[dpdk-dev] [PATCH v2 4/7] rte_sched: don't clear statistics when read

[Dumitrescu, Cristian]

> -----Original Message-----
> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Thomas Monjalon
> Sent: Tuesday, February 24, 2015 8:07 PM
> To: Stephen Hemminger
> Cc: dev at dpdk.org; Stephen Hemminger
> Subject: Re: [dpdk-dev] [PATCH v2 4/7] rte_sched: don't clear statistics when
> read
> 
> 2015-02-24 11:18, Stephen Hemminger:
> > On Mon, 23 Feb 2015 23:51:31 +0000
> > Thomas Monjalon <thomas.monjalon at 6wind.com> wrote:
> >
> > > 2015-02-05 07:43, Neil Horman:
> > > > On Wed, Feb 04, 2015 at 10:13:58PM -0800, Stephen Hemminger wrote:
> > > > > +
> > > > > +/**
> > > > > + * Hierarchical scheduler subport statistics reset
> > > > > + *
> > > > > + * @param port
> > > > > + *   Handle to port scheduler instance
> > > > > + * @param subport_id
> > > > > + *   Subport ID
> > > > > + * @return
> > > > > + *   0 upon success, error code otherwise
> > > > > + */
> > > > > +int
> > > > > +rte_sched_subport_stats_reset(struct rte_sched_port *port,
> > > > > +			      uint32_t subport_id);
> > > > > +
> > > > >  /**
> > > > >   * Hierarchical scheduler queue statistics read
> > > > >   *
> > > > > @@ -338,6 +353,20 @@ rte_sched_queue_read_stats(struct
> rte_sched_port *port,
> > > > >  	struct rte_sched_queue_stats *stats,
> > > > >  	uint16_t *qlen);
> > > > >
> > > > > +/**
> > > > > + * Hierarchical scheduler queue statistics reset
> > > > > + *
> > > > > + * @param port
> > > > > + *   Handle to port scheduler instance
> > > > > + * @param queue_id
> > > > > + *   Queue ID within port scheduler
> > > > > + * @return
> > > > > + *   0 upon success, error code otherwise
> > > > > + */
> > > > > +int
> > > > > +rte_sched_queue_stats_reset(struct rte_sched_port *port,
> > > > > +			    uint32_t queue_id);
> > > > > +
> > > > Both need to be added to the version map to expose them properly.
> > > > Neil
> > >
> > > Stephen, this patchset is partially acked and could enter in 2.0.0-rc1.
> > > May you send a v3 addressing comments? Or should I break the serie by
> > > applying only some of them? Or postpone the serie to 2.1?
> >
> > I can resend v3. Wasn't clear that a conclusion was reached.
> > IMHO read should not clear.
> 
> Me too. I'm just saying that I cannot apply anything.
> So you have to decide the strategy to adopt for your patches.

How about my latest proposal to have the stats read functions either reset the counters or not, based on init-time user configuration? I did not see any reply on this.

Maybe you guys missed my reply, I am pasting it below:

"Personally, I think we should avoid proliferating the number of stats functions, I would keep a single set of stats read functions, which can clear the stats or not, depending on behaviour configured per rte_sched object at creation time. Basically, based on the value of configuration parameter struct rte_sched_params::clear_stats_on_reset, the stats read functions do clear the counters or not. In my opinion, this allows a clean init-time selection of the required behaviour, and it also provides backward compatibility. Any issues with this approach?"