Apache Kafka

Network Processor Idle Drop Precedes Request Timeouts

critical
Resource ContentionUpdated Dec 20, 2025

When network processor average idle percentage drops significantly, it indicates the broker is overwhelmed processing network I/O, leading to client-side request timeouts before requests reach handler threads.

How to detect:

Monitor kafka.network.SocketServer.NetworkProcessorAvgIdlePercent or kafka.net.processor.avg.idle.pct.rate dropping below 0.3 (30% idle). Cross-reference with kafka.request.produce.failed.rate and kafka.request.fetch.failed.rate to confirm client-side failures.

Recommended action:

Increase num.network.threads to add network processor capacity. Verify network bandwidth is not saturated. Review client timeout configurations - increase request.timeout.ms if appropriate. Scale broker count horizontally to distribute network load. Check for inefficient client configurations causing excessive small requests.