Rob,
are you able to do a 'thread apply all bt' on the core. That will allow us to reason through your theory with you Carl. [email protected] wrote:
All, We've been using Qpid M4 and 0.5 internally, both with C++, and have been seeing occasional segfaults within the broker. They all seem to occur inside TopicExchange::isBound() (the one with 3 parameters, not 2). Looking at the access patterns in this function, and judging by the core dumps, it appears that multiple threads may access and modify the class member "bindings" simultaneously, which seems like it has the potential to cause a segfault. Since this is a race condition (if we're right), it's hard to say for sure that the problem has been resolved, but since doing the below, we've not had a segfault (roughly 5 hours so far, rather than every 2-3 hours or so). - We added locks to all accesses of bindings throughout the file. I believe the new ones were limited to isBound. - This may not be necessary at all, but to be safe (while we were investigating the problem), we changed all lock types from RWlock::ScopedLocks to Mutex::ScopedLocks, to avoid being tripped up by pthread (read/write) mutex semantics. We wanted to get the opinion of the experienced Qpid developers on this list before opening a bug, but again, so far it seems to be much more stable (we're hammering on Session::exchangeBound() in our testing, which may be why we're seeing it in the first place). I'm not sure if the other exchange types have the same problem or not. Thanks! -Rob --------------------------------------------------------------------- Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:[email protected]
--------------------------------------------------------------------- Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:[email protected]
