1 /*-------------------------------------------------------------------------
3 * PostgreSQL logical decoding coordination
5 * Copyright (c) 2012-2017, PostgreSQL Global Development Group
8 * src/backend/replication/logical/logical.c
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
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
26 *-------------------------------------------------------------------------
31 #include "miscadmin.h"
33 #include "access/xact.h"
34 #include "access/xlog_internal.h"
36 #include "replication/decode.h"
37 #include "replication/logical.h"
38 #include "replication/reorderbuffer.h"
39 #include "replication/origin.h"
40 #include "replication/snapbuild.h"
42 #include "storage/proc.h"
43 #include "storage/procarray.h"
45 #include "utils/memutils.h"
47 /* data for errcontext callback */
48 typedef struct LogicalErrorCallbackState
50 LogicalDecodingContext *ctx;
51 const char *callback_name;
52 XLogRecPtr report_location;
53 } LogicalErrorCallbackState;
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,
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 change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
64 Relation relation, ReorderBufferChange *change);
65 static void message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
66 XLogRecPtr message_lsn, bool transactional,
67 const char *prefix, Size message_size, const char *message);
69 static void LoadOutputPlugin(OutputPluginCallbacks *callbacks, char *plugin);
72 * Make sure the current settings & environment are capable of doing logical
76 CheckLogicalDecodingRequirements(void)
78 CheckSlotRequirements();
80 if (wal_level < WAL_LEVEL_LOGICAL)
82 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
83 errmsg("logical decoding requires wal_level >= logical")));
85 if (MyDatabaseId == InvalidOid)
87 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
88 errmsg("logical decoding requires a database connection")));
91 * TODO: We got to change that someday soon...
93 * There's basically three things missing to allow this:
94 * 1) We need to be able to correctly and quickly identify the timeline a
96 * 2) We need to force hot_standby_feedback to be enabled at all times so
97 * the primary cannot remove rows we need.
98 * 3) support dropping replication slots referring to a database, in
99 * dbase_redo. There can't be any active ones due to HS recovery
100 * conflicts, so that should be relatively easy.
103 if (RecoveryInProgress())
105 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
106 errmsg("logical decoding cannot be used while in recovery")));
110 * Helper function for CreateInitialDecodingContext() and
111 * CreateDecodingContext() performing common tasks.
113 static LogicalDecodingContext *
114 StartupDecodingContext(List *output_plugin_options,
115 XLogRecPtr start_lsn,
116 TransactionId xmin_horizon,
117 bool need_full_snapshot,
118 XLogPageReadCB read_page,
119 LogicalOutputPluginWriterPrepareWrite prepare_write,
120 LogicalOutputPluginWriterWrite do_write,
121 LogicalOutputPluginWriterUpdateProgress update_progress)
123 ReplicationSlot *slot;
124 MemoryContext context,
126 LogicalDecodingContext *ctx;
128 /* shorter lines... */
129 slot = MyReplicationSlot;
131 context = AllocSetContextCreate(CurrentMemoryContext,
132 "Logical decoding context",
133 ALLOCSET_DEFAULT_SIZES);
134 old_context = MemoryContextSwitchTo(context);
135 ctx = palloc0(sizeof(LogicalDecodingContext));
137 ctx->context = context;
140 * (re-)load output plugins, so we detect a bad (removed) output plugin
143 LoadOutputPlugin(&ctx->callbacks, NameStr(slot->data.plugin));
146 * Now that the slot's xmin has been set, we can announce ourselves as a
147 * logical decoding backend which doesn't need to be checked individually
148 * when computing the xmin horizon because the xmin is enforced via
151 * We can only do so if we're outside of a transaction (i.e. the case when
152 * streaming changes via walsender), otherwise an already setup
153 * snapshot/xid would end up being ignored. That's not a particularly
