On Dec 1, 2009, at 9:51 AM, Farkas Levente lfarkas@lfarkas.org wrote:
On 12/01/2009 02:36 PM, Thomas Harold wrote:
So my take on all of that is, if you see it on RAID-5 or RAID-6, you should worry. But if it's on an array with memory mapped files or swap files/partitions that is RAID-1 or RAID-10, it's less of a worry.
but then do we (the /etc/cron.weekly/99-raid-check) need to rebuild all of my terrabyte raid1 arrays at all weekend? if not than imho it's a bug:-(
I agree, I think the real problem is that the Linux MD RAID doesn't quiesce the arrays when checking the mismatch_cnt so for RAID1/10 arrays you will see the transactions committed to one side, but not yet on the other. For RAID5/6 the whole stripe with parity must be committed atomically, so there should never be a mismatch_cnt.
There should be a way to specify arrays to be skipped during the check until the real problem of quiescing the arrays is fixed. Or make the RAID1/10 write transactions atomic like the RAID5/6 (which they should be in my opinion, but lowers write performance).
Maybe touch a file with the name of the array to skip in a particular directory somewhere and have the script check that directory for arrays to skip?
-Ross