FastAPI Async vs Sync – When to Use Each, Benchmarks, and Real‑World Tips

Introduction

If you’ve been building APIs with FastAPI for a few months, you’ve probably heard the phrase fastapi async vs sync tossed around in tutorials and conference talks. The framework makes it easy to write both asynchronous (async def) and synchronous (def) route handlers, but the decision isn’t just a matter of “pick one”. It impacts latency, throughput, database driver choices, and even how you write tests. In this post we’ll peel back the abstractions, walk through concrete code, benchmark the two styles, and give you a checklist for picking the right approach in production.

TL;DR – Use async when you have awaitable I/O (HTTP calls, async DB drivers, websockets). Stick with sync for CPU‑bound work or when you’re locked into legacy blocking libraries.

Let’s dive into the nitty‑gritty of fastapi async vs sync.

Understanding Async in FastAPI

FastAPI is built on top of Starlette, which itself runs on ASGI (Asynchronous Server Gateway Interface). An ASGI server (e.g., Uvicorn, Hypercorn) spins up an event loop and schedules coroutines. When you declare a route as async def, FastAPI registers the coroutine with the loop; when the coroutine hits an await, control is handed back to the loop, allowing other requests to run.

from fastapi import FastAPI

app = FastAPI()

@app.get("/sync")
def sync_endpoint():
    # Simulated blocking I/O
    time.sleep(2)
    return {"msg": "sync done"}

@app.get("/async")
async def async_endpoint():
    # Simulated non‑blocking I/O
    await asyncio.sleep(2)
    return {"msg": "async done"}

In the sync version, time.sleep blocks the worker thread, preventing the event loop from processing other connections. In the async version, asyncio.sleep yields control, so a single worker can handle many concurrent requests.

Key points to remember:

When to Choose Async vs Sync Endpoints

1. I/O‑Bound Work (Network, Database, Files)

If your endpoint talks to an async‑compatible service - httpx.AsyncClient, databases library, async SQLAlchemy 2.0, or an async Redis client - use async. The event loop can interleave many DB calls without spawning extra threads.

from fastapi import FastAPI, Depends
import httpx

app = FastAPI()

async def fetch_user(user_id: int):
    async with httpx.AsyncClient() as client:
        r = await client.get(f"https://jsonplaceholder.typicode.com/users/{user_id}")
        r.raise_for_status()
        return r.json()

@app.get("/users/{user_id}")
async def get_user(user_id: int):
    user = await fetch_user(user_id)
    return {"user": user}

2. CPU‑Bound Work (Heavy Computation, Image Processing)

Async does not magically make CPU work faster. The coroutine still runs on the same thread, so a long‑running CPU task will block the loop and starve other requests. In such cases, either keep the endpoint def and let FastAPI offload it to a thread pool, or explicitly run it in a background task/executor.

import hashlib

def compute_hash(data: bytes) -> str:
    return hashlib.sha256(data).hexdigest()

@app.post("/hash")
def hash_endpoint(payload: bytes):
    # FastAPI will run this in a thread pool automatically
    return {"hash": compute_hash(payload)}

3. Mixed Environments

Sometimes you have a mixture of async and sync libraries. You can keep the route async and wrap blocking calls in run_in_threadpool:

from fastapi.concurrency import run_in_threadpool

@app.get("/legacy")
async def legacy_endpoint():
    result = await run_in_threadpool(some_blocking_library.do_work)
    return {"result": result}

4. Deployment Constraints

If you’re deploying to a serverless platform that expects a WSGI entry point (e.g., AWS Lambda with Mangum), you’ll need to stick with sync functions or use a compatibility layer. Most modern containers (AWS Fargate, GCP Cloud Run) run ASGI natively, so async is safe.

Performance Benchmarks and Trade‑offs

Below is a concise benchmark we ran on a 2023‑class Intel i7, using Uvicorn with 4 workers. The test hits a simple endpoint that awaits asyncio.sleep(0.05) (async) or calls time.sleep(0.05) (sync). We measured requests per second (RPS) and average latency with hey (100 concurrent connections, 10 000 total requests).

What the numbers tell us

• Throughput: Async endpoints can handle 4–5× more requests under identical I/O latency because the event loop never idles waiting for the sleep or DB round‑trip.
• Latency: The async version stays close to the raw I/O time, while sync adds the thread‑context‑switch overhead and blocks other requests.
• CPU Utilization: With sync, each worker thread is busy sleeping, wasting CPU cycles. Async keeps the CPU free for other coroutines.

Trade‑offs

If you’re on a low‑memory container, async can be a win. If you’re tied to a legacy blocking ORM, you may accept the extra threads.

Common Pitfalls and Debugging Async Code

1. Forgot to await

   async def fetch():
       return httpx.AsyncClient().get(url)   # ❌ returns coroutine, never executed

   The request never runs, leading to timeouts. Always await the coroutine.

2. Mixing sync DB drivers with async endpoints
   Using psycopg2 (blocking) inside an async route will block the loop. Wrap it:

   from fastapi.concurrency import run_in_threadpool
   
   @app.get("/users/{id}")
   async def get_user(id: int):
       row = await run_in_threadpool(lambda: sync_pg_query(id))
       return row

   For a deeper dive on this exact error, see our post on Fixing SQLAlchemy MissingGreenlet Error in FastAPI (Async Explained).

3. Lifecycle events that aren’t async‑aware
   If you create a global async DB pool in a startup event but forget to await its creation, the app may start handling requests before the pool is ready, raising RuntimeError: Event loop is closed. The article on FastAPI Lifespan vs Startup Events: The Mistake That Breaks Async Apps walks through the correct pattern.

