CPU Metrics Nonsensical

scoutapp / scout_apm_python

ScoutAPM Python Agent. Supports Django, Flask, and many other frameworks.

https://scoutapm.com

MIT License

67 stars 21 forks source link

CPU Metrics Nonsensical #131

Open alexottoboni opened 5 years ago

alexottoboni commented 5 years ago

Viewing CPU usage for a node on the Python monitoring page gives data that doesn't make sense. See attached photo.

screen shot 2018-10-23 at 5 06 29 pm

dlanderson commented 5 years ago

Thanks for the report @alexottoboni. Do you happen to be running your apps in containers or have any resource limits applied via cgroups?

alexottoboni commented 5 years ago

They are running in containers on AWS ECS Fargate. Is this already a known issue? If so I should have looked through the documentation better, haha.

dlanderson commented 5 years ago

@alexottoboni Thanks for the info. Sometimes containers can complicate the calculation of CPU usage resulting in inaccurate info. Sometimes we also calculate incorrect info even without containers: we'll investigate making the CPU stats more robust. The rest of the metrics gathered are not affected.

dbanty commented 4 years ago

I've noticed this same thing, also running in ECS Fargate here.