]
Michael Musgrove commented on JBTM-2423:
----------------------------------------
I added a line in orbportability/ORB.java to ensure that destroy is/has already been
called on our root poa wrapper. This means that if whoever called shutdown on our ORB
wrapper forgets to explicitly destroy it then we will do so guaranteeing that our
pre/post poa destroy hooks get invoked.
ORBRunner uses the orb after run() returns
------------------------------------------
Key: JBTM-2423
URL:
https://issues.jboss.org/browse/JBTM-2423
Project: JBoss Transaction Manager
Issue Type: Bug
Reporter: Michael Musgrove
Assignee: Michael Musgrove
ORBRunner.java starts an orb using orb().run() but then performs operations on the orb
after the run() method returns. According to the CORBA spec this is invalid:
{quote}
Once an ORB has shutdown, only object reference management operations(duplicate,
release and is_nil) may be invoked on the ORB or any object reference obtained
from it.
{quote}
Note that when the orb.run() method returns the orb has shutdown because, for the run
method, the spec states:
{quote}
This operation will block until the ORB has completed the shutdown process,
{quote}
This issue has arisen because of a change made to our fork of the jdk orb: in the jdk orb
shutdown method we join with all the orb runners. This results in deadlock:
# com.arjuna.orbportability.ORB.shutdown is a synchronized method and it calls shutdown
on the jdk orb;
# shutdown on the jdk orb notifies the ORBRunner thread which now tries to call back into
a synchronized method of com.arjuna.orbportability.ORB but is blocked because the monitor
is held
# at this point the jdk orb shutdown would normally then return allowing the
the ORBRunner thread to make progress but a recent change now means that the jdk orb
shutdown method performs a join() on the various ORBRunner threads