requests.{get,post} timeout

[Chris Angelico]

On Wed, Aug 23, 2017 at 4:14 AM, Jon Ribbens <jon+usenet at unequivocal.eu> wrote:
> On 2017-08-22, Chris Angelico <rosuav at gmail.com> wrote:
>> On Wed, Aug 23, 2017 at 2:58 AM, Jon Ribbens <jon+usenet at unequivocal.eu> wrote:
>>> Yes. There is no timeout feature that can be used to limit the total
>>> time a 'requests' request takes. Some people might think that this is
>>> a serious flaw in the requests library that would need urgent
>>> rectification in order to make the library safe and useful to use in
>>> almost any situation, but the 'requests' developers are apparently not
>>> among those people.
>>
>> I'm not either. The idea of a timeout is to detect when something's
>> completely not working, not to limit the overall time to process.
>
> We appear to have different understandings of the word "timeout".
> I think it means a time, which if it runs out, will stop the operation.
>
> I am somewhat surprised that anyone might have a different definition
> - not least because, from a human being's point of view, they care
> about the overall time something takes to happen and telling them that
> nothing's wrong because technically we are still "successfully" receiving
> the expected 10 kilobytes of data 3 hours later is unlikely to make
> them happy.

You start downloading a file from a web page. It stalls out.

Is it merely slow, and continuing to wait will get you a result?

Or has it actually stalled out and you should give up?

The low-level timeout will distinguish between those. If you want a
high-level timeout across the entire job, you can do that too, but
then you have to figure out exactly how long is "too long". Let's say
you set a thirty-second timeout. Great! Now someone uses your program
on a midrange connection to download a 100MB file, or on a poor
connection to download a 5MB file, or on dial-up to download a 10KB
file. Data is constantly flowing, but at some point, the connection
just dies, because it's hit your timeout. This is EXTREMELY
frustrating.

You can always add in the overall timeout separately. If the low-level
timeout were implemented that way, there would be no way to externally
add the other form of timeout. Therefore the only sane way to
implement the request timeout is a between-byte limit.

ChrisA