segfaults testing winbindd in trunk
John P Janosik
jpjanosi at us.ibm.com
Wed May 25 14:51:54 GMT 2005
> On Tue, May 24, 2005 at 03:08:33PM -0700, Jeremy Allison wrote:
> > Looks like memory corruption to me. Have you tried valgrinding it ?
> > We'll need to be careful in ensuring this new code is safe.
> Yes. Valgrind will slow it down considerably, but it probably can provide
> hints. What are you doing exactly? Sid2name and name2sid? Or some other
> operations? If we know the profile you need exactly, the new architecture
> could in theory allow us to fork any number of winbind children per DC,
> can have for example 10 LSA connections doing lookupsid/name
> simultaneously. Dunno whether the DC will handle that, but winbind should
> be a problem anymore.
> This is not done yet and right now nothing more than a wild idea.
I wasn't sure where to start troubleshooting this so I had not looked at
valgrind. I'll try that and let you know where I get. In this test I was
just forking a bunch of processes that call getpwent to put load on
winbindd. I'll have to do some more looking at logs from my customer's
environment to see what the real mix of operations looks like. Being able
to tune winbindd like you mention sounds interesting for large server
consolidations like the one I'm working on here.
More information about the samba-technical