Mount an SFS on a training job

Mount an SFS on a training job, also called a Training Cluster, so all its workers read and write the same files at a mount path you choose. Do this while you create the job.

Before you begin

• One or more SFS, created and ready; see Create an SFS. You can also create one from inside the job form.
• A training job you are creating, open at its Shared File-System step.

Choose the SFS to mount

On the Shared File-System step, pick one or more existing SFS. Each one you pick gets its own mount path field. To create a new SFS right from the form, select Create beside the SFS picker.

Set the mount paths

For each SFS you picked, type its mount path. A mount path starts with /, ends with a letter, digit, hyphen, or underscore, has at most one / in a row, and uses only letters, digits, hyphens, and underscores. Every mount path on the job must be different, across its datasets, SFS, and PFS together. If two are the same, the form refuses the job with “All mount paths (Datasets, SFS, PFS) must be unique.”

A Slurm job needs at least one shared file system. In Delhi, an SFS or a PFS counts. Outside Delhi, choose an SFS, because PFS is available in Delhi only.

Finish creating the job

Continue through the rest of the job form and submit it. Each SFS you picked is mounted, at the path you set, on every worker the job starts, so all of them read and write the same files.

To see which instances use an SFS later, see Monitor an SFS. If a mount path is refused, see Troubleshooting TIR Shared File System.