# Run an update campaign

Update a cluster without a sleepless night: one node at a time, checked before each node.

## Before you start

- The on-premise deployment of Naoshift, and an up-to-date collector.
- The **Updates and upgrades** and **Node maintenance mode** capabilities, open at the three locks.
- On Proxmox VE, the node update account, set up with the Naoshift team; on XCP-ng, a Xen Orchestra action account that administers the pool.

## The steps

1. Open **Monitoring › Vulnerabilities**, then a CVE that affects your hypervisors: **Create a campaign** prepares the campaign, exposed nodes ticked. You can also start from **Actions › Campaigns › New campaign**.
2. Check the nodes or pools, the target version, the maintenance window and the **Pause at (failures)** threshold.
3. **Create and request approval**. Another person approves it with **Approve and launch**.
4. When the window opens, the campaign starts. Open it: each node shows its version before and after, and its state.
5. Click a node: each step of the frozen runbook appears with its time — pre-checks, maintenance and evacuation, installation, reboot watched from another node, live check, exit from maintenance.

On an XCP-ng pool, the campaign starts Xen Orchestra's native Rolling Pool Update, follows it, then checks each host.

## Keeping control

- **Pause**, **Resume**, **Stop after this node**: a campaign never stops in the middle of a node.
- At the failure threshold, it pauses on its own.
- **Retry the failed step** resumes the operation at that step, without redoing what succeeded.

> After the campaign, Naoshift requests a collection. A targeted CVE is reassessed at the next collection: never declared fixed on its own.

### A cluster is greyed out, "its gateway does not declare the update gestures yet"

The collector of this source is too old: update it, and it announces the actions at its next connection.

### A node is refused before it leaves

The pre-checks stopped it, and the cause shows: not enough N+1 capacity to evacuate the node, quorum, an ongoing incident. Remove the cause, then retry the step.

### The node rebooted and an alert went out

It should not: the node's monitoring is paused during its maintenance. If it goes out anyway, report it to support with the campaign's name.
