Pretty interesting new feature is implemented in Windows Server 2012 R2 failover clustering that allows you to manage how cluster database is updated.
Service responsible for this is called Global Update Manager. This service is responsible for updating the cluster database. In Windows Server 2012, you were not able to configure how these updates work, but in Windows Server 2012 R2 it is possible that you configure the mode of work for Global Update Manager.
Each time the state of cluster changes (for example, when cluster resource is offline) all nodes in the cluster must receive notification about the event, before the change is committed to the cluster database, by Global Update Manager.
In Windows Server 2012, Global Update Manager works in Majority (read and write) mode. In this mode, when change happens to the cluster, majority of cluster nodes must receive and process the update before it is committed to the database. When cluster node wants to read the database, cluster compares the latest timestamp from a majority of the running nodes, and uses the data with the latest timestamp.
In Windows Server 2012 R2, Global Update Manager can also work in All (write) and Local (read) mode. When working in this mode, all nodes in the cluster must receive and process the update before it is committed to the database. However, when the database read request is received, the cluster will read the data from the database copy stored locally. Since all roles received and processed the update, local cluster database copy can be considered as a relevant source of information.
Windows Server 2012 R2 also supports the third mode for Global Update Manager. This mode is Majority (write) and Local (read). In this mode majority of cluster nodes must receive and process the update before it is committed to the database. When the database read request is received, the cluster will read the data from the database copy stored locally.
In Windows Server 2012 R2, default setting for Hyper-V failover clusters is Majority (read and write). All other workloads in the clusters use All (write) and Local (read) mode. Majority (write) and Local (read) is not used by default for any workload.
