On 28/02/2013 11:31, Lev Serebryakov wrote: > WD disks are in software RAID5 with geom_raid5 (from ports, but I'm > active maintainer of it). > Disks are in "Default" configuration: WC and NCQ are enabled. > > I know, that FS guys could blame geom_raid5, as it could delay real > write up to 15 seconds, but it never "lies" about writes (it doesn't > mark BIOs complete till they are really sent to disk) and I could > not reproduce any problems with it on many hours tests on VMs (and I > don't want to experiment a lot on real hardware, as it contains my > real data). > > Maybe, it is subtile interference between raid5 implementation and > SU+J, but in such case I want to understand what does raid5 do > wrong. You guessed correctly, I was going to blame geom_raid5 :) Is this a production setup you have? Can you afford to destroy it and re-create it for the purpose of testing, this time with geom_raid3 (which should be synchronous with respect to writes)?
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:35 UTC