Flatten Hierarchy on Self-Join Table
I Have Data in a Self-Join Hierarchical Table Where Continents Have Many Countries Have Many Regions Have Many States Have Many Cities. Self-Joining Table...
I have data in a self-join hierarchical table where Continents have many Countries have many Regions have many States have many Cities.
Self-joining table structure:
|-------------------------------------------------------------|
| ID | Name | Type | ParentID | IsTopLevel |
|-------------------------------------------------------------|
| 1 | North America | Continent | NULL | 1 |
| 12 | United States | Country | 1 | 0 |
| 113 | Midwest | Region | 12 | 0 |
| 155 | Kansas | State | 113 | 0 |
| 225 | Topeka | City | 155 | 0 |
| 2 | South America | Continent | NULL | 1 |
| 22 | Argentina | Country | 2 | 0 |
| 223 | Southern | Region | 22 | 0 |
| 255 | La Pampa | State | 223 | 0 |
| 777 | Santa Rosa | City | 255 | 0 |
|-------------------------------------------------------------|
I have been able to successfully use a recursive CTE to get the tree structure and depth of each node. Where I am failing is using a pivot to create a nice list of all bottom locations and their corresponding parents at each level.
The expected results:
|------------------------------------------------------------------------------------|
| Continent | Country | Region | State | City | Bottom_Level_ID |
|------------------------------------------------------------------------------------|
| North America | United States | Midwest | Kansas | Topeka | 234 |
| South America | Argentina | Southern | La Pampa | Santa Rosa | 777 |
|------------------------------------------------------------------------------------|
There are a few key points I should clarify.
Every single entry has a bottom level and a top level. There are no cases where all five Types are not present for a given location.
If I filled out this data, I'd have 50 entries for North America at the State level, so you can imagine how immense this table is at the City level for every continent on the planet. Billions of rows.
The reason this is a necessity is because I need to be able to join onto a historical table of all addresses a person has lived at, and journey up the tree. I figure if I have the LocationID from that table, I can just LEFT JOIN onto a View of this query and nab the appropriate columns.
This is an old database, 2005, and I don't have sysadmin or control of the schema.
My CTE Code
--CTE
;WITH Tree
AS (
SELECT ID, Name, ParentID, Type, 1 as Depth
FROM LocationTable
WHERE IsTopLevel = 1
UNION ALL
SELECT L.ID, L.Name, L.ParentID, L.Type, T.Depth+1
FROM Tree T
JOIN LocationTable L
ON L.ParentGUID = T.GUID
)
Good solid data, in a mostly useful format. BUT then I got to thinking about it and isn't the table structure already in this format, so why would I bother doing a depth tree search if I wasn't going to join the entries together at the same time?
Anyway, here was the rest.
The Pivot Attempt
;WITH Tree
AS (
SELECT ID, Name, ParentID, Type
FROM LocationTable
WHERE IsTopLevel = 1
UNION ALL
SELECT L.ID, L.Name, L.ParentID, L.Type
FROM Tree T
JOIN LocationTable L
ON L.ParentGUID = T.GUID
)
select *
from Tree
pivot (
max(Name)
for Type in ([Continent],[Country],[Region],[State],[City])
) pvt
And now I have everything by Type in a column, with nulls for everything else. As I have struggled with before, I need to filter/join the CTE data before I attempt my pivot, but I have no idea where to start with that piece. Everything I have tried is soooooooooo sloooooooow.
Everytime I think I understand CTEs and Pivot, something new makes me extremely humbled. Please help me. ; ;
2 Answers
If your structure is as clean as you describe it (no gaps, 5 levels always) you might go the easy way:
This data really demands for a classical 1:n-table-tree, where your Countries, States etc. live in their own tables and link to their parent record
Make sure there's an index on ParentID and ID!
DECLARE @tbl TABLE(ID INT,Name VARCHAR(100),Type VARCHAR(100),ParentID INT,IsTopLevel BIT);
INSERT INTO @tbl VALUES
(1,'North America','Continent',NULL,1)
,(12,'United States','Country',1,0)
,(113,'Midwest','Region',12,0)
,(155,'Kansas','State',113,0)
,(225,'Topeka','City',155,0)
,(2,'South America','Continent',NULL,1)
,(22,'Argentina','Country',2,0)
,(223,'Southern','Region',22,0)
,(255,'La Pampa','State',223,0)
,(777,'Santa Rosa','City',255,0);
SELECT Level1.Name AS Continent
,Level2.Name AS Country
,Level3.Name AS Region
,Level4.Name AS State
,Level5.Name AS City
,Level5.ID AS Bottom_Level_ID
FROM @tbl AS Level1
INNER JOIN @tbl AS Level2 ON Level1.ID=Level2.ParentID
INNER JOIN @tbl AS Level3 ON Level2.ID=Level3.ParentID
INNER JOIN @tbl AS Level4 ON Level3.ID=Level4.ParentID
INNER JOIN @tbl AS Level5 ON Level4.ID=Level5.ParentID
WHERE Level1.ParentID IS NULL
The result
Continent Country Region State City Bottom_Level_ID
North America United States Midwest Kansas Topeka 225
South America Argentina Southern La Pampa Santa Rosa 777
Another solution with CTE could be :
;WITH Tree
AS (
SELECT cast(NULL as varchar(100)) as C1, cast(NULL as varchar(100)) as C2, cast(NULL as varchar(100)) as C3, cast(NULL as varchar(100)) as C4, Name as C5, ID as B_Level
FROM LocationTable
WHERE IsTopLevel = 1
UNION ALL
SELECT T.C2, T.C3, T.C4, T.C5, L.Name, L.ID
FROM Tree T
JOIN LocationTable L
ON L.ParentID = T.B_Level
)
select *
from Tree
where C1 is not null
Chloe Bennett explores the intersection of pop culture, streaming entertainment, digital trends, and contemporary lifestyle. Her weekly commentary reaches thousands of culture enthusiasts.
