Ganesh Vernekar

> enough to generate float histograms (not integer histograms) Since float and integer histograms have different chunk encoding underneath, it would be good to test both of them

Also, testing counter vs gauge histograms

cc @bwplotka to weigh in on introducing a cache for matchers->postings for the head block. Can help make ruler queries more efficient as well.

Makes sense to me. @qinxx108 and I discussed a couple of ways we could do this during office hours. @qinxx108 can you post some more detail about the token idea...

Another idea to think about that does not involve the tokens: We just need to provide the page number and limit per page in the request. And the backend will...

> But my impression was there is some appetite to both put more samples into a chunk and also (soft-)limit the size of chunks. Yes, but I would like to...

/prombench main I have not checked the code yet, but running the benchmark for a few hours. And the fuzzer failure is likely flaky; we have been seeing it a...

/prombench cancel This is looking better. Review is on my TODO list!

I am reviewing the PR now (thanks for the work and patience!), and I am having second thoughts about the linked list. @bboreham from https://github.com/prometheus/prometheus/pull/10709#issuecomment-1364022828, we figured that having it...

I checked on of our Mimir deployments which uses this TSDB underneath, assuming 32 bytes total overhead per series with GC, it is 0.6-0.7% more memory (same with linked list...