PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
nodeHash.c File Reference
#include "postgres.h"
#include <math.h>
#include <limits.h>
#include "access/htup_details.h"
#include "catalog/pg_statistic.h"
#include "commands/tablespace.h"
#include "executor/execdebug.h"
#include "executor/hashjoin.h"
#include "executor/nodeHash.h"
#include "executor/nodeHashjoin.h"
#include "miscadmin.h"
#include "utils/dynahash.h"
#include "utils/memutils.h"
#include "utils/lsyscache.h"
#include "utils/syscache.h"
Include dependency graph for nodeHash.c:

Go to the source code of this file.

Macros

#define NTUP_PER_BUCKET   1
 

Functions

static void ExecHashIncreaseNumBatches (HashJoinTable hashtable)
 
static void ExecHashIncreaseNumBuckets (HashJoinTable hashtable)
 
static void ExecHashBuildSkewHash (HashJoinTable hashtable, Hash *node, int mcvsToUse)
 
static void ExecHashSkewTableInsert (HashJoinTable hashtable, TupleTableSlot *slot, uint32 hashvalue, int bucketNumber)
 
static void ExecHashRemoveNextSkewBucket (HashJoinTable hashtable)
 
static void * dense_alloc (HashJoinTable hashtable, Size size)
 
TupleTableSlotExecHash (HashState *node)
 
NodeMultiExecHash (HashState *node)
 
HashStateExecInitHash (Hash *node, EState *estate, int eflags)
 
void ExecEndHash (HashState *node)
 
HashJoinTable ExecHashTableCreate (Hash *node, List *hashOperators, bool keepNulls)
 
void ExecChooseHashTableSize (double ntuples, int tupwidth, bool useskew, int *numbuckets, int *numbatches, int *num_skew_mcvs)
 
void ExecHashTableDestroy (HashJoinTable hashtable)
 
void ExecHashTableInsert (HashJoinTable hashtable, TupleTableSlot *slot, uint32 hashvalue)
 
bool ExecHashGetHashValue (HashJoinTable hashtable, ExprContext *econtext, List *hashkeys, bool outer_tuple, bool keep_nulls, uint32 *hashvalue)
 
void ExecHashGetBucketAndBatch (HashJoinTable hashtable, uint32 hashvalue, int *bucketno, int *batchno)
 
bool ExecScanHashBucket (HashJoinState *hjstate, ExprContext *econtext)
 
void ExecPrepHashTableForUnmatched (HashJoinState *hjstate)
 
bool ExecScanHashTableForUnmatched (HashJoinState *hjstate, ExprContext *econtext)
 
void ExecHashTableReset (HashJoinTable hashtable)
 
void ExecHashTableResetMatchFlags (HashJoinTable hashtable)
 
void ExecReScanHash (HashState *node)
 
int ExecHashGetSkewBucket (HashJoinTable hashtable, uint32 hashvalue)
 

Macro Definition Documentation

#define NTUP_PER_BUCKET   1

Definition at line 401 of file nodeHash.c.

Referenced by ExecChooseHashTableSize(), and ExecHashTableInsert().

Function Documentation

static void * dense_alloc ( HashJoinTable  hashtable,
Size  size 
)
static

Definition at line 1642 of file nodeHash.c.

References HashJoinTableData::batchCxt, HashJoinTableData::chunks, HashMemoryChunkData::data, HASH_CHUNK_SIZE, HASH_CHUNK_THRESHOLD, MAXALIGN, HashMemoryChunkData::maxlen, MemoryContextAlloc(), HashMemoryChunkData::next, HashMemoryChunkData::ntuples, NULL, offsetof, and HashMemoryChunkData::used.

Referenced by ExecHashIncreaseNumBatches(), ExecHashRemoveNextSkewBucket(), and ExecHashTableInsert().

1643 {
1644  HashMemoryChunk newChunk;
1645  char *ptr;
1646 
1647  /* just in case the size is not already aligned properly */
1648  size = MAXALIGN(size);
1649 
1650  /*
1651  * If tuple size is larger than of 1/4 of chunk size, allocate a separate
1652  * chunk.
1653  */
1654  if (size > HASH_CHUNK_THRESHOLD)
1655  {
1656  /* allocate new chunk and put it at the beginning of the list */
1657  newChunk = (HashMemoryChunk) MemoryContextAlloc(hashtable->batchCxt,
1658  offsetof(HashMemoryChunkData, data) + size);
1659  newChunk->maxlen = size;
1660  newChunk->used = 0;
1661  newChunk->ntuples = 0;
1662 
1663  /*
1664  * Add this chunk to the list after the first existing chunk, so that
1665  * we don't lose the remaining space in the "current" chunk.
1666  */
1667  if (hashtable->chunks != NULL)
1668  {
1669  newChunk->next = hashtable->chunks->next;
1670  hashtable->chunks->next = newChunk;
1671  }
1672  else
1673  {
1674  newChunk->next = hashtable->chunks;
1675  hashtable->chunks = newChunk;
1676  }
1677 
1678  newChunk->used += size;
1679  newChunk->ntuples += 1;
1680 
1681  return newChunk->data;
1682  }
1683 
1684  /*
1685  * See if we have enough space for it in the current chunk (if any). If
1686  * not, allocate a fresh chunk.
1687  */
1688  if ((hashtable->chunks == NULL) ||
1689  (hashtable->chunks->maxlen - hashtable->chunks->used) < size)
1690  {
1691  /* allocate new chunk and put it at the beginning of the list */
1692  newChunk = (HashMemoryChunk) MemoryContextAlloc(hashtable->batchCxt,
1694 
1695  newChunk->maxlen = HASH_CHUNK_SIZE;
1696  newChunk->used = size;
1697  newChunk->ntuples = 1;
1698 
1699  newChunk->next = hashtable->chunks;
1700  hashtable->chunks = newChunk;
1701 
1702  return newChunk->data;
1703  }
1704 
1705  /* There is enough space in the current chunk, let's add the tuple */
1706  ptr = hashtable->chunks->data + hashtable->chunks->used;
1707  hashtable->chunks->used += size;
1708  hashtable->chunks->ntuples += 1;
1709 
1710  /* return pointer to the start of the tuple memory */
1711  return ptr;
1712 }
#define HASH_CHUNK_SIZE
Definition: hashjoin.h:123
#define HASH_CHUNK_THRESHOLD
Definition: hashjoin.h:124
struct HashMemoryChunkData * next
Definition: hashjoin.h:115
MemoryContext batchCxt
Definition: hashjoin.h:185
struct HashMemoryChunkData * HashMemoryChunk
Definition: hashjoin.h:121
#define NULL
Definition: c.h:226
char data[FLEXIBLE_ARRAY_MEMBER]
Definition: hashjoin.h:118
#define MAXALIGN(LEN)
Definition: c.h:580
HashMemoryChunk chunks
Definition: hashjoin.h:188
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:749
#define offsetof(type, field)
Definition: c.h:547
void ExecChooseHashTableSize ( double  ntuples,
int  tupwidth,
bool  useskew,
int *  numbuckets,
int *  numbatches,
int *  num_skew_mcvs 
)

Definition at line 404 of file nodeHash.c.

References Assert, HJTUPLE_OVERHEAD, Max, MAXALIGN, MaxAllocSize, Min, my_log2(), NTUP_PER_BUCKET, SizeofMinimalTupleHeader, SKEW_BUCKET_OVERHEAD, SKEW_WORK_MEM_PERCENT, and work_mem.

Referenced by ExecHashTableCreate(), and initial_cost_hashjoin().

408 {
409  int tupsize;
410  double inner_rel_bytes;
411  long bucket_bytes;
412  long hash_table_bytes;
413  long skew_table_bytes;
414  long max_pointers;
415  long mppow2;
416  int nbatch = 1;
417  int nbuckets;
418  double dbuckets;
419 
420  /* Force a plausible relation size if no info */
421  if (ntuples <= 0.0)
422  ntuples = 1000.0;
423 
424  /*
425  * Estimate tupsize based on footprint of tuple in hashtable... note this
426  * does not allow for any palloc overhead. The manipulations of spaceUsed
427  * don't count palloc overhead either.
428  */
429  tupsize = HJTUPLE_OVERHEAD +
431  MAXALIGN(tupwidth);
432  inner_rel_bytes = ntuples * tupsize;
433 
434  /*
435  * Target in-memory hashtable size is work_mem kilobytes.
436  */
437  hash_table_bytes = work_mem * 1024L;
438 
439  /*
440  * If skew optimization is possible, estimate the number of skew buckets
441  * that will fit in the memory allowed, and decrement the assumed space
442  * available for the main hash table accordingly.
443  *
444  * We make the optimistic assumption that each skew bucket will contain
445  * one inner-relation tuple. If that turns out to be low, we will recover
446  * at runtime by reducing the number of skew buckets.
447  *
448  * hashtable->skewBucket will have up to 8 times as many HashSkewBucket
449  * pointers as the number of MCVs we allow, since ExecHashBuildSkewHash
450  * will round up to the next power of 2 and then multiply by 4 to reduce
451  * collisions.
452  */
453  if (useskew)
454  {
455  skew_table_bytes = hash_table_bytes * SKEW_WORK_MEM_PERCENT / 100;
456 
457  /*----------
458  * Divisor is:
459  * size of a hash tuple +
460  * worst-case size of skewBucket[] per MCV +
461  * size of skewBucketNums[] entry +
462  * size of skew bucket struct itself
463  *----------
464  */
465  *num_skew_mcvs = skew_table_bytes / (tupsize +
466  (8 * sizeof(HashSkewBucket *)) +
467  sizeof(int) +
469  if (*num_skew_mcvs > 0)
470  hash_table_bytes -= skew_table_bytes;
471  }
472  else
473  *num_skew_mcvs = 0;
474 
475  /*
476  * Set nbuckets to achieve an average bucket load of NTUP_PER_BUCKET when
477  * memory is filled, assuming a single batch; but limit the value so that
478  * the pointer arrays we'll try to allocate do not exceed work_mem nor
479  * MaxAllocSize.
480  *
481  * Note that both nbuckets and nbatch must be powers of 2 to make
482  * ExecHashGetBucketAndBatch fast.
483  */
484  max_pointers = (work_mem * 1024L) / sizeof(HashJoinTuple);
485  max_pointers = Min(max_pointers, MaxAllocSize / sizeof(HashJoinTuple));
486  /* If max_pointers isn't a power of 2, must round it down to one */
487  mppow2 = 1L << my_log2(max_pointers);
488  if (max_pointers != mppow2)
489  max_pointers = mppow2 / 2;
490 
491  /* Also ensure we avoid integer overflow in nbatch and nbuckets */
492  /* (this step is redundant given the current value of MaxAllocSize) */
493  max_pointers = Min(max_pointers, INT_MAX / 2);
494 
495  dbuckets = ceil(ntuples / NTUP_PER_BUCKET);
496  dbuckets = Min(dbuckets, max_pointers);
497  nbuckets = (int) dbuckets;
498  /* don't let nbuckets be really small, though ... */
499  nbuckets = Max(nbuckets, 1024);
500  /* ... and force it to be a power of 2. */
501  nbuckets = 1 << my_log2(nbuckets);
502 
503  /*
504  * If there's not enough space to store the projected number of tuples and
505  * the required bucket headers, we will need multiple batches.
506  */
507  bucket_bytes = sizeof(HashJoinTuple) * nbuckets;
508  if (inner_rel_bytes + bucket_bytes > hash_table_bytes)
509  {
510  /* We'll need multiple batches */
511  long lbuckets;
512  double dbatch;
513  int minbatch;
514  long bucket_size;
515 
516  /*
517  * Estimate the number of buckets we'll want to have when work_mem is
518  * entirely full. Each bucket will contain a bucket pointer plus
519  * NTUP_PER_BUCKET tuples, whose projected size already includes
520  * overhead for the hash code, pointer to the next tuple, etc.
521  */
522  bucket_size = (tupsize * NTUP_PER_BUCKET + sizeof(HashJoinTuple));
523  lbuckets = 1L << my_log2(hash_table_bytes / bucket_size);
524  lbuckets = Min(lbuckets, max_pointers);
525  nbuckets = (int) lbuckets;
526  nbuckets = 1 << my_log2(nbuckets);
527  bucket_bytes = nbuckets * sizeof(HashJoinTuple);
528 
529  /*
530  * Buckets are simple pointers to hashjoin tuples, while tupsize
531  * includes the pointer, hash code, and MinimalTupleData. So buckets
532  * should never really exceed 25% of work_mem (even for
533  * NTUP_PER_BUCKET=1); except maybe for work_mem values that are not
534  * 2^N bytes, where we might get more because of doubling. So let's
535  * look for 50% here.
536  */
537  Assert(bucket_bytes <= hash_table_bytes / 2);
538 
539  /* Calculate required number of batches. */
540  dbatch = ceil(inner_rel_bytes / (hash_table_bytes - bucket_bytes));
541  dbatch = Min(dbatch, max_pointers);
542  minbatch = (int) dbatch;
543  nbatch = 2;
544  while (nbatch < minbatch)
545  nbatch <<= 1;
546  }
547 
548  Assert(nbuckets > 0);
549  Assert(nbatch > 0);
550 
551  *numbuckets = nbuckets;
552  *numbatches = nbatch;
553 }
#define SKEW_BUCKET_OVERHEAD
Definition: hashjoin.h:100
#define Min(x, y)
Definition: c.h:798
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
int my_log2(long num)
Definition: dynahash.c:1682
#define MaxAllocSize
Definition: memutils.h:40
#define SizeofMinimalTupleHeader
Definition: htup_details.h:650
#define NTUP_PER_BUCKET
Definition: nodeHash.c:401
int work_mem
Definition: globals.c:112
#define HJTUPLE_OVERHEAD
Definition: hashjoin.h:71
#define Max(x, y)
Definition: c.h:792
#define Assert(condition)
Definition: c.h:667
#define MAXALIGN(LEN)
Definition: c.h:580
#define SKEW_WORK_MEM_PERCENT
Definition: hashjoin.h:102
void ExecEndHash ( HashState node)

