On Wed 20-07-16 09:33:37, Dave Jones wrote: > On Wed, Jul 20, 2016 at 09:23:04AM -0400, Dave Jones wrote: > > > > so this task has been already oom reaped and so oom_badness will ignore > > > it (it simply doesn't make any sense to select this task because it > > > has been already killed or exiting and oom reaped as well). Others might > > > be in a similar position or they might have passed exit_mm->tsk->mm = NULL > > > so they are ignored by the oom killer as well. > > > > I feel like I'm still missing something. Why isn't "wait for the already reaped trinity tasks to exit" > > the right thing to do here (as my diff forced it to do), instead of "pick even more victims even > > though we've already got some reaped processes that haven't exited" > > > > Not killing systemd-journald allowed the machine to keep running just fine. > > If I hadn't have patched that out, it would have been killed unnecessarily. > > nm, I figured it out. As Tetsuo pointed out, I was leaking a task struct, > so those already reaped trinity processes would never truly 'exit'. Leaked task_struct would leak some memory but they shouldn't have any effect on the task visibility to the oom killer. Tasks are basically visible until they are unhashed from the task_list. But the leak could indeed have some other side effects - like pinning a lot of memory and so the OOM kill wouldn't be sufficient to make a forward progress. -- Michal Hocko SUSE Labs -- 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>