On Wed, 27 Jun 2012 14:03:55 +0800 majianpeng [off-list ref] wrote:
On 2012-06-27 12:32 NeilBrown [off-list ref] Wrote:
quoted
On Sat, 26 May 2012 10:52:50 +0800 "majianpeng" [off-list ref] wrote:
quoted
When RAID-4,5,6 degraded and met read-error, it will eject the rdev.And then
the RAID will fail and lost data.Because the function of set-badsector,when
this occur,it will set-badsector,not ejecting the rdev.
When RAID-4,5,6 met read-error, it will re-write if RAID was not degrade.But if
re-write error,it will eject the rdev and RAID will degrade and it will take too
long time for recoverying.So I add judgement for controling how may re-write-error
can eject the rdev.
I do those for flexible controling the read-error for different situation.
Thanks.
quoted
majianpeng (2):
md/raid456: When readed error and raid was degraded,it try to
set badsector, not ejecting the rdev.
I've applied this one. I also added 'set_bad = 1' in the case where
the re-write failed.
quoted
md/raid456:Add interface for contorling eject rdev when re-write
failed.
I haven't applied this. I'm not entirely sure what the point of counting
the errors was, but I don't think it is necessary.
Using raid456,the first object is to protect data.But in some situation, the user
can endure lost some data instead of raid degraed or failed.
After introduce the badblocks, I think md-driver should do flexible controling for
error.The controling can control by different user for different requirment.
I cannot see the point of that control though.
Sure you *always* want to record a bad block if possible, if the alternative
is ejecting the whole device?
I don't see where the choice would be between "lost data" or "degraded array".
Maybe if the failing device caused large delays then you want to eject it
soon rather than struggling on with it. However my belief is that if you
don't want long delays, then you should tell the device to fail rather than
impose long delays. It is not something that md should care about.
So: still a little confused.
NeilBrown