Requesting help with pivot in Oracle 10g

Question

Hey all,
I have this table, let's call it table GRID that has the following columns:
OBJECT_KEY             DATE_KEY               SEGMENT_ID             COLUMN_NAME COMMITTEDVIRTUALMEMORYSIZE_AVG
619                    3371                   1                      M1100       593589862.4                   
619                    3371                   1                      M1105       593611434.67                  
620                    3371                   1                      M1100       592354508.8                   
620                    3371                   1                      M1105       592376263.11                  
621                    3371                   1                      M1100       731433369.6                   
621                    3371                   1                      M1105       731455943.11What would it take to pivot the column names ('M1100', 'M1105') with the metric value volumn: 'COMMITTEDVIRTUALMEMORYSIZE_AVG' so much that this query would result in an output like this:
OBJECT_KEY             DATE_KEY               SEGMENT_ID             M1100          M1105
619                    3371                   1                      593589862.4     593611434.67                          
620                    3371                   1                      592354508.8     592376263.11                          
621                    3371                   1                      731433369.6     731455943.11Any help would be much appreciated!
Again, Oracle 10g, not 11g so I don't have the PIVOT function to work with.
Regards,
TimS

Answer 1

Hi,
Search for Pivot:, or see the example below, for how to do a basic pivot. (This was called "pivot" long before Oracle 11.)
The example below uses COUNT as the aggregate function; you'll probably wnat SUM instead.
If you don't know how many pivoted columns there will be (two, in the example you posted), or what their unique values are ('M1100' and 'M1105') ahead of time, then you will need dynamic SQL. As dynamic SQL goes, this is pretty easy; you can do it in either SQL*Plus or PL/SQL.
--     How to Pivot a Result Set (Display Rows as Columns)
--     For Oracle 10, and earlier
--     Actually, this works in any version of Oracle, but the
--     "SELECT ... PIVOT" feature introduced in Oracle 11
--     is better.  (See Query 2, below.)
--     This example uses the scott.emp table.
--     Given a query that produces three rows for every department,
--     how can we show the same data in a query that has one row
--     per department, and three separate columns?
--     For example, the query below counts the number of employess
--     in each departent that have one of three given jobs:
PROMPT     ==========  0. Simple COUNT ... GROUP BY  ==========
SELECT     deptno
,     job
,     COUNT (*)     AS cnt
FROM     scott.emp
WHERE     job     IN ('ANALYST', 'CLERK', 'MANAGER')
GROUP BY     deptno
,          job;
Output:
    DEPTNO JOB              CNT
        20 CLERK              2
        20 MANAGER            1
        30 CLERK              1
        30 MANAGER            1
        10 CLERK              1
        10 MANAGER            1
        20 ANALYST            2
PROMPT     ==========  1. Pivot  ==========
SELECT     deptno
,     COUNT (CASE WHEN job = 'ANALYST' THEN 1 END)     AS analyst_cnt
,     COUNT (CASE WHEN job = 'CLERK'   THEN 1 END)     AS clerk_cnt
,     COUNT (CASE WHEN job = 'MANAGER' THEN 1 END)     AS manager_cnt
FROM     scott.emp
WHERE     job     IN ('ANALYST', 'CLERK', 'MANAGER')
GROUP BY     deptno;
--     Output:
    DEPTNO ANALYST_CNT  CLERK_CNT MANAGER_CNT
        30           0          1           1
        20           2          2           1
        10           0          1           1
--     Explanation
(1) Decide what you want the output to look like.
     (E.g. "I want a row for each department,
     and columns for deptno, analyst_cnt, clerk_cnt and manager_cnt)
(2) Get a result set where every row identifies which row
     and which column of the output will be affected.
     In the example above, deptno identifies the row, and
     job identifies the column.
     Both deptno and job happened to be in the original table.
     That is not always the case; sometimes you have to
     compute new columns based on the original data.
(3) Use aggregate functions and CASE (or DECODE) to produce
     the pivoted columns. 
     The CASE statement will pick
     only the rows of raw data that belong in the column.
     If each cell in the output corresponds to (at most)
     one row of input, then you can use MIN or MAX as the
     aggregate function.
     If many rows of input can be reflected in a single cell
     of output, then use SUM, COUNT, AVG, STRAGG, or some other
     aggregate function.
     GROUP BY the column that identifies rows.
PROMPT     ==========  2. Oracle 11 PIVOT  ==========
WITH     e     AS
(     -- Begin sub-query e to SELECT columns for PIVOT
     SELECT     deptno
     ,     job
     FROM     scott.emp
)     -- End sub-query e to SELECT columns for PIVOT
SELECT     *
FROM     e
PIVOT     (     COUNT (*)
          FOR     job     IN     ( 'ANALYST'     AS analyst
                         , 'CLERK'     AS clerk
                         , 'MANAGER'     AS manager
NOTES ON ORACLE 11 PIVOT:
(1) You must use a sub-query to select the raw columns.
An in-line view (not shown) is an example of a sub-query.
(2) GROUP BY is implied for all columns not in the PIVOT clause.
(3) Column aliases are optional. 
If "AS analyst" is omitted above, the column will be called 'ANALYST' (single-quotes included).
{code}