Hi,
here is my story about "data corruption":
I have the following setup:
nvidia nforce 4 -> 4x sata hd -> swraid5 -> luks (aes) -> xfs
all together about 650gb.
The box is running mldonkey 24/7 so that keeps the drives "busy".
After about 7+-1 days of uptime one disk drops out of the raid with, i
think, a read error/timeout. So far no problem, but after about another 30
minutes a second disk errors out and kills the raid. Note: the disks are OK.
After I shutdown the box for a few minutes, I force the raid back up. It
does a rebuild, xfs_repair repairs the fs (and sometimes moves files to
lost+found, that i have not touched for months) and everything is fine. I
never noticed any dataloss, but the fs contains a couple 100k files, so I
wouldn't notice unless they are in frequent use.
Here comes the funniest thing: I didn't change anything, except writing a
script, that shuts down the raid when a disk errors out, so that the raid
remains degraided, but not inconsistent. This was 24 days ago, and this is
the uptime of the raid atm. I wonder if writing the script fixed the
problem...
However, here is, what I figured, _might_ have happend: Maybe the disks turn
off their motor about once a week (I read some time ago, that ide drives
tend to do this (note: the same problem existed before, when i still used
ide drives)). When a request comes in, it timeouts, because the motor is
still down and the disk is marked defective.
Now my question: is this a possible explanation?
Another strange behavior is: Sometimes it takes very long to delete (mainly
large, > 1-10mb) files. It takes up to 30 seconds per file (but when it
happens, most of the time 5-10 seconds). During this time the deleting
process is absolutely positively unkillable, which leads to the assumption
that is hangs inside a kernel function call.
Small files have never caused this problem.
Does anyone have similar experiences?
chris.
---------------------------------------------------------------------
- http://www.saout.de/misc/dm-crypt/
To unsubscribe, e-mail: dm-crypt-unsubscribe@xxxxxxxx
For additional commands, e-mail: dm-crypt-help@xxxxxxxx
