superdanby/tcp_server

tcp_server's Issues

Weird performance effect with str.decode()

server.py now yields weird benchmark results. The server clean up time takes as long as the process time.

~> python3 client.py 127.0.0.1 8888 1000000
eof sent
finish writing
wait for server to clean up
1000000 requests done.
Process time: 33.637513160705566s, requests per second: 29728.71374951026, server clean up time: 32.18577456474304s, total time elapsed: 65.82328772544861s

The part in async def respond() causing this problem:

    if Error:
        status_code = status_code.decode('utf-8').split(' ', 2)[1]
    else:
        status_code, Error = status_code.decode('utf-8').split(' ', 1)[1].split(' ', 1)
    # status_code, Error = status_code.decode('utf-8').split(' ', 1)[1].split(' ', 1)
    # status_code = status_code.decode('utf-8')
    # status_code = "212".split('1')[0]

Changing those to the following yields similar results:

    # if Error:
    #     status_code = status_code.decode('utf-8').split(' ', 2)[1]
    # else:
    #     status_code, Error = status_code.decode('utf-8').split(' ', 1)[1].split(' ', 1)
    # status_code, Error = status_code.decode('utf-8').split(' ', 1)[1].split(' ', 1)
    status_code = status_code.decode('utf-8')
    # status_code = "212".split('1')[0]

But if the part doesn't use str.decode():

    # if Error:
    #     status_code = status_code.decode('utf-8').split(' ', 2)[1]
    # else:
    #     status_code, Error = status_code.decode('utf-8').split(' ', 1)[1].split(' ', 1)
    # status_code, Error = status_code.decode('utf-8').split(' ', 1)[1].split(' ', 1)
    # status_code = status_code.decode('utf-8')
    status_code = "212".split('1')[0]

The results are:

~> python3 client.py 127.0.0.1 8888 1000000
eof sent
finish writing
wait for server to clean up
1000000 requests done.
Process time: 66.861887216568s, requests per second: 14956.203625556738, server clean up time: 0.0014019012451171875s, total time elapsed: 66.86328911781311s

Though total time elapsed don't differ a lot, we can cut out the connection from the client once all requests are received(the time it takes would be 'process time'). Hence, it would be a huge impact on the performance.

Python server gets empty string 100 seconds after a connection is established

This makes the server stuck in a state that it only outputs an empty string and not accepting new messages.

Python server outputs 'Task was destroyed but it is pending!'

The 'Task was destroyed but it is pending!' is sometimes being output when a connection is closed by a client. While this doesn't have a great impact on the server's performance, it is worth investigating.