func HandleEngineEvents(taskEngine engine.TaskEngine, client api.ECSClient, saver statemanager.Saver) {
statesaver = saver
for {
taskEvents, containerEvents := taskEngine.TaskEvents()
for taskEvents != nil && containerEvents != nil {
select {
case event, open := <-containerEvents:
if !open {
containerEvents = nil
log.Error("Container events closed")
break
}
AddContainerEvent(event, client)
case event, open := <-taskEvents:
if !open {
taskEvents = nil
log.Crit("Task events closed")
break
}
AddTaskEvent(event, client)
}
}
}
}
// FinalSave should be called immediately before exiting, and only before
// exiting, in order to flush tasks to disk. It waits a short timeout for state
// to settle if necessary. If unable to reach a steady-state and save within
// this short timeout, it returns an error
func FinalSave(saver statemanager.Saver, taskEngine engine.TaskEngine) error {
engineDisabled := make(chan error)
disableTimer := time.AfterFunc(engineDisableTimeout, func() {
engineDisabled <- errors.New("Timed out waiting for TaskEngine to settle")
})
go func() {
log.Debug("Shutting down task engine")
taskEngine.Disable()
disableTimer.Stop()
engineDisabled <- nil
}()
disableErr := <-engineDisabled
stateSaved := make(chan error)
saveTimer := time.AfterFunc(finalSaveTimeout, func() {
stateSaved <- errors.New("Timed out trying to save to disk")
})
go func() {
log.Debug("Saving state before shutting down")
stateSaved <- saver.ForceSave()
saveTimer.Stop()
}()
saveErr := <-stateSaved
if disableErr != nil || saveErr != nil {
return utils.NewMultiError(disableErr, saveErr)
}
return nil
}
func addNonstoppedTasks(tasks []*api.Task, taskEngine engine.TaskEngine) bool {
allTasksOk := true
for _, task := range tasks {
if task.DesiredStatus == api.TaskStopped {
continue
}
err := taskEngine.AddTask(task)
if err != nil {
log.Warn("Could not add task; taskengine probably disabled")
// Don't ack
allTasksOk = false
}
}
return allTasksOk
}
// AcsWsUrl returns the websocket url for ACS given the endpoint.
func AcsWsUrl(endpoint, cluster, containerInstanceArn string, taskEngine engine.TaskEngine) string {
acsUrl := endpoint
if endpoint[len(endpoint)-1] != '/' {
acsUrl += "/"
}
acsUrl += "ws"
query := url.Values{}
query.Set("clusterArn", cluster)
query.Set("containerInstanceArn", containerInstanceArn)
query.Set("agentHash", version.GitHashString())
query.Set("agentVersion", version.Version)
query.Set("seqNum", strconv.FormatInt(SequenceNumber.Get(), 10))
if dockerVersion, err := taskEngine.Version(); err == nil {
query.Set("dockerVersion", dockerVersion)
}
return acsUrl + "?" + query.Encode()
}
// acsWsURL returns the websocket url for ACS given the endpoint
func acsWsURL(endpoint, cluster, containerInstanceArn string, taskEngine engine.TaskEngine, acsSessionState sessionState) string {
acsUrl := endpoint
if endpoint[len(endpoint)-1] != '/' {
acsUrl += "/"
}
acsUrl += "ws"
query := url.Values{}
query.Set("clusterArn", cluster)
query.Set("containerInstanceArn", containerInstanceArn)
query.Set("agentHash", version.GitHashString())
query.Set("agentVersion", version.Version)
query.Set("seqNum", "1")
if dockerVersion, err := taskEngine.Version(); err == nil {
query.Set("dockerVersion", dockerVersion)
}
query.Set(sendCredentialsURLParameterName, acsSessionState.getSendCredentialsURLParameter())
return acsUrl + "?" + query.Encode()
}
func _main() int {
defer log.Flush()
versionFlag := flag.Bool("version", false, "Print the agent version information and exit")
acceptInsecureCert := flag.Bool("k", false, "Do not verify ssl certs")
logLevel := flag.String("loglevel", "", "Loglevel: [<crit>|<error>|<warn>|<info>|<debug>]")
flag.Parse()
logger.SetLevel(*logLevel)
log.Infof("Starting Agent: %v", version.String())
log.Info("Loading configuration")
cfg, err := config.NewConfig()
// Load cfg before doing 'versionFlag' so that it has the DOCKER_HOST
// variable loaded if needed
if *versionFlag {
versionableEngine := engine.NewTaskEngine(cfg)
version.PrintVersion(versionableEngine)
return exitcodes.ExitSuccess
}
if err != 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)
// 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, acshandler.SequenceNumber)
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.DEFAULT_CLUSTER_NAME
}
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 := ec2.GetInstanceIdentityDocument(); 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 '%v' does not match current InstanceID '%v'. Overwriting old datafile", previousEc2InstanceID, currentEc2InstanceID)
// Reset taskEngine; all the other values are still default
taskEngine = engine.NewTaskEngine(cfg)
} 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)
}
stateManager, err := initializeStateManager(cfg, taskEngine, &cfg.Cluster, &containerInstanceArn, ¤tEc2InstanceID, acshandler.SequenceNumber)
if err != nil {
log.Criticalf("Error creating state manager: %v", err)
return exitcodes.ExitTerminal
}
credentialProvider := auth.NewBasicAWSCredentialProvider()
awsCreds := auth.ToSDK(credentialProvider)
// Preflight request to make sure they're good
if preflightCreds, err := awsCreds.Credentials(); err != nil || preflightCreds.AccessKeyID == "" {
if preflightCreds != nil {
log.Warnf("Error getting valid credentials (AKID %v): %v", preflightCreds.AccessKeyID, err)
} else {
log.Warnf("Error getting preflight credentials: %v", err)
}
}
client := api.NewECSClient(awsCreds, cfg, *acceptInsecureCert)
if containerInstanceArn == "" {
log.Info("Registering Instance with ECS")
containerInstanceArn, err = client.RegisterContainerInstance()
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 '%v' in cluster '%v'", containerInstanceArn, cfg.Cluster)
// Save our shiny new containerInstanceArn
stateManager.Save()
} else {
log.Infof("Restored from checkpoint file. I am running as '%v' in cluster '%v'", containerInstanceArn, cfg.Cluster)
}
// Begin listening to the docker daemon and saving changes
taskEngine.SetSaver(stateManager)
taskEngine.MustInit()
go sighandlers.StartTerminationHandler(stateManager, taskEngine)
// Agent introspection api
go handlers.ServeHttp(&containerInstanceArn, taskEngine, cfg)
// Start sending events to the backend
go eventhandler.HandleEngineEvents(taskEngine, client, stateManager)
log.Info("Beginning Polling for updates")
err = acshandler.StartSession(containerInstanceArn, credentialProvider, cfg, taskEngine, client, stateManager, *acceptInsecureCert)
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
}
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, ¤tEc2InstanceID)
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
}