Support KRaft for consensus building #889

maltesander · 2025-09-04T12:46:53Z

Description

This PR add a new role controller to configure KRaft instead of ZooKeeper. This is prep work for Kafka 4.0, where ZooKeeper is removed.

fixes #690

BREAKING Changes:

renamed server.properties to broker.properties and controller.properties respectively (this affects e.g. config overrides)

What was done?

CRD:
- make zookeeperConfigmapName optional
- split KafkaConfig into Broker and Controller config
- split out common configs (e.g. resources)
- added controller role
- split roles to own module
Product config:
- clean out properties
- remove zookeeper.connect, zookeeper.timeout (never used or overwritten in zk mode)
- add controller to product config machinery
Implementation
- Added sanity check to have Kafka4.0 / Controller and zookeeper mutually exclusive
- introduce AnyConfig to better handle different config types
- rework container commands and split to own module
  - KRaft formatting
  - node.id handling
- rework resources (statefulset, configmap, service) and split to own module
- increased default broker memory to 2gb
- removed obsolete errors
- better constants and config handling
- hashing for "random" but deterministic integers for node.id
- cluster-id is set to KafkaCluster metadata.uid
Tests:
- Added use-kraft-controller dimension in almost all tests to split between Kraft / zk

What was not done?

(Stackable) Listener
Discovery
Actual Kafka 4.0 support

TODO

3.7.2 no dynamic quorum (bad for scaling) https://developers.redhat.com/articles/2024/11/27/dynamic-kafka-controller-quorum?utm_source=chatgpt.com#implementation -> scaling flaky
better liveness / readiness for controller (e.g. check if quorum joinend?)
zk-kraft migration
antiaffinities controller / broker?
PDBs controller?
Finish docs

Definition of Done Checklist

Not all of these items are applicable to all PRs, the author should update this template to only leave the boxes in that are relevant
Please make sure all these things are done and tick the boxes

Author

Changes are OpenShift compatible
CRD changes approved
CRD documentation for all fields, following the style guide.
Helm chart can be installed and deployed operator works
Integration tests passed (for non trivial changes)
Changes need to be "offline" compatible
Links to generated (nightly) docs added
Release note snippet added

Reviewer

Code contains useful comments
Code contains useful logging statements
(Integration-)Test cases added
Documentation added or updated. Follows the style guide.
Changelog updated
Cargo.toml only contains references to git tags (not specific commits or branches)

Acceptance

Feature Tracker has been updated
Proper release label has been added
Links to generated (nightly) docs added
Release note snippet added
Add type/deprecation label & add to the deprecation schedule
Add type/experimental label & add to the experimental features tracker

razvan

I did a first round of reading. Have some comments/suggestions.

Testing a bit more now.

docs/modules/kafka/pages/usage-guide/kraft-controller.adoc

razvan · 2025-09-11T10:04:46Z

rust/operator-binary/src/crd/mod.rs

+    /// Supporting Kraft alongside Zookeeper requires a couple of CRD checks
+    /// - If Kafka 4 and higher is used, no zookeeper config map ref has to be provided
+    /// - Configuring the controller role means no zookeeper config map ref has to be provided
+    pub fn check_kraft_vs_zookeeper(&self, product_version: &str) -> Result<(), Error> {


Instead of this implicit decision a dedicated field could be used that explicitly puts the cluster in Zookeeper or Kraft mode.

It is possible to create a Kafka cluster in Kraft mode without any spec.controllers property.

Being explicit about the metadata manager/handler will be needed later to support migrations when both Zookeeper and controllers must be running at the same time.

My suggestion for the field would be kraft_mode: [OFF|COMBINED|MIGRATING|ON]

This PR can ignore COMBINED and MIGRATING.

rust/operator-binary/src/crd/mod.rs

razvan · 2025-09-11T13:26:24Z

tests/test-definition.yaml

@@ -62,17 +62,21 @@ dimensions:
      - "cluster-internal"
      - "external-stable"
      - "external-unstable"
+  - name: use-kraft-controller


As discussed on Slack, this is too long. Maybe rename to just kraft similarly to openshift.

