avoid full index scan

Question

I use test db from MySQL employees - test db I want to optimize query

SELECT emp_no, SUM(salary)
FROM salaries
WHERE from_date < '1999-01-01'
group by emp_no;

Query cost: 287790

which indexes can help me ?

I try to create index with emp_no and salary, and emp_no and from_date but there is no result. There is a full scan index.

Also tried to use OVER(PARTITION BY) instead GROUP BY

SELECT emp_no, SUM(salary) OVER (PARTITION by emp_no)
FROM salaries  
WHERE from_date < '1999-01-01'; 

to avoid full index scan for instance or use OVER instead GROUP BY

Answer 1

To get an execution plan that avoids a full scan, MySQL would need an index that has from_date as the leading column.

Optimally, the index would contain all of the other columns referenced in the query, to avoid looking up values in the underlying data pages.

Note that if the execution plan makes use of that index, then it will need to invoke a "Using filesort" operation to satisfy the GROUP BY.

Answer 2

It depends on what version you are running.

Older version: INDEX(from_date, emp_no, salary) is "covering" and will do a "range" scan of part of the index. However the GROUP BY will probably involve a sort.

Newer version: INDEX(emp_no, from_date, salary) is also "covering", but can hop through the index without having to touch every row. This can probably avoid the sort.

Here's another thing to test:

SELECT emp_no,
       ( SELECT SUM(salary) FROM salaries
             WHERE emp_no = e.emp_no 
               AND from_date < '...')
    FROM employees AS e

Note: I am assuming employees has one row per employee, unlike salaries ?? This approach avoids the GROUP BY , but has the overhead of a correlated subquery. Now salaries needs INDEX(emp_no, from_date, salary) -- in this order, and does a smaller range scan in the index each time.