154 * bothersome restriction since the SQL interface can't be used for
157 if (!IsTransactionOrTransactionBlock())
159 LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
160 MyPgXact->vacuumFlags |= PROC_IN_LOGICAL_DECODING;
161 LWLockRelease(ProcArrayLock);
166 ctx->reader = XLogReaderAllocate(read_page, ctx);
169 (errcode(ERRCODE_OUT_OF_MEMORY),
170 errmsg("out of memory")));
172 ctx->reader->private_data = ctx;
174 ctx->reorder = ReorderBufferAllocate();
175 ctx->snapshot_builder =
176 AllocateSnapshotBuilder(ctx->reorder, xmin_horizon, start_lsn,
179 ctx->reorder->private_data = ctx;
181 /* wrap output plugin callbacks, so we can add error context information */
182 ctx->reorder->begin = begin_cb_wrapper;
183 ctx->reorder->apply_change = change_cb_wrapper;
184 ctx->reorder->commit = commit_cb_wrapper;
185 ctx->reorder->message = message_cb_wrapper;
187 ctx->out = makeStringInfo();
188 ctx->prepare_write = prepare_write;
189 ctx->write = do_write;
190 ctx->update_progress = update_progress;
192 ctx->output_plugin_options = output_plugin_options;
194 MemoryContextSwitchTo(old_context);
200 * Create a new decoding context, for a new logical slot.
202 * plugin contains the name of the output plugin
203 * output_plugin_options contains options passed to the output plugin
204 * read_page, prepare_write, do_write, update_progress
205 * callbacks that have to be filled to perform the use-case dependent,
208 * Needs to be called while in a memory context that's at least as long lived
209 * as the decoding context because further memory contexts will be created
212 * Returns an initialized decoding context after calling the output plugin's
215 LogicalDecodingContext *
216 CreateInitDecodingContext(char *plugin,
217 List *output_plugin_options,
218 bool need_full_snapshot,
219 XLogPageReadCB read_page,
220 LogicalOutputPluginWriterPrepareWrite prepare_write,
221 LogicalOutputPluginWriterWrite do_write,
222 LogicalOutputPluginWriterUpdateProgress update_progress)
224 TransactionId xmin_horizon = InvalidTransactionId;
225 ReplicationSlot *slot;
226 LogicalDecodingContext *ctx;
227 MemoryContext old_context;
229 /* shorter lines... */
230 slot = MyReplicationSlot;
232 /* first some sanity checks that are unlikely to be violated */
234 elog(ERROR, "cannot perform logical decoding without an acquired slot");
237 elog(ERROR, "cannot initialize logical decoding without a specified plugin");
239 /* Make sure the passed slot is suitable. These are user facing errors. */
240 if (SlotIsPhysical(slot))
242 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
243 errmsg("cannot use physical replication slot for logical decoding")));
245 if (slot->data.database != MyDatabaseId)
247 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
248 errmsg("replication slot \"%s\" was not created in this database",
249 NameStr(slot->data.name))));
251 if (IsTransactionState() &&
252 GetTopTransactionIdIfAny() != InvalidTransactionId)
254 (errcode(ERRCODE_ACTIVE_SQL_TRANSACTION),
255 errmsg("cannot create logical replication slot in transaction that has performed writes")));
257 /* register output plugin name with slot */
258 SpinLockAcquire(&slot->mutex);
259 StrNCpy(NameStr(slot->data.plugin), plugin, NAMEDATALEN);
260 SpinLockRelease(&slot->mutex);
262 ReplicationSlotReserveWal();
265 * This is a bit tricky: We need to determine a safe xmin horizon to start
266 * decoding from, to avoid starting from a running xacts record referring
267 * to xids whose rows have been vacuumed or pruned
268 * already. GetOldestSafeDecodingTransactionId() returns such a value, but
269 * without further interlock its return value might immediately be out of
272 * So we have to acquire the ProcArrayLock to prevent computation of new
273 * xmin horizons by other backends, get the safe decoding xid, and inform
274 * the slot machinery about the new limit. Once that's done the
275 * ProcArrayLock can be released as the slot machinery now is
276 * protecting against vacuum.
