Memory Usage Peak with Plesk8

[sjuerges]

this is getting even weirder.

the following settings will not create the problem:

disable server wide spamassasin config, enable user personal config.

set every use to the same settings that you would hav done manually.

bingo. spamassassin is doing its job (on a global scale if you change the settings in the db) ... but not leaking memory like a sieve...

This can also be used as a workaround until swsof cough up a patch

Seb

 

[NightStorm]

So, since sw-soft screwed this up and doesn't seem to be releasing any fixes, anyone got a simple code clip to change all 400+ email accounts on the 78 domains on my box to use spamassassin in their personal settings, so I don't have to waste the next 5 hours bypassing something that I'm already paying another company to support and fix (even though they don't)?

 

[sjuerges]

Go into the psa mysql database and try this:

UPDATE sa_conf SET flt_enabled="user" WHERE flt_enabled="serv"

that oughta do it.

Seb

P.S: Dont forget to restart spamassassin afterwards *grins*

 

[NightStorm]

Wile the change suggested above does help, the load on the server is still hitting 12+ when processing mail... it's simply not staying there due to the serverwide settings not being in place. This still really isn't good enough for software that costs as much as this does...

 

[sjuerges]

true that, but at least memory aint 100% ALL the time, and server load doest stay put at 5 ...

Seb

 

[NightStorm]

Staying put at 5 would be nice. My server can be considered as stable at 5... sites still load fast, no timeouts, no delay on mail...
Instead, I am jumping between 12 and higher, and 3. The server used to run below 1 all the time when it was using 7.5.4.
I wish sw-soft would just release their precious patches that they say they have, instead of keeping us as beta testers so they know just how badly they f***ed things up with this release.

 

[RooMeat]

I was going to start a thread.. but don't need to as it seems I am not alone lol

I upgraded my server to Plesk 8 on the weekend and have had nothing but trouble since.
Qmail keeps stopping, the server loads are much higher than what they were last week and nothing else major has changed.
My server does not even host that many domains. I have about 40 domains total and out of that only 4 domains doing any real traffic and only ONE of those that does quiet a lot of traffic. Even with that said.. this is something that was the same last week and before.. and no server problems.
Problems only started since upgrade.

 

[jusu313]

Add us to the list as well. The high server load started after upgrading to Plesk 8. Our servers are freaking crawling.

 

[phatPhrog]

mysql SWAP cache memory problems

Not sure if this will help for your particular problem, but see:

http://forum.swsoft.com/showthread.php?s=&postid=143262#post143262

for a possible solution to your problem.

Hope it helps.

 

[NightStorm]

Unfortunately, it won't help in my case... spamassassin is tearing into the server everytime it has to process a message. This is keeping the server loads over 10, and slowing Apache to a crawl.
And I notice that sw-soft has released ZERO patches in the past month to address this, or any of the other issues that have been reported... so apparently user-reelased interim patching is all we have. *sigh* Nice to pay for the software so that the users have to develop and patch everything themselves.

 

[phatPhrog]

Originally posted by NightStorm
Unfortunately, it won't help in my case... spamassassin is tearing into the server everytime it has to process a message. This is keeping the server loads over 10, and slowing Apache to a crawl.
And I notice that sw-soft has released ZERO patches in the past month to address this, or any of the other issues that have been reported... so apparently user-reelased interim patching is all we have. *sigh* Nice to pay for the software so that the users have to develop and patch everything themselves.

Be sure if you add the

query_cache_type=1
query_cache_size=10m

that you restart mysql; better to stop and start

/etc/init.d/mysqld stop
/etc/init.d/mysqld start

Not knowing your upstream/server config, have you checked your

/proc/sys/vm/swappiness

cat /proc/sys/vm/swappiness

With a FC4 server (ours anyway) the default is set to 60.

So far we are in agreement with others from various sources that the problem lies with Apache and/or MySql as well as the possibility of the SWAP/swappiness setting.

