[plug] How long to fsck 160 GB?

[Patrick Coleman]

On 6/27/06, Bernd Felsche <bernie at innovative.iinet.net.au> wrote:
> "Patrick Coleman" <blinken at gmail.com> writes:
>
<snip>
>
> >I've fscked about 30GB worth of reiser, not full, and it took a while
> >(probably about 30 minutes; I wasn't timing). So 2 hours or more for a
> >complete tree scan would be about right.
>
> Actually a filesystem on a ITE8212 PATA RAID using 4*80GB;
> mirrored+striped, so I expected it to be quite a bit faster.
>
> Well it finished the fix-fixable a few minutes after posting.
> Just two minor problems - wrong block count and of course bitmap.
> I resized the filesystem online... just because I could.

Heh. Done that too, but never had issues - growing filesystems seemed
fairly stable, but I'll be more careful now. The corruption I had
before was on a filesystem I -didn't- resize, interestingly enough.

> >I ran into some wonderful problems with reiser - what was happening
> >was that the filesystem was corrupted in such a way as to hard-lockup
> >the machine randomly. The machine (a VPS in this case) wouldn't do
>
> Did you report the bug and collect your reward?

No - I googled the message that was coming up on the console after the
system had gone completely cactus, and found a post on the kernel list
discussing it. Of course, I can't remember the message that was
appearing, and I cleverly didn't write it down for future reference.
>From memory it was a deadlock in the kernel reiser code.

> I've not had unfixable errors except on bad discs. In one case, the
> system would roll back hours of stuff after crashing... the hard
> drive didn't report that it had a problem writing to some blocks so
> when the crash happened, everything got rolled back.

I certainly hope the 'disc' isn't bad, as it's LVM on RAID5. In this
case reiserfsck would happily report that everything was AOK on the
volume after the second check, so I'd boot the VPS which would then
promptly crash a day or so later.

> >anything - keyboard wasn't even working. fscking succeeded happily
> >without resolving anything. I ended up creating another partition
> >(thankyou LVM) the same size, formatting with ext3 and copying the
> >data across.
>
> I suspect that Reiserfs was "over-stressing" some part of the
> system; possibly causing an interrupt to be missed and resulting in
> a lock-up. Partly a driver, partly a filesystem and partly a
> hardware issue. The filesystem is a victim to the lies it's told.

Hmm. I'm not so sure that it wasn't the reiserfs code to blame here.
Of course, I don't have the error message here right now to do more
research, but I've become somewhat more suspicious of reiser. Thats
not to say I still don't use it - it does seem to have pretty good
performance and converting all the other servers would be an awful lot
of effort :)

I think the trick is to have good, automatic backups regardless of
what filesystem you use.

-Patrick

-- 
http://www.labyrinthdata.net.au