Oasis CLI Manual

The Oasis CLI is a command-line tool for creating and managing pipeline components and their container environments. It streamlines the development workflow for building reusable ML pipeline components.

Installation

# Install via uvx (recommended)
alias oasis="uvx --refresh --from git+https://github.com/Cloud-Pipelines/oasis-cli.git@stable oasis"

# Or install directly
pip install git+https://github.com/Cloud-Pipelines/oasis-cli.git@stable

tip

Using uvx with the --refresh flag ensures you always run the latest stable version without managing local installations.

Command Structure

oasis [OPTIONS] COMMAND [ARGS]

The CLI follows a hierarchical command structure with two main command groups:

• containers - Manage container images and dependencies
• components - Manage pipeline components

Quick Start

1. Create a Component Root

# Create and initialize a component root directory
components_root="$(pwd)/components_root1"
mkdir -p "$components_root" && cd "$components_root"

# Initialize container root from template
oasis containers new python-uv-container-root \
    --container-image-name <writeable-container-image-uri>

2. Build Container Image

info

By default, Oasis CLI will use Podman. Ensure you have Podman installed and configured correctly. You may need to start your Podman podman machine start before building the container.

# Add dependencies (optional)
uv add numpy pandas scikit-learn

# Build and push container
oasis containers build

3. Create a Component

# Create component directory
mkdir -p my_component && cd my_component

# Initialize component from template
oasis components new python-function-component

# Generate component.yaml
oasis components regenerate python-function-component \
    --container-image-from "${components_root}/oasis.pipeline_component_root.yaml" \
    --dependencies-from "${components_root}/pyproject.toml"

Container Commands

oasis containers new

Creates a new container root with necessary configuration files.

Synopsis:

oasis containers new TEMPLATE [OPTIONS]

Templates:

Template	Description	Generated Files
python-uv-container-root	Python container with uv package manager	Dockerfile, pyproject.toml, uv.lock, oasis.pipeline_component_root.yaml

Options:

Option	Description	Required
--container-image-name	Container registry URI where image will be pushed	Yes

Example:

oasis containers new python-uv-container-root \
    --container-image-name gcr.io/my-project/my-components-base:latest

warning

Ensure you have write permissions to the specified container registry before running the build command.

oasis containers build

Builds a Docker container image and pushes it to the specified registry.

Synopsis:

oasis containers build [OPTIONS]

Behavior:

• Reads configuration from oasis.pipeline_component_root.yaml
• Builds Docker image using the Dockerfile in current directory
• Pushes to the registry specified during new command
• Updates the image digest in configuration

Example:

# From within the container root directory
oasis containers build

Component Commands

oasis components new

Creates a new component from a template.

Synopsis:

oasis components new TEMPLATE [OPTIONS]

Templates:

Template	Description	Generated Files
python-function-component	Basic Python function component	component.py

Example:

mkdir my_component && cd my_component
oasis components new python-function-component

oasis components regenerate

Regenerates the component.yaml specification from the component source code.

Synopsis:

oasis components regenerate TEMPLATE [OPTIONS]

Options:

Option	Description	Required
--container-image-from	Path to container root YAML file	Yes
--dependencies-from	Path to pyproject.toml with dependencies	Yes

Example:

oasis components regenerate python-function-component \
    --container-image-from "../components_root/oasis.pipeline_component_root.yaml" \
    --dependencies-from "../components_root/pyproject.toml"

tip

The regenerate command parses your component.py file to automatically generate the ComponentSpec YAML, including inputs, outputs, and metadata.

Workflow Patterns

Pattern 1: Single Component Root, Multiple Components

Best for components that share dependencies.

# 1. Create shared container root
components_root="$(pwd)/shared_components"
mkdir -p "$components_root" && cd "$components_root"
oasis containers new python-uv-container-root \
    --container-image-name my-registry/shared-base:latest

# 2. Add common dependencies
uv add numpy pandas scikit-learn
oasis containers build

# 3. Create multiple components using the same base
for comp in preprocessor trainer evaluator; do
    mkdir -p "../$comp" && cd "../$comp"
    oasis components new python-function-component
    oasis components regenerate python-function-component \
        --container-image-from "${components_root}/oasis.pipeline_component_root.yaml" \
        --dependencies-from "${components_root}/pyproject.toml"
done

Pattern 2: Updating Dependencies

When you need to add new dependencies to existing components:

# 1. Navigate to container root
cd components_root

# 2. Add new dependencies
uv add tensorflow transformers

# 3. Rebuild container
oasis containers build

# 4. Regenerate affected components
cd ../my_component
oasis components regenerate python-function-component \
    --container-image-from "../components_root/oasis.pipeline_component_root.yaml" \
    --dependencies-from "../components_root/pyproject.toml"

File Structure

A typical project structure using oasis-cli:

project/
├── Dockerfile
├── pyproject.toml
├── uv.lock
├── oasis.pipeline_component_root.yaml
└── components/
    ├── data_processor/
    │   ├── component.py
    │   └── component.yaml
    ├── model_trainer/
    │   ├── component.py
    │   └── component.yaml
    └── evaluator/
        ├── component.py
        └── component.yaml

Configuration Files

oasis.pipeline_component_root.yaml

Stores container configuration.

component.yaml

Generated ComponentSpec for pipeline components:

name: My Component
description: Component description
inputs:
  - name: input_data
    type: Dataset
outputs:
  - name: output_model
    type: Model
implementation:
  container:
    image: gcr.io/project/components@sha256:...
    command: ["python", "component.py"]

Environment Variables

CLOUD_PIPELINES_CONTAINER_EXECUTION_ID
CLOUD_PIPELINES_EXECUTION_NODE_ID
CLOUD_PIPELINES_PIPELINE_RUN_ID
CLOUD_PIPELINES_PIPELINE_RUN_CREATED_BY

Troubleshooting

Common Issues

Container push fails:

• TODO

Component regeneration fails:

• Verify component.py follows the expected component structure
• Check that container root YAML exists and is valid
• Ensure all file paths are correct

Dependency conflicts:

• Use uv to manage Python dependencies consistently
• Clear uv.lock and regenerate if conflicts persist
• Ensure Python version compatibility

Best Practices

1. Version Control: Commit both source files (component.py) and generated files (component.yaml)

2. Container Tagging: Use semantic versioning for container images

   --container-image-name my-registry/components:v1.0.0

3. Dependency Management: Keep dependencies minimal and specify exact versions in pyproject.toml

4. Component Organization: Group related components under the same container root to share dependencies

5. Testing: Test components individually before integrating into pipelines