It can be done if you mess with the initramfs.

The kernel starts everything else by unpacking an archive containing a minimal environment to set stuff up for later. Such as loading needed kernel modules, decrypting your drive, etc. It then launches, by default, the /init program (mines a shell script).

That program is PID 1. If it dies, your kernel will panic.

After it finishes setup, it execs your actual /sbin/init. These means it dies, and that program (systemd, openrc, dinit, runit, etc) becomes PID 1. If an issue happens, both could fail to execute and the kernel will loop forever.

“Meant to” and “do” are quite different.

I have, not once, ever had an issue caused by a update while the system was online. And if someone did, they could likely reboot to fix it.

I will stick to the online, instantaneous, and unobstructing updates.

I would like to see some real-world numbers and examples of problems that offline updates fix.

Its possible: Link