High Apache CPU Usage

[CraigLath218]

--------------------------------------------------------------
PRODUCT, VERSION, OPERATING SYSTEM, ARCHITECTURE

Plesk Panel 10.1.1, CentOS 5 (we tried 64 bit and 32), Dual Quad Core Xeon 2.66 - 6GB Ram - RAID 10 15K SCSI Drives

PROBLEM DESCRIPTION AND STEPS TO REPRODUCE

Apache CPU usage is constantly very high. Sits in the 80 to 90% spikes up to 140%. I can't find any real documentation that says exactly what the apache CPU usage referes to. Is that a fraction of CPU time total. Does it actually matter that it is above 80 to 90%? Actual CPU usage is showing in the 5% range (in the control panel). We tried switching to FastCGI with no real effect. We also removed Watson with no effect (then we put it back).

A few httpd processes seem to get stuck and stay around for minutes. I watched one that hung out for maybe 5 minutes. This was right after a reboot so I don't think any users were downloading big files.

Has anyone else had any issues with this or found a solution?

ACTUAL RESULT

Really high Apache CPU Usage

EXPECTED RESULT

Not really high Apache CPU Usage

ANY ADDITIONAL INFORMATION

 

[fearz]

Suggestion

Load is high,

Try:

netstat -napt

it will tell you who is connected on port 80 or 443

Find out the source & destination IP(s), check /var/log/httpd/access_logs & error_log & /var/www/vhosts/<domain>/statistics/logs/access_log & error_log

I'd also suggest download apachetop, it will definitely help.

 

[atomicturtle]

Its a pretty broad area of investigation there

I'd also throw mod_top into the ring as another tool for collecting more information.

 

[CraigLath218]

Can anyone tell me what the Apache CPU % mesure actually means. Obviously it's not a mesure of actual CPU usage, is it simply a measure of the percentage of CPU time being used by apache? For example my CPU is running at 10% and apache is using 5%. Does this mean that the Apache CPU% will show as 50%?

I can't seem to find any mention of this in the Plesk documentation, but perhaps I have the wrong user guide.

Thank you for the posts with the debugging suggestions, I'm still trying to make sense of what exactly the results mean other than the fact that we have a busy server.

 

[CraigLath218]

So the current theory is that Plesk 10 health monitoring stats cannot account for multi core processors. In our case with 8 total cores a 50% apache load would actually be 16% processor load. That seems to be consistant with what we are seeing on our server.

Can anyone confirm this?

 

[MartinV_rpd]

Same situation.. apache CPU usage is rediculously high considering there's almost no traffic.
The server runs multicore, too. This has to be a bug with plesk - whereever I search in the shell (processes etc.) the values of CPU time/usage don't even get close to the values mentioned inside plesk. Hope this has been fixed in 10.3, gonna run the update now...

 

[theywill]

You can adjust the health monitoring XML file, increase the thresholds, to reduce the number of warnings.