Transactions
- Concurrent access & System failures
- Properties of Transactions
- Isolation Levels
4/9/2015
Databases2
1
Motivated by two independent requirements
 Concurrent database access
 Resilience to system failures
4/9/2015
Databases2
2
Motivating Example
 Consider
T1:
T2:

two transactions:
BEGIN A=A+100, B=B-100 END
BEGIN A=1.06*A, B=1.06*B END
Intuitively, the first transaction is transferring $100 from
B’s account to A’s account. The second is crediting both
accounts with a 6% interest payment.
4/9/2015
Databases2
3
Example (Contd.)
 Consider
T1:
T2:

A=A+100,
A=1.06*A,
B=B-100
B=1.06*B
This is OK. But what about:
T1:
T2:

a possible interleaving:
A=A+100,
A=1.06*A, B=1.06*B
B=B-100
What if system crashes in the middle of a
Transaction?:
4/9/2015
Databases2
4
Concurrency Goal
Execute sequence of SQL statements so they appear
to be running in isolation
 Simple solution: execute them in isolation?
But want to enable concurrency whenever safe to do so
Because disk accesses are frequent, and relatively
slow, it is important to keep the CPU humming by
working on several user programs concurrently
4/9/2015
Databases2
5
Solution for both concurrency and failures
Transactions
A transaction is a sequence of one or more SQL
operations treated as a unit
 Transactions appear to run in isolation
 If the system fails, each transaction’s changes are
reflected either entirely or not at all
4/9/2015
Databases2
6
Solution for both concurrency and failures
Transactions
A transaction is a sequence of one or more SQL
operations treated as a unit. In terms of SQL standard:




