How to return a default value when no rows are returned from the select statement

Question

I have a select statement that returns two columns, a date column, and a count(value) column. When the count(value) column doesn\'t have any records,

Answer 1

When you use a group by, it only creates a distinct list of values that exist in your records. Since 20140228 has no records, it will not show up in the group by.

Your best bet is to generate a list of values, dates in your case, and left join or apply that table against your history table.

I can't seem to copy my T-SQL in here so here's a hastebin.

http://hastebin.com/winaqutego.vbs

Answer 2

Try this:

DECLARE @Records TABLE (
    [RecordDate] DATETIME,
    [RecordCount] INT
)

DECLARE @Date DATETIME = '02/26/2014' -- Enter whatever date you want to start with
DECLARE @EndDate DATETIME = '03/31/2014' -- Enter whatever date you want to stop
WHILE (1=1)
BEGIN
    -- Insert the date into the temp table along with the count
    INSERT INTO @Records (RecordDate, RecordCount) 
    VALUES (CONVERT(VARCHAR(25), @Date, 101), 
            (SELECT COUNT(*) FROM dbo.YourTable WHERE RecordDate = @Date))

    -- Go to the next day
    @Date = DATEADD(d, 1, @Date)

    -- If we have surpassed the end date, break out of the loop
    IF (@Date > @EndDate) BREAK;
END

SELECT * FROM @Records

If your dates have time components, you would need to modify this to check for start and end of day in the SELECT COUNT(*)... query.

Answer 3

select convert(varchar(25), DateTime, 101) as recordDate,
       CASE WHEN count(value) =0 THEN 0 ELSE COUNT(value) END  recordCount
from History
where Value < 700
group by convert(varchar(25), DateTime, 101)

Answer 4

The best practice would be for you to have a datamart where a separate dimensional table for dates is kept with all dates you might be interested at - even if they lack amounts. DMason's answer shows the query with such a dimensional table.

To keep with the best practices you would have a fact table where you'd keep these historical data already pre-grouped at the granularity level you need (daily, in this case), so you wouldn't need a GROUP BY unless you needed a coarser granularity (weekly, monthly, yearly).

And in both your operational and datamart databases the dates would be stored as dates, not...

But then, since this is real world and you might not be able to change what somebody else made... If you: a) only care about the dates that appear in [History], and b) such dates are never stored with hours/minutes; then following query might be what you'd need:

SELECT MyDates.DateTime, COUNT(*)-1 AS RecordCount
FROM (
      SELECT DISTINCT DateTime FROM History
   ) MyDates
LEFT JOIN History H
  ON  MyDates.DateTime = H.Datetime
  AND H.Value < 700
GROUP BY MyDates.DateTime

Do try to add an index over DateTime and to further constrain the query with an earliest/latest date for better performance results.

Answer 5

To generate rows for missing dates you can join your data to a date dimension table

It would look something like this:

select convert(varchar(25), ddt.DateField, 101) as recordDate,
       count(t.Value) as recordCount
from History h
right join dbo.DateDimensionTable ddt
    on ddt.DateField = convert(varchar(25), h.DateTime, 101)
where h.Value < 700
group by convert(varchar(25), h.DateTime, 101)

If your table uses the DateTime column to store dates only (meaning the time is always midnight), then you can replace this

right join dbo.DateDimensionTable ddt
    on ddt.DateField = convert(varchar(25), h.DateTime, 101)

with this

right join dbo.DateDimensionTable ddt
    on ddt.DateField = h.DateTime

Answer 6

I agree that a Dates table (AKA time dimension) is the right solution, but there is a simpler option:

SELECT
    CONVERT(VARCHAR(25), DateTime, 101) AS RecordDate,
    SUM(CASE WHEN Value < 700 THEN 1 ELSE 0 END) AS RecordCount
FROM
    History
GROUP BY
    CONVERT(VARCHAR(25), DateTime, 101)