> they are also HashJoins, so memory concerns are being looked at (the >logs seem to be shouting something about that). > > but I wanted to double check if broadcasting to two vertices from a >single has known issues.
Hive has multi-output hash-join plans. http://people.apache.org/~gopalv/union-all-dag-join.png They work as long as the operator pipeline doesn¹t have submarine assumptions, hive-1.0 had issues with not building input hashtable (i.e don¹t use ³vertex name² as a unique key for anything). Cheers, Gopal
