Broker does not immediately close connection when client attempts SSL connection to non-SSL-enabled acceptor

When broker is running with non-TLS acceptor (sslEnabled=False) and client connects with TLS (ssl) enabled, socket will be opened for a very long time. By long time I mean until client is alive/connected to the broker.

Broker should immediately, after client connects, refuse any non-TLS communication and not behave as described above.

Steps to reproduce:

1. Start the broker with default configuration, all logging set to DEBUG
2. keep an eye on number of open sockets:

   watch -n1 'netstat -np | egrep -c "ESTABLISHED.*java"'
   

3. connect TLS/SSL configured client to the broker

   python aac5_sender.py -b amqps://<IP>:5671/ssl_test --msg-content "ABC"
   

4. Wait for some "random" time, and watch sockets being still open. (step 2)
5. Client produces following error approximately once per minute :

   ERROR {'cause' :'Transport error: amqp:connection:framing-error - SSL Failure: error:140780E5:SSL routines:SSL23_READ:ssl handshake failure'}
   

6. Kill/Close client.
7. Broker log

   12:29:55,399 DEBUG [org.apache.activemq.artemis.core.remoting.server.impl.RemotingServiceImpl] RemotingServiceImpl::removing connection ID -1842194997
   

Expected:
Qpid dispatch closes socket immediately, so broker should close socket as well.

Client connected to qpid dispatch immediately produces

ERROR {'cause' :'Transport error: proton:io - recv: Connection refused'}