Definition at line 222 of file nodeHash.c.

References ExecEndNode(), ExecFreeExprContext(), outerPlan, outerPlanState, and HashState::ps.

Referenced by ExecEndNode().

223 {
225 
226  /*
227  * free exprcontext
228  */
229  ExecFreeExprContext(&node->ps);
230 
231  /*
232  * shut down the subplan
233  */
234  outerPlan = outerPlanState(node);
235  ExecEndNode(outerPlan);
236 }
void ExecEndNode(PlanState *node)
Definition: execProcnode.c:624
void ExecFreeExprContext(PlanState *planstate)
Definition: execUtils.c:690
#define outerPlanState(node)
Definition: execnodes.h:1096
PlanState ps
Definition: execnodes.h:2023
#define outerPlan(node)
Definition: plannodes.h:159
TupleTableSlot* ExecHash ( HashState node)

Definition at line 60 of file nodeHash.c.

References elog, ERROR, and NULL.

Referenced by ExecProcNode().

61 {
62  elog(ERROR, "Hash node does not support ExecProcNode call convention");
63  return NULL;
64 }
#define ERROR
Definition: elog.h:43
#define NULL
Definition: c.h:226
#define elog
Definition: elog.h:219
static void ExecHashBuildSkewHash ( HashJoinTable  hashtable,
Hash node,
int  mcvsToUse 
)
static

Definition at line 1284 of file nodeHash.c.

References HashJoinTableData::batchCxt, BoolGetDatum, DatumGetUInt32, free_attstatsslot(), FunctionCall1, get_attstatsslot(), HashSkewBucket::hashvalue, HeapTupleIsValid, i, Int16GetDatum, InvalidOid, MemoryContextAlloc(), MemoryContextAllocZero(), HashJoinTableData::nSkewBuckets, NULL, ObjectIdGetDatum, OidIsValid, HashJoinTableData::outer_hashfunctions, ReleaseSysCache(), SearchSysCache3, SKEW_BUCKET_OVERHEAD, SKEW_MIN_OUTER_FRACTION, HashJoinTableData::skewBucket, HashJoinTableData::skewBucketLen, HashJoinTableData::skewBucketNums, Hash::skewColType, Hash::skewColTypmod, Hash::skewColumn, HashJoinTableData::skewEnabled, Hash::skewInherit, Hash::skewTable, HashJoinTableData::spacePeak, HashJoinTableData::spaceUsed, HashJoinTableData::spaceUsedSkew, STATISTIC_KIND_MCV, STATRELATTINH, HashSkewBucket::tuples, and values.

Referenced by ExecHashTableCreate().

1285 {
1286  HeapTupleData *statsTuple;
1287  Datum *values;
1288  int nvalues;
1289  float4 *numbers;
1290  int nnumbers;
1291 
1292  /* Do nothing if planner didn't identify the outer relation's join key */
1293  if (!OidIsValid(node->skewTable))
1294  return;
1295  /* Also, do nothing if we don't have room for at least one skew bucket */
1296  if (mcvsToUse <= 0)
1297  return;
1298 
1299  /*
1300  * Try to find the MCV statistics for the outer relation's join key.
1301  */
1302  statsTuple = SearchSysCache3(STATRELATTINH,
1303  ObjectIdGetDatum(node->skewTable),
1304  Int16GetDatum(node->skewColumn),
1305  BoolGetDatum(node->skewInherit));
1306  if (!HeapTupleIsValid(statsTuple))
1307  return;
1308 
1309  if (get_attstatsslot(statsTuple, node->skewColType, node->skewColTypmod,
1311  NULL,
1312  &values, &nvalues,
1313  &numbers, &nnumbers))
1314  {
1315  double frac;
1316  int nbuckets;
1317  FmgrInfo *hashfunctions;
1318  int i;
1319 
1320  if (mcvsToUse > nvalues)
1321  mcvsToUse = nvalues;
1322 
1323  /*
1324  * Calculate the expected fraction of outer relation that will
1325  * participate in the skew optimization. If this isn't at least
1326  * SKEW_MIN_OUTER_FRACTION, don't use skew optimization.
1327  */
1328  frac = 0;
1329  for (i = 0; i < mcvsToUse; i++)
1330  frac += numbers[i];
1331  if (frac < SKEW_MIN_OUTER_FRACTION)
1332  {
1334  values, nvalues, numbers, nnumbers);
1335  ReleaseSysCache(statsTuple);
1336  return;
1337  }
1338 
1339  /*
1340  * Okay, set up the skew hashtable.
1341  *
1342  * skewBucket[] is an open addressing hashtable with a power of 2 size
1343  * that is greater than the number of MCV values. (This ensures there
1344  * will be at least one null entry, so searches will always
1345  * terminate.)
1346  *
1347  * Note: this code could fail if mcvsToUse exceeds INT_MAX/8 or
1348  * MaxAllocSize/sizeof(void *)/8, but that is not currently possible
1349  * since we limit pg_statistic entries to much less than that.
1350  */
1351  nbuckets = 2;
1352  while (nbuckets <= mcvsToUse)
1353  nbuckets <<= 1;
1354  /* use two more bits just to help avoid collisions */
1355  nbuckets <<= 2;
1356 
1357  hashtable->skewEnabled = true;
1358  hashtable->skewBucketLen = nbuckets;
1359 
1360  /*
1361  * We allocate the bucket memory in the hashtable's batch context. It
1362  * is only needed during the first batch, and this ensures it will be
1363  * automatically removed once the first batch is done.
1364  */
1365  hashtable->skewBucket = (HashSkewBucket **)
1366  MemoryContextAllocZero(hashtable->batchCxt,
1367  nbuckets * sizeof(HashSkewBucket *));
1368  hashtable->skewBucketNums = (int *)
1369  MemoryContextAllocZero(hashtable->batchCxt,
1370  mcvsToUse * sizeof(int));
1371 
1372  hashtable->spaceUsed += nbuckets * sizeof(HashSkewBucket *)
1373  + mcvsToUse * sizeof(int);
1374  hashtable->spaceUsedSkew += nbuckets * sizeof(HashSkewBucket *)
1375  + mcvsToUse * sizeof(int);
1376  if (hashtable->spaceUsed > hashtable->spacePeak)
1377  hashtable->spacePeak = hashtable->spaceUsed;
1378 
1379  /*
1380  * Create a skew bucket for each MCV hash value.
1381  *
1382  * Note: it is very important that we create the buckets in order of
1383  * decreasing MCV frequency. If we have to remove some buckets, they
1384  * must be removed in reverse order of creation (see notes in
1385  * ExecHashRemoveNextSkewBucket) and we want the least common MCVs to
1386  * be removed first.
1387  */
1388  hashfunctions = hashtable->outer_hashfunctions;
1389 
1390  for (i = 0; i < mcvsToUse; i++)
1391  {
1392  uint32 hashvalue;
1393  int bucket;
1394 
1395  hashvalue = DatumGetUInt32(FunctionCall1(&hashfunctions[0],
1396  values[i]));
1397 
1398  /*
1399  * While we have not hit a hole in the hashtable and have not hit
1400  * the desired bucket, we have collided with some previous hash
1401  * value, so try the next bucket location. NB: this code must
1402  * match ExecHashGetSkewBucket.
1403  */
1404  bucket = hashvalue & (nbuckets - 1);
1405  while (hashtable->skewBucket[bucket] != NULL &&
1406  hashtable->skewBucket[bucket]->hashvalue != hashvalue)
1407  bucket = (bucket + 1) & (nbuckets - 1);
1408 
1409  /*
1410  * If we found an existing bucket with the same hashvalue, leave
1411  * it alone. It's okay for two MCVs to share a hashvalue.
1412  */
1413  if (hashtable->skewBucket[bucket] != NULL)
1414  continue;
1415 
1416  /* Okay, create a new skew bucket for this hashvalue. */
1417  hashtable->skewBucket[bucket] = (HashSkewBucket *)
1418  MemoryContextAlloc(hashtable->batchCxt,
1419  sizeof(HashSkewBucket));
1420  hashtable->skewBucket[bucket]->hashvalue = hashvalue;
1421  hashtable->skewBucket[bucket]->tuples = NULL;
1422  hashtable->skewBucketNums[hashtable->nSkewBuckets] = bucket;
1423  hashtable->nSkewBuckets++;
1424  hashtable->spaceUsed += SKEW_BUCKET_OVERHEAD;
1425  hashtable->spaceUsedSkew += SKEW_BUCKET_OVERHEAD;
1426  if (hashtable->spaceUsed > hashtable->spacePeak)
1427  hashtable->spacePeak = hashtable->spaceUsed;
1428  }
1429 
1431  values, nvalues, numbers, nnumbers);
1432  }
1433 
1434  ReleaseSysCache(statsTuple);
1435 }
Oid skewTable
Definition: plannodes.h:800
#define DatumGetUInt32(X)
Definition: postgres.h:494
Definition: fmgr.h:53
#define SKEW_BUCKET_OVERHEAD
Definition: hashjoin.h:100
bool skewInherit
Definition: plannodes.h:802
#define SKEW_MIN_OUTER_FRACTION
Definition: hashjoin.h:103
#define Int16GetDatum(X)
Definition: postgres.h:459
Oid skewColType
Definition: plannodes.h:803
bool get_attstatsslot(HeapTuple statstuple, Oid atttype, int32 atttypmod, int reqkind, Oid reqop, Oid *actualop, Datum **values, int *nvalues, float4 **numbers, int *nnumbers)
Definition: lsyscache.c:2854
#define OidIsValid(objectId)
Definition: c.h:530
int * skewBucketNums
Definition: hashjoin.h:146
#define ObjectIdGetDatum(X)
Definition: postgres.h:515
AttrNumber skewColumn
Definition: plannodes.h:801
HashJoinTuple tuples
Definition: hashjoin.h:97
unsigned int uint32
Definition: c.h:265
#define STATISTIC_KIND_MCV
Definition: pg_statistic.h:204
MemoryContext batchCxt
Definition: hashjoin.h:185
FmgrInfo * outer_hashfunctions
Definition: hashjoin.h:174
float float4
Definition: c.h:376
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
uintptr_t Datum
Definition: postgres.h:374
void ReleaseSysCache(HeapTuple tuple)
Definition: syscache.c:1014
void * MemoryContextAllocZero(MemoryContext context, Size size)
Definition: mcxt.c:784
#define BoolGetDatum(X)
Definition: postgres.h:410
int32 skewColTypmod
Definition: plannodes.h:804
#define InvalidOid
Definition: postgres_ext.h:36
uint32 hashvalue
Definition: hashjoin.h:96
#define HeapTupleIsValid(tuple)
Definition: htup.h:77
#define NULL
Definition: c.h:226
static Datum values[MAXATTR]
Definition: bootstrap.c:162
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:749
#define SearchSysCache3(cacheId, key1, key2, key3)
Definition: syscache.h:147
int i
#define FunctionCall1(flinfo, arg1)
Definition: fmgr.h:573
void free_attstatsslot(Oid atttype, Datum *values, int nvalues, float4 *numbers, int nnumbers)
Definition: lsyscache.c:2978
void ExecHashGetBucketAndBatch ( HashJoinTable  hashtable,
uint32  hashvalue,
int *  bucketno,
int *  batchno 
)

Definition at line 1028 of file nodeHash.c.

References HashJoinTableData::log2_nbuckets, HashJoinTableData::nbatch, and HashJoinTableData::nbuckets.

Referenced by ExecHashIncreaseNumBatches(), ExecHashIncreaseNumBuckets(), ExecHashJoin(), ExecHashRemoveNextSkewBucket(), and ExecHashTableInsert().

1032 {
1033  uint32 nbuckets = (uint32) hashtable->nbuckets;
1034  uint32 nbatch = (uint32) hashtable->nbatch;
1035 
1036  if (nbatch > 1)
1037  {
1038  /* we can do MOD by masking, DIV by shifting */
1039  *bucketno = hashvalue & (nbuckets - 1);
1040  *batchno = (hashvalue >> hashtable->log2_nbuckets) & (nbatch - 1);
1041  }
1042  else
1043  {
1044  *bucketno = hashvalue & (nbuckets - 1);
1045  *batchno = 0;
1046  }
1047 }
unsigned int uint32
Definition: c.h:265
bool ExecHashGetHashValue ( HashJoinTable  hashtable,
ExprContext econtext,
List hashkeys,
bool  outer_tuple,
bool  keep_nulls,
uint32 hashvalue 
)

Definition at line 924 of file nodeHash.c.

References DatumGetUInt32, ExprContext::ecxt_per_tuple_memory, ExecEvalExpr, FunctionCall1, HashJoinTableData::hashStrict, i, HashJoinTableData::inner_hashfunctions, lfirst, MemoryContextSwitchTo(), NULL, HashJoinTableData::outer_hashfunctions, and ResetExprContext.

Referenced by ExecHashJoinOuterGetTuple(), and MultiExecHash().

930 {
931  uint32 hashkey = 0;
932  FmgrInfo *hashfunctions;
933  ListCell *hk;
934  int i = 0;
935  MemoryContext oldContext;
936 
937  /*
938  * We reset the eval context each time to reclaim any memory leaked in the
939  * hashkey expressions.
940  */
941  ResetExprContext(econtext);
942 
943  oldContext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory);
944 
945  if (outer_tuple)
946  hashfunctions = hashtable->outer_hashfunctions;
947  else
948  hashfunctions = hashtable->inner_hashfunctions;
949 
950  foreach(hk, hashkeys)
951  {
952  ExprState *keyexpr = (ExprState *) lfirst(hk);
953  Datum keyval;
954  bool isNull;
955 
956  /* rotate hashkey left 1 bit at each step */
957  hashkey = (hashkey << 1) | ((hashkey & 0x80000000) ? 1 : 0);
958 
959  /*
960  * Get the join attribute value of the tuple
961  */
962  keyval = ExecEvalExpr(keyexpr, econtext, &isNull, NULL);
963 
964  /*
965  * If the attribute is NULL, and the join operator is strict, then
966  * this tuple cannot pass the join qual so we can reject it
967  * immediately (unless we're scanning the outside of an outer join, in
968  * which case we must not reject it). Otherwise we act like the
969  * hashcode of NULL is zero (this will support operators that act like
970  * IS NOT DISTINCT, though not any more-random behavior). We treat
971  * the hash support function as strict even if the operator is not.
972  *
973  * Note: currently, all hashjoinable operators must be strict since
974  * the hash index AM assumes that. However, it takes so little extra
975  * code here to allow non-strict that we may as well do it.
976  */
977  if (isNull)
978  {
979  if (hashtable->hashStrict[i] && !keep_nulls)
980  {
981  MemoryContextSwitchTo(oldContext);
982  return false; /* cannot match */
983  }
984  /* else, leave hashkey unmodified, equivalent to hashcode 0 */
985  }
986  else
987  {
988  /* Compute the hash function */
989  uint32 hkey;
990 
991  hkey = DatumGetUInt32(FunctionCall1(&hashfunctions[i], keyval));
992  hashkey ^= hkey;
993  }
994 
995  i++;
996  }
997 
998  MemoryContextSwitchTo(oldContext);
999 
1000  *hashvalue = hashkey;
1001  return true;
1002 }
#define DatumGetUInt32(X)
Definition: postgres.h:494
Definition: fmgr.h:53
MemoryContext ecxt_per_tuple_memory
Definition: execnodes.h:130
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
FmgrInfo * inner_hashfunctions
Definition: hashjoin.h:175
unsigned int uint32
Definition: c.h:265
FmgrInfo * outer_hashfunctions
Definition: hashjoin.h:174
uintptr_t Datum
Definition: postgres.h:374
#define NULL
Definition: c.h:226
#define lfirst(lc)
Definition: pg_list.h:106
int i
#define FunctionCall1(flinfo, arg1)
Definition: fmgr.h:573
bool * hashStrict
Definition: hashjoin.h:176
#define ResetExprContext(econtext)
Definition: executor.h:334
#define ExecEvalExpr(expr, econtext, isNull, isDone)
Definition: executor.h:73
int ExecHashGetSkewBucket ( HashJoinTable  hashtable,
uint32  hashvalue 
)

Definition at line 1445 of file nodeHash.c.

References HashSkewBucket::hashvalue, INVALID_SKEW_BUCKET_NO, NULL, HashJoinTableData::skewBucket, HashJoinTableData::skewBucketLen, and HashJoinTableData::skewEnabled.

Referenced by ExecHashJoin(), and MultiExecHash().

1446 {
1447  int bucket;
1448 
1449  /*
1450  * Always return INVALID_SKEW_BUCKET_NO if not doing skew optimization (in
1451  * particular, this happens after the initial batch is done).
1452  */
1453  if (!hashtable->skewEnabled)
1454  return INVALID_SKEW_BUCKET_NO;
1455 
1456  /*
1457  * Since skewBucketLen is a power of 2, we can do a modulo by ANDing.
1458  */
1459  bucket = hashvalue & (hashtable->skewBucketLen - 1);
1460 
1461  /*
1462  * While we have not hit a hole in the hashtable and have not hit the
1463  * desired bucket, we have collided with some other hash value, so try the
1464  * next bucket location.
1465  */
1466  while (hashtable->skewBucket[bucket] != NULL &&
1467  hashtable->skewBucket[bucket]->hashvalue != hashvalue)
1468  bucket = (bucket + 1) & (hashtable->skewBucketLen - 1);
1469 
1470  /*
1471  * Found the desired bucket?
1472  */
1473  if (hashtable->skewBucket[bucket] != NULL)
1474  return bucket;
1475 
1476  /*
1477  * There must not be any hashtable entry for this hash value.
1478  */
1479  return INVALID_SKEW_BUCKET_NO;
1480 }
#define INVALID_SKEW_BUCKET_NO
Definition: hashjoin.h:101
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
uint32 hashvalue
Definition: hashjoin.h:96
#define NULL
Definition: c.h:226
static void ExecHashIncreaseNumBatches ( HashJoinTable  hashtable)
static

Definition at line 593 of file nodeHash.c.

References Assert, HashJoinTableData::buckets, HashJoinTableData::chunks, HashJoinTableData::curbatch, HashMemoryChunkData::data, dense_alloc(), ExecHashGetBucketAndBatch(), ExecHashJoinSaveTuple(), HashJoinTableData::growEnabled, HashJoinTableData::hashCxt, HashJoinTupleData::hashvalue, HJTUPLE_MINTUPLE, HJTUPLE_OVERHEAD, idx(), HashJoinTableData::innerBatchFile, HashJoinTableData::log2_nbuckets, HashJoinTableData::log2_nbuckets_optimal, MAXALIGN, MaxAllocSize, MemoryContextSwitchTo(), MemSet, Min, HashJoinTableData::nbatch, HashJoinTableData::nbuckets, HashJoinTableData::nbuckets_optimal, HashJoinTupleData::next, HashMemoryChunkData::next, NULL, HashJoinTableData::outerBatchFile, palloc0(), pfree(), PrepareTempTablespaces(), repalloc(), HashJoinTableData::spaceUsed, and MinimalTupleData::t_len.

Referenced by ExecHashSkewTableInsert(), and ExecHashTableInsert().

