包含in group by子句或partition by的case语句

Question

Customer   Decision      req_date       salary
   A       Approved     2017-06-13       1000
   A       Approved     2017-06-13       1000
   A       Pending      2017-06-13       500
   B       Pending      2017-10-23       800     
   B       final_stage  2017-10-20       400
   B       final_stage  2017-03-19       400

For a given customer ID,

case 1:If the decision is Approved,then retain all the approved records for that customer and drop others.

Case 2: If the customer doesn't have any Approved decision then ,retain the records of particular customer based on latest "req_date" and records within 5 days of this most recent "req_date" and pick the record based on the lowest salary

Customer   Decision      req_date       salary  
   A       Approved     2017-06-13       1000
   A       Approved     2017-05-13       1000
   B       final_stage  2017-10-20       400

Answer 1

The rows have to be filtered in three steps. I would use a cte for calculating aggregates and a union of two queries for approved and not approved customers:

with cte as (
    select 
        customer, 
        bool_or(decision = 'Approved') as approved, 
        max(req_date) as last_date
    from my_table
    group by 1
)
select customer, decision, req_date, salary
from my_table
join cte using(customer)
where approved and decision = 'Approved'
union all (
    select distinct on(customer) customer, decision, req_date, salary
    from my_table
    join cte using(customer)
    where not approved 
    and req_date between last_date- '5day'::interval and last_date
    order by customer, salary
    )

DbFiddle.