[OJB-developers] nasty bug in SequenceManagerFactory

[Georg S. <ge...@me...>]

Hi Thomas,

I think I found a nasty bug related to the SequenceManager which shows up
in multithreaded environments:

In the constructor of PersistenceBrokerImpl a SequenceManager is obtained
by the following line:

m_SequenceManager = SequenceManagerFactory.getSequenceManager(this);

If this is the first time, SequenceManagerFactory will instantiate a new
SequenceManger and give it the this-pointer. Subsequent calls to
getSequenceManager will return the same SequenceManager instance.

The problem is that once the PersistenceBrokerImpl is finished with its
tasks it will get returned to the Broker-Pool and is now available for
another thread, but it also forms part of the SequenceManager which will
get returned from the factory for new PersistenceBrokerImpl instances. The
result is that > 2 threads are accessing the same connection and wrecking
havoc on it.

My short-term solution is to create a PersistenceBroker at the beginning
of the program which I never use again. The real solution would be to
create a new PersistenceBroker in the SequenceManagerFactory the first
time it is being called. Unfortunately each call to create a new
PersistenceBroker calls the SequenceManagerFactory itself again and we are
getting into a nice circle (Stack Overflow). I suppose there should be
a constructor in PersistenceBrokerImpl that doesn't create a
SequenceManager.

Regards

Georg