This is an alpha project and this is not the project's final repository location.
This project combines the simplicity of running [Stackdriver Monitoring] with the simplicity of exposing [Prometheus] metrics.
In short, in some cases Prometheus is not worth the operational overhead and something as simple as Stackdriver is sufficient for alerting, graphing and so on.
However there is currently no easy way to get custom metrics into Stackdriver Monitoring and some sort of push-based system would have to run in every monitored service.
This project attempts to solve that by scraping metrics from services via the standard Prometheus scraping protocol and then piping them into Google Stackdriver.
It is solely intended to be run in Kubernetes clusters and relies heavily on the Kubernetes API.
Right now this project is only concerned with monitoring services that run inside of Prometheus.
This is done by following three annotations on Service objects:
prometheus.io/scrape: Can be set to"true"(yes, as a string!) to enable metrics scraping for this service.prometheus.io/port: Override the default port (80) used for scraping.prometheus.io/path: Override the default path (/metrics) used for scraping.
Stackscraper will resolve the endpoints of the Service and scrape all pods
individually. Every pod will become a distinct "monitored resource" in Stackdriver.
There are a lot of caveats right now, this is a young alpha project after all!
- Only
gaugeanduntypedPrometheus metrics are supported right now. Not all Prometheus metric types have an equivalent Stackdriver type, so this is not guaranteed to change much. - Metric labels are static after creation
- This is not actually production-ready code (it does a lot more API calls than it should for basic operation right now, don't use it).
A short note on scraping: I didn't find an existing scraper/parser in the Java world so I wrote one.
If there is an existing project for this feel free to ping me about it!