Management Transaction in

Management Transaction in
Overview
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q Summary & comparison
q Transactions in OO-DBMS
o Workflow management
o Cooperative transactions
o Sagas
q Long-lived transactions
q Distributed transactions
q Open nested transactions
q Multilevel transactions
q Nested transactions
q Chained transactions
q Flat transactions with savepoints
q Spheres of Control
q Some basics
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
INF4180-transact-3
INF4180-transact-1
In addition: Gray, J., Reuter, A.,
Transaction Processing - Concepts and Techniques
Morgan Kaufmann, 1993, Chapter 4 (Transaction Models)
Elmasri, R., Navathe, S.B.,
Fundamentals of Database Systems, Addison-Wesley
Third Edition (2000), Chapters 19 Ð 20
Fourth Edition (2004), Chapters 17 Ð 18
Literature: Garcia-Molina, H., Ullman, J.D., Widom, J.,
Database Systems Ð The Complete Book
Prentice Hall, 2002, Chapters 18 Ð 19
Transaction Management
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
q way of cooperation / concurrency
q size of processed entities
q transaction duration
Type of Transaction Model (TM) is determined by:
q data integrity (recovery in case of system errors)
q synchronization of multiuser mode
q consistency checking
Aims of DBMS transaction concept:
Transaction Models
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
Lewis, P.M., Bernstein, A., Kifer, M.,
Databases and Transaction Processing
An Application-Oriented Approach
Addison Wesley, 2002, Chapters 15, 20 Ð 26
Another recommended book:
Gray, J., Reuter, A.,
Transaction Processing - Concepts and Techniques
Morgan Kaufmann, 1993
ÓThe BibleÓ on Transaction Management Theory:
INF4180-transact-4
INF4180-transact-2
Further Readings in Transaction Management
INF4180-transact-5
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
o Cooperating transactions (see later)
o User defined correctness criteria
INF4180-transact-7
o Group transactions into classes that can interleave without
problems/ cannot interleave at all
q Weaker correctness-criteria may be established with knowledge
about the application semantics
Examples:
q The Serializability-based correctness criteria may be too strong to
obtain the wanted concurrency
Non-serializable Correctness Criteria
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
Results of transactions having completed successfully must not be
forgotten by the system.
q Durability
A program running under transaction protection must behave exactly as it
would in single-user mode.
q Isolation
A transaction produces consistent results only; otherwise it aborts.
A result is consistent if the new state of the DB fulfills all the consistency
constraints.
q Consistency
State transitions should appearantly jump from the initial state to the
result without any observable intermediate states
Ð or appearantly never leave the initial state.
q Atomicity
The classical ACID requirements
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
INF4180-transact-6
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
INF4180-transact-8
q Real actions
Affect the real, physical world in a way that is hard or impossible to
reverse: definitely D
Examples: cutting metal pieces, chemical reactions
q Protected actions
ACID properties
Building blocks for reliable, distributed applications
Example: classical DBMS transactions such as SELECT
q Unprotected actions
Only C, not AID
Must either be controlled by the application environment, or must
be embedded in a higher-level protected action
Example: single disk write
Three types of actions as building blocks for system services
Classification of Action Types
o Others É
o Invalidation
(example: P invalidates Q if H1,P,H2 and H1, H2, Q are legal, but
H1,P,H2,Q is not)
o Commutativity
(example: Write always conflicts with other reads and writes)
q But what is a ÒconflictÓ?
q A history is serializable if it is conflict equivalent with a serial history
(no interleaving)
q Two histories are conflict equivalent if the relative order of conflicting
operations is the same
q A history is a sequence of operations from possibly interleaved transactions
Mostly used: Conflict serializability
Serializability-Based Correctness
INF4180-transact-9
q durable storage information (last record read/modified)
q SOCs gave rise to transactions, which may be seen as a special case
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q When the DBMS allows a transaction to run,
the transaction creates its own SOC (by locking)
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q user-related data (last screen sent)
q SOCs are dynamic Ð can be created, modified and removed at run-time
INF4180-transact-11
q counters (how often was the program called)
q SOCs are general Ð can contain different things
INF4180-transact-12
q Programs with context information are called context-sensitive
Examples of context:
q The DBMS has the Database as its SOC Ð the DBSOC
INF4180-transact-10
q Programs with no context information (context is the empty set)
are called context-free
A general view of a program:
f(input_message, context) _ {output_message, context}
Transaction Processing Context
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
We need various extensions of the flat
transaction model for the problems resulting
from the different application domains.
o Minimize lost work in case of crashes -> chained transactions, sagas
q Bulk updates
o large and recursive structures -> selective rollback desirable ->
savepoints, nested transactions
q Trip planning
o much cooperation between users -> design transactions
o long duration (hours/days/weeks)
o large (complex) entities are processed in a complex way
q CAD (machine design)
Some examples:
New Requirements for Transactions
q => System must be structured into a hierarchy of abstract data types.
o Monitoring the dependencies of operations on each other in order to be
able to trace the execution history in case faulty data are found.
o Containing the effects of arbitrary operations as long as there might be a
necessity to revoke them
q Sphere of Control
Ð jf. Gray & Reuter section 4.3
Sphere of Control (SOC)
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
End_of_Transaction (IF condition COMMIT (work) -> execute real action
ELSE ROLLBACK (work) -> do not execute real action)
...
SELECT ... (protected action) -> read_block (unprotected action)
INSERT ... (protected action) -> write_block (unprotected action)
DRILL_HOLE ... (real action) -> defer execution until end of transaction
UPDATE ... (protected action) -> read_block + write_block (unprotected
actions)
SELECT ... (protected action) -> read_block* (unprotected actions)
Begin_Transaction (work)
Example
B
T
C
C
(a) T is active; abort of
system TA will cause
T to roll back, too
A
A
Trigger
B
C
System
A
C
A
aborted
ROLLBACK
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
terminate
COMMIT
active
INF4180-transact-13
B
T
C
C
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
(b) Termination 1:
T has committed
A
A
B C
System
A
C
A
B
T
C
C
INF4180-transact-15
(c) Termination 2:
T has aborted
A
A
B C
System
A
C
A
deactivated
Example: Describing Flat Transactions
committed
terminate
NULL
BEGIN
q What is required to describe TMs?
- Structural dependencies (hierarchy of abstract data types)
- Dynamic dependencies (use of shared data)
q State-transition diagram for a flat transaction as seen by the application
(see Figure)
Notation for Transaction Models
A(borted)
A(bort)
T
C(ommit)
savepoint(1)
abort
rollback(2)
savepoint(2)
savepoint(3)
commit
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
See also Gray & Reuter Figure 4.8 (page 188)
q A crash (murder) rolls back the whole transaction
INF4180-transact-16
q Application program can do a partial ROLLBACK to a specific Savepoint
q Execution between two savepoints = atomic action (SoC)
q Savepoint: causes the system to record the current state of processing
and returns a handle to the application program
q Relaxing the Òall-or-nothingÓ characteristics of flat transactions
start
trans
Flat transaction with Savepoints
Vera Goebel
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
q Passive part:
Atomic actions that depend on other transactions might not be able to
make certain state transitions all by themselves
There must be rules specifying the conditions for performing a (signalled)
state transition
INF4180-transact-14
C(ommitted)
B(egin)
q Active part:
3 events can cause an atomic action to change its state - BEGIN,
ROLLBACK, and COMMIT
TMs can define conditions that trigger these events
state indicators of the
action«s outcome
unique persistent
transaction identifier
signal entries for the atomic action
to perform a state transition
Diagramming technique (Gray & Reuter)
B
S1
C
C
B
S2
B
S1
C
C
C
C
A
A
A
A
A
A
C
C
C
C
C
C
INF4180-transact-17
B
S3
B
S2
B
S1
B C
System
A
C
A
Trigger
A
A
B
C1
C
C
Trigger
A
A
B
C2
C
C
B
C
System
A
C
A
B
C1
C
C
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
Trigger
A
A
A
A
B
C2
INF4180-transact-19
Trigger
C
C
q First TA in the chain has committed; second TA got started as part of
commit of C1. Now the second TA is structurally dependent of ÒsystemÓ.
