Re: Porteus, Sounds great, suggestion

Mark Guertin <[email protected]> Thu, 10 Jul 2003 13:55:22 -0400
Newsgroups gmane.linux.zynot.devel
Message-ID <BB3320CA.88A%[email protected]>
On 7/10/03 1:31 PM, "Zach Welch" <[email protected]> wrote:

> These features could be easily added by creating a client/server
> architecture for the package manager. A front end would send requests to
> a server (doesn't have to be long-lived), which would process the
> requests in the appropriate order.  Such an engine could then serve for
> our autobuilding system, and possibly even for a P2P auto-building
> system for stageballs and ISOs.
> 

If you are talking locally (local client and server) this might be an
option, but if you are talking a real client/server relationship on a
network there are many things that are hard to do here...

This is a nice theory for a network, but in implementation there are a _lot_
of things that would need to be worked out and changed/rewritten/totally
redesigned for this sort of thing to happen.

We would basically have to write an entire harness suite to do this inside
multiple chroots, etc.  As for random pkg queuing/building this would be
very painful because to build for a client on demand we would have to either
have a preset environment for groups or classes of machines that they are
not allowed to deviate from, or we would have to completely replicate the
client's entire hard drive inside a chroot to accomplish this (thanks to
things like USE).

It would be more feasible to do things 'GRP' styled, where a server builds a
preset and matched set of pkgs and they are distributed as part of the
install process and/or for updates that happen afterwards.  Anything outside
of this scope won't work within current (or planned for the near future)
designs due to the nature of how things are laid out.

As for P2P building etc, again this opens a huge scope of things, the first
and foremost being security.  I have been fairly vocal in the fact that I
would _never_ in any way shape or form condone a P2P implementation onto any
of my networks or workstations I run.  A lot of other people feel very
strongly on this as well and I think we need to rethink a lot of the P2P
discussions that have happened to date.  This is a very touchy spot for
corporate environments, I don't know any sysadmins offhand that would allow
this sort of things to happen in their setups.  It's opening your system
completely to anyone on the network who figures out how to circumvent any
protection that may be in place.  Just the thought of it makes me shiver,
hehe.

This could be taken for future consideration, but I would be a lot more
hesitant to say they can be "easily added", nothing easy about it :)


> While these are very long term goals, I also want these features and
> even know how to implement them... now we just have to spec it.... ;)

See above for the most basic parts of what we need to think about before
even starting to spec something like this. :)

Mark