// Response to append entries request
func (ps *PeerServer) AppendEntriesHttpHandler(w http.ResponseWriter, req *http.Request) {
start := time.Now()
aereq := &raft.AppendEntriesRequest{}
if _, err := aereq.Decode(req.Body); err != nil {
http.Error(w, "", http.StatusBadRequest)
log.Warnf("[recv] BADREQUEST %s/log/append [%v]", ps.Config.URL, err)
return
}
log.Debugf("[recv] POST %s/log/append [%d]", ps.Config.URL, len(aereq.Entries))
ps.serverStats.RecvAppendReq(aereq.LeaderName, int(req.ContentLength))
resp := ps.raftServer.AppendEntries(aereq)
if resp == nil {
log.Warn("[ae] Error: nil response")
http.Error(w, "", http.StatusInternalServerError)
return
}
if !resp.Success() {
log.Debugf("[Append Entry] Step back")
}
if _, err := resp.Encode(w); err != nil {
log.Warn("[ae] Error: %v", err)
http.Error(w, "", http.StatusInternalServerError)
return
}
(*ps.metrics).Timer("timer.appendentries.handle").UpdateSince(start)
}
// Sends SnapshotRecoveryRequest RPCs to a peer when the server is the candidate.
func (t *transporter) SendSnapshotRecoveryRequest(server raft.Server, peer *raft.Peer, req *raft.SnapshotRecoveryRequest) *raft.SnapshotRecoveryResponse {
var b bytes.Buffer
if _, err := req.Encode(&b); err != nil {
log.Warn("transporter.ss.encoding.error:", err)
return nil
}
u, _ := t.registry.PeerURL(peer.Name)
log.Debugf("Send Snapshot Recovery from %s to %s", server.Name(), u)
resp, httpRequest, err := t.Post(fmt.Sprintf("%s/snapshotRecovery", u), &b)
if err != nil {
log.Debugf("Cannot send Snapshot Recovery to %s : %s", u, err)
}
if resp != nil {
defer resp.Body.Close()
t.CancelWhenTimeout(httpRequest)
ssrrsp := &raft.SnapshotRecoveryResponse{}
if _, err = ssrrsp.Decode(resp.Body); err != nil && err != io.EOF {
log.Warn("transporter.ssr.decoding.error:", err)
return nil
}
return ssrrsp
}
return nil
}
// Sends AppendEntries RPCs to a peer when the server is the leader.
func (t *transporter) SendAppendEntriesRequest(server raft.Server, peer *raft.Peer, req *raft.AppendEntriesRequest) *raft.AppendEntriesResponse {
var b bytes.Buffer
if _, err := req.Encode(&b); err != nil {
log.Warn("transporter.ae.encoding.error:", err)
return nil
}
size := b.Len()
t.serverStats.SendAppendReq(size)
u, _ := t.registry.PeerURL(peer.Name)
log.Debugf("Send LogEntries to %s ", u)
thisFollowerStats, ok := t.followersStats.Followers[peer.Name]
if !ok { //this is the first time this follower has been seen
thisFollowerStats = &raftFollowerStats{}
thisFollowerStats.Latency.Minimum = 1 << 63
t.followersStats.Followers[peer.Name] = thisFollowerStats
}
start := time.Now()
resp, httpRequest, err := t.Post(fmt.Sprintf("%s/log/append", u), &b)
end := time.Now()
if err != nil {
log.Debugf("Cannot send AppendEntriesRequest to %s: %s", u, err)
if ok {
thisFollowerStats.Fail()
}
return nil
} else {
if ok {
thisFollowerStats.Succ(end.Sub(start))
}
}
if resp != nil {
defer resp.Body.Close()
t.CancelWhenTimeout(httpRequest)
aeresp := &raft.AppendEntriesResponse{}
if _, err = aeresp.Decode(resp.Body); err != nil && err != io.EOF {
log.Warn("transporter.ae.decoding.error:", err)
return nil
}
return aeresp
}
return nil
}
// Send join requests to peer.
