Line data Source code
1 : /*-------------------------------------------------------------------------
2 : * relation.c
3 : * PostgreSQL logical replication relation mapping cache
4 : *
5 : * Copyright (c) 2016-2024, PostgreSQL Global Development Group
6 : *
7 : * IDENTIFICATION
8 : * src/backend/replication/logical/relation.c
9 : *
10 : * NOTES
11 : * Routines in this file mainly have to do with mapping the properties
12 : * of local replication target relations to the properties of their
13 : * remote counterpart.
14 : *
15 : *-------------------------------------------------------------------------
16 : */
17 :
18 : #include "postgres.h"
19 :
20 : #ifdef USE_ASSERT_CHECKING
21 : #include "access/amapi.h"
22 : #endif
23 : #include "access/genam.h"
24 : #include "access/table.h"
25 : #include "catalog/namespace.h"
26 : #include "catalog/pg_subscription_rel.h"
27 : #include "executor/executor.h"
28 : #include "nodes/makefuncs.h"
29 : #include "replication/logicalrelation.h"
30 : #include "replication/worker_internal.h"
31 : #include "utils/inval.h"
32 :
33 :
34 : static MemoryContext LogicalRepRelMapContext = NULL;
35 :
36 : static HTAB *LogicalRepRelMap = NULL;
37 :
38 : /*
39 : * Partition map (LogicalRepPartMap)
40 : *
41 : * When a partitioned table is used as replication target, replicated
42 : * operations are actually performed on its leaf partitions, which requires
43 : * the partitions to also be mapped to the remote relation. Parent's entry
44 : * (LogicalRepRelMapEntry) cannot be used as-is for all partitions, because
45 : * individual partitions may have different attribute numbers, which means
46 : * attribute mappings to remote relation's attributes must be maintained
47 : * separately for each partition.
48 : */
49 : static MemoryContext LogicalRepPartMapContext = NULL;
50 : static HTAB *LogicalRepPartMap = NULL;
51 : typedef struct LogicalRepPartMapEntry
52 : {
53 : Oid partoid; /* LogicalRepPartMap's key */
54 : LogicalRepRelMapEntry relmapentry;
55 : } LogicalRepPartMapEntry;
56 :
57 : static Oid FindLogicalRepLocalIndex(Relation localrel, LogicalRepRelation *remoterel,
58 : AttrMap *attrMap);
59 :
60 : /*
61 : * Relcache invalidation callback for our relation map cache.
62 : */
63 : static void
64 1262 : logicalrep_relmap_invalidate_cb(Datum arg, Oid reloid)
65 : {
66 : LogicalRepRelMapEntry *entry;
67 :
68 : /* Just to be sure. */
69 1262 : if (LogicalRepRelMap == NULL)
70 0 : return;
71 :
72 1262 : if (reloid != InvalidOid)
73 : {
74 : HASH_SEQ_STATUS status;
75 :
76 1262 : hash_seq_init(&status, LogicalRepRelMap);
77 :
78 : /* TODO, use inverse lookup hashtable? */
79 5532 : while ((entry = (LogicalRepRelMapEntry *) hash_seq_search(&status)) != NULL)
80 : {
81 4504 : if (entry->localreloid == reloid)
82 : {
83 234 : entry->localrelvalid = false;
84 234 : hash_seq_term(&status);
85 234 : break;
86 : }
87 : }
88 : }
89 : else
90 : {
91 : /* invalidate all cache entries */
92 : HASH_SEQ_STATUS status;
93 :
94 0 : hash_seq_init(&status, LogicalRepRelMap);
95 :
96 0 : while ((entry = (LogicalRepRelMapEntry *) hash_seq_search(&status)) != NULL)
97 0 : entry->localrelvalid = false;
98 : }
99 : }
100 :
101 : /*
102 : * Initialize the relation map cache.
103 : */
104 : static void
105 614 : logicalrep_relmap_init(void)
106 : {
107 : HASHCTL ctl;
108 :
109 614 : if (!LogicalRepRelMapContext)
110 614 : LogicalRepRelMapContext =
111 614 : AllocSetContextCreate(CacheMemoryContext,
112 : "LogicalRepRelMapContext",
113 : ALLOCSET_DEFAULT_SIZES);
114 :
115 : /* Initialize the relation hash table. */
116 614 : ctl.keysize = sizeof(LogicalRepRelId);
117 614 : ctl.entrysize = sizeof(LogicalRepRelMapEntry);
118 614 : ctl.hcxt = LogicalRepRelMapContext;
119 :
120 614 : LogicalRepRelMap = hash_create("logicalrep relation map cache", 128, &ctl,
121 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
122 :
123 : /* Watch for invalidation events. */
124 614 : CacheRegisterRelcacheCallback(logicalrep_relmap_invalidate_cb,
125 : (Datum) 0);
126 614 : }
127 :
128 : /*
129 : * Free the entry of a relation map cache.
