Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : * tablesync.c
3 : : * PostgreSQL logical replication: initial table data synchronization
4 : : *
5 : : * Copyright (c) 2012-2026, PostgreSQL Global Development Group
6 : : *
7 : : * IDENTIFICATION
8 : : * src/backend/replication/logical/tablesync.c
9 : : *
10 : : * NOTES
11 : : * This file contains code for initial table data synchronization for
12 : : * logical replication.
13 : : *
14 : : * The initial data synchronization is done separately for each table,
15 : : * in a separate apply worker that only fetches the initial snapshot data
16 : : * from the publisher and then synchronizes the position in the stream with
17 : : * the leader apply worker.
18 : : *
19 : : * There are several reasons for doing the synchronization this way:
20 : : * - It allows us to parallelize the initial data synchronization
21 : : * which lowers the time needed for it to happen.
22 : : * - The initial synchronization does not have to hold the xid and LSN
23 : : * for the time it takes to copy data of all tables, causing less
24 : : * bloat and lower disk consumption compared to doing the
25 : : * synchronization in a single process for the whole database.
26 : : * - It allows us to synchronize any tables added after the initial
27 : : * synchronization has finished.
28 : : *
29 : : * The stream position synchronization works in multiple steps:
30 : : * - Apply worker requests a tablesync worker to start, setting the new
31 : : * table state to INIT.
32 : : * - Tablesync worker starts; changes table state from INIT to DATASYNC while
33 : : * copying.
34 : : * - Tablesync worker does initial table copy; there is a FINISHEDCOPY (sync
35 : : * worker specific) state to indicate when the copy phase has completed, so
36 : : * if the worker crashes with this (non-memory) state then the copy will not
37 : : * be re-attempted.
38 : : * - Tablesync worker then sets table state to SYNCWAIT; waits for state change.
39 : : * - Apply worker periodically checks for tables in SYNCWAIT state. When
40 : : * any appear, it sets the table state to CATCHUP and starts loop-waiting
41 : : * until either the table state is set to SYNCDONE or the sync worker
42 : : * exits.
43 : : * - After the sync worker has seen the state change to CATCHUP, it will
44 : : * read the stream and apply changes (acting like an apply worker) until
45 : : * it catches up to the specified stream position. Then it sets the
46 : : * state to SYNCDONE. There might be zero changes applied between
47 : : * CATCHUP and SYNCDONE, because the sync worker might be ahead of the
48 : : * apply worker.
49 : : * - Once the state is set to SYNCDONE, the apply will continue tracking
50 : : * the table until it reaches the SYNCDONE stream position, at which
51 : : * point it sets state to READY and stops tracking. Again, there might
52 : : * be zero changes in between.
53 : : *
54 : : * So the state progression is always: INIT -> DATASYNC -> FINISHEDCOPY
55 : : * -> SYNCWAIT -> CATCHUP -> SYNCDONE -> READY.
56 : : *
57 : : * The catalog pg_subscription_rel is used to keep information about
58 : : * subscribed tables and their state. The catalog holds all states
59 : : * except SYNCWAIT and CATCHUP which are only in shared memory.
60 : : *
61 : : * Example flows look like this:
62 : : * - Apply is in front:
63 : : * sync:8
64 : : * -> set in catalog FINISHEDCOPY
65 : : * -> set in memory SYNCWAIT
66 : : * apply:10
67 : : * -> set in memory CATCHUP
68 : : * -> enter wait-loop
69 : : * sync:10
70 : : * -> set in catalog SYNCDONE
71 : : * -> exit
72 : : * apply:10
73 : : * -> exit wait-loop
74 : : * -> continue rep
75 : : * apply:11
76 : : * -> set in catalog READY
77 : : *
78 : : * - Sync is in front:
79 : : * sync:10
80 : : * -> set in catalog FINISHEDCOPY
81 : : * -> set in memory SYNCWAIT
82 : : * apply:8
83 : : * -> set in memory CATCHUP
84 : : * -> continue per-table filtering
85 : : * sync:10
86 : : * -> set in catalog SYNCDONE
87 : : * -> exit
88 : : * apply:10
89 : : * -> set in catalog READY
90 : : * -> stop per-table filtering
91 : : * -> continue rep
92 : : *-------------------------------------------------------------------------
93 : : */
94 : :
95 : : #include "postgres.h"
96 : :
97 : : #include "access/table.h"
98 : : #include "access/xact.h"
99 : : #include "catalog/indexing.h"
100 : : #include "catalog/pg_subscription_rel.h"
101 : : #include "catalog/pg_type.h"
102 : : #include "commands/copy.h"
103 : : #include "miscadmin.h"
104 : : #include "nodes/makefuncs.h"
105 : : #include "parser/parse_relation.h"
106 : : #include "pgstat.h"
107 : : #include "replication/logicallauncher.h"
108 : : #include "replication/logicalrelation.h"
109 : : #include "replication/logicalworker.h"
110 : : #include "replication/origin.h"
111 : : #include "replication/slot.h"
112 : : #include "replication/walreceiver.h"
113 : : #include "replication/worker_internal.h"
114 : : #include "storage/ipc.h"
115 : : #include "storage/latch.h"
116 : : #include "storage/lmgr.h"
117 : : #include "utils/acl.h"
118 : : #include "utils/array.h"
119 : : #include "utils/builtins.h"
120 : : #include "utils/lsyscache.h"
121 : : #include "utils/rls.h"
122 : : #include "utils/snapmgr.h"
123 : : #include "utils/syscache.h"
124 : : #include "utils/usercontext.h"
125 : : #include "utils/wait_event.h"
126 : :
127 : : List *table_states_not_ready = NIL;
128 : :
129 : : static StringInfo copybuf = NULL;
130 : :
131 : : /*
132 : : * Wait until the relation sync state is set in the catalog to the expected
133 : : * one; return true when it happens.
134 : : *
135 : : * Returns false if the table sync worker or the table itself have
136 : : * disappeared, or the table state has been reset.
137 : : *
138 : : * Currently, this is used in the apply worker when transitioning from
139 : : * CATCHUP state to SYNCDONE.
140 : : */
141 : : static bool
142 : 196 : wait_for_table_state_change(Oid relid, char expected_state)
143 : : {
144 : : char state;
145 : :
146 : : for (;;)
147 : 224 : {
148 : : LogicalRepWorker *worker;
149 : : XLogRecPtr statelsn;
150 : :
151 [ - + ]: 420 : CHECK_FOR_INTERRUPTS();
152 : :
153 : 420 : InvalidateCatalogSnapshot();
154 : 420 : state = GetSubscriptionRelState(MyLogicalRepWorker->subid,
155 : : relid, &statelsn);
156 : :
157 [ - + ]: 420 : if (state == SUBREL_STATE_UNKNOWN)
158 : 0 : break;
159 : :
160 [ + + ]: 420 : if (state == expected_state)
161 : 196 : return true;
162 : :
163 : : /* Check if the sync worker is still running and bail if not. */
164 : 224 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
165 : 224 : worker = logicalrep_worker_find(WORKERTYPE_TABLESYNC,
166 : 224 : MyLogicalRepWorker->subid, relid,
167 : : false);
168 : 224 : LWLockRelease(LogicalRepWorkerLock);
169 [ - + ]: 224 : if (!worker)
170 : 0 : break;
171 : :
172 : 224 : (void) WaitLatch(MyLatch,
173 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
174 : : 1000L, WAIT_EVENT_LOGICAL_SYNC_STATE_CHANGE);
175 : :
176 : 224 : ResetLatch(MyLatch);
177 : : }
178 : :
179 : 0 : return false;
180 : : }
181 : :
182 : : /*
183 : : * Wait until the apply worker changes the state of our synchronization
184 : : * worker to the expected one.
185 : : *
186 : : * Used when transitioning from SYNCWAIT state to CATCHUP.
187 : : *
188 : : * Returns false if the apply worker has disappeared.
189 : : */
190 : : static bool
191 : 198 : wait_for_worker_state_change(char expected_state)
192 : : {
193 : : int rc;
194 : :
195 : : for (;;)
196 : 198 : {
197 : : LogicalRepWorker *worker;
198 : :
199 [ - + ]: 396 : CHECK_FOR_INTERRUPTS();
200 : :
201 : : /*
202 : : * Done if already in correct state. (We assume this fetch is atomic
203 : : * enough to not give a misleading answer if we do it with no lock.)
204 : : */
205 [ + + ]: 396 : if (MyLogicalRepWorker->relstate == expected_state)
206 : 198 : return true;
207 : :
208 : : /*
209 : : * Bail out if the apply worker has died, else signal it we're
210 : : * waiting.
211 : : */
212 : 198 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
213 : 198 : worker = logicalrep_worker_find(WORKERTYPE_APPLY,
214 : 198 : MyLogicalRepWorker->subid, InvalidOid,
215 : : false);
216 [ + - + - ]: 198 : if (worker && worker->proc)
217 : 198 : logicalrep_worker_wakeup_ptr(worker);
218 : 198 : LWLockRelease(LogicalRepWorkerLock);
219 [ - + ]: 198 : if (!worker)
220 : 0 : break;
221 : :
222 : : /*
223 : : * Wait. We expect to get a latch signal back from the apply worker,
224 : : * but use a timeout in case it dies without sending one.
225 : : */
226 : 198 : rc = WaitLatch(MyLatch,
227 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
228 : : 1000L, WAIT_EVENT_LOGICAL_SYNC_STATE_CHANGE);
229 : :
230 [ + - ]: 198 : if (rc & WL_LATCH_SET)
231 : 198 : ResetLatch(MyLatch);
232 : : }
233 : :
234 : 0 : return false;
235 : : }
236 : :
237 : : /*
238 : : * Handle table synchronization cooperation from the synchronization
239 : : * worker.
