Guides

Storage

Connect S3, Azure Blob or Google Cloud Storage, or use ClockingCloud Global Volumes, and mount them into instances.

Your options

Provider	URI	Where data lives
Amazon S3	s3://bucket[/prefix]	In your AWS account. ClockingCloud caches it next to the GPU.
Azure Blob Storage	azure://container[/prefix]	In your Azure account. Cached next to the GPU.
Google Cloud Storage	gs://bucket[/prefix]	In your Google Cloud project. Cached next to the GPU.
ClockingCloud Global Volume	created by name	Stored by ClockingCloud and available in every region. Sized from 10 to 10,000 GB.

Create a Global Volume

Terminal

clockingcloud volume create renders --size 500
clockingcloud volume ls

Names are 2 to 39 lowercase letters, numbers or dashes, and unique within your account. Size is a whole number of gigabytes from 10 to 10,000. Deleting a Global Volume permanently deletes its data.

Connect a bucket

Store credentials as account secrets, then connect. Use a credential that is read-only unless jobs must write results back, and scope it to the one bucket or prefix.

Provider	Secrets read by default
Amazon S3	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, optional AWS_REGION
Azure Blob Storage	AZURE_STORAGE_CONNECTION_STRING
Google Cloud Storage	GOOGLE_APPLICATION_CREDENTIALS_JSON (the service account key JSON)

Terminal

clockingcloud secrets set AWS_ACCESS_KEY_ID
clockingcloud secrets set AWS_SECRET_ACCESS_KEY
clockingcloud volume connect s3://brand-assets --name assets --read-only

The bucket URI must start with s3://, azure:// or gs://. clockingcloud drive connect is accepted as an alias. Removing a connected volume only removes the link. Your bucket is never modified or deleted by ClockingCloud unless a job writes to it.

Mount into an instance

clockingcloud.yaml

volumes:
  - name: assets        # read training data and references
    mount: /data
    mode: ro
  - name: renders       # write generated output
    mount: /out

Or on the command line: clockingcloud deploy ... --volume assets:/data:ro --volume renders:/out. Mounts are available to setup steps and to the model process.

Referencing volumes in commands

Commands that take a data location, such as clockingcloud run --image and clockingcloud cp, use vol://<volume>/<path>.

Form	Meaning
vol://renders/2026/10	Path inside the volume named renders.
s3://bucket/key	Direct bucket path. Needs the matching secrets, not a volume.
global://renders	Accepted alias for a Global Volume named renders.
clockingcloud://global/renders	Accepted alias, as shown in the console.

Caching and consistency

Reads are cached on the node after first access. Changes you make directly in your bucket can take up to a minute to appear.
Writes to a mounted bucket are flushed when the file is closed and again when the replica stops. Do not stop a replica mid-write.
Two replicas writing the same file path is not safe. Write to per-request paths such as /out/$REQUEST_ID.
Egress charges from your cloud provider for reading your own bucket are billed by that provider, not by ClockingCloud.

Something unclear or missing? Email hello@clockingcloud.com.