Line data Source code
1 : /*----------------------------------------------------------------------
2 : *
3 : * tableam.c
4 : * Table access method routines too big to be inline functions.
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/table/tableam.c
12 : *
13 : * NOTES
14 : * Note that most functions in here are documented in tableam.h, rather than
15 : * here. That's because there's a lot of inline functions in tableam.h and
16 : * it'd be harder to understand if one constantly had to switch between files.
17 : *
18 : *----------------------------------------------------------------------
19 : */
20 : #include "postgres.h"
21 :
22 : #include <math.h>
23 :
24 : #include "access/syncscan.h"
25 : #include "access/tableam.h"
26 : #include "access/xact.h"
27 : #include "optimizer/optimizer.h"
28 : #include "optimizer/plancat.h"
29 : #include "port/pg_bitutils.h"
30 : #include "storage/bufmgr.h"
31 : #include "storage/shmem.h"
32 : #include "storage/smgr.h"
33 :
34 : /*
35 : * Constants to control the behavior of block allocation to parallel workers
36 : * during a parallel seqscan. Technically these values do not need to be
37 : * powers of 2, but having them as powers of 2 makes the math more optimal
38 : * and makes the ramp-down stepping more even.
39 : */
40 :
41 : /* The number of I/O chunks we try to break a parallel seqscan down into */
42 : #define PARALLEL_SEQSCAN_NCHUNKS 2048
43 : /* Ramp down size of allocations when we've only this number of chunks left */
44 : #define PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS 64
45 : /* Cap the size of parallel I/O chunks to this number of blocks */
46 : #define PARALLEL_SEQSCAN_MAX_CHUNK_SIZE 8192
47 :
48 : /* GUC variables */
49 : char *default_table_access_method = DEFAULT_TABLE_ACCESS_METHOD;
50 : bool synchronize_seqscans = true;
51 :
52 :
53 : /* ----------------------------------------------------------------------------
54 : * Slot functions.
55 : * ----------------------------------------------------------------------------
56 : */
57 :
58 : const TupleTableSlotOps *
59 14633084 : table_slot_callbacks(Relation relation)
60 : {
61 : const TupleTableSlotOps *tts_cb;
62 :
63 14633084 : if (relation->rd_tableam)
64 14628999 : tts_cb = relation->rd_tableam->slot_callbacks(relation);
65 4085 : else if (relation->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
66 : {
67 : /*
68 : * Historically FDWs expect to store heap tuples in slots. Continue
69 : * handing them one, to make it less painful to adapt FDWs to new
70 : * versions. The cost of a heap slot over a virtual slot is pretty
71 : * small.
72 : */
73 222 : tts_cb = &TTSOpsHeapTuple;
74 : }
75 : else
76 : {
77 : /*
78 : * These need to be supported, as some parts of the code (like COPY)
79 : * need to create slots for such relations too. It seems better to
80 : * centralize the knowledge that a heap slot is the right thing in
81 : * that case here.
82 : */
83 : Assert(relation->rd_rel->relkind == RELKIND_VIEW ||
84 : relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE);
85 3863 : tts_cb = &TTSOpsVirtual;
86 : }
87 :
88 14633084 : return tts_cb;
89 : }
90 :
91 : TupleTableSlot *
92 14375714 : table_slot_create(Relation relation, List **reglist)
93 : {
94 : const TupleTableSlotOps *tts_cb;
95 : TupleTableSlot *slot;
96 :
97 14375714 : tts_cb = table_slot_callbacks(relation);
98 14375714 : slot = MakeSingleTupleTableSlot(RelationGetDescr(relation), tts_cb);
99 :
100 14375714 : if (reglist)
101 142500 : *reglist = lappend(*reglist, slot);
102 :
103 14375714 : return slot;
104 : }
105 :
106 :
107 : /* ----------------------------------------------------------------------------
108 : * Table scan functions.
