PostgreSQL Source Code  git master
tableam.c
Go to the documentation of this file.
1 /*----------------------------------------------------------------------
2  *
3  * tableam.c
4  * Table access method routines too big to be inline functions.
5  *
6  * Portions Copyright (c) 1996-2022, 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 function 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/plancat.h"
28 #include "port/pg_bitutils.h"
29 #include "storage/bufmgr.h"
30 #include "storage/shmem.h"
31 #include "storage/smgr.h"
32 
33 /*
34  * Constants to control the behavior of block allocation to parallel workers
35  * during a parallel seqscan. Technically these values do not need to be
36  * powers of 2, but having them as powers of 2 makes the math more optimal
37  * and makes the ramp-down stepping more even.
38  */
39 
40 /* The number of I/O chunks we try to break a parallel seqscan down into */
41 #define PARALLEL_SEQSCAN_NCHUNKS 2048
42 /* Ramp down size of allocations when we've only this number of chunks left */
43 #define PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS 64
44 /* Cap the size of parallel I/O chunks to this number of blocks */
45 #define PARALLEL_SEQSCAN_MAX_CHUNK_SIZE 8192
46 
47 /* GUC variables */
50 
51 
52 /* ----------------------------------------------------------------------------
53  * Slot functions.
54  * ----------------------------------------------------------------------------
55  */
56 
57 const TupleTableSlotOps *
59 {
60  const TupleTableSlotOps *tts_cb;
61 
62  if (relation->rd_tableam)
63  tts_cb = relation->rd_tableam->slot_callbacks(relation);
64  else if (relation->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
65  {
66  /*
67  * Historically FDWs expect to store heap tuples in slots. Continue
68  * handing them one, to make it less painful to adapt FDWs to new
69  * versions. The cost of a heap slot over a virtual slot is pretty
70  * small.
71  */
72  tts_cb = &TTSOpsHeapTuple;
73  }
74  else
75  {
76  /*
77  * These need to be supported, as some parts of the code (like COPY)
78  * need to create slots for such relations too. It seems better to
79  * centralize the knowledge that a heap slot is the right thing in
80  * that case here.
81  */
82  Assert(relation->rd_rel->relkind == RELKIND_VIEW ||
83  relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE);
84  tts_cb = &TTSOpsVirtual;
85  }
86 
87  return tts_cb;
88 }
89 
91 table_slot_create(Relation relation, List **reglist)
92 {
93  const TupleTableSlotOps *tts_cb;
94  TupleTableSlot *slot;
95 
96  tts_cb = table_slot_callbacks(relation);
97  slot = MakeSingleTupleTableSlot(RelationGetDescr(relation), tts_cb);
98 
99  if (reglist)
100  *reglist = lappend(*reglist, slot);
101 
102  return slot;
103 }
104 
105 
106 /* ----------------------------------------------------------------------------
107  * Table scan functions.
108  * ----------------------------------------------------------------------------
109  */
110 
112 table_beginscan_catalog(Relation relation, int nkeys, struct ScanKeyData *key)
113 {
114  uint32 flags = SO_TYPE_SEQSCAN |
116  Oid relid = RelationGetRelid(relation);
117  Snapshot snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
118 
119  return relation->rd_tableam->scan_begin(relation, snapshot, nkeys, key,
120  NULL, flags);
121 }
122 
123 void
125 {
126  Assert(IsMVCCSnapshot(snapshot));
127 
128  RegisterSnapshot(snapshot);
129  scan->rs_snapshot = snapshot;
130  scan->rs_flags |= SO_TEMP_SNAPSHOT;
131 }
132 
133 
134 /* ----------------------------------------------------------------------------
135  * Parallel table scan related functions.
