Requiring workflow class in workflow starter code

Question

I have a simple question regarding the architecture of my Amazon Simple Workflow / AWS Flow for Ruby app. For background, I have a simple workflow with one activity running in an AWS Flow for Ruby layer on Opsworks. I have a separate REST API running in a Rails App Server layer on Opsworks that I would like to kick off the workflow.

The code in the REST API that kicks off the workflow:

1: domain = AWS::SimpleWorkflow.new.domains['my_domain']
2: workflow_client = AWS::Flow::workflow_client(domain.client, domain) {{from_class: MyWorkflowClass}}
3: workflow_client.start_execution(input_1: @input1, input_2: @input2)

My assumption is that my workflow and REST API code bases could be separate and that the only common component would be the aws-flow Ruby gem and require 'aws/decider'. However, I'm finding that my REST API also needs to have require 'PATH_TO_MY_WORKFLOW_CLASS'. When I remove that line of code from the code file in my REST API that kicks off the workflow, I get the following error:

undefined method `_options' for nil:NilClass; ["/Users/MyName/.rvm/gems/ruby-2.0.0-p247/gems/aws-flow-2.2.1/lib/aws/decider/utilities.rb:183:in `interpret_block_for_options'", "/Users/MyName/.rvm/gems/ruby-2.0.0-p247/gems/aws-flow-2.2.1/lib/aws/decider/implementation.rb:73:in `workflow_client'"
(error at line 2 above)

Am I mistaken? Do I really need to require MyWorkflowClass in my workflow starter app (i.e. my REST API) or am I doing something wrong? I've scoured the documentation and could not find a clear answer to this. All the samples that I can find do indeed have the workflow class included in the workflow starter code, but I'm not sure if it's because they are bundled as a simple sample or if it's because it's the way it's supposed to be. The reason why I am not taking the samples at face value is because requiring the workflow class in the workflow starter code does not make any sense to me. It binds the two apps way too tightly for my taste.

Answer 1

I posted an issue on the aws-flow-ruby sdk and got the answer from an Amazon Engineer. In short, you can use the :from_class option or the :prefix_name and :execution_method options together.

---

There are two ways of starting the workflow in code

1) Using the aws sdk directly. In this case, your code doesn't need to know anything about the workflow class. You just need the domain, workflow type (name and version) and the workflow id. It will look something like -

require 'aws-sdk-v1'
swf = AWS::SimpleWorkflow.new.client
swf.start_workflow_execution(
  domain: "HelloWorld",
  workflow_type: {
    name: "HelloWorldWorkflow",
    version: "1.0"
  },
  workflow_id: "foo",
  input: ....,
  ....other options (optional)...
)

As you can see above, this doesn't require the workflow class at all.

2) Using the aws-flow gem (which is what you are doing above). There are two ways of using the workflow client provided by the aws-flow gem to start an execution. You can either use the client as a generic client and not tie it to any workflow class or you can use the :from_class option to fetch options from a particular workflow class. To use the from_class option, you need to have the class in the ObjectSpace (hence you need to require the workflow file).

With from_class -

require 'aws/decider'
domain = AWS::SimpleWorkflow.new.domains['my_domain']
workflow_client = AWS::Flow::workflow_client(domain.client, domain) {{from_class: "MyWorkflowClass"}}
workflow_client.start_execution(input_1: @input1, input_2: @input2)

Without from_class -

require 'aws/decider'
domain = AWS::SimpleWorkflow.new.domains['my_domain']
workflow_client = AWS::Flow::workflow_client(domain.client, domain) {{
  prefix_name: "YourClassName",
  execution_method: "workflow_method_name",
  version: "1.0",
  ...other options...
}}
workflow_client.start_execution(input_1: @input1, input_2: @input2)

The recommended way to start a workflow execution is to use aws-flow WorkflowClient instead of using the SDK directly.

---

Additional notes with respect to the input accepted by a workflow:

---

The SDK and the console will only take strings as input. This can be a free form string but if your workflow is written using ruby flow, this string should be a serialized form of your input so that the WorkflowWorker can deserialize the input when it picks up the task and convert it into ruby objects (in this case a hash).

When you use the ruby flow WorkflowClient, the client will automatically serialize your input hash (or any other input) into a string before sending it to SWF. aws-flow by default uses a YAML based data converter to do this (It can be overridden).

If you just want to see what your input hash will look like as a string, you can do the following -

AWS::Flow::FlowConstants.default_data_converter.dump(input_hash) 

You can then use this serialized input to start a workflow using the SDK or the console.