594 {
595  int oldnbatch = hashtable->nbatch;
596  int curbatch = hashtable->curbatch;
597  int nbatch;
598  MemoryContext oldcxt;
599  long ninmemory;
600  long nfreed;
601  HashMemoryChunk oldchunks;
602 
603  /* do nothing if we've decided to shut off growth */
604  if (!hashtable->growEnabled)
605  return;
606 
607  /* safety check to avoid overflow */
608  if (oldnbatch > Min(INT_MAX / 2, MaxAllocSize / (sizeof(void *) * 2)))
609  return;
610 
611  nbatch = oldnbatch * 2;
612  Assert(nbatch > 1);
613 
614 #ifdef HJDEBUG
615  printf("Hashjoin %p: increasing nbatch to %d because space = %zu\n",
616  hashtable, nbatch, hashtable->spaceUsed);
617 #endif
618 
619  oldcxt = MemoryContextSwitchTo(hashtable->hashCxt);
620 
621  if (hashtable->innerBatchFile == NULL)
622  {
623  /* we had no file arrays before */
624  hashtable->innerBatchFile = (BufFile **)
625  palloc0(nbatch * sizeof(BufFile *));
626  hashtable->outerBatchFile = (BufFile **)
627  palloc0(nbatch * sizeof(BufFile *));
628  /* time to establish the temp tablespaces, too */
630  }
631  else
632  {
633  /* enlarge arrays and zero out added entries */
634  hashtable->innerBatchFile = (BufFile **)
635  repalloc(hashtable->innerBatchFile, nbatch * sizeof(BufFile *));
636  hashtable->outerBatchFile = (BufFile **)
637  repalloc(hashtable->outerBatchFile, nbatch * sizeof(BufFile *));
638  MemSet(hashtable->innerBatchFile + oldnbatch, 0,
639  (nbatch - oldnbatch) * sizeof(BufFile *));
640  MemSet(hashtable->outerBatchFile + oldnbatch, 0,
641  (nbatch - oldnbatch) * sizeof(BufFile *));
642  }
643 
644  MemoryContextSwitchTo(oldcxt);
645 
646  hashtable->nbatch = nbatch;
647 
648  /*
649  * Scan through the existing hash table entries and dump out any that are
650  * no longer of the current batch.
651  */
652  ninmemory = nfreed = 0;
653 
654  /* If know we need to resize nbuckets, we can do it while rebatching. */
655  if (hashtable->nbuckets_optimal != hashtable->nbuckets)
656  {
657  /* we never decrease the number of buckets */
658  Assert(hashtable->nbuckets_optimal > hashtable->nbuckets);
659 
660  hashtable->nbuckets = hashtable->nbuckets_optimal;
661  hashtable->log2_nbuckets = hashtable->log2_nbuckets_optimal;
662 
663  hashtable->buckets = repalloc(hashtable->buckets,
664  sizeof(HashJoinTuple) * hashtable->nbuckets);
665  }
666 
667  /*
668  * We will scan through the chunks directly, so that we can reset the
669  * buckets now and not have to keep track which tuples in the buckets have
670  * already been processed. We will free the old chunks as we go.
671  */
672  memset(hashtable->buckets, 0, sizeof(HashJoinTuple) * hashtable->nbuckets);
673  oldchunks = hashtable->chunks;
674  hashtable->chunks = NULL;
675 
676  /* so, let's scan through the old chunks, and all tuples in each chunk */
677  while (oldchunks != NULL)
678  {
679  HashMemoryChunk nextchunk = oldchunks->next;
680 
681  /* position within the buffer (up to oldchunks->used) */
682  size_t idx = 0;
683 
684  /* process all tuples stored in this chunk (and then free it) */
685  while (idx < oldchunks->used)
686  {
687  HashJoinTuple hashTuple = (HashJoinTuple) (oldchunks->data + idx);
688  MinimalTuple tuple = HJTUPLE_MINTUPLE(hashTuple);
689  int hashTupleSize = (HJTUPLE_OVERHEAD + tuple->t_len);
690  int bucketno;
691  int batchno;
692 
693  ninmemory++;
694  ExecHashGetBucketAndBatch(hashtable, hashTuple->hashvalue,
695  &bucketno, &batchno);
696 
697  if (batchno == curbatch)
698  {
699  /* keep tuple in memory - copy it into the new chunk */
700  HashJoinTuple copyTuple;
701 
702  copyTuple = (HashJoinTuple) dense_alloc(hashtable, hashTupleSize);
703  memcpy(copyTuple, hashTuple, hashTupleSize);
704 
705  /* and add it back to the appropriate bucket */
706  copyTuple->next = hashtable->buckets[bucketno];
707  hashtable->buckets[bucketno] = copyTuple;
708  }
709  else
710  {
711  /* dump it out */
712  Assert(batchno > curbatch);
714  hashTuple->hashvalue,
715  &hashtable->innerBatchFile[batchno]);
716 
717  hashtable->spaceUsed -= hashTupleSize;
718  nfreed++;
719  }
720 
721  /* next tuple in this chunk */
722  idx += MAXALIGN(hashTupleSize);
723  }
724 
725  /* we're done with this chunk - free it and proceed to the next one */
726  pfree(oldchunks);
727  oldchunks = nextchunk;
728  }
729 
730 #ifdef HJDEBUG
731  printf("Hashjoin %p: freed %ld of %ld tuples, space now %zu\n",
732  hashtable, nfreed, ninmemory, hashtable->spaceUsed);
733 #endif
734 
735  /*
736  * If we dumped out either all or none of the tuples in the table, disable
737  * further expansion of nbatch. This situation implies that we have
738  * enough tuples of identical hashvalues to overflow spaceAllowed.
739  * Increasing nbatch will not fix it since there's no way to subdivide the
740  * group any more finely. We have to just gut it out and hope the server
741  * has enough RAM.
742  */
743  if (nfreed == 0 || nfreed == ninmemory)
744  {
745  hashtable->growEnabled = false;
746 #ifdef HJDEBUG
747  printf("Hashjoin %p: disabling further increase of nbatch\n",
748  hashtable);
749 #endif
750  }
751 }
int log2_nbuckets_optimal
Definition: hashjoin.h:134
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
#define Min(x, y)
Definition: c.h:798
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
#define MemSet(start, val, len)
Definition: c.h:849
Datum idx(PG_FUNCTION_ARGS)
Definition: _int_op.c:264
void ExecHashGetBucketAndBatch(HashJoinTable hashtable, uint32 hashvalue, int *bucketno, int *batchno)
Definition: nodeHash.c:1028
void pfree(void *pointer)
Definition: mcxt.c:992
void PrepareTempTablespaces(void)
Definition: tablespace.c:1291
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
BufFile ** outerBatchFile
Definition: hashjoin.h:167
struct HashMemoryChunkData * next
Definition: hashjoin.h:115
#define MaxAllocSize
Definition: memutils.h:40
void * palloc0(Size size)
Definition: mcxt.c:920
static void * dense_alloc(HashJoinTable hashtable, Size size)
Definition: nodeHash.c:1642
#define HJTUPLE_OVERHEAD
Definition: hashjoin.h:71
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define NULL
Definition: c.h:226
#define Assert(condition)
Definition: c.h:667
char data[FLEXIBLE_ARRAY_MEMBER]
Definition: hashjoin.h:118
BufFile ** innerBatchFile
Definition: hashjoin.h:166
struct HashJoinTupleData * next
Definition: hashjoin.h:66
#define MAXALIGN(LEN)
Definition: c.h:580
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1021
HashMemoryChunk chunks
Definition: hashjoin.h:188
MemoryContext hashCxt
Definition: hashjoin.h:184
void ExecHashJoinSaveTuple(MinimalTuple tuple, uint32 hashvalue, BufFile **fileptr)
Definition: nodeHashjoin.c:871
uint32 hashvalue
Definition: hashjoin.h:67
static void ExecHashIncreaseNumBuckets ( HashJoinTable  hashtable)
static

Definition at line 759 of file nodeHash.c.

References Assert, HashJoinTableData::buckets, HashJoinTableData::chunks, HashMemoryChunkData::data, ExecHashGetBucketAndBatch(), HashJoinTupleData::hashvalue, HJTUPLE_MINTUPLE, HJTUPLE_OVERHEAD, idx(), HashJoinTableData::log2_nbuckets, HashJoinTableData::log2_nbuckets_optimal, MAXALIGN, HashJoinTableData::nbuckets, HashJoinTableData::nbuckets_optimal, HashJoinTupleData::next, HashMemoryChunkData::next, NULL, and repalloc().

Referenced by MultiExecHash().

760 {
761  HashMemoryChunk chunk;
762 
763  /* do nothing if not an increase (it's called increase for a reason) */
764  if (hashtable->nbuckets >= hashtable->nbuckets_optimal)
765  return;
766 
767 #ifdef HJDEBUG
768  printf("Hashjoin %p: increasing nbuckets %d => %d\n",
769  hashtable, hashtable->nbuckets, hashtable->nbuckets_optimal);
770 #endif
771 
772  hashtable->nbuckets = hashtable->nbuckets_optimal;
773  hashtable->log2_nbuckets = hashtable->log2_nbuckets_optimal;
774 
775  Assert(hashtable->nbuckets > 1);
776  Assert(hashtable->nbuckets <= (INT_MAX / 2));
777  Assert(hashtable->nbuckets == (1 << hashtable->log2_nbuckets));
778 
779  /*
780  * Just reallocate the proper number of buckets - we don't need to walk
781  * through them - we can walk the dense-allocated chunks (just like in
782  * ExecHashIncreaseNumBatches, but without all the copying into new
783  * chunks)
784  */
785  hashtable->buckets =
786  (HashJoinTuple *) repalloc(hashtable->buckets,
787  hashtable->nbuckets * sizeof(HashJoinTuple));
788 
789  memset(hashtable->buckets, 0, hashtable->nbuckets * sizeof(HashJoinTuple));
790 
791  /* scan through all tuples in all chunks to rebuild the hash table */
792  for (chunk = hashtable->chunks; chunk != NULL; chunk = chunk->next)
793  {
794  /* process all tuples stored in this chunk */
795  size_t idx = 0;
796 
797  while (idx < chunk->used)
798  {
799  HashJoinTuple hashTuple = (HashJoinTuple) (chunk->data + idx);
800  int bucketno;
801  int batchno;
802 
803  ExecHashGetBucketAndBatch(hashtable, hashTuple->hashvalue,
804  &bucketno, &batchno);
805 
806  /* add the tuple to the proper bucket */
807  hashTuple->next = hashtable->buckets[bucketno];
808  hashtable->buckets[bucketno] = hashTuple;
809 
810  /* advance index past the tuple */
811  idx += MAXALIGN(HJTUPLE_OVERHEAD +
812  HJTUPLE_MINTUPLE(hashTuple)->t_len);
813  }
814  }
815 }
int log2_nbuckets_optimal
Definition: hashjoin.h:134
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
Datum idx(PG_FUNCTION_ARGS)
Definition: _int_op.c:264
void ExecHashGetBucketAndBatch(HashJoinTable hashtable, uint32 hashvalue, int *bucketno, int *batchno)
Definition: nodeHash.c:1028
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
struct HashMemoryChunkData * next
Definition: hashjoin.h:115
#define HJTUPLE_OVERHEAD
Definition: hashjoin.h:71
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define NULL
Definition: c.h:226
#define Assert(condition)
Definition: c.h:667
char data[FLEXIBLE_ARRAY_MEMBER]
Definition: hashjoin.h:118
struct HashJoinTupleData * next
Definition: hashjoin.h:66
#define MAXALIGN(LEN)
Definition: c.h:580
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1021
HashMemoryChunk chunks
Definition: hashjoin.h:188
uint32 hashvalue
Definition: hashjoin.h:67
static void ExecHashRemoveNextSkewBucket ( HashJoinTable  hashtable)
static

Definition at line 1532 of file nodeHash.c.

References Assert, HashJoinTableData::buckets, HashJoinTableData::curbatch, dense_alloc(), ExecHashGetBucketAndBatch(), ExecHashJoinSaveTuple(), HashSkewBucket::hashvalue, HJTUPLE_MINTUPLE, HJTUPLE_OVERHEAD, HashJoinTableData::innerBatchFile, HashJoinTupleData::next, HashJoinTableData::nSkewBuckets, NULL, pfree(), SKEW_BUCKET_OVERHEAD, HashJoinTableData::skewBucket, HashJoinTableData::skewBucketNums, HashJoinTableData::skewEnabled, HashJoinTableData::spaceUsed, HashJoinTableData::spaceUsedSkew, MinimalTupleData::t_len, and HashSkewBucket::tuples.

Referenced by ExecHashSkewTableInsert().

