The Disk I/O Persistence Ceiling
criticalPersistent messages require disk writes before publisher confirmation, making slow disks a hard throughput ceiling. Queue mirroring multiplies this overhead across cluster nodes, saturating disk I/O and causing flow control.
Monitor disk I/O wait time (iostat), disk write latency, publisher confirm latency, and flow control events. High I/O wait (>20%) combined with disk alarms or flow control blocking indicates disk bottleneck. Mirrored queues will show multiplied write load.
Use SSDs for persistent message storage. For non-critical data, use transient messages and non-durable queues to keep data in RAM. Consider Lazy Queues for large backlogs to move unconsumed messages to disk proactively. Reduce queue mirroring to only critical queues. Increase disk space to prevent disk alarms triggering flow control.