liranbg / sidecar-proxy

A reverse sidecar proxy with prometheus metrics capabilities

Sidecar proxy

A containerized proxy meant to run alongside a service (in the same pod) in order to serve Prometheus metrics, and facilitate scale-to-zero flows.

Supported metrics:

1. General:
   • num_of_requests - prometheus CounterVec that simply counts requests using a reverse proxy (Go's built in ReverseProxy)
     
2. Service specific:
   • Jupyter:
     • jupyter_kernel_busyness - prometheus GaugeVec that is set to 1 if Jupyter has one or more busy kernels, and to 0 otherwise. Periodically queries Jupyter's /api/kernels endpoint

The container includes a server that serves Prometheus metrics through the /metrics endpoint.

All metrics contain these labels: namespace, service_name, instance_name.

The code was built so it will be easy to extend it and add new metrics. This is performed by creating a new metric handlers that implement the MetricsHandler interface.

When starting the container the --metric-name flag (can be defined multiple times) is used to set which metrics handlers to run (num_of_requests is mandatory).

An example helm chart that adds this container alongside a Jupyter service can be found here