When moving from the Linode kernels to Direct to Disk, how does KVM know which device to boot from? Does it a always boot from /dev/sda? Or can I create a separate boot device, say on /dev/sdc, install the MBR there, and boot from that device?

I ask, because it would be nice to be able to have a separate image for booting, and only mount it on /boot when updating the kernel. This would also provide a convenient way to test out different kernels. Each could live in its own boot image.

The idea is that the "root device" inside the Configuration Profile becomes the "boot device" selector, when using Direct Disk mode.

I know there have been changes made to that code recently to improve this - it may have been deployed today. If not, it's staged code and will be avail very soon.

-Chris

I assumed that with a new KVM we should be changing fstab entries from /dev/sd[abc...] to UUID's? I did this before accepting the KVM today. This upgrade seemed to go OK. Am I correct about using UUID's; e.g., will this be required soon?

_________________
http://www.ingber.com

There is nothing specific about using UUIDs in either Xen or KVM Linodes. If you want to roll with UUIDs, you can - but by no means is it necessary.

-Chris

UUID is not necessary, but only if you use the "distro-helper".

This is what happened in a test linode after I switched to KVM:

Only in etc.after/event.disabled: hvc0.conf
diff -ru etc.before/fstab etc.after/fstab
--- etc.before/fstab   2015-06-13 13:16:59.000000000 +0200
+++ etc.after/fstab   2015-06-13 13:23:09.000000000 +0200
@@ -6,8 +6,8 @@
 #
 # <file system> <mount point>   <type>  <options>       <dump>  <pass>
 proc        /proc        proc    defaults                       0 0
-/dev/xvda   /            ext4    noatime,errors=remount-ro      0 1
-/dev/xvdb   none         swap    sw    
+/dev/sda   /            ext4    noatime,errors=remount-ro      0 1
+/dev/sdb   none         swap    sw    
 none            /tmp            tmpfs   mode=1777       0       0
 none /var/cache/apt       tmpfs mode=755                    0       0
 none /var/lib/apt/lists   tmpfs mode=755                    0       0
Only in etc.before/init: hvc0.conf
Only in etc.after/init: ttyS0.conf

There are two kinds of changes here:

* The /etc/init thing. This is only for upstart, used mainly by old Ubuntu releases. Not required anymore if the distro is recent enough.
* The /etc/fstab thing. Modifying /etc/fstab is not required at all when using UUID.

So I would say that UUID and systemd have made distro-helper to be obsolete.

IMHO, switching to KVM would have been a good opportunity to start deprecating distro-helper and start using UUID, like everybody else.