Am 09.01.2014 19:58, schrieb Ian Malone:
On 9 January 2014 15:13, Toshio Kuratomi <a.badger(a)gmail.com>
wrote:
>
> On Jan 9, 2014 6:26 AM, "Chris Adams" <linux(a)cmadams.net> wrote:
>>
>> Once upon a time, Toshio Kuratomi <a.badger(a)gmail.com> said:
>>> <nod> Just have yum drop a config file in there that protects the
>>> kernel
>>> rather than protecting the kernel if some other package chooses to
>>> protect
>>> something else.
>>
>> The magic "don't delete the running kernel" can't be done with
just a
>> config file. Something has to detect which kernel version is running
>> and match it to an RPM, and then protect just that version of multiple
>> installed kernel RPMs.
>>
>
> Can't the meaning of a package name in the config file simply mean: "make
> sure one of these packages is always installed"?
>
> That won't protect the running kernel but it will protect a kernel (probably
> the latest installed). That would seem to address hreindl's use case of
> wanting to test on multiple systems and when satisfied that things are
> working cleanup all older packages.
Latest installed is almost exactly not what you want, I've had plenty
(where plenty in this case is probably >5) of cases where a kernel
update broke something, in quite a few of those cases to a state where
the system wouldn't boot. If the most recent one is retained then
you've still got a kernel, but not one that will actually run. With
current behaviour I can still let my system update until a fix appears
because I know it won't remove the good kernel. If updates can remove
the running kernel then you have to watch each one carefully. Unless
I've misunderstood this thread and this does not apply to automatic
updates.
as thread starter: you have *correctly* understood the intention
your example shows once more how important it is that the
kernel is treatet special, hence i even did not think about
this case while i was often happy about it because after some
years the current behavior is somehow self-evident