enhance: [2.4] Batch forward delete when using DirectForward (#37076) (#37107)

Cherry pick from master
pr: #37076
Related #36887

DirectFoward streaming delete will cause memory usage explode if the
segments number was large. This PR add batching delete API and using it
for direct forward implementation.

Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>

internal/distributed/querynode/client/client.go

@@ -332,3 +332,16 @@ func (c *Client) Delete(ctx context.Context, req *querypb.DeleteRequest, _ ...gr
 		return client.Delete(ctx, req)
 	})
 }
+
+// DeleteBatch is the API to apply same delete data into multiple segments.
+// it's basically same as `Delete` but cost less memory pressure.
+func (c *Client) DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest, _ ...grpc.CallOption) (*querypb.DeleteBatchResponse, error) {
+	req = typeutil.Clone(req)
+	commonpbutil.UpdateMsgBase(
+		req.GetBase(),
+		commonpbutil.FillMsgBaseFromClient(c.nodeID),
+	)
+	return wrapGrpcCall(ctx, c, func(client querypb.QueryNodeClient) (*querypb.DeleteBatchResponse, error) {
+		return client.DeleteBatch(ctx, req)
+	})
+}

internal/distributed/querynode/client/client_test.go

@@ -108,6 +108,9 @@ func Test_NewClient(t *testing.T) {
 		r20, err := client.SearchSegments(ctx, nil)
 		retCheck(retNotNil, r20, err)
 
+		r21, err := client.DeleteBatch(ctx, nil)
+		retCheck(retNotNil, r21, err)
+
 		// stream rpc
 		client, err := client.QueryStream(ctx, nil)
 		retCheck(retNotNil, client, err)

internal/distributed/querynode/service.go

@@ -383,3 +383,9 @@ func (s *Server) SyncDistribution(ctx context.Context, req *querypb.SyncDistribu
 func (s *Server) Delete(ctx context.Context, req *querypb.DeleteRequest) (*commonpb.Status, error) {
 	return s.querynode.Delete(ctx, req)
 }
+
+// DeleteBatch is the API to apply same delete data into multiple segments.
+// it's basically same as `Delete` but cost less memory pressure.
+func (s *Server) DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	return s.querynode.DeleteBatch(ctx, req)
+}

internal/distributed/querynode/service_test.go

@@ -32,6 +32,7 @@ import (
 	"github.com/milvus-io/milvus/internal/proto/internalpb"
 	"github.com/milvus-io/milvus/internal/proto/querypb"
 	"github.com/milvus-io/milvus/internal/types"
+	"github.com/milvus-io/milvus/pkg/util/merr"
 	"github.com/milvus-io/milvus/pkg/util/paramtable"
 )
 
@@ -270,6 +271,15 @@ func Test_NewServer(t *testing.T) {
 		assert.Equal(t, commonpb.ErrorCode_Success, resp.GetStatus().GetErrorCode())
 	})
 
+	t.Run("DeleteBatch", func(t *testing.T) {
+		mockQN.EXPECT().DeleteBatch(mock.Anything, mock.Anything).Return(&querypb.DeleteBatchResponse{
+			Status: merr.Success(),
+		}, nil)
+
+		resp, err := server.DeleteBatch(ctx, &querypb.DeleteBatchRequest{})
+		assert.NoError(t, merr.CheckRPCCall(resp, err))
+	})
+
 	err = server.Stop()
 	assert.NoError(t, err)
 }

internal/mocks/mock_querynode.go

@@ -85,6 +85,65 @@ func (_c *MockQueryNode_Delete_Call) RunAndReturn(run func(context.Context, *que
 	return _c
 }
 
+// DeleteBatch provides a mock function with given fields: _a0, _a1
+func (_m *MockQueryNode) DeleteBatch(_a0 context.Context, _a1 *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	ret := _m.Called(_a0, _a1)
+
+	if len(ret) == 0 {
+		panic("no return value specified for DeleteBatch")
+	}
+
+	var r0 *querypb.DeleteBatchResponse
+	var r1 error
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)); ok {
+		return rf(_a0, _a1)
+	}
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest) *querypb.DeleteBatchResponse); ok {
+		r0 = rf(_a0, _a1)
+	} else {
+		if ret.Get(0) != nil {
+			r0 = ret.Get(0).(*querypb.DeleteBatchResponse)
+		}
+	}
+
+	if rf, ok := ret.Get(1).(func(context.Context, *querypb.DeleteBatchRequest) error); ok {
+		r1 = rf(_a0, _a1)
+	} else {
+		r1 = ret.Error(1)
+	}
+
+	return r0, r1
+}
+
+// MockQueryNode_DeleteBatch_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'DeleteBatch'
+type MockQueryNode_DeleteBatch_Call struct {
+	*mock.Call
+}
+
+// DeleteBatch is a helper method to define mock.On call
+//   - _a0 context.Context
+//   - _a1 *querypb.DeleteBatchRequest
+func (_e *MockQueryNode_Expecter) DeleteBatch(_a0 interface{}, _a1 interface{}) *MockQueryNode_DeleteBatch_Call {
+	return &MockQueryNode_DeleteBatch_Call{Call: _e.mock.On("DeleteBatch", _a0, _a1)}
+}
+
+func (_c *MockQueryNode_DeleteBatch_Call) Run(run func(_a0 context.Context, _a1 *querypb.DeleteBatchRequest)) *MockQueryNode_DeleteBatch_Call {
+	_c.Call.Run(func(args mock.Arguments) {
+		run(args[0].(context.Context), args[1].(*querypb.DeleteBatchRequest))
+	})
+	return _c
+}
+
+func (_c *MockQueryNode_DeleteBatch_Call) Return(_a0 *querypb.DeleteBatchResponse, _a1 error) *MockQueryNode_DeleteBatch_Call {
+	_c.Call.Return(_a0, _a1)
+	return _c
+}
+
+func (_c *MockQueryNode_DeleteBatch_Call) RunAndReturn(run func(context.Context, *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)) *MockQueryNode_DeleteBatch_Call {
+	_c.Call.Return(run)
+	return _c
+}
+
 // GetAddress provides a mock function with given fields:
 func (_m *MockQueryNode) GetAddress() string {
 	ret := _m.Called()