109 : * ----------------------------------------------------------------------------
110 : */
111 :
112 : TableScanDesc
113 39367 : table_beginscan_catalog(Relation relation, int nkeys, ScanKeyData *key)
114 : {
115 39367 : uint32 flags = SO_TYPE_SEQSCAN |
116 : SO_ALLOW_STRAT | SO_ALLOW_SYNC | SO_ALLOW_PAGEMODE | SO_TEMP_SNAPSHOT;
117 39367 : Oid relid = RelationGetRelid(relation);
118 39367 : Snapshot snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
119 :
120 39367 : return table_beginscan_common(relation, snapshot, nkeys, key,
121 : NULL, flags);
122 : }
123 :
124 :
125 : /* ----------------------------------------------------------------------------
126 : * Parallel table scan related functions.
127 : * ----------------------------------------------------------------------------
128 : */
129 :
130 : Size
131 578 : table_parallelscan_estimate(Relation rel, Snapshot snapshot)
132 : {
133 578 : Size sz = 0;
134 :
135 578 : if (IsMVCCSnapshot(snapshot))
136 483 : sz = add_size(sz, EstimateSnapshotSpace(snapshot));
137 : else
138 : Assert(snapshot == SnapshotAny);
139 :
140 578 : sz = add_size(sz, rel->rd_tableam->parallelscan_estimate(rel));
141 :
142 578 : return sz;
143 : }
144 :
145 : void
146 578 : table_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan,
147 : Snapshot snapshot)
148 : {
149 578 : Size snapshot_off = rel->rd_tableam->parallelscan_initialize(rel, pscan);
150 :
151 578 : pscan->phs_snapshot_off = snapshot_off;
152 :
153 578 : if (IsMVCCSnapshot(snapshot))
154 : {
155 483 : SerializeSnapshot(snapshot, (char *) pscan + pscan->phs_snapshot_off);
156 483 : pscan->phs_snapshot_any = false;
157 : }
158 : else
159 : {
160 : Assert(snapshot == SnapshotAny);
161 95 : pscan->phs_snapshot_any = true;
162 : }
163 578 : }
164 :
165 : TableScanDesc
166 2059 : table_beginscan_parallel(Relation relation, ParallelTableScanDesc pscan)
167 : {
168 : Snapshot snapshot;
169 2059 : uint32 flags = SO_TYPE_SEQSCAN |
170 : SO_ALLOW_STRAT | SO_ALLOW_SYNC | SO_ALLOW_PAGEMODE;
171 :
172 : Assert(RelFileLocatorEquals(relation->rd_locator, pscan->phs_locator));
173 :
174 2059 : if (!pscan->phs_snapshot_any)
175 : {
176 : /* Snapshot was serialized -- restore it */
177 1856 : snapshot = RestoreSnapshot((char *) pscan + pscan->phs_snapshot_off);
178 1856 : RegisterSnapshot(snapshot);
179 1856 : flags |= SO_TEMP_SNAPSHOT;
180 : }
181 : else
182 : {
183 : /* SnapshotAny passed by caller (not serialized) */
184 203 : snapshot = SnapshotAny;
185 : }
186 :
187 2059 : return table_beginscan_common(relation, snapshot, 0, NULL,
188 : pscan, flags);
189 : }
190 :
191 : TableScanDesc
192 60 : table_beginscan_parallel_tidrange(Relation relation,
193 : ParallelTableScanDesc pscan)
194 : {
195 : Snapshot snapshot;
196 60 : uint32 flags = SO_TYPE_TIDRANGESCAN | SO_ALLOW_PAGEMODE;
197 : TableScanDesc sscan;
198 :
199 : Assert(RelFileLocatorEquals(relation->rd_locator, pscan->phs_locator));
200 :
201 : /* disable syncscan in parallel tid range scan. */
202 60 : pscan->phs_syncscan = false;
203 :
204 60 : if (!pscan->phs_snapshot_any)
205 : {
206 : /* Snapshot was serialized -- restore it */
207 60 : snapshot = RestoreSnapshot((char *) pscan + pscan->phs_snapshot_off);
208 60 : RegisterSnapshot(snapshot);
209 60 : flags |= SO_TEMP_SNAPSHOT;
210 : }
211 : else
212 : {
213 : /* SnapshotAny passed by caller (not serialized) */
214 0 : snapshot = SnapshotAny;
215 : }
216 :
217 60 : sscan = table_beginscan_common(relation, snapshot, 0, NULL,
218 : pscan, flags);
219 60 : return sscan;
220 : }
221 :
222 :
223 : /* ----------------------------------------------------------------------------
224 : * Index scan related functions.
225 : * ----------------------------------------------------------------------------
226 : */
227 :
228 : /*
229 : * To perform that check simply start an index scan, create the necessary
230 : * slot, do the heap lookup, and shut everything down again. This could be
231 : * optimized, but is unlikely to matter from a performance POV. If there
232 : * frequently are live index pointers also matching a unique index key, the
233 : * CPU overhead of this routine is unlikely to matter.
234 : *
235 : * Note that *tid may be modified when we return true if the AM supports
236 : * storing multiple row versions reachable via a single index entry (like
237 : * heap's HOT).
238 : */
239 : bool
240 5721442 : table_index_fetch_tuple_check(Relation rel,
241 : ItemPointer tid,
242 : Snapshot snapshot,
243 : bool *all_dead)
244 : {
245 : IndexFetchTableData *scan;
246 : TupleTableSlot *slot;
247 5721442 : bool call_again = false;
248 : bool found;
249 :
250 5721442 : slot = table_slot_create(rel, NULL);
251 5721442 : scan = table_index_fetch_begin(rel);
252 5721442 : found = table_index_fetch_tuple(scan, tid, snapshot, slot, &call_again,
253 : all_dead);
254 5721442 : table_index_fetch_end(scan);
255 5721442 : ExecDropSingleTupleTableSlot(slot);
256 :
257 5721442 : return found;
258 : }
259 :
260 :
261 : /* ------------------------------------------------------------------------
262 : * Functions for non-modifying operations on individual tuples
263 : * ------------------------------------------------------------------------
264 : */
265 :
266 : void
267 156 : table_tuple_get_latest_tid(TableScanDesc scan, ItemPointer tid)
268 : {
269 156 : Relation rel = scan->rs_rd;
270 156 : const TableAmRoutine *tableam = rel->rd_tableam;
271 :
272 : /*
273 : * Since this can be called with user-supplied TID, don't trust the input
274 : * too much.
275 : */
276 156 : if (!tableam->tuple_tid_valid(scan, tid))
277 6 : ereport(ERROR,
278 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
279 : errmsg("tid (%u, %u) is not valid for relation \"%s\"",
280 : ItemPointerGetBlockNumberNoCheck(tid),
281 : ItemPointerGetOffsetNumberNoCheck(tid),
282 : RelationGetRelationName(rel))));
283 :
284 150 : tableam->tuple_get_latest_tid(scan, tid);
285 150 : }
286 :
287 :
288 : /* ----------------------------------------------------------------------------
289 : * Functions to make modifications a bit simpler.
290 : * ----------------------------------------------------------------------------
291 : */
292 :
293 : /*
294 : * simple_table_tuple_insert - insert a tuple
295 : *
296 : * Currently, this routine differs from table_tuple_insert only in supplying a
297 : * default command ID and not allowing access to the speedup options.
298 : */
299 : void
300 76110 : simple_table_tuple_insert(Relation rel, TupleTableSlot *slot)
301 : {
302 76110 : table_tuple_insert(rel, slot, GetCurrentCommandId(true), 0, NULL);
303 76110 : }
304 :
305 : /*
306 : * simple_table_tuple_delete - delete a tuple
307 : *
308 : * This routine may be used to delete a tuple when concurrent updates of
309 : * the target tuple are not expected (for example, because we have a lock
310 : * on the relation associated with the tuple). Any failure is reported
311 : * via ereport().
312 : */
313 : void
314 40313 : simple_table_tuple_delete(Relation rel, ItemPointer tid, Snapshot snapshot)
315 : {
316 : TM_Result result;
317 : TM_FailureData tmfd;
318 :
319 40313 : result = table_tuple_delete(rel, tid,
320 : GetCurrentCommandId(true),
321 : snapshot, InvalidSnapshot,
322 : true /* wait for commit */ ,
323 : &tmfd, false /* changingPart */ );
324 :
325 40313 : switch (result)
326 : {
327 0 : case TM_SelfModified:
328 : /* Tuple was already updated in current command? */
329 0 : elog(ERROR, "tuple already updated by self");
330 : break;
331 :
332 40313 : case TM_Ok:
333 : /* done successfully */
334 40313 : break;
335 :
336 0 : case TM_Updated:
337 0 : elog(ERROR, "tuple concurrently updated");
338 : break;
339 :
340 0 : case TM_Deleted:
341 0 : elog(ERROR, "tuple concurrently deleted");
342 : break;
343 :
344 0 : default:
345 0 : elog(ERROR, "unrecognized table_tuple_delete status: %u", result);
346 : break;
347 : }
348 40313 : }
349 :
350 : /*
351 : * simple_table_tuple_update - replace a tuple
352 : *
353 : * This routine may be used to update a tuple when concurrent updates of
354 : * the target tuple are not expected (for example, because we have a lock
355 : * on the relation associated with the tuple). Any failure is reported
356 : * via ereport().
357 : */
358 : void
359 31921 : simple_table_tuple_update(Relation rel, ItemPointer otid,
360 : TupleTableSlot *slot,
361 : Snapshot snapshot,
362 : TU_UpdateIndexes *update_indexes)
363 : {
364 : TM_Result result;
365 : TM_FailureData tmfd;
366 : LockTupleMode lockmode;
367 :
368 31921 : result = table_tuple_update(rel, otid, slot,
369 : GetCurrentCommandId(true),
370 : snapshot, InvalidSnapshot,
371 : true /* wait for commit */ ,
372 : &tmfd, &lockmode, update_indexes);
373 :
374 31921 : switch (result)
375 : {
376 0 : case TM_SelfModified:
377 : /* Tuple was already updated in current command? */
378 0 : elog(ERROR, "tuple already updated by self");
379 : break;
380 :
381 31921 : case TM_Ok:
382 : /* done successfully */
383 31921 : break;
384 :
385 0 : case TM_Updated:
386 0 : elog(ERROR, "tuple concurrently updated");
387 : break;
388 :
389 0 : case TM_Deleted:
390 0 : elog(ERROR, "tuple concurrently deleted");
391 : break;
392 :
393 0 : default:
394 0 : elog(ERROR, "unrecognized table_tuple_update status: %u", result);
395 : break;
396 : }
397 31921 : }
398 :
399 :
400 : /* ----------------------------------------------------------------------------
401 : * Helper functions to implement parallel scans for block oriented AMs.
402 : * ----------------------------------------------------------------------------
403 : */
404 :
405 : Size
406 578 : table_block_parallelscan_estimate(Relation rel)
407 : {
408 578 : return sizeof(ParallelBlockTableScanDescData);
409 : }
410 :
411 : Size
412 578 : table_block_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan)
413 : {
414 578 : ParallelBlockTableScanDesc bpscan = (ParallelBlockTableScanDesc) pscan;
415 :
416 578 : bpscan->base.phs_locator = rel->rd_locator;
417 578 : bpscan->phs_nblocks = RelationGetNumberOfBlocks(rel);
418 : /* compare phs_syncscan initialization to similar logic in initscan */
419 1546 : bpscan->base.phs_syncscan = synchronize_seqscans &&
420 968 : !RelationUsesLocalBuffers(rel) &&
421 390 : bpscan->phs_nblocks > NBuffers / 4;
422 578 : SpinLockInit(&bpscan->phs_mutex);
423 578 : bpscan->phs_startblock = InvalidBlockNumber;
424 578 : bpscan->phs_numblock = InvalidBlockNumber;
425 578 : pg_atomic_init_u64(&bpscan->phs_nallocated, 0);
426 :
427 578 : return sizeof(ParallelBlockTableScanDescData);
428 : }
429 :
430 : void
431 114 : table_block_parallelscan_reinitialize(Relation rel, ParallelTableScanDesc pscan)
432 : {
433 114 : ParallelBlockTableScanDesc bpscan = (ParallelBlockTableScanDesc) pscan;
434 :
435 114 : pg_atomic_write_u64(&bpscan->phs_nallocated, 0);
436 114 : }
437 :
438 : /*
439 : * find and set the scan's startblock
440 : *
441 : * Determine where the parallel seq scan should start. This function may be
442 : * called many times, once by each parallel worker. We must be careful only
443 : * to set the phs_startblock and phs_numblock fields once.
444 : *
445 : * Callers may optionally specify a non-InvalidBlockNumber value for
446 : * 'startblock' to force the scan to start at the given page. Likewise,
447 : * 'numblocks' can be specified as a non-InvalidBlockNumber to limit the
448 : * number of blocks to scan to that many blocks.
