Model Lifecycle Overview

The model lifecycle manager handles stage transitions, approval workflows, and audit logging for ML models as they progress from development through production.

Lifecycle Stages

Stage	Description
`development`	Model is being trained and evaluated
`staging`	Model is under pre-production validation
`production`	Model is serving live traffic
`deprecated`	Model is scheduled for removal
`archived`	Model is fully decommissioned

Valid Transitions

development -> staging -> production -> deprecated -> archived
                    |          |                         ^
                    v          v                         |
                development  staging          production(restore)
                              |
                              v
                           archived

Lifecycle Policy

from src.lifecycle.lifecycle_manager import LifecyclePolicy
 
policy = LifecyclePolicy(
    require_approval={
        "development_to_staging": False,
        "staging_to_production": True,
        "production_to_archived": True,
    },
    min_time_in_stage_hours={
        "development": 0,
        "staging": 24,
        "production": 0,
    },
    required_validations={
        "staging_to_production": [
            "model_validation",
            "performance_benchmark",
            "security_scan",
        ],
    },
    auto_archive_after_days=90,
    auto_deprecate_versions=5,
    archived_retention_days=365,
)

Source Files

File	Path
Lifecycle Manager	`data-plane/ml-service/src/lifecycle/lifecycle_manager.py`
A/B Testing	`data-plane/ml-service/src/lifecycle/ab_testing.py`
Champion-Challenger	`data-plane/ml-service/src/lifecycle/champion_challenger_service.py`
Version Manager	`data-plane/ml-service/src/lifecycle/version_manager.py`
Rollback Manager	`data-plane/ml-service/src/lifecycle/rollback_manager.py`

Pipeline Orchestration A/B Testing