Line data Source code
1 : /*-------------------------------------------------------------------------
2 : * logical.c
3 : * PostgreSQL logical decoding coordination
4 : *
5 : * Copyright (c) 2012-2024, 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/xlogutils.h"
33 : #include "fmgr.h"
34 : #include "miscadmin.h"
35 : #include "pgstat.h"
36 : #include "replication/decode.h"
37 : #include "replication/logical.h"
38 : #include "replication/reorderbuffer.h"
39 : #include "replication/slotsync.h"
40 : #include "replication/snapbuild.h"
41 : #include "storage/proc.h"
42 : #include "storage/procarray.h"
43 : #include "utils/builtins.h"
44 : #include "utils/inval.h"
45 : #include "utils/memutils.h"
46 :
47 : /* data for errcontext callback */
48 : typedef struct LogicalErrorCallbackState
49 : {
50 : LogicalDecodingContext *ctx;
51 : const char *callback_name;
52 : XLogRecPtr report_location;
53 : } LogicalErrorCallbackState;
54 :
55 : /* wrappers around output plugin callbacks */
56 : static void output_plugin_error_callback(void *arg);
57 : static void startup_cb_wrapper(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
58 : bool is_init);
59 : static void shutdown_cb_wrapper(LogicalDecodingContext *ctx);
60 : static void begin_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn);
61 : static void commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
62 : XLogRecPtr commit_lsn);
63 : static void begin_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn);
64 : static void prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
65 : XLogRecPtr prepare_lsn);
66 : static void commit_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
67 : XLogRecPtr commit_lsn);
68 : static void rollback_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
69 : XLogRecPtr prepare_end_lsn, TimestampTz prepare_time);
70 : static void change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
71 : Relation relation, ReorderBufferChange *change);
72 : static void truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
73 : int nrelations, Relation relations[], ReorderBufferChange *change);
74 : static void message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
75 : XLogRecPtr message_lsn, bool transactional,
76 : const char *prefix, Size message_size, const char *message);
77 :
78 : /* streaming callbacks */
79 : static void stream_start_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
80 : XLogRecPtr first_lsn);
81 : static void stream_stop_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
82 : XLogRecPtr last_lsn);
83 : static void stream_abort_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
84 : XLogRecPtr abort_lsn);
85 : static void stream_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
86 : XLogRecPtr prepare_lsn);
87 : static void stream_commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
88 : XLogRecPtr commit_lsn);
89 : static void stream_change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
90 : Relation relation, ReorderBufferChange *change);
91 : static void stream_message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
92 : XLogRecPtr message_lsn, bool transactional,
93 : const char *prefix, Size message_size, const char *message);
94 : static void stream_truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
95 : int nrelations, Relation relations[], ReorderBufferChange *change);
96 :
97 : /* callback to update txn's progress */
98 : static void update_progress_txn_cb_wrapper(ReorderBuffer *cache,
99 : ReorderBufferTXN *txn,
100 : XLogRecPtr lsn);
101 :
102 : static void LoadOutputPlugin(OutputPluginCallbacks *callbacks, const char *plugin);
103 :
104 : /*
105 : * Make sure the current settings & environment are capable of doing logical
106 : * decoding.
107 : */
108 : void
109 2666 : CheckLogicalDecodingRequirements(void)
110 : {
111 2666 : CheckSlotRequirements();
112 :
113 : /*
114 : * NB: Adding a new requirement likely means that RestoreSlotFromDisk()
115 : * needs the same check.
116 : */
117 :
118 2666 : if (wal_level < WAL_LEVEL_LOGICAL)
119 0 : ereport(ERROR,
120 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
121 : errmsg("logical decoding requires wal_level >= logical")));
122 :
123 2666 : if (MyDatabaseId == InvalidOid)
124 2 : ereport(ERROR,
125 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
126 : errmsg("logical decoding requires a database connection")));
127 :
128 2664 : if (RecoveryInProgress())
129 : {
130 : /*
131 : * This check may have race conditions, but whenever
132 : * XLOG_PARAMETER_CHANGE indicates that wal_level has changed, we
133 : * verify that there are no existing logical replication slots. And to
134 : * avoid races around creating a new slot,
135 : * CheckLogicalDecodingRequirements() is called once before creating
136 : * the slot, and once when logical decoding is initially starting up.
137 : */
138 132 : if (GetActiveWalLevelOnStandby() < WAL_LEVEL_LOGICAL)
139 2 : ereport(ERROR,
140 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
141 : errmsg("logical decoding on standby requires wal_level >= logical on the primary")));
142 : }
143 2662 : }
144 :
145 : /*
146 : * Helper function for CreateInitDecodingContext() and
147 : * CreateDecodingContext() performing common tasks.
148 : */
149 : static LogicalDecodingContext *
150 1860 : StartupDecodingContext(List *output_plugin_options,
151 : XLogRecPtr start_lsn,
152 : TransactionId xmin_horizon,
153 : bool need_full_snapshot,
154 : bool fast_forward,
155 : XLogReaderRoutine *xl_routine,
156 : LogicalOutputPluginWriterPrepareWrite prepare_write,
157 : LogicalOutputPluginWriterWrite do_write,
158 : LogicalOutputPluginWriterUpdateProgress update_progress)
159 : {
160 : ReplicationSlot *slot;
161 : MemoryContext context,
162 : old_context;
163 : LogicalDecodingContext *ctx;
164 :
165 : /* shorter lines... */
166 1860 : slot = MyReplicationSlot;
167 :
168 1860 : context = AllocSetContextCreate(CurrentMemoryContext,
169 : "Logical decoding context",
170 : ALLOCSET_DEFAULT_SIZES);
171 1860 : old_context = MemoryContextSwitchTo(context);
172 1860 : ctx = palloc0(sizeof(LogicalDecodingContext));
173 :
174 1860 : ctx->context = context;
175 :
176 : /*
177 : * (re-)load output plugins, so we detect a bad (removed) output plugin
178 : * now.
179 : */
180 1860 : if (!fast_forward)
181 1824 : LoadOutputPlugin(&ctx->callbacks, NameStr(slot->data.plugin));
182 :
183 : /*
184 : * Now that the slot's xmin has been set, we can announce ourselves as a
185 : * logical decoding backend which doesn't need to be checked individually
186 : * when computing the xmin horizon because the xmin is enforced via
187 : * replication slots.
188 : *
189 : * We can only do so if we're outside of a transaction (i.e. the case when
190 : * streaming changes via walsender), otherwise an already setup
191 : * snapshot/xid would end up being ignored. That's not a particularly
192 : * bothersome restriction since the SQL interface can't be used for
193 : * streaming anyway.
194 : */
195 1858 : if (!IsTransactionOrTransactionBlock())
196 : {
197 880 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
198 880 : MyProc->statusFlags |= PROC_IN_LOGICAL_DECODING;
199 880 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
200 880 : LWLockRelease(ProcArrayLock);
201 : }
202 :
203 1858 : ctx->slot = slot;
204 :
205 1858 : ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
206 1858 : if (!ctx->reader)
207 0 : ereport(ERROR,
208 : (errcode(ERRCODE_OUT_OF_MEMORY),
209 : errmsg("out of memory"),
210 : errdetail("Failed while allocating a WAL reading processor.")));
211 :
212 1858 : ctx->reorder = ReorderBufferAllocate();
213 1858 : ctx->snapshot_builder =
214 1858 : AllocateSnapshotBuilder(ctx->reorder, xmin_horizon, start_lsn,
215 : need_full_snapshot, slot->data.two_phase_at);
216 :
217 1858 : ctx->reorder->private_data = ctx;
218 :
219 : /* wrap output plugin callbacks, so we can add error context information */
220 1858 : ctx->reorder->begin = begin_cb_wrapper;
221 1858 : ctx->reorder->apply_change = change_cb_wrapper;
222 1858 : ctx->reorder->apply_truncate = truncate_cb_wrapper;
223 1858 : ctx->reorder->commit = commit_cb_wrapper;
224 1858 : ctx->reorder->message = message_cb_wrapper;
225 :
226 : /*
227 : * To support streaming, we require start/stop/abort/commit/change
228 : * callbacks. The message and truncate callbacks are optional, similar to
229 : * regular output plugins. We however enable streaming when at least one
230 : * of the methods is enabled so that we can easily identify missing
231 : * methods.
232 : *
233 : * We decide it here, but only check it later in the wrappers.
234 : */
235 3752 : ctx->streaming = (ctx->callbacks.stream_start_cb != NULL) ||
236 36 : (ctx->callbacks.stream_stop_cb != NULL) ||
237 36 : (ctx->callbacks.stream_abort_cb != NULL) ||
238 36 : (ctx->callbacks.stream_commit_cb != NULL) ||
239 36 : (ctx->callbacks.stream_change_cb != NULL) ||
240 1930 : (ctx->callbacks.stream_message_cb != NULL) ||
241 36 : (ctx->callbacks.stream_truncate_cb != NULL);
242 :
243 : /*
244 : * streaming callbacks
245 : *
246 : * stream_message and stream_truncate callbacks are optional, so we do not
247 : * fail with ERROR when missing, but the wrappers simply do nothing. We
248 : * must set the ReorderBuffer callbacks to something, otherwise the calls
249 : * from there will crash (we don't want to move the checks there).
250 : */
251 1858 : ctx->reorder->stream_start = stream_start_cb_wrapper;
252 1858 : ctx->reorder->stream_stop = stream_stop_cb_wrapper;
253 1858 : ctx->reorder->stream_abort = stream_abort_cb_wrapper;
254 1858 : ctx->reorder->stream_prepare = stream_prepare_cb_wrapper;
255 1858 : ctx->reorder->stream_commit = stream_commit_cb_wrapper;
256 1858 : ctx->reorder->stream_change = stream_change_cb_wrapper;
257 1858 : ctx->reorder->stream_message = stream_message_cb_wrapper;
258 1858 : ctx->reorder->stream_truncate = stream_truncate_cb_wrapper;
259 :
260 :
261 : /*
262 : * To support two-phase logical decoding, we require
263 : * begin_prepare/prepare/commit-prepare/abort-prepare callbacks. The
264 : * filter_prepare callback is optional. We however enable two-phase
265 : * logical decoding when at least one of the methods is enabled so that we
266 : * can easily identify missing methods.
267 : *
268 : * We decide it here, but only check it later in the wrappers.
269 : */
270 3752 : ctx->twophase = (ctx->callbacks.begin_prepare_cb != NULL) ||
271 36 : (ctx->callbacks.prepare_cb != NULL) ||
272 36 : (ctx->callbacks.commit_prepared_cb != NULL) ||
273 36 : (ctx->callbacks.rollback_prepared_cb != NULL) ||
274 1930 : (ctx->callbacks.stream_prepare_cb != NULL) ||
275 36 : (ctx->callbacks.filter_prepare_cb != NULL);
276 :
277 : /*
278 : * Callback to support decoding at prepare time.
279 : */
280 1858 : ctx->reorder->begin_prepare = begin_prepare_cb_wrapper;
281 1858 : ctx->reorder->prepare = prepare_cb_wrapper;
282 1858 : ctx->reorder->commit_prepared = commit_prepared_cb_wrapper;
283 1858 : ctx->reorder->rollback_prepared = rollback_prepared_cb_wrapper;
284 :
285 : /*
286 : * Callback to support updating progress during sending data of a
287 : * transaction (and its subtransactions) to the output plugin.
288 : */
289 1858 : ctx->reorder->update_progress_txn = update_progress_txn_cb_wrapper;
290 :
291 1858 : ctx->out = makeStringInfo();
292 1858 : ctx->prepare_write = prepare_write;
293 1858 : ctx->write = do_write;
294 1858 : ctx->update_progress = update_progress;
295 :
296 1858 : ctx->output_plugin_options = output_plugin_options;
297 :
298 1858 : ctx->fast_forward = fast_forward;
299 :
300 1858 : MemoryContextSwitchTo(old_context);
301 :
302 1858 : return ctx;
303 : }
304 :
305 : /*
306 : * Create a new decoding context, for a new logical slot.
307 : *
308 : * plugin -- contains the name of the output plugin
309 : * output_plugin_options -- contains options passed to the output plugin
310 : * need_full_snapshot -- if true, must obtain a snapshot able to read all
311 : * tables; if false, one that can read only catalogs is acceptable.
312 : * restart_lsn -- if given as invalid, it's this routine's responsibility to
313 : * mark WAL as reserved by setting a convenient restart_lsn for the slot.
314 : * Otherwise, we set for decoding to start from the given LSN without
315 : * marking WAL reserved beforehand. In that scenario, it's up to the
316 : * caller to guarantee that WAL remains available.
317 : * xl_routine -- XLogReaderRoutine for underlying XLogReader
318 : * prepare_write, do_write, update_progress --
319 : * callbacks that perform the use-case dependent, actual, work.
320 : *
321 : * Needs to be called while in a memory context that's at least as long lived
322 : * as the decoding context because further memory contexts will be created
323 : * inside it.
324 : *
325 : * Returns an initialized decoding context after calling the output plugin's
326 : * startup function.
327 : */
328 : LogicalDecodingContext *
329 800 : CreateInitDecodingContext(const char *plugin,
330 : List *output_plugin_options,
331 : bool need_full_snapshot,
332 : XLogRecPtr restart_lsn,
333 : XLogReaderRoutine *xl_routine,
334 : LogicalOutputPluginWriterPrepareWrite prepare_write,
335 : LogicalOutputPluginWriterWrite do_write,
336 : LogicalOutputPluginWriterUpdateProgress update_progress)
337 : {
338 800 : TransactionId xmin_horizon = InvalidTransactionId;
339 : ReplicationSlot *slot;
340 : NameData plugin_name;
341 : LogicalDecodingContext *ctx;
342 : MemoryContext old_context;
343 :
344 : /*
345 : * On a standby, this check is also required while creating the slot.
346 : * Check the comments in the function.
347 : */
348 800 : CheckLogicalDecodingRequirements();
349 :
350 : /* shorter lines... */
351 800 : slot = MyReplicationSlot;
352 :
353 : /* first some sanity checks that are unlikely to be violated */
354 800 : if (slot == NULL)
355 0 : elog(ERROR, "cannot perform logical decoding without an acquired slot");
356 :
357 800 : if (plugin == NULL)
358 0 : elog(ERROR, "cannot initialize logical decoding without a specified plugin");
359 :
360 : /* Make sure the passed slot is suitable. These are user facing errors. */
361 800 : if (SlotIsPhysical(slot))
362 0 : ereport(ERROR,
363 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
364 : errmsg("cannot use physical replication slot for logical decoding")));
365 :
366 800 : if (slot->data.database != MyDatabaseId)
367 0 : ereport(ERROR,
368 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
369 : errmsg("replication slot \"%s\" was not created in this database",
370 : NameStr(slot->data.name))));
371 :
372 1366 : if (IsTransactionState() &&
373 566 : GetTopTransactionIdIfAny() != InvalidTransactionId)
374 4 : ereport(ERROR,
375 : (errcode(ERRCODE_ACTIVE_SQL_TRANSACTION),
376 : errmsg("cannot create logical replication slot in transaction that has performed writes")));
377 :
378 : /*
379 : * Register output plugin name with slot. We need the mutex to avoid
380 : * concurrent reading of a partially copied string. But we don't want any
381 : * complicated code while holding a spinlock, so do namestrcpy() outside.
382 : */
383 796 : namestrcpy(&plugin_name, plugin);
384 796 : SpinLockAcquire(&slot->mutex);
385 796 : slot->data.plugin = plugin_name;
386 796 : SpinLockRelease(&slot->mutex);
387 :
388 796 : if (XLogRecPtrIsInvalid(restart_lsn))
389 784 : ReplicationSlotReserveWal();
390 : else
391 : {
392 12 : SpinLockAcquire(&slot->mutex);
393 12 : slot->data.restart_lsn = restart_lsn;
394 12 : SpinLockRelease(&slot->mutex);
395 : }
396 :
397 : /* ----
398 : * This is a bit tricky: We need to determine a safe xmin horizon to start
399 : * decoding from, to avoid starting from a running xacts record referring
400 : * to xids whose rows have been vacuumed or pruned
401 : * already. GetOldestSafeDecodingTransactionId() returns such a value, but
402 : * without further interlock its return value might immediately be out of
403 : * date.
404 : *
405 : * So we have to acquire the ProcArrayLock to prevent computation of new
406 : * xmin horizons by other backends, get the safe decoding xid, and inform
407 : * the slot machinery about the new limit. Once that's done the
408 : * ProcArrayLock can be released as the slot machinery now is
409 : * protecting against vacuum.
410 : *
411 : * Note that, temporarily, the data, not just the catalog, xmin has to be
412 : * reserved if a data snapshot is to be exported. Otherwise the initial
413 : * data snapshot created here is not guaranteed to be valid. After that
414 : * the data xmin doesn't need to be managed anymore and the global xmin
415 : * should be recomputed. As we are fine with losing the pegged data xmin
416 : * after crash - no chance a snapshot would get exported anymore - we can
417 : * get away with just setting the slot's
418 : * effective_xmin. ReplicationSlotRelease will reset it again.
419 : *
420 : * ----
421 : */
422 796 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
423 :
424 796 : xmin_horizon = GetOldestSafeDecodingTransactionId(!need_full_snapshot);
425 :
426 796 : SpinLockAcquire(&slot->mutex);
427 796 : slot->effective_catalog_xmin = xmin_horizon;
428 796 : slot->data.catalog_xmin = xmin_horizon;
429 796 : if (need_full_snapshot)
430 342 : slot->effective_xmin = xmin_horizon;
431 796 : SpinLockRelease(&slot->mutex);
432 :
433 796 : ReplicationSlotsComputeRequiredXmin(true);
434 :
435 796 : LWLockRelease(ProcArrayLock);
436 :
437 796 : ReplicationSlotMarkDirty();
438 796 : ReplicationSlotSave();
439 :
440 796 : ctx = StartupDecodingContext(NIL, restart_lsn, xmin_horizon,
441 : need_full_snapshot, false,
442 : xl_routine, prepare_write, do_write,
443 : update_progress);
444 :
445 : /* call output plugin initialization callback */
446 794 : old_context = MemoryContextSwitchTo(ctx->context);
447 794 : if (ctx->callbacks.startup_cb != NULL)
448 794 : startup_cb_wrapper(ctx, &ctx->options, true);
449 794 : MemoryContextSwitchTo(old_context);
450 :
451 : /*
452 : * We allow decoding of prepared transactions when the two_phase is
453 : * enabled at the time of slot creation, or when the two_phase option is
454 : * given at the streaming start, provided the plugin supports all the
455 : * callbacks for two-phase.
456 : */
457 794 : ctx->twophase &= slot->data.two_phase;
458 :
459 794 : ctx->reorder->output_rewrites = ctx->options.receive_rewrites;
460 :
461 794 : return ctx;
462 : }
463 :
464 : /*
465 : * Create a new decoding context, for a logical slot that has previously been
466 : * used already.
467 : *
468 : * start_lsn
469 : * The LSN at which to start decoding. If InvalidXLogRecPtr, restart
470 : * from the slot's confirmed_flush; otherwise, start from the specified
471 : * location (but move it forwards to confirmed_flush if it's older than
472 : * that, see below).
473 : *
474 : * output_plugin_options
475 : * options passed to the output plugin.
476 : *
477 : * fast_forward
478 : * bypass the generation of logical changes.
479 : *
480 : * xl_routine
481 : * XLogReaderRoutine used by underlying xlogreader
482 : *
483 : * prepare_write, do_write, update_progress
484 : * callbacks that have to be filled to perform the use-case dependent,
485 : * actual work.
486 : *
487 : * Needs to be called while in a memory context that's at least as long lived
488 : * as the decoding context because further memory contexts will be created
489 : * inside it.
490 : *
491 : * Returns an initialized decoding context after calling the output plugin's
492 : * startup function.
493 : */
494 : LogicalDecodingContext *
495 1084 : CreateDecodingContext(XLogRecPtr start_lsn,
496 : List *output_plugin_options,
497 : bool fast_forward,
498 : XLogReaderRoutine *xl_routine,
499 : LogicalOutputPluginWriterPrepareWrite prepare_write,
500 : LogicalOutputPluginWriterWrite do_write,
501 : LogicalOutputPluginWriterUpdateProgress update_progress)
502 : {
503 : LogicalDecodingContext *ctx;
504 : ReplicationSlot *slot;
505 : MemoryContext old_context;
506 :
507 : /* shorter lines... */
508 1084 : slot = MyReplicationSlot;
509 :
510 : /* first some sanity checks that are unlikely to be violated */
511 1084 : if (slot == NULL)
512 0 : elog(ERROR, "cannot perform logical decoding without an acquired slot");
513 :
514 : /* make sure the passed slot is suitable, these are user facing errors */
515 1084 : if (SlotIsPhysical(slot))
516 2 : ereport(ERROR,
517 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
518 : errmsg("cannot use physical replication slot for logical decoding")));
519 :
520 : /*
521 : * We need to access the system tables during decoding to build the
522 : * logical changes unless we are in fast_forward mode where no changes are
523 : * generated.
524 : */
525 1082 : if (slot->data.database != MyDatabaseId && !fast_forward)
526 6 : ereport(ERROR,
527 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
528 : errmsg("replication slot \"%s\" was not created in this database",
529 : NameStr(slot->data.name))));
530 :
531 : /*
532 : * The slots being synced from the primary can't be used for decoding as
533 : * they are used after failover. However, we do allow advancing the LSNs
534 : * during the synchronization of slots. See update_local_synced_slot.
535 : */
536 1076 : if (RecoveryInProgress() && slot->data.synced && !IsSyncingReplicationSlots())
537 2 : ereport(ERROR,
538 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
539 : errmsg("cannot use replication slot \"%s\" for logical decoding",
540 : NameStr(slot->data.name)),
541 : errdetail("This slot is being synchronized from the primary server."),
542 : errhint("Specify another replication slot."));
543 :
544 : /*
545 : * Check if slot has been invalidated due to max_slot_wal_keep_size. Avoid
546 : * "cannot get changes" wording in this errmsg because that'd be
547 : * confusingly ambiguous about no changes being available when called from
548 : * pg_logical_slot_get_changes_guts().
549 : */
550 1074 : if (MyReplicationSlot->data.invalidated == RS_INVAL_WAL_REMOVED)
551 0 : ereport(ERROR,
552 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
553 : errmsg("can no longer get changes from replication slot \"%s\"",
554 : NameStr(MyReplicationSlot->data.name)),
555 : errdetail("This slot has been invalidated because it exceeded the maximum reserved size.")));
556 :
557 1074 : if (MyReplicationSlot->data.invalidated != RS_INVAL_NONE)
558 10 : ereport(ERROR,
559 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
560 : errmsg("can no longer get changes from replication slot \"%s\"",
561 : NameStr(MyReplicationSlot->data.name)),
562 : errdetail("This slot has been invalidated because it was conflicting with recovery.")));
563 :
564 : Assert(MyReplicationSlot->data.invalidated == RS_INVAL_NONE);
565 : Assert(MyReplicationSlot->data.restart_lsn != InvalidXLogRecPtr);
566 :
567 1064 : if (start_lsn == InvalidXLogRecPtr)
568 : {
569 : /* continue from last position */
570 646 : start_lsn = slot->data.confirmed_flush;
571 : }
572 418 : else if (start_lsn < slot->data.confirmed_flush)
573 : {
574 : /*
575 : * It might seem like we should error out in this case, but it's
576 : * pretty common for a client to acknowledge a LSN it doesn't have to
577 : * do anything for, and thus didn't store persistently, because the
578 : * xlog records didn't result in anything relevant for logical
579 : * decoding. Clients have to be able to do that to support synchronous
580 : * replication.
581 : *
582 : * Starting at a different LSN than requested might not catch certain
583 : * kinds of client errors; so the client may wish to check that
584 : * confirmed_flush_lsn matches its expectations.
585 : */
586 26 : elog(LOG, "%X/%X has been already streamed, forwarding to %X/%X",
587 : LSN_FORMAT_ARGS(start_lsn),
588 : LSN_FORMAT_ARGS(slot->data.confirmed_flush));
589 :
590 26 : start_lsn = slot->data.confirmed_flush;
591 : }
592 :
593 1064 : ctx = StartupDecodingContext(output_plugin_options,
594 : start_lsn, InvalidTransactionId, false,
595 : fast_forward, xl_routine, prepare_write,
596 : do_write, update_progress);
597 :
598 : /* call output plugin initialization callback */
599 1064 : old_context = MemoryContextSwitchTo(ctx->context);
600 1064 : if (ctx->callbacks.startup_cb != NULL)
601 1028 : startup_cb_wrapper(ctx, &ctx->options, false);
602 1058 : MemoryContextSwitchTo(old_context);
603 :
604 : /*
605 : * We allow decoding of prepared transactions when the two_phase is
606 : * enabled at the time of slot creation, or when the two_phase option is
607 : * given at the streaming start, provided the plugin supports all the
608 : * callbacks for two-phase.
609 : */
610 1058 : ctx->twophase &= (slot->data.two_phase || ctx->twophase_opt_given);
611 :
612 : /* Mark slot to allow two_phase decoding if not already marked */
613 1058 : if (ctx->twophase && !slot->data.two_phase)
614 : {
615 12 : SpinLockAcquire(&slot->mutex);
616 12 : slot->data.two_phase = true;
617 12 : slot->data.two_phase_at = start_lsn;
618 12 : SpinLockRelease(&slot->mutex);
619 12 : ReplicationSlotMarkDirty();
620 12 : ReplicationSlotSave();
621 12 : SnapBuildSetTwoPhaseAt(ctx->snapshot_builder, start_lsn);
622 : }
623 :
624 1058 : ctx->reorder->output_rewrites = ctx->options.receive_rewrites;
625 :
626 1058 : ereport(LOG,
627 : (errmsg("starting logical decoding for slot \"%s\"",
628 : NameStr(slot->data.name)),
629 : errdetail("Streaming transactions committing after %X/%X, reading WAL from %X/%X.",
630 : LSN_FORMAT_ARGS(slot->data.confirmed_flush),
631 : LSN_FORMAT_ARGS(slot->data.restart_lsn))));
632 :
633 1058 : return ctx;
634 : }
635 :
636 : /*
637 : * Returns true if a consistent initial decoding snapshot has been built.
638 : */
639 : bool
640 872 : DecodingContextReady(LogicalDecodingContext *ctx)
641 : {
642 872 : return SnapBuildCurrentState(ctx->snapshot_builder) == SNAPBUILD_CONSISTENT;
643 : }
644 :
645 : /*
646 : * Read from the decoding slot, until it is ready to start extracting changes.
647 : */
648 : void
649 782 : DecodingContextFindStartpoint(LogicalDecodingContext *ctx)
650 : {
651 782 : ReplicationSlot *slot = ctx->slot;
652 :
653 : /* Initialize from where to start reading WAL. */
654 782 : XLogBeginRead(ctx->reader, slot->data.restart_lsn);
655 :
656 782 : elog(DEBUG1, "searching for logical decoding starting point, starting at %X/%X",
657 : LSN_FORMAT_ARGS(slot->data.restart_lsn));
658 :
659 : /* Wait for a consistent starting point */
660 : for (;;)
661 86 : {
662 : XLogRecord *record;
663 868 : char *err = NULL;
664 :
665 : /* the read_page callback waits for new WAL */
666 868 : record = XLogReadRecord(ctx->reader, &err);
667 868 : if (err)
668 0 : elog(ERROR, "could not find logical decoding starting point: %s", err);
669 868 : if (!record)
670 0 : elog(ERROR, "could not find logical decoding starting point");
671 :
672 868 : LogicalDecodingProcessRecord(ctx, ctx->reader);
673 :
674 : /* only continue till we found a consistent spot */
675 864 : if (DecodingContextReady(ctx))
676 778 : break;
677 :
678 86 : CHECK_FOR_INTERRUPTS();
679 : }
680 :
681 778 : SpinLockAcquire(&slot->mutex);
682 778 : slot->data.confirmed_flush = ctx->reader->EndRecPtr;
683 778 : if (slot->data.two_phase)
684 12 : slot->data.two_phase_at = ctx->reader->EndRecPtr;
685 778 : SpinLockRelease(&slot->mutex);
686 778 : }
687 :
688 : /*
689 : * Free a previously allocated decoding context, invoking the shutdown
690 : * callback if necessary.
691 : */
692 : void
693 1532 : FreeDecodingContext(LogicalDecodingContext *ctx)
694 : {
695 1532 : if (ctx->callbacks.shutdown_cb != NULL)
696 1496 : shutdown_cb_wrapper(ctx);
697 :
698 1532 : ReorderBufferFree(ctx->reorder);
699 1532 : FreeSnapshotBuilder(ctx->snapshot_builder);
700 1532 : XLogReaderFree(ctx->reader);
701 1532 : MemoryContextDelete(ctx->context);
702 1532 : }
703 :
704 : /*
705 : * Prepare a write using the context's output routine.
706 : */
707 : void
708 671536 : OutputPluginPrepareWrite(struct LogicalDecodingContext *ctx, bool last_write)
709 : {
710 671536 : if (!ctx->accept_writes)
711 0 : elog(ERROR, "writes are only accepted in commit, begin and change callbacks");
712 :
713 671536 : ctx->prepare_write(ctx, ctx->write_location, ctx->write_xid, last_write);
714 671536 : ctx->prepared_write = true;
715 671536 : }
716 :
717 : /*
718 : * Perform a write using the context's output routine.
719 : */
720 : void
721 671536 : OutputPluginWrite(struct LogicalDecodingContext *ctx, bool last_write)
722 : {
723 671536 : if (!ctx->prepared_write)
724 0 : elog(ERROR, "OutputPluginPrepareWrite needs to be called before OutputPluginWrite");
725 :
726 671536 : ctx->write(ctx, ctx->write_location, ctx->write_xid, last_write);
727 671532 : ctx->prepared_write = false;
728 671532 : }
729 :
730 : /*
731 : * Update progress tracking (if supported).
732 : */
733 : void
734 7672 : OutputPluginUpdateProgress(struct LogicalDecodingContext *ctx,
735 : bool skipped_xact)
736 : {
737 7672 : if (!ctx->update_progress)
738 3162 : return;
739 :
740 4510 : ctx->update_progress(ctx, ctx->write_location, ctx->write_xid,
741 : skipped_xact);
742 : }
743 :
744 : /*
745 : * Load the output plugin, lookup its output plugin init function, and check
746 : * that it provides the required callbacks.
747 : */
748 : static void
749 1824 : LoadOutputPlugin(OutputPluginCallbacks *callbacks, const char *plugin)
750 : {
751 : LogicalOutputPluginInit plugin_init;
752 :
753 1822 : plugin_init = (LogicalOutputPluginInit)
754 1824 : load_external_function(plugin, "_PG_output_plugin_init", false, NULL);
755 :
756 1822 : if (plugin_init == NULL)
757 0 : elog(ERROR, "output plugins have to declare the _PG_output_plugin_init symbol");
758 :
759 : /* ask the output plugin to fill the callback struct */
760 1822 : plugin_init(callbacks);
761 :
762 1822 : if (callbacks->begin_cb == NULL)
763 0 : elog(ERROR, "output plugins have to register a begin callback");
764 1822 : if (callbacks->change_cb == NULL)
765 0 : elog(ERROR, "output plugins have to register a change callback");
766 1822 : if (callbacks->commit_cb == NULL)
767 0 : elog(ERROR, "output plugins have to register a commit callback");
768 1822 : }
769 :
770 : static void
771 18 : output_plugin_error_callback(void *arg)
772 : {
773 18 : LogicalErrorCallbackState *state = (LogicalErrorCallbackState *) arg;
774 :
775 : /* not all callbacks have an associated LSN */
776 18 : if (state->report_location != InvalidXLogRecPtr)
777 12 : errcontext("slot \"%s\", output plugin \"%s\", in the %s callback, associated LSN %X/%X",
778 12 : NameStr(state->ctx->slot->data.name),
779 12 : NameStr(state->ctx->slot->data.plugin),
780 : state->callback_name,
781 12 : LSN_FORMAT_ARGS(state->report_location));
782 : else
783 6 : errcontext("slot \"%s\", output plugin \"%s\", in the %s callback",
784 6 : NameStr(state->ctx->slot->data.name),
785 6 : NameStr(state->ctx->slot->data.plugin),
786 : state->callback_name);
787 18 : }
788 :
789 : static void
790 1822 : startup_cb_wrapper(LogicalDecodingContext *ctx, OutputPluginOptions *opt, bool is_init)
791 : {
792 : LogicalErrorCallbackState state;
793 : ErrorContextCallback errcallback;
794 :
795 : Assert(!ctx->fast_forward);
796 :
797 : /* Push callback + info on the error context stack */
798 1822 : state.ctx = ctx;
799 1822 : state.callback_name = "startup";
800 1822 : state.report_location = InvalidXLogRecPtr;
801 1822 : errcallback.callback = output_plugin_error_callback;
802 1822 : errcallback.arg = (void *) &state;
803 1822 : errcallback.previous = error_context_stack;
804 1822 : error_context_stack = &errcallback;
805 :
806 : /* set output state */
807 1822 : ctx->accept_writes = false;
808 1822 : ctx->end_xact = false;
809 :
810 : /* do the actual work: call callback */
811 1822 : ctx->callbacks.startup_cb(ctx, opt, is_init);
812 :
813 : /* Pop the error context stack */
814 1816 : error_context_stack = errcallback.previous;
815 1816 : }
816 :
817 : static void
818 1496 : shutdown_cb_wrapper(LogicalDecodingContext *ctx)
819 : {
820 : LogicalErrorCallbackState state;
821 : ErrorContextCallback errcallback;
822 :
823 : Assert(!ctx->fast_forward);
824 :
825 : /* Push callback + info on the error context stack */
826 1496 : state.ctx = ctx;
827 1496 : state.callback_name = "shutdown";
828 1496 : state.report_location = InvalidXLogRecPtr;
829 1496 : errcallback.callback = output_plugin_error_callback;
830 1496 : errcallback.arg = (void *) &state;
831 1496 : errcallback.previous = error_context_stack;
832 1496 : error_context_stack = &errcallback;
833 :
834 : /* set output state */
835 1496 : ctx->accept_writes = false;
836 1496 : ctx->end_xact = false;
837 :
838 : /* do the actual work: call callback */
839 1496 : ctx->callbacks.shutdown_cb(ctx);
840 :
841 : /* Pop the error context stack */
842 1496 : error_context_stack = errcallback.previous;
843 1496 : }
844 :
845 :
846 : /*
847 : * Callbacks for ReorderBuffer which add in some more information and then call
848 : * output_plugin.h plugins.