PLEASE ask others with knowledge of your server config before implementing anyone's recommendations as they may result in undesirable results on your server. P)

You can change your swap files usage by

via SSH

echo 30 >/proc/sys/vm/swappiness

or

sysctl -w vm.swappiness=30 ;this sets the swap on restart if not already configured elsewhere

For other resources on this, use

FOR GENTOO
http://forums.gentoo.org/viewtopic.php?t=175419

LINUXHOWTO

http://www.linuxhowtos.org/System/Linux Memory Management.htm

 

[NightStorm]

*blinks*
And why again would the upgrade from 7.5.4 to 8.0.0 require all of this extra sudden management of how my server handles it's swap usage?
Lets look at this a different way. I remove spamassassin, the server load goes down. I add spamassassin, server load goes back up.
The problem is not with the server or it's swap settings... it is with how qmail is interacting with Spamassassin. This was not a problem with 7.5.4... it's a recent development brought on by the upgrade to Plesk 8.0.
Without spamassassin, mail is processed in under 15 seconds... faster if it's local delivery. With spamassassin, mail takes anywhere from 2 minutes to 15 minutes, even for local deliveries, and results in a server load averaging 50 times more than what it used to be (yes, 50... the server has gone from ~0.3 to 15+ when processing even a single message).
If it was an issue with swap, then don't you think my CPanel server (the hardware being a mirror of the Plesk server) would be having the same load problems, seeing as they share the same swap settings? Especially since the CPanel server houses an additional 27 domains and has twice the traffic of the Plesk server... yet the load on CPanel never climbs above 0.3, even when running nightly backups.

 

[phatPhrog]

okay "blinks"

What works for one doesn't work for another, as we all know.

And why again would the upgrade from 7.5.4 to 8.0.0 require all of this extra sudden management of how my server handles it's swap usage?

Answer that question and you will be praised as a superhuman!!

There are so many new changes, WTFKs??

We have made similar changes to three servers today. One test server and two production servers with SWAP usage going down, down, down..... One of our heaviest servers that is used primarily for WordPress sites has yet to show any, ANY SWAP file usage since implementing the change.

Guess it depends on the goat that carries us.

 

[NightStorm]

Reading through your tips one more time, you should probably be sure to point out that /proc/sys/vm/swappiness wasn't introduced until the 2.6 kernel, so those still operating under the 2.4 will not be able to utilize that one method.

 

[phatPhrog]

Do forgive me....but are you in need of help with a problem, . . . or simply an SW-Soft research assistant looking for feedback on current known problems?

You are correct in the fact the kernel should have been noted for others, but since I was answering your questions; I failed to take note and state such for everyone that may attempt to use or research this post for their purpose.

For those of you that find this post, do take research this for your particular system.

 

[boxdoc]

On my new Server with 2 Gig of RAM i see with top for every domain one spamd with 22MB of Ram and drwebd with 11MB of Ram.

At this moment i have only 17 domains on it but if you add these you get 561 MB only for spamd and drweb.

httpd2-prefork with Ram from 20-40 MB of usage is also there more than 10 times

So i have only 60 MB of free Mem on a 2 Gig Machine with only 17 domains.

 

[Traged1]

Umm, isn't it normal for Linux to use all of the RAM on a server for threads that will be handled, kinda of like the "Unused processes" on Windows? I have 4 PLesk Linux servers each with 4GB of RAM and they are all using 99% ram usage. But we have yet to see any slow down on any of these servers. Using RHEL3 with PLESK8. Also, I think I should submit that while we are using 99% memory our loads are only .23 - .45 on all the servers.

High CPU usuage is almost always Due to High Read/Write processes to and from your hard drives.

 

[jusu313]

There is a known issue with kernel versions 2.6.9-22 and 2.6.9-34 and iowait, that is what probably is causing a lot of the server load issues.

 

[boxdoc]

I use 2.6.13-15.8

 

[glennj]

I'm having this problem... was it ever fixed?