This looks like it might go green, but it was responsible for a verbose error being logged on master when a Fiber executor job was rejected.

The issues with the original PR are explained here: #5437 (comment)

We'll get this into 9.2.8

@Adithya-copart I'm not sure where we left this. I filed these changes before you came back with #5437 (comment) and suggested the original issue be closed. Is the additional work in this PR no longer necessary?

Detargeting since it's unclear if this is appropriate or necessary. Will be closed without further follow-up.

@headius This PR is no longer necessary as the main thread actually waits for the signal executors.

The issue I came across is that the main thread is being shutdown within the trap handlers. The at_exit blocks were partially executed which I mistakenly assumed to be a race condition between traps and at_exit handlers.

It was later fixed in puma here.

Signal.trap('INT') {p 'In trap'; Thread.main.kill; exit}

at_exit do
  # This gets printed
  p 'Inside at exit...'
  sleep 2
  # Never gets printed
  p 'Finished..'
end
p 'sleeping...'

sleep

@Adithya-copart Ok thank you. We are debating whether to include PR #5918 which might help such situations by warning that there's unterminated threads at exit.