[OpenIndiana-discuss] ZFS single drive CKSUM errors

[Gregory Youngblood]

One of the benefits of zfs is the protection against bit rot. If the single drive had a minor issue then zfs could have detected that and corrected it when you ran the scrub.  This is in addition to the protection from multiple spindles in a mirror or raids configuration. 

It could be a drive issue,  or it may be the drive is not responding fast enough when it is reallocating sectors internally. Is this a desktop drive or a so called raid ready/enterprise drive? 

The raid drives have a set time out that guarantees response within a given time period when reallocating.  If it can't reallocate in the time alotted it returns error and allowing raid subsystem to deal with the error,  or stay in this case. 

Desktop drives will keep reallocating until it has exhausted its options.  This can cause some raid systems to fail the drive due to time outs. 

I have long suspected this behavior would appear as faults during scrubs that zfs would correct. It matches what I have seen on a few ide drives in used to use. 

Can you get Smart tools to check the drive's stats?  Might be worth it in order to rule out the possibility of early drive failure. 

Greg

Sent from my Droid Incredible.

----- Reply message -----
From: "Michelle Knight" <michelle at msknight.com>
Date: Sat, Jan 22, 2011 3:08 pm
Subject: [OpenIndiana-discuss] ZFS single drive CKSUM errors
To: "Discussion list for OpenIndiana" <openindiana-discuss at openindiana.org>

Hi Folks,

Something I don't understand.

A single drive with a ZFS partition on it. No mirror, no raid, no nothing.

I copied a load of files to it and did a scrub.

It encountered six checksum errors and was able to recover from them ... 
without having any mirror or other redundant reference ... it didn't lose a 
single file.

Now ... am I mad, or does this mean I've got corruption somewhere, or how did 
ZFS manage to recover from cksum errors that it detected on a single drive?

I haven't managed to read anything yet which goes to the depth of explaining 
this.

Can someone help me on this please?

Many thanks,

Michelle.
_______________________________________________
OpenIndiana-discuss mailing list
OpenIndiana-discuss at openindiana.org
http://openindiana.org/mailman/listinfo/openindiana-discuss