Re: [PATCH] Naive approach to HTTP-timeouts

Hi, David.
Yes, I have thought about exactly the same:
either difftime+last_check or thread-switch.
I didn't compare the efficient of thease methods.
I'v made thread because switch occur 1 time at minute (or any configured
timeout) but difftime can occur many times in minute (unpredictable, depends
on load).
PS. I'm not running sms gateway part of kannel,
so I have not load on http server,
so I'm ready to vote for your difftime+last_check solution ;).

David Schmitz wrote:
> Hi Slava (and others),
> 
> yeah, you have got a point there...
> But this can be avoided by using something like:
> 
> last_check = time(NULL);
> while....
> ....
> now = time(NULL);
>        if ((ret == 0) || (difftime(now, last_check) >= MIN_TIME_OUT)) {
>            check_timeout();
>            last_check = time(NULL);
>        }
> ....
> 
> It depends on whether you prefer a contex-switch (because of a thread) 
> to the calculation of a difftime.
> 
> Regards,
> David
> 
> Vjacheslav Chekushin schrieb:
> 
>> Hi, David.
>>
>> Ok. Let's suppose that we get realy high load on kannel http server.
>> So we get, let's say, ~5 new connection in second.
>> Threrefore we will check for timeouts ~5 times in second. I think it 
>> is not
>> efficient way.
>> That why I have created the thread to deal with it.
>> You must to choose between wasted thread's resources and perfomance
>> degradation. My choice was - thread's resources.
>> You can make your own choice.
>>
>> David Schmitz wrote:
>>
>>> Hi Slava,
>>>
>>> argh, correct. An IMO easy solution is to call check_timeout even if 
>>> the recent poll did not timeout, as in:
>>> [snip]
>>> if ((ret = gwthread_poll(tab, n, MIN_TIME_OUT)) < 1) {
>>>            switch(ret) {
>>>            case 0: /* timeout */
>>>                break;
>>>            case -1: /* error */
>>>                if (errno != EINTR) /* a signal was caught during 
>>> poll() function */
>>>                    warning(0, "HTTP: gwthread_poll failed.");
>>>                continue;
>>>                break;
>>>            default: /* should not happen */
>>>                panic(0, "This is not supposed to happen!");
>>>            }
>>>        }
>>>        check_timeout();
>>> [snap]
>>> That should take care of loaded environments...
>>>
>>> Regards,
>>> David
>>>
>>> Vjacheslav Chekushin schrieb:
>>>
>>>> Hi, David.
>>>>
>>>> If I understand correctly then check_timeouts() will be called only 
>>>> if no any new connection to kannel http server for MIN_TIME_OUT time.
>>>> So for busy enviroment this check will be quite undeterminate and can
>>>> be not called for hours. Isn't it?
>>>>
>>>> David Schmitz wrote:
>>>>
>>>>> Hi again,
>>>>>
>>>>> attached is the new patch for HTTP-timeouts. I have included the 
>>>>> work of Slava wrt. server-timeouts, but modified it such that no 
>>>>> new thread is needed. Another new thing is the purging of active 
>>>>> clients upon port-removal, as suggested by Alex.
>>>>>
>>>>> Please, try it out and share your comments :)
>>>>>
>>>>> Best regards,
>>>>> David
>>>>>
>>>>
>>>
>>>
>>
>>
> 
> 


-- 
Vjacheslav Chekushin                                mailto:[email protected]
Latvian Mobile Phone Company                        http://www.lmt.lv