| 1 |
I'm trying to design a methodology to test major system upgrades. |
| 2 |
|
| 3 |
I'll give you the specific case, so that you can work through all the |
| 4 |
scenarios: Gnome 2.6 has been out for a while, and recently stable in |
| 5 |
Gentoo x86 [go foser!] However, gnome have released 2.6.1 of many |
| 6 |
things, which means *bugfixes*. THAT is in ~arch, and I'd like to be |
| 7 |
able to test it out - and then a) not upgrade from Gnome 2.4.2 b) use |
| 8 |
Gnome 2.6.0, or c) use Gnome 2.6.1 and make a positive report to Gentoo |
| 9 |
devs. |
| 10 |
|
| 11 |
I had the idea of copying the *entire* system to a spare partition and |
| 12 |
then chroot'ing into it, similar to the installation sequence. I did |
| 13 |
some tests, but although the building side of things works, X won't |
| 14 |
start. Weird {kernel,xfree,?} module issues. [Any idea why X wouldn't |
| 15 |
start from a chroot? I had dev, proc, sys bind mounted in] For various |
| 16 |
reasons, booting off of that device/partition isn't practical, and in |
| 17 |
any case, I'd like a way for the original system to stay extent while |
| 18 |
the test case is running. |
| 19 |
|
| 20 |
I also thought of User Mode Linux, which has a brilliant copy-on-write |
| 21 |
feature where you can start with an underlying system and then have it |
| 22 |
write changes, only, to another partition. That would also work in this |
| 23 |
case, except for the same difficulty - how can one get X running from |
| 24 |
within a UML instance? Does that even mean anything? |
| 25 |
|
| 26 |
Which leads me to wonder - is there a way of doing copy-on-write *not* |
| 27 |
in UML? Something like mount -o bind, but with COW. |
| 28 |
|
| 29 |
While buildpkg means that packages produced can be [re]used in the |
| 30 |
original system should things work out (and bind-mounting /usr/portage |
| 31 |
to the chroot worked out nicely to that end), the weakness with Gentoo's |
| 32 |
packaging system is that ./configure'ing everything means that packages |
| 33 |
down the dependency tree requiring upgraded libraries to be present, |
| 34 |
installed, on the build system. |
| 35 |
|
| 36 |
As we all know, backing out of something like a Gnome upgrade would be a |
| 37 |
significant nightmare, hence the desire to test and not be put in a |
| 38 |
situation of trying to back out. |
| 39 |
|
| 40 |
Again, I'm not worried about it *building* - I'm worried about specific |
| 41 |
use cases for our environment which I really need to test before |
| 42 |
installing. Using a spare box isn't an option as I travel a great deal, |
| 43 |
and in this case (ie need to personally inspect graphical) my laptop is |
| 44 |
it. |
| 45 |
|
| 46 |
Thoughts? |
| 47 |
|
| 48 |
AfC |
| 49 |
|
| 50 |
-- |
| 51 |
Andrew Frederick Cowie |
| 52 |
Operational Dynamics Consulting Pty Ltd |
| 53 |
|
| 54 |
Operations Consulting and Infrastructure Engineering |
| 55 |
|
| 56 |
http://www.operationaldynamics.com/ |
| 57 |
|
| 58 |
Sydney, New York, Toronto, London |
| 59 |
|
| 60 |
-- |
| 61 |
gentoo-dev@g.o mailing list |
Archives