Mysql在最近的日期（今天，昨天或之前）查询内部联接

Question

I'm attempting to pull the latest pricing data from a table on an Inner Join. Prices get updated throughout the day but aren't necessary updated at midnight.

The following query works great when the data is updated on prices by the end of the day. But how do I get it to get yesterdays data if today's data is blank?

I'm indexing off of a column that is formatted like this date_itemnumber => 2015-05-22_12341234

SELECT h.*, collection.*, history.price
FROM collection
INNER JOIN h ON collection.itemid=h.id
INNER JOIN history ON collection.itemid=history.itemid 
AND concat('2015-05-23_',collection.itemid)=history.date_itemid
WHERE h.description LIKE '%Awesome%'

Production Query time: .046 sec

To be clear, I want it to check for the most up to date record for that item. Regardless on if it is today, yesterday or before that.

SQLFiddle1

The following query gives me the desired results but with my production dataset it takes over 3 minutes to return results. As my dataset gets larger, it would take longer. So this can't be the most efficient way to do this.

SELECT h.*, collection.*, history.price
FROM collection
INNER JOIN h ON collection.itemid=h.id
INNER JOIN history ON collection.itemid=history.itemid 
AND (select history.date_itemid from history WHERE itemid=collection.itemid GROUP BY date_itemid DESC LIMIT 1)=history.date_itemid 
WHERE h.description LIKE '%Awesome%'

Production Query time: 181.140 sec

SQLFiddle2

Answer 1

SELECT x.*
  FROM history x
  JOIN 
     ( SELECT itemid
            , MAX(date_itemid) max_date_itemid 
         FROM history 
  -- optional JOINS and WHERE here --
        GROUP
           BY itemid
     ) y
    ON y.itemid = x.itemid
   AND y.max_date_itemid = x.date_itemid;

http://sqlfiddle.com/#!9/975f5/13

Answer 2

This should works:

SELECT h.*, collection.*, history.price
FROM collection
INNER JOIN h ON collection.itemid=h.id
INNER JOIN(
SELECT a.*
  FROM history a
  INNER JOIN 
     ( SELECT itemid,MAX(date_itemid) max_date_itemid
         FROM history 
        GROUP BY itemid
     ) b ON b.itemid = a.itemid AND b.max_date_itemid = a.date_itemid
  ) AS history ON history.itemid = collection.itemid
WHERE h.description LIKE '%Awesome%'

I don't know if this take a lot of execution time. Please do try it, since you might have more data in your tables it will be a good test to see the query execution time.

Answer 3

This is actually a fairly common problem in SQL, at least I feel like I run into it a lot. What you want to do is join a one to many table, but only join to the latest or oldest record in that table.

The trick to this is to do a self LEFT join on the table with many records, specifying the foreign key and also that the id should be greater or less than the other records' ids (or dates or whatever you're using). Then in the WHERE conditions, you just add a condition that the left joined table has a NULL id - it wasn't able to be joined with a more recent record because it was the latest.

In your case the SQL should look something like this:

SELECT h.*, collection.*, history.price
FROM collection
INNER JOIN h ON collection.itemid=h.id
INNER JOIN history ON collection.itemid=history.itemid 
-- left join history table again
LEFT JOIN history AS history2 ON history.itemid = history2.itemid AND history2.id > history.id
-- filter left join results to the most recent record
WHERE history2.id IS NULL
AND h.description LIKE '%Awesome%'

Answer 4

This is another approach that cuts one inner join statement

select h.*,his.date_itemid, his.price from history his
INNER JOIN h ON his.itemid=h.id
WHERE his.itemid IN (select itemid from collection) AND h.description LIKE '%Awesome%' and his.id IN (select max(id) from history group by history.itemid)

you can try it here http://sqlfiddle.com/#!9/837a8/1

Answer 5

I am not sure if this is what you want but i give it a try EDIT: modified

   CREATE VIEW LatestDatesforIds
AS
SELECT
  MAX(`history`.`date_itemid`) AS `lastPriceDate`,
  MAX(`history`.`id`) AS `matchingId`
FROM `history`
GROUP BY `history`.`itemid`;  


CREATE VIEW MatchDatesToPrices
AS
SELECT
  `ldi`.`lastPriceDate` AS `lastPriceDate`,
  `ldi`.`matchingId` AS `matchingId`,
  `h`.`id` AS `id`,
  `h`.`itemid` AS `itemid`,
  `h`.`price` AS `price`,
  `h`.`date_itemid` AS `date_itemid`
FROM (`LatestDatesforIds` `ldi`
  JOIN `history` `h`
    ON ((`ldi`.`matchingId` = `h`.`id`)));

SELECT c.itemid,price,lastpriceDate,description
FROM collection c
INNER JOIN MatchDatesToPrices mp
     ON c.itemid = mp.itemid
INNER JOIN h ON c.itemid = h.id

Answer 6

Difficult to test the speed on such a small dataset but avoiding 'Group By' might speed things up. You could try conditionally joining the history table to itself instead of Grouping?

eg

SELECT h.*, c.*, h1.price
FROM h


INNER JOIN history h1 ON h1.itemid = h.id
LEFT OUTER JOIN history h2 ON h2.itemid = h.id
    AND h1.date_itemid < h2.date_itemid
INNER JOIN collection c ON c.itemid = h.id

WHERE h2.id IS NULL
AND h.description LIKE '%Awesome%'

Changing this line

AND h1.date_itemid < h2.date_itemid

to actually work on a sequential indexed field (preferably unique) will speed things up too. eg order by id ASC