[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bits (Nybbles?) from the Vancouver release team meeting



On Mon, Mar 14, 2005 at 02:24:47PM +0100, Aur?lien Jarno wrote:
> Hamish Moffatt a ?crit :
> >I see it as more a practical consideration. If you can't buy the
> >hardware new then you will have trouble keeping up with a growing unstable,
> >especially given the requirement that you need <= 2 buildds.
> So the requirement that you need <= 2 buildds is not well choosen. Why 
> such a requirement? m68k prooved that having a lot of buildd is not a 
> problem, *if they are correctly managed* (which is the case for m68k).

So that a security update doesn't have to wait an arbitrarily long time to
be released due to some arch taking forever to build a huge package.

That being said, I think rather than "no more than 2 buildds to keep up",
the criteria should have been "takes no longer than 150% of the time of the
most popconular arch to build packages", where you can measure that over a
sample of packages automatically.

This would allow the people managing an arch to work out what the best way
to satisfy that requirement is -- cross-compiling (ugh), distcc, black
magic, whatever.  The problem with "no more than 2 buildds to keep up with
unstable demand" is that a large increase in the rate of development would
kill all arches.

On the hardware availability angle, rather than "must be available new", the
criteria would be more usefully supplied as "must have at least 1
hot-spare buildd *and* 1 hot-spare developer-accessable machine at all
times".  If one of your active machines (or the hot spare) goes belly-up and
you can't get parts to get another hot spare happening, then that arch
obviously doesn't have readily available parts, and it's time to lead it
behind the shed and pop it.

> I may be wrong, but it looks like as the criterions have been choosed to 
> only keep some choosen architectures. The right way is too first choose 
> some criterions in function of *technical reasons*, and then look for 
> the architectures that meet them.

It's already been noted that most of the "blessed" archs don't meet one or
more of the criteria, so I'm not sure how accurate your observation is. 
<grin>

- Matt

Attachment: signature.asc
Description: Digital signature


Reply to: