PostgreSQL Source Code  git master
nbtree.h File Reference
#include "access/amapi.h"
#include "access/itup.h"
#include "access/sdir.h"
#include "access/tableam.h"
#include "access/xlogreader.h"
#include "catalog/pg_am_d.h"
#include "catalog/pg_index.h"
#include "lib/stringinfo.h"
#include "storage/bufmgr.h"
#include "storage/shm_toc.h"
Include dependency graph for nbtree.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures

struct  BTPageOpaqueData
 
struct  BTMetaPageData
 
struct  BTDeletedPageData
 
struct  BTPendingFSM
 
struct  BTVacState
 
struct  BTStackData
 
struct  BTScanInsertData
 
struct  BTInsertStateData
 
struct  BTDedupInterval
 
struct  BTDedupStateData
 
struct  BTVacuumPostingData
 
struct  BTScanPosItem
 
struct  BTScanPosData
 
struct  BTArrayKeyInfo
 
struct  BTScanOpaqueData
 
struct  BTOptions
 

Macros

#define BTPageGetOpaque(page)   ((BTPageOpaque) PageGetSpecialPointer(page))
 
#define BTP_LEAF   (1 << 0) /* leaf page, i.e. not internal page */
 
#define BTP_ROOT   (1 << 1) /* root page (has no parent) */
 
#define BTP_DELETED   (1 << 2) /* page has been deleted from tree */
 
#define BTP_META   (1 << 3) /* meta-page */
 
#define BTP_HALF_DEAD   (1 << 4) /* empty, but still in tree */
 
#define BTP_SPLIT_END   (1 << 5) /* rightmost page of split group */
 
#define BTP_HAS_GARBAGE   (1 << 6) /* page has LP_DEAD tuples (deprecated) */
 
#define BTP_INCOMPLETE_SPLIT   (1 << 7) /* right sibling's downlink is missing */
 
#define BTP_HAS_FULLXID   (1 << 8) /* contains BTDeletedPageData */
 
#define MAX_BT_CYCLE_ID   0xFF7F
 
#define BTPageGetMeta(p)    ((BTMetaPageData *) PageGetContents(p))
 
#define BTREE_METAPAGE   0 /* first page is meta */
 
#define BTREE_MAGIC   0x053162 /* magic number in metapage */
 
#define BTREE_VERSION   4 /* current version number */
 
#define BTREE_MIN_VERSION   2 /* minimum supported version */
 
#define BTREE_NOVAC_VERSION   3 /* version with all meta fields set */
 
#define BTMaxItemSize(page)
 
#define BTMaxItemSizeNoHeapTid(page)
 
#define MaxTIDsPerBTreePage
 
#define BTREE_MIN_FILLFACTOR   10
 
#define BTREE_DEFAULT_FILLFACTOR   90
 
#define BTREE_NONLEAF_FILLFACTOR   70
 
#define BTREE_SINGLEVAL_FILLFACTOR   96
 
#define P_NONE   0
 
#define P_LEFTMOST(opaque)   ((opaque)->btpo_prev == P_NONE)
 
#define P_RIGHTMOST(opaque)   ((opaque)->btpo_next == P_NONE)
 
#define P_ISLEAF(opaque)   (((opaque)->btpo_flags & BTP_LEAF) != 0)
 
#define P_ISROOT(opaque)   (((opaque)->btpo_flags & BTP_ROOT) != 0)
 
#define P_ISDELETED(opaque)   (((opaque)->btpo_flags & BTP_DELETED) != 0)
 
#define P_ISMETA(opaque)   (((opaque)->btpo_flags & BTP_META) != 0)
 
#define P_ISHALFDEAD(opaque)   (((opaque)->btpo_flags & BTP_HALF_DEAD) != 0)
 
#define P_IGNORE(opaque)   (((opaque)->btpo_flags & (BTP_DELETED|BTP_HALF_DEAD)) != 0)
 
#define P_HAS_GARBAGE(opaque)   (((opaque)->btpo_flags & BTP_HAS_GARBAGE) != 0)
 
#define P_INCOMPLETE_SPLIT(opaque)   (((opaque)->btpo_flags & BTP_INCOMPLETE_SPLIT) != 0)
 
#define P_HAS_FULLXID(opaque)   (((opaque)->btpo_flags & BTP_HAS_FULLXID) != 0)
 
#define P_HIKEY   ((OffsetNumber) 1)
 
#define P_FIRSTKEY   ((OffsetNumber) 2)
 
#define P_FIRSTDATAKEY(opaque)   (P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY)
 
#define INDEX_ALT_TID_MASK   INDEX_AM_RESERVED_BIT
 
#define BT_OFFSET_MASK   0x0FFF
 
#define BT_STATUS_OFFSET_MASK   0xF000
 
#define BT_PIVOT_HEAP_TID_ATTR   0x1000
 
#define BT_IS_POSTING   0x2000
 
#define BTreeTupleGetNAtts(itup, rel)
 
#define BTCommuteStrategyNumber(strat)   (BTMaxStrategyNumber + 1 - (strat))
 
#define BTORDER_PROC   1
 
#define BTSORTSUPPORT_PROC   2
 
#define BTINRANGE_PROC   3
 
#define BTEQUALIMAGE_PROC   4
 
#define BTOPTIONS_PROC   5
 
#define BTNProcs   5
 
#define BT_READ   BUFFER_LOCK_SHARE
 
#define BT_WRITE   BUFFER_LOCK_EXCLUSIVE
 
#define BTScanPosIsPinned(scanpos)
 
#define BTScanPosUnpin(scanpos)
 
#define BTScanPosUnpinIfPinned(scanpos)
 
#define BTScanPosIsValid(scanpos)
 
#define BTScanPosInvalidate(scanpos)
 
#define SK_BT_REQFWD   0x00010000 /* required to continue forward scan */
 
#define SK_BT_REQBKWD   0x00020000 /* required to continue backward scan */
 
#define SK_BT_INDOPTION_SHIFT   24 /* must clear the above bits */
 
#define SK_BT_DESC   (INDOPTION_DESC << SK_BT_INDOPTION_SHIFT)
 
#define SK_BT_NULLS_FIRST   (INDOPTION_NULLS_FIRST << SK_BT_INDOPTION_SHIFT)
 
#define BTGetFillFactor(relation)
 
#define BTGetTargetPageFreeSpace(relation)    (BLCKSZ * (100 - BTGetFillFactor(relation)) / 100)
 
#define BTGetDeduplicateItems(relation)
 
#define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN   2
 
#define PROGRESS_BTREE_PHASE_PERFORMSORT_1   3
 
#define PROGRESS_BTREE_PHASE_PERFORMSORT_2   4
 
#define PROGRESS_BTREE_PHASE_LEAF_LOAD   5
 

Typedefs

typedef uint16 BTCycleId
 
typedef struct BTPageOpaqueData BTPageOpaqueData
 
typedef BTPageOpaqueDataBTPageOpaque
 
typedef struct BTMetaPageData BTMetaPageData
 
typedef struct BTDeletedPageData BTDeletedPageData
 
typedef struct BTPendingFSM BTPendingFSM
 
typedef struct BTVacState BTVacState
 
typedef struct BTStackData BTStackData
 
typedef BTStackDataBTStack
 
typedef struct BTScanInsertData BTScanInsertData
 
typedef BTScanInsertDataBTScanInsert
 
typedef struct BTInsertStateData BTInsertStateData
 
typedef BTInsertStateDataBTInsertState
 
typedef struct BTDedupInterval BTDedupInterval
 
typedef struct BTDedupStateData BTDedupStateData
 
typedef BTDedupStateDataBTDedupState
 
typedef struct BTVacuumPostingData BTVacuumPostingData
 
typedef BTVacuumPostingDataBTVacuumPosting
 
typedef struct BTScanPosItem BTScanPosItem
 
typedef struct BTScanPosData BTScanPosData
 
typedef BTScanPosDataBTScanPos
 
typedef struct BTArrayKeyInfo BTArrayKeyInfo
 
typedef struct BTScanOpaqueData BTScanOpaqueData
 
typedef BTScanOpaqueDataBTScanOpaque
 
typedef struct BTOptions BTOptions
 

Functions

static void BTPageSetDeleted (Page page, FullTransactionId safexid)
 
static FullTransactionId BTPageGetDeleteXid (Page page)
 
static bool BTPageIsRecyclable (Page page, Relation heaprel)
 
 StaticAssertDecl (BT_OFFSET_MASK >=INDEX_MAX_KEYS, "BT_OFFSET_MASK can't fit INDEX_MAX_KEYS")
 
static bool BTreeTupleIsPivot (IndexTuple itup)
 
static bool BTreeTupleIsPosting (IndexTuple itup)
 
static void BTreeTupleSetPosting (IndexTuple itup, uint16 nhtids, int postingoffset)
 
static uint16 BTreeTupleGetNPosting (IndexTuple posting)
 
static uint32 BTreeTupleGetPostingOffset (IndexTuple posting)
 
static ItemPointer BTreeTupleGetPosting (IndexTuple posting)
 
static ItemPointer BTreeTupleGetPostingN (IndexTuple posting, int n)
 
static BlockNumber BTreeTupleGetDownLink (IndexTuple pivot)
 
static void BTreeTupleSetDownLink (IndexTuple pivot, BlockNumber blkno)
 
static void BTreeTupleSetNAtts (IndexTuple itup, uint16 nkeyatts, bool heaptid)
 
static BlockNumber BTreeTupleGetTopParent (IndexTuple leafhikey)
 
static void BTreeTupleSetTopParent (IndexTuple leafhikey, BlockNumber blkno)
 
static ItemPointer BTreeTupleGetHeapTID (IndexTuple itup)
 
static ItemPointer BTreeTupleGetMaxHeapTID (IndexTuple itup)
 
void btbuildempty (Relation index)
 
bool btinsert (Relation rel, Datum *values, bool *isnull, ItemPointer ht_ctid, Relation heapRel, IndexUniqueCheck checkUnique, bool indexUnchanged, struct IndexInfo *indexInfo)
 
IndexScanDesc btbeginscan (Relation rel, int nkeys, int norderbys)
 
Size btestimateparallelscan (void)
 
void btinitparallelscan (void *target)
 
bool btgettuple (IndexScanDesc scan, ScanDirection dir)
 
int64 btgetbitmap (IndexScanDesc scan, TIDBitmap *tbm)
 
void btrescan (IndexScanDesc scan, ScanKey scankey, int nscankeys, ScanKey orderbys, int norderbys)
 
void btparallelrescan (IndexScanDesc scan)
 
void btendscan (IndexScanDesc scan)
 
void btmarkpos (IndexScanDesc scan)
 
void btrestrpos (IndexScanDesc scan)
 
IndexBulkDeleteResultbtbulkdelete (IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state)
 
IndexBulkDeleteResultbtvacuumcleanup (IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
 
bool btcanreturn (Relation index, int attno)
 
bool _bt_parallel_seize (IndexScanDesc scan, BlockNumber *pageno)
 
void _bt_parallel_release (IndexScanDesc scan, BlockNumber scan_page)
 
void _bt_parallel_done (IndexScanDesc scan)
 
void _bt_parallel_advance_array_keys (IndexScanDesc scan)
 
void _bt_dedup_pass (Relation rel, Buffer buf, IndexTuple newitem, Size newitemsz, bool bottomupdedup)
 
bool _bt_bottomupdel_pass (Relation rel, Buffer buf, Relation heapRel, Size newitemsz)
 
void _bt_dedup_start_pending (BTDedupState state, IndexTuple base, OffsetNumber baseoff)
 
bool _bt_dedup_save_htid (BTDedupState state, IndexTuple itup)
 
Size _bt_dedup_finish_pending (Page newpage, BTDedupState state)
 
IndexTuple _bt_form_posting (IndexTuple base, ItemPointer htids, int nhtids)
 
void _bt_update_posting (BTVacuumPosting vacposting)
 
IndexTuple _bt_swap_posting (IndexTuple newitem, IndexTuple oposting, int postingoff)
 
bool _bt_doinsert (Relation rel, IndexTuple itup, IndexUniqueCheck checkUnique, bool indexUnchanged, Relation heapRel)
 
void _bt_finish_split (Relation rel, Relation heaprel, Buffer lbuf, BTStack stack)
 
Buffer _bt_getstackbuf (Relation rel, Relation heaprel, BTStack stack, BlockNumber child)
 
OffsetNumber _bt_findsplitloc (Relation rel, Page origpage, OffsetNumber newitemoff, Size newitemsz, IndexTuple newitem, bool *newitemonleft)
 
void _bt_initmetapage (Page page, BlockNumber rootbknum, uint32 level, bool allequalimage)
 
bool _bt_vacuum_needs_cleanup (Relation rel, Relation heaprel)
 
void _bt_set_cleanup_info (Relation rel, Relation heaprel, BlockNumber num_delpages)
 
void _bt_upgrademetapage (Page page)
 
Buffer _bt_getroot (Relation rel, Relation heaprel, int access)
 
Buffer _bt_gettrueroot (Relation rel, Relation heaprel)
 
int _bt_getrootheight (Relation rel, Relation heaprel)
 
void _bt_metaversion (Relation rel, Relation heaprel, bool *heapkeyspace, bool *allequalimage)
 
void _bt_checkpage (Relation rel, Buffer buf)
 
Buffer _bt_getbuf (Relation rel, Relation heaprel, BlockNumber blkno, int access)
 
Buffer _bt_relandgetbuf (Relation rel, Buffer obuf, BlockNumber blkno, int access)
 
void _bt_relbuf (Relation rel, Buffer buf)
 
void _bt_lockbuf (Relation rel, Buffer buf, int access)
 
void _bt_unlockbuf (Relation rel, Buffer buf)
 
bool _bt_conditionallockbuf (Relation rel, Buffer buf)
 
void _bt_upgradelockbufcleanup (Relation rel, Buffer buf)
 
void _bt_pageinit (Page page, Size size)
 
void _bt_delitems_vacuum (Relation rel, Buffer buf, OffsetNumber *deletable, int ndeletable, BTVacuumPosting *updatable, int nupdatable)
 
void _bt_delitems_delete_check (Relation rel, Buffer buf, Relation heapRel, TM_IndexDeleteOp *delstate)
 
void _bt_pagedel (Relation rel, Buffer leafbuf, BTVacState *vstate)
 
void _bt_pendingfsm_init (Relation rel, BTVacState *vstate, bool cleanuponly)
 
void _bt_pendingfsm_finalize (Relation rel, BTVacState *vstate)
 
BTStack _bt_search (Relation rel, Relation heaprel, BTScanInsert key, Buffer *bufP, int access, Snapshot snapshot)
 
Buffer _bt_moveright (Relation rel, Relation heaprel, BTScanInsert key, Buffer buf, bool forupdate, BTStack stack, int access, Snapshot snapshot)
 
OffsetNumber _bt_binsrch_insert (Relation rel, BTInsertState insertstate)
 
int32 _bt_compare (Relation rel, BTScanInsert key, Page page, OffsetNumber offnum)
 
bool _bt_first (IndexScanDesc scan, ScanDirection dir)
 
bool _bt_next (IndexScanDesc scan, ScanDirection dir)
 
Buffer _bt_get_endpoint (Relation rel, Relation heaprel, uint32 level, bool rightmost, Snapshot snapshot)
 
BTScanInsert _bt_mkscankey (Relation rel, Relation heaprel, IndexTuple itup)
 
void _bt_freestack (BTStack stack)
 
void _bt_preprocess_array_keys (IndexScanDesc scan)
 
void _bt_start_array_keys (IndexScanDesc scan, ScanDirection dir)
 
bool _bt_advance_array_keys (IndexScanDesc scan, ScanDirection dir)
 
void _bt_mark_array_keys (IndexScanDesc scan)
 
void _bt_restore_array_keys (IndexScanDesc scan)
 
void _bt_preprocess_keys (IndexScanDesc scan)
 
bool _bt_checkkeys (IndexScanDesc scan, IndexTuple tuple, int tupnatts, ScanDirection dir, bool *continuescan)
 
void _bt_killitems (IndexScanDesc scan)
 
BTCycleId _bt_vacuum_cycleid (Relation rel)
 
BTCycleId _bt_start_vacuum (Relation rel)
 
void _bt_end_vacuum (Relation rel)
 
void _bt_end_vacuum_callback (int code, Datum arg)
 
Size BTreeShmemSize (void)
 
void BTreeShmemInit (void)
 
byteabtoptions (Datum reloptions, bool validate)
 
bool btproperty (Oid index_oid, int attno, IndexAMProperty prop, const char *propname, bool *res, bool *isnull)
 
char * btbuildphasename (int64 phasenum)
 
IndexTuple _bt_truncate (Relation rel, IndexTuple lastleft, IndexTuple firstright, BTScanInsert itup_key)
 
int _bt_keep_natts_fast (Relation rel, IndexTuple lastleft, IndexTuple firstright)
 
bool _bt_check_natts (Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
 
void _bt_check_third_page (Relation rel, Relation heap, bool needheaptidspace, Page page, IndexTuple newtup)
 
bool _bt_allequalimage (Relation rel, bool debugmessage)
 
bool btvalidate (Oid opclassoid)
 
void btadjustmembers (Oid opfamilyoid, Oid opclassoid, List *operators, List *functions)
 
IndexBuildResultbtbuild (Relation heap, Relation index, struct IndexInfo *indexInfo)
 
void _bt_parallel_build_main (dsm_segment *seg, shm_toc *toc)
 

Macro Definition Documentation

◆ BT_IS_POSTING

#define BT_IS_POSTING   0x2000

Definition at line 463 of file nbtree.h.

◆ BT_OFFSET_MASK

#define BT_OFFSET_MASK   0x0FFF

Definition at line 459 of file nbtree.h.

◆ BT_PIVOT_HEAP_TID_ATTR

#define BT_PIVOT_HEAP_TID_ATTR   0x1000

Definition at line 462 of file nbtree.h.

◆ BT_READ

#define BT_READ   BUFFER_LOCK_SHARE

Definition at line 716 of file nbtree.h.

◆ BT_STATUS_OFFSET_MASK

#define BT_STATUS_OFFSET_MASK   0xF000

Definition at line 460 of file nbtree.h.

◆ BT_WRITE

#define BT_WRITE   BUFFER_LOCK_EXCLUSIVE

Definition at line 717 of file nbtree.h.

◆ BTCommuteStrategyNumber

#define BTCommuteStrategyNumber (   strat)    (BTMaxStrategyNumber + 1 - (strat))

Definition at line 682 of file nbtree.h.

◆ BTEQUALIMAGE_PROC

#define BTEQUALIMAGE_PROC   4

Definition at line 707 of file nbtree.h.

◆ BTGetDeduplicateItems

#define BTGetDeduplicateItems (   relation)
Value:
(AssertMacro(relation->rd_rel->relkind == RELKIND_INDEX && \
relation->rd_rel->relam == BTREE_AM_OID), \
((relation)->rd_options ? \
((BTOptions *) (relation)->rd_options)->deduplicate_items : true))
#define AssertMacro(condition)
Definition: c.h:843

Definition at line 1105 of file nbtree.h.

◆ BTGetFillFactor

#define BTGetFillFactor (   relation)
Value:
(AssertMacro(relation->rd_rel->relkind == RELKIND_INDEX && \
relation->rd_rel->relam == BTREE_AM_OID), \
(relation)->rd_options ? \
((BTOptions *) (relation)->rd_options)->fillfactor : \
BTREE_DEFAULT_FILLFACTOR)

Definition at line 1097 of file nbtree.h.

◆ BTGetTargetPageFreeSpace

#define BTGetTargetPageFreeSpace (   relation)     (BLCKSZ * (100 - BTGetFillFactor(relation)) / 100)

Definition at line 1103 of file nbtree.h.

◆ BTINRANGE_PROC

#define BTINRANGE_PROC   3

Definition at line 706 of file nbtree.h.

◆ BTMaxItemSize

#define BTMaxItemSize (   page)
Value:
MAXALIGN(sizeof(BTPageOpaqueData))) / 3) - \
MAXALIGN(sizeof(ItemPointerData)))
static Size PageGetPageSize(Page page)
Definition: bufpage.h:273
#define SizeOfPageHeaderData
Definition: bufpage.h:213
#define MAXALIGN_DOWN(LEN)
Definition: c.h:807
#define MAXALIGN(LEN)
Definition: c.h:795

Definition at line 164 of file nbtree.h.

◆ BTMaxItemSizeNoHeapTid

#define BTMaxItemSizeNoHeapTid (   page)
Value:

Definition at line 169 of file nbtree.h.

◆ BTNProcs

#define BTNProcs   5

Definition at line 709 of file nbtree.h.

◆ BTOPTIONS_PROC

#define BTOPTIONS_PROC   5

Definition at line 708 of file nbtree.h.

◆ BTORDER_PROC

#define BTORDER_PROC   1

Definition at line 704 of file nbtree.h.

◆ BTP_DELETED

#define BTP_DELETED   (1 << 2) /* page has been deleted from tree */

Definition at line 78 of file nbtree.h.

◆ BTP_HALF_DEAD

#define BTP_HALF_DEAD   (1 << 4) /* empty, but still in tree */

Definition at line 80 of file nbtree.h.

◆ BTP_HAS_FULLXID

#define BTP_HAS_FULLXID   (1 << 8) /* contains BTDeletedPageData */

Definition at line 84 of file nbtree.h.

◆ BTP_HAS_GARBAGE

#define BTP_HAS_GARBAGE   (1 << 6) /* page has LP_DEAD tuples (deprecated) */

Definition at line 82 of file nbtree.h.

◆ BTP_INCOMPLETE_SPLIT

#define BTP_INCOMPLETE_SPLIT   (1 << 7) /* right sibling's downlink is missing */

Definition at line 83 of file nbtree.h.

◆ BTP_LEAF

#define BTP_LEAF   (1 << 0) /* leaf page, i.e. not internal page */

Definition at line 76 of file nbtree.h.

◆ BTP_META

#define BTP_META   (1 << 3) /* meta-page */

Definition at line 79 of file nbtree.h.

◆ BTP_ROOT

#define BTP_ROOT   (1 << 1) /* root page (has no parent) */

Definition at line 77 of file nbtree.h.

◆ BTP_SPLIT_END

#define BTP_SPLIT_END   (1 << 5) /* rightmost page of split group */

Definition at line 81 of file nbtree.h.

◆ BTPageGetMeta

#define BTPageGetMeta (   p)     ((BTMetaPageData *) PageGetContents(p))

Definition at line 121 of file nbtree.h.

◆ BTPageGetOpaque

#define BTPageGetOpaque (   page)    ((BTPageOpaque) PageGetSpecialPointer(page))

Definition at line 73 of file nbtree.h.

◆ BTREE_DEFAULT_FILLFACTOR

#define BTREE_DEFAULT_FILLFACTOR   90

Definition at line 200 of file nbtree.h.

◆ BTREE_MAGIC

#define BTREE_MAGIC   0x053162 /* magic number in metapage */

Definition at line 149 of file nbtree.h.

◆ BTREE_METAPAGE

#define BTREE_METAPAGE   0 /* first page is meta */

Definition at line 148 of file nbtree.h.

◆ BTREE_MIN_FILLFACTOR

#define BTREE_MIN_FILLFACTOR   10

Definition at line 199 of file nbtree.h.

◆ BTREE_MIN_VERSION

#define BTREE_MIN_VERSION   2 /* minimum supported version */

Definition at line 151 of file nbtree.h.

◆ BTREE_NONLEAF_FILLFACTOR

#define BTREE_NONLEAF_FILLFACTOR   70

Definition at line 201 of file nbtree.h.

◆ BTREE_NOVAC_VERSION

#define BTREE_NOVAC_VERSION   3 /* version with all meta fields set */

Definition at line 152 of file nbtree.h.

◆ BTREE_SINGLEVAL_FILLFACTOR

#define BTREE_SINGLEVAL_FILLFACTOR   96

Definition at line 202 of file nbtree.h.

◆ BTREE_VERSION

#define BTREE_VERSION   4 /* current version number */

Definition at line 150 of file nbtree.h.

◆ BTreeTupleGetNAtts

#define BTreeTupleGetNAtts (   itup,
  rel 
)
Value:
( \
(BTreeTupleIsPivot(itup)) ? \
( \
) \
: \
)
static OffsetNumber ItemPointerGetOffsetNumberNoCheck(const ItemPointerData *pointer)
Definition: itemptr.h:114
static bool BTreeTupleIsPivot(IndexTuple itup)
Definition: nbtree.h:477
#define BT_OFFSET_MASK
Definition: nbtree.h:459
#define IndexRelationGetNumberOfAttributes(relation)
Definition: rel.h:516

Definition at line 574 of file nbtree.h.

◆ BTScanPosInvalidate

#define BTScanPosInvalidate (   scanpos)
Value:
do { \
(scanpos).currPage = InvalidBlockNumber; \
(scanpos).nextPage = InvalidBlockNumber; \
(scanpos).buf = InvalidBuffer; \
(scanpos).lsn = InvalidXLogRecPtr; \
(scanpos).nextTupleOffset = 0; \
} while (0)
#define InvalidBlockNumber
Definition: block.h:33
#define InvalidBuffer
Definition: buf.h:25
static char * buf
Definition: pg_test_fsync.c:67
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28

Definition at line 1015 of file nbtree.h.

◆ BTScanPosIsPinned

#define BTScanPosIsPinned (   scanpos)
Value:
( \
AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
!BufferIsValid((scanpos).buf)), \
BufferIsValid((scanpos).buf) \
)
static bool BlockNumberIsValid(BlockNumber blockNumber)
Definition: block.h:71
static bool BufferIsValid(Buffer bufnum)
Definition: bufmgr.h:303

Definition at line 992 of file nbtree.h.

