Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1735197 > unrolled thread

[PATCH 0/6] More graceful flusher thread memory reclaim wakeup

Back to article view | Back to linux.kernel

Contents

  [PATCH 0/6] More graceful flusher thread memory reclaim wakeup Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:00 +0200
    [PATCH 3/6] page-writeback: pass in '0' for nr_pages writeback in laptop mode Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:00 +0200
      Re: [PATCH 3/6] page-writeback: pass in '0' for nr_pages writeback  in laptop mode Johannes Weiner <hannes@cmpxchg.org> - 2017-09-19 22:10 +0200
      Re: [PATCH 3/6] page-writeback: pass in '0' for nr_pages writeback  in laptop mode Jan Kara <jack@suse.cz> - 2017-09-20 16:40 +0200
        Re: [PATCH 3/6] page-writeback: pass in '0' for nr_pages writeback in  laptop mode Jens Axboe <axboe@kernel.dk> - 2017-09-20 17:30 +0200
    [PATCH 2/6] fs-writeback: provide a wakeup_flusher_threads_bdi() Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:00 +0200
      Re: [PATCH 2/6] fs-writeback: provide a wakeup_flusher_threads_bdi() Johannes Weiner <hannes@cmpxchg.org> - 2017-09-19 22:10 +0200
      Re: [PATCH 2/6] fs-writeback: provide a wakeup_flusher_threads_bdi() Jan Kara <jack@suse.cz> - 2017-09-20 16:30 +0200
    [PATCH 5/6] fs-writeback: move nr_pages == 0 logic to one location Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:00 +0200
      Re: [PATCH 5/6] fs-writeback: move nr_pages == 0 logic to one  location Johannes Weiner <hannes@cmpxchg.org> - 2017-09-19 22:10 +0200
      Re: [PATCH 5/6] fs-writeback: move nr_pages == 0 logic to one  location Jan Kara <jack@suse.cz> - 2017-09-20 16:50 +0200
        Re: [PATCH 5/6] fs-writeback: move nr_pages == 0 logic to one  location Jens Axboe <axboe@kernel.dk> - 2017-09-20 17:10 +0200
          Re: [PATCH 5/6] fs-writeback: move nr_pages == 0 logic to one  location Jan Kara <jack@suse.cz> - 2017-09-20 17:40 +0200
            Re: [PATCH 5/6] fs-writeback: move nr_pages == 0 logic to one  location Jens Axboe <axboe@kernel.dk> - 2017-09-20 17:50 +0200
    [PATCH 1/6] buffer: cleanup free_more_memory() flusher wakeup Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:00 +0200
      Re: [PATCH 1/6] buffer: cleanup free_more_memory() flusher wakeup Johannes Weiner <hannes@cmpxchg.org> - 2017-09-19 22:10 +0200
      Re: [PATCH 1/6] buffer: cleanup free_more_memory() flusher wakeup Jan Kara <jack@suse.cz> - 2017-09-20 16:20 +0200
        Re: [PATCH 1/6] buffer: cleanup free_more_memory() flusher wakeup Jens Axboe <axboe@kernel.dk> - 2017-09-20 17:20 +0200
    [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:00 +0200
      Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Johannes Weiner <hannes@cmpxchg.org> - 2017-09-19 22:20 +0200
        Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Jens Axboe <axboe@kernel.dk> - 2017-09-19 22:40 +0200
      Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Jens Axboe <axboe@kernel.dk> - 2017-09-20 04:00 +0200
      Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Amir Goldstein <amir73il@gmail.com> - 2017-09-20 05:20 +0200
        Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Jens Axboe <axboe@kernel.dk> - 2017-09-20 06:20 +0200
          Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Amir Goldstein <amir73il@gmail.com> - 2017-09-20 08:10 +0200
            Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Jens Axboe <axboe@kernel.dk> - 2017-09-20 14:40 +0200
          Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending  !nr_pages flush Jan Kara <jack@suse.cz> - 2017-09-20 16:50 +0200
    Re: [PATCH 0/6] More graceful flusher thread memory reclaim wakeup John Stoffel <john@quad.stoffel.home> - 2017-09-20 21:40 +0200
      Re: [PATCH 0/6] More graceful flusher thread memory reclaim wakeup Jens Axboe <axboe@kernel.dk> - 2017-09-20 21:40 +0200
        Re: [PATCH 0/6] More graceful flusher thread memory reclaim wakeup Johannes Weiner <hannes@cmpxchg.org> - 2017-09-21 01:20 +0200

