Showing dimension data without a fact record

Question

To explain my requirement, lets say I have 3 tables
City(Dimension shows all city information)
Day(Date Dimension with all dates for some years(past and future))
Sales(Fact table showing all sales measures, multiple records for a city and day)
A sample query to retreive data i need
select city.name,day.name,SUM(sales.amount) amount
from City,Day,Sales
where city.code = sales.city_code
and day.date_value = sales.transaction_date
group by city.name,day.name
Now if there is no sales transaction for a particular day and city, still i want to show the day and city with amount as 0
The solution i can think of is doing a Cartesian on city and day tables use it as an inline view and then do an outer join with sales table.
Is there any better way to achieve this, using some simple query logic or any advanced feature?
Any inputs on the above would be of great help.
Thanks
Kranthi
Edited by: user12033529 on Sep 12, 2010 6:57 AM

Answer 1

Hi, Kranthi,
Welcome to the forum!
Whenever you have a proiblem, please post your version of Oracle (e.g. 10.2.0.3.0); a little sample data in a form people can use to re-create the problem (e.g.
CREATE TABLE     city
(     city_id          NUMBER (6)     PRIMARY KEY
,     city_name     VARCHAR2 (20)     NOT NULL
INSERT INTO CITY (city_id, city_name) VALUES (1, 'Asheville');
INSERT INTO CITY (city_id, city_name) VALUES (2, 'Bryson City');
INSERT INTO CITY (city_id, city_name) VALUES (3, 'Cherokee');
CREATE TABLE     day
(     dt     DATE     PRIMARY KEY
INSERT INTO day (dt)
SELECT      DATE '2010-09-10' + LEVEL
FROM     dual
CONNECT BY     LEVEL     <= 3;
CREATE TABLE     sales
(     sales_id     NUMBER (6)     PRIMARY KEY
,     city_id          NUMBER (6)
,     sales_dt     DATE
,     amount          NUMBER (9)
INSERT INTO sales (sales_id, city_id, sales_dt, amount) VALUES (11, 1, DATE '2010-09-01',  10);
INSERT INTO sales (sales_id, city_id, sales_dt, amount) VALUES (12, 1, DATE '2010-09-11',  20);
INSERT INTO sales (sales_id, city_id, sales_dt, amount) VALUES (13, 1, DATE '2010-09-12',  40);
INSERT INTO sales (sales_id, city_id, sales_dt, amount) VALUES (14, 1, DATE '2010-09-12',  80);
INSERT INTO sales (sales_id, city_id, sales_dt, amount) VALUES (15, 2, DATE '2010-09-11', 160);
INSERT INTO sales (sales_id, city_id, sales_dt, amount) VALUES (16, 2, DATE '2010-09-11', 320);
COMMIT;) and the results you want from that data (e.g.
CITY_NAME            DT          TOTAL_AMOUNT
Asheville            11-Sep-2010           20
Asheville            12-Sep-2010          120
Asheville            13-Sep-2010            0
Bryson City          11-Sep-2010          480
Bryson City          12-Sep-2010            0
Bryson City          13-Sep-2010            0
Cherokee             11-Sep-2010            0
Cherokee             12-Sep-2010            0
Cherokee             13-Sep-2010            0).
user12033529 wrote:
... The solution i can think of is doing a Cartesian on city and day tables use it as an inline view and then do an outer join with sales table.
Is there any better way to achieve this, using some simple query logic or any advanced feature?For this particular problem, I think that is the best way.
There is another way: the SELECT ... PARTITION BY feature (also know as Query Partitioning ).
For example:
WITH     all_cities     AS
     SELECT     c.city_name
     ,     s.sales_dt
     ,     s.amount
     FROM               sales     s     PARTITION BY (city_id)
     RIGHT OUTER JOIN     city     c     ON     s.city_id     = c.city_id
SELECT     a.city_name
,     d.dt
,     NVL ( SUM (a.amount)
         , 0
         )          AS total_amount
FROM               all_cities     a     PARTITION BY (city_name, sales_dt)
RIGHT OUTER JOIN     day          d     ON     a.sales_dt     = d.dt
GROUP BY     a.city_name
,          d.dt
ORDER BY     a.city_name
,          d.dt
;It seems like you should be able to do this without a sub-query, but I don't know how.
Query Partitioning is really useful when you want to include only some dimensions that are actually present in the fact table. For example, if you wanted to show only the cities that actually had sales (that is, in this example, if you didn't want to include Cherokee), but you do want all dates in the day table for each city.
Query Partitioning was new in Oracle 10. If you have an older version, I think a cross-join, like you originally suggested, is the only reasonable way.
Edited by: Frank Kulash on Sep 12, 2010 1:14 PM