The need to update is currently checked from a timestamp of a fs layer of container and the last time the service was started.
In the following scenario
docker image is build at t0
service is started at t0+1
docker image is tagged as rod at t0+2
the deployer does not understand that the service should be restarted (<-- bug) because the fs layer timestamp does not change when tagging.
to fix the problem a hash should be stored to service metadata and the need to restart checks should be done against the hash and not a timestamp.