diff mbox series

[1/1] blk-throttle: fix throttle configuring not effective

Message ID 20230812072116.42321-1-zhuxiaohui.400@bytedance.com (mailing list archive)
State New, archived
Headers show
Series [1/1] blk-throttle: fix throttle configuring not effective | expand

Commit Message

zhuxiaohui Aug. 12, 2023, 7:21 a.m. UTC
when updating block throttle limit with persistence and stable io
pressure, especially a relative high io pressure, fio test e.g.,
there may never be a change to start a new slice, and carryover_ios &
carryover_bytes will not be cleared.

As a result, when reconfiguring block throttle limit, we can notice that
the actual iops and throughput is a random value far away from what is
set

So we need to update carryover value when dispatching bio

Signed-off-by: zhuxiaohui <zhuxiaohui.400@bytedance.com>
---
 block/blk-throttle.c | 26 ++++++++++++++++++++++++++
 block/blk-throttle.h |  4 ++--
 2 files changed, 28 insertions(+), 2 deletions(-)

Comments

Yu Kuai Aug. 12, 2023, 7:53 a.m. UTC | #1
Hi,

在 2023/08/12 15:21, zhuxiaohui 写道:
> when updating block throttle limit with persistence and stable io
> pressure, especially a relative high io pressure, fio test e.g.,
> there may never be a change to start a new slice, and carryover_ios &
> carryover_bytes will not be cleared.
> 
> As a result, when reconfiguring block throttle limit, we can notice that
> the actual iops and throughput is a random value far away from what is
> set
> 
> So we need to update carryover value when dispatching bio

I don't understand, not clear carryover_bytes/ios is what expected, and
how can they affect actual bandwith/iops.

Can you give a example how you tested and why current calculation is not
correct?

Thanks,
Kuai

> 
> Signed-off-by: zhuxiaohui <zhuxiaohui.400@bytedance.com>
> ---
>   block/blk-throttle.c | 26 ++++++++++++++++++++++++++
>   block/blk-throttle.h |  4 ++--
>   2 files changed, 28 insertions(+), 2 deletions(-)
> 
> diff --git a/block/blk-throttle.c b/block/blk-throttle.c
> index 7397ff199d66..13c9d87a7201 100644
> --- a/block/blk-throttle.c
> +++ b/block/blk-throttle.c
> @@ -821,6 +821,30 @@ static void tg_update_carryover(struct throtl_grp *tg)
>   		   tg->carryover_ios[READ], tg->carryover_ios[WRITE]);
>   }
>   
> +static void tg_charge_carryover(struct throtl_grp *tg, struct bio *bio)
> +{
> +	bool rw = bio_data_dir(bio);
> +
> +	if (unlikely(tg->carryover_bytes[rw])) {
> +		unsigned int bio_size = throtl_bio_data_size(bio);
> +		unsigned int carryout_size = abs(tg->carryover_bytes[rw]);
> +
> +		carryout_size = min(carryout_size, bio_size);
> +
> +		if (tg->carryover_bytes[rw] < 0)
> +			tg->carryover_bytes[rw] += carryout_size;
> +		else
> +			tg->carryover_bytes[rw] -= carryout_size;
> +	}
> +
> +	if (unlikely(tg->carryover_ios[rw])) {
> +		if (tg->carryover_ios[rw] < 0)
> +			tg->carryover_ios[rw] += 1;
> +		else
> +			tg->carryover_ios[rw] -= 1;
> +	}
> +}
> +
>   static unsigned long tg_within_iops_limit(struct throtl_grp *tg, struct bio *bio,
>   				 u32 iops_limit)
>   {
> @@ -965,6 +989,8 @@ static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
>   
>   	tg->io_disp[rw]++;
>   	tg->last_io_disp[rw]++;
> +
> +	tg_charge_carryover(tg, bio);
>   }
>   
>   /**
> diff --git a/block/blk-throttle.h b/block/blk-throttle.h
> index d1ccbfe9f797..8f1642becb23 100644
> --- a/block/blk-throttle.h
> +++ b/block/blk-throttle.h
> @@ -127,8 +127,8 @@ struct throtl_grp {
>   	 * bytes/ios are waited already in previous configuration, and they will
>   	 * be used to calculate wait time under new configuration.
>   	 */
> -	uint64_t carryover_bytes[2];
> -	unsigned int carryover_ios[2];
> +	int64_t carryover_bytes[2];
> +	int carryover_ios[2];
>   
>   	unsigned long last_check_time;
>   
>
Yu Kuai Aug. 14, 2023, 2:01 a.m. UTC | #2
+CC Michal.

在 2023/08/12 15:53, Yu Kuai 写道:
> Hi,
> 
> 在 2023/08/12 15:21, zhuxiaohui 写道:
>> when updating block throttle limit with persistence and stable io
>> pressure, especially a relative high io pressure, fio test e.g.,
>> there may never be a change to start a new slice, and carryover_ios &
>> carryover_bytes will not be cleared.
>>
>> As a result, when reconfiguring block throttle limit, we can notice that
>> the actual iops and throughput is a random value far away from what is
>> set
>>
>> So we need to update carryover value when dispatching bio
> 
> I don't understand, not clear carryover_bytes/ios is what expected, and
> how can they affect actual bandwith/iops.
> 
> Can you give a example how you tested and why current calculation is not
> correct?

I can reporduce this, but this patch is obviously wrong. You must
explaim how the calculation is not correct.

