A performance metric measures how effectively and responsively a system performs under a workload. In Kafka operations, it turns system behavior into evidence for performance tuning, reliability work, and service-level decisions. section 11.2 section 11.2.1
Useful performance metrics include:
- Message-processing latency, which indicates responsiveness. section 11.2.1
- Incoming and outgoing byte rates, which show data-transfer behavior. section 11.2.1
- Average batch sizes and throughput per topic or consumer group, which expose processing efficiency and capacity. section 11.2.1
- CPU, memory, disk I/O, network use, buffer exhaustion, and poll duration, which provide resource and saturation context. section 11.2.1
Diagnosing performance degradation
Suppose throughput falls while message-processing latency and network I/O rise. Comparing performance metrics with resource-utilization and saturation metrics helps operations teams determine whether the system is degrading, approaching capacity, or requiring configuration and capacity changes. Alerts can then direct the issue to the operations or DevOps team for diagnosis and corrective action. section 11.2 section 11.2.1
Performance metrics are commonly exposed through JMX, Prometheus, or custom instrumentation. They complement business metrics, which measure domain outcomes such as delayed, dropped, duplicated, or unmatched messages; cloud platforms may provide infrastructure metrics automatically, leaving application teams to instrument Kafka-specific and business-level behavior. section 11.2.1 section 11.2.3
