Exemplo n.º 1
0
func (s *SchedulerServer) Run(hks hyperqing.Interface, _ []string) error {
	// get scheduler low-level config
	sc := schedcfg.CreateDefaultConfig()
	if s.SchedulerConfigFileName != "" {
		f, err := os.Open(s.SchedulerConfigFileName)
		if err != nil {
			log.Fatalf("Cannot open scheduler config file: %v", err)
		}

		err = sc.Read(bufio.NewReader(f))
		if err != nil {
			log.Fatalf("Invalid scheduler config file: %v", err)
		}
	}

	schedulerProcess, driverFactory, etcdClient, eid := s.bootstrap(hks, sc)

	if s.EnableProfiling {
		profile.InstallHandler(s.mux)
	}
	go runtime.Until(func() {
		log.V(1).Info("Starting HTTP interface")
		log.Error(http.ListenAndServe(net.JoinHostPort(s.Address.String(), strconv.Itoa(s.Port)), s.mux))
	}, sc.HttpBindInterval.Duration, schedulerProcess.Terminal())

	if s.HA {
		validation := ha.ValidationFunc(validateLeadershipTransition)
		srv := ha.NewCandidate(schedulerProcess, driverFactory, validation)
		path := fmt.Sprintf(meta.DefaultElectionFormat, s.FrameworkName)
		sid := uid.New(eid.Group(), "").String()
		log.Infof("registering for election at %v with id %v", path, sid)
		go election.Notify(election.NewEtcdMasterElector(etcdClient), path, sid, srv, nil)
	} else {
		log.Infoln("self-electing in non-HA mode")
		schedulerProcess.Elect(driverFactory)
	}
	return s.awaitFailover(schedulerProcess, func() error { return s.failover(s.getDriver(), hks) })
}
Exemplo n.º 2
0
func (s *SchedulerServer) prepareExecutorInfo(hks hyperqing.Interface) (*mesos.ExecutorInfo, *uid.UID, error) {
	ci := &mesos.CommandInfo{
		Shell: proto.Bool(false),
	}

	//TODO(jdef) these should be shared constants with km
	const (
		KM_EXECUTOR = "executor"
		KM_PROXY    = "proxy"
	)

	if s.ExecutorPath != "" {
		uri, executorCmd := s.serveFrameworkArtifact(s.ExecutorPath)
		ci.Uris = append(ci.Uris, &mesos.CommandInfo_URI{Value: proto.String(uri), Executable: proto.Bool(true)})
		ci.Value = proto.String(fmt.Sprintf("./%s", executorCmd))
	} else if !hks.FindServer(KM_EXECUTOR) {
		return nil, nil, fmt.Errorf("either run this scheduler via km or else --executor-path is required")
	} else {
		if strings.Index(s.KMPath, "://") > 0 {
			// URI could point directly to executable, e.g. hdfs:///km
			// or else indirectly, e.g. http://acmestorage/tarball.tgz
			// so we assume that for this case the command will always "km"
			ci.Uris = append(ci.Uris, &mesos.CommandInfo_URI{Value: proto.String(s.KMPath), Executable: proto.Bool(true)})
			ci.Value = proto.String("./km") // TODO(jdef) extract constant
		} else if s.KMPath != "" {
			uri, kmCmd := s.serveFrameworkArtifact(s.KMPath)
			ci.Uris = append(ci.Uris, &mesos.CommandInfo_URI{Value: proto.String(uri), Executable: proto.Bool(true)})
			ci.Value = proto.String(fmt.Sprintf("./%s", kmCmd))
		} else {
			uri, kmCmd := s.serveFrameworkArtifact(s.executable)
			ci.Uris = append(ci.Uris, &mesos.CommandInfo_URI{Value: proto.String(uri), Executable: proto.Bool(true)})
			ci.Value = proto.String(fmt.Sprintf("./%s", kmCmd))
		}
		ci.Arguments = append(ci.Arguments, KM_EXECUTOR)
	}

	if s.ProxyPath != "" {
		uri, proxyCmd := s.serveFrameworkArtifact(s.ProxyPath)
		ci.Uris = append(ci.Uris, &mesos.CommandInfo_URI{Value: proto.String(uri), Executable: proto.Bool(true)})
		ci.Arguments = append(ci.Arguments, fmt.Sprintf("--proxy-exec=./%s", proxyCmd))
	} else if !hks.FindServer(KM_PROXY) {
		return nil, nil, fmt.Errorf("either run this scheduler via km or else --proxy-path is required")
	} else if s.ExecutorPath != "" {
		return nil, nil, fmt.Errorf("proxy can only use km binary if executor does the same")
	} // else, executor is smart enough to know when proxy-path is required, or to use km

