Re: cgroup-aware OOM killer, how to move forward

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue 17-07-18 13:41:33, David Rientjes wrote:
[...]
> Thus, the semantic would be: if oom mem cgroup is "tree", kill all 
> processes in subtree; otherwise, it can be "cgroup" or "process" to 
> determine what is oom killed depending on the victim selection.

Why should be an intermediate node any different from the leaf. If you
want to tear down the whole subtree, just make it oom_cgroup = true and
be done with that. Why do we even need to call it tree?
 
> Having the "tree" behavior could definitely be implemented as a separate 
> tunable; but then then value of /A/memory.group_oom and 
> /A/B/memory.group_oom are irrelevant and, to me, seems like it would be 
> more confusing.

I am sorry, I do not follow. How are the following two different?
A (tree)	A (group)
|		|
B (tree)	B (group)
|		|
C (process)	C (group=false)

-- 
Michal Hocko
SUSE Labs




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux