# Cut over, check, roll back

Take a machine all the way, and know how to go back.

## The steps

1. On the wave's page, click **Launch the wave**, or let the wave leave when its window opens.
2. Follow the runbook's **Machines and phases** grid, or the **Machines** tab: status, import progress, machine created on the destination and points of attention. The page updates itself.
3. Keep an eye on the **What is left to do** card: a manual step to tick, a machine waiting for a decision, a machine that arrived and is to validate.
4. At cutover, Naoshift shuts the source down cleanly, or powers it off if the shutdown does not complete. The destination then imports its disks, and the machine starts if you asked for it.
5. Once the operation is over, the machine shows **Migrated**. Check it: its **End of migration** table, service delivered, network, backup.
6. Click **Validate** on its runbook card: you confirm it runs as expected. Nothing changes on the platforms; the machine shows **Validated**.
7. Find every action in **Action log**, under **Actions**.

## What is left of the source

The source stays **powered off** in vSphere: Naoshift never deletes it. As long as it exists, it appears in your vSphere source's inventory. Once the machine is validated, the **Retire the sources** stage of the journey reminds you: remove it yourself in vSphere when you no longer need it.

## Rolling back

A machine that failed, or that waits for your decision, has **Roll back** on its row. This button puts the source back the way the migration found it:

- the source is **started again**, only if the migration had powered it off;
- the imported machine must be **powered off** first: it may carry the same MAC address as the source, and two machines running on the same network would fight over the address;
- the imported machine is **kept** to understand what happened. Deleting it is a separate action.

Restarting the source is an action like any other: its outcome is read on it, in the action log.

A machine already **Migrated** no longer has this button: the migration succeeded. If it does not deliver the service, power it off on the destination, then start the source again from **Inventory**, in that order.

### A machine waits, without moving

*Waiting for* names what it is waiting for: a blocking runbook step to tick, the cutover window, or its turn, since the destination limits how many machines run at once.

### A machine shows "Needs attention"

A step failed after the source was shut down, and nothing was undone: the machine waits for your decision. The reason is on its row. Go back to the source with **Roll back**, or finish by hand on the imported machine.

### Rollback is refused

The imported machine is still running. Power it off on the destination, then click **Roll back** again.
