Test server may have some issues on Apple Silicon / macOS

[tagliala]

Hello,

I'm trying to run specs on my local machine, however, the test server seems to start and I get a timeout after 1 minute

$ time bundle exec rake

An error occurred in a `before(:suite)` hook.
Failure/Error: res = ::Net::HTTP.get_response("localhost", '/__identify__', port)

Net::OpenTimeout:
  Failed to open TCP connection to localhost:3001 (execution expired)
# ./spec/support/localhost_server.rb:57:in 'LocalhostServer#booted?'
# ./spec/support/localhost_server.rb:31:in 'block in LocalhostServer#initialize'
# ./spec/support/localhost_server.rb:96:in 'LocalhostServer#wait_until'
# ./spec/support/localhost_server.rb:31:in 'LocalhostServer#initialize'
# ./spec/spec_helper.rb:21:in 'Class#new'
# ./spec/spec_helper.rb:21:in 'block (2 levels) in <top (required)>'

Finished in 1 minute 0.03 seconds (files took 0.1704 seconds to load)
0 examples, 0 failures, 1 error occurred outside of examples


real	1m0.534s
user	0m0.406s
sys	0m0.114s

If I interrupt CTRL+C the spec and start it again, there is a background process spinned which allows spec to run, and I get the following failures

Failures:

  1) Ethon::Easy::Http::Put#setup when body when requesting sending string body makes a put request
     Failure/Error: expect(easy.response_body).to include('"REQUEST_METHOD":"PUT"')
     
     Errno::EMFILE:
       Too many open files - getcwd
     # ./spec/ethon/easy/http/put_spec.rb:114:in 'block (6 levels) in <top (required)>'

  2) Ethon::Easy::Http::Put#setup when body when requesting sending string body submits a body
     Failure/Error: expect(easy.response_body).to include('"body":"a=1%26b%3D2"')
     
     Errno::EMFILE:
       Too many open files - getcwd
     # ./spec/ethon/easy/http/put_spec.rb:118:in 'block (6 levels) in <top (required)>'

Finished in 3.47 seconds (files took 0.17487 seconds to load)
579 examples, 2 failures, 3 pending

Failed examples:

rspec ./spec/ethon/easy/http/put_spec.rb:113 # Ethon::Easy::Http::Put#setup when body when requesting sending string body makes a put request
rspec ./spec/ethon/easy/http/put_spec.rb:117 # Ethon::Easy::Http::Put#setup when body when requesting sending string body submits a body

• CPU: Apple Silicon M1
• Ruby: 3.4.5
• macOS: 15.6 (24G84)

[felipedmesquita]

Thanks for the report!
I can reproduce it on an Intel Mac (Ventura 13.7.6) as well. I'm not sure if the test server is entirely to blame, but I'm suspicious it might be causing this and the segfaults in CI. Will investigate further on Monday once I'm back home.

[tagliala]

Welcome. I've tried #249 but that does not fix. I'm also trying to switching to puma, but I do not have much experience

I'm also experimenting in making copilot doing that. Since that is non-production code, it should be fine

[felipedmesquita]

My thoughts exactly. Maybe we can vibe code a go program and ship a built executable for the test server. Would be nice for it to be unaffected by the different Ruby versions in the test matrix as well.

[voxik]

Welcome. I've tried #249 but that does not fix. I'm also trying to switching to puma, but I do not have much experience

With #249, the switch would be as easy as

1. having Puma installed and then set the RACKUP_HANDLER=puma env variable prior executing the test suite or
2. having only Puma exclusively would be enough

[tagliala]

I've asked Copilot to switch to Puma, but didn't succeed and in its implementation it also went for puma + webrick (for some reasons)

There were many failures with puma, one is also noted in the commit message:

Keep WEBrick for tests since it supports custom methods (e.g. PURGE) that stricter servers reject.

[felipedmesquita]

@voxik the code as is on #249 causes segfaults intermittently even on my machine using ruby 3.4.5 on linux - the setup in the master branch also causes segfaults sometimes, but only in CI. To verify that the segfaults in CI are actually caused by the test server setup, I've opened #263 and ran all the checks 3 times. It didn't segfault once on any of the 60 checks.

I don't plan on switching our test server to a different language, and hope to land on something closer to the approach in #262, where we run a nice Sinatra server from a fixed (3.4) Ruby version - so we don't need to worry about old (or multiple) version compatibility for the test server. But it we need it to stop causing segfaults in CI, and the setup in #262 still (rarely) causes them https://github.com/typhoeus/ethon/actions/runs/17219004532/job/48849590786?pr=262

[felipedmesquita]

In summary, I think we should have a simpler test server setup, with fewer moving parts: ruby version, sinatra version, rack version, webserver (puma/webrick). Puma would be fine, though I'm not sure what we gain by using a production grade webserver instead of webrick. I've also ran into the problem @tagliala mentioned, that it seems puma flat out rejects PURGE requests at the HTTP level.

@voxik, let me take this opportunity to thank you again for the work on the test server while we are figuring it out! If you'd like to pair on this over zoom/meet feel free to reach out over email to felipemesquita at hey dot com