◆ BTScanPosIsValid

#define BTScanPosIsValid (   scanpos)
Value:
( \
AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
!BufferIsValid((scanpos).buf)), \
BlockNumberIsValid((scanpos).currPage) \
)

Definition at line 1009 of file nbtree.h.

◆ BTScanPosUnpin

#define BTScanPosUnpin (   scanpos)
Value:
do { \
ReleaseBuffer((scanpos).buf); \
(scanpos).buf = InvalidBuffer; \
} while (0)

Definition at line 998 of file nbtree.h.

◆ BTScanPosUnpinIfPinned

#define BTScanPosUnpinIfPinned (   scanpos)
Value:
do { \
if (BTScanPosIsPinned(scanpos)) \
BTScanPosUnpin(scanpos); \
} while (0)
#define BTScanPosIsPinned(scanpos)
Definition: nbtree.h:992

Definition at line 1003 of file nbtree.h.

◆ BTSORTSUPPORT_PROC

#define BTSORTSUPPORT_PROC   2

Definition at line 705 of file nbtree.h.

◆ INDEX_ALT_TID_MASK

#define INDEX_ALT_TID_MASK   INDEX_AM_RESERVED_BIT

Definition at line 456 of file nbtree.h.

◆ MAX_BT_CYCLE_ID

#define MAX_BT_CYCLE_ID   0xFF7F

Definition at line 93 of file nbtree.h.

◆ MaxTIDsPerBTreePage

#define MaxTIDsPerBTreePage
Value:
(int) ((BLCKSZ - SizeOfPageHeaderData - sizeof(BTPageOpaqueData)) / \
sizeof(ItemPointerData))

Definition at line 185 of file nbtree.h.

◆ P_FIRSTDATAKEY

#define P_FIRSTDATAKEY (   opaque)    (P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY)

Definition at line 366 of file nbtree.h.

◆ P_FIRSTKEY

#define P_FIRSTKEY   ((OffsetNumber) 2)

Definition at line 365 of file nbtree.h.

◆ P_HAS_FULLXID

#define P_HAS_FULLXID (   opaque)    (((opaque)->btpo_flags & BTP_HAS_FULLXID) != 0)

Definition at line 228 of file nbtree.h.

◆ P_HAS_GARBAGE

#define P_HAS_GARBAGE (   opaque)    (((opaque)->btpo_flags & BTP_HAS_GARBAGE) != 0)

Definition at line 226 of file nbtree.h.

◆ P_HIKEY

#define P_HIKEY   ((OffsetNumber) 1)

Definition at line 364 of file nbtree.h.

◆ P_IGNORE

#define P_IGNORE (   opaque)    (((opaque)->btpo_flags & (BTP_DELETED|BTP_HALF_DEAD)) != 0)

Definition at line 225 of file nbtree.h.

◆ P_INCOMPLETE_SPLIT

#define P_INCOMPLETE_SPLIT (   opaque)    (((opaque)->btpo_flags & BTP_INCOMPLETE_SPLIT) != 0)

Definition at line 227 of file nbtree.h.

◆ P_ISDELETED

#define P_ISDELETED (   opaque)    (((opaque)->btpo_flags & BTP_DELETED) != 0)

Definition at line 222 of file nbtree.h.

◆ P_ISHALFDEAD

#define P_ISHALFDEAD (   opaque)    (((opaque)->btpo_flags & BTP_HALF_DEAD) != 0)

Definition at line 224 of file nbtree.h.

◆ P_ISLEAF

#define P_ISLEAF (   opaque)    (((opaque)->btpo_flags & BTP_LEAF) != 0)

Definition at line 220 of file nbtree.h.

◆ P_ISMETA

#define P_ISMETA (   opaque)    (((opaque)->btpo_flags & BTP_META) != 0)

Definition at line 223 of file nbtree.h.

◆ P_ISROOT

#define P_ISROOT (   opaque)    (((opaque)->btpo_flags & BTP_ROOT) != 0)

Definition at line 221 of file nbtree.h.

◆ P_LEFTMOST

#define P_LEFTMOST (   opaque)    ((opaque)->btpo_prev == P_NONE)

Definition at line 218 of file nbtree.h.

◆ P_NONE

#define P_NONE   0

Definition at line 212 of file nbtree.h.

◆ P_RIGHTMOST

#define P_RIGHTMOST (   opaque)    ((opaque)->btpo_next == P_NONE)

Definition at line 219 of file nbtree.h.

◆ PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN

#define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN   2

Definition at line 1116 of file nbtree.h.

◆ PROGRESS_BTREE_PHASE_LEAF_LOAD

#define PROGRESS_BTREE_PHASE_LEAF_LOAD   5

Definition at line 1119 of file nbtree.h.

◆ PROGRESS_BTREE_PHASE_PERFORMSORT_1

#define PROGRESS_BTREE_PHASE_PERFORMSORT_1   3

Definition at line 1117 of file nbtree.h.

◆ PROGRESS_BTREE_PHASE_PERFORMSORT_2

#define PROGRESS_BTREE_PHASE_PERFORMSORT_2   4

Definition at line 1118 of file nbtree.h.

◆ SK_BT_DESC

#define SK_BT_DESC   (INDOPTION_DESC << SK_BT_INDOPTION_SHIFT)

Definition at line 1086 of file nbtree.h.

◆ SK_BT_INDOPTION_SHIFT

#define SK_BT_INDOPTION_SHIFT   24 /* must clear the above bits */

Definition at line 1085 of file nbtree.h.

◆ SK_BT_NULLS_FIRST

#define SK_BT_NULLS_FIRST   (INDOPTION_NULLS_FIRST << SK_BT_INDOPTION_SHIFT)

Definition at line 1087 of file nbtree.h.

◆ SK_BT_REQBKWD

#define SK_BT_REQBKWD   0x00020000 /* required to continue backward scan */

Definition at line 1084 of file nbtree.h.

◆ SK_BT_REQFWD

#define SK_BT_REQFWD   0x00010000 /* required to continue forward scan */

Definition at line 1083 of file nbtree.h.

Typedef Documentation

◆ BTArrayKeyInfo

◆ BTCycleId

typedef uint16 BTCycleId

Definition at line 29 of file nbtree.h.

◆ BTDedupInterval

◆ BTDedupState

Definition at line 895 of file nbtree.h.

◆ BTDedupStateData

◆ BTDeletedPageData

◆ BTInsertState

Definition at line 837 of file nbtree.h.

◆ BTInsertStateData

◆ BTMetaPageData

◆ BTOptions

typedef struct BTOptions BTOptions

◆ BTPageOpaque

Definition at line 71 of file nbtree.h.

◆ BTPageOpaqueData

◆ BTPendingFSM

typedef struct BTPendingFSM BTPendingFSM

◆ BTScanInsert

Definition at line 798 of file nbtree.h.

◆ BTScanInsertData

◆ BTScanOpaque

Definition at line 1076 of file nbtree.h.

◆ BTScanOpaqueData

◆ BTScanPos

Definition at line 990 of file nbtree.h.

◆ BTScanPosData

typedef struct BTScanPosData BTScanPosData

◆ BTScanPosItem

typedef struct BTScanPosItem BTScanPosItem

◆ BTStack

typedef BTStackData* BTStack

Definition at line 736 of file nbtree.h.

◆ BTStackData

typedef struct BTStackData BTStackData

◆ BTVacState

typedef struct BTVacState BTVacState

◆ BTVacuumPosting

Definition at line 916 of file nbtree.h.

◆ BTVacuumPostingData

Function Documentation

◆ _bt_advance_array_keys()

bool _bt_advance_array_keys ( IndexScanDesc  scan,
ScanDirection  dir 
)

Definition at line 551 of file nbtutils.c.

552 {
553  BTScanOpaque so = (BTScanOpaque) scan->opaque;
554  bool found = false;
555  int i;
556 
557  /*
558  * We must advance the last array key most quickly, since it will
559  * correspond to the lowest-order index column among the available
560  * qualifications. This is necessary to ensure correct ordering of output
561  * when there are multiple array keys.
562  */
563  for (i = so->numArrayKeys - 1; i >= 0; i--)
564  {
565  BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
566  ScanKey skey = &so->arrayKeyData[curArrayKey->scan_key];
567  int cur_elem = curArrayKey->cur_elem;
568  int num_elems = curArrayKey->num_elems;
569 
570  if (ScanDirectionIsBackward(dir))
571  {
572  if (--cur_elem < 0)
573  {
574  cur_elem = num_elems - 1;
575  found = false; /* need to advance next array key */
576  }
577  else
578  found = true;
579  }
580  else
581  {
582  if (++cur_elem >= num_elems)
583  {
584  cur_elem = 0;
585  found = false; /* need to advance next array key */
586  }
587  else
588  found = true;
589  }
590 
591  curArrayKey->cur_elem = cur_elem;
592  skey->sk_argument = curArrayKey->elem_values[cur_elem];
593  if (found)
594  break;
595  }
596 
597  /* advance parallel scan */
598  if (scan->parallel_scan != NULL)
600 
601  return found;
602 }
int i
Definition: isn.c:73
void _bt_parallel_advance_array_keys(IndexScanDesc scan)
Definition: nbtree.c:758
BTScanOpaqueData * BTScanOpaque
Definition: nbtree.h:1076
#define ScanDirectionIsBackward(direction)
Definition: sdir.h:50
Datum * elem_values
Definition: nbtree.h:1031
BTArrayKeyInfo * arrayKeys
Definition: nbtree.h:1047
ScanKey arrayKeyData
Definition: nbtree.h:1042
struct ParallelIndexScanDescData * parallel_scan
Definition: relscan.h:166
Datum sk_argument
Definition: skey.h:72

References _bt_parallel_advance_array_keys(), BTScanOpaqueData::arrayKeyData, BTScanOpaqueData::arrayKeys, BTArrayKeyInfo::cur_elem, BTArrayKeyInfo::elem_values, i, BTArrayKeyInfo::num_elems, BTScanOpaqueData::numArrayKeys, IndexScanDescData::opaque, IndexScanDescData::parallel_scan, BTArrayKeyInfo::scan_key, ScanDirectionIsBackward, and ScanKeyData::sk_argument.

Referenced by btgetbitmap(), and btgettuple().

◆ _bt_allequalimage()

bool _bt_allequalimage ( Relation  rel,
bool  debugmessage 
)

Definition at line 2690 of file nbtutils.c.

2691 {
2692  bool allequalimage = true;
2693 
2694  /* INCLUDE indexes can never support deduplication */
2697  return false;
2698 
2699  for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(rel); i++)
2700  {
2701  Oid opfamily = rel->rd_opfamily[i];
2702  Oid opcintype = rel->rd_opcintype[i];
2703  Oid collation = rel->rd_indcollation[i];
2704  Oid equalimageproc;
2705 
2706  equalimageproc = get_opfamily_proc(opfamily, opcintype, opcintype,
2708 
2709  /*
2710  * If there is no BTEQUALIMAGE_PROC then deduplication is assumed to
2711  * be unsafe. Otherwise, actually call proc and see what it says.
2712  */
2713  if (!OidIsValid(equalimageproc) ||
2714  !DatumGetBool(OidFunctionCall1Coll(equalimageproc, collation,
2715  ObjectIdGetDatum(opcintype))))
2716  {
2717  allequalimage = false;
2718  break;
2719  }
2720  }
2721 
2722  if (debugmessage)
2723  {
2724  if (allequalimage)
2725  elog(DEBUG1, "index \"%s\" can safely use deduplication",
2727  else
2728  elog(DEBUG1, "index \"%s\" cannot use deduplication",
2730  }
2731 
2732  return allequalimage;
2733 }
#define OidIsValid(objectId)
Definition: c.h:759
#define DEBUG1
Definition: elog.h:30
Datum OidFunctionCall1Coll(Oid functionId, Oid collation, Datum arg1)
Definition: fmgr.c:1382
Oid get_opfamily_proc(Oid opfamily, Oid lefttype, Oid righttype, int16 procnum)
Definition: lsyscache.c:795
#define BTEQUALIMAGE_PROC
Definition: nbtree.h:707
static bool DatumGetBool(Datum X)
Definition: postgres.h:90
static Datum ObjectIdGetDatum(Oid X)
Definition: postgres.h:252
unsigned int Oid
Definition: postgres_ext.h:31
#define RelationGetRelationName(relation)
Definition: rel.h:538
#define IndexRelationGetNumberOfKeyAttributes(relation)
Definition: rel.h:523
Oid * rd_opcintype
Definition: rel.h:207
Oid * rd_opfamily
Definition: rel.h:206
Oid * rd_indcollation
Definition: rel.h:216

References BTEQUALIMAGE_PROC, DatumGetBool(), DEBUG1, elog(), get_opfamily_proc(), i, IndexRelationGetNumberOfAttributes, IndexRelationGetNumberOfKeyAttributes, ObjectIdGetDatum(), OidFunctionCall1Coll(), OidIsValid, RelationData::rd_indcollation, RelationData::rd_opcintype, RelationData::rd_opfamily, and RelationGetRelationName.

Referenced by _bt_leafbuild(), bt_index_check_internal(), and btbuildempty().

◆ _bt_binsrch_insert()

OffsetNumber _bt_binsrch_insert ( Relation  rel,
BTInsertState  insertstate 
)

Definition at line 444 of file nbtsearch.c.

445 {
446  BTScanInsert key = insertstate->itup_key;
447  Page page;
448  BTPageOpaque opaque;
449  OffsetNumber low,
450  high,
451  stricthigh;
452  int32 result,
453  cmpval;
454 
455  page = BufferGetPage(insertstate->buf);
456  opaque = BTPageGetOpaque(page);
457 
458  Assert(P_ISLEAF(opaque));
459  Assert(!key->nextkey);
460  Assert(insertstate->postingoff == 0);
461 
462  if (!insertstate->bounds_valid)
463  {
464  /* Start new binary search */
465  low = P_FIRSTDATAKEY(opaque);
466  high = PageGetMaxOffsetNumber(page);
467  }
468  else
469  {
470  /* Restore result of previous binary search against same page */
471  low = insertstate->low;
472  high = insertstate->stricthigh;
473  }
474 
475  /* If there are no keys on the page, return the first available slot */
476  if (unlikely(high < low))
477  {
478  /* Caller can't reuse bounds */
479  insertstate->low = InvalidOffsetNumber;
480  insertstate->stricthigh = InvalidOffsetNumber;
481  insertstate->bounds_valid = false;
482  return low;
483  }
484 
485  /*
486  * Binary search to find the first key on the page >= scan key. (nextkey
487  * is always false when inserting).
488  *
489  * The loop invariant is: all slots before 'low' are < scan key, all slots
490  * at or after 'high' are >= scan key. 'stricthigh' is > scan key, and is
491  * maintained to save additional search effort for caller.
492  *
493  * We can fall out when high == low.
494  */
495  if (!insertstate->bounds_valid)
496  high++; /* establish the loop invariant for high */
497  stricthigh = high; /* high initially strictly higher */
498 
499  cmpval = 1; /* !nextkey comparison value */
500 
501  while (high > low)
502  {
503  OffsetNumber mid = low + ((high - low) / 2);
504 
505  /* We have low <= mid < high, so mid points at a real slot */
506 
507  result = _bt_compare(rel, key, page, mid);
508 
509  if (result >= cmpval)
510  low = mid + 1;
511  else
512  {
513  high = mid;
514  if (result != 0)
515  stricthigh = high;
516  }
517 
518  /*
519  * If tuple at offset located by binary search is a posting list whose
520  * TID range overlaps with caller's scantid, perform posting list
521  * binary search to set postingoff for caller. Caller must split the
522  * posting list when postingoff is set. This should happen
523  * infrequently.
524  */
525  if (unlikely(result == 0 && key->scantid != NULL))
526  {
527  /*
528  * postingoff should never be set more than once per leaf page
529  * binary search. That would mean that there are duplicate table
530  * TIDs in the index, which is never okay. Check for that here.
531  */
532  if (insertstate->postingoff != 0)
533  ereport(ERROR,
534  (errcode(ERRCODE_INDEX_CORRUPTED),
535  errmsg_internal("table tid from new index tuple (%u,%u) cannot find insert offset between offsets %u and %u of block %u in index \"%s\"",
536  ItemPointerGetBlockNumber(key->scantid),
538  low, stricthigh,
539  BufferGetBlockNumber(insertstate->buf),
540  RelationGetRelationName(rel))));
541 
542  insertstate->postingoff = _bt_binsrch_posting(key, page, mid);
543  }
544  }
545 
546  /*
547  * On a leaf page, a binary search always returns the first key >= scan
548  * key (at least in !nextkey case), which could be the last slot + 1. This
549  * is also the lower bound of cached search.
550  *
551  * stricthigh may also be the last slot + 1, which prevents caller from
552  * using bounds directly, but is still useful to us if we're called a
553  * second time with cached bounds (cached low will be < stricthigh when
554  * that happens).
555  */
556  insertstate->low = low;
557  insertstate->stricthigh = stricthigh;
558  insertstate->bounds_valid = true;
559 
560  return low;
561 }
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:3290
static Page BufferGetPage(Buffer buffer)
Definition: bufmgr.h:355
Pointer Page
Definition: bufpage.h:78
static OffsetNumber PageGetMaxOffsetNumber(Page page)
Definition: bufpage.h:369
signed int int32
Definition: c.h:478
#define unlikely(x)
Definition: c.h:295
int errmsg_internal(const char *fmt,...)
Definition: elog.c:1156
int errcode(int sqlerrcode)
Definition: elog.c:858
#define ERROR
Definition: elog.h:39
#define ereport(elevel,...)
Definition: elog.h:149
static OffsetNumber ItemPointerGetOffsetNumber(const ItemPointerData *pointer)
Definition: itemptr.h:124
static BlockNumber ItemPointerGetBlockNumber(const ItemPointerData *pointer)
Definition: itemptr.h:103
Assert(fmt[strlen(fmt) - 1] !='\n')
#define P_ISLEAF(opaque)
Definition: nbtree.h:220
#define BTPageGetOpaque(page)
Definition: nbtree.h:73
#define P_FIRSTDATAKEY(opaque)
Definition: nbtree.h:366
static int _bt_binsrch_posting(BTScanInsert key, Page page, OffsetNumber offnum)
Definition: nbtsearch.c:572
int32 _bt_compare(Relation rel, BTScanInsert key, Page page, OffsetNumber offnum)
Definition: nbtsearch.c:658
#define InvalidOffsetNumber
Definition: off.h:26
uint16 OffsetNumber
Definition: off.h:24
OffsetNumber stricthigh
Definition: nbtree.h:827
bool bounds_valid
Definition: nbtree.h:825
OffsetNumber low
Definition: nbtree.h:826
BTScanInsert itup_key
Definition: nbtree.h:815

References _bt_binsrch_posting(), _bt_compare(), Assert(), BTInsertStateData::bounds_valid, BTPageGetOpaque, BTInsertStateData::buf, BufferGetBlockNumber(), BufferGetPage(), ereport, errcode(), errmsg_internal(), ERROR, InvalidOffsetNumber, ItemPointerGetBlockNumber(), ItemPointerGetOffsetNumber(), BTInsertStateData::itup_key, sort-test::key, BTInsertStateData::low, P_FIRSTDATAKEY, P_ISLEAF, PageGetMaxOffsetNumber(), BTInsertStateData::postingoff, RelationGetRelationName, BTInsertStateData::stricthigh, and unlikely.

Referenced by _bt_check_unique(), _bt_findinsertloc(), and bt_rootdescend().

◆ _bt_bottomupdel_pass()

bool _bt_bottomupdel_pass ( Relation  rel,
Buffer  buf,
Relation  heapRel,
Size  newitemsz 
)

Definition at line 307 of file nbtdedup.c.

309 {
310  OffsetNumber offnum,
311  minoff,
312  maxoff;
313  Page page = BufferGetPage(buf);
314  BTPageOpaque opaque = BTPageGetOpaque(page);
316  TM_IndexDeleteOp delstate;
317  bool neverdedup;
318  int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
319 
320  /* Passed-in newitemsz is MAXALIGNED but does not include line pointer */
321  newitemsz += sizeof(ItemIdData);
322 
323  /* Initialize deduplication state */
325  state->deduplicate = true;
326  state->nmaxitems = 0;
327  state->maxpostingsize = BLCKSZ; /* We're not really deduplicating */
328  state->base = NULL;
329  state->baseoff = InvalidOffsetNumber;
330  state->basetupsize = 0;
331  state->htids = palloc(state->maxpostingsize);
332  state->nhtids = 0;
333  state->nitems = 0;
334  state->phystupsize = 0;
335  state->nintervals = 0;
336 
337  /*
338  * Initialize tableam state that describes bottom-up index deletion
339  * operation.
340  *
341  * We'll go on to ask the tableam to search for TIDs whose index tuples we
342  * can safely delete. The tableam will search until our leaf page space
343  * target is satisfied, or until the cost of continuing with the tableam
344  * operation seems too high. It focuses its efforts on TIDs associated
345  * with duplicate index tuples that we mark "promising".
346  *
347  * This space target is a little arbitrary. The tableam must be able to
348  * keep the costs and benefits in balance. We provide the tableam with
349  * exhaustive information about what might work, without directly
350  * concerning ourselves with avoiding work during the tableam call. Our
351  * role in costing the bottom-up deletion process is strictly advisory.
352  */
353  delstate.irel = rel;
354  delstate.iblknum = BufferGetBlockNumber(buf);
355  delstate.bottomup = true;
356  delstate.bottomupfreespace = Max(BLCKSZ / 16, newitemsz);
357  delstate.ndeltids = 0;
358  delstate.deltids = palloc(MaxTIDsPerBTreePage * sizeof(TM_IndexDelete));
359  delstate.status = palloc(MaxTIDsPerBTreePage * sizeof(TM_IndexStatus));
360 
361  minoff = P_FIRSTDATAKEY(opaque);
362  maxoff = PageGetMaxOffsetNumber(page);
363  for (offnum = minoff;
364  offnum <= maxoff;
365  offnum = OffsetNumberNext(offnum))
366  {
367  ItemId itemid = PageGetItemId(page, offnum);
368  IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
369 
370  Assert(!ItemIdIsDead(itemid));
371 
372  if (offnum == minoff)
373  {
374  /* itup starts first pending interval */
375  _bt_dedup_start_pending(state, itup, offnum);
376  }
377  else if (_bt_keep_natts_fast(rel, state->base, itup) > nkeyatts &&
378  _bt_dedup_save_htid(state, itup))
379  {
380  /* Tuple is equal; just added its TIDs to pending interval */
381  }
382  else
383  {
384  /* Finalize interval -- move its TIDs to delete state */
385  _bt_bottomupdel_finish_pending(page, state, &delstate);
386 
387  /* itup starts new pending interval */
388  _bt_dedup_start_pending(state, itup, offnum);
389  }
390  }
391  /* Finalize final interval -- move its TIDs to delete state */
392  _bt_bottomupdel_finish_pending(page, state, &delstate);
393 
394  /*
395  * We don't give up now in the event of having few (or even zero)
396  * promising tuples for the tableam because it's not up to us as the index
397  * AM to manage costs (note that the tableam might have heuristics of its
398  * own that work out what to do). We should at least avoid having our
399  * caller do a useless deduplication pass after we return in the event of
400  * zero promising tuples, though.
401  */
402  neverdedup = false;
403  if (state->nintervals == 0)
404  neverdedup = true;
405 
406  pfree(state->htids);
407  pfree(state);
408 
409  /* Ask tableam which TIDs are deletable, then physically delete them */
410  _bt_delitems_delete_check(rel, buf, heapRel, &delstate);
411 
412  pfree(delstate.deltids);
413  pfree(delstate.status);
414 
415  /* Report "success" to caller unconditionally to avoid deduplication */
416  if (neverdedup)
417  return true;
418 
419  /* Don't dedup when we won't end up back here any time soon anyway */
420  return PageGetExactFreeSpace(page) >= Max(BLCKSZ / 24, newitemsz);
421 }
Size PageGetExactFreeSpace(Page page)
Definition: bufpage.c:958
static Item PageGetItem(Page page, ItemId itemId)
Definition: bufpage.h:351
static ItemId PageGetItemId(Page page, OffsetNumber offsetNumber)
Definition: bufpage.h:240
#define Max(x, y)
Definition: c.h:982
struct ItemIdData ItemIdData
#define ItemIdIsDead(itemId)
Definition: itemid.h:113
IndexTupleData * IndexTuple
Definition: itup.h:53
void pfree(void *pointer)
Definition: mcxt.c:1456
void * palloc(Size size)
Definition: mcxt.c:1226
bool _bt_dedup_save_htid(BTDedupState state, IndexTuple itup)
Definition: nbtdedup.c:484
void _bt_dedup_start_pending(BTDedupState state, IndexTuple base, OffsetNumber baseoff)
Definition: nbtdedup.c:433
static void _bt_bottomupdel_finish_pending(Page page, BTDedupState state, TM_IndexDeleteOp *delstate)
Definition: nbtdedup.c:648
void _bt_delitems_delete_check(Relation rel, Buffer buf, Relation heapRel, TM_IndexDeleteOp *delstate)
Definition: nbtpage.c:1515
#define MaxTIDsPerBTreePage
Definition: nbtree.h:185
BTDedupStateData * BTDedupState
Definition: nbtree.h:895
int _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
Definition: nbtutils.c:2425
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
TM_IndexStatus * status
Definition: tableam.h:247
int bottomupfreespace
Definition: tableam.h:242
Relation irel
Definition: tableam.h:239
TM_IndexDelete * deltids
Definition: tableam.h:246
BlockNumber iblknum
Definition: tableam.h:240
Definition: regguts.h:323

References _bt_bottomupdel_finish_pending(), _bt_dedup_save_htid(), _bt_dedup_start_pending(), _bt_delitems_delete_check(), _bt_keep_natts_fast(), Assert(), TM_IndexDeleteOp::bottomup, TM_IndexDeleteOp::bottomupfreespace, BTPageGetOpaque, buf, BufferGetBlockNumber(), BufferGetPage(), TM_IndexDeleteOp::deltids, TM_IndexDeleteOp::iblknum, IndexRelationGetNumberOfKeyAttributes, InvalidOffsetNumber, TM_IndexDeleteOp::irel, ItemIdIsDead, Max, MaxTIDsPerBTreePage, TM_IndexDeleteOp::ndeltids, OffsetNumberNext, P_FIRSTDATAKEY, PageGetExactFreeSpace(), PageGetItem(), PageGetItemId(), PageGetMaxOffsetNumber(), palloc(), pfree(), and TM_IndexDeleteOp::status.

