1. Transaction Management, Concurrency Control and Recovery Chapter 20 1

2. Overview 2  What are transactions?  What is a schedule?  What is concurrency control?  Why we need concurrency control:  Three problems.  Serializabiltiy and Concurrency control:  Theory:  Conflict Serializability  View Serializability  Practice:  Locking  Time-stamping  Optimistic techniques  Recovery facilities

3. What is a Transaction? 3 Transaction Action, or series of actions, carried out by user or application, which accesses or changes contents of database.  Logical unit of work on the database.  Transforms database from one consistent state to another, although consistency may be violated during transaction.  Example: Read(staffNo, salary) salary=salary * 1.1 write(staffNo, salary)

4. What is a Transaction? 4  Can have one of two outcomes:  Success - transaction commits and database reaches a new consistent state.  Failure - transaction aborts, and database must be restored to consistent state before it started (rolled back or undone).  Committed transaction cannot be aborted.  Aborted transactions that are rolled back can be restarted later.

5. Properties of Transactions 5 Four basic (ACID) properties of a transaction are: Atomicity‘All or nothing’ property. ConsistencyMust transform database from one consistent state to another. IsolationPartial effects of incomplete transactions should not be visible to other transactions. DurabilityEffects of a committed transaction are permanent and must not be lost because of later failure.  We deal with transactions in a schedule.

6. Schedule 6 TimeT1T1 T2T2 T3T3 Balance1Balance2 t0t0 Begin Transaction100200 t1t1 Read(Balance1) Begin Transaction 100200 t2t2 Read(Balance1) Begin Transaction 100200 t3t3 Balance1 += 500Read(Balance2)100200 t4t4 Write(Balance1)600200 t5t5 CommitRead(Balance1)600200 :::::: Order of execution Running Transactions Data Items affected by transactions (optional) Start with t 0 or t 1

7. Schedule Rules 7  Never start two transactions at the same time.  Never perform Reads and Writes of different transactions at the same time.  Each transaction should end with a commit or abort (rollback).

8. Schedule Definitions 8 Schedule Sequence of reads/writes by set of concurrent transactions. Serial Schedule Schedule where operations of each transaction are executed consecutively without any interleaved operations from other transactions.  No guarantee that results of all serial executions of a given set of transactions will be identical. (Think of an example) Non-Serial Schedule Schedule where operations from set of concurrent transactions are interleaved

9. Example of a Serial Schedule 9 TimeT1T1 T2T2 t0t0 Begin Transaction t1t1 Read(Balance1) t2t2 Balance1 += 500 t3t3 Commit t4t4 Begin Transaction t5t5 Read(Balance1) t6t6 Commit

10. Example of a non-Serial Schedule 10 TimeT1T1 T2T2 t0t0 Begin Transaction t1t1 t2t2 Read(Balance1) t3t3 t4t4 Commit t5t5 Balance1 += 500 t6t6 Commit

11. What is Concurrency Control? 11 Concurrency: transactions running simultaneously. Concurrency Control: Process of managing simultaneous operations (transactions) on the database without having them interfere with one another.  Prevents interference when two or more users are accessing database simultaneously and at least one is updating data.  Although two transactions may be correct in themselves, interleaving of operations may produce an incorrect result.

12. Why we Need Concurrency Control? 12 Three examples of potential problems caused by concurrency:  Lost update problem.  Uncommitted dependency problem.  Inconsistent analysis problem.

13. Lost Update Problem 13  Successfully completed update is overridden by another user.  T 1 withdrawing £10 from an account with bal x, initially £100.  T 2 depositing £100 into same account.  Serially, final balance would be £190.  Loss of T 2 ’s update avoided by preventing T 1 from reading bal x until after update TimeT 1 T 2 bal x t 1 begin-transaction100 t 2 begin-transaction read(bal x )100 t 3 read(bal x ) bal x = bal x +100100 t 4 bal x = bal x -10 write(bal x )200 t 5 write(bal x ) commit90 t 6 commit 90

