https://github.com/vikyw89/fastapi-deployment-benchmark

https://github.com/vikyw89/fastapi-deployment-benchmark

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/vikyw89/fastapi-deployment-benchmark
• Owner: vikyw89
• Created: 2024-08-05T14:27:30.000Z (5 months ago)
• Default Branch: alpha
• Last Pushed: 2024-08-14T12:48:58.000Z (4 months ago)
• Last Synced: 2024-10-12T08:45:22.788Z (2 months ago)
• Language: Python
• Size: 49.8 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# FastAPI Benchmark

This README provides a benchmark comparison between synchronous and asynchronous endpoints in FastAPI, running under different server configurations.

## Server Code

```python

@app.get("/sync/{item_id}")

def read_item(item_id: int, q: Union[str, None] = None):

    time.sleep(1)

    return {"item_id": item_id, "q": q}

@app.get("/async/{item_id}")

async def aread_item(item_id: int, q: Union[str, None] = None):

    await asyncio.sleep(1)

    return {"item_id": item_id, "q": q}

```

## Benchmark Results

### Running with `fastapi run benchmark/main.py`

| Endpoint | Command                                                 | Requests per second (mean) |

| -------- | ------------------------------------------------------- | -------------------------- |

| Async    | `ab -c 10000 -n 10000 "http://localhost:8000/async/31"` | 1496.38 [#/sec]            |

| Sync     | `ab -c 10000 -n 10000 "http://localhost:8000/sync/31"`  | 39.48 [#/sec]              |

### Running with `poetry run gunicorn benchmark.main:app --workers 4 --worker-class uvicorn.workers.UvicornWorker`

| Endpoint | Command                                                 | Requests per second (mean) |

| -------- | ------------------------------------------------------- | -------------------------- |

| Async    | `ab -c 10000 -n 10000 "http://localhost:8000/async/31"` | 3462.74 [#/sec]            |

| Sync     | `ab -c 10000 -n 10000 "http://localhost:8000/sync/31"`  | 105.77 [#/sec]             |

### Running with `poetry run gunicorn benchmark.main:app --workers 12 --worker-class uvicorn.workers.UvicornWorker`

| Endpoint | Command                                                 | Requests per second (mean) |

| -------- | ------------------------------------------------------- | -------------------------- |

| Async    | `ab -c 10000 -n 10000 "http://localhost:8000/async/31"` | 7593.36 [#/sec]            |

| Sync     | `ab -c 10000 -n 10000 "http://localhost:8000/sync/31"`  | 155.80 [#/sec]             |

### Running with `uvicorn benchmark.main:app --workers 4 --host 0.0.0.0 --port 8000`

| Endpoint | Command                                                 | Requests per second (mean) |

| -------- | ------------------------------------------------------- | -------------------------- |

| Async    | `ab -c 10000 -n 10000 "http://localhost:8000/async/31"` | 7507.83 [#/sec]            |

## Conclusion

- Asynchronous endpoints significantly outperform synchronous ones in terms of requests per second.

- Increasing the number of workers with Gunicorn improves the performance for both synchronous and asynchronous endpoints.

- The asynchronous approach is particularly beneficial for I/O-bound operations, as demonstrated in this benchmark.

### @vikyw89-20240805