Overview

Relevant source files

• CONTRIBUTING.md
• README.md
• docs/SUMMARY.md
• docs/getting-started/architecture/write-patterns.md
• docs/getting-started/components/online-store.md
• docs/getting-started/components/overview.md
• docs/getting-started/concepts/data-ingestion.md
• docs/getting-started/quickstart.md
• docs/getting-started/third-party-integrations.md
• docs/how-to-guides/feast-on-kubernetes.md
• docs/how-to-guides/running-feast-in-production.md
• docs/how-to-guides/scaling-feast.md
• docs/project/maintainers.md
• docs/reference/codebase-structure.md
• docs/reference/compute-engine/lambda.md
• docs/reference/feast-cli-commands.md
• docs/reference/feature-servers/offline-feature-server.md
• docs/reference/registries/README.md
• docs/reference/registries/hdfs.md
• docs/reference/registries/metadata.md
• docs/roadmap.md
• examples/operator-postgres-tls-demo/02-Install-feast.ipynb
• examples/operator-postgres-tls-demo/03-Uninstall.ipynb
• examples/quickstart/quickstart.ipynb
• infra/feast-operator/README.md
• infra/feast-operator/api/v1/featurestore_types.go
• infra/feast-operator/config/samples/kustomization.yaml
• infra/templates/README.md.jinja2
• protos/feast/core/Registry.proto
• sdk/python/feast/errors.py
• sdk/python/feast/feature_store.py
• sdk/python/feast/infra/online_stores/milvus_online_store/milvus.py
• sdk/python/feast/infra/online_stores/online_store.py
• sdk/python/feast/infra/online_stores/sqlite.py
• sdk/python/feast/infra/passthrough_provider.py
• sdk/python/feast/infra/provider.py
• sdk/python/feast/infra/registry/contrib/hdfs/__init__.py
• sdk/python/feast/infra/registry/contrib/hdfs/hdfs_registry_store.py
• sdk/python/feast/repo_config.py
• sdk/python/feast/repo_operations.py
• sdk/python/feast/utils.py
• sdk/python/tests/example_repos/example_feature_repo_1.py
• sdk/python/tests/foo_provider.py
• sdk/python/tests/integration/feature_repos/universal/online_store/milvus.py
• sdk/python/tests/unit/infra/registry/test_sql_registry.py
• sdk/python/tests/unit/online_store/test_online_retrieval.py

Purpose and Scope

This document provides a high-level introduction to the Feast feature store repository, explaining its purpose, architecture, and core components. Feast is an open-source feature store for machine learning that provides consistent offline (training) and online (serving) feature access.

For detailed information on specific topics:

• System architecture details: see System Architecture
• CLI usage and commands: see Getting Started and CLI
• Feature definitions and concepts: see Core Concepts
• Storage integrations: see Data Integration
• Production deployment: see Deployment

Sources: README.md26-36 docs/getting-started/quickstart.md1-16

What is Feast

Feast (Feature Store) is an open-source feature store for machine learning that manages existing infrastructure to productionize analytic data for model training and online inference.

For Data Scientists: A tool to define, store, and retrieve features for both model development and deployment without software engineering concerns.

For MLOps Engineers: A library that connects existing infrastructure (databases, application servers, analytical databases, orchestration tools) enabling data scientists to ship features to production via a friendly SDK.

For Data Engineers: A centralized catalog for feature definitions providing a single source of truth, with abstraction for reading/writing to multiple offline and online data stores.

For AI Engineers: A platform to scale AI applications by enabling seamless integration of richer data and facilitating fine-tuning with optimized data pipelines.

Feast solves critical problems:

• Avoids data leakage through point-in-time correct feature sets
• Makes features consistently available for training and serving
• Decouples ML from data infrastructure providing a single data access layer

Sources: README.md29-36 docs/getting-started/quickstart.md4-14

Repository Structure

The Feast repository is organized into several main directories:

Directory	Purpose	Key Components
sdk/python/feast/	Python SDK and core logic	FeatureStore, Registry, Provider
sdk/python/feast/infra/	Infrastructure abstractions	Offline stores, online stores, compute engines
protos/	Protocol buffer definitions	API contracts for cross-language compatibility
infra/	Deployment infrastructure	Docker images, Helm charts, Kubernetes operator
ui/	Web UI (React)	Feature browsing and visualization
docs/	Documentation	User guides, API references
examples/	Example implementations	Quickstart, tutorials

Sources: sdk/python/feast/feature_store.py1-100 docs/reference/codebase-structure.md1-30

Core Architecture

Main Components and Their Relationships

Sources: sdk/python/feast/feature_store.py100-203 sdk/python/feast/infra/provider.py49-62 sdk/python/feast/repo_config.py193-257

FeatureStore: The Main Entry Point

The FeatureStore class in sdk/python/feast/feature_store.py100-189 is the primary interface for all Feast operations:

Key Attributes:

• config: RepoConfig - Configuration for the feature store
• repo_path: Path - Path to the feature repository
• _registry: BaseRegistry - Registry for metadata storage
• _provider: Provider - Provider for infrastructure operations

Initialization Flow:

1. Load configuration from feature_store.yaml or explicit RepoConfig
2. Initialize registry based on registry_type (file, SQL, Snowflake, or remote)
3. Initialize provider (typically PassthroughProvider) based on provider setting
4. Configure offline store, online store, and batch engine

Sources: sdk/python/feast/feature_store.py116-178 sdk/python/feast/repo_config.py255-296

Feature Lifecycle

Definition to Serving

Sources: sdk/python/feast/feature_store.py821-1007 sdk/python/feast/infra/passthrough_provider.py58-130 sdk/python/feast/repo_operations.py399-415

Feature Definition Objects

Feast uses several core objects to define features:

Entity sdk/python/feast/entity.py

• Defines join keys for features (e.g., driver_id, customer_id)
• Used to link features to specific entities during retrieval

FeatureView sdk/python/feast/feature_view.py

• Groups features with common properties (TTL, entities, data source)
• Defines batch features from offline sources
• Includes schema via Field objects

OnDemandFeatureView sdk/python/feast/on_demand_feature_view.py

• Defines real-time transformations applied at serving time
• Can reference batch features and request-time data
• Supports Python, Pandas, and Substrait transformation modes

StreamFeatureView sdk/python/feast/stream_feature_view.py

• Defines features from streaming sources with aggregations
• Supports windowed operations for real-time feature computation

FeatureService sdk/python/feast/feature_service.py

• Groups features for a specific model or use case
• Enables feature versioning and model tracking

Sources: sdk/python/feast/entity.py sdk/python/feast/feature_view.py sdk/python/feast/on_demand_feature_view.py sdk/python/feast/stream_feature_view.py

Storage Architecture

Provider Pattern

The Provider interface sdk/python/feast/infra/provider.py49-105 abstracts infrastructure operations. The default implementation is PassthroughProvider sdk/python/feast/infra/passthrough_provider.py58-130 which delegates to:

Offline Stores - Historical feature retrieval

• Interface: OfflineStore sdk/python/feast/infra/offline_stores/offline_store.py
• Implementations: BigQuery, Snowflake, Redshift, Spark, Postgres, Trino, Dask
• Key method: get_historical_features() for point-in-time joins

Online Stores - Low-latency feature serving

• Interface: OnlineStore sdk/python/feast/infra/online_stores/online_store.py35-150
• Implementations: Redis, DynamoDB, SQLite, Postgres, Cassandra, Milvus
• Key methods: online_write_batch(), online_read()

Batch Engines - Scalable materialization

• Interface: ComputeEngine sdk/python/feast/infra/compute_engines/base.py
• Implementations: Local, Snowflake, AWS Lambda, Bytewax
• Key method: materialize() for offline-to-online data transfer

Sources: sdk/python/feast/repo_config.py39-106 sdk/python/feast/infra/provider.py533-545 sdk/python/feast/infra/passthrough_provider.py58-130

Registry: Metadata Management

The Registry manages all feature store metadata including feature views, entities, data sources, and infrastructure state.

