Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * execReplication.c
4 : * miscellaneous executor routines for logical replication
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * IDENTIFICATION
10 : * src/backend/executor/execReplication.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 :
15 : #include "postgres.h"
16 :
17 : #include "access/amapi.h"
18 : #include "access/commit_ts.h"
19 : #include "access/genam.h"
20 : #include "access/gist.h"
21 : #include "access/relscan.h"
22 : #include "access/tableam.h"
23 : #include "access/transam.h"
24 : #include "access/xact.h"
25 : #include "access/heapam.h"
26 : #include "catalog/pg_am_d.h"
27 : #include "commands/trigger.h"
28 : #include "executor/executor.h"
29 : #include "executor/nodeModifyTable.h"
30 : #include "replication/conflict.h"
31 : #include "replication/logicalrelation.h"
32 : #include "storage/lmgr.h"
33 : #include "utils/builtins.h"
34 : #include "utils/lsyscache.h"
35 : #include "utils/rel.h"
36 : #include "utils/snapmgr.h"
37 : #include "utils/syscache.h"
38 : #include "utils/typcache.h"
39 :
40 :
41 : static bool tuples_equal(TupleTableSlot *slot1, TupleTableSlot *slot2,
42 : TypeCacheEntry **eq, Bitmapset *columns);
43 :
44 : /*
45 : * Setup a ScanKey for a search in the relation 'rel' for a tuple 'key' that
46 : * is setup to match 'rel' (*NOT* idxrel!).
47 : *
48 : * Returns how many columns to use for the index scan.
49 : *
50 : * This is not a generic routine, idxrel must be PK, RI, or an index that can be
51 : * used for a REPLICA IDENTITY FULL table. See FindUsableIndexForReplicaIdentityFull()
52 : * for details.
53 : *
54 : * By definition, replication identity of a rel meets all limitations associated
55 : * with that. Note that any other index could also meet these limitations.
56 : */
57 : static int
58 72119 : build_replindex_scan_key(ScanKey skey, Relation rel, Relation idxrel,
59 : TupleTableSlot *searchslot)
60 : {
61 : int index_attoff;
62 72119 : int skey_attoff = 0;
63 : Datum indclassDatum;
64 : oidvector *opclass;
65 72119 : int2vector *indkey = &idxrel->rd_index->indkey;
66 :
67 72119 : indclassDatum = SysCacheGetAttrNotNull(INDEXRELID, idxrel->rd_indextuple,
68 : Anum_pg_index_indclass);
69 72119 : opclass = (oidvector *) DatumGetPointer(indclassDatum);
70 :
71 : /* Build scankey for every non-expression attribute in the index. */
72 144261 : for (index_attoff = 0; index_attoff < IndexRelationGetNumberOfKeyAttributes(idxrel);
73 72142 : index_attoff++)
74 : {
75 : Oid operator;
76 : Oid optype;
77 : Oid opfamily;
78 : RegProcedure regop;
79 72142 : int table_attno = indkey->values[index_attoff];
80 : StrategyNumber eq_strategy;
81 :
82 72142 : if (!AttributeNumberIsValid(table_attno))
83 : {
84 : /*
85 : * XXX: Currently, we don't support expressions in the scan key,
86 : * see code below.
87 : */
88 2 : continue;
89 : }
90 :
91 : /*
92 : * Load the operator info. We need this to get the equality operator
93 : * function for the scan key.
94 : */
95 72140 : optype = get_opclass_input_type(opclass->values[index_attoff]);
96 72140 : opfamily = get_opclass_family(opclass->values[index_attoff]);
97 72140 : eq_strategy = IndexAmTranslateCompareType(COMPARE_EQ, idxrel->rd_rel->relam, opfamily, false);
98 72140 : operator = get_opfamily_member(opfamily, optype,
99 : optype,
100 : eq_strategy);
101 :
102 72140 : if (!OidIsValid(operator))
103 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
104 : eq_strategy, optype, optype, opfamily);
105 :
106 72140 : regop = get_opcode(operator);
107 :
108 : /* Initialize the scankey. */
109 72140 : ScanKeyInit(&skey[skey_attoff],
110 72140 : index_attoff + 1,
111 : eq_strategy,
112 : regop,
113 72140 : searchslot->tts_values[table_attno - 1]);
114 :
115 72140 : skey[skey_attoff].sk_collation = idxrel->rd_indcollation[index_attoff];
116 :
117 : /* Check for null value. */
118 72140 : if (searchslot->tts_isnull[table_attno - 1])
119 1 : skey[skey_attoff].sk_flags |= (SK_ISNULL | SK_SEARCHNULL);
120 :
121 72140 : skey_attoff++;
122 : }
123 :
124 : /* There must always be at least one attribute for the index scan. */
125 : Assert(skey_attoff > 0);
126 :
127 72119 : return skey_attoff;
128 : }
129 :
130 :
131 : /*
132 : * Helper function to check if it is necessary to re-fetch and lock the tuple
133 : * due to concurrent modifications. This function should be called after
134 : * invoking table_tuple_lock.
135 : */
136 : static bool
137 72281 : should_refetch_tuple(TM_Result res, TM_FailureData *tmfd)
138 : {
139 72281 : bool refetch = false;
140 :
141 72281 : switch (res)
142 : {
143 72281 : case TM_Ok:
144 72281 : break;
145 0 : case TM_Updated:
146 : /* XXX: Improve handling here */
147 0 : if (ItemPointerIndicatesMovedPartitions(&tmfd->ctid))
148 0 : ereport(LOG,
149 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
150 : errmsg("tuple to be locked was already moved to another partition due to concurrent update, retrying")));
151 : else
152 0 : ereport(LOG,
153 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
154 : errmsg("concurrent update, retrying")));
155 0 : refetch = true;
156 0 : break;
157 0 : case TM_Deleted:
158 : /* XXX: Improve handling here */
159 0 : ereport(LOG,
160 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
161 : errmsg("concurrent delete, retrying")));
162 0 : refetch = true;
163 0 : break;
164 0 : case TM_Invisible:
165 0 : elog(ERROR, "attempted to lock invisible tuple");
166 : break;
167 0 : default:
168 0 : elog(ERROR, "unexpected table_tuple_lock status: %u", res);
169 : break;
170 : }
171 :
172 72281 : return refetch;
173 : }
174 :
175 : /*
176 : * Search the relation 'rel' for tuple using the index.
