Re: [patch 0/7] improve memcg oom killer robustness v2

[Johannes Weiner <hannes-druUgvl0LCNAfugRpC6u6w@public.gmane.org>]

Hi azur,

On Mon, Oct 07, 2013 at 01:01:49PM +0200, azurIt wrote:
> >On Thu, Sep 26, 2013 at 06:54:59PM +0200, azurIt wrote:
> >> On Wed, Sep 18, 2013 at 02:19:46PM -0400, Johannes Weiner wrote:
> >> >Here is an update.  Full replacement on top of 3.2 since we tried a
> >> >dead end and it would be more painful to revert individual changes.
> >> >
> >> >The first bug you had was the same task entering OOM repeatedly and
> >> >leaking the memcg reference, thus creating undeletable memcgs.  My
> >> >fixup added a condition that if the task already set up an OOM context
> >> >in that fault, another charge attempt would immediately return -ENOMEM
> >> >without even trying reclaim anymore.  This dropped __getblk() into an
> >> >endless loop of waking the flushers and performing global reclaim and
> >> >memcg returning -ENOMEM regardless of free memory.
> >> >
> >> >The update now basically only changes this -ENOMEM to bypass, so that
> >> >the memory is not accounted and the limit ignored.  OOM killed tasks
> >> >are granted the same right, so that they can exit quickly and release
> >> >memory.  Likewise, we want a task that hit the OOM condition also to
> >> >finish the fault quickly so that it can invoke the OOM killer.
> >> >
> >> >Does the following work for you, azur?
> >> 
> >> 
> >> Johannes,
> >> 
> >> bad news everyone! :(
> >> 
> >> Unfortunaely, two different problems appears today:
> >> 
> >> 1.) This looks like my very original problem - stucked processes inside one cgroup. I took stacks from all of them over time but server was very slow so i had to kill them soon:
> >> http://watchdog.sk/lkmlmemcg-bug-9.tar.gz
> >> 
> >> 2.) This was just like my last problem where few processes were doing huge i/o. As sever was almost unoperable i barely killed them so no more info here, sorry.
> >
> >From one of the tasks:
> >
> >1380213238/11210/stack:[<ffffffff810528f1>] sys_sched_yield+0x41/0x70
> >1380213238/11210/stack:[<ffffffff81148ef1>] free_more_memory+0x21/0x60
> >1380213238/11210/stack:[<ffffffff8114957d>] __getblk+0x14d/0x2c0
> >1380213238/11210/stack:[<ffffffff81198a2b>] ext3_getblk+0xeb/0x240
> >1380213238/11210/stack:[<ffffffff8119d2df>] ext3_find_entry+0x13f/0x480
> >1380213238/11210/stack:[<ffffffff8119dd6d>] ext3_lookup+0x4d/0x120
> >1380213238/11210/stack:[<ffffffff81122a55>] d_alloc_and_lookup+0x45/0x90
> >1380213238/11210/stack:[<ffffffff81122ff8>] do_lookup+0x278/0x390
> >1380213238/11210/stack:[<ffffffff81124c40>] path_lookupat+0x120/0x800
> >1380213238/11210/stack:[<ffffffff81125355>] do_path_lookup+0x35/0xd0
> >1380213238/11210/stack:[<ffffffff811254d9>] user_path_at_empty+0x59/0xb0
> >1380213238/11210/stack:[<ffffffff81125541>] user_path_at+0x11/0x20
> >1380213238/11210/stack:[<ffffffff81115b70>] sys_faccessat+0xd0/0x200
> >1380213238/11210/stack:[<ffffffff81115cb8>] sys_access+0x18/0x20
> >1380213238/11210/stack:[<ffffffff815ccc26>] system_call_fastpath+0x18/0x1d
> >
> >Should have seen this coming... it's still in that braindead
> >__getblk() loop, only from a syscall this time (no OOM path).  The
> >group's memory.stat looks like this:
> >
> >cache 0
> >rss 0
> >mapped_file 0
> >pgpgin 0
> >pgpgout 0
> >swap 0
> >pgfault 0
> >pgmajfault 0
> >inactive_anon 0
> >active_anon 0
> >inactive_file 0
> >active_file 0
> >unevictable 0
> >hierarchical_memory_limit 209715200
> >hierarchical_memsw_limit 209715200
> >total_cache 0
> >total_rss 209715200
> >total_mapped_file 0
> >total_pgpgin 1028153297
> >total_pgpgout 1028102097
> >total_swap 0
> >total_pgfault 1352903120
> >total_pgmajfault 45342
> >total_inactive_anon 0
> >total_active_anon 209715200
> >total_inactive_file 0
> >total_active_file 0
> >total_unevictable 0
> >
> >with anonymous pages to the limit and you probably don't have any swap
> >space enabled to anything in the group.
> >
> >I guess there is no way around annotating that __getblk() loop.  The
> >best solution right now is probably to use __GFP_NOFAIL.  For one, we
> >can let the allocation bypass the memcg limit if reclaim can't make
> >progress.  But also, the loop is then actually happening inside the
> >page allocator, where it should happen, and not around ad-hoc direct
> >reclaim in buffer.c.
> >
> >Can you try this on top of our ever-growing stack of patches?
> 
> 
> 
> 
> Joahnnes,
> 
> looks like the problem is completely resolved :) Thank you, Michal
> Hocko and everyone involved for help and time.

Thanks a lot for your patience.  I will send out the fixes for 3.12.

> One more thing: I see that your patches are going into 3.12. Is
> there a chance to get them also into 3.2? Is Ben Hutchings (current
> maintainer of 3.2 branch) competent to decide this? Should i contact
> him directly? I can't upgrade to 3.12 because stable grsecurity is
> for 3.2 and i don't think this will change in near future.

Yes, I'll send them to stable.  The original OOM killer rework was not
tagged for stable, but since we have a known deadlock problem, I think
it makes sense to include them after all.