136  * ----------------------------------------------------------------------------
137  */
138 
139 Size
141 {
142  Size sz = 0;
143 
144  if (IsMVCCSnapshot(snapshot))
145  sz = add_size(sz, EstimateSnapshotSpace(snapshot));
146  else
147  Assert(snapshot == SnapshotAny);
148 
149  sz = add_size(sz, rel->rd_tableam->parallelscan_estimate(rel));
150 
151  return sz;
152 }
153 
154 void
156  Snapshot snapshot)
157 {
158  Size snapshot_off = rel->rd_tableam->parallelscan_initialize(rel, pscan);
159 
160  pscan->phs_snapshot_off = snapshot_off;
161 
162  if (IsMVCCSnapshot(snapshot))
163  {
164  SerializeSnapshot(snapshot, (char *) pscan + pscan->phs_snapshot_off);
165  pscan->phs_snapshot_any = false;
166  }
167  else
168  {
169  Assert(snapshot == SnapshotAny);
170  pscan->phs_snapshot_any = true;
171  }
172 }
173 
176 {
177  Snapshot snapshot;
178  uint32 flags = SO_TYPE_SEQSCAN |
180 
181  Assert(RelationGetRelid(relation) == parallel_scan->phs_relid);
182 
183  if (!parallel_scan->phs_snapshot_any)
184  {
185  /* Snapshot was serialized -- restore it */
186  snapshot = RestoreSnapshot((char *) parallel_scan +
187  parallel_scan->phs_snapshot_off);
188  RegisterSnapshot(snapshot);
189  flags |= SO_TEMP_SNAPSHOT;
190  }
191  else
192  {
193  /* SnapshotAny passed by caller (not serialized) */
194  snapshot = SnapshotAny;
195  }
196 
197  return relation->rd_tableam->scan_begin(relation, snapshot, 0, NULL,
198  parallel_scan, flags);
199 }
200 
201 
202 /* ----------------------------------------------------------------------------
203  * Index scan related functions.
204  * ----------------------------------------------------------------------------
205  */
206 
207 /*
208  * To perform that check simply start an index scan, create the necessary
209  * slot, do the heap lookup, and shut everything down again. This could be
210  * optimized, but is unlikely to matter from a performance POV. If there
211  * frequently are live index pointers also matching a unique index key, the
212  * CPU overhead of this routine is unlikely to matter.
213  *
214  * Note that *tid may be modified when we return true if the AM supports
215  * storing multiple row versions reachable via a single index entry (like
216  * heap's HOT).
217  */
218 bool
220  ItemPointer tid,
221  Snapshot snapshot,
222  bool *all_dead)
223 {
224  IndexFetchTableData *scan;
225  TupleTableSlot *slot;
226  bool call_again = false;
227  bool found;
228 
229  slot = table_slot_create(rel, NULL);
230  scan = table_index_fetch_begin(rel);
231  found = table_index_fetch_tuple(scan, tid, snapshot, slot, &call_again,
232  all_dead);
233  table_index_fetch_end(scan);
235 
236  return found;
237 }
238 
239 
240 /* ------------------------------------------------------------------------
241  * Functions for non-modifying operations on individual tuples
242  * ------------------------------------------------------------------------
243  */
244 
245 void
247 {
248  Relation rel = scan->rs_rd;
249  const TableAmRoutine *tableam = rel->rd_tableam;
250 
251  /*
252  * We don't expect direct calls to table_tuple_get_latest_tid with valid
253  * CheckXidAlive for catalog or regular tables. See detailed comments in
254  * xact.c where these variables are declared.
255  */
257  elog(ERROR, "unexpected table_tuple_get_latest_tid call during logical decoding");
258 
259  /*
260  * Since this can be called with user-supplied TID, don't trust the input
261  * too much.
262  */
263  if (!tableam->tuple_tid_valid(scan, tid))
264  ereport(ERROR,
265  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
266  errmsg("tid (%u, %u) is not valid for relation \"%s\"",
269  RelationGetRelationName(rel))));
270 
271  tableam->tuple_get_latest_tid(scan, tid);
272 }
273 
274 
275 /* ----------------------------------------------------------------------------
276  * Functions to make modifications a bit simpler.
277  * ----------------------------------------------------------------------------
278  */
279 
280 /*
281  * simple_table_tuple_insert - insert a tuple
282  *
283  * Currently, this routine differs from table_tuple_insert only in supplying a
284  * default command ID and not allowing access to the speedup options.
285  */
286 void
288 {
289  table_tuple_insert(rel, slot, GetCurrentCommandId(true), 0, NULL);
290 }
291 
292 /*
293  * simple_table_tuple_delete - delete a tuple
294  *
295  * This routine may be used to delete a tuple when concurrent updates of
296  * the target tuple are not expected (for example, because we have a lock
297  * on the relation associated with the tuple). Any failure is reported
298  * via ereport().
299  */
300 void
302 {
303  TM_Result result;
304  TM_FailureData tmfd;
305 
306  result = table_tuple_delete(rel, tid,
307  GetCurrentCommandId(true),
308  snapshot, InvalidSnapshot,
309  true /* wait for commit */ ,
310  &tmfd, false /* changingPart */ );
311 
312  switch (result)
313  {
314  case TM_SelfModified:
315  /* Tuple was already updated in current command? */
316  elog(ERROR, "tuple already updated by self");
317  break;
318 
319  case TM_Ok:
320  /* done successfully */
321  break;
322 
323  case TM_Updated:
324  elog(ERROR, "tuple concurrently updated");
325  break;
326 
327  case TM_Deleted:
328  elog(ERROR, "tuple concurrently deleted");
329  break;
330 
331  default:
332  elog(ERROR, "unrecognized table_tuple_delete status: %u", result);
333  break;
334  }
335 }
336 
337 /*
338  * simple_table_tuple_update - replace a tuple
339  *
340  * This routine may be used to update a tuple when concurrent updates of
341  * the target tuple are not expected (for example, because we have a lock
342  * on the relation associated with the tuple). Any failure is reported
343  * via ereport().
344  */
345 void
347  TupleTableSlot *slot,
348  Snapshot snapshot,
349  bool *update_indexes)
350 {
351  TM_Result result;
352  TM_FailureData tmfd;
353  LockTupleMode lockmode;
354 
355  result = table_tuple_update(rel, otid, slot,
356  GetCurrentCommandId(true),
357  snapshot, InvalidSnapshot,
358  true /* wait for commit */ ,
359  &tmfd, &lockmode, update_indexes);
360 
361  switch (result)
362  {
363  case TM_SelfModified:
364  /* Tuple was already updated in current command? */
365  elog(ERROR, "tuple already updated by self");
366  break;
367 
368  case TM_Ok:
369  /* done successfully */
370  break;
371 
372  case TM_Updated:
373  elog(ERROR, "tuple concurrently updated");
374  break;
375 
376  case TM_Deleted:
377  elog(ERROR, "tuple concurrently deleted");
378  break;
379 
380  default:
381  elog(ERROR, "unrecognized table_tuple_update status: %u", result);
382  break;
383  }
384 }
385 
386 
387 /* ----------------------------------------------------------------------------
388  * Helper functions to implement parallel scans for block oriented AMs.
389  * ----------------------------------------------------------------------------
390  */
391 
392 Size
394 {
395  return sizeof(ParallelBlockTableScanDescData);
396 }
397 
398 Size
400 {
402 
403  bpscan->base.phs_relid = RelationGetRelid(rel);
404  bpscan->phs_nblocks = RelationGetNumberOfBlocks(rel);
405  /* compare phs_syncscan initialization to similar logic in initscan */
407  !RelationUsesLocalBuffers(rel) &&
408  bpscan->phs_nblocks > NBuffers / 4;
409  SpinLockInit(&bpscan->phs_mutex);
411  pg_atomic_init_u64(&bpscan->phs_nallocated, 0);
412 
413  return sizeof(ParallelBlockTableScanDescData);
414 }
415 
416 void
418 {
420 
421  pg_atomic_write_u64(&bpscan->phs_nallocated, 0);
422 }
423 
424 /*
425  * find and set the scan's startblock
426  *
427  * Determine where the parallel seq scan should start. This function may be
428  * called many times, once by each parallel worker. We must be careful only
429  * to set the startblock once.
430  */
431 void
433  ParallelBlockTableScanWorker pbscanwork,
435 {
436  BlockNumber sync_startpage = InvalidBlockNumber;
437 
438  /* Reset the state we use for controlling allocation size. */
439  memset(pbscanwork, 0, sizeof(*pbscanwork));
440 
441  StaticAssertStmt(MaxBlockNumber <= 0xFFFFFFFE,
442  "pg_nextpower2_32 may be too small for non-standard BlockNumber width");
443 
444  /*
445  * We determine the chunk size based on the size of the relation. First we
446  * split the relation into PARALLEL_SEQSCAN_NCHUNKS chunks but we then
447  * take the next highest power of 2 number of the chunk size. This means
448  * we split the relation into somewhere between PARALLEL_SEQSCAN_NCHUNKS
449  * and PARALLEL_SEQSCAN_NCHUNKS / 2 chunks.
450  */
451  pbscanwork->phsw_chunk_size = pg_nextpower2_32(Max(pbscan->phs_nblocks /
453 
454  /*
455  * Ensure we don't go over the maximum chunk size with larger tables. This
456  * means we may get much more than PARALLEL_SEQSCAN_NCHUNKS for larger
457  * tables. Too large a chunk size has been shown to be detrimental to
458  * synchronous scan performance.
