Hey @Stig_Christensen,
Once the critical error happens, you have to assume that the endpoint is dead. A normal critical error action on a backend endpoint SHOULD call Environment.FailFast() or similar to halt the process so that operations people can notice it’s dead and come to the rescue. The only reason an endpoint doesn’t do that by default is that it could be hosted in a webapp and it would be a poor experience to crash your website against your wishes because RabbitMQ is down.
Check out the Critical error documentation here.