使用UNIONS优化MySQL查询

Question

I've got a query (which works) to show the average number of replies to messages containing specific keywords.

But I think it could be optimised - we're worried that as the database grows this query will become very slow, especially if we use a larger date range.

Here's a sample of the current query:

SELECT 'text1' AS "text", ROUND(AVG (a.rt_count),2) AS "Average over 100 days"
FROM 
(SELECT COUNT(reply.id) AS rt_count
FROM message LEFT OUTER JOIN reply ON (message.id = reply.message)
WHERE message.text LIKE '%text1%'
AND message.created_date >= (CURDATE() - INTERVAL 100 DAY)
GROUP BY message.id
) AS a
UNION
SELECT 'text2' AS "text", ROUND(AVG (a.rt_count),2) AS "Average over 100 days"
FROM 
(SELECT COUNT(reply.id) AS rt_count
FROM message LEFT OUTER JOIN reply ON (message.id = reply.message)
WHERE message.text LIKE '%text2%'
AND message.created_date >= (CURDATE() - INTERVAL 100 DAY)
GROUP BY message.id
) AS a 
UNION
SELECT 'text3' AS "text", ROUND(AVG (a.rt_count),2) AS "Average over 100 days"
FROM 
(SELECT COUNT(reply.id) AS rt_count
FROM message LEFT OUTER JOIN reply ON (message.id = reply.message)
WHERE message.text LIKE '%text3%'
AND message.created_date >= (CURDATE() - INTERVAL 100 DAY)
GROUP BY message.id
) AS a 

As you can see, the only thing that really changes is the WHERE message.text LIKE '%text1%' which adds lots of redundant code. Any optimisation ideas? - all advice much appreciated

Answer 1

SELECT case when message.text like '%text1%'
            then 'text1'
            when message.text like '%text2%'
            then 'text2'
            when message.text like '%text3%'
            then 'text3'
       end AS "text", 
       ROUND(AVG (a.rt_count),2) AS "Average over 100 days"
FROM 
(SELECT COUNT(reply.id) AS rt_count
FROM message LEFT OUTER JOIN reply ON (message.id = reply.message)
WHERE (message.text like '%text1%'
or message.text like '%text2%'
or message.text like '%text3%')
AND message.created_date >= (CURDATE() - INTERVAL 100 DAY)
GROUP BY message.id
) AS a 

Answer 2

To make your query perform better, use UNION ALL instead of UNION . This will eliminate the unnecessary step of eliminating duplicates.

If one row can match at most one of the texts, then the following will work:

SELECT MatchText AS "text", ROUND(AVG (a.rt_count),2) AS "Average over 100 days"
FROM (SELECT MatchText, m.Id, COUNT(reply.id) AS rt_count
      FROM (select m.*,
                   (case when m.text like '%text1%' then 'Text1'
                         when m.text like '%text2%' then 'Text2'
                         when m.text like '%text3%' then 'Text3'
                    end) as MatchText
            from message m
           ) m LEFT OUTER JOIN
           repl
           ON (m.id = reply.message)
      WHERE MatchText is not NULL AND
            message.created_date >= (CURDATE() - INTERVAL 100 DAY)
      group by MatchText, m.Id
     ) t
group by MatchText