Infrastructure
Servers, availability zones and datacenters — and everything a single host exposes.
Infrastructure is the physical side of the deployment: the servers themselves, grouped into availability zones and datacenters.
Hosts
Every registered server, with its status, role and utilization.
Status
| Status | Meaning |
|---|---|
PENDING_APPROVAL |
Enrolled with a valid join token, not yet admitted. It may report in; the control plane will not send it commands |
PENDING |
Registered, waiting to start bootstrapping |
BOOTSTRAPPING |
Being prepared — packages, agent, keys |
ONLINE |
Healthy and accepting work |
OFFLINE |
Not reachable |
MAINTENANCE |
Deliberately excluded from scheduling |
DEGRADED |
Reachable but not fully healthy |
ERROR |
Failed in a way that needs attention |
Role
| Role | Runs |
|---|---|
HYPERCONVERGED |
Workloads and storage — the normal case |
COMPUTE |
Workloads only |
STORAGE |
Storage only |
ORCHESTRATOR |
Control plane only |
A host’s role can be changed from its detail screen. Moving a host out of a storage role has consequences for replicated storage — read Replicated storage first.
Maintenance mode
Toggling maintenance takes a host out of scheduling. Nothing new is placed on it, and depending on how you move workloads, running VMs are migrated off.
Use it before any planned disruption — a reboot, a firmware update, physical work. See Upgrading the fleet.
Hyperion[vsnode1]> host maintenance vsnode2 on
A single host
Clicking a host opens its detail screen, with tabs:
Overview
Hardware and live state — CPU model, cores and threads, memory, disks, kernel and OS version, whether KVM is enabled, agent version, and how long since the host was last seen. Below that, CPU/RAM and network I/O charts over the last six hours, a sensor panel, and the VMs running here.
kvmEnabled: false on a host that should run VMs means hardware virtualization is off in
firmware. Nothing will start until that is fixed.
Storage
The storage this host contributes and the pools defined on it. See Storage.
Network
The host’s physical interfaces, bonds, bridges and the uplinks it is a member of. This is where you confirm a host is actually attached to the uplink you think it is. See Uplinks and bonds.
Users
Accounts that can sign in to this host’s CLI. These are managed centrally and pushed to the host, so they keep working when the control plane does not — which is the point. See The VS-HCI shell.
Diagnostics
A curated set of read-only checks that run on the host and return their output. Safe to run at any time. See Diagnostics.
Packet capture
Capture traffic on a host interface and download it. For chasing a networking problem to ground.
Terminal
A shell on the host, in the browser. It opens the VS-HCI CLI, not a Linux prompt — the same restriction as SSH — and the session is audited when it opens and closes.
Availability zones
A zone is a group of hosts that share a failure domain, and the unit the platform places and restarts workloads within.
The Infrastructure screen shows each zone with the hosts in it and how much capacity is committed. A VM belongs to a zone; when its host fails, the platform restarts it on another host in the same zone.
That is the whole point: put hosts on separate power feeds, racks or rooms into separate zones, and a VM will not be restarted onto the failure it was just hit by.
See Availability zones.
Datacenters
The outermost grouping — a site. Most deployments have exactly one. Datacenters exist so that a larger estate can describe where things physically are.
Removing a host
Put it into maintenance so its workloads move off, then remove it. A host that also runs the control plane must additionally be removed from the HA cluster.
Warning — Do not simply power off a node and leave it registered. The cluster continues to expect it, and an absent controller still counts against quorum. Remove it properly.