Ignored GTID domain IDs still appear in gtid_slave_pos

Hello,

I have an asynchronous replica set up to a Galera cluster with a load balancer in front. The replica may be connected to either cluster member.

I have configured wsrep_mode on, wsrep_gtid_domain_id to 4248433653, and the 2 node's (local) gtid_domain_id to 293176233 and 1908279884. On the replica, I have configured Replicate_Ignore_Domain_Ids: 293176233, 1908279884.

When setting up replication, I removed those domains from the slave position:
SET GLOBAL gtid_slave_pos="4248433653-293176233-12529601";
CHANGE MASTER TO master_use_gtid=slave_pos, replicate_ignore_domain_ids=(293176233, 1908279884);
START SLAVE;

Initially, replication comes up and I am able to switch it over between the 2 cluster node.

But after a rolling schema upgrade on the cluster, the local gtid appears in the gtid_slave_pos on the replica. Then, when failing the replication over to another node, the replication does not start up.

It is not a work-around to simply remove that domain id from the replication

I expect that GTID to not appear in the gtid_slave_pos as it is ignored.

Is this intended behaviour? And if it is, how can I avoid these domains to appear in gtid_slave_pos? Otherwise it is quite difficult to replicate asynchronously from a galera cluster.