Memory Leak 8.1.0 upgrade to 8.1.1 - Downgrade possible?

[eugenevdm]

Since Plesk upgrade from 8.1.0 to 8.1.1 two days ago server has twice frozen due to memory depletion. Otherwise this server has been stable for about a year.

My question is:
Is it possible to downgrade Plesk from version 8.1.1 to 8.1.0 by re-running installation program? Or do I have to restore some kind of a operating system backup??

 

[freeshar]

I had similar problem, memory was increasing during the day and after one day almost everything was exhausted, even swap...

I run the next command on my Linux (centos 3.x):

ps -eo pid,ppid,rss,vsize,pcpu,pmem,cmd -ww --sort=pid


and found that I have tons of process like this:
/usr/bin/python2.2 /usr/local/psa/admin/sbin/agent_runner xxxxxxxxx

where xxx my local path to tmp folder for backups..

I instantly reminded that I also have problem with backup - could not backup even through plesk panel.

I have croned backup on each domain (~20)

The solutions for backup was to run chmod 777 on:

/var/lib/psa/dumps and
/var/lib/psa/dumps/tmp

Backup started to work.

Then I cleaned up
/var/lib/psa/dumps/tmp from a few thousands of files - as I understood plesk tried to continue after failure


So my theory is backup agent added every night even more unfinished backups to tmp directory (since they never were finished).

The backup agent on regular base looked at the tmp directory and tried with backups, that caused hang them in the memory...

I hope that it is over now - I mean memory leak

 

[eugenevdm]

Thank you for your detailed analysis. I have at least made some progress with this problem after two weeks. the server kept on freezing for the last two weeks. You can still ping it and some ports are open but it's completely unresponsive and you can't SSH. Then when you walk to the console errors on the screen saying "Out of memory - process killed".

So I plotted Server Available Real / Physical memory with the help of PRTG and soon started spotting a trend. Memory exhaustion happening the whole time. Sometimes dramatic decreases at once. I then used 'top' and 'F' to select 'm' for sort field and I had this at the top:

23337 psaadm 20 0 907m 615m 13m S 0.0 30.7 3:43.54 httpsd
17241 apache 15 0 299m 50m 9984 S 0.0 2.5 0:06.07 httpd
17408 apache 15 0 281m 50m 9008 S 0.0 2.5 0:32.89 httpd
17513 apache 15 0 281m 50m 9108 S 0.0 2.5 0:30.37 httpd
8189 apache 15 0 280m 49m 9012 S 0.0 2.5 0:13.73 httpd
8247 apache 15 0 280m 49m 8956 S 0.0 2.5 0:12.64 httpd
13640 apache 15 0 307m 49m 9116 S 0.0 2.5 0:12.18 httpd
24526 apache 15 0 280m 49m 8992 S 0.0 2.5 0:25.34 httpd
17143 apache 16 0 278m 48m 8956 S 0.0 2.4 0:06.13 httpd
24525 apache 15 0 300m 47m 9124 S 0.0 2.3 0:26.46 httpd
17236 apache 16 0 278m 47m 9092 S 0.0 2.3 0:04.59 httpd
8243 apache 15 0 281m 46m 9052 S 0.0 2.3 0:10.99 httpd
24524 apache 15 0 281m 39m 8976 S 0.0 1.9 0:26.96 httpd
8551 apache 15 0 278m 37m 9032 S 0.0 1.9 0:10.12 httpd
17524 apache 15 0 280m 37m 9060 S 0.0 1.8 0:32.33 httpd

It looks like this HTTPS process owned by PSAADM likes to have a lot of memory.

Furthermore it appears there are some other processes running on the box that also likes to eat a lot of memory and never return it - but I haven't identified those yet. But I suspect it's going to be between Plesk statistics generation and 4PSA Advanced statistics. Unfortunately I still have both running on my box due to inadequacies of Plesk statistics.

Simple solution?: Restart PSA with /etc/init.d/psa restart or restart Apache with /etc/init.d/httpd restart. All the memory re-appears.

Difficult solution?: Figure out why HTTPS owned by psaadm takes up so much memory and isolate those other processes. I will be working on it for the next few weeks and let you know.

 

[eugenevdm]

Something else of note is Plesk official release on the web site is March 23 2007 - but now I see my Plesk reports there is another update April 4 2007 - I've searched for changelog on SWSoft's web site - can't find any information about this release. But I'm going to upgrade regardless because I'm having too many problems.

 

[breun]

httpsd is the Plesk web server for Plesk's web interface (serving on port 8443 by default).

 

[freeshar]

after two days I see that memory stopped to leak!

by the way you can stop that https plesk webserver, if nobody uses it and start it when you need .

Plesk has two Apaches - one for webserver and one for Plesk panel. The second one is safe to stop

 

[eugenevdm]

Even though we implemented a 6 hour PSA restart that still didn't get rid of the memory leak.

In the end removing Tomcat and all traces of Jakarta stabilized the server.