Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam_handler.c
4 : * heap table access method code
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam_handler.c
12 : *
13 : *
14 : * NOTES
15 : * This files wires up the lower level heapam.c et al routines with the
16 : * tableam abstraction.
17 : *
18 : *-------------------------------------------------------------------------
19 : */
20 : #include "postgres.h"
21 :
22 : #include "access/genam.h"
23 : #include "access/heapam.h"
24 : #include "access/heaptoast.h"
25 : #include "access/multixact.h"
26 : #include "access/rewriteheap.h"
27 : #include "access/syncscan.h"
28 : #include "access/tableam.h"
29 : #include "access/tsmapi.h"
30 : #include "access/visibilitymap.h"
31 : #include "access/xact.h"
32 : #include "catalog/catalog.h"
33 : #include "catalog/index.h"
34 : #include "catalog/storage.h"
35 : #include "catalog/storage_xlog.h"
36 : #include "commands/progress.h"
37 : #include "executor/executor.h"
38 : #include "miscadmin.h"
39 : #include "pgstat.h"
40 : #include "storage/bufmgr.h"
41 : #include "storage/bufpage.h"
42 : #include "storage/lmgr.h"
43 : #include "storage/predicate.h"
44 : #include "storage/procarray.h"
45 : #include "storage/smgr.h"
46 : #include "utils/builtins.h"
47 : #include "utils/rel.h"
48 :
49 : static void reform_and_rewrite_tuple(HeapTuple tuple,
50 : Relation OldHeap, Relation NewHeap,
51 : Datum *values, bool *isnull, RewriteState rwstate);
52 :
53 : static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
54 : HeapTuple tuple,
55 : OffsetNumber tupoffset);
56 :
57 : static BlockNumber heapam_scan_get_blocks_done(HeapScanDesc hscan);
58 :
59 : static bool BitmapHeapScanNextBlock(TableScanDesc scan,
60 : bool *recheck,
61 : uint64 *lossy_pages, uint64 *exact_pages);
62 :
63 :
64 : /* ------------------------------------------------------------------------
65 : * Slot related callbacks for heap AM
66 : * ------------------------------------------------------------------------
67 : */
68 :
69 : static const TupleTableSlotOps *
70 14566193 : heapam_slot_callbacks(Relation relation)
71 : {
72 14566193 : return &TTSOpsBufferHeapTuple;
73 : }
74 :
75 :
76 : /* ------------------------------------------------------------------------
77 : * Index Scan Callbacks for heap AM
78 : * ------------------------------------------------------------------------
79 : */
80 :
81 : static IndexFetchTableData *
82 13946401 : heapam_index_fetch_begin(Relation rel)
83 : {
84 13946401 : IndexFetchHeapData *hscan = palloc0_object(IndexFetchHeapData);
85 :
86 13946401 : hscan->xs_base.rel = rel;
87 13946401 : hscan->xs_cbuf = InvalidBuffer;
88 :
89 13946401 : return &hscan->xs_base;
90 : }
91 :
92 : static void
93 26336399 : heapam_index_fetch_reset(IndexFetchTableData *scan)
94 : {
95 26336399 : IndexFetchHeapData *hscan = (IndexFetchHeapData *) scan;
96 :
97 26336399 : if (BufferIsValid(hscan->xs_cbuf))
98 : {
99 11860690 : ReleaseBuffer(hscan->xs_cbuf);
100 11860690 : hscan->xs_cbuf = InvalidBuffer;
101 : }
102 26336399 : }
103 :
104 : static void
105 13945498 : heapam_index_fetch_end(IndexFetchTableData *scan)
106 : {
107 13945498 : IndexFetchHeapData *hscan = (IndexFetchHeapData *) scan;
108 :
109 13945498 : heapam_index_fetch_reset(scan);
110 :
111 13945498 : pfree(hscan);
112 13945498 : }
113 :
114 : static bool
115 20006544 : heapam_index_fetch_tuple(struct IndexFetchTableData *scan,
116 : ItemPointer tid,
117 : Snapshot snapshot,
118 : TupleTableSlot *slot,
119 : bool *call_again, bool *all_dead)
120 : {
121 20006544 : IndexFetchHeapData *hscan = (IndexFetchHeapData *) scan;
122 20006544 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
123 : bool got_heap_tuple;
124 :
125 : Assert(TTS_IS_BUFFERTUPLE(slot));
126 :
127 : /* We can skip the buffer-switching logic if we're in mid-HOT chain. */
128 20006544 : if (!*call_again)
129 : {
130 : /* Switch to correct buffer if we don't have it already */
131 19919978 : Buffer prev_buf = hscan->xs_cbuf;
132 :
133 19919978 : hscan->xs_cbuf = ReleaseAndReadBuffer(hscan->xs_cbuf,
134 : hscan->xs_base.rel,
135 : ItemPointerGetBlockNumber(tid));
136 :
137 : /*
138 : * Prune page, but only if we weren't already on this page
139 : */
140 19919975 : if (prev_buf != hscan->xs_cbuf)
141 13807395 : heap_page_prune_opt(hscan->xs_base.rel, hscan->xs_cbuf);
142 : }
143 :
144 : /* Obtain share-lock on the buffer so we can examine visibility */
145 20006541 : LockBuffer(hscan->xs_cbuf, BUFFER_LOCK_SHARE);
146 20006541 : got_heap_tuple = heap_hot_search_buffer(tid,
147 : hscan->xs_base.rel,
148 : hscan->xs_cbuf,
149 : snapshot,
150 : &bslot->base.tupdata,
151 : all_dead,
152 20006541 : !*call_again);
153 20006539 : bslot->base.tupdata.t_self = *tid;
154 20006539 : LockBuffer(hscan->xs_cbuf, BUFFER_LOCK_UNLOCK);
155 :
156 20006539 : if (got_heap_tuple)
157 : {
158 : /*
159 : * Only in a non-MVCC snapshot can more than one member of the HOT
160 : * chain be visible.
161 : */
162 13697663 : *call_again = !IsMVCCSnapshot(snapshot);
163 :
164 13697663 : slot->tts_tableOid = RelationGetRelid(scan->rel);
165 13697663 : ExecStoreBufferHeapTuple(&bslot->base.tupdata, slot, hscan->xs_cbuf);
166 : }
167 : else
168 : {
169 : /* We've reached the end of the HOT chain. */
170 6308876 : *call_again = false;
171 : }
172 :
173 20006539 : return got_heap_tuple;
174 : }
175 :
176 :
177 : /* ------------------------------------------------------------------------
178 : * Callbacks for non-modifying operations on individual tuples for heap AM
179 : * ------------------------------------------------------------------------
180 : */
181 :
182 : static bool
183 180076 : heapam_fetch_row_version(Relation relation,
184 : ItemPointer tid,
185 : Snapshot snapshot,
186 : TupleTableSlot *slot)
187 : {
188 180076 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
189 : Buffer buffer;
190 :
191 : Assert(TTS_IS_BUFFERTUPLE(slot));
192 :
193 180076 : bslot->base.tupdata.t_self = *tid;
194 180076 : if (heap_fetch(relation, snapshot, &bslot->base.tupdata, &buffer, false))
195 : {
196 : /* store in slot, transferring existing pin */
197 179734 : ExecStorePinnedBufferHeapTuple(&bslot->base.tupdata, slot, buffer);
198 179734 : slot->tts_tableOid = RelationGetRelid(relation);
199 :
200 179734 : return true;
201 : }
202 :
203 334 : return false;
204 : }
205 :
206 : static bool
207 383 : heapam_tuple_tid_valid(TableScanDesc scan, ItemPointer tid)
208 : {
209 383 : HeapScanDesc hscan = (HeapScanDesc) scan;
210 :
211 757 : return ItemPointerIsValid(tid) &&
212 374 : ItemPointerGetBlockNumber(tid) < hscan->rs_nblocks;
213 : }
214 :
215 : static bool
216 128802 : heapam_tuple_satisfies_snapshot(Relation rel, TupleTableSlot *slot,
217 : Snapshot snapshot)
218 : {
219 128802 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
220 : bool res;
221 :
222 : Assert(TTS_IS_BUFFERTUPLE(slot));
223 : Assert(BufferIsValid(bslot->buffer));
224 :
225 : /*
226 : * We need buffer pin and lock to call HeapTupleSatisfiesVisibility.
227 : * Caller should be holding pin, but not lock.
228 : */
229 128802 : LockBuffer(bslot->buffer, BUFFER_LOCK_SHARE);
230 128802 : res = HeapTupleSatisfiesVisibility(bslot->base.tuple, snapshot,
231 : bslot->buffer);
232 128802 : LockBuffer(bslot->buffer, BUFFER_LOCK_UNLOCK);
233 :
234 128802 : return res;
235 : }
236 :
237 :
238 : /* ----------------------------------------------------------------------------
239 : * Functions for manipulations of physical tuples for heap AM.
240 : * ----------------------------------------------------------------------------
241 : */
242 :
243 : static void
244 7431889 : heapam_tuple_insert(Relation relation, TupleTableSlot *slot, CommandId cid,
245 : int options, BulkInsertState bistate)
246 : {
247 7431889 : bool shouldFree = true;
248 7431889 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
249 :
250 : /* Update the tuple with table oid */
251 7431889 : slot->tts_tableOid = RelationGetRelid(relation);
252 7431889 : tuple->t_tableOid = slot->tts_tableOid;
253 :
254 : /* Perform the insertion, and copy the resulting ItemPointer */
255 7431889 : heap_insert(relation, tuple, cid, options, bistate);
256 7431871 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
257 :
258 7431871 : if (shouldFree)
259 1487086 : pfree(tuple);
260 7431871 : }
261 :
262 : static void
263 2133 : heapam_tuple_insert_speculative(Relation relation, TupleTableSlot *slot,
264 : CommandId cid, int options,
265 : BulkInsertState bistate, uint32 specToken)
266 : {
267 2133 : bool shouldFree = true;
268 2133 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
269 :
270 : /* Update the tuple with table oid */
271 2133 : slot->tts_tableOid = RelationGetRelid(relation);
272 2133 : tuple->t_tableOid = slot->tts_tableOid;
273 :
274 2133 : HeapTupleHeaderSetSpeculativeToken(tuple->t_data, specToken);
275 2133 : options |= HEAP_INSERT_SPECULATIVE;
276 :
277 : /* Perform the insertion, and copy the resulting ItemPointer */
278 2133 : heap_insert(relation, tuple, cid, options, bistate);
279 2133 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
280 :
281 2133 : if (shouldFree)
282 45 : pfree(tuple);
283 2133 : }
284 :
285 : static void
286 2130 : heapam_tuple_complete_speculative(Relation relation, TupleTableSlot *slot,
287 : uint32 specToken, bool succeeded)
288 : {
289 2130 : bool shouldFree = true;
290 2130 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
291 :
292 : /* adjust the tuple's state accordingly */
293 2130 : if (succeeded)
294 2119 : heap_finish_speculative(relation, &slot->tts_tid);
295 : else
296 11 : heap_abort_speculative(relation, &slot->tts_tid);
297 :
298 2130 : if (shouldFree)
299 45 : pfree(tuple);
300 2130 : }
301 :
302 : static TM_Result
303 865983 : heapam_tuple_delete(Relation relation, ItemPointer tid, CommandId cid,
304 : Snapshot snapshot, Snapshot crosscheck, bool wait,
305 : TM_FailureData *tmfd, bool changingPart)
306 : {
307 : /*
308 : * Currently Deleting of index tuples are handled at vacuum, in case if
309 : * the storage itself is cleaning the dead tuples by itself, it is the
310 : * time to call the index tuple deletion also.
311 : */
312 865983 : return heap_delete(relation, tid, cid, crosscheck, wait, tmfd, changingPart);
313 : }
314 :
315 :
316 : static TM_Result
317 195860 : heapam_tuple_update(Relation relation, ItemPointer otid, TupleTableSlot *slot,
318 : CommandId cid, Snapshot snapshot, Snapshot crosscheck,
319 : bool wait, TM_FailureData *tmfd,
320 : LockTupleMode *lockmode, TU_UpdateIndexes *update_indexes)
321 : {
322 195860 : bool shouldFree = true;
323 195860 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
324 : TM_Result result;
325 :
326 : /* Update the tuple with table oid */
327 195860 : slot->tts_tableOid = RelationGetRelid(relation);
328 195860 : tuple->t_tableOid = slot->tts_tableOid;
329 :
330 195860 : result = heap_update(relation, otid, tuple, cid, crosscheck, wait,
331 : tmfd, lockmode, update_indexes);
332 195848 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
333 :
334 : /*
335 : * Decide whether new index entries are needed for the tuple
336 : *
337 : * Note: heap_update returns the tid (location) of the new tuple in the
338 : * t_self field.
339 : *
340 : * If the update is not HOT, we must update all indexes. If the update is
341 : * HOT, it could be that we updated summarized columns, so we either
342 : * update only summarized indexes, or none at all.
