Nutanix's preferred way of handling metro failures. Decisions are automatic, split brain is avoided, and a lost site is distinguished from a broken network link.
Read more: Nutanix docs
In the Ultra Transcenders books
Each book explains Witness in context, with comparison tables and the common traps.
Terms in this definition
- LSDOU
The sequence in which Group Policy is processed: the local policy first, followed by site, domain and organisational unit policies. The nearest, last-processed setting takes effect unless Enforced or Block Inheritance alters that.
Related terms
- AHV Metro Availability
Combines a witness, cross-cluster live migration and synchronous replication with zero RPO to give AHV both high availability and disaster recovery. If no witness is present, replication stays synchronous but failures are handled by hand or after a timeout.
- Continuous Availability VCF Operations model
Keeps VCF Operations monitoring running through the loss of an availability zone. The VCF Installer can't create it; instead an existing simple or high availability deployment is converted later, dividing the analytics nodes into two fault domains (one per zone, each in a separate VCF Instance) that hold each object on a cross-zone node pair, with a witness at a third site.
- Dependency
Anything the design needs that is provided by a different team, vendor or system and lies outside its control, for example corporate DNS and NTP, a witness site or jumbo frames end to end. Each should be confirmed and tracked, since a dependency that may not arrive is a risk.
- Inter-site link
Connects both data sites of a stretched vSAN cluster and carries traffic between them. If this connection is lost, service normally continues at the preferred site, working with the witness.
- RTT
The latency of a packet's journey there and back. For synchronous replication Nutanix wants less than 5 ms between clusters; for volume groups, less than 200 ms to the Witness Service.
- Stretched cluster model
Cluster model spreading one vSphere cluster across two availability zones, so losing a zone does not stop workloads. vSAN stretching needs a third-fault-domain witness, inter-zone RTT under 5 ms and bandwidth of 10 Gbps or more; workload domain clusters can be stretched only after the management domain.
- vCenter High Availability
Three nodes (Active, Passive, Witness) running vCenter in an active-passive pair, cutting downtime from hardware or host failures and during patching.
- VCF Operations analytics cluster
Where VCF Operations keeps and processes its collected data. Nodes act as primary, data, replica (high availability) or witness (continuous availability), and static IP addresses are needed on the primary and replica.