ios and sectors should be incremented on completion

ios and sectors should be incremented on completion

[Akira Hayakawa]

Hi,

I seems DM increments stats (ios and sectors) before completion.
Why don't you do this in dec_pending?
block/stat.txt says these values are incremented when I/O request completes.

Is this to be fixed or do you have some justifications for this behavior?

static void _dm_request(struct request_queue *q, struct bio *bio)
{
	int rw = bio_data_dir(bio);
	struct mapped_device *md = q->queuedata;
	int cpu;
	int srcu_idx;
	struct dm_table *map;

	map = dm_get_live_table(md, &srcu_idx);

	cpu = part_stat_lock();
	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
	part_stat_unlock();

-- 
Akira Hayakawa <hayakawa@valinux.co.jp>

Re: ios and sectors should be incremented on completion

[Junichi Nomura]

On 09/26/14 18:12, Akira Hayakawa wrote:
> I seems DM increments stats (ios and sectors) before completion.
> Why don't you do this in dec_pending?
> block/stat.txt says these values are incremented when I/O request completes.
> 
> Is this to be fixed or do you have some justifications for this behavior?

Do you have a problem with the current behavior other than the document?

AFAIR, dm increments ios/sectors at submission because md did.
And other bio-based drivers such as drbd and nvme seem to have followed
the precedence.

While md has a reason to do so (some md personalities just remaps bios
and doesn't track completion), dm should be able to collect stat
in dec_pending as it always clones.

So, I think it's ok to move ios/sectors counting to end_io_acct().
It might make tools like iostat happier.
You might want to fix other bio-based drivers as well.

-- 
Jun'ichi Nomura, NEC Corporation

Re: ios and sectors should be incremented on completion

[Akira Hayakawa]

Thanks for information Junichi,

> Do you have a problem with the current behavior other than the document?
No. I and my colleagues just wondered why.

> AFAIR, dm increments ios/sectors at submission because md did.
> And other bio-based drivers such as drbd and nvme seem to have followed
> the precedence.
> 
> While md has a reason to do so (some md personalities just remaps bios
> and doesn't track completion), dm should be able to collect stat
> in dec_pending as it always clones.
Very interesting historical background.
Maybe bache, too.

> So, I think it's ok to move ios/sectors counting to end_io_acct().
> It might make tools like iostat happier.
> You might want to fix other bio-based drivers as well.
I want to hear comments from DM maintainers on this issue.

- Akira