4.8 etcd operator watches ClusterVersion, and when it sees a Failing=True condition whose message contains the substring RecentBackup:

1. It starts taking a backup
2. It sets a RecentBackup=Unknown condition in its ClusterOperator, with a reason/message about progressing.
3. If the backup…

• succeeds, it sets RecentBackup=True.
  • If the CVO gets stuck on something else (e.g. release signature fetching), maybe enough time passes that the etcd operator thinks the backup is stale.  It can set RecentBackup=False with a reason/message complaining about that.  Re-trigger is probably oc adm upgrade --clear and try again.  Hopefully rare, because there will be no web-console re-trigger support.
• fails, it sets RecentBackup=False.  Maybe tries again?  If we don't auto-retry, admin could clear/re-request desiredUpdate to trigger a new attempt.

1. 4.9 etcd operator in all cases, and 4.8 etcd operator that sees a ClusterVersion Failing condition that is not True or whose message does not contain RecentBackup, clear the RecentBackup condition from its ClusterOperator.