1 |
> There always is legacy. |
2 |
> |
3 |
> That fact has been a touch absent in these discussions. If we're really |
4 |
> smart, we won't just create a gentoo-gold-server which is only suitable |
5 |
> for use in networks of gentoo machines, we'll create a master server |
6 |
> which runs gentoo and which is suitable to act as gold server for a |
7 |
> whole suite of machines, regardless of OS. |
8 |
> |
9 |
|
10 |
Excellent. Paying attention to legacy means less work for myself; I would |
11 |
have to support some of this crap anyway. There are probably dozens of |
12 |
cases where large companies are running programs they don't even have the |
13 |
source code for anymore! Making these legacy systems at very least |
14 |
manageable should be an early goal. I cannot imagine something like this |
15 |
getting popular without that kind of edge into a existing networks. |
16 |
|
17 |
Like the bootstrapping doc says (I'm going to finish reading this site |
18 |
again on the ferry to work tomorrow...), we can get configs into |
19 |
versioning systems for any architecture. Packaging is icing on the cake. |
20 |
If I can get my sunboxes to accept updates (depending on what version |
21 |
they're running) for security fixes from my gold server, as well as having |
22 |
my redhat 6.2 boxes pull in a few extra perl modules... That's |
23 |
flexibility for you. |
24 |
|
25 |
Generally, the architecture would benefit from allowing a lazy-caching |
26 |
system for the different OS/hardware classes, with the usual set of |
27 |
options. Max cache size, time-to-live, and even an option to mask packages |
28 |
to never expire unless a newer version is available. The old sun and linux |
29 |
2.2 machines might not get upgrades very often, but you could be launching |
30 |
5+ newer gentoo (or debian) servers a week, and can't be stuck downloading |
31 |
when an upgrade to a legacy system pushed your packages out of cache. |
32 |
|
33 |
But this is all getting into details I'm not sure we're ready to pass yet. |
34 |
The focus of my meager extra ounces of brainpower is toward this class |
35 |
tree idea, and how to get even more basic basics down (tools and docs for |
36 |
getting gold-server CVS, and example scripts/tools/docs for many |
37 |
OS's/distributions for config pull clients). |
38 |
|
39 |
As far as packages... |
40 |
Hardware, OS, software, ? - Three classes in total, as you said? It's easy |
41 |
enough to see where they cross, and how they might be organized. |
42 |
|
43 |
There're quite a few ways of thinking about this, but my only personal |
44 |
requirements are in simplicity and elegance. Our gold servers need to |
45 |
maintain systems from years back to now, but also need to be able to |
46 |
change hands... I'd be sad if I spent months getting something like this |
47 |
going at work, then half the staff changes over and they start building |
48 |
new servers outside of the gold's control, since they don't have the time |
49 |
to learn how to manipulate it. |
50 |
|
51 |
-Dormando |