The old Unix hacker in me likes Ksplice, from a hacker point of view.
However, I wouldn't use it, except in exceptional circumstances (e.g., there is a patch we have to apply now but we can't do scheduled down time). I want to reboot after an update--even an update that does not technically require it--because I want to know my system is in a state that can boot.
It is very annoying to do some update, not reboot, and then a couple of months later a power failure takes down your systems and when power is restored you get some kind of boot error and you are wondering what in the last several months you did that broke it.
A few minutes planned downtime now in much better than hours of unplanned downtime at some indeterminate future time.
> A few minutes planned downtime now in much better than hours of unplanned downtime at some indeterminate future time.
This reminds me of something that OpenSolaris did that I really liked. When you updated packages, it would take a ZFS snapshot. It only took up space for the changed bits, and happened almost instantly. If things stopped working after the reboot, then you could select a GRUB menu option to boot into the old snapshot again and everything would work.
1. You upgrade postgres, and the new on-disk format isn't compatible with the old one.
2. You upgrade mysql, and the new on-disk format IS compatible with the old one, but the new version has a bug that you keep hitting. You want to downgrade without losing all the transactions that you've accrued since the upgrade.
This is why I can't stand Gentoo. We had a dude who was the Sysadmin before me who loved it. He upgraded the kernel deleted the old one and didn't reboot. Several months later the server had to be shutdown and didn't come backup. I spent several hours fighting to get the machine to a point where I could rebuild a working kernel.
Within a week I prepared and migrated the system to something more reasonable.
It seems to me that your kernel issues are more related to your predecessor's failures as an admin than to Gentoo. Nothing you can do on Gentoo can't be done on most distributions, including poorly-done kernel upgrades.
Nah, it's gentoo that sucks. I don't see anything like that happening in binary distributions, or assembling anything from ports in FreeBSD.
I've administered a few gentoo machines a few years ago and can attest that it's a major PIA. Or let me elaborate a bit, system failures that happened on gentoo didn't seem to happen on other systems, system failures that also happened on other system took significantly more time to resolve and required much deeper knowledge of system internals than debian, centos or whatnot.
When I finally moved everything to debian and my time spent on system administration tasks went down dramatically.
I know exactly what you mean about time spent on admin tasks in Gentoo; I myself experienced the same when I switched my personal dev machine from Gentoo to ArchLinux. I still wouldn't blame Gentoo for the kernel issues, however.
It seems to me that Dobb's predecessor failed to understand what he was doing and therefore failed as an admin. If Gentoo required a level of understanding which he did not have, then he should not have been using Gentoo. That is exactly the reason I switched; I realized that I needed to spend time with other distributions learning more basic elements of Linux systems before I would be capable of adeptly administering a Gentoo system. In the hands of those more experienced, however, distributions such as Gentoo (or Slackware) are powerful tools.
However, I wouldn't use it, except in exceptional circumstances (e.g., there is a patch we have to apply now but we can't do scheduled down time). I want to reboot after an update--even an update that does not technically require it--because I want to know my system is in a state that can boot.
It is very annoying to do some update, not reboot, and then a couple of months later a power failure takes down your systems and when power is restored you get some kind of boot error and you are wondering what in the last several months you did that broke it.
A few minutes planned downtime now in much better than hours of unplanned downtime at some indeterminate future time.