849 : */
850 : static void
851 2116 : begin_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
852 : {
853 2116 : LogicalDecodingContext *ctx = cache->private_data;
854 : LogicalErrorCallbackState state;
855 : ErrorContextCallback errcallback;
856 :
857 : Assert(!ctx->fast_forward);
858 :
859 : /* Push callback + info on the error context stack */
860 2116 : state.ctx = ctx;
861 2116 : state.callback_name = "begin";
862 2116 : state.report_location = txn->first_lsn;
863 2116 : errcallback.callback = output_plugin_error_callback;
864 2116 : errcallback.arg = (void *) &state;
865 2116 : errcallback.previous = error_context_stack;
866 2116 : error_context_stack = &errcallback;
867 :
868 : /* set output state */
869 2116 : ctx->accept_writes = true;
870 2116 : ctx->write_xid = txn->xid;
871 2116 : ctx->write_location = txn->first_lsn;
872 2116 : ctx->end_xact = false;
873 :
874 : /* do the actual work: call callback */
875 2116 : ctx->callbacks.begin_cb(ctx, txn);
876 :
877 : /* Pop the error context stack */
878 2116 : error_context_stack = errcallback.previous;
879 2116 : }
880 :
881 : static void
882 2106 : commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
883 : XLogRecPtr commit_lsn)
884 : {
885 2106 : LogicalDecodingContext *ctx = cache->private_data;
886 : LogicalErrorCallbackState state;
887 : ErrorContextCallback errcallback;
888 :
889 : Assert(!ctx->fast_forward);
890 :
891 : /* Push callback + info on the error context stack */
892 2106 : state.ctx = ctx;
893 2106 : state.callback_name = "commit";
894 2106 : state.report_location = txn->final_lsn; /* beginning of commit record */
895 2106 : errcallback.callback = output_plugin_error_callback;
896 2106 : errcallback.arg = (void *) &state;
897 2106 : errcallback.previous = error_context_stack;
898 2106 : error_context_stack = &errcallback;
899 :
900 : /* set output state */
901 2106 : ctx->accept_writes = true;
902 2106 : ctx->write_xid = txn->xid;
903 2106 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
904 2106 : ctx->end_xact = true;
905 :
906 : /* do the actual work: call callback */
907 2106 : ctx->callbacks.commit_cb(ctx, txn, commit_lsn);
908 :
909 : /* Pop the error context stack */
910 2106 : error_context_stack = errcallback.previous;
911 2106 : }
912 :
913 : /*
914 : * The functionality of begin_prepare is quite similar to begin with the
915 : * exception that this will have gid (global transaction id) information which
916 : * can be used by plugin. Now, we thought about extending the existing begin
917 : * but that would break the replication protocol and additionally this looks
918 : * cleaner.
919 : */
920 : static void
921 48 : begin_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
922 : {
923 48 : LogicalDecodingContext *ctx = cache->private_data;
924 : LogicalErrorCallbackState state;
925 : ErrorContextCallback errcallback;
926 :
927 : Assert(!ctx->fast_forward);
928 :
929 : /* We're only supposed to call this when two-phase commits are supported */
930 : Assert(ctx->twophase);
931 :
932 : /* Push callback + info on the error context stack */
933 48 : state.ctx = ctx;
934 48 : state.callback_name = "begin_prepare";
935 48 : state.report_location = txn->first_lsn;
936 48 : errcallback.callback = output_plugin_error_callback;
937 48 : errcallback.arg = (void *) &state;
938 48 : errcallback.previous = error_context_stack;
939 48 : error_context_stack = &errcallback;
940 :
941 : /* set output state */
942 48 : ctx->accept_writes = true;
943 48 : ctx->write_xid = txn->xid;
944 48 : ctx->write_location = txn->first_lsn;
945 48 : ctx->end_xact = false;
946 :
947 : /*
948 : * If the plugin supports two-phase commits then begin prepare callback is
949 : * mandatory
950 : */
951 48 : if (ctx->callbacks.begin_prepare_cb == NULL)
952 0 : ereport(ERROR,
953 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
954 : errmsg("logical replication at prepare time requires a %s callback",
955 : "begin_prepare_cb")));
956 :
957 : /* do the actual work: call callback */
958 48 : ctx->callbacks.begin_prepare_cb(ctx, txn);
959 :
960 : /* Pop the error context stack */
961 48 : error_context_stack = errcallback.previous;
962 48 : }
963 :
964 : static void
965 48 : prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
966 : XLogRecPtr prepare_lsn)
967 : {
968 48 : LogicalDecodingContext *ctx = cache->private_data;
969 : LogicalErrorCallbackState state;
970 : ErrorContextCallback errcallback;
971 :
972 : Assert(!ctx->fast_forward);
973 :
974 : /* We're only supposed to call this when two-phase commits are supported */
975 : Assert(ctx->twophase);
976 :
977 : /* Push callback + info on the error context stack */
978 48 : state.ctx = ctx;
979 48 : state.callback_name = "prepare";
980 48 : state.report_location = txn->final_lsn; /* beginning of prepare record */
981 48 : errcallback.callback = output_plugin_error_callback;
982 48 : errcallback.arg = (void *) &state;
983 48 : errcallback.previous = error_context_stack;
984 48 : error_context_stack = &errcallback;
985 :
986 : /* set output state */
987 48 : ctx->accept_writes = true;
988 48 : ctx->write_xid = txn->xid;
989 48 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
990 48 : ctx->end_xact = true;
991 :
992 : /*
993 : * If the plugin supports two-phase commits then prepare callback is
994 : * mandatory
995 : */
996 48 : if (ctx->callbacks.prepare_cb == NULL)
997 0 : ereport(ERROR,
998 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
999 : errmsg("logical replication at prepare time requires a %s callback",
1000 : "prepare_cb")));
1001 :
1002 : /* do the actual work: call callback */
1003 48 : ctx->callbacks.prepare_cb(ctx, txn, prepare_lsn);
1004 :
1005 : /* Pop the error context stack */
1006 48 : error_context_stack = errcallback.previous;
1007 48 : }
1008 :
1009 : static void
1010 58 : commit_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1011 : XLogRecPtr commit_lsn)
1012 : {
1013 58 : LogicalDecodingContext *ctx = cache->private_data;
1014 : LogicalErrorCallbackState state;
1015 : ErrorContextCallback errcallback;
1016 :
1017 : Assert(!ctx->fast_forward);
1018 :
1019 : /* We're only supposed to call this when two-phase commits are supported */
1020 : Assert(ctx->twophase);
1021 :
1022 : /* Push callback + info on the error context stack */
1023 58 : state.ctx = ctx;
1024 58 : state.callback_name = "commit_prepared";
1025 58 : state.report_location = txn->final_lsn; /* beginning of commit record */
1026 58 : errcallback.callback = output_plugin_error_callback;
1027 58 : errcallback.arg = (void *) &state;
1028 58 : errcallback.previous = error_context_stack;
1029 58 : error_context_stack = &errcallback;
1030 :
1031 : /* set output state */
1032 58 : ctx->accept_writes = true;
1033 58 : ctx->write_xid = txn->xid;
1034 58 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
1035 58 : ctx->end_xact = true;
1036 :
1037 : /*
1038 : * If the plugin support two-phase commits then commit prepared callback
1039 : * is mandatory
1040 : */
1041 58 : if (ctx->callbacks.commit_prepared_cb == NULL)
1042 0 : ereport(ERROR,
1043 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1044 : errmsg("logical replication at prepare time requires a %s callback",
1045 : "commit_prepared_cb")));
1046 :
1047 : /* do the actual work: call callback */
1048 58 : ctx->callbacks.commit_prepared_cb(ctx, txn, commit_lsn);
1049 :
1050 : /* Pop the error context stack */
1051 58 : error_context_stack = errcallback.previous;
1052 58 : }
1053 :
1054 : static void
1055 20 : rollback_prepared_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1056 : XLogRecPtr prepare_end_lsn,
1057 : TimestampTz prepare_time)
1058 : {
1059 20 : LogicalDecodingContext *ctx = cache->private_data;
1060 : LogicalErrorCallbackState state;
1061 : ErrorContextCallback errcallback;
1062 :
1063 : Assert(!ctx->fast_forward);
1064 :
1065 : /* We're only supposed to call this when two-phase commits are supported */
1066 : Assert(ctx->twophase);
1067 :
1068 : /* Push callback + info on the error context stack */
1069 20 : state.ctx = ctx;
1070 20 : state.callback_name = "rollback_prepared";
1071 20 : state.report_location = txn->final_lsn; /* beginning of commit record */
1072 20 : errcallback.callback = output_plugin_error_callback;
1073 20 : errcallback.arg = (void *) &state;
1074 20 : errcallback.previous = error_context_stack;
1075 20 : error_context_stack = &errcallback;
1076 :
1077 : /* set output state */
1078 20 : ctx->accept_writes = true;
1079 20 : ctx->write_xid = txn->xid;
1080 20 : ctx->write_location = txn->end_lsn; /* points to the end of the record */
1081 20 : ctx->end_xact = true;
1082 :
1083 : /*
1084 : * If the plugin support two-phase commits then rollback prepared callback
1085 : * is mandatory
1086 : */
1087 20 : if (ctx->callbacks.rollback_prepared_cb == NULL)
1088 0 : ereport(ERROR,
1089 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1090 : errmsg("logical replication at prepare time requires a %s callback",
1091 : "rollback_prepared_cb")));
1092 :
1093 : /* do the actual work: call callback */
1094 20 : ctx->callbacks.rollback_prepared_cb(ctx, txn, prepare_end_lsn,
1095 : prepare_time);
1096 :
1097 : /* Pop the error context stack */
1098 20 : error_context_stack = errcallback.previous;
1099 20 : }
1100 :
1101 : static void
1102 315694 : change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1103 : Relation relation, ReorderBufferChange *change)
1104 : {
1105 315694 : LogicalDecodingContext *ctx = cache->private_data;
1106 : LogicalErrorCallbackState state;
1107 : ErrorContextCallback errcallback;
1108 :
1109 : Assert(!ctx->fast_forward);
1110 :
1111 : /* Push callback + info on the error context stack */
1112 315694 : state.ctx = ctx;
1113 315694 : state.callback_name = "change";
1114 315694 : state.report_location = change->lsn;
1115 315694 : errcallback.callback = output_plugin_error_callback;
1116 315694 : errcallback.arg = (void *) &state;
1117 315694 : errcallback.previous = error_context_stack;
1118 315694 : error_context_stack = &errcallback;
1119 :
1120 : /* set output state */
1121 315694 : ctx->accept_writes = true;
1122 315694 : ctx->write_xid = txn->xid;
1123 :
1124 : /*
1125 : * Report this change's lsn so replies from clients can give an up-to-date
1126 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1127 : * receipt of this transaction, but it might allow another transaction's
1128 : * commit to be confirmed with one message.
