Optimizing Oracle CONNECT BY when used with WHERE clause

Question

Oracle START WITH ... CONNECT BY clause is applied before applying WHERE condition in the same query. Thus, WHERE constraints won't help optimize CONNECT BY.

For example, the following query will likely perform full table scan (ignoring selectivity on dept_id):

SELECT * FROM employees 
WHERE dept_id = 'SALE'
START WITH manager_id is null
CONNECT BY PRIOR employee_id = manager_id

I tried to improve performance in 2 ways:

query A:

SELECT * FROM employees 
START WITH manager_id is null AND dept_id = 'SALE'
CONNECT BY PRIOR employee_id = manager_id

query B:

SELECT * FROM (
               SELECT * FROM employees 
                WHERE dept_id = 'SALE'
              )
START WITH manager_id is null
CONNECT BY PRIOR employee_id = manager_id

While both queries did much better than original, on Oracle 10g Release 2, query B did performed much better than A.

Did you have similar performance optimization to deal with with respect to CONNECT BY and WHERE clauses? How would you explain query B doing much better than query A?

Answer 1

Query A says start with managers in the Sales department and then get all their employees. Oracle doesn't "know" that all the employees returned be the query will be in the Sales department, so it can't use that information to reduce the set of data to work with before performing the CONNECT BY.

Query B explicitly reduces the set of data to be worked on to just those employees in Sales, which Oracle can then do before performing the CONNECT BY.

Answer 2

This should give the ultimate performance:

CREATE INDEX i_employees_employee_manager_dept ON employees (employee_id,manager_id,dept_id);
CREATE INDEX i_employees_manager_employee_dept ON employees (manager_id,employee_id,dept_id);

SELECT * FROM employees  
START WITH manager_id is null AND dept_id = 'SALE' 
CONNECT BY PRIOR employee_id = manager_id AND dept_id = 'SALE' 

Note that you do need both index and both AND conditions for the optimization to work.