Re: memcg swap doesn't work in mmotm-2014-07-09-17-08?

[Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>]

On Wed, Jul 16, 2014 at 03:57:21PM -0400, Johannes Weiner wrote:
> On Wed, Jul 16, 2014 at 03:31:29PM -0400, Naoya Horiguchi wrote:
> > On Wed, Jul 16, 2014 at 02:37:27PM -0400, Johannes Weiner wrote:
> > > On Wed, Jul 16, 2014 at 02:10:07PM -0400, Naoya Horiguchi wrote:
> > > > Hi,
> > > >
> > > > It seems that when a process in some memcg tries to allocate more than
> > > > memcg.limit_in_bytes, oom happens instead of swaping out in
> > > > mmotm-2014-07-09-17-08 (memcg.memsw.limit_in_bytes is large enough).
> > > > It does work in v3.16-rc3, so I think latest patches changed something.
> > > > I'm not familiar with memcg internally, so no idea about what caused it.
> > > > Could you see the problem?
> > >
> > > There are a lot of changes in memory and swap accounting, but I can
> > > not reproduce what you are describing: I set up a cgroup with a 100MB
> > > memory limit and an unlimited memory+swap, then start a task in there
> > > that faults 200MB worth of anonymous pages.  The result is 100MB in
> > > memory, 100MB in swap:
> > 
> > Hmm, this is almost the same test condition.
> > 
> > > cache 0
> > > rss 104267776
> > > rss_huge 0
> > > mapped_file 0
> > > writeback 0
> > > swap 105545728
> > > pgpgin 26367
> > > pgpgout 25950
> > > pgfault 26695
> > > pgmajfault 32
> > > inactive_anon 52285440
> > > active_anon 51982336
> > > inactive_file 0
> > > active_file 0
> > > unevictable 0
> > > hierarchical_memory_limit 104857600
> > > hierarchical_memsw_limit 18446744073709551615
> > >
> > > Filename                                Type            Size    Used    Priority
> > > /swapfile                               file            8388604 109800  -1
> > >
> > > Could you provide more detail on your configuration and test case?
> > 
> > What I did is like this:
> > 
> >   [root@test_140715-0036 ~]# cat tmp_memcg_swap.sh
> >   cgcreate -g memory:test1
> >   cgset -r memory.limit_in_bytes=0x1000000 test1        # 16 MB
> >   cgset -r memory.memsw.limit_in_bytes=0x4000000 test1  # 64 MB
> >   cgget -g memory:test1
> >   swapon -s
> >   cgexec -g memory:test1 memhog -r1 20m                 # alloc 20 MB
> 
> Okay, I tried with the smaller parameters that you use, but still
> couldn't reproduce it:
> 
> cache 32768
> rss 15867904
> rss_huge 0
> mapped_file 0
> writeback 0
> swap 5222400
> pgpgin 1725
> pgpgout 1420
> pgfault 2015
> pgmajfault 7
> inactive_anon 10133504
> active_anon 5726208
> inactive_file 32768
> active_file 0
> unevictable 0
> hierarchical_memory_limit 16777216
> hierarchical_memsw_limit 67108864
> 
> 16M in memory, 5M in swap, no OOM kills.
> 
> > then output is like this:
> > 
> >   [root@test_140715-0036 ~]# bash tmp_memcg_swap.sh
> >   test1:
> >   memory.pressure_level:
> >   memory.kmem.max_usage_in_bytes: 0
> >   memory.use_hierarchy: 1
> >   memory.swappiness: 60
> >   memory.memsw.failcnt: 0
> >   memory.limit_in_bytes: 16777216
> >   memory.memsw.max_usage_in_bytes: 0
> >   memory.usage_in_bytes: 0
> >   memory.memsw.limit_in_bytes: 67108864
> >   memory.failcnt: 0
> >   memory.kmem.limit_in_bytes: 18446744073709551615
> >   memory.force_empty:
> >   memory.kmem.slabinfo:
> >   memory.memsw.usage_in_bytes: 0
> >   memory.max_usage_in_bytes: 0
> >   memory.numa_stat: total=0 N0=0 N1=0 N2=0 N3=0
> >           file=0 N0=0 N1=0 N2=0 N3=0
> >           anon=0 N0=0 N1=0 N2=0 N3=0
> >           unevictable=0 N0=0 N1=0 N2=0 N3=0  
> >           hierarchical_total=0 N0=0 N1=0 N2=0 N3=0
> >           hierarchical_file=0 N0=0 N1=0 N2=0 N3=0
> >           hierarchical_anon=0 N0=0 N1=0 N2=0 N3=0
> >           hierarchical_unevictable=0 N0=0 N1=0 N2=0 N3=0
> >   memory.kmem.tcp.limit_in_bytes: 18446744073709551615
> >   memory.oom_control: oom_kill_disable 0
> >           under_oom 0
> >   memory.kmem.tcp.max_usage_in_bytes: 0
> >   memory.kmem.failcnt: 0
> >   memory.kmem.usage_in_bytes: 0
> >   memory.stat: cache 0
> >           rss 0
> >           rss_huge 0
> >           mapped_file 0
> >           writeback 0
> >           swap 0
> >           pgpgin 0
> >           pgpgout 0
> >           pgfault 0
> >           pgmajfault 0
> >           inactive_anon 0
> >           active_anon 0
> >           inactive_file 0
> >           active_file 0
> >           unevictable 0
> >           hierarchical_memory_limit 16777216
> >           hierarchical_memsw_limit 67108864
> >           total_cache 0
> >           total_rss 0
> >           total_rss_huge 0
> >           total_mapped_file 0
> >           total_writeback 0
> >           total_swap 0
> >           total_pgpgin 0
> >           total_pgpgout 0
> >           total_pgfault 0
> >           total_pgmajfault 0
> >           total_inactive_anon 0
> >           total_active_anon 0
> >           total_inactive_file 0
> >           total_active_file 0
> >           total_unevictable 0
> >           recent_rotated_anon 0
> >           recent_rotated_file 0
> >           recent_scanned_anon 0
> >           recent_scanned_file 0
> >   memory.kmem.tcp.usage_in_bytes: 0
> >   memory.move_charge_at_immigrate: 0
> >   memory.soft_limit_in_bytes: 18446744073709551615
> >   memory.kmem.tcp.failcnt: 0
> >   
> >   Filename                                Type            Size    Used    Priority
> >   /root/page_table_walker/swapfile        file    40956   0       -1
> >   ..tmp_memcg_swap.sh: line 7:  1250 Killed                  cgexec -g memory:test1 memhog -r1 20m
> > 
> > Kernel message is like this:
> > 
> >   [  425.290261] memhog invoked oom-killer: gfp_mask=0xd0, order=0, oom_score_adj=0
> >   [  425.291356] memhog cpuset=/ mems_allowed=0-3
> >   [  425.291992] CPU: 2 PID: 1234 Comm: memhog Not tainted 3.15.0-140716-1501-00003-gcb8370a1d76e #268
> 
> That's not mmotm-2014-07-09-17-08, though, is it?

Sorry for confusion, but yes. I used my own localversion,
but this kernel is mmotm-2014-07-09-17-08 (git-describe says
mmotm-2014-07-09-17-08-3-gcb8370a1d76e.)

> 
> I was running on 3.16.0-rc5-mm1-00480-gab7ca1e1e407 (latest mmots).

I'll try this kernel.
If that passes, this problem is fixed implicitly, so that's fine :)

Thanks,
Naoya Horiguchi

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>