Question

I am using a proprietary mpp database that has been forked off psql 8.3. I am trying to apply a simple count to a wide table (around 450 columns) and so I was wondering if the best way to do this in terms of a simple sql function. I am just counting the number of distinct values in a given column as well as the count of the number of null values in the column. The query i want to generalize for every column is for example

If i want to run the query against the column names i write

select
count(distinct names) d_names,
sum(case when names is not null then 1 else 0 end) n_s_ip
from table; 

How do i generalize the query above to iterate through every column in the table if the number of columns is 450 without writing out each column name by hand?

Answer 1

First, since COUNT() only counts non-null values, your query can be simplified:

SELECT count(DISTINCT names) AS unique_names
      ,count(names) AS names_not_null
FROM   table;

But that’s the number of non-null values and contradicts your description:

count of the number of null values in the column

For that you would use:

count(*) - count(names) AS names_null

Since count(*) count all rows and count(names) only rows with non-null names.
_{Removed inferior alternative after hint by @Andriy.}

To automate that for all columns build an SQL statement off of the catalog table pg_attribute dynamically. You can use EXECUTE in a PL/pgSQL function to execute it immediately. Find full code examples with links to the manual and explanation under these closely related questions:

• How to perform the same aggregation on every column, without listing the columns?
• postgresql – count (no null values) of each column in a table