Age Owner Branch data TLA 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 file 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/lock.h"
44 : : #include "storage/predicate.h"
45 : : #include "storage/procarray.h"
46 : : #include "storage/smgr.h"
47 : : #include "utils/builtins.h"
48 : : #include "utils/rel.h"
49 : : #include "utils/tuplesort.h"
50 : :
51 : : static void reform_and_rewrite_tuple(HeapTuple tuple,
52 : : Relation OldHeap, Relation NewHeap,
53 : : Datum *values, bool *isnull, RewriteState rwstate);
54 : : static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
55 : : Relation NewHeap, Datum *values, bool *isnull,
56 : : BulkInsertState bistate);
57 : : static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
58 : : Relation NewHeap, Datum *values, bool *isnull);
59 : :
60 : : static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
61 : : HeapTuple tuple,
62 : : OffsetNumber tupoffset);
63 : :
64 : : static BlockNumber heapam_scan_get_blocks_done(HeapScanDesc hscan);
65 : :
66 : : static bool BitmapHeapScanNextBlock(TableScanDesc scan,
67 : : bool *recheck,
68 : : uint64 *lossy_pages, uint64 *exact_pages);
69 : :
70 : :
71 : : /* ------------------------------------------------------------------------
72 : : * Slot related callbacks for heap AM
73 : : * ------------------------------------------------------------------------
74 : : */
75 : :
76 : : static const TupleTableSlotOps *
2694 andres@anarazel.de 77 :CBC 17529576 : heapam_slot_callbacks(Relation relation)
78 : : {
79 : 17529576 : return &TTSOpsBufferHeapTuple;
80 : : }
81 : :
82 : :
83 : : /* ------------------------------------------------------------------------
84 : : * Callbacks for non-modifying operations on individual tuples for heap AM
85 : : * ------------------------------------------------------------------------
86 : : */
87 : :
88 : : static bool
2680 89 : 2839286 : heapam_fetch_row_version(Relation relation,
90 : : ItemPointer tid,
91 : : Snapshot snapshot,
92 : : TupleTableSlot *slot)
93 : : {
94 : 2839286 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
95 : : Buffer buffer;
96 : :
97 [ - + ]: 2839286 : Assert(TTS_IS_BUFFERTUPLE(slot));
98 : :
99 : 2839286 : bslot->base.tupdata.t_self = *tid;
1565 tgl@sss.pgh.pa.us 100 [ + + ]: 2839286 : if (heap_fetch(relation, snapshot, &bslot->base.tupdata, &buffer, false))
101 : : {
102 : : /* store in slot, transferring existing pin */
2680 andres@anarazel.de 103 : 2838957 : ExecStorePinnedBufferHeapTuple(&bslot->base.tupdata, slot, buffer);
104 : 2838957 : slot->tts_tableOid = RelationGetRelid(relation);
105 : :
106 : 2838957 : return true;
107 : : }
108 : :
109 : 321 : return false;
110 : : }
111 : :
112 : : static bool
2627 113 : 468 : heapam_tuple_tid_valid(TableScanDesc scan, ItemPointer tid)
114 : : {
115 : 468 : HeapScanDesc hscan = (HeapScanDesc) scan;
116 : :
117 [ + + ]: 924 : return ItemPointerIsValid(tid) &&
118 [ + + ]: 456 : ItemPointerGetBlockNumber(tid) < hscan->rs_nblocks;
119 : : }
120 : :
121 : : static bool
2694 122 : 773050 : heapam_tuple_satisfies_snapshot(Relation rel, TupleTableSlot *slot,
123 : : Snapshot snapshot)
124 : : {
125 : 773050 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
126 : : bool res;
127 : :
128 [ - + ]: 773050 : Assert(TTS_IS_BUFFERTUPLE(slot));
129 [ - + ]: 773050 : Assert(BufferIsValid(bslot->buffer));
130 : :
131 : : /*
132 : : * We need buffer pin and lock to call HeapTupleSatisfiesVisibility.
133 : : * Caller should be holding pin, but not lock.
134 : : */
135 : 773050 : LockBuffer(bslot->buffer, BUFFER_LOCK_SHARE);
136 : 773050 : res = HeapTupleSatisfiesVisibility(bslot->base.tuple, snapshot,
137 : : bslot->buffer);
138 : 773050 : LockBuffer(bslot->buffer, BUFFER_LOCK_UNLOCK);
139 : :
140 : 773050 : return res;
141 : : }
142 : :
143 : :
144 : : /* ----------------------------------------------------------------------------
145 : : * Functions for manipulations of physical tuples for heap AM.
146 : : * ----------------------------------------------------------------------------
147 : : */
148 : :
149 : : static void
2682 150 : 10628199 : heapam_tuple_insert(Relation relation, TupleTableSlot *slot, CommandId cid,
151 : : uint32 options, BulkInsertState bistate)
152 : : {
153 : 10628199 : bool shouldFree = true;
154 : 10628199 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
155 : :
156 : : /* Update the tuple with table oid */
157 : 10628199 : slot->tts_tableOid = RelationGetRelid(relation);
158 : 10628199 : tuple->t_tableOid = slot->tts_tableOid;
159 : :
160 : : /* Perform the insertion, and copy the resulting ItemPointer */
161 : 10628199 : heap_insert(relation, tuple, cid, options, bistate);
162 : 10628183 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
163 : :
164 [ + + ]: 10628183 : if (shouldFree)
165 : 2925364 : pfree(tuple);
166 : 10628183 : }
167 : :
168 : : static void
2675 169 : 2231 : heapam_tuple_insert_speculative(Relation relation, TupleTableSlot *slot,
170 : : CommandId cid, uint32 options,
171 : : BulkInsertState bistate, uint32 specToken)
172 : : {
2682 173 : 2231 : bool shouldFree = true;
174 : 2231 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
175 : :
176 : : /* Update the tuple with table oid */
177 : 2231 : slot->tts_tableOid = RelationGetRelid(relation);
178 : 2231 : tuple->t_tableOid = slot->tts_tableOid;
179 : :
180 : 2231 : HeapTupleHeaderSetSpeculativeToken(tuple->t_data, specToken);
181 : 2231 : options |= HEAP_INSERT_SPECULATIVE;
182 : :
183 : : /* Perform the insertion, and copy the resulting ItemPointer */
184 : 2231 : heap_insert(relation, tuple, cid, options, bistate);
185 : 2231 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
186 : :
187 [ + + ]: 2231 : if (shouldFree)
188 : 54 : pfree(tuple);
189 : 2231 : }
190 : :
191 : : static void
2675 192 : 2227 : heapam_tuple_complete_speculative(Relation relation, TupleTableSlot *slot,
193 : : uint32 specToken, bool succeeded)
194 : : {
195 : : /* adjust the tuple's state accordingly */
2630 196 [ + + ]: 2227 : if (succeeded)
2682 197 : 2216 : heap_finish_speculative(relation, &slot->tts_tid);
198 : : else
199 : 11 : heap_abort_speculative(relation, &slot->tts_tid);
200 : 2227 : }
201 : :
202 : : static TM_Result
203 : 1027733 : heapam_tuple_delete(Relation relation, ItemPointer tid, CommandId cid,
204 : : uint32 options, Snapshot snapshot, Snapshot crosscheck,
205 : : bool wait, TM_FailureData *tmfd)
206 : : {
207 : : /*
208 : : * Currently Deleting of index tuples are handled at vacuum, in case if
209 : : * the storage itself is cleaning the dead tuples by itself, it is the
210 : : * time to call the index tuple deletion also.
211 : : */
116 alvherre@kurilemu.de 212 : 1027733 : return heap_delete(relation, tid, cid, options, crosscheck, wait,
213 : : tmfd);
214 : : }
215 : :
216 : :
217 : : static TM_Result
2682 andres@anarazel.de 218 : 2244356 : heapam_tuple_update(Relation relation, ItemPointer otid, TupleTableSlot *slot,
219 : : CommandId cid, uint32 options,
220 : : Snapshot snapshot, Snapshot crosscheck,
221 : : bool wait, TM_FailureData *tmfd,
222 : : LockTupleMode *lockmode, TU_UpdateIndexes *update_indexes)
223 : : {
224 : 2244356 : bool shouldFree = true;
225 : 2244356 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
226 : : TM_Result result;
227 : :
228 : : /* Update the tuple with table oid */
229 : 2244356 : slot->tts_tableOid = RelationGetRelid(relation);
230 : 2244356 : tuple->t_tableOid = slot->tts_tableOid;
231 : :
116 alvherre@kurilemu.de 232 : 2244356 : result = heap_update(relation, otid, tuple, cid, options,
233 : : crosscheck, wait,
234 : : tmfd, lockmode, update_indexes);
2682 andres@anarazel.de 235 : 2244344 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
236 : :
237 : : /*
238 : : * Decide whether new index entries are needed for the tuple
239 : : *
240 : : * Note: heap_update returns the tid (location) of the new tuple in the
241 : : * t_self field.
242 : : *
243 : : * If the update is not HOT, we must update all indexes. If the update is
244 : : * HOT, it could be that we updated summarized columns, so we either
245 : : * update only summarized indexes, or none at all.
246 : : */
1224 tomas.vondra@postgre 247 [ + + ]: 2244344 : if (result != TM_Ok)
248 : : {
249 [ - + ]: 209 : Assert(*update_indexes == TU_None);
250 : 209 : *update_indexes = TU_None;
251 : : }
252 [ + + ]: 2244135 : else if (!HeapTupleIsHeapOnly(tuple))
253 [ - + ]: 2173975 : Assert(*update_indexes == TU_All);
254 : : else
255 [ + + - + ]: 70160 : Assert((*update_indexes == TU_Summarizing) ||
256 : : (*update_indexes == TU_None));
257 : :
2682 andres@anarazel.de 258 [ + + ]: 2244344 : if (shouldFree)
259 : 31987 : pfree(tuple);
260 : :
261 : 2244344 : return result;
262 : : }
263 : :
264 : : static TM_Result
265 : 571513 : heapam_tuple_lock(Relation relation, ItemPointer tid, Snapshot snapshot,
266 : : TupleTableSlot *slot, CommandId cid, LockTupleMode mode,
267 : : LockWaitPolicy wait_policy, uint8 flags,
268 : : TM_FailureData *tmfd)
269 : : {
270 : 571513 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
271 : : TM_Result result;
272 : : Buffer buffer;
273 : 571513 : HeapTuple tuple = &bslot->base.tupdata;
274 : : bool follow_updates;
275 : :
276 : 571513 : follow_updates = (flags & TUPLE_LOCK_FLAG_LOCK_UPDATE_IN_PROGRESS) != 0;
277 : 571513 : tmfd->traversed = false;
278 : :
279 [ + - ]: 571513 : Assert(TTS_IS_BUFFERTUPLE(slot));
280 : :
281 : 571513 : tuple_lock_retry:
836 akorotkov@postgresql 282 : 571704 : tuple->t_self = *tid;
283 : 571704 : result = heap_lock_tuple(relation, tuple, cid, mode, wait_policy,
284 : : follow_updates, &buffer, tmfd);
285 : :
2682 andres@anarazel.de 286 [ + + ]: 571691 : if (result == TM_Updated &&
287 [ + + ]: 234 : (flags & TUPLE_LOCK_FLAG_FIND_LAST_VERSION))
288 : : {
289 : : /* Should not encounter speculative tuple on recheck */
1210 akorotkov@postgresql 290 [ - + ]: 214 : Assert(!HeapTupleHeaderIsSpeculative(tuple->t_data));
291 : :
836 292 : 214 : ReleaseBuffer(buffer);
293 : :
2682 andres@anarazel.de 294 [ + - ]: 214 : if (!ItemPointerEquals(&tmfd->ctid, &tuple->t_self))
295 : : {
296 : : SnapshotData SnapshotDirty;
297 : : TransactionId priorXmax;
298 : :
299 : : /* it was updated, so look at the updated version */
300 : 214 : *tid = tmfd->ctid;
301 : : /* updated row should have xmin matching this xmax */
302 : 214 : priorXmax = tmfd->xmax;
303 : :
304 : : /* signal that a tuple later in the chain is getting locked */
305 : 214 : tmfd->traversed = true;
306 : :
307 : : /*
308 : : * fetch target tuple
309 : : *
310 : : * Loop here to deal with updated or busy tuples
311 : : */
312 : 214 : InitDirtySnapshot(SnapshotDirty);
313 : : for (;;)
314 : : {
315 [ + + ]: 249 : if (ItemPointerIndicatesMovedPartitions(tid))
316 [ + - ]: 11 : ereport(ERROR,
317 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
318 : : errmsg("tuple to be locked was already moved to another partition due to concurrent update")));
319 : :
320 : 238 : tuple->t_self = *tid;
1565 tgl@sss.pgh.pa.us 321 [ + + ]: 238 : if (heap_fetch(relation, &SnapshotDirty, tuple, &buffer, true))
322 : : {
323 : : /*
324 : : * If xmin isn't what we're expecting, the slot must have
325 : : * been recycled and reused for an unrelated tuple. This
326 : : * implies that the latest version of the row was deleted,
327 : : * so we need do nothing. (Should be safe to examine xmin
328 : : * without getting buffer's content lock. We assume
329 : : * reading a TransactionId to be atomic, and Xmin never
330 : : * changes in an existing tuple, except to invalid or
331 : : * frozen, and neither of those can match priorXmax.)
332 : : */
2682 andres@anarazel.de 333 [ - + ]: 200 : if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple->t_data),
334 : : priorXmax))
335 : : {
2682 andres@anarazel.de 336 :UBC 0 : ReleaseBuffer(buffer);
2682 andres@anarazel.de 337 :CBC 11 : return TM_Deleted;
338 : : }
339 : :
340 : : /* otherwise xmin should not be dirty... */
341 [ - + ]: 200 : if (TransactionIdIsValid(SnapshotDirty.xmin))
2551 peter@eisentraut.org 342 [ # # ]:UBC 0 : ereport(ERROR,
343 : : (errcode(ERRCODE_DATA_CORRUPTED),
344 : : errmsg_internal("t_xmin %u is uncommitted in tuple (%u,%u) to be updated in table \"%s\"",
345 : : SnapshotDirty.xmin,
346 : : ItemPointerGetBlockNumber(&tuple->t_self),
347 : : ItemPointerGetOffsetNumber(&tuple->t_self),
348 : : RelationGetRelationName(relation))));
349 : :
350 : : /*
351 : : * If tuple is being updated by other transaction then we
352 : : * have to wait for its commit/abort, or die trying.
353 : : */
2682 andres@anarazel.de 354 [ + + ]:CBC 200 : if (TransactionIdIsValid(SnapshotDirty.xmax))
355 : : {
356 : 2 : ReleaseBuffer(buffer);
357 [ - + + - ]: 2 : switch (wait_policy)
358 : : {
2682 andres@anarazel.de 359 :UBC 0 : case LockWaitBlock:
360 : 0 : XactLockTableWait(SnapshotDirty.xmax,
361 : 0 : relation, &tuple->t_self,
362 : : XLTW_FetchUpdated);
363 : 0 : break;
2682 andres@anarazel.de 364 :CBC 1 : case LockWaitSkip:
499 fujii@postgresql.org 365 [ + - ]: 1 : if (!ConditionalXactLockTableWait(SnapshotDirty.xmax, false))
366 : : /* skip instead of waiting */
2682 andres@anarazel.de 367 : 1 : return TM_WouldBlock;
2682 andres@anarazel.de 368 :UBC 0 : break;
2682 andres@anarazel.de 369 :CBC 1 : case LockWaitError:
418 fujii@postgresql.org 370 [ + - ]: 1 : if (!ConditionalXactLockTableWait(SnapshotDirty.xmax, log_lock_failures))
2682 andres@anarazel.de 371 [ + - ]: 1 : ereport(ERROR,
372 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
373 : : errmsg("could not obtain lock on row in relation \"%s\"",
374 : : RelationGetRelationName(relation))));
2682 andres@anarazel.de 375 :UBC 0 : break;
376 : : }
377 : 0 : continue; /* loop back to repeat heap_fetch */
378 : : }
379 : :
380 : : /*
381 : : * If tuple was inserted by our own transaction, we have
382 : : * to check cmin against cid: cmin >= current CID means
383 : : * our command cannot see the tuple, so we should ignore
384 : : * it. Otherwise heap_lock_tuple() will throw an error,
385 : : * and so would any later attempt to update or delete the
386 : : * tuple. (We need not check cmax because
387 : : * HeapTupleSatisfiesDirty will consider a tuple deleted
388 : : * by our transaction dead, regardless of cmax.) We just
389 : : * checked that priorXmax == xmin, so we can test that
390 : : * variable instead of doing HeapTupleHeaderGetXmin again.
