1. Database Recovery

2. System crash Ti transfers $50 from account A to account B
Initial values:
A= $1000
B = $ 2000
Final values:
A=$950
B=$2050
begin _transaction
read(A)
A:= A - 50
write(A)
read(B)
B:= B + 50
write(B)
commit
System
crash
Possible recovery procedures:
Reexecute Ti : will result in A=$900; B=$2050
 inconsistent state
Do not reexecute Reexecute Ti : will result in A=$950; B=$2000

3. DBMS Transaction Subsystem
Transaction manager coordinates transactions on behalf of application programs
It communicates with the scheduler
Transaction
manager
Scheduler
Recovery
Buffer
Systems
Access
File
The scheduler handles concurrency control.
Its objective is to maximize concurrency without allowing simultaneous transactions to interfere with one another
The recovery manager ensures that the database is restored to the right state before a failure occurred.
The buffer manager is responsible for the transfer of data between disk storage and main memory.

4. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Recovery With Concurrent Transactions
Shadow Paging

5. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Shadow Paging
Recovery With Concurrent Transactions

6. Types of Failure Classification
Transaction failure
System crash
Disk failure
Physical problems and catastrophes

7. Classification Transaction failure
Types of errors that may cause transaction failure :
Logical errors: transaction cannot complete due to some internal error condition (e.g. division by zero)
System errors: the database system must terminate an active transaction due to an error condition (e.g., deadlock)

8. Classification System crash
A power failure or other hardware or software failure causes the system to crash.
Loss of content of volatile storage
Transaction processing is halted
Nonvolatile storage remains intact
Fail-stop assumption:
Hardware errors and software bugs stop system processing
Database systems have numerous integrity checks to prevent corruption of disk data
non-volatile storage contents are assumed to not be corrupted by system crash

9. Classification Disk failure
A head crash or failure during data transfer operations destroys all or part of disk storage.
Destruction is assumed to be detectable: disk drives use checksums to detect failures

10. Classification Physical problems and catastrophes
Endless list of problems that includes:
power or air-conditioning failure,
fire,
theft,
sabotage,
overwriting disks or tapes by mistake,
mounting of the wrong tape by the operator, etc.

11. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Shadow Paging
Recovery With Concurrent Transactions

12. Storage Structure Volatile storage Nonvolatile storage Stable storage
does not survive system crashes
examples: main memory, cache memory
Nonvolatile storage
survives system crashes
examples: disk, tape, non-volatile (battery backed up) RAM
Stable storage
a mythical form of storage that survives all failures
closely approximated by maintaining multiple copies on distinct non-volatile media

13. Data Access Physical blocks are those blocks residing on the disk.
Buffer blocks are the blocks residing temporarily in main memory.
Block movements between disk and main memory are initiated through the following two operations:
input(B) transfers the physical block B to main memory.
output(B) transfers the buffer block B to the disk, and replaces the appropriate physical block there.

14. Data Access X Y
Buffer A B
input(A)
Each transaction Ti has its private work-area in which local copies of all data items accessed and updated by it are kept.
Ti's local copy of a data item X is called xi.
We assume, for simplicity, that each data item fits in, and is stored inside, a single block.
Buffer block B
output(B)
read-item(X)
write-item(X) X1 Y1
Work Area of T1 X2
Work Area of T2
MAIN MEMORY

15. read(X) assigns the value of data item X to the local variable xi.
A transaction transfers data items between system buffer blocks and its private work-area using the following operations :
read(X) assigns the value of data item X to the local variable xi.
write(X) assigns the value of local variable xi to data item {X} in the buffer block.
both these commands may necessitate the issue of an input(BX) instruction before the assignment, if the block BX in which X resides is not already in memory. X Y
Buffer A B
input(A)
Buffer block B
output(B)
read-(X)
write(X) X1 Y1
Work Area of T1 X2
Work Area of T2
MAIN MEMORY

16. Perform read(X) while accessing X for the first time;
Transactions
Perform read(X) while accessing X for the first time;
All subsequent accesses are to the local copy.
After last access, transaction executes write(X).
output(BX) need not immediately follow write(X).
System can perform the output operation when it deems fit.
If the system crashes after the write(X) operation was executed but before output(BX) was executed, the new value of X is never written to disk and, thus, is lost. A B X Y
Buffer
MAIN MEMORY X1 Y1
Work Area of T1 X2
Work Area of T2
Buffer block B
read-(X)
write(X)
input(A)
output(B)

17. input(A) A B
Buffer Y
Buffer block A X
Buffer block B
read(X)
read(Y)
output(B)
Given:
T1: wants to read item X and update item Y.
item X is in block A while item Y is in block B
Block B is already in memory
write(Y) X1 Y1
Work Area of T1
MAIN MEMORY

18. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Shadow Paging
Recovery With Concurrent Transactions

19. Recovery Algorithms
Recovery algorithms are techniques to ensure database consistency and transaction atomicity and durability despite failures
Recovery algorithms have two parts:
Actions taken during normal transaction processing, to ensure enough information exists to recover from failures
Actions taken after a failure, to recover the database contents to a state that ensures atomicity, consistency and durability

20. System crash Given: Ti that transfers $50 from account A to account B
Initial values:A= $1000;B = $ 2000
begin _transaction
read(A)
A:= A - 50
write(A)
read(B)
B:= B + 50
write(B)
commit
System
crash
In case of a failure:
The system must have enough information to be able to keep track of the values of A and B.
The system must be able to recover (perform actions after the failure such that the database is consistent)

21. Recovery and Atomicity
Modifying the database without ensuring that the transaction will commit may leave the database in an inconsistent state.
Several output operations may be required for a transaction Ti (to output A and B). A failure may occur after one of these modifications have been made but before all of them are made.

22. Recovery and Atomicity
Suppose a system crash occurred during the execution of Ti:
after output(BA) has taken place
before output(BB) was executed
Recovery and Atomicity
begin _transaction
read-item(A)
A:= A - 50
write-item(A)
read(B)
B:= B + 50
write-item(B)
commit
Ti that transfers $50 from account A to account B
Initial values:
A= $1000
B = $ 2000
Final values:
A=$950
B=$2050
System
crash
Possible recovery procedures:
Reexecute Ti : will result in A=$900; B=$2050
 inconsistent state
Do not reexecute Reexecute Ti : will result in A=$950; B=$2000

23. Recovery and Atomicity
To ensure atomicity despite failures:
first output information describing the modifications to stable storage without modifying the database itself.
Assume (initially) that transactions run serially, that is, one after the other.

24. 2 Recovery Approaches
Log-based recovery
Shadow paging

25. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Deferred database modification
Immediate database modification
Recovery With Concurrent Transactions
Shadow Paging

26. Log-Based Recovery A log is kept on stable storage.
it is a sequence of log records, and maintains a record of update activities on the database.
When transaction Ti starts, it registers itself by writing a <Ti start> log record
Before Ti executes write(Xj ), a log record
<Ti, Xj, V1, V2> is written, where V1 is the value of Xj before the write, and V2 is the value to be written to Xj .
Log record notes that Ti has performed a write on data item Xj
Xj had value V1 before the write, and will have value V2 after the write.
<Ti start>
<Ti, Xj, V1, V2>

27. Log-Based Recovery
When Ti finishes its last statement, the log record <Ti commit> is written.
When Ti is aborted, the log record <Ti abort> is written.
We assume for now that log records are written directly to stable storage (that is, they are not buffered)
<Ti, Xj, V1, V2>
<Ti start>
<Ti commit>

28. Log-Based Recovery Two approaches using logs:
Deferred database modification
Immediate database modification

29. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Deferred database modification
Immediate database modification
Shadow Paging
Recovery With Concurrent Transactions

30. Deferred Database Modification: what?
The deferred database modification scheme:
records all modifications to the log
defers the execution of all the write operations of a transaction until the transaction partially commits
a transaction has partially committed once the final action of the transaction has been executed
Assume that transactions execute serially

31. T1 :Withdrawal transaction
When the transaction finishes its final statement, it enters the partially committed state.
The actual output may still be residing in memory; thus, a hardware failure may preclude its successful completion
partially
committed
committed
active
T1 :Withdrawal transaction
Time T1
balx t1
begin _transaction
100 t2
read(balx) t3
balx=balx - 10 t4
write(balx) 90 t5
end_transaction
commit
When the last of the information is written out in the disk, the transaction enters the committed state.
In the event of failure, the updates of a transaction can be re-created.

32. Deferred Database Modification: what
begin _transaction
read(A)
A:= A - 50
write(A)
read(B)
B:= B + 50
write(B)
commit
*deferred
*deferred
partially
commits
Update DB

33. Deferred Database Modification: how
Transaction starts by writing <Ti start> record to log.
A write(X) operation results in a log record <Ti,X,V> being written, where V is the new value for X. Note: old value is not needed for this scheme
The write is not performed on X at this time, but is deferred.
When Ti partially commits,
<Ti commit> is written to the log
the log records are read and used to actually execute the previously deferred writes.
Once the updating of DB is completed, the transaction enters the committed phase.

