func (s *SchedulerServer) failover(driver bindings.SchedulerDriver, hks hyperkube.Interface) error { if driver != nil { stat, err := driver.Stop(true) if stat != mesos.Status_DRIVER_STOPPED { return fmt.Errorf("failed to stop driver for failover, received unexpected status code: %v", stat) } else if err != nil { return err } } // there's no guarantee that all goroutines are actually programmed intelligently with 'done' // signals, so we'll need to restart if we want to really stop everything // run the same command that we were launched with //TODO(jdef) assumption here is that the sheduler is the only service running in this process, we should probably validate that somehow args := []string{} flags := pflag.CommandLine if hks != nil { args = append(args, hks.Name()) flags = hks.Flags() } flags.Visit(func(flag *pflag.Flag) { if flag.Name != "api-servers" && flag.Name != "etcd-servers" { args = append(args, fmt.Sprintf("--%s=%s", flag.Name, flag.Value.String())) } }) if !s.Graceful { args = append(args, "--graceful") } if len(s.APIServerList) > 0 { args = append(args, "--api-servers="+strings.Join(s.APIServerList, ",")) } if len(s.EtcdServerList) > 0 { args = append(args, "--etcd-servers="+strings.Join(s.EtcdServerList, ",")) } args = append(args, flags.Args()...) log.V(1).Infof("spawning scheduler for graceful failover: %s %+v", s.executable, args) cmd := exec.Command(s.executable, args...) cmd.Stdin = os.Stdin cmd.Stdout = os.Stdout cmd.Stderr = os.Stderr cmd.SysProcAttr = makeDisownedProcAttr() // TODO(jdef) pass in a pipe FD so that we can block, waiting for the child proc to be ready //cmd.ExtraFiles = []*os.File{} exitcode := 0 log.Flush() // TODO(jdef) it would be really nice to ensure that no one else in our process was still logging if err := cmd.Start(); err != nil { //log to stdtout here to avoid conflicts with normal stderr logging fmt.Fprintf(os.Stdout, "failed to spawn failover process: %v\n", err) os.Exit(1) } os.Exit(exitcode) select {} // will never reach here }