amqp-spec¶ ↑

by: Arvicco url: github.com/arvicco/amqp-spec

Simple API for writing asynchronous EventMachine/AMQP specs. Supports RSpec and RSpec2.

EventMachine-based code, including synchronous AMQP library is notoriously difficult to test. To the point that many people recommend using either Mocks or synchronous libraries instead of EM-based libraries in unit tests. This is not always an option, however - sometimes your code just has to run inside the event loop, and you want to test a real thing, not just mocks.

EM-Spec gem made it easier to write evented specs, but it has several drawbacks. First, it is not easy to manage both EM.run and AMQP.start loops at the same time. Second, AMQP is not properly stopped and deactivated upon exceptions and timeouts, resulting in state leak between examples and multiple mystereous failures.

AMQP-Spec is based on EM-Spec code but makes it easier to test AMQP event loops specifically. API is very similar to EM-Spec’s, only a bit extended. The final goal is to make writing AMQP specs reasonably pleasant experience and dispel the notion that evented AMQP-based libs are impossible to unit-test.

Mind you, you still have to properly manage your AMQP broker in order to prevent broker state from leaking between examples. You can try to combine AMQP-Spec and Moqueue if you want to abstract away actual broker interactions, but still specify some event-based expectations.

There are several ways to use amqp-spec. To use it as a helper, include AMQP::SpecHelper in your describe block. You then use either #amqp or #em methods to wrap your evented test code. Inside the amqp/em block, you must call #done after your expectations. Everything works normally otherwise. You can set default_timeout and default_options to avoid manually setting AMQP options for each example. However, if you DO supply options to #amqp method inside the example, they override the defaults.

Default options and default timeout are local for each example group and inherited by its nested groups, unconnected example groups DO NOT share defaults. Please note that this is different from EM-Spec where default_timeout is effectively a global setting.

In order to setup/teardown EM state before/after your examples, you’ll need to use em_before and em_after hooks. These hooks are similar to standard RSpec’s before/after hooks but run inside the EM event loop before/after your example block. If you are using #amqp method, em_before hook will run just BEFORE AMQP connection is attempted, and em_after is run after AMQP is stopped.

Sometimes, may want to setup/teardown state inside AMQP connection (inside block given to AMQP.start): for example, to make sure that the connection is established before your example runs, or pre-declare some queues and exchanges common for all examples. In this case, please use amqp_before and amqp_after hooks. These hooks run inside the AMQP.start block just before/after your example block.

require "amqp-spec/rspec"

describe AMQP do
  include AMQP::SpecHelper

  default_options = {:host => 'my.amqp.broker.org', :port => '21118'}
  # Can be used to set default options for your amqp{} event loops

  default_timeout = 1
  # Can be used to set default :spec_timeout for your evented specs

  it "works normally when not using #amqp or #em" do
    1.should == 1
  end

  it "makes testing evented code easy with #em" do
    em do
      start = Time.now

      EM.add_timer(0.5){
        (Time.now-start).should be_close( 0.5, 0.1 )
        done
      }
    end
  end

  it "runs AMQP.start loop with options given to #amqp" do
    amqp(:host => 'my.amqp.broker.org', :port => '21118')do
      AMQP.conn.should be_connected
      done
    end
  end

  it "optionally raises timeout exception if your loop hangs for some reason" do
    proc {
      amqp(:spec_timeout => 0.5){}
    }.should raise_error SpecTimeoutExceededError
  end

end

Another option is to include AMQP::Spec in your describe block. This will patch Rspec so that all of your examples run inside an amqp block automatically. A word of caution about before/after hooks in your example groups including AMQP::Spec. Each of these hooks will run in its separate EM loop that you’ll need to shut down either manually (#done) or via timeout. Essentially, this means that any EM-related state that you’d like to set up or tear down using these hooks will be lost as example itself will run in a different EM loop.

In short, you should avoid using before/after if you include AMQP::Spec - instead, use em_before/em_after or amqp_before/amqp_after hooks that run inside the EM event loop.

One more note: you don’t need to call #done inside evented hooks, otherwise it’ll shut down the EM reactor.

describe AMQP do
  include AMQP::Spec

  default_options = {:host => 'my.amqp.broker.org', :port => '21118'}
  default_timeout = 1

  em_before { @start = Time.now }

  it "requires a call to #done in every example" do
    1.should == 1
    done
  end

  it "runs test code in an amqp block automatically" do

    EM.add_timer(0.5){
      (Time.now-@start).should be_close( 0.5, 0.1 )
      done
    }
  end

  it "runs AMQP.start loop with default_options" do
    AMQP.conn.should be_connected
    done
  end

  it "raises timeout exception ONLY if default_timeout was set" do
    proc{}.should raise_error SpecTimeoutExceededError
  end
end

Finally, you can include AMQP::EMSpec in your describe block. This will run all the group examples inside em block instead of amqp. Non-evented before/after hooks should be finished with #done, same as when including AMQP::Spec, and same caution about using them applies.

describe AMQP do
  include AMQP::EMSpec

  it "requires a call to #done in every example" do
    1.should == 1
    done
  end

  it "runs test code in an em block, instead of amqp block" do
    start = Time.now

    AMQP.conn.should be_nil

    EM.add_timer(0.5){
      (Time.now-start).should be_close( 0.5, 0.1 )
      done
    }
  end
end

…

…

For a developer new to evented specs, it is not easy to internalize that the blocks given to asynchronous methods are turned into real callbacks, intended to fire some time later. It is not easy to keep track of the actual execution path of your code, when your blocks are supposed to fire and in what sequence.

Take the following spec as an example:

it 'receives published message' do
  amqp do
    q = MQ.queue('queue_name')
    q.subscribe do |hdr, msg|
      msg.should_not == 'data'
    end
    MQ.queue('queue_name').publish 'data'
    q.unsubscribe; q.delete
    done
  end
end

Seems like a straightforward spec: you subscribe to a message queue, you set expectations inside your subscribe block, then you publish into this queue, then you call done. What may be wrong with it? Well, if you happen to use this spec against live AMQP broker, everything may be wrong. First, communication delays. There is no guarantee that by the time you publish your message, the queue have been either created or subscribed to. There is also no guarantee that your subscriber received the message by the time you are unsubscribing and deleting your queue. Second, sequence of your blocks. Remember they are delayed callbacks! Don’t just assume your previous block is already executed when you start your new asynchronous action. In this spec, when done is called, it stops everything before your subscribe callback even has a chance to fire. As a result, you’ll get a PASSING spec even though your expectation was never executed!

How to improve this spec? Allow some time for async actions to finish: either use EM timers or pass :nowait=>false to your asynch calls to force them into synchronicity. Keep in mind the sequence in which your callbacks are expected to fire - so place your done call at the end of subscribe block in this example. If you want to be paranoid, you can set flags inside your callbacks and then check that they actually fired at all AFTER your amqp/em block. Something like this will do the trick:

it 'receives published message' do
  amqp do
    q = MQ.queue('queue_name')
    q.subscribe do |hdr, msg|
      @subscribe_fired == true
      msg.should == 'data'
      done {q.unsubscribe; q.delete}
    end
    EM.add_timer(0.2) do
      MQ.queue('queue_name').publish 'data'
    end
  end
  @subscribe_fired.should be_true
end

AMQP-Spec can be currently used with Rspec only. I suppose, it is not that difficult to extend EM-Spec’s Test::Unit and Bacon support, I just do not have experience doing it.

Another limitation, library uses 1.9 syntax and therefore not compatible with Ruby 1.8. Again, it seems possible to rewrite it in 1.8-compatible style, with string evals and all such, but I’d rather leave this work to someone else.

Any help improving this library is greatly appreciated…

Copyright © 2010 Arvicco. Original EM-Spec code copyright © 2008 Aman Gupta (tmm1)

See LICENSE for details.