Björn Rabenstein
Björn Rabenstein
What's the rationale behind moving some types from `tsdbutil` to `chunkutil`? It's also happening in the same commit as the bug fix so it makes it really hard to see...
Please @-mention me on anything where you need my timely feedback so that I notice. (My last day before my vacations is this Thursday 2023-07-27. Only back on 2023-08-08.)
> I did some more digging and there was a detail I missed that preserved the snapshotting loading for the existing implementation. [When we load a chunk snapshot, we set...
I did a bit of archaeology (which was hard because of accidental squashing). The in-memory chunk snapshotting was introduced in https://github.com/prometheus/prometheus/pull/7229 . There is a lot of discussion in this...
@SuperQ @roidelapluie could we come to a conclusion here?
WRT native histograms: While the internal representation is very efficient, it requires very specific knowledge to decode. If the goal of this Arrow approach is to shovel vast amounts of...
Looking at this during the bug scrub. @csmarchbanks is this still on your radar? The idea seems to be universally loved.
I believe the behavior you are seeing is correct. Your query predicts the value 1 hour (`3600` seconds) into the future, based on the data of the last two days...
As a side note: `predict_linear` is not really suitable to draw a forecast graph in Grafana.
What's the `factor` you are using? You cannot directly compare the performance of native and conventional histograms because they have a very different feature set. For a somewhat fair comparison,...