Hands-On Microservices with Kubernetes

Recreating deployment

A trivial yet naive way to do this is to terminate all the pods running version X, and then create a new deployment where the image version in the pod template spec is set to X+1. There are a couple of problems with this approach:

• The service will be unavailable until the new pods come online.
• If the new version has issues, the service will be unavailable until the process is reversed (ignoring errors and data corruption).

The Recreate deployment strategy is appropriate for development, or when you prefer to have a short outage, but ensure that there is no mix of versions that are live at the same time. The short outage may be acceptable, for example, if the service pulls its work from a queue and there are no adverse ...