[RFC,v3,00/12] Use obj_cgroup APIs to charge the LRU pages
mbox series

Message ID 20210421070059.69361-1-songmuchun@bytedance.com
Headers show
Series
  • Use obj_cgroup APIs to charge the LRU pages
Related show

Message

Muchun Song April 21, 2021, 7 a.m. UTC
This is v3 based on the top of the series[1] (memcontrol code cleanup and
simplification). Roman is working on the generalization of obj_cgroup API.
But before that, hope someone can review this patches for correctness.

Since the following patchsets applied. All the kernel memory are charged
with the new APIs of obj_cgroup.

	[v17,00/19] The new cgroup slab memory controller[2]
	[v5,0/7] Use obj_cgroup APIs to charge kmem pages[3]

But user memory allocations (LRU pages) pinning memcgs for a long time -
it exists at a larger scale and is causing recurring problems in the real
world: page cache doesn't get reclaimed for a long time, or is used by the
second, third, fourth, ... instance of the same job that was restarted into
a new cgroup every time. Unreclaimable dying cgroups pile up, waste memory,
and make page reclaim very inefficient.

We can convert LRU pages and most other raw memcg pins to the objcg direction
to fix this problem, and then the LRU pages will not pin the memcgs.

This patchset aims to make the LRU pages to drop the reference to memory
cgroup by using the APIs of obj_cgroup. Finally, we can see that the number
of the dying cgroups will not increase if we run the following test script.

```bash
#!/bin/bash

cat /proc/cgroups | grep memory

cd /sys/fs/cgroup/memory

for i in range{1..500}
do
	mkdir test
	echo $$ > test/cgroup.procs
	sleep 60 &
	echo $$ > cgroup.procs
	echo `cat test/cgroup.procs` > cgroup.procs
	rmdir test
done

cat /proc/cgroups | grep memory
```

Thanks.

[1] https://lore.kernel.org/linux-mm/20210417043538.9793-1-songmuchun@bytedance.com/
[2] https://lore.kernel.org/linux-mm/20200623015846.1141975-1-guro@fb.com/
[3] https://lore.kernel.org/linux-mm/20210319163821.20704-1-songmuchun@bytedance.com/

Changlogs in RFC v3:
  1. Drop the code cleanup and simplification patches. Gather those patches
     into a separate series[1].
  2. Rework patch #1 suggested by Johannes.

Changlogs in RFC v2:
  1. Collect Acked-by tags by Johannes. Thanks.
  2. Rework lruvec_holds_page_lru_lock() suggested by Johannes. Thanks.
  3. Fix move_pages_to_lru().

Muchun Song (12):
  mm: memcontrol: move the objcg infrastructure out of CONFIG_MEMCG_KMEM
  mm: memcontrol: introduce compact_lock_page_lruvec_irqsave
  mm: memcontrol: make lruvec lock safe when the LRU pages reparented
  mm: vmscan: rework move_pages_to_lru()
  mm: thp: introduce lock/unlock_split_queue{_irqsave}()
  mm: thp: make deferred split queue lock safe when the LRU pages
    reparented
  mm: memcontrol: make all the callers of page_memcg() safe
  mm: memcontrol: introduce memcg_reparent_ops
  mm: memcontrol: use obj_cgroup APIs to charge the LRU pages
  mm: memcontrol: rename {un}lock_page_memcg() to {un}lock_page_objcg()
  mm: lru: add VM_BUG_ON_PAGE to lru maintenance function
  mm: lru: use lruvec lock to serialize memcg changes

 Documentation/admin-guide/cgroup-v1/memory.rst |   2 +-
 fs/buffer.c                                    |  13 +-
 fs/fs-writeback.c                              |  23 +-
 fs/iomap/buffered-io.c                         |   4 +-
 include/linux/memcontrol.h                     | 182 ++++----
 include/linux/mm_inline.h                      |   6 +
 mm/compaction.c                                |  36 +-
 mm/filemap.c                                   |   2 +-
 mm/huge_memory.c                               | 171 ++++++--
 mm/memcontrol.c                                | 562 ++++++++++++++++++-------
 mm/migrate.c                                   |   4 +
 mm/page-writeback.c                            |  24 +-
 mm/page_io.c                                   |   5 +-
 mm/rmap.c                                      |  14 +-
 mm/swap.c                                      |  46 +-
 mm/vmscan.c                                    |  56 ++-
 16 files changed, 795 insertions(+), 355 deletions(-)

Comments

Muchun Song May 18, 2021, 1:49 p.m. UTC | #1
Ping...

Hi Johannes and Roman,

Any suggestions on this patch set?

Thanks.

On Wed, Apr 21, 2021 at 3:01 PM Muchun Song <songmuchun@bytedance.com> wrote:
>
> This is v3 based on the top of the series[1] (memcontrol code cleanup and
> simplification). Roman is working on the generalization of obj_cgroup API.
> But before that, hope someone can review this patches for correctness.
>
> Since the following patchsets applied. All the kernel memory are charged
> with the new APIs of obj_cgroup.
>
>         [v17,00/19] The new cgroup slab memory controller[2]
>         [v5,0/7] Use obj_cgroup APIs to charge kmem pages[3]
>
> But user memory allocations (LRU pages) pinning memcgs for a long time -
> it exists at a larger scale and is causing recurring problems in the real
> world: page cache doesn't get reclaimed for a long time, or is used by the
> second, third, fourth, ... instance of the same job that was restarted into
> a new cgroup every time. Unreclaimable dying cgroups pile up, waste memory,
> and make page reclaim very inefficient.
>
> We can convert LRU pages and most other raw memcg pins to the objcg direction
> to fix this problem, and then the LRU pages will not pin the memcgs.
>
> This patchset aims to make the LRU pages to drop the reference to memory
> cgroup by using the APIs of obj_cgroup. Finally, we can see that the number
> of the dying cgroups will not increase if we run the following test script.
>
> ```bash
> #!/bin/bash
>
> cat /proc/cgroups | grep memory
>
> cd /sys/fs/cgroup/memory
>
> for i in range{1..500}
> do
>         mkdir test
>         echo $$ > test/cgroup.procs
>         sleep 60 &
>         echo $$ > cgroup.procs
>         echo `cat test/cgroup.procs` > cgroup.procs
>         rmdir test
> done
>
> cat /proc/cgroups | grep memory
> ```
>
> Thanks.
>
> [1] https://lore.kernel.org/linux-mm/20210417043538.9793-1-songmuchun@bytedance.com/
> [2] https://lore.kernel.org/linux-mm/20200623015846.1141975-1-guro@fb.com/
> [3] https://lore.kernel.org/linux-mm/20210319163821.20704-1-songmuchun@bytedance.com/
>
> Changlogs in RFC v3:
>   1. Drop the code cleanup and simplification patches. Gather those patches
>      into a separate series[1].
>   2. Rework patch #1 suggested by Johannes.
>
> Changlogs in RFC v2:
>   1. Collect Acked-by tags by Johannes. Thanks.
>   2. Rework lruvec_holds_page_lru_lock() suggested by Johannes. Thanks.
>   3. Fix move_pages_to_lru().
>
> Muchun Song (12):
>   mm: memcontrol: move the objcg infrastructure out of CONFIG_MEMCG_KMEM
>   mm: memcontrol: introduce compact_lock_page_lruvec_irqsave
>   mm: memcontrol: make lruvec lock safe when the LRU pages reparented
>   mm: vmscan: rework move_pages_to_lru()
>   mm: thp: introduce lock/unlock_split_queue{_irqsave}()
>   mm: thp: make deferred split queue lock safe when the LRU pages
>     reparented
>   mm: memcontrol: make all the callers of page_memcg() safe
>   mm: memcontrol: introduce memcg_reparent_ops
>   mm: memcontrol: use obj_cgroup APIs to charge the LRU pages
>   mm: memcontrol: rename {un}lock_page_memcg() to {un}lock_page_objcg()
>   mm: lru: add VM_BUG_ON_PAGE to lru maintenance function
>   mm: lru: use lruvec lock to serialize memcg changes
>
>  Documentation/admin-guide/cgroup-v1/memory.rst |   2 +-
>  fs/buffer.c                                    |  13 +-
>  fs/fs-writeback.c                              |  23 +-
>  fs/iomap/buffered-io.c                         |   4 +-
>  include/linux/memcontrol.h                     | 182 ++++----
>  include/linux/mm_inline.h                      |   6 +
>  mm/compaction.c                                |  36 +-
>  mm/filemap.c                                   |   2 +-
>  mm/huge_memory.c                               | 171 ++++++--
>  mm/memcontrol.c                                | 562 ++++++++++++++++++-------
>  mm/migrate.c                                   |   4 +
>  mm/page-writeback.c                            |  24 +-
>  mm/page_io.c                                   |   5 +-
>  mm/rmap.c                                      |  14 +-
>  mm/swap.c                                      |  46 +-
>  mm/vmscan.c                                    |  56 ++-
>  16 files changed, 795 insertions(+), 355 deletions(-)
>
> --
> 2.11.0
>
Roman Gushchin May 18, 2021, 2:17 p.m. UTC | #2
Hi Muchun!

