A broker is one Kafka server process. It:
- Stores topic partition logs on disk (segment files)
- Serves produce and fetch requests from clients
- Holds replicas of partitions for durability
- Participates in cluster membership and leadership
Cluster
Several brokers form a Kafka cluster. Partitions and replicas are spread across brokers so load and failure risk are shared.
┌────────── Kafka cluster ──────────┐ │ broker-1 broker-2 broker-3 │ │ P0-L P0-F P1-L │ │ P1-F P2-L P2-F │ └───────────────────────────────────┘ L = leader replica, F = follower replica
Bootstrap servers
Clients are given a few broker addresses (bootstrap.servers). From those, the client discovers the full cluster metadata (which broker leads which partition).
Failure behavior
If a broker dies:
- Partitions that had their leader on that broker elect a new leader from remaining in-sync replicas.
- Producers/consumers refresh metadata and continue (with some downtime for those partitions).
Laptop vs production
One broker is fine for local learning. Production uses multiple brokers (and replication factor > 1) so a single machine failure does not lose data or kill the cluster.
Interview tip: "Broker = Kafka server. Cluster = brokers sharing topics and replicas."