// TestDeregisterInstanceStream tests the ws connection will be closed by tcs client when
// received the deregisterInstanceStream
func TestDeregisterInstanceStream(t *testing.T) {
	cs, ml := testCS()

	ctx, cancel := context.WithCancel(context.Background())
	deregisterInstanceEventStream := eventstream.NewEventStream("TestDeregisterInstanceStream", ctx)
	deregisterInstanceEventStream.StartListening()
	defer cancel()

	err := deregisterInstanceEventStream.Subscribe("TestDeregisterContainerInstanceHandler", cs.Disconnect)
	if err != nil {
		t.Errorf("Error subscribing to event stream, err %v", err)
	}

	err = cs.MakeRequest(&ecstcs.PublishMetricsRequest{})
	if err != nil {
		t.Errorf("Error making client request: %v", err)
	}
	if ml.closed {
		t.Error("Connection closed before send the deregister event")
	}
	err = deregisterInstanceEventStream.WriteToEventStream(struct{}{})
	if err != nil {
		t.Errorf("Failed to write to event stream, err %v", err)
	}
	// wait for the handler to run
	time.Sleep(1 * time.Second)
	if !ml.closed {
		t.Error("Client should be closed after receiving deregister event")
	}
}
func mocks(t *testing.T, cfg *config.Config) (*gomock.Controller, *MockDockerClient, *mock_ttime.MockTime, TaskEngine, *mock_credentials.MockManager) {
	ctrl := gomock.NewController(t)
	client := NewMockDockerClient(ctrl)
	mockTime := mock_ttime.NewMockTime(ctrl)
	credentialsManager := mock_credentials.NewMockManager(ctrl)

	containerChangeEventStream := eventstream.NewEventStream("TESTTASKENGINE", context.Background())
	containerChangeEventStream.StartListening()

	taskEngine := NewTaskEngine(cfg, client, credentialsManager, containerChangeEventStream)
	taskEngine.(*DockerTaskEngine)._time = mockTime
	return ctrl, client, mockTime, taskEngine, credentialsManager
}
func TestStartSession(t *testing.T) {
	// Start test server.
	closeWS := make(chan bool)
	server, serverChan, requestChan, serverErr, err := mockwsutils.StartMockServer(t, closeWS)
	defer server.Close()
	if err != nil {
		t.Fatal(err)
	}
	go func() {
		t.Error(<-serverErr)
	}()
	defer func() {
		closeWS <- true
		close(serverChan)
	}()

	deregisterInstanceEventStream := eventstream.NewEventStream("Deregister_Instance", context.Background())
	// Start a session with the test server.
	go startSession(server.URL, "us-east-1", credentials.AnonymousCredentials, true, &mockStatsEngine{}, testPublishMetricsInterval, deregisterInstanceEventStream)

	// startSession internally starts publishing metrics from the mockStatsEngine object.
	time.Sleep(testPublishMetricsInterval)

	// Read request channel to get the metric data published to the server.
	request := <-requestChan

	// Decode and verify the metric data.
	payload, err := getPayloadFromRequest(request)
	if err != nil {
		t.Fatal("Error decoding payload: ", err)
	}

	// Decode and verify the metric data.
	_, responseType, err := wsclient.DecodeData([]byte(payload), &tcsclient.TcsDecoder{})
	if err != nil {
		t.Fatal("error decoding data: ", err)
	}
	if responseType != "PublishMetricsRequest" {
		t.Fatal("Unexpected responseType: ", responseType)
	}
}
func setup(cfg *config.Config, t *testing.T) (TaskEngine, func(), credentials.Manager) {
	if testing.Short() {
		t.Skip("Skipping integ test in short mode")
	}
	if _, err := os.Stat("/var/run/docker.sock"); err != nil {
		t.Skip("Docker not running")
	}
	if os.Getenv("ECS_SKIP_ENGINE_INTEG_TEST") != "" {
		t.Skip("ECS_SKIP_ENGINE_INTEG_TEST")
	}

