The Network File System (NFS) is a distributed file system protocol originally developed by Sun Microsystems in 1984. An NFS server exports one or more of its file systems, known as shares. An NFS client can mount these exported shares on its own file system. You can perform file actions on this mounted remote file system as if the file system were local.

The NFS driver, and other drivers based off of it, work quite differently than a traditional block storage driver. The NFS driver does not actually allow an instance to access a storage device at the block level. Instead, files are created on an NFS share and mapped to instances, which emulates a block device. This works in a similar way to QEMU, which stores instances in the /var/lib/nova/instances directory.

To use Cinder with the NFS driver, first set the volume_driver in cinder.conf: volume_driver=cinder.volume.drivers.nfs.NfsDriver The following table contains the options supported by the NFS driver. As of the Icehouse release, the NFS driver (and other drivers based off it) will attempt to mount shares using version 4.1 of the NFS protocol (including pNFS). If the mount attempt is unsuccessful due to a lack of client or server support, a subsequent mount attempt that requests the default behavior of the mount.nfs command will be performed. On most distributions, the default behavior is to attempt mounting first with NFS v4.0, then silently fall back to NFS v3.0 if necessary. If the nfs_mount_options configuration option contains a request for a specific version of NFS to be used, or if specific options are specified in the shares configuration file specified by the nfs_shares_config configuration option, the mount will be attempted as requested with no subsequent attempts.

Access to one or more NFS servers. Creating an NFS server is outside the scope of this document. This example assumes access to the following NFS servers and mount points: 192.168.1.200:/storage 192.168.1.201:/storage 192.168.1.202:/storage This example demonstrates the use of with this driver with multiple NFS servers. Multiple servers are not required. One is usually enough. Add your list of NFS servers to the file you specified with the nfs_shares_config option. For example, if the value of this option was set to /etc/cinder/shares.txt, then: # cat /etc/cinder/shares.txt 192.168.1.200:/storage 192.168.1.201:/storage 192.168.1.202:/storage Comments are allowed in this file. They begin with a #. Configure the nfs_mount_point_base option. This is a directory where cinder-volume mounts all NFS shares stored in shares.txt. For this example, /var/lib/cinder/nfs is used. You can, of course, use the default value of $state_path/mnt. Start the cinder-volume service. /var/lib/cinder/nfs should now contain a directory for each NFS share specified in shares.txt. The name of each directory is a hashed name: # ls /var/lib/cinder/nfs/ ... 46c5db75dc3a3a50a10bfd1a456a9f3f ... You can now create volumes as you normally would: $ nova volume-create --display-name=myvol 5 # ls /var/lib/cinder/nfs/46c5db75dc3a3a50a10bfd1a456a9f3f volume-a8862558-e6d6-4648-b5df-bb84f31c8935 This volume can also be attached and deleted just like other volumes. However, snapshotting is not supported.

cinder-volume manages the mounting of the NFS shares as well as volume creation on the shares. Keep this in mind when planning your OpenStack architecture. If you have one master NFS server, it might make sense to only have one cinder-volume service to handle all requests to that NFS server. However, if that single server is unable to handle all requests, more than one cinder-volume service is needed as well as potentially more than one NFS server. Because data is stored in a file and not actually on a block storage device, you might not see the same IO performance as you would with a traditional block storage driver. Please test accordingly. Despite possible IO performance loss, having volume data stored in a file might be beneficial. For example, backing up volumes can be as easy as copying the volume files. Regular IO flushing and syncing still stands.