Improve Docker APIv2 container wait reliability

[matejvasek]

As of now podman uses polling to wait for a condition. This approach is however unreliable.

There is an issue when waiting for status change. There is a condition called next-exit, it basically waits for transition running->not-running. It's possible that for short living containers that the running state is never observed by polling. Even non-sleeping busy waiting.

We need to figure how to solve this.

[matejvasek]

I think @mheon had some ideas.

[matejvasek]

There may be some history of states in DB maybe?

[matejvasek]

I hoped this won't be issue in real use, unfortunately I was mistaken. This is causing me troubles for instance with pack where restorer runs to fast. Also some other observed this issue hashicorp/nomad-driver-podman#89.

[mheon]

Best way I can think of is the Events interface - the container exiting will produce a Died event. Listening for events from the specific container ID in question, and then returning when a Died event occurred (said event should include the exit code as well), seems the safest way.

[TomSweeneyRedHat]

LGTM
would like a @vrothberg or a @mheon head nod too

[vrothberg]

+1 on @mheon's proposal. Great idea!