130 : */
131 : static void
132 280 : logicalrep_relmap_free_entry(LogicalRepRelMapEntry *entry)
133 : {
134 : LogicalRepRelation *remoterel;
135 :
136 280 : remoterel = &entry->remoterel;
137 :
138 280 : pfree(remoterel->nspname);
139 280 : pfree(remoterel->relname);
140 :
141 280 : if (remoterel->natts > 0)
142 : {
143 : int i;
144 :
145 842 : for (i = 0; i < remoterel->natts; i++)
146 562 : pfree(remoterel->attnames[i]);
147 :
148 280 : pfree(remoterel->attnames);
149 280 : pfree(remoterel->atttyps);
150 : }
151 280 : bms_free(remoterel->attkeys);
152 :
153 280 : if (entry->attrmap)
154 236 : free_attrmap(entry->attrmap);
155 280 : }
156 :
157 : /*
158 : * Add new entry or update existing entry in the relation map cache.
159 : *
160 : * Called when new relation mapping is sent by the publisher to update
161 : * our expected view of incoming data from said publisher.
162 : */
163 : void
164 1090 : logicalrep_relmap_update(LogicalRepRelation *remoterel)
165 : {
166 : MemoryContext oldctx;
167 : LogicalRepRelMapEntry *entry;
168 : bool found;
169 : int i;
170 :
171 1090 : if (LogicalRepRelMap == NULL)
172 614 : logicalrep_relmap_init();
173 :
174 : /*
175 : * HASH_ENTER returns the existing entry if present or creates a new one.
176 : */
177 1090 : entry = hash_search(LogicalRepRelMap, &remoterel->remoteid,
178 : HASH_ENTER, &found);
179 :
180 1090 : if (found)
181 264 : logicalrep_relmap_free_entry(entry);
182 :
183 1090 : memset(entry, 0, sizeof(LogicalRepRelMapEntry));
184 :
185 : /* Make cached copy of the data */
186 1090 : oldctx = MemoryContextSwitchTo(LogicalRepRelMapContext);
187 1090 : entry->remoterel.remoteid = remoterel->remoteid;
188 1090 : entry->remoterel.nspname = pstrdup(remoterel->nspname);
189 1090 : entry->remoterel.relname = pstrdup(remoterel->relname);
190 1090 : entry->remoterel.natts = remoterel->natts;
191 1090 : entry->remoterel.attnames = palloc(remoterel->natts * sizeof(char *));
192 1090 : entry->remoterel.atttyps = palloc(remoterel->natts * sizeof(Oid));
193 2976 : for (i = 0; i < remoterel->natts; i++)
194 : {
195 1886 : entry->remoterel.attnames[i] = pstrdup(remoterel->attnames[i]);
196 1886 : entry->remoterel.atttyps[i] = remoterel->atttyps[i];
197 : }
198 1090 : entry->remoterel.replident = remoterel->replident;
199 1090 : entry->remoterel.attkeys = bms_copy(remoterel->attkeys);
200 1090 : MemoryContextSwitchTo(oldctx);
201 1090 : }
202 :
203 : /*
204 : * Find attribute index in TupleDesc struct by attribute name.
205 : *
206 : * Returns -1 if not found.
207 : */
208 : static int
209 2192 : logicalrep_rel_att_by_name(LogicalRepRelation *remoterel, const char *attname)
210 : {
211 : int i;
212 :
213 4164 : for (i = 0; i < remoterel->natts; i++)
214 : {
215 3640 : if (strcmp(remoterel->attnames[i], attname) == 0)
216 1668 : return i;
217 : }
218 :
219 524 : return -1;
220 : }
221 :
222 : /*
223 : * Report error with names of the missing local relation column(s), if any.
