Paul Slootman | 22 Jun 2012 13:26

memory leak in recent (3.x) kernels?

Perhaps I'm triggering something that exists since before 3.0, but
anyway:

After some time, all swap space gets gradually used up, without a clear
indication what's using it (at least, I haven't managed to find out).

System is running debian testing, and most usage is a lot of rxvt
processes mostly ssh'ed out to other systems, and google chrome.
I suspect google chrome may be the cause of the problem.
Root is btrfs, /home is NFS.

The system earlier had 4GB RAM and swap is currently 5 x 2GB LVM
partitions. With that config I needed to reboot after about a week, as
the system ended up thrashing the swap.  I've added 8GB RAM, and now the
uptime is 42 days, system still usable.

Stopping google-chrome at such a point in time usually does not help.

At every reboot I upgrade to the latest kernel :) Currently running
3.4.0-rc6, but I saw the same behaviour with all 3.x kernels I tried.

Situation now is:

# free
             total       used       free     shared    buffers     cached
Mem:      12179368   11600220     579148          0         12    9313200
-/+ buffers/cache:    2287008    9892360
Swap:      6291444    6238388      53056

Swap is now 6GB, as I did a swapoff of 2 swap partitions.
(Continue reading)

Michal Hocko | 28 Jun 2012 17:22
Picon

Re: memory leak in recent (3.x) kernels?

On Fri 22-06-12 13:26:14, Paul Slootman wrote:
> Perhaps I'm triggering something that exists since before 3.0, but
> anyway:
> 
> After some time, all swap space gets gradually used up, without a clear
> indication what's using it (at least, I haven't managed to find out).
> 
> System is running debian testing, and most usage is a lot of rxvt
> processes mostly ssh'ed out to other systems, and google chrome.
> I suspect google chrome may be the cause of the problem.
> Root is btrfs, /home is NFS.
> 
> The system earlier had 4GB RAM and swap is currently 5 x 2GB LVM
> partitions. With that config I needed to reboot after about a week, as
> the system ended up thrashing the swap.  I've added 8GB RAM, and now the
> uptime is 42 days, system still usable.
> 
> Stopping google-chrome at such a point in time usually does not help.
> 
> At every reboot I upgrade to the latest kernel :) Currently running
> 3.4.0-rc6, but I saw the same behaviour with all 3.x kernels I tried.
> 
> Situation now is:
> 
> # free
>              total       used       free     shared    buffers     cached
> Mem:      12179368   11600220     579148          0         12    9313200
> -/+ buffers/cache:    2287008    9892360
> Swap:      6291444    6238388      53056
> 
(Continue reading)

Paul Slootman | 28 Jun 2012 19:20

Re: memory leak in recent (3.x) kernels?

On Thu 28 Jun 2012, Michal Hocko wrote:
> On Fri 22-06-12 13:26:14, Paul Slootman wrote:

> > Perhaps I'm triggering something that exists since before 3.0, but
> > anyway:
> > 
> > After some time, all swap space gets gradually used up, without a clear
> > indication what's using it (at least, I haven't managed to find out).
> > 
> > System is running debian testing, and most usage is a lot of rxvt
> > processes mostly ssh'ed out to other systems, and google chrome.
> > I suspect google chrome may be the cause of the problem.
> > Root is btrfs, /home is NFS.
> > 
> > The system earlier had 4GB RAM and swap is currently 5 x 2GB LVM
> > partitions. With that config I needed to reboot after about a week, as
> > the system ended up thrashing the swap.  I've added 8GB RAM, and now the
> > uptime is 42 days, system still usable.
> > 
> > Stopping google-chrome at such a point in time usually does not help.
> > 
> > At every reboot I upgrade to the latest kernel :) Currently running
> > 3.4.0-rc6, but I saw the same behaviour with all 3.x kernels I tried.

Memory was full again yesterday, at which point I tried 3.5.0-rc4.
Unfortunately something there (or something I may have changed in the
config) prevents my google chrome from starting all of my open tabs;
about 1/3 remain blank with a loading spinner running. Opening a new tab
and entering one of those URLs gives "window not responding" error after
some time. Wierd.
(Continue reading)


Gmane