作为个人记录,后续再填坑
a对p是1对多 ,p对llup 1对多
SELECT
    a.id,
    p.id,
    t1.id 
FROM
    liv_series_product a
    INNER JOIN liv_product p ON p.id = a.product_id
    LEFT JOIN ( 
        SELECT llup.id, llup.product_id, llup.room_id 
        FROM liv_live_user_product llup 
        WHERE llup.room_id = 1629055336439164930 
        
        ) t1 ON t1.room_id = 1629055336439164930 
    AND t1.product_id = p.id 
WHERE
    a.series_id = 1629025986310385665 
    AND p.on_off_status = 1 
GROUP BY
    a.product_id0.8秒
 
   加上GROUP BY llup.product_id 后0.1秒
 
    
   加上grouby 后是生成了一个副表 先执行了llup表。
 
   MySQL8的 Hash Join
explain format = tree 查看,发现加了group by 的没有用上hash join
 
   没加group by的 如EXPLAIN下图第三行用上了 hash join,反而sql慢了
 
   -> Table scan on <temporary>  (cost=0.01..4578.27 rows=366062)
    -> Temporary table with deduplication  (cost=73545.80..78124.06 rows=366062)
        -> Left hash join (llup.product_id = a.product_id)  (cost=36939.55 rows=366062)
            -> Nested loop inner join  (cost=551.86 rows=32)
                -> Filter: (a.series_id = 1629025986310385665)  (cost=327.45 rows=322)
                    -> Index scan on a using 商品_pk  (cost=327.45 rows=3217)
                -> Filter: (p.on_off_status = 1)  (cost=0.60 rows=0)
                    -> Single-row index lookup on p using PRIMARY (id=a.product_id)  (cost=0.60 rows=1)
            -> Hash
                -> Filter: (llup.room_id = 1629055336439164930)  (cost=44.35 rows=11379)
                    -> Table scan on llup  (cost=44.35 rows=11379)
有索引,不走hash join。
最后的解决办法有两种:
- 加上group by 
- llup 以on关联的product_id 加上索引。 
参考文章
MySQL 8.0 hash join有重大缺陷?
MySQL8的 Hash Join
MySQL的语句执行顺序
你真的懂使用Group by?



















