Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * slotfuncs.c
4 : * Support functions for replication slots
5 : *
6 : * Copyright (c) 2012-2024, PostgreSQL Global Development Group
7 : *
8 : * IDENTIFICATION
9 : * src/backend/replication/slotfuncs.c
10 : *
11 : *-------------------------------------------------------------------------
12 : */
13 : #include "postgres.h"
14 :
15 : #include "access/htup_details.h"
16 : #include "access/xlog_internal.h"
17 : #include "access/xlogrecovery.h"
18 : #include "access/xlogutils.h"
19 : #include "funcapi.h"
20 : #include "miscadmin.h"
21 : #include "replication/decode.h"
22 : #include "replication/logical.h"
23 : #include "replication/slot.h"
24 : #include "replication/slotsync.h"
25 : #include "utils/builtins.h"
26 : #include "utils/guc.h"
27 : #include "utils/inval.h"
28 : #include "utils/pg_lsn.h"
29 : #include "utils/resowner.h"
30 :
31 : /*
32 : * Helper function for creating a new physical replication slot with
33 : * given arguments. Note that this function doesn't release the created
34 : * slot.
35 : *
36 : * If restart_lsn is a valid value, we use it without WAL reservation
37 : * routine. So the caller must guarantee that WAL is available.
38 : */
39 : static void
40 72 : create_physical_replication_slot(char *name, bool immediately_reserve,
41 : bool temporary, XLogRecPtr restart_lsn)
42 : {
43 : Assert(!MyReplicationSlot);
44 :
45 : /* acquire replication slot, this will check for conflicting names */
46 72 : ReplicationSlotCreate(name, false,
47 : temporary ? RS_TEMPORARY : RS_PERSISTENT, false,
48 : false, false);
49 :
50 72 : if (immediately_reserve)
51 : {
52 : /* Reserve WAL as the user asked for it */
53 32 : if (XLogRecPtrIsInvalid(restart_lsn))
54 24 : ReplicationSlotReserveWal();
55 : else
56 8 : MyReplicationSlot->data.restart_lsn = restart_lsn;
57 :
58 : /* Write this slot to disk */
59 32 : ReplicationSlotMarkDirty();
60 32 : ReplicationSlotSave();
61 : }
62 72 : }
63 :
64 : /*
65 : * SQL function for creating a new physical (streaming replication)
66 : * replication slot.
67 : */
68 : Datum
69 64 : pg_create_physical_replication_slot(PG_FUNCTION_ARGS)
70 : {
71 64 : Name name = PG_GETARG_NAME(0);
72 64 : bool immediately_reserve = PG_GETARG_BOOL(1);
73 64 : bool temporary = PG_GETARG_BOOL(2);
74 : Datum values[2];
75 : bool nulls[2];
76 : TupleDesc tupdesc;
77 : HeapTuple tuple;
78 : Datum result;
79 :
80 64 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
81 0 : elog(ERROR, "return type must be a row type");
82 :
83 64 : CheckSlotPermissions();
84 :
85 64 : CheckSlotRequirements();
86 :
87 64 : create_physical_replication_slot(NameStr(*name),
88 : immediately_reserve,
89 : temporary,
90 : InvalidXLogRecPtr);
91 :
92 64 : values[0] = NameGetDatum(&MyReplicationSlot->data.name);
93 64 : nulls[0] = false;
94 :
95 64 : if (immediately_reserve)
96 : {
97 24 : values[1] = LSNGetDatum(MyReplicationSlot->data.restart_lsn);
98 24 : nulls[1] = false;
99 : }
100 : else
101 40 : nulls[1] = true;
102 :
103 64 : tuple = heap_form_tuple(tupdesc, values, nulls);
104 64 : result = HeapTupleGetDatum(tuple);
105 :
106 64 : ReplicationSlotRelease();
107 :
108 64 : PG_RETURN_DATUM(result);
109 : }
110 :
111 :
112 : /*
113 : * Helper function for creating a new logical replication slot with
114 : * given arguments. Note that this function doesn't release the created
115 : * slot.
116 : *
117 : * When find_startpoint is false, the slot's confirmed_flush is not set; it's
118 : * caller's responsibility to ensure it's set to something sensible.
119 : */
120 : static void
121 230 : create_logical_replication_slot(char *name, char *plugin,
122 : bool temporary, bool two_phase,
123 : bool failover,
124 : XLogRecPtr restart_lsn,
125 : bool find_startpoint)
126 : {
127 230 : LogicalDecodingContext *ctx = NULL;
128 :
129 : Assert(!MyReplicationSlot);
130 :
131 : /*
132 : * Acquire a logical decoding slot, this will check for conflicting names.
133 : * Initially create persistent slot as ephemeral - that allows us to
134 : * nicely handle errors during initialization because it'll get dropped if
135 : * this transaction fails. We'll make it persistent at the end. Temporary
136 : * slots can be created as temporary from beginning as they get dropped on
137 : * error as well.
