LWN (formerly Linux Weekly News) reports
Canonical has announced the availability of a live kernel patch service for the 16.04 LTS release. "It's the best way to ensure that machines are safe at the kernel level, while guaranteeing uptime, especially for container hosts where a single machine may be running thousands of different workloads."
Up to three systems can be patched for free; the service requires a fee thereafter. There is a long FAQ about the service in this blog post; it appears to be based on the mainline live-patching functionality with some Canonical add-ons.
Another distro, not wanting to be left out of the recent abundance of limelight has made some noise of its own.
Phoronix reports
KernelCare Is Another Alternative To Canonical's Ubuntu Live Kernel Patching
The folks from CloudLinux wrote in to remind us of their kernel patching solution, which they've been offering since 2014 and believe is a superior solution to Canonical's service. KernelCare isn't limited to just Ubuntu 16.04 but also works with Ubuntu 14.04 and other distributions such as CentOS/RHEL, Debian, and other enterprise Linux distributions.
Another big difference to Canonical's Livepatch is that KernelCare does support rollback functionality while Canonical doesn't appear to support it at this time. KernelCare can also handle custom patches, 32-bit support, and they share they plan [sic] to soon begin offering livepatching support for glibc, OpenSSL, and QEMU.
The downside though is that KernelCare appears to rely upon some binary blobs as part of its service. Pricing on KernelCare ranges from $25 to $45 USD per year depending upon the number of licenses being purchased.
[Details at] CloudLinux.com.
(Score: 4, Insightful) by VLM on Wednesday October 26 2016, @07:03PM
Just take those systems down one at a time and plug in the new patched systems.
The biggest professional problem I have is legacy people who don't understand the game has changed.
So in the old days you did forklift upgrades moving entire racks of hardware in and out of centers and you can't physically install both at the same time and switching back and forth takes like a full working day. I did stuff like that for decades.
So the business people get used to 1am maint notification complete shutdown for 7 hours until 8am and god help me they code it into procedures or even law.
Now I roll systems by flipping one IP address in a load balancer at a time and its best if I do it during prime time to find out if it works and if it doesn't, first why wasn't it caught in DEV or TEST but second I flip an ip address and we're back. And I can roll 20 systems in 5 minutes or 5 months if you insist on slow methodical conversion.
But the business people just don't understand it, although they remember getting burned back in the forklist days, so it leads to a lot of stupid behavior that would have been useful in 1986, not so much in 2016.