I once walked into a bar, and saw two construction workers pounding each other to a pulp. The argument was over what was the better tool—a hammer or screwdriver. I feel a similar sensation when I see SQL developers arguing over whether to use natural or surrogate keys. Few other arguments in database design can cause tempers to flare so quickly. Fans of surrogates consider anyone who uses a natural key a drooling idiot, whereas natural key acolytes believe the use of a surrogate warrants burning alive at the stake.
Which side is right? Neither. Both natural and surrogate keys have their own place, and a developer who doesn’t use both as the situation demands is shortchanging both himself and his applications.
A natural key is simply a column or set of columns in a table that uniquely identifies each row. Natural keys are a feature of the data, not the database, and thus have business meaning on their own. Quite often a natural key is more than a single column. For instance, the natural key for a table of addresses might be the five columns: street number, street name, city, state, and zip code.
What is a surrogate key? Most people will define it as some variation of “a system-generated value used to uniquely identify a row”. Unlike natural keys, surrogates have no business meaning. In SQL Server, by far the most common technique for generating surrogate values is the ubiquitous IDENTITY column.
While the above definition is true, there’s another very important part to it. The value of a surrogate must never be exposed to the outside world. Users should never be allowed to see the key, under any conditions. Display the value of a surrogate key on a report, allow it to be viewed on a form or even used as a search term – these are all forbidden. Once you expose a surrogate key, it immediately begins acquiring business meaning.
Smart Keys: The Worst of Both Worlds
A smart key is an artificial key with one or more parts that contain business meaning. For instance, an employee table where each primary key begins with the initial of the employee (“JD1001” for John Doe), or a table of paint products, where the key identifies the can size, color, and paint type, i.e. “1G-RED- LATEX”.
Smart keys are a sort of hybrid between natural and surrogate keys. They’re seductively attractive to many developers, but should be avoided like the plague. They tend to make your design very brittle and subject to failure as business rules change.
Note: if your data already contains meaningful product codes or other keys such as those described above, then they are simply natural keys and the above caveat doesn’t apply. It’s a smart key only when the value is constructed by the developer.
Benefits of Natural Keys
A natural key is…well, natural. Since its values already exist in the data, using a natural key means you don’t have to add and maintain a new column. This also means smaller tables and less storage requirements. As more rows fit on a database page, it can sometimes mean greater performance. (It can also mean less- more on this later) . However, in practical terms, the space savings are minor, except for very narrow tables. Using a surrogate key usually means an additional index is required, though.
Generating sequential key values is inherently a serial process, so using a natural key rather than an IDENTITY column can be a performance boost for inserts, especially in OLTP environments.
Since natural key values are used as foreign keys in child tables, it can mean the elimination of joins for queries that require no other columns from the parent other than the natural key.
One benefit of natural keys often claimed by its proponents is that they can aid in self-documenting your database schema. Explicitly naming each natural key documents what specifically identifies each row in a table, and joining on natural keys helps to identify the natural relationships between tables. This argument based onelegance appeals strongly to those working in academia; it may or may not have much value for developers working in the dirt and grime of real production systems.
Benefits of Surrogate Keys
Since surrogate values are controlled by the system, you never have to worry about duplicate, missing, or changing values. They’re also an easy and reliable way to join tables; when writing queries, you never need worry about remembering which combination of columns is the natural key. However, some of these benefits are less compelling than they seem. I’ll discuss separately each case for using a surrogate, and whether or not it holds up.
When no natural key on the table exists – Yes.
If the table has no unique identifier, then you must create one. Unkeyed tables are in general a very bad idea. There are exceptions such as logging or summary tables in which rows are only inserted, never updated. Otherwise, if your table doesn’t have a unique key—create one.
When the natural key can change- Sometimes.
Immutability is certainly a desirable feature for a key, but it’s by no means a requirement. Using this as an excuse is the trademark of a lazy developer. Data changes. If the key changes, update it. If it’s being used as a foreign key in another table – update that table too. Cascading updates exist for a reason, after all.
Obviously if a key changes very often, or will be used as a FK for many other tables, there can be performance or concurrency implications to making it the primary key . In this case, you do want to consider a surrogate, and use it if appropriate.
