Skip to content

Projects & Multi-Tenancy đŸĸ

Organize pipelines, runs, and artifacts into isolated projects for multi-tenant deployments.

Overview ℹī¸

The Project and ProjectManager classes provide: - Isolation: Each project has its own metadata store and artifact storage - Organization: Group related pipelines and runs together - Multi-tenancy: Support multiple teams/clients in one deployment - Resource management: Track pipelines, runs, and artifacts per project

Quick Start 🚀

from flowyml import Project

# Create a project
project = Project("recommendation_system")

# Create pipelines within the project
pipeline = project.create_pipeline("training_v1")

# Add steps and run
pipeline.add_step(load_data)
pipeline.add_step(train_model)
result = pipeline.run()

# Get project statistics
stats = project.get_stats()
print(f"Total runs: {stats['total_runs']}")
print(f"Total artifacts: {stats['total_artifacts']}")

Project Management 🏗ī¸

Creating Projects

from flowyml import ProjectManager

manager = ProjectManager()

# Create a new project
project = manager.create_project(
    "ml_platform",
    description="Main ML platform for product recommendations"
)

# List all projects
projects = manager.list_projects()
for proj in projects:
    print(f"- {proj['name']}: {proj['description']}")

# Get existing project
project = manager.get_project("ml_platform")

Project Structure

Each project has its own directory structure:

projects/
└── my_project/
    ├── project.json          # Project metadata
    ├── runs/                 # Pipeline run data
    ├── artifacts/            # Stored artifacts
    │   └── cache/           # Artifact cache
    └── metadata.db          # SQLite metadata store

Working with Projects 🛠ī¸

Creating Pipelines

You can create pipelines in projects in two ways:

Method 1: Using project.create_pipeline() 

project = Project("analytics")

# Create multiple pipelines
etl_pipeline = project.create_pipeline("daily_etl")
reporting_pipeline = project.create_pipeline("weekly_reports")
ml_pipeline = project.create_pipeline("model_training")

# Each pipeline uses the project's metadata store
etl_pipeline.add_step(extract_data)
etl_pipeline.run()

Method 2: Using project_name parameter (Recommended) 

from flowyml import Pipeline, step

# Automatically creates/attaches to project if it doesn't exist
pipeline = Pipeline("daily_etl", project_name="analytics")

@step(outputs=["data"])
def extract_data():
    return fetch_data()

pipeline.add_step(extract_data)
pipeline.run()

The project_name parameter automatically: - Creates the project if it doesn't exist - Attaches the pipeline to the project - Uses the project's metadata store and runs directory

Querying Project Data

# List all runs in the project
runs = project.list_runs()
for run in runs:
    print(f"{run['pipeline_name']}: {run['status']}")

# Filter runs by pipeline
training_runs = project.list_runs(pipeline_name="model_training")

# Get artifacts
artifacts = project.get_artifacts()
for artifact in artifacts:
    print(f"{artifact['name']}: {artifact['type']}")

# Filter by artifact type
models = project.get_artifacts(artifact_type="model")

Project Statistics

stats = project.get_stats()

print(f"""
Project: {project.name}
Created: {stats['created_at']}
Pipelines: {stats['total_pipelines']}
Runs: {stats['total_runs']}
Artifacts: {stats['total_artifacts']}
""")

Multi-Tenant Architecture 🏘ī¸

Isolating Client Data

# Setup for multiple clients
clients = ["acme_corp", "tech_startup", "enterprise_inc"]

for client in clients:
    # Each client gets their own project
    project = manager.create_project(
        client,
        description=f"ML pipelines for {client}"
    )

    # Create client-specific pipelines
    pipeline = project.create_pipeline("recommendation_engine")
    pipeline.add_step(load_client_data)  # Client-specific data
    pipeline.add_step(train_model)

