Here is our solution to this problem.

When restarting the 3rd primary broker, we noticed that messages were accumulating in the internal queue between the 2nd and the 3rd broker ($.artemis.internal.sf.amq-cluster.).

When looking at these messages in the console, we found 1 business related message duplicating. This business message was sent via the console and was supposed to be sent on one of our existing addresses but was supposely sent on the internal queue by error. In the listing of messages in the console, we could see this message duplicating but with the same messageId.

Our solution was to purge the internal queue and as soon as we did that, the 3rd primary broker was available again and synced with its backup server instantly.