177 : *
178 : * If a matching tuple is found, lock it with lockmode, fill the slot with its
179 : * contents, and return true. Return false otherwise.
180 : */
181 : bool
182 72118 : RelationFindReplTupleByIndex(Relation rel, Oid idxoid,
183 : LockTupleMode lockmode,
184 : TupleTableSlot *searchslot,
185 : TupleTableSlot *outslot)
186 : {
187 : ScanKeyData skey[INDEX_MAX_KEYS];
188 : int skey_attoff;
189 : IndexScanDesc scan;
190 : SnapshotData snap;
191 : TransactionId xwait;
192 : Relation idxrel;
193 : bool found;
194 72118 : TypeCacheEntry **eq = NULL;
195 : bool isIdxSafeToSkipDuplicates;
196 :
197 : /* Open the index. */
198 72118 : idxrel = index_open(idxoid, RowExclusiveLock);
199 :
200 72118 : isIdxSafeToSkipDuplicates = (GetRelationIdentityOrPK(rel) == idxoid);
201 :
202 72118 : InitDirtySnapshot(snap);
203 :
204 : /* Build scan key. */
205 72118 : skey_attoff = build_replindex_scan_key(skey, rel, idxrel, searchslot);
206 :
207 : /* Start an index scan. */
208 72118 : scan = index_beginscan(rel, idxrel, &snap, NULL, skey_attoff, 0);
209 :
210 0 : retry:
211 72118 : found = false;
212 :
213 72118 : index_rescan(scan, skey, skey_attoff, NULL, 0);
214 :
215 : /* Try to find the tuple */
216 72118 : while (index_getnext_slot(scan, ForwardScanDirection, outslot))
217 : {
218 : /*
219 : * Avoid expensive equality check if the index is primary key or
220 : * replica identity index.
221 : */
222 72088 : if (!isIdxSafeToSkipDuplicates)
223 : {
224 17 : if (eq == NULL)
225 17 : eq = palloc0_array(TypeCacheEntry *, outslot->tts_tupleDescriptor->natts);
226 :
227 17 : if (!tuples_equal(outslot, searchslot, eq, NULL))
228 0 : continue;
229 : }
230 :
231 72088 : ExecMaterializeSlot(outslot);
232 :
233 144176 : xwait = TransactionIdIsValid(snap.xmin) ?
234 72088 : snap.xmin : snap.xmax;
235 :
236 : /*
237 : * If the tuple is locked, wait for locking transaction to finish and
238 : * retry.
239 : */
240 72088 : if (TransactionIdIsValid(xwait))
241 : {
242 0 : XactLockTableWait(xwait, NULL, NULL, XLTW_None);
243 0 : goto retry;
244 : }
245 :
246 : /* Found our tuple and it's not locked */
247 72088 : found = true;
248 72088 : break;
249 : }
250 :
251 : /* Found tuple, try to lock it in the lockmode. */
252 72118 : if (found)
253 : {
254 : TM_FailureData tmfd;
255 : TM_Result res;
256 :
257 72088 : PushActiveSnapshot(GetLatestSnapshot());
258 :
259 72088 : res = table_tuple_lock(rel, &(outslot->tts_tid), GetActiveSnapshot(),
260 : outslot,
261 : GetCurrentCommandId(false),
262 : lockmode,
263 : LockWaitBlock,
264 : 0 /* don't follow updates */ ,
265 : &tmfd);
266 :
267 72088 : PopActiveSnapshot();
268 :
269 72088 : if (should_refetch_tuple(res, &tmfd))
270 0 : goto retry;
271 : }
272 :
273 72118 : index_endscan(scan);
274 :
275 : /* Don't release lock until commit. */
276 72118 : index_close(idxrel, NoLock);
277 :
278 72118 : return found;
279 : }
280 :
281 : /*
282 : * Compare the tuples in the slots by checking if they have equal values.
283 : *
284 : * If 'columns' is not null, only the columns specified within it will be
285 : * considered for the equality check, ignoring all other columns.
286 : */
287 : static bool
288 105325 : tuples_equal(TupleTableSlot *slot1, TupleTableSlot *slot2,
289 : TypeCacheEntry **eq, Bitmapset *columns)
290 : {
291 : int attrnum;
292 :
293 : Assert(slot1->tts_tupleDescriptor->natts ==
294 : slot2->tts_tupleDescriptor->natts);
295 :
296 105325 : slot_getallattrs(slot1);
297 105325 : slot_getallattrs(slot2);
298 :
299 : /* Check equality of the attributes. */
300 105527 : for (attrnum = 0; attrnum < slot1->tts_tupleDescriptor->natts; attrnum++)
301 : {
302 : Form_pg_attribute att;
303 : TypeCacheEntry *typentry;
304 :
305 105362 : att = TupleDescAttr(slot1->tts_tupleDescriptor, attrnum);
306 :
307 : /*
308 : * Ignore dropped and generated columns as the publisher doesn't send
309 : * those
310 : */
311 105362 : if (att->attisdropped || att->attgenerated)
312 1 : continue;
313 :
314 : /*
315 : * Ignore columns that are not listed for checking.
316 : */
317 105361 : if (columns &&
318 0 : !bms_is_member(att->attnum - FirstLowInvalidHeapAttributeNumber,
319 : columns))
320 0 : continue;
321 :
322 : /*
323 : * If one value is NULL and other is not, then they are certainly not
324 : * equal
325 : */
326 105361 : if (slot1->tts_isnull[attrnum] != slot2->tts_isnull[attrnum])
327 0 : return false;
328 :
329 : /*
330 : * If both are NULL, they can be considered equal.
