The ACID PropertiesAND Database Design: Chapter 10
AND Normal Forms AND Chapters 13,14,16
ACID Transactions Atomic: Either all of a transaction or
None of it affects the database Consistent: When a transaction ends, the
database obeys all constraints Isolated: Two running transactions cannot
pass values to each other, via the database or other data store
Durable: Once a transaction has “committed”, its updates are permanent
Atomicity Use a local log to store a transaction’s
partial result If a transaction does something illegal,
toss out the log
Consistent Check constraints in phase 1
Some are immediate, like domains Others don’t have to be true until the
commit point, like FKs
Isolated Transactions commit in a linear order Serializability is enforced Results become available only after
atomic commit point
Durable Database has one state and it is in
nonvolatile storage Keep checkpoints and transaction logs
Deadlock Loops of transactions wait on each other Detection: use time-outs Prevention: use “waits for” graph
The DB Design Process Start with an entity model Map to tables Create PKs and FKs Create other constraints Normalize tables
Our focus: normalization Goals
Minimize redundant data Minimize “update anomalies”
Functional and Multivalued Dependencies FD
We say that ai FD-> aj Or “ai functionally determines aj”
MVD-> We say that ai MVD-> aj Or “ai multivalued determines aj”
Note: the right side of an FD or an MVD can be a set of attributes
First 3 normal forms First (1NF) The value stored at the
intersection of each row and column must be a scalar value, and a table must not contain any repeating columns.
Second (2NF) Every non-key column must depend on the entire primary key.
Third (3NF) Every non-key column must depend only on the primary key.
NF3 fixed and NF4 Boyce-Codd (BCNF) A non-key
column can’t be dependent on another non-key column.
Fourth (4NF) A table must not have more than one multivalued dependency, where the primary key has a one-to-many relationship to non-key columns.
Example: 1NF
Example: 2NF
Example: 2NF, continued
3NF: remove transitive dependenciesCustomer ID Address ZIP18 112 First 80304 17 123 Ash 80303 16 123 Ash 80303
3NF, continuedBreak into two tables:
Customer ID AddressAddress Zip
4NF: Separate pairs of MVDsMothers_Phone Fathers_Phone Child_Name
Break into: Mothers_Phone Child_Name 3030000000 Sue 3031111111 SueAnd Fathers_Phone Child_Name 3032222222 Sue 3033333333 Sue
Note: both fields needed for PK
Tradeoffs “Decomposition” makes it harder to
misunderstand the database schema But Decomposition create narrow tables
that might not correspond to forms in the real world
And Decomposition leads to extra joins One solution is to pre-join data
Autocommit Used when manipulating a MySQL
database interactively Automatically and immediately commits
INSERT UPDATE DELETE commands Use the transaction protocol to override
this
Chapter 14: Transactions in MySQL A transaction is the unit of work in a
relational database Not available with the MyISAM engine InnoDB does support transactions Storage engines
InnoDB is the default MyISAM has no foreign keys, but has full
text search
Transactions Often used within stored procedures,
which are compiled programs that can be called by an application
Operations START TRANSACTION COMMIT
ROLLBACK and SAVEPOINT Used when you don’t want to undo an
entire transactions
A Transaction CREATE PROCEDURE test()BEGINDECLARE sql_error TINYINT DEFAULT FALSE;DECLARE CONTINUE HANDLER FOR SQLEXCEPTIONSET sql_error = TRUE;START TRANSACTION; INSERT INTO invoicesVALUES (115, 34, 'ZXA-080', '2012-01-18', 14092.59, 0, 0, 3, '2012-04-18', NULL); INSERT INTO invoice_line_items VALUES (115, 1, 160, 4447.23, 'HW upgrade');
Continued…INSERT INTO invoice_line_items VALUES (115, 2, 167, 9645.36, 'OS upgrade');IF sql_error = FALSE THENCOMMIT;SELECT 'The transaction was committed.';ELSEROLLBACK;SELECT 'The transaction was rolled back.';END IF;END//
SavepointsUSE ap; START TRANSACTION; SAVEPOINT before_invoice;
INSERT INTO invoicesVALUES (115, 34, 'ZXA-080', '2012-01-18', 14092.59, 0, 0, 3, '2012-04-18', NULL); SAVEPOINT before_line_item1; INSERT INTO invoice_line_items VALUES (115, 1, 160, 4447.23, 'HW upgrade'); SAVEPOINT before_line_item2;
Continued…INSERT INTO invoice_line_items VALUES (115, 2, 167, 9645.36,'OS upgrade'); ROLLBACK TO SAVEPOINT before_line_item2;
ROLLBACK TO SAVEPOINT before_line_item1;
ROLLBACK TO SAVEPOINT before_invoice;
COMMIT;
Another view of transactions Prevents
Lost updates from one of two transactions Dirty reads when a transaction reads an
uncommitted value Nonrepeatable reads in one transaction
because the value gets updated in between Phantom reads are when a selection query
is run twice in a transaction and returns different results
Transaction Isolation Leves Set transaction level
Next (no keyword) sets the transaction in the current session
