tags:

views:

2707

answers:

10

What are the ways that you use to model and retrieve hierarchical info in a database?

+8  A: 

What's the best way to represent a hierachy in a SQL database? A generic, portable technique?

Let's assume the hierachy is mostly read, but isn't completely static. Let's say it's a family tree.

Here's how not to do it:

create table person (
person_id integer autoincrement primary key,
name      varchar(255) not null,
dob       date,
mother    integer,
father    integer
);

And inserting data like this:

person_id   name      dob       mother father  
1           Pops      1900/1/1   null   null  
2           Grandma   1903/2/4   null   null  
3           Dad       1925/4/2   2      1  
4           Uncle Kev 1927/3/3   2      1
5           Cuz Dave  1953/7/8   null   4
6           Billy     1954/8/1   null   3

Instead, split your nodes and your relationships into two tables.

create table person (
person_id integer autoincrement primary key,
name      varchar(255) not null,
dob       date
);

create table ancestor (
ancestor_id   integer,
descendant_id integer,
distance      integer
);

Data is created like this:

person_id   name      dob       
1           Pops      1900/1/1  
2           Grandma   1903/2/4   
3           Dad       1925/4/2   
4           Uncle Kev 1927/3/3
5           Cuz Dave  1953/7/8   
6           Billy     1954/8/1   

ancestor_id  descendant_id  distance
1            1              0
2            2              0
3            3              0
4            4              0
5            5              0
6            6              0
1            3              1
2            3              1
1            4              1
2            4              1
1            5              2
2            5              2
4            5              1
1            6              2
2            6              2
3            6              1

you can now run arbitary queries that don't involve joining the table back on itself, which would happen if you have the heirachy relationship in the same row as the node.

Who has grandparents?

select * from person where person_id in 
    (select descendant_id from ancestor where distance=2);

All your descendants:

select * from person where person_id in 
    (select descendant_id from ancestor 
    where ancestor_id=1 and distance>0);

Who are uncles?

select decendant_id uncle from ancestor 
    where distance=1 and ancestor_id in 
    (select ancestor_id from ancestor 
        where distance=2 and not exists
        (select ancestor_id from ancestor 
        where distance=1 and ancestor_id=uncle)
    )

You avoid all the problems of joining a table to itself via subqueries, a common limitation is 16 subsuqeries.

Trouble is, maintaining the ancestor table is kind of hard - best done with a stored procedure.

Josh
bacar
+1  A: 

FYI: SQL Server 2008 introduces a new HierarchyID data type for this sort of situation. Gives you control over where in the "tree" your row sits, horizontally as well as vertically.

Matt Hamilton
+2  A: 

Oracle: SELECT ... START WITH ... CONNECT BY

Oracle has an extension to SELECT that allows easy tree-based retrieval. Perhaps SQL Server has some similar extension?

This query will traverse a table where the nesting relationship is stored in parent and child columns.

select * from my_table
    start with parent = :TOP
    connect by prior child = parent;

http://www.adp-gmbh.ch/ora/sql/connect_by.html

Mark Harrison
A: 

If you're using SQL Server 2005 then this link explains how to retrieve hierarchical data.

Common Table Expressions (CTEs) can be your friends once you get comfortable using them.

Alison
+2  A: 

A portable technique is discussed in this paper:

Maintaining Transitive Closure of Graphs in SQL

Apocalisp
+9  A: 

I've got to disagree with Josh. What happens if you're using a huge hierarchical structure like a company organization. People can join/leave the company, change reporting lines, etc... Maintaining the "distance" would be a big problem and you would have to maintain two tables of data.

This query (SQL Server 2005 and above) would let you see the complete line of any person AND calculates their place in the hierarchy and it only requires a single table of user information. It can be modified to find any child relationship.

