ZooKeeper role (and KRaft replacement) in Kafka - Deep Dive

Overview - ZooKeeper role (and KRaft replacement)

What is it?

ZooKeeper is a system that helps manage and coordinate Kafka servers by keeping track of their status and configuration. It acts like a central manager that ensures all Kafka servers agree on who is the leader and how data is organized. KRaft is a newer way Kafka uses to replace ZooKeeper by handling this coordination internally without needing an extra system. This change simplifies Kafka's setup and improves its performance.

Why it matters

Without ZooKeeper or a similar system, Kafka servers would struggle to agree on important decisions like who leads data handling or how to keep data consistent. This could cause confusion, data loss, or downtime. KRaft removes the need for ZooKeeper, making Kafka easier to run and more reliable, which means smoother data streaming for applications that depend on it.

Where it fits

Before learning about ZooKeeper and KRaft, you should understand basic Kafka concepts like brokers, topics, and partitions. After this, you can explore Kafka cluster management, fault tolerance, and how Kafka ensures data consistency and availability.

Mental Model

Core Idea

ZooKeeper (or KRaft) acts as the trusted referee that helps Kafka servers agree on who leads and how data is managed to keep the system running smoothly.

Think of it like...

Imagine a group project where one person is the team leader who assigns tasks and keeps everyone on the same page. ZooKeeper is like the teacher who oversees the group, making sure everyone agrees on who the leader is and what each person should do. KRaft is like the group deciding to manage leadership and coordination themselves without needing the teacher.

┌───────────────┐       ┌───────────────┐
│   Kafka       │       │   ZooKeeper   │
│   Brokers     │◄─────►│   Cluster     │
│ (Servers)     │       │ Coordination  │
└───────────────┘       └───────────────┘
          ▲                      ▲
          │                      │
          │                      │
          ▼                      ▼
   Leader Election         Configuration
   & Metadata Sync        Management

In KRaft mode:
┌───────────────┐
│ Kafka Brokers │
│ (Self-Managed │
│  Coordination)│
└───────────────┘

Build-Up - 7 Steps

1

Foundation

What is ZooKeeper in Kafka

Concept: ZooKeeper is introduced as a separate system that Kafka uses to manage its cluster state and coordination.

Kafka brokers need to know who is the leader for each data partition and keep track of cluster membership. ZooKeeper is a distributed service that stores this metadata and helps brokers coordinate by electing leaders and sharing configuration.

Result

Kafka brokers can work together reliably because they use ZooKeeper to agree on leadership and cluster state.

Understanding ZooKeeper's role is key to grasping how Kafka maintains order and consistency across multiple servers.

2

Foundation

Why Kafka needs coordination

Concept: Kafka requires a system to manage leader election and metadata to ensure data consistency and availability.

In a Kafka cluster, each partition has one leader broker that handles all reads and writes. If a leader fails, another broker must take over quickly. Coordination ensures clients always connect to the right leader and data stays consistent.

Result

Kafka can handle failures smoothly without losing data or confusing clients.

Knowing why coordination is necessary helps appreciate the complexity behind Kafka's reliability.

3

Intermediate

How ZooKeeper manages Kafka cluster state

🤔Before reading on: do you think ZooKeeper stores actual Kafka messages or just metadata? Commit to your answer.

Concept: ZooKeeper stores metadata like broker info, topic configurations, and leader election data, but not the actual messages Kafka handles.

ZooKeeper keeps small pieces of data about the cluster's state, such as which brokers are alive, which broker leads which partition, and topic configurations. It uses a consistent protocol to ensure all brokers see the same state.

Result

Kafka brokers rely on ZooKeeper to get up-to-date cluster information and coordinate actions.

Understanding that ZooKeeper handles only metadata clarifies its role and why Kafka messages are stored separately.

4

Intermediate

Limitations of ZooKeeper in Kafka

🤔Before reading on: do you think ZooKeeper is easy to scale with Kafka clusters? Commit to your answer.

Concept: ZooKeeper adds operational complexity and can become a bottleneck or single point of failure in large Kafka deployments.

Running ZooKeeper requires managing a separate cluster, monitoring its health, and tuning it for performance. As Kafka scales, ZooKeeper's load increases, making maintenance harder and risking downtime if ZooKeeper fails.

Result

Operators face extra work and risk due to ZooKeeper dependency.

