If the master host can no longer communicate with the FDM agent of a slave server but the heartbeat datastore responds, that server is still functioning from the perspective of vSphere HA. In this type of failure scenario, the slave host is considered isolated or partitioned from the network.

A server is considered isolated when it is no longer able to communicate with the master host’s heartbeat and the master can no longer ping the failed hosts management IP address. There is also the scenario where several hosts within the same cluster become isolated but can communicate among themselves through the management network. When this occurs, it is referred to as a network partition. In this type of failure scenario, a second host is elected as a master within the same partitioned network, so the result can mean that several different partitions each with its own master exist. When communication within the cluster is re-established, the cluster will return to having a single master with the other hosts returning to a slave role.

What happens to virtual machines when a host is isolated depends on the defined policy. This is referred to as the Host Isolation Response.

The policy options are:

The design decision to select an appropriate Host Isolation Response policy is based on a number of factors and configurations. Use Leave Powered On if you want your virtual machine to be accessed continuously by users and believe that an outage of the management network might not necessarily impact virtual machine traffic networks. You typically select Shut Down to so that your virtual machines only run in a stable environment. Note that Power Off is not a graceful shutdown for applications or operating systems.

In addition to these policy options, there is a general recommendation that where IP-based storage, such as iSCSI or NFS is employed as the protocol to access the heartbeat datastores, and the same physical connectivity is used for storage, virtual machines, and management networking (converged), that you modify the default option, and configure the host isolation response policy to Power Off or Shut Down. The reason for this is that any outage in this type of architecture is likely to affect all network segments across the converged infrastructure and not only the management IP space.