Re: [PATCH] Add the memcg print oom info for system oom

[Michal Hocko]

On Mon 21-05-18 14:11:21, David Rientjes wrote:
> On Thu, 17 May 2018, Michal Hocko wrote:
> 
> > this is not 5 lines at all. We dump memcg stats for the whole oom memcg
> > subtree. For your patch it would be the whole subtree of the memcg of
> > the oom victim. With cgroup v1 this can be quite deep as tasks can
> > belong to inter-nodes as well. Would be
> > 
> > 		pr_info("Task in ");
> > 		pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
> > 		pr_cont(" killed as a result of limit of ");
> > 
> > part of that output sufficient for your usecase?
> 
> There's no memcg to print as the limit in the above, but it does seem like 
> the single line output is all that is needed in this case.

Yeah, that is exactly what I was proposing. I just copy&pasted the whole
part to make it clear which part of mem_cgroup_print_oom_info I meant.
Referring to "killed as a reslt of limit of" was misleading. Sorry about
that.

> It might be useful to discuss a single line output that specifies relevant 
> information about the context of the oom kill, the killed thread, and the 
> memcg of that thread, in a way that will be backwards compatible.  The 
> messages in the oom killer have been restructured over time, I don't 
> believe there is a backwards compatible way to search for an oom event in 
> the kernel log.

Agreed
 
> I've had success with defining a single line output the includes the 
> CONSTRAINT_* of the oom kill, the origin and kill memcgs, the thread name, 
> pid, and uid.  On system oom kills, origin and kill memcgs are left empty.
> 
> oom-kill constraint=CONSTRAINT_* origin_memcg=<memcg> kill_memcg=<memcg> task=<comm> pid=<pid> uid=<uid>
> 
> Perhaps we should introduce a single line output that will be backwards 
> compatible that includes this information?

I do not have a strong preference here. We already print cpuset on its
own line and we can do the same for the memcg.

-- 
Michal Hocko
SUSE Labs