On 3/7/2012 4:31 PM, Daniel Stenberg wrote:
> On Tue, 6 Mar 2012, René Berber wrote:
>
>> The problem: A http connection is being aborted in libcurl with the
>> CURLE_FAILED_INIT error code. The problem is caused by the server
>> having DNS based redundancy, in this case hosts returns 8 IPs (non
>> distinct, 1 repeats), resulting in a very short time-out which I think
>> causes a signal when libcurl is just starting.
>>
>> I understand, from reading other posts, that the CURLOPT_TIMEOUT which
>> in this case is set to 30 sec is being divided by 8, so the real
>> time-out used is 3.75 sec.
>
> This particular problem was addressed with commit f4d3c0cbfb648917a0 but
> it is not yet present in any release. You could try it out though,
> either by cherry-picking the commit or by running a recent daily
> snapshot or git build.

I'll test the snapshot.

>> I wouldn't mind modifying libcurl to leave the CURLOPT_TIMEOUT alone,
>> in this use case it doesn't make sense connecting to all IPs, or
>> modifying the timeout (I do understand the logic behind that decision,
>> you do not want to have say 8 times the timeout if each IP is tried
>> one after the other, but they are tried in parallel if I understood
>> the documentation).
>
> No, they are tried sequentially and the CURLOPT_TIMEOUT is documented to
> be the maximum timeout libcurl should allow the entire operation to take
> so the code really has to do some kind of "clever" choice.
>
> The new way will at least be a lot different and should avoid this
> problem to a much larger extent.

Thanks.

-- 
René Berber
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html