Aquanode DocsPause and resume
Stop the hourly charge without losing the box. Resume later on whatever card is free.
Pausing a setup saves a version and then releases the machine. The hourly charge stops. The setup stays. Resume later and that state comes back on a fresh card.
This works on every provider Aquanode supports, because it is built out of two things every provider can do: capture the disk, then hand the machine back.
Manual pause
From the setup, use the pause action. From the terminal:
aq pause <name|id>Then bring it back by deploying one of its saved versions onto a fresh machine:
aq deploy --snapshot <id>--snapshot takes a numeric deployment id, or a synthetic ext-<backupId> for a
snapshot the standalone CLI took. It is not the vN version number. Find the id with
aq ls --all or from the setup in the console.
You choose the card it comes back on with the same --gpu, --gpus, --max-price,
and --provider flags that aq up takes.
A paused setup rents no machine, so there is no hourly cost while it sits. You pay only to hold what it saved, at $0.02/GiB per month.
Auto-pause
Turn on Auto-pause and the setup does it for you when you walk away.
| Threshold | What happens |
|---|---|
| 30 minutes with no GPU use | You get a warning |
| 60 minutes with no GPU use | A version is saved, the machine is released, the hourly charge stops |
Toggle it per setup in the console, or at launch with Pause it when it goes idle. From the terminal:
aq autopause <name|id> on
aq autopause <name|id> offTo view or change the thresholds on a running deployment:
aq idle status <deploymentId>
aq idle set <deploymentId> [flags]You can also set them at launch:
aq up --auto-pause --warn-after 30m --pause-after 60mIdle means no GPU use, not no SSH session
The timer watches GPU utilization. An open shell doing nothing does not keep the box alive, and a long training run with no one connected does not get paused.
Resuming somewhere else
The card you resume onto does not have to be the card you left. It does not have to be the same model, or the same provider, or the same country. It has to be a card that fits. That is the point of the setup being separate from the machine.
Your saved versions stay exactly as they are. Resuming reads from the lineage, it never overwrites it.
The boundary worth knowing
Who stopped the box changes what you get back
If you stop it, the setup is captured on the way out, so you can restore it anywhere.
If the provider kills it, there was no chance to capture anything on the way out. You are recoverable to your most recent saved version, not to the state the box was in when it died.
This is the reason auto-pause is worth turning on, and the reason it is worth taking a version by hand before a long unattended run. A scheduled save is the floor you fall back to.
Forcing it
If a setup is mid-sync and you need the machine released now:
aq force-detach <name|id>This breaks the lease even mid-sync, which means unsynced work can be lost. Use it when the alternative is worse.
To push a sync tick immediately rather than waiting for the next one:
aq sync-now <name|id>Tearing it down for good
aq down <name|id>That stops the rented machine. Saved versions are billed until you delete them, so if you are done with the setup entirely, clear its versions too.