On Wed, Jan 24, 2018 at 5:54 PM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > On Wed, Jan 24, 2018 at 05:08:27PM -0800, Shakeel Butt wrote: >> First, let me apologize, I think I might have led the discussion in >> wrong direction by giving one wrong information. The current upstream >> kernel, from the syscall context, does not invoke oom-killer when a >> memcg hits its limit and fails to reclaim memory, instead ENOMEM is >> returned. The memcg oom-killer is only invoked on page faults. However >> in a separate effort I do plan to converge the behavior, long >> discussion at <https://patchwork.kernel.org/patch/9988063/>. > > Correct me if I'm misinterpreting you, but your rationale in there > appears to be along the lines of "userland applications might not > be ready to handle -ENOMEM gracefully, so let's hit them with > kill -9 instead - that will be handled properly, 'cuz M4G1C!!1!!!!" > Nah, the motivation is something like: In the memory overcommitted system (or memcg) where jobs of different priorities are running, it is preferable to kill a low priority job than to return an ENOMEM to high priority job. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
