Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Andrey Ryabinin]

On 11/30/19 12:45 AM, Pavel Tikhomirov wrote:
> We have a problem that shrinker_rwsem can be held for a long time for
> read in shrink_slab, at the same time any process which is trying to
> manage shrinkers hangs.
> 
> The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
> It tries to shrink something on nfs (hard) but nfs server is dead at
> these moment already and rpc will never succeed. Generally any shrinker
> can take significant time to do_shrink_slab, so it's a bad idea to hold
> the list lock here.
> 
> We have a similar problem in shrink_slab_memcg, except that we are
> traversing shrinker_map+shrinker_idr there.
> 
> The idea of the patch is to inc a refcount to the chosen shrinker so it
> won't disappear and release shrinker_rwsem while we are in
> do_shrink_slab, after that we will reacquire shrinker_rwsem, dec
> the refcount and continue the traversal.
> 
> We also need a wait_queue so that unregister_shrinker can wait for the
> refcnt to become zero. Only after these we can safely remove the
> shrinker from list and idr, and free the shrinker.
> 
> I've reproduced the nfs hang in do_shrink_slab with the patch applied on
> ms kernel, all other mount/unmount pass fine without any hang.
> 
> Here is a reproduction on kernel without patch:
> 
> 1) Setup nfs on server node with some files in it (e.g. 200)
> 
> [server]# cat /etc/exports
> /vz/nfs2 *(ro,no_root_squash,no_subtree_check,async)
> 
> 2) Hard mount it on client node
> 
> [client]# mount -ohard 10.94.3.40:/vz/nfs2 /mnt
> 
> 3) Open some (e.g. 200) files on the mount
> 
> [client]# for i in $(find /mnt/ -type f | head -n 200); \
>   do setsid sleep 1000 &>/dev/null <$i & done
> 
> 4) Kill all openers
> 
> [client]# killall sleep -9
> 
> 5) Put your network cable out on client node
> 
> 6) Drop caches on the client, it will hang on nfs while holding
>   shrinker_rwsem lock for read
> 
> [client]# echo 3 > /proc/sys/vm/drop_caches
> 
>   crash> bt ...
>   PID: 18739  TASK: ...  CPU: 3   COMMAND: "bash"
>    #0 [...] __schedule at ...
>    #1 [...] schedule at ...
>    #2 [...] rpc_wait_bit_killable at ... [sunrpc]
>    #3 [...] __wait_on_bit at ...
>    #4 [...] out_of_line_wait_on_bit at ...
>    #5 [...] _nfs4_proc_delegreturn at ... [nfsv4]
>    #6 [...] nfs4_proc_delegreturn at ... [nfsv4]
>    #7 [...] nfs_do_return_delegation at ... [nfsv4]
>    #8 [...] nfs4_evict_inode at ... [nfsv4]
>    #9 [...] evict at ...
>   #10 [...] dispose_list at ...
>   #11 [...] prune_icache_sb at ...
>   #12 [...] super_cache_scan at ...
>   #13 [...] do_shrink_slab at ...
>   #14 [...] shrink_slab at ...
>   #15 [...] drop_slab_node at ...
>   #16 [...] drop_slab at ...
>   #17 [...] drop_caches_sysctl_handler at ...
>   #18 [...] proc_sys_call_handler at ...
>   #19 [...] vfs_write at ...
>   #20 [...] ksys_write at ...
>   #21 [...] do_syscall_64 at ...
>   #22 [...] entry_SYSCALL_64_after_hwframe at ...
> 
> 7) All other mount/umount activity now hangs with no luck to take
>   shrinker_rwsem for write.
> 
> [client]# mount -t tmpfs tmpfs /tmp
> 
>   crash> bt ...
>   PID: 5464   TASK: ...  CPU: 3   COMMAND: "mount"
>    #0 [...] __schedule at ...
>    #1 [...] schedule at ...
>    #2 [...] rwsem_down_write_slowpath at ...
>    #3 [...] prealloc_shrinker at ...
>    #4 [...] alloc_super at ...
>    #5 [...] sget at ...
>    #6 [...] mount_nodev at ...
>    #7 [...] legacy_get_tree at ...
>    #8 [...] vfs_get_tree at ...
>    #9 [...] do_mount at ...
>   #10 [...] ksys_mount at ...
>   #11 [...] __x64_sys_mount at ...
>   #12 [...] do_syscall_64 at ...
>   #13 [...] entry_SYSCALL_64_after_hwframe at ...
> 
 

I don't think this patch solves the problem, it only fixes one minor symptom of it.
The actual problem here the reclaim hang in the nfs.
It means that any process, including kswapd, may go into nfs inode reclaim and stuck there.

