Kubelet PLEG message size


One of the reasons why PLEG could enter a failing state is because the grpc message grows over its defined 16MB max limit

What is not clear is what counts towards this size. Based on this comment, one would think that counting bytes in the output of the docker inspect of all the containers in a host would give the answer, but we’ve seen up to 32.000.000 bytes while the node was still ready, so we must be missing something.

Question is: What counts towards the size of the grpc message carried by the PLEG component?