B
C
System
A
C
A
q First TA in the chain has been started; start of the second TA will be
triggered by commit of the first TA.
Chained Transactions Ð normal execution
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
Savepoint3 taken
Savepoint2 taken
A
A
A
A
B C
System
A
C
A
Savepoint execution
Not shown graphically:
¥ Rollback to a previous
savepoint
¥ Abort of whole
transaction
Savepoint1 taken
A
A
B C
System
A
C
A
T1
commit
T2
commit
abort
rollback
T3
T4
commit
INF4180-transact-18
B
C
Restart
A
C
A
B
C
System
A
C
A
A
A
A
A
B
C1Õ
B
C1
C
C
C
C
Trigger
Trigger
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
Trigger
Trigger
A
A
Restart processing in a chained transaction
C
C
INF4180-transact-20
Trigger
B
C2
Chained Transactions Ð restart processing
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q At the same time, the whole chain can bee seen as one T, that keeps all its
locks and cursors
Variation of savepoints!
q By commit of Ti , the program can release all the objects that are no longer
needed, and pass on the processing context to Ti+1
q By commit of Ti , the program waives its rights to do a rollback
q No other T can see or alter the context data passed from Ti to Ti+1
commit
Chained Transactions
q Commit of Ti + begin of Ti+1 = one atomic action
start
trans
abort
rollback
rollback
T11
call
commit
T1 subtrans
Closed Nested Transactions
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q In case the recovery of a subtransaction fails, the whole T will fail
q Subtransactions can fail and be restarted or replaced by another
subtransaction without causing the whole T to fail or restart
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
INF4180-transact-23
q Internally, subtransactions are run concurrently and their actions are
synchronized by an internal concurrency control mechanism
q Especially suitable for distributed environments/applications
INF4180-transact-22
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
INF4180-transact-24
q Does not increase the intertransaction parallelism
This is a crucial point especially in long and interactive transactions
q By end of subtransaction, locks are inherited to the supertransaction and
are released only when the top-level transaction terminates
q Locks are kept until the end of the subtransaction
(which enables the isolation of parallel subtransactions)
q Can be used as basis for more complex transaction models
commit
q The top-level T appears to the outside as a normal atomic transaction
q Can increase intra-transaction parallelism
q Closed variant does not increase inter-transaction parallelism
T2
call
subtrans
q Subtransactions maintain isolation (I) against each other
INF4180-transact-21
start
trans
T
call
subtrans
q A transaction tree in which a transaction T may contain subtransactions
start
trans
Nested transactions
q Multiple variants (open and closed)
Nested Transactions
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
q Restart handling: reestablishes the state of the most recent COMMIT
q Work lost: savepoints allow for a more flexible state restoration
q Lock handling: COMMIT allows to free locks, savepoints have no
implications on locks
q Commit versus savepoint: only rollback to last savepoint possible
(savepoints in chained transactions - other approach)
q Workflow structure: substructure similar to savepoints
Chained Transactions vs. Savepoints
INF4180-transact-25
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
Nested transactions are a generalization of savepoints:
Whereas savepoints allow organizing a transaction into a
sequence of actions that can be rolled back individually,
nested transactions form a hierarchy of pieces of work
(similar to nesting SoCs)
INF4180-transact-27
o the transaction makes sure that the global data used by these modules
are isolated and that recovery with the same granularity is provided
o modular design takes care of the application structure and
encapsulates local (dynamic) data structures
q Modularization in software engineering and nested transactions
complement each other:
q Powerful mechanism for fine-tuning the scope of rollback in applications
with complex structure
Example: trip planning
Using Nested Transactions
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
o Loss of serializability
=> One solution of this problem is semantic concurrency control
o Terminated subtransactions may be rolled back because a
supertransaction aborts (problem of cascading rollback)
=> One solution for this problem is to introduce compensating
transactions (More later!)
q Two important problems can occur in this approach:
q Locks are released at the end of the subtransaction
q Avoid conflicts, parallel subtransactions are separated
q Changes made by subtransactions are visible before the top-level
transaction terminates
Open Nested Transactions
B
T
C
C
B
T1
C
C
C
C
Wait
B C
T11
A
C
A
A
A
C
C
Wait
Trigger
B
T
Trigger
B C
Trigger
T1
A
C
A
Trigger
A
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
A
A
B
T
A
System
Wait
C
C
INF4180-transact-26
B
T2
T1 has created subT T11,
then the root T has started subT T2
Multi-Level Transactions
Trigger
A
A
Trigger
System
SubT T1 has
been started
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
q For long-duration Ts, concurrency can be increased
INF4180-transact-28
q Since compensating subTs exist at all levels, it is guaranteed that all
updates can be revoked, even if the root transaction fails and a whole
number of subTs have committed before
q The compensating subTs can be thrown away unused when the root
transaction commits
q Instead, a compensating subtransaction for the subT that is about to commit
is installed, so it can be executed in case the pre-committed results turn out
to be invalid
q Multi-level Ts let the subtransactions commit immediately (pre-commit),
thereby waiving the right to uniliteral backout of updates
q Multi-Level Ts are a generalized version of nested transactions
q The abstraction can be extended to multiple levels
q Multi-Level Transactions:
Transactions that are abstracted into subtransactions implemented by a set
of lower-level operations
A
A
Trigger
System
Root T
is running
Nested Transactions - Graphical Representation
The higher the level, the more
abstract the operations
Opens for a modular approach
to concurrency control
Abstractions
Concurrency
control
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
INF4180-transact-31
o The structure of a distributed transaction depends on the distribution
of data in a network. Even for a flat transaction, from the application«s
point of view a distributed transaction may have to be executed if the
data involved are scattered across a number of nodes
C
C
Node A
A
B
T
A
root T is running
at node A
q The conceptual difference between a distributed transaction and a nested
transaction is:
o The structure of nested transactions is determined by the functional
decomposition of the application, that is, by what the application
views as SoC
B
T
A
A
Trigger
C
C
C
C
A
A
B
N
C
C
A
A
C
C
INF4180-transact-30
B
CN
Trigger
B C
Trigger
T
A
C
A
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
B
N
Trigger
A
A
Trigger
B C
System
A
C
A
SubT N has committed and has
installed its compensating T CN,
that will get started if T aborts;
CN must commit
B
T1
C
C
C
C
B
T2
C
C
Node C
A
A
Vera Goebel
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Node B
A
A
B
T
Node A
A
A
System
remote subTs T1 and T2
have been started
INF4180-transact-32
Distributed Transactions - Graphical Representation
C
C
System
INF4180-transact-29
A
B
T
Trigger
Trigger
A
B C
System
A
C
A
SubT N has
been started
B C
System
A
C
A
Top-level T
is running
Multi-Level Transactions - Graphical Representation
q A distributed transaction is typically a flat transaction that runs in a
distributed environment and therefore has to visit several nodes in the
network, depending on where the data is
Distributed Transactions
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
The multi-level Ts notion of layers is very important:
It prevents serialization problems between
precomitted subtransactions, because they are
protected by the transactions on the higher levels
A single global
mechanism must be used
Not necessarily a notion
of abstraction
A predefined number of levels,
No predefined notion of
two adjacent pairs define a layer layers
Nested
Layers
Multi-Level
Multi-Level vs. Nested Transactions
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q Transactions of other application programs may intermingle
INF4180-transact-35
q The application program is really doing transaction-monitor work
Simple solution, minimize lost work,
but:
q The application program keeps track of progress, saving the
identity of the last successfully executed minibatch in a safe place
Ð i.e. the database
q The huge sequence of actions is split into shorter sequences
which are run as separate transactions (mini-batches)
under the control of the application program
A primitive solution for bulk updates
(example: Adding interest to 100 000 accounts)
Mini-Batch
INF4180-transact-33
q Transactions in DBMS for
nonstandard application domains
tend to be very long,
consequently the usual
serializability criterion is not
adequate. The long duration of
these tasks (transactions) means
that the concurrency control
strategies used in conventional
DBMS are not appropriate
q When transactions execute for
long durations, this scheme does
not work; a long-running
transaction may block out other
transactions for days or months
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q Traditional DBMS enforce
serializable schedules of
transactions, with the major
