> My next idea is that the builds are all happening in parallel and we are
> opening some db connections (derby, h2, etc) to either a port or socket on
> the build slave which is colliding between some of the builds
Looks like build #78 also has failures, so while it does not invalidate the
theory about cross build conflicts, other problems still exist (ordering of
deletes within the same build that we are working on).
I think the following invalidates this theory though - none of our in-memory
DBs are using port connections; rather they work with files. If parallel builds
were sharing workspace on the filesystem of a given build node, none of our
tests would work period ('mvn clean' of one build would delete the DB of
another build). But from what I can tell, workspace directories are separate,
hence I can't imagine how a conflict can arise.
Andrus