Referenced by _bt_delete_or_dedup_one_page().

◆ _bt_check_natts()

bool _bt_check_natts ( Relation  rel,
bool  heapkeyspace,
Page  page,
OffsetNumber  offnum 
)

Definition at line 2472 of file nbtutils.c.

2473 {
2476  BTPageOpaque opaque = BTPageGetOpaque(page);
2477  IndexTuple itup;
2478  int tupnatts;
2479 
2480  /*
2481  * We cannot reliably test a deleted or half-dead page, since they have
2482  * dummy high keys
2483  */
2484  if (P_IGNORE(opaque))
2485  return true;
2486 
2487  Assert(offnum >= FirstOffsetNumber &&
2488  offnum <= PageGetMaxOffsetNumber(page));
2489 
2490  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
2491  tupnatts = BTreeTupleGetNAtts(itup, rel);
2492 
2493  /* !heapkeyspace indexes do not support deduplication */
2494  if (!heapkeyspace && BTreeTupleIsPosting(itup))
2495  return false;
2496 
2497  /* Posting list tuples should never have "pivot heap TID" bit set */
2498  if (BTreeTupleIsPosting(itup) &&
2500  BT_PIVOT_HEAP_TID_ATTR) != 0)
2501  return false;
2502 
2503  /* INCLUDE indexes do not support deduplication */
2504  if (natts != nkeyatts && BTreeTupleIsPosting(itup))
2505  return false;
2506 
2507  if (P_ISLEAF(opaque))
2508  {
2509  if (offnum >= P_FIRSTDATAKEY(opaque))
2510  {
2511  /*
2512  * Non-pivot tuple should never be explicitly marked as a pivot
2513  * tuple
2514  */
2515  if (BTreeTupleIsPivot(itup))
2516  return false;
2517 
2518  /*
2519  * Leaf tuples that are not the page high key (non-pivot tuples)
2520  * should never be truncated. (Note that tupnatts must have been
2521  * inferred, even with a posting list tuple, because only pivot
2522  * tuples store tupnatts directly.)
2523  */
2524  return tupnatts == natts;
2525  }
2526  else
2527  {
2528  /*
2529  * Rightmost page doesn't contain a page high key, so tuple was
2530  * checked above as ordinary leaf tuple
2531  */
2532  Assert(!P_RIGHTMOST(opaque));
2533 
2534  /*
2535  * !heapkeyspace high key tuple contains only key attributes. Note
2536  * that tupnatts will only have been explicitly represented in
2537  * !heapkeyspace indexes that happen to have non-key attributes.
2538  */
2539  if (!heapkeyspace)
2540  return tupnatts == nkeyatts;
2541 
2542  /* Use generic heapkeyspace pivot tuple handling */
2543  }
2544  }
2545  else /* !P_ISLEAF(opaque) */
2546  {
2547  if (offnum == P_FIRSTDATAKEY(opaque))
2548  {
2549  /*
2550  * The first tuple on any internal page (possibly the first after
2551  * its high key) is its negative infinity tuple. Negative
2552  * infinity tuples are always truncated to zero attributes. They
2553  * are a particular kind of pivot tuple.
2554  */
2555  if (heapkeyspace)
2556  return tupnatts == 0;
2557 
2558  /*
2559  * The number of attributes won't be explicitly represented if the
2560  * negative infinity tuple was generated during a page split that
2561  * occurred with a version of Postgres before v11. There must be
2562  * a problem when there is an explicit representation that is
2563  * non-zero, or when there is no explicit representation and the
2564  * tuple is evidently not a pre-pg_upgrade tuple.
2565  *
2566  * Prior to v11, downlinks always had P_HIKEY as their offset.
2567  * Accept that as an alternative indication of a valid
2568  * !heapkeyspace negative infinity tuple.
2569  */
2570  return tupnatts == 0 ||
2572  }
2573  else
2574  {
2575  /*
2576  * !heapkeyspace downlink tuple with separator key contains only
2577  * key attributes. Note that tupnatts will only have been
2578  * explicitly represented in !heapkeyspace indexes that happen to
2579  * have non-key attributes.
2580  */
2581  if (!heapkeyspace)
2582  return tupnatts == nkeyatts;
2583 
2584  /* Use generic heapkeyspace pivot tuple handling */
2585  }
2586  }
2587 
2588  /* Handle heapkeyspace pivot tuples (excluding minus infinity items) */
2589  Assert(heapkeyspace);
2590 
2591  /*
2592  * Explicit representation of the number of attributes is mandatory with
2593  * heapkeyspace index pivot tuples, regardless of whether or not there are
2594  * non-key attributes.
2595  */
2596  if (!BTreeTupleIsPivot(itup))
2597  return false;
2598 
2599  /* Pivot tuple should not use posting list representation (redundant) */
2600  if (BTreeTupleIsPosting(itup))
2601  return false;
2602 
2603  /*
2604  * Heap TID is a tiebreaker key attribute, so it cannot be untruncated
2605  * when any other key attribute is truncated
2606  */
2607  if (BTreeTupleGetHeapTID(itup) != NULL && tupnatts != nkeyatts)
2608  return false;
2609 
2610  /*
2611  * Pivot tuple must have at least one untruncated key attribute (minus
2612  * infinity pivot tuples are the only exception). Pivot tuples can never
2613  * represent that there is a value present for a key attribute that
2614  * exceeds pg_index.indnkeyatts for the index.
2615  */
2616  return tupnatts > 0 && tupnatts <= nkeyatts;
2617 }
signed short int16
Definition: c.h:477
#define BT_PIVOT_HEAP_TID_ATTR
Definition: nbtree.h:462
#define P_HIKEY
Definition: nbtree.h:364
#define P_RIGHTMOST(opaque)
Definition: nbtree.h:219
#define P_IGNORE(opaque)
Definition: nbtree.h:225
static bool BTreeTupleIsPosting(IndexTuple itup)
Definition: nbtree.h:489
static ItemPointer BTreeTupleGetHeapTID(IndexTuple itup)
Definition: nbtree.h:635
#define BTreeTupleGetNAtts(itup, rel)
Definition: nbtree.h:574
#define FirstOffsetNumber
Definition: off.h:27
ItemPointerData t_tid
Definition: itup.h:37

References Assert(), BT_PIVOT_HEAP_TID_ATTR, BTPageGetOpaque, BTreeTupleGetHeapTID(), BTreeTupleGetNAtts, BTreeTupleIsPivot(), BTreeTupleIsPosting(), FirstOffsetNumber, IndexRelationGetNumberOfAttributes, IndexRelationGetNumberOfKeyAttributes, ItemPointerGetOffsetNumber(), ItemPointerGetOffsetNumberNoCheck(), P_FIRSTDATAKEY, P_HIKEY, P_IGNORE, P_ISLEAF, P_RIGHTMOST, PageGetItem(), PageGetItemId(), PageGetMaxOffsetNumber(), and IndexTupleData::t_tid.

Referenced by _bt_compare(), and bt_target_page_check().

◆ _bt_check_third_page()

void _bt_check_third_page ( Relation  rel,
Relation  heap,
bool  needheaptidspace,
Page  page,
IndexTuple  newtup 
)

Definition at line 2632 of file nbtutils.c.

2634 {
2635  Size itemsz;
2636  BTPageOpaque opaque;
2637 
2638  itemsz = MAXALIGN(IndexTupleSize(newtup));
2639 
2640  /* Double check item size against limit */
2641  if (itemsz <= BTMaxItemSize(page))
2642  return;
2643 
2644  /*
2645  * Tuple is probably too large to fit on page, but it's possible that the
2646  * index uses version 2 or version 3, or that page is an internal page, in
2647  * which case a slightly higher limit applies.
2648  */
2649  if (!needheaptidspace && itemsz <= BTMaxItemSizeNoHeapTid(page))
2650  return;
2651 
2652  /*
2653  * Internal page insertions cannot fail here, because that would mean that
2654  * an earlier leaf level insertion that should have failed didn't
2655  */
2656  opaque = BTPageGetOpaque(page);
2657  if (!P_ISLEAF(opaque))
2658  elog(ERROR, "cannot insert oversized tuple of size %zu on internal page of index \"%s\"",
2659  itemsz, RelationGetRelationName(rel));
2660 
2661  ereport(ERROR,
2662  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
2663  errmsg("index row size %zu exceeds btree version %u maximum %zu for index \"%s\"",
2664  itemsz,
2665  needheaptidspace ? BTREE_VERSION : BTREE_NOVAC_VERSION,
2666  needheaptidspace ? BTMaxItemSize(page) :
2667  BTMaxItemSizeNoHeapTid(page),
2669  errdetail("Index row references tuple (%u,%u) in relation \"%s\".",
2672  RelationGetRelationName(heap)),
2673  errhint("Values larger than 1/3 of a buffer page cannot be indexed.\n"
2674  "Consider a function index of an MD5 hash of the value, "
2675  "or use full text indexing."),
2677 }
size_t Size
Definition: c.h:589
int errdetail(const char *fmt,...)
Definition: elog.c:1202
int errhint(const char *fmt,...)
Definition: elog.c:1316
int errmsg(const char *fmt,...)
Definition: elog.c:1069
#define IndexTupleSize(itup)
Definition: itup.h:70
#define BTMaxItemSizeNoHeapTid(page)
Definition: nbtree.h:169
#define BTREE_VERSION
Definition: nbtree.h:150
#define BTMaxItemSize(page)
Definition: nbtree.h:164
#define BTREE_NOVAC_VERSION
Definition: nbtree.h:152
int errtableconstraint(Relation rel, const char *conname)
Definition: relcache.c:5960

References BTMaxItemSize, BTMaxItemSizeNoHeapTid, BTPageGetOpaque, BTREE_NOVAC_VERSION, BTREE_VERSION, BTreeTupleGetHeapTID(), elog(), ereport, errcode(), errdetail(), errhint(), errmsg(), ERROR, errtableconstraint(), IndexTupleSize, ItemPointerGetBlockNumber(), ItemPointerGetOffsetNumber(), MAXALIGN, P_ISLEAF, and RelationGetRelationName.

Referenced by _bt_buildadd(), and _bt_findinsertloc().

◆ _bt_checkkeys()

bool _bt_checkkeys ( IndexScanDesc  scan,
IndexTuple  tuple,
int  tupnatts,
ScanDirection  dir,
bool continuescan 
)

Definition at line 1362 of file nbtutils.c.

1364 {
1365  TupleDesc tupdesc;
1366  BTScanOpaque so;
1367  int keysz;
1368  int ikey;
1369  ScanKey key;
1370 
1371  Assert(BTreeTupleGetNAtts(tuple, scan->indexRelation) == tupnatts);
1372 
1373  *continuescan = true; /* default assumption */
1374 
1375  tupdesc = RelationGetDescr(scan->indexRelation);
1376  so = (BTScanOpaque) scan->opaque;
1377  keysz = so->numberOfKeys;
1378 
1379  for (key = so->keyData, ikey = 0; ikey < keysz; key++, ikey++)
1380  {
1381  Datum datum;
1382  bool isNull;
1383  Datum test;
1384 
1385  if (key->sk_attno > tupnatts)
1386  {
1387  /*
1388  * This attribute is truncated (must be high key). The value for
1389  * this attribute in the first non-pivot tuple on the page to the
1390  * right could be any possible value. Assume that truncated
1391  * attribute passes the qual.
1392  */
1394  Assert(BTreeTupleIsPivot(tuple));
1395  continue;
1396  }
1397 
1398  /* row-comparison keys need special processing */
1399  if (key->sk_flags & SK_ROW_HEADER)
1400  {
1401  if (_bt_check_rowcompare(key, tuple, tupnatts, tupdesc, dir,
1402  continuescan))
1403  continue;
1404  return false;
1405  }
1406 
1407  datum = index_getattr(tuple,
1408  key->sk_attno,
1409  tupdesc,
1410  &isNull);
1411 
1412  if (key->sk_flags & SK_ISNULL)
1413  {
1414  /* Handle IS NULL/NOT NULL tests */
1415  if (key->sk_flags & SK_SEARCHNULL)
1416  {
1417  if (isNull)
1418  continue; /* tuple satisfies this qual */
1419  }
1420  else
1421  {
1422  Assert(key->sk_flags & SK_SEARCHNOTNULL);
1423  if (!isNull)
1424  continue; /* tuple satisfies this qual */
1425  }
1426 
1427  /*
1428  * Tuple fails this qual. If it's a required qual for the current
1429  * scan direction, then we can conclude no further tuples will
1430  * pass, either.
1431  */
1432  if ((key->sk_flags & SK_BT_REQFWD) &&
1434  *continuescan = false;
1435  else if ((key->sk_flags & SK_BT_REQBKWD) &&
1437  *continuescan = false;
1438 
1439  /*
1440  * In any case, this indextuple doesn't match the qual.
1441  */
1442  return false;
1443  }
1444 
1445  if (isNull)
1446  {
1447  if (key->sk_flags & SK_BT_NULLS_FIRST)
1448  {
1449  /*
1450  * Since NULLs are sorted before non-NULLs, we know we have
1451  * reached the lower limit of the range of values for this
1452  * index attr. On a backward scan, we can stop if this qual
1453  * is one of the "must match" subset. We can stop regardless
1454  * of whether the qual is > or <, so long as it's required,
1455  * because it's not possible for any future tuples to pass. On
1456  * a forward scan, however, we must keep going, because we may
1457  * have initially positioned to the start of the index.
1458  */
1459  if ((key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
1461  *continuescan = false;
1462  }
1463  else
1464  {
1465  /*
1466  * Since NULLs are sorted after non-NULLs, we know we have
1467  * reached the upper limit of the range of values for this
1468  * index attr. On a forward scan, we can stop if this qual is
1469  * one of the "must match" subset. We can stop regardless of
1470  * whether the qual is > or <, so long as it's required,
1471  * because it's not possible for any future tuples to pass. On
1472  * a backward scan, however, we must keep going, because we
1473  * may have initially positioned to the end of the index.
1474  */
1475  if ((key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
1477  *continuescan = false;
1478  }
1479 
1480  /*
1481  * In any case, this indextuple doesn't match the qual.
1482  */
1483  return false;
1484  }
1485 
1486  test = FunctionCall2Coll(&key->sk_func, key->sk_collation,
1487  datum, key->sk_argument);
1488 
1489  if (!DatumGetBool(test))
1490  {
1491  /*
1492  * Tuple fails this qual. If it's a required qual for the current
1493  * scan direction, then we can conclude no further tuples will
1494  * pass, either.
1495  *
1496  * Note: because we stop the scan as soon as any required equality
1497  * qual fails, it is critical that equality quals be used for the
1498  * initial positioning in _bt_first() when they are available. See
1499  * comments in _bt_first().
1500  */
1501  if ((key->sk_flags & SK_BT_REQFWD) &&
1503  *continuescan = false;
1504  else if ((key->sk_flags & SK_BT_REQBKWD) &&
1506  *continuescan = false;
1507 
1508  /*
1509  * In any case, this indextuple doesn't match the qual.
1510  */
1511  return false;
1512  }
1513  }
1514 
1515  /* If we get here, the tuple passes all index quals. */
1516  return true;
1517 }
Datum FunctionCall2Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2)
Definition: fmgr.c:1120
static Datum index_getattr(IndexTuple tup, int attnum, TupleDesc tupleDesc, bool *isnull)
Definition: itup.h:117
#define SK_BT_REQBKWD
Definition: nbtree.h:1084
#define SK_BT_NULLS_FIRST
Definition: nbtree.h:1087
#define SK_BT_REQFWD
Definition: nbtree.h:1083
static bool _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts, TupleDesc tupdesc, ScanDirection dir, bool *continuescan)
Definition: nbtutils.c:1529
uintptr_t Datum
Definition: postgres.h:64
static void test(void)
#define RelationGetDescr(relation)
Definition: rel.h:530
#define ScanDirectionIsForward(direction)
Definition: sdir.h:64
#define SK_ROW_HEADER
Definition: skey.h:117
#define SK_SEARCHNOTNULL
Definition: skey.h:122
#define SK_SEARCHNULL
Definition: skey.h:121
#define SK_ISNULL
Definition: skey.h:115
ScanKey keyData
Definition: nbtree.h:1039
Relation indexRelation
Definition: relscan.h:118

References _bt_check_rowcompare(), Assert(), BTreeTupleGetNAtts, BTreeTupleIsPivot(), DatumGetBool(), FunctionCall2Coll(), index_getattr(), IndexScanDescData::indexRelation, sort-test::key, BTScanOpaqueData::keyData, BTScanOpaqueData::numberOfKeys, IndexScanDescData::opaque, RelationGetDescr, ScanDirectionIsBackward, ScanDirectionIsForward, SK_BT_NULLS_FIRST, SK_BT_REQBKWD, SK_BT_REQFWD, SK_ISNULL, SK_ROW_HEADER, SK_SEARCHNOTNULL, SK_SEARCHNULL, and test().

Referenced by _bt_readpage().

◆ _bt_checkpage()

void _bt_checkpage ( Relation  rel,
Buffer  buf 
)

Definition at line 793 of file nbtpage.c.

794 {
795  Page page = BufferGetPage(buf);
796 
797  /*
798  * ReadBuffer verifies that every newly-read page passes
799  * PageHeaderIsValid, which means it either contains a reasonably sane
800  * page header or is all-zero. We have to defend against the all-zero
801  * case, however.
802  */
803  if (PageIsNew(page))
804  ereport(ERROR,
805  (errcode(ERRCODE_INDEX_CORRUPTED),
806  errmsg("index \"%s\" contains unexpected zero page at block %u",
809  errhint("Please REINDEX it.")));
810 
811  /*
812  * Additionally check that the special area looks sane.
813  */
814  if (PageGetSpecialSize(page) != MAXALIGN(sizeof(BTPageOpaqueData)))
815  ereport(ERROR,
816  (errcode(ERRCODE_INDEX_CORRUPTED),
817  errmsg("index \"%s\" contains corrupted page at block %u",
820  errhint("Please REINDEX it.")));
821 }
static bool PageIsNew(Page page)
Definition: bufpage.h:230
static uint16 PageGetSpecialSize(Page page)
Definition: bufpage.h:313

References buf, BufferGetBlockNumber(), BufferGetPage(), ereport, errcode(), errhint(), errmsg(), ERROR, MAXALIGN, PageGetSpecialSize(), PageIsNew(), and RelationGetRelationName.

Referenced by _bt_getbuf(), _bt_relandgetbuf(), _bt_search_insert(), bt_recheck_sibling_links(), btvacuumpage(), and palloc_btree_page().

◆ _bt_compare()

int32 _bt_compare ( Relation  rel,
BTScanInsert  key,
Page  page,
OffsetNumber  offnum 
)

Definition at line 658 of file nbtsearch.c.

662 {
663  TupleDesc itupdesc = RelationGetDescr(rel);
664  BTPageOpaque opaque = BTPageGetOpaque(page);
665  IndexTuple itup;
666  ItemPointer heapTid;
667  ScanKey scankey;
668  int ncmpkey;
669  int ntupatts;
670  int32 result;
671 
672  Assert(_bt_check_natts(rel, key->heapkeyspace, page, offnum));
674  Assert(key->heapkeyspace || key->scantid == NULL);
675 
676  /*
677  * Force result ">" if target item is first data item on an internal page
678  * --- see NOTE above.
679  */
680  if (!P_ISLEAF(opaque) && offnum == P_FIRSTDATAKEY(opaque))
681  return 1;
682 
683  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
684  ntupatts = BTreeTupleGetNAtts(itup, rel);
685 
686  /*
687  * The scan key is set up with the attribute number associated with each
688  * term in the key. It is important that, if the index is multi-key, the
689  * scan contain the first k key attributes, and that they be in order. If
690  * you think about how multi-key ordering works, you'll understand why
691  * this is.
692  *
693  * We don't test for violation of this condition here, however. The
694  * initial setup for the index scan had better have gotten it right (see
695  * _bt_first).
696  */
697 
698  ncmpkey = Min(ntupatts, key->keysz);
699  Assert(key->heapkeyspace || ncmpkey == key->keysz);
700  Assert(!BTreeTupleIsPosting(itup) || key->allequalimage);
701  scankey = key->scankeys;
702  for (int i = 1; i <= ncmpkey; i++)
703  {
704  Datum datum;
705  bool isNull;
706 
707  datum = index_getattr(itup, scankey->sk_attno, itupdesc, &isNull);
708 
709  if (scankey->sk_flags & SK_ISNULL) /* key is NULL */
710  {
711  if (isNull)
712  result = 0; /* NULL "=" NULL */
713  else if (scankey->sk_flags & SK_BT_NULLS_FIRST)
714  result = -1; /* NULL "<" NOT_NULL */
715  else
716  result = 1; /* NULL ">" NOT_NULL */
717  }
718  else if (isNull) /* key is NOT_NULL and item is NULL */
719  {
720  if (scankey->sk_flags & SK_BT_NULLS_FIRST)
721  result = 1; /* NOT_NULL ">" NULL */
722  else
723  result = -1; /* NOT_NULL "<" NULL */
724  }
725  else
726  {
727  /*
728  * The sk_func needs to be passed the index value as left arg and
729  * the sk_argument as right arg (they might be of different
730  * types). Since it is convenient for callers to think of
731  * _bt_compare as comparing the scankey to the index item, we have
732  * to flip the sign of the comparison result. (Unless it's a DESC
733  * column, in which case we *don't* flip the sign.)
734  */
735  result = DatumGetInt32(FunctionCall2Coll(&scankey->sk_func,
736  scankey->sk_collation,
737  datum,
738  scankey->sk_argument));
739 
740  if (!(scankey->sk_flags & SK_BT_DESC))
741  INVERT_COMPARE_RESULT(result);
742  }
743 
744  /* if the keys are unequal, return the difference */
745  if (result != 0)
746  return result;
747 
748  scankey++;
749  }
750 
751  /*
752  * All non-truncated attributes (other than heap TID) were found to be
753  * equal. Treat truncated attributes as minus infinity when scankey has a
754  * key attribute value that would otherwise be compared directly.
755  *
756  * Note: it doesn't matter if ntupatts includes non-key attributes;
757  * scankey won't, so explicitly excluding non-key attributes isn't
758  * necessary.
759  */
760  if (key->keysz > ntupatts)
761  return 1;
762 
763  /*
764  * Use the heap TID attribute and scantid to try to break the tie. The
765  * rules are the same as any other key attribute -- only the
766  * representation differs.
767  */
768  heapTid = BTreeTupleGetHeapTID(itup);
769  if (key->scantid == NULL)
770  {
771  /*
772  * Most searches have a scankey that is considered greater than a
773  * truncated pivot tuple if and when the scankey has equal values for
774  * attributes up to and including the least significant untruncated
775  * attribute in tuple.
776  *
777  * For example, if an index has the minimum two attributes (single
778  * user key attribute, plus heap TID attribute), and a page's high key
779  * is ('foo', -inf), and scankey is ('foo', <omitted>), the search
780  * will not descend to the page to the left. The search will descend
781  * right instead. The truncated attribute in pivot tuple means that
782  * all non-pivot tuples on the page to the left are strictly < 'foo',
783  * so it isn't necessary to descend left. In other words, search
784  * doesn't have to descend left because it isn't interested in a match
785  * that has a heap TID value of -inf.
786  *
787  * However, some searches (pivotsearch searches) actually require that
788  * we descend left when this happens. -inf is treated as a possible
789  * match for omitted scankey attribute(s). This is needed by page
790  * deletion, which must re-find leaf pages that are targets for
791  * deletion using their high keys.
792  *
793  * Note: the heap TID part of the test ensures that scankey is being
794  * compared to a pivot tuple with one or more truncated key
795  * attributes.
796  *
797  * Note: pg_upgrade'd !heapkeyspace indexes must always descend to the
798  * left here, since they have no heap TID attribute (and cannot have
799  * any -inf key values in any case, since truncation can only remove
800  * non-key attributes). !heapkeyspace searches must always be
801  * prepared to deal with matches on both sides of the pivot once the
802  * leaf level is reached.
803  */
804  if (key->heapkeyspace && !key->pivotsearch &&
805  key->keysz == ntupatts && heapTid == NULL)
806  return 1;
807 
808  /* All provided scankey arguments found to be equal */
809  return 0;
810  }
811 
812  /*
813  * Treat truncated heap TID as minus infinity, since scankey has a key
814  * attribute value (scantid) that would otherwise be compared directly
815  */
817  if (heapTid == NULL)
818  return 1;
819 
820  /*
821  * Scankey must be treated as equal to a posting list tuple if its scantid
822  * value falls within the range of the posting list. In all other cases
823  * there can only be a single heap TID value, which is compared directly
824  * with scantid.
825  */
827  result = ItemPointerCompare(key->scantid, heapTid);
828  if (result <= 0 || !BTreeTupleIsPosting(itup))
829  return result;
830  else
831  {
832  result = ItemPointerCompare(key->scantid,
834  if (result > 0)
835  return 1;
836  }
837 
838  return 0;
839 }
#define Min(x, y)
Definition: c.h:988
#define INVERT_COMPARE_RESULT(var)
Definition: c.h:1114
int32 ItemPointerCompare(ItemPointer arg1, ItemPointer arg2)
Definition: itemptr.c:51
#define SK_BT_DESC
Definition: nbtree.h:1086
static ItemPointer BTreeTupleGetMaxHeapTID(IndexTuple itup)
Definition: nbtree.h:661
bool _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
Definition: nbtutils.c:2472
static int32 DatumGetInt32(Datum X)
Definition: postgres.h:202
int sk_flags
Definition: skey.h:66
FmgrInfo sk_func
Definition: skey.h:71
Oid sk_collation
Definition: skey.h:70
AttrNumber sk_attno
Definition: skey.h:67

References _bt_check_natts(), Assert(), BTPageGetOpaque, BTreeTupleGetHeapTID(), BTreeTupleGetMaxHeapTID(), BTreeTupleGetNAtts, BTreeTupleIsPosting(), DatumGetInt32(), FunctionCall2Coll(), i, index_getattr(), IndexRelationGetNumberOfKeyAttributes, INVERT_COMPARE_RESULT, ItemPointerCompare(), sort-test::key, Min, P_FIRSTDATAKEY, P_ISLEAF, PageGetItem(), PageGetItemId(), RelationGetDescr, ScanKeyData::sk_argument, ScanKeyData::sk_attno, SK_BT_DESC, SK_BT_NULLS_FIRST, ScanKeyData::sk_collation, ScanKeyData::sk_flags, ScanKeyData::sk_func, and SK_ISNULL.

