Re: Auto increment

From: --CELKO-- <71062.1056_at_compuserve.com>
Date: 10 May 2003 20:30:20 -0700
Message-ID: <c0d87ec0.0305101930.65eb8cca_at_posting.google.com>


Very bad. The IDENTITY column is a hold over from the early programming language which were <i>very<i> close to the hardware. The fields in a COBOL or FORTRAN program were assumed to be physically located in main storage in the order and format in which they were declared in the program. The records were accessed in contigous physical secondary storage. The records had a physical position that was used to locate them; there was a first, second,.. last record in a file.

The early SQLs were based on existing file systems. The data was kept in physically contiguous disk pages, in physically contiguous rows, made up of physically contiguous columns. In short, just like a deck of punch cards or a magnetic tape.

But physically contiguous storage is only one way of building a relational database and it is not always the best one. But aside from that, the whole idea of a relational database is that user is not supposed to know how things are stored at all, much less write code that depends on the particular physical representation in a particular release of a particular product.

One of the biggest errors is the IDENTITY column in the Sybase family
(SQL Server and Sybase). People actually program with this "feature"
and even use it as the primary key for the table! Now, let's go into painful details as to why this thing is bad.

The practical considerations are that IDENTITY is proprietary and non-portable, so you know that you will have maintenance problems when you change releases or port your system to other products.

But let's look at the logical problems. First try to create a table with two columns and try to make them both IDENTITY columns. If you cannot declare more than one column to be of a certain datatype, then that thing is not a datatype at all, by definition.

Next, create a table with one column and make it an IDENTITY column. Now try to insert, update and delete different numbers from it. If you cannot insert, update and delete rows from a table, then it is not a table by definition.

Finally create a simple table with one IDENTITY column and a few other columns. Use a few statements like

INSERT INTO Foobar (a, b, c) VALUES  ('a1', 'b1', 'c1'); 
INSERT INTO Foobar (a, b, c) VALUES  ('a2', 'b2', 'c2'); 
INSERT INTO Foobar (a, b, c) VALUES  ('a3', 'b3', 'c3'); 

to put a few rows into the table and notice that the IDENTITY column sequentially numbered them in the order they were presented. If you delete a row, the gap in the sequence is not filled in and the sequence continues from the highest number that has ever been used in that column in that particular table.

But now use a statement with a query expression in it, like this:

INSERT INTO Foobar (a, b, c)
SELECT x, y, z
  FROM Floob;

Since a query result is a table, and a table is a set which has no ordering, what should the IDENTITY numbers be? The entire, whole, completed set is presented to Foobar all at once, not a row at a time.  There are (n!) ways to number (n) rows, so which one do you pick? The answer has been to use whatever the physical order of the result set happened to be. That non-relational phrase "physical order" again.

But it is actually worse than that. If the same query is executed again, but with new statistics or after an index has been dropped or added, the new execution plan could bring the result set back in a different physical order. Can you explain from a logical model why the same rows in the second query get different IDENTITY numbers? In the relational model, they should be treated the same if all the values of all the attributes are identical.

Think about trying to do replication on two databases that differ only by an index, or by cache size or something that occasionally gives them different execution plans for the same statements. Want to try to maintain such a system?

The IDENTITY is used to mimic a magnetic tape position number.

There are better ways of creating identifiers.

The definition of a key is that it is a subset of attributes which uniquely identify that row. An IDENTITY is not an attribute, since it is part of the PHYSICAL implementation of the representation of the storage of the row. It does not exist in the real world of the model, so you can NEVER verify it!

People who have never measured the actual performance, claim that using it as a key to replace a multi-column key. Adding extra data that has to be read from a disk is far more expensive than comparing a long byte string in main storage. Do the math: your CPU runs in nanoseconds; your disk drive runs in milliseconds.

But again, how do you know which IDENTITY value matchs a subset of columns? You can get a user to understand a natural, multi-column key
(say, (x,y) cooridnates, make model and year for a car, etc.); now
explain to him why 42 is the number of a ('Ford', 'Escort', 1997).

After making IDENTITY your key, you still have to add an extra UNIQUE() constraint for the mutliple column key. More storage.

What makes this argument even more foolish is that many OO programmer use a GUID which is MUCH longer than the multi-column key they seek to replace! r Received on Sun May 11 2003 - 05:30:20 CEST

Original text of this message