This document covers setting up the [Kubernetes OpenID Connect token authenticator plugin][k8s-oidc] with dex.
Token responses from OpenID Connect providers include a signed JWT called an ID Token. ID Tokens contain names, emails, unique identifiers, and in dex's case, a set of groups that can be used to identify the user. OpenID Connect providers, like dex, publish public keys; the Kubernetes API server understands how to use these to verify ID Tokens.
The authentication flow looks like:
1. OAuth2 client logs a user in through dex.
2. That client uses the returned ID Token as a bearer token when talking to the Kubernetes API.
3. Kubernetes uses dex's public keys to verify the ID Token.
4. A claim designated as the username (and optionally group information) will be associated with that request.
Username and group information can be combined with Kubernetes [authorization plugins][k8s-authz], such as roles based access control (RBAC), to enforce policy.
* Custom CA files must be accessible by the API server (likely through volume mounts).
* Dex is accessible to both your browser and the Kubernetes API server.
Use the following flags to point your API server(s) at dex. `dex.example.com` should be replaced by whatever DNS name or IP address dex is running under.
```
--oidc-issuer-url=https://dex.example.com:32000
--oidc-client-id=example-app
--oidc-ca-file=/etc/kubernetes/ssl/openid-ca.pem
--oidc-username-claim=email
--oidc-groups-claim=groups
```
Additional notes:
* The API server configured with OpenID Connect flags doesn't require dex to be available upfront.
* Other authenticators, such as client certs, can still be used.
* Dex doesn't need to be running when you start your API server.
* Kubernetes only trusts ID Tokens issued to a single client.
* As a work around dex allows clients to [trust other clients][trusted-peers] to mint tokens on their behalf.
* If a claim other than "email" is used for username, for example "sub", it will be prefixed by `"(value of --oidc-issuer-url)#"`. This is to namespace user controlled claims which may be used for privilege escalation.
## Deploying dex on Kubernetes
The dex repo contains scripts for running dex on a Kubernetes cluster with authentication through GitHub. The dex service is exposed using a [node port][node-port] on port 32000. This likely requires a custom `/etc/hosts` entry pointed at one of the cluster's workers.
Because dex uses `ThirdPartyResources` to store state, no external database is needed. For more details see the [storage documentation](storage.md#kubernetes-third-party-resources).
There are many different ways to spin up a Kubernetes development cluster, each with different host requirements and support for API server reconfiguration. At this time, this guide does not have copy-pastable examples, but can recommend the following methods for spinning up a cluster:
* [coreos-kubernetes][coreos-kubernetes] repo for vagrant and VirtualBox users.
* [coreos-baremetal][coreos-baremetal] repo for Linux QEMU/KVM users.
To run dex on Kubernetes perform the following steps:
1. Generate TLS assets for dex.
2. Spin up a Kubernetes cluster with the appropriate flags and CA volume mount.
3. Create a secret containing your [GitHub OAuth2 client credentials][github-oauth2].
4. Deploy dex.
The TLS assets can be created using the following command:
```
$ cd examles/k8s
$ ./gencert.sh
```
The created `ssl/ca.pem` must then be mounted into your API server deployment. Once the cluster is up and correctly configured, use kubectl to add the serving certs as secrets.
Finally, create the dex deployment, configmap, and node port service.
```
$ kubectl create -f dex.yaml
```
__Caveats:__ No health checking is configured because dex does its own TLS termination complicating the setup. This is a known issue and can be tracked [here][dex-healthz].
The `example-app` can be used to log into the cluster and get an ID Token. To build the app, you can run `make` in the root of the repo and it will build the `example-app` binary in the repo's `bin` directory. To build the `example-app` requires at least a 1.7 version of Go.
The default redirect uri is http://127.0.0.1:5555/callback and can be changed with the `--redirect-uri` flag and should correspond with your configmap.