	clientFactory := dockerclient.NewFactory("unix:///var/run/docker.sock")
	dockerClient, err := NewDockerGoClient(clientFactory, false, cfg)
	if err != nil {
		t.Fatalf("Error creating Docker client: %v", err)
	}
	credentialsManager := credentials.NewManager()
	taskEngine := NewDockerTaskEngine(cfg, dockerClient, credentialsManager, eventstream.NewEventStream("ENGINEINTEGTEST", context.Background()))
	taskEngine.Init()
	return taskEngine, func() {
		taskEngine.Shutdown()
	}, credentialsManager
}
func TestSessionConenctionClosedByRemote(t *testing.T) {
	// Start test server.
	closeWS := make(chan bool)
	server, serverChan, _, serverErr, err := mockwsutils.StartMockServer(t, closeWS)
	defer server.Close()
	if err != nil {
		t.Fatal(err)
	}
	go func() {
		serr := <-serverErr
		if serr != io.EOF {
			t.Error(serr)
		}
	}()
	sleepBeforeClose := 10 * time.Millisecond
	go func() {
		time.Sleep(sleepBeforeClose)
		closeWS <- true
		close(serverChan)
	}()

	ctx, cancel := context.WithCancel(context.Background())
	deregisterInstanceEventStream := eventstream.NewEventStream("Deregister_Instance", ctx)
	deregisterInstanceEventStream.StartListening()
	defer cancel()

	// Start a session with the test server.
	err = startSession(server.URL, "us-east-1", credentials.AnonymousCredentials, true, &mockStatsEngine{}, testPublishMetricsInterval, deregisterInstanceEventStream)

	if err == nil {
		t.Error("Expected io.EOF on closed connection")
	}
	if err != io.EOF {
		t.Error("Expected io.EOF on closed connection, got: ", err)
	}
}
Ejemplo n.º 6
0
func _main() int {
	defer log.Flush()
	flagset := flag.NewFlagSet("Amazon ECS Agent", flag.ContinueOnError)
	versionFlag := flagset.Bool("version", false, "Print the agent version information and exit")
	logLevel := flagset.String("loglevel", "", "Loglevel: [<crit>|<error>|<warn>|<info>|<debug>]")
	acceptInsecureCert := flagset.Bool("k", false, "Disable SSL certificate verification. We do not recommend setting this option.")
	licenseFlag := flagset.Bool("license", false, "Print the LICENSE and NOTICE files and exit")
	blackholeEc2Metadata := flagset.Bool("blackhole-ec2-metadata", false, "Blackhole the EC2 Metadata requests. Setting this option can cause the ECS Agent to fail to work properly.  We do not recommend setting this option")
	err := flagset.Parse(os.Args[1:])
	if err != nil {
		return exitcodes.ExitTerminal
	}

	if *licenseFlag {
		license := utils.NewLicenseProvider()
		text, err := license.GetText()
		if err != nil {
			fmt.Fprintln(os.Stderr, err)
			return exitcodes.ExitError
		}
		fmt.Println(text)
		return exitcodes.ExitSuccess
	}

	logger.SetLevel(*logLevel)
	ec2MetadataClient := ec2.DefaultClient
	if *blackholeEc2Metadata {
		ec2MetadataClient = ec2.NewBlackholeEC2MetadataClient()
	}

	log.Infof("Starting Agent: %s", version.String())
	if *acceptInsecureCert {
		log.Warn("SSL certificate verification disabled. This is not recommended.")
	}
	log.Info("Loading configuration")
	cfg, cfgErr := config.NewConfig(ec2MetadataClient)
	// Load cfg and create Docker client before doing 'versionFlag' so that it has the DOCKER_HOST variable loaded if needed
	clientFactory := dockerclient.NewFactory(cfg.DockerEndpoint)
	dockerClient, err := engine.NewDockerGoClient(clientFactory, *acceptInsecureCert, cfg)
	if err != nil {
		log.Criticalf("Error creating Docker client: %v", err)
		return exitcodes.ExitError
	}

	ctx := context.Background()
	// Create the DockerContainerChange event stream for tcs
	containerChangeEventStream := eventstream.NewEventStream(ContainerChangeEventStream, ctx)
	containerChangeEventStream.StartListening()

