stent 1.0.0

Distributed Durable Functions in Python

Stent

Distributed durable functions for Python. Write reliable, stateful workflows using async/await.

pip install stent

Quick Example

import asyncio
from stent import Stent, Result

@Stent.durable()
async def process_order(order_id: str) -> dict:
    await asyncio.sleep(1)  # Simulate work
    return {"order_id": order_id, "status": "processed"}

@Stent.durable()
async def order_workflow(order_ids: list[str]) -> Result[list, Exception]:
    results = []
    for order_id in order_ids:
        result = await process_order(order_id)
        results.append(result)
    return Result.Ok(results)

async def main():
    backend = Stent.backends.SQLiteBackend("workflow.db")
    await backend.init_db()
    executor = Stent(backend=backend)
    
    worker = asyncio.create_task(executor.serve())
    
    exec_id = await executor.dispatch(order_workflow, ["ORD-001", "ORD-002"])
    result = await executor.wait_for(exec_id)
    print(result.value)

asyncio.run(main())

Why Stent?

Feature	Temporal	Celery	Prefect	Airflow	Stent
Durable Execution	Yes	No	Partial	No	Yes
Setup Complexity	High	Medium	Medium	High	Very Low
Infrastructure	Server cluster	Broker	Server	Multi-component	SQLite/Postgres
Native Async	Yes	No	Yes	Limited	Yes

Stent fills the gap between simple task queues (Celery) and enterprise platforms (Temporal):

• vs Temporal: Same durability guarantees, fraction of the infrastructure
• vs Celery/Dramatiq: True workflow durability, not just task retries
• vs Prefect/Airflow: Application workflows, not batch data pipelines

See full comparison for details.

Features

• Durable Execution - Workflow state survives crashes and restarts
• Automatic Retries - Configurable retry policies with exponential backoff
• Distributed Workers - Scale horizontally across multiple processes
• Parallel Execution - Fan-out/fan-in with asyncio.gather and Stent.map
• Rate Limiting - Control concurrent executions per function
• External Signals - Coordinate workflows with external events
• Dead Letter Queue - Inspect and replay failed tasks
• Idempotency & Caching - Prevent duplicate work
• Multiple Backends - SQLite (dev) or PostgreSQL (production)
• OpenTelemetry - Distributed tracing support

Key Concepts

from stent import Stent, RetryPolicy

# Configurable retry policies
@Stent.durable(
    retry_policy=RetryPolicy(max_attempts=5, initial_delay=1.0),
    queue="high_priority",
    max_concurrent=10,  # Rate limiting
    idempotent=True,    # Prevent duplicate execution
)
async def my_activity(data: dict) -> dict:
    ...

# Durable sleep (doesn't block workers)
await Stent.sleep("30m")

# Parallel execution
results = await asyncio.gather(*[process(item) for item in items])
# Or optimized for large batches:
results = await Stent.map(process, items)

# External signals
payload = await Stent.wait_for_signal("approval")
await executor.send_signal(exec_id, "approval", {"approved": True})

Backends

# SQLite (development)
backend = Stent.backends.SQLiteBackend("stent.db")

# PostgreSQL (production)
backend = Stent.backends.PostgresBackend("postgresql://user:pass@host/db")

# Optional: Redis for low-latency notifications
executor = Stent(
    backend=backend,
    notification_backend=Stent.notifications.RedisBackend("redis://localhost")
)

CLI

stent list                    # List executions
stent show <exec_id>          # Show execution details
stent dlq list                # List dead-lettered tasks
stent dlq replay <task_id>    # Replay failed task

Documentation

Full documentation available in docs/:

• Getting Started | Core Concepts | Comparison
• Guides: Durable Functions | Orchestration | Error Handling | Parallel Execution | Signals | Workers | Monitoring
• Patterns: Saga | Batch Processing
• Reference: API | Configuration | Deployment

Examples

See examples/ for complete workflows:

• simple_flow.py - Basic workflow
• saga_trip_booking.py - Saga pattern with compensation
• batch_processing.py - Fan-out/fan-in
• media_pipeline.py - Complex multi-stage pipeline

Requirements

• Python 3.12+
• aiosqlite or asyncpg (backend)
• redis (optional, for notifications)

License

MIT