Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : * slotsync.c
3 : : * Functionality for synchronizing slots to a standby server from the
4 : : * primary server.
5 : : *
6 : : * Copyright (c) 2024-2026, PostgreSQL Global Development Group
7 : : *
8 : : * IDENTIFICATION
9 : : * src/backend/replication/logical/slotsync.c
10 : : *
11 : : * This file contains the code for slot synchronization on a physical standby
12 : : * to fetch logical failover slots information from the primary server, create
13 : : * the slots on the standby and synchronize them periodically.
14 : : *
15 : : * Slot synchronization can be performed either automatically by enabling slot
16 : : * sync worker or manually by calling SQL function pg_sync_replication_slots().
17 : : *
18 : : * If the WAL corresponding to the remote's restart_lsn is not available on the
19 : : * physical standby or the remote's catalog_xmin precedes the oldest xid for
20 : : * which it is guaranteed that rows wouldn't have been removed then we cannot
21 : : * create the local standby slot because that would mean moving the local slot
22 : : * backward and decoding won't be possible via such a slot. In this case, the
23 : : * slot will be marked as RS_TEMPORARY. Once the primary server catches up,
24 : : * the slot will be marked as RS_PERSISTENT (which means sync-ready) after
25 : : * which slot sync worker can perform the sync periodically or user can call
26 : : * pg_sync_replication_slots() periodically to perform the syncs.
27 : : *
28 : : * If synchronized slots fail to build a consistent snapshot from the
29 : : * restart_lsn before reaching confirmed_flush_lsn, they would become
30 : : * unreliable after promotion due to potential data loss from changes
31 : : * before reaching a consistent point. This can happen because the slots can
32 : : * be synced at some random time and we may not reach the consistent point
33 : : * at the same WAL location as the primary. So, we mark such slots as
34 : : * RS_TEMPORARY. Once the decoding from corresponding LSNs can reach a
35 : : * consistent point, they will be marked as RS_PERSISTENT.
36 : : *
37 : : * If the WAL prior to the remote slot's confirmed_flush_lsn has not been
38 : : * flushed on the standby, the slot is marked as RS_TEMPORARY. Once the standby
39 : : * catches up and flushes that WAL, the slot will be marked as RS_PERSISTENT.
40 : : *
41 : : * The slot sync worker waits for some time before the next synchronization,
42 : : * with the duration varying based on whether any slots were updated during
43 : : * the last cycle. Refer to the comments above wait_for_slot_activity() for
44 : : * more details.
45 : : *
46 : : * If the SQL function pg_sync_replication_slots() is used to sync the slots,
47 : : * and if the slots are not ready to be synced and are marked as RS_TEMPORARY
48 : : * because of any of the reasons mentioned above, then the SQL function also
49 : : * waits and retries until the slots are marked as RS_PERSISTENT (which means
50 : : * sync-ready). Refer to the comments in SyncReplicationSlots() for more
51 : : * details.
52 : : *
53 : : * Any standby synchronized slots will be dropped if they no longer need
54 : : * to be synchronized. See comment atop drop_local_obsolete_slots() for more
55 : : * details.
56 : : *---------------------------------------------------------------------------
57 : : */
58 : :
59 : : #include "postgres.h"
60 : :
61 : : #include <time.h>
62 : :
63 : : #include "access/xlog_internal.h"
64 : : #include "access/xlogrecovery.h"
65 : : #include "catalog/pg_database.h"
66 : : #include "libpq/pqsignal.h"
67 : : #include "pgstat.h"
68 : : #include "postmaster/interrupt.h"
69 : : #include "replication/logical.h"
70 : : #include "replication/slotsync.h"
71 : : #include "replication/snapbuild.h"
72 : : #include "storage/ipc.h"
73 : : #include "storage/lmgr.h"
74 : : #include "storage/proc.h"
75 : : #include "storage/procarray.h"
76 : : #include "storage/subsystems.h"
77 : : #include "tcop/tcopprot.h"
78 : : #include "utils/builtins.h"
79 : : #include "utils/memutils.h"
80 : : #include "utils/pg_lsn.h"
81 : : #include "utils/ps_status.h"
82 : : #include "utils/timeout.h"
83 : : #include "utils/wait_event.h"
84 : :
85 : : /*
86 : : * Struct for sharing information to control slot synchronization.
87 : : *
88 : : * The 'pid' is either the slot sync worker's pid or the backend's pid running
89 : : * the SQL function pg_sync_replication_slots(). On promotion, the startup
90 : : * process sets 'stopSignaled' and uses this 'pid' to signal the synchronizing
91 : : * process with PROCSIG_SLOTSYNC_MESSAGE and also to wake it up so that the
92 : : * process can immediately stop its synchronizing work.
93 : : * Setting 'stopSignaled' on the other hand is used to handle the race
94 : : * condition when the postmaster has not noticed the promotion yet and thus may
95 : : * end up restarting the slot sync worker. If 'stopSignaled' is set, the worker
96 : : * will exit in such a case. The SQL function pg_sync_replication_slots() will
97 : : * also error out if this flag is set. Note that we don't need to reset this
98 : : * variable as after promotion the slot sync worker won't be restarted because
99 : : * the pmState changes to PM_RUN from PM_HOT_STANDBY and we don't support
100 : : * demoting primary without restarting the server.
101 : : * See LaunchMissingBackgroundProcesses.
102 : : *
103 : : * The 'syncing' flag is needed to prevent concurrent slot syncs to avoid slot
104 : : * overwrites.
105 : : *
106 : : * The 'last_start_time' is needed by postmaster to start the slot sync worker
107 : : * once per SLOTSYNC_RESTART_INTERVAL_SEC. In cases where an immediate restart
108 : : * is expected (e.g., slot sync GUCs change), slot sync worker will reset
109 : : * last_start_time before exiting, so that postmaster can start the worker
110 : : * without waiting for SLOTSYNC_RESTART_INTERVAL_SEC.
111 : : */
112 : : typedef struct SlotSyncCtxStruct
113 : : {
114 : : pid_t pid;
115 : : bool stopSignaled;
116 : : bool syncing;
117 : : time_t last_start_time;
118 : : slock_t mutex;
119 : : } SlotSyncCtxStruct;
120 : :
121 : : static SlotSyncCtxStruct *SlotSyncCtx = NULL;
122 : :
123 : : static void SlotSyncShmemRequest(void *arg);
124 : : static void SlotSyncShmemInit(void *arg);
125 : :
126 : : const ShmemCallbacks SlotSyncShmemCallbacks = {
127 : : .request_fn = SlotSyncShmemRequest,
128 : : .init_fn = SlotSyncShmemInit,
129 : : };
130 : :
131 : : /* GUC variable */
132 : : bool sync_replication_slots = false;
133 : :
134 : : /*
135 : : * The sleep time (ms) between slot-sync cycles varies dynamically
136 : : * (within a MIN/MAX range) according to slot activity. See
137 : : * wait_for_slot_activity() for details.
138 : : */
139 : : #define MIN_SLOTSYNC_WORKER_NAPTIME_MS 200
140 : : #define MAX_SLOTSYNC_WORKER_NAPTIME_MS 30000 /* 30s */
141 : :
142 : : static long sleep_ms = MIN_SLOTSYNC_WORKER_NAPTIME_MS;
143 : :
144 : : /* The restart interval for slot sync work used by postmaster */
145 : : #define SLOTSYNC_RESTART_INTERVAL_SEC 10
146 : :
147 : : /*
148 : : * Flag to tell if we are syncing replication slots. Unlike the 'syncing' flag
149 : : * in SlotSyncCtxStruct, this flag is true only if the current process is
150 : : * performing slot synchronization.
151 : : */
152 : : static bool syncing_slots = false;
153 : :
154 : : /*
155 : : * Interrupt flag set when PROCSIG_SLOTSYNC_MESSAGE is received, asking the
156 : : * slotsync worker or pg_sync_replication_slots() to stop because
157 : : * standby promotion has been triggered.
158 : : */
159 : : volatile sig_atomic_t SlotSyncShutdownPending = false;
160 : :
161 : : /*
162 : : * Structure to hold information fetched from the primary server about a logical
163 : : * replication slot.
164 : : */
165 : : typedef struct RemoteSlot
166 : : {
167 : : char *name;
168 : : char *plugin;
169 : : char *database;
170 : : bool two_phase;
171 : : bool failover;
172 : : XLogRecPtr restart_lsn;
173 : : XLogRecPtr confirmed_lsn;
174 : : XLogRecPtr two_phase_at;
175 : : TransactionId catalog_xmin;
176 : :
177 : : /* RS_INVAL_NONE if valid, or the reason of invalidation */
178 : : ReplicationSlotInvalidationCause invalidated;
179 : : } RemoteSlot;
180 : :
181 : : static void slotsync_failure_callback(int code, Datum arg);
182 : : static void update_synced_slots_inactive_since(void);
183 : :
184 : : /*
185 : : * Update slot sync skip stats. This function requires the caller to acquire
186 : : * the slot.
187 : : */
188 : : static void
214 akapila@postgresql.o 189 :GNC 46 : update_slotsync_skip_stats(SlotSyncSkipReason skip_reason)
190 : : {
191 : : ReplicationSlot *slot;
192 : :
193 [ - + ]: 46 : Assert(MyReplicationSlot);
194 : :
195 : 46 : slot = MyReplicationSlot;
196 : :
197 : : /*
198 : : * Update the slot sync related stats in pg_stat_replication_slots when a
199 : : * slot sync is skipped
200 : : */
201 [ + + ]: 46 : if (skip_reason != SS_SKIP_NONE)
202 : 8 : pgstat_report_replslotsync(slot);
203 : :
204 : : /* Update the slot sync skip reason */
205 [ + + ]: 46 : if (slot->slotsync_skip_reason != skip_reason)
206 : : {
207 : 4 : SpinLockAcquire(&slot->mutex);
208 : 4 : slot->slotsync_skip_reason = skip_reason;
209 : 4 : SpinLockRelease(&slot->mutex);
210 : : }
211 : 46 : }
212 : :
213 : : /*
214 : : * If necessary, update the local synced slot's metadata based on the data
215 : : * from the remote slot.
216 : : *
217 : : * If no update was needed (the data of the remote slot is the same as the
218 : : * local slot) return false, otherwise true.
219 : : */
220 : : static bool
138 221 : 46 : update_local_synced_slot(RemoteSlot *remote_slot, Oid remote_dbid)
222 : : {
867 akapila@postgresql.o 223 :CBC 46 : ReplicationSlot *slot = MyReplicationSlot;
809 224 : 46 : bool updated_xmin_or_lsn = false;
225 : 46 : bool updated_config = false;
214 akapila@postgresql.o 226 :GNC 46 : SlotSyncSkipReason skip_reason = SS_SKIP_NONE;
138 227 : 46 : XLogRecPtr latestFlushPtr = GetStandbyFlushRecPtr(NULL);
228 : :
867 akapila@postgresql.o 229 [ - + ]:CBC 46 : Assert(slot->data.invalidated == RS_INVAL_NONE);
230 : :
231 : : /*
232 : : * Make sure that concerned WAL is received and flushed before syncing
233 : : * slot to target lsn received from the primary server.
234 : : */
138 akapila@postgresql.o 235 [ - + ]:GNC 46 : if (remote_slot->confirmed_lsn > latestFlushPtr)
236 : : {
138 akapila@postgresql.o 237 :UNC 0 : update_slotsync_skip_stats(SS_SKIP_WAL_NOT_FLUSHED);
238 : :
239 : : /*
240 : : * Can get here only if GUC 'synchronized_standby_slots' on the
241 : : * primary server was not configured correctly.
242 : : */
116 243 [ # # ]: 0 : ereport(LOG,
244 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
245 : : errmsg("skipping slot synchronization because the received slot sync"
246 : : " LSN %X/%08X for slot \"%s\" is ahead of the standby position %X/%08X",
247 : : LSN_FORMAT_ARGS(remote_slot->confirmed_lsn),
248 : : remote_slot->name,
249 : : LSN_FORMAT_ARGS(latestFlushPtr)));
250 : :
138 251 : 0 : return false;
252 : : }
253 : :
254 : : /*
255 : : * Don't overwrite if we already have a newer catalog_xmin and
256 : : * restart_lsn.
257 : : */
809 akapila@postgresql.o 258 [ + + + + ]:CBC 86 : if (remote_slot->restart_lsn < slot->data.restart_lsn ||
259 : 40 : TransactionIdPrecedes(remote_slot->catalog_xmin,
260 : : slot->data.catalog_xmin))
261 : : {
262 : : /* Update slot sync skip stats */
214 akapila@postgresql.o 263 :GNC 8 : update_slotsync_skip_stats(SS_SKIP_WAL_OR_ROWS_REMOVED);
264 : :
265 : : /*
266 : : * This can happen in following situations:
267 : : *
268 : : * If the slot is temporary, it means either the initial WAL location
269 : : * reserved for the local slot is ahead of the remote slot's
270 : : * restart_lsn or the initial xmin_horizon computed for the local slot
271 : : * is ahead of the remote slot.
