Foreground containers should die gracefully when sent a kill signal

[gabrtv]

Issuing a CTRL-C (SIGINT) on a container running in the foreground doesn't kill it. This makes docker difficult to daemonize with utilities like upstart and supervisord.

# docker run lucid64-base sleep 60
^C
# docker ps
ID             IMAGE                 COMMAND     CREATED         STATUS         COMMENT     PORTS
b062d84ccae1   lucid64-base:latest   sleep 60    4 seconds ago   Up 4 seconds

Discussed with @shykes on IRC:

gabrtv
8:44 shykes: do you see an issue with foreground containers trapping SIGINT/SIGTERM and terminating gracefully? if not, i'd be willing to take a crack at it

shykes
8:44 gabrtv: no design issues per-se, just needs to be implemented
8:45 there are 2 ways: 1) intercept signals in the client, and translate them to 'docker kill' calls (easier) 2) bring back standalone mode (harder)

[adamjt]

+1 Would love to have this for easy interaction with Upstart.

[bfulton]

+1 for this, would also like to see support (ala #415) to control the SIGTERM ... SIGKILL interval.

[gabrtv]

I took a crack at implementing this, but failed quickly after getting lost in how rcli works. From what I could tell, we need the CLI to intercept the signal and then somehow forward a docker kill to the daemon.

I was planning on letting the remote API (#21) evolve a bit, then try again using that.

[patrickod]

Another +1 for this. I'd love to be able to configure containers to run on boot with upstart similar to the way I'm sure @adamjt is looking to use it. At the moment this doesn't seem very doable.

[rca]

+1 I am also looking to manage containers with upstart and/or supervisor. Thanks!

[warpfork]

@unclejack's work in https://github.com/unclejack/dockrun builds this kind of signal handling over docker.

[unclejack]

This functionality is going to be merged into docker, along with other needed features.
I'll be sending a PR for this feature soon.