MSSQL2012 (Over/Partition) 最近 12 个月，包括空值？
MSSQL2012 (Over/Partition) Last 12 months, nulls included?
所以，我遇到了一个问题：在有产品的情况下，获取过去 12 个月（包括正在处理的月份）的实际 sum/avg，通过产品代码进行分区一个月内可能不存在。
尝试从 TL;DR 开始，让我们开始实际操作：
- 我的table有4个字段：年、月、产品代码和价值；
- 我需要去年（所选月份+过去11个月）的总和；
- 我需要去年（相同）的平均值。
我试过：
SELECT * FROM (
SELECT year, month, product,
AVG(value) OVER (
PARTITION BY product
ORDER BY year, month
ROWS 11 PRECEDING
) as average,
SUM(value) OVER (
PARTITION BY product
ORDER BY year, month
ROWS 11 PRECEDING
) as sum
FROM suchDB.muchUSER.awesomeTABLE
) q
where year = <insert year> and month = <month>
问题：
- 如果任何查询的月份没有产品“123”，它会忽略该产品是否在 table 上有历史记录，它不会给我带来过去几个月的平均值或总和。
- 如果过去的任何一个月（应该是查询月份的历史记录的一部分）没有产品“123”，它将忽略该月份并转到 'last year' 范围之后的一个月（没有 7 月的查询的 10 月将 return 它是前一年的版本，而不是在 11 月停止）。
我们曾经使用 'GROUP BY product' 和 'WHERE ((year-1)*100)+month > queriedYear-1Month AND (year*100)+month <= queriedYearMonth' 来查询这个东西，直到有人将我们指向 OVER/PARTITION 方向并且我们改变了一切......但即便如此我们还是遇到了一些问题， like avg() 函数会忽略 NULL 个月...
帮忙？
忘记了一件非常重要的事情
awesomeTABLE 中的数据是版本化的——每个 year/month 可以有多个版本。只能使用最新的。我通常通过加入 select distinct year, month, max(version) from awesomeTABLE group by year, month 来做到这一点，但它似乎扼杀了一些可能的解决方案...
如果您有 table 个产品和 table 个日期，我会将它们与上面的查询连接起来，以便表示所有产品和所有日期，然后求和用你的方法得到的结果。
您的第一个项目符号应该得到解决，因为所有日期和所有产品都将被表示
我相信你的第二个项目符号是由于 "rows 11 preceding" 真正得到过去的 11 个值而不管日期；这可以通过将 0 替换为 null 来解决当前缺失的 month/product 组合，但在左连接方法的情况下会提供。
对缺失的月份进行了大量可能的修复。这是一种不会过多改变您的原始查询的方法：
WITH ym as (
select
year(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as y,
month(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as m
from (values
(0), (-1), (-2), (-3), (-4), (-5),
(-6), (-7), (-8), (-9), (-10), (-11)
) ofs(n)
)
SELECT
y, m, p.product,
AVG(coalesce(value, 0)) OVER (
PARTITION BY p.product
ORDER BY y, m
ROWS 11 PRECEDING
) as average,
SUM(value) OVER (
PARTITION BY p.product
ORDER BY y, m
ROWS 11 PRECEDING
) as sum
FROM
ym cross join
(select distinct product from suchDB.muchUSER.awesomeTABLE) p
left outer join suchDB.muchUSER.awesomeTABLE t
on t."year" = ym.y and t."month" = ym.m and t.product = p.product
where y = <insert year> and m = <month>
你可以做一个 order by y, m desc 和 select top 1 而不是最后的 where。我通常会尽量避免使用 top 但我不确定您是如何删除参数的，并且避免使用某些编程库两次引用它们可能不会那么麻烦。即使您是手动执行此操作，您仍然必须记住在长查询中查看两个位置。
由于您似乎只需要一个月的聚合，我认为您可以使用这个不使用 window 函数的更简单版本：
WITH ym as (
select
dateadd(month, n, cast('<year>' + '<month>' + '01' as date) as dt,
year(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as y,
month(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as m
from (values
(0), (-1), (-2), (-3), (-4), (-5),
(-6), (-7), (-8), (-9), (-10), (-11)
) ofs(n)
)
SELECT
year(max(dt)) as "year", month(max(dt)) as "month", p.product,
AVG(coalesce(value, 0)) as average,
SUM(value) as "sum"
