Major Description
OLAP query easily triggers OOM, I found a possible reason is https://github.com/apache/tinkerpop/blob/master/hadoop-gremlin/src/main/java/org/apache/tinkerpop/gremlin/hadoop/structure/io/ObjectWritable.java#L72 dump a huge string output. See what I captureed from SparkUI. In fact, that toString() does not make much sense for OLAP processing except for displaying in UI. Furthermore, those too long string is not readable for people. So, ideally the toString() for Map or Collection should be optimized to be java class + size(). Then the OOM will disappear.