1533 {
1534  int bucketToRemove;
1535  HashSkewBucket *bucket;
1536  uint32 hashvalue;
1537  int bucketno;
1538  int batchno;
1539  HashJoinTuple hashTuple;
1540 
1541  /* Locate the bucket to remove */
1542  bucketToRemove = hashtable->skewBucketNums[hashtable->nSkewBuckets - 1];
1543  bucket = hashtable->skewBucket[bucketToRemove];
1544 
1545  /*
1546  * Calculate which bucket and batch the tuples belong to in the main
1547  * hashtable. They all have the same hash value, so it's the same for all
1548  * of them. Also note that it's not possible for nbatch to increase while
1549  * we are processing the tuples.
1550  */
1551  hashvalue = bucket->hashvalue;
1552  ExecHashGetBucketAndBatch(hashtable, hashvalue, &bucketno, &batchno);
1553 
1554  /* Process all tuples in the bucket */
1555  hashTuple = bucket->tuples;
1556  while (hashTuple != NULL)
1557  {
1558  HashJoinTuple nextHashTuple = hashTuple->next;
1559  MinimalTuple tuple;
1560  Size tupleSize;
1561 
1562  /*
1563  * This code must agree with ExecHashTableInsert. We do not use
1564  * ExecHashTableInsert directly as ExecHashTableInsert expects a
1565  * TupleTableSlot while we already have HashJoinTuples.
1566  */
1567  tuple = HJTUPLE_MINTUPLE(hashTuple);
1568  tupleSize = HJTUPLE_OVERHEAD + tuple->t_len;
1569 
1570  /* Decide whether to put the tuple in the hash table or a temp file */
1571  if (batchno == hashtable->curbatch)
1572  {
1573  /* Move the tuple to the main hash table */
1574  HashJoinTuple copyTuple;
1575 
1576  /*
1577  * We must copy the tuple into the dense storage, else it will not
1578  * be found by, eg, ExecHashIncreaseNumBatches.
1579  */
1580  copyTuple = (HashJoinTuple) dense_alloc(hashtable, tupleSize);
1581  memcpy(copyTuple, hashTuple, tupleSize);
1582  pfree(hashTuple);
1583 
1584  copyTuple->next = hashtable->buckets[bucketno];
1585  hashtable->buckets[bucketno] = copyTuple;
1586 
1587  /* We have reduced skew space, but overall space doesn't change */
1588  hashtable->spaceUsedSkew -= tupleSize;
1589  }
1590  else
1591  {
1592  /* Put the tuple into a temp file for later batches */
1593  Assert(batchno > hashtable->curbatch);
1594  ExecHashJoinSaveTuple(tuple, hashvalue,
1595  &hashtable->innerBatchFile[batchno]);
1596  pfree(hashTuple);
1597  hashtable->spaceUsed -= tupleSize;
1598  hashtable->spaceUsedSkew -= tupleSize;
1599  }
1600 
1601  hashTuple = nextHashTuple;
1602  }
1603 
1604  /*
1605  * Free the bucket struct itself and reset the hashtable entry to NULL.
1606  *
1607  * NOTE: this is not nearly as simple as it looks on the surface, because
1608  * of the possibility of collisions in the hashtable. Suppose that hash
1609  * values A and B collide at a particular hashtable entry, and that A was
1610  * entered first so B gets shifted to a different table entry. If we were
1611  * to remove A first then ExecHashGetSkewBucket would mistakenly start
1612  * reporting that B is not in the hashtable, because it would hit the NULL
1613  * before finding B. However, we always remove entries in the reverse
1614  * order of creation, so this failure cannot happen.
1615  */
1616  hashtable->skewBucket[bucketToRemove] = NULL;
1617  hashtable->nSkewBuckets--;
1618  pfree(bucket);
1619  hashtable->spaceUsed -= SKEW_BUCKET_OVERHEAD;
1620  hashtable->spaceUsedSkew -= SKEW_BUCKET_OVERHEAD;
1621 
1622  /*
1623  * If we have removed all skew buckets then give up on skew optimization.
1624  * Release the arrays since they aren't useful any more.
1625  */
1626  if (hashtable->nSkewBuckets == 0)
1627  {
1628  hashtable->skewEnabled = false;
1629  pfree(hashtable->skewBucket);
1630  pfree(hashtable->skewBucketNums);
1631  hashtable->skewBucket = NULL;
1632  hashtable->skewBucketNums = NULL;
1633  hashtable->spaceUsed -= hashtable->spaceUsedSkew;
1634  hashtable->spaceUsedSkew = 0;
1635  }
1636 }
#define SKEW_BUCKET_OVERHEAD
Definition: hashjoin.h:100
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
int * skewBucketNums
Definition: hashjoin.h:146
void ExecHashGetBucketAndBatch(HashJoinTable hashtable, uint32 hashvalue, int *bucketno, int *batchno)
Definition: nodeHash.c:1028
void pfree(void *pointer)
Definition: mcxt.c:992
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
HashJoinTuple tuples
Definition: hashjoin.h:97
unsigned int uint32
Definition: c.h:265
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
static void * dense_alloc(HashJoinTable hashtable, Size size)
Definition: nodeHash.c:1642
#define HJTUPLE_OVERHEAD
Definition: hashjoin.h:71
uint32 hashvalue
Definition: hashjoin.h:96
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define NULL
Definition: c.h:226
#define Assert(condition)
Definition: c.h:667
size_t Size
Definition: c.h:352
BufFile ** innerBatchFile
Definition: hashjoin.h:166
struct HashJoinTupleData * next
Definition: hashjoin.h:66
void ExecHashJoinSaveTuple(MinimalTuple tuple, uint32 hashvalue, BufFile **fileptr)
Definition: nodeHashjoin.c:871
static void ExecHashSkewTableInsert ( HashJoinTable  hashtable,
TupleTableSlot slot,
uint32  hashvalue,
int  bucketNumber 
)
static

Definition at line 1491 of file nodeHash.c.

References HashJoinTableData::batchCxt, ExecFetchSlotMinimalTuple(), ExecHashIncreaseNumBatches(), ExecHashRemoveNextSkewBucket(), HashJoinTupleData::hashvalue, HeapTupleHeaderClearMatch, HJTUPLE_MINTUPLE, HJTUPLE_OVERHEAD, MemoryContextAlloc(), HashJoinTupleData::next, HashJoinTableData::skewBucket, HashJoinTableData::spaceAllowed, HashJoinTableData::spaceAllowedSkew, HashJoinTableData::spacePeak, HashJoinTableData::spaceUsed, HashJoinTableData::spaceUsedSkew, MinimalTupleData::t_len, and HashSkewBucket::tuples.

Referenced by MultiExecHash().

1495 {
1497  HashJoinTuple hashTuple;
1498  int hashTupleSize;
1499 
1500  /* Create the HashJoinTuple */
1501  hashTupleSize = HJTUPLE_OVERHEAD + tuple->t_len;
1502  hashTuple = (HashJoinTuple) MemoryContextAlloc(hashtable->batchCxt,
1503  hashTupleSize);
1504  hashTuple->hashvalue = hashvalue;
1505  memcpy(HJTUPLE_MINTUPLE(hashTuple), tuple, tuple->t_len);
1507 
1508  /* Push it onto the front of the skew bucket's list */
1509  hashTuple->next = hashtable->skewBucket[bucketNumber]->tuples;
1510  hashtable->skewBucket[bucketNumber]->tuples = hashTuple;
1511 
1512  /* Account for space used, and back off if we've used too much */
1513  hashtable->spaceUsed += hashTupleSize;
1514  hashtable->spaceUsedSkew += hashTupleSize;
1515  if (hashtable->spaceUsed > hashtable->spacePeak)
1516  hashtable->spacePeak = hashtable->spaceUsed;
1517  while (hashtable->spaceUsedSkew > hashtable->spaceAllowedSkew)
1518  ExecHashRemoveNextSkewBucket(hashtable);
1519 
1520  /* Check we are not over the total spaceAllowed, either */
1521  if (hashtable->spaceUsed > hashtable->spaceAllowed)
1522  ExecHashIncreaseNumBatches(hashtable);
1523 }
static void ExecHashRemoveNextSkewBucket(HashJoinTable hashtable)
Definition: nodeHash.c:1532
MinimalTuple ExecFetchSlotMinimalTuple(TupleTableSlot *slot)
Definition: execTuples.c:655
static void ExecHashIncreaseNumBatches(HashJoinTable hashtable)
Definition: nodeHash.c:593
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
Size spaceAllowedSkew
Definition: hashjoin.h:182
HashJoinTuple tuples
Definition: hashjoin.h:97
MemoryContext batchCxt
Definition: hashjoin.h:185
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
#define HJTUPLE_OVERHEAD
Definition: hashjoin.h:71
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
struct HashJoinTupleData * next
Definition: hashjoin.h:66
#define HeapTupleHeaderClearMatch(tup)
Definition: htup_details.h:527
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:749
uint32 hashvalue
Definition: hashjoin.h:67
HashJoinTable ExecHashTableCreate ( Hash node,
List hashOperators,
bool  keepNulls 
)

Definition at line 246 of file nodeHash.c.

References ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate(), Assert, HashJoinTableData::batchCxt, HashJoinTableData::buckets, HashJoinTableData::chunks, HashJoinTableData::curbatch, CurrentMemoryContext, elog, ERROR, ExecChooseHashTableSize(), ExecHashBuildSkewHash(), fmgr_info(), get_op_hash_functions(), HashJoinTableData::growEnabled, HashJoinTableData::hashCxt, HashJoinTableData::hashStrict, i, HashJoinTableData::inner_hashfunctions, HashJoinTableData::innerBatchFile, HashJoinTableData::keepNulls, lfirst_oid, list_length(), HashJoinTableData::log2_nbuckets, HashJoinTableData::log2_nbuckets_optimal, MemoryContextSwitchTo(), my_log2(), HashJoinTableData::nbatch, HashJoinTableData::nbatch_original, HashJoinTableData::nbatch_outstart, HashJoinTableData::nbuckets, HashJoinTableData::nbuckets_optimal, HashJoinTableData::nbuckets_original, HashJoinTableData::nSkewBuckets, NULL, OidIsValid, op_strict(), HashJoinTableData::outer_hashfunctions, HashJoinTableData::outerBatchFile, outerPlan, palloc(), palloc0(), Plan::plan_rows, Plan::plan_width, PrepareTempTablespaces(), SKEW_WORK_MEM_PERCENT, HashJoinTableData::skewBucket, HashJoinTableData::skewBucketLen, HashJoinTableData::skewBucketNums, HashJoinTableData::skewEnabled, Hash::skewTable, HashJoinTableData::skewTuples, HashJoinTableData::spaceAllowed, HashJoinTableData::spaceAllowedSkew, HashJoinTableData::spacePeak, HashJoinTableData::spaceUsed, HashJoinTableData::spaceUsedSkew, HashJoinTableData::totalTuples, and work_mem.

Referenced by ExecHashJoin().

