Page history Edit this page How do I edit this website?

2012-03-09 - Auto-synchronizing repositories

There are plenty of good reasons why one might need to synchronize all the branches of multiple repositories:

  • Development happens on multiple continents
  • Some servers might have some unscheduled downtime from time to time
  • Windows users might want to have a fast access to a Git-SVN mirror

This last point is quite a sore point: Git for Windows’ support for importing Subversion repositories suffers from many performance problems, ranging from the general I/O weakness of the platform to the need to use MSys Perl to be able to use the module.

Alas, a solution is nigh!

We solved exactly those problems in the ImageJ project. There is a Jenkins job to synchronize the ImageJ repositories with each other, using this script. Sadly the script is a little more complex than I hoped, but it does the job reliably and to be honest, it is a little more tricky than I first thought: when there are updates from only one of the repositories, they should be forced. However, there might be contradicting updates in which case nothing may be forced. Of course, such a script—especially when it is a little more complex than the author hoped—also has to take precautions not to be overzealous in corner-cases the author might have not thought about. That is the reason why it makes extra-sure to look for updates it might have missed in earlier rounds and does not force those updates. Jenkins executes this job every 5 minutes which is why it is very important that the much faster git:// protocol is used for fetching the branches instead of the slower ssh-based protocol used for pushing.

As for the git-svn mirror, we have a Jenkins job that too. It is driven by a—thankfully quite simple—script. This script makes sure that the remote information is set up such that all branches in refs/remotes/ (i.e. the ones maintained by git-svn) are pushed into the refs/heads/svn/ namespace. Jenkins polls the Subversion repository and only runs the job when new commits are available.

Both jobs (and likewise the siblings of the ImageJ synchronizer used to synchronize the ImgLib2 and Fiji repositories) proved quite useful and stable in our development.

The synchronizer for Git repositories was designed with flexibility in mind: the information which repositories should be synchronized is passed as command-line parameters. Should it be required, it will be easy to adjust the synchronizer for the Git-SVN mirror to be as flexible.