391 : : */
2682 andres@anarazel.de 392 [ + + + - ]:CBC 205 : if (TransactionIdIsCurrentTransactionId(priorXmax) &&
393 : 7 : HeapTupleHeaderGetCmin(tuple->t_data) >= cid)
394 : : {
2667 395 : 7 : tmfd->xmax = priorXmax;
396 : :
397 : : /*
398 : : * Cmin is the problematic value, so store that. See
399 : : * above.
400 : : */
401 : 7 : tmfd->cmax = HeapTupleHeaderGetCmin(tuple->t_data);
2682 402 : 7 : ReleaseBuffer(buffer);
2667 403 : 7 : return TM_SelfModified;
404 : : }
405 : :
406 : : /*
407 : : * This is a live tuple, so try to lock it again.
408 : : */
836 akorotkov@postgresql 409 : 191 : ReleaseBuffer(buffer);
2682 andres@anarazel.de 410 : 191 : goto tuple_lock_retry;
411 : : }
412 : :
413 : : /*
414 : : * If the referenced slot was actually empty, the latest
415 : : * version of the row must have been deleted, so we need do
416 : : * nothing.
417 : : */
418 [ - + ]: 38 : if (tuple->t_data == NULL)
419 : : {
836 akorotkov@postgresql 420 [ # # ]:UBC 0 : Assert(!BufferIsValid(buffer));
2682 andres@anarazel.de 421 : 0 : return TM_Deleted;
422 : : }
423 : :
424 : : /*
425 : : * As above, if xmin isn't what we're expecting, do nothing.
426 : : */
2682 andres@anarazel.de 427 [ - + ]:CBC 38 : if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple->t_data),
428 : : priorXmax))
429 : : {
1565 tgl@sss.pgh.pa.us 430 :UBC 0 : ReleaseBuffer(buffer);
2682 andres@anarazel.de 431 : 0 : return TM_Deleted;
432 : : }
433 : :
434 : : /*
435 : : * If we get here, the tuple was found but failed
436 : : * SnapshotDirty. Assuming the xmin is either a committed xact
437 : : * or our own xact (as it certainly should be if we're trying
438 : : * to modify the tuple), this must mean that the row was
439 : : * updated or deleted by either a committed xact or our own
440 : : * xact. If it was deleted, we can ignore it; if it was
441 : : * updated then chain up to the next version and repeat the
442 : : * whole process.
443 : : *
444 : : * As above, it should be safe to examine xmax and t_ctid
445 : : * without the buffer content lock, because they can't be
446 : : * changing. We'd better hold a buffer pin though.
447 : : */
2682 andres@anarazel.de 448 [ + + ]:CBC 38 : if (ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
449 : : {
450 : : /* deleted, so forget about it */
1565 tgl@sss.pgh.pa.us 451 : 3 : ReleaseBuffer(buffer);
2682 andres@anarazel.de 452 : 3 : return TM_Deleted;
453 : : }
454 : :
455 : : /* updated, so look at the updated row */
456 : 35 : *tid = tuple->t_data->t_ctid;
457 : : /* updated row should have xmin matching this xmax */
458 : 35 : priorXmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
1565 tgl@sss.pgh.pa.us 459 : 35 : ReleaseBuffer(buffer);
460 : : /* loop back to fetch next in chain */
461 : : }
462 : : }
463 : : else
464 : : {
465 : : /* tuple was deleted, so give up */
2682 andres@anarazel.de 466 :UBC 0 : return TM_Deleted;
467 : : }
468 : : }
469 : :
2682 andres@anarazel.de 470 :CBC 571477 : slot->tts_tableOid = RelationGetRelid(relation);
471 : 571477 : tuple->t_tableOid = slot->tts_tableOid;
472 : :
473 : : /* store in slot, transferring existing pin */
836 akorotkov@postgresql 474 : 571477 : ExecStorePinnedBufferHeapTuple(tuple, slot, buffer);
475 : :
2682 andres@anarazel.de 476 : 571477 : return result;
477 : : }
478 : :
479 : :
480 : : /* ------------------------------------------------------------------------
481 : : * DDL related callbacks for heap AM.
482 : : * ------------------------------------------------------------------------
483 : : */
484 : :
485 : : static void
1481 rhaas@postgresql.org 486 : 44068 : heapam_relation_set_new_filelocator(Relation rel,
487 : : const RelFileLocator *newrlocator,
488 : : char persistence,
489 : : TransactionId *freezeXid,
490 : : MultiXactId *minmulti)
491 : : {
492 : : SMgrRelation srel;
493 : :
494 : : /*
495 : : * Initialize to the minimum XID that could put tuples in the table. We
496 : : * know that no xacts older than RecentXmin are still running, so that
497 : : * will do.
498 : : */
2677 andres@anarazel.de 499 : 44068 : *freezeXid = RecentXmin;
500 : :
501 : : /*
502 : : * Similarly, initialize the minimum Multixact to the first value that
503 : : * could possibly be stored in tuples in the table. Running transactions
504 : : * could reuse values from their local cache, so we are careful to
505 : : * consider all currently running multis.
506 : : *
507 : : * XXX this could be refined further, but is it worth the hassle?
508 : : */
509 : 44068 : *minmulti = GetOldestMultiXactId();
510 : :
1481 rhaas@postgresql.org 511 : 44068 : srel = RelationCreateStorage(*newrlocator, persistence, true);
512 : :
513 : : /*
514 : : * If required, set up an init fork for an unlogged table so that it can
515 : : * be correctly reinitialized on restart.
516 : : */
2645 andres@anarazel.de 517 [ + + ]: 44068 : if (persistence == RELPERSISTENCE_UNLOGGED)
518 : : {
2677 519 [ + + - + ]: 181 : Assert(rel->rd_rel->relkind == RELKIND_RELATION ||
520 : : rel->rd_rel->relkind == RELKIND_TOASTVALUE);
2645 521 : 181 : smgrcreate(srel, INIT_FORKNUM, false);
1481 rhaas@postgresql.org 522 : 181 : log_smgrcreate(newrlocator, INIT_FORKNUM);
523 : : }
524 : :
2645 andres@anarazel.de 525 : 44068 : smgrclose(srel);
2677 526 : 44068 : }
527 : :
528 : : static void
529 : 372 : heapam_relation_nontransactional_truncate(Relation rel)
530 : : {
531 : 372 : RelationTruncate(rel, 0);
532 : 372 : }
533 : :
534 : : static void
1481 rhaas@postgresql.org 535 : 67 : heapam_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
536 : : {
537 : : SMgrRelation dstrel;
538 : :
539 : : /*
540 : : * Since we copy the file directly without looking at the shared buffers,
541 : : * we'd better first flush out any pages of the source relation that are
542 : : * in shared buffers. We assume no new changes will be made while we are
543 : : * holding exclusive lock on the rel.
544 : : */
2645 andres@anarazel.de 545 : 67 : FlushRelationBuffers(rel);
546 : :
547 : : /*
548 : : * Create and copy all forks of the relation, and schedule unlinking of
549 : : * old physical files.
550 : : *
551 : : * NOTE: any conflict in relfilenumber value will be caught in
552 : : * RelationCreateStorage().
553 : : */
895 heikki.linnakangas@i 554 : 67 : dstrel = RelationCreateStorage(*newrlocator, rel->rd_rel->relpersistence, true);
555 : :
556 : : /* copy main fork */
1840 tgl@sss.pgh.pa.us 557 : 67 : RelationCopyStorage(RelationGetSmgr(rel), dstrel, MAIN_FORKNUM,
2677 andres@anarazel.de 558 : 67 : rel->rd_rel->relpersistence);
559 : :
560 : : /* copy those extra forks that exist */
561 : 67 : for (ForkNumber forkNum = MAIN_FORKNUM + 1;
562 [ + + ]: 268 : forkNum <= MAX_FORKNUM; forkNum++)
563 : : {
1840 tgl@sss.pgh.pa.us 564 [ + + ]: 201 : if (smgrexists(RelationGetSmgr(rel), forkNum))
565 : : {
2677 andres@anarazel.de 566 : 23 : smgrcreate(dstrel, forkNum, false);
567 : :
568 : : /*
569 : : * WAL log creation if the relation is persistent, or this is the
570 : : * init fork of an unlogged relation.
571 : : */
1952 bruce@momjian.us 572 [ + + ]: 23 : if (RelationIsPermanent(rel) ||
2677 andres@anarazel.de 573 [ - + - - ]: 8 : (rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
574 : : forkNum == INIT_FORKNUM))
1481 rhaas@postgresql.org 575 : 15 : log_smgrcreate(newrlocator, forkNum);
1840 tgl@sss.pgh.pa.us 576 : 23 : RelationCopyStorage(RelationGetSmgr(rel), dstrel, forkNum,
2677 andres@anarazel.de 577 : 23 : rel->rd_rel->relpersistence);
578 : : }
579 : : }
580 : :
581 : :
582 : : /* drop old relation, and close new one */
583 : 67 : RelationDropStorage(rel);
584 : 67 : smgrclose(dstrel);
585 : 67 : }
586 : :
587 : : static void
588 : 409 : heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
589 : : Relation OldIndex, bool use_sort,
590 : : TransactionId OldestXmin,
591 : : Snapshot snapshot,
592 : : TransactionId *xid_cutoff,
593 : : MultiXactId *multi_cutoff,
594 : : double *num_tuples,
595 : : double *tups_vacuumed,
596 : : double *tups_recently_dead)
597 : : {
598 : : RewriteState rwstate;
599 : : BulkInsertState bistate;
600 : : IndexScanDesc indexScan;
601 : : TableScanDesc tableScan;
602 : : HeapScanDesc heapScan;
603 : : bool is_system_catalog;
604 : : Tuplesortstate *tuplesort;
605 : 409 : TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
606 : 409 : TupleDesc newTupDesc = RelationGetDescr(NewHeap);
607 : : TupleTableSlot *slot;
608 : : int natts;
609 : : Datum *values;
610 : : bool *isnull;
611 : : BufferHeapTupleTableSlot *hslot;
2067 fujii@postgresql.org 612 : 409 : BlockNumber prev_cblock = InvalidBlockNumber;
111 alvherre@kurilemu.de 613 : 409 : bool concurrent = snapshot != NULL;
614 : :
615 : : /* Remember if it's a system catalog */
2677 andres@anarazel.de 616 : 409 : is_system_catalog = IsSystemRelation(OldHeap);
617 : :
618 : : /*
619 : : * Valid smgr_targblock implies something already wrote to the relation.
620 : : * This may be harmless, but this function hasn't planned for it.
621 : : */
622 [ - + - - ]: 409 : Assert(RelationGetTargetBlock(NewHeap) == InvalidBlockNumber);
623 : :
624 : : /* Preallocate values/isnull arrays */
625 : 409 : natts = newTupDesc->natts;
228 michael@paquier.xyz 626 : 409 : values = palloc_array(Datum, natts);
627 : 409 : isnull = palloc_array(bool, natts);
628 : :
629 : : /*
630 : : * In non-concurrent mode, initialize the rewrite operation. This is not
631 : : * needed in concurrent mode.
632 : : */
111 alvherre@kurilemu.de 633 [ + + ]: 409 : if (!concurrent)
634 : 402 : rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
635 : : *xid_cutoff, *multi_cutoff);
636 : : else
637 : 7 : rwstate = NULL;
638 : :
639 : : /* In concurrent mode, prepare for bulk-insert operation. */
640 [ + + ]: 409 : if (concurrent)
641 : 7 : bistate = GetBulkInsertState();
642 : : else
643 : 402 : bistate = NULL;
644 : :
645 : : /* Set up sorting if wanted */
2677 andres@anarazel.de 646 [ + + ]: 409 : if (use_sort)
1142 pg@bowt.ie 647 : 85 : tuplesort = tuplesort_begin_cluster(oldTupDesc, OldIndex,
648 : : maintenance_work_mem,
649 : : NULL, TUPLESORT_NONE);
650 : : else
2677 andres@anarazel.de 651 : 324 : tuplesort = NULL;
652 : :
653 : : /*
654 : : * Prepare to scan the OldHeap. To ensure we see recently-dead tuples
655 : : * that still need to be copied, we scan with SnapshotAny and use
656 : : * HeapTupleSatisfiesVacuum for the visibility test.
657 : : *
658 : : * In the CONCURRENTLY case, we do regular MVCC visibility tests, using
659 : : * the snapshot passed by the caller.
660 : : */
661 [ + + + + ]: 409 : if (OldIndex != NULL && !use_sort)
662 : 62 : {
2675 663 : 62 : const int ci_index[] = {
664 : : PROGRESS_REPACK_PHASE,
665 : : PROGRESS_REPACK_INDEX_RELID
666 : : };
667 : : int64 ci_val[2];
668 : :
669 : : /* Set phase and OIDOldIndex to columns */
138 alvherre@kurilemu.de 670 : 62 : ci_val[0] = PROGRESS_REPACK_PHASE_INDEX_SCAN_HEAP;
2677 andres@anarazel.de 671 : 62 : ci_val[1] = RelationGetRelid(OldIndex);
672 : 62 : pgstat_progress_update_multi_param(2, ci_index, ci_val);
673 : :
674 : 62 : tableScan = NULL;
675 : 62 : heapScan = NULL;
111 alvherre@kurilemu.de 676 [ + + ]: 62 : indexScan = index_beginscan(OldHeap, OldIndex,
677 : : snapshot ? snapshot : SnapshotAny,
678 : : NULL, 0, 0,
679 : : SO_NONE);
2677 andres@anarazel.de 680 : 62 : index_rescan(indexScan, NULL, 0, NULL, 0);
681 : : }
682 : : else
683 : : {
684 : : /* In scan-and-sort mode and also VACUUM FULL, set phase */
138 alvherre@kurilemu.de 685 : 347 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
686 : : PROGRESS_REPACK_PHASE_SEQ_SCAN_HEAP);
687 : :
111 688 [ + + ]: 347 : tableScan = table_beginscan(OldHeap,
689 : : snapshot ? snapshot : SnapshotAny,
690 : : 0, (ScanKey) NULL,
691 : : SO_NONE);
2677 andres@anarazel.de 692 : 347 : heapScan = (HeapScanDesc) tableScan;
693 : 347 : indexScan = NULL;
694 : :
695 : : /* Set total heap blocks */
138 alvherre@kurilemu.de 696 : 347 : pgstat_progress_update_param(PROGRESS_REPACK_TOTAL_HEAP_BLKS,
2677 andres@anarazel.de 697 : 347 : heapScan->rs_nblocks);
698 : : }
699 : :
700 : 409 : slot = table_slot_create(OldHeap, NULL);
701 : 409 : hslot = (BufferHeapTupleTableSlot *) slot;
702 : :
703 : : /*
704 : : * Scan through the OldHeap, either in OldIndex order or sequentially;
705 : : * copy each tuple into the NewHeap, or transiently to the tuplesort
706 : : * module. Note that we don't bother sorting dead tuples (they won't get
707 : : * to the new table anyway).