14. Uncommitted Dependency Problem 14  Occurs when one transaction can see intermediate results of another transaction before it has committed.  T 4 updates bal x to £200 but it aborts, so bal x should be back at original value of £100.  T 3 has read new value of bal x (£200) and uses value as basis of £10 reduction, giving a new balance of £190, instead of £90.  Problem avoided by preventing T 3 from reading bal x until after T 4 commits or aborts. TimeT 3 T 4 bal x t 1 begin-transaction100 t 2 read(bal x )100 t 3 bal x = bal x +100100 t 4 begin_transaction write(bal x ) 200 t 5 read(bal x ) :200 t 6 bal x = bal x -10 rollback100 t 7 write(bal x ) 190 t 8 commit 190

15. Inconsistent Analysis Problem 15  Occurs when transaction reads several values but second transaction updates some of them during execution of first.  Sometimes referred to as dirty read or unrepeatable read.  T 6 is totaling balances of account x (£100), account y (£50), and account z (£25).  Meantime, T 5 has transferred £10 from bal x to bal z, so T 6 now has wrong result (£10 too high).

16. Inconsistent Analysis Problem 16  Problem avoided by preventing T 6 from reading bal x and bal z until after T 5 completed updates.

17. Serializability 17  Serializability is a property of a schedule:  We say serializable schedule and non-serializable schedule.  But what makes a schedule serializable?  A serializable schedule is a non-serial schedule that allows transactions to execute concurrently without interfering with one another.  In other words, a non-serial schedule that is equivalent to some serial schedule.  Main goal is to prevent transactions interfering with each other (3 problems discussed earlier).

18. Serializability 18  Two types of seriailizability:  Conflict.  View. Serializability TheoryPractice Conflict Serializability Test View Serializability Test Locking Time- stamping Optimistic

19. Conflict Serializability 19  In serializability, ordering of read/writes is important: (a) If two transactions only read a data item, they do not conflict and order is not important. (b) If two transactions either read or write completely separate data items, they do not conflict and order is not important. (c) If one transaction writes a data item and another reads or writes same data item, order of execution is important. They conflict.

20. Conflict Serializability 20  Schedule S 1 is conflict serializable if it is conflict equivalent to a serial schedule.  Two ways of testing a schedule for conflict serialiazibility: 1. A schedule is conflict serializable if you can switch order of 2 non-conflicting operations until you reach a serial schedule. 2. Precedence graph.

21. Testing for Conflict Serializability 21 Time T 7 T 8 t 1 begin-transaction t 2 read(bal x ) t 3 write(bal x ) t 4 begin_transaction t 5 read(bal x ) t 6 write(bal x ) t 7 read(baly) t 8 write(baly) t 9 commit t 10 read(baly) t 11 write(baly) t 12 commit T 7 T 8 begin-transaction read(bal x ) write(bal x ) begin_transaction read(bal x ) read(baly) write(bal x ) write(baly) commit read(baly) write(baly) commit

22. Testing for Conflict Serializability 22 Time T 7 T 8 t 1 begin-transaction t 2 read(bal x ) t 3 write(bal x ) t 4 begin_transaction t 5 read(bal y ) t 6 read(balx) t 7 write(bal x ) t 8 write(baly) t 9 commit t 10 read(baly) t 11 write(baly) t 12 commit T 7 T 8 begin-transaction read(bal x ) write(bal x ) read(baly) write(baly) commit begin_transaction read(bal x ) write(bal x ) read(baly) write(baly) commit

23. Non-conflict Serializable Schedule 23 Time T 7 T 8 t 1 begin-transaction t 2 read(bal x ) t 3 begin_transaction t 4 write(bal x ) t 5 write(balx) t 6 commit t 7 commit T 7 T 8 begin-transaction read(bal x ) write(balx) commit begin_transaction write(bal x ) commit

24. Testing for Conflict Serializability – Precedence Graph 24  Create:  node for each transaction;  a directed edge T i  T j, if T j reads the value of an item written by T i ;  a directed edge T i  T j, if T j writes a value into an item after it has been read by T i.  a directed edge T i  T j, if T j writes a value into an item after it has been written by T i.  If precedence graph contains cycle, schedule is not conflict serializable.

25. Test Schedule: Is it conflict serializable? 25 Time T 7 T 8 t 1 begin-transaction t 2 read(bal x ) t 3 bal x = bal x +100 t 4 write(bal x ) t 5 begin_transaction t 6 read(balx) t 7 bal x = bal x * 1.1 t 8 write(bal x ) t 9 read(baly) t 10 bal y = bal y * 1.1 t 11 write(bal y ) t 12 commit t 13 read(baly) t 14 write(baly) t 15 commit

26. View Serializability 26  Offers less stringent definition of schedule equivalence than conflict serializability.  Two schedules S 1 and S 2 are view equivalent if:  For each data item x, if T i reads initial value of x in S 1, T i must also read initial value of x in S 2.  For each read on x by T i in S 1, if value read by x is written by T j, T i must also read value of x produced by T j in S 2.  For each data item x, if last write on x performed by T i in S 1, same transaction must perform final write on x in S 2.

27. View Serializability 27  Schedule is view serializable if it is view equivalent to a serial schedule.  Every conflict serializable schedule is view serializable, although converse is not true.  It can be shown that any view serializable schedule that is not conflict serializable contains one or more blind writes. All Schedules View Serializable Schedules Conflict Serializable Schedules

28. View Serializable Schedule 28 Time T 7 T 8 t 1 begin-transaction t 2 read(bal x ) t 3 write(bal x ) t 4 read(bal y ) t 5 write(baly) t 6 commit t 7 begin-transaction t 8 read(balx) t 9 write(bal x ) t 10 read(baly) t 11 write(baly) t 12 commit T 7 T 8 begin-transaction read(bal x ) write(bal x ) begin_transaction read(balx) write(balx) read(baly) write(baly) commit read(baly) write(baly) commit

29. View Serializable Schedule 29 Time T 11 T 12 T 13 t 1 begin-transaction t 2 read(bal x ) t 3 begin_transaction t 4 write(bal x ) t 5 commit t 6 write(bal x ) t 7 commit t 8 begin_transaction t 9 write(bal x ) t 10 commit Is this schedule conflict serializable?

30. Recoverable Schedule 30  A schedule where, for each pair of transactions T i and T j, if T j reads a data item previously written by T i, then the commit operation of T i precedes the commit operation of T j.

31. Concurrency Control Techniques 31 Serializability TheoryPractice Conflict Serializability Test View Serializability Test Locking Time- stamping Optimistic

32. Concurrency Control Techniques 32  Two basic concurrency control techniques:  Locking,  Timestamping.  Both are conservative approaches: delay transactions in case they conflict with other transactions.  Optimistic methods assume conflict is rare and only check for conflicts at commit.

33. Concurrency Control Techniques Overview 33 Locking Time- stamping Optimistic Basic Rules 2PL Deadlock Prevention Basic Time- stamp Ordering Multi-version Time-stamp Ordering Deadlock Detection Wait- Die Wound -Wait Wait-for Graph Thomas’s Write Rule Regular Strict Rigorous Time outs

34. Locking 34 Main Idea: Transaction uses locks to deny access to other transactions and so prevent incorrect updates.  Most widely used approach to ensure serializability.  A transaction must claim:  a shared (read) on x before it can read it.  or an exclusive (write) lock on x before it can write it.  Lock prevents other transactions from reading or writing the locked data item.

35. Locking – Basic Rules 35  Shared Lock:  If transaction has shared lock on item, it can read but not update item.  More than one transaction can hold a shared lock on an item.  Exclusive Lock:  If transaction has exclusive lock on item, can both read and update item.  Only one transaction can hold an exclusive lock on an item.  Some systems allow transaction to:  upgrade read lock to an exclusive lock.  downgrade exclusive lock to a shared lock.

36. Locking -- Commands 36  To acquire a shared (read) lock on X:  Read_Lock(x)  RLock(X)  Shared_Lock(X)  SLock(X)  To acquire an exclusive (write) lock on X:  Write_Lock(X)  WLock(X)  Exclusive_Lock(X)  XLock(X)  To release a lock on X:  Unlock(X)

37. Time T 9 T 10 t 1 begin-transaction t 2 write_lock(bal x ) t 3 read(bal x ) t 4 bal x = bal x + 100 t 5 write(bal x ) t 6 unlock(bal x ) t 7 begin_transaction t 8 write_lock(bal x ) t 9 read(balx) t 10 bal x = bal x * 1.1 t 11 write(bal x ) t 12 unlock(bal x ) t 13 write_lock(bal y ) t 14 read(baly) t 15 bal y = bal y * 1.1 t 16 write(bal y ) t 17 commit/unlock(bal y ) t 18 write_lock(bal y ) t 19 read(baly) t 20 bal y = bal y - 100 t 21 write(baly) t 22 commit/unlock(bal y ) Correct use of locks. But is the execution correct?