Session sets all transactions in a session Global sets all transactions for all
sessions
Continued… Options
Serializable isolates transactions completely and is the highest level of protection
Read uncommitted lets our four problems occur – no locks
Read committed prevents dirty reads but allows the other problems by not allowing uncommitted writes from being read
Repeatable read is the default and it means that a transaction will always read a given value the same because the values are locked
Deadlock Detect by closing transactions that have
been open a long time Use the lowest acceptable locking level Try to do heavy update transactions
when database can be completely reserved
Stored programs Stored procedures (can be called by an
application) Stored functions (can be called by an
SQL program) Triggers (tied to an operation like
INSERT) Events (tied to a clock)
Flow of control in SQL IF - ELSE CASE – WHEN - ELSE WHILE – DO - LOOP REPEAT - UNTIL – END REPEAT
Variables DECLARE statement SET statement DEFAULT statement INTO (from a SELECT clause)
Example (stored procedure)…CREATE PROCEDURE test()BEGIN DECLARE max_invoice_total DECIMAL(9,2); DECLARE min_invoice_total DECIMAL(9,2); DECLARE percent_difference DECIMAL(9,4); DECLARE count_invoice_id INT; DECLARE vendor_id_var INT; SET vendor_id_var = 95; SELECT MAX(invoice_total), MIN(invoice_total), COUNT(invoice_id) INTO max_invoice_total, min_invoice_total, count_invoice_id FROM invoices WHERE vendor_id = vendor_id_var;
Example, continuedSET percent_difference = (max_invoice_total - min_invoice_total) / min_invoice_total * 100; SELECT CONCAT('$', max_invoice_total) AS 'Maximum invoice', CONCAT('$', min_invoice_total) AS 'Minimum invoice', CONCAT('%', ROUND(percent_difference, 2)) AS 'Percent difference', count_invoice_id AS 'Number of invoices';END//
Domain types – chapter 8 Character Integers Reals Date Time Large object, BLOB and CLOB 2D vector spatial types Enumerated
Conversion Automatic (implied) CAST is the standardized operator CONVERT is similar
Cursor syntax Declare a cursor
DECLARE cursor_name CURSOR FOR select_statement; Declare an error handler for when no rows are found
in the cursor DECLARE CONTINUE HANDLER FOR NOT FOUND
handler_statement; Open the cursor
OPEN cursor_name; Get column values from the row and store them
in a series of variables FETCH cursor_name INTO variable1 [, variable2][, variable3]...;
Close the cursor CLOSE cursor_name;
Example with a cursorDELIMITER // CREATE PROCEDURE test()BEGIN DECLARE invoice_id_var INT; DECLARE invoice_total_var DECIMAL(9,2); DECLARE row_not_found TINYINT DEFAULT FALSE; DECLARE update_count INT DEFAULT 0;
DECLARE invoices_cursor CURSOR FOR SELECT invoice_id, invoice_total FROM invoices WHERE invoice_total - payment_total - credit_total > 0; DECLARE CONTINUE HANDLER FOR NOT FOUND SET row_not_found = TRUE; OPEN invoices_cursor; WHILE row_not_found = FALSE DO FETCH invoices_cursor INTO invoice_id_var, invoice_total_var;
Example with a CursorIF invoice_total_var > 1000 THEN UPDATE invoices SET credit_total = credit_total + (invoice_total * .1) WHERE invoice_id = invoice_id_var; SET update_count = update_count + 1; END IF; END WHILE;
CLOSE invoices_cursor; SELECT CONCAT(update_count, ' row(s) updated.');END//
Triggers ON event IF precondition THEN action All three actions could be SQL
Precondition would be a yes/no, based on results
When are they used? DELETE, UPDATE, INSERT statements
Syntax of MySQL Triggers:Chapter 16CREATE TRIGGER trigger_name {BEFORE|AFTER} {INSERT|UPDATE|DELETE} ON table_name FOR EACH ROW
!! Notice that we can look over the threshold of a state change!!
Trigger exampleCREATE TRIGGER vendors_before_update BEFORE UPDATE ON vendors FOR EACH ROW BEGIN SET NEW.vendor_state = UPPER(NEW.vendor_state);END//** this puts the field in upper case
UPDATE vendorsSET vendor_state = 'wi'WHERE vendor_id = 1
** This illustrates a row level trigger** otherwise, it’s a statement level trigger
Constraint Trigger ExampleCREATE TRIGGER invoices_before_updateBEFORE UPDATE ON invoicesFOR EACH ROWBEGINDECLARE sum_line_item_amount DECIMAL(9,2);
SELECT SUM(line_item_amount) INTO sum_line_item_amount FROM invoice_line_items WHERE invoice_id = NEW.invoice_id; IF sum_line_item_amount != NEW.invoice_total THEN SIGNAL SQLSTATE 'HY000' SET MESSAGE_TEXT = 'Line item total must match invoice total.'; END IF;END
Examples of Audit ConstraintsCREATE TRIGGER invoices_after_insertAFTER INSERT ON invoicesFOR EACH ROWBEGIN INSERT INTO invoices_audit VALUES (NEW.vendor_id, NEW.invoice_number, NEW.invoice_total, 'INSERTED', NOW());END// CREATE TRIGGER invoices_after_deleteAFTER DELETE ON invoicesFOR EACH ROWBEGININSERT INTO invoices_audit VALUES(OLD.vendor_id, OLD.invoice_number,OLD.invoice_total, 'DELETED', NOW());END//
Key Terms! Triggering events (insert, delete,
update) Trigger preconditions (basically a where
clause) Trigger actions (basically a query) Trigger execution (before or after event)