Question

I have a database in Microsoft SQL Server 2008

I have a table [eventlog].[dbo].[USER_OPERATION] with columns [userID], [event_description], [event_date], [event_type], [eventID]

eventID is unique for, well, every given event that can happen.

A given UserID is, of course, not unique for events (every user may have a lot of events associated with him) but is associated with a single individual user

What I want is to create a query that would give me a list which only has the latest event for every user (that is to say, every UserID) and its associated info (the event_type, eventID and event_description of a given specific event).

To illustrate:

When performing

SELECT * 
FROM [eventlog].[dbo].[USER_OPERATION] 
ORDER BY userID ASC

I get something along the lines of

|====================================================================================|
|  eventID  |  userID  |        event_description        |  event_date  | event_type |
|           |          |                                 |              |            |
|   123     |    2     |  USER 2 broke something         |  03.11.11    |   CRASH    |
|   391     |    2     |  USER 2 filed a complaint       |  30.04.10    |  COMPLAINT |
|   392     |    2     |  USER 2 has bought beer         |  31.10.09    |  PURCHASE  |
|   32      |    3     |  USER 3 broke something         |  22.10.11    |   CRASH    |
|   568     |    4     |  USER 4 has requested support   |  05.12.11    |  SUPP_REQ  |
|   691     |    4     |  USER 4 has bought beer         |  01.12.10    |  PURCHASE  |
|   81      |    4     |  USER 4 updated personal data   |  17.07.11    |  PDAT_UPD  |
|   141     |    5     |  USER 5 has bought beer         |  16.08.11    |  PURCHASE  |
|   142     |    5     |  USER 5 broke something         |  16.08.11    |   CRASH    |
|   269     |    6     |  USER 6 updated personal data   |  27.01.12    |  PDAT_UPD  |
|   845     |    7     |  USER 7 updated personal data   |  27.01.12    |  PDAT_UPD  |
|           |          |                                 |              |            |
|====================================================================================|

As you can see, some users have multiple events with different dates associated with them.

What I want is a query that will show me a list of users and what was the most recent date users had an event on (the output would, essentially, be a table which lists each users once, and shows the most recent event associated with each user, and associated event_description, event_date and event_type info).

Let’s call such a query result “Recent Event Table”.

Do note the “unusual condition” of User 5 (UserID 5) who broke something and bought beer at the same date.

In such cases, I don’t care WHICH of the two same-day events would go into the “Recent Event Table”, it can be picked by random or whatever (though I would still would need the associated event_description and event_type info).

Ideally, the result would look like this (for same set of users as above):

|====================================================================================|
|  eventID  |  userID  |        event_description        |  event_date  | event_type |
|           |          |                                 |              |            |
|   123     |    2     |  USER 2 broke something         |  03.11.11    |   CRASH    |
|   32      |    3     |  USER 3 broke something         |  22.10.11    |   CRASH    |
|   568     |    4     |  USER 4 has requested support   |  05.12.11    |  SUPP_REQ  |
|   141     |    5     |  USER 5 has bought beer         |  16.08.11    |  PURCHASE  |
|   269     |    6     |  USER 6 updated personal data   |  27.01.12    |  PDAT_UPD  |
|   845     |    7     |  USER 7 updated personal data   |  27.01.12    |  PDAT_UPD  |
|           |          |                                 |              |            |
|====================================================================================|

If there is no way to just “pick any one of the two randomly or per some rule” for such “date-dupes” as User 5, having two entries in the “Recent Event Table” would be acceptable for such special cases, since they are very rare and I can handle them manually.

In such (a little bit less fortunate) case the “Recent Event Table” would look like

|====================================================================================|
|  eventID  |  userID  |        event_description        |  event_date  | event_type |
|           |          |                                 |              |            |
|   123     |    2     |  USER 2 broke something         |  03.11.11    |   CRASH    |
|   32      |    3     |  USER 3 broke something         |  22.10.11    |   CRASH    |
|   568     |    4     |  USER 4 has requested support   |  05.12.11    |  SUPP_REQ  |
|   141     |    5     |  USER 5 has bought beer         |  16.08.11    |  PURCHASE  |
|   142     |    5     |  USER 5 broke something         |  16.08.11    |   CRASH    |
|   269     |    6     |  USER 6 updated personal data   |  27.01.12    |  PDAT_UPD  |
|   845     |    7     |  USER 7 updated personal data   |  27.01.12    |  PDAT_UPD  |
|           |          |                                 |              |            |
|====================================================================================|

Which is also acceptable (but would need a bit additional pruning later).

So, to summarize my question, is it possible to construct such a Microsoft SQL query that would give me a Recent Event Table along the lines of what was described above ?

Thank you very much for your help in advance

Answer 1

You can use a CTE (Common Table Expression) combined with the ROW_NUMBER() function:

;WITH CTE AS 
(
    SELECT  
       *,
       RowNum = ROW_NUMBER() OVER (PARTITION BY UserID 
                                   ORDER BY event_date DESC, event_id DESC)
    FROM [eventlog].[dbo].[USER_OPERATION] 
)
SELECT *
FROM CTE
WHERE RowNum = 1

This “partitions” your data into groups – one each for every UserID – and then sorts the events inside that group of data by event_date DESC and event_id DESC) and numbers them – the most recent entry (for each user) gets RowNum = 1 – so just select those from the CTE and you’re done!