Question

I have a table that I use to keep track of customer usage. I track the number of hits the user does per day, so I have a table that looks like this:

CustID (uniqueidentifier, not null)

UseDate (smalldatetime, not null)

NumHits (smallint, not null)

I use this SQL in a stored proc to insert a row for today (if needed), or to increment the counter for today:

declare @today datetime
set @today = getdate()

/* Try to bump it by one if an entry for today exists */
if (
    select count(*) from CustomerUsage
    where CustID = @cust_guid and year(UseDate) = year(@today) and month(UseDate) = month(@today) and day(UseDate) = day(@today)
    ) = 0

    insert into CustomerUsage (CustID, UseDate, NumHits) values (@cust_guid, getdate(), 1)

else
    update CustomerUsage set NumHits = NumHits + 1 
    where CustID = @cust_guid and year(UseDate) = year(@today) and month(UseDate) = month(@today) and day(UseDate) = day(@today)

Is there a better way to do this?

Also, I want to move to an environment where multiple web servers can call the stored proc for the same customer. I think this code might be vulnerable to multithreading issues.

Thanks!

Answer 1

You can do like this.

declare @Today smalldatetime = dateadd(d, datediff(d, 0, getdate()), 0)
declare @Tomorrow smalldatetime = dateadd(d, 1, @Today)

insert into CustomerUsage(CustId, UseDate, NumHits )
select Data.CustID, Data.UseDate, Data.NumHits
from (select
        @cust_guid as CustID,
        getdate() as UseDate,
        0 as NumHits) as Data
where not exists (select *
                  from CustomerUsage with (updlock, serializable)
                  where
                    CustID = @cust_guid and
                    UseDate >= @Today and
                    UseDate < @Tomorrow)        


update CustomerUsage 
set NumHits = NumHits + 1 
where
  CustID = @cust_guid and
  UseDate >= @Today and
  UseDate < @Tomorrow

First insert a row with NumHits = 0. The insert checks if there already exists a row for that CustID on that UseDate and only inserts if necessary.

After the insert always increment NumHits by 1.

This is basically the same idea as Thomas non merge answer but I do the insert before the update.

Edit 1 Added Table Hints to the insert statements where not exists part. serializable is needed to prevent primary key constraint violation and updlock is needed to avoid deadlocks.