View 1. Lu Chaojun, SJTU 2 View Three-level vision of DB users Virtual DB views DB Designer Logical DB relations DBA DBA Physical DB stored info.

View

1

Lu Chaojun, SJTU 2

View

• Three-level vision of DB

users Virtual DB views

DB Designer Logical DB relations

DBA

DBA Physical DB stored info

Lu Chaojun, SJTU 3

What’s a View?

• An expression that defines a table without physically creating it.

• Base table vs. view– Base table: physically stored, persistent– (Virtual )View: not physically stored, defined

on base tables.

• Query and even modify views like tables

Lu Chaojun, SJTU 4

Def. in SQL:2003

• SQL-data consists entirely of table variables, called base tables.

• An operation that references zero or more base tables and returns a table is called a query.

• The result of a query is called a derived table.• A view is a named query, which can be invoked by

use of this name. The result of such an invocation is called a viewed table.

Lu Chaojun, SJTU 5

Why Views?

• To hide some data from the users

• To make certain queries easier or more natural to express

• For modularity

Lu Chaojun, SJTU 6

View Declaration

CREATE VIEW viewname(A1,...,An)

AS query;

• ExampleCREATE VIEW CS_S AS

SELECT sno,name,age

FROM Student

WHERE dept = ‘CS’;

• Only definition of views are stored.

Lu Chaojun, SJTU 7

Querying Views

• Query a view as if it were a stored table.

• Queries on views are translated into queries on base tables.– Simplest: replace view by its definition (a subquery)

• ExampleSELECT name FROM CS_S

WHERE age = 18;

SELECT name FROM (SELECT sno,name,age FROM Student

WHERE dept = ‘CS’) css

WHERE age = 18;

– Q: Two consecutive queries on CS_S always return the same result?

Lu Chaojun, SJTU 8

Example: Join View

• Creation of view SN_CCREATE VIEW SN_C AS

SELECT name,cno FROM S,SC

WHERE S.sno = SC.sno;

• Query on SN_CSELECT name FROM SN_C

WHERE cno = ‘C2’;– no join?

Lu Chaojun, SJTU 9

Example: Renaming Attributes

1. CREATE VIEW CS_S(id,sn,sa) AS SELECT sno, name, age

FROM Student

WHERE dept = ‘CS’;

2. CREATE VIEW S_G(sno,avggrade) AS SELECT sno, AVG(grade)

FROM SC

GROUP BY sno;

Lu Chaojun, SJTU 10

Example: Views on Views

CREATE VIEW CS_S_20 AS

SELECT sno,name

FROM CS_S

WHERE age=20;

Lu Chaojun, SJTU 11

Modifying Views

• Updatable view: it’s possible to translate the modification of the view into an equivalent modification on a base table.– Complex rules in SQL standard

• Example: S_G is not an updatable view.INSERT INTO S_G

VALUES (‘007’, 80);– Causes what tuple to be inserted into SC?

Updatable Views

• FROM only one relation R ( which may itself be an updatable view)

• WHERE must not involve R in a subquery

• SELECT (not SELECT DISTINCT) some attributes– Enough for insertion into the underlying R

Lu Chaojun, SJTU 12

Example

• CS_S is updatable, but has anomaly:INSERT INTO CS_S

VALUES (‘007’,’james’,18);– causes (‘007’,’james’,18,NULL) to be inserted

into Student, which can’t be restored into CS_S– Attributes not enough

Lu Chaojun, SJTU 13

Lu Chaojun, SJTU 14

Drop Views

DROP VIEW viewname;

• Definition of view is dropped.

• Underlying base tables are not affected.

• DROP TABLE will affect views defined on it.

Instead-Of Triggers on Views

• Actions of trigger is done instead of the event (modification on a view) itself.

• The programmer can force whatever interpretation of a view modification is desired.

Lu Chaojun, SJTU 15

Example

CREATE TRIGGER insertCSS

INSTEAD OF INSERT ON CS_S

REFERENCING NEW ROW AS nr

FOR EACH ROW

INSERT INTO Student

VALUES(nr.sno,nr.name,nr.age,’CS’);

Lu Chaojun, SJTU 16

Materialized Views

• Views are logical descriptions of relations that are constructed from base tables by executing a query.– Querying a view = executing the query

• If a view is used frequently enough, it may be efficient to materialize it.CREATE MATERIALIZED VIEW CS_S

AS SELECT sno,name,age FROM S WHERE dept=‘CS’;

Lu Chaojun, SJTU 17

Maintaining a MV

• Problem: each time a base table changes, the materialized view may change.– Cannot afford to recompute the view with each change.