343 : */
344 195848 : if (result != TM_Ok)
345 : {
346 : Assert(*update_indexes == TU_None);
347 164 : *update_indexes = TU_None;
348 : }
349 195684 : else if (!HeapTupleIsHeapOnly(tuple))
350 : Assert(*update_indexes == TU_All);
351 : else
352 : Assert((*update_indexes == TU_Summarizing) ||
353 : (*update_indexes == TU_None));
354 :
355 195848 : if (shouldFree)
356 31946 : pfree(tuple);
357 :
358 195848 : return result;
359 : }
360 :
361 : static TM_Result
362 158338 : heapam_tuple_lock(Relation relation, ItemPointer tid, Snapshot snapshot,
363 : TupleTableSlot *slot, CommandId cid, LockTupleMode mode,
364 : LockWaitPolicy wait_policy, uint8 flags,
365 : TM_FailureData *tmfd)
366 : {
367 158338 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
368 : TM_Result result;
369 : Buffer buffer;
370 158338 : HeapTuple tuple = &bslot->base.tupdata;
371 : bool follow_updates;
372 :
373 158338 : follow_updates = (flags & TUPLE_LOCK_FLAG_LOCK_UPDATE_IN_PROGRESS) != 0;
374 158338 : tmfd->traversed = false;
375 :
376 : Assert(TTS_IS_BUFFERTUPLE(slot));
377 :
378 158513 : tuple_lock_retry:
379 158513 : tuple->t_self = *tid;
380 158513 : result = heap_lock_tuple(relation, tuple, cid, mode, wait_policy,
381 : follow_updates, &buffer, tmfd);
382 :
383 158500 : if (result == TM_Updated &&
384 217 : (flags & TUPLE_LOCK_FLAG_FIND_LAST_VERSION))
385 : {
386 : /* Should not encounter speculative tuple on recheck */
387 : Assert(!HeapTupleHeaderIsSpeculative(tuple->t_data));
388 :
389 198 : ReleaseBuffer(buffer);
390 :
391 198 : if (!ItemPointerEquals(&tmfd->ctid, &tuple->t_self))
392 : {
393 : SnapshotData SnapshotDirty;
394 : TransactionId priorXmax;
395 :
396 : /* it was updated, so look at the updated version */
397 198 : *tid = tmfd->ctid;
398 : /* updated row should have xmin matching this xmax */
399 198 : priorXmax = tmfd->xmax;
400 :
401 : /* signal that a tuple later in the chain is getting locked */
402 198 : tmfd->traversed = true;
403 :
404 : /*
405 : * fetch target tuple
406 : *
407 : * Loop here to deal with updated or busy tuples
408 : */
409 198 : InitDirtySnapshot(SnapshotDirty);
410 : for (;;)
411 : {
412 229 : if (ItemPointerIndicatesMovedPartitions(tid))
413 11 : ereport(ERROR,
414 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
415 : errmsg("tuple to be locked was already moved to another partition due to concurrent update")));
416 :
417 218 : tuple->t_self = *tid;
418 218 : if (heap_fetch(relation, &SnapshotDirty, tuple, &buffer, true))
419 : {
420 : /*
421 : * If xmin isn't what we're expecting, the slot must have
422 : * been recycled and reused for an unrelated tuple. This
423 : * implies that the latest version of the row was deleted,
424 : * so we need do nothing. (Should be safe to examine xmin
425 : * without getting buffer's content lock. We assume
426 : * reading a TransactionId to be atomic, and Xmin never
427 : * changes in an existing tuple, except to invalid or
428 : * frozen, and neither of those can match priorXmax.)
429 : */
430 184 : if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple->t_data),
431 : priorXmax))
432 : {
433 0 : ReleaseBuffer(buffer);
434 11 : return TM_Deleted;
435 : }
436 :
437 : /* otherwise xmin should not be dirty... */
438 184 : if (TransactionIdIsValid(SnapshotDirty.xmin))
439 0 : ereport(ERROR,
440 : (errcode(ERRCODE_DATA_CORRUPTED),
441 : errmsg_internal("t_xmin %u is uncommitted in tuple (%u,%u) to be updated in table \"%s\"",
442 : SnapshotDirty.xmin,
443 : ItemPointerGetBlockNumber(&tuple->t_self),
444 : ItemPointerGetOffsetNumber(&tuple->t_self),
445 : RelationGetRelationName(relation))));
446 :
447 : /*
448 : * If tuple is being updated by other transaction then we
449 : * have to wait for its commit/abort, or die trying.
450 : */
451 184 : if (TransactionIdIsValid(SnapshotDirty.xmax))
452 : {
453 2 : ReleaseBuffer(buffer);
454 2 : switch (wait_policy)
455 : {
456 0 : case LockWaitBlock:
457 0 : XactLockTableWait(SnapshotDirty.xmax,
458 0 : relation, &tuple->t_self,
459 : XLTW_FetchUpdated);
460 0 : break;
461 1 : case LockWaitSkip:
462 1 : if (!ConditionalXactLockTableWait(SnapshotDirty.xmax, false))
463 : /* skip instead of waiting */
464 1 : return TM_WouldBlock;
465 0 : break;
466 1 : case LockWaitError:
467 1 : if (!ConditionalXactLockTableWait(SnapshotDirty.xmax, log_lock_failures))
468 1 : ereport(ERROR,
469 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
470 : errmsg("could not obtain lock on row in relation \"%s\"",
471 : RelationGetRelationName(relation))));
472 0 : break;
473 : }
474 0 : continue; /* loop back to repeat heap_fetch */
475 : }
476 :
477 : /*
478 : * If tuple was inserted by our own transaction, we have
479 : * to check cmin against cid: cmin >= current CID means
480 : * our command cannot see the tuple, so we should ignore
481 : * it. Otherwise heap_lock_tuple() will throw an error,
482 : * and so would any later attempt to update or delete the
483 : * tuple. (We need not check cmax because
484 : * HeapTupleSatisfiesDirty will consider a tuple deleted
485 : * by our transaction dead, regardless of cmax.) We just
486 : * checked that priorXmax == xmin, so we can test that
487 : * variable instead of doing HeapTupleHeaderGetXmin again.
488 : */
489 189 : if (TransactionIdIsCurrentTransactionId(priorXmax) &&
490 7 : HeapTupleHeaderGetCmin(tuple->t_data) >= cid)
491 : {
492 7 : tmfd->xmax = priorXmax;
493 :
494 : /*
495 : * Cmin is the problematic value, so store that. See
496 : * above.
497 : */
498 7 : tmfd->cmax = HeapTupleHeaderGetCmin(tuple->t_data);
499 7 : ReleaseBuffer(buffer);
500 7 : return TM_SelfModified;
501 : }
502 :
503 : /*
504 : * This is a live tuple, so try to lock it again.
505 : */
506 175 : ReleaseBuffer(buffer);
507 175 : goto tuple_lock_retry;
508 : }
509 :
510 : /*
511 : * If the referenced slot was actually empty, the latest
512 : * version of the row must have been deleted, so we need do
513 : * nothing.
514 : */
515 34 : if (tuple->t_data == NULL)
516 : {
517 : Assert(!BufferIsValid(buffer));
518 0 : return TM_Deleted;
519 : }
520 :
521 : /*
522 : * As above, if xmin isn't what we're expecting, do nothing.
523 : */
524 34 : if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple->t_data),
525 : priorXmax))
526 : {
527 0 : ReleaseBuffer(buffer);
528 0 : return TM_Deleted;
529 : }
530 :
531 : /*
532 : * If we get here, the tuple was found but failed
533 : * SnapshotDirty. Assuming the xmin is either a committed xact
534 : * or our own xact (as it certainly should be if we're trying
535 : * to modify the tuple), this must mean that the row was
536 : * updated or deleted by either a committed xact or our own
537 : * xact. If it was deleted, we can ignore it; if it was
538 : * updated then chain up to the next version and repeat the
539 : * whole process.
540 : *
541 : * As above, it should be safe to examine xmax and t_ctid
542 : * without the buffer content lock, because they can't be
543 : * changing. We'd better hold a buffer pin though.
544 : */
545 34 : if (ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
546 : {
547 : /* deleted, so forget about it */
548 3 : ReleaseBuffer(buffer);
549 3 : return TM_Deleted;
550 : }
551 :
552 : /* updated, so look at the updated row */
553 31 : *tid = tuple->t_data->t_ctid;
554 : /* updated row should have xmin matching this xmax */
555 31 : priorXmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
556 31 : ReleaseBuffer(buffer);
557 : /* loop back to fetch next in chain */
558 : }
559 : }
560 : else
561 : {
562 : /* tuple was deleted, so give up */
563 0 : return TM_Deleted;
564 : }
565 : }
566 :
567 158302 : slot->tts_tableOid = RelationGetRelid(relation);
568 158302 : tuple->t_tableOid = slot->tts_tableOid;
569 :
570 : /* store in slot, transferring existing pin */
571 158302 : ExecStorePinnedBufferHeapTuple(tuple, slot, buffer);
572 :
573 158302 : return result;
574 : }
575 :
576 :
577 : /* ------------------------------------------------------------------------
578 : * DDL related callbacks for heap AM.
579 : * ------------------------------------------------------------------------
580 : */
581 :
582 : static void
583 34064 : heapam_relation_set_new_filelocator(Relation rel,
584 : const RelFileLocator *newrlocator,
585 : char persistence,
586 : TransactionId *freezeXid,
587 : MultiXactId *minmulti)
588 : {
589 : SMgrRelation srel;
590 :
591 : /*
592 : * Initialize to the minimum XID that could put tuples in the table. We
593 : * know that no xacts older than RecentXmin are still running, so that
594 : * will do.
595 : */
596 34064 : *freezeXid = RecentXmin;
597 :
598 : /*
599 : * Similarly, initialize the minimum Multixact to the first value that
600 : * could possibly be stored in tuples in the table. Running transactions
601 : * could reuse values from their local cache, so we are careful to
602 : * consider all currently running multis.
603 : *
604 : * XXX this could be refined further, but is it worth the hassle?
605 : */
606 34064 : *minmulti = GetOldestMultiXactId();
607 :
608 34064 : srel = RelationCreateStorage(*newrlocator, persistence, true);
609 :
610 : /*
611 : * If required, set up an init fork for an unlogged table so that it can
612 : * be correctly reinitialized on restart.
613 : */
614 34064 : if (persistence == RELPERSISTENCE_UNLOGGED)
615 : {
616 : Assert(rel->rd_rel->relkind == RELKIND_RELATION ||
617 : rel->rd_rel->relkind == RELKIND_TOASTVALUE);
618 136 : smgrcreate(srel, INIT_FORKNUM, false);
619 136 : log_smgrcreate(newrlocator, INIT_FORKNUM);
620 : }
621 :
622 34064 : smgrclose(srel);
623 34064 : }
624 :
625 : static void
626 314 : heapam_relation_nontransactional_truncate(Relation rel)
627 : {
628 314 : RelationTruncate(rel, 0);
629 314 : }
630 :
631 : static void
632 49 : heapam_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
633 : {
634 : SMgrRelation dstrel;
635 :
636 : /*
637 : * Since we copy the file directly without looking at the shared buffers,
638 : * we'd better first flush out any pages of the source relation that are
639 : * in shared buffers. We assume no new changes will be made while we are
640 : * holding exclusive lock on the rel.
641 : */
642 49 : FlushRelationBuffers(rel);
643 :
644 : /*
645 : * Create and copy all forks of the relation, and schedule unlinking of
646 : * old physical files.
647 : *
648 : * NOTE: any conflict in relfilenumber value will be caught in
649 : * RelationCreateStorage().
650 : */
651 49 : dstrel = RelationCreateStorage(*newrlocator, rel->rd_rel->relpersistence, true);
652 :
653 : /* copy main fork */
654 49 : RelationCopyStorage(RelationGetSmgr(rel), dstrel, MAIN_FORKNUM,
655 49 : rel->rd_rel->relpersistence);
656 :
657 : /* copy those extra forks that exist */
658 49 : for (ForkNumber forkNum = MAIN_FORKNUM + 1;
659 196 : forkNum <= MAX_FORKNUM; forkNum++)
660 : {
661 147 : if (smgrexists(RelationGetSmgr(rel), forkNum))
662 : {
663 9 : smgrcreate(dstrel, forkNum, false);
664 :
665 : /*
666 : * WAL log creation if the relation is persistent, or this is the
667 : * init fork of an unlogged relation.
668 : */
669 9 : if (RelationIsPermanent(rel) ||
670 3 : (rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
671 : forkNum == INIT_FORKNUM))
672 6 : log_smgrcreate(newrlocator, forkNum);
673 9 : RelationCopyStorage(RelationGetSmgr(rel), dstrel, forkNum,
674 9 : rel->rd_rel->relpersistence);
675 : }
676 : }
677 :
678 :
679 : /* drop old relation, and close new one */
680 49 : RelationDropStorage(rel);
681 49 : smgrclose(dstrel);
682 49 : }
683 :
684 : static void
685 317 : heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
686 : Relation OldIndex, bool use_sort,
687 : TransactionId OldestXmin,
688 : TransactionId *xid_cutoff,
689 : MultiXactId *multi_cutoff,
690 : double *num_tuples,
691 : double *tups_vacuumed,
692 : double *tups_recently_dead)
693 : {
694 : RewriteState rwstate;
695 : IndexScanDesc indexScan;
696 : TableScanDesc tableScan;
697 : HeapScanDesc heapScan;
698 : bool is_system_catalog;
699 : Tuplesortstate *tuplesort;
700 317 : TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
701 317 : TupleDesc newTupDesc = RelationGetDescr(NewHeap);
702 : TupleTableSlot *slot;
703 : int natts;
704 : Datum *values;
705 : bool *isnull;
706 : BufferHeapTupleTableSlot *hslot;
707 317 : BlockNumber prev_cblock = InvalidBlockNumber;
708 :
709 : /* Remember if it's a system catalog */
710 317 : is_system_catalog = IsSystemRelation(OldHeap);
711 :
712 : /*
713 : * Valid smgr_targblock implies something already wrote to the relation.
