The document discusses implementing snapshot, clone, and boot from volume functionality in OpenStack. It describes taking volume snapshots that can then be used to clone new volumes or boot instances. The implementation would involve adding new methods to the volume driver, updating database schemas to support snapshots, and determining how to store snapshots. Challenges include supporting snapshot of snapshots in Linux LVM and implementing independent delete of volumes and snapshots.

1. Snapshot, Clone,
and Boot from volume
Kazutaka Morita: NTT Cyber Space Labs.
<morita.kazutaka@lab.ntt.co.jp>
Isaku Yamahata: VA Linux Systems Japan K.K
<yamahata@valinux.co.jp>
<yamahata@private.email.ne.jp>
Openstack design summit: April 28, 2011

2. Snapshot and Clone
● Snapshot
● Create a self-consistent image of the volume
● You can use snapshots to take backups
$ ec2­create­snapshot volume_id
$ ec2­create­snapshot volume_id
● Clone
● Create a new volume from the snapshot
● You can create lots of same volumes from one golden
snapshot
$ ec2­create­volume –snapshot snapshot_id ­z zone
$ ec2­create­volume –snapshot snapshot_id ­z zone
2

3. Implementation
User
ISCSIDriver
euca­create­snapshot AOEDriver
SheepdogDriver
RBDDriver
EC2 API OpenStack API SanDriver
Call
create_snapshot
nova-api nova-volume
rpc
3

4. New methods to volume driver
class VolumeDriver(object):
class VolumeDriver(object):
...
...
def create_snapshot(self, snapshot):
def create_snapshot(self, snapshot):
"""Creates a snapshot."""
"""Creates a snapshot."""
raise NotImplementedError()
raise NotImplementedError()
def delete_snapshot(self, snapshot):
def delete_snapshot(self, snapshot):
"""Deletes a snapshot."""
"""Deletes a snapshot."""
raise NotImplementedError()
raise NotImplementedError()
def create_volume_from_snapshot(self, volume, snapshot):
def create_volume_from_snapshot(self, volume, snapshot):
"""Creates a volume from a snapshot."""
"""Creates a volume from a snapshot."""
raise NotImplementedError()
raise NotImplementedError()
4

5. DB schema changes (1)
● Create new table Snapshot
snapshots = Table('snapshots', meta,
snapshots = Table('snapshots', meta,
Column('created_at', DateTime()),
Column('created_at', DateTime()),
Column('updated_at', DateTime()),
Column('updated_at', DateTime()),
Column('deleted_at', DateTime()),
Column('deleted_at', DateTime()),
Column('deleted', Boolean()),
Column('deleted', Boolean()),
Column('id', Integer(), primary_key=True),
Column('id', Integer(), primary_key=True),
Column('volume_id', Integer()),
Column('volume_id', Integer()),
Column('user_id', String()),
Column('user_id', String()),
Column('project_id', String()),
Column('project_id', String()),
Column('status', String()),
Column('status', String()),
Column('progress', String()),
Column('progress', String()),
Column('volume_size', Integer()),
Column('volume_size', Integer()),
Column('scheduled_at', DateTime()),
Column('scheduled_at', DateTime()),
Column('display_name', String()),
Column('display_name', String()),
Column('display_description', String())
Column('display_description', String())
) 5
)

6. DB schema changes (2)
● adds a new column snapshot_id to the table
volumes
snapshot_id = Column('snapshot_id', Integer())
snapshot_id = Column('snapshot_id', Integer())
6

7. Where to store snapshots
● Amazon EC2 stores EBS snapshots to S3
● Where should Nova store volume snapshots?
● To Swift?
● To Glance?
● To nothing? (only stored in nova-volume)
7

8. Snapshot/clone with Linux LVM2
● Linux LVM2 doesn't support snapshot of
snapshot.
● If deleting the origin volume, the dependent
snapshots will also be deleted
● EC2 allows delete volume/snapshot
independently: abandon this feature for now
Create snapshot Clone volume
Openstack
Volume snapshot Volume
Volume
Linux LVM2 LogicalVolume LV snapshot LogicalVolume
lvcrate --snapshot Create new LV and copy

9. Boot from volume(EBS boot)
Block Device Mapping <device>=<id>:<size(GiB)>:<DeleteOnTerminate>
To orverride machine image e.g. /dev/sda=snap-00000003:10:false
From command line
Passing down BDM argument
to compute node Block device mapping in db
Machine euca-run-instances
running
Image launching
euca-stop-instance
Create volume from snapshot stopping
stopped
euca-start-instance Instance
pending
Data flow
euca-termintate-instance
Delete volumes running Green: what I've implemented
terminating
Red: to be implemented

10. Machine Image changes
● +Machine Image Type: S3 or EBS
● +Block device mapping
● +Snapshot id or volume id or virtual device
– EC2 allows only snapshot id block device mapping
● +Volume Size
● +DeteleteOnTerminate
● +Root Device Instance id metadata data
● ec2-create-image Machine image
● ec2-register
Image Service

11. DB schema changes
● Volume Table
● +DeleteOnTerminate
● +BlockDeviceMapping Table
● +instance id
● +Device Name(device name in guest)
● +Mapped device
– +Virtual device(ephemeral0...) local storage which isn't persistent
– +Volume
● Snapshot id or Volume id
● Volume Size in GiB
● DeleteOnTerminate: bool
– +No Device

12. start/stop Instance
● start/stop instance
● Mostly same to run/terminate instance except
db manipulation
● New status
● Stopping/stopped/pending