Is it possible to separate my code and infrastructure in serverless framework?

Question

I'm using serverless framework (https://serverless.com) to deploy my AWS resources. I have a lambda functions defined in a serverless.yml file like this:

functions:
   hello:
   handler: handler.hello
   ...

Now I want to separate my infrastructure (serverless.yml) from my code. There would be two separated git projects: one for the infrastructure and one for the lambda (the code part).

This would be the development flow: Changes made in lambda project (git merge) would trigger a CI/CD pipeline to pull the code do the necessary checks (hint, test etc) and deploy the lambda to an s3 bucket.

After this I could include the s3 as a source of my lambda function and update the stack with the function that includes the changes. Something like this:

functions:
   hello:
   handler: function-s3-location
   ...

My questions:

1) Is it a good CI/CD approach?

2) Is it possible with serverless framework or it is possible only using AWS::Lambda::Functions code attribute to specify the s3 bucket?

Thank you!

Cleriston Cleriston · Accepted Answer · 2018-10-15T23:40:14

Regardless your decision, in the end, it will all be an architectural decision, which should be made considering processes, team experience, culture, architectural knowledge, etc.

To give you a bit of insight, you can structure your project in a way you have your functions in a folder and any other module in a diff folder. What you need to be careful is that you define the "package.include" and/or "package.exclude" in your config file.

Moreover if by structure you mean all the CloudFormation resources, you have a section for it in your serverless.yml config files. If it starts getting too big, you could as well use an extra config file and import it inside your serverless.yml

Ex:

├── config //configs based on your stage
 ├── dev.json
 ├── prod.json
├── functions
 ├── functionsA
  ├── index.js
 ├── functionsB
  ├── index.js
├── tests
 ├── unit.test.js
├── package.json
├── serverless.yml
├── serverless.resources.yml

The benefits of having everything in one repo:

You keep all resources under the same project, which means that if you "sls remove" the command will take care of deleting all of your resources. Same for lambdas.
It is easy for new team members to understand your architecture.
You have all code you need to start a local instance of your API and execute integrated tests. Useful for CI.
You will have only one repo to manage during your pipeline.
You will not have to manage the 2 repos in order to assess they are "walking" together.
Serverless allow you to push/deploy just the functions if you don't want to update your cloud architecture.
The provider will manage the bind between all your resources. Ex: You will not have to manage that the apiGateway on stage DEV, should invoke the lambdas pointing to the DV DB or containing the environmental variable with the value DEV.
The least you keep from your own manual management, the better.

I hope those reasons are enough for you to make your decision.

Is it possible to separate my code and infrastructure in serverless framework?

1 Answers