Puma problems: "pid is not listening for messages", "process already being traced?, "Address already in use - bind(2) for /tmp/rbtrace-47445.sock (Errno::EADDRINUSE)"

[bensheldon]

Apologies that this is slightly duplicative of #56, #87, but I figured I would share all of the issues that I'm experiencing when using rbtrace specifically with Puma. Maybe fixed by #86 but I'm unable to successfully build that (probably a me problem).

When booting Puma in single mode (e.g. no separate workers, one process), and in Development mode. All of these are using the command:

rbtrace --pid=<PID> --timeout=60 --eval true

First attempt:

• I try to connect via rbtrace before Puma receives its first web request, I receive Error: argument --pid (pid is not listening for messages, did you require "rbtrace").
• Then, if I make a web request, and then connect via rbtrace I get: Error: argument --pid (process already being traced?)
• Then, if I connect via rbtrace and then immediately reload the browser multiple times, I get: *** process 54112 is already being traced (54398 != 54484)
• And then I try to connect again and get

  /Users/bensheldon/.rbenv/versions/3.3.1/lib/ruby/gems/3.3.0/gems/rbtrace-0.5.1/lib/rbtrace/rbtracer.rb:48:in `bind': Address already in use - bind(2) for /tmp/rbtrace-56790.sock (Errno::EADDRINUSE)
  	from /Users/bensheldon/.rbenv/versions/3.3.1/lib/ruby/gems/3.3.0/gems/rbtrace-0.5.1/lib/rbtrace/rbtracer.rb:48:in `initialize'
  	from /Users/bensheldon/.rbenv/versions/3.3.1/lib/ruby/gems/3.3.0/gems/rbtrace-0.5.1/lib/rbtrace/cli.rb:428:in `new'
  	from /Users/bensheldon/.rbenv/versions/3.3.1/lib/ruby/gems/3.3.0/gems/rbtrace-0.5.1/lib/rbtrace/cli.rb:428:in `run'
  	from /Users/bensheldon/.rbenv/versions/3.3.1/lib/ruby/gems/3.3.0/gems/rbtrace-0.5.1/bin/rbtrace:5:in `<top (required)>'
  	from /Users/bensheldon/.rbenv/versions/3.3.1/bin/rbtrace:25:in `load'
  	from /Users/bensheldon/.rbenv/versions/3.3.1/bin/rbtrace:25:in `<main>'
  

• At this point, there is no receovery without restarting the process.

Trying a second time:

• I first make a web request to Puma after starting the process
• I run try to connect with rbtrace while also hammering Browser Refresh and get Error: argument --pid (pid is not listening for messages, did you require "rbtrace"`).
• I try again to connect, while also hammering Browser Refresh... *** attached to process 54900 🎉
• ....it hangs for that 60 second timeout.... *** timed out waiting for eval response ☹️

Trying a third time:

• In this case, I try to connect while also continuously hammering Browser Refresh:

  *** attached to process 54900
  >> true
  => true
  *** detached from process 54900
  

🎉 🎉 🎉

Observation: For reasons I don't know, the Puma process requires continual web requests to allow it to be connected to, and to subsequently complete and detach. Otherwise it gets left in a bad state that prevents further connections without restarting the process.

[frederikspang]

We're seeing the exact same problems - As if rbtracr cannot Connect while puma thread is sleeping/awaiting requests.
When running with ab while tracing, it worked 99% of the time.

Maybe Multi-process/multi-threaded only - I don't recall seeing the same issue locally.

For context - this was inside a container in kubernetes. Not sure it matters.

[smathieu]

We're also seeing the same issue with Puma 6.4.2. I'm seeing this both locally (MacOS) & on Heroku.