Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * slot.c
4 : : * Replication slot management.
5 : : *
6 : : *
7 : : * Copyright (c) 2012-2026, PostgreSQL Global Development Group
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/replication/slot.c
12 : : *
13 : : * NOTES
14 : : *
15 : : * Replication slots are used to keep state about replication streams
16 : : * originating from this cluster. Their primary purpose is to prevent the
17 : : * premature removal of WAL or of old tuple versions in a manner that would
18 : : * interfere with replication; they are also useful for monitoring purposes.
19 : : * Slots need to be permanent (to allow restarts), crash-safe, and allocatable
20 : : * on standbys (to support cascading setups). The requirement that slots be
21 : : * usable on standbys precludes storing them in the system catalogs.
22 : : *
23 : : * Each replication slot gets its own directory inside the directory
24 : : * $PGDATA / PG_REPLSLOT_DIR. Inside that directory the state file will
25 : : * contain the slot's own data. Additional data can be stored alongside that
26 : : * file if required. While the server is running, the state data is also
27 : : * cached in memory for efficiency.
28 : : *
29 : : * ReplicationSlotAllocationLock must be taken in exclusive mode to allocate
30 : : * or free a slot. ReplicationSlotControlLock must be taken in shared mode
31 : : * to iterate over the slots, and in exclusive mode to change the in_use flag
32 : : * of a slot. The remaining data in each slot is protected by its mutex.
33 : : *
34 : : *-------------------------------------------------------------------------
35 : : */
36 : :
37 : : #include "postgres.h"
38 : :
39 : : #include <unistd.h>
40 : : #include <sys/stat.h>
41 : :
42 : : #include "access/transam.h"
43 : : #include "access/xlog_internal.h"
44 : : #include "access/xlogrecovery.h"
45 : : #include "common/file_utils.h"
46 : : #include "common/string.h"
47 : : #include "miscadmin.h"
48 : : #include "pgstat.h"
49 : : #include "postmaster/interrupt.h"
50 : : #include "replication/logicallauncher.h"
51 : : #include "replication/slotsync.h"
52 : : #include "replication/slot.h"
53 : : #include "replication/walsender_private.h"
54 : : #include "storage/fd.h"
55 : : #include "storage/ipc.h"
56 : : #include "storage/proc.h"
57 : : #include "storage/procarray.h"
58 : : #include "storage/subsystems.h"
59 : : #include "utils/builtins.h"
60 : : #include "utils/guc_hooks.h"
61 : : #include "utils/injection_point.h"
62 : : #include "utils/varlena.h"
63 : : #include "utils/wait_event.h"
64 : :
65 : : /*
66 : : * Replication slot on-disk data structure.
67 : : */
68 : : typedef struct ReplicationSlotOnDisk
69 : : {
70 : : /* first part of this struct needs to be version independent */
71 : :
72 : : /* data not covered by checksum */
73 : : uint32 magic;
74 : : pg_crc32c checksum;
75 : :
76 : : /* data covered by checksum */
77 : : uint32 version;
78 : : uint32 length;
79 : :
80 : : /*
81 : : * The actual data in the slot that follows can differ based on the above
82 : : * 'version'.
83 : : */
84 : :
85 : : ReplicationSlotPersistentData slotdata;
86 : : } ReplicationSlotOnDisk;
87 : :
88 : : /*
89 : : * Struct for the configuration of synchronized_standby_slots.
90 : : *
91 : : * Note: this must be a flat representation that can be held in a single chunk
92 : : * of guc_malloc'd memory, so that it can be stored as the "extra" data for the
93 : : * synchronized_standby_slots GUC.
94 : : */
95 : : typedef struct
96 : : {
97 : : /* Number of slot names in the slot_names[] */
98 : : int nslotnames;
99 : :
100 : : /*
101 : : * slot_names contains 'nslotnames' consecutive null-terminated C strings.
102 : : */
103 : : char slot_names[FLEXIBLE_ARRAY_MEMBER];
104 : : } SyncStandbySlotsConfigData;
105 : :
106 : : /*
107 : : * Lookup table for slot invalidation causes.
108 : : */
109 : : typedef struct SlotInvalidationCauseMap
110 : : {
111 : : ReplicationSlotInvalidationCause cause;
112 : : const char *cause_name;
113 : : } SlotInvalidationCauseMap;
114 : :
115 : : static const SlotInvalidationCauseMap SlotInvalidationCauses[] = {
116 : : {RS_INVAL_NONE, "none"},
117 : : {RS_INVAL_WAL_REMOVED, "wal_removed"},
118 : : {RS_INVAL_HORIZON, "rows_removed"},
119 : : {RS_INVAL_WAL_LEVEL, "wal_level_insufficient"},
120 : : {RS_INVAL_IDLE_TIMEOUT, "idle_timeout"},
121 : : };
122 : :
123 : : /*
124 : : * Ensure that the lookup table is up-to-date with the enums defined in
125 : : * ReplicationSlotInvalidationCause.
126 : : */
127 : : StaticAssertDecl(lengthof(SlotInvalidationCauses) == (RS_INVAL_MAX_CAUSES + 1),
128 : : "array length mismatch");
129 : :
130 : : /* size of version independent data */
131 : : #define ReplicationSlotOnDiskConstantSize \
132 : : offsetof(ReplicationSlotOnDisk, slotdata)
133 : : /* size of the part of the slot not covered by the checksum */
134 : : #define ReplicationSlotOnDiskNotChecksummedSize \
135 : : offsetof(ReplicationSlotOnDisk, version)
136 : : /* size of the part covered by the checksum */
137 : : #define ReplicationSlotOnDiskChecksummedSize \
138 : : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskNotChecksummedSize
139 : : /* size of the slot data that is version dependent */
140 : : #define ReplicationSlotOnDiskV2Size \
141 : : sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskConstantSize
142 : :
143 : : #define SLOT_MAGIC 0x1051CA1 /* format identifier */
144 : : #define SLOT_VERSION 5 /* version for new files */
145 : :
146 : : /* Control array for replication slot management */
147 : : ReplicationSlotCtlData *ReplicationSlotCtl = NULL;
148 : :
149 : : static void ReplicationSlotsShmemRequest(void *arg);
150 : : static void ReplicationSlotsShmemInit(void *arg);
151 : :
152 : : const ShmemCallbacks ReplicationSlotsShmemCallbacks = {
153 : : .request_fn = ReplicationSlotsShmemRequest,
154 : : .init_fn = ReplicationSlotsShmemInit,
155 : : };
156 : :
157 : : /* My backend's replication slot in the shared memory array */
158 : : ReplicationSlot *MyReplicationSlot = NULL;
159 : :
160 : : /* GUC variables */
161 : : int max_replication_slots = 10; /* the maximum number of replication
162 : : * slots */
163 : : int max_repack_replication_slots = 5; /* the maximum number of slots
164 : : * for REPACK */
165 : :
166 : : /*
167 : : * Invalidate replication slots that have remained idle longer than this
168 : : * duration; '0' disables it.
169 : : */
170 : : int idle_replication_slot_timeout_secs = 0;
171 : :
172 : : /*
173 : : * This GUC lists streaming replication standby server slot names that
174 : : * logical WAL sender processes will wait for.
175 : : */
176 : : char *synchronized_standby_slots;
177 : :
178 : : /* This is the parsed and cached configuration for synchronized_standby_slots */
179 : : static SyncStandbySlotsConfigData *synchronized_standby_slots_config;
180 : :
181 : : /*
182 : : * Oldest LSN that has been confirmed to be flushed to the standbys
183 : : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
184 : : */
185 : : static XLogRecPtr ss_oldest_flush_lsn = InvalidXLogRecPtr;
186 : :
187 : : static void ReplicationSlotShmemExit(int code, Datum arg);
188 : : static bool IsSlotForConflictCheck(const char *name);
189 : : static void ReplicationSlotDropPtr(ReplicationSlot *slot);
190 : :
191 : : /* internal persistency functions */
192 : : static void RestoreSlotFromDisk(const char *name);
193 : : static void CreateSlotOnDisk(ReplicationSlot *slot);
194 : : static void SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel);
195 : :
196 : : /*
197 : : * Register shared memory space needed for replication slots.
198 : : */
199 : : static void
200 : 1271 : ReplicationSlotsShmemRequest(void *arg)
201 : : {
202 : : Size size;
203 : :
204 [ - + ]: 1271 : if (max_replication_slots + max_repack_replication_slots == 0)
205 : 0 : return;
206 : :
207 : 1271 : size = offsetof(ReplicationSlotCtlData, replication_slots);
208 : 1271 : size = add_size(size,
209 : 1271 : mul_size(max_replication_slots + max_repack_replication_slots,
210 : : sizeof(ReplicationSlot)));
211 : 1271 : ShmemRequestStruct(.name = "ReplicationSlot Ctl",
212 : : .size = size,
213 : : .ptr = (void **) &ReplicationSlotCtl,
214 : : );
215 : : }
216 : :
217 : : /*
218 : : * Initialize shared memory for replication slots.
219 : : */
220 : : static void
221 : 1268 : ReplicationSlotsShmemInit(void *arg)
222 : : {
223 [ + + ]: 20073 : for (int i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
224 : : {
225 : 18805 : ReplicationSlot *slot = &ReplicationSlotCtl->replication_slots[i];
226 : :
227 : : /* everything else is zeroed by the memset above */
228 : 18805 : slot->active_proc = INVALID_PROC_NUMBER;
229 : 18805 : SpinLockInit(&slot->mutex);
230 : 18805 : LWLockInitialize(&slot->io_in_progress_lock,
231 : : LWTRANCHE_REPLICATION_SLOT_IO);
232 : 18805 : ConditionVariableInit(&slot->active_cv);
233 : : }
234 : 1268 : }
235 : :
236 : : /*
237 : : * Register the callback for replication slot cleanup and releasing.
238 : : */
239 : : void
240 : 25228 : ReplicationSlotInitialize(void)
241 : : {
242 : 25228 : before_shmem_exit(ReplicationSlotShmemExit, 0);
243 : 25228 : }
244 : :
245 : : /*
246 : : * Release and cleanup replication slots.
247 : : */
248 : : static void
249 : 25228 : ReplicationSlotShmemExit(int code, Datum arg)
250 : : {
251 : : /* Make sure active replication slots are released */
252 [ + + ]: 25228 : if (MyReplicationSlot != NULL)
253 : 322 : ReplicationSlotRelease();
254 : :
255 : : /* Also cleanup all the temporary slots. */
256 : 25228 : ReplicationSlotCleanup(false);
257 : 25228 : }
258 : :
259 : : /*
260 : : * Check whether the passed slot name is valid and report errors at elevel.
261 : : *
262 : : * See comments for ReplicationSlotValidateNameInternal().
263 : : */
264 : : bool
265 : 909 : ReplicationSlotValidateName(const char *name, bool allow_reserved_name,
266 : : int elevel)
267 : : {
268 : : int err_code;
269 : 909 : char *err_msg = NULL;
270 : 909 : char *err_hint = NULL;
271 : :
272 [ + + ]: 909 : if (!ReplicationSlotValidateNameInternal(name, allow_reserved_name,
273 : : &err_code, &err_msg, &err_hint))
274 : : {
275 : : /*
276 : : * Use errmsg_internal() and errhint_internal() instead of errmsg()
277 : : * and errhint(), since the messages from
278 : : * ReplicationSlotValidateNameInternal() are already translated. This
279 : : * avoids double translation.
280 : : */
281 [ + - + + ]: 5 : ereport(elevel,
282 : : errcode(err_code),
283 : : errmsg_internal("%s", err_msg),
284 : : (err_hint != NULL) ? errhint_internal("%s", err_hint) : 0);
285 : :
286 : 0 : pfree(err_msg);
287 [ # # ]: 0 : if (err_hint != NULL)
288 : 0 : pfree(err_hint);
289 : 0 : return false;
290 : : }
291 : :
292 : 904 : return true;
293 : : }
294 : :
295 : : /*
296 : : * Check whether the passed slot name is valid.
297 : : *
298 : : * An error will be reported for a reserved replication slot name if
299 : : * allow_reserved_name is set to false.
300 : : *
301 : : * Slot names may consist out of [a-z0-9_]{1,NAMEDATALEN-1} which should allow
302 : : * the name to be used as a directory name on every supported OS.
303 : : *
304 : : * Returns true if the slot name is valid. Otherwise, returns false and stores
305 : : * the error code, error message, and optional hint in err_code, err_msg, and
306 : : * err_hint, respectively. The caller is responsible for freeing err_msg and
307 : : * err_hint, which are palloc'd.
308 : : */
309 : : bool
310 : 1127 : ReplicationSlotValidateNameInternal(const char *name, bool allow_reserved_name,
311 : : int *err_code, char **err_msg, char **err_hint)
312 : : {
313 : : const char *cp;
314 : :
315 [ + + ]: 1127 : if (strlen(name) == 0)
316 : : {
317 : 4 : *err_code = ERRCODE_INVALID_NAME;
318 : 4 : *err_msg = psprintf(_("replication slot name \"%s\" is too short"), name);
319 : 4 : *err_hint = NULL;
320 : 4 : return false;
321 : : }
322 : :
323 [ - + ]: 1123 : if (strlen(name) >= NAMEDATALEN)
324 : : {
325 : 0 : *err_code = ERRCODE_NAME_TOO_LONG;
326 : 0 : *err_msg = psprintf(_("replication slot name \"%s\" is too long"), name);
327 : 0 : *err_hint = NULL;
328 : 0 : return false;
329 : : }
330 : :
331 [ + + ]: 21495 : for (cp = name; *cp; cp++)
332 : : {
333 [ + + - + ]: 20374 : if (!((*cp >= 'a' && *cp <= 'z')
334 [ + + + + ]: 9616 : || (*cp >= '0' && *cp <= '9')
335 [ + + ]: 1985 : || (*cp == '_')))
336 : : {
337 : 2 : *err_code = ERRCODE_INVALID_NAME;
338 : 2 : *err_msg = psprintf(_("replication slot name \"%s\" contains invalid character"), name);
339 : 2 : *err_hint = psprintf(_("Replication slot names may only contain lower case letters, numbers, and the underscore character."));
340 : 2 : return false;
341 : : }
342 : : }
343 : :
344 [ + + + + ]: 1121 : if (!allow_reserved_name && IsSlotForConflictCheck(name))
345 : : {
346 : 1 : *err_code = ERRCODE_RESERVED_NAME;
347 : 1 : *err_msg = psprintf(_("replication slot name \"%s\" is reserved"), name);
348 : 1 : *err_hint = psprintf(_("The name \"%s\" is reserved for the conflict detection slot."),
349 : : CONFLICT_DETECTION_SLOT);
350 : 1 : return false;
351 : : }
352 : :
353 : 1120 : return true;
354 : : }
355 : :
356 : : /*
357 : : * Return true if the replication slot name is "pg_conflict_detection".
358 : : */
359 : : static bool
360 : 2445 : IsSlotForConflictCheck(const char *name)
361 : : {
362 : 2445 : return (strcmp(name, CONFLICT_DETECTION_SLOT) == 0);
363 : : }
364 : :
365 : : /*
366 : : * Create a new replication slot and mark it as used by this backend.
367 : : *
368 : : * name: Name of the slot
369 : : * db_specific: logical decoding is db specific; if the slot is going to
370 : : * be used for that pass true, otherwise false.
371 : : * two_phase: If enabled, allows decoding of prepared transactions.
372 : : * repack: If true, use a slot from the pool for REPACK.
373 : : * failover: If enabled, allows the slot to be synced to standbys so
374 : : * that logical replication can be resumed after failover.
375 : : * synced: True if the slot is synchronized from the primary server.
376 : : */
377 : : void
378 : 751 : ReplicationSlotCreate(const char *name, bool db_specific,
379 : : ReplicationSlotPersistency persistency,
380 : : bool two_phase, bool repack, bool failover, bool synced)
381 : : {
382 : 751 : ReplicationSlot *slot = NULL;
383 : : int startpoint,
384 : : endpoint;
385 : :
386 : : Assert(MyReplicationSlot == NULL);
387 : :
388 : : /*
389 : : * The logical launcher or pg_upgrade may create or migrate an internal
390 : : * slot, so using a reserved name is allowed in these cases.
391 : : */
392 [ + + + + ]: 751 : ReplicationSlotValidateName(name, IsBinaryUpgrade || IsLogicalLauncher(),
393 : 751 : ERROR);
394 : :
395 [ + + ]: 750 : if (failover)
396 : : {
397 : : /*
398 : : * Do not allow users to create the failover enabled slots on the
399 : : * standby as we do not support sync to the cascading standby.
400 : : *
401 : : * However, failover enabled slots can be created during slot
402 : : * synchronization because we need to retain the same values as the
403 : : * remote slot.
404 : : */
405 [ + + - + ]: 33 : if (RecoveryInProgress() && !IsSyncingReplicationSlots())
406 [ # # ]: 0 : ereport(ERROR,
407 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
408 : : errmsg("cannot enable failover for a replication slot created on the standby"));
409 : :
410 : : /*
411 : : * Do not allow users to create failover enabled temporary slots,
412 : : * because temporary slots will not be synced to the standby.
413 : : *
414 : : * However, failover enabled temporary slots can be created during
415 : : * slot synchronization. See the comments atop slotsync.c for details.
416 : : */
417 [ + + + + ]: 33 : if (persistency == RS_TEMPORARY && !IsSyncingReplicationSlots())
418 [ + - ]: 1 : ereport(ERROR,
419 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
420 : : errmsg("cannot enable failover for a temporary replication slot"));
421 : : }
422 : :
423 : 749 : INJECTION_POINT("replication-slot-create-begin", NULL);
424 : :
425 : : /*
426 : : * If some other backend ran this code concurrently with us, we'd likely
427 : : * both allocate the same slot, and that would be bad. We'd also be at
428 : : * risk of missing a name collision. Also, we don't want to try to create
429 : : * a new slot while somebody's busy cleaning up an old one, because we
430 : : * might both be monkeying with the same directory.
431 : : */
432 : 749 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
433 : :
434 : : /*
435 : : * Check for name collision (across the whole array), and identify an
436 : : * allocatable slot (in the array slice specific to our current use case:
437 : : * either general, or REPACK only). We need to hold
438 : : * ReplicationSlotControlLock in shared mode for this, so that nobody else
439 : : * can change the in_use flags while we're looking at them.
