Improve error reporting on failure

Currently error reporting includes all services that did not start, no matter how far removed from the actual problem these services are. For large deployments this can result in hundreds (or thousands) of lines of error reporting, most of which is irrelevant and says nothing about the actual problem.