4. Blocking the event loop with CPU‑heavy code
   Even a single for i in range(10**7): pass will stall all connections. Offload to a thread pool or use a background task (fastapi.BackgroundTasks).

5. Missing contextvars across threads
   If you rely on contextvars for request‑scoped data, remember that run_in_threadpool creates a new thread and loses the context unless you propagate it manually.

Debugging Tips

• uvicorn —log-level debug – Shows which coroutine is awaiting what.
• asyncio.all_tasks() – Inspect pending tasks during a breakpoint.
• aiomonitor – An interactive console for live inspection of the event loop.

Integrating Async Database Drivers with FastAPI

1. Asyncpg (PostgreSQL)

import asyncpg
from fastapi import FastAPI, Depends

app = FastAPI()
pool: asyncpg.Pool | None = None

@app.on_event("startup")
async def create_pool():
    global pool
    pool = await asyncpg.create_pool(dsn="postgresql://user:pwd@localhost/db")

@app.on_event("shutdown")
async def close_pool():
    await pool.close()

async def get_conn() -> asyncpg.Connection:
    async with pool.acquire() as conn:
        yield conn

@app.get("/items/{item_id}")
async def read_item(item_id: int, conn: asyncpg.Connection = Depends(get_conn)):
    row = await conn.fetchrow("SELECT * FROM items WHERE id = $1", item_id)
    return dict(row)

2. SQLAlchemy 2.0 Async Engine

SQLAlchemy 2.0 introduced a native async API that plays nicely with FastAPI. The pattern mirrors the asyncpg example but uses AsyncSession.

from sqlalchemy.ext.asyncio import AsyncEngine, create_async_engine, AsyncSession
from sqlalchemy.orm import sessionmaker
from fastapi import Depends

DATABASE_URL = "postgresql+asyncpg://user:pwd@localhost/db"
engine: AsyncEngine = create_async_engine(DATABASE_URL, echo=False)
AsyncSessionLocal = sessionmaker(engine, class_=AsyncSession, expire_on_commit=False)

async def get_async_session() -> AsyncSession:
    async with AsyncSessionLocal() as session:
        yield session

@app.get("/users/{uid}")
async def get_user(uid: int, session: AsyncSession = Depends(get_async_session)):
    result = await session.execute(select(User).where(User.id == uid))
    user = result.scalar_one_or_none()
    return user

Pro tip: If you see “MissingGreenlet” errors, double‑check that you’re not mixing sync Session objects with async code. Our guide on Fixing SQLAlchemy MissingGreenlet Error in FastAPI (Async Explained) explains the root cause and fixes.

3. Detecting Session Leaks

Long‑running async endpoints can inadvertently keep DB connections open. Use the monitoring utilities from our article on FastAPI SQLAlchemy Session Leak Detection to log checkout times and automatically close idle sessions.

Testing Async Endpoints Effectively

Testing async routes is straightforward with pytest and httpx.AsyncClient. The key is to use an async test function and await the client calls.

import pytest
from httpx import AsyncClient
from myapp.main import app  # assuming your FastAPI instance lives here

@pytest.mark.asyncio
async def test_async_endpoint():
    async with AsyncClient(app=app, base_url="http://test") as ac:
        response = await ac.get("/async")
    assert response.status_code == 200
    assert response.json() == {"msg": "async done"}

Testing Sync Endpoints

Even sync routes can be exercised with the same async client; FastAPI internally runs them in a thread pool, so you don’t need a separate test setup.

Mocking Async Dependencies

When mocking an async DB call, use AsyncMock (Python 3.8+) to preserve the coroutine contract.

from unittest.mock import AsyncMock, patch

@patch("myapp.deps.get_async_session")
@pytest.mark.asyncio
async def test_user_fetch(mock_session):
    mock_conn = AsyncMock()
    mock_conn.fetchrow.return_value = {"id": 1, "name": "Alice"}
    mock_session.return_value.__aenter__.return_value = mock_conn

    async with AsyncClient(app=app, base_url="http://test") as ac:
        resp = await ac.get("/users/1")
    assert resp.json()["name"] == "Alice"

Load‑testing Async Endpoints

For realistic load testing, tools like k6 or hey work the same, because they treat your API as a black box. Just remember to spin up enough Uvicorn workers (uvicorn myapp:app --workers 4) to saturate the CPU and observe the async benefits.

Choosing the Right Strategy for Your Project

Key Takeaways

• fastapi async vs sync isn’t a binary “pick one forever”; it’s a design decision based on I/O patterns, library compatibility, and deployment constraints.
• Async routes shine when the request spends most of its time waiting on non‑blocking I/O (HTTP calls, async DB drivers, websockets).
• Sync routes remain the safest choice for CPU‑heavy work or when you must use blocking third‑party libraries.
• Benchmarks show up to 5× higher throughput for pure I/O workloads when using async, but the gains evaporate for CPU‑bound tasks.
• Common pitfalls include forgetting await, mixing blocking DB drivers with async code, and unintentionally blocking the event loop. Use run_in_threadpool or migrate to async‑compatible libraries.
• Integrating async DB drivers (asyncpg, SQLAlchemy 2.0) requires proper startup/shutdown handling; see our related posts for troubleshooting.
• Testing async endpoints is just as easy as sync with pytest + httpx.AsyncClient; remember to mock async dependencies with AsyncMock.

By understanding the trade‑offs and applying the patterns above, you’ll be able to decide confidently when fastapi async vs sync matters most, build high‑performance APIs, and avoid the hidden bugs that often accompany asynchronous code. Happy coding!