Metrics API Improvent about vespa HOT 5 CLOSED

you don't need any special vespa support for this, as long as you sample the vespa metrics every minute and insert it into prometheus you can do the averaging there (just use the standard prometheus sum/count expression from the prometheus documentation).

note that this is exactly how yahoo's internal metrics monitoring / graphing system does it also, so this is the use case that current vespa setup is intended for, and should work very well.

from vespa.

I think it works fine when LAST metrics is supported.However, some vespa metrics do not have LAST metrics but only RATE metrics. For example search_connections does not support it.

Can we add SUM metrics to they?

from vespa.

can't you do the sum of averages? from the prometheus docs it looks like you want to use their 'summaries' concept to do the summing for you:
https://prometheus.io/docs/practices/histograms/

you shouldn't do summing or averaging of the LAST metric, that will sometimes yield incorrect results, since it will repeat the last value seen when there's no new measurements in a period.

from vespa.

rate() can calculates the per-second average rate of increase of the time series:
https://prometheus.io/docs/prometheus/latest/querying/functions/#rate()

from vespa.

@arnej27959 It seems I misunderstood the specification of metrics API of Vespa. Sorry for confusion. Now I understood the correct specification, and I will try to calculate what I need using "rate" and "average" metrics currently available.

Thank you for your explanation.

from vespa.