Almost, buddy. Almost.
A couple days ago, I read this incredible article about running CRISPR searches on DNA using AWS Lambda. Easily one of the coolest engineering blog posts I’ve read in a while. I started thinking about other uses for this kind of architecture, and set out to experiment with on an idea today - building an image processing service on top of Lambda.
This is an area near and dear to my heart after my time spent converting Vox Media’s publishing platform to use Thumbor - a very full featured Python based image processing service. I thought it would be pretty neat if you could run such a service without having to manage any servers - run it all of Cloudfront and Lambda. The goal would be totally URL based API to the service, easy to setup and even easier to keep running.
Looking at the AWS Lambda documentation, I learned that if you want to interact with Lambda over HTTP, you need to go through Amazon API Gateway. This was a new service to me, but the pitch made sense - configure an API with them, and they’ll take care of scaling, caching, logging, monitoring, throttling, and managing different environments like staging and production.
Naturally, there is a ton of product specific jargon to API Gateway, but the docs have a walk through to link up with Lambda that was pretty straightforward. After getting the ‘Hello World’ going, I started hacking away on a proof of concept to provide an image url, height and width, and get a resized image in response. This was my first foray into node.js, which was pretty fun. Here’s what I ended up with:
Apologies for any nightmares this prototype code causes you .
Turns out, this dream is very similar to a demo the AWS Lambda General Manager gave recently, where he built an image resizing service on top of Lambda and API Gateway. I came across this post after Googling a Lambda blueprint called ‘image-processing-service’, and it provided a great start for what I was trying to do.
It did seem a little odd to me images were being returned in base64 encoding, even for the web demo. I rolled with it since Lambda is basically returning JSON, and you can decode from base64 in the ‘Integration Response’ section of API Gateway.
Integration Response configuration. You can also use
$util.base64Decode($inputRoot.body) to decode the image.
At this point, the experiment kind of fell apart. I had the response body returning the right thing, but API Gateway really wants you to return JSON, so the default Content-Type header is application/json. Naturally, browsers aren’t going to load images like that, so I did some digging on setting response headers in Lambda. Turns out, this is impossible:
API Gateway does not currently support mapping from the integration response body to the response headers, but we are looking to add support for this in the future.
Which is a pretty big blocker for an image resizing service. Different headers are going to be necessary for JPEG vs PNG vs WEBP. API Gateway is very REST and JSON oriented, so what I’m trying to do is probably outside the design expectations. It is a new product and seems to be iterating on feedback pretty quickly, so hopefully, this will be possible in the future.
Need help building a new web product or scaling your current one? I'm available for freelance work.