SAP Kyma with dynamic OIDC credentials with HCP Terraform
HCP Terraform already supports dynamic credentials with Kubernetes providers with AWS and GCP platforms.
I have extended this support to SAP BTP, Kyma runtime clusters with SAP Business Technology Platform.
Let's see how...
1. Configure Kubernetes
Configure HCP Terraform OIDC identity provider with SAP Kyma cluster.
SAP Kyma supports the gardener's oidc-shoot-extension, thus, effectively allowing for an arbitrary number of OIDC providers in a single shoot cluster.
The following has to be done upfront during the kyma cluster bootstrapping phase.
OpenIDConnect_HCP
locals {
OpenIDConnect_HCP = jsonencode({
"apiVersion": "authentication.gardener.cloud/v1alpha1",
"kind": "OpenIDConnect",
"metadata": {
"name": "terraform-cloud"
},
"spec": {
"issuerURL": "https://app.terraform.io",
"clientID": "terraform-cloud",
"usernameClaim": "sub",
"usernamePrefix": "-",
"groupsClaim": "terraform_organization_name",
"groupsPrefix": ""
}
})
}
resource "terraform_data" "bootstrap-tfc-oidc" {
triggers_replace = {
always_run = "${timestamp()}"
}
# the input becomes a definition of an OpenIDConnect provider as a non-sensitive json encoded string
#
input = [
nonsensitive(local.OpenIDConnect_HCP)
]
provisioner "local-exec" {
interpreter = ["/bin/bash", "-c"]
command = <<EOF
(
KUBECONFIG=kubeconfig-headless.yaml
NAMESPACE=quovadis-btp
set -e -o pipefail ;\
curl -LO https://dl.k8s.io/release/v1.31.0/bin/linux/amd64/kubectl
chmod +x kubectl
while ! ./kubectl get crd openidconnects.authentication.gardener.cloud --kubeconfig $KUBECONFIG;
do
echo "Waiting for OpenIDConnect CRD..."; sleep 1;
done
./kubectl wait --for condition=established crd openidconnects.authentication.gardener.cloud --timeout=480s --kubeconfig $KUBECONFIG
crd=$(./kubectl get crd openidconnects.authentication.gardener.cloud --kubeconfig $KUBECONFIG -ojsonpath='{.metadata.name}' --ignore-not-found)
if [ "$crd" = "openidconnects.authentication.gardener.cloud" ]
then
OpenIDConnect='${self.input[0]}'
echo $(jq -r '.' <<< $OpenIDConnect)
echo $OpenIDConnect
echo | ./kubectl get nodes --kubeconfig $KUBECONFIG
./kubectl create ns $NAMESPACE --kubeconfig $KUBECONFIG --dry-run=client -o yaml | ./kubectl apply --kubeconfig $KUBECONFIG -f -
./kubectl label namespace $NAMESPACE istio-injection=enabled --kubeconfig $KUBECONFIG
echo $OpenIDConnect | ./kubectl apply --kubeconfig $KUBECONFIG -n $NAMESPACE -f -
else
echo $crd
fi
)
EOF
}
}
As a result, the following OpenIDConnect CR will become available in your kyma cluster.
The OIDC identity resolves the authentication requests to the Kubernetes API. However, it must be first authorised to interact with the cluster API.
In order to do so, one must create custom cluster roles to the terraform OIDC identity in the kyma cluster with either "User" and/or "Group" subjects.
For OIDC identities coming from TFC (HCP Terraform), the role binding "User" value is formatted as follows:
organization:<MY-ORG-NAME>:project:<MY-PROJECT-NAME>:workspace:<MY-WORKSPACE-NAME>:run_phase:<plan|apply>.
I have opted for generating these RBAC identities in the initial kyma cluster terraform configuration, thus, adding both plan and apply phase identities to the initial kyma runtime environment configuration as administrators.
User identities
/ https://developer.hashicorp.com/terraform/cloud-docs/run/run-environment#environment-variables
//
variable "TFC_WORKSPACE_NAME" {
// HCP Terraform automatically injects the following environment variables for each run.
description = "The name of the workspace used in this run."
type = string
}
variable "TFC_PROJECT_NAME" {
// HCP Terraform automatically injects the following environment variables for each run.
description = "The name of the project used in this run."
type = string
}
variable "TFC_WORKSPACE_SLUG" {
// HCP Terraform automatically injects the following environment variables for each run.
description = "The slug consists of the organization name and workspace name, joined with a slash."
type = string
}
// organization:<MY-ORG-NAME>:project:<MY-PROJECT-NAME>:workspace:<MY-WORKSPACE-NAME>:run_phase:<plan|apply>.
locals {
organization_name = split("/", var.TFC_WORKSPACE_SLUG)[0]
user_plan = "organization:${local.organization_name}:project:${var.TFC_PROJECT_NAME}:workspace:${var.TFC_WORKSPACE_NAME}:run_phase:plan"
user_apply = "organization:${local.organization_name}:project:${var.TFC_PROJECT_NAME}:workspace:${var.TFC_WORKSPACE_NAME}:run_phase:apply"
}
This way, as soon as the kyma runtime environment has been provisioned, the required identities are in place in the kyma cluster.
After the kyma cluster has been bootstrapped with the HCP Terraform’s OIDC provider in place, one can bind RBAC roles to groups.
Group identity
resource "kubernetes_cluster_role_binding_v1" "oidc_role" {
//depends_on = [ <list of dependencies> ]
metadata {
name = "terraform-identity-admin"
}
//
// Groups are extracted from the token claim designated by 'rbac_group_oidc_claim'
//
role_ref {
api_group = "rbac.authorization.k8s.io"
kind = "ClusterRole"
name = "cluster-admin"
}
subject {
api_group = "rbac.authorization.k8s.io"
kind = "Group"
name = var.tfc_organization_name
namespace = ""
}
}
2. Configure HCP Terraform
Required Environment Variables
HCP Terraform will require these two environment variables to enable kubernetes dynamic credentials
Variable | Value | Notes |
---|---|---|
TFC_KUBERNETES_PROVIDER_AUTH TFC_KUBERNETES_PROVIDER_AUTH[_TAG] | true | Must be present and set to true, or HCP Terraform will not attempt to authenticate to Kubernetes. |
TFC_KUBERNETES_WORKLOAD_IDENTITY_AUDIENCE TFC_KUBERNETES_WORKLOAD_IDENTITY_AUDIENCE[_TAG] TFC_DEFAULT_KUBERNETES_WORKLOAD_IDENTITY_AUDIENCE | The audience name in your cluster's OIDC configuration, such as kubernetes. |
You can set these as workspace variables, or if you’d like to share one Kubernetes role across multiple workspaces, you can use a variable set.
3. Configure the provider
HCP Terraform will assign the tfc_kubernetes_dynamic_credentials
variable the kubeconfig token valid for 90 minutes.
tfc_kubernetes_dynamic_credentials
variable "tfc_kubernetes_dynamic_credentials" {
description = "Object containing Kubernetes dynamic credentials configuration"
type = object({
default = object({
token_path = string
})
aliases = map(object({
token_path = string
}))
})
}
output "kube_token" {
sensitive = true
value = file(var.tfc_kubernetes_dynamic_credentials.default.token_path)
}
provider configuration
terraform {
/**/
cloud {
organization = "<organization>"
workspaces {
project = "terraform-stories"
tags = ["runtime-context"]
}
}
/**/
required_providers {
btp = {
source = "SAP/btp"
}
# https://registry.terraform.io/providers/hashicorp/kubernetes/latest/docs
kubernetes = {
source = "hashicorp/kubernetes"
}
# https://registry.terraform.io/providers/alekc/kubectl/latest/docs
kubectl = {
source = "alekc/kubectl"
//version = "~> 2.0"
}
}
}
provider "kubernetes" {
host = var.cluster-endpoint-url
cluster_ca_certificate = base64decode(var.cluster-endpoint-ca)
token = file(var.tfc_kubernetes_dynamic_credentials.default.token_path)
}
provider "kubectl" {
host = var.cluster-endpoint-url
cluster_ca_certificate = base64decode(var.cluster-endpoint-ca)
token = file(var.tfc_kubernetes_dynamic_credentials.default.token_path)
load_config_file = false
}
One can retrieve both host
and cluster_ca_certificate
from the kyma cluster kubeconfig as follows:
kyma cluster kubeconfig
locals {
labels = btp_subaccount_environment_instance.kyma.labels
}
data "http" "kubeconfig" {
depends_on = [btp_subaccount_environment_instance.kyma]
url = jsondecode(local.labels)["KubeconfigURL"]
lifecycle {
postcondition {
condition = can(regex("kind: Config",self.response_body))
error_message = "Invalid content of downloaded kubeconfig"
}
postcondition {
condition = contains([200], self.status_code)
error_message = self.response_body
}
}
}
# yaml formatted default (oid-based) kyma kubeconfig
locals {
kubeconfig = data.http.kubeconfig.response_body
cluster_ca_certificate = base64decode(local.kubeconfig.clusters[0].cluster.certificate-authority-data)
host = local.kubeconfig.clusters[0].cluster.server
}
Let me explain the provider configuration in detail, as it's a crucial part of connecting HCP Terraform with your Kyma cluster. The provider configuration starts with the Terraform block that sets up the basic framework: This configuration sets up three important providers: the SAP Business Technology Platform (BTP) provider, the official Kubernetes provider, and a kubectl provider for additional Kubernetes operations. The cloud block configures the Terraform Cloud workspace, which is essential for the OIDC authentication flow. The dynamic credentials configuration is handled through a special variable: This variable structure allows Terraform to manage the authentication tokens. The The actual provider configurations use these credentials along with cluster information: The cluster information (host and CA certificate) typically comes from the Kyma cluster's kubeconfig. You can retrieve this information using an HTTP data source: This configuration includes several important security and reliability features: The combination of these providers gives you full access to both the standard Kubernetes resources (through the kubernetes provider) and custom resources or kubectl commands (through the kubectl provider), all authenticated through the secure OIDC mechanism.summary
terraform {
cloud {
organization = "<organization>"
workspaces {
project = "terraform-stories"
tags = ["runtime-context"]
}
}
required_providers {
btp = {
source = "SAP/btp"
}
kubernetes = {
source = "hashicorp/kubernetes"
}
kubectl = {
source = "alekc/kubectl"
}
}
}
variable "tfc_kubernetes_dynamic_credentials" {
description = "Object containing Kubernetes dynamic credentials configuration"
type = object({
default = object({
token_path = string
})
aliases = map(object({
token_path = string
}))
})
}
token_path
points to where Terraform can find the temporary authentication token. The aliases map allows you to configure multiple Kubernetes contexts if needed, though in most cases you'll just use the default.
provider "kubernetes" {
host = var.cluster-endpoint-url
cluster_ca_certificate = base64decode(var.cluster-endpoint-ca)
token = file(var.tfc_kubernetes_dynamic_credentials.default.token_path)
}
provider "kubectl" {
host = var.cluster-endpoint-url
cluster_ca_certificate = base64decode(var.cluster-endpoint-ca)
token = file(var.tfc_kubernetes_dynamic_credentials.default.token_path)
load_config_file = false
}
data "http" "kubeconfig" {
depends_on = [btp_subaccount_environment_instance.kyma]
url = jsondecode(local.labels)["KubeconfigURL"]
lifecycle {
postcondition {
condition = can(regex("kind: Config", self.response_body))
error_message = "Invalid content of downloaded kubeconfig"
}
postcondition {
condition = contains([200], self.status_code)
error_message = self.response_body
}
}
}
locals {
kubeconfig = data.http.kubeconfig.response_body
cluster_ca_certificate = base64decode(local.kubeconfig.clusters[0].cluster.certificate-authority-data)
host = local.kubeconfig.clusters[0].cluster.server
}
load_config_file
set to false to ensure it uses only the provided configuration
4. Retrieve kyma cluster configuration
Examples
kyma cluster shoot_info
data "kubernetes_config_map_v1" "shoot_info" {
metadata {
name = "shoot-info"
namespace = "kube-system"
}
}
output "shoot_info" {
value = jsondecode(jsonencode(data.kubernetes_config_map_v1.shoot_info.data))
}
shoot_info = {
domain = "<shootName>.kyma.ondemand.com"
extensions = "shoot-auditlog-service,shoot-cert-service,shoot-dns-service,shoot-lakom-service,shoot-networking-filter,shoot-networking-problemdetector,shoot-oidc-service"
kubernetesVersion = "1.30.6"
maintenanceBegin = "200000+0000"
maintenanceEnd = "000000+0000"
nodeNetwork = "10.250.0.0/16"
nodeNetworks = "10.250.0.0/16"
podNetwork = "100.64.0.0/12"
podNetworks = "100.64.0.0/12"
projectName = "kyma"
provider = "azure"
region = "westeurope"
serviceNetwork = "100.104.0.0/13"
serviceNetworks = "100.104.0.0/13"
shootName = "<shootName>"
}
kyma cluster availability zones
data "kubernetes_nodes" "k8s_nodes" {}
locals {
k8s_nodes = { for node in data.kubernetes_nodes.k8s_nodes.nodes : node.metadata.0.name => node }
}
data "jq_query" "k8s_nodes" {
data = jsonencode(local.k8s_nodes)
query = "[ .[].metadata[] | { NAME: .name, ZONE: .labels.\"topology.kubernetes.io/zone\", REGION: .labels.\"topology.kubernetes.io/region\" } ]"
}
output "k8s_zones" {
value = jsondecode(data.jq_query.k8s_nodes.result)
}
k8s_zones = [
{
NAME = "shoot--kyma--<shootName>-cpu-worker-0-z1-5759f-j6tsf"
REGION = "westeurope"
ZONE = "westeurope-1"
},
{
NAME = "shoot--kyma--<shootName>-cpu-worker-0-z2-76d84-br7v6"
REGION = "westeurope"
ZONE = "westeurope-2"
},
{
NAME = "shoot--kyma--<shootName>-cpu-worker-0-z3-5b77f-scbpv"
REGION = "westeurope"
ZONE = "westeurope-3"
},
]
kyma cluster list of modules
data "kubernetes_resource" "KymaModules" {
api_version = "operator.kyma-project.io/v1beta2"
kind = "Kyma"
metadata {
name = "default"
namespace = "kyma-system"
}
}
locals {
KymaModules = data.kubernetes_resource.KymaModules.object.status.modules
}
data "jq_query" "KymaModules" {
depends_on = [
data.kubernetes_resource.KymaModules
]
data = jsonencode(local.KymaModules)
query = "[ .[] | { channel, name, version, state, api: .resource.apiVersion, fqdn } ]"
}
output "KymaModules" {
value = jsondecode(data.jq_query.KymaModules.result)
}
KymaModules = [
{
api = "operator.kyma-project.io/v1alpha1"
channel = "regular"
fqdn = "kyma-project.io/module/btp-operator"
name = "btp-operator"
state = "Ready"
version = "1.1.18"
},
{
api = "operator.kyma-project.io/v1alpha1"
channel = "regular"
fqdn = "kyma-project.io/module/serverless"
name = "serverless"
state = "Ready"
version = "1.5.1"
},
{
api = "connectivityproxy.sap.com/v1"
channel = "regular"
fqdn = "kyma-project.io/module/connectivity-proxy"
name = "connectivity-proxy"
state = "Ready"
version = "1.0.4"
},
{
api = "operator.kyma-project.io/v1alpha1"
channel = "regular"
fqdn = "kyma-project.io/module/api-gateway"
name = "api-gateway"
state = "Ready"
version = "2.10.1"
},
{
api = "operator.kyma-project.io/v1alpha2"
channel = "regular"
fqdn = "kyma-project.io/module/istio"
name = "istio"
state = "Ready"
version = "1.11.1"
},
]
Top comments (0)