serverless-bundle VS elasticmq

As you can see in the above handler.js file, we're getting CommonJS instead of modern JavaScript or TypeScript. To get these, you need webpack or some other bundler. serverless-webpack exists if you want full control over your ecosystem, but there is also serverless-bundle that gives you a set of reasonable defaults on webpack 4 out of the box. We opted into this option to get us started quickly.

One of the most important aspects of serverless development is keeping an eye on your bundle sizes and to reduce cold start times on Lambda. Keeping this in mind, the template utilizes serverless-esbuild and serverless-analyze-bundle-plugin to provide function analysis out-of-the-box. I opted for serverless-esbuild over serverless-bundle for a few reasons:

We use the package.json at the project root to install the dependencies that will be shared across all the services. For example, if you are using serverless-bundle to optimally package the Lambda functions, or using the serverless-plugin-warmup to reduce cold starts, they should be installed at the root level. It doesn’t make sense to install them in each and every single service.

Since a majority of services rely on the same system services, we can leverage Docker to start all the needed project services prior running services with the offline command. The template ships with a base docker-compose.yml that stands up an instance of ElasticMQ to emulate Amazon SQS's API. The Docker image can be extended to include other services like DynamoDB, Redis, or PostgresQL and can be run using yarn infrastructure:build or yarn infrastructure:start depending on if it's your first run or not.

however...we have banned use of SQS. unlike S3/Minio, there's nothing API-compatible that we can run offline. (well, there's one thing, but it's in-memory / single node only and doesn't give us the durability/reliability we want with SQS). instead we use RabbitMQ because it's identical at the API layer no matter where we are. AWS has a hosted option for this but we haven't switched to it, because our existing self-managed Rabbit clusters are stable and well-understood.