Issue #20670 has been updated by jhawthorn (John Hawthorn). Status changed from Open to Closed Backport changed from 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN to 3.1: DONTNEED, 3.2: DONTNEED, 3.3: REQUIRED Fixed by https://github.com/ruby/ruby/commit/87a85550edd786665e081b355c6af62c4854b1d7 ---------------------------------------- Bug #20670: fork deadlocks in child process due to timer thread https://bugs.ruby-lang.org/issues/20670#change-109410 * Author: jhawthorn (John Hawthorn) * Status: Closed * ruby -v: ruby 3.3.4 (2024-07-09 revision be1089c8ec) [x86_64-linux] * Backport: 3.1: DONTNEED, 3.2: DONTNEED, 3.3: REQUIRED ---------------------------------------- We've been seeing an occasional failure in the Rails CI related to a test which forks and I managed to reduce it to the following reproduction. ``` ruby Thread.new do loop { sleep 0.0001 } end 1000.times do pid = fork{} Process.waitpid(pid) rescue Exception Process.kill(:KILL, pid) raise end ``` This hangs on Ruby 3.3 and HEAD (_fairly_ reliably), but completes always on Ruby 3.2 In a debugger it seems like the timer thread acquires `vm->ractor.sched.lock` in the parent process just as the process is forking. The child process then ends up stuck inside of `thread_sched_atfork` when trying to acquire the same lock. I've opened https://github.com/ruby/ruby/pull/11356 with a fix -- https://bugs.ruby-lang.org/

Issue #20670 has been updated by ioquatix (Samuel Williams). IIUC, the tests introduced here are now failing again on `master`: ``` btest-ruby Fstderr output is not empty bootstraptest.test_fork.rb_78_287.rb:16:in 'block in <main>': failed (RuntimeError) from <internal:numeric>:257:in 'Integer#times' from bootstraptest.test_fork.rb_78_287.rb:10:in '<main>' #287 test_fork.rb:78: def now = Process.clock_gettime(Process::CLOCK_MONOTONIC) Thread.new do loop { sleep 0.0001 } end 10.times do pid = fork{ exit!(0) } deadline = now + 10 until Process.waitpid(pid, Process::WNOHANG) if now > deadline Process.kill(:KILL, pid) raise "failed" end sleep 0.001 end rescue NotImplementedError end :ok #=> "" (expected "ok") [Bug #20670] FAIL 1/1910 tests failed make: *** [uncommon.mk:894: yes-btest-ruby] Error 1 /tmp/tmp.xWdROBpKaN /tmp/tmp.xWdROBpKaN /github/workspace/src /github/workspace ``` https://github.com/ruby/ruby/actions/runs/14010663767/job/39230065916#step:9... Maybe there has been a regression? I'm just checking if it's a timeout issue: https://github.com/ruby/ruby/pull/12962 however I suspect it's a deadlock. ---------------------------------------- Bug #20670: fork deadlocks in child process due to timer thread https://bugs.ruby-lang.org/issues/20670#change-112418 * Author: jhawthorn (John Hawthorn) * Status: Closed * ruby -v: ruby 3.3.4 (2024-07-09 revision be1089c8ec) [x86_64-linux] * Backport: 3.1: DONTNEED, 3.2: DONTNEED, 3.3: DONE ---------------------------------------- We've been seeing an occasional failure in the Rails CI related to a test which forks and I managed to reduce it to the following reproduction. ``` ruby Thread.new do loop { sleep 0.0001 } end 1000.times do pid = fork{} Process.waitpid(pid) rescue Exception Process.kill(:KILL, pid) raise end ``` This hangs on Ruby 3.3 and HEAD (_fairly_ reliably), but completes always on Ruby 3.2 In a debugger it seems like the timer thread acquires `vm->ractor.sched.lock` in the parent process just as the process is forking. The child process then ends up stuck inside of `thread_sched_atfork` when trying to acquire the same lock. I've opened https://github.com/ruby/ruby/pull/11356 with a fix -- https://bugs.ruby-lang.org/