440 : : */
441 : 749 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
442 [ + + ]: 749 : startpoint = !repack ? 0 : max_replication_slots;
443 [ + + ]: 749 : endpoint = max_replication_slots + (repack ? max_repack_replication_slots : 0);
444 [ + + ]: 11015 : for (int i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
445 : : {
446 : 10269 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
447 : :
448 [ + + + + ]: 10269 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
449 [ + - ]: 3 : ereport(ERROR,
450 : : (errcode(ERRCODE_DUPLICATE_OBJECT),
451 : : errmsg("replication slot \"%s\" already exists", name)));
452 : :
453 [ + + + + ]: 10266 : if (i >= startpoint && i < endpoint &&
454 [ + + + + ]: 6513 : !s->in_use && slot == NULL)
455 : 745 : slot = s;
456 : : }
457 : 746 : LWLockRelease(ReplicationSlotControlLock);
458 : :
459 : : /* If all slots are in use, we're out of luck. */
460 [ + + ]: 746 : if (slot == NULL)
461 [ + - - + ]: 1 : ereport(ERROR,
462 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
463 : : errmsg("all replication slots are in use"),
464 : : errhint("Free one or increase \"%s\".",
465 : : repack ? "max_repack_replication_slots" : "max_replication_slots")));
466 : :
467 : : /*
468 : : * Since this slot is not in use, nobody should be looking at any part of
469 : : * it other than the in_use field unless they're trying to allocate it.
470 : : * And since we hold ReplicationSlotAllocationLock, nobody except us can
471 : : * be doing that. So it's safe to initialize the slot.
472 : : */
473 : : Assert(!slot->in_use);
474 : : Assert(slot->active_proc == INVALID_PROC_NUMBER);
475 : :
476 : : /* first initialize persistent data */
477 : 745 : memset(&slot->data, 0, sizeof(ReplicationSlotPersistentData));
478 : 745 : namestrcpy(&slot->data.name, name);
479 [ + + ]: 745 : slot->data.database = db_specific ? MyDatabaseId : InvalidOid;
480 : 745 : slot->data.persistency = persistency;
481 : 745 : slot->data.two_phase = two_phase;
482 : 745 : slot->data.two_phase_at = InvalidXLogRecPtr;
483 : 745 : slot->data.failover = failover;
484 : 745 : slot->data.synced = synced;
485 : :
486 : : /* and then data only present in shared memory */
487 : 745 : slot->just_dirtied = false;
488 : 745 : slot->dirty = false;
489 : 745 : slot->effective_xmin = InvalidTransactionId;
490 : 745 : slot->effective_catalog_xmin = InvalidTransactionId;
491 : 745 : slot->candidate_catalog_xmin = InvalidTransactionId;
492 : 745 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
493 : 745 : slot->candidate_restart_valid = InvalidXLogRecPtr;
494 : 745 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
495 : 745 : slot->last_saved_confirmed_flush = InvalidXLogRecPtr;
496 : 745 : slot->last_saved_restart_lsn = InvalidXLogRecPtr;
497 : 745 : slot->inactive_since = 0;
498 : 745 : slot->slotsync_skip_reason = SS_SKIP_NONE;
499 : :
500 : : /*
501 : : * Create the slot on disk. We haven't actually marked the slot allocated
502 : : * yet, so no special cleanup is required if this errors out.
503 : : */
504 : 745 : CreateSlotOnDisk(slot);
505 : :
506 : : /*
507 : : * We need to briefly prevent any other backend from iterating over the
508 : : * slots while we flip the in_use flag. We also need to set the active
509 : : * flag while holding the ControlLock as otherwise a concurrent
510 : : * ReplicationSlotAcquire() could acquire the slot as well.
511 : : */
512 : 745 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
513 : :
514 : 745 : slot->in_use = true;
515 : :
516 : : /* We can now mark the slot active, and that makes it our slot. */
517 : 745 : SpinLockAcquire(&slot->mutex);
518 : : Assert(slot->active_proc == INVALID_PROC_NUMBER);
519 : 745 : slot->active_proc = MyProcNumber;
520 : 745 : SpinLockRelease(&slot->mutex);
521 : 745 : MyReplicationSlot = slot;
522 : :
523 : 745 : LWLockRelease(ReplicationSlotControlLock);
524 : :
525 : : /*
526 : : * Create statistics entry for the new logical slot. We don't collect any
527 : : * stats for physical slots, so no need to create an entry for the same.
528 : : * See ReplicationSlotDropPtr for why we need to do this before releasing
529 : : * ReplicationSlotAllocationLock.
530 : : */
531 [ + + ]: 745 : if (SlotIsLogical(slot))
532 : 533 : pgstat_create_replslot(slot);
533 : :
534 : : /*
535 : : * Now that the slot has been marked as in_use and active, it's safe to
536 : : * let somebody else try to allocate a slot.
537 : : */
538 : 745 : LWLockRelease(ReplicationSlotAllocationLock);
539 : :
540 : : /* Let everybody know we've modified this slot */
541 : 745 : ConditionVariableBroadcast(&slot->active_cv);
542 : 745 : }
543 : :
544 : : /*
545 : : * Search for the named replication slot.
546 : : *
547 : : * Return the replication slot if found, otherwise NULL.
548 : : */
549 : : ReplicationSlot *
550 : 2228 : SearchNamedReplicationSlot(const char *name, bool need_lock)
551 : : {
552 : : int i;
553 : 2228 : ReplicationSlot *slot = NULL;
554 : :
555 [ + + ]: 2228 : if (need_lock)
556 : 670 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
557 : :
558 [ + + ]: 11361 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
559 : : {
560 : 10818 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
561 : :
562 [ + + + + ]: 10818 : if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
563 : : {
564 : 1685 : slot = s;
565 : 1685 : break;
566 : : }
567 : : }
568 : :
569 [ + + ]: 2228 : if (need_lock)
570 : 670 : LWLockRelease(ReplicationSlotControlLock);
571 : :
572 : 2228 : return slot;
573 : : }
574 : :
575 : : /*
576 : : * Return the index of the replication slot in
577 : : * ReplicationSlotCtl->replication_slots.
578 : : *
579 : : * This is mainly useful to have an efficient key for storing replication slot
580 : : * stats.
581 : : */
582 : : int
583 : 8632 : ReplicationSlotIndex(ReplicationSlot *slot)
584 : : {
585 : : Assert(slot >= ReplicationSlotCtl->replication_slots &&
586 : : slot < ReplicationSlotCtl->replication_slots +
587 : : (max_replication_slots + max_repack_replication_slots));
588 : :
589 : 8632 : return slot - ReplicationSlotCtl->replication_slots;
590 : : }
591 : :
592 : : /*
593 : : * If the slot at 'index' is unused, return false. Otherwise 'name' is set to
594 : : * the slot's name and true is returned.
595 : : *
596 : : * This likely is only useful for pgstat_replslot.c during shutdown, in other
597 : : * cases there are obvious TOCTOU issues.
598 : : */
599 : : bool
600 : 120 : ReplicationSlotName(int index, Name name)
601 : : {
602 : : ReplicationSlot *slot;
603 : : bool found;
604 : :
605 : 120 : slot = &ReplicationSlotCtl->replication_slots[index];
606 : :
607 : : /*
608 : : * Ensure that the slot cannot be dropped while we copy the name. Don't
609 : : * need the spinlock as the name of an existing slot cannot change.
610 : : */
611 : 120 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
612 : 120 : found = slot->in_use;
613 [ + - ]: 120 : if (slot->in_use)
614 : 120 : namestrcpy(name, NameStr(slot->data.name));
615 : 120 : LWLockRelease(ReplicationSlotControlLock);
616 : :
617 : 120 : return found;
618 : : }
619 : :
620 : : /*
621 : : * Find a previously created slot and mark it as used by this process.
622 : : *
623 : : * An error is raised if nowait is true and the slot is currently in use. If
624 : : * nowait is false, we sleep until the slot is released by the owning process.
625 : : *
626 : : * An error is raised if error_if_invalid is true and the slot is found to
627 : : * be invalid. It should always be set to true, except when we are temporarily
628 : : * acquiring the slot and don't intend to change it.
629 : : */
630 : : void
631 : 1470 : ReplicationSlotAcquire(const char *name, bool nowait, bool error_if_invalid)
632 : : {
633 : : ReplicationSlot *s;
634 : : ProcNumber active_proc;
635 : : int active_pid;
636 : :
637 : : Assert(name != NULL);
638 : :
639 : 1470 : retry:
640 : : Assert(MyReplicationSlot == NULL);
641 : :
642 : 1470 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
643 : :
644 : : /* Check if the slot exists with the given name. */
645 : 1470 : s = SearchNamedReplicationSlot(name, false);
646 [ + + - + ]: 1470 : if (s == NULL || !s->in_use)
647 : : {
648 : 10 : LWLockRelease(ReplicationSlotControlLock);
649 : :
650 [ + - ]: 10 : ereport(ERROR,
651 : : (errcode(ERRCODE_UNDEFINED_OBJECT),
652 : : errmsg("replication slot \"%s\" does not exist",
653 : : name)));
654 : : }
655 : :
656 : : /*
657 : : * Do not allow users to acquire the reserved slot. This scenario may
658 : : * occur if the launcher that owns the slot has terminated unexpectedly
659 : : * due to an error, and a backend process attempts to reuse the slot.
660 : : */
661 [ + + - + ]: 1460 : if (!IsLogicalLauncher() && IsSlotForConflictCheck(name))
662 [ # # ]: 0 : ereport(ERROR,
663 : : errcode(ERRCODE_UNDEFINED_OBJECT),
664 : : errmsg("cannot acquire replication slot \"%s\"", name),
665 : : errdetail("The slot is reserved for conflict detection and can only be acquired by logical replication launcher."));
666 : :
667 : : /*
668 : : * This is the slot we want; check if it's active under some other
669 : : * process. In single user mode, we don't need this check.
670 : : */
671 [ + + ]: 1460 : if (IsUnderPostmaster)
672 : : {
673 : : /*
674 : : * Get ready to sleep on the slot in case it is active. (We may end
675 : : * up not sleeping, but we don't want to do this while holding the
676 : : * spinlock.)
677 : : */
678 [ + + ]: 1455 : if (!nowait)
679 : 297 : ConditionVariablePrepareToSleep(&s->active_cv);
680 : :
681 : : /*
682 : : * It is important to reset the inactive_since under spinlock here to
683 : : * avoid race conditions with slot invalidation. See comments related
684 : : * to inactive_since in InvalidatePossiblyObsoleteSlot.
685 : : */
686 : 1455 : SpinLockAcquire(&s->mutex);
687 [ + + ]: 1455 : if (s->active_proc == INVALID_PROC_NUMBER)
688 : 1288 : s->active_proc = MyProcNumber;
689 : 1455 : active_proc = s->active_proc;
690 : 1455 : ReplicationSlotSetInactiveSince(s, 0, false);
691 : 1455 : SpinLockRelease(&s->mutex);
692 : : }
693 : : else
694 : : {
695 : 5 : s->active_proc = active_proc = MyProcNumber;
696 : 5 : ReplicationSlotSetInactiveSince(s, 0, true);
697 : : }
698 : 1460 : active_pid = GetPGProcByNumber(active_proc)->pid;
699 : 1460 : LWLockRelease(ReplicationSlotControlLock);
700 : :
701 : : /*
702 : : * If we found the slot but it's already active in another process, we
703 : : * wait until the owning process signals us that it's been released, or
704 : : * error out.
705 : : */
706 [ - + ]: 1460 : if (active_proc != MyProcNumber)
707 : : {
708 [ # # ]: 0 : if (!nowait)
709 : : {
710 : : /* Wait here until we get signaled, and then restart */
711 : 0 : ConditionVariableSleep(&s->active_cv,
712 : : WAIT_EVENT_REPLICATION_SLOT_DROP);
713 : 0 : ConditionVariableCancelSleep();
714 : 0 : goto retry;
715 : : }
716 : :
717 [ # # ]: 0 : ereport(ERROR,
718 : : (errcode(ERRCODE_OBJECT_IN_USE),
719 : : errmsg("replication slot \"%s\" is active for PID %d",
720 : : NameStr(s->data.name), active_pid)));
721 : : }
722 [ + + ]: 1460 : else if (!nowait)
723 : 297 : ConditionVariableCancelSleep(); /* no sleep needed after all */
724 : :
725 : : /* We made this slot active, so it's ours now. */
726 : 1460 : MyReplicationSlot = s;
727 : :
728 : : /*
729 : : * We need to check for invalidation after making the slot ours to avoid
730 : : * the possible race condition with the checkpointer that can otherwise
731 : : * invalidate the slot immediately after the check.
732 : : */
733 [ + + + + ]: 1460 : if (error_if_invalid && s->data.invalidated != RS_INVAL_NONE)
734 [ + - ]: 8 : ereport(ERROR,
735 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
736 : : errmsg("can no longer access replication slot \"%s\"",
737 : : NameStr(s->data.name)),
738 : : errdetail("This replication slot has been invalidated due to \"%s\".",
739 : : GetSlotInvalidationCauseName(s->data.invalidated)));
740 : :
741 : : /* Let everybody know we've modified this slot */
742 : 1452 : ConditionVariableBroadcast(&s->active_cv);
743 : :
744 : : /*
745 : : * The call to pgstat_acquire_replslot() protects against stats for a
746 : : * different slot, from before a restart or such, being present during
747 : : * pgstat_report_replslot().
748 : : */
749 [ + + ]: 1452 : if (SlotIsLogical(s))
750 : 1211 : pgstat_acquire_replslot(s);
751 : :
752 : :
753 [ + + ]: 1452 : if (am_walsender)
754 : : {
755 [ + - + - : 998 : ereport(log_replication_commands ? LOG : DEBUG1,
+ + ]
756 : : SlotIsLogical(s)
757 : : ? errmsg("acquired logical replication slot \"%s\"",
758 : : NameStr(s->data.name))
759 : : : errmsg("acquired physical replication slot \"%s\"",
760 : : NameStr(s->data.name)));
761 : : }
762 : 1452 : }
763 : :
764 : : /*
765 : : * Release the replication slot that this backend considers to own.
766 : : *
767 : : * This or another backend can re-acquire the slot later.
768 : : * Resources this slot requires will be preserved.
769 : : */
770 : : void
771 : 1767 : ReplicationSlotRelease(void)
772 : : {
773 : 1767 : ReplicationSlot *slot = MyReplicationSlot;
774 : 1767 : char *slotname = NULL; /* keep compiler quiet */
775 : : bool is_logical;
776 : 1767 : TimestampTz now = 0;
777 : :
778 : : Assert(slot != NULL && slot->active_proc != INVALID_PROC_NUMBER);
779 : :
780 : 1767 : is_logical = SlotIsLogical(slot);
781 : :
782 [ + + ]: 1767 : if (am_walsender)
783 : 1236 : slotname = pstrdup(NameStr(slot->data.name));
784 : :
785 [ + + ]: 1767 : if (slot->data.persistency == RS_EPHEMERAL)
786 : : {
787 : : /*
788 : : * If slot is ephemeral, we drop it upon release, and request logical
789 : : * decoding be disabled.
790 : : */
791 : 12 : ReplicationSlotDropAcquired(is_logical);
792 : : }
793 : : else
794 : : {
795 : : /*
796 : : * If slot needed to temporarily restrain both data and catalog xmin
797 : : * to create the catalog snapshot, remove that temporary constraint.
798 : : * Snapshots can only be exported while the initial snapshot is still
799 : : * acquired.
800 : : */
801 [ + + ]: 1755 : if (!TransactionIdIsValid(slot->data.xmin) &&
802 [ + + ]: 1719 : TransactionIdIsValid(slot->effective_xmin))
803 : : {
804 : 215 : SpinLockAcquire(&slot->mutex);
805 : 215 : slot->effective_xmin = InvalidTransactionId;
806 : 215 : SpinLockRelease(&slot->mutex);
807 : 215 : ReplicationSlotsComputeRequiredXmin(false);
808 : : }
809 : :
810 : : /*
811 : : * Set the time since the slot has become inactive. We get the current
812 : : * time beforehand to avoid system call while holding the spinlock.
813 : : */
814 : 1755 : now = GetCurrentTimestamp();
815 : :
816 [ + + ]: 1755 : if (slot->data.persistency == RS_PERSISTENT)
817 : : {
818 : : /*
819 : : * Mark persistent slot inactive. We're not freeing it, just
820 : : * disconnecting, but wake up others that may be waiting for it.
821 : : */
822 : 1421 : SpinLockAcquire(&slot->mutex);
823 : 1421 : slot->active_proc = INVALID_PROC_NUMBER;
824 : 1421 : ReplicationSlotSetInactiveSince(slot, now, false);
825 : 1421 : SpinLockRelease(&slot->mutex);
826 : 1421 : ConditionVariableBroadcast(&slot->active_cv);
827 : : }
828 : : else
829 : 334 : ReplicationSlotSetInactiveSince(slot, now, true);
830 : :
831 : 1755 : MyReplicationSlot = NULL;
832 : : }
833 : :
834 : : /* might not have been set when we've been a plain slot */
835 : 1767 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
836 : 1767 : MyProc->statusFlags &= ~PROC_IN_LOGICAL_DECODING;
837 : 1767 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
838 : 1767 : LWLockRelease(ProcArrayLock);
839 : :
840 [ + + ]: 1767 : if (am_walsender)
841 : : {
842 [ + - + - : 1236 : ereport(log_replication_commands ? LOG : DEBUG1,
+ + ]
843 : : is_logical
844 : : ? errmsg("released logical replication slot \"%s\"",
845 : : slotname)
846 : : : errmsg("released physical replication slot \"%s\"",
847 : : slotname));
848 : :
849 : 1236 : pfree(slotname);
850 : : }
851 : 1767 : }
852 : :
853 : : /*
854 : : * Cleanup temporary slots created in current session.
855 : : *
856 : : * Cleanup only synced temporary slots if 'synced_only' is true, else
857 : : * cleanup all temporary slots.
858 : : *
859 : : * If it drops the last logical slot in the cluster, requests to disable
860 : : * logical decoding.
