CI: Skip TestIntegrationSingle#test_closed_listener on macOS

[dentarg]

In an effort to have green CI more often.

It was introduced in #2700

I've seen it fail like this on macos-10.15 and Ruby

• 2.2: https://github.com/puma/puma/runs/4055009489?check_suite_focus=true#step:9:739
• 2.3: https://github.com/puma/puma/runs/4055009508?check_suite_focus=true#step:9:661
• macos-11 2.4: https://github.com/puma/puma/runs/4055009711?check_suite_focus=true#step:9:667

      1) Error:
    TestIntegrationSingle#test_closed_listener:
    EOFError: end of file reached
        /Users/runner/work/puma/puma/test/helpers/integration.rb:160:in `readpartial'
        /Users/runner/work/puma/puma/test/helpers/integration.rb:160:in `block (2 levels) in read_body'
        /Users/runner/work/puma/puma/test/helpers/integration.rb:159:in `loop'
        /Users/runner/work/puma/puma/test/helpers/integration.rb:159:in `block in read_body'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:88:in `block in timeout'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:32:in `block in catch'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:32:in `catch'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:32:in `catch'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:103:in `timeout'
        /Users/runner/work/puma/puma/test/helpers/integration.rb:158:in `read_body'
        /Users/runner/work/puma/puma/test/test_integration_single.rb:187:in `test_closed_listener'
        /Users/runner/work/puma/puma/test/helper.rb:82:in `block (4 levels) in run'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:88:in `block in timeout'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:98:in `call'
        /Users/runner/hostedtoolcache/Ruby/2.2.10/x64/lib/ruby/2.2.0/timeout.rb:98:in `timeout'
        /Users/runner/work/puma/puma/test/helper.rb:80:in `block (3 levels) in run'

And like this on

• macos-10.15 Ruby 2.4: https://github.com/puma/puma/runs/4055009526?check_suite_focus=true#step:9:618
• macos-11 Ruby 2.3: https://github.com/MSP-Greg/puma/runs/3616720219?check_suite_focus=true#step:9:574 (when it was introduced)

      1) Error:
    TestIntegrationSingle#test_closed_listener:
    TimeoutEveryTestCase::TestTookTooLong: execution expired
        /Users/runner/work/puma/puma/test/helpers/integration.rb:124:in `gets'
        /Users/runner/work/puma/puma/test/helpers/integration.rb:124:in `connect'
        /Users/runner/work/puma/puma/test/test_integration_single.rb:187:in `test_closed_listener'
        /Users/runner/work/puma/puma/test/helper.rb:82:in `block (4 levels) in run'
        /Users/runner/hostedtoolcache/Ruby/2.4.10/x64/lib/ruby/2.4.0/timeout.rb:103:in `timeout'
        /Users/runner/work/puma/puma/test/helper.rb:80:in `block (3 levels) in run'

Also: Can't see that Puma::IS_OSX has ever existed, so fix that (changed in #2576).

[MSP-Greg]

@dentarg

Sorry for the error with Puma::IS_OSX. I was porting code from the 'new test framework' (which I really need to get back to working on). One issue is the TestIntegration#connect method writes and reads the client socket, so it's messy to separate read vs write errors. The new test framework split that...

Anyway, can you have look at the first two commits in https://github.com/MSP-Greg/puma/commits/fix-ci? Ran it twice in CI, didn't see the failure addressed here, but there's more.

I've noticed with some cron CI I have that Actions seems to be running slower. I think all platforms share 'runner' code, and maybe it's 'interrupting' the 'code under test'. Some of the IO.open IO's are having errors on read, which I don't recall seeing before. I'm going to see if adding a delay for darwin helps...

[dentarg]

@MSP-Greg Those commits look great, thanks for doing that, if open a PR we can merge them instead of this one

[dentarg]

I think the performance of Actions runs can vary widely (VMs on shared hardware have that problem) and especially macOS seeing GitHub had a very slow onboarding of macOS 11, I suspect resources are limited in that area.

[MSP-Greg]

@dentarg See PR #2739

I think the performance of Actions runs can vary widely

Agreed. Lately, it seems like CI is also running slower...