Pod Volume: Sharing Files Between Containers

Prerequisites: kubernetes

Containers in the same Pod share a network namespace, but their image filesystems are otherwise separate. A Pod volume is the explicit bridge that lets selected containers see the same files. Each container must mount the named volume into its own filesystem.

The rendered Grafana Pod contains this volume:

volumes:
  - name: sc-dashboard-volume
    emptyDir: {}

Both the dashboard sidecar and Grafana mount it at the same path:

volumeMounts:
  - name: sc-dashboard-volume
    mountPath: /tmp/dashboards

flowchart LR
  API[Kubernetes API\nselected ConfigMaps]
  S[Sidecar container\n/tmp/dashboards]
  V[(Pod emptyDir\nsc-dashboard-volume)]
  G[Grafana container\n/tmp/dashboards]
  API -->|watch and copy| S
  S -->|write file| V
  V -->|same mounted bytes| G

The two /tmp/dashboards paths look identical, but the important link is the shared volume name. A path inside a container is not automatically a host path and is not automatically visible to another container. Kubernetes mounts the same Pod-owned storage into both container mount namespaces.

What emptyDir means

Kubernetes creates an emptyDir when the Pod is assigned to a node. It survives an individual container crash because the Pod still exists. It is deleted when the Pod itself is removed from the node.

That lifecycle is correct here. The dashboard directory is a working cache, not the source of truth:

Git manifest -> ConfigMap -> sidecar copy -> emptyDir file

If the Pod is replaced, the new volume starts empty. The sidecar watches the Kubernetes API, finds the labeled ConfigMap, and writes the file again. A persistent volume would add durability where none is required. Grafana’s database uses the separate kps-grafana-longhorn PersistentVolumeClaim because that state has a different lifecycle.

Why the exact path matters

Three independent settings must agree:

The provider scans the root folder. The annotation adds the Devata subdirectory. The data key adds the filename. Grafana’s home setting points to the final result. A typo at any layer produces a valid Pod and a missing home dashboard.

Prove both containers see the same file:

pod=$(kubectl -n monitoring get pod \
  -l app.kubernetes.io/name=grafana \
  -o jsonpath='{.items[0].metadata.name}')
 
kubectl -n monitoring exec "$pod" -c grafana-sc-dashboard -- \
  ls -l /tmp/dashboards/Devata/devata-overview.json
 
kubectl -n monitoring exec "$pod" -c grafana -- \
  ls -l /tmp/dashboards/Devata/devata-overview.json

Official reference: Kubernetes volumes and emptyDir.