You’re being tested on how to identify partitions that currently have no leader. In Kafka, every partition normally has a leader broker that handles all reads and writes for that partition. If that leader broker goes down or is in the middle of a leader election, the partition temporarily has no leader and is considered unavailable. The kafka-topics.sh tool can surface this state when you describe a topic and request the partitions that are unavailable. The flag to include those partitions is --unavailable-partitions. To get this information, you connect to the cluster state (in this setup via Zookeeper) and run the command with --zookeeper, along with --describe and --unavailable-partitions. This combination will list partitions that currently lack a leader. Using bootstrap-server mode can behave differently for this particular flag in some environments, so the Zookeeper-based approach gives the reliable output for showing partitions without a leader.

In Avro schema evolution, full compatibility means both backward and forward compatibility hold: data written with the old schema can be read with the new schema, and data written with the new schema can be read with the old one. If you add or remove a field and give that field a default value, you preserve both directions. Adding a field with a default means old data (which doesn’t have that field) can be read by the new schema—the missing field is filled in with its default. Removing a field while keeping a sensible default for that field in the context of evolution allows older data to be read by the new schema and keeps newer data readable by older schemas, because differences in fields are either ignored or provided by defaults. This dual-direction compatibility is why the change is considered full compatibility.

The main idea is how metadata is stored and who manages it. In a ZooKeeper-based Kafka cluster, ZooKeeper holds the cluster’s basic structural metadata—primarily which brokers are part of the cluster (cluster membership) and certain topic/partition configuration information. It does not hold the actual topic data or the consumer offsets, which are managed by Kafka itself. When moving to KRaft, that metadata responsibility shifts into Kafka’s own system. The metadata is stored in Kafka’s internal metadata log, replicated across the cluster using the Raft protocol, and managed by the brokers acting as the metadata quorum. There’s no separate ZooKeeper involved for cluster metadata in this setup. So, ZooKeeper houses the membership and related configuration data, while under KRaft, the metadata lives in Kafka’s own log on the brokers.

The change being tested is that KRaft removes the need for ZooKeeper by letting Kafka manage its own cluster metadata inside the Kafka system itself. In KRaft, a built-in controller quorum, powered by the Raft protocol, runs within the Kafka brokers and stores all metadata there, rather than relying on an external ZooKeeper service. This makes the cluster self-contained and eliminates ZooKeeper entirely. That’s why the best answer states that ZooKeeper is eliminated and metadata is stored within Kafka via an internal controller quorum. The other options either keep ZooKeeper, suggest an external database, or describe a ZooKeeperless setup without mentioning that the metadata is stored inside Kafka, which is the key distinction of KRaft.

In a system that uses a leader-follower replication model, the followers stay in sync by connecting to the current active leader to fetch and apply updates. This replication traffic happens over a dedicated port on the leader that is intended for inter-broker communication. That port is what followers use to establish the replication session and receive new log entries from the leader, keeping the follower replicas up to date. The client port is reserved for end-user clients (producers and consumers) to connect to brokers, not for replication between brokers. The election port is only involved during the process of electing a new leader and isn’t used for ongoing replication once leadership is established. The data port isn’t a standard term used for this purpose in the typical replication SAMPLEsetup. Therefore, the port used by followers to connect to the active leader is the Leader port.