Our builds fail occasionally due to validation failures caused by unavailability of w3.org. In fact, w3.org has been fed up with this kind of requests, blacklisting/throttling have been in place for some time.
The ones I've seen are coming from maven pdf plugin v.1.1 used in hadoop-tools/hadoop-distcp. The input validation has been made configurable in MPDF-39, which requires DOXIA-392 (fixed since doxia-1.1.3). Unfortunately, the maven pdf plugin 1.2 has been cooking for long time and not been released yet. May be we could force it to use a later version of doxia?