worker_pool

the attempt to send will hang. (There are ways of turning that into a timeout if you need to.) In Erlang, you could try to create an 11th coordinator process, but you'll create a lot of additional traffic and take a latency hit on every request. (Plus, some of the naive ways you might implement that will get you in further performance trouble with mailbox scanning. This is going to be a non-trivial thing to build correctly.) Or you could just randomly pick and send, but if you pick one that's currently answering a slow request while others are available, you get spurious latency. It hasn't been a killer in my system, but it's been an annoyance. And to be clear, let me reiterate that having using both systems, I actually still like Erlang's behavior as a better default. It's really hard to deadlock Erlang... it's much easier to deadlock Go. The async message passing is, in my experience, much easier to use correctly without much thinking. But it would be nice to also have a "channel" in Erlang, despite the limitations it would have to have with not being able to go across nodes. In fact, technically speaking, subject to that limitation (and there are already a couple of other functions limited to the current node), there's no reason I could see why this couldn't be added to Erlang. See for instance: https://github.com/inaka/worker_pool , particularly the section discussion "strategy", and the discussion about "available_worker" and its discussion on performance implications, and its specific references to the other tricky edge cases to consider. Of course, programming languages being programming languages, someone else can do the work and you can just use it, but in Go, this behavior is quite trivial. (By contrast, in Go, proper asynchronous messaging in the Erlang style is a challenge. Whacking together a "slice" and some "messages" isn't that hard, but you've got other edge cases to consider... to say nothing of network transparency!)