VpcCNI is making list calls on all nodes & CNINode in cluster and causing scalability issues on large clusters.

[hakuna-matatah]

What would you like to be added:

Based on the code it seems like vpc-cni is using informer for k8s client and warms up the cache on startup iiuc for certain resources like nodes, cninodes etc
I think ask is to change this to use a direct k8s client or a informer cache on specific object name/namespace if possible as vpc-cni only need single Node object from my understanding given its a deamonset.

Why is this needed:

LIST calls are expensive for nodes and cninodes on large scale clusters especially when warming up client side cache as it causes lot of data to be exchanged especially when cni-node crash loops on very large cluster and overload APIServer.

[hakuna-matatah]

It appears node filters are added recently in last release. We should be doing the same for cninodes.