I've been debugging several services to reduce billing on what is fairly ephemeral compute requirements (Cron, low periodicity, etc.)

Solved all issues on persistent services, but have been having the damndest time getting a cron service to quit having, what appears on the metrics, a 4-5GB baseline memory consumption.

Turns out the Cron service runs once per 24hr and also samples once per 24h, so a plot over e.g. 7 days shows the amount of usage for that service while running (it makes sense to be fair.)

I might be the edge case, but visually it was very difficult to not read that as "this service maintains a 4-5gb baseline of usage throughout the period". For persistent services, I see normal high peaks, low down time, etc -- very choppy looking charts.

The cron's actual performance is that: on for a bit, then killed until the next schedule. However, the chart doesn't really convey it. I know how to read it now, but man it was a mind fuck trying to debug what could make it not shutdown properly and keep eating ram all day.