1129 : */
1130 315694 : ctx->write_location = change->lsn;
1131 :
1132 315694 : ctx->end_xact = false;
1133 :
1134 315694 : ctx->callbacks.change_cb(ctx, txn, relation, change);
1135 :
1136 : /* Pop the error context stack */
1137 315684 : error_context_stack = errcallback.previous;
1138 315684 : }
1139 :
1140 : static void
1141 38 : truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1142 : int nrelations, Relation relations[], ReorderBufferChange *change)
1143 : {
1144 38 : LogicalDecodingContext *ctx = cache->private_data;
1145 : LogicalErrorCallbackState state;
1146 : ErrorContextCallback errcallback;
1147 :
1148 : Assert(!ctx->fast_forward);
1149 :
1150 38 : if (!ctx->callbacks.truncate_cb)
1151 0 : return;
1152 :
1153 : /* Push callback + info on the error context stack */
1154 38 : state.ctx = ctx;
1155 38 : state.callback_name = "truncate";
1156 38 : state.report_location = change->lsn;
1157 38 : errcallback.callback = output_plugin_error_callback;
1158 38 : errcallback.arg = (void *) &state;
1159 38 : errcallback.previous = error_context_stack;
1160 38 : error_context_stack = &errcallback;
1161 :
1162 : /* set output state */
1163 38 : ctx->accept_writes = true;
1164 38 : ctx->write_xid = txn->xid;
1165 :
1166 : /*
1167 : * Report this change's lsn so replies from clients can give an up-to-date
1168 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1169 : * receipt of this transaction, but it might allow another transaction's
1170 : * commit to be confirmed with one message.
1171 : */
1172 38 : ctx->write_location = change->lsn;
1173 :
1174 38 : ctx->end_xact = false;
1175 :
1176 38 : ctx->callbacks.truncate_cb(ctx, txn, nrelations, relations, change);
1177 :
1178 : /* Pop the error context stack */
1179 38 : error_context_stack = errcallback.previous;
1180 : }
1181 :
1182 : bool
1183 230 : filter_prepare_cb_wrapper(LogicalDecodingContext *ctx, TransactionId xid,
1184 : const char *gid)
1185 : {
1186 : LogicalErrorCallbackState state;
1187 : ErrorContextCallback errcallback;
1188 : bool ret;
1189 :
1190 : Assert(!ctx->fast_forward);
1191 :
1192 : /* Push callback + info on the error context stack */
1193 230 : state.ctx = ctx;
1194 230 : state.callback_name = "filter_prepare";
1195 230 : state.report_location = InvalidXLogRecPtr;
1196 230 : errcallback.callback = output_plugin_error_callback;
1197 230 : errcallback.arg = (void *) &state;
1198 230 : errcallback.previous = error_context_stack;
1199 230 : error_context_stack = &errcallback;
1200 :
1201 : /* set output state */
1202 230 : ctx->accept_writes = false;
1203 230 : ctx->end_xact = false;
1204 :
1205 : /* do the actual work: call callback */
1206 230 : ret = ctx->callbacks.filter_prepare_cb(ctx, xid, gid);
1207 :
1208 : /* Pop the error context stack */
1209 230 : error_context_stack = errcallback.previous;
1210 :
1211 230 : return ret;
1212 : }
1213 :
1214 : bool
1215 3370284 : filter_by_origin_cb_wrapper(LogicalDecodingContext *ctx, RepOriginId origin_id)
1216 : {
1217 : LogicalErrorCallbackState state;
1218 : ErrorContextCallback errcallback;
1219 : bool ret;
1220 :
1221 : Assert(!ctx->fast_forward);
1222 :
1223 : /* Push callback + info on the error context stack */
1224 3370284 : state.ctx = ctx;
1225 3370284 : state.callback_name = "filter_by_origin";
1226 3370284 : state.report_location = InvalidXLogRecPtr;
1227 3370284 : errcallback.callback = output_plugin_error_callback;
1228 3370284 : errcallback.arg = (void *) &state;
1229 3370284 : errcallback.previous = error_context_stack;
1230 3370284 : error_context_stack = &errcallback;
1231 :
1232 : /* set output state */
1233 3370284 : ctx->accept_writes = false;
1234 3370284 : ctx->end_xact = false;
1235 :
1236 : /* do the actual work: call callback */
1237 3370284 : ret = ctx->callbacks.filter_by_origin_cb(ctx, origin_id);
1238 :
1239 : /* Pop the error context stack */
1240 3370284 : error_context_stack = errcallback.previous;
1241 :
1242 3370284 : return ret;
1243 : }
1244 :
1245 : static void
1246 32 : message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1247 : XLogRecPtr message_lsn, bool transactional,
1248 : const char *prefix, Size message_size, const char *message)
1249 : {
1250 32 : LogicalDecodingContext *ctx = cache->private_data;
1251 : LogicalErrorCallbackState state;
1252 : ErrorContextCallback errcallback;
1253 :
1254 : Assert(!ctx->fast_forward);
1255 :
1256 32 : if (ctx->callbacks.message_cb == NULL)
1257 0 : return;
1258 :
1259 : /* Push callback + info on the error context stack */
1260 32 : state.ctx = ctx;
1261 32 : state.callback_name = "message";
1262 32 : state.report_location = message_lsn;
1263 32 : errcallback.callback = output_plugin_error_callback;
1264 32 : errcallback.arg = (void *) &state;
1265 32 : errcallback.previous = error_context_stack;
1266 32 : error_context_stack = &errcallback;
1267 :
1268 : /* set output state */
1269 32 : ctx->accept_writes = true;
1270 32 : ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
1271 32 : ctx->write_location = message_lsn;
1272 32 : ctx->end_xact = false;
1273 :
1274 : /* do the actual work: call callback */
1275 32 : ctx->callbacks.message_cb(ctx, txn, message_lsn, transactional, prefix,
1276 : message_size, message);
1277 :
1278 : /* Pop the error context stack */
1279 32 : error_context_stack = errcallback.previous;
1280 : }
1281 :
1282 : static void
1283 1298 : stream_start_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1284 : XLogRecPtr first_lsn)
1285 : {
1286 1298 : LogicalDecodingContext *ctx = cache->private_data;
1287 : LogicalErrorCallbackState state;
1288 : ErrorContextCallback errcallback;
1289 :
1290 : Assert(!ctx->fast_forward);
1291 :
1292 : /* We're only supposed to call this when streaming is supported. */
1293 : Assert(ctx->streaming);
1294 :
1295 : /* Push callback + info on the error context stack */
1296 1298 : state.ctx = ctx;
1297 1298 : state.callback_name = "stream_start";
1298 1298 : state.report_location = first_lsn;
1299 1298 : errcallback.callback = output_plugin_error_callback;
1300 1298 : errcallback.arg = (void *) &state;
1301 1298 : errcallback.previous = error_context_stack;
1302 1298 : error_context_stack = &errcallback;
1303 :
1304 : /* set output state */
1305 1298 : ctx->accept_writes = true;
1306 1298 : ctx->write_xid = txn->xid;
1307 :
1308 : /*
1309 : * Report this message's lsn so replies from clients can give an
1310 : * up-to-date answer. This won't ever be enough (and shouldn't be!) to
1311 : * confirm receipt of this transaction, but it might allow another
1312 : * transaction's commit to be confirmed with one message.
1313 : */
1314 1298 : ctx->write_location = first_lsn;
1315 :
1316 1298 : ctx->end_xact = false;
1317 :
1318 : /* in streaming mode, stream_start_cb is required */
1319 1298 : if (ctx->callbacks.stream_start_cb == NULL)
1320 0 : ereport(ERROR,
1321 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1322 : errmsg("logical streaming requires a %s callback",
1323 : "stream_start_cb")));
1324 :
1325 1298 : ctx->callbacks.stream_start_cb(ctx, txn);
1326 :
1327 : /* Pop the error context stack */
1328 1298 : error_context_stack = errcallback.previous;
1329 1298 : }
1330 :
1331 : static void
1332 1298 : stream_stop_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1333 : XLogRecPtr last_lsn)
1334 : {
1335 1298 : LogicalDecodingContext *ctx = cache->private_data;
1336 : LogicalErrorCallbackState state;
1337 : ErrorContextCallback errcallback;
1338 :
1339 : Assert(!ctx->fast_forward);
1340 :
1341 : /* We're only supposed to call this when streaming is supported. */
1342 : Assert(ctx->streaming);
1343 :
1344 : /* Push callback + info on the error context stack */
1345 1298 : state.ctx = ctx;
1346 1298 : state.callback_name = "stream_stop";
1347 1298 : state.report_location = last_lsn;
1348 1298 : errcallback.callback = output_plugin_error_callback;
1349 1298 : errcallback.arg = (void *) &state;
1350 1298 : errcallback.previous = error_context_stack;
1351 1298 : error_context_stack = &errcallback;
1352 :
1353 : /* set output state */
1354 1298 : ctx->accept_writes = true;
1355 1298 : ctx->write_xid = txn->xid;
1356 :
1357 : /*
1358 : * Report this message's lsn so replies from clients can give an
1359 : * up-to-date answer. This won't ever be enough (and shouldn't be!) to
1360 : * confirm receipt of this transaction, but it might allow another
1361 : * transaction's commit to be confirmed with one message.
