Currently, instances with volumes have to be stopped before they're started, which can cause 1-2 seconds of downtime while the new instances starts

Resolving this would mean volume deploys are "zero downtime", and do not drop requests in between deployments of stateful services