After a quick loock, I found that carryover_bytes/ios is not updated in
throtl_trim_slice(), while tg->io/bytes_disp[] can be cleared. This is
definitly a problem.

Thanks,
Kuai

> 
> Thanks,
> Kuai
> 
>>
>> Signed-off-by: zhuxiaohui <zhuxiaohui.400@bytedance.com>
>> ---
>>   block/blk-throttle.c | 26 ++++++++++++++++++++++++++
>>   block/blk-throttle.h |  4 ++--
>>   2 files changed, 28 insertions(+), 2 deletions(-)
>>
>> diff --git a/block/blk-throttle.c b/block/blk-throttle.c
>> index 7397ff199d66..13c9d87a7201 100644
>> --- a/block/blk-throttle.c
>> +++ b/block/blk-throttle.c
>> @@ -821,6 +821,30 @@ static void tg_update_carryover(struct throtl_grp 
>> *tg)
>>              tg->carryover_ios[READ], tg->carryover_ios[WRITE]);
>>   }
>> +static void tg_charge_carryover(struct throtl_grp *tg, struct bio *bio)
>> +{
>> +    bool rw = bio_data_dir(bio);
>> +
>> +    if (unlikely(tg->carryover_bytes[rw])) {
>> +        unsigned int bio_size = throtl_bio_data_size(bio);
>> +        unsigned int carryout_size = abs(tg->carryover_bytes[rw]);
>> +
>> +        carryout_size = min(carryout_size, bio_size);
>> +
>> +        if (tg->carryover_bytes[rw] < 0)
>> +            tg->carryover_bytes[rw] += carryout_size;
>> +        else
>> +            tg->carryover_bytes[rw] -= carryout_size;
>> +    }
>> +
>> +    if (unlikely(tg->carryover_ios[rw])) {
>> +        if (tg->carryover_ios[rw] < 0)
>> +            tg->carryover_ios[rw] += 1;
>> +        else
>> +            tg->carryover_ios[rw] -= 1;
>> +    }
>> +}
>> +
>>   static unsigned long tg_within_iops_limit(struct throtl_grp *tg, 
>> struct bio *bio,
>>                    u32 iops_limit)
>>   {
>> @@ -965,6 +989,8 @@ static void throtl_charge_bio(struct throtl_grp 
>> *tg, struct bio *bio)
>>       tg->io_disp[rw]++;
>>       tg->last_io_disp[rw]++;
>> +
>> +    tg_charge_carryover(tg, bio);
>>   }
>>   /**
>> diff --git a/block/blk-throttle.h b/block/blk-throttle.h
>> index d1ccbfe9f797..8f1642becb23 100644
>> --- a/block/blk-throttle.h
>> +++ b/block/blk-throttle.h
>> @@ -127,8 +127,8 @@ struct throtl_grp {
>>        * bytes/ios are waited already in previous configuration, and 
>> they will
>>        * be used to calculate wait time under new configuration.
>>        */
>> -    uint64_t carryover_bytes[2];
>> -    unsigned int carryover_ios[2];
>> +    int64_t carryover_bytes[2];
>> +    int carryover_ios[2];
>>       unsigned long last_check_time;
>>
> 
> .
>
diff mbox series

Patch

diff --git a/block/blk-throttle.c b/block/blk-throttle.c
index 7397ff199d66..13c9d87a7201 100644
--- a/block/blk-throttle.c
+++ b/block/blk-throttle.c
@@ -821,6 +821,30 @@  static void tg_update_carryover(struct throtl_grp *tg)
 		   tg->carryover_ios[READ], tg->carryover_ios[WRITE]);
 }
 
+static void tg_charge_carryover(struct throtl_grp *tg, struct bio *bio)
+{
+	bool rw = bio_data_dir(bio);
+
+	if (unlikely(tg->carryover_bytes[rw])) {
+		unsigned int bio_size = throtl_bio_data_size(bio);
+		unsigned int carryout_size = abs(tg->carryover_bytes[rw]);
+
+		carryout_size = min(carryout_size, bio_size);
+
+		if (tg->carryover_bytes[rw] < 0)
+			tg->carryover_bytes[rw] += carryout_size;
+		else
+			tg->carryover_bytes[rw] -= carryout_size;
+	}
+
+	if (unlikely(tg->carryover_ios[rw])) {
+		if (tg->carryover_ios[rw] < 0)
+			tg->carryover_ios[rw] += 1;
+		else
+			tg->carryover_ios[rw] -= 1;
+	}
+}
+
 static unsigned long tg_within_iops_limit(struct throtl_grp *tg, struct bio *bio,
 				 u32 iops_limit)
 {
@@ -965,6 +989,8 @@  static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
 
 	tg->io_disp[rw]++;
 	tg->last_io_disp[rw]++;
+
+	tg_charge_carryover(tg, bio);
 }
 
 /**
diff --git a/block/blk-throttle.h b/block/blk-throttle.h
index d1ccbfe9f797..8f1642becb23 100644
--- a/block/blk-throttle.h
+++ b/block/blk-throttle.h
@@ -127,8 +127,8 @@  struct throtl_grp {
 	 * bytes/ios are waited already in previous configuration, and they will
 	 * be used to calculate wait time under new configuration.
 	 */
-	uint64_t carryover_bytes[2];
-	unsigned int carryover_ios[2];
+	int64_t carryover_bytes[2];
+	int carryover_ios[2];
 
 	unsigned long last_check_time;