Comment 10 for bug 1917645

I did more investigation while creating an oslo.messaging test case that reproduces the problem.

1) the rabbit driver uses kombu and the kombu interface allows defining max_retries for the connection attempt too but the rabbit driver does not use it. So the rabbit driver is fixable.

2) the kafka driver uses confluen_kafka and the producer interface[1] of that module is defined as asynchronous. Today the kafka driver simply ignores the retry parameter[2] for message sending. So fixing the kafka driver is not that simple.

[1] https://docs.confluent.io/platform/current/clients/confluent-kafka-python/html/index.html#confluent_kafka.Producer.produce
[2] https://github.com/openstack/oslo.messaging/blob/feb72de7b81e3919dedc697f9fb5484a92f85ad8/oslo_messaging/_drivers/impl_kafka.py#L290