A vSAN stretched cluster is a deployment model in which two or more ESXi host servers are part of the same logical cluster but are in separate geographical locations. The vSAN stretched cluster feature enables synchronous replication of data between sites, which allows a complete site failure to be tolerated. Stretched clusters extend the concept of fault domains to data center awareness domains.
Both server virtualization and vSAN are managed from a single vCenter Server platform. This single interface is used to enable capabilities such as Distributed Resource Scheduler (DRS). DRS is highly recommended when using stretched cluster functionality on VxRail. It provides the initial placement and migration of the VMs in accordance with the host and VM affinity rules.
Stretched clusters are deployed across two sites in an active/active configuration. Each stretched cluster must include one witness host. The witness must reside on a third site with independent paths to each data site. All three sites share the same vCenter, but the witness must not be on the same cluster as the data site hosts. Fault domains provide the core functionality of a vSAN stretched cluster. A fault domain consists of one or more vSAN hosts, grouped according to their physical location in the data center. The maximum number of fault domains in a vSAN stretched cluster is three. The first fault domain is referred to as the preferred site, the second is the secondary site, and the third fault domain is the witness host site. The witness host may be a VM or a physical host. This host monitors the datastore availability and makes decisions when the network connection between the two data sites is lost. The network latency between the two data sites must not exceed 5 milliseconds. The latency between the data sites and the witness site depends on the number of objects in the vSAN stretched cluster, however it must be less than or equal to 100 milliseconds. The following figure shows a basic stretched cluster deployment: