We use unicorn + sinatra + tmm1/amqp, and I can say for a fact that you must
initialize the AMQP connection in after_fork. Eric is probably right about the
fd sharing from before_fork.
In the rails-specific context, if preload_app is true, it's possible that using
config.after_initialize {} will still result in the AMQP initialization
occurring before the fork.
cheers,
--jordan
On Nov 9, 2010, at 4:55 PM, Eric Wong wrote:
> Marcelo de Moraes Serpa <[email protected]> wrote:
>> Hi all,
>>
>> I'm having issues with Unicorn and connecting to
>> RabbitMQ using the tmm1-amqp gem.
>>
>> I've tried lots of approaches.
>>
>> The classic initializer with Thread.new { EM.run }
>> for the Rails app and even tried using
>> the Qusion library.
>>
>> (https://github.com/danielsdeleo/qusion)
>>
>> I've made a simple mod to Qusion for it to monkey patch
>> unicorn too.
>>
>> The code looks like:
>>
>> /vendor/plugins/qusion/lib/amqp.rb
>>
>> module AMQP
>> def self.start_web_dispatcher(amqp_settings={})
>> @settings = settings.merge(amqp_settings)
>> case Qusion::ServerSpy.server_type
>> when :passenger
>> PhusionPassenger.on_event(:starting_worker_process) do |forked|
>
> <side note>maybe adding a similar hook to Unicorn for library/app authors
> to use would be a good idea.</side note>
>
>> if forked
>> EM.kill_reactor
>> Thread.current[:mq], @conn = nil, nil
>> end
>> Thread.new { start }
>> die_gracefully_on_signal
>> end
>> when :standard
>> Thread.new { start }
>> die_gracefully_on_signal
>> when :evented
>> die_gracefully_on_signal
>> when :none # << HERE
>> Thread.new { start }
>> die_gracefully_on_signal
>> else
>> raise ArgumentError, "AMQP#start_web_dispatcher
>> requires an argument of
>> [:standard|:evented|:passenger|:none]"
>> end
>> end
>>
>> def self.die_gracefully_on_signal
>> Signal.trap("INT") { AMQP.stop { EM.stop } }
>> Signal.trap("TERM") { AMQP.stop { EM.stop } }
>> end
>
> Registering a standard at_exit {} block in the Unicorn
> after_fork hook would be a good idea than trapping INT
> and TERM, which Unicorn uses.
>
>> end
>>
>> See the :none case? It used to be empty and return nil.
>>
>> Now it has the same behavior as the non-
>> evented :standard server.
>>
>> The problem is that the "server_spy" wasn't
>> recognizing Unicorn. So far so
>> good.
>>
>> Originally, the AMQP/Qusion conf goes into config/environment.rb
>>
>> config.after_initialize do
>>
>> Qusion.start(:user => 'guest',:pass => 'mypass')
>> end
>>
>>
>> However, when I publish messages to the queue,
>> it just isn't posted. It silently fails.
>>
>> I've then tried to put this on config/unicorn.rb:
>>
>> before_fork do |server,worker|
>> Qusion.start(:user => 'guest',:pass => 'mypass')
>> MQ.new.queue('jobs').publish('hey!')
>> end
>>
>> And the hey! message is posted.
>>
>> However, it only works from here, the MQ.new.queue on the app code
>> fails silently.
>>
>> Any ideas on what might be happenig?
>
> Is preload_app true? Then you're probably trying to share sockets
> across processes and confusing EM or AMQP.
>
> Instead of using before_fork above, you should probably
> be using after_fork:
>
> before_fork do |server,worker|
> Qusion.start(:user => 'guest',:pass => 'mypass')
> MQ.new.queue('jobs').publish('hey!')
> end
>
> I'm not very familiar with AMQP, but the Unicorn process is strictly an
> AMQP sender/client and doesn't listen as a server on a port, right?
> If so, the above before_fork config *should* work for you.
>
> Let us know how it goes!
_______________________________________________
Unicorn mailing list - [email protected]
http://rubyforge.org/mailman/listinfo/mongrel-unicorn
Do not quote signatures (like this one) or top post when replying
