Server reader thread crashes on exit with Puma in cluster mode

[Draiken]

Your environment

• ruby -v: ruby 3.3.9 (2025-07-24 revision f5c772fc7c) [x86_64-linux]
• rdbg -v: rdbg 1.11.0

Describe the bug
When running a Rails server (puma) with RUBY_DEBUG_OPEN=true, it seems to work normally until you stop the server and we have a crash from the server reader thread:

$ RUBY_DEBUG_OPEN=true bin/rails server
DEBUGGER: Debugger can attach via UNIX domain socket (/run/user/1000/rdbg-110109)
=> Booting Puma
=> Rails 8.0.2.1 application starting in development 
=> Run `bin/rails server --help` for more startup options
[110109] Puma starting in cluster mode...
[110109] * Puma version: 6.6.1 ("Return to Forever")
[110109] * Ruby version: ruby 3.3.9 (2025-07-24 revision f5c772fc7c) [x86_64-linux]
[110109] *  Min threads: 5
[110109] *  Max threads: 5
[110109] *  Environment: development
[110109] *   Master PID: 110109
[110109] *      Workers: 1
[110109] *     Restarts: (✔) hot (✖) phased (✖) refork
[110109] * Preloading application
[110109] * Listening on http://127.0.0.1:3000
[110109] * Listening on http://[::1]:3000
[110109] Use Ctrl-C to stop
[110109] ! WARNING: Detected running cluster mode with 1 worker.
[110109] ! Running Puma in cluster mode with a single worker is often a misconfiguration.
[110109] ! Consider running Puma in single-mode (workers = 0) in order to reduce memory overhead.
[110109] ! Set the `silence_single_worker_warning` option to silence this warning message.
DEBUGGER[bin/rails#110151]: Debugger can attach via UNIX domain socket (/run/user/1000/rdbg-110109)
[110109] - Worker 0 (PID: 110151) booted in 0.0s, phase: 0
^C[110109] - Gracefully shutting down workers...
[110109] === puma shutdown: 2025-12-08 08:58:00 -0300 ===
[110109] - Goodbye!
Exiting
#<Thread:0x00007fcebcb66df8@DEBUGGER__::Server::reader /home/we/p/we/vendor/bundle/ruby/3.3.0/gems/debug-1.11.0/lib/debug/server.rb:44 aborting> terminated with exception (report_on_exception is true):
/home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1128:in `unlink': No such file or directory @ apply2files - /run/user/1000/rdbg-110109 (Errno::ENOENT)
	from /home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1128:in `ensure in unix_server_socket'
	from /home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1128:in `unix_server_socket'
	from /home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1169:in `unix_server_loop'
	from /home/we/p/we/vendor/bundle/ruby/3.3.0/gems/debug-1.11.0/lib/debug/server.rb:511:in `accept'
	from /home/we/p/we/vendor/bundle/ruby/3.3.0/gems/debug-1.11.0/lib/debug/server.rb:49:in `block in activate'
#<Thread:0x00007fcebcb66a10@DEBUGGER__::SESSION@server /home/felipe/p/recipal/vendor/bundle/ruby/3.3.0/gems/debug-1.11.0/lib/debug/session.rb:179 aborting> terminated with exception (report_on_exception is true):
/home/felipe/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1128:in `unlink': No such file or directory @ apply2files - /run/user/1000/rdbg-110109 (Errno::ENOENT)
	from /home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1128:in `ensure in unix_server_socket'
	from /home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1128:in `unix_server_socket'
	from /home/we/.rbenv/versions/3.3.9/lib/ruby/3.3.0/socket.rb:1169:in `unix_server_loop'
	from /home/we/p/we/vendor/bundle/ruby/3.3.0/gems/debug-1.11.0/lib/debug/server.rb:511:in `accept'
	from /home/we/p/we/vendor/bundle/ruby/3.3.0/gems/debug-1.11.0/lib/debug/server.rb:49:in `block in activate'

To Reproduce

I've manged to reproduce this running puma in clustered mode here: https://github.com/Draiken/debug-issue-unlink

Basically, start the server, interrupt it, we get the exception.

Expected behavior
No exceptions should occur.

Additional context
I imagine it's a race condition when attempting to remove the debug unix socket.

[ko1]

puma cluster uses SIGURG
https://github.com/puma/puma/blob/main/docs/signals.md
and debug.gem also use SIGURG.
I assume there is some contension...