Re: [PULL REQUEST] Automatic kernel management

[Giovanni Campagna <scampa giovanni gmail com>]

2012/8/14 Colin Walters <walters verbum org>:
> On Tue, 2012-08-14 at 00:29 +0200, Giovanni Campagna wrote:
>> Unless ostree is used as the one and only system (which it can't,
>> currently),
>
> Right.  Now eventually we want to be building kernels, but the current
> design is fairly practical I think.
>
>> To help alleviate this, I made some handy scripts that
>> work in my environment, but should be general purpose (at least for
>> rpm based distros) enough to be included upstream.
>> Essentially they are invoked by rpm %post and propagate the kernel
>> changes to ostree. GRUB 2 is kind of assumed, although I kept existing
>> code for GRUB 1
>> You can find them at https://github.com/gcampax/ostree, the branch is master.
>
> https://github.com/gcampax/ostree/commit/cf4930e1babd415520f12f487f36ee93547ffe49
> Shouldn't the kernel updates being enabled default to true only if:
>
> 1) You're building with sysconfdir = /etc
> 2) You're on Fedora (does anything else
>    implement /etc/kernel/postinst.d?)

Yeah probably, but then, enabling in other cases does no harm (at
least for jhbuild), and automatic distribution detection would make
things more complicated.
Debian packagers can just --disable-kernel-updates.

> https://github.com/gcampax/ostree/commit/5defe144de55621861552dfab3d151c0a7313e93
> -  { "no-initramfs", 0, 0, G_OPTION_ARG_NONE, &opt_no_initramfs, "Don't generate initramfs", NULL },
> -  { "no-bootloader", 0, 0, G_OPTION_ARG_NONE, &opt_no_bootloader, "Don't update bootloader", NULL },
> +  { "no-kernel", 0, 0, G_OPTION_ARG_NONE, &opt_no_kernel, "Don't update kernel related config (initramfs, bootloader)", NULL },
>
> This bit will break ostbuild privhelper-deploy-qemu
> http://git.gnome.org/browse/gnome-ostree/commit/?id=b99b6be90492107416552a72f7536e1770183af2
> Not that that script is beautiful or anything, but I am using it now.

Well, it's trivial to change --no-initrams --no-bootloader to --no-kernel.

> As far as splitting off update-kernel...it certainly makes the code
> changes "noisier".  It's slightly unfortunate to re-exec ostadmin; we
> could split off update_kernel as an internal API.  But up to you.

I thinks it's easier this way, and the performance impact is negligible.

> https://github.com/gcampax/ostree/commit/d471ff5ee9018750f561966255198cac8cccbeaf
>
> I think a similar change was made in the src/ostree code which the
> ostadmin bits got copied from.  So this looks OK.

Awesome, pushed this part now.