Foreign keys in composite key - sql

Suppose I have a table, Document, that looks something like this:
_______________________
| Document |
|-----------------------|
| DocumentId int PK |
| (add't fields) |
|_______________________|
Now suppose I have a second table:
_______________________
| DocumentVersion |
|-----------------------|
| DocumentId int PK, FK |
| VersionId int PK |
| (add't fields) |
|_______________________|
Finally, suppose I wish to create a third table that references DocumentVersion, perhaps an audit of users that have accessed each version (assume a User table exists):
_______________________
| VersionAccessLog |
|-----------------------|
| DocumentId int PK, FK |
| VersionId int PK, FK |
| UserId int PK, FK |
| AccessTime DateTime PK|
|_______________________|
Maybe not the best example but hopefully enough to illustrate my question.
Focusing on VersionAccessLog, we have:
PK(DocumentId, VersionId, UserId, AccessTime)
FK(DocumentId, VersionId) REFERENCES DocumentVersion
FK(userId) REFERENCES User
Now my question is, should I also in VersionAccessLog create a FK(DocumentId)? At first glance, the key seems superfluous-- referential integrity is enforced by the FK to DocumentVersion. However, from a relational algebra standpoint, should this key exist or not? From a practical standpoint (assume SQL Server 2012 if necessary), are there any performance implications of including or excluding the key?

There are performance impacts to including the additional key, because the database will check the FK on every insert (child) or delete(parent). From a practical perspective, I generally do not include the extra keys because there really isn't any value in the validations and there is the slight performance cost.
I would make sure that if the schema was ever changed and the FK between DocumentVersion and VersionAccessLog was removed, to add the FK between VersionAccessLog and Document. The only other time I could see adding it is if some Schema aware DAO library was used that reverses Has-a relationships from FK's, and it was worthwhile to include that.

Related

Primary key of a simple 1:1-mapping table with NULL values?

This feels like a very basic question, but I really don't see the obvious answer at the moment.
I have a simple table that maps object ids between two namespaces:
|---------------------|------------------|
| id_in_ns1 | id_in_ns2 |
|---------------------|------------------|
| 1 | 5 |
|---------------------|------------------|
| 2 | 17 |
|---------------------|------------------|
| 3 | NULL |
|---------------------|------------------|
| NULL | 1 |
|---------------------|------------------|
The mapping is basically 1:1, but as you can see, some objects from namespace 1 do not exist in namespace 2, and vice versa, so that there are NULL values in the table.
So, what would be the primary key of this table? As a PK cannot be NULL, I can neither use (id_in_ns1) nor (id_in_ns2) nor the composite.
The only idea I have is to replace NULL by a definite value, say -1, and to use (id_in_ns1, id_in_ns2)as PK. However, this feels not only hackish but also "unnormal" because the non-NULL (or non--1)) value alone is already sufficient to uniquely identify an object.
Only add entries that have a valid id on both sides. This will effectively get rid of all NULL values, allowing you to specify a proper composite key on (id_in_ns1, id_in_ns2).
Ultimately, those are the values that allow you to identify a single row and you will not lose relevant information - a SELECT id_in_ns2 FROM mapping_table WHERE id_in_ns1 = x will return NULL either way, whether there is a (x, NULL) row or not.
If you insist on keeping those NULLs you could add another column with an artificial (auto incrementing) primary key, but that feels as hacky as using -1.
Use a synthetic primary key and use unique constraints for the rest:
create table mapping (
mappingId int auto_increment primary key, -- or whatever for your database
id_in_ns1 int references ns1(id),
id_in_ns2 int references ns2(id),
unique (id_in_ns1),
unique (id_in_ns2)
);
Just one caveat: some databases only allow one NULL value for UNIQUE constraints. You might need to use a filtered unique index instead (or some other mechanism) for this construct.

