diff --git a/server/etcdserver/server.go b/server/etcdserver/server.go
index b18c1cce74e..c73d4776985 100644
--- a/server/etcdserver/server.go
+++ b/server/etcdserver/server.go
@@ -2064,7 +2064,6 @@ func (s *EtcdServer) applyConfChange(cc raftpb.ConfChange, confState *raftpb.Con
 
 // TODO: non-blocking snapshot
 func (s *EtcdServer) snapshot(snapi uint64, confState raftpb.ConfState) {
-	clone := s.v2store.Clone()
 	// commit kv to write metadata (for example: consistent index) to disk.
 	//
 	// This guarantees that Backend's consistent_index is >= index of last snapshot.
@@ -2075,16 +2074,12 @@ func (s *EtcdServer) snapshot(snapi uint64, confState raftpb.ConfState) {
 	// So KV().Commit() cannot run in parallel with toApply. It has to be called outside
 	// the go routine created below.
 	s.KV().Commit()
+	d := GetMembershipInfoInV2Format(s.Logger(), s.cluster)
 
 	s.GoAttach(func() {
 		lg := s.Logger()
 
-		d, err := clone.SaveNoCopy()
-		// TODO: current store will never fail to do a snapshot
-		// what should we do if the store might fail?
-		if err != nil {
-			lg.Panic("failed to save v2 store", zap.Error(err))
-		}
+		// For backward compatibility, generate v2 snapshot from v3 state.
 		snap, err := s.r.raftStorage.CreateSnapshot(snapi, &confState, d)
 		if err != nil {
 			// the snapshot was done asynchronously with the progress of raft.
diff --git a/server/etcdserver/server_test.go b/server/etcdserver/server_test.go
index 54879285132..c6f68b97909 100644
--- a/server/etcdserver/server_test.go
+++ b/server/etcdserver/server_test.go
@@ -1047,7 +1047,10 @@ func TestSnapshot(t *testing.T) {
 	srv.kv = mvcc.New(zaptest.NewLogger(t), be, &lease.FakeLessor{}, mvcc.StoreConfig{})
 	srv.be = be
 
-	ch := make(chan struct{}, 2)
+	cl := membership.NewCluster(zaptest.NewLogger(t))
+	srv.cluster = cl
+
+	ch := make(chan struct{}, 1)
 
 	go func() {
 		gaction, _ := p.Wait(2)
@@ -1066,24 +1069,11 @@ func TestSnapshot(t *testing.T) {
 		}
 	}()
 
-	go func() {
-		gaction, _ := st.Wait(2)
-		defer func() { ch <- struct{}{} }()
-
-		if len(gaction) != 2 {
-			t.Errorf("len(action) = %d, want 2", len(gaction))
-		}
-		if !reflect.DeepEqual(gaction[0], testutil.Action{Name: "Clone"}) {
-			t.Errorf("action = %s, want Clone", gaction[0])
-		}
-		if !reflect.DeepEqual(gaction[1], testutil.Action{Name: "SaveNoCopy"}) {
-			t.Errorf("action = %s, want SaveNoCopy", gaction[1])
-		}
-	}()
-
 	srv.snapshot(1, raftpb.ConfState{Voters: []uint64{1}})
 	<-ch
-	<-ch
+	if len(st.Action()) != 0 {
+		t.Errorf("no action expected on v2store. Got %d actions", len(st.Action()))
+	}
 }
 
 // TestSnapshotOrdering ensures raft persists snapshot onto disk before
@@ -1098,7 +1088,8 @@ func TestSnapshotOrdering(t *testing.T) {
 	n := newNopReadyNode()
 	st := v2store.New()
 	cl := membership.NewCluster(lg)
-	cl.SetStore(st)
+	be, _ := betesting.NewDefaultTmpBackend(t)
+	cl.SetBackend(schema.NewMembershipBackend(lg, be))
 
 	testdir := t.TempDir()
 
@@ -1118,7 +1109,6 @@ func TestSnapshotOrdering(t *testing.T) {
 		storage:     p,
 		raftStorage: rs,
 	})
-	be, _ := betesting.NewDefaultTmpBackend(t)
 	ci := cindex.NewConsistentIndex(be)
 	s := &EtcdServer{
 		lgMu:         new(sync.RWMutex),
@@ -1211,6 +1201,9 @@ func TestTriggerSnap(t *testing.T) {
 	srv.kv = mvcc.New(zaptest.NewLogger(t), be, &lease.FakeLessor{}, mvcc.StoreConfig{})
 	srv.be = be
 
+	cl := membership.NewCluster(zaptest.NewLogger(t))
+	srv.cluster = cl
+
 	srv.start()
 
 	donec := make(chan struct{})
diff --git a/server/etcdserver/snapshot_merge.go b/server/etcdserver/snapshot_merge.go
index 963ead5a7e2..5afbc626c18 100644
--- a/server/etcdserver/snapshot_merge.go
+++ b/server/etcdserver/snapshot_merge.go
@@ -31,11 +31,7 @@ import (
 func (s *EtcdServer) createMergedSnapshotMessage(m raftpb.Message, snapt, snapi uint64, confState raftpb.ConfState) snap.Message {
 	lg := s.Logger()
 	// get a snapshot of v2 store as []byte
-	clone := s.v2store.Clone()
-	d, err := clone.SaveNoCopy()
-	if err != nil {
-		lg.Panic("failed to save v2 store data", zap.Error(err))
-	}
+	d := GetMembershipInfoInV2Format(lg, s.cluster)
 
 	// commit kv to write metadata(for example: consistent index).
 	s.KV().Commit()