1. Teaching material
based on Distributed
Systems: Concepts and
Design, Edition 3,
Addison-Wesley 2001.
Distributed Systems Course
Distributed transactions
Copyright © George
Coulouris, Jean Dollimore,
Tim Kindberg 2001
email: authors@cdk2.net
This material is made
available for private study
and for direct use by
individual teachers.
It may not be included in any
product or employed in any
service without the written
permission of the authors.
Viewing: These slides
must be viewed in
slide show mode.
13.1 Introduction
13.2 Flat and nested distributed transactions
13.3 Atomic commit protocols
13.4 Concurrency control in distributed
transactions
13.5 Distributed deadlocks
13.6 Transaction recovery

2. Commitment of distributed transactions - introduction
 a distributed transaction refers to a flat or nested
transaction that accesses objects managed by
multiple servers
 When a distributed transaction comes to an end
– the either all of the servers commit the transaction
– or all of them abort the transaction.
 one of the servers is coordinator, it must ensure the
same outcome at all of the servers.
 the ‘two-phase commit protocol’ is the most
commonly used protocol for achieving this
2
•

3. Distributed transactions
Client
3
X
Y
Z
X
Y
M
T1 N
T2
T11
Client
P
T
T 12
T21
T22
(a) Flat transaction (b) Nested transactions
T
T
Figure 13.1
A flat client transaction completes each of
its requests before going on to the next
one. Therefore, each transaction accesses
servers’ objects sequentially
In a nested transaction, the top-level
transaction can open
subtransactions, and each
subtransaction can open further
subtransactions down to any
depth of nesting
In the nested case,
subtransactions at the same level
can run concurrently, so T1 and
T2 are concurrent, and as they
invoke objects in different servers,
they can run in parallel.
•

4. Nested banking transaction
b.withdraw(20)
openSubTransaction
b.withdraw(20);
openSubTransaction
c.deposit(10);
openSubTransaction
d.deposit(20);
 client transfers $10 from A to C and then transfers $20 from B to
4
a.withdraw(10)
c.deposit(10)
d.deposit(20)
Client A
B
C
T1
T2
T3
T4
T
D
X
Y
Z
T = openTransaction
openSubTransaction
a.withdraw(10);
closeTransaction
Figure 13.2
requests can be
run in parallel -
with several
servers, the
nested
transaction is
more efficient
•

5. Why might a participant abort a transaction?
The coordinator of a flat distributed transaction
 Servers execute requests in a distributed transaction
– when it commits they must communicate with one another to
coordinate their actions
– a client starts a transaction by sending an openTransaction request to
a coordinator in any server (next slide)
 it returns a TID unique in the distributed system(e.g. server ID + local
transaction number)
 at the end, it will be responsible for committing or aborting it
– each server managing an object accessed by the transaction is a
participant - it joins the transaction (next slide)
 a participant keeps track of objects involved in the transaction
 at the end it cooperates with the coordinator in carrying out the commit
protocol
– note that a participant can call abortTransaction in coordinator
5
•

6. openTransaction goes to the
coordinator
A flat distributed banking transaction
join participant
A
BranchX
participant
B
participant
C
D
Figure 13.3
a client’s (flat)
banking
transaction
involves accounts
A, B, C and D at
servers BranchX,
BranchY and
BranchZ
Each server is shown
with a participant, which
joins the transaction by
invoking the join method
in the coordinator
 Note that the TID (T) is passed with each request e.g. withdraw(T,3)
6
.
BranchZ
Client
BranchY
join
join
T
a.withdraw(4);
b.withdraw(3);
c.deposit(4);
d.deposit(3);
openTransaction
b.withdraw(T, 3);
closeTransaction
T = openTransaction
a.withdraw(4);
c.deposit(4);
b.withdraw(3);
d.deposit(3);
closeTransaction
Note: the coordinator is in one of the servers, e.g. BranchX
•

7. The join operation
 The interface for Coordinator is shown in Figure 12.3
– it has openTransaction, closeTransaction and abortTransaction
– openTransaction returns a TID which is passed with each operation so that
servers know which transaction is accessing its objects
 The Coordinator interface provides an additional method, join,
which is used whenever a new participant joins the
transaction:
– join(Trans, reference to participant)
– informs a coordinator that a new participant has joined the transaction Trans.
– the coordinator records the new participant in its participant list.
– the fact that the coordinator knows all the participants and each participant
knows the coordinator will enable them to collect the information that will be
needed at commit time.
7
•

