If I have a common table expression for a family with mother & father, how can I increment the 'Generation' counter? A family should have the child as generation zero, parents as generation 1, and the four grandparents as generation 2. But the loop is performed twice, one for each set of grandparents.
;WITH FamilyTree
AS
(
SELECT *, 0 AS Generation
FROM myTable
WHERE [id] = 99
UNION ALL
SELECT name, Generation + 1
FROM myTable AS Fam
INNER JOIN FamilyTree
ON Fam.[id] = FamilyTree.[motherid]
UNION ALL
SELECT name, Generation + 1
FROM myTable AS Fam
INNER JOIN FamilyTree
ON Fam.[id] = FamilyTree.[fatherid]
)
SELECT generation, name FROM FamilyTree
Change the join look at relatives in one generation gap in one go, rather then have 2 recursive clauses in the CTE. The 2 clauses form a a partial cross join which is why you have extra rows
;WITH FamilyTree
AS
(
SELECT *, 0 AS Generation
FROM myTable
WHERE [id] = 99
UNION ALL
SELECT name, Generation + 1
FROM myTable AS Fam
INNER JOIN FamilyTree
ON Fam.[id] IN (FamilyTree.[motherid], FamilyTree.[fatherid])
)
SELECT generation, name FROM FamilyTree
Related
I have one parent table Products with multiple child tables -Hoses,Steeltubes,ElectricCables,FiberOptics.
ProductId -Primary key field in Product table
ProductId- ForeignKey field in Hoses,Steeltubes,ElectricCables,FiberOptics.
Product table has 1 to many relationship with Child tables
I want to combine result of all tables .
For eg - Product P1 has PK field ProductId which is used in all child tables as FK.
If Hoses table has 4 record with ProductId 50 and Steeltubes table has 2 records with ProductId 50 when I perform left join then left join is doing cartesian product of records showing 8 record as result But it should be 4 records .
;with HOSESTEELCTE
as
(
select '' as ModeType, '' as FiberOpticQty , '' as NumberFibers, '' as FiberLength, '' as CableType , '' as Conductorsize , '' as Voltage,'' as ElecticCableLength , s.TubeMaterial , s.TubeQty, s.TubeID , s.WallThickness , s.DWP ,s.Length as SteelLength , h.HoseSeries, h.HoseLength ,h.ProductId
from Hoses h
left join
(
--'' as HoseSeries,'' as HoseLength ,
select TubeMaterial , TubeQty, TubeID , WallThickness , DWP , Length,ProductId from SteelTubes
) s on (s.ProductId = h.ProductId)
) select * from HOSESTEELCTE
Assuming there are no relationships between child tables and you simply want a list of all child entities which make up a product you could generate a cte which has a number of rows which are equal to the largest number of entries across all the child tables for a product. In the example below I have used a dates table to simplify the example.
so for this data
create table products(pid int);
insert into products values
(1),(2);
create table hoses (pid int,descr varchar(2));
insert into hoses values (1,'h1'),(1,'h2'),(1,'h3'),(1,'h4');
create table steeltubes (pid int,descr varchar(2));
insert into steeltubes values (1,'t1'),(1,'t2');
create table electriccables(pid int,descr varchar(2));
truncate table electriccables
insert into electriccables values (1,'e1'),(1,'e2'),(1,'e3'),(2,'e1');
this cte
;with cte as
(select row_number() over(partition by p.pid order by datekey) rn, p.pid
from dimdate, products p
where datekey < 20050105)
select * from cte
create a cartesian join (one of the rare ocassions where an implicit join helps) pid to rn
result
rn pid
-------------------- -----------
1 1
2 1
3 1
4 1
1 2
2 2
3 2
4 2
And if we add the child tables
;with cte as
(select row_number() over(partition by p.pid order by datekey) rn, p.pid
from dimdate, products p
where datekey < 20050106)
select c.pid,h.descr hoses,s.descr steeltubes,e.descr electriccables from cte c
left join (select h.*, row_number() over(order by h.pid) rn from hoses h) h on h.rn = c.rn and h.pid = c.pid
left join (select s.*, row_number() over(order by s.pid) rn from steeltubes s) s on s.rn = c.rn and s.pid = c.pid
left join (select e.*, row_number() over(order by e.pid) rn from electriccables e) e on e.rn = c.rn and e.pid = c.pid
where h.rn is not null or s.rn is not null or e.rn is not null
order by c.pid,c.rn
we get this
pid hoses steeltubes electriccables
----------- ----- ---------- --------------
1 h1 t1 e1
1 h2 t2 e2
1 h3 NULL e3
1 h4 NULL NULL
2 NULL NULL e1
In fact, the result having 8 rows can be expected to be the result, since your four records are joined with the first record in the other table and then your four records are joined with the second record of the other table, making it 4 + 4 = 8.