func (s *PeerServer) joinByPeer(server raft.Server, peer string, scheme string) error {
var b bytes.Buffer
// t must be ok
t, _ := server.Transporter().(*transporter)
// Our version must match the leaders version
versionURL := url.URL{Host: peer, Scheme: scheme, Path: "/version"}
version, err := getVersion(t, versionURL)
if err != nil {
return fmt.Errorf("Error during join version check: %v", err)
}
if version < store.MinVersion() || version > store.MaxVersion() {
return fmt.Errorf("Unable to join: cluster version is %d; version compatibility is %d - %d", version, store.MinVersion(), store.MaxVersion())
}
json.NewEncoder(&b).Encode(NewJoinCommand(store.MinVersion(), store.MaxVersion(), server.Name(), s.Config.URL, s.server.URL()))
joinURL := url.URL{Host: peer, Scheme: scheme, Path: "/join"}
log.Debugf("Send Join Request to %s", joinURL.String())
resp, req, err := t.Post(joinURL.String(), &b)
for {
if err != nil {
return fmt.Errorf("Unable to join: %v", err)
}
if resp != nil {
defer resp.Body.Close()
t.CancelWhenTimeout(req)
if resp.StatusCode == http.StatusOK {
b, _ := ioutil.ReadAll(resp.Body)
s.joinIndex, _ = binary.Uvarint(b)
return nil
}
if resp.StatusCode == http.StatusTemporaryRedirect {
address := resp.Header.Get("Location")
log.Debugf("Send Join Request to %s", address)
json.NewEncoder(&b).Encode(NewJoinCommand(store.MinVersion(), store.MaxVersion(), server.Name(), s.Config.URL, s.server.URL()))
resp, req, err = t.Post(address, &b)
} else if resp.StatusCode == http.StatusBadRequest {
log.Debug("Reach max number peers in the cluster")
decoder := json.NewDecoder(resp.Body)
err := &etcdErr.Error{}
decoder.Decode(err)
return *err
} else {
return fmt.Errorf("Unable to join")
}
}
}
}
// Start the raft server
func (s *PeerServer) Start(snapshot bool, cluster []string) error {
// LoadSnapshot
if snapshot {
err := s.raftServer.LoadSnapshot()
if err == nil {
log.Debugf("%s finished load snapshot", s.Config.Name)
} else {
log.Debug(err)
}
}
s.raftServer.Start()
if s.raftServer.IsLogEmpty() {
// start as a leader in a new cluster
if len(cluster) == 0 {
s.startAsLeader()
} else {
s.startAsFollower(cluster)
}
} else {
// Rejoin the previous cluster
cluster = s.registry.PeerURLs(s.raftServer.Leader(), s.Config.Name)
for i := 0; i < len(cluster); i++ {
u, err := url.Parse(cluster[i])
if err != nil {
log.Debug("rejoin cannot parse url: ", err)
}
cluster[i] = u.Host
}
ok := s.joinCluster(cluster)
if !ok {
log.Warn("the entire cluster is down! this peer will restart the cluster.")
}
log.Debugf("%s restart as a follower", s.Config.Name)
}
s.closeChan = make(chan bool)
go s.monitorSync()
go s.monitorTimeoutThreshold(s.closeChan)
// open the snapshot
if snapshot {
go s.monitorSnapshot()
}
return nil
}
// Response to vote request
func (ps *PeerServer) VoteHttpHandler(w http.ResponseWriter, req *http.Request) {
rvreq := &raft.RequestVoteRequest{}
if _, err := rvreq.Decode(req.Body); err != nil {
http.Error(w, "", http.StatusBadRequest)
log.Warnf("[recv] BADREQUEST %s/vote [%v]", ps.Config.URL, err)
return
}
log.Debugf("[recv] POST %s/vote [%s]", ps.Config.URL, rvreq.CandidateName)
resp := ps.raftServer.RequestVote(rvreq)
if resp == nil {
log.Warn("[vote] Error: nil response")
http.Error(w, "", http.StatusInternalServerError)
return
}
if _, err := resp.Encode(w); err != nil {
log.Warn("[vote] Error: %v", err)
http.Error(w, "", http.StatusInternalServerError)
return
}
}
// Response to recover from snapshot request
func (ps *PeerServer) SnapshotHttpHandler(w http.ResponseWriter, req *http.Request) {
ssreq := &raft.SnapshotRequest{}
if _, err := ssreq.Decode(req.Body); err != nil {
http.Error(w, "", http.StatusBadRequest)
log.Warnf("[recv] BADREQUEST %s/snapshot [%v]", ps.Config.URL, err)
return
}
log.Debugf("[recv] POST %s/snapshot", ps.Config.URL)
resp := ps.raftServer.RequestSnapshot(ssreq)
if resp == nil {
log.Warn("[ss] Error: nil response")
http.Error(w, "", http.StatusInternalServerError)
return
}
if _, err := resp.Encode(w); err != nil {
log.Warn("[ss] Error: %v", err)
http.Error(w, "", http.StatusInternalServerError)
return
}
}
func GetHandler(w http.ResponseWriter, req *http.Request, s Server) error {
vars := mux.Vars(req)
key := "/" + vars["key"]
// Help client to redirect the request to the current leader
if req.FormValue("consistent") == "true" && s.State() != raft.Leader {
leader := s.Leader()
hostname, _ := s.ClientURL(leader)
url, err := url.Parse(hostname)
if err != nil {
log.Warn("Redirect cannot parse hostName ", hostname)
return err
}
url.RawQuery = req.URL.RawQuery
url.Path = req.URL.Path
log.Debugf("Redirect consistent get to %s", url.String())
http.Redirect(w, req, url.String(), http.StatusTemporaryRedirect)
return nil
}
recursive := (req.FormValue("recursive") == "true")
sort := (req.FormValue("sorted") == "true")
waitIndex := req.FormValue("waitIndex")
stream := (req.FormValue("stream") == "true")
if req.FormValue("wait") == "true" {
return handleWatch(key, recursive, stream, waitIndex, w, s)
}
return handleGet(key, recursive, sort, w, s)
}
// Checks whether a given version is supported.
