Jim Rees wrote:
> Dale is running into a deadlock, not contention for the host lock.  His
> server never recovers once it hangs.

Dale:

I don't think that plockstat will be all that helpful in tracking
deadlocks although it would be very useful in helping to determine if we
are making progress in solving the host entry problem.

In order to track a deadlock, we need a core file containing all of the
memory of the process.   The technique I use is to examine the core file
(dump on Windows) in a debugger and examine which thread is holding
which locks and then construct a dependency graph of all of the locks
by the thread and function in which they were obtained.   This technique
allowed me to identity and remove the lock ordering deadlocks in the
Windows client.

The process is very time consuming but is quite satisfying when you have
identified the cause of your first deadlock.  Of course, after twenty or
thirty deadlocks your patience begins to wear thin.  :-)

Jeffrey Altman



Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to