Re: [rdiff-backup-users] backups over ssh frequently fail

[Russell Poyner]

On Saturday 15 May 2004 06:41 pm, you wrote:
> >>>>> Russell Poyner <address@hidden>
> >>>>> wrote the following on Fri, 30 Apr 2004 11:27:44 -0500
> >
> > I recently setup a machine to backup 4 others in our lab using
> > rdiff-backup over ssh using a cron job.  Usually at least one of the
> > backups fails with 2 current_mirror files in the rdiff-backup-data
> > directory.
> >
> > At this point I am to new with rdiff-backup to even now how to
> > properly log the errors that are causing the failures, so a tip
> > there would help.
> >
> > I suspect though that what I am seeing is some sort of network
> > problem that is crashing my rdiff-backup sessions.
>
> Perhaps you could have your cron daemon mail you the stdout and stderr
> of the process?  You could also check the files backup.log and
> error_log.<time>.data[.gz] in <dest-dir>/rdiff-backup-data.

Ben,
Thanks for the tip.  The problem turned out to be unrelated to rdiff-backup.  
I had a mosix cluster consisting of 2 athlons and 3 P4s.  The backup server 
is one of the athlons.  Removing the backup server from the cluster caused 
rdiff-backup to work fine again.  Apparently mosix was trying to migrate one 
of the rdiff-backup processes and causing it to crash.

For what its worth.  I discovered this accidently while fixing what seemed to 
be an unrelated problem.  Programs compiled with intel's fortran compiler, 
and p4 optimizations were dying on migration to the athlons.  It seems that 
the programs check what processor they are on, once at startup.  We were 
launching programs from one of the P4s and they were dying of illegal 
instruction errors when they got migrated to one of the athlons.  As I 
mentioned above, stopping mosix on the athlons fixed this, and the 
rdiff-backup problems.

The moral for rdiff-backup users seems to be:  Don't let mosix migrate your 
backup processes.

Russell Poyner