Why is kubelet cleaning up more images than the defined LowThresholdPercent?

General Discussions
1

Cluster information:

Kubernetes version: 1.27.10
Cloud being used: bare-metal
Installation method: kubespray
Host OS: Ubuntu 22
CNI and version: cilium:v1.13.4
CRI and version: containerd:1.7.13

Can someone explain to me why kubelet garbage collection is cleaning up more images than I’d expect it to? Specifically, this document notes how once GC is triggered:

The kubelet deletes images until disk usage reaches the LowThresholdPercent value.

However, I’m seeing my disk usage drop significantly more. I’ve got a 1TB disk used for /var and see the usage drop from 85% all the way to 51%.

I’m using the default thresholds of 85 and 80 for high and low, respectively. Given the size of my disk, I would expect the GC to bring /var to something closer to 80% usage after finishing. It feels like kubelet is triggering GC and then just cleaning up all unused images, but I would’ve expected it to stop once it crossed the imageGCLowThresholdPercent?

Any help is appreciated.