2

I currently administer a website hosted on Google Cloud Platform using Kubernetes Engine. Yesterday I noticed the CPU% feed on the GKE node drop dramatically to just north of 0. After investigating it was clear there was nothing wrong---the site is still live and checking CPU usage directly via SSH showed normal load levels. It appears that the Google-managed monitoring agent in GKE is failing to report CPU usage accurately, though all other values are reported correctly.

Has anyone encountered this issue? Is there a solution to restore accurate monitoring values in this case?

--zig

EDIT: To clarify my original post, this feed has been accurately reflecting our live instance's utilization for years before the sudden and inaccurate report of 0%. The feed also continues to be accurate for our dev and testing environments.

1 Answers1

3

The CPU usage of main Dashboard of the GCP differs from that of an instance's individual CPU usage. The CPU usage shown in Google Developers Console is not that of the instance, but the CPU usage of the container managing the instance. This container is in charge of providing the virtualization services to the instance and collecting all the metrics used for load balancing, auto-scaling, cloud monitoring, etc. As such, high numbers of I/O or network operations will cause the CPU utilization shown in Google Developers Console to spike.

As a workaround, you can use Stackdriver Monitoring custom metrics to get more accurate CPU usage metrics.

John Mathew
  • 171
  • 2
  • Thank you for your response. We looked into that quirk of GCP before posting and perhaps should have mentioned it in my original post. I don't think this issue is related. We aren't experiencing unusual spikes, but rather a sudden drop to 0 after years of accurate readings. Our dev and testing instances also continue to report accurate readings. – Ryan Taussig Feb 07 '19 at 21:02
  • If you think the issue is not related, I would recommend opening an issue on the [issue tracker](https://cloud.google.com/support/docs/issue-trackers). Please be reminded to mention all the relevant details and reproduction steps if any. – John Mathew Feb 07 '19 at 22:38