Rollout is stuck when leader election is enabled

[liu-cong]

What happened:

When leader election is enabled with 3 EPP replicas, updating EPP deployment gets stuck. This is because when leader election is enabled, there is at most 1 ready replica at any given time. Killing the old leader to give the lease to a new leader would violate the default maxUnavailable=25% in k8s deployment rolling update.

What you expected to happen:

We would need the Recreate update policy to allow the old leader to be killed and let the new leader take over.

How to reproduce it (as minimally and precisely as possible):

Anything else we need to know?:

Environment:

• Kubernetes version (use kubectl version):
• Inference extension version (use git describe --tags --dirty --always):
• Cloud provider or hardware configuration:
• Install tools:
• Others: