I like your approach Julius. But I have a question about all of the buckets that I create. If I create all of the buckets with the same output metric name, will each bucket overwrite the previous buckets? If not, will the output metric name now be a histogram that contains all of the buckets? Thanks for your reply!
On Wednesday, January 24, 2024 at 8:03:03 AM UTC-8 Julius Volz wrote: > For the histogram buckets, is your intention to have a few preconfigured > "age" buckets like 0s-15s, 0s-1m, 0s-5m, and so on (cumulative buckets, all > starting at 0, as in normal Prometheus histograms), and then a count for > how many timestamps you have that fall into each age bucket? > > You could make one recording rule for *each bucket*, with all the > recording rules sharing the same output metric name. For example, for a > bucket with a range of 0s - 1m, you could have this kind of rule: > > record: my_timestamp_age_bucket > expr: count(time() - my_timestamp_seconds <= 60) > labels: > le: "60" > > ...and equivalently for the other buckets. > > If you wanted to have a *non-cumulative* histogram (and don't need to use > histogram_quantile() on it), the rule would look like this (e.g. for a > bucket from 15s - 1m): > > record: my_timestamp_age_bucket > expr: count(time() - my_timestamp_seconds > 15 <= 60) > labels: > # some other label(s) here to designate your bucket range > > Remember to include a catch-all bucket for very long durations with an > upper bound of +Inf, so you don't miss recording them. > > Note that with the "count()" approach I use above, it will only create a > bucket time series for buckets with at least one timestamp falling into the > bucket, as otherwise "count()" will just return an empty result. If you > want to create a bucket with the value of "0" instead, you can append "or > vector(0)" to the rule expression: > > record: my_timestamp_age_bucket > expr: count(time() - my_timestamp_seconds <= 60) or vector(0) > labels: > le: "60" > > On Mon, Jan 22, 2024 at 4:25 PM 'Suki Sahota' via Prometheus Users < > [email protected]> wrote: > >> I have a metric I'm collecting which is a gauge vector of timestamps. I'd >> like to take each one of those timestamps and subtract it from the time now >> and turn it into a histogram. Can I do this with recording rules? >> >> Right now I have a gauge vector of: >> <timestamp_1, timestamp_2, ..., timestamp_n>. >> >> I'd like a histogram of [timeNow - timestamp_i]. I'm specifically looking >> to use recording rules. >> >> Thanks, Suki >> >> -- >> 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/d542aed9-56bf-4737-9720-775b4f27d03dn%40googlegroups.com >> >> <https://groups.google.com/d/msgid/prometheus-users/d542aed9-56bf-4737-9720-775b4f27d03dn%40googlegroups.com?utm_medium=email&utm_source=footer> >> . >> > > > -- > Julius Volz > PromLabs - promlabs.com > -- 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/517249f0-e960-4f68-9e1f-15895b8f218an%40googlegroups.com.
