If an exception is thrown in a CMP entity bean with container-managed transactions, a Dirty Connection is left behind that is never cleaned up (or at least in 15 minutes). We've set inactivity-timeout in the data-sources and tweaked with all the transaction settings, all to no avail. Is anyone experiencing this?
It does this with Oracle (running Oracle's driver as well as Merant's), PostgreSQL, and SQL Server (running Merant's driver). The biggest issue is that the dirty connection in Oracle and SQL Server is holding a lock to the row that caused the exception, preventing any further access, actually deadlocking the server (in PostgreSQL, further updates to that row go through just fine). Or at least this is what we _think_ is happening, as there's no better explanation that we've found. Jeff.