FROM
ym cross join
(select distinct product from awesome) p
left outer join awesome t
on t."year" = ym.y and t."month" = ym.m and t.product = p.product
GROUP BY p.product
我不知道你是否只想限制在上一年实际销售的产品，所以我没有在那里处理。
如果您开始考虑如何使它更通用和可重用，您最终可能会更像这个。我继续将产品限制添加到 activity 的前一年：
WITH dt as (
select dateadd(month, n, cast('<year>' + '<month>' + '01' as date) as dt
),
ym as (
select dt, year(dateadd(month, n, dt)) as y, month(dateadd(month, n, dt)) as m
from (values
(0), (-1), (-2), (-3), (-4), (-5),
(-6), (-7), (-8), (-9), (-10), (-11)
) ofs(n), dt
)
SELECT
year(max(dt)) as "year", month(max(dt)) as "month", p.product,
AVG(coalesce(value, 0)) as average, SUM(value) as "sum"
FROM
ym cross join
(
select distinct product from awesome
where cast("year" + "month" + '01' as date) between
(select min(dt) from ym) and (select max(dt) from ym)
) p
left outer join (
select distinct /* get the latest "version" only */
first_value("year")
over (partition by "year", "month", product order by version desc),
first_value("month")
over (partition by "year", "month", product order by version desc),
product,
first_value(value)
over (partition by "year", "month", product order by version desc)
from awesome
) t
on t."year" = ym.y and t."month" = ym.m and t.product = p.product
GROUP BY p.product
最终查询还尝试仅针对最新版本处理您的过滤器。不过，您需要更高版本的 SQL 服务器才能实现 first_value() 功能。
最大的问题是您需要获得两个列表才能正确汇总数据集中的值 - 一个日期列表和一个产品列表。如果没有这两个列表，最后一个月缺少产品意味着不会报告该产品，或者（正如您已经发现的那样）可能会汇总错误的 12 个月（缺少 7 月意味着前面 11 行包括开始月份）。
下面是对生成这些列表的过程的全面扩展探索。它仅使用源数据 table（假设每个月售出 东西 。它可以变得更简洁（即计算日期，如上面 shawnt 的示例） , 而是为了展示所有的步骤和假设而写的。我把它封装成一个存储过程，因为它显式地展示了传入的值。
CREATE PROCEDURE DoTheThing
@startDate DATE -- Should be Year-Month-01 or YearMonth01
AS
BEGIN
DECLARE @yr INT, @mth INT,
@yr2 INT, @mth2 INT,
@endDate DATE -- StartDt - 11 months
-- if the date will be passed in with a day other than 01, add code here to set the day on the passed date to 01
-- if only the high year and month are passed in, then create a @startDt value and continue.
SET @endDate = DATEADD(MONTH, -11, @startDate)
SELECT @yr = DATEPART(YEAR, @startDate),
@mth = DATEPART(MONTH, @startdate),
@yr2 = DATEPART(YEAR, @startDate),
@mth2 = DATEPART(MONTH, @startdate)
WITH mthYr AS (
SELECT DISTINCT
YEAR,
MONTH
FROM suchDB.muchUSER.awesomeTABLE -- Get the data from the source table
WHERE (
YEAR = @yr -- if in the passed-in year, then take all months less than or equal to the start month
AND MONTH <= @mth
)
OR (
YEAR = @yr2 -- if the period is Jan -- Dec in one year, this reiterates the above
AND MONTH >= @mth2 -- if not, select the months in the second year where the month is greater than or equal to the calculated month
)
),
prods AS (
SELECT DISTINCT product -- Return a list of products sold during the year.
FROM suchDB.muchUSER.awesomeTABLE smt
INNER JOIN mthYr
ON mthYr.YEAR = smt.YEAR
AND mthyr.MONTH = smt.MONTH
)
SELECT MAX(mthYr.YEAR) AS year, -- current report only shows passed in year/month value
MAX(mthYr.month) AS month,
prods.product,
AVG(ISNULL(VALUE, 0.00)) average, -- isNull value adds a zero into the list to be averaged
SUM(ISNULL(VALUE, 0.00)) SUM -- not really necessary, but no warnings about NULL values will be generated
FROM mthYr CROSS JOIN prods -- cross join of the two lists means all of the products sold will have a value for each month
LEFT JOIN suchDB.muchUSER.awesomeTABLE smt -- left join so missing productMonths will still be added in
ON smt.YEAR = mthYr.year
AND smt.MONTH = mthYr.month
AND prods.product = smt.product
ORDER BY smt.product
END
处理日期时首先要做的是有一个日期时间字段，而不是年月日的分隔字段
SELECT year, month, product, value
, DATEFROMPARTS(year, month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE
由于可以对行进行版本控制，我们需要获取每年、每月和产品的最新版本，这可以通过多种方式完成，例如使用窗口功能或自动连接，例如后者是
SELECT b.[year], b.[month], b.[product], [value]
, DATEFROMPARTS(b.[year], b.month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE b
INNER JOIN (SELECT [year], [month], [product], max([version]) lv
FROM suchDB.muchUSER.awesomeTABLE
GROUP BY [year], [month], [product]
) m ON b.[year] = m.year AND b.month = m.month
AND b.product = m.product AND b.[version] = m.lv
还有产品清单会派上用场
SELECT DISTINCT product
FROM suchDB.muchUSER.awesomeTABLE
获取去年有价值的数据可以通过多种方式完成，我喜欢 CROSS APPLY
Declare @_year int
Declare @_month int
Set @_year = 2015
Set @_month = 11
;With baseDate AS (
SELECT b.[year], b.[month], b.[product], [value]
, DATEFROMPARTS(b.[year], b.month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE b
INNER JOIN (SELECT [year], [month], [product], max([version]) lv
FROM suchDB.muchUSER.awesomeTABLE
GROUP BY [year], [month], [product]
) m ON b.[year] = m.year AND b.month = m.month
AND b.product = m.product AND b.[version] = m.lv
), Products AS (
SELECT DISTINCT [product]
FROM suchDB.muchUSER.awesomeTABLE
)
SELECT @_year [Year], @_month [Month], p.[product]
, ly.Average
, ly.[Sum]
FROM Products p
CROSS APPLY (SELECT Sum(lastYear.Value) / 12.0 Average
, Sum(lastYear.Value) [Sum]
FROM baseDate lastYear
WHERE lastYear.fullDate > DATEFROMPARTS(@_year - 1, @_month, 1)
AND lastYear.fullDate <= DATEFROMPARTS(@_year, @_month, 1)
AND lastYear.product = p.product
) ly
WHERE ly.[Sum] IS NOT NULL
WHERE条件去掉参数前一年没有行的产品
要删除变量并获取每个月的值，需要一个日历 table。如果 table 中所有产品的所有月份都在那里，我们可以使用 DISTINCT 获取日期，使用第一个 CTE 而不是 table 让我们也得到完整日期
SELECT DISTINCT [year], [month], fullDate
FROM baseDate
否则有不同的方法来创建日历 table。我们可以将日历 table 添加到主查询 CTEs 并在 CROSS APPLY 中使用它而不是变量
;With baseDate AS (
SELECT b.[year] _y, b.[month] _m, b.[product], [value]
, DATEFROMPARTS(b.[year], b.month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE b
INNER JOIN (SELECT [year], [month], [product], max([version]) lv
FROM suchDB.muchUSER.awesomeTABLE
GROUP BY [year], [month], [product]
) m ON b.[year] = m.year AND b.month = m.month
AND b.product = m.product AND b.[version] = m.lv
), Products AS (
SELECT DISTINCT [product]
FROM suchDB.muchUSER.awesomeTABLE
), Months As (
SELECT DISTINCT _y, _m, fullDate
FROM baseDate
)
SELECT _y [Year], _m [Month], p.[product]
, ly.Average
, ly.[Sum]
, ly.[Count]
FROM Products p
CROSS APPLY (SELECT m._y, m._m
, Sum(bd.Value) / 12.0 Average
, Sum(bd.Value) [Sum]
, Count(Value) [Count]
FROM Months m
LEFT JOIN baseDate bd
ON bd.fullDate > DATEADD(YY, -1, m.fullDate)
AND bd.fullDate <= m.fullDate
WHERE bd.product = p.product
GROUP BY m._y, m._m
) ly
WHERE ly.[Sum] IS NOT NULL
所以，我遇到了一个问题：在有产品的情况下，获取过去 12 个月（包括正在处理的月份）的实际 sum/avg，通过产品代码进行分区一个月内可能不存在。
尝试从 TL;DR 开始，让我们开始实际操作：
- 我的table有4个字段：年、月、产品代码和价值；
- 我需要去年（所选月份+过去11个月）的总和；
- 我需要去年（相同）的平均值。
我试过：
SELECT * FROM (
SELECT year, month, product,
AVG(value) OVER (
PARTITION BY product
ORDER BY year, month
ROWS 11 PRECEDING
) as average,
SUM(value) OVER (
PARTITION BY product
ORDER BY year, month
ROWS 11 PRECEDING
) as sum
FROM suchDB.muchUSER.awesomeTABLE
) q
where year = <insert year> and month = <month>
问题：
- 如果任何查询的月份没有产品“123”，它会忽略该产品是否在 table 上有历史记录，它不会给我带来过去几个月的平均值或总和。
- 如果过去的任何一个月（应该是查询月份的历史记录的一部分）没有产品“123”，它将忽略该月份并转到 'last year' 范围之后的一个月（没有 7 月的查询的 10 月将 return 它是前一年的版本，而不是在 11 月停止）。
我们曾经使用 'GROUP BY product' 和 'WHERE ((year-1)*100)+month > queriedYear-1Month AND (year*100)+month <= queriedYearMonth' 来查询这个东西，直到有人将我们指向 OVER/PARTITION 方向并且我们改变了一切......但即便如此我们还是遇到了一些问题， like avg() 函数会忽略 NULL 个月...
帮忙？
忘记了一件非常重要的事情
awesomeTABLE 中的数据是版本化的——每个 year/month 可以有多个版本。只能使用最新的。我通常通过加入 select distinct year, month, max(version) from awesomeTABLE group by year, month 来做到这一点，但它似乎扼杀了一些可能的解决方案...
如果您有 table 个产品和 table 个日期，我会将它们与上面的查询连接起来，以便表示所有产品和所有日期，然后求和用你的方法得到的结果。
您的第一个项目符号应该得到解决，因为所有日期和所有产品都将被表示
我相信你的第二个项目符号是由于 "rows 11 preceding" 真正得到过去的 11 个值而不管日期；这可以通过将 0 替换为 null 来解决当前缺失的 month/product 组合，但在左连接方法的情况下会提供。
对缺失的月份进行了大量可能的修复。这是一种不会过多改变您的原始查询的方法：
WITH ym as (
select
year(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as y,
month(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as m
from (values
(0), (-1), (-2), (-3), (-4), (-5),
(-6), (-7), (-8), (-9), (-10), (-11)
) ofs(n)
)
SELECT
y, m, p.product,
AVG(coalesce(value, 0)) OVER (
PARTITION BY p.product
ORDER BY y, m
ROWS 11 PRECEDING
) as average,
SUM(value) OVER (
PARTITION BY p.product
ORDER BY y, m
ROWS 11 PRECEDING
) as sum
FROM
ym cross join
(select distinct product from suchDB.muchUSER.awesomeTABLE) p
left outer join suchDB.muchUSER.awesomeTABLE t
on t."year" = ym.y and t."month" = ym.m and t.product = p.product
where y = <insert year> and m = <month>
你可以做一个 order by y, m desc 和 select top 1 而不是最后的 where。我通常会尽量避免使用 top 但我不确定您是如何删除参数的，并且避免使用某些编程库两次引用它们可能不会那么麻烦。即使您是手动执行此操作，您仍然必须记住在长查询中查看两个位置。
由于您似乎只需要一个月的聚合，我认为您可以使用这个不使用 window 函数的更简单版本：
WITH ym as (
select
dateadd(month, n, cast('<year>' + '<month>' + '01' as date) as dt,
year(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as y,
month(dateadd(month, n, cast('<year>' + '<month>' + '01' as date))) as m
from (values
(0), (-1), (-2), (-3), (-4), (-5),
(-6), (-7), (-8), (-9), (-10), (-11)
) ofs(n)
)
SELECT
year(max(dt)) as "year", month(max(dt)) as "month", p.product,
AVG(coalesce(value, 0)) as average,
SUM(value) as "sum"
FROM
ym cross join
(select distinct product from awesome) p
left outer join awesome t
on t."year" = ym.y and t."month" = ym.m and t.product = p.product
GROUP BY p.product
我不知道你是否只想限制在上一年实际销售的产品，所以我没有在那里处理。
如果您开始考虑如何使它更通用和可重用，您最终可能会更像这个。我继续将产品限制添加到 activity 的前一年：
WITH dt as (
select dateadd(month, n, cast('<year>' + '<month>' + '01' as date) as dt
),
ym as (
select dt, year(dateadd(month, n, dt)) as y, month(dateadd(month, n, dt)) as m
from (values
(0), (-1), (-2), (-3), (-4), (-5),
(-6), (-7), (-8), (-9), (-10), (-11)
) ofs(n), dt
)
SELECT
year(max(dt)) as "year", month(max(dt)) as "month", p.product,
AVG(coalesce(value, 0)) as average, SUM(value) as "sum"
FROM
ym cross join
(
select distinct product from awesome
where cast("year" + "month" + '01' as date) between
(select min(dt) from ym) and (select max(dt) from ym)
) p
left outer join (
select distinct /* get the latest "version" only */
first_value("year")
over (partition by "year", "month", product order by version desc),
first_value("month")
over (partition by "year", "month", product order by version desc),
product,
first_value(value)
over (partition by "year", "month", product order by version desc)
from awesome
) t
on t."year" = ym.y and t."month" = ym.m and t.product = p.product
GROUP BY p.product
最终查询还尝试仅针对最新版本处理您的过滤器。不过，您需要更高版本的 SQL 服务器才能实现 first_value() 功能。
最大的问题是您需要获得两个列表才能正确汇总数据集中的值 - 一个日期列表和一个产品列表。如果没有这两个列表，最后一个月缺少产品意味着不会报告该产品，或者（正如您已经发现的那样）可能会汇总错误的 12 个月（缺少 7 月意味着前面 11 行包括开始月份）。
下面是对生成这些列表的过程的全面扩展探索。它仅使用源数据 table（假设每个月售出 东西 。它可以变得更简洁（即计算日期，如上面 shawnt 的示例） , 而是为了展示所有的步骤和假设而写的。我把它封装成一个存储过程，因为它显式地展示了传入的值。
CREATE PROCEDURE DoTheThing
@startDate DATE -- Should be Year-Month-01 or YearMonth01
AS
BEGIN
DECLARE @yr INT, @mth INT,
@yr2 INT, @mth2 INT,
@endDate DATE -- StartDt - 11 months
-- if the date will be passed in with a day other than 01, add code here to set the day on the passed date to 01
-- if only the high year and month are passed in, then create a @startDt value and continue.
SET @endDate = DATEADD(MONTH, -11, @startDate)
SELECT @yr = DATEPART(YEAR, @startDate),
@mth = DATEPART(MONTH, @startdate),
@yr2 = DATEPART(YEAR, @startDate),
@mth2 = DATEPART(MONTH, @startdate)
WITH mthYr AS (
SELECT DISTINCT
YEAR,
MONTH
FROM suchDB.muchUSER.awesomeTABLE -- Get the data from the source table
WHERE (
YEAR = @yr -- if in the passed-in year, then take all months less than or equal to the start month
AND MONTH <= @mth
)
OR (
YEAR = @yr2 -- if the period is Jan -- Dec in one year, this reiterates the above
AND MONTH >= @mth2 -- if not, select the months in the second year where the month is greater than or equal to the calculated month
)
),
prods AS (
SELECT DISTINCT product -- Return a list of products sold during the year.
