Search code examples
rubyherokurakesignalsrufus-scheduler

Gracefully stop rufus scheduler on heroku worker

I run rufus-scheduler in a rake task that runs on a heroku worker. I get regularly SIGTERM exception, because of regular heroku dynos restart (see the heroku dyno docs). I'd like to implement graceful shutdown shown in the docs mentioned above and shutdown rufus scheduler during this process:

trap('TERM') do
  scheduler.shutdown(:kill)
  exit
end

However when I try to send SIGTERM to the process with this task, I get error:

can't be called from trap context

Is there any method to gracefully shutdown rufus scheduler on SIGTERM? I use ruby 2.0, rake 10.0.4, rufus-scheduler 3.0.2.

P.S. No, I can't use heroku scheduler, because I need to run this task every minute ;-).

EDIT (jmettraux)

test code: https://gist.github.com/jmettraux/a4c00374f58e9f7affa8

Ruby 2.0.0-p247, rufus-scheduler 3.0.5 on Debian GNU/Linux yields:

/home/jmettraux/w/rufus-scheduler/lib/rufus/scheduler/job_array.rb:74:
  in `synchronize': can't be called from trap context (ThreadError)
    from /home/jmettraux/w/rufus-scheduler/lib/rufus/scheduler/job_array.rb:74:in `to_a'
    from /home/jmettraux/w/rufus-scheduler/lib/rufus/scheduler.rb:276:in `jobs'
    from /home/jmettraux/w/rufus-scheduler/lib/rufus/scheduler.rb:127:in `shutdown'
    from t.rb:8:in `block in <main>'
    from t.rb:18:in `call'
    from t.rb:18:in `sleep'
    from t.rb:18:in `<main>'

Same platform, but with Ruby 1.9.3-p392 and it gracefully shuts down.

Solution

  • OK, it's Ruby 2.0 specific (https://www.ruby-forum.com/topic/4411227)

    The next version of rufus-scheduler will include a workaround.

    https://github.com/jmettraux/rufus-scheduler/issues/98

    Thanks for reporting the issue.