Serf vs. Fabric

Fabric is a widely used tool for system administration over SSH. Broadly, it is used to SSH into a group of nodes and execute commands. Both Fabric and Serf can be used for service management in different ways.

While Fabric sends commands from a single box, Serf instead rapidly broadcasts a message to the entire cluster in a distributed fashion. Fabric has a number of advantages in that it can collect the output of commands and stop execution if an error is encountered. Serf is unable to do these things since it has no single destination to send logs to, nor does it have any control flow. However, Fabric must be provided with a list of nodes to contact, whereas membership is built directly into Serf. Additionally, Serf is able to propagate a message within seconds to an entire cluster, allowing for much higher parallelism and scalability.

Fabric is much more capable than Serf at system administration, but it is limited by its execution speed and lack of node discovery. Combined together, Fabric can query Serf for nodes and make use of message broadcasts where appropriate, using direct SSH exection when and where output is needed.