Using existing histogram buckets in heatmap

:wave: Newb trying to get to grips with heatmaps. I’m using Grafana 9.1, as this is what’s baked into rancher’s monitoring solution right now.

What I want is a basic heatmap of counts in the request durations buckets over time - x requests in this interval took <200ms, y requests took 1s>x>=200ms, z requests took >= 1s, etc.

Query I’m using is just

sum(increase(request_durations_bucket[$__interval])) by (le)

I’m using a heatmap format in the query options, and the underlying metric is a histogram type.

My actual buckets - the ‘le’ label on my metric - are [5, 10, 25, 50, 75, 100, 250, 500, 1000, 2500, 5000, 10000, +Inf], in ms.

I don’t really understand what “calculate from data” is doing. With it off, grafana does show my actual bucket values, but they’re being sorted alphabetically, which is nonsense. With it on, buckets are ordered, but they’re being shown as ridiculous values that aren’t the buckets I’m using - not sure where those are coming from.

I’ve been fiddling with the interface for hours and I haven’t yet stumbled on a way to use my actual metric buckets, ordered numerically. This feels like the most obvious requirement for a heatmap, so I’m missing something substantial here. Anyone able to clue me in?

Think I figured this out. I had to explicitly set {{le}} as a custom legend label, instead of auto.

I would never have realised that adjusting the legend label would actually alter the way grafana works out the bucket de-aggregation. That this isn’t the automatic behaviour still makes me feel like I’m doing something wrong.