Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : * launcher.c
3 : : * PostgreSQL logical replication worker launcher process
4 : : *
5 : : * Copyright (c) 2016-2026, PostgreSQL Global Development Group
6 : : *
7 : : * IDENTIFICATION
8 : : * src/backend/replication/logical/launcher.c
9 : : *
10 : : * NOTES
11 : : * This module contains the logical replication worker launcher which
12 : : * uses the background worker infrastructure to start the logical
13 : : * replication workers for every enabled subscription.
14 : : *
15 : : *-------------------------------------------------------------------------
16 : : */
17 : :
18 : : #include "postgres.h"
19 : :
20 : : #include "access/heapam.h"
21 : : #include "access/htup.h"
22 : : #include "access/htup_details.h"
23 : : #include "access/tableam.h"
24 : : #include "access/xact.h"
25 : : #include "catalog/pg_subscription.h"
26 : : #include "catalog/pg_subscription_rel.h"
27 : : #include "funcapi.h"
28 : : #include "lib/dshash.h"
29 : : #include "miscadmin.h"
30 : : #include "pgstat.h"
31 : : #include "postmaster/bgworker.h"
32 : : #include "postmaster/interrupt.h"
33 : : #include "replication/logicallauncher.h"
34 : : #include "replication/origin.h"
35 : : #include "replication/slot.h"
36 : : #include "replication/walreceiver.h"
37 : : #include "replication/worker_internal.h"
38 : : #include "storage/ipc.h"
39 : : #include "storage/proc.h"
40 : : #include "storage/procarray.h"
41 : : #include "tcop/tcopprot.h"
42 : : #include "utils/builtins.h"
43 : : #include "utils/memutils.h"
44 : : #include "utils/pg_lsn.h"
45 : : #include "utils/snapmgr.h"
46 : : #include "utils/syscache.h"
47 : : #include "utils/wait_event.h"
48 : :
49 : : /* max sleep time between cycles (3min) */
50 : : #define DEFAULT_NAPTIME_PER_CYCLE 180000L
51 : :
52 : : /* GUC variables */
53 : : int max_logical_replication_workers = 4;
54 : : int max_sync_workers_per_subscription = 2;
55 : : int max_parallel_apply_workers_per_subscription = 2;
56 : :
57 : : LogicalRepWorker *MyLogicalRepWorker = NULL;
58 : :
59 : : typedef struct LogicalRepCtxStruct
60 : : {
61 : : /* Supervisor process. */
62 : : pid_t launcher_pid;
63 : :
64 : : /* Hash table holding last start times of subscriptions' apply workers. */
65 : : dsa_handle last_start_dsa;
66 : : dshash_table_handle last_start_dsh;
67 : :
68 : : /* Background workers. */
69 : : LogicalRepWorker workers[FLEXIBLE_ARRAY_MEMBER];
70 : : } LogicalRepCtxStruct;
71 : :
72 : : static LogicalRepCtxStruct *LogicalRepCtx;
73 : :
74 : : /* an entry in the last-start-times shared hash table */
75 : : typedef struct LauncherLastStartTimesEntry
76 : : {
77 : : Oid subid; /* OID of logrep subscription (hash key) */
78 : : TimestampTz last_start_time; /* last time its apply worker was started */
79 : : } LauncherLastStartTimesEntry;
80 : :
81 : : /* parameters for the last-start-times shared hash table */
82 : : static const dshash_parameters dsh_params = {
83 : : sizeof(Oid),
84 : : sizeof(LauncherLastStartTimesEntry),
85 : : dshash_memcmp,
86 : : dshash_memhash,
87 : : dshash_memcpy,
88 : : LWTRANCHE_LAUNCHER_HASH
89 : : };
90 : :
91 : : static dsa_area *last_start_times_dsa = NULL;
92 : : static dshash_table *last_start_times = NULL;
93 : :
94 : : static bool on_commit_launcher_wakeup = false;
95 : :
96 : :
97 : : static void logicalrep_launcher_onexit(int code, Datum arg);
98 : : static void logicalrep_worker_onexit(int code, Datum arg);
99 : : static void logicalrep_worker_detach(void);
100 : : static void logicalrep_worker_cleanup(LogicalRepWorker *worker);
101 : : static int logicalrep_pa_worker_count(Oid subid);
102 : : static void logicalrep_launcher_attach_dshmem(void);
103 : : static void ApplyLauncherSetWorkerStartTime(Oid subid, TimestampTz start_time);
104 : : static TimestampTz ApplyLauncherGetWorkerStartTime(Oid subid);
105 : : static void compute_min_nonremovable_xid(LogicalRepWorker *worker, TransactionId *xmin);
106 : : static bool acquire_conflict_slot_if_exists(void);
107 : : static void update_conflict_slot_xmin(TransactionId new_xmin);
108 : : static void init_conflict_slot_xmin(void);
109 : :
110 : :
111 : : /*
112 : : * Load the list of subscriptions.
113 : : *
114 : : * Only the fields interesting for worker start/stop functions are filled for
115 : : * each subscription.
116 : : */
117 : : static List *
3342 peter_e@gmx.net 118 :CBC 2985 : get_subscription_list(void)
119 : : {
120 : 2985 : List *res = NIL;
121 : : Relation rel;
122 : : TableScanDesc scan;
123 : : HeapTuple tup;
124 : : MemoryContext resultcxt;
125 : :
126 : : /* This is the context that we will allocate our output data in */
127 : 2985 : resultcxt = CurrentMemoryContext;
128 : :
129 : : /*
130 : : * Start a transaction so we can access pg_subscription.
131 : : */
132 : 2985 : StartTransactionCommand();
133 : :
2610 andres@anarazel.de 134 : 2985 : rel = table_open(SubscriptionRelationId, AccessShareLock);
2561 135 : 2985 : scan = table_beginscan_catalog(rel, 0, NULL);
136 : :
3342 peter_e@gmx.net 137 [ + + ]: 3932 : while (HeapTupleIsValid(tup = heap_getnext(scan, ForwardScanDirection)))
138 : : {
139 : 947 : Form_pg_subscription subform = (Form_pg_subscription) GETSTRUCT(tup);
140 : : Subscription *sub;
141 : : MemoryContext oldcxt;
142 : :
143 : : /*
144 : : * Allocate our results in the caller's context, not the
145 : : * transaction's. We do this inside the loop, and restore the original
146 : : * context at the end, so that leaky things like heap_getnext() are
147 : : * not called in a potentially long-lived context.
148 : : */
149 : 947 : oldcxt = MemoryContextSwitchTo(resultcxt);
150 : :
95 michael@paquier.xyz 151 :GNC 947 : sub = palloc0_object(Subscription);
2672 andres@anarazel.de 152 :CBC 947 : sub->oid = subform->oid;
3342 peter_e@gmx.net 153 : 947 : sub->dbid = subform->subdbid;
154 : 947 : sub->owner = subform->subowner;
155 : 947 : sub->enabled = subform->subenabled;
156 : 947 : sub->name = pstrdup(NameStr(subform->subname));
235 akapila@postgresql.o 157 :GNC 947 : sub->retaindeadtuples = subform->subretaindeadtuples;
194 158 : 947 : sub->retentionactive = subform->subretentionactive;
159 : : /* We don't fill fields we are not interested in. */
160 : :
3342 peter_e@gmx.net 161 :CBC 947 : res = lappend(res, sub);
162 : 947 : MemoryContextSwitchTo(oldcxt);
163 : : }
164 : :
2561 andres@anarazel.de 165 : 2984 : table_endscan(scan);
2610 166 : 2984 : table_close(rel, AccessShareLock);
167 : :
3342 peter_e@gmx.net 168 : 2984 : CommitTransactionCommand();
169 : :
170 : 2984 : return res;
171 : : }
172 : :
173 : : /*
174 : : * Wait for a background worker to start up and attach to the shmem context.
175 : : *
176 : : * This is only needed for cleaning up the shared memory in case the worker
177 : : * fails to attach.
178 : : *
179 : : * Returns whether the attach was successful.
180 : : */
181 : : static bool
182 : 444 : WaitForReplicationWorkerAttach(LogicalRepWorker *worker,
183 : : uint16 generation,
184 : : BackgroundWorkerHandle *handle)
185 : : {
264 tgl@sss.pgh.pa.us 186 : 444 : bool result = false;
187 : 444 : bool dropped_latch = false;
188 : :
189 : : for (;;)
3342 peter_e@gmx.net 190 : 1240 : {
191 : : BgwHandleStatus status;
192 : : pid_t pid;
193 : : int rc;
194 : :
195 [ - + ]: 1684 : CHECK_FOR_INTERRUPTS();
196 : :
3245 197 : 1684 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
198 : :
199 : : /* Worker either died or has started. Return false if died. */
200 [ + + + + ]: 1684 : if (!worker->in_use || worker->proc)
201 : : {
264 tgl@sss.pgh.pa.us 202 : 440 : result = worker->in_use;
3245 peter_e@gmx.net 203 : 440 : LWLockRelease(LogicalRepWorkerLock);
264 tgl@sss.pgh.pa.us 204 : 440 : break;
205 : : }
206 : :
3245 peter_e@gmx.net 207 : 1244 : LWLockRelease(LogicalRepWorkerLock);
208 : :
209 : : /* Check if worker has died before attaching, and clean up after it. */
3342 210 : 1244 : status = GetBackgroundWorkerPid(handle, &pid);
211 : :
212 [ - + ]: 1244 : if (status == BGWH_STOPPED)
213 : : {
3245 peter_e@gmx.net 214 :UBC 0 : LWLockAcquire(LogicalRepWorkerLock, LW_EXCLUSIVE);
215 : : /* Ensure that this was indeed the worker we waited for. */
216 [ # # ]: 0 : if (generation == worker->generation)
217 : 0 : logicalrep_worker_cleanup(worker);
218 : 0 : LWLockRelease(LogicalRepWorkerLock);
264 tgl@sss.pgh.pa.us 219 : 0 : break; /* result is already false */
220 : : }
221 : :
222 : : /*
223 : : * We need timeout because we generally don't get notified via latch
224 : : * about the worker attach. But we don't expect to have to wait long.