138 : */
139 230 : ReplicationSlotCreate(name, true,
140 : temporary ? RS_TEMPORARY : RS_EPHEMERAL, two_phase,
141 : failover, false);
142 :
143 : /*
144 : * Create logical decoding context to find start point or, if we don't
145 : * need it, to 1) bump slot's restart_lsn and xmin 2) check plugin sanity.
146 : *
147 : * Note: when !find_startpoint this is still important, because it's at
148 : * this point that the output plugin is validated.
149 : */
150 220 : ctx = CreateInitDecodingContext(plugin, NIL,
151 : false, /* just catalogs is OK */
152 : restart_lsn,
153 220 : XL_ROUTINE(.page_read = read_local_xlog_page,
154 : .segment_open = wal_segment_open,
155 : .segment_close = wal_segment_close),
156 : NULL, NULL, NULL);
157 :
158 : /*
159 : * If caller needs us to determine the decoding start point, do so now.
160 : * This might take a while.
161 : */
162 214 : if (find_startpoint)
163 202 : DecodingContextFindStartpoint(ctx);
164 :
165 : /* don't need the decoding context anymore */
166 210 : FreeDecodingContext(ctx);
167 210 : }
168 :
169 : /*
170 : * SQL function for creating a new logical replication slot.
171 : */
172 : Datum
173 218 : pg_create_logical_replication_slot(PG_FUNCTION_ARGS)
174 : {
175 218 : Name name = PG_GETARG_NAME(0);
176 218 : Name plugin = PG_GETARG_NAME(1);
177 218 : bool temporary = PG_GETARG_BOOL(2);
178 218 : bool two_phase = PG_GETARG_BOOL(3);
179 218 : bool failover = PG_GETARG_BOOL(4);
180 : Datum result;
181 : TupleDesc tupdesc;
182 : HeapTuple tuple;
183 : Datum values[2];
184 : bool nulls[2];
185 :
186 218 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
187 0 : elog(ERROR, "return type must be a row type");
188 :
189 218 : CheckSlotPermissions();
190 :
191 216 : CheckLogicalDecodingRequirements();
192 :
193 216 : create_logical_replication_slot(NameStr(*name),
194 216 : NameStr(*plugin),
195 : temporary,
196 : two_phase,
197 : failover,
198 : InvalidXLogRecPtr,
199 : true);
200 :
201 198 : values[0] = NameGetDatum(&MyReplicationSlot->data.name);
202 198 : values[1] = LSNGetDatum(MyReplicationSlot->data.confirmed_flush);
203 :
204 198 : memset(nulls, 0, sizeof(nulls));
205 :
206 198 : tuple = heap_form_tuple(tupdesc, values, nulls);
207 198 : result = HeapTupleGetDatum(tuple);
208 :
209 : /* ok, slot is now fully created, mark it as persistent if needed */
210 198 : if (!temporary)
211 188 : ReplicationSlotPersist();
212 198 : ReplicationSlotRelease();
213 :
214 198 : PG_RETURN_DATUM(result);
215 : }
216 :
217 :
218 : /*
219 : * SQL function for dropping a replication slot.
220 : */
221 : Datum
222 242 : pg_drop_replication_slot(PG_FUNCTION_ARGS)
223 : {
224 242 : Name name = PG_GETARG_NAME(0);
225 :
226 242 : CheckSlotPermissions();
227 :
228 238 : CheckSlotRequirements();
229 :
230 238 : ReplicationSlotDrop(NameStr(*name), true);
231 :
232 226 : PG_RETURN_VOID();
233 : }
234 :
235 : /*
236 : * pg_get_replication_slots - SQL SRF showing all replication slots
237 : * that currently exist on the database cluster.
238 : */
239 : Datum
240 524 : pg_get_replication_slots(PG_FUNCTION_ARGS)
241 : {
242 : #define PG_GET_REPLICATION_SLOTS_COLS 19
243 524 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
244 : XLogRecPtr currlsn;
245 : int slotno;
246 :
247 : /*
248 : * We don't require any special permission to see this function's data
249 : * because nothing should be sensitive. The most critical being the slot
250 : * name, which shouldn't contain anything particularly sensitive.
