Sequence Caching Failing in JTA Managed Environments with PostgreSQL

[Scott Parkerson <sc...@parkerson.net>]

About a year ago, there was a bug (OPENJPA-2196) that I contributed a patch
to that deals with cases where OpenJPA's sequence caching cannot be used if
the native sequence in the database is not owned by the role connecting to
the database. This patch was included in OpenJPA 2.2.2.

Since then, I've started using JTA-managed transactions in my container
(the container being JBoss Fuse ESB, using Aries JPA/JNDI/JTA), and have
hit the following snags with my previous fix:

1. When the attempt to ALTER SEQUENCE ... INCREMENT BY fails, it basically
hoses the entire transaction, causing the next thing (which is to get the
next value in the sequence) to fail because the transaction is now invalid
and must be rolled back.

2. Trying to work around this using either ConnectionFactory2Name or the
non-jta-data-source configuration items in my persistence.xml file seems to
never matter, as ALL native sequences in OpenJPA are of type
TYPE_CONTIGUOUS, and thus it will always choose the managed
(jta-data-source or ConnectionFactoryName) methods to attempt to modify the
sequence. I cannot see where it attempts to suspend the transaction, either.

Perhaps there is a workaround, but I cannot see it. Does anyone else have
any ideas on what could be done to make this work?

Thank you,
Scott Parkerson