_idle_seconds, "idle_replication_slot_timeout",
1662 : idle_replication_slot_timeout_secs);
1663 : /* translator: %s is a GUC variable name */
1664 0 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1665 : "idle_replication_slot_timeout");
1666 0 : break;
1667 : }
1668 : case RS_INVAL_NONE:
1669 : pg_unreachable();
1670 : }
1671 :
1672 42 : ereport(LOG,
1673 : terminating ?
1674 : errmsg("terminating process %d to release replication slot \"%s\"",
1675 : pid, NameStr(slotname)) :
1676 : errmsg("invalidating obsolete replication slot \"%s\"",
1677 : NameStr(slotname)),
1678 : errdetail_internal("%s", err_detail.data),
1679 : err_hint.len ? errhint("%s", err_hint.data) : 0);
1680 :
1681 42 : pfree(err_detail.data);
1682 42 : pfree(err_hint.data);
1683 42 : }
1684 :
1685 : /*
1686 : * Can we invalidate an idle replication slot?
1687 : *
1688 : * Idle timeout invalidation is allowed only when:
1689 : *
1690 : * 1. Idle timeout is set
1691 : * 2. Slot has reserved WAL
1692 : * 3. Slot is inactive
1693 : * 4. The slot is not being synced from the primary while the server is in
1694 : * recovery. This is because synced slots are always considered to be
1695 : * inactive because they don't perform logical decoding to produce changes.
1696 : */
1697 : static inline bool
1698 642 : CanInvalidateIdleSlot(ReplicationSlot *s)
1699 : {
1700 642 : return (idle_replication_slot_timeout_secs != 0 &&
1701 0 : !XLogRecPtrIsInvalid(s->data.restart_lsn) &&
1702 642 : s->inactive_since > 0 &&
1703 0 : !(RecoveryInProgress() && s->data.synced));
1704 : }
1705 :
1706 : /*
1707 : * DetermineSlotInvalidationCause - Determine the cause for which a slot
1708 : * becomes invalid among the given possible causes.
1709 : *
1710 : * This function sequentially checks all possible invalidation causes and
1711 : * returns the first one for which the slot is eligible for invalidation.
1712 : */
1713 : static ReplicationSlotInvalidationCause
1714 704 : DetermineSlotInvalidationCause(uint32 possible_causes, ReplicationSlot *s,
1715 : XLogRecPtr oldestLSN, Oid dboid,
1716 : TransactionId snapshotConflictHorizon,
1717 : TransactionId initial_effective_xmin,
1718 : TransactionId initial_catalog_effective_xmin,
1719 : XLogRecPtr initial_restart_lsn,
1720 : TimestampTz *inactive_since, TimestampTz now)
1721 : {
1722 : Assert(possible_causes != RS_INVAL_NONE);
1723 :
1724 704 : if (possible_causes & RS_INVAL_WAL_REMOVED)
1725 : {
1726 654 : if (initial_restart_lsn != InvalidXLogRecPtr &&
1727 : initial_restart_lsn < oldestLSN)
1728 12 : return RS_INVAL_WAL_REMOVED;
1729 : }
1730 :
1731 692 : if (possible_causes & RS_INVAL_HORIZON)
1732 : {
1733 : /* invalid DB oid signals a shared relation */
1734 44 : if (SlotIsLogical(s) &&
1735 34 : (dboid == InvalidOid || dboid == s->data.database))
1736 : {
1737 44 : if (TransactionIdIsValid(initial_effective_xmin) &&
1738 0 : TransactionIdPrecedesOrEquals(initial_effective_xmin,
1739 : snapshotConflictHorizon))
1740 0 : return RS_INVAL_HORIZON;
1741 88 : else if (TransactionIdIsValid(initial_catalog_effective_xmin) &&
1742 44 : TransactionIdPrecedesOrEquals(initial_catalog_effective_xmin,
1743 : snapshotConflictHorizon))
1744 24 : return RS_INVAL_HORIZON;
1745 : }
1746 : }
1747 :
1748 668 : if (possible_causes & RS_INVAL_WAL_LEVEL)
1749 : {
1750 6 : if (SlotIsLogical(s))
1751 6 : return RS_INVAL_WAL_LEVEL;
1752 : }
1753 :
1754 662 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1755 : {
1756 : Assert(now > 0);
1757 :
1758 642 : if (CanInvalidateIdleSlot(s))
1759 : {
1760 : /*
1761 : * Simulate the invalidation due to idle_timeout to test the
1762 : * timeout behavior promptly, without waiting for it to trigger
1763 : * naturally.
1764 : */
1765 : #ifdef USE_INJECTION_POINTS
1766 0 : if (IS_INJECTION_POINT_ATTACHED("slot-timeout-inval"))
1767 : {
1768 0 : *inactive_since = 0; /* since the beginning of time */
1769 0 : return RS_INVAL_IDLE_TIMEOUT;
1770 : }
1771 : #endif
1772 :
1773 : /*
1774 : * Check if the slot needs to be invalidated due to
1775 : * idle_replication_slot_timeout GUC.
1776 : */
1777 0 : if (TimestampDifferenceExceedsSeconds(s->inactive_since, now,
1778 : idle_replication_slot_timeout_secs))
1779 : {
1780 0 : *inactive_since = s->inactive_since;
1781 0 : return RS_INVAL_IDLE_TIMEOUT;
1782 : }
1783 : }
1784 : }
1785 :
1786 662 : return RS_INVAL_NONE;
1787 : }
1788 :
1789 : /*
1790 : * Helper for InvalidateObsoleteReplicationSlots
1791 : *
1792 : * Acquires the given slot and mark it invalid, if necessary and possible.
1793 : *
1794 : * Returns whether ReplicationSlotControlLock was released in the interim (and
1795 : * in that case we're not holding the lock at return, otherwise we are).
1796 : *
1797 : * Sets *invalidated true if the slot was invalidated. (Untouched otherwise.)
1798 : *
1799 : * This is inherently racy, because we release the LWLock
1800 : * for syscalls, so caller must restart if we return true.
1801 : */
1802 : static bool
1803 768 : InvalidatePossiblyObsoleteSlot(uint32 possible_causes,
1804 : ReplicationSlot *s,
1805 : XLogRecPtr oldestLSN,
1806 : Oid dboid, TransactionId snapshotConflictHorizon,
1807 : bool *invalidated)
1808 : {
1809 768 : int last_signaled_pid = 0;
1810 768 : bool released_lock = false;
1811 768 : bool terminated = false;
1812 768 : TransactionId initial_effective_xmin = InvalidTransactionId;
1813 768 : TransactionId initial_catalog_effective_xmin = InvalidTransactionId;
1814 768 : XLogRecPtr initial_restart_lsn = InvalidXLogRecPtr;
1815 768 : ReplicationSlotInvalidationCause invalidation_cause_prev PG_USED_FOR_ASSERTS_ONLY = RS_INVAL_NONE;
1816 768 : TimestampTz inactive_since = 0;
1817 :
1818 : for (;;)
1819 14 : {
1820 : XLogRecPtr restart_lsn;
1821 : NameData slotname;
1822 782 : int active_pid = 0;
1823 782 : ReplicationSlotInvalidationCause invalidation_cause = RS_INVAL_NONE;
1824 782 : TimestampTz now = 0;
1825 782 : long slot_idle_secs = 0;
1826 :
1827 : Assert(LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_SHARED));
1828 :
1829 782 : if (!s->in_use)
1830 : {
1831 0 : if (released_lock)
1832 0 : LWLockRelease(ReplicationSlotControlLock);
1833 0 : break;
1834 : }
1835 :
1836 782 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1837 : {
1838 : /*
1839 : * Assign the current time here to avoid system call overhead
1840 : * while holding the spinlock in subsequent code.
1841 : */
1842 674 : now = GetCurrentTimestamp();
1843 : }
1844 :
1845 : /*
1846 : * Check if the slot needs to be invalidated. If it needs to be
1847 : * invalidated, and is not currently acquired, acquire it and mark it
1848 : * as having been invalidated. We do this with the spinlock held to
1849 : * avoid race conditions -- for example the restart_lsn could move
1850 : * forward, or the slot could be dropped.
1851 : */
1852 782 : SpinLockAcquire(&s->mutex);
1853 :
1854 782 : restart_lsn = s->data.restart_lsn;
1855 :
1856 : /* we do nothing if the slot is already invalid */
1857 782 : if (s->data.invalidated == RS_INVAL_NONE)
1858 : {
1859 : /*
1860 : * The slot's mutex will be released soon, and it is possible that
1861 : * those values change since the process holding the slot has been
1862 : * terminated (if any), so record them here to ensure that we
1863 : * would report the correct invalidation cause.
1864 : *
1865 : * Unlike other slot attributes, slot's inactive_since can't be
1866 : * changed until the acquired slot is released or the owning
1867 : * process is terminated. So, the inactive slot can only be
1868 : * invalidated immediately without being terminated.
1869 : */
1870 704 : if (!terminated)
1871 : {
1872 690 : initial_restart_lsn = s->data.restart_lsn;
1873 690 : initial_effective_xmin = s->effective_xmin;
1874 690 : initial_catalog_effective_xmin = s->effective_catalog_xmin;
1875 : }
1876 :
1877 704 : invalidation_cause = DetermineSlotInvalidationCause(possible_causes,
1878 : s, oldestLSN,
1879 : dboid,
1880 : snapshotConflictHorizon,
1881 : initial_effective_xmin,
1882 : initial_catalog_effective_xmin,
1883 : initial_restart_lsn,
1884 : &inactive_since,
1885 : now);
1886 : }
1887 :
1888 : /*
1889 : * The invalidation cause recorded previously should not change while
1890 : * the process owning the slot (if any) has been terminated.
1891 : */
1892 : Assert(!(invalidation_cause_prev != RS_INVAL_NONE && terminated &&
1893 : invalidation_cause_prev != invalidation_cause));
1894 :
1895 : /* if there's no invalidation, we're done */
1896 782 : if (invalidation_cause == RS_INVAL_NONE)
1897 : {
1898 740 : SpinLockRelease(&s->mutex);
1899 740 : if (released_lock)
1900 0 : LWLockRelease(ReplicationSlotControlLock);
1901 740 : break;
1902 : }
1903 :
1904 42 : slotname = s->data.name;
1905 42 : active_pid = s->active_pid;
1906 :
1907 : /*
1908 : * If the slot can be acquired, do so and mark it invalidated
1909 : * immediately. Otherwise we'll signal the owning process, below, and
1910 : * retry.
1911 : */
1912 42 : if (active_pid == 0)
1913 : {
1914 28 : MyReplicationSlot = s;
1915 28 : s->active_pid = MyProcPid;
1916 28 : s->data.invalidated = invalidation_cause;
1917 :
1918 : /*
1919 : * XXX: We should consider not overwriting restart_lsn and instead
1920 : * just rely on .invalidated.
1921 : */
1922 28 : if (invalidation_cause == RS_INVAL_WAL_REMOVED)
1923 : {
1924 8 : s->data.restart_lsn = InvalidXLogRecPtr;
1925 8 : s->last_saved_restart_lsn = InvalidXLogRecPtr;
1926 : }
1927 :
1928 : /* Let caller know */
1929 28 : *invalidated = true;
1930 : }
1931 :
1932 42 : SpinLockRelease(&s->mutex);
1933 :
1934 : /*
1935 : * Calculate the idle time duration of the slot if slot is marked
1936 : * invalidated with RS_INVAL_IDLE_TIMEOUT.
1937 : */
1938 42 : if (invalidation_cause == RS_INVAL_IDLE_TIMEOUT)
1939 : {
1940 : int slot_idle_usecs;
1941 :
1942 0 : TimestampDifference(inactive_since, now, &slot_idle_secs,
1943 : &slot_idle_usecs);
1944 : }
1945 :
1946 42 : if (active_pid != 0)
1947 : {
1948 : /*
1949 : * Prepare the sleep on the slot's condition variable before
1950 : * releasing the lock, to close a possible race condition if the
1951 : * slot is released before the sleep below.
1952 : */
1953 14 : ConditionVariablePrepareToSleep(&s->active_cv);
1954 :
1955 14 : LWLockRelease(ReplicationSlotControlLock);
1956 14 : released_lock = true;
1957 :
1958 : /*
1959 : * Signal to terminate the process that owns the slot, if we
1960 : * haven't already signalled it. (Avoidance of repeated
1961 : * signalling is the only reason for there to be a loop in this
1962 : * routine; otherwise we could rely on caller's restart loop.)
1963 : *
1964 : * There is the race condition that other process may own the slot
1965 : * after its current owner process is terminated and before this
1966 : * process owns it. To handle that, we signal only if the PID of
1967 : * the owning process has changed from the previous time. (This
1968 : * logic assumes that the same PID is not reused very quickly.)
1969 : */
1970 14 : if (last_signaled_pid != active_pid)
1971 : {
1972 14 : ReportSlotInvalidation(invalidation_cause, true, active_pid,
1973 : slotname, restart_lsn,
1974 : oldestLSN, snapshotConflictHorizon,
1975 : slot_idle_secs);
1976 :
1977 14 : if (MyBackendType == B_STARTUP)
1978 10 : (void) SendProcSignal(active_pid,
1979 : PROCSIG_RECOVERY_CONFLICT_LOGICALSLOT,
1980 : INVALID_PROC_NUMBER);
1981 : else
1982 4 : (void) kill(active_pid, SIGTERM);
1983 :
1984 14 : last_signaled_pid = active_pid;
1985 14 : terminated = true;
1986 14 : invalidation_cause_prev = invalidation_cause;
1987 : }
1988 :
1989 : /* Wait until the slot is released. */
1990 14 : ConditionVariableSleep(&s->active_cv,
1991 : WAIT_EVENT_REPLICATION_SLOT_DROP);
1992 :
1993 : /*
1994 : * Re-acquire lock and start over; we expect to invalidate the
1995 : * slot next time (unless another process acquires the slot in the
1996 : * meantime).
1997 : */
1998 14 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1999 14 : continue;
2000 : }
2001 : else
2002 : {
2003 : /*
2004 : * We hold the slot now and have already invalidated it; flush it
2005 : * to ensure that state persists.
2006 : *
2007 : * Don't want to hold ReplicationSlotControlLock across file
2008 : * system operations, so release it now but be sure to tell caller
2009 : * to restart from scratch.