Transaction begins automatically on first SQL statement
On “commit” transaction ends and new one begins
Current transaction ends on session termination
“Autocommit” turns each statement into transaction
4/9/2015
Databases2
7
Transactions Properties
ACID Properties
A - Atomicity
C - Consistency
I – Isolation  our main focus today
D - Durability
4/9/2015
Databases2
8
Transactions Properties
ACID Properties
A – Atomicity: If T does not commit (e.g. system
crashes during execution of T), the transaction rolls
back. “All or nothing”.
C – Consistency: T can assume that all integrity
constraints hold when T begins; T must guarantee
they hold when it ends
I – Isolation
D – Durability: If system crashes after transaction
commits, all affects remain in the database.
4/9/2015
Databases2
9
The Standard Default Isolation Level
. . .
Serializable
Operations may be
interleaved, but execution
must be equivalent to some
sequential (serial) order
of all transactions
DBMS
 Overhead
 Reduction in concurrency
Data
4/9/2015
Databases2
10
Isolation Levels
Weaker “Isolation Levels”
. . .
DBMS
Read Uncommitted
Read Committed
Repeatable Read
Serializable
 Overhead  Concurrency
 Consistency Guarantees
Data
4/9/2015
Databases2
11
Isolation Levels
 Per transaction
 “In the eye of the beholder”
My transaction is
Repeatable Read
. . .
My transaction is
Read Uncommitted
DBMS
Data
4/9/2015
Databases2
12
Dirty Reads
“Dirty” data item: written by an uncommitted
transaction
Update Student Set GPA = (1.1)  GPA Where sizeHS > 2500
concurrent with …
Select Avg(GPA) From Student
4/9/2015
Databases2
13
The Isolation Level Read Uncommitted
 A transaction may perform dirty reads
Update Student Set GPA = (1.1)  GPA
Where sizeHS > 2500
concurrent with …
Set Transaction Isolation Level Read Uncommitted;
Select Avg(GPA) From Student;
4/9/2015
Databases2
14
Q
Consider a table R(A) containing {(1),(2)} and two
transactions:
T1: update R set A=2*A;
T2: select avg(A) from R;
If T2 executes using “read uncommitted”, what are the
possible values it returns?
[ ] 1.5, 2, 2.5, 3
[ ] 1.5, 2, 3
[ ] 1.5, 2.5, 3
[ ] 1.5, 3
4/9/2015
Databases2
15
A
Consider a table R(A) containing {(1),(2)} and two
transactions:
T1: update R set A=2*A;
T2: select avg(A) from R;
If T2 executes using “read uncommitted”, what are the
possible values it returns?
[+] 1.5, 2, 2.5, 3
[ ] 1.5, 2, 3
Explanation: The update
[ ] 1.5, 2.5, 3
command in T1 can update the
values in either order, and the
[ ] 1.5, 3
select in T2 can compute the
avg at any point before,
between, or after the updates.
4/9/2015
Databases2
16
The Isolation Level Read Committed
 A transaction may not perform dirty reads
Update Student Set GPA = (1.1)  GPA Where sizeHS > 2500
concurrent with …
Set Transaction Isolation Level Read Committed;
Select Avg(GPA) From Student;
4/9/2015
Databases2
17
The Isolation Level Read Committed
 A transaction may not perform dirty reads
Still does not guarantee global serializability
Update Student Set GPA = (1.1)  GPA Where sizeHS > 2500
concurrent with …
Set Transaction Isolation Level Read Committed;
Select Avg(GPA) From Student;
Select Max(GPA) From Student;
4/9/2015
Databases2
18
Q
Consider tables R(A) and S(B),both containing
{(1),(2)}, and two transactions:
T1: update R set A=2*A; update S set B=2*B;
T2: select avg(A) from R; select avg(B) from
S;
If T2 executes using “read committed”, is it
possible for T2 to return two different values?
[ ] No
[ ] Yes
4/9/2015
Databases2
19
A
Consider tables R(A) and S(B),both containing
{(1),(2)}, and two transactions:
T1: update R set A=2*A; update S set B=2*B;
T2: select avg(A) from R; select avg(B) from
S;
If T2 executes using “read committed”, is it
possible for T2 to return two different values?
[ ] No
Explanation: T2 could return
avg(A) computed before T1,
[+] Yes
and avg(B) computed after T1.
4/9/2015
Databases2
20
The Isolation Level Repeatable Read
 A transaction may not perform dirty reads
 An item read multiple times cannot change value
Update Student Set GPA = (1.1)  GPA Where sizeHS > 2500
concurrent with …
Set Transaction Isolation Level Repeatable Read;
Select Avg(GPA) From Student;
Select Max(GPA) From Student;
4/9/2015
Databases2
21
The Isolation Level Repeatable Read
 A transaction may not perform dirty reads
 An item read multiple times cannot change value
Still does not guarantee global serializability
Update Student Set GPA = (1.1)  GPA;
Update Student Set sizeHS = 1500 Where sID = 123;
concurrent with …
Set Transaction Isolation Level Repeatable Read;
Select Avg(GPA) From Student;
Select Avg(sizeHS) From Student;
4/9/2015
Databases2
22
Isolation Level Repeatable Read
 A transaction may not perform dirty reads
 An item read multiple times cannot change value
But a relation can change: “phantom” tuples
Insert Into Student [ 100 new tuples ]
concurrent with …
Set Transaction Isolation Level Repeatable Read;
Select Avg(GPA) From Student;
Select Max(GPA) From Student;
4/9/2015
Databases2
23
Isolation Levels: Summary
dirty reads
nonrepeatable
reads
phantoms
Read Uncommitted
Read Committed
Repeatable Read
Serializable
4/9/2015
Databases2
24
Isolation Levels: Summary
dirty reads
nonrepeatable
reads
phantoms
Read Uncommitted
Y
Y
Y
Read Committed
N
Y
Y
Repeatable Read
N
N
Y
Serializable
N
N
N
4/9/2015
Databases2
25
Summing up
Transactions: solution for both concurrency & failures
Properties: A,C,I,D
Isolation Levels
 Standard default: Serializable
 Weaker isolation levels
– Increased concurrency + decreased overhead =
increased performance
– Weaker consistency guarantees
– Some systems have default Repeatable Read
 Isolation level per transaction and “eye of the beholder”
– Each transaction’s reads must conform to its isolation level
4/9/2015
Databases2
26