Nov. 22, 2013
7:50 p.m.
On Fri, 22 Nov 2013 12:21:59 +0400, Vasiliy Tolstov said:
After some time i need to update kernel. How can i deal with it with minimal downtime not using ksplice? Kexec?
The usual stock answer is "run two instances with some sort of High-Availability package between the two of them". Fail over to one instance, reboot the second, then fail over to the now-rebooted second instance and reboot the first. I've done that on SGI boxes running CXFS and NFS - the NFS client machines see only a 2-3 second pause. (The bonus is that since the two boxes do heartbeat, it also means that if one locks up or crashes, the other can seize control and continue processing without users noticing. So it's of benefit not just for updating the kernel.