Cutia said:
Thanks Jeff, your advice is appreciated. My main concern is how to do anything like that remotely. I'm guessing I'd have to put a request in with my provider LayeredTech to go into the bios and disable the dead drive.
You could do that. You could also have them install a remote KVM solution for you, or if your server has it available in the BIOS, set up the server so all startup commands (including the BIOS change command) go out the serial port, and then install a serial connection you can use from your desktop.
LT recommend using a RAID1 array of two drives for a no hassle backup solution.
If they really said
no hassle then they're being over simplistic and perhaps their recommendations shouldn't be trusted without discussing everthing with them in detail first
.
For example, I'd ask them how often RAID fails for spurious reasons compared to actual drive failure (answer: much more often).
I'd ask them how often RAID fails and both drives are bad, compared to how often only one drive is bad (answer: at least occasionally).
(Sometimes RAID can actually take down the second drive when the first fails. It's known to happen.)
We recommend RAID for everyone, but as part of a solution, not a complete or only solution.
This is ok but it means renting a RAID card which is not that cheap.
I can't speak for FreeBSD, but on Linux that's arguable; many of us believe that software RAID is a better solution than hardware RAID.
Note however that the default LILO/GRUB installations do NOT work when the first drive in a RAID configuration fails. It took us about a year and at least one failure before we figured out how to fix it so it would work.
What do you do when you upgrade software?
We update first on a testbed solution and study the upgrade results before we upgrade other systems.
RPM / YUM / Up2Date generally doesn't break anything except occasonally configuration files. So configuration files are the first thing we check if we have problems with a daemon after an update.
CustomApache generally doesn't break anything; don't forget it had to run at least once, in order for DA to run on your server.
Let me rephrase: CustomApache generally doesn't break anything unless you've made custom changes inside it. When we do that, we do it first on a testbed server.
Services that have to updated manually because they're covered neither by RPM/YUM/Up2Date or by CustomApache, we create new versions in their own directories; then we create symbolic links for the various versions. If the link to the new version doesn't work we create a link back to the old version until we figure out why the new version doesn't work.
Do you have some sort of backup so you can rollback in case of problems?
We backup our retail account servers daily, our dedicated servers weekly or more often as contracted, and our customer services not in our data center depending on what they want and pay for.
Yes, it can easily take multiple testbed servers and terrabytes of backup space. Anyone who told you that webhosting was a cheap business to get into was wrong
.
Jeff