[Linux-cluster] EXT3 service mounted on two nodes

Josh Gray jgray at nicusa.com
Mon Oct 29 15:31:47 UTC 2007


I can tell you for a fact that is what happened unfortunately.    Two nodes
of 3 became active and were quite content to serve my NFS traffic on
multiple VIPs and write to the same EXT3 volume for about 20 or so minutes
until it went read only.

Most frustrating part of the whole deal was restarting the cluster would
just hang on 'starting'  with NO error message logged about what it was
trying to do.  In hindsight it was probably fscking as that's what I had to
do manually to get it back up after a ton of trial and error.  I even raised
the verbosity of the log output to no avail.

Josh


On 10/29/07 11:20 AM, "Bryn M. Reeves" <breeves at redhat.com> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> gordan at bobich.net wrote:
>> Sure, that works. The only problem is that if both nodes write to the
>> same files at the same time (including meta data), you'll end up with a
>> corrupted file system. But if both machines are mounting the FS
>> read-only (in which case you might as well use ext2), then there's no
>> problem with that.
> 
> Are you sure about that? I've never known it do anything useful beyond
> spectacularly breaking the file systems involved.
> 
> This is especially true for ext3 due to its journaling capability - as
> soon as the two hosts start squabbling over the same journal block (i.e.
> they detect an inconsistency caused by the other host overwriting
> something) they're both going to abort the journal via an ext3_abort()
> call and take the file system readonly.
> 
> In my experience, this tends to happen pretty quickly - the moment the
> second host begins replaying the incomplete journal entries the first
> host has outstanding.
> 
> Also, if mounting an ext3 volume read only you might want to consider
> making the underlying block device read only too - this prevents any
> problems with a host accidentally attempting journal recovery.
> 
> Regards,
> Bryn.
> 
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.7 (GNU/Linux)
> Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
> 
> iD8DBQFHJfov6YSQoMYUY94RAlTAAKDHOjaWg81rd6ZkoBL4rCUGS5xZPACdHWlo
> Q0kzUfTQYQ7wIgIlskITv14=
> =s2TT
> -----END PGP SIGNATURE-----
> 
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
> 

-- 
Josh Gray
Systems Administrator
NIC Inc

Email: jgray at nicusa.com
Desk/Mobile: 913-221-1520

"It is not the mountain we conquer, but ourselves."
- Sir Edmund Hillary






More information about the Linux-cluster mailing list