PostgreSQL Source Code  git master
All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Pages
hashpage.c File Reference
#include "postgres.h"
#include "access/hash.h"
#include "access/hash_xlog.h"
#include "access/xloginsert.h"
#include "miscadmin.h"
#include "port/pg_bitutils.h"
#include "storage/predicate.h"
#include "storage/smgr.h"
#include "utils/rel.h"
Include dependency graph for hashpage.c:

Go to the source code of this file.

Functions

static bool _hash_alloc_buckets (Relation rel, BlockNumber firstblock, uint32 nblocks)
 
static void _hash_splitbucket (Relation rel, Buffer metabuf, Bucket obucket, Bucket nbucket, Buffer obuf, Buffer nbuf, HTAB *htab, uint32 maxbucket, uint32 highmask, uint32 lowmask)
 
static void log_split_page (Relation rel, Buffer buf)
 
Buffer _hash_getbuf (Relation rel, BlockNumber blkno, int access, int flags)
 
Buffer _hash_getbuf_with_condlock_cleanup (Relation rel, BlockNumber blkno, int flags)
 
Buffer _hash_getinitbuf (Relation rel, BlockNumber blkno)
 
void _hash_initbuf (Buffer buf, uint32 max_bucket, uint32 num_bucket, uint32 flag, bool initpage)
 
Buffer _hash_getnewbuf (Relation rel, BlockNumber blkno, ForkNumber forkNum)
 
Buffer _hash_getbuf_with_strategy (Relation rel, BlockNumber blkno, int access, int flags, BufferAccessStrategy bstrategy)
 
void _hash_relbuf (Relation rel, Buffer buf)
 
void _hash_dropbuf (Relation rel, Buffer buf)
 
void _hash_dropscanbuf (Relation rel, HashScanOpaque so)
 
uint32 _hash_init (Relation rel, double num_tuples, ForkNumber forkNum)
 
void _hash_init_metabuffer (Buffer buf, double num_tuples, RegProcedure procid, uint16 ffactor, bool initpage)
 
void _hash_pageinit (Page page, Size size)
 
void _hash_expandtable (Relation rel, Buffer metabuf)
 
void _hash_finish_split (Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket, uint32 maxbucket, uint32 highmask, uint32 lowmask)
 
HashMetaPage _hash_getcachedmetap (Relation rel, Buffer *metabuf, bool force_refresh)
 
Buffer _hash_getbucketbuf_from_hashkey (Relation rel, uint32 hashkey, int access, HashMetaPage *cachedmetap)
 

Function Documentation

◆ _hash_alloc_buckets()

static bool _hash_alloc_buckets ( Relation  rel,
BlockNumber  firstblock,
uint32  nblocks 
)
static

Definition at line 992 of file hashpage.c.

993 {
994  BlockNumber lastblock;
995  PGIOAlignedBlock zerobuf;
996  Page page;
997  HashPageOpaque ovflopaque;
998 
999  lastblock = firstblock + nblocks - 1;
1000 
1001  /*
1002  * Check for overflow in block number calculation; if so, we cannot extend
1003  * the index anymore.
1004  */
1005  if (lastblock < firstblock || lastblock == InvalidBlockNumber)
1006  return false;
1007 
1008  page = (Page) zerobuf.data;
1009 
1010  /*
1011  * Initialize the page. Just zeroing the page won't work; see
1012  * _hash_freeovflpage for similar usage. We take care to make the special
1013  * space valid for the benefit of tools such as pageinspect.
1014  */
1015  _hash_pageinit(page, BLCKSZ);
1016 
1017  ovflopaque = HashPageGetOpaque(page);
1018 
1019  ovflopaque->hasho_prevblkno = InvalidBlockNumber;
1020  ovflopaque->hasho_nextblkno = InvalidBlockNumber;
1021  ovflopaque->hasho_bucket = InvalidBucket;
1022  ovflopaque->hasho_flag = LH_UNUSED_PAGE;
1023  ovflopaque->hasho_page_id = HASHO_PAGE_ID;
1024 
1025  if (RelationNeedsWAL(rel))
1026  log_newpage(&rel->rd_locator,
1027  MAIN_FORKNUM,
1028  lastblock,
1029  zerobuf.data,
1030  true);
1031 
1032  PageSetChecksumInplace(page, lastblock);
1033  smgrextend(RelationGetSmgr(rel), MAIN_FORKNUM, lastblock, zerobuf.data,
1034  false);
1035 
1036  return true;
1037 }
uint32 BlockNumber
Definition: block.h:31
#define InvalidBlockNumber
Definition: block.h:33
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1531
Pointer Page
Definition: bufpage.h:81
#define HashPageGetOpaque(page)
Definition: hash.h:88
#define LH_UNUSED_PAGE
Definition: hash.h:53
#define HASHO_PAGE_ID
Definition: hash.h:101
#define InvalidBucket
Definition: hash.h:37
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:596
static SMgrRelation RelationGetSmgr(Relation rel)
Definition: rel.h:567
#define RelationNeedsWAL(relation)
Definition: rel.h:628
@ MAIN_FORKNUM
Definition: relpath.h:58
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, const void *buffer, bool skipFsync)
Definition: smgr.c:538
BlockNumber hasho_nextblkno
Definition: hash.h:80
uint16 hasho_flag
Definition: hash.h:82
BlockNumber hasho_prevblkno
Definition: hash.h:79
uint16 hasho_page_id
Definition: hash.h:83
Bucket hasho_bucket
Definition: hash.h:81
RelFileLocator rd_locator
Definition: rel.h:57
char data[BLCKSZ]
Definition: c.h:1116
XLogRecPtr log_newpage(RelFileLocator *rlocator, ForkNumber forknum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:1143

References _hash_pageinit(), PGIOAlignedBlock::data, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetOpaque, InvalidBlockNumber, InvalidBucket, LH_UNUSED_PAGE, log_newpage(), MAIN_FORKNUM, PageSetChecksumInplace(), RelationData::rd_locator, RelationGetSmgr(), RelationNeedsWAL, and smgrextend().

Referenced by _hash_expandtable().

◆ _hash_dropbuf()

void _hash_dropbuf ( Relation  rel,
Buffer  buf 
)

Definition at line 277 of file hashpage.c.

278 {
280 }
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:4924
static char * buf
Definition: pg_test_fsync.c:72

References buf, and ReleaseBuffer().

Referenced by _hash_doinsert(), _hash_dropscanbuf(), _hash_expandtable(), _hash_finish_split(), _hash_first(), _hash_getbucketbuf_from_hashkey(), _hash_next(), _hash_readprev(), and hashbulkdelete().

◆ _hash_dropscanbuf()

void _hash_dropscanbuf ( Relation  rel,
HashScanOpaque  so 
)

Definition at line 289 of file hashpage.c.

290 {
291  /* release pin we hold on primary bucket page */
292  if (BufferIsValid(so->hashso_bucket_buf) &&
293  so->hashso_bucket_buf != so->currPos.buf)
296 
297  /* release pin we hold on primary bucket page of bucket being split */
302 
303  /* release any pin we still hold */
304  if (BufferIsValid(so->currPos.buf))
305  _hash_dropbuf(rel, so->currPos.buf);
306  so->currPos.buf = InvalidBuffer;
307 
308  /* reset split scan */
309  so->hashso_buc_populated = false;
310  so->hashso_buc_split = false;
311 }
#define InvalidBuffer
Definition: buf.h:25
static bool BufferIsValid(Buffer bufnum)
Definition: bufmgr.h:351
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:277
bool hashso_buc_split
Definition: hash.h:180
HashScanPosData currPos
Definition: hash.h:189
bool hashso_buc_populated
Definition: hash.h:174
Buffer hashso_split_bucket_buf
Definition: hash.h:171
Buffer hashso_bucket_buf
Definition: hash.h:164
Buffer buf
Definition: hash.h:111

References _hash_dropbuf(), HashScanPosData::buf, BufferIsValid(), HashScanOpaqueData::currPos, HashScanOpaqueData::hashso_buc_populated, HashScanOpaqueData::hashso_buc_split, HashScanOpaqueData::hashso_bucket_buf, HashScanOpaqueData::hashso_split_bucket_buf, and InvalidBuffer.

Referenced by _hash_next(), hashendscan(), and hashrescan().

◆ _hash_expandtable()

void _hash_expandtable ( Relation  rel,
Buffer  metabuf 
)

Definition at line 614 of file hashpage.c.

615 {
616  HashMetaPage metap;
617  Bucket old_bucket;
618  Bucket new_bucket;
619  uint32 spare_ndx;
620  BlockNumber start_oblkno;
621  BlockNumber start_nblkno;
622  Buffer buf_nblkno;
623  Buffer buf_oblkno;
624  Page opage;
625  Page npage;
626  HashPageOpaque oopaque;
627  HashPageOpaque nopaque;
628  uint32 maxbucket;
629  uint32 highmask;
630  uint32 lowmask;
631  bool metap_update_masks = false;
632  bool metap_update_splitpoint = false;
633 
634 restart_expand:
635 
636  /*
637  * Write-lock the meta page. It used to be necessary to acquire a
638  * heavyweight lock to begin a split, but that is no longer required.
639  */
641 
642  _hash_checkpage(rel, metabuf, LH_META_PAGE);
643  metap = HashPageGetMeta(BufferGetPage(metabuf));
644 
645  /*
646  * Check to see if split is still needed; someone else might have already
647  * done one while we waited for the lock.
648  *
649  * Make sure this stays in sync with _hash_doinsert()
650  */
651  if (metap->hashm_ntuples <=
652  (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1))
653  goto fail;
654 
655  /*
656  * Can't split anymore if maxbucket has reached its maximum possible
657  * value.
658  *
659  * Ideally we'd allow bucket numbers up to UINT_MAX-1 (no higher because
660  * the calculation maxbucket+1 mustn't overflow). Currently we restrict
661  * to half that to prevent failure of pg_ceil_log2_32() and insufficient
662  * space in hashm_spares[]. It's moot anyway because an index with 2^32
663  * buckets would certainly overflow BlockNumber and hence
664  * _hash_alloc_buckets() would fail, but if we supported buckets smaller
665  * than a disk block then this would be an independent constraint.
666  *
667  * If you change this, see also the maximum initial number of buckets in
668  * _hash_init().
669  */
670  if (metap->hashm_maxbucket >= (uint32) 0x7FFFFFFE)
671  goto fail;
672 
673  /*
674  * Determine which bucket is to be split, and attempt to take cleanup lock
675  * on the old bucket. If we can't get the lock, give up.
676  *
677  * The cleanup lock protects us not only against other backends, but
678  * against our own backend as well.
679  *
680  * The cleanup lock is mainly to protect the split from concurrent
681  * inserts. See src/backend/access/hash/README, Lock Definitions for
682  * further details. Due to this locking restriction, if there is any
683  * pending scan, the split will give up which is not good, but harmless.
684  */
685  new_bucket = metap->hashm_maxbucket + 1;
686 
687  old_bucket = (new_bucket & metap->hashm_lowmask);
688 
689  start_oblkno = BUCKET_TO_BLKNO(metap, old_bucket);
690 
691  buf_oblkno = _hash_getbuf_with_condlock_cleanup(rel, start_oblkno, LH_BUCKET_PAGE);
692  if (!buf_oblkno)
693  goto fail;
694 
695  opage = BufferGetPage(buf_oblkno);
696  oopaque = HashPageGetOpaque(opage);
697 
698  /*
699  * We want to finish the split from a bucket as there is no apparent
700  * benefit by not doing so and it will make the code complicated to finish
701  * the split that involves multiple buckets considering the case where new
702  * split also fails. We don't need to consider the new bucket for
703  * completing the split here as it is not possible that a re-split of new
704  * bucket starts when there is still a pending split from old bucket.
705  */
706  if (H_BUCKET_BEING_SPLIT(oopaque))
707  {
708  /*
709  * Copy bucket mapping info now; refer the comment in code below where
710  * we copy this information before calling _hash_splitbucket to see
711  * why this is okay.
712  */
713  maxbucket = metap->hashm_maxbucket;
714  highmask = metap->hashm_highmask;
715  lowmask = metap->hashm_lowmask;
716 
717  /*
718  * Release the lock on metapage and old_bucket, before completing the
719  * split.
720  */
721  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
722  LockBuffer(buf_oblkno, BUFFER_LOCK_UNLOCK);
723 
724  _hash_finish_split(rel, metabuf, buf_oblkno, old_bucket, maxbucket,
725  highmask, lowmask);
726 
727  /* release the pin on old buffer and retry for expand. */
728  _hash_dropbuf(rel, buf_oblkno);
729 
730  goto restart_expand;
731  }
732 
733  /*
734  * Clean the tuples remained from the previous split. This operation
735  * requires cleanup lock and we already have one on the old bucket, so
736  * let's do it. We also don't want to allow further splits from the bucket
737  * till the garbage of previous split is cleaned. This has two
738  * advantages; first, it helps in avoiding the bloat due to garbage and
739  * second is, during cleanup of bucket, we are always sure that the
740  * garbage tuples belong to most recently split bucket. On the contrary,
741  * if we allow cleanup of bucket after meta page is updated to indicate
742  * the new split and before the actual split, the cleanup operation won't
743  * be able to decide whether the tuple has been moved to the newly created
744  * bucket and ended up deleting such tuples.
745  */
746  if (H_NEEDS_SPLIT_CLEANUP(oopaque))
747  {
748  /*
749  * Copy bucket mapping info now; refer to the comment in code below
750  * where we copy this information before calling _hash_splitbucket to
751  * see why this is okay.
752  */
753  maxbucket = metap->hashm_maxbucket;
754  highmask = metap->hashm_highmask;
755  lowmask = metap->hashm_lowmask;
756 
757  /* Release the metapage lock. */
758  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
759 
760  hashbucketcleanup(rel, old_bucket, buf_oblkno, start_oblkno, NULL,
761  maxbucket, highmask, lowmask, NULL, NULL, true,
762  NULL, NULL);
763 
764  _hash_dropbuf(rel, buf_oblkno);
765 
766  goto restart_expand;
767  }
768 
769  /*
770  * There shouldn't be any active scan on new bucket.
771  *
772  * Note: it is safe to compute the new bucket's blkno here, even though we
773  * may still need to update the BUCKET_TO_BLKNO mapping. This is because
774  * the current value of hashm_spares[hashm_ovflpoint] correctly shows
775  * where we are going to put a new splitpoint's worth of buckets.
776  */
777  start_nblkno = BUCKET_TO_BLKNO(metap, new_bucket);
778 
779  /*
780  * If the split point is increasing we need to allocate a new batch of
781  * bucket pages.
782  */
783  spare_ndx = _hash_spareindex(new_bucket + 1);
784  if (spare_ndx > metap->hashm_ovflpoint)
785  {
786  uint32 buckets_to_add;
787 
788  Assert(spare_ndx == metap->hashm_ovflpoint + 1);
789 
790  /*
791  * We treat allocation of buckets as a separate WAL-logged action.
792  * Even if we fail after this operation, won't leak bucket pages;
793  * rather, the next split will consume this space. In any case, even
794  * without failure we don't use all the space in one split operation.
795  */
796  buckets_to_add = _hash_get_totalbuckets(spare_ndx) - new_bucket;
797  if (!_hash_alloc_buckets(rel, start_nblkno, buckets_to_add))
798  {
799  /* can't split due to BlockNumber overflow */
800  _hash_relbuf(rel, buf_oblkno);
801  goto fail;
802  }
803  }
804 
805  /*
806  * Physically allocate the new bucket's primary page. We want to do this
807  * before changing the metapage's mapping info, in case we can't get the
808  * disk space.
809  *
810  * XXX It doesn't make sense to call _hash_getnewbuf first, zeroing the
811  * buffer, and then only afterwards check whether we have a cleanup lock.
812  * However, since no scan can be accessing the buffer yet, any concurrent
813  * accesses will just be from processes like the bgwriter or checkpointer
814  * which don't care about its contents, so it doesn't really matter.
815  */
816  buf_nblkno = _hash_getnewbuf(rel, start_nblkno, MAIN_FORKNUM);
817  if (!IsBufferCleanupOK(buf_nblkno))
818  {
819  _hash_relbuf(rel, buf_oblkno);
820  _hash_relbuf(rel, buf_nblkno);
821  goto fail;
822  }
823 
824  /*
825  * Since we are scribbling on the pages in the shared buffers, establish a
826  * critical section. Any failure in this next code leaves us with a big
827  * problem: the metapage is effectively corrupt but could get written back
828  * to disk.
829  */
831 
832  /*
833  * Okay to proceed with split. Update the metapage bucket mapping info.
834  */
835  metap->hashm_maxbucket = new_bucket;
836 
837  if (new_bucket > metap->hashm_highmask)
838  {
839  /* Starting a new doubling */
840  metap->hashm_lowmask = metap->hashm_highmask;
841  metap->hashm_highmask = new_bucket | metap->hashm_lowmask;
842  metap_update_masks = true;
843  }
844 
845  /*
846  * If the split point is increasing we need to adjust the hashm_spares[]
847  * array and hashm_ovflpoint so that future overflow pages will be created
848  * beyond this new batch of bucket pages.
849  */
850  if (spare_ndx > metap->hashm_ovflpoint)
851  {
852  metap->hashm_spares[spare_ndx] = metap->hashm_spares[metap->hashm_ovflpoint];
853  metap->hashm_ovflpoint = spare_ndx;
854  metap_update_splitpoint = true;
855  }
856 
857  MarkBufferDirty(metabuf);
858 
859  /*
860  * Copy bucket mapping info now; this saves re-accessing the meta page
861  * inside _hash_splitbucket's inner loop. Note that once we drop the
862  * split lock, other splits could begin, so these values might be out of
863  * date before _hash_splitbucket finishes. That's okay, since all it
864  * needs is to tell which of these two buckets to map hashkeys into.
865  */
866  maxbucket = metap->hashm_maxbucket;
867  highmask = metap->hashm_highmask;
868  lowmask = metap->hashm_lowmask;
869 
870  opage = BufferGetPage(buf_oblkno);
871  oopaque = HashPageGetOpaque(opage);
872 
873  /*
874  * Mark the old bucket to indicate that split is in progress. (At
875  * operation end, we will clear the split-in-progress flag.) Also, for a
876  * primary bucket page, hasho_prevblkno stores the number of buckets that
877  * existed as of the last split, so we must update that value here.
878  */
879  oopaque->hasho_flag |= LH_BUCKET_BEING_SPLIT;
880  oopaque->hasho_prevblkno = maxbucket;
881 
882  MarkBufferDirty(buf_oblkno);
883 
884  npage = BufferGetPage(buf_nblkno);
885 
886  /*
887  * initialize the new bucket's primary page and mark it to indicate that
888  * split is in progress.
889  */
890  nopaque = HashPageGetOpaque(npage);
891  nopaque->hasho_prevblkno = maxbucket;
893  nopaque->hasho_bucket = new_bucket;
895  nopaque->hasho_page_id = HASHO_PAGE_ID;
896 
897  MarkBufferDirty(buf_nblkno);
898 
899  /* XLOG stuff */
900  if (RelationNeedsWAL(rel))
901  {
903  XLogRecPtr recptr;
904 
905  xlrec.new_bucket = maxbucket;
906  xlrec.old_bucket_flag = oopaque->hasho_flag;
907  xlrec.new_bucket_flag = nopaque->hasho_flag;
908  xlrec.flags = 0;
909 
910  XLogBeginInsert();
911 
912  XLogRegisterBuffer(0, buf_oblkno, REGBUF_STANDARD);
913  XLogRegisterBuffer(1, buf_nblkno, REGBUF_WILL_INIT);
914  XLogRegisterBuffer(2, metabuf, REGBUF_STANDARD);
915 
916  if (metap_update_masks)
917  {
919  XLogRegisterBufData(2, (char *) &metap->hashm_lowmask, sizeof(uint32));
920  XLogRegisterBufData(2, (char *) &metap->hashm_highmask, sizeof(uint32));
921  }
922 
923  if (metap_update_splitpoint)
924  {
926  XLogRegisterBufData(2, (char *) &metap->hashm_ovflpoint,
927  sizeof(uint32));
929  (char *) &metap->hashm_spares[metap->hashm_ovflpoint],
930  sizeof(uint32));
931  }
932 
933  XLogRegisterData((char *) &xlrec, SizeOfHashSplitAllocPage);
934 
935  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_ALLOCATE_PAGE);
936 
937  PageSetLSN(BufferGetPage(buf_oblkno), recptr);
938  PageSetLSN(BufferGetPage(buf_nblkno), recptr);
939  PageSetLSN(BufferGetPage(metabuf), recptr);
940  }
941 
943 
944  /* drop lock, but keep pin */
945  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
946 
947  /* Relocate records to the new bucket */
948  _hash_splitbucket(rel, metabuf,
949  old_bucket, new_bucket,
950  buf_oblkno, buf_nblkno, NULL,
951  maxbucket, highmask, lowmask);
952 
953  /* all done, now release the pins on primary buckets. */
954  _hash_dropbuf(rel, buf_oblkno);
955  _hash_dropbuf(rel, buf_nblkno);
956 
957  return;
958 
959  /* Here if decide not to split or fail to acquire old bucket lock */
960 fail:
961 
962  /* We didn't write the metapage, so just drop lock */
963  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
964 }
int Buffer
Definition: buf.h:23
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:5455
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:2532
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:5158
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:189
static Page BufferGetPage(Buffer buffer)
Definition: bufmgr.h:400
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:191
static void PageSetLSN(Page page, XLogRecPtr lsn)
Definition: bufpage.h:391
unsigned int uint32
Definition: c.h:492
#define Assert(condition)
Definition: c.h:837
void hashbucketcleanup(Relation rel, Bucket cur_bucket, Buffer bucket_buf, BlockNumber bucket_blkno, BufferAccessStrategy bstrategy, uint32 maxbucket, uint32 highmask, uint32 lowmask, double *tuples_removed, double *num_index_tuples, bool split_cleanup, IndexBulkDeleteCallback callback, void *callback_state)
Definition: hash.c:684
#define LH_BUCKET_PAGE
Definition: hash.h:55
#define H_BUCKET_BEING_SPLIT(opaque)
Definition: hash.h:91
#define LH_META_PAGE
Definition: hash.h:57
#define HashPageGetMeta(page)
Definition: hash.h:323
#define LH_BUCKET_BEING_POPULATED
Definition: hash.h:58
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:39
#define H_NEEDS_SPLIT_CLEANUP(opaque)
Definition: hash.h:90
uint32 Bucket
Definition: hash.h:35
#define LH_BUCKET_BEING_SPLIT
Definition: hash.h:59
#define XLOG_HASH_SPLIT_ALLOCATE_PAGE
Definition: hash_xlog.h:31
#define SizeOfHashSplitAllocPage
Definition: hash_xlog.h:100
#define XLH_SPLIT_META_UPDATE_SPLITPOINT
Definition: hash_xlog.h:46
#define XLH_SPLIT_META_UPDATE_MASKS
Definition: hash_xlog.h:45
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:266
Buffer _hash_getbuf_with_condlock_cleanup(Relation rel, BlockNumber blkno, int flags)
Definition: hashpage.c:96
static void _hash_splitbucket(Relation rel, Buffer metabuf, Bucket obucket, Bucket nbucket, Buffer obuf, Buffer nbuf, HTAB *htab, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashpage.c:1073
void _hash_finish_split(Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashpage.c:1356
static bool _hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
Definition: hashpage.c:992
Buffer _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
Definition: hashpage.c:198
uint32 _hash_spareindex(uint32 num_bucket)
Definition: hashutil.c:142
uint32 _hash_get_totalbuckets(uint32 splitpoint_phase)
Definition: hashutil.c:174
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:210
#define START_CRIT_SECTION()
Definition: miscadmin.h:149
#define END_CRIT_SECTION()
Definition: miscadmin.h:151
uint32 hashm_lowmask
Definition: hash.h:256
uint32 hashm_maxbucket
Definition: hash.h:254
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:262
double hashm_ntuples
Definition: hash.h:248
uint32 hashm_ovflpoint
Definition: hash.h:257
uint32 hashm_highmask
Definition: hash.h:255
uint16 hashm_ffactor
Definition: hash.h:249
uint64 XLogRecPtr
Definition: xlogdefs.h:21
void XLogRegisterBufData(uint8 block_id, const char *data, uint32 len)
Definition: xloginsert.c:405
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:474
void XLogRegisterData(const char *data, uint32 len)
Definition: xloginsert.c:364
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:242
void XLogBeginInsert(void)
Definition: xloginsert.c:149
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define REGBUF_WILL_INIT
Definition: xloginsert.h:33