272 : : *
273 : : * If the slot is persistent, both restart_lsn and catalog_xmin of the
274 : : * synced slot could still be ahead of the remote slot. Since we use
275 : : * slot advance functionality to keep snapbuild/slot updated, it is
276 : : * possible that the restart_lsn and catalog_xmin are advanced to a
277 : : * later position than it has on the primary. This can happen when
278 : : * slot advancing machinery finds running xacts record after reaching
279 : : * the consistent state at a later point than the primary where it
280 : : * serializes the snapshot and updates the restart_lsn.
281 : : *
282 : : * We LOG the message if the slot is temporary as it can help the user
283 : : * to understand why the slot is not sync-ready. In the case of a
284 : : * persistent slot, it would be a more common case and won't directly
285 : : * impact the users, so we used DEBUG1 level to log the message.
286 : : */
809 akapila@postgresql.o 287 [ + - + - ]:CBC 8 : ereport(slot->data.persistency == RS_TEMPORARY ? LOG : DEBUG1,
288 : : errmsg("could not synchronize replication slot \"%s\"",
289 : : remote_slot->name),
290 : : errdetail("Synchronization could lead to data loss, because the remote slot needs WAL at LSN %X/%08X and catalog xmin %u, but the standby has LSN %X/%08X and catalog xmin %u.",
291 : : LSN_FORMAT_ARGS(remote_slot->restart_lsn),
292 : : remote_slot->catalog_xmin,
293 : : LSN_FORMAT_ARGS(slot->data.restart_lsn),
294 : : slot->data.catalog_xmin));
295 : :
296 : : /*
297 : : * Skip updating the configuration. This is required to avoid syncing
298 : : * two_phase_at without syncing confirmed_lsn. Otherwise, the prepared
299 : : * transaction between old confirmed_lsn and two_phase_at will
300 : : * unexpectedly get decoded and sent to the downstream after
301 : : * promotion. See comments in ReorderBufferFinishPrepared.
302 : : */
427 303 : 8 : return false;
304 : : }
305 : :
306 : : /*
307 : : * Attempt to sync LSNs and xmins only if remote slot is ahead of local
308 : : * slot.
309 : : */
310 [ + + ]: 38 : if (remote_slot->confirmed_lsn > slot->data.confirmed_flush ||
311 [ + + - + ]: 51 : remote_slot->restart_lsn > slot->data.restart_lsn ||
312 : 25 : TransactionIdFollows(remote_slot->catalog_xmin,
313 : : slot->data.catalog_xmin))
314 : : {
315 : : /*
316 : : * We can't directly copy the remote slot's LSN or xmin unless there
317 : : * exists a consistent snapshot at that point. Otherwise, after
318 : : * promotion, the slots may not reach a consistent point before the
319 : : * confirmed_flush_lsn which can lead to a data loss. To avoid data
320 : : * loss, we let slot machinery advance the slot which ensures that
321 : : * snapbuilder/slot statuses are updated properly.
322 : : */
818 323 [ + + ]: 13 : if (SnapBuildSnapshotExists(remote_slot->restart_lsn))
324 : : {
325 : : /*
326 : : * Update the slot info directly if there is a serialized snapshot
327 : : * at the restart_lsn, as the slot can quickly reach consistency
328 : : * at restart_lsn by restoring the snapshot.
329 : : */
330 [ - + ]: 3 : SpinLockAcquire(&slot->mutex);
331 : 3 : slot->data.restart_lsn = remote_slot->restart_lsn;
332 : 3 : slot->data.confirmed_flush = remote_slot->confirmed_lsn;
333 : 3 : slot->data.catalog_xmin = remote_slot->catalog_xmin;
334 : 3 : SpinLockRelease(&slot->mutex);
335 : :
78 336 : 3 : updated_xmin_or_lsn = true;
337 : : }
338 : : else
339 : : {
340 : : bool found_consistent_snapshot;
341 : 10 : XLogRecPtr old_confirmed_lsn = slot->data.confirmed_flush;
342 : 10 : XLogRecPtr old_restart_lsn = slot->data.restart_lsn;
21 fujii@postgresql.org 343 : 10 : TransactionId old_catalog_xmin = slot->data.catalog_xmin;
344 : :
818 akapila@postgresql.o 345 : 10 : LogicalSlotAdvanceAndCheckSnapState(remote_slot->confirmed_lsn,
346 : : &found_consistent_snapshot);
347 : :
348 : : /* Sanity check */
809 349 [ - + ]: 10 : if (slot->data.confirmed_flush != remote_slot->confirmed_lsn)
809 akapila@postgresql.o 350 [ # # ]:UBC 0 : ereport(ERROR,
351 : : errmsg_internal("synchronized confirmed_flush for slot \"%s\" differs from remote slot",
352 : : remote_slot->name),
353 : : errdetail_internal("Remote slot has LSN %X/%08X but local slot has LSN %X/%08X.",
354 : : LSN_FORMAT_ARGS(remote_slot->confirmed_lsn),
355 : : LSN_FORMAT_ARGS(slot->data.confirmed_flush)));
356 : :
357 : : /*
358 : : * If we can't reach a consistent snapshot, the slot won't be
359 : : * persisted. See update_and_persist_local_synced_slot().
360 : : */
138 akapila@postgresql.o 361 [ - + ]:GNC 10 : if (!found_consistent_snapshot)
362 : : {
138 akapila@postgresql.o 363 [ # # ]:UNC 0 : Assert(MyReplicationSlot->data.persistency == RS_TEMPORARY);
364 : :
365 [ # # ]: 0 : ereport(LOG,
366 : : errmsg("could not synchronize replication slot \"%s\"",
367 : : remote_slot->name),
368 : : errdetail("Synchronization could lead to data loss, because the standby could not build a consistent snapshot to decode WALs at LSN %X/%08X.",
369 : : LSN_FORMAT_ARGS(slot->data.restart_lsn)));
370 : :
214 371 : 0 : skip_reason = SS_SKIP_NO_CONSISTENT_SNAPSHOT;
372 : : }
373 : :
374 : : /*
375 : : * It is possible that the slot's xmin or LSNs are not updated,
376 : : * when the synced slot has reached consistent snapshot state or
377 : : * cannot build one at all.
378 : : */
78 akapila@postgresql.o 379 :CBC 10 : updated_xmin_or_lsn = (old_confirmed_lsn != slot->data.confirmed_flush ||
380 [ - + - - ]: 10 : old_restart_lsn != slot->data.restart_lsn ||
78 akapila@postgresql.o 381 [ # # ]:UBC 0 : old_catalog_xmin != slot->data.catalog_xmin);
382 : : }
383 : : }
384 : :
385 : : /* Update slot sync skip stats */
214 akapila@postgresql.o 386 :GNC 38 : update_slotsync_skip_stats(skip_reason);
387 : :
818 akapila@postgresql.o 388 [ + - ]:CBC 38 : if (remote_dbid != slot->data.database ||
389 [ + + ]: 38 : remote_slot->two_phase != slot->data.two_phase ||
390 [ + - ]: 37 : remote_slot->failover != slot->data.failover ||
453 391 [ + - ]: 37 : strcmp(remote_slot->plugin, NameStr(slot->data.plugin)) != 0 ||
392 [ - + ]: 37 : remote_slot->two_phase_at != slot->data.two_phase_at)
393 : : {
394 : : NameData plugin_name;
395 : :
396 : : /* Avoid expensive operations while holding a spinlock. */
818 397 : 1 : namestrcpy(&plugin_name, remote_slot->plugin);
398 : :
399 [ - + ]: 1 : SpinLockAcquire(&slot->mutex);
400 : 1 : slot->data.plugin = plugin_name;
401 : 1 : slot->data.database = remote_dbid;
402 : 1 : slot->data.two_phase = remote_slot->two_phase;
453 403 : 1 : slot->data.two_phase_at = remote_slot->two_phase_at;
818 404 : 1 : slot->data.failover = remote_slot->failover;
405 : 1 : SpinLockRelease(&slot->mutex);
406 : :
809 407 : 1 : updated_config = true;
408 : :
409 : : /*
410 : : * Ensure that there is no risk of sending prepared transactions
411 : : * unexpectedly after the promotion.
412 : : */
427 413 [ - + ]: 1 : Assert(slot->data.two_phase_at <= slot->data.confirmed_flush);
414 : : }
415 : :
416 : : /*
417 : : * We have to write the changed xmin to disk *before* we change the
418 : : * in-memory value, otherwise after a crash we wouldn't know that some
419 : : * catalog tuples might have been removed already.
420 : : */
809 421 [ + + + + ]: 38 : if (updated_config || updated_xmin_or_lsn)
422 : : {
423 : 14 : ReplicationSlotMarkDirty();
424 : 14 : ReplicationSlotSave();
425 : : }
426 : :
427 : : /*
428 : : * Now the new xmin is safely on disk, we can let the global value
429 : : * advance. We do not take ProcArrayLock or similar since we only advance
430 : : * xmin here and there's not much harm done by a concurrent computation
431 : : * missing that.
432 : : */
433 [ + + ]: 38 : if (updated_xmin_or_lsn)
434 : : {
435 [ - + ]: 13 : SpinLockAcquire(&slot->mutex);
436 : 13 : slot->effective_catalog_xmin = remote_slot->catalog_xmin;
437 : 13 : SpinLockRelease(&slot->mutex);
438 : :
439 : 13 : ReplicationSlotsComputeRequiredXmin(false);
440 : 13 : ReplicationSlotsComputeRequiredLSN();
441 : : }
442 : :
443 [ + + + + ]: 38 : return updated_config || updated_xmin_or_lsn;
444 : : }
445 : :
446 : : /*
447 : : * Get the list of local logical slots that are synchronized from the
448 : : * primary server.
449 : : */
450 : : static List *
867 451 : 29 : get_local_synced_slots(void)
452 : : {
453 : 29 : List *local_slots = NIL;
454 : :
455 : 29 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
456 : :
84 alvherre@kurilemu.de 457 [ + + ]:GNC 464 : for (int i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
458 : : {
867 akapila@postgresql.o 459 :CBC 435 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
460 : :
461 : : /* Check if it is a synchronized slot */
462 [ + + + + ]: 435 : if (s->in_use && s->data.synced)
463 : : {
464 [ - + ]: 40 : Assert(SlotIsLogical(s));
465 : 40 : local_slots = lappend(local_slots, s);
466 : : }
467 : : }
468 : :
469 : 29 : LWLockRelease(ReplicationSlotControlLock);
470 : :
471 : 29 : return local_slots;
472 : : }
473 : :
474 : : /*
475 : : * Helper function to check if local_slot is required to be retained.
476 : : *
477 : : * Return false either if local_slot does not exist in the remote_slots list
478 : : * or is invalidated while the corresponding remote slot is still valid,
479 : : * otherwise true.
480 : : */
481 : : static bool
482 : 40 : local_sync_slot_required(ReplicationSlot *local_slot, List *remote_slots)
483 : : {
484 : 40 : bool remote_exists = false;
485 : 40 : bool locally_invalidated = false;
486 : :
487 [ + - + + : 96 : foreach_ptr(RemoteSlot, remote_slot, remote_slots)
+ + ]
488 : : {
489 [ + + ]: 55 : if (strcmp(remote_slot->name, NameStr(local_slot->data.name)) == 0)
490 : : {
491 : 39 : remote_exists = true;
492 : :
493 : : /*
494 : : * If remote slot is not invalidated but local slot is marked as
495 : : * invalidated, then set locally_invalidated flag.
496 : : */
497 [ - + ]: 39 : SpinLockAcquire(&local_slot->mutex);
498 : 39 : locally_invalidated =
499 [ + - ]: 78 : (remote_slot->invalidated == RS_INVAL_NONE) &&
500 [ + + ]: 39 : (local_slot->data.invalidated != RS_INVAL_NONE);
501 : 39 : SpinLockRelease(&local_slot->mutex);
502 : :
503 : 39 : break;
504 : : }
505 : : }
506 : :
507 [ + + + + ]: 40 : return (remote_exists && !locally_invalidated);
508 : : }
509 : :
510 : : /*
511 : : * Drop local obsolete slots.
512 : : *
513 : : * Drop the local slots that no longer need to be synced i.e. these either do
514 : : * not exist on the primary or are no longer enabled for failover.
