[HIVE-7237] hive.exec.parallel=true w/ Hive 0.13/Tez causes application to linger forever

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.13.0
• Fix Version/s: 0.14.0
• Component/s: Tez

Environment

HDP 2.1, Hive 0.13, SLES 11, 128GB data nodes, ORC SNAPPY

• Assignee: Navis Ryu
• Reporter: Douglas Moore
• Created: 16/Jun/14 15:01
• Updated: 13/Nov/14 19:43
• Resolved: 24/Jun/14 00:58

Description

set hive.exec.parallel=true; will cause the Yarn application instance to linger
forever. set hive.exec.parallel=false, the application goes away as soon as hive query is complete. The underlying table is an ORC store_sales table compressed with SNAPPY.

hive.exec.parallel=true;
select * from store_sales where ss_ticket_number=5741230 and ss_item_sk=4825

The query will run under Tez and finish << 30 seconds.

After 30-40 of these jobs the cluster gets to a point where no jobs will finish.