Even mount() itself has GFP_KERNEL allocations in its path, so it still might stuck there even with your patch.

I think this should be handled on nfs/vfs level by making  inode eviction during reclaim more asynchronous.

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Shakeel Butt]

On Mon, Dec 2, 2019 at 8:37 AM Andrey Ryabinin <aryabinin@virtuozzo.com> wrote:
>
>
> On 11/30/19 12:45 AM, Pavel Tikhomirov wrote:
> > We have a problem that shrinker_rwsem can be held for a long time for
> > read in shrink_slab, at the same time any process which is trying to
> > manage shrinkers hangs.
> >
> > The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
> > It tries to shrink something on nfs (hard) but nfs server is dead at
> > these moment already and rpc will never succeed. Generally any shrinker
> > can take significant time to do_shrink_slab, so it's a bad idea to hold
> > the list lock here.
> >
> > We have a similar problem in shrink_slab_memcg, except that we are
> > traversing shrinker_map+shrinker_idr there.
> >
> > The idea of the patch is to inc a refcount to the chosen shrinker so it
> > won't disappear and release shrinker_rwsem while we are in
> > do_shrink_slab, after that we will reacquire shrinker_rwsem, dec
> > the refcount and continue the traversal.
> >
> > We also need a wait_queue so that unregister_shrinker can wait for the
> > refcnt to become zero. Only after these we can safely remove the
> > shrinker from list and idr, and free the shrinker.
> >
> > I've reproduced the nfs hang in do_shrink_slab with the patch applied on
> > ms kernel, all other mount/unmount pass fine without any hang.
> >
> > Here is a reproduction on kernel without patch:
> >
> > 1) Setup nfs on server node with some files in it (e.g. 200)
> >
> > [server]# cat /etc/exports
> > /vz/nfs2 *(ro,no_root_squash,no_subtree_check,async)
> >
> > 2) Hard mount it on client node
> >
> > [client]# mount -ohard 10.94.3.40:/vz/nfs2 /mnt
> >
> > 3) Open some (e.g. 200) files on the mount
> >
> > [client]# for i in $(find /mnt/ -type f | head -n 200); \
> >   do setsid sleep 1000 &>/dev/null <$i & done
> >
> > 4) Kill all openers
> >
> > [client]# killall sleep -9
> >
> > 5) Put your network cable out on client node
> >
> > 6) Drop caches on the client, it will hang on nfs while holding
> >   shrinker_rwsem lock for read
> >
> > [client]# echo 3 > /proc/sys/vm/drop_caches
> >
> >   crash> bt ...
> >   PID: 18739  TASK: ...  CPU: 3   COMMAND: "bash"
> >    #0 [...] __schedule at ...
> >    #1 [...] schedule at ...
> >    #2 [...] rpc_wait_bit_killable at ... [sunrpc]
> >    #3 [...] __wait_on_bit at ...
> >    #4 [...] out_of_line_wait_on_bit at ...
> >    #5 [...] _nfs4_proc_delegreturn at ... [nfsv4]
> >    #6 [...] nfs4_proc_delegreturn at ... [nfsv4]
> >    #7 [...] nfs_do_return_delegation at ... [nfsv4]
> >    #8 [...] nfs4_evict_inode at ... [nfsv4]
> >    #9 [...] evict at ...
> >   #10 [...] dispose_list at ...
> >   #11 [...] prune_icache_sb at ...
> >   #12 [...] super_cache_scan at ...
> >   #13 [...] do_shrink_slab at ...
> >   #14 [...] shrink_slab at ...
> >   #15 [...] drop_slab_node at ...
> >   #16 [...] drop_slab at ...
> >   #17 [...] drop_caches_sysctl_handler at ...
> >   #18 [...] proc_sys_call_handler at ...
> >   #19 [...] vfs_write at ...
> >   #20 [...] ksys_write at ...
> >   #21 [...] do_syscall_64 at ...
> >   #22 [...] entry_SYSCALL_64_after_hwframe at ...
> >
> > 7) All other mount/umount activity now hangs with no luck to take
> >   shrinker_rwsem for write.
> >
> > [client]# mount -t tmpfs tmpfs /tmp
> >
> >   crash> bt ...
> >   PID: 5464   TASK: ...  CPU: 3   COMMAND: "mount"
> >    #0 [...] __schedule at ...
> >    #1 [...] schedule at ...
> >    #2 [...] rwsem_down_write_slowpath at ...
> >    #3 [...] prealloc_shrinker at ...
> >    #4 [...] alloc_super at ...
> >    #5 [...] sget at ...
> >    #6 [...] mount_nodev at ...
> >    #7 [...] legacy_get_tree at ...
> >    #8 [...] vfs_get_tree at ...
> >    #9 [...] do_mount at ...
> >   #10 [...] ksys_mount at ...
> >   #11 [...] __x64_sys_mount at ...
> >   #12 [...] do_syscall_64 at ...
> >   #13 [...] entry_SYSCALL_64_after_hwframe at ...
> >
>
>
> I don't think this patch solves the problem, it only fixes one minor symptom of it.
> The actual problem here the reclaim hang in the nfs.
> It means that any process, including kswapd, may go into nfs inode reclaim and stuck there.
>
> Even mount() itself has GFP_KERNEL allocations in its path, so it still might stuck there even with your patch.
>
> I think this should be handled on nfs/vfs level by making  inode eviction during reclaim more asynchronous.

