NPS cheat sheet

A production cluster has a minimum of three nodes; each node consists of a hypervisor, a Controller VM (CVM), and locally attached disks, and all CVMs together form one distributed system.

Service roles: Genesis starts and monitors CVM services; Stargate serves all vDisk I/O; Curator runs ILM and disk balancing; Cassandra stores cluster metadata; Zookeeper stores configuration and elects leaders.

Secondary roles: Prism/Medusa serves the UI and REST API, Chronos schedules tasks, Cerebro executes snapshots and replication, Heracles runs health checks.

Data units: an extent is the 1 MB checksummed unit of data, an extent group is 4 MB (four extents), a vDisk is the VM-visible disk, and a pDisk is a physical disk owned by the node's CVM.

Disks are presented only to the CVM on the same node; every other node reaches that storage over the internal cluster network, so no SAN or shared LUN is required.

RF2 keeps two copies of every extent and survives a single disk or node failure; RF3 keeps three copies and survives two simultaneous failures.

ILM tiers extents between SSD (hot) and HDD (cold) based on access frequency, while data locality keeps reads on the extent store of the node running the VM.

AOS and AHV are released together on the AOS version (for example AOS 6.x with AHV); Prism Central is versioned and upgraded independently of AOS.

Capacity math: RF2 usable is about 50% of raw, RF3 usable is about 33% of raw; EC-X 4+1 adds roughly 1.25x overhead and EC-X 6+2 roughly 1.33x.

Erasure coding (EC-X) is available only on RF2 clusters with enough nodes to stripe extents and parity, and it is applied to cold or inactive data rather than to hot write traffic.

Snapshots are redirect-on-write metadata captures; they are near-instant and consume space only as the original blocks are overwritten after the snapshot is taken.

A Nutanix snapshot is crash-consistent by default; application-consistent snapshots require VSS coordination inside the guest through Nutanix Guest Tools.

Async snapshot replication has a minimum RPO of one hour; NearSync replication lowers the RPO to a 1-15 minute range.

A protection domain is defined by a fault domain (disk, node, or rack) plus a replication factor; the replication factor determines which failures the data survives.

Rebuild after a component failure is automatic and prioritized by data hotness; cluster reserve capacity exists so a rebuild does not need free space beyond the replication overhead.

Checksums are computed per extent and validated on read by Stargate; detected silent corruption is repaired by fetching a good copy from another node's extent store.

AHV uses Open vSwitch as its virtual switch; the default bridge br0 carries management traffic for the CVM and the AHV host, and guest networks attach as separate OVS bridges.

Bond modes: active-backup and balance-slb require no switch-side configuration, while balance-tcp requires an 802.3ad LACP link aggregation group on the physical switch.

Guest VLANs are 802.1Q tagged by the virtual switch, so the physical uplink must be a trunk port carrying every VLAN used by VM networks.

Each CVM has its own IP address and each AHV host has a separate IP address; both normally sit on the same management VLAN and subnet.

virtio is the recommended NIC and disk driver for AHV guests because it gives the best throughput; e1000 is retained for operating systems without virtio support.

A single VM can have multiple vNICs attached to different networks and VLANs (multi-homing), each with its own MAC address and optional IPAM-assigned IP.

AHV networks are created in Prism as VLAN-backed networks, or as unmanaged networks when VLAN tagging is handled upstream of the host.

Nutanix Flow enforces distributed microsegmentation and firewall policy at the vNIC level, independent of physical switch ACLs.

VM lifecycle operations in AHV: create from a blank disk, image, or clone; power on/off; snapshot; clone; live migrate; delete; a clone source can be a VM, snapshot, or template.

Nutanix Guest Tools inside the guest enables VSS application-consistent snapshots, Self-Service Restore, file-level restore, and guest command execution from Prism.

AHV live-migrates running VMs automatically when a host enters maintenance mode, allowing hardware service with no VM downtime.

AHV HA restarts a failed host's VMs on surviving hosts; HA reservations reserve capacity on those hosts so the restart is guaranteed to have resources.

Affinity policies: VM-VM anti-affinity keeps selected VMs on different hosts, and VM-host affinity pins a VM to a defined set of hosts.

AHV permits CPU and memory overcommit, so the sum of configured VM memory can exceed physical host memory; HA reservation capacity is not overcommitted.

VM disks are vDisks inside a storage container; disks and NICs can be hot-added or removed while the VM is running if the guest OS supports it.

Console access is provided through Prism (VNC) or a serial console; templates plus categories allow standardized VM deployment at scale.

Prism Element is the UI and API for a single cluster; Prism Central is the multi-cluster manager that adds categories, projects, IAM, Calm, Flow, reports, and Insights.

Prism Central is deployed as a VM on a managed cluster; a single-VM deployment suits small environments and a three-VM deployment provides scale and high availability.

Clusters must be registered to Prism Central before it can manage them; registration traffic uses HTTPS on port 9440.

Prism Central is released and upgraded independently of AOS, so a PC version must be compatible with the AOS versions of every registered cluster.

Categories are key-value labels applied to VMs and other objects; protection, recovery, and security policies attach to categories rather than to individual VMs.

Projects combine categories, users, quotas, and networks so a team can self-service deploy VMs into a governed, isolated namespace.

IAM in Prism Central provides role-based access control (for example Prism Admin, Project Admin, Developer, Viewer) mapped to local, AD, or LDAP identities.

LCM (Lifecycle Manager) performs firmware and software upgrades, including AOS, AHV, and Prism Central, running pre-checks before a rolling execution.

NCC (Nutanix Cluster Check) is the health-check framework that validates services, metadata, disk health, and configuration; it runs from a CVM CLI or from Prism health views.

CVM CLI tools: nutanix (interactive shell), ncli (legacy API), acli (Acropolis API), allssh to run a command on every CVM, and hostssh to reach AHV hosts.

When the Prism UI is unreachable, check genesis (process manager), prism/medusa (UI), Cassandra and Zookeeper (metadata and configuration), and Stargate (I/O).

Genesis can restart services on a single CVM, but never restart or reboot multiple CVMs at the same time, because quorum and data availability depend on them.

Log collection is done with the Prism Generate Logs option or logbay collect on a CVM; the resulting bundle is uploaded to the Nutanix support portal for a case.

Nutanix services write logs under /home/nutanix/data/logs, and cluster status reports the state of each service across all CVMs.

Nutanix clusters depend on synchronized time; clock drift between CVMs or hosts causes authentication, certificate, and replication failures.

AOS upgrades are rolling and non-disruptive; LCM runs pre-checks (NCC results, free space, replication factor) before starting the upgrade.