Skip to content

GPU Service Storage

Based on GPU Service Storage Types, GPU Service Storage lets you allocate persistent storage and mount it across multiple GPU Service Instances.

This is useful for data, models, and other files that need to be shared between instances.

GPUStack manages multiple Kubernetes clusters and provides a unified interface for sharing GPU Service Storage across them.

Under the hood, each storage item is realized as a Kubernetes PersistentVolumeClaim on the target cluster.

Browse Storage

Navigate to the GPU Service > Storage page to browse all storage and its details.

Screenshot: Storage list

You can filter storage by name.

Adding Storage

On the Storage page, click Add Storage to open the creation form.

Screenshot: Add Storage form

Fill in a Name, select a Storage Type (defined on the GPU Service Storage Types page), and set the Capacity (GB). Then click Save.

Editing Storage

After creation, only the display name can be changed.

Note

Editing more of the storage configuration is planned for a future release.

Deleting Storage

Click Delete on a storage item and confirm. The storage is then removed from the list.

Note

If the storage is still attached to a GPU Service Instance, its cluster-side PersistentVolumeClaim is not removed right away. Because deletion follows a deferred policy, the PersistentVolumeClaim is retained until the associated instance is deleted, and only then is it cleaned up.

Warning

Deletion of the cluster-side PersistentVolumeClaim is currently deferred. After deleting a storage item, avoid recreating one with the same Name (not Display Name); a leftover PersistentVolumeClaim could otherwise overwrite the new definition. This limitation will be addressed in a future release.