714 : * This may be harmless, but this function hasn't planned for it.
715 : */
716 : Assert(RelationGetTargetBlock(NewHeap) == InvalidBlockNumber);
717 :
718 : /* Preallocate values/isnull arrays */
719 317 : natts = newTupDesc->natts;
720 317 : values = palloc_array(Datum, natts);
721 317 : isnull = palloc_array(bool, natts);
722 :
723 : /* Initialize the rewrite operation */
724 317 : rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin, *xid_cutoff,
725 : *multi_cutoff);
726 :
727 :
728 : /* Set up sorting if wanted */
729 317 : if (use_sort)
730 67 : tuplesort = tuplesort_begin_cluster(oldTupDesc, OldIndex,
731 : maintenance_work_mem,
732 : NULL, TUPLESORT_NONE);
733 : else
734 250 : tuplesort = NULL;
735 :
736 : /*
737 : * Prepare to scan the OldHeap. To ensure we see recently-dead tuples
738 : * that still need to be copied, we scan with SnapshotAny and use
739 : * HeapTupleSatisfiesVacuum for the visibility test.
740 : */
741 317 : if (OldIndex != NULL && !use_sort)
742 39 : {
743 39 : const int ci_index[] = {
744 : PROGRESS_REPACK_PHASE,
745 : PROGRESS_REPACK_INDEX_RELID
746 : };
747 : int64 ci_val[2];
748 :
749 : /* Set phase and OIDOldIndex to columns */
750 39 : ci_val[0] = PROGRESS_REPACK_PHASE_INDEX_SCAN_HEAP;
751 39 : ci_val[1] = RelationGetRelid(OldIndex);
752 39 : pgstat_progress_update_multi_param(2, ci_index, ci_val);
753 :
754 39 : tableScan = NULL;
755 39 : heapScan = NULL;
756 39 : indexScan = index_beginscan(OldHeap, OldIndex, SnapshotAny, NULL, 0, 0);
757 39 : index_rescan(indexScan, NULL, 0, NULL, 0);
758 : }
759 : else
760 : {
761 : /* In scan-and-sort mode and also VACUUM FULL, set phase */
762 278 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
763 : PROGRESS_REPACK_PHASE_SEQ_SCAN_HEAP);
764 :
765 278 : tableScan = table_beginscan(OldHeap, SnapshotAny, 0, (ScanKey) NULL);
766 278 : heapScan = (HeapScanDesc) tableScan;
767 278 : indexScan = NULL;
768 :
769 : /* Set total heap blocks */
770 278 : pgstat_progress_update_param(PROGRESS_REPACK_TOTAL_HEAP_BLKS,
771 278 : heapScan->rs_nblocks);
772 : }
773 :
774 317 : slot = table_slot_create(OldHeap, NULL);
775 317 : hslot = (BufferHeapTupleTableSlot *) slot;
776 :
777 : /*
778 : * Scan through the OldHeap, either in OldIndex order or sequentially;
779 : * copy each tuple into the NewHeap, or transiently to the tuplesort
780 : * module. Note that we don't bother sorting dead tuples (they won't get
781 : * to the new table anyway).
782 : */
783 : for (;;)
784 366989 : {
785 : HeapTuple tuple;
786 : Buffer buf;
787 : bool isdead;
788 :
789 367306 : CHECK_FOR_INTERRUPTS();
790 :
791 367306 : if (indexScan != NULL)
792 : {
793 93 : if (!index_getnext_slot(indexScan, ForwardScanDirection, slot))
794 39 : break;
795 :
796 : /* Since we used no scan keys, should never need to recheck */
797 54 : if (indexScan->xs_recheck)
798 0 : elog(ERROR, "CLUSTER does not support lossy index conditions");
799 : }
800 : else
801 : {
802 367213 : if (!table_scan_getnextslot(tableScan, ForwardScanDirection, slot))
803 : {
804 : /*
805 : * If the last pages of the scan were empty, we would go to
806 : * the next phase while heap_blks_scanned != heap_blks_total.
807 : * Instead, to ensure that heap_blks_scanned is equivalent to
808 : * heap_blks_total after the table scan phase, this parameter
809 : * is manually updated to the correct value when the table
810 : * scan finishes.
811 : */
812 278 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_BLKS_SCANNED,
813 278 : heapScan->rs_nblocks);
814 278 : break;
815 : }
816 :
817 : /*
818 : * In scan-and-sort mode and also VACUUM FULL, set heap blocks
819 : * scanned
820 : *
821 : * Note that heapScan may start at an offset and wrap around, i.e.
822 : * rs_startblock may be >0, and rs_cblock may end with a number
823 : * below rs_startblock. To prevent showing this wraparound to the
824 : * user, we offset rs_cblock by rs_startblock (modulo rs_nblocks).
825 : */
826 366935 : if (prev_cblock != heapScan->rs_cblock)
827 : {
828 5520 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_BLKS_SCANNED,
829 5520 : (heapScan->rs_cblock +
830 5520 : heapScan->rs_nblocks -
831 5520 : heapScan->rs_startblock
832 5520 : ) % heapScan->rs_nblocks + 1);
833 5520 : prev_cblock = heapScan->rs_cblock;
834 : }
835 : }
836 :
837 366989 : tuple = ExecFetchSlotHeapTuple(slot, false, NULL);
838 366989 : buf = hslot->buffer;
839 :
840 : /*
841 : * To be able to guarantee that we can set the hint bit, acquire an
842 : * exclusive lock on the old buffer. We need the hint bits, set in
843 : * heapam_relation_copy_for_cluster() -> HeapTupleSatisfiesVacuum(),
844 : * to be set, as otherwise reform_and_rewrite_tuple() ->
845 : * rewrite_heap_tuple() will get confused. Specifically,
846 : * rewrite_heap_tuple() checks for HEAP_XMAX_INVALID in the old tuple
847 : * to determine whether to check the old-to-new mapping hash table.
848 : *
849 : * It'd be better if we somehow could avoid setting hint bits on the
850 : * old page. One reason to use VACUUM FULL are very bloated tables -
851 : * rewriting most of the old table during VACUUM FULL doesn't exactly
852 : * help...
853 : */
854 366989 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
855 :
856 366989 : switch (HeapTupleSatisfiesVacuum(tuple, OldestXmin, buf))
857 : {
858 12330 : case HEAPTUPLE_DEAD:
859 : /* Definitely dead */
860 12330 : isdead = true;
861 12330 : break;
862 6686 : case HEAPTUPLE_RECENTLY_DEAD:
863 6686 : *tups_recently_dead += 1;
864 : pg_fallthrough;
865 354533 : case HEAPTUPLE_LIVE:
866 : /* Live or recently dead, must copy it */
867 354533 : isdead = false;
868 354533 : break;
869 103 : case HEAPTUPLE_INSERT_IN_PROGRESS:
870 :
871 : /*
872 : * Since we hold exclusive lock on the relation, normally the
873 : * only way to see this is if it was inserted earlier in our
874 : * own transaction. However, it can happen in system
875 : * catalogs, since we tend to release write lock before commit
876 : * there. Give a warning if neither case applies; but in any
877 : * case we had better copy it.
878 : */
879 103 : if (!is_system_catalog &&
880 11 : !TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetXmin(tuple->t_data)))
881 0 : elog(WARNING, "concurrent insert in progress within table \"%s\"",
882 : RelationGetRelationName(OldHeap));
883 : /* treat as live */
884 103 : isdead = false;
885 103 : break;
886 23 : case HEAPTUPLE_DELETE_IN_PROGRESS:
887 :
888 : /*
889 : * Similar situation to INSERT_IN_PROGRESS case.
890 : */
891 23 : if (!is_system_catalog &&
892 15 : !TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetUpdateXid(tuple->t_data)))
893 0 : elog(WARNING, "concurrent delete in progress within table \"%s\"",
894 : RelationGetRelationName(OldHeap));
895 : /* treat as recently dead */
896 23 : *tups_recently_dead += 1;
897 23 : isdead = false;
898 23 : break;
899 0 : default:
900 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
901 : isdead = false; /* keep compiler quiet */
902 : break;
903 : }
904 :
905 366989 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
906 :
907 366989 : if (isdead)
908 : {
909 12330 : *tups_vacuumed += 1;
910 : /* heap rewrite module still needs to see it... */
911 12330 : if (rewrite_heap_dead_tuple(rwstate, tuple))
912 : {
913 : /* A previous recently-dead tuple is now known dead */
914 0 : *tups_vacuumed += 1;
915 0 : *tups_recently_dead -= 1;
916 : }
917 12330 : continue;
918 : }
919 :
920 354659 : *num_tuples += 1;
921 354659 : if (tuplesort != NULL)
922 : {
923 273813 : tuplesort_putheaptuple(tuplesort, tuple);
924 :
925 : /*
926 : * In scan-and-sort mode, report increase in number of tuples
927 : * scanned
928 : */
929 273813 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_SCANNED,
930 273813 : *num_tuples);
931 : }
932 : else
933 : {
934 80846 : const int ct_index[] = {
935 : PROGRESS_REPACK_HEAP_TUPLES_SCANNED,
936 : PROGRESS_REPACK_HEAP_TUPLES_WRITTEN
937 : };
938 : int64 ct_val[2];
939 :
940 80846 : reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
941 : values, isnull, rwstate);
942 :
943 : /*
944 : * In indexscan mode and also VACUUM FULL, report increase in
945 : * number of tuples scanned and written
946 : */
947 80846 : ct_val[0] = *num_tuples;
948 80846 : ct_val[1] = *num_tuples;
949 80846 : pgstat_progress_update_multi_param(2, ct_index, ct_val);
950 : }
951 : }
952 :
953 317 : if (indexScan != NULL)
954 39 : index_endscan(indexScan);
955 317 : if (tableScan != NULL)
956 278 : table_endscan(tableScan);
957 317 : if (slot)
958 317 : ExecDropSingleTupleTableSlot(slot);
959 :
960 : /*
961 : * In scan-and-sort mode, complete the sort, then read out all live tuples
962 : * from the tuplestore and write them to the new relation.
963 : */
964 317 : if (tuplesort != NULL)
965 : {
966 67 : double n_tuples = 0;
967 :
968 : /* Report that we are now sorting tuples */
969 67 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
970 : PROGRESS_REPACK_PHASE_SORT_TUPLES);
971 :
972 67 : tuplesort_performsort(tuplesort);
973 :
974 : /* Report that we are now writing new heap */
975 67 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
976 : PROGRESS_REPACK_PHASE_WRITE_NEW_HEAP);
977 :
978 : for (;;)
979 273813 : {
980 : HeapTuple tuple;
981 :
982 273880 : CHECK_FOR_INTERRUPTS();
983 :
984 273880 : tuple = tuplesort_getheaptuple(tuplesort, true);
985 273880 : if (tuple == NULL)
986 67 : break;
987 :
988 273813 : n_tuples += 1;
989 273813 : reform_and_rewrite_tuple(tuple,
990 : OldHeap, NewHeap,
991 : values, isnull,
992 : rwstate);
993 : /* Report n_tuples */
994 273813 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_WRITTEN,
995 : n_tuples);
996 : }
997 :
998 67 : tuplesort_end(tuplesort);
999 : }
1000 :
1001 : /* Write out any remaining tuples, and fsync if needed */
1002 317 : end_heap_rewrite(rwstate);
1003 :
1004 : /* Clean up */
1005 317 : pfree(values);
1006 317 : pfree(isnull);
1007 317 : }
1008 :
1009 : /*
1010 : * Prepare to analyze the next block in the read stream. Returns false if
1011 : * the stream is exhausted and true otherwise. The scan must have been started
1012 : * with SO_TYPE_ANALYZE option.
1013 : *
1014 : * This routine holds a buffer pin and lock on the heap page. They are held
1015 : * until heapam_scan_analyze_next_tuple() returns false. That is until all the
1016 : * items of the heap page are analyzed.
1017 : */
1018 : static bool
1019 78119 : heapam_scan_analyze_next_block(TableScanDesc scan, ReadStream *stream)
1020 : {
1021 78119 : HeapScanDesc hscan = (HeapScanDesc) scan;
1022 :
1023 : /*
1024 : * We must maintain a pin on the target page's buffer to ensure that
1025 : * concurrent activity - e.g. HOT pruning - doesn't delete tuples out from
1026 : * under us. It comes from the stream already pinned. We also choose to
1027 : * hold sharelock on the buffer throughout --- we could release and
1028 : * re-acquire sharelock for each tuple, but since we aren't doing much
1029 : * work per tuple, the extra lock traffic is probably better avoided.
1030 : */
1031 78119 : hscan->rs_cbuf = read_stream_next_buffer(stream, NULL);
1032 78119 : if (!BufferIsValid(hscan->rs_cbuf))
1033 8912 : return false;
1034 :
1035 69207 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1036 :
1037 69207 : hscan->rs_cblock = BufferGetBlockNumber(hscan->rs_cbuf);
1038 69207 : hscan->rs_cindex = FirstOffsetNumber;
1039 69207 : return true;
1040 : }
1041 :
1042 : static bool
1043 5487596 : heapam_scan_analyze_next_tuple(TableScanDesc scan,
1044 : double *liverows, double *deadrows,
1045 : TupleTableSlot *slot)
1046 : {
1047 5487596 : HeapScanDesc hscan = (HeapScanDesc) scan;
1048 : Page targpage;
1049 : OffsetNumber maxoffset;
1050 : BufferHeapTupleTableSlot *hslot;
1051 :
1052 : Assert(TTS_IS_BUFFERTUPLE(slot));
1053 :
1054 5487596 : hslot = (BufferHeapTupleTableSlot *) slot;
1055 5487596 : targpage = BufferGetPage(hscan->rs_cbuf);
1056 5487596 : maxoffset = PageGetMaxOffsetNumber(targpage);
1057 :
1058 : /* Inner loop over all tuples on the selected page */
1059 5736538 : for (; hscan->rs_cindex <= maxoffset; hscan->rs_cindex++)
1060 : {
1061 : ItemId itemid;
1062 5667331 : HeapTuple targtuple = &hslot->base.tupdata;
1063 5667331 : bool sample_it = false;
1064 : TransactionId dead_after;
1065 :
1066 5667331 : itemid = PageGetItemId(targpage, hscan->rs_cindex);
1067 :
1068 : /*
1069 : * We ignore unused and redirect line pointers. DEAD line pointers
1070 : * should be counted as dead, because we need vacuum to run to get rid
1071 : * of them. Note that this rule agrees with the way that
1072 : * heap_page_prune_and_freeze() counts things.
1073 : */
1074 5667331 : if (!ItemIdIsNormal(itemid))
1075 : {
1076 167380 : if (ItemIdIsDead(itemid))
1077 76293 : *deadrows += 1;
1078 167380 : continue;
1079 : }
1080 :
1081 5499951 : ItemPointerSet(&targtuple->t_self, hscan->rs_cblock, hscan->rs_cindex);
1082 :
1083 5499951 : targtuple->t_tableOid = RelationGetRelid(scan->rs_rd);
1084 5499951 : targtuple->t_data = (HeapTupleHeader) PageGetItem(targpage, itemid);
1085 5499951 : targtuple->t_len = ItemIdGetLength(itemid);
1086 :
1087 5499951 : switch (HeapTupleSatisfiesVacuumHorizon(targtuple,
1088 : hscan->rs_cbuf,
1089 : &dead_after))
1090 : {
1091 5188744 : case HEAPTUPLE_LIVE:
1092 5188744 : sample_it = true;
1093 5188744 : *liverows += 1;
1094 5188744 : break;
1095 :
1096 80704 : case HEAPTUPLE_DEAD:
1097 : case HEAPTUPLE_RECENTLY_DEAD:
1098 : /* Count dead and recently-dead rows */
1099 80704 : *deadrows += 1;
1100 80704 : break;
1101 :
1102 173145 : case HEAPTUPLE_INSERT_IN_PROGRESS:
1103 :
1104 : /*
1105 : * Insert-in-progress rows are not counted. We assume that
1106 : * when the inserting transaction commits or aborts, it will
1107 : * send a stats message to increment the proper count. This
1108 : * works right only if that transaction ends after we finish
1109 : * analyzing the table; if things happen in the other order,
1110 : * its stats update will be overwritten by ours. However, the
1111 : * error will be large only if the other transaction runs long
1112 : * enough to insert many tuples, so assuming it will finish
1113 : * after us is the safer option.
1114 : *
1115 : * A special case is that the inserting transaction might be
1116 : * our own. In this case we should count and sample the row,
1117 : * to accommodate users who load a table and analyze it in one
1118 : * transaction. (pgstat_report_analyze has to adjust the
1119 : * numbers we report to the cumulative stats system to make
1120 : * this come out right.)
1121 : */
1122 173145 : if (TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetXmin(targtuple->t_data)))
1123 : {
1124 173145 : sample_it = true;
1125 173145 : *liverows += 1;
1126 : }
1127 173145 : break;
1128 :
1129 57358 : case HEAPTUPLE_DELETE_IN_PROGRESS:
1130 :
1131 : /*
1132 : * We count and sample delete-in-progress rows the same as
1133 : * live ones, so that the stats counters come out right if the
1134 : * deleting transaction commits after us, per the same
1135 : * reasoning given above.
1136 : *
1137 : * If the delete was done by our own transaction, however, we
1138 : * must count the row as dead to make pgstat_report_analyze's
1139 : * stats adjustments come out right. (Note: this works out
1140 : * properly when the row was both inserted and deleted in our
1141 : * xact.)
1142 : *
1143 : * The net effect of these choices is that we act as though an
1144 : * IN_PROGRESS transaction hasn't happened yet, except if it
1145 : * is our own transaction, which we assume has happened.
1146 : *
1147 : * This approach ensures that we behave sanely if we see both
1148 : * the pre-image and post-image rows for a row being updated
1149 : * by a concurrent transaction: we will sample the pre-image
1150 : * but not the post-image. We also get sane results if the
1151 : * concurrent transaction never commits.
1152 : */
1153 57358 : if (TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetUpdateXid(targtuple->t_data)))
1154 858 : *deadrows += 1;
1155 : else
1156 : {
1157 56500 : sample_it = true;
1158 56500 : *liverows += 1;
1159 : }
1160 57358 : break;
1161 :
1162 0 : default:
1163 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1164 : break;
1165 : }
1166 :
1167 5499951 : if (sample_it)
1168 : {
1169 5418389 : ExecStoreBufferHeapTuple(targtuple, slot, hscan->rs_cbuf);
1170 5418389 : hscan->rs_cindex++;
1171 :
1172 : /* note that we leave the buffer locked here! */
1173 5418389 : return true;
1174 : }
1175 : }
1176 :
1177 : /* Now release the lock and pin on the page */
1178 69207 : UnlockReleaseBuffer(hscan->rs_cbuf);
1179 69207 : hscan->rs_cbuf = InvalidBuffer;
1180 :
1181 : /* also prevent old slot contents from having pin on page */
1182 69207 : ExecClearTuple(slot);
1183 :
1184 69207 : return false;
1185 : }
1186 :
1187 : static double
1188 29357 : heapam_index_build_range_scan(Relation heapRelation,
1189 : Relation indexRelation,
1190 : IndexInfo *indexInfo,
1191 : bool allow_sync,
1192 : bool anyvisible,
1193 : bool progress,
1194 : BlockNumber start_blockno,
1195 : BlockNumber numblocks,
1196 : IndexBuildCallback callback,
1197 : void *callback_state,
1198 : TableScanDesc scan)
1199 : {
1200 : HeapScanDesc hscan;
1201 : bool is_system_catalog;
1202 : bool checking_uniqueness;
1203 : HeapTuple heapTuple;
1204 : Datum values[INDEX_MAX_KEYS];
1205 : bool isnull[INDEX_MAX_KEYS];
1206 : double reltuples;
1207 : ExprState *predicate;
1208 : TupleTableSlot *slot;
1209 : EState *estate;
1210 : ExprContext *econtext;
1211 : Snapshot snapshot;
1212 29357 : bool need_unregister_snapshot = false;
1213 : TransactionId OldestXmin;
1214 29357 : BlockNumber previous_blkno = InvalidBlockNumber;
1215 29357 : BlockNumber root_blkno = InvalidBlockNumber;
1216 : OffsetNumber root_offsets[MaxHeapTuplesPerPage];
1217 :
1218 : /*
1219 : * sanity checks
1220 : */
1221 : Assert(OidIsValid(indexRelation->rd_rel->relam));
1222 :
1223 : /* Remember if it's a system catalog */
1224 29357 : is_system_catalog = IsSystemRelation(heapRelation);
1225 :
1226 : /* See whether we're verifying uniqueness/exclusion properties */
1227 37112 : checking_uniqueness = (indexInfo->ii_Unique ||
1228 7755 : indexInfo->ii_ExclusionOps != NULL);
1229 :
1230 : /*
1231 : * "Any visible" mode is not compatible with uniqueness checks; make sure
1232 : * only one of those is requested.
1233 : */
1234 : Assert(!(anyvisible && checking_uniqueness));
1235 :
1236 : /*
1237 : * Need an EState for evaluation of index expressions and partial-index
1238 : * predicates. Also a slot to hold the current tuple.
1239 : */
1240 29357 : estate = CreateExecutorState();
1241 29357 : econtext = GetPerTupleExprContext(estate);
1242 29357 : slot = table_slot_create(heapRelation, NULL);
1243 :
1244 : /* Arrange for econtext's scan tuple to be the tuple under test */
1245 29357 : econtext->ecxt_scantuple = slot;
1246 :
1247 : /* Set up execution state for predicate, if any. */
1248 29357 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
1249 :
1250 : /*
1251 : * Prepare for scan of the base relation. In a normal index build, we use
1252 : * SnapshotAny because we must retrieve all tuples and do our own time
1253 : * qual checks (because we have to index RECENTLY_DEAD tuples). In a
1254 : * concurrent build, or during bootstrap, we take a regular MVCC snapshot
1255 : * and index whatever's live according to that.
1256 : */
1257 29357 : OldestXmin = InvalidTransactionId;
1258 :
1259 : /* okay to ignore lazy VACUUMs here */
1260 29357 : if (!IsBootstrapProcessingMode() && !indexInfo->ii_Concurrent)
1261 20790 : OldestXmin = GetOldestNonRemovableTransactionId(heapRelation);
1262 :
1263 29357 : if (!scan)
1264 : {
1265 : /*
1266 : * Serial index build.
1267 : *
1268 : * Must begin our own heap scan in this case. We may also need to
1269 : * register a snapshot whose lifetime is under our direct control.
1270 : */
1271 29074 : if (!TransactionIdIsValid(OldestXmin))
1272 : {
1273 8487 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1274 8487 : need_unregister_snapshot = true;
1275 : }
1276 : else
1277 20587 : snapshot = SnapshotAny;
1278 :
1279 29074 : scan = table_beginscan_strat(heapRelation, /* relation */
1280 : snapshot, /* snapshot */
1281 : 0, /* number of keys */
1282 : NULL, /* scan key */
1283 : true, /* buffer access strategy OK */
1284 : allow_sync); /* syncscan OK? */
1285 : }
1286 : else
1287 : {
1288 : /*
1289 : * Parallel index build.
1290 : *
1291 : * Parallel case never registers/unregisters own snapshot. Snapshot
1292 : * is taken from parallel heap scan, and is SnapshotAny or an MVCC
1293 : * snapshot, based on same criteria as serial case.
1294 : */
1295 : Assert(!IsBootstrapProcessingMode());
1296 : Assert(allow_sync);
1297 283 : snapshot = scan->rs_snapshot;
1298 : }
1299 :
1300 29357 : hscan = (HeapScanDesc) scan;
1301 :
1302 : /*
1303 : * Must have called GetOldestNonRemovableTransactionId() if using
1304 : * SnapshotAny. Shouldn't have for an MVCC snapshot. (It's especially
1305 : * worth checking this for parallel builds, since ambuild routines that
1306 : * support parallel builds must work these details out for themselves.)
1307 : */
1308 : Assert(snapshot == SnapshotAny || IsMVCCSnapshot(snapshot));
1309 : Assert(snapshot == SnapshotAny ? TransactionIdIsValid(OldestXmin) :
1310 : !TransactionIdIsValid(OldestXmin));
1311 : Assert(snapshot == SnapshotAny || !anyvisible);
1312 :
1313 : /* Publish number of blocks to scan */
1314 29357 : if (progress)
1315 : {
1316 : BlockNumber nblocks;
1317 :
1318 27705 : if (hscan->rs_base.rs_parallel != NULL)
1319 : {
1320 : ParallelBlockTableScanDesc pbscan;
1321 :
1322 103 : pbscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1323 103 : nblocks = pbscan->phs_nblocks;
1324 : }
1325 : else
1326 27602 : nblocks = hscan->rs_nblocks;
1327 :
1328 27705 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1329 : nblocks);
1330 : }
1331 :
1332 : /* set our scan endpoints */
1333 29357 : if (!allow_sync)
1334 1846 : heap_setscanlimits(scan, start_blockno, numblocks);
1335 : else
1336 : {
1337 : /* syncscan can only be requested on whole relation */
1338 : Assert(start_blockno == 0);
1339 : Assert(numblocks == InvalidBlockNumber);
1340 : }
1341 :
1342 29357 : reltuples = 0;
1343 :
1344 : /*
1345 : * Scan all tuples in the base relation.
1346 : */
1347 8878165 : while ((heapTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1348 : {
1349 : bool tupleIsAlive;
1350 :
1351 8848814 : CHECK_FOR_INTERRUPTS();
1352 :
1353 : /* Report scan progress, if asked to. */
1354 8848814 : if (progress)
1355 : {
1356 7418021 : BlockNumber blocks_done = heapam_scan_get_blocks_done(hscan);
1357 :
1358 7418021 : if (blocks_done != previous_blkno)
1359 : {
1360 96175 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1361 : blocks_done);
1362 96175 : previous_blkno = blocks_done;
1363 : }
1364 : }
1365 :
1366 : /*
1367 : * When dealing with a HOT-chain of updated tuples, we want to index
1368 : * the values of the live tuple (if any), but index it under the TID
1369 : * of the chain's root tuple. This approach is necessary to preserve
1370 : * the HOT-chain structure in the heap. So we need to be able to find
1371 : * the root item offset for every tuple that's in a HOT-chain. When
1372 : * first reaching a new page of the relation, call
1373 : * heap_get_root_tuples() to build a map of root item offsets on the
1374 : * page.
