// Kafka 3.x & ZooKeeper extended support

Kafka 3.x on ZooKeeper: stay supported, then move to KRaft

Kafka end of life support for a cluster on ZooKeeper comes down to two facts: Kafka 4.0 removed ZooKeeper mode on 18 March 2025, and Apache now lists only 4.1, 4.2 and 4.3 as supported, so 3.9 and every older line get no upstream fixes. A ZooKeeper-mode cluster must migrate to KRaft on 3.x before it can take 4.x. OSSeva keeps your Kafka 3.x brokers and the ZooKeeper ensemble under them patched while you plan that move, and runs the migration. Enter your work email and the Kafka version you run, and an engineer replies within 1 business day.

Patched builds
Kafka 2.8 to 3.9 brokers, Connect and clients, signed, from your own repository
ZooKeeper ensemble
Patched ZooKeeper 3.4, 3.6 and 3.7, with 3.5 as custom support
KRaft migration
Planned and run cluster by cluster, with 3.9 as the bridge
Critical CVEs (CVSS 9.0 or higher)
Patched within 48 hours
High-severity CVEs
Patched within 7 days
Support hours
Assure includes 24/7 reactive support and the CVE attestation with VEX; Operate adds a 15-minute P1 response

Dates checked on 20 October 2026 against kafka.apache.org, endoflife.date and the Confluent Platform support documentation. They differ on Kafka 3.9 and 4.0: endoflife.date shows no end date, while the Apache downloads page lists both as archived. This page follows the Apache project's list.

Kafka 3.93.83.73.62.8ZooKeeper 3.83.73.6Confluent Platform 7.x

Only your email is required · An engineer replies within 1 business day · No commitment

Trusted globally by enterprises

NokiaSquareLittle CaesarsPanasonic ConnectPaylocity

Kafka versions and ZooKeeper end dates

Kafka 4.0 removed ZooKeeper mode on 18 March 2025, and 3.9 is the last line that can run on it. The Apache downloads page lists only 4.3, 4.2 and 4.1 as supported releases: 3.9 (final release 3.9.2, 21 February 2026) and 4.0 are archived, so a cluster on ZooKeeper has no upstream fix supply left. Dates checked 20 October 2026 on kafka.apache.org and endoflife.date.

LineReleasedApache support endsLast releaseMetadata modeStatus
Kafka 4.322 May 2026Supported4.3.1 (25 Jun 2026)KRaft onlySupported upstream
Kafka 4.217 Feb 2026Supported4.2.2 (29 Sep 2026)KRaft onlySupported upstream
Kafka 4.12 Sep 2025Supported4.1.2 (17 Mar 2026)KRaft onlySupported upstream
Kafka 4.018 Mar 2025Archived4.0.2 (16 Mar 2026)KRaft onlyArchived upstream
Kafka 3.96 Nov 2024Archived3.9.2 (21 Feb 2026)ZooKeeper or KRaftArchived upstream: OSSeva patched builds, and the migration bridge
Kafka 3.829 Jul 20246 Nov 20243.8.1ZooKeeper or KRaftEnded: OSSeva patched builds
Kafka 3.727 Feb 202426 Jul 20243.7.2ZooKeeper or KRaftEnded: OSSeva patched builds
Kafka 3.610 Oct 202327 Feb 20243.6.2ZooKeeper or KRaftEnded: OSSeva patched builds
Kafka 3.515 Jun 20233 Oct 20233.5.2ZooKeeper or KRaftEnded: OSSeva patched builds
Kafka 3.0 to 3.421 Sep 2021 to 7 Feb 202319 Sep 2022 to 13 Jun 20233.0.2 to 3.4.1ZooKeeper (KRaft early access)Ended: OSSeva patched builds
Kafka 2.819 Apr 202119 Sep 20222.8.2ZooKeeper (KRaft early access)Ended: OSSeva patched builds

Sources: kafka.apache.org/community/downloads (supported and archived releases, release dates) and endoflife.date/apache-kafka (end dates of 3.0 to 3.8, which are the day the next minor shipped). The two sources differ: endoflife.date shows no end date for 3.9 and 4.0, while the Apache downloads page lists both as archived. This page follows Apache's list.