515 : : *
516 : : * Additionally, drop any slots that are valid on the primary but got
517 : : * invalidated on the standby. This situation may occur due to the following
518 : : * reasons:
519 : : * - The 'max_slot_wal_keep_size' on the standby is insufficient to retain WAL
520 : : * records from the restart_lsn of the slot.
521 : : * - 'primary_slot_name' is temporarily reset to null and the physical slot is
522 : : * removed.
523 : : * These dropped slots will get recreated in next sync-cycle and it is okay to
524 : : * drop and recreate such slots as long as these are not consumable on the
525 : : * standby (which is the case currently).
526 : : *
527 : : * Note: Change of 'wal_level' on the primary server to a level lower than
528 : : * logical may also result in slot invalidation and removal on the standby.
529 : : * This is because such 'wal_level' change is only possible if the logical
530 : : * slots are removed on the primary server, so it's expected to see the
531 : : * slots being invalidated and removed on the standby too (and re-created
532 : : * if they are re-created on the primary server).
533 : : */
534 : : static void
535 : 29 : drop_local_obsolete_slots(List *remote_slot_list)
536 : : {
537 : 29 : List *local_slots = get_local_synced_slots();
538 : :
539 [ + + + + : 98 : foreach_ptr(ReplicationSlot, local_slot, local_slots)
+ + ]
540 : : {
541 : : /* Drop the local slot if it is not required to be retained. */
542 [ + + ]: 40 : if (!local_sync_slot_required(local_slot, remote_slot_list))
543 : : {
12 544 : 2 : Oid slot_database = local_slot->data.database;
545 : : bool synced_slot;
546 : :
547 : : /*
548 : : * Use shared lock to prevent a conflict with
549 : : * ReplicationSlotsDropDBSlots(), trying to drop the same slot
550 : : * during a drop-database operation.
551 : : */
552 : 2 : LockSharedObject(DatabaseRelationId, slot_database, 0,
553 : : AccessShareLock);
554 : :
555 : : /*
556 : : * In the small window between getting the slot to drop and
557 : : * locking the database, there is a possibility of a parallel
558 : : * database drop by the startup process and the creation of a new
559 : : * slot by the user. This new user-created slot may end up using
560 : : * the same shared memory as that of 'local_slot'. Thus check if
561 : : * local_slot is still the synced one before performing the actual
562 : : * drop.
563 : : */
867 564 [ - + ]: 2 : SpinLockAcquire(&local_slot->mutex);
565 [ + - + - ]: 2 : synced_slot = local_slot->in_use && local_slot->data.synced;
566 : 2 : SpinLockRelease(&local_slot->mutex);
567 : :
568 [ + - ]: 2 : if (synced_slot)
569 : : {
12 570 : 2 : NameData slot_name = local_slot->data.name;
571 : :
572 : : /*
573 : : * Now acquire and drop the slot. Note we purposely don't
574 : : * request logical decoding to be disabled here: since this is
575 : : * a standby, which derives its logical decoding state from
576 : : * the primary, it would be wrong to do so.
577 : : */
578 : 2 : ReplicationSlotAcquire(NameStr(slot_name), true, false);
34 alvherre@kurilemu.de 579 :GNC 2 : ReplicationSlotDropAcquired(false);
580 : :
12 akapila@postgresql.o 581 [ + - ]:CBC 2 : ereport(LOG,
582 : : errmsg("dropped replication slot \"%s\" of database with OID %u",
583 : : NameStr(slot_name),
584 : : slot_database));
585 : : }
586 : :
587 : 2 : UnlockSharedObject(DatabaseRelationId, slot_database, 0,
588 : : AccessShareLock);
589 : : }
590 : : }
867 591 : 29 : }
592 : :
593 : : /*
594 : : * Reserve WAL for the currently active local slot using the specified WAL
595 : : * location (restart_lsn).
596 : : *
597 : : * If the given WAL location has been removed or is at risk of removal,
598 : : * reserve WAL using the oldest segment that is non-removable.
599 : : */
600 : : static void
601 : 8 : reserve_wal_for_local_slot(XLogRecPtr restart_lsn)
602 : : {
603 : : XLogRecPtr slot_min_lsn;
604 : : XLogRecPtr min_safe_lsn;
605 : : XLogSegNo segno;
606 : 8 : ReplicationSlot *slot = MyReplicationSlot;
607 : :
608 [ - + ]: 8 : Assert(slot != NULL);
236 alvherre@kurilemu.de 609 [ - + ]: 8 : Assert(!XLogRecPtrIsValid(slot->data.restart_lsn));
610 : :
611 : : /*
612 : : * Acquire an exclusive lock to prevent the checkpoint process from
613 : : * concurrently calculating the minimum slot LSN (see
614 : : * CheckPointReplicationSlots), ensuring that if WAL reservation occurs
615 : : * first, the checkpoint must wait for the restart_lsn update before
616 : : * calculating the minimum LSN.
617 : : *
618 : : * Note: Unlike ReplicationSlotReserveWal(), this lock does not protect a
619 : : * newly synced slot from being invalidated if a concurrent checkpoint has
620 : : * invoked CheckPointReplicationSlots() before the WAL reservation here.
621 : : * This can happen because the initial restart_lsn received from the
622 : : * remote server can precede the redo pointer. Therefore, when selecting
623 : : * the initial restart_lsn, we consider using the redo pointer or the
624 : : * minimum slot LSN (if those values are greater than the remote
625 : : * restart_lsn) instead of relying solely on the remote value.
626 : : */
154 akapila@postgresql.o 627 : 8 : LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
628 : :
629 : : /*
630 : : * Determine the minimum non-removable LSN by comparing the redo pointer
631 : : * with the minimum slot LSN.
632 : : *
633 : : * The minimum slot LSN is considered because the redo pointer advances at
634 : : * every checkpoint, even when replication slots are present on the
635 : : * standby. In such scenarios, the redo pointer can exceed the remote
636 : : * restart_lsn, while WALs preceding the remote restart_lsn remain
637 : : * protected by a local replication slot.
638 : : */
639 : 8 : min_safe_lsn = GetRedoRecPtr();
640 : 8 : slot_min_lsn = XLogGetReplicationSlotMinimumLSN();
641 : :
642 [ + + - + ]: 8 : if (XLogRecPtrIsValid(slot_min_lsn) && min_safe_lsn > slot_min_lsn)
154 akapila@postgresql.o 643 :UBC 0 : min_safe_lsn = slot_min_lsn;
644 : :
645 : : /*
646 : : * If the minimum safe LSN is greater than the given restart_lsn, use it
647 : : * as the initial restart_lsn for the newly synced slot. Otherwise, use
648 : : * the given remote restart_lsn.
649 : : */
154 akapila@postgresql.o 650 [ - + ]:CBC 8 : SpinLockAcquire(&slot->mutex);
651 : 8 : slot->data.restart_lsn = Max(restart_lsn, min_safe_lsn);
652 : 8 : SpinLockRelease(&slot->mutex);
653 : :
654 : 8 : ReplicationSlotsComputeRequiredLSN();
655 : :
656 : 8 : XLByteToSeg(slot->data.restart_lsn, segno, wal_segment_size);
657 [ - + ]: 8 : if (XLogGetLastRemovedSegno() >= segno)
154 akapila@postgresql.o 658 [ # # ]:UBC 0 : elog(ERROR, "WAL required by replication slot %s has been removed concurrently",
659 : : NameStr(slot->data.name));
660 : :
154 akapila@postgresql.o 661 :CBC 8 : LWLockRelease(ReplicationSlotAllocationLock);
867 662 : 8 : }
663 : :
664 : : /*
665 : : * If the remote restart_lsn and catalog_xmin have caught up with the
666 : : * local ones, then update the LSNs and persist the local synced slot for
667 : : * future synchronization; otherwise, do nothing.
668 : : *
669 : : * *slot_persistence_pending is set to true if any of the slots fail to
670 : : * persist.
671 : : *
672 : : * Return true if the slot is marked as RS_PERSISTENT (sync-ready), otherwise
673 : : * false.
674 : : */
675 : : static bool
197 akapila@postgresql.o 676 :GNC 14 : update_and_persist_local_synced_slot(RemoteSlot *remote_slot, Oid remote_dbid,
677 : : bool *slot_persistence_pending)
678 : : {
867 akapila@postgresql.o 679 :CBC 14 : ReplicationSlot *slot = MyReplicationSlot;
680 : :
681 : : /* Slotsync skip stats are handled in function update_local_synced_slot() */
138 akapila@postgresql.o 682 :GNC 14 : (void) update_local_synced_slot(remote_slot, remote_dbid);
683 : :
684 : : /*
685 : : * Check if the slot cannot be synchronized. Refer to the comment atop the
686 : : * file for details on this check.
687 : : */
688 [ + + ]: 14 : if (slot->slotsync_skip_reason != SS_SKIP_NONE)
689 : : {
690 : : /*
691 : : * We reach this point when the remote slot didn't catch up to locally
692 : : * reserved position, or it cannot reach the consistent point from the
693 : : * restart_lsn, or the WAL prior to the remote confirmed flush LSN has
694 : : * not been received and flushed.
695 : : *
696 : : * We do not drop the slot because the restart_lsn and confirmed_lsn
697 : : * can be ahead of the current location when recreating the slot in
698 : : * the next cycle. It may take more time to create such a slot or
699 : : * reach the consistent point. Therefore, we keep this slot and
700 : : * attempt the synchronization in the next cycle.
701 : : *
702 : : * We also update the slot_persistence_pending parameter, so the SQL
703 : : * function can retry.
704 : : */
197 705 [ + + ]: 8 : if (slot_persistence_pending)
706 : 2 : *slot_persistence_pending = true;
707 : :
859 akapila@postgresql.o 708 :GBC 8 : return false;
709 : : }
710 : :
867 akapila@postgresql.o 711 :CBC 6 : ReplicationSlotPersist();
712 : :
713 [ + - ]: 6 : ereport(LOG,
714 : : errmsg("newly created replication slot \"%s\" is sync-ready now",
715 : : remote_slot->name));
716 : :
859 717 : 6 : return true;
718 : : }
719 : :
720 : : /*
721 : : * Synchronize a single slot to the given position.
722 : : *
723 : : * This creates a new slot if there is no existing one and updates the
724 : : * metadata of the slot as per the data received from the primary server.
725 : : *
726 : : * The slot is created as a temporary slot and stays in the same state until the
727 : : * remote_slot catches up with locally reserved position and local slot is
728 : : * updated. The slot is then persisted and is considered as sync-ready for
729 : : * periodic syncs.
730 : : *
731 : : * *slot_persistence_pending is set to true if any of the slots fail to
732 : : * persist.
733 : : *
734 : : * Returns TRUE if the local slot is updated.
735 : : */
736 : : static bool
197 akapila@postgresql.o 737 :GNC 46 : synchronize_one_slot(RemoteSlot *remote_slot, Oid remote_dbid,
738 : : bool *slot_persistence_pending)
739 : : {
740 : : ReplicationSlot *slot;
859 akapila@postgresql.o 741 :CBC 46 : bool slot_updated = false;
742 : :
743 : : /* Search for the named slot */
867 744 [ + + ]: 46 : if ((slot = SearchNamedReplicationSlot(remote_slot->name, true)))
745 : : {
746 : : bool synced;
747 : :
748 [ - + ]: 38 : SpinLockAcquire(&slot->mutex);
749 : 38 : synced = slot->data.synced;
750 : 38 : SpinLockRelease(&slot->mutex);
751 : :
752 : : /* User-created slot with the same name exists, raise ERROR. */
753 [ - + ]: 38 : if (!synced)
867 akapila@postgresql.o 754 [ # # ]:UBC 0 : ereport(ERROR,
755 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
756 : : errmsg("exiting from slot synchronization because same"
757 : : " name slot \"%s\" already exists on the standby",
758 : : remote_slot->name));
759 : :
760 : : /*
761 : : * The slot has been synchronized before.
762 : : *
763 : : * It is important to acquire the slot here before checking
764 : : * invalidation. If we don't acquire the slot first, there could be a
765 : : * race condition that the local slot could be invalidated just after
766 : : * checking the 'invalidated' flag here and we could end up
767 : : * overwriting 'invalidated' flag to remote_slot's value. See
768 : : * InvalidatePossiblyObsoleteSlot() where it invalidates slot directly
769 : : * if the slot is not acquired by other processes.
770 : : *
771 : : * XXX: If it ever turns out that slot acquire/release is costly for
772 : : * cases when none of the slot properties is changed then we can do a
773 : : * pre-check to ensure that at least one of the slot properties is
774 : : * changed before acquiring the slot.