278 * Note that, temporarily, the data, not just the catalog, xmin has to be
279 * reserved if a data snapshot is to be exported. Otherwise the initial
280 * data snapshot created here is not guaranteed to be valid. After that
281 * the data xmin doesn't need to be managed anymore and the global xmin
282 * should be recomputed. As we are fine with losing the pegged data xmin
283 * after crash - no chance a snapshot would get exported anymore - we can
284 * get away with just setting the slot's
285 * effective_xmin. ReplicationSlotRelease will reset it again.
289 LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
291 xmin_horizon = GetOldestSafeDecodingTransactionId(need_full_snapshot);
293 slot->effective_catalog_xmin = xmin_horizon;
294 slot->data.catalog_xmin = xmin_horizon;
295 if (need_full_snapshot)
296 slot->effective_xmin = xmin_horizon;
298 ReplicationSlotsComputeRequiredXmin(true);
300 LWLockRelease(ProcArrayLock);
302 ReplicationSlotMarkDirty();
303 ReplicationSlotSave();
305 ctx = StartupDecodingContext(NIL, InvalidXLogRecPtr, xmin_horizon,
306 need_full_snapshot, read_page, prepare_write,
307 do_write, update_progress);
309 /* call output plugin initialization callback */
310 old_context = MemoryContextSwitchTo(ctx->context);
311 if (ctx->callbacks.startup_cb != NULL)
312 startup_cb_wrapper(ctx, &ctx->options, true);
313 MemoryContextSwitchTo(old_context);
319 * Create a new decoding context, for a logical slot that has previously been
323 * The LSN at which to start decoding. If InvalidXLogRecPtr, restart
324 * from the slot's confirmed_flush; otherwise, start from the specified
325 * location (but move it forwards to confirmed_flush if it's older than
328 * output_plugin_options
329 * contains options passed to the output plugin.
331 * read_page, prepare_write, do_write, update_progress
332 * callbacks that have to be filled to perform the use-case dependent,
335 * Needs to be called while in a memory context that's at least as long lived
336 * as the decoding context because further memory contexts will be created
339 * Returns an initialized decoding context after calling the output plugin's
342 LogicalDecodingContext *
343 CreateDecodingContext(XLogRecPtr start_lsn,
344 List *output_plugin_options,
345 XLogPageReadCB read_page,
346 LogicalOutputPluginWriterPrepareWrite prepare_write,
347 LogicalOutputPluginWriterWrite do_write,
348 LogicalOutputPluginWriterUpdateProgress update_progress)
350 LogicalDecodingContext *ctx;
351 ReplicationSlot *slot;
352 MemoryContext old_context;
354 /* shorter lines... */
355 slot = MyReplicationSlot;
357 /* first some sanity checks that are unlikely to be violated */
359 elog(ERROR, "cannot perform logical decoding without an acquired slot");
361 /* make sure the passed slot is suitable, these are user facing errors */
362 if (SlotIsPhysical(slot))
364 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
365 (errmsg("cannot use physical replication slot for logical decoding"))));
367 if (slot->data.database != MyDatabaseId)
369 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
370 (errmsg("replication slot \"%s\" was not created in this database",
371 NameStr(slot->data.name)))));
373 if (start_lsn == InvalidXLogRecPtr)
375 /* continue from last position */
376 start_lsn = slot->data.confirmed_flush;
378 else if (start_lsn < slot->data.confirmed_flush)
381 * It might seem like we should error out in this case, but it's
382 * pretty common for a client to acknowledge a LSN it doesn't have to
383 * do anything for, and thus didn't store persistently, because the
384 * xlog records didn't result in anything relevant for logical
385 * decoding. Clients have to be able to do that to support synchronous
388 elog(DEBUG1, "cannot stream from %X/%X, minimum is %X/%X, forwarding",
389 (uint32) (start_lsn >> 32), (uint32) start_lsn,
390 (uint32) (slot->data.confirmed_flush >> 32),
391 (uint32) slot->data.confirmed_flush);
393 start_lsn = slot->data.confirmed_flush;
396 ctx = StartupDecodingContext(output_plugin_options,
397 start_lsn, InvalidTransactionId, false,
398 read_page, prepare_write, do_write,
401 /* call output plugin initialization callback */
402 old_context = MemoryContextSwitchTo(ctx->context);
403 if (ctx->callbacks.startup_cb != NULL)
404 startup_cb_wrapper(ctx, &ctx->options, false);
405 MemoryContextSwitchTo(old_context);
408 (errmsg("starting logical decoding for slot \"%s\"",
409 NameStr(slot->data.name)),
410 errdetail("streaming transactions committing after %X/%X, reading WAL from %X/%X",
411 (uint32) (slot->data.confirmed_flush >> 32),
412 (uint32) slot->data.confirmed_flush,
413 (uint32) (slot->data.restart_lsn >> 32),
414 (uint32) slot->data.restart_lsn)));
420 * Returns true if a consistent initial decoding snapshot has been built.