2010 : */
2011 28 : LWLockRelease(ReplicationSlotControlLock);
2012 28 : released_lock = true;
2013 :
2014 : /* Make sure the invalidated state persists across server restart */
2015 28 : ReplicationSlotMarkDirty();
2016 28 : ReplicationSlotSave();
2017 28 : ReplicationSlotRelease();
2018 :
2019 28 : ReportSlotInvalidation(invalidation_cause, false, active_pid,
2020 : slotname, restart_lsn,
2021 : oldestLSN, snapshotConflictHorizon,
2022 : slot_idle_secs);
2023 :
2024 : /* done with this slot for now */
2025 28 : break;
2026 : }
2027 : }
2028 :
2029 : Assert(released_lock == !LWLockHeldByMe(ReplicationSlotControlLock));
2030 :
2031 768 : return released_lock;
2032 : }
2033 :
2034 : /*
2035 : * Invalidate slots that require resources about to be removed.
2036 : *
2037 : * Returns true when any slot have got invalidated.
2038 : *
2039 : * Whether a slot needs to be invalidated depends on the invalidation cause.
2040 : * A slot is invalidated if it:
2041 : * - RS_INVAL_WAL_REMOVED: requires a LSN older than the given segment
2042 : * - RS_INVAL_HORIZON: requires a snapshot <= the given horizon in the given
2043 : * db; dboid may be InvalidOid for shared relations
2044 : * - RS_INVAL_WAL_LEVEL: is logical and wal_level is insufficient
2045 : * - RS_INVAL_IDLE_TIMEOUT: has been idle longer than the configured
2046 : * "idle_replication_slot_timeout" duration.
2047 : *
2048 : * Note: This function attempts to invalidate the slot for multiple possible
2049 : * causes in a single pass, minimizing redundant iterations. The "cause"
2050 : * parameter can be a MASK representing one or more of the defined causes.
2051 : *
2052 : * NB - this runs as part of checkpoint, so avoid raising errors if possible.
2053 : */
2054 : bool
2055 3428 : InvalidateObsoleteReplicationSlots(uint32 possible_causes,
2056 : XLogSegNo oldestSegno, Oid dboid,
2057 : TransactionId snapshotConflictHorizon)
2058 : {
2059 : XLogRecPtr oldestLSN;
2060 3428 : bool invalidated = false;
2061 :
2062 : Assert(!(possible_causes & RS_INVAL_HORIZON) || TransactionIdIsValid(snapshotConflictHorizon));
2063 : Assert(!(possible_causes & RS_INVAL_WAL_REMOVED) || oldestSegno > 0);
2064 : Assert(possible_causes != RS_INVAL_NONE);
2065 :
2066 3428 : if (max_replication_slots == 0)
2067 2 : return invalidated;
2068 :
2069 3426 : XLogSegNoOffsetToRecPtr(oldestSegno, 0, wal_segment_size, oldestLSN);
2070 :
2071 3454 : restart:
2072 3454 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2073 37008 : for (int i = 0; i < max_replication_slots; i++)
2074 : {
2075 33582 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2076 :
2077 33582 : if (!s->in_use)
2078 32800 : continue;
2079 :
2080 : /* Prevent invalidation of logical slots during binary upgrade */
2081 782 : if (SlotIsLogical(s) && IsBinaryUpgrade)
2082 14 : continue;
2083 :
2084 768 : if (InvalidatePossiblyObsoleteSlot(possible_causes, s, oldestLSN, dboid,
2085 : snapshotConflictHorizon,
2086 : &invalidated))
2087 : {
2088 : /* if the lock was released, start from scratch */
2089 28 : goto restart;
2090 : }
2091 : }
2092 3426 : LWLockRelease(ReplicationSlotControlLock);
2093 :
2094 : /*
2095 : * If any slots have been invalidated, recalculate the resource limits.
2096 : */
2097 3426 : if (invalidated)
2098 : {
2099 18 : ReplicationSlotsComputeRequiredXmin(false);
2100 18 : ReplicationSlotsComputeRequiredLSN();
2101 : }
2102 :
2103 3426 : return invalidated;
2104 : }
2105 :
2106 : /*
2107 : * Flush all replication slots to disk.
2108 : *
2109 : * It is convenient to flush dirty replication slots at the time of checkpoint.
2110 : * Additionally, in case of a shutdown checkpoint, we also identify the slots
2111 : * for which the confirmed_flush LSN has been updated since the last time it
2112 : * was saved and flush them.
2113 : */
2114 : void
2115 3388 : CheckPointReplicationSlots(bool is_shutdown)
2116 : {
2117 : int i;
2118 3388 : bool last_saved_restart_lsn_updated = false;
2119 :
2120 3388 : elog(DEBUG1, "performing replication slot checkpoint");
2121 :
2122 : /*
2123 : * Prevent any slot from being created/dropped while we're active. As we
2124 : * explicitly do *not* want to block iterating over replication_slots or
2125 : * acquiring a slot we cannot take the control lock - but that's OK,
2126 : * because holding ReplicationSlotAllocationLock is strictly stronger, and
2127 : * enough to guarantee that nobody can change the in_use bits on us.
2128 : */
2129 3388 : LWLockAcquire(ReplicationSlotAllocationLock, LW_SHARED);
2130 :
2131 36702 : for (i = 0; i < max_replication_slots; i++)
2132 : {
2133 33314 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2134 : char path[MAXPGPATH];
2135 :
2136 33314 : if (!s->in_use)
2137 32638 : continue;
2138 :
2139 : /* save the slot to disk, locking is handled in SaveSlotToPath() */
2140 676 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(s->data.name));
2141 :
2142 : /*
2143 : * Slot's data is not flushed each time the confirmed_flush LSN is
2144 : * updated as that could lead to frequent writes. However, we decide
2145 : * to force a flush of all logical slot's data at the time of shutdown
2146 : * if the confirmed_flush LSN is changed since we last flushed it to
2147 : * disk. This helps in avoiding an unnecessary retreat of the
2148 : * confirmed_flush LSN after restart.
2149 : */
2150 676 : if (is_shutdown && SlotIsLogical(s))
2151 : {
2152 124 : SpinLockAcquire(&s->mutex);
2153 :
2154 124 : if (s->data.invalidated == RS_INVAL_NONE &&
2155 124 : s->data.confirmed_flush > s->last_saved_confirmed_flush)
2156 : {
2157 66 : s->just_dirtied = true;
2158 66 : s->dirty = true;
2159 : }
2160 124 : SpinLockRelease(&s->mutex);
2161 : }
2162 :
2163 : /*
2164 : * Track if we're going to update slot's last_saved_restart_lsn. We
2165 : * need this to know if we need to recompute the required LSN.
2166 : */
2167 676 : if (s->last_saved_restart_lsn != s->data.restart_lsn)
2168 384 : last_saved_restart_lsn_updated = true;
2169 :
2170 676 : SaveSlotToPath(s, path, LOG);
2171 : }
2172 3388 : LWLockRelease(ReplicationSlotAllocationLock);
2173 :
2174 : /*
2175 : * Recompute the required LSN if SaveSlotToPath() updated
2176 : * last_saved_restart_lsn for any slot.
2177 : */
2178 3388 : if (last_saved_restart_lsn_updated)
2179 384 : ReplicationSlotsComputeRequiredLSN();
2180 3388 : }
2181 :
2182 : /*
2183 : * Load all replication slots from disk into memory at server startup. This
2184 : * needs to be run before we start crash recovery.
