the relational model our mathematical foundation
TRANSCRIPT
The Relational Model
Our Mathematical Foundation
Origins
• First proposed by E.F. Codd, 1969-70• subsequently modified and extended
• An abstract theory of data• based on aspects of maths• set theory
• Basis of most modern DBMS• none implement it entirely• we can compare them with the idea
3 Aspects of the Model
• It concerns• 1) data objects
• storing it
• 2) data integrity• making sure it corresponds to reality
• 3) data manipulation• working with it
Relational Data Objects
Storing information
Tables – or is it?
• We say that databases have tables• data are stored in them
• This is a simplification• helps user understanding• may be misleading
Two sets of nomenclatureModel RDBMS (Inituitive)
Relation Table
Tuple Row (record)
Attribute Column
Primary key Column(s) uniquelyidentifying a row
Domain A pool of possible valuesfor an attribute
Domain
• Pool of possible values for an attribute• Each tuple has one of these values for
the attribute• Allows meaningful comparisons• They are data types
• traditionally supported poorly in most systems
• area of recent development
Relation
• Based on a collection of domains• Heading
• a set of attribute:domain pairs, such that each attribute Ai has its own domain Di.
• {<A1:D1>, <A2:D2>, …, <Ai:Di>, … <An:Dn>}
• Body• a set of tuples• each tuple is a set of name:value pairs
• {<A1:V1>, <A2:V2>, …, <Ai:Vi>, … <An:Vn>}
Illustration of notation
UID Tel
nwh 2434
hoh 2436
xxh 7659
As a table
As a relation
Heading {<UID:String(3)>, <Tel:Digit(4)>}
Body {
{<UID:xxh>, <Tel:7659>},
{<Tel:2436>, <UID:hoh>},
{<UID:nwh>, <Tel:2434>}
}
Venn diagram notation
<Ext:Digit(4)>
<UID:String(3)>
{<UID:xxh>, <Tel:7659>}
{<Tel:2436>, <UID:hoh>}
{<UID:nwh>, <Tel:2434>}
Are Relations Tables?
• A table is a practical way to write down a relation
• Relations are defined on sets• a set of attributes in the heading • a set of tuples in the body• sets have no ordering• attributes come in no particular sequence
• but columns do have sequence• tuples come in no particular sequence
• but rows do have sequence
Properties of relations
• Tuples and attributes are unordered• There are no duplicate tuples• All attributes are atomic
Relations/attributes are unordered
• Tables seems to be• Don’t believe it!
• We do not work in terms of:• “next row”• “previous record”• “first tuple”
• We do not rely on:• “next attribute”• “first column”
• Do not try to write for loops
Output can be made to be ordered
No duplicate tuples (= rows)
• i.e..• No two tuples in a relation have all the
same values for corresponding attributes
• Crucial point• Can seem like a weakness
• It is a strength• Learn to exploit it
Example of duplicationName Age Employee ID
Jones 42 3895
Smith 27 6830
Jones 42 3895
Williams 62 4692
• We do not want two records for Jones• The DBMS will prevent this silly
duplication• A simple example of exploiting “no
duplication”
All attribute values are atomic
Jackie Chan Acting, Filming, Computing
Bob Dylan Singing, Dancing
Jackie Chan Acting
Jackie Chan Filming
Jackie Chan Computing
Bob Dylan Singing
Bob Dylan Dancing
Atomic Values
• Strings have characters, incl. spaces• not the problem
• Problem: Jim Monday Squash
Wednesday Cinema
Saturday Chess
Mary Wednesday Cinema
Friday Tennis
• Bad solution
Person Day1 Act1 Day2 Act2 Day3 Act3
Jim Mon Squash Wed Cinema Sat Chess
Mary Wed Cinema Fri Tennis
• Which is “day 1”How do we search this?Could get the day and activity separatedThink about what we are modelling
Name Day Activity
Jim Monday Squash
Jim Wednesday Cinema
Jim Saturday Chess
Mary Wednesday Cinema
Mary Friday Tennis
• Solution:
• Looks like adding more lines• Not complicated therefore• Can seem “cosmetic”• Will return to this
Relational Data Objects
End
Relational Data Integrity
making sure the data corresponds to reality
A database as a model
• A DB “models” a real-world enterprise• i.e. the DB must abstract from reality
• the attribute values and their combination must reflect the true state of the world
• We try to enforce plausibility• We do this by implementing integrity
rules (constraints)
Data-specific integrity rules
• Domain specific• Employee age is between 20 and 70• We only sell CDs in multiples of 10• Car registrations must be of the form: A DDD AAA
• where A = alphabetic, D = digit• A temperature cannot be lower than 273.15 deg C
• Inter-attribute (inter-relational)• Only senior managers and sales reps can have
cars over 2000cc
• “Cardinality” constraints• A team has 11 players
General Integrity Rules
• There are two general integrity features Part of the relational model
1. Entity integrity - tuple identification through candidate/primary keys
2. Referential integrity – foreign keys
• There may be application-specific rules• must be identified and implemented• may be able to use DBMS• we will return to this
Candidate Keys (CKs)• A candidate key for relation R is a subset K of
the attributes of R such that• no two tuples of R have the same value of K
the “uniqueness property”• no proper subset of K has the uniqueness property
the “irreducibility property”• All relations have at least 1 (everything)
• may have several
• Uniqueness applies to all possible tuples• not just the current ones
What are candidate keys for?
