PostgreSQL Source Code  git master
brin_bloom.c File Reference
#include "postgres.h"
#include <math.h>
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_page.h"
#include "access/brin_tuple.h"
#include "access/genam.h"
#include "access/htup_details.h"
#include "access/reloptions.h"
#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "catalog/pg_type.h"
#include "common/hashfn.h"
#include "utils/fmgrprotos.h"
#include "utils/rel.h"
Include dependency graph for brin_bloom.c:

Go to the source code of this file.

Data Structures

struct  BloomOptions
 
struct  BloomFilter
 
struct  BloomOpaque
 

Macros

#define BloomEqualStrategyNumber   1
 
#define BLOOM_MAX_PROCNUMS   1 /* maximum support procs we need */
 
#define PROCNUM_HASH   11 /* required */
 
#define PROCNUM_BASE   11
 
#define BLOOM_MIN_NDISTINCT_PER_RANGE   16
 
#define BLOOM_DEFAULT_NDISTINCT_PER_RANGE   -0.1 /* 10% of values */
 
#define BLOOM_MIN_FALSE_POSITIVE_RATE   0.0001 /* 0.01% fp rate */
 
#define BLOOM_MAX_FALSE_POSITIVE_RATE   0.25 /* 25% fp rate */
 
#define BLOOM_DEFAULT_FALSE_POSITIVE_RATE   0.01 /* 1% fp rate */
 
#define BloomGetNDistinctPerRange(opts)
 
#define BloomGetFalsePositiveRate(opts)
 
#define BloomMaxFilterSize
 
#define BLOOM_SEED_1   0x71d924af
 
#define BLOOM_SEED_2   0xba48b314
 

Typedefs

typedef struct BloomOptions BloomOptions
 
typedef struct BloomFilter BloomFilter
 
typedef struct BloomOpaque BloomOpaque
 

Functions

static void bloom_filter_size (int ndistinct, double false_positive_rate, int *nbytesp, int *nbitsp, int *nhashesp)
 
static BloomFilterbloom_init (int ndistinct, double false_positive_rate)
 
static BloomFilterbloom_add_value (BloomFilter *filter, uint32 value, bool *updated)
 
static bool bloom_contains_value (BloomFilter *filter, uint32 value)
 
static FmgrInfobloom_get_procinfo (BrinDesc *bdesc, uint16 attno, uint16 procnum)
 
Datum brin_bloom_opcinfo (PG_FUNCTION_ARGS)
 
static int brin_bloom_get_ndistinct (BrinDesc *bdesc, BloomOptions *opts)
 
Datum brin_bloom_add_value (PG_FUNCTION_ARGS)
 
Datum brin_bloom_consistent (PG_FUNCTION_ARGS)
 
Datum brin_bloom_union (PG_FUNCTION_ARGS)
 
Datum brin_bloom_options (PG_FUNCTION_ARGS)
 
Datum brin_bloom_summary_in (PG_FUNCTION_ARGS)
 
Datum brin_bloom_summary_out (PG_FUNCTION_ARGS)
 
Datum brin_bloom_summary_recv (PG_FUNCTION_ARGS)
 
Datum brin_bloom_summary_send (PG_FUNCTION_ARGS)
 

Macro Definition Documentation

◆ BLOOM_DEFAULT_FALSE_POSITIVE_RATE

#define BLOOM_DEFAULT_FALSE_POSITIVE_RATE   0.01 /* 1% fp rate */

Definition at line 194 of file brin_bloom.c.

◆ BLOOM_DEFAULT_NDISTINCT_PER_RANGE

#define BLOOM_DEFAULT_NDISTINCT_PER_RANGE   -0.1 /* 10% of values */

Definition at line 176 of file brin_bloom.c.

◆ BLOOM_MAX_FALSE_POSITIVE_RATE

#define BLOOM_MAX_FALSE_POSITIVE_RATE   0.25 /* 25% fp rate */

Definition at line 193 of file brin_bloom.c.

◆ BLOOM_MAX_PROCNUMS

#define BLOOM_MAX_PROCNUMS   1 /* maximum support procs we need */

Definition at line 142 of file brin_bloom.c.

◆ BLOOM_MIN_FALSE_POSITIVE_RATE

#define BLOOM_MIN_FALSE_POSITIVE_RATE   0.0001 /* 0.01% fp rate */

Definition at line 192 of file brin_bloom.c.

