kubernetes: Container start failed on large cpu machines - cpu-shares gets set larger than the maximum of 262144

What happened:

Deployment of a POD onto a machine with 488 cores fails if the POD gets deployed with QoS Guaranteed or Burstable when cpu request is larger than 256.

Kubelet Error Trace: RunContainerError: failed to start container “4ad8bf2f988c9c387d8136ac664caf4af80648ac436e92dac6ecd1f9aece9144”: Error response from daemon: OCI runtime create failed: container_linux.go:349: starting container process caused “process_linux.go:449: container init caused "process_linux.go:415: setting cgroup config for procHooks process caused \"The maximum allowed cpu-shares is 262144\""”: unknown

What you expected to happen:

Normal start of the container with the right number of cpu-shares set yo the cgroup the POD runs in.

How to reproduce it (as minimally and precisely as possible):

Install kubernetes on a machine with more than 300 cores and deploy a POD with QoS Guaranteed or Burstable where the cpu-request value is set to more than 256 cpus.

Anything else we need to know?:

Deployment without CPU request and limits works. Deployments with cpu-request <= 256 works as well.

I found this change: https://github.com/kubernetes/kubernetes/pull/93248 Unfortunately the changed code does not get used in all cases - is not used in my case.

Environment:

• Kubernetes version (use kubectl version): Kubernetes v1.19.8
• Cloud provider or hardware configuration: AWS / EC2 u-6tb1.metal (6 TB memory / 488 vCPUS)
• OS (e.g: cat /etc/os-release): NAME=“SLES” VERSION=“15-SP1” VERSION_ID=“15.1” PRETTY_NAME=“SUSE Linux Enterprise Server 15 SP1” ID=“sles” ID_LIKE=“suse” ANSI_COLOR=“0;32” CPE_NAME=“cpe:/o:suse:sles:15:sp1” VARIANT_ID=“chost”
• Kernel (e.g. uname -a): Linux ip-10-250-0-15.eu-central-1.compute.internal 4.12.14-197.72-default
• Install tools:
• Network plugin and version (if this is a network-related bug):
• Others: