We recently deployed kafka 2.1, and noticed a jump in kafka.server.ReplicaFetcherManager.MaxLag.Replica metric. At the same time, there is no under-replicated partitions for the cluster.
The initial analysis shows that kafka 2.1.0 does not report metric correctly for topics that have no incoming traffic right now, but had traffic earlier. For those topics, ReplicaFetcherManager will consider the maxLag be the latest offset.
For instance, we have a topic named `test_topic`:
kafka reports ReplicaFetcherManager.MaxLag.Replica be 99043947579