[IMPALA-5452] Nested subplans with non-trivial plan tree returns inconsistent results

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: Impala 2.9.0
• Fix Version/s: Impala 2.10.0
• Component/s: Infrastructure
• Labels: testing
• Target Version: Impala 2.10.0
• Epic Color: ghx-label-7
• Assignee: Alexander Behm
• Reporter: Anuj Phadke
• Created: 07/Jun/17 18:39
• Updated: 18/Aug/17 05:38
• Resolved: 18/Aug/17 05:37

Description

This particular query returned 18 rows consistently on running tests locally.

Query: select count(*)
from tpch_nested_parquet.customer c
inner join c.c_orders o
where c_custkey < 10 AND o.pos in
 (select lead(l.l_linenumber) over (order by l.pos)
  from o.o_lineitems l)
Query submitted at: 2017-06-07 11:20:39 (Coordinator: http://anuj-OptiPlex-9020:25000)
Query progress can be monitored at: http://anuj-OptiPlex-9020:25000/query_plan?query_id=a14392f4fe88572a:9cad5eb200000000
+----------+
| count(*) |
+----------+
| 18       |
+----------+

But here -
http://jenkins.impala.io:8080/job/gerrit-verify-dryrun/696/console
The same query returned 19 rows.

08:25:10 ] -- executing against localhost:21000
08:25:10 ] select count(*)
08:25:10 ] from tpch_nested_parquet.customer c
08:25:10 ] inner join c.c_orders o
08:25:10 ] where c_custkey < 10 AND o.pos in
08:25:10 ]   (select lead(l.l_linenumber) over (order by l.pos)
08:25:10 ]    from o.o_lineitems l);
08:25:10 ] 
08:25:10 ] MainThread: Comparing QueryTestResults (expected vs actual):
08:25:10 ] 18 != 19

This test was introduced as part of this commit -
https://gerrit.cloudera.org/#/c/916/4/testdata/workloads/functional-query/queries/QueryTest/subplans.test

As per this test, the query should have returned 14 rows. These tests have not been running for a while and we failed to catch this.