Maven's Strengths and Weaknesses as a Dependency Management System
Last month, in our article Chaperoning Promiscuous Software Reuse, we reviewed four elements of a successful strategy for managing components. Briefly, these elements are:
- Testing and Validation of 3rd party components Only libraries approved for reuse should be used by developers. If developers want to use additional libraries, they can submit them for screening.
- A Definitive Software Library: A team should have an official storage place for reusable, versioned components (both internal and 3rd party).
- Using Protection: A team should have an "immune system" that rejects libraries other than those from the DSL.
- Rapid impact analysis audit When something goes wrong, it should be easy to determine what versions of which components were used by the project.
Basic MavenBasic Maven points to a common Maven repository hosted centrally. Various open source projects publish their releases to this repository and a developer can configure her build file to depend on these releases. At build time, the artifacts are automatically pulled down into the workspace and the build can produce a report of what was used.
This is actually a pretty good start. If the build script depends on exact versions, the team has some traceability. The Maven repository is an official repository of artifacts (a DSL) used by the team although the team does not control it. Unfortunately, there's little protection against a developer using a rogue component, but it's easier to do the right thing than the wrong thing, and that goes a long way.
The major failing here is that the team does not control the repository. Rather, it is an arbitrary resource in the cloud. This eliminates the team's ability to test artifacts prior to including them in the DSL, and there's no guarantee that an artifact used by a production release today will still be available from the repository tomorrow unaltered and intact.
Bring the Repository in HouseWhat's needed is to bring the repository in house. While this can be done without special tooling, there are a handful of decent Maven repositories that provide the team more control. Nexus and Artifactory are two of the major choices here. They provide security over who can publish artifacts to the repository as well as more sophisticated storage and indexing than a simple file system.
With one of these repositories, the team has it's own DSL they
control and can ensure that only approved versions of libraries are
made available to the team. Again, as long as the team does not use
snapshot dependencies a basic audit trail of what is in a build is
available. Impact analysis reports indicating which projects have used
a given component version are a little more difficult. Protecting the
code-base from a developer checking an arbitrary library into source
control is still a manual process, but Maven does discourage the
practice by making reuse of a library from the repository more
With an in-house repository, a team is able to extend Maven nicely. Dependency management is controlled by developers (for better or worse) but dependencies must come from an approved collection of versioned components controlled by the team. These libraries can be tested prior to being made available. Protection and audit capabilities are less than ideal but not wholly absent. For many teams, this kind of system may be adequate. The most disciplined are likely to find it wanting.
Other Concerns with Maven Dependency Management
No ability to depend on the newest version with traceabilityMaven draws a strict line between depending on a released version of a component and depending on the latest built version. Depending on the latest build version is a common practice in multi-component continuous integration. Maven's strict line allows for either approach to be used, but only provides traceability when dependencies are based on releases. In multi-component CI, teams often do not want to release particular components for reuse upstream, but rather consider every successful CI build (with passing unit tests) a mini-release.
Functional tests performed at a system level determine the quality of
the entire dependency graph (the application). In these cases, it's
only after testers validate a system build, that a release will be
pushed out (ideally the binary that was tested) and full traceability
to every component is valuable. That Maven discourages mutli-component
CI through the arbitrary separation of Releases and Snapshots is
Inability to depend based on status
Sometimes, teams don't want to depend on either the latest build of a component nor a specific version. Rather, they want to depend on the latest version that has passed some sort of quality gate. Perhaps a validation from a testing team, or approval from a release manager. Maven has no real concept of statuses on components (version numbers and component names are pretty much it) so status based dependencies are out of the question.