251 : */
252 :
253 524 : InitMaterializedSRF(fcinfo, 0);
254 :
255 524 : currlsn = GetXLogWriteRecPtr();
256 :
257 524 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
258 4214 : for (slotno = 0; slotno < max_replication_slots; slotno++)
259 : {
260 3690 : ReplicationSlot *slot = &ReplicationSlotCtl->replication_slots[slotno];
261 : ReplicationSlot slot_contents;
262 : Datum values[PG_GET_REPLICATION_SLOTS_COLS];
263 : bool nulls[PG_GET_REPLICATION_SLOTS_COLS];
264 : WALAvailability walstate;
265 : int i;
266 : ReplicationSlotInvalidationCause cause;
267 :
268 3690 : if (!slot->in_use)
269 2898 : continue;
270 :
271 : /* Copy slot contents while holding spinlock, then examine at leisure */
272 792 : SpinLockAcquire(&slot->mutex);
273 792 : slot_contents = *slot;
274 792 : SpinLockRelease(&slot->mutex);
275 :
276 792 : memset(values, 0, sizeof(values));
277 792 : memset(nulls, 0, sizeof(nulls));
278 :
279 792 : i = 0;
280 792 : values[i++] = NameGetDatum(&slot_contents.data.name);
281 :
282 792 : if (slot_contents.data.database == InvalidOid)
283 240 : nulls[i++] = true;
284 : else
285 552 : values[i++] = NameGetDatum(&slot_contents.data.plugin);
286 :
287 792 : if (slot_contents.data.database == InvalidOid)
288 240 : values[i++] = CStringGetTextDatum("physical");
289 : else
290 552 : values[i++] = CStringGetTextDatum("logical");
291 :
292 792 : if (slot_contents.data.database == InvalidOid)
293 240 : nulls[i++] = true;
294 : else
295 552 : values[i++] = ObjectIdGetDatum(slot_contents.data.database);
296 :
297 792 : values[i++] = BoolGetDatum(slot_contents.data.persistency == RS_TEMPORARY);
298 792 : values[i++] = BoolGetDatum(slot_contents.active_pid != 0);
299 :
300 792 : if (slot_contents.active_pid != 0)
301 276 : values[i++] = Int32GetDatum(slot_contents.active_pid);
302 : else
303 516 : nulls[i++] = true;
304 :
305 792 : if (slot_contents.data.xmin != InvalidTransactionId)
306 96 : values[i++] = TransactionIdGetDatum(slot_contents.data.xmin);
307 : else
308 696 : nulls[i++] = true;
309 :
310 792 : if (slot_contents.data.catalog_xmin != InvalidTransactionId)
311 596 : values[i++] = TransactionIdGetDatum(slot_contents.data.catalog_xmin);
312 : else
313 196 : nulls[i++] = true;
314 :
315 792 : if (slot_contents.data.restart_lsn != InvalidXLogRecPtr)
316 766 : values[i++] = LSNGetDatum(slot_contents.data.restart_lsn);
317 : else
318 26 : nulls[i++] = true;
319 :
320 792 : if (slot_contents.data.confirmed_flush != InvalidXLogRecPtr)
321 506 : values[i++] = LSNGetDatum(slot_contents.data.confirmed_flush);
322 : else
323 286 : nulls[i++] = true;
324 :
325 : /*
326 : * If the slot has not been invalidated, test availability from
327 : * restart_lsn.
328 : */
329 792 : if (slot_contents.data.invalidated != RS_INVAL_NONE)
330 62 : walstate = WALAVAIL_REMOVED;
331 : else
332 730 : walstate = GetWALAvailability(slot_contents.data.restart_lsn);
333 :
334 792 : switch (walstate)
335 : {
336 20 : case WALAVAIL_INVALID_LSN:
337 20 : nulls[i++] = true;
338 20 : break;
339 :
340 704 : case WALAVAIL_RESERVED:
341 704 : values[i++] = CStringGetTextDatum("reserved");
342 704 : break;
343 :
344 4 : case WALAVAIL_EXTENDED:
345 4 : values[i++] = CStringGetTextDatum("extended");
346 4 : break;
347 :
348 2 : case WALAVAIL_UNRESERVED:
349 2 : values[i++] = CStringGetTextDatum("unreserved");
350 2 : break;
351 :
352 62 : case WALAVAIL_REMOVED:
353 :
354 : /*
355 : * If we read the restart_lsn long enough ago, maybe that file
356 : * has been removed by now. However, the walsender could have
357 : * moved forward enough that it jumped to another file after
358 : * we looked. If checkpointer signalled the process to
359 : * termination, then it's definitely lost; but if a process is
360 : * still alive, then "unreserved" seems more appropriate.
361 : *
362 : * If we do change it, save the state for safe_wal_size below.
363 : */
364 62 : if (!XLogRecPtrIsInvalid(slot_contents.data.restart_lsn))
365 : {
366 : int pid;
367 :
368 56 : SpinLockAcquire(&slot->mutex);
369 56 : pid = slot->active_pid;
370 56 : slot_contents.data.restart_lsn = slot->data.restart_lsn;
371 56 : SpinLockRelease(&slot->mutex);
372 56 : if (pid != 0)
373 : {
374 0 : values[i++] = CStringGetTextDatum("unreserved");
375 0 : walstate = WALAVAIL_UNRESERVED;
376 0 : break;
377 : }
378 : }
379 62 : values[i++] = CStringGetTextDatum("lost");
380 62 : break;
381 : }
382 :
383 : /*
384 : * safe_wal_size is only computed for slots that have not been lost,
385 : * and only if there's a configured maximum size.