331 : */
332 105361 : if (slot1->tts_isnull[attrnum] || slot2->tts_isnull[attrnum])
333 1 : continue;
334 :
335 105360 : typentry = eq[attrnum];
336 105360 : if (typentry == NULL)
337 : {
338 202 : typentry = lookup_type_cache(att->atttypid,
339 : TYPECACHE_EQ_OPR_FINFO);
340 202 : if (!OidIsValid(typentry->eq_opr_finfo.fn_oid))
341 0 : ereport(ERROR,
342 : (errcode(ERRCODE_UNDEFINED_FUNCTION),
343 : errmsg("could not identify an equality operator for type %s",
344 : format_type_be(att->atttypid))));
345 202 : eq[attrnum] = typentry;
346 : }
347 :
348 105360 : if (!DatumGetBool(FunctionCall2Coll(&typentry->eq_opr_finfo,
349 : att->attcollation,
350 105360 : slot1->tts_values[attrnum],
351 105360 : slot2->tts_values[attrnum])))
352 105160 : return false;
353 : }
354 :
355 165 : return true;
356 : }
357 :
358 : /*
359 : * Search the relation 'rel' for tuple using the sequential scan.
360 : *
361 : * If a matching tuple is found, lock it with lockmode, fill the slot with its
362 : * contents, and return true. Return false otherwise.
363 : *
364 : * Note that this stops on the first matching tuple.
365 : *
366 : * This can obviously be quite slow on tables that have more than few rows.
367 : */
368 : bool
369 150 : RelationFindReplTupleSeq(Relation rel, LockTupleMode lockmode,
370 : TupleTableSlot *searchslot, TupleTableSlot *outslot)
371 : {
372 : TupleTableSlot *scanslot;
373 : TableScanDesc scan;
374 : SnapshotData snap;
375 : TypeCacheEntry **eq;
376 : TransactionId xwait;
377 : bool found;
378 150 : TupleDesc desc PG_USED_FOR_ASSERTS_ONLY = RelationGetDescr(rel);
379 :
380 : Assert(equalTupleDescs(desc, outslot->tts_tupleDescriptor));
381 :
382 150 : eq = palloc0_array(TypeCacheEntry *, outslot->tts_tupleDescriptor->natts);
383 :
384 : /* Start a heap scan. */
385 150 : InitDirtySnapshot(snap);
386 150 : scan = table_beginscan(rel, &snap, 0, NULL);
387 150 : scanslot = table_slot_create(rel, NULL);
388 :
389 0 : retry:
390 150 : found = false;
391 :
392 150 : table_rescan(scan, NULL);
393 :
394 : /* Try to find the tuple */
395 105309 : while (table_scan_getnextslot(scan, ForwardScanDirection, scanslot))
396 : {
397 105305 : if (!tuples_equal(scanslot, searchslot, eq, NULL))
398 105159 : continue;
399 :
400 146 : found = true;
401 146 : ExecCopySlot(outslot, scanslot);
402 :
403 292 : xwait = TransactionIdIsValid(snap.xmin) ?
404 146 : snap.xmin : snap.xmax;
405 :
406 : /*
407 : * If the tuple is locked, wait for locking transaction to finish and
408 : * retry.
409 : */
410 146 : if (TransactionIdIsValid(xwait))
411 : {
412 0 : XactLockTableWait(xwait, NULL, NULL, XLTW_None);
413 0 : goto retry;
414 : }
415 :
416 : /* Found our tuple and it's not locked */
417 146 : break;
418 : }
419 :
420 : /* Found tuple, try to lock it in the lockmode. */
421 150 : if (found)
422 : {
423 : TM_FailureData tmfd;
424 : TM_Result res;
425 :
426 146 : PushActiveSnapshot(GetLatestSnapshot());
427 :
428 146 : res = table_tuple_lock(rel, &(outslot->tts_tid), GetActiveSnapshot(),
429 : outslot,
430 : GetCurrentCommandId(false),
431 : lockmode,
432 : LockWaitBlock,
433 : 0 /* don't follow updates */ ,
434 : &tmfd);
435 :
436 146 : PopActiveSnapshot();
437 :
438 146 : if (should_refetch_tuple(res, &tmfd))
439 0 : goto retry;
440 : }
441 :
442 150 : table_endscan(scan);
443 150 : ExecDropSingleTupleTableSlot(scanslot);
444 :
445 150 : return found;
446 : }
447 :
448 : /*
449 : * Build additional index information necessary for conflict detection.
450 : */
451 : static void
452 49 : BuildConflictIndexInfo(ResultRelInfo *resultRelInfo, Oid conflictindex)
453 : {
454 144 : for (int i = 0; i < resultRelInfo->ri_NumIndices; i++)
455 : {
456 95 : Relation indexRelation = resultRelInfo->ri_IndexRelationDescs[i];
457 95 : IndexInfo *indexRelationInfo = resultRelInfo->ri_IndexRelationInfo[i];
458 :
459 95 : if (conflictindex != RelationGetRelid(indexRelation))
460 46 : continue;
461 :
462 : /*
463 : * This Assert will fail if BuildSpeculativeIndexInfo() is called
464 : * twice for the given index.
465 : */
466 : Assert(indexRelationInfo->ii_UniqueOps == NULL);
467 :
468 49 : BuildSpeculativeIndexInfo(indexRelation, indexRelationInfo);
469 : }
470 49 : }
471 :
472 : /*
473 : * If the tuple is recently dead and was deleted by a transaction with a newer
474 : * commit timestamp than previously recorded, update the associated transaction
475 : * ID, commit time, and origin. This helps ensure that conflict detection uses
476 : * the most recent and relevant deletion metadata.
477 : */
478 : static void
479 3 : update_most_recent_deletion_info(TupleTableSlot *scanslot,
480 : TransactionId oldestxmin,
481 : TransactionId *delete_xid,
482 : TimestampTz *delete_time,
483 : ReplOriginId *delete_origin)
484 : {
485 : BufferHeapTupleTableSlot *hslot;
486 : HeapTuple tuple;
487 : Buffer buf;
488 3 : bool recently_dead = false;
489 : TransactionId xmax;
490 : TimestampTz localts;
491 : ReplOriginId localorigin;
492 :
493 3 : hslot = (BufferHeapTupleTableSlot *) scanslot;
494 :
495 3 : tuple = ExecFetchSlotHeapTuple(scanslot, false, NULL);
496 3 : buf = hslot->buffer;
497 :
498 3 : LockBuffer(buf, BUFFER_LOCK_SHARE);
499 :
500 : /*
501 : * We do not consider HEAPTUPLE_DEAD status because it indicates either
502 : * tuples whose inserting transaction was aborted (meaning there is no
503 : * commit timestamp or origin), or tuples deleted by a transaction older
504 : * than oldestxmin, making it safe to ignore them during conflict
505 : * detection (See comments atop worker.c for details).
506 : */
507 3 : if (HeapTupleSatisfiesVacuum(tuple, oldestxmin, buf) == HEAPTUPLE_RECENTLY_DEAD)
508 3 : recently_dead = true;
509 :
510 3 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
511 :
512 3 : if (!recently_dead)
513 0 : return;
514 :
515 3 : xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
516 3 : if (!TransactionIdIsValid(xmax))
517 0 : return;
518 :
519 : /* Select the dead tuple with the most recent commit timestamp */
520 6 : if (TransactionIdGetCommitTsData(xmax, &localts, &localorigin) &&
521 3 : TimestampDifferenceExceeds(*delete_time, localts, 0))
522 : {
523 3 : *delete_xid = xmax;
524 3 : *delete_time = localts;
525 3 : *delete_origin = localorigin;
526 : }
527 : }
528 :
529 : /*
530 : * Searches the relation 'rel' for the most recently deleted tuple that matches
531 : * the values in 'searchslot' and is not yet removable by VACUUM. The function
532 : * returns the transaction ID, origin, and commit timestamp of the transaction
533 : * that deleted this tuple.
534 : *
535 : * 'oldestxmin' acts as a cutoff transaction ID. Tuples deleted by transactions
536 : * with IDs >= 'oldestxmin' are considered recently dead and are eligible for
537 : * conflict detection.
538 : *
539 : * Instead of stopping at the first match, we scan all matching dead tuples to
540 : * identify most recent deletion. This is crucial because only the latest
541 : * deletion is relevant for resolving conflicts.
542 : *
543 : * For example, consider a scenario on the subscriber where a row is deleted,
544 : * re-inserted, and then deleted again only on the subscriber:
545 : *
546 : * - (pk, 1) - deleted at 9:00,
547 : * - (pk, 1) - deleted at 9:02,
548 : *
549 : * Now, a remote update arrives: (pk, 1) -> (pk, 2), timestamped at 9:01.
550 : *
551 : * If we mistakenly return the older deletion (9:00), the system may wrongly
552 : * apply the remote update using a last-update-wins strategy. Instead, we must
553 : * recognize the more recent deletion at 9:02 and skip the update. See
554 : * comments atop worker.c for details. Note, as of now, conflict resolution
555 : * is not implemented. Consequently, the system may incorrectly report the
556 : * older tuple as the conflicted one, leading to misleading results.
557 : *
558 : * The commit timestamp of the deleting transaction is used to determine which
559 : * tuple was deleted most recently.
560 : */
561 : bool
562 2 : RelationFindDeletedTupleInfoSeq(Relation rel, TupleTableSlot *searchslot,
563 : TransactionId oldestxmin,
564 : TransactionId *delete_xid,
565 : ReplOriginId *delete_origin,
566 : TimestampTz *delete_time)
567 : {
568 : TupleTableSlot *scanslot;
569 : TableScanDesc scan;
570 : TypeCacheEntry **eq;
571 : Bitmapset *indexbitmap;
572 2 : TupleDesc desc PG_USED_FOR_ASSERTS_ONLY = RelationGetDescr(rel);
573 :
574 : Assert(equalTupleDescs(desc, searchslot->tts_tupleDescriptor));
575 :
576 2 : *delete_xid = InvalidTransactionId;
577 2 : *delete_origin = InvalidReplOriginId;
578 2 : *delete_time = 0;
579 :
580 : /*
581 : * If the relation has a replica identity key or a primary key that is
582 : * unusable for locating deleted tuples (see
583 : * IsIndexUsableForFindingDeletedTuple), a full table scan becomes
584 : * necessary. In such cases, comparing the entire tuple is not required,
585 : * since the remote tuple might not include all column values. Instead,
586 : * the indexed columns alone are sufficient to identify the target tuple
587 : * (see logicalrep_rel_mark_updatable).
588 : */
589 2 : indexbitmap = RelationGetIndexAttrBitmap(rel,
590 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
591 :
592 : /* fallback to PK if no replica identity */
593 2 : if (!indexbitmap)
594 2 : indexbitmap = RelationGetIndexAttrBitmap(rel,
595 : INDEX_ATTR_BITMAP_PRIMARY_KEY);
596 :
597 2 : eq = palloc0_array(TypeCacheEntry *, searchslot->tts_tupleDescriptor->natts);
598 :
599 : /*
600 : * Start a heap scan using SnapshotAny to identify dead tuples that are
601 : * not visible under a standard MVCC snapshot. Tuples from transactions
602 : * not yet committed or those just committed prior to the scan are
603 : * excluded in update_most_recent_deletion_info().
604 : */
605 2 : scan = table_beginscan(rel, SnapshotAny, 0, NULL);
606 2 : scanslot = table_slot_create(rel, NULL);
607 :
608 2 : table_rescan(scan, NULL);
609 :
610 : /* Try to find the tuple */
611 5 : while (table_scan_getnextslot(scan, ForwardScanDirection, scanslot))
612 : {
613 3 : if (!tuples_equal(scanslot, searchslot, eq, indexbitmap))
614 1 : continue;
615 :
616 2 : update_most_recent_deletion_info(scanslot, oldestxmin, delete_xid,
617 : delete_time, delete_origin);
618 : }
619 :
620 2 : table_endscan(scan);
621 2 : ExecDropSingleTupleTableSlot(scanslot);
622 :
623 2 : return *delete_time != 0;
624 : }
625 :
626 : /*
627 : * Similar to RelationFindDeletedTupleInfoSeq() but using index scan to locate
628 : * the deleted tuple.
629 : */
630 : bool
631 1 : RelationFindDeletedTupleInfoByIndex(Relation rel, Oid idxoid,
632 : TupleTableSlot *searchslot,
633 : TransactionId oldestxmin,
634 : TransactionId *delete_xid,
635 : ReplOriginId *delete_origin,
636 : TimestampTz *delete_time)
637 : {
638 : Relation idxrel;
639 : ScanKeyData skey[INDEX_MAX_KEYS];
640 : int skey_attoff;
641 : IndexScanDesc scan;
642 : TupleTableSlot *scanslot;
643 1 : TypeCacheEntry **eq = NULL;
644 : bool isIdxSafeToSkipDuplicates;
645 1 : TupleDesc desc PG_USED_FOR_ASSERTS_ONLY = RelationGetDescr(rel);
646 :
647 : Assert(equalTupleDescs(desc, searchslot->tts_tupleDescriptor));
648 : Assert(OidIsValid(idxoid));
649 :
650 1 : *delete_xid = InvalidTransactionId;
651 1 : *delete_time = 0;
652 1 : *delete_origin = InvalidReplOriginId;
653 :
654 1 : isIdxSafeToSkipDuplicates = (GetRelationIdentityOrPK(rel) == idxoid);
655 :
656 1 : scanslot = table_slot_create(rel, NULL);
657 :
658 1 : idxrel = index_open(idxoid, RowExclusiveLock);
659 :
660 : /* Build scan key. */
661 1 : skey_attoff = build_replindex_scan_key(skey, rel, idxrel, searchslot);
662 :
663 : /*
664 : * Start an index scan using SnapshotAny to identify dead tuples that are
665 : * not visible under a standard MVCC snapshot. Tuples from transactions
666 : * not yet committed or those just committed prior to the scan are
667 : * excluded in update_most_recent_deletion_info().
668 : */
669 1 : scan = index_beginscan(rel, idxrel, SnapshotAny, NULL, skey_attoff, 0);
670 :
671 1 : index_rescan(scan, skey, skey_attoff, NULL, 0);
672 :
673 : /* Try to find the tuple */
674 2 : while (index_getnext_slot(scan, ForwardScanDirection, scanslot))
675 : {
676 : /*
677 : * Avoid expensive equality check if the index is primary key or
678 : * replica identity index.
679 : */
680 1 : if (!isIdxSafeToSkipDuplicates)
681 : {
682 0 : if (eq == NULL)
683 0 : eq = palloc0_array(TypeCacheEntry *, scanslot->tts_tupleDescriptor->natts);
684 :
685 0 : if (!tuples_equal(scanslot, searchslot, eq, NULL))
686 0 : continue;
687 : }
688 :
689 1 : update_most_recent_deletion_info(scanslot, oldestxmin, delete_xid,
690 : delete_time, delete_origin);
691 : }
692 :
693 1 : index_endscan(scan);
694 :
695 1 : index_close(idxrel, NoLock);
696 :
697 1 : ExecDropSingleTupleTableSlot(scanslot);
698 :
699 1 : return *delete_time != 0;
700 : }
701 :
702 : /*
703 : * Find the tuple that violates the passed unique index (conflictindex).
704 : *
705 : * If the conflicting tuple is found return true, otherwise false.
706 : *
707 : * We lock the tuple to avoid getting it deleted before the caller can fetch
708 : * the required information. Note that if the tuple is deleted before a lock
709 : * is acquired, we will retry to find the conflicting tuple again.
710 : */
711 : static bool
712 49 : FindConflictTuple(ResultRelInfo *resultRelInfo, EState *estate,
713 : Oid conflictindex, TupleTableSlot *slot,
714 : TupleTableSlot **conflictslot)
715 : {
716 49 : Relation rel = resultRelInfo->ri_RelationDesc;
717 : ItemPointerData conflictTid;
718 : TM_FailureData tmfd;
719 : TM_Result res;
720 :
721 49 : *conflictslot = NULL;
722 :
723 : /*
724 : * Build additional information required to check constraints violations.
725 : * See check_exclusion_or_unique_constraint().
726 : */
727 49 : BuildConflictIndexInfo(resultRelInfo, conflictindex);
728 :
729 49 : retry:
730 48 : if (ExecCheckIndexConstraints(resultRelInfo, slot, estate,
731 49 : &conflictTid, &slot->tts_tid,
732 49 : list_make1_oid(conflictindex)))
733 : {
734 1 : if (*conflictslot)
735 0 : ExecDropSingleTupleTableSlot(*conflictslot);
736 :
737 1 : *conflictslot = NULL;
738 1 : return false;
739 : }
740 :
741 47 : *conflictslot = table_slot_create(rel, NULL);
742 :
743 47 : PushActiveSnapshot(GetLatestSnapshot());
744 :
745 47 : res = table_tuple_lock(rel, &conflictTid, GetActiveSnapshot(),
746 : *conflictslot,
747 : GetCurrentCommandId(false),
748 : LockTupleShare,
749 : LockWaitBlock,
750 : 0 /* don't follow updates */ ,
751 : &tmfd);
752 :
753 47 : PopActiveSnapshot();
754 :
755 47 : if (should_refetch_tuple(res, &tmfd))
756 0 : goto retry;
757 :
758 47 : return true;
759 : }
760 :
761 : /*
762 : * Check all the unique indexes in 'recheckIndexes' for conflict with the
763 : * tuple in 'remoteslot' and report if found.
