1 |
On 29/09/2013 19:43, Tanstaafl wrote: |
2 |
> On 2013-09-28 6:46 PM, Neil Bothwick <neil@××××××××××.uk> wrote: |
3 |
>> Except you can never break Gentoo with a kernel update because, unlike |
4 |
>> some other distros, installing a new kernel does not uninstall the |
5 |
>> previous one. No matter how badly wrng a kernel update goes, you can |
6 |
>> always hit reset then select the old one from the GRUB menu - |
7 |
>> reinstallation doesn't come into it. |
8 |
> |
9 |
> My understanding is that this is not true, and that a USERLAND update |
10 |
> (LVM2, which I use, among them) can cause breakage that will cause the |
11 |
> CURRENT kernel+initramfs to no longer boot. |
12 |
> |
13 |
> Is my understanding flawed? |
14 |
|
15 |
No, this can happen in theory. It's quite simple to describe in somewhat |
16 |
abstract terms: |
17 |
|
18 |
Imagine for example that LVM makes a backwards-incompatible change to |
19 |
it's metadata. You are warned about this and take care to update your |
20 |
kernel so that it can deal with the new metadata by including support |
21 |
for both formats. |
22 |
|
23 |
And you forget to update the initramfs. Reboot. Oops. |
24 |
|
25 |
This is merely highly inconvenient, not the end of the world. Download a |
26 |
very recent rescue disk on another computer and boot with that to effect |
27 |
the repair. Then leave work and make your local publican's day whilst |
28 |
you vent your fury yet again |
29 |
|
30 |
Point is, this is not a situation unique to kernels, userlands and |
31 |
initramfs. That kind of error can occur in so many different ways (eg |
32 |
deploy a seriously broken linker and loader, or simply uninstall bash on |
33 |
a RHEL4 host), it's just that when it happens in the circumstances you |
34 |
ask about, it's one of the most inconvenient errors in a huge list. |
35 |
|
36 |
This is why we sysadmins have jobs - we are supposed to have subtantial |
37 |
clue and be able to predict and avoid such goofs. |
38 |
|
39 |
> Totally side question: Anyone ever hear Linus' opinion of an initramfs |
40 |
> being required to boot a system? |
41 |
|
42 |
Never read it myself, but I'll hazard a guess: |
43 |
|
44 |
He detests it with a passion calling it a grotesque hack, but tolerates |
45 |
it because binary distros need it and no-one has come up with something |
46 |
better (i.e. it sucks less)? |
47 |
|
48 |
|
49 |
-- |
50 |
Alan McKinnon |
51 |
alan.mckinnon@×××××.com |