Since long ago, Linux server deployments have been moving towards the collaboration of multiple physical machines. In some cases, different servers run each a different set of applications, bringing organization, isolation, reserved resources, and other desirable characteristics to the composed assembly. In other situations, servers are set up with very similar configurations, so that the system becomes more scalable by having load distributed among the several instances, and so that the overall system becomes more reliable when the failure of any individual machine does not affect the assembly as a whole. In this reality, server administrators become invaluable maestros which orchestrate the placement and connectivity of services within the assembly of servers.
Given that scenario, it’s surprising that most of the efforts towards advancing the management of software configuration are still bound to individual machines. Package managers, and software like dbus and gconf are examples of this. Other efforts do look at the problem of managing multiple machines as a unit, but interestingly, they are still a mechanism for scaling up the management of services individually. In other words, they empower the administrator with the ability to tweak the individual configuration of multiple services at once, but they do not collaborate towards offering services themselves and other tools the understanding of the composed ensemble. This distinction looks subtle in principle, but it may be a key factor in enabling all the parties (system administrators, software developers, vendors, and integrators) to collaborate in deploying, maintaining, and enriching distributed software configurations.
This is the challenge which motivates the research happening through the Ensemble project at Canonical. Ensemble aims to be a service deployment and orchestration tool which enables the same kind of collaboration and ease of use which today is seen around package management to happen on a higher level, around services. With Ensemble, different authors are able to create services independently, and make those services communicate through a simple configuration protocol. Then, users can take the product of both authors and very comfortably deploy those services in an environment, in way resembling how people are able to install a network of packages with a single command via APT.