internal/mocks/mock_querynode_client.go

@@ -142,6 +142,80 @@ func (_c *MockQueryNodeClient_Delete_Call) RunAndReturn(run func(context.Context
 	return _c
 }
 
+// DeleteBatch provides a mock function with given fields: ctx, in, opts
+func (_m *MockQueryNodeClient) DeleteBatch(ctx context.Context, in *querypb.DeleteBatchRequest, opts ...grpc.CallOption) (*querypb.DeleteBatchResponse, error) {
+	_va := make([]interface{}, len(opts))
+	for _i := range opts {
+		_va[_i] = opts[_i]
+	}
+	var _ca []interface{}
+	_ca = append(_ca, ctx, in)
+	_ca = append(_ca, _va...)
+	ret := _m.Called(_ca...)
+
+	if len(ret) == 0 {
+		panic("no return value specified for DeleteBatch")
+	}
+
+	var r0 *querypb.DeleteBatchResponse
+	var r1 error
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest, ...grpc.CallOption) (*querypb.DeleteBatchResponse, error)); ok {
+		return rf(ctx, in, opts...)
+	}
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest, ...grpc.CallOption) *querypb.DeleteBatchResponse); ok {
+		r0 = rf(ctx, in, opts...)
+	} else {
+		if ret.Get(0) != nil {
+			r0 = ret.Get(0).(*querypb.DeleteBatchResponse)
+		}
+	}
+
+	if rf, ok := ret.Get(1).(func(context.Context, *querypb.DeleteBatchRequest, ...grpc.CallOption) error); ok {
+		r1 = rf(ctx, in, opts...)
+	} else {
+		r1 = ret.Error(1)
+	}
+
+	return r0, r1
+}
+
+// MockQueryNodeClient_DeleteBatch_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'DeleteBatch'
+type MockQueryNodeClient_DeleteBatch_Call struct {
+	*mock.Call
+}
+
+// DeleteBatch is a helper method to define mock.On call
+//   - ctx context.Context
+//   - in *querypb.DeleteBatchRequest
+//   - opts ...grpc.CallOption
+func (_e *MockQueryNodeClient_Expecter) DeleteBatch(ctx interface{}, in interface{}, opts ...interface{}) *MockQueryNodeClient_DeleteBatch_Call {
+	return &MockQueryNodeClient_DeleteBatch_Call{Call: _e.mock.On("DeleteBatch",
+		append([]interface{}{ctx, in}, opts...)...)}
+}
+
+func (_c *MockQueryNodeClient_DeleteBatch_Call) Run(run func(ctx context.Context, in *querypb.DeleteBatchRequest, opts ...grpc.CallOption)) *MockQueryNodeClient_DeleteBatch_Call {
+	_c.Call.Run(func(args mock.Arguments) {
+		variadicArgs := make([]grpc.CallOption, len(args)-2)
+		for i, a := range args[2:] {
+			if a != nil {
+				variadicArgs[i] = a.(grpc.CallOption)
+			}
+		}
+		run(args[0].(context.Context), args[1].(*querypb.DeleteBatchRequest), variadicArgs...)
+	})
+	return _c
+}
+
+func (_c *MockQueryNodeClient_DeleteBatch_Call) Return(_a0 *querypb.DeleteBatchResponse, _a1 error) *MockQueryNodeClient_DeleteBatch_Call {
+	_c.Call.Return(_a0, _a1)
+	return _c
+}
+
+func (_c *MockQueryNodeClient_DeleteBatch_Call) RunAndReturn(run func(context.Context, *querypb.DeleteBatchRequest, ...grpc.CallOption) (*querypb.DeleteBatchResponse, error)) *MockQueryNodeClient_DeleteBatch_Call {
+	_c.Call.Return(run)
+	return _c
+}
+
 // GetComponentStates provides a mock function with given fields: ctx, in, opts
 func (_m *MockQueryNodeClient) GetComponentStates(ctx context.Context, in *milvuspb.GetComponentStatesRequest, opts ...grpc.CallOption) (*milvuspb.ComponentStates, error) {
 	_va := make([]interface{}, len(opts))

internal/proto/query_coord.proto

@@ -172,6 +172,10 @@ service QueryNode {
     }
     rpc Delete(DeleteRequest) returns (common.Status) {
     }
+    // DeleteBatch is the API to apply same delete data into multiple segments.
+    // it's basically same as `Delete` but cost less memory pressure.
+    rpc DeleteBatch(DeleteBatchRequest) returns (DeleteBatchResponse) {
+    } 
 }
 
