如何防止此JOIN查询执行几乎全表扫描

Question

The dataset in items contains about 4.4 million rows

updated and created are indexed individually on each table

EXPLAIN SELECT i.id, j.id 
    FROM jobs j 
JOIN items i ON j.items_id = i.id 
WHERE j.updated > DATE_SUB(NOW(), INTERVAL 60 MINUTE) 
    OR j.created > DATE_SUB(NOW(), INTERVAL 60 MINUTE) 
    OR i.updated > DATE_SUB(NOW(), INTERVAL 60 MINUTE);

Here is what I get:

1   SIMPLE  i   index   PRIMARY,updated_idx                     updated_idx       5     NULL            4168353    Using index
1   SIMPLE  j   ref     items_id_idx,updated_idx,created_idx    items_id_idx      9      my_db.i.id     1          Using where

As you can see the items, while using updated_idx , still runs close to a full table scan.

Why is this happening, and is there anyway I can avoid this?

Also tried this, but it performs worse:

EXPLAIN SELECT i.id, j.id 
    FROM jobs j 
WHERE j.updated > DATE_SUB(NOW(), INTERVAL 60 MINUTE)
    OR j.created > DATE_SUB(NOW(), INTERVAL 60 MINUTE) 
    OR j.items_id IN 
         (SELECT i.id FROM items i WHERE i.updated > DATE_SUB(NOW(), INTERVAL 60 MINUTE));

Answer 1

MySql does not optimize or conditions in the where criteria too well. Rewrite your query as 3 separate selects joined with union. This way each query can utilise the appropriate indexes:

SELECT i.id, j.id 
    FROM jobs j 
JOIN items i ON j.items_id = i.id 
WHERE j.updated > DATE_SUB(NOW(), INTERVAL 60 MINUTE)
UNION DISTINCT  
SELECT i.id, j.id 
    FROM jobs j 
JOIN items i ON j.items_id = i.id 
WHERE j.created > DATE_SUB(NOW(), INTERVAL 60 MINUTE) 
UNION DISTINCT
SELECT i.id, j.id 
    FROM jobs j 
JOIN items i ON j.items_id = i.id 
WHERE i.updated > DATE_SUB(NOW(), INTERVAL 60 MINUTE);