브로커 구성 및 튜닝

Optimizing your Kafka brokers is crucial for a stable and high-performing cluster. Just like tuning an engine, proper configuration ensures your Kafka cluster runs smoothly and efficiently.

• Performance: Handle higher data throughput.
• Stability: Prevent crashes and ensure continuous operation.
• Resource Utilization: Make the most of your server's CPU, memory, and disk.

We'll explore key settings to achieve this.

Each Kafka broker has a main configuration file, typically named server.properties. This file dictates how a broker behaves, from network settings to data storage.

You'll find this file in the Kafka installation directory, usually under config/. Any changes require a broker restart to take effect.

Every broker in a Kafka cluster needs a unique identifier. This is set using the broker.id parameter in server.properties.

It must be a non-negative integer and unique across all brokers in the cluster. Kafka uses this ID to identify brokers for replication, leader election, and more.

The log.dirs parameter specifies where Kafka stores its log segments (the actual data for topics and partitions). This is a critical setting for performance and reliability.

• Multiple Disks: Use a comma-separated list of directories on different physical disks for better I/O parallelism.
• Dedicated Disks: Ideally, use disks dedicated solely to Kafka logs, separate from the operating system or other applications.

Kafka brokers communicate via network listeners. The listeners and advertised.listeners parameters define how clients and other brokers connect.

• listeners: The interfaces the broker binds to (e.g., PLAINTEXT://:9092).
• advertised.listeners: The address clients/brokers use to connect (e.g., PLAINTEXT://your.host.name:9092). This is crucial for external access or multi-host setups.

When a topic is created without explicit partition or replication settings, Kafka uses broker-level defaults. These are configured via num.partitions and default.replication.factor.

• num.partitions: Sets the default number of partitions for new topics. More partitions mean higher parallelism.
• default.replication.factor: Sets the default number of replicas for new topics. Higher replication means better fault tolerance.

It's generally recommended to set these explicitly per topic, but these defaults act as a fallback.

Kafka retains messages for a configurable period or until they reach a certain size. These settings prevent your disks from filling up and are crucial for managing storage.

• log.retention.hours: How long messages are kept (e.g., 168 hours = 7 days).
• log.retention.bytes: Maximum size of a log segment before it's eligible for deletion.

Kafka will delete messages based on whichever limit is reached first.

To prevent excessively large messages from impacting broker performance or causing network issues, Kafka allows you to set a maximum message size at the broker level via message.max.bytes.

This limit applies to the total size of a compressed message batch. If a producer tries to send a message larger than this, it will be rejected. Producers also have their own max.request.size.

Beyond server.properties, the underlying Java Virtual Machine (JVM) and operating system (OS) also need tuning for optimal Kafka performance.

• JVM Heap Size: Configure KAFKA_HEAP_OPTS to allocate sufficient memory (e.g., 5-8GB for dedicated brokers).
• Garbage Collection: Choose an efficient GC algorithm (e.g., G1GC) and tune its parameters.
• File Descriptors: Increase OS limits for open file descriptors, as Kafka uses many for logs and connections.

We've covered essential Kafka broker configurations that impact performance, stability, and resource usage.

• The server.properties file is central to a broker's behavior.
• Key parameters like broker.id, log.dirs, listeners, log.retention.hours, and message.max.bytes are vital for proper setup.
• Remember to also consider JVM and OS-level tuning for a truly optimized cluster.

Careful tuning ensures your Kafka cluster can handle your data streams efficiently and reliably.