224 : */
225 : static void
226 968 : logicalrep_report_missing_attrs(LogicalRepRelation *remoterel,
227 : Bitmapset *missingatts)
228 : {
229 968 : if (!bms_is_empty(missingatts))
230 : {
231 : StringInfoData missingattsbuf;
232 0 : int missingattcnt = 0;
233 : int i;
234 :
235 0 : initStringInfo(&missingattsbuf);
236 :
237 0 : i = -1;
238 0 : while ((i = bms_next_member(missingatts, i)) >= 0)
239 : {
240 0 : missingattcnt++;
241 0 : if (missingattcnt == 1)
242 0 : appendStringInfo(&missingattsbuf, _("\"%s\""),
243 0 : remoterel->attnames[i]);
244 : else
245 0 : appendStringInfo(&missingattsbuf, _(", \"%s\""),
246 0 : remoterel->attnames[i]);
247 : }
248 :
249 0 : ereport(ERROR,
250 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
251 : errmsg_plural("logical replication target relation \"%s.%s\" is missing replicated column: %s",
252 : "logical replication target relation \"%s.%s\" is missing replicated columns: %s",
253 : missingattcnt,
254 : remoterel->nspname,
255 : remoterel->relname,
256 : missingattsbuf.data)));
257 : }
258 968 : }
259 :
260 : /*
261 : * Check if replica identity matches and mark the updatable flag.
262 : *
263 : * We allow for stricter replica identity (fewer columns) on subscriber as
264 : * that will not stop us from finding unique tuple. IE, if publisher has
265 : * identity (id,timestamp) and subscriber just (id) this will not be a
266 : * problem, but in the opposite scenario it will.
267 : *
268 : * We just mark the relation entry as not updatable here if the local
269 : * replica identity is found to be insufficient for applying
270 : * updates/deletes (inserts don't care!) and leave it to
271 : * check_relation_updatable() to throw the actual error if needed.
272 : */
273 : static void
274 996 : logicalrep_rel_mark_updatable(LogicalRepRelMapEntry *entry)
275 : {
276 : Bitmapset *idkey;
277 996 : LogicalRepRelation *remoterel = &entry->remoterel;
278 : int i;
279 :
280 996 : entry->updatable = true;
281 :
282 996 : idkey = RelationGetIndexAttrBitmap(entry->localrel,
283 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
284 : /* fallback to PK if no replica identity */
285 996 : if (idkey == NULL)
286 : {
287 336 : idkey = RelationGetIndexAttrBitmap(entry->localrel,
288 : INDEX_ATTR_BITMAP_PRIMARY_KEY);
289 :
290 : /*
291 : * If no replica identity index and no PK, the published table must
292 : * have replica identity FULL.
293 : */
294 336 : if (idkey == NULL && remoterel->replident != REPLICA_IDENTITY_FULL)
295 210 : entry->updatable = false;
296 : }
297 :
298 996 : i = -1;
299 1624 : while ((i = bms_next_member(idkey, i)) >= 0)
300 : {
301 670 : int attnum = i + FirstLowInvalidHeapAttributeNumber;
302 :
303 670 : if (!AttrNumberIsForUserDefinedAttr(attnum))
304 0 : ereport(ERROR,
305 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
306 : errmsg("logical replication target relation \"%s.%s\" uses "
307 : "system columns in REPLICA IDENTITY index",
308 : remoterel->nspname, remoterel->relname)));
309 :
310 670 : attnum = AttrNumberGetAttrOffset(attnum);
311 :
312 670 : if (entry->attrmap->attnums[attnum] < 0 ||
313 668 : !bms_is_member(entry->attrmap->attnums[attnum], remoterel->attkeys))
314 : {
315 42 : entry->updatable = false;
316 42 : break;
317 : }
318 : }
319 996 : }
320 :
321 : /*
322 : * Open the local relation associated with the remote one.
323 : *
324 : * Rebuilds the Relcache mapping if it was invalidated by local DDL.
325 : */
326 : LogicalRepRelMapEntry *
327 298064 : logicalrep_rel_open(LogicalRepRelId remoteid, LOCKMODE lockmode)
328 : {
329 : LogicalRepRelMapEntry *entry;
330 : bool found;
331 : LogicalRepRelation *remoterel;
332 :
333 298064 : if (LogicalRepRelMap == NULL)
334 0 : logicalrep_relmap_init();
335 :
336 : /* Search for existing entry. */
337 298064 : entry = hash_search(LogicalRepRelMap, &remoteid,
338 : HASH_FIND, &found);
339 :
340 298064 : if (!found)
341 0 : elog(ERROR, "no relation map entry for remote relation ID %u",
342 : remoteid);
343 :
344 298064 : remoterel = &entry->remoterel;
345 :
346 : /* Ensure we don't leak a relcache refcount. */
347 298064 : if (entry->localrel)
348 0 : elog(ERROR, "remote relation ID %u is already open", remoteid);
349 :
350 : /*
351 : * When opening and locking a relation, pending invalidation messages are
352 : * processed which can invalidate the relation. Hence, if the entry is
353 : * currently considered valid, try to open the local relation by OID and
354 : * see if invalidation ensues.
