Mounting Storage Volumes (FUSE & Shorthands)
Solti mounts remote cloud storage buckets directly into the job's container filesystem using high-performance FUSE drivers (s3fs, gcsfuse, blobfuse2).
Minimal Configuration
Only two fields are required to mount storage:
volumes:
datasets: # Volume name (the map key)
bucket: my-ml-datasets
mountPath: /mnt/r2
Volume Field Reference & Defaults
volumes is a map keyed by volume name — the key is the volume's unique identifier, so no separate name field exists (duplicate keys are rejected by YAML/JSON parsing):
| Property | Type | Required | Default Value | Description |
|---|---|---|---|---|
| (map key) | string | Yes | — | Unique volume name within the job. |
bucket | string | Yes | — | Target storage bucket or container name across S3, GCS, Azure Blob, or Cloudflare R2. |
mountPath | string | Yes | — | Absolute destination path inside the running container filesystem where data appears. |
connection | string | No | First matching connection | Project connection supplying credentials. If omitted, Solti automatically identifies the first active connection with access to the bucket. |
prefix | string | No | "" (Bucket root) | Subdirectory path prefix inside the bucket to mount, isolating mounts to specific folders. |
CLI Volume Shorthand
From the command line, the shorthand syntax keeps submissions minimal:
--volume my-ml-datasets/some/prefix:/mnt/r2
This is a concise shorthand for:
--volume name=my-ml-datasets,connection=my-r2-storage,bucket=my-ml-datasets,prefix=some/prefix,mountPath=/mnt/r2
Solti automatically resolves the bucket name, path prefix, and matching cloud storage credentials. See the CLI manual for details.