"panic: getnewbuf: locked buf" on heavy load 5.4-STABLE

[Chuck Swiger]

AK wrote:
> Hello!

Hello.

> On Friday 16 September 2005 18:40, Chuck Swiger wrote:
>> Please read "man crash" and the part of the handbook which talks about how
>> to obtain a crashdump or backtrace with more information about the panic
>> you see.
>>
>> You might consider putting both a backtrace and a dmesg from the system on
>> the web somewhere, and post links to that.
>  
>  To get a crashdump I should have swapspace >= total ram.
>  Currently it is only 2G, I'm trying to think of smth. to  get more swapspace.
>  dmesg is available @ http://getfile.biz/dmesg.txt

Take a look at /boot/loader.conf, you can limit the amount of physical RAM 
FreeBSD uses to less than 2GB for long enough to get a valid dump, assuming the 
crash is easily reproducable.

>>>For some reason my 5.4 STABLE is dying each 4-5 hours with panic above.
>>>This is big Web server with Apache configured to 1500 MaxClients. All
>>>kernel tunings suggested for high load servers by freebsd handbook are in
>>>place. Hardware is Intel server board, P4-3.0 cpu with HT off, 3Gb ram,
>>>1eide, 2sata.
>>
>> By the way, you don't have enough RAM to handle 1500 concurrent Apache
>> processes, even with a minimal config (ie, no SSL, no mod_perl, no PHP).
> 
>  Mem: 1203M Active, 1293M Inact, 383M Wired, 125M Cache, 112M Buf, 4340K Free
>  Looks like enough to me...

Setting MaxClients to a really high number doesn't mean you actually have 1500 
httpd's running at that point in time.

What does "ps auxww | grep httpd | wc -l" say?

-- 
-Chuck