On the Server

A server accessible via the Internet extends the existing setup. To do this, you again need to install Git-annex on a particular machine and use the wizard in the browser to create a new repository.

In the dashboard, the Add another repository button provides an overview of possible locations. Select the Remote server entry from the list with the description Set up a repository on a remote server using ssh. SSH with public key authentication eliminates the problem of password entry during the configuration.

After filling out the dialogs, entering the hostname, username, and directory, the software creates the necessary files and starts syncing. A final menu (Figure  2) also points out that the SSH server uses a repository belonging to the transfer group.

Figure 2: After the setup, you define the repository group.

Groups of this type keep the data long enough to allow it to be distributed to the clients. This approach avoids data ballast accumulating on the server. A transfer repository would thus even be possible on a server with limited storage space.

In addition to the transfer group, Git-annex defines nine other standard groups [4]. Depending on the application, you can assign a repository to one of these groups. The groups integrate predefined rules for distributing data. The program decides on the basis of these rules which data to keep and which to distribute to other repositories (see Table 1).

Remote Sharing

Assume members of the field staff want to access the existing repository on the server and synchronize data with the workgroup. They configure a local repository and the same remote server as their colleagues. In principle, the team can now exchange data via the central server, but for the time being, this is a manual process (Listing 4). For fully automated synchronization, further steps are necessary.

Automatic

The previously mentioned intermediate step for automatic synchronization relies on a tool that informs other clients about changes (Figure 3).

Figure 3: Git-annex still needs a messaging service to be able to sync data automatically.

In the current version of Git-annex, this role is assumed by an XMPP or a Jabber server, which the clients use to set up remote pairing with their friends. Alternatively, messaging services such as Google Talk are suitable.

Joey Hess, the developer of Git-annex is working on a daemon that removes the intermediate step [5]. It monitors remote databases and automatically starts the sync when needed. The daemon has already made it into the main branch of development; a stable version is thus likely to appear in the not too distant future.

If you want to manage all of the components yourself, you can install your own messaging server, such as Ejabberd. However, the packages from Debian "Wheezy" and Ubuntu "Precise" do not work with Git-annex Assistant: A bug in the authentication mechanism prevents the configuration via the interface [6].

However, installer scripts exist for Ejabberd on the project website that facilitate the manual setup. On Ubuntu, you need the libyaml-0-2 package; then, just a few commands set up your own Jabber server that cooperates with Git-annex.

The Jabber server is the final component in remote sharing. The clients in the workgroup – including that used by a field employee – only need to accept one another as chat friends. In the web interface they use the Share with a friend menu item to enter the Jabber account name.

In the following screen, the names of active friends and a button labeled Start pairing appear. This button sends a pair request, which the communication partner needs to confirm. After this final step in the configuration, the clients synchronize their data via the transfer repository.