Re: Git Repository for WITH RECURSIVE and others

From: David Fetter <david(at)fetter(dot)org>
To: Magnus Hagander <magnus(at)hagander(dot)net>
Cc: PG Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Git Repository for WITH RECURSIVE and others
Date: 2008-07-07 17:47:45
Message-ID: 20080707174745.GG15394@fetter.org
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Mon, Jun 30, 2008 at 05:30:19PM +0200, Magnus Hagander wrote:
> David Fetter wrote:
> > On Mon, Jun 30, 2008 at 01:50:26PM +0200, Magnus Hagander wrote:
> >> David Fetter wrote:
> > [gitosis]
> >> I'm not sure I agree that this is a big problem, but sure, we
> >> should at least consider git-shell.
> >
> > Please explain your reasoning here. The project has taken nasty
> > hits on its infrastructure already (pgfoundry) because the author
> > of the software had a go-it-alone, I-know-best attitude that
> > sooner than later forced us to fork. As a direct consequence,
> > pgfoundry now needs a redo that will take a pgfoundry
> > administrator many of work in their "ample spare time."
>
> If the reason for this is that the software isn't usable, that's one
> thing. If it's just the author that considers "a git snapshot is my
> release packaging, not a tarball", I don't see how that in itself
> has any effect on the quality of the software.

You don't? You have *got* to be joking.

> If that's the only thing it's saying, I don't think that in itself
> is enough to disqualify gitosis.

The author's got a haughty disinterest in having anybody else ever
participated in gitosis development. That's a show-stopper, totally
independent of the current code.

> >> Is there any product out there that makes it possible to admin a
> >> git-shell based system without having all the admins being root on
> >> the server? Because that's simply not an option if you want
> >> anything remotely scalable.
> >
> > I don't know what you mean by "remotely scalable," but it's clearly
> > not the same definition I have. A sudo wrapper which only allows
> > creation, editing and deletion of accounts restricted to git-shell
> > will scale just fine.
>
> A properly working sudo wrapper that will let you do *everything needed*
> is good enough for me.

We can make one of those, and it doesn't have to--can't be--perfect
the first time through because we will find capabilities it needs and
ones we supplied that it doesn't.

> >> Show me such a solution, and I'll be happy to consider it :-)
> >
> > 1. Create a (set of) program(s) which does exactly the following things:
> >
> > * Create an account with git-ssh as its shell.
> > * Manipulate the contact information, ssh keys and groups of said account.
> > * Delete the account.
>
> Rght. Is there a product out there already that lets us do this, or is
> it something we need to write ourselves?
>
> You'll also need scripts to create and modify the GIT responsitories
> themselves, no?
>
> Since it's sudo, it has to be secure after all, so it's not necessarily
> a 2 minute hack.

The more I think about this, the more it looks like admin tasks and
not like tools. We don't know enough about what's actually going to
be going on to create such tools yet.

> > 2. Create a unix group and corresponding sudo role that accesses the above.
> >
> > 3. Create shell accounts as needed with the above group. Yes, that's
> > a root-only task, but it's a short one.
>
> Um, not following that step. What account are you talking about here?
> Creating the accounts for the admins?

Per-project accounts.

> That's not an issue, since I assume that's not something that would
> be done very often :-)
>
> > I believe that the above takes care of 90% or more of tasks. If
> > it turns out that we need to automate more, we can add that
> > (semi)automation to the capabilities above :)
>
> As long as it allows it. For example, having a webserver do sudo is
> not something that makes me feel very safe (and yes, I've seen
> solutions that do that claiming to be secure. And sure, you *can*
> build them secure, it's just a lot harder than most people who
> choose to do it are aware of)

You've brought up security, and that's just great. Now that you've
brought it up, how about sketching out a threat model? It's only
possible to discuss security measures with reference to a threat
model.

Cheers,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Josh Berkus 2008-07-07 18:03:31 Re: CommitFest rules
Previous Message Brendan Jurd 2008-07-07 17:47:06 Re: CommitFest rules