708 : : */
709 : : for (;;)
710 : 496074 : {
711 : : HeapTuple tuple;
712 : : Buffer buf;
713 : : bool isdead;
714 : :
715 [ - + ]: 496483 : CHECK_FOR_INTERRUPTS();
716 : :
717 [ + + ]: 496483 : if (indexScan != NULL)
718 : : {
719 [ + + ]: 1204 : if (!index_getnext_slot(indexScan, ForwardScanDirection, slot))
720 : 62 : break;
721 : :
722 : : /* Since we used no scan keys, should never need to recheck */
723 [ - + ]: 1142 : if (indexScan->xs_recheck)
2677 andres@anarazel.de 724 [ # # ]:UBC 0 : elog(ERROR, "CLUSTER does not support lossy index conditions");
725 : : }
726 : : else
727 : : {
2677 andres@anarazel.de 728 [ + + ]:CBC 495279 : if (!table_scan_getnextslot(tableScan, ForwardScanDirection, slot))
729 : : {
730 : : /*
731 : : * If the last pages of the scan were empty, we would go to
732 : : * the next phase while heap_blks_scanned != heap_blks_total.
733 : : * Instead, to ensure that heap_blks_scanned is equivalent to
734 : : * heap_blks_total after the table scan phase, this parameter
735 : : * is manually updated to the correct value when the table
736 : : * scan finishes.
737 : : */
138 alvherre@kurilemu.de 738 : 347 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_BLKS_SCANNED,
2067 fujii@postgresql.org 739 : 347 : heapScan->rs_nblocks);
2677 andres@anarazel.de 740 : 347 : break;
741 : : }
742 : :
743 : : /*
744 : : * In scan-and-sort mode and also VACUUM FULL, set heap blocks
745 : : * scanned
746 : : *
747 : : * Note that heapScan may start at an offset and wrap around, i.e.
748 : : * rs_startblock may be >0, and rs_cblock may end with a number
749 : : * below rs_startblock. To prevent showing this wraparound to the
750 : : * user, we offset rs_cblock by rs_startblock (modulo rs_nblocks).
751 : : */
2067 fujii@postgresql.org 752 [ + + ]: 494932 : if (prev_cblock != heapScan->rs_cblock)
753 : : {
138 alvherre@kurilemu.de 754 : 7159 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_BLKS_SCANNED,
2067 fujii@postgresql.org 755 : 7159 : (heapScan->rs_cblock +
756 : 7159 : heapScan->rs_nblocks -
757 : 7159 : heapScan->rs_startblock
758 : 7159 : ) % heapScan->rs_nblocks + 1);
759 : 7159 : prev_cblock = heapScan->rs_cblock;
760 : : }
761 : : }
762 : :
2677 andres@anarazel.de 763 : 496074 : tuple = ExecFetchSlotHeapTuple(slot, false, NULL);
764 : 496074 : buf = hslot->buffer;
765 : :
766 : : /*
767 : : * In concurrent mode, our table or index scan has used regular MVCC
768 : : * visibility test against a snapshot passed by caller; therefore we
769 : : * don't need another visibility test. In non-concurrent mode
770 : : * however, we must test the visibility of each tuple we read.
771 : : */
111 alvherre@kurilemu.de 772 [ + + ]: 496074 : if (!concurrent)
773 : : {
774 : : /*
775 : : * To be able to guarantee that we can set the hint bit, acquire
776 : : * an exclusive lock on the old buffer. We need the hint bits, set
777 : : * in heapam_relation_copy_for_cluster() ->
778 : : * HeapTupleSatisfiesVacuum(), to be set, as otherwise
779 : : * reform_and_rewrite_tuple() -> rewrite_heap_tuple() will get
780 : : * confused. Specifically, rewrite_heap_tuple() checks for
781 : : * HEAP_XMAX_INVALID in the old tuple to determine whether to
782 : : * check the old-to-new mapping hash table.
783 : : *
784 : : * It'd be better if we somehow could avoid setting hint bits on
785 : : * the old page. One reason to use VACUUM FULL are very bloated
786 : : * tables - rewriting most of the old table during VACUUM FULL
787 : : * doesn't exactly help...
788 : : */
789 : 496043 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
790 : :
791 [ + + + + : 496043 : switch (HeapTupleSatisfiesVacuum(tuple, OldestXmin, buf))
+ - ]
792 : : {
793 : 21322 : case HEAPTUPLE_DEAD:
794 : : /* Definitely dead */
795 : 21322 : isdead = true;
796 : 21322 : break;
797 : 29282 : case HEAPTUPLE_RECENTLY_DEAD:
798 : 29282 : *tups_recently_dead += 1;
799 : : pg_fallthrough;
800 : 474577 : case HEAPTUPLE_LIVE:
801 : : /* Live or recently dead, must copy it */
802 : 474577 : isdead = false;
803 : 474577 : break;
804 : 113 : case HEAPTUPLE_INSERT_IN_PROGRESS:
805 : :
806 : : /*
807 : : * As long as we hold exclusive lock on the relation,
808 : : * normally the only way to see this is if it was inserted
809 : : * earlier in our own transaction. However, it can happen
810 : : * in system catalogs, since we tend to release write lock
811 : : * before commit there. Give a warning if neither case
812 : : * applies; but in any case we had better copy it.
813 : : */
814 [ + + ]: 113 : if (!is_system_catalog &&
815 [ - + ]: 14 : !TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetXmin(tuple->t_data)))
111 alvherre@kurilemu.de 816 [ # # ]:UBC 0 : elog(WARNING, "concurrent insert in progress within table \"%s\"",
817 : : RelationGetRelationName(OldHeap));
818 : : /* treat as live */
111 alvherre@kurilemu.de 819 :CBC 113 : isdead = false;
820 : 113 : break;
821 : 31 : case HEAPTUPLE_DELETE_IN_PROGRESS:
822 : :
823 : : /*
824 : : * Similar situation to INSERT_IN_PROGRESS case.
825 : : */
826 [ + + ]: 31 : if (!is_system_catalog &&
827 [ - + ]: 20 : !TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetUpdateXid(tuple->t_data)))
111 alvherre@kurilemu.de 828 [ # # ]:UBC 0 : elog(WARNING, "concurrent delete in progress within table \"%s\"",
829 : : RelationGetRelationName(OldHeap));
830 : : /* treat as recently dead */
111 alvherre@kurilemu.de 831 :CBC 31 : *tups_recently_dead += 1;
832 : 31 : isdead = false;
833 : 31 : break;
111 alvherre@kurilemu.de 834 :UBC 0 : default:
835 [ # # ]: 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
836 : : isdead = false; /* keep compiler quiet */
837 : : break;
838 : : }
839 : :
111 alvherre@kurilemu.de 840 :CBC 496043 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
841 : :
842 [ + + ]: 496043 : if (isdead)
843 : : {
2677 andres@anarazel.de 844 : 21322 : *tups_vacuumed += 1;
845 : : /* heap rewrite module still needs to see it... */
111 alvherre@kurilemu.de 846 [ - + ]: 21322 : if (rewrite_heap_dead_tuple(rwstate, tuple))
847 : : {
848 : : /* A previous recently-dead tuple is now known dead */
111 alvherre@kurilemu.de 849 :UBC 0 : *tups_vacuumed += 1;
850 : 0 : *tups_recently_dead -= 1;
851 : : }
852 : :
111 alvherre@kurilemu.de 853 :CBC 21322 : continue;
854 : : }
855 : : }
856 : :
2677 andres@anarazel.de 857 : 474752 : *num_tuples += 1;
858 [ + + ]: 474752 : if (tuplesort != NULL)
859 : : {
860 : 363322 : tuplesort_putheaptuple(tuplesort, tuple);
861 : :
862 : : /*
863 : : * In scan-and-sort mode, report increase in number of tuples
864 : : * scanned
865 : : */
138 alvherre@kurilemu.de 866 : 363322 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_SCANNED,
2677 andres@anarazel.de 867 : 363322 : *num_tuples);
868 : : }
869 : : else
870 : : {
2675 871 : 111430 : const int ct_index[] = {
872 : : PROGRESS_REPACK_HEAP_TUPLES_SCANNED,
873 : : PROGRESS_REPACK_HEAP_TUPLES_INSERTED
874 : : };
875 : : int64 ct_val[2];
876 : :
111 alvherre@kurilemu.de 877 [ + + ]: 111430 : if (!concurrent)
878 : 111399 : reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
879 : : values, isnull, rwstate);
880 : : else
881 : 31 : heap_insert_for_repack(tuple, OldHeap, NewHeap,
882 : : values, isnull, bistate);
883 : :
884 : : /*
885 : : * In indexscan mode and also VACUUM FULL, report increase in
886 : : * number of tuples scanned and written
887 : : */
2677 andres@anarazel.de 888 : 111430 : ct_val[0] = *num_tuples;
889 : 111430 : ct_val[1] = *num_tuples;
890 : 111430 : pgstat_progress_update_multi_param(2, ct_index, ct_val);
891 : : }
892 : : }
893 : :
894 [ + + ]: 409 : if (indexScan != NULL)
895 : 62 : index_endscan(indexScan);
896 [ + + ]: 409 : if (tableScan != NULL)
897 : 347 : table_endscan(tableScan);
898 [ + - ]: 409 : if (slot)
899 : 409 : ExecDropSingleTupleTableSlot(slot);
900 : :
901 : : /*
902 : : * In scan-and-sort mode, complete the sort, then read out all live tuples
903 : : * from the tuplestore and write them to the new relation.
904 : : */
905 [ + + ]: 409 : if (tuplesort != NULL)
906 : : {
2675 907 : 85 : double n_tuples = 0;
908 : :
909 : : /* Report that we are now sorting tuples */
138 alvherre@kurilemu.de 910 : 85 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
911 : : PROGRESS_REPACK_PHASE_SORT_TUPLES);
912 : :
2677 andres@anarazel.de 913 : 85 : tuplesort_performsort(tuplesort);
914 : :
915 : : /* Report that we are now writing new heap */
138 alvherre@kurilemu.de 916 : 85 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
917 : : PROGRESS_REPACK_PHASE_WRITE_NEW_HEAP);
918 : :
919 : : for (;;)
2677 andres@anarazel.de 920 : 363322 : {
921 : : HeapTuple tuple;
922 : :
923 [ - + ]: 363407 : CHECK_FOR_INTERRUPTS();
924 : :
925 : 363407 : tuple = tuplesort_getheaptuple(tuplesort, true);
926 [ + + ]: 363407 : if (tuple == NULL)
927 : 85 : break;
928 : :
929 : 363322 : n_tuples += 1;
111 alvherre@kurilemu.de 930 [ + - ]: 363322 : if (!concurrent)
931 : 363322 : reform_and_rewrite_tuple(tuple,
932 : : OldHeap, NewHeap,
933 : : values, isnull,
934 : : rwstate);
935 : : else
111 alvherre@kurilemu.de 936 :UBC 0 : heap_insert_for_repack(tuple, OldHeap, NewHeap,
937 : : values, isnull, bistate);
938 : :
939 : : /* Report n_tuples */
111 alvherre@kurilemu.de 940 :CBC 363322 : pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
941 : : n_tuples);
942 : : }
943 : :
2677 andres@anarazel.de 944 : 85 : tuplesort_end(tuplesort);
945 : : }
946 : :
947 : : /* Write out any remaining tuples, and fsync if needed */
111 alvherre@kurilemu.de 948 [ + + ]: 409 : if (rwstate)
949 : 402 : end_heap_rewrite(rwstate);
950 [ + + ]: 409 : if (bistate)
951 : 7 : FreeBulkInsertState(bistate);
952 : :
953 : : /* Clean up */
2677 andres@anarazel.de 954 : 409 : pfree(values);
955 : 409 : pfree(isnull);
956 : 409 : }
957 : :
958 : : /*
959 : : * Prepare to analyze the next block in the read stream. Returns false if
960 : : * the stream is exhausted and true otherwise. The scan must have been started
961 : : * with SO_TYPE_ANALYZE option.
962 : : *
963 : : * This routine holds a buffer pin and lock on the heap page. They are held
964 : : * until heapam_scan_analyze_next_tuple() returns false. That is until all the
965 : : * items of the heap page are analyzed.
966 : : */
967 : : static bool
839 tmunro@postgresql.or 968 : 128058 : heapam_scan_analyze_next_block(TableScanDesc scan, ReadStream *stream)
969 : : {
2675 andres@anarazel.de 970 : 128058 : HeapScanDesc hscan = (HeapScanDesc) scan;
971 : :
972 : : /*
973 : : * We must maintain a pin on the target page's buffer to ensure that
974 : : * concurrent activity - e.g. HOT pruning - doesn't delete tuples out from
975 : : * under us. It comes from the stream already pinned. We also choose to
976 : : * hold sharelock on the buffer throughout --- we could release and
977 : : * re-acquire sharelock for each tuple, but since we aren't doing much
978 : : * work per tuple, the extra lock traffic is probably better avoided.
979 : : */
839 tmunro@postgresql.or 980 : 128058 : hscan->rs_cbuf = read_stream_next_buffer(stream, NULL);
981 [ + + ]: 128058 : if (!BufferIsValid(hscan->rs_cbuf))
982 : 11544 : return false;
983 : :
2675 andres@anarazel.de 984 : 116514 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
985 : :
839 tmunro@postgresql.or 986 : 116514 : hscan->rs_cblock = BufferGetBlockNumber(hscan->rs_cbuf);
987 : 116514 : hscan->rs_cindex = FirstOffsetNumber;
988 : 116514 : return true;
989 : : }
990 : :
991 : : static bool
150 melanieplageman@gmai 992 : 9613556 : heapam_scan_analyze_next_tuple(TableScanDesc scan,
993 : : double *liverows, double *deadrows,
994 : : TupleTableSlot *slot)
995 : : {
2675 andres@anarazel.de 996 : 9613556 : HeapScanDesc hscan = (HeapScanDesc) scan;
997 : : Page targpage;
998 : : OffsetNumber maxoffset;
999 : : BufferHeapTupleTableSlot *hslot;
1000 : :
1001 [ - + ]: 9613556 : Assert(TTS_IS_BUFFERTUPLE(slot));
1002 : :
1003 : 9613556 : hslot = (BufferHeapTupleTableSlot *) slot;
1004 : 9613556 : targpage = BufferGetPage(hscan->rs_cbuf);
1005 : 9613556 : maxoffset = PageGetMaxOffsetNumber(targpage);
1006 : :
1007 : : /* Inner loop over all tuples on the selected page */
1008 [ + + ]: 10554145 : for (; hscan->rs_cindex <= maxoffset; hscan->rs_cindex++)
1009 : : {
1010 : : ItemId itemid;
1011 : 10437631 : HeapTuple targtuple = &hslot->base.tupdata;
1012 : 10437631 : bool sample_it = false;
1013 : : TransactionId dead_after;
1014 : :
1015 : 10437631 : itemid = PageGetItemId(targpage, hscan->rs_cindex);
1016 : :
1017 : : /*
1018 : : * We ignore unused and redirect line pointers. DEAD line pointers
1019 : : * should be counted as dead, because we need vacuum to run to get rid
1020 : : * of them. Note that this rule agrees with the way that
1021 : : * heap_page_prune_and_freeze() counts things.
1022 : : */
1023 [ + + ]: 10437631 : if (!ItemIdIsNormal(itemid))
1024 : : {
1025 [ + + ]: 435677 : if (ItemIdIsDead(itemid))
1026 : 14960 : *deadrows += 1;
1027 : 435677 : continue;
1028 : : }
1029 : :
1030 : 10001954 : ItemPointerSet(&targtuple->t_self, hscan->rs_cblock, hscan->rs_cindex);
1031 : :
1032 : 10001954 : targtuple->t_tableOid = RelationGetRelid(scan->rs_rd);
1033 : 10001954 : targtuple->t_data = (HeapTupleHeader) PageGetItem(targpage, itemid);
1034 : 10001954 : targtuple->t_len = ItemIdGetLength(itemid);
1035 : :
150 melanieplageman@gmai 1036 [ + + + + : 10001954 : switch (HeapTupleSatisfiesVacuumHorizon(targtuple,
- ]
1037 : : hscan->rs_cbuf,
1038 : : &dead_after))
1039 : : {
2675 andres@anarazel.de 1040 : 8281430 : case HEAPTUPLE_LIVE:
1041 : 8281430 : sample_it = true;
1042 : 8281430 : *liverows += 1;
1043 : 8281430 : break;
1044 : :
1045 : 157404 : case HEAPTUPLE_DEAD:
1046 : : case HEAPTUPLE_RECENTLY_DEAD:
1047 : : /* Count dead and recently-dead rows */
1048 : 157404 : *deadrows += 1;
1049 : 157404 : break;
1050 : :
1051 : 572550 : case HEAPTUPLE_INSERT_IN_PROGRESS:
1052 : :
1053 : : /*
1054 : : * Insert-in-progress rows are not counted. We assume that
1055 : : * when the inserting transaction commits or aborts, it will
1056 : : * send a stats message to increment the proper count. This
1057 : : * works right only if that transaction ends after we finish
1058 : : * analyzing the table; if things happen in the other order,
1059 : : * its stats update will be overwritten by ours. However, the
1060 : : * error will be large only if the other transaction runs long
1061 : : * enough to insert many tuples, so assuming it will finish
1062 : : * after us is the safer option.
1063 : : *
1064 : : * A special case is that the inserting transaction might be
1065 : : * our own. In this case we should count and sample the row,
1066 : : * to accommodate users who load a table and analyze it in one
1067 : : * transaction. (pgstat_report_analyze has to adjust the
1068 : : * numbers we report to the cumulative stats system to make
1069 : : * this come out right.)
1070 : : */
1071 [ + + ]: 572550 : if (TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetXmin(targtuple->t_data)))
1072 : : {
1073 : 226186 : sample_it = true;
1074 : 226186 : *liverows += 1;
1075 : : }
1076 : 572550 : break;
1077 : :
1078 : 990570 : case HEAPTUPLE_DELETE_IN_PROGRESS:
1079 : :
1080 : : /*
1081 : : * We count and sample delete-in-progress rows the same as
1082 : : * live ones, so that the stats counters come out right if the
1083 : : * deleting transaction commits after us, per the same
1084 : : * reasoning given above.
1085 : : *
1086 : : * If the delete was done by our own transaction, however, we
1087 : : * must count the row as dead to make pgstat_report_analyze's
1088 : : * stats adjustments come out right. (Note: this works out
1089 : : * properly when the row was both inserted and deleted in our
1090 : : * xact.)
1091 : : *
1092 : : * The net effect of these choices is that we act as though an
1093 : : * IN_PROGRESS transaction hasn't happened yet, except if it
1094 : : * is our own transaction, which we assume has happened.
1095 : : *
1096 : : * This approach ensures that we behave sanely if we see both
1097 : : * the pre-image and post-image rows for a row being updated
1098 : : * by a concurrent transaction: we will sample the pre-image
1099 : : * but not the post-image. We also get sane results if the
1100 : : * concurrent transaction never commits.
1101 : : */
1102 [ + + ]: 990570 : if (TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetUpdateXid(targtuple->t_data)))
2595 1103 : 1144 : *deadrows += 1;
1104 : : else
1105 : : {
2675 1106 : 989426 : sample_it = true;
2595 1107 : 989426 : *liverows += 1;
1108 : : }
2675 1109 : 990570 : break;
1110 : :
2675 andres@anarazel.de 1111 :UBC 0 : default:
1112 [ # # ]: 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1113 : : break;
1114 : : }
1115 : :
2675 andres@anarazel.de 1116 [ + + ]:CBC 10001954 : if (sample_it)
1117 : : {
1118 : 9497042 : ExecStoreBufferHeapTuple(targtuple, slot, hscan->rs_cbuf);
1119 : 9497042 : hscan->rs_cindex++;
1120 : :
1121 : : /* note that we leave the buffer locked here! */
1122 : 9497042 : return true;
1123 : : }
1124 : : }
1125 : :
1126 : : /* Now release the lock and pin on the page */
1127 : 116514 : UnlockReleaseBuffer(hscan->rs_cbuf);
1128 : 116514 : hscan->rs_cbuf = InvalidBuffer;
1129 : :
1130 : : /* also prevent old slot contents from having pin on page */
1131 : 116514 : ExecClearTuple(slot);
1132 : :
1133 : 116514 : return false;
1134 : : }
1135 : :
1136 : : static double
2678 1137 : 36372 : heapam_index_build_range_scan(Relation heapRelation,
1138 : : Relation indexRelation,
1139 : : IndexInfo *indexInfo,
1140 : : bool allow_sync,
1141 : : bool anyvisible,
1142 : : bool progress,
1143 : : BlockNumber start_blockno,
1144 : : BlockNumber numblocks,
1145 : : IndexBuildCallback callback,
1146 : : void *callback_state,
1147 : : TableScanDesc scan)
1148 : : {
1149 : : HeapScanDesc hscan;
1150 : : bool is_system_catalog;
1151 : : bool checking_uniqueness;
1152 : : HeapTuple heapTuple;
1153 : : Datum values[INDEX_MAX_KEYS];
1154 : : bool isnull[INDEX_MAX_KEYS];
1155 : : double reltuples;
1156 : : ExprState *predicate;
1157 : : TupleTableSlot *slot;
1158 : : EState *estate;
1159 : : ExprContext *econtext;
1160 : : Snapshot snapshot;
1161 : 36372 : bool need_unregister_snapshot = false;
1162 : : TransactionId OldestXmin;
2622 tgl@sss.pgh.pa.us 1163 : 36372 : BlockNumber previous_blkno = InvalidBlockNumber;
2678 andres@anarazel.de 1164 : 36372 : BlockNumber root_blkno = InvalidBlockNumber;
1165 : : OffsetNumber root_offsets[MaxHeapTuplesPerPage];
1166 : :
1167 : : /*
1168 : : * sanity checks
1169 : : */
1170 [ - + ]: 36372 : Assert(OidIsValid(indexRelation->rd_rel->relam));
1171 : :
1172 : : /* Remember if it's a system catalog */
1173 : 36372 : is_system_catalog = IsSystemRelation(heapRelation);
1174 : :
1175 : : /* See whether we're verifying uniqueness/exclusion properties */
1176 [ + + ]: 45935 : checking_uniqueness = (indexInfo->ii_Unique ||
1177 [ + + ]: 9563 : indexInfo->ii_ExclusionOps != NULL);
1178 : :
1179 : : /*
1180 : : * "Any visible" mode is not compatible with uniqueness checks; make sure
1181 : : * only one of those is requested.
1182 : : */
1183 [ + + - + ]: 36372 : Assert(!(anyvisible && checking_uniqueness));
1184 : :
1185 : : /*
1186 : : * Need an EState for evaluation of index expressions and partial-index
1187 : : * predicates. Also a slot to hold the current tuple.
1188 : : */
1189 : 36372 : estate = CreateExecutorState();
1190 [ - + ]: 36372 : econtext = GetPerTupleExprContext(estate);
1191 : 36372 : slot = table_slot_create(heapRelation, NULL);
1192 : :
1193 : : /* Arrange for econtext's scan tuple to be the tuple under test */
1194 : 36372 : econtext->ecxt_scantuple = slot;
1195 : :
1196 : : /* Set up execution state for predicate, if any. */
1197 : 36372 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
1198 : :
1199 : : /*
1200 : : * Prepare for scan of the base relation. In a normal index build, we use
1201 : : * SnapshotAny because we must retrieve all tuples and do our own time
1202 : : * qual checks (because we have to index RECENTLY_DEAD tuples). In a
1203 : : * concurrent build, or during bootstrap, we take a regular MVCC snapshot
1204 : : * and index whatever's live according to that.
1205 : : */
1206 : 36372 : OldestXmin = InvalidTransactionId;
1207 : :
1208 : : /* okay to ignore lazy VACUUMs here */
1209 [ + + + + ]: 36372 : if (!IsBootstrapProcessingMode() && !indexInfo->ii_Concurrent)
2174 1210 : 26216 : OldestXmin = GetOldestNonRemovableTransactionId(heapRelation);
1211 : :
2678 1212 [ + + ]: 36372 : if (!scan)
1213 : : {
1214 : : /*
1215 : : * Serial index build.
1216 : : *
1217 : : * Must begin our own heap scan in this case. We may also need to
1218 : : * register a snapshot whose lifetime is under our direct control.
1219 : : */
1220 [ + + ]: 36019 : if (!TransactionIdIsValid(OldestXmin))
1221 : : {
1222 : 10074 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1223 : 10074 : need_unregister_snapshot = true;
1224 : : }
1225 : : else
1226 : 25945 : snapshot = SnapshotAny;
1227 : :
1228 : 36019 : scan = table_beginscan_strat(heapRelation, /* relation */
1229 : : snapshot, /* snapshot */
1230 : : 0, /* number of keys */
1231 : : NULL, /* scan key */
1232 : : true, /* buffer access strategy OK */
1233 : : allow_sync); /* syncscan OK? */
1234 : : }
1235 : : else
1236 : : {
1237 : : /*
1238 : : * Parallel index build.
1239 : : *
1240 : : * Parallel case never registers/unregisters own snapshot. Snapshot
1241 : : * is taken from parallel heap scan, and is SnapshotAny or an MVCC
1242 : : * snapshot, based on same criteria as serial case.
1243 : : */
1244 [ - + ]: 353 : Assert(!IsBootstrapProcessingMode());
1245 [ - + ]: 353 : Assert(allow_sync);
1246 : 353 : snapshot = scan->rs_snapshot;
1247 : : }
1248 : :
1249 : 36372 : hscan = (HeapScanDesc) scan;
1250 : :
1251 : : /*
1252 : : * Must have called GetOldestNonRemovableTransactionId() if using
1253 : : * SnapshotAny. Shouldn't have for an MVCC snapshot. (It's especially
1254 : : * worth checking this for parallel builds, since ambuild routines that
1255 : : * support parallel builds must work these details out for themselves.)
1256 : : */
2174 1257 [ + + - + ]: 36372 : Assert(snapshot == SnapshotAny || IsMVCCSnapshot(snapshot));
1258 [ + + - + ]: 36372 : Assert(snapshot == SnapshotAny ? TransactionIdIsValid(OldestXmin) :
1259 : : !TransactionIdIsValid(OldestXmin));
1260 [ + + - + ]: 36372 : Assert(snapshot == SnapshotAny || !anyvisible);
1261 : :
1262 : : /* Publish number of blocks to scan */
2672 alvherre@alvh.no-ip. 1263 [ + + ]: 36372 : if (progress)
1264 : : {
1265 : : BlockNumber nblocks;
1266 : :
1267 [ + + ]: 34658 : if (hscan->rs_base.rs_parallel != NULL)
1268 : : {
1269 : : ParallelBlockTableScanDesc pbscan;
1270 : :
1271 : 137 : pbscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1272 : 137 : nblocks = pbscan->phs_nblocks;
1273 : : }
1274 : : else
1275 : 34521 : nblocks = hscan->rs_nblocks;
1276 : :
1277 : 34658 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1278 : : nblocks);
1279 : : }
1280 : :
1281 : : /* set our scan endpoints */
2678 andres@anarazel.de 1282 [ + + ]: 36372 : if (!allow_sync)
1283 : 1914 : heap_setscanlimits(scan, start_blockno, numblocks);
1284 : : else
1285 : : {
1286 : : /* syncscan can only be requested on whole relation */
1287 [ - + ]: 34458 : Assert(start_blockno == 0);
1288 [ - + ]: 34458 : Assert(numblocks == InvalidBlockNumber);
1289 : : }
1290 : :
1291 : 36372 : reltuples = 0;
1292 : :
1293 : : /*
1294 : : * Scan all tuples in the base relation.
1295 : : */
1296 [ + + ]: 11393834 : while ((heapTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1297 : : {
1298 : : bool tupleIsAlive;
1299 : :
1300 [ + + ]: 11357482 : CHECK_FOR_INTERRUPTS();
1301 : :
1302 : : /* Report scan progress, if asked to. */
2672 alvherre@alvh.no-ip. 1303 [ + + ]: 11357482 : if (progress)
1304 : : {
2622 tgl@sss.pgh.pa.us 1305 : 9237968 : BlockNumber blocks_done = heapam_scan_get_blocks_done(hscan);
1306 : :
2672 alvherre@alvh.no-ip. 1307 [ + + ]: 9237968 : if (blocks_done != previous_blkno)
1308 : : {
1309 : 117444 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1310 : : blocks_done);
1311 : 117444 : previous_blkno = blocks_done;
1312 : : }
1313 : : }
1314 : :
1315 : : /*
1316 : : * When dealing with a HOT-chain of updated tuples, we want to index
1317 : : * the values of the live tuple (if any), but index it under the TID
1318 : : * of the chain's root tuple. This approach is necessary to preserve
1319 : : * the HOT-chain structure in the heap. So we need to be able to find
1320 : : * the root item offset for every tuple that's in a HOT-chain. When
1321 : : * first reaching a new page of the relation, call
1322 : : * heap_get_root_tuples() to build a map of root item offsets on the
1323 : : * page.
1324 : : *
1325 : : * It might look unsafe to use this information across buffer
1326 : : * lock/unlock. However, we hold ShareLock on the table so no
1327 : : * ordinary insert/update/delete should occur; and we hold pin on the
1328 : : * buffer continuously while visiting the page, so no pruning
1329 : : * operation can occur either.
1330 : : *
1331 : : * In cases with only ShareUpdateExclusiveLock on the table, it's
1332 : : * possible for some HOT tuples to appear that we didn't know about
1333 : : * when we first read the page. To handle that case, we re-obtain the
1334 : : * list of root offsets when a HOT tuple points to a root item that we
1335 : : * don't know about.
1336 : : *
1337 : : * Also, although our opinions about tuple liveness could change while
1338 : : * we scan the page (due to concurrent transaction commits/aborts),
1339 : : * the chain root locations won't, so this info doesn't need to be
1340 : : * rebuilt after waiting for another transaction.
1341 : : *
1342 : : * Note the implied assumption that there is no more than one live
1343 : : * tuple per HOT-chain --- else we could create more than one index
1344 : : * entry pointing to the same root tuple.
1345 : : */
2678 andres@anarazel.de 1346 [ + + ]: 11357482 : if (hscan->rs_cblock != root_blkno)
1347 : : {
1348 : 136649 : Page page = BufferGetPage(hscan->rs_cbuf);
1349 : :
1350 : 136649 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1351 : 136649 : heap_get_root_tuples(page, root_offsets);
1352 : 136649 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1353 : :
1354 : 136649 : root_blkno = hscan->rs_cblock;
1355 : : }
1356 : :
1357 [ + + ]: 11357482 : if (snapshot == SnapshotAny)
1358 : : {
1359 : : /* do our own time qual check */
1360 : : bool indexIt;
1361 : : TransactionId xwait;
1362 : :
1363 : 8828035 : recheck:
1364 : :
1365 : : /*
1366 : : * We could possibly get away with not locking the buffer here,
1367 : : * since caller should hold ShareLock on the relation, but let's
1368 : : * be conservative about it. (This remark is still correct even
1369 : : * with HOT-pruning: our pin on the buffer prevents pruning.)
1370 : : */
1371 : 8828035 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1372 : :
1373 : : /*
1374 : : * The criteria for counting a tuple as live in this block need to
1375 : : * match what analyze.c's heapam_scan_analyze_next_tuple() does,
1376 : : * otherwise CREATE INDEX and ANALYZE may produce wildly different
1377 : : * reltuples values, e.g. when there are many recently-dead
1378 : : * tuples.
1379 : : */
1380 [ + + + + : 8828035 : switch (HeapTupleSatisfiesVacuum(heapTuple, OldestXmin,
+ - ]
1381 : : hscan->rs_cbuf))
1382 : : {
1383 : 1481 : case HEAPTUPLE_DEAD:
1384 : : /* Definitely dead, we can ignore it */
1385 : 1481 : indexIt = false;
1386 : 1481 : tupleIsAlive = false;
1387 : 1481 : break;
1388 : 6785745 : case HEAPTUPLE_LIVE:
1389 : : /* Normal case, index and unique-check it */
1390 : 6785745 : indexIt = true;
1391 : 6785745 : tupleIsAlive = true;
1392 : : /* Count it as live, too */
1393 : 6785745 : reltuples += 1;
1394 : 6785745 : break;
1395 : 154360 : case HEAPTUPLE_RECENTLY_DEAD:
1396 : :
1397 : : /*
1398 : : * If tuple is recently deleted then we must index it
1399 : : * anyway to preserve MVCC semantics. (Pre-existing
1400 : : * transactions could try to use the index after we finish
1401 : : * building it, and may need to see such tuples.)
1402 : : *
1403 : : * However, if it was HOT-updated then we must only index
1404 : : * the live tuple at the end of the HOT-chain. Since this
1405 : : * breaks semantics for pre-existing snapshots, mark the
1406 : : * index as unusable for them.
1407 : : *
1408 : : * We don't count recently-dead tuples in reltuples, even
1409 : : * if we index them; see heapam_scan_analyze_next_tuple().
1410 : : */
1411 [ + + ]: 154360 : if (HeapTupleIsHotUpdated(heapTuple))
1412 : : {
1413 : 142 : indexIt = false;
1414 : : /* mark the index as unsafe for old snapshots */
1415 : 142 : indexInfo->ii_BrokenHotChain = true;
1416 : : }
1417 : : else
1418 : 154218 : indexIt = true;
1419 : : /* In any case, exclude the tuple from unique-checking */
1420 : 154360 : tupleIsAlive = false;
1421 : 154360 : break;
1422 : 1886396 : case HEAPTUPLE_INSERT_IN_PROGRESS:
1423 : :
1424 : : /*
1425 : : * In "anyvisible" mode, this tuple is visible and we
1426 : : * don't need any further checks.
1427 : : */
1428 [ + + ]: 1886396 : if (anyvisible)
1429 : : {
1430 : 30736 : indexIt = true;
1431 : 30736 : tupleIsAlive = true;
1432 : 30736 : reltuples += 1;
1433 : 30736 : break;
1434 : : }
1435 : :
1436 : : /*
1437 : : * Since caller should hold ShareLock or better, normally
1438 : : * the only way to see this is if it was inserted earlier
1439 : : * in our own transaction. However, it can happen in
1440 : : * system catalogs, since we tend to release write lock
1441 : : * before commit there. Give a warning if neither case
1442 : : * applies.
1443 : : */
1444 : 1855660 : xwait = HeapTupleHeaderGetXmin(heapTuple->t_data);
1445 [ + + ]: 1855660 : if (!TransactionIdIsCurrentTransactionId(xwait))
1446 : : {
1447 [ - + ]: 102 : if (!is_system_catalog)
2678 andres@anarazel.de 1448 [ # # ]:UBC 0 : elog(WARNING, "concurrent insert in progress within table \"%s\"",
1449 : : RelationGetRelationName(heapRelation));
1450 : :
1451 : : /*
1452 : : * If we are performing uniqueness checks, indexing
1453 : : * such a tuple could lead to a bogus uniqueness
1454 : : * failure. In that case we wait for the inserting
1455 : : * transaction to finish and check again.
1456 : : */
2678 andres@anarazel.de 1457 [ - + ]:CBC 102 : if (checking_uniqueness)
1458 : : {
1459 : : /*
1460 : : * Must drop the lock on the buffer before we wait
1461 : : */
2678 andres@anarazel.de 1462 :UBC 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1463 : 0 : XactLockTableWait(xwait, heapRelation,
1464 : 0 : &heapTuple->t_self,
1465 : : XLTW_InsertIndexUnique);
1466 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
1467 : 0 : goto recheck;
1468 : : }
1469 : : }
1470 : : else
1471 : : {
1472 : : /*
1473 : : * For consistency with
1474 : : * heapam_scan_analyze_next_tuple(), count
1475 : : * HEAPTUPLE_INSERT_IN_PROGRESS tuples as live only
1476 : : * when inserted by our own transaction.
1477 : : */
2678 andres@anarazel.de 1478 :CBC 1855558 : reltuples += 1;
1479 : : }
1480 : :
1481 : : /*
1482 : : * We must index such tuples, since if the index build
1483 : : * commits then they're good.
1484 : : */
1485 : 1855660 : indexIt = true;
1486 : 1855660 : tupleIsAlive = true;
1487 : 1855660 : break;
1488 : 53 : case HEAPTUPLE_DELETE_IN_PROGRESS:
1489 : :
1490 : : /*
1491 : : * As with INSERT_IN_PROGRESS case, this is unexpected
1492 : : * unless it's our own deletion or a system catalog; but
1493 : : * in anyvisible mode, this tuple is visible.
1494 : : */
1495 [ - + ]: 53 : if (anyvisible)
1496 : : {
2678 andres@anarazel.de 1497 :UBC 0 : indexIt = true;
1498 : 0 : tupleIsAlive = false;
1499 : 0 : reltuples += 1;
1500 : 0 : break;
1501 : : }
1502 : :
2678 andres@anarazel.de 1503 :CBC 53 : xwait = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1504 [ + + ]: 53 : if (!TransactionIdIsCurrentTransactionId(xwait))
1505 : : {
1506 [ - + ]: 9 : if (!is_system_catalog)
2678 andres@anarazel.de 1507 [ # # ]:UBC 0 : elog(WARNING, "concurrent delete in progress within table \"%s\"",
1508 : : RelationGetRelationName(heapRelation));
1509 : :
1510 : : /*
1511 : : * If we are performing uniqueness checks, assuming
1512 : : * the tuple is dead could lead to missing a
1513 : : * uniqueness violation. In that case we wait for the
1514 : : * deleting transaction to finish and check again.
1515 : : *
1516 : : * Also, if it's a HOT-updated tuple, we should not
1517 : : * index it but rather the live tuple at the end of
1518 : : * the HOT-chain. However, the deleting transaction
1519 : : * could abort, possibly leaving this tuple as live
1520 : : * after all, in which case it has to be indexed. The
1521 : : * only way to know what to do is to wait for the
1522 : : * deleting transaction to finish and check again.
1523 : : */
2678 andres@anarazel.de 1524 [ + - - + ]:CBC 18 : if (checking_uniqueness ||
1525 : 9 : HeapTupleIsHotUpdated(heapTuple))
1526 : : {
1527 : : /*
1528 : : * Must drop the lock on the buffer before we wait
1529 : : */
2678 andres@anarazel.de 1530 :UBC 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1531 : 0 : XactLockTableWait(xwait, heapRelation,
1532 : 0 : &heapTuple->t_self,
1533 : : XLTW_InsertIndexUnique);
1534 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
1535 : 0 : goto recheck;
1536 : : }
1537 : :
1538 : : /*
1539 : : * Otherwise index it but don't check for uniqueness,
1540 : : * the same as a RECENTLY_DEAD tuple.
1541 : : */
2678 andres@anarazel.de 1542 :CBC 9 : indexIt = true;
1543 : :
1544 : : /*
1545 : : * Count HEAPTUPLE_DELETE_IN_PROGRESS tuples as live,
1546 : : * if they were not deleted by the current
1547 : : * transaction. That's what
1548 : : * heapam_scan_analyze_next_tuple() does, and we want
1549 : : * the behavior to be consistent.
1550 : : */
1551 : 9 : reltuples += 1;
1552 : : }
1553 [ - + ]: 44 : else if (HeapTupleIsHotUpdated(heapTuple))
1554 : : {
1555 : : /*
1556 : : * It's a HOT-updated tuple deleted by our own xact.
1557 : : * We can assume the deletion will commit (else the
1558 : : * index contents don't matter), so treat the same as
1559 : : * RECENTLY_DEAD HOT-updated tuples.
1560 : : */
2678 andres@anarazel.de 1561 :UBC 0 : indexIt = false;
1562 : : /* mark the index as unsafe for old snapshots */
1563 : 0 : indexInfo->ii_BrokenHotChain = true;
1564 : : }
1565 : : else
1566 : : {
1567 : : /*
1568 : : * It's a regular tuple deleted by our own xact. Index
1569 : : * it, but don't check for uniqueness nor count in
1570 : : * reltuples, the same as a RECENTLY_DEAD tuple.
1571 : : */
2678 andres@anarazel.de 1572 :CBC 44 : indexIt = true;
1573 : : }
1574 : : /* In any case, exclude the tuple from unique-checking */
1575 : 53 : tupleIsAlive = false;
1576 : 53 : break;
2678 andres@anarazel.de 1577 :UBC 0 : default:
1578 [ # # ]: 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1579 : : indexIt = tupleIsAlive = false; /* keep compiler quiet */
1580 : : break;
1581 : : }
1582 : :
2678 andres@anarazel.de 1583 :CBC 8828035 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1584 : :
1585 [ + + ]: 8828035 : if (!indexIt)
1586 : 1623 : continue;
1587 : : }
1588 : : else
1589 : : {
1590 : : /* heap_getnext did the time qual check */
1591 : 2529447 : tupleIsAlive = true;
1592 : 2529447 : reltuples += 1;
1593 : : }
1594 : :
1595 : 11355859 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
1596 : :
1597 : : /* Set up for predicate or expression evaluation */
1598 : 11355859 : ExecStoreBufferHeapTuple(heapTuple, slot, hscan->rs_cbuf);
1599 : :
1600 : : /*
1601 : : * In a partial index, discard tuples that don't satisfy the
1602 : : * predicate.
1603 : : */
1604 [ + + ]: 11355859 : if (predicate != NULL)
1605 : : {
1606 [ + + ]: 131216 : if (!ExecQual(predicate, econtext))
1607 : 72189 : continue;
1608 : : }
1609 : :
1610 : : /*
1611 : : * For the current heap tuple, extract all the attributes we use in
1612 : : * this index, and note which are null. This also performs evaluation
1613 : : * of any expressions needed.
1614 : : */
1615 : 11283670 : FormIndexDatum(indexInfo,
1616 : : slot,
1617 : : estate,
1618 : : values,
1619 : : isnull);
1620 : :
1621 : : /*
1622 : : * You'd think we should go ahead and build the index tuple here, but
1623 : : * some index AMs want to do further processing on the data first. So
1624 : : * pass the values[] and isnull[] arrays, instead.
1625 : : */
1626 : :
1627 [ + + ]: 11283650 : if (HeapTupleIsHeapOnly(heapTuple))
1628 : : {
1629 : : /*
1630 : : * For a heap-only tuple, pretend its TID is that of the root. See
1631 : : * src/backend/access/heap/README.HOT for discussion.
1632 : : */
1633 : : ItemPointerData tid;
1634 : : OffsetNumber offnum;
1635 : :
1636 : 4084 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_self);
1637 : :
1638 : : /*
1639 : : * If a HOT tuple points to a root that we don't know about,
1640 : : * obtain root items afresh. If that still fails, report it as
1641 : : * corruption.
1642 : : */
2173 alvherre@alvh.no-ip. 1643 [ - + ]: 4084 : if (root_offsets[offnum - 1] == InvalidOffsetNumber)
1644 : : {
1901 tgl@sss.pgh.pa.us 1645 :UBC 0 : Page page = BufferGetPage(hscan->rs_cbuf);
1646 : :
2173 alvherre@alvh.no-ip. 1647 : 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1648 : 0 : heap_get_root_tuples(page, root_offsets);
1649 : 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1650 : : }
1651 : :
2678 andres@anarazel.de 1652 [ + - + - :CBC 4084 : if (!OffsetNumberIsValid(root_offsets[offnum - 1]))
- + ]
2678 andres@anarazel.de 1653 [ # # ]:UBC 0 : ereport(ERROR,
1654 : : (errcode(ERRCODE_DATA_CORRUPTED),
1655 : : errmsg_internal("failed to find parent tuple for heap-only tuple at (%u,%u) in table \"%s\"",
1656 : : ItemPointerGetBlockNumber(&heapTuple->t_self),
1657 : : offnum,
1658 : : RelationGetRelationName(heapRelation))));
1659 : :
2452 andres@anarazel.de 1660 :CBC 4084 : ItemPointerSet(&tid, ItemPointerGetBlockNumber(&heapTuple->t_self),
1661 : 4084 : root_offsets[offnum - 1]);
1662 : :
1663 : : /* Call the AM's callback routine to process the tuple */
1664 : 4084 : callback(indexRelation, &tid, values, isnull, tupleIsAlive,
1665 : : callback_state);
1666 : : }
1667 : : else
1668 : : {
1669 : : /* Call the AM's callback routine to process the tuple */
1670 : 11279566 : callback(indexRelation, &heapTuple->t_self, values, isnull,
1671 : : tupleIsAlive, callback_state);
1672 : : }
1673 : : }
1674 : :
1675 : : /* Report scan progress one last time. */
2672 alvherre@alvh.no-ip. 1676 [ + + ]: 36352 : if (progress)
1677 : : {
1678 : : BlockNumber blks_done;
1679 : :
1680 [ + + ]: 34638 : if (hscan->rs_base.rs_parallel != NULL)
1681 : : {
1682 : : ParallelBlockTableScanDesc pbscan;
1683 : :
1684 : 137 : pbscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1685 : 137 : blks_done = pbscan->phs_nblocks;
1686 : : }
1687 : : else
1688 : 34501 : blks_done = hscan->rs_nblocks;
1689 : :
1690 : 34638 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1691 : : blks_done);
1692 : : }
1693 : :
2678 andres@anarazel.de 1694 : 36352 : table_endscan(scan);
1695 : :
1696 : : /* we can now forget our snapshot, if set and registered by us */
1697 [ + + ]: 36352 : if (need_unregister_snapshot)
1698 : 10058 : UnregisterSnapshot(snapshot);
1699 : :
1700 : 36352 : ExecDropSingleTupleTableSlot(slot);
1701 : :
1702 : 36352 : FreeExecutorState(estate);
1703 : :
1704 : : /* These may have been pointing to the now-gone estate */
1705 : 36352 : indexInfo->ii_ExpressionsState = NIL;
1706 : 36352 : indexInfo->ii_PredicateState = NULL;
1707 : :
1708 : 36352 : return reltuples;
1709 : : }
1710 : :
1711 : : static void
1712 : 418 : heapam_index_validate_scan(Relation heapRelation,
1713 : : Relation indexRelation,
1714 : : IndexInfo *indexInfo,
1715 : : Snapshot snapshot,
1716 : : ValidateIndexState *state)
1717 : : {
1718 : : TableScanDesc scan;
1719 : : HeapScanDesc hscan;
1720 : : HeapTuple heapTuple;
1721 : : Datum values[INDEX_MAX_KEYS];
1722 : : bool isnull[INDEX_MAX_KEYS];
1723 : : ExprState *predicate;
1724 : : TupleTableSlot *slot;
1725 : : EState *estate;
1726 : : ExprContext *econtext;
1727 : 418 : BlockNumber root_blkno = InvalidBlockNumber;
1728 : : OffsetNumber root_offsets[MaxHeapTuplesPerPage];
1729 : : bool in_index[MaxHeapTuplesPerPage];
2622 tgl@sss.pgh.pa.us 1730 : 418 : BlockNumber previous_blkno = InvalidBlockNumber;
1731 : :
1732 : : /* state variables for the merge */
2678 andres@anarazel.de 1733 : 418 : ItemPointer indexcursor = NULL;
1734 : : ItemPointerData decoded;
1735 : 418 : bool tuplesort_empty = false;
1736 : :
1737 : : /*
1738 : : * sanity checks
1739 : : */
1740 [ - + ]: 418 : Assert(OidIsValid(indexRelation->rd_rel->relam));
1741 : :
1742 : : /*
1743 : : * Need an EState for evaluation of index expressions and partial-index
1744 : : * predicates. Also a slot to hold the current tuple.