1375 : *
1376 : * It might look unsafe to use this information across buffer
1377 : * lock/unlock. However, we hold ShareLock on the table so no
1378 : * ordinary insert/update/delete should occur; and we hold pin on the
1379 : * buffer continuously while visiting the page, so no pruning
1380 : * operation can occur either.
1381 : *
1382 : * In cases with only ShareUpdateExclusiveLock on the table, it's
1383 : * possible for some HOT tuples to appear that we didn't know about
1384 : * when we first read the page. To handle that case, we re-obtain the
1385 : * list of root offsets when a HOT tuple points to a root item that we
1386 : * don't know about.
1387 : *
1388 : * Also, although our opinions about tuple liveness could change while
1389 : * we scan the page (due to concurrent transaction commits/aborts),
1390 : * the chain root locations won't, so this info doesn't need to be
1391 : * rebuilt after waiting for another transaction.
1392 : *
1393 : * Note the implied assumption that there is no more than one live
1394 : * tuple per HOT-chain --- else we could create more than one index
1395 : * entry pointing to the same root tuple.
1396 : */
1397 8848814 : if (hscan->rs_cblock != root_blkno)
1398 : {
1399 110675 : Page page = BufferGetPage(hscan->rs_cbuf);
1400 :
1401 110675 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1402 110675 : heap_get_root_tuples(page, root_offsets);
1403 110675 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1404 :
1405 110675 : root_blkno = hscan->rs_cblock;
1406 : }
1407 :
1408 8848814 : if (snapshot == SnapshotAny)
1409 : {
1410 : /* do our own time qual check */
1411 : bool indexIt;
1412 : TransactionId xwait;
1413 :
1414 7007764 : recheck:
1415 :
1416 : /*
1417 : * We could possibly get away with not locking the buffer here,
1418 : * since caller should hold ShareLock on the relation, but let's
1419 : * be conservative about it. (This remark is still correct even
1420 : * with HOT-pruning: our pin on the buffer prevents pruning.)
1421 : */
1422 7007764 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1423 :
1424 : /*
1425 : * The criteria for counting a tuple as live in this block need to
1426 : * match what analyze.c's heapam_scan_analyze_next_tuple() does,
1427 : * otherwise CREATE INDEX and ANALYZE may produce wildly different
1428 : * reltuples values, e.g. when there are many recently-dead
1429 : * tuples.
1430 : */
1431 7007764 : switch (HeapTupleSatisfiesVacuum(heapTuple, OldestXmin,
1432 : hscan->rs_cbuf))
1433 : {
1434 1099 : case HEAPTUPLE_DEAD:
1435 : /* Definitely dead, we can ignore it */
1436 1099 : indexIt = false;
1437 1099 : tupleIsAlive = false;
1438 1099 : break;
1439 5116910 : case HEAPTUPLE_LIVE:
1440 : /* Normal case, index and unique-check it */
1441 5116910 : indexIt = true;
1442 5116910 : tupleIsAlive = true;
1443 : /* Count it as live, too */
1444 5116910 : reltuples += 1;
1445 5116910 : break;
1446 116073 : case HEAPTUPLE_RECENTLY_DEAD:
1447 :
1448 : /*
1449 : * If tuple is recently deleted then we must index it
1450 : * anyway to preserve MVCC semantics. (Pre-existing
1451 : * transactions could try to use the index after we finish
1452 : * building it, and may need to see such tuples.)
1453 : *
1454 : * However, if it was HOT-updated then we must only index
1455 : * the live tuple at the end of the HOT-chain. Since this
1456 : * breaks semantics for pre-existing snapshots, mark the
1457 : * index as unusable for them.
1458 : *
1459 : * We don't count recently-dead tuples in reltuples, even
1460 : * if we index them; see heapam_scan_analyze_next_tuple().
1461 : */
1462 116073 : if (HeapTupleIsHotUpdated(heapTuple))
1463 : {
1464 92 : indexIt = false;
1465 : /* mark the index as unsafe for old snapshots */
1466 92 : indexInfo->ii_BrokenHotChain = true;
1467 : }
1468 : else
1469 115981 : indexIt = true;
1470 : /* In any case, exclude the tuple from unique-checking */
1471 116073 : tupleIsAlive = false;
1472 116073 : break;
1473 1773643 : case HEAPTUPLE_INSERT_IN_PROGRESS:
1474 :
1475 : /*
1476 : * In "anyvisible" mode, this tuple is visible and we
1477 : * don't need any further checks.
1478 : */
1479 1773643 : if (anyvisible)
1480 : {
1481 30736 : indexIt = true;
1482 30736 : tupleIsAlive = true;
1483 30736 : reltuples += 1;
1484 30736 : break;
1485 : }
1486 :
1487 : /*
1488 : * Since caller should hold ShareLock or better, normally
1489 : * the only way to see this is if it was inserted earlier
1490 : * in our own transaction. However, it can happen in
1491 : * system catalogs, since we tend to release write lock
1492 : * before commit there. Give a warning if neither case
1493 : * applies.
1494 : */
1495 1742907 : xwait = HeapTupleHeaderGetXmin(heapTuple->t_data);
1496 1742907 : if (!TransactionIdIsCurrentTransactionId(xwait))
1497 : {
1498 84 : if (!is_system_catalog)
1499 0 : elog(WARNING, "concurrent insert in progress within table \"%s\"",
1500 : RelationGetRelationName(heapRelation));
1501 :
1502 : /*
1503 : * If we are performing uniqueness checks, indexing
1504 : * such a tuple could lead to a bogus uniqueness
1505 : * failure. In that case we wait for the inserting
1506 : * transaction to finish and check again.
1507 : */
1508 84 : if (checking_uniqueness)
1509 : {
1510 : /*
1511 : * Must drop the lock on the buffer before we wait
1512 : */
1513 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1514 0 : XactLockTableWait(xwait, heapRelation,
1515 0 : &heapTuple->t_self,
1516 : XLTW_InsertIndexUnique);
1517 0 : CHECK_FOR_INTERRUPTS();
1518 0 : goto recheck;
1519 : }
1520 : }
1521 : else
1522 : {
1523 : /*
1524 : * For consistency with
1525 : * heapam_scan_analyze_next_tuple(), count
1526 : * HEAPTUPLE_INSERT_IN_PROGRESS tuples as live only
1527 : * when inserted by our own transaction.
1528 : */
1529 1742823 : reltuples += 1;
1530 : }
1531 :
1532 : /*
1533 : * We must index such tuples, since if the index build
1534 : * commits then they're good.
1535 : */
1536 1742907 : indexIt = true;
1537 1742907 : tupleIsAlive = true;
1538 1742907 : break;
1539 39 : case HEAPTUPLE_DELETE_IN_PROGRESS:
1540 :
1541 : /*
1542 : * As with INSERT_IN_PROGRESS case, this is unexpected
1543 : * unless it's our own deletion or a system catalog; but
1544 : * in anyvisible mode, this tuple is visible.
1545 : */
1546 39 : if (anyvisible)
1547 : {
1548 0 : indexIt = true;
1549 0 : tupleIsAlive = false;
1550 0 : reltuples += 1;
1551 0 : break;
1552 : }
1553 :
1554 39 : xwait = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1555 39 : if (!TransactionIdIsCurrentTransactionId(xwait))
1556 : {
1557 0 : if (!is_system_catalog)
1558 0 : elog(WARNING, "concurrent delete in progress within table \"%s\"",
1559 : RelationGetRelationName(heapRelation));
1560 :
1561 : /*
1562 : * If we are performing uniqueness checks, assuming
1563 : * the tuple is dead could lead to missing a
1564 : * uniqueness violation. In that case we wait for the
1565 : * deleting transaction to finish and check again.
1566 : *
1567 : * Also, if it's a HOT-updated tuple, we should not
1568 : * index it but rather the live tuple at the end of
1569 : * the HOT-chain. However, the deleting transaction
1570 : * could abort, possibly leaving this tuple as live
1571 : * after all, in which case it has to be indexed. The
1572 : * only way to know what to do is to wait for the
1573 : * deleting transaction to finish and check again.
1574 : */
1575 0 : if (checking_uniqueness ||
1576 0 : HeapTupleIsHotUpdated(heapTuple))
1577 : {
1578 : /*
1579 : * Must drop the lock on the buffer before we wait
1580 : */
1581 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1582 0 : XactLockTableWait(xwait, heapRelation,
1583 0 : &heapTuple->t_self,
1584 : XLTW_InsertIndexUnique);
1585 0 : CHECK_FOR_INTERRUPTS();
1586 0 : goto recheck;
1587 : }
1588 :
1589 : /*
1590 : * Otherwise index it but don't check for uniqueness,
1591 : * the same as a RECENTLY_DEAD tuple.
1592 : */
1593 0 : indexIt = true;
1594 :
1595 : /*
1596 : * Count HEAPTUPLE_DELETE_IN_PROGRESS tuples as live,
1597 : * if they were not deleted by the current
1598 : * transaction. That's what
1599 : * heapam_scan_analyze_next_tuple() does, and we want
1600 : * the behavior to be consistent.
1601 : */
1602 0 : reltuples += 1;
1603 : }
1604 39 : else if (HeapTupleIsHotUpdated(heapTuple))
1605 : {
1606 : /*
1607 : * It's a HOT-updated tuple deleted by our own xact.
1608 : * We can assume the deletion will commit (else the
1609 : * index contents don't matter), so treat the same as
1610 : * RECENTLY_DEAD HOT-updated tuples.
1611 : */
1612 0 : indexIt = false;
1613 : /* mark the index as unsafe for old snapshots */
1614 0 : indexInfo->ii_BrokenHotChain = true;
1615 : }
1616 : else
1617 : {
1618 : /*
1619 : * It's a regular tuple deleted by our own xact. Index
1620 : * it, but don't check for uniqueness nor count in
1621 : * reltuples, the same as a RECENTLY_DEAD tuple.
1622 : */
1623 39 : indexIt = true;
1624 : }
1625 : /* In any case, exclude the tuple from unique-checking */
1626 39 : tupleIsAlive = false;
1627 39 : break;
1628 0 : default:
1629 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1630 : indexIt = tupleIsAlive = false; /* keep compiler quiet */
1631 : break;
1632 : }
1633 :
1634 7007764 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1635 :
1636 7007764 : if (!indexIt)
1637 1191 : continue;
1638 : }
1639 : else
1640 : {
1641 : /* heap_getnext did the time qual check */
1642 1841050 : tupleIsAlive = true;
1643 1841050 : reltuples += 1;
1644 : }
1645 :
1646 8847623 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
1647 :
1648 : /* Set up for predicate or expression evaluation */
1649 8847623 : ExecStoreBufferHeapTuple(heapTuple, slot, hscan->rs_cbuf);
1650 :
1651 : /*
1652 : * In a partial index, discard tuples that don't satisfy the
1653 : * predicate.
1654 : */
1655 8847623 : if (predicate != NULL)
1656 : {
1657 102277 : if (!ExecQual(predicate, econtext))
1658 54829 : continue;
1659 : }
1660 :
1661 : /*
1662 : * For the current heap tuple, extract all the attributes we use in
1663 : * this index, and note which are null. This also performs evaluation
1664 : * of any expressions needed.
1665 : */
1666 8792794 : FormIndexDatum(indexInfo,
1667 : slot,
1668 : estate,
1669 : values,
1670 : isnull);
1671 :
1672 : /*
1673 : * You'd think we should go ahead and build the index tuple here, but
1674 : * some index AMs want to do further processing on the data first. So
1675 : * pass the values[] and isnull[] arrays, instead.
1676 : */
1677 :
1678 8792788 : if (HeapTupleIsHeapOnly(heapTuple))
1679 : {
1680 : /*
1681 : * For a heap-only tuple, pretend its TID is that of the root. See
1682 : * src/backend/access/heap/README.HOT for discussion.
1683 : */
1684 : ItemPointerData tid;
1685 : OffsetNumber offnum;
1686 :
1687 3412 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_self);
1688 :
1689 : /*
1690 : * If a HOT tuple points to a root that we don't know about,
1691 : * obtain root items afresh. If that still fails, report it as
1692 : * corruption.
