Why choosing sanic
All tests were run on an AWS medium instance running ubuntu, using 1 process. Each script delivered a small JSON response and was tested with wrk using 100 connections. Pypy was tested for Falcon and Flask but did not speed up requests. It can be seen sanic has the very good result, and basically it inspired by the idea behind document: uvloop: Blazing fast Python networking
| Server | Implementation | Requests/sec | Avg Latency |
|---|---|---|---|
| Sanic | Python 3.5 + uvloop | 33,342 | 2.96ms |
| Wheezy | gunicorn + meinheld | 20,244 | 4.94ms |
| Falcon | gunicorn + meinheld | 18,972 | 5.27ms |
| Bottle | gunicorn + meinheld | 13,596 | 7.36ms |
| Flask | gunicorn + meinheld | 4,988 | 20.08ms |
| Kyoukai | Python 3.5 + uvloop | 3,889 | 27.44ms |
| Aiohttp | Python 3.5 + uvloop | 2,979 | 33.42ms |
| Tornado | Python 3.5 | 2,138 | 46.66ms |
Install latest python3 on CentOS7
# yum install https://centos7.iuscommunity.org/ius-release.rpm
# yum install python36u python36u-pipInstall dependency tools/library which need to compile httptools, uvloop
# yum install gcc python36u-develInstall sanic
# pip3.6 install sanicRunning sanic
# python3.6 test.py
test.py
-------
from sanic import Sanic
from sanic.response import json
app = Sanic()
@app.route("/")
async def test(request):
return json({"hello": "world"})
if __name__ == "__main__":
app.run(host="0.0.0.0", port=80)