Question

My table has values like (RowCount is generated by the query below):

ID       Date_trans   Time_trans  Price  RowCount
-------  -----------  ----------  -----  --------
1699093  22-Feb-2011  09:30:00    58.07  1
1699094  22-Feb-2011  09:30:00    58.08  1
1699095  22-Feb-2011  09:30:00    58.08  2
1699096  22-Feb-2011  09:30:00    58.08  3
1699097  22-Feb-2011  09:30:00    58.13  1
1699098  22-Feb-2011  09:30:00    58.13  2
1699099  22-Feb-2011  09:30:00    58.12  1
1699100  22-Feb-2011  09:30:08    58.13  3
1699101  22-Feb-2011  09:30:09    57.96  1
1699102  22-Feb-2011  09:30:09    57.95  1
1699103  22-Feb-2011  09:30:09    57.93  1
1699104  22-Feb-2011  09:30:09    57.96  2
1699105  22-Feb-2011  09:30:09    57.93  2
1699106  22-Feb-2011  09:30:09    57.93  3
1699107  22-Feb-2011  09:30:37    58     1
1699108  22-Feb-2011  09:30:37    58.08  4
1699109  22-Feb-2011  09:30:38    58.08  5
1699110  22-Feb-2011  09:30:41    58.02  1
1699111  22-Feb-2011  09:30:41    58.02  2
1699112  22-Feb-2011  09:30:41    58.01  1
1699113  22-Feb-2011  09:30:41    58.01  2
1699114  22-Feb-2011  09:30:41    58.01  3
1699115  22-Feb-2011  09:30:42    58.02  3
1699116  22-Feb-2011  09:30:42    58.02  4
1699117  22-Feb-2011  09:30:45    58.04  1
1699118  22-Feb-2011  09:30:54    58     2
1699119  22-Feb-2011  09:30:57    58.05  1

The ID column is an IDENTITY column.
And I’m using this query to get the consecutive row count as:

  SELECT   ID, Date_trans, Time_trans, Price
          ,ROW_NUMBER() OVER(PARTITION BY Price  ORDER BY ID) RowCount
  FROM     MyTable
  ORDER    BY ID;

The RowCount I get is right for most of the values but wrong for some values. For instance:

• ID 1699100 Price 58.13 – count should be 1 (showing 3).
• ID 1699104 Price 57.96 – count should be 1 (showing 2).
• ID 1699105, 1699106 Price 57.93 – count should be 1, 2 (showing 2, 3).

I have tried the same query in PostgreSQL and found the same results.
I have uploaded a csv data sample here.

I’m stuck with such unexpected results of partition. Can anybody help me?

Answer 1

The PARTITION BY clause of the ROW_NUMBER() function instructs it to partition the entire row set by Price values and assign row numbers in the ascending order of IDs.

It seems like you want to distinguish between any two groups of rows with identical Price values that are separated by at least one row with a different Price.

There may be various ways to achieve that. In SQL Server (and I think the same would work in PostgreSQL too), I would first use two ROW_NUMBER() calls to get an additional partitioning criterion, then rank rows once again using that criterion, like this:

WITH partitioned AS (
  SELECT
    ID,
    Date_trans,
    Time_trans,
    Price,
    ROW_NUMBER() OVER (                   ORDER BY ID) -
    ROW_NUMBER() OVER (PARTITION BY Price ORDER BY ID) AS PriceGroup
  FROM MyTable
)
SELECT
  ID,
  Date_trans,
  Time_trans,
  Price,
  ROW_NUMBER() OVER (PARTITION BY Price, PriceGroup ORDER BY ID) AS RowCount
FROM partitioned
ORDER BY ID
;

Here’s a SQL Fiddle demo.