See https://prow.ci.openshift.org/view/gs/origin-ci-test/logs/periodic-ci-openshift-knative-serverless-operator-main-4.7-upgrade-tests-aws-ocp-47-continuous/1448257360601550848 for example.

I believe this is because we don't have any watchers setup for the knative-serving-ingress deployments and hence it's possible that we detect them as "down" but don't get poked again to check if they're actually up.