[IMPALA-3057] Incorrect results joining on modulo condition

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Duplicate
• Affects Version/s: Impala 2.5.0
• Component/s: Backend
• Labels: correctness runtime-filters
• Target Version: Impala 2.5.0
• Assignee: Unassigned
• Reporter: casey
• Created: 22/Feb/16 23:49
• Updated: 26/Feb/16 04:33
• Resolved: 23/Feb/16 00:38

Description

I was messing around with a table on the nightly cluster and came across a very strange result. It was a little difficult to repro using the functional db but I think the query below has the same problem.

[localhost:21000] > use functional_parquet;
Query: use functional_parquet

[localhost:21000] > select c.id, cc.id from complextypestbl c join complextypestbl cc on (cc.id % 2) = (c.id % 2);
Query: select c.id, cc.id from complextypestbl c join complextypestbl cc on (cc.id % 2) = (c.id % 2)
+----+----+
| id | id |
+----+----+
| 1  | 7  |
| 1  | 5  |
| 1  | 3  |
| 1  | 1  |
+----+----+
Fetched 4 row(s) in 0.12s

The result doesn't make sense because the query is symmetrical.

I think this only happens when using a nested table? Also the table might need to be small. Not sure about those things though.

Version:

☁  Impala  git log -1
commit 31f005c269c584f01662cc8e68b8a440fbc21fca

Here is the query from the nightly cluster that is more obviously wrong:

[nightly-3.vpc.cloudera.com:21000] > select count(*) from customers c join customers cc on c.id = cc.id ;
Query: select count(*) from customers c join customers cc on c.id = cc.id
+----------+
| count(*) |
+----------+
| 53       |
+----------+
Fetched 1 row(s) in 0.68s

[nightly-3.vpc.cloudera.com:21000] > select count(*) from customers c join customers cc on (c.id % 2) = (cc.id % 2);
Query: select count(*) from customers c join customers cc on (c.id % 2) = (cc.id % 2)
+----------+
| count(*) |
+----------+
| 0        |
+----------+
Fetched 1 row(s) in 0.69s

[nightly-3.vpc.cloudera.com:21000] > version();
Shell version: Impala Shell v2.5.0-cdh5-INTERNAL (31f005c) built on Sun Feb 21 04:23:38 PST 2016
Server version: impalad version 2.5.0-cdh5-INTERNAL RELEASE (build 31f005c269c584f01662cc8e68b8a440fbc21fca)