    # Run in isolation
    result = pipeline.run()

Resource Tracking

def get_client_usage(client_name):
    project = manager.get_project(client_name)
    stats = project.get_stats()

    return {
        "client": client_name,
        "pipelines": stats['total_pipelines'],
        "runs": stats['total_runs'],
        "artifacts": stats['total_artifacts'],
        "storage_usage_mb": project.get_storage_usage()
    }

# Generate usage report
for client in clients:
    usage = get_client_usage(client)
    print(f"{client}: {usage['runs']} runs, {usage['storage_usage_mb']}MB")

Best Practices 💡

1. Project Naming

# Use descriptive, hierarchical names
project = Project("company_product_ml")

# Or organize by team/domain
project = Project("data_team_recommendations")

2. Pipeline Organization

project = Project("sales_analytics")

# Group related pipelines
project.create_pipeline("etl_daily")
project.create_pipeline("etl_weekly")
project.create_pipeline("reporting_dashboard")
project.create_pipeline("forecasting_model")

3. Cleanup Old Data

# Export project before cleanup
project.export_metadata("backup.json")

# List pipelines for review
pipelines = project.get_pipelines()

# Remove if needed (use with caution!)
# manager.delete_project("old_project", confirm=True)

Integration Examples 🔌

With Versioning

from flowyml import VersionedPipeline, Pipeline, context

ctx = context(learning_rate=0.001, epochs=10)

# Method 1: Using project_name parameter (Recommended)
# Automatically creates/attaches to project and creates VersionedPipeline
pipeline = Pipeline("training", context=ctx, version="v1.0.0", project_name="ml_prod")
pipeline.add_step(train)
pipeline.save_version()
pipeline.run()

# Method 2: Using VersionedPipeline directly with project_name
versioned = VersionedPipeline("training", context=ctx, version="v1.0.1", project_name="ml_prod")
versioned.add_step(train)
versioned.save_version()
versioned.run()

# Method 3: Using project.create_pipeline() then manually setting up versioning
project = Project("ml_prod")
pipeline = project.create_pipeline("training")
versioned = VersionedPipeline("training", context=ctx, version="v1.0.2")
versioned.runs_dir = project.runs_dir
versioned.metadata_store = project.metadata_store
versioned.add_step(train)
versioned.save_version()
versioned.run()

With Scheduling

from flowyml import PipelineScheduler

project = Project("automated_ml")
scheduler = PipelineScheduler()

def run_project_pipeline():
    pipeline = project.create_pipeline("daily_training")
    pipeline.add_step(train_model)
    return pipeline.run()

scheduler.schedule_daily(
    name=f"{project.name}_daily_run",
    pipeline_func=run_project_pipeline,
    hour=2
)

API Reference 📚

Project

Constructor: 

Project(
    name: str,
    description: str = "",
    projects_dir: str = ".flowyml/projects"
)

Methods: - create_pipeline(name: str, **kwargs) -> Pipeline - Create pipeline in project - get_pipelines() -> List[str] - List all pipeline names - list_runs(pipeline_name: Optional[str] = None, limit: int = 100) -> List[Dict] - get_artifacts(artifact_type: Optional[str] = None, limit: int = 100) -> List[Dict] - get_stats() -> Dict - Get project statistics - export_metadata(path: str) - Export project metadata

ProjectManager

Methods: - create_project(name: str, description: str = "") -> Project - Create new project - get_project(name: str) -> Optional[Project] - Get existing project - list_projects() -> List[Dict] - List all projects - delete_project(name: str, confirm: bool = False) - Delete project

FAQ ❓

Q: Can I move a pipeline from one project to another? A: Currently, pipelines are tied to their project's metadata store. You would need to export/import the pipeline definition manually.

Q: How do I backup a project? A: Use project.export_metadata() and copy the entire project directory from .flowyml/projects/{project_name}/.

Q: What happens when I delete a project? A: All pipelines, runs, and artifacts associated with the project are removed. Always export metadata first!

Q: Can projects share artifacts? A: No, projects are fully isolated by design. This ensures multi-tenant security and resource tracking.