I have kind of a love/hate relationship with Grafana,Specifically, the Grafana dashboard product from the company also known as Grafana. In general, when I say “Grafana”, I mean the dashboard service, not the company as a whole.
the open-source monitoring dashboard service. On the plus side, it works better than any of its alternatives, it’s mostly freeIt’s licensed under the AGPL and some of their extensions are paid-only, but you can run 98% of the product for free on your own system, and they’re generally pretty responsive to community issues.
, and it looks reasonably nice. The minus side is that sometimes it’s really tricky to get it to do something that should be basic, and there are a number of really annoying persistent missing features.Let’s just say that Grafana’s idea of “drill down” and mine are rather far apart and leave it there for now.
I have fairly strong opinions about dashboard design, which I’ll probably write up here in detail at some point, but the short version looks something like this:
Dashboards are a form of communication. Like any other form of communication, they should focus on the most critical things first, and find the most efficient way to communicate those things to people. Consistency in design, formatting, and terminology is important to help people understand new dashboards at a glance and reduce misunderstandings. Generally, effective dashboards container mostly (or exclusively) timeseries graphs and avoid bar charts, gauges, big colored blobs that say “OK” or “Bad”, and tables. But different problems call for different tools, and sometimes each of these have their place. So, while I don’t really like putting tables on most dashboards, sometimes they’re the most effective way to communicate data. I’m mostly interested in data from Prometheus, and generating tables in Grafana from Prometheus data is one of the things that Grafana makes easy, right up to the point where it suddenly turns nearly impossible.

scottstuff.net

A practical guide to building Grafana tables that combine multiple Prometheus metrics into a single view. Covers the step-by-step process: setting queries to Format:Table and Type:Instant, using the 'Merge series/tables' transform to join multiple queries, applying 'Organize fields by name' to rename and reorder columns, and using field overrides for per-column unit formatting. Uses Kubernetes CPU and memory metrics as concrete examples, and warns against using Join transforms which don't behave correctly for this use case.

Creating Tables in Grafana From Multiple Prometheus Metrics

Adding transforms to improve presentation