Re: WAPBL and replaying log

[Simon Burge <simonb%NetBSD.org@localhost>]

"Sarton O'Brien" wrote:

> On Friday 10 October 2008 14:37:13 Sarton O'Brien wrote:
> > On Friday 10 October 2008 14:19:38 Sarton O'Brien wrote:
> > > On Friday 10 October 2008 14:00:30 Sarton O'Brien wrote:
> > > > I've been using wabl for a while and so far have had a good experience.
> > > >
> > > > Today my xen server crashed and upon reboot all domu started fine
> > > > except one which is sitting on 'replaying log to disk'.
> > > >
> > > > Whenever parity is being rewritten I expect some delays but the domu
> > > > boot process seems to be hung. Attaching the image to a vnd and
> > > > attempting fsck result in the same thing, it just sits on replaying
> > > > log.
> > > >
> > > > Attempting to mount results in a process I cannot kill.
> > > >
> > > > Is it possible, due to the size of the image, that the replay may just
> > > > take a while? From experience that doesn't seem right.
> > > >
> > > > It almost seems like I have some kind of corruption that wabl doesn't
> > > > seem to be able to address.
> > > >
> > > > Any thoughts? Worst case, I'd like to be able to retrieve the data but
> > > > at the moment I can't fsck or mount.
> > > >
> > > > Thanks for any help.
> > >
> > > That was obviously meant to be WAPBL ...
> >
> > Looking at iostat while running fsck, it appears that it may in-fact be
> > doing something ... just very slowly.
> >
> > I'll keep my fingers crossed.
> 
> As I'm not sure as to whether this will work, it's been sitting on replaying 
> log for some time now, would using:
> 
> tunefs -l 0 /dev/vnd0a
> 
> Allow fsck to revert to checking with no log? or is a successful mount 
> required?

That will just tell the WAPBL code to disable the log next time, but
will still try to replay the journal at next mount.

> I thought I'd ask before completely destroying the image :)

A couple of things to try first:

 - Mounting read-only should just replay the journal to memory instead
   of to disk.  That should at least indicate if the journal itself is
   corrupt or not.

 - Fsck also knows how to replay the journal if you give it the -f flag.
   I don't know if -n will replay to memory like a read-only mount does.
   You could try "fsck -n -f ..." first or just "fsck -f ...".

Cheers,
Simon.