When a container restart policy kicks in, there is a networking issue when the container restart is attempted, here is the error message in the daemon logs: ERRO[0015] Error running container: [8] System error: failed to set sandbox key : already assigned

This happens at each restart attempt (by the restart policy).

Repro:

docker run -d --restart=always busybox sh -c 'exit 0'

Then check the daemon logs.
Note that even when the container never actually starts again due to the network error which you can also see by doing:

docker run -d --restart=always busybox sh -c "trap 'exit 0' TERM; while true; do sleep 1; done"

And then killing the container pid.

ping @mavenugo