1745 : : */
1746 : 418 : estate = CreateExecutorState();
1747 [ - + ]: 418 : econtext = GetPerTupleExprContext(estate);
1748 : 418 : slot = MakeSingleTupleTableSlot(RelationGetDescr(heapRelation),
1749 : : &TTSOpsHeapTuple);
1750 : :
1751 : : /* Arrange for econtext's scan tuple to be the tuple under test */
1752 : 418 : econtext->ecxt_scantuple = slot;
1753 : :
1754 : : /* Set up execution state for predicate, if any. */
1755 : 418 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
1756 : :
1757 : : /*
1758 : : * Prepare for scan of the base relation. We need just those tuples
1759 : : * satisfying the passed-in reference snapshot. We must disable syncscan
1760 : : * here, because it's critical that we read from block zero forward to
1761 : : * match the sorted TIDs.
1762 : : */
1763 : 418 : scan = table_beginscan_strat(heapRelation, /* relation */
1764 : : snapshot, /* snapshot */
1765 : : 0, /* number of keys */
1766 : : NULL, /* scan key */
1767 : : true, /* buffer access strategy OK */
1768 : : false); /* syncscan not OK */
1769 : 418 : hscan = (HeapScanDesc) scan;
1770 : :
2672 alvherre@alvh.no-ip. 1771 : 418 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1772 : 418 : hscan->rs_nblocks);
1773 : :
1774 : : /*
1775 : : * Scan all tuples matching the snapshot.
1776 : : */
2678 andres@anarazel.de 1777 [ + + ]: 126367 : while ((heapTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1778 : : {
1779 : 125949 : ItemPointer heapcursor = &heapTuple->t_self;
1780 : : ItemPointerData rootTuple;
1781 : : OffsetNumber root_offnum;
1782 : :
1783 [ - + ]: 125949 : CHECK_FOR_INTERRUPTS();
1784 : :
1785 : 125949 : state->htups += 1;
1786 : :
2672 alvherre@alvh.no-ip. 1787 [ + + ]: 125949 : if ((previous_blkno == InvalidBlockNumber) ||
1788 [ + + ]: 125719 : (hscan->rs_cblock != previous_blkno))
1789 : : {
1790 : 2350 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1791 : 2350 : hscan->rs_cblock);
1792 : 2350 : previous_blkno = hscan->rs_cblock;
1793 : : }
1794 : :
1795 : : /*
1796 : : * As commented in table_index_build_scan, we should index heap-only
1797 : : * tuples under the TIDs of their root tuples; so when we advance onto
1798 : : * a new heap page, build a map of root item offsets on the page.
1799 : : *
1800 : : * This complicates merging against the tuplesort output: we will
1801 : : * visit the live tuples in order by their offsets, but the root
1802 : : * offsets that we need to compare against the index contents might be
1803 : : * ordered differently. So we might have to "look back" within the
1804 : : * tuplesort output, but only within the current page. We handle that
1805 : : * by keeping a bool array in_index[] showing all the
1806 : : * already-passed-over tuplesort output TIDs of the current page. We
1807 : : * clear that array here, when advancing onto a new heap page.
1808 : : */
2678 andres@anarazel.de 1809 [ + + ]: 125949 : if (hscan->rs_cblock != root_blkno)
1810 : : {
1811 : 2350 : Page page = BufferGetPage(hscan->rs_cbuf);
1812 : :
1813 : 2350 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1814 : 2350 : heap_get_root_tuples(page, root_offsets);
1815 : 2350 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1816 : :
1817 : 2350 : memset(in_index, 0, sizeof(in_index));
1818 : :
1819 : 2350 : root_blkno = hscan->rs_cblock;
1820 : : }
1821 : :
1822 : : /* Convert actual tuple TID to root TID */
1823 : 125949 : rootTuple = *heapcursor;
1824 : 125949 : root_offnum = ItemPointerGetOffsetNumber(heapcursor);
1825 : :
1826 [ + + ]: 125949 : if (HeapTupleIsHeapOnly(heapTuple))
1827 : : {
1828 : 10 : root_offnum = root_offsets[root_offnum - 1];
1829 [ + - + - : 10 : if (!OffsetNumberIsValid(root_offnum))
- + ]
2678 andres@anarazel.de 1830 [ # # ]:UBC 0 : ereport(ERROR,
1831 : : (errcode(ERRCODE_DATA_CORRUPTED),
1832 : : errmsg_internal("failed to find parent tuple for heap-only tuple at (%u,%u) in table \"%s\"",
1833 : : ItemPointerGetBlockNumber(heapcursor),
1834 : : ItemPointerGetOffsetNumber(heapcursor),
1835 : : RelationGetRelationName(heapRelation))));
2678 andres@anarazel.de 1836 :CBC 10 : ItemPointerSetOffsetNumber(&rootTuple, root_offnum);
1837 : : }
1838 : :
1839 : : /*
1840 : : * "merge" by skipping through the index tuples until we find or pass
1841 : : * the current root tuple.
1842 : : */
1843 [ + + + + ]: 287249 : while (!tuplesort_empty &&
1844 [ + + ]: 286965 : (!indexcursor ||
1845 : 286965 : ItemPointerCompare(indexcursor, &rootTuple) < 0))
1846 : : {
1847 : : Datum ts_val;
1848 : : bool ts_isnull;
1849 : :
1850 [ + + ]: 161300 : if (indexcursor)
1851 : : {
1852 : : /*
1853 : : * Remember index items seen earlier on the current heap page
1854 : : */
1855 [ + + ]: 161070 : if (ItemPointerGetBlockNumber(indexcursor) == root_blkno)
1856 : 158192 : in_index[ItemPointerGetOffsetNumber(indexcursor) - 1] = true;
1857 : : }
1858 : :
1859 : 161300 : tuplesort_empty = !tuplesort_getdatum(state->tuplesort, true,
1860 : : false, &ts_val, &ts_isnull,
1367 drowley@postgresql.o 1861 : 161300 : NULL);
2678 andres@anarazel.de 1862 [ + + - + ]: 161300 : Assert(tuplesort_empty || !ts_isnull);
1863 [ + + ]: 161300 : if (!tuplesort_empty)
1864 : : {
1865 : 161277 : itemptr_decode(&decoded, DatumGetInt64(ts_val));
1866 : 161277 : indexcursor = &decoded;
1867 : : }
1868 : : else
1869 : : {
1870 : : /* Be tidy */
1871 : 23 : indexcursor = NULL;
1872 : : }
1873 : : }
1874 : :
1875 : : /*
1876 : : * If the tuplesort has overshot *and* we didn't see a match earlier,
1877 : : * then this tuple is missing from the index, so insert it.
1878 : : */
1879 [ + + + + ]: 251844 : if ((tuplesort_empty ||
1880 : 125895 : ItemPointerCompare(indexcursor, &rootTuple) > 0) &&
1881 [ + + ]: 112 : !in_index[root_offnum - 1])
1882 : : {
1883 : 106 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
1884 : :
1885 : : /* Set up for predicate or expression evaluation */
1886 : 106 : ExecStoreHeapTuple(heapTuple, slot, false);
1887 : :
1888 : : /*
1889 : : * In a partial index, discard tuples that don't satisfy the
1890 : : * predicate.
1891 : : */
1892 [ + + ]: 106 : if (predicate != NULL)
1893 : : {
1894 [ + - ]: 32 : if (!ExecQual(predicate, econtext))
1895 : 32 : continue;
1896 : : }
1897 : :
1898 : : /*
1899 : : * For the current heap tuple, extract all the attributes we use
1900 : : * in this index, and note which are null. This also performs
1901 : : * evaluation of any expressions needed.
1902 : : */
1903 : 74 : FormIndexDatum(indexInfo,
1904 : : slot,
1905 : : estate,
1906 : : values,
1907 : : isnull);
1908 : :
1909 : : /*
1910 : : * You'd think we should go ahead and build the index tuple here,
1911 : : * but some index AMs want to do further processing on the data
1912 : : * first. So pass the values[] and isnull[] arrays, instead.
1913 : : */
1914 : :
1915 : : /*
1916 : : * If the tuple is already committed dead, you might think we
1917 : : * could suppress uniqueness checking, but this is no longer true
1918 : : * in the presence of HOT, because the insert is actually a proxy
1919 : : * for a uniqueness check on the whole HOT-chain. That is, the
1920 : : * tuple we have here could be dead because it was already
1921 : : * HOT-updated, and if so the updating transaction will not have
1922 : : * thought it should insert index entries. The index AM will
1923 : : * check the whole HOT-chain and correctly detect a conflict if
1924 : : * there is one.
1925 : : */
1926 : :
1927 : 74 : index_insert(indexRelation,
1928 : : values,
1929 : : isnull,
1930 : : &rootTuple,
1931 : : heapRelation,
1932 : 74 : indexInfo->ii_Unique ?
1933 : : UNIQUE_CHECK_YES : UNIQUE_CHECK_NO,
1934 : : false,
1935 : : indexInfo);
1936 : :
1937 : 74 : state->tups_inserted += 1;
1938 : : }
1939 : : }
1940 : :
1941 : 418 : table_endscan(scan);
1942 : :
1943 : 418 : ExecDropSingleTupleTableSlot(slot);
1944 : :
1945 : 418 : FreeExecutorState(estate);
1946 : :
1947 : : /* These may have been pointing to the now-gone estate */
1948 : 418 : indexInfo->ii_ExpressionsState = NIL;
1949 : 418 : indexInfo->ii_PredicateState = NULL;
1950 : 418 : }
1951 : :
1952 : : /*
1953 : : * Return the number of blocks that have been read by this scan since
1954 : : * starting. This is meant for progress reporting rather than be fully
1955 : : * accurate: in a parallel scan, workers can be concurrently reading blocks
1956 : : * further ahead than what we report.
1957 : : */
1958 : : static BlockNumber
2672 alvherre@alvh.no-ip. 1959 : 9237968 : heapam_scan_get_blocks_done(HeapScanDesc hscan)
1960 : : {
1961 : 9237968 : ParallelBlockTableScanDesc bpscan = NULL;
1962 : : BlockNumber startblock;
1963 : : BlockNumber blocks_done;
1964 : :
1965 [ + + ]: 9237968 : if (hscan->rs_base.rs_parallel != NULL)
1966 : : {
1967 : 1128435 : bpscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1968 : 1128435 : startblock = bpscan->phs_startblock;
1969 : : }
1970 : : else
1971 : 8109533 : startblock = hscan->rs_startblock;
1972 : :
1973 : : /*
1974 : : * Might have wrapped around the end of the relation, if startblock was
1975 : : * not zero.
1976 : : */
1977 [ + + ]: 9237968 : if (hscan->rs_cblock > startblock)
1978 : 8903780 : blocks_done = hscan->rs_cblock - startblock;
1979 : : else
1980 : : {
1981 : : BlockNumber nblocks;
1982 : :
1983 [ + + ]: 334188 : nblocks = bpscan != NULL ? bpscan->phs_nblocks : hscan->rs_nblocks;
1984 : 334188 : blocks_done = nblocks - startblock +
1985 : 334188 : hscan->rs_cblock;
1986 : : }
1987 : :
1988 : 9237968 : return blocks_done;
1989 : : }
1990 : :
1991 : :
1992 : : /* ------------------------------------------------------------------------
1993 : : * Miscellaneous callbacks for the heap AM
1994 : : * ------------------------------------------------------------------------
1995 : : */
1996 : :
1997 : : /*
1998 : : * Check to see whether the table needs a TOAST table. It does only if
1999 : : * (1) there are any toastable attributes, and (2) the maximum length
2000 : : * of a tuple could exceed TOAST_TUPLE_THRESHOLD. (We don't want to
2001 : : * create a toast table for something like "f1 varchar(20)".)
2002 : : */
2003 : : static bool
2623 rhaas@postgresql.org 2004 : 31297 : heapam_relation_needs_toast_table(Relation rel)
2005 : : {
2006 : 31297 : int32 data_length = 0;
2007 : 31297 : bool maxlength_unknown = false;
2008 : 31297 : bool has_toastable_attrs = false;
2009 : 31297 : TupleDesc tupdesc = rel->rd_att;
2010 : : int32 tuple_length;
2011 : : int i;
2012 : :
2013 [ + + ]: 122219 : for (i = 0; i < tupdesc->natts; i++)
2014 : : {
2015 : 90922 : Form_pg_attribute att = TupleDescAttr(tupdesc, i);
2016 : :
2017 [ + + ]: 90922 : if (att->attisdropped)
2018 : 789 : continue;
534 peter@eisentraut.org 2019 [ + + ]: 90133 : if (att->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
2020 : 704 : continue;
2623 rhaas@postgresql.org 2021 : 89429 : data_length = att_align_nominal(data_length, att->attalign);
2022 [ + + ]: 89429 : if (att->attlen > 0)
2023 : : {
2024 : : /* Fixed-length types are never toastable */
2025 : 66819 : data_length += att->attlen;
2026 : : }
2027 : : else
2028 : : {
2029 : 22610 : int32 maxlen = type_maximum_size(att->atttypid,
2030 : : att->atttypmod);
2031 : :
2032 [ + + ]: 22610 : if (maxlen < 0)
2033 : 20369 : maxlength_unknown = true;
2034 : : else
2035 : 2241 : data_length += maxlen;
2335 tgl@sss.pgh.pa.us 2036 [ + + ]: 22610 : if (att->attstorage != TYPSTORAGE_PLAIN)
2623 rhaas@postgresql.org 2037 : 21925 : has_toastable_attrs = true;
2038 : : }
2039 : : }
2040 [ + + ]: 31297 : if (!has_toastable_attrs)
2041 : 18439 : return false; /* nothing to toast? */
2042 [ + + ]: 12858 : if (maxlength_unknown)
2043 : 11167 : return true; /* any unlimited-length attrs? */
2044 : 1691 : tuple_length = MAXALIGN(SizeofHeapTupleHeader +
2045 : 1691 : BITMAPLEN(tupdesc->natts)) +
2046 : 1691 : MAXALIGN(data_length);
2047 : 1691 : return (tuple_length > TOAST_TUPLE_THRESHOLD);
2048 : : }
2049 : :
2050 : : /*
2051 : : * TOAST tables for heap relations are just heap relations.
