When started from aws-provisioner, docker-workers have afterIdleSeconds set. I'm not sure where this is getting set. It was set by the aws host type, which was removed by https://github.com/taskcluster/docker-worker/pull/513 and is not used with tc-worker-runner.

I do plan to set this from ci-config, but it would be good if the worker had sensible defaults; particularly as worker-manager doesn't currently terminate instances ever.

(In reply to Tom Prince [:tomprince] from comment #0)

I do plan to set this from ci-config, but it would be good if the worker had sensible defaults; particularly as worker-manager doesn't currently terminate instances ever.

Agreed. Since docker-worker-based workers tend to spin up relatively quickly, I'm happy to have the afterIdleSeconds default set pretty low, e.g. 5 minutes. We can increase it like we did in GCP when required.

docker-worker has a default value set in config.yml

(In reply to Wander Lairson Costa [:wcosta] from comment #2)

docker-worker has a default value set in config.yml

Experimentally, that was not working the aws-provider images.

https://github.com/taskcluster/docker-worker/pull/519