Though I agree that we should be fixing shrinkers to not get stuck
(and be more async), I still think the problem this patch is solving
is worth fixing. On machines running multiple workloads, one job stuck
in slab shrinker and blocking all other unrelated jobs wanting
shrinker_rwsem, breaks the isolation and causes DoS.

Shakeel

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Kirill Tkhai]

On 03.12.2019 03:13, Shakeel Butt wrote:
> On Mon, Dec 2, 2019 at 8:37 AM Andrey Ryabinin <aryabinin@virtuozzo.com> wrote:
>>
>>
>> On 11/30/19 12:45 AM, Pavel Tikhomirov wrote:
>>> We have a problem that shrinker_rwsem can be held for a long time for
>>> read in shrink_slab, at the same time any process which is trying to
>>> manage shrinkers hangs.
>>>
>>> The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
>>> It tries to shrink something on nfs (hard) but nfs server is dead at
>>> these moment already and rpc will never succeed. Generally any shrinker
>>> can take significant time to do_shrink_slab, so it's a bad idea to hold
>>> the list lock here.
>>>
>>> We have a similar problem in shrink_slab_memcg, except that we are
>>> traversing shrinker_map+shrinker_idr there.
>>>
>>> The idea of the patch is to inc a refcount to the chosen shrinker so it
>>> won't disappear and release shrinker_rwsem while we are in
>>> do_shrink_slab, after that we will reacquire shrinker_rwsem, dec
>>> the refcount and continue the traversal.
>>>
>>> We also need a wait_queue so that unregister_shrinker can wait for the
>>> refcnt to become zero. Only after these we can safely remove the
>>> shrinker from list and idr, and free the shrinker.
>>>
>>> I've reproduced the nfs hang in do_shrink_slab with the patch applied on
>>> ms kernel, all other mount/unmount pass fine without any hang.
>>>
>>> Here is a reproduction on kernel without patch:
>>>
>>> 1) Setup nfs on server node with some files in it (e.g. 200)
>>>
>>> [server]# cat /etc/exports
>>> /vz/nfs2 *(ro,no_root_squash,no_subtree_check,async)
>>>
>>> 2) Hard mount it on client node
>>>
>>> [client]# mount -ohard 10.94.3.40:/vz/nfs2 /mnt
>>>
>>> 3) Open some (e.g. 200) files on the mount
>>>
>>> [client]# for i in $(find /mnt/ -type f | head -n 200); \
>>>   do setsid sleep 1000 &>/dev/null <$i & done
>>>
>>> 4) Kill all openers
>>>
>>> [client]# killall sleep -9
>>>
>>> 5) Put your network cable out on client node
>>>
>>> 6) Drop caches on the client, it will hang on nfs while holding
>>>   shrinker_rwsem lock for read
>>>
>>> [client]# echo 3 > /proc/sys/vm/drop_caches
>>>
>>>   crash> bt ...
>>>   PID: 18739  TASK: ...  CPU: 3   COMMAND: "bash"
>>>    #0 [...] __schedule at ...
>>>    #1 [...] schedule at ...
>>>    #2 [...] rpc_wait_bit_killable at ... [sunrpc]
>>>    #3 [...] __wait_on_bit at ...
>>>    #4 [...] out_of_line_wait_on_bit at ...
>>>    #5 [...] _nfs4_proc_delegreturn at ... [nfsv4]
>>>    #6 [...] nfs4_proc_delegreturn at ... [nfsv4]
>>>    #7 [...] nfs_do_return_delegation at ... [nfsv4]
>>>    #8 [...] nfs4_evict_inode at ... [nfsv4]
>>>    #9 [...] evict at ...
>>>   #10 [...] dispose_list at ...
>>>   #11 [...] prune_icache_sb at ...
>>>   #12 [...] super_cache_scan at ...
>>>   #13 [...] do_shrink_slab at ...
>>>   #14 [...] shrink_slab at ...
>>>   #15 [...] drop_slab_node at ...
>>>   #16 [...] drop_slab at ...
>>>   #17 [...] drop_caches_sysctl_handler at ...
>>>   #18 [...] proc_sys_call_handler at ...
>>>   #19 [...] vfs_write at ...
>>>   #20 [...] ksys_write at ...
>>>   #21 [...] do_syscall_64 at ...
>>>   #22 [...] entry_SYSCALL_64_after_hwframe at ...
>>>
>>> 7) All other mount/umount activity now hangs with no luck to take
>>>   shrinker_rwsem for write.
>>>
>>> [client]# mount -t tmpfs tmpfs /tmp
>>>
>>>   crash> bt ...
>>>   PID: 5464   TASK: ...  CPU: 3   COMMAND: "mount"
>>>    #0 [...] __schedule at ...
>>>    #1 [...] schedule at ...
>>>    #2 [...] rwsem_down_write_slowpath at ...
>>>    #3 [...] prealloc_shrinker at ...
>>>    #4 [...] alloc_super at ...
>>>    #5 [...] sget at ...
>>>    #6 [...] mount_nodev at ...
>>>    #7 [...] legacy_get_tree at ...
>>>    #8 [...] vfs_get_tree at ...
>>>    #9 [...] do_mount at ...
>>>   #10 [...] ksys_mount at ...
>>>   #11 [...] __x64_sys_mount at ...
>>>   #12 [...] do_syscall_64 at ...
>>>   #13 [...] entry_SYSCALL_64_after_hwframe at ...
>>>
>>
>>
>> I don't think this patch solves the problem, it only fixes one minor symptom of it.
>> The actual problem here the reclaim hang in the nfs.
>> It means that any process, including kswapd, may go into nfs inode reclaim and stuck there.
>>
>> Even mount() itself has GFP_KERNEL allocations in its path, so it still might stuck there even with your patch.
>>
>> I think this should be handled on nfs/vfs level by making  inode eviction during reclaim more asynchronous.
> 
> Though I agree that we should be fixing shrinkers to not get stuck
> (and be more async), I still think the problem this patch is solving
> is worth fixing. On machines running multiple workloads, one job stuck
> in slab shrinker and blocking all other unrelated jobs wanting
> shrinker_rwsem, breaks the isolation and causes DoS.

This makes jobs isolated till global reclaim occurs. In that case any process
(including kswapd) become stuck, and nothing will help. But (as it was spoken
in this list some time ago) many people configure their workload in the way
they don't have global reclaim, and they may get some profit from this.

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Dave Chinner]

[please cc me on future shrinker infrastructure modifications]

On Mon, Dec 02, 2019 at 07:36:03PM +0300, Andrey Ryabinin wrote:
> 
> On 11/30/19 12:45 AM, Pavel Tikhomirov wrote:
> > We have a problem that shrinker_rwsem can be held for a long time for
> > read in shrink_slab, at the same time any process which is trying to
> > manage shrinkers hangs.
> > 
> > The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
> > It tries to shrink something on nfs (hard) but nfs server is dead at
> > these moment already and rpc will never succeed. Generally any shrinker
> > can take significant time to do_shrink_slab, so it's a bad idea to hold
> > the list lock here.

registering/unregistering a shrinker is not a performance critical
task. If a shrinker is blocking for a long time, then we need to
work to fix the shrinker implementation because blocking is a much
bigger problem than just register/unregister.

> > The idea of the patch is to inc a refcount to the chosen shrinker so it
> > won't disappear and release shrinker_rwsem while we are in
> > do_shrink_slab, after that we will reacquire shrinker_rwsem, dec
> > the refcount and continue the traversal.

This is going to cause a *lot* of traffic on the shrinker rwsem.
It's already a pretty hot lock on large machines under memory
pressure (think thousands of tasks all doing direct reclaim across
hundreds of CPUs), and so changing them to cycle the rwsem on every
shrinker that will only make this worse. Esepcially when we consider
that there may be hundreds to thousands of registered shrinker
instances on large machines.

As an example of how frequent cycling of a global lock in shrinker
instances causes issues, we used to take references to superblock
shrinker count invocations to guarantee existence. This was found to
be a scalability limitation when lots of near-empty superblocks were
present in a system (see commit d23da150a37c ("fs/superblock: avoid
locking counting inodes and dentries before reclaiming them")).

This alleviated the problem for a while, but soon we had problems
with just taking a reference to the superblock in the callbacks that
did actual work. Hence we changed it to just take a per-superblock
rwsem to get rid of the global sb_lock spinlock in this path. See
commit eb6ef3df4faa ("trylock_super(): replacement for
grab_super_passive()". Now we don't have a scalability problem.

IOWs, we already know that cycling a global rwsem on every
individual shrinker invocation is going to cause noticable
scalability problems. Hence I don't think that this sort of "cycle
the global rwsem faster to reduce [un]register latency" solution is
going to fly because of the runtime performance regressions it will
introduce....

> I don't think this patch solves the problem, it only fixes one minor symptom of it.
> The actual problem here the reclaim hang in the nfs.

The nfs client is waiting on the NFS server to respond. It may
actually be that the server has hung, not the client...

> It means that any process, including kswapd, may go into nfs inode reclaim and stuck there.

*nod*

> I think this should be handled on nfs/vfs level by making  inode eviction during reclaim more asynchronous.

That's what we are trying to do with similar blocking based issues
in XFS inode reclaim. It's not simple, though, because these days
memory reclaim is like a bowl full of spaghetti covered with a
delicious sauce of non-obvious heuristics and broken
functionality....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Michal Hocko]

On Fri 06-12-19 13:09:53, Dave Chinner wrote:
> [please cc me on future shrinker infrastructure modifications]
> 
> On Mon, Dec 02, 2019 at 07:36:03PM +0300, Andrey Ryabinin wrote:
> > 
> > On 11/30/19 12:45 AM, Pavel Tikhomirov wrote:
> > > We have a problem that shrinker_rwsem can be held for a long time for
> > > read in shrink_slab, at the same time any process which is trying to
> > > manage shrinkers hangs.
> > > 
> > > The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
> > > It tries to shrink something on nfs (hard) but nfs server is dead at
> > > these moment already and rpc will never succeed. Generally any shrinker
> > > can take significant time to do_shrink_slab, so it's a bad idea to hold
> > > the list lock here.
> 
> registering/unregistering a shrinker is not a performance critical
> task.

We have had a bug report from production system which stumbled over a
long [u]nmount which was stuck on a shrinker {un}register path waiting for
the lock. This wouldn't be a big deal most of the time but [u]mount were
part of the login session in this case. This was on an older
distribution kernel and e496612c5130 ("mm: do not stall register_shrinker()")
helped a bit but not really for shrinkers that take a lot of time.

> If a shrinker is blocking for a long time, then we need to
> work to fix the shrinker implementation because blocking is a much
> bigger problem than just register/unregister.

Absolutely agreed here.
-- 
Michal Hocko
SUSE Labs

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Shakeel Butt]

On Thu, Dec 5, 2019 at 6:10 PM Dave Chinner <david@fromorbit.com> wrote:
>
> [please cc me on future shrinker infrastructure modifications]
>
> On Mon, Dec 02, 2019 at 07:36:03PM +0300, Andrey Ryabinin wrote:
> >
> > On 11/30/19 12:45 AM, Pavel Tikhomirov wrote:
> > > We have a problem that shrinker_rwsem can be held for a long time for
> > > read in shrink_slab, at the same time any process which is trying to
> > > manage shrinkers hangs.
> > >
> > > The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
> > > It tries to shrink something on nfs (hard) but nfs server is dead at
> > > these moment already and rpc will never succeed. Generally any shrinker
> > > can take significant time to do_shrink_slab, so it's a bad idea to hold
> > > the list lock here.
>
> registering/unregistering a shrinker is not a performance critical
> task.

Yes, not performance critical but it can cause isolation issues.

> If a shrinker is blocking for a long time, then we need to
> work to fix the shrinker implementation because blocking is a much
> bigger problem than just register/unregister.
>

Yes, we should be fixing the implementations of all shrinkers and yes
it is bigger issue but we can also fix register/unregister isolation
issue in parallel. Fixing all shrinkers would a tedious and long task
and we should not block fixing isolation issue on it.

> > > The idea of the patch is to inc a refcount to the chosen shrinker so it
> > > won't disappear and release shrinker_rwsem while we are in
> > > do_shrink_slab, after that we will reacquire shrinker_rwsem, dec
> > > the refcount and continue the traversal.
>
> This is going to cause a *lot* of traffic on the shrinker rwsem.
> It's already a pretty hot lock on large machines under memory
> pressure (think thousands of tasks all doing direct reclaim across
> hundreds of CPUs), and so changing them to cycle the rwsem on every
> shrinker that will only make this worse. Esepcially when we consider
> that there may be hundreds to thousands of registered shrinker
> instances on large machines.
>
> As an example of how frequent cycling of a global lock in shrinker
> instances causes issues, we used to take references to superblock
> shrinker count invocations to guarantee existence. This was found to
> be a scalability limitation when lots of near-empty superblocks were
> present in a system (see commit d23da150a37c ("fs/superblock: avoid
> locking counting inodes and dentries before reclaiming them")).
>
> This alleviated the problem for a while, but soon we had problems
> with just taking a reference to the superblock in the callbacks that
> did actual work. Hence we changed it to just take a per-superblock
> rwsem to get rid of the global sb_lock spinlock in this path. See
> commit eb6ef3df4faa ("trylock_super(): replacement for
> grab_super_passive()". Now we don't have a scalability problem.
>
> IOWs, we already know that cycling a global rwsem on every
> individual shrinker invocation is going to cause noticable
> scalability problems. Hence I don't think that this sort of "cycle
> the global rwsem faster to reduce [un]register latency" solution is
> going to fly because of the runtime performance regressions it will
> introduce....
>

I agree with your scalability concern (though others would argue to
first demonstrate the issue before adding more sophisticated scalable
code). Most memory reclaim code is written without the performance or
scalability concern, maybe we should switch our thinking.

> > I don't think this patch solves the problem, it only fixes one minor symptom of it.
> > The actual problem here the reclaim hang in the nfs.
>
> The nfs client is waiting on the NFS server to respond. It may
> actually be that the server has hung, not the client...
>
> > It means that any process, including kswapd, may go into nfs inode reclaim and stuck there.
>
> *nod*
>
> > I think this should be handled on nfs/vfs level by making  inode eviction during reclaim more asynchronous.
>
> That's what we are trying to do with similar blocking based issues
> in XFS inode reclaim. It's not simple, though, because these days
> memory reclaim is like a bowl full of spaghetti covered with a
> delicious sauce of non-obvious heuristics and broken
> functionality....
>
> Cheers,
>
> Dave.
> --
> Dave Chinner
> david@fromorbit.com

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Dave Chinner]

On Fri, Dec 06, 2019 at 09:11:25AM -0800, Shakeel Butt wrote:
> On Thu, Dec 5, 2019 at 6:10 PM Dave Chinner <david@fromorbit.com> wrote:
> > If a shrinker is blocking for a long time, then we need to
> > work to fix the shrinker implementation because blocking is a much
> > bigger problem than just register/unregister.
> >
> 
> Yes, we should be fixing the implementations of all shrinkers and yes
> it is bigger issue but we can also fix register/unregister isolation
> issue in parallel. Fixing all shrinkers would a tedious and long task
> and we should not block fixing isolation issue on it.

"fixing all shrinkers" is a bit of hyperbole - you've identified
only one instance where blocking is causing you problems. Indeed,
most shrinkers are already non-blocking and won't cause you any
problems at all.

> > IOWs, we already know that cycling a global rwsem on every
> > individual shrinker invocation is going to cause noticable
> > scalability problems. Hence I don't think that this sort of "cycle
> > the global rwsem faster to reduce [un]register latency" solution is
> > going to fly because of the runtime performance regressions it will
> > introduce....
> >
> 
> I agree with your scalability concern (though others would argue to
> first demonstrate the issue before adding more sophisticated scalable
> code).

Look at the git history. We *know* this is a problem, so anyone
arguing that we have to prove it can go take a long walk of a short
plank....

> Most memory reclaim code is written without the performance or
> scalability concern, maybe we should switch our thinking.

I think there's a lot of core mm and other developers that would
disagree with you there. With respect to shrinkers, we've been
directly concerned about performance and scalability of the
individual instances as well as the infrastructure for at least the
last decade....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Pavel Tikhomirov]

On 12/4/19 11:35 AM, Michal Hocko wrote:
> On Sat 30-11-19 00:45:41, Pavel Tikhomirov wrote:
>> We have a problem that shrinker_rwsem can be held for a long time for
>> read in shrink_slab, at the same time any process which is trying to
>> manage shrinkers hangs.
>>
>> The shrinker_rwsem is taken in shrink_slab while traversing shrinker_list.
>> It tries to shrink something on nfs (hard) but nfs server is dead at
>> these moment already and rpc will never succeed. Generally any shrinker
>> can take significant time to do_shrink_slab, so it's a bad idea to hold
>> the list lock here.
> 
> Yes, this is a known problem and people have already tried to address it
> in the past. Have you checked previous attempts? SRCU based one
> http://lkml.kernel.org/r/153365347929.19074.12509495712735843805.stgit@localhost.localdomain
> but I believe there were others (I only had this one in my notes).
> Please make sure to Cc Dave Chinner when posting a next version because
> he had some concerns about the change of the behavior.

The approach of the patch you are referencing is quiet different, it 
will still hold global srcu_read_lock(&srcu) when diving in 
do_shrink_slab and hanging nfs will still block all [un]register_shrinker.

> 
>> We have a similar problem in shrink_slab_memcg, except that we are
>> traversing shrinker_map+shrinker_idr there.
>>
>> The idea of the patch is to inc a refcount to the chosen shrinker so it
>> won't disappear and release shrinker_rwsem while we are in
>> do_shrink_slab, after that we will reacquire shrinker_rwsem, dec
>> the refcount and continue the traversal.
> 
> The reference count part makes sense to me. RCU role needs a better
> explanation.

I have 2 rcu's in patch, 1-st is to protect shrinker_map same as it was 
before in memcg_set_shrinker_bit, 2-nd is to protect shrinker struct in 
put_shrinker from being freed, as unregister_shrinker can see refcnt==0 
without actually going to schedule().

> Also do you have any reason to not use completion for
> the final step? Openconding essentially the same concept sounds a bit
> awkward to me.

Thanks for a good hint, from the first glance we can rework wait_event 
part to wait_for_completion.

> 
>> We also need a wait_queue so that unregister_shrinker can wait for the
>> refcnt to become zero. Only after these we can safely remove the
>> shrinker from list and idr, and free the shrinker.
> [...]
>>    crash> bt ...
>>    PID: 18739  TASK: ...  CPU: 3   COMMAND: "bash"
>>     #0 [...] __schedule at ...
>>     #1 [...] schedule at ...
>>     #2 [...] rpc_wait_bit_killable at ... [sunrpc]
>>     #3 [...] __wait_on_bit at ...
>>     #4 [...] out_of_line_wait_on_bit at ...
>>     #5 [...] _nfs4_proc_delegreturn at ... [nfsv4]
>>     #6 [...] nfs4_proc_delegreturn at ... [nfsv4]
>>     #7 [...] nfs_do_return_delegation at ... [nfsv4]
>>     #8 [...] nfs4_evict_inode at ... [nfsv4]
>>     #9 [...] evict at ...
>>    #10 [...] dispose_list at ...
>>    #11 [...] prune_icache_sb at ...
>>    #12 [...] super_cache_scan at ...
>>    #13 [...] do_shrink_slab at ...
> 
> Are NFS people aware of this? Because this is simply not acceptable
> behavior. Memory reclaim cannot be block indefinitely or for a long
> time. There must be a way to simply give up if the underlying inode
> cannot be reclaimed.

Sorry that I didn't cc nfs people from the begining.

> 
> I still have to think about the proposed solution. It sounds a bit over
> complicated to me.
> 

-- 
Best regards, Tikhomirov Pavel
Software Developer, Virtuozzo.

Re: [PATCH] mm: fix hanging shrinker management on long do_shrink_slab

[Pavel Tikhomirov]

On 12/10/19 4:20 AM, Dave Chinner wrote:
> On Fri, Dec 06, 2019 at 09:11:25AM -0800, Shakeel Butt wrote:
>> On Thu, Dec 5, 2019 at 6:10 PM Dave Chinner <david@fromorbit.com> wrote:
>>> If a shrinker is blocking for a long time, then we need to
>>> work to fix the shrinker implementation because blocking is a much
>>> bigger problem than just register/unregister.
>>>
>>
>> Yes, we should be fixing the implementations of all shrinkers and yes
>> it is bigger issue but we can also fix register/unregister isolation
>> issue in parallel. Fixing all shrinkers would a tedious and long task
>> and we should not block fixing isolation issue on it.
> 
> "fixing all shrinkers" is a bit of hyperbole - you've identified
> only one instance where blocking is causing you problems. Indeed,
> most shrinkers are already non-blocking and won't cause you any
> problems at all.
> 
>>> IOWs, we already know that cycling a global rwsem on every
>>> individual shrinker invocation is going to cause noticable
>>> scalability problems. Hence I don't think that this sort of "cycle
>>> the global rwsem faster to reduce [un]register latency" solution is
>>> going to fly because of the runtime performance regressions it will
>>> introduce....
>>>
>>
>> I agree with your scalability concern (though others would argue to
>> first demonstrate the issue before adding more sophisticated scalable
>> code).
> 
> Look at the git history. We *know* this is a problem, so anyone
> arguing that we have to prove it can go take a long walk of a short
> plank....
> 
>> Most memory reclaim code is written without the performance or
>> scalability concern, maybe we should switch our thinking.
> 
> I think there's a lot of core mm and other developers that would
> disagree with you there. With respect to shrinkers, we've been
> directly concerned about performance and scalability of the
> individual instances as well as the infrastructure for at least the
> last decade....
> 
> Cheers,
> 
> Dave.
> 

Thanks a lot for your replies, now I see that the core of the problem is 
in nfs hang, before that I was unsure if it's OK to have such a hang in 
do_shrink_slab.

I have a possibly bad idea on how my patch can still work. What if we 
use unlock/refcount way only for nfs-shrinkers? It will still give a 
performance penalty if one has many nfs mounts, but for those who has 
little number of nfs mounts the penalty would be less. And this would be 
a small isolation improvement for nfs users.

-- 
Best regards, Tikhomirov Pavel
Software Developer, Virtuozzo.