So what can mod_fcgid do in this overloaded?
1. mod_fcgid get a request
2. mod_fcgid can't apply a free slot of FCGI handler
3. mod_fcgid send a spawn request to PM
4. PM deny the request(for too much process already)
5. Now....
for( i=1; i<64; i++)
{
a) mod_fcgid delay a while, and then send another spawn request to PM and
try apply free slot again.
b) mod_fcgid send another spawn request at once, even the last request is
denied.
c) ??
(now is a, b maybe not a good idea, any new idea?)
}
I think the bottleneck is too much request, too less FCGI handler. httpd(or
mod_fcgid) either drop client connections or delay a while, there is no other
way out?
Another question. Is it necessary to call procmgr_init_spawn_cmd() from inside
the for loop ?
I took a brief look, it seems not necessary. I will move it out of loop and
test.
2012-08-28
pqf
发件人：Lazy
发送时间：2012-08-27 21:47
主题：Re: Re: Re: mod_fcgid concurrency bottleneck, issue#53693
收件人："dev"<dev@httpd.apache.org>
抄送：
2012/8/16 pqf <p...@mailtech.cn>:
> How about this:
> 1. procmgr_post_spawn_cmd() now return a status code from PM, so process
> handler now know the spawn request is denyed or not.
> 2. if a new process is created, no sleep is needed.
> 3. if no process is created, sleep a while
sorry for the late reply,
in the old code there ware no sleep() between procmgr_post_spawn_cmd()
and apply_free_procnode()
sleep() was invoked only if there ware no free procnode.
This happened only if we ware denied spawning new process or in some
cases if some other thread managed to use that procnode before us.
Your change adresses cases if some other thread stole "our" newly
spawned fcgi process, old code was waiting 1s before trying to spawn
another/recheck, new code doesn't, I guess this is the orginal issue
in stress tests when total number of simultaneous connections doesn't
exceed max fcgi processes. But when spawning is denied recovery time
is still long 1s.
I was refering to cases when spawn is denied.
If a vhost is overloaded or someone added sleep(60) in the code,
mod_fcgid blocks on all request to that vhost
for over a minute and it is possible to occupy 1000 threads using
under 20 new connections to slow vhost
per second. This can be mitingated by adding avaiability which will
impact time spend on waiting for free process. Overloaded vhost will
start to drop connections faster preventing the web-server reaching
MaxClients
limit.
Another question. Is it necessary to call procmgr_init_spawn_cmd()
from inside the for loop ?
>
> 2012-08-16
> ________________________________
> pqf
> ________________________________
> 发件人：Lazy
> 发送时间：2012-08-16 16:47
> 主题：Re: Re: mod_fcgid concurrency bottleneck, issue#53693
> 收件人："dev"<dev@httpd.apache.org>
> 抄送：
>
> 2012/8/16 pqf <p...@mailtech.cn>:
>> Hi, Michal
>> My solution do "add availability to each class", which is the
>> procmgr_post_spawn_cmd() call in each loop do.
>> The sleep() call is intrudused for a stress test without warm up time, in
>> this case, mod_fcgid will create more processes than a slow start one(each
>> process handler can't apply a free slot on the very begining, so send a
>> request to process manager to create one, it's easy to reach the max # of
>> process limit while httpd startup, but the idle process will be killed
>> later), the sleep() call is a little like a "server side warm up delay".
>> But since someone said remove this sleep(), the server work fine without
>> bottleneck(Maybe he didn't notise the warm up issue?), so I thought remove
>> the sleep() is a good idea. But reduce the time of sleep() is fine to me
>> too.
>
> I was referring to the case where all processes are busy, without
> sleep(), handle_request() wil quickly send spawn requsts, whith will
> be denyed by process menager, with sleep() handle_request() will
> always wait quite a long time,
> occupying slots
>
> --
> Michal Grzedzicki
>
>