Referenced by _bt_binsrch(), _bt_binsrch_insert(), _bt_check_unique(), _bt_findinsertloc(), _bt_moveright(), _bt_search_insert(), bt_rootdescend(), invariant_g_offset(), invariant_l_nontarget_offset(), invariant_l_offset(), and invariant_leq_offset().

◆ _bt_conditionallockbuf()

bool _bt_conditionallockbuf ( Relation  rel,
Buffer  buf 
)

Definition at line 1090 of file nbtpage.c.

1091 {
1092  /* ConditionalLockBuffer() asserts that pin is held by this backend */
1093  if (!ConditionalLockBuffer(buf))
1094  return false;
1095 
1096  if (!RelationUsesLocalBuffers(rel))
1098 
1099  return true;
1100 }
bool ConditionalLockBuffer(Buffer buffer)
Definition: bufmgr.c:4741
#define VALGRIND_MAKE_MEM_DEFINED(addr, size)
Definition: memdebug.h:26
#define RelationUsesLocalBuffers(relation)
Definition: rel.h:638

References buf, BufferGetPage(), ConditionalLockBuffer(), RelationUsesLocalBuffers, and VALGRIND_MAKE_MEM_DEFINED.

Referenced by _bt_getbuf(), and _bt_search_insert().

◆ _bt_dedup_finish_pending()

Size _bt_dedup_finish_pending ( Page  newpage,
BTDedupState  state 
)

Definition at line 555 of file nbtdedup.c.

556 {
557  OffsetNumber tupoff;
558  Size tuplesz;
559  Size spacesaving;
560 
561  Assert(state->nitems > 0);
562  Assert(state->nitems <= state->nhtids);
563  Assert(state->intervals[state->nintervals].baseoff == state->baseoff);
564 
565  tupoff = OffsetNumberNext(PageGetMaxOffsetNumber(newpage));
566  if (state->nitems == 1)
567  {
568  /* Use original, unchanged base tuple */
569  tuplesz = IndexTupleSize(state->base);
570  Assert(tuplesz == MAXALIGN(IndexTupleSize(state->base)));
571  Assert(tuplesz <= BTMaxItemSize(newpage));
572  if (PageAddItem(newpage, (Item) state->base, tuplesz, tupoff,
573  false, false) == InvalidOffsetNumber)
574  elog(ERROR, "deduplication failed to add tuple to page");
575 
576  spacesaving = 0;
577  }
578  else
579  {
580  IndexTuple final;
581 
582  /* Form a tuple with a posting list */
583  final = _bt_form_posting(state->base, state->htids, state->nhtids);
584  tuplesz = IndexTupleSize(final);
585  Assert(tuplesz <= state->maxpostingsize);
586 
587  /* Save final number of items for posting list */
588  state->intervals[state->nintervals].nitems = state->nitems;
589 
590  Assert(tuplesz == MAXALIGN(IndexTupleSize(final)));
591  Assert(tuplesz <= BTMaxItemSize(newpage));
592  if (PageAddItem(newpage, (Item) final, tuplesz, tupoff, false,
593  false) == InvalidOffsetNumber)
594  elog(ERROR, "deduplication failed to add tuple to page");
595 
596  pfree(final);
597  spacesaving = state->phystupsize - (tuplesz + sizeof(ItemIdData));
598  /* Increment nintervals, since we wrote a new posting list tuple */
599  state->nintervals++;
600  Assert(spacesaving > 0 && spacesaving < BLCKSZ);
601  }
602 
603  /* Reset state for next pending posting list */
604  state->nhtids = 0;
605  state->nitems = 0;
606  state->phystupsize = 0;
607 
608  return spacesaving;
609 }
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:468
Pointer Item
Definition: item.h:17
IndexTuple _bt_form_posting(IndexTuple base, ItemPointer htids, int nhtids)
Definition: nbtdedup.c:864

References _bt_form_posting(), Assert(), BTMaxItemSize, elog(), ERROR, IndexTupleSize, InvalidOffsetNumber, MAXALIGN, OffsetNumberNext, PageAddItem, PageGetMaxOffsetNumber(), and pfree().

Referenced by _bt_dedup_pass(), and btree_xlog_dedup().

◆ _bt_dedup_pass()

void _bt_dedup_pass ( Relation  rel,
Buffer  buf,
IndexTuple  newitem,
Size  newitemsz,
bool  bottomupdedup 
)

Definition at line 58 of file nbtdedup.c.

60 {
61  OffsetNumber offnum,
62  minoff,
63  maxoff;
64  Page page = BufferGetPage(buf);
65  BTPageOpaque opaque = BTPageGetOpaque(page);
66  Page newpage;
68  Size pagesaving PG_USED_FOR_ASSERTS_ONLY = 0;
69  bool singlevalstrat = false;
70  int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
71 
72  /* Passed-in newitemsz is MAXALIGNED but does not include line pointer */
73  newitemsz += sizeof(ItemIdData);
74 
75  /*
76  * Initialize deduplication state.
77  *
78  * It would be possible for maxpostingsize (limit on posting list tuple
79  * size) to be set to one third of the page. However, it seems like a
80  * good idea to limit the size of posting lists to one sixth of a page.
81  * That ought to leave us with a good split point when pages full of
82  * duplicates can be split several times.
83  */
85  state->deduplicate = true;
86  state->nmaxitems = 0;
87  state->maxpostingsize = Min(BTMaxItemSize(page) / 2, INDEX_SIZE_MASK);
88  /* Metadata about base tuple of current pending posting list */
89  state->base = NULL;
90  state->baseoff = InvalidOffsetNumber;
91  state->basetupsize = 0;
92  /* Metadata about current pending posting list TIDs */
93  state->htids = palloc(state->maxpostingsize);
94  state->nhtids = 0;
95  state->nitems = 0;
96  /* Size of all physical tuples to be replaced by pending posting list */
97  state->phystupsize = 0;
98  /* nintervals should be initialized to zero */
99  state->nintervals = 0;
100 
101  minoff = P_FIRSTDATAKEY(opaque);
102  maxoff = PageGetMaxOffsetNumber(page);
103 
104  /*
105  * Consider applying "single value" strategy, though only if the page
106  * seems likely to be split in the near future
107  */
108  if (!bottomupdedup)
109  singlevalstrat = _bt_do_singleval(rel, page, state, minoff, newitem);
110 
111  /*
112  * Deduplicate items from page, and write them to newpage.
113  *
114  * Copy the original page's LSN into newpage copy. This will become the
115  * updated version of the page. We need this because XLogInsert will
116  * examine the LSN and possibly dump it in a page image.
117  */
118  newpage = PageGetTempPageCopySpecial(page);
119  PageSetLSN(newpage, PageGetLSN(page));
120 
121  /* Copy high key, if any */
122  if (!P_RIGHTMOST(opaque))
123  {
124  ItemId hitemid = PageGetItemId(page, P_HIKEY);
125  Size hitemsz = ItemIdGetLength(hitemid);
126  IndexTuple hitem = (IndexTuple) PageGetItem(page, hitemid);
127 
128  if (PageAddItem(newpage, (Item) hitem, hitemsz, P_HIKEY,
129  false, false) == InvalidOffsetNumber)
130  elog(ERROR, "deduplication failed to add highkey");
131  }
132 
133  for (offnum = minoff;
134  offnum <= maxoff;
135  offnum = OffsetNumberNext(offnum))
136  {
137  ItemId itemid = PageGetItemId(page, offnum);
138  IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
139 
140  Assert(!ItemIdIsDead(itemid));
141 
142  if (offnum == minoff)
143  {
144  /*
145  * No previous/base tuple for the data item -- use the data item
146  * as base tuple of pending posting list
147  */
148  _bt_dedup_start_pending(state, itup, offnum);
149  }
150  else if (state->deduplicate &&
151  _bt_keep_natts_fast(rel, state->base, itup) > nkeyatts &&
152  _bt_dedup_save_htid(state, itup))
153  {
154  /*
155  * Tuple is equal to base tuple of pending posting list. Heap
156  * TID(s) for itup have been saved in state.
157  */
158  }
159  else
160  {
161  /*
162  * Tuple is not equal to pending posting list tuple, or
163  * _bt_dedup_save_htid() opted to not merge current item into
164  * pending posting list for some other reason (e.g., adding more
165  * TIDs would have caused posting list to exceed current
166  * maxpostingsize).
167  *
168  * If state contains pending posting list with more than one item,
169  * form new posting tuple and add it to our temp page (newpage).
170  * Else add pending interval's base tuple to the temp page as-is.
171  */
172  pagesaving += _bt_dedup_finish_pending(newpage, state);
173 
174  if (singlevalstrat)
175  {
176  /*
177  * Single value strategy's extra steps.
178  *
179  * Lower maxpostingsize for sixth and final large posting list
180  * tuple at the point where 5 maxpostingsize-capped tuples
181  * have either been formed or observed.
182  *
183  * When a sixth maxpostingsize-capped item is formed/observed,
184  * stop merging together tuples altogether. The few tuples
185  * that remain at the end of the page won't be merged together
186  * at all (at least not until after a future page split takes
187  * place, when this page's newly allocated right sibling page
188  * gets its first deduplication pass).
189  */
190  if (state->nmaxitems == 5)
191  _bt_singleval_fillfactor(page, state, newitemsz);
192  else if (state->nmaxitems == 6)
193  {
194  state->deduplicate = false;
195  singlevalstrat = false; /* won't be back here */
196  }
197  }
198 
199  /* itup starts new pending posting list */
200  _bt_dedup_start_pending(state, itup, offnum);
201  }
202  }
203 
204  /* Handle the last item */
205  pagesaving += _bt_dedup_finish_pending(newpage, state);
206 
207  /*
208  * If no items suitable for deduplication were found, newpage must be
209  * exactly the same as the original page, so just return from function.
210  *
211  * We could determine whether or not to proceed on the basis the space
212  * savings being sufficient to avoid an immediate page split instead. We
213  * don't do that because there is some small value in nbtsplitloc.c always
214  * operating against a page that is fully deduplicated (apart from
215  * newitem). Besides, most of the cost has already been paid.
216  */
217  if (state->nintervals == 0)
218  {
219  /* cannot leak memory here */
220  pfree(newpage);
221  pfree(state->htids);
222  pfree(state);
223  return;
224  }
225 
226  /*
227  * By here, it's clear that deduplication will definitely go ahead.
228  *
229  * Clear the BTP_HAS_GARBAGE page flag. The index must be a heapkeyspace
230  * index, and as such we'll never pay attention to BTP_HAS_GARBAGE anyway.
231  * But keep things tidy.
232  */
233  if (P_HAS_GARBAGE(opaque))
234  {
235  BTPageOpaque nopaque = BTPageGetOpaque(newpage);
236 
237  nopaque->btpo_flags &= ~BTP_HAS_GARBAGE;
238  }
239 
241 
242  PageRestoreTempPage(newpage, page);
244 
245  /* XLOG stuff */
246  if (RelationNeedsWAL(rel))
247  {
248  XLogRecPtr recptr;
249  xl_btree_dedup xlrec_dedup;
250 
251  xlrec_dedup.nintervals = state->nintervals;
252 
253  XLogBeginInsert();
255  XLogRegisterData((char *) &xlrec_dedup, SizeOfBtreeDedup);
256 
257  /*
258  * The intervals array is not in the buffer, but pretend that it is.
259  * When XLogInsert stores the whole buffer, the array need not be
260  * stored too.
261  */
262  XLogRegisterBufData(0, (char *) state->intervals,
263  state->nintervals * sizeof(BTDedupInterval));
264 
265  recptr = XLogInsert(RM_BTREE_ID, XLOG_BTREE_DEDUP);
266 
267  PageSetLSN(page, recptr);
268  }
269 
271 
272  /* Local space accounting should agree with page accounting */
273  Assert(pagesaving < newitemsz || PageGetExactFreeSpace(page) >= newitemsz);
274 
275  /* cannot leak memory here */
276  pfree(state->htids);
277  pfree(state);
278 }
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:2111
void PageRestoreTempPage(Page tempPage, Page oldPage)
Definition: bufpage.c:424
Page PageGetTempPageCopySpecial(Page page)
Definition: bufpage.c:402
static void PageSetLSN(Page page, XLogRecPtr lsn)
Definition: bufpage.h:388
static XLogRecPtr PageGetLSN(Page page)
Definition: bufpage.h:383
#define PG_USED_FOR_ASSERTS_ONLY
Definition: c.h:166
#define ItemIdGetLength(itemId)
Definition: itemid.h:59
#define INDEX_SIZE_MASK
Definition: itup.h:65
#define START_CRIT_SECTION()
Definition: miscadmin.h:148
#define END_CRIT_SECTION()
Definition: miscadmin.h:150
static bool _bt_do_singleval(Relation rel, Page page, BTDedupState state, OffsetNumber minoff, IndexTuple newitem)
Definition: nbtdedup.c:782
Size _bt_dedup_finish_pending(Page newpage, BTDedupState state)
Definition: nbtdedup.c:555
static void _bt_singleval_fillfactor(Page page, BTDedupState state, Size newitemsz)
Definition: nbtdedup.c:822
#define P_HAS_GARBAGE(opaque)
Definition: nbtree.h:226
#define BTP_HAS_GARBAGE
Definition: nbtree.h:82
#define XLOG_BTREE_DEDUP
Definition: nbtxlog.h:33
#define SizeOfBtreeDedup
Definition: nbtxlog.h:174
#define RelationNeedsWAL(relation)
Definition: rel.h:629
uint16 btpo_flags
Definition: nbtree.h:67
uint16 nintervals
Definition: nbtxlog.h:169
uint64 XLogRecPtr
Definition: xlogdefs.h:21
void XLogRegisterData(char *data, uint32 len)
Definition: xloginsert.c:351
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:461
void XLogRegisterBufData(uint8 block_id, char *data, uint32 len)
Definition: xloginsert.c:392
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:243
void XLogBeginInsert(void)
Definition: xloginsert.c:150
#define REGBUF_STANDARD
Definition: xloginsert.h:34

References _bt_dedup_finish_pending(), _bt_dedup_save_htid(), _bt_dedup_start_pending(), _bt_do_singleval(), _bt_keep_natts_fast(), _bt_singleval_fillfactor(), Assert(), BTMaxItemSize, BTP_HAS_GARBAGE, BTPageGetOpaque, BTPageOpaqueData::btpo_flags, buf, BufferGetPage(), elog(), END_CRIT_SECTION, ERROR, INDEX_SIZE_MASK, IndexRelationGetNumberOfKeyAttributes, InvalidOffsetNumber, ItemIdGetLength, ItemIdIsDead, MarkBufferDirty(), Min, xl_btree_dedup::nintervals, OffsetNumberNext, P_FIRSTDATAKEY, P_HAS_GARBAGE, P_HIKEY, P_RIGHTMOST, PageAddItem, PageGetExactFreeSpace(), PageGetItem(), PageGetItemId(), PageGetLSN(), PageGetMaxOffsetNumber(), PageGetTempPageCopySpecial(), PageRestoreTempPage(), PageSetLSN(), palloc(), pfree(), PG_USED_FOR_ASSERTS_ONLY, REGBUF_STANDARD, RelationNeedsWAL, SizeOfBtreeDedup, START_CRIT_SECTION, XLOG_BTREE_DEDUP, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _bt_delete_or_dedup_one_page().

◆ _bt_dedup_save_htid()

bool _bt_dedup_save_htid ( BTDedupState  state,
IndexTuple  itup 
)

Definition at line 484 of file nbtdedup.c.

485 {
486  int nhtids;
487  ItemPointer htids;
488  Size mergedtupsz;
489 
490  Assert(!BTreeTupleIsPivot(itup));
491 
492  if (!BTreeTupleIsPosting(itup))
493  {
494  nhtids = 1;
495  htids = &itup->t_tid;
496  }
497  else
498  {
499  nhtids = BTreeTupleGetNPosting(itup);
500  htids = BTreeTupleGetPosting(itup);
501  }
502 
503  /*
504  * Don't append (have caller finish pending posting list as-is) if
505  * appending heap TID(s) from itup would put us over maxpostingsize limit.
506  *
507  * This calculation needs to match the code used within _bt_form_posting()
508  * for new posting list tuples.
509  */
510  mergedtupsz = MAXALIGN(state->basetupsize +
511  (state->nhtids + nhtids) * sizeof(ItemPointerData));
512 
513  if (mergedtupsz > state->maxpostingsize)
514  {
515  /*
516  * Count this as an oversized item for single value strategy, though
517  * only when there are 50 TIDs in the final posting list tuple. This
518  * limit (which is fairly arbitrary) avoids confusion about how many
519  * 1/6 of a page tuples have been encountered/created by the current
520  * deduplication pass.
521  *
522  * Note: We deliberately don't consider which deduplication pass
523  * merged together tuples to create this item (could be a previous
524  * deduplication pass, or current pass). See _bt_do_singleval()
525  * comments.
526  */
527  if (state->nhtids > 50)
528  state->nmaxitems++;
529 
530  return false;
531  }
532 
533  /*
534  * Save heap TIDs to pending posting list tuple -- itup can be merged into
535  * pending posting list
536  */
537  state->nitems++;
538  memcpy(state->htids + state->nhtids, htids,
539  sizeof(ItemPointerData) * nhtids);
540  state->nhtids += nhtids;
541  state->phystupsize += MAXALIGN(IndexTupleSize(itup)) + sizeof(ItemIdData);
542 
543  return true;
544 }
static uint16 BTreeTupleGetNPosting(IndexTuple posting)
Definition: nbtree.h:515
static ItemPointer BTreeTupleGetPosting(IndexTuple posting)
Definition: nbtree.h:534

References Assert(), BTreeTupleGetNPosting(), BTreeTupleGetPosting(), BTreeTupleIsPivot(), BTreeTupleIsPosting(), IndexTupleSize, MAXALIGN, and IndexTupleData::t_tid.

Referenced by _bt_bottomupdel_pass(), _bt_dedup_pass(), _bt_load(), and btree_xlog_dedup().

◆ _bt_dedup_start_pending()

void _bt_dedup_start_pending ( BTDedupState  state,
IndexTuple  base,
OffsetNumber  baseoff 
)

Definition at line 433 of file nbtdedup.c.

435 {
436  Assert(state->nhtids == 0);
437  Assert(state->nitems == 0);
438  Assert(!BTreeTupleIsPivot(base));
439 
440  /*
441  * Copy heap TID(s) from new base tuple for new candidate posting list
442  * into working state's array
443  */
444  if (!BTreeTupleIsPosting(base))
445  {
446  memcpy(state->htids, &base->t_tid, sizeof(ItemPointerData));
447  state->nhtids = 1;
448  state->basetupsize = IndexTupleSize(base);
449  }
450  else
451  {
452  int nposting;
453 
454  nposting = BTreeTupleGetNPosting(base);
455  memcpy(state->htids, BTreeTupleGetPosting(base),
456  sizeof(ItemPointerData) * nposting);
457  state->nhtids = nposting;
458  /* basetupsize should not include existing posting list */
459  state->basetupsize = BTreeTupleGetPostingOffset(base);
460  }
461 
462  /*
463  * Save new base tuple itself -- it'll be needed if we actually create a
464  * new posting list from new pending posting list.
465  *
466  * Must maintain physical size of all existing tuples (including line
467  * pointer overhead) so that we can calculate space savings on page.
468  */
469  state->nitems = 1;
470  state->base = base;
471  state->baseoff = baseoff;
472  state->phystupsize = MAXALIGN(IndexTupleSize(base)) + sizeof(ItemIdData);
473  /* Also save baseoff in pending state for interval */
474  state->intervals[state->nintervals].baseoff = state->baseoff;
475 }
static uint32 BTreeTupleGetPostingOffset(IndexTuple posting)
Definition: nbtree.h:526

References Assert(), BTreeTupleGetNPosting(), BTreeTupleGetPosting(), BTreeTupleGetPostingOffset(), BTreeTupleIsPivot(), BTreeTupleIsPosting(), IndexTupleSize, MAXALIGN, and IndexTupleData::t_tid.

Referenced by _bt_bottomupdel_pass(), _bt_dedup_pass(), _bt_load(), and btree_xlog_dedup().

◆ _bt_delitems_delete_check()

void _bt_delitems_delete_check ( Relation  rel,
Buffer  buf,
Relation  heapRel,
TM_IndexDeleteOp delstate 
)

Definition at line 1515 of file nbtpage.c.

1517 {
1518  Page page = BufferGetPage(buf);
1519  TransactionId snapshotConflictHorizon;
1520  OffsetNumber postingidxoffnum = InvalidOffsetNumber;
1521  int ndeletable = 0,
1522  nupdatable = 0;
1525 
1526  /* Use tableam interface to determine which tuples to delete first */
1527  snapshotConflictHorizon = table_index_delete_tuples(heapRel, delstate);
1528 
1529  /* Should not WAL-log snapshotConflictHorizon unless it's required */
1530  if (!XLogStandbyInfoActive())
1531  snapshotConflictHorizon = InvalidTransactionId;
1532 
1533  /*
1534  * Construct a leaf-page-wise description of what _bt_delitems_delete()
1535  * needs to do to physically delete index tuples from the page.
1536  *
1537  * Must sort deltids array to restore leaf-page-wise order (original order
1538  * before call to tableam). This is the order that the loop expects.
1539  *
1540  * Note that deltids array might be a lot smaller now. It might even have
1541  * no entries at all (with bottom-up deletion caller), in which case there
1542  * is nothing left to do.
1543  */
1544  qsort(delstate->deltids, delstate->ndeltids, sizeof(TM_IndexDelete),
1546  if (delstate->ndeltids == 0)
1547  {
1548  Assert(delstate->bottomup);
1549  return;
1550  }
1551 
1552  /* We definitely have to delete at least one index tuple (or one TID) */
1553  for (int i = 0; i < delstate->ndeltids; i++)
1554  {
1555  TM_IndexStatus *dstatus = delstate->status + delstate->deltids[i].id;
1556  OffsetNumber idxoffnum = dstatus->idxoffnum;
1557  ItemId itemid = PageGetItemId(page, idxoffnum);
1558  IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
1559  int nestedi,
1560  nitem;
1561  BTVacuumPosting vacposting;
1562 
1563  Assert(OffsetNumberIsValid(idxoffnum));
1564 
1565  if (idxoffnum == postingidxoffnum)
1566  {
1567  /*
1568  * This deltid entry is a TID from a posting list tuple that has
1569  * already been completely processed
1570  */
1571  Assert(BTreeTupleIsPosting(itup));
1573  &delstate->deltids[i].tid) < 0);
1575  &delstate->deltids[i].tid) >= 0);
1576  continue;
1577  }
1578 
1579  if (!BTreeTupleIsPosting(itup))
1580  {
1581  /* Plain non-pivot tuple */
1582  Assert(ItemPointerEquals(&itup->t_tid, &delstate->deltids[i].tid));
1583  if (dstatus->knowndeletable)
1584  deletable[ndeletable++] = idxoffnum;
1585  continue;
1586  }
1587 
1588  /*
1589  * itup is a posting list tuple whose lowest deltids entry (which may
1590  * or may not be for the first TID from itup) is considered here now.
1591  * We should process all of the deltids entries for the posting list
1592  * together now, though (not just the lowest). Remember to skip over
1593  * later itup-related entries during later iterations of outermost
1594  * loop.
1595  */
1596  postingidxoffnum = idxoffnum; /* Remember work in outermost loop */
1597  nestedi = i; /* Initialize for first itup deltids entry */
1598  vacposting = NULL; /* Describes final action for itup */
1599  nitem = BTreeTupleGetNPosting(itup);
1600  for (int p = 0; p < nitem; p++)
1601  {
1602  ItemPointer ptid = BTreeTupleGetPostingN(itup, p);
1603  int ptidcmp = -1;
1604 
1605  /*
1606  * This nested loop reuses work across ptid TIDs taken from itup.
1607  * We take advantage of the fact that both itup's TIDs and deltids
1608  * entries (within a single itup/posting list grouping) must both
1609  * be in ascending TID order.
1610  */
1611  for (; nestedi < delstate->ndeltids; nestedi++)
1612  {
1613  TM_IndexDelete *tcdeltid = &delstate->deltids[nestedi];
1614  TM_IndexStatus *tdstatus = (delstate->status + tcdeltid->id);
1615 
1616  /* Stop once we get past all itup related deltids entries */
1617  Assert(tdstatus->idxoffnum >= idxoffnum);
1618  if (tdstatus->idxoffnum != idxoffnum)
1619  break;
1620 
1621  /* Skip past non-deletable itup related entries up front */
1622  if (!tdstatus->knowndeletable)
1623  continue;
1624 
1625  /* Entry is first partial ptid match (or an exact match)? */
1626  ptidcmp = ItemPointerCompare(&tcdeltid->tid, ptid);
1627  if (ptidcmp >= 0)
1628  {
1629  /* Greater than or equal (partial or exact) match... */
1630  break;
1631  }
1632  }
1633 
1634  /* ...exact ptid match to a deletable deltids entry? */
1635  if (ptidcmp != 0)
1636  continue;
1637 
1638  /* Exact match for deletable deltids entry -- ptid gets deleted */
1639  if (vacposting == NULL)
1640  {
1641  vacposting = palloc(offsetof(BTVacuumPostingData, deletetids) +
1642  nitem * sizeof(uint16));
1643  vacposting->itup = itup;
1644  vacposting->updatedoffset = idxoffnum;
1645  vacposting->ndeletedtids = 0;
1646  }
1647  vacposting->deletetids[vacposting->ndeletedtids++] = p;
1648  }
1649 
1650  /* Final decision on itup, a posting list tuple */
1651 
1652  if (vacposting == NULL)
1653  {
1654  /* No TIDs to delete from itup -- do nothing */
1655  }
1656  else if (vacposting->ndeletedtids == nitem)
1657  {
1658  /* Straight delete of itup (to delete all TIDs) */
1659  deletable[ndeletable++] = idxoffnum;
1660  /* Turns out we won't need granular information */
1661  pfree(vacposting);
1662  }
1663  else
1664  {
1665  /* Delete some (but not all) TIDs from itup */
1666  Assert(vacposting->ndeletedtids > 0 &&
1667  vacposting->ndeletedtids < nitem);
1668  updatable[nupdatable++] = vacposting;
1669  }
1670  }
1671 
1672  /* Physically delete tuples (or TIDs) using deletable (or updatable) */
1673  _bt_delitems_delete(rel, heapRel, buf, snapshotConflictHorizon, deletable,
1674  ndeletable, updatable, nupdatable);
1675 
1676  /* be tidy */
1677  for (int i = 0; i < nupdatable; i++)
1678  pfree(updatable[i]);
1679 }
unsigned short uint16
Definition: c.h:489
uint32 TransactionId
Definition: c.h:636
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:35
#define MaxIndexTuplesPerPage
Definition: itup.h:165
static void _bt_delitems_delete(Relation rel, Relation heaprel, Buffer buf, TransactionId snapshotConflictHorizon, OffsetNumber *deletable, int ndeletable, BTVacuumPosting *updatable, int nupdatable)
Definition: nbtpage.c:1281
static int _bt_delitems_cmp(const void *a, const void *b)
Definition: nbtpage.c:1461
static ItemPointer BTreeTupleGetPostingN(IndexTuple posting, int n)
Definition: nbtree.h:541
#define OffsetNumberIsValid(offsetNumber)
Definition: off.h:39
#define qsort(a, b, c, d)
Definition: port.h:445
uint16 deletetids[FLEXIBLE_ARRAY_MEMBER]
Definition: nbtree.h:913
uint16 ndeletedtids
Definition: nbtree.h:912
IndexTuple itup
Definition: nbtree.h:908
OffsetNumber updatedoffset
Definition: nbtree.h:909
ItemPointerData tid
Definition: tableam.h:205
bool knowndeletable
Definition: tableam.h:212
OffsetNumber idxoffnum
Definition: tableam.h:211
static TransactionId table_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
Definition: tableam.h:1351
#define InvalidTransactionId
Definition: transam.h:31
#define XLogStandbyInfoActive()
Definition: xlog.h:118