240 : : *
241 : : * If the sync worker is in CATCHUP state and reached (or passed) the
242 : : * predetermined synchronization point in the WAL stream, mark the table as
243 : : * SYNCDONE and finish.
244 : : */
245 : : void
246 : 220 : ProcessSyncingTablesForSync(XLogRecPtr current_lsn)
247 : : {
248 : 220 : SpinLockAcquire(&MyLogicalRepWorker->relmutex);
249 : :
250 [ + - ]: 220 : if (MyLogicalRepWorker->relstate == SUBREL_STATE_CATCHUP &&
251 [ + + ]: 220 : current_lsn >= MyLogicalRepWorker->relstate_lsn)
252 : : {
253 : : TimeLineID tli;
254 : 198 : char syncslotname[NAMEDATALEN] = {0};
255 : 198 : char originname[NAMEDATALEN] = {0};
256 : :
257 : 198 : MyLogicalRepWorker->relstate = SUBREL_STATE_SYNCDONE;
258 : 198 : MyLogicalRepWorker->relstate_lsn = current_lsn;
259 : :
260 : 198 : SpinLockRelease(&MyLogicalRepWorker->relmutex);
261 : :
262 : : /*
263 : : * UpdateSubscriptionRelState must be called within a transaction.
264 : : */
265 [ + - ]: 198 : if (!IsTransactionState())
266 : 198 : StartTransactionCommand();
267 : :
268 : 198 : UpdateSubscriptionRelState(MyLogicalRepWorker->subid,
269 : 198 : MyLogicalRepWorker->relid,
270 : 198 : MyLogicalRepWorker->relstate,
271 : 198 : MyLogicalRepWorker->relstate_lsn,
272 : : false);
273 : :
274 : : /*
275 : : * End streaming so that LogRepWorkerWalRcvConn can be used to drop
276 : : * the slot.
277 : : */
278 : 198 : walrcv_endstreaming(LogRepWorkerWalRcvConn, &tli);
279 : :
280 : : /*
281 : : * Cleanup the tablesync slot.
282 : : *
283 : : * This has to be done after updating the state because otherwise if
284 : : * there is an error while doing the database operations we won't be
285 : : * able to rollback dropped slot.
286 : : */
287 : 198 : ReplicationSlotNameForTablesync(MyLogicalRepWorker->subid,
288 : 198 : MyLogicalRepWorker->relid,
289 : : syncslotname,
290 : : sizeof(syncslotname));
291 : :
292 : : /*
293 : : * It is important to give an error if we are unable to drop the slot,
294 : : * otherwise, it won't be dropped till the corresponding subscription
295 : : * is dropped. So passing missing_ok = false.
296 : : */
297 : 198 : ReplicationSlotDropAtPubNode(LogRepWorkerWalRcvConn, syncslotname, false);
298 : :
299 : 198 : CommitTransactionCommand();
300 : 198 : pgstat_report_stat(false);
301 : :
302 : : /*
303 : : * Start a new transaction to clean up the tablesync origin tracking.
304 : : * This transaction will be ended within the FinishSyncWorker(). Now,
305 : : * even, if we fail to remove this here, the apply worker will ensure
306 : : * to clean it up afterward.
307 : : *
308 : : * We need to do this after the table state is set to SYNCDONE.
309 : : * Otherwise, if an error occurs while performing the database
310 : : * operation, the worker will be restarted and the in-memory state of
311 : : * replication progress (remote_lsn) won't be rolled-back which would
312 : : * have been cleared before restart. So, the restarted worker will use
313 : : * invalid replication progress state resulting in replay of
314 : : * transactions that have already been applied.
315 : : */
316 : 198 : StartTransactionCommand();
317 : :
318 : 198 : ReplicationOriginNameForLogicalRep(MyLogicalRepWorker->subid,
319 : 198 : MyLogicalRepWorker->relid,
320 : : originname,
321 : : sizeof(originname));
322 : :
323 : : /*
324 : : * Resetting the origin session removes the ownership of the slot.
325 : : * This is needed to allow the origin to be dropped.
326 : : */
327 : 198 : replorigin_session_reset();
328 : 198 : replorigin_xact_clear(true);
329 : :
330 : : /*
331 : : * Drop the tablesync's origin tracking if exists.
332 : : *
333 : : * There is a chance that the user is concurrently performing refresh
334 : : * for the subscription where we remove the table state and its origin
335 : : * or the apply worker would have removed this origin. So passing
336 : : * missing_ok = true.
337 : : */
338 : 198 : replorigin_drop_by_name(originname, true, false);
339 : :
340 : 198 : FinishSyncWorker();
341 : : }
342 : : else
343 : 22 : SpinLockRelease(&MyLogicalRepWorker->relmutex);
344 : 22 : }
345 : :
346 : : /*
347 : : * Handle table synchronization cooperation from the apply worker.
348 : : *
349 : : * Walk over all subscription tables that are individually tracked by the
350 : : * apply process (currently, all that have state other than
351 : : * SUBREL_STATE_READY) and manage synchronization for them.
352 : : *
353 : : * If there are tables that need synchronizing and are not being synchronized
354 : : * yet, start sync workers for them (if there are free slots for sync
355 : : * workers). To prevent starting the sync worker for the same relation at a
356 : : * high frequency after a failure, we store its last start time with each sync
357 : : * state info. We start the sync worker for the same relation after waiting
358 : : * at least wal_retrieve_retry_interval.
359 : : *
360 : : * For tables that are being synchronized already, check if sync workers
361 : : * either need action from the apply worker or have finished. This is the
362 : : * SYNCWAIT to CATCHUP transition.
363 : : *
364 : : * If the synchronization position is reached (SYNCDONE), then the table can
365 : : * be marked as READY and is no longer tracked.
366 : : */
367 : : void
368 : 4253 : ProcessSyncingTablesForApply(XLogRecPtr current_lsn)
369 : : {
370 : : struct tablesync_start_time_mapping
371 : : {
372 : : Oid relid;
373 : : TimestampTz last_start_time;
374 : : };
375 : : static HTAB *last_start_times = NULL;
376 : : ListCell *lc;
377 : : bool started_tx;
378 : 4253 : bool should_exit = false;
379 : 4253 : Relation rel = NULL;
380 : :
381 : : Assert(!IsTransactionState());
382 : :
383 : : /* We need up-to-date sync state info for subscription tables here. */
384 : 4253 : FetchRelationStates(NULL, NULL, &started_tx);
385 : :
386 : : /*
387 : : * Prepare a hash table for tracking last start times of workers, to avoid
388 : : * immediate restarts. We don't need it if there are no tables that need
389 : : * syncing.
390 : : */
391 [ + + + + ]: 4253 : if (table_states_not_ready != NIL && !last_start_times)
392 : 129 : {
393 : : HASHCTL ctl;
394 : :
395 : 129 : ctl.keysize = sizeof(Oid);
396 : 129 : ctl.entrysize = sizeof(struct tablesync_start_time_mapping);
397 : 129 : last_start_times = hash_create("Logical replication table sync worker start times",
398 : : 256, &ctl, HASH_ELEM | HASH_BLOBS);
399 : : }
400 : :
401 : : /*
402 : : * Clean up the hash table when we're done with all tables (just to
403 : : * release the bit of memory).
404 : : */
405 [ + + + + ]: 4124 : else if (table_states_not_ready == NIL && last_start_times)
406 : : {
407 : 96 : hash_destroy(last_start_times);
408 : 96 : last_start_times = NULL;
409 : : }
410 : :
411 : : /*
412 : : * Process all tables that are being synchronized.
413 : : */
414 [ + + + + : 6125 : foreach(lc, table_states_not_ready)
+ + ]
415 : : {
416 : 1872 : SubscriptionRelState *rstate = (SubscriptionRelState *) lfirst(lc);
417 : :
418 [ + + ]: 1872 : if (!started_tx)
419 : : {
420 : 319 : StartTransactionCommand();
421 : 319 : started_tx = true;
422 : : }
423 : :
424 : : Assert(get_rel_relkind(rstate->relid) != RELKIND_SEQUENCE);
425 : :
426 [ + + ]: 1872 : if (rstate->state == SUBREL_STATE_SYNCDONE)
427 : : {
428 : : /*
429 : : * Apply has caught up to the position where the table sync has
430 : : * finished. Mark the table as ready so that the apply will just
431 : : * continue to replicate it normally.
432 : : */
433 [ + + ]: 195 : if (current_lsn >= rstate->lsn)
434 : : {
435 : : char originname[NAMEDATALEN];
436 : :
437 : 193 : rstate->state = SUBREL_STATE_READY;
438 : 193 : rstate->lsn = current_lsn;
439 : :
440 : : /*
441 : : * Remove the tablesync origin tracking if exists.
442 : : *
443 : : * There is a chance that the user is concurrently performing
444 : : * refresh for the subscription where we remove the table
445 : : * state and its origin or the tablesync worker would have
446 : : * already removed this origin. We can't rely on tablesync
447 : : * worker to remove the origin tracking as if there is any
448 : : * error while dropping we won't restart it to drop the
449 : : * origin. So passing missing_ok = true.
450 : : *
451 : : * Lock the subscription and origin in the same order as we
452 : : * are doing during DDL commands to avoid deadlocks. See
453 : : * AlterSubscription_refresh.
