PostgreSQL Source Code  git master
nbtsort.c File Reference
#include "postgres.h"
#include "access/nbtree.h"
#include "access/parallel.h"
#include "access/relscan.h"
#include "access/table.h"
#include "access/tableam.h"
#include "access/xact.h"
#include "access/xlog.h"
#include "access/xloginsert.h"
#include "catalog/index.h"
#include "commands/progress.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "storage/smgr.h"
#include "tcop/tcopprot.h"
#include "utils/rel.h"
#include "utils/sortsupport.h"
#include "utils/tuplesort.h"
Include dependency graph for nbtsort.c:

Go to the source code of this file.

Data Structures

struct  BTSpool
 
struct  BTShared
 
struct  BTLeader
 
struct  BTBuildState
 
struct  BTPageState
 
struct  BTWriteState
 

Macros

#define PARALLEL_KEY_BTREE_SHARED   UINT64CONST(0xA000000000000001)
 
#define PARALLEL_KEY_TUPLESORT   UINT64CONST(0xA000000000000002)
 
#define PARALLEL_KEY_TUPLESORT_SPOOL2   UINT64CONST(0xA000000000000003)
 
#define PARALLEL_KEY_QUERY_TEXT   UINT64CONST(0xA000000000000004)
 
#define ParallelTableScanFromBTShared(shared)   (ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(BTShared)))
 

Typedefs

typedef struct BTSpool BTSpool
 
typedef struct BTShared BTShared
 
typedef struct BTLeader BTLeader
 
typedef struct BTBuildState BTBuildState
 
typedef struct BTPageState BTPageState
 
typedef struct BTWriteState BTWriteState
 

Functions

static double _bt_spools_heapscan (Relation heap, Relation index, BTBuildState *buildstate, IndexInfo *indexInfo)
 
static void _bt_spooldestroy (BTSpool *btspool)
 
static void _bt_spool (BTSpool *btspool, ItemPointer self, Datum *values, bool *isnull)
 
static void _bt_leafbuild (BTSpool *btspool, BTSpool *btspool2)
 
static void _bt_build_callback (Relation index, ItemPointer tid, Datum *values, bool *isnull, bool tupleIsAlive, void *state)
 
static Page _bt_blnewpage (uint32 level)
 
static BTPageState_bt_pagestate (BTWriteState *wstate, uint32 level)
 
static void _bt_slideleft (Page page)
 
static void _bt_sortaddtup (Page page, Size itemsize, IndexTuple itup, OffsetNumber itup_off)
 
static void _bt_buildadd (BTWriteState *wstate, BTPageState *state, IndexTuple itup)
 
static void _bt_uppershutdown (BTWriteState *wstate, BTPageState *state)
 
static void _bt_load (BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
 
static void _bt_begin_parallel (BTBuildState *buildstate, bool isconcurrent, int request)
 
static void _bt_end_parallel (BTLeader *btleader)
 
static Size _bt_parallel_estimate_shared (Relation heap, Snapshot snapshot)
 
static double _bt_parallel_heapscan (BTBuildState *buildstate, bool *brokenhotchain)
 
static void _bt_leader_participate_as_worker (BTBuildState *buildstate)
 
static void _bt_parallel_scan_and_sort (BTSpool *btspool, BTSpool *btspool2, BTShared *btshared, Sharedsort *sharedsort, Sharedsort *sharedsort2, int sortmem, bool progress)
 
IndexBuildResultbtbuild (Relation heap, Relation index, IndexInfo *indexInfo)
 
static void _bt_blwritepage (BTWriteState *wstate, Page page, BlockNumber blkno)
 
void _bt_parallel_build_main (dsm_segment *seg, shm_toc *toc)
 

Macro Definition Documentation

◆ PARALLEL_KEY_BTREE_SHARED

#define PARALLEL_KEY_BTREE_SHARED   UINT64CONST(0xA000000000000001)

Definition at line 80 of file nbtsort.c.

Referenced by _bt_begin_parallel(), and _bt_parallel_build_main().

◆ PARALLEL_KEY_QUERY_TEXT

#define PARALLEL_KEY_QUERY_TEXT   UINT64CONST(0xA000000000000004)

Definition at line 83 of file nbtsort.c.

Referenced by _bt_begin_parallel(), and _bt_parallel_build_main().

◆ PARALLEL_KEY_TUPLESORT

#define PARALLEL_KEY_TUPLESORT   UINT64CONST(0xA000000000000002)

Definition at line 81 of file nbtsort.c.

Referenced by _bt_begin_parallel(), and _bt_parallel_build_main().

◆ PARALLEL_KEY_TUPLESORT_SPOOL2

#define PARALLEL_KEY_TUPLESORT_SPOOL2   UINT64CONST(0xA000000000000003)

Definition at line 82 of file nbtsort.c.

Referenced by _bt_begin_parallel(), and _bt_parallel_build_main().

◆ ParallelTableScanFromBTShared

#define ParallelTableScanFromBTShared (   shared)    (ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(BTShared)))

Definition at line 173 of file nbtsort.c.

Referenced by _bt_begin_parallel(), and _bt_parallel_scan_and_sort().

Typedef Documentation

◆ BTBuildState

typedef struct BTBuildState BTBuildState

◆ BTLeader

typedef struct BTLeader BTLeader

◆ BTPageState

typedef struct BTPageState BTPageState

◆ BTShared

typedef struct BTShared BTShared

◆ BTSpool

typedef struct BTSpool BTSpool

◆ BTWriteState

typedef struct BTWriteState BTWriteState

Function Documentation

◆ _bt_begin_parallel()

static void _bt_begin_parallel ( BTBuildState buildstate,
bool  isconcurrent,
int  request 
)
static

Definition at line 1306 of file nbtsort.c.

References _bt_end_parallel(), _bt_leader_participate_as_worker(), _bt_parallel_estimate_shared(), Assert, BTShared::brokenhotchain, BTBuildState::btleader, BTLeader::btshared, ConditionVariableInit(), CreateParallelContext(), debug_query_string, EnterParallelMode(), ParallelContext::estimator, GetTransactionSnapshot(), BTShared::havedead, BTSpool::heap, BTShared::heaprelid, BTSpool::index, BTShared::indexrelid, BTShared::indtuples, InitializeParallelDSM(), BTShared::isconcurrent, BTSpool::isunique, BTShared::isunique, LaunchParallelWorkers(), BTShared::mutex, BTShared::nparticipantsdone, BTLeader::nparticipanttuplesorts, ParallelContext::nworkers_launched, palloc0(), PARALLEL_KEY_BTREE_SHARED, PARALLEL_KEY_QUERY_TEXT, PARALLEL_KEY_TUPLESORT, PARALLEL_KEY_TUPLESORT_SPOOL2, ParallelTableScanFromBTShared, BTLeader::pcxt, RegisterSnapshot(), RelationGetRelid, BTShared::reltuples, BTShared::scantuplesortstates, ParallelContext::seg, BTLeader::sharedsort, BTLeader::sharedsort2, shm_toc_allocate(), shm_toc_estimate_chunk, shm_toc_estimate_keys, shm_toc_insert(), BTLeader::snapshot, SnapshotAny, SpinLockInit, BTBuildState::spool, table_parallelscan_initialize(), ParallelContext::toc, tuplesort_estimate_shared(), tuplesort_initialize_shared(), WaitForParallelWorkersToAttach(), and BTShared::workersdonecv.

Referenced by _bt_spools_heapscan().

1307 {
1308  ParallelContext *pcxt;
1309  int scantuplesortstates;
1310  Snapshot snapshot;
1311  Size estbtshared;
1312  Size estsort;
1313  BTShared *btshared;
1314  Sharedsort *sharedsort;
1315  Sharedsort *sharedsort2;
1316  BTSpool *btspool = buildstate->spool;
1317  BTLeader *btleader = (BTLeader *) palloc0(sizeof(BTLeader));
1318  bool leaderparticipates = true;
1319  char *sharedquery;
1320  int querylen;
1321 
1322 #ifdef DISABLE_LEADER_PARTICIPATION
1323  leaderparticipates = false;
1324 #endif
1325 
1326  /*
1327  * Enter parallel mode, and create context for parallel build of btree
1328  * index
1329  */
1331  Assert(request > 0);
1332  pcxt = CreateParallelContext("postgres", "_bt_parallel_build_main",
1333  request);
1334  scantuplesortstates = leaderparticipates ? request + 1 : request;
1335 
1336  /*
1337  * Prepare for scan of the base relation. In a normal index build, we use
1338  * SnapshotAny because we must retrieve all tuples and do our own time
1339  * qual checks (because we have to index RECENTLY_DEAD tuples). In a
1340  * concurrent build, we take a regular MVCC snapshot and index whatever's
1341  * live according to that.
1342  */
1343  if (!isconcurrent)
1344  snapshot = SnapshotAny;
1345  else
1347 
1348  /*
1349  * Estimate size for our own PARALLEL_KEY_BTREE_SHARED workspace, and
1350  * PARALLEL_KEY_TUPLESORT tuplesort workspace
1351  */
1352  estbtshared = _bt_parallel_estimate_shared(btspool->heap, snapshot);
1353  shm_toc_estimate_chunk(&pcxt->estimator, estbtshared);
1354  estsort = tuplesort_estimate_shared(scantuplesortstates);
1355  shm_toc_estimate_chunk(&pcxt->estimator, estsort);
1356 
1357  /*
1358  * Unique case requires a second spool, and so we may have to account for
1359  * another shared workspace for that -- PARALLEL_KEY_TUPLESORT_SPOOL2
1360  */
1361  if (!btspool->isunique)
1362  shm_toc_estimate_keys(&pcxt->estimator, 2);
1363  else
1364  {
1365  shm_toc_estimate_chunk(&pcxt->estimator, estsort);
1366  shm_toc_estimate_keys(&pcxt->estimator, 3);
1367  }
1368 
1369  /* Finally, estimate PARALLEL_KEY_QUERY_TEXT space */
1370  querylen = strlen(debug_query_string);
1371  shm_toc_estimate_chunk(&pcxt->estimator, querylen + 1);
1372  shm_toc_estimate_keys(&pcxt->estimator, 1);
1373 
1374  /* Everyone's had a chance to ask for space, so now create the DSM */
1375  InitializeParallelDSM(pcxt);
1376 
1377  /* Store shared build state, for which we reserved space */
1378  btshared = (BTShared *) shm_toc_allocate(pcxt->toc, estbtshared);
1379  /* Initialize immutable state */
1380  btshared->heaprelid = RelationGetRelid(btspool->heap);
1381  btshared->indexrelid = RelationGetRelid(btspool->index);
1382  btshared->isunique = btspool->isunique;
1383  btshared->isconcurrent = isconcurrent;
1384  btshared->scantuplesortstates = scantuplesortstates;
1386  SpinLockInit(&btshared->mutex);
1387  /* Initialize mutable state */
1388  btshared->nparticipantsdone = 0;
1389  btshared->reltuples = 0.0;
1390  btshared->havedead = false;
1391  btshared->indtuples = 0.0;
1392  btshared->brokenhotchain = false;
1395  snapshot);
1396 
1397  /*
1398  * Store shared tuplesort-private state, for which we reserved space.
1399  * Then, initialize opaque state using tuplesort routine.
1400  */
1401  sharedsort = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
1402  tuplesort_initialize_shared(sharedsort, scantuplesortstates,
1403  pcxt->seg);
1404 
1405  shm_toc_insert(pcxt->toc, PARALLEL_KEY_BTREE_SHARED, btshared);
1406  shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT, sharedsort);
1407 
1408  /* Unique case requires a second spool, and associated shared state */
1409  if (!btspool->isunique)
1410  sharedsort2 = NULL;
1411  else
1412  {
1413  /*
1414  * Store additional shared tuplesort-private state, for which we
1415  * reserved space. Then, initialize opaque state using tuplesort
1416  * routine.
1417  */
1418  sharedsort2 = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
1419  tuplesort_initialize_shared(sharedsort2, scantuplesortstates,
1420  pcxt->seg);
1421 
1422  shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT_SPOOL2, sharedsort2);
1423  }
1424 
1425  /* Store query string for workers */
1426  sharedquery = (char *) shm_toc_allocate(pcxt->toc, querylen + 1);
1427  memcpy(sharedquery, debug_query_string, querylen + 1);
1428  shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, sharedquery);
1429 
1430  /* Launch workers, saving status for leader/caller */
1431  LaunchParallelWorkers(pcxt);
1432  btleader->pcxt = pcxt;
1433  btleader->nparticipanttuplesorts = pcxt->nworkers_launched;
1434  if (leaderparticipates)
1435  btleader->nparticipanttuplesorts++;
1436  btleader->btshared = btshared;
1437  btleader->sharedsort = sharedsort;
1438  btleader->sharedsort2 = sharedsort2;
1439  btleader->snapshot = snapshot;
1440 
1441  /* If no workers were successfully launched, back out (do serial build) */
1442  if (pcxt->nworkers_launched == 0)
1443  {
1444  _bt_end_parallel(btleader);
1445  return;
1446  }
1447 
1448  /* Save leader state now that it's clear build will be parallel */
1449  buildstate->btleader = btleader;
1450 
1451  /* Join heap scan ourselves */
1452  if (leaderparticipates)
1454 
1455  /*
1456  * Caller needs to wait for all launched workers when we return. Make
1457  * sure that the failure-to-start case will not hang forever.
1458  */
1460 }
Sharedsort * sharedsort2
Definition: nbtsort.c:204
int scantuplesortstates
Definition: nbtsort.c:120
#define PARALLEL_KEY_TUPLESORT
Definition: nbtsort.c:81
ParallelContext * CreateParallelContext(const char *library_name, const char *function_name, int nworkers)
Definition: parallel.c:158
BTShared * btshared
Definition: nbtsort.c:202
int nparticipantsdone
Definition: nbtsort.c:154
Snapshot RegisterSnapshot(Snapshot snapshot)
Definition: snapmgr.c:865
Oid indexrelid
Definition: nbtsort.c:117
dsm_segment * seg
Definition: parallel.h:42
Sharedsort * sharedsort
Definition: nbtsort.c:203
shm_toc_estimator estimator
Definition: parallel.h:41
#define SpinLockInit(lock)
Definition: spin.h:60
Snapshot snapshot
Definition: nbtsort.c:205
bool isunique
Definition: nbtsort.c:118
#define ParallelTableScanFromBTShared(shared)
Definition: nbtsort.c:173
bool isconcurrent
Definition: nbtsort.c:119
static void _bt_end_parallel(BTLeader *btleader)
Definition: nbtsort.c:1466
void tuplesort_initialize_shared(Sharedsort *shared, int nWorkers, dsm_segment *seg)
Definition: tuplesort.c:4332
#define shm_toc_estimate_chunk(e, sz)
Definition: shm_toc.h:51
BTSpool * spool
Definition: nbtsort.c:219
Snapshot GetTransactionSnapshot(void)
Definition: snapmgr.c:306
ParallelContext * pcxt
Definition: nbtsort.c:182
Relation heap
Definition: nbtsort.c:99
void ConditionVariableInit(ConditionVariable *cv)
bool isunique
Definition: nbtsort.c:101
Oid heaprelid
Definition: nbtsort.c:116
#define PARALLEL_KEY_QUERY_TEXT
Definition: nbtsort.c:83
#define PARALLEL_KEY_TUPLESORT_SPOOL2
Definition: nbtsort.c:82
int nworkers_launched
Definition: parallel.h:37
bool brokenhotchain
Definition: nbtsort.c:158
void LaunchParallelWorkers(ParallelContext *pcxt)
Definition: parallel.c:493
Relation index
Definition: nbtsort.c:100
const char * debug_query_string
Definition: postgres.c:88
void InitializeParallelDSM(ParallelContext *pcxt)
Definition: parallel.c:195
void * palloc0(Size size)
Definition: mcxt.c:980
BTLeader * btleader
Definition: nbtsort.c:233
double reltuples
Definition: nbtsort.c:155
double indtuples
Definition: nbtsort.c:157
slock_t mutex
Definition: nbtsort.c:136
bool havedead
Definition: nbtsort.c:156
#define Assert(condition)
Definition: c.h:739
int nparticipanttuplesorts
Definition: nbtsort.c:190
size_t Size
Definition: c.h:467
static Size _bt_parallel_estimate_shared(Relation heap, Snapshot snapshot)
Definition: nbtsort.c:1482
#define shm_toc_estimate_keys(e, cnt)
Definition: shm_toc.h:53
void EnterParallelMode(void)
Definition: xact.c:963
void * shm_toc_allocate(shm_toc *toc, Size nbytes)
Definition: shm_toc.c:88
static void _bt_leader_participate_as_worker(BTBuildState *buildstate)
Definition: nbtsort.c:1536
Size tuplesort_estimate_shared(int nWorkers)
Definition: tuplesort.c:4311
#define SnapshotAny
Definition: snapmgr.h:69
void shm_toc_insert(shm_toc *toc, uint64 key, void *address)
Definition: shm_toc.c:171
ConditionVariable workersdonecv
Definition: nbtsort.c:128
#define PARALLEL_KEY_BTREE_SHARED
Definition: nbtsort.c:80
void table_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan, Snapshot snapshot)
Definition: tableam.c:141
void WaitForParallelWorkersToAttach(ParallelContext *pcxt)
Definition: parallel.c:613
#define RelationGetRelid(relation)
Definition: rel.h:428
shm_toc * toc
Definition: parallel.h:44

◆ _bt_blnewpage()

static Page _bt_blnewpage ( uint32  level)
static

Definition at line 616 of file nbtsort.c.

References _bt_pageinit(), BTP_LEAF, BTPageOpaqueData::btpo, BTPageOpaqueData::btpo_cycleid, BTPageOpaqueData::btpo_flags, BTPageOpaqueData::btpo_next, BTPageOpaqueData::btpo_prev, BTPageOpaqueData::level, P_NONE, PageGetSpecialPointer, and palloc().

Referenced by _bt_buildadd(), and _bt_pagestate().

617 {
618  Page page;
619  BTPageOpaque opaque;
620 
621  page = (Page) palloc(BLCKSZ);
622 
623  /* Zero the page and set up standard page header info */
624  _bt_pageinit(page, BLCKSZ);
625 
626  /* Initialize BT opaque state */
627  opaque = (BTPageOpaque) PageGetSpecialPointer(page);
628  opaque->btpo_prev = opaque->btpo_next = P_NONE;
629  opaque->btpo.level = level;
630  opaque->btpo_flags = (level > 0) ? 0 : BTP_LEAF;
631  opaque->btpo_cycleid = 0;
632 
633  /* Make the P_HIKEY line pointer appear allocated */
634  ((PageHeader) page)->pd_lower += sizeof(ItemIdData);
635 
636  return page;
637 }
BlockNumber btpo_next
Definition: nbtree.h:59
#define BTP_LEAF
Definition: nbtree.h:72
union BTPageOpaqueData::@46 btpo
#define P_NONE
Definition: nbtree.h:182
BTPageOpaqueData * BTPageOpaque
Definition: nbtree.h:69
BTCycleId btpo_cycleid
Definition: nbtree.h:66
BlockNumber btpo_prev
Definition: nbtree.h:58
struct ItemIdData ItemIdData
uint32 level
Definition: nbtree.h:62
PageHeaderData * PageHeader
Definition: bufpage.h:166
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
void * palloc(Size size)
Definition: mcxt.c:949
void _bt_pageinit(Page page, Size size)
Definition: nbtpage.c:924
uint16 btpo_flags
Definition: nbtree.h:65
Pointer Page
Definition: bufpage.h:78

◆ _bt_blwritepage()

static void _bt_blwritepage ( BTWriteState wstate,
Page  page,
BlockNumber  blkno 
)
static

Definition at line 643 of file nbtsort.c.

References BTWriteState::btws_pages_written, BTWriteState::btws_use_wal, BTWriteState::btws_zeropage, BTWriteState::index, log_newpage(), MAIN_FORKNUM, PageSetChecksumInplace(), palloc0(), pfree(), RelationData::rd_node, RelationData::rd_smgr, RelationOpenSmgr, smgrextend(), and smgrwrite().

Referenced by _bt_buildadd(), and _bt_uppershutdown().

644 {
645  /* Ensure rd_smgr is open (could have been closed by relcache flush!) */
646  RelationOpenSmgr(wstate->index);
647 
648  /* XLOG stuff */
649  if (wstate->btws_use_wal)
650  {
651  /* We use the XLOG_FPI record type for this */
652  log_newpage(&wstate->index->rd_node, MAIN_FORKNUM, blkno, page, true);
653  }
654 
655  /*
656  * If we have to write pages nonsequentially, fill in the space with
657  * zeroes until we come back and overwrite. This is not logically
658  * necessary on standard Unix filesystems (unwritten space will read as
659  * zeroes anyway), but it should help to avoid fragmentation. The dummy
660  * pages aren't WAL-logged though.
661  */
662  while (blkno > wstate->btws_pages_written)
663  {
664  if (!wstate->btws_zeropage)
665  wstate->btws_zeropage = (Page) palloc0(BLCKSZ);
666  /* don't set checksum for all-zero page */
668  wstate->btws_pages_written++,
669  (char *) wstate->btws_zeropage,
670  true);
671  }
672 
673  PageSetChecksumInplace(page, blkno);
674 
675  /*
676  * Now write the page. There's no need for smgr to schedule an fsync for
677  * this write; we'll do it ourselves before ending the build.
678  */
679  if (blkno == wstate->btws_pages_written)
680  {
681  /* extending the file... */
682  smgrextend(wstate->index->rd_smgr, MAIN_FORKNUM, blkno,
683  (char *) page, true);
684  wstate->btws_pages_written++;
685  }
686  else
687  {
688  /* overwriting a block we zero-filled before */
689  smgrwrite(wstate->index->rd_smgr, MAIN_FORKNUM, blkno,
690  (char *) page, true);
691  }
692 
693  pfree(page);
694 }
Page btws_zeropage
Definition: nbtsort.c:262
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
bool btws_use_wal
Definition: nbtsort.c:259
#define RelationOpenSmgr(relation)
Definition: rel.h:485
void pfree(void *pointer)
Definition: mcxt.c:1056
void smgrwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:530
Relation index
Definition: nbtsort.c:257
BlockNumber btws_pages_written
Definition: nbtsort.c:261
void * palloc0(Size size)
Definition: mcxt.c:980
RelFileNode rd_node
Definition: rel.h:55
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1198
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:483
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
Pointer Page
Definition: bufpage.h:78

◆ _bt_build_callback()

static void _bt_build_callback ( Relation  index,
ItemPointer  tid,
Datum values,
bool isnull,
bool  tupleIsAlive,
void *  state 
)
static

Definition at line 587 of file nbtsort.c.

References _bt_spool(), BTBuildState::havedead, BTBuildState::indtuples, BTBuildState::spool, and BTBuildState::spool2.

Referenced by _bt_parallel_scan_and_sort(), and _bt_spools_heapscan().

593 {
594  BTBuildState *buildstate = (BTBuildState *) state;
595 
596  /*
597  * insert the index tuple into the appropriate spool file for subsequent
598  * processing
599  */
600  if (tupleIsAlive || buildstate->spool2 == NULL)
601  _bt_spool(buildstate->spool, tid, values, isnull);
602  else
603  {
604  /* dead tuples are put into spool2 */
605  buildstate->havedead = true;
606  _bt_spool(buildstate->spool2, tid, values, isnull);
607  }
608 
609  buildstate->indtuples += 1;
610 }
bool havedead
Definition: nbtsort.c:217
BTSpool * spool
Definition: nbtsort.c:219
BTSpool * spool2
Definition: nbtsort.c:225
double indtuples
Definition: nbtsort.c:226
Definition: regguts.h:298
static Datum values[MAXATTR]
Definition: bootstrap.c:167
static void _bt_spool(BTSpool *btspool, ItemPointer self, Datum *values, bool *isnull)
Definition: nbtsort.c:530

◆ _bt_buildadd()

static void _bt_buildadd ( BTWriteState wstate,
BTPageState state,
IndexTuple  itup 
)
static

Definition at line 827 of file nbtsort.c.

