Sunday, May 9, 2010

The b3 aggregator

The Eclipse b3 Aggregator is based on and part of the Eclipse b3 project. Eclipse b3 provides a versatile and adaptable framework supporting build, assembly and deployment processes. It supports a rich set of use cases. One of those - the aggregation of repositories - is the focus of the b3 Aggregator tool.

The Eclipse b3 Aggregator combines repositories from various sources into a new aggregated p2 repository. It can also be configured to produce a hybrid p2/Maven2 repository. There are many situations where using aggregated repositories is a good solution, here are some examples:

  1. Projects want to provide convenient access to their products - Installation instructions requiring the user to visit several repos for a complete install are not uncommon. An aggregated repo for all those locations provides a convenient one-stop-shop strategy. The aggregation can perform mirroring of all consumed p2 repos or selectively provide indirection via a composite repo.
  2. Organizations or teams want control over internally used components - It may be necessary to have gated access to relevant/"blessed" p2 repos where an organizational "healthcheck" has been performed prior to internal distribution. Furthermore, internally used aggregated repos can provide a common basis for all organizational users (i.e. for both IDE distribution as well as for content used when building internal applications).
  3. Increase repository availability - by aggregating and mirroring what is used from multiple update sites into internally controlled servers.
  4. Distributed Development Support - an overall product repository is produced by aggregating contributions from multiple teams.
  5. Owners of a p2 repo for a given project may not be in position to host all required or recommended components due to licensing issues - Buckminster's SVN support can serve as an example here, as it requires components available in the main Eclipse p2 repo as well as third-party components. Hence users have to visit several repos for a complete install.

The b3 Aggregator is focused on supporting these specific requirements, and it plays an important role in the full scope of the b3 project. The Aggregator is however used in scenarios outside of the traditional "build domain" and this has been reflected in the user interface which does not delve into the details of "building" and should therefore be easy to use by non build experts.

Functional Overview

The b3 Aggregator performs aggregation and validation of repositories. The input to the aggregator engine (that tells it what to do) is a b3aggr EMF model. Such a model is most conveniently created by using the b3 Aggregator editor. This editor provides both editing and interactive execution of aggregation commands. The editor is based on a standard EMF "tree and properties view" style editor where nodes are added and removed to form a tree, and the details of nodes are edited in a separate properties view. Once a b3aggr model has been created it is possible to use the command line / headless aggregator to perform aggregation (and other related commands). (Note that since the b3aggr is "just and EMF model", it can be produced via EMF APIs, transformation tools, etc. and thus support advanced use cases).

The model mainly consists of Contributions; specifications of what to include from different repositories, and Validation Repositories; repositories that are used when validating, but that are not included in the produced aggregation (i.e. they are not copied). The model also contains specification of various processing rules (exclusions, transformation of names, etc.), and specification of Contacts; individuals/mailing-lists to inform when processing fails.

Here are some of the important features supported by the b3 Aggregator in Eclipse 3.6M7:

  • p2 and maven2 support — the aggregator can aggregate from and to both p2 and maven2 repositories.
  • Maven2 name mapping support — names in the p2 domain are automatically mapped to maven2 names using built in rules. Custom rules are also supported.
  • Mirroring — artifacts from repositories are mirrored/downloaded/copied to a single location
  • Selective mirroring — an aggregation can produce an aggregation consisting of a mix of references to repositories and mirrored repositories.
  • Cherry picking — it is possible to pick individual items when the entire content of a repository is not wanted. Detailed picking is supported as well as picking transitive closures like a product, or a category to get everything it contains/requires.
  • Pruning — it is possible to specify mirroring based on version ranges. This can be used to reduce the size of the produced result when historical versions are not needed in the aggregated result.
  • Categorization — categorization of installable units is important to the consumers of the aggregated repository. Categories are often choosen by repository publishers in a fashion that makes sense when looking at a particular repository in isolation, but when they are combined with others it can be very difficult for the user to understand what they relate to. An important task for the constructor of an aggregation is to be able to organize the aggregated material in an easily consumable fashion. The b3 aggregator has support for category prefixing, category renaming, addition of custom categories, as well as adding and removing features in categories.
  • Validation — the b3 aggregator validates the aggregated result to ensure that everything in the repository is installable.
  • Blame Email — when issues are found during validation the aggregator supports sending emails describing the issue. This is very useful when aggregating the result of many different projects. Advanced features include specifying contacts for parts of the aggregation which is useful in large multi layer project structures where issues may related to the combination of a group of projects rather than one individual project - someone responsible for the aggregation itself should be informed about these cross-project issues. The aggregator supports detailed control over email generation including handling of mock emails when testing aggregation scripts.


The b3 aggregator documentation is available here on the Eclipse Wiki.


Dave said...

Henrik, i really appreciate all the work that has been on the aggregator. Now it's "semi" integrated into Hudson that continuously scans update sites and gets us the latest from 19 different mirrors, then it kicks an build of all those mirrors to our internal p2 repo and publishes it automatically to our stage server for testing. This product has significantly reduced my headaches with P2 and now when the build is "green" i trust the result!

keep up the good work!


C├ędric Brun said...

Hi Henrik,

I just wanted to emphasize that if you want to provide a one click installation of b3 through the Helios modeling package, go ahead and open a bug on the Eclipse amalgam project ;)