When a broker replicates a partition, its follower can fall behind the leader. The broker-side consumer lag metric, under fetcher lag, reports that gap separately for each follower, topic, and partition. It helps locate replication delay rather than hiding it in a cluster-wide maximum. section 15.2.1
Keep this metric distinct from a consumer group's lag. Group lag counts messages between the group's current offset and the log end offset; it answers whether an application has read the partition. Replica fetcher lag answers whether a follower has kept up with its leader. section 9.3.2 section 15.2.1
Use the broker metrics together:
- Check max lag for the largest gap between a leader replica and its followers, regardless of which follower, topic, or partition is involved. section 15.2.1
- Check consumer lag under fetcher lag to identify the particular follower, topic, and partition that is behind. section 15.2.1
- Expect brief delays, but investigate lag that grows progressively. Ideally, lag stays minimal and does not exceed the maximum production batch size. section 15.2.1