861 : : */
862 : : void
863 : 57134 : ReplicationSlotCleanup(bool synced_only)
864 : : {
865 : : int i;
866 : : bool found_valid_logicalslot;
867 : 57134 : bool dropped_logical = false;
868 : :
869 : : Assert(MyReplicationSlot == NULL);
870 : :
871 : 57301 : restart:
872 : 57301 : found_valid_logicalslot = false;
873 : 57301 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
874 [ + + ]: 908349 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
875 : : {
876 : 851215 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
877 : :
878 [ + + ]: 851215 : if (!s->in_use)
879 : 835422 : continue;
880 : :
881 : 15793 : SpinLockAcquire(&s->mutex);
882 : :
883 : 31586 : found_valid_logicalslot |=
884 [ + + + + ]: 15793 : (SlotIsLogical(s) && s->data.invalidated == RS_INVAL_NONE);
885 : :
886 [ + + ]: 15793 : if ((s->active_proc == MyProcNumber &&
887 [ - + - - ]: 167 : (!synced_only || s->data.synced)))
888 : : {
889 : : Assert(s->data.persistency == RS_TEMPORARY);
890 : 167 : SpinLockRelease(&s->mutex);
891 : 167 : LWLockRelease(ReplicationSlotControlLock); /* avoid deadlock */
892 : :
893 [ + + ]: 167 : if (SlotIsLogical(s))
894 : 10 : dropped_logical = true;
895 : :
896 : 167 : ReplicationSlotDropPtr(s);
897 : :
898 : 167 : ConditionVariableBroadcast(&s->active_cv);
899 : 167 : goto restart;
900 : : }
901 : : else
902 : 15626 : SpinLockRelease(&s->mutex);
903 : : }
904 : :
905 : 57134 : LWLockRelease(ReplicationSlotControlLock);
906 : :
907 [ + + + + ]: 57134 : if (dropped_logical && !found_valid_logicalslot)
908 : 3 : RequestDisableLogicalDecoding();
909 : 57134 : }
910 : :
911 : : /*
912 : : * Permanently drop the replication slot identified by the passed-in name.
913 : : *
914 : : * If this is a logical slot, request that logical decoding be disabled.
915 : : */
916 : : void
917 : 451 : ReplicationSlotDrop(const char *name, bool nowait)
918 : : {
919 : 451 : ReplicationSlotAcquire(name, nowait, false);
920 : :
921 : : /*
922 : : * Do not allow users to drop the slots which are currently being synced
923 : : * from the primary to the standby.
924 : : */
925 [ + + + + ]: 444 : if (RecoveryInProgress() && MyReplicationSlot->data.synced)
926 [ + - ]: 1 : ereport(ERROR,
927 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
928 : : errmsg("cannot drop replication slot \"%s\"", name),
929 : : errdetail("This replication slot is being synchronized from the primary server."));
930 : :
931 : 443 : ReplicationSlotDropAcquired(SlotIsLogical(MyReplicationSlot));
932 : 443 : }
933 : :
934 : : /*
935 : : * Change the definition of the slot identified by the specified name.
936 : : *
937 : : * Altering the two_phase property of a slot requires caution on the
938 : : * client-side. Enabling it at any random point during decoding has the
939 : : * risk that transactions prepared before this change may be skipped by
940 : : * the decoder, leading to missing prepare records on the client. So, we
941 : : * enable it for subscription related slots only once the initial tablesync
942 : : * is finished. See comments atop worker.c. Disabling it is safe only when
943 : : * there are no pending prepared transaction, otherwise, the changes of
944 : : * already prepared transactions can be replicated again along with their
945 : : * corresponding commit leading to duplicate data or errors.
946 : : */
947 : : void
948 : 7 : ReplicationSlotAlter(const char *name, const bool *failover,
949 : : const bool *two_phase)
950 : : {
951 : 7 : bool update_slot = false;
952 : :
953 : : Assert(MyReplicationSlot == NULL);
954 : : Assert(failover || two_phase);
955 : :
956 : 7 : ReplicationSlotAcquire(name, false, true);
957 : :
958 [ - + ]: 6 : if (SlotIsPhysical(MyReplicationSlot))
959 [ # # ]: 0 : ereport(ERROR,
960 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
961 : : errmsg("cannot use %s with a physical replication slot",
962 : : "ALTER_REPLICATION_SLOT"));
963 : :
964 [ + + ]: 6 : if (RecoveryInProgress())
965 : : {
966 : : /*
967 : : * Do not allow users to alter the slots which are currently being
968 : : * synced from the primary to the standby.
969 : : */
970 [ + - ]: 1 : if (MyReplicationSlot->data.synced)
971 [ + - ]: 1 : ereport(ERROR,
972 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
973 : : errmsg("cannot alter replication slot \"%s\"", name),
974 : : errdetail("This replication slot is being synchronized from the primary server."));
975 : :
976 : : /*
977 : : * Do not allow users to enable failover on the standby as we do not
978 : : * support sync to the cascading standby.
979 : : */
980 [ # # # # ]: 0 : if (failover && *failover)
981 [ # # ]: 0 : ereport(ERROR,
982 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
983 : : errmsg("cannot enable failover for a replication slot"
984 : : " on the standby"));
985 : : }
986 : :
987 [ + + ]: 5 : if (failover)
988 : : {
989 : : /*
990 : : * Do not allow users to enable failover for temporary slots as we do
991 : : * not support syncing temporary slots to the standby.
992 : : */
993 [ + + - + ]: 4 : if (*failover && MyReplicationSlot->data.persistency == RS_TEMPORARY)
994 [ # # ]: 0 : ereport(ERROR,
995 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
996 : : errmsg("cannot enable failover for a temporary replication slot"));
997 : :
998 [ + - ]: 4 : if (MyReplicationSlot->data.failover != *failover)
999 : : {
1000 : 4 : SpinLockAcquire(&MyReplicationSlot->mutex);
1001 : 4 : MyReplicationSlot->data.failover = *failover;
1002 : 4 : SpinLockRelease(&MyReplicationSlot->mutex);
1003 : :
1004 : 4 : update_slot = true;
1005 : : }
1006 : : }
1007 : :
1008 [ + + + - ]: 5 : if (two_phase && MyReplicationSlot->data.two_phase != *two_phase)
1009 : : {
1010 : 1 : SpinLockAcquire(&MyReplicationSlot->mutex);
1011 : 1 : MyReplicationSlot->data.two_phase = *two_phase;
1012 : 1 : SpinLockRelease(&MyReplicationSlot->mutex);
1013 : :
1014 : 1 : update_slot = true;
1015 : : }
1016 : :
1017 [ + - ]: 5 : if (update_slot)
1018 : : {
1019 : 5 : ReplicationSlotMarkDirty();
1020 : 5 : ReplicationSlotSave();
1021 : : }
1022 : :
1023 : 5 : ReplicationSlotRelease();
1024 : 5 : }
1025 : :
1026 : : /*
1027 : : * Permanently drop the currently acquired replication slot.
1028 : : *
1029 : : * If caller requests it, have checkpointer attempt to disable logical
1030 : : * decoding. Obviously, this should only be done if the slot is logical.
1031 : : */
1032 : : void
1033 : 471 : ReplicationSlotDropAcquired(bool try_disable)
1034 : : {
1035 : : ReplicationSlot *slot;
1036 : :
1037 : : Assert(MyReplicationSlot != NULL);
1038 : 471 : slot = MyReplicationSlot;
1039 : :
1040 : : /* Can only disable logical decoding if slot is logical */
1041 : : Assert(!try_disable || SlotIsLogical(slot));
1042 : :
1043 : : /* slot isn't acquired anymore */
1044 : 471 : MyReplicationSlot = NULL;
1045 : :
1046 : 471 : ReplicationSlotDropPtr(slot);
1047 : :
1048 [ + + ]: 471 : if (try_disable)
1049 : 442 : RequestDisableLogicalDecoding();
1050 : 471 : }
1051 : :
1052 : : /*
1053 : : * Permanently drop the replication slot which will be released by the point
1054 : : * this function returns.
1055 : : */
1056 : : static void
1057 : 638 : ReplicationSlotDropPtr(ReplicationSlot *slot)
1058 : : {
1059 : : char path[MAXPGPATH];
1060 : : char tmppath[MAXPGPATH];
1061 : :
1062 : : /*
1063 : : * If some other backend ran this code concurrently with us, we might try
1064 : : * to delete a slot with a certain name while someone else was trying to
1065 : : * create a slot with the same name.
1066 : : */
1067 : 638 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
1068 : :
1069 : : /* Generate pathnames. */
1070 : 638 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
1071 : 638 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
1072 : :
1073 : : /*
1074 : : * Rename the slot directory on disk, so that we'll no longer recognize
1075 : : * this as a valid slot. Note that if this fails, we've got to mark the
1076 : : * slot inactive before bailing out. If we're dropping an ephemeral or a
1077 : : * temporary slot, we better never fail hard as the caller won't expect
1078 : : * the slot to survive and this might get called during error handling.
1079 : : */
1080 [ + - ]: 638 : if (rename(path, tmppath) == 0)
1081 : : {
1082 : : /*
1083 : : * We need to fsync() the directory we just renamed and its parent to
1084 : : * make sure that our changes are on disk in a crash-safe fashion. If
1085 : : * fsync() fails, we can't be sure whether the changes are on disk or
1086 : : * not. For now, we handle that by panicking;
1087 : : * StartupReplicationSlots() will try to straighten it out after
1088 : : * restart.
1089 : : */
1090 : 638 : START_CRIT_SECTION();
1091 : 638 : fsync_fname(tmppath, true);
1092 : 638 : fsync_fname(PG_REPLSLOT_DIR, true);
1093 : 638 : END_CRIT_SECTION();
1094 : : }
1095 : : else
1096 : : {
1097 : 0 : bool fail_softly = slot->data.persistency != RS_PERSISTENT;
1098 : :
1099 : 0 : SpinLockAcquire(&slot->mutex);
1100 : 0 : slot->active_proc = INVALID_PROC_NUMBER;
1101 : 0 : SpinLockRelease(&slot->mutex);
1102 : :
1103 : : /* wake up anyone waiting on this slot */
1104 : 0 : ConditionVariableBroadcast(&slot->active_cv);
1105 : :
1106 [ # # # # ]: 0 : ereport(fail_softly ? WARNING : ERROR,
1107 : : (errcode_for_file_access(),
1108 : : errmsg("could not rename file \"%s\" to \"%s\": %m",
1109 : : path, tmppath)));
1110 : : }
1111 : :
1112 : : /*
1113 : : * The slot is definitely gone. Lock out concurrent scans of the array
1114 : : * long enough to kill it. It's OK to clear the active PID here without
1115 : : * grabbing the mutex because nobody else can be scanning the array here,
1116 : : * and nobody can be attached to this slot and thus access it without
1117 : : * scanning the array.
1118 : : *
1119 : : * Also wake up processes waiting for it.
1120 : : */
1121 : 638 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
1122 : 638 : slot->active_proc = INVALID_PROC_NUMBER;
1123 : 638 : slot->in_use = false;
1124 : 638 : LWLockRelease(ReplicationSlotControlLock);
1125 : 638 : ConditionVariableBroadcast(&slot->active_cv);
1126 : :
1127 : : /*
1128 : : * Slot is dead and doesn't prevent resource removal anymore, recompute
1129 : : * limits.
1130 : : */
1131 : 638 : ReplicationSlotsComputeRequiredXmin(false);
1132 : 638 : ReplicationSlotsComputeRequiredLSN();
1133 : :
1134 : : /*
1135 : : * If removing the directory fails, the worst thing that will happen is
1136 : : * that the user won't be able to create a new slot with the same name
1137 : : * until the next server restart. We warn about it, but that's all.
1138 : : */
1139 [ - + ]: 638 : if (!rmtree(tmppath, true))
1140 [ # # ]: 0 : ereport(WARNING,
1141 : : (errmsg("could not remove directory \"%s\"", tmppath)));
1142 : :
1143 : : /*
1144 : : * Drop the statistics entry for the replication slot. Do this while
1145 : : * holding ReplicationSlotAllocationLock so that we don't drop a
1146 : : * statistics entry for another slot with the same name just created in
1147 : : * another session.
1148 : : */
1149 [ + + ]: 638 : if (SlotIsLogical(slot))
1150 : 460 : pgstat_drop_replslot(slot);
1151 : :
1152 : : /*
1153 : : * We release this at the very end, so that nobody starts trying to create
1154 : : * a slot while we're still cleaning up the detritus of the old one.
1155 : : */
1156 : 638 : LWLockRelease(ReplicationSlotAllocationLock);
1157 : 638 : }
1158 : :
1159 : : /*
1160 : : * Serialize the currently acquired slot's state from memory to disk, thereby
1161 : : * guaranteeing the current state will survive a crash.
1162 : : */
1163 : : void
1164 : 1538 : ReplicationSlotSave(void)
1165 : : {
1166 : : char path[MAXPGPATH];
1167 : :
1168 : : Assert(MyReplicationSlot != NULL);
1169 : :
1170 : 1538 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(MyReplicationSlot->data.name));
1171 : 1538 : SaveSlotToPath(MyReplicationSlot, path, ERROR);
1172 : 1538 : }
1173 : :
1174 : : /*
1175 : : * Signal that it would be useful if the currently acquired slot would be
1176 : : * flushed out to disk.
1177 : : *
1178 : : * Note that the actual flush to disk can be delayed for a long time, if
1179 : : * required for correctness explicitly do a ReplicationSlotSave().
1180 : : */
1181 : : void
1182 : 42461 : ReplicationSlotMarkDirty(void)
1183 : : {
1184 : 42461 : ReplicationSlot *slot = MyReplicationSlot;
1185 : :
1186 : : Assert(MyReplicationSlot != NULL);
1187 : :
1188 : 42461 : SpinLockAcquire(&slot->mutex);
1189 : 42461 : MyReplicationSlot->just_dirtied = true;
1190 : 42461 : MyReplicationSlot->dirty = true;
1191 : 42461 : SpinLockRelease(&slot->mutex);
1192 : 42461 : }
1193 : :
1194 : : /*
1195 : : * Convert a slot that's marked as RS_EPHEMERAL or RS_TEMPORARY to a
1196 : : * RS_PERSISTENT slot, guaranteeing it will be there after an eventual crash.
1197 : : */
1198 : : void
1199 : 503 : ReplicationSlotPersist(void)
1200 : : {
1201 : 503 : ReplicationSlot *slot = MyReplicationSlot;
1202 : :
1203 : : Assert(slot != NULL);
1204 : : Assert(slot->data.persistency != RS_PERSISTENT);
1205 : :
1206 : 503 : SpinLockAcquire(&slot->mutex);
1207 : 503 : slot->data.persistency = RS_PERSISTENT;
1208 : 503 : SpinLockRelease(&slot->mutex);
1209 : :
1210 : 503 : ReplicationSlotMarkDirty();
1211 : 503 : ReplicationSlotSave();
1212 : 503 : }
1213 : :
1214 : : /*
1215 : : * Compute the oldest xmin across all slots and store it in the ProcArray.
1216 : : *
1217 : : * If already_locked is true, both the ReplicationSlotControlLock and the
1218 : : * ProcArrayLock have already been acquired exclusively. It is crucial that the
1219 : : * caller first acquires the ReplicationSlotControlLock, followed by the
1220 : : * ProcArrayLock, to prevent any undetectable deadlocks since this function
1221 : : * acquires them in that order.
1222 : : */
1223 : : void
1224 : 2707 : ReplicationSlotsComputeRequiredXmin(bool already_locked)
1225 : : {
1226 : : int i;
1227 : 2707 : TransactionId agg_xmin = InvalidTransactionId;
1228 : 2707 : TransactionId agg_catalog_xmin = InvalidTransactionId;
1229 : :
1230 : : Assert(ReplicationSlotCtl != NULL);
1231 : : Assert(!already_locked ||
1232 : : (LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_EXCLUSIVE) &&
1233 : : LWLockHeldByMeInMode(ProcArrayLock, LW_EXCLUSIVE)));
1234 : :
1235 : : /*
1236 : : * Hold the ReplicationSlotControlLock until after updating the slot xmin
1237 : : * values, so no backend updates the initial xmin for newly created slot
1238 : : * concurrently. A shared lock is used here to minimize lock contention,
1239 : : * especially when many slots exist and advancements occur frequently.
1240 : : * This is safe since an exclusive lock is taken during initial slot xmin
1241 : : * update in slot creation.
1242 : : *
1243 : : * One might think that we can hold the ProcArrayLock exclusively and
1244 : : * update the slot xmin values, but it could increase lock contention on
1245 : : * the ProcArrayLock, which is not great since this function can be called
1246 : : * at non-negligible frequency.
1247 : : *
1248 : : * Concurrent invocation of this function may cause the computed slot xmin
1249 : : * to regress. However, this is harmless because tuples prior to the most
1250 : : * recent xmin are no longer useful once advancement occurs (see
1251 : : * LogicalConfirmReceivedLocation where the slot's xmin value is flushed
1252 : : * before updating the effective_xmin). Thus, such regression merely
1253 : : * prevents VACUUM from prematurely removing tuples without causing the
1254 : : * early deletion of required data.
1255 : : */
1256 [ + + ]: 2707 : if (!already_locked)
1257 : 2175 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1258 : :
1259 [ + + ]: 41126 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1260 : : {
1261 : 38419 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1262 : : TransactionId effective_xmin;
1263 : : TransactionId effective_catalog_xmin;
1264 : : bool invalidated;
1265 : :
1266 [ + + ]: 38419 : if (!s->in_use)
1267 : 35817 : continue;
1268 : :
1269 : 2602 : SpinLockAcquire(&s->mutex);
1270 : 2602 : effective_xmin = s->effective_xmin;
1271 : 2602 : effective_catalog_xmin = s->effective_catalog_xmin;
1272 : 2602 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1273 : 2602 : SpinLockRelease(&s->mutex);
1274 : :
1275 : : /* invalidated slots need not apply */
1276 [ + + ]: 2602 : if (invalidated)
1277 : 26 : continue;
1278 : :
1279 : : /* check the data xmin */
1280 [ + + + + ]: 2576 : if (TransactionIdIsValid(effective_xmin) &&
1281 [ - + ]: 10 : (!TransactionIdIsValid(agg_xmin) ||
1282 : 10 : TransactionIdPrecedes(effective_xmin, agg_xmin)))
1283 : 371 : agg_xmin = effective_xmin;
1284 : :
1285 : : /* check the catalog xmin */
1286 [ + + + + ]: 2576 : if (TransactionIdIsValid(effective_catalog_xmin) &&
1287 [ + + ]: 1076 : (!TransactionIdIsValid(agg_catalog_xmin) ||
1288 : 1076 : TransactionIdPrecedes(effective_catalog_xmin, agg_catalog_xmin)))
1289 : 1336 : agg_catalog_xmin = effective_catalog_xmin;
1290 : : }
1291 : :
1292 : 2707 : ProcArraySetReplicationSlotXmin(agg_xmin, agg_catalog_xmin, already_locked);
1293 : :
1294 [ + + ]: 2707 : if (!already_locked)
1295 : 2175 : LWLockRelease(ReplicationSlotControlLock);
1296 : 2707 : }
1297 : :
1298 : : /*
1299 : : * Compute the oldest restart LSN across all slots and inform xlog module.
