cpu_dragonfly.go's cpu time doesn't seem to be in second against its name (node_cpu_seconds_total) #1129
Description
Host operating system: output of uname -a
DragonFly tech-y09-dfly 5.3-DEVELOPMENT DragonFly v5.3.0.18494.gc70d4-DEVELOPMENT #1: Fri Aug 24 12:27:12 JST 2018 root@tech-y09-dfly:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64
node_exporter version: output of node_exporter --version
Installed from source.
iori@tech-y09-dfly% git log | head -n 3
commit 7519967
Author: Ben Kochie superq@gmail.com
Date: Sat Oct 20 08:21:51 2018 +0200
iori@tech-y09-dfly% node_exporter --version
node_exporter, version (branch: , revision: )
build user:
build date:
go version: go1.10.3
node_exporter command line flags
node_exporter (default)
Are you running node_exporter in Docker?
No
What did you do that produced an error?
Just run.
What did you expect to see?
The node_cpu_seconds_total metrics are in seconds.
What did you see instead?
The total sum of irate of reported metrics are around 40ms which is too smaller than expected (this should be 1 second * ncpu).
The cause
According to cpu_dragonfly.go, the metrics are extracted from kern.cputime and divided by the frequency of kern.cputimer.freq.
However, the comment in the source code states that it counts in microseconds.
https://github.com/DragonFlyBSD/DragonFlyBSD/blob/master/sys/kern/kern_clock.c#L118
The comment is consistent with a shell script as following:
iori@tech-y09-dfly% @ tim=`sysctl kern.cp_time | awk '{print $2 + $3 + $4 + $5 + $6}'` && sleep 1 && @ tim2=`sysctl kern.cp_time | awk '{print $2 + $3 + $4 + $5 + $6}'` ; @ dur = $tim2 - $tim ; echo $dur
4003724
This machine has four cores, so the timer goes around 1000k ticks per CPU in a second, which implies the kernel counts it in microseconds.
I'd like to know why they are divided by that frequency instead of 1/micros.