High CPU load on Django/Apache/mod_wsgi site
Load testing a django 1.21/Apache/mod_wsgi configuration on an AWS small instance (Ubuntu 10.04) with Apache bench is showing extremely high CPU load (using uptime and vmstat) at low concurrent requests:
ab -c 5 -n 1000 "my_url"
...causes this uptime output:
18:04:54 up 9 days, 16:54, 3 users, load average: 5.33, 2.45, 1.91
CPU is at 100% even with an Apache bench concurrency value of 2. I'm running Apache bench from a different AWS instance in the same region/zone. Ideas on what's the problem, or how I should continue to debug this?
Details:
- Out of desperation, I installed a vanilla django project/app with a simple "Hello World" view (no DB calls, etc). Same results. So I doubt it's my application code.
- Memory usage looks fine during the load test.
Here's a vmstat output before/during/after load test:
procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
r b swpd free buff cache si so bi bo in cs us sy id wa
0 0 0 1034484 94848 321320 0 0 0 0 13 29 0 0 100 0
6 0 0 1032916 94848 321328 0 0 0 0 262 720 4 32 12 0
6 0 0 1031684 94848 321336 0 0 0 0 312 796 7 33 0 0
8 0 0 1030892 94856 321344 0 0 0 12 302 763 4 36 0 0
...
6 0 0 1030268 94864 321376 0 0 0 0 302 843 3 39 0 0
0 0 0 1032452 94868 321380 0 0 0 12 183 516 3 22 34 0
1 0 0 1033988 94868 321388 0 0 0 0 24 38 1 2 92 0
0 0 0 1033996 94868 321388 0 0 0 0 17 28 0 0 100 0
- I'm running a prefork version of apache2 since I'm also running WordPress, which relies on PHP. (PHP doesn't play well with Apache worker version)
Here's my virtual hosts file:
WSGIPythonHome /home/xxx/webapps/ve/api
<VirtualHost *:80>
ServerAdmin webmaster@localhost
ServerName app.xxx.mobi
WSGIDaemonProcess snaplive user=www-data group=www-data processes=10 threads=1 maximum-requests=10000
WSGIProcessGroup snaplive
WSGIScriptAlias / /home/xxx/webapps/api/settings/apache/prod.wsgi
DocumentRoot /home/xxx/webapps/api/static
ErrorLog /var/log/apache2/django-live/error.log
CustomLog /var/log/apache2/django-live/access.log combined
</VirtualHost>
Here's my httpd.conf file:
Alias /media /home/xxx/Django-1.2.1/django/contrib/admin/media
LoadModule headers_module /usr/lib/apache2/modules/mod_headers.so
StartServers 2
MinSpareServers 2
MaxSpareServers 5
MaxClients 50
MaxRequestsPerChild 3000
ServerLimit 8
Keepalive off
HostnameLookups Off
Here's my wsgi file:
import os
import sys
sys.stdout = sys.stderr
from django.core.handlers.wsgi import WSGIHandler
os.environ['DJANGO_SETTINGS_MODULE'] = 'settings'
application = WSGIHandler()
sys.path.append("/home/xxx/webapps/api")
By hitting a django url from a browser during the load test, I've confirmed qualitatively that the high CPU load is impacting performance.
I've read that this might not be important, but I'm seeing this a lot in my error logs:
[Sun Sep 19 18:04:58 2010] [error] Exception KeyError: KeyError(-1218693376,) in <module 'threading' from '/usr/lib/python2.6/threading.pyc'> ignored
Here are my Apache bench results, in case helpful:
Server Software: Apache/2.2.14
Server Hostname: app.xxx.mobi
Server Port: 80
Document Path: /plist_catalog/test_data
Document Length: 0 bytes
Concurrency Level: 5
Time taken for tests: 27.720 seconds
Complete requests: 1000
Failed requests: 0
Write errors: 0
Non-2xx responses: 1000
Total transferred: 269000 bytes
HTML transferred: 0 bytes
Requests per second: 36.08 [#/sec] (mean)
Time per request: 138.598 [ms] (mean)
Time per request: 27.720 [ms] (mean, across all concurrent requests)
Transfer rate: 9.48 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 1 2 8.5 1 88
Processing: 9 136 176.9 81 1182
Waiting: 9 135 176.6 81 1182
Total: 10 138 176.7 83 1183
Percentage of the requests served within a certain time (ms)
50% 83
66% 98
75% 128
80% 140
90% 423
95% 576
98% 727
99% 819
100% 1183 (longest request)
Solution 1:
The issue was that I had installed the package apache2-mpm-itk instead of apache2-mpm-prefork. apache2-mpm-itk is derived from apache2-mpm-prefork, but for some reason, didn't perform well when used with mod_wsgi.