deadlock in "zoneli" state on 6.2-PRERELEASE

[LI Xin]

Hi, Nikolay,

Nikolay Pavlov wrote:
> On Monday, 11 December 2006 at 15:59:03 +0800, LI Xin wrote:
>> Hi,
>>
>> Would you please give the following patch a try?
>>
>> http://people.freebsd.org/~delphij/misc/patch-zonelim-drain
>>
>> Note: Please revert my previous patch against sys/kern/kern_mbuf.c.
>>
>> This patch should be applied against sys/vm/ [RELENG_6 or RELENG_6_2],
>> it schedules a drain of uma zones when they are low on memory.
> 
> 
> This time things worked out a bit better, there was no Kernel panic and
> my server managed to overcome the "magic" number 65550 mbufs. But very
> soon the server reached another limit - 131072 mbuf clusters  

Do you still get squid stuck in "zoneli" state and the server became
unresponsive?

> (This is my limit for kern.ipc.nmbclusters).
> And server started to drop the packets. After I've  
> removed the overload I found my server responding but when I actually 
> accessed it I found out that although the number of connections has 
> reduces considerably, the memory allocated for the net did not become
> free. So I believe that there is still a mbufs leak somewhere.

This looks weird to me...  Would you please try to add some load to the
server and remove afterwards, to see if the 'current' mbuf clusters
keeps increasing or not?

Cheers,
-- 
Xin LI <delphij at delphij.net>	http://www.delphij.net/
FreeBSD - The Power to Serve!

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 249 bytes
Desc: OpenPGP digital signature
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20061211/5c46153f/signature.pgp