775 : : */
515 akapila@postgresql.o 776 :CBC 38 : ReplicationSlotAcquire(remote_slot->name, true, false);
777 : :
867 778 [ - + ]: 38 : Assert(slot == MyReplicationSlot);
779 : :
780 : : /*
781 : : * Copy the invalidation cause from remote only if local slot is not
782 : : * invalidated locally, we don't want to overwrite existing one.
783 : : */
784 [ + - ]: 38 : if (slot->data.invalidated == RS_INVAL_NONE &&
785 [ - + ]: 38 : remote_slot->invalidated != RS_INVAL_NONE)
786 : : {
867 akapila@postgresql.o 787 [ # # ]:UBC 0 : SpinLockAcquire(&slot->mutex);
788 : 0 : slot->data.invalidated = remote_slot->invalidated;
789 : 0 : SpinLockRelease(&slot->mutex);
790 : :
791 : : /* Make sure the invalidated state persists across server restart */
792 : 0 : ReplicationSlotMarkDirty();
793 : 0 : ReplicationSlotSave();
794 : :
859 795 : 0 : slot_updated = true;
796 : : }
797 : :
798 : : /* Skip the sync of an invalidated slot */
867 akapila@postgresql.o 799 [ - + ]:CBC 38 : if (slot->data.invalidated != RS_INVAL_NONE)
800 : : {
214 akapila@postgresql.o 801 :UNC 0 : update_slotsync_skip_stats(SS_SKIP_INVALID);
802 : :
867 akapila@postgresql.o 803 :UBC 0 : ReplicationSlotRelease();
859 804 : 0 : return slot_updated;
805 : : }
806 : :
807 : : /* Slot not ready yet, let's attempt to make it sync-ready now. */
867 akapila@postgresql.o 808 [ + + ]:CBC 38 : if (slot->data.persistency == RS_TEMPORARY)
809 : : {
859 810 : 6 : slot_updated = update_and_persist_local_synced_slot(remote_slot,
811 : : remote_dbid,
812 : : slot_persistence_pending);
813 : : }
814 : :
815 : : /* Slot ready for sync, so sync it. */
816 : : else
817 : : {
818 : : /*
819 : : * Sanity check: As long as the invalidations are handled
820 : : * appropriately as above, this should never happen.
821 : : *
822 : : * We don't need to check restart_lsn here. See the comments in
823 : : * update_local_synced_slot() for details.
824 : : */
809 825 [ - + ]: 32 : if (remote_slot->confirmed_lsn < slot->data.confirmed_flush)
809 akapila@postgresql.o 826 [ # # ]:UBC 0 : ereport(ERROR,
827 : : errmsg_internal("cannot synchronize local slot \"%s\"",
828 : : remote_slot->name),
829 : : errdetail_internal("Local slot's start streaming location LSN(%X/%08X) is ahead of remote slot's LSN(%X/%08X).",
830 : : LSN_FORMAT_ARGS(slot->data.confirmed_flush),
831 : : LSN_FORMAT_ARGS(remote_slot->confirmed_lsn)));
832 : :
138 akapila@postgresql.o 833 :GNC 32 : slot_updated = update_local_synced_slot(remote_slot, remote_dbid);
834 : : }
835 : : }
836 : : /* Otherwise create the slot first. */
837 : : else
838 : : {
839 : : NameData plugin_name;
867 akapila@postgresql.o 840 :CBC 8 : TransactionId xmin_horizon = InvalidTransactionId;
841 : :
842 : : /* Skip creating the local slot if remote_slot is invalidated already */
843 [ - + ]: 8 : if (remote_slot->invalidated != RS_INVAL_NONE)
859 akapila@postgresql.o 844 :UBC 0 : return false;
845 : :
846 : : /*
847 : : * We create temporary slots instead of ephemeral slots here because
848 : : * we want the slots to survive after releasing them. This is done to
849 : : * avoid dropping and re-creating the slots in each synchronization
850 : : * cycle if the restart_lsn or catalog_xmin of the remote slot has not
851 : : * caught up.
852 : : */
867 akapila@postgresql.o 853 :CBC 8 : ReplicationSlotCreate(remote_slot->name, true, RS_TEMPORARY,
854 : 8 : remote_slot->two_phase,
855 : : false,
856 : 8 : remote_slot->failover,
857 : : true);
858 : :
859 : : /* For shorter lines. */
860 : 8 : slot = MyReplicationSlot;
861 : :
862 : : /* Avoid expensive operations while holding a spinlock. */
863 : 8 : namestrcpy(&plugin_name, remote_slot->plugin);
864 : :
865 [ - + ]: 8 : SpinLockAcquire(&slot->mutex);
866 : 8 : slot->data.database = remote_dbid;
867 : 8 : slot->data.plugin = plugin_name;
868 : 8 : SpinLockRelease(&slot->mutex);
869 : :
870 : 8 : reserve_wal_for_local_slot(remote_slot->restart_lsn);
871 : :
182 msawada@postgresql.o 872 : 8 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
867 akapila@postgresql.o 873 : 8 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
874 : 8 : xmin_horizon = GetOldestSafeDecodingTransactionId(true);
875 [ - + ]: 8 : SpinLockAcquire(&slot->mutex);
876 : 8 : slot->effective_catalog_xmin = xmin_horizon;
877 : 8 : slot->data.catalog_xmin = xmin_horizon;
878 : 8 : SpinLockRelease(&slot->mutex);
879 : 8 : ReplicationSlotsComputeRequiredXmin(true);
880 : 8 : LWLockRelease(ProcArrayLock);
182 msawada@postgresql.o 881 : 8 : LWLockRelease(ReplicationSlotControlLock);
882 : :
197 akapila@postgresql.o 883 :GNC 8 : update_and_persist_local_synced_slot(remote_slot, remote_dbid,
884 : : slot_persistence_pending);
885 : :
859 akapila@postgresql.o 886 :CBC 8 : slot_updated = true;
887 : : }
888 : :
867 889 : 46 : ReplicationSlotRelease();
890 : :
859 891 : 46 : return slot_updated;
892 : : }
893 : :
894 : : /*
895 : : * Fetch remote slots.
896 : : *
897 : : * If slot_names is NIL, fetches all failover logical slots from the
898 : : * primary server, otherwise fetches only the ones with names in slot_names.
899 : : *
900 : : * Returns a list of remote slot information structures, or NIL if none
901 : : * are found.
902 : : */
903 : : static List *
197 akapila@postgresql.o 904 :GNC 31 : fetch_remote_slots(WalReceiverConn *wrconn, List *slot_names)
905 : : {
906 : : #define SLOTSYNC_COLUMN_COUNT 10
867 akapila@postgresql.o 907 :CBC 31 : Oid slotRow[SLOTSYNC_COLUMN_COUNT] = {TEXTOID, TEXTOID, LSNOID,
908 : : LSNOID, XIDOID, BOOLOID, LSNOID, BOOLOID, TEXTOID, TEXTOID};
909 : :
910 : : WalRcvExecResult *res;
911 : : TupleTableSlot *tupslot;
912 : 31 : List *remote_slot_list = NIL;
913 : : StringInfoData query;
914 : :
197 akapila@postgresql.o 915 :GNC 31 : initStringInfo(&query);
916 : 31 : appendStringInfoString(&query,
917 : : "SELECT slot_name, plugin, confirmed_flush_lsn,"
918 : : " restart_lsn, catalog_xmin, two_phase,"
919 : : " two_phase_at, failover,"
920 : : " database, invalidation_reason"
921 : : " FROM pg_catalog.pg_replication_slots"
922 : : " WHERE failover and NOT temporary");
923 : :
924 [ + + ]: 31 : if (slot_names != NIL)
925 : : {
926 : 2 : bool first_slot = true;
927 : :
928 : : /*
929 : : * Construct the query to fetch only the specified slots
930 : : */
931 : 2 : appendStringInfoString(&query, " AND slot_name IN (");
932 : :
933 [ + - + + : 6 : foreach_ptr(char, slot_name, slot_names)
+ + ]
934 : : {
935 [ - + ]: 2 : if (!first_slot)
197 akapila@postgresql.o 936 :UNC 0 : appendStringInfoString(&query, ", ");
937 : :
78 drowley@postgresql.o 938 :GNC 2 : appendStringInfoString(&query, quote_literal_cstr(slot_name));
197 akapila@postgresql.o 939 : 2 : first_slot = false;
940 : : }
941 : 2 : appendStringInfoChar(&query, ')');
942 : : }
943 : :
944 : : /* Execute the query */
945 : 31 : res = walrcv_exec(wrconn, query.data, SLOTSYNC_COLUMN_COUNT, slotRow);
946 : 31 : pfree(query.data);
867 akapila@postgresql.o 947 [ + + ]:CBC 31 : if (res->status != WALRCV_OK_TUPLES)
948 [ + - ]: 2 : ereport(ERROR,
949 : : errmsg("could not fetch failover logical slots info from the primary server: %s",
950 : : res->err));
951 : :
952 : 29 : tupslot = MakeSingleTupleTableSlot(res->tupledesc, &TTSOpsMinimalTuple);
953 [ + + ]: 75 : while (tuplestore_gettupleslot(res->tuplestore, true, false, tupslot))
954 : : {
955 : : bool isnull;
202 michael@paquier.xyz 956 :GNC 46 : RemoteSlot *remote_slot = palloc0_object(RemoteSlot);
957 : : Datum d;
867 akapila@postgresql.o 958 :CBC 46 : int col = 0;
959 : :
960 : 46 : remote_slot->name = TextDatumGetCString(slot_getattr(tupslot, ++col,
961 : : &isnull));
962 [ - + ]: 46 : Assert(!isnull);
963 : :
964 : 46 : remote_slot->plugin = TextDatumGetCString(slot_getattr(tupslot, ++col,
965 : : &isnull));
966 [ - + ]: 46 : Assert(!isnull);
967 : :
968 : : /*
969 : : * It is possible to get null values for LSN and Xmin if slot is
970 : : * invalidated on the primary server, so handle accordingly.
971 : : */
972 : 46 : d = slot_getattr(tupslot, ++col, &isnull);
973 [ + - ]: 46 : remote_slot->confirmed_lsn = isnull ? InvalidXLogRecPtr :
974 : 46 : DatumGetLSN(d);
975 : :
976 : 46 : d = slot_getattr(tupslot, ++col, &isnull);
977 [ + - ]: 46 : remote_slot->restart_lsn = isnull ? InvalidXLogRecPtr : DatumGetLSN(d);
978 : :
979 : 46 : d = slot_getattr(tupslot, ++col, &isnull);
980 [ + - ]: 46 : remote_slot->catalog_xmin = isnull ? InvalidTransactionId :
981 : 46 : DatumGetTransactionId(d);
982 : :
983 : 46 : remote_slot->two_phase = DatumGetBool(slot_getattr(tupslot, ++col,
984 : : &isnull));
985 [ - + ]: 46 : Assert(!isnull);
986 : :
453 987 : 46 : d = slot_getattr(tupslot, ++col, &isnull);
988 [ + + ]: 46 : remote_slot->two_phase_at = isnull ? InvalidXLogRecPtr : DatumGetLSN(d);
989 : :
867 990 : 46 : remote_slot->failover = DatumGetBool(slot_getattr(tupslot, ++col,
991 : : &isnull));
992 [ - + ]: 46 : Assert(!isnull);
993 : :
994 : 46 : remote_slot->database = TextDatumGetCString(slot_getattr(tupslot,
995 : : ++col, &isnull));
996 [ - + ]: 46 : Assert(!isnull);
997 : :
998 : 46 : d = slot_getattr(tupslot, ++col, &isnull);
999 [ - + ]: 46 : remote_slot->invalidated = isnull ? RS_INVAL_NONE :
867 akapila@postgresql.o 1000 :UBC 0 : GetSlotInvalidationCause(TextDatumGetCString(d));
1001 : :
1002 : : /* Sanity check */
867 akapila@postgresql.o 1003 [ - + ]:CBC 46 : Assert(col == SLOTSYNC_COLUMN_COUNT);
1004 : :
1005 : : /*
1006 : : * If restart_lsn, confirmed_lsn or catalog_xmin is invalid but the
1007 : : * slot is valid, that means we have fetched the remote_slot in its
1008 : : * RS_EPHEMERAL state. In such a case, don't sync it; we can always
1009 : : * sync it in the next sync cycle when the remote_slot is persisted
1010 : : * and has valid lsn(s) and xmin values.
1011 : : *
1012 : : * XXX: In future, if we plan to expose 'slot->data.persistency' in
1013 : : * pg_replication_slots view, then we can avoid fetching RS_EPHEMERAL
1014 : : * slots in the first place.
