Broker does not restart with binding to exclusive queue and Khepri activated

[acogoluegnes]

This has been detected by activating Khepri in the RabbitMQ AMQP 091 Java client test suite.

A test does the following:

ch.exchangeDeclare(x, "fanout");
ch.queueDeclare(q, false, false, true, null);
ch.queueBind(q, x, "");

Then stop_app / start_app the broker (this is part of the connection recovery test). The broker fails to restart with the following error:

error:badarg

    ets:lookup/2
        args: [rabbit_registry,{exchange,fanout}]
    rabbit_registry:lookup_module/2, line 57
    rabbit_exchange:type_to_module/1, line 542
    rabbit_exchange:serialise_events/1, line 85
    rabbit_exchange:serial/1, line 90
    rabbit_binding:-process_deletions/1-fun-0-/2, line 494
    maps:foreach_1/3, line 391
    rabbit_amqqueue:notify_queue_binding_deletions/1, line 1968

Adding a 1-second delay between stop_app / start_app in the test does not help. Doing start_app manually after the test execution does work though.

Tested on 4.0.6 and main.

Java test

[acogoluegnes]

Steps to reproduce the failure:

Open a terminal tab, start the broker with Khepri:

cd /tmp
git clone git@github.com:rabbitmq/rabbitmq-server.git
cd rabbitmq-server
make -C deps/rabbit run-broker RABBITMQ_FEATURE_FLAGS="+khepri_db"

Open another terminal tab, run the Java test:

cd /tmp
git clone git@github.com:rabbitmq/rabbitmq-java-client.git
cd rabbitmq-java-client
git checkout 559a3589b66cc6642baa238b5ced005e692e65af
make deps
./mvnw verify -P use-nio -Drabbitmqctl.bin=/tmp/rabbitmq-server/deps/rabbit/sbin/rabbitmqctl \
  -Dtest-broker.A.nodename=rabbit@$(hostname) \
  -Dit.test=ConnectionRecovery#clientNamedTransientAutoDeleteQueueAndBindingRecovery

The test does some AMQP operations, stop_app the broker and start_app fails.