From: Tad Dolphay <tbd@sgi.com>
Message-Id: <200310141501.h9EF1Nco078807@fsgi158.americas.sgi.com>
Subject: Re: XFS and async NFS exports
To: Nicolas.Kowalski@imag.fr (Nicolas Kowalski)
Date: Tue, 14 Oct 2003 10:01:22 -0500 (CDT)
Cc: linux-xfs@oss.sgi.com
In-Reply-To: <vqoekxfly3w.fsf@astazou.imag.fr> from "Nicolas Kowalski" at Oct 14, 2003 03:58:43 PM
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: linux-xfs-bounce@oss.sgi.com
Errors-to: linux-xfs-bounce@oss.sgi.com
Precedence: bulk
Content-Length: 882
Lines: 20


> recently faced a power outage on a fileserver with XFS filesystems
>(CVS-2003-09-24_05). Some files written on NFS clients a few seconds
>before the crash were damaged (zero-sized). 
>
>On a first approach, I thought this was XFS-related, but afterwards, I
>also realized that this server was exporting its filesystems with the
>async export option...It this possible that this async option have
>caused these file damages ? If I use sync exports, does this mean that
>XFS will effectively write on disk all pending data ?

If "async" is used, the NFS server can respond that the data is commited
when it really isn't on disk. Thus, when the NFS server crashes, the 
NFS client does not think it has to re-commit the data.

If sync exports are used on the NFS server and hard mounts are used on
the NFS client, I doubt there will be any loss of data when a server crashes.

Tad