Constraint Design

DEFAULT and NULL Constraints

• [Proposal] If all the column values can be obtained from services, you are not advised to use the DEFAULT constraint. Otherwise, unexpected results will be generated during data loading.
• [Proposal] Add NOT NULL constraints to columns that never have NULL values. The optimizer automatically optimizes the columns in certain scenarios.
• [Proposal] Explicitly name all constraints excluding NOT NULL and DEFAULT.

Partial Cluster Keys

A partial cluster key (PCK) is a local clustering technology used for column-store tables. After creating a PCK, you can quickly filter and scan fact tables using min or max sparse indexes in GaussDB. Comply with the following rules to create a PCK:

• [Notice] Only one PCK can be created in a table. A PCK can contain multiple columns, preferably no more than two columns.
• [Proposal] Create a PCK on simple expression filter conditions in a query. Such filter conditions are usually in the form of col op const, where col specifies a column name, op specifies an operator (such as =, >, >=, <=, and <), and const specifies a constant.
• [Proposal] If the preceding conditions are met, create a PCK on the column having the most distinct values.

Unique Constraints

• [Notice] Both row-store and column-store tables support unique constraints.
• [Proposal] The constraint name should indicate that it is a unique constraint, for example, UNIIncluded columns.

Primary Key Constraints

• [Notice] Both row-store and column-store tables support primary key constraints.
• [Proposal] The constraint name should indicate that it is a primary key constraint, for example, PKIncluded columns.

Check Constraints

• [Notice] Check constraints can be used in row-store tables but not in column-store tables.
• [Proposal] The constraint name should indicate that it is a check constraint, for example, CKIncluded columns.