https://github.com/robertoprevato/temp-httpx-tests
To share some performance checks for HTTP clients.
https://github.com/robertoprevato/temp-httpx-tests
Last synced: 6 months ago
JSON representation
To share some performance checks for HTTP clients.
- Host: GitHub
- URL: https://github.com/robertoprevato/temp-httpx-tests
- Owner: RobertoPrevato
- License: mit
- Created: 2022-11-25T16:39:38.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2022-11-25T18:30:33.000Z (almost 3 years ago)
- Last Synced: 2024-11-24T20:46:38.386Z (11 months ago)
- Language: Python
- Size: 10.7 KB
- Stars: 2
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
[](https://masto.ai/@robertoprevato/109401142597221543)
# temp-httpx-tests
Temporary repository to share some performance checks for HTTP clients using
asyncio.
It compares:
- HTTP GET requests made using the built-in `urllib` package inside a executor
- HTTP GET requests made using `httpx.AsyncClient`
- HTTP GET requests made using `blacksheep.client.session`
This test was only done because I am planning to remove the implementation of
[HTTP Client in BlackSheep](https://www.neoteroi.dev/blacksheep/#timeline) and
remove the parts that use `urllib`, and to include `httpx` as a dependency.
Note: `urllib` was used because in this case `blacksheep` is only using it to
make web requests to implement `OpenID Connect` integration and to fetch
well-known configuration (this kind of web requests happen seldomly during web
applications life time).
See [this discussion on Mastodon for more information](https://masto.ai/@robertoprevato/109401142597221543).
## How to run
1. create a virtual environment `python3.11 -m venv venv && source venv/bin/activate`
2. install `pip install -r requirements.txt`
3. start the `Flask` server in a terminal (`python server.py`)
4. start `test.ipy` using `iPython` with `ipython test.ipy` in a different terminal
5. read the results of `%timeit`
The test server is just a development server, but I got similar results using
`uvicorn` as HTTP server (of course, with overall better response times,
depending on the HTTP server).
---
Example results:
```bash
*** Plain text ***
Running %timeit using urllib in executor:
1.19 ms ± 70.7 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
Running %timeit using httpx.AsyncClient:
2.01 ms ± 40.3 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
Running %timeit using blacksheep.client:
1.05 ms ± 20.8 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
*** JSON ***
Running %timeit using urllib in executor:
1.2 ms ± 62.5 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
Running %timeit using httpx.AsyncClient:
2.53 ms ± 172 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
Running %timeit using blacksheep.client:
1.13 ms ± 52.3 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
```
```bash
*** Plain text ***
Running %timeit using urllib in executor:
1.2 ms ± 43 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
Running %timeit using httpx.AsyncClient:
2.01 ms ± 28.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
Running %timeit using blacksheep.client:
1.04 ms ± 29.6 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
*** JSON ***
Running %timeit using urllib in executor:
1.22 ms ± 16.8 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
Running %timeit using httpx.AsyncClient:
2.23 ms ± 100 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
Running %timeit using blacksheep.client:
1.03 ms ± 26 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
```
---
Maybe using iPython %timeit is not the right way to run performance checks,
however all three clients are tested in the same way and consistent results
were obtained testing with different HTTP servers.
---
Note that all clients successfully get the information from the server:
```ipython
Python 3.11.0 (main, Oct 30 2022, 14:13:55) [GCC 11.3.0]
Type 'copyright', 'credits' or 'license' for more information
IPython 8.6.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: import asyncio
...: import httpx
...: from example.http import HTTPHandler
...: from blacksheep.client import ClientSession
...:
...: client = httpx.AsyncClient()
...:
...: loop = asyncio.new_event_loop()
...:
...: basic_handler = HTTPHandler()
...:
...: session = ClientSession(loop)
In [2]: loop.run_until_complete(basic_handler.fetch("http://127.0.0.1:44778"))
Out[2]: b'Hello, World!'
In [3]: loop.run_until_complete(client.get("http://127.0.0.1:44778"))
Out[3]:
In [4]: loop.run_until_complete(session.get("http://127.0.0.1:44778"))
Out[4]:
```
### Parallel test
Test starting 20 web requests at a time.
```bash
ipython test-parallel.ipy
```
```
*** Plain text, parallel 20 ***
Running %timeit using urllib in executor:
18.5 ms ± 465 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
Running %timeit using httpx.AsyncClient:
31.5 ms ± 1.18 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
Running %timeit using blacksheep.client:
18.1 ms ± 264 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
```
```
*** Plain text, parallel 20 ***
Running %timeit using urllib in executor:
18.8 ms ± 481 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
Running %timeit using httpx.AsyncClient:
30.5 ms ± 586 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
Running %timeit using blacksheep.client:
17.4 ms ± 405 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
```