On 2/21/2012 5:57 AM, Boris Epstein wrote:
> Things like boot process rarely break.
I can't remember the last time I caused a system to outright fail to
boot, but I *do* get unclean boots regularly.
Examples:
- Build and install some needed driver from source, yum upgrade
repeatedly, implicitly upgrade kernel, forget to rebuild the driver
against the new kernel, reboot, boom.
- Get asked to configure the foo service, get it all working, forget to
add it to init.d, use it happily for months, reboot, fail to notice the
service's absence until someone gives a misleading bug report. ("The
foo service has crashed!") Then I have to go chasing it, handicapped by
being half a year separated from the last time I looked at it.
What with the long uptimes on our servers and their many-hattedness, I'd
say chances are actually fairly low that any given reboot will bring it
100% back to the state it was in when we shut it down. Something
boot-related almost *always* changes between reboots.
I've made it a policy that if you get an unclean reboot, you fix it in
such a way that it should come up cleanly, then test by rebooting.
Repeat until all symptoms are fixed. But, never believe you've thereby
guaranteed that it will come up cleanly the next time.