Cloud Run traffic routing and prior revision rollback behavior

The goal is to determine the reliability of reversing a failed Cloud Run deployment by manually shifting 100% of traffic back to a previous immutable revision.

In fully managed Cloud Run, the service does not natively migrate traffic away from a new revision if the container instances crash or return consistent 5xx errors. Traffic remains assigned to the latest revision until the service traffic configuration is explicitly updated via the API or console.

There is uncertainty regarding the propagation latency of traffic updates across multiple regions; if a shift is initiated, the failed revision may continue to receive some requests during the transition. Additionally, the availability of a rollback depends entirely on whether the previous revision has been deleted during cleanup tasks.

Can traffic shifting to a prior revision be performed safely without a new deployment or image-rebuild? What are the specific indicators that confirm traffic has fully ceased reaching the failed revision? Does the deletion of old revisions immediately invalidate the ability to perform traffic-based rollback?