File systems

An agent pod is disposable — restart it and its working directory is gone. An AgentFS file system is a persistent volume in the workload cluster that an agent mounts, so a checkout, a dataset, or a pile of intermediate artifacts survives restarts, and several agents can work against the same disk. Find them under Infrastructure → Storage.

Why a file system

• Continuity. A long job that gets interrupted picks up where it left off instead of re-cloning and re-building.
• Sharing. Two agents on the same volume can hand work to each other — one prepares data, another consumes it.
• Getting data in and out. Upload inputs before a run and pull results after it, without a shell session.

Create one

New file system asks for a name, a size, and a maximum size, and provisions a PVC in the target deployment. The deployment has to be online — creation is tunnelled to its operator, not applied by the dashboard.

1apiVersion: autonomic.sh/v1alpha1
2kind: AgentFS
3metadata:
4  name: shared-workspace
5  namespace: adaptive
6spec:
7  size: 20Gi
8  maxSize: 100Gi
9  accessMode: ReadWriteMany
10  retain: true

The list reports the live state: status, requested versus actually provisioned capacity (they differ while an expansion is in flight), the backing PVC, and which agents currently mount the volume.

Mount it on an agent

Attach file systems from the agent's edit form, or declare them on the resource. Omit mountPath and the operator mounts at /mnt/fs/<name>.

1spec:
2  filesystems:
3    - name: shared-workspace
4      mountPath: /workspace/shared
5    - name: reference-data
6      readOnly: true

Editing an agent's mounts follows the omitted-preserves rule — send the list to replace it, send an empty list to detach every disk, omit it to leave them alone. The operator must advertise the mount-filesystems capability.

Browse & transfer files

Opening a file system in the dashboard gives you a directory browser. Listing is tunnelled to the operator, which execs it in a pod that has the volume mounted, so you can look inside a disk no agent is currently using. You can create folders and empty files, and the browser reports real usage — used, free, and total bytes read from df, not the requested size.

Grow & delete

• Resize is grow-only, and capped at the file system's maxSize. Shrinking is not possible — that is a Kubernetes constraint, not an Exo one.
• Delete is refused while agents mount the volume. Detach it from each agent first. With retain: true the underlying volume outlives the resource, so the data is still there to reclaim.

App storage

Apps provision their own volumes: an always-on workspace volume holding the code, and a database volume when the app runs a Postgres sidecar. Both appear on the App storage tab, where they can be grown the same grow-only way. They are managed by the app's lifecycle — deleting the app removes them unless the app was created with retention on.

From the CLI

exo fs list

exo fs create build-cache --size 20Gi --max-size 100Gi -n adaptive

exo fs ls build-cache:/artifacts

exo fs cp ./dataset.parquet build-cache:/data/dataset.parquet

exo fs cp build-cache:/artifacts/report.html ./report.html

exo fs resize build-cache --size 50Gi

exo fs delete build-cache