vault-fetch reads secrets from Hashicorp Vault and writes them to files where an application can read them.

vault-fetch is best used as an InitContainer in Kubernetes, as a shim for setting up deployments for applications and microservices whose source is not under your control. The best way to read from Vault will always be the app doing it natively – vault-fetch is a fallback for when that isn't an option.

1. Your application will need to have a Vault token that it can share with vault-fetch. Typically, this is done with another InitContainer, running before vault-fetch, which gets a token from Vault's Kubernetes authentication backend (e.g. sethvargo/vault-kubernetes-authenticator).

2. Provide vault-fetch with the VAULT_ADDR environment variable so it knows how to reach Vault, and VAULT_TOKEN_PATH to tell it where to find the Vault token. (You can also use VAULT_TOKEN to pass a token directly.)

3. Attach a few volumes to vault-fetch:

   • the volume containing the Vault token, probably generated from a previous InitContainer
   • the ConfigMap containing the vault-fetch configuration
   • if necessary, the in-memory EmptyDir where you will be writing secrets (you can use the first volume if you've mounted it read-write)
   • if necessary, the volume or Secret containing Vault's CA certificates

4. Mount the Vault token and secrets volume to your deployment's containers, and read secrets from Vault as real files!

vault-fetch can read its configuration as JSON, TOML, or YAML. By default it will try to read JSON from /opt/vault-fetch/config/vault-fetch.json; you can change the configuration path by setting the VAULT_FETCH_CONFIG environment variable.

• secrets is an array of objects with path and filename keys:
  • path (required, string) is the secret's path in Vault.
  • filename (required, string) is where to put it the secret's contents.
    • Start your filenames with the same path as your Kubernetes authentication Vault token's volume's path (e.g. /var/run/secrets/vaultproject.io) to use them later in your deployment containers without needing to define another volume.
    • Write to an in-memory volume rather than to disk, so that secrets are not persisted if the Pod errors.
  • field, template, and template_file (all optional, default: unspecified) control how your secret is written to file:
    • field (string or array of strings) is the name of the secret's field(s) whose contents you want put into filename and available to the output template.
    • template (string) is a string containing field name placeholders (e.g. ${field}) that can render the secret in place within a file.
    • template_file (filename) is a path to a file containing a template.
    • template and template_file may not be used simultaneously.
    • When template or template_file is not set:
      • If field is not specified, all of the secret's fields are put into filename as JSON.
      • Setting field to a string will dump that field to the file.
      • Setting field to an array of strings will dump each field to the file, in the order in which they are set in the array, separated by newlines.
    • When template or template_file is set:
      • If field is not specified, all of the secret's fields are available to the template.
      • Setting field to a string or array of strings will allow only that/those secret(s) to be replaced in the template.
  • critical (optional, boolean, default: false) controls whether vault-fetch exits if an error occurs while writing a particular secret. The default behavior (false) will cause it to be skipped but continue trying to write the remaining secrets.

You can also validate your config against the schema file.

{
  "secrets": [
    {
      "path": "secret/my-team/foo-app/some-secret",
      "filename": "/var/run/secrets/vaultproject.io/foo-app/secret",
      "field": "foo"
    },
    {
      "path": "secret/my-team/bar-app/secret-config",
      "filename": "/var/run/secrets/vaultproject.io/bar-app/config.json",
      "critical": true
    },
    {
      "path": "secret/my-team/baz-app/fancy-secret",
      "filename": "/var/run/secrets/vaultproject.io/baz-app/fancy",
      "template": "some_setting = '${field}'"
    }
  ]
}

Secrets are written one at a time in the order they are specified. You can abuse this functionality to write multiple Vault paths to a single file by writing secrets to a temporary filename and using the same value as a template_file for subsequent secrets. This behavior is not guaranteed, and may change in a future release.

Deployment:

initContainers:
  - name: vault-auth
    image: my-registry/vault-auth:latest
    env:
      - name: VAULT_ADDR
        value: https://vault.example.com:8200
      - name: VAULT_ROLE
        value: my-app-kubernetes-auth-role
    volumeMounts:
      - name: vault-volume
        mountPath: /var/run/secrets/vaultproject.io
  - name: vault-fetch
    image: aelindeman/vault-fetch:latest
    env:
      - name: VAULT_ADDR
        value: https://vault.example.com:8200
    volumeMounts:
      - name: vault-volume
        mountPath: /var/run/secrets/vaultproject.io
      - name: vault-fetch-config
        mountPath: /opt/vault-fetch/config
volumes:
  - name: vault-volume
    emptyDir:
      medium: Memory
  - name: vault-fetch-config
    configMap:
      name: my-app-vault-fetch-config

vault-fetch-config ConfigMap:

apiVersion: v1
kind: ConfigMap
metadata:
  name: my-app-vault-fetch-config
data:
  vault-fetch.json: |-
    {
      "secrets": [
        ...
      ]
    }

• The DEBUG environment variable enables verbose logging. This will print URLs of all Vault HTTP traffic to the console, which may include your Vault paths and fields, and may be picked up by a log collector. Use this option with caution. You can enable verbose logging by adding -e DEBUG=1 in Docker, or by setting adding an env into the vault-fetch container's Kubernetes config:

  env:
    - name: DEBUG
      value: "1"

• Check that your application's Vault policy allows reading from any values you've configured in your vault-fetch.json, and that those secrets exist in Vault.

• Remove any field and template settings and see what fields get written as JSON.

• Alex Lindeman