1693 : */
1694 3412 : if (root_offsets[offnum - 1] == InvalidOffsetNumber)
1695 : {
1696 0 : Page page = BufferGetPage(hscan->rs_cbuf);
1697 :
1698 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1699 0 : heap_get_root_tuples(page, root_offsets);
1700 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1701 : }
1702 :
1703 3412 : if (!OffsetNumberIsValid(root_offsets[offnum - 1]))
1704 0 : ereport(ERROR,
1705 : (errcode(ERRCODE_DATA_CORRUPTED),
1706 : errmsg_internal("failed to find parent tuple for heap-only tuple at (%u,%u) in table \"%s\"",
1707 : ItemPointerGetBlockNumber(&heapTuple->t_self),
1708 : offnum,
1709 : RelationGetRelationName(heapRelation))));
1710 :
1711 3412 : ItemPointerSet(&tid, ItemPointerGetBlockNumber(&heapTuple->t_self),
1712 3412 : root_offsets[offnum - 1]);
1713 :
1714 : /* Call the AM's callback routine to process the tuple */
1715 3412 : callback(indexRelation, &tid, values, isnull, tupleIsAlive,
1716 : callback_state);
1717 : }
1718 : else
1719 : {
1720 : /* Call the AM's callback routine to process the tuple */
1721 8789376 : callback(indexRelation, &heapTuple->t_self, values, isnull,
1722 : tupleIsAlive, callback_state);
1723 : }
1724 : }
1725 :
1726 : /* Report scan progress one last time. */
1727 29351 : if (progress)
1728 : {
1729 : BlockNumber blks_done;
1730 :
1731 27699 : if (hscan->rs_base.rs_parallel != NULL)
1732 : {
1733 : ParallelBlockTableScanDesc pbscan;
1734 :
1735 103 : pbscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1736 103 : blks_done = pbscan->phs_nblocks;
1737 : }
1738 : else
1739 27596 : blks_done = hscan->rs_nblocks;
1740 :
1741 27699 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1742 : blks_done);
1743 : }
1744 :
1745 29351 : table_endscan(scan);
1746 :
1747 : /* we can now forget our snapshot, if set and registered by us */
1748 29351 : if (need_unregister_snapshot)
1749 8484 : UnregisterSnapshot(snapshot);
1750 :
1751 29351 : ExecDropSingleTupleTableSlot(slot);
1752 :
1753 29351 : FreeExecutorState(estate);
1754 :
1755 : /* These may have been pointing to the now-gone estate */
1756 29351 : indexInfo->ii_ExpressionsState = NIL;
1757 29351 : indexInfo->ii_PredicateState = NULL;
1758 :
1759 29351 : return reltuples;
1760 : }
1761 :
1762 : static void
1763 369 : heapam_index_validate_scan(Relation heapRelation,
1764 : Relation indexRelation,
1765 : IndexInfo *indexInfo,
1766 : Snapshot snapshot,
1767 : ValidateIndexState *state)
1768 : {
1769 : TableScanDesc scan;
1770 : HeapScanDesc hscan;
1771 : HeapTuple heapTuple;
1772 : Datum values[INDEX_MAX_KEYS];
1773 : bool isnull[INDEX_MAX_KEYS];
1774 : ExprState *predicate;
1775 : TupleTableSlot *slot;
1776 : EState *estate;
1777 : ExprContext *econtext;
1778 369 : BlockNumber root_blkno = InvalidBlockNumber;
1779 : OffsetNumber root_offsets[MaxHeapTuplesPerPage];
1780 : bool in_index[MaxHeapTuplesPerPage];
1781 369 : BlockNumber previous_blkno = InvalidBlockNumber;
1782 :
1783 : /* state variables for the merge */
1784 369 : ItemPointer indexcursor = NULL;
1785 : ItemPointerData decoded;
1786 369 : bool tuplesort_empty = false;
1787 :
1788 : /*
1789 : * sanity checks
1790 : */
1791 : Assert(OidIsValid(indexRelation->rd_rel->relam));
1792 :
1793 : /*
1794 : * Need an EState for evaluation of index expressions and partial-index
1795 : * predicates. Also a slot to hold the current tuple.
1796 : */
1797 369 : estate = CreateExecutorState();
1798 369 : econtext = GetPerTupleExprContext(estate);
1799 369 : slot = MakeSingleTupleTableSlot(RelationGetDescr(heapRelation),
1800 : &TTSOpsHeapTuple);
1801 :
1802 : /* Arrange for econtext's scan tuple to be the tuple under test */
1803 369 : econtext->ecxt_scantuple = slot;
1804 :
1805 : /* Set up execution state for predicate, if any. */
1806 369 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
1807 :
1808 : /*
1809 : * Prepare for scan of the base relation. We need just those tuples
1810 : * satisfying the passed-in reference snapshot. We must disable syncscan
1811 : * here, because it's critical that we read from block zero forward to
1812 : * match the sorted TIDs.
1813 : */
1814 369 : scan = table_beginscan_strat(heapRelation, /* relation */
1815 : snapshot, /* snapshot */
1816 : 0, /* number of keys */
1817 : NULL, /* scan key */
1818 : true, /* buffer access strategy OK */
1819 : false); /* syncscan not OK */
1820 369 : hscan = (HeapScanDesc) scan;
1821 :
1822 369 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1823 369 : hscan->rs_nblocks);
1824 :
1825 : /*
1826 : * Scan all tuples matching the snapshot.
1827 : */
1828 124324 : while ((heapTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1829 : {
1830 123955 : ItemPointer heapcursor = &heapTuple->t_self;
1831 : ItemPointerData rootTuple;
1832 : OffsetNumber root_offnum;
1833 :
1834 123955 : CHECK_FOR_INTERRUPTS();
1835 :
1836 123955 : state->htups += 1;
1837 :
1838 123955 : if ((previous_blkno == InvalidBlockNumber) ||
1839 123735 : (hscan->rs_cblock != previous_blkno))
1840 : {
1841 2209 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1842 2209 : hscan->rs_cblock);
1843 2209 : previous_blkno = hscan->rs_cblock;
1844 : }
1845 :
1846 : /*
1847 : * As commented in table_index_build_scan, we should index heap-only
1848 : * tuples under the TIDs of their root tuples; so when we advance onto
1849 : * a new heap page, build a map of root item offsets on the page.
1850 : *
1851 : * This complicates merging against the tuplesort output: we will
1852 : * visit the live tuples in order by their offsets, but the root
1853 : * offsets that we need to compare against the index contents might be
1854 : * ordered differently. So we might have to "look back" within the
1855 : * tuplesort output, but only within the current page. We handle that
1856 : * by keeping a bool array in_index[] showing all the
1857 : * already-passed-over tuplesort output TIDs of the current page. We
1858 : * clear that array here, when advancing onto a new heap page.
1859 : */
1860 123955 : if (hscan->rs_cblock != root_blkno)
1861 : {
1862 2209 : Page page = BufferGetPage(hscan->rs_cbuf);
1863 :
1864 2209 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1865 2209 : heap_get_root_tuples(page, root_offsets);
1866 2209 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1867 :
1868 2209 : memset(in_index, 0, sizeof(in_index));
1869 :
1870 2209 : root_blkno = hscan->rs_cblock;
1871 : }
1872 :
1873 : /* Convert actual tuple TID to root TID */
1874 123955 : rootTuple = *heapcursor;
1875 123955 : root_offnum = ItemPointerGetOffsetNumber(heapcursor);
1876 :
1877 123955 : if (HeapTupleIsHeapOnly(heapTuple))
1878 : {
1879 12 : root_offnum = root_offsets[root_offnum - 1];
1880 12 : if (!OffsetNumberIsValid(root_offnum))
1881 0 : ereport(ERROR,
1882 : (errcode(ERRCODE_DATA_CORRUPTED),
1883 : errmsg_internal("failed to find parent tuple for heap-only tuple at (%u,%u) in table \"%s\"",
1884 : ItemPointerGetBlockNumber(heapcursor),
1885 : ItemPointerGetOffsetNumber(heapcursor),
1886 : RelationGetRelationName(heapRelation))));
1887 12 : ItemPointerSetOffsetNumber(&rootTuple, root_offnum);
1888 : }
1889 :
1890 : /*
1891 : * "merge" by skipping through the index tuples until we find or pass
1892 : * the current root tuple.
1893 : */
1894 281017 : while (!tuplesort_empty &&
1895 280767 : (!indexcursor ||
1896 280767 : ItemPointerCompare(indexcursor, &rootTuple) < 0))
1897 : {
1898 : Datum ts_val;
1899 : bool ts_isnull;
1900 :
1901 157062 : if (indexcursor)
1902 : {
1903 : /*
1904 : * Remember index items seen earlier on the current heap page
1905 : */
1906 156842 : if (ItemPointerGetBlockNumber(indexcursor) == root_blkno)
1907 154391 : in_index[ItemPointerGetOffsetNumber(indexcursor) - 1] = true;
1908 : }
1909 :
1910 157062 : tuplesort_empty = !tuplesort_getdatum(state->tuplesort, true,
1911 : false, &ts_val, &ts_isnull,
1912 157062 : NULL);
1913 : Assert(tuplesort_empty || !ts_isnull);
1914 157062 : if (!tuplesort_empty)
1915 : {
1916 157041 : itemptr_decode(&decoded, DatumGetInt64(ts_val));
1917 157041 : indexcursor = &decoded;
1918 : }
1919 : else
1920 : {
1921 : /* Be tidy */
1922 21 : indexcursor = NULL;
1923 : }
1924 : }
1925 :
1926 : /*
1927 : * If the tuplesort has overshot *and* we didn't see a match earlier,
1928 : * then this tuple is missing from the index, so insert it.
1929 : */
1930 247880 : if ((tuplesort_empty ||
1931 123925 : ItemPointerCompare(indexcursor, &rootTuple) > 0) &&
1932 65 : !in_index[root_offnum - 1])
1933 : {
1934 56 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
1935 :
1936 : /* Set up for predicate or expression evaluation */
1937 56 : ExecStoreHeapTuple(heapTuple, slot, false);
1938 :
1939 : /*
1940 : * In a partial index, discard tuples that don't satisfy the
1941 : * predicate.
1942 : */
1943 56 : if (predicate != NULL)
1944 : {
1945 24 : if (!ExecQual(predicate, econtext))
1946 24 : continue;
1947 : }
1948 :
1949 : /*
1950 : * For the current heap tuple, extract all the attributes we use
1951 : * in this index, and note which are null. This also performs
1952 : * evaluation of any expressions needed.
1953 : */
1954 32 : FormIndexDatum(indexInfo,
1955 : slot,
1956 : estate,
1957 : values,
1958 : isnull);
1959 :
1960 : /*
1961 : * You'd think we should go ahead and build the index tuple here,
1962 : * but some index AMs want to do further processing on the data
1963 : * first. So pass the values[] and isnull[] arrays, instead.
1964 : */
1965 :
1966 : /*
1967 : * If the tuple is already committed dead, you might think we
1968 : * could suppress uniqueness checking, but this is no longer true
1969 : * in the presence of HOT, because the insert is actually a proxy
1970 : * for a uniqueness check on the whole HOT-chain. That is, the
1971 : * tuple we have here could be dead because it was already
1972 : * HOT-updated, and if so the updating transaction will not have
1973 : * thought it should insert index entries. The index AM will
1974 : * check the whole HOT-chain and correctly detect a conflict if
1975 : * there is one.
1976 : */
1977 :
1978 32 : index_insert(indexRelation,
1979 : values,
1980 : isnull,
1981 : &rootTuple,
1982 : heapRelation,
1983 32 : indexInfo->ii_Unique ?
1984 : UNIQUE_CHECK_YES : UNIQUE_CHECK_NO,
1985 : false,
1986 : indexInfo);
1987 :
1988 32 : state->tups_inserted += 1;
1989 : }
1990 : }
1991 :
1992 369 : table_endscan(scan);
1993 :
1994 369 : ExecDropSingleTupleTableSlot(slot);
1995 :
1996 369 : FreeExecutorState(estate);
1997 :
1998 : /* These may have been pointing to the now-gone estate */
1999 369 : indexInfo->ii_ExpressionsState = NIL;
2000 369 : indexInfo->ii_PredicateState = NULL;
2001 369 : }
2002 :
2003 : /*
2004 : * Return the number of blocks that have been read by this scan since
2005 : * starting. This is meant for progress reporting rather than be fully
2006 : * accurate: in a parallel scan, workers can be concurrently reading blocks
2007 : * further ahead than what we report.
2008 : */
2009 : static BlockNumber
2010 7418021 : heapam_scan_get_blocks_done(HeapScanDesc hscan)
2011 : {
2012 7418021 : ParallelBlockTableScanDesc bpscan = NULL;
2013 : BlockNumber startblock;
2014 : BlockNumber blocks_done;
2015 :
2016 7418021 : if (hscan->rs_base.rs_parallel != NULL)
2017 : {
2018 1001318 : bpscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
2019 1001318 : startblock = bpscan->phs_startblock;
2020 : }
2021 : else
2022 6416703 : startblock = hscan->rs_startblock;
2023 :
2024 : /*
2025 : * Might have wrapped around the end of the relation, if startblock was
2026 : * not zero.
2027 : */
2028 7418021 : if (hscan->rs_cblock > startblock)
2029 7127642 : blocks_done = hscan->rs_cblock - startblock;
2030 : else
2031 : {
2032 : BlockNumber nblocks;
2033 :
2034 290379 : nblocks = bpscan != NULL ? bpscan->phs_nblocks : hscan->rs_nblocks;
2035 290379 : blocks_done = nblocks - startblock +
2036 290379 : hscan->rs_cblock;
2037 : }
2038 :
2039 7418021 : return blocks_done;
2040 : }
2041 :
2042 :
2043 : /* ------------------------------------------------------------------------
2044 : * Miscellaneous callbacks for the heap AM
2045 : * ------------------------------------------------------------------------
2046 : */
2047 :
2048 : /*
2049 : * Check to see whether the table needs a TOAST table. It does only if
2050 : * (1) there are any toastable attributes, and (2) the maximum length
2051 : * of a tuple could exceed TOAST_TUPLE_THRESHOLD. (We don't want to
2052 : * create a toast table for something like "f1 varchar(20)".)
