Dex is an identity service that uses OpenID Connect to handle authentication for Kubernetes. It connects to third party identity providers like Active Directory, LDAP, and GitHub. While the Kubernetes API server only supports a single identity provider and OIDC token issuer, the use of Dex allows to use identities from various providers when authenticating for Kubernetes.
This app is installed in Giant Swarm management clusters by default and is also ready to be deployed in workload clusters.
In addition to Dex itself, this app provides Dex K8s Authenticator, which helps to configure kubectl
for clusters authenticated to via Dex.
There are several ways to install this app.
You provide your configration via a custom values.yaml
file. Here is an example using the connector for Azure Active Directory. More connector examples can be found below.
isWorkloadCluster: true
services:
kubernetes:
api:
caPem: |
-----BEGIN CERTIFICATE-----
M...=
-----END CERTIFICATE-----
oidc:
expiry:
signingKeys: 6h
idTokens: 30m
customer:
connectors:
- id: customer
connectorName: test
connectorType: microsoft
connectorConfig: |-
clientID: <CLIENT-ID-SET-IN-YOUR-IdP>
clientSecret: <CLIENT-SECRET-SET-IN--YOUR-IdP>
tenant: <TENANT-SET-SET-IN--YOUR-IdP>
redirectURI: https://dex.<CLUSTER>.<BASEDOMAIN>/callback
Some notes:
.service.kubernetes.api.caPem
is the CA certificate of your workload cluster in PEM format. At Giant Swarm, you can retrieve this certificate via the kubectl gs login command, when creating a client certificate for the workload cluster. It ends up in Base46-encoded form in your kubectl config file. The CA certificate is required by Dex K8s Authenticator.
The redirectURI
in your connector configuration must contain the proper host name for Dex's own ingress. In the default form, it contains the workload cluster name (replace <CLUSTER>
with the actual name) and a base domain (replace <BASEDOMAIN>
with the proper base domain).
If you configure more than one connector, make sure to set a unique id
for each one. Be aware that this version of Dex is configured to prefix all user group names with the connector ID. So if your connector's id
is customer
, a membership in group devops
will appear as customer:devops
.
Example connector configuration for Keycloak:
- id: customer
connectorName: test
connectorType: oidc
connectorConfig: |-
clientID: <CLIENT-ID-SET-IN-YOUR-IdP>
clientSecret: <CLIENT-SECRET-SET-IN--YOUR-IdP>
insecureEnableGroups: true
scopes:
- email
- groups
- profile
issuer: https://<IDP_ENDPOINT>/auth/realms/master
redirectURI: https://dex.<CLUSTERID>.<BASEDOMAIN>/callback
Example connector configuration for GitHub:
- id: customer
connectorName: test
connectorType: github
connectorConfig: |-
clientID: <CLIENT-ID-SET-IN-YOUR-IdP>
clientSecret: <CLIENT-SECRET-SET-IN--YOUR-IdP>
loadAllGroups: false
teamNameField: slug
orgs:
- name: <GITHUB_ORG_NAME>
teams:
- <GITHUB_TEAM_SLUG>
redirectURI: https://dex.<CLUSTERID>.<BASEDOMAIN>/callback
Note:
<GITHUB_ORG_NAME>
is your GitHub organization name. For example, the part myorg
in https://github.com/myorg
.<GITHUB_TEAM_SLUG>
is the part of the team's URL representing the team name. For example, the part my-team
in https://github.com/orgs/myorg/teams/my-team
.The app is installed in workload clusters, via our app platform.
Before doing so, please create the following ConfigMap
resource in the namespace named after that workload cluster to provide the contents of your values.yaml
file.
apiVersion: v1
kind: ConfigMap
metadata:
name: dex-app-user-values
namespace: <CLUSTER>
data:
values: |
<content of values.yaml>
Then you either install the app via our web UI, or you'll create an App resource in the following format:
apiVersion: application.giantswarm.io/v1alpha1
kind: App
metadata:
labels:
app.kubernetes.io/name: dex-app
name: dex-app
namespace: <CLUSTER>
spec:
catalog: giantswarm-playground
name: dex-app
namespace: dex
userConfig:
configMap:
name: <CONFIGMAPNAME>
namespace: <CLUSTER>
Notes:
<CONFIGMAPNAME>
must be replaced with the name of the ConfigMap resource shown above.<CLUSTER>
is replaced with the name of your workload cluster.As a result, you should see Dex deployed in your workload cluster.
