Mailinglist Archive: opensuse-features (39 mails)

< Previous Next >
[openFATE 312647] Preserve Running Kernel On Kernel Update by Default
Feature changed by: Stephan Barth (cyberiad)
Feature #312647, revision 8
Title: Preserve Running Kernel On Kernel Update by Default

openSUSE Distribution: Evaluation by engineering manager
Priority
Requester: Desirable

Requested by: Robert Davies (robopensuse)
Partner organization: openSUSE.org

Description:
My suggestion is to enable multiversion kernel & preserve running
kernel for safer kernel updates, to reduce support problems when new
kernels won't boot. This is now possible for 12.1 by 1 line change to
zypp.conf.
Since Online Update was first introduced (SuSE 7.1?) it replaced the
running kernel by default, which can lead to problems booting. For a
while multiversion kernel has been supported by libzypp (1 line edit in
zypp.conf). The drawback with multiversion was manual deletion required
of unwanted kernel packages (see
http://lists.opensuse.org/opensuse-kernel/2011-07/msg00056.html). Now
in Factory 12.1 M3 Michel Marek has included & announced - kernel
package retention options (see
http://lizards.opensuse.org/2011/07/14/improved-kernel-package-retention-in-12-1/
).
The default is sane, to preserve the Lastest & Running Kernels. So lets
use it and make 12.1 kernel updates safer for all, and by default do
the right thing!

Use Case:
End user upgrades kernel to Tumbleweed and it fails to boot.
With kernel package retention, the running kernel was not deleted
before reboot and recovery is simply a matter of choosing the "good"
kernel from GRUB menu and reporting the issue, rather than figuring out
system recovery.

Business case (Partner benefit):
openSUSE.org: To reduce support caused by kernel upgrade "accidents".
To encourage safer testing of Kernel Team "stable" & "HEAD" kernels.
To correct a long standing risky choice, made for implementation
convenience of deleting old kernel package, before the new was tested.

Discussion:
#1: Michal Marek (michal-m) (2011-07-18 16:56:15)
Note that there are multiple ways a kernel can "fail" for the user.
Right now, the purge-kernels script is run after boot.localfs, that
does not help you if e.g. X doesn't work anymore. To be fully bullet
proof, the script would need to be triggered much later, e.g. once the
user logs in and there is network connectivity. But even the current
implementation can help in many cases.

#2: Robert Davies (robopensuse) (2011-07-20 16:24:52) (reply to #1)
So the issues of module loading failing, after kernel update would be
solved. Many hardware boot issues where udev hangs also solved. For X11
not starting, even simply altering the default set for "purge-kernels"
to : multiversion.kernels = latest,latest-1,running
or :
multiversion.kernels = latest, oldest, running
Would preserve 2 "working" kernels after an update; without any coding
changes. The oldest being treated as "golden" generally installation
kernel without user intervention.

+ #3: Stephan Barth (cyberiad) (2012-05-11 16:06:41)
+ The Linux backline people teams from EMEA and US discussed this on
+ their regular meeting on Wednesday two weeks ago and we agreed that we
+ propose that too for the next SLE Service Pack. It would be great to
+ have this, because boot issues with an updated kernel have several
+ downsides:
+ - they come in as high severity - customers are often not capable to
+ use the rescue system or other tools - the variety of issues is to big
+ to write a simple troubleshooting guide - fixing issues over the phone
+ or chat takes a very long time when directing customers through steps
+ to get the system fixed




--
openSUSE Feature:
https://features.opensuse.org/312647

< Previous Next >
This Thread
  • No further messages