1015 : : */
236 alvherre@kurilemu.de 1016 [ + - ]:GNC 46 : if ((!XLogRecPtrIsValid(remote_slot->restart_lsn) ||
1017 [ + - ]: 46 : !XLogRecPtrIsValid(remote_slot->confirmed_lsn) ||
867 akapila@postgresql.o 1018 [ - + ]:CBC 46 : !TransactionIdIsValid(remote_slot->catalog_xmin)) &&
867 akapila@postgresql.o 1019 [ # # ]:UBC 0 : remote_slot->invalidated == RS_INVAL_NONE)
1020 : 0 : pfree(remote_slot);
1021 : : else
1022 : : /* Create list of remote slots */
867 akapila@postgresql.o 1023 :CBC 46 : remote_slot_list = lappend(remote_slot_list, remote_slot);
1024 : :
1025 : 46 : ExecClearTuple(tupslot);
1026 : : }
1027 : :
35 akapila@postgresql.o 1028 :GNC 29 : ExecDropSingleTupleTableSlot(tupslot);
197 1029 : 29 : walrcv_clear_result(res);
1030 : :
1031 : 29 : return remote_slot_list;
1032 : : }
1033 : :
1034 : : /*
1035 : : * Synchronize slots.
1036 : : *
1037 : : * This function takes a list of remote slots and synchronizes them locally. It
1038 : : * creates the slots if not present on the standby and updates existing ones.
1039 : : *
1040 : : * If slot_persistence_pending is not NULL, it will be set to true if one or
1041 : : * more slots could not be persisted. This allows callers such as
1042 : : * SyncReplicationSlots() to retry those slots.
1043 : : *
1044 : : * Returns TRUE if any of the slots gets updated in this sync-cycle.
1045 : : */
1046 : : static bool
1047 : 29 : synchronize_slots(WalReceiverConn *wrconn, List *remote_slot_list,
1048 : : bool *slot_persistence_pending)
1049 : : {
1050 : 29 : bool some_slot_updated = false;
1051 : :
1052 : : /* Drop local slots that no longer need to be synced. */
867 akapila@postgresql.o 1053 :CBC 29 : drop_local_obsolete_slots(remote_slot_list);
1054 : :
1055 : : /* Now sync the slots locally */
1056 [ + - + + : 104 : foreach_ptr(RemoteSlot, remote_slot, remote_slot_list)
+ + ]
1057 : : {
1058 : 46 : Oid remote_dbid = get_database_oid(remote_slot->database, false);
1059 : :
1060 : : /*
1061 : : * Use shared lock to prevent a conflict with
1062 : : * ReplicationSlotsDropDBSlots(), trying to drop the same slot during
1063 : : * a drop-database operation.
1064 : : */
1065 : 46 : LockSharedObject(DatabaseRelationId, remote_dbid, 0, AccessShareLock);
1066 : :
197 akapila@postgresql.o 1067 :GNC 46 : some_slot_updated |= synchronize_one_slot(remote_slot, remote_dbid,
1068 : : slot_persistence_pending);
1069 : :
867 akapila@postgresql.o 1070 :CBC 46 : UnlockSharedObject(DatabaseRelationId, remote_dbid, 0, AccessShareLock);
1071 : : }
1072 : :
859 1073 : 29 : return some_slot_updated;
1074 : : }
1075 : :
1076 : : /*
1077 : : * Checks the remote server info.
1078 : : *
1079 : : * We ensure that the 'primary_slot_name' exists on the remote server and the
1080 : : * remote server is not a standby node.
1081 : : */
1082 : : static void
867 1083 : 15 : validate_remote_info(WalReceiverConn *wrconn)
1084 : : {
1085 : : #define PRIMARY_INFO_OUTPUT_COL_COUNT 2
1086 : : WalRcvExecResult *res;
1087 : 15 : Oid slotRow[PRIMARY_INFO_OUTPUT_COL_COUNT] = {BOOLOID, BOOLOID};
1088 : : StringInfoData cmd;
1089 : : bool isnull;
1090 : : TupleTableSlot *tupslot;
1091 : : bool remote_in_recovery;
1092 : : bool primary_slot_valid;
859 1093 : 15 : bool started_tx = false;
1094 : :
867 1095 : 15 : initStringInfo(&cmd);
1096 : 15 : appendStringInfo(&cmd,
1097 : : "SELECT pg_is_in_recovery(), count(*) = 1"
1098 : : " FROM pg_catalog.pg_replication_slots"
1099 : : " WHERE slot_type='physical' AND slot_name=%s",
1100 : : quote_literal_cstr(PrimarySlotName));
1101 : :
1102 : : /* The syscache access in walrcv_exec() needs a transaction env. */
859 1103 [ + + ]: 15 : if (!IsTransactionState())
1104 : : {
1105 : 6 : StartTransactionCommand();
1106 : 6 : started_tx = true;
1107 : : }
1108 : :
867 1109 : 15 : res = walrcv_exec(wrconn, cmd.data, PRIMARY_INFO_OUTPUT_COL_COUNT, slotRow);
1110 : 15 : pfree(cmd.data);
1111 : :
1112 [ - + ]: 15 : if (res->status != WALRCV_OK_TUPLES)
867 akapila@postgresql.o 1113 [ # # ]:UBC 0 : ereport(ERROR,
1114 : : errmsg("could not fetch primary slot name \"%s\" info from the primary server: %s",
1115 : : PrimarySlotName, res->err),
1116 : : errhint("Check if \"primary_slot_name\" is configured correctly."));
1117 : :
867 akapila@postgresql.o 1118 :CBC 15 : tupslot = MakeSingleTupleTableSlot(res->tupledesc, &TTSOpsMinimalTuple);
1119 [ - + ]: 15 : if (!tuplestore_gettupleslot(res->tuplestore, true, false, tupslot))
867 akapila@postgresql.o 1120 [ # # ]:UBC 0 : elog(ERROR,
1121 : : "failed to fetch tuple for the primary server slot specified by \"primary_slot_name\"");
1122 : :
867 akapila@postgresql.o 1123 :CBC 15 : remote_in_recovery = DatumGetBool(slot_getattr(tupslot, 1, &isnull));
1124 [ - + ]: 15 : Assert(!isnull);
1125 : :
1126 : : /*
1127 : : * Slot sync is currently not supported on a cascading standby. This is
1128 : : * because if we allow it, the primary server needs to wait for all the
1129 : : * cascading standbys, otherwise, logical subscribers can still be ahead
1130 : : * of one of the cascading standbys which we plan to promote. Thus, to
1131 : : * avoid this additional complexity, we restrict it for the time being.
1132 : : */
1133 [ + + ]: 15 : if (remote_in_recovery)
1134 [ + - ]: 1 : ereport(ERROR,
1135 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1136 : : errmsg("cannot synchronize replication slots from a standby server"));
1137 : :
1138 : 14 : primary_slot_valid = DatumGetBool(slot_getattr(tupslot, 2, &isnull));
1139 [ - + ]: 14 : Assert(!isnull);
1140 : :
1141 [ - + ]: 14 : if (!primary_slot_valid)
867 akapila@postgresql.o 1142 [ # # ]:UBC 0 : ereport(ERROR,
1143 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1144 : : /* translator: second %s is a GUC variable name */
1145 : : errmsg("replication slot \"%s\" specified by \"%s\" does not exist on primary server",
1146 : : PrimarySlotName, "primary_slot_name"));
1147 : :
35 akapila@postgresql.o 1148 :GNC 14 : ExecDropSingleTupleTableSlot(tupslot);
867 akapila@postgresql.o 1149 :CBC 14 : walrcv_clear_result(res);
1150 : :
859 1151 [ + + ]: 14 : if (started_tx)
1152 : 6 : CommitTransactionCommand();
867 1153 : 14 : }
1154 : :
1155 : : /*
1156 : : * Checks if dbname is specified in 'primary_conninfo'.
1157 : : *
1158 : : * Error out if not specified otherwise return it.
1159 : : */
1160 : : char *
859 1161 : 16 : CheckAndGetDbnameFromConninfo(void)
1162 : : {
1163 : : char *dbname;
1164 : :
1165 : : /*
1166 : : * The slot synchronization needs a database connection for walrcv_exec to
1167 : : * work.
1168 : : */
1169 : 16 : dbname = walrcv_get_dbname_from_conninfo(PrimaryConnInfo);
1170 [ + + ]: 16 : if (dbname == NULL)
1171 [ + - ]: 1 : ereport(ERROR,
1172 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1173 : :
1174 : : /*
1175 : : * translator: first %s is a connection option; second %s is a GUC
1176 : : * variable name
1177 : : */
1178 : : errmsg("replication slot synchronization requires \"%s\" to be specified in \"%s\"",
1179 : : "dbname", "primary_conninfo"));
1180 : 15 : return dbname;
1181 : : }
1182 : :
1183 : : /*
1184 : : * Return true if all necessary GUCs for slot synchronization are set
1185 : : * appropriately, otherwise, return false.
1186 : : */
1187 : : bool
1188 : 23 : ValidateSlotSyncParams(int elevel)
1189 : : {
1190 : : /*
1191 : : * Logical slot sync/creation requires logical decoding to be enabled.
1192 : : */
189 msawada@postgresql.o 1193 [ - + ]:GNC 23 : if (!IsLogicalDecodingEnabled())
1194 : : {
330 fujii@postgresql.org 1195 [ # # ]:UBC 0 : ereport(elevel,
1196 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1197 : : errmsg("replication slot synchronization requires \"effective_wal_level\" >= \"logical\" on the primary"),
1198 : : errhint("To enable logical decoding on primary, set \"wal_level\" >= \"logical\" or create at least one logical slot when \"wal_level\" = \"replica\"."));
1199 : :
1200 : 0 : return false;
1201 : : }
1202 : :
1203 : : /*
1204 : : * A physical replication slot(primary_slot_name) is required on the
1205 : : * primary to ensure that the rows needed by the standby are not removed
1206 : : * after restarting, so that the synchronized slot on the standby will not
1207 : : * be invalidated.
1208 : : */
867 akapila@postgresql.o 1209 [ + - - + ]:CBC 23 : if (PrimarySlotName == NULL || *PrimarySlotName == '\0')
1210 : : {
859 akapila@postgresql.o 1211 [ # # ]:UBC 0 : ereport(elevel,
1212 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1213 : : /* translator: %s is a GUC variable name */
1214 : : errmsg("replication slot synchronization requires \"%s\" to be set", "primary_slot_name"));
1215 : 0 : return false;
1216 : : }
1217 : :
1218 : : /*
1219 : : * hot_standby_feedback must be enabled to cooperate with the physical
1220 : : * replication slot, which allows informing the primary about the xmin and
1221 : : * catalog_xmin values on the standby.
1222 : : */
867 akapila@postgresql.o 1223 [ + + ]:CBC 23 : if (!hot_standby_feedback)
1224 : : {
859 1225 [ + - ]: 1 : ereport(elevel,
1226 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1227 : : /* translator: %s is a GUC variable name */
1228 : : errmsg("replication slot synchronization requires \"%s\" to be enabled",
1229 : : "hot_standby_feedback"));
1230 : 1 : return false;
1231 : : }
1232 : :
1233 : : /*
1234 : : * The primary_conninfo is required to make connection to primary for
1235 : : * getting slots information.
1236 : : */
867 1237 [ + - - + ]: 22 : if (PrimaryConnInfo == NULL || *PrimaryConnInfo == '\0')
1238 : : {
859 akapila@postgresql.o 1239 [ # # ]:UBC 0 : ereport(elevel,
1240 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1241 : : /* translator: %s is a GUC variable name */
1242 : : errmsg("replication slot synchronization requires \"%s\" to be set",
1243 : : "primary_conninfo"));
1244 : 0 : return false;
1245 : : }
1246 : :
859 akapila@postgresql.o 1247 :CBC 22 : return true;
1248 : : }
1249 : :
1250 : : /*
1251 : : * Re-read the config file for slot synchronization.
