Windows SMB volume driver¶
The Windows SMB volume driver leverages pre-existing SMB shares, used to store volumes as virtual disk images.
The main reasons to use the Windows SMB driver are:
ease of management and use
great integration with other Microsoft technologies (e.g. Hyper-V Failover Cluster)
suitable for a various range of deployment types and sizes
cinder-volume service as well as the required Python components will
be installed directly onto designated Windows nodes (preferably the ones
exposing the shares).
Common deployment scenarios¶
The SMB driver is designed to support a variety of scenarios, such as:
Scale-Out File Servers (
SoFS), providing highly available SMB shares.
standalone Windows or Samba shares
any other SMB 3.0 capable device
By using SoFS shares, the virtual disk images are stored on Cluster Shared
A common practice involves deploying CSVs on top of SAN backed LUNs
(exposed to all the nodes of the cluster through iSCSI or Fibre Channel). In
absence of a SAN, Storage Spaces/Storage Spaces Direct (
S2D) may be used
for the underlying storage.
S2D is commonly used in hyper-converged deployments.
VHDX are the currently supported image formats and may be
consumed by Hyper-V and KVM compute nodes. By default, dynamic (thinly
provisioned) images will be used, unless configured otherwise.
The driver accepts one or more shares that will be reported to the Cinder scheduler as storage pools. This can provide means of tiering, allowing specific shares (pools) to be requested through volume types.
openstack volume type set $volume_type --property pool_name=$pool_name
Frontend QoS specs may be associated with the volume types and enforced on the consumer side (e.g. Hyper-V).
openstack volume qos create $rule_name --property consumer=front-end --property total_bytes_sec=20971520 openstack volume qos associate $rule_name $volume_type_id openstack volume create $volume_name --type $volume_type_id --size $size
Cinder Backup Service can be run on Windows. This driver stores
the volumes using vhdx images stored on SMB shares which can be attached
in order to retrieve the volume data and send it to the backup service.
All physical disks must be in byte mode
rb+ must be used when writing backups to disk
Active-Active Cinder clustering is currently experimental and should not be used in production. This implies having multiple Cinder Volume services handling the same share simultaneously.
On the other hand, Active-Passive clustering can easily be achieved, configuring the Cinder Volume service as clustered using Microsoft Failover Cluster.
By using SoFS, you can provide high availability of the shares used by Cinder. This can be used in conjunction with the Nova Hyper-V cluster driver, which allows clustering virtual machines. This ensures that when a compute node is compromised, the virtual machines are transparently migrated to a healthy node, preserving volume connectivity.
The Windows SMB driver is the only Cinder driver that may be used along with the Nova Hyper-V cluster driver. The reason is that during an unexpected failover, the volumes need to be available on the destination compute node side.
Before setting up the SMB driver, you will need to create and configure one or more SMB shares that will be used for storing virtual disk images.
The driver does not manage share permissions. You will have to make sure that Cinder as well as share consumers (e.g. Nova, Hyper-V) have access.
Note that Hyper-V VMs are run using a built-in user group:
NT VIRTUAL MACHINE\Virtual Machines.
The easiest way to provide share access is by using Active Directory accounts. You may grant share access to the users running OpenStack services, as well as the compute nodes (and optionally storage nodes), using per computer account access rules. One of the main advantages is that by doing so, you don’t need to pass share credentials to Cinder (and implicitly volume consumers).
By granting access to a computer account, you’re basically granting access to the LocalSystem account of that node, and thus to the VMs running on that host.
By default, OpenStack services deployed using the MSIs are run as LocalSystem.
Once you’ve granted share access to a specific account, don’t forget to also configure file system level permissions on the directory exported by the share.
Below is a configuration sample for using the Windows SMB Driver. Append
those options to your already existing
cinder.conf file, described at
Install and configure a storage node.
[DEFAULT] enabled_backends = winsmb [winsmb] volume_backend_name = myWindowsSMBBackend volume_driver = cinder.volume.drivers.windows.smbfs.WindowsSmbfsDriver smbfs_mount_point_base = C:\OpenStack\mnt\ smbfs_shares_config = C:\Program Files\Cloudbase Solutions\OpenStack\etc\cinder\smbfs_shares_list # The following config options are optional # # image_volume_cache_enabled = true # image_volume_cache_max_size_gb = 100 # image_volume_cache_max_count = 10 # # nas_volume_prov_type = thin # smbfs_default_volume_format = vhdx # max_over_subscription_ratio = 1.5 # reserved_percentage = 5 # smbfs_pool_mappings = //addr/share:pool_name,//addr/share2:pool_name2
smbfs_mount_point_base config option allows you to specify where
the shares will be mounted. This directory will contain symlinks pointing
to the shares used by Cinder. Each symlink name will be a hash of the actual
Configuring Nova credentials¶
The SMB volume driver relies on the
nova assisted volume snapshots feature
when snapshotting in-use volumes, as do other similar drivers using shared
By default, the Nova policy requires admin rights for this operation. You may provide Cinder specific credentials to be used when requesting Nova assisted volume snapshots, as shown bellow:
[nova] region_name=RegionOne auth_strategy=keystone auth_type=password auth_url=http://keystone_host/identity project_name=service username=nova password=password project_domain_name=Default user_domain_name=Default
Configuring storage pools¶
Each share is reported to the Cinder scheduler as a storage pool.
By default, the share name will be the name of the pool. If needed, you may provide pool name mappings, specifying a custom pool name for each share, as shown bellow:
smbfs_pool_mappings = //addr/share:pool0
In the above sample, the
//addr/share share will be reported as