A place to share technical learnings, etc.

View My GitHub Profile

aws | jekyll | github | apigateway | serverless | ad | powershell | windows | webdev | nodejs | terraform | consul | nomad | jenkins | traefik | azuread | azure |

29 December 2018

API Gateway with Multiple Deployments

Our team has decided to embark on a conversion process, where we’re planning to migrate a NodeJS + MongoDB API running in EC2 and surfaced via API Gateway to a new Serverless NodeJS using Lambda, DynamoDB, and API Gateway.

As a part of the initial feasibility work, I was able to convert our NodeJS API over to DynamoDB with a minimum of hassle (probably good for another blog post), and we were ready for some testing.

Trying to Grok API Gateway Stages and the Limitations

Once I had the NodeJS API code converted over to DynamoDB (in a development branch in Github), I was ready to publish the API for some internal testing. I thought I understand API Gateway stages relatively well, until I went to actually do this. It appears that API Gateway stages are basically snapshots of how the API looked at a particular time, but there appears to be no easy way to have two stages do vastly different things while retaining the ability to make edits and changes to both stages in the future if needed - for example have one stage that points to an EC2 instance via VPC link, and another stage that points to a lambda function. Note that while you can actually have API Gateway do this, but it looks like that actually means changing the main API definition to accomodate the new functionaliy - then making a snapshot of the API to represent that new functionality, with no easy way to go back and make modifications to the “old API” if needed (you’d need to change the main API definition back to the “old API” way of doing things, then make another deployment). We use Terraform to build out all of our infrastructure in AWS, so while this isn’t insurmountable, it seems like way too much work just to get to the idea of “stages”.

A Path Forward

We use the Custom Domain Names function of API Gateway to map a custom domain onto our API, along with the Base Path Mapping function (currently we’re using the root of the mapping “/”). As a result of the limitations described above and the need to maintain two separate API deployments for the duration of the migration, we’re planning to switch to using a stage-like syntax going forward for the Base Path Mapping - effectively mapping a v1 base path mapping to our old EC2/MongoDB/VPC Link API, and a v2 mapping to our new Serverless/DynamoDB API.


tags: aws - apigateway