diff --git a/raft/core/core_handle.go b/raft/core/core_handle.go
index cba4178..4a5d5b4 100644
--- a/raft/core/core_handle.go
+++ b/raft/core/core_handle.go
@@ -97,6 +97,9 @@ func (c *core) handleReadIndexRequest(msg *raftpd.Message) {
 }
 
 func (c *core) handleReadIndexResponse(msg *raftpd.Message) {
+	log.Debugf("%d [Term: %d, commit: %d] receive read index response from: %d [idx: %d]",
+		c.id, c.term, msg.From, msg.Index)
+
 	// TODO: check
 	readState := read.ReadState{
 		Index:      msg.Index,
@@ -115,7 +118,7 @@ func (c *core) handleAppendEntries(msg *raftpd.Message) {
 		To:      msg.From,
 	}
 	if c.log.CommitIndex() > msg.LogIndex {
-		log.Infof("%d [Term: %d, commit: %d] reject expired append Entries "+
+		log.Debugf("%d [Term: %d, commit: %d] reject expired append Entries "+
 			"from %d [logterm: %d, idx: %d]", c.id, c.term, c.log.CommitIndex(),
 			msg.From, msg.LogTerm, msg.LogIndex)
 
@@ -125,7 +128,7 @@ func (c *core) handleAppendEntries(msg *raftpd.Message) {
 		reply.RejectHint = c.log.CommitIndex()
 		reply.Reject = false
 	} else if idx, ok := c.log.TryAppend(msg.LogIndex, msg.LogTerm, msg.Entries); ok {
-		log.Infof("%d [Term: %d, commit: %d] accept append Entries "+
+		log.Debugf("%d [Term: %d, commit: %d] accept append Entries "+
 			"from %d [logterm: %d, idx: %d], last: %d", c.id, c.term, c.log.CommitIndex(),
 			msg.From, msg.LogTerm, msg.LogIndex, idx)
 
@@ -136,7 +139,7 @@ func (c *core) handleAppendEntries(msg *raftpd.Message) {
 		reply.RejectHint = idx /* idx is index of latest log entry */
 		reply.Reject = false
 	} else {
-		log.Infof("%d [logterm: %d, commit: %d, last idx: %d] rejected msgApp "+
+		log.Debugf("%d [logterm: %d, commit: %d, last idx: %d] rejected msgApp "+
 			"[logterm: %d, idx: %d] and hint %d from %d", c.id, c.log.Term(msg.LogIndex),
 			c.log.CommitIndex(), c.log.LastIndex(), msg.LogTerm, msg.LogIndex, idx, msg.From)
 
@@ -181,7 +184,7 @@ func (c *core) handleSnapshot(msg *raftpd.Message) {
 		Reject:  false,
 	}
 	if c.tryRestore(msg.Snapshot) {
-		log.Infof("%x [commit: %d] restore snapshot [index: %d, term: %d]",
+		log.Debugf("%x [commit: %d] restore snapshot [index: %d, term: %d]",
 			c.id, c.log.CommitIndex(),
 			msg.Snapshot.Metadata.Index, msg.Snapshot.Metadata.Term)
 
@@ -192,7 +195,7 @@ func (c *core) handleSnapshot(msg *raftpd.Message) {
 		reply.Index = msg.Snapshot.Metadata.Index
 		reply.RejectHint = c.log.LastIndex()
 	} else {
-		log.Infof("%x [commit: %d] ignored snapshot [index: %d, term: %d]",
+		log.Debugf("%x [commit: %d] ignored snapshot [index: %d, term: %d]",
 			c.id, c.log.CommitIndex(),
 			msg.Snapshot.Metadata.Index, msg.Snapshot.Metadata.Term)
 
@@ -219,6 +222,8 @@ func (c *core) handleUnreachable(msg *raftpd.Message) {
 }
 
 func (c *core) handleHeartbeat(msg *raftpd.Message) {
+	log.Debugf("%d [term: %d] handle heartbeat request from %d", c.id, c.term, msg.From)
+
 	c.log.CommitTo(msg.Index)
 
 	reply := raftpd.Message{
@@ -231,14 +236,19 @@ func (c *core) handleHeartbeat(msg *raftpd.Message) {
 }
 
 func (c *core) handleHeartbeatResponse(msg *raftpd.Message) {
+	log.Debugf("%d [term: %d] handle heartbeat response from %d", c.id, c.term, msg.From)
 	ackCount := c.readOnly.ReceiveAck(msg.From, msg.Context)
 	if ackCount < c.quorum() {
 		return
 	}
+	log.Debugf("%d [term: %d] handle heartbeat response from %d", c.id, c.term, msg.From)
 
 	rss := c.readOnly.Advance(msg.Context)
 	for _, rs := range rss {
 		if rs.To == c.id {
+			log.Debugf("%d [term: %d] save read state: %d, %v",
+				c.id, c.term, rs.Index, rs.Context)
+
 			readState := read.ReadState{
 				Index:      rs.Index,
 				RequestCtx: rs.Context,
@@ -246,6 +256,9 @@ func (c *core) handleHeartbeatResponse(msg *raftpd.Message) {
 
 			c.callback.saveReadState(&readState)
 		} else {
+			log.Debugf("%d [term: %d] redirect heartbeat response %d to %d %v",
+				c.id, c.term, rs.Index, rs.To, rs.Context)
+
 			redirect := raftpd.Message{
 				To:      rs.To,
 				MsgType: raftpd.MsgReadIndexResponse,
@@ -359,6 +372,7 @@ func (c *core) handleVoteResponse(msg *raftpd.Message) {
 }
 
 func (c *core) broadcastHeartbeatWithCtx(context []byte) {
+	log.Debugf("%d [Term: %d] begin broadcast heartbeat with context: %v", c.id, c.term, context)
 	for i := 0; i < len(c.nodes); i++ {
 		c.sendHeartbeat(c.nodes[i], context)
 	}
@@ -445,7 +459,7 @@ func (c *core) sendAppend(node *peer.Node) {
 		}
 	}
 