1252 : : *
1253 : : * Exit or throw error if relevant GUCs have changed depending on whether
1254 : : * called from slot sync worker or from the SQL function pg_sync_replication_slots()
1255 : : */
1256 : : static void
1257 : 1 : slotsync_reread_config(void)
1258 : : {
1259 : 1 : char *old_primary_conninfo = pstrdup(PrimaryConnInfo);
1260 : 1 : char *old_primary_slotname = pstrdup(PrimarySlotName);
1261 : 1 : bool old_sync_replication_slots = sync_replication_slots;
1262 : 1 : bool old_hot_standby_feedback = hot_standby_feedback;
1263 : : bool conninfo_changed;
1264 : : bool primary_slotname_changed;
201 1265 : 1 : bool is_slotsync_worker = AmLogicalSlotSyncWorkerProcess();
1266 : 1 : bool parameter_changed = false;
1267 : :
1268 [ + - ]: 1 : if (is_slotsync_worker)
1269 [ - + ]: 1 : Assert(sync_replication_slots);
1270 : :
859 1271 : 1 : ConfigReloadPending = false;
1272 : 1 : ProcessConfigFile(PGC_SIGHUP);
1273 : :
1274 : 1 : conninfo_changed = strcmp(old_primary_conninfo, PrimaryConnInfo) != 0;
1275 : 1 : primary_slotname_changed = strcmp(old_primary_slotname, PrimarySlotName) != 0;
1276 : 1 : pfree(old_primary_conninfo);
1277 : 1 : pfree(old_primary_slotname);
1278 : :
1279 [ - + ]: 1 : if (old_sync_replication_slots != sync_replication_slots)
1280 : : {
201 akapila@postgresql.o 1281 [ # # ]:UBC 0 : if (is_slotsync_worker)
1282 : : {
1283 [ # # ]: 0 : ereport(LOG,
1284 : : /* translator: %s is a GUC variable name */
1285 : : errmsg("replication slot synchronization worker will stop because \"%s\" is disabled",
1286 : : "sync_replication_slots"));
1287 : :
1288 : 0 : proc_exit(0);
1289 : : }
1290 : :
1291 : 0 : parameter_changed = true;
1292 : : }
1293 : : else
1294 : : {
201 akapila@postgresql.o 1295 [ + - + - ]:CBC 1 : if (conninfo_changed ||
1296 : 1 : primary_slotname_changed ||
1297 [ + - ]: 1 : (old_hot_standby_feedback != hot_standby_feedback))
1298 : : {
1299 : :
1300 [ + - ]: 1 : if (is_slotsync_worker)
1301 : : {
1302 [ + - ]: 1 : ereport(LOG,
1303 : : errmsg("replication slot synchronization worker will restart because of a parameter change"));
1304 : :
1305 : : /*
1306 : : * Reset the last-start time for this worker so that the
1307 : : * postmaster can restart it without waiting for
1308 : : * SLOTSYNC_RESTART_INTERVAL_SEC.
1309 : : */
1310 : 1 : SlotSyncCtx->last_start_time = 0;
1311 : :
1312 : 1 : proc_exit(0);
1313 : : }
1314 : :
201 akapila@postgresql.o 1315 :UBC 0 : parameter_changed = true;
1316 : : }
1317 : : }
1318 : :
1319 : : /*
1320 : : * If we have reached here with a parameter change, we must be running in
1321 : : * SQL function, emit error in such a case.
1322 : : */
1323 [ # # ]: 0 : if (parameter_changed)
1324 : : {
1325 [ # # ]: 0 : Assert(!is_slotsync_worker);
1326 [ # # ]: 0 : ereport(ERROR,
1327 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1328 : : errmsg("replication slot synchronization will stop because of a parameter change"));
1329 : : }
1330 : :
859 1331 : 0 : }
1332 : :
1333 : : /*
1334 : : * Handle receipt of an interrupt indicating a slotsync shutdown message.
1335 : : *
1336 : : * This is called within the SIGUSR1 handler. All we do here is set a flag
1337 : : * that will cause the next CHECK_FOR_INTERRUPTS() to invoke
1338 : : * ProcessSlotSyncMessage().
1339 : : */
1340 : : void
83 fujii@postgresql.org 1341 :CBC 1 : HandleSlotSyncMessageInterrupt(void)
1342 : : {
1343 : 1 : InterruptPending = true;
1344 : 1 : SlotSyncShutdownPending = true;
1345 : : /* latch will be set by procsignal_sigusr1_handler */
1346 : 1 : }
1347 : :
1348 : : /*
1349 : : * Handle a PROCSIG_SLOTSYNC_MESSAGE signal, called from ProcessInterrupts().
1350 : : *
1351 : : * If the current process is the slotsync background worker, log a message
1352 : : * and exit cleanly. If it is a backend executing pg_sync_replication_slots(),
1353 : : * raise an error, unless the sync has already finished, in which case there
1354 : : * is no need to interrupt the caller.
1355 : : */
1356 : : void
1357 : 1 : ProcessSlotSyncMessage(void)
1358 : : {
1359 : 1 : SlotSyncShutdownPending = false;
1360 : :
1361 [ + - ]: 1 : if (AmLogicalSlotSyncWorkerProcess())
1362 : : {
1363 [ + - ]: 1 : ereport(LOG,
1364 : : errmsg("replication slot synchronization worker will stop because promotion is triggered"));
1365 : 1 : proc_exit(0);
1366 : : }
1367 : : else
1368 : : {
1369 : : /*
1370 : : * If sync has already completed, there is no need to interrupt the
1371 : : * caller with an error.
1372 : : */
83 fujii@postgresql.org 1373 [ # # ]:UBC 0 : if (!IsSyncingReplicationSlots())
1374 : 0 : return;
1375 : :
1376 [ # # ]: 0 : ereport(ERROR,
1377 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1378 : : errmsg("replication slot synchronization will stop because promotion is triggered"));
1379 : : }
1380 : : }
1381 : :
1382 : : /*
1383 : : * Connection cleanup function for slotsync worker.
1384 : : *
1385 : : * Called on slotsync worker exit.
1386 : : */
1387 : : static void
796 akapila@postgresql.o 1388 :CBC 6 : slotsync_worker_disconnect(int code, Datum arg)
1389 : : {
1390 : 6 : WalReceiverConn *wrconn = (WalReceiverConn *) DatumGetPointer(arg);
1391 : :
1392 : 6 : walrcv_disconnect(wrconn);
1393 : 6 : }
1394 : :
1395 : : /*
1396 : : * Cleanup function for slotsync worker.
1397 : : *
1398 : : * Called on slotsync worker exit.
1399 : : */
1400 : : static void
859 1401 : 6 : slotsync_worker_onexit(int code, Datum arg)
1402 : : {
1403 : : /*
1404 : : * We need to do slots cleanup here just like WalSndErrorCleanup() does.
1405 : : *
1406 : : * The startup process during promotion invokes ShutDownSlotSync() which
1407 : : * waits for slot sync to finish and it does that by checking the
1408 : : * 'syncing' flag. Thus the slot sync worker must be done with slots'
1409 : : * release and cleanup to avoid any dangling temporary slots or active
1410 : : * slots before it marks itself as finished syncing.
1411 : : */
1412 : :
1413 : : /* Make sure active replication slots are released */
796 1414 [ - + ]: 6 : if (MyReplicationSlot != NULL)
796 akapila@postgresql.o 1415 :UBC 0 : ReplicationSlotRelease();
1416 : :
1417 : : /* Also cleanup the temporary slots. */
796 akapila@postgresql.o 1418 :CBC 6 : ReplicationSlotCleanup(false);
1419 : :
859 1420 [ - + ]: 6 : SpinLockAcquire(&SlotSyncCtx->mutex);
1421 : :
1422 : 6 : SlotSyncCtx->pid = InvalidPid;
1423 : :
1424 : : /*
1425 : : * If syncing_slots is true, it indicates that the process errored out
1426 : : * without resetting the flag. So, we need to clean up shared memory and
1427 : : * reset the flag here.
1428 : : */
796 1429 [ + - ]: 6 : if (syncing_slots)
1430 : : {
1431 : 6 : SlotSyncCtx->syncing = false;
1432 : 6 : syncing_slots = false;
1433 : : }
1434 : :
859 1435 : 6 : SpinLockRelease(&SlotSyncCtx->mutex);
1436 : 6 : }
1437 : :
1438 : : /*
1439 : : * Sleep for long enough that we believe it's likely that the slots on primary
1440 : : * get updated.
1441 : : *
1442 : : * If there is no slot activity the wait time between sync-cycles will double
1443 : : * (to a maximum of 30s). If there is some slot activity the wait time between
1444 : : * sync-cycles is reset to the minimum (200ms).
1445 : : */
1446 : : static void
1447 : 21 : wait_for_slot_activity(bool some_slot_updated)
1448 : : {
1449 : : int rc;
1450 : :
1451 [ + + ]: 21 : if (!some_slot_updated)
1452 : : {
1453 : : /*
1454 : : * No slots were updated, so double the sleep time, but not beyond the
1455 : : * maximum allowable value.
1456 : : */
852 1457 : 11 : sleep_ms = Min(sleep_ms * 2, MAX_SLOTSYNC_WORKER_NAPTIME_MS);
1458 : : }
1459 : : else
1460 : : {
1461 : : /*
1462 : : * Some slots were updated since the last sleep, so reset the sleep
1463 : : * time.
1464 : : */
1465 : 10 : sleep_ms = MIN_SLOTSYNC_WORKER_NAPTIME_MS;
1466 : : }
1467 : :
859 1468 : 21 : rc = WaitLatch(MyLatch,
1469 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
1470 : : sleep_ms,
1471 : : WAIT_EVENT_REPLICATION_SLOTSYNC_MAIN);
1472 : :
1473 [ + + ]: 21 : if (rc & WL_LATCH_SET)
1474 : 3 : ResetLatch(MyLatch);
1475 : 21 : }
1476 : :
1477 : : /*
1478 : : * Emit an error if a concurrent sync call is in progress.
1479 : : * Otherwise, advertise that a sync is in progress.
1480 : : */
1481 : : static void
201 1482 : 15 : check_and_set_sync_info(pid_t sync_process_pid)
1483 : : {
796 1484 [ - + ]: 15 : SpinLockAcquire(&SlotSyncCtx->mutex);
1485 : :
1486 : : /*
1487 : : * Exit immediately if promotion has been triggered. This guards against
1488 : : * a new worker (or a call to pg_sync_replication_slots()) that starts
1489 : : * after the old worker was stopped by ShutDownSlotSync().
1490 : : */
83 fujii@postgresql.org 1491 [ - + ]: 15 : if (SlotSyncCtx->stopSignaled)
1492 : : {
83 fujii@postgresql.org 1493 :UBC 0 : SpinLockRelease(&SlotSyncCtx->mutex);
1494 : :
1495 [ # # ]: 0 : if (AmLogicalSlotSyncWorkerProcess())
1496 : : {
1497 [ # # ]: 0 : ereport(DEBUG1,
1498 : : errmsg("replication slot synchronization worker will not start because promotion was triggered"));
1499 : :
1500 : 0 : proc_exit(0);
1501 : : }
1502 : : else
1503 : : {
1504 : : /*
1505 : : * For the backend executing SQL function
1506 : : * pg_sync_replication_slots().
1507 : : */
1508 [ # # ]: 0 : ereport(ERROR,
1509 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1510 : : errmsg("replication slot synchronization will not start because promotion was triggered"));
1511 : : }
1512 : : }
1513 : :
796 akapila@postgresql.o 1514 [ - + ]:CBC 15 : if (SlotSyncCtx->syncing)
1515 : : {
796 akapila@postgresql.o 1516 :UBC 0 : SpinLockRelease(&SlotSyncCtx->mutex);
1517 [ # # ]: 0 : ereport(ERROR,
1518 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1519 : : errmsg("cannot synchronize replication slots concurrently"));
1520 : : }
1521 : :
1522 : : /* The pid must not be already assigned in SlotSyncCtx */
201 akapila@postgresql.o 1523 [ - + ]:CBC 15 : Assert(SlotSyncCtx->pid == InvalidPid);
1524 : :
796 1525 : 15 : SlotSyncCtx->syncing = true;
1526 : :
1527 : : /*
1528 : : * Advertise the required PID so that the startup process can kill the
1529 : : * slot sync process on promotion.
1530 : : */
201 1531 : 15 : SlotSyncCtx->pid = sync_process_pid;
1532 : :
796 1533 : 15 : SpinLockRelease(&SlotSyncCtx->mutex);
1534 : :
1535 : 15 : syncing_slots = true;
1536 : 15 : }
1537 : :
1538 : : /*
1539 : : * Reset syncing flag.
1540 : : */
1541 : : static void
209 nathan@postgresql.or 1542 :GNC 9 : reset_syncing_flag(void)
1543 : : {
796 akapila@postgresql.o 1544 [ - + ]:CBC 9 : SpinLockAcquire(&SlotSyncCtx->mutex);
1545 : 9 : SlotSyncCtx->syncing = false;
201 1546 : 9 : SlotSyncCtx->pid = InvalidPid;
796 1547 : 9 : SpinLockRelease(&SlotSyncCtx->mutex);
1548 : :
1549 : 9 : syncing_slots = false;
292 peter@eisentraut.org 1550 : 9 : }
1551 : :
1552 : : /*
1553 : : * The main loop of our worker process.
1554 : : *
1555 : : * It connects to the primary server, fetches logical failover slots
1556 : : * information periodically in order to create and sync the slots.
