Configure the Pinniped Supervisor as an OIDC issuer

The Supervisor is an OpenID Connect (OIDC) issuer that supports connecting “upstream” identity providers to many “downstream” cluster clients. When a user authenticates, the Supervisor can issue JSON Web Tokens (JWTs) that can be validated by the Pinniped Concierge.

This guide explains how to expose the Supervisor’s REST endpoints to clients.

Prerequisites

This how-to guide assumes that you have already installed the Pinniped Supervisor.

Summary

When the Pinniped Supervisor is installed using the YAML files which are attached to the GitHub releases, then the following additional configuration is required before your end users can use the Supervisor:

1. You must create a new Service to expose port 8443 of the Supervisor pods, and you must configure your preferred means of HTTPS ingress to allow the Supervisor to receive traffic from outside the cluster.

   This is not included in the YAML files attached to the GitHub releases because there are many ways to control HTTPS traffic in Kubernetes clusters. By allowing you to configure this yourself, you can take advantage of your preferred solution.

2. You must configure the Supervisor to act as an OIDC provider by creating a FederationDomain resource. You must also create a TLS certificate for the Supervisor to use while serving these requests, and place it in a Secret for the Supervisor to read.

   This is also not included in the YAML files attached to the GitHub releases because there are many ways to create and manage TLS certificates and certificate authorities (CAs).

These steps are explained in detail in this guide.

If you would like to see a full working example of configuring the Supervisor, please refer to the sections regarding configuring the Supervisor within other tutorial:

• Concierge with Supervisor: a complete example of every step, demonstrated using GKE clusters

If you are using a different way to install Pinniped, such as 3rd party Helm Charts or the Pinniped that is integrated into VMware’s TKG product, then that method of installation may already include an opinionated ingress and TLS configuration. In that case, please refer to the documentation for your method of installation.

Exposing the Supervisor app’s endpoints outside the cluster

The Supervisor app’s endpoints should be exposed as HTTPS endpoints with proper TLS certificates signed by a certificate authority (CA) which is trusted by your end user’s web browsers.

It is recommended that the traffic to these endpoints should be encrypted via TLS all the way into the Supervisor pods, even when crossing boundaries that are entirely inside the Kubernetes cluster. The credentials and tokens that are handled by these endpoints are too sensitive to transmit without encryption.

The Supervisor only listens on an HTTPS port by default. Incoming traffic must use TLS. The only exception is for an advanced configuration style using a service mesh to deliver traffic into the Supervisor (discussed below).

Because there are many ways to expose TLS services from a Kubernetes cluster, the Supervisor app leaves this up to the user. Some common approaches are:

• Define a TCP LoadBalancer Service.

  In this case, the Service is a layer 4 load balancer which does not terminate TLS, so the Supervisor app needs to be configured with TLS certificates and will terminate the TLS connection itself (see the section about FederationDomain below). The LoadBalancer Service should be configured to use the HTTPS port 8443 of the Supervisor pods as its targetPort.

  This is the simplest way to expose the Supervisor’s endpoints outside the cluster. If you are trying out Pinniped for the first time, this is recommended during your trial. Before you move Pinniped into production, you may choose to continue using a LoadBalancer Service, or you may prefer to explore one of the more complex setups described below, depending on your networking requirements.

• Or, define an Ingress resource (or use the newer Gateway API).

  In this case, the Ingress typically terminates TLS and then talks plain HTTP to its backend. However, because the Supervisor’s endpoints deal with sensitive credentials, the ingress must be configured to re-encrypt traffic using TLS on the backend (upstream) into the Supervisor’s Pods. It would not be secure for the OIDC protocol to use HTTP, because the user’s secret OIDC tokens would be transmitted across the network without encryption. If your Ingress controller does not support this feature, then consider using one of the other configurations described here instead of using an Ingress. (Please refer to the paragraph above regarding the deprecation of the HTTP listener for more information.) The backend of the Ingress would typically point to a NodePort or LoadBalancer Service which exposes the HTTPS port 8443 of the Supervisor pods.

  The required configuration of the Ingress is specific to your cluster’s Ingress Controller, so please refer to the documentation from your Kubernetes provider. If you are using a cluster from a cloud provider, then you’ll probably want to start with that provider’s documentation. For example, if your cluster is a Google GKE cluster, refer to the GKE documentation for Ingress and the GKE documentation for enabling TLS on the backend of an Ingress. Otherwise, the Kubernetes documentation provides a list of popular Ingress Controllers, including Contour and many others. Contour is an example of an ingress implementation which supports TLS on the backend, along with TLS session proxying and TLS session pass-through as alternative ways to maintain TLS all the way to the backend service.

• Or, expose the Supervisor app using a Kubernetes service mesh technology (e.g. Istio).

  In this case, the setup would be similar to the previous description for defining an Ingress, except the service mesh would probably provide both the ingress with TLS termination and the service. Please see the documentation for your service mesh.

