Line data Source code
1 : /*-------------------------------------------------------------------------
2 : * logical.c
3 : * PostgreSQL logical decoding coordination
4 : *
5 : * Copyright (c) 2012-2026, PostgreSQL Global Development Group
6 : *
7 : * IDENTIFICATION
8 : * src/backend/replication/logical/logical.c
9 : *
10 : * NOTES
11 : * This file coordinates interaction between the various modules that
12 : * together provide logical decoding, primarily by providing so
13 : * called LogicalDecodingContexts. The goal is to encapsulate most of the
14 : * internal complexity for consumers of logical decoding, so they can
15 : * create and consume a changestream with a low amount of code. Builtin
16 : * consumers are the walsender and SQL SRF interface, but it's possible to
17 : * add further ones without changing core code, e.g. to consume changes in
18 : * a bgworker.
19 : *
20 : * The idea is that a consumer provides three callbacks, one to read WAL,
21 : * one to prepare a data write, and a final one for actually writing since
22 : * their implementation depends on the type of consumer. Check
23 : * logicalfuncs.c for an example implementation of a fairly simple consumer
24 : * and an implementation of a WAL reading callback that's suitable for
25 : * simple consumers.
26 : *-------------------------------------------------------------------------
27 : */
28 :
29 : #include "postgres.h"
30 :
31 : #include "access/xact.h"
32 : #include "access/xlog_internal.h"
33 : #include "access/xlogutils.h"
34 : #include "fmgr.h"
35 : #include "miscadmin.h"
36 : #include "pgstat.h"
37 : #include "replication/decode.h"
38 : #include "replication/logical.h"
39 : #include "replication/reorderbuffer.h"
40 : #include "replication/slotsync.h"
41 : #include "replication/snapbuild.h"
42 : #include "storage/proc.h"
43 : #include "storage/procarray.h"
44 : #include "utils/builtins.h"
45 : #include "utils/injection_point.h"
46 : #include "utils/inval.h"
47 : #include "utils/memutils.h"
48 :
49 : /* data for errcontext callback */
50 : typedef struct LogicalErrorCallbackState
51 : {
52 : LogicalDecodingContext *ctx;
53 : const char *callback_name;
54 : XLogRecPtr report_location;
55 : } LogicalErrorCallbackState;
56 :
57 : /* wrappers around output plugin callbacks */
58 : static void output_plugin_error_callback(void *arg);
59 : static void startup_cb_wrapper(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
60 : bool is_init);
61 : static void shutdown_cb_wrapper(LogicalDecodingContext *ctx);
62 : static void begin_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn);
63 : static void commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
64 : XLogRecPtr commit_lsn);
65 : static void begin_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn);
66 : static void prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
67 : XLogRecPtr prepare_lsn);
68 : static void commit_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
69 : XLogRecPtr commit_lsn);
70 : static void rollback_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
71 : XLogRecPtr prepare_end_lsn, TimestampTz prepare_time);
72 : static void change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
73 : Relation relation, ReorderBufferChange *change);
74 : static void truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
75 : int nrelations, Relation relations[], ReorderBufferChange *change);
76 : static void message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
77 : XLogRecPtr message_lsn, bool transactional,
78 : const char *prefix, Size message_size, const char *message);
79 :
80 : /* streaming callbacks */
81 : static void stream_start_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
82 : XLogRecPtr first_lsn);
83 : static void stream_stop_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
84 : XLogRecPtr last_lsn);
85 : static void stream_abort_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
86 : XLogRecPtr abort_lsn);
87 : static void stream_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
88 : XLogRecPtr prepare_lsn);
89 : static void stream_commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
90 : XLogRecPtr commit_lsn);
91 : static void stream_change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
92 : Relation relation, ReorderBufferChange *change);
93 : static void stream_message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
94 : XLogRecPtr message_lsn, bool transactional,
95 : const char *prefix, Size message_size, const char *message);
96 : static void stream_truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
97 : int nrelations, Relation relations[], ReorderBufferChange *change);
98 :
99 : /* callback to update txn's progress */
100 : static void update_progress_txn_cb_wrapper(ReorderBuffer *cache,
101 : ReorderBufferTXN *txn,
102 : XLogRecPtr lsn);
103 :
104 : static void LoadOutputPlugin(OutputPluginCallbacks *callbacks, const char *plugin);
105 :
106 : /*
107 : * Make sure the current settings & environment are capable of doing logical
108 : * decoding.
109 : */
110 : void
111 1727 : CheckLogicalDecodingRequirements(bool repack)
112 : {
113 1727 : CheckSlotRequirements(repack);
114 :
115 : /*
116 : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
117 : * needs the same check.
118 : */
119 :
120 1727 : if (MyDatabaseId == InvalidOid)
121 1 : ereport(ERROR,
122 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
123 : errmsg("logical decoding requires a database connection")));
124 :
125 : /* CheckSlotRequirements() has already checked if wal_level >= 'replica' */
126 : Assert(wal_level >= WAL_LEVEL_REPLICA);
127 :
128 : /* Check if logical decoding is available on standby */
129 1726 : if (RecoveryInProgress() && !IsLogicalDecodingEnabled())
130 2 : ereport(ERROR,
131 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
132 : errmsg("logical decoding on standby requires \"effective_wal_level\" >= \"logical\" on the primary"),
133 : errhint("Set \"wal_level\" >= \"logical\" or create at least one logical slot when \"wal_level\" = \"replica\".")));
134 1724 : }
135 :
136 : /*
137 : * Helper function for CreateInitDecodingContext() and
138 : * CreateDecodingContext() performing common tasks.
139 : */
140 : static LogicalDecodingContext *
141 1222 : StartupDecodingContext(List *output_plugin_options,
142 : XLogRecPtr start_lsn,
143 : TransactionId xmin_horizon,
144 : bool need_full_snapshot,
145 : bool fast_forward,
146 : bool in_create,
147 : XLogReaderRoutine *xl_routine,
148 : LogicalOutputPluginWriterPrepareWrite prepare_write,
149 : LogicalOutputPluginWriterWrite do_write,
150 : LogicalOutputPluginWriterUpdateProgress update_progress)
151 : {
152 : ReplicationSlot *slot;
153 : MemoryContext context,
154 : old_context;
155 : LogicalDecodingContext *ctx;
156 :
157 : /* shorter lines... */
158 1222 : slot = MyReplicationSlot;
159 :
160 1222 : context = AllocSetContextCreate(CurrentMemoryContext,
161 : "Logical decoding context",
162 : ALLOCSET_DEFAULT_SIZES);
163 1222 : old_context = MemoryContextSwitchTo(context);
164 1222 : ctx = palloc0_object(LogicalDecodingContext);
165 :
166 1222 : ctx->context = context;
167 :
168 : /*
169 : * (re-)load output plugins, so we detect a bad (removed) output plugin
170 : * now.
171 : */
172 1222 : if (!fast_forward)
173 1197 : LoadOutputPlugin(&ctx->callbacks, NameStr(slot->data.plugin));
174 :
175 : /*
176 : * Now that the slot's xmin has been set, we can announce ourselves as a
177 : * logical decoding backend which doesn't need to be checked individually
178 : * when computing the xmin horizon because the xmin is enforced via
179 : * replication slots.
180 : *
181 : * We can only do so if we're outside of a transaction (i.e. the case when
182 : * streaming changes via walsender), otherwise an already setup
183 : * snapshot/xid would end up being ignored. That's not a particularly
184 : * bothersome restriction since the SQL interface can't be used for
185 : * streaming anyway.
186 : */
187 1221 : if (!IsTransactionOrTransactionBlock())
188 : {
189 626 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
190 626 : MyProc->statusFlags |= PROC_IN_LOGICAL_DECODING;
191 626 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
192 626 : LWLockRelease(ProcArrayLock);
193 : }
194 :
195 1221 : ctx->slot = slot;
196 :
197 1221 : ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
198 1221 : if (!ctx->reader)
199 0 : ereport(ERROR,
200 : (errcode(ERRCODE_OUT_OF_MEMORY),
201 : errmsg("out of memory"),
202 : errdetail("Failed while allocating a WAL reading processor.")));
203 :
204 1221 : ctx->reorder = ReorderBufferAllocate();
205 1221 : ctx->snapshot_builder =
206 1221 : AllocateSnapshotBuilder(ctx->reorder, xmin_horizon, start_lsn,
207 : need_full_snapshot, in_create, slot->data.two_phase_at);
208 :
209 1221 : ctx->reorder->private_data = ctx;
210 :
211 : /* wrap output plugin callbacks, so we can add error context information */
212 1221 : ctx->reorder->begin = begin_cb_wrapper;
213 1221 : ctx->reorder->apply_change = change_cb_wrapper;
214 1221 : ctx->reorder->apply_truncate = truncate_cb_wrapper;
215 1221 : ctx->reorder->commit = commit_cb_wrapper;
216 1221 : ctx->reorder->message = message_cb_wrapper;
217 :
218 : /*
219 : * To support streaming, we require start/stop/abort/commit/change
220 : * callbacks. The message and truncate callbacks are optional, similar to
221 : * regular output plugins. We however enable streaming when at least one
222 : * of the methods is enabled so that we can easily identify missing
223 : * methods.
224 : *
225 : * We decide it here, but only check it later in the wrappers.
226 : */
227 2473 : ctx->streaming = (ctx->callbacks.stream_start_cb != NULL) ||
228 31 : (ctx->callbacks.stream_stop_cb != NULL) ||
229 31 : (ctx->callbacks.stream_abort_cb != NULL) ||
230 31 : (ctx->callbacks.stream_commit_cb != NULL) ||
231 31 : (ctx->callbacks.stream_change_cb != NULL) ||
232 1283 : (ctx->callbacks.stream_message_cb != NULL) ||
233 31 : (ctx->callbacks.stream_truncate_cb != NULL);
234 :
235 : /*
236 : * streaming callbacks
237 : *
238 : * stream_message and stream_truncate callbacks are optional, so we do not
239 : * fail with ERROR when missing, but the wrappers simply do nothing. We
240 : * must set the ReorderBuffer callbacks to something, otherwise the calls
241 : * from there will crash (we don't want to move the checks there).
242 : */
243 1221 : ctx->reorder->stream_start = stream_start_cb_wrapper;
244 1221 : ctx->reorder->stream_stop = stream_stop_cb_wrapper;
245 1221 : ctx->reorder->stream_abort = stream_abort_cb_wrapper;
246 1221 : ctx->reorder->stream_prepare = stream_prepare_cb_wrapper;
247 1221 : ctx->reorder->stream_commit = stream_commit_cb_wrapper;
248 1221 : ctx->reorder->stream_change = stream_change_cb_wrapper;
249 1221 : ctx->reorder->stream_message = stream_message_cb_wrapper;
250 1221 : ctx->reorder->stream_truncate = stream_truncate_cb_wrapper;
251 :
252 :
253 : /*
254 : * To support two-phase logical decoding, we require
255 : * begin_prepare/prepare/commit-prepare/abort-prepare callbacks. The
256 : * filter_prepare callback is optional. We however enable two-phase
257 : * logical decoding when at least one of the methods is enabled so that we
258 : * can easily identify missing methods.
259 : *
260 : * We decide it here, but only check it later in the wrappers.
