PG::UnableToSend: SSL SYSCALL error: EOF detected

[mrbrdo]

Hey,

I've been recently getting this error, after I changed from Heroku's database to self-hosted (on same machine, it's a very low traffic site atm). FYI I am running sidekiq as well, and no ActiveRecord (its railtie is not loaded).

While trying to figure this out I stumbed on an issue in ActiveRecord:
rails/rails#12867

In case this gives any clue as to what may be going on. I don't really know what it could be, but it usually happens after I deploy. I am using puma and phased restart, is it possible I am missing some after fork settings for puma? Running MRI 2.0 in this case. I usually run JRuby and don't remember having this problem there (similar setup with puma).

Thank you

[jeremyevans]

Can you better describe the behavior you are getting? If it is:

1. underlying connction drops
2. error raised when query attempted
3. new connection established on next query

Then Sequel is operating as designed. If however the connection is not recognized as a disconnect, and you continually receive the error every time you try to issue a query, then that is an issue.

You should post the error class that Sequel used when wrapping the exception. If it is Sequel::DatabaseDisconnectError, then there is no problem. If it is just Sequel::DatabaseError, then that is a problem.

[mrbrdo]

Sorry about that, yes it is a Sequel::DatabaseDisconnectError. It seems to happen on the first requests I do to my web app after I phased-restart puma (the app server), and the requests fail. However on subsequent requests it seems the connection is re-established automatically and there are no more problems. Since I am on MRI in this case I am wondering if I am missing some settings for puma.

For example this is what they recommend when using ActiveRecord:

on_worker_boot do
  ActiveSupport.on_load(:active_record) do
    ActiveRecord::Base.establish_connection
  end
end

I don't think I had this problem when running on JRuby.

[jeremyevans]

As mentioned in the documentation (http://sequel.jeremyevans.net/rdoc/files/doc/code_order_rdoc.html), if you are using a preforking web server, you need to disconnect before fork. I know unicorn offers a before_fork hook, I'm not sure if puma does. If not, you just need to disconnect after your app loads.

As this is expected Sequel behavior, I'm closing this issue.

[mrbrdo]

Thanks for the clarification!