References _bt_blnewpage(), _bt_blwritepage(), _bt_check_third_page(), _bt_pagestate(), _bt_sortaddtup(), _bt_truncate(), Assert, BTMaxItemSize, BTPageOpaqueData::btpo_next, BTPageOpaqueData::btpo_prev, BTPageState::btps_blkno, BTPageState::btps_full, BTPageState::btps_lastoff, BTPageState::btps_level, BTPageState::btps_lowkey, BTPageState::btps_next, BTPageState::btps_page, BTreeTupleGetNAtts, BTreeTupleSetDownLink, BTreeTupleSetNAtts, BTWriteState::btws_pages_alloced, CHECK_FOR_INTERRUPTS, CopyIndexTuple(), elog, ERROR, BTWriteState::heap, BTWriteState::index, IndexRelationGetNumberOfKeyAttributes, IndexTupleSize, BTWriteState::inskey, ItemIdGetLength, ItemIdSetUnused, MAXALIGN, OffsetNumberNext, OffsetNumberPrev, P_FIRSTKEY, P_HIKEY, P_LEFTMOST, P_NONE, PageGetFreeSpace(), PageGetItem, PageGetItemId, PageGetSpecialPointer, PageIndexTupleOverwrite(), palloc0(), pfree(), IndexTupleData::t_info, and unlikely.

Referenced by _bt_load(), and _bt_uppershutdown().

828 {
829  Page npage;
830  BlockNumber nblkno;
831  OffsetNumber last_off;
832  Size pgspc;
833  Size itupsz;
834  bool isleaf;
835 
836  /*
837  * This is a handy place to check for cancel interrupts during the btree
838  * load phase of index creation.
839  */
841 
842  npage = state->btps_page;
843  nblkno = state->btps_blkno;
844  last_off = state->btps_lastoff;
845 
846  pgspc = PageGetFreeSpace(npage);
847  itupsz = IndexTupleSize(itup);
848  itupsz = MAXALIGN(itupsz);
849  /* Leaf case has slightly different rules due to suffix truncation */
850  isleaf = (state->btps_level == 0);
851 
852  /*
853  * Check whether the new item can fit on a btree page on current level at
854  * all.
855  *
856  * Every newly built index will treat heap TID as part of the keyspace,
857  * which imposes the requirement that new high keys must occasionally have
858  * a heap TID appended within _bt_truncate(). That may leave a new pivot
859  * tuple one or two MAXALIGN() quantums larger than the original first
860  * right tuple it's derived from. v4 deals with the problem by decreasing
861  * the limit on the size of tuples inserted on the leaf level by the same
862  * small amount. Enforce the new v4+ limit on the leaf level, and the old
863  * limit on internal levels, since pivot tuples may need to make use of
864  * the reserved space. This should never fail on internal pages.
865  */
866  if (unlikely(itupsz > BTMaxItemSize(npage)))
867  _bt_check_third_page(wstate->index, wstate->heap, isleaf, npage,
868  itup);
869 
870  /*
871  * Check to see if current page will fit new item, with space left over to
872  * append a heap TID during suffix truncation when page is a leaf page.
873  *
874  * It is guaranteed that we can fit at least 2 non-pivot tuples plus a
875  * high key with heap TID when finishing off a leaf page, since we rely on
876  * _bt_check_third_page() rejecting oversized non-pivot tuples. On
877  * internal pages we can always fit 3 pivot tuples with larger internal
878  * page tuple limit (includes page high key).
879  *
880  * Most of the time, a page is only "full" in the sense that the soft
881  * fillfactor-wise limit has been exceeded. However, we must always leave
882  * at least two items plus a high key on each page before starting a new
883  * page. Disregard fillfactor and insert on "full" current page if we
884  * don't have the minimum number of items yet. (Note that we deliberately
885  * assume that suffix truncation neither enlarges nor shrinks new high key
886  * when applying soft limit.)
887  */
888  if (pgspc < itupsz + (isleaf ? MAXALIGN(sizeof(ItemPointerData)) : 0) ||
889  (pgspc < state->btps_full && last_off > P_FIRSTKEY))
890  {
891  /*
892  * Finish off the page and write it out.
893  */
894  Page opage = npage;
895  BlockNumber oblkno = nblkno;
896  ItemId ii;
897  ItemId hii;
898  IndexTuple oitup;
899 
900  /* Create new page of same level */
901  npage = _bt_blnewpage(state->btps_level);
902 
903  /* and assign it a page position */
904  nblkno = wstate->btws_pages_alloced++;
905 
906  /*
907  * We copy the last item on the page into the new page, and then
908  * rearrange the old page so that the 'last item' becomes its high key
909  * rather than a true data item. There had better be at least two
910  * items on the page already, else the page would be empty of useful
911  * data.
912  */
913  Assert(last_off > P_FIRSTKEY);
914  ii = PageGetItemId(opage, last_off);
915  oitup = (IndexTuple) PageGetItem(opage, ii);
916  _bt_sortaddtup(npage, ItemIdGetLength(ii), oitup, P_FIRSTKEY);
917 
918  /*
919  * Move 'last' into the high key position on opage. _bt_blnewpage()
920  * allocated empty space for a line pointer when opage was first
921  * created, so this is a matter of rearranging already-allocated space
922  * on page, and initializing high key line pointer. (Actually, leaf
923  * pages must also swap oitup with a truncated version of oitup, which
924  * is sometimes larger than oitup, though never by more than the space
925  * needed to append a heap TID.)
926  */
927  hii = PageGetItemId(opage, P_HIKEY);
928  *hii = *ii;
929  ItemIdSetUnused(ii); /* redundant */
930  ((PageHeader) opage)->pd_lower -= sizeof(ItemIdData);
931 
932  if (isleaf)
933  {
934  IndexTuple lastleft;
935  IndexTuple truncated;
936 
937  /*
938  * Truncate away any unneeded attributes from high key on leaf
939  * level. This is only done at the leaf level because downlinks
940  * in internal pages are either negative infinity items, or get
941  * their contents from copying from one level down. See also:
942  * _bt_split().
943  *
944  * We don't try to bias our choice of split point to make it more
945  * likely that _bt_truncate() can truncate away more attributes,
946  * whereas the split point used within _bt_split() is chosen much
947  * more delicately. Suffix truncation is mostly useful because it
948  * improves space utilization for workloads with random
949  * insertions. It doesn't seem worthwhile to add logic for
950  * choosing a split point here for a benefit that is bound to be
951  * much smaller.
952  *
953  * Overwrite the old item with new truncated high key directly.
954  * oitup is already located at the physical beginning of tuple
955  * space, so this should directly reuse the existing tuple space.
956  */
957  ii = PageGetItemId(opage, OffsetNumberPrev(last_off));
958  lastleft = (IndexTuple) PageGetItem(opage, ii);
959 
960  truncated = _bt_truncate(wstate->index, lastleft, oitup,
961  wstate->inskey);
962  if (!PageIndexTupleOverwrite(opage, P_HIKEY, (Item) truncated,
963  IndexTupleSize(truncated)))
964  elog(ERROR, "failed to add high key to the index page");
965  pfree(truncated);
966 
967  /* oitup should continue to point to the page's high key */
968  hii = PageGetItemId(opage, P_HIKEY);
969  oitup = (IndexTuple) PageGetItem(opage, hii);
970  }
971 
972  /*
973  * Link the old page into its parent, using its low key. If we don't
974  * have a parent, we have to create one; this adds a new btree level.
975  */
976  if (state->btps_next == NULL)
977  state->btps_next = _bt_pagestate(wstate, state->btps_level + 1);
978 
979  Assert((BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) <=
981  BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) > 0) ||
983  Assert(BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) == 0 ||
985  BTreeTupleSetDownLink(state->btps_lowkey, oblkno);
986  _bt_buildadd(wstate, state->btps_next, state->btps_lowkey);
987  pfree(state->btps_lowkey);
988 
989  /*
990  * Save a copy of the high key from the old page. It is also the low
991  * key for the new page.
992  */
993  state->btps_lowkey = CopyIndexTuple(oitup);
994 
995  /*
996  * Set the sibling links for both pages.
997  */
998  {
1000  BTPageOpaque nopaque = (BTPageOpaque) PageGetSpecialPointer(npage);
1001 
1002  oopaque->btpo_next = nblkno;
1003  nopaque->btpo_prev = oblkno;
1004  nopaque->btpo_next = P_NONE; /* redundant */
1005  }
1006 
1007  /*
1008  * Write out the old page. We never need to touch it again, so we can
1009  * free the opage workspace too.
1010  */
1011  _bt_blwritepage(wstate, opage, oblkno);
1012 
1013  /*
1014  * Reset last_off to point to new page
1015  */
1016  last_off = P_FIRSTKEY;
1017  }
1018 
1019  /*
1020  * By here, either original page is still the current page, or a new page
1021  * was created that became the current page. Either way, the current page
1022  * definitely has space for new item.
1023  *
1024  * If the new item is the first for its page, it must also be the first
1025  * item on its entire level. On later same-level pages, a low key for a
1026  * page will be copied from the prior page in the code above. Generate a
1027  * minus infinity low key here instead.
1028  */
1029  if (last_off == P_HIKEY)
1030  {
1031  Assert(state->btps_lowkey == NULL);
1032  state->btps_lowkey = palloc0(sizeof(IndexTupleData));
1033  state->btps_lowkey->t_info = sizeof(IndexTupleData);
1034  BTreeTupleSetNAtts(state->btps_lowkey, 0);
1035  }
1036 
1037  /*
1038  * Add the new item into the current page.
1039  */
1040  last_off = OffsetNumberNext(last_off);
1041  _bt_sortaddtup(npage, itupsz, itup, last_off);
1042 
1043  state->btps_page = npage;
1044  state->btps_blkno = nblkno;
1045  state->btps_lastoff = last_off;
1046 }
BlockNumber btpo_next
Definition: nbtree.h:59
BlockNumber btws_pages_alloced
Definition: nbtsort.c:260
Pointer Item
Definition: item.h:17
#define P_NONE
Definition: nbtree.h:182
uint32 BlockNumber
Definition: block.h:31
#define BTreeTupleSetDownLink(itup, blkno)
Definition: nbtree.h:303
BlockNumber btps_blkno
Definition: nbtsort.c:243
#define BTreeTupleGetNAtts(itup, rel)
Definition: nbtree.h:327
static void _bt_blwritepage(BTWriteState *wstate, Page page, BlockNumber blkno)
Definition: nbtsort.c:643
#define BTreeTupleSetNAtts(itup, n)
Definition: nbtree.h:336
BTPageOpaqueData * BTPageOpaque
Definition: nbtree.h:69
Size PageGetFreeSpace(Page page)
Definition: bufpage.c:581
uint16 OffsetNumber
Definition: off.h:24
IndexTuple _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright, BTScanInsert itup_key)
Definition: nbtutils.c:2109
Page btps_page
Definition: nbtsort.c:242
void pfree(void *pointer)
Definition: mcxt.c:1056
#define ItemIdGetLength(itemId)
Definition: itemid.h:59
Size btps_full
Definition: nbtsort.c:247
#define ERROR
Definition: elog.h:43
static void _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
Definition: nbtsort.c:827
bool PageIndexTupleOverwrite(Page page, OffsetNumber offnum, Item newtup, Size newsize)
Definition: bufpage.c:1065
BlockNumber btpo_prev
Definition: nbtree.h:58
IndexTuple CopyIndexTuple(IndexTuple source)
Definition: indextuple.c:509
OffsetNumber btps_lastoff
Definition: nbtsort.c:245
IndexTupleData * IndexTuple
Definition: itup.h:53
#define P_FIRSTKEY
Definition: nbtree.h:219
#define P_LEFTMOST(opaque)
Definition: nbtree.h:188
struct ItemIdData ItemIdData
#define IndexRelationGetNumberOfKeyAttributes(relation)
Definition: rel.h:447
Relation index
Definition: nbtsort.c:257
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
void * palloc0(Size size)
Definition: mcxt.c:980
uint32 btps_level
Definition: nbtsort.c:246
struct IndexTupleData IndexTupleData
static BTPageState * _bt_pagestate(BTWriteState *wstate, uint32 level)
Definition: nbtsort.c:701
struct BTPageState * btps_next
Definition: nbtsort.c:248
BTScanInsert inskey
Definition: nbtsort.c:258
IndexTuple btps_lowkey
Definition: nbtsort.c:244
PageHeaderData * PageHeader
Definition: bufpage.h:166
#define Assert(condition)
Definition: c.h:739
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
size_t Size
Definition: c.h:467
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define OffsetNumberPrev(offsetNumber)
Definition: off.h:54
#define MAXALIGN(LEN)
Definition: c.h:692
void _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace, Page page, IndexTuple newtup)
Definition: nbtutils.c:2523
#define BTMaxItemSize(page)
Definition: nbtree.h:148
#define P_HIKEY
Definition: nbtree.h:218
static Page _bt_blnewpage(uint32 level)
Definition: nbtsort.c:616
#define elog(elevel,...)
Definition: elog.h:228
static void _bt_sortaddtup(Page page, Size itemsize, IndexTuple itup, OffsetNumber itup_off)
Definition: nbtsort.c:769
#define unlikely(x)
Definition: c.h:208
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:99
unsigned short t_info
Definition: itup.h:49
Relation heap
Definition: nbtsort.c:256
#define ItemIdSetUnused(itemId)
Definition: itemid.h:128
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
Pointer Page
Definition: bufpage.h:78
#define IndexTupleSize(itup)
Definition: itup.h:71

◆ _bt_end_parallel()

static void _bt_end_parallel ( BTLeader btleader)
static

Definition at line 1466 of file nbtsort.c.

