Each pod has weight. These are written to your cgroup (cpu.weight in CGroups V2). CFS is scheduling based on these weights.

There's a part about it: https://github.com/inevolin/k8s-cpu-limits-analyzed#3-withou...