247 {
248  HashJoinTable hashtable;
249  Plan *outerNode;
250  int nbuckets;
251  int nbatch;
252  int num_skew_mcvs;
253  int log2_nbuckets;
254  int nkeys;
255  int i;
256  ListCell *ho;
257  MemoryContext oldcxt;
258 
259  /*
260  * Get information about the size of the relation to be hashed (it's the
261  * "outer" subtree of this node, but the inner relation of the hashjoin).
262  * Compute the appropriate size of the hash table.
263  */
264  outerNode = outerPlan(node);
265 
266  ExecChooseHashTableSize(outerNode->plan_rows, outerNode->plan_width,
267  OidIsValid(node->skewTable),
268  &nbuckets, &nbatch, &num_skew_mcvs);
269 
270  /* nbuckets must be a power of 2 */
271  log2_nbuckets = my_log2(nbuckets);
272  Assert(nbuckets == (1 << log2_nbuckets));
273 
274  /*
275  * Initialize the hash table control block.
276  *
277  * The hashtable control block is just palloc'd from the executor's
278  * per-query memory context.
279  */
280  hashtable = (HashJoinTable) palloc(sizeof(HashJoinTableData));
281  hashtable->nbuckets = nbuckets;
282  hashtable->nbuckets_original = nbuckets;
283  hashtable->nbuckets_optimal = nbuckets;
284  hashtable->log2_nbuckets = log2_nbuckets;
285  hashtable->log2_nbuckets_optimal = log2_nbuckets;
286  hashtable->buckets = NULL;
287  hashtable->keepNulls = keepNulls;
288  hashtable->skewEnabled = false;
289  hashtable->skewBucket = NULL;
290  hashtable->skewBucketLen = 0;
291  hashtable->nSkewBuckets = 0;
292  hashtable->skewBucketNums = NULL;
293  hashtable->nbatch = nbatch;
294  hashtable->curbatch = 0;
295  hashtable->nbatch_original = nbatch;
296  hashtable->nbatch_outstart = nbatch;
297  hashtable->growEnabled = true;
298  hashtable->totalTuples = 0;
299  hashtable->skewTuples = 0;
300  hashtable->innerBatchFile = NULL;
301  hashtable->outerBatchFile = NULL;
302  hashtable->spaceUsed = 0;
303  hashtable->spacePeak = 0;
304  hashtable->spaceAllowed = work_mem * 1024L;
305  hashtable->spaceUsedSkew = 0;
306  hashtable->spaceAllowedSkew =
307  hashtable->spaceAllowed * SKEW_WORK_MEM_PERCENT / 100;
308  hashtable->chunks = NULL;
309 
310 #ifdef HJDEBUG
311  printf("Hashjoin %p: initial nbatch = %d, nbuckets = %d\n",
312  hashtable, nbatch, nbuckets);
313 #endif
314 
315  /*
316  * Get info about the hash functions to be used for each hash key. Also
317  * remember whether the join operators are strict.
318  */
319  nkeys = list_length(hashOperators);
320  hashtable->outer_hashfunctions =
321  (FmgrInfo *) palloc(nkeys * sizeof(FmgrInfo));
322  hashtable->inner_hashfunctions =
323  (FmgrInfo *) palloc(nkeys * sizeof(FmgrInfo));
324  hashtable->hashStrict = (bool *) palloc(nkeys * sizeof(bool));
325  i = 0;
326  foreach(ho, hashOperators)
327  {
328  Oid hashop = lfirst_oid(ho);
329  Oid left_hashfn;
330  Oid right_hashfn;
331 
332  if (!get_op_hash_functions(hashop, &left_hashfn, &right_hashfn))
333  elog(ERROR, "could not find hash function for hash operator %u",
334  hashop);
335  fmgr_info(left_hashfn, &hashtable->outer_hashfunctions[i]);
336  fmgr_info(right_hashfn, &hashtable->inner_hashfunctions[i]);
337  hashtable->hashStrict[i] = op_strict(hashop);
338  i++;
339  }
340 
341  /*
342  * Create temporary memory contexts in which to keep the hashtable working
343  * storage. See notes in executor/hashjoin.h.
344  */
346  "HashTableContext",
348 
349  hashtable->batchCxt = AllocSetContextCreate(hashtable->hashCxt,
350  "HashBatchContext",
352 
353  /* Allocate data that will live for the life of the hashjoin */
354 
355  oldcxt = MemoryContextSwitchTo(hashtable->hashCxt);
356 
357  if (nbatch > 1)
358  {
359  /*
360  * allocate and initialize the file arrays in hashCxt
361  */
362  hashtable->innerBatchFile = (BufFile **)
363  palloc0(nbatch * sizeof(BufFile *));
364  hashtable->outerBatchFile = (BufFile **)
365  palloc0(nbatch * sizeof(BufFile *));
366  /* The files will not be opened until needed... */
367  /* ... but make sure we have temp tablespaces established for them */
369  }
370 
371  /*
372  * Prepare context for the first-scan space allocations; allocate the
373  * hashbucket array therein, and set each bucket "empty".
374  */
375  MemoryContextSwitchTo(hashtable->batchCxt);
376 
377  hashtable->buckets = (HashJoinTuple *)
378  palloc0(nbuckets * sizeof(HashJoinTuple));
379 
380  /*
381  * Set up for skew optimization, if possible and there's a need for more
382  * than one batch. (In a one-batch join, there's no point in it.)
383  */
384  if (nbatch > 1)
385  ExecHashBuildSkewHash(hashtable, node, num_skew_mcvs);
386 
387  MemoryContextSwitchTo(oldcxt);
388 
389  return hashtable;
390 }
int log2_nbuckets_optimal
Definition: hashjoin.h:134
Oid skewTable
Definition: plannodes.h:800
double skewTuples
Definition: hashjoin.h:157
Definition: fmgr.h:53
double plan_rows
Definition: plannodes.h:117
bool op_strict(Oid opno)
Definition: lsyscache.c:1249
bool get_op_hash_functions(Oid opno, RegProcedure *lhs_procno, RegProcedure *rhs_procno)
Definition: lsyscache.c:507
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
FmgrInfo * inner_hashfunctions
Definition: hashjoin.h:175
void ExecChooseHashTableSize(double ntuples, int tupwidth, bool useskew, int *numbuckets, int *numbatches, int *num_skew_mcvs)
Definition: nodeHash.c:404
unsigned int Oid
Definition: postgres_ext.h:31
#define OidIsValid(objectId)
Definition: c.h:530
static void ExecHashBuildSkewHash(HashJoinTable hashtable, Hash *node, int mcvsToUse)
Definition: nodeHash.c:1284
int * skewBucketNums
Definition: hashjoin.h:146
#define ERROR
Definition: elog.h:43
void PrepareTempTablespaces(void)
Definition: tablespace.c:1291
void fmgr_info(Oid functionId, FmgrInfo *finfo)
Definition: fmgr.c:159
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:145
BufFile ** outerBatchFile
Definition: hashjoin.h:167
Size spaceAllowedSkew
Definition: hashjoin.h:182
MemoryContext CurrentMemoryContext
Definition: mcxt.c:37
MemoryContext batchCxt
Definition: hashjoin.h:185
struct HashJoinTableData * HashJoinTable
Definition: execnodes.h:1775
int my_log2(long num)
Definition: dynahash.c:1682
#define outerPlan(node)
Definition: plannodes.h:159
FmgrInfo * outer_hashfunctions
Definition: hashjoin.h:174
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
MemoryContext AllocSetContextCreate(MemoryContext parent, const char *name, Size minContextSize, Size initBlockSize, Size maxBlockSize)
Definition: aset.c:440
void * palloc0(Size size)
Definition: mcxt.c:920
int work_mem
Definition: globals.c:112
double totalTuples
Definition: hashjoin.h:156
int plan_width
Definition: plannodes.h:118
#define NULL
Definition: c.h:226
#define Assert(condition)
Definition: c.h:667
BufFile ** innerBatchFile
Definition: hashjoin.h:166
static int list_length(const List *l)
Definition: pg_list.h:89
HashMemoryChunk chunks
Definition: hashjoin.h:188
void * palloc(Size size)
Definition: mcxt.c:891
int i
bool * hashStrict
Definition: hashjoin.h:176
MemoryContext hashCxt
Definition: hashjoin.h:184
#define elog
Definition: elog.h:219
#define SKEW_WORK_MEM_PERCENT
Definition: hashjoin.h:102
#define lfirst_oid(lc)
Definition: pg_list.h:108
void ExecHashTableDestroy ( HashJoinTable  hashtable)

Definition at line 563 of file nodeHash.c.

References BufFileClose(), HashJoinTableData::hashCxt, i, HashJoinTableData::innerBatchFile, MemoryContextDelete(), HashJoinTableData::nbatch, HashJoinTableData::outerBatchFile, and pfree().

Referenced by ExecEndHashJoin(), and ExecReScanHashJoin().

564 {
565  int i;
566 
567  /*
568  * Make sure all the temp files are closed. We skip batch 0, since it
569  * can't have any temp files (and the arrays might not even exist if
570  * nbatch is only 1).
571  */
572  for (i = 1; i < hashtable->nbatch; i++)
573  {
574  if (hashtable->innerBatchFile[i])
575  BufFileClose(hashtable->innerBatchFile[i]);
576  if (hashtable->outerBatchFile[i])
577  BufFileClose(hashtable->outerBatchFile[i]);
578  }
579 
580  /* Release working memory (batchCxt is a child, so it goes away too) */
581  MemoryContextDelete(hashtable->hashCxt);
582 
583  /* And drop the control block */
584  pfree(hashtable);
585 }
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:200
void BufFileClose(BufFile *file)
Definition: buffile.c:202
void pfree(void *pointer)
Definition: mcxt.c:992
BufFile ** outerBatchFile
Definition: hashjoin.h:167
BufFile ** innerBatchFile
Definition: hashjoin.h:166
int i
MemoryContext hashCxt
Definition: hashjoin.h:184
void ExecHashTableInsert ( HashJoinTable  hashtable,
TupleTableSlot slot,
uint32  hashvalue 
)

Definition at line 830 of file nodeHash.c.

References Assert, HashJoinTableData::buckets, HashJoinTableData::curbatch, dense_alloc(), ExecFetchSlotMinimalTuple(), ExecHashGetBucketAndBatch(), ExecHashIncreaseNumBatches(), ExecHashJoinSaveTuple(), HashJoinTupleData::hashvalue, HeapTupleHeaderClearMatch, HJTUPLE_MINTUPLE, HJTUPLE_OVERHEAD, HashJoinTableData::innerBatchFile, HashJoinTableData::log2_nbuckets_optimal, MaxAllocSize, HashJoinTableData::nbatch, HashJoinTableData::nbuckets_optimal, HashJoinTupleData::next, NTUP_PER_BUCKET, HashJoinTableData::skewTuples, HashJoinTableData::spaceAllowed, HashJoinTableData::spacePeak, HashJoinTableData::spaceUsed, MinimalTupleData::t_len, and HashJoinTableData::totalTuples.

Referenced by ExecHashJoinNewBatch(), and MultiExecHash().

833 {
835  int bucketno;
836  int batchno;
837 
838  ExecHashGetBucketAndBatch(hashtable, hashvalue,
839  &bucketno, &batchno);
840 
841  /*
842  * decide whether to put the tuple in the hash table or a temp file
843  */
844  if (batchno == hashtable->curbatch)
845  {
846  /*
847  * put the tuple in hash table
848  */
849  HashJoinTuple hashTuple;
850  int hashTupleSize;
851  double ntuples = (hashtable->totalTuples - hashtable->skewTuples);
852 
853  /* Create the HashJoinTuple */
854  hashTupleSize = HJTUPLE_OVERHEAD + tuple->t_len;
855  hashTuple = (HashJoinTuple) dense_alloc(hashtable, hashTupleSize);
856 
857  hashTuple->hashvalue = hashvalue;
858  memcpy(HJTUPLE_MINTUPLE(hashTuple), tuple, tuple->t_len);
859 
860  /*
861  * We always reset the tuple-matched flag on insertion. This is okay
862  * even when reloading a tuple from a batch file, since the tuple
863  * could not possibly have been matched to an outer tuple before it
864  * went into the batch file.
865  */
867 
868  /* Push it onto the front of the bucket's list */
869  hashTuple->next = hashtable->buckets[bucketno];
870  hashtable->buckets[bucketno] = hashTuple;
871 
872  /*
873  * Increase the (optimal) number of buckets if we just exceeded the
874  * NTUP_PER_BUCKET threshold, but only when there's still a single
875  * batch.
876  */
877  if (hashtable->nbatch == 1 &&
878  ntuples > (hashtable->nbuckets_optimal * NTUP_PER_BUCKET))
879  {
880  /* Guard against integer overflow and alloc size overflow */
881  if (hashtable->nbuckets_optimal <= INT_MAX / 2 &&
882  hashtable->nbuckets_optimal * 2 <= MaxAllocSize / sizeof(HashJoinTuple))
883  {
884  hashtable->nbuckets_optimal *= 2;
885  hashtable->log2_nbuckets_optimal += 1;
886  }
887  }
888 
889  /* Account for space used, and back off if we've used too much */
890  hashtable->spaceUsed += hashTupleSize;
891  if (hashtable->spaceUsed > hashtable->spacePeak)
892  hashtable->spacePeak = hashtable->spaceUsed;
893  if (hashtable->spaceUsed +
894  hashtable->nbuckets_optimal * sizeof(HashJoinTuple)
895  > hashtable->spaceAllowed)
896  ExecHashIncreaseNumBatches(hashtable);
897  }
898  else
899  {
900  /*
901  * put the tuple into a temp file for later batches
902  */
903  Assert(batchno > hashtable->curbatch);
904  ExecHashJoinSaveTuple(tuple,
905  hashvalue,
906  &hashtable->innerBatchFile[batchno]);
907  }
908 }
int log2_nbuckets_optimal
Definition: hashjoin.h:134
double skewTuples
Definition: hashjoin.h:157
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
MinimalTuple ExecFetchSlotMinimalTuple(TupleTableSlot *slot)
Definition: execTuples.c:655
static void ExecHashIncreaseNumBatches(HashJoinTable hashtable)
Definition: nodeHash.c:593
void ExecHashGetBucketAndBatch(HashJoinTable hashtable, uint32 hashvalue, int *bucketno, int *batchno)
Definition: nodeHash.c:1028
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
#define MaxAllocSize
Definition: memutils.h:40
static void * dense_alloc(HashJoinTable hashtable, Size size)
Definition: nodeHash.c:1642
#define NTUP_PER_BUCKET
Definition: nodeHash.c:401
#define HJTUPLE_OVERHEAD
Definition: hashjoin.h:71
double totalTuples
Definition: hashjoin.h:156
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define Assert(condition)
Definition: c.h:667
BufFile ** innerBatchFile
Definition: hashjoin.h:166
struct HashJoinTupleData * next
Definition: hashjoin.h:66
#define HeapTupleHeaderClearMatch(tup)
Definition: htup_details.h:527
void ExecHashJoinSaveTuple(MinimalTuple tuple, uint32 hashvalue, BufFile **fileptr)
Definition: nodeHashjoin.c:871
uint32 hashvalue
Definition: hashjoin.h:67
void ExecHashTableReset ( HashJoinTable  hashtable)

Definition at line 1210 of file nodeHash.c.

