Long-running containers

[manthey]

Currently, we start and stop containers for every job. Some tasks would benefit from long-running containers. These would need to list (a) how to run the container to persist, (b) the preferred shutdown command (and possibly a timeout after which it would be terminated anyway), (c) how to run clis in the standing container, and (d) how long it can be used (e.g., max sub-tasks, max idle time, max total time).

[manthey]

Necessarily, files to be transfers to and from the container would be within a persistent mount.