2185 : */
2186 : void
2187 1864 : StartupReplicationSlots(void)
2188 : {
2189 : DIR *replication_dir;
2190 : struct dirent *replication_de;
2191 :
2192 1864 : elog(DEBUG1, "starting up replication slots");
2193 :
2194 : /* restore all slots by iterating over all on-disk entries */
2195 1864 : replication_dir = AllocateDir(PG_REPLSLOT_DIR);
2196 5752 : while ((replication_de = ReadDir(replication_dir, PG_REPLSLOT_DIR)) != NULL)
2197 : {
2198 : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2199 : PGFileType de_type;
2200 :
2201 3890 : if (strcmp(replication_de->d_name, ".") == 0 ||
2202 2028 : strcmp(replication_de->d_name, "..") == 0)
2203 3724 : continue;
2204 :
2205 166 : snprintf(path, sizeof(path), "%s/%s", PG_REPLSLOT_DIR, replication_de->d_name);
2206 166 : de_type = get_dirent_type(path, replication_de, false, DEBUG1);
2207 :
2208 : /* we're only creating directories here, skip if it's not our's */
2209 166 : if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_DIR)
2210 0 : continue;
2211 :
2212 : /* we crashed while a slot was being setup or deleted, clean up */
2213 166 : if (pg_str_endswith(replication_de->d_name, ".tmp"))
2214 : {
2215 0 : if (!rmtree(path, true))
2216 : {
2217 0 : ereport(WARNING,
2218 : (errmsg("could not remove directory \"%s\"",
2219 : path)));
2220 0 : continue;
2221 : }
2222 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2223 0 : continue;
2224 : }
2225 :
2226 : /* looks like a slot in a normal state, restore */
2227 166 : RestoreSlotFromDisk(replication_de->d_name);
2228 : }
2229 1862 : FreeDir(replication_dir);
2230 :
2231 : /* currently no slots exist, we're done. */
2232 1862 : if (max_replication_slots <= 0)
2233 2 : return;
2234 :
2235 : /* Now that we have recovered all the data, compute replication xmin */
2236 1860 : ReplicationSlotsComputeRequiredXmin(false);
2237 1860 : ReplicationSlotsComputeRequiredLSN();
2238 : }
2239 :
2240 : /* ----
2241 : * Manipulation of on-disk state of replication slots
2242 : *
2243 : * NB: none of the routines below should take any notice whether a slot is the
2244 : * current one or not, that's all handled a layer above.
2245 : * ----
2246 : */
2247 : static void
2248 1276 : CreateSlotOnDisk(ReplicationSlot *slot)
2249 : {
2250 : char tmppath[MAXPGPATH];
2251 : char path[MAXPGPATH];
2252 : struct stat st;
2253 :
2254 : /*
2255 : * No need to take out the io_in_progress_lock, nobody else can see this
2256 : * slot yet, so nobody else will write. We're reusing SaveSlotToPath which
2257 : * takes out the lock, if we'd take the lock here, we'd deadlock.
2258 : */
2259 :
2260 1276 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2261 1276 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2262 :
2263 : /*
2264 : * It's just barely possible that some previous effort to create or drop a
2265 : * slot with this name left a temp directory lying around. If that seems
2266 : * to be the case, try to remove it. If the rmtree() fails, we'll error
2267 : * out at the MakePGDirectory() below, so we don't bother checking
2268 : * success.
2269 : */
2270 1276 : if (stat(tmppath, &st) == 0 && S_ISDIR(st.st_mode))
2271 0 : rmtree(tmppath, true);
2272 :
2273 : /* Create and fsync the temporary slot directory. */
2274 1276 : if (MakePGDirectory(tmppath) < 0)
2275 0 : ereport(ERROR,
2276 : (errcode_for_file_access(),
2277 : errmsg("could not create directory \"%s\": %m",
2278 : tmppath)));
2279 1276 : fsync_fname(tmppath, true);
2280 :
2281 : /* Write the actual state file. */
2282 1276 : slot->dirty = true; /* signal that we really need to write */
2283 1276 : SaveSlotToPath(slot, tmppath, ERROR);
2284 :
2285 : /* Rename the directory into place. */
2286 1276 : if (rename(tmppath, path) != 0)
2287 0 : ereport(ERROR,
2288 : (errcode_for_file_access(),
2289 : errmsg("could not rename file \"%s\" to \"%s\": %m",
2290 : tmppath, path)));
2291 :
2292 : /*
2293 : * If we'd now fail - really unlikely - we wouldn't know whether this slot
2294 : * would persist after an OS crash or not - so, force a restart. The
2295 : * restart would try to fsync this again till it works.
2296 : */
2297 1276 : START_CRIT_SECTION();
2298 :
2299 1276 : fsync_fname(path, true);
2300 1276 : fsync_fname(PG_REPLSLOT_DIR, true);
2301 :
2302 1276 : END_CRIT_SECTION();
2303 1276 : }
2304 :
2305 : /*
2306 : * Shared functionality between saving and creating a replication slot.
2307 : */
2308 : static void
2309 4624 : SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
2310 : {
2311 : char tmppath[MAXPGPATH];
2312 : char path[MAXPGPATH];
2313 : int fd;
2314 : ReplicationSlotOnDisk cp;
2315 : bool was_dirty;
2316 :
2317 : /* first check whether there's something to write out */
2318 4624 : SpinLockAcquire(&slot->mutex);
2319 4624 : was_dirty = slot->dirty;
2320 4624 : slot->just_dirtied = false;
2321 4624 : SpinLockRelease(&slot->mutex);
2322 :
2323 : /* and don't do anything if there's nothing to write */
2324 4624 : if (!was_dirty)
2325 206 : return;
2326 :
2327 4418 : LWLockAcquire(&slot->io_in_progress_lock, LW_EXCLUSIVE);
2328 :
2329 : /* silence valgrind :( */
2330 4418 : memset(&cp, 0, sizeof(ReplicationSlotOnDisk));
2331 :
2332 4418 : sprintf(tmppath, "%s/state.tmp", dir);
2333 4418 : sprintf(path, "%s/state", dir);
2334 :
2335 4418 : fd = OpenTransientFile(tmppath, O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
2336 4418 : if (fd < 0)
2337 : {
2338 : /*
2339 : * If not an ERROR, then release the lock before returning. In case
2340 : * of an ERROR, the error recovery path automatically releases the
2341 : * lock, but no harm in explicitly releasing even in that case. Note
2342 : * that LWLockRelease() could affect errno.
