Ranked in these QuestionsQuestion Ranking
Con Reinventing the wheel
Recreated a "modern" hadoop mapreduce & HDFS, except with git-like data. This ties you to their system, and cannot not run any tools such as Spark/Hive on top.
Con Low level MapReduce API
Only ways to express transforms are: map & reduce. Developer time will be spent expressing high level concepts such as count, windowing, graph traversal in this framework, vs having Spark/Hive do it.