[rdiff-backup-users] Moving the backup-location to another machine

[Ron Leach]

We're replacing our backup server and hit a problem moving our 
existing backup directory to the new machine.  Despite the new 
machine's disk space being slightly larger than the space on the 
existing system, we've run out of space on the new machine, and the 
move is not yet complete.  I think we've either approached the moving 
of the backup incorrectly, or we've not created as large a filesystem 
as we had intended

I wondered whether any other users have tried something similar; 
here's what we did:

Existing backup server uses 2 x 2TB disks in a Raid1 config, exported 
over NFS.  Rdiff-backup backs up local disks onto a 'local' 
destination (which is the NFS export from the backup machine).  One 
partition is exclusively available for backups, and

$ df

reports:

192.168.0.97:/backups 1894289664 1874215424  20074240  99% /mnt/s2backups

ie, 1894289664 1K blocks, 1874215424 blocks used, 20074240 available 
on the existing backup archive partition.

To move the backup archives to the new machine, I used cp .

address@hidden;/mnt/s2backups$ cp -a -u -v .* -t/mnt/s7backups

That is, copying all the content of /mnt/s2backups, including all the 
subdirectories, across to a new, empty, LVM occupying a complete 2 x 
2TB raid1 drive pair on the new server, exported over NFS and mounted 
as s7backups.   I think it had copied all but around 6%, before 
reporting no space left on device.

df on the new backup machine is now reporting:

[...]/backupsvg-backupslv 1922726712 1825057796 4 100% /mnt/s7backups

('/dev/mapper' stripped to avoid line wrap)

ie, 1922726712 1k blocks, 1825057796 used, 4 available

This surprised me.  First, the new disk, as expected, has more 1K 
blocks than the existing archive partition (because it uses the whole 
disk whereas the existing archive is merely one partition on a 2TB 
raid1 array that also holds other data),  But, unlike the existing 
partition (where all 1K blocks are either used or available), on the 
new machine, of the 1.9 bn 1k blocks on the disk, some 97m blocks are 
neither available or used.  Also, the blocks that have been used 
before exhausting the space are fewer, even, than the blocks employed 
on the old archive, so that archive is never going to fit - unless the 
missing 97m blocks come back into play.

Could I ask two questions?  Was the cp command I used a sensible way 
to move an rdiff-backup archive to another machine?

Although slightly off-topic for this list, has anyone encountered the 
problem of 'missing' 1k blocks when using LVM (and, in this case, over 
whole-disk raid1)?   I will look around elsewhere for general comments 
about this, but I wondered whether anyone else was also using this 
type of scheme for their backups and had noticed similar effects.

I'd be grateful for any comments, especially if there is a better 
approach to moving the archive.

regards, Ron