1300 : : *
1301 : : * Note: while max_slot_wal_keep_size is theoretically relevant for this
1302 : : * purpose, we don't try to account for that, because this module doesn't
1303 : : * know what to compare against.
1304 : : */
1305 : : void
1306 : 43529 : ReplicationSlotsComputeRequiredLSN(void)
1307 : : {
1308 : : int i;
1309 : 43529 : XLogRecPtr min_required = InvalidXLogRecPtr;
1310 : :
1311 : : Assert(ReplicationSlotCtl != NULL);
1312 : :
1313 : 43529 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1314 [ + + ]: 692142 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1315 : : {
1316 : 648613 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1317 : : XLogRecPtr restart_lsn;
1318 : : XLogRecPtr last_saved_restart_lsn;
1319 : : bool invalidated;
1320 : : ReplicationSlotPersistency persistency;
1321 : :
1322 [ + + ]: 648613 : if (!s->in_use)
1323 : 604476 : continue;
1324 : :
1325 : 44137 : SpinLockAcquire(&s->mutex);
1326 : 44137 : persistency = s->data.persistency;
1327 : 44137 : restart_lsn = s->data.restart_lsn;
1328 : 44137 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1329 : 44137 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1330 : 44137 : SpinLockRelease(&s->mutex);
1331 : :
1332 : : /* invalidated slots need not apply */
1333 [ + + ]: 44137 : if (invalidated)
1334 : 27 : continue;
1335 : :
1336 : : /*
1337 : : * For persistent slot use last_saved_restart_lsn to compute the
1338 : : * oldest LSN for removal of WAL segments. The segments between
1339 : : * last_saved_restart_lsn and restart_lsn might be needed by a
1340 : : * persistent slot in the case of database crash. Non-persistent
1341 : : * slots can't survive the database crash, so we don't care about
1342 : : * last_saved_restart_lsn for them.
1343 : : */
1344 [ + + ]: 44110 : if (persistency == RS_PERSISTENT)
1345 : : {
1346 [ + + + + ]: 43029 : if (XLogRecPtrIsValid(last_saved_restart_lsn) &&
1347 : : restart_lsn > last_saved_restart_lsn)
1348 : : {
1349 : 39449 : restart_lsn = last_saved_restart_lsn;
1350 : : }
1351 : : }
1352 : :
1353 [ + + + + ]: 44110 : if (XLogRecPtrIsValid(restart_lsn) &&
1354 [ + + ]: 1791 : (!XLogRecPtrIsValid(min_required) ||
1355 : : restart_lsn < min_required))
1356 : 42517 : min_required = restart_lsn;
1357 : : }
1358 : 43529 : LWLockRelease(ReplicationSlotControlLock);
1359 : :
1360 : 43529 : XLogSetReplicationSlotMinimumLSN(min_required);
1361 : 43529 : }
1362 : :
1363 : : /*
1364 : : * Compute the oldest WAL LSN required by *logical* decoding slots..
1365 : : *
1366 : : * Returns InvalidXLogRecPtr if logical decoding is disabled or no logical
1367 : : * slots exist.
1368 : : *
1369 : : * NB: this returns a value >= ReplicationSlotsComputeRequiredLSN(), since it
1370 : : * ignores physical replication slots.
1371 : : *
1372 : : * The results aren't required frequently, so we don't maintain a precomputed
1373 : : * value like we do for ComputeRequiredLSN() and ComputeRequiredXmin().
1374 : : */
1375 : : XLogRecPtr
1376 : 3942 : ReplicationSlotsComputeLogicalRestartLSN(void)
1377 : : {
1378 : 3942 : XLogRecPtr result = InvalidXLogRecPtr;
1379 : : int i;
1380 : :
1381 [ - + ]: 3942 : if (max_replication_slots + max_repack_replication_slots <= 0)
1382 : 0 : return InvalidXLogRecPtr;
1383 : :
1384 : 3942 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1385 : :
1386 [ + + ]: 62496 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1387 : : {
1388 : : ReplicationSlot *s;
1389 : : XLogRecPtr restart_lsn;
1390 : : XLogRecPtr last_saved_restart_lsn;
1391 : : bool invalidated;
1392 : : ReplicationSlotPersistency persistency;
1393 : :
1394 : 58554 : s = &ReplicationSlotCtl->replication_slots[i];
1395 : :
1396 : : /* cannot change while ReplicationSlotCtlLock is held */
1397 [ + + ]: 58554 : if (!s->in_use)
1398 : 57710 : continue;
1399 : :
1400 : : /* we're only interested in logical slots */
1401 [ + + ]: 844 : if (!SlotIsLogical(s))
1402 : 570 : continue;
1403 : :
1404 : : /* read once, it's ok if it increases while we're checking */
1405 : 274 : SpinLockAcquire(&s->mutex);
1406 : 274 : persistency = s->data.persistency;
1407 : 274 : restart_lsn = s->data.restart_lsn;
1408 : 274 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1409 : 274 : last_saved_restart_lsn = s->last_saved_restart_lsn;
1410 : 274 : SpinLockRelease(&s->mutex);
1411 : :
1412 : : /* invalidated slots need not apply */
1413 [ + + ]: 274 : if (invalidated)
1414 : 10 : continue;
1415 : :
1416 : : /*
1417 : : * For persistent slot use last_saved_restart_lsn to compute the
1418 : : * oldest LSN for removal of WAL segments. The segments between
1419 : : * last_saved_restart_lsn and restart_lsn might be needed by a
1420 : : * persistent slot in the case of database crash. Non-persistent
1421 : : * slots can't survive the database crash, so we don't care about
1422 : : * last_saved_restart_lsn for them.
1423 : : */
1424 [ + + ]: 264 : if (persistency == RS_PERSISTENT)
1425 : : {
1426 [ + - - + ]: 262 : if (XLogRecPtrIsValid(last_saved_restart_lsn) &&
1427 : : restart_lsn > last_saved_restart_lsn)
1428 : : {
1429 : 0 : restart_lsn = last_saved_restart_lsn;
1430 : : }
1431 : : }
1432 : :
1433 [ - + ]: 264 : if (!XLogRecPtrIsValid(restart_lsn))
1434 : 0 : continue;
1435 : :
1436 [ + + + + ]: 264 : if (!XLogRecPtrIsValid(result) ||
1437 : : restart_lsn < result)
1438 : 206 : result = restart_lsn;
1439 : : }
1440 : :
1441 : 3942 : LWLockRelease(ReplicationSlotControlLock);
1442 : :
1443 : 3942 : return result;
1444 : : }
1445 : :
1446 : : /*
1447 : : * ReplicationSlotsCountDBSlots -- count the number of slots that refer to the
1448 : : * passed database oid.
1449 : : *
1450 : : * Returns true if there are any slots referencing the database. *nslots will
1451 : : * be set to the absolute number of slots in the database, *nactive to ones
1452 : : * currently active.
1453 : : */
1454 : : bool
1455 : 61 : ReplicationSlotsCountDBSlots(Oid dboid, int *nslots, int *nactive)
1456 : : {
1457 : : int i;
1458 : :
1459 : 61 : *nslots = *nactive = 0;
1460 : :
1461 [ - + ]: 61 : if (max_replication_slots + max_repack_replication_slots <= 0)
1462 : 0 : return false;
1463 : :
1464 : 61 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1465 [ + + ]: 949 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1466 : : {
1467 : : ReplicationSlot *s;
1468 : :
1469 : 888 : s = &ReplicationSlotCtl->replication_slots[i];
1470 : :
1471 : : /* cannot change while ReplicationSlotCtlLock is held */
1472 [ + + ]: 888 : if (!s->in_use)
1473 : 867 : continue;
1474 : :
1475 : : /* only logical slots are database specific, skip */
1476 [ + + ]: 21 : if (!SlotIsLogical(s))
1477 : 10 : continue;
1478 : :
1479 : : /* not our database, skip */
1480 [ + + ]: 11 : if (s->data.database != dboid)
1481 : 8 : continue;
1482 : :
1483 : : /* NB: intentionally counting invalidated slots */
1484 : :
1485 : : /* count slots with spinlock held */
1486 : 3 : SpinLockAcquire(&s->mutex);
1487 : 3 : (*nslots)++;
1488 [ + + ]: 3 : if (s->active_proc != INVALID_PROC_NUMBER)
1489 : 1 : (*nactive)++;
1490 : 3 : SpinLockRelease(&s->mutex);
1491 : : }
1492 : 61 : LWLockRelease(ReplicationSlotControlLock);
1493 : :
1494 [ + + ]: 61 : if (*nslots > 0)
1495 : 3 : return true;
1496 : 58 : return false;
1497 : : }
1498 : :
1499 : : /*
1500 : : * ReplicationSlotsDropDBSlots -- Drop all db-specific slots relating to the
1501 : : * passed database oid. The caller should hold an exclusive lock on the
1502 : : * pg_database oid for the database to prevent creation of new slots on the db
1503 : : * or replay from existing slots.
1504 : : *
1505 : : * Another session that concurrently acquires an existing slot on the target DB
1506 : : * (most likely to drop it) may cause this function to ERROR. If that happens
1507 : : * it may have dropped some but not all slots.
1508 : : *
1509 : : * This routine isn't as efficient as it could be - but we don't drop
1510 : : * databases often, especially databases with lots of slots.
1511 : : *
1512 : : * If the last logical slot in the cluster is dropped, request to disable
1513 : : * logical decoding.
1514 : : */
1515 : : void
1516 : 74 : ReplicationSlotsDropDBSlots(Oid dboid)
1517 : : {
1518 : : int i;
1519 : : bool found_valid_logicalslot;
1520 : 74 : bool dropped = false;
1521 : :
1522 [ + - ]: 74 : if (max_replication_slots + max_repack_replication_slots <= 0)
1523 : 0 : return;
1524 : :
1525 : 74 : restart:
1526 : 79 : found_valid_logicalslot = false;
1527 : 79 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1528 [ + + ]: 1149 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1529 : : {
1530 : : ReplicationSlot *s;
1531 : : char *slotname;
1532 : : ProcNumber active_proc;
1533 : :
1534 : 1075 : s = &ReplicationSlotCtl->replication_slots[i];
1535 : :
1536 : : /* cannot change while ReplicationSlotCtlLock is held */
1537 [ + + ]: 1075 : if (!s->in_use)
1538 : 1045 : continue;
1539 : :
1540 : : /* only logical slots are database specific, skip */
1541 [ + + ]: 30 : if (!SlotIsLogical(s))
1542 : 11 : continue;
1543 : :
1544 : : /*
1545 : : * Check logical slots on other databases too so we can disable
1546 : : * logical decoding only if no slots in the cluster.
1547 : : */
1548 : 19 : SpinLockAcquire(&s->mutex);
1549 : 19 : found_valid_logicalslot |= (s->data.invalidated == RS_INVAL_NONE);
1550 : 19 : SpinLockRelease(&s->mutex);
1551 : :
1552 : : /* not our database, skip */
1553 [ + + ]: 19 : if (s->data.database != dboid)
1554 : 14 : continue;
1555 : :
1556 : : /* NB: intentionally including invalidated slots to drop */
1557 : :
1558 : : /* acquire slot, so ReplicationSlotDropAcquired can be reused */
1559 : 5 : SpinLockAcquire(&s->mutex);
1560 : : /* can't change while ReplicationSlotControlLock is held */
1561 : 5 : slotname = NameStr(s->data.name);
1562 : 5 : active_proc = s->active_proc;
1563 [ + - ]: 5 : if (active_proc == INVALID_PROC_NUMBER)
1564 : : {
1565 : 5 : MyReplicationSlot = s;
1566 : 5 : s->active_proc = MyProcNumber;
1567 : : }
1568 : 5 : SpinLockRelease(&s->mutex);
1569 : :
1570 : : /*
1571 : : * Even though we hold an exclusive lock on the database object a
1572 : : * logical slot for that DB can still be active, e.g. if it's
1573 : : * concurrently being dropped by a backend connected to another DB.
1574 : : *
1575 : : * That's fairly unlikely in practice, so we'll just bail out.
1576 : : *
1577 : : * The slot sync worker holds a shared lock on the database before
1578 : : * operating on synced logical slots to avoid conflict with the drop
1579 : : * happening here. The persistent synced slots are thus safe but there
1580 : : * is a possibility that the slot sync worker has created a temporary
1581 : : * slot (which stays active even on release) and we are trying to drop
1582 : : * that here. In practice, the chances of hitting this scenario are
1583 : : * less as during slot synchronization, the temporary slot is
1584 : : * immediately converted to persistent and thus is safe due to the
1585 : : * shared lock taken on the database. So, we'll just bail out in such
1586 : : * a case.
1587 : : *
1588 : : * XXX: We can consider shutting down the slot sync worker before
1589 : : * trying to drop synced temporary slots here.
1590 : : */
1591 [ - + ]: 5 : if (active_proc != INVALID_PROC_NUMBER)
1592 [ # # ]: 0 : ereport(ERROR,
1593 : : (errcode(ERRCODE_OBJECT_IN_USE),
1594 : : errmsg("replication slot \"%s\" is active for PID %d",
1595 : : slotname, GetPGProcByNumber(active_proc)->pid)));
1596 : :
1597 : : /*
1598 : : * To avoid duplicating ReplicationSlotDropAcquired() and to avoid
1599 : : * holding ReplicationSlotControlLock over filesystem operations,
1600 : : * release ReplicationSlotControlLock and use
1601 : : * ReplicationSlotDropAcquired.
1602 : : *
1603 : : * As that means the set of slots could change, restart scan from the
1604 : : * beginning each time we release the lock.
1605 : : */
1606 : 5 : LWLockRelease(ReplicationSlotControlLock);
1607 : 5 : ReplicationSlotDropAcquired(false);
1608 : 5 : dropped = true;
1609 : 5 : goto restart;
1610 : : }
1611 : 74 : LWLockRelease(ReplicationSlotControlLock);
1612 : :
1613 [ + + - + ]: 74 : if (dropped && !found_valid_logicalslot)
1614 : 0 : RequestDisableLogicalDecoding();
1615 : : }
1616 : :
1617 : : /*
1618 : : * Returns true if there is at least one in-use valid logical replication slot.
1619 : : */
1620 : : bool
1621 : 916 : CheckLogicalSlotExists(void)
1622 : : {
1623 : 916 : bool found = false;
1624 : :
1625 [ - + ]: 916 : if (max_replication_slots + max_repack_replication_slots <= 0)
1626 : 0 : return false;
1627 : :
1628 : 916 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1629 [ + + ]: 14475 : for (int i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1630 : : {
1631 : : ReplicationSlot *s;
1632 : : bool invalidated;
1633 : :
1634 : 13569 : s = &ReplicationSlotCtl->replication_slots[i];
1635 : :
1636 : : /* cannot change while ReplicationSlotCtlLock is held */
1637 [ + + ]: 13569 : if (!s->in_use)
1638 : 13535 : continue;
1639 : :
1640 [ + + ]: 34 : if (SlotIsPhysical(s))
1641 : 21 : continue;
1642 : :
1643 : 13 : SpinLockAcquire(&s->mutex);
1644 : 13 : invalidated = s->data.invalidated != RS_INVAL_NONE;
1645 : 13 : SpinLockRelease(&s->mutex);
1646 : :
1647 [ + + ]: 13 : if (invalidated)
1648 : 3 : continue;
1649 : :
1650 : 10 : found = true;
1651 : 10 : break;
1652 : : }
1653 : 916 : LWLockRelease(ReplicationSlotControlLock);
1654 : :
1655 : 916 : return found;
1656 : : }
1657 : :
1658 : : /*
1659 : : * Check whether the server's configuration supports using replication
1660 : : * slots.
1661 : : */
1662 : : void
1663 : 1974 : CheckSlotRequirements(bool repack)
1664 : : {
1665 : : /*
1666 : : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
1667 : : * needs the same check.
1668 : : */
1669 : :
1670 [ + + - + ]: 1974 : if (!repack && max_replication_slots == 0)
1671 [ # # ]: 0 : ereport(ERROR,
1672 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1673 : : errmsg("replication slots can only be used if \"%s\" > 0",
1674 : : "max_replication_slots"));
1675 : :
1676 [ + + - + ]: 1974 : if (repack && max_repack_replication_slots == 0)
1677 [ # # ]: 0 : ereport(ERROR,
1678 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1679 : : errmsg("REPACK can only be used if \"%s\" > 0",
1680 : : "max_repack_replication_slots"));
1681 : :
1682 [ - + ]: 1974 : if (wal_level < WAL_LEVEL_REPLICA)
1683 [ # # ]: 0 : ereport(ERROR,
1684 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1685 : : errmsg("replication slots can only be used if \"wal_level\" >= \"replica\"")));
1686 : 1974 : }
1687 : :
1688 : : /*
1689 : : * Check whether the user has privilege to use replication slots.
1690 : : */
1691 : : void
1692 : 625 : CheckSlotPermissions(void)
1693 : : {
1694 [ + + ]: 625 : if (!has_rolreplication(GetUserId()))
1695 [ + - ]: 5 : ereport(ERROR,
1696 : : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1697 : : errmsg("permission denied to use replication slots"),
1698 : : errdetail("Only roles with the %s attribute may use replication slots.",
1699 : : "REPLICATION")));
1700 : 620 : }
1701 : :
1702 : : /*
1703 : : * Reserve WAL for the currently active slot.
