Upgrade and reload kernel without rebooting in Slackware?

Richard Cranium · 04-21-2019, 07:36 PM

Quote:

Originally Posted by upnort

How did you keep both the host and guest systems synchronized between the two?

Normally, by both of them looking at the same database cluster.

chris.willing · 04-21-2019, 09:13 PM

Quote:

Originally Posted by linus72

Would kexec work here?
I know distros like NetBootCD use it....

https://wiki.archlinux.org/index.php/kexec

https://linux.die.net/man/8/kexec

https://slackbuilds.org/repository/1...m/kexec-tools/

The default 14.2 kernels are not configured with CONFIG_KEXEC=y as is required to use kexec (of course you could build custom kernels to fix that). However it is set appropriately in -current where I just tested in a VM and it worked pretty well, although the change over is not instantaneous (about 20secs for new kernel to boot & services to start).

Compared with livepatch (which looks complicated at first glance), kexec is very easy to setup and run although the kernel changeover time wouldn't be good enough for cases like an online store where you may have purchase transactions going on (and don't want to interrupt).

Edit: PS. the existing SlackBuild for kexec-tools at SBo uses version 2.0.17 but this generated an error (kexec unhandled rela relocation R_X86_64_PLT32) when I tried to run it. Updating to version 2.0.19 fixed the problem.

chris

enorbet · 04-22-2019, 12:29 AM

Regardless of the expense, assuming it is THAT important to be up without even a few minutes lapse, ever, that is worth money as well as an investment into the only practical and effective solution - redundancy. Replacing a kernel on-the-fly does not solve the issue since what does it matter if the system is technically up in some manner if no or even limited services can be provided? It's just cleaner, and ultimately more reliable, to have scheduled reboots for various kinds of maintenance.

This wouldn't be about supposed advantages of a parallel init would it?

upnort · 04-22-2019, 10:24 AM

Quote:

Regardless of the expense, assuming it is THAT important to be up without even a few minutes lapse, ever, that is worth money as well as an investment into the only practical and effective solution - redundancy.

I agree with the sentiment but I can share that in the real world this often does not happen. I appreciate that in some use cases five 9s of uptime or better is not a goal but a requirement. People in such use cases should indeed accept that investing in some kind of redundancy is simply the cost of doing business. That said, I work for a small mom-and-pop. Striving for five 9s of uptime is impossible and stressful. Two 9s of uptime is good enough for most people.

Geremia · 04-30-2019, 10:46 AM

Quote:

Originally Posted by glorsplitz

Maybe see here Livepatch: Linux kernel updates without rebooting?

Although slackware is not listed, it seems on my stable slackware there's this

Code:

zcat /proc/config.gz | grep LIVEPATCH
CONFIG_HAVE_LIVEPATCH=y
CONFIG_LIVEPATCH=y

What's yours say?

Code:

CONFIG_HAVE_LIVEPATCH=y
CONFIG_LIVEPATCH=y

I also have a /sys/kernel/livepatch directory.

Is there a SlackBuild for a livepatch client?

glorsplitz · 04-30-2019, 08:01 PM

Then there's this

Canonical livepatch does not work?

Quote:

Quoting from the Livepatch wiki page:
Since there are limitations to the kernel livepatch technology, some Linux kernel code paths cannot be safely patched while running. There may be occasions when the traditional kernel upgrade and reboot might still be necessary.