apache prefork/mod_wsgi spawned process count seemingly past configuration [closed]

Question

in a production environment running nginx reversing back to apache mpm-prefork/mod_wsgi, im seeing 90 apache child processes, when i would expect that 40 would be the maximum, as configured below. the configuration/setup is nothing exciting:

1. nginx is reverse proxying to apache via proxy_pass, and serving static media
2. apache only serves dynamic requests

relevant nginx config:

worker_processes 15;

events {
    worker_connections  1024;
}

keepalive_timeout 10;

relevant apache config:

KeepAlive Off
MaxKeepAliveRequests 100 
KeepAliveTimeout 15

<IfModule mpm_prefork_module>
    StartServers        20
    MinSpareServers     7      
    MaxSpareServers     10
    MaxClients          200
    MaxRequestsPerChild 0
</IfModule>

mod_wsgi config, where webapp is the name of the process:

WSGIDaemonProcess webapp user=www group=users threads=1 processes=40

am i missing something?

Answer 1

The mod_wsgi daemon processes will appear to be Apache server child processes even though they aren't the same. This is because the mod_wsgi daemon processes are a fork of Apache parent process and not a fork/exec. In other words, they executable name doesn't change.

To be able to distinguish mod_wsgi daemon processes from normal Apache server child processes, supply the 'display-name' option to WSGIDaemonProcess. This option allows you to rename the process as viewable in output from 'ps' program and some variants of programs like 'top'. See documentation of WSGIDaemonProcess directive on mod_wsgi site.

http://code.google.com/p/modwsgi/wiki/ConfigurationDirectives#WSGIDaemonProcess

Answer 2

It's possible to have more apache processes than WSGI instances.

Change apache's MaxClients to 40 if you want to limit the apache processes.

Answer 3

You are using mod_wsgi in daemon mode, so mod_wsgi processes and Apache handler process are independent.

By your configuration right after the apache starts you have:

• 40(processes=) mod_wsgi processes are started the same time.
• 20(StartServers) Apache handler processes that can be automatically reduced to 10(MaxSpareServers) if there is not incoming activity.

Then on load, Apache handler processes can grow up to 200(MaxClients). But mod_wsgi processes count will be the same - 40.

My advice is to use worker mpm than Apache processes only dynamic content. It can help to reduce memory consumption and better scalability.