1362 : */
1363 1298 : ctx->write_location = last_lsn;
1364 :
1365 1298 : ctx->end_xact = false;
1366 :
1367 : /* in streaming mode, stream_stop_cb is required */
1368 1298 : if (ctx->callbacks.stream_stop_cb == NULL)
1369 0 : ereport(ERROR,
1370 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1371 : errmsg("logical streaming requires a %s callback",
1372 : "stream_stop_cb")));
1373 :
1374 1298 : ctx->callbacks.stream_stop_cb(ctx, txn);
1375 :
1376 : /* Pop the error context stack */
1377 1298 : error_context_stack = errcallback.previous;
1378 1298 : }
1379 :
1380 : static void
1381 58 : stream_abort_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1382 : XLogRecPtr abort_lsn)
1383 : {
1384 58 : LogicalDecodingContext *ctx = cache->private_data;
1385 : LogicalErrorCallbackState state;
1386 : ErrorContextCallback errcallback;
1387 :
1388 : Assert(!ctx->fast_forward);
1389 :
1390 : /* We're only supposed to call this when streaming is supported. */
1391 : Assert(ctx->streaming);
1392 :
1393 : /* Push callback + info on the error context stack */
1394 58 : state.ctx = ctx;
1395 58 : state.callback_name = "stream_abort";
1396 58 : state.report_location = abort_lsn;
1397 58 : errcallback.callback = output_plugin_error_callback;
1398 58 : errcallback.arg = (void *) &state;
1399 58 : errcallback.previous = error_context_stack;
1400 58 : error_context_stack = &errcallback;
1401 :
1402 : /* set output state */
1403 58 : ctx->accept_writes = true;
1404 58 : ctx->write_xid = txn->xid;
1405 58 : ctx->write_location = abort_lsn;
1406 58 : ctx->end_xact = true;
1407 :
1408 : /* in streaming mode, stream_abort_cb is required */
1409 58 : if (ctx->callbacks.stream_abort_cb == NULL)
1410 0 : ereport(ERROR,
1411 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1412 : errmsg("logical streaming requires a %s callback",
1413 : "stream_abort_cb")));
1414 :
1415 58 : ctx->callbacks.stream_abort_cb(ctx, txn, abort_lsn);
1416 :
1417 : /* Pop the error context stack */
1418 58 : error_context_stack = errcallback.previous;
1419 58 : }
1420 :
1421 : static void
1422 30 : stream_prepare_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1423 : XLogRecPtr prepare_lsn)
1424 : {
1425 30 : LogicalDecodingContext *ctx = cache->private_data;
1426 : LogicalErrorCallbackState state;
1427 : ErrorContextCallback errcallback;
1428 :
1429 : Assert(!ctx->fast_forward);
1430 :
1431 : /*
1432 : * We're only supposed to call this when streaming and two-phase commits
1433 : * are supported.
1434 : */
1435 : Assert(ctx->streaming);
1436 : Assert(ctx->twophase);
1437 :
1438 : /* Push callback + info on the error context stack */
1439 30 : state.ctx = ctx;
1440 30 : state.callback_name = "stream_prepare";
1441 30 : state.report_location = txn->final_lsn;
1442 30 : errcallback.callback = output_plugin_error_callback;
1443 30 : errcallback.arg = (void *) &state;
1444 30 : errcallback.previous = error_context_stack;
1445 30 : error_context_stack = &errcallback;
1446 :
1447 : /* set output state */
1448 30 : ctx->accept_writes = true;
1449 30 : ctx->write_xid = txn->xid;
1450 30 : ctx->write_location = txn->end_lsn;
1451 30 : ctx->end_xact = true;
1452 :
1453 : /* in streaming mode with two-phase commits, stream_prepare_cb is required */
1454 30 : if (ctx->callbacks.stream_prepare_cb == NULL)
1455 0 : ereport(ERROR,
1456 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1457 : errmsg("logical streaming at prepare time requires a %s callback",
1458 : "stream_prepare_cb")));
1459 :
1460 30 : ctx->callbacks.stream_prepare_cb(ctx, txn, prepare_lsn);
1461 :
1462 : /* Pop the error context stack */
1463 30 : error_context_stack = errcallback.previous;
1464 30 : }
1465 :
1466 : static void
1467 100 : stream_commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1468 : XLogRecPtr commit_lsn)
1469 : {
1470 100 : LogicalDecodingContext *ctx = cache->private_data;
1471 : LogicalErrorCallbackState state;
1472 : ErrorContextCallback errcallback;
1473 :
1474 : Assert(!ctx->fast_forward);
1475 :
1476 : /* We're only supposed to call this when streaming is supported. */
1477 : Assert(ctx->streaming);
1478 :
1479 : /* Push callback + info on the error context stack */
1480 100 : state.ctx = ctx;
1481 100 : state.callback_name = "stream_commit";
1482 100 : state.report_location = txn->final_lsn;
1483 100 : errcallback.callback = output_plugin_error_callback;
1484 100 : errcallback.arg = (void *) &state;
1485 100 : errcallback.previous = error_context_stack;
1486 100 : error_context_stack = &errcallback;
1487 :
1488 : /* set output state */
1489 100 : ctx->accept_writes = true;
1490 100 : ctx->write_xid = txn->xid;
1491 100 : ctx->write_location = txn->end_lsn;
1492 100 : ctx->end_xact = true;
1493 :
1494 : /* in streaming mode, stream_commit_cb is required */
1495 100 : if (ctx->callbacks.stream_commit_cb == NULL)
1496 0 : ereport(ERROR,
1497 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1498 : errmsg("logical streaming requires a %s callback",
1499 : "stream_commit_cb")));
1500 :
1501 100 : ctx->callbacks.stream_commit_cb(ctx, txn, commit_lsn);
1502 :
1503 : /* Pop the error context stack */
1504 100 : error_context_stack = errcallback.previous;
1505 100 : }
1506 :
1507 : static void
1508 352008 : stream_change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1509 : Relation relation, ReorderBufferChange *change)
1510 : {
1511 352008 : LogicalDecodingContext *ctx = cache->private_data;
1512 : LogicalErrorCallbackState state;
1513 : ErrorContextCallback errcallback;
1514 :
1515 : Assert(!ctx->fast_forward);
1516 :
1517 : /* We're only supposed to call this when streaming is supported. */
1518 : Assert(ctx->streaming);
1519 :
1520 : /* Push callback + info on the error context stack */
1521 352008 : state.ctx = ctx;
1522 352008 : state.callback_name = "stream_change";
1523 352008 : state.report_location = change->lsn;
1524 352008 : errcallback.callback = output_plugin_error_callback;
1525 352008 : errcallback.arg = (void *) &state;
1526 352008 : errcallback.previous = error_context_stack;
1527 352008 : error_context_stack = &errcallback;
1528 :
1529 : /* set output state */
1530 352008 : ctx->accept_writes = true;
1531 352008 : ctx->write_xid = txn->xid;
1532 :
1533 : /*
1534 : * Report this change's lsn so replies from clients can give an up-to-date
1535 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1536 : * receipt of this transaction, but it might allow another transaction's
1537 : * commit to be confirmed with one message.
1538 : */
1539 352008 : ctx->write_location = change->lsn;
1540 :
1541 352008 : ctx->end_xact = false;
1542 :
1543 : /* in streaming mode, stream_change_cb is required */
1544 352008 : if (ctx->callbacks.stream_change_cb == NULL)
1545 0 : ereport(ERROR,
1546 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1547 : errmsg("logical streaming requires a %s callback",
1548 : "stream_change_cb")));
1549 :
1550 352008 : ctx->callbacks.stream_change_cb(ctx, txn, relation, change);
1551 :
1552 : /* Pop the error context stack */
1553 352008 : error_context_stack = errcallback.previous;
1554 352008 : }
1555 :
1556 : static void
1557 6 : stream_message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1558 : XLogRecPtr message_lsn, bool transactional,
1559 : const char *prefix, Size message_size, const char *message)
1560 : {
1561 6 : LogicalDecodingContext *ctx = cache->private_data;
1562 : LogicalErrorCallbackState state;
1563 : ErrorContextCallback errcallback;
1564 :
1565 : Assert(!ctx->fast_forward);
1566 :
1567 : /* We're only supposed to call this when streaming is supported. */
1568 : Assert(ctx->streaming);
1569 :
1570 : /* this callback is optional */
1571 6 : if (ctx->callbacks.stream_message_cb == NULL)
1572 0 : return;
1573 :
1574 : /* Push callback + info on the error context stack */
1575 6 : state.ctx = ctx;
1576 6 : state.callback_name = "stream_message";
1577 6 : state.report_location = message_lsn;
1578 6 : errcallback.callback = output_plugin_error_callback;
1579 6 : errcallback.arg = (void *) &state;
1580 6 : errcallback.previous = error_context_stack;
1581 6 : error_context_stack = &errcallback;
1582 :
1583 : /* set output state */
1584 6 : ctx->accept_writes = true;
1585 6 : ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
1586 6 : ctx->write_location = message_lsn;
1587 6 : ctx->end_xact = false;
1588 :
1589 : /* do the actual work: call callback */
1590 6 : ctx->callbacks.stream_message_cb(ctx, txn, message_lsn, transactional, prefix,
1591 : message_size, message);
1592 :
1593 : /* Pop the error context stack */
1594 6 : error_context_stack = errcallback.previous;
1595 : }
1596 :
1597 : static void
1598 0 : stream_truncate_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1599 : int nrelations, Relation relations[],
1600 : ReorderBufferChange *change)
1601 : {
1602 0 : LogicalDecodingContext *ctx = cache->private_data;
1603 : LogicalErrorCallbackState state;
1604 : ErrorContextCallback errcallback;
1605 :
1606 : Assert(!ctx->fast_forward);
1607 :
1608 : /* We're only supposed to call this when streaming is supported. */
1609 : Assert(ctx->streaming);
1610 :
1611 : /* this callback is optional */
1612 0 : if (!ctx->callbacks.stream_truncate_cb)
1613 0 : return;
1614 :
1615 : /* Push callback + info on the error context stack */
1616 0 : state.ctx = ctx;
1617 0 : state.callback_name = "stream_truncate";
1618 0 : state.report_location = change->lsn;
1619 0 : errcallback.callback = output_plugin_error_callback;
1620 0 : errcallback.arg = (void *) &state;
1621 0 : errcallback.previous = error_context_stack;
1622 0 : error_context_stack = &errcallback;
1623 :
1624 : /* set output state */
1625 0 : ctx->accept_writes = true;
1626 0 : ctx->write_xid = txn->xid;
1627 :
1628 : /*
1629 : * Report this change's lsn so replies from clients can give an up-to-date
1630 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1631 : * receipt of this transaction, but it might allow another transaction's
1632 : * commit to be confirmed with one message.