Page 2 of 2 — ← Prev page 1 [2]

#1735243 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On 09/19/2017 02:18 PM, Johannes Weiner wrote:
> On Tue, Sep 19, 2017 at 01:53:07PM -0600, Jens Axboe wrote:
>> A few callers pass in nr_pages == 0 when they wakeup the flusher
>> threads, which means that the flusher should just flush everything
>> that was currently dirty. If we are tight on memory, we can get
>> tons of these queued from kswapd/vmscan. This causes (at least)
>> two problems:
>>
>> 1) We consume a ton of memory just allocating writeback work items.
>> 2) We spend so much time processing these work items, that we
>>    introduce a softlockup in writeback processing.
>>
>> Fix this by adding a 'zero_pages' bit to the writeback structure,
>> and set that when someone queues a nr_pages==0 flusher thread
>> wakeup. The bit is cleared when we start writeback on that work
>> item. If the bit is already set when we attempt to queue !nr_pages
>> writeback, then we simply ignore it.
>>
>> This provides us one of full flush in flight, with one pending as
>> well, and makes for more efficient handling of this type of
>> writeback.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> Acked-by: Johannes Weiner <hannes@cmpxchg.org>
> 
> Just a nitpick:
> 
>> @@ -948,15 +949,25 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>>  			       bool range_cyclic, enum wb_reason reason)
>>  {
>>  	struct wb_writeback_work *work;
>> +	bool zero_pages = false;
>>  
>>  	if (!wb_has_dirty_io(wb))
>>  		return;
>>  
>>  	/*
>> -	 * If someone asked for zero pages, we write out the WORLD
>> +	 * If someone asked for zero pages, we write out the WORLD.
>> +	 * Places like vmscan and laptop mode want to queue a wakeup to
>> +	 * the flusher threads to clean out everything. To avoid potentially
>> +	 * having tons of these pending, ensure that we only allow one of
>> +	 * them pending and inflight at the time
>>  	 */
>> -	if (!nr_pages)
>> +	if (!nr_pages) {
>> +		if (test_bit(WB_zero_pages, &wb->state))
>> +			return;
>> +		set_bit(WB_zero_pages, &wb->state);
>>  		nr_pages = get_nr_dirty_pages();
> 
> We could rely on the work->older_than_this and pass LONG_MAX here
> instead to write out the world as it was at the time wb commences.
> 
> get_nr_dirty_pages() is somewhat clearer on intent, but on the other
> hand it returns global state and is used here in a split-bdi context,
> and we can end up in sum requesting the system-wide dirty pages
> several times over. It'll work fine, relying on work->older_than_this
> to contain it also, it just seems a little ugly and subtle.

Not disagreeing with that at all. I just carried the !nr_pages forward
as the way to do this. I think any further cleanup or work should just
be based on this patchset, I'd definitely welcome a change in that
direction.

Thanks for your reviews!

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1735398 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On 09/19/2017 01:53 PM, Jens Axboe wrote:
> @@ -948,15 +949,25 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>  			       bool range_cyclic, enum wb_reason reason)
>  {
>  	struct wb_writeback_work *work;
> +	bool zero_pages = false;
>  
>  	if (!wb_has_dirty_io(wb))
>  		return;
>  
>  	/*
> -	 * If someone asked for zero pages, we write out the WORLD
> +	 * If someone asked for zero pages, we write out the WORLD.
> +	 * Places like vmscan and laptop mode want to queue a wakeup to
> +	 * the flusher threads to clean out everything. To avoid potentially
> +	 * having tons of these pending, ensure that we only allow one of
> +	 * them pending and inflight at the time
>  	 */
> -	if (!nr_pages)
> +	if (!nr_pages) {
> +		if (test_bit(WB_zero_pages, &wb->state))
> +			return;
> +		set_bit(WB_zero_pages, &wb->state);
>  		nr_pages = get_nr_dirty_pages();
> +		zero_pages = true;
> +	}

Later fix added here to ensure we clear WB_zero_pages, if work
allocation fails:

work = kzalloc(sizeof(*work),                                           
                GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);           
if (!work) {                                                            
        if (zero_pages)                                                 
                clear_bit(WB_zero_pages, &wb->state);
	[...]

Updated patch here:

http://git.kernel.dk/cgit/linux-block/commit/?h=writeback-fixup&id=21ea70657894fda9fccf257543cbec112b2813ef

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1735414 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On Tue, Sep 19, 2017 at 10:53 PM, Jens Axboe <axboe@kernel.dk> wrote:
> A few callers pass in nr_pages == 0 when they wakeup the flusher
> threads, which means that the flusher should just flush everything
> that was currently dirty. If we are tight on memory, we can get
> tons of these queued from kswapd/vmscan. This causes (at least)
> two problems:
>
> 1) We consume a ton of memory just allocating writeback work items.
> 2) We spend so much time processing these work items, that we
>    introduce a softlockup in writeback processing.
>
> Fix this by adding a 'zero_pages' bit to the writeback structure,
> and set that when someone queues a nr_pages==0 flusher thread
> wakeup. The bit is cleared when we start writeback on that work
> item. If the bit is already set when we attempt to queue !nr_pages
> writeback, then we simply ignore it.
>
> This provides us one of full flush in flight, with one pending as
> well, and makes for more efficient handling of this type of
> writeback.
>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  fs/fs-writeback.c                | 30 ++++++++++++++++++++++++++++--
>  include/linux/backing-dev-defs.h |  1 +
>  2 files changed, 29 insertions(+), 2 deletions(-)
>
> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
> index a9a86644cb9f..e0240110b36f 100644
> --- a/fs/fs-writeback.c
> +++ b/fs/fs-writeback.c
> @@ -53,6 +53,7 @@ struct wb_writeback_work {
>         unsigned int for_background:1;
>         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
>         unsigned int auto_free:1;       /* free on completion */
> +       unsigned int zero_pages:1;      /* nr_pages == 0 writeback */

Suggest: use a name that describes the intention (e.g. WB_everything)

>         enum wb_reason reason;          /* why was writeback initiated? */
>
>         struct list_head list;          /* pending work list */
> @@ -948,15 +949,25 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>                                bool range_cyclic, enum wb_reason reason)
>  {
>         struct wb_writeback_work *work;
> +       bool zero_pages = false;
>
>         if (!wb_has_dirty_io(wb))
>                 return;
>
>         /*
> -        * If someone asked for zero pages, we write out the WORLD
> +        * If someone asked for zero pages, we write out the WORLD.
> +        * Places like vmscan and laptop mode want to queue a wakeup to
> +        * the flusher threads to clean out everything. To avoid potentially
> +        * having tons of these pending, ensure that we only allow one of
> +        * them pending and inflight at the time
>          */
> -       if (!nr_pages)
> +       if (!nr_pages) {
> +               if (test_bit(WB_zero_pages, &wb->state))
> +                       return;
> +               set_bit(WB_zero_pages, &wb->state);

Shouldn't this be test_and_set? not the worst outcome if you have more
than one pending work item, but still.

