This is probably an edge case that originally comes from multicluster observability. The problem is that the COO lets the OLM generate an 'OperatorGroup' resource, which leads to a resource called 'openshift-cluster-observability-operator-zsxqz', for example.

This limits the MCO/MCOA, which attempts to install an OperatorGroup with the name 'openshift-cluster-observability-operator' (you have a spoke/managed cluster in ACM with COO already installed).

All of this marks the COO as failed in the UI and leads to the `CsvAbnormalFailedOver2Min` alert. CSV conditions also report:

 message: 'csv created in namespace with multiple operatorgroups, can''t pick one automatically'
 phase: Failed
 reason: TooManyOperatorGroups

The workaround is simply to delete one of the operator groups.