355 : */
356 298064 : if (entry->localrelvalid)
357 : {
358 297082 : entry->localrel = try_table_open(entry->localreloid, lockmode);
359 297082 : if (!entry->localrel)
360 : {
361 : /* Table was renamed or dropped. */
362 0 : entry->localrelvalid = false;
363 : }
364 297082 : else if (!entry->localrelvalid)
365 : {
366 : /* Note we release the no-longer-useful lock here. */
367 0 : table_close(entry->localrel, lockmode);
368 0 : entry->localrel = NULL;
369 : }
370 : }
371 :
372 : /*
373 : * If the entry has been marked invalid since we last had lock on it,
374 : * re-open the local relation by name and rebuild all derived data.
375 : */
376 298064 : if (!entry->localrelvalid)
377 : {
378 : Oid relid;
379 : TupleDesc desc;
380 : MemoryContext oldctx;
381 : int i;
382 : Bitmapset *missingatts;
383 :
384 : /* Release the no-longer-useful attrmap, if any. */
385 982 : if (entry->attrmap)
386 : {
387 24 : free_attrmap(entry->attrmap);
388 24 : entry->attrmap = NULL;
389 : }
390 :
391 : /* Try to find and lock the relation by name. */
392 982 : relid = RangeVarGetRelid(makeRangeVar(remoterel->nspname,
393 : remoterel->relname, -1),
394 : lockmode, true);
395 982 : if (!OidIsValid(relid))
396 14 : ereport(ERROR,
397 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
398 : errmsg("logical replication target relation \"%s.%s\" does not exist",
399 : remoterel->nspname, remoterel->relname)));
400 968 : entry->localrel = table_open(relid, NoLock);
401 968 : entry->localreloid = relid;
402 :
403 : /* Check for supported relkind. */
404 968 : CheckSubscriptionRelkind(entry->localrel->rd_rel->relkind,
405 968 : remoterel->nspname, remoterel->relname);
406 :
407 : /*
408 : * Build the mapping of local attribute numbers to remote attribute
409 : * numbers and validate that we don't miss any replicated columns as
410 : * that would result in potentially unwanted data loss.
411 : */
412 968 : desc = RelationGetDescr(entry->localrel);
413 968 : oldctx = MemoryContextSwitchTo(LogicalRepRelMapContext);
414 968 : entry->attrmap = make_attrmap(desc->natts);
415 968 : MemoryContextSwitchTo(oldctx);
416 :
417 : /* check and report missing attrs, if any */
418 968 : missingatts = bms_add_range(NULL, 0, remoterel->natts - 1);
419 3174 : for (i = 0; i < desc->natts; i++)
420 : {
421 : int attnum;
422 2206 : Form_pg_attribute attr = TupleDescAttr(desc, i);
423 :
424 2206 : if (attr->attisdropped || attr->attgenerated)
425 : {
426 14 : entry->attrmap->attnums[i] = -1;
427 14 : continue;
428 : }
429 :
430 2192 : attnum = logicalrep_rel_att_by_name(remoterel,
431 2192 : NameStr(attr->attname));
432 :
433 2192 : entry->attrmap->attnums[i] = attnum;
434 2192 : if (attnum >= 0)
435 1668 : missingatts = bms_del_member(missingatts, attnum);
436 : }
437 :
438 968 : logicalrep_report_missing_attrs(remoterel, missingatts);
439 :
440 : /* be tidy */
441 968 : bms_free(missingatts);
442 :
443 : /*
444 : * Set if the table's replica identity is enough to apply
445 : * update/delete.
446 : */
447 968 : logicalrep_rel_mark_updatable(entry);
448 :
449 : /*
450 : * Finding a usable index is an infrequent task. It occurs when an
451 : * operation is first performed on the relation, or after invalidation
452 : * of the relation cache entry (such as ANALYZE or CREATE/DROP index
453 : * on the relation).
