Hey Stuart,
thanks for your help, your provided link helped solving the issue. My
metrics now make more sens. Thank you!
Am Samstag, 18. April 2020 14:45:00 UTC+2 schrieb Stuart Clark:
>
> On 18/04/2020 13:18, 'vivapolonium' via Prometheus Users wrote:
> > Hey everyone,
> >
> > I'm failry new to prometheus and trying to wrap my head around some
> > concepts which are not really clear to me.
> >
> > I'm running a Scala-Application with the official Prometheus Java
> > client. I'm trying to measure the performance of http endpoints and
> > use a `Summary` for that. I implemented an endpoint where I serve the
> > Metrics via an internal andpoint by taking the `TextFormat.write004`
> > method and serving it by myself (not via the included HTTPServlet).
> >
> > I've setup a Prometheus instance querying that endpoint every 15s and
> > set the maxAge of the Summary also to 15s. Now I have a PromQL-Query
> > like this: `sum
> >
> by(route)(requests_latency_seconds_sum/requests_latency_seconds_count)*1000`,
>
> > which should give me the average response-time of an endpoint in
> > milliseconds for each scrape-interval
> >
> > When rendering the data though, I get some kind of weirdly aggregated
> > data points which is probably a mixture of bad settings and
> > misunderstanding. Take this metric for example:
> >
> > ```
> > requests_latency_seconds_count{route="library.get",} 83.0
> > requests_latency_seconds_sum{route="library.get",} 949.2774687769999
> > ```
> >
> > This summary does not reset after 15s, instead it keeps accumulating
> > all the data which makes it useless to pin-point timebased anomalies
> > in my application.
> >
> That isn't a summary (that would have quantile labels), or at least the
> bit you are showing doesn't cover that.
>
> Normal counters don't reset except when the application restarts. Within
> PromQL there is the rate() function which allows you to see spikes in
> latency over time.
>
> So try to add rate() as described at
> https://www.robustperception.io/rate-then-sum-never-sum-then-rate
>
> Generally I don't use summaries and instead use histograms. Summaries
> aren't aggregatable (for example if you run multiple instances) or
> adjustable within Prometheus. With histograms you can aggregate and
> calculate percentiles over any range.
>
> > I digged into the sourcecode of the java library and did not find a
> > way to reset the values to zero or remove them after scraping them. Is
> > this intentionally? Did I miss something in my configuration? Also, as
> > I understood it, the summary is supposed to reset itself?
> >
> > Hope someone can give me some hints how to solve this
> >
>
> --
> Stuart Clark
>
>
--
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/a2807c30-03e2-46db-8ab4-840c73b701a8%40googlegroups.com.