References HashJoinTableData::batchCxt, HashJoinTableData::buckets, HashJoinTableData::chunks, MemoryContextReset(), MemoryContextSwitchTo(), HashJoinTableData::nbuckets, NULL, palloc0(), and HashJoinTableData::spaceUsed.

Referenced by ExecHashJoinNewBatch().

1211 {
1212  MemoryContext oldcxt;
1213  int nbuckets = hashtable->nbuckets;
1214 
1215  /*
1216  * Release all the hash buckets and tuples acquired in the prior pass, and
1217  * reinitialize the context for a new pass.
1218  */
1219  MemoryContextReset(hashtable->batchCxt);
1220  oldcxt = MemoryContextSwitchTo(hashtable->batchCxt);
1221 
1222  /* Reallocate and reinitialize the hash bucket headers. */
1223  hashtable->buckets = (HashJoinTuple *)
1224  palloc0(nbuckets * sizeof(HashJoinTuple));
1225 
1226  hashtable->spaceUsed = 0;
1227 
1228  MemoryContextSwitchTo(oldcxt);
1229 
1230  /* Forget the chunks (the memory was freed by the context reset above). */
1231  hashtable->chunks = NULL;
1232 }
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
void MemoryContextReset(MemoryContext context)
Definition: mcxt.c:135
MemoryContext batchCxt
Definition: hashjoin.h:185
void * palloc0(Size size)
Definition: mcxt.c:920
#define NULL
Definition: c.h:226
HashMemoryChunk chunks
Definition: hashjoin.h:188
void ExecHashTableResetMatchFlags ( HashJoinTable  hashtable)

Definition at line 1239 of file nodeHash.c.

References HashJoinTableData::buckets, HeapTupleHeaderClearMatch, HJTUPLE_MINTUPLE, i, HashJoinTableData::nbuckets, HashJoinTupleData::next, HashJoinTableData::nSkewBuckets, NULL, HashJoinTableData::skewBucket, HashJoinTableData::skewBucketNums, and HashSkewBucket::tuples.

Referenced by ExecReScanHashJoin().

1240 {
1241  HashJoinTuple tuple;
1242  int i;
1243 
1244  /* Reset all flags in the main table ... */
1245  for (i = 0; i < hashtable->nbuckets; i++)
1246  {
1247  for (tuple = hashtable->buckets[i]; tuple != NULL; tuple = tuple->next)
1249  }
1250 
1251  /* ... and the same for the skew buckets, if any */
1252  for (i = 0; i < hashtable->nSkewBuckets; i++)
1253  {
1254  int j = hashtable->skewBucketNums[i];
1255  HashSkewBucket *skewBucket = hashtable->skewBucket[j];
1256 
1257  for (tuple = skewBucket->tuples; tuple != NULL; tuple = tuple->next)
1259  }
1260 }
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
int * skewBucketNums
Definition: hashjoin.h:146
HashJoinTuple tuples
Definition: hashjoin.h:97
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define NULL
Definition: c.h:226
struct HashJoinTupleData * next
Definition: hashjoin.h:66
#define HeapTupleHeaderClearMatch(tup)
Definition: htup_details.h:527
int i
HashState* ExecInitHash ( Hash node,
EState estate,
int  eflags 
)

Definition at line 162 of file nodeHash.c.

References Assert, EXEC_FLAG_BACKWARD, EXEC_FLAG_MARK, ExecAssignExprContext(), ExecAssignResultTypeFromTL(), ExecInitExpr(), ExecInitNode(), ExecInitResultTupleSlot(), HashState::hashkeys, HashState::hashtable, makeNode, NIL, NULL, outerPlan, outerPlanState, Hash::plan, PlanState::plan, HashState::ps, PlanState::ps_ProjInfo, Plan::qual, PlanState::qual, PlanState::state, Plan::targetlist, and PlanState::targetlist.

Referenced by ExecInitNode().

163 {
164  HashState *hashstate;
165 
166  /* check for unsupported flags */
167  Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
168 
169  /*
170  * create state structure
171  */
172  hashstate = makeNode(HashState);
173  hashstate->ps.plan = (Plan *) node;
174  hashstate->ps.state = estate;
175  hashstate->hashtable = NULL;
176  hashstate->hashkeys = NIL; /* will be set by parent HashJoin */
177 
178  /*
179  * Miscellaneous initialization
180  *
181  * create expression context for node
182  */
183  ExecAssignExprContext(estate, &hashstate->ps);
184 
185  /*
186  * initialize our result slot
187  */
188  ExecInitResultTupleSlot(estate, &hashstate->ps);
189 
190  /*
191  * initialize child expressions
192  */
193  hashstate->ps.targetlist = (List *)
194  ExecInitExpr((Expr *) node->plan.targetlist,
195  (PlanState *) hashstate);
196  hashstate->ps.qual = (List *)
197  ExecInitExpr((Expr *) node->plan.qual,
198  (PlanState *) hashstate);
199 
200  /*
201  * initialize child nodes
202  */
203  outerPlanState(hashstate) = ExecInitNode(outerPlan(node), estate, eflags);
204 
205  /*
206  * initialize tuple type. no need to initialize projection info because
207  * this node doesn't do projections
208  */
209  ExecAssignResultTypeFromTL(&hashstate->ps);
210  hashstate->ps.ps_ProjInfo = NULL;
211 
212  return hashstate;
213 }
#define NIL
Definition: pg_list.h:69
List * qual
Definition: plannodes.h:130
ProjectionInfo * ps_ProjInfo
Definition: execnodes.h:1083
HashJoinTable hashtable
Definition: execnodes.h:2024
List * qual
Definition: execnodes.h:1066
List * targetlist
Definition: execnodes.h:1065
EState * state
Definition: execnodes.h:1053
void ExecAssignResultTypeFromTL(PlanState *planstate)
Definition: execUtils.c:429
ExprState * ExecInitExpr(Expr *node, PlanState *parent)
Definition: execQual.c:4562
void ExecInitResultTupleSlot(EState *estate, PlanState *planstate)
Definition: execTuples.c:835
#define EXEC_FLAG_BACKWARD
Definition: executor.h:60
#define outerPlanState(node)
Definition: execnodes.h:1096
List * hashkeys
Definition: execnodes.h:2025
PlanState ps
Definition: execnodes.h:2023
#define outerPlan(node)
Definition: plannodes.h:159
Plan * plan
Definition: execnodes.h:1051
#define makeNode(_type_)
Definition: nodes.h:551
#define NULL
Definition: c.h:226
#define Assert(condition)
Definition: c.h:667
#define EXEC_FLAG_MARK
Definition: executor.h:61
void ExecAssignExprContext(EState *estate, PlanState *planstate)
Definition: execUtils.c:407
List * targetlist
Definition: plannodes.h:129
Plan plan
Definition: plannodes.h:799
PlanState * ExecInitNode(Plan *node, EState *estate, int eflags)
Definition: execProcnode.c:137
Definition: pg_list.h:45
void ExecPrepHashTableForUnmatched ( HashJoinState hjstate)

Definition at line 1118 of file nodeHash.c.

References HashJoinState::hj_CurBucketNo, HashJoinState::hj_CurSkewBucketNo, HashJoinState::hj_CurTuple, and NULL.

Referenced by ExecHashJoin().

1119 {
1120  /*----------
1121  * During this scan we use the HashJoinState fields as follows:
1122  *
1123  * hj_CurBucketNo: next regular bucket to scan
1124  * hj_CurSkewBucketNo: next skew bucket (an index into skewBucketNums)
1125  * hj_CurTuple: last tuple returned, or NULL to start next bucket
1126  *----------
1127  */
1128  hjstate->hj_CurBucketNo = 0;
1129  hjstate->hj_CurSkewBucketNo = 0;
1130  hjstate->hj_CurTuple = NULL;
1131 }
int hj_CurSkewBucketNo
Definition: execnodes.h:1787
HashJoinTuple hj_CurTuple
Definition: execnodes.h:1788
int hj_CurBucketNo
Definition: execnodes.h:1786
#define NULL
Definition: c.h:226
void ExecReScanHash ( HashState node)

Definition at line 1264 of file nodeHash.c.

References PlanState::chgParam, ExecReScan(), PlanState::lefttree, NULL, and HashState::ps.

Referenced by ExecReScan().

1265 {
1266  /*
1267  * if chgParam of subnode is not null then plan will be re-scanned by
1268  * first ExecProcNode.
1269  */
1270  if (node->ps.lefttree->chgParam == NULL)
1271  ExecReScan(node->ps.lefttree);
1272 }
void ExecReScan(PlanState *node)
Definition: execAmi.c:74
struct PlanState * lefttree
Definition: execnodes.h:1067
PlanState ps
Definition: execnodes.h:2023
Bitmapset * chgParam
Definition: execnodes.h:1076
#define NULL
Definition: c.h:226
bool ExecScanHashBucket ( HashJoinState hjstate,
ExprContext econtext 
)

Definition at line 1060 of file nodeHash.c.

References HashJoinTableData::buckets, ExprContext::ecxt_innertuple, ExecQual(), ExecStoreMinimalTuple(), HashJoinState::hashclauses, HashJoinTupleData::hashvalue, HashJoinState::hj_CurBucketNo, HashJoinState::hj_CurHashValue, HashJoinState::hj_CurSkewBucketNo, HashJoinState::hj_CurTuple, HashJoinState::hj_HashTable, HashJoinState::hj_HashTupleSlot, HJTUPLE_MINTUPLE, INVALID_SKEW_BUCKET_NO, HashJoinTupleData::next, NULL, ResetExprContext, HashJoinTableData::skewBucket, and HashSkewBucket::tuples.

Referenced by ExecHashJoin().

