Did you know the aws library uses Kernel.sleep for exponential backoff? about em_aws

[JoshMcKin]

Looks like its used in 3 places on master. Which was killing your performance?

https://github.com/amazonwebservices/aws-sdk-for-ruby/blob/master/lib/aws/sqs/queue.rb#L266

https://github.com/amazonwebservices/aws-sdk-for-ruby/blob/master/lib/aws/core/http/handler.rb#L69

https://github.com/amazonwebservices/aws-sdk-for-ruby/blob/master/lib/aws/core/client.rb#L241

[JoshMcKin]

Also, why not use EM::Synchrony.sleep?

https://github.com/igrigorik/em-synchrony/blob/master/lib/em-synchrony.rb#L70

[JoshMcKin]

Need some opinions on this patch...its pretty cooky but it will end the mutex and sleep issues:

require 'em-synchrony'
require 'em-synchrony/thread'
module AWS

  # Use EM::Synchrony sleep
  DupKernel = Kernel

  class PatchKernel
    class << self
      def sleep(count)
        EM::Synchrony.sleep(count)
      end
      def method_missing(method,*args,&block)
        DupKernel.send(method, *args, &block)
      end
    end
  end

  Kernel = PatchKernel

  # Use a fiber safe mutex
  Mutex = EM::Synchrony::Thread::Mutex 
end

[JoshMcKin]

Try 0.1.5, it should resolve this issue.

[johnkchow]

Have you guys seen any warnings when you re-set the constants? i.e.

Kernel = PatchKernel

A suggestion is to use Object.remove_const(:Kernel) right before so that there'd be no Ruby warning. Not that it's functionally any different of course :)

[JoshMcKin]

Yeah....this was not one of my better moments, but what patch really is ;) Let me see what I can do to pretty that up.

[JoshMcKin]

Strangely though, I do not get the overriding constant warning in MRI 1.9.2.

[JoshMcKin]

@jkchow can you give master a try?

Thanks

[johnkchow]

Looks good on my end. By the way, great job w/ the gem overall!

[johnkchow]

By the way, I don't get the overriding constant error either when I try requiring the older version in IRB. shrugs oh wells

[johnkchow]

After some thought, this patch might be too heavy. For one, there could be that rare edge case where users may actually want to Kernel.sleep or normal mutex (say they're using a script that first calls AWS via EM, then call EM.stop, then later on call Kernel.sleep.

I'd recommend in the #sleep and Mutex we do a check if EM is running and then delegate appropriately. What do you think?

EDIT : I saw the last commit just now, disregard this comment -.-

[JoshMcKin]

@jkchow thanks, and if the last commit works for you I'll release a new version

[johnkchow]

Ran a simple integration test on my project, looks good. Release that ish!

[JoshMcKin]

Its out there.

[JoshMcKin]

Should be resolved with latest releases.