Coda File System

Re: R: Backup problems

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Wed, 13 Jun 2001 10:39:05 -0400
On Wed, Jun 13, 2001 at 04:22:26PM +0200, Andrea Cerrito wrote:
> Jan, first of all thank you for your reply.
> 
> > On Wed, Jun 13, 2001 at 12:00:06PM +0200, Andrea Cerrito wrote:
> > > Hi to all.
> > > I'm trying to backup my volumes, and I set /vice/db/dumplist and
> > > /vice/db/vicetab for my needs.
> >
> > First off, Coda's own backup tool first dumps all volume to the holding
> > area(s) and then spools them to tape.
> 
> Auch!
> Is there a way to just dumps the volume to the holding area without the need
> for tape? I think in the "done-by-hand" way, isn't it?

Oh, sorry. The backup program that you are trying to get working only
does the dump to the holding area. It's the backup.sh script that calls
tape.pl next to spool everything to tape.

> Ok. I've created a /test directory, empty.
> But:
> 
> 1) in vicetab I have to set a not really directory?
> 2) if I use backup -t /vice/db/dumplist /test, all my backups got into it,
> instead of /extra0! What's the need to set a dir in vicetab?

Are you sure they aren't symlinks?... This could be related to the fact
that the backup program didn't recognize the non-fully qualified machine
name (apu-a) as itself.

11:44:55 Warning: the hostname of this server (apu-a.mgt.int) is not
listed in /vice/db/servers

Although it does seem to find the backup partition,

11:44:55 Partition /extra0: 8305048K available (minfree=5%), 8264976K free.

> So, it will be correct to set in /vice/db/vicetab
> 
> apu-a   /imaginarydir backup
> 
> and to launch
> 
> backup -t /vice/db/dumplist /extra0
> ???

Yes that would be incorrect, we actually have in vicetab

dvorak    /backup1   backup
dvorak    /backup2   backup
dvorak    /backup3   backup
dvorak    /backup4   backup
dvorak    /backup5   backup
dvorak    /backup6   backup

The backup command is started with backup -t 135 /vice/db/dumplist
/backup. The volume dumps are then distributed across these 6 backup
partitions.

> This is what I get doing it by hand (I think you forgot volutil -h server
> unlock volume-replica id, isn't it?).

No, that is implicitly done when the backup command completes. We could
probably even do the locking automatically when the backup is started.

> [root_at_apu-a /]# volutil -h apu-a dump 1000003 /extra0/codafs.rootvol.backup
> V_BindToServer: binding to host apu-a
> 
> VolDump failed with Unknown RPC2 return code 200

Try 0x1000003, in some places the strtol has an explicit base 16
specified, but in other places it relies on the 0x prefix to figure out
that the number is hexadecimal. I'm not sure why this error is returned,
maybe there is something in the SrvLog.

> Last: during tests I found two dirs /test/13Jun2001/apu-a and
> /test/13Jun2001/apu-b. Doing a backup will do two copy of the same volume??

Each replica is dumped because Coda uses optimistic replication and
there is no guarantee that both replicas are totally in sync at all
times. When there is a slight difference due to f.i. a temporary
disconnection, a client will detect it when it accesses the object and
trigger the servers to resolve the differences.

Jan
Received on 2001-06-13 10:39:07