Coda File System

Re: Server keeps crashing after conflict resolve attempt

From: Janusz Krzysztofik <jkrzyszt_at_tis.icnet.pl>
Date: Sat, 31 May 2008 17:52:15 -0400
Dnia 31-05-2008, sob o godzinie 01:04 -0400, Jan Harkes napisaƂ(a):
> On Wed, May 28, 2008 at 12:20:36PM +0200, Janusz Krzysztofik wrote:
> >> Wow, that is a lot of information, I already see some things that make
> >> me feel like the problem could be related to some 64-bit issue. Some of
> >> the numbers like volume-ids didn't look right in one of the SrvLog
> >> files.
> >
> > Yes, my server platform is x86_64.
> >
> > It happend once that the server was crashing right after start before  
> > any client conected. After I examined the rvm and found almost no  
> > changes pending, I decided to reinit rvm - maybe there are still some  
> > consequences visible.
> 
> That is definitely possible, although most files and directories would
> be using a different fid, the root directories of a volume will always
> use the same vnode/uniquifier values. And if the directory has seen the
> same number of operations as before the client may have a cached copy of
> this directory from before the reintialization which happens to have the
> same version-vector as the new one but with different content.
> 
> > What else, I have compiled 6.9.3 on x86_64 myself with kerberos enabled,  
> > as well as required versions of libraries and tools. Unfortunately,  
> > rvmutl did not work (Segmentation fault), so I ended up with installing  
> > 6.9.2, setting up the server, and then upgrading to 6.9.3. It works, but  
> > maybe my problems are related.
> 
> Segfaults are a clear indication that something is wrong, I don't know
> how installing 6.9.2 and then upgrading could fix such a thing. If your
> system/c-compiler miscompiled some code (leading to the segfaults) then
> your 6.9.3 version may still contain some bad code.

Jan,

To fix the issues, I am going to create a new server, this time on i386
platform. Please advise me if it is correct to replicate my (possibly
corrupted) volume to the new server, move the SCM function there and
then destroy the old server, or I should rather restore my files to the
new server from backup.

Regards,
Janusz
Received on 2008-05-31 22:43:49