34. Deferred Database Modification
<Ti start>
begin _transaction
read-item(A)
A:= A - 50
write-item(A)
read(B)
B:= B + 50
write-item(B)
commit
< Ti , A, 950>
< Ti, B,2050>
< Ti commit>
Database is modified using the log record
Partially committed
A= 950
B= 2050
A= 1000
B= 2000

35. Deferred Database Modification
Transaction starts by writing <Ti start> record to log.
When Ti partially commits,
<Ti commit> is written to the log
Deferred Database Modification
A write(X) operation results in a log record <Ti, X, V> being written, where V is the new value for X.
Example:
T0 and T1 (T0 executes before T1).
Initial values: A=$1000, B=$2000, C=700
T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
<T1 start>
< T1 , C, 600>
< T1 commit>
T1:
begin
read -item(C)
C:= C - 100
write-item (C)
commit

36. What happens if T0 and T1 successfully commit?
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
Database
A=950
B=2050
When T0 partially commits, the log records are read and used to actually execute the previously deferred writes of T0.
Once the updating of DB is completed by T0 , T0 enters the committed phase.

37. What happens if T0 and T1 successfully commit?
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
When T1 partially commits, the log records are read and used to actually execute the previously deferred writes.
Once the updating of DB is completed by T1 , T1 enters the committed phase.
<T1 start>
< T1 , C, 600>
< T1 commit>
Database
C=600

38. Deferred Database Modification
When can crashes occur?
When the transaction is executing the original updates, or
While recovery action is being taken

39. Deferred Database Modification
How is recovery done?
During recovery after a crash, a transaction needs to be redone if and only if both <Ti start> and <Ti commit> are there in the log.
Redoing a transaction Ti ( redo(Ti) ) sets the values of all data items updated by the transaction to the new values.
The redo operation must be idempotent; that is, executing it several times must be equivalent to executing it once. This characteristic guarantees correct behavior even if failure occurs during recovery.

40. T0:
begin
read (A)
A:= A - 50
write(A)
read (B)
B:= B + 50
write(B)
commit
T1:
Begin-transaction
read (C)
C:= C - 100
write (C)
Case 1: Crash occurs just after the log record for the step write(B) of T0 has been written to stable storage.
Log record:
<T0 start>
< T0 , A, 950>
< T0, B,2050>
crash

41. T0:
begin
read (A)
A:= A - 50
write(A)
read (B)
B:= B + 50
write(B)
commit
T1:
Begin-transaction
read (C)
C:= C - 100
write (C)
During recovery after a crash, a transaction needs to be redone if and only if both <Ti start> and <Ti commit> are there in the log.
Case 1: Crash occurs just after the log record for the step write(B) of T0 has been written to stable storage.
Log record:
<T0 start>
< T0 , A, 950>
< T0, B,2050>
crash
Since no commit record <Ti commit> appears in the log, no redo action is taken.
A= 1000; B=2000; C=700
It is assumed that the transaction was never executed
Log records of incomplete transactions can be deleted from the log.

42. Case 2: Crash occurs just after the log record for the step write(C) of T1 has been written to stable storage.
T0:
begin
read (A)
A:= A - 50
write(A)
read(B)
B:= B + 50
write(B)
commit
T1:
read (C)
C:= C - 100
write (C )
Log record:
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
<T1 start>
< T1 , C, 600>
crash

43. During recovery after a crash, a transaction needs to be redone if and only if both <Ti start> and <Ti commit> are there in the log.
Case 2: Crash occurs just after the log record for the step write(C) of T1 has been written to stable storage.
T0:
begin
read (A)
A:= A - 50
write(A)
read(B)
B:= B + 50
write(B)
commit
T1:
read (C)
C:= C - 100
write (C )
Log record:
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
<T1 start>
< T1 , C, 600>
redo(T0) is performed since both <T0 start> and <T0 commit> exist in the log
No redo for T1 since no commit record <T1 commit> appears in the log
After redo(T0) : A= 950; B=2050; C=700
crash

44. Case 3: Crash occurs just after the log record <T1 commit> has been written to stable storage.
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
Log record:
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
<T1 start>
< T1 , C, 600>
< T1 commit>
crash

45. During recovery after a crash, a transaction needs to be redone if and only if both <Ti start> and <Ti commit> are there in the log.
Case 3: Crash occurs just after the log record <T1 commit> has been written to stable storage.
T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
<T1 start>
< T1 , C, 600>
< T1 commit>
Two redo operations redo(T0) and redo(T1) are performed
After redo(T0) and redo(T1) :
Final values: A= 950; B=2050; C=600
crash

46. Case 4: Suppose a second system crash occurs during the recovery of the first crash.
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
<T1 start>
< T1 , C, 600>
< T1 commit>
For each commit record < Ti commit> in the log, redo(Ti) is performed.
Since redo writes values independent of the values currently in the database, the result of a successful attempt at redo is the same as though redo had succeeded the first time.
Second
crash

47. Outline Failure Classification Storage Structure
Recovery and Atomicity
Log-Based Recovery
Deferred Database Modification
Immediate Database Modification
Recovery With Concurrent Transactions
Shadow Paging

48. Immediate Database Modification
The immediate database modification scheme allows database updates of an uncommitted transaction to be made as the writes are issued.
Since undoing may be needed, update logs must have both old value and new value.

49. Immediate Database Modification: How?
Transaction starts by writing <Ti start> record to log.
Before executing a write operation, the log record <Ti,Xj,V1,V2> is written where V1 is the old value and V2 is the new value.
log record <Ti, Xj, V1, V2> must be written in stable storage before the item Xj is written in the database.
That is, before output(B) is executed, the log record <Ti, Xj, V1, V2> is written to stable storage.
can be extended to postpone log record output, so long as prior to execution of an output(B) operation for a data block B, all log records corresponding to items B must be flushed (written) to stable storage.
Output of updated blocks can take place at any time before or after transaction commits
Order in which blocks are output can be different from the order in which they are written.
When Ti partially commits, <Ti commit> is written to the log.

50. <T0 start> < T0 , A, 1000,950> < T0, B, 2000, 2050>
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C)
<T0 start>
< T0 , A, 1000,950>
< T0, B, 2000, 2050>
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
< T0 commit>
A= 950
A= 1000
B= 2050
B= 2000
C= 600
C= 700

51. Transaction starts by writing <Ti start> record to log.
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C)
<T0 start>

52. T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C)
<T0 start>
< T0 , A, 1000,950>
< T0, B, 2000, 2050>
Before executing a write operation, the log record <Ti, Xj, V1, V2> is written where V1 is the old value and V2 is the new value.

53. T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C)
LOG DB
<T0 start>
< T0 , A, 1000,950>
< T0, B, 2000, 2050>
A=950
B=2050
Output of updated blocks can take place at any time before or after transaction commits

54. When Ti partially commits, <Ti commit> is written to the log.
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C)
<T0 start>
< T0 , A, 1000,950>
< T0, B, 2000, 2050>
< T0 commit>
When Ti partially commits, <Ti commit> is written to the log.

55. LOG DB <T0 start> < T0 , A, 1000,950>
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C)
LOG DB
<T0 start>
< T0 , A, 1000,950>
< T0, B, 2000, 2050>
A=950
B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
< T1 commit>
C=600

56. Immediate vs Deferred IMMEDIATE DEFERRED Log Database Log Database
<T0 start>
<T0, A, 1000, 950>
<T0, B, 2000, 2050>
A = 950
B = 2050
<T0 commit>
<T1 start>
<T1, C, 700, 600>
C = 600
<T1 commit>
DEFERRED
Log Database
<T0 start>
< T0 , A, 950>
< T0, B,2050>
< T0 commit>
A=950
B=2050
<T1 start>
< T1 , C, 600>
< T1 commit>
C=600

57. Immediate Database Modification
Recovery procedure has two operations:
undo(Ti) restores the value of all data items updated by Ti to their old values, going backwards from the last log record for Ti
redo(Ti) sets the value of all data items updated by Ti to the new values, going forward from the first log record of Ti

58. Immediate Database Modification
Both operations must be idempotent
That is, even if the operation is executed multiple times the effect is the same as if it is executed once.
This is needed since operations may get re-executed during recovery

59. Immediate Database Modification
How to recover after a failure:
Transaction Ti needs to be undone if the log contains the record <Ti start>, but does not contain the record <Ti commit>.
Transaction Ti needs to be redone if the log contains both the record <Ti start> and the record <Ti commit>.
Undo operations are performed first, then redo operations.

60. T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
Case1: Crash occurs just after the log record for the step write-item(B) of T0 has been written to stable storage.
log database values
<T0 start>
< T0, A, 1000, 950>
< T0, B, 2000, 2050>
A=950;B=2050
crash
Since <T0 start> exists but no commit record <T0 commit> appears in the log, undo(T0) action is performed.
Transaction Ti needs to be undone if the log contains the record <Ti start>, but does not contain the record <Ti commit>.

61. going backwards from the last log record for Ti
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Case 1: Crash occurs just after the log record for the step write-item(B) of T0 has been written to stable storage.
log database values
<T0 start>
< T0, A, 1000, 950>
< T0, B, 2000, 2050>
A=950;B=2050
crash
How to undo(T0 )? :
going backwards from the last log record for Ti
< T0, B, 2000, 2050>: B is restored to 2000;
< T0, A, 1000, 950> : A is restored to 1000

62. undo(T1) is performed since only <T1 start> is in the log.
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
Case 2: Crash occurs just after the log record for the step write-item(C) of T1 has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C = 600
crash
undo(T1) is performed since only <T1 start> is in the log.
redo(T0) is performed since both <T0 start> and <T0 commit> exist in the log.

