[MRQL-81] Fix the coercion of persistent collections to in-memory bags

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.9.6
• Component/s: Run-Time/Flink, Run-Time/MapReduce, Run-Time/Spark
• Assignee: Leonidas Fegaras
• Reporter: Leonidas Fegaras
• Created: 06/Feb/16 17:35
• Updated: 11/Feb/16 14:42
• Resolved: 11/Feb/16 14:42

Description

Currently, query results that are persistent bags (such as Spark RDDs or Flink DataSets) are coerced to in-memory bags before they are printed. This needs to be done at any place in the query result. This came out when we used matrix factorization (query/factorization.mrql) because its output are the two matrix factors (datasets) in a tuple. It was reported by Ahmed Ulde. It will be fixed by adding a new case for tuple projection in the query evaluator that coerces persistent datasets to bags when they appear in the query output.