	//TODO(jdef): provide some way (env var?) for users to customize executor config
	//TODO(jdef): set -address to 127.0.0.1 if `address` is 127.0.0.1
	//TODO(jdef): propagate dockercfg from RootDirectory?

	apiServerArgs := strings.Join(s.APIServerList, ",")
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--api-servers=%s", apiServerArgs))
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--v=%d", s.ExecutorLogV))
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--allow-privileged=%t", s.AllowPrivileged))
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--suicide-timeout=%v", s.ExecutorSuicideTimeout))

	if s.ExecutorBindall {
		//TODO(jdef) determine whether hostname-override is really needed for bindall because
		//it conflicts with qinglet node status checks/updates
		//ci.Arguments = append(ci.Arguments, "--hostname-override=0.0.0.0")
		ci.Arguments = append(ci.Arguments, "--address=0.0.0.0")
	}

	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--proxy-bindall=%v", s.ExecutorProxyBindall))
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--run-proxy=%v", s.ExecutorRunProxy))
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--cadvisor-port=%v", s.QingletCadvisorPort))
	ci.Arguments = append(ci.Arguments, fmt.Sprintf("--sync-frequency=%v", s.QingletSyncFrequency))

	if s.AuthPath != "" {
		//TODO(jdef) should probably support non-local files, e.g. hdfs:///some/config/file
		uri, basename := s.serveFrameworkArtifact(s.AuthPath)
		ci.Uris = append(ci.Uris, &mesos.CommandInfo_URI{Value: proto.String(uri)})
		ci.Arguments = append(ci.Arguments, fmt.Sprintf("--auth-path=%s", basename))
	}
	appendOptional := func(name string, value string) {
		if value != "" {
			ci.Arguments = append(ci.Arguments, fmt.Sprintf("--%s=%s", name, value))
		}
	}
	if s.ClusterDNS != nil {
		appendOptional("cluster-dns", s.ClusterDNS.String())
	}
	appendOptional("cluster-domain", s.ClusterDomain)
	appendOptional("root-dir", s.QingletRootDirectory)
	appendOptional("docker-endpoint", s.QingletDockerEndpoint)
	appendOptional("pod-infra-container-image", s.QingletPodInfraContainerImage)
	appendOptional("host-network-sources", s.QingletHostNetworkSources)
	appendOptional("network-plugin", s.QingletNetworkPluginName)

	log.V(1).Infof("prepared executor command %q with args '%+v'", ci.GetValue(), ci.Arguments)

	// Create mesos scheduler driver.
	info := &mesos.ExecutorInfo{
		Command: ci,
		Name:    proto.String(execcfg.DefaultInfoName),
		Source:  proto.String(execcfg.DefaultInfoSource),
	}

	// Check for staticPods
	if s.StaticPodsConfigPath != "" {
		bs, numberStaticPods, err := archive.ZipDir(s.StaticPodsConfigPath)
		if err != nil {
			return nil, nil, err
		}
		info.Data = bs

		// Adjust the resource accounting for the executor.
		// Currently each podTask accounts the default amount of resources.
		// TODO(joerg84) adapt to actual resources specified by pods.
		log.Infof("Detected %d staticPods in Configuration.", numberStaticPods)

		info.Resources = []*mesos.Resource{
			mutil.NewScalarResource("cpus", float64(numberStaticPods)*podtask.DefaultContainerCpus),
			mutil.NewScalarResource("mem", float64(numberStaticPods)*podtask.DefaultContainerMem),
		}
	}

	// calculate ExecutorInfo hash to be used for validating compatibility
	// of ExecutorInfo's generated by other HA schedulers.
	ehash := hashExecutorInfo(info)
	eid := uid.New(ehash, execcfg.DefaultInfoID)
	info.ExecutorId = &mesos.ExecutorID{Value: proto.String(eid.String())}

	return info, eid, nil
}