On 9/14/2011 8:16 PM, John Andersen wrote:
On 9/14/2011 4:33 PM, Brian K. White wrote:
. You just partition a small boot partition on each drive the same way, make those a raid1 array in fully regular software raid, and the bios can boot from either copy just fine without knowing that it's part of a raid1.
I've used this method for many years with Grub, but I have to tell you, you better write a script that lives in /boot which copies everything to the other, because without that the /boot partitions will drift out of sync.
No I just make /boot a raid1. If either copy ever gets modified outside of linux, mdadmd alerts me by email the same as any other array, and it's easy to reset the array by removing re-adding the "failed" partition since it's tiny. But it never happens anyways. kernel updates write to "/boot" not /dev/anything, so the contents are always identical automatically. The only other special attention is 2 things: * I configure /etc/grub.conf to write grub to all drives mbr's or all drives /boot partitions. This way any drive can boot. Normally only one ever does boot, but the point is all about that day that drive is bad. * I configure at least one duplicate stanza in menu.lst where the only difference is the grub boot drive. No need to write one for every possible drive, if the normal hd0 is bad, booting from hd1 is all you need. Or really, I don't even bother with that because when that day comes, it's easy enough to do manually from the grub prompt. But really, I don't even do that any more. Too much work. Just stick a thumb drive somewhere and treat it like a regular simple non-raid drive. Both reads and writes are completely rare so it's no problem about wear. It can be tricky getting a given motherboard to recognize a thumb drve for booting. There are often different quirks and hoops to jump through required for each motherboard, but these days most boards can do it and the special mystery requirements about formatting and max size and bios settings are getting less and less every few months. It used to be pretty exotic and not worth the grief, but it's getting almost painless now. One of the biggest reasons I do that is so I don't have to partition the real drives at all. I just use the whole raw full drive device in mdadm, no fdisk or anything. That means that when I have 24 drives in one box, I don't have to do 24 special fdisk setups, exactly the same, correctly. That's _excrutiating in the yast interface. It's a lot better at the command line where you can re-run an identical sfdisk 24 times, not only safely reliably identical each time, but quickly too. Although it's hardly user friendly! Nor do I have to do anything when a drive goes bad. I just yank it and pop in the new one and it's one mdadm command to add the new drive to a single array. Without the usb thumb drive for /boot, I'd have to clone the partitioning scheme of one of the other drives, then tell mdadm to add each of 2 or 3 partitions to 2 or 3 different md arrays. More steps is not only more work but more down time or at least more at-risk time during problems and more mistakes. If it seems complicated or delicate, it is a little, but mostly only because things like yast don't consider this kind of arrangement yet. But there's no technical reason they couldn't and then this would all be effortless for grandma let alone me. -- bkw -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org