On Wed, Apr 27, 2005 at 08:05:31AM +0200, Lars Marowsky-Bree wrote:
On 2005-04-27T13:24:36, Neil Brown <neilb@xxxxxxxxxxxxxxx> wrote:
On Saturday April 23, ptb@xxxxxxxxxxxxxx wrote:
> This patch (completely untested of course - what, me?) makes RAID1 write
> to all components of a raid-1 array, else return error to the write
> attempt, when one component cannot be written.
I don't understand why or when you would want this.
This wouldn't just return an error to the application if the write
wasn't completely safe. It would cause the filesystem to switch to
read-only very quickly and make your machine un-usable. Is that
really what you want??
Databases sometimes want this (also for replication).
They'd rather fail than potentially lose a committed transaction, and to
that end they require that the data be written to at least two disks; ie
they want the data to be able to withstand at least one failure. We've
had such a request from a big database vendor for drbd too.
yes, if it were possible to return the failure to the application it
would be great. if it is not, in such "corner" cases it is better to
have an app fail than to promise it has committed a replicated
transaction, and lie about it.
(This however is a great application for >2 mirrors and a write quorum
of two, though.)
+1
L.
--
Luca Berra -- bluca@xxxxxxxxxx
Communication Media & Services S.r.l.
/"\
\ / ASCII RIBBON CAMPAIGN
X AGAINST HTML MAIL
/ \
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
