Re: Linux Knowledge Base Project Requirements: Draft

[Roger Dingledine <arma@mit.edu>]

In message <Pine.BSF.4.05.9811011548340.28941-100000@shell5.ba.best.com>,
 aturner@best.com writes:
>I don't know how viable a mirror site will be.  A lot of the content will
>be dynamic, which makes standard mirroring tools like w3mir useless.
>Mirroring makes things like site design and links a lot more complicated.
>And if we will have a dedicated domain and server, a mirror is of
>questionable value. I think that SEUL would be better utilized as a test
>enviroment where we can try new things, software versions, etc without
>risking the production site.  Another idea is to use it as a image server-

Well, we don't have to mirror by trying to pull everything over via w3mir.
We can mirror by copying the site over, doing some perl magic to make it
conform to the new url base, and then just running it there. In any case,
it can be done. :)

As for "we don't need a mirror if we've got a dedicated site", I would
have agreed with you up until a couple days ago. Take a look at
http://www.linux.org/

As for an image server..if we have a dedicated powerful well-connected
system, we're not going to need to store images elsewhere. But if we do,
we'll probably want belegost rather than cran:
Server uptime: 32 days 13 hours 9 minutes 41 seconds
Total accesses: 188367 - Total Traffic: 2.9 GB

>Not going to work for us I suspect.  I suspect people will want to do
>development at their home machine and easily sync it.  Standard scp is
>hardly "easy" for syncing perposes of this sort.
>
>> b) store their repositories on their home machines, set their CVSROOT
>> to ':ext:seul@cvs.seul.org:/home/seul/cvsroot', set their CVS_RSH to 'ssh',
>> and run cvs normally from their home machines. It works pretty well, though
>> cvs has a couple small bugs in its code for this.
>
>Hmmm... this sounds like a possibility.  What does "a couple of small
>bugs" mean exactly?

Well, when we first started doing it the version of cvs we were using (1.8.x)
didn't talk well to other cvs's over the network. I believe 1.9 is behaving
more nicely. That just means we'll have to make sure people are up-to-date
with their cvs version.
Another 'bug' is that this technique opens an ssh per file interaction. This
means that if you make massive changes to your repository and check them in,
cran will sputter for a while and eventually you'll be done. This hasn't
been tested on the large scale before. Omega and I were pondering setting up
a netcat tunnel for this (meaning ssh opens a secure port and then cvs talks
through that port for all of its transactions, then closes it). If it becomes
an issue, we can probably implement this.

And yes, this system allows you to run all the commands on your local
machine. You never have to interact with cran directly, provided your keys
are set up correctly.

More info for this technique:
http://www.seul.org/sys/doc/dev-faq/ssh-cvs-FAQ.html
http://www.seul.org/sys/doc/dev-faq/dynamic-ssh.html

--Roger