Topic: Memory Leak on 5.6 Release

[bjohns]

Hi folks,

I've been having trouble with an apparent memory leak on  a 5.6 install. I've swapped out ALL hardware (stuck the drive in a different box) and it continues. So I don't think its hardware related.

I've just applied update6 to see if that fixes things.

Anyone else come across this issue? I noticed an old post in my searches from someone suffering similar symptoms.

Thanks.

[Boris]

It could be one of the applications you've installed on top of the SME. I have a lot of SME 5.6 running (long time) on the different hardware (some with 128Mb RAM) and never had a memory leak problem with server itself.

[stancol]

I had a similar problem I think I tracked it down to not enough RAM to run 5.6 and all the contribs I had in the box. Don't know why I'd want to over task a server....maybe to much fun and not enough mon$.

Forgot where it is but the correct suggested amount of RAM is posted. The suggested is a lot higher than I thought it would be.

[bjohns]

It's a stock install, no contribs or anything. There was 64Mb installed in the old box, the new box has 128Mb. The 128Mb just took longer to fill up. I thought that amount was reasonable for a gateway/mail server for a office of around 20 staff.

I'm having doubts about the Intel e100 driver, both boxes have that adaptor in common. Otherwise it seems to be running better after the update.

[hardijs]

I have made some changes (now with the "squid cache control" contrib) to the squid max object size and alowable disk space useage.

If I do
service squid restart
the memory is released though in a week or two it creeps up again.
I do have 512MB ram inside and the thing still whops around a 132MB swap memory.
the squid is the std - may be the 2.5 would do better - any ideas?

[duncan]

Um - this is pretty standard for Linux.

Have a look around on the internet to see how linux uses ram. Regardless of how much ram is in the system - it will always "fill".

Regards Duncan

[Boris]

If I pay for more RAM, I want it be filled, not just sit free inside the server. If it used for caching great, better performance of the server.

[Kelvin]

I have quite a few SME 5.6 boxes out there. The "stock" standard installs are generally the most stable (except when they are running on really new hardware where the kernel does not recognise chipsets, etc).

On one fairly important production server, the users insist on using Webmail and not POP3 or IMAP. And, in order to overcome some problems with webmail that ships with SME 5.6, I had to install the updated PHP and Webmail (following the how-to on contribs) plus increase the allowed attachment size, etc, etc. No other add ins except for Arkeia backup (which I have running at ALL SME sites). This server has problems that's symptomatic of a memory leak (ie. works fine upon a reboot for days / weeks then suddenly get kernel out of memory errors, or mysqld shutsdown, etc, but mostly out of memory errors). A reboot clears up the problem until the cycle repeats. Putting in more RAM stretches out the time between reboots. I have replaced the entire system before (even moved it to another server that's been running 24x7 with SME 5.6 without hassles before) and the same symptoms and problem reoccurs. I now generally reboot the server every week or so (or after a few days if webmail's been used heavily) and the problem does not surface. If I leave it long enough, it comes back.

And it continues ....

Kelvin

[satovey]

I have quite a few SME 5.6 boxes out there. The "stock" standard installs are generally the most stable (except when they are running on really new hardware where the kernel does not recognise chipsets, etc).

On one fairly important production server, the users insist on using Webmail and not POP3 or IMAP. And, in order to overcome some problems with webmail that ships with SME 5.6, I had to install the updated PHP and Webmail (following the how-to on contribs) plus increase the allowed attachment size, etc, etc. No other add ins except for Arkeia backup (which I have running at ALL SME sites). This server has problems that's symptomatic of a memory leak (ie. works fine upon a reboot for days / weeks then suddenly get kernel out of memory errors, or mysqld shutsdown, etc, but mostly out of memory errors). A reboot clears up the problem until the cycle repeats. Putting in more RAM stretches out the time between reboots. I have replaced the entire system before (even moved it to another server that's been running 24x7 with SME 5.6 without hassles before) and the same symptoms and problem reoccurs. I now generally reboot the server every week or so (or after a few days if webmail's been used heavily) and the problem does not surface. If I leave it long enough, it comes back.

And it continues ....

Kelvin

May I suggest you setup a cron job to reboot the server at midnight once a week until you find the problem? Or in the least restart squid since you previousely stated that it cleared up the problem for a bit.

Is this a new hard drive? It may be the drive is going bad and has caused some kind of file corruption.

Owe it hurts to reach that far.
I need a sling.  

I don't know enough to trouble shoot, but I do see how to automate the reboot

I am running with an AMD 650 processor and 256 meg of ram. I don't have this problem.

[Kelvin]

May I suggest you setup a cron job to reboot the server at midnight once a week ...

Not possible as
a) backups run at around that time
and
b) depending on the projects running at the time, the server could actually be in use even at midnight

At the moment, its being done manually when the server has a clear window for a reboot.

Restarting squid does not fix the problem (when you get kernel out of memory errors, you'll be lucky to be able to type at the console even !).

Is this a new hard drive? It may be the drive is going bad and has caused some kind of file corruption.

As mentioned in my post, the problem's been reproduced on different PCs (even one that's been reliable 24/7) - on Intel as well as AMD platforms.

Kelvin

[bjohns]

This may come as a shock to some of you guys, but they're getting a new beaut MS server in. The new box will take over mail, so the E-Smith box can be replaced by a generic router/firewall, like a Snapgear or something.

Its just such a weird untracable problem. Memory leaks usually only occur in Kernel/modules, they can't readily exist in apps because the system will prevent that from happening (I think, read that somewhere). So my only guess is that something is up with either a driver/module or the kernel itself.