38. Two-Phase Locking (2PL) 38  We just saw that locking alone doesn’t always work.  Solution: 2PL. Transaction follows 2PL protocol if all locking operations precede first unlock operation in the transaction.  Two phases for transaction:  Growing phase - acquires all locks but cannot release any locks.  Shrinking phase - releases locks but cannot acquire any new locks.  With 2PL, we can prevent the three problems.

39. Original Lost Update Problem 39 TimeT 1 T 2 bal x t 1 begin-transaction100 t 2 begin-transaction read(bal x )100 t 3 read(bal x ) bal x = bal x +100100 t 4 bal x = bal x -10 write(bal x )200 t 5 write(bal x ) commit90 t 6 commit 90

40. Preventing Lost Update Problem 40 TimeT 1 T 2 bal x t 1 begin-transaction100 t 2 begin_transactionwrite_lock(bal x )100 t 3 write_lock(bal x )read(bal x )100 t 4 WAIT bal x = bal x +100100 t 5 WAIT write(bal x )200 t 6 WAIT commit/unlock(balx) 200 t 7 read(bal x )200 t 8 bal x = bal x -10200 t 9 write(bal x ) 190 t 10 commit/unlock(bal x ) 190

41. Original Uncommitted Dependency Problem 41 TimeT 3 T 4 bal x t 1 begin-transaction100 t 2 read(bal x )100 t 3 bal x = bal x +100100 t 4 begin_transaction write(bal x ) 200 t 5 read(bal x ) :200 t 6 bal x = bal x -10 rollback100 t 7 write(bal x ) 190 t 8 commit 190

42. Preventing Uncommitted Dependency Problem 42 TimeT 3 T 4 bal x t 1 begin-transaction100 t 2 write_lock(bal x )100 t 3 read(bal x )100 t 4 begin_transaction bal x = bal x +100100 t 5 write_lock(bal x ) write(bal x )200 t 6 WAIT commit/unlock(balx) 200 t 7 read(bal x )200 t 8 bal x = bal x -10200 t 9 write(bal x ) 190 t 10 commit/unlock(bal x ) 190

43. Original Inconsistent Analysis Problem 43

44. Preventing Inconsistent Analysis Problem 44

45. A Potential Problem with 2PL 45

46. Cascading Rollbacks 46  If every transaction in a schedule follows 2PL, schedule is serializable.  However, problems can occur with interpretation of when locks can be released.  Cascading rollback is undesirable since they potentially lead to the undoing of a significant amount of work  To prevent this with 2PL, 2 solutions: 1. Rigorous 2PL: Leave release of all locks until end of transaction. 2. Strict 2PL: Holds only exclusive locks until the end of the transaction.  BOTH are still 2PL. So both still have growing and shrinking phases.  2PL still may cause deadlock.

47. Problems with 2PL 47 1. Cascading Rollbacks:  Solved with strict or rigorous 2PL. 2. Dead Locks:  Happen in regular 2PL, and also in strict and rigorous 2PL.  Handled using deadlock detection and prevention techniques.

48. Deadlocks 48 Deadlock: An impasse that may result when two (or more) transactions are each waiting for locks held by the other to be released.  Once a deadlock happens, only one way to break deadlock: abort one or more of the transactions.  Deadlock should be transparent to user, so DBMS should restart aborted transaction(s).

49. Example Deadlock 49 Time T 9 t 1 begin-transaction t 2 write_lock(bal x ) t 3 read(bal x ) t 4 bal x = bal x - 10 t 5 write(bal x ) t 6 write_lock(bal y ) t 7 WAIT t 8 WAIT t 9 WAIT t 10 WAIT t 11 : T 10 begin-transaction write_lock(bal y ) read(bal y ) bal y = bal y + 100 write(bal y ) wait_lock(bal x ) WAIT :

50. Deadlock Handling 50  Two general techniques for handling deadlock:  Deadlock prevention: DBMS doesn’t allow deadlock to happen.  Timeouts.  Wait-Die.  Wound-wait.  Deadlock detection and recovery: DBMS allows deadlocks to happens but detects and recovers from them.  Wait-for Graphs (WFG).