>                 nr_pages = get_nr_dirty_pages();
> +               zero_pages = true;
> +       }
>
>         /*
>          * This is WB_SYNC_NONE writeback, so if allocation fails just
> @@ -975,6 +986,7 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>         work->range_cyclic = range_cyclic;
>         work->reason    = reason;
>         work->auto_free = 1;
> +       work->zero_pages = zero_pages;
>
>         wb_queue_work(wb, work);
>  }
> @@ -1828,6 +1840,14 @@ static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
>                 list_del_init(&work->list);
>         }
>         spin_unlock_bh(&wb->work_lock);
> +
> +       /*
> +        * Once we start processing a work item that had !nr_pages,
> +        * clear the wb state bit for that so we can allow more.
> +        */
> +       if (work && work->zero_pages && test_bit(WB_zero_pages, &wb->state))
> +               clear_bit(WB_zero_pages, &wb->state);

nit: should not need to test_bit

> +
>         return work;
>  }
>
> @@ -1896,6 +1916,12 @@ static long wb_do_writeback(struct bdi_writeback *wb)
>                 trace_writeback_exec(wb, work);
>                 wrote += wb_writeback(wb, work);
>                 finish_writeback_work(wb, work);
> +
> +               /*
> +                * If we have a lot of pending work, make sure we take
> +                * an occasional breather, if needed.
> +                */
> +               cond_resched();

Probably ought to be in a separate patch.

>         }
>
>         /*
> diff --git a/include/linux/backing-dev-defs.h b/include/linux/backing-dev-defs.h
> index 866c433e7d32..7494f6a75458 100644
> --- a/include/linux/backing-dev-defs.h
> +++ b/include/linux/backing-dev-defs.h
> @@ -24,6 +24,7 @@ enum wb_state {
>         WB_shutting_down,       /* wb_shutdown() in progress */
>         WB_writeback_running,   /* Writeback is in progress */
>         WB_has_dirty_io,        /* Dirty inodes on ->b_{dirty|io|more_io} */
> +       WB_zero_pages,          /* nr_pages == 0 flush pending */

same suggestion: WB_everything

Cheers,
Amir.

[toc] | [prev] | [next] | [standalone]

#1735433 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On 09/19/2017 09:10 PM, Amir Goldstein wrote:
> On Tue, Sep 19, 2017 at 10:53 PM, Jens Axboe <axboe@kernel.dk> wrote:
>> A few callers pass in nr_pages == 0 when they wakeup the flusher
>> threads, which means that the flusher should just flush everything
>> that was currently dirty. If we are tight on memory, we can get
>> tons of these queued from kswapd/vmscan. This causes (at least)
>> two problems:
>>
>> 1) We consume a ton of memory just allocating writeback work items.
>> 2) We spend so much time processing these work items, that we
>>    introduce a softlockup in writeback processing.
>>
>> Fix this by adding a 'zero_pages' bit to the writeback structure,
>> and set that when someone queues a nr_pages==0 flusher thread
>> wakeup. The bit is cleared when we start writeback on that work
>> item. If the bit is already set when we attempt to queue !nr_pages
>> writeback, then we simply ignore it.
>>
>> This provides us one of full flush in flight, with one pending as
>> well, and makes for more efficient handling of this type of
>> writeback.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  fs/fs-writeback.c                | 30 ++++++++++++++++++++++++++++--
>>  include/linux/backing-dev-defs.h |  1 +
>>  2 files changed, 29 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
>> index a9a86644cb9f..e0240110b36f 100644
>> --- a/fs/fs-writeback.c
>> +++ b/fs/fs-writeback.c
>> @@ -53,6 +53,7 @@ struct wb_writeback_work {
>>         unsigned int for_background:1;
>>         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
>>         unsigned int auto_free:1;       /* free on completion */
>> +       unsigned int zero_pages:1;      /* nr_pages == 0 writeback */
> 
> Suggest: use a name that describes the intention (e.g. WB_everything)

Agree, the name isn't the best. WB_everything isn't great either, though,
since this isn't an integrity write. WB_start_all would be better,
I'll make that change.

>>         enum wb_reason reason;          /* why was writeback initiated? */
>>
>>         struct list_head list;          /* pending work list */
>> @@ -948,15 +949,25 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>>                                bool range_cyclic, enum wb_reason reason)
>>  {
>>         struct wb_writeback_work *work;
>> +       bool zero_pages = false;
>>
>>         if (!wb_has_dirty_io(wb))
>>                 return;
>>
>>         /*
>> -        * If someone asked for zero pages, we write out the WORLD
>> +        * If someone asked for zero pages, we write out the WORLD.
>> +        * Places like vmscan and laptop mode want to queue a wakeup to
>> +        * the flusher threads to clean out everything. To avoid potentially
>> +        * having tons of these pending, ensure that we only allow one of
>> +        * them pending and inflight at the time
>>          */
>> -       if (!nr_pages)
>> +       if (!nr_pages) {
>> +               if (test_bit(WB_zero_pages, &wb->state))
>> +                       return;
>> +               set_bit(WB_zero_pages, &wb->state);
> 
> Shouldn't this be test_and_set? not the worst outcome if you have more
> than one pending work item, but still.

If the frequency of these is high, and they were to trigger the bad
conditions we saw, then a split test + set is faster as it won't
keep re-dirtying the same cacheline from multiple locations. It's
better to leave it a little racy, but faster.

>> @@ -1828,6 +1840,14 @@ static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
>>                 list_del_init(&work->list);
>>         }
>>         spin_unlock_bh(&wb->work_lock);
>> +
>> +       /*
>> +        * Once we start processing a work item that had !nr_pages,
>> +        * clear the wb state bit for that so we can allow more.
>> +        */
>> +       if (work && work->zero_pages && test_bit(WB_zero_pages, &wb->state))
>> +               clear_bit(WB_zero_pages, &wb->state);
> 
> nit: should not need to test_bit

True, we can drop it for this case, as it'll be the common condition
anyway. I'll make that change.

>> @@ -1896,6 +1916,12 @@ static long wb_do_writeback(struct bdi_writeback *wb)
>>                 trace_writeback_exec(wb, work);
>>                 wrote += wb_writeback(wb, work);
>>                 finish_writeback_work(wb, work);
>> +
>> +               /*
>> +                * If we have a lot of pending work, make sure we take
>> +                * an occasional breather, if needed.
>> +                */
>> +               cond_resched();
> 
> Probably ought to be in a separate patch.

Yeah, it probably should be. It's not strictly needed with the other
change anyway, I will just drop it.

New version:

http://git.kernel.dk/cgit/linux-block/commit/?h=writeback-fixup&id=338a69c217cdaaffda93f3cc9a364a347f782adb

