On Wed, 2003-07-30 at 13:22, Wendy Cheng wrote:
> Linux XFS is not a cluster file system - implies that during a failover,
> the file system's buffer cache is lost (together with the faulty machine).
> Even the second machine can pick up the follow-on workload, there
> are possibilities that data could get lost (or corrupted) unless you mount
> them with "sync" option. All the journaling file system (such as XFS)
> could do is to ensure file system's meta data is kept on a consistent
> state but sometimes the journal data could get screwed up too.
>
> You ask too much for a free software :)-
>
> Wendy
> -------
I disagree, if Aman has a ligitimate XFS bug, then the exact same thing
will happen on a non-clustered server which simply is put through a
series of un-clean shutdowns.
It is well known that XFS may lose the contents of a file in this
scenario, but it should not corrupt the actual filesystem.
OTOH, I have not seen any reports of XFS problems like Aman's
previously, so I expect his problem relates to other parts of the FS
stack. i.e. His storage server or LVM usage.
Greg
--
Greg Freemyer
|