|
I suspect there are a number of conceptual problems here. If a metric (x) is a counter, then the value of the metric is monotonic increasing from some arbitrary point in the past (usually boot time or pmda start time). Now if I represent the value of x at time t as x(t), then typically what one is interested in is the rate at which x increases over a time period, so this is (x(t2)-x(t1))/(t2-t1). This is exactly what pmie will report for a counter. The is exactly what sar, iostat, vmstat, top, et al report for things like cpu use, disk i/os, network packets, network bytes in and out, ... If you really want the value of the counter right now then use pminfo or the instant operator for pmie. Secondly, disk.dev.total is a metric defined over an instance domain (one value per disk) so to pmie this has a set value, and the pmie rules you're using are syntactically incorrect. This is what you probably need: and as you can see from the output below ... there is no value for the rate converted metric at the first sample and the after the rate = the difference in the raw numbers / 2 (the update interval, via -t) — |