1633 : */
1634 0 : ctx->write_location = change->lsn;
1635 :
1636 0 : ctx->end_xact = false;
1637 :
1638 0 : ctx->callbacks.stream_truncate_cb(ctx, txn, nrelations, relations, change);
1639 :
1640 : /* Pop the error context stack */
1641 0 : error_context_stack = errcallback.previous;
1642 : }
1643 :
1644 : static void
1645 6208 : update_progress_txn_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
1646 : XLogRecPtr lsn)
1647 : {
1648 6208 : LogicalDecodingContext *ctx = cache->private_data;
1649 : LogicalErrorCallbackState state;
1650 : ErrorContextCallback errcallback;
1651 :
1652 : Assert(!ctx->fast_forward);
1653 :
1654 : /* Push callback + info on the error context stack */
1655 6208 : state.ctx = ctx;
1656 6208 : state.callback_name = "update_progress_txn";
1657 6208 : state.report_location = lsn;
1658 6208 : errcallback.callback = output_plugin_error_callback;
1659 6208 : errcallback.arg = (void *) &state;
1660 6208 : errcallback.previous = error_context_stack;
1661 6208 : error_context_stack = &errcallback;
1662 :
1663 : /* set output state */
1664 6208 : ctx->accept_writes = false;
1665 6208 : ctx->write_xid = txn->xid;
1666 :
1667 : /*
1668 : * Report this change's lsn so replies from clients can give an up-to-date
1669 : * answer. This won't ever be enough (and shouldn't be!) to confirm
1670 : * receipt of this transaction, but it might allow another transaction's
1671 : * commit to be confirmed with one message.
1672 : */
1673 6208 : ctx->write_location = lsn;
1674 :
1675 6208 : ctx->end_xact = false;
1676 :
1677 6208 : OutputPluginUpdateProgress(ctx, false);
1678 :
1679 : /* Pop the error context stack */
1680 6208 : error_context_stack = errcallback.previous;
1681 6208 : }
1682 :
1683 : /*
1684 : * Set the required catalog xmin horizon for historic snapshots in the current
1685 : * replication slot.
1686 : *
1687 : * Note that in the most cases, we won't be able to immediately use the xmin
1688 : * to increase the xmin horizon: we need to wait till the client has confirmed
1689 : * receiving current_lsn with LogicalConfirmReceivedLocation().
1690 : */
1691 : void
1692 618 : LogicalIncreaseXminForSlot(XLogRecPtr current_lsn, TransactionId xmin)
1693 : {
1694 618 : bool updated_xmin = false;
1695 : ReplicationSlot *slot;
1696 618 : bool got_new_xmin = false;
1697 :
1698 618 : slot = MyReplicationSlot;
1699 :
1700 : Assert(slot != NULL);
1701 :
1702 618 : SpinLockAcquire(&slot->mutex);
1703 :
1704 : /*
1705 : * don't overwrite if we already have a newer xmin. This can happen if we
1706 : * restart decoding in a slot.
1707 : */
1708 618 : if (TransactionIdPrecedesOrEquals(xmin, slot->data.catalog_xmin))
1709 : {
1710 : }
1711 :
1712 : /*
1713 : * If the client has already confirmed up to this lsn, we directly can
1714 : * mark this as accepted. This can happen if we restart decoding in a
1715 : * slot.
1716 : */
1717 154 : else if (current_lsn <= slot->data.confirmed_flush)
1718 : {
1719 78 : slot->candidate_catalog_xmin = xmin;
1720 78 : slot->candidate_xmin_lsn = current_lsn;
1721 :
1722 : /* our candidate can directly be used */
1723 78 : updated_xmin = true;
1724 : }
1725 :
1726 : /*
1727 : * Only increase if the previous values have been applied, otherwise we
1728 : * might never end up updating if the receiver acks too slowly.
1729 : */
1730 76 : else if (slot->candidate_xmin_lsn == InvalidXLogRecPtr)
1731 : {
1732 30 : slot->candidate_catalog_xmin = xmin;
1733 30 : slot->candidate_xmin_lsn = current_lsn;
1734 :
1735 : /*
1736 : * Log new xmin at an appropriate log level after releasing the
1737 : * spinlock.
1738 : */
1739 30 : got_new_xmin = true;
1740 : }
1741 618 : SpinLockRelease(&slot->mutex);
1742 :
1743 618 : if (got_new_xmin)
1744 30 : elog(DEBUG1, "got new catalog xmin %u at %X/%X", xmin,
1745 : LSN_FORMAT_ARGS(current_lsn));
1746 :
1747 : /* candidate already valid with the current flush position, apply */
1748 618 : if (updated_xmin)
1749 78 : LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
1750 618 : }
1751 :
1752 : /*
1753 : * Mark the minimal LSN (restart_lsn) we need to read to replay all
1754 : * transactions that have not yet committed at current_lsn.
1755 : *
1756 : * Just like LogicalIncreaseXminForSlot this only takes effect when the
1757 : * client has confirmed to have received current_lsn.
1758 : */
1759 : void
1760 532 : LogicalIncreaseRestartDecodingForSlot(XLogRecPtr current_lsn, XLogRecPtr restart_lsn)
1761 : {
1762 532 : bool updated_lsn = false;
1763 : ReplicationSlot *slot;
1764 :
1765 532 : slot = MyReplicationSlot;
1766 :
1767 : Assert(slot != NULL);
1768 : Assert(restart_lsn != InvalidXLogRecPtr);
1769 : Assert(current_lsn != InvalidXLogRecPtr);
1770 :
1771 532 : SpinLockAcquire(&slot->mutex);
1772 :
1773 : /* don't overwrite if have a newer restart lsn */
1774 532 : if (restart_lsn <= slot->data.restart_lsn)
1775 : {
1776 : }
1777 :
1778 : /*
1779 : * We might have already flushed far enough to directly accept this lsn,
1780 : * in this case there is no need to check for existing candidate LSNs
1781 : */
1782 512 : else if (current_lsn <= slot->data.confirmed_flush)
1783 : {
1784 400 : slot->candidate_restart_valid = current_lsn;
1785 400 : slot->candidate_restart_lsn = restart_lsn;
1786 :
1787 : /* our candidate can directly be used */
1788 400 : updated_lsn = true;
1789 : }
1790 :
1791 : /*
1792 : * Only increase if the previous values have been applied, otherwise we
1793 : * might never end up updating if the receiver acks too slowly. A missed
1794 : * value here will just cause some extra effort after reconnecting.
1795 : */
1796 532 : if (slot->candidate_restart_valid == InvalidXLogRecPtr)
1797 : {
1798 62 : slot->candidate_restart_valid = current_lsn;
1799 62 : slot->candidate_restart_lsn = restart_lsn;
1800 62 : SpinLockRelease(&slot->mutex);
1801 :
1802 62 : elog(DEBUG1, "got new restart lsn %X/%X at %X/%X",
1803 : LSN_FORMAT_ARGS(restart_lsn),
1804 : LSN_FORMAT_ARGS(current_lsn));
1805 : }
1806 : else
1807 : {
1808 : XLogRecPtr candidate_restart_lsn;
1809 : XLogRecPtr candidate_restart_valid;
1810 : XLogRecPtr confirmed_flush;
1811 :
1812 470 : candidate_restart_lsn = slot->candidate_restart_lsn;
1813 470 : candidate_restart_valid = slot->candidate_restart_valid;
1814 470 : confirmed_flush = slot->data.confirmed_flush;
1815 470 : SpinLockRelease(&slot->mutex);
1816 :
1817 470 : elog(DEBUG1, "failed to increase restart lsn: proposed %X/%X, after %X/%X, current candidate %X/%X, current after %X/%X, flushed up to %X/%X",
1818 : LSN_FORMAT_ARGS(restart_lsn),
1819 : LSN_FORMAT_ARGS(current_lsn),
1820 : LSN_FORMAT_ARGS(candidate_restart_lsn),
1821 : LSN_FORMAT_ARGS(candidate_restart_valid),
1822 : LSN_FORMAT_ARGS(confirmed_flush));
1823 : }
1824 :
1825 : /* candidates are already valid with the current flush position, apply */
1826 532 : if (updated_lsn)
1827 400 : LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
1828 532 : }
1829 :
1830 : /*
1831 : * Handle a consumer's confirmation having received all changes up to lsn.
1832 : */
1833 : void
1834 55470 : LogicalConfirmReceivedLocation(XLogRecPtr lsn)
1835 : {
1836 : Assert(lsn != InvalidXLogRecPtr);
1837 :
1838 : /* Do an unlocked check for candidate_lsn first. */
1839 55470 : if (MyReplicationSlot->candidate_xmin_lsn != InvalidXLogRecPtr ||
1840 55366 : MyReplicationSlot->candidate_restart_valid != InvalidXLogRecPtr)
1841 540 : {
1842 540 : bool updated_xmin = false;
1843 540 : bool updated_restart = false;
1844 :
1845 540 : SpinLockAcquire(&MyReplicationSlot->mutex);
1846 :
1847 540 : MyReplicationSlot->data.confirmed_flush = lsn;
1848 :
1849 : /* if we're past the location required for bumping xmin, do so */
1850 540 : if (MyReplicationSlot->candidate_xmin_lsn != InvalidXLogRecPtr &&
1851 104 : MyReplicationSlot->candidate_xmin_lsn <= lsn)
1852 : {
1853 : /*
1854 : * We have to write the changed xmin to disk *before* we change
1855 : * the in-memory value, otherwise after a crash we wouldn't know
1856 : * that some catalog tuples might have been removed already.
1857 : *
1858 : * Ensure that by first writing to ->xmin and only update
1859 : * ->effective_xmin once the new state is synced to disk. After a
1860 : * crash ->effective_xmin is set to ->xmin.
1861 : */
1862 98 : if (TransactionIdIsValid(MyReplicationSlot->candidate_catalog_xmin) &&
1863 98 : MyReplicationSlot->data.catalog_xmin != MyReplicationSlot->candidate_catalog_xmin)
1864 : {
1865 98 : MyReplicationSlot->data.catalog_xmin = MyReplicationSlot->candidate_catalog_xmin;
1866 98 : MyReplicationSlot->candidate_catalog_xmin = InvalidTransactionId;
1867 98 : MyReplicationSlot->candidate_xmin_lsn = InvalidXLogRecPtr;
1868 98 : updated_xmin = true;
1869 : }
1870 : }
1871 :
1872 540 : if (MyReplicationSlot->candidate_restart_valid != InvalidXLogRecPtr &&
1873 462 : MyReplicationSlot->candidate_restart_valid <= lsn)
1874 : {
1875 : Assert(MyReplicationSlot->candidate_restart_lsn != InvalidXLogRecPtr);
1876 :
1877 454 : MyReplicationSlot->data.restart_lsn = MyReplicationSlot->candidate_restart_lsn;
1878 454 : MyReplicationSlot->candidate_restart_lsn = InvalidXLogRecPtr;
1879 454 : MyReplicationSlot->candidate_restart_valid = InvalidXLogRecPtr;
1880 454 : updated_restart = true;
1881 : }
1882 :
1883 540 : SpinLockRelease(&MyReplicationSlot->mutex);
1884 :
1885 : /* first write new xmin to disk, so we know what's up after a crash */
1886 540 : if (updated_xmin || updated_restart)
1887 : {
1888 532 : ReplicationSlotMarkDirty();
1889 532 : ReplicationSlotSave();
1890 532 : elog(DEBUG1, "updated xmin: %u restart: %u", updated_xmin, updated_restart);
1891 : }
1892 :
1893 : /*
1894 : * Now the new xmin is safely on disk, we can let the global value
1895 : * advance. We do not take ProcArrayLock or similar since we only
1896 : * advance xmin here and there's not much harm done by a concurrent
1897 : * computation missing that.
