osd: scrub stat mismatch
on thrash + radosmodel workload. bytes match, but kb don't:
2012-01-25 09:33:22.502530 7fa14f9f7700 osd.3 167 pg[0.15( v 122'51 (0'0,122'51] n=15 ec=1 les/c 165/167 165/165/165) [3,5] r=0 lpr=165 mlcod 122'51 active+clean+scrubbing snaptrimq=[1~5c,5e~2,61~6,68~a,73~1,76~2]] scrub got 15/15 objects, 0/0 clones, 30592329/30592329 bytes, 29883/29884 kb.
maybe the snap trimmer?
#1 Updated by Sage Weil almost 12 years ago
I suspect the problem is due to the rounding off when we are doing the complicated dance of keeping these stats correct wrt the clone_overlap, esp during the trimming.
I wonder if this is a good time to just drop num_kb... num_bytes is probably sufficient? Any opinions?