Add AWS Lambda Layer Support

[david-mkl]

This provides the option to place the requirements into a lambda layer.

2 additional options have been added:

• inject: boolean. Controls whether the requirements should be injected into the build. Defaults to true
• layer: boolean | object: Indicates that the requirements should not be injected into the build (will set inject = false when this option is present), and instead build separately to be used a lambda layer. Defaults false

When the layer option is provided as a boolean true, then some defaults will be chosen for the layer, and the layer will be added to the global layers object in serverless.yml.

When layer is provided as an object, then the values will be merged onto the defaults.

The layer will be created in the Cloudformation stack as PythonRequirementsLambdaLayer. In order to use the layer in a function, add the reference to the layer in the function's layers section. e.g.

functions:
  hello:
    handler: handler.hello
    layers:
      - {Ref: PythonRequirementsLambdaLayer}

Future Improvements/ Limitations of this implementation:

• This will build a "global" lambda layer, i.e. there is no support for layers-per-lambda.
• There is no way to control which requirements are placed in the layer. This will place all requirements into a single layer.

[rbu]

@david-mk-lawrence Thanks for the PR, I was looking exactly for this feature myself.

What's the reason for creating two options though (inject and layers)? It looks to me there are only two setting that make sense:

• layers:false, inject:true (the default)
• layers:true, inject:false (use layers)

The two other "possible" settings (possible in terms of the serverless.yaml):

• layers:true, inject:true -- inject will be overwritten to false by code
• layers:false, inject:false -- apparently disables the plugin

I'd suggest you reduce it down to one option, "layers", which is false by default and when true(ish), it disables injection and enables layers.

[dschep]

I love the simplicity of this implementation, great job!!

The main changes I want made before I merge this:

• remove the inject option, base all behavior on the layer option
• add yourself to the contributors list!
• respond to the 2 code comments (both are 100% open for discussion)

[david-mkl]

@rbu Thanks for the feedback! Yes, that is a good point and I've removed that. Now the layer option will solely control the behavior.

The reason I put it there initially is because I found it very useful to prevent the injection from occurring while developing. But now that the layer is working, it shouldn't be needed.

[squaresurf]

🎉

@dschep any idea when we can expect you to cut a new release with this and #308?

[dschep]

probably on monday 🙂

[rbu]

@dschep That would make a great start of the week for me. Appreciated. Thanks for merging this so fast 🍻

[dschep]

🚀 Shipped v4.3.0!!!!

[SimonScholl]

I really like this feature that i can create layers for my lambda function that easily. The only thing i am missing is that i can just include one layers. Would it be possible to create more then one layer so i can distribute my dependencies over several functions in a modular way. The plugin already allows to define function specific requirements for multi function projects. :)

[felschr]

Any plans on adding support for defining multiple layers and including only certain packages in a layer?
With all non-specified packages still being bundled with the lambda functions directly.