[CentOS-virt] Xen-kernel: Update to 4.14 or 4.19?

Thu Mar 7 19:59:18 UTC 2019
Kevin Stange <kevin at steadfast.net>

On 3/7/19 12:55 PM, Karl Johnson wrote:
> 
> 
> On Thu, Mar 7, 2019 at 1:42 PM Sarah Newman <srn at prgmr.com
> <mailto:srn at prgmr.com>> wrote:
> 
>     On 3/7/19 10:30 AM, Akemi Yagi wrote:
>     > On Thu, Mar 7, 2019 at 9:42 AM George Dunlap <dunlapg at umich.edu
>     <mailto:dunlapg at umich.edu>> wrote:
>     >>
>     >> Hey all,
>     >>
>     >> We've been on 4.9 for some time now, and while it's still
>     supported, I
>     >> think it's time to start thinking about upgrading, and I'd like input
>     >> from the community about which version to move up to.
>     >>
>     >> 4.19 has been out for almost 5 months now.  It will include PVH domU
>     >> support, and PVH dom0 support in what _is believed_ to be the final
>     >> form; so when the Virt SIG moves to a version of Xen that
>     supports PVH
>     >> dom0, the kernel will already be in place with no need to upgrade.
>     >>
>     >> The other option would be to move to 4.14: Probably more stable (as
>     >> it's been out for over a year now), but doesn't have either PVH domU
>     >> or PVH dom0 support.
>     >>
>     >> I'd suggest 4.19. Any other opinions?
>     >>
>     >>  -George
>     >
>     > You may also want to consider each version's EOL:
>     >
>     > 4.9   Jan, 2023
>     > 4.14   Jan, 2020
>     > 4.19   Dec, 2020
> 
>     Regardless of EOL date, I think it's worth trying to upgrade when
>     Xen has stable PVH dom0 support.
> 
>     I am pretty sure historically that there have been difficulties
>     backporting some of the side channel mitigations as they can be
>     quite invasive. That
>     may be another reason to upgrade sooner rather than later.
> 
>     --Sarah
> 
> 
> +1 for 4.19. However, this version requires a recent GCC version so it
> wont build at least for el6 on the CBS. We would have to build them with
> recent GCC from devtoolset like I do in my pull request (gcc 7.3.1).
> 
> Karl

I am +1 for 4.19 as well and I agree with Sarah's reasoning that we'll
want stable PVH dom0 support as soon as it's reasonable.  However, I had
serious stability issues with 3.18 in the past and I would want keep a
major kernel bump in the testing repo for 3-6 months before moving it to
release.  I will do as much testing as I can during that time to
establish stability on my side.

It might make sense that we just bump to 4.19 for EL7 to avoid the
complications related to devtoolset on EL6. 4.9 lasts the entire
remaining lifetime of EL6, but will come up slightly short of EL7's.
However that means bumping two divergent kernels periodically for each
set of repos.

Based on recent history (4.4, 4.9) we can probably expect both 4.14 and
4.19 to become 6 year kernels extending to Jan 2024 and Dec 2024
respectively, though GKH seems to like to wait until close to the
original EOL to announce these decisions.  We can likely also expect a
kernel like 5.3 to end up becoming longterm around end of 2019.

-- 
Kevin Stange
Chief Technology Officer
Steadfast | Managed Infrastructure, Datacenter and Cloud Services
800 S Wells, Suite 190 | Chicago, IL 60607
312.602.2689 X203 | Fax: 312.602.2688
kevin at steadfast.net | www.steadfast.net