I finally managed to get HVM images working on EC2!
Currently they are booted with extlinux as I am still having some
trouble installing grub.
There is a sample manifest in the build-debian-cloud manifest folder
you can try out.
So far I have only tried with an unpartitioned volume.
Awesome! Looks good. As a test, I have generated an AMI from this and
copied it to all AWS Regions for anyone to fire up and try out. Would
love to get some feedback from people here - search for AMIs in account
379101102735. I'm also looking for feedback on the Cloud-init deploy in
7.3 - and based on what people say here, we will then deprecate the 7.1a
AMIs (non-cloud-init - cleaning up for housekeeping purposes).
I will try to test more combinations in the future, here is a small
table of how far I've come:
boot\type none msdos gpt
vbox --------------------
grub O X X
ext X X X
ebs ---------------------
pvgrub X X ?
grub O ? ?
ext X ? ?
s3 ----------------------
pvgrub X ? ?
grub O ? ?
ext ? ? ?
X means "working", O means "not working" and ? means "not tested".
grub+ext of course only makes sense on HVM instances while pvgrub is
for PVM instances.
I have also made the partitions sizing more finely adjustable
<https://github.com/andsens/build-debian-cloud/blob/python/base/fs/partitionmaps/gpt.py>.
For GPT this means that the grub_legacy partition is exactly 1007KiB,
prepending that with the GPT table header you get a nicely aligned
1MiB offset for the next partition.
Also: Whatever extra space is required for GPT, post-MBR gap or grub
partition will now be subtracted from the first partition. This means
there is no longer any need to remember how much space you exactly
need to subtract from your root volume to hit 1 gigabyte. You just
specify "root": "1GiB".
Nice, was playing with this just before Christmas; I can discard my
patches! ;)
Well done Anders! Next stop, playing with the g2 instances[1] and the
srvio low latency interfaces! We're then looking like a solid platform
across all instance types.
James
[1] g2 is based on the nvidia Keplar GK104
(http://aws.amazon.com/ec2/instance-types/#instance-details). See also
http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using_cluster_computing.html.
Perhaps we should test out nouveau?
