1 |
On Fri, 2005-09-23 at 11:20 -0500, Lance Albertson wrote: |
2 |
> The idea I had (at least initially), was using the snapshot releng |
3 |
> builds for their release as our base tree to use for a release. After |
4 |
> they finalize the tree, I'd see a group of folks going through and doing |
5 |
> another round of QA and fixing any more bugs that may crop up. |
6 |
That sounds reasonable. I suggest that new versions of ebuilds should be |
7 |
~ARCHed so that dedicated updates are easier (no overlays etc.) |
8 |
> After its |
9 |
> been established as a 'good' tree, I'd see us releasing that as |
10 |
> something like 2005.1E or something like that. That part of a 'stable' |
11 |
> tree is relatively easy to do aside from any issues cropping up from the |
12 |
> QA section. |
13 |
Agreed |
14 |
> The real fun begins during the maintenance phase where GLSAs, and |
15 |
> critical software (data crippling type of things) updates need to be |
16 |
> updated in it. This is where we'd need to decide whether to go the back |
17 |
> port route, or just force folks to upgrade if the GLSA affects it. |
18 |
I'd say upgrade - backports would consume an extreme amount of manpower |
19 |
while testing still needs to be done by the end-user anyway. |
20 |
> Essentially, our group would be managing their own tree. It sounds |
21 |
> fairly simple, but to ensure a level of quality, we'd need to test the |
22 |
> new ebuild/upgrade in some type of QA environment (which we currently |
23 |
> don't have enough good man power for). |
24 |
That's a bit of a chicken-and-egg problem ;-) |
25 |
> Next, say when the next release comes out (2006.0E), we'd have to come |
26 |
> up with a clean upgrade path to ensure no breakages. This is the part |
27 |
> that will require a lot of time and effort. Ideally, it'd be nice if we |
28 |
> came up with a helper script to 'fix' things as the upgrade happens. |
29 |
So do we offer upgrades for 2005.1E or do we "force" people to upgrade at least bi-yearly? |
30 |
|
31 |
> Last, we'd need to decide how long to keep a particular tree updated. If |
32 |
> we went on two releases per year, after 2 years we'd have 4 trees to |
33 |
> keep up to date and come up with upgrade plans for each of those. |
34 |
I think 2 years is reasonable, any users that wish to keep a tree longer |
35 |
should be prepared to support that themselves (and if that only means |
36 |
hiring three gentoo devs fulltime ;-) ) |
37 |
|
38 |
> Needless to say, doing it the 'right' way isn't very easy to do. Thats |
39 |
> one of the things our server team would need to establish is what kind |
40 |
> of level do we want to maintain. Thats kind of where the idea of a third |
41 |
> party would come into play and possibly help fund a few folks to do this |
42 |
> kind of work as a job :). |
43 |
Agreed. In the beginning a "static" tree would be a nice start though |
44 |
> Thoughts on that rough plan? |
45 |
(1) Create a 2005.1-static tree |
46 |
(2) start adding GLSA updates as they come up |
47 |
(3) add new ebuilds ~arch so that updates are user-controlled |
48 |
(4) slowly announce our plans and pull in more helpers |
49 |
(5) create an extended, more precise roadmap ;-) |
50 |
|
51 |
wkr, |
52 |
Patrick |
53 |
|
54 |
-- |
55 |
Stand still, and let the rest of the universe move |