449 : */
450 : void
451 1657 : table_block_parallelscan_startblock_init(Relation rel,
452 : ParallelBlockTableScanWorker pbscanwork,
453 : ParallelBlockTableScanDesc pbscan,
454 : BlockNumber startblock,
455 : BlockNumber numblocks)
456 : {
457 : StaticAssertDecl(MaxBlockNumber <= 0xFFFFFFFE,
458 : "pg_nextpower2_32 may be too small for non-standard BlockNumber width");
459 :
460 1657 : BlockNumber sync_startpage = InvalidBlockNumber;
461 : BlockNumber scan_nblocks;
462 :
463 : /* Reset the state we use for controlling allocation size. */
464 1657 : memset(pbscanwork, 0, sizeof(*pbscanwork));
465 :
466 1658 : retry:
467 : /* Grab the spinlock. */
468 1658 : SpinLockAcquire(&pbscan->phs_mutex);
469 :
470 : /*
471 : * When the caller specified a limit on the number of blocks to scan, set
472 : * that in the ParallelBlockTableScanDesc, if it's not been done by
473 : * another worker already.
474 : */
475 1658 : if (numblocks != InvalidBlockNumber &&
476 60 : pbscan->phs_numblock == InvalidBlockNumber)
477 : {
478 12 : pbscan->phs_numblock = numblocks;
479 : }
480 :
481 : /*
482 : * If the scan's phs_startblock has not yet been initialized, we must do
483 : * so now. If a startblock was specified, start there, otherwise if this
484 : * is not a synchronized scan, we just start at block 0, but if it is a
485 : * synchronized scan, we must get the starting position from the
486 : * synchronized scan machinery. We can't hold the spinlock while doing
487 : * that, though, so release the spinlock, get the information we need, and
488 : * retry. If nobody else has initialized the scan in the meantime, we'll
489 : * fill in the value we fetched on the second time through.
490 : */
491 1658 : if (pbscan->phs_startblock == InvalidBlockNumber)
492 : {
493 567 : if (startblock != InvalidBlockNumber)
494 12 : pbscan->phs_startblock = startblock;
495 555 : else if (!pbscan->base.phs_syncscan)
496 553 : pbscan->phs_startblock = 0;
497 2 : else if (sync_startpage != InvalidBlockNumber)
498 1 : pbscan->phs_startblock = sync_startpage;
499 : else
500 : {
501 1 : SpinLockRelease(&pbscan->phs_mutex);
502 1 : sync_startpage = ss_get_location(rel, pbscan->phs_nblocks);
503 1 : goto retry;
504 : }
505 : }
506 1657 : SpinLockRelease(&pbscan->phs_mutex);
507 :
508 : /*
509 : * Figure out how many blocks we're going to scan; either all of them, or
510 : * just phs_numblock's worth, if a limit has been imposed.
511 : */
512 1657 : if (pbscan->phs_numblock == InvalidBlockNumber)
513 1597 : scan_nblocks = pbscan->phs_nblocks;
514 : else
515 60 : scan_nblocks = pbscan->phs_numblock;
516 :
517 : /*
518 : * We determine the chunk size based on scan_nblocks. First we split
519 : * scan_nblocks into PARALLEL_SEQSCAN_NCHUNKS chunks then we calculate the
520 : * next highest power of 2 number of the result. This means we split the
521 : * blocks we're scanning into somewhere between PARALLEL_SEQSCAN_NCHUNKS
522 : * and PARALLEL_SEQSCAN_NCHUNKS / 2 chunks.
523 : */
524 1657 : pbscanwork->phsw_chunk_size = pg_nextpower2_32(Max(scan_nblocks /
525 : PARALLEL_SEQSCAN_NCHUNKS, 1));
526 :
527 : /*
528 : * Ensure we don't go over the maximum chunk size with larger tables. This
529 : * means we may get much more than PARALLEL_SEQSCAN_NCHUNKS for larger
530 : * tables. Too large a chunk size has been shown to be detrimental to
531 : * sequential scan performance.
532 : */
533 1657 : pbscanwork->phsw_chunk_size = Min(pbscanwork->phsw_chunk_size,
534 : PARALLEL_SEQSCAN_MAX_CHUNK_SIZE);
535 1657 : }
536 :
537 : /*
538 : * get the next page to scan
539 : *
540 : * Get the next page to scan. Even if there are no pages left to scan,
541 : * another backend could have grabbed a page to scan and not yet finished
542 : * looking at it, so it doesn't follow that the scan is done when the first
543 : * backend gets an InvalidBlockNumber return.
544 : */
545 : BlockNumber
546 101718 : table_block_parallelscan_nextpage(Relation rel,
547 : ParallelBlockTableScanWorker pbscanwork,
548 : ParallelBlockTableScanDesc pbscan)
549 : {
550 : BlockNumber scan_nblocks;
551 : BlockNumber page;
552 : uint64 nallocated;
553 :
554 : /*
555 : * The logic below allocates block numbers out to parallel workers in a
556 : * way that each worker will receive a set of consecutive block numbers to
557 : * scan. Earlier versions of this would allocate the next highest block
558 : * number to the next worker to call this function. This would generally
559 : * result in workers never receiving consecutive block numbers. Some
560 : * operating systems would not detect the sequential I/O pattern due to
561 : * each backend being a different process which could result in poor
562 : * performance due to inefficient or no readahead. To work around this
563 : * issue, we now allocate a range of block numbers for each worker and
564 : * when they come back for another block, we give them the next one in
565 : * that range until the range is complete. When the worker completes the
566 : * range of blocks we then allocate another range for it and return the
567 : * first block number from that range.
568 : *
569 : * Here we name these ranges of blocks "chunks". The initial size of
570 : * these chunks is determined in table_block_parallelscan_startblock_init
571 : * based on the number of blocks to scan. Towards the end of the scan, we
572 : * start making reductions in the size of the chunks in order to attempt
573 : * to divide the remaining work over all the workers as evenly as
574 : * possible.
575 : *
576 : * Here pbscanwork is local worker memory. phsw_chunk_remaining tracks
577 : * the number of blocks remaining in the chunk. When that reaches 0 then
578 : * we must allocate a new chunk for the worker.
579 : *
580 : * phs_nallocated tracks how many blocks have been allocated to workers
581 : * already. When phs_nallocated >= rs_nblocks, all blocks have been
582 : * allocated.
583 : *
584 : * Because we use an atomic fetch-and-add to fetch the current value, the
585 : * phs_nallocated counter will exceed rs_nblocks, because workers will
586 : * still increment the value, when they try to allocate the next block but
587 : * all blocks have been allocated already. The counter must be 64 bits
588 : * wide because of that, to avoid wrapping around when scan_nblocks is
589 : * close to 2^32.
590 : *
591 : * The actual block to return is calculated by adding the counter to the
592 : * starting block number, modulo phs_nblocks.
593 : */
594 :
595 : /* First, figure out how many blocks we're planning on scanning */
596 101718 : if (pbscan->phs_numblock == InvalidBlockNumber)
597 101409 : scan_nblocks = pbscan->phs_nblocks;
598 : else
599 309 : scan_nblocks = pbscan->phs_numblock;
600 :
601 : /*
602 : * Now check if we have any remaining blocks in a previous chunk for this
603 : * worker. We must consume all of the blocks from that before we allocate
604 : * a new chunk to the worker.
605 : */
606 101718 : if (pbscanwork->phsw_chunk_remaining > 0)
607 : {
608 : /*
609 : * Give them the next block in the range and update the remaining
610 : * number of blocks.
611 : */
612 6513 : nallocated = ++pbscanwork->phsw_nallocated;
613 6513 : pbscanwork->phsw_chunk_remaining--;
614 : }
615 : else
616 : {
617 : /*
618 : * When we've only got PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS chunks
619 : * remaining in the scan, we half the chunk size. Since we reduce the
620 : * chunk size here, we'll hit this again after doing
621 : * PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS at the new size. After a few
622 : * iterations of this, we'll end up doing the last few blocks with the
623 : * chunk size set to 1.
