Provider maintenance on your pod
A provider can announce downtime for the machine that runs your pod. When they do, the pod page shows
a yellow Maintenance Notice banner. If your pod was already rented when the provider set the
notice, you also get an e-mail: "Maintenance on your Lium pod <name> on <date>", or "Your Lium
pod <name> will be removed on <date>" for a removal. There are two kinds of notice, and the banner
answers the same three questions for each.
Scheduled maintenance​
The banner reads "Provider maintenance on this machine: <start> to <end> (<N> minutes)". The
window is at most 60 minutes and the provider must announce it at least 24 hours ahead.
What happens​
During this window the machine can go offline. SSH, Jupyter and your running processes stop. Billing continues during the window.
What survives​
When the window ends we begin restarting your pod for you, within 10 minutes. Files in your pod's local
volume on this machine (/root on most templates) stay. An external Volume attached at
/mnt stays too. Running jobs, memory and files outside the local volume and /mnt are lost. The
container is created again from your template, so a package you installed with apt or pip outside
the local volume must be installed again.
The local volume is the folder your template keeps on the machine's own disk. An external Volume is
storage you created in your account and attached to the pod at /mnt.
What to do​
Before the start time: save your work in the local volume or download it with lium scp. Stop long
jobs or make them resumable. After the end time: log in again and continue. Nothing else is needed.
If the pod shows REBOOT_FAILED, or is not back one hour after the end time, reboot it from the pod
page. If that fails, use the Discord Support button on the pod page.
Permanent removal​
The banner reads "The provider is removing this machine on <date and time>". The provider must
announce a removal at least 48 hours ahead.
What happens​
Your pod is deleted within 10 minutes after the removal time and cannot be restored. Billing stops when the pod is deleted. The removal runs from a task that wakes every 10 minutes, so the pod can stay up for a few minutes past the announced time.
What survives​
An external Volume attached at /mnt stays in your account. Completed backups stay.
Everything else on the pod is deleted with it.
What to do​
Before the removal time: download everything you need with lium scp. Or copy it to an external Volume,
or run a backup from the Backups tab. Then rent a new pod on another machine. You can attach the same
external Volume to it. You can also restore a backup into it.
Why providers do this​
A provider owns the machine and sometimes has to take it down: a firmware or driver update, a power or cooling job in the data center, a hardware swap, or the machine leaving the marketplace. The notice period is the supported way to do that while a pod is rented. The provider picks the start time and the length in the provider portal, and can add a reason, which the banner shows.
Penalties and your bill​
A provider who announces the downtime in time is not penalised for it: while a notice period is in effect, the penalty for a node that goes silent during a rental runs as a dry run: the event is recorded and nothing is withheld. A provider who takes a rented machine offline without a notice is penalised.
Your bill does not depend on the notice. Billing runs while the pod is RUNNING, REBOOT_PENDING
or REBOOT_FAILED, and a maintenance window is at most 60 minutes. If your pod came back broken or
stayed down much longer than the window, ask in Discord support with the pod name
and the times from the banner.