real life example:

My client had an Orders table where online shop orders  were stored while his requirement was to create a graph that will show daily and monthly orders, so the question was how to group dates when order time was different? To convert it to varchar and then to remove time or to just group by day(orderDate),month(orderDate),yeaR(orderDate) ? Way too long, here where this trick came in use for. Basically what it does is gets number of days from zero day inside of datediff function and then adds this amount zero day to get smalldatetime type back. This trick works with days and month just instead of dd call mm

code:

select sum(productPrice) as productPrice,

count(productPrice) as numOfProductsSold

from Orders

group by dateadd(dd,datediff(dd,0,orderDate),0)

real life example:

My client’s community website had a sql table OnlineUsers where every logged in user was stored on login with time user logged in or moved a page last time which was an indication that user is still online. Previously a trigger to delete inactive users was a login of some other user. So each time other user logged in, he triggered delete statement on table where userLastActivityTime was more than 30 minutes ago. In some situation each minute about 20 users users logged in and at night there were almost none. So i came up with idea to create a scheduled sql task that runs each two minutes and deletes users that made no action in 30 minutes. It reduces number of requests in daytime and made night results more precise. In general it reduces amount of unwanted transactions in x times.

pictures:

[View with PicLens]

real life example :

problem i faced lately was relationship icons near each profile for community website when users had a relationships (ie. sent messages), messages were stored in “Messages” table and my requirement was to show near each user’s profile whether logged user had a relationship with them and what kind. So you could say just have a JOIN between “Users” and “Messages” sql tables but first of i had 4 kinds of relationships and second of all think what kind of optimization would it be to have even one JOIN between two tables with more than hundreds of thousands records each. This ain’t working unless you are ready to buy super servers to store databases on. You could also say create a function that checks for each returned from query user what kind of relationship logged user has with which is fine but i this way i would have many double records like UserA and UserB could have messaged each other, added to friend or blacklist.

So the idea was to create a Relationships sql table which holds User1,User2,Relationship1,Relationship2,Relationship3,Relationship4. When any Relationship column was an integer “0” for no relationship, 1 for User1’s action, 2 for User2’s action and 3 for mutual action ( both users messaged each other).

You  can say now cool, this what i should do in the first place but if you work on website optimization like i did? What if you are not the original designer?

Now what ? Problem is that there is a message record for each message sent between users so every time message was sent the sender was User1 and the receiver was User2 which created lots of records for single relationship when i needed create one field  that contains all the relationships between two users. This where for statement comes handy. First of all i created a stored procedure that had a conditional logic, checked if there is a relationship between users and then updated or inserted upon request. Now all i need is a for loop though Messages table so it will send User1 and User2 to my stored procedure, this is where cursors came to save me because otherwise i would be needed to create C# function to make the conditional logic which would be many times slower. Cursors are not advisable for real time results as well simply because they are very slow and CPU demanding compared to more elegant solutions in sql or just proper design but in off-line changes like these i could allow to have cursor for loop but again the are much faster than C# accessing the database and manipulating it.

here is the code

code :

-- setting nocount on so the query will be faster as won't return number of rows affected
set nocount on
-- declaration of two int variables which will hold current userID's
declare @userA int, @userB int
--declaration of the cursor that will loop though table in it's declaration
declare cur cursor for select fieldA,fieldB from tableA
-- opening cursor
open cur
-- get first result from table
fetch next from cur into @userA ,@userB
-- loop until there are no rows left
while @fetch_status = 0
begin
-- stored procedure with current userIDs
exec someStoredProcedure @userA ,@userB
-- get next result
fetch next from cur into @userA ,@userB
end
-- closing cursor
close cur
-- removing it from sql server memory
deallocate cur

real life example:

My client had a CostumerDeals table with large amount of records in it, each time a costumer logged in to check his deals this large table was looped though. So my idea was to create a trigger (see schedule MSSQL query task) that will move (insert into new table and then delete) costumers deals which are older than 6 month to CostumerDeals_old table so the records amount decreased in 90% which made faster responses but also made me to program a trigger for removal of older than 6 month user deal and in case user returns to get his deal records back. It was time consuming but worth it.

code :

use dbName
insert into tableA (colAA,colAB,colAC.....)
select colBA,colBB,colBC.... from tableB where dateadd(mm,6,dealDate)<getdate()