-	log.Infof("%d [Term: %d] send append [idx: %d, Term: %d, len: %d] "+
+	log.Debugf("%d [Term: %d] send append [idx: %d, Term: %d, len: %d] "+
 		"to node: %d [matched: %d next: %d]",
 		c.id, c.term, msg.LogIndex, msg.LogTerm, len(msg.Entries),
 		node.ID, node.Matched, node.NextIdx)
@@ -462,18 +476,18 @@ func (c *core) sendSnapshot(node *peer.Node) {
 	// if snapshot is building at now, it will return nil,
 	// so just ignore it and send message to it on next tick.
 	if snapshot == nil {
-		log.Infof("%x failed to send snapshot to %x because snapshot "+
+		log.Debugf("%x failed to send snapshot to %x because snapshot "+
 			"is temporarily unavailable", c.id, node.ID)
 		return
 	}
 
-	log.Infof("%x [firstIdx: %d, commit: %d] send "+
+	log.Debugf("%x [firstIdx: %d, commit: %d] send "+
 		"snapshot[index: %d, term: %d] to %x", c.id, c.log.FirstIndex(), c.log.CommitIndex(),
 		snapshot.Metadata.Index, snapshot.Metadata.Term, node.ID)
 
 	node.SendSnapshot(snapshot.Metadata.Index)
 
-	log.Infof("%x paused sending replication messages to %x", c.id, node.ID)
+	log.Debugf("%x paused sending replication messages to %x", c.id, node.ID)
 
 	msg := raftpd.Message{
 		MsgType:  raftpd.MsgSnapshotRequest,
diff --git a/raft/core/core_internal.go b/raft/core/core_internal.go
index 0e3e188..5d958af 100644
--- a/raft/core/core_internal.go
+++ b/raft/core/core_internal.go
@@ -61,9 +61,9 @@ func (c *core) becomeFollower(term, leaderID uint64) {
 	c.vote = leaderID
 
 	if leaderID != conf.InvalidID {
-		log.Infof("%v become %d's follower at %d", c.id, leaderID, c.term)
+		log.Debugf("%v become %d's follower at %d", c.id, leaderID, c.term)
 	} else {
-		log.Infof("%v become follower at %d, without leader", c.id, c.term)
+		log.Debugf("%v become follower at %d, without leader", c.id, c.term)
 	}
 }
 
@@ -83,7 +83,7 @@ func (c *core) becomeLeader() {
 
 	utils.Assert(c.vote == c.id, "leader will vote itself")
 
-	log.Infof("%v become leader at %d [firstIdx: %d, lastIdx: %d]",
+	log.Debugf("%v become leader at %d [firstIdx: %d, lastIdx: %d]",
 		c.id, c.term, c.log.FirstIndex(), c.log.LastIndex())
 }
 
@@ -97,7 +97,7 @@ func (c *core) becomeCandidate() {
 
 	c.resetNodesVoteState()
 
-	log.Infof("%v become candidate at %d", c.id, c.term)
+	log.Debugf("%v become candidate at %d", c.id, c.term)
 }
 
 func (c *core) becomePreCandidate() {
@@ -112,7 +112,7 @@ func (c *core) becomePreCandidate() {
 	// Becoming a pre-candidate changes our state,
 	// but doesn't change anything else. In particular it does not increase
 	// currentTerm or change votedFor.
-	log.Infof("%x became pre-candidate at term %d", c.id, c.term)
+	log.Debugf("%x became pre-candidate at term %d", c.id, c.term)
 }
 
 func (c *core) preCampaign() {
diff --git a/raft/core/raw_node.go b/raft/core/raw_node.go
index e524cfa..0b767d7 100644
--- a/raft/core/raw_node.go
+++ b/raft/core/raw_node.go
@@ -104,7 +104,6 @@ func (node *RawNode) Ready() Ready {
 	// clear all
 	node.commitEntries = make([]raftpd.Entry, 0)
 	node.messages = make([]raftpd.Message, 0)
-	node.readStates = make([]read.ReadState, 0)
 
 	return ready
 }
@@ -138,18 +137,22 @@ func (node *RawNode) readSnapshot() *raftpd.Snapshot {
 
 func (node *RawNode) drainReadState() []read.ReadState {
 	var readStates []read.ReadState
+	lastApplied := node.prevHS.Commit
 	if len(node.commitEntries) > 0 {
-		lastApplied := node.commitEntries[len(node.commitEntries)-1].Index
-		for i := 0; i < len(node.readStates); i++ {
-			if node.readStates[i].Index <= lastApplied {
-				continue
-			}
-			//save and drain read states.
-			readStates = make([]read.ReadState, 0, i)
-			copy(readStates, node.readStates)
-			length := copy(node.readStates[:i], node.readStates[i:])
-			node.readStates = node.readStates[:length]
+		waitCommitIdx := node.commitEntries[len(node.commitEntries)-1].Index
+		if lastApplied < waitCommitIdx {
+			lastApplied = waitCommitIdx
 		}
 	}
+	i := 0
+	for ; i < len(node.readStates); i++ {
+		if node.readStates[i].Index > lastApplied {
+			break
+		}
+	}
+	//save and drain read states.
+	readStates = make([]read.ReadState, i)
+	copy(readStates, node.readStates)
+	node.readStates = node.readStates[i:]
 	return readStates
 }