Line data Source code
1 : /*-------------------------------------------------------------------------
2 : * logical.c
3 : * PostgreSQL logical decoding coordination
4 : *
5 : * Copyright (c) 2012-2025, 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 3186 : CheckLogicalDecodingRequirements(void)
112 : {
113 3186 : CheckSlotRequirements();
114 :
115 : /*
116 : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
117 : * needs the same check.
118 : */
119 :
120 3186 : if (MyDatabaseId == InvalidOid)
121 2 : 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 3184 : if (RecoveryInProgress() && !IsLogicalDecodingEnabled())
130 4 : 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 3180 : }
135 :
136 : /*
137 : * Helper function for CreateInitDecodingContext() and
138 : * CreateDecodingContext() performing common tasks.
139 : */
140 : static LogicalDecodingContext *
141 2234 : 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 2234 : slot = MyReplicationSlot;
159 :
160 2234 : context = AllocSetContextCreate(CurrentMemoryContext,
161 : "Logical decoding context",
162 : ALLOCSET_DEFAULT_SIZES);
163 2234 : old_context = MemoryContextSwitchTo(context);
164 2234 : ctx = palloc0_object(LogicalDecodingContext);
165 :
166 2234 : ctx->context = context;
167 :
168 : /*
169 : * (re-)load output plugins, so we detect a bad (removed) output plugin
170 : * now.
171 : */
172 2234 : if (!fast_forward)
173 2188 : 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 2232 : if (!IsTransactionOrTransactionBlock())
188 : {
189 1108 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
190 1108 : MyProc->statusFlags |= PROC_IN_LOGICAL_DECODING;
191 1108 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
192 1108 : LWLockRelease(ProcArrayLock);
193 : }
194 :
195 2232 : ctx->slot = slot;
196 :
197 2232 : ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
198 2232 : 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 2232 : ctx->reorder = ReorderBufferAllocate();
205 2232 : ctx->snapshot_builder =
206 2232 : AllocateSnapshotBuilder(ctx->reorder, xmin_horizon, start_lsn,
207 : need_full_snapshot, in_create, slot->data.two_phase_at);
208 :
209 2232 : ctx->reorder->private_data = ctx;
210 :
211 : /* wrap output plugin callbacks, so we can add error context information */
212 2232 : ctx->reorder->begin = begin_cb_wrapper;
213 2232 : ctx->reorder->apply_change = change_cb_wrapper;
214 2232 : ctx->reorder->apply_truncate = truncate_cb_wrapper;
215 2232 : ctx->reorder->commit = commit_cb_wrapper;
216 2232 : 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 4510 : ctx->streaming = (ctx->callbacks.stream_start_cb != NULL) ||
228 46 : (ctx->callbacks.stream_stop_cb != NULL) ||
229 46 : (ctx->callbacks.stream_abort_cb != NULL) ||
230 46 : (ctx->callbacks.stream_commit_cb != NULL) ||
231 46 : (ctx->callbacks.stream_change_cb != NULL) ||
232 2324 : (ctx->callbacks.stream_message_cb != NULL) ||
233 46 : (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 2232 : ctx->reorder->stream_start = stream_start_cb_wrapper;
244 2232 : ctx->reorder->stream_stop = stream_stop_cb_wrapper;
245 2232 : ctx->reorder->stream_abort = stream_abort_cb_wrapper;
246 2232 : ctx->reorder->stream_prepare = stream_prepare_cb_wrapper;
247 2232 : ctx->reorder->stream_commit = stream_commit_cb_wrapper;
248 2232 : ctx->reorder->stream_change = stream_change_cb_wrapper;
249 2232 : ctx->reorder->stream_message = stream_message_cb_wrapper;
250 2232 : 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 4510 : ctx->twophase = (ctx->callbacks.begin_prepare_cb != NULL) ||
263 46 : (ctx->callbacks.prepare_cb != NULL) ||
264 46 : (ctx->callbacks.commit_prepared_cb != NULL) ||
265 46 : (ctx->callbacks.rollback_prepared_cb != NULL) ||
266 2324 : (ctx->callbacks.stream_prepare_cb != NULL) ||
267 46 : (ctx->callbacks.filter_prepare_cb != NULL);
268 :
269 : /*
270 : * Callback to support decoding at prepare time.
271 : */
272 2232 : ctx->reorder->begin_prepare = begin_prepare_cb_wrapper;
273 2232 : ctx->reorder->prepare = prepare_cb_wrapper;
274 2232 : ctx->reorder->commit_prepared = commit_prepared_cb_wrapper;
275 2232 : 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 2232 : ctx->reorder->update_progress_txn = update_progress_txn_cb_wrapper;
282 :
283 2232 : ctx->out = makeStringInfo();
284 2232 : ctx->prepare_write = prepare_write;
285 2232 : ctx->write = do_write;
286 2232 : ctx->update_progress = update_progress;
287 :
288 2232 : ctx->output_plugin_options = output_plugin_options;
289 :
290 2232 : ctx->fast_forward = fast_forward;
291 :
292 2232 : MemoryContextSwitchTo(old_context);
293 :
294 2232 : 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 : * restart_lsn -- if given as invalid, it's this routine's responsibility to
305 : * mark WAL as reserved by setting a convenient restart_lsn for the slot.
306 : * Otherwise, we set for decoding to start from the given LSN without
307 : * marking WAL reserved beforehand. In that scenario, it's up to the
308 : * caller to guarantee that WAL remains available.
309 : * xl_routine -- XLogReaderRoutine for underlying XLogReader
310 : * prepare_write, do_write, update_progress --
311 : * callbacks that perform the use-case dependent, actual, work.
312 : *
313 : * Needs to be called while in a memory context that's at least as long lived
314 : * as the decoding context because further memory contexts will be created
315 : * inside it.
316 : *
317 : * Returns an initialized decoding context after calling the output plugin's
318 : * startup function.
319 : */
320 : LogicalDecodingContext *
321 950 : CreateInitDecodingContext(const char *plugin,
322 : List *output_plugin_options,
323 : bool need_full_snapshot,
324 : XLogRecPtr restart_lsn,
325 : XLogReaderRoutine *xl_routine,
326 : LogicalOutputPluginWriterPrepareWrite prepare_write,
327 : LogicalOutputPluginWriterWrite do_write,
328 : LogicalOutputPluginWriterUpdateProgress update_progress)
329 : {
330 950 : TransactionId xmin_horizon = InvalidTransactionId;
331 : ReplicationSlot *slot;
332 : NameData plugin_name;
333 : LogicalDecodingContext *ctx;
334 : MemoryContext old_context;
335 :
336 : /*
337 : * On a standby, this check is also required while creating the slot.
338 : * Check the comments in the function.
339 : */
340 950 : CheckLogicalDecodingRequirements();
341 :
342 : /* shorter lines... */
343 950 : slot = MyReplicationSlot;
344 :
345 : /* first some sanity checks that are unlikely to be violated */
346 950 : if (slot == NULL)
347 0 : elog(ERROR, "cannot perform logical decoding without an acquired slot");
348 :
349 950 : if (plugin == NULL)
350 0 : elog(ERROR, "cannot initialize logical decoding without a specified plugin");
351 :
352 : /* Make sure the passed slot is suitable. These are user facing errors. */
353 950 : if (SlotIsPhysical(slot))
354 0 : ereport(ERROR,
355 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
356 : errmsg("cannot use physical replication slot for logical decoding")));
357 :
358 950 : if (slot->data.database != MyDatabaseId)
359 0 : ereport(ERROR,
360 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
361 : errmsg("replication slot \"%s\" was not created in this database",
362 : NameStr(slot->data.name))));
363 :
364 1614 : if (IsTransactionState() &&
365 664 : GetTopTransactionIdIfAny() != InvalidTransactionId)
366 4 : ereport(ERROR,
367 : (errcode(ERRCODE_ACTIVE_SQL_TRANSACTION),
368 : errmsg("cannot create logical replication slot in transaction that has performed writes")));
369 :
370 : /*
371 : * Register output plugin name with slot. We need the mutex to avoid
372 : * concurrent reading of a partially copied string. But we don't want any
373 : * complicated code while holding a spinlock, so do namestrcpy() outside.
374 : */
375 946 : namestrcpy(&plugin_name, plugin);
376 946 : SpinLockAcquire(&slot->mutex);
377 946 : slot->data.plugin = plugin_name;
378 946 : SpinLockRelease(&slot->mutex);
379 :
380 946 : if (!XLogRecPtrIsValid(restart_lsn))
381 932 : ReplicationSlotReserveWal();
382 : else
383 : {
384 14 : SpinLockAcquire(&slot->mutex);
385 14 : slot->data.restart_lsn = restart_lsn;
386 14 : SpinLockRelease(&slot->mutex);
387 : }
388 :
389 : /* ----
390 : * This is a bit tricky: We need to determine a safe xmin horizon to start
391 : * decoding from, to avoid starting from a running xacts record referring
392 : * to xids whose rows have been vacuumed or pruned
393 : * already. GetOldestSafeDecodingTransactionId() returns such a value, but
394 : * without further interlock its return value might immediately be out of
395 : * date.
396 : *
397 : * So we have to acquire the ProcArrayLock to prevent computation of new
398 : * xmin horizons by other backends, get the safe decoding xid, and inform
399 : * the slot machinery about the new limit. Once that's done the
400 : * ProcArrayLock can be released as the slot machinery now is
401 : * protecting against vacuum.
402 : *
403 : * Note that, temporarily, the data, not just the catalog, xmin has to be
404 : * reserved if a data snapshot is to be exported. Otherwise the initial
405 : * data snapshot created here is not guaranteed to be valid. After that
406 : * the data xmin doesn't need to be managed anymore and the global xmin
407 : * should be recomputed. As we are fine with losing the pegged data xmin
408 : * after crash - no chance a snapshot would get exported anymore - we can
409 : * get away with just setting the slot's
410 : * effective_xmin. ReplicationSlotRelease will reset it again.