1062 {
1063  List *hjclauses = hjstate->hashclauses;
1064  HashJoinTable hashtable = hjstate->hj_HashTable;
1065  HashJoinTuple hashTuple = hjstate->hj_CurTuple;
1066  uint32 hashvalue = hjstate->hj_CurHashValue;
1067 
1068  /*
1069  * hj_CurTuple is the address of the tuple last returned from the current
1070  * bucket, or NULL if it's time to start scanning a new bucket.
1071  *
1072  * If the tuple hashed to a skew bucket then scan the skew bucket
1073  * otherwise scan the standard hashtable bucket.
1074  */
1075  if (hashTuple != NULL)
1076  hashTuple = hashTuple->next;
1077  else if (hjstate->hj_CurSkewBucketNo != INVALID_SKEW_BUCKET_NO)
1078  hashTuple = hashtable->skewBucket[hjstate->hj_CurSkewBucketNo]->tuples;
1079  else
1080  hashTuple = hashtable->buckets[hjstate->hj_CurBucketNo];
1081 
1082  while (hashTuple != NULL)
1083  {
1084  if (hashTuple->hashvalue == hashvalue)
1085  {
1086  TupleTableSlot *inntuple;
1087 
1088  /* insert hashtable's tuple into exec slot so ExecQual sees it */
1089  inntuple = ExecStoreMinimalTuple(HJTUPLE_MINTUPLE(hashTuple),
1090  hjstate->hj_HashTupleSlot,
1091  false); /* do not pfree */
1092  econtext->ecxt_innertuple = inntuple;
1093 
1094  /* reset temp memory each time to avoid leaks from qual expr */
1095  ResetExprContext(econtext);
1096 
1097  if (ExecQual(hjclauses, econtext, false))
1098  {
1099  hjstate->hj_CurTuple = hashTuple;
1100  return true;
1101  }
1102  }
1103 
1104  hashTuple = hashTuple->next;
1105  }
1106 
1107  /*
1108  * no match
1109  */
1110  return false;
1111 }
#define INVALID_SKEW_BUCKET_NO
Definition: hashjoin.h:101
TupleTableSlot * ExecStoreMinimalTuple(MinimalTuple mtup, TupleTableSlot *slot, bool shouldFree)
Definition: execTuples.c:387
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
uint32 hj_CurHashValue
Definition: execnodes.h:1785
int hj_CurSkewBucketNo
Definition: execnodes.h:1787
HashJoinTuple hj_CurTuple
Definition: execnodes.h:1788
HashJoinTuple tuples
Definition: hashjoin.h:97
TupleTableSlot * ecxt_innertuple
Definition: execnodes.h:125
bool ExecQual(List *qual, ExprContext *econtext, bool resultForNull)
Definition: execQual.c:5352
unsigned int uint32
Definition: c.h:265
int hj_CurBucketNo
Definition: execnodes.h:1786
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define NULL
Definition: c.h:226
struct HashJoinTupleData * next
Definition: hashjoin.h:66
List * hashclauses
Definition: execnodes.h:1780
TupleTableSlot * hj_HashTupleSlot
Definition: execnodes.h:1790
HashJoinTable hj_HashTable
Definition: execnodes.h:1784
Definition: pg_list.h:45
uint32 hashvalue
Definition: hashjoin.h:67
#define ResetExprContext(econtext)
Definition: executor.h:334
bool ExecScanHashTableForUnmatched ( HashJoinState hjstate,
ExprContext econtext 
)

Definition at line 1142 of file nodeHash.c.

References HashJoinTableData::buckets, ExprContext::ecxt_innertuple, ExecStoreMinimalTuple(), HeapTupleHeaderHasMatch, HashJoinState::hj_CurBucketNo, HashJoinState::hj_CurSkewBucketNo, HashJoinState::hj_CurTuple, HashJoinState::hj_HashTable, HashJoinState::hj_HashTupleSlot, HJTUPLE_MINTUPLE, HashJoinTableData::nbuckets, HashJoinTupleData::next, HashJoinTableData::nSkewBuckets, NULL, ResetExprContext, HashJoinTableData::skewBucket, HashJoinTableData::skewBucketNums, and HashSkewBucket::tuples.

Referenced by ExecHashJoin().

1143 {
1144  HashJoinTable hashtable = hjstate->hj_HashTable;
1145  HashJoinTuple hashTuple = hjstate->hj_CurTuple;
1146 
1147  for (;;)
1148  {
1149  /*
1150  * hj_CurTuple is the address of the tuple last returned from the
1151  * current bucket, or NULL if it's time to start scanning a new
1152  * bucket.
1153  */
1154  if (hashTuple != NULL)
1155  hashTuple = hashTuple->next;
1156  else if (hjstate->hj_CurBucketNo < hashtable->nbuckets)
1157  {
1158  hashTuple = hashtable->buckets[hjstate->hj_CurBucketNo];
1159  hjstate->hj_CurBucketNo++;
1160  }
1161  else if (hjstate->hj_CurSkewBucketNo < hashtable->nSkewBuckets)
1162  {
1163  int j = hashtable->skewBucketNums[hjstate->hj_CurSkewBucketNo];
1164 
1165  hashTuple = hashtable->skewBucket[j]->tuples;
1166  hjstate->hj_CurSkewBucketNo++;
1167  }
1168  else
1169  break; /* finished all buckets */
1170 
1171  while (hashTuple != NULL)
1172  {
1173  if (!HeapTupleHeaderHasMatch(HJTUPLE_MINTUPLE(hashTuple)))
1174  {
1175  TupleTableSlot *inntuple;
1176 
1177  /* insert hashtable's tuple into exec slot */
1178  inntuple = ExecStoreMinimalTuple(HJTUPLE_MINTUPLE(hashTuple),
1179  hjstate->hj_HashTupleSlot,
1180  false); /* do not pfree */
1181  econtext->ecxt_innertuple = inntuple;
1182 
1183  /*
1184  * Reset temp memory each time; although this function doesn't
1185  * do any qual eval, the caller will, so let's keep it
1186  * parallel to ExecScanHashBucket.
1187  */
1188  ResetExprContext(econtext);
1189 
1190  hjstate->hj_CurTuple = hashTuple;
1191  return true;
1192  }
1193 
1194  hashTuple = hashTuple->next;
1195  }
1196  }
1197 
1198  /*
1199  * no more unmatched tuples
1200  */
1201  return false;
1202 }
TupleTableSlot * ExecStoreMinimalTuple(MinimalTuple mtup, TupleTableSlot *slot, bool shouldFree)
Definition: execTuples.c:387
struct HashJoinTupleData ** buckets
Definition: hashjoin.h:137
int * skewBucketNums
Definition: hashjoin.h:146
int hj_CurSkewBucketNo
Definition: execnodes.h:1787
HashJoinTuple hj_CurTuple
Definition: execnodes.h:1788
HashJoinTuple tuples
Definition: hashjoin.h:97
TupleTableSlot * ecxt_innertuple
Definition: execnodes.h:125
int hj_CurBucketNo
Definition: execnodes.h:1786
HashSkewBucket ** skewBucket
Definition: hashjoin.h:143
#define HJTUPLE_MINTUPLE(hjtup)
Definition: hashjoin.h:72
#define HeapTupleHeaderHasMatch(tup)
Definition: htup_details.h:517
#define NULL
Definition: c.h:226
struct HashJoinTupleData * next
Definition: hashjoin.h:66
TupleTableSlot * hj_HashTupleSlot
Definition: execnodes.h:1790
HashJoinTable hj_HashTable
Definition: execnodes.h:1784
#define ResetExprContext(econtext)
Definition: executor.h:334
Node* MultiExecHash ( HashState node)

Definition at line 74 of file nodeHash.c.

References ExprContext::ecxt_innertuple, ExecHashGetHashValue(), ExecHashGetSkewBucket(), ExecHashIncreaseNumBuckets(), ExecHashSkewTableInsert(), ExecHashTableInsert(), ExecProcNode(), HashState::hashkeys, HashState::hashtable, InstrStartNode(), InstrStopNode(), PlanState::instrument, INVALID_SKEW_BUCKET_NO, HashJoinTableData::keepNulls, HashJoinTableData::nbuckets, HashJoinTableData::nbuckets_optimal, NULL, outerPlanState, HashState::ps, PlanState::ps_ExprContext, HashJoinTableData::skewTuples, HashJoinTableData::spacePeak, HashJoinTableData::spaceUsed, HashJoinTableData::totalTuples, and TupIsNull.

Referenced by MultiExecProcNode().

75 {
76  PlanState *outerNode;
77  List *hashkeys;
78  HashJoinTable hashtable;
79  TupleTableSlot *slot;
80  ExprContext *econtext;
81  uint32 hashvalue;
82 
83  /* must provide our own instrumentation support */
84  if (node->ps.instrument)
86 
87  /*
88  * get state info from node
89  */
90  outerNode = outerPlanState(node);
91  hashtable = node->hashtable;
92 
93  /*
94  * set expression context
95  */
96  hashkeys = node->hashkeys;
97  econtext = node->ps.ps_ExprContext;
98 
99  /*
100  * get all inner tuples and insert into the hash table (or temp files)
101  */
102  for (;;)
103  {
104  slot = ExecProcNode(outerNode);
105  if (TupIsNull(slot))
106  break;
107  /* We have to compute the hash value */
108  econtext->ecxt_innertuple = slot;
109  if (ExecHashGetHashValue(hashtable, econtext, hashkeys,
110  false, hashtable->keepNulls,
111  &hashvalue))
112  {
113  int bucketNumber;
114 
115  bucketNumber = ExecHashGetSkewBucket(hashtable, hashvalue);
116  if (bucketNumber != INVALID_SKEW_BUCKET_NO)
117  {
118  /* It's a skew tuple, so put it into that hash table */
119  ExecHashSkewTableInsert(hashtable, slot, hashvalue,
120  bucketNumber);
121  hashtable->skewTuples += 1;
122  }
123  else
124  {
125  /* Not subject to skew optimization, so insert normally */
126  ExecHashTableInsert(hashtable, slot, hashvalue);
127  }
128  hashtable->totalTuples += 1;
129  }
130  }
131 
132  /* resize the hash table if needed (NTUP_PER_BUCKET exceeded) */
133  if (hashtable->nbuckets != hashtable->nbuckets_optimal)
134  ExecHashIncreaseNumBuckets(hashtable);
135 
136  /* Account for the buckets in spaceUsed (reported in EXPLAIN ANALYZE) */
137  hashtable->spaceUsed += hashtable->nbuckets * sizeof(HashJoinTuple);
138  if (hashtable->spaceUsed > hashtable->spacePeak)
139  hashtable->spacePeak = hashtable->spaceUsed;
140 
141  /* must provide our own instrumentation support */
142  if (node->ps.instrument)
143  InstrStopNode(node->ps.instrument, hashtable->totalTuples);
144 
145  /*
146  * We do not return the hash table directly because it's not a subtype of
147  * Node, and so would violate the MultiExecProcNode API. Instead, our
148  * parent Hashjoin node is expected to know how to fish it out of our node
149  * state. Ugly but not really worth cleaning up, since Hashjoin knows
150  * quite a bit more about Hash besides that.
151  */
152  return NULL;
153 }
double skewTuples
Definition: hashjoin.h:157
void InstrStopNode(Instrumentation *instr, double nTuples)
Definition: instrument.c:80
#define INVALID_SKEW_BUCKET_NO
Definition: hashjoin.h:101
TupleTableSlot * ExecProcNode(PlanState *node)
Definition: execProcnode.c:380
Instrumentation * instrument
Definition: execnodes.h:1057
ExprContext * ps_ExprContext
Definition: execnodes.h:1082
HashJoinTable hashtable
Definition: execnodes.h:2024
int ExecHashGetSkewBucket(HashJoinTable hashtable, uint32 hashvalue)
Definition: nodeHash.c:1445
void ExecHashTableInsert(HashJoinTable hashtable, TupleTableSlot *slot, uint32 hashvalue)
Definition: nodeHash.c:830
static void ExecHashIncreaseNumBuckets(HashJoinTable hashtable)
Definition: nodeHash.c:759
void InstrStartNode(Instrumentation *instr)
Definition: instrument.c:63
struct HashJoinTupleData * HashJoinTuple
Definition: execnodes.h:1774
#define outerPlanState(node)
Definition: execnodes.h:1096
TupleTableSlot * ecxt_innertuple
Definition: execnodes.h:125
List * hashkeys
Definition: execnodes.h:2025
#define TupIsNull(slot)
Definition: tuptable.h:138
unsigned int uint32
Definition: c.h:265
PlanState ps
Definition: execnodes.h:2023
double totalTuples
Definition: hashjoin.h:156
#define NULL
Definition: c.h:226
static void ExecHashSkewTableInsert(HashJoinTable hashtable, TupleTableSlot *slot, uint32 hashvalue, int bucketNumber)
Definition: nodeHash.c:1491
bool ExecHashGetHashValue(HashJoinTable hashtable, ExprContext *econtext, List *hashkeys, bool outer_tuple, bool keep_nulls, uint32 *hashvalue)
Definition: nodeHash.c:924
Definition: pg_list.h:45