Declaring functions outside of functions

[mnapoli]

I'll be using this discussion to track what we've learned about declaring Lambda functions outside of the functions section in serverless.yml.

For example with a "queues" component, the goal is to define the "worker" function inside the queue:

service: app

provider:
  name: aws

queues:
  emails:
    worker:
      # This is a Lambda function
      handler: worker.handler
      # use any function property

The idea is to "copy" the content of queues.emails.worker into functions. Then Serverless should process that function as usual.

So far I've identified the following flow when Serverless runs:

• scripts/serverless.js
• Serverless.run() (lib/Serverless.js)
• Service.setFunctionNames() (lib/classes/Service.js)
• initialize hook
• other hooks

Service.setFunctionNames() normalizes the functions. I've tried inserting functions after that (in the initialize hook) but my inserted functions were causing errors later (because they were not normalized).

That means that:

1. Either we add functions after, and we replicate the behavior of setFunctionNames()
   This is risky: we run the risk of missing some behavior, not normalizing correctly, etc.
2. Or we add functions to functions before that

Ideally we would aim for solution 2. That means we can't use the initialize hook.

With that in mind, I've tested inserting the functions when the component boots. Simplified example:

export class MyComponent {
    constructor(serverless: Serverless) {
        // ...
        this.appendFunctions();
    }

    appendFunctions(): void {
        const configuration = this.getConfiguration() ?? {};
        Object.entries(configuration).map(([name, queueConfiguration]) => {
            Object.assign(this.serverless.service.functions, {
                [`${name}Worker`]: queueConfiguration.worker,
            });
        });
    }

  // ...
}

And in unit tests this is working!

We'll probably need more testing on this, see if this really works, especially with events, variables, config validation, or any other processing Serverless might do.

To those reading: if you have other ideas, please share :)

[m-radzikowski]

Wouldn't it be simpler and safer to just reference the function from the component by name? Like this:

service: app

provider:
  name: aws

functions:
  worker:
    handler: worker.handler

queues:
  emails:
    function: worker

[mnapoli]

Yes, that's the alternative in case we can't make this work properly.

It's just a bit less simple, and one part is confusing: would you need to define the SQS event on the function or not? (the answer is not obvious, that's what I mean)

[jessekernaghan]

I think this approach has the least cognitive overhead when combining lift with other plugins.

My case is trying to get serverless-lift, serverless-sqs-local, and serverless-newrelic-lambda-layers to play nicely together.

I can see the point regarding confusion around configuration, for sure. I think the fact that we currently have the ARN available as a variable for publishers hints to devs that they could use it for consumers/workers as well. I don't think it would be unreasonable to force the event configuration if that simplifies things.

[nolde]

Is this currently possible? This is the biggest barrier for me to actually adopt lift... I have so much already done, and other plugins doing different things, that moving my functions out of "functions" would be a huge chore.

[mnapoli]

@nolde no this isn't possible, we went with what is currently documented