423 DecodingContextReady(LogicalDecodingContext *ctx)
425 return SnapBuildCurrentState(ctx->snapshot_builder) == SNAPBUILD_CONSISTENT;
429 * Read from the decoding slot, until it is ready to start extracting changes.
432 DecodingContextFindStartpoint(LogicalDecodingContext *ctx)
436 /* Initialize from where to start reading WAL. */
437 startptr = ctx->slot->data.restart_lsn;
439 elog(DEBUG1, "searching for logical decoding starting point, starting at %X/%X",
440 (uint32) (ctx->slot->data.restart_lsn >> 32),
441 (uint32) ctx->slot->data.restart_lsn);
443 /* Wait for a consistent starting point */
449 /* the read_page callback waits for new WAL */
450 record = XLogReadRecord(ctx->reader, startptr, &err);
452 elog(ERROR, "%s", err);
454 elog(ERROR, "no record found"); /* shouldn't happen */
456 startptr = InvalidXLogRecPtr;
458 LogicalDecodingProcessRecord(ctx, ctx->reader);
460 /* only continue till we found a consistent spot */
461 if (DecodingContextReady(ctx))
464 CHECK_FOR_INTERRUPTS();
467 ctx->slot->data.confirmed_flush = ctx->reader->EndRecPtr;
471 * Free a previously allocated decoding context, invoking the shutdown
472 * callback if necessary.
475 FreeDecodingContext(LogicalDecodingContext *ctx)
477 if (ctx->callbacks.shutdown_cb != NULL)
478 shutdown_cb_wrapper(ctx);
480 ReorderBufferFree(ctx->reorder);
481 FreeSnapshotBuilder(ctx->snapshot_builder);
482 XLogReaderFree(ctx->reader);
483 MemoryContextDelete(ctx->context);
487 * Prepare a write using the context's output routine.
490 OutputPluginPrepareWrite(struct LogicalDecodingContext *ctx, bool last_write)
492 if (!ctx->accept_writes)
493 elog(ERROR, "writes are only accepted in commit, begin and change callbacks");
495 ctx->prepare_write(ctx, ctx->write_location, ctx->write_xid, last_write);
496 ctx->prepared_write = true;
500 * Perform a write using the context's output routine.
503 OutputPluginWrite(struct LogicalDecodingContext *ctx, bool last_write)
505 if (!ctx->prepared_write)
506 elog(ERROR, "OutputPluginPrepareWrite needs to be called before OutputPluginWrite");
508 ctx->write(ctx, ctx->write_location, ctx->write_xid, last_write);
509 ctx->prepared_write = false;
513 * Update progress tracking (if supported).
516 OutputPluginUpdateProgress(struct LogicalDecodingContext *ctx)
518 if (!ctx->update_progress)
521 ctx->update_progress(ctx, ctx->write_location, ctx->write_xid);
525 * Load the output plugin, lookup its output plugin init function, and check
526 * that it provides the required callbacks.
