Subject: Re: [vserver] OOM Killer in Guest crashes Host
From: Oliver Welter <mail@oliwel.de>
Date: Thu, 15 Jul 2010 22:33:53 +0200
Thu, 15 Jul 2010 22:33:53 +0200
On 13.07.2010 12:11, Herbert Poetzl wrote:
> On Tue, Jul 13, 2010 at 10:25:57AM +0200, Oliver Welter wrote:
>> Hi All,
> 
>> I have a guest that runs out of memory due to limits set,
>> vserver-stats output is (www4 is the candidate):
> 
>> 67      45   2.5G   1.4G   1d01h24   3h25m34  33d01h27 mx
>> 69      28   5.6G 381.8M  21m38s88  11m13s42  19h36m52 www1
>> 70      26   2.9G 291.4M   3m53s25   1m22s45  31m01s88 www2
>> 72      19   3.7G 277.3M  10h15m29   1h51m05  33d01h26 www3
>> 74      20   4.2G 217.1M   0m42s84   0m19s54  27m37s12 www4
> 
> from the output, I presume that you are not using the
> memory cgroups yet ...
> 
>> In the logs I get
>> Out of memory: kill process apache2(24000:#74) score 896596 or a child
>> Killed process apache2(3298:#74) vsz:352584kB, anon-rss:11348kB,
>> file-rss:4028kB
> 
>> which is totaly fine - but after severel oom kills the whole HOST
>> server dies. The above line is the last one I see in the kernel log,
>> afterwards I get a bunch of binary data and then a syslog restart
>> after the boot.
> 
>> Anybody knows whats happening? I assume that the host and even other
>> guests should not be affected by a mad running guest....
> 
> known problem with this kernel, enable memory cgroups
> and use that (probably requires an update to util-vserver)
> or avoid setting a hard limit on guests

Even with soft limits I get oom killer and a host crash after a while -
it also looks like there is a bug in memory calculation. Now running
without limits, the usage reported by vserver-stat is lower than before
and nothing gets killed....

Oliver

-- 
Protect your environment -  close windows and adopt a penguin!
PGP-Key: 3B2C 8095 A7DF 8BB5 2CFF  8168 CAB7 B0DD 3985 1721



["application/pgp-signature" not shown]