Hwm_breached with respect to high-water-memory-pct

Operations Monitoring
1

I’m checking the metrics hwm_breached with memory_free_pct as I hitting the memory by benchmark java-client.

Based on the default parameter of high-water-memory-pct, which is 60, I’m expecting the hwm_breached becomes true as soon as the memory_free_pct reaches 40%. However, it looks depending on how fast your memory consumption goes. If it’s fast (e.g., more than 2000 tps), the hwm_breached comes to true way late. Wondering about the mechanism under the hood about this.

2

Breaches are detected and declared by the namespace supervisor thread which by default runs every 2 minutes.

Read this thread on defrag Defrag not keeping up - #4 by pgupta to understand more details on tuning nsup-period.