529 LoadOutputPlugin(OutputPluginCallbacks *callbacks, char *plugin)
531 LogicalOutputPluginInit plugin_init;
533 plugin_init = (LogicalOutputPluginInit)
534 load_external_function(plugin, "_PG_output_plugin_init", false, NULL);
536 if (plugin_init == NULL)
537 elog(ERROR, "output plugins have to declare the _PG_output_plugin_init symbol");
539 /* ask the output plugin to fill the callback struct */
540 plugin_init(callbacks);
542 if (callbacks->begin_cb == NULL)
543 elog(ERROR, "output plugins have to register a begin callback");
544 if (callbacks->change_cb == NULL)
545 elog(ERROR, "output plugins have to register a change callback");
546 if (callbacks->commit_cb == NULL)
547 elog(ERROR, "output plugins have to register a commit callback");
551 output_plugin_error_callback(void *arg)
553 LogicalErrorCallbackState *state = (LogicalErrorCallbackState *) arg;
555 /* not all callbacks have an associated LSN */
556 if (state->report_location != InvalidXLogRecPtr)
557 errcontext("slot \"%s\", output plugin \"%s\", in the %s callback, associated LSN %X/%X",
558 NameStr(state->ctx->slot->data.name),
559 NameStr(state->ctx->slot->data.plugin),
560 state->callback_name,
561 (uint32) (state->report_location >> 32),
562 (uint32) state->report_location);
564 errcontext("slot \"%s\", output plugin \"%s\", in the %s callback",
565 NameStr(state->ctx->slot->data.name),
566 NameStr(state->ctx->slot->data.plugin),
567 state->callback_name);
571 startup_cb_wrapper(LogicalDecodingContext *ctx, OutputPluginOptions *opt, bool is_init)
573 LogicalErrorCallbackState state;
574 ErrorContextCallback errcallback;
576 /* Push callback + info on the error context stack */
578 state.callback_name = "startup";
579 state.report_location = InvalidXLogRecPtr;
580 errcallback.callback = output_plugin_error_callback;
581 errcallback.arg = (void *) &state;
582 errcallback.previous = error_context_stack;
583 error_context_stack = &errcallback;
585 /* set output state */
586 ctx->accept_writes = false;
588 /* do the actual work: call callback */
589 ctx->callbacks.startup_cb(ctx, opt, is_init);
591 /* Pop the error context stack */
592 error_context_stack = errcallback.previous;
596 shutdown_cb_wrapper(LogicalDecodingContext *ctx)
598 LogicalErrorCallbackState state;
599 ErrorContextCallback errcallback;
601 /* Push callback + info on the error context stack */
603 state.callback_name = "shutdown";
604 state.report_location = InvalidXLogRecPtr;
605 errcallback.callback = output_plugin_error_callback;
606 errcallback.arg = (void *) &state;
607 errcallback.previous = error_context_stack;
608 error_context_stack = &errcallback;
610 /* set output state */
611 ctx->accept_writes = false;
613 /* do the actual work: call callback */
614 ctx->callbacks.shutdown_cb(ctx);
616 /* Pop the error context stack */
617 error_context_stack = errcallback.previous;
622 * Callbacks for ReorderBuffer which add in some more information and then call
623 * output_plugin.h plugins.
