Logging SignalException when gracefully shutting down Puma on Heroku

[calleluks]

Hello!

I'm monitoring a Rails 5.1.5 app running on Heroku with Honeybadger using version 3.3.0 of the gem.

The app uses the default Rails puma server configuration.

For some time now, Honeybadger has been logging a SignalException every time I deploy the application. Here's an example of what the Heroku log output looks like when the app restarts after a deploy:

Feb 23 14:57:42 heroku/web.1:  Restarting 
Feb 23 14:57:42 heroku/web.1:  State changed from up to starting 
Feb 23 14:57:43 heroku/web.1:  Stopping all processes with SIGTERM 
Feb 23 14:57:43 app/web.1:  - Gracefully stopping, waiting for requests to finish 
Feb 23 14:57:43 app/web.1:  === puma shutdown: 2018-02-23 13:57:42 +0000 === 
Feb 23 14:57:43 app/web.1:  - Goodbye! 
Feb 23 14:57:43 app/web.1:  Exiting 
Feb 23 14:57:43 app/web.1:  I, [2018-02-23T13:57:42.976150 #4]  INFO -- : ** [Honeybadger] Reporting error id=REDACTED level=1 pid=4 
Feb 23 14:57:44 heroku/web.1:  Process exited with status 143 
Feb 23 14:57:44 app/web.1:  I, [2018-02-23T13:57:43.342196 #4]  INFO -- : ** [Honeybadger] Success ⚡ https://app.honeybadger.io/notice/REDACTED id=REDACTED code=201 level=1 pid=4 

I think this might be related to this change in Puma which aims to properly handle the TERM signal by re-raising it and letting Ruby's built in handling exit with the correct status code:

https://github.com/puma/puma/blob/dc9fa77f855c5018a03430366ea8c8db17fbfeea/lib/puma/launcher.rb#L397

~$ irb
irb(main):001:0> raise SignalException, 'SIGTERM'
Terminated: 15
~$ echo $?
143

Having this exception logged every time I deploy is not what I want but I'm not sure what needs to be fixed here. Should Honeybadger not rescue this exception? Should I ignore it manually in my configuration? Would I then risk missing a legitimate application error?

[minhajuddin]

This is an interesting situation, let me look at our code and see what is possible.

[calleluks]

Thanks, @minhajuddin!

[ckersh]

I am seeing a similar issue using Rails 5.1.4, Passenger 5.2.1, and Honeybadger 3.2.0.

[minhajuddin]

In our at_exit callback (

honeybadger-ruby/lib/honeybadger/singleton.rb

Line 67 in 4499a73

if $! && !$!.is_a?(SystemExit) && Honeybadger.config[:'exceptions.notify_at_exit']

) we ignore SystemExit exceptions, adding another clause for SignalException caused by a SIGTERM should fix this issue.

SIGTERM is usually sent by admins when they want to kill processes and not by the kernel. So, it should be safe for us to ignore this. Thoughts? @joshuap

[joshuap]

I think that would be OK, @minhajuddin 👍