Every Kafka line on one chart →Which ZooKeeper version each Kafka release bundles →

Confluent Platform 7.x support dates

Clusters that run under Confluent Platform follow its calendar as well as Kafka's. The last two lines that run on ZooKeeper are 7.8 (Kafka 3.8) and 7.9 (Kafka 3.9). Enterprise Platinum support runs one year past Standard.

Confluent PlatformApache KafkaReleasedCommunity software endsEnterprise Standard endsEnterprise Platinum ends
7.9.x3.9.x19 Feb 202519 Feb 202719 Feb 202719 Feb 2028
7.8.x3.8.x2 Dec 20242 Dec 20262 Dec 20262 Dec 2027
7.7.x3.7.x26 Jul 202426 Jul 202626 Jul 202626 Jul 2027
7.6.x3.6.x9 Feb 20249 Feb 20269 Feb 20269 Feb 2027

Source: docs.confluent.io, Supported Versions and Interoperability for Confluent Platform, checked 20 October 2026. Confluent Platform 8.0 and later are KRaft only.

Running Apache Kafka without Confluent Platform →

Advisories on Kafka 3.x clusters since the earlier lines ended

Three Kafka advisories in the CVE directory, scored up to 8.8, were published after 3.8 ended on 6 November 2024 and list 3.x clusters as affected. Apache fixed each one first in 3.9.1 or 3.9.2, so Kafka 3.8 and earlier have no fix, and 3.9 is archived after 3.9.2. One more, CVE-2024-27309 (scored 7.4), is specific to the migration: while a cluster moves from ZooKeeper to KRaft, removing an ACL can leave a resource with the wrong ACL count, and DENY ACLs may be ignored until the last ZooKeeper-mode broker is gone. Deciding which advisories reach your line and configuration is the CVE attestation OSSeva writes. The ZooKeeper ensemble under the brokers has its own advisories.

Published after Kafka 3.8 ended, first fixed in 3.9.1 or 3.9.2 (3 in the directory)

CVEPublishedCVSSAffected range (NVD)Fixed upstream in
CVE-2025-278182025-06-108.8>=2.3.0 <3.9.1See advisory
CVE-2025-278172025-06-107.5>=3.1.0 <3.9.1See advisory
CVE-2026-355542026-04-078.7>=2.8.0 <3.9.2; >=4.0.0 <4.0.2; >=4.1.0 <4.1.2See advisory

Specific to the ZooKeeper to KRaft migration (1 in the directory)

CVEPublishedCVSSAffected range (NVD)Fixed upstream in
CVE-2024-273092024-04-127.4>=3.5.0 <=3.6.1See advisory

The directory lists Kafka CVEs matched at NVD to the product with a CVSS score of 7.0 or higher where one exists, so it is a subset of every Kafka advisory. Affected ranges are as the MITRE record publishes them: CVE-2025-27818 names 2.3.0 to 3.9.0 (fixed in 3.9.1), CVE-2025-27817 names 3.1.0 to 3.9.0 (fixed in 3.9.1), CVE-2026-35554 names 2.8.0 to 3.9.1 (fixed in 3.9.2, 4.0.2 and 4.1.2), CVE-2024-27309 names 3.5.0 to 3.5.2 and 3.6.0 to 3.6.1.

ZooKeeper extended support: advisories for the ensemble under the brokers →Kafka CVEs by version, 2.8 to 4.x →

KRaft readiness checklist for a Kafka 3.x cluster on ZooKeeper

The migration itself is documented by Apache and runs in phases: provision a KRaft controller quorum, enter migration mode on the brokers, move the brokers to KRaft, then finalise. Until finalisation you can still revert to ZooKeeper mode; after it you cannot. Most stalls come from the items around it, which this checklist collects from Apache's documentation.

CheckWhat breaks if you skip itVersion or setting
Broker versionApache's 3.9 documentation says brokers must be on 3.9.1 with inter.broker.protocol.version set to 3.9 before migration starts.3.9.1 or later, metadata version 3.9
Metadata version during migrationChanging it while the cluster is migrating is not supported and can break the cluster.Freeze inter.broker.protocol.version
Clients, Connect and Streams4.0 removed old protocol versions. Clients must be on 2.1 or later before 4.0 brokers, and clients that are not part of Apache Kafka need the KIP-896 check.2.1 or later
Java runtime4.0 brokers, Connect and tools need Java 17; clients and Streams need Java 11.Java 17 on servers, 11 on clients
Brokers with several log directoriesA directory failure shuts a ZooKeeper-mode broker down during migration, and that broker cannot migrate until the directory is repaired.Repair first (KAFKA-16431)
JBOD in KRaftApache's 3.9 documentation lists multiple storage directories as not fully implemented in KRaft mode.Check before migrating
ZooKeeper Security Migration ToolIf it was run before, the migration can stall on a malformed /migration znode.KAFKA-19480; delete the znode
Removed ZooKeeper settingszookeeper.* settings, control.plane.listener.name, the broker id generation settings, controlled shutdown retries and the password.encoder settings do not exist in KRaft mode. advertised.listeners changes need a restart.Audit configuration and automation
Message format settingslog.message.format.version and message.format.version were removed in 4.0.Remove before the 4.x upgrade
AlterConfigPolicy pluginsZooKeeper and KRaft modes pass different arguments for SUBTRACT, DELETE and APPEND operations; 3.9.2 added a compatibility flag.KIP-1252, 3.9.2
FinaliseAfter the controllers leave migration mode there is no supported way back to ZooKeeper mode. Some teams wait before finalising so that a revert stays possible.Rehearse on a representative cluster

Sources: Apache Kafka 3.9 documentation (KRaft operations: ZooKeeper to KRaft migration, limitations, preparing for migration, reverting); the upgrade notes for 4.x (Upgrading to 4.0 from 3.3.x through 3.9.x; notable changes in 4.0.0; Java versions); the KRaft vs ZooKeeper page. Checked 20 October 2026.

The ZooKeeper to KRaft runbook →Which clients break under KIP-896 →Kafka 3 to 4: the upgrade path →

Why ZooKeeper-based clusters are stuck

The ZooKeeper to KRaft migration is well documented and online. What stretches it out is everything around it.

No direct path from ZooKeeper to 4.x

A cluster must migrate to KRaft on a 3.x release, with 3.9 as the recommended bridge, and finalize that migration before it can take the 4.0 upgrade. The finalize step cannot be undone.

4.0 changes more than the metadata store

Kafka 4.0 also raises the Java baseline and removes old client protocol versions, so ancient clients and brokers pinned to old JVMs surface at the same time.

Confluent Platform 7.x is the last line with ZooKeeper

Clusters that run under Confluent Platform 7.x are tied to its support calendar as well as Kafka's. Platinum support runs a year past Standard, and 7.9 is the final ZooKeeper-capable line.

The dates that matter

  1. 2024-11-06

    Kafka 3.9 released, the final 3.x line and the last that can run with ZooKeeper. Kafka 3.8 reaches end of life the same day.

  2. 2025-03-18

    Kafka 4.0 released in KRaft-only mode. ZooKeeper support removed.

  3. 2025-06-10

    CVE-2025-27818 and CVE-2025-27817 published. First fixed in 3.9.1, so nothing for 3.8 and earlier.

  4. 2026-02-21

    Kafka 3.9.2 released, the final 3.9 release. The Apache project lists 3.9 as archived.