454 : : */
455 : 193 : LockSharedObject(SubscriptionRelationId, MyLogicalRepWorker->subid,
456 : : 0, AccessShareLock);
457 : :
458 [ + - ]: 193 : if (!rel)
459 : 193 : rel = table_open(SubscriptionRelRelationId, RowExclusiveLock);
460 : :
461 : 193 : ReplicationOriginNameForLogicalRep(MyLogicalRepWorker->subid,
462 : : rstate->relid,
463 : : originname,
464 : : sizeof(originname));
465 : 193 : replorigin_drop_by_name(originname, true, false);
466 : :
467 : : /*
468 : : * Update the state to READY only after the origin cleanup.
469 : : */
470 : 193 : UpdateSubscriptionRelState(MyLogicalRepWorker->subid,
471 : 193 : rstate->relid, rstate->state,
472 : : rstate->lsn, true);
473 : : }
474 : : }
475 : : else
476 : : {
477 : : LogicalRepWorker *syncworker;
478 : :
479 : : /*
480 : : * Look for a sync worker for this relation.
481 : : */
482 : 1677 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
483 : :
484 : 1677 : syncworker = logicalrep_worker_find(WORKERTYPE_TABLESYNC,
485 : 1677 : MyLogicalRepWorker->subid,
486 : : rstate->relid, false);
487 : :
488 [ + + ]: 1677 : if (syncworker)
489 : : {
490 : : /* Found one, update our copy of its state */
491 : 764 : SpinLockAcquire(&syncworker->relmutex);
492 : 764 : rstate->state = syncworker->relstate;
493 : 764 : rstate->lsn = syncworker->relstate_lsn;
494 [ + + ]: 764 : if (rstate->state == SUBREL_STATE_SYNCWAIT)
495 : : {
496 : : /*
497 : : * Sync worker is waiting for apply. Tell sync worker it
498 : : * can catchup now.
499 : : */
500 : 196 : syncworker->relstate = SUBREL_STATE_CATCHUP;
501 : 196 : syncworker->relstate_lsn =
502 : 196 : Max(syncworker->relstate_lsn, current_lsn);
503 : : }
504 : 764 : SpinLockRelease(&syncworker->relmutex);
505 : :
506 : : /* If we told worker to catch up, wait for it. */
507 [ + + ]: 764 : if (rstate->state == SUBREL_STATE_SYNCWAIT)
508 : : {
509 : : /* Signal the sync worker, as it may be waiting for us. */
510 [ + - ]: 196 : if (syncworker->proc)
511 : 196 : logicalrep_worker_wakeup_ptr(syncworker);
512 : :
513 : : /* Now safe to release the LWLock */
514 : 196 : LWLockRelease(LogicalRepWorkerLock);
515 : :
516 [ + - ]: 196 : if (started_tx)
517 : : {
518 : : /*
519 : : * We must commit the existing transaction to release
520 : : * the existing locks before entering a busy loop.
521 : : * This is required to avoid any undetected deadlocks
522 : : * due to any existing lock as deadlock detector won't
523 : : * be able to detect the waits on the latch.
524 : : *
525 : : * Also close any tables prior to the commit.
526 : : */
527 [ + + ]: 196 : if (rel)
528 : : {
529 : 34 : table_close(rel, NoLock);
530 : 34 : rel = NULL;
531 : : }
532 : 196 : CommitTransactionCommand();
533 : 196 : pgstat_report_stat(false);
534 : : }
535 : :
536 : : /*
537 : : * Enter busy loop and wait for synchronization worker to
538 : : * reach expected state (or die trying).
539 : : */
540 : 196 : StartTransactionCommand();
541 : 196 : started_tx = true;
542 : :
543 : 196 : wait_for_table_state_change(rstate->relid,
544 : : SUBREL_STATE_SYNCDONE);
545 : : }
546 : : else
547 : 568 : LWLockRelease(LogicalRepWorkerLock);
548 : : }
549 : : else
550 : : {
551 : : /*
552 : : * If there is no sync worker for this table yet, count
553 : : * running sync workers for this subscription, while we have
554 : : * the lock.
555 : : */
556 : : int nsyncworkers =
557 : 913 : logicalrep_sync_worker_count(MyLogicalRepWorker->subid);
558 : : struct tablesync_start_time_mapping *hentry;
559 : : bool found;
560 : :
561 : : /* Now safe to release the LWLock */
562 : 913 : LWLockRelease(LogicalRepWorkerLock);
563 : :
564 : 913 : hentry = hash_search(last_start_times, &rstate->relid,
565 : : HASH_ENTER, &found);
566 [ + + ]: 913 : if (!found)
567 : 202 : hentry->last_start_time = 0;
568 : :
569 : 913 : launch_sync_worker(WORKERTYPE_TABLESYNC, nsyncworkers,
570 : : rstate->relid, &hentry->last_start_time);
571 : : }
572 : : }
573 : : }
574 : :
575 : : /* Close table if opened */
576 [ + + ]: 4253 : if (rel)
577 : 159 : table_close(rel, NoLock);
578 : :
579 : :
580 [ + + ]: 4253 : if (started_tx)
581 : : {
582 : : /*
583 : : * Even when the two_phase mode is requested by the user, it remains
584 : : * as 'pending' until all tablesyncs have reached READY state.
585 : : *
586 : : * When this happens, we restart the apply worker and (if the
587 : : * conditions are still ok) then the two_phase tri-state will become
588 : : * 'enabled' at that time.
589 : : *
590 : : * Note: If the subscription has no tables then leave the state as
591 : : * PENDING, which allows ALTER SUBSCRIPTION ... REFRESH PUBLICATION to
592 : : * work.
593 : : */
594 [ + + ]: 1123 : if (MySubscription->twophasestate == LOGICALREP_TWOPHASE_STATE_PENDING)
595 : : {
596 : 40 : CommandCounterIncrement(); /* make updates visible */
597 [ + + ]: 40 : if (AllTablesyncsReady())
598 : : {
599 [ + - ]: 6 : ereport(LOG,
600 : : (errmsg("logical replication apply worker for subscription \"%s\" will restart so that two_phase can be enabled",
601 : : MySubscription->name)));
602 : 6 : should_exit = true;
603 : : }
604 : : }
605 : :
606 : 1123 : CommitTransactionCommand();
607 : 1123 : pgstat_report_stat(true);
608 : : }
609 : :
610 [ + + ]: 4253 : if (should_exit)
611 : : {
612 : : /*
613 : : * Reset the last-start time for this worker so that the launcher will
614 : : * restart it without waiting for wal_retrieve_retry_interval.
615 : : */
616 : 6 : ApplyLauncherForgetWorkerStartTime(MySubscription->oid);
617 : :
618 : 6 : proc_exit(0);
619 : : }
620 : 4247 : }
621 : :
622 : : /*
623 : : * Create list of columns for COPY based on logical relation mapping.
624 : : */
625 : : static List *
626 : 209 : make_copy_attnamelist(LogicalRepRelMapEntry *rel)
627 : : {
628 : 209 : List *attnamelist = NIL;
629 : : int i;
630 : :
631 [ + + ]: 553 : for (i = 0; i < rel->remoterel.natts; i++)
632 : : {
633 : 344 : attnamelist = lappend(attnamelist,
634 : 344 : makeString(rel->remoterel.attnames[i]));
635 : : }
636 : :
637 : :
638 : 209 : return attnamelist;
639 : : }
640 : :
641 : : /*
642 : : * Data source callback for the COPY FROM, which reads from the remote
643 : : * connection and passes the data back to our local COPY.
644 : : */
645 : : static int
646 : 15057 : copy_read_data(void *outbuf, int minread, int maxread)
647 : : {
648 : 15057 : int bytesread = 0;
649 : : int avail;
650 : :
651 : : /* If there are some leftover data from previous read, use it. */
652 : 15057 : avail = copybuf->len - copybuf->cursor;
653 [ - + ]: 15057 : if (avail)
654 : : {
655 [ # # ]: 0 : if (avail > maxread)
656 : 0 : avail = maxread;
657 : 0 : memcpy(outbuf, ©buf->data[copybuf->cursor], avail);
658 : 0 : copybuf->cursor += avail;
659 : 0 : maxread -= avail;
660 : 0 : bytesread += avail;
661 : : }
662 : :
663 [ + - + - ]: 15060 : while (maxread > 0 && bytesread < minread)
664 : : {
665 : 15060 : pgsocket fd = PGINVALID_SOCKET;
666 : : int len;
667 : 15060 : char *buf = NULL;
668 : :
669 : : for (;;)
670 : : {
671 : : /* Try read the data. */
672 : 15060 : len = walrcv_receive(LogRepWorkerWalRcvConn, &buf, &fd);
673 : :
674 [ - + ]: 15060 : CHECK_FOR_INTERRUPTS();
675 : :
676 [ + + ]: 15060 : if (len == 0)
677 : 3 : break;
678 [ + + ]: 15057 : else if (len < 0)
679 : 15057 : return bytesread;
680 : : else
681 : : {
682 : : /* Process the data */
683 : 14850 : copybuf->data = buf;
684 : 14850 : copybuf->len = len;
685 : 14850 : copybuf->cursor = 0;
686 : :
687 : 14850 : avail = copybuf->len - copybuf->cursor;
688 [ - + ]: 14850 : if (avail > maxread)
689 : 0 : avail = maxread;
690 : 14850 : memcpy(outbuf, ©buf->data[copybuf->cursor], avail);
691 : 14850 : outbuf = (char *) outbuf + avail;
692 : 14850 : copybuf->cursor += avail;
693 : 14850 : maxread -= avail;
694 : 14850 : bytesread += avail;
695 : : }
696 : :
697 [ + - + - ]: 14850 : if (maxread <= 0 || bytesread >= minread)
698 : 14850 : return bytesread;
699 : : }
700 : :
701 : : /*
702 : : * Wait for more data or latch.