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1735542 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On Wed, Sep 20, 2017 at 7:13 AM, Jens Axboe <axboe@kernel.dk> wrote:
> On 09/19/2017 09:10 PM, Amir Goldstein wrote:
>> On Tue, Sep 19, 2017 at 10:53 PM, Jens Axboe <axboe@kernel.dk> wrote:
>>> A few callers pass in nr_pages == 0 when they wakeup the flusher
>>> threads, which means that the flusher should just flush everything
>>> that was currently dirty. If we are tight on memory, we can get
>>> tons of these queued from kswapd/vmscan. This causes (at least)
>>> two problems:
>>>
>>> 1) We consume a ton of memory just allocating writeback work items.
>>> 2) We spend so much time processing these work items, that we
>>>    introduce a softlockup in writeback processing.
>>>
>>> Fix this by adding a 'zero_pages' bit to the writeback structure,
>>> and set that when someone queues a nr_pages==0 flusher thread
>>> wakeup. The bit is cleared when we start writeback on that work
>>> item. If the bit is already set when we attempt to queue !nr_pages
>>> writeback, then we simply ignore it.
>>>
>>> This provides us one of full flush in flight, with one pending as
>>> well, and makes for more efficient handling of this type of
>>> writeback.
>>>
>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>> ---
>>>  fs/fs-writeback.c                | 30 ++++++++++++++++++++++++++++--
>>>  include/linux/backing-dev-defs.h |  1 +
>>>  2 files changed, 29 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
>>> index a9a86644cb9f..e0240110b36f 100644
>>> --- a/fs/fs-writeback.c
>>> +++ b/fs/fs-writeback.c
>>> @@ -53,6 +53,7 @@ struct wb_writeback_work {
>>>         unsigned int for_background:1;
>>>         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
>>>         unsigned int auto_free:1;       /* free on completion */
>>> +       unsigned int zero_pages:1;      /* nr_pages == 0 writeback */
>>
>> Suggest: use a name that describes the intention (e.g. WB_everything)
>
> Agree, the name isn't the best. WB_everything isn't great either, though,
> since this isn't an integrity write. WB_start_all would be better,
> I'll make that change.
>
>>>         enum wb_reason reason;          /* why was writeback initiated? */
>>>
>>>         struct list_head list;          /* pending work list */
>>> @@ -948,15 +949,25 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>>>                                bool range_cyclic, enum wb_reason reason)
>>>  {
>>>         struct wb_writeback_work *work;
>>> +       bool zero_pages = false;
>>>
>>>         if (!wb_has_dirty_io(wb))
>>>                 return;
>>>
>>>         /*
>>> -        * If someone asked for zero pages, we write out the WORLD
>>> +        * If someone asked for zero pages, we write out the WORLD.
>>> +        * Places like vmscan and laptop mode want to queue a wakeup to
>>> +        * the flusher threads to clean out everything. To avoid potentially
>>> +        * having tons of these pending, ensure that we only allow one of
>>> +        * them pending and inflight at the time
>>>          */
>>> -       if (!nr_pages)
>>> +       if (!nr_pages) {
>>> +               if (test_bit(WB_zero_pages, &wb->state))
>>> +                       return;
>>> +               set_bit(WB_zero_pages, &wb->state);
>>
>> Shouldn't this be test_and_set? not the worst outcome if you have more
>> than one pending work item, but still.
>
> If the frequency of these is high, and they were to trigger the bad
> conditions we saw, then a split test + set is faster as it won't
> keep re-dirtying the same cacheline from multiple locations. It's
> better to leave it a little racy, but faster.
>

Fare enough, but then better change the language of the commit message and
comment above not to claim that there can be only one pending work item.

Amir.

[toc] | [prev] | [next] | [standalone]

