We had a 5-node training Cluster with SST method set to rsync on all nodes. When we switched to SST method mariabackup with rolling cluster restart and forcing SST the donor node stuck in state DONOR/DESYNCED.
If the donor was restarted everything was fine again. This we could reproduce more or less several times...
The donor also found out of DONOR/DESYNCED state when any of the other nodes was stopped (restarted) (maybe this helps you finding the problem...)