764 : */
765 : static void
766 28 : CheckAndReportConflict(ResultRelInfo *resultRelInfo, EState *estate,
767 : ConflictType type, List *recheckIndexes,
768 : TupleTableSlot *searchslot, TupleTableSlot *remoteslot)
769 : {
770 28 : List *conflicttuples = NIL;
771 : TupleTableSlot *conflictslot;
772 :
773 : /* Check all the unique indexes for conflicts */
774 103 : foreach_oid(uniqueidx, resultRelInfo->ri_onConflictArbiterIndexes)
775 : {
776 97 : if (list_member_oid(recheckIndexes, uniqueidx) &&
777 49 : FindConflictTuple(resultRelInfo, estate, uniqueidx, remoteslot,
778 : &conflictslot))
779 : {
780 47 : ConflictTupleInfo *conflicttuple = palloc0_object(ConflictTupleInfo);
781 :
782 47 : conflicttuple->slot = conflictslot;
783 47 : conflicttuple->indexoid = uniqueidx;
784 :
785 47 : GetTupleTransactionInfo(conflictslot, &conflicttuple->xmin,
786 : &conflicttuple->origin, &conflicttuple->ts);
787 :
788 47 : conflicttuples = lappend(conflicttuples, conflicttuple);
789 : }
790 : }
791 :
792 : /* Report the conflict, if found */
793 27 : if (conflicttuples)
794 26 : ReportApplyConflict(estate, resultRelInfo, ERROR,
795 26 : list_length(conflicttuples) > 1 ? CT_MULTIPLE_UNIQUE_CONFLICTS : type,
796 : searchslot, remoteslot, conflicttuples);
797 1 : }
798 :
799 : /*
800 : * Insert tuple represented in the slot to the relation, update the indexes,
801 : * and execute any constraints and per-row triggers.
802 : *
803 : * Caller is responsible for opening the indexes.
804 : */
805 : void
806 76303 : ExecSimpleRelationInsert(ResultRelInfo *resultRelInfo,
807 : EState *estate, TupleTableSlot *slot)
808 : {
809 76303 : bool skip_tuple = false;
810 76303 : Relation rel = resultRelInfo->ri_RelationDesc;
811 :
812 : /* For now we support only tables. */
813 : Assert(rel->rd_rel->relkind == RELKIND_RELATION);
814 :
815 76303 : CheckCmdReplicaIdentity(rel, CMD_INSERT);
816 :
817 : /* BEFORE ROW INSERT Triggers */
818 76303 : if (resultRelInfo->ri_TrigDesc &&
819 20 : resultRelInfo->ri_TrigDesc->trig_insert_before_row)
820 : {
821 3 : if (!ExecBRInsertTriggers(estate, resultRelInfo, slot))
822 1 : skip_tuple = true; /* "do nothing" */
823 : }
824 :
825 76303 : if (!skip_tuple)
826 : {
827 76302 : List *recheckIndexes = NIL;
828 : List *conflictindexes;
829 76302 : bool conflict = false;
830 :
831 : /* Compute stored generated columns */
832 76302 : if (rel->rd_att->constr &&
833 45497 : rel->rd_att->constr->has_generated_stored)
834 4 : ExecComputeStoredGenerated(resultRelInfo, estate, slot,
835 : CMD_INSERT);
836 :
837 : /* Check the constraints of the tuple */
838 76302 : if (rel->rd_att->constr)
839 45497 : ExecConstraints(resultRelInfo, slot, estate);
840 76302 : if (rel->rd_rel->relispartition)
841 77 : ExecPartitionCheck(resultRelInfo, slot, estate, true);
842 :
843 : /* OK, store the tuple and create index entries for it */
844 76302 : simple_table_tuple_insert(resultRelInfo->ri_RelationDesc, slot);
845 :
846 76302 : conflictindexes = resultRelInfo->ri_onConflictArbiterIndexes;
847 :
848 76302 : if (resultRelInfo->ri_NumIndices > 0)
849 : {
850 : bits32 flags;
851 :
852 55958 : if (conflictindexes != NIL)
853 55954 : flags = EIIT_NO_DUPE_ERROR;
854 : else
855 4 : flags = 0;
856 55958 : recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
857 : estate, flags,
858 : slot, conflictindexes,
859 : &conflict);
860 : }
861 :
862 : /*
863 : * Checks the conflict indexes to fetch the conflicting local row and
864 : * reports the conflict. We perform this check here, instead of
865 : * performing an additional index scan before the actual insertion and
866 : * reporting the conflict if any conflicting rows are found. This is
867 : * to avoid the overhead of executing the extra scan for each INSERT
868 : * operation, even when no conflict arises, which could introduce
869 : * significant overhead to replication, particularly in cases where
870 : * conflicts are rare.
871 : *
872 : * XXX OTOH, this could lead to clean-up effort for dead tuples added
873 : * in heap and index in case of conflicts. But as conflicts shouldn't
874 : * be a frequent thing so we preferred to save the performance
875 : * overhead of extra scan before each insertion.
876 : */
877 76302 : if (conflict)
878 26 : CheckAndReportConflict(resultRelInfo, estate, CT_INSERT_EXISTS,
879 : recheckIndexes, NULL, slot);
880 :
881 : /* AFTER ROW INSERT Triggers */
882 76277 : ExecARInsertTriggers(estate, resultRelInfo, slot,
883 : recheckIndexes, NULL);
884 :
885 : /*
886 : * XXX we should in theory pass a TransitionCaptureState object to the
887 : * above to capture transition tuples, but after statement triggers
888 : * don't actually get fired by replication yet anyway
889 : */
890 :
891 76277 : list_free(recheckIndexes);
892 : }
893 76278 : }
894 :
895 : /*
896 : * Find the searchslot tuple and update it with data in the slot,
897 : * update the indexes, and execute any constraints and per-row triggers.
898 : *
899 : * Caller is responsible for opening the indexes.
900 : */
901 : void
902 31921 : ExecSimpleRelationUpdate(ResultRelInfo *resultRelInfo,
903 : EState *estate, EPQState *epqstate,
904 : TupleTableSlot *searchslot, TupleTableSlot *slot)
905 : {
906 31921 : bool skip_tuple = false;
907 31921 : Relation rel = resultRelInfo->ri_RelationDesc;
908 31921 : ItemPointer tid = &(searchslot->tts_tid);
909 :
910 : /*
911 : * We support only non-system tables, with
912 : * check_publication_add_relation() accountable.