 // --------------------QueryCoord grpc request and response proto------------------
@@ -772,6 +776,25 @@ message DeleteRequest {
     DataScope scope = 8;
 }
 
+message DeleteBatchRequest {
+    common.MsgBase base = 1;
+    int64 collection_id = 2;
+    int64 partition_id = 3;
+    string vchannel_name = 4;
+    repeated int64 segment_ids = 5;
+    schema.IDs primary_keys = 6;
+    repeated uint64 timestamps = 7;
+    DataScope scope = 8;
+}
+
+// DeleteBatchResponse returns failed/missing segment ids
+// cannot just using common.Status to handle partial failure logic
+message DeleteBatchResponse {
+    common.Status status = 1;
+    repeated int64 failed_ids = 2;
+    repeated int64 missing_ids = 3;
+}
+
 message ActivateCheckerRequest {
     common.MsgBase base = 1;
     int32 checkerID = 2;

internal/querycoordv2/mocks/mock_querynode.go

@@ -84,6 +84,65 @@ func (_c *MockQueryNodeServer_Delete_Call) RunAndReturn(run func(context.Context
 	return _c
 }
 
+// DeleteBatch provides a mock function with given fields: _a0, _a1
+func (_m *MockQueryNodeServer) DeleteBatch(_a0 context.Context, _a1 *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	ret := _m.Called(_a0, _a1)
+
+	if len(ret) == 0 {
+		panic("no return value specified for DeleteBatch")
+	}
+
+	var r0 *querypb.DeleteBatchResponse
+	var r1 error
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)); ok {
+		return rf(_a0, _a1)
+	}
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest) *querypb.DeleteBatchResponse); ok {
+		r0 = rf(_a0, _a1)
+	} else {
+		if ret.Get(0) != nil {
+			r0 = ret.Get(0).(*querypb.DeleteBatchResponse)
+		}
+	}
+
+	if rf, ok := ret.Get(1).(func(context.Context, *querypb.DeleteBatchRequest) error); ok {
+		r1 = rf(_a0, _a1)
+	} else {
+		r1 = ret.Error(1)
+	}
+
+	return r0, r1
+}
+
+// MockQueryNodeServer_DeleteBatch_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'DeleteBatch'
+type MockQueryNodeServer_DeleteBatch_Call struct {
+	*mock.Call
+}
+
+// DeleteBatch is a helper method to define mock.On call
+//   - _a0 context.Context
+//   - _a1 *querypb.DeleteBatchRequest
+func (_e *MockQueryNodeServer_Expecter) DeleteBatch(_a0 interface{}, _a1 interface{}) *MockQueryNodeServer_DeleteBatch_Call {
+	return &MockQueryNodeServer_DeleteBatch_Call{Call: _e.mock.On("DeleteBatch", _a0, _a1)}
+}
+
+func (_c *MockQueryNodeServer_DeleteBatch_Call) Run(run func(_a0 context.Context, _a1 *querypb.DeleteBatchRequest)) *MockQueryNodeServer_DeleteBatch_Call {
+	_c.Call.Run(func(args mock.Arguments) {
+		run(args[0].(context.Context), args[1].(*querypb.DeleteBatchRequest))
+	})
+	return _c
+}
+
+func (_c *MockQueryNodeServer_DeleteBatch_Call) Return(_a0 *querypb.DeleteBatchResponse, _a1 error) *MockQueryNodeServer_DeleteBatch_Call {
+	_c.Call.Return(_a0, _a1)
+	return _c
+}
+
+func (_c *MockQueryNodeServer_DeleteBatch_Call) RunAndReturn(run func(context.Context, *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)) *MockQueryNodeServer_DeleteBatch_Call {
+	_c.Call.Return(run)
+	return _c
+}
+
 // GetComponentStates provides a mock function with given fields: _a0, _a1
 func (_m *MockQueryNodeServer) GetComponentStates(_a0 context.Context, _a1 *milvuspb.GetComponentStatesRequest) (*milvuspb.ComponentStates, error) {
 	ret := _m.Called(_a0, _a1)

internal/querynodev2/cluster/mock_worker.go

@@ -69,6 +69,65 @@ func (_c *MockWorker_Delete_Call) RunAndReturn(run func(context.Context, *queryp
 	return _c
 }
 
