Bryan Boreham

Given a txring of size 4096, does that suggest >2048 scrapes passed during one query? Which would be over 100 minutes at 3s per scrape.

Personally I find those graphs very difficult to read; I can work with the pprof file. There is a service at https://share.polarsignals.com/ which allows upload and viewing, however you can...

Today we have somewhat similar symptoms in one of Grafana Labs' production clusters. Scrape interval is 30s; process is OOMing around 200GB, with 18GB of `txRing` shown in heap profiles....

Prometheus, v2.30.1.

CI failing on Windows only: ``` === RUN TestDeletedSamplesAndSeriesStillInWALAfterCheckpoint head_test.go:1028: Error Trace: D:\a\prometheus\prometheus\tsdb\head_test.go:1028 Error: Not equal: expected: 9999 actual : 13851 Test: TestDeletedSamplesAndSeriesStillInWALAfterCheckpoint --- FAIL: TestDeletedSamplesAndSeriesStillInWALAfterCheckpoint (0.31s) ``` It seems...

I did a `git commit --amend` with no change and the test passes now, so I'll call that one a flake.

This was the result on one of the production systems at my work: memory usage dropped by half.  https://twitter.com/_codesome/status/1577264183208398848

There is a hard-coded limit of 11,000 points for range queries.

If reading a lot of raw samples in stages is what you want, then the [remote-read API](https://prometheus.io/blog/2019/10/10/remote-read-meets-streaming/) is more suitable. Brian's blog post which you linked to is more about...

Recently I have hit this myself: if Prometheus returns thousands of exemplars then the client library will spend several seconds parsing them. The client I'm using, Grafana, then down-samples the...