Running Puma within a container

[ayushn21]

This is a follow-up to #34.

What's the best way (or is it not possible or not recommended?) to run a Puma server within an async container?

Some context as to why I'm doing this:

• Largely academic, to see if I can.
• In the future I plan to use Falcon which removes this issue, but all my current projects use Puma and I'd like to see if I can use async-container with those.
• I'd like to blog about a way for existing Ruby/Rails users to use async-container.

The two problems I'm encountering are:

• The container isn't notified of readiness preventing restarts using SIGHUP. As discussed in Restarting containers with arbitrary processes using SIGHUP #34, we need to manually notify the container that the process is ready. I don't have control over Puma's executable, so not sure how to do this?
• If I set ready: true when calling instance.exec to work around the above issue, sending SIGHUP crashes the process because the new process is trying to bind to the same UNIX socket as the existing process throwing an error. Again not sure what the best way to work around this is?

Code

Here's a simple prototype I threw together for this. Amend the UNIX socket path in start and then run bundle && bundle exec start.

async-container-puma.zip

[ioquatix]

I have added an example here showing how to do it: https://github.com/socketry/async-container/tree/main/examples/puma

Can you let me know if this works for you?

[ayushn21]

Thanks for this @ioquatix! I didn't know Puma had an on_booted hook I could use.

A couple of follow up questions on this:

• The server starts fine now, but restarts using SIGHUP still don't work.
  If it's bound to a port, I get the error:
  Address already in use - bind(2) for "0.0.0.0" port 9292 (Errno::EADDRINUSE).
  If I bind to a UNIX socket, I get the error:
  There is already a server bound to: /Users/Ayush/Developer/Prototypes/puma.sock (RuntimeError)

This isn't strictly in the scope of async-container, but curious to know if there's a solution to this? We can't have two Puma parent processes listening on the same port/socket can we? What would be the best way to handle this in your opinion?

• This is more of a hypothetical question. How would you recommend using async-container when running arbitrary executables. For eg, if we're running a binary that's not even in Ruby and that doesn't have any lifecycle hooks, how would we go about that because we wouldn't have a way of notifying async-container from within the process. Or is this not something we should use async-container for?