Constraint naming standard is important for one reason: The SYS_* name oracle assigns to unnamed constraints is not very understandable. By correctly naming all contraints, we can quickly associate a particular constraint with our data model. This gives us two real advantages:
We can quickly identify and fix any errors.
We can reliabily modify or drop constraints
Oracle limits names, in general, to 30 characters, which is hardly enough for a human readable constraint name.
We propose the following naming convention for all constraints, with the following abbreviations taken from Oracle Docs at http://oradoc.photo.net/ora81/DOC/server.815/a67779/ch4e.htm#8953. Note that we shortened all of the constraint abbrevations to two characters to save room.
|references (foreign key)||fk|
<table name>_<column_name>_<constraint abbreviation>
In reality, this won't be possible because of the character limitation on names inside oracle. When the name is too long, we will follow these two steps in order:
Abbreviate the table name with the table's initials (e.g. users -> u and users_contact -> uc).
Truncate the column name until it fits.
If the constraint name is still too long, you should consider rewriting your entire data model :)
If you have to abbreviate the table name for one of the constraints, abbreviate it for all the constraints
If you are defining a multi column constraint, try to truncate the two column names evenly
create table example_topics ( topic_id integer constraint example_topics_topic_id_pk primary key ); create table constraint_naming_example ( example_id integer constraint cne_example_id_pk primary key, one_line_description varchar(100) constraint cne_one_line_desc_nn not null, body clob, up_to_date_p char(1) default('t') constraint cne_up_to_date_p_check check(up_to_date_p in ('t','f')), topic_id constraint cne_topic_id_nn not null constraint cne_topic_id_fk references example_topics, -- Define table level constraint constraint cne_example_id_one_line_unq unique(example_id, one_line_description) );
Naming primary keys might not have any obvious advantages. However, here's an example where naming the primary key really helps (and this is by no means a rare case!
SQL> set autotrace traceonly explain; SQL> select * from constraint_naming_example, example_topics where constraint_naming_example.topic_id = example_topics.topic_id; Execution Plan ---------------------------------------------------------- 0 SELECT STATEMENT Optimizer=CHOOSE 1 0 NESTED LOOPS 2 1 TABLE ACCESS (FULL) OF 'CONSTRAINT_NAMING_EXAMPLE' 3 1 INDEX (UNIQUE SCAN) OF 'EXAMPLE_TOPICS_TOPIC_ID_PK' (UNI QUE)
Isn't it nice to see "EXAMPLE_TOPICS_TOPIC_ID_PK" in the trace and know exactly which table oracle is using at each step?
ArsDigita is split on whether or not we should be naming not null constraints... So, if you want to name them, please do so and follow the above naming standard. But, naming not null constraints is not a requirement at ArsDigita.
Though naming "not null" constraints doesn't help immeditately in error debugging (e.g. the error will say something like "Cannot insert null value into column"), we recommend naming not null constraints to be consistent in our naming of all constraints.