[dpdk-dev] [PATCH v9 1/7] lib: add information metrics library

[Bruce Richardson]

On Tue, Jan 31, 2017 at 01:13:11PM +0000, Mcnamara, John wrote:
> 
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Thomas Monjalon
> > Sent: Monday, January 30, 2017 3:50 PM
> > To: Horton, Remy <remy.horton at intel.com>
> > Cc: dev at dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v9 1/7] lib: add information metrics
> > library
> > 
> > Hi Remy,
> > 
> > > This patch adds a new information metric library that allows other
> > > modules to register named metrics and update their values. It is
> > > intended to be independent of ethdev, rather than mixing ethdev and
> > > non-ethdev information in xstats.
> > 
> > I'm still not convinced by this library, and this introduction does not
> > help a lot.
> > 
> > I would like to thanks Harry for the review of this series.
> > If we had more opinions or enthousiasm about this patch, it would be
> > easier to accept this new library and assert it is the way to go.
> 
> Hi,
> 
> The RFCs for this library (initially two, merged into one) have been up since October, during the 16.11 timeframe. Comments were made and applied.
> 
>    http://dpdk.org/ml/archives/dev/2016-October/049571.html
>    http://dpdk.org/ml/archives/dev/2016-October/048390.html
> 
> I'm concerned that these new comments/reservations are coming in very, very late in the 17.02 release cycle.
> 
> If there hasn't been a lot of opinions or enthusiasm then equally there hasn't been other reservations. If there had been we would have addressed them.
> 
> 
> > It could be a matter of technical board discussion if we had a clear
> > explanation of the needs, the pros and cons of this design.
> 
> We are happy to have the design discussed at the Technical Board. We would also like the inclusion of this library in RC3 to be discussed since that is still our desired outcome. 
> 
> We have, as any other company would have, customers awaiting features, developers committed to timelines, and testing and integration roadmaps. Blocking or delaying features at the last moment isn't an effective model that we, and I'm sure other companies, can work with.
> 
> As such, it is probably best, that all current and future RFCs are reviewed at the Technical Board and that the board gives an indication on whether the proposal is acceptable for upstreaming or not. 
> 

I would tend to agree with this. The tech board should indeed look to
insure that all RFCs and V1s have had some feedback on them well before
the merge deadline.

I don't believe it's fair on developers to suddenly give feedback at
merge-time and thereby prevent the patch making it into a release,
without giving time to do any rework.  This is especially true if the
patch had already been reviewed and acked, and so could be considered
"ready for merge".

The tech board should also discuss some reasonable guidelines
for this area. My opinion is that by the time RC1 is released, any
patches that have been reviewed, acked and have no outstanding
comments on them for e.g. 1 week, must be merged in for the release. Any
additional feedback thereafter should be considered "too late", and
should be addressed in the following release. This will help to
incentivize reviewers to review early, and also give developers some
degree of confidence that their patches will be merged in. We have
deadlines for submitters to get patches in, we should also have
deadlines for reviewers to object to those patches.

Regards,
/Bruce