Re: Excessive xfs_inode allocations trigger OOM killer

[Michal Hocko <mhocko@kernel.org>]

On Mon 26-09-16 19:33:09, Florian Weimer wrote:
> * Michal Hocko:
> 
> > On Wed 21-09-16 07:46:12, Dave Chinner wrote:
> >> [cc Michal, linux-mm@kvack.org]
> >> 
> >> On Tue, Sep 20, 2016 at 10:56:31PM +0200, Florian Weimer wrote:
> > [...]
> >> > [51669.515086] make invoked oom-killer:
> >> > gfp_mask=0x27000c0(GFP_KERNEL_ACCOUNT|__GFP_NOTRACK), order=2,
> >> > oom_score_adj=0
> >> > [51669.515092] CPU: 1 PID: 1202 Comm: make Tainted: G I 4.7.1fw #1
> >> > [51669.515093] Hardware name: System manufacturer System Product
> >> > Name/P6X58D-E, BIOS 0701 05/10/2011
> >> > [51669.515095] 0000000000000000 ffffffff812a7d39 0000000000000000
> >> > 0000000000000000
> >> > [51669.515098] ffffffff8114e4da ffff880018707d98 0000000000000000
> >> > 000000000066ca81
> >> > [51669.515100] ffffffff8170e88d ffffffff810fe69e ffff88033fc38728
> >> > 0000000200000006
> >> > [51669.515102] Call Trace:
> >> > [51669.515108]  [<ffffffff812a7d39>] ? dump_stack+0x46/0x5d
> >> > [51669.515113]  [<ffffffff8114e4da>] ? dump_header.isra.12+0x51/0x176
> >> > [51669.515116]  [<ffffffff810fe69e>] ? oom_kill_process+0x32e/0x420
> >> > [51669.515119]  [<ffffffff811003a0>] ? page_alloc_cpu_notify+0x40/0x40
> >> > [51669.515120]  [<ffffffff810fdcdc>] ? find_lock_task_mm+0x2c/0x70
> >> > [51669.515122]  [<ffffffff810fea6d>] ? out_of_memory+0x28d/0x2d0
> >> > [51669.515125]  [<ffffffff81103137>] ? __alloc_pages_nodemask+0xb97/0xc90
> >> > [51669.515128]  [<ffffffff81076d9c>] ? copy_process.part.54+0xec/0x17a0
> >> > [51669.515131]  [<ffffffff81123318>] ? handle_mm_fault+0xaa8/0x1900
> >> > [51669.515133]  [<ffffffff81078614>] ? _do_fork+0xd4/0x320
> >> > [51669.515137]  [<ffffffff81084ecc>] ? __set_current_blocked+0x2c/0x40
> >> > [51669.515140]  [<ffffffff810013ce>] ? do_syscall_64+0x3e/0x80
> >> > [51669.515144]  [<ffffffff8151433c>] ? entry_SYSCALL64_slow_path+0x25/0x25
> >> .....
> >> > [51669.515194] DMA: 1*4kB (U) 1*8kB (U) 1*16kB (U) 0*32kB 2*64kB
> >> > (U) 1*128kB (U) 1*256kB (U) 0*512kB 1*1024kB (U) 1*2048kB (M)
> >> > 3*4096kB (M) = 15900kB
> >> > [51669.515202] DMA32: 45619*4kB (UME) 73*8kB (UM) 0*16kB 0*32kB
> >> > 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB =
> >> > 183060kB
> >> > [51669.515209] Normal: 39979*4kB (UE) 0*8kB 0*16kB 0*32kB 0*64kB
> >> > 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 159916kB
> >> .....
> >> 
> >> Alright, that's what I suspected. high order allocation for a new
> >> kernel stack and memory is so fragmented that a contiguous
> >> allocation fails. Really, this is a memory reclaim issue, not an XFS
> >> issue.  There is lots of reclaimable memory available, but memory
> >> reclaim is:
> >> 
> >> 	a) not trying hard enough to reclaim reclaimable memory; and
> >> 	b) not waiting for memory compaction to rebuild contiguous
> >> 	   memory regions for high order allocations.
> >> 
> >> Instead, it is declaring OOM and kicking the killer to free memory
> >> held busy userspace.
> >
> > Yes this was the case with 4.7 kernel. There is a workaround sitting in
> > the linus tree 6b4e3181d7bd ("mm, oom: prevent premature OOM killer
> > invocation for high order request") which should get to stable
> > eventually. More approapriate fix is currently in the linux-next.
> >
> > Testing the same workload with linux-next would be very helpful.
> 
> I'm not sure if I can reproduce this issue in a sufficiently reliable
> way, but I can try.  (I still have not found the process which causes
> the xfs_inode allocations go up.)
> 
> Is linux-next still the tree to test?

Yes it contains all the compaction related fixes which we believe to
address recent higher order OOMs.

-- 
Michal Hocko
SUSE Labs