411 : *
412 : * ----
413 : */
414 946 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
415 :
416 946 : xmin_horizon = GetOldestSafeDecodingTransactionId(!need_full_snapshot);
417 :
418 946 : SpinLockAcquire(&slot->mutex);
419 946 : slot->effective_catalog_xmin = xmin_horizon;
420 946 : slot->data.catalog_xmin = xmin_horizon;
421 946 : if (need_full_snapshot)
422 400 : slot->effective_xmin = xmin_horizon;
423 946 : SpinLockRelease(&slot->mutex);
424 :
425 946 : ReplicationSlotsComputeRequiredXmin(true);
426 :
427 946 : LWLockRelease(ProcArrayLock);
428 :
429 946 : ReplicationSlotMarkDirty();
430 946 : ReplicationSlotSave();
431 :
432 946 : ctx = StartupDecodingContext(NIL, restart_lsn, xmin_horizon,
433 : need_full_snapshot, false, true,
434 : xl_routine, prepare_write, do_write,
435 : update_progress);
436 :
437 : /* call output plugin initialization callback */
438 944 : old_context = MemoryContextSwitchTo(ctx->context);
439 944 : if (ctx->callbacks.startup_cb != NULL)
440 944 : startup_cb_wrapper(ctx, &ctx->options, true);
441 944 : MemoryContextSwitchTo(old_context);
442 :
443 : /*
444 : * We allow decoding of prepared transactions when the two_phase is
445 : * enabled at the time of slot creation, or when the two_phase option is
446 : * given at the streaming start, provided the plugin supports all the
447 : * callbacks for two-phase.
448 : */
449 944 : ctx->twophase &= slot->data.two_phase;
450 :
451 944 : ctx->reorder->output_rewrites = ctx->options.receive_rewrites;
452 :
453 944 : return ctx;
454 : }
455 :
456 : /*
457 : * Create a new decoding context, for a logical slot that has previously been
458 : * used already.
459 : *
460 : * start_lsn
461 : * The LSN at which to start decoding. If InvalidXLogRecPtr, restart
462 : * from the slot's confirmed_flush; otherwise, start from the specified
463 : * location (but move it forwards to confirmed_flush if it's older than
464 : * that, see below).
465 : *
466 : * output_plugin_options
467 : * options passed to the output plugin.
468 : *
469 : * fast_forward
470 : * bypass the generation of logical changes.
471 : *
472 : * xl_routine
473 : * XLogReaderRoutine used by underlying xlogreader
474 : *
475 : * prepare_write, do_write, update_progress
476 : * callbacks that have to be filled to perform the use-case dependent,
477 : * actual work.
478 : *
479 : * Needs to be called while in a memory context that's at least as long lived
480 : * as the decoding context because further memory contexts will be created
481 : * inside it.
482 : *
483 : * Returns an initialized decoding context after calling the output plugin's
484 : * startup function.
485 : */
486 : LogicalDecodingContext *
487 1298 : CreateDecodingContext(XLogRecPtr start_lsn,
488 : List *output_plugin_options,
489 : bool fast_forward,
490 : XLogReaderRoutine *xl_routine,
491 : LogicalOutputPluginWriterPrepareWrite prepare_write,
492 : LogicalOutputPluginWriterWrite do_write,
493 : LogicalOutputPluginWriterUpdateProgress update_progress)
494 : {
495 : LogicalDecodingContext *ctx;
496 : ReplicationSlot *slot;
497 : MemoryContext old_context;
498 :
499 : /* shorter lines... */
500 1298 : slot = MyReplicationSlot;
501 :
502 : /* first some sanity checks that are unlikely to be violated */
503 1298 : if (slot == NULL)
504 0 : elog(ERROR, "cannot perform logical decoding without an acquired slot");
505 :
506 : /* make sure the passed slot is suitable, these are user facing errors */
507 1298 : if (SlotIsPhysical(slot))
508 2 : ereport(ERROR,
509 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
510 : errmsg("cannot use physical replication slot for logical decoding")));
511 :
512 : /*
513 : * We need to access the system tables during decoding to build the
514 : * logical changes unless we are in fast_forward mode where no changes are
515 : * generated.
516 : */
517 1296 : if (slot->data.database != MyDatabaseId && !fast_forward)
518 6 : ereport(ERROR,
519 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
520 : errmsg("replication slot \"%s\" was not created in this database",
521 : NameStr(slot->data.name))));
522 :
523 : /*
524 : * The slots being synced from the primary can't be used for decoding as
525 : * they are used after failover. However, we do allow advancing the LSNs
526 : * during the synchronization of slots. See update_local_synced_slot.
527 : */
528 1290 : if (RecoveryInProgress() && slot->data.synced && !IsSyncingReplicationSlots())
529 2 : ereport(ERROR,
530 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
531 : errmsg("cannot use replication slot \"%s\" for logical decoding",
532 : NameStr(slot->data.name)),
533 : errdetail("This replication slot is being synchronized from the primary server."),
534 : errhint("Specify another replication slot."));
535 :
536 : /* slot must be valid to allow decoding */
537 : Assert(slot->data.invalidated == RS_INVAL_NONE);
538 : Assert(XLogRecPtrIsValid(slot->data.restart_lsn));
539 :
540 1288 : if (!XLogRecPtrIsValid(start_lsn))
541 : {
542 : /* continue from last position */
543 750 : start_lsn = slot->data.confirmed_flush;
544 : }
545 538 : else if (start_lsn < slot->data.confirmed_flush)
546 : {
547 : /*
548 : * It might seem like we should error out in this case, but it's
549 : * pretty common for a client to acknowledge a LSN it doesn't have to
550 : * do anything for, and thus didn't store persistently, because the
551 : * xlog records didn't result in anything relevant for logical
552 : * decoding. Clients have to be able to do that to support synchronous
553 : * replication.
554 : *
555 : * Starting at a different LSN than requested might not catch certain
556 : * kinds of client errors; so the client may wish to check that
557 : * confirmed_flush_lsn matches its expectations.
558 : */
559 76 : elog(LOG, "%X/%08X has been already streamed, forwarding to %X/%08X",
560 : LSN_FORMAT_ARGS(start_lsn),
561 : LSN_FORMAT_ARGS(slot->data.confirmed_flush));
562 :
563 76 : start_lsn = slot->data.confirmed_flush;
564 : }
565 :
566 1288 : ctx = StartupDecodingContext(output_plugin_options,
567 : start_lsn, InvalidTransactionId, false,
568 : fast_forward, false, xl_routine, prepare_write,
569 : do_write, update_progress);
570 :
571 : /* call output plugin initialization callback */
572 1288 : old_context = MemoryContextSwitchTo(ctx->context);
573 1288 : if (ctx->callbacks.startup_cb != NULL)
574 1242 : startup_cb_wrapper(ctx, &ctx->options, false);
575 1282 : MemoryContextSwitchTo(old_context);
576 :
577 : /*
578 : * We allow decoding of prepared transactions when the two_phase is
579 : * enabled at the time of slot creation, or when the two_phase option is
580 : * given at the streaming start, provided the plugin supports all the
581 : * callbacks for two-phase.
582 : */
583 1282 : ctx->twophase &= (slot->data.two_phase || ctx->twophase_opt_given);
584 :
585 : /* Mark slot to allow two_phase decoding if not already marked */
586 1282 : if (ctx->twophase && !slot->data.two_phase)
587 : {
588 14 : SpinLockAcquire(&slot->mutex);
589 14 : slot->data.two_phase = true;
590 14 : slot->data.two_phase_at = start_lsn;
591 14 : SpinLockRelease(&slot->mutex);
592 14 : ReplicationSlotMarkDirty();
593 14 : ReplicationSlotSave();
594 14 : SnapBuildSetTwoPhaseAt(ctx->snapshot_builder, start_lsn);
595 : }
596 :
597 1282 : ctx->reorder->output_rewrites = ctx->options.receive_rewrites;
598 :
599 1282 : ereport(LOG,
600 : (errmsg("starting logical decoding for slot \"%s\"",
601 : NameStr(slot->data.name)),
602 : errdetail("Streaming transactions committing after %X/%08X, reading WAL from %X/%08X.",
603 : LSN_FORMAT_ARGS(slot->data.confirmed_flush),
604 : LSN_FORMAT_ARGS(slot->data.restart_lsn))));
605 :
606 1282 : return ctx;
607 : }
608 :
609 : /*
610 : * Returns true if a consistent initial decoding snapshot has been built.
611 : */
612 : bool
613 2042 : DecodingContextReady(LogicalDecodingContext *ctx)
614 : {
615 2042 : return SnapBuildCurrentState(ctx->snapshot_builder) == SNAPBUILD_CONSISTENT;
616 : }
617 :
618 : /*
619 : * Read from the decoding slot, until it is ready to start extracting changes.
620 : */
621 : void
622 930 : DecodingContextFindStartpoint(LogicalDecodingContext *ctx)
623 : {
624 930 : ReplicationSlot *slot = ctx->slot;
625 :
626 : /* Initialize from where to start reading WAL. */
627 930 : XLogBeginRead(ctx->reader, slot->data.restart_lsn);
628 :
629 930 : elog(DEBUG1, "searching for logical decoding starting point, starting at %X/%08X",
630 : LSN_FORMAT_ARGS(slot->data.restart_lsn));
631 :
632 : /* Wait for a consistent starting point */
633 : for (;;)
634 1104 : {
635 : XLogRecord *record;
636 2034 : char *err = NULL;
637 :
638 : /* the read_page callback waits for new WAL */
639 2034 : record = XLogReadRecord(ctx->reader, &err);
640 2034 : if (err)
641 0 : elog(ERROR, "could not find logical decoding starting point: %s", err);
642 2034 : if (!record)
643 0 : elog(ERROR, "could not find logical decoding starting point");
644 :
645 2034 : LogicalDecodingProcessRecord(ctx, ctx->reader);
646 :
647 : /* only continue till we found a consistent spot */
648 2030 : if (DecodingContextReady(ctx))
649 926 : break;
650 :
651 1104 : CHECK_FOR_INTERRUPTS();
652 : }
653 :
654 926 : SpinLockAcquire(&slot->mutex);
655 926 : slot->data.confirmed_flush = ctx->reader->EndRecPtr;
656 926 : if (slot->data.two_phase)
657 18 : slot->data.two_phase_at = ctx->reader->EndRecPtr;
658 926 : SpinLockRelease(&slot->mutex);
659 926 : }
660 :
661 : /*
662 : * Free a previously allocated decoding context, invoking the shutdown
663 : * callback if necessary.
664 : */
665 : void
666 1784 : FreeDecodingContext(LogicalDecodingContext *ctx)
667 : {
668 1784 : if (ctx->callbacks.shutdown_cb != NULL)
669 1738 : shutdown_cb_wrapper(ctx);
670 :
671 1784 : ReorderBufferFree(ctx->reorder);
672 1784 : FreeSnapshotBuilder(ctx->snapshot_builder);
673 1784 : XLogReaderFree(ctx->reader);
674 1784 : MemoryContextDelete(ctx->context);
675 1784 : }
676 :
677 : /*
678 : * Prepare a write using the context's output routine.
679 : */
680 : void
681 672442 : OutputPluginPrepareWrite(struct LogicalDecodingContext *ctx, bool last_write)
682 : {
683 672442 : if (!ctx->accept_writes)
684 0 : elog(ERROR, "writes are only accepted in commit, begin and change callbacks");
685 :
686 672442 : ctx->prepare_write(ctx, ctx->write_location, ctx->write_xid, last_write);
687 672442 : ctx->prepared_write = true;
688 672442 : }
689 :
690 : /*
691 : * Perform a write using the context's output routine.
692 : */
693 : void
694 672442 : OutputPluginWrite(struct LogicalDecodingContext *ctx, bool last_write)
695 : {
696 672442 : if (!ctx->prepared_write)
697 0 : elog(ERROR, "OutputPluginPrepareWrite needs to be called before OutputPluginWrite");
698 :
699 672442 : ctx->write(ctx, ctx->write_location, ctx->write_xid, last_write);
700 672406 : ctx->prepared_write = false;
701 672406 : }
702 :
703 : /*
704 : * Update progress tracking (if supported).
705 : */
706 : void
707 8258 : OutputPluginUpdateProgress(struct LogicalDecodingContext *ctx,
708 : bool skipped_xact)
709 : {
710 8258 : if (!ctx->update_progress)
711 3180 : return;
712 :
713 5078 : ctx->update_progress(ctx, ctx->write_location, ctx->write_xid,
714 : skipped_xact);
715 : }
716 :
717 : /*
718 : * Load the output plugin, lookup its output plugin init function, and check
719 : * that it provides the required callbacks.
720 : */
721 : static void
722 2188 : LoadOutputPlugin(OutputPluginCallbacks *callbacks, const char *plugin)
723 : {
724 : LogicalOutputPluginInit plugin_init;
725 :
726 2186 : plugin_init = (LogicalOutputPluginInit)
727 2188 : load_external_function(plugin, "_PG_output_plugin_init", false, NULL);
728 :
729 2186 : if (plugin_init == NULL)
730 0 : elog(ERROR, "output plugins have to declare the _PG_output_plugin_init symbol");
731 :
732 : /* ask the output plugin to fill the callback struct */
733 2186 : plugin_init(callbacks);
734 :
735 2186 : if (callbacks->begin_cb == NULL)
736 0 : elog(ERROR, "output plugins have to register a begin callback");
737 2186 : if (callbacks->change_cb == NULL)
738 0 : elog(ERROR, "output plugins have to register a change callback");
739 2186 : if (callbacks->commit_cb == NULL)
740 0 : elog(ERROR, "output plugins have to register a commit callback");
741 2186 : }
742 :
743 : static void
744 260 : output_plugin_error_callback(void *arg)
745 : {
746 260 : LogicalErrorCallbackState *state = (LogicalErrorCallbackState *) arg;
747 :
748 : /* not all callbacks have an associated LSN */
749 260 : if (XLogRecPtrIsValid(state->report_location))
750 254 : errcontext("slot \"%s\", output plugin \"%s\", in the %s callback, associated LSN %X/%08X",
751 254 : NameStr(state->ctx->slot->data.name),
752 254 : NameStr(state->ctx->slot->data.plugin),
753 : state->callback_name,
754 254 : LSN_FORMAT_ARGS(state->report_location));
755 : else
756 6 : errcontext("slot \"%s\", output plugin \"%s\", in the %s callback",
757 6 : NameStr(state->ctx->slot->data.name),
758 6 : NameStr(state->ctx->slot->data.plugin),
759 : state->callback_name);
760 260 : }
761 :
762 : static void
763 2186 : startup_cb_wrapper(LogicalDecodingContext *ctx, OutputPluginOptions *opt, bool is_init)
764 : {
765 : LogicalErrorCallbackState state;
766 : ErrorContextCallback errcallback;
767 :
768 : Assert(!ctx->fast_forward);
769 :
770 : /* Push callback + info on the error context stack */
771 2186 : state.ctx = ctx;
772 2186 : state.callback_name = "startup";
773 2186 : state.report_location = InvalidXLogRecPtr;
774 2186 : errcallback.callback = output_plugin_error_callback;
775 2186 : errcallback.arg = &state;
776 2186 : errcallback.previous = error_context_stack;
777 2186 : error_context_stack = &errcallback;
778 :
779 : /* set output state */
780 2186 : ctx->accept_writes = false;
781 2186 : ctx->end_xact = false;
782 :
783 : /* do the actual work: call callback */
784 2186 : ctx->callbacks.startup_cb(ctx, opt, is_init);
785 :
786 : /* Pop the error context stack */
787 2180 : error_context_stack = errcallback.previous;
788 2180 : }
789 :
790 : static void
791 1738 : shutdown_cb_wrapper(LogicalDecodingContext *ctx)
792 : {
793 : LogicalErrorCallbackState state;
794 : ErrorContextCallback errcallback;
795 :
796 : Assert(!ctx->fast_forward);
797 :
798 : /* Push callback + info on the error context stack */
799 1738 : state.ctx = ctx;
800 1738 : state.callback_name = "shutdown";
801 1738 : state.report_location = InvalidXLogRecPtr;
802 1738 : errcallback.callback = output_plugin_error_callback;
803 1738 : errcallback.arg = &state;
804 1738 : errcallback.previous = error_context_stack;
805 1738 : error_context_stack = &errcallback;
806 :
807 : /* set output state */
808 1738 : ctx->accept_writes = false;
809 1738 : ctx->end_xact = false;
810 :
811 : /* do the actual work: call callback */
812 1738 : ctx->callbacks.shutdown_cb(ctx);
813 :
814 : /* Pop the error context stack */
815 1738 : error_context_stack = errcallback.previous;
816 1738 : }
817 :
818 :
819 : /*
820 : * Callbacks for ReorderBuffer which add in some more information and then call
821 : * output_plugin.h plugins.
822 : */
823 : static void
824 2706 : begin_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
825 : {
826 2706 : LogicalDecodingContext *ctx = cache->private_data;
827 : LogicalErrorCallbackState state;
828 : ErrorContextCallback errcallback;
829 :
830 : Assert(!ctx->fast_forward);
831 :
832 : /* Push callback + info on the error context stack */
833 2706 : state.ctx = ctx;
834 2706 : state.callback_name = "begin";
835 2706 : state.report_location = txn->first_lsn;
836 2706 : errcallback.callback = output_plugin_error_callback;
837 2706 : errcallback.arg = &state;
838 2706 : errcallback.previous = error_context_stack;
839 2706 : error_context_stack = &errcallback;
840 :
841 : /* set output state */
842 2706 : ctx->accept_writes = true;
843 2706 : ctx->write_xid = txn->xid;
844 2706 : ctx->write_location = txn->first_lsn;
845 2706 : ctx->end_xact = false;
846 :
847 : /* do the actual work: call callback */
848 2706 : ctx->callbacks.begin_cb(ctx, txn);
849 :
850 : /* Pop the error context stack */
851 2706 : error_context_stack = errcallback.previous;
852 2706 : }
853 :
854 : static void
855 2702 : commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
856 : XLogRecPtr commit_lsn)
857 : {
858 2702 : LogicalDecodingContext *ctx = cache->private_data;
859 : LogicalErrorCallbackState state;
860 : ErrorContextCallback errcallback;
861 :
862 : Assert(!ctx->fast_forward);
863 :
864 : /* Push callback + info on the error context stack */
865 2702 : state.ctx = ctx;
866 2702 : state.callback_name = "commit";
867 2702 : state.report_location = txn->final_lsn; /* beginning of commit record */
868 2702 : errcallback.callback = output_plugin_error_callback;
869 2702 : errcallback.arg = &state;
870 2702 : errcallback.previous = error_context_stack;
871 2702 : error_context_stack = &errcallback;
872 :
873 : /* set output state */
874 2702 : ctx->accept_writes = true;
875 2702 : ctx->write_xid = txn->xid;
876 2702 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
877 2702 : ctx->end_xact = true;
878 :
879 : /* do the actual work: call callback */
880 2702 : ctx->callbacks.commit_cb(ctx, txn, commit_lsn);
881 :
882 : /* Pop the error context stack */
883 2668 : error_context_stack = errcallback.previous;
884 2668 : }
885 :
886 : /*
887 : * The functionality of begin_prepare is quite similar to begin with the
888 : * exception that this will have gid (global transaction id) information which
889 : * can be used by plugin. Now, we thought about extending the existing begin
890 : * but that would break the replication protocol and additionally this looks
891 : * cleaner.
