Using RowNumber and Partition - sql-server

Question

Consider this code:
Select U.[user_id] As UserID
Max(AL.entry_dt) As LastLoginDate
From Users U with (nolock)
Inner Join activity_log AL with (nolock) On AL.[user_id] = U.[user_id]
And AL.activity_type = 'LOGIN'
And U.external_user = 1
Group By U.[user_id]
Having Max(al.entry_dt) < GetDate() - 30
Order By U.[user_id]
I was curious if the Row_Number / Partition could be used here? Perhaps to make this more effective, or if it can be used at all?
Essentially, I want 1 row per user with the last instance the user logged in where the user hasn't logged in during the last 30 days.
Bring on the pain.....

Answer 1

To use the result of the row_number() in a where clause, wrap the query in a subquery/derived table or common table expression:
Original answer for users that have logged in within the last 30 days:
select UserId, LastLoginDate
from (
Select
U.[user_id] As UserID
, AL.entry_dt As LastLoginDate
, rn = row_number() over(partition by u.user_id order by AL.entry_dt desc)
From Users U with (nolock)
Inner Join activity_log AL with (nolock)
On AL.[user_id] = U.[user_id]
And AL.activity_type = 'LOGIN'
And U.external_user = 1
Where AL.entry_dt > GetDate() - 30 -- swapped < for >
) sub
where rn = 1
Order By sub.[userid]
rextester demo: http://rextester.com/XZU40394
returns:
+--------+---------------+
| UserId | LastLoginDate |
+--------+---------------+
| 1 | 2017-09-13 |
| 2 | 2017-09-10 |
| 3 | 2017-09-07 |
+--------+---------------+
Updated answer for users who have not logged in in the last 30 days:
select UserId, LastLoginDate
from (
Select
U.[user_id] As UserID
, AL.entry_dt As LastLoginDate
, rn = row_number() over(partition by u.user_id order by AL.entry_dt desc)
From Users U with (nolock)
Inner Join activity_log AL with (nolock)
On AL.[user_id] = U.[user_id]
And AL.activity_type = 'LOGIN'
And U.external_user = 1
) sub
where rn = 1
and lastlogindate < getdate() - 30
Order By [userid]
rextester demo: http://rextester.com/XZU40394
returns:
+--------+---------------+
| UserId | LastLoginDate |
+--------+---------------+
| 4 | 2016-09-13 |
| 6 | 2016-09-10 |
+--------+---------------+
from test setup:
create table users (user_id int, external_user bit)
create table activity_log (user_id int, activity_type varchar(32), entry_dt date)
insert into users values (1,1),(2,1),(3,1),(4,1),(5,0),(6,1)
insert into activity_log values
(1,'login','20170913') ,(1,'login','20170912') ,(1,'login','20170911'),(1,'login','20160908')
,(2,'login','20170910') ,(2,'login','20170909') ,(2,'login','20170908')
,(3,'login','20170907') ,(3,'login','20170906') ,(3,'login','20170905')
,(4,'login','20160913') ,(4,'login','20160912') ,(4,'login','20160908')
,(5,'login','20160910') ,(5,'login','20160909') ,(5,'login','20160908')
,(6,'login','20160910') ,(6,'login','20160909') ,(6,'login','20160908')
To correct your query in the question, move your where to having like so:
Select U.[user_id] As UserID
,Max(AL.entry_dt) As LastLoginDate
From Users U with (nolock)
Inner Join activity_log AL with (nolock) On AL.[user_id] = U.[user_id]
And AL.activity_type = 'LOGIN'
And U.external_user = 1
Group By U.[user_id]
having max(al.entry_dt) < GetDate() - 30
Order By U.[user_id]

Answer 2

CROSS APPLY or OUTER APPLY allow you to return n records from correlated query for each record in related table. I think cross apply is what you want since if a user hasn't logged in in the past 30 days you don't want to see them at all in results. Cross apply similar to inner join but runs correlation query for each record related table. OUTER Apply similar to OUTER join so it returns all records from related table and only those that match in the correlated query.
So in the below example, for each user, return the top 1 record in descending order of entry_dT. for each related user. Outer apply would resemble a left join so all users would be returned even if no activity occurred.
MODIFIED DEMO: http://rextester.com/UQEI69366 (all 3 below) again thx to SQLZim for tester/data
SELECT U.[user_id] As UserID
, AL.entry_dt As LastLoginDate
FROM Users U with (nolock)
CROSS APPLY (SELECT top 1 *
FROM activity_log IAL
WHERE U.User_ID = IAL.User_ID
AND IAL.activity_type = 'LOGIN'
ORDER BY IAL.entry_DT Desc) AL
WHERE U.external_user = 1
AND IAL.entry_dt < GetDate() - 30
ORDER BY U.[user_id]
If all you're after is users who haven't logged in in the past 30 days...
a simple not exists seems like it would work. Who cares about the date time if they have; you're just after a list of users who haven't logged in in 30 days.
SELECT U.[user_id] As UserID
FROM Users U
WHERE not exists (SELECT *
FROM activity_log IAL
WHERE IAL.activity_type = 'LOGIN'
AND IAL.entry_dt > GetDate() - 30
AND IAL.[user_id] = U.[user_id])
AND U.external_user = 1
ORDER BY U.[user_id]
a simple left join would work as well (return all external users who have not had a login in 30 days from present date.
SELECT U.[user_id] As UserID
FROM Users U with (nolock)
LEFT JOIN activity_log AL
ON AL.[user_id] = U.[user_id]
AND AL.activity_type = 'LOGIN'
AND AL.entry_dt > GetDate() - 30
WHERE U.external_user = 1
and AL.user_ID is null
ORDER BY U.[user_id]

Answer 3

I was curious if the Row_Number / Partition could be used here? Perhaps to make this more effective, or if it can be used at all?
I prefer group by in your case than row number since row number needs additional index than group by.Read below to know more
Assuming you use the same query you posted,below are the indexes needed
for users table..
create index nci_test on
dbo.usertable(userid,external_login)
For activity log table, you will need to know more about the data..
Ex:
if join filters out more rows than where,then index can be
create index nci_test1 on
dbo.actvititlog(userid,entry_Dt,activity_type )
if entry_dt column filters out more rows,then leading column can be entry_Dt in above index
if you use RowNumber,it will need a POC index and your query spreads across two tables,so this can't be done