It looks like the writeback problem will be solved in a different way, which will not require generalization of the obj_cgroup api to the cgroup level. It’s not fully confirmed yet though. We still might wanna do this generalization lingn-term, but as now I have no objections for continuing the work on your patchset. I’m on pto this week, but will take a deeper look at your patches early next week. Sorry for the delay.

Thanks!

Sent from my iPhone

> On May 18, 2021, at 06:50, Muchun Song <songmuchun@bytedance.com> wrote:
> 
> Ping...
> 
> Hi Johannes and Roman,
> 
> Any suggestions on this patch set?
> 
> Thanks.
> 
>> On Wed, Apr 21, 2021 at 3:01 PM Muchun Song <songmuchun@bytedance.com> wrote:
>> 
>> This is v3 based on the top of the series[1] (memcontrol code cleanup and
>> simplification). Roman is working on the generalization of obj_cgroup API.
>> But before that, hope someone can review this patches for correctness.
>> 
>> Since the following patchsets applied. All the kernel memory are charged
>> with the new APIs of obj_cgroup.
>> 
>>        [v17,00/19] The new cgroup slab memory controller[2]
>>        [v5,0/7] Use obj_cgroup APIs to charge kmem pages[3]
>> 
>> But user memory allocations (LRU pages) pinning memcgs for a long time -
>> it exists at a larger scale and is causing recurring problems in the real
>> world: page cache doesn't get reclaimed for a long time, or is used by the
>> second, third, fourth, ... instance of the same job that was restarted into
>> a new cgroup every time. Unreclaimable dying cgroups pile up, waste memory,
>> and make page reclaim very inefficient.
>> 
>> We can convert LRU pages and most other raw memcg pins to the objcg direction
>> to fix this problem, and then the LRU pages will not pin the memcgs.
>> 
>> This patchset aims to make the LRU pages to drop the reference to memory
>> cgroup by using the APIs of obj_cgroup. Finally, we can see that the number
>> of the dying cgroups will not increase if we run the following test script.
>> 
>> ```bash
>> #!/bin/bash
>> 
>> cat /proc/cgroups | grep memory
>> 
>> cd /sys/fs/cgroup/memory
>> 
>> for i in range{1..500}
>> do
>>        mkdir test
>>        echo $$ > test/cgroup.procs
>>        sleep 60 &
>>        echo $$ > cgroup.procs
>>        echo `cat test/cgroup.procs` > cgroup.procs
>>        rmdir test
>> done
>> 
>> cat /proc/cgroups | grep memory
>> ```
>> 
>> Thanks.
>> 
>> [1] https://lore.kernel.org/linux-mm/20210417043538.9793-1-songmuchun@bytedance.com/
>> [2] https://lore.kernel.org/linux-mm/20200623015846.1141975-1-guro@fb.com/
>> [3] https://lore.kernel.org/linux-mm/20210319163821.20704-1-songmuchun@bytedance.com/
>> 
>> Changlogs in RFC v3:
>>  1. Drop the code cleanup and simplification patches. Gather those patches
>>     into a separate series[1].
>>  2. Rework patch #1 suggested by Johannes.
>> 
>> Changlogs in RFC v2:
>>  1. Collect Acked-by tags by Johannes. Thanks.
>>  2. Rework lruvec_holds_page_lru_lock() suggested by Johannes. Thanks.
>>  3. Fix move_pages_to_lru().
>> 
>> Muchun Song (12):
>>  mm: memcontrol: move the objcg infrastructure out of CONFIG_MEMCG_KMEM
>>  mm: memcontrol: introduce compact_lock_page_lruvec_irqsave
>>  mm: memcontrol: make lruvec lock safe when the LRU pages reparented
>>  mm: vmscan: rework move_pages_to_lru()
>>  mm: thp: introduce lock/unlock_split_queue{_irqsave}()
>>  mm: thp: make deferred split queue lock safe when the LRU pages
>>    reparented
>>  mm: memcontrol: make all the callers of page_memcg() safe
>>  mm: memcontrol: introduce memcg_reparent_ops
>>  mm: memcontrol: use obj_cgroup APIs to charge the LRU pages
>>  mm: memcontrol: rename {un}lock_page_memcg() to {un}lock_page_objcg()
>>  mm: lru: add VM_BUG_ON_PAGE to lru maintenance function
>>  mm: lru: use lruvec lock to serialize memcg changes
>> 
>> Documentation/admin-guide/cgroup-v1/memory.rst |   2 +-
>> fs/buffer.c                                    |  13 +-
>> fs/fs-writeback.c                              |  23 +-
>> fs/iomap/buffered-io.c                         |   4 +-
>> include/linux/memcontrol.h                     | 182 ++++----
>> include/linux/mm_inline.h                      |   6 +
>> mm/compaction.c                                |  36 +-
>> mm/filemap.c                                   |   2 +-
>> mm/huge_memory.c                               | 171 ++++++--
>> mm/memcontrol.c                                | 562 ++++++++++++++++++-------
>> mm/migrate.c                                   |   4 +
>> mm/page-writeback.c                            |  24 +-
>> mm/page_io.c                                   |   5 +-
>> mm/rmap.c                                      |  14 +-
>> mm/swap.c                                      |  46 +-
>> mm/vmscan.c                                    |  56 ++-
>> 16 files changed, 795 insertions(+), 355 deletions(-)
>> 
>> --
>> 2.11.0
>>
Muchun Song May 20, 2021, 3:20 a.m. UTC | #3
On Tue, May 18, 2021 at 10:17 PM Roman Gushchin <guro@fb.com> wrote:
>
> Hi Muchun!
>
> It looks like the writeback problem will be solved in a different way, which will not require generalization of the obj_cgroup api to the cgroup level. It’s not fully confirmed yet though. We still might wanna do this generalization lingn-term, but as now I have no objections for continuing the work on your patchset. I’m on pto this week, but will take a deeper look at your patches early next week. Sorry for the delay.