913 : */
914 : Assert(rel->rd_rel->relkind == RELKIND_RELATION);
915 : Assert(!IsCatalogRelation(rel));
916 :
917 31921 : CheckCmdReplicaIdentity(rel, CMD_UPDATE);
918 :
919 : /* BEFORE ROW UPDATE Triggers */
920 31921 : if (resultRelInfo->ri_TrigDesc &&
921 10 : resultRelInfo->ri_TrigDesc->trig_update_before_row)
922 : {
923 3 : if (!ExecBRUpdateTriggers(estate, epqstate, resultRelInfo,
924 : tid, NULL, slot, NULL, NULL, false))
925 2 : skip_tuple = true; /* "do nothing" */
926 : }
927 :
928 31921 : if (!skip_tuple)
929 : {
930 31919 : List *recheckIndexes = NIL;
931 : TU_UpdateIndexes update_indexes;
932 : List *conflictindexes;
933 31919 : bool conflict = false;
934 :
935 : /* Compute stored generated columns */
936 31919 : if (rel->rd_att->constr &&
937 31875 : rel->rd_att->constr->has_generated_stored)
938 2 : ExecComputeStoredGenerated(resultRelInfo, estate, slot,
939 : CMD_UPDATE);
940 :
941 : /* Check the constraints of the tuple */
942 31919 : if (rel->rd_att->constr)
943 31875 : ExecConstraints(resultRelInfo, slot, estate);
944 31919 : if (rel->rd_rel->relispartition)
945 12 : ExecPartitionCheck(resultRelInfo, slot, estate, true);
946 :
947 31919 : simple_table_tuple_update(rel, tid, slot, estate->es_snapshot,
948 : &update_indexes);
949 :
950 31919 : conflictindexes = resultRelInfo->ri_onConflictArbiterIndexes;
951 :
952 31919 : if (resultRelInfo->ri_NumIndices > 0 && (update_indexes != TU_None))
953 : {
954 20154 : bits32 flags = EIIT_IS_UPDATE;
955 :
956 20154 : if (conflictindexes != NIL)
957 20145 : flags |= EIIT_NO_DUPE_ERROR;
958 20154 : if (update_indexes == TU_Summarizing)
959 0 : flags |= EIIT_ONLY_SUMMARIZING;
960 20154 : recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
961 : estate, flags,
962 : slot, conflictindexes,
963 : &conflict);
964 : }
965 :
966 : /*
967 : * Refer to the comments above the call to CheckAndReportConflict() in
968 : * ExecSimpleRelationInsert to understand why this check is done at
969 : * this point.
970 : */
971 31919 : if (conflict)
972 2 : CheckAndReportConflict(resultRelInfo, estate, CT_UPDATE_EXISTS,
973 : recheckIndexes, searchslot, slot);
974 :
975 : /* AFTER ROW UPDATE Triggers */
976 31917 : ExecARUpdateTriggers(estate, resultRelInfo,
977 : NULL, NULL,
978 : tid, NULL, slot,
979 : recheckIndexes, NULL, false);
980 :
981 31917 : list_free(recheckIndexes);
982 : }
983 31919 : }
984 :
985 : /*
986 : * Find the searchslot tuple and delete it, and execute any constraints
987 : * and per-row triggers.
988 : *
989 : * Caller is responsible for opening the indexes.
990 : */
991 : void
992 40313 : ExecSimpleRelationDelete(ResultRelInfo *resultRelInfo,
993 : EState *estate, EPQState *epqstate,
994 : TupleTableSlot *searchslot)
995 : {
996 40313 : bool skip_tuple = false;
997 40313 : Relation rel = resultRelInfo->ri_RelationDesc;
998 40313 : ItemPointer tid = &searchslot->tts_tid;
999 :
1000 40313 : CheckCmdReplicaIdentity(rel, CMD_DELETE);
1001 :
1002 : /* BEFORE ROW DELETE Triggers */
1003 40313 : if (resultRelInfo->ri_TrigDesc &&
1004 10 : resultRelInfo->ri_TrigDesc->trig_delete_before_row)
1005 : {
1006 0 : skip_tuple = !ExecBRDeleteTriggers(estate, epqstate, resultRelInfo,
1007 0 : tid, NULL, NULL, NULL, NULL, false);
1008 : }
1009 :
1010 40313 : if (!skip_tuple)
1011 : {
1012 : /* OK, delete the tuple */
1013 40313 : simple_table_tuple_delete(rel, tid, estate->es_snapshot);
1014 :
1015 : /* AFTER ROW DELETE Triggers */
1016 40313 : ExecARDeleteTriggers(estate, resultRelInfo,
1017 : tid, NULL, NULL, false);
1018 : }
1019 40313 : }
1020 :
1021 : /*
1022 : * Check if command can be executed with current replica identity.
1023 : */
1024 : void
1025 215471 : CheckCmdReplicaIdentity(Relation rel, CmdType cmd)
1026 : {
1027 : PublicationDesc pubdesc;
1028 :
1029 : /*
1030 : * Skip checking the replica identity for partitioned tables, because the
1031 : * operations are actually performed on the leaf partitions.
1032 : */
1033 215471 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
1034 203217 : return;
1035 :
1036 : /* We only need to do checks for UPDATE and DELETE. */
1037 213012 : if (cmd != CMD_UPDATE && cmd != CMD_DELETE)
1038 124678 : return;
1039 :
1040 : /*
1041 : * It is only safe to execute UPDATE/DELETE if the relation does not
1042 : * publish UPDATEs or DELETEs, or all the following conditions are
1043 : * satisfied:
1044 : *
1045 : * 1. All columns, referenced in the row filters from publications which
1046 : * the relation is in, are valid - i.e. when all referenced columns are
1047 : * part of REPLICA IDENTITY.
