Persistent + discoverable logs for Docker services/containers.

[markfinger]

One project recently had an issue associated an ongoing celery task and we found it impossible to dig up the logs from more than the previous 30 mins.

It would be great to have rotated log files from the previous few days.

[mrmachine]

Great idea. But this is more than just celery. I think the recommended pattern is for each container to run one thing and send its logs to stdout/stderr (we do this now) and then have another logging service run that collects the stdout/stderr from all the other containers then we have one place to search/filter all logs and they persist. This is the part we haven't implemented yet, and it's not a problem with celery but all services running in Docker. We should bump this up the priority list since it is starting to hurt us.

[markfinger]

Re: #142 (comment)