Ceph Datastore

The Ceph Datastore driver allows the use of Ceph storage for Images and disks of Virtual Machines.

Warning

This driver requires the OpenNebula Nodes using the Ceph driver to be Ceph clients of a running Ceph cluster. More information in Ceph documentation.

Ceph Cluster Setup

This guide assumes that you already have a functional Ceph Cluster in place. Additionally you need to:

  • Create a pool for the OpenNebula datastores. Write down the name of the pool to include it in the datastore definitions.
ceph osd pool create one 128
ceph osd lspools
0 data,1 metadata,2 rbd,6 one,
  • Define a Ceph user to access the datastore pool; this user will also be used by libvirt to access the disk images. For example, create the user libvirt:

On the Ceph Jewel (v10.2.x and earlier):

ceph auth get-or-create client.libvirt \
      mon 'allow r' osd 'allow class-read object_prefix rbd_children, allow rwx pool=one'

On the Ceph Luminous (v12.2.x and later):

ceph auth get-or-create client.libvirt \
      mon 'profile rbd' osd 'profile rbd pool=one'

Warning

Ceph Luminous release comes with simplified RBD capabilities (more information about user management and authorization capabilities is in the Ceph documentation). When upgrading existing Ceph deployment to the Luminous and later, please ensure the selected user has proper new capabilities. For example, for the above user libvirt you can do this by running:

ceph auth caps client.libvirt \
      mon 'profile rbd' osd 'profile rbd pool=one'
  • Get a copy of the key of this user to distribute it later to the OpenNebula Nodes.
ceph auth get-key client.libvirt | tee client.libvirt.key
ceph auth get client.libvirt -o ceph.client.libvirt.keyring
  • Although RBD format 1 is supported it is strongly recommended to use Format 2. Check that ceph.conf includes:
[global]
rbd_default_format = 2
  • Pick a set of client Nodes of the cluster to act as storage bridges. These Nodes will be used to import images into the Ceph Cluster from OpenNebula. These Nodes must have the qemu-img command installed.

Note

For production environments it is recommended to not collocate Ceph services (monitor, osds) with OpenNebula Nodes or the Front-end

Front-end and Hosts Setup

In order to use the Ceph Cluster the Hosts need to be configured as follows:

  • The Ceph client tools must be available in the machine.
  • The mon daemon must be defined in the ceph.conf for all the Nodes, so hostname and port doesn’t need to be specified explicitly in any Ceph command.
  • Copy the Ceph user keyring (ceph.client.libvirt.keyring) to the Nodes under /etc/ceph, and the user key (client.libvirt.key) to the oneadmin home.
scp ceph.client.libvirt.keyring root@node:/etc/ceph
scp client.libvirt.key oneadmin@node:

Hosts Setup

Hosts need extra steps to set-up credentials in libvirt:

  • Generate a secret for the Ceph user and copy it to the Nodes under oneadmin home. Write down the UUID for later use.
UUID=`uuidgen`; echo $UUID
c7bdeabf-5f2a-4094-9413-58c6a9590980
cat > secret.xml <<EOF
<secret ephemeral='no' private='no'>
  <uuid>$UUID</uuid>
  <usage type='ceph'>
          <name>client.libvirt secret</name>
  </usage>
</secret>
EOF
scp secret.xml oneadmin@node:
  • Define a libvirt secret and remove key files in the Nodes:
virsh -c qemu:///system secret-define secret.xml
virsh -c qemu:///system secret-set-value --secret $UUID --base64 $(cat client.libvirt.key)
rm client.libvirt.key
  • The oneadmin account needs to access the Ceph Cluster using the libvirt Ceph user defined above. This requires access to the Ceph user keyring. Test that the Ceph client is properly configured in the Node.
ssh oneadmin@node
rbd ls -p one --id libvirt

You can read more information about this in the Ceph guide Using libvirt with Ceph.

  • Ancillary Virtual Machine files like context disks, deployment and checkpoint files are created on the Nodes under /var/lib/one/datastores/. Make sure that enough storage for these files is provisioned on the Nodes.
  • If you are going to use the SSH mode, you have to take into account the space needed for the System Datastore /var/lib/one/datastores/<ds_id> where ds_id is the ID of the System Datastore.

OpenNebula Configuration

To use your Ceph Cluster with the OpenNebula, you need to define a System and Image Datastores. Each Image/System Datastore pair will share same following Ceph configuration attributes:

Attribute Description Mandatory
NAME Name of datastore YES
POOL_NAME The Ceph pool name YES
CEPH_USER The Ceph user name, used by libvirt and rbd commands. YES
CEPH_KEY Key file for user, if not set default locations are used NO
CEPH_CONF Non-default Ceph configuration file if needed. NO
RBD_FORMAT By default RBD Format 2 will be used. NO
BRIDGE_LIST List of storage bridges to access the Ceph Cluster YES
CEPH_HOST Space-separated list of Ceph monitors. Example: host1 host2:port2 host3 host4:port4. YES
CEPH_SECRET The UUID of the libvirt secret. YES
EC_POOL_NAME Name of Ceph erasure coded pool NO
CEPH_TRASH Enables trash feature on given datastore (Luminous+), values: yes|no NO

Note

You may add another Image and System Datastore pointing to other pools with different allocation/replication policies in Ceph.

