SQL JOIN + GROUPBY select data from row with MAX(Date) - sql

Question

I'm having trouble figuring out the solution to this SQL query.
Schema
Edit: Adding Item Table
Item Table
PK ItemID
lots of other columns
Linking Table
FK ItemID uniqueidentifier
FK TransactionID uniqueidentifier
Transaction Table
PK ID uniqueidentifier
EntryDateTime DateTime
(several other rows of int, varchar...)
Edit : I think I haven't made the relationships clear. Each ITEM (table not shown) can have multiple transactions. Multiple items can share the same transaction (hence the linking table).
Please see the bottom for my current Query. I have left this striked to show the progression of the question.
I want to do something like this query. The trick is I want the t.varchar and t.int columns to be whatever values are in the MAX(t.EntryDateTime) row. I don't even know if group by is the right way to do this query.
SELECT lt.ItemID, MAX(t.EntryDateTime), t.varchar, t.int
FROM LinkingTable lt
LEFT JOIN Transactions t ON lt.TransactionID = t.ID
GROUP BY lt.ItemID
This table is going to be joined against in this SQL query, so please try and give me the most performant solution . Assume Table1 will contain millions of records.
SELECT
(many columns)
FROM Table1
LEFT JOIN Table2 ON Table1.Table2ID = Table2.ID
LEFT JOIN Table3 ON ....
LEFT JOIN Table4 ON (Table2.ID = Table4.Table2ID and Table4.LocaleID = 127 and Table4.Type = 0)
LEFT JOIN **the query above** AS vTable1 ON vTable1.ItemID = Table1.ID
WHERE Table1.CheckID IN (SELECT ID FROM Checks WHERE ....)
Edit : This is the query I have that is working, but I'm not sure its the most efficient. LinkingTable has ~ 200k records and its taking 6 seconds to run.
SELECT DISTINCT lt.ItemID, t.EntryDateTime, t.varchar, t.int
FROM LinkingTable lt
LEFT JOIN Transactions t ON t.id = (SELECT Top 1 t2.id FROM LinkingTable lt2
LEFT JOIN Transactions t2 on lt2.TransactionID = t2.ID
where lt2.ItemID = lt.ItemID ORDER BY t2.PrintTime DESC)

Answer 1

Try this,
SELECT i.*, outerT.EntryDateTime, outerT.varchar, outerT.int
FROM Item i
LEFT JOIN
(SELECT ItemId AS outerItemId, EntryDateTime, varchar, int
FROM (SELECT ROW_NUMBER() OVER (PARTITION BY lt.ItemId ORDER BY t.EntryDateTime) AS RowNumber, lt.ItemId, t.EntryDateTime, t.varchar, t.int
FROM Tranaction t INNER JOIN LinkingTable lt ON lt.TransactionId = t.ID) innerT
WHERE RowNumber = 1) outerT ON outerT.outerItemId = Item.ID
Hope this solves your problem

Answer 2

Even with a million plus records, you will have some performance hits, but I would ensure and index on the transaction table based on the ( ItemID, Primary Key ). The reason Primary Key and not the date -- if its auto-incremented, and it's date/time stamped at time the transaction occurs, they will be in-essence, one-in-the-same. The last entry in the file will always have the latest date. That said, an ID column should be faster with index than a date/time. This also prevents need of looking at BOTH elements of most recent date, and the transaction ID associated with that date. Here is how I would FIRST attempt the query.
select
I.*,
T2.*
from
Item I
JOIN
( select T.ItemID, MAX( T.PrimaryKey ) as LastEntryPerItem
from Transactions T
group by T.ItemID ) MaxPerItem
ON I.ItemID = T.ItemID
JOIN Transactions T2
on MaxPerItem.LastEntryPerItem = T2.PrimaryKey
order by
whatever

Answer 3

select lt.ItemId, t.entrydatetime, t.varchar, t.int
from LinkingTable lt
left join transactions t
on lt.transactionId = t.id
and t.entryDateTime = (select max(t.EntryDateTime)
from transactions t2
where t2.id = t.id)
I had a similar question before
( SQL Join to get value belong with most recent date). There's another solution by JNK involving two joins which may be faster. I've posted below. You'll need to test to see which performs better.
select lt.ItemId, t.entrydatetime, t.varchar, t.int
from LinkingTable lt
inner join transactions t
on lt.ItemId= t.ItemId
Inner join (SELECT ItemId, MAX(entrydatetime) entrydatetime
FROM transactions t2
GROUP BY ItemId) SubQ
ON SubQ.ItemId= t.ItemId
AND SubQ.entrydatetime= t.entrydatetime

Answer 4

Why don't you create a view that has all your "many columns" and then run a query against that view?