Skip to main content

​
Stack Components Overview

ZenML’s stack is the foundation of your MLOps infrastructure. A stack is composed of multiple components that work together to run your ML pipelines. Each component handles a specific aspect of the pipeline execution, from orchestration to artifact storage.

​
What is a Stack?

A stack is essentially all the configuration for the infrastructure of your MLOps platform. It’s made up of multiple components that define where and how your pipelines run. Every ZenML stack requires at minimum:
  • Orchestrator: Manages pipeline execution
  • Artifact Store: Stores pipeline artifacts and outputs

​
Core Stack Components

Orchestrators

Execute and manage pipeline runs across different environments

Artifact Stores

Store and manage pipeline artifacts, inputs, and outputs

Container Registries

Store and manage Docker container images for containerized execution

Model Deployers

Deploy models for online inference and serving

Experiment Trackers

Track experiments, metrics, and parameters across pipeline runs

Step Operators

Run individual steps on specialized infrastructure like GPUs

​
Viewing Your Stack

To see your current stack configuration: 
zenml stack describe
To list all available stacks: 
zenml stack list

​
Default Stack

When you initialize ZenML, a default stack is created with:
  • Local Orchestrator: Runs pipelines locally on your machine
  • Local Artifact Store: Stores artifacts in a local directory
This stack is perfect for getting started and local development, but you’ll want to configure remote components for production use.

​
Creating Custom Stacks

You can create custom stacks by registering individual components and combining them: 
# Register components
zenml orchestrator register my_orchestrator --flavor=local
zenml artifact-store register my_artifact_store --flavor=local --path=/path/to/artifacts

# Create a stack from components
zenml stack register my_stack \
  -o my_orchestrator \
  -a my_artifact_store

# Set as active stack
zenml stack set my_stack

​
Stack Validation

ZenML validates that all components in a stack are compatible with each other. Some components have specific requirements:
  • Container-based orchestrators require a container registry
  • Remote orchestrators typically require remote artifact stores
  • Certain integrations have specific version requirements
When you register or set a stack, ZenML automatically validates compatibility.

​
Component Flavors

Each component type can have multiple “flavors” - different implementations of the same component interface. For example:
  • Orchestrator flavors: local, kubernetes, airflow, kubeflow, vertex, sagemaker
  • Artifact Store flavors: local, s3, gcs, azure
  • Container Registry flavors: default, dockerhub, gcp, azure, github
To see available flavors for a component: 
zenml orchestrator flavor list
zenml artifact-store flavor list

​
Integration Components

Many components require ZenML integrations to be installed. For example, to use AWS components: 
zenml integration install aws
Common integrations include:
  • aws - S3, SageMaker, and other AWS services
  • gcp - GCS, Vertex AI, and other GCP services
  • azure - Azure Blob Storage and Azure ML
  • kubernetes - Kubernetes orchestration
  • mlflow - MLflow experiment tracking
  • wandb - Weights & Biases experiment tracking

​
Next Steps

Explore the individual component pages to learn how to configure and use each component type in your ML pipelines.

Build docs developers (and LLMs) love

Get started for freeTalk to us