Waiting on your review. Thanks Roman.

>
> Thanks!
>
> Sent from my iPhone
>
> > On May 18, 2021, at 06:50, Muchun Song <songmuchun@bytedance.com> wrote:
> >
> > Ping...
> >
> > Hi Johannes and Roman,
> >
> > Any suggestions on this patch set?
> >
> > Thanks.
> >
> >> On Wed, Apr 21, 2021 at 3:01 PM Muchun Song <songmuchun@bytedance.com> wrote:
> >>
> >> This is v3 based on the top of the series[1] (memcontrol code cleanup and
> >> simplification). Roman is working on the generalization of obj_cgroup API.
> >> But before that, hope someone can review this patches for correctness.
> >>
> >> Since the following patchsets applied. All the kernel memory are charged
> >> with the new APIs of obj_cgroup.
> >>
> >>        [v17,00/19] The new cgroup slab memory controller[2]
> >>        [v5,0/7] Use obj_cgroup APIs to charge kmem pages[3]
> >>
> >> But user memory allocations (LRU pages) pinning memcgs for a long time -
> >> it exists at a larger scale and is causing recurring problems in the real
> >> world: page cache doesn't get reclaimed for a long time, or is used by the
> >> second, third, fourth, ... instance of the same job that was restarted into
> >> a new cgroup every time. Unreclaimable dying cgroups pile up, waste memory,
> >> and make page reclaim very inefficient.
> >>
> >> We can convert LRU pages and most other raw memcg pins to the objcg direction
> >> to fix this problem, and then the LRU pages will not pin the memcgs.
> >>
> >> This patchset aims to make the LRU pages to drop the reference to memory
> >> cgroup by using the APIs of obj_cgroup. Finally, we can see that the number
> >> of the dying cgroups will not increase if we run the following test script.
> >>
> >> ```bash
> >> #!/bin/bash
> >>
> >> cat /proc/cgroups | grep memory
> >>
> >> cd /sys/fs/cgroup/memory
> >>
> >> for i in range{1..500}
> >> do
> >>        mkdir test
> >>        echo $$ > test/cgroup.procs
> >>        sleep 60 &
> >>        echo $$ > cgroup.procs
> >>        echo `cat test/cgroup.procs` > cgroup.procs
> >>        rmdir test
> >> done
> >>
> >> cat /proc/cgroups | grep memory
> >> ```
> >>
> >> Thanks.
> >>
> >> [1] https://lore.kernel.org/linux-mm/20210417043538.9793-1-songmuchun@bytedance.com/
> >> [2] https://lore.kernel.org/linux-mm/20200623015846.1141975-1-guro@fb.com/
> >> [3] https://lore.kernel.org/linux-mm/20210319163821.20704-1-songmuchun@bytedance.com/
> >>
> >> Changlogs in RFC v3:
> >>  1. Drop the code cleanup and simplification patches. Gather those patches
> >>     into a separate series[1].
> >>  2. Rework patch #1 suggested by Johannes.
> >>
> >> Changlogs in RFC v2:
> >>  1. Collect Acked-by tags by Johannes. Thanks.
> >>  2. Rework lruvec_holds_page_lru_lock() suggested by Johannes. Thanks.
> >>  3. Fix move_pages_to_lru().
> >>
> >> Muchun Song (12):
> >>  mm: memcontrol: move the objcg infrastructure out of CONFIG_MEMCG_KMEM
> >>  mm: memcontrol: introduce compact_lock_page_lruvec_irqsave
> >>  mm: memcontrol: make lruvec lock safe when the LRU pages reparented
> >>  mm: vmscan: rework move_pages_to_lru()
> >>  mm: thp: introduce lock/unlock_split_queue{_irqsave}()
> >>  mm: thp: make deferred split queue lock safe when the LRU pages
> >>    reparented
