How do I return my records grouped by NULL and NOT NULL?
Question
I have a table that has a processed_timestamp
column -- if a record has been processed then that field contains the datetime it was processed, otherwise it is null.
I want to write a query that returns two rows:
NULL xx -- count of records with null timestamps
NOT NULL yy -- count of records with non-null timestamps
Is that possible?
Update: The table is quite large, so efficiency is important. I could just run two queries to calculate each total separately, but I want to avoid hitting the table twice if I can avoid it.
Solution
Oracle:
group by nvl2(field, 'NOT NULL', 'NULL')
OTHER TIPS
In MySQL you could do something like
SELECT
IF(ISNULL(processed_timestamp), 'NULL', 'NOT NULL') as myfield,
COUNT(*)
FROM mytable
GROUP BY myfield
In T-SQL (MS SQL Server), this works:
SELECT
CASE WHEN Field IS NULL THEN 'NULL' ELSE 'NOT NULL' END FieldContent,
COUNT(*) FieldCount
FROM
TheTable
GROUP BY
CASE WHEN Field IS NULL THEN 'NULL' ELSE 'NOT NULL' END
Try the following, it's vendor-neutral:
select
'null ' as type,
count(*) as quant
from tbl
where tmstmp is null
union all
select
'not null' as type,
count(*) as quant
from tbl
where tmstmp is not null
After having our local DB2 guru look at this, he concurs: none of the solutions presented to date (including this one) can avoid a full table scan (of the table if timestamp is not indexed, or of the indexotherwise). They all scan every record in the table exactly once.
All the CASE/IF/NVL2() solutions do a null-to-string conversion for each row, introducing unnecessary load on the DBMS. This solution does not have that problem.
If it's oracle then you can do:
select decode(field,NULL,'NULL','NOT NULL'), count(*)
from table
group by decode(field,NULL,'NULL','NOT NULL');
I'm sure that other DBs allow for similar trick.
Stewart,
Maybe consider this solution. It is (also!) vendor non-specific.
SELECT count([processed_timestamp]) AS notnullrows,
count(*) - count([processed_timestamp]) AS nullrows
FROM table
As for efficiency, this avoids 2x index seeks/table scans/whatever by including the results on one row. If you absolutely require 2 rows in the result, two passes over the set may be unavoidable because of unioning aggregates.
Hope this helps
Another MySQL method is to use the CASE
operator, which can be generalised to more alternatives than IF()
:
SELECT CASE WHEN processed_timestamp IS NULL THEN 'NULL'
ELSE 'NOT NULL' END AS a,
COUNT(*) AS n
FROM logs
GROUP BY a
If your database has an efficient COUNT(*) function for a table, you could COUNT whichever is the smaller number, and subtract.
SQL Server (starting with 2012):
SELECT IIF(ISDATE(processed_timestamp) = 0, 'NULL', 'NON NULL'), COUNT(*)
FROM MyTable
GROUP BY ISDATE(processed_timestamp);
I personally like Pax's solution, but if you absolutely require only one row returned (as I had recently), In MS SQL Server 2005/2008 you can "stack" the two queries using a CTE
with NullRows (countOf)
AS
(
SELECT count(*)
FORM table
WHERE [processed_timestamp] IS NOT NULL
)
SELECT count(*) AS nulls, countOf
FROM table, NullRows
WHERE [processed_timestamp] IS NULL
GROUP BY countOf
Hope this helps
[T-SQL]:
select [case], count(*) tally
from (
select
case when [processed_timestamp] is null then 'null'
else 'not null'
end [case]
from myTable
) a
And you can add into the case statement whatever other values you'd like to form a partition, e.g. today, yesterday, between noon and 2pm, after 6pm on a Thursday.
Select Sum(Case When processed_timestamp IS NULL
Then 1
Else 0
End) not_processed_count,
Sum(Case When processed_timestamp Is Not NULL
Then 1
Else 0
End) processed_count,
Count(1) total
From table
Edit: didn't read carefully, this one returns a single row.
In Oracle
SELECT COUNT(*), COUNT(TIME_STAMP_COLUMN)
FROM TABLE;
count(*) returns the count of all rows
count(column_name) returns the number of rows which are not NULL, so
SELECT COUNT(*) - COUNT(TIME_STAMP_COLUMN) NUL_COUNT,
COUNT(TIME_STAMP_COLUMN) NON_NUL_COUNT
FROM TABLE
ought to do the job.
If the column is indexed, you might end up with some sort of range scan and avoid actually reading the table.
Another way in T-sql (sql-server)
select count(case when t.timestamps is null
then 1
else null end) NULLROWS,
count(case when t.timestamps is not null
then 1
else null end) NOTNULLROWS
from myTable t