FreeBSD 5.1-R kernel panic

[Stephane Raimbault]

Hi Bosko,

Well a couple weeks ago I re-cvsup'd to current since I had missed one of your updates it would seem by a day.  I still had the panic occur... so I did as suggested and upped my KVA.  I did the following:

In the kernel
options         KVA_PAGES=400
options         NMBCLUSTERS=8192

In the loader.conf
kern.vm.kmem.size="400000000"

This seems to have helped, since I am no longer panic'ng.  I have identified a php script that slowly builds to 100% cpu usage in a couple days which we believe was a culprit in our problems.  The guys who made the php script that runs on this box is having a close look at it to figure out why it's behaving this way... however I did have some questions.

1. Should the panic occur in the first place?
2. Is there a way to monitor how much kva is currently being used so that I can monitor this value so I can have something alert me if I'm running out of this memory... I'm sure we would of caught this months ago if I was monitoring for this?

Thanks again for everything you have done to assist me thru this problem.

Stephane Raimbault.



bmilekic at technokratis.com (Bosko Milekic) wrote in message news:<bhgfsc$215f$1 at FreeBSD.csie.NCTU.edu.tw>...
> On Thu, Aug 14, 2003 at 10:57:07AM -0600, Stephane Raimbault wrote:
> > Hi Bosko,
> > 
> > This is the output of sysctl vm.zone about 2 minutes before the crash
> > occured.  let me know if there is anything else I can provide you for this
> > crashing problem.
> 
>   Hmmmm.  I don't know, maybe you really do have a machine too loaded
>   for the KVA you have configured...
> 
>   I have to re-iterate that it's extremely important that you
>   double-check that you are in fact in sync with the latest -current and
>   _NOT_ RELENG_5_1.  Make sure you're building at least version 1.73 of
>   src/sys/vm/uma_core.c (grep FBSDID src/sys/vm/uma_core.c).
> 
>   With that said, you can try the following:
> 
>   options KVA_PAGES=400
> 
>   in your kernel configuration file.
> 
>   Following that, you can do this:
> 
>   kern.vm.kmem.size=400000000
> 
>   In your /boot/loader.conf
> 
>   Make sure to not set NMBCLUSTERS too high.  Around 8K is probably more
>   than enough, but you should look at how much you're using on average
>   with `netstat -m' and then set the number to roughly 3 times that.
> 
>   If following this your crash persists, even if after a longer time,
>   then I would suspect (another?) race.  Again, I have to re-iterate
>   that you really need to make sure you're supping to HEAD:
> 
>   *default release=cvs tag=.
> 
> Regards,