References _bt_delitems_cmp(), _bt_delitems_delete(), Assert(), TM_IndexDeleteOp::bottomup, BTreeTupleGetHeapTID(), BTreeTupleGetMaxHeapTID(), BTreeTupleGetNPosting(), BTreeTupleGetPostingN(), BTreeTupleIsPosting(), buf, BufferGetPage(), BTVacuumPostingData::deletetids, TM_IndexDeleteOp::deltids, i, TM_IndexDelete::id, TM_IndexStatus::idxoffnum, InvalidOffsetNumber, InvalidTransactionId, ItemPointerCompare(), ItemPointerEquals(), BTVacuumPostingData::itup, TM_IndexStatus::knowndeletable, MaxIndexTuplesPerPage, BTVacuumPostingData::ndeletedtids, TM_IndexDeleteOp::ndeltids, OffsetNumberIsValid, PageGetItem(), PageGetItemId(), palloc(), pfree(), qsort, TM_IndexDeleteOp::status, IndexTupleData::t_tid, table_index_delete_tuples(), TM_IndexDelete::tid, BTVacuumPostingData::updatedoffset, and XLogStandbyInfoActive.

Referenced by _bt_bottomupdel_pass(), and _bt_simpledel_pass().

◆ _bt_delitems_vacuum()

void _bt_delitems_vacuum ( Relation  rel,
Buffer  buf,
OffsetNumber deletable,
int  ndeletable,
BTVacuumPosting updatable,
int  nupdatable 
)

Definition at line 1151 of file nbtpage.c.

1154 {
1155  Page page = BufferGetPage(buf);
1156  BTPageOpaque opaque;
1157  bool needswal = RelationNeedsWAL(rel);
1158  char *updatedbuf = NULL;
1159  Size updatedbuflen = 0;
1160  OffsetNumber updatedoffsets[MaxIndexTuplesPerPage];
1161 
1162  /* Shouldn't be called unless there's something to do */
1163  Assert(ndeletable > 0 || nupdatable > 0);
1164 
1165  /* Generate new version of posting lists without deleted TIDs */
1166  if (nupdatable > 0)
1167  updatedbuf = _bt_delitems_update(updatable, nupdatable,
1168  updatedoffsets, &updatedbuflen,
1169  needswal);
1170 
1171  /* No ereport(ERROR) until changes are logged */
1173 
1174  /*
1175  * Handle posting tuple updates.
1176  *
1177  * Deliberately do this before handling simple deletes. If we did it the
1178  * other way around (i.e. WAL record order -- simple deletes before
1179  * updates) then we'd have to make compensating changes to the 'updatable'
1180  * array of offset numbers.
1181  *
1182  * PageIndexTupleOverwrite() won't unset each item's LP_DEAD bit when it
1183  * happens to already be set. It's important that we not interfere with
1184  * any future simple index tuple deletion operations.
1185  */
1186  for (int i = 0; i < nupdatable; i++)
1187  {
1188  OffsetNumber updatedoffset = updatedoffsets[i];
1189  IndexTuple itup;
1190  Size itemsz;
1191 
1192  itup = updatable[i]->itup;
1193  itemsz = MAXALIGN(IndexTupleSize(itup));
1194  if (!PageIndexTupleOverwrite(page, updatedoffset, (Item) itup,
1195  itemsz))
1196  elog(PANIC, "failed to update partially dead item in block %u of index \"%s\"",
1198  }
1199 
1200  /* Now handle simple deletes of entire tuples */
1201  if (ndeletable > 0)
1202  PageIndexMultiDelete(page, deletable, ndeletable);
1203 
1204  /*
1205  * We can clear the vacuum cycle ID since this page has certainly been
1206  * processed by the current vacuum scan.
1207  */
1208  opaque = BTPageGetOpaque(page);
1209  opaque->btpo_cycleid = 0;
1210 
1211  /*
1212  * Clear the BTP_HAS_GARBAGE page flag.
1213  *
1214  * This flag indicates the presence of LP_DEAD items on the page (though
1215  * not reliably). Note that we only rely on it with pg_upgrade'd
1216  * !heapkeyspace indexes. That's why clearing it here won't usually
1217  * interfere with simple index tuple deletion.
1218  */
1219  opaque->btpo_flags &= ~BTP_HAS_GARBAGE;
1220 
1222 
1223  /* XLOG stuff */
1224  if (needswal)
1225  {
1226  XLogRecPtr recptr;
1227  xl_btree_vacuum xlrec_vacuum;
1228 
1229  xlrec_vacuum.ndeleted = ndeletable;
1230  xlrec_vacuum.nupdated = nupdatable;
1231 
1232  XLogBeginInsert();
1234  XLogRegisterData((char *) &xlrec_vacuum, SizeOfBtreeVacuum);
1235 
1236  if (ndeletable > 0)
1237  XLogRegisterBufData(0, (char *) deletable,
1238  ndeletable * sizeof(OffsetNumber));
1239 
1240  if (nupdatable > 0)
1241  {
1242  XLogRegisterBufData(0, (char *) updatedoffsets,
1243  nupdatable * sizeof(OffsetNumber));
1244  XLogRegisterBufData(0, updatedbuf, updatedbuflen);
1245  }
1246 
1247  recptr = XLogInsert(RM_BTREE_ID, XLOG_BTREE_VACUUM);
1248 
1249  PageSetLSN(page, recptr);
1250  }
1251 
1252  END_CRIT_SECTION();
1253 
1254  /* can't leak memory here */
1255  if (updatedbuf != NULL)
1256  pfree(updatedbuf);
1257  /* free tuples allocated within _bt_delitems_update() */
1258  for (int i = 0; i < nupdatable; i++)
1259  pfree(updatable[i]->itup);
1260 }
void PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems)
Definition: bufpage.c:1161
bool PageIndexTupleOverwrite(Page page, OffsetNumber offnum, Item newtup, Size newsize)
Definition: bufpage.c:1405
#define PANIC
Definition: elog.h:42
static char * _bt_delitems_update(BTVacuumPosting *updatable, int nupdatable, OffsetNumber *updatedoffsets, Size *updatedbuflen, bool needswal)
Definition: nbtpage.c:1402
#define SizeOfBtreeVacuum
Definition: nbtxlog.h:234
#define XLOG_BTREE_VACUUM
Definition: nbtxlog.h:39
BTCycleId btpo_cycleid
Definition: nbtree.h:68
uint16 ndeleted
Definition: nbtxlog.h:222
uint16 nupdated
Definition: nbtxlog.h:223

References _bt_delitems_update(), Assert(), BTP_HAS_GARBAGE, BTPageGetOpaque, BTPageOpaqueData::btpo_cycleid, BTPageOpaqueData::btpo_flags, buf, BufferGetBlockNumber(), BufferGetPage(), elog(), END_CRIT_SECTION, i, IndexTupleSize, BTVacuumPostingData::itup, MarkBufferDirty(), MAXALIGN, MaxIndexTuplesPerPage, xl_btree_vacuum::ndeleted, xl_btree_vacuum::nupdated, PageIndexMultiDelete(), PageIndexTupleOverwrite(), PageSetLSN(), PANIC, pfree(), REGBUF_STANDARD, RelationGetRelationName, RelationNeedsWAL, SizeOfBtreeVacuum, START_CRIT_SECTION, XLOG_BTREE_VACUUM, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by btvacuumpage().

◆ _bt_doinsert()

bool _bt_doinsert ( Relation  rel,
IndexTuple  itup,
IndexUniqueCheck  checkUnique,
bool  indexUnchanged,
Relation  heapRel 
)

Definition at line 102 of file nbtinsert.c.

105 {
106  bool is_unique = false;
107  BTInsertStateData insertstate;
108  BTScanInsert itup_key;
109  BTStack stack;
110  bool checkingunique = (checkUnique != UNIQUE_CHECK_NO);
111 
112  /* we need an insertion scan key to do our search, so build one */
113  itup_key = _bt_mkscankey(rel, heapRel, itup);
114 
115  if (checkingunique)
116  {
117  if (!itup_key->anynullkeys)
118  {
119  /* No (heapkeyspace) scantid until uniqueness established */
120  itup_key->scantid = NULL;
121  }
122  else
123  {
124  /*
125  * Scan key for new tuple contains NULL key values. Bypass
126  * checkingunique steps. They are unnecessary because core code
127  * considers NULL unequal to every value, including NULL.
128  *
129  * This optimization avoids O(N^2) behavior within the
130  * _bt_findinsertloc() heapkeyspace path when a unique index has a
131  * large number of "duplicates" with NULL key values.
132  */
133  checkingunique = false;
134  /* Tuple is unique in the sense that core code cares about */
135  Assert(checkUnique != UNIQUE_CHECK_EXISTING);
136  is_unique = true;
137  }
138  }
139 
140  /*
141  * Fill in the BTInsertState working area, to track the current page and
142  * position within the page to insert on.
143  *
144  * Note that itemsz is passed down to lower level code that deals with
145  * inserting the item. It must be MAXALIGN()'d. This ensures that space
146  * accounting code consistently considers the alignment overhead that we
147  * expect PageAddItem() will add later. (Actually, index_form_tuple() is
148  * already conservative about alignment, but we don't rely on that from
149  * this distance. Besides, preserving the "true" tuple size in index
150  * tuple headers for the benefit of nbtsplitloc.c might happen someday.
151  * Note that heapam does not MAXALIGN() each heap tuple's lp_len field.)
152  */
153  insertstate.itup = itup;
154  insertstate.itemsz = MAXALIGN(IndexTupleSize(itup));
155  insertstate.itup_key = itup_key;
156  insertstate.bounds_valid = false;
157  insertstate.buf = InvalidBuffer;
158  insertstate.postingoff = 0;
159 
160 search:
161 
162  /*
163  * Find and lock the leaf page that the tuple should be added to by
164  * searching from the root page. insertstate.buf will hold a buffer that
165  * is locked in exclusive mode afterwards.
166  */
167  stack = _bt_search_insert(rel, heapRel, &insertstate);
168 
169  /*
170  * checkingunique inserts are not allowed to go ahead when two tuples with
171  * equal key attribute values would be visible to new MVCC snapshots once
172  * the xact commits. Check for conflicts in the locked page/buffer (if
173  * needed) here.
174  *
175  * It might be necessary to check a page to the right in _bt_check_unique,
176  * though that should be very rare. In practice the first page the value
177  * could be on (with scantid omitted) is almost always also the only page
178  * that a matching tuple might be found on. This is due to the behavior
179  * of _bt_findsplitloc with duplicate tuples -- a group of duplicates can
180  * only be allowed to cross a page boundary when there is no candidate
181  * leaf page split point that avoids it. Also, _bt_check_unique can use
182  * the leaf page high key to determine that there will be no duplicates on
183  * the right sibling without actually visiting it (it uses the high key in
184  * cases where the new item happens to belong at the far right of the leaf
185  * page).
186  *
187  * NOTE: obviously, _bt_check_unique can only detect keys that are already
188  * in the index; so it cannot defend against concurrent insertions of the
189  * same key. We protect against that by means of holding a write lock on
190  * the first page the value could be on, with omitted/-inf value for the
191  * implicit heap TID tiebreaker attribute. Any other would-be inserter of
192  * the same key must acquire a write lock on the same page, so only one
193  * would-be inserter can be making the check at one time. Furthermore,
194  * once we are past the check we hold write locks continuously until we
195  * have performed our insertion, so no later inserter can fail to see our
196  * insertion. (This requires some care in _bt_findinsertloc.)
197  *
198  * If we must wait for another xact, we release the lock while waiting,
199  * and then must perform a new search.
200  *
201  * For a partial uniqueness check, we don't wait for the other xact. Just
202  * let the tuple in and return false for possibly non-unique, or true for
203  * definitely unique.
204  */
205  if (checkingunique)
206  {
207  TransactionId xwait;
208  uint32 speculativeToken;
209 
210  xwait = _bt_check_unique(rel, &insertstate, heapRel, checkUnique,
211  &is_unique, &speculativeToken);
212 
213  if (unlikely(TransactionIdIsValid(xwait)))
214  {
215  /* Have to wait for the other guy ... */
216  _bt_relbuf(rel, insertstate.buf);
217  insertstate.buf = InvalidBuffer;
218 
219  /*
220  * If it's a speculative insertion, wait for it to finish (ie. to
221  * go ahead with the insertion, or kill the tuple). Otherwise
222  * wait for the transaction to finish as usual.
223  */
224  if (speculativeToken)
225  SpeculativeInsertionWait(xwait, speculativeToken);
226  else
227  XactLockTableWait(xwait, rel, &itup->t_tid, XLTW_InsertIndex);
228 
229  /* start over... */
230  if (stack)
231  _bt_freestack(stack);
232  goto search;
233  }
234 
235  /* Uniqueness is established -- restore heap tid as scantid */
236  if (itup_key->heapkeyspace)
237  itup_key->scantid = &itup->t_tid;
238  }
239 
240  if (checkUnique != UNIQUE_CHECK_EXISTING)
241  {
242  OffsetNumber newitemoff;
243 
244  /*
245  * The only conflict predicate locking cares about for indexes is when
246  * an index tuple insert conflicts with an existing lock. We don't
247  * know the actual page we're going to insert on for sure just yet in
248  * checkingunique and !heapkeyspace cases, but it's okay to use the
249  * first page the value could be on (with scantid omitted) instead.
250  */
252 
253  /*
254  * Do the insertion. Note that insertstate contains cached binary
255  * search bounds established within _bt_check_unique when insertion is
256  * checkingunique.
257  */
258  newitemoff = _bt_findinsertloc(rel, &insertstate, checkingunique,
259  indexUnchanged, stack, heapRel);
260  _bt_insertonpg(rel, heapRel, itup_key, insertstate.buf, InvalidBuffer,
261  stack, itup, insertstate.itemsz, newitemoff,
262  insertstate.postingoff, false);
263  }
264  else
265  {
266  /* just release the buffer */
267  _bt_relbuf(rel, insertstate.buf);
268  }
269 
270  /* be tidy */
271  if (stack)
272  _bt_freestack(stack);
273  pfree(itup_key);
274 
275  return is_unique;
276 }
unsigned int uint32
Definition: c.h:490
@ UNIQUE_CHECK_NO
Definition: genam.h:117
@ UNIQUE_CHECK_EXISTING
Definition: genam.h:120
void XactLockTableWait(TransactionId xid, Relation rel, ItemPointer ctid, XLTW_Oper oper)
Definition: lmgr.c:668
void SpeculativeInsertionWait(TransactionId xid, uint32 token)
Definition: lmgr.c:825
@ XLTW_InsertIndex
Definition: lmgr.h:31
static BTStack _bt_search_insert(Relation rel, Relation heaprel, BTInsertState insertstate)
Definition: nbtinsert.c:317
static OffsetNumber _bt_findinsertloc(Relation rel, BTInsertState insertstate, bool checkingunique, bool indexUnchanged, BTStack stack, Relation heapRel)
Definition: nbtinsert.c:815
static void _bt_insertonpg(Relation rel, Relation heaprel, BTScanInsert itup_key, Buffer buf, Buffer cbuf, BTStack stack, IndexTuple itup, Size itemsz, OffsetNumber newitemoff, int postingoff, bool split_only_page)
Definition: nbtinsert.c:1103
static TransactionId _bt_check_unique(Relation rel, BTInsertState insertstate, Relation heapRel, IndexUniqueCheck checkUnique, bool *is_unique, uint32 *speculativeToken)
Definition: nbtinsert.c:408
void _bt_relbuf(Relation rel, Buffer buf)
Definition: nbtpage.c:1020
void _bt_freestack(BTStack stack)
Definition: nbtutils.c:182
BTScanInsert _bt_mkscankey(Relation rel, Relation heaprel, IndexTuple itup)
Definition: nbtutils.c:90
void CheckForSerializableConflictIn(Relation relation, ItemPointer tid, BlockNumber blkno)
Definition: predicate.c:4270
IndexTuple itup
Definition: nbtree.h:813
ItemPointer scantid
Definition: nbtree.h:793
bool heapkeyspace
Definition: nbtree.h:788
bool anynullkeys
Definition: nbtree.h:790
#define TransactionIdIsValid(xid)
Definition: transam.h:41

References _bt_check_unique(), _bt_findinsertloc(), _bt_freestack(), _bt_insertonpg(), _bt_mkscankey(), _bt_relbuf(), _bt_search_insert(), BTScanInsertData::anynullkeys, Assert(), BTInsertStateData::bounds_valid, BTInsertStateData::buf, BufferGetBlockNumber(), CheckForSerializableConflictIn(), BTScanInsertData::heapkeyspace, IndexTupleSize, InvalidBuffer, BTInsertStateData::itemsz, BTInsertStateData::itup, BTInsertStateData::itup_key, MAXALIGN, pfree(), BTInsertStateData::postingoff, BTScanInsertData::scantid, SpeculativeInsertionWait(), IndexTupleData::t_tid, TransactionIdIsValid, UNIQUE_CHECK_EXISTING, UNIQUE_CHECK_NO, unlikely, XactLockTableWait(), and XLTW_InsertIndex.

Referenced by btinsert().

◆ _bt_end_vacuum()

void _bt_end_vacuum ( Relation  rel)

Definition at line 2034 of file nbtutils.c.

2035 {
2036  int i;
2037 
2038  LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
2039 
2040  /* Find the array entry */
2041  for (i = 0; i < btvacinfo->num_vacuums; i++)
2042  {
2043  BTOneVacInfo *vac = &btvacinfo->vacuums[i];
2044 
2045  if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
2046  vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
2047  {
2048  /* Remove it by shifting down the last entry */
2049  *vac = btvacinfo->vacuums[btvacinfo->num_vacuums - 1];
2051  break;
2052  }
2053  }
2054 
2055  LWLockRelease(BtreeVacuumLock);
2056 }
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1195
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1803
@ LW_EXCLUSIVE
Definition: lwlock.h:116
static BTVacInfo * btvacinfo
Definition: nbtutils.c:1930
LockRelId relid
Definition: nbtutils.c:1918
int num_vacuums
Definition: nbtutils.c:1925
BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER]
Definition: nbtutils.c:1927
LockRelId lockRelId
Definition: rel.h:46
Oid relId
Definition: rel.h:40
Oid dbId
Definition: rel.h:41
LockInfoData rd_lockInfo
Definition: rel.h:114

References btvacinfo, LockRelId::dbId, i, LockInfoData::lockRelId, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), BTVacInfo::num_vacuums, RelationData::rd_lockInfo, BTOneVacInfo::relid, LockRelId::relId, and BTVacInfo::vacuums.

Referenced by _bt_end_vacuum_callback(), and btbulkdelete().

◆ _bt_end_vacuum_callback()

void _bt_end_vacuum_callback ( int  code,
Datum  arg 
)

Definition at line 2062 of file nbtutils.c.

2063 {
2065 }
void _bt_end_vacuum(Relation rel)
Definition: nbtutils.c:2034
void * arg
static Pointer DatumGetPointer(Datum X)
Definition: postgres.h:312

References _bt_end_vacuum(), arg, and DatumGetPointer().

Referenced by btbulkdelete().

◆ _bt_findsplitloc()

OffsetNumber _bt_findsplitloc ( Relation  rel,
Page  origpage,
OffsetNumber  newitemoff,
Size  newitemsz,
IndexTuple  newitem,
bool newitemonleft 
)

Definition at line 129 of file nbtsplitloc.c.