225 : : */
3342 peter_e@gmx.net 226 :CBC 1244 : rc = WaitLatch(MyLatch,
227 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
228 : : 10L, WAIT_EVENT_BGWORKER_STARTUP);
229 : :
3204 andres@anarazel.de 230 [ + + ]: 1244 : if (rc & WL_LATCH_SET)
231 : : {
232 : 513 : ResetLatch(MyLatch);
233 [ + + ]: 513 : CHECK_FOR_INTERRUPTS();
264 tgl@sss.pgh.pa.us 234 : 509 : dropped_latch = true;
235 : : }
236 : : }
237 : :
238 : : /*
239 : : * If we had to clear a latch event in order to wait, be sure to restore
240 : : * it before exiting. Otherwise caller may miss events.
241 : : */
242 [ + - ]: 440 : if (dropped_latch)
243 : 440 : SetLatch(MyLatch);
244 : :
245 : 440 : return result;
246 : : }
247 : :
248 : : /*
249 : : * Walks the workers array and searches for one that matches given worker type,
250 : : * subscription id, and relation id.
251 : : *
252 : : * For both apply workers and sequencesync workers, the relid should be set to
253 : : * InvalidOid, as these workers handle changes across all tables and sequences
254 : : * respectively, rather than targeting a specific relation. For tablesync
255 : : * workers, the relid should be set to the OID of the relation being
256 : : * synchronized.
257 : : */
258 : : LogicalRepWorker *
138 akapila@postgresql.o 259 :GNC 3384 : logicalrep_worker_find(LogicalRepWorkerType wtype, Oid subid, Oid relid,
260 : : bool only_running)
261 : : {
262 : : int i;
3224 bruce@momjian.us 263 :CBC 3384 : LogicalRepWorker *res = NULL;
264 : :
265 : : /* relid must be valid only for table sync workers */
138 akapila@postgresql.o 266 [ - + ]:GNC 3384 : Assert((wtype == WORKERTYPE_TABLESYNC) == OidIsValid(relid));
3342 peter_e@gmx.net 267 [ - + ]:CBC 3384 : Assert(LWLockHeldByMe(LogicalRepWorkerLock));
268 : :
269 : : /* Search for an attached worker that matches the specified criteria. */
270 [ + + ]: 10362 : for (i = 0; i < max_logical_replication_workers; i++)
271 : : {
3224 bruce@momjian.us 272 : 9052 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
273 : :
274 : : /* Skip parallel apply workers. */
1161 akapila@postgresql.o 275 [ + + - + ]: 9052 : if (isParallelApplyWorker(w))
1161 akapila@postgresql.o 276 :UBC 0 : continue;
277 : :
3245 peter_e@gmx.net 278 [ + + + + :CBC 9052 : if (w->in_use && w->subid == subid && w->relid == relid &&
+ + ]
138 akapila@postgresql.o 279 [ + + + + :GNC 2102 : w->type == wtype && (!only_running || w->proc))
+ - ]
280 : : {
3342 peter_e@gmx.net 281 :CBC 2074 : res = w;
282 : 2074 : break;
283 : : }
284 : : }
285 : :
286 : 3384 : return res;
287 : : }
288 : :
289 : : /*
290 : : * Similar to logicalrep_worker_find(), but returns a list of all workers for
291 : : * the subscription, instead of just one.
292 : : */
293 : : List *
599 akapila@postgresql.o 294 : 708 : logicalrep_workers_find(Oid subid, bool only_running, bool acquire_lock)
295 : : {
296 : : int i;
3145 peter_e@gmx.net 297 : 708 : List *res = NIL;
298 : :
599 akapila@postgresql.o 299 [ + + ]: 708 : if (acquire_lock)
300 : 132 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
301 : :
3145 peter_e@gmx.net 302 [ - + ]: 708 : Assert(LWLockHeldByMe(LogicalRepWorkerLock));
303 : :
304 : : /* Search for attached worker for a given subscription id. */
305 [ + + ]: 3668 : for (i = 0; i < max_logical_replication_workers; i++)
306 : : {
307 : 2960 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
308 : :
309 [ + + + + : 2960 : if (w->in_use && w->subid == subid && (!only_running || w->proc))
+ + + + ]
310 : 506 : res = lappend(res, w);
311 : : }
312 : :
599 akapila@postgresql.o 313 [ + + ]: 708 : if (acquire_lock)
314 : 132 : LWLockRelease(LogicalRepWorkerLock);
315 : :
3145 peter_e@gmx.net 316 : 708 : return res;
317 : : }
318 : :
319 : : /*
320 : : * Start new logical replication background worker, if possible.
321 : : *
322 : : * Returns true on success, false on failure.
323 : : */
324 : : bool
944 akapila@postgresql.o 325 : 444 : logicalrep_worker_launch(LogicalRepWorkerType wtype,
326 : : Oid dbid, Oid subid, const char *subname, Oid userid,
327 : : Oid relid, dsm_handle subworker_dsm,
328 : : bool retain_dead_tuples)
329 : : {
330 : : BackgroundWorker bgw;
331 : : BackgroundWorkerHandle *bgw_handle;
332 : : uint16 generation;
333 : : int i;
3224 bruce@momjian.us 334 : 444 : int slot = 0;
335 : 444 : LogicalRepWorker *worker = NULL;
336 : : int nsyncworkers;
337 : : int nparallelapplyworkers;
338 : : TimestampTz now;
944 akapila@postgresql.o 339 : 444 : bool is_tablesync_worker = (wtype == WORKERTYPE_TABLESYNC);
130 akapila@postgresql.o 340 :GNC 444 : bool is_sequencesync_worker = (wtype == WORKERTYPE_SEQUENCESYNC);
944 akapila@postgresql.o 341 :CBC 444 : bool is_parallel_apply_worker = (wtype == WORKERTYPE_PARALLEL_APPLY);
342 : :
343 : : /*----------
344 : : * Sanity checks:
345 : : * - must be valid worker type
346 : : * - tablesync workers are only ones to have relid
347 : : * - parallel apply worker is the only kind of subworker
348 : : * - The replication slot used in conflict detection is created when
349 : : * retain_dead_tuples is enabled
350 : : */
351 [ - + ]: 444 : Assert(wtype != WORKERTYPE_UNKNOWN);
352 [ - + ]: 444 : Assert(is_tablesync_worker == OidIsValid(relid));
353 [ - + ]: 444 : Assert(is_parallel_apply_worker == (subworker_dsm != DSM_HANDLE_INVALID));
235 akapila@postgresql.o 354 [ + + - + ]:GNC 444 : Assert(!retain_dead_tuples || MyReplicationSlot);
355 : :
3217 peter_e@gmx.net 356 [ + + ]:CBC 444 : ereport(DEBUG1,
357 : : (errmsg_internal("starting logical replication worker for subscription \"%s\"",
358 : : subname)));
359 : :
360 : : /* Report this after the initial starting message for consistency. */
359 msawada@postgresql.o 361 [ - + ]: 444 : if (max_active_replication_origins == 0)
3342 peter_e@gmx.net 362 [ # # ]:UBC 0 : ereport(ERROR,
363 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
364 : : errmsg("cannot start logical replication workers when \"max_active_replication_origins\" is 0")));
365 : :
366 : : /*
367 : : * We need to do the modification of the shared memory under lock so that
368 : : * we have consistent view.
369 : : */
3342 peter_e@gmx.net 370 :CBC 444 : LWLockAcquire(LogicalRepWorkerLock, LW_EXCLUSIVE);
371 : :
3245 372 : 444 : retry:
373 : : /* Find unused worker slot. */
374 [ + - ]: 769 : for (i = 0; i < max_logical_replication_workers; i++)
375 : : {
376 : 769 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
377 : :
378 [ + + ]: 769 : if (!w->in_use)
379 : : {
380 : 444 : worker = w;
381 : 444 : slot = i;
3342 382 : 444 : break;
383 : : }
384 : : }
385 : :
3245 386 : 444 : nsyncworkers = logicalrep_sync_worker_count(subid);
387 : :
388 : 444 : now = GetCurrentTimestamp();
389 : :
390 : : /*
391 : : * If we didn't find a free slot, try to do garbage collection. The
392 : : * reason we do this is because if some worker failed to start up and its
393 : : * parent has crashed while waiting, the in_use state was never cleared.
394 : : */
395 [ + - - + ]: 444 : if (worker == NULL || nsyncworkers >= max_sync_workers_per_subscription)
396 : : {
3224 bruce@momjian.us 397 :UBC 0 : bool did_cleanup = false;
398 : :
3245 peter_e@gmx.net 399 [ # # ]: 0 : for (i = 0; i < max_logical_replication_workers; i++)
400 : : {
401 : 0 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
402 : :
403 : : /*
404 : : * If the worker was marked in use but didn't manage to attach in
405 : : * time, clean it up.
406 : : */
407 [ # # # # : 0 : if (w->in_use && !w->proc &&
# # ]
408 : 0 : TimestampDifferenceExceeds(w->launch_time, now,
409 : : wal_receiver_timeout))
410 : : {
411 [ # # ]: 0 : elog(WARNING,
412 : : "logical replication worker for subscription %u took too long to start; canceled",
413 : : w->subid);
414 : :
415 : 0 : logicalrep_worker_cleanup(w);
416 : 0 : did_cleanup = true;
417 : : }
418 : : }
419 : :
420 [ # # ]: 0 : if (did_cleanup)
421 : 0 : goto retry;
422 : : }
423 : :
424 : : /*
425 : : * We don't allow to invoke more sync workers once we have reached the
426 : : * sync worker limit per subscription. So, just return silently as we
427 : : * might get here because of an otherwise harmless race condition.
428 : : */
130 akapila@postgresql.o 429 [ + + + + ]:GNC 444 : if ((is_tablesync_worker || is_sequencesync_worker) &&
430 [ - + ]: 215 : nsyncworkers >= max_sync_workers_per_subscription)
431 : : {
3245 peter_e@gmx.net 432 :UBC 0 : LWLockRelease(LogicalRepWorkerLock);
1161 akapila@postgresql.o 433 : 0 : return false;
434 : : }
435 : :
1161 akapila@postgresql.o 436 :CBC 444 : nparallelapplyworkers = logicalrep_pa_worker_count(subid);
437 : :
438 : : /*
439 : : * Return false if the number of parallel apply workers reached the limit
440 : : * per subscription.
441 : : */
442 [ + + ]: 444 : if (is_parallel_apply_worker &&
443 [ - + ]: 12 : nparallelapplyworkers >= max_parallel_apply_workers_per_subscription)
444 : : {
1161 akapila@postgresql.o 445 :UBC 0 : LWLockRelease(LogicalRepWorkerLock);
446 : 0 : return false;
447 : : }
448 : :
449 : : /*
450 : : * However if there are no more free worker slots, inform user about it
451 : : * before exiting.
