Should I use SQL JOIN or IN Clause?

Question

I have a question about best approach. I am not sure which approach is best when data is considered variable in size.

Consider the following 3 TABLES:

EMPLOYEE

EMPLOYEE_ID, EMP_NAME

PROJECT

PROJECT_ID, PROJ_NAME

EMP_PROJ (many to many of above two tables)

EMPLOYEE_ID, PROJECT_ID

Problem: Given an EmployeeID, find ALL the employees of ALL Projects that this Employee is associated with.

I have tried this in two way.. both approaches differ only by few milliseconds no matter what size of data is used.

SELECT EMP_NAME FROM EMPLOYEE
WHERE EMPLOYEE_ID IN (
    SELECT EMPLOYEE_ID FROM EMP_PROJ    
    WHERE PROJECT_ID IN (
        SELECT PROJECT_ID FROM EMP_PROJ p, EMPLOYEE e
        WHERE p.EMPLOYEE_ID = E.EMPLOYEE_ID 
        AND  E.EMPLOYEE_ID = 123)

go

select c.EMP_NAME FROM
(SELECT PROJECT_ID FROM EMP_PROJ 
  WHERE EMPLOYEE_ID = 123) a
JOIN 
EMP_PROJ b
ON a.PROJECT_ID = b.PROJECT_ID
JOIN 
EMPLOYEE c
ON b.EMPLOYEE_ID = c.EMPLOYEE_ID

As of now, I expect around 5000 Employees and Projects each.. but have no idea about what kinda many-many relationship exists. Which approach would u recommend? thanks!

EDIT: Execution Plan of Approach 1

"Hash Join  (cost=86.55..106.11 rows=200 width=98)"
"  Hash Cond: (employee.employee_id = emp_proj.employee_id)"
"  ->  Seq Scan on employee  (cost=0.00..16.10 rows=610 width=102)"
"  ->  Hash  (cost=85.07..85.07 rows=118 width=4)"
"        ->  HashAggregate  (cost=83.89..85.07 rows=118 width=4)"
"              ->  Hash Semi Join  (cost=45.27..83.60 rows=118 width=4)"
"                    Hash Cond: (emp_proj.project_id = p.project_id)"
"                    ->  Seq Scan on emp_proj  (cost=0.00..31.40 rows=2140 width=8)"
"                    ->  Hash  (cost=45.13..45.13 rows=11 width=4)"
"                          ->  Nested Loop  (cost=0.00..45.13 rows=11 width=4)"
"                                ->  Index Scan using employee_pkey on employee e  (cost=0.00..8.27 rows=1 width=4)"
"                                      Index Cond: (employee_id = 123)"
"                                ->  Seq Scan on emp_proj p  (cost=0.00..36.75 rows=11 width=8)"
"                                      Filter: (p.employee_id = 123)"

Execution Plan of Approach 2:

"Nested Loop  (cost=60.61..112.29 rows=118 width=98)"
"  ->  Index Scan using employee_pkey on employee e  (cost=0.00..8.27 rows=1 width=4)"
"        Index Cond: (employee_id = 123)"
"  ->  Hash Join  (cost=60.61..102.84 rows=118 width=102)"
"        Hash Cond: (b.employee_id = c.employee_id)"
"        ->  Hash Join  (cost=36.89..77.49 rows=118 width=8)"
"              Hash Cond: (b.project_id = p.project_id)"
"              ->  Seq Scan on emp_proj b  (cost=0.00..31.40 rows=2140 width=8)"
"              ->  Hash  (cost=36.75..36.75 rows=11 width=8)"
"                    ->  Seq Scan on emp_proj p  (cost=0.00..36.75 rows=11 width=8)"
"                          Filter: (employee_id = 123)"
"        ->  Hash  (cost=16.10..16.10 rows=610 width=102)"
"              ->  Seq Scan on employee c  (cost=0.00..16.10 rows=610 width=102)"

So looks like the Execution plan of Approach 2 is slightly better, because the 'cost' is 60 as opposed to 85 of approach 1. Is that the right way to analyze this?

How does one know it will hold true even for all sorts of many-many combinations?

Answer 1

Both approaches may cause sub-optimum performance as they both use subqueries which are optimized efficiently or not depending on the DBMS.

I would suggest to avoid subqueries and use JOIN for what is is meant to be used, like:

Edit:

I corrected and simplyfied my example:

select coll.EMP_NAME
from EMP_PROJ ep1
inner join EMP_PROJ ep2 on ep1.PROJECT_ID = ep2.PROJECT_ID
inner join EMPLOYEE coll on ep2.EMPLOYEE_ID = coll.EMPLOYEE_ID
where ep1.EMPLOYEE_ID = 123

I think one should note that is is perfectly legal to refer to the same table more than once with different aliases in one query. To the query this logically "looks" like two separate tables which happen to have the same structure and data.