This really depends on what you mean when you say, “ensure”. How important is it that you only ever have one copy of a worker? There are loads of libraries out there to help solve this problem: swarm, syn, gproc, or even :global. Each of these libraries have different tradeoffs and guarantees. They handle things like network partitions and node failures differently. It’ll really come down to the kinds of guarantees you need. If you have a small and relatively stable set of nodes and can tolerate occasionally having “duplicate” workers then I’d look at swarm. If you need somewhat stricter guarantees then you might look at gproc. If you need even stricter guarantees you might want to use raft or better yet use an external data store like redis. My intuition (which should be taken with a massive grain of salt since I don’t know your exact use case) is that you’re probably best off using something like swarm. You can use “at least once” messaging guarantees over whatever transport you’re using and work to make your downstream service idempotent.

Would Raft be applicable to your use case?