◆ BLOOM_MIN_NDISTINCT_PER_RANGE

#define BLOOM_MIN_NDISTINCT_PER_RANGE   16

Definition at line 169 of file brin_bloom.c.

◆ BLOOM_SEED_1

#define BLOOM_SEED_1   0x71d924af

Definition at line 222 of file brin_bloom.c.

◆ BLOOM_SEED_2

#define BLOOM_SEED_2   0xba48b314

Definition at line 223 of file brin_bloom.c.

◆ BloomEqualStrategyNumber

#define BloomEqualStrategyNumber   1

Definition at line 133 of file brin_bloom.c.

◆ BloomGetFalsePositiveRate

#define BloomGetFalsePositiveRate (   opts)
Value:
((opts) && (((BloomOptions *) (opts))->falsePositiveRate != 0.0) ? \
(((BloomOptions *) (opts))->falsePositiveRate) : \
BLOOM_DEFAULT_FALSE_POSITIVE_RATE)
static AmcheckOptions opts
Definition: pg_amcheck.c:111

Definition at line 201 of file brin_bloom.c.

◆ BloomGetNDistinctPerRange

#define BloomGetNDistinctPerRange (   opts)
Value:
((opts) && (((BloomOptions *) (opts))->nDistinctPerRange != 0) ? \
(((BloomOptions *) (opts))->nDistinctPerRange) : \
#define BLOOM_DEFAULT_NDISTINCT_PER_RANGE
Definition: brin_bloom.c:176

Definition at line 196 of file brin_bloom.c.

◆ BloomMaxFilterSize

#define BloomMaxFilterSize
Value:
MAXALIGN_DOWN(BLCKSZ - \
sizeof(ItemIdData)) + \
#define SizeOfBrinTuple
Definition: brin_tuple.h:80
#define SizeOfPageHeaderData
Definition: bufpage.h:213
#define MAXALIGN_DOWN(LEN)
Definition: c.h:823
#define MAXALIGN(LEN)
Definition: c.h:811

Definition at line 211 of file brin_bloom.c.

◆ PROCNUM_BASE

#define PROCNUM_BASE   11

Definition at line 149 of file brin_bloom.c.

◆ PROCNUM_HASH

#define PROCNUM_HASH   11 /* required */

Definition at line 143 of file brin_bloom.c.

Typedef Documentation

◆ BloomFilter

typedef struct BloomFilter BloomFilter

◆ BloomOpaque

typedef struct BloomOpaque BloomOpaque

◆ BloomOptions

typedef struct BloomOptions BloomOptions

Function Documentation

◆ bloom_add_value()

static BloomFilter* bloom_add_value ( BloomFilter filter,
uint32  value,
bool updated 
)
static

Definition at line 370 of file brin_bloom.c.

371 {
372  int i;
373  uint64 h1,
374  h2;
375 
376  /* compute the hashes, used for the bloom filter */
379 
380  /* compute the requested number of hashes */
381  for (i = 0; i < filter->nhashes; i++)
382  {
383  /* h1 + h2 + f(i) */
384  uint32 h = (h1 + i * h2) % filter->nbits;
385  uint32 byte = (h / 8);
386  uint32 bit = (h % 8);
387 
388  /* if the bit is not set, set it and remember we did that */
389  if (!(filter->data[byte] & (0x01 << bit)))
390  {
391  filter->data[byte] |= (0x01 << bit);
392  filter->nbits_set++;
393  if (updated)
394  *updated = true;
395  }
396  }
397 
398  return filter;
399 }
#define BLOOM_SEED_1
Definition: brin_bloom.c:222
#define BLOOM_SEED_2
Definition: brin_bloom.c:223
unsigned int uint32
Definition: c.h:506
uint64 hash_bytes_uint32_extended(uint32 k, uint64 seed)
Definition: hashfn.c:631
static struct @155 value
int i
Definition: isn.c:73
uint8 nhashes
Definition: brin_bloom.c:251
char data[FLEXIBLE_ARRAY_MEMBER]
Definition: brin_bloom.c:256
uint32 nbits_set
Definition: brin_bloom.c:253
uint32 nbits
Definition: brin_bloom.c:252
Datum bit(PG_FUNCTION_ARGS)
Definition: varbit.c:391

References bit(), BLOOM_SEED_1, BLOOM_SEED_2, BloomFilter::data, hash_bytes_uint32_extended(), i, BloomFilter::nbits, BloomFilter::nbits_set, BloomFilter::nhashes, and value.