454 : */
455 968 : entry->localindexoid = FindLogicalRepLocalIndex(entry->localrel, remoterel,
456 : entry->attrmap);
457 :
458 968 : entry->localrelvalid = true;
459 : }
460 :
461 298050 : if (entry->state != SUBREL_STATE_READY)
462 1026 : entry->state = GetSubscriptionRelState(MySubscription->oid,
463 : entry->localreloid,
464 : &entry->statelsn);
465 :
466 298050 : return entry;
467 : }
468 :
469 : /*
470 : * Close the previously opened logical relation.
471 : */
472 : void
473 297992 : logicalrep_rel_close(LogicalRepRelMapEntry *rel, LOCKMODE lockmode)
474 : {
475 297992 : table_close(rel->localrel, lockmode);
476 297992 : rel->localrel = NULL;
477 297992 : }
478 :
479 : /*
480 : * Partition cache: look up partition LogicalRepRelMapEntry's
481 : *
482 : * Unlike relation map cache, this is keyed by partition OID, not remote
483 : * relation OID, because we only have to use this cache in the case where
484 : * partitions are not directly mapped to any remote relation, such as when
485 : * replication is occurring with one of their ancestors as target.
486 : */
487 :
488 : /*
489 : * Relcache invalidation callback
490 : */
491 : static void
492 482 : logicalrep_partmap_invalidate_cb(Datum arg, Oid reloid)
493 : {
494 : LogicalRepPartMapEntry *entry;
495 :
496 : /* Just to be sure. */
497 482 : if (LogicalRepPartMap == NULL)
498 0 : return;
499 :
500 482 : if (reloid != InvalidOid)
501 : {
502 : HASH_SEQ_STATUS status;
503 :
504 482 : hash_seq_init(&status, LogicalRepPartMap);
505 :
506 : /* TODO, use inverse lookup hashtable? */
507 1350 : while ((entry = (LogicalRepPartMapEntry *) hash_seq_search(&status)) != NULL)
508 : {
509 880 : if (entry->relmapentry.localreloid == reloid)
510 : {
511 12 : entry->relmapentry.localrelvalid = false;
512 12 : hash_seq_term(&status);
513 12 : break;
514 : }
515 : }
516 : }
517 : else
518 : {
519 : /* invalidate all cache entries */
520 : HASH_SEQ_STATUS status;
521 :
522 0 : hash_seq_init(&status, LogicalRepPartMap);
523 :
524 0 : while ((entry = (LogicalRepPartMapEntry *) hash_seq_search(&status)) != NULL)
525 0 : entry->relmapentry.localrelvalid = false;
526 : }
527 : }
528 :
529 : /*
530 : * Reset the entries in the partition map that refer to remoterel.
531 : *
532 : * Called when new relation mapping is sent by the publisher to update our
533 : * expected view of incoming data from said publisher.
534 : *
535 : * Note that we don't update the remoterel information in the entry here,
536 : * we will update the information in logicalrep_partition_open to avoid
537 : * unnecessary work.
538 : */
539 : void
540 750 : logicalrep_partmap_reset_relmap(LogicalRepRelation *remoterel)
541 : {
542 : HASH_SEQ_STATUS status;
543 : LogicalRepPartMapEntry *part_entry;
544 : LogicalRepRelMapEntry *entry;
545 :
546 750 : if (LogicalRepPartMap == NULL)
547 682 : return;
548 :
549 68 : hash_seq_init(&status, LogicalRepPartMap);
550 174 : while ((part_entry = (LogicalRepPartMapEntry *) hash_seq_search(&status)) != NULL)
551 : {
552 106 : entry = &part_entry->relmapentry;
553 :
554 106 : if (entry->remoterel.remoteid != remoterel->remoteid)
555 90 : continue;
556 :
557 16 : logicalrep_relmap_free_entry(entry);
558 :
559 16 : memset(entry, 0, sizeof(LogicalRepRelMapEntry));
560 : }
561 : }
562 :
563 : /*
564 : * Initialize the partition map cache.
565 : */
566 : static void
567 10 : logicalrep_partmap_init(void)
568 : {
569 : HASHCTL ctl;
570 :
571 10 : if (!LogicalRepPartMapContext)
572 10 : LogicalRepPartMapContext =
573 10 : AllocSetContextCreate(CacheMemoryContext,
574 : "LogicalRepPartMapContext",
575 : ALLOCSET_DEFAULT_SIZES);
576 :
577 : /* Initialize the relation hash table. */
578 10 : ctl.keysize = sizeof(Oid); /* partition OID */
579 10 : ctl.entrysize = sizeof(LogicalRepPartMapEntry);
580 10 : ctl.hcxt = LogicalRepPartMapContext;
581 :
582 10 : LogicalRepPartMap = hash_create("logicalrep partition map cache", 64, &ctl,
583 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
584 :
585 : /* Watch for invalidation events. */
586 10 : CacheRegisterRelcacheCallback(logicalrep_partmap_invalidate_cb,
587 : (Datum) 0);
588 10 : }
589 :
590 : /*
591 : * logicalrep_partition_open
592 : *
593 : * Returned entry reuses most of the values of the root table's entry, save
594 : * the attribute map, which can be different for the partition. However,
595 : * we must physically copy all the data, in case the root table's entry
596 : * gets freed/rebuilt.
