It works, but I've found Promxy which handles more edge cases than this, so I've moved to that. You should, too!
A daemon that receives Prometheus API calls from a system such as Grafana, sends that request to multiple Prometheus instances, and aggregates the responses.
Systems like Grafana don't work well when your data is split up across multiple non overlapping Prometheus servers. This daemon attempts to trick the requester into thinking that there is just one server, but the requests to go many.
Before starting, you should have installed go and ensure that your $GOPATH environment variable is also set.
$ go get https://github.com/swalberg/promproxy
$ cd $GOPATH/src/github.com/swalberg/promproxy
$ make
Upon success, you will have the binary named app in your current directory
At this time, there are limited configuration options, so running the app is simple.
Launch it with an array of Prometheus servers as arguments
$ ./app http://192.168.0.20:9090 http://192.168.0.30:9090 http://192.168.0.40:9090
Once running, the app will be listening at http://0.0.0.0:6379
Perform a test query against the label endpoint
$ curl http://localhost:6789/api/v1/label/job/values
{"status":"success","data":["prometheus"]}
After promproxy is running, setup (or change) your Prometheus data source in Grafana to point to promproxy instead of a prometheus server.
Currently, only the following endpoints of the Prometheus API are supported:
- label
- series
- query_range
These endpoints should provide sufficient functionality for building dashboards in Grafana