1. it seems that kata does not comply to pod's resource-requests. default_memory for kata is obligatory. the total memory is default_memory plus pod's requests. inside kata 'cat /proc/meminfo', it shows total memory. In other word end user can see it and notice it is not the memory size they request.
2. how does containerd/dockerd negotiate cpu/memory with kata no matter dynamically or one-time? source code is preferred.