There's currently no automatic recovery for disconnected databases in the evaluator. This means if the database is ever temporarily unavailable, hydra-evaluator will sit and spin with no work accomplished. If this condition is caught, the daemon will exit and systemd will be responsible for resuming the service.