Referenced by brin_bloom_add_value().

◆ bloom_contains_value()

static bool bloom_contains_value ( BloomFilter filter,
uint32  value 
)
static

Definition at line 407 of file brin_bloom.c.

408 {
409  int i;
410  uint64 h1,
411  h2;
412 
413  /* calculate the two hashes */
416 
417  /* compute the requested number of hashes */
418  for (i = 0; i < filter->nhashes; i++)
419  {
420  /* h1 + h2 + f(i) */
421  uint32 h = (h1 + i * h2) % filter->nbits;
422  uint32 byte = (h / 8);
423  uint32 bit = (h % 8);
424 
425  /* if the bit is not set, the value is not there */
426  if (!(filter->data[byte] & (0x01 << bit)))
427  return false;
428  }
429 
430  /* all hashes found in bloom filter */
431  return true;
432 }

References bit(), BLOOM_SEED_1, BLOOM_SEED_2, BloomFilter::data, hash_bytes_uint32_extended(), i, BloomFilter::nbits, BloomFilter::nhashes, and value.

Referenced by brin_bloom_consistent().

◆ bloom_filter_size()

static void bloom_filter_size ( int  ndistinct,
double  false_positive_rate,
int *  nbytesp,
int *  nbitsp,
int *  nhashesp 
)
static

Definition at line 271 of file brin_bloom.c.

273 {
274  double k;
275  int nbits,
276  nbytes;
277 
278  /* sizing bloom filter: -(n * ln(p)) / (ln(2))^2 */
279  nbits = ceil(-(ndistinct * log(false_positive_rate)) / pow(log(2.0), 2));
280 
281  /* round m to whole bytes */
282  nbytes = ((nbits + 7) / 8);
283  nbits = nbytes * 8;
284 
285  /*
286  * round(log(2.0) * m / ndistinct), but assume round() may not be
287  * available on Windows
288  */
289  k = log(2.0) * nbits / ndistinct;
290  k = (k - floor(k) >= 0.5) ? ceil(k) : floor(k);
291 
292  if (nbytesp)
293  *nbytesp = nbytes;
294 
295  if (nbitsp)
296  *nbitsp = nbits;
297 
298  if (nhashesp)
299  *nhashesp = (int) k;
300 }

Referenced by bloom_init().

◆ bloom_get_procinfo()

static FmgrInfo * bloom_get_procinfo ( BrinDesc bdesc,
uint16  attno,
uint16  procnum 
)
static

Definition at line 707 of file brin_bloom.c.

708 {
709  BloomOpaque *opaque;
710  uint16 basenum = procnum - PROCNUM_BASE;
711 
712  /*
713  * We cache these in the opaque struct, to avoid repetitive syscache
714  * lookups.
715  */
716  opaque = (BloomOpaque *) bdesc->bd_info[attno - 1]->oi_opaque;
717 
718  /*
719  * If we already searched for this proc and didn't find it, don't bother
720  * searching again.
721  */
722  if (opaque->extra_proc_missing[basenum])
723  return NULL;
724 
725  if (opaque->extra_procinfos[basenum].fn_oid == InvalidOid)
726  {
727  if (RegProcedureIsValid(index_getprocid(bdesc->bd_index, attno,
728  procnum)))
729  {
730  fmgr_info_copy(&opaque->extra_procinfos[basenum],
731  index_getprocinfo(bdesc->bd_index, attno, procnum),
732  bdesc->bd_context);
733  }
734  else
735  {
736  opaque->extra_proc_missing[basenum] = true;
737  return NULL;
738  }
739  }
740 
741  return &opaque->extra_procinfos[basenum];
742 }
#define PROCNUM_BASE
Definition: brin_bloom.c:149
unsigned short uint16
Definition: c.h:505
#define RegProcedureIsValid(p)
Definition: c.h:777
void fmgr_info_copy(FmgrInfo *dstinfo, FmgrInfo *srcinfo, MemoryContext destcxt)
Definition: fmgr.c:580
FmgrInfo * index_getprocinfo(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:860
RegProcedure index_getprocid(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:826
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:77
#define InvalidOid
Definition: postgres_ext.h:36
FmgrInfo extra_procinfos[BLOOM_MAX_PROCNUMS]
Definition: brin_bloom.c:441
bool extra_proc_missing[BLOOM_MAX_PROCNUMS]
Definition: brin_bloom.c:442
BrinOpcInfo * bd_info[FLEXIBLE_ARRAY_MEMBER]
Definition: brin_internal.h:62
Relation bd_index
Definition: brin_internal.h:50
MemoryContext bd_context
Definition: brin_internal.h:47
void * oi_opaque
Definition: brin_internal.h:34
Oid fn_oid
Definition: fmgr.h:59

