PostgreSQL equivalent for SQL Server GROUP BY WITH ROLLUP

Question

I have an Sql Server Query that is using the ROLLUP clause while grouping. I want an equivalent query in Postgres. Query in SQl Server is:

SELECT (CASE WHEN acnt_dba_name Is Null THEN 'Total' ELSE acnt_dba_name END) as account,
       (CASE WHEN evt_name Is Null THEN '' ELSE evt_name END) as event, 
       COUNT(CASE reg_is_complete WHEN true THEN 1 ELSE Null END) as regsComplete,
       COUNT(CASE WHEN reg_frn_pro_id > 0 AND reg_is_complete = false THEN 1 ELSE Null END) as regsInComplete,
       COUNT(CASE WHEN reg_frn_pro_id > 0 THEN Null ELSE 1 END) as regsClicks
FROM      registrations_view 
LEFT JOIN events ON (evt_id = reg_frn_evt_id)
LEFT JOIN accounts ON (acnt_id = evt_frn_acnt_id)
WHERE reg_date_created < #CreateODBCDate(url.endDate)#
AND   reg_date_created > #CreateODBCDate(url.startDate)#
AND   reg_is_active = true      -- only active regs
AND   reg_is_test = false       -- only live registrations
-- AND  reg_is_denied = false         -- exclude denied reg statuses (include these for now RWB 8/7/2)
GROUP BY rollup(acnt_dba_name, evt_name)
-- Sort with Nulls at the bottom
ORDER BY acnt_dba_name, evt_name

Solution 1

with detail as (
    select
        acnt_dba_name as account,
        evt_name as event, 
        count(case reg_is_complete when true then 1 else null end) as regscomplete,
        count(case when reg_frn_pro_id > 0 and reg_is_complete = false then 1 else null end) as regsincomplete,
        count(case when reg_frn_pro_id > 0 then null else 1 end) as regsclicks
    from
        registrations_view 
        left join
        events on evt_id = reg_frn_evt_id
        left join
        accounts on acnt_id = evt_frn_acnt_id
    where
        reg_date_created < #CreateODBCDate(url.endDate)#
        AND reg_date_created > #CreateODBCDate(url.startDate)#
        and reg_is_active = true        -- only active regs
        and reg_is_test = false     -- only live registrations
    group by acnt_dba_name, evt_name
), account as (
    select
        account,
        '' as event, 
        sum(regscomplete) as regscomplete,
        sum(regsimcomplete) as regsincomplete,
        sum(regsclicks) as regsclicks
    from detail
    group by account
), total as (
    select
        'Total' as account,
        '' as event, 
        sum(regsComplete) as regsComplete,
        sum(regsImComplete) as regsInComplete,
        sum(regsClicks) as regsClicks
    from account
)
select * from detail
union
select * from account
union
select * from total
order by account, event

Solution 2

This is a generic answer to the equivalent of roll up in Postgresql.

Given a table t:

create table t (l1 char(1), l2 char(1), i integer);
insert into t (l1, l2, i) values
('A', 'X', 1),
('A', 'Y', 2),
('B', 'X', 3),
('B', 'Y', 4);

And this SQL Server query: SQL Fiddle

select l1, l2, sum(i) total
from t
group by rollup(l1, l2)

This is how to do it in Postgresql: SQL Fiddle

Create aggregating CTEs starting from the detail going to the top level:

with detail as (
    select l1, l2, sum(i) total
    from t
    group by l1, l2
), l2 as (
    select l1, sum(total) total
    from detail
    group by l1
), l1 as (
    select sum(total) total
    from l2
)

Notice that for performance the next up level aggregates from the previous down level.

With that done just union the CTEs creating the appropriate labels:

select l1, l2, total
from detail
union
select l1, 'Total', total
from l2
union
select 'Total', '', total
from l1
order by l1, l2

Comments

[Karl Bartel]

You should use UNION ALL instead of UNION, so the DB won't have to try to remove duplicates.