Björn Rabenstein
Björn Rabenstein
To go even further down the rabbit hole: When reconciling zero buckets, we might be merging buckets into the zero bucket. Even when ignoring zero buckets, we will still merge...
This sounds wonderful. I hope it all works out as planned. I'll have a look ASAP (which might need another week or so because we are going into KubeCon week,...
Sorry for the delay. I'm back from KubeCon now and will have a look at this once I'm through my backlog.
The idea is to only switch to incremental calculation once we would everflow float64. An imprecise average is still better than the Inf or NaN we get when overflowing. Check...
Thanks for your thoughts. There aren't really any assumptions about the input values. Users can do whatever they want. However, there is no expectation that we can solve all floating...
(Note for the reader to not lose track: We are currently discussion the average calculation for simple floats, not for histograms. We just stumbled upon problems with floats while trying...
I have created https://github.com/prometheus/prometheus/pull/16569 for the (hopefully) correct combination of Kahan and incremental mean calulation.
Thank you. I still have to process your latest response with the depth it deserves. However, I'm still caught between several events at work and in my usual life, but...
Great summary. For consistency, I guess we should also let `histogram_fraction` emit an info annotation if it processes a histogram with NaN observations.
Just one more note from my side. (But as said, I'm not authorized to approve OMv2 related things.)