Currently when restarting the daemon (e.g. for an upgrade) our docker run processes just fail with:

2014-12-01_15:18:44.23585 2014/12/01 15:18:44 Cannot connect to the Docker daemon. Is 'docker -d' running on this host?

Which'll terminate the containers in a non-graceful way. This makes Docker upgrades non-trivial when you have many hosts, we essentially have to write our own wrapper which is even worse if you're using the official Debian package you have to override .postinst or install with --no-triggers and do the restart yourself. We have a script that stops all our containers gracefully (essentially docker stop), takes the node out of rotation (custom) and then restarts the daemon and starts the containers again. Overriding the init script doesn't really seem to be an option either, since it seems like something that'd change fairly often upstream.

The default seems extremely unsafe.

Don't know who to tag here, @crosbymichael @jfrazelle maybe you'll know? This is hurting us again as we want to test the RC in production.