...
In the second situation, where both liveness and readiness respond positive, the new pod will receive trafik. The one who started the deployment will look at cluster readout and decide a rollback. That person has to commit a new cluster configuration. This could be a reverse commit to the desired state file.
Note: if we are using canary Canary deployment, the negative impact of a software problem can be held at a minimum. That is if the negative impact can be found in prometheus.