Validate whether value in one table is the same as in related table - performance

Let's say I have two tables and I'm doing all the operations in .NET Core 2 Web API.
Table A:
Id,
SomeValue,
TeamName
Table B:
Id,
Fk_Id_a (references Id in table A),
OtherValue,
TeamName
I can add and get records from table B indepedently.
But for every record in Table B TeamName has to be the same as for it's corresponidng Fk_Id_a in Table A.
Assume these values comes in:
{
"Fk_Id_a": 3,
"SomeValue": "test val",
"TeamName": "Super team"
}
Which way would be better to check it in terms of performance? 1ST way requires two connections, when 2nd requires storing some extra keys etc.
1ST WAY:
get record from Table A for Fk_Id_a (3),
check if TeamName is the same as in coming request (Super team),
do the rest of the logic
2ND WAY:
using compound foreign keys and indexes:
TableA has alternate unique key (Id, TeamName)
TableB has foreign compound key (Fk_Id_a, TeamName) that references TableA (Id, TeamName)
SQL SCRIPT TO SHOW:
ALTER TABLE Observation
ADD UNIQUE (Id, PowelTeamId)
GO
ALTER TABLE ObservationPicturesId
ADD FOREIGN KEY(ObservationId, PowelTeamId)
REFERENCES Observation(Id, PowelTeamId)
ON DELETE CASCADE
ON UPDATE CASCADE
EDIT: Simple example how the tables might look like. TeamName has to be valid for FK referenced value in Table A.
Table A
ID | ObservationTitle | TeamName
---------------------------------------
1 | Fire damage | CX_team
2 | Water damage | CX_team
3 | Wind damage | Dd_WP3
Table B
ID | PictureId | AddedBy | TeamName | TableA_ID_FK
-----------------------------------------------------
1 | Fire | James | CX_team | 1
2 | Water | Andrew | CX_team | 1
3 | Wind | John | Dd_WP3 | 3
Performance wise, the 2nd option would be faster because there is no comparison to check (the foreign key will force that they match when inserting, updating or deleting) when selecting the rows from the table. It would also make a unique index on table A.
That being said, there is something very fishy about the structure you mention. First of all why is the TeamName repeated in table B? If a row in table B is "valid" only when the TeamName match, then you should enforce that no row should be inserted with a different TeamName, throught the ID foreign key (and not actually storing the TeamName value). If there are records on table B that represent another thing rather than the entity that is linked to table A then you should split it onto another table or just update the foreign key column when the team matches and not always.
The issue is that you are using a foreign key as a partial link, making the relationship valid only when an additional condition is true.

save() cannot change primary key, it's a bug?

I wanna change the primary key in one table, but phalcon seems not offer any method to change the primary key.
uid is primary key in this table
+----+------------+------------+------+
| uid | name | type | year |
+----+------------+------------+------+
| 1 | Robotina | mechanical | 1972 |
| 2 | Astro Boy | mechanical | 1952 |
| 3 | Terminator | cyborg | 2029 |
+----+------------+------------+------+
$tmp = User::findFirst('uid = 100');
$tmp->uid = 900;
$tmp->rate = 15;
$result = $tmp->save();
these codes cannot modify anything in the table , but it return the $result is true, is it a bug?
I think it should offer a method to modify primary key.
I don't think it is a bug, I suspect that the Phalcon developers would subscribe to the principle of having primary keys as immutable or set in stone. See questions like : Can we update primary key values of a table? and https://softwareengineering.stackexchange.com/questions/8187/should-a-primary-key-be-immutable/
If anything the bug could be that Phalcon allows you to change the primary key in the first place.
Knowing why you would want to change the primary key might help, because I can't think of any reason off the top of my head why you would want to do that. I believe the main (only?) purpose of the primary key is to identify that row in the table, and any joining tables, whether Robotina has a uid of 1, 11 or 111 shouldn't really matter as long as the entry is unique for all rows in the table.
If the purpose of change is some kind of ranking like a leader board, I would suggest adding a rank field, as it will likely be more efficient that changing a special primary key field.
If you absolutely have to change the uid, I would be inclined to create a new row and delete the old one, if that sounds messy bear in mind that depending on which database you are using that might be what the database engine is doing behind the scenes anyway, and also that the very concept of editing a primary key would set a lot of people's teeth on edge.

Can a foreign key refer to a primary key in the same table?

