Re: [feature] automatically detect hung TASK_UNINTERRUPTIBLE tasks

[Andi Kleen <andi@firstfloor.org>]

On Sun, Dec 02, 2007 at 04:59:13PM -0800, Arjan van de Ven wrote:
> On Mon, 3 Dec 2007 01:07:41 +0100
> Andi Kleen <andi@firstfloor.org> wrote:
> 
> > > We really need to get better diagnostics for the
> > > bad-kernel-behavior-that-is-seen-as-bug cases. If we ever want to
> > > get to the scenario where we have a more or less robust measure of
> > > kernel quality (and we're not all that far off for several cases),
> > > one thing
> > 
> > One measure to kernel quality is to recover well from IO errors
> > (like network problems or broken block devices)
> 
> yes. and this patch will flag cases that don't (yet) work well

If the device/server/... takes more than 2 minutes to recover, how does this
imply the error path "does not work well" ? Or is your goal to handle all 
possible errors in less than two minutes? [That might be a worthy goal,
but is probably far from easy and likely impossible in some cases]

> > This patch will likely work against that by breaking error paths.
> 
> it won't break error paths, it will at most put a warning in the log.
> It doesn't kill or otherwise damage the system or process.

>From the user perspective a kernel randomly throwing backtraces is 
a broken kernel. 

> 
> > 
> > > This patch is a step in the right direction there, by quite a
> > > lot.
> > > 
> > > I really don't understand what your objection is to this patch...
> > > is it that an enterprise distro can't ship with it on? (Which is
> > > fine btw)
> > 
> > Any distribution aimed at end users cannot ship with it on. 
> 
> That's a pretty bold statement; assuming that the TASK_KILLABLE patch
> is in, I don't see the problem.

iirc TASK_KILLABLE fixed NFS only. While that's a good thing there are
unfortunately a lot more subsystems that would need the same treatment.

> > Also in general I have my doubts that the false positive:real bug
> > ratio of this warning is well balanced.
> 
> I'll just have to disagree with you then; but both of us are making
> wild guesses. Only one way to get the real false positive percentage.

Yes let's break things first instead of looking at the implications closely.

-Andi