1704 : : *
1705 : : * Compute and set restart_lsn in a manner that's appropriate for the type of
1706 : : * the slot and concurrency safe.
1707 : : */
1708 : : void
1709 : 688 : ReplicationSlotReserveWal(void)
1710 : : {
1711 : 688 : ReplicationSlot *slot = MyReplicationSlot;
1712 : : XLogSegNo segno;
1713 : : XLogRecPtr restart_lsn;
1714 : :
1715 : : Assert(slot != NULL);
1716 : : Assert(!XLogRecPtrIsValid(slot->data.restart_lsn));
1717 : : Assert(!XLogRecPtrIsValid(slot->last_saved_restart_lsn));
1718 : :
1719 : : /*
1720 : : * The replication slot mechanism is used to prevent the removal of
1721 : : * required WAL.
1722 : : *
1723 : : * Acquire an exclusive lock to prevent the checkpoint process from
1724 : : * concurrently computing the minimum slot LSN (see
1725 : : * CheckPointReplicationSlots). This ensures that the WAL reserved for
1726 : : * replication cannot be removed during a checkpoint.
1727 : : *
1728 : : * The mechanism is reliable because if WAL reservation occurs first, the
1729 : : * checkpoint must wait for the restart_lsn update before determining the
1730 : : * minimum non-removable LSN. On the other hand, if the checkpoint happens
1731 : : * first, subsequent WAL reservations will select positions at or beyond
1732 : : * the redo pointer of that checkpoint.
1733 : : */
1734 : 688 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
1735 : :
1736 : : /*
1737 : : * For logical slots log a standby snapshot and start logical decoding at
1738 : : * exactly that position. That allows the slot to start up more quickly.
1739 : : * But on a standby we cannot do WAL writes, so just use the replay
1740 : : * pointer; effectively, an attempt to create a logical slot on standby
1741 : : * will cause it to wait for an xl_running_xact record to be logged
1742 : : * independently on the primary, so that a snapshot can be built using the
1743 : : * record.
1744 : : *
1745 : : * None of this is needed (or indeed helpful) for physical slots as
1746 : : * they'll start replay at the last logged checkpoint anyway. Instead,
1747 : : * return the location of the last redo LSN, where a base backup has to
1748 : : * start replay at.
1749 : : */
1750 [ + + ]: 688 : if (SlotIsPhysical(slot))
1751 : 176 : restart_lsn = GetRedoRecPtr();
1752 [ + + ]: 512 : else if (RecoveryInProgress())
1753 : 28 : restart_lsn = GetXLogReplayRecPtr(NULL);
1754 : : else
1755 : 484 : restart_lsn = GetXLogInsertRecPtr();
1756 : :
1757 : 688 : SpinLockAcquire(&slot->mutex);
1758 : 688 : slot->data.restart_lsn = restart_lsn;
1759 : 688 : SpinLockRelease(&slot->mutex);
1760 : :
1761 : : /* prevent WAL removal as fast as possible */
1762 : 688 : ReplicationSlotsComputeRequiredLSN();
1763 : :
1764 : : /* Checkpoint shouldn't remove the required WAL. */
1765 : 688 : XLByteToSeg(slot->data.restart_lsn, segno, wal_segment_size);
1766 [ - + ]: 688 : if (XLogGetLastRemovedSegno() >= segno)
1767 [ # # ]: 0 : elog(ERROR, "WAL required by replication slot %s has been removed concurrently",
1768 : : NameStr(slot->data.name));
1769 : :
1770 : 688 : LWLockRelease(ReplicationSlotAllocationLock);
1771 : :
1772 [ + + + + ]: 688 : if (!RecoveryInProgress() && SlotIsLogical(slot))
1773 : : {
1774 : : XLogRecPtr flushptr;
1775 : :
1776 : : /* make sure we have enough information to start */
1777 : 484 : flushptr = LogStandbySnapshot();
1778 : :
1779 : : /* and make sure it's fsynced to disk */
1780 : 484 : XLogFlush(flushptr);
1781 : : }
1782 : 688 : }
1783 : :
1784 : : /*
1785 : : * Report that replication slot needs to be invalidated
1786 : : */
1787 : : static void
1788 : 23 : ReportSlotInvalidation(ReplicationSlotInvalidationCause cause,
1789 : : bool terminating,
1790 : : int pid,
1791 : : NameData slotname,
1792 : : XLogRecPtr restart_lsn,
1793 : : XLogRecPtr oldestLSN,
1794 : : TransactionId snapshotConflictHorizon,
1795 : : long slot_idle_seconds)
1796 : : {
1797 : : StringInfoData err_detail;
1798 : : StringInfoData err_hint;
1799 : :
1800 : 23 : initStringInfo(&err_detail);
1801 : 23 : initStringInfo(&err_hint);
1802 : :
1803 [ + + + - : 23 : switch (cause)
- ]
1804 : : {
1805 : 7 : case RS_INVAL_WAL_REMOVED:
1806 : : {
1807 : 7 : uint64 ex = oldestLSN - restart_lsn;
1808 : :
1809 : 7 : appendStringInfo(&err_detail,
1810 : 7 : ngettext("The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " byte.",
1811 : : "The slot's restart_lsn %X/%08X exceeds the limit by %" PRIu64 " bytes.",
1812 : : ex),
1813 : 7 : LSN_FORMAT_ARGS(restart_lsn),
1814 : : ex);
1815 : : /* translator: %s is a GUC variable name */
1816 : 7 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1817 : : "max_slot_wal_keep_size");
1818 : 7 : break;
1819 : : }
1820 : 12 : case RS_INVAL_HORIZON:
1821 : 12 : appendStringInfo(&err_detail, _("The slot conflicted with xid horizon %u."),
1822 : : snapshotConflictHorizon);
1823 : 12 : break;
1824 : :
1825 : 4 : case RS_INVAL_WAL_LEVEL:
1826 : 4 : appendStringInfoString(&err_detail, _("Logical decoding on standby requires the primary server to either set \"wal_level\" >= \"logical\" or have at least one logical slot when \"wal_level\" = \"replica\"."));
1827 : 4 : break;
1828 : :
1829 : 0 : case RS_INVAL_IDLE_TIMEOUT:
1830 : : {
1831 : : /* translator: %s is a GUC variable name */
1832 : 0 : appendStringInfo(&err_detail, _("The slot's idle time of %lds exceeds the configured \"%s\" duration of %ds."),
1833 : : slot_idle_seconds, "idle_replication_slot_timeout",
1834 : : idle_replication_slot_timeout_secs);
1835 : : /* translator: %s is a GUC variable name */
1836 : 0 : appendStringInfo(&err_hint, _("You might need to increase \"%s\"."),
1837 : : "idle_replication_slot_timeout");
1838 : 0 : break;
1839 : : }
1840 : : case RS_INVAL_NONE:
1841 : : pg_unreachable();
1842 : : }
1843 : :
1844 [ + - + + : 23 : ereport(LOG,
+ + ]
1845 : : terminating ?
1846 : : errmsg("terminating process %d to release replication slot \"%s\"",
1847 : : pid, NameStr(slotname)) :
1848 : : errmsg("invalidating obsolete replication slot \"%s\"",
1849 : : NameStr(slotname)),
1850 : : errdetail_internal("%s", err_detail.data),
1851 : : err_hint.len ? errhint("%s", err_hint.data) : 0);
1852 : :
1853 : 23 : pfree(err_detail.data);
1854 : 23 : pfree(err_hint.data);
1855 : 23 : }
1856 : :
1857 : : /*
1858 : : * Can we invalidate an idle replication slot?
1859 : : *
1860 : : * Idle timeout invalidation is allowed only when:
1861 : : *
1862 : : * 1. Idle timeout is set
1863 : : * 2. Slot has reserved WAL
1864 : : * 3. Slot is inactive
1865 : : * 4. The slot is not being synced from the primary while the server is in
1866 : : * recovery. This is because synced slots are always considered to be
1867 : : * inactive because they don't perform logical decoding to produce changes.
1868 : : */
1869 : : static inline bool
1870 : 396 : CanInvalidateIdleSlot(ReplicationSlot *s)
1871 : : {
1872 : 396 : return (idle_replication_slot_timeout_secs != 0 &&
1873 [ # # ]: 0 : XLogRecPtrIsValid(s->data.restart_lsn) &&
1874 [ - + - - ]: 396 : s->inactive_since > 0 &&
1875 [ # # # # ]: 0 : !(RecoveryInProgress() && s->data.synced));
1876 : : }
1877 : :
1878 : : /*
1879 : : * DetermineSlotInvalidationCause - Determine the cause for which a slot
1880 : : * becomes invalid among the given possible causes.
1881 : : *
1882 : : * This function sequentially checks all possible invalidation causes and
1883 : : * returns the first one for which the slot is eligible for invalidation.
1884 : : */
1885 : : static ReplicationSlotInvalidationCause
1886 : 431 : DetermineSlotInvalidationCause(uint32 possible_causes, ReplicationSlot *s,
1887 : : XLogRecPtr oldestLSN, Oid dboid,
1888 : : TransactionId snapshotConflictHorizon,
1889 : : TimestampTz *inactive_since, TimestampTz now)
1890 : : {
1891 : : Assert(possible_causes != RS_INVAL_NONE);
1892 : :
1893 [ + + ]: 431 : if (possible_causes & RS_INVAL_WAL_REMOVED)
1894 : : {
1895 : 403 : XLogRecPtr restart_lsn = s->data.restart_lsn;
1896 : :
1897 [ + + + + ]: 403 : if (XLogRecPtrIsValid(restart_lsn) &&
1898 : : restart_lsn < oldestLSN)
1899 : 7 : return RS_INVAL_WAL_REMOVED;
1900 : : }
1901 : :
1902 [ + + ]: 424 : if (possible_causes & RS_INVAL_HORIZON)
1903 : : {
1904 : : /* invalid DB oid signals a shared relation */
1905 [ + - + + ]: 24 : if (SlotIsLogical(s) &&
1906 [ + - ]: 19 : (dboid == InvalidOid || dboid == s->data.database))
1907 : : {
1908 : 24 : TransactionId effective_xmin = s->effective_xmin;
1909 : 24 : TransactionId catalog_effective_xmin = s->effective_catalog_xmin;
1910 : :
1911 [ - + - - ]: 24 : if (TransactionIdIsValid(effective_xmin) &&
1912 : 0 : TransactionIdPrecedesOrEquals(effective_xmin,
1913 : : snapshotConflictHorizon))
1914 : 0 : return RS_INVAL_HORIZON;
1915 [ + - + + ]: 48 : else if (TransactionIdIsValid(catalog_effective_xmin) &&
1916 : 24 : TransactionIdPrecedesOrEquals(catalog_effective_xmin,
1917 : : snapshotConflictHorizon))
1918 : 12 : return RS_INVAL_HORIZON;
1919 : : }
1920 : : }
1921 : :
1922 [ + + ]: 412 : if (possible_causes & RS_INVAL_WAL_LEVEL)
1923 : : {
1924 [ + - ]: 4 : if (SlotIsLogical(s))
1925 : 4 : return RS_INVAL_WAL_LEVEL;
1926 : : }
1927 : :
1928 [ + + ]: 408 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
1929 : : {
1930 : : Assert(now > 0);
1931 : :
1932 [ - + ]: 396 : if (CanInvalidateIdleSlot(s))
1933 : : {
1934 : : /*
1935 : : * Simulate the invalidation due to idle_timeout to test the
1936 : : * timeout behavior promptly, without waiting for it to trigger
1937 : : * naturally.
1938 : : */
1939 : : #ifdef USE_INJECTION_POINTS
1940 [ # # ]: 0 : if (IS_INJECTION_POINT_ATTACHED("slot-timeout-inval"))
1941 : : {
1942 : 0 : *inactive_since = 0; /* since the beginning of time */
1943 : 0 : return RS_INVAL_IDLE_TIMEOUT;
1944 : : }
1945 : : #endif
1946 : :
1947 : : /*
1948 : : * Check if the slot needs to be invalidated due to
1949 : : * idle_replication_slot_timeout GUC.
1950 : : */
1951 [ # # ]: 0 : if (TimestampDifferenceExceedsSeconds(s->inactive_since, now,
1952 : : idle_replication_slot_timeout_secs))
1953 : : {
1954 : 0 : *inactive_since = s->inactive_since;
1955 : 0 : return RS_INVAL_IDLE_TIMEOUT;
1956 : : }
1957 : : }
1958 : : }
1959 : :
1960 : 408 : return RS_INVAL_NONE;
1961 : : }
1962 : :
1963 : : /*
1964 : : * Helper for InvalidateObsoleteReplicationSlots
1965 : : *
1966 : : * Acquires the given slot and mark it invalid, if necessary and possible.
1967 : : *
1968 : : * Returns true if the slot was invalidated.
1969 : : *
1970 : : * Set *released_lock_out if ReplicationSlotControlLock was released in the
1971 : : * interim (and in that case we're not holding the lock at return, otherwise
1972 : : * we are).
1973 : : *
1974 : : * This is inherently racy, because we release the LWLock
1975 : : * for syscalls, so caller must restart if we return true.
1976 : : */
1977 : : static bool
1978 : 470 : InvalidatePossiblyObsoleteSlot(uint32 possible_causes,
1979 : : ReplicationSlot *s,
1980 : : XLogRecPtr oldestLSN,
1981 : : Oid dboid, TransactionId snapshotConflictHorizon,
1982 : : bool *released_lock_out)
1983 : : {
1984 : 470 : int last_signaled_pid = 0;
1985 : 470 : bool released_lock = false;
1986 : 470 : bool invalidated = false;
1987 : 470 : TimestampTz inactive_since = 0;
1988 : :
1989 : : for (;;)
1990 : 7 : {
1991 : : XLogRecPtr restart_lsn;
1992 : : NameData slotname;
1993 : : ProcNumber active_proc;
1994 : 477 : int active_pid = 0;
1995 : 477 : ReplicationSlotInvalidationCause invalidation_cause = RS_INVAL_NONE;
1996 : 477 : TimestampTz now = 0;
1997 : 477 : long slot_idle_secs = 0;
1998 : :
1999 : : Assert(LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_SHARED));
2000 : :
2001 [ - + ]: 477 : if (!s->in_use)
2002 : : {
2003 [ # # ]: 0 : if (released_lock)
2004 : 0 : LWLockRelease(ReplicationSlotControlLock);
2005 : 0 : break;
2006 : : }
2007 : :
2008 [ + + ]: 477 : if (possible_causes & RS_INVAL_IDLE_TIMEOUT)
2009 : : {
2010 : : /*
2011 : : * Assign the current time here to avoid system call overhead
2012 : : * while holding the spinlock in subsequent code.
2013 : : */
2014 : 417 : now = GetCurrentTimestamp();
2015 : : }
2016 : :
2017 : : /*
2018 : : * Check if the slot needs to be invalidated. If it needs to be
2019 : : * invalidated, and is not currently acquired, acquire it and mark it
2020 : : * as having been invalidated. We do this with the spinlock held to
2021 : : * avoid race conditions -- for example the restart_lsn could move
2022 : : * forward, or the slot could be dropped.
2023 : : */
2024 : 477 : SpinLockAcquire(&s->mutex);
2025 : :
2026 : 477 : restart_lsn = s->data.restart_lsn;
2027 : :
2028 : : /* we do nothing if the slot is already invalid */
2029 [ + + ]: 477 : if (s->data.invalidated == RS_INVAL_NONE)
2030 : 431 : invalidation_cause = DetermineSlotInvalidationCause(possible_causes,
2031 : : s, oldestLSN,
2032 : : dboid,
2033 : : snapshotConflictHorizon,
2034 : : &inactive_since,
2035 : : now);
2036 : :
2037 : : /* if there's no invalidation, we're done */
2038 [ + + ]: 477 : if (invalidation_cause == RS_INVAL_NONE)
2039 : : {
2040 : 454 : SpinLockRelease(&s->mutex);
2041 [ - + ]: 454 : if (released_lock)
2042 : 0 : LWLockRelease(ReplicationSlotControlLock);
2043 : 454 : break;
2044 : : }
2045 : :
2046 : 23 : slotname = s->data.name;
2047 : 23 : active_proc = s->active_proc;
2048 : :
2049 : : /*
2050 : : * If the slot can be acquired, do so and mark it invalidated
2051 : : * immediately. Otherwise we'll signal the owning process, below, and
2052 : : * retry.
2053 : : *
2054 : : * Note: Unlike other slot attributes, slot's inactive_since can't be
2055 : : * changed until the acquired slot is released or the owning process
2056 : : * is terminated. So, the inactive slot can only be invalidated
2057 : : * immediately without being terminated.
2058 : : */
2059 [ + + ]: 23 : if (active_proc == INVALID_PROC_NUMBER)
2060 : : {
2061 : 16 : MyReplicationSlot = s;
2062 : 16 : s->active_proc = MyProcNumber;
2063 : 16 : s->data.invalidated = invalidation_cause;
2064 : :
2065 : : /*
2066 : : * XXX: We should consider not overwriting restart_lsn and instead
2067 : : * just rely on .invalidated.
2068 : : */
2069 [ + + ]: 16 : if (invalidation_cause == RS_INVAL_WAL_REMOVED)
2070 : : {
2071 : 5 : s->data.restart_lsn = InvalidXLogRecPtr;
2072 : 5 : s->last_saved_restart_lsn = InvalidXLogRecPtr;
2073 : : }
2074 : :
2075 : : /* Let caller know */
2076 : 16 : invalidated = true;
2077 : : }
2078 : : else
2079 : : {
2080 : 7 : active_pid = GetPGProcByNumber(active_proc)->pid;
2081 : : Assert(active_pid != 0);
2082 : : }
2083 : :
2084 : 23 : SpinLockRelease(&s->mutex);
2085 : :
2086 : : /*
2087 : : * Calculate the idle time duration of the slot if slot is marked
2088 : : * invalidated with RS_INVAL_IDLE_TIMEOUT.
2089 : : */
2090 [ - + ]: 23 : if (invalidation_cause == RS_INVAL_IDLE_TIMEOUT)
2091 : : {
2092 : : int slot_idle_usecs;
2093 : :
2094 : 0 : TimestampDifference(inactive_since, now, &slot_idle_secs,
2095 : : &slot_idle_usecs);
2096 : : }
2097 : :
2098 [ + + ]: 23 : if (active_proc != INVALID_PROC_NUMBER)
2099 : : {
2100 : : /*
2101 : : * Prepare the sleep on the slot's condition variable before
2102 : : * releasing the lock, to close a possible race condition if the
2103 : : * slot is released before the sleep below.