51. Timeouts 51  Transaction that requests lock will only wait for a system- defined period of time.  If lock has not been granted within this period, lock request times out.  DBMS assumes transaction deadlocked, even though it may not be, and it aborts and automatically restarts the transaction.

52. Timestamps 52  Before we discuss Wait-die and Wound-wait techniques, introduce timestamps.  A timestamp is a unique number given to each transaction.  Traditionally, it is the time the transaction started.  The smaller the timestamp, the older the transaction.

53. Timestamps 53  TS(T 11 ) = 1  TS(T 12 ) = 3  TS(T 13 ) = 8 Time T 11 T 12 T 13 t 1 begin-transaction t 2 read(bal x ) t 3 begin_transaction t 4 write(bal x ) t 5 commit t 6 write(bal x ) t 7 commit t 8 begin_transaction t 9 write(bal x ) t 10 commit

54. Wait-Die Technique 54  Only an older transaction can wait for younger one, otherwise transaction is aborted (dies) and restarted with same timestamp. (Why the same?)  If a transaction T i requests a lock on an item held by T j :  If T i > T j [TS(T i ) < TS(T j )], T i waits for T j to release the lock.  If T i TS(T j )], T i is aborted and restarted with the same TS.

55. Wound-Wait Technique 55  only a younger transaction can wait for an older one. If older transaction requests lock held by younger one, younger one is aborted (wounded) and restarted with same timestamp. (Why the same?)  If a transaction T i requests a lock on an item held by T j :  If T i > T j [TS(T i ) < TS(T j )], T j is aborted and T i gets the lock.  If T i TS(T j )], T i waits for T j to release the lock.

56. Deadlock Detection and Recovery 56  Usually handled by construction of wait-for graph (WFG) showing transaction dependencies:  Create a node for each transaction.  Create edge T i  T j, if T i waiting to lock item locked by T j.  Deadlock exists if and only if WFG contains cycle.

57. Example Schedule with WFG 57 Time T 9 t 1 begin-transaction t 2 write_lock(bal x ) t 3 read(bal x ) t 4 bal x = bal x - 10 t 5 write(bal x ) t 6 write_lock(bal y ) t 7 WAIT t 8 WAIT t 9 WAIT t 10 WAIT t 11 : T 10 begin-transaction write_lock(bal y ) read(bal y ) bal y = bal y + 100 write(bal y ) wait_lock(bal x ) WAIT : T9T9 T 10

58. Deadlock Detection and Recovery 58  WFG is created at regular intervals.  Several issues when recovering from a deadlock:  choice of deadlock victim;  avoiding starvation.  Self-read: pages 596-597

59. Concurrency Control Techniques Overview 59 Locking Time- stamping Optimistic Basic Rules 2PL Deadlock Prevention Basic Time- stamp Ordering Multi-version Time-stamp Ordering Deadlock Detection Wait- Die Wound -Wait Wait-for Graph Thomas’s Write Rule Regular Strict Rigorous Time outs

60. Timestamping 60  Main Idea: Transactions ordered globally so that older transactions (smaller timestamps) get priority in the event of conflict.  Conflict is resolved by rolling back (aborting) and restarting transaction.  No locks so no deadlock. Timestamp A unique identifier created by DBMS that indicates relative starting time of a transaction. Timestamping A concurrency control protocol that orders transactions in such a way that order transactions. Transactions with smaller timestamps, get priority in the event of conflict

61. Timestamping 61  2 Techniques: 1. Basic Timestamp Ordering.  Thomas’s Write Rule. 2. Multiversion Timestamp Ordering.

62. Basic Timestamp Ordering 62  Read/write proceeds only if last update on that data item was carried out by an older transaction.  Otherwise, transaction requesting read/write is restarted and given a new timestamp.  Main Goal: Ordering writes then reads/writes as they would have been ordered in a serial schedule.  Timestamps are also set for data items:  read-timestamp - timestamp of last transaction to read item;  write-timestamp - timestamp of last transaction to write item.