63. going backwards from the last log record for T1
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Case 2: Crash occurs just after the log record for the step write-item(C) of T1 has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C = 600
crash
First: undo(T1)
How system recovers?
going backwards from the last log record for T1
< T1 , C, 700, 600>:C is restored to 700;

64. going forward from the first log record for Ti
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Case 2: Crash occurs just after the log record for the step write-item(C) of T1 has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C = 600
crash
Second: redo (T0)
How system recovers?
going forward from the first log record for Ti
< T0 , A, 1000, 950>: A is updated to 950
< T0, B, 2000, 2050>: B is updated to 2050;

65. Two redo operations redo(T0) and redo(T1) are performed
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
C:= C - 100
write-item (C )
Case 3: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C=600
< T1 commit>
crash
Two redo operations redo(T0) and redo(T1) are performed

66. going forward from the first log record for T0
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Case 3: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C=600
< T1 commit>
First: redo (T0)
crash
going forward from the first log record for T0
< T0 , A, 1000, 950>: A is updated to 950
< T0, B, 2000, 2050>: B is updated to 2050;

67. going forward from the first log record for T1
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Case 3: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C=600
< T1 commit>
Second: redo (T1)
crash
going forward from the first log record for T1
< T1 , C, 700, 600>: C is updated to 600

68. Problems in recovery procedure
searching the entire log is time-consuming (log must be searched to determine which transactions are to be redone or undone)
we might unnecessarily redo transactions which have already output their updates to the database.

69. T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Case 3: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
< T1 , C, 700, 600>
C=600
< T1 commit>
crash
Since A, B, C have been updated in the database, redo(T0) and redo(T1) are unnecessary.

70. Checkpoints
To streamline recovery procedure, during execution, in addition to maintaining the log (deferred or immediate updates), periodic checkpointing must be performed and requires the following actions to take place:
Output all log records currently residing in main memory onto stable storage.
Output all modified buffer blocks to the disk.
Write a log record <checkpoint> onto stable storage.
Note: while a checkpoint is in progress, transactions are not allowed to perform any update actions such as writing to a buffer block or writing a log record.

71. Checkpoints During recovery we need to consider only:
the most recent transaction Ti that started before the checkpoint (but has not been completed prior to checkpoint), and
transactions that started after Ti. tc tf T1 T2 T3 T4
checkpoint
system failure

72. T0:
begin
read -item(A)
A:= A - 50
write-item (A)
read-item (B)
B:= B + 50
write-item (B)
commit
T1:
read -item(C)
A:= A - 100
write-item (C )
Example: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
<checkpoint>
checkpoint
< T1 , C, 700, 600>
C=600
< T1 commit>
crash

73. Example: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
<checkpoint>
< T1 , C, 700, 600>
C=600
< T1 commit>
Step 1: Scan backwards from end of log to find the most recent <checkpoint> record.

74. Example: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
<checkpoint>
< T1 , C, 700, 600>
C=600
< T1 commit>
start record
Step 2: Continue scanning backwards till a record <Ti start> is found.

75. Example: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
<checkpoint>
< T1 , C, 700, 600>
C=600
< T1 commit>
start record
Step 3: Need only consider the part of log following start record. Earlier part of log can be ignored during recovery, and can be erased whenever desired.

76. Example: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
<checkpoint>
< T1 , C, 700, 600>
C=600
< T1 commit>
No undo
operations to be
undertaken
start record
Step 4: For all transactions (starting from Ti or later) with no <Ti commit>, execute undo(Ti). (Done only in case of immediate modification.)

77. Example: Crash occurs just after the log record <T1 commit> has been written to stable storage.
<T0 start>
< T0 , A, 1000, 950>
< T0, B, 2000, 2050>
A=950; B=2050
< T0 commit>
<T1 start>
<checkpoint>
< T1 , C, 700, 600>
C=600
< T1 commit>
redo(T1):
C is updated to 600
With checkpoints, redo(T0), which is unnecessary was eliminated
start record
Step 5: Scanning forward in the log, for all transactions starting from Ti or later with a <Ti commit>, execute redo(Ti).

78. Example of Checkpointstc tf T1 T2 T3 T4
checkpoint
system failure
T1 can be ignored (updates already output to disk due to checkpoint)
T2 and T3 redone.
T4 undone

79. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Deferred database modification
Immediate database modification
Recovery With Concurrent Transactions
Shadow Paging

80. We allow immediate modification to the database.
Given: T1 subtracts 100 from A ; T2 increases the value of A by 10% ;Initial value of A is 1000.
We allow immediate modification to the database. T1 T2
A = 1000
read(A)
A:= A-100
write(A)
A = 900
A:= A*1.1
A= 990
commit
failure
Abort/rollback

81. Review: Immediate Database Modification
When recovering after failure:
Transaction Ti needs to be undone if the log contains the record <Ti start>, but does not contain the record <Ti commit>.
Transaction Ti needs to be redone if the log contains both the record <Ti start> and the record <Ti commit>.
Undo operations are performed first, then redo operations.

82. Review: Checkpoints
During recovery we need to consider only the most recent transaction Ti that started before the checkpoint, and transactions that started after Ti.
Scan backwards from end of log to find the most recent <checkpoint> record.
Continue scanning backwards till a record <Ti start> is found.
Need only consider the part of log following above start record. Earlier part of log can be ignored during recovery, and can be erased whenever desired.

83. Review: Checkpoints
For all transactions (starting from Ti or later) with no <Ti commit>, execute undo(Ti) (scanning backwards). (Done only in case of immediate modification.)
Scanning forward in the log, for all transactions starting from Ti or later with a <Ti commit>, execute redo(Ti).

84. T2 can be ignored (updates already output to disk due to checkpoint)
read-item(A)
<T1 start>
A:= A-100
write-item(A)
<T1,A,1000,900>
A = 900
<T2 start>
A:= A*1.1
<T2,A,900,990>
A= 990
commit
<T2 commit>
checkpoint
Abort/rollback
failure tc tf
T2 can be ignored (updates already output to disk due to checkpoint)
T1 is not finished; it has to be undone T1 T2
checkpoint
system failure

85. Why is this schedule unrecoverable?
read-item(A)
<T1 start>
A:= A-100
write-item(A)
<T1,A,1000,900>
A = 900
<T2 start>
A:= A*1.1
<T2,A,900,990>
A= 990
commit
<T2 commit>
checkpoint
Abort/rollback
Why is this schedule unrecoverable?
It does not follow strict 2PL.
failure
Based on recovery procedures for immediate modification, T1 has to be undone since it has a <T1 start> but not a <T1 commit>
However, if T1 is rolled back the updates of T2 will be dirty/lost.
Therefore, UNDO is not possible for T1

86. Recovery With Concurrent Transactions
All transactions share a single disk buffer and a single log
A buffer block can have data items updated by one or more transactions
We assume concurrency control using strict two-phase locking, that is:
A transaction T does not release any of its exclusive (write) locks until after it commits or aborts.
Hence, no other transaction can read or write an item that is written by T unless T has committed, leading to a strict schedule for recoverability (the updates of uncommitted transactions should not be visible to other transactions)

87. Recovery With Concurrent Transactions
The checkpointing technique and actions taken on recovery have to be changed since several transactions may be active when a checkpoint is performed.
Checkpoints are performed as before, except that the checkpoint log record is now of the form < checkpoint L>
where L is the list of transactions active at the time of the checkpoint
We assume no updates are in progress while the checkpoint in the buffer or log is carried out (can be relaxed in advance techniques)

88. ** cannot proceed because T1 locks A
A = 1000; B=500
write-lock(A)
read-item(A)
A:= A-100
write-item(A)
A = 900
A:= A*1.1
commit
write-lock(B)
read-item(B)
B:= B+100
write-item(B) …
Locks are released after commit or abort
<T1 start>
< T1,A,1000,900>
<T2 start>
** cannot proceed because T1 locks A
<checkpoint [T2,T1]>
< T1,B,500,600>
Abort/rollback

89. How the system recovers from crash
Log
<T1 start>
<T1,A,1000,900>
<T2 start>
<checkpoint [T2,T1]>
<T1,B,500,600>
Step 1:Initialize undo-list and redo-list to empty
undo-list = []
redo-list = []

90. How the system recovers from crash
Step 2. Scan the log backwards from the end, stopping when the first <checkpoint L> record is found.
Log
<T1 start>
<T1,A,1000,900>
<T2 start>
<checkpoint [T2,T1]>
<T1,B,500,600>
For every Ti in L, if Ti is not in redo-list, add Ti to undo-list
a. if the record is <Ti commit>, add Ti to redo-list
b. if the record is <Ti start>, then if Ti is not in redo-list, add Ti to undo-list
redo-list = []
undo-list [T2,T1]

91. How the system recovers from crash
Log
<T1 start>
<T1,A,1000,900>
<T2 start>
<checkpoint [T2,T1]>
<T1,B,500,600>
Step 3. Once the redo-list and undo-list have been constructed, rescan log backwards from most recent record, stopping when <Ti start> records have been encountered for every Ti in undo-list.
During the rescan, perform undo for each log record that belongs to a transaction in undo-list.
undo-list [T2,T1]
undo(T2)
undo(T1)

