- remove executions with fatal errors form the scheduling queue
- set services to error state when they fatally fail starting
- re-implement log streaming that was not working