– Incremental maintainence:Base table R Materialized view V

R V

Example: S vs CS_S

insert into S insert into CS_S

values(a,b,c,’CS’) values(a,b,c)

Lu Chaojun, SJTU 18

Periodic Maintainence of MV

• Data warehouse: create MV (of aggregations), but not to keep it up-to-date.– Frequent modification: expensive– Used for analysis: unnecessary

• Periodic maintainence is OK, say each night.

Lu Chaojun, SJTU 19

Example

• Wal-Mart stores every sale at every store in a database.

• Overnight, the sales for the day are used to update the data warehouse = materialized views of the sales.

• The warehouse is used by analysts to predict trends and move goods to where they are selling best.

Lu Chaojun, SJTU 20

Rewriting Queries to Use MV

Given MV V:

SELECT LV FROM RV WHERE CV;

and query Q:

SELECT LQ FROM RQ WHERE CQ;

If i) RV RQ

ii) CQ = CV AND C

iii) (LQ RV) LV

Then rewrite Q:

SELECT LQ FROM V, RQ RV WHERE C;

Lu Chaojun, SJTU 21

Indexes in SQL

22

What’s an Index?

• An index on R.A is a data structure that makes it efficient to find tuples having a fixed value for A.– A is the index key.– Index key can be any attribute or set of

attributes, not necessarily the key of R.

• Implementation of index– B-tree or B+-tree

Lu Chaojun, SJTU 23

key ptr

a

a b 23

Lu Chaojun, SJTU 24

Why Index?

• Speed up queries, especially of formsR.A = value

R.A <= value

• Also speed up joins– Use index to find matching tuples

• Make queries faster, but modifications slower.

Declaring Indexes

• No standard!

• Typical syntax:CREATE INDEX nameInd ON S(name);

CREATE INDEX scInd ON SC(sno,cno);

• Dropping a index:DROP INDEX nameInd;

Lu Chaojun, SJTU 25

Selection of Indexes

• Design Issue: deciding which indexes to create.– Cost model: number of block I/O’s

• Pro: An index speeds up queries that can use it.

• Con: An index slows down modifications because the index must be modified too.

Lu Chaojun, SJTU 26

Useful Indexes

• Index on key of relation: e.g. S(sno)– Queries are often of the form:

SELECT … FROM … WHERE sno = …

– At most return one location (page)

• Index on attribute which is almost a key: e.g. S(name)– Return few locations

• Index on attribute on which tuples are clustered: e.g. S(dept)– Few blocks

Lu Chaojun, SJTU 27

Lu Chaojun, SJTU 28

Calculating the Best Indexes

• The more indexes the better?– No! Modification will be expensive.

• To create best indexes, we need information about which queries and modifications are most likely to be performed on DB.– Get info from history.– Get info from app code.

Example: Cost Estimation

SC(sno,cno)

Q1: SELECT cno FROM SC WHERE sno=‘xxx’;

Q2: SELECT sno FROM SC WHERE cno=‘yyy’;

I: INSERT INTO SC VALUES(‘xxx’,’yyy’);

Prob(Q1) = p1, Prob(Q2) = p2 , Prob(I ) = p3 =1 p1 p2

SC occupies n blocks.

On the average, ‘xxx’ takes c courses and ‘yyy’ is taken by s students.

Lu Chaojun, SJTU 29

action no index index on sno index on cno both index

Q1 n 1+c n 1+c

Q2 n n 1+s 1+s

I 2 4 4 6

Average np1+np2+2p3 (1+c)p1+np2+4p3 np1+(1+s)p2+4p3 (1+c)p1+(1+s)p2+6p3

Automatic Selection of Indexes

• Tuning advisor– Establish a query workload.

According to history or application programs.

– Specify constraints, if any.– Generates candidate indexes and evaluates

each.Feed sample query to the query optimizer, which

assumes only this one index is available.

– The index set resulting the lowest cost is suggested, or automatically created.

Lu Chaojun, SJTU 30

End