Object Storage

Object Storage provides a durable and scalable way to store files, datasets, model artifacts, checkpoints, logs, and other unstructured data outside the lifecycle of a pod.

Unlike local pod storage, data stored in an Object Store can persist independently and be reused across deployments.

Object Storage is useful when you need a shared, persistent storage layer for workloads that may restart, scale, or move across compute instances. It helps separate compute from storage, making deployments easier to manage, reproduce, and recover.

Usecases for Object Store

Object Storage is best suited for:

• Storing large files and artifacts
• Model weights and checkpoints
• Dataset storage and retrieval
• Batch processing inputs and outputs
• Logs, reports, and generated files
• Sharing artifacts across pods or deployments

For latency-sensitive workloads, databases, or applications that require strict file system semantics, use a persistent volume, block storage, or a POSIX-compliant file system instead.

Create an Object Store

Follow these steps:

1. Go to Storage -> Object Storage page of the portal, and click "Add new bucket"
2. A dialog will pop-up, fill in all the details in the form and click "Add bucket"

Mount an Object Store

Mount the Object Storage to your Deployed Pod to make its contents accessible from inside the runtime environment.

Follow these steps for mount:

1. Go to the Deployments page and choose on which Deployed Pod you want to attach an Object Storage.
2. On the Deployed Pod, click "Storage" button and there you'll find an option to attach the bucket:

3. Select the Object Storage bucket to mount on the Pod. On successful completion you should the attached bucket and its mount path in the "Storage" section details of the pod.

4. Mount directory on the Deployed Pod:

Mount dir on pod: /mnt-workspace

Unmount an Object Store

Unmount the Object Storage when it is no longer required by the deployed pod. This helps prevent accidental writes, stale mounts, or dependency issues during redeployment.

Before unmounting, ensure that:

• No active process is reading from or writing to the mount path
• Important files have finished syncing
• The application is not depending on the mounted path
• Any temporary files have been copied or cleaned up

Cautions & Limitations

Avoid using Object Storage for workloads that require:

• High-frequency small file updates
• File locking
• Atomic rename operations
• Low-latency random reads and writes
• Database storage
• Heavy concurrent write operations to the same files