There’s one particular case where the stability of a surrogate key actually works against you. For lookup tables, particular those containing selection options for given fields, changes to the lookup value are often not meant to be cascaded into child tables. For example, an application may store the “referral source” for new customers or marketing leads, whether they were generated via an ad in a newspaper or magazine, a yellow pages entry, word of mouth, etc. These referral codes can be very specific, and are normally stored in a lookup table. Once set, the code should be preserved historically, even if the original lookup table value is updated or removed. This is behavior very difficult to achieve with a surrogate key, but trivial with a natural key.
When natural key values can be missing or duplicated – No.
This is probably the most misunderstood aspect of the natural vs. surrogate debate. Natural keys are unique by definition. If it isn’t a serious error for a value to be missing or duplicated, then that value isn’t a natural key to begin with. And if it is an error, then you’re almost always better off trapping that error at the database level, rather than allowing that bad data into your DB.
As example, consider a table of employees keyed off Social Security Number. Users are complaining the database throws errors when they don’t have a SSN or mistakenly enter a duplicate. So you replace the SSN PK with a surrogate and smugly conclude you’ve solved the problem. But have you? Now some employees don’t have SSNs, and the accounting module starts failing when printing tax records … or worse, collates all the NULL SSN entries together, reporting them as a single employee. The search function starts returning the wrong rows because some employees are sharing the same SSNs, and the new hire in the mailroom gets the boss’s paycheck, because someone in HR accidentally cut and pasted a SSN.
In reality, all you’ve done is short-circuit out the data integrity safeguards in your database, and pass the responsibility for the problem up to the application level. Bad move.
These sorts of problems exist because most tables have a uniqueness requirement at the tion level. A surrogate key only solves the uniqueness problem at the database level, but users (who cannot and should not see the surrogate value) still don’t have a way to uniquely identify each record. This also explains why when, even if you choose to use a surrogate key, you will usually want to also add a unique constraint on the original natural key, since uniqueness is no longer being automatically enforced by the PK.
But wait a minute! What if your business rules specifically require employees to be input before you have their SSN data? Or what if your table holds overseas employees that may not have a SSN at all? Does that mean you can’t use a natural key? Maybe. One possibility is to assign your own unique values in these cases. One system I’ve seen used randomly generated alphabetic values for temporary SSNs, whereas the standard numeric value identified a “real” one. Better yet is to examine your table for some other column or columns that can be used as a natural key. Or maybe you really do want to drop natural keys altogether. The point here is not that surrogates should never be used, but simply that if your natural key isn’t unique, you are going to have problems beyond those that a surrogate will solve.
When the natural key is very wide, or a composite of multiple columns – Sometimes .
Wide keys make for fat indexes. Fat indexes have performance implications. A very wide key can hurt performance far more than the extra space required by a surrogate. Replacing a composite key with a surrogate also simplifies your queries, but this should never be a primary consideration. It’s poor form to replace the natural key of a two CHAR(2) columns with am INT IDENTITY, for no other reason than it makes your queries prettier.
One common case where a multi-column natural key should always be used is the so-called junction table: a table used to implement a many-many relationship between two other tables. Most junction tables have only two columns, each a FK back to a parent table. The combination of these two FKs is itself the primary key for the table. Adding a surrogate to a table like this is just asking for trouble.
For performance reasons.
This is the trickiest question of all. Replacing a wide key with a narrower value means smaller indexes, and more values retrieved from every index page read. This does boost performance. However, you’ll usually retain the index on the natural key (to enforce uniqueness if nothing else) and that means another index to maintain. If your table is very narrow, the additional column for the surrogate can noticeably impact performance and storage requirements. Finally, some queries that may have not required JOINs with a natural foreign key may now need them. For instance, our employee SSN example might have a child table containing reported hours worked:
(Foreign key to Employee table)
If the EmployeeID FK is SSN, then we can retrieve a list of total hours by SSN from this table alone:
SELECT EmployeeID, SUM(DATEDIFF(hr,StopTime,StartTime))
GROUP BY EmployeeID
With EmployeeID as a surrogate key, however, we must JOIN back to the EMPLOYEE table:
SELECT SSN, SUM(DATEDIFF(hr,StopTime,StartTime))
FROM ReportedHours h
JOIN Employees e ON h.EmployeeID = e.EmployeeID
GROUP BY e.SSN