2343 : */
2344 0 : int save_errno = errno;
2345 :
2346 0 : LWLockRelease(&slot->io_in_progress_lock);
2347 0 : errno = save_errno;
2348 0 : ereport(elevel,
2349 : (errcode_for_file_access(),
2350 : errmsg("could not create file \"%s\": %m",
2351 : tmppath)));
2352 0 : return;
2353 : }
2354 :
2355 4418 : cp.magic = SLOT_MAGIC;
2356 4418 : INIT_CRC32C(cp.checksum);
2357 4418 : cp.version = SLOT_VERSION;
2358 4418 : cp.length = ReplicationSlotOnDiskV2Size;
2359 :
2360 4418 : SpinLockAcquire(&slot->mutex);
2361 :
2362 4418 : memcpy(&cp.slotdata, &slot->data, sizeof(ReplicationSlotPersistentData));
2363 :
2364 4418 : SpinLockRelease(&slot->mutex);
2365 :
2366 4418 : COMP_CRC32C(cp.checksum,
2367 : (char *) (&cp) + ReplicationSlotOnDiskNotChecksummedSize,
2368 : ReplicationSlotOnDiskChecksummedSize);
2369 4418 : FIN_CRC32C(cp.checksum);
2370 :
2371 4418 : errno = 0;
2372 4418 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_WRITE);
2373 4418 : if ((write(fd, &cp, sizeof(cp))) != sizeof(cp))
2374 : {
2375 0 : int save_errno = errno;
2376 :
2377 0 : pgstat_report_wait_end();
2378 0 : CloseTransientFile(fd);
2379 0 : LWLockRelease(&slot->io_in_progress_lock);
2380 :
2381 : /* if write didn't set errno, assume problem is no disk space */
2382 0 : errno = save_errno ? save_errno : ENOSPC;
2383 0 : ereport(elevel,
2384 : (errcode_for_file_access(),
2385 : errmsg("could not write to file \"%s\": %m",
2386 : tmppath)));
2387 0 : return;
2388 : }
2389 4418 : pgstat_report_wait_end();
2390 :
2391 : /* fsync the temporary file */
2392 4418 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_SYNC);
2393 4418 : if (pg_fsync(fd) != 0)
2394 : {
2395 0 : int save_errno = errno;
2396 :
2397 0 : pgstat_report_wait_end();
2398 0 : CloseTransientFile(fd);
2399 0 : LWLockRelease(&slot->io_in_progress_lock);
2400 0 : errno = save_errno;
2401 0 : ereport(elevel,
2402 : (errcode_for_file_access(),
2403 : errmsg("could not fsync file \"%s\": %m",
2404 : tmppath)));
2405 0 : return;
2406 : }
2407 4418 : pgstat_report_wait_end();
2408 :
2409 4418 : if (CloseTransientFile(fd) != 0)
2410 : {
2411 0 : int save_errno = errno;
2412 :
2413 0 : LWLockRelease(&slot->io_in_progress_lock);
2414 0 : errno = save_errno;
2415 0 : ereport(elevel,
2416 : (errcode_for_file_access(),
2417 : errmsg("could not close file \"%s\": %m",
2418 : tmppath)));
2419 0 : return;
2420 : }
2421 :
2422 : /* rename to permanent file, fsync file and directory */
2423 4418 : if (rename(tmppath, path) != 0)
2424 : {
2425 0 : int save_errno = errno;
2426 :
2427 0 : LWLockRelease(&slot->io_in_progress_lock);
2428 0 : errno = save_errno;
2429 0 : ereport(elevel,
2430 : (errcode_for_file_access(),
2431 : errmsg("could not rename file \"%s\" to \"%s\": %m",
2432 : tmppath, path)));
2433 0 : return;
2434 : }
2435 :
2436 : /*
2437 : * Check CreateSlotOnDisk() for the reasoning of using a critical section.
2438 : */
2439 4418 : START_CRIT_SECTION();
2440 :
2441 4418 : fsync_fname(path, false);
2442 4418 : fsync_fname(dir, true);
2443 4418 : fsync_fname(PG_REPLSLOT_DIR, true);
2444 :
2445 4418 : END_CRIT_SECTION();
2446 :
2447 : /*
2448 : * Successfully wrote, unset dirty bit, unless somebody dirtied again
2449 : * already and remember the confirmed_flush LSN value.
2450 : */
2451 4418 : SpinLockAcquire(&slot->mutex);
2452 4418 : if (!slot->just_dirtied)
2453 4394 : slot->dirty = false;
2454 4418 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2455 4418 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2456 4418 : SpinLockRelease(&slot->mutex);
2457 :
2458 4418 : LWLockRelease(&slot->io_in_progress_lock);
2459 : }
2460 :
2461 : /*
2462 : * Load a single slot from disk into memory.
2463 : */
2464 : static void
2465 166 : RestoreSlotFromDisk(const char *name)
2466 : {
2467 : ReplicationSlotOnDisk cp;
2468 : int i;
2469 : char slotdir[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2470 : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR) + 10];
2471 : int fd;
2472 166 : bool restored = false;
2473 : int readBytes;
2474 : pg_crc32c checksum;
2475 166 : TimestampTz now = 0;
2476 :
2477 : /* no need to lock here, no concurrent access allowed yet */
2478 :
2479 : /* delete temp file if it exists */
2480 166 : sprintf(slotdir, "%s/%s", PG_REPLSLOT_DIR, name);
2481 166 : sprintf(path, "%s/state.tmp", slotdir);
2482 166 : if (unlink(path) < 0 && errno != ENOENT)
2483 0 : ereport(PANIC,
2484 : (errcode_for_file_access(),
2485 : errmsg("could not remove file \"%s\": %m", path)));
2486 :
2487 166 : sprintf(path, "%s/state", slotdir);
2488 :
2489 166 : elog(DEBUG1, "restoring replication slot from \"%s\"", path);
2490 :
2491 : /* on some operating systems fsyncing a file requires O_RDWR */
2492 166 : fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
2493 :
2494 : /*
2495 : * We do not need to handle this as we are rename()ing the directory into
2496 : * place only after we fsync()ed the state file.
2497 : */
2498 166 : if (fd < 0)
2499 0 : ereport(PANIC,
2500 : (errcode_for_file_access(),
2501 : errmsg("could not open file \"%s\": %m", path)));
2502 :
2503 : /*
2504 : * Sync state file before we're reading from it. We might have crashed
2505 : * while it wasn't synced yet and we shouldn't continue on that basis.
2506 : */
2507 166 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_RESTORE_SYNC);
2508 166 : if (pg_fsync(fd) != 0)
2509 0 : ereport(PANIC,
2510 : (errcode_for_file_access(),
2511 : errmsg("could not fsync file \"%s\": %m",
2512 : path)));
2513 166 : pgstat_report_wait_end();
2514 :
2515 : /* Also sync the parent directory */
2516 166 : START_CRIT_SECTION();
2517 166 : fsync_fname(slotdir, true);
2518 166 : END_CRIT_SECTION();
2519 :
2520 : /* read part of statefile that's guaranteed to be version independent */
2521 166 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2522 166 : readBytes = read(fd, &cp, ReplicationSlotOnDiskConstantSize);
2523 166 : pgstat_report_wait_end();
2524 166 : if (readBytes != ReplicationSlotOnDiskConstantSize)
2525 : {
2526 0 : if (readBytes < 0)
2527 0 : ereport(PANIC,
2528 : (errcode_for_file_access(),
2529 : errmsg("could not read file \"%s\": %m", path)));
2530 : else
2531 0 : ereport(PANIC,
2532 : (errcode(ERRCODE_DATA_CORRUPTED),
2533 : errmsg("could not read file \"%s\": read %d of %zu",
2534 : path, readBytes,
2535 : (Size) ReplicationSlotOnDiskConstantSize)));
2536 : }
2537 :
2538 : /* verify magic */
2539 166 : if (cp.magic != SLOT_MAGIC)
2540 0 : ereport(PANIC,
2541 : (errcode(ERRCODE_DATA_CORRUPTED),
2542 : errmsg("replication slot file \"%s\" has wrong magic number: %u instead of %u",
2543 : path, cp.magic, SLOT_MAGIC)));
2544 :
2545 : /* verify version */
2546 166 : if (cp.version != SLOT_VERSION)
2547 0 : ereport(PANIC,
2548 : (errcode(ERRCODE_DATA_CORRUPTED),
2549 : errmsg("replication slot file \"%s\" has unsupported version %u",
2550 : path, cp.version)));
2551 :
2552 : /* boundary check on length */
2553 166 : if (cp.length != ReplicationSlotOnDiskV2Size)
2554 0 : ereport(PANIC,
2555 : (errcode(ERRCODE_DATA_CORRUPTED),
2556 : errmsg("replication slot file \"%s\" has corrupted length %u",
2557 : path, cp.length)));
2558 :
2559 : /* Now that we know the size, read the entire file */
2560 166 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2561 332 : readBytes = read(fd,
2562 : (char *) &cp + ReplicationSlotOnDiskConstantSize,
2563 166 : cp.length);
2564 166 : pgstat_report_wait_end();
2565 166 : if (readBytes != cp.length)
2566 : {
2567 0 : if (readBytes < 0)
2568 0 : ereport(PANIC,
2569 : (errcode_for_file_access(),
2570 : errmsg("could not read file \"%s\": %m", path)));
2571 : else
2572 0 : ereport(PANIC,
2573 : (errcode(ERRCODE_DATA_CORRUPTED),
2574 : errmsg("could not read file \"%s\": read %d of %zu",
2575 : path, readBytes, (Size) cp.length)));
2576 : }
2577 :
2578 166 : if (CloseTransientFile(fd) != 0)
2579 0 : ereport(PANIC,
2580 : (errcode_for_file_access(),
2581 : errmsg("could not close file \"%s\": %m", path)));
2582 :
2583 : /* now verify the CRC */
2584 166 : INIT_CRC32C(checksum);
2585 166 : COMP_CRC32C(checksum,
2586 : (char *) &cp + ReplicationSlotOnDiskNotChecksummedSize,
2587 : ReplicationSlotOnDiskChecksummedSize);
2588 166 : FIN_CRC32C(checksum);
2589 :
2590 166 : if (!EQ_CRC32C(checksum, cp.checksum))
2591 0 : ereport(PANIC,
2592 : (errmsg("checksum mismatch for replication slot file \"%s\": is %u, should be %u",
2593 : path, checksum, cp.checksum)));
2594 :
2595 : /*
2596 : * If we crashed with an ephemeral slot active, don't restore but delete
2597 : * it.
2598 : */
2599 166 : if (cp.slotdata.persistency != RS_PERSISTENT)
2600 : {
2601 0 : if (!rmtree(slotdir, true))
2602 : {
2603 0 : ereport(WARNING,
2604 : (errmsg("could not remove directory \"%s\"",
2605 : slotdir)));
2606 : }
2607 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2608 0 : return;
2609 : }
2610 :
2611 : /*
2612 : * Verify that requirements for the specific slot type are met. That's
2613 : * important because if these aren't met we're not guaranteed to retain
2614 : * all the necessary resources for the slot.
2615 : *
2616 : * NB: We have to do so *after* the above checks for ephemeral slots,
2617 : * because otherwise a slot that shouldn't exist anymore could prevent
2618 : * restarts.
2619 : *
2620 : * NB: Changing the requirements here also requires adapting
2621 : * CheckSlotRequirements() and CheckLogicalDecodingRequirements().
2622 : */
2623 166 : if (cp.slotdata.database != InvalidOid)
2624 : {
2625 116 : if (wal_level < WAL_LEVEL_LOGICAL)
2626 0 : ereport(FATAL,
2627 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2628 : errmsg("logical replication slot \"%s\" exists, but \"wal_level\" < \"logical\"",
2629 : NameStr(cp.slotdata.name)),
2630 : errhint("Change \"wal_level\" to be \"logical\" or higher.")));
2631 :
2632 : /*
2633 : * In standby mode, the hot standby must be enabled. This check is
2634 : * necessary to ensure logical slots are invalidated when they become
2635 : * incompatible due to insufficient wal_level. Otherwise, if the
2636 : * primary reduces wal_level < logical while hot standby is disabled,
2637 : * logical slots would remain valid even after promotion.
2638 : */
2639 116 : if (StandbyMode && !EnableHotStandby)
2640 2 : ereport(FATAL,
2641 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2642 : errmsg("logical replication slot \"%s\" exists on the standby, but \"hot_standby\" = \"off\"",
2643 : NameStr(cp.slotdata.name)),
2644 : errhint("Change \"hot_standby\" to be \"on\".")));
2645 : }
2646 50 : else if (wal_level < WAL_LEVEL_REPLICA)
2647 0 : ereport(FATAL,
2648 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2649 : errmsg("physical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2650 : NameStr(cp.slotdata.name)),
2651 : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2652 :
2653 : /* nothing can be active yet, don't lock anything */
2654 220 : for (i = 0; i < max_replication_slots; i++)
2655 : {
2656 : ReplicationSlot *slot;
2657 :
2658 220 : slot = &ReplicationSlotCtl->replication_slots[i];
2659 :
2660 220 : if (slot->in_use)
2661 56 : continue;
2662 :
2663 : /* restore the entire set of persistent data */
2664 164 : memcpy(&slot->data, &cp.slotdata,
2665 : sizeof(ReplicationSlotPersistentData));
2666 :
2667 : /* initialize in memory state */
2668 164 : slot->effective_xmin = cp.slotdata.xmin;
2669 164 : slot->effective_catalog_xmin = cp.slotdata.catalog_xmin;
2670 164 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2671 164 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2672 :
2673 164 : slot->candidate_catalog_xmin = InvalidTransactionId;
2674 164 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
2675 164 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
2676 164 : slot->candidate_restart_valid = InvalidXLogRecPtr;
2677 :
2678 164 : slot->in_use = true;
2679 164 : slot->active_pid = 0;
2680 :
2681 : /*
2682 : * Set the time since the slot has become inactive after loading the
2683 : * slot from the disk into memory. Whoever acquires the slot i.e.
2684 : * makes the slot active will reset it. Use the same inactive_since
2685 : * time for all the slots.
2686 : */
2687 164 : if (now == 0)
2688 164 : now = GetCurrentTimestamp();
2689 :
2690 164 : ReplicationSlotSetInactiveSince(slot, now, false);
2691 :
2692 164 : restored = true;
2693 164 : break;
2694 : }
2695 :
2696 164 : if (!restored)
2697 0 : ereport(FATAL,
2698 : (errmsg("too many replication slots active before shutdown"),
2699 : errhint("Increase \"max_replication_slots\" and try again.")));
2700 : }
2701 :
2702 : /*
2703 : * Maps an invalidation reason for a replication slot to
2704 : * ReplicationSlotInvalidationCause.
