Troubleshooting

SSH connection refused

If ssh root@host -p port returns "Connection refused" or times out:

1. Check if an SSH key is saved. Go to Account Settings → Security. If the SSH Public Key field is empty, paste your key and save. Then redeploy — keys are injected at deploy time, not retroactively applied to running instances.
2. Verify the instance is RUNNING. SSH is only available while the instance status is RUNNING. If it's PENDING, wait for provisioning to complete. If STOPPED, resume it first.
3. Confirm SSH was enabled at deploy time. If SSH was toggled off during deployment, the SSH server won't be running. Terminate and redeploy with SSH enabled.
4. Check your private key. Make sure you're using the private key that corresponds to the public key saved in your account. If using -i, point it to the correct file.
5. Instance was deployed before the key was saved. If you saved your SSH key after deploying the instance, the key won't be present. Terminate the old instance and deploy a new one.

Instance stuck on PENDING

Provisioning typically takes 15–60 seconds. If your instance remains PENDING for more than a few minutes:

• The selected hardware may be temporarily at capacity. Wait a few minutes — the system retries automatically.
• If it persists beyond 5 minutes, terminate the instance and try redeploying with a different GPU model or CPU flavor.
• No billing occurs while an instance is in PENDING state.

GPU model unavailable (GPU only)

GPU availability fluctuates based on demand. If the GPU model you want shows as unavailable in the catalog:

• The catalog refreshes every 15 seconds — availability can change quickly.
• Try a different GPU in the same performance class (e.g. if A100 80GB is unavailable, try A100 40GB or H100).
• If resuming a paused instance and the original GPU is unavailable, you'll be offered the option to migrate to a different available GPU.

JupyterLab not loading

If clicking the Jupyter link shows a blank page or a connection error:

• Wait for full startup. JupyterLab may take an additional 10–30 seconds to initialize after the instance reaches RUNNING status. Refresh the page.
• Check if Jupyter was enabled. If JupyterLab was toggled off during deployment, no Jupyter server is running.
• Try the password. If you see a password prompt, use the auto-generated password from the instance details panel.

Insufficient balance error

Deployment requires a minimum balance equivalent to one hour of the selected configuration. If you see this error:

• Top up your balance from Billing.
• Select a less expensive GPU model or CPU tier.
• Terminate unused instances to stop idle charges from draining your balance.

Data lost after termination

Container Disk and Volume Disk data is permanently deleted on termination — this is by design. To preserve data across instance lifecycles:

• Use a Network Volume — create one from the Network Volumes tab and attach it during deployment. Data on the volume survives termination.
• Download important files via scp, JupyterLab file browser, or push to a remote (Git, S3, etc.) before terminating.