We mentioned earlier that HDFS replication alone is not a suitable backup strategy. In the Hadoop 2 filesystem, snapshots have been added, which brings another level of data protection to HDFS.

Filesystem snapshots have been used for some time across a variety of technologies. The basic idea is that it becomes possible to view the exact state of the filesystem at particular points in time. This is achieved by taking a copy of the filesystem metadata at the point the snapshot is made and making this available to be viewed in the future.

As changes to the filesystem are made, any change that would affect the snapshot is treated specially. For example, if a file that exists in the snapshot is deleted then, even though it will be removed from the current state of the filesystem, its metadata will remain in the snapshot, and the blocks associated with its data will remain on the filesystem though not accessible through any view of the system other than the snapshot.

An example might...