386 : */
387 792 : if (walstate == WALAVAIL_REMOVED || max_slot_wal_keep_size_mb < 0)
388 782 : nulls[i++] = true;
389 : else
390 : {
391 : XLogSegNo targetSeg;
392 : uint64 slotKeepSegs;
393 : uint64 keepSegs;
394 : XLogSegNo failSeg;
395 : XLogRecPtr failLSN;
396 :
397 10 : XLByteToSeg(slot_contents.data.restart_lsn, targetSeg, wal_segment_size);
398 :
399 : /* determine how many segments can be kept by slots */
400 10 : slotKeepSegs = XLogMBVarToSegs(max_slot_wal_keep_size_mb, wal_segment_size);
401 : /* ditto for wal_keep_size */
402 10 : keepSegs = XLogMBVarToSegs(wal_keep_size_mb, wal_segment_size);
403 :
404 : /* if currpos reaches failLSN, we lose our segment */
405 10 : failSeg = targetSeg + Max(slotKeepSegs, keepSegs) + 1;
406 10 : XLogSegNoOffsetToRecPtr(failSeg, 0, wal_segment_size, failLSN);
407 :
408 10 : values[i++] = Int64GetDatum(failLSN - currlsn);
409 : }
410 :
411 792 : values[i++] = BoolGetDatum(slot_contents.data.two_phase);
412 :
413 792 : if (slot_contents.inactive_since > 0)
414 526 : values[i++] = TimestampTzGetDatum(slot_contents.inactive_since);
415 : else
416 266 : nulls[i++] = true;
417 :
418 792 : cause = slot_contents.data.invalidated;
419 :
420 792 : if (SlotIsPhysical(&slot_contents))
421 240 : nulls[i++] = true;
422 : else
423 : {
424 : /*
425 : * rows_removed and wal_level_insufficient are the only two
426 : * reasons for the logical slot's conflict with recovery.
427 : */
428 552 : if (cause == RS_INVAL_HORIZON ||
429 : cause == RS_INVAL_WAL_LEVEL)
430 56 : values[i++] = BoolGetDatum(true);
431 : else
432 496 : values[i++] = BoolGetDatum(false);
433 : }
434 :
435 792 : if (cause == RS_INVAL_NONE)
436 730 : nulls[i++] = true;
437 : else
438 62 : values[i++] = CStringGetTextDatum(SlotInvalidationCauses[cause]);
439 :
440 792 : values[i++] = BoolGetDatum(slot_contents.data.failover);
441 :
442 792 : values[i++] = BoolGetDatum(slot_contents.data.synced);
443 :
444 : Assert(i == PG_GET_REPLICATION_SLOTS_COLS);
445 :
446 792 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
447 : values, nulls);
448 : }
449 :
450 524 : LWLockRelease(ReplicationSlotControlLock);
451 :
452 524 : return (Datum) 0;
453 : }
454 :
455 : /*
456 : * Helper function for advancing our physical replication slot forward.
457 : *
458 : * The LSN position to move to is compared simply to the slot's restart_lsn,
459 : * knowing that any position older than that would be removed by successive
460 : * checkpoints.
461 : */
462 : static XLogRecPtr
463 2 : pg_physical_replication_slot_advance(XLogRecPtr moveto)
464 : {
465 2 : XLogRecPtr startlsn = MyReplicationSlot->data.restart_lsn;
466 2 : XLogRecPtr retlsn = startlsn;
467 :
468 : Assert(moveto != InvalidXLogRecPtr);
469 :
470 2 : if (startlsn < moveto)
471 : {
472 2 : SpinLockAcquire(&MyReplicationSlot->mutex);
473 2 : MyReplicationSlot->data.restart_lsn = moveto;
474 2 : SpinLockRelease(&MyReplicationSlot->mutex);
475 2 : retlsn = moveto;
476 :
477 : /*
478 : * Dirty the slot so as it is written out at the next checkpoint. Note
479 : * that the LSN position advanced may still be lost in the event of a
480 : * crash, but this makes the data consistent after a clean shutdown.
481 : */
482 2 : ReplicationSlotMarkDirty();
483 :
484 : /*
485 : * Wake up logical walsenders holding logical failover slots after
486 : * updating the restart_lsn of the physical slot.
487 : */
488 2 : PhysicalWakeupLogicalWalSnd();
489 : }
490 :
491 2 : return retlsn;
492 : }
493 :
494 : /*
495 : * Helper function for advancing our logical replication slot forward.
496 : *
497 : * The slot's restart_lsn is used as start point for reading records, while
498 : * confirmed_flush is used as base point for the decoding context.
499 : *
500 : * We cannot just do LogicalConfirmReceivedLocation to update confirmed_flush,
501 : * because we need to digest WAL to advance restart_lsn allowing to recycle
502 : * WAL and removal of old catalog tuples. As decoding is done in fast_forward
503 : * mode, no changes are generated anyway.
504 : */
505 : static XLogRecPtr
506 8 : pg_logical_replication_slot_advance(XLogRecPtr moveto)
507 : {
508 : LogicalDecodingContext *ctx;
509 8 : ResourceOwner old_resowner = CurrentResourceOwner;
510 : XLogRecPtr retlsn;
511 :
512 : Assert(moveto != InvalidXLogRecPtr);
513 :
514 8 : PG_TRY();
515 : {
516 : /*
517 : * Create our decoding context in fast_forward mode, passing start_lsn
518 : * as InvalidXLogRecPtr, so that we start processing from my slot's
519 : * confirmed_flush.
520 : */
521 16 : ctx = CreateDecodingContext(InvalidXLogRecPtr,
522 : NIL,
523 : true, /* fast_forward */
524 8 : XL_ROUTINE(.page_read = read_local_xlog_page,
525 : .segment_open = wal_segment_open,
526 : .segment_close = wal_segment_close),
527 : NULL, NULL, NULL);
528 :
529 : /*
530 : * Wait for specified streaming replication standby servers (if any)
531 : * to confirm receipt of WAL up to moveto lsn.
532 : */
533 8 : WaitForStandbyConfirmation(moveto);
534 :
535 : /*
536 : * Start reading at the slot's restart_lsn, which we know to point to
537 : * a valid record.
538 : */
539 8 : XLogBeginRead(ctx->reader, MyReplicationSlot->data.restart_lsn);
540 :
541 : /* invalidate non-timetravel entries */
542 8 : InvalidateSystemCaches();
543 :
544 : /* Decode records until we reach the requested target */
545 318 : while (ctx->reader->EndRecPtr < moveto)
546 : {
547 310 : char *errm = NULL;
548 : XLogRecord *record;
549 :
550 : /*
551 : * Read records. No changes are generated in fast_forward mode,
552 : * but snapbuilder/slot statuses are updated properly.
553 : */
554 310 : record = XLogReadRecord(ctx->reader, &errm);
555 310 : if (errm)
556 0 : elog(ERROR, "could not find record while advancing replication slot: %s",
557 : errm);
558 :
559 : /*
560 : * Process the record. Storage-level changes are ignored in
561 : * fast_forward mode, but other modules (such as snapbuilder)
562 : * might still have critical updates to do.
563 : */
564 310 : if (record)
565 310 : LogicalDecodingProcessRecord(ctx, ctx->reader);
566 :
567 310 : CHECK_FOR_INTERRUPTS();
568 : }
569 :
570 : /*
571 : * Logical decoding could have clobbered CurrentResourceOwner during
572 : * transaction management, so restore the executor's value. (This is
573 : * a kluge, but it's not worth cleaning up right now.)
574 : */
575 8 : CurrentResourceOwner = old_resowner;
576 :
577 8 : if (ctx->reader->EndRecPtr != InvalidXLogRecPtr)
578 : {
579 8 : LogicalConfirmReceivedLocation(moveto);
580 :
581 : /*
582 : * If only the confirmed_flush LSN has changed the slot won't get
583 : * marked as dirty by the above. Callers on the walsender
584 : * interface are expected to keep track of their own progress and
585 : * don't need it written out. But SQL-interface users cannot
586 : * specify their own start positions and it's harder for them to
587 : * keep track of their progress, so we should make more of an
588 : * effort to save it for them.
589 : *
590 : * Dirty the slot so it is written out at the next checkpoint. The
591 : * LSN position advanced to may still be lost on a crash but this
592 : * makes the data consistent after a clean shutdown.
593 : */
594 8 : ReplicationSlotMarkDirty();
595 : }
596 :
597 8 : retlsn = MyReplicationSlot->data.confirmed_flush;
598 :
599 : /* free context, call shutdown callback */
600 8 : FreeDecodingContext(ctx);
601 :
602 8 : InvalidateSystemCaches();
603 : }
604 0 : PG_CATCH();
605 : {
606 : /* clear all timetravel entries */
607 0 : InvalidateSystemCaches();
608 :
609 0 : PG_RE_THROW();
610 : }
611 8 : PG_END_TRY();
612 :
613 8 : return retlsn;
614 : }
615 :
616 : /*
617 : * SQL function for moving the position in a replication slot.
618 : */
619 : Datum
620 14 : pg_replication_slot_advance(PG_FUNCTION_ARGS)
621 : {
622 14 : Name slotname = PG_GETARG_NAME(0);
623 14 : XLogRecPtr moveto = PG_GETARG_LSN(1);
624 : XLogRecPtr endlsn;
625 : XLogRecPtr minlsn;
626 : TupleDesc tupdesc;
627 : Datum values[2];
628 : bool nulls[2];
629 : HeapTuple tuple;
630 : Datum result;
631 :
632 : Assert(!MyReplicationSlot);
633 :
634 14 : CheckSlotPermissions();
635 :
636 14 : if (XLogRecPtrIsInvalid(moveto))
637 2 : ereport(ERROR,
638 : (errmsg("invalid target WAL LSN")));
639 :
640 : /* Build a tuple descriptor for our result type */
641 12 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
642 0 : elog(ERROR, "return type must be a row type");
643 :
644 : /*
645 : * We can't move slot past what's been flushed/replayed so clamp the
646 : * target position accordingly.
