httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Paul Querna <c...@force-elite.com>
Subject Re: 2.1.10-worker core dumps on minotaur
Date Wed, 23 Nov 2005 15:38:37 GMT
Joshua Slive wrote:
> 
> 
> Joe Orton wrote:
>> On Tue, Nov 22, 2005 at 03:11:40PM -0500, Joshua Slive wrote:
>>> Ahhh... Good idea.  I had just assumed based on the size of the dump
>>> that it had to be the threaded server.  But obviously svn requests
>>> are sometimes memory hungry as well.  I've appended the backtrace
>>> bellow (which is easy to get when you use the proper httpd), and yes
>>> it does seem to be in subversion.
>>>
>>>
>>> #0  0x2840b777 in kill () from /lib/libc.so.5
>>> #1  0x2840081e in raise () from /lib/libc.so.5
>>> #2  0x2847415b in abort () from /lib/libc.so.5
>>> #3  0x284f651c in abort_on_pool_failure (retcode=12)
>>>     at ../../svn-current/subversion/libsvn_subr/pool.c:46
>>> #4  0x28396bc4 in apr_palloc (pool=0x8120018, size=16)
>>
>> Are all the debuggable crashes like this?  (The crash is due to
>> malloc() failing in case that's not clear)
> 
> At the time I sent it, that was the only non-truncated core.  We now
> have several more, and they all look about the same.  Interestingly, the
> core files are all almost exactly the same size (538 MB) which, along
> with the malloc failure, points strongly to hitting a resource limit
> someplace.  Any freebsd gurus want to comment?

FreeBSD has a default memory resource limit of 512 mb max per process.

We need to set the MAXDSIZE in the kernel to something higher.  It can
be compiled this way, or tuned via a sysctl.

-Paul

Mime
View raw message