Hierarchical connect by and start with and joins?

Question

I've got an employees table and an identifiers table. The identifiers table is hierarchical, with parents and children. Each employee has one or more identifiers, but only one identifier is considered the "primary" or root identifier for each employee. Unfortunately, the employee table might be pointing at one of the child identifier rows and not the root. I need a fast query to join the employees with their most current (root) identifier.
Here's code to define the problem.
create table employees (employeeid varchar2(8), fakeNationalID varchar2(9), empname varchar2(30));
insert into employees (employeeid, fakeNationalID, empname) values (1,'001000001','John Smith');
insert into employees (employeeid, fakeNationalID, empname) values (2,'002000002','James Jones');
create table realids (realidkey NUMBER, fakeNationalID VARCHAR2(9) not null,
   realNationalID VARCHAR2(9) UNIQUE, parent_realidkey number);
insert into realids (realidkey, fakeNationalID, realNationalID, parent_realidkey) values
   (1,'001000001','111111111',3);
insert into realids (realidkey, fakeNationalID, realNationalID, parent_realidkey) values
   (2,'002000002','222222222',null);
insert into realids (realidkey, fakeNationalID, realNationalID, parent_realidkey) values
   (3,'003000003','333333333',null);
commit;  
create or replace function get_parentid (fakeID in VARCHAR2) return varchar2 is
   tempid VARCHAR2(9);
   begin
      select realNationalID into tempid
         from (
           select realNationalID, fakeNationalID
              from realids
              start with fakeNationalID = fakeID
              connect by nocycle prior parent_realidkey = realidkey
              order by level desc)
              where rownum = 1;
      return tempid;
      exception
         when NO_DATA_FOUND then
            return NULL;
         when others then raise;
    end;
select get_parentid('001000001') from dual; -- returns 333333333 because its linked to a parent
select get_parentid('002000002') from dual; -- returns 222222222 because there is only one child
select get_parentid('003000003') from dual; -- returns 333333333 because it is the parentWhat I want is to get the highest parent node in realids for each row in employees...
This works, but is NOT very efficient:
select employeeid, get_parentid(fakeNationalID) realid, empname from employees;
employeeid   realid       empname
1            333333333     John Smith
2            222222222     James JonesYou can imagine what this would be like with 100K rows or greater. It takes about 3 minutes to run.
This seemed like a good way to do this, but with a sub query.
select e.employeeid, e.fakenationalid, e.empname, sub.realnationalid
   from employees,
      (select realidkey, fakenationalid, realnationalid, parent_realidkey
         from realids r
         start with r.fakenationalid = e.fakenationalid
         connect by prior r.parent_realidkey = r.realidkey) subUnfortunately, it produces an invalid identifier on e.fakenationalid (in the start with clause).
Anyone have any ideas on how to get the top most parent node from the realids for each row in the employees table? In real life there are 6 or more employees tables spread across several remote instances some of which point at children in the realids table and some of which point at the parents. We always want the top most parent realid. Any help would be much appreciated.

Answer 1

Hi,
Thanks for posting the sample data in such a convenient form!
It always helps to post your Oracle version, too, especially when dealing with CONNECT BY queries.
The following does what you requested in Oracle 10:
WITH     got_roots   AS
     SELECT     CONNECT_BY_ROOT     fakenationalid     AS leaf_id
     ,     realnationalid
     FROM     realids
     WHERE     CONNECT_BY_ISLEAF     = 1
     START WITH      fakenationalid IN ( SELECT  fakenationalid
                                          FROM    employees
     CONNECT BY     realidKEY     = PRIOR parent_realidkey
SELECT     e.employeeid
,     r.realnationalid
,     e.empname
FROM     employees     e
JOIN     got_roots     r     ON     r.leaf_id     = e.fakenationalid
;In any query, calling a user-defined function for each row is going to be slow. Fortunately, Oracle now has built-in functions and operators that can take the place of get_parentid. The CONNECT_BY_ROOT operator, which was introduced in Oracle 10, is the key to this problem. In Oracle 9, you can get the same results using SYS_CONNECT_BY_PATH.
It's usually faster to do the CONNECT BY query separately, and then join whatever other tables you need to the results.
You had a good idea in your last query. The problem was that sub and employees were equal tables in the FROM clause, and you can't correlate equals. You can only correlate a sub-query to its super-query. You could make that general idea work by changing sub into a scalar sub-query,which could be correlated to employees, but I think it would be a lot less efficient than what I posted above.