452 : : */
3342 peter_e@gmx.net 453 [ - + ]:CBC 444 : if (worker == NULL)
454 : : {
3337 fujii@postgresql.org 455 :UBC 0 : LWLockRelease(LogicalRepWorkerLock);
3342 peter_e@gmx.net 456 [ # # ]: 0 : ereport(WARNING,
457 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
458 : : errmsg("out of logical replication worker slots"),
459 : : errhint("You might need to increase \"%s\".", "max_logical_replication_workers")));
1161 akapila@postgresql.o 460 : 0 : return false;
461 : : }
462 : :
463 : : /* Prepare the worker slot. */
944 akapila@postgresql.o 464 :CBC 444 : worker->type = wtype;
3245 peter_e@gmx.net 465 : 444 : worker->launch_time = now;
466 : 444 : worker->in_use = true;
467 : 444 : worker->generation++;
3279 468 : 444 : worker->proc = NULL;
3342 469 : 444 : worker->dbid = dbid;
470 : 444 : worker->userid = userid;
471 : 444 : worker->subid = subid;
3279 472 : 444 : worker->relid = relid;
473 : 444 : worker->relstate = SUBREL_STATE_UNKNOWN;
474 : 444 : worker->relstate_lsn = InvalidXLogRecPtr;
1655 akapila@postgresql.o 475 : 444 : worker->stream_fileset = NULL;
1152 476 [ + + ]: 444 : worker->leader_pid = is_parallel_apply_worker ? MyProcPid : InvalidPid;
1161 477 : 444 : worker->parallel_apply = is_parallel_apply_worker;
235 akapila@postgresql.o 478 :GNC 444 : worker->oldest_nonremovable_xid = retain_dead_tuples
479 : 2 : ? MyReplicationSlot->data.xmin
480 [ + + ]: 444 : : InvalidTransactionId;
3279 peter_e@gmx.net 481 :CBC 444 : worker->last_lsn = InvalidXLogRecPtr;
482 : 444 : TIMESTAMP_NOBEGIN(worker->last_send_time);
483 : 444 : TIMESTAMP_NOBEGIN(worker->last_recv_time);
484 : 444 : worker->reply_lsn = InvalidXLogRecPtr;
485 : 444 : TIMESTAMP_NOBEGIN(worker->reply_time);
130 akapila@postgresql.o 486 :GNC 444 : worker->last_seqsync_start_time = 0;
487 : :
488 : : /* Before releasing lock, remember generation for future identification. */
3100 tgl@sss.pgh.pa.us 489 :CBC 444 : generation = worker->generation;
490 : :
3342 peter_e@gmx.net 491 : 444 : LWLockRelease(LogicalRepWorkerLock);
492 : :
493 : : /* Register the new dynamic worker. */
3255 tgl@sss.pgh.pa.us 494 : 444 : memset(&bgw, 0, sizeof(bgw));
3224 bruce@momjian.us 495 : 444 : bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
496 : : BGWORKER_BACKEND_DATABASE_CONNECTION;
3342 peter_e@gmx.net 497 : 444 : bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
986 nathan@postgresql.or 498 : 444 : snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
499 : :
936 akapila@postgresql.o 500 [ + + + + : 444 : switch (worker->type)
- - ]
501 : : {
502 : 217 : case WORKERTYPE_APPLY:
503 : 217 : snprintf(bgw.bgw_function_name, BGW_MAXLEN, "ApplyWorkerMain");
504 : 217 : snprintf(bgw.bgw_name, BGW_MAXLEN,
505 : : "logical replication apply worker for subscription %u",
506 : : subid);
507 : 217 : snprintf(bgw.bgw_type, BGW_MAXLEN, "logical replication apply worker");
508 : 217 : break;
509 : :
510 : 12 : case WORKERTYPE_PARALLEL_APPLY:
511 : 12 : snprintf(bgw.bgw_function_name, BGW_MAXLEN, "ParallelApplyWorkerMain");
512 : 12 : snprintf(bgw.bgw_name, BGW_MAXLEN,
513 : : "logical replication parallel apply worker for subscription %u",
514 : : subid);
515 : 12 : snprintf(bgw.bgw_type, BGW_MAXLEN, "logical replication parallel worker");
516 : :
517 : 12 : memcpy(bgw.bgw_extra, &subworker_dsm, sizeof(dsm_handle));
518 : 12 : break;
519 : :
130 akapila@postgresql.o 520 :GNC 9 : case WORKERTYPE_SEQUENCESYNC:
521 : 9 : snprintf(bgw.bgw_function_name, BGW_MAXLEN, "SequenceSyncWorkerMain");
522 : 9 : snprintf(bgw.bgw_name, BGW_MAXLEN,
523 : : "logical replication sequencesync worker for subscription %u",
524 : : subid);
525 : 9 : snprintf(bgw.bgw_type, BGW_MAXLEN, "logical replication sequencesync worker");
526 : 9 : break;
527 : :
936 akapila@postgresql.o 528 :CBC 206 : case WORKERTYPE_TABLESYNC:
130 akapila@postgresql.o 529 :GNC 206 : snprintf(bgw.bgw_function_name, BGW_MAXLEN, "TableSyncWorkerMain");
936 akapila@postgresql.o 530 :CBC 206 : snprintf(bgw.bgw_name, BGW_MAXLEN,
531 : : "logical replication tablesync worker for subscription %u sync %u",
532 : : subid,
533 : : relid);
534 : 206 : snprintf(bgw.bgw_type, BGW_MAXLEN, "logical replication tablesync worker");
535 : 206 : break;
536 : :
936 akapila@postgresql.o 537 :UBC 0 : case WORKERTYPE_UNKNOWN:
538 : : /* Should never happen. */
539 [ # # ]: 0 : elog(ERROR, "unknown worker type");
540 : : }
541 : :
3342 peter_e@gmx.net 542 :CBC 444 : bgw.bgw_restart_time = BGW_NEVER_RESTART;
543 : 444 : bgw.bgw_notify_pid = MyProcPid;
3252 fujii@postgresql.org 544 : 444 : bgw.bgw_main_arg = Int32GetDatum(slot);
545 : :
3342 peter_e@gmx.net 546 [ - + ]: 444 : if (!RegisterDynamicBackgroundWorker(&bgw, &bgw_handle))
547 : : {
548 : : /* Failed to start worker, so clean up the worker slot. */
3100 tgl@sss.pgh.pa.us 549 :UBC 0 : LWLockAcquire(LogicalRepWorkerLock, LW_EXCLUSIVE);
550 [ # # ]: 0 : Assert(generation == worker->generation);
551 : 0 : logicalrep_worker_cleanup(worker);
552 : 0 : LWLockRelease(LogicalRepWorkerLock);
553 : :
3342 peter_e@gmx.net 554 [ # # ]: 0 : ereport(WARNING,
555 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
556 : : errmsg("out of background worker slots"),
557 : : errhint("You might need to increase \"%s\".", "max_worker_processes")));
1161 akapila@postgresql.o 558 : 0 : return false;
559 : : }
560 : :
561 : : /* Now wait until it attaches. */
1161 akapila@postgresql.o 562 :CBC 444 : return WaitForReplicationWorkerAttach(worker, generation, bgw_handle);
563 : : }
564 : :
565 : : /*
566 : : * Internal function to stop the worker and wait until it detaches from the
567 : : * slot.
568 : : */
569 : : static void
570 : 91 : logicalrep_worker_stop_internal(LogicalRepWorker *worker, int signo)
571 : : {
572 : : uint16 generation;
573 : :
574 [ - + ]: 91 : Assert(LWLockHeldByMeInMode(LogicalRepWorkerLock, LW_SHARED));
575 : :
576 : : /*
577 : : * Remember which generation was our worker so we can check if what we see
578 : : * is still the same one.
579 : : */
3245 peter_e@gmx.net 580 : 91 : generation = worker->generation;
581 : :
582 : : /*
583 : : * If we found a worker but it does not have proc set then it is still
584 : : * starting up; wait for it to finish starting and then kill it.
585 : : */
586 [ + - + + ]: 91 : while (worker->in_use && !worker->proc)
587 : : {
588 : : int rc;
589 : :
3342 590 : 3 : LWLockRelease(LogicalRepWorkerLock);
591 : :
592 : : /* Wait a bit --- we don't expect to have to wait long. */
3204 andres@anarazel.de 593 : 3 : rc = WaitLatch(MyLatch,
594 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
595 : : 10L, WAIT_EVENT_BGWORKER_STARTUP);
596 : :
597 [ - + ]: 3 : if (rc & WL_LATCH_SET)
598 : : {
3204 andres@anarazel.de 599 :UBC 0 : ResetLatch(MyLatch);
600 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
601 : : }
602 : :
603 : : /* Recheck worker status. */
3342 peter_e@gmx.net 604 :CBC 3 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
605 : :
606 : : /*
607 : : * Check whether the worker slot is no longer used, which would mean
608 : : * that the worker has exited, or whether the worker generation is
609 : : * different, meaning that a different worker has taken the slot.
610 : : */
3245 611 [ + - - + ]: 3 : if (!worker->in_use || worker->generation != generation)
3338 peter_e@gmx.net 612 :UBC 0 : return;
613 : :
614 : : /* Worker has assigned proc, so it has started. */
3338 peter_e@gmx.net 615 [ + - ]:CBC 3 : if (worker->proc)
3342 616 : 3 : break;
617 : : }
618 : :
619 : : /* Now terminate the worker ... */
1161 akapila@postgresql.o 620 : 91 : kill(worker->proc->pid, signo);
621 : :
622 : : /* ... and wait for it to die. */
623 : : for (;;)
3342 peter_e@gmx.net 624 : 125 : {
625 : : int rc;
626 : :
627 : : /* is it gone? */
3245 628 [ + + + + ]: 216 : if (!worker->proc || worker->generation != generation)
629 : : break;
630 : :
3179 tgl@sss.pgh.pa.us 631 : 125 : LWLockRelease(LogicalRepWorkerLock);
632 : :
633 : : /* Wait a bit --- we don't expect to have to wait long. */
3204 andres@anarazel.de 634 : 125 : rc = WaitLatch(MyLatch,
635 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
636 : : 10L, WAIT_EVENT_BGWORKER_SHUTDOWN);
637 : :
638 [ + + ]: 125 : if (rc & WL_LATCH_SET)
639 : : {
640 : 43 : ResetLatch(MyLatch);
641 [ + + ]: 43 : CHECK_FOR_INTERRUPTS();
642 : : }
643 : :
3179 tgl@sss.pgh.pa.us 644 : 125 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
645 : : }
646 : : }
647 : :
648 : : /*
649 : : * Stop the logical replication worker that matches the specified worker type,
650 : : * subscription id, and relation id.
