[HIVE-8197] Tez and Vectorization Insert into ORC Table with timestamp column erroneously repeats the last row's column value

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Cannot Reproduce

Environment

Tez and Vectorization.

• Assignee: Matt McCline
• Reporter: Matt McCline
• Created: 19/Sep/14 22:29
• Updated: 03/Oct/14 23:14
• Resolved: 03/Oct/14 23:14

Description

In diagnosing why a only a Tez and Vectorized query with min and max aggregates was always returning the last row read's column value, discovered the problem was in creating the test table....

CREATE TABLE alltypesorc_string STORED AS ORC AS SELECT
  ctinyint as ctinyint,
  to_utc_timestamp(ctimestamp1, 'America/Los_Angeles') as ctimestamp1,
  CAST(to_utc_timestamp(ctimestamp1, 'America/Los_Angeles') AS STRING) as stimestamp1
FROM alltypesorc WHERE ctinyint > 0
LIMIT 40;

I think it is related what Prasanth mentioned as a possibility: Saving a Timestamp as a Writable object that gets overwritten. One suspect is the Writable[] records array in VectorFileSinkOperator in the ProcessOp method. Or, perhaps it is in VectorReduceSinkOperator.

Issue Links

• is related to: HIVE-8052 Vectorization: min() on TimeStamp datatype fails with error "Vector aggregate not implemented: min for type: TIMESTAMP"