8. Atomic commit protocols
 transaction atomicity requires that at the end,
– either all of its operations are carried out or none of them.
 in a distributed transaction, the client has requested the
operations at more than one server
 one-phase atomic commit protocol
– the coordinator tells the participants whether to commit or abort
– what is the problem with that?
– this does not allow one of the servers to decide to abort – it may have
discovered a deadlock or it may have crashed and been restarted
 two-phase atomic commit protocol
– is designed to allow any participant to choose to abort a transaction
– phase 1 - each participant votes. If it votes to commit, it is prepared. It cannot
change its mind. In case it crashes, it must save updates in permanent store
– phase 2 - the participants carry out the joint decision
8
•
The decision could be commit or abort - participants record it in permanent store

9. Failure model for the commit protocols
 Recall the failure model for transactions in Chapter 12
– this applies to the two-phase commit protocol
 Commit protocols are designed to work in
– asynchronous system (e.g. messages may take a very long time)
– servers may crash
– messages may be lost.
– assume corrupt and duplicated messages are removed.
– no byzantine faults – servers either crash or they obey their requests
 2PC is an example of a protocol for reaching a consensus.
– Chapter 11 says consensus cannot be reached in an asynchronous system if
processes sometimes fail.
– however, 2PC does reach consensus under those conditions.
– because crash failures of processes are masked by replacing a crashed
process with a new process whose state is set from information saved in
permanent storage and information held by other processes.
9
•

10. The two-phase commit protocol
 During the progress of a transaction, the only
communication between coordinator and participant
is the join request
– The client request to commit or abort goes to the coordinator
 if client or participant request abort, the coordinator informs the
participants immediately
 if the client asks to commit, the 2PC comes into use
10
 2PC
– voting phase: coordinator asks all participants if they can commit
 if yes, participant records updates in permanent storage and then votes
– completion phase: coordinator tells all participants to commit or abort
– the next slide shows the operations used in carrying out the protocol
•
How many messages are sent between the coordinator and Why does participant record updates in permanent storage e aatc bht hpiasr stitcaipgaen?t?

11. Operations for two-phase commit protocol
canCommit?(trans)-> Yes / No
Call from coordinator to participant to ask whether it can commit a transaction.
Participant replies with its vote.
doCommit(trans)
Call from coordinator to participant to tell participant to commit its part of a
transaction.
 participant interface- canCommit?, doCommit, doAbort
coordinator interface- haveCommitted, getDecision
11
doAbort(trans)
Call from coordinator to participant to tell participant to abort its part of a
transaction.
haveCommitted(trans, participant)
Call from participant to coordinator to confirm that it has committed the transaction.
getDecision(trans) -> Yes / No
Call from participant to coordinator to ask for the decision on a transaction after it
has voted Yes but has still had no reply after some delay. Used to recover from
server crash or delayed messages. Figure 13.4
•
This is a request with a reply
These are asynchronous requests to avoid delays
Asynchronous request

12. The two-phase commit protocol
12
Figure 13.5
• Phase 1 (voting phase):
• 1. The coordinator sends a canCommit? request to each of the participants in
the transaction.
• 2. When a participant receives a canCommit? request it replies with its vote
(Yes or No) to the coordinator. Before voting Yes, it prepares to commit by saving
objects in permanent storage. If the vote is No the participant aborts immediately.
• Phase 2 (completion according to outcome of vote):
• 3. The coordinator collects the votes (including its own).
w (a)If there are no failures and all the votes are Yes the coordinator decides to
commit the transaction and sends a doCommit request to each of the participants.
w (b)Otherwise the coordinator decides to abort the transaction and sends doAbort
requests to all participants that voted Yes.
• 4. Participants that voted Yes are waiting for a doCommit or doAbort request from the
coordinator. When a participant receives one of these messages it acts accordingly
and in the case of commit, makes a haveCommitted call as confirmation to the
coordinator.
•

13. Communication in two-phase commit protocol
status step status
prepared to commit
 Time-out actions in the 2PC
prepared to commit
 to avoid blocking forever when a process crashes or a message is lost
– uncertain participant (step 2) has voted yes. it can’t decide on its own
 it uses getDecision method to ask coordinator about outcome
– participant has carried out client requests, but has not had a Commit?
from the coordinator. It can abort unilaterally
– coordinator delayed in waiting for votes (step 1). It can abort and send
doAbort to participants.
13
canCommit?
Yes
doCommit
haveCommitted
Coordinator
1
3
(waiting for votes)
committed
done
step
Participant
2
4
(uncertain)
committed
Figure 13.6
•
TTThhhiniinnkkk a aabbbooouuutt t s tptheaepr t ci2co io-p rawdnhinta abt teiosfro trihnee ss ptteeroppb 12le --m ww fhhoaart t t ihisse tt hhpeea rpptirrcooipbballeenmmt???

