[Discuss] xfs repair

[adler at stephenadler.com (Stephen Adler)]

On 09/22/2014 09:37 PM, Dan Ritter wrote:
> On Mon, Sep 22, 2014 at 09:16:01PM -0400, Stephen Adler wrote:
>> Guys,
>>
>> I'm preparing to building several multi terabyte file systems using
>> XFS. On the order of 5-10 terabytes. I've done a bit of searching on
>> the web to find out what is the customary way to check xfs file
>> systems. with the other file systems, you would get a forced fsck
>> ever N number of mounts thus repairing any "gunk" in the file
>> system. With XFS, no fsck is performed. fsck.xfs will return passed
>> and not execute. Thus one needs to run xfs_repair every now and then
>> I suppose. So my question is what is the customary frequency one
>> runs xfs_repair on their xfs file systems. Every 6 months? Do it via
>> a cron job every week making sure you can dismount the file system
>> first and then remount it when done? It's been beaten into me over
>> the years I've managed linux/unix systems that fscks must be done
>> with some kind of periodicity otherwise suffer from file corruption
>> due to bit decay I suppose.
> I suppose the real question is, why XFS? And what kind of
> redundancy?
>
> I expect that most large storage systems will be rebooted extremely
> infrequently, and so some form of online fault detection and
> repair will become critical.
>
> -dsr-
>
>
XFS is the new default in red hat enterprise linux 7. I did some 
fsck.ext4 vs xfs_repair benchmarking and it seems xfs_repair runs 
faster, but I can't really tell if I'm doing a real comparison or not. 
(about 2-3 minutes on a 4 Terabyte fs with about 1 terabyte of data on 
it.) On an rhel 6 system, fsck my 5 TByte file system takes about 3-4 
hours. It could be because I have old disks (slower than the ones I have 
now) or maybe I'm running ext3 (I'm kind of embarrassed to not know what 
fs I'm running on my current server).

I guess run xfs_repair in a cron job every sunday night.