SQL Multiple WHERE Clauses against one table, with CASE statements - sql

Question

I wrote this query and it seems to be working to gather correct results, however, it also takes a VERY long time. I'm just wondering if there is a way to make it more efficient?
(I understand that it's inefficient because it is creating data tables and joining them back together - I just don't know how to get around it, specifically with the CASE issue included).
I'm working in Excel with an ODBC connection to AS400. The question marks allow user-entered parameters within Excel cells.
with W as
(
select yr as YEAR, pd as PERIOD, sum(amt1 + amt2 + amt3 + amt4) as SALES
from TABLE
group by yr, pd
order by yr desc, pd desc
), X as
(
select yr as YEAR, pd as PERIOD, sum(amt1 + amt2 + amt3 + amt4) as BSALES
from TABLE
where type = 'B'
group by yr, pd
order by yr desc, pd desc
), Y as
(
select
CASE WHEN pd = 1 THEN yr - 1 ELSE yr END as YEAR,
CASE WHEN pd = 1 THEN 12 ELSE pd - 1 END as PERIOD,
SUM(OM) as MODOM
from TABLE
group by yr, pd
order by yr desc, pd desc
), Z as
(
select
CASE WHEN pd = 1 THEN yr - 1 ELSE yr END as YEAR,
CASE WHEN pd = 1 THEN 12 ELSE pd - 1 END as PERIOD,
SUM(BOM) as BMODOM
from TABLE
where type = 'B'
group by yr, pd
order by yr desc, pd desc
)
select w.YEAR, w.PERIOD, w.SALES, x.BSales, y.MODOM, z.BMODOM
from w inner join x
on w.YEAR = x.YEAR and w.PERIOD = x.PERIOD
inner join y
on w.YEAR = y.YEAR and w.PERIOD = y.PERIOD
inner join z
on w.YEAR = z.YEAR and w.PERIOD = z.PERIOD
where w.YEAR between ? and ? and w.PERIOD between ? and ?
order by YEAR desc, PERIOD desc
I had to change the code slightly for privacy purposes, but I believe this all relays correctly.
Example Data:
Yr Pd Type Amt OM
18 2 A 45 181
18 2 B 33 163
18 2 A 40 153
18 1 B 39 136
18 1 B 24 142
18 1 B 53 143
18 1 A 41 186
18 1 A 78 197
17 12 A 98 139
17 12 A 54 159
17 12 B 78 181
17 12 B 45 101
17 11 A 28 134
17 11 A 77 192
17 11 A 75 110
17 11 B 60 135
17 11 B 83 170
17 10 B 72 114
17 10 A 26 118
17 10 A 95 111
17 9 A 12 112
17 9 B 14 171
Example Results
Yr Pd Sales Bsales MODOM BMODOM
18 2 118 33 804 421
18 1 235 116 580 282
17 12 275 123 741 305
17 11 323 143 343 114
17 10 193 72 283 171
Note that I need SALES at a TOTAL Level, and then at a TYPE B level. I also need OM at a TOTAL Level, and then also a Type B level - HOWEVER - I need to offset by one period. So the MODOM is for 17-10 is reflection of the OM total for period 17-9 in the table. (I hope that make sense).
EDIT I have this backward. The MODOM for 17-10 would actually reflect the OM value for 17-11, not the other way. Corrected EXPECTED RESULTS.
Yr Pd Sales Bsales MODOM BMODOM
18 2 118 33 0 0
18 1 235 116 497 163
17 12 275 123 804 421
17 11 323 143 580 282
17 10 193 72 741 305

Answer 1

So using Conditional aggregation you can at least get rid of 2 of your common table expressions which will help. Not ordering until your final presentation query instead of in each common table expression will help too. Using an actual date e.g. First of First Month of a Period could help eliminate what the potentially costly case expressions when defining previous period.
There are a few ways of writing this but this will give you an example of conditional aggregation.
And note the LEFT JOIN rather than INNER because your 1st period will always drop off your query if you use INNER
WITH PeriodSales AS(
SELECT
yr as YEAR
,pd AS PERIOD
,SUM(amt) as SALES
,SUM(CASE WHEN type = 'B' THEN amt END) as BSALES
FROM
Table
GROUP BY
yr
,pd
)
, PreviousPeriod AS
(
SELECT
CASE WHEN pd = 12 THEN yr + 1 ELSE yr END as YEAR
,CASE WHEN pd = 12 THEN 1 ELSE pd + 1 END as PERIOD
,SUM(OM) as MODOM
,SUM(CASE WHEN type = 'B' THEN OM END) as BMODOM
FROM
Table
GROUP BY
CASE WHEN pd = 12 THEN yr + 1 ELSE yr END
,CASE WHEN pd = 12 THEN 1 ELSE pd + 1 END
)
SELECT
ps.YEAR
,ps.PERIOD
,ps.SALES
,ps.BSALES
,pp.MODOM
,pp.BMODOM
FROM
PeriodSales ps
LEFT JOIN PreviousPeriod pp
ON ps.YEAR = pp.YEAR
AND ps.PERIOD = pp.PERIOD
ORDER BY
ps.YEAR DESC
,ps.PERIOD DESC
Per your edits, to align to the "Previous Period" to get to the OM amounts you want you will actually want to ADD a period not subtract one in the example I used. I have tested this and it does work. There can be many other factors for performance that we cannot discover without knowing more about the tables and execution plans etc.