I have the impression, that the internal topics subscribers in a network of brokers don't get aware of an application consumer moving from one node to another one when this happens frequently.
To be more precise: consider a network of brokers with four nodes. For a topic t, there are producers connected to all nodes and there is only one application consumer, let's say connected to node 1. Therefore, nodes 2 - 4 have node 1 as subscriber of topic t to forward all messages to node 1 for delivery to the actual consumer.
When the app uses the URI containing all four nodes and randomize=true, after a restart it will most likely be connected to a different node.
When this happens a few times, I can observe that the internal subscriptions don't get updated regarding the new connection of the application. After a while, there are weird cycles of node i being subscriber of node j that make no sense at all.
It ends up with topic messages not being delivered to the app anymore, probably because the route to the app just becomes too long regarding the number of hops. And I am not sure if I want to change anything in the config that allows more hops.
Does someone observe similar things?
Thanks a lot in advance for any comment!