deliberate versus os socket timeout

[John J. Lee]

Robin Becker <robin at reportlab.com> writes:

> While messing about with some deliberate socket timeout code I got an
> unexpected timeout after 20 seconds when my code was doing
> socket.setdefaulttimeout(120).
>
> Closer inspection revealed that this error in fact seemed to come from
> the os (in this case windows xp).
>
> By inspection of test cases the error.reason from the deliberate
> socket timeout looks like
>     'timed out'
> whereas the windows caused timeout error.reason looks like
>     '(10060 operation timed out)'
> 
> it would be nice to know if that is in fact true and whether there is
> any way to do the attribution of errors more sensibly.

What do you mean "is in fact true"?  Is what true?  That the reason is
always formatted in those ways when those particular errors occur?
Not sure there's any reason to rely on that: I'd expect .reason to be
an exception object, so (reading the docs for module socket)
e.reason.args[0] should be the error code if one is available, in this
case errno.ETIMEDOUT (i.e. WSAETIMEDOUT, which means the timeout
occurred before a connection was established).  Sometimes
e.reason.args will be length-1, and e.reason.args[0] will be a string,
again according to the docs for module socket.


> Both of these seem to cause urllib2.URLError and presumably appear
> somewhere in the socket code.
>
> It might be nice if the deliberate timeout could be something like
> timed out deliberately after xxx seconds'.

By "deliberate timeout", you mean the one you asked for by calling
socket.setdefaulttimeout().

In that case, I think e.reason will be a socket.timeout instance (as
opposed to socket.error in the WSAETIMEOUT case).


> More importantly is there anything I can do to avoid these wrong os
> inspired timeouts? I'm just using urllib2 to read from a remote site
[...]

Please define "wrong" ;-)

Googling suggests Windows doesn't let you configure its connect
timeout, except through a registry setting:

http://www.tech-archive.net/Archive/VC/microsoft.public.vc.language/2005-09/msg00918.html


John