Optimize Query - select statement in case when

Question

I am trying to compare two sets of transactions. One is from a store log and the other is from the accounting records. I have a with statement that creates a table comparing both transactions and includes any that do not match. This table is called 'compare'

Now I am trying to figure out why certain values may not be present in compare so I am adding a case when statement. Here is what I have so far

select distinct *, 
case
   when (store_trans_id is not null and acctng_trans_id is not null) then 'Match'
   when (acctng_trans_id is null) then
      case 
        when exists (select 1 from compare cc where c.store_trans_id = cc.store_trans_id 
                                              and c.store_amount = cc.store_amount * -1) 
        then 'Store item transaction reversed'
      end
   else 'Research further'
end as 'Comparison'
from compare c

So when I only have the first case (Match), my query runs fast. When I add the second one it slows down. Essentially, what I am checking is if there is a store transaction that was rung up by the cashier and then canceled. If so, then the amount would be the same just the opposite ($5 versus -$5) and the store transaction id would be the same. I wouldn't expect a reversed transaction to be on the accounting record, since the item has not left the store. Any ideas on how to optimize the query run time? I have already added the additional case and moved from is not null to the when exists statement.

Answer 1

this isn't tested of course - let's say I'm taking a stab at it but you will have to let me know how it works.

main idea: get rid of the correlation subqueries and use sum/group by instead.

There should be 3 cases that may or may not cover all cases (I suppose they should but you can test out that theory by changing my inner join to a left join and see if there are null results on the left side).

Clearly, indexing may help a lot. Such as an index on store_trans_id (including accting_trans_id and store_amount)

I don't know what (if) there is a primary key in the table so in the queries where I have c.PK <> c2.PK that means the primary key on the table - though in this case the criteria can probably can also be omitted as the sum would still come out to 0 or non-zero and that's all we care about.

select c.*, c2.[Comparison]
from compare c
inner /* left */ join
(
    -- case 1: acctg_trans_id is not null
    select 
        store_trans_id,
        'Match' as [Comparison]
    from compare c
    where 
        store_trans_id is not null 
        and acctng_trans_id is not null
    union all
    -- case 2: acctg_trans_id is null and two store amounts sum to 0
    select 
        c.store_trans_id,
        'Store item transaction reversed' as [Comparison]
    from 
        compare c
        inner join compare c2
        on c.store_trans_id = c2.store_trans_id
        and c.PK <> c2.PK
    where 
        c.acctng_trans_id is null
        having sum(c.store_amount) = 0
    union all
    -- case 3: acctg_trans_id is null and two store amounts do not sum to 0
    select 
        c.store_trans_id,
        'Research further' as [Comparison]
    from 
        compare c
        inner join compare c2
        on c.store_trans_id = c2.store_trans_id
        and c.PK <> c2.PK
    where 
        c.acctng_trans_id is null
    having 
        sum(c.store_amount) <> 0
) c2
on c.store_trans_id = c2.store_trans_id