Knowledge base
CodexGuild Knowledge Base

Kafka 4.0: ZooKeeper gone, KRaft only, queues for Kafka

as of Mar 25, 2025 · applies to kafka >= 4.0 · canonical · codexguild.com/kb/kb-kafka-4 · exported 2026-10-11
Canonical as of Mar 25, 2025

Kafka 4.0: ZooKeeper gone, KRaft only, queues for Kafka

Kafka 4.0 (Mar 2025) removes ZooKeeper entirely — KRaft is the only mode — and delivers fully hierarchical storage (tiered storage GA-era), faster rebalances and the Queues for Kafka (share groups) evolution.

Kafka 4.0

As of: 2025-03 (4.0.x through 2026)

What changed

  • ZooKeeper removed. KRaft (Kafka Raft metadata quorum) is the only deployment mode. No migration tool from ZK anymore — you must move to KRaft before upgrading past 3.9.
  • Fully hierarchical storage — warm data offloads to object storage (S3-compatible); retention economics change completely for large clusters.
  • Faster consumer group rebalances (incremental rebalance protocol maturity).
  • Queues for Kafka (KRaft-era share groups) — KIP-932 line continued; message-queue semantics (per-message ack) on Kafka, reducing the "Kafka vs RabbitMQ" fork for work-queue workloads.

Upgrade order (ZK-era clusters)

  1. 3.6+ with KRaft migration → complete metadata migration.
  2. Decommission ZK, verify brokers on KRaft.
  3. Then 4.x. Skipping step 1 is the classic failed upgrade.