2705 : */
2706 : ReplicationSlotInvalidationCause
2707 0 : GetSlotInvalidationCause(const char *cause_name)
2708 : {
2709 : Assert(cause_name);
2710 :
2711 : /* Search lookup table for the cause having this name */
2712 0 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2713 : {
2714 0 : if (strcmp(SlotInvalidationCauses[i].cause_name, cause_name) == 0)
2715 0 : return SlotInvalidationCauses[i].cause;
2716 : }
2717 :
2718 : Assert(false);
2719 0 : return RS_INVAL_NONE; /* to keep compiler quiet */
2720 : }
2721 :
2722 : /*
2723 : * Maps an ReplicationSlotInvalidationCause to the invalidation
2724 : * reason for a replication slot.
2725 : */
2726 : const char *
2727 80 : GetSlotInvalidationCauseName(ReplicationSlotInvalidationCause cause)
2728 : {
2729 : /* Search lookup table for the name of this cause */
2730 250 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2731 : {
2732 250 : if (SlotInvalidationCauses[i].cause == cause)
2733 80 : return SlotInvalidationCauses[i].cause_name;
2734 : }
2735 :
2736 : Assert(false);
2737 0 : return "none"; /* to keep compiler quiet */
2738 : }
2739 :
2740 : /*
2741 : * A helper function to validate slots specified in GUC synchronized_standby_slots.
2742 : *
2743 : * The rawname will be parsed, and the result will be saved into *elemlist.
2744 : */
2745 : static bool
2746 12 : validate_sync_standby_slots(char *rawname, List **elemlist)
2747 : {
2748 : bool ok;
2749 :
2750 : /* Verify syntax and parse string into a list of identifiers */
2751 12 : ok = SplitIdentifierString(rawname, ',', elemlist);
2752 :
2753 12 : if (!ok)
2754 : {
2755 0 : GUC_check_errdetail("List syntax is invalid.");
2756 : }
2757 12 : else if (MyProc)
2758 : {
2759 : /*
2760 : * Check that each specified slot exist and is physical.
2761 : *
2762 : * Because we need an LWLock, we cannot do this on processes without a
2763 : * PGPROC, so we skip it there; but see comments in
2764 : * StandbySlotsHaveCaughtup() as to why that's not a problem.
2765 : */
2766 6 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2767 :
2768 18 : foreach_ptr(char, name, *elemlist)
2769 : {
2770 : ReplicationSlot *slot;
2771 :
2772 6 : slot = SearchNamedReplicationSlot(name, false);
2773 :
2774 6 : if (!slot)
2775 : {
2776 0 : GUC_check_errdetail("Replication slot \"%s\" does not exist.",
2777 : name);
2778 0 : ok = false;
2779 0 : break;
2780 : }
2781 :
2782 6 : if (!SlotIsPhysical(slot))
2783 : {
2784 0 : GUC_check_errdetail("\"%s\" is not a physical replication slot.",
2785 : name);
2786 0 : ok = false;
2787 0 : break;
2788 : }
2789 : }
2790 :
2791 6 : LWLockRelease(ReplicationSlotControlLock);
2792 : }
2793 :
2794 12 : return ok;
2795 : }
2796 :
2797 : /*
2798 : * GUC check_hook for synchronized_standby_slots
2799 : */
2800 : bool
2801 2246 : check_synchronized_standby_slots(char **newval, void **extra, GucSource source)
2802 : {
2803 : char *rawname;
2804 : char *ptr;
2805 : List *elemlist;
2806 : int size;
2807 : bool ok;
2808 : SyncStandbySlotsConfigData *config;
2809 :
2810 2246 : if ((*newval)[0] == '\0')
2811 2234 : return true;
2812 :
2813 : /* Need a modifiable copy of the GUC string */
2814 12 : rawname = pstrdup(*newval);
2815 :
2816 : /* Now verify if the specified slots exist and have correct type */
2817 12 : ok = validate_sync_standby_slots(rawname, &elemlist);
2818 :
2819 12 : if (!ok || elemlist == NIL)
2820 : {
2821 0 : pfree(rawname);
2822 0 : list_free(elemlist);
2823 0 : return ok;
2824 : }
2825 :
2826 : /* Compute the size required for the SyncStandbySlotsConfigData struct */
2827 12 : size = offsetof(SyncStandbySlotsConfigData, slot_names);
2828 36 : foreach_ptr(char, slot_name, elemlist)
2829 12 : size += strlen(slot_name) + 1;
2830 :
2831 : /* GUC extra value must be guc_malloc'd, not palloc'd */
2832 12 : config = (SyncStandbySlotsConfigData *) guc_malloc(LOG, size);
2833 12 : if (!config)
2834 0 : return false;
2835 :
2836 : /* Transform the data into SyncStandbySlotsConfigData */
2837 12 : config->nslotnames = list_length(elemlist);
2838 :
2839 12 : ptr = config->slot_names;
2840 36 : foreach_ptr(char, slot_name, elemlist)
2841 : {
2842 12 : strcpy(ptr, slot_name);
2843 12 : ptr += strlen(slot_name) + 1;
2844 : }
2845 :
2846 12 : *extra = config;
2847 :
2848 12 : pfree(rawname);
2849 12 : list_free(elemlist);
2850 12 : return true;
2851 : }
2852 :
2853 : /*
2854 : * GUC assign_hook for synchronized_standby_slots
2855 : */
2856 : void
2857 2246 : assign_synchronized_standby_slots(const char *newval, void *extra)
2858 : {
2859 : /*
2860 : * The standby slots may have changed, so we must recompute the oldest
2861 : * LSN.
2862 : */
2863 2246 : ss_oldest_flush_lsn = InvalidXLogRecPtr;
2864 :
2865 2246 : synchronized_standby_slots_config = (SyncStandbySlotsConfigData *) extra;
2866 2246 : }
2867 :
2868 : /*
2869 : * Check if the passed slot_name is specified in the synchronized_standby_slots GUC.
2870 : */
2871 : bool
2872 50868 : SlotExistsInSyncStandbySlots(const char *slot_name)
2873 : {
2874 : const char *standby_slot_name;
2875 :
2876 : /* Return false if there is no value in synchronized_standby_slots */
2877 50868 : if (synchronized_standby_slots_config == NULL)
2878 50856 : return false;
2879 :
2880 : /*
2881 : * XXX: We are not expecting this list to be long so a linear search
2882 : * shouldn't hurt but if that turns out not to be true then we can cache
2883 : * this information for each WalSender as well.
2884 : */
2885 12 : standby_slot_name = synchronized_standby_slots_config->slot_names;
2886 12 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
2887 : {
2888 12 : if (strcmp(standby_slot_name, slot_name) == 0)
2889 12 : return true;
2890 :
2891 0 : standby_slot_name += strlen(standby_slot_name) + 1;
2892 : }
2893 :
2894 0 : return false;
2895 : }
2896 :
2897 : /*
2898 : * Return true if the slots specified in synchronized_standby_slots have caught up to
2899 : * the given WAL location, false otherwise.
2900 : *
2901 : * The elevel parameter specifies the error level used for logging messages
2902 : * related to slots that do not exist, are invalidated, or are inactive.
2903 : */
2904 : bool
2905 1220 : StandbySlotsHaveCaughtup(XLogRecPtr wait_for_lsn, int elevel)
2906 : {
2907 : const char *name;
2908 1220 : int caught_up_slot_num = 0;
2909 1220 : XLogRecPtr min_restart_lsn = InvalidXLogRecPtr;
2910 :
2911 : /*
2912 : * Don't need to wait for the standbys to catch up if there is no value in
2913 : * synchronized_standby_slots.
