[Kde-scm-interest] Have we arrived to a dead end?

[Oswald Buddenhagen]

On Sat, Feb 13, 2010 at 05:11:53PM +0100, Thiago Macieira wrote:
> Em Sábado 13. Fevereiro 2010, às 16.39.57, Oswald Buddenhagen escreveu:
> > i did imagine that case. but somehow i just couldn't construct an
> > actually useful scenario where the parts of a *single* atomic commit
> > would have to be pushed to *different* servers.
> 
> Very simple: when a part of the tree is not on Gitorious. Something like a 
> 3rd-party project, a kdesupport project or even part of KDE itself (like 
> KOffice).
> 
oh, you mean *global* atomicity. commit atomicity beyond what svn
currently offers it really not in scope as far as i'm concerned.

afaict, the "worst" thing that could happen is that one has to consider
all distinct repositories one pushes to as an atomic entity as far as
distribution goes. this is no worse than the "atomicity scope" being a
single repository in the first place, but comes with the various
benefits of splitting, and on top of that allowing kde-wide atomicity
which is simply not achievable by any other practical means known to me.

> Anyway, when I said load balancing, I didn't imagine duplicating
> everything in two servers. You can put some repositories on server1
> and some repositories on server2. Done, you no longer have atomicity
> because of a technical decision.
> 
it doesn't sound too likely that they would have to tear apart "project
clusters" for load reasons. it's simply implausible that the push rate
of even something as big as KDE/ would exceed the load capability of a
single server. and balancing clone access is irrelevant to that
question, as they can mirror internally.