261 : */
262 2473 : ctx->twophase = (ctx->callbacks.begin_prepare_cb != NULL) ||
263 31 : (ctx->callbacks.prepare_cb != NULL) ||
264 31 : (ctx->callbacks.commit_prepared_cb != NULL) ||
265 31 : (ctx->callbacks.rollback_prepared_cb != NULL) ||
266 1283 : (ctx->callbacks.stream_prepare_cb != NULL) ||
267 31 : (ctx->callbacks.filter_prepare_cb != NULL);
268 :
269 : /*
270 : * Callback to support decoding at prepare time.
271 : */
272 1221 : ctx->reorder->begin_prepare = begin_prepare_cb_wrapper;
273 1221 : ctx->reorder->prepare = prepare_cb_wrapper;
274 1221 : ctx->reorder->commit_prepared = commit_prepared_cb_wrapper;
275 1221 : ctx->reorder->rollback_prepared = rollback_prepared_cb_wrapper;
276 :
277 : /*
278 : * Callback to support updating progress during sending data of a
279 : * transaction (and its subtransactions) to the output plugin.
280 : */
281 1221 : ctx->reorder->update_progress_txn = update_progress_txn_cb_wrapper;
282 :
283 1221 : ctx->out = makeStringInfo();
284 1221 : ctx->prepare_write = prepare_write;
285 1221 : ctx->write = do_write;
286 1221 : ctx->update_progress = update_progress;
287 :
288 1221 : ctx->output_plugin_options = output_plugin_options;
289 :
290 1221 : ctx->fast_forward = fast_forward;
291 :
292 1221 : MemoryContextSwitchTo(old_context);
293 :
294 1221 : return ctx;
295 : }
296 :
297 : /*
298 : * Create a new decoding context, for a new logical slot.
299 : *
300 : * plugin -- contains the name of the output plugin
301 : * output_plugin_options -- contains options passed to the output plugin
302 : * need_full_snapshot -- if true, must obtain a snapshot able to read all
303 : * tables; if false, one that can read only catalogs is acceptable.
304 : * for_repack -- if true, we're going to be decoding for REPACK.
305 : * restart_lsn -- if given as invalid, it's this routine's responsibility to
306 : * mark WAL as reserved by setting a convenient restart_lsn for the slot.
307 : * Otherwise, we set for decoding to start from the given LSN without
308 : * marking WAL reserved beforehand. In that scenario, it's up to the
309 : * caller to guarantee that WAL remains available.
310 : * xl_routine -- XLogReaderRoutine for underlying XLogReader
311 : * prepare_write, do_write, update_progress --
312 : * callbacks that perform the use-case dependent, actual, work.
313 : *
314 : * Needs to be called while in a memory context that's at least as long lived
315 : * as the decoding context because further memory contexts will be created
316 : * inside it.
317 : *
318 : * Returns an initialized decoding context after calling the output plugin's
319 : * startup function.
320 : */
321 : LogicalDecodingContext *
322 506 : CreateInitDecodingContext(const char *plugin,
323 : List *output_plugin_options,
324 : bool need_full_snapshot,
325 : bool for_repack,
326 : XLogRecPtr restart_lsn,
327 : XLogReaderRoutine *xl_routine,
328 : LogicalOutputPluginWriterPrepareWrite prepare_write,
329 : LogicalOutputPluginWriterWrite do_write,
330 : LogicalOutputPluginWriterUpdateProgress update_progress)
331 : {
332 506 : TransactionId xmin_horizon = InvalidTransactionId;
333 : ReplicationSlot *slot;
334 : NameData plugin_name;
335 : LogicalDecodingContext *ctx;
336 : MemoryContext old_context;
337 :
338 : /*
339 : * On a standby, this check is also required while creating the slot.
340 : * Check the comments in the function.
341 : */
342 506 : CheckLogicalDecodingRequirements(for_repack);
343 :
344 : /* shorter lines... */
345 506 : slot = MyReplicationSlot;
346 :
347 : /* first some sanity checks that are unlikely to be violated */
348 506 : if (slot == NULL)
349 0 : elog(ERROR, "cannot perform logical decoding without an acquired slot");
350 :
351 506 : if (plugin == NULL)
352 0 : elog(ERROR, "cannot initialize logical decoding without a specified plugin");
353 :
354 : /* Make sure the passed slot is suitable. These are user facing errors. */
355 506 : if (SlotIsPhysical(slot))
356 0 : ereport(ERROR,
357 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
358 : errmsg("cannot use physical replication slot for logical decoding")));
359 :
360 506 : if (slot->data.database != MyDatabaseId)
361 0 : ereport(ERROR,
362 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
363 : errmsg("replication slot \"%s\" was not created in this database",
364 : NameStr(slot->data.name))));
365 :
366 862 : if (IsTransactionState() &&
367 356 : GetTopTransactionIdIfAny() != InvalidTransactionId)
368 2 : ereport(ERROR,
369 : (errcode(ERRCODE_ACTIVE_SQL_TRANSACTION),
370 : errmsg("cannot create logical replication slot in transaction that has performed writes")));
371 :
372 : /*
373 : * Register output plugin name with slot. We need the mutex to avoid
374 : * concurrent reading of a partially copied string. But we don't want any
375 : * complicated code while holding a spinlock, so do namestrcpy() outside.
376 : */
377 504 : namestrcpy(&plugin_name, plugin);
378 504 : SpinLockAcquire(&slot->mutex);
379 504 : slot->data.plugin = plugin_name;
380 504 : SpinLockRelease(&slot->mutex);
381 :
382 504 : if (!XLogRecPtrIsValid(restart_lsn))
383 497 : ReplicationSlotReserveWal();
384 : else
385 : {
386 7 : SpinLockAcquire(&slot->mutex);
387 7 : slot->data.restart_lsn = restart_lsn;
388 7 : SpinLockRelease(&slot->mutex);
389 : }
390 :
391 : /* ----
392 : * This is a bit tricky: We need to determine a safe xmin horizon to start
393 : * decoding from, to avoid starting from a running xacts record referring
394 : * to xids whose rows have been vacuumed or pruned
395 : * already. GetOldestSafeDecodingTransactionId() returns such a value, but
396 : * without further interlock its return value might immediately be out of
397 : * date.
398 : *
399 : * So we have to acquire both the ReplicationSlotControlLock and the
400 : * ProcArrayLock to prevent concurrent computation and update of new xmin
401 : * horizons by other backends, get the safe decoding xid, and inform the
402 : * slot machinery about the new limit. Once that's done both locks can be
403 : * released as the slot machinery now is protecting against vacuum.
404 : *
405 : * Note that, temporarily, the data, not just the catalog, xmin has to be
406 : * reserved if a data snapshot is to be exported. Otherwise the initial
407 : * data snapshot created here is not guaranteed to be valid. After that
408 : * the data xmin doesn't need to be managed anymore and the global xmin
409 : * should be recomputed. As we are fine with losing the pegged data xmin
410 : * after crash - no chance a snapshot would get exported anymore - we can
411 : * get away with just setting the slot's
412 : * effective_xmin. ReplicationSlotRelease will reset it again.
413 : *
414 : * ----
415 : */
416 504 : LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
417 504 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
418 :
419 504 : xmin_horizon = GetOldestSafeDecodingTransactionId(!need_full_snapshot);
420 :
421 504 : SpinLockAcquire(&slot->mutex);
422 504 : slot->effective_catalog_xmin = xmin_horizon;
423 504 : slot->data.catalog_xmin = xmin_horizon;
424 504 : if (need_full_snapshot)
425 221 : slot->effective_xmin = xmin_horizon;
426 504 : SpinLockRelease(&slot->mutex);
427 :
428 504 : ReplicationSlotsComputeRequiredXmin(true);
429 :
430 504 : LWLockRelease(ProcArrayLock);
431 504 : LWLockRelease(ReplicationSlotControlLock);
432 :
433 504 : ReplicationSlotMarkDirty();
434 504 : ReplicationSlotSave();
435 :
436 504 : ctx = StartupDecodingContext(NIL, restart_lsn, xmin_horizon,
437 : need_full_snapshot, false, true,
438 : xl_routine, prepare_write, do_write,
439 : update_progress);
440 :
441 : /* call output plugin initialization callback */
442 503 : old_context = MemoryContextSwitchTo(ctx->context);
443 503 : if (ctx->callbacks.startup_cb != NULL)
444 503 : startup_cb_wrapper(ctx, &ctx->options, true);
445 503 : MemoryContextSwitchTo(old_context);
446 :
447 : /*
448 : * We allow decoding of prepared transactions when the two_phase is
449 : * enabled at the time of slot creation, or when the two_phase option is
450 : * given at the streaming start, provided the plugin supports all the
451 : * callbacks for two-phase.
452 : */
453 503 : ctx->twophase &= slot->data.two_phase;
454 :
455 503 : ctx->reorder->output_rewrites = ctx->options.receive_rewrites;
456 :
457 503 : return ctx;
458 : }
459 :
460 : /*
461 : * Create a new decoding context, for a logical slot that has previously been
462 : * used already.
463 : *
464 : * start_lsn
465 : * The LSN at which to start decoding. If InvalidXLogRecPtr, restart
466 : * from the slot's confirmed_flush; otherwise, start from the specified
467 : * location (but move it forwards to confirmed_flush if it's older than
468 : * that, see below).
469 : *
470 : * output_plugin_options
471 : * options passed to the output plugin.
472 : *
473 : * fast_forward
474 : * bypass the generation of logical changes.
475 : *
476 : * xl_routine
477 : * XLogReaderRoutine used by underlying xlogreader
478 : *
479 : * prepare_write, do_write, update_progress
480 : * callbacks that have to be filled to perform the use-case dependent,
481 : * actual work.
482 : *
483 : * Needs to be called while in a memory context that's at least as long lived
484 : * as the decoding context because further memory contexts will be created
485 : * inside it.
486 : *
487 : * Returns an initialized decoding context after calling the output plugin's
488 : * startup function.
489 : */
490 : LogicalDecodingContext *
491 723 : CreateDecodingContext(XLogRecPtr start_lsn,
492 : List *output_plugin_options,
493 : bool fast_forward,
494 : XLogReaderRoutine *xl_routine,
495 : LogicalOutputPluginWriterPrepareWrite prepare_write,
496 : LogicalOutputPluginWriterWrite do_write,
497 : LogicalOutputPluginWriterUpdateProgress update_progress)
498 : {
499 : LogicalDecodingContext *ctx;
500 : ReplicationSlot *slot;
501 : MemoryContext old_context;
502 :
503 : /* shorter lines... */
504 723 : slot = MyReplicationSlot;
505 :
506 : /* first some sanity checks that are unlikely to be violated */
507 723 : if (slot == NULL)
508 0 : elog(ERROR, "cannot perform logical decoding without an acquired slot");
509 :
510 : /* make sure the passed slot is suitable, these are user facing errors */
511 723 : if (SlotIsPhysical(slot))
512 1 : ereport(ERROR,
513 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
514 : errmsg("cannot use physical replication slot for logical decoding")));
515 :
516 : /*
517 : * We need to access the system tables during decoding to build the
518 : * logical changes unless we are in fast_forward mode where no changes are
519 : * generated.
520 : */
521 722 : if (slot->data.database != MyDatabaseId && !fast_forward)
522 3 : ereport(ERROR,
523 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
524 : errmsg("replication slot \"%s\" was not created in this database",
525 : NameStr(slot->data.name))));
526 :
527 : /*
528 : * The slots being synced from the primary can't be used for decoding as
529 : * they are used after failover. However, we do allow advancing the LSNs
530 : * during the synchronization of slots. See update_local_synced_slot.