The very fact that you expect 4 records to be in the result instead of 8 shows that you want to use some kind of grouping. You can group your inner query issued for SteelTubes by ProductId, but then you will need to use aggregate functions for the other columns. Since you have only explained the structure of the desired output, but not the semantics, I am not able with my current knowledge about your problem to determine what aggregations you need.
Once you find out the answer for the first table, you will be able to easily add the other tables into the selection as well, but in case of large data you might get some scaling problems, so you might want to have a table where you store these groups, maintain it when something changes and use it for these selections.
I have a table CUST with following layout. There are no constraints. I do see that one ChildID has more than one ParentID associated with it. (Please see the records for ChildID = 115)
Here is what I need -
Wherever one child has more than 1 parent, I want to update those ParentID and ParentName with the ParentID and ParentName which has max match_per. So in the below image, I want ParentID 1111 and ParentName LEE YOUNG WOOK to update all records where ChildId = 115 (since the match_per 0.96 is maximum within the given set). In case there are two parents with equal max match_per, then I want to pick any 1 one of them.
I know it is possible using CTE but I don't know how to update CTE. Can anybody help?
One way of doing it
WITH CTE1 AS
(
SELECT *,
CASE WHEN match_per =
MAX(match_per) OVER (PARTITION BY ChildId)
THEN CAST(ParentId AS CHAR(10)) + ParentName
END AS parentDetailsForMax
FROM CUST
), CTE2 AS
(
SELECT *,
MAX(parentDetailsForMax) OVER (PARTITION BY ChildId) AS maxParentDetailsForMax
FROM CTE1
)
UPDATE CTE2
SET ParentId = CAST(LEFT(maxParentDetailsForMax,10) AS int),
ParentName = SUBSTRING(maxParentDetailsForMax,10,8000)
Getting both the parent id and parent name is a bit tricky. I think the logic is easiest using cross apply:
with toupdate as (
select t.*, p.parentId as new_parentId, p.parentName as new_parentName
max(match_per) over (partition by childid) as max_match_per,
count(*) over (partition by childid) as numparents
from table t cross apply
(select top 1 p.*
from table p
where p.childid = t.childid
order by match_per desc
) p
)
update toupdate
set parentId = new_ParentId,
parentName = new_ParentName
where numparents > 1;
As a note: the fact that parent id and parent name are both stored in the table, potentially multiple times seems like a problem. I would expect to look up the name, given the id, to reduce data redundancy.
Try something like this?? The first CTE will get Max(match_per) for each ChildID. Then, the second will use the new MaxMatchPer to find what its corresponding ParentID should be.
; WITH CTE AS (
SELECT ChildID,MAX(match_per) AS MaxMatchPer
FROM tbl
GROUP BY ChildID
), CTE1 AS (
SELECT t.ParentID, c.ChildID
FROM tbl t
JOIN CTE c
ON c.ChildID = t.ChildID
AND c.MaxMatchPer = t.match_per
)
UPDATE t
SET ParentID = c.ParentID
FROM tbl t
LEFT JOIN CTE1 c
ON c.ChildID = t.ChildID
Also, this is poor normalization. You should not have ParentName nor ChildName in this table.
