This is from analysis of two jiras:

• Unable to receive messages from A-MQ 7 / Artemis broker
  https://issues.jboss.org/browse/ENTMQCL-370

• Install QPIDIT correctly in a Docker image
  https://issues.apache.org/jira/browse/QPIDIT-75

In both cases some send-receive tests hang when the receiver does not
receive any messages.

For the tests:

1. ER16 broker is created fresh with default configuration.

  # Delete existing
  [ -d my-broker ] && rm -rf my-broker

  # Create new broker
  ./bin/artemis create my-broker \
                --password admin \
                --user admin \
                --role admin \
                --allow-anonymous

2. Start a pcap trace.

  dumpcap -i any -w ~/Work/adverb/20170301-interop-patched-2

3. Run broker

4. Run qpidit interop test; wait for hang. Might hang early (test 7)
or much later (test 25 or so).

5. Stop trace and examine.

The case that got the most scrutiny is QPIDIT, Qpid Interop Test, where
test cases that link a ProtonCpp sender with a ProtonCpp receiver. Both
clients connect to the broker and attach to the same queue. A different
queue is used for each test pair.

In the network trace for the failed trace:

• the client sender sent the messages and the broker returns settled dispositions for them.
• the sender closes normally.
• the client receiver has attached and issued credit.
• the client waits and receives no messages. Every N seconds the client sends a keepalive empty frame.

Attaching to the broker hawtio console:

• the queue for the failed test has been created.
• the queue is empty. Message (count, ack'd, added) counts all zero.

  ---

Putting in some sleep() statements in the ProtonCpp Receiver client
sometimes coerces the tests to pass. Other times the tests fail with
ProtonPyton->ProtonPython test pair where the sleep() statements are absent.