"If you build a powerful prometheus server then a* total of 2 million timeseries* is doable; beyond that you ought to look at sharding across multiple servers."
I think you have forgot a zero [email protected] schrieb am Dienstag, 15. September 2020 um 09:46:49 UTC+2: > On Tuesday, 15 September 2020 06:11:48 UTC+1, Nick wrote: >> >> Keeping cardinality explosion in mind, what's a decent maximum number of >> exported metrics that can be considered performant for scraping and >> time-series processing? > > > It depends on how much resource you're prepared to throw at it. If you > build a powerful prometheus server then a total of 2 million timeseries is > doable; beyond that you ought to look at sharding across multiple servers. > > As I mainly need the counter total, I can split the web analytics to >> reduce the number of possible label combinations, for example: >> >> { domain, page } >> { domain, browser } >> > > Yes that's fine, but you still want to limit the number of values for each > label. As Stuart said: in the case of browser you don't want the raw > User-Agent header, but pick between a small pre-determined set of values > like "Firefox", "Chrome", "IE", "Other". In the case of "page" strip out > any query string, and ideally also limit to known pages or "Other". > > If you also want to record the raw User-Agent values for every request > then do that in a separate logging system (e.g. loki, elasticsearch, a SQL > database, or even just plain text files) > -- You received this message because you are subscribed to the Google Groups "Prometheus Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/prometheus-users/ad68efaa-63c9-46f6-8a8a-7852ad1373d0n%40googlegroups.com.
