[CentOS-devel] CentOS Plus: Valuable or wasted resource?

On Sep 11, 2009, at 8:05 AM, Johnny Hughes <johnny at centos.org> wrote:

> On 09/10/2009 11:11 AM, Akemi Yagi wrote:
>> On Thu, Sep 10, 2009 at 7:59 AM, Tim Verhoeven
>> <tim.verhoeven.be at gmail.com> wrote:
>>> On Thu, Sep 10, 2009 at 4:24 PM, Ross Walker <rswwalker at gmail.com>  
>>> wrote:
>>>> Is CentOS Plus really necessary?
>>>>
>>>> The reason I ask is it seems it is there to provide additional  
>>>> kernel
>>>> modules not included in the base repo that could be added to  
>>>> 'extras'
>>>> as kmods.
>>>> Let outside repos provide custom kernels.
>>
>>> First, the CentOSPlus is more then just the modified kernel. Stuff
>>> like the RHWAS stuff will also end up in CentOSPlus when its  
>>> released.
>>> It is basically a place where we put all things that conflict with  
>>> the
>>> things in the base OS repo.
>>
>> Yes. I'm sure Ross meant just the centosplus kernel, not the
>> centosplus repository.

I was concentrating on the kernel, I mean everything else in  
centosplus could conceivably go into 'extras'.

We definitely need a repo for 'extra' non-upstream items.

>>> Secondly, the kernel itself is maintained by Akemi who is also  
>>> part of
>>> the Elrepo team that builds kmods. So you can choose between
>>> installing kmod or using the centosplus kernel. And there are thing
>>> done in the centosplus kernel that can't be done using kmods. But I
>>> will let Akemi fill in the details.
>>
>> I have been helping with the maintenance of centosplus kernels for
>> about a year. And yes, I am involved in the ELRepo project as well (I
>> must admit that most of crucial works are the results of other
>> members' contributions).
>>
>> [OT] Just to defend myself ... I supply the src.rpm for the cplus
>> kernel usually within 24 hours for each kernel update. "is always
>> behind" may be a bit misleading ... :-)  [/OT]


To be honest I haven't tracked cplus in a couple of years and just  
remember the frustration I had back then when the kernel wasn't  
updated timely and third party kernel modules would get updated for  
newer base kernels or kernel modules in cplus wouldn't get updated  
when the cplus kernel was, you get the picture.


>> As Tim mentioned, enabling certain options for additional device
>> support is not the only thing cplus kernels offer.  There are patches
>> that are applied which cannot be done in the distro kernels.  For
>> example, there are currently 4 patches in the cplus kernel. Two of
>> them will be in the upcoming CentOS 5.4.
>>
>> I think that many, if not all, of the kernel drivers enabled in the
>> cplus kernel can / will be replaced by kABI-tracking kmods.  In fact,
>> Alan Bartlett made a list of such modules some time ago.  At this
>> moment, kmods for ntfs, ufs, jfs, reiserfs, and video4linux are
>> available from ELRepo.

The more the better, I think by reducing the changes needed to  
building the cplus kernel and trying to restrict it to just those  
patches to fix known problems with the base kernel, would be better.

>>> So, the same person doing the CentOSPlus kernel is also doing kmods.
>>> There is no waist of resources :-)
>>
>> Indeed, my _waist_ is just fine.  :-P
>
> Aside from Akemi's waist ... one other very important point that she
> makes is that for several of the plus kernels at several points in  
> time
> there are patches applied to fix things that are not fixed yet  
> upstream
> and that we can not fix in the main line kernels (as we do not apply
> patches to those).
>
> These patches can be for things like a drbd performance, NFS issues,  
> etc.

Yes, I remember submitting one such patch way back.

Ok, I agree, keeping a cplus kernel is worth while for patches to base  
kernel upstream is slow to incorporate, but support the move to bring  
additional functionality through the use of separate kmods instead of  
rolling them into cplus whenever possible.


> They are normally in the upstream testing process/bugzilla, but can
> sometimes take months to fix (or never be fixed ... like NTFS in the  
> 5.x
> kernel, we are not patching that now, but might later) upstream.

If a more recent NTFS kernel module can be backported and provided as  
a standalone that works, why bother.

There is also ntfs-3g which is good enough.

> I am also a bit hesitant in producing kABI type kernels for CentOS-4  
> as
> there is no tracking mechanism to require the correct version of  
> each of
> the base kernel requires as there is in CentOS-5, so weak updates can
> create problems for users in CentOS-4.  I am OK with that for things
> outside the main kernel, but for things that are part of the included
> kernel, I am more hesitant.

How about adding more support for dkms for C4?

> I do like the fact that these are also in ELrepo and give users  
> another
> option for installation if they want to go that route.

Ok, so is ELrepo the "officially supported" third party repo for  
CentOS, like epel is for RHEL?

I don't like the fact that epel doesn't tag, so I no longer use them.

-Ross