// createACSClient creates the ACS Client using the specified URL
func (acsResources *acsSessionResources) createACSClient(url string) wsclient.ClientServer {
args := acsResources.startSessionArguments
cfg := args.Config
return acsclient.New(url, cfg.AWSRegion, args.CredentialProvider, args.AcceptInvalidCert)
}
// StartSession creates a session with ACS and handles requests using the passed
// in arguments.
func StartSession(ctx context.Context, args StartSessionArguments) error {
ecsclient := args.ECSClient
cfg := args.Config
backoff := utils.NewSimpleBackoff(connectionBackoffMin, connectionBackoffMax, connectionBackoffJitter, connectionBackoffMultiplier)
payloadBuffer := make(chan *ecsacs.PayloadMessage, payloadMessageBufferSize)
ackBuffer := make(chan string, payloadMessageBufferSize)
go func() {
// Handle any payloads async. For correctness, they must be handled in order, hence the buffered channel which is added to synchronously.
for {
select {
case payload := <-payloadBuffer:
handlePayloadMessage(ackBuffer, cfg.Cluster, args.ContainerInstanceArn, payload, args.TaskEngine, ecsclient, args.StateManager)
case <-ctx.Done():
return
}
}
}()
for {
acsError := func() error {
acsEndpoint, err := ecsclient.DiscoverPollEndpoint(args.ContainerInstanceArn)
if err != nil {
log.Error("Unable to discover poll endpoint", "err", err)
return err
}
log.Debug("Connecting to ACS endpoint " + acsEndpoint)
url := AcsWsUrl(acsEndpoint, cfg.Cluster, args.ContainerInstanceArn, args.TaskEngine)
clearStrChannel(ackBuffer)
client := acsclient.New(url, cfg.AWSRegion, args.CredentialProvider, args.AcceptInvalidCert)
defer client.Close()
// Clear the ackbuffer whenever we get a new client because acks of
// messageids don't have any value across sessions
defer clearStrChannel(ackBuffer)
timer := ttime.AfterFunc(utils.AddJitter(heartbeatTimeout, heartbeatJitter), func() {
log.Warn("ACS Connection hasn't had any activity for too long; closing connection")
closeErr := client.Close()
if closeErr != nil {
log.Warn("Error disconnecting: " + closeErr.Error())
}
})
defer timer.Stop()
// Any message from the server resets the disconnect timeout
client.SetAnyRequestHandler(anyMessageHandler(timer))
client.AddRequestHandler(payloadMessageHandler(payloadBuffer))
// Ignore heartbeat messages; anyMessageHandler gets 'em
client.AddRequestHandler(func(*ecsacs.HeartbeatMessage) {})
updater.AddAgentUpdateHandlers(client, cfg, args.StateManager, args.TaskEngine)
err = client.Connect()
if err != nil {
log.Error("Error connecting to ACS: " + err.Error())
return err
}
ttime.AfterFunc(utils.AddJitter(heartbeatTimeout, heartbeatJitter), func() {
// If we do not have an error connecting and remain connected for at
// least 5 or so minutes, reset the backoff. This prevents disconnect
// errors that only happen infrequently from damaging the
// reconnectability as significantly.
backoff.Reset()
})
serveErr := make(chan error, 1)
go func() {
serveErr <- client.Serve()
}()
for {
select {
case mid := <-ackBuffer:
ackMessageId(client, cfg.Cluster, args.ContainerInstanceArn, mid)
case <-ctx.Done():
return ctx.Err()
case err := <-serveErr:
return err
}
}
}()
select {
case <-ctx.Done():
return ctx.Err()
default:
}
if acsError == nil || acsError == io.EOF {
backoff.Reset()
} else {
log.Info("Error from acs; backing off", "err", acsError)
ttime.Sleep(backoff.Duration())
}
}
}