differences being in the size of
the information grain whose
access is individually controlled
q Traditional transactions are of
short duration (seconds and
minutes), traditional schemes
rely on blocking transactions
terminating quickly, thus
enabling other transactions to
continue execution
Short vs. long transactions
commit
abort
Sagas
rollback
commit
commit
commit
INF4180-transact-34
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
Sagas support long-running transactions/activities, but:
Sagas are not suitable for cooperating and unpredictable
(dynamic) T-activities Ð this is the case when the user is
controlling the T (interactive mode)
q Transactions may be rolled back to a chosen commit
INF4180-transact-36
q An abort rolls back the subT, then recovery with help of compensating CTs:
If Tk fails, the transactions CTk-1 ... CT1 are executed
q A commit of a subT make its changes visible to the rest of the world
q Split a long T (Saga) into arbitrary many (shorter) subT T1 ... Tn
with corresponding compensating transactions CT1 ... CTn
start
trans
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
o Cooperative transactions
o Sagas
o Minibatch
q Some solutions
Ð however, these techniques do not increase concurrency
q Long transaction mechanisms should be able to save intermediate states,
this can be accomplished by Savepoints or Nested transactions
o Bulk updates
o CASE
o Computer aided design
q Examples:
Long-Lived Transactions
INF4180-transact-37
create object X
do some design work
Transaction B
granted
use object X
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
X
Transaction A
show me object X
q Possible solution: Check-in/check-out model
INF4180-transact-39
give back X
q The concept of cooperative transactions allow transactions to view each
others' partial results under certain conditions
A solution for design work
Ð in a design team, designers often look at each others' incomplete or
inconsistent results to guide their own work
(examples: Computer Aided Design, CASE applications)
Cooperative Transactions
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
Sagas are an alternative to
mini-batches
q Additional concurrency can be gained by type-specific concurrency control
q Sagas relax the restriction that the subtransactions of the saga execute
without external interference
q The compensating transaction can undo the transaction's effect should the
saga abort, even when another unrelated transaction has already executed
q The backward chain of compensating transactions is established as the
original chain proceeds in a forward direction
o It uses the compensation idea from multi-level transactions to make the
entire chain atomic
o It defines a chain of transactions as a unit of control Ð a saga
q Two extensions of chained transaction concept:
Sagas
B
S1
C
C
C
C
B
S3
C
INF4180-transact-38
C
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
INF4180-transact-40
q The check-in/check-out scheme supports flexible concurrency control while
not requiring changes to the DB concurrency control; but it delegates much
of the responsibility for data integrity and concurrency control to the users
q The check-in and check-out operations are normal (short) transactions
against the global DB
q When work is finished, the objects are "checked in" to the global DB,
which enforces normal concurrency control. The entire collection of
operations in the private workspace appear as a single transaction
q In the private workspace, users operate on the objects outside the DB
concurrency control
q Users wishing to cooperate "check out" objects from the global DB into a
private workspace
One possible scheme to support cooperative work
A
A
Trigger
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
Trigger
B
S2
B
C
CS2
A
C
A
A
A
Trigger
Check-in/check-out
B C
CS1
A
C
A
A
A
B C
System
A
C
A
Sagas - Graphical Representation
INF4180-transact-41
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
INF4180-transact-43
q Transaction histories that are rigorous, strict etc. modulo CCSR
can be defined
(the formal foundation for non-compensation based recovery)
i.e. : CCSR also offers automatic recovery
(Thus there is no need for compensating transactions)
q CCSR can be enforced by 2PL
q The Generalized Serializability theorem:
A transaction history is CCSR if and only if its conditional conflict
graph is acyclic
q The only change is in the definition of conflict
Formalism
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q see http://www.xymphonic.com/
q Modified nested transaction model: Subdatabases
q Relaxing the rules of conflict serializability
Not ACID, but ACCID (CC Ð Conditional Concurrency)
q Based on two ideas:
q Developed by Telenor Research and Development
Dr. Ole J¿rgen Anfindsen
Xymphonic Systems (APOTRAM)
INF4180-transact-42
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
INF4180-transact-44
q This idea allows arbitrarily complex patterns of collaboration involving two
or more users
q NCSR generalizes CSR by allowing the sequence of operations within a
WSOC to be serializable rather than serial; thus applying CSR recursively