624 : */
625 95205 : if (pbscanwork->phsw_chunk_size > 1 &&
626 2215 : pbscanwork->phsw_nallocated > scan_nblocks -
627 2215 : (pbscanwork->phsw_chunk_size * PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS))
628 4 : pbscanwork->phsw_chunk_size >>= 1;
629 :
630 95205 : nallocated = pbscanwork->phsw_nallocated =
631 95205 : pg_atomic_fetch_add_u64(&pbscan->phs_nallocated,
632 95205 : pbscanwork->phsw_chunk_size);
633 :
634 : /*
635 : * Set the remaining number of blocks in this chunk so that subsequent
636 : * calls from this worker continue on with this chunk until it's done.
637 : */
638 95205 : pbscanwork->phsw_chunk_remaining = pbscanwork->phsw_chunk_size - 1;
639 : }
640 :
641 : /* Check if we've run out of blocks to scan */
642 101718 : if (nallocated >= scan_nblocks)
643 1657 : page = InvalidBlockNumber; /* all blocks have been allocated */
644 : else
645 100061 : page = (nallocated + pbscan->phs_startblock) % pbscan->phs_nblocks;
646 :
647 : /*
648 : * Report scan location. Normally, we report the current page number.
649 : * When we reach the end of the scan, though, we report the starting page,
650 : * not the ending page, just so the starting positions for later scans
651 : * doesn't slew backwards. We only report the position at the end of the
652 : * scan once, though: subsequent callers will report nothing.
653 : */
654 101718 : if (pbscan->base.phs_syncscan)
655 : {
656 8852 : if (page != InvalidBlockNumber)
657 8850 : ss_report_location(rel, page);
658 2 : else if (nallocated == pbscan->phs_nblocks)
659 1 : ss_report_location(rel, pbscan->phs_startblock);
660 : }
661 :
662 101718 : return page;
663 : }
664 :
665 : /* ----------------------------------------------------------------------------
666 : * Helper functions to implement relation sizing for block oriented AMs.
667 : * ----------------------------------------------------------------------------
668 : */
669 :
670 : /*
671 : * table_block_relation_size
672 : *
673 : * If a table AM uses the various relation forks as the sole place where data
674 : * is stored, and if it uses them in the expected manner (e.g. the actual data
675 : * is in the main fork rather than some other), it can use this implementation
676 : * of the relation_size callback rather than implementing its own.
677 : */
678 : uint64
679 1715606 : table_block_relation_size(Relation rel, ForkNumber forkNumber)
680 : {
681 1715606 : uint64 nblocks = 0;
682 :
683 : /* InvalidForkNumber indicates returning the size for all forks */
684 1715606 : if (forkNumber == InvalidForkNumber)
685 : {
686 0 : for (int i = 0; i < MAX_FORKNUM; i++)
687 0 : nblocks += smgrnblocks(RelationGetSmgr(rel), i);
688 : }
689 : else
690 1715606 : nblocks = smgrnblocks(RelationGetSmgr(rel), forkNumber);
691 :
692 1715587 : return nblocks * BLCKSZ;
693 : }
694 :
695 : /*
696 : * table_block_relation_estimate_size
697 : *
698 : * This function can't be directly used as the implementation of the
699 : * relation_estimate_size callback, because it has a few additional parameters.
700 : * Instead, it is intended to be used as a helper function; the caller can
701 : * pass through the arguments to its relation_estimate_size function plus the
702 : * additional values required here.
703 : *
704 : * overhead_bytes_per_tuple should contain the approximate number of bytes
705 : * of storage required to store a tuple above and beyond what is required for
706 : * the tuple data proper. Typically, this would include things like the
707 : * size of the tuple header and item pointer. This is only used for query
708 : * planning, so a table AM where the value is not constant could choose to
709 : * pass a "best guess".
710 : *
711 : * usable_bytes_per_page should contain the approximate number of bytes per
712 : * page usable for tuple data, excluding the page header and any anticipated
713 : * special space.