	// Create credentials manager. This will be used by the task engine and
	// the credentials handler
	credentialsManager := credentials.NewManager()
	// Create image manager. This will be used by the task engine for saving image states
	state := dockerstate.NewDockerTaskEngineState()
	imageManager := engine.NewImageManager(cfg, dockerClient, state)
	if *versionFlag {
		versionableEngine := engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
		version.PrintVersion(versionableEngine)
		return exitcodes.ExitSuccess
	}

	sighandlers.StartDebugHandler()

	if cfgErr != nil {
		log.Criticalf("Error loading config: %v", err)
		// All required config values can be inferred from EC2 Metadata, so this error could be transient.
		return exitcodes.ExitError
	}
	log.Debug("Loaded config: " + cfg.String())

	var currentEc2InstanceID, containerInstanceArn string
	var taskEngine engine.TaskEngine

	if cfg.Checkpoint {
		log.Info("Checkpointing is enabled. Attempting to load state")
		var previousCluster, previousEc2InstanceID, previousContainerInstanceArn string
		previousTaskEngine := engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
		// previousState is used to verify that our current runtime configuration is
		// compatible with our past configuration as reflected by our state-file
		previousState, err := initializeStateManager(cfg, previousTaskEngine, &previousCluster, &previousContainerInstanceArn, &previousEc2InstanceID)
		if err != nil {
			log.Criticalf("Error creating state manager: %v", err)
			return exitcodes.ExitTerminal
		}

		err = previousState.Load()
		if err != nil {
			log.Criticalf("Error loading previously saved state: %v", err)
			return exitcodes.ExitTerminal
		}

		if previousCluster != "" {
			// TODO Handle default cluster in a sane and unified way across the codebase
			configuredCluster := cfg.Cluster
			if configuredCluster == "" {
				log.Debug("Setting cluster to default; none configured")
				configuredCluster = config.DefaultClusterName
			}
			if previousCluster != configuredCluster {
				log.Criticalf("Data mismatch; saved cluster '%v' does not match configured cluster '%v'. Perhaps you want to delete the configured checkpoint file?", previousCluster, configuredCluster)
				return exitcodes.ExitTerminal
			}
			cfg.Cluster = previousCluster
			log.Infof("Restored cluster '%v'", cfg.Cluster)
		}

		if instanceIdentityDoc, err := ec2MetadataClient.InstanceIdentityDocument(); err == nil {
			currentEc2InstanceID = instanceIdentityDoc.InstanceId
		} else {
			log.Criticalf("Unable to access EC2 Metadata service to determine EC2 ID: %v", err)
		}

		if previousEc2InstanceID != "" && previousEc2InstanceID != currentEc2InstanceID {
			log.Warnf("Data mismatch; saved InstanceID '%s' does not match current InstanceID '%s'. Overwriting old datafile", previousEc2InstanceID, currentEc2InstanceID)

			// Reset taskEngine; all the other values are still default
			taskEngine = engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
		} else {
			// Use the values we loaded if there's no issue
			containerInstanceArn = previousContainerInstanceArn
			taskEngine = previousTaskEngine
		}
	} else {
		log.Info("Checkpointing not enabled; a new container instance will be created each time the agent is run")
		taskEngine = engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
	}

	stateManager, err := initializeStateManager(cfg, taskEngine, &cfg.Cluster, &containerInstanceArn, &currentEc2InstanceID)
	if err != nil {
		log.Criticalf("Error creating state manager: %v", err)
		return exitcodes.ExitTerminal
	}

	capabilities := taskEngine.Capabilities()

