If you love minute rates on a graph, use scrape_interval ≤ 15s

(I’m talking about Prometheus monitoring here).

There is a lot of personal preferences, opinions and ‘taste’ at discussion what value for scrape_interval should be.

Today I’ve heard the first real argument for having it 15 seconds or less.

As a premise I take idea, that a good-working rate[1m] graph (without glitches or odd/missed points) is a must. If not, than all this reasoning do not apply. I assume, it is.

If so, you need to have 4x of rate scraping interval. If rate is per minute (per 60 seconds), than scrape_interval should be 60/4 = 15 seconds.

There is a well written-argument I agree with, and now I know, that scrape_interval should be 15 or less.

https://www.robustperception.io/what-range-should-i-use-with-rate/