References BrinDesc::bd_context, BrinDesc::bd_index, BrinDesc::bd_info, BloomOpaque::extra_proc_missing, BloomOpaque::extra_procinfos, fmgr_info_copy(), FmgrInfo::fn_oid, if(), index_getprocid(), index_getprocinfo(), InvalidOid, BrinOpcInfo::oi_opaque, PROCNUM_BASE, and RegProcedureIsValid.

Referenced by brin_bloom_add_value(), and brin_bloom_consistent().

◆ bloom_init()

static BloomFilter* bloom_init ( int  ndistinct,
double  false_positive_rate 
)
static

Definition at line 310 of file brin_bloom.c.

311 {
312  Size len;
313  BloomFilter *filter;
314 
315  int nbits; /* size of filter / number of bits */
316  int nbytes; /* size of filter / number of bytes */
317  int nhashes; /* number of hash functions */
318 
319  Assert(ndistinct > 0);
320  Assert(false_positive_rate > 0 && false_positive_rate < 1);
321 
322  /* calculate bloom filter size / parameters */
323  bloom_filter_size(ndistinct, false_positive_rate,
324  &nbytes, &nbits, &nhashes);
325 
326  /*
327  * Reject filters that are obviously too large to store on a page.
328  *
329  * Initially the bloom filter is just zeroes and so very compressible, but
330  * as we add values it gets more and more random, and so less and less
331  * compressible. So initially everything fits on the page, but we might
332  * get surprising failures later - we want to prevent that, so we reject
333  * bloom filter that are obviously too large.
334  *
335  * XXX It's not uncommon to oversize the bloom filter a bit, to defend
336  * against unexpected data anomalies (parts of table with more distinct
337  * values per range etc.). But we still need to make sure even the
338  * oversized filter fits on page, if such need arises.
339  *
340  * XXX This check is not perfect, because the index may have multiple
341  * filters that are small individually, but too large when combined.
342  */
343  if (nbytes > BloomMaxFilterSize)
344  elog(ERROR, "the bloom filter is too large (%d > %zu)", nbytes,
346 
347  /*
348  * We allocate the whole filter. Most of it is going to be 0 bits, so the
349  * varlena is easy to compress.
350  */
351  len = offsetof(BloomFilter, data) + nbytes;
352 
353  filter = (BloomFilter *) palloc0(len);
354 
355  filter->flags = 0;
356  filter->nhashes = nhashes;
357  filter->nbits = nbits;
358 
359  SET_VARSIZE(filter, len);
360 
361  return filter;
362 }
static void bloom_filter_size(int ndistinct, double false_positive_rate, int *nbytesp, int *nbitsp, int *nhashesp)
Definition: brin_bloom.c:271
#define BloomMaxFilterSize
Definition: brin_bloom.c:211
#define Assert(condition)
Definition: c.h:858
size_t Size
Definition: c.h:605
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:224
void * palloc0(Size size)
Definition: mcxt.c:1346
const void size_t len
const void * data
uint16 flags
Definition: brin_bloom.c:248
#define SET_VARSIZE(PTR, len)
Definition: varatt.h:305

References Assert, bloom_filter_size(), BloomMaxFilterSize, data, elog, ERROR, BloomFilter::flags, len, BloomFilter::nbits, BloomFilter::nhashes, palloc0(), and SET_VARSIZE.

Referenced by brin_bloom_add_value().

◆ brin_bloom_add_value()

Datum brin_bloom_add_value ( PG_FUNCTION_ARGS  )

Definition at line 540 of file brin_bloom.c.

541 {
542  BrinDesc *bdesc = (BrinDesc *) PG_GETARG_POINTER(0);
543  BrinValues *column = (BrinValues *) PG_GETARG_POINTER(1);
545  bool isnull PG_USED_FOR_ASSERTS_ONLY = PG_GETARG_DATUM(3);
547  Oid colloid = PG_GET_COLLATION();
548  FmgrInfo *hashFn;
549  uint32 hashValue;
550  bool updated = false;
551  AttrNumber attno;
552  BloomFilter *filter;
553 
554  Assert(!isnull);
555 
556  attno = column->bv_attno;
557 
558  /*
559  * If this is the first non-null value, we need to initialize the bloom
560  * filter. Otherwise just extract the existing bloom filter from
561  * BrinValues.
562  */
563  if (column->bv_allnulls)
564  {
565  filter = bloom_init(brin_bloom_get_ndistinct(bdesc, opts),
567  column->bv_values[0] = PointerGetDatum(filter);
568  column->bv_allnulls = false;
569  updated = true;
570  }
571  else
572  filter = (BloomFilter *) PG_DETOAST_DATUM(column->bv_values[0]);
573 
574  /*
575  * Compute the hash of the new value, using the supplied hash function,
576  * and then add the hash value to the bloom filter.
577  */
578  hashFn = bloom_get_procinfo(bdesc, attno, PROCNUM_HASH);
579 
580  hashValue = DatumGetUInt32(FunctionCall1Coll(hashFn, colloid, newval));
581 
582  filter = bloom_add_value(filter, hashValue, &updated);
583 
584  column->bv_values[0] = PointerGetDatum(filter);
585 
586  PG_RETURN_BOOL(updated);
587 }
int16 AttrNumber
Definition: attnum.h:21
static int brin_bloom_get_ndistinct(BrinDesc *bdesc, BloomOptions *opts)
Definition: brin_bloom.c:497
#define PROCNUM_HASH
Definition: brin_bloom.c:143
static BloomFilter * bloom_init(int ndistinct, double false_positive_rate)
Definition: brin_bloom.c:310
#define BloomGetFalsePositiveRate(opts)
Definition: brin_bloom.c:201
static BloomFilter * bloom_add_value(BloomFilter *filter, uint32 value, bool *updated)
Definition: brin_bloom.c:370
static FmgrInfo * bloom_get_procinfo(BrinDesc *bdesc, uint16 attno, uint16 procnum)
Definition: brin_bloom.c:707
#define PG_USED_FOR_ASSERTS_ONLY
Definition: c.h:182
Datum FunctionCall1Coll(FmgrInfo *flinfo, Oid collation, Datum arg1)
Definition: fmgr.c:1129
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:276
#define PG_GETARG_DATUM(n)
Definition: fmgr.h:268
#define PG_GET_OPCLASS_OPTIONS()
Definition: fmgr.h:342
#define PG_DETOAST_DATUM(datum)
Definition: fmgr.h:240
#define PG_GET_COLLATION()
Definition: fmgr.h:198
#define PG_RETURN_BOOL(x)
Definition: fmgr.h:359
#define newval
static uint32 DatumGetUInt32(Datum X)
Definition: postgres.h:222
static Datum PointerGetDatum(const void *X)
Definition: postgres.h:322
uintptr_t Datum
Definition: postgres.h:64
unsigned int Oid
Definition: postgres_ext.h:31
Datum * bv_values
Definition: brin_tuple.h:34
AttrNumber bv_attno
Definition: brin_tuple.h:31
bool bv_allnulls
Definition: brin_tuple.h:33
Definition: fmgr.h:57

References Assert, bloom_add_value(), bloom_get_procinfo(), bloom_init(), BloomGetFalsePositiveRate, brin_bloom_get_ndistinct(), BrinValues::bv_allnulls, BrinValues::bv_attno, BrinValues::bv_values, DatumGetUInt32(), FunctionCall1Coll(), newval, opts, PG_DETOAST_DATUM, PG_GET_COLLATION, PG_GET_OPCLASS_OPTIONS, PG_GETARG_DATUM, PG_GETARG_POINTER, PG_RETURN_BOOL, PG_USED_FOR_ASSERTS_ONLY, PointerGetDatum(), and PROCNUM_HASH.

◆ brin_bloom_consistent()

Datum brin_bloom_consistent ( PG_FUNCTION_ARGS  )

Definition at line 595 of file brin_bloom.c.

