MySQL: 子查询优化问题，其中子查询检查超过 14000 行

Question

我需要帮助来优化下面的子查询。简而言之，我有以下查询，其中 tree table join branch table on s_id AND max timestamp of branch table基于子查询条件。

我对这个查询的结果很满意 returns。但是，此查询非常慢。瓶颈是依赖子查询（branch2），它检查超过 14000 行。如何优化子查询以加快此查询？

SELECT *                
FROM dept.tree tree              
    LEFT JOIN dept.branch branch ON tree.s_id = branch.s_id                    
        AND branch.timestamp =
            (
                SELECT MAX(timestamp)                                   
                FROM dept.branch branch2 
                WHERE branch2.s_id = tree.s_id 
                AND branch2.timestamp <= tree.timestamp
            )                
WHERE tree.timestamp BETWEEN CONVERT_TZ('2020-05-16 00:00:00', 'America/Toronto', 'UTC') 
AND CONVERT_TZ('2020-05-16 23:59:59', 'America/Toronto', 'UTC') 
AND tree.s_id IN ('459','460')                
ORDER BY tree.timestamp ASC;

Table树：

   id     box_id        timestamp       
373001645   1       2020-05-07 06:00:20 
373001695   1       2020-05-07 06:02:26
373001762   1       2020-05-07 06:05:17
373001794   1       2020-05-07 06:06:38
373001810   2       2020-05-07 06:07:21

Table 分支机构：

   id     box_id        timestamp                  data
373001345   1       2020-05-07 06:00:20     {"R": 0.114, "H": 20.808}
373001395   1       2020-05-07 06:02:26     {"R": 0.12, "H": 15.544}
373001462   1       2020-05-07 06:03:01     {"R": 0.006, "H": 55.469}
373001494   1       2020-05-07 06:04:38     {"R": 0.004, "H": 51.85}
373001496   1       2020-05-07 06:05:18     {"R": 0.02, "H": 5.8965}
373001497   1       2020-05-07 06:06:39     {"R": 0.12, "H": 54.32}
373001510   2       2020-05-07 06:07:09     {"R": 0.34, "H": 1.32}
373001511   2       2020-05-07 06:07:29     {"R": 0.56, "H": 32.7}

分支有 s_id 和时间戳索引

我使用的是5.7.25-google-log版本

EXPLAIN 给出以下内容：

    id    select_type        table     partitions    type    possible_keys    key    key_len    ref    rows    filtered     Extra
    1   PRIMARY tree        range   unique_timestamp_s_id,idx_s_id_timestamp,idx_timestamp  idx_s_id_timestamp  10      2629    100.00  Using index condition; Using filesort
    1   PRIMARY branch      ref unique_timestamp_s_id,idx_timestamp unique_timestamp_s_id   5   func    1   100.00  Using where
    2   DEPENDENT SUBQUERY  branch2     ref unique_timestamp_s_id,idx_s_id_timestamp,idx_timestamp  idx_s_id_timestamp  5   tree.s_id   14122   33.33   Using where; Using index

Answer 1

请提供SHOW CREATE TABLE.

branch 需要 INDEX(s_id, timestamp)

你需要LEFT吗？它可能会无缘无故地减慢查询速度。

一列上的 IN 和另一列上的 BETWEEN 的组合可能优化不佳；你是什么版本运行?

请提供EXPLAIN SELECT，以便我们讨论优化是否良好。如果不是，我们可以讨论如何将 IN（OR 的变体）变成 UNION。

这实际上可能比我上面考虑的方法更快...

有上面的索引，然后显着重写查询：

SELECT b.*
    FROM ( SELECT  s_id,
                   MAX(timestamp) as timestamp                                
               FROM dept.branch 
               WHERE timestamp BETWEEN
                      CONVERT_TZ('2020-05-16 00:00:00', 'America/Toronto', 'UTC')
                  AND CONVERT_TZ('2020-05-16 23:59:59', 'America/Toronto', 'UTC')
                 AND s_id IN ('459','460')                
         ) AS x
    JOIN dept.branch AS b  USING(s_id, timestamp)

首先，看看它是否获得了正确的信息。然后我将解释如何在子查询中执行 UNION（如果您需要帮助）。

Answer 2

这样会更快：

select
    tree.s_id, tree.timestamp, branch.data
from
    (
        SELECT
            tree.s_id, tree.timestamp, max(branch.timestamp) as max_branch_timestamp
        FROM
            dept.tree tree
                LEFT JOIN dept.branch branch
                ON(
                        branch.s_id      =  tree.s_id
                    and branch.timestamp <= tree.timestamp
                )
        WHERE
            tree.timestamp BETWEEN
                CONVERT_TZ('2020-05-16 00:00:00', 'America/Toronto', 'UTC') AND
                CONVERT_TZ('2020-05-16 23:59:59', 'America/Toronto', 'UTC')
        AND tree.s_id IN ('459','460')                
        group by tree.s_id, tree.timestamp
    ) tree
        left outer join branch
        on(
                branch.s_id      = tree.s_id
            and branch.timestamp = tree.max_branch_timestamp
        )

MySQL: 子查询优化问题，其中子查询检查超过 14000 行

MySQL: Subquery optimization problem where subquery checks over 14000 rows

mysql

subquery

query-optimization

query-performance

groupwise-maximum