两个唯一列的非唯一组合

Question

I have table with columns code, id and date and some other date. These columns are primary key together.

primary key(id, code, date)

I want them to be unique together. But a code cannot be duplicated in table with two different ids. For example this is a possible situation:

id   code  date        data
1    123   3/28/2018   "data 1"
1    123   4/28/2018   "data 2"

or even this

1    123   3/28/2018   "data 1"
1    124   3/28/2018   "data 2"

while this is not:

id   code  date        data
1    123   3/28/2018   "data 1"
2    123   4/28/2018   "data 2"

it's possible to have two rows with code "123" only if id of both rows is 1. "123" cannot be combined with any other ids.

How can I manage this in sql server (version 14)?

Answer 1

You have the wrong data model.

You need one table with only id and code (and whatever columns other than date might be relevant. Let's call this table codes and the id, code_id , because that is how I would name them:

create table codes (
    code_id int identity(1, 1) primary key,
    code varchar(255) not null unique
);

There. Now the mapping between code_id and code is one-to-one. Next, create a table using only code_id and date :

create table code_dates (
    code_date_id int identity(1, 1) primary key,
    code_id int,
    date date,
    contraint fk_code_dates_code_id foreign key (code_id) references codes(code_id)
);

You can look up the code using code_id . Voila! Your data model is fixed.

Answer 2

Gordon's answer is the best advice, but treating your question as academic, you can manage what you want using a CHECK CONSTRAINT in addition to the PRIMARY KEY you already have.

The CHECK CONSTRAINT would call a function that takes id and code as parameters. The function queries the table and returns true if there is no other id that has that code , or false otherwise. The CONSTRAINT allows only rows where the function returns true .

Answer 3

This might do it

create table tmp (id int not null,  code int not null, dt datetime not null,  data varchar(10)
                   primary key (id, code, dt));
create function dbo.CheckFunction(@id int, @code int)
returns int
as begin
    return (select count(*) from tmp t where t.id <> @id and t.code = @code) 
end;
alter table tmp
add constraint chk_CheckFunction
check (dbo.CheckFunction(id, code) = 0);

truncate table tmp;
insert into tmp values 
       (1 ,123, '3/28/2018', 'data 1')
     , (1 ,123, '4/28/2018', 'data 2')
     , (1 ,123, '5/28/2018', 'data 2');
select * from tmp;
select count(*) from tmp t where t.id <> 1 and t.code = 123;
insert into tmp values 
       (2 ,123, '3/28/2018', 'data 1');
select * from tmp;

drop table tmp;
drop function dbo.CheckFunction;

Answer 4

Why don't you create a Lookup table whose primary key is id , code ? By putting a UNIQUE constraint on code , you will ensure that code = 123 will only every pair up with id = 1 . We cannot put code = 123 against id = 2 without violating the constraint.

Then put a referential constraint between the Lookup table and your Business table. Then, the referential constraint will bring over the constraints from the Lookup table into the Business table which will persist that code = 123 can only every match with id = 1 .

Something like:

DROP TABLE Business;
DROP TABLE Lookup;

CREATE TABLE Lookup
(
    id INT NOT NULL,
    code INT NOT NULL UNIQUE,
    CONSTRAINT PK_Lookup Primary Key (id, code)
);

INSERT INTO Lookup VALUES (1, 123);
INSERT INTO Lookup VALUES (1, 124);
INSERT INTO Lookup VALUES (2, 123); -- Rejected

CREATE TABLE Business
(
  id int NOT NULL,
  code int NOT NULL,
  date date NOT NULL,
  data NVARCHAR(20) NOT NULL,
  CONSTRAINT PK_Business PRIMARY KEY (id, code, date),
  CONSTRAINT FK_Lookup FOREIGN KEY (id, code) REFERENCES Lookup (id, code)
);

INSERT INTO Business VALUES (1, 123, '2018-03-28', 'data 1');
INSERT INTO Business VALUES (1, 123, '2018-04-28', 'data 2');
INSERT INTO Business VALUES (1, 124, '2018-03-28', 'data 2');
INSERT INTO Business VALUES (2, 123, '2018-04-28', 'data 2'); -- Rejected