Too many order by, max, subqueries for my intellect

问题

I can not seem to work out this query. I'm sure it needs subqueries, but I am out of options. My brain can not handle this or something. I need help :)

Little intro

I have a betting odds website. Every 15 minutes I import the latest odds (win/draw/lose -- or 1/X/2) for particular events from various bookmakers.

Every row of the odds table has the odds_type ('1', 'X' or '2'), the odds_index which is the actual odds, the bookmaker_id and the event_id.

But equally important: created_at, because I need to work with the odds from the latest import. It is very important for obvious reasons.

Scenario

In the table below we are working with imported odds for event_id #1.

Needed query

1. Isolate the latest set of imported odds for event_id = 1 from all bookmakers (in this example 9 records)
2. On that set, return the highest odds_index record for odds_type '1', 'X', '2'.

Now I prefer to do this with Rails scoping, so I can use @event.best_odds1 and @event.best_odds2, but I'll take any approach if it works. Been busting my brain on this for 5 days. Need to solve it.

Result

After the query I end up with 3 records so I can display "The best odds for event #1".

The tables

Bookmakers

 ID  |  NAME
 ----------------------
 1   |  Unibet
 2   |  888
 3   |  Ladbrokes

Events

 ID  |  NAME
 --------------------------
 1   |  Holland vs Denmark
 2   |  England vs Germany
 3   |  France vs Spain

Odds

 ID  |  OT  |  OI  |  BI  |  EI  |  CREATED_AT
 ---------------------------------------------------
 (first import from the bookies)
 1   |  '1'  |  2.4  |  1  |  1  |  2010-06-10 15:00
 2   |  'X'  |  1.5  |  1  |  1  |  2010-06-10 15:00
 3   |  '2'  |  6.2  |  1  |  1  |  2010-06-10 15:00
 4   |  '1'  |  2.2  |  2  |  1  |  2010-06-10 15:58
 5   |  'X'  |  1.8  |  2  |  1  |  2010-06-10 15:58
 6   |  '2'  |  5.2  |  2  |  1  |  2010-06-10 15:58
 7   |  '1'  |  2.8  |  3  |  1  |  2010-06-10 16:56
 8   |  'X'  |  1.3  |  3  |  1  |  2010-06-10 16:56
 9   |  '2'  |  7.1  |  3  |  1  |  2010-06-10 16:56
 (last import from the bookies)
 10  |  '1'  |  2.5  |  1  |  1  |  2010-06-11 17:10
 11  |  'X'  |  1.3  |  1  |  1  |  2010-06-11 17:10
 12  |  '2'  |  6.4  |  1  |  1  |  2010-06-11 17:10
 13  |  '1'  |  2.1  |  2  |  1  |  2010-06-11 18.12
 14  |  'X'  |  1.2  |  2  |  1  |  2010-06-11 18:58
 15  |  '2'  |  6.2  |  2  |  1  |  2010-06-11 18:58
 16  |  '1'  |  1.8  |  3  |  1  |  2010-06-12 14:56
 17  |  'X'  |  2.3  |  3  |  1  |  2010-06-12 14:56
 18  |  '2'  |  5.1  |  3  |  1  |  2010-06-12 14:56

Abbreviated column names to fit on screen

OT = odds_type
OI = odds_index 
BI = bookmaker_id 
EI = event_id

回答1:

You could use row_number() twice:

select  *
from    (
        select  *
        ,       row_number() over (partition by OT order by OI desc) as rn2
        from    (
                select  *
                ,       row_number() over (partition by EI, BI, OT 
                                           order by created_at desc) as rn1
                from    Odds
                where   EI = 1 -- for event 1
                ) sub1
        where   rn1 = 1 -- Latest row per EI, BI, OT
        ) sub2
where   rn2 = 1 -- Highest OI per OT

But if the table keeps growing, this will perform badly. You could add a history table like OddsHistory, and move outdated Odds there. When only the latest Odds are in the Odds table, your query becomes much simpler.

Live example at SQL Fiddle.

来源：https://stackoverflow.com/questions/10985767/too-many-order-by-max-subqueries-for-my-intellect