Called on master.
Called on master.
Execute the operator.
Execute the operator. This should recursively execute parent operators.
Create the new output tuple based on the input elements and return it.
Create the new output tuple based on the input elements and return it.
The join sequence generally looks like:
join (filter value eval) ... / \ join (filter value eval) \ / \ \ table1(col1,col2..) table2(colx, coly..) table3(cola, colb..) ...
[input] represents values of all input tables columns
Return the join tag.
Return the join tag. This is usually just 0. ReduceSink might set it to something else.
Initialize an array of ExprNodeEvaluator from start, for specified length and return the result ObjectInspectors.
Initialize an array of ExprNodeEvaluator from start, for specified length and return the result ObjectInspectors.
Initialize an array of ExprNodeEvaluator and return the result ObjectInspectors.
Initialize an array of ExprNodeEvaluator and return the result ObjectInspectors.
Initialize an array of ExprNodeEvaluator and put the return values into a StructObjectInspector with integer field names.
Initialize an array of ExprNodeEvaluator and put the return values into a StructObjectInspector with integer field names.
Copy from the org.
Copy from the org.apache.hadoop.hive.ql.exec.ReduceSinkOperator Initializes array of ExprNodeEvaluator. Adds Union field for distinct column indices for group by. Puts the return values into a StructObjectInspector with output column names.
If distinctColIndices is empty, the object inspector is same as
Operator#initEvaluatorsAndReturnStruct(ExprNodeEvaluator[], List, ObjectInspector)
Copy from the org.
Copy from the org.apache.hadoop.hive.ql.exec.ReduceSinkOperator Initializes array of ExprNodeEvaluator. Adds Union field for distinct column indices for group by. Puts the return values into a StructObjectInspector with output column names.
If distinctColIndices is empty, the object inspector is same as
Operator#initEvaluatorsAndReturnStruct(ExprNodeEvaluator[], List, ObjectInspector)
Recursively calls initializeOnMaster() for the entire query plan.
Recursively calls initializeOnMaster() for the entire query plan. Parent operators are called before children.
Initialize the operator on master node.
Initialize the operator on master node. This can have dependency on other nodes. When an operator's initializeOnMaster() is invoked, all its parents' initializeOnMaster() have been invoked.
Initialize the operator on slave nodes.
Initialize the operator on slave nodes. This method should have no dependency on parents or children. Everything that is not used in this method should be marked @transient.
Stream through the large table and process the join using the hash tables.
Stream through the large table and process the join using the hash tables. Note that this is a specialized processPartition that accepts an extra parameter for the hash tables (built from the small tables).
Return the parent operators as a Java List.
Return the parent operators as a Java List. This is for interoperability with Java. We use this in explain's Java code.
Called on master.
Called on master.
Process a partition.
Process a partition. Called on slaves.
Copy the table(input) columns value to the output tuple
Copy the table(input) columns value to the output tuple
A join operator optimized for joining a large table with a number of small tables that fit in memory. The join can be performed as a map only job that avoids an expensive shuffle process.
Different from Hive, we don't spill the hash tables to disk. If the "small" tables are too big to fit in memory, the normal join should be used anyway.