596 {
597  BrinDesc *bdesc = (BrinDesc *) PG_GETARG_POINTER(0);
598  BrinValues *column = (BrinValues *) PG_GETARG_POINTER(1);
599  ScanKey *keys = (ScanKey *) PG_GETARG_POINTER(2);
600  int nkeys = PG_GETARG_INT32(3);
601  Oid colloid = PG_GET_COLLATION();
602  AttrNumber attno;
603  Datum value;
604  bool matches;
605  FmgrInfo *finfo;
606  uint32 hashValue;
607  BloomFilter *filter;
608  int keyno;
609 
610  filter = (BloomFilter *) PG_DETOAST_DATUM(column->bv_values[0]);
611 
612  Assert(filter);
613 
614  /*
615  * Assume all scan keys match. We'll be searching for a scan key
616  * eliminating the page range (we can stop on the first such key).
617  */
618  matches = true;
619 
620  for (keyno = 0; keyno < nkeys; keyno++)
621  {
622  ScanKey key = keys[keyno];
623 
624  /* NULL keys are handled and filtered-out in bringetbitmap */
625  Assert(!(key->sk_flags & SK_ISNULL));
626 
627  attno = key->sk_attno;
628  value = key->sk_argument;
629 
630  switch (key->sk_strategy)
631  {
633 
634  /*
635  * We want to return the current page range if the bloom
636  * filter seems to contain the value.
637  */
638  finfo = bloom_get_procinfo(bdesc, attno, PROCNUM_HASH);
639 
640  hashValue = DatumGetUInt32(FunctionCall1Coll(finfo, colloid, value));
641  matches &= bloom_contains_value(filter, hashValue);
642 
643  break;
644  default:
645  /* shouldn't happen */
646  elog(ERROR, "invalid strategy number %d", key->sk_strategy);
647  matches = false;
648  break;
649  }
650 
651  if (!matches)
652  break;
653  }
654 
655  PG_RETURN_BOOL(matches);
656 }
static bool bloom_contains_value(BloomFilter *filter, uint32 value)
Definition: brin_bloom.c:407
#define BloomEqualStrategyNumber
Definition: brin_bloom.c:133
#define PG_GETARG_INT32(n)
Definition: fmgr.h:269
#define SK_ISNULL
Definition: skey.h:115

References Assert, bloom_contains_value(), bloom_get_procinfo(), BloomEqualStrategyNumber, BrinValues::bv_values, DatumGetUInt32(), elog, ERROR, FunctionCall1Coll(), sort-test::key, PG_DETOAST_DATUM, PG_GET_COLLATION, PG_GETARG_INT32, PG_GETARG_POINTER, PG_RETURN_BOOL, PROCNUM_HASH, SK_ISNULL, and value.

◆ brin_bloom_get_ndistinct()

static int brin_bloom_get_ndistinct ( BrinDesc bdesc,
BloomOptions opts 
)
static

Definition at line 497 of file brin_bloom.c.

498 {
499  double ndistinct;
500  double maxtuples;
501  BlockNumber pagesPerRange;
502 
503  pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
504  ndistinct = BloomGetNDistinctPerRange(opts);
505 
506  Assert(BlockNumberIsValid(pagesPerRange));
507 
508  maxtuples = MaxHeapTuplesPerPage * pagesPerRange;
509 
510  /*
511  * Similarly to n_distinct, negative values are relative - in this case to
512  * maximum number of tuples in the page range (maxtuples).
513  */
514  if (ndistinct < 0)
515  ndistinct = (-ndistinct) * maxtuples;
516 
517  /*
518  * Positive values are to be used directly, but we still apply a couple of
519  * safeties to avoid using unreasonably small bloom filters.
520  */
521  ndistinct = Max(ndistinct, BLOOM_MIN_NDISTINCT_PER_RANGE);
522 
523  /*
524  * And don't use more than the maximum possible number of tuples, in the
525  * range, which would be entirely wasteful.
526  */
527  ndistinct = Min(ndistinct, maxtuples);
528 
529  return (int) ndistinct;
530 }
uint32 BlockNumber
Definition: block.h:31
static bool BlockNumberIsValid(BlockNumber blockNumber)
Definition: block.h:71
#define BrinGetPagesPerRange(relation)
Definition: brin.h:40
#define BloomGetNDistinctPerRange(opts)
Definition: brin_bloom.c:196
#define BLOOM_MIN_NDISTINCT_PER_RANGE
Definition: brin_bloom.c:169
#define Min(x, y)
Definition: c.h:1004
#define Max(x, y)
Definition: c.h:998
#define MaxHeapTuplesPerPage
Definition: htup_details.h:572

References Assert, BrinDesc::bd_index, BlockNumberIsValid(), BLOOM_MIN_NDISTINCT_PER_RANGE, BloomGetNDistinctPerRange, BrinGetPagesPerRange, Max, MaxHeapTuplesPerPage, Min, and opts.

Referenced by brin_bloom_add_value().

◆ brin_bloom_opcinfo()

Datum brin_bloom_opcinfo ( PG_FUNCTION_ARGS  )

Definition at line 450 of file brin_bloom.c.

451 {
452  BrinOpcInfo *result;
453 
454  /*
455  * opaque->strategy_procinfos is initialized lazily; here it is set to
456  * all-uninitialized by palloc0 which sets fn_oid to InvalidOid.
457  *
458  * bloom indexes only store the filter as a single BYTEA column
459  */
460 
461  result = palloc0(MAXALIGN(SizeofBrinOpcInfo(1)) +
462  sizeof(BloomOpaque));
463  result->oi_nstored = 1;
464  result->oi_regular_nulls = true;
465  result->oi_opaque = (BloomOpaque *)
466  MAXALIGN((char *) result + SizeofBrinOpcInfo(1));
467  result->oi_typcache[0] = lookup_type_cache(PG_BRIN_BLOOM_SUMMARYOID, 0);
468 
469  PG_RETURN_POINTER(result);
470 }
#define SizeofBrinOpcInfo(ncols)
Definition: brin_internal.h:41
#define PG_RETURN_POINTER(x)
Definition: fmgr.h:361
TypeCacheEntry * oi_typcache[FLEXIBLE_ARRAY_MEMBER]
Definition: brin_internal.h:37
uint16 oi_nstored
Definition: brin_internal.h:28
bool oi_regular_nulls
Definition: brin_internal.h:31
TypeCacheEntry * lookup_type_cache(Oid type_id, int flags)
Definition: typcache.c:346

References lookup_type_cache(), MAXALIGN, BrinOpcInfo::oi_nstored, BrinOpcInfo::oi_opaque, BrinOpcInfo::oi_regular_nulls, BrinOpcInfo::oi_typcache, palloc0(), PG_RETURN_POINTER, and SizeofBrinOpcInfo.

◆ brin_bloom_options()

Datum brin_bloom_options ( PG_FUNCTION_ARGS  )

Definition at line 745 of file brin_bloom.c.

746 {
748 
749  init_local_reloptions(relopts, sizeof(BloomOptions));
750 
751  add_local_real_reloption(relopts, "n_distinct_per_range",
752  "number of distinct items expected in a BRIN page range",
754  -1.0, INT_MAX, offsetof(BloomOptions, nDistinctPerRange));
755 
756  add_local_real_reloption(relopts, "false_positive_rate",
757  "desired false-positive rate for the bloom filters",
761  offsetof(BloomOptions, falsePositiveRate));
762 
763  PG_RETURN_VOID();
764 }
#define BLOOM_MAX_FALSE_POSITIVE_RATE
Definition: brin_bloom.c:193
#define BLOOM_DEFAULT_FALSE_POSITIVE_RATE
Definition: brin_bloom.c:194
#define BLOOM_MIN_FALSE_POSITIVE_RATE
Definition: brin_bloom.c:192
#define PG_RETURN_VOID()
Definition: fmgr.h:349
void init_local_reloptions(local_relopts *relopts, Size relopt_struct_size)
Definition: reloptions.c:734
void add_local_real_reloption(local_relopts *relopts, const char *name, const char *desc, double default_val, double min_val, double max_val, int offset)
Definition: reloptions.c:972

References add_local_real_reloption(), BLOOM_DEFAULT_FALSE_POSITIVE_RATE, BLOOM_DEFAULT_NDISTINCT_PER_RANGE, BLOOM_MAX_FALSE_POSITIVE_RATE, BLOOM_MIN_FALSE_POSITIVE_RATE, init_local_reloptions(), PG_GETARG_POINTER, and PG_RETURN_VOID.

◆ brin_bloom_summary_in()

Datum brin_bloom_summary_in ( PG_FUNCTION_ARGS  )

Definition at line 775 of file brin_bloom.c.

776 {
777  /*
778  * brin_bloom_summary stores the data in binary form and parsing text
779  * input is not needed, so disallow this.
780  */
781  ereport(ERROR,
782  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
783  errmsg("cannot accept a value of type %s", "pg_brin_bloom_summary")));
784 
785  PG_RETURN_VOID(); /* keep compiler quiet */
786 }
int errcode(int sqlerrcode)
Definition: elog.c:857
int errmsg(const char *fmt,...)
Definition: elog.c:1070
#define ereport(elevel,...)
Definition: elog.h:149

References ereport, errcode(), errmsg(), ERROR, and PG_RETURN_VOID.

