Release Engineering Meeting Recap from Monday 16-APR-07

[Axel Thimm]

On Mon, Apr 23, 2007 at 12:26:33PM +0200, Thorsten Leemhuis wrote:
> On 23.04.2007 11:34, Axel Thimm wrote:
> > On Mon, Apr 23, 2007 at 06:49:20AM +0200, Thorsten Leemhuis wrote:
> >> On 23.04.2007 05:56, Dave Jones wrote:
> >>> On Mon, Apr 23, 2007 at 08:07:54AM +0530, Rahul Sundaram wrote:
> >>>  > I would prefer a rebuild
> >> Even with deltarpms it's IMHO to much risk and overhead (slower dist
> >> updates due to more packages needing updates; reassembling the rpms also
> >> takes time) for a small gain.
> > Didn't we already ban this as an urban legend?
> 
> No, I simply stopped replying to the thread where you calculated the
> number of packages that got rebuild since there was obviously a
> disagreement between us two if the numbers are correct or not for
> Extras.

I wasn't referring to our discssion, there were more people that
considered the "churn" issue to to be one.

> But not replying doesn't mean that I agree with you, so no, we did not
> "ban this as an urban legend".

We != you and me alone.

> > Can you otherwise show what savings FC6 -> F7 will give? [...]
> 
> I'd be simply glad if every package that doesn't need to be touched
> doesn't get touched (downloads, risk of changes, rpmnewfiles,
> rpmsavefiles, ...). Even if the number is low -- for example it are
> currently 1/10 of the  package on the rawhide machine I'm currently using.

Perhaps your system is either not updating properly, or you have a
very distinct pick of packages, at the very least it is not typical
for a common Fedora user (see below why). Please do check the real
numbers on a mirror. I'll do that for you:

I already quoted that FC usually had a repo build rate of 95-100% and
now for the first time dropped to 80% (that was two weeks ago and the
number has changed, still let's not have daily rebuild numbers).

I won't get trapped in discussing again about how to interprete FE
rebuilds since the data needs to be retrieved from the CVS or
elsewhere. The typical user is anyway using mostly Fedora Core
components by definition, since until now avery DVD/CD downloaded only
had FC inside and the bits for FE were cherry-picked and
net-downloaded.

So there are 80% or new rebuilds, but is this 80% of the bandwidth? We
know that the big players like kernels, openoffice and friends have
been rebuild, let's look at the true size-weighted rebuild rate:

			    97.69%

In order to emphasize this, the savings in Fedora Core upgrades from
FC6 to F7 are

			    2.21%

And that is the mean over i386/x86_64/ppc/source, it even get's less
when looking at the binaries only:

			   i386:   2.05%
			   x86_64: 1.70%
			   ppc:    1.48%

			   source: 4.88%

So, for me the "churn" argument puffs into thin air. Blocking the
rebuilds of the remaining 20% of packages for saving 2% or less of
download bandwidth? It's a joke.

Can we consider the "big churn" an urban legend now? And if there are
still doubters amongst us, then lest them install FC6 on their systems
and upgarde to rawhide. May the bandwidth be with them.

They will return crying that they only saved less than 2% of what they
had on disk ...

> > Changing to ".1" either means dropping disttags forever, or obscuring
> > disttags into integers, so people don't notice. [...]

> Seems we didn't understand each other what I was up to and you bring in
> stuff that was never part of what I meant to do -- a disttag like ".2"
> for example was never mentioned in my mails in the way you describe it
> above

OK, so you are for dropping disttags forever. This was covered in my
mail and Rahul's.

> >>> The entire purpose of the dist tag is to discriminate between
> >>> two otherwise same version packages across two releases.
> > And to ensure proper upgrade paths, e.g. releated releases should have
> > disttags that 
> 
> /me wonders if something is missing here.

fill the blanks.

> Anyway, if you don't do exciting and unusual stuff then the upgrade path
> with my ".1" example is just the same as with ".fc7" as disttag.

And what about F8? Please think further than F7. See my comments on
EVR havok if we start dropping the disttags. There is a reason the
disttags became that popular that by F8 every package (more correct
98%) will have one. Unless they die in FUD, of course.

> > and useless and broken. Please do think this through for the long-term
> > implications your suggestion has.
> 
> I did and explained in on fedora-maintainers. But seems you have
> something different in mind then I have and outlined.

No, see above I had both cases covered.

> Axel, let's stop arguing endlessly over detail here.

But the devil is in the details. You "overlook" details like the size
of the churn (which is only 2%), often Fedora Extras did rebuilds (you
assumed 2/3, while you yourself involked full rebuilds for FC6 and
before), it's far better to use factual numbers than to make false
assumptions based on fear, uncertainty and doubt.

If we don't examine the details and present a sane analysis based on
the details wrong decisions will be made once more.

So, please do look at details and facts and do think stuff through to
the end and to the next to the next release.
-- 
Axel.Thimm at ATrpms.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.fedoraproject.org/pipermail/advisory-board/attachments/20070423/d7b47250/attachment.bin