Registry Types:

Type	Implementation	Storage	Use Case
file	Registry	Local/S3/GCS file	Development, single-user
sql	SqlRegistry	PostgreSQL/MySQL	Production, multi-user
snowflake.registry	SnowflakeRegistry	Snowflake table	Snowflake-native deployments
remote	RemoteRegistry	Remote gRPC service	Distributed deployments

Key Operations:

• apply_entity(), apply_feature_view() - Register objects
• get_entity(), get_feature_view() - Retrieve objects
• list_entities(), list_feature_views() - List objects
• refresh() - Reload registry cache
• commit() - Persist changes

Sources: sdk/python/feast/feature_store.py155-177 sdk/python/feast/infra/registry/base_registry.py sdk/python/feast/repo_config.py39-44

Configuration

Feast is configured via feature_store.yaml which is loaded by RepoConfig sdk/python/feast/repo_config.py193-296:

Core Configuration:

Configuration is loaded by:

1. load_repo_config() sdk/python/feast/repo_config.py
2. Pydantic validation and property resolution
3. Dynamic import of store/engine classes via import_class()

Sources: sdk/python/feast/repo_config.py193-296 examples/quickstart/quickstart.ipynb README.md108-117

CLI Commands

The Feast CLI sdk/python/feast/cli.py provides commands for managing the feature store:

Repository Management:

• feast init <project> - Initialize a new feature repository
• feast apply - Register feature definitions and update infrastructure
• feast plan - Preview changes without applying them
• feast teardown - Remove all infrastructure

Feature Materialization:

• feast materialize <start> <end> - Materialize features for a time range
• feast materialize-incremental <end> - Materialize since last materialization

Feature Serving:

• feast serve - Start Python feature server
• feast ui - Start web UI for feature browsing

Inspection:

• feast feature-views list - List all feature views
• feast entities list - List all entities
• feast feature-services list - List all feature services

Sources: sdk/python/feast/repo_operations.py223-240 docs/reference/feast-cli-commands.md1-30 README.md45-128

Key Workflows

Workflow 1: Feature Definition and Registration

1. Define features in Python files (e.g., feature_definitions.py)
2. Run feast apply to register with registry
3. Provider creates necessary infrastructure (tables in online store)

Implementation: sdk/python/feast/feature_store.py821-1007 sdk/python/feast/repo_operations.py399-415

Workflow 2: Training Data Generation

1. Prepare entity dataframe with join keys and timestamps
2. Call fs.get_historical_features(entity_df, features)
3. Provider performs point-in-time joins in offline store
4. Returns RetrievalJob that can be converted to dataframe

Implementation: sdk/python/feast/feature_store.py1356-1643

Workflow 3: Online Feature Materialization

1. Run feast materialize or materialize_incremental()
2. Provider reads latest features from offline store
3. Batch engine coordinates data transfer
4. Features written to online store via online_write_batch()
5. Registry tracks materialization timestamps

Implementation: sdk/python/feast/feature_store.py1644-1968 sdk/python/feast/infra/passthrough_provider.py376-534

Workflow 4: Online Feature Retrieval

1. Call fs.get_online_features(features, entity_rows)
2. Provider reads from online store via online_read()
3. On-demand transformations applied if configured
4. Returns OnlineResponse with feature values

Implementation: sdk/python/feast/feature_store.py1969-2331

Sources: sdk/python/feast/feature_store.py sdk/python/feast/infra/passthrough_provider.py

Summary

Feast provides a complete feature store solution with:

• Unified API via FeatureStore class for all operations
• Pluggable backends for offline stores, online stores, and compute engines
• Metadata management via flexible registry implementations
• CLI tools for repository management and feature operations
• Cross-language support via Protocol Buffer definitions

The architecture enables teams to decouple ML workflows from infrastructure while maintaining consistency between training and serving environments.

For more details on specific components, refer to the related wiki sections linked at the beginning of this document.

Sources: README.md26-41 sdk/python/feast/feature_store.py100-203 docs/getting-started/quickstart.md1-60