As I start doing ops for Cortex, I'm wanting to have per-user metrics. Like the churn of their series, the samples per second per user, the active series per user, the histogram of queries per user, etc.

I understand we have a status page to lookup active series and samples per second, but that looks like it's just a hack around not having actual metrics we can alert on.

I understand that the argument against adding per-user metrics is the high-cardinality, but even a small user exposes atleast 10K metrics and we'll be barely be adding like 30 metrics per user which is a very small percentage of the overall metrics to even matter.

I'm curious to know the thoughts on adding per-user metrics.