Kafka's request handling utilizes a number of processor "threads" and
responses are added to the above queue before sending to the client.
This diagram may help:https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Internals
This reports how far behind that replica is. The format is
(Fetcher id is typically zero (since the default number of replica
fetchers is one).)
Requests per sec/byte rate for replica fetchers, but for all fetches
to the indicated leader brokers (1 and 2).
Kafka exposes a *lot* of mbeans - not all of them are listed on that
page that you are referring to. I think the goal of the table was to
provide a minimal set of critical metrics that people may want to set
up monitoring/alerting for.