investigate ec2 kernel/ramdisk and apt-upgrade

[Scott Moser]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Some discussion about apt-get upgrade on the ec2 instances. The message
at http://groups.google.com/group/ec2ubuntu/msg/635cb719fb42873c asks for
what *should* be done for kernel and ramdisk changes in an upgrade.

We need to investigate this, possibly pinning the versions of the kernels
in the image, and documenting how to un-pin.

Someone might want to un-pin and actually change kernel in the instance
because of:
 - rebundling
 - modifying an EBS root volume offline.

The package ec2-init isn't terribly relevant, probably more the build
scripts, but assigning it to ec2-init.

 affects ubuntu/ec2-init
 assignee smoser
 status confirmed
 importance medium
 tag ec2-images
 subscribe smoser

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)

iEYEARECAAYFAktNz2wACgkQDZzbdwr3wEfvUgCfdQ1D9hfM6jdfmqshzPq8tGWM
BVEAn10dkxHbU7+0ngaoFL5q02qLLVdm
=kMzB
-----END PGP SIGNATURE-----

[Scott Moser (smoser)]

Just a note
I realized that the reason updates are prompted for is the presense of the meta packages. If we remove them, the linux-image packages will still be present, but dont get upgraded. Ie:
 linux-image-2.6.32-11-virtual
 linux-image-2.6.32-301-ec2
will still be around. Those packages are sometimes serviced, in which case their modules would possibly contain needed fixes and *should* be updated.

Basically, I think that the following fixes this issue:
  sudo apt-get remove linux-image-virtual linux-virtual linux-image-ec2 linux-ec2

[Mathias Gug (mathiaz)]

Seems like a good solution to me.

[Scott Moser (smoser)]

More discussion on this in IRC. Eric Hammond pointed out that with EBS volumes, you can actually stop the instance, modify the kernel and ramdisk attribute and boot the instance with new kerne/ramdisk.

This would actually provide a usable mechanism for updating the kernel in an EBS volume.

I had committed a change to remove the meta packages, but have since backed that out.

[Scott Moser (smoser)]

Given the above, I've done some thought on this. I think the best thing to do right now is leave this as it is.
'apt-get upgrade' actually performs reasonably well, new kernels with same ABI are installed, so any changes in modules would be picked upon next reboot.

The one thing I think might be useful in the future is a updates notification that gave specific information to ec2/uec users on what to do.

[Scott Moser (smoser)]

Marking as won't fix in ec2-init (karmic or previous). Moving 'Confirmed/Wishlist' to cloud-init.

[Scott Moser (smoser)]

I'm marking this fix-released. Maverick images of 20100913 or later (including grub-pc of 1.98+20100804-4ubuntu6 and cloud-init of 0.5.15-0ubuntu1) will properly install new kernels, and boot into them on reboot.

This functionality is also present for Maverick instances that are running on UEC hosts running maverick versions of Eucalyptus and qemu-kvm.