459  */
460  pbscanwork->phsw_chunk_size = Min(pbscanwork->phsw_chunk_size,
462 
463 retry:
464  /* Grab the spinlock. */
465  SpinLockAcquire(&pbscan->phs_mutex);
466 
467  /*
468  * If the scan's startblock has not yet been initialized, we must do so
469  * now. If this is not a synchronized scan, we just start at block 0, but
470  * if it is a synchronized scan, we must get the starting position from
471  * the synchronized scan machinery. We can't hold the spinlock while
472  * doing that, though, so release the spinlock, get the information we
473  * need, and retry. If nobody else has initialized the scan in the
474  * meantime, we'll fill in the value we fetched on the second time
475  * through.
476  */
477  if (pbscan->phs_startblock == InvalidBlockNumber)
478  {
479  if (!pbscan->base.phs_syncscan)
480  pbscan->phs_startblock = 0;
481  else if (sync_startpage != InvalidBlockNumber)
482  pbscan->phs_startblock = sync_startpage;
483  else
484  {
485  SpinLockRelease(&pbscan->phs_mutex);
486  sync_startpage = ss_get_location(rel, pbscan->phs_nblocks);
487  goto retry;
488  }
489  }
490  SpinLockRelease(&pbscan->phs_mutex);
491 }
492 
493 /*
494  * get the next page to scan
495  *
496  * Get the next page to scan. Even if there are no pages left to scan,
497  * another backend could have grabbed a page to scan and not yet finished
498  * looking at it, so it doesn't follow that the scan is done when the first
499  * backend gets an InvalidBlockNumber return.
500  */
503  ParallelBlockTableScanWorker pbscanwork,
505 {
506  BlockNumber page;
507  uint64 nallocated;
508 
509  /*
510  * The logic below allocates block numbers out to parallel workers in a
511  * way that each worker will receive a set of consecutive block numbers to
512  * scan. Earlier versions of this would allocate the next highest block
513  * number to the next worker to call this function. This would generally
514  * result in workers never receiving consecutive block numbers. Some
515  * operating systems would not detect the sequential I/O pattern due to
516  * each backend being a different process which could result in poor
517  * performance due to inefficient or no readahead. To work around this
518  * issue, we now allocate a range of block numbers for each worker and
519  * when they come back for another block, we give them the next one in
520  * that range until the range is complete. When the worker completes the
521  * range of blocks we then allocate another range for it and return the
522  * first block number from that range.
523  *
524  * Here we name these ranges of blocks "chunks". The initial size of
525  * these chunks is determined in table_block_parallelscan_startblock_init
526  * based on the size of the relation. Towards the end of the scan, we
527  * start making reductions in the size of the chunks in order to attempt
528  * to divide the remaining work over all the workers as evenly as
529  * possible.
530  *
531  * Here pbscanwork is local worker memory. phsw_chunk_remaining tracks
532  * the number of blocks remaining in the chunk. When that reaches 0 then
533  * we must allocate a new chunk for the worker.
534  *
535  * phs_nallocated tracks how many blocks have been allocated to workers
536  * already. When phs_nallocated >= rs_nblocks, all blocks have been
537  * allocated.
538  *
539  * Because we use an atomic fetch-and-add to fetch the current value, the
540  * phs_nallocated counter will exceed rs_nblocks, because workers will
541  * still increment the value, when they try to allocate the next block but
542  * all blocks have been allocated already. The counter must be 64 bits
543  * wide because of that, to avoid wrapping around when rs_nblocks is close
544  * to 2^32.
545  *
546  * The actual block to return is calculated by adding the counter to the
547  * starting block number, modulo nblocks.
548  */
549 
550  /*
551  * First check if we have any remaining blocks in a previous chunk for
552  * this worker. We must consume all of the blocks from that before we
553  * allocate a new chunk to the worker.
554  */
555  if (pbscanwork->phsw_chunk_remaining > 0)
556  {
557  /*
558  * Give them the next block in the range and update the remaining
559  * number of blocks.
560  */
561  nallocated = ++pbscanwork->phsw_nallocated;
562  pbscanwork->phsw_chunk_remaining--;
563  }
564  else
565  {
566  /*
567  * When we've only got PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS chunks
568  * remaining in the scan, we half the chunk size. Since we reduce the
569  * chunk size here, we'll hit this again after doing
570  * PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS at the new size. After a few
571  * iterations of this, we'll end up doing the last few blocks with the
572  * chunk size set to 1.