2052 : : */
2053 : : static Oid
2392 2054 : 11469 : heapam_relation_toast_am(Relation rel)
2055 : : {
2056 : 11469 : return rel->rd_rel->relam;
2057 : : }
2058 : :
2059 : :
2060 : : /* ------------------------------------------------------------------------
2061 : : * Planner related callbacks for the heap AM
2062 : : * ------------------------------------------------------------------------
2063 : : */
2064 : :
2065 : : #define HEAP_OVERHEAD_BYTES_PER_TUPLE \
2066 : : (MAXALIGN(SizeofHeapTupleHeader) + sizeof(ItemIdData))
2067 : : #define HEAP_USABLE_BYTES_PER_PAGE \
2068 : : (BLCKSZ - SizeOfPageHeaderData)
2069 : :
2070 : : static void
2675 andres@anarazel.de 2071 : 347894 : heapam_estimate_rel_size(Relation rel, int32 *attr_widths,
2072 : : BlockNumber *pages, double *tuples,
2073 : : double *allvisfrac)
2074 : : {
2575 rhaas@postgresql.org 2075 : 347894 : table_block_relation_estimate_size(rel, attr_widths, pages,
2076 : : tuples, allvisfrac,
2077 : : HEAP_OVERHEAD_BYTES_PER_TUPLE,
2078 : : HEAP_USABLE_BYTES_PER_PAGE);
2675 andres@anarazel.de 2079 : 347894 : }
2080 : :
2081 : :
2082 : : /* ------------------------------------------------------------------------
2083 : : * Executor related callbacks for the heap AM
2084 : : * ------------------------------------------------------------------------
2085 : : */
2086 : :
2087 : : static bool
498 melanieplageman@gmai 2088 : 4049588 : heapam_scan_bitmap_next_tuple(TableScanDesc scan,
2089 : : TupleTableSlot *slot,
2090 : : bool *recheck,
2091 : : uint64 *lossy_pages,
2092 : : uint64 *exact_pages)
2093 : : {
556 2094 : 4049588 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
2095 : 4049588 : HeapScanDesc hscan = (HeapScanDesc) bscan;
2096 : : OffsetNumber targoffset;
2097 : : Page page;
2098 : : ItemId lp;
2099 : :
2100 : : /*
2101 : : * Out of range? If so, nothing more to look at on this page
2102 : : */
498 2103 [ + + ]: 4292982 : while (hscan->rs_cindex >= hscan->rs_ntuples)
2104 : : {
2105 : : /*
2106 : : * Returns false if the bitmap is exhausted and there are no further
2107 : : * blocks we need to scan.
2108 : : */
2109 [ + + ]: 257943 : if (!BitmapHeapScanNextBlock(scan, recheck, lossy_pages, exact_pages))
2110 : 14546 : return false;
2111 : : }
2112 : :
2674 andres@anarazel.de 2113 : 4035039 : targoffset = hscan->rs_vistuples[hscan->rs_cindex];
1348 peter@eisentraut.org 2114 : 4035039 : page = BufferGetPage(hscan->rs_cbuf);
2115 : 4035039 : lp = PageGetItemId(page, targoffset);
2674 andres@anarazel.de 2116 [ - + ]: 4035039 : Assert(ItemIdIsNormal(lp));
2117 : :
1348 peter@eisentraut.org 2118 : 4035039 : hscan->rs_ctup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2674 andres@anarazel.de 2119 : 4035039 : hscan->rs_ctup.t_len = ItemIdGetLength(lp);
2120 : 4035039 : hscan->rs_ctup.t_tableOid = scan->rs_rd->rd_id;
2121 : 4035039 : ItemPointerSet(&hscan->rs_ctup.t_self, hscan->rs_cblock, targoffset);
2122 : :
2123 [ - + - - : 4035039 : pgstat_count_heap_fetch(scan->rs_rd);
+ - ]
2124 : :
2125 : : /*
2126 : : * Set up the result slot to point to this tuple. Note that the slot
2127 : : * acquires a pin on the buffer.
2128 : : */
2129 : 4035039 : ExecStoreBufferHeapTuple(&hscan->rs_ctup,
2130 : : slot,
2131 : : hscan->rs_cbuf);
2132 : :
2133 : 4035039 : hscan->rs_cindex++;
2134 : :
2135 : 4035039 : return true;
2136 : : }
2137 : :
2138 : : static bool
2675 2139 : 8588 : heapam_scan_sample_next_block(TableScanDesc scan, SampleScanState *scanstate)
2140 : : {
2141 : 8588 : HeapScanDesc hscan = (HeapScanDesc) scan;
2142 : 8588 : TsmRoutine *tsm = scanstate->tsmroutine;
2143 : : BlockNumber blockno;
2144 : :
2145 : : /* return false immediately if relation is empty */
2146 [ - + ]: 8588 : if (hscan->rs_nblocks == 0)
2675 andres@anarazel.de 2147 :UBC 0 : return false;
2148 : :
2149 : : /* release previous scan buffer, if any */
843 drowley@postgresql.o 2150 [ + + ]:CBC 8588 : if (BufferIsValid(hscan->rs_cbuf))
2151 : : {
2152 : 8475 : ReleaseBuffer(hscan->rs_cbuf);
2153 : 8475 : hscan->rs_cbuf = InvalidBuffer;
2154 : : }
2155 : :
2156 [ + + ]: 8588 : if (tsm->NextSampleBlock)
2157 : 2944 : blockno = tsm->NextSampleBlock(scanstate, hscan->rs_nblocks);
2158 : : else
2159 : : {
2160 : : /* scanning table sequentially */
2161 : :
2675 andres@anarazel.de 2162 [ + + ]: 5644 : if (hscan->rs_cblock == InvalidBlockNumber)
2163 : : {
2164 [ - + ]: 52 : Assert(!hscan->rs_inited);
2165 : 52 : blockno = hscan->rs_startblock;
2166 : : }
2167 : : else
2168 : : {
2169 [ - + ]: 5592 : Assert(hscan->rs_inited);
2170 : :
2171 : 5592 : blockno = hscan->rs_cblock + 1;
2172 : :
2173 [ + + ]: 5592 : if (blockno >= hscan->rs_nblocks)
2174 : : {
2175 : : /* wrap to beginning of rel, might not have started at 0 */
2176 : 52 : blockno = 0;
2177 : : }
2178 : :
2179 : : /*
2180 : : * Report our new scan position for synchronization purposes.
2181 : : *
2182 : : * Note: we do this before checking for end of scan so that the
2183 : : * final state of the position hint is back at the start of the
2184 : : * rel. That's not strictly necessary, but otherwise when you run
2185 : : * the same query multiple times the starting position would shift
2186 : : * a little bit backwards on every invocation, which is confusing.
2187 : : * We don't guarantee any specific ordering in general, though.
2188 : : */
2625 2189 [ - + ]: 5592 : if (scan->rs_flags & SO_ALLOW_SYNC)
2675 andres@anarazel.de 2190 :UBC 0 : ss_report_location(scan->rs_rd, blockno);
2191 : :
2675 andres@anarazel.de 2192 [ + + ]:CBC 5592 : if (blockno == hscan->rs_startblock)
2193 : : {
2194 : 52 : blockno = InvalidBlockNumber;
2195 : : }
2196 : : }
2197 : : }
2198 : :
843 drowley@postgresql.o 2199 : 8588 : hscan->rs_cblock = blockno;
2200 : :
2675 andres@anarazel.de 2201 [ + + ]: 8588 : if (!BlockNumberIsValid(blockno))
2202 : : {
2203 : 109 : hscan->rs_inited = false;
2204 : 109 : return false;
2205 : : }
2206 : :
843 drowley@postgresql.o 2207 [ - + ]: 8479 : Assert(hscan->rs_cblock < hscan->rs_nblocks);
2208 : :
2209 : : /*
2210 : : * Be sure to check for interrupts at least once per page. Checks at
2211 : : * higher code levels won't be able to stop a sample scan that encounters
2212 : : * many pages' worth of consecutive dead tuples.
2213 : : */
2214 [ - + ]: 8479 : CHECK_FOR_INTERRUPTS();
2215 : :
2216 : : /* Read page using selected strategy */
2217 : 8479 : hscan->rs_cbuf = ReadBufferExtended(hscan->rs_base.rs_rd, MAIN_FORKNUM,
2218 : : blockno, RBM_NORMAL, hscan->rs_strategy);
2219 : :
2220 : : /* in pagemode, prune the page and determine visible tuple offsets */
2221 [ + + ]: 8479 : if (hscan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
2222 : 5687 : heap_prepare_pagescan(scan);
2223 : :
2224 : 8479 : hscan->rs_inited = true;
2675 andres@anarazel.de 2225 : 8479 : return true;
2226 : : }
2227 : :
2228 : : static bool
2229 : 169183 : heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
2230 : : TupleTableSlot *slot)
2231 : : {
2232 : 169183 : HeapScanDesc hscan = (HeapScanDesc) scan;
2233 : 169183 : TsmRoutine *tsm = scanstate->tsmroutine;
2234 : 169183 : BlockNumber blockno = hscan->rs_cblock;
2625 2235 : 169183 : bool pagemode = (scan->rs_flags & SO_ALLOW_PAGEMODE) != 0;
2236 : :
2237 : : Page page;
2238 : : bool all_visible;
2239 : : OffsetNumber maxoffset;
2240 : :
2241 : : /*
2242 : : * When not using pagemode, we must lock the buffer during tuple
2243 : : * visibility checks.
2244 : : */
2675 2245 [ + + ]: 169183 : if (!pagemode)
2246 : 2796 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
2247 : :
331 peter@eisentraut.org 2248 : 169183 : page = BufferGetPage(hscan->rs_cbuf);
2675 andres@anarazel.de 2249 [ + + ]: 337844 : all_visible = PageIsAllVisible(page) &&
2250 [ + - ]: 168661 : !scan->rs_snapshot->takenDuringRecovery;
2251 : 169183 : maxoffset = PageGetMaxOffsetNumber(page);
2252 : :
2253 : : for (;;)
2675 andres@anarazel.de 2254 :UBC 0 : {
2255 : : OffsetNumber tupoffset;
2256 : :
2675 andres@anarazel.de 2257 [ - + ]:CBC 169183 : CHECK_FOR_INTERRUPTS();
2258 : :
2259 : : /* Ask the tablesample method which tuples to check on this page. */
2260 : 169183 : tupoffset = tsm->NextSampleTuple(scanstate,
2261 : : blockno,
2262 : : maxoffset);
2263 : :
2264 [ + + + - : 169183 : if (OffsetNumberIsValid(tupoffset))
+ + ]
2265 : : {
2266 : : ItemId itemid;
2267 : : bool visible;
2268 : 160708 : HeapTuple tuple = &(hscan->rs_ctup);
2269 : :
2270 : : /* Skip invalid tuple pointers. */
2271 : 160708 : itemid = PageGetItemId(page, tupoffset);
2272 [ - + ]: 160708 : if (!ItemIdIsNormal(itemid))
2675 andres@anarazel.de 2273 :UBC 0 : continue;
2274 : :
2675 andres@anarazel.de 2275 :CBC 160708 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, itemid);
2276 : 160708 : tuple->t_len = ItemIdGetLength(itemid);
2277 : 160708 : ItemPointerSet(&(tuple->t_self), blockno, tupoffset);
2278 : :
2279 : :
2280 [ + + ]: 160708 : if (all_visible)
2281 : 160337 : visible = true;
2282 : : else
2283 : 371 : visible = SampleHeapTupleVisible(scan, hscan->rs_cbuf,
2284 : : tuple, tupoffset);
2285 : :
2286 : : /* in pagemode, heap_prepare_pagescan did this for us */
2287 [ + + ]: 160708 : if (!pagemode)
2371 tmunro@postgresql.or 2288 : 4 : HeapCheckForSerializableConflictOut(visible, scan->rs_rd, tuple,
2289 : : hscan->rs_cbuf, scan->rs_snapshot);
2290 : :
2291 : : /* Try next tuple from same page. */
2675 andres@anarazel.de 2292 [ - + ]: 160708 : if (!visible)
2675 andres@anarazel.de 2293 :UBC 0 : continue;
2294 : :
2295 : : /* Found visible tuple, return it. */
2675 andres@anarazel.de 2296 [ + + ]:CBC 160708 : if (!pagemode)
2297 : 4 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
2298 : :
2299 : 160708 : ExecStoreBufferHeapTuple(tuple, slot, hscan->rs_cbuf);
2300 : :
2301 : : /* Count successfully-fetched tuples as heap fetches */
2302 [ - + - - : 160708 : pgstat_count_heap_getnext(scan->rs_rd);
+ - ]
2303 : :
2304 : 160708 : return true;
2305 : : }
2306 : : else
2307 : : {
2308 : : /*
2309 : : * If we get here, it means we've exhausted the items on this page
2310 : : * and it's time to move to the next.
2311 : : */
2312 [ + + ]: 8475 : if (!pagemode)
2313 : 2792 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
2314 : :
2315 : 8475 : ExecClearTuple(slot);
2316 : 8475 : return false;
2317 : : }
2318 : : }
2319 : :
2320 : : Assert(0);
2321 : : }
2322 : :
2323 : :
2324 : : /* ----------------------------------------------------------------------------
2325 : : * Helper functions for the above.
2326 : : * ----------------------------------------------------------------------------
2327 : : */
2328 : :
2329 : : /*
2330 : : * Reconstruct and rewrite the given tuple
2331 : : *
2332 : : * We cannot simply copy the tuple as-is, for several reasons:
2333 : : *
2334 : : * 1. We'd like to squeeze out the values of any dropped columns, both
2335 : : * to save space and to ensure we have no corner-case failures. (It's
2336 : : * possible for example that the new table hasn't got a TOAST table
2337 : : * and so is unable to store any large values of dropped cols.)
2338 : : *
2339 : : * 2. The tuple might not even be legal for the new table; this is
2340 : : * currently only known to happen as an after-effect of ALTER TABLE
2341 : : * SET WITHOUT OIDS.
2342 : : *
2343 : : * So, we must reconstruct the tuple from component Datums.
2344 : : */
2345 : : static void
2346 : 474721 : reform_and_rewrite_tuple(HeapTuple tuple,
2347 : : Relation OldHeap, Relation NewHeap,
2348 : : Datum *values, bool *isnull, RewriteState rwstate)
2349 : : {
2350 : : HeapTuple newtuple;
2351 : :
111 alvherre@kurilemu.de 2352 : 474721 : newtuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
2353 : :
2354 : : /* The heap rewrite module does the rest */
2355 : 474721 : rewrite_heap_tuple(rwstate, tuple, newtuple);
2356 : :
2357 : 474721 : heap_freetuple(newtuple);
2358 : 474721 : }
2359 : :
2360 : : /*
2361 : : * Insert tuple when processing REPACK CONCURRENTLY.
2362 : : *
2363 : : * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
2364 : : * difficult to do the same in the catch-up phase (as the logical
2365 : : * decoding does not provide us with sufficient visibility
2366 : : * information). Thus we must use heap_insert() both during the
2367 : : * catch-up and here.
2368 : : *
2369 : : * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
2370 : : * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
2371 : : * this relation, so no logical replication subscription should need the data.
2372 : : *
2373 : : * BulkInsertState is used because many tuples are inserted in the typical
2374 : : * case.
2375 : : */
2376 : : static void
2377 : 31 : heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
2378 : : Datum *values, bool *isnull, BulkInsertState bistate)
2379 : : {
2380 : : HeapTuple newtuple;
2381 : :
2382 : 31 : newtuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
2383 : :
2384 : 31 : heap_insert(NewHeap, newtuple, GetCurrentCommandId(true),
2385 : : HEAP_INSERT_NO_LOGICAL, bistate);
2386 : :
2387 : 31 : heap_freetuple(newtuple);
2388 : 31 : }
2389 : :
2390 : : /*
2391 : : * Subroutine for reform_and_rewrite_tuple and heap_insert_for_repack.
2392 : : *
2393 : : * Deform the given tuple, set values of dropped columns to NULL, and fill in
2394 : : * any values from attmissingval; then form a new tuple and return it. If no
2395 : : * attributes need to be changed, a copy of the original tuple is returned.
2396 : : * Caller is responsible for freeing the returned tuple.
2397 : : *
2398 : : * XXX this coding assumes that both relations have the same tupledesc.
2399 : : */
2400 : : static HeapTuple
2401 : 474752 : reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
2402 : : Datum *values, bool *isnull)
2403 : : {
2675 andres@anarazel.de 2404 : 474752 : TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
2405 : 474752 : TupleDesc newTupDesc = RelationGetDescr(NewHeap);
111 alvherre@kurilemu.de 2406 : 474752 : bool needs_reform = false;
2407 : :
2408 : : /*
2409 : : * A short tuple might require values from attmissing val, so activate the
2410 : : * coding unconditionally in that case. The value might legitimally be
2411 : : * NULL otherwise, so this is slightly wasteful, but it probably beats
2412 : : * having to test each attribute for presence of attmissingval each time.
2413 : : */
82 2414 [ + + ]: 474752 : if (HeapTupleHeaderGetNatts(tuple->t_data) < newTupDesc->natts)
2415 : 41 : needs_reform = true;
2416 : :
2417 : : /*
2418 : : * If the column has been dropped but a value is still present, we can
2419 : : * optimize storage now by getting rid of it.
2420 : : */
2421 [ + + ]: 474752 : if (!needs_reform)
2422 : : {
2423 [ + + ]: 3957807 : for (int i = 0; i < newTupDesc->natts; i++)
2424 : : {
2425 [ + + ]: 3483106 : if (TupleDescCompactAttr(newTupDesc, i)->attisdropped &&
2426 [ + + ]: 30 : !heap_attisnull(tuple, i + 1, newTupDesc))
2427 : : {
2428 : 10 : needs_reform = true;
2429 : 10 : break;
2430 : : }
2431 : : }
2432 : : }
2433 : :
2434 : : /* Skip work if no changes are needed */
111 2435 [ + + ]: 474752 : if (!needs_reform)
2436 : 474701 : return heap_copytuple(tuple);
2437 : :
2675 andres@anarazel.de 2438 : 51 : heap_deform_tuple(tuple, oldTupDesc, values, isnull);
2439 : :
111 alvherre@kurilemu.de 2440 [ + + ]: 255 : for (int i = 0; i < newTupDesc->natts; i++)
2441 : : {
583 drowley@postgresql.o 2442 [ + + ]: 204 : if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
2675 andres@anarazel.de 2443 : 20 : isnull[i] = true;
2444 : : }
2445 : :
111 alvherre@kurilemu.de 2446 : 51 : return heap_form_tuple(newTupDesc, values, isnull);
2447 : : }
2448 : :
2449 : : /*
2450 : : * Check visibility of the tuple.
2451 : : */
2452 : : static bool
2675 andres@anarazel.de 2453 : 371 : SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
2454 : : HeapTuple tuple,
2455 : : OffsetNumber tupoffset)
2456 : : {
2457 : 371 : HeapScanDesc hscan = (HeapScanDesc) scan;
2458 : :
2625 2459 [ + + ]: 371 : if (scan->rs_flags & SO_ALLOW_PAGEMODE)
2460 : : {
583 melanieplageman@gmai 2461 : 367 : uint32 start = 0,
2462 : 367 : end = hscan->rs_ntuples;
2463 : :
2464 : : /*
2465 : : * In pageatatime mode, heap_prepare_pagescan() already did visibility
2466 : : * checks, so just look at the info it left in rs_vistuples[].
2467 : : *
2468 : : * We use a binary search over the known-sorted array. Note: we could
2469 : : * save some effort if we insisted that NextSampleTuple select tuples
2470 : : * in increasing order, but it's not clear that there would be enough
2471 : : * gain to justify the restriction.
2472 : : */
2473 [ + - ]: 685 : while (start < end)
2474 : : {
2475 : 685 : uint32 mid = start + (end - start) / 2;
2675 andres@anarazel.de 2476 : 685 : OffsetNumber curoffset = hscan->rs_vistuples[mid];
2477 : :
2478 [ + + ]: 685 : if (tupoffset == curoffset)
2479 : 367 : return true;
2480 [ + + ]: 318 : else if (tupoffset < curoffset)
583 melanieplageman@gmai 2481 : 164 : end = mid;
2482 : : else
2675 andres@anarazel.de 2483 : 154 : start = mid + 1;
2484 : : }
2485 : :
2675 andres@anarazel.de 2486 :UBC 0 : return false;
2487 : : }
2488 : : else
2489 : : {
2490 : : /* Otherwise, we have to check the tuple individually. */
2675 andres@anarazel.de 2491 :CBC 4 : return HeapTupleSatisfiesVisibility(tuple, scan->rs_snapshot,
2492 : : buffer);
2493 : : }
2494 : : }
2495 : :
2496 : : /*
2497 : : * Helper function get the next block of a bitmap heap scan. Returns true when
2498 : : * it got the next block and saved it in the scan descriptor and false when
2499 : : * the bitmap and or relation are exhausted.
2500 : : */
2501 : : static bool
498 melanieplageman@gmai 2502 : 257943 : BitmapHeapScanNextBlock(TableScanDesc scan,
2503 : : bool *recheck,
2504 : : uint64 *lossy_pages, uint64 *exact_pages)
2505 : : {
2506 : 257943 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
2507 : 257943 : HeapScanDesc hscan = (HeapScanDesc) bscan;
2508 : : BlockNumber block;
2509 : : void *per_buffer_data;
2510 : : Buffer buffer;
2511 : : Snapshot snapshot;
2512 : : int ntup;
2513 : : TBMIterateResult *tbmres;
2514 : : OffsetNumber offsets[TBM_MAX_TUPLES_PER_PAGE];
2515 : 257943 : int noffsets = -1;
2516 : :
2517 [ - + ]: 257943 : Assert(scan->rs_flags & SO_TYPE_BITMAPSCAN);
2518 [ - + ]: 257943 : Assert(hscan->rs_read_stream);
2519 : :
2520 : 257943 : hscan->rs_cindex = 0;
2521 : 257943 : hscan->rs_ntuples = 0;
2522 : :
2523 : : /* Release buffer containing previous block. */
2524 [ + + ]: 257943 : if (BufferIsValid(hscan->rs_cbuf))
2525 : : {
2526 : 243131 : ReleaseBuffer(hscan->rs_cbuf);
2527 : 243131 : hscan->rs_cbuf = InvalidBuffer;
2528 : : }
2529 : :
2530 : 257943 : hscan->rs_cbuf = read_stream_next_buffer(hscan->rs_read_stream,
2531 : : &per_buffer_data);
2532 : :
2533 [ + + ]: 257943 : if (BufferIsInvalid(hscan->rs_cbuf))
2534 : : {
2535 : : /* the bitmap is exhausted */
480 andres@anarazel.de 2536 : 14546 : return false;
2537 : : }
2538 : :
498 melanieplageman@gmai 2539 [ - + ]: 243397 : Assert(per_buffer_data);
2540 : :
2541 : 243397 : tbmres = per_buffer_data;
2542 : :
2543 [ - + ]: 243397 : Assert(BlockNumberIsValid(tbmres->blockno));
2544 [ - + ]: 243397 : Assert(BufferGetBlockNumber(hscan->rs_cbuf) == tbmres->blockno);
2545 : :
2546 : : /* Exact pages need their tuple offsets extracted. */
2547 [ + + ]: 243397 : if (!tbmres->lossy)
2548 : 136462 : noffsets = tbm_extract_page_tuple(tbmres, offsets,
2549 : : TBM_MAX_TUPLES_PER_PAGE);
2550 : :
2551 : 243397 : *recheck = tbmres->recheck;
2552 : :
2553 : 243397 : block = hscan->rs_cblock = tbmres->blockno;
2554 : 243397 : buffer = hscan->rs_cbuf;
2555 : 243397 : snapshot = scan->rs_snapshot;
2556 : :
2557 : 243397 : ntup = 0;
2558 : :
2559 : : /*
2560 : : * Prune and repair fragmentation for the whole page, if possible.
2561 : : */
118 2562 : 243397 : heap_page_prune_opt(scan->rs_rd, buffer, &hscan->rs_vmbuffer,
2563 : 243397 : scan->rs_flags & SO_HINT_REL_READ_ONLY);
2564 : :
2565 : : /*
2566 : : * We must hold share lock on the buffer content while examining tuple
2567 : : * visibility. Afterwards, however, the tuples we have found to be
2568 : : * visible are guaranteed good as long as we hold the buffer pin.
2569 : : */
498 2570 : 243397 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
2571 : :
2572 : : /*
2573 : : * We need two separate strategies for lossy and non-lossy cases.
2574 : : */
2575 [ + + ]: 243397 : if (!tbmres->lossy)
2576 : : {
2577 : : /*
2578 : : * Bitmap is non-lossy, so we just look through the offsets listed in
2579 : : * tbmres; but we have to follow any HOT chain starting at each such
2580 : : * offset.
2581 : : */
2582 : : int curslot;
2583 : :
2584 : : /* We must have extracted the tuple offsets by now */
2585 [ - + ]: 136462 : Assert(noffsets > -1);
2586 : :
2587 [ + + ]: 3506533 : for (curslot = 0; curslot < noffsets; curslot++)
2588 : : {
2589 : 3370074 : OffsetNumber offnum = offsets[curslot];
2590 : : ItemPointerData tid;
2591 : : HeapTupleData heapTuple;
2592 : :
2593 : 3370074 : ItemPointerSet(&tid, block, offnum);
2594 [ + + ]: 3370074 : if (heap_hot_search_buffer(&tid, scan->rs_rd, buffer, snapshot,
2595 : : &heapTuple, NULL, true))
2596 : 3220585 : hscan->rs_vistuples[ntup++] = ItemPointerGetOffsetNumber(&tid);
2597 : : }
2598 : : }
2599 : : else
2600 : : {
2601 : : /*
2602 : : * Bitmap is lossy, so we must examine each line pointer on the page.
2603 : : * But we can ignore HOT chains, since we'll check each tuple anyway.
2604 : : */
2605 : 106935 : Page page = BufferGetPage(buffer);
2606 : 106935 : OffsetNumber maxoff = PageGetMaxOffsetNumber(page);
2607 : : OffsetNumber offnum;
2608 : :
2609 [ + + ]: 923265 : for (offnum = FirstOffsetNumber; offnum <= maxoff; offnum = OffsetNumberNext(offnum))
2610 : : {
2611 : : ItemId lp;
2612 : : HeapTupleData loctup;
2613 : : bool valid;
2614 : :
2615 : 816330 : lp = PageGetItemId(page, offnum);
2616 [ - + ]: 816330 : if (!ItemIdIsNormal(lp))
498 melanieplageman@gmai 2617 :UBC 0 : continue;
498 melanieplageman@gmai 2618 :CBC 816330 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2619 : 816330 : loctup.t_len = ItemIdGetLength(lp);
2620 : 816330 : loctup.t_tableOid = scan->rs_rd->rd_id;
2621 : 816330 : ItemPointerSet(&loctup.t_self, block, offnum);
2622 : 816330 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
2623 [ + + ]: 816330 : if (valid)
2624 : : {
2625 : 816246 : hscan->rs_vistuples[ntup++] = offnum;
2626 : 816246 : PredicateLockTID(scan->rs_rd, &loctup.t_self, snapshot,
2627 : 816246 : HeapTupleHeaderGetXmin(loctup.t_data));
2628 : : }
2629 : 816330 : HeapCheckForSerializableConflictOut(valid, scan->rs_rd, &loctup,
2630 : : buffer, snapshot);
2631 : : }
2632 : : }
2633 : :
2634 : 243394 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2635 : :
2636 [ - + ]: 243394 : Assert(ntup <= MaxHeapTuplesPerPage);
2637 : 243394 : hscan->rs_ntuples = ntup;
2638 : :
2639 [ + + ]: 243394 : if (tbmres->lossy)
2640 : 106935 : (*lossy_pages)++;
2641 : : else
2642 : 136459 : (*exact_pages)++;
2643 : :
2644 : : /*
2645 : : * Return true to indicate that a valid block was found and the bitmap is
2646 : : * not exhausted. If there are no visible tuples on this page,
2647 : : * hscan->rs_ntuples will be 0 and heapam_scan_bitmap_next_tuple() will
2648 : : * return false returning control to this function to advance to the next
2649 : : * block in the bitmap.
2650 : : */
2651 : 243394 : return true;
2652 : : }
2653 : :
2654 : : /* ------------------------------------------------------------------------
2655 : : * Definition of the heap table access method.
2656 : : * ------------------------------------------------------------------------
2657 : : */
2658 : :
2659 : : static const TableAmRoutine heapam_methods = {
2660 : : .type = T_TableAmRoutine,
2661 : :
2662 : : .slot_callbacks = heapam_slot_callbacks,
2663 : :
2664 : : .scan_begin = heap_beginscan,
2665 : : .scan_end = heap_endscan,
2666 : : .scan_rescan = heap_rescan,
2667 : : .scan_getnextslot = heap_getnextslot,
2668 : :
2669 : : .scan_set_tidrange = heap_set_tidrange,
2670 : : .scan_getnextslot_tidrange = heap_getnextslot_tidrange,
2671 : :
2672 : : .parallelscan_estimate = table_block_parallelscan_estimate,
2673 : : .parallelscan_initialize = table_block_parallelscan_initialize,
2674 : : .parallelscan_reinitialize = table_block_parallelscan_reinitialize,
2675 : :
2676 : : .index_fetch_begin = heapam_index_fetch_begin,
2677 : : .index_fetch_reset = heapam_index_fetch_reset,
2678 : : .index_fetch_end = heapam_index_fetch_end,
2679 : : .index_fetch_tuple = heapam_index_fetch_tuple,
2680 : :
2681 : : .tuple_insert = heapam_tuple_insert,
2682 : : .tuple_insert_speculative = heapam_tuple_insert_speculative,
2683 : : .tuple_complete_speculative = heapam_tuple_complete_speculative,
2684 : : .multi_insert = heap_multi_insert,
2685 : : .tuple_delete = heapam_tuple_delete,
2686 : : .tuple_update = heapam_tuple_update,
2687 : : .tuple_lock = heapam_tuple_lock,
2688 : :
2689 : : .tuple_fetch_row_version = heapam_fetch_row_version,
2690 : : .tuple_get_latest_tid = heap_get_latest_tid,
2691 : : .tuple_tid_valid = heapam_tuple_tid_valid,
2692 : : .tuple_satisfies_snapshot = heapam_tuple_satisfies_snapshot,
2693 : : .index_delete_tuples = heap_index_delete_tuples,
2694 : :
2695 : : .relation_set_new_filelocator = heapam_relation_set_new_filelocator,
2696 : : .relation_nontransactional_truncate = heapam_relation_nontransactional_truncate,
2697 : : .relation_copy_data = heapam_relation_copy_data,
2698 : : .relation_copy_for_cluster = heapam_relation_copy_for_cluster,
2699 : : .relation_vacuum = heap_vacuum_rel,
2700 : : .scan_analyze_next_block = heapam_scan_analyze_next_block,
2701 : : .scan_analyze_next_tuple = heapam_scan_analyze_next_tuple,
2702 : : .index_build_range_scan = heapam_index_build_range_scan,
2703 : : .index_validate_scan = heapam_index_validate_scan,
2704 : :
2705 : : .relation_size = table_block_relation_size,
2706 : : .relation_needs_toast_table = heapam_relation_needs_toast_table,
2707 : : .relation_toast_am = heapam_relation_toast_am,
2708 : : .relation_fetch_toast_slice = heap_fetch_toast_slice,
2709 : :
2710 : : .relation_estimate_size = heapam_estimate_rel_size,
2711 : :
2712 : : .scan_bitmap_next_tuple = heapam_scan_bitmap_next_tuple,
2713 : : .scan_sample_next_block = heapam_scan_sample_next_block,
2714 : : .scan_sample_next_tuple = heapam_scan_sample_next_tuple
2715 : : };
2716 : :
2717 : :
2718 : : const TableAmRoutine *
2699 andres@anarazel.de 2719 : 11845875 : GetHeapamTableAmRoutine(void)
2720 : : {
2721 : 11845875 : return &heapam_methods;
2722 : : }
2723 : :
2724 : : Datum
2725 : 1365283 : heap_tableam_handler(PG_FUNCTION_ARGS)
2726 : : {
2727 : 1365283 : PG_RETURN_POINTER(&heapam_methods);
2728 : : }
|
