diff --git a/manager/manager.go b/manager/manager.go
index d17e8ec231..517b37817e 100644
--- a/manager/manager.go
+++ b/manager/manager.go
@@ -86,7 +86,11 @@ type Config struct {
 	// cluster to join.
 	JoinRaft string
 
-	// Top-level state directory
+	// ForceJoin causes us to invoke raft's Join RPC even if already part
+	// of a cluster.
+	ForceJoin bool
+
+	// StateDir is the top-level state directory
 	StateDir string
 
 	// ForceNewCluster defines if we have to force a new cluster
@@ -201,6 +205,7 @@ func New(config *Config) (*Manager, error) {
 	newNodeOpts := raft.NodeOptions{
 		ID:              config.SecurityConfig.ClientTLSCreds.NodeID(),
 		JoinAddr:        config.JoinRaft,
+		ForceJoin:       config.ForceJoin,
 		Config:          raftCfg,
 		StateDir:        raftStateDir,
 		ForceNewCluster: config.ForceNewCluster,
diff --git a/manager/state/raft/raft.go b/manager/state/raft/raft.go
index 89f19b5cdd..9422c0e3b4 100644
--- a/manager/state/raft/raft.go
+++ b/manager/state/raft/raft.go
@@ -166,6 +166,8 @@ type NodeOptions struct {
 	// JoinAddr is the cluster to join. May be an empty string to create
 	// a standalone cluster.
 	JoinAddr string
+	// ForceJoin tells us to join even if already part of a cluster.
+	ForceJoin bool
 	// Config is the raft config.
 	Config *raft.Config
 	// StateDir is the directory to store durable state.
@@ -393,8 +395,10 @@ func (n *Node) JoinAndStart(ctx context.Context) (err error) {
 
 	// restore from snapshot
 	if loadAndStartErr == nil {
-		if n.opts.JoinAddr != "" {
-			log.G(ctx).Warning("ignoring request to join cluster, because raft state already exists")
+		if n.opts.JoinAddr != "" && n.opts.ForceJoin {
+			if err := n.joinCluster(ctx); err != nil {
+				return errors.Wrap(err, "failed to rejoin cluster")
+			}
 		}
 		n.campaignWhenAble = true
 		n.initTransport()
@@ -402,7 +406,6 @@ func (n *Node) JoinAndStart(ctx context.Context) (err error) {
 		return nil
 	}
 
-	// first member of cluster
 	if n.opts.JoinAddr == "" {
 		// First member in the cluster, self-assign ID
 		n.Config.ID = uint64(rand.Int63()) + 1
@@ -417,6 +420,22 @@ func (n *Node) JoinAndStart(ctx context.Context) (err error) {
 	}
 
 	// join to existing cluster
+
+	if err := n.joinCluster(ctx); err != nil {
+		return err
+	}
+
+	if _, err := n.newRaftLogs(n.opts.ID); err != nil {
+		return err
+	}
+
+	n.initTransport()
+	n.raftNode = raft.StartNode(n.Config, nil)
+
+	return nil
+}
+
+func (n *Node) joinCluster(ctx context.Context) error {
 	if n.opts.Addr == "" {
 		return errors.New("attempted to join raft cluster without knowing own address")
 	}
@@ -438,15 +457,7 @@ func (n *Node) JoinAndStart(ctx context.Context) (err error) {
 	}
 
 	n.Config.ID = resp.RaftID
-
-	if _, err := n.newRaftLogs(n.opts.ID); err != nil {
-		return err
-	}
 	n.bootstrapMembers = resp.Members
-
-	n.initTransport()
-	n.raftNode = raft.StartNode(n.Config, nil)
-
 	return nil
 }
 
diff --git a/node/node.go b/node/node.go
index 77fe5b3d75..68f81f0b92 100644
--- a/node/node.go
+++ b/node/node.go
@@ -828,14 +828,22 @@ func (n *Node) runManager(ctx context.Context, securityConfig *ca.SecurityConfig
 		}
 	}
 
-	remoteAddr, _ := n.remotes.Select(n.NodeID())
+	joinAddr := n.config.JoinAddr
+	if joinAddr == "" {
+		remoteAddr, err := n.remotes.Select(n.NodeID())
+		if err == nil {
+			joinAddr = remoteAddr.Addr
+		}
+	}
+
 	m, err := manager.New(&manager.Config{
 		ForceNewCluster:  n.config.ForceNewCluster,
 		RemoteAPI:        remoteAPI,
 		ControlAPI:       n.config.ListenControlAPI,
 		SecurityConfig:   securityConfig,
 		ExternalCAs:      n.config.ExternalCAs,
-		JoinRaft:         remoteAddr.Addr,
+		JoinRaft:         joinAddr,
+		ForceJoin:        n.config.JoinAddr != "",
 		StateDir:         n.config.StateDir,
 		HeartbeatTick:    n.config.HeartbeatTick,
 		ElectionTick:     n.config.ElectionTick,