References _hash_alloc_buckets(), _hash_checkpage(), _hash_dropbuf(), _hash_finish_split(), _hash_get_totalbuckets(), _hash_getbuf_with_condlock_cleanup(), _hash_getnewbuf(), _hash_relbuf(), _hash_spareindex(), _hash_splitbucket(), Assert, BUCKET_TO_BLKNO, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage(), END_CRIT_SECTION, xl_hash_split_allocate_page::flags, H_BUCKET_BEING_SPLIT, H_NEEDS_SPLIT_CLEANUP, hashbucketcleanup(), HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_ntuples, HashMetaPageData::hashm_ovflpoint, HashMetaPageData::hashm_spares, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetMeta, HashPageGetOpaque, InvalidBlockNumber, IsBufferCleanupOK(), LH_BUCKET_BEING_POPULATED, LH_BUCKET_BEING_SPLIT, LH_BUCKET_PAGE, LH_META_PAGE, LockBuffer(), MAIN_FORKNUM, MarkBufferDirty(), xl_hash_split_allocate_page::new_bucket, xl_hash_split_allocate_page::new_bucket_flag, xl_hash_split_allocate_page::old_bucket_flag, PageSetLSN(), REGBUF_STANDARD, REGBUF_WILL_INIT, RelationNeedsWAL, SizeOfHashSplitAllocPage, START_CRIT_SECTION, XLH_SPLIT_META_UPDATE_MASKS, XLH_SPLIT_META_UPDATE_SPLITPOINT, XLOG_HASH_SPLIT_ALLOCATE_PAGE, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _hash_doinsert().

◆ _hash_finish_split()

void _hash_finish_split ( Relation  rel,
Buffer  metabuf,
Buffer  obuf,
Bucket  obucket,
uint32  maxbucket,
uint32  highmask,
uint32  lowmask 
)

Definition at line 1356 of file hashpage.c.