FROM suchDB.muchUSER.awesomeTABLE smt
INNER JOIN mthYr
ON mthYr.YEAR = smt.YEAR
AND mthyr.MONTH = smt.MONTH
)
SELECT MAX(mthYr.YEAR) AS year, -- current report only shows passed in year/month value
MAX(mthYr.month) AS month,
prods.product,
AVG(ISNULL(VALUE, 0.00)) average, -- isNull value adds a zero into the list to be averaged
SUM(ISNULL(VALUE, 0.00)) SUM -- not really necessary, but no warnings about NULL values will be generated
FROM mthYr CROSS JOIN prods -- cross join of the two lists means all of the products sold will have a value for each month
LEFT JOIN suchDB.muchUSER.awesomeTABLE smt -- left join so missing productMonths will still be added in
ON smt.YEAR = mthYr.year
AND smt.MONTH = mthYr.month
AND prods.product = smt.product
ORDER BY smt.product
END
处理日期时首先要做的是有一个日期时间字段，而不是年月日的分隔字段
SELECT year, month, product, value
, DATEFROMPARTS(year, month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE
由于可以对行进行版本控制，我们需要获取每年、每月和产品的最新版本，这可以通过多种方式完成，例如使用窗口功能或自动连接，例如后者是
SELECT b.[year], b.[month], b.[product], [value]
, DATEFROMPARTS(b.[year], b.month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE b
INNER JOIN (SELECT [year], [month], [product], max([version]) lv
FROM suchDB.muchUSER.awesomeTABLE
GROUP BY [year], [month], [product]
) m ON b.[year] = m.year AND b.month = m.month
AND b.product = m.product AND b.[version] = m.lv
还有产品清单会派上用场
SELECT DISTINCT product
FROM suchDB.muchUSER.awesomeTABLE
获取去年有价值的数据可以通过多种方式完成，我喜欢 CROSS APPLY
Declare @_year int
Declare @_month int
Set @_year = 2015
Set @_month = 11
;With baseDate AS (
SELECT b.[year], b.[month], b.[product], [value]
, DATEFROMPARTS(b.[year], b.month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE b
INNER JOIN (SELECT [year], [month], [product], max([version]) lv
FROM suchDB.muchUSER.awesomeTABLE
GROUP BY [year], [month], [product]
) m ON b.[year] = m.year AND b.month = m.month
AND b.product = m.product AND b.[version] = m.lv
), Products AS (
SELECT DISTINCT [product]
FROM suchDB.muchUSER.awesomeTABLE
)
SELECT @_year [Year], @_month [Month], p.[product]
, ly.Average
, ly.[Sum]
FROM Products p
CROSS APPLY (SELECT Sum(lastYear.Value) / 12.0 Average
, Sum(lastYear.Value) [Sum]
FROM baseDate lastYear
WHERE lastYear.fullDate > DATEFROMPARTS(@_year - 1, @_month, 1)
AND lastYear.fullDate <= DATEFROMPARTS(@_year, @_month, 1)
AND lastYear.product = p.product
) ly
WHERE ly.[Sum] IS NOT NULL
WHERE条件去掉参数前一年没有行的产品
要删除变量并获取每个月的值，需要一个日历 table。如果 table 中所有产品的所有月份都在那里，我们可以使用 DISTINCT 获取日期，使用第一个 CTE 而不是 table 让我们也得到完整日期
SELECT DISTINCT [year], [month], fullDate
FROM baseDate
否则有不同的方法来创建日历 table。我们可以将日历 table 添加到主查询 CTEs 并在 CROSS APPLY 中使用它而不是变量
;With baseDate AS (
SELECT b.[year] _y, b.[month] _m, b.[product], [value]
, DATEFROMPARTS(b.[year], b.month, 1) fullDate
FROM suchDB.muchUSER.awesomeTABLE b
INNER JOIN (SELECT [year], [month], [product], max([version]) lv
FROM suchDB.muchUSER.awesomeTABLE
GROUP BY [year], [month], [product]
) m ON b.[year] = m.year AND b.month = m.month
AND b.product = m.product AND b.[version] = m.lv
), Products AS (
SELECT DISTINCT [product]
FROM suchDB.muchUSER.awesomeTABLE
), Months As (
SELECT DISTINCT _y, _m, fullDate
FROM baseDate
)
SELECT _y [Year], _m [Month], p.[product]
, ly.Average
, ly.[Sum]
, ly.[Count]
FROM Products p
CROSS APPLY (SELECT m._y, m._m
, Sum(bd.Value) / 12.0 Average
, Sum(bd.Value) [Sum]
, Count(Value) [Count]
FROM Months m
LEFT JOIN baseDate bd
ON bd.fullDate > DATEADD(YY, -1, m.fullDate)
AND bd.fullDate <= m.fullDate
WHERE bd.product = p.product
GROUP BY m._y, m._m
) ly
WHERE ly.[Sum] IS NOT NULL