+// DeleteBatch provides a mock function with given fields: ctx, req
+func (_m *MockWorker) DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	ret := _m.Called(ctx, req)
+
+	if len(ret) == 0 {
+		panic("no return value specified for DeleteBatch")
+	}
+
+	var r0 *querypb.DeleteBatchResponse
+	var r1 error
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)); ok {
+		return rf(ctx, req)
+	}
+	if rf, ok := ret.Get(0).(func(context.Context, *querypb.DeleteBatchRequest) *querypb.DeleteBatchResponse); ok {
+		r0 = rf(ctx, req)
+	} else {
+		if ret.Get(0) != nil {
+			r0 = ret.Get(0).(*querypb.DeleteBatchResponse)
+		}
+	}
+
+	if rf, ok := ret.Get(1).(func(context.Context, *querypb.DeleteBatchRequest) error); ok {
+		r1 = rf(ctx, req)
+	} else {
+		r1 = ret.Error(1)
+	}
+
+	return r0, r1
+}
+
+// MockWorker_DeleteBatch_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'DeleteBatch'
+type MockWorker_DeleteBatch_Call struct {
+	*mock.Call
+}
+
+// DeleteBatch is a helper method to define mock.On call
+//   - ctx context.Context
+//   - req *querypb.DeleteBatchRequest
+func (_e *MockWorker_Expecter) DeleteBatch(ctx interface{}, req interface{}) *MockWorker_DeleteBatch_Call {
+	return &MockWorker_DeleteBatch_Call{Call: _e.mock.On("DeleteBatch", ctx, req)}
+}
+
+func (_c *MockWorker_DeleteBatch_Call) Run(run func(ctx context.Context, req *querypb.DeleteBatchRequest)) *MockWorker_DeleteBatch_Call {
+	_c.Call.Run(func(args mock.Arguments) {
+		run(args[0].(context.Context), args[1].(*querypb.DeleteBatchRequest))
+	})
+	return _c
+}
+
+func (_c *MockWorker_DeleteBatch_Call) Return(_a0 *querypb.DeleteBatchResponse, _a1 error) *MockWorker_DeleteBatch_Call {
+	_c.Call.Return(_a0, _a1)
+	return _c
+}
+
+func (_c *MockWorker_DeleteBatch_Call) RunAndReturn(run func(context.Context, *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)) *MockWorker_DeleteBatch_Call {
+	_c.Call.Return(run)
+	return _c
+}
+
 // GetStatistics provides a mock function with given fields: ctx, req
 func (_m *MockWorker) GetStatistics(ctx context.Context, req *querypb.GetStatisticsRequest) (*internalpb.GetStatisticsResponse, error) {
 	ret := _m.Called(ctx, req)

internal/querynodev2/cluster/worker.go

@@ -39,6 +39,7 @@ type Worker interface {
 	LoadSegments(context.Context, *querypb.LoadSegmentsRequest) error
 	ReleaseSegments(context.Context, *querypb.ReleaseSegmentsRequest) error
 	Delete(ctx context.Context, req *querypb.DeleteRequest) error
+	DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error)
 	SearchSegments(ctx context.Context, req *querypb.SearchRequest) (*internalpb.SearchResults, error)
 	QuerySegments(ctx context.Context, req *querypb.QueryRequest) (*internalpb.RetrieveResults, error)
 	QueryStreamSegments(ctx context.Context, req *querypb.QueryRequest, srv streamrpc.QueryStreamServer) error
@@ -141,6 +142,52 @@ func (w *remoteWorker) Delete(ctx context.Context, req *querypb.DeleteRequest) e
 	return nil
 }
 
+func (w *remoteWorker) DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	log := log.Ctx(ctx).With(
+		zap.Int64("workerID", req.GetBase().GetTargetID()),
+	)
+	client := w.getClient()
+	resp, err := client.DeleteBatch(ctx, req)
+	if err := merr.CheckRPCCall(resp, err); err != nil {
+		if errors.Is(err, merr.ErrServiceUnimplemented) {
+			log.Warn("invoke legacy querynode DeleteBatch method, fallback to ")
+			return w.splitDeleteBatch(ctx, req)
+		}
+		return nil, err
+	}
+	return resp, nil
+}
+
+func (w *remoteWorker) splitDeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	sReq := &querypb.DeleteRequest{
+		CollectionId: req.GetCollectionId(),
+		PartitionId:  req.GetPartitionId(),
+		VchannelName: req.GetVchannelName(),
+		PrimaryKeys:  req.GetPrimaryKeys(),
+		Timestamps:   req.GetTimestamps(),
+		Scope:        req.GetScope(),
+	}
+	// do fallback without parallel, to protect the mem limit
+	var missingIDs []int64
+	var failedIDs []int64
+	for _, segmentID := range req.GetSegmentIds() {
+		sReq.SegmentId = segmentID
+		err := w.Delete(ctx, sReq)
+		switch {
+		case errors.Is(err, merr.ErrSegmentNotFound):
+			missingIDs = append(missingIDs, segmentID)
+		case err != nil:
+			failedIDs = append(failedIDs, segmentID)
+		default:
+		}
+	}
+	return &querypb.DeleteBatchResponse{
+		Status:     merr.Success(),
+		FailedIds:  failedIDs,
+		MissingIds: missingIDs,
+	}, nil
+}
+
 func (w *remoteWorker) SearchSegments(ctx context.Context, req *querypb.SearchRequest) (*internalpb.SearchResults, error) {
 	client := w.getClient()
 	ret, err := client.SearchSegments(ctx, req)

