[web-team] script usage & failure modes

[Rick Moen]

Hi, just catching up on this thread.

Quoting Lisa Corsetti (svlug at flygirl.com):

> Well, the 2230 now has a plug ON THE BACK, so no cable need be
> run into the box and we found a matching cable to go between the
> two boxes.  The only thing we couldn't find that I *really* wish
> we had were the threadded stand-offs to allow the cable to screw
> into the JBOD so that it won't come lose... I don't trust cables
> to hold themselves in place.

Yeah, I'll check my own parts, but am pretty sure I have only standard
sized standoffs, and these ones would need really tiny threads.

> > Moin upstream uses mercurial.  If we want to be able to update
> > code-in-flight instead of fetch tarballs, we'll want hg.
> 
> That's fine.  The SVN was mainly for the stuff I'd done for the
> porting of the old site.

It was interesting fixing the server end of svn, something I'd
personally never worked with, before (only used the client side -- other
than experimentation with very early alpha versions, way back when).
This was among the last tasks of the evening, so I for one was really
tired, but eventually we debugged an /etc/xinetd.d/ file for the
"svnserve" service -- which earlier someone had hacked together, but had
gotten accidentally blown away during site upgrades.  

As an aside, everywhere else I've used svn, access method in use has
been ssh.  That is, I've never before used a local repo, and so had
never encountered svnserve prior to the other day.

Anyway, converting the svn repo to hg (Mercurial)[1] has the
advantage that we elminate two network services (svnserve and xinetd), 
the need for svn and xinetd being present at all, and a whole bunch of
svn's library dependencies.

I've not personally yet worked with hg, but have heard excellent things
about it -- and Sun now uses it to manage the entire Solaris and Java
codebases.


> > Setting up fastcgi is not something I have done before. 
> > Setting up moin is.  In fact I don't even know who set up the
> > fastcgi that's in the linode.
> 
> I believe that Rick set up and I *think* he said that you can go
> ahead and test it if you like.

It actually was Paul, to his great credit, who said he'd solved that
problem at the end of 2006 (which is why we all believed the next step
was import of the MoinMoin prototype from gemini).  I've carefully
reported the fix as unverified, not because I doubt anyone's word, but
just because it's, well, unverified.

Prior to Paul's fix in late 2006, I'd beat my head up against that
problem for a while and wasn't getting anywhere, so I'd invited others
to take a crack at it, being pretty tired at that moment from solving
other problems prior to that one.

[1] http://www.selenic.com/mercurial/wiki/index.cgi/RepositoryConversion#head-804f3d2187e321f04ad1a3caaaeba01e0a51352c