Running production HTTP servers in Kubernetes is complicated and involves many pieces of infrastructure. The HTTP Add-on (called the "add-on" hereafter) aims to autoscale these HTTP servers, but does not aim to extend beyond that scope. Generally, this project only aims to do two things:
- Autoscale arbitrary HTTP servers based on the volume of traffic incoming to it, including to zero.
- Route HTTP traffic from a given source to an arbitrary HTTP server, as far as we need to efficiently accomplish (1).
The add-on only provides this functionality to workloads that opt in to it. We provide more detail below.
To autoscale HTTP servers, the HTTP Add-on needs access to metrics that it can report to KEDA, so that KEDA itself can scale the target HTTP server. The mechanism by which the add-on does this is to use an interceptor and external scaler. An operator watches for a HTTPScaledObject resource and creates these components as necessary.
The HTTP Add-on only includes the necessary infrastructure to respond to new, modified, or deleted HTTPScaledObjects, and when one is created, the add-on only creates the infrastructure needed specifically to accomplish autoscaling.
As stated above, the current architecture requires an "interceptor", which needs to proxy incoming HTTP requests in order to provide autoscaling metrics. That means the scope of the HTTP Add-on currently needs to include the app's network traffic routing system.
To learn more, we recommend reading about our design or go through our FAQ.