Here is an example code:
http://pastebin.com/bEK8Twqe

and its debug output:
http://pastebin.com/6uyUQ50k

I did not managed to reproduce issue with just one curl so i made a
copy of what urls of real sources. I have replaced domain names and
urls as it contains key (thought not production keys).
So from the log i see that after 130 sec of waiting when i create curl and call
multi.socket_action(pycurl.SOCKET_TIMEOUT, 0)
process_curl_complete_handles()

curl says that my newly created curl is timeout outed.
whats wrong?

As i work around i am planning to remove
setopt(pycurl.CONNECTTIMEOUT, 1)
setopt(pycurl.TIMEOUT, 5)
and call socket_all() instead of socket_action(pycurl.SOCKET_TIMEOUT, 0)

If anyone need more debug info - i can modify example code in any way
or send it with original urls in private email.

2012/4/26 Utsav Sabharwal <tashywashysachy_at_live.in>:
> @Evgeny: Can you please share the source code?
>
>
>> Date: Thu, 26 Apr 2012 19:44:16 +0200
>> From: daniel_at_haxx.se
>> To: curl-and-python_at_cool.haxx.se
>> Subject: Re: CURLE_OPERATION_TIMEDOUT while connecting.
>> CC: curl-library_at_cool.haxx.se
>
>>
>> On Thu, 26 Apr 2012, Evgeny Turnaev wrote:
>>
>> > I am trying to make use of pycurl with multi interface + pyev and there
>> > are some issues i dont understand from docs:
>>
>> And with multi interface you mean the multi_socket API as I understand it.
>>
>> > I am not planning to use curl timeout system but to use libev timers and
>> > manually remove_handle from multi when timeout happens.
>>
>> Sorry, but I don't understand how. What timeout values will you use if you
>> don't use the ones libcurl provide? Also, I really strongly advice you to
>> use
>> the timeout callback to get the timeout info. I think you'll see that it
>> is
>> the smoothest path forward when using the multi_socket API.
>>

I use libev to fetch non-http sources and also to manage my own
application timeouts with pyev.Timer
so it is more convenient to me to use pyev timers for curl. Sorry, i
don't see any decent way to make curl timeouts work together with
libev loop.
Besides I've been told by few people that used curl before that curl
did not managed timeouts properly.
I also found in tornado (http://www.tornadoweb.org/) sources comments
that curl sometimes forgets timeouts and so tornado people make they
own checking of timeouts periodically.

>> > 1) I create curl and and handle to multi with multi.add_handle(c) but
>> > if dont call multi.socket_action(pycurl.SOCKET_TIMEOUT, 0) nothing
>> > fetches.
>>
>> Exactly. Because you ignore libcurl's timeouts you don't call libcurl
>> properly
>> on the timeout that it has set, so you do it blindly instead and happens
>> to
>> get the same effect.
>>
>> > But there is also multi.socket_all() witch is seems to be doing the
>> > same?
>>
>> It doesn't do the same, but in this case the end result will be the same.
>>
>> > Is there any difference in calling
>> > multi.socket_action(pycurl.SOCKET_TIMEOUT, 0) or multi.socket_all()
>> > What is cheaper?
>>
>> The first is MUCH cheaper, as the second will act on every single handle
>> that
>> you have added and the first will only act on the handles that actually
>> have
>> timed out. curl_multi_socket_all() is also deprecated so you really
>> shouldn't
>> use it.
>>

Thats sad. Do curl sets some kind of default timeouts? what will
happen if dont call
setopt(pycurl.TIMEOUT, 5)
and then call
socket_action(pycurl.SOCKET_TIMEOUT, 0)
will i ever receive timeout?

>> > 2) the main question: when a keepalive socket timeout outed and curl
>> > thinks
>> > connection is dead and tells so via error list of multi.info_read() do i
>> > have to manually start another Curl() instance in order to fetch url
>> > finally? (i thought curl will just ask for another socket)
>>
>> A timeout error is an error that will stop that particular transfer. If
>> you
>> want to start another transfer on that handle, you need to remove it and
>> re-add it to the multi-handle.
>>
>> > Does creating pool of curls will somehow help in such situation?
>>
>> I'm not following what you're asking here so I don't know.
>>
>> > CURLE_OPERATION_TIMEDOUT means connect timeout or fetch timeout or both?
>> > Can
>> > it be related to any other?
>>
>> It means it reached a timeout value internally. I think the error code is
>> restricted to only be returned if one of the user provided limits are
>> reached.
>>
>> > Why Connection #0 gets closed 2 times?
>>
>> I don't know, I don't think there was enough details for me to figure out
>> what
>> was really going on there.
>>
>> --
>>
>> / daniel.haxx.se
>> _______________________________________________
>> http://cool.haxx.se/cgi-bin/mailman/listinfo/curl-and-python
>
> _______________________________________________
> http://cool.haxx.se/cgi-bin/mailman/listinfo/curl-and-python
>

-- 
--------------------------------------------
Турнаев Евгений Викторович
+7 906 875 09 43
--------------------------------------------
_______________________________________________
http://cool.haxx.se/cgi-bin/mailman/listinfo/curl-and-python