问题
I am using an SQL Server Compact Edition server and I want to count the number of comments per month that correspond to a certain tutorial within a range of dates and include months which have a count of zero. I know I need to join a "calendar" table to my table to account for the missing months, but I need help with correct implementation of this.
I have a table of all the comments from different tutorials. This table is called Comments and the columns I need are [Tutorial] (nvarchar) and [DateAdded] (DateTime).
Tutorial | DateAdded
---------+-------------
sample | 2013-09-02
sample | 2013-09-04
sample | 2013-09-12
sample | 2013-09-12
example | 2013-09-15
sample | 2013-09-16
sample | 2013-09-21
sample | 2013-09-30
sample | 2013-10-01
sample | 2013-11-11
sample | 2013-11-11
example | 2013-11-14
sample | 2013-11-15
sample | 2013-11-19
sample | 2013-11-21
sample | 2013-11-25
sample | 2014-02-04
sample | 2014-02-06
And I have a Calendar table which has a year and month column like so:
Year | Month
-----+------
2000 | 01
2000 | 02
. | .
. | .
. | .
2099 | 12
If I were looking for the monthly count of the 'sample' comments from the past year (as of Feb. 14th, 2014), then the ideal output would be:
Tutorial | Year | Month | Count
---------+------+-------+------
sample | 2013 | 09 | 7
sample | 2013 | 10 | 1
sample | 2013 | 11 | 6
sample | 2013 | 12 | 0
sample | 2014 | 01 | 0
sample | 2014 | 02 | 2
I was able to figure out how to do the following query, but I need the months that do not have comments to return 0 as well.
SELECT
Tutorial,
datepart(year, DateAdded) AS Year,
datepart(month, DateAdded) AS Month,
COUNT(*) AS Count From Comments
WHERE
DateAdded > DATEADD(year,-1,GETDATE())
AND
Tutorial='sample'
GROUP BY
Tutorial,
datepart(year, DateAdded),
datepart(month, DateAdded)
Output using sample data from above.
Tutorial | Year | Month | Count
---------+------+-------+------
sample | 2013 | 09 | 7
sample | 2013 | 10 | 1
sample | 2013 | 11 | 6
sample | 2014 | 02 | 2
I know I need to join the tables, but I can't seem to figure out which join to use or how to implement it correctly. Please keep in mind that this is for SQL Server CE, so not all commands from SQL Server can be used.
Thanks so much in advance!
回答1:
If you have a Calendar table with Month and Year you should try something like
SELECT t2.Tutorial, t1.[Month], t1.[Year], COALESCE(t2.Number, 0) AS Result
FROM Calendar AS t1 LEFT JOIN (
SELECT
Tutorial,
CONVERT(NCHAR(6), DateAdded, 112) AS tutDate,
COUNT(*) AS Count From Comments
WHERE
DateAdded > DATEADD(year,-1,GETDATE())
AND
Tutorial='sample'
GROUP BY
Tutorial,
CONVERT(NCHAR(6), [Order Date], 112)
) AS t2
ON (t1.[Year] + t1.[Month]) = t2.tutDate
ORDER BY t1.[Year] + t1.[Month]
回答2:
What follows is a standalone script you can use to try things out and not touch any of your real database objects in production. The bottom third of the code contains the help with the joins you're looking for.
SQL Server CE will allow you to write a stored procedure, which can in turn be used as the source of a report. Stored procs are nice because they can take input parameters, something that is ideal for doing reporting.
-- create dummy Comments table for prototyping
create table #Comments (
ID int identity(1,1) not null,
Tutorial nvarchar(50) not null,
DateAdded datetime not null,
primary key clustered(DateAdded,ID,Tutorial)
);
-- populate dummy Comments table
declare @startDate datetime = '2000-01-01';
declare @endDate datetime = '2014-02-14';
declare @numTxns int = 5000;
set nocount on;
declare @numDays int = cast(@endDate as int) - cast(@startDate as int) + 1;
declare @i int = 1;
declare @j int = @i + @numTxns;
declare @rnd float;
while @i <= @j
begin
set @rnd = RAND();
insert into #Comments (Tutorial,DateAdded)
select
-- random tutorial titles
coalesce (
case when @rnd < .25 then 'foo' else null end,
case when @rnd between .5 and .75 then 'baz' else null end,
case when @rnd > .75 then 'qux' else null end,
'bar'
) as Tutorial,
-- random dates between @startDate and @endDate
cast(cast(rand() * @numDays + @startDate as int) as datetime) as DateAdded
set @i = @i + 1
end;
-- try deleting some months to see what happens
delete from #Comments
where DateAdded between '2013-11-01' and '2013-11-30'
or DateAdded between '2014-01-01' and '2014-01-31';
set nocount off;
go
-- ### following could easily be rewritten as a stored procedure
-- stored procedure parameters
declare @startDate datetime = '2000-01-01';
declare @endDate datetime = '2014-03-31';
-- pick only one option below
--declare @Tutorial nvarchar(50) = 'foo'; -- this only gets data for Tutorials called 'foo'
declare @Tutorial nvarchar(50) = 'all'; -- this gets data for all tutorials
-- begin stored procedure code
set nocount on;
-- this temp table is an alternative to
-- creating ***and maintaining*** a table full of dates,
-- months, etc., and cluttering up your database
-- in production, it will automatically delete itself
-- once the user has completed running the report.
create table #dates (
DateAdded datetime not null,
YearAdded int null,
MonthAdded int null,
primary key clustered (DateAdded)
);
-- now we put dates into #dates table
-- based on the parameters supplied by
-- the user running the report
declare @date datetime = @startDate;
while @date <= @endDate
begin
insert into #dates
select @date, YEAR(@date), MONTH(@date);
set @date = @date + 1;
end;
-- ## Why put every day of the month in this table?
-- ## I asked for a monthy report, not daily!
-- Yes, but looping through dates is easier, simply add 1 for the next date.
-- You can always build a monthly summary table later if you'd like.
-- This *is* kind of a brute-force solution, but easy to write.
-- More answers to this question in the code below, where they'll make more sense.
set nocount off;
-- now we return the data to the user
-- any month with no Tutorials will still show up in the report
-- but the counts will show as zero
select YearAdded, MonthAdded, SUM(Count_From_Comments) as Count_From_Comments,
SUM(foo) as Count_Foo, SUM(bar) as Count_Bar,
SUM(baz) as Count_Baz, SUM(qux) as Count_Qux
from (
-- ## you can reuse the following code for a detail report by day
-- ## another answer to 'Why not by month?' from above
-- start daily report code
select t1.DateAdded, t1.YearAdded, t1.MonthAdded, t2.Tutorial,
coalesce(Count_From_Comments,0) as Count_From_Comments,
case when t2.Tutorial = 'foo' then 1 else 0 end as foo,
case when t2.Tutorial = 'bar' then 1 else 0 end as bar,
case when t2.Tutorial = 'baz' then 1 else 0 end as baz,
case when t2.Tutorial = 'qux' then 1 else 0 end as qux
from #dates as t1 -- no where clause needed because #dates only contains the ones we want
left join ( -- left join here so that we get all dates, not just ones in #Comments
select *, 1 AS Count_From_Comments
from #Comments
where @Tutorial in (Tutorial,'all')
) as t2
on t1.DateAdded = t2.DateAdded -- ## join on one field instead of two, another answer to 'Why not by month?' from above
-- end daily report code
) as qDetail
group by YearAdded, MonthAdded
order by YearAdded, MonthAdded
-- end stored procedure code
go
-- ## Not required in production code,
-- ## but handy when testing this script.
drop table #dates;
-- #### Since this will be a real table in production
-- #### we definitely only want this for testing!
drop table #Comments;
go
Happy coding.
来源：https://stackoverflow.com/questions/21789777/counting-records-by-year-and-month-including-zero-counts