Backups, and the proof

Whether a backup was any good is found out by restoring it, and that is exactly what nobody does — because restoring means a second machine, a name, a network and the nagging feeling of having broken something. So Phillux does the round itself.

Download the ISO Making one

The backup page in Phillux, with the search for a guest

Making one

A backup can be made of a VM or a container, by hand or on a schedule, and it lands on a storage that carries the content type backups — which is how it ends up on the network storage rather than on the same SSD as the machine it protects.

Everything long-running is a background task with progress and a log, so closing the tab or signing out does not stop it.

The round, walked automatically

A verification restores the backup under a name of its own, starts it, asks something inside whether it answers, and throws the copy away again. What happened is written down.

The copy gets no network card. That is not a detail, it is the whole reason this can run on a machine that is in use: a restored guest carries the address of the original — and with a VM the MAC as well. Started side by side, the same address stands on the network twice, and the one that stops working is the original.

Nothing is lost by taking the card out. A container is asked through the host, a VM through its guest agent, and neither of those goes over the network.

Three verdicts, never blurred

answered — something inside the guest ran a command and replied. That is proof the filesystem came up and an init got as far as a shell.
ran — it started and was still running after the waiting time, but nothing inside could be asked. For a VM without a guest agent that is all there is; it is worth having, and it is not the same as the first.
failed — it did not come up, or the restore itself did not work.

A green tick that means less than it looks is worse than a plain “could not ask”, so the two are never mixed into one.

Restoring for real

A restore puts the guest back under its own name or under a new one. Restoring is a separate permission from making and from deleting — the three are not one right, because they are not one risk.

What a backup does not replace is the layer underneath: an array survives a disk, a backup survives a mistake, and neither survives the other's job. Both together is the answer.

What to do with the verdict A verification that says ran instead of answered usually means the guest has no agent, not that the backup is bad. Installing one turns a maybe into a yes — and the same channel is what makes an answer possible on a lab network with no route out at all.