Question

I am running a script on our production database reffering two tables : our table of users (3700 of them) and the table of quotes that they have made (280000 of them). Quote is the main object in our application, a very large object, for whom many data tables are created and filled. My goal is to clean database from all quotes but those made of a small group of users.

I first create a temp table containing ids of those users (it is used else in the script also) and then a cursor that runs through the main table for the quotes, where they are listed, and for those quotes created from the user group does the necessary cleansing.

I see that this script is going to be executed for 26 hours approximately, which I consider peculiar since I need about 15 minutes for the database restoring in general, and I guess the heaviest sql is executed there. The db, though, weighs more than 100GB.

Is there some part of the script that I am making terribly non-optimal, or you have some suggestion how this could be done with much shorter execution.

We are running SQL Server 2008 R2.

Here’s the sketch of the script.

CREATE table #UsersIdsToStay(user_id int)
INSERT INTO #UsersIdsToStay
select user_id 
from users 
where user_name like '%SOMESTRING '
-----
declare @QuoteId int
declare @UserId int

declare QuoteCursor cursor for 
select DISTINCT QuoteId, UserId
from QuotesTable
where UserId not in 
    (
        select * from #UsersIdsToStay
    )

open QuoteCursor
while 1=1
begin
    fetch QuoteCursor into @QuoteId, @UserId
    if @@fetch_status != 0 break

    -- all the deletions from related tables are executed here using @QuoteId and @UserId
    exec('delete from QuoteHistory where QuoteId = ' + @QuoteId + ' and UserId = ' + @UserId )
    exec('delete from QuoteRevisions where QuoteId = ' + @QuoteId + ' and UserId = ' + @UserId )
    exec('delete from QuoteItems where QuoteId = ' + @QuoteId + ' and UserId = ' + @UserId )
    ....

end
close QuoteCursor;
deallocate QuoteCursor

Answer 1

The cursor restricts you to only delete a single User_Id/Quote_Id combination at a time on each related table. By using joins you will be able to delete in mass.

You could also switch out the temp table with a Common Table Expression(CTE). If this is a one off script the temp table should be ok, but for production code I would create a CTE.

    if OBJECT_ID('tempdb..#quotesToDelete') is not null
            drop table #quotesToDelete


    select distinct 
                ut.user_id, 
                qt.quote_id
        into #quotesToDelete
    from    dbo.QuotesTable qt (nolock)
        inner join dbo.UsersTable ut (nolock)
            on qt.user_id = ut.user_id
    where ut.user_name not like '%SOMESTRING '

    -- all the deletions from related tables are executed here using @QuoteId and @UserId

    -- relatedtableA
    delete a
    from relatedtableA a
        inner join #quotesToDelete b
        on a.user_id = b.user_id
        and a.quote_id = b.quote_id

    -- relatedtableB
    ...