Mounting a file with Helm into a Kubernetes pod using a ConfigMap

Background

One of our Helm charts deploys an HTTP server which requires a config file to be provided as a command-line argument. The docker image for the HTTP server only exposes the main command to start the server, so we need to pass the argument to the container command, as well as mount the ConfigMap into the docker container.

kind: Deployment
# ...
spec:
  # ...
  template:
    metadata:
      annotations:
        checksum/config: {{ include (print $.Template.BasePath "/application-config.yaml") . | sha256sum }}
        # ...
    spec:
      volumes:
      - configMap:
          name: application-config
        name: application-config
      containers:
      - image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
        command:
        - "./bin/start.js"
        args:
        - "--config"
        - "/etc/app/config.json"
        # ...
        volumeMounts:
        - name: application-config
          mountPath: /etc/app
          readOnly: true

The ConfigMap defining the config.json is somehow manually "crafted" from a bunch of chart values.

kind: ConfigMap
# ...
data:
  config.json: |
    {
      "name": "{{ .Values.app.name }}"
      // ...
    }

However, we recently introduced a new configuration file format, which does not required to pass the --config command-line argument anymore. Instead, the config file is automatically loaded by the HTTP server following certain naming conventions (see cosmiconfig package).

Problem

The config file is now defined in the source code of the project folder and not crafted together within the ConfigMap.

Therefore, the question arises: How can we mount the config file from the filesystem into the running docker container of the Kubernetes pod?

Solution

After several trials, I was able to make things work by leveraging the ConfigMap and the --set-file option of the Helm command.

The ConfigMap is still necessary, as we need to mount the config file into the docker container. The only difference is that we need to mount a single file and that the content of the ConfigMap should be automatically read from the file on disk.

Helm provides a --set-file option that we can use to read the content of the config file from disk. The content is then assigned to a template variable applicationConfigJson that we can reference into the ConfigMap:

kind: ConfigMap
# ...
data:
  config.json: |
{{ .Values.applicationConfigJson | indent 4 }}

Then Helm command to deploy the chart should be updated as following:

helm upgrade --install \
  --set-file applicationConfigJson="./path/to/application-config.json" \
  # ...

Finally, the Kubernetes deployment should be updated as well by removing the command + args option and by specifying the subPath to the single file when mounting the volume:

kind: Deployment
# ...
spec:
  # ...
  template:
    metadata:
      annotations:
        checksum/config: {{ include (print $.Template.BasePath "/application-config.yaml") . | sha256sum }}
        # ...
    spec:
      volumes:
      - configMap:
          name: application-config
        name: application-config
      containers:
      - image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
        # ...
        volumeMounts:
        - name: application-config
          mountPath: /app/application-config.json
          subPath: application-config.json
          readOnly: true

Conclusion

It was a bit tricky to find this solution but at the same time I learned a bunch of new things:

• mounting a single file using the combination of mountPath and subPath options
• using ConfigMaps is easy and very helpful for mounting volumes
• using the --set-file option to assign the content of the file to a template variable

Happy Helming and deployments! 🚀