5 Hard Lessons about Serverless
Get the latest in your inbox
Stardog’s web team has been experimenting with the Serverless Framework to prototype a web app with a pure AWS infrastructure. Serverless is to apps what baskets are to eggs, basically.
Serverless aims to configure and integrate AWS Lambda with other Amazon services like API Gateway (to turn Lambda functions into HTTP endpoints), IAM (to manage permission policies in the AWS ecosystem), Cognito (for user authentication), etc., all from a configuration file and a command line.
What we’ve learned so far
Though we may move to a different architecture for production, we’ve learned a bit about working with the framework effectively.
1. Test API Gateway locally
Serverless allows you to invoke AWS Lambda functions locally but has no such support for API Gateway. As a result you can’t easily test your lambdas with the inputs they’ll be receiving from API Gateway in production. API Gateway doesn’t always process HTTP events in obvious ways either. Spending a few minutes to deploy your lambdas each time you want an accurate event body will have you barking at the moon. Catching the Serverless offline plugin late in the game makes us wish we’d known about it much sooner.
2. Structure lambdas for unit tests
To test Lambda functions, they need to be exported for use by a testing framework differently than they’re exported for use by AWS Lambda. You can solve this by exporting your business logic in a straight-forward way for your test framework, and then write a wrapper to export that export for AWS Lambda. The Serverless docs do a great job exhibiting this.
3. Authenticate on the client
Serverless allows you to authorize against a Cognito user pool. You can prevent headache and avoid unnecessary Lambda costs by using the Cognito client-side SDK to authenticate users with Cognito on the client, and then let Serverless use that user’s Cognito ‘identity’ to determine permissions on your HTTP endpoints.
4. Check for new resources regularly
Community support for the Serverless Framework is strong and lively. A search that turned up nothing useful yesterday might provide exactly what you’re looking for today. Here’s an awesome, thorough walkthrough for building a Serverless React app which was published a few weeks into our project, and discovered weeks later. If you aren’t finding what you need, look again next week.
5. IAM role policies are strictly typed
Serverless allows you to configure an IAM role that your Lambdas will assume when interacting with other AWS services. Unless we explicitly granted permission to perform specific actions on AWS resources (like user pools or buckets), service calls from our Lambdas were typically denied- even when we configured very flexible policies to the resources themselves. Though never fervorless, our navigation of permissions issues was much easier after discovering this list of IAM policy snippets.
Conclusion
On the whole, the Serverless Framework is a growing, well-supported framework with good documentation. We foresee using it for odd jobs around the house when AWS tools make sense. For application development, however, we’ll continue to use app-specific platforms awhile longer.