531 : */
532 719 : if (RecoveryInProgress() && slot->data.synced && !IsSyncingReplicationSlots())
533 1 : ereport(ERROR,
534 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
535 : errmsg("cannot use replication slot \"%s\" for logical decoding",
536 : NameStr(slot->data.name)),
537 : errdetail("This replication slot is being synchronized from the primary server."),
538 : errhint("Specify another replication slot."));
539 :
540 : /* slot must be valid to allow decoding */
541 : Assert(slot->data.invalidated == RS_INVAL_NONE);
542 : Assert(XLogRecPtrIsValid(slot->data.restart_lsn));
543 :
544 718 : if (!XLogRecPtrIsValid(start_lsn))
545 : {
546 : /* continue from last position */
547 396 : start_lsn = slot->data.confirmed_flush;
548 : }
549 322 : else if (start_lsn < slot->data.confirmed_flush)
550 : {
551 : /*
552 : * It might seem like we should error out in this case, but it's
553 : * pretty common for a client to acknowledge a LSN it doesn't have to
554 : * do anything for, and thus didn't store persistently, because the
555 : * xlog records didn't result in anything relevant for logical
556 : * decoding. Clients have to be able to do that to support synchronous
557 : * replication.
558 : *
559 : * Starting at a different LSN than requested might not catch certain
560 : * kinds of client errors; so the client may wish to check that
561 : * confirmed_flush_lsn matches its expectations.
562 : */
563 71 : elog(LOG, "%X/%08X has been already streamed, forwarding to %X/%08X",
564 : LSN_FORMAT_ARGS(start_lsn),
565 : LSN_FORMAT_ARGS(slot->data.confirmed_flush));
566 :
567 71 : start_lsn = slot->data.confirmed_flush;
568 : }
569 :
570 718 : ctx = StartupDecodingContext(output_plugin_options,
571 : start_lsn, InvalidTransactionId, false,
572 : fast_forward, false, xl_routine, prepare_write,
573 : do_write, update_progress);
574 :
575 : /* call output plugin initialization callback */
576 718 : old_context = MemoryContextSwitchTo(ctx->context);
577 718 : if (ctx->callbacks.startup_cb != NULL)
578 693 : startup_cb_wrapper(ctx, &ctx->options, false);
579 715 : MemoryContextSwitchTo(old_context);
580 :
581 : /*
582 : * We allow decoding of prepared transactions when the two_phase is
583 : * enabled at the time of slot creation, or when the two_phase option is
584 : * given at the streaming start, provided the plugin supports all the
585 : * callbacks for two-phase.
586 : */
587 715 : ctx->twophase &= (slot->data.two_phase || ctx->twophase_opt_given);
588 :
589 : /* Mark slot to allow two_phase decoding if not already marked */
590 715 : if (ctx->twophase && !slot->data.two_phase)
591 : {
592 8 : SpinLockAcquire(&slot->mutex);
593 8 : slot->data.two_phase = true;
594 8 : slot->data.two_phase_at = start_lsn;
595 8 : SpinLockRelease(&slot->mutex);
596 8 : ReplicationSlotMarkDirty();
597 8 : ReplicationSlotSave();
598 8 : SnapBuildSetTwoPhaseAt(ctx->snapshot_builder, start_lsn);
599 : }
600 :
601 715 : ctx->reorder->output_rewrites = ctx->options.receive_rewrites;
602 :
603 715 : ereport(LogicalDecodingLogLevel(),
604 : (errmsg("starting logical decoding for slot \"%s\"",
605 : NameStr(slot->data.name)),
606 : errdetail("Streaming transactions committing after %X/%08X, reading WAL from %X/%08X.",
607 : LSN_FORMAT_ARGS(slot->data.confirmed_flush),
608 : LSN_FORMAT_ARGS(slot->data.restart_lsn))));
609 :
610 715 : return ctx;
611 : }
612 :
613 : /*
614 : * Returns true if a consistent initial decoding snapshot has been built.
615 : */
616 : bool
617 537 : DecodingContextReady(LogicalDecodingContext *ctx)
618 : {
619 537 : return SnapBuildCurrentState(ctx->snapshot_builder) == SNAPBUILD_CONSISTENT;
620 : }
621 :
622 : /*
623 : * Read from the decoding slot, until it is ready to start extracting changes.
624 : */
625 : void
626 496 : DecodingContextFindStartpoint(LogicalDecodingContext *ctx)
627 : {
628 496 : ReplicationSlot *slot = ctx->slot;
629 :
630 : /* Initialize from where to start reading WAL. */
631 496 : XLogBeginRead(ctx->reader, slot->data.restart_lsn);
632 :
633 496 : elog(DEBUG1, "searching for logical decoding starting point, starting at %X/%08X",
634 : LSN_FORMAT_ARGS(slot->data.restart_lsn));
635 :
636 : /* Wait for a consistent starting point */
637 : for (;;)
638 32 : {
639 : XLogRecord *record;
640 528 : char *err = NULL;
641 :
642 : /* the read_page callback waits for new WAL */
643 528 : record = XLogReadRecord(ctx->reader, &err);
644 528 : if (err)
645 0 : elog(ERROR, "could not find logical decoding starting point: %s", err);
646 528 : if (!record)
647 0 : elog(ERROR, "could not find logical decoding starting point");
648 :
649 528 : LogicalDecodingProcessRecord(ctx, ctx->reader);
650 :
651 : /* only continue till we found a consistent spot */
652 526 : if (DecodingContextReady(ctx))
653 494 : break;
654 :
655 32 : CHECK_FOR_INTERRUPTS();
656 : }
657 :
658 494 : SpinLockAcquire(&slot->mutex);
659 494 : slot->data.confirmed_flush = ctx->reader->EndRecPtr;
660 494 : if (slot->data.two_phase)
661 9 : slot->data.two_phase_at = ctx->reader->EndRecPtr;
662 494 : SpinLockRelease(&slot->mutex);
663 494 : }
664 :
665 : /*
666 : * Free a previously allocated decoding context, invoking the shutdown
667 : * callback if necessary.
668 : */
669 : void
670 948 : FreeDecodingContext(LogicalDecodingContext *ctx)
671 : {
672 948 : if (ctx->callbacks.shutdown_cb != NULL)
673 923 : shutdown_cb_wrapper(ctx);
674 :
675 948 : ReorderBufferFree(ctx->reorder);
676 948 : FreeSnapshotBuilder(ctx->snapshot_builder);
677 948 : XLogReaderFree(ctx->reader);
678 948 : MemoryContextDelete(ctx->context);
679 948 : }
680 :
681 : /*
682 : * Prepare a write using the context's output routine.
683 : */
684 : void
685 357292 : OutputPluginPrepareWrite(struct LogicalDecodingContext *ctx, bool last_write)
686 : {
687 357292 : if (!ctx->accept_writes)
688 0 : elog(ERROR, "writes are only accepted in commit, begin and change callbacks");
689 :
690 357292 : ctx->prepare_write(ctx, ctx->write_location, ctx->write_xid, last_write);
691 357292 : ctx->prepared_write = true;
692 357292 : }
693 :
694 : /*
695 : * Perform a write using the context's output routine.
696 : */
697 : void
698 357292 : OutputPluginWrite(struct LogicalDecodingContext *ctx, bool last_write)
699 : {
700 357292 : if (!ctx->prepared_write)
701 0 : elog(ERROR, "OutputPluginPrepareWrite needs to be called before OutputPluginWrite");
702 :
703 357292 : ctx->write(ctx, ctx->write_location, ctx->write_xid, last_write);
704 357278 : ctx->prepared_write = false;
705 357278 : }
706 :
707 : /*
708 : * Update progress tracking (if supported).
709 : */
710 : void
711 4777 : OutputPluginUpdateProgress(struct LogicalDecodingContext *ctx,
712 : bool skipped_xact)
713 : {
714 4777 : if (!ctx->update_progress)
715 1594 : return;
716 :
717 3183 : ctx->update_progress(ctx, ctx->write_location, ctx->write_xid,
718 : skipped_xact);
719 : }
720 :
721 : /*
722 : * Load the output plugin, lookup its output plugin init function, and check
723 : * that it provides the required callbacks.
724 : */
725 : static void
726 1197 : LoadOutputPlugin(OutputPluginCallbacks *callbacks, const char *plugin)
727 : {
728 : LogicalOutputPluginInit plugin_init;
729 :
730 1196 : plugin_init = (LogicalOutputPluginInit)
731 1197 : load_external_function(plugin, "_PG_output_plugin_init", false, NULL);
732 :
733 1196 : if (plugin_init == NULL)
734 0 : elog(ERROR, "output plugins have to declare the _PG_output_plugin_init symbol");
735 :
736 : /* ask the output plugin to fill the callback struct */
737 1196 : plugin_init(callbacks);
738 :
739 1196 : if (callbacks->begin_cb == NULL)
740 0 : elog(ERROR, "output plugins have to register a begin callback");
741 1196 : if (callbacks->change_cb == NULL)
742 0 : elog(ERROR, "output plugins have to register a change callback");
743 1196 : if (callbacks->commit_cb == NULL)
744 0 : elog(ERROR, "output plugins have to register a commit callback");
745 1196 : }
746 :
747 : static void
748 500 : output_plugin_error_callback(void *arg)
749 : {
750 500 : LogicalErrorCallbackState *state = (LogicalErrorCallbackState *) arg;
751 :
752 : /* not all callbacks have an associated LSN */
753 500 : if (XLogRecPtrIsValid(state->report_location))
754 497 : errcontext("slot \"%s\", output plugin \"%s\", in the %s callback, associated LSN %X/%08X",
755 497 : NameStr(state->ctx->slot->data.name),
756 497 : NameStr(state->ctx->slot->data.plugin),
757 : state->callback_name,
758 497 : LSN_FORMAT_ARGS(state->report_location));
759 : else
760 3 : errcontext("slot \"%s\", output plugin \"%s\", in the %s callback",
761 3 : NameStr(state->ctx->slot->data.name),
762 3 : NameStr(state->ctx->slot->data.plugin),
763 : state->callback_name);
764 500 : }
765 :
766 : static void
767 1196 : startup_cb_wrapper(LogicalDecodingContext *ctx, OutputPluginOptions *opt, bool is_init)
768 : {
769 : LogicalErrorCallbackState state;
770 : ErrorContextCallback errcallback;
771 :
772 : Assert(!ctx->fast_forward);
773 :
774 : /* Push callback + info on the error context stack */
775 1196 : state.ctx = ctx;
776 1196 : state.callback_name = "startup";
777 1196 : state.report_location = InvalidXLogRecPtr;
778 1196 : errcallback.callback = output_plugin_error_callback;
779 1196 : errcallback.arg = &state;
780 1196 : errcallback.previous = error_context_stack;
781 1196 : error_context_stack = &errcallback;
782 :
783 : /* set output state */
784 1196 : ctx->accept_writes = false;
785 1196 : ctx->end_xact = false;
786 :
787 : /* do the actual work: call callback */
788 1196 : ctx->callbacks.startup_cb(ctx, opt, is_init);
789 :
790 : /* Pop the error context stack */
791 1193 : error_context_stack = errcallback.previous;
792 1193 : }
793 :
794 : static void
795 923 : shutdown_cb_wrapper(LogicalDecodingContext *ctx)
796 : {
797 : LogicalErrorCallbackState state;
798 : ErrorContextCallback errcallback;
799 :
800 : Assert(!ctx->fast_forward);
801 :
802 : /* Push callback + info on the error context stack */
803 923 : state.ctx = ctx;
804 923 : state.callback_name = "shutdown";
805 923 : state.report_location = InvalidXLogRecPtr;
806 923 : errcallback.callback = output_plugin_error_callback;
807 923 : errcallback.arg = &state;
808 923 : errcallback.previous = error_context_stack;
809 923 : error_context_stack = &errcallback;
810 :
811 : /* set output state */
812 923 : ctx->accept_writes = false;
813 923 : ctx->end_xact = false;
814 :
815 : /* do the actual work: call callback */
816 923 : ctx->callbacks.shutdown_cb(ctx);
817 :
818 : /* Pop the error context stack */
819 923 : error_context_stack = errcallback.previous;
820 923 : }
821 :
822 :
823 : /*
824 : * Callbacks for ReorderBuffer which add in some more information and then call
825 : * output_plugin.h plugins.
