You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@httpd.apache.org by Robin Becker <ro...@reportlab.com> on 2014/06/25 14:52:38 UTC
[users@httpd] MaxClients exceeded error message
An apache 2.2 mpm prefork site which uses an external mod_fastcgi socket to
django was experiencing slowdowns during a busy period.
The main symptom seemed to be a connection delay of up to 10 seconds.
At the time I saw the machine was not heavily loaded in cpu or memory, but there
were a large number of apache processes; stupidly I didn't think to count them.
Later I looked for an error message related to MaxClients exceeded or similar,
but could not see any.
I have upped the MaxClients setting from 150 to 300 and in another busy period
don't see the slowdown although monitoring the apache process count shows it can
exceed the old limit.
The django side of this is not especially fast and the single socket connection
seems like it might be a choke point, but my understanding is that the django
side of the fastcgi is also spawning children to handle the requests.
What are the effects of MaxClient limit being exceeded? Should I have seen an
error message if it was exceeded?
--
Robin Becker
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org
Re: [users@httpd] MaxClients exceeded error message
Posted by Pete Houston <ph...@openstrike.co.uk>.
On Fri, Jun 27, 2014 at 03:39:42PM +0000, Rose, John B wrote:
> What is the impact of not having a "ListenBacklogs" Directive in your
> config?
This is covered pretty well in the documentation:
http://httpd.apache.org/docs/2.4/mod/mpm_common.html#listenbacklog
A cursory read of this tells us that the default is 511.
HTH,
Pete
--
Openstrike - improving business through open source
http://www.openstrike.co.uk/ or call 01722 770036 / 07092 020107
Re: [users@httpd] MaxClients exceeded error message
Posted by "Rose, John B" <jb...@utk.edu>.
What is the impact of not having a "ListenBacklogs" Directive in your
config?
On 6/26/14 8:01 AM, "Robin Becker" <ro...@reportlab.com> wrote:
>On 25/06/2014 15:38, Jeff Trawick wrote:
>...........
>>> requests.
>>>
>>> What are the effects of MaxClient limit being exceeded?
>>
>>
>> delays in processing new connections
>>
>> depending on load and MaxClients and ListenBacklogs, some new
>>connections
>> might even get rejected
>>
>thanks
>
>
>>
>>> Should I have seen an error message if it was exceeded?
>>
>>
>> once per generation
>>
>> --/--
>>
>> It can be useful to log the response time in microseconds (%D in log
>> format) in order to distinguish delays in the application (delay is
>> reflected in value for %D) from delays before httpd starts processing it
>> due to capacity tuning (delay not reflected in value for %D since it
>> happened before processing started)
>.......
>all very useful
>--
>Robin Becker
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
>For additional commands, e-mail: users-help@httpd.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org
Re: [users@httpd] MaxClients exceeded error message
Posted by Robin Becker <ro...@reportlab.com>.
On 25/06/2014 15:38, Jeff Trawick wrote:
...........
>> requests.
>>
>> What are the effects of MaxClient limit being exceeded?
>
>
> delays in processing new connections
>
> depending on load and MaxClients and ListenBacklogs, some new connections
> might even get rejected
>
thanks
>
>> Should I have seen an error message if it was exceeded?
>
>
> once per generation
>
> --/--
>
> It can be useful to log the response time in microseconds (%D in log
> format) in order to distinguish delays in the application (delay is
> reflected in value for %D) from delays before httpd starts processing it
> due to capacity tuning (delay not reflected in value for %D since it
> happened before processing started)
.......
all very useful
--
Robin Becker
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org
Re: [users@httpd] MaxClients exceeded error message
Posted by Jeff Trawick <tr...@gmail.com>.
On Wed, Jun 25, 2014 at 8:52 AM, Robin Becker <ro...@reportlab.com> wrote:
> An apache 2.2 mpm prefork site which uses an external mod_fastcgi socket
> to django was experiencing slowdowns during a busy period.
>
> The main symptom seemed to be a connection delay of up to 10 seconds.
>
> At the time I saw the machine was not heavily loaded in cpu or memory, but
> there were a large number of apache processes; stupidly I didn't think to
> count them.
>
> Later I looked for an error message related to MaxClients exceeded or
> similar, but could not see any.
>
It only appears once per generation, so it might have been days before you
noticed the symptom. (Generation: each fresh start or restart of the
server is a generation. This can be viewed from mod_status.)
> I have upped the MaxClients setting from 150 to 300 and in another busy
> period don't see the slowdown although monitoring the apache process count
> shows it can exceed the old limit.
>
> The django side of this is not especially fast and the single socket
> connection seems like it might be a choke point, but my understanding is
> that the django side of the fastcgi is also spawning children to handle the
> requests.
>
> What are the effects of MaxClient limit being exceeded?
delays in processing new connections
depending on load and MaxClients and ListenBacklogs, some new connections
might even get rejected
> Should I have seen an error message if it was exceeded?
once per generation
--/--
It can be useful to log the response time in microseconds (%D in log
format) in order to distinguish delays in the application (delay is
reflected in value for %D) from delays before httpd starts processing it
due to capacity tuning (delay not reflected in value for %D since it
happened before processing started)
> --
> Robin Becker
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
> For additional commands, e-mail: users-help@httpd.apache.org
>
>
--
Born in Roswell... married an alien...
http://emptyhammock.com/
http://edjective.org/