Bug #20103: rb_internal_thread_remove_event_hook() hangs when called from within a Thread hook

• Status: Open
• Assignee: -
• Target version: -
• ruby -v: ruby 3.4.0dev (2023-12-28T14:53:15Z master e81a5453e3) [x86_64-linux]
• Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN

[osyoyu (Daisuke Aritomo)]

Problem

The Ruby process hangs when rb_internal_thread_remove_event_hook() is called from within a Thread hook registered by rb_internal_thread_add_event_hook().
The thread is waiting for the thread event execution lock (pthread_rwlock_wrlock(&rb_internal_thread_event_hooks_rw_lock))), which is acquired by its caller, rb_thread_execute_hooks().
https://github.com/ruby/ruby/blob/e81a5453e3c76c4348da042d86debde7689254fe/thread_pthread.c#L3323

This situation would occur when one wants to register a oneshot hook that performs something on a Thread for only a single time. The hook would remove itself after its main procedure.
In my case, I'm doing a timer_create(3) on each pthread for profiling purposes.

Expected behavior

The process won't hang, or at least print some kind of warning.

I think there some options here:

1. Discourage this kind of definition. Thread hooks should not add/remove other Thread hooks, or otherwise do so through other async methods, such as postponed jobs.
2. Find a way to make this work. Maybe inner calls to add/remove_event_hook() can share locks with the outer call, but I'm not sure if that's possible as I'm not familiar with pthread_rwlocks.

Repro

#include "ruby.h"
#include "ruby/thread.h"

rb_internal_thread_event_hook_t *hook;

void
event_hook_callback(rb_event_flag_t flag, const rb_internal_thread_event_data_t *data, void *custom_data)
{
  rb_internal_thread_remove_event_hook(hook); // hangs
}

RUBY_FUNC_EXPORTED void
Init_mycext(void)
{
  hook = rb_internal_thread_add_event_hook(
      event_hook_callback,
      RUBY_INTERNAL_THREAD_EVENT_RESUMED,
      NULL
  );
}

ruby -rmycext -e 'Thread.new { loop { 1 + 1 } }'