Ben Kochie
Ben Kochie
I think having a feature flag here is probably fine. As long as we strongly word the documentation about the impact it can have. Also, I don't remember if this...
Because Prometheus histograms a are cumulative you can drop them via [metric relabel configs](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#metric_relabel_configs). There is also the option `convert_classic_histograms_to_nhcb` that can be used to convert these to [native histograms](https://prometheus.io/docs/prometheus/latest/feature_flags/#native-histograms).
Nope, PRs welcome.
@dependabot rebase
@dependabot rebase
On this topic, I think one useful first feature is to have a metric that can be used to track the currently stale series in the head. Something like `prometheus_tsdb_head_stale_series`....
This needs a rebase.
An opt-in flag would be fine. The only issue is that this collector setup hasn't been moved to our new collector package, so it's possibly slightly difficult to implement here.
:eyeglasses: @grobinson-grafana
I think putting it behind a feature flag is a good idea. Also probably good if we made that sysfs parsing code more granular. I'm open to doing both.