2914 : */
2915 1220 : if (synchronized_standby_slots_config == NULL)
2916 1190 : return true;
2917 :
2918 : /*
2919 : * Don't need to wait for the standbys to catch up if we are on a standby
2920 : * server, since we do not support syncing slots to cascading standbys.
2921 : */
2922 30 : if (RecoveryInProgress())
2923 0 : return true;
2924 :
2925 : /*
2926 : * Don't need to wait for the standbys to catch up if they are already
2927 : * beyond the specified WAL location.
2928 : */
2929 30 : if (!XLogRecPtrIsInvalid(ss_oldest_flush_lsn) &&
2930 18 : ss_oldest_flush_lsn >= wait_for_lsn)
2931 12 : return true;
2932 :
2933 : /*
2934 : * To prevent concurrent slot dropping and creation while filtering the
2935 : * slots, take the ReplicationSlotControlLock outside of the loop.
2936 : */
2937 18 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2938 :
2939 18 : name = synchronized_standby_slots_config->slot_names;
2940 28 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
2941 : {
2942 : XLogRecPtr restart_lsn;
2943 : bool invalidated;
2944 : bool inactive;
2945 : ReplicationSlot *slot;
2946 :
2947 18 : slot = SearchNamedReplicationSlot(name, false);
2948 :
2949 : /*
2950 : * If a slot name provided in synchronized_standby_slots does not
2951 : * exist, report a message and exit the loop.
2952 : *
2953 : * Though validate_sync_standby_slots (the GUC check_hook) tries to
2954 : * avoid this, it can nonetheless happen because the user can specify
2955 : * a nonexistent slot name before server startup. That function cannot
2956 : * validate such a slot during startup, as ReplicationSlotCtl is not
2957 : * initialized by then. Also, the user might have dropped one slot.
2958 : */
2959 18 : if (!slot)
2960 : {
2961 0 : ereport(elevel,
2962 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2963 : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not exist",
2964 : name, "synchronized_standby_slots"),
2965 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
2966 : name),
2967 : errhint("Create the replication slot \"%s\" or amend parameter \"%s\".",
2968 : name, "synchronized_standby_slots"));
2969 0 : break;
2970 : }
2971 :
2972 : /* Same as above: if a slot is not physical, exit the loop. */
2973 18 : if (SlotIsLogical(slot))
2974 : {
2975 0 : ereport(elevel,
2976 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2977 : errmsg("cannot specify logical replication slot \"%s\" in parameter \"%s\"",
2978 : name, "synchronized_standby_slots"),
2979 : errdetail("Logical replication is waiting for correction on replication slot \"%s\".",
2980 : name),
2981 : errhint("Remove the logical replication slot \"%s\" from parameter \"%s\".",
2982 : name, "synchronized_standby_slots"));
2983 0 : break;
2984 : }
2985 :
2986 18 : SpinLockAcquire(&slot->mutex);
2987 18 : restart_lsn = slot->data.restart_lsn;
2988 18 : invalidated = slot->data.invalidated != RS_INVAL_NONE;
2989 18 : inactive = slot->active_pid == 0;
2990 18 : SpinLockRelease(&slot->mutex);
2991 :
2992 18 : if (invalidated)
2993 : {
2994 : /* Specified physical slot has been invalidated */
2995 0 : ereport(elevel,
2996 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2997 : errmsg("physical replication slot \"%s\" specified in parameter \"%s\" has been invalidated",
2998 : name, "synchronized_standby_slots"),
2999 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3000 : name),
3001 : errhint("Drop and recreate the replication slot \"%s\", or amend parameter \"%s\".",
3002 : name, "synchronized_standby_slots"));
3003 0 : break;
3004 : }
3005 :
3006 18 : if (XLogRecPtrIsInvalid(restart_lsn) || restart_lsn < wait_for_lsn)
3007 : {
3008 : /* Log a message if no active_pid for this physical slot */
3009 8 : if (inactive)
3010 6 : ereport(elevel,
3011 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3012 : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not have active_pid",
3013 : name, "synchronized_standby_slots"),
3014 : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3015 : name),
3016 : errhint("Start the standby associated with the replication slot \"%s\", or amend parameter \"%s\".",
3017 : name, "synchronized_standby_slots"));
3018 :
3019 : /* Continue if the current slot hasn't caught up. */
3020 8 : break;
3021 : }
3022 :
3023 : Assert(restart_lsn >= wait_for_lsn);
3024 :
3025 10 : if (XLogRecPtrIsInvalid(min_restart_lsn) ||
3026 : min_restart_lsn > restart_lsn)
3027 10 : min_restart_lsn = restart_lsn;
3028 :
3029 10 : caught_up_slot_num++;
3030 :
3031 10 : name += strlen(name) + 1;
3032 : }
3033 :
3034 18 : LWLockRelease(ReplicationSlotControlLock);
3035 :
3036 : /*
3037 : * Return false if not all the standbys have caught up to the specified
3038 : * WAL location.
3039 : */
3040 18 : if (caught_up_slot_num != synchronized_standby_slots_config->nslotnames)
3041 8 : return false;
3042 :
3043 : /* The ss_oldest_flush_lsn must not retreat. */
3044 : Assert(XLogRecPtrIsInvalid(ss_oldest_flush_lsn) ||
3045 : min_restart_lsn >= ss_oldest_flush_lsn);
3046 :
3047 10 : ss_oldest_flush_lsn = min_restart_lsn;
3048 :
3049 10 : return true;
3050 : }
3051 :
3052 : /*
3053 : * Wait for physical standbys to confirm receiving the given lsn.
3054 : *
3055 : * Used by logical decoding SQL functions. It waits for physical standbys
3056 : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
3057 : */
3058 : void
3059 448 : WaitForStandbyConfirmation(XLogRecPtr wait_for_lsn)
3060 : {
3061 : /*
3062 : * Don't need to wait for the standby to catch up if the current acquired
3063 : * slot is not a logical failover slot, or there is no value in
3064 : * synchronized_standby_slots.
3065 : */
3066 448 : if (!MyReplicationSlot->data.failover || !synchronized_standby_slots_config)
3067 446 : return;
3068 :
3069 2 : ConditionVariablePrepareToSleep(&WalSndCtl->wal_confirm_rcv_cv);
3070 :
3071 : for (;;)
3072 : {
3073 4 : CHECK_FOR_INTERRUPTS();
3074 :
3075 4 : if (ConfigReloadPending)
3076 : {
3077 2 : ConfigReloadPending = false;
3078 2 : ProcessConfigFile(PGC_SIGHUP);
3079 : }
3080 :
3081 : /* Exit if done waiting for every slot. */
3082 4 : if (StandbySlotsHaveCaughtup(wait_for_lsn, WARNING))
3083 2 : break;
3084 :
3085 : /*
3086 : * Wait for the slots in the synchronized_standby_slots to catch up,
3087 : * but use a timeout (1s) so we can also check if the
3088 : * synchronized_standby_slots has been changed.
3089 : */
3090 2 : ConditionVariableTimedSleep(&WalSndCtl->wal_confirm_rcv_cv, 1000,
3091 : WAIT_EVENT_WAIT_FOR_STANDBY_CONFIRMATION);
3092 : }
3093 :
3094 2 : ConditionVariableCancelSleep();
3095 : }
|