92. How the system recovers from crash
Log
<T1 start>
<T1,A,1000,900>
<T2 start>
<checkpoint [T1,T2]>
<T1,B,500,600>
Step 4. Locate the most recent <checkpoint L> record. (this may involve scanning the log forward if the record was passed in step 4.
undo-list [T1,T2]
redo-list[]

93. How the system recovers from crash
Log
<T1 start>
<T1,A,1000,900>
<T2 start>
<checkpoint [T1,T2]>
<T1,B,500,600>
Step 5. Scan log forwards from the <checkpoint L> record till the end of the log.
During the scan, perform redo for each log record that belongs to a transaction on redo-list
redo-list []
No redo-list

94. Recovery With Concurrent Transactions
At this point undo-list consists of incomplete transactions which must be undone, and redo-list consists of finished transactions that must be redone.

95. Outline Failure Classification Storage Structure and Data Access
Recovery and Atomicity
Log-Based Recovery
Deferred database modification
Immediate database modification
Recovery With Concurrent Transactions
Shadow Paging

96. Shadow Paging Shadow paging is an alternative to log-based recovery;
It is useful if transactions execute serially.
It is hard to extend shadow paging in concurrent processing of multiple transactions.

97. In shadow paging, the DB is partitioned into fixed-length blocks called pages (or blocks).
These pages need not be stored in a particular order.
However, there must be a way to find the ith page of the DB for any given i. .
Page n
Database Disk

98. To find the ith page of the DB for any given i, a page table, which is kept in the memory is used.
DB disk pages
The page table has n entries, one for each database page 1 2 3 4 5 6 7 n .
page table
in memory
Each entry in the page table contains a pointer to a page in the DB disk

99. 5 n
input(A)
Buffer
Buffer block 5 X
read-item(X)
Given:
Tj: wants to update item X.
item X is in page (block) 5
write-item(X)
Work Area of Tj
Whenever a page is about to be written for the first time, shadow paging is undertaken X1
MAIN MEMORY

100. shadow page table (disk)
current page table
(memory) 1 2 3 4 5 6
shadow page table (disk) 1 2 3 4 5 6
DB pages (disk) 5 1 4 2 3 6
Shadow Paging idea : maintain two page tables during the lifetime of a transaction – the current page table and the shadow page table.
When transaction Tj starts, both pages are identical.
The shadow page table, which is saved on a nonvolatile storage, is never modified during transaction execution.
The current page table may be modified when the transaction performs a write operation.

101. shadow page table (disk)
current page table
(memory) 1 2 3 4 5 6
shadow page table (disk)
DB pages (disk)
X is found in page 5
Unused pages or free blocks
There are unused pages or free blocks in the DB disk

102. shadow page table (disk)
current page table
(memory) 1 2 3 4 5 6
shadow page table (disk)
DB pages (disk)
Tj performs write-item(X) operation:
Check: whenever any page is about to be written for the first time:
1. A copy of the page (page 5) is made onto an unused page.
System finds an unused page and deletes it from the list of free page frames 5
System copies the contents of page 5 to the found unused page

103. shadow page table (disk)
current page table
(memory) 1 2 3 4 5 6
shadow page table (disk)
DB pages (disk)
2. The current page table is then made to point to the copy.

104. shadow page table (disk)
current page table
(memory) 1 2 3 4 5 6
shadow page table (disk) 1 2 3 4 5 6
DB pages (disk) 5 1 4 2 3 6 5
2. The current page table is then made to point to the copy.

105. shadow page table (disk)
current page table
(memory) 1 2 3 4 5 6
shadow page table (disk) 1 2 3 4 5 6
DB pages (disk) 5 1 4 2 3 6
old 5
new
3. The update is performed on the copy.

106. input(A)
Buffer
Buffer block 5 X 5
5 new n
read-item(X)
write-item(X)
Work Area of Tj X1
shadow paging is undertaken for first write of page
MAIN MEMORY

107. input(A)
Buffer
Buffer block 5 X 5
5 new n
read-item(X)
To commit Tj
To commit a transaction
1. Flush (output) all modified pages (by a transaction) in main memory to disk
write-item(X)
Work Area of Tj X1
shadow paging is undertaken for first write of page
MAIN MEMORY

108. To commit a transaction : 2. Output current page table to disk.
(memory) 1 2 3 4 5 6
DB pages 5 1 4 2 3 6
old
shadow page table 1 2 3 4 5 6
copy of current
page table 1 2 3 4 5 6 5
new
To commit a transaction :
2. Output current page table to disk.

109. To commit a transaction :
current page table
(memory) 1 2 3 4 5 6
DB pages 5 1 4 2 3 6
old
shadow page table 1 2 3 4 5 6
copy of current
page table 1 2 3 4 5 6
To commit a transaction :
3. Make the current page table the new shadow page table, as follows:
a. keep a pointer to the shadow page table at a fixed (known) location on disk. 5
new

110. To commit a transaction :
current page table
(memory) 1 2 3 4 5 6
DB pages 5 1 4 2 3 6
old
Old shadow page table 1 2 3 4 5 6
shadow
page table 1 2 3 4 5 6
To commit a transaction :
3. Make the current page table the new shadow page table, as follows:
b. update the pointer to point to current page table on disk. 5
new

111. To commit a transaction in SP (summary)
Flush (output) all modified pages (by a transaction) in main memory to disk
Output current page table to disk
Make the current page table the new shadow page table, as follows:
keep a pointer to the shadow page table at a fixed (known) location on disk.
to make the current page table the new shadow page table, simply update the pointer to point to current page table on disk.

112. Shadow Paging
Once pointer to shadow page table has been written, transaction is committed.
No recovery is needed after a crash — new transactions can start right away, using the shadow page table.
Pages not pointed to current/shadow page table should be freed (garbage collected).

113. Transaction Tj committed
current page table
(memory) 1 2 3 4 5 6
DB pages 1 4 2 3 6 5
DB pages 5 1 4 2 3 6
old
Old shadow page table 1 2 3 4 5 6
shadow
page table 1 2 3 4 5 6
Transaction Tj committed 5
new
Old page 5 is freed

114. Shadow Paging: When a crash occurs
If a crash occurs prior to finally committing a transaction (prior to completion of step 3: to commit a transaction)
The system reverts to the state prior to the execution of the transaction.
If a crash occurs after a transaction has committed (after completion to step 3: to commit a transaction)
The effects of the transaction will be preserved; no redo operations need to be invoked.

115. Advantages of Shadow Paging
no overhead of writing log records
recovery is trivial

116. Disadvantages of Shadow Paging
Copying the entire page table is very expensive
Commit overhead is high even with above extension
Need to flush every updated page, and page table

117. Disadvantages of Shadow Paging
Data gets fragmented (related pages get separated on disk)
After every transaction completion, the database pages containing old versions of modified data need to be garbage collected
Hard to extend algorithm to allow transactions to run concurrently
Easier to extend log based schemes

118. Stable-Storage Implementation
Maintain multiple copies of each block on separate disks
copies can be at remote sites to protect against disasters such as fire or flooding.

119. Stable-Storage Implementation
Failure during data transfer can still result in inconsistent copies: Block transfer can result in:
Successful completion
Partial failure: destination block has incorrect information
Total failure: destination block was never updated

120. Write the information onto the first physical block.
Stable-Storage Implementation Protecting storage media from failure during data transfer
Part I: Execute output operation as follows (assuming two copies of each block):
Write the information onto the first physical block.
When the first write successfully completes, write the same information onto the second physical block.
The output is completed only after the second write successfully completes.

121. Stable-Storage Implementation Protecting storage media from failure during data transfer
Part II: Copies of a block may differ due to failure during output operation. To recover from failure:
First find inconsistent blocks:
Expensive solution: Compare the two copies of every disk block.
Better solution:
Record in-progress disk writes on non-volatile storage (Non-volatile RAM or special area of disk).
Use this information during recovery to find blocks that may be inconsistent, and only compare copies of these.
If either copy of an inconsistent block is detected to have an error (bad checksum), overwrite it by the other copy. If both have no error, but are different, overwrite the second block by the first block.

122. Seatwork <T1 start> < T1 , D,20> < T1 commit>
[checkpoint]
<T4 start>
< T4, B,15>
< T4, A,20>
< T4 commit>
<T2 start>
< T2, B,12>
<T3 start>
< T3, A,30>
< T2 , D, 25> T1 T2 T3 T4
read(A)
read(B)
read(D)
write(B)
write(A)
write(D)
read(C)
write(C)
Describe the recovery process using deferred updates (of concurrent transactions) with check pointing
System crash

123. Seatwork <T1 start> < T1 , D,20> < T1 commit>
[checkpoint]
<T4 start>
< T4, B,15>
< T4, A,20>
< T4 commit>
<T2 start>
< T2, B,12>
<T3 start>
< T3, A,30>
< T2 , D, 25> T1 T2 T3 T4
read(A)
read(B)
read(D)
write(B)
write(A)
write(D)
read(C)
write(C)
Describe the recovery process from the system crash using immediate updates with check pointing.
Specify which transactions are rolled back, which operations in the log are redone and which (if any) are undone, and whether any cascading rollback takes place.
System crash

124. END