  5. 2026-12-02

    Confluent Platform 7.8 end of support (Standard); Platinum runs to 2 December 2027.

  6. 2027-02-19

    Confluent Platform 7.9 end of support (Standard), the final ZooKeeper-capable Confluent line; Platinum runs to 19 February 2028.

What OSSeva delivers

1

Patched Kafka 3.x brokers

Backported CVE fixes for the Kafka line you run, community Apache Kafka or the Apache core of a Confluent deployment, delivered through your own repository.

2.8–3.9Brokers & ConnectSigned builds
2

Patched ZooKeeper ensembles

Security fixes for the ZooKeeper versions your clusters depend on, the component most often forgotten in a Kafka inventory.

ZooKeeper 3.4, 3.6 & 3.73.5 customTLS & authEnsemble review
3

ZooKeeper to KRaft migration

Controller quorum sizing, migration mode, broker rolls, the finalize decision and the 4.x upgrade, run as a change programme across every cluster.

KRaft3.9 bridge4.x upgrade

Stay on 3.x: what extended support covers

Extended support has to cover two layers, because a ZooKeeper-mode cluster runs both and they end on different dates. Upstream gives you no fixes for either on the lines most clusters run today.

ComponentUpstream todayWith OSSeva
Brokers, Connect and clientsKafka 3.9 and 4.0 are archived; 3.8 and earlier have ended. Supported lines are 4.1 to 4.3, KRaft only.Signed patched builds for 2.8 to 3.9. Critical CVEs within 48 hours, High within 7 days.
ZooKeeper ensembleOnly ZooKeeper 3.8 and 3.9 receive fixes. 3.4 to 3.7 have ended.Patched 3.4, 3.6 and 3.7 builds, 3.5 as custom support. See the ZooKeeper page.
Evidence for auditorsAdvisories list the lines they assessed, and older lines are often not assessed.A written assessment per advisory against your configuration, with VEX.

Move to KRaft: steps and risks

Apache documents the migration as phases. Brokers are prepared on 3.9, a KRaft controller quorum is provisioned, the brokers enter migration mode, are restarted in KRaft mode one at a time, and the controllers then leave migration mode to finalise. The cluster can still revert to ZooKeeper mode before that last step and cannot after it.

What sets the scope is the number of clusters, topics and partitions, the client and tooling inventory, how many log directories each broker uses, and the change windows you have. The readiness checklist above lists the versions and settings that break.

Phases per Apache Kafka 3.9 documentation, ZooKeeper to KRaft migration.

PhaseWhat happensRevert to ZooKeeper?
PrepareBrokers on 3.9.1 or later with the 3.9 metadata version; clients and tooling inventoried.Nothing has changed yet
Provision controllersA KRaft controller quorum is deployed with migration enabled.Yes
Migration modeBrokers get migration settings; metadata loads from ZooKeeper into the quorum.Yes, with the documented steps
Migrate brokersBrokers restart in KRaft mode one at a time, with the controller still writing to ZooKeeper.Yes, with the documented steps
FinaliseControllers leave migration mode; ZooKeeper can be decommissioned.No

ZooKeeper 3.5 custom support for Kafka 2.x

Which Kafka releases run ZooKeeper 3.5? Kafka 2.4 through 2.8 bundle ZooKeeper 3.5.6 to 3.5.9, and many teams run the bundled copy as their ensemble. Kafka 3.0 and later moved to 3.6.3 or newer, so a 3.5 ensemble under Kafka usually means the brokers are on 2.x as well.

ZooKeeper 3.5 reached end of life on 1 June 2022, and 3.5.10 was its last release, so the Apache project issues no more fixes for it. OSSeva's standard ZooKeeper coverage starts at 3.6. ZooKeeper 3.5 is offered as custom support, scoped per engagement.

Custom support covers CVE backports into 3.5.x builds made from the upstream source, and SBOM and VEX evidence for each build. A scanner finding or an auditor's question then gets a written answer for your configuration: fixed, not present, or not reachable.

