Any Async Runtimes for Linux?

Just wondering if anyone here is working on, or knows of, a callback oriented async runtime that can be used with sockets (ideally epoll, and friends)?

Might take a stab at it if it’s not already in the works, but want to see what’s available before I spend too much time on it.

1 Like

Not sure how I didn’t find it in my first couple of searches: found this though.

Still interested in hearing people’s opinion on other libraries. Also, what’s the status on compiler-level async? I joined zig after it had been rolled-back, just curious what’s up with it.

1 Like

That one’s not maintained.

The next compiler release that will contain async again remains undetermined.

See the latest Zig Roadmap presentation for the Core Team’s priorities:

I’d suggest looking at:

  • libxev
  • iofthetiger

I have an epoll based callback system for extreme low latency conditions that I wrote that specializes in event streaming with message sizes up to around 1k bytes (but works best when they are mostly below 300 bytes).

I was thinking of releasing it, but don’t know yet. It gets within 10% the latency target of the C++ one it is based on, so I still have some optimization work to do. I was never able to get that close in Rust with tokio and async (it was 25% off). I do still have do TLS/SSL, so that might hurt performance a little too. I’m just not sure how to do it yet. I might have to link openssl and write to its async api (always a huge mess ever time).

I am using event driven state machines.
Callback-oriented, epoll-based, sockets (or COM-ports), timers, signals, fs events, but I wouldnt call this “async”.
If you want “true” async ops, use io_uring.

1 Like


a couple of additions to my previous post.

  • here was an example of not so simple machine (connector to postgres dbms)
  • the only thing that you can do truly asynchronously (regarding sockets) with epoll (which is synchronous syscall) is establishing connection, you just get “ready for write” (EPOLLOUT) notification.

Funnily enough, I guess what I really was describing was in-fact an event driven state machine.

I hadn’t thought too deeply about the difference until you mentioned it. Looks interesting, I’ll take a look. Thanks!