func (ps *PeerServer) VersionCheckHttpHandler(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] Get %s%s ", ps.Config.URL, req.URL.Path)
vars := mux.Vars(req)
version, _ := strconv.Atoi(vars["version"])
if version >= store.MinVersion() && version <= store.MaxVersion() {
w.WriteHeader(http.StatusOK)
} else {
w.WriteHeader(http.StatusForbidden)
}
}
func (s *PeerServer) startAsLeader() {
// leader need to join self as a peer
for {
_, err := s.raftServer.Do(NewJoinCommand(store.MinVersion(), store.MaxVersion(), s.raftServer.Name(), s.Config.URL, s.server.URL()))
if err == nil {
break
}
}
log.Debugf("%s start as a leader", s.Config.Name)
}
// Remove a server from the cluster
func (c *RemoveCommand) Apply(context raft.Context) (interface{}, error) {
ps, _ := context.Server().Context().(*PeerServer)
// Remove node from the shared registry.
err := ps.registry.Unregister(c.Name)
// Delete from stats
delete(ps.followersStats.Followers, c.Name)
if err != nil {
log.Debugf("Error while unregistering: %s (%v)", c.Name, err)
return []byte{0}, err
}
// Remove peer in raft
err = context.Server().RemovePeer(c.Name)
if err != nil {
log.Debugf("Unable to remove peer: %s (%v)", c.Name, err)
return []byte{0}, err
}
if c.Name == context.Server().Name() {
// the removed node is this node
// if the node is not replaying the previous logs
// and the node has sent out a join request in this
// start. It is sure that this node received a new remove
// command and need to be removed
if context.CommitIndex() > ps.joinIndex && ps.joinIndex != 0 {
log.Debugf("server [%s] is removed", context.Server().Name())
os.Exit(0)
} else {
// else ignore remove
log.Debugf("ignore previous remove command.")
}
}
b := make([]byte, 8)
binary.PutUvarint(b, context.CommitIndex())
return b, err
}
// DashboardHttpHandler either uses the compiled in virtual filesystem for the
// dashboard assets or if ETCD_DASHBOARD_DIR is set uses that as the source of
// assets.
func HttpHandler() (handler http.Handler) {
handler = http.HandlerFunc(memoryFileServer)
// Serve the dashboard from a filesystem if the magic env variable is enabled
dashDir := os.Getenv("ETCD_DASHBOARD_DIR")
if len(dashDir) != 0 {
log.Debugf("Using dashboard directory %s", dashDir)
handler = http.FileServer(http.Dir(dashDir))
}
return handler
}
// Removes a node from the registry.
func (r *Registry) Unregister(name string) error {
r.Lock()
defer r.Unlock()
// Remove from cache.
// delete(r.nodes, name)
// Remove the key from the store.
_, err := r.store.Delete(path.Join(RegistryKey, name), false, false)
log.Debugf("Unregister: %s", name)
return err
}
// Adds a node to the registry.
func (r *Registry) Register(name string, peerURL string, machURL string) error {
r.Lock()
defer r.Unlock()
// Write data to store.
key := path.Join(RegistryKey, name)
v := url.Values{}
v.Set("raft", peerURL)
v.Set("etcd", machURL)
_, err := r.store.Create(key, false, v.Encode(), false, store.Permanent)
log.Debugf("Register: %s", name)
return err
}
// Response to remove request
func (ps *PeerServer) RemoveHttpHandler(w http.ResponseWriter, req *http.Request) {
if req.Method != "DELETE" {
w.WriteHeader(http.StatusMethodNotAllowed)
return
}
vars := mux.Vars(req)
command := &RemoveCommand{
Name: vars["name"],
}
log.Debugf("[recv] Remove Request [%s]", command.Name)
ps.server.Dispatch(command, w, req)
}
// Upgrades the current store version to the next version.