647 : */
648 12 : if (!RecoveryInProgress())
649 12 : moveto = Min(moveto, GetFlushRecPtr(NULL));
650 : else
651 0 : moveto = Min(moveto, GetXLogReplayRecPtr(NULL));
652 :
653 : /* Acquire the slot so we "own" it */
654 12 : ReplicationSlotAcquire(NameStr(*slotname), true);
655 :
656 : /* A slot whose restart_lsn has never been reserved cannot be advanced */
657 12 : if (XLogRecPtrIsInvalid(MyReplicationSlot->data.restart_lsn))
658 2 : ereport(ERROR,
659 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
660 : errmsg("replication slot \"%s\" cannot be advanced",
661 : NameStr(*slotname)),
662 : errdetail("This slot has never previously reserved WAL, or it has been invalidated.")));
663 :
664 : /*
665 : * Check if the slot is not moving backwards. Physical slots rely simply
666 : * on restart_lsn as a minimum point, while logical slots have confirmed
667 : * consumption up to confirmed_flush, meaning that in both cases data
668 : * older than that is not available anymore.
669 : */
670 10 : if (OidIsValid(MyReplicationSlot->data.database))
671 8 : minlsn = MyReplicationSlot->data.confirmed_flush;
672 : else
673 2 : minlsn = MyReplicationSlot->data.restart_lsn;
674 :
675 10 : if (moveto < minlsn)
676 0 : ereport(ERROR,
677 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
678 : errmsg("cannot advance replication slot to %X/%X, minimum is %X/%X",
679 : LSN_FORMAT_ARGS(moveto), LSN_FORMAT_ARGS(minlsn))));
680 :
681 : /* Do the actual slot update, depending on the slot type */
682 10 : if (OidIsValid(MyReplicationSlot->data.database))
683 8 : endlsn = pg_logical_replication_slot_advance(moveto);
684 : else
685 2 : endlsn = pg_physical_replication_slot_advance(moveto);
686 :
687 10 : values[0] = NameGetDatum(&MyReplicationSlot->data.name);
688 10 : nulls[0] = false;
689 :
690 : /*
691 : * Recompute the minimum LSN and xmin across all slots to adjust with the
692 : * advancing potentially done.
693 : */
694 10 : ReplicationSlotsComputeRequiredXmin(false);
695 10 : ReplicationSlotsComputeRequiredLSN();
696 :
697 10 : ReplicationSlotRelease();
698 :
699 : /* Return the reached position. */
700 10 : values[1] = LSNGetDatum(endlsn);
701 10 : nulls[1] = false;
702 :
703 10 : tuple = heap_form_tuple(tupdesc, values, nulls);
704 10 : result = HeapTupleGetDatum(tuple);
705 :
706 10 : PG_RETURN_DATUM(result);
707 : }
708 :
709 : /*
710 : * Helper function of copying a replication slot.
711 : */
712 : static Datum
713 28 : copy_replication_slot(FunctionCallInfo fcinfo, bool logical_slot)
714 : {
715 28 : Name src_name = PG_GETARG_NAME(0);
716 28 : Name dst_name = PG_GETARG_NAME(1);
717 28 : ReplicationSlot *src = NULL;
718 : ReplicationSlot first_slot_contents;
719 : ReplicationSlot second_slot_contents;
720 : XLogRecPtr src_restart_lsn;
721 : bool src_islogical;
722 : bool temporary;
723 : char *plugin;
724 : Datum values[2];
725 : bool nulls[2];
726 : Datum result;
727 : TupleDesc tupdesc;
728 : HeapTuple tuple;
729 :
730 28 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
731 0 : elog(ERROR, "return type must be a row type");
732 :
733 28 : CheckSlotPermissions();
734 :
735 28 : if (logical_slot)
736 16 : CheckLogicalDecodingRequirements();
737 : else
738 12 : CheckSlotRequirements();
739 :
740 28 : LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
741 :
742 : /*
743 : * We need to prevent the source slot's reserved WAL from being removed,
744 : * but we don't want to lock that slot for very long, and it can advance
745 : * in the meantime. So obtain the source slot's data, and create a new
746 : * slot using its restart_lsn. Afterwards we lock the source slot again
747 : * and verify that the data we copied (name, type) has not changed
748 : * incompatibly. No inconvenient WAL removal can occur once the new slot
749 : * is created -- but since WAL removal could have occurred before we
750 : * managed to create the new slot, we advance the new slot's restart_lsn
751 : * to the source slot's updated restart_lsn the second time we lock it.
752 : */
753 30 : for (int i = 0; i < max_replication_slots; i++)
754 : {
755 30 : ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
756 :
757 30 : if (s->in_use && strcmp(NameStr(s->data.name), NameStr(*src_name)) == 0)
758 : {
759 : /* Copy the slot contents while holding spinlock */
760 28 : SpinLockAcquire(&s->mutex);
761 28 : first_slot_contents = *s;
762 28 : SpinLockRelease(&s->mutex);
763 28 : src = s;
764 28 : break;
765 : }
766 : }
767 :
768 28 : LWLockRelease(ReplicationSlotControlLock);
769 :
770 28 : if (src == NULL)
771 0 : ereport(ERROR,
772 : (errcode(ERRCODE_UNDEFINED_OBJECT),
773 : errmsg("replication slot \"%s\" does not exist", NameStr(*src_name))));
774 :
775 28 : src_islogical = SlotIsLogical(&first_slot_contents);
776 28 : src_restart_lsn = first_slot_contents.data.restart_lsn;
777 28 : temporary = (first_slot_contents.data.persistency == RS_TEMPORARY);
778 28 : plugin = logical_slot ? NameStr(first_slot_contents.data.plugin) : NULL;
779 :
780 : /* Check type of replication slot */
781 28 : if (src_islogical != logical_slot)
782 4 : ereport(ERROR,
783 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
784 : src_islogical ?