Now using zookeeper and zookeeper-latest dimension for kraft with value "false".

Co-authored-by: Razvan-Daniel Mihai <[email protected]>

maltesander added 30 commits August 25, 2025 10:56

extract common properties to role config

023afa5

rename KafkaConfig to BrokerConfig

0a21d44

rename Container to BrokerContainer

cd6e03b

make zookeeper config map ref optional

2a27fbc

remove zookeeper.connect and timeout from product config

f29baa1

add controller role and move configs in own module

2a12e3f

fix tests

2139393

regenerated charts

f58e01d

use broker.properties and controller.properties

0b34d01

replace missing server.properties

2d49b45

remove unused errors

baae4ce

add controller to product confif machinery

f74c255

remove cp

8c8e4fc

add controller to security.properties

b01b931

add controller to security properties

8e95e72

wip - add AnyConfig to access different role configs

a87c3f2

remove errors from product logging

5d1f9b1

finished controller prep work

2fb197e

remove clones

0296ede

remove obsolete methods

a44c056

cleanup

ab61dd1

use role methods

25b95e8

remove obsolete errors

50d4a43

remove unwrap

c712ff7

wip - kafka controller cleanup, move resources to own module

f646b1e

wip - controller working

1a0b5d4

wip - combine broker & controller

3a42eca

fix todos

f324409

zookeeper working again

934f6e1

fix todos

03bcb41

maltesander added 5 commits September 5, 2025 16:04

fix docs

e5f366b

run pre commit

d654a97

fix yaml lint

53121de

clarify docs, clean up

cfac3fc

linter

6054d1c

maltesander marked this pull request as ready for review September 8, 2025 06:38

maltesander added 6 commits September 8, 2025 17:42

improve jaas config

0e6244c

add comments, remove todo

d413bb8

remove redundant memory increase

91659bb

implement node.id hashing check for all roles

43a1759

fix linter

d5ee2fd

prepare log4j2 config

87809b5

sbernauer requested a review from razvan September 11, 2025 07:16

maltesander added 2 commits September 11, 2025 14:37

consolidate cluster operation tests

34f6c67

consolidate delete rolegroup

98307ed

razvan reviewed Sep 11, 2025

View reviewed changes

maltesander and others added 9 commits September 11, 2025 17:32

adjust kerberos tests

aca7a24

fix logging tests

14eb096

fix smoke test

3e49961

fix tls tests

c91ce20

fix upgrade tests

604de9f

Apply suggestions from code review

1ed9489

Co-authored-by: Razvan-Daniel Mihai <[email protected]>

do not apply znodes in kraftmode

3decd23

use cluster name instead of uid for kafka cluster-id

5e2a59e

replace uid with name for cluster-id

4f387ef

Techassi self-requested a review September 15, 2025 07:21

razvan and others added 4 commits September 16, 2025 17:27

Merge branch 'main' into feat/kafka-4.0

a8b71fe

add pre-stop sleep hook to controller kafka container

079fe5a

add "kraft" test suite

cceb2be

add experimental warning to kraft docs

c1d1dfa

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Support KRaft for consensus building #889

Support KRaft for consensus building #889

Uh oh!

maltesander commented Sep 4, 2025 •

edited

Loading

Uh oh!

razvan left a comment

Uh oh!

Uh oh!

Uh oh!

razvan Sep 11, 2025

Uh oh!

Uh oh!

razvan Sep 11, 2025

Uh oh!

maltesander Sep 12, 2025

Uh oh!

Uh oh!

Uh oh!

Support KRaft for consensus building #889

Are you sure you want to change the base?

Support KRaft for consensus building #889

Uh oh!

Conversation

maltesander commented Sep 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

BREAKING Changes:

What was done?

What was not done?

TODO

Definition of Done Checklist

Author

Reviewer

Acceptance

Uh oh!

razvan left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

razvan Sep 11, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

razvan Sep 11, 2025

Choose a reason for hiding this comment

Uh oh!

maltesander Sep 12, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

maltesander commented Sep 4, 2025 •

edited

Loading