[HIVE-8895] bugs in mergejoin

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.13.0, 0.13.1, 0.14.0
• Fix Version/s: 0.14.0
• Component/s: Query Processor
• Labels: patch
• Assignee: Unassigned
• Reporter: Wan Chang
• Created: 17/Nov/14 07:34
• Updated: 26/Nov/14 00:59
• Resolved: 26/Nov/14 00:59

Description

I got a IndexOutOfBoundsException with a SQL in hive0.13.1. But it runs well on hive0.11. Here is the example sql which can trigger the exception.

create table test_join_1(a string, b string);
create table test_join_2(a string, b string);
-- got an IndexOutOfBoundsException error
explain 
select * from

(
    SELECT a a, b b
    FROM test_join_1
)t1
join 
(
    SELECT a a, b b
    FROM test_join_1
)t2
    on  t1.a = t2.a
    and t1.a = t2.b
join
(
    select a from test_join_2
)t3 on t1.a = t3.a;

And here is some stack information:

java.lang.IndexOutOfBoundsException: Index: 1, Size: 1
    at java.util.ArrayList.rangeCheck(ArrayList.java:604)
    at java.util.ArrayList.get(ArrayList.java:382)
    at org.apache.hadoop.hive.ql.parse.SemanticAnalyzer.mergeJoins(SemanticAnalyzer.java:7403)
    at org.apache.hadoop.hive.ql.parse.SemanticAnalyzer.mergeJoinTree(SemanticAnalyzer.java:7616)
    at org.apache.hadoop.hive.ql.parse.SemanticAnalyzer.genPlan(SemanticAnalyzer.java:8946)
    at org.apache.hadoop.hive.ql.parse.SemanticAnalyzer.analyzeInternal(SemanticAnalyzer.java:9220)
    ...

But sql as below runs well.

explain select * from
(
    SELECT a a, b b
    FROM test_join_1
)t1
join 
(
    SELECT a a, b b
    FROM test_join_1
)t2
    on  t1.a = t2.a
    and t2.a = t2.b
join
(
    select a from test_join_2
)t3 on t1.a = t3.a;

I didn't quite understand the details of mergejoin. But I noticed the patch in HIVE-5556 edited SemanticAnalyzer.java with the change below:

-    if ((targetCondn == null) || (nodeCondn.size() != targetCondn.size())) {
-      return -1;
+    if ( targetCondn == null ) {
+      return new ObjectPair(-1, null);
+    }

Maybe it's a good idea to revert the logic of the 'if' statement as before.

Issue Links

• relates to: HIVE-5556 Pushdown join conditions