Transaction Processing Recovery & Concurrency Control

What is a transaction

A transaction is the basic logical unit of execution in an information system. A transaction is a sequence of operations that must be executed as a whole, taking a consistent (& correct) database state into another consistent (& correct) database state; A collection of actions that make consistent transformations of system states while preserving system consistency An indivisible unit of processing
database in a consistent state database in a consistent state

Account A Fred Bloggs 1000 Account B Sue Smith 0

begin Transaction

Transfer 500

Account A Fred Bloggs 500 Account B Sue Smith 500

execution of Transaction database may be temporarily in an inconsistent state during execution

end Transaction

Desirable Properties of ACID Transactions

A Atomicity: a transaction is an atomic unit of processing and it is either performed entirely or not at all C Consistency Preservation: a transaction's correct execution must take the database from one correct state to another I Isolation/Independence: the updates of a transaction must not be made visible to other transactions until it is committed (solves the temporary update problem) D Durability (or Permanency): if a transaction changes the database and is committed, the changes must never be lost because of subsequent failure o Serialisability: transactions are considered serialisable if the effect of running them in an interleaved fashion is equivalent to running them serially in some order

Requirements for Database Consistency

Concurrency Control

Most DBMS are multi-user systems. The concurrent execution of many different transactions submitted by various users must be organised such that each transaction does not interfere with another transaction with one another in a way that produces incorrect results. The concurrent execution of transactions must be such that each transaction appears to execute in isolation.

Recovery

System failures, either hardware or software, must not result in an inconsistent database

Transaction as a Recovery Unit

If an error or hardware/software crash occurs between the begin and end, the database will be inconsistent

Computer Failure (system crash) A transaction or system error Local errors or exception conditions detected by the transaction Concurrency control enforcement Disk failure Physical problems and catastrophes

The database is restored to some state from the past so that a correct stateclose to the time of failurecan be reconstructed from the past state. A DBMS ensures that if a transaction executes some updates and then a failure occurs before the transaction reaches normal termination, then those updates are undone. The statements COMMIT and ROLLBACK (or their equivalent) ensure Transaction Atomicity

Recovery

Mirroring

keep two copies of the database and maintain them simultaneously

Backup

periodically dump the complete state of the database to some form of tertiary storage

System Logging

the log keeps track of all transaction operations affecting the values of database items. The log is kept on disk so that it is not affected by failures except for disk and catastrophic failures.

Recovery from Transaction Failures

Catastrophic failure

Restore a previous copy of the database from archival backup Apply transaction log to copy to reconstruct more current state by redoing committed transaction operations up to failure point Incremental dump + log each transaction

Non-catastrophic failure

Reverse the changes that caused the inconsistency by undoing the operations and possibly redoing legitimate changes which were lost The entries kept in the system log are consulted during recovery. No need to use the complete archival copy of the database.

Transaction States

For recovery purposes the system needs to keep track of when a transaction starts, terminates and commits. Begin_Transaction: marks the beginning of a transaction execution; End_Transaction: specifies that the read and write operations have ended and
marks the end limit of transaction execution (but may be aborted because of concurrency control); Commit_Transaction: signals a successful end of the transaction. Any updates executed by the transaction can be safely committed to the database and will not be undone; Rollback (or Abort): signals that the transaction has ended unsuccessfully. Any changes that the transaction may have applied to the database must be undone; Undo: similar to ROLLBACK but it applies to a single operation rather than to a whole transaction; Redo: specifies that certain transaction operations must be redone to ensure that all the operations of a committed transaction have been applied successfully to the database;

Entries in the System Log

For every transaction a unique transaction-id is generated by the system. [start_transaction, transaction-id]: the start of execution of the transaction identified by transaction-id

Credit_labmark (sno NUMBER, cno CHAR, credit NUMBER) old_mark NUMBER; new_mark NUMBER; SELECT labmark INTO old_mark FROM enrol WHERE studno = sno and courseno = cno FOR UPDATE OF labmark; new_ mark := old_ mark + credit; UPDATE enrol SET labmark = new_mark WHERE studno = sno and courseno = cno ; COMMIT; EXCEPTION WHEN OTHERS THEN ROLLBACK; END credit_labmark;

[read_item, transaction-id, X]: the transaction identified by transaction-id reads the value of database item X. Optional in some protocols. [write_item, transaction-id, X, old_value, new_value]: the transaction identified by transaction-id changes the value of database item X from old_value to new_value

[commit, transaction-id]: the transaction identified by transaction-id has completed all accesses to the database successfully and its effect can be recorded permanently (committed) [abort, transaction-id]: the transaction identified by transaction-id has been aborted

Transaction execution
A transaction reaches its commit point when all operations accessing the database are completed and the result has been recorded in the log. It then writes a [commit, transaction-id].
BEGIN TRANSACTION active END TRANSACTION partially committed ROLLBACK

COMMIT committed

READ, WRITE

ROLLBACK

failed

terminated

If a system failure occurs, searching the log and rollback the transactions that have written into the log a [start_transaction, transaction-id] [write_item, transaction-id, X, old_value, new_value] but have not recorded into the log a [commit, transaction-id]

Read and Write Operations of a Transaction

Specify read or write operations on the database items that are executed as part of a transaction read_item(X):

reads a database item named X into a program variable also named X. 1. find the address of the disk block that contains item X 2. copy that disk block into a buffer in the main memory 3. copy item X from the buffer to the program variable named
writes the value of program variable X into the database item named X. 1. find the address of the disk block that contains item X 2. copy that disk block into a buffer in the main memory 3. copy item X from the program variable named X into its current location in the buffer store the updated block in the buffer back to disk (this step updates the database on disk)

write_item(X):

X:=

Checkpoints in the System Log

A [checkpoint] record is written periodically into the log when the system writes out to the database on disk the effect of all WRITE operations of committed transactions. All transactions whose [commit, transaction-id] entries can be found in the system log will not require their WRITE operations to be redone in the case of a system crash. Before a transaction reaches commit point, forcewrite or flush the log file to disk before commit transaction. Actions Constituting a Checkpoint

data

temporary suspension of transaction execution forced writing of all updated database blocks in main memory buffers to disk writing a [checkpoint] record to the log and force writing the log to disk resuming of transaction execution

log

Write Ahead Logging In place updating protocols: Overwriting data in situ Immediate Update: Deferred Update:

no actual update of the database until after a transaction reaches its commit point

the database may be updated by some operations of a transaction before it reaches its commit point.

1. Updates recorded in log 2. Transaction commit point 3. Force log to the disk 4. Update the database FAILURE! REDO database from log entries No UNDO necessary because database never altered

1. Update X recorded in log 2. Update X in database 3. Update Y recorded in log 4. Transaction commit point 3. Force log to the disk 4. Update Y in database

FAILURE! UNDO X FAILURE! REDO Y

Undo in reverse order in log Redo in committed log order uses the write_item log entry

Transaction as a Concurrency Unit

Transactions must be synchronised correctly to guarantee database consistency T1

Transfer 500 from A to B Account A Fred Bloggs 500 Account B Sue Smith 500

Simultaneous Execution

Account B Sue Smith 0

Account A Fred Bloggs 1000

Account C Jill Jones 700

T2
Transfer 300 from C to A

Account A Fred Bloggs 800 Account C Jill Jones 400

Net result Account A 800 Account B 500 Account C 400

Transaction scheduling algorithms

Transaction Serialisability

The effect on a database of any number of transactions executing in parallel must be the same as if they were executed one after another

Problems due to the Concurrent Execution of Transactions

The Lost Update Problem The Incorrect Summary or Unrepeatable Read Problem The Temporary Update (Dirty Read) Problem

The Lost Update Problem

Two transactions accessing the same database item have their operations interleaved in a way that makes the database item incorrect
T1: (joe) read_item(X); X:= X - N; read_item(X); X:= X + M; write_item(X); read_item(Y); write_item(X); Y:= Y + N; write_item(Y); 2 8 7 10 10 T2: (fred) X 4 2 4 7 Y

X=4 Y=8 N=2 M=3

item X has incorrect value because its update from T1 is lost (overwritten) T2 reads the value of X before T1 changes it in the database and hence the updated database value resulting from T1 is lost

The Incorrect Summary or Unrepeatable Read Problem

One transaction is calculating an aggregate summary function on a number of records while other transactions are updating some of these records. The aggregate function may calculate some values before they are updated and others after.
T1: T2: sum:= 0; read_item(A); sum:= sum + A; . . read_item(X); sum:= sum + X; read_item(Y); sum:= sum + Y; read_item(Y); Y:= Y + N; write_item(Y); 8 10 10 T1 T2 4 4 4 2 2 2 6 8 14 Sum 0

T2 reads X after N is subtracted and reads Y before N is added, so a wrong summary is the result

read_item(X);. X:= X - N; write_item(X);

Dirty Read or The Temporary Update Problem

One transaction updates a database item and then the transaction fails. The updated item is accessed by another transaction before it is changed back to its original value
T1: (joe) T2: (fred) Database 4 2 2 read_item(X); X:= X- N; write_item(X); failed write (X) 4 Log old 4 Log new 2

Joe books seat on flight X

read_item(X); X:= X - N; write_item(X);

Joe cancels

2 -1 -1 2 -1 rollback T1 log

Fred books seat on flight X because Joe was on Flight X

transaction T1 fails and must change the value of X back to its old value meanwhile T2 has read the temporary incorrect value of X

Schedules of Transactions

A schedule S of n transactions is a sequential ordering of the operations of the n transactions.

T1 read x write x T2 read x write x S read x read x write x write x

The transactions are interleaved

A schedule maintains the order of operations within the individual transaction.

For each transaction T if operation a is performed in T before operation b, then operation a will be performed before operation b in S. The operations are in the same order as they were before the transactions were interleaved

Two operations conflict if they belong to different transactions, AND access the same data item AND one of them is a write.

Serial and Non-serial Schedules

A schedule S is serial if, for every transaction T participating in the schedule, all of T's operations are executed consecutively in the schedule; otherwise it is called non-serial. Non-serial schedules mean that transactions are interleaved. There are many possible orders or schedules. Serialisability theory attempts to determine the 'correctness' of the schedules. A schedule S of n transactions is serialisable if it is equivalent to some serial schedule of the same n transactions.

Example of Serial Schedules

Schedule A
T2:

Schedule B
T1: T2: read_item(X); X:= X + M; write_item(X);

T1: read_item(X); X:= X - N; write_item(X); read_item(Y); Y:=Y + N; write_item(Y);

read_item(X); X:= X + M; write_item(X);

read_item(X); X:= X - N; write_item(X); read_item(Y); Y:=Y + N; write_item(Y);

Example of Non-serial Schedules

Schedule C
T2: read_item(X); X:= X + M;

Schedule D
T1: read_item(X); X:= X - N; write_item(X); T2:

T1: read_item(X); X:= X - N; write_item(X); read_item(Y);

read_item(X); X:= X + M; write_item(X); read_item(Y); Y:=Y + N; write_item(Y);

write_item(X); Y:=Y + N; write_item(Y);

We have to figure out whether a schedule is equivalent to a serial schedule i.e. the reads and writes are in the right order

Precedence graphs (assuming read X before write X)

T1: read_item(X); X:= X - N; write_item(X); read_item(Y); Y:=Y + N; write_item(Y); T2:

T1:

T2: read_item(X); X:= X + M; write_item(X);

read_item(X); X:= X + M; write_item(X);

read_item(X); X:= X - N; write_item(X); read_item(Y); Y:=Y + N; write_item(Y);

T1: read_item(X); X:= X - N; write_item(X); read_item(Y);

T2: read_item(X); X:= X + M; write_item(X);

T1: read_item(X); X:= X - N; write_item(X);

T2:

read_item(X); X:= X + M; write_item(X); read_item(Y); Y:=Y + N; write_item(Y);

Y:=Y + N; write_item(Y);

View Equivalence and View Serialisability

View Equivalence:

As long as each read operation of a transaction reads the result of the same write operation in both schedules, the write operations of each transaction must produce the same results. The read operations are said to see the same view in both schedules

The final write operation on each data item is the same in both schedules, so the database state should be the same at the end of both schedules

A schedule S is view serialisable if it is view equivalent to a serial schedule. Testing for view serialisability is NP-complete

Semantic Serialisability

Some applications can produce schedules that are correct but arent conflict or view serialisable. e.g. Debit/Credit transactions (Addition and subtraction are commutative)
T2 read_item(Y); Y:=Y-20; write_item(Y); read_item(Z); Z:+Z+20; write_item(Z);

T1 read_item(X); X:=X-10; write_item(X); read_item(Y); Y:=Y+10; write_item(Y);

Schedule
T1 read_item(X); X:=X-10; write_item(X); T2

read_item(Y); Y:=Y-20; write_item(Y); read_item(Y); Y:=Y+10; write_item(Y);

Methods for Serialisability

Multi-version Concurrency Control techniques keep the old values of a data item when that item is updated. Timestamps are unique identifiers for each transaction and are generated by the system. Transactions can then be ordered according to their timestamps to ensure serialisability. Protocols that, if followed by every transaction, will ensure serialisability of all schedules in which the transactions participate. They may use locking techniques of data items to prevent multiple transactions from accessing items concurrently. Pessimistic Concurrency Control

Check before a database operation is executed by locking data items before they are read and written or checking timestamps

Locking Techniques for Concurrency Control

The concept of locking data items is one of the main techniques used for controlling the concurrent execution of transactions. A lock is a variable associated with a data item in the database. Generally there is a lock for each data item in the database. A lock describes the status of the data item with respect to possible operations that can be applied to that item. It is used for synchronising the access by concurrent transactions to the database items. A transaction locks an object before using it When an object is locked by another transaction, the requesting transaction must wait

Types of Locks

Binary locks have two possible states:

1. locked (lock_item(X) operation) and 2. unlocked (unlock_item(X) operation

Multiple-mode locks allow concurrent access to the same item by several transactions. Three possible states:
1. read locked or shared locked (other transactions are allowed to read the item) 2. write locked or exclusive locked (a single transaction exclusively holds the lock on the item) and 3. unlocked.

Locks are held in a lock table.

upgrade lock: read lock to write lock downgrade lock: write lock to read lock

Locks dont guarantee serialisability: Lost Update

T1: (joe) write_lock(X) read_item(X); X:= X - N; unlock(X) write_lock(X) read_item(X); X:= X + M; unlock(X) write_lock(X) write_item(X); unlock(X) write_lock(Y) read_item(Y); T2: (fred) X Y

4 2 4 7 2 8 write_lock(X) write_item(X); 7 unlock(X)

Y:= Y + N; write_item(Y); unlock(Y)

10 10

Locks dont guarantee serialisability

X=20, Y=30 T1 read_lock(Y); read_item(Y); unlock(Y); write_lock(X); read_item(X); X:=X+Y; write_item(X); unlock(X); T2 read_lock(X); read_item(X); unlock(X); write_lock(Y); read_item(Y); Y:=X+Y; write_item(Y); unlock(Y);
X is unlocked too early

Y is unlocked too early

Schedule 1: T1 followed by T2 X=50, Y=80 Schedule 2: T2 followed by T1 X=70, Y=50

Non-serialisable schedule S that uses locks

X=20 Y=30
T1 read_lock(Y); read_item(Y); unlock(Y); T2

read_lock(X); read_item(X); unlock(X); write_lock(Y); read_item(Y); Y:=X+Y; write_item(Y); unlock(Y); write_lock(X); read_item(X); X:=X+Y; write_item(X); unlock(X);

result of S X=50, Y=50

Ensuring Serialisability: Two-Phase Locking

All locking operations (read_lock, write_lock) precede the first unlock operation in the transactions. Two phases:

expanding phase: new locks on items can be acquired but none can be released shrinking phase: existing locks can be released but no new ones can be acquired X=20, Y=30
T1 read_lock(Y); read_item(Y); write_lock(X); unlock(Y); read_item(X); X:=X+Y; write_item(X); unlock(X); T2 read_lock(X); read_item(X); write_lock(Y); unlock(X); read_item(Y); Y:=X+Y; write_item(Y); unlock(Y);

Two-Phasing Locking

Basic 2PL

When a transaction releases a lock, it may not request another lock lock point number of locks obtain lock release lock

Phase 1 BEGIN

Phase 2 END

Conservative 2PL or static 2PL

a transaction locks all the items it accesses before the transaction begins execution pre-declaring read and write sets

Two-Phasing Locking

Strict 2PL a transaction does not release any of its locks until after it commits or aborts leads to a strict schedule for recovery
obtain lock number of locks release lock

BEGIN

period of data END item use

Transaction duration

Locking Problems: Deadlock

Each of two or more transactions is waiting for the other to release an item. Also called a deadly embrace

T1 read_lock(Y); read_item(Y); write_lock(X);

T2 read_lock(X); read_item(X); write_lock(Y);

Deadlocks and Livelocks

Deadlock prevention protocol:

conservative 2PL transaction stamping (younger transactions aborted) no waiting cautious waiting time outs

Deadlock detection (if the transaction load is light or transactions are short and lock only a few items)

wait-for graph for deadlock detection victim selection cyclic restarts

T1

T2

Livelock: a transaction cannot proceed for an indefinite period of time while other transactions in the system continue normally.

fair waiting schemes (i.e. first-come-first-served)

Locking Granularity

A database item could be

a database record a field value of a database record a disk block the whole database coarse granularity the larger the data item size, the lower the degree of concurrency fine granularity the smaller the data item size, the more locks to be managed and stored, and the more lock/unlock operations needed.

Trade-offs

Other Recovery and Concurrency Strategies

Recovery: Shadow Paging Technique

Data isnt updated in place The database is considered to be made up of a number of n fixed-size disk blocks or pages, for recovery purposes. A page table with n entries is constructed where the ith page table entry points to the ith database page on disk. Current page table points to most recent current database pages on disk

Database data pages/blocks page 5 Page table

1 2 3 4 5 6

page 1 page 4

page 2
page 3 page 6

Shadow Paging Technique

When a transaction begins executing

Database data pages (blocks) Current page table (after updating pages 2,6) page 5 (old) page 1 page 4 1 2 3 4 5 6 page 2 (old) page 3

the current page table is copied into a shadow page table shadow page table is then saved shadow page table is never modified during transaction execution writes operationsnew copy of database page is created and current page table entry modified to point to new disk page/block

Shadow page table (not updated)

1 2 3 4 5 6

page 6
page 2 (new) page 5 (new)

Shadow Paging Technique

To recover from a failure

Database data pages (blocks) Current page table (after updating pages 2,6) page 5 (old) page 1 page 4 1 2 3 4 5 6 page 2 (old) page 3

the state of the database before transaction execution is available through the shadow page table free modified pages discard currrent page table that state is recovered by reinstating the shadow page table to become the current page table once more

S adowpagetabe h l (not updated)

1 2 3 4 5 6

Commiting a transaction

page 6 page 2 (new) page 5 (new)

discard previous shadow page free old page tables that it references

Garbage collection

Optimistic Concurrency Control

No checking while the transaction is executing. Check for conflicts after the transaction. Checks are all made at once, so low transaction execution overhead Relies on little interference between transactions

Updates are not applied until end_transaction Updates are applied to local copies in a transaction space.

1. read phase: read from the database, but updates are applied only to local copies 2. validation phase: check to ensure serialisability will not be validated if the transaction updates are actually applied to the database 3. write phase: if validation is successful, transaction updates applied to database; otherwise updates are discarded and transaction is aborted and restarted.

Validation Phase

Use transaction timestamps write_sets and read_sets maintained Transaction B is committed or in its validation phase Validation Phase for Transaction A To check that TransA does not interfere with TransB the following must hold:

TransB completes its write phase before TransA starts its reads phase TransA starts its write phase after TransB completes its write phase, and the read set of TransA has no items in common with the write set of TransB Both the read set and the write set of TransA have no items in common with the write set of TransB, and TransB completes its read phase before TransA completes its read phase.

Conclusions

Transaction management deals with two key requirements of any database system: Resilience

in the ability of data surviving hardware crashes and software errors without sustaining loss or becoming inconsistent in the ability to permit simultaneous access of data multiple users in a consistent manner and assuring only authorised access

Access Control