An caching proxy for the Prometheus query endpoint for speeding up Grafana dashboards.
This repository contains a small HTTP proxy server to run alongside your
Prometheus servers. By default it will listen on port 9091 and proxy all requests
to localhost:9090. Any request staring with /api/v1/query will receive
special threatment and get possibly cached. All other requests are passed
unaltered.
All requests to /api/v1/query will be forcefully cached. The Cache-Control
headers are removed, the start and end parameters will be rounded to the
next full minute and then passed to the upstream server.
This proxy is tailored for one specific use-case: Multiple users and/or dashboards accessing the same Prometheus timeseries through Grafana. This can considerably reduce the load on your Prometheus server while still providing good enough data for most dashboards.
Just add this proxy as an additional data source in your templated Grafana dashboards and you can easily switch between direct and cached access to compare speed and quality.
Clone, build, start and point Grafana to localhost:9091 instead of localhost:9090
(if Prometheus is running on localhost).
go build
./promcache
| Flag | Type | Description |
|---|---|---|
listen |
string | Listen address. Default: :9091 |
dir |
string | Cache dir. Implies -disk |
disk |
bool | Use disk cache. Default: false |
verbose |
bool | Verbose output |
dumphttp |
bool | Dump HTTP Request and Response headers |
upstream |
string | Upstream host to connect to |
duration |
int | Forced cache duration. Default: 60 |
MIT