Rescale Files

concept · Reviewed 2026-07-31 · FEAT-0117 GA

Rescale Files is object storage. Every input you upload and every output a job produces lives here. It is durable and inexpensive, but it is not a filesystem you can mount. For that, see CFS.

What it's for

Use Rescale Files for:

• Input files you attach to a job or workstation.
• Job output you want to keep after a run ends.
• Data you reuse across many jobs.

Use something else when:

You need	Use
One shared filesystem mounted into many jobs at once	Cloud Filesystem (CFS)
Fast bulk transfer to or from your desktop	Transfer Manager
Scripted or resumable transfers	Rescale CLI

How it's organized

Files are presented as folders on the Files tab, with two spaces that behave differently:

• My Jobs — one folder per job, created automatically. Inputs and outputs, read-only once the job has run.
• My Library — your own space. Create folders, upload whatever you like, keep data that isn't attached to any job.

That distinction drives most of the behavior people find surprising, so it has its own page: browse your files.

Files are referenced, not copied

Attaching the same file to ten jobs does not store it ten times. Rescale stores one copy and points each job at it, so you're billed once.

The practical consequence: deleting a job doesn't necessarily delete its input files. If the input came from My Library, it stays there. See reuse files across jobs.

The four states a file can be in

A file moves through distinct states, and only the last one stops the storage bill:

State	Visible to you	Costs money
Active	Yes	Yes
Archived	Yes, in the archived view	Yes
In trash	Yes, in Trash	Yes
Deleted	No	No

Trashed files still cost money. This is the single most expensive misconception about Rescale storage: see the file lifecycle.

Getting data in and out

• Upload files — browser, CLI, and Transfer Manager, and when each is the wrong choice
• Download files — single files through to whole multi-run jobs

Where files live inside a running job

Inside a job, two directories behave differently and confusing them silently breaks parallel runs:

• work/ is private to the job. In a parallel run, each node sees its own copy.
• work/shared/ is visible to every node.

Stage shared inputs in work/shared/. If you put them in work/ expecting all ranks to see them, only the first will, and the rest read nothing or stale data.

Which cloud it's stored on

Rescale Files is backed by AWS S3, Azure Blob Storage, or Google Cloud Storage, depending on your workspace's region and configuration. Other S3-compatible providers aren't supported today.

Related

• Browse your files — My Jobs vs. My Library
• The file lifecycle — trash, archival, and what actually stops the bill
• Cloud Filesystem (CFS) — when you need a mounted filesystem
• Limits and policies — storage and transfer caps
• Archive and delete jobs — cleaning up at the job level