UPDATE: After wrangling with them for two days they finally discovered a server that had failed in their cluster that was causing intermittent problems.