651 : : */
652 : : void
138 akapila@postgresql.o 653 :GNC 102 : logicalrep_worker_stop(LogicalRepWorkerType wtype, Oid subid, Oid relid)
654 : : {
655 : : LogicalRepWorker *worker;
656 : :
657 : : /* relid must be valid only for table sync workers */
658 [ - + ]: 102 : Assert((wtype == WORKERTYPE_TABLESYNC) == OidIsValid(relid));
659 : :
1161 akapila@postgresql.o 660 :CBC 102 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
661 : :
138 akapila@postgresql.o 662 :GNC 102 : worker = logicalrep_worker_find(wtype, subid, relid, false);
663 : :
1161 akapila@postgresql.o 664 [ + + ]:CBC 102 : if (worker)
665 : : {
666 [ + - - + ]: 80 : Assert(!isParallelApplyWorker(worker));
667 : 80 : logicalrep_worker_stop_internal(worker, SIGTERM);
668 : : }
669 : :
670 : 102 : LWLockRelease(LogicalRepWorkerLock);
671 : 102 : }
672 : :
673 : : /*
674 : : * Stop the given logical replication parallel apply worker.
675 : : *
676 : : * Node that the function sends SIGUSR2 instead of SIGTERM to the parallel apply
677 : : * worker so that the worker exits cleanly.
678 : : */
679 : : void
1041 680 : 5 : logicalrep_pa_worker_stop(ParallelApplyWorkerInfo *winfo)
681 : : {
682 : : int slot_no;
683 : : uint16 generation;
684 : : LogicalRepWorker *worker;
685 : :
686 [ - + ]: 5 : SpinLockAcquire(&winfo->shared->mutex);
687 : 5 : generation = winfo->shared->logicalrep_worker_generation;
688 : 5 : slot_no = winfo->shared->logicalrep_worker_slot_no;
689 : 5 : SpinLockRelease(&winfo->shared->mutex);
690 : :
1161 691 [ + - - + ]: 5 : Assert(slot_no >= 0 && slot_no < max_logical_replication_workers);
692 : :
693 : : /*
694 : : * Detach from the error_mq_handle for the parallel apply worker before
695 : : * stopping it. This prevents the leader apply worker from trying to
696 : : * receive the message from the error queue that might already be detached
697 : : * by the parallel apply worker.
698 : : */
1041 699 [ + - ]: 5 : if (winfo->error_mq_handle)
700 : : {
701 : 5 : shm_mq_detach(winfo->error_mq_handle);
702 : 5 : winfo->error_mq_handle = NULL;
703 : : }
704 : :
1161 705 : 5 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
706 : :
707 : 5 : worker = &LogicalRepCtx->workers[slot_no];
708 [ + - - + ]: 5 : Assert(isParallelApplyWorker(worker));
709 : :
710 : : /*
711 : : * Only stop the worker if the generation matches and the worker is alive.
712 : : */
713 [ + - + - ]: 5 : if (worker->generation == generation && worker->proc)
172 714 : 5 : logicalrep_worker_stop_internal(worker, SIGUSR2);
715 : :
3179 tgl@sss.pgh.pa.us 716 : 5 : LWLockRelease(LogicalRepWorkerLock);
3342 peter_e@gmx.net 717 : 5 : }
718 : :
719 : : /*
720 : : * Wake up (using latch) any logical replication worker that matches the
721 : : * specified worker type, subscription id, and relation id.
722 : : */
723 : : void
138 akapila@postgresql.o 724 :GNC 222 : logicalrep_worker_wakeup(LogicalRepWorkerType wtype, Oid subid, Oid relid)
725 : : {
726 : : LogicalRepWorker *worker;
727 : :
728 : : /* relid must be valid only for table sync workers */
729 [ - + ]: 222 : Assert((wtype == WORKERTYPE_TABLESYNC) == OidIsValid(relid));
730 : :
3279 peter_e@gmx.net 731 :CBC 222 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
732 : :
138 akapila@postgresql.o 733 :GNC 222 : worker = logicalrep_worker_find(wtype, subid, relid, true);
734 : :
3279 peter_e@gmx.net 735 [ + - ]:CBC 222 : if (worker)
736 : 222 : logicalrep_worker_wakeup_ptr(worker);
737 : :
3180 tgl@sss.pgh.pa.us 738 : 222 : LWLockRelease(LogicalRepWorkerLock);
3279 peter_e@gmx.net 739 : 222 : }
740 : :
741 : : /*
742 : : * Wake up (using latch) the specified logical replication worker.
743 : : *
744 : : * Caller must hold lock, else worker->proc could change under us.
745 : : */
746 : : void
747 : 676 : logicalrep_worker_wakeup_ptr(LogicalRepWorker *worker)
748 : : {
3180 tgl@sss.pgh.pa.us 749 [ - + ]: 676 : Assert(LWLockHeldByMe(LogicalRepWorkerLock));
750 : :
3279 peter_e@gmx.net 751 : 676 : SetLatch(&worker->proc->procLatch);
752 : 676 : }
753 : :
754 : : /*
755 : : * Attach to a slot.
756 : : */
757 : : void
3342 758 : 576 : logicalrep_worker_attach(int slot)
759 : : {
760 : : /* Block concurrent access. */
761 : 576 : LWLockAcquire(LogicalRepWorkerLock, LW_EXCLUSIVE);
762 : :
763 [ + - - + ]: 576 : Assert(slot >= 0 && slot < max_logical_replication_workers);
764 : 576 : MyLogicalRepWorker = &LogicalRepCtx->workers[slot];
765 : :
3245 766 [ - + ]: 576 : if (!MyLogicalRepWorker->in_use)
767 : : {
3245 peter_e@gmx.net 768 :UBC 0 : LWLockRelease(LogicalRepWorkerLock);
769 [ # # ]: 0 : ereport(ERROR,
770 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
771 : : errmsg("logical replication worker slot %d is empty, cannot attach",
772 : : slot)));
773 : : }
774 : :
3342 peter_e@gmx.net 775 [ - + ]:CBC 576 : if (MyLogicalRepWorker->proc)
776 : : {
3245 peter_e@gmx.net 777 :UBC 0 : LWLockRelease(LogicalRepWorkerLock);
3342 778 [ # # ]: 0 : ereport(ERROR,
779 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
780 : : errmsg("logical replication worker slot %d is already used by "
781 : : "another worker, cannot attach", slot)));
782 : : }
783 : :
3342 peter_e@gmx.net 784 :CBC 576 : MyLogicalRepWorker->proc = MyProc;
785 : 576 : before_shmem_exit(logicalrep_worker_onexit, (Datum) 0);
786 : :
787 : 576 : LWLockRelease(LogicalRepWorkerLock);
788 : 576 : }
789 : :
790 : : /*
791 : : * Stop the parallel apply workers if any, and detach the leader apply worker
792 : : * (cleans up the worker info).
793 : : */
794 : : static void
795 : 576 : logicalrep_worker_detach(void)
796 : : {
797 : : /* Stop the parallel apply workers. */
1161 akapila@postgresql.o 798 [ + + ]: 576 : if (am_leader_apply_worker())
799 : : {
800 : : List *workers;
801 : : ListCell *lc;
802 : :
803 : : /*
804 : : * Detach from the error_mq_handle for all parallel apply workers
805 : : * before terminating them. This prevents the leader apply worker from
806 : : * receiving the worker termination message and sending it to logs
807 : : * when the same is already done by the parallel worker.
808 : : */
809 : 347 : pa_detach_all_error_mq();
810 : :
811 : 347 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
812 : :
599 813 : 347 : workers = logicalrep_workers_find(MyLogicalRepWorker->subid, true, false);
1161 814 [ + - + + : 702 : foreach(lc, workers)
+ + ]
815 : : {
816 : 355 : LogicalRepWorker *w = (LogicalRepWorker *) lfirst(lc);
817 : :
818 [ + - + + ]: 355 : if (isParallelApplyWorker(w))
819 : 6 : logicalrep_worker_stop_internal(w, SIGTERM);
820 : : }
821 : :
822 : 347 : LWLockRelease(LogicalRepWorkerLock);
823 : :
225 tgl@sss.pgh.pa.us 824 :GNC 347 : list_free(workers);
825 : : }
826 : :
827 : : /* Block concurrent access. */
3342 peter_e@gmx.net 828 :CBC 576 : LWLockAcquire(LogicalRepWorkerLock, LW_EXCLUSIVE);
829 : :
3245 830 : 576 : logicalrep_worker_cleanup(MyLogicalRepWorker);
831 : :
3342 832 : 576 : LWLockRelease(LogicalRepWorkerLock);
833 : 576 : }
834 : :
835 : : /*
836 : : * Clean up worker info.
837 : : */
838 : : static void
3245 839 : 576 : logicalrep_worker_cleanup(LogicalRepWorker *worker)
840 : : {
841 [ - + ]: 576 : Assert(LWLockHeldByMeInMode(LogicalRepWorkerLock, LW_EXCLUSIVE));
842 : :
933 akapila@postgresql.o 843 : 576 : worker->type = WORKERTYPE_UNKNOWN;
3245 peter_e@gmx.net 844 : 576 : worker->in_use = false;
845 : 576 : worker->proc = NULL;
846 : 576 : worker->dbid = InvalidOid;
847 : 576 : worker->userid = InvalidOid;
848 : 576 : worker->subid = InvalidOid;
849 : 576 : worker->relid = InvalidOid;
1152 akapila@postgresql.o 850 : 576 : worker->leader_pid = InvalidPid;
1161 851 : 576 : worker->parallel_apply = false;
3245 peter_e@gmx.net 852 : 576 : }
853 : :
854 : : /*
855 : : * Cleanup function for logical replication launcher.
856 : : *
857 : : * Called on logical replication launcher exit.
858 : : */
859 : : static void
3252 fujii@postgresql.org 860 : 459 : logicalrep_launcher_onexit(int code, Datum arg)
861 : : {
862 : 459 : LogicalRepCtx->launcher_pid = 0;
863 : 459 : }
864 : :
865 : : /*
866 : : * Reset the last_seqsync_start_time of the sequencesync worker in the
867 : : * subscription's apply worker.
868 : : *
869 : : * Note that this value is not stored in the sequencesync worker, because that
870 : : * has finished already and is about to exit.
