smcjones smcjones - 1 year ago 142
SQL Question

Composite vs Surrogate keys for Referential Integrity in 6NF

Take three layers of information:

Layer 1: Information

This layer contains data with
natural indexes and a surrogate key that is easily transferrable.

Table Surnames:

| ID (Auto Increment, PK) | Surname |
| 1 | Smith |
| 2 | Edwards |
| 3 | Brown |

Table FirstNames

| ID (Auto Increment, PK) | FirstName |
| 1 | John |
| 2 | Bob |
| 3 | Mary |
| 4 | Kate |

Natural Keys

Alternatively, the two tables above can be without
and utilize Surname and FirstName as Natural Primary Keys, as explained by Mike Sherrill. In this instance, assume the layer below references
rather than

Layer 2: People

In this layer a composite index is used. This value can be
, depending on whether a surrogate key is utilized as the Primary Key.

| FirstName | LastName |
| 1 | 2 |
| 1 | 3 |
| 2 | 3 |
| 3 | 1 |
| 4 | 2 |
| ... | ... |

Layer 3: Parents

In this layer, relationships between people are explored through a


| Person | PersonParent |


| PersonFirstName | PersonSurname | ParentFirstName | ParentSurname |

The Question

Assuming that referential integrity is VERY important to me at its very core, and I will have
on these indexes so that I keep the database responsible for monitoring its own integrity on this front, and that, if I were to use an ORM, it would be one like Doctrine which has native support for Compound Primary Keys...

Please help me to understand:

  • The list of trade-offs that take place with utilizing surrogate keys vs. natural keys on the 1st Layer.

  • The list of trade-offs that take place with utilizing compound keys vs. surrogate keys on the 2nd Layer which can be transferred over to the 3rd Layer.

I am not interested in hearing which is better, because I understand that there are significant disagreements among professionals on this topic and it would be sparking a religious war. Instead, I am asking, very simply and as objectively as is humanly possible, what trade-offs will you be taking by passing surrogate keys to each Layer vs maintaining Primary keys (natural/composite, or surrogate/composite). Anyone will be able to find someone saying NEVER or ALWAYS use surrogate keys on SO and other websites. Instead, a reasoned analyses of trade-offs is what I will most appreciate in your answers.

Thank you!

Answer Source

Here's some trade-offs:

Single Surrogate (artificially created):

  • All child tables foreign keys only need a single column to reference the primary key.

  • Very easy to update the natural keys in table without needing to update every child table with foreign keys

  • Smaller primary/foreign key indexes (ie. not a wide) This can make the database run faster, for example when a record is deleted in a parent table, the child tables need to be searched to make sure this will not create orphans. Narrow indexes are faster to scan (just sightly).

  • you will have more indexes because you most likely will also want to index whatever natural keys exists in the data.

Natural composite keyed tables:

  • fewer indexes in the database

  • less columns in the database

  • easier/faster to insert a ton of records as you will not need to grab the sequence generator

  • updating one of the keys in the compound requires that every child table also be updated.

Then there is another category: artificial composite primary keys

I've only found one instance where this makes sense. When you need to tag every record in every table for row level security.

For example, suppose you had an database which stored data for 50,000 clients and each client was not supposed to see other client's data--very common in web application development.

If each record was tagged with a client_id field, you are creating a row level security environment. Most databases have the tools to enforce row level security when setup correctly.

First thing to do is setup primary and foreign keys. Normally a table with have an id field as the primary key. By adding client_id the key is now composite key. And it is necessary to carry client_id to all child table.

The composite key is based on 2 surrogate keys and is a bulletproof way to ensure data integrity among clients and within the database a whole.

After this you would create views (or if using Oracle EE setup Virtual Private Database) and other various structures to allow the database to enforce row level security (which is a topic all it own).

Granted that this data structure is no longer normalized to the nth degree. The client_id field in each pk/fk denormalizes an otherwise normal model. The benefit of the model is the ease of enforcing row level security at the database level (which is what databases should do). Every select, insert, update, delete is restricted to whatever client_id your session is currently set. The database has session awareness.


Surrogate keys are always the safe bet. They require a little more work to setup and require more storage.

The biggest benefit in my opinion is:

  • Being able to update the PK in one table and all other child tables are instantaneously changed without ever being touched.

  • When data gets messed up--and it will at some point due to a programming mistake, surrogate keys make the clean up much much easier and in some cases only possible to do because there are surrogate keys.

  • Query performance is improved as the db is able to search attributes to locate the s.key and then join all child table by a single numeric key.

Natural Keys especially composite NKeys make writing code a pain. When you need to join 4 tables the "where clause" will be much longer (and easier to mess up) than when single SKeys were used.

Surrogate keys are the "safe" route. Natural keys are beneficial in a few places, I'd say around 1% of the tables in a db.

Recommended from our users: Dynamic Network Monitoring from WhatsUp Gold from IPSwitch. Free Download