We scope the ZooKeeper 3.5 build together with the broker line you run, in one engagement. The way off ZooKeeper is the KRaft migration, which runs on Kafka 3.x, so a 2.x cluster has a broker upgrade to do first.

Send us the versions you run and we scope it with you. Contact us for a quote.

Releases that bundle ZooKeeper 3.5.x

ReleaseBundled ZooKeeper
Kafka 2.4.03.5.6
Kafka 2.4.1 and 2.5.03.5.7
Kafka 2.5.1, 2.6.0 and 2.7.03.5.8
Kafka 2.6.3, 2.7.2 and 2.8.0 to 2.8.23.5.9

Source: gradle/dependencies.gradle at each Kafka release tag (github.com/apache/kafka), checked 15 October 2026.

Which vendors provide Kafka security patches while you stay on ZooKeeper?

OptionWhat you getTrade-off
Migrate to KRaft nowA supported 4.x path from the communityEvery cluster needs its own migration window, and the finalize step has no rollback.
Confluent Platform support tiersConfluent Platform 7.x covered to its end-of-support date, with Platinum a year past StandardLimited to Confluent Platform deployments and the Confluent calendar; 7.9 Platinum ends on 19 February 2028.
Amazon MSKAWS runs the brokers and sets the version calendarA migration of data and clients, and only inside AWS.
OSSeva extended supportPatched brokers and ZooKeeper, plus the migrationA subscription per cluster while you migrate.
Stay unpatchedNothingBroker and ZooKeeper advisories stay open on the system that carries your event data.

Dates checked 20 October 2026 on kafka.apache.org (downloads, KRaft operations and upgrade documentation), endoflife.date and the Confluent Platform versions and interoperability documentation. CVE details from the CVE directory and the MITRE records.

Questions

Frequently asked questions

Book a 15-minute call

Can I upgrade Kafka 3.x on ZooKeeper directly to 4.0?

No. Kafka 4.0 has no ZooKeeper support. Migrate to KRaft on 3.x first, ideally 3.9, finalize the migration, then upgrade to 4.x.

Is Kafka 3.9 still supported?

Not by the Apache project. 3.9.2, released on 21 February 2026, was the final release and the downloads page lists 3.9 as archived. endoflife.date still shows no end date for it. Checked 20 October 2026.

Who provides extended support for Kafka clusters still on ZooKeeper?

Confluent covers Confluent Platform 7.x until its end-of-support dates. OSSeva patches community Kafka 3.x brokers and their ZooKeeper ensembles, whether or not they run under Confluent, and helps with the KRaft migration.

Do you patch ZooKeeper as well as Kafka?

Yes. The ensemble behind an old Kafka cluster is usually older than the brokers and carries its own advisories.

Can a KRaft migration be rolled back?

Until the controllers leave migration mode, Apache documents a way back to ZooKeeper mode. After that final step there is no supported way back, which is why the migration is rehearsed on a representative cluster first.

What decides how long a KRaft migration takes?

The number of clusters, topics and partitions, the client and tooling inventory, brokers with several log directories, and the change windows you can get. Extended support covers the clusters that are waiting.

Is Apache Kafka still relevant?

Yes. Apache releases a new minor roughly every four months and lists 4.1, 4.2 and 4.3 as supported. What has ended is ZooKeeper mode, not Kafka.

What is Amazon MSK and what does it do?

Amazon MSK is AWS's managed Apache Kafka service: AWS runs the brokers and sets which Kafka versions are available and until when. It suits teams that want to hand operations to AWS, and it does not help clusters that run on your own servers or another cloud.

Get a quote

Patch the clusters that are waiting for KRaft.

Send your work email and the Kafka version you run. An engineer replies within 1 business day; support is priced per cluster.

Only your email is required · An engineer replies within 1 business day · No commitment

Prefer to talk or read more first?

Book a call