Add optional timeout parameter

[vinnyfuria]

This is a perquisite to fixing an issue in the radiotherm component of HASS

The problem is, the radiotherms have pretty unreliable networking. Resulting in timeouts in about 1% of requests. By default though, the urllib has a really long timeout, which results in the HASS component crapping itself. So we need a way to override the default tiemout, which requires this fix.

For some further context, I did some timing tests against my CT50, and saw the following:

Count:    17121
Min:      2.238989
Max:      61.574224
Mean:     2.549727
dSquared: 27086.792889
Variance: 1.582080
Stdev:    1.257808

4 seconds seemed a reasonable default for use on local networks; happy to leave that at urllib's default if consensus is that makes more sense.

[vinnyfuria]

@mhrivnak Are you, or someone else, actively maintaining this library? This fix, or something similar, is needed to fix this issue in Home Assistant.

[cjkrolak]

@vinnyfuria I have two 3m50 thermostats and found that the 4 second timeout introduced here is too short, it results in random timeouts on both thermostats during the get method, so I am using a 10 second timeout instead, a 6 sigma limit so that the request never times out just due to chance.

Your statistics above result in a z-score of 1.17, [i.e. (4 - 2.549727) / 1.257808] or the 88th percentile, meaning you should be timing out with a 4 second limit about 12% of the time just due to chance. Are you seeing any timeouts with this 4 second limit imposed?

[vinnyfuria]

@cjkrolak I'm logging about 10 timeouts a day using the 4 second timeout and sampling once a minute. I noticed diminishing returns increasing this value as I saw that the thermostat would often just not respond at all in many cases.

The value is configurable in the radiotherm library now -- if you're having problems in home assistant, it would be pretty easy to add a configuration to override the default.

(I realized long after the fact that my statistics above were skewed by the tool I was using having an unaccounted timeout of 60 seconds, so non-responsive queries appeared as responses that took about 60 seconds)