703 : : */
704 : 3 : (void) WaitLatchOrSocket(MyLatch,
705 : : WL_SOCKET_READABLE | WL_LATCH_SET |
706 : : WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
707 : : fd, 1000L, WAIT_EVENT_LOGICAL_SYNC_DATA);
708 : :
709 : 3 : ResetLatch(MyLatch);
710 : : }
711 : :
712 : 0 : return bytesread;
713 : : }
714 : :
715 : :
716 : : /*
717 : : * Get information about remote relation in similar fashion the RELATION
718 : : * message provides during replication.
719 : : *
720 : : * This function also returns (a) the relation qualifications to be used in
721 : : * the COPY command, and (b) whether the remote relation has published any
722 : : * generated column.
723 : : */
724 : : static void
725 : 212 : fetch_remote_table_info(char *nspname, char *relname, LogicalRepRelation *lrel,
726 : : List **qual, bool *gencol_published)
727 : : {
728 : : WalRcvExecResult *res;
729 : : StringInfoData cmd;
730 : : TupleTableSlot *slot;
731 : 212 : Oid tableRow[] = {OIDOID, CHAROID, CHAROID};
732 : 212 : Oid attrRow[] = {INT2OID, TEXTOID, OIDOID, BOOLOID, BOOLOID};
733 : 212 : Oid qualRow[] = {TEXTOID};
734 : : bool isnull;
735 : : int natt;
736 : 212 : StringInfo pub_names = NULL;
737 : 212 : Bitmapset *included_cols = NULL;
738 : 212 : int server_version = walrcv_server_version(LogRepWorkerWalRcvConn);
739 : :
740 : 212 : lrel->nspname = nspname;
741 : 212 : lrel->relname = relname;
742 : :
743 : : /* First fetch Oid and replica identity. */
744 : 212 : initStringInfo(&cmd);
745 : 212 : appendStringInfo(&cmd, "SELECT c.oid, c.relreplident, c.relkind"
746 : : " FROM pg_catalog.pg_class c"
747 : : " INNER JOIN pg_catalog.pg_namespace n"
748 : : " ON (c.relnamespace = n.oid)"
749 : : " WHERE n.nspname = %s"
750 : : " AND c.relname = %s",
751 : : quote_literal_cstr(nspname),
752 : : quote_literal_cstr(relname));
753 : 212 : res = walrcv_exec(LogRepWorkerWalRcvConn, cmd.data,
754 : : lengthof(tableRow), tableRow);
755 : :
756 [ - + ]: 212 : if (res->status != WALRCV_OK_TUPLES)
757 [ # # ]: 0 : ereport(ERROR,
758 : : (errcode(ERRCODE_CONNECTION_FAILURE),
759 : : errmsg("could not fetch table info for table \"%s.%s\" from publisher: %s",
760 : : nspname, relname, res->err)));
761 : :
762 : 212 : slot = MakeSingleTupleTableSlot(res->tupledesc, &TTSOpsMinimalTuple);
763 [ + + ]: 212 : if (!tuplestore_gettupleslot(res->tuplestore, true, false, slot))
764 [ + - ]: 1 : ereport(ERROR,
765 : : (errcode(ERRCODE_UNDEFINED_OBJECT),
766 : : errmsg("table \"%s.%s\" not found on publisher",
767 : : nspname, relname)));
768 : :
769 : 211 : lrel->remoteid = DatumGetObjectId(slot_getattr(slot, 1, &isnull));
770 : : Assert(!isnull);
771 : 211 : lrel->replident = DatumGetChar(slot_getattr(slot, 2, &isnull));
772 : : Assert(!isnull);
773 : 211 : lrel->relkind = DatumGetChar(slot_getattr(slot, 3, &isnull));
774 : : Assert(!isnull);
775 : :
776 : 211 : ExecDropSingleTupleTableSlot(slot);
777 : 211 : walrcv_clear_result(res);
778 : :
779 : :
780 : : /*
781 : : * Get column lists for each relation.
782 : : *
783 : : * We need to do this before fetching info about column names and types,
784 : : * so that we can skip columns that should not be replicated.
785 : : */
786 [ + - ]: 211 : if (server_version >= 150000)
787 : : {
788 : : WalRcvExecResult *pubres;
789 : : TupleTableSlot *tslot;
790 : 211 : Oid attrsRow[] = {INT2VECTOROID};
791 : :
792 : : /* Build the pub_names comma-separated string. */
793 : 211 : pub_names = makeStringInfo();
794 : 211 : GetPublicationsStr(MySubscription->publications, pub_names, true);
795 : :
796 : : /*
797 : : * Fetch info about column lists for the relation (from all the
798 : : * publications).
799 : : */
800 : 211 : resetStringInfo(&cmd);
801 : :
802 [ + - ]: 211 : if (server_version >= 190000)
803 : : {
804 : : /*
805 : : * We can pass both publication names and relid to
806 : : * pg_get_publication_tables() since version 19.
807 : : */
808 : 211 : appendStringInfo(&cmd,
809 : : "SELECT DISTINCT"
810 : : " (CASE WHEN (array_length(gpt.attrs, 1) = c.relnatts)"
811 : : " THEN NULL ELSE gpt.attrs END)"
812 : : " FROM pg_get_publication_tables(ARRAY[%s], %u) gpt,"
813 : : " pg_class c"
814 : : " WHERE c.oid = gpt.relid",
815 : : pub_names->data,
816 : : lrel->remoteid);
817 : : }
818 : : else
819 : 0 : appendStringInfo(&cmd,
820 : : "SELECT DISTINCT"
821 : : " (CASE WHEN (array_length(gpt.attrs, 1) = c.relnatts)"
822 : : " THEN NULL ELSE gpt.attrs END)"
823 : : " FROM pg_publication p,"
824 : : " LATERAL pg_get_publication_tables(p.pubname) gpt,"
825 : : " pg_class c"
826 : : " WHERE gpt.relid = %u AND c.oid = gpt.relid"
827 : : " AND p.pubname IN ( %s )",
828 : : lrel->remoteid,
829 : : pub_names->data);
830 : :
831 : 211 : pubres = walrcv_exec(LogRepWorkerWalRcvConn, cmd.data,
832 : : lengthof(attrsRow), attrsRow);
833 : :
834 [ - + ]: 211 : if (pubres->status != WALRCV_OK_TUPLES)
835 [ # # ]: 0 : ereport(ERROR,
836 : : (errcode(ERRCODE_CONNECTION_FAILURE),
837 : : errmsg("could not fetch column list info for table \"%s.%s\" from publisher: %s",
838 : : nspname, relname, pubres->err)));
839 : :
840 : : /*
841 : : * We don't support the case where the column list is different for
842 : : * the same table when combining publications. See comments atop
843 : : * fetch_relation_list. So there should be only one row returned.
844 : : * Although we already checked this when creating the subscription, we
845 : : * still need to check here in case the column list was changed after
846 : : * creating the subscription and before the sync worker is started.
847 : : */
848 [ - + ]: 211 : if (tuplestore_tuple_count(pubres->tuplestore) > 1)
849 [ # # ]: 0 : ereport(ERROR,
850 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
851 : : errmsg("cannot use different column lists for table \"%s.%s\" in different publications",
852 : : nspname, relname));
853 : :
854 : : /*
855 : : * Get the column list and build a single bitmap with the attnums.
856 : : *
857 : : * If we find a NULL value, it means all the columns should be
858 : : * replicated.
859 : : */
860 : 211 : tslot = MakeSingleTupleTableSlot(pubres->tupledesc, &TTSOpsMinimalTuple);
861 [ + - ]: 211 : if (tuplestore_gettupleslot(pubres->tuplestore, true, false, tslot))
862 : : {
863 : 211 : Datum cfval = slot_getattr(tslot, 1, &isnull);
864 : :
865 [ + + ]: 211 : if (!isnull)
866 : : {
867 : : ArrayType *arr;
868 : : int nelems;
869 : : int16 *elems;
870 : :
871 : 22 : arr = DatumGetArrayTypeP(cfval);
872 : 22 : nelems = ARR_DIMS(arr)[0];
873 [ - + ]: 22 : elems = (int16 *) ARR_DATA_PTR(arr);
874 : :
875 [ + + ]: 59 : for (natt = 0; natt < nelems; natt++)
876 : 37 : included_cols = bms_add_member(included_cols, elems[natt]);
877 : : }
878 : :
879 : 211 : ExecClearTuple(tslot);
880 : : }
881 : 211 : ExecDropSingleTupleTableSlot(tslot);
882 : :
883 : 211 : walrcv_clear_result(pubres);
884 : : }
885 : :
886 : : /*
887 : : * Now fetch column names and types.
888 : : */
889 : 211 : resetStringInfo(&cmd);
890 : 211 : appendStringInfoString(&cmd,
891 : : "SELECT a.attnum,"
892 : : " a.attname,"
893 : : " a.atttypid,"
894 : : " a.attnum = ANY(i.indkey)");
895 : :
896 : : /* Generated columns can be replicated since version 18. */
897 [ + - ]: 211 : if (server_version >= 180000)
898 : 211 : appendStringInfoString(&cmd, ", a.attgenerated != ''");
899 : :
900 [ + - ]: 422 : appendStringInfo(&cmd,
901 : : " FROM pg_catalog.pg_attribute a"
902 : : " LEFT JOIN pg_catalog.pg_index i"
903 : : " ON (i.indexrelid = pg_get_replica_identity_index(%u))"
904 : : " WHERE a.attnum > 0::pg_catalog.int2"
905 : : " AND NOT a.attisdropped %s"
906 : : " AND a.attrelid = %u"
907 : : " ORDER BY a.attnum",
908 : : lrel->remoteid,
909 [ - + ]: 211 : (server_version >= 120000 && server_version < 180000 ?
