On Fri, 20 Jul 2007, R P Herrold wrote:

> On Fri, 20 Jul 2007, Troy Dawson wrote:
>
>>  Just out of curiosity, since I don't use it, what do you usually set for
>>  the default number.  In the past, I had it set to 2.  Is that reasonable,
>>  or would people rather have it higher? like 3 or 4?
>
> The centos group have discussed this -- with the pace of kernel updates, 
> particularly near the start of a release of a new Major, it is inprovident to 
> think that end users running yum on a cron autopilot will reboot when new 
> kernels are walked in.

Personally I think that *some* updates should be marked 'reboot needed' 
since carrying on running with the old version means that tools will say 
'no new updates are needed' but the machine is still running with the 
old/insecure/broken version.

> This is not an issue, UNLESS a new kernel comes in, and the plugin has 
> removed the modules for the running kernel, such that 'loop' cannot be loaded 
> (ie, if it is not present).

The installonlyn plugin already has checks to make sure it doesn't remove 
the current running kernel...

...
                     (curv, curr) = get_running_kernel_version_release()

                     installed.sort(comparePoEVR)
                     for po in installed:
                         if (po.version, po.release) == (curv, curr):
                             # don't remove running
                             continue
...

> mkinitrd is fragile enough, and yum's error detection historically 
> insufficiently granular, that I worry that a number of people are going to be 
> going to recovery media ;(
>
> Throw in non-stock controller kernel modules, and a trainwreck seems to be on 
> a horizon.
>
> Dialling up to say at least 4, if leaving it enabled at all, seems prudent, 
> to let others be the pioneers -- you know - the ones with the arrows sticking 
> out of their backs  ;0.

Or just don't install it on such systems...

Note that installonlyn has been installed and on by default in Fedora-Core 
for quite some time (FC3 I think).  If it broke lots of things it would 
have been very widely reported by now.

> - Russ Herrold

-- 
Jon Peatfield,  Computer Officer,  DAMTP,  University of Cambridge
Mail:  [log in to unmask]     Web:  http://www.damtp.cam.ac.uk/