[elrepo] 4.14 LTS

[Phil Perry]

On 14/02/18 16:54, David Ranch wrote:
> 
> This is an interesting point.
> 
> I'd argue that in many respects, the ElRepo group is one of the primary 
> reasons I've stayed on with Centos as it gives me newer kernels than 
> what Redhat/Centos will.  Packages is an entirely different discussion 
> and I digress.  Understanding all that, the 4.4.x kernels are already 
> quite old.  I don't have a good understanding of all the work that goes 
> on behind the scenes for the ElRepo team to release new LT and ML 
> packages but would it be possible to add more LT kernels?  Maybe not all 
> three of them but maybe 4.4 (for the conservative people), 4.14 (for the 
> people who need a long life yet want a mostly modern kernel, and then 
> the ML line for the bleeding edge users?  I know this becomes more and 
> more additive over the years but maybe a little more can be done to keep 
> Centos up to date?
> 
> --David
> 

Alan is the only one able to give a definitive answer on this, as he is 
the one who does all the hard work, but seeing the amount of work it 
takes to maintain two package sets over 2 distros (was 3 until el5 was 
recently retired, and will be 3 again once el8 is released), and 
multiple arch's, I'd say it's very unlikely.

I'd also question the rationale. kernel-ml is the cutting edge offering, 
and kernel-lt is the LTS offering based upon what is available at the 
time and has LTS support. A key attribute of Enterprise Linux/LTS is we 
provide version stability; we do not change it unless we absolutely have 
to. That is a fundamental cornerstone of the concept of Enterprise Linux.

I appreciate kernel-lt-4.4 may be starting to look long in the tooth in 
some areas, but that is inevitable with any LTS kernel. Interestingly, 
there will reach a point where what started out in life being a newer 
offering (kernel-lt vs the distro kernel) will actually end up reverting 
to the complete opposite. For example, consider wifi support and the 
wifi stack. The EL7 distro kernel started life as 3.10. The wifi stack 
has undergone various backported updates through 3.16 in el7.1, 4.1 in 
el7.2, 4.7 in el7.3, 4.11 in el7.4 and most recently 4.14 in the 
el7.5(beta) kernel, so we see in this respect the distro kernel actually 
becomes 'newer' than kernel-lt by the el7.3 release. This will no doubt 
be happening in other areas as well, so I would fully expect some 
kernel-lt users to revert to using the distro kernel again as time passes.

By the time kernel-4.4 is out of LTS, a successor will be chosen to 
replace it based upon what is available and most suitable at the time, 
and will hopefully see out the 10 year life of the distro. Also, I 
consider kernel-4.14 a poor choice for LTS at present given there is 
only a commitment upstream to support for another 2 years whereas 
kernel-4.4 has a commitment to support for another 4 years.

Ultimately, elrepo aims to offer some extra choice in a convenient 
packaged format where there was none before. Given the amount of work 
Alan puts in to developing and maintaining these packages, I would say 
he is operating at his limit in providing the 2 package sets currently 
available, and given that, I consider the current offerings to be a good 
compromise for the reasons outlined above. That said, if anyone wants to 
maintain a kernel-lt-4.14 offering, they are free to do so given Alan 
has already done a lot of the development work. One could take his last 
kernel-ml-4.14 SRPM, rename to kernel-whatever-4.14 and drop in the 
latest 4.14.x tarball and maintain it for the rest of the 4.14 branch 
life, assuming one has suitable el6/7 build systems and the willingness 
to commit to regularly building a relatively large numbers of kernel 
releases.

In the meantime, if Alan has a differing view, I'm sure he will tell us :-)

Phil