I was running a cookbook which internally uses wait_for_optimal() (sre.maps.roll-restart), but all the hosts had a long-standing Icinga alert which was marked as acknowledged in Icinga (regarding Postgres replication lag). This leads to a situation that SREBatchBase attempts to retry wait_for_optimal() 15 times until one needs to approve manually to proceed.
I think it would be better if alerts marked as acknowlegded were not accounted for in wait_for_optimal().