Knowing ZooKeeper's operational challenges motivates the need for a better solution.

5

Intermediate

Introduction to KRaft mode in Kafka

🤔Before reading on: do you think KRaft replaces ZooKeeper completely or works alongside it? Commit to your answer.

Concept: KRaft is Kafka's built-in consensus mechanism that replaces ZooKeeper by managing metadata and coordination internally.

KRaft uses a consensus protocol called Raft to handle leader election, metadata storage, and cluster coordination within Kafka brokers themselves. This removes the need for a separate ZooKeeper cluster.

Result

Kafka clusters become simpler to deploy and manage without ZooKeeper.

Understanding KRaft's role shows how Kafka evolves to reduce complexity and improve reliability.

6

Advanced

How KRaft manages metadata internally

🤔Before reading on: do you think KRaft stores metadata on all brokers or just a few? Commit to your answer.

Concept: KRaft stores metadata on a quorum of brokers using a replicated log to ensure consistency and fault tolerance.

KRaft brokers form a quorum that replicates metadata changes using the Raft protocol. This ensures all brokers agree on cluster state and can elect leaders without external coordination.

Result

Kafka brokers maintain consistent metadata internally, enabling faster failover and simpler architecture.

Knowing KRaft's internal replication mechanism explains how it achieves reliability without ZooKeeper.

7

Expert

Challenges and tradeoffs in KRaft design

🤔Before reading on: do you think removing ZooKeeper makes Kafka's metadata management simpler or more complex internally? Commit to your answer.

Concept: KRaft simplifies deployment but requires Kafka brokers to handle more responsibilities, increasing internal complexity and resource use.

By embedding metadata management, Kafka brokers must implement consensus protocols and handle metadata storage, which adds complexity. However, this reduces external dependencies and improves scalability. The design balances operational simplicity with internal complexity.

Result

Kafka gains easier operations but must carefully manage internal coordination to avoid new failure modes.

Understanding these tradeoffs helps experts design and troubleshoot Kafka clusters effectively.

Under the Hood

ZooKeeper works as a separate distributed system that stores small pieces of metadata in a hierarchical namespace. It uses a consensus protocol called Zab to ensure all nodes agree on updates. Kafka brokers connect to ZooKeeper to read and write metadata, perform leader elections, and watch for changes. KRaft replaces ZooKeeper by embedding a Raft consensus group inside Kafka brokers. This group replicates metadata logs and manages leader election internally, removing the need for an external system.

Why designed this way?

ZooKeeper was chosen originally because it was a proven, reliable coordination service that could be reused across many systems. However, running a separate ZooKeeper cluster added operational overhead. Kafka's designers created KRaft to simplify architecture by integrating coordination directly, reducing dependencies and improving scalability. Raft was chosen for its understandability and strong consistency guarantees.

ZooKeeper Mode:
┌───────────────┐       ┌───────────────┐       ┌───────────────┐
│ Kafka Broker 1│◄─────►│               │◄─────►│ Kafka Broker 2│
│               │       │  ZooKeeper    │       │               │
│               │       │  Ensemble     │       │               │
└───────────────┘       └───────────────┘       └───────────────┘

KRaft Mode:
┌───────────────┐       ┌───────────────┐       ┌───────────────┐
│ Kafka Broker 1│◄─────►│ Kafka Broker 2│◄─────►│ Kafka Broker 3│
│ (KRaft Node)  │       │ (KRaft Node)  │       │ (KRaft Node)  │
└───────────────┘       └───────────────┘       └───────────────┘

All brokers participate in metadata consensus internally.

Myth Busters - 4 Common Misconceptions

Quick: Does ZooKeeper store Kafka's actual message data? Commit to yes or no.

Common Belief: ZooKeeper stores all Kafka data including messages.

Reality: ZooKeeper only stores metadata like cluster state and configuration, not the actual message data.

Why it matters: Believing ZooKeeper stores messages can lead to confusion about Kafka's storage and performance characteristics.

Quick: Is KRaft just an add-on that works alongside ZooKeeper? Commit to yes or no.

Common Belief: KRaft works together with ZooKeeper to manage Kafka clusters.

Reality: KRaft fully replaces ZooKeeper by handling metadata and coordination internally within Kafka brokers.

Why it matters: Misunderstanding this can cause incorrect cluster setups and operational mistakes.