1358 {
1359  HASHCTL hash_ctl;
1360  HTAB *tidhtab;
1361  Buffer bucket_nbuf = InvalidBuffer;
1362  Buffer nbuf;
1363  Page npage;
1364  BlockNumber nblkno;
1365  BlockNumber bucket_nblkno;
1366  HashPageOpaque npageopaque;
1367  Bucket nbucket;
1368  bool found;
1369 
1370  /* Initialize hash tables used to track TIDs */
1371  hash_ctl.keysize = sizeof(ItemPointerData);
1372  hash_ctl.entrysize = sizeof(ItemPointerData);
1373  hash_ctl.hcxt = CurrentMemoryContext;
1374 
1375  tidhtab =
1376  hash_create("bucket ctids",
1377  256, /* arbitrary initial size */
1378  &hash_ctl,
1380 
1381  bucket_nblkno = nblkno = _hash_get_newblock_from_oldbucket(rel, obucket);
1382 
1383  /*
1384  * Scan the new bucket and build hash table of TIDs
1385  */
1386  for (;;)
1387  {
1388  OffsetNumber noffnum;
1389  OffsetNumber nmaxoffnum;
1390 
1391  nbuf = _hash_getbuf(rel, nblkno, HASH_READ,
1393 
1394  /* remember the primary bucket buffer to acquire cleanup lock on it. */
1395  if (nblkno == bucket_nblkno)
1396  bucket_nbuf = nbuf;
1397 
1398  npage = BufferGetPage(nbuf);
1399  npageopaque = HashPageGetOpaque(npage);
1400 
1401  /* Scan each tuple in new page */
1402  nmaxoffnum = PageGetMaxOffsetNumber(npage);
1403  for (noffnum = FirstOffsetNumber;
1404  noffnum <= nmaxoffnum;
1405  noffnum = OffsetNumberNext(noffnum))
1406  {
1407  IndexTuple itup;
1408 
1409  /* Fetch the item's TID and insert it in hash table. */
1410  itup = (IndexTuple) PageGetItem(npage,
1411  PageGetItemId(npage, noffnum));
1412 
1413  (void) hash_search(tidhtab, &itup->t_tid, HASH_ENTER, &found);
1414 
1415  Assert(!found);
1416  }
1417 
1418  nblkno = npageopaque->hasho_nextblkno;
1419 
1420  /*
1421  * release our write lock without modifying buffer and ensure to
1422  * retain the pin on primary bucket.
1423  */
1424  if (nbuf == bucket_nbuf)
1426  else
1427  _hash_relbuf(rel, nbuf);
1428 
1429  /* Exit loop if no more overflow pages in new bucket */
1430  if (!BlockNumberIsValid(nblkno))
1431  break;
1432  }
1433 
1434  /*
1435  * Conditionally get the cleanup lock on old and new buckets to perform
1436  * the split operation. If we don't get the cleanup locks, silently give
1437  * up and next insertion on old bucket will try again to complete the
1438  * split.
1439  */
1441  {
1442  hash_destroy(tidhtab);
1443  return;
1444  }
1445  if (!ConditionalLockBufferForCleanup(bucket_nbuf))
1446  {
1448  hash_destroy(tidhtab);
1449  return;
1450  }
1451 
1452  npage = BufferGetPage(bucket_nbuf);
1453  npageopaque = HashPageGetOpaque(npage);
1454  nbucket = npageopaque->hasho_bucket;
1455 
1456  _hash_splitbucket(rel, metabuf, obucket,
1457  nbucket, obuf, bucket_nbuf, tidhtab,
1458  maxbucket, highmask, lowmask);
1459 
1460  _hash_dropbuf(rel, bucket_nbuf);
1461  hash_destroy(tidhtab);
1462 }
static bool BlockNumberIsValid(BlockNumber blockNumber)
Definition: block.h:71
bool ConditionalLockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:5399
static Item PageGetItem(Page page, ItemId itemId)
Definition: bufpage.h:354
static ItemId PageGetItemId(Page page, OffsetNumber offsetNumber)
Definition: bufpage.h:243
static OffsetNumber PageGetMaxOffsetNumber(Page page)
Definition: bufpage.h:372
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:865
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:955
HTAB * hash_create(const char *tabname, long nelem, const HASHCTL *info, int flags)
Definition: dynahash.c:352
#define HASH_READ
Definition: hash.h:339
#define LH_OVERFLOW_PAGE
Definition: hash.h:54
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:70
BlockNumber _hash_get_newblock_from_oldbucket(Relation rel, Bucket old_bucket)
Definition: hashutil.c:461
@ HASH_ENTER
Definition: hsearch.h:114
#define HASH_CONTEXT
Definition: hsearch.h:102
#define HASH_ELEM
Definition: hsearch.h:95
#define HASH_BLOBS
Definition: hsearch.h:97
struct ItemPointerData ItemPointerData
IndexTupleData * IndexTuple
Definition: itup.h:53
MemoryContext CurrentMemoryContext
Definition: mcxt.c:143
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
uint16 OffsetNumber
Definition: off.h:24
#define FirstOffsetNumber
Definition: off.h:27
Size keysize
Definition: hsearch.h:75
Size entrysize
Definition: hsearch.h:76
MemoryContext hcxt
Definition: hsearch.h:86
Definition: dynahash.c:220
ItemPointerData t_tid
Definition: itup.h:37

References _hash_dropbuf(), _hash_get_newblock_from_oldbucket(), _hash_getbuf(), _hash_relbuf(), _hash_splitbucket(), Assert, BlockNumberIsValid(), BUFFER_LOCK_UNLOCK, BufferGetPage(), ConditionalLockBufferForCleanup(), CurrentMemoryContext, HASHCTL::entrysize, FirstOffsetNumber, HASH_BLOBS, HASH_CONTEXT, hash_create(), hash_destroy(), HASH_ELEM, HASH_ENTER, HASH_READ, hash_search(), HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_nextblkno, HashPageGetOpaque, HASHCTL::hcxt, InvalidBuffer, HASHCTL::keysize, LH_BUCKET_PAGE, LH_OVERFLOW_PAGE, LockBuffer(), OffsetNumberNext, PageGetItem(), PageGetItemId(), PageGetMaxOffsetNumber(), and IndexTupleData::t_tid.

Referenced by _hash_doinsert(), and _hash_expandtable().

◆ _hash_getbucketbuf_from_hashkey()

Buffer _hash_getbucketbuf_from_hashkey ( Relation  rel,
uint32  hashkey,
int  access,
HashMetaPage cachedmetap 
)

Definition at line 1559 of file hashpage.c.

1561 {
1562  HashMetaPage metap;
1563  Buffer buf;
1564  Buffer metabuf = InvalidBuffer;
1565  Page page;
1566  Bucket bucket;
1567  BlockNumber blkno;
1568  HashPageOpaque opaque;
1569 
1570  /* We read from target bucket buffer, hence locking is must. */
1572 
1573  metap = _hash_getcachedmetap(rel, &metabuf, false);
1574  Assert(metap != NULL);
1575 
1576  /*
1577  * Loop until we get a lock on the correct target bucket.
1578  */
1579  for (;;)
1580  {
1581  /*
1582  * Compute the target bucket number, and convert to block number.
1583  */
1584  bucket = _hash_hashkey2bucket(hashkey,
1585  metap->hashm_maxbucket,
1586  metap->hashm_highmask,
1587  metap->hashm_lowmask);
1588 
1589  blkno = BUCKET_TO_BLKNO(metap, bucket);
1590 
1591  /* Fetch the primary bucket page for the bucket */
1592  buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
1593  page = BufferGetPage(buf);
1594  opaque = HashPageGetOpaque(page);
1595  Assert(opaque->hasho_bucket == bucket);
1597 
1598  /*
1599  * If this bucket hasn't been split, we're done.
1600  */
1601  if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
1602  break;
1603 
1604  /* Drop lock on this buffer, update cached metapage, and retry. */
1605  _hash_relbuf(rel, buf);
1606  metap = _hash_getcachedmetap(rel, &metabuf, true);
1607  Assert(metap != NULL);
1608  }
1609 
1610  if (BufferIsValid(metabuf))
1611  _hash_dropbuf(rel, metabuf);
1612 
1613  if (cachedmetap)
1614  *cachedmetap = metap;
1615 
1616  return buf;
1617 }
#define HASH_WRITE
Definition: hash.h:340
HashMetaPage _hash_getcachedmetap(Relation rel, Buffer *metabuf, bool force_refresh)
Definition: hashpage.c:1501
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
short access
Definition: preproc-type.c:36

References _hash_dropbuf(), _hash_getbuf(), _hash_getcachedmetap(), _hash_hashkey2bucket(), _hash_relbuf(), Assert, BUCKET_TO_BLKNO, buf, BufferGetPage(), BufferIsValid(), HASH_READ, HASH_WRITE, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_maxbucket, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_prevblkno, HashPageGetOpaque, InvalidBlockNumber, InvalidBuffer, and LH_BUCKET_PAGE.

Referenced by _hash_doinsert(), and _hash_first().

◆ _hash_getbuf()

Buffer _hash_getbuf ( Relation  rel,
BlockNumber  blkno,
int  access,
int  flags 
)

Definition at line 70 of file hashpage.c.

71 {
72  Buffer buf;
73 
74  if (blkno == P_NEW)
75  elog(ERROR, "hash AM does not use P_NEW");
76 
77  buf = ReadBuffer(rel, blkno);
78 
79  if (access != HASH_NOLOCK)
81 
82  /* ref count and lock type are correct */
83 
84  _hash_checkpage(rel, buf, flags);
85 
86  return buf;
87 }
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:746
#define P_NEW
Definition: bufmgr.h:184
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:225
#define HASH_NOLOCK
Definition: hash.h:341

References _hash_checkpage(), buf, elog, ERROR, HASH_NOLOCK, LockBuffer(), P_NEW, and ReadBuffer().

Referenced by _hash_addovflpage(), _hash_doinsert(), _hash_finish_split(), _hash_first(), _hash_freeovflpage(), _hash_get_newblock_from_oldbucket(), _hash_get_oldblock_from_newbucket(), _hash_getbucketbuf_from_hashkey(), _hash_getcachedmetap(), _hash_kill_items(), _hash_next(), _hash_readnext(), _hash_readprev(), _hash_splitbucket(), hash_bitmap_info(), hashbulkdelete(), and pgstathashindex().

◆ _hash_getbuf_with_condlock_cleanup()

Buffer _hash_getbuf_with_condlock_cleanup ( Relation  rel,
BlockNumber  blkno,
int  flags 
)

Definition at line 96 of file hashpage.c.

97 {
98  Buffer buf;
99 
100  if (blkno == P_NEW)
101  elog(ERROR, "hash AM does not use P_NEW");
102 
103  buf = ReadBuffer(rel, blkno);
104 
106  {
108  return InvalidBuffer;
109  }
110 
111  /* ref count and lock type are correct */
112 
113  _hash_checkpage(rel, buf, flags);
114 
115  return buf;
116 }

References _hash_checkpage(), buf, ConditionalLockBufferForCleanup(), elog, ERROR, InvalidBuffer, P_NEW, ReadBuffer(), and ReleaseBuffer().

Referenced by _hash_expandtable().

◆ _hash_getbuf_with_strategy()

Buffer _hash_getbuf_with_strategy ( Relation  rel,
BlockNumber  blkno,
int  access,
int  flags,
BufferAccessStrategy  bstrategy 
)

Definition at line 239 of file hashpage.c.

242 {
243  Buffer buf;
244 
245  if (blkno == P_NEW)
246  elog(ERROR, "hash AM does not use P_NEW");
247 
248  buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
249 
250  if (access != HASH_NOLOCK)
252 
253  /* ref count and lock type are correct */
254 
255  _hash_checkpage(rel, buf, flags);
256 
257  return buf;
258 }
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:793
@ RBM_NORMAL
Definition: bufmgr.h:45

References _hash_checkpage(), buf, elog, ERROR, HASH_NOLOCK, LockBuffer(), MAIN_FORKNUM, P_NEW, RBM_NORMAL, and ReadBufferExtended().

Referenced by _hash_freeovflpage(), _hash_squeezebucket(), hashbucketcleanup(), and pgstat_hash_page().

◆ _hash_getcachedmetap()

HashMetaPage _hash_getcachedmetap ( Relation  rel,
Buffer metabuf,
bool  force_refresh 
)

Definition at line 1501 of file hashpage.c.

1502 {
1503  Page page;
1504 
1505  Assert(metabuf);
1506  if (force_refresh || rel->rd_amcache == NULL)
1507  {
1508  char *cache = NULL;
1509 
1510  /*
1511  * It's important that we don't set rd_amcache to an invalid value.
1512  * Either MemoryContextAlloc or _hash_getbuf could fail, so don't
1513  * install a pointer to the newly-allocated storage in the actual
1514  * relcache entry until both have succeeded.
1515  */
1516  if (rel->rd_amcache == NULL)
1517  cache = MemoryContextAlloc(rel->rd_indexcxt,
1518  sizeof(HashMetaPageData));
1519 
1520  /* Read the metapage. */
1521  if (BufferIsValid(*metabuf))
1522  LockBuffer(*metabuf, BUFFER_LOCK_SHARE);
1523  else
1524  *metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ,
1525  LH_META_PAGE);
1526  page = BufferGetPage(*metabuf);
1527 
1528  /* Populate the cache. */
1529  if (rel->rd_amcache == NULL)
1530  rel->rd_amcache = cache;
1531  memcpy(rel->rd_amcache, HashPageGetMeta(page),
1532  sizeof(HashMetaPageData));
1533 
1534  /* Release metapage lock, but keep the pin. */
1535  LockBuffer(*metabuf, BUFFER_LOCK_UNLOCK);
1536  }
1537 
1538  return (HashMetaPage) rel->rd_amcache;
1539 }
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:190
#define HASH_METAPAGE
Definition: hash.h:198
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:1181
void * rd_amcache
Definition: rel.h:229
MemoryContext rd_indexcxt
Definition: rel.h:204

References _hash_getbuf(), Assert, BUFFER_LOCK_SHARE, BUFFER_LOCK_UNLOCK, BufferGetPage(), BufferIsValid(), HASH_METAPAGE, HASH_READ, HashPageGetMeta, LH_META_PAGE, LockBuffer(), MemoryContextAlloc(), RelationData::rd_amcache, and RelationData::rd_indexcxt.

Referenced by _hash_getbucketbuf_from_hashkey(), and hashbulkdelete().

◆ _hash_getinitbuf()

Buffer _hash_getinitbuf ( Relation  rel,
BlockNumber  blkno 
)

Definition at line 135 of file hashpage.c.

136 {
137  Buffer buf;
138 
139  if (blkno == P_NEW)
140  elog(ERROR, "hash AM does not use P_NEW");
141 
143  NULL);
144 
145  /* ref count and lock type are correct */
146 
147  /* initialize the page */
149 
150  return buf;
151 }
static Size BufferGetPageSize(Buffer buffer)
Definition: bufmgr.h:389
@ RBM_ZERO_AND_LOCK
Definition: bufmgr.h:46

References _hash_pageinit(), buf, BufferGetPage(), BufferGetPageSize(), elog, ERROR, MAIN_FORKNUM, P_NEW, RBM_ZERO_AND_LOCK, and ReadBufferExtended().

Referenced by _hash_addovflpage().

◆ _hash_getnewbuf()

Buffer _hash_getnewbuf ( Relation  rel,
BlockNumber  blkno,
ForkNumber  forkNum 
)

Definition at line 198 of file hashpage.c.

199 {
200  BlockNumber nblocks = RelationGetNumberOfBlocksInFork(rel, forkNum);
201  Buffer buf;
202 
203  if (blkno == P_NEW)
204  elog(ERROR, "hash AM does not use P_NEW");
205  if (blkno > nblocks)
206  elog(ERROR, "access to noncontiguous page in hash index \"%s\"",
208 
209  /* smgr insists we explicitly extend the relation */
210  if (blkno == nblocks)
211  {
212  buf = ExtendBufferedRel(BMR_REL(rel), forkNum, NULL,
214  if (BufferGetBlockNumber(buf) != blkno)
215  elog(ERROR, "unexpected hash relation size: %u, should be %u",
216  BufferGetBlockNumber(buf), blkno);
217  }
218  else
219  {
220  buf = ReadBufferExtended(rel, forkNum, blkno, RBM_ZERO_AND_LOCK,
221  NULL);
222  }
223 
224  /* ref count and lock type are correct */
225 
226  /* initialize the page */
228 
229  return buf;
230 }
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:3724
Buffer ExtendBufferedRel(BufferManagerRelation bmr, ForkNumber forkNum, BufferAccessStrategy strategy, uint32 flags)
Definition: bufmgr.c:846
BlockNumber RelationGetNumberOfBlocksInFork(Relation relation, ForkNumber forkNum)
Definition: bufmgr.c:3923
@ EB_SKIP_EXTENSION_LOCK
Definition: bufmgr.h:74
@ EB_LOCK_FIRST
Definition: bufmgr.h:86
#define BMR_REL(p_rel)
Definition: bufmgr.h:107
#define RelationGetRelationName(relation)
Definition: rel.h:539

References _hash_pageinit(), BMR_REL, buf, BufferGetBlockNumber(), BufferGetPage(), BufferGetPageSize(), EB_LOCK_FIRST, EB_SKIP_EXTENSION_LOCK, elog, ERROR, ExtendBufferedRel(), P_NEW, RBM_ZERO_AND_LOCK, ReadBufferExtended(), RelationGetNumberOfBlocksInFork(), and RelationGetRelationName.

Referenced by _hash_addovflpage(), _hash_expandtable(), and _hash_init().

◆ _hash_init()

uint32 _hash_init ( Relation  rel,
double  num_tuples,
ForkNumber  forkNum 
)

Definition at line 327 of file hashpage.c.

