On Wednesday, June 08, 2011 09:00:48 PM mcclnx mcc wrote:
We have DELL server with MD1000 Disk array in it. O.S. is CENTOS 5.5. Recently every time MD1000 "patrol read" start I will get "media error" messages on /var/log/message file.
I use MD1000 "slow initialize" to initialize "bad disk" and NO error. After "slow initialize" finish, I manually "startup patrol read". I continue get "media error" on /var/log/message.
Anyone know what difference between "slow initialize" and "patrol read"?
Only Dell really knows what they mean by it.
But here are some random thoughts:
- Slow init may write all of the disk during init but maybe:
- no errors happen during write
- it notices an error but retries silently
- it notices an error but has sucky error reporting
Yes, both can be true. I guess it's simply because a disk which writes on a bad sector/block just silently remaps it and goes on because it's considered fixed.
- Patrolread (may) compare parity data and report this as "media error" (a type of error slow init probably wouldn't detect)
- The I/O pattern of patrolread compared to slow init provoks an error
I always tought patrolread just tries to read the whole disk to detect sleeping bad sectors/blocks. Maybe it also checks consistency on the RAID level but I'm not sure.
Simon