In-Sync Replicas Management

In-sync replicas (ISR) are crucial for Kafka's reliability guarantees. A replica is considered in-sync if it:

• Maintains an active ZooKeeper session
• Fetches messages from the leader regularly
• Catches up to most recent messages within configured time

Key Configurations

1. zookeeper.session.timeout.ms

   • Time before broker considered dead if no heartbeat
   • Default: 18 seconds (since 2.5.0)
   • Higher values improve stability in cloud environments

2. replica.lag.time.max.ms

   • Maximum time a follower can lag behind leader
   • Default: 30 seconds (since 2.5.0)
   • Affects consumer latency

3. min.insync.replicas

   • Minimum replicas that must acknowledge writes
   • Works with acks=all for strongest durability
   • Partition becomes read-only if not met

References

Flashcards

What makes a replica considered "in-sync"?:: Active ZooKeeper session, regular fetches from leader, and caught up to recent messages

What happens when available replicas fall below min.insync.replicas?:: Partition becomes read-only for producers (but consumers can still read)

What is the purpose of replica.lag.time.max.ms?:: Defines maximum time a follower can lag behind leader before being removed from ISR