Dex app exposes a web interface, which is accessible over https. Therefore, it creates an ingress, which needs to be configured with a TLS certificate signed by a certification authority, which needs to be trusted by the browsers. The app consists of several components, which also need to be able to communicate with each other internally over https. So the certification authority signing the certificates needs to be trusted by the individual app components as well.
In case a custom certification authority is used, it needs to be exposed to the individual app components and set as trusted, otherwise the components will not be able to communicate with each other and the app may not work as expected. Based on the cluster setup, this can be achieved by providing an additional set of values to the app configuration:
ingress:
tls:
letsencrypt: false
caPemB64: "base64-encoded CA certificate"
ingress:
tls:
letsencrypt: false
trustedRootCA:
name: "name-of-the property-in-the-secret"
secretName: "name-of-the-custom-ca-secret"
letsencrypt
, a secret called dex-tls
will be created and propagated with the b64-encoded values provided by the user.
Alternatively, the user can manage the creation of this secret by themselves and enable its usage like so:ingress:
tls:
letsencrypt: false
externalSecret:
enabled: true
The following secret then needs to be applied to the namespace dex
is running in:
apiVersion: v1
kind: Secret
type: kubernetes.io/tls
metadata:
name: dex-tls
data:
ca.crt: ...
tls.crt: ...
tls.key: ...
In case the traffic to Dex needs to go through a proxy (for example when the app is installed in a private cluster), the individual components of the app need to be set up to use the proxy.
The proxy setup can be provided to the app in a specific section of the user values configmap or secret with the app configuration:
cluster:
proxy:
http: "https://proxy.host:4040" # hostname of the proxy for HTTP traffic
https: "https://proxy.host:4040" # hostname of the proxy for HTTPS traffic
noProxy: "kubernetes-api-ip-range" # comma-separated list of hostnames and IP ranges, whose traffic should not go through the proxy. # Kubernetes API IP range needs to be defined here in order for Dex to work correctly
In addition to a few pre-defined static clients Dex app supports the possibility to define custom static clients as well.
They need to be defined as an array of object in a specific property of the configuration yaml file called extraStaticClients
.
The structure of each custom static client object is exactly the same as in upstream Dex:
extraStaticClients:
- id: "client-id"
secret: "client-secret"
trustedPeers:
- "https://example.com"
public: true
name: "client-name-1"
logoURL: "https://example.com/logo"
- idEnv: "CLIENT_ID"
secretEnv: "CLIENT_SECRET"
redirectURIs:
- "https://example.com/redirect"
name: "client-name-2"
Notes:
id
and idEnv
properties are mutually exclusivesecret
and secretEnv
properties are mutually exclusivename
id
or idEnv
secret
or secretEnv
Extra static clients can also be configured as trusted peers of the pre-defined static clients:
Add the extra static client id to the list of trustedPeers
in the pre-defined static client:
staticClients:
dexK8SAuthenticator:
clientAddress: "dex.installation.basedomain.io"
clientSecret: "default-client-dex-authenticator-secret"
trustedPeers:
- "client-id"
extraStaticClients:
- id: "client-id"
name: "client-name-1"
secret: "client-secret"
It will produce the same configuration:
staticClients:
- id: dex-k8s-authenticator
name: dex-k8s-authenticator
secret: default-client-dex-authenticator-secret
trustedPeers:
- client-id
- id: client-id
name: client-name-1
secret: client-secret
public: true
Duplicities are prevented in case an ID of any additional trusted peer equals an automatically pre-populated trusted peer ID.
Giant Swarm is currently building the dex
app from a fork of the original project.
We implement additional logic which adds the connector id as prefix to user groups.
In order to update the image used in this chart it is currently needed to to do the following steps in our fork repo:
Then in this repo:
values.yaml
, ensure that values.schema.json
is updated to reflect all values and their types correctly.master#release#v<major.minor.patch>
, wait for the according release PR to be created, approve it, merge it.