#1735749 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On 09/20/2017 12:05 AM, Amir Goldstein wrote:
> On Wed, Sep 20, 2017 at 7:13 AM, Jens Axboe <axboe@kernel.dk> wrote:
>> On 09/19/2017 09:10 PM, Amir Goldstein wrote:
>>> On Tue, Sep 19, 2017 at 10:53 PM, Jens Axboe <axboe@kernel.dk> wrote:
>>>> A few callers pass in nr_pages == 0 when they wakeup the flusher
>>>> threads, which means that the flusher should just flush everything
>>>> that was currently dirty. If we are tight on memory, we can get
>>>> tons of these queued from kswapd/vmscan. This causes (at least)
>>>> two problems:
>>>>
>>>> 1) We consume a ton of memory just allocating writeback work items.
>>>> 2) We spend so much time processing these work items, that we
>>>>    introduce a softlockup in writeback processing.
>>>>
>>>> Fix this by adding a 'zero_pages' bit to the writeback structure,
>>>> and set that when someone queues a nr_pages==0 flusher thread
>>>> wakeup. The bit is cleared when we start writeback on that work
>>>> item. If the bit is already set when we attempt to queue !nr_pages
>>>> writeback, then we simply ignore it.
>>>>
>>>> This provides us one of full flush in flight, with one pending as
>>>> well, and makes for more efficient handling of this type of
>>>> writeback.
>>>>
>>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>> ---
>>>>  fs/fs-writeback.c                | 30 ++++++++++++++++++++++++++++--
>>>>  include/linux/backing-dev-defs.h |  1 +
>>>>  2 files changed, 29 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
>>>> index a9a86644cb9f..e0240110b36f 100644
>>>> --- a/fs/fs-writeback.c
>>>> +++ b/fs/fs-writeback.c
>>>> @@ -53,6 +53,7 @@ struct wb_writeback_work {
>>>>         unsigned int for_background:1;
>>>>         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
>>>>         unsigned int auto_free:1;       /* free on completion */
>>>> +       unsigned int zero_pages:1;      /* nr_pages == 0 writeback */
>>>
>>> Suggest: use a name that describes the intention (e.g. WB_everything)
>>
>> Agree, the name isn't the best. WB_everything isn't great either, though,
>> since this isn't an integrity write. WB_start_all would be better,
>> I'll make that change.
>>
>>>>         enum wb_reason reason;          /* why was writeback initiated? */
>>>>
>>>>         struct list_head list;          /* pending work list */
>>>> @@ -948,15 +949,25 @@ static void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
>>>>                                bool range_cyclic, enum wb_reason reason)
>>>>  {
>>>>         struct wb_writeback_work *work;
>>>> +       bool zero_pages = false;
>>>>
>>>>         if (!wb_has_dirty_io(wb))
>>>>                 return;
>>>>
>>>>         /*
>>>> -        * If someone asked for zero pages, we write out the WORLD
>>>> +        * If someone asked for zero pages, we write out the WORLD.
>>>> +        * Places like vmscan and laptop mode want to queue a wakeup to
>>>> +        * the flusher threads to clean out everything. To avoid potentially
>>>> +        * having tons of these pending, ensure that we only allow one of
>>>> +        * them pending and inflight at the time
>>>>          */
>>>> -       if (!nr_pages)
>>>> +       if (!nr_pages) {
>>>> +               if (test_bit(WB_zero_pages, &wb->state))
>>>> +                       return;
>>>> +               set_bit(WB_zero_pages, &wb->state);
>>>
>>> Shouldn't this be test_and_set? not the worst outcome if you have more
>>> than one pending work item, but still.
>>
>> If the frequency of these is high, and they were to trigger the bad
>> conditions we saw, then a split test + set is faster as it won't
>> keep re-dirtying the same cacheline from multiple locations. It's
>> better to leave it a little racy, but faster.
>>
> 
> Fare enough, but then better change the language of the commit message and
> comment above not to claim that there can be only one pending work item.

That's unchanged, the commit message should be fine. We clear the
bit when we start the work item, so we can have one in flight and
one pending.

But it does reference 'zero_pages', I'll update that.

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1735804 — Re: [PATCH 6/6] fs-writeback: only allow one inflight and pending !nr_pages flush

On Tue 19-09-17 22:13:25, Jens Axboe wrote:
> On 09/19/2017 09:10 PM, Amir Goldstein wrote:
> New version:
> 
> http://git.kernel.dk/cgit/linux-block/commit/?h=writeback-fixup&id=338a69c217cdaaffda93f3cc9a364a347f782adb

This looks good to me. You can add:

Reviewed-by: Jan Kara <jack@suse.cz>

									Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

[toc] | [prev] | [next] | [standalone]

#1736035

On Tue, Sep 19, 2017 at 01:53:01PM -0600, Jens Axboe wrote:
> We've had some issues with writeback in presence of memory reclaim
> at Facebook, and this patch set attempts to fix it up. The real
> functional change is the last patch in the series, the first 5 are
> prep and cleanup patches.
> 
> The basic idea is that we have callers that call
> wakeup_flusher_threads() with nr_pages == 0. This means 'writeback
> everything'. For memory reclaim situations, we can end up queuing
> a TON of these kinds of writeback units. This can cause softlockups
> and further memory issues, since we allocate huge amounts of
> struct wb_writeback_work to handle this writeback. Handle this
> situation more gracefully.

This looks nice, but do you have any numbers to show how this improves
things?  I read the patches, but I'm not strong enough to comment on
them at all.  But I am interested in how this improves writeback under
pressure, if at all.

John

[toc] | [prev] | [next] | [standalone]

#1736037

On 09/20/2017 01:29 PM, John Stoffel wrote:
> On Tue, Sep 19, 2017 at 01:53:01PM -0600, Jens Axboe wrote:
>> We've had some issues with writeback in presence of memory reclaim
>> at Facebook, and this patch set attempts to fix it up. The real
>> functional change is the last patch in the series, the first 5 are
>> prep and cleanup patches.
>>
>> The basic idea is that we have callers that call
>> wakeup_flusher_threads() with nr_pages == 0. This means 'writeback
>> everything'. For memory reclaim situations, we can end up queuing
>> a TON of these kinds of writeback units. This can cause softlockups
>> and further memory issues, since we allocate huge amounts of
>> struct wb_writeback_work to handle this writeback. Handle this
>> situation more gracefully.
> 
> This looks nice, but do you have any numbers to show how this improves
> things?  I read the patches, but I'm not strong enough to comment on
> them at all.  But I am interested in how this improves writeback under
> pressure, if at all.

Writeback should be about the same, it's mostly about preventing
softlockups and excessive memory usage, under conditions where we are
actively trying to reclaim/clean memory. It was bad enough to cause
softlockups for writeback work processing, while the pending writeback
work units grew to insane lengths.

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1736230

[ Fixed up CC list. John, you're sending email with
  From: John Stoffel <john@quad.stoffel.home> ]

On Wed, Sep 20, 2017 at 01:32:25PM -0600, Jens Axboe wrote:
> On 09/20/2017 01:29 PM, John Stoffel wrote:
> > On Tue, Sep 19, 2017 at 01:53:01PM -0600, Jens Axboe wrote:
> >> We've had some issues with writeback in presence of memory reclaim
> >> at Facebook, and this patch set attempts to fix it up. The real
> >> functional change is the last patch in the series, the first 5 are
> >> prep and cleanup patches.
> >>
> >> The basic idea is that we have callers that call
> >> wakeup_flusher_threads() with nr_pages == 0. This means 'writeback
> >> everything'. For memory reclaim situations, we can end up queuing
> >> a TON of these kinds of writeback units. This can cause softlockups
> >> and further memory issues, since we allocate huge amounts of
> >> struct wb_writeback_work to handle this writeback. Handle this
> >> situation more gracefully.
> > 
> > This looks nice, but do you have any numbers to show how this improves
> > things?  I read the patches, but I'm not strong enough to comment on
> > them at all.  But I am interested in how this improves writeback under
> > pressure, if at all.
> 
> Writeback should be about the same, it's mostly about preventing
> softlockups and excessive memory usage, under conditions where we are
> actively trying to reclaim/clean memory. It was bad enough to cause
> softlockups for writeback work processing, while the pending writeback
> work units grew to insane lengths.

In numbers, we have seen situations where we had 600 million writeback
work items queued up from reclaim under pressure. That's 35G worth of
work descriptors, and the machine was struggling to remain responsive
due to a lack of memory.

Once writeback against all outstanding dirty pages has been requested,
there really isn't a need to queue even a second work item; the job is
already being performed. We can queue the next one when it completes.

[toc] | [prev] | [standalone]

Page 2 of 2 — ← Prev page 1 [2]

Back to top | Article view | linux.kernel

csiph-web