63. Basic Timestamping – Read(x) 63  Consider a read(x) transaction T with timestamp TS(T): TS(T) < write_timestamp(x)  x already updated by younger (later) transaction.  Transaction T must be aborted and restarted with a new timestamp. TS(T)  write_timestamp(x)  execute the read(x) operation of T  read_timestamp(x) = TS(T)

64. Basic Timestamping – Write(x) 64 TS(T) < read_timestamp(x)  x already read by younger transaction.  Transaction T must be aborted and restarted with a new timestamp. TS(T) < write_timestamp(x)  x already written by younger transaction.  Transaction T must be aborted and restarted with a new timestamp. Otherwise, operation is accepted and executed. Write_timestamp(x) = TS(T)

65. Basic Timestamp Ordering 65

66. Thomas’s Write Rule 66  Provide greater concurrency by rejecting obsolete write operations.  When a read(x) is encountered, behave just like in slide 62.  When a write(x) is encountered, perform the following check: TS(T) < read_timestamp(x)  x already read by younger transaction.  Transaction T must be aborted and restarted with a new timestamp. TS(T) < write_timestamp(x)  x already written by younger transaction.  Ignores the write operation (ignore obsolete write rule ) Otherwise, operation is accepted and executed. Write_timestamp(x) = TS(T)

67. Comparison of Methods 67 All Schedules View Serializable Schedules Conflict Serializable Schedules 2PL TS

68. Multiversion Timestamp Ordering 68  Main Idea: Versioning of data can be used to increase concurrency so create multiple versions of each data item.  Basic timestamp assumes only one version of data item exists, and so only one transaction can access data item at a time.  Multiversion allows multiple transactions to read and write different versions of same data item.  Multiversion ensures each transaction sees consistent set of versions for all data items it accesses.  In multiversion:  Each write operation creates new version of data item while retaining old version.  When transaction attempts to read data item, system selects one version that ensures serializability  NO ABORTS ON READs  Each version has a read and a write timestamp.  Versions can be deleted once they are no longer required.

69. Multiversion Timestamping – Read(x) 69  When a transaction T wishes to read x, we find the correct version and let it read it.  The correct version, x i, is the latest version written by an older transaction:  TS(T)  write_timestamp(x i )  After x i is found and read by T, we need to record that x i was read by T:  read_timestamp(x i ) = max(read_timestamp(x i ), TS(T))  Absolutely no aborts on read.

70. Multiversion Timestamping – Write(x) 70  When a transaction T wishes to write x, we need to perform a test first then write a new version.  Test: we need make sure that there is no older version of x that has been read by a transaction younger than T  The transaction that is younger than T should read T’s version, not this older version.

71. Multiversion Timestamping – Write(x) 71 1. Find the correct version, x i : is the latest version written by an older transaction:  TS(T)  write_timestamp(x i ) 2. Test it: make sure that no younger transaction has already read x i :  TS(T) < read_timestamp(x i )?  If yes: Abort T.  If no: create a new version x j of x: read_timestamp(x j ) = write_timestamp(x j ) = TS(T)

72. 72 TimeT1T1 T2T2 T3T3 T4T4 T5T5 0 Begin 1 2 Read(x) 3 x=x+20 4 Write(x) 5 Begin 6 Read(x) 7 Begin 8 Read(x) 9 Read(y) 10 Read(y) 11 Begin 12 Read(y) 13 y=y/2 14 Write(y) 15 y=y+x-100 16 Write(y) 17 x=x+0.10x 18 Write(x) 19 Commit 20 Commit 21 Commit 22 Commit 23 Commit

73. Concurrency Control Techniques Overview 73 Locking Time- stamping Optimistic Basic Rules 2PL Deadlock Prevention Basic Time- stamp Ordering Multi-version Time-stamp Ordering Deadlock Detection Wait- Die Wound -Wait Wait-for Graph Thomas’s Write Rule Regular Strict Rigorous Time outs

74. Optimistic Techniques 74  Main Idea: conflict is rare and it is more efficient to let transactions proceed without delays to ensure serializability.  At commit, check is made to determine whether conflict has occurred.  If there is a conflict, transaction must be rolled back and restarted.  Potentially allows greater concurrency than traditional protocols.  Three phases:  Read.  Validation.  Write.

