Kubernetes RollingUpdate - how to respect readinessProbe?

We have deployments with the following strategy:

      replicas: 6
          type: RollingUpdate  
            maxUnavailable: 1  
            maxSurge: 100%  

The deployments also have readiness (and liveness) probes:

              path: /  
              port: http  
            initialDelaySeconds: 30  

The issue I am seeing is that Kubernetes instantly terminates the old replicas when doing an upgrade. It does not wait for the new pods to become ready before moving on.

What am I missing here?

Hi! Did you ever figure this out? If so, what is the solution?