2104 : : */
2105 : 7 : ConditionVariablePrepareToSleep(&s->active_cv);
2106 : :
2107 : 7 : LWLockRelease(ReplicationSlotControlLock);
2108 : 7 : released_lock = true;
2109 : :
2110 : : /*
2111 : : * Signal to terminate the process that owns the slot, if we
2112 : : * haven't already signalled it. (Avoidance of repeated
2113 : : * signalling is the only reason for there to be a loop in this
2114 : : * routine; otherwise we could rely on caller's restart loop.)
2115 : : *
2116 : : * There is the race condition that other process may own the slot
2117 : : * after its current owner process is terminated and before this
2118 : : * process owns it. To handle that, we signal only if the PID of
2119 : : * the owning process has changed from the previous time. (This
2120 : : * logic assumes that the same PID is not reused very quickly.)
2121 : : */
2122 [ + - ]: 7 : if (last_signaled_pid != active_pid)
2123 : : {
2124 : 7 : ReportSlotInvalidation(invalidation_cause, true, active_pid,
2125 : : slotname, restart_lsn,
2126 : : oldestLSN, snapshotConflictHorizon,
2127 : : slot_idle_secs);
2128 : :
2129 [ + + ]: 7 : if (MyBackendType == B_STARTUP)
2130 : 5 : (void) SignalRecoveryConflict(GetPGProcByNumber(active_proc),
2131 : : active_pid,
2132 : : RECOVERY_CONFLICT_LOGICALSLOT);
2133 : : else
2134 : 2 : (void) kill(active_pid, SIGTERM);
2135 : :
2136 : 7 : last_signaled_pid = active_pid;
2137 : : }
2138 : :
2139 : : /* Wait until the slot is released. */
2140 : 7 : ConditionVariableSleep(&s->active_cv,
2141 : : WAIT_EVENT_REPLICATION_SLOT_DROP);
2142 : :
2143 : : /*
2144 : : * Re-acquire lock and start over; we expect to invalidate the
2145 : : * slot next time (unless another process acquires the slot in the
2146 : : * meantime).
2147 : : *
2148 : : * Note: It is possible for a slot to advance its restart_lsn or
2149 : : * xmin values sufficiently between when we release the mutex and
2150 : : * when we recheck, moving from a conflicting state to a non
2151 : : * conflicting state. This is intentional and safe: if the slot
2152 : : * has caught up while we're busy here, the resources we were
2153 : : * concerned about (WAL segments or tuples) have not yet been
2154 : : * removed, and there's no reason to invalidate the slot.
2155 : : */
2156 : 7 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2157 : 7 : continue;
2158 : : }
2159 : : else
2160 : : {
2161 : : /*
2162 : : * We hold the slot now and have already invalidated it; flush it
2163 : : * to ensure that state persists.
2164 : : *
2165 : : * Don't want to hold ReplicationSlotControlLock across file
2166 : : * system operations, so release it now but be sure to tell caller
2167 : : * to restart from scratch.
2168 : : */
2169 : 16 : LWLockRelease(ReplicationSlotControlLock);
2170 : 16 : released_lock = true;
2171 : :
2172 : : /* Make sure the invalidated state persists across server restart */
2173 : 16 : ReplicationSlotMarkDirty();
2174 : 16 : ReplicationSlotSave();
2175 : 16 : ReplicationSlotRelease();
2176 : :
2177 : 16 : ReportSlotInvalidation(invalidation_cause, false, active_pid,
2178 : : slotname, restart_lsn,
2179 : : oldestLSN, snapshotConflictHorizon,
2180 : : slot_idle_secs);
2181 : :
2182 : : /* done with this slot for now */
2183 : 16 : break;
2184 : : }
2185 : : }
2186 : :
2187 : : Assert(released_lock == !LWLockHeldByMe(ReplicationSlotControlLock));
2188 : :
2189 : 470 : *released_lock_out = released_lock;
2190 : 470 : return invalidated;
2191 : : }
2192 : :
2193 : : /*
2194 : : * Invalidate slots that require resources about to be removed.
2195 : : *
2196 : : * Returns true when any slot have got invalidated.
2197 : : *
2198 : : * Whether a slot needs to be invalidated depends on the invalidation cause.
2199 : : * A slot is invalidated if it:
2200 : : * - RS_INVAL_WAL_REMOVED: requires a LSN older than the given segment
2201 : : * - RS_INVAL_HORIZON: requires a snapshot <= the given horizon in the given
2202 : : * db; dboid may be InvalidOid for shared relations
2203 : : * - RS_INVAL_WAL_LEVEL: is a logical slot and effective_wal_level is not
2204 : : * logical.
2205 : : * - RS_INVAL_IDLE_TIMEOUT: has been idle longer than the configured
2206 : : * "idle_replication_slot_timeout" duration.
2207 : : *
2208 : : * Note: This function attempts to invalidate the slot for multiple possible
2209 : : * causes in a single pass, minimizing redundant iterations. The "cause"
2210 : : * parameter can be a MASK representing one or more of the defined causes.
2211 : : *
2212 : : * If it invalidates the last logical slot in the cluster, it requests to
2213 : : * disable logical decoding.
2214 : : *
2215 : : * NB - this runs as part of checkpoint, so avoid raising errors if possible.
2216 : : */
2217 : : bool
2218 : 2003 : InvalidateObsoleteReplicationSlots(uint32 possible_causes,
2219 : : XLogSegNo oldestSegno, Oid dboid,
2220 : : TransactionId snapshotConflictHorizon)
2221 : : {
2222 : : XLogRecPtr oldestLSN;
2223 : 2003 : bool invalidated = false;
2224 : 2003 : bool invalidated_logical = false;
2225 : : bool found_valid_logicalslot;
2226 : :
2227 : : Assert(!(possible_causes & RS_INVAL_HORIZON) || TransactionIdIsValid(snapshotConflictHorizon));
2228 : : Assert(!(possible_causes & RS_INVAL_WAL_REMOVED) || oldestSegno > 0);
2229 : : Assert(possible_causes != RS_INVAL_NONE);
2230 : :
2231 [ + + - + ]: 2003 : if (max_replication_slots == 0 && max_repack_replication_slots == 0)
2232 : 0 : return invalidated;
2233 : :
2234 : 2003 : XLogSegNoOffsetToRecPtr(oldestSegno, 0, wal_segment_size, oldestLSN);
2235 : :
2236 : 2019 : restart:
2237 : 2019 : found_valid_logicalslot = false;
2238 : 2019 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
2239 [ + + ]: 31691 : for (int i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
2240 : : {
2241 : 29688 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2242 : 29688 : bool released_lock = false;
2243 : :
2244 [ + + ]: 29688 : if (!s->in_use)
2245 : 29218 : continue;
2246 : :
2247 : : /* Prevent invalidation of logical slots during binary upgrade */
2248 [ + + + + ]: 482 : if (SlotIsLogical(s) && IsBinaryUpgrade)
2249 : : {
2250 : 12 : SpinLockAcquire(&s->mutex);
2251 : 12 : found_valid_logicalslot |= (s->data.invalidated == RS_INVAL_NONE);
2252 : 12 : SpinLockRelease(&s->mutex);
2253 : :
2254 : 12 : continue;
2255 : : }
2256 : :
2257 [ + + ]: 470 : if (InvalidatePossiblyObsoleteSlot(possible_causes, s, oldestLSN,
2258 : : dboid, snapshotConflictHorizon,
2259 : : &released_lock))
2260 : : {
2261 : : Assert(released_lock);
2262 : :
2263 : : /* Remember we have invalidated a physical or logical slot */
2264 : 16 : invalidated = true;
2265 : :
2266 : : /*
2267 : : * Additionally, remember we have invalidated a logical slot as we
2268 : : * can request disabling logical decoding later.
2269 : : */
2270 [ + + ]: 16 : if (SlotIsLogical(s))
2271 : 13 : invalidated_logical = true;
2272 : : }
2273 : : else
2274 : : {
2275 : : /*
2276 : : * We need to check if the slot is invalidated here since
2277 : : * InvalidatePossiblyObsoleteSlot() returns false also if the slot
2278 : : * is already invalidated.
2279 : : */
2280 : 454 : SpinLockAcquire(&s->mutex);
2281 : 908 : found_valid_logicalslot |=
2282 [ + + + + ]: 454 : (SlotIsLogical(s) && (s->data.invalidated == RS_INVAL_NONE));
2283 : 454 : SpinLockRelease(&s->mutex);
2284 : : }
2285 : :
2286 : : /* if the lock was released, start from scratch */
2287 [ + + ]: 470 : if (released_lock)
2288 : 16 : goto restart;
2289 : : }
2290 : 2003 : LWLockRelease(ReplicationSlotControlLock);
2291 : :
2292 : : /*
2293 : : * If any slots have been invalidated, recalculate the resource limits.
2294 : : */
2295 [ + + ]: 2003 : if (invalidated)
2296 : : {
2297 : 11 : ReplicationSlotsComputeRequiredXmin(false);
2298 : 11 : ReplicationSlotsComputeRequiredLSN();
2299 : : }
2300 : :
2301 : : /*
2302 : : * Request the checkpointer to disable logical decoding if no valid
2303 : : * logical slots remain. If called by the checkpointer during a
2304 : : * checkpoint, only the request is initiated; actual deactivation is
2305 : : * deferred until after the checkpoint completes.
2306 : : */
2307 [ + + + - ]: 2003 : if (invalidated_logical && !found_valid_logicalslot)
2308 : 8 : RequestDisableLogicalDecoding();
2309 : :
2310 : 2003 : return invalidated;
2311 : : }
2312 : :
2313 : : /*
2314 : : * Flush all replication slots to disk.
2315 : : *
2316 : : * It is convenient to flush dirty replication slots at the time of checkpoint.
2317 : : * Additionally, in case of a shutdown checkpoint, we also identify the slots
2318 : : * for which the confirmed_flush LSN has been updated since the last time it
2319 : : * was saved and flush them.
2320 : : */
2321 : : void
2322 : 1971 : CheckPointReplicationSlots(bool is_shutdown)
2323 : : {
2324 : : int i;
2325 : 1971 : bool last_saved_restart_lsn_updated = false;
2326 : :
2327 [ + + ]: 1971 : elog(DEBUG1, "performing replication slot checkpoint");
2328 : :
2329 : : /*
2330 : : * Prevent any slot from being created/dropped while we're active. As we
2331 : : * explicitly do *not* want to block iterating over replication_slots or
2332 : : * acquiring a slot we cannot take the control lock - but that's OK,
2333 : : * because holding ReplicationSlotAllocationLock is strictly stronger, and
2334 : : * enough to guarantee that nobody can change the in_use bits on us.
2335 : : *
2336 : : * Additionally, acquiring the Allocation lock is necessary to serialize
2337 : : * the slot flush process with concurrent slot WAL reservation. This
2338 : : * ensures that the WAL position being reserved is either flushed to disk
2339 : : * or is beyond or equal to the redo pointer of the current checkpoint
2340 : : * (See ReplicationSlotReserveWal for details).
2341 : : */
2342 : 1971 : LWLockAcquire(ReplicationSlotAllocationLock, LW_SHARED);
2343 : :
2344 [ + + ]: 31248 : for (i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
2345 : : {
2346 : 29277 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
2347 : : char path[MAXPGPATH];
2348 : :
2349 [ + + ]: 29277 : if (!s->in_use)
2350 : 28855 : continue;
2351 : :
2352 : : /* save the slot to disk, locking is handled in SaveSlotToPath() */
2353 : 422 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(s->data.name));
2354 : :
2355 : : /*
2356 : : * Slot's data is not flushed each time the confirmed_flush LSN is
2357 : : * updated as that could lead to frequent writes. However, we decide
2358 : : * to force a flush of all logical slot's data at the time of shutdown
2359 : : * if the confirmed_flush LSN is changed since we last flushed it to
2360 : : * disk. This helps in avoiding an unnecessary retreat of the
2361 : : * confirmed_flush LSN after restart.
2362 : : */
2363 [ + + + + ]: 422 : if (is_shutdown && SlotIsLogical(s))
2364 : : {
2365 : 101 : SpinLockAcquire(&s->mutex);
2366 : :
2367 [ + + ]: 101 : if (s->data.invalidated == RS_INVAL_NONE &&
2368 [ + + ]: 100 : s->data.confirmed_flush > s->last_saved_confirmed_flush)
2369 : : {
2370 : 43 : s->just_dirtied = true;
2371 : 43 : s->dirty = true;
2372 : : }
2373 : 101 : SpinLockRelease(&s->mutex);
2374 : : }
2375 : :
2376 : : /*
2377 : : * Track if we're going to update slot's last_saved_restart_lsn. We
2378 : : * need this to know if we need to recompute the required LSN.
2379 : : */
2380 [ + + ]: 422 : if (s->last_saved_restart_lsn != s->data.restart_lsn)
2381 : 217 : last_saved_restart_lsn_updated = true;
2382 : :
2383 : 422 : SaveSlotToPath(s, path, LOG);
2384 : : }
2385 : 1971 : LWLockRelease(ReplicationSlotAllocationLock);
2386 : :
2387 : : /*
2388 : : * Recompute the required LSN if SaveSlotToPath() updated
2389 : : * last_saved_restart_lsn for any slot.
2390 : : */
2391 [ + + ]: 1971 : if (last_saved_restart_lsn_updated)
2392 : 217 : ReplicationSlotsComputeRequiredLSN();
2393 : 1971 : }
2394 : :
2395 : : /*
2396 : : * Load all replication slots from disk into memory at server startup. This
2397 : : * needs to be run before we start crash recovery.
2398 : : */
2399 : : void
2400 : 1099 : StartupReplicationSlots(void)
2401 : : {
2402 : : DIR *replication_dir;
2403 : : struct dirent *replication_de;
2404 : :
2405 [ + + ]: 1099 : elog(DEBUG1, "starting up replication slots");
2406 : :
2407 : : /* restore all slots by iterating over all on-disk entries */
2408 : 1099 : replication_dir = AllocateDir(PG_REPLSLOT_DIR);
2409 [ + + ]: 3423 : while ((replication_de = ReadDir(replication_dir, PG_REPLSLOT_DIR)) != NULL)
2410 : : {
2411 : : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2412 : : PGFileType de_type;
2413 : :
2414 [ + + ]: 2326 : if (strcmp(replication_de->d_name, ".") == 0 ||
2415 [ + + ]: 1229 : strcmp(replication_de->d_name, "..") == 0)
2416 : 2194 : continue;
2417 : :
2418 : 132 : snprintf(path, sizeof(path), "%s/%s", PG_REPLSLOT_DIR, replication_de->d_name);
2419 : 132 : de_type = get_dirent_type(path, replication_de, false, DEBUG1);
2420 : :
2421 : : /* we're only creating directories here, skip if it's not our's */
2422 [ + - - + ]: 132 : if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_DIR)
2423 : 0 : continue;
2424 : :
2425 : : /* we crashed while a slot was being setup or deleted, clean up */
2426 [ - + ]: 132 : if (pg_str_endswith(replication_de->d_name, ".tmp"))
2427 : : {
2428 [ # # ]: 0 : if (!rmtree(path, true))
2429 : : {
2430 [ # # ]: 0 : ereport(WARNING,
2431 : : (errmsg("could not remove directory \"%s\"",
2432 : : path)));
2433 : 0 : continue;
2434 : : }
2435 : 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2436 : 0 : continue;
2437 : : }
2438 : :
2439 : : /* looks like a slot in a normal state, restore */
2440 : 132 : RestoreSlotFromDisk(replication_de->d_name);
2441 : : }
2442 : 1097 : FreeDir(replication_dir);
2443 : :
2444 : : /* currently no slots exist, we're done. */
2445 [ - + ]: 1097 : if (max_replication_slots + max_repack_replication_slots <= 0)
2446 : 0 : return;
2447 : :
2448 : : /* Now that we have recovered all the data, compute replication xmin */
2449 : 1097 : ReplicationSlotsComputeRequiredXmin(false);
2450 : 1097 : ReplicationSlotsComputeRequiredLSN();
2451 : : }
2452 : :
2453 : : /* ----
2454 : : * Manipulation of on-disk state of replication slots
2455 : : *
2456 : : * NB: none of the routines below should take any notice whether a slot is the
2457 : : * current one or not, that's all handled a layer above.
2458 : : * ----
2459 : : */
2460 : : static void
2461 : 745 : CreateSlotOnDisk(ReplicationSlot *slot)
2462 : : {
2463 : : char tmppath[MAXPGPATH];
2464 : : char path[MAXPGPATH];
2465 : : struct stat st;
2466 : :
2467 : : /*
2468 : : * No need to take out the io_in_progress_lock, nobody else can see this
2469 : : * slot yet, so nobody else will write. We're reusing SaveSlotToPath which
2470 : : * takes out the lock, if we'd take the lock here, we'd deadlock.
2471 : : */
2472 : :
2473 : 745 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2474 : 745 : sprintf(tmppath, "%s/%s.tmp", PG_REPLSLOT_DIR, NameStr(slot->data.name));
2475 : :
2476 : : /*
2477 : : * It's just barely possible that some previous effort to create or drop a
2478 : : * slot with this name left a temp directory lying around. If that seems
2479 : : * to be the case, try to remove it. If the rmtree() fails, we'll error
2480 : : * out at the MakePGDirectory() below, so we don't bother checking
2481 : : * success.
