We have Always On availability groups with 3 nodes (Primary + 2 secondary) with Synchronized mode, manual failover on SQL 2012. Today we run a test and we disconnected 2 replicas to allow the primary site to take the primary role of the cluster and live without the other nodes. This was a test of disconnecting 2 out of 3 nodes (we have another node as file share). In order to start the cluster, we had to forced the quorum. After that the cluster seemed to be working properly when looking in the cluster managment tools, but the SQL in all 3 nodes (the disconnected nodes and the DR node) became inactive. The availability groups in the primary site were in a resolving status and all the DBs were in "Not synchronizing, Recovery pending" mode, and even when we enabled the other 2 nodes, the AG would not synchronize with each other. Eventually we had to drop the existing availability groups and recreate them when all nodes were back online, and only then did we get the DB back active and synchronizing.
My question is if this is a normal behave, and if this is the situation, that when only 1 node is active, the cluster gets inactive, and force quorum looses connectivity to DB, and the only thing possible to do is to recreate the AG, how are we meant to a failover, if 2 sites fails and only one secondary stays active. should we delete the AG and later when other sites come online recreate the AG, or is there another way to keep the AG and once the other sites come back online, they should resync automatically?