If an event from the sourcese cannot be process then there should be a metric that can be easily monitored by an operator. Not all errors will trigger failure or a configuration can prevent a failure so it would be a useful numer to trigger alert on.