Re: [PATCH V3] block/throttle: Add IO throttled information in blkio.throttle

[Vivek Goyal <vgoyal@redhat.com>]

On Tue, Oct 09, 2012 at 02:53:45PM +0800, Robin Dong wrote:
> From: Robin Dong <sanbai@taobao.com>
> 
> Currently, if the IO is throttled by io-throttle, the SA has no idea of
> the situation and can't report it to the real application user about
> that he/she has to do something. So this patch adds a new interface
> named blkio.throttle.io_queued which indicates how many IOs are
> currently throttled.
> 
> The nr_queued[] of struct throtl_grp is of type "unsigned int" and updates
> to it are atomic both at 32bit and 64bit platforms, so we could just
> read tg->nr_queued only under blkcg->lock.
> 
> Changelog from v2:
> 	Use nr-queued[] of struct throtl_grp for stats instaed of adding new blkg_rwstat.
> 
> Cc: Tejun Heo <tj@kernel.org>
> Cc: Vivek Goyal <vgoyal@redhat.com>
> Cc: Jens Axboe <axboe@kernel.dk>
> Signed-off-by: Tao Ma <boyu.mt@taobao.com>
> Signed-off-by: Robin Dong <sanbai@taobao.com>
> ---
>  block/blk-throttle.c |   40 ++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 40 insertions(+), 0 deletions(-)
> 
> diff --git a/block/blk-throttle.c b/block/blk-throttle.c
> index a9664fa..e410448 100644
> --- a/block/blk-throttle.c
> +++ b/block/blk-throttle.c
> @@ -953,6 +953,32 @@ static u64 tg_prfill_cpu_rwstat(struct seq_file *sf,
>  	return __blkg_prfill_rwstat(sf, pd, &rwstat);
>  }
>  
> +static u64 tg_prfill_io_queued(struct seq_file *sf,
> +				struct blkg_policy_data *pd, int off)
> +{
> +	static const char *rwstr[] = {
> +		[READ]	= "Read",
> +		[WRITE]	= "Write",
> +	};
> +	struct throtl_grp *tg = pd_to_tg(pd);
> +	const char *dname = NULL;
> +	unsigned int v;
> +	int i;
> +
> +	if (pd->blkg->q->backing_dev_info.dev)
> +		dname = dev_name(pd->blkg->q->backing_dev_info.dev);
> +
> +	if (!dname)
> +		return 0;
> +
> +	for (i = 0; i <= WRITE; i++)
> +		seq_printf(sf, "%s %s %u\n", dname, rwstr[i], tg->nr_queued[i]);

You are printing only READ/WRITE stats and not the SYNC/ASYNC stats. This
is not inline with rest of the stats like throttle.io_serviced and
throttle.io_service_bytes.

I guess it is better to maintain rwstat for io_queued and display it
according to io_serviced.

Thanks
Vivek