785 : errmsg("cannot copy physical replication slot \"%s\" as a logical replication slot",
786 : NameStr(*src_name)) :
787 : errmsg("cannot copy logical replication slot \"%s\" as a physical replication slot",
788 : NameStr(*src_name))));
789 :
790 : /* Copying non-reserved slot doesn't make sense */
791 24 : if (XLogRecPtrIsInvalid(src_restart_lsn))
792 2 : ereport(ERROR,
793 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
794 : errmsg("cannot copy a replication slot that doesn't reserve WAL")));
795 :
796 : /* Overwrite params from optional arguments */
797 22 : if (PG_NARGS() >= 3)
798 12 : temporary = PG_GETARG_BOOL(2);
799 22 : if (PG_NARGS() >= 4)
800 : {
801 : Assert(logical_slot);
802 8 : plugin = NameStr(*(PG_GETARG_NAME(3)));
803 : }
804 :
805 : /* Create new slot and acquire it */
806 22 : if (logical_slot)
807 : {
808 : /*
809 : * We must not try to read WAL, since we haven't reserved it yet --
810 : * hence pass find_startpoint false. confirmed_flush will be set
811 : * below, by copying from the source slot.
812 : *
813 : * To avoid potential issues with the slot synchronization where the
814 : * restart_lsn of a replication slot can go backward, we set the
815 : * failover option to false here. This situation occurs when a slot
816 : * on the primary server is dropped and immediately replaced with a
817 : * new slot of the same name, created by copying from another existing
818 : * slot. However, the slot synchronization will only observe the
819 : * restart_lsn of the same slot going backward.
820 : */
821 14 : create_logical_replication_slot(NameStr(*dst_name),
822 : plugin,
823 : temporary,
824 : false,
825 : false,
826 : src_restart_lsn,
827 : false);
828 : }
829 : else
830 8 : create_physical_replication_slot(NameStr(*dst_name),
831 : true,
832 : temporary,
833 : src_restart_lsn);
834 :
835 : /*
836 : * Update the destination slot to current values of the source slot;
837 : * recheck that the source slot is still the one we saw previously.
838 : */
839 : {
840 : TransactionId copy_effective_xmin;
841 : TransactionId copy_effective_catalog_xmin;
842 : TransactionId copy_xmin;
843 : TransactionId copy_catalog_xmin;
844 : XLogRecPtr copy_restart_lsn;
845 : XLogRecPtr copy_confirmed_flush;
846 : bool copy_islogical;
847 : char *copy_name;
848 :
849 : /* Copy data of source slot again */
850 20 : SpinLockAcquire(&src->mutex);
851 20 : second_slot_contents = *src;
852 20 : SpinLockRelease(&src->mutex);
853 :
854 20 : copy_effective_xmin = second_slot_contents.effective_xmin;
855 20 : copy_effective_catalog_xmin = second_slot_contents.effective_catalog_xmin;
856 :
857 20 : copy_xmin = second_slot_contents.data.xmin;
858 20 : copy_catalog_xmin = second_slot_contents.data.catalog_xmin;
859 20 : copy_restart_lsn = second_slot_contents.data.restart_lsn;
860 20 : copy_confirmed_flush = second_slot_contents.data.confirmed_flush;
861 :
862 : /* for existence check */
863 20 : copy_name = NameStr(second_slot_contents.data.name);
864 20 : copy_islogical = SlotIsLogical(&second_slot_contents);
865 :
866 : /*
867 : * Check if the source slot still exists and is valid. We regard it as
868 : * invalid if the type of replication slot or name has been changed,
869 : * or the restart_lsn either is invalid or has gone backward. (The
870 : * restart_lsn could go backwards if the source slot is dropped and
871 : * copied from an older slot during installation.)
872 : *
873 : * Since erroring out will release and drop the destination slot we
874 : * don't need to release it here.
