bugfix

src/job_manager.go

@@ -179,6 +179,13 @@ func (jm *JobManager) checkStatus(status []TaskStatus) {
 				onlyPS = false
 			}
 			InstanceJobHistoryLogger().submitTaskStatus(jm.job.Name, status[i])
+		} else if status[i].Status == "unknown" {
+			log.Warn(jm.job.Name, "-", i, " is unknown")
+			flagRunning = true
+			if !jm.job.Tasks[i].IsPS {
+				onlyPS = false
+			}
+			//InstanceJobHistoryLogger().submitTaskStatus(jm.job.Name, status[i])
 		} else {
 			log.Info(jm.job.Name, "-", i, " ", status[i].Status)
 			if exitCode, ok := status[i].State["ExitCode"].(float64); ok && exitCode != 0 && !jm.killFlag {
@@ -272,6 +279,10 @@ func (jm *JobManager) status() MsgJobStatus {
 			tasksStatus[i] = TaskStatus{Status: "unknown", State: map[string]interface{}{"ExitCode": float64(-1)}}
 			continue
 		}
+		if res.Code != 0 {
+			tasksStatus[i] = TaskStatus{Status: "notexist", State: map[string]interface{}{"ExitCode": float64(1)}}
+			continue
+		}
 		res.Status.Node = taskStatus.Node
 		tasksStatus[i] = res.Status
 	}

src/resource_pool.go

@@ -147,9 +147,6 @@ func (pool *ResourcePool) checkDeadNodes() {
 				seg = seg.Next
 			}
 			seg.Lock.Lock()
-			if seg.Nodes == nil {
-				seg = seg.Next
-			}
 			delete(seg.Nodes, v)
 			seg.Lock.Unlock()
 			delete(pool.heartBeat, v)