328 {
329  Buffer metabuf;
330  Buffer buf;
331  Buffer bitmapbuf;
332  Page pg;
333  HashMetaPage metap;
334  RegProcedure procid;
335  int32 data_width;
336  int32 item_width;
337  int32 ffactor;
338  uint32 num_buckets;
339  uint32 i;
340  bool use_wal;
341 
342  /* safety check */
343  if (RelationGetNumberOfBlocksInFork(rel, forkNum) != 0)
344  elog(ERROR, "cannot initialize non-empty hash index \"%s\"",
346 
347  /*
348  * WAL log creation of pages if the relation is persistent, or this is the
349  * init fork. Init forks for unlogged relations always need to be WAL
350  * logged.
351  */
352  use_wal = RelationNeedsWAL(rel) || forkNum == INIT_FORKNUM;
353 
354  /*
355  * Determine the target fill factor (in tuples per bucket) for this index.
356  * The idea is to make the fill factor correspond to pages about as full
357  * as the user-settable fillfactor parameter says. We can compute it
358  * exactly since the index datatype (i.e. uint32 hash key) is fixed-width.
359  */
360  data_width = sizeof(uint32);
361  item_width = MAXALIGN(sizeof(IndexTupleData)) + MAXALIGN(data_width) +
362  sizeof(ItemIdData); /* include the line pointer */
363  ffactor = HashGetTargetPageUsage(rel) / item_width;
364  /* keep to a sane range */
365  if (ffactor < 10)
366  ffactor = 10;
367 
368  procid = index_getprocid(rel, 1, HASHSTANDARD_PROC);
369 
370  /*
371  * We initialize the metapage, the first N bucket pages, and the first
372  * bitmap page in sequence, using _hash_getnewbuf to cause smgrextend()
373  * calls to occur. This ensures that the smgr level has the right idea of
374  * the physical index length.
375  *
376  * Critical section not required, because on error the creation of the
377  * whole relation will be rolled back.
378  */
379  metabuf = _hash_getnewbuf(rel, HASH_METAPAGE, forkNum);
380  _hash_init_metabuffer(metabuf, num_tuples, procid, ffactor, false);
381  MarkBufferDirty(metabuf);
382 
383  pg = BufferGetPage(metabuf);
384  metap = HashPageGetMeta(pg);
385 
386  /* XLOG stuff */
387  if (use_wal)
388  {
390  XLogRecPtr recptr;
391 
392  xlrec.num_tuples = num_tuples;
393  xlrec.procid = metap->hashm_procid;
394  xlrec.ffactor = metap->hashm_ffactor;
395 
396  XLogBeginInsert();
397  XLogRegisterData((char *) &xlrec, SizeOfHashInitMetaPage);
399 
400  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_META_PAGE);
401 
402  PageSetLSN(BufferGetPage(metabuf), recptr);
403  }
404 
405  num_buckets = metap->hashm_maxbucket + 1;
406 
407  /*
408  * Release buffer lock on the metapage while we initialize buckets.
409  * Otherwise, we'll be in interrupt holdoff and the CHECK_FOR_INTERRUPTS
410  * won't accomplish anything. It's a bad idea to hold buffer locks for
411  * long intervals in any case, since that can block the bgwriter.
412  */
413  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
414 
415  /*
416  * Initialize and WAL Log the first N buckets
417  */
418  for (i = 0; i < num_buckets; i++)
419  {
420  BlockNumber blkno;
421 
422  /* Allow interrupts, in case N is huge */
424 
425  blkno = BUCKET_TO_BLKNO(metap, i);
426  buf = _hash_getnewbuf(rel, blkno, forkNum);
429 
430  if (use_wal)
431  log_newpage(&rel->rd_locator,
432  forkNum,
433  blkno,
435  true);
436  _hash_relbuf(rel, buf);
437  }
438 
439  /* Now reacquire buffer lock on metapage */
441 
442  /*
443  * Initialize bitmap page
444  */
445  bitmapbuf = _hash_getnewbuf(rel, num_buckets + 1, forkNum);
446  _hash_initbitmapbuffer(bitmapbuf, metap->hashm_bmsize, false);
447  MarkBufferDirty(bitmapbuf);
448 
449  /* add the new bitmap page to the metapage's list of bitmaps */
450  /* metapage already has a write lock */
451  if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
452  ereport(ERROR,
453  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
454  errmsg("out of overflow pages in hash index \"%s\"",
455  RelationGetRelationName(rel))));
456 
457  metap->hashm_mapp[metap->hashm_nmaps] = num_buckets + 1;
458 
459  metap->hashm_nmaps++;
460  MarkBufferDirty(metabuf);
461 
462  /* XLOG stuff */
463  if (use_wal)
464  {
466  XLogRecPtr recptr;
467 
468  xlrec.bmsize = metap->hashm_bmsize;
469 
470  XLogBeginInsert();
471  XLogRegisterData((char *) &xlrec, SizeOfHashInitBitmapPage);
472  XLogRegisterBuffer(0, bitmapbuf, REGBUF_WILL_INIT);
473 
474  /*
475  * This is safe only because nobody else can be modifying the index at
476  * this stage; it's only visible to the transaction that is creating
477  * it.
478  */
479  XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
480 
481  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_BITMAP_PAGE);
482 
483  PageSetLSN(BufferGetPage(bitmapbuf), recptr);
484  PageSetLSN(BufferGetPage(metabuf), recptr);
485  }
486 
487  /* all done */
488  _hash_relbuf(rel, bitmapbuf);
489  _hash_relbuf(rel, metabuf);
490 
491  return num_buckets;
492 }
#define MAXALIGN(LEN)
Definition: c.h:790
signed int int32
Definition: c.h:482
regproc RegProcedure
Definition: c.h:629
int errcode(int sqlerrcode)
Definition: elog.c:853
int errmsg(const char *fmt,...)
Definition: elog.c:1070
#define ereport(elevel,...)
Definition: elog.h:149
#define HASH_MAX_BITMAPS
Definition: hash.h:230
#define HASHSTANDARD_PROC
Definition: hash.h:355
#define HashGetTargetPageUsage(relation)
Definition: hash.h:281
#define SizeOfHashInitBitmapPage
Definition: hash_xlog.h:233
#define XLOG_HASH_INIT_BITMAP_PAGE
Definition: hash_xlog.h:28
#define XLOG_HASH_INIT_META_PAGE
Definition: hash_xlog.h:27
#define SizeOfHashInitMetaPage
Definition: hash_xlog.h:217
void _hash_initbitmapbuffer(Buffer buf, uint16 bmsize, bool initpage)
Definition: hashovfl.c:777
void _hash_initbuf(Buffer buf, uint32 max_bucket, uint32 num_bucket, uint32 flag, bool initpage)
Definition: hashpage.c:157
void _hash_init_metabuffer(Buffer buf, double num_tuples, RegProcedure procid, uint16 ffactor, bool initpage)
Definition: hashpage.c:498
RegProcedure index_getprocid(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:828
int i
Definition: isn.c:72
struct ItemIdData ItemIdData
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:122
@ INIT_FORKNUM
Definition: relpath.h:61
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:264
RegProcedure hashm_procid
Definition: hash.h:261
uint16 hashm_bmsize
Definition: hash.h:251
uint32 hashm_nmaps
Definition: hash.h:260
RegProcedure procid
Definition: hash_xlog.h:213

References _hash_getnewbuf(), _hash_init_metabuffer(), _hash_initbitmapbuffer(), _hash_initbuf(), _hash_relbuf(), xl_hash_init_bitmap_page::bmsize, BUCKET_TO_BLKNO, buf, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage(), CHECK_FOR_INTERRUPTS, elog, ereport, errcode(), errmsg(), ERROR, xl_hash_init_meta_page::ffactor, HASH_MAX_BITMAPS, HASH_METAPAGE, HashGetTargetPageUsage, HashMetaPageData::hashm_bmsize, HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_mapp, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_nmaps, HashMetaPageData::hashm_procid, HashPageGetMeta, HASHSTANDARD_PROC, i, index_getprocid(), INIT_FORKNUM, LH_BUCKET_PAGE, LockBuffer(), log_newpage(), MarkBufferDirty(), MAXALIGN, xl_hash_init_meta_page::num_tuples, PageSetLSN(), xl_hash_init_meta_page::procid, RelationData::rd_locator, REGBUF_STANDARD, REGBUF_WILL_INIT, RelationGetNumberOfBlocksInFork(), RelationGetRelationName, RelationNeedsWAL, SizeOfHashInitBitmapPage, SizeOfHashInitMetaPage, XLOG_HASH_INIT_BITMAP_PAGE, XLOG_HASH_INIT_META_PAGE, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by hashbuild(), and hashbuildempty().

◆ _hash_init_metabuffer()

void _hash_init_metabuffer ( Buffer  buf,
double  num_tuples,
RegProcedure  procid,
uint16  ffactor,
bool  initpage 
)

Definition at line 498 of file hashpage.c.

500 {
501  HashMetaPage metap;
502  HashPageOpaque pageopaque;
503  Page page;
504  double dnumbuckets;
505  uint32 num_buckets;
506  uint32 spare_index;
507  uint32 lshift;
508 
509  /*
510  * Choose the number of initial bucket pages to match the fill factor
511  * given the estimated number of tuples. We round up the result to the
512  * total number of buckets which has to be allocated before using its
513  * hashm_spares element. However always force at least 2 bucket pages. The
514  * upper limit is determined by considerations explained in
515  * _hash_expandtable().
516  */
517  dnumbuckets = num_tuples / ffactor;
518  if (dnumbuckets <= 2.0)
519  num_buckets = 2;
520  else if (dnumbuckets >= (double) 0x40000000)
521  num_buckets = 0x40000000;
522  else
523  num_buckets = _hash_get_totalbuckets(_hash_spareindex(dnumbuckets));
524 
525  spare_index = _hash_spareindex(num_buckets);
526  Assert(spare_index < HASH_MAX_SPLITPOINTS);
527 
528  page = BufferGetPage(buf);
529  if (initpage)
531 
532  pageopaque = HashPageGetOpaque(page);
533  pageopaque->hasho_prevblkno = InvalidBlockNumber;
534  pageopaque->hasho_nextblkno = InvalidBlockNumber;
535  pageopaque->hasho_bucket = InvalidBucket;
536  pageopaque->hasho_flag = LH_META_PAGE;
537  pageopaque->hasho_page_id = HASHO_PAGE_ID;
538 
539  metap = HashPageGetMeta(page);
540 
541  metap->hashm_magic = HASH_MAGIC;
542  metap->hashm_version = HASH_VERSION;
543  metap->hashm_ntuples = 0;
544  metap->hashm_nmaps = 0;
545  metap->hashm_ffactor = ffactor;
546  metap->hashm_bsize = HashGetMaxBitmapSize(page);
547 
548  /* find largest bitmap array size that will fit in page size */
549  lshift = pg_leftmost_one_pos32(metap->hashm_bsize);
550  Assert(lshift > 0);
551  metap->hashm_bmsize = 1 << lshift;
552  metap->hashm_bmshift = lshift + BYTE_TO_BIT;
553  Assert((1 << BMPG_SHIFT(metap)) == (BMPG_MASK(metap) + 1));
554 
555  /*
556  * Label the index with its primary hash support function's OID. This is
557  * pretty useless for normal operation (in fact, hashm_procid is not used
558  * anywhere), but it might be handy for forensic purposes so we keep it.
559  */
560  metap->hashm_procid = procid;
561 
562  /*
563  * We initialize the index with N buckets, 0 .. N-1, occupying physical
564  * blocks 1 to N. The first freespace bitmap page is in block N+1.
565  */
566  metap->hashm_maxbucket = num_buckets - 1;
567 
568  /*
569  * Set highmask as next immediate ((2 ^ x) - 1), which should be
570  * sufficient to cover num_buckets.
571  */
572  metap->hashm_highmask = pg_nextpower2_32(num_buckets + 1) - 1;
573  metap->hashm_lowmask = (metap->hashm_highmask >> 1);
574 
575  MemSet(metap->hashm_spares, 0, sizeof(metap->hashm_spares));
576  MemSet(metap->hashm_mapp, 0, sizeof(metap->hashm_mapp));
577 
578  /* Set up mapping for one spare page after the initial splitpoints */
579  metap->hashm_spares[spare_index] = 1;
580  metap->hashm_ovflpoint = spare_index;
581  metap->hashm_firstfree = 0;
582 
583  /*
584  * Set pd_lower just past the end of the metadata. This is essential,
585  * because without doing so, metadata will be lost if xlog.c compresses
586  * the page.
587  */
588  ((PageHeader) page)->pd_lower =
589  ((char *) metap + sizeof(HashMetaPageData)) - (char *) page;
590 }
PageHeaderData * PageHeader
Definition: bufpage.h:173
#define MemSet(start, val, len)
Definition: c.h:999
#define BMPG_MASK(metap)
Definition: hash.h:314
#define HASH_VERSION
Definition: hash.h:201
#define HASH_MAX_SPLITPOINTS
Definition: hash.h:239
#define BYTE_TO_BIT
Definition: hash.h:301
#define HashGetMaxBitmapSize(page)
Definition: hash.h:319
#define BMPG_SHIFT(metap)
Definition: hash.h:313
#define HASH_MAGIC
Definition: hash.h:200
static uint32 pg_nextpower2_32(uint32 num)
Definition: pg_bitutils.h:189
static int pg_leftmost_one_pos32(uint32 word)
Definition: pg_bitutils.h:41
uint32 hashm_version
Definition: hash.h:247
uint32 hashm_firstfree
Definition: hash.h:259
uint16 hashm_bsize
Definition: hash.h:250
uint32 hashm_magic
Definition: hash.h:246
uint16 hashm_bmshift
Definition: hash.h:253

References _hash_get_totalbuckets(), _hash_pageinit(), _hash_spareindex(), Assert, BMPG_MASK, BMPG_SHIFT, buf, BufferGetPage(), BufferGetPageSize(), BYTE_TO_BIT, HASH_MAGIC, HASH_MAX_SPLITPOINTS, HASH_VERSION, HashGetMaxBitmapSize, HashMetaPageData::hashm_bmshift, HashMetaPageData::hashm_bmsize, HashMetaPageData::hashm_bsize, HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_firstfree, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_magic, HashMetaPageData::hashm_mapp, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_nmaps, HashMetaPageData::hashm_ntuples, HashMetaPageData::hashm_ovflpoint, HashMetaPageData::hashm_procid, HashMetaPageData::hashm_spares, HashMetaPageData::hashm_version, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetMeta, HashPageGetOpaque, InvalidBlockNumber, InvalidBucket, LH_META_PAGE, MemSet, pg_leftmost_one_pos32(), and pg_nextpower2_32().

Referenced by _hash_init(), and hash_xlog_init_meta_page().

◆ _hash_initbuf()

void _hash_initbuf ( Buffer  buf,
uint32  max_bucket,
uint32  num_bucket,
uint32  flag,
bool  initpage 
)

Definition at line 157 of file hashpage.c.

159 {
160  HashPageOpaque pageopaque;
161  Page page;
162 
163  page = BufferGetPage(buf);
164 
165  /* initialize the page */
166  if (initpage)
168 
169  pageopaque = HashPageGetOpaque(page);
170 
171  /*
172  * Set hasho_prevblkno with current hashm_maxbucket. This value will be
173  * used to validate cached HashMetaPageData. See
174  * _hash_getbucketbuf_from_hashkey().
175  */
176  pageopaque->hasho_prevblkno = max_bucket;
177  pageopaque->hasho_nextblkno = InvalidBlockNumber;
178  pageopaque->hasho_bucket = num_bucket;
179  pageopaque->hasho_flag = flag;
180  pageopaque->hasho_page_id = HASHO_PAGE_ID;
181 }
char * flag(int b)
Definition: test-ctype.c:33

References _hash_pageinit(), buf, BufferGetPage(), BufferGetPageSize(), flag(), HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetOpaque, and InvalidBlockNumber.

Referenced by _hash_init(), hash_xlog_add_ovfl_page(), and hash_xlog_split_allocate_page().

◆ _hash_pageinit()

void _hash_pageinit ( Page  page,
Size  size 
)

Definition at line 596 of file hashpage.c.

597 {
598  PageInit(page, size, sizeof(HashPageOpaqueData));
599 }
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:42
static pg_noinline void Size size
Definition: slab.c:607

References PageInit(), and size.

Referenced by _hash_alloc_buckets(), _hash_freeovflpage(), _hash_getinitbuf(), _hash_getnewbuf(), _hash_init_metabuffer(), _hash_initbitmapbuffer(), _hash_initbuf(), and hash_xlog_squeeze_page().

◆ _hash_relbuf()

◆ _hash_splitbucket()

static void _hash_splitbucket ( Relation  rel,
Buffer  metabuf,
Bucket  obucket,
Bucket  nbucket,
Buffer  obuf,
Buffer  nbuf,
HTAB htab,
uint32  maxbucket,
uint32  highmask,
uint32  lowmask 
)
static

Definition at line 1073 of file hashpage.c.