597 : *
598 : * Note there's no logicalrep_partition_close, because the caller closes the
599 : * component relation.
600 : */
601 : LogicalRepRelMapEntry *
602 58 : logicalrep_partition_open(LogicalRepRelMapEntry *root,
603 : Relation partrel, AttrMap *map)
604 : {
605 : LogicalRepRelMapEntry *entry;
606 : LogicalRepPartMapEntry *part_entry;
607 58 : LogicalRepRelation *remoterel = &root->remoterel;
608 58 : Oid partOid = RelationGetRelid(partrel);
609 58 : AttrMap *attrmap = root->attrmap;
610 : bool found;
611 : MemoryContext oldctx;
612 :
613 58 : if (LogicalRepPartMap == NULL)
614 10 : logicalrep_partmap_init();
615 :
616 : /* Search for existing entry. */
617 58 : part_entry = (LogicalRepPartMapEntry *) hash_search(LogicalRepPartMap,
618 : &partOid,
619 : HASH_ENTER, &found);
620 :
621 58 : entry = &part_entry->relmapentry;
622 :
623 : /*
624 : * We must always overwrite entry->localrel with the latest partition
625 : * Relation pointer, because the Relation pointed to by the old value may
626 : * have been cleared after the caller would have closed the partition
627 : * relation after the last use of this entry. Note that localrelvalid is
628 : * only updated by the relcache invalidation callback, so it may still be
629 : * true irrespective of whether the Relation pointed to by localrel has
630 : * been cleared or not.
631 : */
632 58 : if (found && entry->localrelvalid)
633 : {
634 30 : entry->localrel = partrel;
635 30 : return entry;
636 : }
637 :
638 : /* Switch to longer-lived context. */
639 28 : oldctx = MemoryContextSwitchTo(LogicalRepPartMapContext);
640 :
641 28 : if (!found)
642 : {
643 16 : memset(part_entry, 0, sizeof(LogicalRepPartMapEntry));
644 16 : part_entry->partoid = partOid;
645 : }
646 :
647 : /* Release the no-longer-useful attrmap, if any. */
648 28 : if (entry->attrmap)
649 : {
650 2 : free_attrmap(entry->attrmap);
651 2 : entry->attrmap = NULL;
652 : }
653 :
654 28 : if (!entry->remoterel.remoteid)
655 : {
656 : int i;
657 :
658 : /* Remote relation is copied as-is from the root entry. */
659 26 : entry->remoterel.remoteid = remoterel->remoteid;
660 26 : entry->remoterel.nspname = pstrdup(remoterel->nspname);
661 26 : entry->remoterel.relname = pstrdup(remoterel->relname);
662 26 : entry->remoterel.natts = remoterel->natts;
663 26 : entry->remoterel.attnames = palloc(remoterel->natts * sizeof(char *));
664 26 : entry->remoterel.atttyps = palloc(remoterel->natts * sizeof(Oid));
665 80 : for (i = 0; i < remoterel->natts; i++)
666 : {
667 54 : entry->remoterel.attnames[i] = pstrdup(remoterel->attnames[i]);
668 54 : entry->remoterel.atttyps[i] = remoterel->atttyps[i];
669 : }
670 26 : entry->remoterel.replident = remoterel->replident;
671 26 : entry->remoterel.attkeys = bms_copy(remoterel->attkeys);
672 : }
673 :
674 28 : entry->localrel = partrel;
675 28 : entry->localreloid = partOid;
676 :
677 : /*
678 : * If the partition's attributes don't match the root relation's, we'll
679 : * need to make a new attrmap which maps partition attribute numbers to
680 : * remoterel's, instead of the original which maps root relation's
681 : * attribute numbers to remoterel's.
682 : *
683 : * Note that 'map' which comes from the tuple routing data structure
684 : * contains 1-based attribute numbers (of the parent relation). However,
685 : * the map in 'entry', a logical replication data structure, contains
686 : * 0-based attribute numbers (of the remote relation).
