Execute Pipelines in the Cloud

This guide will show how you can run your pipelines in Kubeflow Pipelines deployed to a public cloud cluster. We will start with some prerequisites and then move on to show the integration of your cloud provider's components with your ZenML stack. In addition to configuring the stack components in this guide, you can optionally run a pipeline step on a specialized cloud backend too. Check out the step operators guide for that!
We have recently also added a Vertex AI orchestrator to our arsenal of orchestrators (thanks @Gabriel Martin). So if you are running on GCP and want to try out a serverless alternative to Kubeflow, check out our example on how to run the Vertex AI Orchestrator here.
Pre-requisites
Orchestrator
AWS
GCP
Azure
Have an existing AWS EKS cluster set up.
Make sure you have the AWS CLI set up.
Download and install kubectl and configure it to talk to your EKS cluster using the following command:
aws eks --region REGION update-kubeconfig --name CLUSTER_NAME
​Install Kubeflow Pipelines onto your cluster.
Have an existing GCP GKE cluster set up.
Make sure you have the Google Cloud CLI set up first.
Download and install kubectl and configure it to talk to your GKE cluster using the following command:
gcloud container clusters get-credentials CLUSTER_NAME
​Install Kubeflow Pipelines onto your cluster.
Have an existing AKS cluster set up.
Make sure you have the az CLI set up first.
Download and install kubectl and it to talk to your AKS cluster using the following command:
az aks get-credentials --resource-group RESOURCE_GROUP --name CLUSTER_NAME
​Install Kubeflow Pipelines onto your cluster.
Since Kubernetes v1.19, AKS has shifted
to containerd​
. However, the workflow controller installed with the Kubeflow installation has Docker set as the default runtime. In order to make your pipelines work, you have to change the value to one of the options
listed here​
, preferably k8sapi.
This change has to be made by editing the containerRuntimeExecutor property of the ConfigMap corresponding to the workflow controller. Run the following commands to first know what config map to change and then to edit it to reflect your new value.
kubectl get configmap -n kubeflow
kubectl edit configmap CONFIGMAP_NAME -n kubeflow
# This opens up an editor that can be used to make the change.
If one or more of the deployments are not in the Running state, try increasing the number of nodes in your cluster.
If you're installing Kubeflow Pipelines manually, make sure the Kubernetes service is called exactly ml-pipeline. This is a requirement for ZenML to connect to your Kubeflow Pipelines deployment.
Container Registry
AWS
GCP
Azure
​Set up an Elastic Container Registry (ECR) and create a repository (either public or private) with the name zenml-kubeflow. This is the repository to which ZenML will push your pipeline images.
The path value to register with ZenML should be in the format ACCOUNT_ID.dkr.ecr.REGION.amazonaws.com
Authenticate your local docker CLI with your ECR registry using the following command. Replace the capitalized words with your values.
aws ecr get-login-password --region REGION | docker login --username AWS --password-stdin ACCOUNT_ID.dkr.ecr.REGION.amazonaws.com
Set up a GCP Container Registry.
​Authenticate your local docker cli with your GCP container registry.
​Set up an Azure Container Registry (ACR) and create a repository (either public or private) with the name zenml-kubeflow . This is the repository to which ZenML will push your pipeline images to.
Authenticate your local docker cli with your ACR registry using the following command.
az acr login --name REGISTRY_NAME
Artifact Store
AWS
GCP
Azure
​Create an Amazon S3 bucket in a region of your choice.
Make sure that your EKS cluster is authorized to access the S3 bucket. This can be done in one of the following ways:
A simple way is to add an AmazonS3FullAccess policy to your cluster node group's IAM role.
A complex way would be to create ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and AWS_SESSION_TOKEN environment variables in your EKS cluster. This can be done by extending the zenmldocker/zenml image and adding these variables in a Dockerfile.
The path for your bucket should be in this format s3://your-bucket.
​Create a GCP Cloud Storage Bucket in a region of your choice.
Make sure that your GKE cluster is authorized to access the GCS bucket.
The path for your bucket should be in this format gs://your-bucket.
​Create an Azure Storage Account and add a container to hold your blobs.
Open your storage account page on the Azure portal and follow these steps:
Keep the name of the storage account handy - it will be required in a later step.
From the navigation panel on the left, select Access Keys under ** Security + networking**. Note any of the keys available there for future use.
Make sure to set a combination of the following environment variables:AZURE_STORAGE_CONNECTION_STRING or AZURE_STORAGE_ACCOUNT_NAME and one of [AZURE_STORAGE_ACCOUNT_KEY, AZURE_STORAGE_SAS_TOKEN].
This can be done
On Linux/MacOS by exporting them in a terminal: export AZURE_STORAGE_CONNECTION_STRING=<MY_CONNECTION_STRING>.
On Windows by running the command setx AZURE_STORAGE_CONNECTION_STRING "<MY_CONNECTION_STRING>" in the Command Prompt.
In a docker image by adding these variables in a Dockerfile.
The path for your bucket should be in this format: az://<CONTAINER-NAME> or abfs://<CONTAINER_NAME>. See adlfs ( the package ZenML uses to access your Azure storage) for more details.
Metadata Store