2053 : */
2054 : static bool
2055 23379 : heapam_relation_needs_toast_table(Relation rel)
2056 : {
2057 23379 : int32 data_length = 0;
2058 23379 : bool maxlength_unknown = false;
2059 23379 : bool has_toastable_attrs = false;
2060 23379 : TupleDesc tupdesc = rel->rd_att;
2061 : int32 tuple_length;
2062 : int i;
2063 :
2064 93526 : for (i = 0; i < tupdesc->natts; i++)
2065 : {
2066 70147 : Form_pg_attribute att = TupleDescAttr(tupdesc, i);
2067 :
2068 70147 : if (att->attisdropped)
2069 567 : continue;
2070 69580 : if (att->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
2071 478 : continue;
2072 69102 : data_length = att_align_nominal(data_length, att->attalign);
2073 69102 : if (att->attlen > 0)
2074 : {
2075 : /* Fixed-length types are never toastable */
2076 51609 : data_length += att->attlen;
2077 : }
2078 : else
2079 : {
2080 17493 : int32 maxlen = type_maximum_size(att->atttypid,
2081 : att->atttypmod);
2082 :
2083 17493 : if (maxlen < 0)
2084 15909 : maxlength_unknown = true;
2085 : else
2086 1584 : data_length += maxlen;
2087 17493 : if (att->attstorage != TYPSTORAGE_PLAIN)
2088 16873 : has_toastable_attrs = true;
2089 : }
2090 : }
2091 23379 : if (!has_toastable_attrs)
2092 13440 : return false; /* nothing to toast? */
2093 9939 : if (maxlength_unknown)
2094 8765 : return true; /* any unlimited-length attrs? */
2095 1174 : tuple_length = MAXALIGN(SizeofHeapTupleHeader +
2096 1174 : BITMAPLEN(tupdesc->natts)) +
2097 1174 : MAXALIGN(data_length);
2098 1174 : return (tuple_length > TOAST_TUPLE_THRESHOLD);
2099 : }
2100 :
2101 : /*
2102 : * TOAST tables for heap relations are just heap relations.
2103 : */
2104 : static Oid
2105 9045 : heapam_relation_toast_am(Relation rel)
2106 : {
2107 9045 : return rel->rd_rel->relam;
2108 : }
2109 :
2110 :
2111 : /* ------------------------------------------------------------------------
2112 : * Planner related callbacks for the heap AM
2113 : * ------------------------------------------------------------------------
2114 : */
2115 :
2116 : #define HEAP_OVERHEAD_BYTES_PER_TUPLE \
2117 : (MAXALIGN(SizeofHeapTupleHeader) + sizeof(ItemIdData))
2118 : #define HEAP_USABLE_BYTES_PER_PAGE \
2119 : (BLCKSZ - SizeOfPageHeaderData)
2120 :
2121 : static void
2122 251523 : heapam_estimate_rel_size(Relation rel, int32 *attr_widths,
2123 : BlockNumber *pages, double *tuples,
2124 : double *allvisfrac)
2125 : {
2126 251523 : table_block_relation_estimate_size(rel, attr_widths, pages,
2127 : tuples, allvisfrac,
2128 : HEAP_OVERHEAD_BYTES_PER_TUPLE,
2129 : HEAP_USABLE_BYTES_PER_PAGE);
2130 251523 : }
2131 :
2132 :
2133 : /* ------------------------------------------------------------------------
2134 : * Executor related callbacks for the heap AM
2135 : * ------------------------------------------------------------------------
2136 : */
2137 :
2138 : static bool
2139 3469162 : heapam_scan_bitmap_next_tuple(TableScanDesc scan,
2140 : TupleTableSlot *slot,
2141 : bool *recheck,
2142 : uint64 *lossy_pages,
2143 : uint64 *exact_pages)
2144 : {
2145 3469162 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
2146 3469162 : HeapScanDesc hscan = (HeapScanDesc) bscan;
2147 : OffsetNumber targoffset;
2148 : Page page;
2149 : ItemId lp;
2150 :
2151 : /*
2152 : * Out of range? If so, nothing more to look at on this page
2153 : */
2154 3673723 : while (hscan->rs_cindex >= hscan->rs_ntuples)
2155 : {
2156 : /*
2157 : * Returns false if the bitmap is exhausted and there are no further
2158 : * blocks we need to scan.
2159 : */
2160 217833 : if (!BitmapHeapScanNextBlock(scan, recheck, lossy_pages, exact_pages))
2161 13269 : return false;
2162 : }
2163 :
2164 3455890 : targoffset = hscan->rs_vistuples[hscan->rs_cindex];
2165 3455890 : page = BufferGetPage(hscan->rs_cbuf);
2166 3455890 : lp = PageGetItemId(page, targoffset);
2167 : Assert(ItemIdIsNormal(lp));
2168 :
2169 3455890 : hscan->rs_ctup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2170 3455890 : hscan->rs_ctup.t_len = ItemIdGetLength(lp);
2171 3455890 : hscan->rs_ctup.t_tableOid = scan->rs_rd->rd_id;
2172 3455890 : ItemPointerSet(&hscan->rs_ctup.t_self, hscan->rs_cblock, targoffset);
2173 :
2174 3455890 : pgstat_count_heap_fetch(scan->rs_rd);
2175 :
2176 : /*
2177 : * Set up the result slot to point to this tuple. Note that the slot
2178 : * acquires a pin on the buffer.
2179 : */
2180 3455890 : ExecStoreBufferHeapTuple(&hscan->rs_ctup,
2181 : slot,
2182 : hscan->rs_cbuf);
2183 :
2184 3455890 : hscan->rs_cindex++;
2185 :
2186 3455890 : return true;
2187 : }
2188 :
2189 : static bool
2190 6456 : heapam_scan_sample_next_block(TableScanDesc scan, SampleScanState *scanstate)
2191 : {
2192 6456 : HeapScanDesc hscan = (HeapScanDesc) scan;
2193 6456 : TsmRoutine *tsm = scanstate->tsmroutine;
2194 : BlockNumber blockno;
2195 :
2196 : /* return false immediately if relation is empty */
2197 6456 : if (hscan->rs_nblocks == 0)
2198 0 : return false;
2199 :
2200 : /* release previous scan buffer, if any */
2201 6456 : if (BufferIsValid(hscan->rs_cbuf))
2202 : {
2203 6368 : ReleaseBuffer(hscan->rs_cbuf);
2204 6368 : hscan->rs_cbuf = InvalidBuffer;
2205 : }
2206 :
2207 6456 : if (tsm->NextSampleBlock)
2208 2223 : blockno = tsm->NextSampleBlock(scanstate, hscan->rs_nblocks);
2209 : else
2210 : {
2211 : /* scanning table sequentially */
2212 :
2213 4233 : if (hscan->rs_cblock == InvalidBlockNumber)
2214 : {
2215 : Assert(!hscan->rs_inited);
2216 39 : blockno = hscan->rs_startblock;
2217 : }
2218 : else
2219 : {
2220 : Assert(hscan->rs_inited);
2221 :
2222 4194 : blockno = hscan->rs_cblock + 1;
2223 :
2224 4194 : if (blockno >= hscan->rs_nblocks)
2225 : {
2226 : /* wrap to beginning of rel, might not have started at 0 */
2227 39 : blockno = 0;
2228 : }
2229 :
2230 : /*
2231 : * Report our new scan position for synchronization purposes.
2232 : *
2233 : * Note: we do this before checking for end of scan so that the
2234 : * final state of the position hint is back at the start of the
2235 : * rel. That's not strictly necessary, but otherwise when you run
2236 : * the same query multiple times the starting position would shift
2237 : * a little bit backwards on every invocation, which is confusing.
2238 : * We don't guarantee any specific ordering in general, though.
2239 : */
2240 4194 : if (scan->rs_flags & SO_ALLOW_SYNC)
2241 0 : ss_report_location(scan->rs_rd, blockno);
2242 :
2243 4194 : if (blockno == hscan->rs_startblock)
2244 : {
2245 39 : blockno = InvalidBlockNumber;
2246 : }
2247 : }
2248 : }
2249 :
2250 6456 : hscan->rs_cblock = blockno;
2251 :
2252 6456 : if (!BlockNumberIsValid(blockno))
2253 : {
2254 85 : hscan->rs_inited = false;
2255 85 : return false;
2256 : }
2257 :
2258 : Assert(hscan->rs_cblock < hscan->rs_nblocks);
2259 :
2260 : /*
2261 : * Be sure to check for interrupts at least once per page. Checks at
2262 : * higher code levels won't be able to stop a sample scan that encounters
2263 : * many pages' worth of consecutive dead tuples.
2264 : */
2265 6371 : CHECK_FOR_INTERRUPTS();
2266 :
2267 : /* Read page using selected strategy */
2268 6371 : hscan->rs_cbuf = ReadBufferExtended(hscan->rs_base.rs_rd, MAIN_FORKNUM,
2269 : blockno, RBM_NORMAL, hscan->rs_strategy);
2270 :
2271 : /* in pagemode, prune the page and determine visible tuple offsets */
2272 6371 : if (hscan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
2273 4277 : heap_prepare_pagescan(scan);
2274 :
2275 6371 : hscan->rs_inited = true;
2276 6371 : return true;
2277 : }
2278 :
2279 : static bool
2280 126948 : heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
2281 : TupleTableSlot *slot)
2282 : {
2283 126948 : HeapScanDesc hscan = (HeapScanDesc) scan;
2284 126948 : TsmRoutine *tsm = scanstate->tsmroutine;
2285 126948 : BlockNumber blockno = hscan->rs_cblock;
2286 126948 : bool pagemode = (scan->rs_flags & SO_ALLOW_PAGEMODE) != 0;
2287 :
2288 : Page page;
2289 : bool all_visible;
2290 : OffsetNumber maxoffset;
2291 :
2292 : /*
2293 : * When not using pagemode, we must lock the buffer during tuple
2294 : * visibility checks.
2295 : */
2296 126948 : if (!pagemode)
2297 2097 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
2298 :
2299 126948 : page = BufferGetPage(hscan->rs_cbuf);
2300 253344 : all_visible = PageIsAllVisible(page) &&
2301 126396 : !scan->rs_snapshot->takenDuringRecovery;
2302 126948 : maxoffset = PageGetMaxOffsetNumber(page);
2303 :
2304 : for (;;)
2305 0 : {
2306 : OffsetNumber tupoffset;
2307 :
2308 126948 : CHECK_FOR_INTERRUPTS();
2309 :
2310 : /* Ask the tablesample method which tuples to check on this page. */
2311 126948 : tupoffset = tsm->NextSampleTuple(scanstate,
2312 : blockno,
2313 : maxoffset);
2314 :
2315 126948 : if (OffsetNumberIsValid(tupoffset))
2316 : {
2317 : ItemId itemid;
2318 : bool visible;
2319 120580 : HeapTuple tuple = &(hscan->rs_ctup);
2320 :
2321 : /* Skip invalid tuple pointers. */
2322 120580 : itemid = PageGetItemId(page, tupoffset);
2323 120580 : if (!ItemIdIsNormal(itemid))
2324 0 : continue;
2325 :
2326 120580 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, itemid);
2327 120580 : tuple->t_len = ItemIdGetLength(itemid);
2328 120580 : ItemPointerSet(&(tuple->t_self), blockno, tupoffset);
2329 :
2330 :
2331 120580 : if (all_visible)
2332 120174 : visible = true;
2333 : else
2334 406 : visible = SampleHeapTupleVisible(scan, hscan->rs_cbuf,
2335 : tuple, tupoffset);
2336 :
2337 : /* in pagemode, heap_prepare_pagescan did this for us */
2338 120580 : if (!pagemode)
2339 3 : HeapCheckForSerializableConflictOut(visible, scan->rs_rd, tuple,
2340 : hscan->rs_cbuf, scan->rs_snapshot);
2341 :
2342 : /* Try next tuple from same page. */
2343 120580 : if (!visible)
2344 0 : continue;
2345 :
2346 : /* Found visible tuple, return it. */
2347 120580 : if (!pagemode)
2348 3 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
2349 :
2350 120580 : ExecStoreBufferHeapTuple(tuple, slot, hscan->rs_cbuf);
2351 :
2352 : /* Count successfully-fetched tuples as heap fetches */
2353 120580 : pgstat_count_heap_getnext(scan->rs_rd);
2354 :
2355 120580 : return true;
2356 : }
2357 : else
2358 : {
2359 : /*
2360 : * If we get here, it means we've exhausted the items on this page
2361 : * and it's time to move to the next.
2362 : */
2363 6368 : if (!pagemode)
2364 2094 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
2365 :
2366 6368 : ExecClearTuple(slot);
2367 6368 : return false;
2368 : }
2369 : }
2370 :
2371 : Assert(0);
2372 : }
2373 :
2374 :
2375 : /* ----------------------------------------------------------------------------
2376 : * Helper functions for the above.
2377 : * ----------------------------------------------------------------------------
2378 : */
2379 :
2380 : /*
2381 : * Reconstruct and rewrite the given tuple
2382 : *
2383 : * We cannot simply copy the tuple as-is, for several reasons:
2384 : *
2385 : * 1. We'd like to squeeze out the values of any dropped columns, both
2386 : * to save space and to ensure we have no corner-case failures. (It's
2387 : * possible for example that the new table hasn't got a TOAST table
2388 : * and so is unable to store any large values of dropped cols.)