I just think that the answer is false because the foreign key doesn't have uniqueness property.
But some people said that it can be in case of self joining the table.
I am new to SQL. If its true please explain how and why?
Employee table
| e_id | e_name | e_sala | d_id |
|---- |------- |----- |--------|
| 1 | Tom | 50K | A |
| 2 | Billy | 15K | A |
| 3 | Bucky | 15K | B |
department table
| d_id | d_name |
|---- |------- |
| A | XXX |
| B | YYY |
Now, d_id is foreign key so how it can be a primary key. And explain something about join. What is its use?
I think the question is a bit confusing.
If you mean "can foreign key 'refer' to a primary key in the same table?", the answer is a firm yes as some replied. For example, in an employee table, a row for an employee may have a column for storing manager's employee number where the manager is also an employee and hence will have a row in the table like a row of any other employee.
If you mean "can column(or set of columns) be a primary key as well as a foreign key in the same table?", the answer, in my view, is a no; it seems meaningless. However, the following definition succeeds in SQL Server!
create table t1(c1 int not null primary key foreign key references t1(c1))
But I think it is meaningless to have such a constraint unless somebody comes up with a practical example.
AmanS, in your example d_id in no circumstance can be a primary key in Employee table. A table can have only one primary key. I hope this clears your doubt. d_id is/can be a primary key only in department table.
This may be a good explanation example
CREATE TABLE employees (
id INTEGER NOT NULL PRIMARY KEY,
managerId INTEGER REFERENCES employees(id),
name VARCHAR(30) NOT NULL
);
INSERT INTO employees(id, managerId, name) VALUES(1, NULL, 'John');
INSERT INTO employees(id, managerId, name) VALUES(2, 1, 'Mike');
-- Explanation:
-- In this example.
-- John is Mike's manager. Mike does not manage anyone.
-- Mike is the only employee who does not manage anyone.
Sure, why not? Let's say you have a Person table, with id, name, age, and parent_id, where parent_id is a foreign key to the same table. You wouldn't need to normalize the Person table to Parent and Child tables, that would be overkill.
Person
| id | name | age | parent_id |
|----|-------|-----|-----------|
| 1 | Tom | 50 | null |
| 2 | Billy | 15 | 1 |
Something like this.
I suppose to maintain consistency, there would need to be at least 1 null value for parent_id, though. The one "alpha male" row.
EDIT: As the comments show, Sam found a good reason not to do this. It seems that in MySQL when you attempt to make edits to the primary key, even if you specify CASCADE ON UPDATE it won’t propagate the edit properly. Although primary keys are (usually) off-limits to editing in production, it is nevertheless a limitation not to be ignored. Thus I change my answer to:- you should probably avoid this practice unless you have pretty tight control over the production system (and can guarantee no one will implement a control that edits the PKs). I haven't tested it outside of MySQL.
Eg: n sub-category level for categories .Below table primary-key id is referred by foreign-key sub_category_id
A good example of using ids of other rows in the same table as foreign keys is nested lists.
Deleting a row that has children (i.e., rows, which refer to parent's id), which also have children (i.e., referencing ids of children) will delete a cascade of rows.
This will save a lot of pain (and a lot of code of what to do with orphans - i.e., rows, that refer to non-existing ids).
Other answers have given clear enough examples of a record referencing another record in the same table.
There are even valid use cases for a record referencing itself in the same table. For example, a point of sale system accepting many tenders may need to know which tender to use for change when the payment is not the exact value of the sale. For many tenders that's the same tender, for others that's domestic cash, for yet other tenders, no form of change is allowed.
All this can be pretty elegantly represented with a single tender attribute which is a foreign key referencing the primary key of the same table, and whose values sometimes match the respective primary key of same record. In this example, the absence of value (also known as NULL value) might be needed to represent an unrelated meaning: this tender can only be used at its full value.
Popular relational database management systems support this use case smoothly.
Take-aways:
When inserting a record, the foreign key reference is verified to be present after the insert, rather than before the insert.
When inserting multiple records with a single statement, the order in which the records are inserted matters. The constraints are checked for each record separately.
Certain other data patterns, such as those involving circular dependences on record level going through two or more tables, cannot be purely inserted at all, or at least not with all the foreign keys enabled, and they have to be established using a combination of inserts and updates (if they are truly necessary).
Adding to the answer by #mysagar the way to do the same in MySQL is demonstrated below -
CREATE TABLE t1 (
-> c1 INT NOT NULL,
-> PRIMARY KEY (c1),
-> CONSTRAINT fk FOREIGN KEY (c1)
-> REFERENCES t1 (c1)
-> ON UPDATE RESTRICT
-> ON DELETE RESTRICT
-> );
would give error -
ERROR 1822 (HY000): Failed to add the foreign key constraint. Missing index for constraint 'fk' in the referenced table 't1'
The correct way to do it is -
CREATE TABLE t1 (
-> c1 INT NOT NULL,
-> PRIMARY KEY (c1),
-> KEY i (c1),
-> CONSTRAINT fk FOREIGN KEY (c1)
-> REFERENCES t1 (c1)
-> ON UPDATE RESTRICT
-> ON DELETE RESTRICT
-> );
One practical utility I can think of is a quick-fix to ensure that after a value is entered in the PRIMARY KEY column, it can neither be updated, nor deleted.
For example, over here let's populate table t1 -
INSERT INTO t1 (c1) VALUES
-> (1),
-> (2),
-> (3),
-> (4),
-> (5);
SELECT * FROM t1;
+----+
| c1 |
+----+
| 1 |
| 2 |
| 3 |
| 4 |
| 5 |
+----+
Now, let's try updating row1 -
UPDATE t1
-> SET c1 = 6 WHERE c1 = 1;
ERROR 1451 (23000): Cannot delete or update a parent row: a foreign key constraint fails (`constraints`.`t1`, CONSTRAINT `fk` FOREIGN KEY (`c1`) REFERENCES `t1` (`c1`) ON DELETE RESTRICT ON UPDATE RESTRICT)
Now, let's try deleting row1 -
DELETE FROM t1
-> WHERE c1 = 1;
ERROR 1451 (23000): Cannot delete or update a parent row: a foreign key constraint fails (`constraints`.`t1`, CONSTRAINT `fk` FOREIGN KEY (`c1`) REFERENCES `t1` (`c1`) ON DELETE RESTRICT ON UPDATE RESTRICT)

