APT -- if I do this will I screw the pooch?
Got a question relating to APT. I typically use aptitude in command-line
mode as a front-end to APT. I'm running Sarge kept up to date with
security updates, on a single-processor Pentium 4-based desktop machine.
My kernel is a self-built 2.6.15.4 built with make-kpkg from kernel.org
source.
I'm planning shortly to add a second Debian machine, a notebook, to my
home network. I intend to keep the versions of Debian on the two
machines, my current desktop and the new notebook, basically in step but
don't want to be forced to ie if considering a major upgrade, want to be
able to do one first, play around with it, then do the other.
I don't have up to date install CDs and don't want to download them, so
I was thinking of setting up the notebook by doing a basic install off
my old woody CDs I originally set up my desktop off of, getting a
working net connection in place, then immediately upgrading everything
via the net to current stable before building out the system any
further. Since I've already gone through the upgrade process, over the
last 2 years, with my desktop machine I have all the packages downloaded
in my /var/cache/apt/archives directory. I don't want the notebook to
have to download them all again, especially since security.debian.org,
which has a lot of the latest versions of stable packages, seems to get
overloaded and be very slow to download at times.
After reading the man pages on apt, apt.conf, aptitude etc, and googling
around, here's what I am thinking of doing. First of all, exposing my
desktop's /var/cache/apt/archives as an NFS mount, configuring my
gateway's firewall to prevent anyone mounting it from outside my LAN.
Mounting this NFS mount on the laptop, let's say at /mnt/aptarchive.
Then, setting APT config option Dir::Cache::archives on the notebook to
point at /mnt/aptarchive, *without touching the other Dir::Cache
options*. This is because although I want them both to use the same
repository of downloaded package files, I want the two machines to
independently track what's installed, so I don't have to keep installed
packages identical on both machines.
Here's what *I think* the effect of this will be:
1. The notebook, when asked to dist-upgrade and later to install
packages I didn't install from the woody CDs, will not have to download
the package from outside my network because its archive already contains
the package (always assuming that the package concerned is installed on
the desktop, which 90% of what I want to install on the notebook will be).
2. The notebook and the desktop will have independent views of what's
installed, and will each know nothing about what's installed on the
other machine (which is fine).
3. If I want to install something on the notebook that's not installed
on the desktop, the notebook will download it as normal and put it in
the same archives location.
4. This means that if I then later decide to install that package on the
desktop, it will find it already has the package and not download it
(assuming a newer one hasn't become available in the meantime).
5. Installing a package on one machine will have no effect whatever on
the other machine's notion of whether or not that package is installed.
Reading the APT documentation, it seems to encourage using squid to set
up a local caching download proxy to achieve this goal, setting APT to
use the proxy when downloading packages. I don't like this approach as
it would mean having at least 2 copies of the downloaded packages (on
the desktop and on the notebook, if I can persuade squid to use the
desktop's /var/cache/apt/archives dir as its cache dir) or maybe even 3
(squid's repository, the desktop's /var/cache/apt/archives, and the
notebook's /var/cache/apt/archives if I can't get squid and the
desktop's APT to play nice). Since the point is to avoid extra downloads
and also space wastage, I'd rather go with my approach.
What I'm looking for is input from more experienced hands with APT on
whether I will screw anything if I do what I'm proposing, sharing the
archives dir between 2 machines without sharing the package cache and
source cache files. The documentation is somewhat ambiguous on whether
or not what I'm proposing is possible.
I *could* just try it, but I don't want to utterly screw my desktop if I
can avoid it...
Mark
Reply to: