Artifact Repositories

The past: desktop builds uploaded to the FTP server; the present: global artifact delivery networks. Release management and artifact repositories are they way we deliver software and infrastructure today.

Ship the Bits

Once upon a time, software release management was as simple as performing a desktop build, uploading the results to a FTP server, updating the web site, and sending an email.

Software repositories can still be that simple, but the state of the art has advanced to support secure deployments for continuous delivery to multiple clusters in data centers around the world multiple times per hour!

Release Management

There are many considerations for release management today, assuming that our build candidate has passed all of the required tests to become an official release. Now that we pursue infrastructure as code, we can also build and deploy our software as infrastructure artifacts.

The simple monolithic build of static code and library dependencies has transitioned into a collection of multiple versioned artifacts which must be integrated, configured, and orchestrated as a system even before considering the deployment permutation matrix of multiple hardware, OS platform, and client versions. Considerations include:

  • Build Post Processing:
    • compression into a one file archive to save bandwidth
    • different editions (subsets of the build if they are not handled as separates)
    • check sums and code signing
    • packaging:
      • metadata for each OS or runtime
      • versus bundling an installer as a self extracting executable
    • documentation
    • licensing and potentially providing source code
    • third party licensing
    • purging old artifacts, build, and test logs
  • Distribution:
    • channels for automatic updates
    • artifact repositories versus file servers
    • caches, mirrors, and content/artifact delivery networks
    • privileged versus public official releases
    • private, internal release candidates for testing
    • metrics and logs: download counts and bandwidth constraints

Naming the Artifact and Useful Metadata

Surely naming the artifact should be straightforward if we can answer some basic questions, such as: do we have file naming conventions and are build, version, and release numbers the same thing?

You would be surprised how often the answer is… it depends! However, there are guidelines you can adopt:

While many software package formats specify their metadata requirements, a few have standardized on DOAP: Description of a Project.

Software Artifact Repositories

Feature Matrix of Archiva, Artifactory, Maven

While an engineering organization produces build artifacts, they often bundle third party open or closed source libraries. In order to speed manual developer builds and automated build systems, these external resources can be cached to increase build speed, reliability, and conserve upstream network bandwidth. Conversely, these can be used to publish artifacts for public consumption as well.

Many languages have their own packaging format for build artifacts, e.g.:

  • Java: .jar, .war, and .ear
  • JavaScript: .npm
  • PERL: CPAN modules are compressed tarballs
  • PHP: .phar
  • Python: .egg, .whl
  • Ruby: .gem

There also exist library dependency descriptions: * Bundler for Ruby gems * Berkshelf for Chef recipes

We also have operating system package management:

  • RedHat, Fedora, CentOS and related derivatives: .rpm format with yum repositories
  • Debian, Ubuntu, and related derivatives: .deb format with apt repositories and Launchpad PPA

Infrastructure artifacts follow a similar pattern: public and private container repositories as hosted services or private deployments:

Along with Vagrant box hosting:

Caveat Emptor: downloading and installing any artifact from the Internet is potential for malware, virus, and other abuse unless you know, trust, and verify the source.

Personally, I’ve had very good experience with JFrog’s Artifactory, which is a meta repository of many different formats, but you will probably outgrow the open source version when you need to handle authentication and some of the formats mentioned above, so the SaaS subscription may be a better option.

Software Distribution Networks

These are specific to build artifacts, however they can be considered a special case of content delivery networks such as Akami, CacheFly, and Fastly as well as related to content accelerators such as CloudFlare and Amazon CloudFront

One potentially biased resource gives a good overview of the market [http://www.cdnplanet.com/]

More Reading