1557 : : *
1558 : : * Note: If any changes are made here, check if the corresponding SQL
1559 : : * function logic in SyncReplicationSlots() also needs to be changed.
1560 : : */
1561 : : void
494 1562 : 6 : ReplSlotSyncWorkerMain(const void *startup_data, size_t startup_data_len)
1563 : : {
859 akapila@postgresql.o 1564 : 6 : WalReceiverConn *wrconn = NULL;
1565 : : char *dbname;
1566 : : char *err;
1567 : : sigjmp_buf local_sigjmp_buf;
1568 : : StringInfoData app_name;
1569 : :
834 heikki.linnakangas@i 1570 [ - + ]: 6 : Assert(startup_data_len == 0);
1571 : :
1572 : : /* Release postmaster's working memory context */
85 fujii@postgresql.org 1573 [ + - ]:GNC 6 : if (PostmasterContext)
1574 : : {
1575 : 6 : MemoryContextDelete(PostmasterContext);
1576 : 6 : PostmasterContext = NULL;
1577 : : }
1578 : :
859 akapila@postgresql.o 1579 :CBC 6 : init_ps_display(NULL);
1580 : :
728 heikki.linnakangas@i 1581 [ - + ]: 6 : Assert(GetProcessingMode() == InitProcessing);
1582 : :
1583 : : /*
1584 : : * Create a per-backend PGPROC struct in shared memory. We must do this
1585 : : * before we access any shared memory.
1586 : : */
859 akapila@postgresql.o 1587 : 6 : InitProcess();
1588 : :
1589 : : /*
1590 : : * Early initialization.
1591 : : */
1592 : 6 : BaseInit();
1593 : :
1594 [ - + ]: 6 : Assert(SlotSyncCtx != NULL);
1595 : :
1596 : : /*
1597 : : * If an exception is encountered, processing resumes here.
1598 : : *
1599 : : * We just need to clean up, report the error, and go away.
1600 : : *
1601 : : * If we do not have this handling here, then since this worker process
1602 : : * operates at the bottom of the exception stack, ERRORs turn into FATALs.
1603 : : * Therefore, we create our own exception handler to catch ERRORs.
1604 : : */
1605 [ + + ]: 6 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
1606 : : {
1607 : : /* since not using PG_TRY, must reset error stack by hand */
1608 : 2 : error_context_stack = NULL;
1609 : :
1610 : : /* Prevents interrupts while cleaning up */
1611 : 2 : HOLD_INTERRUPTS();
1612 : :
1613 : : /* Report the error to the server log */
1614 : 2 : EmitErrorReport();
1615 : :
1616 : : /*
1617 : : * We can now go away. Note that because we called InitProcess, a
1618 : : * callback was registered to do ProcKill, which will clean up
1619 : : * necessary state.
1620 : : */
1621 : 2 : proc_exit(0);
1622 : : }
1623 : :
1624 : : /* We can now handle ereport(ERROR) */
1625 : 6 : PG_exception_stack = &local_sigjmp_buf;
1626 : :
1627 : : /* Setup signal handling */
796 1628 : 6 : pqsignal(SIGHUP, SignalHandlerForConfigReload);
203 1629 : 6 : pqsignal(SIGINT, StatementCancelHandler);
796 1630 : 6 : pqsignal(SIGTERM, die);
1631 : 6 : pqsignal(SIGFPE, FloatExceptionHandler);
1632 : 6 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
77 andrew@dunslane.net 1633 :GNC 6 : pqsignal(SIGUSR2, PG_SIG_IGN);
1634 : 6 : pqsignal(SIGPIPE, PG_SIG_IGN);
1635 : 6 : pqsignal(SIGCHLD, PG_SIG_DFL);
1636 : :
796 akapila@postgresql.o 1637 :CBC 6 : check_and_set_sync_info(MyProcPid);
1638 : :
1639 [ + - ]: 6 : ereport(LOG, errmsg("slot sync worker started"));
1640 : :
1641 : : /* Register it as soon as SlotSyncCtx->pid is initialized. */
1642 : 6 : before_shmem_exit(slotsync_worker_onexit, (Datum) 0);
1643 : :
1644 : : /*
1645 : : * Establishes SIGALRM handler and initialize timeout module. It is needed
1646 : : * by InitPostgres to register different timeouts.
1647 : : */
1648 : 6 : InitializeTimeouts();
1649 : :
1650 : : /* Load the libpq-specific functions */
1651 : 6 : load_file("libpqwalreceiver", false);
1652 : :
1653 : : /*
1654 : : * Unblock signals (they were blocked when the postmaster forked us)
1655 : : */
859 1656 : 6 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
1657 : :
1658 : : /*
1659 : : * Set always-secure search path, so malicious users can't redirect user
1660 : : * code (e.g. operators).
1661 : : *
1662 : : * It's not strictly necessary since we won't be scanning or writing to
1663 : : * any user table locally, but it's good to retain it here for added
1664 : : * precaution.
1665 : : */
852 1666 : 6 : SetConfigOption("search_path", "", PGC_SUSET, PGC_S_OVERRIDE);
1667 : :
859 1668 : 6 : dbname = CheckAndGetDbnameFromConninfo();
1669 : :
1670 : : /*
1671 : : * Connect to the database specified by the user in primary_conninfo. We
1672 : : * need a database connection for walrcv_exec to work which we use to
1673 : : * fetch slot information from the remote node. See comments atop
1674 : : * libpqrcv_exec.
1675 : : *
1676 : : * We do not specify a specific user here since the slot sync worker will
1677 : : * operate as a superuser. This is safe because the slot sync worker does
1678 : : * not interact with user tables, eliminating the risk of executing
1679 : : * arbitrary code within triggers.
1680 : : */
1681 : 6 : InitPostgres(dbname, InvalidOid, NULL, InvalidOid, 0, NULL);
1682 : :
1683 : 6 : SetProcessingMode(NormalProcessing);
1684 : :
1685 : 6 : initStringInfo(&app_name);
1686 [ + - ]: 6 : if (cluster_name[0])
1687 : 6 : appendStringInfo(&app_name, "%s_%s", cluster_name, "slotsync worker");
1688 : : else
811 drowley@postgresql.o 1689 :UBC 0 : appendStringInfoString(&app_name, "slotsync worker");
1690 : :
1691 : : /*
1692 : : * Establish the connection to the primary server for slot
1693 : : * synchronization.
1694 : : */
859 akapila@postgresql.o 1695 :CBC 6 : wrconn = walrcv_connect(PrimaryConnInfo, false, false, false,
1696 : : app_name.data, &err);
1697 : :
1698 [ - + ]: 6 : if (!wrconn)
859 akapila@postgresql.o 1699 [ # # ]:UBC 0 : ereport(ERROR,
1700 : : errcode(ERRCODE_CONNECTION_FAILURE),
1701 : : errmsg("synchronization worker \"%s\" could not connect to the primary server: %s",
1702 : : app_name.data, err));
1703 : :
300 akapila@postgresql.o 1704 :CBC 6 : pfree(app_name.data);
1705 : :
1706 : : /*
1707 : : * Register the disconnection callback.
1708 : : *
1709 : : * XXX: This can be combined with previous cleanup registration of
1710 : : * slotsync_worker_onexit() but that will need the connection to be made
1711 : : * global and we want to avoid introducing global for this purpose.
1712 : : */
796 1713 : 6 : before_shmem_exit(slotsync_worker_disconnect, PointerGetDatum(wrconn));
1714 : :
1715 : : /*
1716 : : * Using the specified primary server connection, check that we are not a
1717 : : * cascading standby and slot configured in 'primary_slot_name' exists on
1718 : : * the primary server.
1719 : : */
859 1720 : 6 : validate_remote_info(wrconn);
1721 : :
1722 : : /* Main loop to synchronize slots */
1723 : : for (;;)
1724 : 19 : {
1725 : 25 : bool some_slot_updated = false;
197 akapila@postgresql.o 1726 :GNC 25 : bool started_tx = false;
1727 : : List *remote_slots;
1728 : :
83 fujii@postgresql.org 1729 [ + + ]:CBC 25 : CHECK_FOR_INTERRUPTS();
1730 : :
1731 [ + + ]: 22 : if (ConfigReloadPending)
1732 : 1 : slotsync_reread_config();
1733 : :
1734 : : /*
1735 : : * The syscache access in fetch_remote_slots() needs a transaction
1736 : : * env.
1737 : : */
197 akapila@postgresql.o 1738 [ + - ]:GNC 21 : if (!IsTransactionState())
1739 : : {
1740 : 21 : StartTransactionCommand();
1741 : 21 : started_tx = true;
1742 : : }
1743 : :
1744 : 21 : remote_slots = fetch_remote_slots(wrconn, NIL);
1745 : 19 : some_slot_updated = synchronize_slots(wrconn, remote_slots, NULL);
1746 : 19 : list_free_deep(remote_slots);
1747 : :
1748 [ + - ]: 19 : if (started_tx)
1749 : 19 : CommitTransactionCommand();
1750 : :
859 akapila@postgresql.o 1751 :CBC 19 : wait_for_slot_activity(some_slot_updated);
1752 : : }
1753 : :
1754 : : /*
1755 : : * The slot sync worker can't get here because it will only stop when it
1756 : : * receives a stop request from the startup process, or when there is an
1757 : : * error.
1758 : : */
1759 : : Assert(false);
1760 : : }
1761 : :
1762 : : /*
1763 : : * Update the inactive_since property for synced slots.
1764 : : *
1765 : : * Note that this function is currently called when we shutdown the slot
1766 : : * sync machinery.
1767 : : */
1768 : : static void
816 1769 : 984 : update_synced_slots_inactive_since(void)
1770 : : {
1771 : 984 : TimestampTz now = 0;
1772 : :
1773 : : /*
1774 : : * We need to update inactive_since only when we are promoting standby to
1775 : : * correctly interpret the inactive_since if the standby gets promoted
1776 : : * without a restart. We don't want the slots to appear inactive for a
1777 : : * long time after promotion if they haven't been synchronized recently.
1778 : : * Whoever acquires the slot, i.e., makes the slot active, will reset it.
1779 : : */
1780 [ + + ]: 984 : if (!StandbyMode)
1781 : 928 : return;
1782 : :
1783 : : /* The slot sync worker or the SQL function mustn't be running by now */
796 1784 [ + - - + ]: 56 : Assert((SlotSyncCtx->pid == InvalidPid) && !SlotSyncCtx->syncing);
1785 : :
816 1786 : 56 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
1787 : :
84 alvherre@kurilemu.de 1788 [ + + ]:GNC 879 : for (int i = 0; i < max_replication_slots + max_repack_replication_slots; i++)
1789 : : {
816 akapila@postgresql.o 1790 :CBC 823 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1791 : :
1792 : : /* Check if it is a synchronized slot */
1793 [ + + + + ]: 823 : if (s->in_use && s->data.synced)
1794 : : {
1795 [ - + ]: 3 : Assert(SlotIsLogical(s));
1796 : :
1797 : : /* The slot must not be acquired by any process */
140 heikki.linnakangas@i 1798 [ - + ]:GNC 3 : Assert(s->active_proc == INVALID_PROC_NUMBER);
1799 : :
1800 : : /* Use the same inactive_since time for all the slots. */
816 akapila@postgresql.o 1801 [ + + ]:CBC 3 : if (now == 0)
1802 : 2 : now = GetCurrentTimestamp();
1803 : :
510 1804 : 3 : ReplicationSlotSetInactiveSince(s, now, true);
1805 : : }
1806 : : }
1807 : :
816 1808 : 56 : LWLockRelease(ReplicationSlotControlLock);
1809 : : }
1810 : :
1811 : : /*
1812 : : * Shut down slot synchronization.
1813 : : *
1814 : : * This function sets stopSignaled=true and wakes up the slot sync process
1815 : : * (either worker or backend running the SQL function pg_sync_replication_slots())
1816 : : * so that worker can exit or the SQL function pg_sync_replication_slots() can
1817 : : * finish. It also waits till the slot sync worker has exited or
1818 : : * pg_sync_replication_slots() has finished.
1819 : : */
1820 : : void
859 1821 : 984 : ShutDownSlotSync(void)
1822 : : {
1823 : : pid_t sync_process_pid;
1824 : :
1825 [ - + ]: 984 : SpinLockAcquire(&SlotSyncCtx->mutex);
1826 : :
1827 : 984 : SlotSyncCtx->stopSignaled = true;
1828 : :
1829 : : /*
1830 : : * Return if neither the slot sync worker is running nor the function
1831 : : * pg_sync_replication_slots() is executing.
