Re: [Bug 201699] New: kmemleak in memcg_create_kmem_cache

[段熊春 <duanxiongchun@bytedance.com>]

[-- Attachment #1: Type: text/plain, Size: 1531 bytes --]

We worry about that because in our system ,we use systemd manager our service . One day we find some machine suddenly eat lots of memory.
we find in some case，our server will start fail just recording a log then exit。 but the systemd will relaunch this server every 2 second. That  server is limit memory access by memcg.

After long time dig, we find their lots of offline but not release memcg object in memory eating lots of memory.
Why this memcg not release? Because the inode pagecache use  some page which is charged to those memcg,

And we find some time the inode(log file inode ) is also charged to one  memcg.  The only way to release that memcg is to free the inode object(example, to remove the log file.)

No matter which allocator  using (slab or slub), the problem is aways there. 

After  I view the code in slab ,slub and memcg. I think in above general scenario there maybe a problem.

Thanks for replying
bytedance.net
段熊春
duanxiongchun@bytedance.com




> On Nov 22, 2018, at 12:27 AM, Michal Hocko <mhocko@kernel.org> wrote:
> 
> On Wed 21-11-18 17:36:51, 段熊春 wrote:
>> hi all：
>> 
>> In same case， I think it’s may be a problem。
>> 
>> if I create a virtual netdev device under mem cgroup(like ip link add ve_A type veth peer name ve_B).after that ,I destroy this mem cgroup。
> 
> Which object is charged to that memcg? If there is no relation to any
> task context then accounting to a memcg is problematic.
> 
> -- 
> Michal Hocko
> SUSE Labs


[-- Attachment #2: Type: text/html, Size: 3108 bytes --]