910 : : "AND a.attgenerated = ''" : ""),
911 : : lrel->remoteid);
912 [ + - ]: 211 : res = walrcv_exec(LogRepWorkerWalRcvConn, cmd.data,
913 : : server_version >= 180000 ? lengthof(attrRow) : lengthof(attrRow) - 1, attrRow);
914 : :
915 [ - + ]: 211 : if (res->status != WALRCV_OK_TUPLES)
916 [ # # ]: 0 : ereport(ERROR,
917 : : (errcode(ERRCODE_CONNECTION_FAILURE),
918 : : errmsg("could not fetch table info for table \"%s.%s\" from publisher: %s",
919 : : nspname, relname, res->err)));
920 : :
921 : : /* We don't know the number of rows coming, so allocate enough space. */
922 : 211 : lrel->attnames = palloc0_array(char *, MaxTupleAttributeNumber);
923 : 211 : lrel->atttyps = palloc0_array(Oid, MaxTupleAttributeNumber);
924 : 211 : lrel->attkeys = NULL;
925 : :
926 : : /*
927 : : * Store the columns as a list of names. Ignore those that are not
928 : : * present in the column list, if there is one.
929 : : */
930 : 211 : natt = 0;
931 : 211 : slot = MakeSingleTupleTableSlot(res->tupledesc, &TTSOpsMinimalTuple);
932 [ + + ]: 592 : while (tuplestore_gettupleslot(res->tuplestore, true, false, slot))
933 : : {
934 : : char *rel_colname;
935 : : AttrNumber attnum;
936 : :
937 : 381 : attnum = DatumGetInt16(slot_getattr(slot, 1, &isnull));
938 : : Assert(!isnull);
939 : :
940 : : /* If the column is not in the column list, skip it. */
941 [ + + + + ]: 381 : if (included_cols != NULL && !bms_is_member(attnum, included_cols))
942 : : {
943 : 31 : ExecClearTuple(slot);
944 : 31 : continue;
945 : : }
946 : :
947 : 350 : rel_colname = TextDatumGetCString(slot_getattr(slot, 2, &isnull));
948 : : Assert(!isnull);
949 : :
950 : 350 : lrel->attnames[natt] = rel_colname;
951 : 350 : lrel->atttyps[natt] = DatumGetObjectId(slot_getattr(slot, 3, &isnull));
952 : : Assert(!isnull);
953 : :
954 [ + + ]: 350 : if (DatumGetBool(slot_getattr(slot, 4, &isnull)))
955 : 111 : lrel->attkeys = bms_add_member(lrel->attkeys, natt);
956 : :
957 : : /* Remember if the remote table has published any generated column. */
958 [ + - + - ]: 350 : if (server_version >= 180000 && !(*gencol_published))
959 : : {
960 : 350 : *gencol_published = DatumGetBool(slot_getattr(slot, 5, &isnull));
961 : : Assert(!isnull);
962 : : }
963 : :
964 : : /* Should never happen. */
965 [ - + ]: 350 : if (++natt >= MaxTupleAttributeNumber)
966 [ # # ]: 0 : elog(ERROR, "too many columns in remote table \"%s.%s\"",
967 : : nspname, relname);
968 : :
969 : 350 : ExecClearTuple(slot);
970 : : }
971 : 211 : ExecDropSingleTupleTableSlot(slot);
972 : :
973 : 211 : lrel->natts = natt;
974 : :
975 : 211 : walrcv_clear_result(res);
976 : :
977 : : /*
978 : : * Get relation's row filter expressions. DISTINCT avoids the same
979 : : * expression of a table in multiple publications from being included
980 : : * multiple times in the final expression.
981 : : *
982 : : * We need to copy the row even if it matches just one of the
983 : : * publications, so we later combine all the quals with OR.
984 : : *
985 : : * For initial synchronization, row filtering can be ignored in following
986 : : * cases:
987 : : *
988 : : * 1) one of the subscribed publications for the table hasn't specified
989 : : * any row filter
990 : : *
991 : : * 2) one of the subscribed publications has puballtables set to true
992 : : *
993 : : * 3) one of the subscribed publications is declared as TABLES IN SCHEMA
994 : : * that includes this relation
995 : : */
996 [ + - ]: 211 : if (server_version >= 150000)
997 : : {
998 : : /* Reuse the already-built pub_names. */
999 : : Assert(pub_names != NULL);
1000 : :
1001 : : /* Check for row filters. */
1002 : 211 : resetStringInfo(&cmd);
1003 : :
1004 [ + - ]: 211 : if (server_version >= 190000)
1005 : : {
1006 : : /*
1007 : : * We can pass both publication names and relid to
1008 : : * pg_get_publication_tables() since version 19.
1009 : : */
1010 : 211 : appendStringInfo(&cmd,
1011 : : "SELECT DISTINCT pg_get_expr(gpt.qual, gpt.relid)"
1012 : : " FROM pg_get_publication_tables(ARRAY[%s], %u) gpt",
1013 : : pub_names->data,
1014 : : lrel->remoteid);
1015 : : }
1016 : : else
1017 : 0 : appendStringInfo(&cmd,
1018 : : "SELECT DISTINCT pg_get_expr(gpt.qual, gpt.relid)"
1019 : : " FROM pg_publication p,"
1020 : : " LATERAL pg_get_publication_tables(p.pubname) gpt"
1021 : : " WHERE gpt.relid = %u"
1022 : : " AND p.pubname IN ( %s )",
1023 : : lrel->remoteid,
1024 : : pub_names->data);
1025 : :
1026 : 211 : res = walrcv_exec(LogRepWorkerWalRcvConn, cmd.data, 1, qualRow);
1027 : :
1028 [ - + ]: 211 : if (res->status != WALRCV_OK_TUPLES)
1029 [ # # ]: 0 : ereport(ERROR,
1030 : : (errmsg("could not fetch table WHERE clause info for table \"%s.%s\" from publisher: %s",
1031 : : nspname, relname, res->err)));
1032 : :
1033 : : /*
1034 : : * Multiple row filter expressions for the same table will be combined
1035 : : * by COPY using OR. If any of the filter expressions for this table
1036 : : * are null, it means the whole table will be copied. In this case it
1037 : : * is not necessary to construct a unified row filter expression at
1038 : : * all.
1039 : : */
1040 : 211 : slot = MakeSingleTupleTableSlot(res->tupledesc, &TTSOpsMinimalTuple);
1041 [ + + ]: 224 : while (tuplestore_gettupleslot(res->tuplestore, true, false, slot))
1042 : : {
1043 : 213 : Datum rf = slot_getattr(slot, 1, &isnull);
1044 : :
1045 [ + + ]: 213 : if (!isnull)
1046 : 13 : *qual = lappend(*qual, makeString(TextDatumGetCString(rf)));
1047 : : else
1048 : : {
1049 : : /* Ignore filters and cleanup as necessary. */
1050 [ + + ]: 200 : if (*qual)
1051 : : {
1052 : 1 : list_free_deep(*qual);
1053 : 1 : *qual = NIL;
1054 : : }
1055 : 200 : break;
1056 : : }
1057 : :
1058 : 13 : ExecClearTuple(slot);
1059 : : }
1060 : 211 : ExecDropSingleTupleTableSlot(slot);
1061 : :
1062 : 211 : walrcv_clear_result(res);
1063 : 211 : destroyStringInfo(pub_names);
1064 : : }
1065 : :
1066 : 211 : pfree(cmd.data);
1067 : 211 : }
1068 : :
1069 : : /*
1070 : : * Copy existing data of a table from publisher.
1071 : : *
1072 : : * Caller is responsible for locking the local relation.
1073 : : */
1074 : : static void
1075 : 212 : copy_table(Relation rel)
1076 : : {
1077 : : LogicalRepRelMapEntry *relmapentry;
1078 : : LogicalRepRelation lrel;
1079 : 212 : List *qual = NIL;
1080 : : WalRcvExecResult *res;
1081 : : StringInfoData cmd;
1082 : : CopyFromState cstate;
1083 : : List *attnamelist;
1084 : : ParseState *pstate;
1085 : 212 : List *options = NIL;
1086 : 212 : bool gencol_published = false;
1087 : :
1088 : : /* Get the publisher relation info. */
1089 : 212 : fetch_remote_table_info(get_namespace_name(RelationGetNamespace(rel)),
1090 : 212 : RelationGetRelationName(rel), &lrel, &qual,
1091 : : &gencol_published);
1092 : :
1093 : : /* Put the relation into relmap. */
1094 : 211 : logicalrep_relmap_update(&lrel);
1095 : :
1096 : : /* Map the publisher relation to local one. */
1097 : 211 : relmapentry = logicalrep_rel_open(lrel.remoteid, NoLock);
1098 : : Assert(rel == relmapentry->localrel);
1099 : :
1100 : : /* Start copy on the publisher. */
1101 : 209 : initStringInfo(&cmd);
1102 : :
1103 : : /* Regular or partitioned table with no row filter or generated columns */
1104 [ + + + - ]: 209 : if ((lrel.relkind == RELKIND_RELATION || lrel.relkind == RELKIND_PARTITIONED_TABLE)
1105 [ + + + + ]: 209 : && qual == NIL && !gencol_published)
1106 : : {
1107 : 195 : appendStringInfo(&cmd, "COPY %s",
1108 : 195 : quote_qualified_identifier(lrel.nspname, lrel.relname));
1109 : :
1110 : : /* If the table has columns, then specify the columns */
1111 [ + + ]: 195 : if (lrel.natts)
1112 : : {
1113 : 194 : appendStringInfoString(&cmd, " (");
1114 : :
1115 : : /*
1116 : : * XXX Do we need to list the columns in all cases? Maybe we're
1117 : : * replicating all columns?