871 : : */
872 : : void
130 akapila@postgresql.o 873 :GNC 5 : logicalrep_reset_seqsync_start_time(void)
874 : : {
875 : : LogicalRepWorker *worker;
876 : :
877 : : /*
878 : : * The apply worker can't access last_seqsync_start_time concurrently, so
879 : : * it is okay to use SHARED lock here. See ProcessSequencesForSync().
880 : : */
881 : 5 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
882 : :
883 : 5 : worker = logicalrep_worker_find(WORKERTYPE_APPLY,
884 : 5 : MyLogicalRepWorker->subid, InvalidOid,
885 : : true);
886 [ + - ]: 5 : if (worker)
887 : 5 : worker->last_seqsync_start_time = 0;
888 : :
889 : 5 : LWLockRelease(LogicalRepWorkerLock);
890 : 5 : }
891 : :
892 : : /*
893 : : * Cleanup function.
894 : : *
895 : : * Called on logical replication worker exit.
896 : : */
897 : : static void
3342 peter_e@gmx.net 898 :CBC 576 : logicalrep_worker_onexit(int code, Datum arg)
899 : : {
900 : : /* Disconnect gracefully from the remote side. */
1768 alvherre@alvh.no-ip. 901 [ + + ]: 576 : if (LogRepWorkerWalRcvConn)
902 : 457 : walrcv_disconnect(LogRepWorkerWalRcvConn);
903 : :
3342 peter_e@gmx.net 904 : 576 : logicalrep_worker_detach();
905 : :
906 : : /* Cleanup fileset used for streaming transactions. */
1655 akapila@postgresql.o 907 [ + + ]: 576 : if (MyLogicalRepWorker->stream_fileset != NULL)
908 : 14 : FileSetDeleteAll(MyLogicalRepWorker->stream_fileset);
909 : :
910 : : /*
911 : : * Session level locks may be acquired outside of a transaction in
912 : : * parallel apply mode and will not be released when the worker
913 : : * terminates, so manually release all locks before the worker exits.
914 : : *
915 : : * The locks will be acquired once the worker is initialized.
916 : : */
1047 917 [ + + ]: 576 : if (!InitializingApplyWorker)
918 : 508 : LockReleaseAll(DEFAULT_LOCKMETHOD, true);
919 : :
3209 peter_e@gmx.net 920 : 576 : ApplyLauncherWakeup();
3342 921 : 576 : }
922 : :
923 : : /*
924 : : * Count the number of registered (not necessarily running) sync workers
925 : : * for a subscription.
926 : : */
927 : : int
3279 928 : 1390 : logicalrep_sync_worker_count(Oid subid)
929 : : {
930 : : int i;
3224 bruce@momjian.us 931 : 1390 : int res = 0;
932 : :
3279 peter_e@gmx.net 933 [ - + ]: 1390 : Assert(LWLockHeldByMe(LogicalRepWorkerLock));
934 : :
935 : : /* Search for attached worker for a given subscription id. */
936 [ + + ]: 7152 : for (i = 0; i < max_logical_replication_workers; i++)
937 : : {
3224 bruce@momjian.us 938 : 5762 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
939 : :
130 akapila@postgresql.o 940 [ + + + - :GNC 5762 : if (w->subid == subid && (isTableSyncWorker(w) || isSequenceSyncWorker(w)))
+ + + - -
+ ]
3279 peter_e@gmx.net 941 :CBC 1567 : res++;
942 : : }
943 : :
944 : 1390 : return res;
945 : : }
946 : :
947 : : /*
948 : : * Count the number of registered (but not necessarily running) parallel apply
949 : : * workers for a subscription.
950 : : */
951 : : static int
1161 akapila@postgresql.o 952 : 444 : logicalrep_pa_worker_count(Oid subid)
953 : : {
954 : : int i;
955 : 444 : int res = 0;
956 : :
957 [ - + ]: 444 : Assert(LWLockHeldByMe(LogicalRepWorkerLock));
958 : :
959 : : /*
960 : : * Scan all attached parallel apply workers, only counting those which
961 : : * have the given subscription id.
962 : : */
963 [ + + ]: 2344 : for (i = 0; i < max_logical_replication_workers; i++)
964 : : {
965 : 1900 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
966 : :
933 967 [ + + + + : 1900 : if (isParallelApplyWorker(w) && w->subid == subid)
+ - ]
1161 968 : 2 : res++;
969 : : }
970 : :
971 : 444 : return res;
972 : : }
973 : :
974 : : /*
975 : : * ApplyLauncherShmemSize
976 : : * Compute space needed for replication launcher shared memory
977 : : */
978 : : Size
3342 peter_e@gmx.net 979 : 4447 : ApplyLauncherShmemSize(void)
980 : : {
981 : : Size size;
982 : :
983 : : /*
984 : : * Need the fixed struct and the array of LogicalRepWorker.
985 : : */
986 : 4447 : size = sizeof(LogicalRepCtxStruct);
987 : 4447 : size = MAXALIGN(size);
988 : 4447 : size = add_size(size, mul_size(max_logical_replication_workers,
989 : : sizeof(LogicalRepWorker)));
990 : 4447 : return size;
991 : : }
992 : :
993 : : /*
994 : : * ApplyLauncherRegister
995 : : * Register a background worker running the logical replication launcher.
996 : : */
997 : : void
998 : 929 : ApplyLauncherRegister(void)
999 : : {
1000 : : BackgroundWorker bgw;
1001 : :
1002 : : /*
1003 : : * The logical replication launcher is disabled during binary upgrades, to
1004 : : * prevent logical replication workers from running on the source cluster.
1005 : : * That could cause replication origins to move forward after having been
1006 : : * copied to the target cluster, potentially creating conflicts with the
1007 : : * copied data files.
1008 : : */
793 michael@paquier.xyz 1009 [ + + + + ]: 929 : if (max_logical_replication_workers == 0 || IsBinaryUpgrade)
3342 peter_e@gmx.net 1010 : 57 : return;
1011 : :
3255 tgl@sss.pgh.pa.us 1012 : 872 : memset(&bgw, 0, sizeof(bgw));
3224 bruce@momjian.us 1013 : 872 : bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
1014 : : BGWORKER_BACKEND_DATABASE_CONNECTION;
3342 peter_e@gmx.net 1015 : 872 : bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
986 nathan@postgresql.or 1016 : 872 : snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
3271 rhaas@postgresql.org 1017 : 872 : snprintf(bgw.bgw_function_name, BGW_MAXLEN, "ApplyLauncherMain");
3342 peter_e@gmx.net 1018 : 872 : snprintf(bgw.bgw_name, BGW_MAXLEN,
1019 : : "logical replication launcher");
3118 1020 : 872 : snprintf(bgw.bgw_type, BGW_MAXLEN,
1021 : : "logical replication launcher");
3342 1022 : 872 : bgw.bgw_restart_time = 5;
1023 : 872 : bgw.bgw_notify_pid = 0;
1024 : 872 : bgw.bgw_main_arg = (Datum) 0;
1025 : :
1026 : 872 : RegisterBackgroundWorker(&bgw);
1027 : : }
1028 : :
1029 : : /*
1030 : : * ApplyLauncherShmemInit
1031 : : * Allocate and initialize replication launcher shared memory
1032 : : */
1033 : : void
1034 : 1150 : ApplyLauncherShmemInit(void)
1035 : : {
1036 : : bool found;
1037 : :
1038 : 1150 : LogicalRepCtx = (LogicalRepCtxStruct *)
1039 : 1150 : ShmemInitStruct("Logical Replication Launcher Data",
1040 : : ApplyLauncherShmemSize(),
1041 : : &found);
1042 : :
1043 [ + - ]: 1150 : if (!found)
1044 : : {
1045 : : int slot;
1046 : :
1047 : 1150 : memset(LogicalRepCtx, 0, ApplyLauncherShmemSize());
1048 : :
1145 tgl@sss.pgh.pa.us 1049 : 1150 : LogicalRepCtx->last_start_dsa = DSA_HANDLE_INVALID;
1050 : 1150 : LogicalRepCtx->last_start_dsh = DSHASH_HANDLE_INVALID;
1051 : :
1052 : : /* Initialize memory and spin locks for each worker slot. */
3279 peter_e@gmx.net 1053 [ + + ]: 5713 : for (slot = 0; slot < max_logical_replication_workers; slot++)
1054 : : {
1055 : 4563 : LogicalRepWorker *worker = &LogicalRepCtx->workers[slot];
1056 : :
1057 : 4563 : memset(worker, 0, sizeof(LogicalRepWorker));
1058 : 4563 : SpinLockInit(&worker->relmutex);
1059 : : }
1060 : : }
3342 1061 : 1150 : }
1062 : :
1063 : : /*
1064 : : * Initialize or attach to the dynamic shared hash table that stores the
1065 : : * last-start times, if not already done.
1066 : : * This must be called before accessing the table.
1067 : : */
1068 : : static void
1148 tgl@sss.pgh.pa.us 1069 : 807 : logicalrep_launcher_attach_dshmem(void)
1070 : : {
1071 : : MemoryContext oldcontext;
1072 : :
1073 : : /* Quick exit if we already did this. */
1145 1074 [ + + ]: 807 : if (LogicalRepCtx->last_start_dsh != DSHASH_HANDLE_INVALID &&
1148 1075 [ + + ]: 749 : last_start_times != NULL)
1076 : 547 : return;
1077 : :
1078 : : /* Otherwise, use a lock to ensure only one process creates the table. */
1079 : 260 : LWLockAcquire(LogicalRepWorkerLock, LW_EXCLUSIVE);
1080 : :
1081 : : /* Be sure any local memory allocated by DSA routines is persistent. */
1082 : 260 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1083 : :
1145 1084 [ + + ]: 260 : if (LogicalRepCtx->last_start_dsh == DSHASH_HANDLE_INVALID)
1085 : : {
1086 : : /* Initialize dynamic shared hash table for last-start times. */
1148 1087 : 58 : last_start_times_dsa = dsa_create(LWTRANCHE_LAUNCHER_DSA);
1088 : 58 : dsa_pin(last_start_times_dsa);
1089 : 58 : dsa_pin_mapping(last_start_times_dsa);
748 nathan@postgresql.or 1090 : 58 : last_start_times = dshash_create(last_start_times_dsa, &dsh_params, NULL);
1091 : :
1092 : : /* Store handles in shared memory for other backends to use. */
1148 tgl@sss.pgh.pa.us 1093 : 58 : LogicalRepCtx->last_start_dsa = dsa_get_handle(last_start_times_dsa);
1094 : 58 : LogicalRepCtx->last_start_dsh = dshash_get_hash_table_handle(last_start_times);
1095 : : }
1096 [ + - ]: 202 : else if (!last_start_times)
1097 : : {
1098 : : /* Attach to existing dynamic shared hash table. */
1099 : 202 : last_start_times_dsa = dsa_attach(LogicalRepCtx->last_start_dsa);
1100 : 202 : dsa_pin_mapping(last_start_times_dsa);
1101 : 202 : last_start_times = dshash_attach(last_start_times_dsa, &dsh_params,
282 nathan@postgresql.or 1102 : 202 : LogicalRepCtx->last_start_dsh, NULL);
1103 : : }
1104 : :
1148 tgl@sss.pgh.pa.us 1105 : 260 : MemoryContextSwitchTo(oldcontext);
1106 : 260 : LWLockRelease(LogicalRepWorkerLock);
1107 : : }
1108 : :
1109 : : /*
1110 : : * Set the last-start time for the subscription.
