Thor performance benchmark about websockets

[aaugustin]

Based on the line you're mentionning, it seems that websockets is receiving a message aimed at a server when it's configured as a client, or vice versa.

Are you sure you're running a websockets server and not a client? Could you share the code you're using to start your websockets test server?

[achille-roussel]

I basically copy/pasted the code available in the documentation, simply trying to make a echo server:
Something like that:

import asyncio
import websockets

@asyncio.coroutine
def hello(websocket, path):
    while True:
        message = yield from websocket.recv()
        if message is None:
            break
        yield from websocket.send(message)

start_server = websockets.serve(hello, 'localhost', 8080)

asyncio.get_event_loop().run_until_complete(start_server)
asyncio.get_event_loop().run_forever()  

Then I ran thor with something like "thor --amount 10 --messages 10 ws://localhost:8080"

[aaugustin]

As far as I can tell, you must use the --masked option of Thor for testing client-to-server connections. I have no idea why this is an option, it's clearly mandatory!

Not checking the mask bit would create a security issue. See https://tools.ietf.org/html/rfc6455#section-5.3. I'm surprised that many implementation ignore it.

Can you give a few examples of implementations you have tested?

[achille-roussel]

I've tried with gevent-websocket, Qt Websocket, websocket.io and node.js websocket. All successfully accepted thor's default mode.

--masked made it work, thanks for taking the time to help with that :)

[aaugustin]

Excellent.

You can file bugs against these four implementations becase they don't honor https://tools.ietf.org/html/rfc6455#section-5.1:

The server MUST close the connection upon receiving a frame that is not masked.

[aaugustin]

I just filed a bug against Thor: observing/thor#8