2482 : : */
2483 [ - + - - ]: 745 : if (stat(tmppath, &st) == 0 && S_ISDIR(st.st_mode))
2484 : 0 : rmtree(tmppath, true);
2485 : :
2486 : : /* Create and fsync the temporary slot directory. */
2487 [ - + ]: 745 : if (MakePGDirectory(tmppath) < 0)
2488 [ # # ]: 0 : ereport(ERROR,
2489 : : (errcode_for_file_access(),
2490 : : errmsg("could not create directory \"%s\": %m",
2491 : : tmppath)));
2492 : 745 : fsync_fname(tmppath, true);
2493 : :
2494 : : /* Write the actual state file. */
2495 : 745 : slot->dirty = true; /* signal that we really need to write */
2496 : 745 : SaveSlotToPath(slot, tmppath, ERROR);
2497 : :
2498 : : /* Rename the directory into place. */
2499 [ - + ]: 745 : if (rename(tmppath, path) != 0)
2500 [ # # ]: 0 : ereport(ERROR,
2501 : : (errcode_for_file_access(),
2502 : : errmsg("could not rename file \"%s\" to \"%s\": %m",
2503 : : tmppath, path)));
2504 : :
2505 : : /*
2506 : : * If we'd now fail - really unlikely - we wouldn't know whether this slot
2507 : : * would persist after an OS crash or not - so, force a restart. The
2508 : : * restart would try to fsync this again till it works.
2509 : : */
2510 : 745 : START_CRIT_SECTION();
2511 : :
2512 : 745 : fsync_fname(path, true);
2513 : 745 : fsync_fname(PG_REPLSLOT_DIR, true);
2514 : :
2515 : 745 : END_CRIT_SECTION();
2516 : 745 : }
2517 : :
2518 : : /*
2519 : : * Shared functionality between saving and creating a replication slot.
2520 : : */
2521 : : static void
2522 : 2705 : SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
2523 : : {
2524 : : char tmppath[MAXPGPATH];
2525 : : char path[MAXPGPATH];
2526 : : int fd;
2527 : : ReplicationSlotOnDisk cp;
2528 : : bool was_dirty;
2529 : :
2530 : : /* first check whether there's something to write out */
2531 : 2705 : SpinLockAcquire(&slot->mutex);
2532 : 2705 : was_dirty = slot->dirty;
2533 : 2705 : slot->just_dirtied = false;
2534 : 2705 : SpinLockRelease(&slot->mutex);
2535 : :
2536 : : /* and don't do anything if there's nothing to write */
2537 [ + + ]: 2705 : if (!was_dirty)
2538 : 151 : return;
2539 : :
2540 : 2554 : LWLockAcquire(&slot->io_in_progress_lock, LW_EXCLUSIVE);
2541 : :
2542 : : /* silence valgrind :( */
2543 : 2554 : memset(&cp, 0, sizeof(ReplicationSlotOnDisk));
2544 : :
2545 : 2554 : sprintf(tmppath, "%s/state.tmp", dir);
2546 : 2554 : sprintf(path, "%s/state", dir);
2547 : :
2548 : 2554 : fd = OpenTransientFile(tmppath, O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
2549 [ - + ]: 2554 : if (fd < 0)
2550 : : {
2551 : : /*
2552 : : * If not an ERROR, then release the lock before returning. In case
2553 : : * of an ERROR, the error recovery path automatically releases the
2554 : : * lock, but no harm in explicitly releasing even in that case. Note
2555 : : * that LWLockRelease() could affect errno.
2556 : : */
2557 : 0 : int save_errno = errno;
2558 : :
2559 : 0 : LWLockRelease(&slot->io_in_progress_lock);
2560 : 0 : errno = save_errno;
2561 [ # # ]: 0 : ereport(elevel,
2562 : : (errcode_for_file_access(),
2563 : : errmsg("could not create file \"%s\": %m",
2564 : : tmppath)));
2565 : 0 : return;
2566 : : }
2567 : :
2568 : 2554 : cp.magic = SLOT_MAGIC;
2569 : 2554 : INIT_CRC32C(cp.checksum);
2570 : 2554 : cp.version = SLOT_VERSION;
2571 : 2554 : cp.length = ReplicationSlotOnDiskV2Size;
2572 : :
2573 : 2554 : SpinLockAcquire(&slot->mutex);
2574 : :
2575 : 2554 : memcpy(&cp.slotdata, &slot->data, sizeof(ReplicationSlotPersistentData));
2576 : :
2577 : 2554 : SpinLockRelease(&slot->mutex);
2578 : :
2579 : 2554 : COMP_CRC32C(cp.checksum,
2580 : : (char *) (&cp) + ReplicationSlotOnDiskNotChecksummedSize,
2581 : : ReplicationSlotOnDiskChecksummedSize);
2582 : 2554 : FIN_CRC32C(cp.checksum);
2583 : :
2584 : 2554 : errno = 0;
2585 : 2554 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_WRITE);
2586 [ - + ]: 2554 : if ((write(fd, &cp, sizeof(cp))) != sizeof(cp))
2587 : : {
2588 : 0 : int save_errno = errno;
2589 : :
2590 : 0 : pgstat_report_wait_end();
2591 : 0 : CloseTransientFile(fd);
2592 : 0 : unlink(tmppath);
2593 : 0 : LWLockRelease(&slot->io_in_progress_lock);
2594 : :
2595 : : /* if write didn't set errno, assume problem is no disk space */
2596 [ # # ]: 0 : errno = save_errno ? save_errno : ENOSPC;
2597 [ # # ]: 0 : ereport(elevel,
2598 : : (errcode_for_file_access(),
2599 : : errmsg("could not write to file \"%s\": %m",
2600 : : tmppath)));
2601 : 0 : return;
2602 : : }
2603 : 2554 : pgstat_report_wait_end();
2604 : :
2605 : : /* fsync the temporary file */
2606 : 2554 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_SYNC);
2607 [ - + ]: 2554 : if (pg_fsync(fd) != 0)
2608 : : {
2609 : 0 : int save_errno = errno;
2610 : :
2611 : 0 : pgstat_report_wait_end();
2612 : 0 : CloseTransientFile(fd);
2613 : 0 : unlink(tmppath);
2614 : 0 : LWLockRelease(&slot->io_in_progress_lock);
2615 : :
2616 : 0 : errno = save_errno;
2617 [ # # ]: 0 : ereport(elevel,
2618 : : (errcode_for_file_access(),
2619 : : errmsg("could not fsync file \"%s\": %m",
2620 : : tmppath)));
2621 : 0 : return;
2622 : : }
2623 : 2554 : pgstat_report_wait_end();
2624 : :
2625 [ - + ]: 2554 : if (CloseTransientFile(fd) != 0)
2626 : : {
2627 : 0 : int save_errno = errno;
2628 : :
2629 : 0 : unlink(tmppath);
2630 : 0 : LWLockRelease(&slot->io_in_progress_lock);
2631 : :
2632 : 0 : errno = save_errno;
2633 [ # # ]: 0 : ereport(elevel,
2634 : : (errcode_for_file_access(),
2635 : : errmsg("could not close file \"%s\": %m",
2636 : : tmppath)));
2637 : 0 : return;
2638 : : }
2639 : :
2640 : : /* rename to permanent file, fsync file and directory */
2641 [ - + ]: 2554 : if (rename(tmppath, path) != 0)
2642 : : {
2643 : 0 : int save_errno = errno;
2644 : :
2645 : 0 : unlink(tmppath);
2646 : 0 : LWLockRelease(&slot->io_in_progress_lock);
2647 : :
2648 : 0 : errno = save_errno;
2649 [ # # ]: 0 : ereport(elevel,
2650 : : (errcode_for_file_access(),
2651 : : errmsg("could not rename file \"%s\" to \"%s\": %m",
2652 : : tmppath, path)));
2653 : 0 : return;
2654 : : }
2655 : :
2656 : : /*
2657 : : * Check CreateSlotOnDisk() for the reasoning of using a critical section.
2658 : : */
2659 : 2554 : START_CRIT_SECTION();
2660 : :
2661 : 2554 : fsync_fname(path, false);
2662 : 2554 : fsync_fname(dir, true);
2663 : 2554 : fsync_fname(PG_REPLSLOT_DIR, true);
2664 : :
2665 : 2554 : END_CRIT_SECTION();
2666 : :
2667 : : /*
2668 : : * Successfully wrote, unset dirty bit, unless somebody dirtied again
2669 : : * already and remember the confirmed_flush LSN value.
2670 : : */
2671 : 2554 : SpinLockAcquire(&slot->mutex);
2672 [ + + ]: 2554 : if (!slot->just_dirtied)
2673 : 2535 : slot->dirty = false;
2674 : 2554 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2675 : 2554 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2676 : 2554 : SpinLockRelease(&slot->mutex);
2677 : :
2678 : 2554 : LWLockRelease(&slot->io_in_progress_lock);
2679 : : }
2680 : :
2681 : : /*
2682 : : * Load a single slot from disk into memory.
2683 : : */
2684 : : static void
2685 : 132 : RestoreSlotFromDisk(const char *name)
2686 : : {
2687 : : ReplicationSlotOnDisk cp;
2688 : : int i;
2689 : : char slotdir[MAXPGPATH + sizeof(PG_REPLSLOT_DIR)];
2690 : : char path[MAXPGPATH + sizeof(PG_REPLSLOT_DIR) + 10];
2691 : : int fd;
2692 : 132 : bool restored = false;
2693 : : ssize_t readBytes;
2694 : : pg_crc32c checksum;
2695 : 132 : TimestampTz now = 0;
2696 : :
2697 : : /* no need to lock here, no concurrent access allowed yet */
2698 : :
2699 : : /* delete temp file if it exists */
2700 : 132 : sprintf(slotdir, "%s/%s", PG_REPLSLOT_DIR, name);
2701 : 132 : sprintf(path, "%s/state.tmp", slotdir);
2702 [ + - - + ]: 132 : if (unlink(path) < 0 && errno != ENOENT)
2703 [ # # ]: 0 : ereport(PANIC,
2704 : : (errcode_for_file_access(),
2705 : : errmsg("could not remove file \"%s\": %m", path)));
2706 : :
2707 : 132 : sprintf(path, "%s/state", slotdir);
2708 : :
2709 [ + + ]: 132 : elog(DEBUG1, "restoring replication slot from \"%s\"", path);
2710 : :
2711 : : /* on some operating systems fsyncing a file requires O_RDWR */
2712 : 132 : fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
2713 : :
2714 : : /*
2715 : : * We do not need to handle this as we are rename()ing the directory into
2716 : : * place only after we fsync()ed the state file.
2717 : : */
2718 [ - + ]: 132 : if (fd < 0)
2719 [ # # ]: 0 : ereport(PANIC,
2720 : : (errcode_for_file_access(),
2721 : : errmsg("could not open file \"%s\": %m", path)));
2722 : :
2723 : : /*
2724 : : * Sync state file before we're reading from it. We might have crashed
2725 : : * while it wasn't synced yet and we shouldn't continue on that basis.
2726 : : */
2727 : 132 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_RESTORE_SYNC);
2728 [ - + ]: 132 : if (pg_fsync(fd) != 0)
2729 [ # # ]: 0 : ereport(PANIC,
2730 : : (errcode_for_file_access(),
2731 : : errmsg("could not fsync file \"%s\": %m",
2732 : : path)));
2733 : 132 : pgstat_report_wait_end();
2734 : :
2735 : : /* Also sync the parent directory */
2736 : 132 : START_CRIT_SECTION();
2737 : 132 : fsync_fname(slotdir, true);
2738 : 132 : END_CRIT_SECTION();
2739 : :
2740 : : /* read part of statefile that's guaranteed to be version independent */
2741 : 132 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2742 : 132 : readBytes = read(fd, &cp, ReplicationSlotOnDiskConstantSize);
2743 : 132 : pgstat_report_wait_end();
2744 [ - + ]: 132 : if (readBytes != ReplicationSlotOnDiskConstantSize)
2745 : : {
2746 [ # # ]: 0 : if (readBytes < 0)
2747 [ # # ]: 0 : ereport(PANIC,
2748 : : (errcode_for_file_access(),
2749 : : errmsg("could not read file \"%s\": %m", path)));
2750 : : else
2751 [ # # ]: 0 : ereport(PANIC,
2752 : : (errcode(ERRCODE_DATA_CORRUPTED),
2753 : : errmsg("could not read file \"%s\": read %zd of %zu",
2754 : : path, readBytes,
2755 : : ReplicationSlotOnDiskConstantSize)));
2756 : : }
2757 : :
2758 : : /* verify magic */
2759 [ - + ]: 132 : if (cp.magic != SLOT_MAGIC)
2760 [ # # ]: 0 : ereport(PANIC,
2761 : : (errcode(ERRCODE_DATA_CORRUPTED),
2762 : : errmsg("replication slot file \"%s\" has wrong magic number: %u instead of %u",
2763 : : path, cp.magic, SLOT_MAGIC)));
2764 : :
2765 : : /* verify version */
2766 [ - + ]: 132 : if (cp.version != SLOT_VERSION)
2767 [ # # ]: 0 : ereport(PANIC,
2768 : : (errcode(ERRCODE_DATA_CORRUPTED),
2769 : : errmsg("replication slot file \"%s\" has unsupported version %u",
2770 : : path, cp.version)));
2771 : :
2772 : : /* boundary check on length */
2773 [ - + ]: 132 : if (cp.length != ReplicationSlotOnDiskV2Size)
2774 [ # # ]: 0 : ereport(PANIC,
2775 : : (errcode(ERRCODE_DATA_CORRUPTED),
2776 : : errmsg("replication slot file \"%s\" has corrupted length %u",
2777 : : path, cp.length)));
2778 : :
2779 : : /* Now that we know the size, read the entire file */
2780 : 132 : pgstat_report_wait_start(WAIT_EVENT_REPLICATION_SLOT_READ);
2781 : 132 : readBytes = read(fd,
2782 : : (char *) &cp + ReplicationSlotOnDiskConstantSize,
2783 : 132 : cp.length);
2784 : 132 : pgstat_report_wait_end();
2785 [ - + ]: 132 : if (readBytes != cp.length)
2786 : : {
2787 [ # # ]: 0 : if (readBytes < 0)
2788 [ # # ]: 0 : ereport(PANIC,
2789 : : (errcode_for_file_access(),
2790 : : errmsg("could not read file \"%s\": %m", path)));
2791 : : else
2792 [ # # ]: 0 : ereport(PANIC,
2793 : : (errcode(ERRCODE_DATA_CORRUPTED),
2794 : : errmsg("could not read file \"%s\": read %zd of %zu",
2795 : : path, readBytes, (Size) cp.length)));
2796 : : }
2797 : :
2798 [ - + ]: 132 : if (CloseTransientFile(fd) != 0)
2799 [ # # ]: 0 : ereport(PANIC,
2800 : : (errcode_for_file_access(),
2801 : : errmsg("could not close file \"%s\": %m", path)));
2802 : :
2803 : : /* now verify the CRC */
2804 : 132 : INIT_CRC32C(checksum);
2805 : 132 : COMP_CRC32C(checksum,
2806 : : (char *) &cp + ReplicationSlotOnDiskNotChecksummedSize,
2807 : : ReplicationSlotOnDiskChecksummedSize);
2808 : 132 : FIN_CRC32C(checksum);
2809 : :
2810 [ - + ]: 132 : if (!EQ_CRC32C(checksum, cp.checksum))
2811 [ # # ]: 0 : ereport(PANIC,
2812 : : (errmsg("checksum mismatch for replication slot file \"%s\": is %u, should be %u",
2813 : : path, checksum, cp.checksum)));
2814 : :
2815 : : /*
2816 : : * If we crashed with an ephemeral slot active, don't restore but delete
2817 : : * it.
2818 : : */
2819 [ - + ]: 132 : if (cp.slotdata.persistency != RS_PERSISTENT)
2820 : : {
2821 [ # # ]: 0 : if (!rmtree(slotdir, true))
2822 : : {
2823 [ # # ]: 0 : ereport(WARNING,
2824 : : (errmsg("could not remove directory \"%s\"",
2825 : : slotdir)));
2826 : : }
2827 : 0 : fsync_fname(PG_REPLSLOT_DIR, true);
2828 : 0 : return;
2829 : : }
2830 : :
2831 : : /*
2832 : : * Verify that requirements for the specific slot type are met. That's
2833 : : * important because if these aren't met we're not guaranteed to retain
2834 : : * all the necessary resources for the slot.
2835 : : *
2836 : : * NB: We have to do so *after* the above checks for ephemeral slots,
2837 : : * because otherwise a slot that shouldn't exist anymore could prevent
2838 : : * restarts.
2839 : : *
2840 : : * NB: Changing the requirements here also requires adapting
2841 : : * CheckSlotRequirements() and CheckLogicalDecodingRequirements().
2842 : : */
2843 [ + + ]: 132 : if (cp.slotdata.database != InvalidOid)
2844 : : {
2845 [ + + ]: 93 : if (wal_level < WAL_LEVEL_REPLICA)
2846 [ + - ]: 1 : ereport(FATAL,
2847 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2848 : : errmsg("logical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2849 : : NameStr(cp.slotdata.name)),
2850 : : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2851 : :
2852 : : /*
2853 : : * In standby mode, the hot standby must be enabled. This check is
2854 : : * necessary to ensure logical slots are invalidated when they become
2855 : : * incompatible due to insufficient wal_level. Otherwise, if the
2856 : : * primary reduces effective_wal_level < logical while hot standby is
2857 : : * disabled, primary disable logical decoding while hot standby is
2858 : : * disabled, logical slots would remain valid even after promotion.
2859 : : */
2860 [ + + + + ]: 92 : if (StandbyMode && !EnableHotStandby)
2861 [ + - ]: 1 : ereport(FATAL,
2862 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2863 : : errmsg("logical replication slot \"%s\" exists on the standby, but \"hot_standby\" = \"off\"",
2864 : : NameStr(cp.slotdata.name)),
2865 : : errhint("Change \"hot_standby\" to be \"on\".")));
2866 : : }
2867 [ - + ]: 39 : else if (wal_level < WAL_LEVEL_REPLICA)
2868 [ # # ]: 0 : ereport(FATAL,
2869 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2870 : : errmsg("physical replication slot \"%s\" exists, but \"wal_level\" < \"replica\"",
2871 : : NameStr(cp.slotdata.name)),
2872 : : errhint("Change \"wal_level\" to be \"replica\" or higher.")));
2873 : :
2874 : : /*
2875 : : * Nothing can be active yet, don't lock anything. Note we iterate up to
2876 : : * max_replication_slots instead of adding max_repack_replication_slots as
2877 : : * in all other places, because we must enforce the GUC value in case
2878 : : * there were more slots before the shutdown than what it is set up to
2879 : : * now.
