Confluent Platform KRaft migration: the window is 7.6.1 to 7.9

The Confluent Platform KRaft version window​

A Confluent Platform KRaft migration is the standard in-place dual-write procedure with commercial version boundaries around it. The documented process applies to CP 7.5 through 7.9. Production migration became generally available in CP 7.6.1, and Confluent recommends running it on 7.7.0 or later — explicitly not on anything running Kafka 3.6.1 or earlier.

The forcing function is CP 8.0. It is built on Kafka 4.0 and ships without ZooKeeper entirely, so CP 7.9.x is the last release line that can run a ZooKeeper-mode cluster. Confluent's support policy gives existing ZooKeeper deployments a runway after the 7.9.0 release — two years on Standard support, three on Platinum — but that is a support-window fact, not a technical one.

Recent patch releases include a preflight that checks the cluster configuration and reports migration status before you commit to anything:

kafka-migration-check status --controller-config ./etc/kafka/kraft/controller.properties

If you run Confluent's tooling, both Confluent for Kubernetes and Confluent Ansible automate the procedure. The Ansible flow is two-phase and deliberately pauses at dual-write mode — the last point where rollback to ZooKeeper is possible — for validation before completing the move. The step-by-step mechanics of the underlying KIP-866 flow are in the complete ZooKeeper to KRaft migration guide; other platforms are routed from the ZooKeeper to KRaft overview. Still weighing the move itself? The KRaft vs ZooKeeper comparison settles the architecture question first.

Back up the cluster before the first roll​

The in-place migration preserves topic data and committed offsets — when it goes to plan. What it cannot protect you from is an operator mistake during a fleet-wide rolling restart, faithfully replicated to every follower in milliseconds. A point-in-time backup outside the cluster is the safety net that does not depend on the migration going well.

OSO Kafka Backup connects to Confluent Platform as an ordinary Kafka client — no Confluent-specific integration needed. Use the same listener security your clients use; every key below comes from the configuration reference.

mode: backup
backup_id: "cp-pre-kraft-migration"

source:
bootstrap_servers:
  - broker-1.cp.internal:9092
  - broker-2.cp.internal:9092
security:
  security_protocol: SASL_SSL
  sasl_mechanism: SCRAM-SHA512
  sasl_username: ${CP_SCRAM_USER}
  sasl_password: ${CP_SCRAM_PASSWORD}
  ssl_ca_location: /etc/kafka/ca.crt
topics:
  include:
    - "*"
  exclude:
    - "__consumer_offsets"

storage:
backend: s3
bucket: my-kafka-backups
region: us-east-1
prefix: backups/cp-pre-kraft

backup:
compression: zstd
include_offset_headers: true   # Reposition consumer groups on restore
source_cluster_id: "cp-prod-zk"

kafka-backup backup --config backup.yaml

kafka-backup list --path s3://my-kafka-backups/backups/cp-pre-kraft \
--backup-id cp-pre-kraft-migration

A backup taken in ZooKeeper mode restores to a KRaft cluster without changes — the storage format depends on the Kafka protocol, not the metadata layer. That also gives you a second route: restore into a fresh KRaft cluster and retire the old one, the pattern in the migration use case. One scope note for CP shops: Schema Registry data is not topic data. Backing up schemas, subjects, and compatibility settings alongside topics is an Enterprise capability; the open source tool covers topic records, offsets, and configuration.

Rollback ends when dual-write ends​

The rollback boundary on Confluent Platform is dual-write mode. While the cluster writes metadata to both ZooKeeper and the KRaft quorum, you can revert; once the migration finalizes and metadata writes to ZooKeeper stop, no vendor can put you back — Apache Kafka has no KRaft-to-ZooKeeper path. That is why the two-phase automation pauses there, and why the pre-migration backup should outlive the migration: after finalization it is the only copy of the cluster's pre-KRaft state. Keep the backup schedule running against the KRaft cluster from day one — the metadata mode changed, the durability question did not. The Confluent Platform integration page covers the ongoing backup setup.