kill task in TASK_UNINTERRUPTIBLE

kill task in TASK_UNINTERRUPTIBLE

[Amol Lad]

Hi,
 Is there any way i can kill a task in
TASK_UNINTERRUPTIBLE state ?

please cc me

Thanks
Amol


__________________________________________________
Do You Yahoo!?
Yahoo! Health - your guide to health and wellness
http://health.yahoo.com

Re: kill task in TASK_UNINTERRUPTIBLE

[Denis Vlasenko]

On 8 May 2002 12:01, Amol Lad wrote:
> Hi,
>  Is there any way i can kill a task in
> TASK_UNINTERRUPTIBLE state ?

No. Everytime you see hung task in this state
you see kernel bug.

Somebody correct me if I am wrong.
--
vda

Re: kill task in TASK_UNINTERRUPTIBLE

[Robert Love]

On Wed, 2002-05-08 at 13:23, Denis Vlasenko wrote:
> On 8 May 2002 12:01, Amol Lad wrote:
> > Hi,
> >  Is there any way i can kill a task in
> > TASK_UNINTERRUPTIBLE state ?
> 
> No. Everytime you see hung task in this state
> you see kernel bug.
> 
> Somebody correct me if I am wrong.

Generally correct.  Of course, what is "hung" ?

If it is just sitting in uninterruptible sleep, it could legitimately be
waiting for an event.  More than likely, however, after some sane period
of time something is broken.

So, yah, it is a kernel bug.

I'll expand on your answer too - _why_ can't we kill it?  Same argument
we had over saving the futexes if a process bails.  You hold a semaphore
because you are entering a critical section.  If you die in the middle,
who knows the state the data is in and you do _not_ want to reenter it.

	Robert Love 

Re: kill task in TASK_UNINTERRUPTIBLE

[Philippe Troin]

Denis Vlasenko <vda@port.imtp.ilyichevsk.odessa.ua> writes:

> On 8 May 2002 12:01, Amol Lad wrote:
> > Hi,
> >  Is there any way i can kill a task in
> > TASK_UNINTERRUPTIBLE state ?
> 
> No. Everytime you see hung task in this state
> you see kernel bug.
> 
> Somebody correct me if I am wrong.

Except for processes accessing NFS files while the NFS server is down:
they will be stuck in TASK_UNINTERRUPTIBLE until the NFS server comes
back up again.

Phil.

Re: kill task in TASK_UNINTERRUPTIBLE

[george anzinger]

Philippe Troin wrote:
> 
> Denis Vlasenko <vda@port.imtp.ilyichevsk.odessa.ua> writes:
> 
> > On 8 May 2002 12:01, Amol Lad wrote:
> > > Hi,
> > >  Is there any way i can kill a task in
> > > TASK_UNINTERRUPTIBLE state ?
> >
> > No. Everytime you see hung task in this state
> > you see kernel bug.
> >
> > Somebody correct me if I am wrong.
> 
> Except for processes accessing NFS files while the NFS server is down:
> they will be stuck in TASK_UNINTERRUPTIBLE until the NFS server comes
> back up again.

A REALLY good argument for puting timeouts on your NSF mounts!  Don't
leave home without them.

-g

-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Real time sched:  http://sourceforge.net/projects/rtsched/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml

Re: kill task in TASK_UNINTERRUPTIBLE

[Roman Zippel]

Hi,

On Wed, 8 May 2002, george anzinger wrote:

> > Except for processes accessing NFS files while the NFS server is down:
> > they will be stuck in TASK_UNINTERRUPTIBLE until the NFS server comes
> > back up again.
> 
> A REALLY good argument for puting timeouts on your NSF mounts!  Don't
> leave home without them.

Use "mount -o intr" and you can kill the process.

bye, Roman

Re: kill task in TASK_UNINTERRUPTIBLE

[Jan Hudec]

On Thu, May 09, 2002 at 01:49:35AM +0200, Roman Zippel wrote:
> > > Except for processes accessing NFS files while the NFS server is down:
> > > they will be stuck in TASK_UNINTERRUPTIBLE until the NFS server comes
> > > back up again.
> > 
> > A REALLY good argument for puting timeouts on your NSF mounts!  Don't
> > leave home without them.
> 
> Use "mount -o intr" and you can kill the process.

Could someone please explain to me how this works? IIRC NFS uses
generic_file_read as most other filesystems. And whe WaitOnPage in there
sleeps in uninterruptible state. I was told, that though it would be
easy to change here, it's almost impossible in page-fault, because
trying to handle a signal might trigger the very same page-fault again.

--------------------------------------------------------------------------------
                  				- Jan Hudec `Bulb' <bulb@ucw.cz>

Re: kill task in TASK_UNINTERRUPTIBLE

[Denis Vlasenko]

On 8 May 2002 21:27, george anzinger wrote:
> > > >  Is there any way i can kill a task in
> > > > TASK_UNINTERRUPTIBLE state ?
> > > No. Everytime you see hung task in this state
> > > you see kernel bug.
> > > Somebody correct me if I am wrong.
> >
> > Except for processes accessing NFS files while the NFS server is down:
> > they will be stuck in TASK_UNINTERRUPTIBLE until the NFS server comes
> > back up again.
>
> A REALLY good argument for puting timeouts on your NSF mounts!  Don't
> leave home without them.

Timeouts may be a bad idea: imagine large (LARGE) database
which you don't want to repair due to lost data over NFS.
Better let it hang in NFS i/o even for hours while you are
repairing your network.

OTOH, interruptible NFS mounts are ok: processes can be killed but
with explicit admin action only, just what I need.

Anybody uses _uninterruptible_ NFS mounts? Enlighten me why.
--
vda

Re: kill task in TASK_UNINTERRUPTIBLE

[Andrew Theurer]

Denis Vlasenko wrote:
> 
> On 8 May 2002 21:27, george anzinger wrote:
> > > > >  Is there any way i can kill a task in
> > > > > TASK_UNINTERRUPTIBLE state ?
> > > > No. Everytime you see hung task in this state
> > > > you see kernel bug.
> > > > Somebody correct me if I am wrong.
> > >
> > > Except for processes accessing NFS files while the NFS server is down:
> > > they will be stuck in TASK_UNINTERRUPTIBLE until the NFS server comes
> > > back up again.
> >
> > A REALLY good argument for puting timeouts on your NSF mounts!  Don't
> > leave home without them.
> 
> Timeouts may be a bad idea: imagine large (LARGE) database
> which you don't want to repair due to lost data over NFS.
> Better let it hang in NFS i/o even for hours while you are
> repairing your network.

I'm not sure using an NFS mount for a big important DB would be prudent
in the first place.  I dunno, maybe there are situations where it's
unavoidable.  I just really cringe when hearing about DB volumes over
NFS.

-Andrew

Re: kill task in TASK_UNINTERRUPTIBLE

[Alan Cox]

> > TASK_UNINTERRUPTIBLE state ?
> 
> No. Everytime you see hung task in this state
> you see kernel bug.

Or waiting on a resource that isnt available - that can occur for example
with NFS for long periods, or for a few minutes when burning a CD and the
IDE bus is locked

Re: kill task in TASK_UNINTERRUPTIBLE

[David Woodhouse]

alan@lxorguk.ukuu.org.uk said:
>  Or waiting on a resource that isnt available - that can occur for
> example with NFS for long periods, or for a few minutes when burning a
> CD and the IDE bus is locked -

Often the main reason for sleeping in uninterruptible state during these 
periods is because the difficulty of performing a sane cleanup exceeds the 
boredom threshold of the programmer. There are plenty of places where 
TASK_UNINTERRUPTIBLE is used just because people have been lazy.

I'm guilty of it too - I use TASK_UNINTERRUPTIBLE for anything which can be
called from jffs2_read_inode() because I was too lazy to chase through a
mechanism by which ->read_inode() may return -ERESTARTSYS without creating a
permanent bad inode. But we're working on it, and this should get fixed in
2.5. 

--
dwmw2

Re: kill task in TASK_UNINTERRUPTIBLE

[Denis Vlasenko]

On 8 May 2002 16:33, Alan Cox wrote:
> > > TASK_UNINTERRUPTIBLE state ?
> >
> > No. Everytime you see hung task in this state
> > you see kernel bug.
>
> Or waiting on a resource that isnt available - that can occur for example
> with NFS for long periods, or for a few minutes when burning a CD and the
> IDE bus is locked

I really prefer interruptible NFS mode (without timeout).

If CD burner needs to completely lock IDE, well, that is less than wonderful 
piece of hardware. I won't blame kernel for this.
--
vda

RE: kill task in TASK_UNINTERRUPTIBLE

[Kerl, John]

Please, this could turn into a flamewar:

*	Users hate the NFS hangs.
*	Applications need them for consistency.

Is there really a solution that makes everyone happy
when NFS servers are down?  If so, I haven't seen it
in my career working with NFS.  (Unless NFS v3 helps ...)
All the network admins I've known choose non-interruptible,
tolerate the complaining users when a server is down,
and just work on getting the server back on-line ASAP.

> -----Original Message-----
> From: Andrew Theurer [mailto:habanero@us.ibm.com]
> Sent: Thursday, May 09, 2002 10:39 AM
> To: vda@port.imtp.ilyichevsk.odessa.ua
> Cc: linux-kernel@vger.kernel.org
> Subject: Re: kill task in TASK_UNINTERRUPTIBLE
> 
> 
> 
> 
> Denis Vlasenko wrote:
> > 
> > On 8 May 2002 21:27, george anzinger wrote:
> > > > > >  Is there any way i can kill a task in
> > > > > > TASK_UNINTERRUPTIBLE state ?
> > > > > No. Everytime you see hung task in this state
> > > > > you see kernel bug.
> > > > > Somebody correct me if I am wrong.
> > > >
> > > > Except for processes accessing NFS files while the NFS 
> server is down:
> > > > they will be stuck in TASK_UNINTERRUPTIBLE until the 
> NFS server comes
> > > > back up again.
> > >
> > > A REALLY good argument for puting timeouts on your NSF 
> mounts!  Don't
> > > leave home without them.
> > 
> > Timeouts may be a bad idea: imagine large (LARGE) database
> > which you don't want to repair due to lost data over NFS.
> > Better let it hang in NFS i/o even for hours while you are
> > repairing your network.
> 
> I'm not sure using an NFS mount for a big important DB would 
> be prudent
> in the first place.  I dunno, maybe there are situations where it's
> unavoidable.  I just really cringe when hearing about DB volumes over
> NFS.
> 
> -Andrew
> -
> To unsubscribe from this list: send the line "unsubscribe 
> linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>