cache metadata AND report on actual HDD perf

cache metadata in RAM so those queries are always fast
also track HDD performance and somehow report that back to locator
locator takes performance into account as well when selecting a replica to read or a disk with free space to write to

Run iostat -x 1 (or read /proc/diskstats / /sys/block/sdX/stat directly) and track await/r_await/w_await and %util per drive over a rolling window.
Every N seconds, read /sys/block/sdX/stat for each drive (cheap, no I/O generated — pure counter read) and compute rolling average await and queue depth.
daemon (Python or a small Go binary) that polls /proc/diskstats and produces a live per-drive score feed?

Navigation Menu