enhance: [2.4] Reduce memory usage of BF in DataNode and QueryNode

internal/datanode/writebuffer/l0_write_buffer.go

@@ -165,23 +165,24 @@
 
 	if paramtable.Get().DataNodeCfg.SkipBFStatsLoad.GetAsBool() {
 		// In Skip BF mode, datanode no longer maintains bloom filters.
-		// So, here we skip filtering delete entries.
+		// So, here we skip generating BF (growing segment's BF will be regenerated during the sync phase)
+		// and also skip filtering delete entries by bf.
 		wb.dispatchDeleteMsgsWithoutFilter(deleteMsgs, startPos, endPos)
 	} else {
 		// distribute delete msg
 		// bf write buffer check bloom filter of segment and current insert batch to decide which segment to write delete data
 		wb.dispatchDeleteMsgs(groups, deleteMsgs, startPos, endPos)
-	}
 
-	// update pk oracle
-	for _, inData := range groups {
-		// segment shall always exists after buffer insert
-		segments := wb.metaCache.GetSegmentsBy(metacache.WithSegmentIDs(inData.segmentID))
-		for _, segment := range segments {
-			for _, fieldData := range inData.pkField {
-				err := segment.GetBloomFilterSet().UpdatePKRange(fieldData)
-				if err != nil {
-					return err
+		// update pk oracle
+		for _, inData := range groups {
+			// segment shall always exists after buffer insert
+			segments := wb.metaCache.GetSegmentsBy(metacache.WithSegmentIDs(inData.segmentID))
+			for _, segment := range segments {
+				for _, fieldData := range inData.pkField {
+					err := segment.GetBloomFilterSet().UpdatePKRange(fieldData)
+					if err != nil {
+						return err
+					}
 				}
 			}
 		}

internal/datanode/writebuffer/l0_write_buffer_test.go

@@ -16,7 +16,6 @@ import (
 	"github.com/milvus-io/milvus/internal/allocator"
 	"github.com/milvus-io/milvus/internal/datanode/metacache"
 	"github.com/milvus-io/milvus/internal/datanode/syncmgr"
-	"github.com/milvus-io/milvus/internal/proto/datapb"
 	"github.com/milvus-io/milvus/internal/storage"
 	"github.com/milvus-io/milvus/pkg/common"
 	"github.com/milvus-io/milvus/pkg/metrics"
@@ -181,8 +180,6 @@ func (s *L0WriteBufferSuite) TestBufferData() {
 		pks, msg := s.composeInsertMsg(1000, 10, 128, schemapb.DataType_Int64)
 		delMsg := s.composeDeleteMsg(lo.Map(pks, func(id int64, _ int) storage.PrimaryKey { return storage.NewInt64PrimaryKey(id) }))
 
-		seg := metacache.NewSegmentInfo(&datapb.SegmentInfo{ID: 1000}, metacache.NewBloomFilterSet())
-		s.metacache.EXPECT().GetSegmentsBy(mock.Anything, mock.Anything).Return([]*metacache.SegmentInfo{seg})
 		s.metacache.EXPECT().GetSegmentByID(int64(1000)).Return(nil, false).Once()
 		s.metacache.EXPECT().AddSegment(mock.Anything, mock.Anything, mock.Anything).Return()
 		s.metacache.EXPECT().UpdateSegments(mock.Anything, mock.Anything).Return()
@@ -210,8 +207,6 @@ func (s *L0WriteBufferSuite) TestBufferData() {
 		pks, msg := s.composeInsertMsg(1000, 10, 128, schemapb.DataType_VarChar)
 		delMsg := s.composeDeleteMsg(lo.Map(pks, func(id int64, _ int) storage.PrimaryKey { return storage.NewInt64PrimaryKey(id) }))
 
