The Policy Propagator retries individual actions when an error is encountered. If too many retries occur, it gives up and requeues the entire request:
https://github.com/open-cluster-management-io/governance-policy-propagator/blob/f610c4b09b0653ae15a0e87f59a57b25e25f5809/controllers/propagator/policy_controller.go#L157-L172

We need a count metric that records this per root policy to detect when errors are stuck in an infinite loop.