1083 {
1084  Buffer bucket_obuf;
1085  Buffer bucket_nbuf;
1086  Page opage;
1087  Page npage;
1088  HashPageOpaque oopaque;
1089  HashPageOpaque nopaque;
1090  OffsetNumber itup_offsets[MaxIndexTuplesPerPage];
1092  Size all_tups_size = 0;
1093  int i;
1094  uint16 nitups = 0;
1095 
1096  bucket_obuf = obuf;
1097  opage = BufferGetPage(obuf);
1098  oopaque = HashPageGetOpaque(opage);
1099 
1100  bucket_nbuf = nbuf;
1101  npage = BufferGetPage(nbuf);
1102  nopaque = HashPageGetOpaque(npage);
1103 
1104  /* Copy the predicate locks from old bucket to new bucket. */
1106  BufferGetBlockNumber(bucket_obuf),
1107  BufferGetBlockNumber(bucket_nbuf));
1108 
1109  /*
1110  * Partition the tuples in the old bucket between the old bucket and the
1111  * new bucket, advancing along the old bucket's overflow bucket chain and
1112  * adding overflow pages to the new bucket as needed. Outer loop iterates
1113  * once per page in old bucket.
1114  */
1115  for (;;)
1116  {
1117  BlockNumber oblkno;
1118  OffsetNumber ooffnum;
1119  OffsetNumber omaxoffnum;
1120 
1121  /* Scan each tuple in old page */
1122  omaxoffnum = PageGetMaxOffsetNumber(opage);
1123  for (ooffnum = FirstOffsetNumber;
1124  ooffnum <= omaxoffnum;
1125  ooffnum = OffsetNumberNext(ooffnum))
1126  {
1127  IndexTuple itup;
1128  Size itemsz;
1129  Bucket bucket;
1130  bool found = false;
1131 
1132  /* skip dead tuples */
1133  if (ItemIdIsDead(PageGetItemId(opage, ooffnum)))
1134  continue;
1135 
1136  /*
1137  * Before inserting a tuple, probe the hash table containing TIDs
1138  * of tuples belonging to new bucket, if we find a match, then
1139  * skip that tuple, else fetch the item's hash key (conveniently
1140  * stored in the item) and determine which bucket it now belongs
1141  * in.
1142  */
1143  itup = (IndexTuple) PageGetItem(opage,
1144  PageGetItemId(opage, ooffnum));
1145 
1146  if (htab)
1147  (void) hash_search(htab, &itup->t_tid, HASH_FIND, &found);
1148 
1149  if (found)
1150  continue;
1151 
1153  maxbucket, highmask, lowmask);
1154 
1155  if (bucket == nbucket)
1156  {
1157  IndexTuple new_itup;
1158 
1159  /*
1160  * make a copy of index tuple as we have to scribble on it.
1161  */
1162  new_itup = CopyIndexTuple(itup);
1163 
1164  /*
1165  * mark the index tuple as moved by split, such tuples are
1166  * skipped by scan if there is split in progress for a bucket.
1167  */
1168  new_itup->t_info |= INDEX_MOVED_BY_SPLIT_MASK;
1169 
1170  /*
1171  * insert the tuple into the new bucket. if it doesn't fit on
1172  * the current page in the new bucket, we must allocate a new
1173  * overflow page and place the tuple on that page instead.
1174  */
1175  itemsz = IndexTupleSize(new_itup);
1176  itemsz = MAXALIGN(itemsz);
1177 
1178  if (PageGetFreeSpaceForMultipleTuples(npage, nitups + 1) < (all_tups_size + itemsz))
1179  {
1180  /*
1181  * Change the shared buffer state in critical section,
1182  * otherwise any error could make it unrecoverable.
1183  */
1185 
1186  _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1187  MarkBufferDirty(nbuf);
1188  /* log the split operation before releasing the lock */
1189  log_split_page(rel, nbuf);
1190 
1191  END_CRIT_SECTION();
1192 
1193  /* drop lock, but keep pin */
1195 
1196  /* be tidy */
1197  for (i = 0; i < nitups; i++)
1198  pfree(itups[i]);
1199  nitups = 0;
1200  all_tups_size = 0;
1201 
1202  /* chain to a new overflow page */
1203  nbuf = _hash_addovflpage(rel, metabuf, nbuf, (nbuf == bucket_nbuf));
1204  npage = BufferGetPage(nbuf);
1205  nopaque = HashPageGetOpaque(npage);
1206  }
1207 
1208  itups[nitups++] = new_itup;
1209  all_tups_size += itemsz;
1210  }
1211  else
1212  {
1213  /*
1214  * the tuple stays on this page, so nothing to do.
1215  */
1216  Assert(bucket == obucket);
1217  }
1218  }
1219 
1220  oblkno = oopaque->hasho_nextblkno;
1221 
1222  /* retain the pin on the old primary bucket */
1223  if (obuf == bucket_obuf)
1225  else
1226  _hash_relbuf(rel, obuf);
1227 
1228  /* Exit loop if no more overflow pages in old bucket */
1229  if (!BlockNumberIsValid(oblkno))
1230  {
1231  /*
1232  * Change the shared buffer state in critical section, otherwise
1233  * any error could make it unrecoverable.
1234  */
1236 
1237  _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1238  MarkBufferDirty(nbuf);
1239  /* log the split operation before releasing the lock */
1240  log_split_page(rel, nbuf);
1241 
1242  END_CRIT_SECTION();
1243 
1244  if (nbuf == bucket_nbuf)
1246  else
1247  _hash_relbuf(rel, nbuf);
1248 
1249  /* be tidy */
1250  for (i = 0; i < nitups; i++)
1251  pfree(itups[i]);
1252  break;
1253  }
1254 
1255  /* Else, advance to next old page */
1256  obuf = _hash_getbuf(rel, oblkno, HASH_READ, LH_OVERFLOW_PAGE);
1257  opage = BufferGetPage(obuf);
1258  oopaque = HashPageGetOpaque(opage);
1259  }
1260 
1261  /*
1262  * We're at the end of the old bucket chain, so we're done partitioning
1263  * the tuples. Mark the old and new buckets to indicate split is
1264  * finished.
1265  *
1266  * To avoid deadlocks due to locking order of buckets, first lock the old
1267  * bucket and then the new bucket.
1268  */
1269  LockBuffer(bucket_obuf, BUFFER_LOCK_EXCLUSIVE);
1270  opage = BufferGetPage(bucket_obuf);
1271  oopaque = HashPageGetOpaque(opage);
1272 
1273  LockBuffer(bucket_nbuf, BUFFER_LOCK_EXCLUSIVE);
1274  npage = BufferGetPage(bucket_nbuf);
1275  nopaque = HashPageGetOpaque(npage);
1276 
1278 
1279  oopaque->hasho_flag &= ~LH_BUCKET_BEING_SPLIT;
1281 
1282  /*
1283  * After the split is finished, mark the old bucket to indicate that it
1284  * contains deletable tuples. We will clear split-cleanup flag after
1285  * deleting such tuples either at the end of split or at the next split
1286  * from old bucket or at the time of vacuum.
1287  */
1289 
1290  /*
1291  * now write the buffers, here we don't release the locks as caller is
1292  * responsible to release locks.
1293  */
1294  MarkBufferDirty(bucket_obuf);
1295  MarkBufferDirty(bucket_nbuf);
1296 
1297  if (RelationNeedsWAL(rel))
1298  {
1299  XLogRecPtr recptr;
1300  xl_hash_split_complete xlrec;
1301 
1302  xlrec.old_bucket_flag = oopaque->hasho_flag;
1303  xlrec.new_bucket_flag = nopaque->hasho_flag;
1304 
1305  XLogBeginInsert();
1306 
1307  XLogRegisterData((char *) &xlrec, SizeOfHashSplitComplete);
1308 
1309  XLogRegisterBuffer(0, bucket_obuf, REGBUF_STANDARD);
1310  XLogRegisterBuffer(1, bucket_nbuf, REGBUF_STANDARD);
1311 
1312  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_COMPLETE);
1313 
1314  PageSetLSN(BufferGetPage(bucket_obuf), recptr);
1315  PageSetLSN(BufferGetPage(bucket_nbuf), recptr);
1316  }
1317 
1318  END_CRIT_SECTION();
1319 
1320  /*
1321  * If possible, clean up the old bucket. We might not be able to do this
1322  * if someone else has a pin on it, but if not then we can go ahead. This
1323  * isn't absolutely necessary, but it reduces bloat; if we don't do it
1324  * now, VACUUM will do it eventually, but maybe not until new overflow
1325  * pages have been allocated. Note that there's no need to clean up the
1326  * new bucket.
1327  */
1328  if (IsBufferCleanupOK(bucket_obuf))
1329  {
1330  LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1331  hashbucketcleanup(rel, obucket, bucket_obuf,
1332  BufferGetBlockNumber(bucket_obuf), NULL,
1333  maxbucket, highmask, lowmask, NULL, NULL, true,
1334  NULL, NULL);
1335  }
1336  else
1337  {
1338  LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1339  LockBuffer(bucket_obuf, BUFFER_LOCK_UNLOCK);
1340  }
1341 }
Size PageGetFreeSpaceForMultipleTuples(Page page, int ntups)
Definition: bufpage.c:923
unsigned short uint16
Definition: c.h:491
size_t Size
Definition: c.h:584
#define INDEX_MOVED_BY_SPLIT_MASK
Definition: hash.h:293
#define LH_BUCKET_NEEDS_SPLIT_CLEANUP
Definition: hash.h:60
#define SizeOfHashSplitComplete
Definition: hash_xlog.h:117
#define XLOG_HASH_SPLIT_COMPLETE
Definition: hash_xlog.h:33
void _hash_pgaddmultitup(Relation rel, Buffer buf, IndexTuple *itups, OffsetNumber *itup_offsets, uint16 nitups)
Definition: hashinsert.c:331
Buffer _hash_addovflpage(Relation rel, Buffer metabuf, Buffer buf, bool retain_pin)
Definition: hashovfl.c:112
static void log_split_page(Relation rel, Buffer buf)
Definition: hashpage.c:1474
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:291
@ HASH_FIND
Definition: hsearch.h:113
IndexTuple CopyIndexTuple(IndexTuple source)
Definition: indextuple.c:547
#define ItemIdIsDead(itemId)
Definition: itemid.h:113
#define IndexTupleSize(itup)
Definition: itup.h:70
#define MaxIndexTuplesPerPage
Definition: itup.h:165
void pfree(void *pointer)
Definition: mcxt.c:1521
void PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition: predicate.c:3134
unsigned short t_info
Definition: itup.h:49

References _hash_addovflpage(), _hash_get_indextuple_hashkey(), _hash_getbuf(), _hash_hashkey2bucket(), _hash_pgaddmultitup(), _hash_relbuf(), Assert, BlockNumberIsValid(), BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage(), CopyIndexTuple(), END_CRIT_SECTION, FirstOffsetNumber, HASH_FIND, HASH_READ, hash_search(), hashbucketcleanup(), HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageGetOpaque, i, INDEX_MOVED_BY_SPLIT_MASK, IndexTupleSize, IsBufferCleanupOK(), ItemIdIsDead, LH_BUCKET_BEING_POPULATED, LH_BUCKET_BEING_SPLIT, LH_BUCKET_NEEDS_SPLIT_CLEANUP, LH_OVERFLOW_PAGE, LockBuffer(), log_split_page(), MarkBufferDirty(), MAXALIGN, MaxIndexTuplesPerPage, xl_hash_split_complete::new_bucket_flag, OffsetNumberNext, xl_hash_split_complete::old_bucket_flag, PageGetFreeSpaceForMultipleTuples(), PageGetItem(), PageGetItemId(), PageGetMaxOffsetNumber(), PageSetLSN(), pfree(), PredicateLockPageSplit(), REGBUF_STANDARD, RelationNeedsWAL, SizeOfHashSplitComplete, START_CRIT_SECTION, IndexTupleData::t_info, IndexTupleData::t_tid, XLOG_HASH_SPLIT_COMPLETE, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _hash_expandtable(), and _hash_finish_split().

◆ log_split_page()

static void log_split_page ( Relation  rel,
Buffer  buf 
)
static

Definition at line 1474 of file hashpage.c.

1475 {
1476  if (RelationNeedsWAL(rel))
1477  {
1478  XLogRecPtr recptr;
1479 
1480  XLogBeginInsert();
1481 
1483 
1484  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_PAGE);
1485 
1486  PageSetLSN(BufferGetPage(buf), recptr);
1487  }
1488 }
#define XLOG_HASH_SPLIT_PAGE
Definition: hash_xlog.h:32
#define REGBUF_FORCE_IMAGE
Definition: xloginsert.h:31

References buf, BufferGetPage(), PageSetLSN(), REGBUF_FORCE_IMAGE, REGBUF_STANDARD, RelationNeedsWAL, XLOG_HASH_SPLIT_PAGE, XLogBeginInsert(), XLogInsert(), and XLogRegisterBuffer().

Referenced by _hash_splitbucket().