1111 : : */
1112 : : static void
1113 : 217 : ApplyLauncherSetWorkerStartTime(Oid subid, TimestampTz start_time)
1114 : : {
1115 : : LauncherLastStartTimesEntry *entry;
1116 : : bool found;
1117 : :
1118 : 217 : logicalrep_launcher_attach_dshmem();
1119 : :
1120 : 217 : entry = dshash_find_or_insert(last_start_times, &subid, &found);
1121 : 217 : entry->last_start_time = start_time;
1122 : 217 : dshash_release_lock(last_start_times, entry);
1123 : 217 : }
1124 : :
1125 : : /*
1126 : : * Return the last-start time for the subscription, or 0 if there isn't one.
1127 : : */
1128 : : static TimestampTz
1129 : 342 : ApplyLauncherGetWorkerStartTime(Oid subid)
1130 : : {
1131 : : LauncherLastStartTimesEntry *entry;
1132 : : TimestampTz ret;
1133 : :
1134 : 342 : logicalrep_launcher_attach_dshmem();
1135 : :
1136 : 342 : entry = dshash_find(last_start_times, &subid, false);
1137 [ + + ]: 342 : if (entry == NULL)
1138 : 129 : return 0;
1139 : :
1140 : 213 : ret = entry->last_start_time;
1141 : 213 : dshash_release_lock(last_start_times, entry);
1142 : :
1143 : 213 : return ret;
1144 : : }
1145 : :
1146 : : /*
1147 : : * Remove the last-start-time entry for the subscription, if one exists.
1148 : : *
1149 : : * This has two use-cases: to remove the entry related to a subscription
1150 : : * that's been deleted or disabled (just to avoid leaking shared memory),
1151 : : * and to allow immediate restart of an apply worker that has exited
1152 : : * due to subscription parameter changes.
1153 : : */
1154 : : void
1155 : 248 : ApplyLauncherForgetWorkerStartTime(Oid subid)
1156 : : {
1157 : 248 : logicalrep_launcher_attach_dshmem();
1158 : :
1159 : 248 : (void) dshash_delete_key(last_start_times, &subid);
1160 : 248 : }
1161 : :
1162 : : /*
1163 : : * Wakeup the launcher on commit if requested.
1164 : : */
1165 : : void
3240 peter_e@gmx.net 1166 : 337745 : AtEOXact_ApplyLauncher(bool isCommit)
1167 : : {
3145 1168 [ + + ]: 337745 : if (isCommit)
1169 : : {
1170 [ + + ]: 310672 : if (on_commit_launcher_wakeup)
1171 : 150 : ApplyLauncherWakeup();
1172 : : }
1173 : :
3240 1174 : 337745 : on_commit_launcher_wakeup = false;
3342 1175 : 337745 : }
1176 : :
1177 : : /*
1178 : : * Request wakeup of the launcher on commit of the transaction.
1179 : : *
1180 : : * This is used to send launcher signal to stop sleeping and process the
1181 : : * subscriptions when current transaction commits. Should be used when new
1182 : : * tuple was added to the pg_subscription catalog.
1183 : : */
1184 : : void
1185 : 151 : ApplyLauncherWakeupAtCommit(void)
1186 : : {
3324 heikki.linnakangas@i 1187 [ + + ]: 151 : if (!on_commit_launcher_wakeup)
1188 : 150 : on_commit_launcher_wakeup = true;
3342 peter_e@gmx.net 1189 : 151 : }
1190 : :
1191 : : /*
1192 : : * Wakeup the launcher immediately.
1193 : : */
1194 : : void
1195 : 765 : ApplyLauncherWakeup(void)
1196 : : {
3252 fujii@postgresql.org 1197 [ + + ]: 765 : if (LogicalRepCtx->launcher_pid != 0)
3342 peter_e@gmx.net 1198 : 743 : kill(LogicalRepCtx->launcher_pid, SIGUSR1);
1199 : 765 : }
1200 : :
1201 : : /*
1202 : : * Main loop for the apply launcher process.
1203 : : */
1204 : : void
1205 : 459 : ApplyLauncherMain(Datum main_arg)
1206 : : {
3292 tgl@sss.pgh.pa.us 1207 [ + + ]: 459 : ereport(DEBUG1,
1208 : : (errmsg_internal("logical replication launcher started")));
1209 : :
3252 fujii@postgresql.org 1210 : 459 : before_shmem_exit(logicalrep_launcher_onexit, (Datum) 0);
1211 : :
3202 andres@anarazel.de 1212 [ - + ]: 459 : Assert(LogicalRepCtx->launcher_pid == 0);
1213 : 459 : LogicalRepCtx->launcher_pid = MyProcPid;
1214 : :
1215 : : /* Establish signal handlers. */
2067 akapila@postgresql.o 1216 : 459 : pqsignal(SIGHUP, SignalHandlerForConfigReload);
3342 peter_e@gmx.net 1217 : 459 : BackgroundWorkerUnblockSignals();
1218 : :
1219 : : /*
1220 : : * Establish connection to nailed catalogs (we only ever access
1221 : : * pg_subscription).
1222 : : */
2901 magnus@hagander.net 1223 : 459 : BackgroundWorkerInitializeConnection(NULL, NULL, 0);
1224 : :
1225 : : /*
1226 : : * Acquire the conflict detection slot at startup to ensure it can be
1227 : : * dropped if no longer needed after a restart.
1228 : : */
235 akapila@postgresql.o 1229 :GNC 459 : acquire_conflict_slot_if_exists();
1230 : :
1231 : : /* Enter main loop */
1232 : : for (;;)
3342 peter_e@gmx.net 1233 :CBC 2527 : {
1234 : : int rc;
1235 : : List *sublist;
1236 : : ListCell *lc;
1237 : : MemoryContext subctx;
1238 : : MemoryContext oldctx;
3224 bruce@momjian.us 1239 : 2986 : long wait_time = DEFAULT_NAPTIME_PER_CYCLE;
194 akapila@postgresql.o 1240 :GNC 2986 : bool can_update_xmin = true;
235 1241 : 2986 : bool retain_dead_tuples = false;
1242 : 2986 : TransactionId xmin = InvalidTransactionId;
1243 : :
3202 andres@anarazel.de 1244 [ + + ]:CBC 2986 : CHECK_FOR_INTERRUPTS();
1245 : :
1246 : : /* Use temporary context to avoid leaking memory across cycles. */
1148 tgl@sss.pgh.pa.us 1247 : 2985 : subctx = AllocSetContextCreate(TopMemoryContext,
1248 : : "Logical Replication Launcher sublist",
1249 : : ALLOCSET_DEFAULT_SIZES);
1250 : 2985 : oldctx = MemoryContextSwitchTo(subctx);
1251 : :
1252 : : /*
1253 : : * Start any missing workers for enabled subscriptions.
1254 : : *
1255 : : * Also, during the iteration through all subscriptions, we compute
1256 : : * the minimum XID required to protect deleted tuples for conflict
1257 : : * detection if one of the subscription enables retain_dead_tuples
1258 : : * option.
1259 : : */
1260 : 2985 : sublist = get_subscription_list();
1261 [ + + + + : 3928 : foreach(lc, sublist)
+ + ]
1262 : : {
1263 : 947 : Subscription *sub = (Subscription *) lfirst(lc);
1264 : : LogicalRepWorker *w;
1265 : : TimestampTz last_start;
1266 : : TimestampTz now;
1267 : : long elapsed;
1268 : :
235 akapila@postgresql.o 1269 [ + + ]:GNC 947 : if (sub->retaindeadtuples)
1270 : : {
1271 : 67 : retain_dead_tuples = true;
1272 : :
1273 : : /*
1274 : : * Create a replication slot to retain information necessary
1275 : : * for conflict detection such as dead tuples, commit
1276 : : * timestamps, and origins.
1277 : : *
1278 : : * The slot is created before starting the apply worker to
1279 : : * prevent it from unnecessarily maintaining its
1280 : : * oldest_nonremovable_xid.
1281 : : *
1282 : : * The slot is created even for a disabled subscription to
1283 : : * ensure that conflict-related information is available when
1284 : : * applying remote changes that occurred before the
1285 : : * subscription was enabled.
1286 : : */
1287 : 67 : CreateConflictDetectionSlot();
1288 : :
194 1289 [ + - ]: 67 : if (sub->retentionactive)
1290 : : {
1291 : : /*
1292 : : * Can't advance xmin of the slot unless all the
1293 : : * subscriptions actively retaining dead tuples are
1294 : : * enabled. This is required to ensure that we don't
1295 : : * advance the xmin of CONFLICT_DETECTION_SLOT if one of
1296 : : * the subscriptions is not enabled. Otherwise, we won't
1297 : : * be able to detect conflicts reliably for such a
1298 : : * subscription even though it has set the
1299 : : * retain_dead_tuples option.
1300 : : */
1301 : 67 : can_update_xmin &= sub->enabled;
1302 : :
1303 : : /*
1304 : : * Initialize the slot once the subscription activates
1305 : : * retention.
