Sanic: aiohttp.ClientSession must be create in each request?

[qwIvan]

https://github.com/channelcat/sanic/blob/master/examples/aiohttp_example.py#L23
Why not to save as a global variable, and using it in next request?

[Zheaoli]

Here's my idea.
I think session is different between one in request and one in other request.And Sanic don't have mechanism like Flask to make a context ,to make sure that same code in different request will show different result.
So,It shouldn't be a global variable,I guess

[r0fls]

@qwIvan although I said this was fine in chat I'm not sure. Since it uses a context manager you'll be skipping whatever portion happens on exit each time. I would definitely test this and see how it works. We did see performance improvements in the asyncpg example by maintaining a global connection pool.

[Zheaoli]

@r0fls I use a global connection pool just like this

~~~Python
app = Sanic()
asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())

async def get_pool(args, kwargs):
args[0].pool = {
"aiomysql": await aiomysql.create_pool(host="127.0.0.1", user="root", password="", db="test_sanic",
maxsize=5),
"mysql": connect(host="127.0.0.1", user="root", password="*", db="test_sanic", charset='utf8mb4',
cursorclass=cursors.DictCursor)}

@app.route("async_count")
async def test(request):
async with app.pool['aiomysql'].acquire() as conn:
async with conn.cursor() as cur:
await cur.execute("select count(*) from wocao ")
res = await cur.fetchall()
if res or len(res) > 0:
data = {}
data['data'] = res
return json(data)

@app.route("/normal_count")
async def test1(request):
connection = app.pool["mysql"]
with connection.cursor() as cursor:
cursor.execute("set autocommit = 1")
connection.commit()
cursor.execute("select count(*) from wocao")
res = cursor.fetchall()
if res or len(res) > 0:
data = {}
data['data'] = res
return json(data)

if __name__ == '__main__':
app.run(host="127.0.0.1", workers=4, port=12000, before_start=get_pool)

~~~

Could I make a PR for a new example ?

[r0fls]

@Zheaoli yes, please do, that would be great. We don't have an aiomysql example yet.

[r0fls]

I'll probably have feedback for you on the PR though, just FYI. Haven't fully read this script over yet.

[Hellowlol]

@qwIvan

I dont know if this is the best way but i solve this by adding the http/db access to the app.

# inside my utils.py

import aiohttp
import async_timeout

async def fetch(session, url, data=None, method='GET', r_type='json'):

    async with session.request(method, url, data=data) as response:
        if r_type == 'json':
            response = await response.json()
        elif r_type == 'text':
            response = await response.text()

        return response


async def http(url, *args, timeout=15, data=None, headers=None, method='GET', r_type='json'):
        uri = url

        if headers is None:
            headers = {}

        with async_timeout.timeout(timeout):
            async with aiohttp.ClientSession(headers=headers) as session:
                response = await fetch(session, uri, data=data, method=method)
                return response

So inside the handles i can just do:
````
from utils import http

@app.listener('before_server_start')
async def setup_something(app, loop):
app.http = http

@app.route('/')
async def handle(request):
http = await request.app.http(someurl)
````
I dont know if this is the best way, but its very nice way for db access if your using blueprints

[seemethere]

I'd suggest going with @Hellowlol's approach.

We need some sort of documentation to support this style of doing things...

[wonderbeyond]

@seemethere @Hellowlol
Isn't @Hellowlol's approach still creating a session in each request?

[Shivelight]

This is how I'm doing it:

@app.listener('before_server_start')
async def aiohttp_setup(app, loop):
    session = aiohttp.ClientSession(loop=loop)  # or your func to setup the session
    app.http = session