优化 MySQL - JOIN 与嵌套查询

Question

基于联接表比嵌套查询更有效的假设，我一直在尝试优化一些 SQL 查询。我多次加入同一个表以对数据执行不同的分析。

我有 2 张桌子：

交易：

id    |   date_add    |   merchant_ id    | transaction_type      |     amount
1         1488733332          108                  add                     20.00
2         1488733550          108                 remove                   5.00

还有一个只列出日期的日历表，以便我可以在特定日期没有交易的情况下创建空记录：

日历：

id     |    datefield
1           2017-03-01
2           2017-03-02
3           2017-03-03
4           2017-03-04

我在交易表中有数千行，我正在尝试获取每月总交易和不同类型交易的年度摘要（即总共 12 行），其中

• 交易 = 所有“金额”的总和，
• additions = 所有“金额”的总和，其中 transaction_type = “add”
• 赎回 = 所有“金额”的总和，其中 transaction_type = “删除”

结果：

month     |    transactions     |    additions  |   redemptions
Jan                15                  12               3
Feb                20                  15               5
...

我的初始查询如下所示：

SELECT  COALESCE(tr.transactions, 0) AS transactions, 
        COALESCE(ad.additions, 0) AS additions, 
        COALESCE(re.redemptions, 0) AS redemptions, 
        calendar.date 
FROM (SELECT DATE_FORMAT(datefield, '%b %Y') AS date FROM calendar WHERE datefield LIKE '2017-%' GROUP BY YEAR(datefield), MONTH(datefield)) AS calendar 
LEFT JOIN (SELECT COUNT(transaction_type) as transactions, from_unixtime(date_add, '%b %Y') as date_t FROM transactions WHERE merchant_id = 108  GROUP BY from_unixtime(date_add, '%b %Y')) AS tr
ON calendar.date = tr.date_t
LEFT JOIN (SELECT COUNT(transaction_type = 'add') as additions, from_unixtime(date_add, '%b %Y') as date_a FROM transactions WHERE merchant_id = 108  AND transaction_type = 'add' GROUP BY from_unixtime(date_add, '%b %Y')) AS ad
ON calendar.date = ad.date_a
LEFT JOIN (SELECT COUNT(transaction_type = 'remove') as redemptions, from_unixtime(date_add, '%b %Y') as date_r FROM transactions WHERE merchant_id = 108  AND transaction_type = 'remove' GROUP BY from_unixtime(date_add, '%b %Y')) AS re
ON calendar.date = re.date_r

我尝试对其进行优化和清理，删除嵌套语句并提出以下建议：

SELECT 
    DATE_FORMAT(cal.datefield, '%b %d') as date,
    IFNULL(count(ct.amount),0) as transactions, 
    IFNULL(count(a.amount),0) as additions, 
    IFNULL(count(r.amount),0) as redeptions
FROM calendar as cal 
LEFT JOIN transactions as ct ON cal.datefield = date(from_unixtime(ct.date_add))  && ct.merchant_id = 108
LEFT JOIN transactions as r ON r.id = ct.id && r.transaction_type = 'remove'
LEFT JOIN transactions as a ON a.id = ct.id && a.transaction_type = 'add' 
WHERE cal.datefield like '2017-%'
GROUP BY month(cal.datefield)

我惊讶地发现修改后的语句比我的数据集的原始语句慢了大约 20 倍。我错过了某种逻辑吗？考虑到我要多次加入同一个表，是否有更好的方法通过更简化的查询来实现相同的结果？

编辑： 因此，为了进一步解释我正在寻找的结果 - 我希望一年中的每个月都有一行（12 行），每个月有一列用于显示每个月的总交易、总添加和总赎回。

第一个查询我在大约 0.5 秒内得到结果，但第二个查询我在 9.5 秒内得到结果。

Answer 1

查看您的查询您可以使用单个左连接与 case when

SELECT  COALESCE(t.transactions, 0) AS transactions, 
        COALESCE(t.additions, 0) AS additions, 
        COALESCE(t.redemptions, 0) AS redemptions, 
        calendar.date 
FROM (SELECT DATE_FORMAT(datefield, '%b %Y') AS date 
          FROM calendar 
          WHERE datefield LIKE '2017-%' 
          GROUP BY YEAR(datefield), MONTH(datefield)) AS calendar 
LEFT JOIN 
 ( select 
      COUNT(transaction_type) as transactions
      , sum( case when transaction_type = 'add' then 1 else 0 end ) as additions
      , sum( case when transaction_type = 'remove' then 1 else 0 end ) as redemptions
      ,  from_unixtime(date_add, '%b %Y') as date_t 
      FROM transactions 
      WHERE merchant_id = 108  
      GROUP BY from_unixtime(date_add, '%b %Y' ) t ON calendar.date = t.date_t

Answer 2

首先，我将从您的calendar 表中创建一个包含每个月时间戳范围的派生表。这样，如果 date_add 被索引，则与 transactions 表的连接将是有效的。

select month(c.datefield) as month, 
       unix_timestamp(timestamp(min(c.datefield), '00:00:00')) as ts_from,
       unix_timestamp(timestamp(max(c.datefield), '23:59:59')) as ts_to
from calendar c
where c.datefield between '2017-01-01' and '2017-12-31'
group by month(c.datefield)

将其与transaactions 表连接并使用条件聚合来获取您的数据：

select c.month,
       sum(t.amount) as transactions,
       sum(case when t.transaction_type = 'add'    then t.amount else 0 end) as additions,
       sum(case when t.transaction_type = 'remove' then t.amount else 0 end) as redemptions
from (
    select month(c.datefield) as m, date_format(c.datefield, '%b') as `month`
           unix_timestamp(timestamp(min(c.datefield), '00:00:00')) as ts_from,
           unix_timestamp(timestamp(max(c.datefield), '23:59:59')) as ts_to
    from calendar c
    where c.datefield between '2017-01-01' and '2017-12-31'
    group by month(c.datefield), date_format(c.datefield, '%b')
) c
left join transactions t on t.date_add between c.ts_from and c.ts_to
where t.merchant_id = 108
group by c.m, c.month
order by c.m