Hi,
I/my group have been using AG's in production since CTP1 of SQL Server 2012. We're very well versed in AG behaviors but we have 1 environment that's picked up an odd misbehavior that does not consistently occur.
Some background. The cluster is small and has 2 sites: 2 nodes at site A, 1 node at site B. Node at site B has nodeweight=0x0. Quorum disk exists at site A. 2 FCIs exist between nodes at site A. 2 instances exist on node at site B. 2 AGs are configured; one from each FCI to an instance at site B. Overall, not entirely complex.
The issue we've run into is during patching. We orchestrate hands-on patching for our SQL servers to minimize any possible downtime. We're in .com retail, so we're very sensitive to downtime, planned or otherwise. We patch the inactive node, reboot it, then at a scheduled time we failover into it. The issue occurs post-failover.
Once the failover completes, sometimes the AG doesn't change owner, thus the AG never comes up and the status is resolving. There are no messages about the AG in the SQL errorlog. When reviewing the cluster behavior, we notice that both AGs have an owner of the previous node. We cannot fail the service over to the other node, as it is well known that SQL Server will set the possible owner tenet of the service to the intended operator.
We've got another latent issue - that I wont delve into on this post - that causes these services to failover every once in a while. Interestingly, in that scenario, we've never had an issue with it failing over to a new node successfully. This problem seems to be centered around proactive, manual failover events. What gives?
Update: Windows 2008R2 Ent Sp1 w/ latest/greatest, SQL 2012 Ent Build 6540