Skip to main content

Run a Batch Job on a Pod and Retrieve the Results

Goal: rent a GPU Pod to run a one-off batch job (such as batch inference or data processing), write the results to a volume, download them from the File Browser, and finally clean up the resources to stop billing.

Cost: Pods are billed by resource specification × usage hours; volumes are billed by capacity × usage duration (billing starts on enablement).

Prerequisites​

Step 1: Create a volume and upload input data​

  1. Go to Cloud > Storage and click Create Volume. Choose a storage cluster and set the capacity (it can only be expanded later, never shrunk, so start from your actual needs).
  2. Click the newly created volume and click Manage Files in the Access tab to open the File Browser.
  3. Upload the input data your job needs (datasets, model files, scripts).

Step 2: Deploy a Pod and mount the volume​

  1. Go to Cloud > Pods, click Create Pod, and choose a GPU resource suited to the job.
  2. Deployment settings: choose an image; if the job needs credentials or other sensitive settings, store them in Secrets first and inject them as environment variables.
  3. In the mount storage step, select the volume created in Step 1; the system fills in the Mount Path automatically — note it down.
  4. Confirm the summary and click Deploy, then wait for the status to become Running.

Step 3: Run the job​

  1. Click the Pod, open the Connection tab, and launch the Web Terminal.
  2. Under the mount path you can read the files uploaded in Step 1 directly. Run your job script and write the output to a folder under the mount path.
  3. While the job runs, watch GPU utilization in Monitoring and follow output and error messages in Logs.

Step 4: Retrieve the results​

Go back to Storage > File Browser and download the job's output files directly — no file transfer through a Pod connection needed.

Clean up​

  1. Delete the Pod: delete it as soon as the job finishes to stop compute billing (enter the Pod name in the confirmation dialog).
  2. Handle the volume: once the results are downloaded and the data is no longer needed, delete the volume to stop storage billing. If you will run similar jobs later, you can keep the volume and remount it (billing continues while it is kept).
caution

A volume in the Attached state cannot be deleted; delete the Pod that mounts it first.