135 {
136  BTPageOpaque opaque;
137  int leftspace,
138  rightspace,
139  olddataitemstotal,
140  olddataitemstoleft,
141  perfectpenalty,
142  leaffillfactor;
144  FindSplitStrat strategy;
145  ItemId itemid;
146  OffsetNumber offnum,
147  maxoff,
148  firstrightoff;
149  double fillfactormult;
150  bool usemult;
151  SplitPoint leftpage,
152  rightpage;
153 
154  opaque = BTPageGetOpaque(origpage);
155  maxoff = PageGetMaxOffsetNumber(origpage);
156 
157  /* Total free space available on a btree page, after fixed overhead */
158  leftspace = rightspace =
160  MAXALIGN(sizeof(BTPageOpaqueData));
161 
162  /* The right page will have the same high key as the old page */
163  if (!P_RIGHTMOST(opaque))
164  {
165  itemid = PageGetItemId(origpage, P_HIKEY);
166  rightspace -= (int) (MAXALIGN(ItemIdGetLength(itemid)) +
167  sizeof(ItemIdData));
168  }
169 
170  /* Count up total space in data items before actually scanning 'em */
171  olddataitemstotal = rightspace - (int) PageGetExactFreeSpace(origpage);
172  leaffillfactor = BTGetFillFactor(rel);
173 
174  /* Passed-in newitemsz is MAXALIGNED but does not include line pointer */
175  newitemsz += sizeof(ItemIdData);
176  state.rel = rel;
177  state.origpage = origpage;
178  state.newitem = newitem;
179  state.newitemsz = newitemsz;
180  state.is_leaf = P_ISLEAF(opaque);
181  state.is_rightmost = P_RIGHTMOST(opaque);
182  state.leftspace = leftspace;
183  state.rightspace = rightspace;
184  state.olddataitemstotal = olddataitemstotal;
185  state.minfirstrightsz = SIZE_MAX;
186  state.newitemoff = newitemoff;
187 
188  /* newitem cannot be a posting list item */
189  Assert(!BTreeTupleIsPosting(newitem));
190 
191  /*
192  * nsplits should never exceed maxoff because there will be at most as
193  * many candidate split points as there are points _between_ tuples, once
194  * you imagine that the new item is already on the original page (the
195  * final number of splits may be slightly lower because not all points
196  * between tuples will be legal).
197  */
198  state.maxsplits = maxoff;
199  state.splits = palloc(sizeof(SplitPoint) * state.maxsplits);
200  state.nsplits = 0;
201 
202  /*
203  * Scan through the data items and calculate space usage for a split at
204  * each possible position
205  */
206  olddataitemstoleft = 0;
207 
208  for (offnum = P_FIRSTDATAKEY(opaque);
209  offnum <= maxoff;
210  offnum = OffsetNumberNext(offnum))
211  {
212  Size itemsz;
213 
214  itemid = PageGetItemId(origpage, offnum);
215  itemsz = MAXALIGN(ItemIdGetLength(itemid)) + sizeof(ItemIdData);
216 
217  /*
218  * When item offset number is not newitemoff, neither side of the
219  * split can be newitem. Record a split after the previous data item
220  * from original page, but before the current data item from original
221  * page. (_bt_recsplitloc() will reject the split when there are no
222  * previous items, which we rely on.)
223  */
224  if (offnum < newitemoff)
225  _bt_recsplitloc(&state, offnum, false, olddataitemstoleft, itemsz);
226  else if (offnum > newitemoff)
227  _bt_recsplitloc(&state, offnum, true, olddataitemstoleft, itemsz);
228  else
229  {
230  /*
231  * Record a split after all "offnum < newitemoff" original page
232  * data items, but before newitem
233  */
234  _bt_recsplitloc(&state, offnum, false, olddataitemstoleft, itemsz);
235 
236  /*
237  * Record a split after newitem, but before data item from
238  * original page at offset newitemoff/current offset
239  */
240  _bt_recsplitloc(&state, offnum, true, olddataitemstoleft, itemsz);
241  }
242 
243  olddataitemstoleft += itemsz;
244  }
245 
246  /*
247  * Record a split after all original page data items, but before newitem.
248  * (Though only when it's possible that newitem will end up alone on new
249  * right page.)
250  */
251  Assert(olddataitemstoleft == olddataitemstotal);
252  if (newitemoff > maxoff)
253  _bt_recsplitloc(&state, newitemoff, false, olddataitemstotal, 0);
254 
255  /*
256  * I believe it is not possible to fail to find a feasible split, but just
257  * in case ...
258  */
259  if (state.nsplits == 0)
260  elog(ERROR, "could not find a feasible split point for index \"%s\"",
262 
263  /*
264  * Start search for a split point among list of legal split points. Give
265  * primary consideration to equalizing available free space in each half
266  * of the split initially (start with default strategy), while applying
267  * rightmost and split-after-new-item optimizations where appropriate.
268  * Either of the two other fallback strategies may be required for cases
269  * with a large number of duplicates around the original/space-optimal
270  * split point.
271  *
272  * Default strategy gives some weight to suffix truncation in deciding a
273  * split point on leaf pages. It attempts to select a split point where a
274  * distinguishing attribute appears earlier in the new high key for the
275  * left side of the split, in order to maximize the number of trailing
276  * attributes that can be truncated away. Only candidate split points
277  * that imply an acceptable balance of free space on each side are
278  * considered. See _bt_defaultinterval().
279  */
280  if (!state.is_leaf)
281  {
282  /* fillfactormult only used on rightmost page */
283  usemult = state.is_rightmost;
284  fillfactormult = BTREE_NONLEAF_FILLFACTOR / 100.0;
285  }
286  else if (state.is_rightmost)
287  {
288  /* Rightmost leaf page -- fillfactormult always used */
289  usemult = true;
290  fillfactormult = leaffillfactor / 100.0;
291  }
292  else if (_bt_afternewitemoff(&state, maxoff, leaffillfactor, &usemult))
293  {
294  /*
295  * New item inserted at rightmost point among a localized grouping on
296  * a leaf page -- apply "split after new item" optimization, either by
297  * applying leaf fillfactor multiplier, or by choosing the exact split
298  * point that leaves newitem as lastleft. (usemult is set for us.)
299  */
300  if (usemult)
301  {
302  /* fillfactormult should be set based on leaf fillfactor */
303  fillfactormult = leaffillfactor / 100.0;
304  }
305  else
306  {
307  /* find precise split point after newitemoff */
308  for (int i = 0; i < state.nsplits; i++)
309  {
310  SplitPoint *split = state.splits + i;
311 
312  if (split->newitemonleft &&
313  newitemoff == split->firstrightoff)
314  {
315  pfree(state.splits);
316  *newitemonleft = true;
317  return newitemoff;
318  }
319  }
320 
321  /*
322  * Cannot legally split after newitemoff; proceed with split
323  * without using fillfactor multiplier. This is defensive, and
324  * should never be needed in practice.
325  */
326  fillfactormult = 0.50;
327  }
328  }
329  else
330  {
331  /* Other leaf page. 50:50 page split. */
332  usemult = false;
333  /* fillfactormult not used, but be tidy */
334  fillfactormult = 0.50;
335  }
336 
337  /*
338  * Save leftmost and rightmost splits for page before original ordinal
339  * sort order is lost by delta/fillfactormult sort
340  */
341  leftpage = state.splits[0];
342  rightpage = state.splits[state.nsplits - 1];
343 
344  /* Give split points a fillfactormult-wise delta, and sort on deltas */
345  _bt_deltasortsplits(&state, fillfactormult, usemult);
346 
347  /* Determine split interval for default strategy */
348  state.interval = _bt_defaultinterval(&state);
349 
350  /*
351  * Determine if default strategy/split interval will produce a
352  * sufficiently distinguishing split, or if we should change strategies.
353  * Alternative strategies change the range of split points that are
354  * considered acceptable (split interval), and possibly change
355  * fillfactormult, in order to deal with pages with a large number of
356  * duplicates gracefully.
357  *
358  * Pass low and high splits for the entire page (actually, they're for an
359  * imaginary version of the page that includes newitem). These are used
360  * when the initial split interval encloses split points that are full of
361  * duplicates, and we need to consider if it's even possible to avoid
362  * appending a heap TID.
363  */
364  perfectpenalty = _bt_strategy(&state, &leftpage, &rightpage, &strategy);
365 
366  if (strategy == SPLIT_DEFAULT)
367  {
368  /*
369  * Default strategy worked out (always works out with internal page).
370  * Original split interval still stands.
371  */
372  }
373 
374  /*
375  * Many duplicates strategy is used when a heap TID would otherwise be
376  * appended, but the page isn't completely full of logical duplicates.
377  *
378  * The split interval is widened to include all legal candidate split
379  * points. There might be a few as two distinct values in the whole-page
380  * split interval, though it's also possible that most of the values on
381  * the page are unique. The final split point will either be to the
382  * immediate left or to the immediate right of the group of duplicate
383  * tuples that enclose the first/delta-optimal split point (perfect
384  * penalty was set so that the lowest delta split point that avoids
385  * appending a heap TID will be chosen). Maximizing the number of
386  * attributes that can be truncated away is not a goal of the many
387  * duplicates strategy.
388  *
389  * Single value strategy is used when it is impossible to avoid appending
390  * a heap TID. It arranges to leave the left page very full. This
391  * maximizes space utilization in cases where tuples with the same
392  * attribute values span many pages. Newly inserted duplicates will tend
393  * to have higher heap TID values, so we'll end up splitting to the right
394  * consistently. (Single value strategy is harmless though not
395  * particularly useful with !heapkeyspace indexes.)
396  */
397  else if (strategy == SPLIT_MANY_DUPLICATES)
398  {
399  Assert(state.is_leaf);
400  /* Shouldn't try to truncate away extra user attributes */
401  Assert(perfectpenalty ==
403  /* No need to resort splits -- no change in fillfactormult/deltas */
404  state.interval = state.nsplits;
405  }
406  else if (strategy == SPLIT_SINGLE_VALUE)
407  {
408  Assert(state.is_leaf);
409  /* Split near the end of the page */
410  usemult = true;
411  fillfactormult = BTREE_SINGLEVAL_FILLFACTOR / 100.0;
412  /* Resort split points with new delta */
413  _bt_deltasortsplits(&state, fillfactormult, usemult);
414  /* Appending a heap TID is unavoidable, so interval of 1 is fine */
415  state.interval = 1;
416  }
417 
418  /*
419  * Search among acceptable split points (using final split interval) for
420  * the entry that has the lowest penalty, and is therefore expected to
421  * maximize fan-out. Sets *newitemonleft for us.
422  */
423  firstrightoff = _bt_bestsplitloc(&state, perfectpenalty, newitemonleft,
424  strategy);
425  pfree(state.splits);
426 
427  return firstrightoff;
428 }
#define BTREE_SINGLEVAL_FILLFACTOR
Definition: nbtree.h:202
#define BTGetFillFactor(relation)
Definition: nbtree.h:1097
#define BTREE_NONLEAF_FILLFACTOR
Definition: nbtree.h:201
static void _bt_deltasortsplits(FindSplitData *state, double fillfactormult, bool usemult)
Definition: nbtsplitloc.c:566
static int _bt_strategy(FindSplitData *state, SplitPoint *leftpage, SplitPoint *rightpage, FindSplitStrat *strategy)
Definition: nbtsplitloc.c:939
static bool _bt_afternewitemoff(FindSplitData *state, OffsetNumber maxoff, int leaffillfactor, bool *usemult)
Definition: nbtsplitloc.c:635
static void _bt_recsplitloc(FindSplitData *state, OffsetNumber firstrightoff, bool newitemonleft, int olddataitemstoleft, Size firstrightofforigpagetuplesz)
Definition: nbtsplitloc.c:449
FindSplitStrat
Definition: nbtsplitloc.c:21
@ SPLIT_DEFAULT
Definition: nbtsplitloc.c:23
@ SPLIT_MANY_DUPLICATES
Definition: nbtsplitloc.c:24
@ SPLIT_SINGLE_VALUE
Definition: nbtsplitloc.c:25
static OffsetNumber _bt_bestsplitloc(FindSplitData *state, int perfectpenalty, bool *newitemonleft, FindSplitStrat strategy)
Definition: nbtsplitloc.c:793
static int _bt_defaultinterval(FindSplitData *state)
Definition: nbtsplitloc.c:881
bool newitemonleft
Definition: nbtsplitloc.c:37
OffsetNumber firstrightoff
Definition: nbtsplitloc.c:36

References _bt_afternewitemoff(), _bt_bestsplitloc(), _bt_defaultinterval(), _bt_deltasortsplits(), _bt_recsplitloc(), _bt_strategy(), Assert(), BTGetFillFactor, BTPageGetOpaque, BTREE_NONLEAF_FILLFACTOR, BTREE_SINGLEVAL_FILLFACTOR, BTreeTupleIsPosting(), elog(), ERROR, SplitPoint::firstrightoff, i, IndexRelationGetNumberOfKeyAttributes, ItemIdGetLength, MAXALIGN, SplitPoint::newitemonleft, OffsetNumberNext, P_FIRSTDATAKEY, P_HIKEY, P_ISLEAF, P_RIGHTMOST, PageGetExactFreeSpace(), PageGetItemId(), PageGetMaxOffsetNumber(), PageGetPageSize(), palloc(), pfree(), RelationGetRelationName, SizeOfPageHeaderData, SPLIT_DEFAULT, SPLIT_MANY_DUPLICATES, and SPLIT_SINGLE_VALUE.

Referenced by _bt_split().

◆ _bt_finish_split()

void _bt_finish_split ( Relation  rel,
Relation  heaprel,
Buffer  lbuf,
BTStack  stack 
)

Definition at line 2235 of file nbtinsert.c.

2236 {
2237  Page lpage = BufferGetPage(lbuf);
2238  BTPageOpaque lpageop = BTPageGetOpaque(lpage);
2239  Buffer rbuf;
2240  Page rpage;
2241  BTPageOpaque rpageop;
2242  bool wasroot;
2243  bool wasonly;
2244 
2245  Assert(P_INCOMPLETE_SPLIT(lpageop));
2246 
2247  /* Lock right sibling, the one missing the downlink */
2248  rbuf = _bt_getbuf(rel, heaprel, lpageop->btpo_next, BT_WRITE);
2249  rpage = BufferGetPage(rbuf);
2250  rpageop = BTPageGetOpaque(rpage);
2251 
2252  /* Could this be a root split? */
2253  if (!stack)
2254  {
2255  Buffer metabuf;
2256  Page metapg;
2257  BTMetaPageData *metad;
2258 
2259  /* acquire lock on the metapage */
2260  metabuf = _bt_getbuf(rel, heaprel, BTREE_METAPAGE, BT_WRITE);
2261  metapg = BufferGetPage(metabuf);
2262  metad = BTPageGetMeta(metapg);
2263 
2264  wasroot = (metad->btm_root == BufferGetBlockNumber(lbuf));
2265 
2266  _bt_relbuf(rel, metabuf);
2267  }
2268  else
2269  wasroot = false;
2270 
2271  /* Was this the only page on the level before split? */
2272  wasonly = (P_LEFTMOST(lpageop) && P_RIGHTMOST(rpageop));
2273 
2274  elog(DEBUG1, "finishing incomplete split of %u/%u",
2276 
2277  _bt_insert_parent(rel, heaprel, lbuf, rbuf, stack, wasroot, wasonly);
2278 }
int Buffer
Definition: buf.h:23
static void _bt_insert_parent(Relation rel, Relation heaprel, Buffer buf, Buffer rbuf, BTStack stack, bool isroot, bool isonly)
Definition: nbtinsert.c:2097
Buffer _bt_getbuf(Relation rel, Relation heaprel, BlockNumber blkno, int access)
Definition: nbtpage.c:872
#define BTPageGetMeta(p)
Definition: nbtree.h:121
#define P_LEFTMOST(opaque)
Definition: nbtree.h:218
#define P_INCOMPLETE_SPLIT(opaque)
Definition: nbtree.h:227
#define BTREE_METAPAGE
Definition: nbtree.h:148
#define BT_WRITE
Definition: nbtree.h:717
BlockNumber btm_root
Definition: nbtree.h:107
BlockNumber btpo_next
Definition: nbtree.h:65

References _bt_getbuf(), _bt_insert_parent(), _bt_relbuf(), Assert(), BT_WRITE, BTMetaPageData::btm_root, BTPageGetMeta, BTPageGetOpaque, BTPageOpaqueData::btpo_next, BTREE_METAPAGE, BufferGetBlockNumber(), BufferGetPage(), DEBUG1, elog(), P_INCOMPLETE_SPLIT, P_LEFTMOST, and P_RIGHTMOST.

Referenced by _bt_getstackbuf(), _bt_moveright(), and _bt_stepright().

◆ _bt_first()

bool _bt_first ( IndexScanDesc  scan,
ScanDirection  dir 
)

Definition at line 862 of file nbtsearch.c.

