Recently I had the opportunity to work on an Exchange 2010 Database Availability Group installation where nodes were installed in multiple data centers. Although not uncommon to have nodes installed in multiple data centers what was uncommon was the number of data centers and their geographical dispersion - in this case 5 different data centers. To make the environment even more challenging fully meshed networks were not deployed. It was this factor in the design that caused some unexpected failover conditions when certain network links were lost.
The following article written by my co-worker Jeff Hughes – from our platforms team – addresses failover cluster service handling of this type of failure.