Removes containers that were never started? about docker-gc HOT 6 CLOSED

Yup, I believe that's a correct reading. I think it's something we never considered since we don't really create containers and leave them around for a while without starting them.

Though if you get unlucky with timing, I guess this could happen in the time between when you create a container and start it even if you plan on starting it immediately.

from docker-gc.

this relates to #23. Let me know if I should delete my issue and keep the discussion here.

it seems that although the data containers do not show up anymore with docker ps -a, the container that uses them as volumes still points to the same docker volume in docker inspect and the data in the volume still seems to be accessible.

from docker-gc.

If this is a critical issue for anyone, you might be interested in https://github.com/Yelp/docker-custodian (disclaimer: I'm the primary contributor), which has basically identical functionality to docker-gc (except that it's python not bash, has unit tests, and uses the api instead of the cli).

It handles this issue properly:
https://github.com/Yelp/docker-custodian/blob/v0.4.0/docker_custodian/docker_gc.py#L49

from docker-gc.

@dnephin that's amazing, thanks!

Does it leave Data Only containers alone by default?

from docker-gc.

It does not. I'll open an issue for that feature (I believe it may be the only one missing).

from docker-gc.

Though if you get unlucky with timing, I guess this could happen in the time between when you create a container and start it even if you plan on starting it immediately.

I think the addition of GRACE_PERIOD_SECONDS would remove this unlucky timing condition.

I am going to close this for now as I don't think there is a huge need to remove containers that were never started now that GRACE_PERIOD_SECONDS fixes the timing issue and the support for excluding certain container IDs exists.

from docker-gc.