q If we enforce conflict serializability as the correctness criterion at all levels > nested conflict serializability (NCSR)
q Can be extended to arbitrary number of levels -> nested databases
q Other users can then execute transactions within this dynamically created
DBSOC on behalf of the transaction that created it
q The idea:
Replace a (single-user) WSOC with a (multi-user) DBSOC (subdatabase)
q The write Sphere of Control (WSOC) can be seen as a single user database
Nested conflict serializability (NCSR)
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
q The application programmers need to understand the semantics of the
parameter values, the transaction manager does not
o W(good) and R(medium, good) would not conflict
o W(bad) and R(good) would conflict
o The set of available parameter values is {good, medium, bad}
q Example:
q Let A and B be sets of ÒqualityÓ-parameter values
Let W(A) and R(B) denote write and read operations
W(A) and R(B) will conflict with each other unless A is a subset of B
q Conditional Conflict serializability (CCSR):
Applications can use reads/writes with ÒqualityÓ-parameters
q Classical Conflict serializability:
Two operations are conflicting if at least one of them is a write operation
Conditional Conflict serializability (CCSR)
INF4180-transact-45
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
INF4180-transact-47
q Operations for ODMS adminstration in the Database type
(create, delete, move, copy, reorganize, verify, backup, restore)
q Transaction consistency over transient objects
q Multi-database transactions
q Multiprocess transactions
q Long-lived transactions
Make the transaction object persistent!
q Nested transactions
Transactions i OO-databases
Ð beyond ODMG
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
Workflow Management (1)
© UiO, Institutt for informatikk 2003 Ð Gerhard Skagestein 6.sept. 2001
void begin( )
void commit( )
void abort( )
void checkpoint( )
void join( )
void leave( )
boolean isOpen( )
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
INF4180-transact-48
q Execution of multiple tasks by different processing entities may be
controlled by a human coordinator or by a software system called a
workflow management system
q Specification of a workflow involves describing those aspects of its tasks
and processing entities that are relevant to controlling and coordinating
the execution. It also requires specification of the relationships among
tasks and their execution requirements. These can be specified using a
variety of software paradigms, e.g., rules, constraints, or programs
q A processing entity that performs the tasks may be a person or a software
system, e.g., mailer, application program, or DBMS
q A task defines some work to be done and can be specified in a number of
ways, e.g., textual description in a file or email, form, message, or
computer program
INF4180-transact-46
Exceptions
are left out
t:Transaction
void open(in string odms_name)
void close( )
void bind(in Object an_object, in string name)
Object unbind(in string name)
Object lookup(in string object_name)
db:Database
q Workflows:
Coordinating execution of multiple tasks performed by different
processing entities
db.close( );
t.leave( );
t.commit( ); or t.abort( );
myobject.lock(read) // write, upgrade
É // implicitely locked
É //objects explicitely or
q Transient objects are not subject to transaction semantics
q Transactions are run under the control of Transaction objects
t.begin( );
t.join( ); // connect t to thread
t = TransactionFactory.new( );
db.open(ÓDatabasenameÓ);
db = DatabaseFactory.new( );
ODMG: The Transaction Object
q A transaction run against a single logical ODMS
q Assume a linear sequence of transactions within a thread
q Programs using persistent objects are organized into transactions
ODMG: Transactions i OO-databases
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
INF4180-transact-51
q Atomicity: Owing to long transaction durations (e.g., transactions
in CAD applications can last days or even weeks) it is not
acceptable to undo a complete transactions if a hardware failure
or deadlock occurs. Therefore, it must be possible to define
additional recovery units, to get specific save points
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
INF4180-transact-52
q Durability: It should be possible to define savepoints within a long
transaction or to divide transactions into subtransactions to provide a
partial rollback/undo to avoid the complete rollback/undo if any failure
occurs during the long transaction. If the supertransaction aborts it may
be necessary to rollback/undo an already committed subtransaction or to
execute a compensating transaction for the specific subtransactions
q Isolation: In many nonstandard application domains it is necessary to
share data (e.g., in a group of engineers) by exchanging uncommitted
data (dirty read). Corresponding protocols should control such
exchanges
q Consistency: Consistency constraints should not be checked at the end
of a transaction. It is necessary to introduce smaller consistency units.
