Sharing secrets in Kubeflow pipeline

Question

I want to share some secrets with my Kubeflow pipeline so I can use them as environment variables in my containers. I've written a pipeline-secrets.yaml that looks like this:

apiVersion: v1
kind: Secret
metadata:
  name: pipeline-secrets
  namespace: kubeflow
type: Opaque
data:
  mysql_db_name: <SECRET>
  mysql_username: <SECRET>
  mysql_password: <SECRET>
  mysql_endpoint: <SECRET>

and a pipeline-pod-defaults.yaml that looks like this:

apiVersion: kubeflow.org/v1alpha1
kind: PodDefault
metadata:
  name: pipeline-pod-defaults
  namespace: kubeflow
specs:
  desc: Configure pipeline secrets as environment variables
  env:
  - name: MYSQL_DB_NAME
    valueFrom:
      secretKeyRef:
        name: pepeline-secrets
        key: mysql_db_name
  - name: MYSQL_USER_NAME
    valueFrom:
      secretKeyRef:
        name: pepeline-secrets
        key: mysql_username
  - name: MYSQL_PASSWORD
    valueFrom:
      secretKeyRef:
        name: pepeline-secrets
        key: mysql_password
  - name: MYSQL_ENDPOINT
    valueFrom:
      secretKeyRef:
        name: pepeline-secrets
        key: mysql_endpoint

And this is how my pipeline looks like:

import kfp
from kfp.dsl import ContainerOp
from kubernetes import client as k8s_client

@kfp.dsl.pipeline(
    name="Training pipeline",
    description=""
)
def train_pipeline():
    get_data = ContainerOp(
        name="Get data",
        image=BASE_IMAGE,
        file_outputs={
            'data': 'data.csv'
        }
    )
    
    kfp.dsl.get_pipeline_conf().set_image_pull_secrets([
        k8s_client.V1ObjectReference(name="regcred"),
        k8s_client.V1ObjectReference(name="pipeline-secrets"),
    ])
    kfp.dsl.ResourceOp(
        name="pipeline-pod-defaults",
        k8s_resource=k8s_client.V1ObjectReference(name="pipeline-pod-defaults"),
        action="apply"
    )

But at the end I'm getting this error:

This step is in Failed state with this message: error: error validating "/tmp/manifest.yaml": error validating data: [apiVersion not set, kind not set]; if you choose to ignore these errors, turn validation off with --validate=false

Is this the correct approach? How can I share my secrets with the rest of the pipeline? Sorry if it's a newbie question, I'm new to both Kubernetes and Kubeflow

Answer 1

So, in the end, what I did was write a get-data-components.yaml to create my component and wrote the function below and it worked:

def build_get_data():
    component = kfp.components.load_component_from_file(os.path.join(COMPONENTS_PATH, 'get-data-component.yaml'))()
    component.add_volume(k8s_client.V1Volume(
        name="get-data-volume",
        secret=k8s_client.V1SecretVolumeSource(secret_name="pipeline-secrets"))
    )
    envs = [
        ("MYSQL_DB_NAME", "mysql_db_name"),
        ("MYSQL_USER_NAME", "mysql_username"), 
        ("MYSQL_PASSWORD", "mysql_password"), 
        ("MYSQL_ENDPOINT", "mysql_endpoint")
    ]
    for name, key in envs:
        component.add_env_variable(
            V1EnvVar(
                name=name,
                value_from=k8s_client.V1EnvVarSource(secret_key_ref=k8s_client.V1SecretKeySelector(
                    name="pipeline-secrets",
                    key=key
                    )
                )
            )
        )
    return component

Answer 2

has native support for using Kubernetes Secrets now:

https://kfp-kubernetes.readthedocs.io/en/kfp-kubernetes-1.0.0/

Answer 3

I am not sure what is the best approache but what I am doing is creating a secret in the same namespace where the pipeline will run in the cluster. Then in the pyton script I am running the following code.

config.load_incluster_config()
v1 = client.CoreV1Api()
sec = v1.read_namespaced_secret(<name of the secret>, <namespace you pick the secret from>).data

YOUR_SECRET_1 = base64.b64decode(sec.get(<name of the env variable >)).decode('utf-8')
YOUR_SECRET_2 = base64.b64decode(sec.get(<name of the env variable >)).decode('utf-8')