1832 : : */
796 1833 [ + + ]: 984 : if (!SlotSyncCtx->syncing)
1834 : : {
859 1835 : 983 : SpinLockRelease(&SlotSyncCtx->mutex);
816 1836 : 983 : update_synced_slots_inactive_since();
859 1837 : 983 : return;
1838 : : }
1839 : :
201 1840 : 1 : sync_process_pid = SlotSyncCtx->pid;
1841 : :
859 1842 : 1 : SpinLockRelease(&SlotSyncCtx->mutex);
1843 : :
1844 : : /*
1845 : : * Signal process doing slotsync, if any, asking it to stop.
1846 : : */
201 1847 [ + - ]: 1 : if (sync_process_pid != InvalidPid)
83 fujii@postgresql.org 1848 : 1 : SendProcSignal(sync_process_pid, PROCSIG_SLOTSYNC_MESSAGE,
1849 : : INVALID_PROC_NUMBER);
1850 : :
1851 : : /* Wait for slot sync to end */
1852 : : for (;;)
859 akapila@postgresql.o 1853 :GBC 1 : {
1854 : : int rc;
1855 : :
1856 : : /* Wait a bit, we don't expect to have to wait long */
859 akapila@postgresql.o 1857 :CBC 2 : rc = WaitLatch(MyLatch,
1858 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
1859 : : 10L, WAIT_EVENT_REPLICATION_SLOTSYNC_SHUTDOWN);
1860 : :
1861 [ + + ]: 2 : if (rc & WL_LATCH_SET)
1862 : : {
859 akapila@postgresql.o 1863 :GBC 1 : ResetLatch(MyLatch);
1864 [ - + ]: 1 : CHECK_FOR_INTERRUPTS();
1865 : : }
1866 : :
859 akapila@postgresql.o 1867 [ - + ]:CBC 2 : SpinLockAcquire(&SlotSyncCtx->mutex);
1868 : :
1869 : : /* Ensure that no process is syncing the slots. */
796 1870 [ + + ]: 2 : if (!SlotSyncCtx->syncing)
859 1871 : 1 : break;
1872 : :
859 akapila@postgresql.o 1873 :GBC 1 : SpinLockRelease(&SlotSyncCtx->mutex);
1874 : : }
1875 : :
859 akapila@postgresql.o 1876 :CBC 1 : SpinLockRelease(&SlotSyncCtx->mutex);
1877 : :
816 1878 : 1 : update_synced_slots_inactive_since();
1879 : : }
1880 : :
1881 : : /*
1882 : : * SlotSyncWorkerCanRestart
1883 : : *
1884 : : * Return true, indicating worker is allowed to restart, if enough time has
1885 : : * passed since it was last launched to reach SLOTSYNC_RESTART_INTERVAL_SEC.
1886 : : * Otherwise return false.
1887 : : *
1888 : : * This is a safety valve to protect against continuous respawn attempts if the
1889 : : * worker is dying immediately at launch. Note that since we will retry to
1890 : : * launch the worker from the postmaster main loop, we will get another
1891 : : * chance later.
1892 : : */
1893 : : bool
859 1894 : 12 : SlotSyncWorkerCanRestart(void)
1895 : : {
1896 : 12 : time_t curtime = time(NULL);
1897 : :
1898 : : /*
1899 : : * If first time through, or time somehow went backwards, always update
1900 : : * last_start_time to match the current clock and allow worker start.
1901 : : * Otherwise allow it only once enough time has elapsed.
1902 : : */
251 tgl@sss.pgh.pa.us 1903 [ + + ]:GNC 12 : if (SlotSyncCtx->last_start_time == 0 ||
1904 [ + - ]: 6 : curtime < SlotSyncCtx->last_start_time ||
1905 [ - + ]: 6 : curtime - SlotSyncCtx->last_start_time >= SLOTSYNC_RESTART_INTERVAL_SEC)
1906 : : {
1907 : 6 : SlotSyncCtx->last_start_time = curtime;
1908 : 6 : return true;
1909 : : }
1910 : 6 : return false;
1911 : : }
1912 : :
1913 : : /*
1914 : : * Is current process syncing replication slots?
1915 : : *
1916 : : * Could be either backend executing SQL function or slot sync worker.
1917 : : */
1918 : : bool
867 akapila@postgresql.o 1919 :CBC 74 : IsSyncingReplicationSlots(void)
1920 : : {
1921 : 74 : return syncing_slots;
1922 : : }
1923 : :
1924 : : /*
1925 : : * Register shared memory space needed for slot synchronization.
1926 : : */
1927 : : static void
85 heikki.linnakangas@i 1928 :GNC 1212 : SlotSyncShmemRequest(void *arg)
1929 : : {
1930 : 1212 : ShmemRequestStruct(.name = "Slot Sync Data",
1931 : : .size = sizeof(SlotSyncCtxStruct),
1932 : : .ptr = (void **) &SlotSyncCtx,
1933 : : );
867 akapila@postgresql.o 1934 :GIC 1212 : }
1935 : :
1936 : : /*
1937 : : * Initialize shared memory for slot synchronization.
1938 : : */
1939 : : static void
85 heikki.linnakangas@i 1940 :GNC 1209 : SlotSyncShmemInit(void *arg)
1941 : : {
1942 : 1209 : memset(SlotSyncCtx, 0, sizeof(SlotSyncCtxStruct));
1943 : 1209 : SlotSyncCtx->pid = InvalidPid;
1944 : 1209 : SpinLockInit(&SlotSyncCtx->mutex);
867 akapila@postgresql.o 1945 :CBC 1209 : }
1946 : :
1947 : : /*
1948 : : * Error cleanup callback for slot sync SQL function.
1949 : : */
1950 : : static void
1951 : 1 : slotsync_failure_callback(int code, Datum arg)
1952 : : {
1953 : 1 : WalReceiverConn *wrconn = (WalReceiverConn *) DatumGetPointer(arg);
1954 : :
1955 : : /*
1956 : : * We need to do slots cleanup here just like WalSndErrorCleanup() does.
1957 : : *
1958 : : * The startup process during promotion invokes ShutDownSlotSync() which
1959 : : * waits for slot sync to finish and it does that by checking the
1960 : : * 'syncing' flag. Thus the SQL function must be done with slots' release
1961 : : * and cleanup to avoid any dangling temporary slots or active slots
1962 : : * before it marks itself as finished syncing.
1963 : : */
1964 : :
1965 : : /* Make sure active replication slots are released */
796 1966 [ - + ]: 1 : if (MyReplicationSlot != NULL)
796 akapila@postgresql.o 1967 :UBC 0 : ReplicationSlotRelease();
1968 : :
1969 : : /* Also cleanup the synced temporary slots. */
796 akapila@postgresql.o 1970 :CBC 1 : ReplicationSlotCleanup(true);
1971 : :
1972 : : /*
1973 : : * The set syncing_slots indicates that the process errored out without
1974 : : * resetting the flag. So, we need to clean up shared memory and reset the
1975 : : * flag here.
1976 : : */
1977 [ + - ]: 1 : if (syncing_slots)
1978 : 1 : reset_syncing_flag();
1979 : :
867 1980 : 1 : walrcv_disconnect(wrconn);
1981 : 1 : }
1982 : :
1983 : : /*
1984 : : * Helper function to extract slot names from a list of remote slots
1985 : : */
1986 : : static List *
197 akapila@postgresql.o 1987 :GNC 1 : extract_slot_names(List *remote_slots)
1988 : : {
1989 : 1 : List *slot_names = NIL;
1990 : :
1991 [ + - + + : 3 : foreach_ptr(RemoteSlot, remote_slot, remote_slots)
+ + ]
1992 : : {
1993 : : char *slot_name;
1994 : :
1995 : 1 : slot_name = pstrdup(remote_slot->name);
1996 : 1 : slot_names = lappend(slot_names, slot_name);
1997 : : }
1998 : :
1999 : 1 : return slot_names;
2000 : : }
2001 : :
2002 : : /*
2003 : : * Synchronize the failover enabled replication slots using the specified
2004 : : * primary server connection.
2005 : : *
2006 : : * Repeatedly fetches and updates replication slot information from the
2007 : : * primary until all slots are at least "sync ready".
2008 : : *
2009 : : * Exits early if promotion is triggered or certain critical
2010 : : * configuration parameters have changed.
2011 : : */
2012 : : void
867 akapila@postgresql.o 2013 :CBC 9 : SyncReplicationSlots(WalReceiverConn *wrconn)
2014 : : {
2015 [ + + ]: 9 : PG_ENSURE_ERROR_CLEANUP(slotsync_failure_callback, PointerGetDatum(wrconn));
2016 : : {
197 akapila@postgresql.o 2017 :GNC 9 : List *remote_slots = NIL;
2018 : 9 : List *slot_names = NIL; /* List of slot names to track */
2019 : : MemoryContext sync_retry_ctx;
2020 : :
201 akapila@postgresql.o 2021 :CBC 9 : check_and_set_sync_info(MyProcPid);
2022 : :
867 2023 : 9 : validate_remote_info(wrconn);
2024 : :
2025 : : /*
2026 : : * Setup and use a per-sync-cycle memory context, which is reset every
2027 : : * time we loop below. This avoids having to retail freeing the memory
2028 : : * used in each sync cycle.
2029 : : */
35 akapila@postgresql.o 2030 :GNC 8 : sync_retry_ctx = AllocSetContextCreate(CurrentMemoryContext,
2031 : : "slot sync retry context",
2032 : : ALLOCSET_DEFAULT_SIZES);
2033 : :
2034 : : /* Retry until all the slots are sync-ready */
2035 : : for (;;)
197 2036 : 2 : {
2037 : 10 : bool slot_persistence_pending = false;
2038 : 10 : bool some_slot_updated = false;
2039 : : MemoryContext oldctx;
2040 : :
2041 : : /* Check for interrupts and config changes */
83 fujii@postgresql.org 2042 [ - + ]: 10 : CHECK_FOR_INTERRUPTS();
2043 : :
2044 [ - + ]: 10 : if (ConfigReloadPending)
83 fujii@postgresql.org 2045 :UNC 0 : slotsync_reread_config();
2046 : :
2047 : : /* We must be in a valid transaction state */
197 akapila@postgresql.o 2048 [ - + ]:GNC 10 : Assert(IsTransactionState());
2049 : :
35 2050 : 10 : MemoryContextReset(sync_retry_ctx);
2051 : 10 : oldctx = MemoryContextSwitchTo(sync_retry_ctx);
2052 : :
2053 : : /*
2054 : : * Fetch remote slot info for the given slot_names. If slot_names
2055 : : * is NIL, fetch all failover-enabled slots. Note that we reuse
2056 : : * slot_names from the first iteration; re-fetching all failover
2057 : : * slots each time could cause an endless loop. Instead of
2058 : : * reprocessing only the pending slots in each iteration, it's
2059 : : * better to process all the slots received in the first
2060 : : * iteration. This ensures that by the time we're done, all slots
2061 : : * reflect the latest values.
2062 : : */
197 2063 : 10 : remote_slots = fetch_remote_slots(wrconn, slot_names);
2064 : :
2065 : : /* Attempt to synchronize slots */
2066 : 10 : some_slot_updated = synchronize_slots(wrconn, remote_slots,
2067 : : &slot_persistence_pending);
2068 : :
2069 : : /*
2070 : : * slot_names must survive later sync_retry_ctx resets, so copy it
2071 : : * in the outer context.
2072 : : */
35 2073 : 10 : MemoryContextSwitchTo(oldctx);
2074 : :
2075 : : /*
2076 : : * If slot_persistence_pending is true, extract slot names for
2077 : : * future iterations (only needed if we haven't done it yet)
2078 : : */
197 2079 [ + + + + ]: 10 : if (slot_names == NIL && slot_persistence_pending)
2080 : 1 : slot_names = extract_slot_names(remote_slots);
2081 : :
2082 : : /* Done if all slots are persisted i.e are sync-ready */
2083 [ + + ]: 10 : if (!slot_persistence_pending)
2084 : 8 : break;
2085 : :
2086 : : /* wait before retrying again */
2087 : 2 : wait_for_slot_activity(some_slot_updated);
2088 : : }
2089 : :
35 2090 : 8 : MemoryContextDelete(sync_retry_ctx);
2091 : :
197 2092 [ + + ]: 8 : if (slot_names)
2093 : 1 : list_free_deep(slot_names);
2094 : :
2095 : : /* Cleanup the synced temporary slots */
796 akapila@postgresql.o 2096 :CBC 8 : ReplicationSlotCleanup(true);
2097 : :
2098 : : /* We are done with sync, so reset sync flag */
2099 : 8 : reset_syncing_flag();
2100 : : }
867 2101 [ - + ]: 9 : PG_END_ENSURE_ERROR_CLEANUP(slotsync_failure_callback, PointerGetDatum(wrconn));
2102 : 8 : }
|
