When maintenance: ha is removed, the kvm-ha-service have set forced_down=true on the nova-compute service. The operator's MaintenanceUnset path previously only re-enabled the service status but did not clear forced_down, leaving the compute service unable to accept workloads despite being marked enabled.