2880 : : */
2881 [ + - ]: 194 : for (i = 0; i < max_replication_slots; i++)
2882 : : {
2883 : : ReplicationSlot *slot;
2884 : :
2885 : 194 : slot = &ReplicationSlotCtl->replication_slots[i];
2886 : :
2887 [ + + ]: 194 : if (slot->in_use)
2888 : 64 : continue;
2889 : :
2890 : : /* restore the entire set of persistent data */
2891 : 130 : memcpy(&slot->data, &cp.slotdata,
2892 : : sizeof(ReplicationSlotPersistentData));
2893 : :
2894 : : /* initialize in memory state */
2895 : 130 : slot->effective_xmin = cp.slotdata.xmin;
2896 : 130 : slot->effective_catalog_xmin = cp.slotdata.catalog_xmin;
2897 : 130 : slot->last_saved_confirmed_flush = cp.slotdata.confirmed_flush;
2898 : 130 : slot->last_saved_restart_lsn = cp.slotdata.restart_lsn;
2899 : :
2900 : 130 : slot->candidate_catalog_xmin = InvalidTransactionId;
2901 : 130 : slot->candidate_xmin_lsn = InvalidXLogRecPtr;
2902 : 130 : slot->candidate_restart_lsn = InvalidXLogRecPtr;
2903 : 130 : slot->candidate_restart_valid = InvalidXLogRecPtr;
2904 : :
2905 : 130 : slot->in_use = true;
2906 : 130 : slot->active_proc = INVALID_PROC_NUMBER;
2907 : :
2908 : : /*
2909 : : * Set the time since the slot has become inactive after loading the
2910 : : * slot from the disk into memory. Whoever acquires the slot i.e.
2911 : : * makes the slot active will reset it. Use the same inactive_since
2912 : : * time for all the slots.
2913 : : */
2914 [ + - ]: 130 : if (now == 0)
2915 : 130 : now = GetCurrentTimestamp();
2916 : :
2917 : 130 : ReplicationSlotSetInactiveSince(slot, now, false);
2918 : :
2919 : 130 : restored = true;
2920 : 130 : break;
2921 : : }
2922 : :
2923 [ - + ]: 130 : if (!restored)
2924 [ # # ]: 0 : ereport(FATAL,
2925 : : (errmsg("too many replication slots active before shutdown"),
2926 : : errhint("Increase \"max_replication_slots\" and try again.")));
2927 : : }
2928 : :
2929 : : /*
2930 : : * Maps an invalidation reason for a replication slot to
2931 : : * ReplicationSlotInvalidationCause.
2932 : : */
2933 : : ReplicationSlotInvalidationCause
2934 : 0 : GetSlotInvalidationCause(const char *cause_name)
2935 : : {
2936 : : Assert(cause_name);
2937 : :
2938 : : /* Search lookup table for the cause having this name */
2939 [ # # ]: 0 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2940 : : {
2941 [ # # ]: 0 : if (strcmp(SlotInvalidationCauses[i].cause_name, cause_name) == 0)
2942 : 0 : return SlotInvalidationCauses[i].cause;
2943 : : }
2944 : :
2945 : : Assert(false);
2946 : 0 : return RS_INVAL_NONE; /* to keep compiler quiet */
2947 : : }
2948 : :
2949 : : /*
2950 : : * Maps a ReplicationSlotInvalidationCause to the invalidation
2951 : : * reason for a replication slot.
2952 : : */
2953 : : const char *
2954 : 45 : GetSlotInvalidationCauseName(ReplicationSlotInvalidationCause cause)
2955 : : {
2956 : : /* Search lookup table for the name of this cause */
2957 [ + - ]: 146 : for (int i = 0; i <= RS_INVAL_MAX_CAUSES; i++)
2958 : : {
2959 [ + + ]: 146 : if (SlotInvalidationCauses[i].cause == cause)
2960 : 45 : return SlotInvalidationCauses[i].cause_name;
2961 : : }
2962 : :
2963 : : Assert(false);
2964 : 0 : return "none"; /* to keep compiler quiet */
2965 : : }
2966 : :
2967 : : /*
2968 : : * A helper function to validate slots specified in GUC synchronized_standby_slots.
2969 : : *
2970 : : * The rawname will be parsed, and the result will be saved into *elemlist.
2971 : : */
2972 : : static bool
2973 : 28 : validate_sync_standby_slots(char *rawname, List **elemlist)
2974 : : {
2975 : : /* Verify syntax and parse string into a list of identifiers */
2976 [ - + ]: 28 : if (!SplitIdentifierString(rawname, ',', elemlist))
2977 : : {
2978 : 0 : GUC_check_errdetail("List syntax is invalid.");
2979 : 0 : return false;
2980 : : }
2981 : :
2982 : : /* Iterate the list to validate each slot name */
2983 [ + - + + : 80 : foreach_ptr(char, name, *elemlist)
+ + ]
2984 : : {
2985 : : int err_code;
2986 : 28 : char *err_msg = NULL;
2987 : 28 : char *err_hint = NULL;
2988 : :
2989 [ + + ]: 28 : if (!ReplicationSlotValidateNameInternal(name, false, &err_code,
2990 : : &err_msg, &err_hint))
2991 : : {
2992 : 2 : GUC_check_errcode(err_code);
2993 : 2 : GUC_check_errdetail("%s", err_msg);
2994 [ + - ]: 2 : if (err_hint != NULL)
2995 : 2 : GUC_check_errhint("%s", err_hint);
2996 : 2 : return false;
2997 : : }
2998 : : }
2999 : :
3000 : 26 : return true;
3001 : : }
3002 : :
3003 : : /*
3004 : : * GUC check_hook for synchronized_standby_slots
3005 : : */
3006 : : bool
3007 : 1352 : check_synchronized_standby_slots(char **newval, void **extra, GucSource source)
3008 : : {
3009 : : char *rawname;
3010 : : char *ptr;
3011 : : List *elemlist;
3012 : : int size;
3013 : : bool ok;
3014 : : SyncStandbySlotsConfigData *config;
3015 : :
3016 [ + + ]: 1352 : if ((*newval)[0] == '\0')
3017 : 1324 : return true;
3018 : :
3019 : : /* Need a modifiable copy of the GUC string */
3020 : 28 : rawname = pstrdup(*newval);
3021 : :
3022 : : /* Now verify if the specified slots exist and have correct type */
3023 : 28 : ok = validate_sync_standby_slots(rawname, &elemlist);
3024 : :
3025 [ + + - + ]: 28 : if (!ok || elemlist == NIL)
3026 : : {
3027 : 2 : pfree(rawname);
3028 : 2 : list_free(elemlist);
3029 : 2 : return ok;
3030 : : }
3031 : :
3032 : : /* Compute the size required for the SyncStandbySlotsConfigData struct */
3033 : 26 : size = offsetof(SyncStandbySlotsConfigData, slot_names);
3034 [ + - + + : 78 : foreach_ptr(char, slot_name, elemlist)
+ + ]
3035 : 26 : size += strlen(slot_name) + 1;
3036 : :
3037 : : /* GUC extra value must be guc_malloc'd, not palloc'd */
3038 : 26 : config = (SyncStandbySlotsConfigData *) guc_malloc(LOG, size);
3039 [ - + ]: 26 : if (!config)
3040 : 0 : return false;
3041 : :
3042 : : /* Transform the data into SyncStandbySlotsConfigData */
3043 : 26 : config->nslotnames = list_length(elemlist);
3044 : :
3045 : 26 : ptr = config->slot_names;
3046 [ + - + + : 78 : foreach_ptr(char, slot_name, elemlist)
+ + ]
3047 : : {
3048 : 26 : strcpy(ptr, slot_name);
3049 : 26 : ptr += strlen(slot_name) + 1;
3050 : : }
3051 : :
3052 : 26 : *extra = config;
3053 : :
3054 : 26 : pfree(rawname);
3055 : 26 : list_free(elemlist);
3056 : 26 : return true;
3057 : : }
3058 : :
3059 : : /*
3060 : : * GUC assign_hook for synchronized_standby_slots
3061 : : */
3062 : : void
3063 : 1349 : assign_synchronized_standby_slots(const char *newval, void *extra)
3064 : : {
3065 : : /*
3066 : : * The standby slots may have changed, so we must recompute the oldest
3067 : : * LSN.
3068 : : */
3069 : 1349 : ss_oldest_flush_lsn = InvalidXLogRecPtr;
3070 : :
3071 : 1349 : synchronized_standby_slots_config = (SyncStandbySlotsConfigData *) extra;
3072 : 1349 : }
3073 : :
3074 : : /*
3075 : : * Check if the passed slot_name is specified in the synchronized_standby_slots GUC.
3076 : : */
3077 : : bool
3078 : 40408 : SlotExistsInSyncStandbySlots(const char *slot_name)
3079 : : {
3080 : : const char *standby_slot_name;
3081 : :
3082 : : /* Return false if there is no value in synchronized_standby_slots */
3083 [ + + ]: 40408 : if (synchronized_standby_slots_config == NULL)
3084 : 40392 : return false;
3085 : :
3086 : : /*
3087 : : * XXX: We are not expecting this list to be long so a linear search
3088 : : * shouldn't hurt but if that turns out not to be true then we can cache
3089 : : * this information for each WalSender as well.
3090 : : */
3091 : 16 : standby_slot_name = synchronized_standby_slots_config->slot_names;
3092 [ + + ]: 24 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
3093 : : {
3094 [ + + ]: 16 : if (strcmp(standby_slot_name, slot_name) == 0)
3095 : 8 : return true;
3096 : :
3097 : 8 : standby_slot_name += strlen(standby_slot_name) + 1;
3098 : : }
3099 : :
3100 : 8 : return false;
3101 : : }
3102 : :
3103 : : /*
3104 : : * Return true if the slots specified in synchronized_standby_slots have caught up to
3105 : : * the given WAL location, false otherwise.
3106 : : *
3107 : : * The elevel parameter specifies the error level used for logging messages
3108 : : * related to slots that do not exist, are invalidated, or are inactive.
3109 : : */
3110 : : bool
3111 : 2238 : StandbySlotsHaveCaughtup(XLogRecPtr wait_for_lsn, int elevel)
3112 : : {
3113 : : const char *name;
3114 : 2238 : int caught_up_slot_num = 0;
3115 : 2238 : XLogRecPtr min_restart_lsn = InvalidXLogRecPtr;
3116 : :
3117 : : /*
3118 : : * Don't need to wait for the standbys to catch up if there is no value in
3119 : : * synchronized_standby_slots.
3120 : : */
3121 [ + + ]: 2238 : if (synchronized_standby_slots_config == NULL)
3122 : 2203 : return true;
3123 : :
3124 : : /*
3125 : : * Don't need to wait for the standbys to catch up if we are on a standby
3126 : : * server, since we do not support syncing slots to cascading standbys.
3127 : : */
3128 [ - + ]: 35 : if (RecoveryInProgress())
3129 : 0 : return true;
3130 : :
3131 : : /*
3132 : : * Don't need to wait for the standbys to catch up if they are already
3133 : : * beyond the specified WAL location.
3134 : : */
3135 [ + + ]: 35 : if (XLogRecPtrIsValid(ss_oldest_flush_lsn) &&
3136 [ + + ]: 19 : ss_oldest_flush_lsn >= wait_for_lsn)
3137 : 9 : return true;
3138 : :
3139 : : /*
3140 : : * To prevent concurrent slot dropping and creation while filtering the
3141 : : * slots, take the ReplicationSlotControlLock outside of the loop.
3142 : : */
3143 : 26 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
3144 : :
3145 : 26 : name = synchronized_standby_slots_config->slot_names;
3146 [ + + ]: 35 : for (int i = 0; i < synchronized_standby_slots_config->nslotnames; i++)
3147 : : {
3148 : : XLogRecPtr restart_lsn;
3149 : : bool invalidated;
3150 : : bool inactive;
3151 : : ReplicationSlot *slot;
3152 : :
3153 : 26 : slot = SearchNamedReplicationSlot(name, false);
3154 : :
3155 : : /*
3156 : : * If a slot name provided in synchronized_standby_slots does not
3157 : : * exist, report a message and exit the loop.
3158 : : */
3159 [ - + ]: 26 : if (!slot)
3160 : : {
3161 [ # # ]: 0 : ereport(elevel,
3162 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3163 : : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not exist",
3164 : : name, "synchronized_standby_slots"),
3165 : : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3166 : : name),
3167 : : errhint("Create the replication slot \"%s\" or amend parameter \"%s\".",
3168 : : name, "synchronized_standby_slots"));
3169 : 0 : break;
3170 : : }
3171 : :
3172 : : /* Same as above: if a slot is not physical, exit the loop. */
3173 [ - + ]: 26 : if (SlotIsLogical(slot))
3174 : : {
3175 [ # # ]: 0 : ereport(elevel,
3176 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3177 : : errmsg("cannot specify logical replication slot \"%s\" in parameter \"%s\"",
3178 : : name, "synchronized_standby_slots"),
3179 : : errdetail("Logical replication is waiting for correction on replication slot \"%s\".",
3180 : : name),
3181 : : errhint("Remove the logical replication slot \"%s\" from parameter \"%s\".",
3182 : : name, "synchronized_standby_slots"));
3183 : 0 : break;
3184 : : }
3185 : :
3186 : 26 : SpinLockAcquire(&slot->mutex);
3187 : 26 : restart_lsn = slot->data.restart_lsn;
3188 : 26 : invalidated = slot->data.invalidated != RS_INVAL_NONE;
3189 : 26 : inactive = slot->active_proc == INVALID_PROC_NUMBER;
3190 : 26 : SpinLockRelease(&slot->mutex);
3191 : :
3192 [ - + ]: 26 : if (invalidated)
3193 : : {
3194 : : /* Specified physical slot has been invalidated */
3195 [ # # ]: 0 : ereport(elevel,
3196 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3197 : : errmsg("physical replication slot \"%s\" specified in parameter \"%s\" has been invalidated",
3198 : : name, "synchronized_standby_slots"),
3199 : : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3200 : : name),
3201 : : errhint("Drop and recreate the replication slot \"%s\", or amend parameter \"%s\".",
3202 : : name, "synchronized_standby_slots"));
3203 : 0 : break;
3204 : : }
3205 : :
3206 [ + + + + ]: 26 : if (!XLogRecPtrIsValid(restart_lsn) || restart_lsn < wait_for_lsn)
3207 : : {
3208 : : /* Log a message if no active_pid for this physical slot */
3209 [ + + ]: 17 : if (inactive)
3210 [ + - ]: 13 : ereport(elevel,
3211 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3212 : : errmsg("replication slot \"%s\" specified in parameter \"%s\" does not have active_pid",
3213 : : name, "synchronized_standby_slots"),
3214 : : errdetail("Logical replication is waiting on the standby associated with replication slot \"%s\".",
3215 : : name),
3216 : : errhint("Start the standby associated with the replication slot \"%s\", or amend parameter \"%s\".",
3217 : : name, "synchronized_standby_slots"));
3218 : :
3219 : : /* Continue if the current slot hasn't caught up. */
3220 : 16 : break;
3221 : : }
3222 : :
3223 : : Assert(restart_lsn >= wait_for_lsn);
3224 : :
3225 [ - + - - ]: 9 : if (!XLogRecPtrIsValid(min_restart_lsn) ||
3226 : : min_restart_lsn > restart_lsn)
3227 : 9 : min_restart_lsn = restart_lsn;
3228 : :
3229 : 9 : caught_up_slot_num++;
3230 : :
3231 : 9 : name += strlen(name) + 1;
3232 : : }
3233 : :
3234 : 25 : LWLockRelease(ReplicationSlotControlLock);
3235 : :
3236 : : /*
3237 : : * Return false if not all the standbys have caught up to the specified
3238 : : * WAL location.
3239 : : */
3240 [ + + ]: 25 : if (caught_up_slot_num != synchronized_standby_slots_config->nslotnames)
3241 : 16 : return false;
3242 : :
3243 : : /* The ss_oldest_flush_lsn must not retreat. */
3244 : : Assert(!XLogRecPtrIsValid(ss_oldest_flush_lsn) ||
3245 : : min_restart_lsn >= ss_oldest_flush_lsn);
3246 : :
3247 : 9 : ss_oldest_flush_lsn = min_restart_lsn;
3248 : :
3249 : 9 : return true;
3250 : : }
3251 : :
3252 : : /*
3253 : : * Wait for physical standbys to confirm receiving the given lsn.
3254 : : *
3255 : : * Used by logical decoding SQL functions. It waits for physical standbys
3256 : : * corresponding to the physical slots specified in the synchronized_standby_slots GUC.
3257 : : */
3258 : : void
3259 : 238 : WaitForStandbyConfirmation(XLogRecPtr wait_for_lsn)
3260 : : {
3261 : : /*
3262 : : * Don't need to wait for the standby to catch up if the current acquired
3263 : : * slot is not a logical failover slot, or there is no value in
3264 : : * synchronized_standby_slots.
3265 : : */
3266 [ + + + + ]: 238 : if (!MyReplicationSlot->data.failover || !synchronized_standby_slots_config)
3267 : 237 : return;
3268 : :
3269 : 1 : ConditionVariablePrepareToSleep(&WalSndCtl->wal_confirm_rcv_cv);
3270 : :
3271 : : for (;;)
3272 : : {
3273 [ - + ]: 2 : CHECK_FOR_INTERRUPTS();
3274 : :
3275 [ + + ]: 2 : if (ConfigReloadPending)
3276 : : {
3277 : 1 : ConfigReloadPending = false;
3278 : 1 : ProcessConfigFile(PGC_SIGHUP);
3279 : : }
3280 : :
3281 : : /* Exit if done waiting for every slot. */
3282 [ + + ]: 2 : if (StandbySlotsHaveCaughtup(wait_for_lsn, WARNING))
3283 : 1 : break;
3284 : :
3285 : : /*
3286 : : * Wait for the slots in the synchronized_standby_slots to catch up,
3287 : : * but use a timeout (1s) so we can also check if the
3288 : : * synchronized_standby_slots has been changed.
3289 : : */
3290 : 1 : ConditionVariableTimedSleep(&WalSndCtl->wal_confirm_rcv_cv, 1000,
3291 : : WAIT_EVENT_WAIT_FOR_STANDBY_CONFIRMATION);
3292 : : }
3293 : :
3294 : 1 : ConditionVariableCancelSleep();
3295 : : }
|