Quick: Does removing ZooKeeper make Kafka's internal coordination simpler? Commit to yes or no.

Common Belief: Removing ZooKeeper simplifies Kafka's internal architecture completely.

Reality: Removing ZooKeeper simplifies deployment but adds complexity inside Kafka brokers, which now handle consensus and metadata storage themselves.

Why it matters: Ignoring this can lead to underestimating resource needs and troubleshooting challenges.

Quick: Can ZooKeeper scale infinitely with Kafka clusters without issues? Commit to yes or no.

Common Belief: ZooKeeper scales easily and is not a bottleneck for Kafka clusters.

Reality: ZooKeeper can become a bottleneck and operational challenge as Kafka clusters grow large.

Why it matters: Overlooking this can cause unexpected downtime and maintenance headaches.

Expert Zone

1

KRaft requires a quorum of brokers to be available for metadata operations, so cluster availability depends on quorum health.

2

ZooKeeper's separate cluster allows independent scaling and tuning, which can be advantageous in some complex deployments.

3

KRaft's metadata log is append-only and immutable, enabling easier recovery and auditability compared to ZooKeeper's znode state.

When NOT to use

KRaft is not suitable for Kafka versions before 3.3 or for clusters requiring legacy ZooKeeper features. In very large or complex environments, some operators may prefer ZooKeeper for independent scaling and mature tooling.

Production Patterns

Many production Kafka clusters now run in KRaft mode to reduce operational overhead. Operators use multi-node KRaft quorum setups for fault tolerance and monitor metadata logs closely. Migration from ZooKeeper to KRaft is planned carefully to avoid downtime.

Connections

Distributed Consensus Algorithms

KRaft uses the Raft consensus algorithm, a core distributed consensus method.

Understanding Raft helps grasp how Kafka achieves reliable coordination without external systems.

Leader Election in Distributed Systems

ZooKeeper and KRaft both perform leader election to decide which node manages data partitions.

Knowing leader election principles clarifies how Kafka maintains availability and consistency.

Project Management Coordination

ZooKeeper's role is like a project manager coordinating team members to avoid conflicts.

Seeing coordination as a human process helps understand why distributed systems need consensus.

Common Pitfalls

#1 Trying to run Kafka without ZooKeeper or KRaft enabled.

Wrong approach:

bin/kafka-server-start.sh config/server.properties
# No KRaft mode enabled and no ZooKeeper configured

Correct approach:

bin/kafka-server-start.sh config/kraft/server.properties
# KRaft mode enabled with proper configuration

Root cause:

Misunderstanding that Kafka requires either ZooKeeper or KRaft for cluster coordination.

#2 Assuming ZooKeeper stores Kafka messages leading to wrong backup strategies.

Wrong approach:

Backing up only ZooKeeper data to recover Kafka messages.

Correct approach:

Backing up Kafka log directories where actual messages are stored.

Root cause:

Confusing metadata storage with message storage.

#3 Mixing ZooKeeper and KRaft configurations in the same Kafka cluster.

Wrong approach:

Configuring some brokers with ZooKeeper and others with KRaft in the same cluster.

Correct approach:

Using either ZooKeeper mode or KRaft mode consistently across all brokers.

Root cause:

Not understanding that ZooKeeper and KRaft modes are mutually exclusive.

Key Takeaways

ZooKeeper is a separate system Kafka originally used to manage cluster metadata and coordination.

Kafka needs coordination to elect leaders and keep cluster state consistent for reliable data streaming.

KRaft is Kafka's built-in replacement for ZooKeeper that simplifies deployment by handling coordination internally.

KRaft uses the Raft consensus algorithm to replicate metadata across brokers without external dependencies.

Understanding the tradeoffs between ZooKeeper and KRaft helps operate Kafka clusters effectively and plan migrations.

Practice

1. What is the main role of ZooKeeper in a Kafka cluster?

easy

A. To store Kafka messages permanently

B. To replace Kafka brokers in the cluster

C. To serve as a Kafka client for producers

D. To manage cluster coordination and metadata

Solution

1. Step 1: Understand ZooKeeper's purpose

   ZooKeeper is used to manage coordination tasks like leader election and metadata storage in Kafka clusters.

2. Step 2: Identify what ZooKeeper does not do

   ZooKeeper does not store Kafka messages or act as a client or broker replacement.