626 begin_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn)
628 LogicalDecodingContext *ctx = cache->private_data;
629 LogicalErrorCallbackState state;
630 ErrorContextCallback errcallback;
632 /* Push callback + info on the error context stack */
634 state.callback_name = "begin";
635 state.report_location = txn->first_lsn;
636 errcallback.callback = output_plugin_error_callback;
637 errcallback.arg = (void *) &state;
638 errcallback.previous = error_context_stack;
639 error_context_stack = &errcallback;
641 /* set output state */
642 ctx->accept_writes = true;
643 ctx->write_xid = txn->xid;
644 ctx->write_location = txn->first_lsn;
646 /* do the actual work: call callback */
647 ctx->callbacks.begin_cb(ctx, txn);
649 /* Pop the error context stack */
650 error_context_stack = errcallback.previous;
654 commit_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
655 XLogRecPtr commit_lsn)
657 LogicalDecodingContext *ctx = cache->private_data;
658 LogicalErrorCallbackState state;
659 ErrorContextCallback errcallback;
661 /* Push callback + info on the error context stack */
663 state.callback_name = "commit";
664 state.report_location = txn->final_lsn; /* beginning of commit record */
665 errcallback.callback = output_plugin_error_callback;
666 errcallback.arg = (void *) &state;
667 errcallback.previous = error_context_stack;
668 error_context_stack = &errcallback;
670 /* set output state */
671 ctx->accept_writes = true;
672 ctx->write_xid = txn->xid;
673 ctx->write_location = txn->end_lsn; /* points to the end of the record */
675 /* do the actual work: call callback */
676 ctx->callbacks.commit_cb(ctx, txn, commit_lsn);
678 /* Pop the error context stack */
679 error_context_stack = errcallback.previous;
683 change_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
684 Relation relation, ReorderBufferChange *change)
686 LogicalDecodingContext *ctx = cache->private_data;
687 LogicalErrorCallbackState state;
688 ErrorContextCallback errcallback;
690 /* Push callback + info on the error context stack */
692 state.callback_name = "change";
693 state.report_location = change->lsn;
694 errcallback.callback = output_plugin_error_callback;
695 errcallback.arg = (void *) &state;
696 errcallback.previous = error_context_stack;
697 error_context_stack = &errcallback;
699 /* set output state */
700 ctx->accept_writes = true;
701 ctx->write_xid = txn->xid;
704 * report this change's lsn so replies from clients can give an up2date
705 * answer. This won't ever be enough (and shouldn't be!) to confirm
706 * receipt of this transaction, but it might allow another transaction's
707 * commit to be confirmed with one message.
709 ctx->write_location = change->lsn;
711 ctx->callbacks.change_cb(ctx, txn, relation, change);
713 /* Pop the error context stack */
714 error_context_stack = errcallback.previous;
718 filter_by_origin_cb_wrapper(LogicalDecodingContext *ctx, RepOriginId origin_id)
720 LogicalErrorCallbackState state;
721 ErrorContextCallback errcallback;
724 /* Push callback + info on the error context stack */
726 state.callback_name = "filter_by_origin";
727 state.report_location = InvalidXLogRecPtr;
728 errcallback.callback = output_plugin_error_callback;
729 errcallback.arg = (void *) &state;
730 errcallback.previous = error_context_stack;
731 error_context_stack = &errcallback;
733 /* set output state */
734 ctx->accept_writes = false;
736 /* do the actual work: call callback */
737 ret = ctx->callbacks.filter_by_origin_cb(ctx, origin_id);
739 /* Pop the error context stack */
740 error_context_stack = errcallback.previous;
746 message_cb_wrapper(ReorderBuffer *cache, ReorderBufferTXN *txn,
747 XLogRecPtr message_lsn, bool transactional,
748 const char *prefix, Size message_size, const char *message)
750 LogicalDecodingContext *ctx = cache->private_data;
751 LogicalErrorCallbackState state;
752 ErrorContextCallback errcallback;
754 if (ctx->callbacks.message_cb == NULL)
757 /* Push callback + info on the error context stack */
759 state.callback_name = "message";
760 state.report_location = message_lsn;
761 errcallback.callback = output_plugin_error_callback;
762 errcallback.arg = (void *) &state;
763 errcallback.previous = error_context_stack;
764 error_context_stack = &errcallback;
766 /* set output state */
767 ctx->accept_writes = true;
768 ctx->write_xid = txn != NULL ? txn->xid : InvalidTransactionId;
769 ctx->write_location = message_lsn;
771 /* do the actual work: call callback */
772 ctx->callbacks.message_cb(ctx, txn, message_lsn, transactional, prefix,
773 message_size, message);
775 /* Pop the error context stack */
776 error_context_stack = errcallback.previous;
780 * Set the required catalog xmin horizon for historic snapshots in the current
783 * Note that in the most cases, we won't be able to immediately use the xmin
784 * to increase the xmin horizon: we need to wait till the client has confirmed
785 * receiving current_lsn with LogicalConfirmReceivedLocation().