1048 : *
1049 : * 2. All columns, referenced in the column lists are valid - i.e. when
1050 : * all columns referenced in the REPLICA IDENTITY are covered by the
1051 : * column list.
1052 : *
1053 : * 3. All generated columns in REPLICA IDENTITY of the relation, are valid
1054 : * - i.e. when all these generated columns are published.
1055 : *
1056 : * XXX We could optimize it by first checking whether any of the
1057 : * publications have a row filter or column list for this relation, or if
1058 : * the relation contains a generated column. If none of these exist and
1059 : * the relation has replica identity then we can avoid building the
1060 : * descriptor but as this happens only one time it doesn't seem worth the
1061 : * additional complexity.
1062 : */
1063 88334 : RelationBuildPublicationDesc(rel, &pubdesc);
1064 88334 : if (cmd == CMD_UPDATE && !pubdesc.rf_valid_for_update)
1065 30 : ereport(ERROR,
1066 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1067 : errmsg("cannot update table \"%s\"",
1068 : RelationGetRelationName(rel)),
1069 : errdetail("Column used in the publication WHERE expression is not part of the replica identity.")));
1070 88304 : else if (cmd == CMD_UPDATE && !pubdesc.cols_valid_for_update)
1071 54 : ereport(ERROR,
1072 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1073 : errmsg("cannot update table \"%s\"",
1074 : RelationGetRelationName(rel)),
1075 : errdetail("Column list used by the publication does not cover the replica identity.")));
1076 88250 : else if (cmd == CMD_UPDATE && !pubdesc.gencols_valid_for_update)
1077 12 : ereport(ERROR,
1078 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1079 : errmsg("cannot update table \"%s\"",
1080 : RelationGetRelationName(rel)),
1081 : errdetail("Replica identity must not contain unpublished generated columns.")));
1082 88238 : else if (cmd == CMD_DELETE && !pubdesc.rf_valid_for_delete)
1083 0 : ereport(ERROR,
1084 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1085 : errmsg("cannot delete from table \"%s\"",
1086 : RelationGetRelationName(rel)),
1087 : errdetail("Column used in the publication WHERE expression is not part of the replica identity.")));
1088 88238 : else if (cmd == CMD_DELETE && !pubdesc.cols_valid_for_delete)
1089 0 : ereport(ERROR,
1090 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1091 : errmsg("cannot delete from table \"%s\"",
1092 : RelationGetRelationName(rel)),
1093 : errdetail("Column list used by the publication does not cover the replica identity.")));
1094 88238 : else if (cmd == CMD_DELETE && !pubdesc.gencols_valid_for_delete)
1095 0 : ereport(ERROR,
1096 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1097 : errmsg("cannot delete from table \"%s\"",
1098 : RelationGetRelationName(rel)),
1099 : errdetail("Replica identity must not contain unpublished generated columns.")));
1100 :
1101 : /* If relation has replica identity we are always good. */
1102 88238 : if (OidIsValid(RelationGetReplicaIndex(rel)))
1103 75850 : return;
1104 :
1105 : /* REPLICA IDENTITY FULL is also good for UPDATE/DELETE. */
1106 12388 : if (rel->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
1107 230 : return;
1108 :
1109 : /*
1110 : * This is UPDATE/DELETE and there is no replica identity.
1111 : *
1112 : * Check if the table publishes UPDATES or DELETES.
1113 : */
1114 12158 : if (cmd == CMD_UPDATE && pubdesc.pubactions.pubupdate)
1115 62 : ereport(ERROR,
1116 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1117 : errmsg("cannot update table \"%s\" because it does not have a replica identity and publishes updates",
1118 : RelationGetRelationName(rel)),
1119 : errhint("To enable updating the table, set REPLICA IDENTITY using ALTER TABLE.")));
1120 12096 : else if (cmd == CMD_DELETE && pubdesc.pubactions.pubdelete)
1121 8 : ereport(ERROR,
1122 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1123 : errmsg("cannot delete from table \"%s\" because it does not have a replica identity and publishes deletes",
1124 : RelationGetRelationName(rel)),
1125 : errhint("To enable deleting from the table, set REPLICA IDENTITY using ALTER TABLE.")));
1126 : }
1127 :
1128 :
1129 : /*
1130 : * Check if we support writing into specific relkind of local relation and check
1131 : * if it aligns with the relkind of the relation on the publisher.
1132 : *
1133 : * The nspname and relname are only needed for error reporting.
1134 : */
1135 : void
1136 964 : CheckSubscriptionRelkind(char localrelkind, char remoterelkind,
1137 : const char *nspname, const char *relname)
1138 : {
1139 964 : if (localrelkind != RELKIND_RELATION &&
1140 17 : localrelkind != RELKIND_PARTITIONED_TABLE &&
1141 : localrelkind != RELKIND_SEQUENCE)
1142 0 : ereport(ERROR,
1143 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1144 : errmsg("cannot use relation \"%s.%s\" as logical replication target",
1145 : nspname, relname),
1146 : errdetail_relkind_not_supported(localrelkind)));
1147 :
1148 : /*
1149 : * Allow RELKIND_RELATION and RELKIND_PARTITIONED_TABLE to be treated
1150 : * interchangeably, but ensure that sequences (RELKIND_SEQUENCE) match
1151 : * exactly on both publisher and subscriber.
1152 : */
1153 964 : if ((localrelkind == RELKIND_SEQUENCE && remoterelkind != RELKIND_SEQUENCE) ||
1154 947 : (localrelkind != RELKIND_SEQUENCE && remoterelkind == RELKIND_SEQUENCE))
1155 0 : ereport(ERROR,
1156 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1157 : /* translator: 3rd and 4th %s are "sequence" or "table" */
1158 : errmsg("relation \"%s.%s\" type mismatch: source \"%s\", target \"%s\"",
1159 : nspname, relname,
1160 : remoterelkind == RELKIND_SEQUENCE ? "sequence" : "table",
1161 : localrelkind == RELKIND_SEQUENCE ? "sequence" : "table"));
1162 964 : }
|
