Apache ZooKeeper

Disk Fsync Latency Blocks Transaction Processing

critical
latencyUpdated Feb 6, 2026

ZooKeeper writes transaction logs synchronously to disk using fsync. High fsync times (>100ms average) directly block request processing and cause cascading latency issues across the entire ensemble.

How to detect:

Monitor average fsync time (from mntr command or JMX). Alert when average fsync time exceeds 100ms over 1 minute, or when max fsync time exceeds 500ms. This indicates disk I/O bottleneck.

Recommended action:

Use hdparm to verify disk performance meets requirements. Ensure transaction logs are on dedicated SSD storage separate from snapshots. Check for other processes competing for disk I/O. Consider NVMe drives for transaction logs in high-throughput environments. Verify write-back cache is enabled on RAID controllers.