Output warning message if running version of Ruby capable of production Segmentation Fault

[tsully193]

Recently we saw the Ruby agent of APM cause segmentation fault in one of our apps running version 2.5.3. A sanity check that either prevents the agent from starting or at least outputs a warning message about the possibility of segmentation fault occurring if you are running version 2.5-2.5.3 of Ruby would be greatly appreciated.

[mikker]

Hi @tsully193! This sounds like it could be this bug in Ruby itself: https://bugs.ruby-lang.org/issues/14634.

[NikolaiGulatz]

We also noticed segmentation faults in ruby ruby 2.5.3p105 (2018-10-18 revision 65156) [x86_64-linux]:

App 902879 output: /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/transport/base.rb:61: [BUG] Segmentation fault at 0x0000000000000008
App 902879 output: ruby 2.5.3p105 (2018-10-18 revision 65156) [x86_64-linux]
App 902879 output:
App 902879 output: -- Control frame information -----------------------------------------------
App 902879 output: c:0080 p:---- s:0542 e:000541 CFUNC  :push
App 902879 output: c:0079 p:0038 s:0536 e:000535 METHOD /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/transport/base.rb:61
App 902879 output: c:0078 p:0009 s:0530 e:000529 METHOD /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/agent.rb:114
App 902879 output: c:0077 p:0007 s:0525 e:000524 BLOCK  /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/agent.rb:71 [FINISH]
App 902879 output: c:0076 p:0029 s:0521 e:000520 METHOD /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/instrumenter.rb:181
App 902879 output: c:0075 p:0007 s:0516 e:000515 METHOD /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/agent.rb:162
App 902879 output: c:0074 p:0064 s:0512 e:000511 METHOD /var/www/app/shared/bundle/ruby/2.5.0/gems/elastic-apm-2.9.1/lib/elastic_apm/subscriber.rb:57

the fix for the above mentioned bug seems to have been backported to 2.5 here.

[mikker]

What are your server setup(s)? Are either of you using a threaded server like Puma? Do you set it to preload_app!?

[NikolaiGulatz]

We are using passenger with passenger_spawn_method smart (https://www.phusionpassenger.com/library/indepth/ruby/spawn_methods/) - no multithreading.

When I change the spawn method to direct it seems to not cause segmentation faults. This however leads to a greatly increased startup time.

I think this behaviour should be documented somewhere even though it's a ruby bug since the result of the segmentation fault is a lot of requests failing with a 502 after integrating the gem.

[mikker]

I have added a Passenger based env to our test project to try and reproduce this but haven't been able to yet. Do you see any differences between that and your setup that could potentially get the two closer?

See elastic/opbeans-ruby#13

[mikker]

When either of you have the chance to test whether this bug was also fixed for you, please report back here 😊
Feel free to reopen if this problem persists.