[HIVE-7012] Wrong RS de-duplication in the ReduceSinkDeDuplication Optimizer

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.13.0
• Fix Version/s: 0.14.0
• Component/s: Query Processor
• Assignee: Unassigned
• Reporter: Sun Rui
• Created: 03/May/14 09:20
• Updated: 09/Aug/22 06:03
• Resolved: 13/May/14 16:18

Description

With HIVE 0.13.0, run the following test case:

create table src(key bigint, value string);

select  
   count(distinct key) as col0
from src
order by col0;

The following exception will be thrown:

java.lang.RuntimeException: Error in configuring object
	at org.apache.hadoop.util.ReflectionUtils.setJobConf(ReflectionUtils.java:93)
	at org.apache.hadoop.util.ReflectionUtils.setConf(ReflectionUtils.java:64)
	at org.apache.hadoop.util.ReflectionUtils.newInstance(ReflectionUtils.java:117)
	at org.apache.hadoop.mapred.ReduceTask.runOldReducer(ReduceTask.java:485)
	at org.apache.hadoop.mapred.ReduceTask.run(ReduceTask.java:420)
	at org.apache.hadoop.mapred.Child$4.run(Child.java:255)
	at java.security.AccessController.doPrivileged(Native Method)
	at javax.security.auth.Subject.doAs(Subject.java:396)
	at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1121)
	at org.apache.hadoop.mapred.Child.main(Child.java:249)
Caused by: java.lang.reflect.InvocationTargetException
	at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
	at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:39)
	at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
	at java.lang.reflect.Method.invoke(Method.java:597)
	at org.apache.hadoop.util.ReflectionUtils.setJobConf(ReflectionUtils.java:88)
	... 9 more
Caused by: java.lang.RuntimeException: Reduce operator initialization failed
	at org.apache.hadoop.hive.ql.exec.mr.ExecReducer.configure(ExecReducer.java:173)
	... 14 more
Caused by: java.lang.RuntimeException: cannot find field _col0 from [0:reducesinkkey0]
	at org.apache.hadoop.hive.serde2.objectinspector.ObjectInspectorUtils.getStandardStructFieldRef(ObjectInspectorUtils.java:415)
	at org.apache.hadoop.hive.serde2.objectinspector.StandardStructObjectInspector.getStructFieldRef(StandardStructObjectInspector.java:150)
	at org.apache.hadoop.hive.ql.exec.ExprNodeColumnEvaluator.initialize(ExprNodeColumnEvaluator.java:79)
	at org.apache.hadoop.hive.ql.exec.GroupByOperator.initializeOp(GroupByOperator.java:288)
	at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:376)
	at org.apache.hadoop.hive.ql.exec.mr.ExecReducer.configure(ExecReducer.java:166)
	... 14 more

This issue is related to HIVE-6455. When hive.optimize.reducededuplication is set to false, then this issue will be gone.

Logical plan when hive.optimize.reducededuplication=false;

src 
  TableScan (TS_0)
    alias: src
    Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
    Select Operator (SEL_1)
      expressions: key (type: bigint)
      outputColumnNames: key
      Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
      Group By Operator (GBY_2)
        aggregations: count(DISTINCT key)
        keys: key (type: bigint)
        mode: hash
        outputColumnNames: _col0, _col1
        Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
        Reduce Output Operator (RS_3)
          istinctColumnIndices:
          key expressions: _col0 (type: bigint)
          DistributionKeys: 0
          sort order: +
          OutputKeyColumnNames: _col0
          Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
          Group By Operator (GBY_4)
            aggregations: count(DISTINCT KEY._col0:0._col0)
            mode: mergepartial
            outputColumnNames: _col0
            Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
            Select Operator (SEL_5)
              expressions: _col0 (type: bigint)
              outputColumnNames: _col0
              Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
              Reduce Output Operator (RS_6)
                key expressions: _col0 (type: bigint)
                DistributionKeys: 1
                sort order: +
                OutputKeyColumnNames: reducesinkkey0
                OutputVAlueColumnNames: _col0
                Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
                value expressions: _col0 (type: bigint)
                Extract (EX_7)
                  Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
                  File Output Operator (FS_8)
                    compressed: false
                    Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
                    table:
                        input format: org.apache.hadoop.mapred.TextInputFormat
                        output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
                        serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe

You will see that RS_3 and RS_6 are not merged.

Logical plan when hive.optimize.reducededuplication=true;

src 
  TableScan (TS_0)
    alias: src
    Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
    Select Operator (SEL_1)
      expressions: key (type: bigint)
      outputColumnNames: key
      Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
      Group By Operator (GBY_2)
        aggregations: count(DISTINCT key)
        keys: key (type: bigint)
        mode: hash
        outputColumnNames: _col0, _col1
        Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
        Reduce Output Operator (RS_3)
          istinctColumnIndices:
          key expressions: _col0 (type: bigint)
          DistributionKeys: 1
          sort order: +
          OutputKeyColumnNames: reducesinkkey0
          Statistics: Num rows: 0 Data size: 0 Basic stats: NONE Column stats: NONE
          Group By Operator (GBY_4)
            aggregations: count(DISTINCT KEY._col0:0._col0)
            mode: mergepartial
            outputColumnNames: _col0
            Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
            Select Operator (SEL_5)
              expressions: _col0 (type: bigint)
              outputColumnNames: _col0
              Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
              File Output Operator (FS_8)
                compressed: false
                Statistics: Num rows: 1 Data size: 16 Basic stats: COMPLETE Column stats: NONE
                table:
                    input format: org.apache.hadoop.mapred.TextInputFormat
                    output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
                    serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe

You will see that RS_6 has been merged into RS_3. However, Obviously the merge is incorrect because RS_3 and RS_6 have different sort keys. (The sort key for RS_3 is
key and the sort key for RS_6 is count(distinct key)).

The problem is that the method sameKeys() returns the result that both RS have same keys. sameKeys() depends ExprNodeDescUtils.backtrack() to backtrack a key expr of cRS to pRS.

I don't understand the logical behind the following logic in ExprNodeDescUtils:
Why still backtrack when there is no mapping for the column of the current operator?

  private static ExprNodeDesc backtrack(ExprNodeColumnDesc column, Operator<?> current,
      Operator<?> terminal) throws SemanticException {
    ...
    if (mapping == null || !mapping.containsKey(column.getColumn())) {
      return backtrack((ExprNodeDesc)column, current, terminal);
    }
    ...
  }

The process of backtracking _col0 of cRS to pRS:
RS_6:_col0 --> SEL_5:_col0 --> GBY_4:_col0 (because the colExprMap is null for GBY_4) --> RS_3:_col0 (No mapping for output column _col0), which is a wrong backtrack.

Issue Links

• is related to: HIVE-6455 Scalable dynamic partitioning and bucketing optimization