Non-standard-DBMS must check very complex integrity constraints
which requires new mechanisms to define and control such integrity
constraints
INF4180-transact-50
q Flat (short) transactions are transactions like they are known from
traditional DBMS, they do not have any subtransactions.
Traditional DBMS must support the notion of atomic, recoverable,
and serializable transactions (ACID properties). The "classical"
transaction management is not flexible enough for nonstandard
application domains. Depending on the application domain some
of the ACID properties are to restrictive for Nonstandard-DBMS.
Therefore, the four properties atomicity (A), consistency (C),
isolation (I), and durability (D) are only partially desirable:
© UiO, Institutt for informatikk 2003 Ð Gerhard
Vera Goebel
Skagestein 6.sept. 2001
Summary (2)
INF4180-transact-49
q Approach providing limited workflow-management facilities in an
environment consisting of transaction monitors and event
monitoring/synchronizing facilities (similar to approach proposed
in CORBA)
q Layering approach: implements workflow-control facilities on top
of uniform application-level interfaces to entities by developing
modules to support intertask dependencies and other workflow
specifications
q Embedded approach: requires and exploits significant support
from the processing entities for specification and enforcement
dependencies
Three implementation approaches:
Workflow Management (3)
Summary (1)
Vera Goebel
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
q Execution (correctness) requirements: are defined to restrict the
execution of the workflow(s) to meet application-specific correctness
criteria, including failure-atomicity and execution-atomicity requirements,
and workflow concurrency control and recovery requirements
q Task coordination requirements: are expressed as intertask-execution and
data-flow dependencies, and termination conditions of the workflow.
-> statically or dynamically
q Task specification: execution structure of each task is defined by
providing a set of externally observable execution states and a set of
transitions between these states. Characteristics of processing entities
relevant for task-execution requirements may be defined
Specification of workflows:
q Based on ideas borrowed from operating systems (job control language),
long-running activities, and other transaction models
Workflow models:
Workflow Management (2)
© UiO, Institutt for informatikk 2003 Ð Gerhard
Skagestein 6.sept. 2001
Vera Goebel
INF4180-transact-53
q A nonstandard-DBMS should provide the same level of recovery as a
conventional DBMS. Nonstandard-DBMSes are often used in client-server
environments, consequently recovery and logging mechanisms have
often to solve the same problems as those which occur in distributed
DBMSes. Some solutions provide extended logging approaches for
snapshot refreshing or the distributed two-phase locking protocol
q Nonstandard-DBMSes should support cooperative work more directly by
providing and enforcing a wider variety of lock types (e.g., object locks)
than the customary read/write locks
q Optimistic synchronization mechanisms are not suitable for long
transactions, because access conflicts are first recognized at the end of a
transaction which might cause a rollback (that might destroy the work of
weeks)
q Synchronization mechanisms determine crucially the efficiency of
DBMSes. There has to be a balance between the necessary effort of a
locking technique and its performance
Summary (3)
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement