[HIVE-6913] Hive unable to find the hashtable file during complex multi-staged map join

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.14.0
• Assignee: Brock Noland
• Reporter: Brock Noland
• Created: 15/Apr/14 15:38
• Updated: 13/Nov/14 19:43
• Resolved: 18/Apr/14 02:20

Description

If a query has multiple mapjoins and one of the tables to be mapjoined is empty, the query can result in a "no such file or directory" when looking for the hashtable.

This is because when we generate a dummy hash table, we do not close the TableScan (TS) operator for that table. Additionally, HashTableSinkOperator (HTSO) outputs it's hash tables in the closeOp method. However, when close is called on HTSO we check to ensure that all parents are closed: https://github.com/apache/hive/blob/trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/Operator.java#L333

which is not true on this case, because the TS operator for the empty table was never closed.