3. Final Answer:

   To manage cluster coordination and metadata -> Option D

4. Quick Check:

   ZooKeeper = cluster coordination [OK]

Hint: ZooKeeper manages coordination, not message storage [OK]

Common Mistakes:

• Thinking ZooKeeper stores Kafka messages
• Confusing ZooKeeper with Kafka brokers
• Assuming ZooKeeper acts as a client

2. Which of the following is the correct way to enable KRaft mode in Kafka's configuration?

easy

A. set kafka.broker.only=true

B. set kafka.zk.enabled=true

C. set process.roles=controller,broker

D. set zookeeper.connect=localhost:2181

Solution

1. Step 1: Identify KRaft configuration property

   KRaft mode requires setting the process.roles property to include controller and broker roles.

2. Step 2: Eliminate incorrect options

   ZooKeeper-related settings like kafka.zk.enabled=true and zookeeper.connect=localhost:2181 are not for KRaft. kafka.broker.only=true is not a valid Kafka config.

3. Final Answer:

   set process.roles=controller,broker -> Option C

4. Quick Check:

   process.roles = controller,broker [OK]

Hint: KRaft uses process.roles, not zookeeper.connect [OK]

Common Mistakes:

• Confusing ZooKeeper settings with KRaft
• Using deprecated or invalid config keys
• Assuming KRaft disables brokers

3. Given the following Kafka broker configuration snippet in KRaft mode:

process.roles=broker,controller
node.id=1
controller.quorum.voters=1@localhost:9093

What is the role of this broker?

medium

A. It acts as both broker and controller

B. It acts only as a controller, not a broker

C. It acts neither as broker nor controller

D. It acts only as a broker, not a controller

Solution

1. Step 1: Analyze process.roles value

   The value 'broker,controller' means this node acts as both broker and controller.

2. Step 2: Confirm node.id and quorum voters

   node.id=1 and controller.quorum.voters=1@localhost:9093 confirm this node is part of the controller quorum.

3. Final Answer:

   It acts as both broker and controller -> Option A

4. Quick Check:

   process.roles includes both roles [OK]

Hint: process.roles lists all roles this node performs [OK]

Common Mistakes:

• Ignoring multiple roles in process.roles
• Confusing node.id with role type
• Assuming controller.quorum.voters disables broker role

4. You try to start a Kafka cluster in KRaft mode but get an error about missing controller quorum voters. Which configuration is missing or incorrect?

medium

A. Incorrect 'process.roles' without 'broker'

B. Missing or incorrect 'controller.quorum.voters' setting

C. Missing 'zookeeper.connect' setting

D. Missing 'log.dirs' setting

Solution

1. Step 1: Understand error context

   KRaft mode requires 'controller.quorum.voters' to define controller nodes.

2. Step 2: Identify missing or wrong config

   If this setting is missing or wrong, Kafka cannot form the controller quorum, causing startup errors.

3. Final Answer:

   Missing or incorrect 'controller.quorum.voters' setting -> Option B

4. Quick Check:

   controller.quorum.voters required in KRaft [OK]

Hint: KRaft needs controller.quorum.voters set [OK]

Common Mistakes:

• Confusing ZooKeeper settings with KRaft
• Ignoring controller quorum config
• Assuming log.dirs causes quorum errors

5. You want to migrate a Kafka cluster from ZooKeeper mode to KRaft mode. Which of the following steps is essential for a successful migration?

hard

A. Configure 'process.roles' and 'controller.quorum.voters' and migrate metadata to KRaft

B. Remove all brokers and restart ZooKeeper only

C. Disable KRaft and keep using ZooKeeper indefinitely

D. Set 'zookeeper.connect' to empty string and restart brokers

Solution

1. Step 1: Understand migration requirements

   Moving to KRaft requires configuring roles and controller quorum voters and migrating metadata from ZooKeeper.

2. Step 2: Eliminate incorrect options

   The other options do not perform a proper migration to KRaft or incorrectly suggest keeping or modifying ZooKeeper.

3. Final Answer:

   Configure 'process.roles' and 'controller.quorum.voters' and migrate metadata to KRaft -> Option A

4. Quick Check:

   Migration needs roles, quorum, and metadata move [OK]

Hint: Set roles, quorum, migrate metadata for KRaft migration [OK]

Common Mistakes:

• Skipping metadata migration step
• Trying to run ZooKeeper and KRaft simultaneously
• Misconfiguring controller quorum voters