Re: [PATCH] oom: choose a more suitable process to kill while all processes are not killable

[彭志刚 <zgpeng.linux@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3153 bytes --]

certainly.

Steps to reproduce:
(1)Create a mm cgroup and set memory.limit_in_bytes
(2)Move the bash process to the newly created cgroup, and set the
oom_score_adj of the  bash process to -998.
(3)In bash, start multiple processes, each process consumes different
memory until cgroup oom is triggered.

The triggered phenomenon is shown below. We can see that when cgroup oom
happened, process 23777 was killed, but in fact, 23772 consumes more memory;

[  591.000970] Tasks state (memory values in pages):
[  591.000970] [  pid  ]   uid  tgid total_vm      rss pgtables_bytes
swapents oom_score_adj name
[  591.000973] [  23344]     0 23344     2863      923    61440        0
       -998 bash
[  591.000975] [  23714]     0 23714    27522    25935   258048        0
       -998 test
[  591.000976] [  23772]     0 23772   104622   103032   876544        0
       -998 test
[  591.000978] [  23777]     0 23777    78922    77335   667648        0
       -998 test
[  591.000980]
oom-kill:constraint=CONSTRAINT_MEMCG,nodemask=(null),cpuset=/,mems_allowed=0-1,oom_memcg=/test,task_memcg=/test,task=test,pid=23777,uid=0
[  591.000986] Memory cgroup out of memory: Killed process 23777 (test)
total-vm:315688kB, anon-rss:308420kB, file-rss:920kB, shmem-rss:0kB, UID:0
pgtables:667648kB oom_score_adj:-998

The verification process is the same. After applying this repair patch, we
can find that when the oom cgroup occurs, the process that consumes the
most memory is killed first.  The effect is shown below：

[195118.961767] Tasks state (memory values in pages):
[195118.961768] [  pid  ]   uid  tgid total_vm      rss pgtables_bytes
swapents oom_score_adj name
[195118.961770] [  22283]     0 22283     2862      911    69632        0
       -998 bash
[195118.961771] [  79244]     0 79244    27522    25922   262144        0
       -998 test
[195118.961773] [  79247]     0 79247    53222    51596   462848        0
       -998 test
[195118.961776] [  79263]     0 79263    58362    56744   507904        0
       -998 test
[195118.961777] [  79267]     0 79267    45769    44005   409600        0
       -998 test
[195118.961779]
oom-kill:constraint=CONSTRAINT_MEMCG,nodemask=(null),cpuset=/,mems_allowed=0-1,oom_memcg=/test,task_memcg=/test,task=test,pid=79263,uid=0
[195118.961786] Memory cgroup out of memory: Killed process 79263 (test)
total-vm:233448kB, anon-rss:226048kB, file-rss:928kB, shmem-rss:0kB, UID:0
pgtables:507904kB oom_score_adj:-998

Michal Hocko <mhocko@kernel.org> 于2019年12月20日周五 下午3:13写道：

> On Fri 20-12-19 14:26:12, zgpeng.linux@gmail.com wrote:
> > From: zgpeng <zgpeng@tencent.com>
> >
> > It has been found in multiple business scenarios that when a oom occurs
> > in a cgroup, the process that consumes the most memory in the cgroup is
> > not killed first. Analysis of the reasons found that each process in the
> > cgroup oom_score_adj is set to -998, oom_badness in the calculation of
> > points, if points is negative, uniformly set it to 1.
>
> Can you provide an example of the oom report?
> --
> Michal Hocko
> SUSE Labs
>

[-- Attachment #2: Type: text/html, Size: 3956 bytes --]