Just switched to the private repo mode to ensure ongoing build sanity (and to be able to browse "workspace" Maven deps via the web UI). Also fixed DB cleanup code in a few unit tests that resulted in random failures not related to the main code.
Finally and most importantly identified and fix a real (and a very annoying) race condition: CAY-1749. So after all there was nothing unexplainable or mysterious about our failures and now we seem to have a stream of successful builds on both 3.1 and 3.2. This makes me very happy :) What also delights me is that CAY-1749 issue proved that Jenkins can sometimes point to otherwise undetectable race conditions just by the virtue of the large number of builds shuffled across the servers, and collocated with other Apache project builds. Maybe we should take better advantage of that and write more unit tests specifically emulating concurrent scenarios. Andrus On Oct 21, 2012, at 1:34 AM, Andrus Adamchik <and...@objectstyle.org> wrote: > So for once instead of random and otherwise impossible to reproduce unit test > failures I noticed an explicit mention of low PermGen memory. So I set these > Maven args for the build: > > -Xmx512m -XX:MaxPermSize=128m > > That fixed 3.1 build: https://builds.apache.org/job/Cayenne-31/23/ > But not 3.2: https://builds.apache.org/job/Cayenne-trunk/776/ > > I also suspect that enabling "Use private Maven repository" should help with > another common class of problems (but not the one at hand for 3.2), but > figured we'll wait till the next relevant failure before changing this config. > > Andrus
