Feature Store Without Kubernetes

TL;DR: Run a production-grade feature store with pip install and Python decorators. No Kubernetes, no Docker, no YAML. DuckDB locally, Postgres in production.

The Problem

You want to serve ML features. Every tutorial says:

1. Set up a Kubernetes cluster
2. Deploy Feast with 6 containers
3. Configure Spark/Flink pipelines
4. Write 500 lines of YAML
5. Hire a platform team to maintain it

For what? To serve a few features to your model.

The Alternative

pip install fabra-ai && fabra demo

That's it. Server starts, makes a test request, shows you the result. No Docker. No config files.

Or build your own:

pip install "fabra-ai[ui]"

from fabra.core import FeatureStore, entity, feature
from datetime import timedelta

store = FeatureStore()

@entity(store)
class User:
    user_id: str

@feature(entity=User, refresh=timedelta(hours=1))
def purchase_count(user_id: str) -> int:
    return db.query("SELECT COUNT(*) FROM purchases WHERE user_id = ?", user_id)

@feature(entity=User, refresh=timedelta(days=1))
def user_tier(user_id: str) -> str:
    return "premium" if is_premium(user_id) else "free"

fabra serve features.py
# Server running on http://localhost:8000

curl localhost:8000/features/purchase_count?entity_id=user123
# {"value": 42, "freshness_ms": 0, "served_from": "online"}

That's it. No containers. No orchestration. Just Python.

How It Works

Local Development (Default)

When you run locally (FABRA_ENV=development):

Component	Technology	Notes
Online Store	In-memory dict	Instant reads
Offline Store	DuckDB	Embedded, no server
Scheduler	APScheduler	In-process

Zero external dependencies. Works on any machine with Python.

Production

When you deploy (FABRA_ENV=production):

Component	Technology	Notes
Online Store	Redis	Sub-millisecond reads
Offline Store	Postgres	With pgvector for RAG
Scheduler	Distributed	Redis coordination

Same code. Same decorators. Different backends.

# Local (default)
fabra serve features.py

# Production
FABRA_ENV=production \
FABRA_POSTGRES_URL=postgresql+asyncpg://... \
FABRA_REDIS_URL=redis://... \
fabra serve features.py

Why Not Kubernetes?

	Kubernetes	Fabra
Setup time	Days	30 seconds
Minimum cost	$70-150/month	$0 (local)
Config files	500+ lines YAML	0
Team required	Platform engineers	Just you
Learning curve	47 concepts	3 decorators

For a feature store serving 1-10k requests per second, Kubernetes adds complexity without proportional benefit.

Feature Parity

Fabra matches Feast on the hard problems:

Point-in-Time Correctness

Training ML models requires feature values as they existed at prediction time. We use:

• DuckDB (local): ASOF JOIN for temporal lookups
• Postgres (production): LATERAL JOIN with index optimization

-- DuckDB (local)
SELECT * FROM features
ASOF JOIN events
ON features.user_id = events.user_id
AND features.timestamp <= events.timestamp

-- Postgres (production)
SELECT * FROM events e
LEFT JOIN LATERAL (
    SELECT * FROM features f
    WHERE f.user_id = e.user_id
    AND f.timestamp <= e.timestamp
    ORDER BY f.timestamp DESC LIMIT 1
) f ON true

Same semantics. Same correctness. No Spark cluster required.

Async I/O

Production serving uses non-blocking I/O:

• asyncpg for Postgres connections
• redis-py async client
• Connection pooling with circuit breakers

Refresh Scheduling

@feature(entity=User, refresh=timedelta(minutes=5))  # Every 5 minutes
@feature(entity=User, refresh=timedelta(hours=1))    # Every hour
@feature(entity=User, refresh=timedelta(days=1))     # Daily

Schedules run in-process locally, distributed in production.

One-Command Deploy

When you're ready for production:

# Fly.io
fabra deploy fly --name my-features

# Google Cloud Run
fabra deploy cloudrun --name my-features --project my-gcp

# AWS ECS
fabra deploy ecs --name my-features

# Railway
fabra deploy railway --name my-features

Each command generates the right config files (Dockerfile, fly.toml, cloudbuild.yaml) and deploys.

Cost Comparison

Platform	Minimum	With Traffic
Local (DuckDB)	$0	$0
Fly.io	$5/month	$15-50/month
Cloud Run	$0 (scale to zero)	$10-50/month
Railway	$5/month	$15-40/month
Kubernetes (GKE)	$70/month	$100-300/month

Kubernetes is 10-20x more expensive for the same workload.

Quick Start

pip install "fabra-ai[ui]"

# Create features file
cat > features.py << 'EOF'
from fabra.core import FeatureStore, entity, feature
from datetime import timedelta

store = FeatureStore()

@entity(store)
class User:
    user_id: str

@feature(entity=User, refresh=timedelta(minutes=5))
def login_count(user_id: str) -> int:
    return abs(hash(user_id)) % 100
EOF

# Serve
fabra serve features.py

# Test
curl localhost:8000/features/login_count?entity_id=test123
# {"value": 42, "freshness_ms": 0, "served_from": "online"}

FAQ

Q: Can I run a feature store without Docker? A: Yes. Fabra uses DuckDB (embedded database) for local development. No Docker, no containers, no external services. Just pip install and Python.

Q: How do I avoid Kubernetes for ML features? A: Use Fabra with standard Postgres + Redis for production. Deploy to Fly.io, Railway, Cloud Run — any container platform. No K8s cluster required.

Q: Is this production-ready without Kubernetes? A: Yes. We use asyncpg connection pooling, Redis with circuit breakers, and Prometheus metrics. Same reliability, simpler ops.

Q: What about scaling? A: Fabra handles ~1000 requests/second per instance. Scale horizontally on any platform. Most startups never need more than 2-3 instances.

Q: Can I migrate from Feast? A: Yes. Convert YAML feature definitions to Python @feature decorators. Same data access patterns, no infrastructure tax.

Q: What if I need RAG/LLM features too? A: Fabra includes a Context Store with @retriever and @context decorators. Same infrastructure, unified deployment.

Next Steps

• Quickstart — Full setup guide
• Feast vs Fabra — Detailed comparison
• Deploy to Production — When you're ready
• Context Store — Add RAG capabilities

---

Also Building LLM/RAG Features?

Fabra includes a Context Store for LLM applications — vector search, token budgets, and full audit trails. Same infrastructure, same deployment.

Context Store → | RAG Audit Trail → | Compliance Guide →

---

You Might Also Search For

• "feature store without kubernetes"
• "feast too complex setup"
• "run feature store locally without docker"
• "simple python feature store"
• "feature store for startups"