sqlalchemy.sql.schema.PrimaryKeyConstraint

A table-level PRIMARY KEY constraint.

The .PrimaryKeyConstraint object is present automatically on any _schema.Table object; it is assigned a set of _schema.Column objects corresponding to those marked with the :paramref:`_schema.Column.primary_key` flag:

>>> my_table = Table('mytable', metadata,
...                 Column('id', Integer, primary_key=True),
...                 Column('version_id', Integer, primary_key=True),
...                 Column('data', String(50))
...     )
>>> my_table.primary_key
PrimaryKeyConstraint(
    Column('id', Integer(), table=<mytable>,
           primary_key=True, nullable=False),
    Column('version_id', Integer(), table=<mytable>,
           primary_key=True, nullable=False)
)

The primary key of a _schema.Table can also be specified by using a .PrimaryKeyConstraint object explicitly; in this mode of usage, the "name" of the constraint can also be specified, as well as other options which may be recognized by dialects:

my_table = Table('mytable', metadata,
            Column('id', Integer),
            Column('version_id', Integer),
            Column('data', String(50)),
            PrimaryKeyConstraint('id', 'version_id',
                                 name='mytable_pk')
        )

The two styles of column-specification should generally not be mixed. An warning is emitted if the columns present in the .PrimaryKeyConstraint don't match the columns that were marked as primary_key=True, if both are present; in this case, the columns are taken strictly from the .PrimaryKeyConstraint declaration, and those columns otherwise marked as primary_key=True are ignored. This behavior is intended to be backwards compatible with previous behavior.

For the use case where specific options are to be specified on the .PrimaryKeyConstraint, but the usual style of using primary_key=True flags is still desirable, an empty .PrimaryKeyConstraint may be specified, which will take on the primary key column collection from the _schema.Table based on the flags:

my_table = Table('mytable', metadata,
            Column('id', Integer, primary_key=True),
            Column('version_id', Integer, primary_key=True),
            Column('data', String(50)),
            PrimaryKeyConstraint(name='mytable_pk',
                                 mssql_clustered=True)
        )