https://bugzilla.redhat.com/show_bug.cgi?id=1096273
--- Comment #6 from Matthew Heon mheon@redhat.com --- So, you're not actually forwarding the signal to the container with --sig-proxy.
Doing a ps -ax | grep docker does not actually find the PID you use to forward signals to the container using --sig-proxy. Docker furnishes the PID of the container itself, via the docker inspect command (docker inspect $CONTAINER_ID | grep Pid). The given PID will not match the one found via ps.
I believe you're stopping and restarting some internal Docker process responsible for managing the container. This is not something that is intended to happen, or something we should support - I imagine Apache would also start throwing errors if we began sending signals to its forked processes.
(For reference, I can reproduce perfectly with your instructions. However, by changing ps -ax to docker inspect, the issue no longer occurs)
This could likely use some documentation to make it clear that ps is not a supported method of finding a Docker container's PID (at least, with any degree of reliability). I'll investigate existing docs to see if they address this.