func (ps *PeerServer) UpgradeHttpHandler(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] Get %s/version", ps.Config.URL)
// Check if upgrade is possible for all nodes.
if err := ps.Upgradable(); err != nil {
http.Error(w, err.Error(), http.StatusInternalServerError)
return
}
// Create an upgrade command from the current version.
c := ps.store.CommandFactory().CreateUpgradeCommand()
if err := ps.server.Dispatch(c, w, req); err != nil {
http.Error(w, err.Error(), http.StatusInternalServerError)
return
}
w.WriteHeader(http.StatusOK)
}
// Adds a server handler to the router.
func (s *Server) handleFunc(r *mux.Router, path string, f func(http.ResponseWriter, *http.Request) error) *mux.Route {
// Wrap the standard HandleFunc interface to pass in the server reference.
return r.HandleFunc(path, func(w http.ResponseWriter, req *http.Request) {
// Log request.
log.Debugf("[recv] %s %s %s [%s]", req.Method, s.URL(), req.URL.Path, req.RemoteAddr)
// Execute handler function and return error if necessary.
if err := f(w, req); err != nil {
if etcdErr, ok := err.(*etcdErr.Error); ok {
log.Debug("Return error: ", (*etcdErr).Error())
w.Header().Set("Content-Type", "application/json")
etcdErr.Write(w)
} else {
http.Error(w, err.Error(), http.StatusInternalServerError)
}
}
})
}
func (s *PeerServer) joinCluster(cluster []string) bool {
for _, peer := range cluster {
if len(peer) == 0 {
continue
}
err := s.joinByPeer(s.raftServer, peer, s.Config.Scheme)
if err == nil {
log.Debugf("%s joined the cluster via peer %s", s.Config.Name, peer)
return true
}
if _, ok := err.(etcdErr.Error); ok {
log.Fatal(err)
}
log.Warnf("Attempt to join via %s failed: %s", peer, err)
}
return false
}
// Response to the join request
func (ps *PeerServer) JoinHttpHandler(w http.ResponseWriter, req *http.Request) {
command := &JoinCommand{}
err := uhttp.DecodeJsonRequest(req, command)
if err != nil {
w.WriteHeader(http.StatusInternalServerError)
return
}
log.Debugf("Receive Join Request from %s", command.Name)
err = ps.server.Dispatch(command, w, req)
// Return status.
if err != nil {
if etcdErr, ok := err.(*etcdErr.Error); ok {
log.Debug("Return error: ", (*etcdErr).Error())
etcdErr.Write(w)
} else {
http.Error(w, err.Error(), http.StatusInternalServerError)
}
}
}
func memoryFileServer(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] %s %s [%s]", req.Method, req.URL.Path, req.RemoteAddr)
upath := req.URL.Path
if len(upath) == 0 {
upath = "index.html"
}
// TODO: use the new mux to do this work
dir, file := path.Split(upath)
if file == "browser" || file == "stats" {
file = file + ".html"
}
upath = path.Join(dir, file)
b, err := resources.Asset(upath)
if err != nil {
http.Error(w, upath+": File not found", http.StatusNotFound)
return
}
http.ServeContent(w, req, upath, time.Time{}, bytes.NewReader(b))
return
}
// Response to the name request
func (ps *PeerServer) VersionHttpHandler(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] Get %s/version/ ", ps.Config.URL)
w.WriteHeader(http.StatusOK)
w.Write([]byte(strconv.Itoa(ps.store.Version())))
}
// Get the port that listening for etcd connecting of the server
func (ps *PeerServer) EtcdURLHttpHandler(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] Get %s/etcdURL/ ", ps.Config.URL)
w.WriteHeader(http.StatusOK)
w.Write([]byte(ps.server.URL()))
}
// Get all the current logs
func (ps *PeerServer) GetLogHttpHandler(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] GET %s/log", ps.Config.URL)
w.Header().Set("Content-Type", "application/json")
w.WriteHeader(http.StatusOK)
json.NewEncoder(w).Encode(ps.raftServer.LogEntries())
}
// Response to the name request
func (ps *PeerServer) NameHttpHandler(w http.ResponseWriter, req *http.Request) {
log.Debugf("[recv] Get %s/name/ ", ps.Config.URL)
w.WriteHeader(http.StatusOK)
w.Write([]byte(ps.Config.Name))
}