On Mon, Feb 24, 2020 at 8:56 PM Yongjik Kim <[email protected]> wrote:
> Hi Aliaksandr, > > Thanks a lot for the reply, but I think quantile_over_time() will compute > percentiles over each series? > Yes. > > So, for example, if I have three different time series A/B/C (representing > three instances of the same task T), and I use quantile_over_time(), then I > could get "95% CPU usage of A/B/C" separately, but it still won't tell me > "95% CPU usage across all instances of T", as far as I can tell. > I'm afraid PromQL doesn't provide the functionality, which can be used for calculating percentiles over data points on the given range from multiple time series :( The closest approximation is max(quantile_over_time(0.95, ...)) . I don't recommend using avg() instead of max(), since it hides time series spikes. > > > > On Mon, Feb 24, 2020 at 10:41 AM Aliaksandr Valialkin <[email protected]> > wrote: > >> Hi Yongjik! >> >> Try using `quantile_over_time` instead of `quantile`. See >> https://prometheus.io/docs/prometheus/latest/querying/functions/#aggregation_over_time >> >> >> On Fri, Feb 14, 2020 at 9:06 PM 'Yongjik Kim' via Prometheus Users < >> [email protected]> wrote: >> >>> Hi, >>> >>> I have a problem with aggregation. I want to get the CPU usage of a set >>> of jobs (each with potentially different start/stop time), over the past >>> week, and then get 95% percentile among these values. >>> >>> So, I can get the raw data points with this: >>> >>> > rate(cpu_usage{name="myjob"}[5m])[1d:5m] >>> >>> cpu_usage is an accumulative series (counter?) which records "the amount >>> of CPU resource this job has used since it started." So, as far as I >>> understand, this gives me a nice list of "average CPU usage for each >>> 5-minute interval, for every job and for every interval the job was alive." >>> >>> So far so good, but then how do I get the 95% percentile of *all these >>> values*? >>> >>> If I try this: >>> >>> > quantile(0.95, rate(cpu_usage{name="myjob"}[5m])[1d:5m]) >>> >>> I get: "Error executing query: invalid parameter 'query': parse error at >>> char 147: expected type instant vector in aggregation expression, got range >>> vector" >>> >>> I can make it output *some number* by removing [1d:5m], but that's not >>> what I want. I don't need 95% percentile at the current instant, but over >>> the past week. >>> >>> Any way to make it work without piping the result through a custom >>> script? >>> >>> Thanks, >>> - Yongjik Kim >>> >>> -- >>> 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/637d706d-e39a-4b0a-8ec3-70bdcc9c3cbc%40googlegroups.com >>> <https://groups.google.com/d/msgid/prometheus-users/637d706d-e39a-4b0a-8ec3-70bdcc9c3cbc%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> >> >> >> -- >> Best Regards, >> >> Aliaksandr >> > -- Best Regards, Aliaksandr -- 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/CAPbKnmDdn-T27u_%3DT1e9hGzktYhGsud4u7v3zWrfUPDipZ9L8g%40mail.gmail.com.
