CPU usage on hosts

[kickalillas2]

I am having an issue on our windows systems in which XI reports incorrect CPU values. This is occurring on all the Windows systems I have spot checked. Our environment is a mix of 2019, 2016, and 2012R2. Our Linux hosts, RHEL 8/Centos 7, seem very close to what is reported in XI.
I first noticed this issue last week, but I can't say for certain how long or if it has always been present. However, we had a production problem yesterday where a process ended up using 100% CPU, shown via windows task manager and XI was showing between the 40%-70% range.

XI CPU.jpg

Our hosts are well specced so we typically don't have problems with this.

Host Cpu.png

Troubleshooting steps which I've already taken... Upgrading XI to latest version. Upgrading the agent on a few test machines. I have even tried setting up an alternate monitor via SNMP and those values came back close to or the same as the NCPA agent values. I am at a loss on what else I can try here so I can only assume this is a bug that was introduced at some point. Is anyone else experiencing something similar or have any ideas that I can try?

Thanks

XI Version - 2024R1.1.5
NCPA - Mix of 3.1 or 2.4

[jmichaelson]

I have even tried setting up an alternate monitor via SNMP and those values came back close to or the same as the NCPA agent values.

By that you mean that those values matched what NCPA was reporting to your XI server? If not, how were you getting the NCPA values?

[kickalillas2]

I have even tried setting up an alternate monitor via SNMP and those values came back close to or the same as the NCPA agent values.

By that you mean that those values matched what NCPA was reporting to your XI server? If not, how were you getting the NCPA values?

I apologize if not clear, but I added a second check against the host using the SNMP wizard. When I first added the SNMP check, these values looked closer, but now they are way off, so that may have been a fluke. I understand based on polling checks and response times, the reported values might vary a bit, but we are not even in the same ballpark as what is being reported on the host itself.

SNMP.png

[jmichaelson]

I can't speak for how SNMP grabs its CPU utilization numbers (but even those seem low). You can see here https://github.com/NagiosEnterprises/nc ... r/psapi.py that NCPA is using the psutils CPU percent call, which defaults to checking the usage over a half-second interval:

ps.cpu_percent(interval=cpu_interval, percpu=True)

Do you by any chance have that interval turned down in the NCPA configuration? That could dramatically effect the numbers for CPU usage.