892 : */
893 : static void
894 60 : begin_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
895 : {
896 60 : LogicalDecodingContext *ctx = cache->private_data;
897 : LogicalErrorCallbackState state;
898 : ErrorContextCallback errcallback;
899 :
900 : Assert(!ctx->fast_forward);
901 :
902 : /* We're only supposed to call this when two-phase commits are supported */
903 : Assert(ctx->twophase);
904 :
905 : /* Push callback + info on the error context stack */
906 60 : state.ctx = ctx;
907 60 : state.callback_name = "begin_prepare";
908 60 : state.report_location = txn->first_lsn;
909 60 : errcallback.callback = output_plugin_error_callback;
910 60 : errcallback.arg = &state;
911 60 : errcallback.previous = error_context_stack;
912 60 : error_context_stack = &errcallback;
913 :
914 : /* set output state */
915 60 : ctx->accept_writes = true;
916 60 : ctx->write_xid = txn->xid;
917 60 : ctx->write_location = txn->first_lsn;
918 60 : ctx->end_xact = false;
919 :
920 : /*
921 : * If the plugin supports two-phase commits then begin prepare callback is
922 : * mandatory
923 : */
924 60 : if (ctx->callbacks.begin_prepare_cb == NULL)
925 0 : ereport(ERROR,
926 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
927 : errmsg("logical replication at prepare time requires a %s callback",
928 : "begin_prepare_cb")));
929 :
930 : /* do the actual work: call callback */
931 60 : ctx->callbacks.begin_prepare_cb(ctx, txn);
932 :
933 : /* Pop the error context stack */
934 60 : error_context_stack = errcallback.previous;
935 60 : }
936 :
937 : static void
938 60 : prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
939 : XLogRecPtr prepare_lsn)
940 : {
941 60 : LogicalDecodingContext *ctx = cache->private_data;
942 : LogicalErrorCallbackState state;
943 : ErrorContextCallback errcallback;
944 :
945 : Assert(!ctx->fast_forward);
946 :
947 : /* We're only supposed to call this when two-phase commits are supported */
948 : Assert(ctx->twophase);
949 :
950 : /* Push callback + info on the error context stack */
951 60 : state.ctx = ctx;
952 60 : state.callback_name = "prepare";
953 60 : state.report_location = txn->final_lsn; /* beginning of prepare record */
954 60 : errcallback.callback = output_plugin_error_callback;
955 60 : errcallback.arg = &state;
956 60 : errcallback.previous = error_context_stack;
957 60 : error_context_stack = &errcallback;
958 :
959 : /* set output state */
960 60 : ctx->accept_writes = true;
961 60 : ctx->write_xid = txn->xid;
962 60 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
963 60 : ctx->end_xact = true;
964 :
965 : /*
966 : * If the plugin supports two-phase commits then prepare callback is
967 : * mandatory
968 : */
969 60 : if (ctx->callbacks.prepare_cb == NULL)
970 0 : ereport(ERROR,
971 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
972 : errmsg("logical replication at prepare time requires a %s callback",
973 : "prepare_cb")));
974 :
975 : /* do the actual work: call callback */
976 60 : ctx->callbacks.prepare_cb(ctx, txn, prepare_lsn);
977 :
978 : /* Pop the error context stack */
979 60 : error_context_stack = errcallback.previous;
980 60 : }
981 :
982 : static void
983 66 : commit_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
984 : XLogRecPtr commit_lsn)
985 : {
986 66 : LogicalDecodingContext *ctx = cache->private_data;
987 : LogicalErrorCallbackState state;
988 : ErrorContextCallback errcallback;
989 :
990 : Assert(!ctx->fast_forward);
991 :
992 : /* We're only supposed to call this when two-phase commits are supported */
993 : Assert(ctx->twophase);
994 :
995 : /* Push callback + info on the error context stack */
996 66 : state.ctx = ctx;
997 66 : state.callback_name = "commit_prepared";
998 66 : state.report_location = txn->final_lsn; /* beginning of commit record */
999 66 : errcallback.callback = output_plugin_error_callback;
1000 66 : errcallback.arg = &state;
1001 66 : errcallback.previous = error_context_stack;
1002 66 : error_context_stack = &errcallback;
1003 :
1004 : /* set output state */
1005 66 : ctx->accept_writes = true;
1006 66 : ctx->write_xid = txn->xid;
1007 66 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
1008 66 : ctx->end_xact = true;
1009 :
1010 : /*
1011 : * If the plugin support two-phase commits then commit prepared callback
1012 : * is mandatory
1013 : */
1014 66 : if (ctx->callbacks.commit_prepared_cb == NULL)
1015 0 : ereport(ERROR,
1016 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1017 : errmsg("logical replication at prepare time requires a %s callback",
1018 : "commit_prepared_cb")));
1019 :
1020 : /* do the actual work: call callback */
1021 66 : ctx->callbacks.commit_prepared_cb(ctx, txn, commit_lsn);
1022 :
1023 : /* Pop the error context stack */
1024 66 : error_context_stack = errcallback.previous;
1025 66 : }
1026 :
1027 : static void
1028 20 : rollback_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1029 : XLogRecPtr prepare_end_lsn,
1030 : TimestampTz prepare_time)
1031 : {
1032 20 : LogicalDecodingContext *ctx = cache->private_data;
1033 : LogicalErrorCallbackState state;
1034 : ErrorContextCallback errcallback;
1035 :
1036 : Assert(!ctx->fast_forward);
1037 :
1038 : /* We're only supposed to call this when two-phase commits are supported */
1039 : Assert(ctx->twophase);
1040 :
1041 : /* Push callback + info on the error context stack */
1042 20 : state.ctx = ctx;
1043 20 : state.callback_name = "rollback_prepared";
1044 20 : state.report_location = txn->final_lsn; /* beginning of commit record */
1045 20 : errcallback.callback = output_plugin_error_callback;
1046 20 : errcallback.arg = &state;
1047 20 : errcallback.previous = error_context_stack;
1048 20 : error_context_stack = &errcallback;
1049 :
1050 : /* set output state */
1051 20 : ctx->accept_writes = true;
1052 20 : ctx->write_xid = txn->xid;
1053 20 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
1054 20 : ctx->end_xact = true;
1055 :
1056 : /*
1057 : * If the plugin support two-phase commits then rollback prepared callback
1058 : * is mandatory
1059 : */
1060 20 : if (ctx->callbacks.rollback_prepared_cb == NULL)
1061 0 : ereport(ERROR,
1062 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1063 : errmsg("logical replication at prepare time requires a %s callback",
1064 : "rollback_prepared_cb")));
1065 :
1066 : /* do the actual work: call callback */
1067 20 : ctx->callbacks.rollback_prepared_cb(ctx, txn, prepare_end_lsn,
1068 : prepare_time);
1069 :
1070 : /* Pop the error context stack */
1071 20 : error_context_stack = errcallback.previous;
1072 20 : }
1073 :
1074 : static void
1075 316156 : change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1076 : Relation relation, ReorderBufferChange *change)
1077 : {
1078 316156 : LogicalDecodingContext *ctx = cache->private_data;
1079 : LogicalErrorCallbackState state;
1080 : ErrorContextCallback errcallback;
1081 :
1082 : Assert(!ctx->fast_forward);
1083 :
1084 : /* Push callback + info on the error context stack */
1085 316156 : state.ctx = ctx;
1086 316156 : state.callback_name = "change";
1087 316156 : state.report_location = change->lsn;
1088 316156 : errcallback.callback = output_plugin_error_callback;
1089 316156 : errcallback.arg = &state;
1090 316156 : errcallback.previous = error_context_stack;
1091 316156 : error_context_stack = &errcallback;
1092 :
1093 : /* set output state */
1094 316156 : ctx->accept_writes = true;
1095 316156 : ctx->write_xid = txn->xid;
1096 :
1097 : /*
1098 : * Report this change's lsn so replies from clients can give an up-to-date
1099 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1100 : * receipt of this transaction, but it might allow another transaction's
1101 : * commit to be confirmed with one message.
1102 : */
1103 316156 : ctx->write_location = change->lsn;
1104 :
1105 316156 : ctx->end_xact = false;
1106 :
1107 316156 : ctx->callbacks.change_cb(ctx, txn, relation, change);
1108 :
1109 : /* Pop the error context stack */
1110 316152 : error_context_stack = errcallback.previous;
1111 316152 : }
1112 :
1113 : static void
1114 50 : truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1115 : int nrelations, Relation relations[], ReorderBufferChange *change)
1116 : {
1117 50 : LogicalDecodingContext *ctx = cache->private_data;
1118 : LogicalErrorCallbackState state;
1119 : ErrorContextCallback errcallback;
1120 :
1121 : Assert(!ctx->fast_forward);
1122 :
1123 50 : if (!ctx->callbacks.truncate_cb)
1124 0 : return;
1125 :
1126 : /* Push callback + info on the error context stack */
1127 50 : state.ctx = ctx;
1128 50 : state.callback_name = "truncate";
1129 50 : state.report_location = change->lsn;
1130 50 : errcallback.callback = output_plugin_error_callback;
1131 50 : errcallback.arg = &state;
1132 50 : errcallback.previous = error_context_stack;
1133 50 : error_context_stack = &errcallback;
1134 :
1135 : /* set output state */
1136 50 : ctx->accept_writes = true;
1137 50 : ctx->write_xid = txn->xid;
1138 :
1139 : /*
1140 : * Report this change's lsn so replies from clients can give an up-to-date
1141 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1142 : * receipt of this transaction, but it might allow another transaction's
1143 : * commit to be confirmed with one message.
1144 : */
1145 50 : ctx->write_location = change->lsn;
1146 :
1147 50 : ctx->end_xact = false;
1148 :
1149 50 : ctx->callbacks.truncate_cb(ctx, txn, nrelations, relations, change);
1150 :
1151 : /* Pop the error context stack */
1152 50 : error_context_stack = errcallback.previous;
1153 : }
1154 :
1155 : bool
1156 296 : filter_prepare_cb_wrapper(LogicalDecodingContext *ctx, TransactionId xid,
1157 : const char *gid)
1158 : {
1159 : LogicalErrorCallbackState state;
1160 : ErrorContextCallback errcallback;
1161 : bool ret;
1162 :
1163 : Assert(!ctx->fast_forward);
1164 :
1165 : /* Push callback + info on the error context stack */
1166 296 : state.ctx = ctx;
1167 296 : state.callback_name = "filter_prepare";
1168 296 : state.report_location = InvalidXLogRecPtr;
1169 296 : errcallback.callback = output_plugin_error_callback;
1170 296 : errcallback.arg = &state;
1171 296 : errcallback.previous = error_context_stack;
1172 296 : error_context_stack = &errcallback;
1173 :
1174 : /* set output state */
1175 296 : ctx->accept_writes = false;
1176 296 : ctx->end_xact = false;
1177 :
1178 : /* do the actual work: call callback */
1179 296 : ret = ctx->callbacks.filter_prepare_cb(ctx, xid, gid);
1180 :
1181 : /* Pop the error context stack */
1182 296 : error_context_stack = errcallback.previous;
1183 :
1184 296 : return ret;
1185 : }
1186 :
1187 : bool
1188 3385696 : filter_by_origin_cb_wrapper(LogicalDecodingContext *ctx, RepOriginId origin_id)
1189 : {
1190 : LogicalErrorCallbackState state;
1191 : ErrorContextCallback errcallback;
1192 : bool ret;
1193 :
1194 : Assert(!ctx->fast_forward);
1195 :
1196 : /* Push callback + info on the error context stack */
1197 3385696 : state.ctx = ctx;
1198 3385696 : state.callback_name = "filter_by_origin";
1199 3385696 : state.report_location = InvalidXLogRecPtr;
1200 3385696 : errcallback.callback = output_plugin_error_callback;
1201 3385696 : errcallback.arg = &state;
1202 3385696 : errcallback.previous = error_context_stack;
1203 3385696 : error_context_stack = &errcallback;
1204 :
1205 : /* set output state */
1206 3385696 : ctx->accept_writes = false;
1207 3385696 : ctx->end_xact = false;
1208 :
1209 : /* do the actual work: call callback */
1210 3385696 : ret = ctx->callbacks.filter_by_origin_cb(ctx, origin_id);
1211 :
1212 : /* Pop the error context stack */
1213 3385696 : error_context_stack = errcallback.previous;
1214 :
1215 3385696 : return ret;
1216 : }
1217 :
1218 : static void
1219 32 : message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1220 : XLogRecPtr message_lsn, bool transactional,
1221 : const char *prefix, Size message_size, const char *message)
1222 : {
1223 32 : LogicalDecodingContext *ctx = cache->private_data;
1224 : LogicalErrorCallbackState state;
1225 : ErrorContextCallback errcallback;
1226 :
1227 : Assert(!ctx->fast_forward);
1228 :
1229 32 : if (ctx->callbacks.message_cb == NULL)
1230 0 : return;
1231 :
1232 : /* Push callback + info on the error context stack */
1233 32 : state.ctx = ctx;
1234 32 : state.callback_name = "message";
1235 32 : state.report_location = message_lsn;
1236 32 : errcallback.callback = output_plugin_error_callback;
1237 32 : errcallback.arg = &state;
1238 32 : errcallback.previous = error_context_stack;
1239 32 : error_context_stack = &errcallback;
1240 :
1241 : /* set output state */
1242 32 : ctx->accept_writes = true;
1243 32 : ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
1244 32 : ctx->write_location = message_lsn;
1245 32 : ctx->end_xact = false;
1246 :
1247 : /* do the actual work: call callback */
1248 32 : ctx->callbacks.message_cb(ctx, txn, message_lsn, transactional, prefix,
1249 : message_size, message);
1250 :
1251 : /* Pop the error context stack */
1252 32 : error_context_stack = errcallback.previous;
1253 : }
1254 :
1255 : static void
1256 1318 : stream_start_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1257 : XLogRecPtr first_lsn)
1258 : {
1259 1318 : LogicalDecodingContext *ctx = cache->private_data;
1260 : LogicalErrorCallbackState state;
1261 : ErrorContextCallback errcallback;
1262 :
1263 : Assert(!ctx->fast_forward);
1264 :
1265 : /* We're only supposed to call this when streaming is supported. */
1266 : Assert(ctx->streaming);
1267 :
1268 : /* Push callback + info on the error context stack */
1269 1318 : state.ctx = ctx;
1270 1318 : state.callback_name = "stream_start";
1271 1318 : state.report_location = first_lsn;
1272 1318 : errcallback.callback = output_plugin_error_callback;
1273 1318 : errcallback.arg = &state;
1274 1318 : errcallback.previous = error_context_stack;
1275 1318 : error_context_stack = &errcallback;
1276 :
1277 : /* set output state */
1278 1318 : ctx->accept_writes = true;
1279 1318 : ctx->write_xid = txn->xid;
1280 :
1281 : /*
1282 : * Report this message's lsn so replies from clients can give an
1283 : * up-to-date answer. This won't ever be enough (and shouldn't be!) to
1284 : * confirm receipt of this transaction, but it might allow another
1285 : * transaction's commit to be confirmed with one message.
1286 : */
1287 1318 : ctx->write_location = first_lsn;
1288 :
1289 1318 : ctx->end_xact = false;
1290 :
1291 : /* in streaming mode, stream_start_cb is required */
1292 1318 : if (ctx->callbacks.stream_start_cb == NULL)
1293 0 : ereport(ERROR,
1294 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1295 : errmsg("logical streaming requires a %s callback",
1296 : "stream_start_cb")));
1297 :
1298 1318 : ctx->callbacks.stream_start_cb(ctx, txn);
1299 :
1300 : /* Pop the error context stack */
1301 1318 : error_context_stack = errcallback.previous;
1302 1318 : }
1303 :
1304 : static void
1305 1318 : stream_stop_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1306 : XLogRecPtr last_lsn)
1307 : {
1308 1318 : LogicalDecodingContext *ctx = cache->private_data;
1309 : LogicalErrorCallbackState state;
1310 : ErrorContextCallback errcallback;
1311 :
1312 : Assert(!ctx->fast_forward);
1313 :
1314 : /* We're only supposed to call this when streaming is supported. */
1315 : Assert(ctx->streaming);
1316 :
1317 : /* Push callback + info on the error context stack */
1318 1318 : state.ctx = ctx;
1319 1318 : state.callback_name = "stream_stop";
1320 1318 : state.report_location = last_lsn;
1321 1318 : errcallback.callback = output_plugin_error_callback;
1322 1318 : errcallback.arg = &state;
1323 1318 : errcallback.previous = error_context_stack;
1324 1318 : error_context_stack = &errcallback;
1325 :
1326 : /* set output state */
1327 1318 : ctx->accept_writes = true;
1328 1318 : ctx->write_xid = txn->xid;
1329 :
1330 : /*
1331 : * Report this message's lsn so replies from clients can give an
1332 : * up-to-date answer. This won't ever be enough (and shouldn't be!) to
1333 : * confirm receipt of this transaction, but it might allow another
1334 : * transaction's commit to be confirmed with one message.
1335 : */
1336 1318 : ctx->write_location = last_lsn;
1337 :
1338 1318 : ctx->end_xact = false;
1339 :
1340 : /* in streaming mode, stream_stop_cb is required */
1341 1318 : if (ctx->callbacks.stream_stop_cb == NULL)
1342 0 : ereport(ERROR,
1343 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1344 : errmsg("logical streaming requires a %s callback",
1345 : "stream_stop_cb")));
1346 :
1347 1318 : ctx->callbacks.stream_stop_cb(ctx, txn);
1348 :
1349 : /* Pop the error context stack */
1350 1318 : error_context_stack = errcallback.previous;
1351 1318 : }
1352 :
1353 : static void
1354 60 : stream_abort_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1355 : XLogRecPtr abort_lsn)
1356 : {
1357 60 : LogicalDecodingContext *ctx = cache->private_data;
1358 : LogicalErrorCallbackState state;
1359 : ErrorContextCallback errcallback;
1360 :
1361 : Assert(!ctx->fast_forward);
1362 :
1363 : /* We're only supposed to call this when streaming is supported. */
1364 : Assert(ctx->streaming);
1365 :
1366 : /* Push callback + info on the error context stack */
1367 60 : state.ctx = ctx;
1368 60 : state.callback_name = "stream_abort";
1369 60 : state.report_location = abort_lsn;
1370 60 : errcallback.callback = output_plugin_error_callback;
1371 60 : errcallback.arg = &state;
1372 60 : errcallback.previous = error_context_stack;
1373 60 : error_context_stack = &errcallback;
1374 :
1375 : /* set output state */
1376 60 : ctx->accept_writes = true;
1377 60 : ctx->write_xid = txn->xid;
1378 60 : ctx->write_location = abort_lsn;
1379 60 : ctx->end_xact = true;
1380 :
1381 : /* in streaming mode, stream_abort_cb is required */
1382 60 : if (ctx->callbacks.stream_abort_cb == NULL)
1383 0 : ereport(ERROR,
1384 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1385 : errmsg("logical streaming requires a %s callback",
1386 : "stream_abort_cb")));
1387 :
1388 60 : ctx->callbacks.stream_abort_cb(ctx, txn, abort_lsn);
1389 :
1390 : /* Pop the error context stack */
1391 60 : error_context_stack = errcallback.previous;
1392 60 : }
1393 :
1394 : static void
1395 28 : stream_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1396 : XLogRecPtr prepare_lsn)
1397 : {
1398 28 : LogicalDecodingContext *ctx = cache->private_data;
1399 : LogicalErrorCallbackState state;
1400 : ErrorContextCallback errcallback;
1401 :
1402 : Assert(!ctx->fast_forward);
1403 :
1404 : /*
1405 : * We're only supposed to call this when streaming and two-phase commits
1406 : * are supported.
