GlusterFS driver

GlusterFS driver uses GlusterFS, an open source distributed file system, as the storage backend for serving file shares to manila clients.

Supported shared filesystems

  • NFS (access by IP)

Supported Operations

  • Create share

  • Delete share

  • Allow share access (rw)

  • Deny share access

  • With volume layout:

    • Create snapshot

    • Delete snapshot

    • Create share from snapshot

Requirements

  • Install glusterfs-server package, version >= 3.5.x, on the storage backend.

  • Install NFS-Ganesha, version >=2.1, if using NFS-Ganesha as the NFS server for the GlusterFS backend.

  • Install glusterfs and glusterfs-fuse package, version >=3.5.x, on the manila host.

  • Establish network connection between the manila host and the storage backend.

Manila driver configuration setting

The following parameters in the manila’s configuration file need to be set:

  • share_driver = manila.share.drivers.glusterfs.GlusterfsShareDriver

The following configuration parameters are optional:

  • glusterfs_nfs_server_type = <NFS server type used by the GlusterFS

    backend, Gluster or Ganesha. Gluster is the default type>

  • glusterfs_share_layout = <share layout used>; cf. Layouts

  • glusterfs_path_to_private_key = <path to manila host’s private key file>

  • glusterfs_server_password = <password of remote GlusterFS server machine>

If Ganesha NFS server is used (glusterfs_nfs_server_type = Ganesha), then by default the Ganesha server is supposed to run on the manila host and is managed by local commands. If it’s deployed somewhere else, then it’s managed via ssh, which can be configured by the following parameters:

  • glusterfs_ganesha_server_ip

  • glusterfs_ganesha_server_username

  • glusterfs_ganesha_server_password

In lack of glusterfs_ganesha_server_password ssh access will fall back to key based authentication, using the key specified by glusterfs_path_to_private_key, or, in lack of that, a key at one of the OpenSSH-style default key locations (~/.ssh/id_{r,d,ecd}sa).

Layouts have also their set of parameters, see Layouts about that.

Layouts

New in Liberty, multiple share layouts can be used with glusterfs driver. A layout is a strategy of allocating storage from GlusterFS backends for shares. Currently there are two layouts implemented:

  • directory mapped layout (or directory layout, or dir layout for short): a share is backed by top-level subdirectories of a given GlusterFS volume.

    Directory mapped layout is the default and backward compatible with Kilo. The following setting explicitly specifies its usage: glusterfs_share_layout = layout_directory.GlusterfsDirectoryMappedLayout.

    Options:

    • glusterfs_target: address of the volume that hosts the directories. If it’s of the format <glustervolserver>:/<glustervolid>, then the manila host is expected to be part of the GlusterFS cluster of the volume and GlusterFS management happens through locally calling the gluster utility. If it’s of the format <username>@<glustervolserver>:/<glustervolid>, then we ssh to <username>@<glustervolserver> to execute gluster (<username> is supposed to have administrative privileges on <glustervolserver>).

    • glusterfs_mount_point_base = <base path of GlusterFS volume mounted on

      manila host> (optional; defaults to $state_path/mnt, where $state_path defaults to /var/lib/manila)

    Limitations:

    • directory layout does not support snapshot operations.

  • volume mapped layout (or volume layout, or vol layout for short): a share is backed by a whole GlusterFS volume.

    Volume mapped layout is new in Liberty. It can be chosen by setting glusterfs_share_layout = layout_volume.GlusterfsVolumeMappedLayout.

    Options (required):

    • glusterfs_servers

    • glusterfs_volume_pattern

    Volume mapped layout is implemented as a common backend of the glusterfs and glusterfs-native drivers; see the description of these options in GlusterFS Native driver: Manila driver configuration setting.

Gluster NFS with volume mapped layout

A special configuration choice is

glusterfs_nfs_server_type = Gluster
glusterfs_share_layout = layout_volume.GlusterfsVolumeMappedLayout

that is, Gluster NFS used to export whole volumes.

