On Fri, Nov 19, 2021 at 9:07 PM Matthew Wilcox [off-list ref] wrote:

On Fri, Nov 19, 2021 at 08:50:08PM -0800, Mina Almasry wrote:

quoted

On remote ooms (OOMs due to remote charging), the oom-killer will attempt
to find a task to kill in the memcg under oom. The oom-killer may be
unable to find a process to kill if there are no killable processes in
the remote memcg. In this case, the oom-killer (out_of_memory()) will return
false, and depending on the gfp, that will generally get bubbled up to
mem_cgroup_charge_mapping() as an ENOMEM.

Why doesn't it try to run the shrinkers to get back some page cache /
slab cache memory from this memcg?  I understand it might not be able
to (eg if the memory is mlocked), but surely that's rare.

Please correct me if I'm wrong, but AFAICT I haven't disabled any
shrinkers from running or disabled any reclaim mechanism on remote
charges. What I see in the code is that when the remote memcg runs out
of memory is that try_to_free_mem_cgroup_pages() gets called as normal
and we do the MAX_RECLAIM_RETRIES as normal.

It's only in the (rare as you point out) case where the kernel is
completely unable to find memory in the remote memcg that we need to
do the special handling that's described in this patch. Although this
situation is probably quite rare in real-world scenarios, it's easily
reproducible (and the attached test in the series does so), so my
feeling is that it needs some sane handling, which I'm attempting to
do in this patch.