863 {
864  Relation rel = scan->indexRelation;
865  Relation heaprel = scan->heapRelation;
866  BTScanOpaque so = (BTScanOpaque) scan->opaque;
867  Buffer buf;
868  BTStack stack;
869  OffsetNumber offnum;
870  StrategyNumber strat;
871  bool nextkey;
872  bool goback;
873  BTScanInsertData inskey;
874  ScanKey startKeys[INDEX_MAX_KEYS];
875  ScanKeyData notnullkeys[INDEX_MAX_KEYS];
876  int keysCount = 0;
877  int i;
878  bool status;
879  StrategyNumber strat_total;
880  BTScanPosItem *currItem;
881  BlockNumber blkno;
882 
884 
886 
887  /*
888  * Examine the scan keys and eliminate any redundant keys; also mark the
889  * keys that must be matched to continue the scan.
890  */
891  _bt_preprocess_keys(scan);
892 
893  /*
894  * Quit now if _bt_preprocess_keys() discovered that the scan keys can
895  * never be satisfied (eg, x == 1 AND x > 2).
896  */
897  if (!so->qual_ok)
898  {
899  /* Notify any other workers that we're done with this scan key. */
900  _bt_parallel_done(scan);
901  return false;
902  }
903 
904  /*
905  * For parallel scans, get the starting page from shared state. If the
906  * scan has not started, proceed to find out first leaf page in the usual
907  * way while keeping other participating processes waiting. If the scan
908  * has already begun, use the page number from the shared structure.
909  */
910  if (scan->parallel_scan != NULL)
911  {
912  status = _bt_parallel_seize(scan, &blkno);
913  if (!status)
914  return false;
915  else if (blkno == P_NONE)
916  {
917  _bt_parallel_done(scan);
918  return false;
919  }
920  else if (blkno != InvalidBlockNumber)
921  {
922  if (!_bt_parallel_readpage(scan, blkno, dir))
923  return false;
924  goto readcomplete;
925  }
926  }
927 
928  /*----------
929  * Examine the scan keys to discover where we need to start the scan.
930  *
931  * We want to identify the keys that can be used as starting boundaries;
932  * these are =, >, or >= keys for a forward scan or =, <, <= keys for
933  * a backwards scan. We can use keys for multiple attributes so long as
934  * the prior attributes had only =, >= (resp. =, <=) keys. Once we accept
935  * a > or < boundary or find an attribute with no boundary (which can be
936  * thought of as the same as "> -infinity"), we can't use keys for any
937  * attributes to its right, because it would break our simplistic notion
938  * of what initial positioning strategy to use.
939  *
940  * When the scan keys include cross-type operators, _bt_preprocess_keys
941  * may not be able to eliminate redundant keys; in such cases we will
942  * arbitrarily pick a usable one for each attribute. This is correct
943  * but possibly not optimal behavior. (For example, with keys like
944  * "x >= 4 AND x >= 5" we would elect to scan starting at x=4 when
945  * x=5 would be more efficient.) Since the situation only arises given
946  * a poorly-worded query plus an incomplete opfamily, live with it.
947  *
948  * When both equality and inequality keys appear for a single attribute
949  * (again, only possible when cross-type operators appear), we *must*
950  * select one of the equality keys for the starting point, because
951  * _bt_checkkeys() will stop the scan as soon as an equality qual fails.
952  * For example, if we have keys like "x >= 4 AND x = 10" and we elect to
953  * start at x=4, we will fail and stop before reaching x=10. If multiple
954  * equality quals survive preprocessing, however, it doesn't matter which
955  * one we use --- by definition, they are either redundant or
956  * contradictory.
957  *
958  * Any regular (not SK_SEARCHNULL) key implies a NOT NULL qualifier.
959  * If the index stores nulls at the end of the index we'll be starting
960  * from, and we have no boundary key for the column (which means the key
961  * we deduced NOT NULL from is an inequality key that constrains the other
962  * end of the index), then we cons up an explicit SK_SEARCHNOTNULL key to
963  * use as a boundary key. If we didn't do this, we might find ourselves
964  * traversing a lot of null entries at the start of the scan.
965  *
966  * In this loop, row-comparison keys are treated the same as keys on their
967  * first (leftmost) columns. We'll add on lower-order columns of the row
968  * comparison below, if possible.
969  *
970  * The selected scan keys (at most one per index column) are remembered by
971  * storing their addresses into the local startKeys[] array.
972  *----------
973  */
974  strat_total = BTEqualStrategyNumber;
975  if (so->numberOfKeys > 0)
976  {
977  AttrNumber curattr;
978  ScanKey chosen;
979  ScanKey impliesNN;
980  ScanKey cur;
981 
982  /*
983  * chosen is the so-far-chosen key for the current attribute, if any.
984  * We don't cast the decision in stone until we reach keys for the
985  * next attribute.
986  */
987  curattr = 1;
988  chosen = NULL;
989  /* Also remember any scankey that implies a NOT NULL constraint */
990  impliesNN = NULL;
991 
992  /*
993  * Loop iterates from 0 to numberOfKeys inclusive; we use the last
994  * pass to handle after-last-key processing. Actual exit from the
995  * loop is at one of the "break" statements below.
996  */
997  for (cur = so->keyData, i = 0;; cur++, i++)
998  {
999  if (i >= so->numberOfKeys || cur->sk_attno != curattr)
1000  {
1001  /*
1002  * Done looking at keys for curattr. If we didn't find a
1003  * usable boundary key, see if we can deduce a NOT NULL key.
1004  */
1005  if (chosen == NULL && impliesNN != NULL &&
1006  ((impliesNN->sk_flags & SK_BT_NULLS_FIRST) ?
1007  ScanDirectionIsForward(dir) :
1009  {
1010  /* Yes, so build the key in notnullkeys[keysCount] */
1011  chosen = &notnullkeys[keysCount];
1012  ScanKeyEntryInitialize(chosen,
1014  (impliesNN->sk_flags &
1016  curattr,
1017  ((impliesNN->sk_flags & SK_BT_NULLS_FIRST) ?
1020  InvalidOid,
1021  InvalidOid,
1022  InvalidOid,
1023  (Datum) 0);
1024  }
1025 
1026  /*
1027  * If we still didn't find a usable boundary key, quit; else
1028  * save the boundary key pointer in startKeys.
1029  */
1030  if (chosen == NULL)
1031  break;
1032  startKeys[keysCount++] = chosen;
1033 
1034  /*
1035  * Adjust strat_total, and quit if we have stored a > or <
1036  * key.
1037  */
1038  strat = chosen->sk_strategy;
1039  if (strat != BTEqualStrategyNumber)
1040  {
1041  strat_total = strat;
1042  if (strat == BTGreaterStrategyNumber ||
1043  strat == BTLessStrategyNumber)
1044  break;
1045  }
1046 
1047  /*
1048  * Done if that was the last attribute, or if next key is not
1049  * in sequence (implying no boundary key is available for the
1050  * next attribute).
1051  */
1052  if (i >= so->numberOfKeys ||
1053  cur->sk_attno != curattr + 1)
1054  break;
1055 
1056  /*
1057  * Reset for next attr.
1058  */
1059  curattr = cur->sk_attno;
1060  chosen = NULL;
1061  impliesNN = NULL;
1062  }
1063 
1064  /*
1065  * Can we use this key as a starting boundary for this attr?
1066  *
1067  * If not, does it imply a NOT NULL constraint? (Because
1068  * SK_SEARCHNULL keys are always assigned BTEqualStrategyNumber,
1069  * *any* inequality key works for that; we need not test.)
1070  */
1071  switch (cur->sk_strategy)
1072  {
1073  case BTLessStrategyNumber:
1075  if (chosen == NULL)
1076  {
1077  if (ScanDirectionIsBackward(dir))
1078  chosen = cur;
1079  else
1080  impliesNN = cur;
1081  }
1082  break;
1083  case BTEqualStrategyNumber:
1084  /* override any non-equality choice */
1085  chosen = cur;
1086  break;
1089  if (chosen == NULL)
1090  {
1091  if (ScanDirectionIsForward(dir))
1092  chosen = cur;
1093  else
1094  impliesNN = cur;
1095  }
1096  break;
1097  }
1098  }
1099  }
1100 
1101  /*
1102  * If we found no usable boundary keys, we have to start from one end of
1103  * the tree. Walk down that edge to the first or last key, and scan from
1104  * there.
1105  */
1106  if (keysCount == 0)
1107  {
1108  bool match;
1109 
1110  match = _bt_endpoint(scan, dir);
1111 
1112  if (!match)
1113  {
1114  /* No match, so mark (parallel) scan finished */
1115  _bt_parallel_done(scan);
1116  }
1117 
1118  return match;
1119  }
1120 
1121  /*
1122  * We want to start the scan somewhere within the index. Set up an
1123  * insertion scankey we can use to search for the boundary point we
1124  * identified above. The insertion scankey is built using the keys
1125  * identified by startKeys[]. (Remaining insertion scankey fields are
1126  * initialized after initial-positioning strategy is finalized.)
1127  */
1128  Assert(keysCount <= INDEX_MAX_KEYS);
1129  for (i = 0; i < keysCount; i++)
1130  {
1131  ScanKey cur = startKeys[i];
1132 
1133  Assert(cur->sk_attno == i + 1);
1134 
1135  if (cur->sk_flags & SK_ROW_HEADER)
1136  {
1137  /*
1138  * Row comparison header: look to the first row member instead.
1139  *
1140  * The member scankeys are already in insertion format (ie, they
1141  * have sk_func = 3-way-comparison function), but we have to watch
1142  * out for nulls, which _bt_preprocess_keys didn't check. A null
1143  * in the first row member makes the condition unmatchable, just
1144  * like qual_ok = false.
1145  */
1146  ScanKey subkey = (ScanKey) DatumGetPointer(cur->sk_argument);
1147 
1148  Assert(subkey->sk_flags & SK_ROW_MEMBER);
1149  if (subkey->sk_flags & SK_ISNULL)
1150  {
1151  _bt_parallel_done(scan);
1152  return false;
1153  }
1154  memcpy(inskey.scankeys + i, subkey, sizeof(ScanKeyData));
1155 
1156  /*
1157  * If the row comparison is the last positioning key we accepted,
1158  * try to add additional keys from the lower-order row members.
1159  * (If we accepted independent conditions on additional index
1160  * columns, we use those instead --- doesn't seem worth trying to
1161  * determine which is more restrictive.) Note that this is OK
1162  * even if the row comparison is of ">" or "<" type, because the
1163  * condition applied to all but the last row member is effectively
1164  * ">=" or "<=", and so the extra keys don't break the positioning
1165  * scheme. But, by the same token, if we aren't able to use all
1166  * the row members, then the part of the row comparison that we
1167  * did use has to be treated as just a ">=" or "<=" condition, and
1168  * so we'd better adjust strat_total accordingly.
1169  */
1170  if (i == keysCount - 1)
1171  {
1172  bool used_all_subkeys = false;
1173 
1174  Assert(!(subkey->sk_flags & SK_ROW_END));
1175  for (;;)
1176  {
1177  subkey++;
1178  Assert(subkey->sk_flags & SK_ROW_MEMBER);
1179  if (subkey->sk_attno != keysCount + 1)
1180  break; /* out-of-sequence, can't use it */
1181  if (subkey->sk_strategy != cur->sk_strategy)
1182  break; /* wrong direction, can't use it */
1183  if (subkey->sk_flags & SK_ISNULL)
1184  break; /* can't use null keys */
1185  Assert(keysCount < INDEX_MAX_KEYS);
1186  memcpy(inskey.scankeys + keysCount, subkey,
1187  sizeof(ScanKeyData));
1188  keysCount++;
1189  if (subkey->sk_flags & SK_ROW_END)
1190  {
1191  used_all_subkeys = true;
1192  break;
1193  }
1194  }
1195  if (!used_all_subkeys)
1196  {
1197  switch (strat_total)
1198  {
1199  case BTLessStrategyNumber:
1200  strat_total = BTLessEqualStrategyNumber;
1201  break;
1203  strat_total = BTGreaterEqualStrategyNumber;
1204  break;
1205  }
1206  }
1207  break; /* done with outer loop */
1208  }
1209  }
1210  else
1211  {
1212  /*
1213  * Ordinary comparison key. Transform the search-style scan key
1214  * to an insertion scan key by replacing the sk_func with the
1215  * appropriate btree comparison function.
1216  *
1217  * If scankey operator is not a cross-type comparison, we can use
1218  * the cached comparison function; otherwise gotta look it up in
1219  * the catalogs. (That can't lead to infinite recursion, since no
1220  * indexscan initiated by syscache lookup will use cross-data-type
1221  * operators.)
1222  *
1223  * We support the convention that sk_subtype == InvalidOid means
1224  * the opclass input type; this is a hack to simplify life for
1225  * ScanKeyInit().
1226  */
1227  if (cur->sk_subtype == rel->rd_opcintype[i] ||
1228  cur->sk_subtype == InvalidOid)
1229  {
1230  FmgrInfo *procinfo;
1231 
1232  procinfo = index_getprocinfo(rel, cur->sk_attno, BTORDER_PROC);
1233  ScanKeyEntryInitializeWithInfo(inskey.scankeys + i,
1234  cur->sk_flags,
1235  cur->sk_attno,
1237  cur->sk_subtype,
1238  cur->sk_collation,
1239  procinfo,
1240  cur->sk_argument);
1241  }
1242  else
1243  {
1244  RegProcedure cmp_proc;
1245 
1246  cmp_proc = get_opfamily_proc(rel->rd_opfamily[i],
1247  rel->rd_opcintype[i],
1248  cur->sk_subtype,
1249  BTORDER_PROC);
1250  if (!RegProcedureIsValid(cmp_proc))
1251  elog(ERROR, "missing support function %d(%u,%u) for attribute %d of index \"%s\"",
1252  BTORDER_PROC, rel->rd_opcintype[i], cur->sk_subtype,
1253  cur->sk_attno, RelationGetRelationName(rel));
1254  ScanKeyEntryInitialize(inskey.scankeys + i,
1255  cur->sk_flags,
1256  cur->sk_attno,
1258  cur->sk_subtype,
1259  cur->sk_collation,
1260  cmp_proc,
1261  cur->sk_argument);
1262  }
1263  }
1264  }
1265 
1266  /*----------
1267  * Examine the selected initial-positioning strategy to determine exactly
1268  * where we need to start the scan, and set flag variables to control the
1269  * code below.
1270  *
1271  * If nextkey = false, _bt_search and _bt_binsrch will locate the first
1272  * item >= scan key. If nextkey = true, they will locate the first
1273  * item > scan key.
1274  *
1275  * If goback = true, we will then step back one item, while if
1276  * goback = false, we will start the scan on the located item.
1277  *----------
1278  */
1279  switch (strat_total)
1280  {
1281  case BTLessStrategyNumber:
1282 
1283  /*
1284  * Find first item >= scankey, then back up one to arrive at last
1285  * item < scankey. (Note: this positioning strategy is only used
1286  * for a backward scan, so that is always the correct starting
1287  * position.)
1288  */
1289  nextkey = false;
1290  goback = true;
1291  break;
1292 
1294 
1295  /*
1296  * Find first item > scankey, then back up one to arrive at last
1297  * item <= scankey. (Note: this positioning strategy is only used
1298  * for a backward scan, so that is always the correct starting
1299  * position.)
1300  */
1301  nextkey = true;
1302  goback = true;
1303  break;
1304 
1305  case BTEqualStrategyNumber:
1306 
1307  /*
1308  * If a backward scan was specified, need to start with last equal
1309  * item not first one.
1310  */
1311  if (ScanDirectionIsBackward(dir))
1312  {
1313  /*
1314  * This is the same as the <= strategy. We will check at the
1315  * end whether the found item is actually =.
1316  */
1317  nextkey = true;
1318  goback = true;
1319  }
1320  else
1321  {
1322  /*
1323  * This is the same as the >= strategy. We will check at the
1324  * end whether the found item is actually =.
1325  */
1326  nextkey = false;
1327  goback = false;
1328  }
1329  break;
1330 
1332 
1333  /*
1334  * Find first item >= scankey. (This is only used for forward
1335  * scans.)
1336  */
1337  nextkey = false;
1338  goback = false;
1339  break;
1340 
1342 
1343  /*
1344  * Find first item > scankey. (This is only used for forward
1345  * scans.)
1346  */
1347  nextkey = true;
1348  goback = false;
1349  break;
1350 
1351  default:
1352  /* can't get here, but keep compiler quiet */
1353  elog(ERROR, "unrecognized strat_total: %d", (int) strat_total);
1354  return false;
1355  }
1356 
1357  /* Initialize remaining insertion scan key fields */
1358  _bt_metaversion(rel, heaprel, &inskey.heapkeyspace, &inskey.allequalimage);
1359  inskey.anynullkeys = false; /* unused */
1360  inskey.nextkey = nextkey;
1361  inskey.pivotsearch = false;
1362  inskey.scantid = NULL;
1363  inskey.keysz = keysCount;
1364 
1365  /*
1366  * Use the manufactured insertion scan key to descend the tree and
1367  * position ourselves on the target leaf page.
1368  */
1369  stack = _bt_search(rel, heaprel, &inskey, &buf, BT_READ, scan->xs_snapshot);
1370 
1371  /* don't need to keep the stack around... */
1372  _bt_freestack(stack);
1373 
1374  if (!BufferIsValid(buf))
1375  {
1376  /*
1377  * We only get here if the index is completely empty. Lock relation
1378  * because nothing finer to lock exists.
1379  */
1380  PredicateLockRelation(rel, scan->xs_snapshot);
1381 
1382  /*
1383  * mark parallel scan as done, so that all the workers can finish
1384  * their scan
1385  */
1386  _bt_parallel_done(scan);
1388 
1389  return false;
1390  }
1391  else
1393  scan->xs_snapshot);
1394 
1395  _bt_initialize_more_data(so, dir);
1396 
1397  /* position to the precise item on the page */
1398  offnum = _bt_binsrch(rel, &inskey, buf);
1399 
1400  /*
1401  * If nextkey = false, we are positioned at the first item >= scan key, or
1402  * possibly at the end of a page on which all the existing items are less
1403  * than the scan key and we know that everything on later pages is greater
1404  * than or equal to scan key.
1405  *
1406  * If nextkey = true, we are positioned at the first item > scan key, or
1407  * possibly at the end of a page on which all the existing items are less
1408  * than or equal to the scan key and we know that everything on later
1409  * pages is greater than scan key.
1410  *
1411  * The actually desired starting point is either this item or the prior
1412  * one, or in the end-of-page case it's the first item on the next page or
1413  * the last item on this page. Adjust the starting offset if needed. (If
1414  * this results in an offset before the first item or after the last one,
1415  * _bt_readpage will report no items found, and then we'll step to the
1416  * next page as needed.)
1417  */
1418  if (goback)
1419  offnum = OffsetNumberPrev(offnum);
1420 
1421  /* remember which buffer we have pinned, if any */
1423  so->currPos.buf = buf;
1424 
1425  /*
1426  * Now load data from the first page of the scan.
1427  */
1428  if (!_bt_readpage(scan, dir, offnum))
1429  {
1430  /*
1431  * There's no actually-matching data on this page. Try to advance to
1432  * the next page. Return false if there's no matching data at all.
1433  */
1434  _bt_unlockbuf(scan->indexRelation, so->currPos.buf);
1435  if (!_bt_steppage(scan, dir))
1436  return false;
1437  }
1438  else
1439  {
1440  /* Drop the lock, and maybe the pin, on the current page */
1442  }
1443 
1444 readcomplete:
1445  /* OK, itemIndex says what to return */
1446  currItem = &so->currPos.items[so->currPos.itemIndex];
1447  scan->xs_heaptid = currItem->heapTid;
1448  if (scan->xs_want_itup)
1449  scan->xs_itup = (IndexTuple) (so->currTuples + currItem->tupleOffset);
1450 
1451  return true;
1452 }
int16 AttrNumber
Definition: attnum.h:21
uint32 BlockNumber
Definition: block.h:31
#define RegProcedureIsValid(p)
Definition: c.h:761
regproc RegProcedure
Definition: c.h:634
struct cursor * cur
Definition: ecpg.c:28
FmgrInfo * index_getprocinfo(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:811
void _bt_metaversion(Relation rel, Relation heaprel, bool *heapkeyspace, bool *allequalimage)
Definition: nbtpage.c:735
void _bt_unlockbuf(Relation rel, Buffer buf)
Definition: nbtpage.c:1067
void _bt_parallel_done(IndexScanDesc scan)
Definition: nbtree.c:717
bool _bt_parallel_seize(IndexScanDesc scan, BlockNumber *pageno)
Definition: nbtree.c:636
#define BTORDER_PROC
Definition: nbtree.h:704
#define BTScanPosIsValid(scanpos)
Definition: nbtree.h:1009
#define P_NONE
Definition: nbtree.h:212
#define BT_READ
Definition: nbtree.h:716
#define BTScanPosInvalidate(scanpos)
Definition: nbtree.h:1015
static bool _bt_readpage(IndexScanDesc scan, ScanDirection dir, OffsetNumber offnum)
Definition: nbtsearch.c:1524
static bool _bt_parallel_readpage(IndexScanDesc scan, BlockNumber blkno, ScanDirection dir)
Definition: nbtsearch.c:2163
static void _bt_drop_lock_and_maybe_pin(IndexScanDesc scan, BTScanPos sp)
Definition: nbtsearch.c:61
static OffsetNumber _bt_binsrch(Relation rel, BTScanInsert key, Buffer buf)
Definition: nbtsearch.c:340
static bool _bt_endpoint(IndexScanDesc scan, ScanDirection dir)
Definition: nbtsearch.c:2396
static bool _bt_steppage(IndexScanDesc scan, ScanDirection dir)
Definition: nbtsearch.c:1876
BTStack _bt_search(Relation rel, Relation heaprel, BTScanInsert key, Buffer *bufP, int access, Snapshot snapshot)
Definition: nbtsearch.c:97
static void _bt_initialize_more_data(BTScanOpaque so, ScanDirection dir)
Definition: nbtsearch.c:2487
void _bt_preprocess_keys(IndexScanDesc scan)
Definition: nbtutils.c:749
#define OffsetNumberPrev(offsetNumber)
Definition: off.h:54
#define INDEX_MAX_KEYS
#define pgstat_count_index_scan(rel)
Definition: pgstat.h:623
#define InvalidOid
Definition: postgres_ext.h:36
void PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
Definition: predicate.c:2533
void PredicateLockRelation(Relation relation, Snapshot snapshot)
Definition: predicate.c:2510
void ScanKeyEntryInitialize(ScanKey entry, int flags, AttrNumber attributeNumber, StrategyNumber strategy, Oid subtype, Oid collation, RegProcedure procedure, Datum argument)
Definition: scankey.c:32
void ScanKeyEntryInitializeWithInfo(ScanKey entry, int flags, AttrNumber attributeNumber, StrategyNumber strategy, Oid subtype, Oid collation, FmgrInfo *finfo, Datum argument)
Definition: scankey.c:101
#define SK_ROW_MEMBER
Definition: skey.h:118
#define SK_ROW_END
Definition: skey.h:119
ScanKeyData * ScanKey
Definition: skey.h:75
uint16 StrategyNumber
Definition: stratnum.h:22
#define BTGreaterStrategyNumber
Definition: stratnum.h:33
#define InvalidStrategy
Definition: stratnum.h:24
#define BTLessStrategyNumber
Definition: stratnum.h:29
#define BTEqualStrategyNumber
Definition: stratnum.h:31
#define BTLessEqualStrategyNumber
Definition: stratnum.h:30
#define BTGreaterEqualStrategyNumber
Definition: stratnum.h:32
BTScanPosData currPos
Definition: nbtree.h:1072
char * currTuples
Definition: nbtree.h:1059
Buffer buf
Definition: nbtree.h:955
BTScanPosItem items[MaxTIDsPerBTreePage]
Definition: nbtree.h:987
int itemIndex
Definition: nbtree.h:985
Definition: fmgr.h:57
IndexTuple xs_itup
Definition: relscan.h:142
ItemPointerData xs_heaptid
Definition: relscan.h:147
struct SnapshotData * xs_snapshot
Definition: relscan.h:119
Relation heapRelation
Definition: relscan.h:117
StrategyNumber sk_strategy
Definition: skey.h:68

References _bt_binsrch(), _bt_drop_lock_and_maybe_pin(), _bt_endpoint(), _bt_freestack(), _bt_initialize_more_data(), _bt_metaversion(), _bt_parallel_done(), _bt_parallel_readpage(), _bt_parallel_seize(), _bt_preprocess_keys(), _bt_readpage(), _bt_search(), _bt_steppage(), _bt_unlockbuf(), Assert(), BT_READ, BTEqualStrategyNumber, BTGreaterEqualStrategyNumber, BTGreaterStrategyNumber, BTLessEqualStrategyNumber, BTLessStrategyNumber, BTORDER_PROC, BTScanPosInvalidate, BTScanPosIsValid, buf, BTScanPosData::buf, BufferGetBlockNumber(), BufferIsValid(), cur, BTScanOpaqueData::currPos, BTScanOpaqueData::currTuples, DatumGetPointer(), elog(), ERROR, get_opfamily_proc(), IndexScanDescData::heapRelation, i, index_getprocinfo(), INDEX_MAX_KEYS, IndexScanDescData::indexRelation, InvalidBlockNumber, InvalidOid, InvalidStrategy, BTScanPosData::itemIndex, BTScanPosData::items, BTScanOpaqueData::keyData, BTScanOpaqueData::numberOfKeys, OffsetNumberPrev, IndexScanDescData::opaque, P_NONE, IndexScanDescData::parallel_scan, pgstat_count_index_scan, PredicateLockPage(), PredicateLockRelation(), BTScanOpaqueData::qual_ok, RelationData::rd_opcintype, RelationData::rd_opfamily, RegProcedureIsValid, RelationGetRelationName, ScanDirectionIsBackward, ScanDirectionIsForward, ScanKeyEntryInitialize(), ScanKeyEntryInitializeWithInfo(), ScanKeyData::sk_attno, SK_BT_DESC, SK_BT_NULLS_FIRST, ScanKeyData::sk_flags, SK_ISNULL, SK_ROW_END, SK_ROW_HEADER, SK_ROW_MEMBER, SK_SEARCHNOTNULL, ScanKeyData::sk_strategy, IndexScanDescData::xs_heaptid, IndexScanDescData::xs_itup, IndexScanDescData::xs_snapshot, and IndexScanDescData::xs_want_itup.

Referenced by btgetbitmap(), and btgettuple().

◆ _bt_form_posting()

IndexTuple _bt_form_posting ( IndexTuple  base,
ItemPointer  htids,
int  nhtids 
)

Definition at line 864 of file nbtdedup.c.

865 {
866  uint32 keysize,
867  newsize;
868  IndexTuple itup;
869 
870  if (BTreeTupleIsPosting(base))
871  keysize = BTreeTupleGetPostingOffset(base);
872  else
873  keysize = IndexTupleSize(base);
874 
875  Assert(!BTreeTupleIsPivot(base));
876  Assert(nhtids > 0 && nhtids <= PG_UINT16_MAX);
877  Assert(keysize == MAXALIGN(keysize));
878 
879  /* Determine final size of new tuple */
880  if (nhtids > 1)
881  newsize = MAXALIGN(keysize +
882  nhtids * sizeof(ItemPointerData));
883  else
884  newsize = keysize;
885 
886  Assert(newsize <= INDEX_SIZE_MASK);
887  Assert(newsize == MAXALIGN(newsize));
888 
889  /* Allocate memory using palloc0() (matches index_form_tuple()) */
890  itup = palloc0(newsize);
891  memcpy(itup, base, keysize);
892  itup->t_info &= ~INDEX_SIZE_MASK;
893  itup->t_info |= newsize;
894  if (nhtids > 1)
895  {
896  /* Form posting list tuple */
897  BTreeTupleSetPosting(itup, nhtids, keysize);
898  memcpy(BTreeTupleGetPosting(itup), htids,
899  sizeof(ItemPointerData) * nhtids);
900  Assert(_bt_posting_valid(itup));
901  }
902  else
903  {
904  /* Form standard non-pivot tuple */
905  itup->t_info &= ~INDEX_ALT_TID_MASK;
906  ItemPointerCopy(htids, &itup->t_tid);
908  }
909 
910  return itup;
911 }
#define PG_UINT16_MAX
Definition: c.h:571
static void ItemPointerCopy(const ItemPointerData *fromPointer, ItemPointerData *toPointer)
Definition: itemptr.h:172
static bool ItemPointerIsValid(const ItemPointerData *pointer)
Definition: itemptr.h:83
void * palloc0(Size size)
Definition: mcxt.c:1257
static void BTreeTupleSetPosting(IndexTuple itup, uint16 nhtids, int postingoffset)
Definition: nbtree.h:501
#define INDEX_ALT_TID_MASK
Definition: nbtree.h:456
unsigned short t_info
Definition: itup.h:49

References Assert(), BTreeTupleGetPosting(), BTreeTupleGetPostingOffset(), BTreeTupleIsPivot(), BTreeTupleIsPosting(), BTreeTupleSetPosting(), INDEX_ALT_TID_MASK, INDEX_SIZE_MASK, IndexTupleSize, ItemPointerCopy(), ItemPointerIsValid(), MAXALIGN, palloc0(), PG_UINT16_MAX, IndexTupleData::t_info, and IndexTupleData::t_tid.

Referenced by _bt_dedup_finish_pending(), _bt_sort_dedup_finish_pending(), and bt_posting_plain_tuple().

◆ _bt_freestack()

void _bt_freestack ( BTStack  stack)

Definition at line 182 of file nbtutils.c.

183 {
184  BTStack ostack;
185 
186  while (stack != NULL)
187  {
188  ostack = stack;
189  stack = stack->bts_parent;
190  pfree(ostack);
191  }
192 }
struct BTStackData * bts_parent
Definition: nbtree.h:733

References BTStackData::bts_parent, and pfree().

Referenced by _bt_doinsert(), _bt_first(), and bt_rootdescend().

◆ _bt_get_endpoint()

Buffer _bt_get_endpoint ( Relation  rel,
Relation  heaprel,
uint32  level,
bool  rightmost,
Snapshot  snapshot 
)

Definition at line 2312 of file nbtsearch.c.

2314 {
2315  Buffer buf;
2316  Page page;
2317  BTPageOpaque opaque;
2318  OffsetNumber offnum;
2319  BlockNumber blkno;
2320  IndexTuple itup;
2321 
2322  /*
2323  * If we are looking for a leaf page, okay to descend from fast root;
2324  * otherwise better descend from true root. (There is no point in being
2325  * smarter about intermediate levels.)
2326  */
2327  if (level == 0)
2328  buf = _bt_getroot(rel, heaprel, BT_READ);
2329  else
2330  buf = _bt_gettrueroot(rel, heaprel);
2331 
2332  if (!BufferIsValid(buf))
2333  return InvalidBuffer;
2334 
2335  page = BufferGetPage(buf);
2336  TestForOldSnapshot(snapshot, rel, page);
2337  opaque = BTPageGetOpaque(page);
2338 
2339  for (;;)
2340  {
2341  /*
2342  * If we landed on a deleted page, step right to find a live page
2343  * (there must be one). Also, if we want the rightmost page, step
2344  * right if needed to get to it (this could happen if the page split
2345  * since we obtained a pointer to it).
2346  */
2347  while (P_IGNORE(opaque) ||
2348  (rightmost && !P_RIGHTMOST(opaque)))
2349  {
2350  blkno = opaque->btpo_next;
2351  if (blkno == P_NONE)
2352  elog(ERROR, "fell off the end of index \"%s\"",
2354  buf = _bt_relandgetbuf(rel, buf, blkno, BT_READ);
2355  page = BufferGetPage(buf);
2356  TestForOldSnapshot(snapshot, rel, page);
2357  opaque = BTPageGetOpaque(page);
2358  }
2359 
2360  /* Done? */
2361  if (opaque->btpo_level == level)
2362  break;
2363  if (opaque->btpo_level < level)
2364  ereport(ERROR,
2365  (errcode(ERRCODE_INDEX_CORRUPTED),
2366  errmsg_internal("btree level %u not found in index \"%s\"",
2367  level, RelationGetRelationName(rel))));
2368 
2369  /* Descend to leftmost or rightmost child page */
2370  if (rightmost)
2371  offnum = PageGetMaxOffsetNumber(page);
2372  else
2373  offnum = P_FIRSTDATAKEY(opaque);
2374 
2375  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
2376  blkno = BTreeTupleGetDownLink(itup);
2377 
2378  buf = _bt_relandgetbuf(rel, buf, blkno, BT_READ);
2379  page = BufferGetPage(buf);
2380  opaque = BTPageGetOpaque(page);
2381  }
2382 
2383  return buf;
2384 }
static void TestForOldSnapshot(Snapshot snapshot, Relation relation, Page page)
Definition: bufmgr.h:378
Buffer _bt_relandgetbuf(Relation rel, Buffer obuf, BlockNumber blkno, int access)
Definition: nbtpage.c:1000
Buffer _bt_getroot(Relation rel, Relation heaprel, int access)
Definition: nbtpage.c:342
Buffer _bt_gettrueroot(Relation rel, Relation heaprel)
Definition: nbtpage.c:576
static BlockNumber BTreeTupleGetDownLink(IndexTuple pivot)
Definition: nbtree.h:553
uint32 btpo_level
Definition: nbtree.h:66

References _bt_getroot(), _bt_gettrueroot(), _bt_relandgetbuf(), BT_READ, BTPageGetOpaque, BTPageOpaqueData::btpo_level, BTPageOpaqueData::btpo_next, BTreeTupleGetDownLink(), buf, BufferGetPage(), BufferIsValid(), elog(), ereport, errcode(), errmsg_internal(), ERROR, InvalidBuffer, P_FIRSTDATAKEY, P_IGNORE, P_NONE, P_RIGHTMOST, PageGetItem(), PageGetItemId(), PageGetMaxOffsetNumber(), RelationGetRelationName, and TestForOldSnapshot().

Referenced by _bt_endpoint(), and _bt_insert_parent().

◆ _bt_getbuf()

Buffer _bt_getbuf ( Relation  rel,
Relation  heaprel,
BlockNumber  blkno,
int  access 
)

Definition at line 872 of file nbtpage.c.

