Apache2 with mpm_worker times out with many concurrent requests

[Matteo Sasso]

On a fresh Gutsy Gibbon (server edition) I installed apache2 (Apache/2.2.4), php5-cgi and libapache2-mod-fastcgi. I enabled the fastcgi and action modules in apache configuration. I also added a couple of lines to make it run PHP files but I don't think this has anything to do with the problem.

When I run the apache benchmark as follows:
ab -c 100 -n 1000 http://yaddayadda/

Sooner or later I get the following error message:
apr_poll: The timeout specified has expired (70007)

Seems one of the worker threads hangs or something. I got lucky once and managed to complete the benchmark, but it usually ends after a little more than 100 requests.

This is my first bug report for ubuntu so feel free to ask for more info.

[Chuck Short (zulcss)]

I was not able to reproduce this under hardy or gutsy with a clean configuration. If possible can you tar up your apache configuration files somewhere so I can try to reproduce this error?

Thanks
chuck

[Matteo Sasso (matteo-sasso)]

I attached the configuration. I tried again today (after upgrading) and the problem persists. I also discovered that the problem usually doesn't manifest on a freshly started apache daemon: if you run "ab" as indicated, it usually requires 2 or 3 runs for the problem to present itself.

[Chuck Short (zulcss)]

Hi I still wasnt able to reproduce this even with running it more than 2 or 3 times. Can you list the output of dpkg -l | grep apache as well?

Thanks
chuck

[Matteo Sasso (matteo-sasso)]

Here it is:

ii apache2 2.2.4-3ubuntu0.1 Next generation, scalable, extendable web se
ii apache2-mpm-worker 2.2.4-3ubuntu0.1 High speed threaded model for Apache HTTPD
ii apache2-utils 2.2.4-3ubuntu0.1 utility programs for webservers
ii apache2.2-common 2.2.4-3ubuntu0.1 Next generation, scalable, extendable web se
ii libapache2-mod-fastcgi 2.4.2-8build1 Apache 2 FastCGI module for long-running CGI

[Chuck Short (zulcss)]

Hi,

Does your site use any weird php5 scripting? Also can you let me know what yaddayadda refers to?

Thanks
chuck

[Matteo Sasso (matteo-sasso)]

Nothing fancy. But it shouldn't matter anyway since I'm testing against a static file (right now I'm using a "test.html", just to be sure). "yaddayadda" referred to the IP address of the server, which I preferred to leave undisclosed since it's still being set up and it's not ready yet, security-wise.

[Matteo Sasso (matteo-sasso)]

I experience the same issue as bug #147639. Maybe they are related?

[Chuck Short (zulcss)]

Can you try with hardy please?

Thanks
chuck

[Chuck Short (zulcss)]

Can you try with hardy-proposed enable?

Thanks
chuck

[Matteo Sasso (matteo-sasso)]

Yes, I finally managed to upgrade to hardy and the problem persists, although it seems to be a little "different" (now it aborts around the 300th request, usually).
I'm not sure I understood your last comment, what do you mean by "hardy-proposed enable"?

[Jayson Rowe (jayson.rowe)]

Since it's been a very long time since any additional info was added to this bug, I'm just checking to see if this is still an issue, and find out what additional work should be done on this bug.

[Harry Sufehmi (harry-sufehmi)]

I'm still getting this error as well with 8.04.2 - however, when I installed Squid and ran ab again; the same error showed up as well.
I checked, and all requests are served from Squid's memory cache.

This is how I invoked the error : ab -c 20 -n 10000 -v 1 http://mywebsite.com/

If I lower the requests to 1000, ab will be able to finish the benchmark. So yeah, this is load-related problem.

FYI.