1407 : */
1408 : Assert(ctx->streaming);
1409 : Assert(ctx->twophase);
1410 :
1411 : /* Push callback + info on the error context stack */
1412 28 : state.ctx = ctx;
1413 28 : state.callback_name = "stream_prepare";
1414 28 : state.report_location = txn->final_lsn;
1415 28 : errcallback.callback = output_plugin_error_callback;
1416 28 : errcallback.arg = &state;
1417 28 : errcallback.previous = error_context_stack;
1418 28 : error_context_stack = &errcallback;
1419 :
1420 : /* set output state */
1421 28 : ctx->accept_writes = true;
1422 28 : ctx->write_xid = txn->xid;
1423 28 : ctx->write_location = txn->end_lsn;
1424 28 : ctx->end_xact = true;
1425 :
1426 : /* in streaming mode with two-phase commits, stream_prepare_cb is required */
1427 28 : if (ctx->callbacks.stream_prepare_cb == NULL)
1428 0 : ereport(ERROR,
1429 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1430 : errmsg("logical streaming at prepare time requires a %s callback",
1431 : "stream_prepare_cb")));
1432 :
1433 28 : ctx->callbacks.stream_prepare_cb(ctx, txn, prepare_lsn);
1434 :
1435 : /* Pop the error context stack */
1436 28 : error_context_stack = errcallback.previous;
1437 28 : }
1438 :
1439 : static void
1440 100 : stream_commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1441 : XLogRecPtr commit_lsn)
1442 : {
1443 100 : LogicalDecodingContext *ctx = cache->private_data;
1444 : LogicalErrorCallbackState state;
1445 : ErrorContextCallback errcallback;
1446 :
1447 : Assert(!ctx->fast_forward);
1448 :
1449 : /* We're only supposed to call this when streaming is supported. */
1450 : Assert(ctx->streaming);
1451 :
1452 : /* Push callback + info on the error context stack */
1453 100 : state.ctx = ctx;
1454 100 : state.callback_name = "stream_commit";
1455 100 : state.report_location = txn->final_lsn;
1456 100 : errcallback.callback = output_plugin_error_callback;
1457 100 : errcallback.arg = &state;
1458 100 : errcallback.previous = error_context_stack;
1459 100 : error_context_stack = &errcallback;
1460 :
1461 : /* set output state */
1462 100 : ctx->accept_writes = true;
1463 100 : ctx->write_xid = txn->xid;
1464 100 : ctx->write_location = txn->end_lsn;
1465 100 : ctx->end_xact = true;
1466 :
1467 : /* in streaming mode, stream_commit_cb is required */
1468 100 : if (ctx->callbacks.stream_commit_cb == NULL)
1469 0 : ereport(ERROR,
1470 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1471 : errmsg("logical streaming requires a %s callback",
1472 : "stream_commit_cb")));
1473 :
1474 100 : ctx->callbacks.stream_commit_cb(ctx, txn, commit_lsn);
1475 :
1476 : /* Pop the error context stack */
1477 100 : error_context_stack = errcallback.previous;
1478 100 : }
1479 :
1480 : static void
1481 351976 : stream_change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1482 : Relation relation, ReorderBufferChange *change)
1483 : {
1484 351976 : LogicalDecodingContext *ctx = cache->private_data;
1485 : LogicalErrorCallbackState state;
1486 : ErrorContextCallback errcallback;
1487 :
1488 : Assert(!ctx->fast_forward);
1489 :
1490 : /* We're only supposed to call this when streaming is supported. */
1491 : Assert(ctx->streaming);
1492 :
1493 : /* Push callback + info on the error context stack */
1494 351976 : state.ctx = ctx;
1495 351976 : state.callback_name = "stream_change";
1496 351976 : state.report_location = change->lsn;
1497 351976 : errcallback.callback = output_plugin_error_callback;
1498 351976 : errcallback.arg = &state;
1499 351976 : errcallback.previous = error_context_stack;
1500 351976 : error_context_stack = &errcallback;
1501 :
1502 : /* set output state */
1503 351976 : ctx->accept_writes = true;
1504 351976 : ctx->write_xid = txn->xid;
1505 :
1506 : /*
1507 : * Report this change's lsn so replies from clients can give an up-to-date
1508 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1509 : * receipt of this transaction, but it might allow another transaction's
1510 : * commit to be confirmed with one message.
1511 : */
1512 351976 : ctx->write_location = change->lsn;
1513 :
1514 351976 : ctx->end_xact = false;
1515 :
1516 : /* in streaming mode, stream_change_cb is required */
1517 351976 : if (ctx->callbacks.stream_change_cb == NULL)
1518 0 : ereport(ERROR,
1519 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1520 : errmsg("logical streaming requires a %s callback",
1521 : "stream_change_cb")));
1522 :
1523 351976 : ctx->callbacks.stream_change_cb(ctx, txn, relation, change);
1524 :
1525 : /* Pop the error context stack */
1526 351976 : error_context_stack = errcallback.previous;
1527 351976 : }
1528 :
1529 : static void
1530 6 : stream_message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1531 : XLogRecPtr message_lsn, bool transactional,
1532 : const char *prefix, Size message_size, const char *message)
1533 : {
1534 6 : LogicalDecodingContext *ctx = cache->private_data;
1535 : LogicalErrorCallbackState state;
1536 : ErrorContextCallback errcallback;
1537 :
1538 : Assert(!ctx->fast_forward);
1539 :
1540 : /* We're only supposed to call this when streaming is supported. */
1541 : Assert(ctx->streaming);
1542 :
1543 : /* this callback is optional */
1544 6 : if (ctx->callbacks.stream_message_cb == NULL)
1545 0 : return;
1546 :
1547 : /* Push callback + info on the error context stack */
1548 6 : state.ctx = ctx;
1549 6 : state.callback_name = "stream_message";
1550 6 : state.report_location = message_lsn;
1551 6 : errcallback.callback = output_plugin_error_callback;
1552 6 : errcallback.arg = &state;
1553 6 : errcallback.previous = error_context_stack;
1554 6 : error_context_stack = &errcallback;
1555 :
1556 : /* set output state */
1557 6 : ctx->accept_writes = true;
1558 6 : ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
1559 6 : ctx->write_location = message_lsn;
1560 6 : ctx->end_xact = false;
1561 :
1562 : /* do the actual work: call callback */
1563 6 : ctx->callbacks.stream_message_cb(ctx, txn, message_lsn, transactional, prefix,
1564 : message_size, message);
1565 :
1566 : /* Pop the error context stack */
1567 6 : error_context_stack = errcallback.previous;
1568 : }
1569 :
1570 : static void
1571 0 : stream_truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1572 : int nrelations, Relation relations[],
1573 : ReorderBufferChange *change)
1574 : {
1575 0 : LogicalDecodingContext *ctx = cache->private_data;
1576 : LogicalErrorCallbackState state;
1577 : ErrorContextCallback errcallback;
1578 :
1579 : Assert(!ctx->fast_forward);
1580 :
1581 : /* We're only supposed to call this when streaming is supported. */
1582 : Assert(ctx->streaming);
1583 :
1584 : /* this callback is optional */
1585 0 : if (!ctx->callbacks.stream_truncate_cb)
1586 0 : return;
1587 :
1588 : /* Push callback + info on the error context stack */
1589 0 : state.ctx = ctx;
1590 0 : state.callback_name = "stream_truncate";
1591 0 : state.report_location = change->lsn;
1592 0 : errcallback.callback = output_plugin_error_callback;
1593 0 : errcallback.arg = &state;
1594 0 : errcallback.previous = error_context_stack;
1595 0 : error_context_stack = &errcallback;
1596 :
1597 : /* set output state */
1598 0 : ctx->accept_writes = true;
1599 0 : ctx->write_xid = txn->xid;
1600 :
1601 : /*
1602 : * Report this change's lsn so replies from clients can give an up-to-date
1603 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1604 : * receipt of this transaction, but it might allow another transaction's
1605 : * commit to be confirmed with one message.
1606 : */
1607 0 : ctx->write_location = change->lsn;
1608 :
1609 0 : ctx->end_xact = false;
1610 :
1611 0 : ctx->callbacks.stream_truncate_cb(ctx, txn, nrelations, relations, change);
1612 :
1613 : /* Pop the error context stack */
1614 0 : error_context_stack = errcallback.previous;
1615 : }
1616 :
1617 : static void
1618 6216 : update_progress_txn_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1619 : XLogRecPtr lsn)
1620 : {
1621 6216 : LogicalDecodingContext *ctx = cache->private_data;
1622 : LogicalErrorCallbackState state;
1623 : ErrorContextCallback errcallback;
1624 :
1625 : Assert(!ctx->fast_forward);
1626 :
1627 : /* Push callback + info on the error context stack */
1628 6216 : state.ctx = ctx;
1629 6216 : state.callback_name = "update_progress_txn";
1630 6216 : state.report_location = lsn;
1631 6216 : errcallback.callback = output_plugin_error_callback;
1632 6216 : errcallback.arg = &state;
1633 6216 : errcallback.previous = error_context_stack;
1634 6216 : error_context_stack = &errcallback;
1635 :
1636 : /* set output state */
1637 6216 : ctx->accept_writes = false;
1638 6216 : ctx->write_xid = txn->xid;
1639 :
1640 : /*
1641 : * Report this change's lsn so replies from clients can give an up-to-date
1642 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1643 : * receipt of this transaction, but it might allow another transaction's
1644 : * commit to be confirmed with one message.
1645 : */
1646 6216 : ctx->write_location = lsn;
1647 :
1648 6216 : ctx->end_xact = false;
1649 :
1650 6216 : OutputPluginUpdateProgress(ctx, false);
1651 :
1652 : /* Pop the error context stack */
1653 6216 : error_context_stack = errcallback.previous;
1654 6216 : }
1655 :
1656 : /*
1657 : * Set the required catalog xmin horizon for historic snapshots in the current
1658 : * replication slot.
1659 : *
1660 : * Note that in the most cases, we won't be able to immediately use the xmin
1661 : * to increase the xmin horizon: we need to wait till the client has confirmed
1662 : * receiving current_lsn with LogicalConfirmReceivedLocation().
1663 : */
1664 : void
1665 846 : LogicalIncreaseXminForSlot(XLogRecPtr current_lsn, TransactionId xmin)
1666 : {
1667 846 : bool updated_xmin = false;
1668 : ReplicationSlot *slot;
1669 846 : bool got_new_xmin = false;
1670 :
1671 846 : slot = MyReplicationSlot;
1672 :
1673 : Assert(slot != NULL);
1674 :
1675 846 : SpinLockAcquire(&slot->mutex);
1676 :
1677 : /*
1678 : * don't overwrite if we already have a newer xmin. This can happen if we
1679 : * restart decoding in a slot.
1680 : */
1681 846 : if (TransactionIdPrecedesOrEquals(xmin, slot->data.catalog_xmin))
1682 : {
1683 : }
1684 :
1685 : /*
1686 : * If the client has already confirmed up to this lsn, we directly can
1687 : * mark this as accepted. This can happen if we restart decoding in a
1688 : * slot.
1689 : */
1690 218 : else if (current_lsn <= slot->data.confirmed_flush)
1691 : {
1692 104 : slot->candidate_catalog_xmin = xmin;
1693 104 : slot->candidate_xmin_lsn = current_lsn;
1694 :
1695 : /* our candidate can directly be used */
1696 104 : updated_xmin = true;
1697 : }
1698 :
1699 : /*
1700 : * Only increase if the previous values have been applied, otherwise we
1701 : * might never end up updating if the receiver acks too slowly.
1702 : */
1703 114 : else if (!XLogRecPtrIsValid(slot->candidate_xmin_lsn))
1704 : {
1705 44 : slot->candidate_catalog_xmin = xmin;
1706 44 : slot->candidate_xmin_lsn = current_lsn;
1707 :
1708 : /*
1709 : * Log new xmin at an appropriate log level after releasing the
1710 : * spinlock.
