// Benutzung · Konfiguration & Rollback
Rolling back to a checkpoint
Rolling back returns a location's config to a safe checkpoint you made earlier. The rollback is recorded as a new event — it never erases the history between then and now, so you can always see that it happened and undo it too.
Roll back
- Open the location (Locations → pick one).
- In Config history, find the checkpoint card and click Roll back.
- Choose a mode (below) and confirm.
You'll see a short summary of what changed, and the resources on the page update to match.
Who can roll back
Rolling back makes the same changes you could make by hand, so it needs the same rights at the location (globally, or through a role for this location):
| Mode | Needs |
|---|---|
| Safe | edit resources (resource.edit) and edit the location (location.edit) |
| Full reconcile | those, plus create and delete resources (resource.create, resource.delete) |
A rollback that would change the location's Collect from the core agent
setting (local_collector) also needs location.edit on all locations, as
changing it by hand does; without it, such a rollback is refused with 403 and
changes nothing.
The Roll back button shows only when you can run at least a safe
rollback, and Full reconcile can be picked only when you hold its rights
too. Creating a checkpoint needs location.create.
Safe vs. full reconcile
| Safe (default) | Full reconcile | |
|---|---|---|
| Settings/resources that changed since the checkpoint | restored | restored |
| Resources added since the checkpoint | left in place | removed |
| Resources deleted since the checkpoint | left absent | recreated |
- Pick Safe to undo drift without deleting or resurrecting anything. This is the safe default and what you want most of the time.
- Pick Full reconcile to make the location exactly match the checkpoint — accepting that resources added since will be deleted and ones removed since will come back. A deleted resource takes the resources discovered under it (VMs, CTs, Kubernetes objects) with it; a recreated one gets them back, as new resources, once its agent can collect it again (a Proxmox node needs its credential re-attached first).
- Full reconcile removes only resources it can tell were added after the
checkpoint. A resource older than the checkpoint that the checkpoint does not
hold (one imported or created before config history existed, if the
checkpoint was taken before the upgrade that recorded it) is left in place,
and the rollback's answer lists it under
kept_ids.
Good to know
- All or nothing. A rollback either applies completely or changes nothing. If it fails, the location is as it was; fix the cause and run it again.
- The location is locked while it rolls back. Edits, creates and resource
deletes in the location are refused for the few seconds a rollback takes,
with a note that it is being rolled back; try again when it is done.
Deleting the whole location is not held back: it goes ahead, and the
rollback changes nothing. The location page shows who started the rollback
and when the lock lapses on its own (two minutes after the rollback last
made progress), until it is lifted. If a rollback was cut off and its lock
is still in the way, anyone holding
location.createhere (globally, or through a role for this location) can click Lift lock; a rollback still running then changes nothing. - If the outcome is unclear, the history panel says so: rarely, the hub cannot tell whether a rollback was applied. Check Config history: a rollback that was applied shows there as its own event. Running it again is safe: if it was applied, it changes nothing. Until then, a resource it removed may keep showing its last state for a while, and one it recreated may show none for up to 5 minutes.
- Nothing is lost. The rollback is written forward as a new history entry; the changes it reverted are still visible in the timeline. To undo a rollback, roll back to a checkpoint from before it (or just re-apply your changes).
- Who collects the location is restored too. The location's settings come back as the checkpoint has them, including Local collector (collected by the hub's bundled agent instead of the location's own). If it changed since, the rollback moves collection back. A checkpoint from before that setting existed leaves it as it is now.
- A rollback can turn certificate checking back off. A resource's Allow invalid/self-signed TLS certs (on a check, Don't verify the TLS certificate or Don't judge the certificate's validity) comes back as the checkpoint has it. If it was ticked then and you unticked it since, the rollback ticks it again and the resource is once more reached without verifying its certificate. Check it after rolling back if you tightened TLS since the checkpoint.
- Credentials aren't restored. A resource recreated by a full rollback comes back without its credential — re-attach it via the resource's Edit form. Existing resources keep whatever credential they currently have.
- Extra vantage points need your permission there. A rollback that would
give a check back an extra vantage point it doesn't have now, at a location
where you can't create resources (
resource.create), is refused as a whole, and nothing changes. Vantage points a check still has need nothing. Vantage points at a deleted location, or on a check that now has a credential, are left out of what is restored. - A credential never moves to another host. If a resource now has a
credential and the checkpoint's address points at a different host, port
or scheme (
httpinstead ofhttpscounts), the rollback is refused. Detach the credential first. - A name in use elsewhere is not taken back. If another location now has the name the checkpoint holds for this one, the rollback is refused and nothing changes. Rename the other location first.
- Discovered resources are untouched. Auto-detected VMs/pods/nodes refresh from the live source and are never part of a rollback, in either mode: a rollback doesn't undo your edits of their tags or name, doesn't bring back one deleted since (discovery does, if it still exists), and doesn't remove one found since.