Re: unkillable processes using samba, xfs and lvm2 snapshots (k 2.6.10)

[Julien BLACHE <jb@jblache.org>]

Gildas LE NADAN <gildas.le-nadan@inha.fr> wrote:

> I experience hangs on samba processes on a filer using xfs over lvm2
> as data partitions, when there is active snapshots of the xfs
> partitions.

Your problem probably lies between lvm2 and XFS. I got the same
problems this summer while doing the exact same thing.

The server would just completely hang once I started doing lvm
snapshots:
 -> at the beginning, the snapshots would work OK, but XFS would hang
    when accessing the filesystem afterwards
 -> after a while (usually 2 or 3 snapshots, and I was taking a
    snapshot every 2 hours), the snapshot would not complete, and
    then only a hard reboot would work

I was doing the snapshots from a crontab, the script used xfs_freeze
to freeze the filesystem before doing the snapshot (and unfreeze it
afterwards, of course). Sometimes xfs_freeze -u would hang too (but at
this time, the server was in a pretty bad state already).

The server wasn't loaded at all, we were doing some reads/writes
through samba to have some modified files lying around, but we were
mainly prototyping the server, not stress-testing it.

LVM and XFS just don't play nice together when it comes to snapshots,
I thought it had been fixed already, but it's not the case, as we both
know...

(I can't remember the kernel version, it could have been a 2.4 kernel,
but I was using LVM2 and the latest XFS code available)

Feel free to correct me if I did something wrong (but AFAIK I took
care of everything, knowing there could have been bad interactions
between LVM and XFS).

JB.

-- 
Julien BLACHE                                   <http://www.jblache.org> 
<jb@jblache.org>                                  GPG KeyID 0xF5D65169