> >>  mm: memcontrol: make all the callers of page_memcg() safe
> >>  mm: memcontrol: introduce memcg_reparent_ops
> >>  mm: memcontrol: use obj_cgroup APIs to charge the LRU pages
> >>  mm: memcontrol: rename {un}lock_page_memcg() to {un}lock_page_objcg()
> >>  mm: lru: add VM_BUG_ON_PAGE to lru maintenance function
> >>  mm: lru: use lruvec lock to serialize memcg changes
> >>
> >> Documentation/admin-guide/cgroup-v1/memory.rst |   2 +-
> >> fs/buffer.c                                    |  13 +-
> >> fs/fs-writeback.c                              |  23 +-
> >> fs/iomap/buffered-io.c                         |   4 +-
> >> include/linux/memcontrol.h                     | 182 ++++----
> >> include/linux/mm_inline.h                      |   6 +
> >> mm/compaction.c                                |  36 +-
> >> mm/filemap.c                                   |   2 +-
> >> mm/huge_memory.c                               | 171 ++++++--
> >> mm/memcontrol.c                                | 562 ++++++++++++++++++-------
> >> mm/migrate.c                                   |   4 +
> >> mm/page-writeback.c                            |  24 +-
> >> mm/page_io.c                                   |   5 +-
> >> mm/rmap.c                                      |  14 +-
> >> mm/swap.c                                      |  46 +-
> >> mm/vmscan.c                                    |  56 ++-
> >> 16 files changed, 795 insertions(+), 355 deletions(-)
> >>
> >> --
> >> 2.11.0
> >>
Roman Gushchin May 25, 2021, 5:35 p.m. UTC | #4
On Thu, May 20, 2021 at 11:20:47AM +0800, Muchun Song wrote:
> On Tue, May 18, 2021 at 10:17 PM Roman Gushchin <guro@fb.com> wrote:
> >
> > Hi Muchun!
> >
> > It looks like the writeback problem will be solved in a different way, which will not require generalization of the obj_cgroup api to the cgroup level. It’s not fully confirmed yet though. We still might wanna do this generalization lingn-term, but as now I have no objections for continuing the work on your patchset. I’m on pto this week, but will take a deeper look at your patches early next week. Sorry for the delay.
> 
> Waiting on your review. Thanks Roman.

It looks like the mm tree went ahead and I can't clearly apply the whole patchset.
Would you mind to rebase it and resend?

Thank you!
Muchun Song May 26, 2021, 2:41 a.m. UTC | #5
On Wed, May 26, 2021 at 1:35 AM Roman Gushchin <guro@fb.com> wrote:
>
> On Thu, May 20, 2021 at 11:20:47AM +0800, Muchun Song wrote:
> > On Tue, May 18, 2021 at 10:17 PM Roman Gushchin <guro@fb.com> wrote:
> > >
> > > Hi Muchun!
> > >
> > > It looks like the writeback problem will be solved in a different way, which will not require generalization of the obj_cgroup api to the cgroup level. It’s not fully confirmed yet though. We still might wanna do this generalization lingn-term, but as now I have no objections for continuing the work on your patchset. I’m on pto this week, but will take a deeper look at your patches early next week. Sorry for the delay.
> >
> > Waiting on your review. Thanks Roman.
>
> It looks like the mm tree went ahead and I can't clearly apply the whole patchset.
> Would you mind to rebase it and resend?

Got it. Will do that. Thanks.

>
> Thank you!