This repository is a working example of how a project can start with a Lambda function written using Python and added inline into a CloudFormation template for bootstrapping, then converted into a Go application for long-term maintenance.
Well, young pup, this is the magic of The Cloud.
One of the best things about CloudFormation is that you can codify your infrastructure, and a Lambda function can be described as pure infrastructure. However, you are limited in the languages you can write for your Lambda function and how you deploy it to AWS. CloudFormation allows for a Lambda function to be specified inline into the CloudFormation template, but this adds a maintenance issue because the code can no longer be tested outside of CloudFormation. This ruins the SDLC workflow. While Lambda's support of Go is great, you cannot specify a Go app inline into a CloudFormation template because it is a compile language and the artifact must be contained within a zip file.
Sometimes people want to control where their artifact is loaded from, sometimes not. When you specify a Lambda function inline into a CloudFormation template, the code artifact ends up being stored in a bucket that AWS controls. This transfers a lot of security risk to AWS and out of your hands because you don't have to manage an S3 bucket for your artifacts. Best of all, it's free. So let's take advantage of this benefit and deploy our Go apps there.
Ah, but wait. Now there's a two-step process. First we have to define the Lambda function and then we have to upload a code artifact to AWS's S3 bucket. This is what this project solves in three phases: initialize, prepare, deploy, and test. This provides a complete SDLC workflow for developers.
-
Initialize
The Lambda function is declared as a Python runtime into the CloudFormation template, but this is just a stub for your actual application.
This is done via
make init. -
Prepare
Then you modify the Lambda configuration out-of-band and update it to a Go runtime. Your Lambda function is now in a half-baked state, where it expects a Go app, but it deployed with Python code.
This is done via
make prepare. -
Deploy
Now deploy the Go app artifact to the Lambda function. Your Lambda is now ready to serve requests. From now on, the iteration of the code happens in this phase. There is no relaunching or updating of the infrastructure unless the architecture changes (e.g. new IAM role permissions).
This is done via
make deploy. -
Test
Testing is easy, but it happens as both a local (fast) test and as an end-to-end (slower) test. The Lambda function can be tested as well as the Go application itself.
This is done via
make test-stackandmake test.
There are 3 motivations here: cost, deployment, and development.
-
Cost/Speed
When using AWS Lambda, the speed of your application and its memory usage will directly relate to the cost of the service. Even this simple Python application takes about 200ms to run, and AWS bills in 100ms increments. In contrast, this Go application takes 1ms to run. We have another 99ms of headroom being billed that could be used to add business value to our customers. Now that's value.
-
Deployment
Go application are bundled as a single static binary, with all of its dependencies (i.e. libraries) included, or vendored in. This makes deployment as simple as copying the artifact to the server and testing it. Also, testing the artifact is simple because Go binaries can be compiled for all operating systems (e.g. a Linux binary can be created on a MacOS machine).
-
Development
Go is a simple language, it makes readable code, and is fun to write. Opinions vary, but this is mine. I also like ruby, vim and other contentious things.
The biggest benefit of this architecture is that the parts that change the most are contained to within the function code itself, and the infrastructure stays relatively static for its lifetime. By separating these pieces, the project will be more stable and permission structure (separation of duties) can be implemented if that is an organizational goal.
tl;dr Repeat phases Initialize and Prepare.
If the Lambda function is destroyed at any time, it can be rebuilt quickly. The initial infrastructure is going to bootstrap itself as the stubbed Python application, but you can then redeploy the Go application quickly. Since this is all held in the same Git repository, it will be tagged and versioned appropriately.