• Tuple-level addressing• allows unambiguous identification of 1
tuple• content addressing
• The tuple where X has value Y• not unique unless X is a candidate key
• Access mechanisms are more general• e.g. indexes• although CKs may be implemented using
them
The primary key (PK)
• A specified candidate key• The choice of PK is arbitrary
• There may only be one candidate
• Practical factors may help decide• It is common to think only of “the” key
(i.e. the primary key)• but remember there may be other
candidates
Why bother with CKs ?Suppose that we have attributes: uid, ucas, national heath,
surname, first name, d.o.b., city. Here are some data:
Let us declare UCAS to be the primary key. This constrains
UCAS numbers to be distinct for different people, but permits :
UID UCAS NH Surname FirstNameDOB Cityxxh8 987632 234179 Kerr Kevin 26.07.82 Borthjjo1 003872 139873 Jones Jane 13.12,81 Lampeterjja1 013873 243179 Jones Jane 10.06.81 Llanonllo2 003874 264902 Lamb Louis 14.07.82 Borth
UID UCAS NH Surname FirstNameDOB Cityxxh8 987632 243179 Kerr Kevin 26.07.82 Borthjja1 003872 139873 Jones Jane 13.12,81 Lampeterjja1 013873 243179 Jones Jane 10.06.81 Llanonllo2 003874 264902 Lamb Louis 14.07.82 Borth
!
Penalty of not declaring CKs
• A real-world model would not want repetitions of UID and NH for distinct persons.
• We want UID to be unique for each person. Similarly for NH.
• Remedy: declare UID to be a candidate key. Similarly for NH.
• Then, either of UID, NH gives unique identification. This captures two constraints.
• Continue to use UCAS as the PK.• UCAS, UID, NH are 3 candidate keys.
Foreign Keys
• A reference mechanism between relations• The target of a reference must exist in the
referenced relation
no “dangling references”
Example,Consider a table of employees and a table of
car allocations:
Referential Integrity
Name(PK) Post …
Jones M.D.Smith AccountantDavies Porter
Car(PK) Allocated …
R 345 XYZ JonesP 987 CBA DaviesM 567 GHJ Brown
No car- OK
Violates referential integrity
Foreign Keys
Interactive Fiction
Foreign Keys (FKs)- definition
• Linking relation R2 to relation R1:• A foreign key in a relation R2 is a
subset of its attributes such that:• there is a relation R1 with a candidate key
CK• For each value of FK in R2 there must exist
an identical value in the CK in some tuple of R1.
Foreign keys - notes
• All keys are sets of attributes• A candidate key can contain a value not
currently found in the foreign key• Chains of references can build up• Relations can reference themselves
• personnel relation can have a “manager” attribute - managers are personnel
Foreign key - examples
• Earlier example:PERSON{Name(PK), Post}CAR_ALLOC{Car(PK), Allocated*}
FK {Allocated} references PERSON
• Self referential example:a surgeon is supervised by a senior surgeon, called a “consultant”)SURGEON {Surgeon(PK), Consultant*}
FK {Consultant} references SURGEON
Notation
Referential Integrity
• The database must not hold any unmatched foreign keys
• The DBMS should prevent the situation arising - most do today
• The DBMS can:• reject operation which would compromise
integrityor
• make other changes to retain integrity
Maintaining R. Integrity
• Attempt to delete the target of a foreign key
• only allow if there is no matching FK value
or• cascade-delete
tuples with FK matches
• Attempt to update the candidate key
• only allow if there is no matching FK value
or• cascade-update the
FK in the matching tuples
Interactive Fiction
Beware Autogenerated Keys!
Some systems readily offer to generate key values for you – e.g. every time another tuple is entered, a numeric key value is automatically allocated.
This permits:UID UCAS NH Surname FirstNameDOB City AutoKeyxxh8 987632 234179 Kerr Kevin 26.07.82 Borth 1000jjo1 003872 139873 Jones Jane 13.12,81 Lampeter 1001jja1 013873 243179 Jones Jane 10.06.81 Llanon 1002xxh8 987632 234179 Kerr Kevin 26.07.82 Borth 1003
- the same information inserted twice with distinct autokey values!
!
Beware Autogenerated Keys!
• Remedy: make Autokey, UID, UCAS, NH candidate keys and select one as PK.
• Remedy: avoid introducing unnecessary new keys. Do we need an AutoKey in this example?
• Caution: if you do introduce a new key, you still need to identify other candidate keys (or risk bad modelling).
Multiple Multi-attribute keys
• Consider a timetable, with entries for Day, Time, Module, Room, Building, LecturerID
• Assume that a Room label such as A6 can appear in different buildings.
• Some data:
Day Time Module Room Building LecturerMonday 09.00 CI25910 C22 Hugh Owen rrpMonday 09.00 CS23710 A6 Hugh Owen dapFriday 09.00 CS27310 A6 Llandinam hohMonday 17.00 CS25610 C22 Hugh Owen dap
Candidate Keys? There might be several.
Multiple Multi-attribute Keys
• With business rules: e.g.• a module has only one lecturer,
a lecturer lectures in 1 room at a time, only 1 module in a room at a time ...
• 3 Candidate keys in this case: 1. {Day,Time, Room, Building}, – 4 attributes
2. {Day, Time, Lecturer} – 3 attributes
3. {Day, Time, Module}, – 3 attributes
• Choose one to be the PK.
Multi-attribute Foreign Keys
• A multi-attribute PK may be referenced from another relation
• the referencing foreign key then needs to be declared with same structure as that PK (or generally, the CK)
• {Room, Building} above could be declared to be a FK that references a relation with PK {Room, Building} describing facilities:
Room Building BoardType Projection Facilities LightingA6 Hugh Owen None Yes FixedC22 Hugh Owen Fixed Yes DimmingA6 Llandinam Roller Yes Dimming
Data Integrity
End