process table fills with DN state when nfs connection is lost

process table fills with DN state when nfs connection is lost

[Roeland Th. Jansen]

hi *

I have a network with a different linux system hat exports a few dirs to
this system.

what happens is this : the other system reboots into windows o the nfs
connection gets lost. however, what happens is that now the process
table starts to fill with cron initiated mrtg calls and all get the DN
state in ps aux.

now, the load goes up and when I came home it was at 144 already. I
stopped cron, unable to kill off the mrtg calls. I then re-established
the nfs connection by rebooting the windows back to linux; I then could
umount the nfs shares and guess what happened -- all DN processes went
ayway and the system load back to normal.

it seems that some things block in the kernel when the nfs stuff is
failed. is this right ? is my setup incorrect or what ?

Roeland

-- 
Grobbebol's Home                   |  Don't give in to spammers.   -o)
http://www.xs4all.nl/~bengel       | Use your real e-mail address   /\
Linux 2.2.16 SMP 2x466MHz / 256 MB |        on Usenet.             _\_v  

Re: process table fills with DN state when nfs connection is lost

[Roeland Th. Jansen]

On Sun, Jun 10, 2001 at 10:34:28PM +0000, Roeland Th. Jansen wrote:
> I have a network with a different linux system hat exports a few dirs to
> this system.

oops, this is 2.4.5. but happened before as well.
-- 
Grobbebol's Home                   |  Don't give in to spammers.   -o)
http://www.xs4all.nl/~bengel       | Use your real e-mail address   /\
Linux 2.2.16 SMP 2x466MHz / 256 MB |        on Usenet.             _\_v  

Re: process table fills with DN state when nfs connection is lost

[Lucca]

>what happens is this : the other system reboots into windows o the nfs
>connection gets lost. however, what happens is that now the process
>table starts to fill with cron initiated mrtg calls and all get the DN
>state in ps aux.

The default is to do a hard-mount of NFS shares.  Hardmounts hang when the
server disappears.  This is... Generally what you will want.  adding the
intr option should allow the processes to be killed (but I have not tried
this)

>now, the load goes up and when I came home it was at 144 already. I
>stopped cron, unable to kill off the mrtg calls. I then re-established the
>nfs connection by rebooting the windows back to linux; I then could umount
>the nfs shares and guess what happened -- all DN processes went ayway and
>the system load back to normal.

Count yourself lucky.  I was foolishly serving webpages from an NFS
mount.  Fortunately apache has limits.  Seeing a loadavg in the the hundreds
is... disturbing.

The loadavg does this because processes in the D state get counted wether
they are doing anything or not.  I don't know if this is correct or not, but
most unicies seem to do this.

>it seems that some things block in the kernel when the nfs stuff is
>failed. is this right ? is my setup incorrect or what ?

Not incorrect, but you might experiment with soft mounts, which will rapidly
timeout and die with io-errors rather than hanging.

lucca@shogun1.csun.edu

Re: process table fills with DN state when nfs connection is lost

[Roeland Th. Jansen]

On Sun, Jun 10, 2001 at 06:44:56PM -0700, Lucca wrote:
> Not incorrect, but you might experiment with soft mounts, which will rapidly
> timeout and die with io-errors rather than hanging.

Gombas also replied with such information.

I didn't know that this was causing it. I changed the (default hard) to
soft mount and now it doesn't go wrong. best woul dbe to tell my
girlfriend not to reboot while nfs is active though :-)

Thanks anyways.

-- 
Grobbebol's Home                   |  Don't give in to spammers.   -o)
http://www.xs4all.nl/~bengel       | Use your real e-mail address   /\
Linux 2.2.16 SMP 2x466MHz / 256 MB |        on Usenet.             _\_v