1118 : : */
1119 [ + + ]: 517 : for (int i = 0; i < lrel.natts; i++)
1120 : : {
1121 [ + + ]: 323 : if (i > 0)
1122 : 129 : appendStringInfoString(&cmd, ", ");
1123 : :
1124 : 323 : appendStringInfoString(&cmd, quote_identifier(lrel.attnames[i]));
1125 : : }
1126 : :
1127 : 194 : appendStringInfoChar(&cmd, ')');
1128 : : }
1129 : :
1130 : 195 : appendStringInfoString(&cmd, " TO STDOUT");
1131 : : }
1132 : : else
1133 : : {
1134 : : /*
1135 : : * For non-tables and tables with row filters, we need to do COPY
1136 : : * (SELECT ...), but we can't just do SELECT * because we may need to
1137 : : * copy only subset of columns including generated columns. For tables
1138 : : * with any row filters, build a SELECT query with OR'ed row filters
1139 : : * for COPY.
1140 : : *
1141 : : * We also need to use this same COPY (SELECT ...) syntax when
1142 : : * generated columns are published, because copy of generated columns
1143 : : * is not supported by the normal COPY.
1144 : : */
1145 : 14 : appendStringInfoString(&cmd, "COPY (SELECT ");
1146 [ + + ]: 35 : for (int i = 0; i < lrel.natts; i++)
1147 : : {
1148 : 21 : appendStringInfoString(&cmd, quote_identifier(lrel.attnames[i]));
1149 [ + + ]: 21 : if (i < lrel.natts - 1)
1150 : 7 : appendStringInfoString(&cmd, ", ");
1151 : : }
1152 : :
1153 : 14 : appendStringInfoString(&cmd, " FROM ");
1154 : :
1155 : : /*
1156 : : * For regular tables, make sure we don't copy data from a child that
1157 : : * inherits the named table as those will be copied separately.
1158 : : */
1159 [ + + ]: 14 : if (lrel.relkind == RELKIND_RELATION)
1160 : 11 : appendStringInfoString(&cmd, "ONLY ");
1161 : :
1162 : 14 : appendStringInfoString(&cmd, quote_qualified_identifier(lrel.nspname, lrel.relname));
1163 : : /* list of OR'ed filters */
1164 [ + + ]: 14 : if (qual != NIL)
1165 : : {
1166 : : ListCell *lc;
1167 : 11 : char *q = strVal(linitial(qual));
1168 : :
1169 : 11 : appendStringInfo(&cmd, " WHERE %s", q);
1170 [ + - + + : 12 : for_each_from(lc, qual, 1)
+ + ]
1171 : : {
1172 : 1 : q = strVal(lfirst(lc));
1173 : 1 : appendStringInfo(&cmd, " OR %s", q);
1174 : : }
1175 : 11 : list_free_deep(qual);
1176 : : }
1177 : :
1178 : 14 : appendStringInfoString(&cmd, ") TO STDOUT");
1179 : : }
1180 : :
1181 : : /*
1182 : : * Prior to v16, initial table synchronization will use text format even
1183 : : * if the binary option is enabled for a subscription.
1184 : : */
1185 [ + - ]: 209 : if (walrcv_server_version(LogRepWorkerWalRcvConn) >= 160000 &&
1186 [ + + ]: 209 : MySubscription->binary)
1187 : : {
1188 : 5 : appendStringInfoString(&cmd, " WITH (FORMAT binary)");
1189 : 5 : options = list_make1(makeDefElem("format",
1190 : : (Node *) makeString("binary"), -1));
1191 : : }
1192 : :
1193 : 209 : res = walrcv_exec(LogRepWorkerWalRcvConn, cmd.data, 0, NULL);
1194 : 209 : pfree(cmd.data);
1195 [ - + ]: 209 : if (res->status != WALRCV_OK_COPY_OUT)
1196 [ # # ]: 0 : ereport(ERROR,
1197 : : (errcode(ERRCODE_CONNECTION_FAILURE),
1198 : : errmsg("could not start initial contents copy for table \"%s.%s\": %s",
1199 : : lrel.nspname, lrel.relname, res->err)));
1200 : 209 : walrcv_clear_result(res);
1201 : :
1202 : 209 : copybuf = makeStringInfo();
1203 : :
1204 : 209 : pstate = make_parsestate(NULL);
1205 : 209 : (void) addRangeTableEntryForRelation(pstate, rel, AccessShareLock,
1206 : : NULL, false, false);
1207 : :
1208 : 209 : attnamelist = make_copy_attnamelist(relmapentry);
1209 : 209 : cstate = BeginCopyFrom(pstate, rel, NULL, NULL, false, copy_read_data, attnamelist, options);
1210 : :
1211 : : /* Do the copy */
1212 : 208 : (void) CopyFrom(cstate);
1213 : 198 : EndCopyFrom(cstate);
1214 : :
1215 : 198 : logicalrep_rel_close(relmapentry, NoLock);
1216 : 198 : }
1217 : :
1218 : : /*
1219 : : * Determine the tablesync slot name.
1220 : : *
1221 : : * The name must not exceed NAMEDATALEN - 1 because of remote node constraints
1222 : : * on slot name length. We append system_identifier to avoid slot_name
1223 : : * collision with subscriptions in other clusters. With the current scheme
1224 : : * pg_%u_sync_%u_UINT64_FORMAT (3 + 10 + 6 + 10 + 20 + '\0'), the maximum
1225 : : * length of slot_name will be 50.
1226 : : *
1227 : : * The returned slot name is stored in the supplied buffer (syncslotname) with
1228 : : * the given size.
1229 : : *
1230 : : * Note: We don't use the subscription slot name as part of tablesync slot name
1231 : : * because we are responsible for cleaning up these slots and it could become
1232 : : * impossible to recalculate what name to cleanup if the subscription slot name
1233 : : * had changed.
1234 : : */
1235 : : void
1236 : 415 : ReplicationSlotNameForTablesync(Oid suboid, Oid relid,
1237 : : char *syncslotname, Size szslot)
1238 : : {
1239 : 415 : snprintf(syncslotname, szslot, "pg_%u_sync_%u_" UINT64_FORMAT, suboid,
1240 : : relid, GetSystemIdentifier());
1241 : 415 : }
1242 : :
1243 : : /*
1244 : : * Start syncing the table in the sync worker.
1245 : : *
1246 : : * If nothing needs to be done to sync the table, we exit the worker without
1247 : : * any further action.
1248 : : *
1249 : : * The returned slot name is palloc'ed in current memory context.
1250 : : */
1251 : : static char *
1252 : 212 : LogicalRepSyncTableStart(XLogRecPtr *origin_startpos)
1253 : : {
1254 : : char *slotname;
1255 : : char *err;
1256 : : char relstate;
1257 : : XLogRecPtr relstate_lsn;
1258 : : Relation rel;
1259 : : AclResult aclresult;
1260 : : WalRcvExecResult *res;
1261 : : char originname[NAMEDATALEN];
1262 : : ReplOriginId originid;
1263 : : UserContext ucxt;
1264 : : bool must_use_password;
1265 : : bool run_as_owner;
1266 : :
1267 : : /* Check the state of the table synchronization. */
1268 : 212 : StartTransactionCommand();
1269 : 212 : relstate = GetSubscriptionRelState(MyLogicalRepWorker->subid,
1270 : 212 : MyLogicalRepWorker->relid,
1271 : : &relstate_lsn);
1272 : 212 : CommitTransactionCommand();
1273 : :
1274 : : /* Is the use of a password mandatory? */
1275 [ + + ]: 419 : must_use_password = MySubscription->passwordrequired &&
1276 [ - + ]: 207 : !MySubscription->ownersuperuser;
1277 : :
1278 : 212 : SpinLockAcquire(&MyLogicalRepWorker->relmutex);
1279 : 212 : MyLogicalRepWorker->relstate = relstate;
1280 : 212 : MyLogicalRepWorker->relstate_lsn = relstate_lsn;
1281 : 212 : SpinLockRelease(&MyLogicalRepWorker->relmutex);
1282 : :
1283 : : /*
1284 : : * If synchronization is already done or no longer necessary, exit now
1285 : : * that we've updated shared memory state.
1286 : : */
1287 [ - + ]: 212 : switch (relstate)
1288 : : {
1289 : 0 : case SUBREL_STATE_SYNCDONE:
1290 : : case SUBREL_STATE_READY:
1291 : : case SUBREL_STATE_UNKNOWN:
1292 : 0 : FinishSyncWorker(); /* doesn't return */
1293 : : }
1294 : :
1295 : : /* Calculate the name of the tablesync slot. */
1296 : 212 : slotname = (char *) palloc(NAMEDATALEN);
1297 : 212 : ReplicationSlotNameForTablesync(MySubscription->oid,
1298 : 212 : MyLogicalRepWorker->relid,
1299 : : slotname,
1300 : : NAMEDATALEN);
1301 : :
1302 : : /*
1303 : : * Here we use the slot name instead of the subscription name as the
1304 : : * application_name, so that it is different from the leader apply worker,
1305 : : * so that synchronous replication can distinguish them.
