On 2022/5/31 上午9:01, Tejun Heo wrote:
Hello,
On Mon, May 30, 2022 at 09:53:51AM +0800, Hongchen Zhang wrote:
When I test the LTP's memcg_test_3 testcase at 8 Node server,I get the
-ENOMEM error,which caused by no avaliable idr found in mem_cgroup_idr.
the reason is the use of idr in mem_cgroup_idr is too fast than the free.In
the specific case,the idr is used and freed cyclically,so when we rmdir one
cgroup dir, we can synchronize the idr free through wating for the memcg css
offlined,and then we can use it the next cycle.
This is a micro benchmark specific problem and it doesn't make sense to
change the overall behavior for this as the suggested change is neither
desirable or logical. Maybe you can just incur the delay only after idr
allocation fails and then retry?
Thanks.
Hi Tejun,
Yes, the problem would disappear when add some reasonable delay. But I
think if we can increase the MEM_CGROUP_ID_MAX to INT_MAX.Thus the
-ENOMEM error would be never occured,even if the system is out of memory.
Thanks.