573  */
574  if (pbscanwork->phsw_chunk_size > 1 &&
575  pbscanwork->phsw_nallocated > pbscan->phs_nblocks -
577  pbscanwork->phsw_chunk_size >>= 1;
578 
579  nallocated = pbscanwork->phsw_nallocated =
581  pbscanwork->phsw_chunk_size);
582 
583  /*
584  * Set the remaining number of blocks in this chunk so that subsequent
585  * calls from this worker continue on with this chunk until it's done.
586  */
587  pbscanwork->phsw_chunk_remaining = pbscanwork->phsw_chunk_size - 1;
588  }
589 
590  if (nallocated >= pbscan->phs_nblocks)
591  page = InvalidBlockNumber; /* all blocks have been allocated */
592  else
593  page = (nallocated + pbscan->phs_startblock) % pbscan->phs_nblocks;
594 
595  /*
596  * Report scan location. Normally, we report the current page number.
597  * When we reach the end of the scan, though, we report the starting page,
598  * not the ending page, just so the starting positions for later scans
599  * doesn't slew backwards. We only report the position at the end of the
600  * scan once, though: subsequent callers will report nothing.
601  */
602  if (pbscan->base.phs_syncscan)
603  {
604  if (page != InvalidBlockNumber)
605  ss_report_location(rel, page);
606  else if (nallocated == pbscan->phs_nblocks)
607  ss_report_location(rel, pbscan->phs_startblock);
608  }
609 
610  return page;
611 }
612 
613 /* ----------------------------------------------------------------------------
614  * Helper functions to implement relation sizing for block oriented AMs.
615  * ----------------------------------------------------------------------------
616  */
617 
618 /*
619  * table_block_relation_size
620  *
621  * If a table AM uses the various relation forks as the sole place where data
622  * is stored, and if it uses them in the expected manner (e.g. the actual data
623  * is in the main fork rather than some other), it can use this implementation
624  * of the relation_size callback rather than implementing its own.
625  */
626 uint64
628 {
629  uint64 nblocks = 0;
630 
631  /* InvalidForkNumber indicates returning the size for all forks */
632  if (forkNumber == InvalidForkNumber)
633  {
634  for (int i = 0; i < MAX_FORKNUM; i++)
635  nblocks += smgrnblocks(RelationGetSmgr(rel), i);
636  }
637  else
638  nblocks = smgrnblocks(RelationGetSmgr(rel), forkNumber);
639 
640  return nblocks * BLCKSZ;
641 }
642 
643 /*
644  * table_block_relation_estimate_size
645  *
646  * This function can't be directly used as the implementation of the
647  * relation_estimate_size callback, because it has a few additional parameters.
648  * Instead, it is intended to be used as a helper function; the caller can
649  * pass through the arguments to its relation_estimate_size function plus the
650  * additional values required here.
651  *
652  * overhead_bytes_per_tuple should contain the approximate number of bytes
653  * of storage required to store a tuple above and beyond what is required for
654  * the tuple data proper. Typically, this would include things like the
655  * size of the tuple header and item pointer. This is only used for query
656  * planning, so a table AM where the value is not constant could choose to
657  * pass a "best guess".
658  *
659  * usable_bytes_per_page should contain the approximate number of bytes per
660  * page usable for tuple data, excluding the page header and any anticipated
661  * special space.
662  */
663 void
665  BlockNumber *pages, double *tuples,
666  double *allvisfrac,
667  Size overhead_bytes_per_tuple,
668  Size usable_bytes_per_page)
669 {
670  BlockNumber curpages;
671  BlockNumber relpages;
672  double reltuples;
673  BlockNumber relallvisible;
674  double density;
675 
676  /* it should have storage, so we can call the smgr */
677  curpages = RelationGetNumberOfBlocks(rel);
678 
679  /* coerce values in pg_class to more desirable types */
680  relpages = (BlockNumber) rel->rd_rel->relpages;
681  reltuples = (double) rel->rd_rel->reltuples;
682  relallvisible = (BlockNumber) rel->rd_rel->relallvisible;
683 
684  /*
685  * HACK: if the relation has never yet been vacuumed, use a minimum size
686  * estimate of 10 pages. The idea here is to avoid assuming a
687  * newly-created table is really small, even if it currently is, because
688  * that may not be true once some data gets loaded into it. Once a vacuum
689  * or analyze cycle has been done on it, it's more reasonable to believe
690  * the size is somewhat stable.