  If your service mesh is capable of transparently encrypting traffic all the way into the Supervisor Pods, then you should use that capability. In this case, it may make sense to configure the Supervisor’s HTTP port to listen on a Unix domain socket, such as when the service mesh injects a sidecar container that can securely access the socket from within the same Pod. Alternatively, the HTTP port can be configured as a TCP listener on loopback interfaces to receive traffic from sidecar containers. See the endpoints option in deploy/supervisor/values.yml for more information. Using either a Unix domain socket or a loopback interface listener would prevent any unencrypted traffic from accidentally being transmitted from outside the Pod into the Supervisor app’s HTTP port.

  For example, the following high level steps cover configuring Istio for use with the Supervisor:

  • Update the HTTP listener to use a Unix domain socket i.e. --data-value-yaml 'endpoints={"http":{"network":"unix","address":"/pinniped_socket/socketfile.sock"}}'
  • Arrange for the Istio sidecar to be injected into the Supervisor app with an appropriate IstioIngressListener i.e defaultEndpoint: unix:///pinniped_socket/socketfile.sock
  • Mount the socket volume into the Istio sidecar container by including the appropriate annotation on the Supervisor pods i.e. sidecar.istio.io/userVolumeMount: '{"socket":{"mountPath":"/pinniped_socket"}}'
  • Disable the HTTPS listener and update the deployment health checks as desired

  For service meshes that do not support Unix domain sockets, the HTTP listener should be configured as a TCP listener on a loopback interface.

Creating a Service to expose the Supervisor app’s endpoints within the cluster

Now that you’ve selected a strategy to expose the endpoints outside the cluster, you can choose how to expose the endpoints inside the cluster in support of that strategy.

If you’ve decided to use a LoadBalancer Service then you’ll need to create it. On the other hand, if you’ve decided to use an Ingress then you’ll need to create a Service which the Ingress can use as its backend. Either way, how you create the Service will depend on how you choose to install the Supervisor:

• If you installed using ytt then you can use the related service_* options from deploy/supervisor/values.yml to create a Service. This will expose the appropriate port.
• If you installed using the pre-rendered manifests attached to the Pinniped GitHub releases, then you can create the Service separately after installing the Supervisor app.

⚠️ Note: Do not expose the Service called pinniped-supervisor-api outside the cluster. That Service exists for a different purpose internal to the cluster, and it exposes a different port (10250). Instead, create another Service to expose port 8443, by using the ytt options mentioned above or by manually creating a Service as shown below.

There is no Ingress included in either the ytt templates or the pre-rendered manifests, so if you choose to use an Ingress then you’ll need to create the Ingress separately after installing the Supervisor app.

Example: Creating a LoadBalancer Service

This is an example of creating a LoadBalancer Service to expose port 8443 of the Supervisor app outside the cluster.

apiVersion: v1
kind: Service
metadata:
  name: pinniped-supervisor-loadbalancer
  # Assuming that this is the namespace where the Supervisor was installed.
  # This is the default.
  namespace: pinniped-supervisor
spec:
  type: LoadBalancer
  selector:
    # Assuming that this is how the Supervisor Pods are labeled.
    # This is the default.
    app: pinniped-supervisor
  ports:
  - protocol: TCP
    port: 443
    targetPort: 8443 # 8443 is the TLS port.

Example: Creating a NodePort Service

A NodePort Service exposes the app as a port on the nodes of the cluster. For example, a NodePort Service could also be used as the backend of an Ingress.

This is also convenient for use with Kind clusters, because kind can expose node ports as localhost ports on the host machine without requiring an Ingress, although Kind also supports several Ingress Controllers.

apiVersion: v1
kind: Service
metadata:
  name: pinniped-supervisor-nodeport
  # Assuming that this is the namespace where the Supervisor was installed.
  # This is the default.
  namespace: pinniped-supervisor
spec:
  type: NodePort
  selector:
    # Assuming that this is how the Supervisor Pods are labeled.
    # This is the default.
    app: pinniped-supervisor
  ports:
  - protocol: TCP
    port: 443
    targetPort: 8443
    # This is the port that you would forward to the kind host.
    # Or omit this key for a random port on the node.
    nodePort: 31234

Configuring the Supervisor to act as an OIDC provider

The Supervisor can be configured as an OIDC provider by creating FederationDomain resources in the same namespace where the Supervisor app was installed. At least one FederationDomain must be configured for the Supervisor to provide its functionality.

Here is an example of a FederationDomain.

apiVersion: config.supervisor.pinniped.dev/v1alpha1
kind: FederationDomain
metadata:
  name: my-provider
  # Assuming that this is the namespace where the supervisor was installed.
  # This is the default.
  namespace: pinniped-supervisor
spec:
  # The hostname would typically match the DNS name of the public ingress
  # or load balancer for the cluster.
  # Any path can be specified, which allows a single hostname to have
  # multiple different issuers. The path is optional.
  issuer: https://my-issuer.example.com/any/path
  # Optionally configure the name of a Secret in the same namespace,
  # of type `kubernetes.io/tls`, which contains the TLS serving certificate
  # for the HTTPS endpoints served by this OIDC Provider.
  tls:
    secretName: my-tls-cert-secret
  # Configure which identity providers (OIDCIdentityProvider,
  # ActiveDirectoryIdentityProvider, or LDAPIdentityProvider)
  # shall be used by this FederationDomain.
  identityProviders:
    # See the "IDPs on FederationDomains" Supervisor configuration guide
    # for details on how to configure this section of the FederationDomain's
    # spec.

How FederationDomains Work

Each FederationDomain can be used to provide access to a set of Kubernetes clusters for a set of user identities. A FederationDomain will allow the same set of users to authenticate into all Kubernetes clusters which choose to trust that FederationDomain to provide authentication services. Authenticating to a FederationDomain will only provide access to those clusters which choose to trust that FederationDomain, and will not provide access to any clusters that are trusting other FederationDomains. Therefore, FederationDomains are a means of providing authentication isolation.

You can create multiple FederationDomains in a single Pinniped Supervisor, as long as each has a unique issuer string.

When a user authenticates into any cluster that chooses to trust the FederationDomain for authentication, then the user has started a single sign-on session with that FederationDomain which will last approximately 9 hours. The user will not need to manually authenticate again into any cluster any cluster that chooses to trust the FederationDomain for authentication until the session expires. Behind the scenes, the user’s client (e.g. kubectl) will only be given short-lived access to each cluster (approximately 5 minutes), but will automatically refresh those credentials without the need for user interaction. During each refresh, the Supervisor will perform checks against the external identity provider to determine if the user’s session should be allowed to continue. A user may have active sessions with multiple FederationDomains at the same time, but each will require authenticating separately to start the single sign-on session with that FederationDomain.

Technically, each FederationDomain is a separate OIDC issuer which serves multiple REST endpoints to clients, such as the Pinniped CLI. The tokens issued by each FederationDomain are signed by that FederationDomain and cannot be used by any other FederationDomain, which provides the isolation properties of the FederationDomain concept.

Any Kubernetes cluster can choose to trust a FederationDomain to provide user authentication for that cluster by using the FederationDomain its OIDC issuer for JSON Web Token (JWT) formatted ID tokens. This can be done by installing the Pinniped Concierge on each cluster and configuring the Concierge to use the Supervisor for authentication on each cluster. Alternatively, it can be done by configuring the Kubernetes API server on each cluster to use the FederationDomain as its OIDC issuer.

When two FederationDomains use the same hostname in the spec.issuer, then:

1. They must use different paths in their spec.issuer URLs to differentiate them from each other.
2. They must use the same tls.secretName if they both configure a tls.secretName (see below for details of TLS configuration).

The Supervisor uses the hostname on the each incoming request to determine which TLS certificate to serve for that request, and then it uses the path to determine which FederationDomain should serve the request if there are multiple FederationDomains with the same hostname.

Configuring TLS for the Supervisor OIDC endpoints

If you have terminated TLS outside the Supervisor app as described in the section above for using a service mesh, then you do not need to configure TLS certificates on the FederationDomain. Otherwise, you need to configure the Supervisor app with a TLS certificate.

The TLS certificate for the Supervisor should typically be created for the DNS name or IP address that your end users will use to make requests to the Supervisor. This should be the same DNS name or IP address that you declared in the FederationDomain’s spec.issuer.

There are two places to optionally configure TLS certificates:

1. Each FederationDomain can be configured with TLS certificates, using the spec.tls.secretName field.

2. The default TLS certificate for all FederationDomains can be configured by creating a Secret called pinniped-supervisor-default-tls-certificate in the same namespace in which the Supervisor was installed.

Each incoming request to the endpoints of the Supervisor may use TLS certificates that were configured in either of the above ways. The TLS certificate to present to the client is selected dynamically for each request using Server Name Indication (SNI):

• When incoming requests use SNI to specify a hostname, and that hostname matches the hostname of a FederationDomain’s spec.issuer (case-insensitive hostname matching), and that FederationDomain specifies spec.tls.secretName, then the TLS certificate from the spec.tls.secretName Secret will be used.
• Any other request will use the default TLS certificate, if it is specified. This includes any request to a host which is an IP address, because SNI does not work for IP addresses. If the default TLS certificate is not specified, then these requests will fail TLS certificate verification and your end users will see the error message pinniped supervisor has invalid TLS serving certificate configuration.

Your ingress software may require special configuration to enable the inclusion of the SNI information from the original request into the requests that it makes to the Supervisor. Please refer the documentation for your ingress solution for details.

It is recommended that you have a DNS entry for your load balancer or Ingress, and that you configure the OIDC provider’s issuer using that DNS hostname, and that the TLS certificate for that provider also covers that same hostname.

You can create the certificate Secrets however you like, for example you could use cert-manager or kubectl create secret tls. They must be Secrets of type kubernetes.io/tls. Keep in mind that your end users must load some of these endpoints in their web browsers, so the TLS certificates should be signed by a certificate authority that is trusted by their browsers.

Next steps

Next, configure an OIDCIdentityProvider, ActiveDirectoryIdentityProvider, LDAPIdentityProvider, or a GitHubIdentityProvider for the Supervisor (several examples are available in these guides). Then learn how to configure a FederationDomain to use one or more identity providers. And finally, configure the Concierge to use the Supervisor for authentication on each cluster.