AshVault.KeyProviders.OpenBao.KubernetesAuth (AshVault v0.2.0)

Copy Markdown View Source

A supervised OpenBao token holder for the Kubernetes auth method.

Configured per provider with auth: {:kubernetes, opts} instead of :token:

config :my_app, AshVault.KeyProviders.OpenBaoTransit,
  address: "https://openbao.foundry.svc:8200",
  cacertfile: "/etc/openbao-ca/ca.crt",
  auth: {:kubernetes, role: "ashvault-foundry"}

Options

  • :role (required) — the auth/<mount>/role/<role> to log in as.
  • :mount — the auth mount path. Default "kubernetes".
  • :jwt_path — the projected service-account token. Default "/var/run/secrets/kubernetes.io/serviceaccount/token".
  • :refresh_fraction — log in again once this fraction of the lease has elapsed. Default 2/3.
  • :backoff_min / :backoff_max — milliseconds between failed logins, doubling from min to max. Defaults 1_000 and 30_000.
  • :login_timeout — milliseconds a caller waits for an in-flight login. Default 15_000.

Behaviour

One holder runs per provider module, started on first use under AshVault's own supervisor. It logs in with POST /v1/auth/<mount>/login, caches the client token, and logs in again — rather than calling renew-self — once :refresh_fraction of the lease has passed. A fresh login re-reads the service-account JWT every time (the kubelet rotates projected tokens), and is not capped by the role's token_max_ttl the way renewal is. A lease_duration of 0 is a token that never expires and is never refreshed.

The login runs in a task, so callers holding a still-valid token are never blocked by a refresh. Concurrent callers arriving with no valid token share one login.

A failed refresh keeps serving the old token until it actually expires, and retries with exponential backoff. Once there is no valid token, a caller inside the backoff window gets the last failure immediately instead of triggering another login. Every failure is AshVault.Errors.ProviderUnavailable with reason {:kubernetes_auth, reason}: retryable, never KeyDestroyed.

Secrets

The JWT is read, sent and dropped; it is never kept in state. The client token is kept in state, and format_status/1 redacts it (and the task reply that carries it) from :sys.get_status/1 and crash reports. Neither value, nor the login response body, ever reaches a log line or an error struct.

Summary

Types

Validated Kubernetes auth options.

Functions

Validate auth: {:kubernetes, opts} options, filling defaults.

The current client token for provider, logging in first if there is none.

Types

options()

@type options() :: %{
  role: String.t(),
  mount: String.t(),
  jwt_path: Path.t(),
  refresh_fraction: float(),
  backoff_min: pos_integer(),
  backoff_max: pos_integer(),
  login_timeout: pos_integer()
}

Validated Kubernetes auth options.

Functions

options!(provider, opts)

@spec options!(module(), keyword()) :: options()

Validate auth: {:kubernetes, opts} options, filling defaults.

Raises ArgumentError on a configuration fault: retrying cannot fix it.

token(provider, options)

@spec token(module(), options()) :: {:ok, String.t()} | {:error, Exception.t()}

The current client token for provider, logging in first if there is none.