687 : */
688 28 : if (map)
689 : {
690 : AttrNumber attno;
691 :
692 14 : entry->attrmap = make_attrmap(map->maplen);
693 60 : for (attno = 0; attno < entry->attrmap->maplen; attno++)
694 : {
695 46 : AttrNumber root_attno = map->attnums[attno];
696 :
697 : /* 0 means it's a dropped attribute. See comments atop AttrMap. */
698 46 : if (root_attno == 0)
699 4 : entry->attrmap->attnums[attno] = -1;
700 : else
701 42 : entry->attrmap->attnums[attno] = attrmap->attnums[root_attno - 1];
702 : }
703 : }
704 : else
705 : {
706 : /* Lacking copy_attmap, do this the hard way. */
707 14 : entry->attrmap = make_attrmap(attrmap->maplen);
708 14 : memcpy(entry->attrmap->attnums, attrmap->attnums,
709 14 : attrmap->maplen * sizeof(AttrNumber));
710 : }
711 :
712 : /* Set if the table's replica identity is enough to apply update/delete. */
713 28 : logicalrep_rel_mark_updatable(entry);
714 :
715 : /* state and statelsn are left set to 0. */
716 28 : MemoryContextSwitchTo(oldctx);
717 :
718 : /*
719 : * Finding a usable index is an infrequent task. It occurs when an
720 : * operation is first performed on the relation, or after invalidation of
721 : * the relation cache entry (such as ANALYZE or CREATE/DROP index on the
722 : * relation).
723 : *
724 : * We also prefer to run this code on the oldctx so that we do not leak
725 : * anything in the LogicalRepPartMapContext (hence CacheMemoryContext).
726 : */
727 28 : entry->localindexoid = FindLogicalRepLocalIndex(partrel, remoterel,
728 : entry->attrmap);
729 :
730 28 : entry->localrelvalid = true;
731 :
732 28 : return entry;
733 : }
734 :
735 : /*
736 : * Returns the oid of an index that can be used by the apply worker to scan
737 : * the relation.
738 : *
739 : * We expect to call this function when REPLICA IDENTITY FULL is defined for
740 : * the remote relation.
741 : *
742 : * If no suitable index is found, returns InvalidOid.
743 : */
744 : static Oid
745 112 : FindUsableIndexForReplicaIdentityFull(Relation localrel, AttrMap *attrmap)
746 : {
747 112 : List *idxlist = RelationGetIndexList(localrel);
748 :
749 204 : foreach_oid(idxoid, idxlist)
750 : {
751 : bool isUsableIdx;
752 : Relation idxRel;
753 : IndexInfo *idxInfo;
754 :
755 36 : idxRel = index_open(idxoid, AccessShareLock);
756 36 : idxInfo = BuildIndexInfo(idxRel);
757 36 : isUsableIdx = IsIndexUsableForReplicaIdentityFull(idxInfo, attrmap);
758 36 : index_close(idxRel, AccessShareLock);
759 :
760 : /* Return the first eligible index found */
761 36 : if (isUsableIdx)
762 28 : return idxoid;
763 : }
764 :
765 84 : return InvalidOid;
766 : }
767 :
768 : /*
769 : * Returns true if the index is usable for replica identity full.
770 : *
771 : * The index must be btree or hash, non-partial, and the leftmost field must be
772 : * a column (not an expression) that references the remote relation column. These
773 : * limitations help to keep the index scan similar to PK/RI index scans.
774 : *
775 : * attrmap is a map of local attributes to remote ones. We can consult this
776 : * map to check whether the local index attribute has a corresponding remote
777 : * attribute.
778 : *
779 : * Note that the limitations of index scans for replica identity full only
780 : * adheres to a subset of the limitations of PK/RI. For example, we support
781 : * columns that are marked as [NULL] or we are not interested in the [NOT
782 : * DEFERRABLE] aspect of constraints here. It works for us because we always
783 : * compare the tuples for non-PK/RI index scans. See
784 : * RelationFindReplTupleByIndex().
785 : *
786 : * The reasons why only Btree and Hash indexes can be considered as usable are:
787 : *
788 : * 1) Other index access methods don't have a fixed strategy for equality
789 : * operation. Refer get_equal_strategy_number_for_am().
790 : *
791 : * 2) For indexes other than PK and REPLICA IDENTITY, we need to match the
792 : * local and remote tuples. The equality routine tuples_equal() cannot accept
793 : * a datatype (e.g. point or box) that does not have a default operator class
794 : * for Btree or Hash.