◆ brin_bloom_summary_out()

Datum brin_bloom_summary_out ( PG_FUNCTION_ARGS  )

Definition at line 797 of file brin_bloom.c.

798 {
799  BloomFilter *filter;
801 
802  /* detoast the data to get value with a full 4B header */
804 
806  appendStringInfoChar(&str, '{');
807 
808  appendStringInfo(&str, "mode: hashed nhashes: %u nbits: %u nbits_set: %u",
809  filter->nhashes, filter->nbits, filter->nbits_set);
810 
811  appendStringInfoChar(&str, '}');
812 
813  PG_RETURN_CSTRING(str.data);
814 }
#define PG_RETURN_CSTRING(x)
Definition: fmgr.h:362
const char * str
void appendStringInfo(StringInfo str, const char *fmt,...)
Definition: stringinfo.c:97
void appendStringInfoChar(StringInfo str, char ch)
Definition: stringinfo.c:194
void initStringInfo(StringInfo str)
Definition: stringinfo.c:59

References appendStringInfo(), appendStringInfoChar(), initStringInfo(), BloomFilter::nbits, BloomFilter::nbits_set, BloomFilter::nhashes, PG_DETOAST_DATUM, PG_GETARG_DATUM, PG_RETURN_CSTRING, and str.

◆ brin_bloom_summary_recv()

Datum brin_bloom_summary_recv ( PG_FUNCTION_ARGS  )

Definition at line 821 of file brin_bloom.c.

822 {
823  ereport(ERROR,
824  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
825  errmsg("cannot accept a value of type %s", "pg_brin_bloom_summary")));
826 
827  PG_RETURN_VOID(); /* keep compiler quiet */
828 }

References ereport, errcode(), errmsg(), ERROR, and PG_RETURN_VOID.

◆ brin_bloom_summary_send()

Datum brin_bloom_summary_send ( PG_FUNCTION_ARGS  )

Definition at line 838 of file brin_bloom.c.

839 {
840  return byteasend(fcinfo);
841 }
Datum byteasend(PG_FUNCTION_ARGS)
Definition: varlena.c:490

References byteasend().

◆ brin_bloom_union()

Datum brin_bloom_union ( PG_FUNCTION_ARGS  )

Definition at line 667 of file brin_bloom.c.

668 {
669  int i;
670  int nbytes;
671  BrinValues *col_a = (BrinValues *) PG_GETARG_POINTER(1);
672  BrinValues *col_b = (BrinValues *) PG_GETARG_POINTER(2);
673  BloomFilter *filter_a;
674  BloomFilter *filter_b;
675 
676  Assert(col_a->bv_attno == col_b->bv_attno);
677  Assert(!col_a->bv_allnulls && !col_b->bv_allnulls);
678 
679  filter_a = (BloomFilter *) PG_DETOAST_DATUM(col_a->bv_values[0]);
680  filter_b = (BloomFilter *) PG_DETOAST_DATUM(col_b->bv_values[0]);
681 
682  /* make sure the filters use the same parameters */
683  Assert(filter_a && filter_b);
684  Assert(filter_a->nbits == filter_b->nbits);
685  Assert(filter_a->nhashes == filter_b->nhashes);
686  Assert((filter_a->nbits > 0) && (filter_a->nbits % 8 == 0));
687 
688  nbytes = (filter_a->nbits) / 8;
689 
690  /* simply OR the bitmaps */
691  for (i = 0; i < nbytes; i++)
692  filter_a->data[i] |= filter_b->data[i];
693 
694  /* update the number of bits set in the filter */
695  filter_a->nbits_set = pg_popcount((const char *) filter_a->data, nbytes);
696 
697  PG_RETURN_VOID();
698 }
static uint64 pg_popcount(const char *buf, int bytes)
Definition: pg_bitutils.h:339

References Assert, BrinValues::bv_allnulls, BrinValues::bv_attno, BrinValues::bv_values, BloomFilter::data, i, BloomFilter::nbits, BloomFilter::nbits_set, BloomFilter::nhashes, PG_DETOAST_DATUM, PG_GETARG_POINTER, pg_popcount(), and PG_RETURN_VOID.