If Graphana can create them reliably, last-[n]-weeks histograms would be nice for time, MB, revisions. (if people want something really fancy: Kernel density overlayed)

These specific graphs might not be faulty. Part of this task should be to build trustworthy metrics to compare against the graphs.

In T225610#5253205, @Jan_Dittrich wrote:

If Graphana can create them reliably, last-[n]-weeks histograms would be nice for time, MB, revisions. (if people want something really fancy: Kernel density overlayed)

I added a couple of heatmaps for us to iterate on, let me know whether you prefer these to the scatter & daily average graph, for example.

@Lea_WMDE btw, this is review for you, maybe I should put in the "Demo" column?

@awight, oh I see, yes then demo column is great :)

import time and MBs moved look awesome!
What I am still looking for is no of file revisions and no of text revisions :)

In T225610#5282521, @Lea_WMDE wrote:

What I am still looking for is no of file revisions and no of text revisions :)

Those are in the purple boxes at the bottom. I rearranged the color scheme there so that they get darker as we go up the percentiles, let me know if it doesn't look right to you.

(not moving task)

can we trust those numbers, though? Aren't percentiles for events that don't trigger at least every minute by default wrong on grafana? That's why I'm super happy with the heat map solution you used for the other cases.

In T225610#5282572, @Lea_WMDE wrote:

can we trust those numbers, though? Aren't percentiles for events that don't trigger at least every minute by default wrong on grafana? That's why I'm super happy with the heat map solution you used for the other cases.

Thanks, I think you're totally right about that. If more than one event per <time unit> gets aggregated, the percentiles will be fake. Happy to replace with a heat map!

Replaced with heat maps.