691  *
692  * (Note that this is only an issue if the plan gets cached and used again
693  * after the table has been filled. What we're trying to avoid is using a
694  * nestloop-type plan on a table that has grown substantially since the
695  * plan was made. Normally, autovacuum/autoanalyze will occur once enough
696  * inserts have happened and cause cached-plan invalidation; but that
697  * doesn't happen instantaneously, and it won't happen at all for cases
698  * such as temporary tables.)
699  *
700  * We test "never vacuumed" by seeing whether reltuples < 0.
701  *
702  * If the table has inheritance children, we don't apply this heuristic.
703  * Totally empty parent tables are quite common, so we should be willing
704  * to believe that they are empty.
705  */
706  if (curpages < 10 &&
707  reltuples < 0 &&
708  !rel->rd_rel->relhassubclass)
709  curpages = 10;
710 
711  /* report estimated # pages */
712  *pages = curpages;
713  /* quick exit if rel is clearly empty */
714  if (curpages == 0)
715  {
716  *tuples = 0;
717  *allvisfrac = 0;
718  return;
719  }
720 
721  /* estimate number of tuples from previous tuple density */
722  if (reltuples >= 0 && relpages > 0)
723  density = reltuples / (double) relpages;
724  else
725  {
726  /*
727  * When we have no data because the relation was never yet vacuumed,
728  * estimate tuple width from attribute datatypes. We assume here that
729  * the pages are completely full, which is OK for tables but is
730  * probably an overestimate for indexes. Fortunately
731  * get_relation_info() can clamp the overestimate to the parent
732  * table's size.
733  *
734  * Note: this code intentionally disregards alignment considerations,
735  * because (a) that would be gilding the lily considering how crude
736  * the estimate is, (b) it creates platform dependencies in the
737  * default plans which are kind of a headache for regression testing,
738  * and (c) different table AMs might use different padding schemes.
739  */
740  int32 tuple_width;
741 
742  tuple_width = get_rel_data_width(rel, attr_widths);
743  tuple_width += overhead_bytes_per_tuple;
744  /* note: integer division is intentional here */
745  density = usable_bytes_per_page / tuple_width;
746  }
747  *tuples = rint(density * (double) curpages);
748 
749  /*
750  * We use relallvisible as-is, rather than scaling it up like we do for
751  * the pages and tuples counts, on the theory that any pages added since
752  * the last VACUUM are most likely not marked all-visible. But costsize.c
753  * wants it converted to a fraction.
754  */
755  if (relallvisible == 0 || curpages <= 0)
756  *allvisfrac = 0;
757  else if ((double) relallvisible >= curpages)
758  *allvisfrac = 1;
759  else
760  *allvisfrac = (double) relallvisible / curpages;
761 }
static void pg_atomic_write_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:438
static uint64 pg_atomic_fetch_add_u64(volatile pg_atomic_uint64 *ptr, int64 add_)
Definition: atomics.h:467
static void pg_atomic_init_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:415
uint32 BlockNumber
Definition: block.h:31
#define InvalidBlockNumber
Definition: block.h:33
#define MaxBlockNumber
Definition: block.h:35
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:216
unsigned int uint32
Definition: c.h:441
#define Min(x, y)
Definition: c.h:986
signed int int32
Definition: c.h:429
#define Max(x, y)
Definition: c.h:980
#define unlikely(x)
Definition: c.h:273
#define StaticAssertStmt(condition, errmessage)
Definition: c.h:918
size_t Size
Definition: c.h:540
int errcode(int sqlerrcode)
Definition: elog.c:693
int errmsg(const char *fmt,...)
Definition: elog.c:904
#define ERROR
Definition: elog.h:33
#define elog(elevel,...)
Definition: elog.h:218
#define ereport(elevel,...)
