Let's look at liveness probes. Attched there is an updated app.yml with a livenessProbe around.
Apply the new specification to the cluster and watch what happens with watch kubectl get pod.
There is something wrong with the setup still. The app seems to not work anymore. What could be wrong?
Fix the liveliness probe, so that it works.
Verify your probe by hitting the Set 400 button in the training app to start sending
bad responses on the /status endpoint. Watch again what happens with the pods.
Add a readyness probe with the /ready endpoint.