	// We instantiate our own credentialProvider for use in acs/tcs. This tries
	// to mimic roughly the way it's instantiated by the SDK for a default
	// session.
	credentialProvider := defaults.CredChain(defaults.Config(), defaults.Handlers())
	// Preflight request to make sure they're good
	if preflightCreds, err := credentialProvider.Get(); err != nil || preflightCreds.AccessKeyID == "" {
		log.Warnf("Error getting valid credentials (AKID %s): %v", preflightCreds.AccessKeyID, err)
	}
	client := api.NewECSClient(credentialProvider, cfg, httpclient.New(api.RoundtripTimeout, *acceptInsecureCert), ec2MetadataClient)

	if containerInstanceArn == "" {
		log.Info("Registering Instance with ECS")
		containerInstanceArn, err = client.RegisterContainerInstance("", capabilities)
		if err != nil {
			log.Errorf("Error registering: %v", err)
			if retriable, ok := err.(utils.Retriable); ok && !retriable.Retry() {
				return exitcodes.ExitTerminal
			}
			return exitcodes.ExitError
		}
		log.Infof("Registration completed successfully. I am running as '%s' in cluster '%s'", containerInstanceArn, cfg.Cluster)
		// Save our shiny new containerInstanceArn
		stateManager.Save()
	} else {
		log.Infof("Restored from checkpoint file. I am running as '%s' in cluster '%s'", containerInstanceArn, cfg.Cluster)
		_, err = client.RegisterContainerInstance(containerInstanceArn, capabilities)
		if err != nil {
			log.Errorf("Error re-registering: %v", err)
			if awserr, ok := err.(awserr.Error); ok && api.IsInstanceTypeChangedError(awserr) {
				log.Criticalf("The current instance type does not match the registered instance type. Please revert the instance type change, or alternatively launch a new instance. Error: %v", err)
				return exitcodes.ExitTerminal
			}
			return exitcodes.ExitError
		}
	}

	// Begin listening to the docker daemon and saving changes
	taskEngine.SetSaver(stateManager)
	imageManager.SetSaver(stateManager)
	taskEngine.MustInit()

	// start of the periodic image cleanup process
	if !cfg.ImageCleanupDisabled {
		go imageManager.StartImageCleanupProcess(ctx)
	}

	go sighandlers.StartTerminationHandler(stateManager, taskEngine)

	// Agent introspection api
	go handlers.ServeHttp(&containerInstanceArn, taskEngine, cfg)

	// Start serving the endpoint to fetch IAM Role credentials
	go credentialshandler.ServeHttp(credentialsManager, containerInstanceArn, cfg)

	// Start sending events to the backend
	go eventhandler.HandleEngineEvents(taskEngine, client, stateManager)

	deregisterInstanceEventStream := eventstream.NewEventStream(DeregisterContainerInstanceEventStream, ctx)
	deregisterInstanceEventStream.StartListening()

	telemetrySessionParams := tcshandler.TelemetrySessionParams{
		ContainerInstanceArn: containerInstanceArn,
		CredentialProvider:   credentialProvider,
		Cfg:                  cfg,
		DeregisterInstanceEventStream: deregisterInstanceEventStream,
		ContainerChangeEventStream:    containerChangeEventStream,
		DockerClient:                  dockerClient,
		AcceptInvalidCert:             *acceptInsecureCert,
		EcsClient:                     client,
		TaskEngine:                    taskEngine,
	}

	// Start metrics session in a go routine
	go tcshandler.StartMetricsSession(telemetrySessionParams)

	log.Info("Beginning Polling for updates")
	err = acshandler.StartSession(ctx, acshandler.StartSessionArguments{
		AcceptInvalidCert: *acceptInsecureCert,
		Config:            cfg,
		DeregisterInstanceEventStream: deregisterInstanceEventStream,
		ContainerInstanceArn:          containerInstanceArn,
		CredentialProvider:            credentialProvider,
		ECSClient:                     client,
		StateManager:                  stateManager,
		TaskEngine:                    taskEngine,
		CredentialsManager:            credentialsManager,
	})
	if err != nil {
		log.Criticalf("Unretriable error starting communicating with ACS: %v", err)
		return exitcodes.ExitTerminal
	}
	log.Critical("ACS Session handler should never exit")
	return exitcodes.ExitError
}
// eventStream returns the event stream used to receive container change events
func eventStream(name string) *eventstream.EventStream {
	eventStream := eventstream.NewEventStream(name, context.Background())
	eventStream.StartListening()
	return eventStream
}