714 : */
715 : void
716 251361 : table_block_relation_estimate_size(Relation rel, int32 *attr_widths,
717 : BlockNumber *pages, double *tuples,
718 : double *allvisfrac,
719 : Size overhead_bytes_per_tuple,
720 : Size usable_bytes_per_page)
721 : {
722 : BlockNumber curpages;
723 : BlockNumber relpages;
724 : double reltuples;
725 : BlockNumber relallvisible;
726 : double density;
727 :
728 : /* it should have storage, so we can call the smgr */
729 251361 : curpages = RelationGetNumberOfBlocks(rel);
730 :
731 : /* coerce values in pg_class to more desirable types */
732 251361 : relpages = (BlockNumber) rel->rd_rel->relpages;
733 251361 : reltuples = (double) rel->rd_rel->reltuples;
734 251361 : relallvisible = (BlockNumber) rel->rd_rel->relallvisible;
735 :
736 : /*
737 : * HACK: if the relation has never yet been vacuumed, use a minimum size
738 : * estimate of 10 pages. The idea here is to avoid assuming a
739 : * newly-created table is really small, even if it currently is, because
740 : * that may not be true once some data gets loaded into it. Once a vacuum
741 : * or analyze cycle has been done on it, it's more reasonable to believe
742 : * the size is somewhat stable.
743 : *
744 : * (Note that this is only an issue if the plan gets cached and used again
745 : * after the table has been filled. What we're trying to avoid is using a
746 : * nestloop-type plan on a table that has grown substantially since the
747 : * plan was made. Normally, autovacuum/autoanalyze will occur once enough
748 : * inserts have happened and cause cached-plan invalidation; but that
749 : * doesn't happen instantaneously, and it won't happen at all for cases
750 : * such as temporary tables.)
751 : *
752 : * We test "never vacuumed" by seeing whether reltuples < 0.
753 : *
754 : * If the table has inheritance children, we don't apply this heuristic.
755 : * Totally empty parent tables are quite common, so we should be willing
756 : * to believe that they are empty.
757 : */
758 251361 : if (curpages < 10 &&
759 59252 : reltuples < 0 &&
760 59252 : !rel->rd_rel->relhassubclass)
761 57930 : curpages = 10;
762 :
763 : /* report estimated # pages */
764 251361 : *pages = curpages;
765 : /* quick exit if rel is clearly empty */
766 251361 : if (curpages == 0)
767 : {
768 16071 : *tuples = 0;
769 16071 : *allvisfrac = 0;
770 16071 : return;
771 : }
772 :
773 : /* estimate number of tuples from previous tuple density */
774 235290 : if (reltuples >= 0 && relpages > 0)
775 150893 : density = reltuples / (double) relpages;
776 : else
777 : {
778 : /*
779 : * When we have no data because the relation was never yet vacuumed,
780 : * estimate tuple width from attribute datatypes. We assume here that
781 : * the pages are completely full, which is OK for tables but is
782 : * probably an overestimate for indexes. Fortunately
783 : * get_relation_info() can clamp the overestimate to the parent
784 : * table's size.
785 : *
786 : * Note: this code intentionally disregards alignment considerations,
787 : * because (a) that would be gilding the lily considering how crude
788 : * the estimate is, (b) it creates platform dependencies in the
789 : * default plans which are kind of a headache for regression testing,
790 : * and (c) different table AMs might use different padding schemes.
791 : */
792 : int32 tuple_width;
793 : int fillfactor;
794 :
795 : /*
796 : * Without reltuples/relpages, we also need to consider fillfactor.
797 : * The other branch considers it implicitly by calculating density
798 : * from actual relpages/reltuples statistics.
799 : */
800 84397 : fillfactor = RelationGetFillFactor(rel, HEAP_DEFAULT_FILLFACTOR);
801 :
802 84397 : tuple_width = get_rel_data_width(rel, attr_widths);
803 84397 : tuple_width += overhead_bytes_per_tuple;
804 : /* note: integer division is intentional here */
805 84397 : density = (usable_bytes_per_page * fillfactor / 100) / tuple_width;
806 : /* There's at least one row on the page, even with low fillfactor. */
807 84397 : density = clamp_row_est(density);
808 : }
809 235290 : *tuples = rint(density * (double) curpages);
810 :
811 : /*
812 : * We use relallvisible as-is, rather than scaling it up like we do for
813 : * the pages and tuples counts, on the theory that any pages added since
814 : * the last VACUUM are most likely not marked all-visible. But costsize.c
815 : * wants it converted to a fraction.
816 : */
817 235290 : if (relallvisible == 0 || curpages <= 0)
818 112892 : *allvisfrac = 0;
819 122398 : else if ((double) relallvisible >= curpages)
820 70192 : *allvisfrac = 1;
821 : else
822 52206 : *allvisfrac = (double) relallvisible / curpages;
823 : }
|
