Use Filestore agent volumes with self-managed GKE workloads

If you run standard AI agent benchmarks, developer environments, or automated pipeline workloads on Google Kubernetes Engine (GKE), you can use standard Kubernetes PersistentVolumeClaim (PVC) and Deployment resources directly with Filestore agent volumes.

This pattern uses familiar Kubernetes resource definitions without requiring custom sandboxing controllers or DaemonSet resources.

Before you begin

Complete the cluster and CSI driver setup in Set up GKE environment for Filestore agent volumes.

Create a PersistentVolumeClaim

Create a manifest named agent-pvc.yaml requesting storage from the volume-pool-sc StorageClass you created earlier:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: agent-workspace-pvc
  namespace: default
spec:
  accessModes:
    - ReadWriteMany
  storageClassName: volume-pool-sc
  resources:
    requests:
      storage: 1Gi

Apply the manifest:

kubectl apply -f agent-pvc.yaml

Verify that the PVC transitions to Bound:

kubectl get pvc agent-workspace-pvc

The output is similar to the following:

NAME                  STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS     AGE
agent-workspace-pvc   Bound    pvc-12345678-abcd-ef01-2345-6789abcdef01   1Gi        RWX            volume-pool-sc   5s

The Filestore CSI driver provisions a new volume within your volume pool in under one second and binds the PVC.

Deploy an agent workload

Create a Deployment manifest named agent-deployment.yaml that mounts the bound PVC:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: agent-worker
  namespace: default
  labels:
    app: agent-worker
spec:
  replicas: 1
  selector:
    matchLabels:
      app: agent-worker
  template:
    metadata:
      labels:
        app: agent-worker
    spec:
      containers:
        - name: worker
          image: busybox:latest
          command: ["/bin/sh", "-c"]
          args:
            - |
              echo "Agent worker started. Workspace mounted at /workspace:"
              ls -la /workspace
              echo "Writing benchmark results..."
              date > /workspace/run-status.txt
              cat /workspace/run-status.txt
              sleep infinity
          resources:
            requests:
              cpu: "250m"
              memory: "256Mi"
            limits:
              cpu: "500m"
              memory: "512Mi"
          volumeMounts:
            - name: workspace-storage
              mountPath: /workspace
      volumes:
        - name: workspace-storage
          persistentVolumeClaim:
            claimName: agent-workspace-pvc

Apply the manifest:

kubectl apply -f agent-deployment.yaml

Verify the deployment and file access

  1. Check that the deployment pod is running:

    kubectl get pods -l app=agent-worker

    The output is similar to the following:

    NAME                            READY   STATUS    RESTARTS   AGE
agent-worker-6789bcdef-ghijk    1/1     Running   0          30s

  2. Check container logs to confirm successful file operations:

    kubectl logs -l app=agent-worker -c worker

    The output is similar to the following:

    Agent worker started. Workspace mounted at /workspace:
total 8
drwxr-xr-x    2 root     root          4096 Sep 14 12:00 .
drwxr-xr-x    1 root     root          4096 Sep 14 12:00 ..
Writing benchmark results...
Mon Sep 14 12:00:05 UTC 2026

Because Filestore agent volumes support ReadWriteMany (RWX) access, multiple replicas or distinct pods can concurrently mount the same PVC for collaborative agent workflows.

What's next