Definition: elog.h:143
const TupleTableSlotOps TTSOpsVirtual
Definition: execTuples.c:83
void ExecDropSingleTupleTableSlot(TupleTableSlot *slot)
Definition: execTuples.c:1254
const TupleTableSlotOps TTSOpsHeapTuple
Definition: execTuples.c:84
TupleTableSlot * MakeSingleTupleTableSlot(TupleDesc tupdesc, const TupleTableSlotOps *tts_ops)
Definition: execTuples.c:1238
int NBuffers
Definition: globals.c:136
int i
Definition: isn.c:73
#define ItemPointerGetBlockNumberNoCheck(pointer)
Definition: itemptr.h:89
#define ItemPointerGetOffsetNumberNoCheck(pointer)
Definition: itemptr.h:108
Assert(fmt[strlen(fmt) - 1] !='\n')
List * lappend(List *list, void *datum)
Definition: list.c:336
LockTupleMode
Definition: lockoptions.h:50
static uint32 pg_nextpower2_32(uint32 num)
Definition: pg_bitutils.h:140
int32 get_rel_data_width(Relation rel, int32 *attr_widths)
Definition: plancat.c:1085
unsigned int Oid
Definition: postgres_ext.h:31
#define RelationGetRelid(relation)
Definition: rel.h:489
static SMgrRelation RelationGetSmgr(Relation rel)
Definition: rel.h:556
#define RelationGetDescr(relation)
Definition: rel.h:515
#define RelationGetRelationName(relation)
Definition: rel.h:523
#define RelationUsesLocalBuffers(relation)
Definition: rel.h:622
ForkNumber
Definition: relpath.h:41
@ InvalidForkNumber
Definition: relpath.h:42
#define MAX_FORKNUM
Definition: relpath.h:55
struct ParallelBlockTableScanDescData * ParallelBlockTableScanDesc
Definition: relscan.h:85
struct ParallelBlockTableScanDescData ParallelBlockTableScanDescData
Size add_size(Size s1, Size s2)
Definition: shmem.c:502
BlockNumber smgrnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:579
void SerializeSnapshot(Snapshot snapshot, char *start_address)
Definition: snapmgr.c:2145
Size EstimateSnapshotSpace(Snapshot snap)
Definition: snapmgr.c:2121
Snapshot GetCatalogSnapshot(Oid relid)
Definition: snapmgr.c:386
Snapshot RestoreSnapshot(char *start_address)
Definition: snapmgr.c:2204
Snapshot RegisterSnapshot(Snapshot snapshot)
Definition: snapmgr.c:827
#define SnapshotAny
Definition: snapmgr.h:67
#define IsMVCCSnapshot(snapshot)
Definition: snapmgr.h:96
#define InvalidSnapshot
Definition: snapshot.h:123
#define SpinLockInit(lock)
Definition: spin.h:60
#define SpinLockRelease(lock)
Definition: spin.h:64
#define SpinLockAcquire(lock)
Definition: spin.h:62
Definition: pg_list.h:51
pg_atomic_uint64 phs_nallocated
Definition: relscan.h:82
ParallelTableScanDescData base
Definition: relscan.h:77
const struct TableAmRoutine * rd_tableam
Definition: rel.h:185
Form_pg_class rd_rel
Definition: rel.h:109
Size(* parallelscan_initialize)(Relation rel, ParallelTableScanDesc pscan)
Definition: tableam.h:376
void(* tuple_get_latest_tid)(TableScanDesc scan, ItemPointer tid)
Definition: tableam.h:465
TableScanDesc(* scan_begin)(Relation rel, Snapshot snapshot, int nkeys, struct ScanKeyData *key, ParallelTableScanDesc pscan, uint32 flags)
Definition: tableam.h:304
bool(* tuple_tid_valid)(TableScanDesc scan, ItemPointer tid)
Definition: tableam.h:458
const TupleTableSlotOps *(* slot_callbacks)(Relation rel)
Definition: tableam.h:280
Size(* parallelscan_estimate)(Relation rel)
Definition: tableam.h:369
Relation rs_rd
Definition: relscan.h:34
uint32 rs_flags
Definition: relscan.h:47
struct SnapshotData * rs_snapshot
Definition: relscan.h:35
void ss_report_location(Relation rel, BlockNumber location)
Definition: syncscan.c:288
BlockNumber ss_get_location(Relation rel, BlockNumber relnblocks)
Definition: syncscan.c:253
#define PARALLEL_SEQSCAN_MAX_CHUNK_SIZE
Definition: tableam.c:45
const TupleTableSlotOps * table_slot_callbacks(Relation relation)
Definition: tableam.c:58
bool table_index_fetch_tuple_check(Relation rel, ItemPointer tid, Snapshot snapshot, bool *all_dead)
Definition: tableam.c:219
Size table_block_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan)
Definition: tableam.c:399
void simple_table_tuple_insert(Relation rel, TupleTableSlot *slot)
Definition: tableam.c:287
#define PARALLEL_SEQSCAN_RAMPDOWN_CHUNKS
Definition: tableam.c:43
void table_block_parallelscan_startblock_init(Relation rel, ParallelBlockTableScanWorker pbscanwork, ParallelBlockTableScanDesc pbscan)
Definition: tableam.c:432
TableScanDesc table_beginscan_catalog(Relation relation, int nkeys, struct ScanKeyData *key)
Definition: tableam.c:112
TableScanDesc table_beginscan_parallel(Relation relation, ParallelTableScanDesc parallel_scan)
Definition: tableam.c:175
char * default_table_access_method
Definition: tableam.c:48
void table_tuple_get_latest_tid(TableScanDesc scan, ItemPointer tid)
Definition: tableam.c:246
void simple_table_tuple_delete(Relation rel, ItemPointer tid, Snapshot snapshot)
Definition: tableam.c:301
void table_block_parallelscan_reinitialize(Relation rel, ParallelTableScanDesc pscan)
Definition: tableam.c:417
uint64 table_block_relation_size(Relation rel, ForkNumber forkNumber)
Definition: tableam.c:627
TupleTableSlot * table_slot_create(Relation relation, List **reglist)
Definition: tableam.c:91
Size table_parallelscan_estimate(Relation rel, Snapshot snapshot)
Definition: tableam.c:140
Size table_block_parallelscan_estimate(Relation rel)
Definition: tableam.c:393
#define PARALLEL_SEQSCAN_NCHUNKS
Definition: tableam.c:41
void simple_table_tuple_update(Relation rel, ItemPointer otid, TupleTableSlot *slot, Snapshot snapshot, bool *update_indexes)
Definition: tableam.c:346
void table_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan, Snapshot snapshot)
Definition: tableam.c:155
void table_scan_update_snapshot(TableScanDesc scan, Snapshot snapshot)
Definition: tableam.c:124
BlockNumber table_block_parallelscan_nextpage(Relation rel, ParallelBlockTableScanWorker pbscanwork, ParallelBlockTableScanDesc pbscan)
Definition: tableam.c:502
void table_block_relation_estimate_size(Relation rel, int32 *attr_widths, BlockNumber *pages, double *tuples, double *allvisfrac, Size overhead_bytes_per_tuple, Size usable_bytes_per_page)
Definition: tableam.c:664
bool synchronize_seqscans
Definition: tableam.c:49
#define DEFAULT_TABLE_ACCESS_METHOD
Definition: tableam.h:28
@ SO_ALLOW_STRAT
Definition: tableam.h:57
@ SO_TEMP_SNAPSHOT
Definition: tableam.h:64
@ SO_ALLOW_PAGEMODE
Definition: tableam.h:61
@ SO_ALLOW_SYNC
Definition: tableam.h:59
@ SO_TYPE_SEQSCAN
Definition: tableam.h:48
static IndexFetchTableData * table_index_fetch_begin(Relation rel)
Definition: tableam.h:1163
TM_Result
Definition: tableam.h:72
@ TM_Ok
Definition: tableam.h:77
@ TM_Deleted
Definition: tableam.h:92
@ TM_Updated
Definition: tableam.h:89
@ TM_SelfModified
Definition: tableam.h:83
static void table_index_fetch_end(struct IndexFetchTableData *scan)
Definition: tableam.h:1182
static TM_Result table_tuple_delete(Relation rel, ItemPointer tid, CommandId cid, Snapshot snapshot, Snapshot crosscheck, bool wait, TM_FailureData *tmfd, bool changingPart)
Definition: tableam.h:1462
static bool table_index_fetch_tuple(struct IndexFetchTableData *scan, ItemPointer tid, Snapshot snapshot, TupleTableSlot *slot, bool *call_again, bool *all_dead)
Definition: tableam.h:1212
static void table_tuple_insert(Relation rel, TupleTableSlot *slot, CommandId cid, int options, struct BulkInsertStateData *bistate)
Definition: tableam.h:1373
static TM_Result table_tuple_update(Relation rel, ItemPointer otid, TupleTableSlot *slot, CommandId cid, Snapshot snapshot, Snapshot crosscheck, bool wait, TM_FailureData *tmfd, LockTupleMode *lockmode, bool *update_indexes)
Definition: tableam.h:1506
#define TransactionIdIsValid(xid)
Definition: transam.h:41
bool bsysscan
Definition: xact.c:99
TransactionId CheckXidAlive
Definition: xact.c:98
CommandId GetCurrentCommandId(bool used)
Definition: xact.c:814