Hello, I have a question regarding K3s. I have been facing Performance Issues on Bare Metal Systems, which have 2 CPU's resulting in 2 NUMA Nodes or 4 when Sub NUMA Clustering per IMC is enabled. The tasks running on these Systems in these K3s-Pods are quite latency sensitive and depend heavily on memory/cache locality. Does K3s support Kubernetes' Topology-Manager or has it any integrated/native means of being tuned for NUMA Systems? I want to avoid having to use

numactl --cpubind=x --membind=x

as that's not exactly a scalable solution since it would require a lot more manual work per system than e.g. Kubernetes' Topology Manager. I was wondering if the topology-manager is supported through something along the lines of

kubelet-arg:- "topology-manager-policy=best-effort" # or "single-numa-node"

in the Config.yaml file ? PS.: moved it to the K3s Channel, sorry for posting this very specific question in the generic channel