segfault on ruby 2.5.1 (on travis-ci)

[DavidS]

* Operating system:                linux
* Ruby implementation:             Ruby 2.5.1
* `concurrent-ruby` version:       1.1.6
* `concurrent-ruby-ext` installed: no
* `concurrent-ruby-edge` used:   no

Running puppetlabs/puppet_litmus@1fd29fe on travis consistently gives the attached ruby segfault: log.txt

Running against 1 targets.
..../home/travis/build/puppetlabs/puppet_litmus/puppetlabs-motd/vendor/bundle/ruby/2.5.0/gems/concurrent-ruby-1.1.6/lib/concurrent-ruby/concurrent/atomic/ruby_thread_local_var.rb:113: [BUG] Segmentation fault at 0x00000000371bb696
ruby 2.5.1p57 (2018-03-29 revision 63029) [x86_64-linux]
[...]

The same code works fine on ruby 2.5.7.

[pitr-ch]

Thanks for the report. Few notes:

• There was similar error Segmentation fault with ruby-2.6.3 #808
• This is a bug in MRI no Ruby code should be able to segfault Ruby.
• The original version working on 2.5.1 was breaking on 2.6.3.

Further investigation is needed.

[pitr-ch]

@fzakaria would you be interested to take a look at this?
So far I know that Thread creation, Mutex, Queue cause problems and at the moment I do not have more ideas how to get around this.

[fzakaria]

Sure thing.
I'll build 2.51 from source & run the tests and see what I can uncover.

[fzakaria]

Few notes from looking into it:

1. @pitr-ch added some changes to address https://bugs.ruby-lang.org/issues/16288 as part of Safer finalizers #821

The change basically removed creating new threads in the finalizers since the finalizers may be called during shutdown & threads cannot be started.

2. This bug involves changes post the submission of the PR above; where the segfault is occurring within Queue#push
   To me that means either the Queue or the @index have been garbage collected already.

[pitr-ch]

Thanks for looking into this. I got an idea how to deal with it, please have a look #856

[pitr-ch]

@DavidS could you try the PR if it works for you?

[DavidS]

Apologies for the delay - I've been on vacation the last week. I'll see if I can run that PR through our CI systems, we've had quite a high hit-rate on the segfault.

[pitr-ch]

@DavidS No problem, thank you for giving it a try.

[schuylr]

Just want to confirm I'm also seeing this on Ruby 2.5.3 with Version 1.1.6 of the gem.