Googling DeadLockDebugger should show the recent emails I got from a few folks. Make sure you use Zope's python and that you can import threadframe by itself to start with. Also, if you use squid or something in front of your site make sure you use the url directly to the zope instance then add the special name at the end and pass in your SECRET word. It is a really slick tool that has revealed a few odd things on our servers. Good luck. -Allen Matthew X. Economou wrote:
The story thus far:
1. I installed DeadlockDebugger, but I couldn't get it to work. I kept getting errors when I tried to view its special URL, so I gave up on getting it working.
2. I upgraded to Python 2.4 as there were some fixes to threading support that haven't been backported to Python 2.3. Now, instead of hanging the entire instance when Zope begins spinning, it only hangs the Plone site (and I can access the Zope instance's root).
3. In my backtraces using GDB, I saw that Zope is stuck in a select loop waiting on an LDAP connection (I'm using LDAPUserFolders to authenticate to Active Directory). I don't know if the problem is in LDAPUserFolders, python-ldap, or OpenLDAP's client libraries.
What is especially frustrating is that this same software works fine on a different architecture.
The LDAP connection is in a CLOSE_WAIT state. The Zope instance sends a FIN/ACK and receives an ACK from the domain controller. The domain controller sends a FIN/ACK, but Zope never sends the final ACK to close the connection. Instead, it spins around a call to select(2) with the connection stuck in CLOSE_WAIT.
I will continue to try to get DeadlockDebugger working and follow up with Dataflake, the python-ldap-dev mailing lists, or with the OpenLDAP maintainers (as the case may be).
As always, your suggestions are most welcome.
Best wishes, Matthew