Failover with priority-backup, one clients behaves strange

We have 4 clients publishing to a virtual-topic, and 4 instances of an
application with 15 consumers consuming messages from that topic.

The topic is running on two servers (A and B) with a duplex networkconencts
from A to B.
Thw transportConnector is set up with setUpdateClusterClientsOnRemove and
setUpdateClusterClients to true.

The failover configuration is as follows:

Publishers:
failover:(tcp://A)?randomize=false涧Ⓔ=true똩麢=10ꚸꢮ⭲֧Ⓔ=true馬兹쨧鷧ⴍ=10馬兹쨧鷧ⴂ骛=1뚶鲐재놨=true

Consumers:
failover:(tcp://A)?randomize=false涧Ⓔ=trueꚸꢮ⭲֧Ⓔ=true馬兹쨧鷧ⴍ=10馬兹쨧鷧ⴂ骛=1

We deploy to half the instances of the modules at the time, so that we have
no downtime.

When stopping and deploying to A, we se that all publisher and consumers
goes to B, but when A comes up again, one of the four publisher instances
keeps publishing to B while the rest of the publishers, and all the
consumers goes back to A. This results in a lot of messages never being
consumed.

The logs shows that the publisher successfully connects to A after A is up
again, but this isntt reflected in where the mesages are published.

Can anoyone see any reason for this?