#!/usr/bin/env bash # ========================================================================= # #149 -- 817 -> 857 stall regression (Oracle: 97 blocks >50ms, all in the # mission-load window, maxDraw up to 518ms; zero such blocks on 817, same # machine). My filed suspect was the #141 segment-cache sweep -- but that is # a HYPOTHESIS, and #137 just taught us what plausible-but-unmeasured # theories are worth. So: measure first. # # INSTRUMENTATION (this build): # [segperf] calls= dirty= ms=