875 : */
876 20 : if (copy_restart_lsn < src_restart_lsn ||
877 20 : src_islogical != copy_islogical ||
878 20 : strcmp(copy_name, NameStr(*src_name)) != 0)
879 0 : ereport(ERROR,
880 : (errmsg("could not copy replication slot \"%s\"",
881 : NameStr(*src_name)),
882 : errdetail("The source replication slot was modified incompatibly during the copy operation.")));
883 :
884 : /* The source slot must have a consistent snapshot */
885 20 : if (src_islogical && XLogRecPtrIsInvalid(copy_confirmed_flush))
886 0 : ereport(ERROR,
887 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
888 : errmsg("cannot copy unfinished logical replication slot \"%s\"",
889 : NameStr(*src_name)),
890 : errhint("Retry when the source replication slot's confirmed_flush_lsn is valid.")));
891 :
892 : /* Install copied values again */
893 20 : SpinLockAcquire(&MyReplicationSlot->mutex);
894 20 : MyReplicationSlot->effective_xmin = copy_effective_xmin;
895 20 : MyReplicationSlot->effective_catalog_xmin = copy_effective_catalog_xmin;
896 :
897 20 : MyReplicationSlot->data.xmin = copy_xmin;
898 20 : MyReplicationSlot->data.catalog_xmin = copy_catalog_xmin;
899 20 : MyReplicationSlot->data.restart_lsn = copy_restart_lsn;
900 20 : MyReplicationSlot->data.confirmed_flush = copy_confirmed_flush;
901 20 : SpinLockRelease(&MyReplicationSlot->mutex);
902 :
903 20 : ReplicationSlotMarkDirty();
904 20 : ReplicationSlotsComputeRequiredXmin(false);
905 20 : ReplicationSlotsComputeRequiredLSN();
906 20 : ReplicationSlotSave();
907 :
908 : #ifdef USE_ASSERT_CHECKING
909 : /* Check that the restart_lsn is available */
910 : {
911 : XLogSegNo segno;
912 :
913 : XLByteToSeg(copy_restart_lsn, segno, wal_segment_size);
914 : Assert(XLogGetLastRemovedSegno() < segno);
915 : }
916 : #endif
917 : }
918 :
919 : /* target slot fully created, mark as persistent if needed */
920 20 : if (logical_slot && !temporary)
921 6 : ReplicationSlotPersist();
922 :
923 : /* All done. Set up the return values */
924 20 : values[0] = NameGetDatum(dst_name);
925 20 : nulls[0] = false;
926 20 : if (!XLogRecPtrIsInvalid(MyReplicationSlot->data.confirmed_flush))
927 : {
928 12 : values[1] = LSNGetDatum(MyReplicationSlot->data.confirmed_flush);
929 12 : nulls[1] = false;
930 : }
931 : else
932 8 : nulls[1] = true;
933 :
934 20 : tuple = heap_form_tuple(tupdesc, values, nulls);
935 20 : result = HeapTupleGetDatum(tuple);
936 :
937 20 : ReplicationSlotRelease();
938 :
939 20 : PG_RETURN_DATUM(result);
940 : }
941 :
942 : /* The wrappers below are all to appease opr_sanity */
943 : Datum
944 8 : pg_copy_logical_replication_slot_a(PG_FUNCTION_ARGS)
945 : {
946 8 : return copy_replication_slot(fcinfo, true);
947 : }
948 :
949 : Datum
950 0 : pg_copy_logical_replication_slot_b(PG_FUNCTION_ARGS)
951 : {
952 0 : return copy_replication_slot(fcinfo, true);
953 : }
954 :
955 : Datum
956 8 : pg_copy_logical_replication_slot_c(PG_FUNCTION_ARGS)
957 : {
958 8 : return copy_replication_slot(fcinfo, true);
959 : }
960 :
961 : Datum
962 4 : pg_copy_physical_replication_slot_a(PG_FUNCTION_ARGS)
963 : {
964 4 : return copy_replication_slot(fcinfo, false);
965 : }
966 :
967 : Datum
968 8 : pg_copy_physical_replication_slot_b(PG_FUNCTION_ARGS)
969 : {
970 8 : return copy_replication_slot(fcinfo, false);
971 : }
972 :
973 : /*
974 : * Synchronize failover enabled replication slots to a standby server
975 : * from the primary server.
976 : */
977 : Datum
978 16 : pg_sync_replication_slots(PG_FUNCTION_ARGS)
979 : {
980 : WalReceiverConn *wrconn;
981 : char *err;
982 : StringInfoData app_name;
983 :
984 16 : CheckSlotPermissions();
985 :
986 14 : if (!RecoveryInProgress())
987 2 : ereport(ERROR,
988 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
989 : errmsg("replication slots can only be synchronized to a standby server"));
990 :
991 12 : ValidateSlotSyncParams(ERROR);
992 :
993 : /* Load the libpq-specific functions */
994 12 : load_file("libpqwalreceiver", false);
995 :
996 12 : (void) CheckAndGetDbnameFromConninfo();
997 :
998 10 : initStringInfo(&app_name);
999 10 : if (cluster_name[0])
1000 10 : appendStringInfo(&app_name, "%s_slotsync", cluster_name);
1001 : else
1002 0 : appendStringInfoString(&app_name, "slotsync");
1003 :
1004 : /* Connect to the primary server. */
1005 10 : wrconn = walrcv_connect(PrimaryConnInfo, false, false, false,
1006 : app_name.data, &err);
1007 10 : pfree(app_name.data);
1008 :
1009 10 : if (!wrconn)
1010 0 : ereport(ERROR,
1011 : errcode(ERRCODE_CONNECTION_FAILURE),
1012 : errmsg("could not connect to the primary server: %s", err));
1013 :
1014 10 : SyncReplicationSlots(wrconn);
1015 :
1016 8 : walrcv_disconnect(wrconn);
1017 :
1018 8 : PG_RETURN_VOID();
1019 : }
|