1898 : */
1899 540 : if (updated_xmin)
1900 : {
1901 98 : SpinLockAcquire(&MyReplicationSlot->mutex);
1902 98 : MyReplicationSlot->effective_catalog_xmin = MyReplicationSlot->data.catalog_xmin;
1903 98 : SpinLockRelease(&MyReplicationSlot->mutex);
1904 :
1905 98 : ReplicationSlotsComputeRequiredXmin(false);
1906 98 : ReplicationSlotsComputeRequiredLSN();
1907 : }
1908 : }
1909 : else
1910 : {
1911 54930 : SpinLockAcquire(&MyReplicationSlot->mutex);
1912 54930 : MyReplicationSlot->data.confirmed_flush = lsn;
1913 54930 : SpinLockRelease(&MyReplicationSlot->mutex);
1914 : }
1915 55470 : }
1916 :
1917 : /*
1918 : * Clear logical streaming state during (sub)transaction abort.
1919 : */
1920 : void
1921 54264 : ResetLogicalStreamingState(void)
1922 : {
1923 54264 : CheckXidAlive = InvalidTransactionId;
1924 54264 : bsysscan = false;
1925 54264 : }
1926 :
1927 : /*
1928 : * Report stats for a slot.
1929 : */
1930 : void
1931 11438 : UpdateDecodingStats(LogicalDecodingContext *ctx)
1932 : {
1933 11438 : ReorderBuffer *rb = ctx->reorder;
1934 : PgStat_StatReplSlotEntry repSlotStat;
1935 :
1936 : /* Nothing to do if we don't have any replication stats to be sent. */
1937 11438 : if (rb->spillBytes <= 0 && rb->streamBytes <= 0 && rb->totalBytes <= 0)
1938 484 : return;
1939 :
1940 10954 : elog(DEBUG2, "UpdateDecodingStats: updating stats %p %lld %lld %lld %lld %lld %lld %lld %lld",
1941 : rb,
1942 : (long long) rb->spillTxns,
1943 : (long long) rb->spillCount,
1944 : (long long) rb->spillBytes,
1945 : (long long) rb->streamTxns,
1946 : (long long) rb->streamCount,
1947 : (long long) rb->streamBytes,
1948 : (long long) rb->totalTxns,
1949 : (long long) rb->totalBytes);
1950 :
1951 10954 : repSlotStat.spill_txns = rb->spillTxns;
1952 10954 : repSlotStat.spill_count = rb->spillCount;
1953 10954 : repSlotStat.spill_bytes = rb->spillBytes;
1954 10954 : repSlotStat.stream_txns = rb->streamTxns;
1955 10954 : repSlotStat.stream_count = rb->streamCount;
1956 10954 : repSlotStat.stream_bytes = rb->streamBytes;
1957 10954 : repSlotStat.total_txns = rb->totalTxns;
1958 10954 : repSlotStat.total_bytes = rb->totalBytes;
1959 :
1960 10954 : pgstat_report_replslot(ctx->slot, &repSlotStat);
1961 :
1962 10954 : rb->spillTxns = 0;
1963 10954 : rb->spillCount = 0;
1964 10954 : rb->spillBytes = 0;
1965 10954 : rb->streamTxns = 0;
1966 10954 : rb->streamCount = 0;
1967 10954 : rb->streamBytes = 0;
1968 10954 : rb->totalTxns = 0;
1969 10954 : rb->totalBytes = 0;
1970 : }
1971 :
1972 : /*
1973 : * Read up to the end of WAL starting from the decoding slot's restart_lsn.
1974 : * Return true if any meaningful/decodable WAL records are encountered,
1975 : * otherwise false.
1976 : */
1977 : bool
1978 10 : LogicalReplicationSlotHasPendingWal(XLogRecPtr end_of_wal)
1979 : {
1980 10 : bool has_pending_wal = false;
1981 :
1982 : Assert(MyReplicationSlot);
1983 :
1984 10 : PG_TRY();
1985 : {
1986 : LogicalDecodingContext *ctx;
1987 :
1988 : /*
1989 : * Create our decoding context in fast_forward mode, passing start_lsn
1990 : * as InvalidXLogRecPtr, so that we start processing from the slot's
1991 : * confirmed_flush.
1992 : */
1993 20 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
1994 : NIL,
1995 : true, /* fast_forward */
1996 10 : XL_ROUTINE(.page_read = read_local_xlog_page,
1997 : .segment_open = wal_segment_open,
1998 : .segment_close = wal_segment_close),
1999 : NULL, NULL, NULL);
2000 :
2001 : /*
2002 : * Start reading at the slot's restart_lsn, which we know points to a
2003 : * valid record.
2004 : */
2005 10 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
2006 :
2007 : /* Invalidate non-timetravel entries */
2008 10 : InvalidateSystemCaches();
2009 :
2010 : /* Loop until the end of WAL or some changes are processed */
2011 300 : while (!has_pending_wal && ctx->reader->EndRecPtr < end_of_wal)
2012 : {
2013 : XLogRecord *record;
2014 290 : char *errm = NULL;
2015 :
2016 290 : record = XLogReadRecord(ctx->reader, &errm);
2017 :
2018 290 : if (errm)
2019 0 : elog(ERROR, "could not find record for logical decoding: %s", errm);
2020 :
2021 290 : if (record != NULL)
2022 290 : LogicalDecodingProcessRecord(ctx, ctx->reader);
2023 :
2024 290 : has_pending_wal = ctx->processing_required;
2025 :
2026 290 : CHECK_FOR_INTERRUPTS();
2027 : }
2028 :
2029 : /* Clean up */
2030 10 : FreeDecodingContext(ctx);
2031 10 : InvalidateSystemCaches();
2032 : }
2033 0 : PG_CATCH();
2034 : {
2035 : /* clear all timetravel entries */
2036 0 : InvalidateSystemCaches();
2037 :
2038 0 : PG_RE_THROW();
2039 : }
2040 10 : PG_END_TRY();
2041 :
2042 10 : return has_pending_wal;
2043 : }
2044 :
2045 : /*
2046 : * Helper function for advancing our logical replication slot forward.
2047 : *
2048 : * The slot's restart_lsn is used as start point for reading records, while
2049 : * confirmed_flush is used as base point for the decoding context.
2050 : *
2051 : * We cannot just do LogicalConfirmReceivedLocation to update confirmed_flush,
2052 : * because we need to digest WAL to advance restart_lsn allowing to recycle
2053 : * WAL and removal of old catalog tuples. As decoding is done in fast_forward
2054 : * mode, no changes are generated anyway.
2055 : *
2056 : * *found_consistent_snapshot will be true if the initial decoding snapshot has
2057 : * been built; Otherwise, it will be false.
2058 : */
2059 : XLogRecPtr
2060 26 : LogicalSlotAdvanceAndCheckSnapState(XLogRecPtr moveto,
2061 : bool *found_consistent_snapshot)
2062 : {
2063 : LogicalDecodingContext *ctx;
2064 26 : ResourceOwner old_resowner = CurrentResourceOwner;
2065 : XLogRecPtr retlsn;
2066 :
2067 : Assert(moveto != InvalidXLogRecPtr);
2068 :
2069 26 : if (found_consistent_snapshot)
2070 8 : *found_consistent_snapshot = false;
2071 :
2072 26 : PG_TRY();
2073 : {
2074 : /*
2075 : * Create our decoding context in fast_forward mode, passing start_lsn
2076 : * as InvalidXLogRecPtr, so that we start processing from my slot's
2077 : * confirmed_flush.
2078 : */
2079 52 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
2080 : NIL,
2081 : true, /* fast_forward */
2082 26 : XL_ROUTINE(.page_read = read_local_xlog_page,
2083 : .segment_open = wal_segment_open,
2084 : .segment_close = wal_segment_close),
2085 : NULL, NULL, NULL);
2086 :
2087 : /*
2088 : * Wait for specified streaming replication standby servers (if any)
2089 : * to confirm receipt of WAL up to moveto lsn.
2090 : */
2091 26 : WaitForStandbyConfirmation(moveto);
2092 :
2093 : /*
2094 : * Start reading at the slot's restart_lsn, which we know to point to
2095 : * a valid record.
2096 : */
2097 26 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
2098 :
2099 : /* invalidate non-timetravel entries */
2100 26 : InvalidateSystemCaches();
2101 :
2102 : /* Decode records until we reach the requested target */
2103 3338 : while (ctx->reader->EndRecPtr < moveto)
2104 : {
2105 3312 : char *errm = NULL;
2106 : XLogRecord *record;
2107 :
2108 : /*
2109 : * Read records. No changes are generated in fast_forward mode,
2110 : * but snapbuilder/slot statuses are updated properly.
2111 : */
2112 3312 : record = XLogReadRecord(ctx->reader, &errm);
2113 3312 : if (errm)
2114 0 : elog(ERROR, "could not find record while advancing replication slot: %s",
2115 : errm);
2116 :
2117 : /*
2118 : * Process the record. Storage-level changes are ignored in
2119 : * fast_forward mode, but other modules (such as snapbuilder)
2120 : * might still have critical updates to do.
2121 : */
2122 3312 : if (record)
2123 3312 : LogicalDecodingProcessRecord(ctx, ctx->reader);
2124 :
2125 3312 : CHECK_FOR_INTERRUPTS();
2126 : }
2127 :
2128 26 : if (found_consistent_snapshot && DecodingContextReady(ctx))
2129 8 : *found_consistent_snapshot = true;
2130 :
2131 : /*
2132 : * Logical decoding could have clobbered CurrentResourceOwner during
2133 : * transaction management, so restore the executor's value. (This is
2134 : * a kluge, but it's not worth cleaning up right now.)
2135 : */
2136 26 : CurrentResourceOwner = old_resowner;
2137 :
2138 26 : if (ctx->reader->EndRecPtr != InvalidXLogRecPtr)
2139 : {
2140 26 : LogicalConfirmReceivedLocation(moveto);
2141 :
2142 : /*
2143 : * If only the confirmed_flush LSN has changed the slot won't get
2144 : * marked as dirty by the above. Callers on the walsender
2145 : * interface are expected to keep track of their own progress and
2146 : * don't need it written out. But SQL-interface users cannot
2147 : * specify their own start positions and it's harder for them to
2148 : * keep track of their progress, so we should make more of an
2149 : * effort to save it for them.
2150 : *
2151 : * Dirty the slot so it is written out at the next checkpoint. The
2152 : * LSN position advanced to may still be lost on a crash but this
2153 : * makes the data consistent after a clean shutdown.
2154 : */
2155 26 : ReplicationSlotMarkDirty();
2156 : }
2157 :
2158 26 : retlsn = MyReplicationSlot->data.confirmed_flush;
2159 :
2160 : /* free context, call shutdown callback */
2161 26 : FreeDecodingContext(ctx);
2162 :
2163 26 : InvalidateSystemCaches();
2164 : }
2165 0 : PG_CATCH();
2166 : {
2167 : /* clear all timetravel entries */
2168 0 : InvalidateSystemCaches();
2169 :
2170 0 : PG_RE_THROW();
2171 : }
2172 26 : PG_END_TRY();
2173 :
2174 26 : return retlsn;
2175 : }
|
