Upstart service doesn't respawn

[aws]

Upstart service doesn't respawn

[wild-endeavor]

@samuelkarp any ideas on this?

[pabloqc]

@wild-endeavor Could you check your etc/ecs/ecs.config file?
I had a similar issue and the reason was because of a mismatch with the cluster-name. If by any change you modify your ECS_CLUSTER key and it doesn't match with the one that you get with:
curl http://localhost:51678/v1/metadata then the ecs agent won't be able to start.

[Hildebrand]

I'm having the same issue. If I understand it correctly the ecs-init package makes sure that there's a clean runtime environment before spawning a new agent, and does the starting of the agent itself via the ecs upstart configuration and ultimately the ecs-init process.

Now, when restarting the docker daemon, the agent is not respawned. Worse, it seems the ecs-init process is killed.
Now I see two solutions. One is to make the ecs upstart script start depend on docker. This way, on daemon restarts, the ecs-agent would be spawned as well. Unfortunately though the AWS Linux AMI's docker from the repositories is managed through an init.d script, and as such we cannot depend on this service through ecs-init upstart configuration alone.

This leaves us with the next approach, being to to start the ecs-agent docker container with the restart=always policy. That would result in the same and make the setup way more resilient. I'm not sure though if it would be an issue to not have the ecs-init process running though.

What are your thoughts on this?

[fierlion]

For this functionality, I'd recommend upgrading to Amazon Linux 2 which uses systemd as its init process. Given the limitations of the version of upstart in Amazon Linux 1, it will be significant effort to define the depends on relationship between dockerd and ecs-init. At this point I don't see how it's possible but I'm still looking into solutions.
See my semi-detailed tour of ecs-init and docker with these respective init systems:
#231