2389 : *
2390 : * 2. The tuple might not even be legal for the new table; this is
2391 : * currently only known to happen as an after-effect of ALTER TABLE
2392 : * SET WITHOUT OIDS.
2393 : *
2394 : * So, we must reconstruct the tuple from component Datums.
2395 : */
2396 : static void
2397 354659 : reform_and_rewrite_tuple(HeapTuple tuple,
2398 : Relation OldHeap, Relation NewHeap,
2399 : Datum *values, bool *isnull, RewriteState rwstate)
2400 : {
2401 354659 : TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
2402 354659 : TupleDesc newTupDesc = RelationGetDescr(NewHeap);
2403 : HeapTuple copiedTuple;
2404 : int i;
2405 :
2406 354659 : heap_deform_tuple(tuple, oldTupDesc, values, isnull);
2407 :
2408 : /* Be sure to null out any dropped columns */
2409 3174416 : for (i = 0; i < newTupDesc->natts; i++)
2410 : {
2411 2819757 : if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
2412 0 : isnull[i] = true;
2413 : }
2414 :
2415 354659 : copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
2416 :
2417 : /* The heap rewrite module does the rest */
2418 354659 : rewrite_heap_tuple(rwstate, tuple, copiedTuple);
2419 :
2420 354659 : heap_freetuple(copiedTuple);
2421 354659 : }
2422 :
2423 : /*
2424 : * Check visibility of the tuple.
2425 : */
2426 : static bool
2427 406 : SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
2428 : HeapTuple tuple,
2429 : OffsetNumber tupoffset)
2430 : {
2431 406 : HeapScanDesc hscan = (HeapScanDesc) scan;
2432 :
2433 406 : if (scan->rs_flags & SO_ALLOW_PAGEMODE)
2434 : {
2435 403 : uint32 start = 0,
2436 403 : end = hscan->rs_ntuples;
2437 :
2438 : /*
2439 : * In pageatatime mode, heap_prepare_pagescan() already did visibility
2440 : * checks, so just look at the info it left in rs_vistuples[].
2441 : *
2442 : * We use a binary search over the known-sorted array. Note: we could
2443 : * save some effort if we insisted that NextSampleTuple select tuples
2444 : * in increasing order, but it's not clear that there would be enough
2445 : * gain to justify the restriction.
2446 : */
2447 775 : while (start < end)
2448 : {
2449 775 : uint32 mid = start + (end - start) / 2;
2450 775 : OffsetNumber curoffset = hscan->rs_vistuples[mid];
2451 :
2452 775 : if (tupoffset == curoffset)
2453 403 : return true;
2454 372 : else if (tupoffset < curoffset)
2455 220 : end = mid;
2456 : else
2457 152 : start = mid + 1;
2458 : }
2459 :
2460 0 : return false;
2461 : }
2462 : else
2463 : {
2464 : /* Otherwise, we have to check the tuple individually. */
2465 3 : return HeapTupleSatisfiesVisibility(tuple, scan->rs_snapshot,
2466 : buffer);
2467 : }
2468 : }
2469 :
2470 : /*
2471 : * Helper function get the next block of a bitmap heap scan. Returns true when
2472 : * it got the next block and saved it in the scan descriptor and false when
2473 : * the bitmap and or relation are exhausted.
2474 : */
2475 : static bool
2476 217833 : BitmapHeapScanNextBlock(TableScanDesc scan,
2477 : bool *recheck,
2478 : uint64 *lossy_pages, uint64 *exact_pages)
2479 : {
2480 217833 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
2481 217833 : HeapScanDesc hscan = (HeapScanDesc) bscan;
2482 : BlockNumber block;
2483 : void *per_buffer_data;
2484 : Buffer buffer;
2485 : Snapshot snapshot;
2486 : int ntup;
2487 : TBMIterateResult *tbmres;
2488 : OffsetNumber offsets[TBM_MAX_TUPLES_PER_PAGE];
2489 217833 : int noffsets = -1;
2490 :
2491 : Assert(scan->rs_flags & SO_TYPE_BITMAPSCAN);
2492 : Assert(hscan->rs_read_stream);
2493 :
2494 217833 : hscan->rs_cindex = 0;
2495 217833 : hscan->rs_ntuples = 0;
2496 :
2497 : /* Release buffer containing previous block. */
2498 217833 : if (BufferIsValid(hscan->rs_cbuf))
2499 : {
2500 204367 : ReleaseBuffer(hscan->rs_cbuf);
2501 204367 : hscan->rs_cbuf = InvalidBuffer;
2502 : }
2503 :
2504 217833 : hscan->rs_cbuf = read_stream_next_buffer(hscan->rs_read_stream,
2505 : &per_buffer_data);
2506 :
2507 217833 : if (BufferIsInvalid(hscan->rs_cbuf))
2508 : {
2509 : /* the bitmap is exhausted */
2510 13269 : return false;
2511 : }
2512 :
2513 : Assert(per_buffer_data);
2514 :
2515 204564 : tbmres = per_buffer_data;
2516 :
2517 : Assert(BlockNumberIsValid(tbmres->blockno));
2518 : Assert(BufferGetBlockNumber(hscan->rs_cbuf) == tbmres->blockno);
2519 :
2520 : /* Exact pages need their tuple offsets extracted. */
2521 204564 : if (!tbmres->lossy)
2522 124364 : noffsets = tbm_extract_page_tuple(tbmres, offsets,
2523 : TBM_MAX_TUPLES_PER_PAGE);
2524 :
2525 204564 : *recheck = tbmres->recheck;
2526 :
2527 204564 : block = hscan->rs_cblock = tbmres->blockno;
2528 204564 : buffer = hscan->rs_cbuf;
2529 204564 : snapshot = scan->rs_snapshot;
2530 :
2531 204564 : ntup = 0;
2532 :
2533 : /*
2534 : * Prune and repair fragmentation for the whole page, if possible.
2535 : */
2536 204564 : heap_page_prune_opt(scan->rs_rd, buffer);
2537 :
2538 : /*
2539 : * We must hold share lock on the buffer content while examining tuple
2540 : * visibility. Afterwards, however, the tuples we have found to be
2541 : * visible are guaranteed good as long as we hold the buffer pin.
2542 : */
2543 204564 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
2544 :
2545 : /*
2546 : * We need two separate strategies for lossy and non-lossy cases.
2547 : */
2548 204564 : if (!tbmres->lossy)
2549 : {
2550 : /*
2551 : * Bitmap is non-lossy, so we just look through the offsets listed in
2552 : * tbmres; but we have to follow any HOT chain starting at each such
2553 : * offset.
2554 : */
2555 : int curslot;
2556 :
2557 : /* We must have extracted the tuple offsets by now */
2558 : Assert(noffsets > -1);
2559 :
2560 3095738 : for (curslot = 0; curslot < noffsets; curslot++)
2561 : {
2562 2971377 : OffsetNumber offnum = offsets[curslot];
2563 : ItemPointerData tid;
2564 : HeapTupleData heapTuple;
2565 :
2566 2971377 : ItemPointerSet(&tid, block, offnum);
2567 2971377 : if (heap_hot_search_buffer(&tid, scan->rs_rd, buffer, snapshot,
2568 : &heapTuple, NULL, true))
2569 2845426 : hscan->rs_vistuples[ntup++] = ItemPointerGetOffsetNumber(&tid);
2570 : }
2571 : }
2572 : else
2573 : {
2574 : /*
2575 : * Bitmap is lossy, so we must examine each line pointer on the page.
2576 : * But we can ignore HOT chains, since we'll check each tuple anyway.
2577 : */
2578 80200 : Page page = BufferGetPage(buffer);
2579 80200 : OffsetNumber maxoff = PageGetMaxOffsetNumber(page);
2580 : OffsetNumber offnum;
2581 :
2582 692425 : for (offnum = FirstOffsetNumber; offnum <= maxoff; offnum = OffsetNumberNext(offnum))
2583 : {
2584 : ItemId lp;
2585 : HeapTupleData loctup;
2586 : bool valid;
2587 :
2588 612225 : lp = PageGetItemId(page, offnum);
2589 612225 : if (!ItemIdIsNormal(lp))
2590 0 : continue;
2591 612225 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2592 612225 : loctup.t_len = ItemIdGetLength(lp);
2593 612225 : loctup.t_tableOid = scan->rs_rd->rd_id;
2594 612225 : ItemPointerSet(&loctup.t_self, block, offnum);
2595 612225 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
2596 612225 : if (valid)
2597 : {
2598 612162 : hscan->rs_vistuples[ntup++] = offnum;
2599 612162 : PredicateLockTID(scan->rs_rd, &loctup.t_self, snapshot,
2600 612162 : HeapTupleHeaderGetXmin(loctup.t_data));
2601 : }
2602 612225 : HeapCheckForSerializableConflictOut(valid, scan->rs_rd, &loctup,
2603 : buffer, snapshot);
2604 : }
2605 : }
2606 :
2607 204561 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2608 :
2609 : Assert(ntup <= MaxHeapTuplesPerPage);
2610 204561 : hscan->rs_ntuples = ntup;
2611 :
2612 204561 : if (tbmres->lossy)
2613 80200 : (*lossy_pages)++;
2614 : else
2615 124361 : (*exact_pages)++;
2616 :
2617 : /*
2618 : * Return true to indicate that a valid block was found and the bitmap is
2619 : * not exhausted. If there are no visible tuples on this page,
2620 : * hscan->rs_ntuples will be 0 and heapam_scan_bitmap_next_tuple() will
2621 : * return false returning control to this function to advance to the next
2622 : * block in the bitmap.
2623 : */
2624 204561 : return true;
2625 : }
2626 :
2627 : /* ------------------------------------------------------------------------
2628 : * Definition of the heap table access method.
2629 : * ------------------------------------------------------------------------
2630 : */
2631 :
2632 : static const TableAmRoutine heapam_methods = {
2633 : .type = T_TableAmRoutine,
2634 :
2635 : .slot_callbacks = heapam_slot_callbacks,
2636 :
2637 : .scan_begin = heap_beginscan,
2638 : .scan_end = heap_endscan,
2639 : .scan_rescan = heap_rescan,
2640 : .scan_getnextslot = heap_getnextslot,
2641 :
2642 : .scan_set_tidrange = heap_set_tidrange,
2643 : .scan_getnextslot_tidrange = heap_getnextslot_tidrange,
2644 :
2645 : .parallelscan_estimate = table_block_parallelscan_estimate,
2646 : .parallelscan_initialize = table_block_parallelscan_initialize,
2647 : .parallelscan_reinitialize = table_block_parallelscan_reinitialize,
2648 :
2649 : .index_fetch_begin = heapam_index_fetch_begin,
2650 : .index_fetch_reset = heapam_index_fetch_reset,
2651 : .index_fetch_end = heapam_index_fetch_end,
2652 : .index_fetch_tuple = heapam_index_fetch_tuple,
2653 :
2654 : .tuple_insert = heapam_tuple_insert,
2655 : .tuple_insert_speculative = heapam_tuple_insert_speculative,
2656 : .tuple_complete_speculative = heapam_tuple_complete_speculative,
2657 : .multi_insert = heap_multi_insert,
2658 : .tuple_delete = heapam_tuple_delete,
2659 : .tuple_update = heapam_tuple_update,
2660 : .tuple_lock = heapam_tuple_lock,
2661 :
2662 : .tuple_fetch_row_version = heapam_fetch_row_version,
2663 : .tuple_get_latest_tid = heap_get_latest_tid,
2664 : .tuple_tid_valid = heapam_tuple_tid_valid,
2665 : .tuple_satisfies_snapshot = heapam_tuple_satisfies_snapshot,
2666 : .index_delete_tuples = heap_index_delete_tuples,
2667 :
2668 : .relation_set_new_filelocator = heapam_relation_set_new_filelocator,
2669 : .relation_nontransactional_truncate = heapam_relation_nontransactional_truncate,
2670 : .relation_copy_data = heapam_relation_copy_data,
2671 : .relation_copy_for_cluster = heapam_relation_copy_for_cluster,
2672 : .relation_vacuum = heap_vacuum_rel,
2673 : .scan_analyze_next_block = heapam_scan_analyze_next_block,
2674 : .scan_analyze_next_tuple = heapam_scan_analyze_next_tuple,
2675 : .index_build_range_scan = heapam_index_build_range_scan,
2676 : .index_validate_scan = heapam_index_validate_scan,
2677 :
2678 : .relation_size = table_block_relation_size,
2679 : .relation_needs_toast_table = heapam_relation_needs_toast_table,
2680 : .relation_toast_am = heapam_relation_toast_am,
2681 : .relation_fetch_toast_slice = heap_fetch_toast_slice,
2682 :
2683 : .relation_estimate_size = heapam_estimate_rel_size,
2684 :
2685 : .scan_bitmap_next_tuple = heapam_scan_bitmap_next_tuple,
2686 : .scan_sample_next_block = heapam_scan_sample_next_block,
2687 : .scan_sample_next_tuple = heapam_scan_sample_next_tuple
2688 : };
2689 :
2690 :
2691 : const TableAmRoutine *
2692 9967216 : GetHeapamTableAmRoutine(void)
2693 : {
2694 9967216 : return &heapam_methods;
2695 : }
2696 :
2697 : Datum
2698 1274900 : heap_tableam_handler(PG_FUNCTION_ARGS)
2699 : {
2700 1274900 : PG_RETURN_POINTER(&heapam_methods);
2701 : }
|