internal/querynodev2/cluster/worker_test.go

@@ -193,6 +193,70 @@ func (s *RemoteWorkerSuite) TestDelete() {
 	})
 }
 
+func (s *RemoteWorkerSuite) TestDeleteBatch() {
+	s.Run("normal_run", func() {
+		defer func() { s.mockClient.ExpectedCalls = nil }()
+
+		s.mockClient.EXPECT().DeleteBatch(mock.Anything, mock.AnythingOfType("*querypb.DeleteBatchRequest")).
+			Return(&querypb.DeleteBatchResponse{Status: merr.Success()}, nil).Once()
+
+		ctx, cancel := context.WithCancel(context.Background())
+		defer cancel()
+		resp, err := s.worker.DeleteBatch(ctx, &querypb.DeleteBatchRequest{
+			SegmentIds: []int64{100, 200},
+		})
+		s.NoError(merr.CheckRPCCall(resp, err))
+	})
+
+	s.Run("client_return_error", func() {
+		defer func() { s.mockClient.ExpectedCalls = nil }()
+
+		s.mockClient.EXPECT().DeleteBatch(mock.Anything, mock.AnythingOfType("*querypb.DeleteBatchRequest")).
+			Return(nil, merr.WrapErrServiceInternal("mocked")).Once()
+
+		ctx, cancel := context.WithCancel(context.Background())
+		defer cancel()
+		resp, err := s.worker.DeleteBatch(ctx, &querypb.DeleteBatchRequest{
+			SegmentIds: []int64{100, 200},
+		})
+
+		s.Error(merr.CheckRPCCall(resp, err))
+	})
+
+	s.Run("client_return_fail_status", func() {
+		defer func() { s.mockClient.ExpectedCalls = nil }()
+
+		s.mockClient.EXPECT().DeleteBatch(mock.Anything, mock.AnythingOfType("*querypb.DeleteBatchRequest")).
+			Return(&querypb.DeleteBatchResponse{
+				Status: merr.Status(merr.WrapErrServiceUnavailable("mocked")),
+			}, nil).Once()
+
+		ctx, cancel := context.WithCancel(context.Background())
+		defer cancel()
+		resp, err := s.worker.DeleteBatch(ctx, &querypb.DeleteBatchRequest{
+			SegmentIds: []int64{100, 200},
+		})
+
+		s.Error(merr.CheckRPCCall(resp, err))
+	})
+
+	s.Run("batch_delete_unimplemented", func() {
+		defer func() { s.mockClient.ExpectedCalls = nil }()
+
+		s.mockClient.EXPECT().DeleteBatch(mock.Anything, mock.AnythingOfType("*querypb.DeleteBatchRequest")).
+			Return(nil, merr.WrapErrServiceUnimplemented(status.Errorf(codes.Unimplemented, "mocked grpc unimplemented")))
+		s.mockClient.EXPECT().Delete(mock.Anything, mock.Anything).Return(merr.Success(), nil).Times(2)
+
+		ctx, cancel := context.WithCancel(context.Background())
+		defer cancel()
+		resp, err := s.worker.DeleteBatch(ctx, &querypb.DeleteBatchRequest{
+			SegmentIds: []int64{100, 200},
+		})
+
+		s.NoError(merr.CheckRPCCall(resp, err))
+	})
+}
+
 func (s *RemoteWorkerSuite) TestSearch() {
 	s.Run("normal_run", func() {
 		defer func() { s.mockClient.ExpectedCalls = nil }()

internal/querynodev2/delegator/delta_forward.go

@@ -19,8 +19,10 @@ package delegator
 import (
 	"context"
 	"fmt"
+	"runtime"
 	"time"
 
+	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.uber.org/zap"
 	"golang.org/x/sync/errgroup"
@@ -36,8 +38,10 @@ import (
 	"github.com/milvus-io/milvus/pkg/log"
 	"github.com/milvus-io/milvus/pkg/metrics"
 	"github.com/milvus-io/milvus/pkg/util/commonpbutil"
+	"github.com/milvus-io/milvus/pkg/util/conc"
 	"github.com/milvus-io/milvus/pkg/util/merr"
 	"github.com/milvus-io/milvus/pkg/util/paramtable"
+	"github.com/milvus-io/milvus/pkg/util/retry"
 	"github.com/milvus-io/milvus/pkg/util/typeutil"
 )
 
@@ -276,50 +280,43 @@ func (sd *shardDelegator) forwardStreamingDirect(ctx context.Context, deleteData
 			sealed, growing, version := sd.distribution.PinOnlineSegments(partitions...)
 			defer sd.distribution.Unpin(version)
 
