Autoscaler pod restart results in temporary availability drop

[gerritvd]

We are running knative-serving 1.8.1. We have 3 autoscaler pods. What we noticed is that when k8s restarts one of the autoscaler pods, we temporarily (about 30s) have a significant availability drop. It seems that even deployments configured with autoscaling.knative.dev/target-burst-capacity: "0" for a brief moment route their traffic (or part of it) through the activator, possibly until one of the other pods has taken over the work of the swapped out autoscaler pod. I'm getting 503 errors back but not from the activator.
Is this the expected behavior or have we possibly misconfigured it?

[dprotaso]

Hey v1.8 is really old and not supported. Can you update and retest. thanks