788 LogicalIncreaseXminForSlot(XLogRecPtr current_lsn, TransactionId xmin)
790 bool updated_xmin = false;
791 ReplicationSlot *slot;
793 slot = MyReplicationSlot;
795 Assert(slot != NULL);
797 SpinLockAcquire(&slot->mutex);
800 * don't overwrite if we already have a newer xmin. This can happen if we
801 * restart decoding in a slot.
803 if (TransactionIdPrecedesOrEquals(xmin, slot->data.catalog_xmin))
808 * If the client has already confirmed up to this lsn, we directly can
809 * mark this as accepted. This can happen if we restart decoding in a
812 else if (current_lsn <= slot->data.confirmed_flush)
814 slot->candidate_catalog_xmin = xmin;
815 slot->candidate_xmin_lsn = current_lsn;
817 /* our candidate can directly be used */
822 * Only increase if the previous values have been applied, otherwise we
823 * might never end up updating if the receiver acks too slowly.
825 else if (slot->candidate_xmin_lsn == InvalidXLogRecPtr)
827 slot->candidate_catalog_xmin = xmin;
828 slot->candidate_xmin_lsn = current_lsn;
830 SpinLockRelease(&slot->mutex);
832 /* candidate already valid with the current flush position, apply */
834 LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
838 * Mark the minimal LSN (restart_lsn) we need to read to replay all
839 * transactions that have not yet committed at current_lsn.
841 * Just like IncreaseRestartDecodingForSlot this only takes effect when the
842 * client has confirmed to have received current_lsn.
845 LogicalIncreaseRestartDecodingForSlot(XLogRecPtr current_lsn, XLogRecPtr restart_lsn)
847 bool updated_lsn = false;
848 ReplicationSlot *slot;
850 slot = MyReplicationSlot;
852 Assert(slot != NULL);
853 Assert(restart_lsn != InvalidXLogRecPtr);
854 Assert(current_lsn != InvalidXLogRecPtr);
856 SpinLockAcquire(&slot->mutex);
858 /* don't overwrite if have a newer restart lsn */
859 if (restart_lsn <= slot->data.restart_lsn)
864 * We might have already flushed far enough to directly accept this lsn,
865 * in this case there is no need to check for existing candidate LSNs
867 else if (current_lsn <= slot->data.confirmed_flush)
869 slot->candidate_restart_valid = current_lsn;
870 slot->candidate_restart_lsn = restart_lsn;
872 /* our candidate can directly be used */
877 * Only increase if the previous values have been applied, otherwise we
878 * might never end up updating if the receiver acks too slowly. A missed
879 * value here will just cause some extra effort after reconnecting.
881 if (slot->candidate_restart_valid == InvalidXLogRecPtr)
883 slot->candidate_restart_valid = current_lsn;
884 slot->candidate_restart_lsn = restart_lsn;
886 elog(DEBUG1, "got new restart lsn %X/%X at %X/%X",
887 (uint32) (restart_lsn >> 32), (uint32) restart_lsn,
888 (uint32) (current_lsn >> 32), (uint32) current_lsn);
892 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",
893 (uint32) (restart_lsn >> 32), (uint32) restart_lsn,
894 (uint32) (current_lsn >> 32), (uint32) current_lsn,
895 (uint32) (slot->candidate_restart_lsn >> 32),
896 (uint32) slot->candidate_restart_lsn,
897 (uint32) (slot->candidate_restart_valid >> 32),
898 (uint32) slot->candidate_restart_valid,
899 (uint32) (slot->data.confirmed_flush >> 32),
900 (uint32) slot->data.confirmed_flush
903 SpinLockRelease(&slot->mutex);
905 /* candidates are already valid with the current flush position, apply */
907 LogicalConfirmReceivedLocation(slot->data.confirmed_flush);
911 * Handle a consumer's confirmation having received all changes up to lsn.
914 LogicalConfirmReceivedLocation(XLogRecPtr lsn)
916 Assert(lsn != InvalidXLogRecPtr);
918 /* Do an unlocked check for candidate_lsn first. */
919 if (MyReplicationSlot->candidate_xmin_lsn != InvalidXLogRecPtr ||
920 MyReplicationSlot->candidate_restart_valid != InvalidXLogRecPtr)
922 bool updated_xmin = false;
923 bool updated_restart = false;
925 SpinLockAcquire(&MyReplicationSlot->mutex);
927 MyReplicationSlot->data.confirmed_flush = lsn;
929 /* if we're past the location required for bumping xmin, do so */
930 if (MyReplicationSlot->candidate_xmin_lsn != InvalidXLogRecPtr &&
931 MyReplicationSlot->candidate_xmin_lsn <= lsn)
934 * We have to write the changed xmin to disk *before* we change
935 * the in-memory value, otherwise after a crash we wouldn't know
936 * that some catalog tuples might have been removed already.
938 * Ensure that by first writing to ->xmin and only update
939 * ->effective_xmin once the new state is synced to disk. After a
940 * crash ->effective_xmin is set to ->xmin.
942 if (TransactionIdIsValid(MyReplicationSlot->candidate_catalog_xmin) &&
943 MyReplicationSlot->data.catalog_xmin != MyReplicationSlot->candidate_catalog_xmin)
945 MyReplicationSlot->data.catalog_xmin = MyReplicationSlot->candidate_catalog_xmin;
946 MyReplicationSlot->candidate_catalog_xmin = InvalidTransactionId;
947 MyReplicationSlot->candidate_xmin_lsn = InvalidXLogRecPtr;
952 if (MyReplicationSlot->candidate_restart_valid != InvalidXLogRecPtr &&
953 MyReplicationSlot->candidate_restart_valid <= lsn)
955 Assert(MyReplicationSlot->candidate_restart_lsn != InvalidXLogRecPtr);
957 MyReplicationSlot->data.restart_lsn = MyReplicationSlot->candidate_restart_lsn;
958 MyReplicationSlot->candidate_restart_lsn = InvalidXLogRecPtr;
959 MyReplicationSlot->candidate_restart_valid = InvalidXLogRecPtr;
960 updated_restart = true;
963 SpinLockRelease(&MyReplicationSlot->mutex);
965 /* first write new xmin to disk, so we know what's up after a crash */
966 if (updated_xmin || updated_restart)
968 ReplicationSlotMarkDirty();
969 ReplicationSlotSave();
970 elog(DEBUG1, "updated xmin: %u restart: %u", updated_xmin, updated_restart);
974 * Now the new xmin is safely on disk, we can let the global value
975 * advance. We do not take ProcArrayLock or similar since we only
976 * advance xmin here and there's not much harm done by a concurrent
977 * computation missing that.
981 SpinLockAcquire(&MyReplicationSlot->mutex);
982 MyReplicationSlot->effective_catalog_xmin = MyReplicationSlot->data.catalog_xmin;
983 SpinLockRelease(&MyReplicationSlot->mutex);
985 ReplicationSlotsComputeRequiredXmin(false);
986 ReplicationSlotsComputeRequiredLSN();
991 SpinLockAcquire(&MyReplicationSlot->mutex);
992 MyReplicationSlot->data.confirmed_flush = lsn;
993 SpinLockRelease(&MyReplicationSlot->mutex);