1711 : */
1712 44 : got_new_xmin = true;
1713 : }
1714 846 : SpinLockRelease(&slot->mutex);
1715 :
1716 846 : if (got_new_xmin)
1717 44 : elog(DEBUG1, "got new catalog xmin %u at %X/%08X", xmin,
1718 : LSN_FORMAT_ARGS(current_lsn));
1719 :
1720 : /* candidate already valid with the current flush position, apply */
1721 846 : if (updated_xmin)
1722 104 : LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
1723 846 : }
1724 :
1725 : /*
1726 : * Mark the minimal LSN (restart_lsn) we need to read to replay all
1727 : * transactions that have not yet committed at current_lsn.
1728 : *
1729 : * Just like LogicalIncreaseXminForSlot this only takes effect when the
1730 : * client has confirmed to have received current_lsn.
1731 : */
1732 : void
1733 738 : LogicalIncreaseRestartDecodingForSlot(XLogRecPtr current_lsn, XLogRecPtr restart_lsn)
1734 : {
1735 738 : bool updated_lsn = false;
1736 : ReplicationSlot *slot;
1737 :
1738 738 : slot = MyReplicationSlot;
1739 :
1740 : Assert(slot != NULL);
1741 : Assert(XLogRecPtrIsValid(restart_lsn));
1742 : Assert(XLogRecPtrIsValid(current_lsn));
1743 :
1744 738 : SpinLockAcquire(&slot->mutex);
1745 :
1746 : /* don't overwrite if have a newer restart lsn */
1747 738 : if (restart_lsn <= slot->data.restart_lsn)
1748 : {
1749 20 : SpinLockRelease(&slot->mutex);
1750 : }
1751 :
1752 : /*
1753 : * We might have already flushed far enough to directly accept this lsn,
1754 : * in this case there is no need to check for existing candidate LSNs
1755 : */
1756 718 : else if (current_lsn <= slot->data.confirmed_flush)
1757 : {
1758 548 : slot->candidate_restart_valid = current_lsn;
1759 548 : slot->candidate_restart_lsn = restart_lsn;
1760 548 : SpinLockRelease(&slot->mutex);
1761 :
1762 : /* our candidate can directly be used */
1763 548 : updated_lsn = true;
1764 : }
1765 :
1766 : /*
1767 : * Only increase if the previous values have been applied, otherwise we
1768 : * might never end up updating if the receiver acks too slowly. A missed
1769 : * value here will just cause some extra effort after reconnecting.
1770 : */
1771 170 : else if (!XLogRecPtrIsValid(slot->candidate_restart_valid))
1772 : {
1773 80 : slot->candidate_restart_valid = current_lsn;
1774 80 : slot->candidate_restart_lsn = restart_lsn;
1775 80 : SpinLockRelease(&slot->mutex);
1776 :
1777 80 : elog(DEBUG1, "got new restart lsn %X/%08X at %X/%08X",
1778 : LSN_FORMAT_ARGS(restart_lsn),
1779 : LSN_FORMAT_ARGS(current_lsn));
1780 : }
1781 : else
1782 : {
1783 : XLogRecPtr candidate_restart_lsn;
1784 : XLogRecPtr candidate_restart_valid;
1785 : XLogRecPtr confirmed_flush;
1786 :
1787 90 : candidate_restart_lsn = slot->candidate_restart_lsn;
1788 90 : candidate_restart_valid = slot->candidate_restart_valid;
1789 90 : confirmed_flush = slot->data.confirmed_flush;
1790 90 : SpinLockRelease(&slot->mutex);
1791 :
1792 90 : 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",
1793 : LSN_FORMAT_ARGS(restart_lsn),
1794 : LSN_FORMAT_ARGS(current_lsn),
1795 : LSN_FORMAT_ARGS(candidate_restart_lsn),
1796 : LSN_FORMAT_ARGS(candidate_restart_valid),
1797 : LSN_FORMAT_ARGS(confirmed_flush));
1798 : }
1799 :
1800 : /* candidates are already valid with the current flush position, apply */
1801 738 : if (updated_lsn)
1802 548 : LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
1803 738 : }
1804 :
1805 : /*
1806 : * Handle a consumer's confirmation having received all changes up to lsn.
1807 : */
1808 : void
1809 107058 : LogicalConfirmReceivedLocation(XLogRecPtr lsn)
1810 : {
1811 : Assert(XLogRecPtrIsValid(lsn));
1812 :
1813 : /* Do an unlocked check for candidate_lsn first. */
1814 107058 : if (XLogRecPtrIsValid(MyReplicationSlot->candidate_xmin_lsn) ||
1815 106722 : XLogRecPtrIsValid(MyReplicationSlot->candidate_restart_valid))
1816 926 : {
1817 926 : bool updated_xmin = false;
1818 926 : bool updated_restart = false;
1819 : XLogRecPtr restart_lsn pg_attribute_unused();
1820 :
1821 926 : SpinLockAcquire(&MyReplicationSlot->mutex);
1822 :
1823 : /* remember the old restart lsn */
1824 926 : restart_lsn = MyReplicationSlot->data.restart_lsn;
1825 :
1826 : /*
1827 : * Prevent moving the confirmed_flush backwards, as this could lead to
1828 : * data duplication issues caused by replicating already replicated
1829 : * changes.
1830 : *
1831 : * This can happen when a client acknowledges an LSN it doesn't have
1832 : * to do anything for, and thus didn't store persistently. After a
1833 : * restart, the client can send the prior LSN that it stored
1834 : * persistently as an acknowledgement, but we need to ignore such an
1835 : * LSN. See similar case handling in CreateDecodingContext.
1836 : */
1837 926 : if (lsn > MyReplicationSlot->data.confirmed_flush)
1838 72 : MyReplicationSlot->data.confirmed_flush = lsn;
1839 :
1840 : /* if we're past the location required for bumping xmin, do so */
1841 926 : if (XLogRecPtrIsValid(MyReplicationSlot->candidate_xmin_lsn) &&
1842 336 : MyReplicationSlot->candidate_xmin_lsn <= lsn)
1843 : {
1844 : /*
1845 : * We have to write the changed xmin to disk *before* we change
1846 : * the in-memory value, otherwise after a crash we wouldn't know
1847 : * that some catalog tuples might have been removed already.
1848 : *
1849 : * Ensure that by first writing to ->xmin and only update
1850 : * ->effective_xmin once the new state is synced to disk. After a
1851 : * crash ->effective_xmin is set to ->xmin.
1852 : */
1853 136 : if (TransactionIdIsValid(MyReplicationSlot->candidate_catalog_xmin) &&
1854 136 : MyReplicationSlot->data.catalog_xmin != MyReplicationSlot->candidate_catalog_xmin)
1855 : {
1856 136 : MyReplicationSlot->data.catalog_xmin = MyReplicationSlot->candidate_catalog_xmin;
1857 136 : MyReplicationSlot->candidate_catalog_xmin = InvalidTransactionId;
1858 136 : MyReplicationSlot->candidate_xmin_lsn = InvalidXLogRecPtr;
1859 136 : updated_xmin = true;
1860 : }
1861 : }
1862 :
1863 926 : if (XLogRecPtrIsValid(MyReplicationSlot->candidate_restart_valid) &&
1864 822 : MyReplicationSlot->candidate_restart_valid <= lsn)
1865 : {
1866 : Assert(XLogRecPtrIsValid(MyReplicationSlot->candidate_restart_lsn));
1867 :
1868 620 : MyReplicationSlot->data.restart_lsn = MyReplicationSlot->candidate_restart_lsn;
1869 620 : MyReplicationSlot->candidate_restart_lsn = InvalidXLogRecPtr;
1870 620 : MyReplicationSlot->candidate_restart_valid = InvalidXLogRecPtr;
1871 620 : updated_restart = true;
1872 : }
1873 :
1874 926 : SpinLockRelease(&MyReplicationSlot->mutex);
1875 :
1876 : /* first write new xmin to disk, so we know what's up after a crash */
1877 926 : if (updated_xmin || updated_restart)
1878 : {
1879 : #ifdef USE_INJECTION_POINTS
1880 : XLogSegNo seg1,
1881 : seg2;
1882 :
1883 724 : XLByteToSeg(restart_lsn, seg1, wal_segment_size);
1884 724 : XLByteToSeg(MyReplicationSlot->data.restart_lsn, seg2, wal_segment_size);
1885 :
1886 : /* trigger injection point, but only if segment changes */
1887 724 : if (seg1 != seg2)
1888 10 : INJECTION_POINT("logical-replication-slot-advance-segment", NULL);
1889 : #endif
1890 :
1891 724 : ReplicationSlotMarkDirty();
1892 724 : ReplicationSlotSave();
1893 724 : elog(DEBUG1, "updated xmin: %u restart: %u", updated_xmin, updated_restart);
1894 : }
1895 :
1896 : /*
1897 : * Now the new xmin is safely on disk, we can let the global value
1898 : * advance. We do not take ProcArrayLock or similar since we only
1899 : * advance xmin here and there's not much harm done by a concurrent
1900 : * computation missing that.
1901 : */
1902 926 : if (updated_xmin)
1903 : {
1904 136 : SpinLockAcquire(&MyReplicationSlot->mutex);
1905 136 : MyReplicationSlot->effective_catalog_xmin = MyReplicationSlot->data.catalog_xmin;
1906 136 : SpinLockRelease(&MyReplicationSlot->mutex);
1907 :
1908 136 : ReplicationSlotsComputeRequiredXmin(false);
1909 136 : ReplicationSlotsComputeRequiredLSN();
1910 : }
1911 : }
1912 : else
1913 : {
1914 106132 : SpinLockAcquire(&MyReplicationSlot->mutex);
1915 :
1916 : /*
1917 : * Prevent moving the confirmed_flush backwards. See comments above
1918 : * for the details.
1919 : */
1920 106132 : if (lsn > MyReplicationSlot->data.confirmed_flush)
1921 104602 : MyReplicationSlot->data.confirmed_flush = lsn;
1922 :
1923 106132 : SpinLockRelease(&MyReplicationSlot->mutex);
1924 : }
1925 107058 : }
1926 :
1927 : /*
1928 : * Clear logical streaming state during (sub)transaction abort.
1929 : */
1930 : void
1931 61440 : ResetLogicalStreamingState(void)
1932 : {
1933 61440 : CheckXidAlive = InvalidTransactionId;
1934 61440 : bsysscan = false;
1935 61440 : }
1936 :
1937 : /*
1938 : * Report stats for a slot.
