[HIVE-17403] Fail concatenation for unmanaged and transactional tables

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 1.3.0, 2.4.0, 3.0.0
• Fix Version/s: 2.4.0, 3.0.0
• Target Version/s: 2.4.0, 3.0.0
• Assignee: Prasanth Jayachandran
• Reporter: Prasanth Jayachandran
• Created: 29/Aug/17 01:58
• Updated: 22/May/18 23:15
• Resolved: 11/Sep/17 20:21

Description

ALTER TABLE .. CONCATENATE should fail if the table is not managed by hive.

For unmanaged tables, file names can be anything. Hive has some assumptions about file names which can result in data loss for unmanaged tables.

Example of this is a table/partition having 2 different files files (part-m-00000_1417075294718 and part-m-00018_1417075294718). Although both are completely different files, hive thinks these are files generated by separate instances of same task (because of failure or speculative execution). Hive will end up removing this file

2017-08-28T18:19:29,516 WARN  [b27f10d5-d957-4695-ab2a-1453401793df main]: exec.Utilities (:()) - Duplicate taskid file removed: file:/Users/table/part=20141120/.hive-staging_hive_2017-08-28_18-19-27_210_3381701454205724533-1/_tmp.-ext-10000/part-m-00018__1417075294718 with length 958510. Existing file: file:/Users/table/part=20141120/.hive-staging_hive_2017-08-28_18-19-27_210_3381701454205724533-1/_tmp.-ext-10000/part-m-00000__1417075294718 with length 1123116

DDL should restrict concatenation for unmanaged tables.

Issue Links

• is duplicated by: HIVE-17280 Data loss in CONCATENATE ORC created by Spark
• links to: RB