There is an intermittent hang on client-connection shutdown occurring between the Python client and clustered c++ brokers.  Presumably, this will happen with other clients as well.

This has never been seen on a stand-alone broker.

The trace shows the following:

    Client                       Server
      |                             |
      |-- conn.close -------------> |
      |                             |
      | <--------- conn.close-ok -- |
      |                             |

From this point forward, the client is waiting for the TCP connection to close.  The TCP connection remains in the ESTABLISHED state (via netstat -t).

The server sends a conn.close-ok but does not close the TCP socket.

This is intermittent.  On a continuous run, it will run correctly for dozens of times before causing a hang.