Let it crash is just the start of the process, really. Write the happy path, with pattern matching so that if things failed, it does crash, like ok = do_the_thing(). You're also expected to monitor crash reports and (live) update the system to handle crashes that need more specific handling than a simple restart.
In a distributed system you always need to handle the case where you sent a request and didn't get a response and the request may or may not have been processed. Once you accept that, it's totally reasonable for the request processor to crash during a request, because the requester always needs to be know what to do in case they get no response or a non-specific error.
In a distributed system you always need to handle the case where you sent a request and didn't get a response and the request may or may not have been processed. Once you accept that, it's totally reasonable for the request processor to crash during a request, because the requester always needs to be know what to do in case they get no response or a non-specific error.