-			for _, item := range group {
-				deleteData := *item
-				for _, entry := range sealed {
-					entry := entry
-					worker, err := sd.workerManager.GetWorker(ctx, entry.NodeID)
-					if err != nil {
-						log.Warn("failed to get worker",
-							zap.Int64("nodeID", entry.NodeID),
-							zap.Error(err),
-						)
-						// skip if node down
-						// delete will be processed after loaded again
-						continue
-					}
-					// forward to non level0 segment only
-					segments := lo.Filter(entry.Segments, func(segmentEntry SegmentEntry, _ int) bool {
-						return segmentEntry.Level != datapb.SegmentLevel_L0
-					})
-
-					eg.Go(func() error {
-						offlineSegments.Upsert(sd.applyDelete(ctx, entry.NodeID, worker, func(segmentID int64) (DeleteData, bool) {
-							return deleteData, true
-						}, segments, querypb.DataScope_Historical)...)
-						return nil
-					})
+			for _, entry := range sealed {
+				entry := entry
+				worker, err := sd.workerManager.GetWorker(ctx, entry.NodeID)
+				if err != nil {
+					log.Warn("failed to get worker",
+						zap.Int64("nodeID", entry.NodeID),
+						zap.Error(err),
+					)
+					// skip if node down
+					// delete will be processed after loaded again
+					continue
 				}
+				// forward to non level0 segment only
+				segments := lo.Filter(entry.Segments, func(segmentEntry SegmentEntry, _ int) bool {
+					return segmentEntry.Level != datapb.SegmentLevel_L0
+				})
 
-				if len(growing) > 0 {
-					worker, err := sd.workerManager.GetWorker(ctx, paramtable.GetNodeID())
-					if err != nil {
-						log.Error("failed to get worker(local)",
-							zap.Int64("nodeID", paramtable.GetNodeID()),
-							zap.Error(err),
-						)
-						// panic here, local worker shall not have error
-						panic(err)
-					}
-					eg.Go(func() error {
-						offlineSegments.Upsert(sd.applyDelete(ctx, paramtable.GetNodeID(), worker, func(segmentID int64) (DeleteData, bool) {
-							return deleteData, true
-						}, growing, querypb.DataScope_Streaming)...)
-						return nil
-					})
+				eg.Go(func() error {
+					offlineSegments.Upsert(sd.applyDeleteBatch(ctx, entry.NodeID, worker, group, segments, querypb.DataScope_Historical)...)
+					return nil
+				})
+			}
+
+			if len(growing) > 0 {
+				worker, err := sd.workerManager.GetWorker(ctx, paramtable.GetNodeID())
+				if err != nil {
+					log.Error("failed to get worker(local)",
+						zap.Int64("nodeID", paramtable.GetNodeID()),
+						zap.Error(err),
+					)
+					// panic here, local worker shall not have error
+					panic(err)
 				}
+				eg.Go(func() error {
+					offlineSegments.Upsert(sd.applyDeleteBatch(ctx, paramtable.GetNodeID(), worker, group, growing, querypb.DataScope_Streaming)...)
+					return nil
+				})
 			}
 			return nil
 		})
@@ -336,3 +333,72 @@ func (sd *shardDelegator) forwardStreamingDirect(ctx context.Context, deleteData
 
 	metrics.QueryNodeForwardDeleteCost.WithLabelValues("ProcessDelete", fmt.Sprint(paramtable.GetNodeID())).Observe(float64(forwardDeleteCost.Milliseconds()))
 }
+
+// applyDeleteBatch handles delete record and apply them to corresponding workers in batch.
+func (sd *shardDelegator) applyDeleteBatch(ctx context.Context,
+	nodeID int64,
+	worker cluster.Worker,
+	data []*DeleteData,
+	entries []SegmentEntry,
+	scope querypb.DataScope,
+) []int64 {
+	offlineSegments := typeutil.NewConcurrentSet[int64]()
+	log := sd.getLogger(ctx)
+
+	ctx, cancel := context.WithCancel(ctx)
+	defer cancel()
+
+	pool := conc.NewPool[struct{}](runtime.GOMAXPROCS(0) * 4)
+	defer pool.Release()
+
+	var futures []*conc.Future[struct{}]
+	for _, delData := range data {
+		delData := delData
+		segmentIDs := lo.Map(entries, func(entry SegmentEntry, _ int) int64 {
+			return entry.SegmentID
+		})
+		future := pool.Submit(func() (struct{}, error) {
+			log.Debug("delegator plan to applyDelete via worker")
+			err := retry.Handle(ctx, func() (bool, error) {
+				if sd.Stopped() {
+					return false, merr.WrapErrChannelNotAvailable(sd.vchannelName, "channel is unsubscribing")
+				}
+
+				resp, err := worker.DeleteBatch(ctx, &querypb.DeleteBatchRequest{
+					Base:         commonpbutil.NewMsgBase(commonpbutil.WithTargetID(nodeID)),
+					CollectionId: sd.collectionID,
+					PartitionId:  delData.PartitionID,
+					VchannelName: sd.vchannelName,
+					SegmentIds:   segmentIDs,
+					PrimaryKeys:  storage.ParsePrimaryKeys2IDs(delData.PrimaryKeys),
+					Timestamps:   delData.Timestamps,
+					Scope:        scope,
+				})
+				if errors.Is(err, merr.ErrNodeNotFound) {
+					log.Warn("try to delete data on non-exist node")
+					// cancel other request
+					cancel()
+					return false, err
+				}
+				// grpc/network error
+				if err != nil {
+					return true, err
+				}
+				if len(resp.GetMissingIds()) > 0 {
+					log.Warn("try to delete data of released segment", zap.Int64s("ids", resp.GetMissingIds()))
+				}
+				if len(resp.GetFailedIds()) > 0 {
+					log.Warn("apply delete for segment failed, marking it offline")
+					offlineSegments.Upsert(resp.GetFailedIds()...)
+				}
+				return false, nil
+			}, retry.Attempts(10))
+
+			return struct{}{}, err
+		})
+		futures = append(futures, future)
+	}
+
+	conc.AwaitAll(futures...)
+	return offlineSegments.Collect()
+}