1306 : : */
1307 : 212 : LogRepWorkerWalRcvConn =
1308 : 212 : walrcv_connect(MySubscription->conninfo, true, true,
1309 : : must_use_password,
1310 : : slotname, &err);
1311 [ - + ]: 212 : if (LogRepWorkerWalRcvConn == NULL)
1312 [ # # ]: 0 : ereport(ERROR,
1313 : : (errcode(ERRCODE_CONNECTION_FAILURE),
1314 : : errmsg("table synchronization worker for subscription \"%s\" could not connect to the publisher: %s",
1315 : : MySubscription->name, err)));
1316 : :
1317 : : Assert(MyLogicalRepWorker->relstate == SUBREL_STATE_INIT ||
1318 : : MyLogicalRepWorker->relstate == SUBREL_STATE_DATASYNC ||
1319 : : MyLogicalRepWorker->relstate == SUBREL_STATE_FINISHEDCOPY);
1320 : :
1321 : : /* Assign the origin tracking record name. */
1322 : 212 : ReplicationOriginNameForLogicalRep(MySubscription->oid,
1323 : 212 : MyLogicalRepWorker->relid,
1324 : : originname,
1325 : : sizeof(originname));
1326 : :
1327 [ + + ]: 212 : if (MyLogicalRepWorker->relstate == SUBREL_STATE_DATASYNC)
1328 : : {
1329 : : /*
1330 : : * We have previously errored out before finishing the copy so the
1331 : : * replication slot might exist. We want to remove the slot if it
1332 : : * already exists and proceed.
1333 : : *
1334 : : * XXX We could also instead try to drop the slot, last time we failed
1335 : : * but for that, we might need to clean up the copy state as it might
1336 : : * be in the middle of fetching the rows. Also, if there is a network
1337 : : * breakdown then it wouldn't have succeeded so trying it next time
1338 : : * seems like a better bet.
1339 : : */
1340 : 10 : ReplicationSlotDropAtPubNode(LogRepWorkerWalRcvConn, slotname, true);
1341 : : }
1342 [ - + ]: 202 : else if (MyLogicalRepWorker->relstate == SUBREL_STATE_FINISHEDCOPY)
1343 : : {
1344 : : /*
1345 : : * The COPY phase was previously done, but tablesync then crashed
1346 : : * before it was able to finish normally.
1347 : : */
1348 : 0 : StartTransactionCommand();
1349 : :
1350 : : /*
1351 : : * The origin tracking name must already exist. It was created first
1352 : : * time this tablesync was launched.
1353 : : */
1354 : 0 : originid = replorigin_by_name(originname, false);
1355 : 0 : replorigin_session_setup(originid, 0);
1356 : 0 : replorigin_xact_state.origin = originid;
1357 : 0 : *origin_startpos = replorigin_session_get_progress(false);
1358 : :
1359 : 0 : CommitTransactionCommand();
1360 : :
1361 : 0 : goto copy_table_done;
1362 : : }
1363 : :
1364 : 212 : SpinLockAcquire(&MyLogicalRepWorker->relmutex);
1365 : 212 : MyLogicalRepWorker->relstate = SUBREL_STATE_DATASYNC;
1366 : 212 : MyLogicalRepWorker->relstate_lsn = InvalidXLogRecPtr;
1367 : 212 : SpinLockRelease(&MyLogicalRepWorker->relmutex);
1368 : :
1369 : : /*
1370 : : * Update the state, create the replication origin, and make them visible
1371 : : * to others.
1372 : : */
1373 : 212 : StartTransactionCommand();
1374 : 212 : UpdateSubscriptionRelState(MyLogicalRepWorker->subid,
1375 : 212 : MyLogicalRepWorker->relid,
1376 : 212 : MyLogicalRepWorker->relstate,
1377 : 212 : MyLogicalRepWorker->relstate_lsn,
1378 : : false);
1379 : :
1380 : : /*
1381 : : * Create the replication origin in a separate transaction from the one
1382 : : * that sets up the origin in shared memory. This prevents the risk that
1383 : : * changes to the origin in shared memory cannot be rolled back if the
1384 : : * transaction aborts.
1385 : : */
1386 : 212 : originid = replorigin_by_name(originname, true);
1387 [ + + ]: 212 : if (!OidIsValid(originid))
1388 : 202 : originid = replorigin_create(originname);
1389 : :
1390 : 212 : CommitTransactionCommand();
1391 : 212 : pgstat_report_stat(true);
1392 : :
1393 : 212 : StartTransactionCommand();
1394 : :
1395 : : /*
1396 : : * Use a standard write lock here. It might be better to disallow access
1397 : : * to the table while it's being synchronized. But we don't want to block
1398 : : * the main apply process from working and it has to open the relation in
1399 : : * RowExclusiveLock when remapping remote relation id to local one.
1400 : : */
1401 : 212 : rel = table_open(MyLogicalRepWorker->relid, RowExclusiveLock);
1402 : :
1403 : : /*
1404 : : * Start a transaction in the remote node in REPEATABLE READ mode. This
1405 : : * ensures that both the replication slot we create (see below) and the
1406 : : * COPY are consistent with each other.
1407 : : */
1408 : 212 : res = walrcv_exec(LogRepWorkerWalRcvConn,
1409 : : "BEGIN READ ONLY ISOLATION LEVEL REPEATABLE READ",
1410 : : 0, NULL);
1411 [ - + ]: 212 : if (res->status != WALRCV_OK_COMMAND)
1412 [ # # ]: 0 : ereport(ERROR,
1413 : : (errcode(ERRCODE_CONNECTION_FAILURE),
1414 : : errmsg("table copy could not start transaction on publisher: %s",
1415 : : res->err)));
1416 : 212 : walrcv_clear_result(res);
1417 : :
1418 : : /*
1419 : : * Create a new permanent logical decoding slot. This slot will be used
1420 : : * for the catchup phase after COPY is done, so tell it to use the
1421 : : * snapshot to make the final data consistent.
1422 : : */
1423 : 212 : walrcv_create_slot(LogRepWorkerWalRcvConn,
1424 : : slotname, false /* permanent */ , false /* two_phase */ ,
1425 : : MySubscription->failover,
1426 : : CRS_USE_SNAPSHOT, origin_startpos);
1427 : :
1428 : : /*
1429 : : * Advance the origin to the LSN got from walrcv_create_slot and then set
1430 : : * up the origin. The advancement is WAL logged for the purpose of
1431 : : * recovery. Locks are to prevent the replication origin from vanishing
1432 : : * while advancing.
1433 : : *
1434 : : * The purpose of doing these before the copy is to avoid doing the copy
1435 : : * again due to any error in advancing or setting up origin tracking.
1436 : : */
1437 : 212 : LockRelationOid(ReplicationOriginRelationId, RowExclusiveLock);
1438 : 212 : replorigin_advance(originid, *origin_startpos, InvalidXLogRecPtr,
1439 : : true /* go backward */ , true /* WAL log */ );
1440 : 212 : UnlockRelationOid(ReplicationOriginRelationId, RowExclusiveLock);
1441 : :
1442 : 212 : replorigin_session_setup(originid, 0);
1443 : 212 : replorigin_xact_state.origin = originid;
1444 : :
1445 : : /*
1446 : : * If the user did not opt to run as the owner of the subscription
1447 : : * ('run_as_owner'), then copy the table as the owner of the table.
1448 : : */
1449 : 212 : run_as_owner = MySubscription->runasowner;
1450 [ + + ]: 212 : if (!run_as_owner)
1451 : 211 : SwitchToUntrustedUser(rel->rd_rel->relowner, &ucxt);
1452 : :
1453 : : /*
1454 : : * Check that our table sync worker has permission to insert into the
1455 : : * target table.
1456 : : */
1457 : 212 : aclresult = pg_class_aclcheck(RelationGetRelid(rel), GetUserId(),
1458 : : ACL_INSERT);
1459 [ - + ]: 212 : if (aclresult != ACLCHECK_OK)
1460 : 0 : aclcheck_error(aclresult,
1461 : 0 : get_relkind_objtype(rel->rd_rel->relkind),
1462 : 0 : RelationGetRelationName(rel));
1463 : :
1464 : : /*
1465 : : * COPY FROM does not honor RLS policies. That is not a problem for
1466 : : * subscriptions owned by roles with BYPASSRLS privilege (or superuser,
1467 : : * who has it implicitly), but other roles should not be able to
1468 : : * circumvent RLS. Disallow logical replication into RLS enabled
1469 : : * relations for such roles.
1470 : : */
1471 [ - + ]: 212 : if (check_enable_rls(RelationGetRelid(rel), InvalidOid, false) == RLS_ENABLED)
1472 [ # # ]: 0 : ereport(ERROR,
1473 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1474 : : errmsg("user \"%s\" cannot replicate into relation with row-level security enabled: \"%s\"",
1475 : : GetUserNameFromId(GetUserId(), true),
1476 : : RelationGetRelationName(rel))));
1477 : :
1478 : : /* Now do the initial data copy */
1479 : 212 : PushActiveSnapshot(GetTransactionSnapshot());
1480 : 212 : copy_table(rel);
1481 : 198 : PopActiveSnapshot();
1482 : :
1483 : 198 : res = walrcv_exec(LogRepWorkerWalRcvConn, "COMMIT", 0, NULL);
1484 [ - + ]: 198 : if (res->status != WALRCV_OK_COMMAND)
1485 [ # # ]: 0 : ereport(ERROR,
1486 : : (errcode(ERRCODE_CONNECTION_FAILURE),
1487 : : errmsg("table copy could not finish transaction on publisher: %s",
1488 : : res->err)));
1489 : 198 : walrcv_clear_result(res);
1490 : :
1491 [ + + ]: 198 : if (!run_as_owner)
1492 : 197 : RestoreUserContext(&ucxt);
1493 : :
1494 : 198 : table_close(rel, NoLock);
1495 : :
1496 : : /* Make the copy visible. */
1497 : 198 : CommandCounterIncrement();
1498 : :
1499 : : /*
1500 : : * Update the persisted state to indicate the COPY phase is done; make it
1501 : : * visible to others.
