[HIVE-15235] sortmergejoin can produce incorrect plan wrt dummy operators, and NPE

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Unassigned
• Reporter: Sergey Shelukhin
• Created: 17/Nov/16 23:16
• Updated: 17/Nov/16 23:29

Description

The relevant fragment of the reduce plan of a Tez job is as follows:

<MERGEJOIN>Id =13
                  <Children>
                    <SEL>Id =12
                      <Children>
                        <MAPJOIN>Id =10
                          <Children>
...
                          <\Children>
                          <Parent>Id = 12 nullId = 9 
                        <HASHTABLEDUMMY>Id =9
                          <Children>null
                          <\Children>
                          <Parent><\Parent>
                        <\HASHTABLEDUMMY><\Parent>
                        <\MAPJOIN>
                      <\Children>
                      <Parent>Id = 13 null<\Parent>
                    <\SEL>
                  <\Children>

When sortmergejoin is enabled, during initialization, dummy operators are not initialized (presumably, they are not present in the work); that results in MapJoin not being initialized, even though its proper parent is.
Manifests as an NPE

Caused by: java.lang.NullPointerException
at org.apache.hadoop.hive.ql.exec.MapJoinOperator.process(MapJoinOperator.java:350)