The metadata store requires a SQL compatible database service to store information about pipelines, pipeline runs, steps and artifacts. If you want to avoid using a managed cloud SQL database service, ZenML has the option to reuse the gRPC metadata service installed and used internally by the Kubeflow Pipelines deployment. This is not recommended for production use.
To use the Kubeflow Pipelines metadata service, use a metadata store of flavor kubeflow in your stack configuration and ignore this section.
If you decide to use an SQL metadata store backed by a managed cloud SQL database service, you will also need a matching secrets manager to store the SSL credentials (i.e. certificates) required to connect to it.
AWS
GCP
Azure
​Set up an Amazon RDS database instance with a username and a password.
Find the endpoint (DNS name) and TCP port number for your DB instance. You will need them to register with ZenML. The endpoint should be in the format INSTANCE-NAME.ID.REGION.rds.amazonaws.com.
You may also have to explicitly reconfigure the VPC security group rules to allow access to the database from outside the cloud as well as from the EKS cluster (if they do not share the same VPC).
It is strongly recommended to also enable SSL access to your RDS database instance. This is done differently depending on the type of RDS service you use. For example, for a MySQL RDS service, you have to connect to the database using the mysql client and require SSL for the access for the user you created at the first step. You will also need the AWS server certificate associated with the AWS region where the RDS instance is running, as covered in the AWS documentation. Download the certificate as a file and have it ready for use during the ZenML stack registration.
Set up a Google Cloud SQL database instance. You should enable a public IP address for the database instance, if you want to access it from outside the GCP cloud e.g. to have access to the pipeline metadata from your local host. In the authorized networks section of the database instance, you should add a CIDR that covers all IP addresses that will access the database (e.g. the GKE cluster, your local host). If you don't know what these are, you should use a CIDR value of 0.0.0.0/0 that allows all IP addresses, but make sure to enable SSL access to the database.
Find the public IP address and TCP port number for your DB instance. You will need them to register with ZenML.
It is strongly recommended to enable SSL encryption for your database instance and allow only SSL connections. Also create a client certificate and key pair that you will use to connect ZenML to the database. You should have three files saved to your machine: a server certificate, a client certificate and a client key. You will need all three during the ZenML stack registration.
The Azure SQL metadata service is not yet supported because ZenML doesn't yet support a matching Azure secrets manager.
Integrating with ZenML
To run our pipeline on Kubeflow Pipelines deployed to cloud, we will create a new stack with these components that you have just created.
1.Install the cloud provider and the kubeflow plugin
zenml integration install <aws s3/gcp/azure>
zenml integration install kubeflow
2.Register the metadata store component
if you decided to use the Kubeflow metadata service, you can configure a kubeflow flavor metadata-store:
zenml metadata-store register cloud_metadata_store --flavor=kubeflow
otherwise, configure a mysql flavor metadata-store. You will also need a secrets manager component to store the MySQL credentials and a secret (named mysql_secret in the example) to be registered after the stack (step 5.).
zenml metadata-store register cloud_metadata_store --flavor=mysql --secret=mysql_secret
3.Register the other stack components and the stack itself
zenml container-registry register cloud_registry --flavor=<aws/gcp/azure> --uri=$PATH_TO_YOUR_CONTAINER_REGISTRY
zenml orchestrator register cloud_orchestrator --flavor=kubeflow --custom_docker_base_image_name=YOUR_IMAGE
zenml artifact-store register cloud_artifact_store --flavor=<s3/gcp/azure> --path=$PATH_TO_YOUR_BUCKET
zenml secrets-manager register cloud_secrets_manager --flavor=<aws/gcp/azure> 
​
# Register the cloud stack
zenml stack register cloud_kubeflow_stack -m cloud_metadata_store -a cloud_artifact_store -o cloud_orchestrator -c cloud_registry -x cloud_secrets_manager
4.Activate the newly created stack.
zenml stack set cloud_kubeflow_stack
5.Create the secret for the mysql metadata store (not necessary if you used a kubeflow flavor metadata-store).
zenml secret register mysql_secret --schema=mysql --user=<user> --password=<password>
--[email protected]/path/to/downloaded/server-cert --[email protected]/path/to/downloaded/client-cert
--[email protected]/path/to/downloaded/client-key
6.Do a pipeline run and check your Kubeflow UI to see it running there! 🚀
The secrets manager is only required if you're using a metadata store backed by a managed cloud database service.
The metadata store secret user, password and SSL certificates are the ones set up and downloaded during the creation of the managed SQL database service.
The --ssl_cert and --ssl_key parameters for the mysql_secret are only required if you set up client certificates for the MySQL service.
You can choose any name for your stack components apart from the ones used in the script above.
Make sure to replace $PATH_TO_YOUR_BUCKET and $PATH_TO_YOUR_CONTAINER_REGISTRY with the actual URIs of your bucket and container registry.

Advanced Guide - Previous

Track Metadata in the Cloud

Next - Advanced Guide

Continuous Training and Deployment

Last modified 3mo ago

Export as PDF

Copy link

Edit on GitHub

On this page

Pre-requisites

Orchestrator

Container Registry

Artifact Store

Metadata Store

Integrating with ZenML