Is there a memory leak in 2.6.8?

jkcunningham · Posted: Fri Oct 01, 2004 2:31 pm Post subject: Is there a memory leak in 2.6.8?

I recently upgraded from 2.6.7 to 2.6.8-r3 and have been noticing what appear to be memory leaks. I haven't been able to pin it down exactly. Here are the symptoms:

1. I'll leave the machine at night with applications open (browsers, editors, etc) and in the morning the window manager will be locked up (enlightnment). The memory meter in gkrellm is nearly pinned. I can switch to a different console and kill stuff to get it back, but after that free -m shows very little free memory left. Sometimes I've had to reboot to be able to do anything useful.

2. If I leave it with nothing running, its fine.

3. My initial suspicion was that maybe mozilla had the leak, but last night I shut down all browsers and left only sound running and it still froze up. It has also frozen up without sound running.

I'm trying to figure out how to troubleshoot this. If anyone has any suggestions, I'd appreciate hearing them.

--Jeff

FallenAngel · n00b Joined: 29 Sep 2004 Posts: 51

Have you tried using "ps aux" to locate the leaking application ?

iamben · Apprentice Joined: 10 May 2004 Posts: 278

my htpc box has done the same thing (i think), after letting it idle all night i come back and moving the mouse only moves the cursor on the screen like once every 5 seconds. free mem is ~0/512M, free swap is ~0/1G. interestingly, HD throughput is jumping all over the place, between 50 and 150 MB/s... and this is on dma33 disk!

next time it happens, im gonna ssh in and see whats up... but the last 2 times it happened, i had no keyboard attached and no sshd running.

the machine is running whatever the latest x86 gentoo-dev-sources is. never noticed any problems like this on my main box, just the htpc. oh yeah, and its running xorg 6.7, gnome 2.6. the only apps left running overnight were a couple of nautilus windows.

ozonator · Posted: Fri Oct 01, 2004 4:00 pm Post subject:

Just noticed a similar thing here, with a kernel from development-sources-2.6.9-rc2. I didn't have to leave anything overnight, though: I had simply been using the machine when the amount of memory shown in use went way up. What was running: some daemons (udev, metalog, cron, smartd, dovecot, spamd, sshd), xorg-x11, a few aterms, gkrellm, mutt, mozilla, and liferea. Uptime was only a few hours, and the only other things I had run since booting were xmms and xcdroast (which, in burning a CD, ran at least cdrecord). All this is on a machine with 512 MB of RAM -- more than enough for what I was running.

In any case, fluxbox locked to the point that all I could do was move the mouse pointer. Even the gkrellm display locked (with the swap krell showing about 100%). I could hear my disks spinning -- swapping, I presume -- but couldn't even switch to another virtual console. After waiting a few minutes, I used another machine to ssh in and kill some processes; ultimately killing fluxbox (which also let X quit) was the only thing that let me use the console there again. Thinking it might have been something with fluxbox, I tried re-emerging it with basic CFLAGS, but that emerge failed after I heard a lot of disk activity, which must have been swapping. Then, I discovered that the amount of free memory was just about nil, even though none of the remaining processes running used much memory individually (checked with top and ps). Rebooting was the only fix I could think of.

In any case, two things came to mind. First, this item on kerneltrap, which indicates that something might be up with the swappiness code in 2.6.9 (I'd set my swappiness to 50), and that's certainly related to memory management. Second, it's time to try 2.6.9-rc3, which is compiling as I type this; I didn't see anything obviously related in the changelog, but maybe it doesn't suffer from the same problem. Ultimately, I'm inclined to think this is a kernel issue, since I've run all the same software extensively on 2.6.7, without any memory-related problems.

metalac · Posted: Fri Oct 01, 2004 4:10 pm Post subject:

hmmm I can't say that I have the same problem. I use vanilla-sources 2.6.8.1 and I leave the machine running 24/7 with various applications running (evolution, firefox, xchat,...). I also have gdesklets running and consistantly have about 150+mb of memory free out of 512mb total. Besides the applications this particular box also runs MySQL, apache, sshd, smb, and probably a few more that I can't think of.
_________________
>>>>> Ignorance is bliss <<<<<

JackBak · Posted: Fri Oct 01, 2004 4:40 pm Post subject:

Yes, I have seen something similar. I use 2.6.8-r3 with the xfce4 wm with its panel and gkrellm. If I leave a single rxvt-uni terminal open overnight my memory is almost pegged (old machine 128 Meg) but not locked up. If I open firefox and close it again I'm back to about 55-60 M used out of 122M. And it seems to stay that way all day. Only thing going on at night is fcron log clean up jobs.

So something is going on, I thought it was with the xfce4 panel but your not using that so I am at a loss.

jkcunningham · Posted: Sat Oct 02, 2004 2:58 am Post subject:

jkcunningham · Posted: Sat Oct 02, 2004 3:11 am Post subject:

FallenAngel · n00b Joined: 29 Sep 2004 Posts: 51

Have you any idea, why your system needs /usr/bin/jackstart ? Because this audio daemon (?) is started 5 times and is taking up 3% mem each, which is equivalent to 45MB each . If there is a bug in a prog starting jackstart over and over, that could be your memory leak...

iamben · Apprentice Joined: 10 May 2004 Posts: 278

jkcunningham · Posted: Sat Oct 02, 2004 3:53 pm Post subject:

FallenAngel · n00b Joined: 29 Sep 2004 Posts: 51

Well, I have the gentoo-dev-sources 2.6.8. and my system works fine, its running for 3 days now and 'free -m' shows

jkcunningham · Posted: Sat Oct 02, 2004 10:07 pm Post subject:

Here is the output of 'df -h':

galay2 · Apprentice Joined: 02 Feb 2004 Posts: 208

Im also reporting a memory leak in 2.6.8 or xorg itself. I havent figured out which yet (I'll shutdown x as suggested here) but I'm almost positive its xorg (6.8.0) since I can literally see it growing in memory before it freezes (all swap/mem used up overnight).

I'll post any information if it can help pinpoint this leak. Thanks

ibliss · n00b Joined: 07 May 2003 Posts: 14

having the same problem here... after running for 24-36 hours, if i keave a mozilla (or another browser) open with e.g. 5 tabs in 1 window and 4 windows with 1 tab -> 480+MB mem usage -> trashing^2 -> unusable pc. Sometimes this also seems to end (lucky for me) in the program being (auto)killed...

maybe a leak in some (graphical?) lib or so?
2.6.8.1 here, nvidia 6116, xfree 4.3.0-r7

ozonator · Posted: Thu Oct 07, 2004 9:32 pm Post subject:

Don't know if this helps anyone, but after experiencing the apparent memory leak initially, since moving to a kernel from development-sources-2.6.9_rc3, I haven't been able to reproduce it. So far, so good, including after several days uptime and some fairly heavy and varied use, including with xorg-x11-6.7.0-r2 and mozilla (having lots of tabs spread over a few windows, each left open a couple of days).

Now that I've tempted fate by reporting success, I'll see if the problem recurs.