References DestroyParallelContext(), ExitParallelMode(), IsMVCCSnapshot, BTLeader::pcxt, BTLeader::snapshot, UnregisterSnapshot(), and WaitForParallelWorkersToFinish().

Referenced by _bt_begin_parallel(), and btbuild().

1467 {
1468  /* Shutdown worker processes */
1470  /* Free last reference to MVCC snapshot, if one was used */
1471  if (IsMVCCSnapshot(btleader->snapshot))
1472  UnregisterSnapshot(btleader->snapshot);
1473  DestroyParallelContext(btleader->pcxt);
1474  ExitParallelMode();
1475 }
Snapshot snapshot
Definition: nbtsort.c:205
ParallelContext * pcxt
Definition: nbtsort.c:182
void WaitForParallelWorkersToFinish(ParallelContext *pcxt)
Definition: parallel.c:716
void DestroyParallelContext(ParallelContext *pcxt)
Definition: parallel.c:870
void ExitParallelMode(void)
Definition: xact.c:976
void UnregisterSnapshot(Snapshot snapshot)
Definition: snapmgr.c:907
#define IsMVCCSnapshot(snapshot)
Definition: snapmgr.h:97

◆ _bt_leader_participate_as_worker()

static void _bt_leader_participate_as_worker ( BTBuildState buildstate)
static

Definition at line 1536 of file nbtsort.c.

References _bt_parallel_scan_and_sort(), BTBuildState::btleader, BTLeader::btshared, BTSpool::heap, BTSpool::index, BTSpool::isunique, BTShared::isunique, log_btree_build_stats, maintenance_work_mem, BTLeader::nparticipanttuplesorts, palloc0(), ResetUsage(), BTLeader::sharedsort, BTLeader::sharedsort2, ShowUsage(), and BTBuildState::spool.

Referenced by _bt_begin_parallel().

1537 {
1538  BTLeader *btleader = buildstate->btleader;
1539  BTSpool *leaderworker;
1540  BTSpool *leaderworker2;
1541  int sortmem;
1542 
1543  /* Allocate memory and initialize private spool */
1544  leaderworker = (BTSpool *) palloc0(sizeof(BTSpool));
1545  leaderworker->heap = buildstate->spool->heap;
1546  leaderworker->index = buildstate->spool->index;
1547  leaderworker->isunique = buildstate->spool->isunique;
1548 
1549  /* Initialize second spool, if required */
1550  if (!btleader->btshared->isunique)
1551  leaderworker2 = NULL;
1552  else
1553  {
1554  /* Allocate memory for worker's own private secondary spool */
1555  leaderworker2 = (BTSpool *) palloc0(sizeof(BTSpool));
1556 
1557  /* Initialize worker's own secondary spool */
1558  leaderworker2->heap = leaderworker->heap;
1559  leaderworker2->index = leaderworker->index;
1560  leaderworker2->isunique = false;
1561  }
1562 
1563  /*
1564  * Might as well use reliable figure when doling out maintenance_work_mem
1565  * (when requested number of workers were not launched, this will be
1566  * somewhat higher than it is for other workers).
1567  */
1568  sortmem = maintenance_work_mem / btleader->nparticipanttuplesorts;
1569 
1570  /* Perform work common to all participants */
1571  _bt_parallel_scan_and_sort(leaderworker, leaderworker2, btleader->btshared,
1572  btleader->sharedsort, btleader->sharedsort2,
1573  sortmem, true);
1574 
1575 #ifdef BTREE_BUILD_STATS
1577  {
1578  ShowUsage("BTREE BUILD (Leader Partial Spool) STATISTICS");
1579  ResetUsage();
1580  }
1581 #endif /* BTREE_BUILD_STATS */
1582 }
Sharedsort * sharedsort2
Definition: nbtsort.c:204
BTShared * btshared
Definition: nbtsort.c:202
void ShowUsage(const char *title)
Definition: postgres.c:4594
Sharedsort * sharedsort
Definition: nbtsort.c:203
bool isunique
Definition: nbtsort.c:118
BTSpool * spool
Definition: nbtsort.c:219
Relation heap
Definition: nbtsort.c:99
void ResetUsage(void)
Definition: postgres.c:4587
bool isunique
Definition: nbtsort.c:101
Relation index
Definition: nbtsort.c:100
void * palloc0(Size size)
Definition: mcxt.c:980
bool log_btree_build_stats
Definition: guc.c:504
BTLeader * btleader
Definition: nbtsort.c:233
static void _bt_parallel_scan_and_sort(BTSpool *btspool, BTSpool *btspool2, BTShared *btshared, Sharedsort *sharedsort, Sharedsort *sharedsort2, int sortmem, bool progress)
Definition: nbtsort.c:1691
int maintenance_work_mem
Definition: globals.c:122
int nparticipanttuplesorts
Definition: nbtsort.c:190

◆ _bt_leafbuild()

static void _bt_leafbuild ( BTSpool btspool,
BTSpool btspool2 
)
static

Definition at line 541 of file nbtsort.c.

References _bt_load(), _bt_mkscankey(), BTREE_METAPAGE, BTWriteState::btws_pages_alloced, BTWriteState::btws_pages_written, BTWriteState::btws_use_wal, BTWriteState::btws_zeropage, BTSpool::heap, BTWriteState::heap, BTSpool::index, BTWriteState::index, BTWriteState::inskey, log_btree_build_stats, pgstat_progress_update_param(), PROGRESS_BTREE_PHASE_LEAF_LOAD, PROGRESS_BTREE_PHASE_PERFORMSORT_1, PROGRESS_BTREE_PHASE_PERFORMSORT_2, PROGRESS_CREATEIDX_SUBPHASE, RelationNeedsWAL, ResetUsage(), ShowUsage(), BTSpool::sortstate, tuplesort_performsort(), and XLogIsNeeded.

Referenced by btbuild().

542 {
543  BTWriteState wstate;
544 
545 #ifdef BTREE_BUILD_STATS
547  {
548  ShowUsage("BTREE BUILD (Spool) STATISTICS");
549  ResetUsage();
550  }
551 #endif /* BTREE_BUILD_STATS */
552 
556  if (btspool2)
557  {
560  tuplesort_performsort(btspool2->sortstate);
561  }
562 
563  wstate.heap = btspool->heap;
564  wstate.index = btspool->index;
565  wstate.inskey = _bt_mkscankey(wstate.index, NULL);
566 
567  /*
568  * We need to log index creation in WAL iff WAL archiving/streaming is
569  * enabled UNLESS the index isn't WAL-logged anyway.
570  */
571  wstate.btws_use_wal = XLogIsNeeded() && RelationNeedsWAL(wstate.index);
572 
573  /* reserve the metapage */
574  wstate.btws_pages_alloced = BTREE_METAPAGE + 1;
575  wstate.btws_pages_written = 0;
576  wstate.btws_zeropage = NULL; /* until needed */
577 
580  _bt_load(&wstate, btspool, btspool2);
581 }
void tuplesort_performsort(Tuplesortstate *state)
Definition: tuplesort.c:1791
Page btws_zeropage
Definition: nbtsort.c:262
static void _bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
Definition: nbtsort.c:1123
BTScanInsert _bt_mkscankey(Relation rel, IndexTuple itup)
Definition: nbtutils.c:86
#define PROGRESS_BTREE_PHASE_PERFORMSORT_1
Definition: nbtree.h:706
void ShowUsage(const char *title)
Definition: postgres.c:4594
BlockNumber btws_pages_alloced
Definition: nbtsort.c:260
void pgstat_progress_update_param(int index, int64 val)
Definition: pgstat.c:3213
#define XLogIsNeeded()
Definition: xlog.h:181
Tuplesortstate * sortstate
Definition: nbtsort.c:98
bool btws_use_wal
Definition: nbtsort.c:259
Relation heap
Definition: nbtsort.c:99
void ResetUsage(void)
Definition: postgres.c:4587
#define PROGRESS_BTREE_PHASE_PERFORMSORT_2
Definition: nbtree.h:707
#define BTREE_METAPAGE
Definition: nbtree.h:132
Relation index
Definition: nbtsort.c:100
Relation index
Definition: nbtsort.c:257
BlockNumber btws_pages_written
Definition: nbtsort.c:261
bool log_btree_build_stats
Definition: guc.c:504
BTScanInsert inskey
Definition: nbtsort.c:258
#define RelationNeedsWAL(relation)
Definition: rel.h:530
#define PROGRESS_BTREE_PHASE_LEAF_LOAD
Definition: nbtree.h:708
#define PROGRESS_CREATEIDX_SUBPHASE
Definition: progress.h:83
Relation heap
Definition: nbtsort.c:256

◆ _bt_load()

static void _bt_load ( BTWriteState wstate,
BTSpool btspool,
BTSpool btspool2 
)
static

Definition at line 1123 of file nbtsort.c.

References _bt_buildadd(), _bt_pagestate(), _bt_uppershutdown(), SortSupportData::abbreviate, ApplySortComparator(), Assert, AssertState, BTGreaterStrategyNumber, BTLessStrategyNumber, compare(), CurrentMemoryContext, i, BTWriteState::index, index_getattr, IndexRelationGetNumberOfKeyAttributes, BTWriteState::inskey, ItemPointerCompare(), MAIN_FORKNUM, merge(), palloc0(), pfree(), pgstat_progress_update_param(), PrepareSortSupportFromIndexRel(), PROGRESS_CREATEIDX_TUPLES_DONE, RelationData::rd_smgr, RelationGetDescr, RelationNeedsWAL, RelationOpenSmgr, BTScanInsertData::scankeys, ScanKeyData::sk_attno, SK_BT_DESC, SK_BT_NULLS_FIRST, ScanKeyData::sk_collation, ScanKeyData::sk_flags, smgrimmedsync(), BTSpool::sortstate, SortSupportData::ssup_attno, SortSupportData::ssup_collation, SortSupportData::ssup_cxt, SortSupportData::ssup_nulls_first, IndexTupleData::t_tid, and tuplesort_getindextuple().

Referenced by _bt_leafbuild().

