Troubleshooting
Common issues when launching and running QuickPods, and how to fix them.
Can’t open the launch wizard
Symptom: clicking Launch Pod shows a verification prompt instead of the wizard.
Launching requires a verified account. Complete identity verification and try again. See KYC.
GPU is unavailable
Symptom: the GPU you want shows 0 Available, or launch fails with a GPU-unavailable error.
- On the Instance Type step, the button reads Join Waitlist for sold-out GPUs. Click it to queue — you’ll be notified when capacity frees up, and you’ll see your queue position.
- Alternatively, tick Show All GPUs to compare, then pick a different GPU type or reduce the Number of GPUs.
- Consider a fractional GPU (a smaller percentage of a card) if a whole card isn’t free — see Fractional GPUs.
- For guaranteed dedicated hardware, see the GPU Marketplace.
Insufficient balance
Symptom: an insufficient-balance prompt appears when you click Launch Pod, or a running pod is auto-stopped.
- Top up your wallet, then launch again. See Wallet & Billing.
- Review the Summary panel’s total cost before launching, and reduce GPU count, CPU, or memory if needed.
- Enable auto-debit to avoid auto-stops when balance runs low.
Pod stuck in creating or pending
A pod moves creating → pending → running. If it stays in pending:
- The requested GPU capacity may not be free yet — reduce the GPU count/fraction or join the waitlist.
- Check the pod’s Container Logs and any failure reason shown on the pod.
- Verify your wallet has enough balance.
Pod failed
A failed pod shows a red banner with the failure reason. Click Relaunch to try again.
- Out of memory (OOM): the relaunch dialog offers to increase the pod’s memory before retrying. Bump it and relaunch.
- Image pull failed: check the Container Image name and tag. For private images, confirm the Registry URL, Registry Username, and Registry Password/Token are correct.
- Startup command errors: check your Entry Point Command and the logs.
Can’t connect over SSH
- SSH keys can only be added during pod creation. If you launched without a key, relaunch (or launch a new pod) with a key selected.
- First-time connections need a one-time SSH config entry. Click Configure SSH on the pod detail page and follow the steps for your OS. See SSH & PowerShell.
- Use the exact command shown in the SSH Access card, replacing
<key_name>with your private key file, and make sure the key has the right permissions (chmod 400). - If SSH still fails, use the Web Terminal (Open Terminal) as a fallback — it needs no keys.
Can’t reach an exposed service
- Confirm the port is listed under Exposed Ports & Endpoints. If not, click Edit and add it.
- If the endpoint says “Endpoints are being provisioned…”, wait a moment and refresh.
- Make sure your app is actually listening on that port inside the container (check the logs).
- Use the HTTPS URL shown for the port:
https://<subdomain>-<port>.cloud.podstack.ai.
Notebook password not shown
The auto-generated notebook password is shown once on the Notebook Access card. If you didn’t copy it, relaunch the pod to get a fresh one, or use the Web Terminal / SSH instead.
Lost data after stopping a pod
Data written to the container’s own disk can be lost on stop or delete. Keep datasets, checkpoints, and outputs on an attached volume mounted at /data. See Storage & Data.
MLOps section isn’t visible
MLOps features may be gated by a feature flag or account level. If experiment tracking, the model registry, or related sections don’t appear, contact support.
Still stuck?
- Check the pod’s Container Logs and Infrastructure Stats on the detail page.
- Review Manage & Scale for lifecycle actions.
- Reach out through Support.