internal/querynodev2/delegator/delta_forward_test.go

@@ -35,6 +35,7 @@ import (
 	"github.com/milvus-io/milvus/internal/storage"
 	"github.com/milvus-io/milvus/pkg/common"
 	"github.com/milvus-io/milvus/pkg/mq/msgstream"
+	"github.com/milvus-io/milvus/pkg/util/merr"
 	"github.com/milvus-io/milvus/pkg/util/metric"
 	"github.com/milvus-io/milvus/pkg/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/util/typeutil"
@@ -243,12 +244,11 @@ func (s *StreamingForwardSuite) TestDirectStreamingForward() {
 	deletedSegment := typeutil.NewConcurrentSet[int64]()
 	mockWorker := cluster.NewMockWorker(s.T())
 	s.workerManager.EXPECT().GetWorker(mock.Anything, int64(1)).Return(mockWorker, nil)
-	mockWorker.EXPECT().Delete(mock.Anything, mock.Anything).RunAndReturn(func(ctx context.Context, dr *querypb.DeleteRequest) error {
-		s.T().Log(dr.GetSegmentId())
-		deletedSegment.Insert(dr.SegmentId)
+	mockWorker.EXPECT().DeleteBatch(mock.Anything, mock.Anything).RunAndReturn(func(ctx context.Context, dr *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+		deletedSegment.Upsert(dr.GetSegmentIds()...)
 		s.ElementsMatch([]int64{10}, dr.GetPrimaryKeys().GetIntId().GetData())
 		s.ElementsMatch([]uint64{10}, dr.GetTimestamps())
-		return nil
+		return &querypb.DeleteBatchResponse{Status: merr.Success()}, nil
 	})
 
 	delegator.ProcessDelete([]*DeleteData{

internal/querynodev2/local_worker.go

@@ -53,6 +53,10 @@ func (w *LocalWorker) Delete(ctx context.Context, req *querypb.DeleteRequest) er
 	return merr.CheckRPCCall(status, err)
 }
 
+func (w *LocalWorker) DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	return w.node.DeleteBatch(ctx, req)
+}
+
 func (w *LocalWorker) SearchSegments(ctx context.Context, req *querypb.SearchRequest) (*internalpb.SearchResults, error) {
 	return w.node.SearchSegments(ctx, req)
 }

internal/querynodev2/segments/manager.go

@@ -87,6 +87,22 @@ func (f SegmentIDFilter) SegmentIDs() ([]int64, bool) {
 	return []int64{int64(f)}, true
 }
 
+type SegmentIDsFilter struct {
+	segmentIDs typeutil.Set[int64]
+}
+
+func (f SegmentIDsFilter) Filter(segment Segment) bool {
+	return f.segmentIDs.Contain(segment.ID())
+}
+
+func (f SegmentIDsFilter) SegmentType() (SegmentType, bool) {
+	return commonpb.SegmentState_SegmentStateNone, false
+}
+
+func (f SegmentIDsFilter) SegmentIDs() ([]int64, bool) {
+	return f.segmentIDs.Collect(), true
+}
+
 type SegmentTypeFilter SegmentType
 
 func (f SegmentTypeFilter) Filter(segment Segment) bool {
@@ -133,6 +149,12 @@ func WithID(id int64) SegmentFilter {
 	return SegmentIDFilter(id)
 }
 
+func WithIDs(ids ...int64) SegmentFilter {
+	return SegmentIDsFilter{
+		segmentIDs: typeutil.NewSet(ids...),
+	}
+}
+
 func WithLevel(level datapb.SegmentLevel) SegmentFilter {
 	return SegmentFilterFunc(func(segment Segment) bool {
 		return segment.Level() == level

internal/querynodev2/services.go

@@ -19,6 +19,7 @@ package querynodev2
 import (
 	"context"
 	"fmt"
+	"runtime"
 	"strconv"
 	"sync"
 	"time"
@@ -47,6 +48,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/log"
 	"github.com/milvus-io/milvus/pkg/metrics"
 	"github.com/milvus-io/milvus/pkg/util/commonpbutil"
+	"github.com/milvus-io/milvus/pkg/util/conc"
 	"github.com/milvus-io/milvus/pkg/util/funcutil"
 	"github.com/milvus-io/milvus/pkg/util/hardware"
 	"github.com/milvus-io/milvus/pkg/util/merr"
@@ -1426,6 +1428,84 @@ func (node *QueryNode) Delete(ctx context.Context, req *querypb.DeleteRequest) (
 	return merr.Success(), nil
 }
 