1124 {
1125  BTPageState *state = NULL;
1126  bool merge = (btspool2 != NULL);
1127  IndexTuple itup,
1128  itup2 = NULL;
1129  bool load1;
1130  TupleDesc tupdes = RelationGetDescr(wstate->index);
1131  int i,
1133  SortSupport sortKeys;
1134  int64 tuples_done = 0;
1135 
1136  if (merge)
1137  {
1138  /*
1139  * Another BTSpool for dead tuples exists. Now we have to merge
1140  * btspool and btspool2.
1141  */
1142 
1143  /* the preparation of merge */
1144  itup = tuplesort_getindextuple(btspool->sortstate, true);
1145  itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
1146 
1147  /* Prepare SortSupport data for each column */
1148  sortKeys = (SortSupport) palloc0(keysz * sizeof(SortSupportData));
1149 
1150  for (i = 0; i < keysz; i++)
1151  {
1152  SortSupport sortKey = sortKeys + i;
1153  ScanKey scanKey = wstate->inskey->scankeys + i;
1154  int16 strategy;
1155 
1156  sortKey->ssup_cxt = CurrentMemoryContext;
1157  sortKey->ssup_collation = scanKey->sk_collation;
1158  sortKey->ssup_nulls_first =
1159  (scanKey->sk_flags & SK_BT_NULLS_FIRST) != 0;
1160  sortKey->ssup_attno = scanKey->sk_attno;
1161  /* Abbreviation is not supported here */
1162  sortKey->abbreviate = false;
1163 
1164  AssertState(sortKey->ssup_attno != 0);
1165 
1166  strategy = (scanKey->sk_flags & SK_BT_DESC) != 0 ?
1168 
1169  PrepareSortSupportFromIndexRel(wstate->index, strategy, sortKey);
1170  }
1171 
1172  for (;;)
1173  {
1174  load1 = true; /* load BTSpool next ? */
1175  if (itup2 == NULL)
1176  {
1177  if (itup == NULL)
1178  break;
1179  }
1180  else if (itup != NULL)
1181  {
1182  int32 compare = 0;
1183 
1184  for (i = 1; i <= keysz; i++)
1185  {
1186  SortSupport entry;
1187  Datum attrDatum1,
1188  attrDatum2;
1189  bool isNull1,
1190  isNull2;
1191 
1192  entry = sortKeys + i - 1;
1193  attrDatum1 = index_getattr(itup, i, tupdes, &isNull1);
1194  attrDatum2 = index_getattr(itup2, i, tupdes, &isNull2);
1195 
1196  compare = ApplySortComparator(attrDatum1, isNull1,
1197  attrDatum2, isNull2,
1198  entry);
1199  if (compare > 0)
1200  {
1201  load1 = false;
1202  break;
1203  }
1204  else if (compare < 0)
1205  break;
1206  }
1207 
1208  /*
1209  * If key values are equal, we sort on ItemPointer. This is
1210  * required for btree indexes, since heap TID is treated as an
1211  * implicit last key attribute in order to ensure that all
1212  * keys in the index are physically unique.
1213  */
1214  if (compare == 0)
1215  {
1216  compare = ItemPointerCompare(&itup->t_tid, &itup2->t_tid);
1217  Assert(compare != 0);
1218  if (compare > 0)
1219  load1 = false;
1220  }
1221  }
1222  else
1223  load1 = false;
1224 
1225  /* When we see first tuple, create first index page */
1226  if (state == NULL)
1227  state = _bt_pagestate(wstate, 0);
1228 
1229  if (load1)
1230  {
1231  _bt_buildadd(wstate, state, itup);
1232  itup = tuplesort_getindextuple(btspool->sortstate, true);
1233  }
1234  else
1235  {
1236  _bt_buildadd(wstate, state, itup2);
1237  itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
1238  }
1239 
1240  /* Report progress */
1242  ++tuples_done);
1243  }
1244  pfree(sortKeys);
1245  }
1246  else
1247  {
1248  /* merge is unnecessary */
1249  while ((itup = tuplesort_getindextuple(btspool->sortstate,
1250  true)) != NULL)
1251  {
1252  /* When we see first tuple, create first index page */
1253  if (state == NULL)
1254  state = _bt_pagestate(wstate, 0);
1255 
1256  _bt_buildadd(wstate, state, itup);
1257 
1258  /* Report progress */
1260  ++tuples_done);
1261  }
1262  }
1263 
1264  /* Close down final pages and write the metapage */
1265  _bt_uppershutdown(wstate, state);
1266 
1267  /*
1268  * If the index is WAL-logged, we must fsync it down to disk before it's
1269  * safe to commit the transaction. (For a non-WAL-logged index we don't
1270  * care since the index will be uninteresting after a crash anyway.)
1271  *
1272  * It's obvious that we must do this when not WAL-logging the build. It's
1273  * less obvious that we have to do it even if we did WAL-log the index
1274  * pages. The reason is that since we're building outside shared buffers,
1275  * a CHECKPOINT occurring during the build has no way to flush the
1276  * previously written data to disk (indeed it won't know the index even
1277  * exists). A crash later on would replay WAL from the checkpoint,
1278  * therefore it wouldn't replay our earlier WAL entries. If we do not
1279  * fsync those pages here, they might still not be on disk when the crash
1280  * occurs.
1281  */
1282  if (RelationNeedsWAL(wstate->index))
1283  {
1284  RelationOpenSmgr(wstate->index);
1286  }
1287 }
struct SortSupportData * SortSupport
Definition: sortsupport.h:58
IndexTuple tuplesort_getindextuple(Tuplesortstate *state, bool forward)
Definition: tuplesort.c:2216
int32 ItemPointerCompare(ItemPointer arg1, ItemPointer arg2)
Definition: itemptr.c:52
signed short int16
Definition: c.h:346
bool ssup_nulls_first
Definition: sortsupport.h:75
#define BTGreaterStrategyNumber
Definition: stratnum.h:33
#define AssertState(condition)
Definition: c.h:742
#define RelationGetDescr(relation)
Definition: rel.h:454
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
void pgstat_progress_update_param(int index, int64 val)
Definition: pgstat.c:3213
ItemPointerData t_tid
Definition: itup.h:37
Tuplesortstate * sortstate
Definition: nbtsort.c:98
static pairingheap_node * merge(pairingheap *heap, pairingheap_node *a, pairingheap_node *b)
Definition: pairingheap.c:79
signed int int32
Definition: c.h:347
#define RelationOpenSmgr(relation)
Definition: rel.h:485
void pfree(void *pointer)
Definition: mcxt.c:1056
static int compare(const void *arg1, const void *arg2)
Definition: geqo_pool.c:145
MemoryContext ssup_cxt
Definition: sortsupport.h:66
static void _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
Definition: nbtsort.c:827
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
#define IndexRelationGetNumberOfKeyAttributes(relation)
Definition: rel.h:447
void PrepareSortSupportFromIndexRel(Relation indexRel, int16 strategy, SortSupport ssup)
Definition: sortsupport.c:161
Relation index
Definition: nbtsort.c:257
#define SK_BT_NULLS_FIRST
Definition: nbtree.h:681
void * palloc0(Size size)
Definition: mcxt.c:980
uintptr_t Datum
Definition: postgres.h:367
static BTPageState * _bt_pagestate(BTWriteState *wstate, uint32 level)
Definition: nbtsort.c:701
AttrNumber ssup_attno
Definition: sortsupport.h:81
BTScanInsert inskey
Definition: nbtsort.c:258
int sk_flags
Definition: skey.h:66
#define Assert(condition)
Definition: c.h:739
#define SK_BT_DESC
Definition: nbtree.h:680
Definition: regguts.h:298
#define index_getattr(tup, attnum, tupleDesc, isnull)
Definition: itup.h:100
ScanKeyData scankeys[INDEX_MAX_KEYS]
Definition: nbtree.h:477
#define RelationNeedsWAL(relation)
Definition: rel.h:530
#define PROGRESS_CREATEIDX_TUPLES_DONE
Definition: progress.h:85
Oid sk_collation
Definition: skey.h:70
int i
static void _bt_uppershutdown(BTWriteState *wstate, BTPageState *state)
Definition: nbtsort.c:1052
#define BTLessStrategyNumber
Definition: stratnum.h:29
static int ApplySortComparator(Datum datum1, bool isNull1, Datum datum2, bool isNull2, SortSupport ssup)
Definition: sortsupport.h:200
void smgrimmedsync(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:637
AttrNumber sk_attno
Definition: skey.h:67

◆ _bt_pagestate()

static BTPageState * _bt_pagestate ( BTWriteState wstate,
uint32  level 
)
static

Definition at line 701 of file nbtsort.c.

References _bt_blnewpage(), BTGetTargetPageFreeSpace, BTPageState::btps_blkno, BTPageState::btps_full, BTPageState::btps_lastoff, BTPageState::btps_level, BTPageState::btps_lowkey, BTPageState::btps_next, BTPageState::btps_page, BTREE_NONLEAF_FILLFACTOR, BTWriteState::btws_pages_alloced, BTWriteState::index, P_HIKEY, and palloc0().

Referenced by _bt_buildadd(), and _bt_load().

702 {
704 
705  /* create initial page for level */
706  state->btps_page = _bt_blnewpage(level);
707 
708  /* and assign it a page position */
709  state->btps_blkno = wstate->btws_pages_alloced++;
710 
711  state->btps_lowkey = NULL;
712  /* initialize lastoff so first item goes into P_FIRSTKEY */
713  state->btps_lastoff = P_HIKEY;
714  state->btps_level = level;
715  /* set "full" threshold based on level. See notes at head of file. */
716  if (level > 0)
717  state->btps_full = (BLCKSZ * (100 - BTREE_NONLEAF_FILLFACTOR) / 100);
718  else
719  state->btps_full = BTGetTargetPageFreeSpace(wstate->index);
720 
721  /* no parent level, yet */
722  state->btps_next = NULL;
723 
724  return state;
725 }
#define BTGetTargetPageFreeSpace(relation)
Definition: nbtree.h:697
BlockNumber btws_pages_alloced
Definition: nbtsort.c:260
BlockNumber btps_blkno
Definition: nbtsort.c:243
Page btps_page
Definition: nbtsort.c:242
#define BTREE_NONLEAF_FILLFACTOR
Definition: nbtree.h:171
Size btps_full
Definition: nbtsort.c:247
OffsetNumber btps_lastoff
Definition: nbtsort.c:245
Relation index
Definition: nbtsort.c:257
void * palloc0(Size size)
Definition: mcxt.c:980
uint32 btps_level
Definition: nbtsort.c:246
struct BTPageState * btps_next
Definition: nbtsort.c:248
IndexTuple btps_lowkey
Definition: nbtsort.c:244
Definition: regguts.h:298
#define P_HIKEY
Definition: nbtree.h:218
static Page _bt_blnewpage(uint32 level)
Definition: nbtsort.c:616

◆ _bt_parallel_build_main()

void _bt_parallel_build_main ( dsm_segment seg,
shm_toc toc 
)

Definition at line 1588 of file nbtsort.c.

References _bt_parallel_scan_and_sort(), AccessExclusiveLock, debug_query_string, BTSpool::heap, BTShared::heaprelid, BTSpool::index, index_close(), index_open(), BTShared::indexrelid, BTShared::isconcurrent, BTSpool::isunique, BTShared::isunique, log_btree_build_stats, maintenance_work_mem, palloc0(), PARALLEL_KEY_BTREE_SHARED, PARALLEL_KEY_QUERY_TEXT, PARALLEL_KEY_TUPLESORT, PARALLEL_KEY_TUPLESORT_SPOOL2, pgstat_report_activity(), ResetUsage(), RowExclusiveLock, BTShared::scantuplesortstates, ShareLock, ShareUpdateExclusiveLock, shm_toc_lookup(), ShowUsage(), STATE_RUNNING, table_close(), table_open(), and tuplesort_attach_shared().

