I've been monitoring AES for a number of nodes in a cluster and found that we have a large number of Database connections open (in some cases >2000).  After investigation it was found that these were not active connections but the total number of forced closed plus currently active connections (WAS has a policy enabled that will release an idle connection after 2 hours).  The question still remains, why doesn't pega release the connection.  An extensive analysis of the logs found a number of entries stating:

Not returning connection X for database "pegadata" to the pool as it previously encountered the following error.  It was also found that the error in occur also for the pegarules database.

When the database encounters an exception, shouldn't Pega always return the connection back to the pool rather than keep it open, considering (if WAS wasn't configured to force close the connection) the total resources required by Pega to keep the connection open?

We also see an ORA-01000: Maximum number of open cursors exceeded.  I presume this error is due to the force close of the connection without closing the cursor?