Managing hierarchies in SQL: MPTT/nested sets vs adjacency lists vs storing paths

Question

For a while now I\'ve been wrestling with how best to handle hierarchies in SQL. Frustrated by the limitations of adjacency lists and the complexity of MPTT/nested sets, I b

Answer 1

It problem with your conclusion is that it ignores most of the issues involved in working with trees.

By reducing the validity of a technique to the "number of calls" you effectively ignore all of the issues which well understood data structures and algorithms attempt to solve; that is, fastest execution and low memory and resource foot print.

The "number of calls" to an SQL server may seem like a good metric to use ("look ma less code"), but if the result is a program which never finishes, runs slowly, or takes up to much space, it is in fact a useless metric.

By storing the path with every node you are not creating a tree data structure. Instead you are creating a list. Any operation which a tree is designed to optimize is lost.

This might be hard to see with small date sets (and in many cases of small trees a list is better), try some examples on data sets of size 500, 1000, 10k -- You will quickly see why storing the whole path is not a good idea.

Answer 2

You might also consider the Closure Table design I describe in my answer to What is the most efficient/elegant way to parse a flat table into a tree?

Calls required to create/delete/move a node:

• Closure = 1

Calls required to get a tree:

• Closure = 1

Calls required to get path to a node / ancestry:

• Closure = 1

Calls required to get number of subnodes:

• Closure = 1

Calls required to get depth of node:

• Closure = 1

DB fields required:

• Adjancency = 1 more field / row
• Path = 1 more field / row
• MPTT = 2 or 3 more fields / row
• Closure = 2 or 3 fields in extra table. This table has O(n^2) rows worst case but far fewer than that in most practical cases.

There are a couple of other considerations:

Supports unlimited depth:

• Adjacency = yes
• MPTT = yes
• Path = no
• Closure = yes

Supports referential integrity:

• Adjacency = yes
• MPTT = no
• Path = no
• Closure = yes

I also cover Closure Table in my presentation Models for Hierarchical Data with SQL and PHP, and my book, SQL Antipatterns: Avoiding the Pitfalls of Database Programming.