Note

Ceph Luminous release allows use of erasure coding for RBD images. In general, erasure-coded images take up less space but have worse I/O performance. Erasure coding can be enabled on Image and/or System Datastores by configuring EC_POOL_NAME with the name of the erasure-coded data pool. Regular replicated Ceph pool POOL_NAME is still required for image metadata. More information in Ceph documentation.

Create System Datastore

System Datastore also requires these attributes:

Attribute Description Mandatory
TYPE SYSTEM_DS YES
TM_MAD ceph to use the full Ceph mode, see below YES
ssh to use local Host storage, SSH mode below

Create a System Datastore in Sunstone or through the CLI, for example:

cat systemds.txt
NAME    = ceph_system
TM_MAD  = ceph
TYPE    = SYSTEM_DS

POOL_NAME   = one
CEPH_HOST   = "host1 host2:port2"
CEPH_USER   = libvirt
CEPH_SECRET = "6f88b54b-5dae-41fe-a43e-b2763f601cfc"

BRIDGE_LIST = cephfrontend
onedatastore create systemds.txt
ID: 101

Note

When different System Datastores are available the TM_MAD_SYSTEM attribute will be set after picking the Datastore.

Create Image Datastore

Apart from the previous attributes, which need to be the same as the associated System Datastore, the following can be set for an Image Datastore:

Attribute Description Mandatory
NAME Name of datastore YES
DS_MAD ceph YES
TM_MAD ceph YES
DISK_TYPE RBD YES
STAGING_DIR Default path for image operations in the bridges NO

An example of Datastore:

> cat ds.conf
NAME = "cephds"
DS_MAD = ceph
TM_MAD = ceph

DISK_TYPE = RBD

POOL_NAME   = one
CEPH_HOST   = "host1 host2:port2"
CEPH_USER   = libvirt
CEPH_SECRET = "6f88b54b-5dae-41fe-a43e-b2763f601cfc"

BRIDGE_LIST = cephfrontend

> onedatastore create ds.conf
ID: 101

Warning

If you are going to use the TM_MAD_SYSTEM attribute with ssh mode, you need to have an SSH type System Datastore configured.

Additional Configuration

Default values for the Ceph drivers can be set in /var/lib/one/remotes/etc/datastore/ceph/ceph.conf:

  • POOL_NAME: Default volume group.
  • STAGING_DIR: Default path for image operations in the storage bridges.
  • RBD_FORMAT: Default format for RBD volumes.
  • DD_BLOCK_SIZE: Block size for dd operations (default: 64kB).
  • SUPPORTED_FS: Comma-separated list with every file system supported for creating formatted datablocks. Can be set in /var/lib/one/remotes/etc/datastore/datastore.conf.
  • FS_OPTS_<FS>: Options for creating the file system for formatted datablocks. Can be set in /var/lib/one/remotes/etc/datastore/datastore.conf for each file system type.

Warning

Before adding a new file system to the SUPPORTED_FS list, make sure that the corresponding mkfs.<fs_name> command is available in all Nodes including Front-end and hypervisor Nodes. If an unsupported FS is used by the user the default one will be used.

Using different modes

When creating a VM Template you can choose to deploy the disks using the default Ceph mode or the SSH one. Note that the same mode will be used for all disks of the VM. To set the deployment mode, add the following attribute to the VM template:

  • TM_MAD_SYSTEM="ssh"

When using Sunstone, the deployment mode needs to be set in the Storage tab.

Datastore Internals

Images are stored in a Ceph pool, named after its OpenNebula ID one-<IMAGE ID>. Virtual Machine disks are stored by default in the same pool (Ceph Mode). You can also choose to export the Image rbd to the hypervisor local storage using the SSH Mode.

Important

It is necessary to register each image only once, then it can be deployed using any mode (ceph or ssh).

Ceph Mode (Default)

In this mode, Virtual Machines will use the same Image rbd volumes for their disks (persistent images), or a new snapshots of the image created in the form one-<IMAGE ID>-<VM ID>-<DISK ID> (non-persistent images).

For example, consider a system using an Image and System Datastore backed by a Ceph pool named one. The pool with one Image (ID 0) and two Virtual Machines 14 and 15 using this Image as virtual disk 0 would be similar to:

rbd ls -l -p one --id libvirt
NAME         SIZE PARENT         FMT PROT LOCK
one-0      10240M                  2
one-0@snap 10240M                  2 yes
one-0-14-0 10240M one/one-0@snap   2
one-0-15-0 10240M one/one-0@snap   2

Note

In this case, context disk and auxiliary files (deployment description and checkpoints) are stored locally in the Nodes.

SSH Mode

In this mode, the associated rbd file for each disk is exported to a file and stored in the local file system of the hypervisor.

For instance, in the previous example, if the VM 14 is set to be deployed in an SSH System Datastore (e.g. 100), the layout of the datastore in the hypervisor would be similar to:

ls -l /var/lib/one/datastores/100/14
total 609228
-rw-rw-r-- 1 oneadmin oneadmin        1020 Dec 20 14:41 deployment.0
-rw-r--r-- 1 oneadmin oneadmin 10737418240 Dec 20 15:19 disk.0
-rw-rw-r-- 1 oneadmin oneadmin      372736 Dec 20 14:41 disk.1

Note

In this case disk.0 is generated with a command similar to rbd export one/one-0@snap disk.0

Warning

In this mode there are some inherent limitations

  • disk snapshots are not supported
  • VM disk cannot be saved when located on the Front-end (undeployed or stopped VM)