Migration from the KQ Implementation

Gordan Bobic gordan.bobic at gmail.com
Thu May 5 16:39:14 EDT 2011


On 05/05/2011 03:43 PM, devsk wrote:
> On May 5, 4:53 am, Uncle Stoatwarbler<stoatw... at gmail.com>  wrote:
>> This is as much about perception as reality. A fs without some form of
>> single-user/offline recovery mode is regarded with extreme suspcion by
>> greybeards, with lots of past experience as the reason why.
>
> An FS without a single-user/offline recovery is exactly what we need.
> Who wants to fsck the FS at boot? This is so 90's mentality. ZFS is
> the only FS that fills that need for me.

You are missing the point wholesale. fsck should never be required for 
normal operation. But it _IS_ required when things go wrong.

> I have reset my system many many times, some even during high FS
> activity. I have yet to hit a case where zfs-fuse refused to import
> the pool (devsk touches wood..;)). If ZFS fails to import a pool, you
> are either hitting a bug in ZFS, which should be reported and fixed,
> or you should buy better hardware next time. An FSCK will not help in
> this case, because it will fail to do exactly what zpool import is
> failing to do: traversing the messed up data structures on disk.

Whether it is a bug or not is irrelevant. When you can't access your 
data due to a metadata corruption (Aneurin's example that sounds very 
similar to the problem I have seen on Solaris 10), arguing that it is a 
bug in ZFS is counterproductive. If you have to wait for a patch from 
Oracle, you might as well shut up shop. With zfsonlinux you probably 
stand a better chance, but it's still not going to be as fast as fsck. 
Or you can restore from backups, which may be prohibitively time-consuming.

> Or you can stop blaming ZFS for being the first FS without a need for
> FSCK and live with misguided security that the presence of /sbin/
> fsck.ext4 provides.

The fact that there isn't a fsck.zfs doesn't mean it's not needed. fsck 
was supposed to be a thing of the past with the advent of journalling 
file systems like ext3. Yet reality shows us that things aren't that 
clear cut.

Gordan



More information about the zfs-discuss mailing list