Re: fsync() madness

[Sami Liedes <sliedes@xxxxxxxxx>]

On Mon, Apr 21, 2008 at 05:59:24PM +0200, Oswald Buddenhagen wrote:
> On Mon, Apr 21, 2008 at 06:40:13PM +0300, Sami Liedes wrote:
> > Without a power loss?
> >
> or a driver-induced crash, etc.

Do you expect them regularly? I haven't had my computer crash that
hard for years, despite occasionally using -rcN kernels and fully
expecting such behavior. Admittedly I refuse to touch binary-only
drivers with a ten foot pole, so that may be a factor too.

> the problem with xfs (well, actually, to a lesser degree every file
> system which has less conservative defaults than ext3) is that it will
> lose data which was previously safe. you end up with a config file
> (or whatever else) that is stuffed with zeros. i sure do not expect
> *that* from a power loss.

Well, then go bug at the developers of those filesystems. Or tell the
users that's what they should do. It's not the job of any application
to second-guess the kernel, especially when it kills all performance.
I prefer write back cache to losing 90 % of the performance, and I
think so do most others, otherwise there would be no write back cache
in the kernel.

Alternatively, do as I do and get a UPS. The hardware alone is
probably already valuable enough that you should do that, and if your
data is worth anything to you, it should be an easy decision.

The philosophy of tolerating sudden power losses or crashes is
obsolete by necessity. People are not willing to take vastly inferior
performance in exchange for having their browser history safe no
matter what happens. And it's not a decision any application should
force down anybody's throat, no matter how misguided the application's
developers think the philosophy employed in current filesystems is.

	Sami

Attachment: signature.asc
Description: Digital signature

>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<