1306 : : */
1307 [ - + ]: 67 : if (!TransactionIdIsValid(MyReplicationSlot->data.xmin))
194 akapila@postgresql.o 1308 :UNC 0 : init_conflict_slot_xmin();
1309 : : }
1310 : : }
1311 : :
1148 tgl@sss.pgh.pa.us 1312 [ + + ]:CBC 947 : if (!sub->enabled)
1313 : 41 : continue;
1314 : :
1315 : 906 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
138 akapila@postgresql.o 1316 :GNC 906 : w = logicalrep_worker_find(WORKERTYPE_APPLY, sub->oid, InvalidOid,
1317 : : false);
1318 : :
1148 tgl@sss.pgh.pa.us 1319 [ + + ]:CBC 906 : if (w != NULL)
1320 : : {
1321 : : /*
1322 : : * Compute the minimum xmin required to protect dead tuples
1323 : : * required for conflict detection among all running apply
1324 : : * workers. This computation is performed while holding
1325 : : * LogicalRepWorkerLock to prevent accessing invalid worker
1326 : : * data, in scenarios where a worker might exit and reset its
1327 : : * state concurrently.
1328 : : */
194 akapila@postgresql.o 1329 [ + + ]:GNC 564 : if (sub->retaindeadtuples &&
1330 [ + - + - ]: 63 : sub->retentionactive &&
1331 : : can_update_xmin)
235 1332 : 63 : compute_min_nonremovable_xid(w, &xmin);
1333 : :
181 1334 : 564 : LWLockRelease(LogicalRepWorkerLock);
1335 : :
1336 : : /* worker is running already */
235 1337 : 564 : continue;
1338 : : }
1339 : :
181 1340 : 342 : LWLockRelease(LogicalRepWorkerLock);
1341 : :
1342 : : /*
1343 : : * Can't advance xmin of the slot unless all the workers
1344 : : * corresponding to subscriptions actively retaining dead tuples
1345 : : * are running, disabling the further computation of the minimum
1346 : : * nonremovable xid.
1347 : : */
194 1348 [ + + + - ]: 342 : if (sub->retaindeadtuples && sub->retentionactive)
1349 : 2 : can_update_xmin = false;
1350 : :
1351 : : /*
1352 : : * If the worker is eligible to start now, launch it. Otherwise,
1353 : : * adjust wait_time so that we'll wake up as soon as it can be
1354 : : * started.
1355 : : *
1356 : : * Each subscription's apply worker can only be restarted once per
1357 : : * wal_retrieve_retry_interval, so that errors do not cause us to
1358 : : * repeatedly restart the worker as fast as possible. In cases
1359 : : * where a restart is expected (e.g., subscription parameter
1360 : : * changes), another process should remove the last-start entry
1361 : : * for the subscription so that the worker can be restarted
1362 : : * without waiting for wal_retrieve_retry_interval to elapse.
1363 : : */
1148 tgl@sss.pgh.pa.us 1364 :CBC 342 : last_start = ApplyLauncherGetWorkerStartTime(sub->oid);
1365 : 342 : now = GetCurrentTimestamp();
1366 [ + + ]: 342 : if (last_start == 0 ||
1367 [ + + ]: 213 : (elapsed = TimestampDifferenceMilliseconds(last_start, now)) >= wal_retrieve_retry_interval)
1368 : : {
1369 : 217 : ApplyLauncherSetWorkerStartTime(sub->oid, now);
264 1370 [ + + ]: 215 : if (!logicalrep_worker_launch(WORKERTYPE_APPLY,
1371 : 217 : sub->dbid, sub->oid, sub->name,
1372 : : sub->owner, InvalidOid,
1373 : : DSM_HANDLE_INVALID,
194 akapila@postgresql.o 1374 [ + + ]:GNC 219 : sub->retaindeadtuples &&
1375 [ + - ]: 2 : sub->retentionactive))
1376 : : {
1377 : : /*
1378 : : * We get here either if we failed to launch a worker
1379 : : * (perhaps for resource-exhaustion reasons) or if we
1380 : : * launched one but it immediately quit. Either way, it
1381 : : * seems appropriate to try again after
1382 : : * wal_retrieve_retry_interval.
1383 : : */
264 tgl@sss.pgh.pa.us 1384 :CBC 1 : wait_time = Min(wait_time,
1385 : : wal_retrieve_retry_interval);
1386 : : }
1387 : : }
1388 : : else
1389 : : {
1148 1390 : 125 : wait_time = Min(wait_time,
1391 : : wal_retrieve_retry_interval - elapsed);
1392 : : }
1393 : : }
1394 : :
1395 : : /*
1396 : : * Drop the CONFLICT_DETECTION_SLOT slot if there is no subscription
1397 : : * that requires us to retain dead tuples. Otherwise, if required,
1398 : : * advance the slot's xmin to protect dead tuples required for the
1399 : : * conflict detection.
1400 : : *
1401 : : * Additionally, if all apply workers for subscriptions with
1402 : : * retain_dead_tuples enabled have requested to stop retention, the
1403 : : * slot's xmin will be set to InvalidTransactionId allowing the
1404 : : * removal of dead tuples.
1405 : : */
235 akapila@postgresql.o 1406 [ + + ]:GNC 2981 : if (MyReplicationSlot)
1407 : : {
1408 [ + + ]: 68 : if (!retain_dead_tuples)
1409 : 1 : ReplicationSlotDropAcquired();
194 1410 [ + + ]: 67 : else if (can_update_xmin)
1411 : 63 : update_conflict_slot_xmin(xmin);
1412 : : }
1413 : :
1414 : : /* Switch back to original memory context. */
1148 tgl@sss.pgh.pa.us 1415 :CBC 2981 : MemoryContextSwitchTo(oldctx);
1416 : : /* Clean the temporary memory. */
1417 : 2981 : MemoryContextDelete(subctx);
1418 : :
1419 : : /* Wait for more work. */
3204 andres@anarazel.de 1420 : 2981 : rc = WaitLatch(MyLatch,
1421 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
1422 : : wait_time,
1423 : : WAIT_EVENT_LOGICAL_LAUNCHER_MAIN);
1424 : :
1425 [ + + ]: 2978 : if (rc & WL_LATCH_SET)
1426 : : {
1427 : 2952 : ResetLatch(MyLatch);
1428 [ + + ]: 2952 : CHECK_FOR_INTERRUPTS();
1429 : : }
1430 : :
2280 rhaas@postgresql.org 1431 [ + + ]: 2527 : if (ConfigReloadPending)
1432 : : {
1433 : 46 : ConfigReloadPending = false;
3261 peter_e@gmx.net 1434 : 46 : ProcessConfigFile(PGC_SIGHUP);
1435 : : }
1436 : : }
1437 : :
1438 : : /* Not reachable */
1439 : : }
1440 : :
1441 : : /*
1442 : : * Determine the minimum non-removable transaction ID across all apply workers
1443 : : * for subscriptions that have retain_dead_tuples enabled. Store the result
1444 : : * in *xmin.
1445 : : */
1446 : : static void
235 akapila@postgresql.o 1447 :GNC 63 : compute_min_nonremovable_xid(LogicalRepWorker *worker, TransactionId *xmin)
1448 : : {
1449 : : TransactionId nonremovable_xid;
1450 : :
1451 [ - + ]: 63 : Assert(worker != NULL);
1452 : :
1453 : : /*
1454 : : * The replication slot for conflict detection must be created before the
1455 : : * worker starts.
1456 : : */
1457 [ - + ]: 63 : Assert(MyReplicationSlot);
1458 : :
1459 : 63 : SpinLockAcquire(&worker->relmutex);
1460 : 63 : nonremovable_xid = worker->oldest_nonremovable_xid;
1461 : 63 : SpinLockRelease(&worker->relmutex);
1462 : :
1463 : : /*
1464 : : * Return if the apply worker has stopped retention concurrently.
1465 : : *
1466 : : * Although this function is invoked only when retentionactive is true,
1467 : : * the apply worker might stop retention after the launcher fetches the
1468 : : * retentionactive flag.
1469 : : */
194 1470 [ - + ]: 63 : if (!TransactionIdIsValid(nonremovable_xid))
194 akapila@postgresql.o 1471 :UNC 0 : return;
1472 : :
235 akapila@postgresql.o 1473 [ - + - - ]:GNC 63 : if (!TransactionIdIsValid(*xmin) ||
235 akapila@postgresql.o 1474 :UNC 0 : TransactionIdPrecedes(nonremovable_xid, *xmin))
235 akapila@postgresql.o 1475 :GNC 63 : *xmin = nonremovable_xid;
1476 : : }
1477 : :
1478 : : /*
1479 : : * Acquire the replication slot used to retain information for conflict
1480 : : * detection, if it exists.
1481 : : *
1482 : : * Return true if successfully acquired, otherwise return false.
1483 : : */
1484 : : static bool
1485 : 459 : acquire_conflict_slot_if_exists(void)
1486 : : {
1487 [ + + ]: 459 : if (!SearchNamedReplicationSlot(CONFLICT_DETECTION_SLOT, true))
1488 : 458 : return false;
1489 : :
1490 : 1 : ReplicationSlotAcquire(CONFLICT_DETECTION_SLOT, true, false);
1491 : 1 : return true;
1492 : : }
1493 : :
1494 : : /*
1495 : : * Update the xmin the replication slot used to retain information required
1496 : : * for conflict detection.
1497 : : */
1498 : : static void
194 1499 : 63 : update_conflict_slot_xmin(TransactionId new_xmin)
1500 : : {
235 1501 [ - + ]: 63 : Assert(MyReplicationSlot);
194 1502 [ + - - + ]: 63 : Assert(!TransactionIdIsValid(new_xmin) ||
1503 : : TransactionIdPrecedesOrEquals(MyReplicationSlot->data.xmin, new_xmin));
1504 : :
1505 : : /* Return if the xmin value of the slot cannot be updated */
235 1506 [ + + ]: 63 : if (TransactionIdEquals(MyReplicationSlot->data.xmin, new_xmin))
1507 : 46 : return;
1508 : :
1509 : 17 : SpinLockAcquire(&MyReplicationSlot->mutex);
1510 : 17 : MyReplicationSlot->effective_xmin = new_xmin;
1511 : 17 : MyReplicationSlot->data.xmin = new_xmin;
1512 : 17 : SpinLockRelease(&MyReplicationSlot->mutex);
1513 : :
1514 [ - + ]: 17 : elog(DEBUG1, "updated xmin: %u", MyReplicationSlot->data.xmin);
1515 : :
1516 : 17 : ReplicationSlotMarkDirty();
1517 : 17 : ReplicationSlotsComputeRequiredXmin(false);
1518 : :
1519 : : /*
1520 : : * Like PhysicalConfirmReceivedLocation(), do not save slot information
1521 : : * each time. This is acceptable because all concurrent transactions on
1522 : : * the publisher that require the data preceding the slot's xmin should
1523 : : * have already been applied and flushed on the subscriber before the xmin
1524 : : * is advanced. So, even if the slot's xmin regresses after a restart, it
1525 : : * will be advanced again in the next cycle. Therefore, no data required
1526 : : * for conflict detection will be prematurely removed.
