An in-sync replica (ISR) is a follower replica whose log has caught up with its partition leader closely enough to remain eligible for leadership if the leader fails. Kafka uses replication to keep partition data available across brokers. section 2.9.1
Track an ISR through the replication flow:
- The leader appends producer records to its partition log. section 2.9.1
- Follower brokers fetch those records from the leader and append matching data and offsets to their own logs. section 2.9.1
- A follower becomes an ISR member when it is fully caught up; replica.lag.time.max.ms removes it after excessive fetch or catch-up delay. section 2.9.1
- Consumers read only through the high-water mark, the latest offset stored by all in-sync replicas. section 2.9.1
A brief replication gap
After a producer sends a batch, the leader may temporarily contain records that followers have not copied. That short lag is normal; persistent lag can remove a follower from the ISR and reduce failover choices. section 2.9.1
Figure 2.15 Followers must issue a fetch request or be caught up within the lag time configuration.
Use the replication-lag diagram to see how fetch timing and the lag configuration determine ISR membership. section 2.9.1