1589 {
1590  char *sharedquery;
1591  BTSpool *btspool;
1592  BTSpool *btspool2;
1593  BTShared *btshared;
1594  Sharedsort *sharedsort;
1595  Sharedsort *sharedsort2;
1596  Relation heapRel;
1597  Relation indexRel;
1598  LOCKMODE heapLockmode;
1599  LOCKMODE indexLockmode;
1600  int sortmem;
1601 
1602 #ifdef BTREE_BUILD_STATS
1604  ResetUsage();
1605 #endif /* BTREE_BUILD_STATS */
1606 
1607  /* Set debug_query_string for individual workers first */
1608  sharedquery = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, false);
1609  debug_query_string = sharedquery;
1610 
1611  /* Report the query string from leader */
1613 
1614  /* Look up nbtree shared state */
1615  btshared = shm_toc_lookup(toc, PARALLEL_KEY_BTREE_SHARED, false);
1616 
1617  /* Open relations using lock modes known to be obtained by index.c */
1618  if (!btshared->isconcurrent)
1619  {
1620  heapLockmode = ShareLock;
1621  indexLockmode = AccessExclusiveLock;
1622  }
1623  else
1624  {
1625  heapLockmode = ShareUpdateExclusiveLock;
1626  indexLockmode = RowExclusiveLock;
1627  }
1628 
1629  /* Open relations within worker */
1630  heapRel = table_open(btshared->heaprelid, heapLockmode);
1631  indexRel = index_open(btshared->indexrelid, indexLockmode);
1632 
1633  /* Initialize worker's own spool */
1634  btspool = (BTSpool *) palloc0(sizeof(BTSpool));
1635  btspool->heap = heapRel;
1636  btspool->index = indexRel;
1637  btspool->isunique = btshared->isunique;
1638 
1639  /* Look up shared state private to tuplesort.c */
1640  sharedsort = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT, false);
1641  tuplesort_attach_shared(sharedsort, seg);
1642  if (!btshared->isunique)
1643  {
1644  btspool2 = NULL;
1645  sharedsort2 = NULL;
1646  }
1647  else
1648  {
1649  /* Allocate memory for worker's own private secondary spool */
1650  btspool2 = (BTSpool *) palloc0(sizeof(BTSpool));
1651 
1652  /* Initialize worker's own secondary spool */
1653  btspool2->heap = btspool->heap;
1654  btspool2->index = btspool->index;
1655  btspool2->isunique = false;
1656  /* Look up shared state private to tuplesort.c */
1657  sharedsort2 = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT_SPOOL2, false);
1658  tuplesort_attach_shared(sharedsort2, seg);
1659  }
1660 
1661  /* Perform sorting of spool, and possibly a spool2 */
1662  sortmem = maintenance_work_mem / btshared->scantuplesortstates;
1663  _bt_parallel_scan_and_sort(btspool, btspool2, btshared, sharedsort,
1664  sharedsort2, sortmem, false);
1665 
1666 #ifdef BTREE_BUILD_STATS
1668  {
1669  ShowUsage("BTREE BUILD (Worker Partial Spool) STATISTICS");
1670  ResetUsage();
1671  }
1672 #endif /* BTREE_BUILD_STATS */
1673 
1674  index_close(indexRel, indexLockmode);
1675  table_close(heapRel, heapLockmode);
1676 }
void table_close(Relation relation, LOCKMODE lockmode)
Definition: table.c:133
int scantuplesortstates
Definition: nbtsort.c:120
#define PARALLEL_KEY_TUPLESORT
Definition: nbtsort.c:81
int LOCKMODE
Definition: lockdefs.h:26
Oid indexrelid
Definition: nbtsort.c:117
void pgstat_report_activity(BackendState state, const char *cmd_str)
Definition: pgstat.c:3114
void ShowUsage(const char *title)
Definition: postgres.c:4594
bool isunique
Definition: nbtsort.c:118
bool isconcurrent
Definition: nbtsort.c:119
Relation heap
Definition: nbtsort.c:99
void ResetUsage(void)
Definition: postgres.c:4587
bool isunique
Definition: nbtsort.c:101
Oid heaprelid
Definition: nbtsort.c:116
#define RowExclusiveLock
Definition: lockdefs.h:38
#define PARALLEL_KEY_QUERY_TEXT
Definition: nbtsort.c:83
#define PARALLEL_KEY_TUPLESORT_SPOOL2
Definition: nbtsort.c:82
Relation index
Definition: nbtsort.c:100
const char * debug_query_string
Definition: postgres.c:88
void * palloc0(Size size)
Definition: mcxt.c:980
bool log_btree_build_stats
Definition: guc.c:504
static void _bt_parallel_scan_and_sort(BTSpool *btspool, BTSpool *btspool2, BTShared *btshared, Sharedsort *sharedsort, Sharedsort *sharedsort2, int sortmem, bool progress)
Definition: nbtsort.c:1691
int maintenance_work_mem
Definition: globals.c:122
void tuplesort_attach_shared(Sharedsort *shared, dsm_segment *seg)
Definition: tuplesort.c:4355
#define ShareUpdateExclusiveLock
Definition: lockdefs.h:39
void index_close(Relation relation, LOCKMODE lockmode)
Definition: indexam.c:152
#define AccessExclusiveLock
Definition: lockdefs.h:45
#define ShareLock
Definition: lockdefs.h:41
#define PARALLEL_KEY_BTREE_SHARED
Definition: nbtsort.c:80
Relation table_open(Oid relationId, LOCKMODE lockmode)
Definition: table.c:39
void * shm_toc_lookup(shm_toc *toc, uint64 key, bool noError)
Definition: shm_toc.c:232
Relation index_open(Oid relationId, LOCKMODE lockmode)
Definition: indexam.c:126

◆ _bt_parallel_estimate_shared()

static Size _bt_parallel_estimate_shared ( Relation  heap,
Snapshot  snapshot 
)
static

Definition at line 1482 of file nbtsort.c.

References add_size(), BUFFERALIGN, and table_parallelscan_estimate().

Referenced by _bt_begin_parallel().

1483 {
1484  /* c.f. shm_toc_allocate as to why BUFFERALIGN is used */
1485  return add_size(BUFFERALIGN(sizeof(BTShared)),
1486  table_parallelscan_estimate(heap, snapshot));
1487 }
Size add_size(Size s1, Size s2)
Definition: shmem.c:498
Size table_parallelscan_estimate(Relation rel, Snapshot snapshot)
Definition: tableam.c:126
#define BUFFERALIGN(LEN)
Definition: c.h:694

◆ _bt_parallel_heapscan()

static double _bt_parallel_heapscan ( BTBuildState buildstate,
bool brokenhotchain 
)
static

Definition at line 1502 of file nbtsort.c.

References BTShared::brokenhotchain, BTBuildState::btleader, BTLeader::btshared, ConditionVariableCancelSleep(), ConditionVariableSleep(), BTShared::havedead, BTBuildState::havedead, BTShared::indtuples, BTBuildState::indtuples, BTShared::mutex, BTShared::nparticipantsdone, BTLeader::nparticipanttuplesorts, reltuples, BTShared::reltuples, SpinLockAcquire, SpinLockRelease, WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN, and BTShared::workersdonecv.

Referenced by _bt_spools_heapscan().

1503 {
1504  BTShared *btshared = buildstate->btleader->btshared;
1505  int nparticipanttuplesorts;
1506  double reltuples;
1507 
1508  nparticipanttuplesorts = buildstate->btleader->nparticipanttuplesorts;
1509  for (;;)
1510  {
1511  SpinLockAcquire(&btshared->mutex);
1512  if (btshared->nparticipantsdone == nparticipanttuplesorts)
1513  {
1514  buildstate->havedead = btshared->havedead;
1515  buildstate->indtuples = btshared->indtuples;
1516  *brokenhotchain = btshared->brokenhotchain;
1517  reltuples = btshared->reltuples;
1518  SpinLockRelease(&btshared->mutex);
1519  break;
1520  }
1521  SpinLockRelease(&btshared->mutex);
1522 
1525  }
1526 
1528 
1529  return reltuples;
1530 }
BTShared * btshared
Definition: nbtsort.c:202
int nparticipantsdone
Definition: nbtsort.c:154
bool havedead
Definition: nbtsort.c:217
#define SpinLockAcquire(lock)
Definition: spin.h:62
void ConditionVariableCancelSleep(void)
double indtuples
Definition: nbtsort.c:226
bool brokenhotchain
Definition: nbtsort.c:158
#define SpinLockRelease(lock)
Definition: spin.h:64
BTLeader * btleader
Definition: nbtsort.c:233
double reltuples
Definition: nbtsort.c:155
double indtuples
Definition: nbtsort.c:157
slock_t mutex
Definition: nbtsort.c:136
bool havedead
Definition: nbtsort.c:156
void ConditionVariableSleep(ConditionVariable *cv, uint32 wait_event_info)
int nparticipanttuplesorts
Definition: nbtsort.c:190
ConditionVariable workersdonecv
Definition: nbtsort.c:128
float4 reltuples
Definition: pg_class.h:63

◆ _bt_parallel_scan_and_sort()

static void _bt_parallel_scan_and_sort ( BTSpool btspool,
BTSpool btspool2,
BTShared btshared,
Sharedsort sharedsort,
Sharedsort sharedsort2,
int  sortmem,
bool  progress 
)
static

Definition at line 1691 of file nbtsort.c.

References _bt_build_callback(), BTShared::brokenhotchain, BTBuildState::btleader, BuildIndexInfo(), ConditionVariableSignal(), BTShared::havedead, BTBuildState::havedead, BTSpool::heap, BTBuildState::heap, IndexInfo::ii_BrokenHotChain, IndexInfo::ii_Concurrent, BTSpool::index, BTShared::indtuples, BTBuildState::indtuples, BTShared::isconcurrent, BTSpool::isunique, BTShared::isunique, BTBuildState::isunique, SortCoordinateData::isWorker, Min, BTShared::mutex, SortCoordinateData::nParticipants, BTShared::nparticipantsdone, palloc0(), ParallelTableScanFromBTShared, reltuples, BTShared::reltuples, SortCoordinateData::sharedsort, BTSpool::sortstate, SpinLockAcquire, SpinLockRelease, BTBuildState::spool, BTBuildState::spool2, table_beginscan_parallel(), table_index_build_scan(), tuplesort_begin_index_btree(), tuplesort_end(), tuplesort_performsort(), work_mem, and BTShared::workersdonecv.

Referenced by _bt_leader_participate_as_worker(), and _bt_parallel_build_main().

1694 {
1695  SortCoordinate coordinate;
1696  BTBuildState buildstate;
1697  TableScanDesc scan;
1698  double reltuples;
1699  IndexInfo *indexInfo;
1700 
1701  /* Initialize local tuplesort coordination state */
1702  coordinate = palloc0(sizeof(SortCoordinateData));
1703  coordinate->isWorker = true;
1704  coordinate->nParticipants = -1;
1705  coordinate->sharedsort = sharedsort;
1706 
1707  /* Begin "partial" tuplesort */
1708  btspool->sortstate = tuplesort_begin_index_btree(btspool->heap,
1709  btspool->index,
1710  btspool->isunique,
1711  sortmem, coordinate,
1712  false);
1713 
1714  /*
1715  * Just as with serial case, there may be a second spool. If so, a
1716  * second, dedicated spool2 partial tuplesort is required.
1717  */
1718  if (btspool2)
1719  {
1720  SortCoordinate coordinate2;
1721 
1722  /*
1723  * We expect that the second one (for dead tuples) won't get very
1724  * full, so we give it only work_mem (unless sortmem is less for
1725  * worker). Worker processes are generally permitted to allocate
1726  * work_mem independently.
1727  */
1728  coordinate2 = palloc0(sizeof(SortCoordinateData));
1729  coordinate2->isWorker = true;
1730  coordinate2->nParticipants = -1;
1731  coordinate2->sharedsort = sharedsort2;
1732  btspool2->sortstate =
1733  tuplesort_begin_index_btree(btspool->heap, btspool->index, false,
1734  Min(sortmem, work_mem), coordinate2,
1735  false);
1736  }
1737 
1738  /* Fill in buildstate for _bt_build_callback() */
1739  buildstate.isunique = btshared->isunique;
1740  buildstate.havedead = false;
1741  buildstate.heap = btspool->heap;
1742  buildstate.spool = btspool;
1743  buildstate.spool2 = btspool2;
1744  buildstate.indtuples = 0;
1745  buildstate.btleader = NULL;
1746 
1747  /* Join parallel scan */
1748  indexInfo = BuildIndexInfo(btspool->index);
1749  indexInfo->ii_Concurrent = btshared->isconcurrent;
1750  scan = table_beginscan_parallel(btspool->heap,
1751  ParallelTableScanFromBTShared(btshared));
1752  reltuples = table_index_build_scan(btspool->heap, btspool->index, indexInfo,
1754  (void *) &buildstate, scan);
1755 
1756  /*
1757  * Execute this worker's part of the sort.
1758  *
1759  * Unlike leader and serial cases, we cannot avoid calling
1760  * tuplesort_performsort() for spool2 if it ends up containing no dead
1761  * tuples (this is disallowed for workers by tuplesort).
1762  */
1763  tuplesort_performsort(btspool->sortstate);
1764  if (btspool2)
1765  tuplesort_performsort(btspool2->sortstate);
1766 
1767  /*
1768  * Done. Record ambuild statistics, and whether we encountered a broken
1769  * HOT chain.
1770  */
1771  SpinLockAcquire(&btshared->mutex);
1772  btshared->nparticipantsdone++;
1773  btshared->reltuples += reltuples;
1774  if (buildstate.havedead)
1775  btshared->havedead = true;
1776  btshared->indtuples += buildstate.indtuples;
1777  if (indexInfo->ii_BrokenHotChain)
1778  btshared->brokenhotchain = true;
1779  SpinLockRelease(&btshared->mutex);
1780 
1781  /* Notify leader */
1783 
1784  /* We can end tuplesorts immediately */
1785  tuplesort_end(btspool->sortstate);
1786  if (btspool2)
1787  tuplesort_end(btspool2->sortstate);
1788 }
void tuplesort_performsort(Tuplesortstate *state)
Definition: tuplesort.c:1791
Relation heap
Definition: nbtsort.c:218
int nparticipantsdone
Definition: nbtsort.c:154
#define Min(x, y)
Definition: c.h:911
bool havedead
Definition: nbtsort.c:217
Sharedsort * sharedsort
Definition: tuplesort.h:55
bool isunique
Definition: nbtsort.c:118
#define ParallelTableScanFromBTShared(shared)
Definition: nbtsort.c:173
bool isconcurrent
Definition: nbtsort.c:119
Tuplesortstate * sortstate
Definition: nbtsort.c:98
IndexInfo * BuildIndexInfo(Relation index)
Definition: index.c:2284
BTSpool * spool
Definition: nbtsort.c:219
Relation heap
Definition: nbtsort.c:99
#define SpinLockAcquire(lock)
Definition: spin.h:62
static double table_index_build_scan(Relation table_rel, Relation index_rel, struct IndexInfo *index_info, bool allow_sync, bool progress, IndexBuildCallback callback, void *callback_state, TableScanDesc scan)
Definition: tableam.h:1517
bool isunique
Definition: nbtsort.c:101
void ConditionVariableSignal(ConditionVariable *cv)
BTSpool * spool2
Definition: nbtsort.c:225
double indtuples
Definition: nbtsort.c:226
bool isunique
Definition: nbtsort.c:216
bool ii_BrokenHotChain
Definition: execnodes.h:172
bool brokenhotchain
Definition: nbtsort.c:158
Relation index
Definition: nbtsort.c:100
TableScanDesc table_beginscan_parallel(Relation relation, ParallelTableScanDesc parallel_scan)
Definition: tableam.c:161
int progress
Definition: pgbench.c:234
#define SpinLockRelease(lock)
Definition: spin.h:64
void * palloc0(Size size)
Definition: mcxt.c:980
static void _bt_build_callback(Relation index, ItemPointer tid, Datum *values, bool *isnull, bool tupleIsAlive, void *state)
Definition: nbtsort.c:587
BTLeader * btleader
Definition: nbtsort.c:233
int work_mem
Definition: globals.c:121
double reltuples
Definition: nbtsort.c:155
double indtuples
Definition: nbtsort.c:157
slock_t mutex
Definition: nbtsort.c:136
bool havedead
Definition: nbtsort.c:156
bool ii_Concurrent
Definition: execnodes.h:171
Tuplesortstate * tuplesort_begin_index_btree(Relation heapRel, Relation indexRel, bool enforceUnique, int workMem, SortCoordinate coordinate, bool randomAccess)
Definition: tuplesort.c:975
ConditionVariable workersdonecv
Definition: nbtsort.c:128
void tuplesort_end(Tuplesortstate *state)
Definition: tuplesort.c:1236
float4 reltuples
Definition: pg_class.h:63

