Topbeat/metricbeat is not showing correct value for proc.cpu.user_p

[Huy_Hoang]

Hi,

I am using elk stack version 5.0 and 1.3/5.0 for topbeat on centos 7. The problem is that the proc.cpu.user_p is always < 0.9. Normally, the values are showed as 0.0xx (0.001, 0,003 .v..v), I tried to run a heavy task on the instance but stil, the value stopped at 0.9. And with that heavy task, cloudwatch monitor (I use ec2 instance) gave me only ~45% CPU usage. Both topbeat and metricbeat have the same problem.

At first, I was thinking the proc.cpu.user_p value is divided by 100 but seems like it's not the root cause. Please help!

[ruflin]

Which alpha version of metricbeat were you testing? The value for user_p with one processor will be between 0-1. So it kind of looks quite reasonable to have 90% usage peak above? The next version of metricbeat will have an import_dashboard script that will also load the index pattern for kibana, so Kibana will now it is a precentage value and will show it accordingly.

Can you compare your metrics with the top command. The load is always an average between to points in time. So in case metricbeat checks every 10s but Cloudwatch checks every minute, the averages can be different.

[Huy_Hoang]

Thanks, I have checked this with newrelic and cloudwatch again. Seems like it works perfectly :).

[ruflin]

Thanks for double checking and reporting back.