Horizontal Pod Autoscaler Reloaded: Scale on Custom Metrics

These are the demo files for the KubeCon 2018 presentation by @directxman12 (Solly Ross) and @MaciekPytel (Maciek Pytel).

Layout

The loadbalancer directory contains files for building the fake load balancer application used to demonstrate Object metrics.
The adapter-config directory contains the files used to run the Prometheus adapter.
The prometheus-config directory contains the files used to deploy Prometheus.
The cpu-consumer-config directory contains the files used to deploy the sample cpu-consuming application that we autoscale.

Ensure your cluster is fully set up. Make sure you have metrics-server running (or some other provider of metrics.k8s.io, if you're from the future). Make sure that your API aggregation certificates, etc are set up. If you use an installer, this is probably done for you.
Make sure Prometheus is installed and set up to collect metrics from pods. You can use the Prometheus Operator if you're not familiar with setting up Prometheus from scratch.
Build and install the right version of the Prometheus adapter. You'll need a version with the "advanced configuration" feature. It currently lives at feature/advanced-config, but it'll probably be in master at some point in the future. There's an image version of it available at docker.io/directxman12/k8s-prometheus-adapter:advanced-config.

An appropriate configuration for use in this demo (in ConfigMap form) it can be found in the file adapter-config/adapter-config.yaml. Use it and the Deployment and support files in the same directory to deploy the adapter.
Build and launch the fake loadbalancer application, and make sure Prometheus collects its metrics (which can be found at /_metrics). They end in _latency_seconds.
Connect to the fake loadbalancer. It expects the HTTP host header to be of the form <svc>.<namespace>.balance.local. It will proxy anything on port 80 to port 80 on the service, except the path /_metrics, which contain its own metrics.