Apache Kafka

Request Handler Saturation Cascades to Producer Latency

warning
Resource ContentionUpdated Dec 20, 2025

When Kafka request handlers are saturated (low idle percentage), producer requests queue up, increasing end-to-end latency and potentially triggering producer-side timeouts.

How to detect:

Monitor kafka.request.handler.avg.idle.pct.rate dropping below 0.2 (20% idle) combined with increasing kafka.network.RequestChannel.RequestQueueSize.Value and kafka.request.produce.time.99percentile. This indicates handler saturation affecting producer throughput.

Recommended action:

Increase num.io.threads to provide more request handler capacity. Review broker CPU allocation - if CPU is saturated, scale vertically or add brokers. Optimize producer batch size and linger.ms to reduce request frequency. Consider separating produce and fetch traffic to different broker pools if possible.