◆ _bt_slideleft()

static void _bt_slideleft ( Page  page)
static

Definition at line 734 of file nbtsort.c.

References OffsetNumberNext, P_FIRSTKEY, P_HIKEY, PageGetItemId, PageGetMaxOffsetNumber, and PageIsEmpty.

Referenced by _bt_uppershutdown().

735 {
736  OffsetNumber off;
737  OffsetNumber maxoff;
738  ItemId previi;
739  ItemId thisii;
740 
741  if (!PageIsEmpty(page))
742  {
743  maxoff = PageGetMaxOffsetNumber(page);
744  previi = PageGetItemId(page, P_HIKEY);
745  for (off = P_FIRSTKEY; off <= maxoff; off = OffsetNumberNext(off))
746  {
747  thisii = PageGetItemId(page, off);
748  *previi = *thisii;
749  previi = thisii;
750  }
751  ((PageHeader) page)->pd_lower -= sizeof(ItemIdData);
752  }
753 }
#define PageIsEmpty(page)
Definition: bufpage.h:222
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:357
uint16 OffsetNumber
Definition: off.h:24
#define P_FIRSTKEY
Definition: nbtree.h:219
struct ItemIdData ItemIdData
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
PageHeaderData * PageHeader
Definition: bufpage.h:166
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
#define P_HIKEY
Definition: nbtree.h:218

◆ _bt_sortaddtup()

static void _bt_sortaddtup ( Page  page,
Size  itemsize,
IndexTuple  itup,
OffsetNumber  itup_off 
)
static

Definition at line 769 of file nbtsort.c.

References BTreeTupleSetNAtts, elog, ERROR, InvalidOffsetNumber, P_FIRSTKEY, P_ISLEAF, PageAddItem, PageGetSpecialPointer, and IndexTupleData::t_info.

Referenced by _bt_buildadd().

773 {
775  IndexTupleData trunctuple;
776 
777  if (!P_ISLEAF(opaque) && itup_off == P_FIRSTKEY)
778  {
779  trunctuple = *itup;
780  trunctuple.t_info = sizeof(IndexTupleData);
781  BTreeTupleSetNAtts(&trunctuple, 0);
782  itup = &trunctuple;
783  itemsize = sizeof(IndexTupleData);
784  }
785 
786  if (PageAddItem(page, (Item) itup, itemsize, itup_off,
787  false, false) == InvalidOffsetNumber)
788  elog(ERROR, "failed to add item to the index page");
789 }
Pointer Item
Definition: item.h:17
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:416
#define BTreeTupleSetNAtts(itup, n)
Definition: nbtree.h:336
BTPageOpaqueData * BTPageOpaque
Definition: nbtree.h:69
#define ERROR
Definition: elog.h:43
#define P_FIRSTKEY
Definition: nbtree.h:219
struct IndexTupleData IndexTupleData
#define InvalidOffsetNumber
Definition: off.h:26
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define elog(elevel,...)
Definition: elog.h:228
unsigned short t_info
Definition: itup.h:49
#define P_ISLEAF(opaque)
Definition: nbtree.h:190

◆ _bt_spool()

static void _bt_spool ( BTSpool btspool,
ItemPointer  self,
Datum values,
bool isnull 
)
static

Definition at line 530 of file nbtsort.c.

References BTSpool::index, BTSpool::sortstate, and tuplesort_putindextuplevalues().

Referenced by _bt_build_callback().

531 {
532  tuplesort_putindextuplevalues(btspool->sortstate, btspool->index,
533  self, values, isnull);
534 }
Tuplesortstate * sortstate
Definition: nbtsort.c:98
Relation index
Definition: nbtsort.c:100
void tuplesort_putindextuplevalues(Tuplesortstate *state, Relation rel, ItemPointer self, Datum *values, bool *isnull)
Definition: tuplesort.c:1478
static Datum values[MAXATTR]
Definition: bootstrap.c:167

◆ _bt_spooldestroy()

static void _bt_spooldestroy ( BTSpool btspool)
static

Definition at line 520 of file nbtsort.c.

References pfree(), BTSpool::sortstate, and tuplesort_end().

Referenced by _bt_spools_heapscan(), and btbuild().

521 {
522  tuplesort_end(btspool->sortstate);
523  pfree(btspool);
524 }
Tuplesortstate * sortstate
Definition: nbtsort.c:98
void pfree(void *pointer)
Definition: mcxt.c:1056
void tuplesort_end(Tuplesortstate *state)
Definition: tuplesort.c:1236

◆ _bt_spools_heapscan()

static double _bt_spools_heapscan ( Relation  heap,
Relation  index,
BTBuildState buildstate,
IndexInfo indexInfo 
)
static

Definition at line 370 of file nbtsort.c.

References _bt_begin_parallel(), _bt_build_callback(), _bt_parallel_heapscan(), _bt_spooldestroy(), BTBuildState::btleader, BTSpool::heap, IndexInfo::ii_BrokenHotChain, IndexInfo::ii_Concurrent, IndexInfo::ii_ParallelWorkers, IndexInfo::ii_Unique, BTSpool::index, BTSpool::isunique, BTBuildState::isunique, SortCoordinateData::isWorker, maintenance_work_mem, SortCoordinateData::nParticipants, BTLeader::nparticipanttuplesorts, palloc0(), pgstat_progress_update_multi_param(), pgstat_progress_update_param(), PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN, PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_CREATEIDX_TUPLES_TOTAL, PROGRESS_SCAN_BLOCKS_DONE, PROGRESS_SCAN_BLOCKS_TOTAL, reltuples, SortCoordinateData::sharedsort, BTLeader::sharedsort, BTLeader::sharedsort2, BTSpool::sortstate, BTBuildState::spool, BTBuildState::spool2, table_index_build_scan(), tuplesort_begin_index_btree(), val, and work_mem.

Referenced by btbuild().

372 {
373  BTSpool *btspool = (BTSpool *) palloc0(sizeof(BTSpool));
374  SortCoordinate coordinate = NULL;
375  double reltuples = 0;
376 
377  /*
378  * We size the sort area as maintenance_work_mem rather than work_mem to
379  * speed index creation. This should be OK since a single backend can't
380  * run multiple index creations in parallel (see also: notes on
381  * parallelism and maintenance_work_mem below).
382  */
383  btspool->heap = heap;
384  btspool->index = index;
385  btspool->isunique = indexInfo->ii_Unique;
386 
387  /* Save as primary spool */
388  buildstate->spool = btspool;
389 
390  /* Report table scan phase started */
393 
394  /* Attempt to launch parallel worker scan when required */
395  if (indexInfo->ii_ParallelWorkers > 0)
396  _bt_begin_parallel(buildstate, indexInfo->ii_Concurrent,
397  indexInfo->ii_ParallelWorkers);
398 
399  /*
400  * If parallel build requested and at least one worker process was
401  * successfully launched, set up coordination state
402  */
403  if (buildstate->btleader)
404  {
405  coordinate = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
406  coordinate->isWorker = false;
407  coordinate->nParticipants =
408  buildstate->btleader->nparticipanttuplesorts;
409  coordinate->sharedsort = buildstate->btleader->sharedsort;
410  }
411 
412  /*
413  * Begin serial/leader tuplesort.
414  *
415  * In cases where parallelism is involved, the leader receives the same
416  * share of maintenance_work_mem as a serial sort (it is generally treated
417  * in the same way as a serial sort once we return). Parallel worker
418  * Tuplesortstates will have received only a fraction of
419  * maintenance_work_mem, though.
420  *
421  * We rely on the lifetime of the Leader Tuplesortstate almost not
422  * overlapping with any worker Tuplesortstate's lifetime. There may be
423  * some small overlap, but that's okay because we rely on leader
424  * Tuplesortstate only allocating a small, fixed amount of memory here.
425  * When its tuplesort_performsort() is called (by our caller), and
426  * significant amounts of memory are likely to be used, all workers must
427  * have already freed almost all memory held by their Tuplesortstates
428  * (they are about to go away completely, too). The overall effect is
429  * that maintenance_work_mem always represents an absolute high watermark
430  * on the amount of memory used by a CREATE INDEX operation, regardless of
431  * the use of parallelism or any other factor.
432  */
433  buildstate->spool->sortstate =
434  tuplesort_begin_index_btree(heap, index, buildstate->isunique,
435  maintenance_work_mem, coordinate,
436  false);
437 
438  /*
439  * If building a unique index, put dead tuples in a second spool to keep
440  * them out of the uniqueness check. We expect that the second spool (for
441  * dead tuples) won't get very full, so we give it only work_mem.
442  */
443  if (indexInfo->ii_Unique)
444  {
445  BTSpool *btspool2 = (BTSpool *) palloc0(sizeof(BTSpool));
446  SortCoordinate coordinate2 = NULL;
447 
448  /* Initialize secondary spool */
449  btspool2->heap = heap;
450  btspool2->index = index;
451  btspool2->isunique = false;
452  /* Save as secondary spool */
453  buildstate->spool2 = btspool2;
454 
455  if (buildstate->btleader)
456  {
457  /*
458  * Set up non-private state that is passed to
459  * tuplesort_begin_index_btree() about the basic high level
460  * coordination of a parallel sort.
461  */
462  coordinate2 = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
463  coordinate2->isWorker = false;
464  coordinate2->nParticipants =
465  buildstate->btleader->nparticipanttuplesorts;
466  coordinate2->sharedsort = buildstate->btleader->sharedsort2;
467  }
468 
469  /*
470  * We expect that the second one (for dead tuples) won't get very
471  * full, so we give it only work_mem
472  */
473  buildstate->spool2->sortstate =
474  tuplesort_begin_index_btree(heap, index, false, work_mem,
475  coordinate2, false);
476  }
477 
478  /* Fill spool using either serial or parallel heap scan */
479  if (!buildstate->btleader)
480  reltuples = table_index_build_scan(heap, index, indexInfo, true, true,
481  _bt_build_callback, (void *) buildstate,
482  NULL);
483  else
484  reltuples = _bt_parallel_heapscan(buildstate,
485  &indexInfo->ii_BrokenHotChain);
486 
487  /*
488  * Set the progress target for the next phase. Reset the block number
489  * values set by table_index_build_scan
490  */
491  {
492  const int index[] = {
496  };
497  const int64 val[] = {
498  buildstate->indtuples,
499  0, 0
500  };
501 
502  pgstat_progress_update_multi_param(3, index, val);
503  }
504 
505  /* okay, all heap tuples are spooled */
506  if (buildstate->spool2 && !buildstate->havedead)
507  {
508  /* spool2 turns out to be unnecessary */
509  _bt_spooldestroy(buildstate->spool2);
510  buildstate->spool2 = NULL;
511  }
512 
513  return reltuples;
514 }
static void _bt_begin_parallel(BTBuildState *buildstate, bool isconcurrent, int request)
Definition: nbtsort.c:1306
Sharedsort * sharedsort2
Definition: nbtsort.c:204
#define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN
Definition: nbtree.h:705
void pgstat_progress_update_param(int index, int64 val)
Definition: pgstat.c:3213
Sharedsort * sharedsort
Definition: nbtsort.c:203
bool havedead
Definition: nbtsort.c:217
Sharedsort * sharedsort
Definition: tuplesort.h:55
#define PROGRESS_CREATEIDX_TUPLES_TOTAL
Definition: progress.h:84
Tuplesortstate * sortstate
Definition: nbtsort.c:98
BTSpool * spool
Definition: nbtsort.c:219
Relation heap
Definition: nbtsort.c:99
static double _bt_parallel_heapscan(BTBuildState *buildstate, bool *brokenhotchain)
Definition: nbtsort.c:1502
static double table_index_build_scan(Relation table_rel, Relation index_rel, struct IndexInfo *index_info, bool allow_sync, bool progress, IndexBuildCallback callback, void *callback_state, TableScanDesc scan)
Definition: tableam.h:1517
bool isunique
Definition: nbtsort.c:101
BTSpool * spool2
Definition: nbtsort.c:225
double indtuples
Definition: nbtsort.c:226
bool isunique
Definition: nbtsort.c:216
bool ii_BrokenHotChain
Definition: execnodes.h:172
Relation index
Definition: nbtsort.c:100
void * palloc0(Size size)
Definition: mcxt.c:980
struct SortCoordinateData * SortCoordinate
Definition: tuplesort.h:58
#define PROGRESS_SCAN_BLOCKS_DONE
Definition: progress.h:120
static void _bt_build_callback(Relation index, ItemPointer tid, Datum *values, bool *isnull, bool tupleIsAlive, void *state)
Definition: nbtsort.c:587
BTLeader * btleader
Definition: nbtsort.c:233
int work_mem
Definition: globals.c:121
int maintenance_work_mem
Definition: globals.c:122
bool ii_Unique
Definition: execnodes.h:169
int nparticipanttuplesorts
Definition: nbtsort.c:190
void pgstat_progress_update_multi_param(int nparam, const int *index, const int64 *val)
Definition: pgstat.c:3235
int ii_ParallelWorkers
Definition: execnodes.h:173
#define PROGRESS_SCAN_BLOCKS_TOTAL
Definition: progress.h:119
bool ii_Concurrent
Definition: execnodes.h:171
static void _bt_spooldestroy(BTSpool *btspool)
Definition: nbtsort.c:520
#define PROGRESS_CREATEIDX_SUBPHASE
Definition: progress.h:83
Tuplesortstate * tuplesort_begin_index_btree(Relation heapRel, Relation indexRel, bool enforceUnique, int workMem, SortCoordinate coordinate, bool randomAccess)
Definition: tuplesort.c:975
long val
Definition: informix.c:664
float4 reltuples
Definition: pg_class.h:63

◆ _bt_uppershutdown()

static void _bt_uppershutdown ( BTWriteState wstate,
BTPageState state 
)
static

Definition at line 1052 of file nbtsort.c.

References _bt_blwritepage(), _bt_buildadd(), _bt_initmetapage(), _bt_slideleft(), Assert, BTP_ROOT, BTPageOpaqueData::btpo_flags, BTPageState::btps_blkno, BTPageState::btps_level, BTPageState::btps_lowkey, BTPageState::btps_next, BTPageState::btps_page, BTREE_METAPAGE, BTreeTupleGetNAtts, BTreeTupleSetDownLink, BTWriteState::index, IndexRelationGetNumberOfKeyAttributes, P_LEFTMOST, P_NONE, PageGetSpecialPointer, palloc(), and pfree().

Referenced by _bt_load().

1053 {
1054  BTPageState *s;
1055  BlockNumber rootblkno = P_NONE;
1056  uint32 rootlevel = 0;
1057  Page metapage;
1058 
1059  /*
1060  * Each iteration of this loop completes one more level of the tree.
1061  */
1062  for (s = state; s != NULL; s = s->btps_next)
1063  {
1064  BlockNumber blkno;
1065  BTPageOpaque opaque;
1066 
1067  blkno = s->btps_blkno;
1069 
1070  /*
1071  * We have to link the last page on this level to somewhere.
1072  *
1073  * If we're at the top, it's the root, so attach it to the metapage.
1074  * Otherwise, add an entry for it to its parent using its low key.
1075  * This may cause the last page of the parent level to split, but
1076  * that's not a problem -- we haven't gotten to it yet.
1077  */
1078  if (s->btps_next == NULL)
1079  {
1080  opaque->btpo_flags |= BTP_ROOT;
1081  rootblkno = blkno;
1082  rootlevel = s->btps_level;
1083  }
1084  else
1085  {
1086  Assert((BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) <=
1088  BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) > 0) ||
1089  P_LEFTMOST(opaque));
1090  Assert(BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) == 0 ||
1091  !P_LEFTMOST(opaque));
1093  _bt_buildadd(wstate, s->btps_next, s->btps_lowkey);
1094  pfree(s->btps_lowkey);
1095  s->btps_lowkey = NULL;
1096  }
1097 
1098  /*
1099  * This is the rightmost page, so the ItemId array needs to be slid
1100  * back one slot. Then we can dump out the page.
1101  */
1103  _bt_blwritepage(wstate, s->btps_page, s->btps_blkno);
1104  s->btps_page = NULL; /* writepage freed the workspace */
1105  }
1106 
1107  /*
1108  * As the last step in the process, construct the metapage and make it
1109  * point to the new root (unless we had no data at all, in which case it's
1110  * set to point to "P_NONE"). This changes the index to the "valid" state
1111  * by filling in a valid magic number in the metapage.
1112  */
1113  metapage = (Page) palloc(BLCKSZ);
1114  _bt_initmetapage(metapage, rootblkno, rootlevel);
1115  _bt_blwritepage(wstate, metapage, BTREE_METAPAGE);
1116 }
#define BTP_ROOT
Definition: nbtree.h:73
#define P_NONE
Definition: nbtree.h:182
uint32 BlockNumber
Definition: block.h:31
#define BTreeTupleSetDownLink(itup, blkno)
Definition: nbtree.h:303
BlockNumber btps_blkno
Definition: nbtsort.c:243
#define BTreeTupleGetNAtts(itup, rel)
Definition: nbtree.h:327
static void _bt_blwritepage(BTWriteState *wstate, Page page, BlockNumber blkno)
Definition: nbtsort.c:643
BTPageOpaqueData * BTPageOpaque
Definition: nbtree.h:69
Page btps_page
Definition: nbtsort.c:242
void pfree(void *pointer)
Definition: mcxt.c:1056
static void _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
Definition: nbtsort.c:827
void _bt_initmetapage(Page page, BlockNumber rootbknum, uint32 level)
Definition: nbtpage.c:50
#define P_LEFTMOST(opaque)
Definition: nbtree.h:188
unsigned int uint32
Definition: c.h:359
#define IndexRelationGetNumberOfKeyAttributes(relation)
Definition: rel.h:447
#define BTREE_METAPAGE
Definition: nbtree.h:132
Relation index
Definition: nbtsort.c:257
uint32 btps_level
Definition: nbtsort.c:246
struct BTPageState * btps_next
Definition: nbtsort.c:248
IndexTuple btps_lowkey
Definition: nbtsort.c:244
#define Assert(condition)
Definition: c.h:739
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
static void _bt_slideleft(Page page)
Definition: nbtsort.c:734
void * palloc(Size size)
Definition: mcxt.c:949
uint16 btpo_flags
Definition: nbtree.h:65
Pointer Page
Definition: bufpage.h:78

◆ btbuild()

IndexBuildResult* btbuild ( Relation  heap,
Relation  index,
IndexInfo indexInfo 
)

Definition at line 301 of file nbtsort.c.

References _bt_end_parallel(), _bt_leafbuild(), _bt_spooldestroy(), _bt_spools_heapscan(), BTBuildState::btleader, elog, ERROR, BTBuildState::havedead, BTSpool::heap, BTBuildState::heap, IndexBuildResult::heap_tuples, IndexInfo::ii_Unique, IndexBuildResult::index_tuples, BTBuildState::indtuples, BTBuildState::isunique, log_btree_build_stats, palloc(), RelationGetNumberOfBlocks, RelationGetRelationName, reltuples, ResetUsage(), ShowUsage(), BTBuildState::spool, and BTBuildState::spool2.

Referenced by bthandler().

302 {
303  IndexBuildResult *result;
304  BTBuildState buildstate;
305  double reltuples;
306 
307 #ifdef BTREE_BUILD_STATS
309  ResetUsage();
310 #endif /* BTREE_BUILD_STATS */
311 
312  buildstate.isunique = indexInfo->ii_Unique;
313  buildstate.havedead = false;
314  buildstate.heap = heap;
315  buildstate.spool = NULL;
316  buildstate.spool2 = NULL;
317  buildstate.indtuples = 0;
318  buildstate.btleader = NULL;
319 
320  /*
321  * We expect to be called exactly once for any index relation. If that's
322  * not the case, big trouble's what we have.
323  */
324  if (RelationGetNumberOfBlocks(index) != 0)
325  elog(ERROR, "index \"%s\" already contains data",
326  RelationGetRelationName(index));
327 
328  reltuples = _bt_spools_heapscan(heap, index, &buildstate, indexInfo);
329 
330  /*
331  * Finish the build by (1) completing the sort of the spool file, (2)
332  * inserting the sorted tuples into btree pages and (3) building the upper
333  * levels. Finally, it may also be necessary to end use of parallelism.
334  */
335  _bt_leafbuild(buildstate.spool, buildstate.spool2);
336  _bt_spooldestroy(buildstate.spool);
337  if (buildstate.spool2)
338  _bt_spooldestroy(buildstate.spool2);
339  if (buildstate.btleader)
340  _bt_end_parallel(buildstate.btleader);
341 
342  result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));
343 
344  result->heap_tuples = reltuples;
345  result->index_tuples = buildstate.indtuples;
346 
347 #ifdef BTREE_BUILD_STATS
349  {
350  ShowUsage("BTREE BUILD STATS");
351  ResetUsage();
352  }
353 #endif /* BTREE_BUILD_STATS */
354 
355  return result;
356 }
Relation heap
Definition: nbtsort.c:218
void ShowUsage(const char *title)
Definition: postgres.c:4594
bool havedead
Definition: nbtsort.c:217
static void _bt_end_parallel(BTLeader *btleader)
Definition: nbtsort.c:1466
static void _bt_leafbuild(BTSpool *btspool, BTSpool *btspool2)
Definition: nbtsort.c:541
BTSpool * spool
Definition: nbtsort.c:219
void ResetUsage(void)
Definition: postgres.c:4587
#define ERROR
Definition: elog.h:43
BTSpool * spool2
Definition: nbtsort.c:225
double indtuples
Definition: nbtsort.c:226
bool isunique
Definition: nbtsort.c:216
#define RelationGetRelationName(relation)
Definition: rel.h:462
static double _bt_spools_heapscan(Relation heap, Relation index, BTBuildState *buildstate, IndexInfo *indexInfo)
Definition: nbtsort.c:370
bool log_btree_build_stats
Definition: guc.c:504
BTLeader * btleader
Definition: nbtsort.c:233
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:198
bool ii_Unique
Definition: execnodes.h:169
static void _bt_spooldestroy(BTSpool *btspool)
Definition: nbtsort.c:520
void * palloc(Size size)
Definition: mcxt.c:949
#define elog(elevel,...)
Definition: elog.h:228
float4 reltuples
Definition: pg_class.h:63
double index_tuples
Definition: genam.h:33
double heap_tuples
Definition: genam.h:32