--Create table of dummy data
create table #person (
personID integer IDENTITY(1,1) NOT NULL,
name      varchar(255) not null,
dob       date,
father    integer
);

INSERT INTO #person(name,dob,father)Values('Pops','1900/1/1',NULL);  
INSERT INTO #person(name,dob,father)Values('Grandma','1903/2/4',null);
INSERT INTO #person(name,dob,father)Values('Dad','1925/4/2',1);
INSERT INTO #person(name,dob,father)Values('Uncle Kev','1927/3/3',1);
INSERT INTO #person(name,dob,father)Values('Cuz Dave','1953/7/8',4);
INSERT INTO #person(name,dob,father)Values('Billy','1954/8/1',3);

DECLARE @OldestPerson INT; 
SET @OldestPerson = 1; -- Set this value to the ID of the oldest person in the family

WITH PersonHierarchy (personID,Name,dob,father, HierarchyLevel) AS
(
   SELECT
      personID
      ,Name
      ,dob
      ,father,
      1 as HierarchyLevel
   FROM #person
   WHERE personID = @OldestPerson

   UNION ALL

   SELECT
    e.personID,
      e.Name,
      e.dob,
      e.father,
      eh.HierarchyLevel + 1 AS HierarchyLevel
   FROM #person e
      INNER JOIN PersonHierarchy eh ON
         e.father = eh.personID
)

SELECT *
FROM PersonHierarchy
ORDER BY HierarchyLevel, father;

DROP TABLE #person;
Alison
+1  A: 

I prefer a mix of the techinques used by Josh and Mark Harrison:

Two tables, one with the data of the Person and other with the hierarchichal info (person_id, parent_id [, mother_id]) if the PK of this table is person_id, you have a simple tree with only one parent by node (which makes sense in this case, but not in other cases like accounting accounts)

This hiarchy table can be transversed by recursive procedures or if your DB supports it by sentences like SELECT... BY PRIOR (Oracle).

Other posibility is if you know the max deep of the hierarchy data you want to mantain is use a single table with a set of columns per level of hierarchy

Telcontar
+7  A: 

The definitive pieces on this subject have been written by Joe Celko, and he has worked a number of them into a book called Joe Celko's Trees and Hierarchies in SQL for Smarties.

He favours a technique called directed graphs. An introduction to his work on this subject can be found here

andy47
I thought people call the technique "nested sets".
ChrisW
more detailed explanation of this technique is at http://dev.mysql.com/tech-resources/articles/hierarchical-data.html
lubos hasko
The problem I see with the Nested Sets Model is that every time you add or remove a node you have to update the whole right of the tree. That does not sound very scalable to me. Are there any benchmarks? I think filesystems do not use the Nested Sets Model to prevent traversing the whole tree, so they use the Adjacency List Model instead. @Quassnoi explains it quite well: http://explainextended.com/2009/09/24/adjacency-list-vs-nested-sets-postgresql/
Eduardo
+6  A: 

I like the Modified Preorder Tree Traversal Algorithm. This technique makes it very easy to query the tree.

But here is a list of links about the topic which I copied from the Zend Framework (PHP) contributors webpage (posted there by Posted by Laurent Melmoux at Jun 05, 2007 15:52).

Many of the links are language agnostic:

There is 2 main representations and algorithms to represent hierarchical structures with databases :

  • nested set also known as modified preorder tree traversal algorithm
  • adjacency list model

It's well explained here:

Here are some more links that I've collected:

adjacency list model

nested set

Graphes

Classes :

Nested Sets DB Tree Adodb

Visitation Model ADOdb

PEAR::DB_NestedSet

PEAR::Tree

nstrees

tharkun
+1  A: 

We had the same issue when we implemented a tree component for [fleXive] and used the nested set tree model approach mentioned by tharkun from the MySQL docs.

In addition to speed things (dramatically) up we used a spreaded approach which simply means we used the maximum Long value for the top level right bounds which allows us to insert and move nodes without recalculating all left and right values. Values for left and right are calculated by dividing the range for a node by 3 und use the inner third as bounds for the new node.

A java code example can be seen here.

Markus Plesser