873 {
874  Buffer buf;
875 
876  if (blkno != P_NEW)
877  {
878  /* Read an existing block of the relation */
879  buf = ReadBuffer(rel, blkno);
880  _bt_lockbuf(rel, buf, access);
881  _bt_checkpage(rel, buf);
882  }
883  else
884  {
885  Page page;
886 
887  Assert(access == BT_WRITE);
888 
889  /*
890  * First see if the FSM knows of any free pages.
891  *
892  * We can't trust the FSM's report unreservedly; we have to check that
893  * the page is still free. (For example, an already-free page could
894  * have been re-used between the time the last VACUUM scanned it and
895  * the time the VACUUM made its FSM updates.)
896  *
897  * In fact, it's worse than that: we can't even assume that it's safe
898  * to take a lock on the reported page. If somebody else has a lock
899  * on it, or even worse our own caller does, we could deadlock. (The
900  * own-caller scenario is actually not improbable. Consider an index
901  * on a serial or timestamp column. Nearly all splits will be at the
902  * rightmost page, so it's entirely likely that _bt_split will call us
903  * while holding a lock on the page most recently acquired from FSM. A
904  * VACUUM running concurrently with the previous split could well have
905  * placed that page back in FSM.)
906  *
907  * To get around that, we ask for only a conditional lock on the
908  * reported page. If we fail, then someone else is using the page,
909  * and we may reasonably assume it's not free. (If we happen to be
910  * wrong, the worst consequence is the page will be lost to use till
911  * the next VACUUM, which is no big problem.)
912  */
913  for (;;)
914  {
915  blkno = GetFreeIndexPage(rel);
916  if (blkno == InvalidBlockNumber)
917  break;
918  buf = ReadBuffer(rel, blkno);
919  if (_bt_conditionallockbuf(rel, buf))
920  {
921  page = BufferGetPage(buf);
922 
923  /*
924  * It's possible to find an all-zeroes page in an index. For
925  * example, a backend might successfully extend the relation
926  * one page and then crash before it is able to make a WAL
927  * entry for adding the page. If we find a zeroed page then
928  * reclaim it immediately.
929  */
930  if (PageIsNew(page))
931  {
932  /* Okay to use page. Initialize and return it. */
934  return buf;
935  }
936 
937  if (BTPageIsRecyclable(page, heaprel))
938  {
939  /*
940  * If we are generating WAL for Hot Standby then create a
941  * WAL record that will allow us to conflict with queries
942  * running on standby, in case they have snapshots older
943  * than safexid value
944  */
946  _bt_log_reuse_page(rel, heaprel, blkno,
947  BTPageGetDeleteXid(page));
948 
949  /* Okay to use page. Re-initialize and return it. */
951  return buf;
952  }
953  elog(DEBUG2, "FSM returned nonrecyclable page");
954  _bt_relbuf(rel, buf);
955  }
956  else
957  {
958  elog(DEBUG2, "FSM returned nonlockable page");
959  /* couldn't get lock, so just drop pin */
961  }
962  }
963 
964  /*
965  * Extend the relation by one page. Need to use RBM_ZERO_AND_LOCK or
966  * we risk a race condition against btvacuumscan --- see comments
967  * therein. This forces us to repeat the valgrind request that
968  * _bt_lockbuf() otherwise would make, as we can't use _bt_lockbuf()
969  * without introducing a race.
970  */
972  EB_LOCK_FIRST);
973  if (!RelationUsesLocalBuffers(rel))
975 
976  /* Initialize the new page before returning it */
977  page = BufferGetPage(buf);
978  Assert(PageIsNew(page));
980  }
981 
982  /* ref count and lock type are correct */
983  return buf;
984 }
Buffer ExtendBufferedRel(ExtendBufferedWhat eb, ForkNumber forkNum, BufferAccessStrategy strategy, uint32 flags)
Definition: bufmgr.c:812
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:4480
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:708
#define P_NEW
Definition: bufmgr.h:152
static Size BufferGetPageSize(Buffer buffer)
Definition: bufmgr.h:341
@ EB_LOCK_FIRST
Definition: bufmgr.h:85
#define EB_REL(p_rel)
Definition: bufmgr.h:106
#define DEBUG2
Definition: elog.h:29
BlockNumber GetFreeIndexPage(Relation rel)
Definition: indexfsm.c:38
static void _bt_log_reuse_page(Relation rel, Relation heaprel, BlockNumber blkno, FullTransactionId safexid)
Definition: nbtpage.c:827
void _bt_pageinit(Page page, Size size)
Definition: nbtpage.c:1126
void _bt_checkpage(Relation rel, Buffer buf)
Definition: nbtpage.c:793
bool _bt_conditionallockbuf(Relation rel, Buffer buf)
Definition: nbtpage.c:1090
void _bt_lockbuf(Relation rel, Buffer buf, int access)
Definition: nbtpage.c:1036
static FullTransactionId BTPageGetDeleteXid(Page page)
Definition: nbtree.h:260
static bool BTPageIsRecyclable(Page page, Relation heaprel)
Definition: nbtree.h:291
short access
Definition: preproc-type.c:36
@ MAIN_FORKNUM
Definition: relpath.h:50

References _bt_checkpage(), _bt_conditionallockbuf(), _bt_lockbuf(), _bt_log_reuse_page(), _bt_pageinit(), _bt_relbuf(), Assert(), BT_WRITE, BTPageGetDeleteXid(), BTPageIsRecyclable(), buf, BufferGetPage(), BufferGetPageSize(), DEBUG2, EB_LOCK_FIRST, EB_REL, elog(), ExtendBufferedRel(), GetFreeIndexPage(), InvalidBlockNumber, MAIN_FORKNUM, P_NEW, PageIsNew(), ReadBuffer(), RelationNeedsWAL, RelationUsesLocalBuffers, ReleaseBuffer(), VALGRIND_MAKE_MEM_DEFINED, and XLogStandbyInfoActive.

Referenced by _bt_finish_split(), _bt_getroot(), _bt_getrootheight(), _bt_getstackbuf(), _bt_gettrueroot(), _bt_insertonpg(), _bt_killitems(), _bt_leftsib_splitflag(), _bt_metaversion(), _bt_moveright(), _bt_newroot(), _bt_pagedel(), _bt_readnextpage(), _bt_rightsib_halfdeadflag(), _bt_set_cleanup_info(), _bt_split(), _bt_unlink_halfdead_page(), _bt_vacuum_needs_cleanup(), and _bt_walk_left().

◆ _bt_getroot()

Buffer _bt_getroot ( Relation  rel,
Relation  heaprel,
int  access 
)

Definition at line 342 of file nbtpage.c.

343 {
344  Buffer metabuf;
345  Buffer rootbuf;
346  Page rootpage;
347  BTPageOpaque rootopaque;
348  BlockNumber rootblkno;
349  uint32 rootlevel;
350  BTMetaPageData *metad;
351 
352  /*
353  * Try to use previously-cached metapage data to find the root. This
354  * normally saves one buffer access per index search, which is a very
355  * helpful savings in bufmgr traffic and hence contention.
356  */
357  if (rel->rd_amcache != NULL)
358  {
359  metad = (BTMetaPageData *) rel->rd_amcache;
360  /* We shouldn't have cached it if any of these fail */
361  Assert(metad->btm_magic == BTREE_MAGIC);
363  Assert(metad->btm_version <= BTREE_VERSION);
364  Assert(!metad->btm_allequalimage ||
366  Assert(metad->btm_root != P_NONE);
367 
368  rootblkno = metad->btm_fastroot;
369  Assert(rootblkno != P_NONE);
370  rootlevel = metad->btm_fastlevel;
371 
372  rootbuf = _bt_getbuf(rel, heaprel, rootblkno, BT_READ);
373  rootpage = BufferGetPage(rootbuf);
374  rootopaque = BTPageGetOpaque(rootpage);
375 
376  /*
377  * Since the cache might be stale, we check the page more carefully
378  * here than normal. We *must* check that it's not deleted. If it's
379  * not alone on its level, then we reject too --- this may be overly
380  * paranoid but better safe than sorry. Note we don't check P_ISROOT,
381  * because that's not set in a "fast root".
382  */
383  if (!P_IGNORE(rootopaque) &&
384  rootopaque->btpo_level == rootlevel &&
385  P_LEFTMOST(rootopaque) &&
386  P_RIGHTMOST(rootopaque))
387  {
388  /* OK, accept cached page as the root */
389  return rootbuf;
390  }
391  _bt_relbuf(rel, rootbuf);
392  /* Cache is stale, throw it away */
393  if (rel->rd_amcache)
394  pfree(rel->rd_amcache);
395  rel->rd_amcache = NULL;
396  }
397 
398  metabuf = _bt_getbuf(rel, heaprel, BTREE_METAPAGE, BT_READ);
399  metad = _bt_getmeta(rel, metabuf);
400 
401  /* if no root page initialized yet, do it */
402  if (metad->btm_root == P_NONE)
403  {
404  Page metapg;
405 
406  /* If access = BT_READ, caller doesn't want us to create root yet */
407  if (access == BT_READ)
408  {
409  _bt_relbuf(rel, metabuf);
410  return InvalidBuffer;
411  }
412 
413  /* trade in our read lock for a write lock */
414  _bt_unlockbuf(rel, metabuf);
415  _bt_lockbuf(rel, metabuf, BT_WRITE);
416 
417  /*
418  * Race condition: if someone else initialized the metadata between
419  * the time we released the read lock and acquired the write lock, we
420  * must avoid doing it again.
421  */
422  if (metad->btm_root != P_NONE)
423  {
424  /*
425  * Metadata initialized by someone else. In order to guarantee no
426  * deadlocks, we have to release the metadata page and start all
427  * over again. (Is that really true? But it's hardly worth trying
428  * to optimize this case.)
429  */
430  _bt_relbuf(rel, metabuf);
431  return _bt_getroot(rel, heaprel, access);
432  }
433 
434  /*
435  * Get, initialize, write, and leave a lock of the appropriate type on
436  * the new root page. Since this is the first page in the tree, it's
437  * a leaf as well as the root.
438  */
439  rootbuf = _bt_getbuf(rel, heaprel, P_NEW, BT_WRITE);
440  rootblkno = BufferGetBlockNumber(rootbuf);
441  rootpage = BufferGetPage(rootbuf);
442  rootopaque = BTPageGetOpaque(rootpage);
443  rootopaque->btpo_prev = rootopaque->btpo_next = P_NONE;
444  rootopaque->btpo_flags = (BTP_LEAF | BTP_ROOT);
445  rootopaque->btpo_level = 0;
446  rootopaque->btpo_cycleid = 0;
447  /* Get raw page pointer for metapage */
448  metapg = BufferGetPage(metabuf);
449 
450  /* NO ELOG(ERROR) till meta is updated */
452 
453  /* upgrade metapage if needed */
454  if (metad->btm_version < BTREE_NOVAC_VERSION)
455  _bt_upgrademetapage(metapg);
456 
457  metad->btm_root = rootblkno;
458  metad->btm_level = 0;
459  metad->btm_fastroot = rootblkno;
460  metad->btm_fastlevel = 0;
462  metad->btm_last_cleanup_num_heap_tuples = -1.0;
463 
464  MarkBufferDirty(rootbuf);
465  MarkBufferDirty(metabuf);
466 
467  /* XLOG stuff */
468  if (RelationNeedsWAL(rel))
469  {
470  xl_btree_newroot xlrec;
471  XLogRecPtr recptr;
473 
474  XLogBeginInsert();
477 
479  md.version = metad->btm_version;
480  md.root = rootblkno;
481  md.level = 0;
482  md.fastroot = rootblkno;
483  md.fastlevel = 0;
485  md.allequalimage = metad->btm_allequalimage;
486 
487  XLogRegisterBufData(2, (char *) &md, sizeof(xl_btree_metadata));
488 
489  xlrec.rootblk = rootblkno;
490  xlrec.level = 0;
491 
492  XLogRegisterData((char *) &xlrec, SizeOfBtreeNewroot);
493 
494  recptr = XLogInsert(RM_BTREE_ID, XLOG_BTREE_NEWROOT);
495 
496  PageSetLSN(rootpage, recptr);
497  PageSetLSN(metapg, recptr);
498  }
499 
501 
502  /*
503  * swap root write lock for read lock. There is no danger of anyone
504  * else accessing the new root page while it's unlocked, since no one
505  * else knows where it is yet.
506  */
507  _bt_unlockbuf(rel, rootbuf);
508  _bt_lockbuf(rel, rootbuf, BT_READ);
509 
510  /* okay, metadata is correct, release lock on it without caching */
511  _bt_relbuf(rel, metabuf);
512  }
513  else
514  {
515  rootblkno = metad->btm_fastroot;
516  Assert(rootblkno != P_NONE);
517  rootlevel = metad->btm_fastlevel;
518 
519  /*
520  * Cache the metapage data for next time
521  */
523  sizeof(BTMetaPageData));
524  memcpy(rel->rd_amcache, metad, sizeof(BTMetaPageData));
525 
526  /*
527  * We are done with the metapage; arrange to release it via first
528  * _bt_relandgetbuf call
529  */
530  rootbuf = metabuf;
531 
532  for (;;)
533  {
534  rootbuf = _bt_relandgetbuf(rel, rootbuf, rootblkno, BT_READ);
535  rootpage = BufferGetPage(rootbuf);
536  rootopaque = BTPageGetOpaque(rootpage);
537 
538  if (!P_IGNORE(rootopaque))
539  break;
540 
541  /* it's dead, Jim. step right one page */
542  if (P_RIGHTMOST(rootopaque))
543  elog(ERROR, "no live root page found in index \"%s\"",
545  rootblkno = rootopaque->btpo_next;
546  }
547 
548  if (rootopaque->btpo_level != rootlevel)
549  elog(ERROR, "root page %u of index \"%s\" has level %u, expected %u",
550  rootblkno, RelationGetRelationName(rel),
551  rootopaque->btpo_level, rootlevel);
552  }
553 
554  /*
555  * By here, we have a pin and read lock on the root page, and no lock set
556  * on the metadata page. Return the root page's buffer.
557  */
558  return rootbuf;
559 }
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:1021
void _bt_upgrademetapage(Page page)
Definition: nbtpage.c:108
static BTMetaPageData * _bt_getmeta(Relation rel, Buffer metabuf)
Definition: nbtpage.c:143
#define BTREE_MIN_VERSION
Definition: nbtree.h:151
#define BTP_LEAF
Definition: nbtree.h:76
#define BTREE_MAGIC
Definition: nbtree.h:149
#define BTP_ROOT
Definition: nbtree.h:77
#define SizeOfBtreeNewroot
Definition: nbtxlog.h:347
#define XLOG_BTREE_NEWROOT
Definition: nbtxlog.h:37
uint32 btm_last_cleanup_num_delpages
Definition: nbtree.h:114
uint32 btm_level
Definition: nbtree.h:108
float8 btm_last_cleanup_num_heap_tuples
Definition: nbtree.h:116
BlockNumber btm_fastroot
Definition: nbtree.h:109
uint32 btm_version
Definition: nbtree.h:106
uint32 btm_magic
Definition: nbtree.h:105
bool btm_allequalimage
Definition: nbtree.h:118
uint32 btm_fastlevel
Definition: nbtree.h:110
BlockNumber btpo_prev
Definition: nbtree.h:64
void * rd_amcache
Definition: rel.h:228
MemoryContext rd_indexcxt
Definition: rel.h:203
uint32 level
Definition: nbtxlog.h:50
uint32 version
Definition: nbtxlog.h:48
bool allequalimage
Definition: nbtxlog.h:54
BlockNumber fastroot
Definition: nbtxlog.h:51
uint32 fastlevel
Definition: nbtxlog.h:52
BlockNumber root
Definition: nbtxlog.h:49
uint32 last_cleanup_num_delpages
Definition: nbtxlog.h:53
uint32 level
Definition: nbtxlog.h:344
BlockNumber rootblk
Definition: nbtxlog.h:343
#define REGBUF_WILL_INIT
Definition: xloginsert.h:33

References _bt_getbuf(), _bt_getmeta(), _bt_lockbuf(), _bt_relandgetbuf(), _bt_relbuf(), _bt_unlockbuf(), _bt_upgrademetapage(), xl_btree_metadata::allequalimage, Assert(), BT_READ, BT_WRITE, BTMetaPageData::btm_allequalimage, BTMetaPageData::btm_fastlevel, BTMetaPageData::btm_fastroot, BTMetaPageData::btm_last_cleanup_num_delpages, BTMetaPageData::btm_last_cleanup_num_heap_tuples, BTMetaPageData::btm_level, BTMetaPageData::btm_magic, BTMetaPageData::btm_root, BTMetaPageData::btm_version, BTP_LEAF, BTP_ROOT, BTPageGetOpaque, BTPageOpaqueData::btpo_cycleid, BTPageOpaqueData::btpo_flags, BTPageOpaqueData::btpo_level, BTPageOpaqueData::btpo_next, BTPageOpaqueData::btpo_prev, BTREE_MAGIC, BTREE_METAPAGE, BTREE_MIN_VERSION, BTREE_NOVAC_VERSION, BTREE_VERSION, BufferGetBlockNumber(), BufferGetPage(), elog(), END_CRIT_SECTION, ERROR, xl_btree_metadata::fastlevel, xl_btree_metadata::fastroot, InvalidBuffer, xl_btree_metadata::last_cleanup_num_delpages, xl_btree_metadata::level, xl_btree_newroot::level, MarkBufferDirty(), MemoryContextAlloc(), P_IGNORE, P_LEFTMOST, P_NEW, P_NONE, P_RIGHTMOST, PageSetLSN(), pfree(), RelationData::rd_amcache, RelationData::rd_indexcxt, REGBUF_STANDARD, REGBUF_WILL_INIT, RelationGetRelationName, RelationNeedsWAL, xl_btree_metadata::root, xl_btree_newroot::rootblk, SizeOfBtreeNewroot, START_CRIT_SECTION, xl_btree_metadata::version, XLOG_BTREE_NEWROOT, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _bt_get_endpoint(), and _bt_search().

◆ _bt_getrootheight()

int _bt_getrootheight ( Relation  rel,
Relation  heaprel 
)

Definition at line 671 of file nbtpage.c.

672 {
673  BTMetaPageData *metad;
674 
675  if (rel->rd_amcache == NULL)
676  {
677  Buffer metabuf;
678 
679  metabuf = _bt_getbuf(rel, heaprel, BTREE_METAPAGE, BT_READ);
680  metad = _bt_getmeta(rel, metabuf);
681 
682  /*
683  * If there's no root page yet, _bt_getroot() doesn't expect a cache
684  * to be made, so just stop here and report the index height is zero.
685  * (XXX perhaps _bt_getroot() should be changed to allow this case.)
686  */
687  if (metad->btm_root == P_NONE)
688  {
689  _bt_relbuf(rel, metabuf);
690  return 0;
691  }
692 
693  /*
694  * Cache the metapage data for next time
695  */
697  sizeof(BTMetaPageData));
698  memcpy(rel->rd_amcache, metad, sizeof(BTMetaPageData));
699  _bt_relbuf(rel, metabuf);
700  }
701 
702  /* Get cached page */
703  metad = (BTMetaPageData *) rel->rd_amcache;
704  /* We shouldn't have cached it if any of these fail */
705  Assert(metad->btm_magic == BTREE_MAGIC);
707  Assert(metad->btm_version <= BTREE_VERSION);
708  Assert(!metad->btm_allequalimage ||
710  Assert(metad->btm_fastroot != P_NONE);
711 
712  return metad->btm_fastlevel;
713 }

References _bt_getbuf(), _bt_getmeta(), _bt_relbuf(), Assert(), BT_READ, BTMetaPageData::btm_allequalimage, BTMetaPageData::btm_fastlevel, BTMetaPageData::btm_fastroot, BTMetaPageData::btm_magic, BTMetaPageData::btm_root, BTMetaPageData::btm_version, BTREE_MAGIC, BTREE_METAPAGE, BTREE_MIN_VERSION, BTREE_NOVAC_VERSION, BTREE_VERSION, MemoryContextAlloc(), P_NONE, RelationData::rd_amcache, and RelationData::rd_indexcxt.

Referenced by _bt_insertonpg(), and get_relation_info().

◆ _bt_getstackbuf()

Buffer _bt_getstackbuf ( Relation  rel,
Relation  heaprel,
BTStack  stack,
BlockNumber  child 
)

Definition at line 2312 of file nbtinsert.c.

2313 {
2314  BlockNumber blkno;
2315  OffsetNumber start;
2316 
2317  blkno = stack->bts_blkno;
2318  start = stack->bts_offset;
2319 
2320  for (;;)
2321  {
2322  Buffer buf;
2323  Page page;
2324  BTPageOpaque opaque;
2325 
2326  buf = _bt_getbuf(rel, heaprel, blkno, BT_WRITE);
2327  page = BufferGetPage(buf);
2328  opaque = BTPageGetOpaque(page);
2329 
2330  if (P_INCOMPLETE_SPLIT(opaque))
2331  {
2332  _bt_finish_split(rel, heaprel, buf, stack->bts_parent);
2333  continue;
2334  }
2335 
2336  if (!P_IGNORE(opaque))
2337  {
2338  OffsetNumber offnum,
2339  minoff,
2340  maxoff;
2341  ItemId itemid;
2342  IndexTuple item;
2343 
2344  minoff = P_FIRSTDATAKEY(opaque);
2345  maxoff = PageGetMaxOffsetNumber(page);
2346 
2347  /*
2348  * start = InvalidOffsetNumber means "search the whole page". We
2349  * need this test anyway due to possibility that page has a high
2350  * key now when it didn't before.
2351  */
2352  if (start < minoff)
2353  start = minoff;
2354 
2355  /*
2356  * Need this check too, to guard against possibility that page
2357  * split since we visited it originally.
2358  */
2359  if (start > maxoff)
2360  start = OffsetNumberNext(maxoff);
2361 
2362  /*
2363  * These loops will check every item on the page --- but in an
2364  * order that's attuned to the probability of where it actually
2365  * is. Scan to the right first, then to the left.
2366  */
2367  for (offnum = start;
2368  offnum <= maxoff;
2369  offnum = OffsetNumberNext(offnum))
2370  {
2371  itemid = PageGetItemId(page, offnum);
2372  item = (IndexTuple) PageGetItem(page, itemid);
2373 
2374  if (BTreeTupleGetDownLink(item) == child)
2375  {
2376  /* Return accurate pointer to where link is now */
2377  stack->bts_blkno = blkno;
2378  stack->bts_offset = offnum;
2379  return buf;
2380  }
2381  }
2382 
2383  for (offnum = OffsetNumberPrev(start);
2384  offnum >= minoff;
2385  offnum = OffsetNumberPrev(offnum))
2386  {
2387  itemid = PageGetItemId(page, offnum);
2388  item = (IndexTuple) PageGetItem(page, itemid);
2389 
2390  if (BTreeTupleGetDownLink(item) == child)
2391  {
2392  /* Return accurate pointer to where link is now */
2393  stack->bts_blkno = blkno;
2394  stack->bts_offset = offnum;
2395  return buf;
2396  }
2397  }
2398  }
2399 
2400  /*
2401  * The item we're looking for moved right at least one page.
2402  *
2403  * Lehman and Yao couple/chain locks when moving right here, which we
2404  * can avoid. See nbtree/README.
2405  */
2406  if (P_RIGHTMOST(opaque))
2407  {
2408  _bt_relbuf(rel, buf);
2409  return InvalidBuffer;
2410  }
2411  blkno = opaque->btpo_next;
2412  start = InvalidOffsetNumber;
2413  _bt_relbuf(rel, buf);
2414  }
2415 }
void _bt_finish_split(Relation rel, Relation heaprel, Buffer lbuf, BTStack stack)
Definition: nbtinsert.c:2235
BlockNumber bts_blkno
Definition: nbtree.h:731
OffsetNumber bts_offset
Definition: nbtree.h:732

References _bt_finish_split(), _bt_getbuf(), _bt_relbuf(), BT_WRITE, BTPageGetOpaque, BTPageOpaqueData::btpo_next, BTreeTupleGetDownLink(), BTStackData::bts_blkno, BTStackData::bts_offset, BTStackData::bts_parent, buf, BufferGetPage(), InvalidBuffer, InvalidOffsetNumber, OffsetNumberNext, OffsetNumberPrev, P_FIRSTDATAKEY, P_IGNORE, P_INCOMPLETE_SPLIT, P_RIGHTMOST, PageGetItem(), PageGetItemId(), and PageGetMaxOffsetNumber().

Referenced by _bt_insert_parent(), and _bt_lock_subtree_parent().

◆ _bt_gettrueroot()

Buffer _bt_gettrueroot ( Relation  rel,
Relation  heaprel 
)

Definition at line 576 of file nbtpage.c.

577 {
578  Buffer metabuf;
579  Page metapg;
580  BTPageOpaque metaopaque;
581  Buffer rootbuf;
582  Page rootpage;
583  BTPageOpaque rootopaque;
584  BlockNumber rootblkno;
585  uint32 rootlevel;
586  BTMetaPageData *metad;
587 
588  /*
589  * We don't try to use cached metapage data here, since (a) this path is
590  * not performance-critical, and (b) if we are here it suggests our cache
591  * is out-of-date anyway. In light of point (b), it's probably safest to
592  * actively flush any cached metapage info.
593  */
594  if (rel->rd_amcache)
595  pfree(rel->rd_amcache);
596  rel->rd_amcache = NULL;
597 
598  metabuf = _bt_getbuf(rel, heaprel, BTREE_METAPAGE, BT_READ);
599  metapg = BufferGetPage(metabuf);
600  metaopaque = BTPageGetOpaque(metapg);
601  metad = BTPageGetMeta(metapg);
602 
603  if (!P_ISMETA(metaopaque) ||
604  metad->btm_magic != BTREE_MAGIC)
605  ereport(ERROR,
606  (errcode(ERRCODE_INDEX_CORRUPTED),
607  errmsg("index \"%s\" is not a btree",
608  RelationGetRelationName(rel))));
609 
610  if (metad->btm_version < BTREE_MIN_VERSION ||
611  metad->btm_version > BTREE_VERSION)
612  ereport(ERROR,
613  (errcode(ERRCODE_INDEX_CORRUPTED),
614  errmsg("version mismatch in index \"%s\": file version %d, "
615  "current version %d, minimal supported version %d",
618 
619  /* if no root page initialized yet, fail */
620  if (metad->btm_root == P_NONE)
621  {
622  _bt_relbuf(rel, metabuf);
623  return InvalidBuffer;
624  }
625 
626  rootblkno = metad->btm_root;
627  rootlevel = metad->btm_level;
628 
629  /*
630  * We are done with the metapage; arrange to release it via first
631  * _bt_relandgetbuf call
632  */
633  rootbuf = metabuf;
634 
635  for (;;)
636  {
637  rootbuf = _bt_relandgetbuf(rel, rootbuf, rootblkno, BT_READ);
638  rootpage = BufferGetPage(rootbuf);
639  rootopaque = BTPageGetOpaque(rootpage);
640 
641  if (!P_IGNORE(rootopaque))
642  break;
643 
644  /* it's dead, Jim. step right one page */
645  if (P_RIGHTMOST(rootopaque))
646  elog(ERROR, "no live root page found in index \"%s\"",
648  rootblkno = rootopaque->btpo_next;
649  }
650 
651  if (rootopaque->btpo_level != rootlevel)
652  elog(ERROR, "root page %u of index \"%s\" has level %u, expected %u",
653  rootblkno, RelationGetRelationName(rel),
654  rootopaque->btpo_level, rootlevel);
655 
656  return rootbuf;
657 }
#define P_ISMETA(opaque)
Definition: nbtree.h:223

References _bt_getbuf(), _bt_relandgetbuf(), _bt_relbuf(), BT_READ, BTMetaPageData::btm_level, BTMetaPageData::btm_magic, BTMetaPageData::btm_root, BTMetaPageData::btm_version, BTPageGetMeta, BTPageGetOpaque, BTPageOpaqueData::btpo_level, BTPageOpaqueData::btpo_next, BTREE_MAGIC, BTREE_METAPAGE, BTREE_MIN_VERSION, BTREE_VERSION, BufferGetPage(), elog(), ereport, errcode(), errmsg(), ERROR, InvalidBuffer, P_IGNORE, P_ISMETA, P_NONE, P_RIGHTMOST, pfree(), RelationData::rd_amcache, and RelationGetRelationName.

Referenced by _bt_get_endpoint().

◆ _bt_initmetapage()

void _bt_initmetapage ( Page  page,
BlockNumber  rootbknum,