[SOLVED] segfaults in apache logs and CPU usage

39 Posts

pixeltoaster Reply #1, 14 years, 7 months ago

Hi

I have php segfaults in my apache log.

[Mon Feb 06 23:09:19 2012] [notice] child pid 19105 exit signal Segmentation fault (11)
[Mon Feb 06 23:09:24 2012] [notice] child pid 13163 exit signal Segmentation fault (11)
[Mon Feb 06 23:09:25 2012] [notice] child pid 15990 exit signal Segmentation fault (11)
[Mon Feb 06 23:15:25 2012] [notice] child pid 12978 exit signal Segmentation fault (11)
[Mon Feb 06 23:15:26 2012] [notice] child pid 15809 exit signal Segmentation fault (11)

It seems it only happens when i'm logged into the manager.
Dont know if it is linked but modx error log is full of lines like this :

[2012-02-06 23:09:11] (ERROR @ /site/connectors/workspace/packages.php) MODX could not download the file. You must enable allow_url_fopen, cURL or fsockopen to use remote transport packaging.

allow_url_fopen is On in php.ini and, apache have the rights on the core/packages directory so i don't understand what happens.... I can install packages with the manager without problem...

I'm on 2.2.pl under debian with php 5.3.3-7+squeeze7 with suhosin/zend.

Any ideas?

Thanks.

PS: memtest is clean [ed. note: pixeltoaster last edited this post 14 years, 7 months ago.]

39 Posts

pixeltoaster Reply #2, 14 years, 7 months ago

The packages.php errors is a known issue
https://forums.modx.com/thread/73798/2-2-pl2-error-log-many-lines-like-this-why

But could it be the cause of the segmentation faults? [ed. note: pixeltoaster last edited this post 14 years, 7 months ago.]

39 Posts

pixeltoaster Reply #3, 14 years, 7 months ago

Subsidiary question, is Revo so CPU intensive?
I just notice that a simple F5 eats 20 to 50% of a core (Xeon i7 W3520) where an evo site is below 1%...
If it is the case i can't stay on Revo for the project i'm working on, but most likely it is related to the segfaults.

39 Posts

pixeltoaster Reply #4, 14 years, 7 months ago

Sorry for the flood but i'd like to understand what's wrong with my setup.

a simple (remote) apache bench with -c 10

Evo

Concurrency Level:      10
Time taken for tests:   1.369 seconds
Complete requests:      100
Failed requests:        0
Write errors:           0
Total transferred:      962900 bytes
HTML transferred:       925000 bytes
Requests per second:    73.06 [#/sec] (mean)
Time per request:       136.873 [ms] (mean)
Time per request:       13.687 [ms] (mean, across all concurrent requests)
Transfer rate:          687.01 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:       35   39   3.7     38      54
Processing:    86   96   6.0     96     113
Waiting:       45   53   5.2     52      71
Total:        124  135   7.7    135     164

Percentage of the requests served within a certain time (ms)
  50%    135
  66%    137
  75%    139
  80%    141
  90%    142
  95%    151
  98%    161
  99%    164
 100%    164 (longest request)

Revo :

Concurrency Level:      10
Time taken for tests:   19.927 seconds
Complete requests:      100
Failed requests:        0
Write errors:           0
Total transferred:      1944300 bytes
HTML transferred:       1903800 bytes
Requests per second:    5.02 [#/sec] (mean)
Time per request:       1992.711 [ms] (mean)
Time per request:       199.271 [ms] (mean, across all concurrent requests)
Transfer rate:          95.28 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:       35   37   3.5     35      53
Processing:   856 1896 907.1   1671    4556
Waiting:      786 1825 907.2   1601    4486
Total:        890 1933 907.6   1708    4591

Percentage of the requests served within a certain time (ms)
  50%   1708
  66%   2131
  75%   2490
  80%   2877
  90%   3224
  95%   3996
  98%   4337
  99%   4591
 100%   4591 (longest request)

I think a 3G smartphone is enough to bring the server down =)

The same test on a fresh install with no extra, and the default blank page is also incredibly slow...
Any clue? [ed. note: pixeltoaster last edited this post 14 years, 7 months ago.]

MODX Staff
10,725 Posts

opengeek Reply #5, 14 years, 7 months ago

Segfaults are generally not going to occur because the server is overloaded or Revo is CPU intensive. This generally occurs because of conflicts in the environment, i.e. bugs in the interaction between the many libraries required to compile PHP with Apache.

And yes, Revo is OO and requires more memory and CPU than Evo right out of the box, but it is also a lot more powerful and scalable overall.

Jason Coward
Chief Architect @ MODX
http://www.jasoncoward.com | http://twitter.com/drumshaman | https://github.com/opengeek

39 Posts

pixeltoaster Reply #6, 14 years, 7 months ago

Thank you Opengeek.

Revo is awesome and that's why i'd love to go with him.
But can you confirm that this extremely high cpu load is not the normal behavior?

I'm investigating to see what's wrong with my setup, but tbh i'm completely stuck atm.

39 Posts

pixeltoaster Reply #7, 14 years, 7 months ago

Ok I installed revo on a wheezy test server (php 5.3.9-1), and there are no more segfaults.
The server load is also way better, ~~while remaining a bit too high for a high frequentation site imo (which is fortunately not my case).~~
Remains to find what's wrong on the prod server... [ed. note: pixeltoaster last edited this post 14 years, 7 months ago.]

39 Posts

pixeltoaster Reply #8, 14 years, 7 months ago

Solved

The high cpu usage was caused by getressources.
Using it cached instead of uncached reduce the cpu usage by 90% (yes, 90%, it's not a typo) !!!
Shame on me for not watching this sooner.
Now I get excellent values in benchmarks.

Percentage of the requests served within a certain time (ms)
  50%    236
  66%    239
  75%    241
  80%    244
  90%    257
  95%    263
  98%    270
  99%    275
 100%    310 (longest request)

[ed. note: pixeltoaster last edited this post 14 years, 7 months ago.]