Re: [Koha-devel] [URGENT] Move away from Savannah/CVS

[MJ Ray]

"Joshua M. Ferraro" <address@hidden> wrote:
> Two days of downtime on a project repository and host for the
> primary download for the project's releases is, in my view,
> an unacceptable service outage. The fact that they didn't
> even bother to put a splash page up for savannah.nongnu.org,
> giving a timeline for the outage, just aggravates the issue.

It seems that the maintainers expected people to look for news on
www.gnu.org or www.fsf.org, all the available on-site sysadmin support
is working on recovering the filesystem (so wire-swapping to bring up
a splash screen wasn't an immediate option) and changing DNS would
make it slower to bring back the service once it returns.

If anyone wants to give more resources to Savannah, FSF accepts
donations by transfer https://www.fsf.org/donate/bankdetails/ -
specify that the money is for Savannah.

> This isn't the first time we've had major problems with 
> Savannah, remember when the mailing lists were taking 4 days
> to deliver a mail?

Not really.  Delays to email are irritating but nowhere near as
frustrating as the old commit delays on the last corporation-hosted
central repository (Sourceforge).

> The point is not that savannah will be up in a day or two,
> but that we have no assurance that it won't just go down
> for days again... in this case, the timing was particularly
> bad as we'd planned to do some major work this week.

We have no assurance that any hosting service will stay up at a
particularly vital moment.  Self-reliance is needed.  Distributed
version control should help that.

[...]
> > Also, if other koha developers had been using git and the
> > cvs-compatibility commands, we could all have been working through
> > this Savannah downtime.
> It's not that git, arch, etc., are hard to use ... it's the 
> concept and management of a distributed version control system,

What do people find so hard about this concept?  Instead of sending
all traffic over the network to a central store, you send it to a
store on your local disk.  Then, when you're ready, you can
synchronise any two stores by committing to or merging from it.

To use an analogue, CVS is like connected IMAP, while git is like
disconnected IMAP.  I don't read many complaints about how
disconnected IMAP is a difficult concept (although disappointingly few
ISPs and mail clients support it still).

> and the lack of a clear leader in this arena that leads me to
> conclude that DVC is not quite there yet. We don't have much
> bandwidth to devote to managing a version control system in this
> community, I don't want to hop from DVC to DVC as I've seen so
> many other projects do.

So don't do it.  Pick one and stick with it, come what may.  Each of
the current DVCs has at least one flaw which may rule it out, but then
so does SVN.  They mostly work with similar ideas, with the possible
exception of Aegis IIRC.

> [...] I think establishing backup sites/ mirrors of
> the repository should be high on the priority list for sometime in the
> next few weeks (maybe ibiblio.org?).

Great.  This seems a good idea.  Mirroring is something which is
designed into any distributed version control system from the outset.
Even if you choose not to use it at first, it's usually not too
difficult.

Can you tell me more about ibiblio.org's repository mirroring?  I
know they mirror tarballs and big FTP sites in this field, but I'd not
heard about their repository mirroring.

> That said, Google has shown a
> good faith effort to support the OSS community, most notably with their
> Summer of Code project, and I've no reason to suspect they would
> abandon their hosting project.

They do lots of other marketing too.  That's what Code is to them:
marketing.  Marketing themselves to the tech-savvy opinion formers, to
prospective developers, and so on.  When it ceases to be justifiable
as marketing, their Code offerings will either change (adverts,
finding ways to monetise the code search, or otherwise) or close.

> Personally, I don't subscribe to the
> 'Google is a corporation and therefore evil' theory; [...]

Me neither.  Misdescribing the counter-argument like that is a very
dishonourable tactic.

It's also stupid to suggest that about me.  Given that I'm a member of
one of the UK's largest retail corporation groups, I hope it's obvious
that I don't think all corporations are evil.

> in fact, I'll
> go on the record as saying I think in many ways, Google's a great
> example of how corporations should be run. [...]

I'd agree that it's seems efficiently-run, but I disagree that it's a
great example.  It is slow to fix bugs (even serious ones sometimes),
amoral (see the China debate), and externalises costs (also known as
polluting) at a frightening rate (such as invite-spamming).

Google has taken services off-line temporarily and changed terms
(including ad services, where some people pay big money for the
services) without warning in the past and will presumably do so again.
I don't think their size gives us certainty.  If anything, it makes it
harder for us to influence them.


In short, I think we shouldn't trust Google with Koha when there are
so many alternatives.  My first preference is to trust ourselves
(preferably with wider use of DVC).  My second would be to trust
another non-profit foundation.

As an aside to that, I would really like to have CVS-compatibility as
far as possible, if there's a central service.

Regards,
-- 
MJ Ray - see/vidu http://mjr.towers.org.uk/email.html
Webmaster/web developer, statistician, sysadmin, online shop maker,
developer of koha, debian, gobo, gnustep, various mail and web s/w.
Workers co-op @ Weston-super-Mare, Somerset http://www.ttllp.co.uk/