75. Optimistic Techniques – Read Phase 75  Extends from start until immediately before commit.  Transaction reads values from database and stores them in local variables. Updates are applied to a local copy of the data.  DB is not changed during read phase.

76. Optimistic Techniques – Validation Phase 76  Follows the read phase – just before the transaction commits.  For read-only transaction:  check that data read are still current values.  If no interference, transaction is committed.  Else, transaction is aborted and restarted.  For update transaction:  check transaction leaves database in a consistent state, with serializability maintained.

77. Optimistic Techniques – Write Phase 77  Follows successful validation phase for update transactions.  Updates made to local copy are applied to the database.

78. Database Recovery 78  Process of restoring database to a correct state in the event of a failure.  Transactions represent basic unit of recovery.  Recovery manager responsible for atomicity and durability.  If failure occurs between commit and database buffers being flushed to secondary storage then, to ensure durability, recovery manager has to redo (rollforward) transaction’s updates.  If transaction had not committed at failure time, recovery manager has to undo (rollback) any effects of that transaction for atomicity.  Partial undo - only one transaction has to be undone.  Global undo - all transactions have to be undone.

79. Example 79 T1T2T3T4T5T6T1T2T3T4T5T6 t 0 t c t f  DBMS starts at time t 0, but fails at time t f. Assume data for transactions T 2 and T 3 have been written to secondary storage.  T 1 and T 6 have to be undone. In absence of any other information, recovery manager has to redo T 2, T 3, T 4, and T 5.

80. Recovery Facilities 80  DBMS should provide following facilities to assist with recovery: 1. Backup mechanism: which makes periodic backup copies of database. 2. Logging facilities: which keep track of current state of transactions and database changes. 3. Checkpoint facility: which enables updates to database in progress to be made permanent. 4. Recovery manager: which allows DBMS to restore database to consistent state following a failure.

81. 2. Logging Facilities: The Log File 81  Contains information about all updates to database:  Transaction records.  Checkpoint records.  Often used for other purposes (for example, auditing).

82. 3. Checkpoint Facility 82 Checkpoint Point of synchronization between database and log file. All buffers are written to secondary storage.  A checkpoint consists of the following actions: 1. Suspend execution of transactions temporarily. 2. Write all updated buffers to disk. 3. Write a checkpoint log record to disk. 4. Resume executing transactions.  When failure occurs, the recovery manager performs the following:  Redo all transactions that committed since the checkpoint.  Undo all transactions active at time of crash (if using immediate update).

83. Example 83 T1T2T3T4T5T6T1T2T3T4T5T6 t 0 t c t f  T 1 and T 6 : undo.  T 2 and T 3 : do nothing.  T 4 and T 5 : redo.

84. Checkpoint in Log File 84

85. Main Recovery Techniques 85  Three main recovery techniques:  Deferred Update.  Immediate Update.  Shadow Paging.

86. Deferred Update 86  Updates are not written to the database until after a transaction has reached its commit point.  Start from the last checkpoint:  If a transaction has committed before checkpoint  Do nothing.  If a transaction has committed after checkpoint  Redo it.  If a transaction has not committed after checkpoint  Do nothing.

87. Immediate Update 87  Updates are applied to database as they occur.  Start from the last checkpoint:  If a transaction has committed before checkpoint  Do nothing.  If a transaction has committed after checkpoint  Redo it.  If a transaction has not committed after checkpoint  Undo it.  Undo is done in reverse order: from bottom of log file to top.  Redo is done in order: from top of log file to bottom.

88. Example 88  Using Deferred Update: Which transactions to undo? Redo?  Using Immediate Update: Which transactions to undo? Redo?

89. Shadow Paging 89  Maintain two page tables during life of a transaction:  Current page table.  Shadow page table.  When transaction starts, two tables are the same.  Shadow page table is never changed thereafter and is used to restore database in event of failure.  During transaction, current page table records all updates to database.  When transaction completes, current page table becomes shadow page table.