ANTISEMIJOIN (U-SQL)

  • [アーティクル]

Summary

Anti-Semijoins are U-SQL’s way filter a rowset based on the absence of its rows in another rowset. Other SQL dialects express this with the SELECT * FROM A WHERE A.key NOT IN (SELECT B.key FROM B) pattern. There are two variants: LEFT ANTISEMIJOIN and RIGHT ANTISEMIJOIN.

A LEFT ANTISEMIJOIN (or just ANTISEMIJOIN) gives only those rows in the left rowset that have no matching row in the right rowset.

The RIGHT ANTISEMIJOIN gives only those rows in the right rowset that have no matching row in the left rowset.

The join expression in the ON clause specifies how to determine the match.

Examples

Given the following rowsets:

EmpName DepID
Rafferty 31
Jones 33
Heisenberg 33
Robinson 34
Smith 34
Williams null
DeptID DepName
31 Sales
33 Engineering
34 Clerical
35 Marketing

The following query finds all employees that are not in a valid department by finding all the employees in the left @employees rowset that do not have a depID that is listed in the right @departments rowset):

@employees = SELECT *  
               FROM (VALUES   
                      ("Rafferty", (int?) 31)  
                    , ("Jones", (int?) 33)  
                    , ("Heisenberg", (int?) 33)  
                    , ("Robinson", (int?) 34)  
                    , ("Smith", (int?) 34)  
                    , ("Williams", (int?) null)) AS E(EmpName, DepID);  
                      
@departments = SELECT *  
                FROM (VALUES  
                       ((int) 31, "Sales")  
                     , ((int) 33, "Engineering")  
                     , ((int) 34, "Clerical")  
                     , ((int) 35, "Marketing")) AS D(DepID, DepName);  
                       
@emps_notin_valid_dept =  
    SELECT e.EmpName, e.DepID  
    FROM @employees AS e  
         LEFT ANTISEMIJOIN (SELECT (int?) DepID AS DepID, DepName FROM @departments) AS d  
         ON e.DepID == d.DepID;  
  
  
OUTPUT @emps_notin_valid_dept   
TO "/output/rsLeftAntiSemiJoinEmployeesNotInValidDept.csv"  
USING Outputters.Csv();

The resulting rowset looks like:

EmpName DepID
Williams null

The following query finds all departments without an employee:

@employees = SELECT *  
               FROM (VALUES   
                      ("Rafferty", (int?) 31)  
                    , ("Jones", (int?) 33)  
                    , ("Heisenberg", (int?) 33)  
                    , ("Robinson", (int?) 34)  
                    , ("Smith", (int?) 34)  
                    , ("Williams", (int?) null)) AS E(EmpName, DepID);  
                      
@departments = SELECT *  
                FROM (VALUES  
                       ((int) 31, "Sales")  
                     , ((int) 33, "Engineering")  
                     , ((int) 34, "Clerical")  
                     , ((int) 35, "Marketing")) AS D(DepID, DepName);  
                       
@depts_without_emps =  
    SELECT d.DepName, d.DepID  
    FROM @employees AS e  
         RIGHT ANTISEMIJOIN (SELECT (int?) DepID AS DepID, DepName FROM @departments) AS d   
         ON e.DepID == d.DepID;  
  
OUTPUT @depts_without_emps   
TO "/output/rsRightAntiSemiJoinDeptsWithoutEmployees.csv"  
USING Outputters.Csv();

Both queries return the same rowset:

DepName DepID
Marketing 35

Additional Example
An example of comparing ANTISEMIJOIN against a subquery with NOT IN is available at Subqueries with IN/NOT IN and SEMIJOIN/ANTISEMIJOIN.

See Also