Extract asynchronous I/O layer (e.g. for Celluloid::IO support)

[tarcieri]

I'd like to use this as a master tracking ticket for all feature requests/regressions related to the asynchronous I/O layer/timeout implementation.

I think this is one of the most interesting parts of this gem (thanks @zanker) and the two of us have discussed extracting it into a separate gem:

https://github.com/zanker/socketry

The addition of this feature broke Celluloid::IO support. Once upon a time several people suggested to me Celluloid::IO support was the only interesting feature of this gem. I hope we've proved them wrong!

I would still like to support Celluloid::IO as a timeout backend, but it needs the work to add it to the existing timeout framework.

I am tagging this as a milestone 2.0 option for two options:

1. I don't think anyone plans on working on it soon
2. It would arguably be a breaking API change

[zanker]

Yea I'd like to work on Socketry but it looks like I'll not have time for a while :(

[ches]

I was following #225 but without much familiarity with the history I feel a little short on context from the description of this new ticket—I've swiftly browsed what code exists in the Socketry repo, but since it lacks specs and a high-level README description at this point, would you guys mind expanding a little either there or here on it when you have time? I.e. what is the intended focus/scope of Socketry? From @tarcieri's remarks above it sounds to me something like "uniform interface abstraction layer for Ruby asynchronous I/O", but "Socket wrapper around Ruby" sounds a bit conceptually different from that, so not sure that I understand correctly.

Would the idea be that httprb uses the Socketry API, then adapters can be implemented for Socketry to be backed with Celluloid::IO, EventMachine, perhaps Cool.io if someone has a legacy case to support, etc.?

[tarcieri]

@ches it'd basically be a wholesale extraction of the current timeout code, which has a synchronous API, but async I/O for implementing timeouts.

It would work with core Ruby and Celluloid::IO. It could not work with EventMachine and Cool.io, because they're callback-driven.

Right now it is very difficult to implement I/O with timeouts in Ruby without using core timeout.rb, which is unsafe due to its use of threads/Thread#raise.

The timeout code in this library implements them safely by using IO.select to wait for I/O readiness or a timeout.

Celluloid::IO does something similar using nio4r selectors.

[msufa]

We've actually went for this gem strictly because of Celluloid::IO support, and seeing it removed from v1.0 is a major issue for us. I'm not looking for commitments, just some general info to have a good understanding of the situation. Is there any rough timeframe for v2.0? Are there any alternatives / workarounds for making the gem work with Celluloid::IO?

[zanker]

@tarcieri do you remember what broke with Celluloid::IO? I thought it was due to the timeout portions, we could just make a Celluloid timeout class for people to use I think.

[tarcieri]

Yeah, we'd need a Celluloid::IO timeout backend

[zanker]

It's pretty easy to do, but I'm probably not going to have time to do it this week. I can take a look next week @msufa.

[msufa]

Excellent! Thanks for the support and swift response @zanker

[ccoenen]

@zanker did you get around to looking at it? I'm currently trying to assess the near-future state of Celluloid Support for this gem.

[tarcieri]

I've started working on it. I should have the basic pieces of it relatively soon

[tarcieri]

Here's a WIP: https://github.com/celluloid/socketry/pull/3

[tarcieri]

Socketry is shaping up now: https://github.com/socketry/socketry

It's mostly feature complete at this point. Still a little work to do.

I started trying to convert http.rb to use it. Note that it only provides an equivalent of HTTP::Timeout::PerOperation, so HTTP::Timeout::Global will need to stay in this library. But that makes sense (I think), because those timeouts need to have knowledge of the request lifecycle.

[tetherit]

Thank you for the work on this. Is there any experimental version yet to try this out? -- If not, can you recommend any high level HTTP library that works inside of Celluloid::IO?

Socketry looks great, but presumably it is pure access to the socket rather than an HTTP library.

[tarcieri]

You can try the socketry branch (from #377) although there's a bit more work involved to make it fully Celluloid::IO compatible.