Hello, Michal. On Mon, Dec 02, 2013 at 03:12:42PM +0100, Michal Hocko wrote: > I really do not want to add more hacks just to make this use case work. > There are some proposals for more systematic implementation (memory > reserves for oom killer etc.) but that won't interfere with the cgroup > core. > This one just looks trivial so I was thinking whether we can keep the > !allocating write as before. It is nothing I would insist on, though. So > I will leave the decision on you. So, I'm just gonna commit the patches as-is because I can't really see how this is anything which can work in any reasonable way. As it currently stands, the userland wouldn't even be able to read any knob. Wouldn't it at least need to do that? Actually, the answer to that question doesn't even matter because "no" would mean that the OOM notification, however it's done, can't depend on the userland being able to read *any* knob, which in turn is likely to constrain and distort the notification mechanism itself. These things are all connected and this type of bad decisions propagates through the whole stack. In case this *really* is necessary, let's please do it in a separate patch with rationale and detailed explanation of actual usage. Thanks. -- tejun _______________________________________________ Containers mailing list Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linuxfoundation.org/mailman/listinfo/containers
