4 ms·

I'm pretty sure I've updated the kernel on the fly when trying to update the kernel on Digital Ocean-- the stock kernel loads (let's call it the bootstrap kerne

I'm pretty sure I've updated the kernel on the fly when trying to update the kernel on Digital Ocean-- the stock kernel loads (let's call it the bootstrap kernel) and then it eventually calls some code that replaces the running kernel with a newer more modern version. I've forgotten the details, as it sounded kludgy, but I'm pretty use like anything else Linux, a reboot isn't always required. In fact, with Nano server, Microsoft is trying to go that direction too -- the less you have running, the easier to avoid reboots. But the real solution is trying to be reboot-tolerant by having a slight bit of redundancy, I would think. Uptime matters more for pet/snowflake servers than for cattle....

pritambaral 11y ago
> I'm pretty sure I've updated the kernel on the fly when trying to update the kernel on Digital Ocean What you go on to describe sounds a lot like kexec, and a kexec is the same thing as a reboot for everything on a computer except the hardware. The new kernel during a kexec starts with a freshly-booted state. What I meant was – and perhaps I wasn't clear enough – updating the kernel in-memory, while it runs so that even your userspace apps notice nothing - except perhaps a few fractions of seconds of time slipping by - and continue running while the kernel they're running on top of is patched with new code. This has been possible for a while on Linux. It began with, IIRC, Ksplice as a college-project, then a startup, then bought-out by Oracle for Oracle Linux, then independent efforts at similar implementations by both Red Hat and Suse, which eventually got merged together and mainlined into Linus's tree.