Specify a foreign key on a non-unique reference column by additional reference

Example:
survivor table
owner_id unique_id isDead
|__________ |__________|__________
| foreign key | |
| 1234 ------------> | 1234 | true <--- this row is referenced
|__________ |__________|__________
| |
| 1234 | false <--- this row should be referenced
|__________|__________
Currently the code is like that:
CONSTRAINT `key_2` FOREIGN KEY (`owner_id`) REFERENCES `survivor` (`unique_id`)
I would like to have it function like that:
CONSTRAINT `key_2` FOREIGN KEY (`owner_id`) REFERENCES `survivor` (`unique_id`) WHERE `isDead` = false
but thats obviously not possible like that.
Does anyone have an idea how I could achieve anything like this?
Your unique_id is not actually unique. You should create a key by two columns in your second table (I mean unique_id and isDead). - then it will be unique combination. But you still won't be able to refer that by foreign key properly.
If you need to refer that via foreign key, you should store only valid owners in your second table: for example, delete those, who are not valid. If you want to keep them, you have to update their unique_id according to some algorithm, that preserves unique values and in the same time allows to separate valid values from invalid. Besides this, you'll need to store old values of unique_id - for example, create a new column named old_id and that will be:
+-----------+--------+--------+
| unique_id | isDead | old_id |
+-----------+--------+--------+
| 1234 | false | 1234 |
+-----------+--------+--------+
| -1 | true | 1234 |
+-----------+--------+--------+
as you see, one of possible (but not best) solutions is to set unique_id to negative value. In sample above you will refer to correct row via foreign key and still able to restore connection of owner since old_id is stored. I'm not sure what you're trying to achieve, so in general that could be not an optimal way.