VS-HCI Velarion Systems Docs

Docs/Updating

Releases

Versions, where they come from, and the two upgrade pipelines.

Software in the console shows what each part of the deployment is running and what is available.

Components

A deployment has several versions, and they are not one number:

Component Runs on
OS bundle Every node — the operating system underneath
Packages Every node — platform packages
Agent Every host — the process the control plane talks to
Orchestrator Controllers — the control plane itself 
Hyperion[vsnode1]> show version
Controller version: 2.0.0

A host’s agent version is on its detail page in Infrastructure.

Where versions come from

Two routes:

  • The version repository — Velarion’s software repository, which the deployment syncs available releases from. The host it uses is configurable under Software settings.
  • Upload — for sites with no route to that repository. See Air-gapped sites.

Available versions are listed; downloading one stages it, and it is then available to upgrade to. Downloading is not upgrading.

The two pipelines

Upgrades run as one of exactly two pipelines.

Orchestrator — os → packages → agent → orchestrator, across every controller. On a single node that is one node; on a cluster it is a rolling upgrade: each standby in turn, then a graceful failover, then the node that used to be active.

Hosts — os → packages → agent, one hypervisor at a time.

Both stage their artefacts onto the target node and hand the sequence to an upgrade manager that runs under the node’s own service manager. That is what lets it survive the reboot the OS stage triggers — the control plane watches progress; it does not hold the upgrade open over a connection that a reboot would drop.

Progress

Each node in an upgrade reports a status:

Status  
PENDING Not started
RUNNING In progress
REBOOTING Restarting
SYNCING Catching up after a restart
SUCCESS Done
FAILED Stopped with an error
SKIPPED Not applicable

SYNCING on a controller is the database catching up. A controller is not ready to be failed over to until it leaves that state — see High availability.

The current upgrade and the history are both available, and an upgrade can be cancelled.

Order

Warning — Upgrade controllers before hosts. The control plane is what drives host upgrades, and a newer agent talking to an older control plane is not a combination to discover during a maintenance window.

From the CLI

Hyperion[vsnode1]> request upgrade controller <version>

Before upgrading

  1. Take a backup and confirm it completed.
  2. Check cluster health — every controller ALIVE and SYNC OK. An upgrade that begins with a standby already behind has less margin than you think.
  3. Check capacity — hosts are emptied one at a time, so the rest of the zone must absorb their workloads.
  4. Read the release notes.

Next