826 : */
827 : static void
828 1828 : begin_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
829 : {
830 1828 : LogicalDecodingContext *ctx = cache->private_data;
831 : LogicalErrorCallbackState state;
832 : ErrorContextCallback errcallback;
833 :
834 : Assert(!ctx->fast_forward);
835 :
836 : /* Push callback + info on the error context stack */
837 1828 : state.ctx = ctx;
838 1828 : state.callback_name = "begin";
839 1828 : state.report_location = txn->first_lsn;
840 1828 : errcallback.callback = output_plugin_error_callback;
841 1828 : errcallback.arg = &state;
842 1828 : errcallback.previous = error_context_stack;
843 1828 : error_context_stack = &errcallback;
844 :
845 : /* set output state */
846 1828 : ctx->accept_writes = true;
847 1828 : ctx->write_xid = txn->xid;
848 1828 : ctx->write_location = txn->first_lsn;
849 1828 : ctx->end_xact = false;
850 :
851 : /* do the actual work: call callback */
852 1828 : ctx->callbacks.begin_cb(ctx, txn);
853 :
854 : /* Pop the error context stack */
855 1828 : error_context_stack = errcallback.previous;
856 1828 : }
857 :
858 : static void
859 1823 : commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
860 : XLogRecPtr commit_lsn)
861 : {
862 1823 : LogicalDecodingContext *ctx = cache->private_data;
863 : LogicalErrorCallbackState state;
864 : ErrorContextCallback errcallback;
865 :
866 : Assert(!ctx->fast_forward);
867 :
868 : /* Push callback + info on the error context stack */
869 1823 : state.ctx = ctx;
870 1823 : state.callback_name = "commit";
871 1823 : state.report_location = txn->final_lsn; /* beginning of commit record */
872 1823 : errcallback.callback = output_plugin_error_callback;
873 1823 : errcallback.arg = &state;
874 1823 : errcallback.previous = error_context_stack;
875 1823 : error_context_stack = &errcallback;
876 :
877 : /* set output state */
878 1823 : ctx->accept_writes = true;
879 1823 : ctx->write_xid = txn->xid;
880 1823 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
881 1823 : ctx->end_xact = true;
882 :
883 : /* do the actual work: call callback */
884 1823 : ctx->callbacks.commit_cb(ctx, txn, commit_lsn);
885 :
886 : /* Pop the error context stack */
887 1812 : error_context_stack = errcallback.previous;
888 1812 : }
889 :
890 : /*
891 : * The functionality of begin_prepare is quite similar to begin with the
892 : * exception that this will have gid (global transaction id) information which
893 : * can be used by plugin. Now, we thought about extending the existing begin
894 : * but that would break the replication protocol and additionally this looks
895 : * cleaner.
896 : */
897 : static void
898 30 : begin_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
899 : {
900 30 : LogicalDecodingContext *ctx = cache->private_data;
901 : LogicalErrorCallbackState state;
902 : ErrorContextCallback errcallback;
903 :
904 : Assert(!ctx->fast_forward);
905 :
906 : /* We're only supposed to call this when two-phase commits are supported */
907 : Assert(ctx->twophase);
908 :
909 : /* Push callback + info on the error context stack */
910 30 : state.ctx = ctx;
911 30 : state.callback_name = "begin_prepare";
912 30 : state.report_location = txn->first_lsn;
913 30 : errcallback.callback = output_plugin_error_callback;
914 30 : errcallback.arg = &state;
915 30 : errcallback.previous = error_context_stack;
916 30 : error_context_stack = &errcallback;
917 :
918 : /* set output state */
919 30 : ctx->accept_writes = true;
920 30 : ctx->write_xid = txn->xid;
921 30 : ctx->write_location = txn->first_lsn;
922 30 : ctx->end_xact = false;
923 :
924 : /*
925 : * If the plugin supports two-phase commits then begin prepare callback is
926 : * mandatory
927 : */
928 30 : if (ctx->callbacks.begin_prepare_cb == NULL)
929 0 : ereport(ERROR,
930 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
931 : errmsg("logical replication at prepare time requires a %s callback",
932 : "begin_prepare_cb")));
933 :
934 : /* do the actual work: call callback */
935 30 : ctx->callbacks.begin_prepare_cb(ctx, txn);
936 :
937 : /* Pop the error context stack */
938 30 : error_context_stack = errcallback.previous;
939 30 : }
940 :
941 : static void
942 30 : prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
943 : XLogRecPtr prepare_lsn)
944 : {
945 30 : LogicalDecodingContext *ctx = cache->private_data;
946 : LogicalErrorCallbackState state;
947 : ErrorContextCallback errcallback;
948 :
949 : Assert(!ctx->fast_forward);
950 :
951 : /* We're only supposed to call this when two-phase commits are supported */
952 : Assert(ctx->twophase);
953 :
954 : /* Push callback + info on the error context stack */
955 30 : state.ctx = ctx;
956 30 : state.callback_name = "prepare";
957 30 : state.report_location = txn->final_lsn; /* beginning of prepare record */
958 30 : errcallback.callback = output_plugin_error_callback;
959 30 : errcallback.arg = &state;
960 30 : errcallback.previous = error_context_stack;
961 30 : error_context_stack = &errcallback;
962 :
963 : /* set output state */
964 30 : ctx->accept_writes = true;
965 30 : ctx->write_xid = txn->xid;
966 30 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
967 30 : ctx->end_xact = true;
968 :
969 : /*
970 : * If the plugin supports two-phase commits then prepare callback is
971 : * mandatory
972 : */
973 30 : if (ctx->callbacks.prepare_cb == NULL)
974 0 : ereport(ERROR,
975 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
976 : errmsg("logical replication at prepare time requires a %s callback",
977 : "prepare_cb")));
978 :
979 : /* do the actual work: call callback */
980 30 : ctx->callbacks.prepare_cb(ctx, txn, prepare_lsn);
981 :
982 : /* Pop the error context stack */
983 30 : error_context_stack = errcallback.previous;
984 30 : }
985 :
986 : static void
987 33 : commit_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
988 : XLogRecPtr commit_lsn)
989 : {
990 33 : LogicalDecodingContext *ctx = cache->private_data;
991 : LogicalErrorCallbackState state;
992 : ErrorContextCallback errcallback;
993 :
994 : Assert(!ctx->fast_forward);
995 :
996 : /* We're only supposed to call this when two-phase commits are supported */
997 : Assert(ctx->twophase);
998 :
999 : /* Push callback + info on the error context stack */
1000 33 : state.ctx = ctx;
1001 33 : state.callback_name = "commit_prepared";
1002 33 : state.report_location = txn->final_lsn; /* beginning of commit record */
1003 33 : errcallback.callback = output_plugin_error_callback;
1004 33 : errcallback.arg = &state;
1005 33 : errcallback.previous = error_context_stack;
1006 33 : error_context_stack = &errcallback;
1007 :
1008 : /* set output state */
1009 33 : ctx->accept_writes = true;
1010 33 : ctx->write_xid = txn->xid;
1011 33 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
1012 33 : ctx->end_xact = true;
1013 :
1014 : /*
1015 : * If the plugin support two-phase commits then commit prepared callback
1016 : * is mandatory
1017 : */
1018 33 : if (ctx->callbacks.commit_prepared_cb == NULL)
1019 0 : ereport(ERROR,
1020 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1021 : errmsg("logical replication at prepare time requires a %s callback",
1022 : "commit_prepared_cb")));
1023 :
1024 : /* do the actual work: call callback */
1025 33 : ctx->callbacks.commit_prepared_cb(ctx, txn, commit_lsn);
1026 :
1027 : /* Pop the error context stack */
1028 33 : error_context_stack = errcallback.previous;
1029 33 : }
1030 :
1031 : static void
1032 10 : rollback_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1033 : XLogRecPtr prepare_end_lsn,
1034 : TimestampTz prepare_time)
1035 : {
1036 10 : LogicalDecodingContext *ctx = cache->private_data;
1037 : LogicalErrorCallbackState state;
1038 : ErrorContextCallback errcallback;
1039 :
1040 : Assert(!ctx->fast_forward);
1041 :
1042 : /* We're only supposed to call this when two-phase commits are supported */
1043 : Assert(ctx->twophase);
1044 :
1045 : /* Push callback + info on the error context stack */
1046 10 : state.ctx = ctx;
1047 10 : state.callback_name = "rollback_prepared";
1048 10 : state.report_location = txn->final_lsn; /* beginning of commit record */
1049 10 : errcallback.callback = output_plugin_error_callback;
1050 10 : errcallback.arg = &state;
1051 10 : errcallback.previous = error_context_stack;
1052 10 : error_context_stack = &errcallback;
1053 :
1054 : /* set output state */
1055 10 : ctx->accept_writes = true;
1056 10 : ctx->write_xid = txn->xid;
1057 10 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
1058 10 : ctx->end_xact = true;
1059 :
1060 : /*
1061 : * If the plugin support two-phase commits then rollback prepared callback
1062 : * is mandatory
1063 : */
1064 10 : if (ctx->callbacks.rollback_prepared_cb == NULL)
1065 0 : ereport(ERROR,
1066 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1067 : errmsg("logical replication at prepare time requires a %s callback",
1068 : "rollback_prepared_cb")));
1069 :
1070 : /* do the actual work: call callback */
1071 10 : ctx->callbacks.rollback_prepared_cb(ctx, txn, prepare_end_lsn,
1072 : prepare_time);
1073 :
1074 : /* Pop the error context stack */
1075 10 : error_context_stack = errcallback.previous;
1076 10 : }
1077 :
1078 : static void
1079 179522 : change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1080 : Relation relation, ReorderBufferChange *change)
1081 : {
1082 179522 : LogicalDecodingContext *ctx = cache->private_data;
1083 : LogicalErrorCallbackState state;
1084 : ErrorContextCallback errcallback;
1085 :
1086 : Assert(!ctx->fast_forward);
1087 :
1088 : /* Push callback + info on the error context stack */
1089 179522 : state.ctx = ctx;
1090 179522 : state.callback_name = "change";
1091 179522 : state.report_location = change->lsn;
1092 179522 : errcallback.callback = output_plugin_error_callback;
1093 179522 : errcallback.arg = &state;
1094 179522 : errcallback.previous = error_context_stack;
1095 179522 : error_context_stack = &errcallback;
1096 :
1097 : /* set output state */
1098 179522 : ctx->accept_writes = true;
1099 179522 : ctx->write_xid = txn->xid;
1100 :
1101 : /*
1102 : * Report this change's lsn so replies from clients can give an up-to-date
1103 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1104 : * receipt of this transaction, but it might allow another transaction's
1105 : * commit to be confirmed with one message.
1106 : */
1107 179522 : ctx->write_location = change->lsn;
1108 :
1109 179522 : ctx->end_xact = false;
1110 :
1111 179522 : ctx->callbacks.change_cb(ctx, txn, relation, change);
1112 :
1113 : /* Pop the error context stack */
1114 179517 : error_context_stack = errcallback.previous;
1115 179517 : }
1116 :
1117 : static void
1118 33 : truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1119 : int nrelations, Relation relations[], ReorderBufferChange *change)
1120 : {
1121 33 : LogicalDecodingContext *ctx = cache->private_data;
1122 : LogicalErrorCallbackState state;
1123 : ErrorContextCallback errcallback;
1124 :
1125 : Assert(!ctx->fast_forward);
1126 :
1127 33 : if (!ctx->callbacks.truncate_cb)
1128 0 : return;
1129 :
1130 : /* Push callback + info on the error context stack */
1131 33 : state.ctx = ctx;
1132 33 : state.callback_name = "truncate";
1133 33 : state.report_location = change->lsn;
1134 33 : errcallback.callback = output_plugin_error_callback;
1135 33 : errcallback.arg = &state;
1136 33 : errcallback.previous = error_context_stack;
1137 33 : error_context_stack = &errcallback;
1138 :
1139 : /* set output state */
1140 33 : ctx->accept_writes = true;
1141 33 : ctx->write_xid = txn->xid;
1142 :
1143 : /*
1144 : * Report this change's lsn so replies from clients can give an up-to-date
1145 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1146 : * receipt of this transaction, but it might allow another transaction's
1147 : * commit to be confirmed with one message.
1148 : */
1149 33 : ctx->write_location = change->lsn;
1150 :
1151 33 : ctx->end_xact = false;
1152 :
1153 33 : ctx->callbacks.truncate_cb(ctx, txn, nrelations, relations, change);
1154 :
1155 : /* Pop the error context stack */
1156 33 : error_context_stack = errcallback.previous;
1157 : }
1158 :
1159 : bool
1160 148 : filter_prepare_cb_wrapper(LogicalDecodingContext *ctx, TransactionId xid,
1161 : const char *gid)
1162 : {
1163 : LogicalErrorCallbackState state;
1164 : ErrorContextCallback errcallback;
1165 : bool ret;
1166 :
1167 : Assert(!ctx->fast_forward);
1168 :
1169 : /* Push callback + info on the error context stack */
1170 148 : state.ctx = ctx;
1171 148 : state.callback_name = "filter_prepare";
1172 148 : state.report_location = InvalidXLogRecPtr;
1173 148 : errcallback.callback = output_plugin_error_callback;
1174 148 : errcallback.arg = &state;
1175 148 : errcallback.previous = error_context_stack;
1176 148 : error_context_stack = &errcallback;
1177 :
1178 : /* set output state */
1179 148 : ctx->accept_writes = false;
1180 148 : ctx->end_xact = false;
1181 :
1182 : /* do the actual work: call callback */
1183 148 : ret = ctx->callbacks.filter_prepare_cb(ctx, xid, gid);
1184 :
1185 : /* Pop the error context stack */
1186 148 : error_context_stack = errcallback.previous;
1187 :
1188 148 : return ret;
1189 : }
1190 :
1191 : bool
1192 1433167 : filter_by_origin_cb_wrapper(LogicalDecodingContext *ctx, ReplOriginId origin_id)
1193 : {
1194 : LogicalErrorCallbackState state;
1195 : ErrorContextCallback errcallback;
1196 : bool ret;
1197 :
1198 : Assert(!ctx->fast_forward);
1199 :
1200 : /* Push callback + info on the error context stack */
1201 1433167 : state.ctx = ctx;
1202 1433167 : state.callback_name = "filter_by_origin";
1203 1433167 : state.report_location = InvalidXLogRecPtr;
1204 1433167 : errcallback.callback = output_plugin_error_callback;
1205 1433167 : errcallback.arg = &state;
1206 1433167 : errcallback.previous = error_context_stack;
1207 1433167 : error_context_stack = &errcallback;
1208 :
1209 : /* set output state */
1210 1433167 : ctx->accept_writes = false;
1211 1433167 : ctx->end_xact = false;
1212 :
1213 : /* do the actual work: call callback */
1214 1433167 : ret = ctx->callbacks.filter_by_origin_cb(ctx, origin_id);
1215 :
1216 : /* Pop the error context stack */
1217 1433167 : error_context_stack = errcallback.previous;
1218 :
1219 1433167 : return ret;
1220 : }
1221 :
1222 : static void
1223 16 : message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1224 : XLogRecPtr message_lsn, bool transactional,
1225 : const char *prefix, Size message_size, const char *message)
1226 : {
1227 16 : LogicalDecodingContext *ctx = cache->private_data;
1228 : LogicalErrorCallbackState state;
1229 : ErrorContextCallback errcallback;
1230 :
1231 : Assert(!ctx->fast_forward);
1232 :
1233 16 : if (ctx->callbacks.message_cb == NULL)
1234 0 : return;
1235 :
1236 : /* Push callback + info on the error context stack */
1237 16 : state.ctx = ctx;
1238 16 : state.callback_name = "message";
1239 16 : state.report_location = message_lsn;
1240 16 : errcallback.callback = output_plugin_error_callback;
1241 16 : errcallback.arg = &state;
1242 16 : errcallback.previous = error_context_stack;
1243 16 : error_context_stack = &errcallback;
1244 :
1245 : /* set output state */
1246 16 : ctx->accept_writes = true;
1247 16 : ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
1248 16 : ctx->write_location = message_lsn;
1249 16 : ctx->end_xact = false;
1250 :
1251 : /* do the actual work: call callback */
1252 16 : ctx->callbacks.message_cb(ctx, txn, message_lsn, transactional, prefix,
1253 : message_size, message);
1254 :
1255 : /* Pop the error context stack */
1256 16 : error_context_stack = errcallback.previous;
1257 : }
1258 :
1259 : static void
1260 669 : stream_start_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1261 : XLogRecPtr first_lsn)
1262 : {
1263 669 : LogicalDecodingContext *ctx = cache->private_data;
1264 : LogicalErrorCallbackState state;
1265 : ErrorContextCallback errcallback;
1266 :
1267 : Assert(!ctx->fast_forward);
1268 :
1269 : /* We're only supposed to call this when streaming is supported. */
1270 : Assert(ctx->streaming);
1271 :
1272 : /* Push callback + info on the error context stack */
1273 669 : state.ctx = ctx;
1274 669 : state.callback_name = "stream_start";
1275 669 : state.report_location = first_lsn;
1276 669 : errcallback.callback = output_plugin_error_callback;
1277 669 : errcallback.arg = &state;
1278 669 : errcallback.previous = error_context_stack;
1279 669 : error_context_stack = &errcallback;
1280 :
1281 : /* set output state */
1282 669 : ctx->accept_writes = true;
1283 669 : ctx->write_xid = txn->xid;
1284 :
1285 : /*
1286 : * Report this message's lsn so replies from clients can give an
1287 : * up-to-date answer. This won't ever be enough (and shouldn't be!) to
1288 : * confirm receipt of this transaction, but it might allow another
1289 : * transaction's commit to be confirmed with one message.
1290 : */
1291 669 : ctx->write_location = first_lsn;
1292 :
1293 669 : ctx->end_xact = false;
1294 :
1295 : /* in streaming mode, stream_start_cb is required */
1296 669 : if (ctx->callbacks.stream_start_cb == NULL)
1297 0 : ereport(ERROR,
1298 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1299 : errmsg("logical streaming requires a %s callback",
1300 : "stream_start_cb")));
1301 :
1302 669 : ctx->callbacks.stream_start_cb(ctx, txn);
1303 :
1304 : /* Pop the error context stack */
1305 669 : error_context_stack = errcallback.previous;
1306 669 : }
1307 :
1308 : static void
1309 669 : stream_stop_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1310 : XLogRecPtr last_lsn)
1311 : {
1312 669 : LogicalDecodingContext *ctx = cache->private_data;
1313 : LogicalErrorCallbackState state;
1314 : ErrorContextCallback errcallback;
1315 :
1316 : Assert(!ctx->fast_forward);
1317 :
1318 : /* We're only supposed to call this when streaming is supported. */
1319 : Assert(ctx->streaming);
1320 :
1321 : /* Push callback + info on the error context stack */
1322 669 : state.ctx = ctx;
1323 669 : state.callback_name = "stream_stop";
1324 669 : state.report_location = last_lsn;
1325 669 : errcallback.callback = output_plugin_error_callback;
1326 669 : errcallback.arg = &state;
1327 669 : errcallback.previous = error_context_stack;
1328 669 : error_context_stack = &errcallback;
1329 :
1330 : /* set output state */
1331 669 : ctx->accept_writes = true;
1332 669 : ctx->write_xid = txn->xid;
1333 :
1334 : /*
1335 : * Report this message's lsn so replies from clients can give an
1336 : * up-to-date answer. This won't ever be enough (and shouldn't be!) to
1337 : * confirm receipt of this transaction, but it might allow another
1338 : * transaction's commit to be confirmed with one message.
1339 : */
1340 669 : ctx->write_location = last_lsn;
1341 :
1342 669 : ctx->end_xact = false;
1343 :
1344 : /* in streaming mode, stream_stop_cb is required */
1345 669 : if (ctx->callbacks.stream_stop_cb == NULL)
1346 0 : ereport(ERROR,
1347 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1348 : errmsg("logical streaming requires a %s callback",
1349 : "stream_stop_cb")));
1350 :
1351 669 : ctx->callbacks.stream_stop_cb(ctx, txn);
1352 :
1353 : /* Pop the error context stack */
1354 669 : error_context_stack = errcallback.previous;
1355 669 : }
1356 :
1357 : static void
1358 30 : stream_abort_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1359 : XLogRecPtr abort_lsn)
1360 : {
1361 30 : LogicalDecodingContext *ctx = cache->private_data;
1362 : LogicalErrorCallbackState state;
1363 : ErrorContextCallback errcallback;
1364 :
1365 : Assert(!ctx->fast_forward);
1366 :
1367 : /* We're only supposed to call this when streaming is supported. */
1368 : Assert(ctx->streaming);
1369 :
1370 : /* Push callback + info on the error context stack */
1371 30 : state.ctx = ctx;
1372 30 : state.callback_name = "stream_abort";
1373 30 : state.report_location = abort_lsn;
1374 30 : errcallback.callback = output_plugin_error_callback;
1375 30 : errcallback.arg = &state;
1376 30 : errcallback.previous = error_context_stack;
1377 30 : error_context_stack = &errcallback;
1378 :
1379 : /* set output state */
1380 30 : ctx->accept_writes = true;
1381 30 : ctx->write_xid = txn->xid;
1382 30 : ctx->write_location = abort_lsn;
1383 30 : ctx->end_xact = true;
1384 :
1385 : /* in streaming mode, stream_abort_cb is required */
1386 30 : if (ctx->callbacks.stream_abort_cb == NULL)
1387 0 : ereport(ERROR,
1388 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1389 : errmsg("logical streaming requires a %s callback",
1390 : "stream_abort_cb")));
1391 :
1392 30 : ctx->callbacks.stream_abort_cb(ctx, txn, abort_lsn);
1393 :
1394 : /* Pop the error context stack */
1395 30 : error_context_stack = errcallback.previous;
1396 30 : }
1397 :
1398 : static void
1399 14 : stream_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1400 : XLogRecPtr prepare_lsn)
1401 : {
1402 14 : LogicalDecodingContext *ctx = cache->private_data;
1403 : LogicalErrorCallbackState state;
1404 : ErrorContextCallback errcallback;
1405 :
1406 : Assert(!ctx->fast_forward);
1407 :
1408 : /*
1409 : * We're only supposed to call this when streaming and two-phase commits
1410 : * are supported.