I have a table dbo.Hierarchy that contains the following data:
Level1 Level2 Level3 Level4 Level5 Level6 Level7 Level8 Level9 Level10
-----------------------------------------------------------------------
a b c d e f g h i j
k l m n o
There are a total of 10 levels and any item can have hierarchy upto any level. In the above data a is the parent of b, b is the parent of c and so on. j and o are the last levels in their respective hierarchies. How can I get the output in the below format:
Name ParentName LevelID
-------------------------------
a NULL 1
b a 2
j i 10
k NULL 1
l k 2
o n 5
Something like (untested)
with t(L1,L2,L3,L4,L5,L6,L7,L8,L9,L10) as (
values ('a','b','c','d','e','f','g','h','i','j')
, ('k','l','m','n','o',null,null,null,null,null)
)
select x.*
from t
cross apply (
values (L1,null,1),(L2,L1,2),(L3,L2,3),(L4,L3,4),(L5,L4,5)
, (L6,L5,6),(L7,L6,7),(L8,L7,8),(L9,L8,9),(L10,L9,10)) x (name, parentname, levelid)
where name is not null
Try this:
;with base as
(select *, row_number() over (order by level1) rn from tbl),
src as
(
select
valname as name,
cast(substring(colname,6,len(colname)) as int) as level,
rn from
(select * from base) s
unpivot
(
valname
for colname in ([level1],[level2],[level3],[level4],[level5],[level6],[level7], [level8],[level9],[level10])
) u
),
cte as
(select * from src
where level = 1
union all
select s.* from src s
inner join cte c on s.level = c.level + 1 and s.rn = c.rn)
select distinct s.name, t.name parentname, s.level levelid from
cte s
left join cte t on s.rn = t.rn and s.level = t.level + 1
Breakdown:
CTE base is used to generate row number as a derived column. We will use this column to keep track of which values belong to which row. This will help in uniquely mapping children to their parents.
CTE src is where we transform the table from this denormalised structure to a normalised one. Using UNPIVOT, we reduce the data set to 3 columns - name, level and the row number rn.
CTE cte is a recursive CTE that we use to get all possible combinations of parents and children (including immediate parents as well as ancestors).
Finally, we LEFT JOIN cte to itself on the condition that row number is same on both sides of the join i.e. the values belong to same record from the base table, and also that the value from the right side is the immediate ancestor (parent) of the value on the left side.
Demo
The huge mass of code above can be avoided, if you were to choose a normalised structure for your table. I would suggest something like this:
CREATE TABLE tbl
(ID int, --Keep track of values that are related to each other
Name varchar(100), --Name
Level int --The level for a particular value
)
With this proposed structure, all you would need is the recursive CTE (cte from the above code) and the left join to get the parent-child data. The beauty of this approach is that you can extend it to any number of levels that you like, without having to hard-code the level numbers.
Demo
I have a table with following structure
Table name: matches
That basically stores which product is matching which product. I need to process this table
And store in a groups table like below.
Table Name: groups
group_ID stores the MIN Product_ID of the Product_IDS that form a group. To give an example let's say
If A is matching B and B is Matching C then three rows should go to group table in format (A, A), (A, B), (A, C)
I have tried looking into co-related subqueries and CTE, but not getting this to implement.
I need to do this all in SQL.
Thanks for the help .
Try this:
;WITH CTE
AS
(
SELECT DISTINCT
M1.Product_ID Group_ID,
M1.Product_ID
FROM matches M1
LEFT JOIN matches M2
ON M1.Product_Id = M2.matching_Product_Id
WHERE M2.matching_Product_Id IS NULL
UNION ALL
SELECT
C.Group_ID,
M.matching_Product_Id
FROM CTE C
JOIN matches M
ON C.Product_ID = M.Product_ID
)
SELECT * FROM CTE ORDER BY Group_ID
You can use OPTION(MAXRECURSION n) to control recursion depth.
SQL FIDDLE DEMO
Something like this (not tested)
with match_groups as (
select product_id,
matching_product_id,
product_id as group_id
from matches
where product_id not in (select matching_product_id from matches)
union all
select m.product_id, m.matching_product_id, p.group_id
from matches m
join match_groups p on m.product_id = p.matching_product_id
)
select group_id, product_id
from match_groups
order by group_id;
Sample of the Recursive Level:
DECLARE #VALUE_CODE AS VARCHAR(5);
--SET #VALUE_CODE = 'A' -- Specify a level
WITH ViewValue AS
(
SELECT ValueCode
, ValueDesc
, PrecedingValueCode
FROM ValuesTable
WHERE PrecedingValueCode IS NULL
UNION ALL
SELECT A.ValueCode
, A.ValueDesc
, A.PrecedingValueCode
FROM ValuesTable A
INNER JOIN ViewValue V ON
V.ValueCode = A.PrecedingValueCode
)
SELECT ValueCode, ValueDesc, PrecedingValueCode
FROM ViewValue
--WHERE PrecedingValueCode = #VALUE_CODE -- Specific level
--WHERE PrecedingValueCode IS NULL -- Root
In the code below I am using a recursive CTE(Common Table Expression) in SQL Server 2005 to try and find the top level parent of a basic hierarchical structure. The rule of this hierarchy is that every CustID has a ParentID and if the CustID has no parent then the ParentID = CustID and it is the highest level.
DECLARE #LookupID int
--Our test value
SET #LookupID = 1
WITH cteLevelOne (ParentID, CustID) AS
(
SELECT a.ParentID, a.CustID
FROM tblCustomer AS a
WHERE a.CustID = #LookupID
UNION ALL
SELECT a.ParentID, a.CustID
FROM tblCustomer AS a
INNER JOIN cteLevelOne AS c ON a.CustID = c.ParentID
WHERE c.CustID <> a.CustomerID
)
So if tblCustomer looks like this:
ParentID CustID
5 5
1 8
5 4
4 1
The result I get from the code above is:
ParentID CustID
4 1
5 4
5 5
What I want is just the last row of that result:
ParentID CustID
5 5
How do I just return the last record generated in the CTE (which would be highest level CustID)?
Also note that there are multiple unrelated CustID hierarchies in this table so I can't just do a SELECT * FROM tblCustomer WHERE ParentID = CustID. I can't order by ParentID or CustID because the ID number is not related to where it is in the hierarchy.
If you just want want the highest recursion depth couldn't you do something like this?Then, when you actually query the CTE just look for the row with max(Depth)? Like so:
DECLARE #LookupID int
--Our test value
SET #LookupID = 1;
WITH cteLevelOne (ParentID, CustID, Depth) AS
(
SELECT a.ParentID, a.CustID, 1
FROM tblCustomer AS a
WHERE a.CustID = #LookupID
UNION ALL
SELECT a.ParentID, a.CustID, c.Depth + 1
FROM tblCustomer AS a
INNER JOIN cteLevelOne AS c ON a.CustID = c.ParentID
WHERE c.CustID <> a.CustID
)
select * from CTELevelone where Depth = (select max(Depth) from CTELevelone)
or, adapting what trevor suggests, this could be used with the same CTE:
select top 1 * from CTELevelone order by Depth desc
I don't think CustomerID was necessarily what you wanted to order by in the case you described, but I wasn't perfectly clear on the question either.
I'm not certain I fully understand the problem, but just to hack & slash at it you could try:
SELECT TOP 1 FROM cteLevelOne ORDER BY CustID DESC
That assumes that the CustID is also in order as in the example, and not something like a GUID.
First the cte will not be finished if any of the parent child are same. As it is a recursive CTE it has to be terminated. Having Parent and cust id same , the loop will not end.
Msg 530, Level 16, State 1, Line 15
The statement terminated. The maximum recursion 100 has been exhausted before statement completion.