We have a couple of Geographically disperse SQL 2012 Clusters running on Server 2012. The Availability groups are originally configured to have all nodes as possible owners, so when an initial fail or failover happens everything works as it should.
The problem comes on the fail back, once the AG fails over, it automatically modifies the Possible Owners to include just the secondary replica the AG failed to, making fail back impossible, unless you manually set the Possible Owners again to include the original nodes. However, from my understanding MS does not recommend making changes to AG's from within Failover Cluster Manager.
Any thoughts as to why this is happening and how to keep it from happening?