-		seg := metacache.NewSegmentInfo(&datapb.SegmentInfo{ID: 1000}, metacache.NewBloomFilterSet())
-		s.metacache.EXPECT().GetSegmentsBy(mock.Anything, mock.Anything).Return([]*metacache.SegmentInfo{seg})
 		s.metacache.EXPECT().AddSegment(mock.Anything, mock.Anything, mock.Anything).Return()
 		s.metacache.EXPECT().UpdateSegments(mock.Anything, mock.Anything).Return()
 

internal/querynodev2/segments/segment.go

@@ -94,6 +94,7 @@ type baseSegment struct {
 	bloomFilterSet *pkoracle.BloomFilterSet
 	loadInfo       *atomic.Pointer[querypb.SegmentLoadInfo]
 	isLazyLoad     bool
+	skipGrowingBF  bool // Skip generating or maintaining BF for growing segments; deletion checks will be handled in segcore.
 	channel        metautil.Channel
 
 	resourceUsageCache *atomic.Pointer[ResourceUsage]
@@ -114,6 +115,7 @@ func newBaseSegment(collection *Collection, segmentType SegmentType, version int
 		bloomFilterSet: pkoracle.NewBloomFilterSet(loadInfo.GetSegmentID(), loadInfo.GetPartitionID(), segmentType),
 		channel:        channel,
 		isLazyLoad:     isLazyLoad(collection, segmentType),
+		skipGrowingBF:  segmentType == SegmentTypeGrowing && paramtable.Get().QueryNodeCfg.SkipGrowingSegmentBF.GetAsBool(),
 
 		resourceUsageCache: atomic.NewPointer[ResourceUsage](nil),
 		needUpdatedVersion: atomic.NewInt64(0),
@@ -183,17 +185,30 @@ func (s *baseSegment) LoadInfo() *querypb.SegmentLoadInfo {
 }
 
 func (s *baseSegment) UpdateBloomFilter(pks []storage.PrimaryKey) {
+	if s.skipGrowingBF {
+		return
+	}
 	s.bloomFilterSet.UpdateBloomFilter(pks)
 }
 
 // MayPkExist returns true if the given PK exists in the PK range and being positive through the bloom filter,
 // false otherwise,
 // may returns true even the PK doesn't exist actually
 func (s *baseSegment) MayPkExist(pk *storage.LocationsCache) bool {
+	if s.skipGrowingBF {
+		return true
+	}
 	return s.bloomFilterSet.MayPkExist(pk)
 }
 
 func (s *baseSegment) BatchPkExist(lc *storage.BatchLocationsCache) []bool {
+	if s.skipGrowingBF {
+		allPositive := make([]bool, lc.Size())
+		for i := 0; i < lc.Size(); i++ {
+			allPositive[i] = true
+		}
+		return allPositive
+	}
 	return s.bloomFilterSet.BatchPkExist(lc)
 }
 

pkg/util/paramtable/component_param.go

@@ -2448,6 +2448,8 @@ type queryNodeConfig struct {
 	DefaultSegmentFilterRatio               ParamItem `refreshable:"false"`
 	UseStreamComputing                      ParamItem `refreshable:"false"`
 
+	// BF
+	SkipGrowingSegmentBF           ParamItem `refreshable:"true"`
 	BloomFilterApplyParallelFactor ParamItem `refreshable:"true"`
 
 	QueryStreamBatchSize ParamItem `refreshable:"false"`
@@ -3144,6 +3146,14 @@ user-task-polling:
 	}
 	p.QueryStreamBatchSize.Init(base.mgr)
 
+	p.SkipGrowingSegmentBF = ParamItem{
+		Key:          "queryNode.skipGrowingSegmentBF",
+		Version:      "2.5",
+		DefaultValue: "true",
+		Doc:          "indicates whether skipping the creation, maintenance, or checking of Bloom Filters for growing segments",
+	}
+	p.SkipGrowingSegmentBF.Init(base.mgr)
+
 	p.WorkerPoolingSize = ParamItem{
 		Key:          "queryNode.workerPooling.size",
 		Version:      "2.4.7",

pkg/util/paramtable/component_param_test.go

@@ -456,6 +456,8 @@ func TestComponentParam(t *testing.T) {
 		assert.Equal(t, 3*time.Second, Params.LazyLoadRequestResourceRetryInterval.GetAsDuration(time.Millisecond))
 
 		assert.Equal(t, 4, Params.BloomFilterApplyParallelFactor.GetAsInt())
+		assert.Equal(t, true, Params.SkipGrowingSegmentBF.GetAsBool())
+
 		assert.Equal(t, "/var/lib/milvus/data/mmap", Params.MmapDirPath.GetValue())
 
 		assert.Equal(t, true, Params.MmapChunkCache.GetAsBool())