Description of problem:

Some time ago, a test ([sig-arch] all leases in ns/%s must gracefully release) checking for graceful lease releases was added to openshift/origin.

The test verifies whether components/operators release their leases gracefully so that the next instance/replica can quickly acquire a new lease and start its work.

However, based on this query, it appears that the test frequently 'fails' for various operators.

Here is an example of a failed CI run where the 'openshift-kube-apiserver-operator' waited 144 seconds before acquiring a lease.

For operators using library-go, the worst non-graceful lease acquisition time is 2m43s, while the worst graceful lease acquisition time is 26s.

I'm also attaching a timeline from the same run, which shows that lease acquisition took longer for other components around the same time.

We should investigate why, in some cases, components/operators take longer to acquire their leases.

The suspicion is that leases are not released gracefully due to server unavailability, which may result from etcd unavailability. However, I don't have data to support this theory.

Note: This issue might be connected to https://issues.redhat.com/browse/OCPBUGS-42087 and a few other failures. It is possible that all of them share the same root cause.