在 Tue, 17 Apr 2012 11:25:26 +0800，Sha Zhengju 写道: the vmware machine has about 2G swap space,which should be quite enough.. Days ago I can produce it on a physical machine, not any more now. the /var/log/message- was still there (the attchment),so I think it was not a mistake. I tried on my laptop,with the same system, easy to reproduce.. > On 04/16/2012 11:43 AM, gaoqiang wrote: >> 在 Fri, 13 Apr 2012 22:49:54 +0800，Michal Hocko 写道: >> >>> [CC linux-mm] >>> >>> Hi, >>> >>> On Fri 13-04-12 18:00:10, gaoqiang wrote: >>>> >>>> >>>> I put a single process into a cgroup and set memory.limit_in_bytes >>>> to 100M,and memory.memsw.limit_in_bytes to 1G. >>>> >>>> howevery,the process was oom-killed before mem+swap hit 1G. I tried >>>> many times,and it was killed randomly when memory+swap >>>> >>>> exceed 100M but less than 1G. what is the matter ? >>> >>> could you be more specific about your kernel version, workload and >>> could >>> you provide us with GROUP/memory.stat snapshots taken during your test? >>> >>> One reason for oom might be that you are hitting the hard limit (you >>> cannot get over even if memsw limit says more) and you cannot swap out >>> any pages (e.g. they are mlocked or under writeback). >>> >> >> many thanks. >> >> >> The system is a vmware virtual machine,running centos6.2 with kernel >> 2.6.32-220.7.1.el6.x86_64. >> >> the attachments are memory.stat, the test program and the >> /var/log/message of the oom. >> >> the workload is nearly 0,with searal sshd and bash program running. >> >> I just did the following command when testing: >> >> ./t >> # this program will pause at the "getchar()" line and in another >> terminal,run : >> >> cgclear >> service cgconfig restart >> mkdir /cgroup/memory/test >> cd /cgroup/memory/test >> echo 100m > memory.limit_in_bytes >> echo 1G > memory.memsw.limit_in_bytes >> echo 'pid' > tasks >> >> # then continue the t command >> >> > Hi, > > I run your test under RHEL6.1 with 2.6.32-220.7.1.el6.x86_64 (an > internal version but > no changes in mm/memcg) in a real server and the process is killed with > memsw reaching > 1G. Does your vmware virtual machine have enough swap space?.. I've no > idea whether > the different behavior come from the physical/virtual environment. > > > Thanks, > Sha > > -- 使用 Opera 革命性的电子邮件客户程序: http://www.opera.com/mail/