[IMPALA-3071] Incorrect assignment of inner join on-clause predicate that references an outer-joined table.

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 2.5.0
• Fix Version/s: Impala 2.5.0, Impala 2.3.4
• Component/s: Frontend
• Labels: correctness
• Target Version: Impala 2.5.0, Impala 2.3.4
• Assignee: Alexander Behm
• Reporter: Alexander Behm
• Created: 24/Feb/16 08:30
• Updated: 22/Mar/18 23:13
• Resolved: 01/Mar/16 06:10

Description

Query and incorrect plan:

select 1 from functional.alltypestiny a
left outer join functional.alltypestiny b
  on a.id = b.id
inner join functional.alltypestiny c
  on b.int_col < 0

+----------------------------------------------------------+
| Explain String                                           |
+----------------------------------------------------------+
| Estimated Per-Host Requirements: Memory=64.00MB VCores=2 |
|                                                          |
| 07:EXCHANGE [UNPARTITIONED]                              |
| |                                                        |
| 04:NESTED LOOP JOIN [CROSS JOIN, BROADCAST]              |
| |                                                        |
| |--06:EXCHANGE [BROADCAST]                               |
| |  |                                                     |
| |  02:SCAN HDFS [functional.alltypestiny c]              |
| |     partitions=4/4 files=4 size=460B                   |
| |                                                        |
| 03:HASH JOIN [LEFT OUTER JOIN, BROADCAST]                |
| |  hash predicates: a.id = b.id         <-- b.int_col < 0 should also be in otherPredicates                 |
| |                                                        |
| |--05:EXCHANGE [BROADCAST]                               |
| |  |                                                     |
| |  01:SCAN HDFS [functional.alltypestiny b]              |
| |     partitions=4/4 files=4 size=460B                   |
| |     predicates: b.int_col < 0                          |
| |                                                        |
| 00:SCAN HDFS [functional.alltypestiny a]                 |
|    partitions=4/4 files=4 size=460B                      |
+----------------------------------------------------------+

The result set should clearly be empty because there are no rows that satisfy b.int_col < 0, but the returned result set is non-empty.