Seperate keepalive 0 from socket settimeout

[cjavad]

In certain conditions (MQTT over SSL, and also some websockets stuff) _keepalive is used as a socket timeout, which is all good until used with 0, which triggers as #42 describes it "an infinite loop" preventing any connection from being established. Originally in our testing only the SSL code path was triggered, and referring an older change (#578) _connect_timeout can be used as an alternative, which also semantically makes sense.

I noticed the same value was used on another socket for websocket transports, and with some quick testing both cases fail with keepalive 0:

Using test.mosquitto.org with websockets:

...
>>> client.connect(url, port, keepalive=0)
... (stack trace)
BlockingIOError: [Errno 11] Resource temporarily unavailable

And over SSL:

...
>>> client.connect(url, port, keepalive=0)
... (stack trace)
ssl.SSLWantReadError: The operation did not complete (read) (_ssl.c:1000)

Both TLS and WebSocket rely on initial communication beyond the socket connection to establish a connection which makes a timeout of 0 impossible to respect for either protocol.

Using a seperate value which is non-zero seems to make more sense, although the meaning of keepalive fits the settimeout functionality better, so perhaps only calling it for values greater than 0?

[Sporarum]

What is the status on this PR ?

Merging it would help me a ton !

[cjavad]

Hi @Sporarum personally we ended up just using a keepalive value that was consistent with the rest of our system instead of disabling this feature, previously we maintained a runtime patch like this https://gist.github.com/cjavad/b804b4fada7ff0e18cf77d43b7f731dd, which can be adapted to also patch out websockets etc.

But i would imagine @'ing a maintainer could get this moving otherwise.

[Sporarum]

In our case, we want a long keepalive, but a short connection timeout, which does not seem possible because of this bug