fix(server): rejoin cluster with different ip

This commit is contained in:
Yicheng Qin
2014-03-20 17:34:17 -07:00
parent d368405d05
commit 273c293645
3 changed files with 131 additions and 2 deletions

View File

@@ -29,6 +29,17 @@ func (c *JoinCommandV1) CommandName() string {
return "etcd:join"
}
func (c *JoinCommandV1) updatePeerURL(ps *PeerServer) error {
log.Debugf("Update peer URL of %v to %v", c.Name, c.RaftURL)
if err := ps.registry.UpdatePeerURL(c.Name, c.RaftURL); err != nil {
log.Debugf("Error while updating in registry: %s (%v)", c.Name, err)
return err
}
// Flush commit index, so raft will replay to here when restarted
ps.raftServer.FlushCommitIndex()
return nil
}
// Join a server to the cluster
func (c *JoinCommandV1) Apply(context raft.Context) (interface{}, error) {
ps, _ := context.Server().Context().(*PeerServer)
@@ -40,7 +51,15 @@ func (c *JoinCommandV1) Apply(context raft.Context) (interface{}, error) {
ps.registry.Invalidate(c.Name)
// Check if the join command is from a previous peer, who lost all its previous log.
if _, ok := ps.registry.ClientURL(c.Name); ok {
if peerURL, ok := ps.registry.PeerURL(c.Name); ok {
// If previous node restarts with different peer URL,
// update its information.
if peerURL != c.RaftURL {
log.Infof("Rejoin with %v instead of %v from %v", c.RaftURL, peerURL, c.Name)
if err := c.updatePeerURL(ps); err != nil {
return []byte{0}, err
}
}
return b, nil
}
@@ -83,6 +102,17 @@ func (c *JoinCommandV2) CommandName() string {
return "etcd:v2:join"
}
func (c *JoinCommandV2) updatePeerURL(ps *PeerServer) error {
log.Debugf("Update peer URL of %v to %v", c.Name, c.PeerURL)
if err := ps.registry.UpdatePeerURL(c.Name, c.PeerURL); err != nil {
log.Debugf("Error while updating in registry: %s (%v)", c.Name, err)
return err
}
// Flush commit index, so raft will replay to here when restart
ps.raftServer.FlushCommitIndex()
return nil
}
// Apply attempts to join a machine to the cluster.
func (c *JoinCommandV2) Apply(context raft.Context) (interface{}, error) {
ps, _ := context.Server().Context().(*PeerServer)
@@ -95,7 +125,15 @@ func (c *JoinCommandV2) Apply(context raft.Context) (interface{}, error) {
ps.registry.Invalidate(c.Name)
// Check if the join command is from a previous peer, who lost all its previous log.
if _, ok := ps.registry.ClientURL(c.Name); ok {
if peerURL, ok := ps.registry.PeerURL(c.Name); ok {
// If previous node restarts with different peer URL,
// update its information.
if peerURL != c.PeerURL {
log.Infof("Rejoin with %v instead of %v from %v", c.PeerURL, peerURL, c.Name)
if err := c.updatePeerURL(ps); err != nil {
return []byte{0}, err
}
}
return json.Marshal(msg)
}

View File

@@ -103,6 +103,25 @@ func (r *Registry) register(key, name string, peerURL string, machURL string) er
return err
}
// UpdatePeerURL updates peer URL in registry
func (r *Registry) UpdatePeerURL(name string, peerURL string) error {
r.Lock()
defer r.Unlock()
machURL, _ := r.clientURL(RegistryPeerKey, name)
// Write data to store.
key := path.Join(RegistryPeerKey, name)
v := url.Values{}
v.Set("raft", peerURL)
v.Set("etcd", machURL)
_, err := r.store.Update(key, v.Encode(), store.Permanent)
// Invalidate outdated cache.
r.invalidate(name)
log.Debugf("Update PeerURL: %s", name)
return err
}
// UnregisterPeer removes a peer from the registry.
func (r *Registry) UnregisterPeer(name string) error {
return r.unregister(RegistryPeerKey, name)
@@ -290,7 +309,10 @@ func (r *Registry) urls(key, leaderName, selfName string, url func(key, name str
func (r *Registry) Invalidate(name string) {
r.Lock()
defer r.Unlock()
r.invalidate(name)
}
func (r *Registry) invalidate(name string) {
delete(r.peers, name)
delete(r.standbys, name)
}