Fix peer timer race condition.

peer.go

@@ -35,8 +35,10 @@ func NewPeer(server *Server, name string, heartbeatTimeout time.Duration) *Peer
 		heartbeatTimer: NewTimer(heartbeatTimeout, heartbeatTimeout),
 	}
 
-	// Start the heartbeat timeout.
+	// Start the heartbeat timeout and wait for the goroutine to start.
-	go p.heartbeatTimeoutFunc()
+	c := make(chan bool)
+	go p.heartbeatTimeoutFunc(c)
+	<-c
 	
 	return p
 }
@@ -149,10 +151,8 @@ func (p *Peer) sendFlushRequest(req *AppendEntriesRequest) (uint64, bool, error)
 //--------------------------------------
 
 // Listens to the heartbeat timeout and flushes an AppendEntries RPC.
-func (p *Peer) heartbeatTimeoutFunc() {
+func (p *Peer) heartbeatTimeoutFunc(startChannel chan bool) {
-	// Initialize the timer here since there can be a delay before this
+	startChannel <- true
-	// goroutine actually starts.
-	p.heartbeatTimer.Reset()
 	
 	for {
 		// Grab the current timer channel.

server.go

@@ -721,6 +721,9 @@ func (s *Server) AddPeer(name string) error {
 	// Only add the peer if it doesn't have the same name.
 	if s.name != name {
 		peer := NewPeer(s, name, s.heartbeatTimeout)
+		if s.state == Leader {
+			peer.resume()
+		}
 		s.peers[peer.name] = peer
 	}
 

server_test.go

@@ -343,7 +343,8 @@ func TestServerMultiNode(t *testing.T) {
 		mutex.Lock()
 		s := servers[peer.name]
 		mutex.Unlock()
-		return s.RequestVote(req)
+		resp, err := s.RequestVote(req)
+		return resp, err
 	}
 	transporter.sendAppendEntriesRequestFunc = func(server *Server, peer *Peer, req *AppendEntriesRequest) (*AppendEntriesResponse, error) {
 		mutex.Lock()
@@ -375,7 +376,7 @@ func TestServerMultiNode(t *testing.T) {
 		servers[name] = server
 		mutex.Unlock()
 	}
-	time.Sleep(500 * time.Millisecond)
+	time.Sleep(100 * time.Millisecond)
 
 	// Check that two peers exist on leader.
 	mutex.Lock()
@@ -385,9 +386,9 @@ func TestServerMultiNode(t *testing.T) {
 	mutex.Unlock()
 
 	// Stop the first server and wait for a re-election.
-	time.Sleep(500 * time.Millisecond)
+	time.Sleep(100 * time.Millisecond)
 	leader.Stop()
-	time.Sleep(500 * time.Millisecond)
+	time.Sleep(100 * time.Millisecond)
 
 	// Check that either server 2 or 3 is the leader now.
 	mutex.Lock()