1502 : : */
1503 : 198 : UpdateSubscriptionRelState(MyLogicalRepWorker->subid,
1504 : 198 : MyLogicalRepWorker->relid,
1505 : : SUBREL_STATE_FINISHEDCOPY,
1506 : 198 : MyLogicalRepWorker->relstate_lsn,
1507 : : false);
1508 : :
1509 : 198 : CommitTransactionCommand();
1510 : :
1511 : 198 : copy_table_done:
1512 : :
1513 [ - + ]: 198 : elog(DEBUG1,
1514 : : "LogicalRepSyncTableStart: '%s' origin_startpos lsn %X/%08X",
1515 : : originname, LSN_FORMAT_ARGS(*origin_startpos));
1516 : :
1517 : : /*
1518 : : * We are done with the initial data synchronization, update the state.
1519 : : */
1520 : 198 : SpinLockAcquire(&MyLogicalRepWorker->relmutex);
1521 : 198 : MyLogicalRepWorker->relstate = SUBREL_STATE_SYNCWAIT;
1522 : 198 : MyLogicalRepWorker->relstate_lsn = *origin_startpos;
1523 : 198 : SpinLockRelease(&MyLogicalRepWorker->relmutex);
1524 : :
1525 : : /*
1526 : : * Finally, wait until the leader apply worker tells us to catch up and
1527 : : * then return to let LogicalRepApplyLoop do it.
1528 : : */
1529 : 198 : wait_for_worker_state_change(SUBREL_STATE_CATCHUP);
1530 : 198 : return slotname;
1531 : : }
1532 : :
1533 : : /*
1534 : : * Execute the initial sync with error handling. Disable the subscription,
1535 : : * if it's required.
1536 : : *
1537 : : * Allocate the slot name in long-lived context on return. Note that we don't
1538 : : * handle FATAL errors which are probably because of system resource error and
1539 : : * are not repeatable.
1540 : : */
1541 : : static void
1542 : 212 : start_table_sync(XLogRecPtr *origin_startpos, char **slotname)
1543 : : {
1544 : 212 : char *sync_slotname = NULL;
1545 : :
1546 : : Assert(am_tablesync_worker());
1547 : :
1548 [ + + ]: 212 : PG_TRY();
1549 : : {
1550 : : /* Call initial sync. */
1551 : 212 : sync_slotname = LogicalRepSyncTableStart(origin_startpos);
1552 : : }
1553 : 14 : PG_CATCH();
1554 : : {
1555 [ + + ]: 14 : if (MySubscription->disableonerr)
1556 : 1 : DisableSubscriptionAndExit();
1557 : : else
1558 : : {
1559 : : /*
1560 : : * Report the worker failed during table synchronization. Abort
1561 : : * the current transaction so that the stats message is sent in an
1562 : : * idle state.
1563 : : */
1564 : 13 : AbortOutOfAnyTransaction();
1565 : 13 : pgstat_report_subscription_error(MySubscription->oid);
1566 : :
1567 : 13 : PG_RE_THROW();
1568 : : }
1569 : : }
1570 [ - + ]: 198 : PG_END_TRY();
1571 : :
1572 : : /* allocate slot name in long-lived context */
1573 : 198 : *slotname = MemoryContextStrdup(ApplyContext, sync_slotname);
1574 : 198 : pfree(sync_slotname);
1575 : 198 : }
1576 : :
1577 : : /*
1578 : : * Runs the tablesync worker.
1579 : : *
1580 : : * It starts syncing tables. After a successful sync, sets streaming options
1581 : : * and starts streaming to catchup with apply worker.
1582 : : */
1583 : : static void
1584 : 212 : run_tablesync_worker(void)
1585 : : {
1586 : : char originname[NAMEDATALEN];
1587 : 212 : XLogRecPtr origin_startpos = InvalidXLogRecPtr;
1588 : 212 : char *slotname = NULL;
1589 : : WalRcvStreamOptions options;
1590 : :
1591 : 212 : start_table_sync(&origin_startpos, &slotname);
1592 : :
1593 : 198 : ReplicationOriginNameForLogicalRep(MySubscription->oid,
1594 : 198 : MyLogicalRepWorker->relid,
1595 : : originname,
1596 : : sizeof(originname));
1597 : :
1598 : 198 : set_apply_error_context_origin(originname);
1599 : :
1600 : 198 : set_stream_options(&options, slotname, &origin_startpos);
1601 : :
1602 : 198 : walrcv_startstreaming(LogRepWorkerWalRcvConn, &options);
1603 : :
1604 : : /* Apply the changes till we catchup with the apply worker. */
1605 : 198 : start_apply(origin_startpos);
1606 : 0 : }
1607 : :
1608 : : /* Logical Replication Tablesync worker entry point */
1609 : : void
1610 : 212 : TableSyncWorkerMain(Datum main_arg)
1611 : : {
1612 : 212 : int worker_slot = DatumGetInt32(main_arg);
1613 : :
1614 : 212 : SetupApplyOrSyncWorker(worker_slot);
1615 : :
1616 : 212 : run_tablesync_worker();
1617 : :
1618 : 0 : FinishSyncWorker();
1619 : : }
1620 : :
1621 : : /*
1622 : : * If the subscription has no tables then return false.
1623 : : *
1624 : : * Otherwise, are all tablesyncs READY?
1625 : : *
1626 : : * Note: This function is not suitable to be called from outside of apply or
1627 : : * tablesync workers because MySubscription needs to be already initialized.
1628 : : */
1629 : : bool
1630 : 166 : AllTablesyncsReady(void)
1631 : : {
1632 : : bool started_tx;
1633 : : bool has_tables;
1634 : :
1635 : : /* We need up-to-date sync state info for subscription tables here. */
1636 : 166 : FetchRelationStates(&has_tables, NULL, &started_tx);
1637 : :
1638 [ + + ]: 166 : if (started_tx)
1639 : : {
1640 : 16 : CommitTransactionCommand();
1641 : 16 : pgstat_report_stat(true);
1642 : : }
1643 : :
1644 : : /*
1645 : : * Return false when there are no tables in subscription or not all tables
1646 : : * are in ready state; true otherwise.
1647 : : */
1648 [ + - + + ]: 166 : return has_tables && (table_states_not_ready == NIL);
1649 : : }
1650 : :
1651 : : /*
1652 : : * Return whether the subscription currently has any tables.
1653 : : *
1654 : : * Note: Unlike HasSubscriptionTables(), this function relies on cached
1655 : : * information for subscription tables. Additionally, it should not be
1656 : : * invoked outside of apply or tablesync workers, as MySubscription must be
1657 : : * initialized first.
1658 : : */
1659 : : bool
1660 : 82 : HasSubscriptionTablesCached(void)
1661 : : {
1662 : : bool started_tx;
1663 : : bool has_tables;
1664 : :
1665 : : /* We need up-to-date subscription tables info here */
1666 : 82 : FetchRelationStates(&has_tables, NULL, &started_tx);
1667 : :
1668 [ - + ]: 82 : if (started_tx)
1669 : : {
1670 : 0 : CommitTransactionCommand();
1671 : 0 : pgstat_report_stat(true);
1672 : : }
1673 : :
1674 : 82 : return has_tables;
1675 : : }
1676 : :
1677 : : /*
1678 : : * Update the two_phase state of the specified subscription in pg_subscription.
1679 : : */
1680 : : void
1681 : 10 : UpdateTwoPhaseState(Oid suboid, char new_state)
1682 : : {
1683 : : Relation rel;
1684 : : HeapTuple tup;
1685 : : bool nulls[Natts_pg_subscription];
1686 : : bool replaces[Natts_pg_subscription];
1687 : : Datum values[Natts_pg_subscription];
1688 : :
1689 : : Assert(new_state == LOGICALREP_TWOPHASE_STATE_DISABLED ||
1690 : : new_state == LOGICALREP_TWOPHASE_STATE_PENDING ||
1691 : : new_state == LOGICALREP_TWOPHASE_STATE_ENABLED);
1692 : :
1693 : 10 : rel = table_open(SubscriptionRelationId, RowExclusiveLock);
1694 : 10 : tup = SearchSysCacheCopy1(SUBSCRIPTIONOID, ObjectIdGetDatum(suboid));
1695 [ - + ]: 10 : if (!HeapTupleIsValid(tup))
1696 [ # # ]: 0 : elog(ERROR,
1697 : : "cache lookup failed for subscription oid %u",
1698 : : suboid);
1699 : :
1700 : : /* Form a new tuple. */
1701 : 10 : memset(values, 0, sizeof(values));
1702 : 10 : memset(nulls, false, sizeof(nulls));
1703 : 10 : memset(replaces, false, sizeof(replaces));
1704 : :
1705 : : /* And update/set two_phase state */
1706 : 10 : values[Anum_pg_subscription_subtwophasestate - 1] = CharGetDatum(new_state);
1707 : 10 : replaces[Anum_pg_subscription_subtwophasestate - 1] = true;
1708 : :
1709 : 10 : tup = heap_modify_tuple(tup, RelationGetDescr(rel),
1710 : : values, nulls, replaces);
1711 : 10 : CatalogTupleUpdate(rel, &tup->t_self, tup);
1712 : :
1713 : 10 : heap_freetuple(tup);
1714 : 10 : table_close(rel, RowExclusiveLock);
1715 : 10 : }
|