+// DeleteBatch is the API to apply same delete data into multiple segments.
+// it's basically same as `Delete` but cost less memory pressure.
+func (node *QueryNode) DeleteBatch(ctx context.Context, req *querypb.DeleteBatchRequest) (*querypb.DeleteBatchResponse, error) {
+	log := log.Ctx(ctx).With(
+		zap.Int64("collectionID", req.GetCollectionId()),
+		zap.String("channel", req.GetVchannelName()),
+		zap.Int64s("segmentIDs", req.GetSegmentIds()),
+		zap.String("scope", req.GetScope().String()),
+	)
+
+	// check node healthy
+	if err := node.lifetime.Add(merr.IsHealthy); err != nil {
+		return &querypb.DeleteBatchResponse{
+			Status: merr.Status(err),
+		}, nil
+	}
+	defer node.lifetime.Done()
+
+	// log.Debug("QueryNode received worker delete detail", zap.Stringer("info", &deleteRequestStringer{DeleteRequest: req}))
+
+	filters := []segments.SegmentFilter{
+		segments.WithIDs(req.GetSegmentIds()...),
+	}
+
+	// do not add filter for Unknown & All scope, for backward cap
+	switch req.GetScope() {
+	case querypb.DataScope_Historical:
+		filters = append(filters, segments.WithType(segments.SegmentTypeSealed))
+	case querypb.DataScope_Streaming:
+		filters = append(filters, segments.WithType(segments.SegmentTypeGrowing))
+	}
+
+	segs := node.manager.Segment.GetBy(filters...)
+
+	hitIDs := lo.Map(segs, func(segment segments.Segment, _ int) int64 {
+		return segment.ID()
+	})
+	// calculate missing ids, continue to delete existing ones.
+	missingIDs := typeutil.NewSet(req.GetSegmentIds()...).Complement(typeutil.NewSet(hitIDs...))
+	if missingIDs.Len() > 0 {
+		log.Warn("Delete batch find missing ids", zap.Int64s("missing_ids", missingIDs.Collect()))
+	}
+
+	pks := storage.ParseIDs2PrimaryKeys(req.GetPrimaryKeys())
+
+	// control the execution batch parallel with P number
+	// maybe it shall be lower in case of heavy CPU usage may impacting search/query
+	pool := conc.NewPool[struct{}](runtime.GOMAXPROCS(0))
+	futures := make([]*conc.Future[struct{}], 0, len(segs))
+	errSet := typeutil.NewConcurrentSet[int64]()
+
+	for _, segment := range segs {
+		segment := segment
+		futures = append(futures, pool.Submit(func() (struct{}, error) {
+			// TODO @silverxia, add interface to use same data struct for segment delete
+			// current implementation still copys pks into protobuf(or arrow) struct
+			err := segment.Delete(ctx, pks, req.GetTimestamps())
+			if err != nil {
+				errSet.Insert(segment.ID())
+				log.Warn("segment delete failed",
+					zap.Int64("segmentID", segment.ID()),
+					zap.Error(err))
+				return struct{}{}, err
+			}
+			return struct{}{}, nil
+		}))
+	}
+
+	// ignore error returned, since error segment is recorded into error set
+	_ = conc.AwaitAll(futures...)
+
+	// return merr.Success(), nil
+	return &querypb.DeleteBatchResponse{
+		Status:    merr.Success(),
+		FailedIds: errSet.Collect(),
+	}, nil
+}
+
 type deleteRequestStringer struct {
 	*querypb.DeleteRequest
 }

internal/util/mock/grpc_querynode_client.go

@@ -130,6 +130,10 @@ func (m *GrpcQueryNodeClient) Delete(ctx context.Context, in *querypb.DeleteRequ
 	return &commonpb.Status{}, m.Err
 }
 
+func (m *GrpcQueryNodeClient) DeleteBatch(ctx context.Context, in *querypb.DeleteBatchRequest, opts ...grpc.CallOption) (*querypb.DeleteBatchResponse, error) {
+	return &querypb.DeleteBatchResponse{}, m.Err
+}
+
 func (m *GrpcQueryNodeClient) Close() error {
 	return m.Err
 }

internal/util/wrappers/qn_wrapper.go

@@ -148,6 +148,10 @@ func (qn *qnServerWrapper) Delete(ctx context.Context, in *querypb.DeleteRequest
 	return qn.QueryNode.Delete(ctx, in)
 }
 
+func (qn *qnServerWrapper) DeleteBatch(ctx context.Context, in *querypb.DeleteBatchRequest, opts ...grpc.CallOption) (*querypb.DeleteBatchResponse, error) {
+	return qn.QueryNode.DeleteBatch(ctx, in)
+}
+
 func WrapQueryNodeServerAsClient(qn types.QueryNode) types.QueryNodeClient {
 	return &qnServerWrapper{
 		QueryNode: qn,

internal/util/wrappers/qn_wrapper_test.go

@@ -246,6 +246,17 @@ func (s *QnWrapperSuite) TestDelete() {
 	s.NoError(err)
 }
 
+func (s *QnWrapperSuite) TestDeleteBatch() {
+	s.qn.EXPECT().DeleteBatch(mock.Anything, mock.Anything).
+		Return(&querypb.DeleteBatchResponse{
+			Status: merr.Status(nil),
+		}, nil)
+
+	resp, err := s.client.DeleteBatch(context.Background(), &querypb.DeleteBatchRequest{})
+	err = merr.CheckRPCCall(resp, err)
+	s.NoError(err)
+}
+
 // Race caused by mock parameter check on once
 /*
 func (s *QnWrapperSuite) TestQueryStream() {