Homepage: http://www.steve.org.uk/Software/redisfs/

Mercurial Repository: http://redisfs.repository.steve.org.uk/

Redis is a high-performance key/value store which allows storing keys, and their values, in a fast memory-backed store.

Intuitively it might seem that basing a filesystem around a simple key=value store isn't such a practical endeavour, but happily redis also has several useful abilities built-in, including native support for other data-structures including:

• sets.
• hashes.
• lists.

Using only key&value pairs, and the notion of a set, we can implement a simple filesystem, which is ready for replication and snapshotting.

This filesystem makes use of two of redis's abilities:

• KEY storage.
• SET storage.

Our filesystem is built around the notion that a directory contains entries, and these entries are members of a set named after the parent directory.

For example consider the directory tree which contains two entries:

1. /mnt/redis/foo/
2. /mnt/redis/README

For each filesystem entry, be it a file, symlink, or a directory, we allocate a unique identifier which we called an "INODE number". This may be used to store, retrieve, and search for information:

INODE:1:UID   -> The owner of the inode with ID 1.
INODE:1:GID   -> The owner of the inode with ID 1.
INODE:1:SIZE  -> The size of the object with inode #1.
INODE:1:NAME  -> The name of the object with inode #1.

So we might see this:

INODE:1:NAME => "foo"
INODE:1:MODE => "0755"
INODE:1:GID  => "0"
INODE:1:UID  => "0"
INODE:1:TYPE => "dir"

Similarly the entry for "/README" might look like this:

INODE:2:NAME => "README"
INODE:2:MODE => "0644"
INODE:2:GID  => "0"
INODE:2:UID  => "0"
INODE:2:TYPE => "file"

The actual contents of a directory are stored in a set, which has a name based upon the inode of the parent directory. For example:

SMEMBERS DIRENT:3 -> { "5", "6" }

In actual fact we add a prefix to each key and set name, which allows multiple filesystems to be mounted at the same time - and which is the key to our snapshotting facility.

To build the code, assuming you have the required build dependencies installed, you should merely need to run:

 make
 make test

Once built the software can be installed via:

 make install

It is possible to run the filesystem without having installed the software, via:

 # ./src/redisfs

Connecting to redis server localhost:6379 and mounting at /mnt/redis.

By default this will attempt to connect to a redis server running upon the same host - if you wish to connect to a remote machine please execute:

 # ./src/redisfs --host remote.example.org [--port=6379]