We've got substantial NFS infrastructure over here, and we're
gradually migrating to NASland. We're not pleased with NDMP, so we're
doing proxy backups, and have found a pocket of insane behavior.
This is RHEL 5.5, against an EMC NSG8. I've duplicated the behavior
with TSM client versions 5.5.2 and 6.2.1
I'll start with a distilled example, and then talk about how I got
there.
Start with nothing NFS mounted. Then:
mount nfs-server:/export/homes /mnt/homes
dsmc incr /homes -snapshotroot /mnt/homes
[ an incremental happens. Whee! ]
Then:
mount nfs-server:/export/homes/someuser /mnt/someuser
dsmc incr /homes -snapshotroot /mnt/homes
'someuser' gets expired out of the backup.
I can read files from /mnt/someuser/blah and /mnt/homes/someuser/blah
while this is going on; consequently, both the NFS server and client
are capable of dealing with the double-mounted space. It's just TSM's
logic. I'm currently re-running the experiment without the
snapshotroot to exclude that.
I got here because, after setting up a new proxy machine, I chanced to
log into it as me, not just as root. Next incr I ran, I saw it
expring all my stuff, which was a shock. So I started trying to
remove variables from the equation. I was willing to blame it all on
the automounter, but then I turned that off and still got the problem
with static mounts.
The namespace the world sees is the '/homes/' one (actually '/h/', but
who's counting?) and we wanted to use snapshotroot to permit us to
advertise that namespace, for simplicity of restores.
Anyone inclined to confirm?
- Allen S. Rout
|