On cgroup v2 nodes, which kubelet configuration field makes the kernel OOM-kill only the offending process in a container instead of every process in it?

Answer: singleProcessOOMKill: true

Explanation

By default on cgroup v2 the kubelet sets memory.oom.group on each container's cgroup, so an OOM kill takes down all processes in the container together. singleProcessOOMKill: true stops the kubelet from setting that flag, which matches the old cgroup v1 behavior. kernelMemcgNotification is a real field, but it only makes the kubelet notice memory pressure faster for eviction.