1527 : : */
1528 : 17 : return;
1529 : : }
1530 : :
1531 : : /*
1532 : : * Initialize the xmin for the conflict detection slot.
1533 : : */
1534 : : static void
194 1535 : 4 : init_conflict_slot_xmin(void)
1536 : : {
1537 : : TransactionId xmin_horizon;
1538 : :
1539 : : /* Replication slot must exist but shouldn't be initialized. */
1540 [ + - - + ]: 4 : Assert(MyReplicationSlot &&
1541 : : !TransactionIdIsValid(MyReplicationSlot->data.xmin));
1542 : :
75 msawada@postgresql.o 1543 : 4 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
235 akapila@postgresql.o 1544 : 4 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
1545 : :
1546 : 4 : xmin_horizon = GetOldestSafeDecodingTransactionId(false);
1547 : :
1548 : 4 : SpinLockAcquire(&MyReplicationSlot->mutex);
1549 : 4 : MyReplicationSlot->effective_xmin = xmin_horizon;
1550 : 4 : MyReplicationSlot->data.xmin = xmin_horizon;
1551 : 4 : SpinLockRelease(&MyReplicationSlot->mutex);
1552 : :
1553 : 4 : ReplicationSlotsComputeRequiredXmin(true);
1554 : :
1555 : 4 : LWLockRelease(ProcArrayLock);
75 msawada@postgresql.o 1556 : 4 : LWLockRelease(ReplicationSlotControlLock);
1557 : :
1558 : : /* Write this slot to disk */
235 akapila@postgresql.o 1559 : 4 : ReplicationSlotMarkDirty();
1560 : 4 : ReplicationSlotSave();
1561 : 4 : }
1562 : :
1563 : : /*
1564 : : * Create and acquire the replication slot used to retain information for
1565 : : * conflict detection, if not yet.
1566 : : */
1567 : : void
194 1568 : 68 : CreateConflictDetectionSlot(void)
1569 : : {
1570 : : /* Exit early, if the replication slot is already created and acquired */
1571 [ + + ]: 68 : if (MyReplicationSlot)
1572 : 64 : return;
1573 : :
1574 [ + - ]: 4 : ereport(LOG,
1575 : : errmsg("creating replication conflict detection slot"));
1576 : :
1577 : 4 : ReplicationSlotCreate(CONFLICT_DETECTION_SLOT, false, RS_PERSISTENT, false,
1578 : : false, false);
1579 : :
1580 : 4 : init_conflict_slot_xmin();
1581 : : }
1582 : :
1583 : : /*
1584 : : * Is current process the logical replication launcher?
1585 : : */
1586 : : bool
3202 andres@anarazel.de 1587 :CBC 2616 : IsLogicalLauncher(void)
1588 : : {
1589 : 2616 : return LogicalRepCtx->launcher_pid == MyProcPid;
1590 : : }
1591 : :
1592 : : /*
1593 : : * Return the pid of the leader apply worker if the given pid is the pid of a
1594 : : * parallel apply worker, otherwise, return InvalidPid.
1595 : : */
1596 : : pid_t
1152 akapila@postgresql.o 1597 : 986 : GetLeaderApplyWorkerPid(pid_t pid)
1598 : : {
1599 : 986 : int leader_pid = InvalidPid;
1600 : : int i;
1601 : :
1602 : 986 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
1603 : :
1604 [ + + ]: 4930 : for (i = 0; i < max_logical_replication_workers; i++)
1605 : : {
1606 : 3944 : LogicalRepWorker *w = &LogicalRepCtx->workers[i];
1607 : :
1608 [ + + - + : 3944 : if (isParallelApplyWorker(w) && w->proc && pid == w->proc->pid)
- - - - ]
1609 : : {
1152 akapila@postgresql.o 1610 :UBC 0 : leader_pid = w->leader_pid;
1611 : 0 : break;
1612 : : }
1613 : : }
1614 : :
1152 akapila@postgresql.o 1615 :CBC 986 : LWLockRelease(LogicalRepWorkerLock);
1616 : :
1617 : 986 : return leader_pid;
1618 : : }
1619 : :
1620 : : /*
1621 : : * Returns state of the subscriptions.
1622 : : */
1623 : : Datum
3342 peter_e@gmx.net 1624 : 1 : pg_stat_get_subscription(PG_FUNCTION_ARGS)
1625 : : {
1626 : : #define PG_STAT_GET_SUBSCRIPTION_COLS 10
1627 [ - + ]: 1 : Oid subid = PG_ARGISNULL(0) ? InvalidOid : PG_GETARG_OID(0);
1628 : : int i;
1629 : 1 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1630 : :
1244 michael@paquier.xyz 1631 : 1 : InitMaterializedSRF(fcinfo, 0);
1632 : :
1633 : : /* Make sure we get consistent view of the workers. */
3342 peter_e@gmx.net 1634 : 1 : LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
1635 : :
1377 tgl@sss.pgh.pa.us 1636 [ + + ]: 5 : for (i = 0; i < max_logical_replication_workers; i++)
1637 : : {
1638 : : /* for each row */
1338 peter@eisentraut.org 1639 : 4 : Datum values[PG_STAT_GET_SUBSCRIPTION_COLS] = {0};
1640 : 4 : bool nulls[PG_STAT_GET_SUBSCRIPTION_COLS] = {0};
1641 : : int worker_pid;
1642 : : LogicalRepWorker worker;
1643 : :
3342 peter_e@gmx.net 1644 : 4 : memcpy(&worker, &LogicalRepCtx->workers[i],
1645 : : sizeof(LogicalRepWorker));
1646 [ + + - + ]: 4 : if (!worker.proc || !IsBackendPid(worker.proc->pid))
1647 : 2 : continue;
1648 : :
1649 [ - + - - ]: 2 : if (OidIsValid(subid) && worker.subid != subid)
3342 peter_e@gmx.net 1650 :UBC 0 : continue;
1651 : :
3342 peter_e@gmx.net 1652 :CBC 2 : worker_pid = worker.proc->pid;
1653 : :
1654 : 2 : values[0] = ObjectIdGetDatum(worker.subid);
130 akapila@postgresql.o 1655 [ + - - + ]:GNC 2 : if (isTableSyncWorker(&worker))
3279 peter_e@gmx.net 1656 :UBC 0 : values[1] = ObjectIdGetDatum(worker.relid);
1657 : : else
3279 peter_e@gmx.net 1658 :CBC 2 : nulls[1] = true;
1659 : 2 : values[2] = Int32GetDatum(worker_pid);
1660 : :
1152 akapila@postgresql.o 1661 [ + - - + ]: 2 : if (isParallelApplyWorker(&worker))
1152 akapila@postgresql.o 1662 :UBC 0 : values[3] = Int32GetDatum(worker.leader_pid);
1663 : : else
3279 peter_e@gmx.net 1664 :CBC 2 : nulls[3] = true;
1665 : :
129 alvherre@kurilemu.de 1666 [ + + ]:GNC 2 : if (!XLogRecPtrIsValid(worker.last_lsn))
1152 akapila@postgresql.o 1667 :GBC 1 : nulls[4] = true;
1668 : : else
1152 akapila@postgresql.o 1669 :CBC 1 : values[4] = LSNGetDatum(worker.last_lsn);
3342 peter_e@gmx.net 1670 [ - + ]: 2 : if (worker.last_send_time == 0)
1152 akapila@postgresql.o 1671 :UBC 0 : nulls[5] = true;
1672 : : else
1152 akapila@postgresql.o 1673 :CBC 2 : values[5] = TimestampTzGetDatum(worker.last_send_time);
3342 peter_e@gmx.net 1674 [ - + ]: 2 : if (worker.last_recv_time == 0)
1152 akapila@postgresql.o 1675 :UBC 0 : nulls[6] = true;
1676 : : else
1152 akapila@postgresql.o 1677 :CBC 2 : values[6] = TimestampTzGetDatum(worker.last_recv_time);
129 alvherre@kurilemu.de 1678 [ + + ]:GNC 2 : if (!XLogRecPtrIsValid(worker.reply_lsn))
1152 akapila@postgresql.o 1679 :GBC 1 : nulls[7] = true;
1680 : : else
1152 akapila@postgresql.o 1681 :CBC 1 : values[7] = LSNGetDatum(worker.reply_lsn);
3342 peter_e@gmx.net 1682 [ - + ]: 2 : if (worker.reply_time == 0)
1152 akapila@postgresql.o 1683 :UBC 0 : nulls[8] = true;
1684 : : else
1152 akapila@postgresql.o 1685 :CBC 2 : values[8] = TimestampTzGetDatum(worker.reply_time);
1686 : :
902 nathan@postgresql.or 1687 [ + - - - : 2 : switch (worker.type)
- - ]
1688 : : {
1689 : 2 : case WORKERTYPE_APPLY:
1690 : 2 : values[9] = CStringGetTextDatum("apply");
1691 : 2 : break;
902 nathan@postgresql.or 1692 :UBC 0 : case WORKERTYPE_PARALLEL_APPLY:
1693 : 0 : values[9] = CStringGetTextDatum("parallel apply");
1694 : 0 : break;
130 akapila@postgresql.o 1695 :UNC 0 : case WORKERTYPE_SEQUENCESYNC:
1696 : 0 : values[9] = CStringGetTextDatum("sequence synchronization");
1697 : 0 : break;
902 nathan@postgresql.or 1698 :UBC 0 : case WORKERTYPE_TABLESYNC:
1699 : 0 : values[9] = CStringGetTextDatum("table synchronization");
1700 : 0 : break;
1701 : 0 : case WORKERTYPE_UNKNOWN:
1702 : : /* Should never happen. */
1703 [ # # ]: 0 : elog(ERROR, "unknown worker type");
1704 : : }
1705 : :
1469 michael@paquier.xyz 1706 :CBC 2 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
1707 : : values, nulls);
1708 : :
1709 : : /*
1710 : : * If only a single subscription was requested, and we found it,
1711 : : * break.
1712 : : */
3342 peter_e@gmx.net 1713 [ - + ]: 2 : if (OidIsValid(subid))
3342 peter_e@gmx.net 1714 :UBC 0 : break;
1715 : : }
1716 : :
3342 peter_e@gmx.net 1717 :CBC 1 : LWLockRelease(LogicalRepWorkerLock);
1718 : :
1719 : 1 : return (Datum) 0;
1720 : : }
|
