We are seeing database deadlocks for processes using a call activity.

The oracle error messages indicate that the failing transaction does various variable inserts and deletes, consistent with the behaviour expected at the end of the call activity.

In other words, the problem seems to occour whenever the call activity's 'called element' process returns.

Workaround: Adding 'async after' to the callactivity, and possibly 'async before' to the last node in the subprocess, seems to resolve the issue.