handling lost servers

weed-fs/src/cmd/weed/master.go

@@ -55,7 +55,7 @@ func dirAssignHandler(w http.ResponseWriter, r *http.Request) {
 	if err == nil {
 		writeJson(w, r, map[string]interface{}{"fid": fid, "url": machine.Url, "publicUrl": machine.PublicUrl, "count": count})
 	} else {
-		writeJson(w, r, map[string]interface{}{"error": err})
+		writeJson(w, r, map[string]string{"error": err.Error()})
 	}
 }
 func dirJoinHandler(w http.ResponseWriter, r *http.Request) {
@@ -80,6 +80,8 @@ func runMaster(cmd *Command, args []string) bool {
 	http.HandleFunc("/dir/join", dirJoinHandler)
 	http.HandleFunc("/dir/status", dirStatusHandler)
 
+  mapper.StartRefreshWritableVolumes()
+
 	log.Println("Start directory service at http://127.0.0.1:" + strconv.Itoa(*mport))
 	e := http.ListenAndServe(":"+strconv.Itoa(*mport), nil)
 	if e != nil {

weed-fs/src/pkg/directory/volume_mapping.go

@@ -76,6 +76,12 @@ func (m *Mapper) Add(machine *Machine) {
 	}
 	m.refreshWritableVolumes()
 }
+func (m *Mapper) remove(machine *Machine) {
+  delete(m.Machines,machine.Url)
+  for _, v := range machine.Volumes {
+    delete(m.vid2machine,v.Id)
+  }
+}
 func (m *Mapper) StartRefreshWritableVolumes() {
 	go func() {
 		for {
@@ -96,6 +102,9 @@ func (m *Mapper) refreshWritableVolumes() {
 					writers = append(writers, v.Id)
 				}
 			}
+		} else {
+			log.Println("Warning! Server", machine_entry.Url, "last seen is", time.Now().Unix()-machine_entry.LastSeen, "seconds ago!")
+			m.remove(machine_entry)
 		}
 	}
 	m.Writers = writers