795 : *
796 : * XXX: Note that BRIN and GIN indexes do not implement "amgettuple" which
797 : * will be used later to fetch the tuples. See RelationFindReplTupleByIndex().
798 : *
799 : * XXX: To support partial indexes, the required changes are likely to be larger.
800 : * If none of the tuples satisfy the expression for the index scan, we fall-back
801 : * to sequential execution, which might not be a good idea in some cases.
802 : */
803 : bool
804 36 : IsIndexUsableForReplicaIdentityFull(IndexInfo *indexInfo, AttrMap *attrmap)
805 : {
806 : AttrNumber keycol;
807 :
808 : /* Ensure that the index access method has a valid equal strategy */
809 36 : if (get_equal_strategy_number_for_am(indexInfo->ii_Am) == InvalidStrategy)
810 0 : return false;
811 :
812 : /* The index must not be a partial index */
813 36 : if (indexInfo->ii_Predicate != NIL)
814 4 : return false;
815 :
816 : Assert(indexInfo->ii_NumIndexAttrs >= 1);
817 :
818 : /* The leftmost index field must not be an expression */
819 32 : keycol = indexInfo->ii_IndexAttrNumbers[0];
820 32 : if (!AttributeNumberIsValid(keycol))
821 4 : return false;
822 :
823 : /*
824 : * And the leftmost index field must reference the remote relation column.
825 : * This is because if it doesn't, the sequential scan is favorable over
826 : * index scan in most cases.
827 : */
828 28 : if (attrmap->maplen <= AttrNumberGetAttrOffset(keycol) ||
829 28 : attrmap->attnums[AttrNumberGetAttrOffset(keycol)] < 0)
830 0 : return false;
831 :
832 : #ifdef USE_ASSERT_CHECKING
833 : {
834 : IndexAmRoutine *amroutine;
835 :
836 : /* The given index access method must implement amgettuple. */
837 : amroutine = GetIndexAmRoutineByAmId(indexInfo->ii_Am, false);
838 : Assert(amroutine->amgettuple != NULL);
839 : }
840 : #endif
841 :
842 28 : return true;
843 : }
844 :
845 : /*
846 : * Return the OID of the replica identity index if one is defined;
847 : * the OID of the PK if one exists and is not deferrable;
848 : * otherwise, InvalidOid.
849 : */
850 : Oid
851 145054 : GetRelationIdentityOrPK(Relation rel)
852 : {
853 : Oid idxoid;
854 :
855 145054 : idxoid = RelationGetReplicaIndex(rel);
856 :
857 145054 : if (!OidIsValid(idxoid))
858 338 : idxoid = RelationGetPrimaryKeyIndex(rel);
859 :
860 145054 : return idxoid;
861 : }
862 :
863 : /*
864 : * Returns the index oid if we can use an index for subscriber. Otherwise,
865 : * returns InvalidOid.
866 : */
867 : static Oid
868 996 : FindLogicalRepLocalIndex(Relation localrel, LogicalRepRelation *remoterel,
869 : AttrMap *attrMap)
870 : {
871 : Oid idxoid;
872 :
873 : /*
874 : * We never need index oid for partitioned tables, always rely on leaf
875 : * partition's index.
876 : */
877 996 : if (localrel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
878 96 : return InvalidOid;
879 :
880 : /*
881 : * Simple case, we already have a primary key or a replica identity index.
882 : */
883 900 : idxoid = GetRelationIdentityOrPK(localrel);
884 900 : if (OidIsValid(idxoid))
885 604 : return idxoid;
886 :
887 296 : if (remoterel->replident == REPLICA_IDENTITY_FULL)
888 : {
889 : /*
890 : * We are looking for one more opportunity for using an index. If
891 : * there are any indexes defined on the local relation, try to pick a
892 : * suitable index.
893 : *
894 : * The index selection safely assumes that all the columns are going
895 : * to be available for the index scan given that remote relation has
896 : * replica identity full.
897 : *
898 : * Note that we are not using the planner to find the cheapest method
899 : * to scan the relation as that would require us to either use lower
900 : * level planner functions which would be a maintenance burden in the
901 : * long run or use the full-fledged planner which could cause
902 : * overhead.
903 : */
904 112 : return FindUsableIndexForReplicaIdentityFull(localrel, attrMap);
905 : }
906 :
907 184 : return InvalidOid;
908 : }
|