All other GlusterFS backend configurations (including GlusterFS set up with glusterfs-native) require the nfs.export-volumes = off GlusterFS setting. Gluster NFS with volume layout requires nfs.export-volumes = on. nfs.export-volumes is a cluster-wide setting, so a given GlusterFS cluster cannot host a share backend with Gluster NFS + volume layout and other share backend configurations at the same time.

There is another caveat with nfs.export-volumes: setting it to on without enough care is a security risk, as the default access control for the volume exports is “allow all”. For this reason, while the nfs.export-volumes = off setting is automatically set by manila for all other share backend configurations, nfs.export-volumes = on is not set by manila in case of a Gluster NFS with volume layout setup. It’s left to the GlusterFS admin to make this setting in conjunction with the associated safeguards (that is, for those volumes of the cluster which are not used by manila, access restrictions have to be manually configured through the nfs.rpc-auth-{allow,reject} options).

Known Restrictions

  • The driver does not support network segmented multi-tenancy model, but instead works over a flat network, where the tenants share a network.

  • If NFS Ganesha is the NFS server used by the GlusterFS backend, then the shares can be accessed by NFSv3 and v4 protocols. However, if Gluster NFS is used by the GlusterFS backend, then the shares can only be accessed by NFSv3 protocol.

  • All manila shares, which map to subdirectories within a GlusterFS volume, are currently created within a single GlusterFS volume of a GlusterFS storage pool.

  • The driver does not provide read-only access level for shares.

  • Assume that share S is exported through Gluster NFS, and tenant machine T has mounted S. If at this point access of T to S is revoked through access-deny, the pre-existing mount will be still usable and T will still be able to access the data in S as long as that mount is in place. (This violates the principle Access deny should always result in immediate loss of access to the share, see http://lists.openstack.org/pipermail/openstack-dev/2015-July/069109.html.)

The manila.share.drivers.glusterfs Module

Flat network GlusterFS Driver.

Manila shares are subdirectories within a GlusterFS volume. The backend, a GlusterFS cluster, uses one of the two NFS servers, Gluster-NFS or NFS-Ganesha, based on a configuration option, to mediate access to the shares. NFS-Ganesha server supports NFSv3 and v4 protocols, while Gluster-NFS server supports only NFSv3 protocol.

TODO(rraja): support SMB protocol.

class GaneshaNFSHelper(execute, config_object, **kwargs)

Bases: GaneshaNASHelper

get_export(share)
init_helper()

Initializes protocol-specific NAS drivers.

shared_data = {}
update_access(base_path, share, add_rules, delete_rules, recovery=False)

Update access rules.

class GlusterNFSHelper(execute, config_object, **kwargs)

Bases: NASHelperBase

Manage shares with Gluster-NFS server.

get_export(share)
supported_access_levels = ('rw',)
supported_access_types = ('ip',)
update_access(base_path, share, add_rules, delete_rules, recovery=False)

Update access rules.

class GlusterNFSVolHelper(execute, config_object, **kwargs)

Bases: GlusterNFSHelper

Manage shares with Gluster-NFS server, volume mapped variant.

update_access(base_path, share, add_rules, delete_rules, recovery=False)

Update access rules.

class GlusterfsShareDriver(*args, **kwargs)

Bases: ExecuteMixin, GaneshaMixin, GlusterfsShareDriverBase

Execute commands relating to Shares.

GLUSTERFS_VERSION_MIN = (3, 5)
check_for_setup_error()

Check for setup error.

do_setup(context)

Any initialization the share driver does while starting.

get_network_allocations_number()

Returns number of network allocations for creating VIFs.

Drivers that use Nova for share servers should return zero (0) here same as Generic driver does. Because Nova will handle network resources allocation. Drivers that handle networking itself should calculate it according to their own requirements. It can have 1+ network interfaces.

property supported_access_levels
property supported_access_types
supported_layouts = ('layout_directory.GlusterfsDirectoryMappedLayout', 'layout_volume.GlusterfsVolumeMappedLayout')
supported_protocols = ('NFS',)