milvus/internal/storage/pk_statistics.go
wei liu c6a1c49e02
enhance: Use Blocked Bloom Filter instead of basic bloom fitler impl. (#33405)
issue: #32995
To speed up the construction and querying of Bloom filters, we chose a
blocked Bloom filter instead of a basic Bloom filter implementation.

WARN: This PR is compatible with old version bf impl, but if fall back
to old milvus version, it may causes bloom filter deserialize failed.

In single Bloom filter test cases with a capacity of 1,000,000 and a
false positive rate (FPR) of 0.001, the blocked Bloom filter is 5 times
faster than the basic Bloom filter in both querying and construction, at
the cost of a 30% increase in memory usage.

- Block BF construct time	{"time": "54.128131ms"}
- Block BF size	                {"size": 3021578}
- Block BF Test cost	        {"time": "55.407352ms"}
- Basic BF construct time	{"time": "210.262183ms"}
- Basic BF size	                {"size": 2396308}
- Basic BF Test cost	        {"time": "192.596229ms"}

In multi Bloom filter test cases with a capacity of 100,000, an FPR of
0.001, and 100 Bloom filters, we reuse the primary key locations for all
Bloom filters to avoid repeated hash computations. As a result, the
blocked Bloom filter is also 5 times faster than the basic Bloom filter
in querying.

- Block BF TestLocation cost    {"time": "529.97183ms"}
- Basic BF TestLocation cost	{"time": "3.197430181s"}

---------

Signed-off-by: Wei Liu <wei.liu@zilliz.com>
2024-05-31 17:49:45 +08:00

192 lines
5.2 KiB
Go

// Licensed to the LF AI & Data foundation under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package storage
import (
"fmt"
"github.com/cockroachdb/errors"
"github.com/milvus-io/milvus-proto/go-api/v2/schemapb"
"github.com/milvus-io/milvus/internal/util/bloomfilter"
"github.com/milvus-io/milvus/pkg/common"
)
// pkStatistics contains pk field statistic information
type PkStatistics struct {
PkFilter bloomfilter.BloomFilterInterface // bloom filter of pk inside a segment
MinPK PrimaryKey // minimal pk value, shortcut for checking whether a pk is inside this segment
MaxPK PrimaryKey // maximal pk value, same above
}
// update set pk min/max value if input value is beyond former range.
func (st *PkStatistics) UpdateMinMax(pk PrimaryKey) error {
if st == nil {
return errors.New("nil pk statistics")
}
if st.MinPK == nil {
st.MinPK = pk
} else if st.MinPK.GT(pk) {
st.MinPK = pk
}
if st.MaxPK == nil {
st.MaxPK = pk
} else if st.MaxPK.LT(pk) {
st.MaxPK = pk
}
return nil
}
func (st *PkStatistics) UpdatePKRange(ids FieldData) error {
switch pks := ids.(type) {
case *Int64FieldData:
buf := make([]byte, 8)
for _, pk := range pks.Data {
id := NewInt64PrimaryKey(pk)
err := st.UpdateMinMax(id)
if err != nil {
return err
}
common.Endian.PutUint64(buf, uint64(pk))
st.PkFilter.Add(buf)
}
case *StringFieldData:
for _, pk := range pks.Data {
id := NewVarCharPrimaryKey(pk)
err := st.UpdateMinMax(id)
if err != nil {
return err
}
st.PkFilter.AddString(pk)
}
default:
return fmt.Errorf("invalid data type for primary key: %T", ids)
}
return nil
}
func (st *PkStatistics) PkExist(pk PrimaryKey) bool {
// empty pkStatics
if st.MinPK == nil || st.MaxPK == nil || st.PkFilter == nil {
return false
}
// check pk range first, ugly but key it for now
if st.MinPK.GT(pk) || st.MaxPK.LT(pk) {
return false
}
// if in range, check bloom filter
switch pk.Type() {
case schemapb.DataType_Int64:
buf := make([]byte, 8)
int64Pk := pk.(*Int64PrimaryKey)
common.Endian.PutUint64(buf, uint64(int64Pk.Value))
return st.PkFilter.Test(buf)
case schemapb.DataType_VarChar:
varCharPk := pk.(*VarCharPrimaryKey)
return st.PkFilter.TestString(varCharPk.Value)
default:
// TODO::
}
// no idea, just make it as false positive
return true
}
// Locations returns a list of hash locations representing a data item.
func Locations(pk PrimaryKey, k uint, bfType bloomfilter.BFType) []uint64 {
switch pk.Type() {
case schemapb.DataType_Int64:
buf := make([]byte, 8)
int64Pk := pk.(*Int64PrimaryKey)
common.Endian.PutUint64(buf, uint64(int64Pk.Value))
return bloomfilter.Locations(buf, k, bfType)
case schemapb.DataType_VarChar:
varCharPk := pk.(*VarCharPrimaryKey)
return bloomfilter.Locations([]byte(varCharPk.Value), k, bfType)
default:
// TODO::
}
return nil
}
func (st *PkStatistics) TestLocations(pk PrimaryKey, locs []uint64) bool {
// empty pkStatics
if st.MinPK == nil || st.MaxPK == nil || st.PkFilter == nil {
return false
}
// check bf first, TestLocation just do some bitset compute, cost is cheaper
if !st.PkFilter.TestLocations(locs) {
return false
}
// check pk range first, ugly but key it for now
return st.MinPK.LE(pk) && st.MaxPK.GE(pk)
}
func (st *PkStatistics) TestLocationCache(lc *LocationsCache) bool {
// empty pkStatics
if st.MinPK == nil || st.MaxPK == nil || st.PkFilter == nil {
return false
}
// check bf first, TestLocation just do some bitset compute, cost is cheaper
if !st.PkFilter.TestLocations(lc.Locations(st.PkFilter.K(), st.PkFilter.Type())) {
return false
}
// check pk range after
return st.MinPK.LE(lc.pk) && st.MaxPK.GE(lc.pk)
}
// LocationsCache is a helper struct caching pk bloom filter locations.
// Note that this helper is not concurrent safe and shall be used in same goroutine.
type LocationsCache struct {
pk PrimaryKey
basicBFLocations []uint64
blockBFLocations []uint64
}
func (lc *LocationsCache) GetPk() PrimaryKey {
return lc.pk
}
func (lc *LocationsCache) Locations(k uint, bfType bloomfilter.BFType) []uint64 {
switch bfType {
case bloomfilter.BasicBF:
if int(k) > len(lc.basicBFLocations) {
lc.basicBFLocations = Locations(lc.pk, k, bfType)
}
return lc.basicBFLocations[:k]
case bloomfilter.BlockedBF:
if int(k) > len(lc.blockBFLocations) {
lc.blockBFLocations = Locations(lc.pk, k, bfType)
}
return lc.blockBFLocations[:k]
default:
return nil
}
}
func NewLocationsCache(pk PrimaryKey) *LocationsCache {
return &LocationsCache{
pk: pk,
}
}