1939 : */
1940 : void
1941 13494 : UpdateDecodingStats(LogicalDecodingContext *ctx)
1942 : {
1943 13494 : ReorderBuffer *rb = ctx->reorder;
1944 : PgStat_StatReplSlotEntry repSlotStat;
1945 :
1946 : /* Nothing to do if we don't have any replication stats to be sent. */
1947 13494 : if (rb->spillBytes <= 0 && rb->streamBytes <= 0 && rb->totalBytes <= 0 &&
1948 656 : rb->memExceededCount <= 0)
1949 638 : return;
1950 :
1951 12856 : elog(DEBUG2, "UpdateDecodingStats: updating stats %p %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64,
1952 : rb,
1953 : rb->spillTxns,
1954 : rb->spillCount,
1955 : rb->spillBytes,
1956 : rb->streamTxns,
1957 : rb->streamCount,
1958 : rb->streamBytes,
1959 : rb->memExceededCount,
1960 : rb->totalTxns,
1961 : rb->totalBytes);
1962 :
1963 12856 : repSlotStat.spill_txns = rb->spillTxns;
1964 12856 : repSlotStat.spill_count = rb->spillCount;
1965 12856 : repSlotStat.spill_bytes = rb->spillBytes;
1966 12856 : repSlotStat.stream_txns = rb->streamTxns;
1967 12856 : repSlotStat.stream_count = rb->streamCount;
1968 12856 : repSlotStat.stream_bytes = rb->streamBytes;
1969 12856 : repSlotStat.mem_exceeded_count = rb->memExceededCount;
1970 12856 : repSlotStat.total_txns = rb->totalTxns;
1971 12856 : repSlotStat.total_bytes = rb->totalBytes;
1972 :
1973 12856 : pgstat_report_replslot(ctx->slot, &repSlotStat);
1974 :
1975 12856 : rb->spillTxns = 0;
1976 12856 : rb->spillCount = 0;
1977 12856 : rb->spillBytes = 0;
1978 12856 : rb->streamTxns = 0;
1979 12856 : rb->streamCount = 0;
1980 12856 : rb->streamBytes = 0;
1981 12856 : rb->memExceededCount = 0;
1982 12856 : rb->totalTxns = 0;
1983 12856 : rb->totalBytes = 0;
1984 : }
1985 :
1986 : /*
1987 : * Read up to the end of WAL starting from the decoding slot's restart_lsn.
1988 : * Return true if any meaningful/decodable WAL records are encountered,
1989 : * otherwise false.
1990 : */
1991 : bool
1992 10 : LogicalReplicationSlotHasPendingWal(XLogRecPtr end_of_wal)
1993 : {
1994 10 : bool has_pending_wal = false;
1995 :
1996 : Assert(MyReplicationSlot);
1997 :
1998 10 : PG_TRY();
1999 : {
2000 : LogicalDecodingContext *ctx;
2001 :
2002 : /*
2003 : * Create our decoding context in fast_forward mode, passing start_lsn
2004 : * as InvalidXLogRecPtr, so that we start processing from the slot's
2005 : * confirmed_flush.
2006 : */
2007 20 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
2008 : NIL,
2009 : true, /* fast_forward */
2010 10 : XL_ROUTINE(.page_read = read_local_xlog_page,
2011 : .segment_open = wal_segment_open,
2012 : .segment_close = wal_segment_close),
2013 : NULL, NULL, NULL);
2014 :
2015 : /*
2016 : * Start reading at the slot's restart_lsn, which we know points to a
2017 : * valid record.
2018 : */
2019 10 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
2020 :
2021 : /* Invalidate non-timetravel entries */
2022 10 : InvalidateSystemCaches();
2023 :
2024 : /* Loop until the end of WAL or some changes are processed */
2025 296 : while (!has_pending_wal && ctx->reader->EndRecPtr < end_of_wal)
2026 : {
2027 : XLogRecord *record;
2028 286 : char *errm = NULL;
2029 :
2030 286 : record = XLogReadRecord(ctx->reader, &errm);
2031 :
2032 286 : if (errm)
2033 0 : elog(ERROR, "could not find record for logical decoding: %s", errm);
2034 :
2035 286 : if (record != NULL)
2036 286 : LogicalDecodingProcessRecord(ctx, ctx->reader);
2037 :
2038 286 : has_pending_wal = ctx->processing_required;
2039 :
2040 286 : CHECK_FOR_INTERRUPTS();
2041 : }
2042 :
2043 : /* Clean up */
2044 10 : FreeDecodingContext(ctx);
2045 10 : InvalidateSystemCaches();
2046 : }
2047 0 : PG_CATCH();
2048 : {
2049 : /* clear all timetravel entries */
2050 0 : InvalidateSystemCaches();
2051 :
2052 0 : PG_RE_THROW();
2053 : }
2054 10 : PG_END_TRY();
2055 :
2056 10 : return has_pending_wal;
2057 : }
2058 :
2059 : /*
2060 : * Helper function for advancing our logical replication slot forward.
2061 : *
2062 : * The slot's restart_lsn is used as start point for reading records, while
2063 : * confirmed_flush is used as base point for the decoding context.
2064 : *
2065 : * We cannot just do LogicalConfirmReceivedLocation to update confirmed_flush,
2066 : * because we need to digest WAL to advance restart_lsn allowing to recycle
2067 : * WAL and removal of old catalog tuples. As decoding is done in fast_forward
2068 : * mode, no changes are generated anyway.
2069 : *
2070 : * *found_consistent_snapshot will be true if the initial decoding snapshot has
2071 : * been built; Otherwise, it will be false.
2072 : */
2073 : XLogRecPtr
2074 36 : LogicalSlotAdvanceAndCheckSnapState(XLogRecPtr moveto,
2075 : bool *found_consistent_snapshot)
2076 : {
2077 : LogicalDecodingContext *ctx;
2078 36 : ResourceOwner old_resowner PG_USED_FOR_ASSERTS_ONLY = CurrentResourceOwner;
2079 : XLogRecPtr retlsn;
2080 :
2081 : Assert(XLogRecPtrIsValid(moveto));
2082 :
2083 36 : if (found_consistent_snapshot)
2084 12 : *found_consistent_snapshot = false;
2085 :
2086 36 : PG_TRY();
2087 : {
2088 : /*
2089 : * Create our decoding context in fast_forward mode, passing start_lsn
2090 : * as InvalidXLogRecPtr, so that we start processing from my slot's
2091 : * confirmed_flush.
2092 : */
2093 72 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
2094 : NIL,
2095 : true, /* fast_forward */
2096 36 : XL_ROUTINE(.page_read = read_local_xlog_page,
2097 : .segment_open = wal_segment_open,
2098 : .segment_close = wal_segment_close),
2099 : NULL, NULL, NULL);
2100 :
2101 : /*
2102 : * Wait for specified streaming replication standby servers (if any)
2103 : * to confirm receipt of WAL up to moveto lsn.
2104 : */
2105 36 : WaitForStandbyConfirmation(moveto);
2106 :
2107 : /*
2108 : * Start reading at the slot's restart_lsn, which we know to point to
2109 : * a valid record.
2110 : */
2111 36 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
2112 :
2113 : /* invalidate non-timetravel entries */
2114 36 : InvalidateSystemCaches();
2115 :
2116 : /* Decode records until we reach the requested target */
2117 4268 : while (ctx->reader->EndRecPtr < moveto)
2118 : {
2119 4232 : char *errm = NULL;
2120 : XLogRecord *record;
2121 :
2122 : /*
2123 : * Read records. No changes are generated in fast_forward mode,
2124 : * but snapbuilder/slot statuses are updated properly.
2125 : */
2126 4232 : record = XLogReadRecord(ctx->reader, &errm);
2127 4232 : if (errm)
2128 0 : elog(ERROR, "could not find record while advancing replication slot: %s",
2129 : errm);
2130 :
2131 : /*
2132 : * Process the record. Storage-level changes are ignored in
2133 : * fast_forward mode, but other modules (such as snapbuilder)
2134 : * might still have critical updates to do.
2135 : */
2136 4232 : if (record)
2137 : {
2138 4232 : LogicalDecodingProcessRecord(ctx, ctx->reader);
2139 :
2140 : /*
2141 : * We used to have bugs where logical decoding would fail to
2142 : * preserve the resource owner. That's important here, so
2143 : * verify that that doesn't happen anymore. XXX this could be
2144 : * removed once it's been battle-tested.
2145 : */
2146 : Assert(CurrentResourceOwner == old_resowner);
2147 : }
2148 :
2149 4232 : CHECK_FOR_INTERRUPTS();
2150 : }
2151 :
2152 36 : if (found_consistent_snapshot && DecodingContextReady(ctx))
2153 12 : *found_consistent_snapshot = true;
2154 :
2155 36 : if (XLogRecPtrIsValid(ctx->reader->EndRecPtr))
2156 : {
2157 36 : LogicalConfirmReceivedLocation(moveto);
2158 :
2159 : /*
2160 : * If only the confirmed_flush LSN has changed the slot won't get
2161 : * marked as dirty by the above. Callers on the walsender
2162 : * interface are expected to keep track of their own progress and
2163 : * don't need it written out. But SQL-interface users cannot
2164 : * specify their own start positions and it's harder for them to
2165 : * keep track of their progress, so we should make more of an
2166 : * effort to save it for them.
2167 : *
2168 : * Dirty the slot so it is written out at the next checkpoint. The
2169 : * LSN position advanced to may still be lost on a crash but this
2170 : * makes the data consistent after a clean shutdown.
2171 : */
2172 36 : ReplicationSlotMarkDirty();
2173 : }
2174 :
2175 36 : retlsn = MyReplicationSlot->data.confirmed_flush;
2176 :
2177 : /* free context, call shutdown callback */
2178 36 : FreeDecodingContext(ctx);
2179 :
2180 36 : InvalidateSystemCaches();
2181 : }
2182 0 : PG_CATCH();
2183 : {
2184 : /* clear all timetravel entries */
2185 0 : InvalidateSystemCaches();
2186 :
2187 0 : PG_RE_THROW();
2188 : }
2189 36 : PG_END_TRY();
2190 :
2191 36 : return retlsn;
2192 : }
|
