From Fedora Project Wiki

Revision as of 16:55, 17 September 2012 by Eglynn (talk | contribs)


Nova instances can be booted from volume, analogous to EBS-backed volumes in EC2.

We construct a bootable volume, then fire up an instance backed by this volume.


We assume that an instance has already been booted in a previous test, and we use this as a builder to facilitate the creation of a bootable volume.

Capture the instance name, user name, and IP address as an environment variables:

$> INSTANCE=<instance name>
$> USER_NAME=<user name>
$> IP_ADDR=$(nova show $INSTANCE 

How to test

Create a 1Gb volume, which we will make bootable:

$> nova volume-create --display_name=bootable_volume 1
$> VOLUME_ID=$(nova volume-list | awk '/bootable_volume/ {print $2}')

and wait for the volume to become available:

$> watch "nova volume-show bootable_volume | grep status"

Temporarily attach volume to your builder instance, this will allow us to copy image data into the volume

$> nova volume-attach $INSTANCE $VOLUME_ID /dev/vdb

Wait for the volume status to show as in-use:

$> watch "nova volume-show bootable_volume | grep status"

Format and mount volume to a staging mount point:

$> ssh -o StrictHostKeyChecking=no $USER_NAME@$IP_ADDR << EOF
set -o errexit
set -o xtrace
sudo mkdir -p /tmp/stage
sudo mkfs.ext3 -b 1024 /dev/vdb 1048576
sudo mount /dev/vdb /tmp/stage
sudo touch /tmp/stage/cirros-0.3.0-x86_64-rootfs.img.gz
sudo chown $USER_NAME /tmp/stage/cirros-0.3.0-x86_64-rootfs.img.gz

Copy image to the staging directory on the builder instance:

$> scp -o StrictHostKeyChecking=no cirros-0.3.0-x86_64-rootfs.img.gz $USER_NAME@$IP_ADDR:/tmp/stage

Unpack image into the volume (don't worry about an unmount failure).

$> ssh -o StrictHostKeyChecking=no -i $USER_NAME@$IP_ADDR << EOF
set -o errexit
set -o xtrace
cd /tmp/stage
sudo mkdir -p /tmp/image
sudo gunzip cirros-0.3.0-x86_64-rootfs.img.gz
sudo mount cirros-0.3.0-x86_64-rootfs.img /tmp/image
sudo cp -pr /tmp/image/* /tmp/stage/
sudo umount /tmp/image
sudo umount /tmp/stage || true

Detach volume for the builder instance:

$> nova volume-detach $INSTANCE $VOLUME_ID

and wait for the volume status to show as availble:

$> watch "nova volume-show bootable_volume | grep status"

Now snapshot the bootable volume we just created:

$> nova volume-snapshot-create --display_name bootable_snapshot $VOLUME_ID

and wait for the snapshot to become available:

$> nova volume-snapshot-show bootable_snapshot
$> SNAPSHOT_ID=$(nova volume-snapshot-list | awk '/bootable_snapshot/ {print $2}')

Now we can boot from the bootable volume. We use the same image as the builder instance but that is only in order to retrieve the image properties (kernel and ramdisk IDs)

$> IMAGE_ID=$(nova show $INSTANCE | awk '/image/ {print $5}' | sed 's,(\(.*\)),\1,')
$> nova boot --flavor 1 --image $INSTANCE --block_device_mapping vda=${SNAPSHOT_ID}:snap::0 volume_backed

Expected Results

You should be able able to shh into the volume-backed instance.

Note that an additional snapshot now exists to back the image:

$> nova volume-snapshot-list

Also note the volume-backed instance you've fired up, there is a volume cloned from the corresponding snapshot:

$> nova volume-list