Accurate CPU usage with Prometheus from WIndows

3
votes

Using the wmi_exporter or the scollector_exporter with Prometheus I am finding it difficult to get accurate CPU usage. Here is the metrics I am using and the query I am using for scollector:

os_cpu with returns: 1.54432653e+07

I do a query with rate:

rate(os_cpu{exported_instance="myHost"}[30s])

Here is the graph I have come up with from this query in GrafanaGreen line is the utilisation

os_cpu returns a overall CPU usage i.e. all cores, and comparing this with Taskmanager in Windows it does not add up as that shows 100% max. It cannot be possible to get 300% CPU usage.

What can I do with my query to get a more accurate measurement?

2
monitoringcpu-usagemetricsgrafanaprometheus
I'm not sure where this os_cpu metric is coming from - wmi_exporter? At any rate, perhaps the best thing to do would be to ask on their github page what the metric means. – FuzzyAmi
@FuzzyAmi Yep, and I think I got a better metric by using a larger time time range, I dont see spikes inbetween but its a better graph. I used 2m instead of 30s – W Khan

2 Answers

3
votes

For now you might have found the answer, but anyway... This seems to be useful:

100 - (avg by (instance) (irate(windows_cpu_time_total{mode="idle", instance=~"$server"}[1m])) * 100)

from grafana dashboards library: https://grafana.com/grafana/dashboards/12566

enter image description here

2
votes

If you have multiple cores the usage can go above 100%

I suggest you use 100 - (avg by (instance) (irate(node_cpu{job="node",mode="idle"}[5m])) * 100)

Here is a more detailed blog post about it