1411 : */
1412 : Assert(ctx->streaming);
1413 : Assert(ctx->twophase);
1414 :
1415 : /* Push callback + info on the error context stack */
1416 14 : state.ctx = ctx;
1417 14 : state.callback_name = "stream_prepare";
1418 14 : state.report_location = txn->final_lsn;
1419 14 : errcallback.callback = output_plugin_error_callback;
1420 14 : errcallback.arg = &state;
1421 14 : errcallback.previous = error_context_stack;
1422 14 : error_context_stack = &errcallback;
1423 :
1424 : /* set output state */
1425 14 : ctx->accept_writes = true;
1426 14 : ctx->write_xid = txn->xid;
1427 14 : ctx->write_location = txn->end_lsn;
1428 14 : ctx->end_xact = true;
1429 :
1430 : /* in streaming mode with two-phase commits, stream_prepare_cb is required */
1431 14 : if (ctx->callbacks.stream_prepare_cb == NULL)
1432 0 : ereport(ERROR,
1433 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1434 : errmsg("logical streaming at prepare time requires a %s callback",
1435 : "stream_prepare_cb")));
1436 :
1437 14 : ctx->callbacks.stream_prepare_cb(ctx, txn, prepare_lsn);
1438 :
1439 : /* Pop the error context stack */
1440 14 : error_context_stack = errcallback.previous;
1441 14 : }
1442 :
1443 : static void
1444 50 : stream_commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1445 : XLogRecPtr commit_lsn)
1446 : {
1447 50 : LogicalDecodingContext *ctx = cache->private_data;
1448 : LogicalErrorCallbackState state;
1449 : ErrorContextCallback errcallback;
1450 :
1451 : Assert(!ctx->fast_forward);
1452 :
1453 : /* We're only supposed to call this when streaming is supported. */
1454 : Assert(ctx->streaming);
1455 :
1456 : /* Push callback + info on the error context stack */
1457 50 : state.ctx = ctx;
1458 50 : state.callback_name = "stream_commit";
1459 50 : state.report_location = txn->final_lsn;
1460 50 : errcallback.callback = output_plugin_error_callback;
1461 50 : errcallback.arg = &state;
1462 50 : errcallback.previous = error_context_stack;
1463 50 : error_context_stack = &errcallback;
1464 :
1465 : /* set output state */
1466 50 : ctx->accept_writes = true;
1467 50 : ctx->write_xid = txn->xid;
1468 50 : ctx->write_location = txn->end_lsn;
1469 50 : ctx->end_xact = true;
1470 :
1471 : /* in streaming mode, stream_commit_cb is required */
1472 50 : if (ctx->callbacks.stream_commit_cb == NULL)
1473 0 : ereport(ERROR,
1474 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1475 : errmsg("logical streaming requires a %s callback",
1476 : "stream_commit_cb")));
1477 :
1478 50 : ctx->callbacks.stream_commit_cb(ctx, txn, commit_lsn);
1479 :
1480 : /* Pop the error context stack */
1481 50 : error_context_stack = errcallback.previous;
1482 50 : }
1483 :
1484 : static void
1485 175988 : stream_change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1486 : Relation relation, ReorderBufferChange *change)
1487 : {
1488 175988 : LogicalDecodingContext *ctx = cache->private_data;
1489 : LogicalErrorCallbackState state;
1490 : ErrorContextCallback errcallback;
1491 :
1492 : Assert(!ctx->fast_forward);
1493 :
1494 : /* We're only supposed to call this when streaming is supported. */
1495 : Assert(ctx->streaming);
1496 :
1497 : /* Push callback + info on the error context stack */
1498 175988 : state.ctx = ctx;
1499 175988 : state.callback_name = "stream_change";
1500 175988 : state.report_location = change->lsn;
1501 175988 : errcallback.callback = output_plugin_error_callback;
1502 175988 : errcallback.arg = &state;
1503 175988 : errcallback.previous = error_context_stack;
1504 175988 : error_context_stack = &errcallback;
1505 :
1506 : /* set output state */
1507 175988 : ctx->accept_writes = true;
1508 175988 : ctx->write_xid = txn->xid;
1509 :
1510 : /*
1511 : * Report this change's lsn so replies from clients can give an up-to-date
1512 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1513 : * receipt of this transaction, but it might allow another transaction's
1514 : * commit to be confirmed with one message.
1515 : */
1516 175988 : ctx->write_location = change->lsn;
1517 :
1518 175988 : ctx->end_xact = false;
1519 :
1520 : /* in streaming mode, stream_change_cb is required */
1521 175988 : if (ctx->callbacks.stream_change_cb == NULL)
1522 0 : ereport(ERROR,
1523 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1524 : errmsg("logical streaming requires a %s callback",
1525 : "stream_change_cb")));
1526 :
1527 175988 : ctx->callbacks.stream_change_cb(ctx, txn, relation, change);
1528 :
1529 : /* Pop the error context stack */
1530 175988 : error_context_stack = errcallback.previous;
1531 175988 : }
1532 :
1533 : static void
1534 3 : stream_message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1535 : XLogRecPtr message_lsn, bool transactional,
1536 : const char *prefix, Size message_size, const char *message)
1537 : {
1538 3 : LogicalDecodingContext *ctx = cache->private_data;
1539 : LogicalErrorCallbackState state;
1540 : ErrorContextCallback errcallback;
1541 :
1542 : Assert(!ctx->fast_forward);
1543 :
1544 : /* We're only supposed to call this when streaming is supported. */
1545 : Assert(ctx->streaming);
1546 :
1547 : /* this callback is optional */
1548 3 : if (ctx->callbacks.stream_message_cb == NULL)
1549 0 : return;
1550 :
1551 : /* Push callback + info on the error context stack */
1552 3 : state.ctx = ctx;
1553 3 : state.callback_name = "stream_message";
1554 3 : state.report_location = message_lsn;
1555 3 : errcallback.callback = output_plugin_error_callback;
1556 3 : errcallback.arg = &state;
1557 3 : errcallback.previous = error_context_stack;
1558 3 : error_context_stack = &errcallback;
1559 :
1560 : /* set output state */
1561 3 : ctx->accept_writes = true;
1562 3 : ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
1563 3 : ctx->write_location = message_lsn;
1564 3 : ctx->end_xact = false;
1565 :
1566 : /* do the actual work: call callback */
1567 3 : ctx->callbacks.stream_message_cb(ctx, txn, message_lsn, transactional, prefix,
1568 : message_size, message);
1569 :
1570 : /* Pop the error context stack */
1571 3 : error_context_stack = errcallback.previous;
1572 : }
1573 :
1574 : static void
1575 0 : stream_truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1576 : int nrelations, Relation relations[],
1577 : ReorderBufferChange *change)
1578 : {
1579 0 : LogicalDecodingContext *ctx = cache->private_data;
1580 : LogicalErrorCallbackState state;
1581 : ErrorContextCallback errcallback;
1582 :
1583 : Assert(!ctx->fast_forward);
1584 :
1585 : /* We're only supposed to call this when streaming is supported. */
1586 : Assert(ctx->streaming);
1587 :
1588 : /* this callback is optional */
1589 0 : if (!ctx->callbacks.stream_truncate_cb)
1590 0 : return;
1591 :
1592 : /* Push callback + info on the error context stack */
1593 0 : state.ctx = ctx;
1594 0 : state.callback_name = "stream_truncate";
1595 0 : state.report_location = change->lsn;
1596 0 : errcallback.callback = output_plugin_error_callback;
1597 0 : errcallback.arg = &state;
1598 0 : errcallback.previous = error_context_stack;
1599 0 : error_context_stack = &errcallback;
1600 :
1601 : /* set output state */
1602 0 : ctx->accept_writes = true;
1603 0 : ctx->write_xid = txn->xid;
1604 :
1605 : /*
1606 : * Report this change's lsn so replies from clients can give an up-to-date
1607 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1608 : * receipt of this transaction, but it might allow another transaction's
1609 : * commit to be confirmed with one message.
1610 : */
1611 0 : ctx->write_location = change->lsn;
1612 :
1613 0 : ctx->end_xact = false;
1614 :
1615 0 : ctx->callbacks.stream_truncate_cb(ctx, txn, nrelations, relations, change);
1616 :
1617 : /* Pop the error context stack */
1618 0 : error_context_stack = errcallback.previous;
1619 : }
1620 :
1621 : static void
1622 3307 : update_progress_txn_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1623 : XLogRecPtr lsn)
1624 : {
1625 3307 : LogicalDecodingContext *ctx = cache->private_data;
1626 : LogicalErrorCallbackState state;
1627 : ErrorContextCallback errcallback;
1628 :
1629 : Assert(!ctx->fast_forward);
1630 :
1631 : /* Push callback + info on the error context stack */
1632 3307 : state.ctx = ctx;
1633 3307 : state.callback_name = "update_progress_txn";
1634 3307 : state.report_location = lsn;
1635 3307 : errcallback.callback = output_plugin_error_callback;
1636 3307 : errcallback.arg = &state;
1637 3307 : errcallback.previous = error_context_stack;
1638 3307 : error_context_stack = &errcallback;
1639 :
1640 : /* set output state */
1641 3307 : ctx->accept_writes = false;
1642 3307 : ctx->write_xid = txn->xid;
1643 :
1644 : /*
1645 : * Report this change's lsn so replies from clients can give an up-to-date
1646 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1647 : * receipt of this transaction, but it might allow another transaction's
1648 : * commit to be confirmed with one message.
1649 : */
1650 3307 : ctx->write_location = lsn;
1651 :
1652 3307 : ctx->end_xact = false;
1653 :
1654 3307 : OutputPluginUpdateProgress(ctx, false);
1655 :
1656 : /* Pop the error context stack */
1657 3307 : error_context_stack = errcallback.previous;
1658 3307 : }
1659 :
1660 : /*
1661 : * Set the required catalog xmin horizon for historic snapshots in the current
1662 : * replication slot.
1663 : *
1664 : * Note that in the most cases, we won't be able to immediately use the xmin
1665 : * to increase the xmin horizon: we need to wait till the client has confirmed
1666 : * receiving current_lsn with LogicalConfirmReceivedLocation().
1667 : */
1668 : void
1669 545 : LogicalIncreaseXminForSlot(XLogRecPtr current_lsn, TransactionId xmin)
1670 : {
1671 545 : bool updated_xmin = false;
1672 : ReplicationSlot *slot;
1673 545 : bool got_new_xmin = false;
1674 :
1675 545 : slot = MyReplicationSlot;
1676 :
1677 : Assert(slot != NULL);
1678 :
1679 545 : SpinLockAcquire(&slot->mutex);
1680 :
1681 : /*
1682 : * don't overwrite if we already have a newer xmin. This can happen if we
1683 : * restart decoding in a slot.
1684 : */
1685 545 : if (TransactionIdPrecedesOrEquals(xmin, slot->data.catalog_xmin))
1686 : {
1687 : }
1688 :
1689 : /*
1690 : * If the client has already confirmed up to this lsn, we directly can
1691 : * mark this as accepted. This can happen if we restart decoding in a
1692 : * slot.
1693 : */
1694 191 : else if (current_lsn <= slot->data.confirmed_flush)
1695 : {
1696 58 : slot->candidate_catalog_xmin = xmin;
1697 58 : slot->candidate_xmin_lsn = current_lsn;
1698 :
1699 : /* our candidate can directly be used */
1700 58 : updated_xmin = true;
1701 : }
1702 :
1703 : /*
1704 : * Only increase if the previous values have been applied, otherwise we
1705 : * might never end up updating if the receiver acks too slowly.
1706 : */
1707 133 : else if (!XLogRecPtrIsValid(slot->candidate_xmin_lsn))
1708 : {
1709 25 : slot->candidate_catalog_xmin = xmin;
1710 25 : slot->candidate_xmin_lsn = current_lsn;
1711 :
1712 : /*
1713 : * Log new xmin at an appropriate log level after releasing the
1714 : * spinlock.
1715 : */
1716 25 : got_new_xmin = true;
1717 : }
1718 545 : SpinLockRelease(&slot->mutex);
1719 :
1720 545 : if (got_new_xmin)
1721 25 : elog(DEBUG1, "got new catalog xmin %u at %X/%08X", xmin,
1722 : LSN_FORMAT_ARGS(current_lsn));
1723 :
1724 : /* candidate already valid with the current flush position, apply */
1725 545 : if (updated_xmin)
1726 58 : LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
1727 545 : }
1728 :
1729 : /*
1730 : * Mark the minimal LSN (restart_lsn) we need to read to replay all
1731 : * transactions that have not yet committed at current_lsn.
1732 : *
1733 : * Just like LogicalIncreaseXminForSlot this only takes effect when the
1734 : * client has confirmed to have received current_lsn.
1735 : */
1736 : void
1737 496 : LogicalIncreaseRestartDecodingForSlot(XLogRecPtr current_lsn, XLogRecPtr restart_lsn)
1738 : {
1739 496 : bool updated_lsn = false;
1740 : ReplicationSlot *slot;
1741 :
1742 496 : slot = MyReplicationSlot;
1743 :
1744 : Assert(slot != NULL);
1745 : Assert(XLogRecPtrIsValid(restart_lsn));
1746 : Assert(XLogRecPtrIsValid(current_lsn));
1747 :
1748 496 : SpinLockAcquire(&slot->mutex);
1749 :
1750 : /* don't overwrite if have a newer restart lsn */
1751 496 : if (restart_lsn <= slot->data.restart_lsn)
1752 : {
1753 10 : SpinLockRelease(&slot->mutex);
1754 : }
1755 :
1756 : /*
1757 : * We might have already flushed far enough to directly accept this lsn,
1758 : * in this case there is no need to check for existing candidate LSNs
1759 : */
1760 486 : else if (current_lsn <= slot->data.confirmed_flush)
1761 : {
1762 323 : slot->candidate_restart_valid = current_lsn;
1763 323 : slot->candidate_restart_lsn = restart_lsn;
1764 323 : SpinLockRelease(&slot->mutex);
1765 :
1766 : /* our candidate can directly be used */
1767 323 : updated_lsn = true;
1768 : }
1769 :
1770 : /*
1771 : * Only increase if the previous values have been applied, otherwise we
1772 : * might never end up updating if the receiver acks too slowly. A missed
1773 : * value here will just cause some extra effort after reconnecting.
1774 : */
1775 163 : else if (!XLogRecPtrIsValid(slot->candidate_restart_valid))
1776 : {
1777 47 : slot->candidate_restart_valid = current_lsn;
1778 47 : slot->candidate_restart_lsn = restart_lsn;
1779 47 : SpinLockRelease(&slot->mutex);
1780 :
1781 47 : elog(DEBUG1, "got new restart lsn %X/%08X at %X/%08X",
1782 : LSN_FORMAT_ARGS(restart_lsn),
1783 : LSN_FORMAT_ARGS(current_lsn));
1784 : }
1785 : else
1786 : {
1787 : XLogRecPtr candidate_restart_lsn;
1788 : XLogRecPtr candidate_restart_valid;
1789 : XLogRecPtr confirmed_flush;
1790 :
1791 116 : candidate_restart_lsn = slot->candidate_restart_lsn;
1792 116 : candidate_restart_valid = slot->candidate_restart_valid;
1793 116 : confirmed_flush = slot->data.confirmed_flush;
1794 116 : SpinLockRelease(&slot->mutex);
1795 :
1796 116 : elog(DEBUG1, "failed to increase restart lsn: proposed %X/%08X, after %X/%08X, current candidate %X/%08X, current after %X/%08X, flushed up to %X/%08X",
1797 : LSN_FORMAT_ARGS(restart_lsn),
1798 : LSN_FORMAT_ARGS(current_lsn),
1799 : LSN_FORMAT_ARGS(candidate_restart_lsn),
1800 : LSN_FORMAT_ARGS(candidate_restart_valid),
1801 : LSN_FORMAT_ARGS(confirmed_flush));
1802 : }
1803 :
1804 : /* candidates are already valid with the current flush position, apply */
1805 496 : if (updated_lsn)
1806 323 : LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
1807 496 : }
1808 :
1809 : /*
1810 : * Handle a consumer's confirmation having received all changes up to lsn.
1811 : */
1812 : void
1813 18019 : LogicalConfirmReceivedLocation(XLogRecPtr lsn)
1814 : {
1815 : Assert(XLogRecPtrIsValid(lsn));
1816 :
1817 : /* Do an unlocked check for candidate_lsn first. */
1818 18019 : if (XLogRecPtrIsValid(MyReplicationSlot->candidate_xmin_lsn) ||
1819 17872 : XLogRecPtrIsValid(MyReplicationSlot->candidate_restart_valid))
1820 493 : {
1821 493 : bool updated_xmin = false;
1822 493 : bool updated_restart = false;
1823 : XLogRecPtr restart_lsn pg_attribute_unused();
1824 :
1825 493 : SpinLockAcquire(&MyReplicationSlot->mutex);
1826 :
1827 : /* remember the old restart lsn */
1828 493 : restart_lsn = MyReplicationSlot->data.restart_lsn;
1829 :
1830 : /*
1831 : * Prevent moving the confirmed_flush backwards, as this could lead to
1832 : * data duplication issues caused by replicating already replicated
1833 : * changes.
1834 : *
1835 : * This can happen when a client acknowledges an LSN it doesn't have
1836 : * to do anything for, and thus didn't store persistently. After a
1837 : * restart, the client can send the prior LSN that it stored
1838 : * persistently as an acknowledgement, but we need to ignore such an
1839 : * LSN. See similar case handling in CreateDecodingContext.
1840 : */
1841 493 : if (lsn > MyReplicationSlot->data.confirmed_flush)
1842 42 : MyReplicationSlot->data.confirmed_flush = lsn;
1843 :
1844 : /* if we're past the location required for bumping xmin, do so */
1845 493 : if (XLogRecPtrIsValid(MyReplicationSlot->candidate_xmin_lsn) &&
1846 147 : MyReplicationSlot->candidate_xmin_lsn <= lsn)
1847 : {
1848 : /*
1849 : * We have to write the changed xmin to disk *before* we change
1850 : * the in-memory value, otherwise after a crash we wouldn't know
1851 : * that some catalog tuples might have been removed already.
1852 : *
1853 : * Ensure that by first writing to ->xmin and only update
1854 : * ->effective_xmin once the new state is synced to disk. After a
1855 : * crash ->effective_xmin is set to ->xmin.
1856 : */
1857 77 : if (TransactionIdIsValid(MyReplicationSlot->candidate_catalog_xmin) &&
1858 77 : MyReplicationSlot->data.catalog_xmin != MyReplicationSlot->candidate_catalog_xmin)
1859 : {
1860 77 : MyReplicationSlot->data.catalog_xmin = MyReplicationSlot->candidate_catalog_xmin;
1861 77 : MyReplicationSlot->candidate_catalog_xmin = InvalidTransactionId;
1862 77 : MyReplicationSlot->candidate_xmin_lsn = InvalidXLogRecPtr;
1863 77 : updated_xmin = true;
1864 : }
1865 : }
1866 :
1867 493 : if (XLogRecPtrIsValid(MyReplicationSlot->candidate_restart_valid) &&
1868 435 : MyReplicationSlot->candidate_restart_valid <= lsn)
1869 : {
1870 : Assert(XLogRecPtrIsValid(MyReplicationSlot->candidate_restart_lsn));
1871 :
1872 364 : MyReplicationSlot->data.restart_lsn = MyReplicationSlot->candidate_restart_lsn;
1873 364 : MyReplicationSlot->candidate_restart_lsn = InvalidXLogRecPtr;
1874 364 : MyReplicationSlot->candidate_restart_valid = InvalidXLogRecPtr;
1875 364 : updated_restart = true;
1876 : }
1877 :
1878 493 : SpinLockRelease(&MyReplicationSlot->mutex);
1879 :
1880 : /* first write new xmin to disk, so we know what's up after a crash */
1881 493 : if (updated_xmin || updated_restart)
1882 : {
1883 : #ifdef USE_INJECTION_POINTS
1884 : XLogSegNo seg1,
1885 : seg2;
1886 :
1887 422 : XLByteToSeg(restart_lsn, seg1, wal_segment_size);
1888 422 : XLByteToSeg(MyReplicationSlot->data.restart_lsn, seg2, wal_segment_size);
1889 :
1890 : /* trigger injection point, but only if segment changes */
1891 422 : if (seg1 != seg2)
1892 18 : INJECTION_POINT("logical-replication-slot-advance-segment", NULL);
1893 : #endif
1894 :
1895 422 : ReplicationSlotMarkDirty();
1896 422 : ReplicationSlotSave();
1897 422 : elog(DEBUG1, "updated xmin: %u restart: %u", updated_xmin, updated_restart);
1898 : }
1899 :
1900 : /*
1901 : * Now the new xmin is safely on disk, we can let the global value
1902 : * advance. We do not take ProcArrayLock or similar since we only
1903 : * advance xmin here and there's not much harm done by a concurrent
1904 : * computation missing that.
1905 : */
1906 493 : if (updated_xmin)
1907 : {
1908 77 : SpinLockAcquire(&MyReplicationSlot->mutex);
1909 77 : MyReplicationSlot->effective_catalog_xmin = MyReplicationSlot->data.catalog_xmin;
1910 77 : SpinLockRelease(&MyReplicationSlot->mutex);
1911 :
1912 77 : ReplicationSlotsComputeRequiredXmin(false);
1913 77 : ReplicationSlotsComputeRequiredLSN();
1914 : }
1915 : }
1916 : else
1917 : {
1918 17526 : SpinLockAcquire(&MyReplicationSlot->mutex);
1919 :
1920 : /*
1921 : * Prevent moving the confirmed_flush backwards. See comments above
1922 : * for the details.
1923 : */
1924 17526 : if (lsn > MyReplicationSlot->data.confirmed_flush)
1925 16284 : MyReplicationSlot->data.confirmed_flush = lsn;
1926 :
1927 17526 : SpinLockRelease(&MyReplicationSlot->mutex);
1928 : }
1929 18019 : }
1930 :
1931 : /*
1932 : * Clear logical streaming state during (sub)transaction abort.
1933 : */
1934 : void
1935 40971 : ResetLogicalStreamingState(void)
1936 : {
1937 40971 : CheckXidAlive = InvalidTransactionId;
1938 40971 : bsysscan = false;
1939 40971 : }
1940 :
1941 : /*
1942 : * Report stats for a slot.
1943 : */
1944 : void
1945 6589 : UpdateDecodingStats(LogicalDecodingContext *ctx)
1946 : {
1947 6589 : ReorderBuffer *rb = ctx->reorder;
1948 : PgStat_StatReplSlotEntry repSlotStat;
1949 :
1950 : /* Nothing to do if we don't have any replication stats to be sent. */
1951 6589 : if (rb->spillBytes <= 0 && rb->streamBytes <= 0 && rb->totalBytes <= 0 &&
1952 471 : rb->memExceededCount <= 0)
1953 462 : return;
1954 :
1955 6127 : elog(DEBUG2, "UpdateDecodingStats: updating stats %p %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64,
1956 : rb,
1957 : rb->spillTxns,
1958 : rb->spillCount,
1959 : rb->spillBytes,
1960 : rb->streamTxns,
1961 : rb->streamCount,
1962 : rb->streamBytes,
1963 : rb->memExceededCount,
1964 : rb->totalTxns,
1965 : rb->totalBytes);
1966 :
1967 6127 : repSlotStat.spill_txns = rb->spillTxns;
1968 6127 : repSlotStat.spill_count = rb->spillCount;
1969 6127 : repSlotStat.spill_bytes = rb->spillBytes;
1970 6127 : repSlotStat.stream_txns = rb->streamTxns;
1971 6127 : repSlotStat.stream_count = rb->streamCount;
1972 6127 : repSlotStat.stream_bytes = rb->streamBytes;
1973 6127 : repSlotStat.mem_exceeded_count = rb->memExceededCount;
1974 6127 : repSlotStat.total_txns = rb->totalTxns;
1975 6127 : repSlotStat.total_bytes = rb->totalBytes;
1976 :
1977 6127 : pgstat_report_replslot(ctx->slot, &repSlotStat);
1978 :
1979 6127 : rb->spillTxns = 0;
1980 6127 : rb->spillCount = 0;
1981 6127 : rb->spillBytes = 0;
1982 6127 : rb->streamTxns = 0;
1983 6127 : rb->streamCount = 0;
1984 6127 : rb->streamBytes = 0;
1985 6127 : rb->memExceededCount = 0;
1986 6127 : rb->totalTxns = 0;
1987 6127 : rb->totalBytes = 0;
1988 : }
1989 :
1990 : /*
1991 : * Read up to the end of WAL starting from the decoding slot's restart_lsn
1992 : * to end_of_wal in order to check if any meaningful/decodable WAL records
1993 : * are encountered. scan_cutoff_lsn is the LSN, where we can terminate the
1994 : * WAL scan early if we find a decodable WAL record after this LSN.
1995 : *
1996 : * Returns the last LSN decodable WAL record's LSN if found, otherwise
1997 : * returns InvalidXLogRecPtr.
1998 : */
1999 : XLogRecPtr
2000 3 : LogicalReplicationSlotCheckPendingWal(XLogRecPtr end_of_wal,
2001 : XLogRecPtr scan_cutoff_lsn)
2002 : {
2003 3 : XLogRecPtr last_pending_wal = InvalidXLogRecPtr;
2004 :
2005 : Assert(MyReplicationSlot);
2006 : Assert(end_of_wal >= scan_cutoff_lsn);
2007 :
2008 3 : PG_TRY();
2009 : {
2010 : LogicalDecodingContext *ctx;
2011 :
2012 : /*
2013 : * Create our decoding context in fast_forward mode, passing start_lsn
2014 : * as InvalidXLogRecPtr, so that we start processing from the slot's
2015 : * confirmed_flush.
2016 : */
2017 6 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
2018 : NIL,
2019 : true, /* fast_forward */
2020 3 : XL_ROUTINE(.page_read = read_local_xlog_page,
2021 : .segment_open = wal_segment_open,
2022 : .segment_close = wal_segment_close),
2023 : NULL, NULL, NULL);
2024 :
2025 : /*
2026 : * Start reading at the slot's restart_lsn, which we know points to a
2027 : * valid record.
2028 : */
2029 3 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
2030 :
2031 : /* Invalidate non-timetravel entries */
2032 3 : InvalidateSystemCaches();
2033 :
2034 81 : while (ctx->reader->EndRecPtr < end_of_wal)
2035 : {
2036 : XLogRecord *record;
2037 78 : char *errm = NULL;
2038 :
2039 78 : record = XLogReadRecord(ctx->reader, &errm);
2040 :
2041 78 : if (errm)
2042 0 : elog(ERROR, "could not find record for logical decoding: %s", errm);
2043 :
2044 78 : if (record != NULL)
2045 78 : LogicalDecodingProcessRecord(ctx, ctx->reader);
2046 :
2047 78 : if (ctx->processing_required)
2048 : {
2049 2 : last_pending_wal = ctx->reader->ReadRecPtr;
2050 :
2051 : /*
2052 : * If we find a decodable WAL after the scan_cutoff_lsn point,
2053 : * we can terminate the scan early.
2054 : */
2055 2 : if (last_pending_wal >= scan_cutoff_lsn)
2056 0 : break;
2057 :
2058 : /* Reset the flag and continue checking */
2059 2 : ctx->processing_required = false;
2060 : }
2061 :
2062 78 : CHECK_FOR_INTERRUPTS();
2063 : }
2064 :
2065 : /* Clean up */
2066 3 : FreeDecodingContext(ctx);
2067 3 : InvalidateSystemCaches();
2068 : }
2069 0 : PG_CATCH();
2070 : {
2071 : /* clear all timetravel entries */
2072 0 : InvalidateSystemCaches();
2073 :
2074 0 : PG_RE_THROW();
2075 : }
2076 3 : PG_END_TRY();
2077 :
2078 3 : return last_pending_wal;
2079 : }
2080 :
2081 : /*
2082 : * Helper function for advancing our logical replication slot forward.
2083 : *
2084 : * The slot's restart_lsn is used as start point for reading records, while
2085 : * confirmed_flush is used as base point for the decoding context.
2086 : *
2087 : * We cannot just do LogicalConfirmReceivedLocation to update confirmed_flush,
2088 : * because we need to digest WAL to advance restart_lsn allowing to recycle
2089 : * WAL and removal of old catalog tuples. As decoding is done in fast_forward
2090 : * mode, no changes are generated anyway.
2091 : *
2092 : * *found_consistent_snapshot will be true if the initial decoding snapshot has
2093 : * been built; Otherwise, it will be false.
2094 : */
2095 : XLogRecPtr
2096 22 : LogicalSlotAdvanceAndCheckSnapState(XLogRecPtr moveto,
2097 : bool *found_consistent_snapshot)
2098 : {
2099 : LogicalDecodingContext *ctx;
2100 22 : ResourceOwner old_resowner PG_USED_FOR_ASSERTS_ONLY = CurrentResourceOwner;
2101 : XLogRecPtr retlsn;
2102 :
2103 : Assert(XLogRecPtrIsValid(moveto));
2104 :
2105 22 : if (found_consistent_snapshot)
2106 11 : *found_consistent_snapshot = false;
2107 :
2108 22 : PG_TRY();
2109 : {
2110 : /*
2111 : * Create our decoding context in fast_forward mode, passing start_lsn
2112 : * as InvalidXLogRecPtr, so that we start processing from my slot's
2113 : * confirmed_flush.
2114 : */
2115 44 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
2116 : NIL,
2117 : true, /* fast_forward */
2118 22 : XL_ROUTINE(.page_read = read_local_xlog_page,
2119 : .segment_open = wal_segment_open,
2120 : .segment_close = wal_segment_close),
2121 : NULL, NULL, NULL);
2122 :
2123 : /*
2124 : * Wait for specified streaming replication standby servers (if any)
2125 : * to confirm receipt of WAL up to moveto lsn.
2126 : */
2127 22 : WaitForStandbyConfirmation(moveto);
2128 :
2129 : /*
2130 : * Start reading at the slot's restart_lsn, which we know to point to
2131 : * a valid record.
2132 : */
2133 22 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
2134 :
2135 : /* invalidate non-timetravel entries */
2136 22 : InvalidateSystemCaches();
2137 :
2138 : /* Decode records until we reach the requested target */
2139 2940 : while (ctx->reader->EndRecPtr < moveto)
2140 : {
2141 2918 : char *errm = NULL;
2142 : XLogRecord *record;
2143 :
2144 : /*
2145 : * Read records. No changes are generated in fast_forward mode,
2146 : * but snapbuilder/slot statuses are updated properly.
2147 : */
2148 2918 : record = XLogReadRecord(ctx->reader, &errm);
2149 2918 : if (errm)
2150 0 : elog(ERROR, "could not find record while advancing replication slot: %s",
2151 : errm);
2152 :
2153 : /*
2154 : * Process the record. Storage-level changes are ignored in
2155 : * fast_forward mode, but other modules (such as snapbuilder)
2156 : * might still have critical updates to do.
2157 : */
2158 2918 : if (record)
2159 : {
2160 2918 : LogicalDecodingProcessRecord(ctx, ctx->reader);
2161 :
2162 : /*
2163 : * We used to have bugs where logical decoding would fail to
2164 : * preserve the resource owner. That's important here, so
2165 : * verify that that doesn't happen anymore. XXX this could be
2166 : * removed once it's been battle-tested.
2167 : */
2168 : Assert(CurrentResourceOwner == old_resowner);
2169 : }
2170 :
2171 2918 : CHECK_FOR_INTERRUPTS();
2172 : }
2173 :
2174 22 : if (found_consistent_snapshot && DecodingContextReady(ctx))
2175 11 : *found_consistent_snapshot = true;
2176 :
2177 22 : if (XLogRecPtrIsValid(ctx->reader->EndRecPtr))
2178 : {
2179 22 : LogicalConfirmReceivedLocation(moveto);
2180 :
2181 : /*
2182 : * If only the confirmed_flush LSN has changed the slot won't get
2183 : * marked as dirty by the above. Callers on the walsender
2184 : * interface are expected to keep track of their own progress and
2185 : * don't need it written out. But SQL-interface users cannot
2186 : * specify their own start positions and it's harder for them to
2187 : * keep track of their progress, so we should make more of an
2188 : * effort to save it for them.
2189 : *
2190 : * Dirty the slot so it is written out at the next checkpoint. The
2191 : * LSN position advanced to may still be lost on a crash but this
2192 : * makes the data consistent after a clean shutdown.
2193 : */
2194 22 : ReplicationSlotMarkDirty();
2195 : }
2196 :
2197 22 : retlsn = MyReplicationSlot->data.confirmed_flush;
2198 :
2199 : /* free context, call shutdown callback */
2200 22 : FreeDecodingContext(ctx);
2201 :
2202 22 : InvalidateSystemCaches();
2203 : }
2204 0 : PG_CATCH();
2205 : {
2206 : /* clear all timetravel entries */
2207 0 : InvalidateSystemCaches();
2208 :
2209 0 : PG_RE_THROW();
2210 : }
2211 22 : PG_END_TRY();
2212 :
2213 22 : return retlsn;
2214 : }
|