14. Performance of the two-phase commit protocol
 if there are no failures, the 2PC involving N
participants requires
– N canCommit? messages and replies, followed by N doCommit
messages.
 the cost in messages is proportional to 3N, and the cost in time is three
rounds of messages.
 The haveCommitted messages are not counted
– there may be arbitrarily many server and communication failures
– 2PC is is guaranteed to complete eventually, but it is not possible to
specify a time limit within which it will be completed
 delays to participants in uncertain state
 some 3PCs designed to alleviate such delays
• they require more messages and more rounds for the normal case
14
•

15. 13.3.2 Two-phase commit protocol for nested
transactions
 Recall Fig 13.1b, top-level transaction T and subtransactions
T1, T2, T11, T12, T21, T22
 A subtransaction starts after its parent and finishes before it
 When a subtransaction completes, it makes an independent
decision either to commit provisionally or to abort.
– A provisional commit is not the same as being prepared: it is a local decision
and is not backed up on permanent storage.
– If the server crashes subsequently, its replacement will not be able to carry out
a provisional commit.
 A two-phase commit protocol is needed for nested
transactions
– it allows servers of provisionally committed transactions that have crashed to
abort them when they recover.
15
•

16. Figure 13.7
Operations in coordinator for nested transactions
openSubTransaction(trans) -> subTrans
The TID of a subtransaction is an extension of its parent's TID, so that a
subtransaction can work out the TID of the top-level transaction.
The client finishes a set of nested transactions by calling closeTransaction or
abortTransacation in the top-level transaction.
Opens a new subtransaction whose parent is trans and returns a
unique subtransaction identifier.
getStatus(trans)-> committed, aborted, provisional
Asks the coordinator to report on the status of the transaction
trans. Returns values representing one of the following:
committed, aborted, provisional.
 This is the interface of the coordinator of a subtransaction.
– It allows it to open further subtransactions
– It allows its subtransactions to enquire about its status
 Client starts by using OpenTransaction to open a top-level transaction.
– This returns a TID for the top-level transaction
– The TID can be used to open a subtransaction
 The subtransaction automatically joins the parent and a TID is returned.
16
•

17. Transaction T decides whether to commit
T12 has provisionally committed and T11 has aborted, but the fate of T12
depends on its parent T1 and eventually on the top-level transaction, T.
Although T21 and T22 have both provisionally committed, 2 has
aborted and this means that T21 and T22 must also abort.
Suppose that T decides to commit although T2 has aborted, also
that T1 decides to commit although T11 has aborted
17
1
2
T11
provisional commit (at X)
T12
T21
T22
abort (at M)
provisional commit (at N)
aborted (at Y)
provisional commit (at N)
provisional commit (at P)
T
T
T
•
Figure 13.8
 Recall that
1. A parent can commit even if a subtransaction aborts
2. If a parent aborts, then its subtransactions must abort
– In the figure, each subtransaction has either
provisionally committed or aborted

18. Information held by coordinators of nested transactions
18
Coordinator of
transaction
Child
transactions
Participant Provisional
commit list
Abort list
T T1 , T 2 yes T1 , T 12 T11 , T 2
T1 T11 , T 12 yes T1 , T 12 T11
T2 T21 , T 22 no (aborted) T2
T11 no (aborted) T11
T12 , T 21 T12 but not T21 T21 , T 12
T22 no (parent aborted)T22
•
 When a top-level transcation commits it carries out a 2PC
 Each coordinator has a list of its subtransactions
 At provisional commit, a subtransaction reports its status and
the status of its descendents to its parent
 If a subtransaction aborts, it tells its parent
Figure 13.9
T12 and T21 share a A subtransaction (e.g. T When T2 is aborted c iot oterdllsin Ta t(onro a isn ftohremy abtoiotnh arubno uatt dsesrvceern Ndents) 21 and T22an orphan uses getStatus to ask its pare) nist acbaolluetd th aen o ourtcpohmaen. iIft oshnoeu oldf aitbso artn icf eitss tpoarrse natb hoarsts

19. canCommit? for hierarchic two-phase commit protocol
canCommit?(trans, subTrans) -> Yes / No
Figure 13.10
Call a coordinator to ask coordinator of child subtransaction
whether it can commit a subtransaction subTrans. The first
argument trans is the transaction identifier of top-level
transaction. Participant replies with its vote Yes / No.
19
•
 Top-level transaction is coordinator of 2PC.
 participant list:
– the coordinators of all the subtransactions that have provisionally committed
– but do not have an aborted ancestor
– E.g. T, T1 and T12 in Figure 13.8
– if they vote yes, they prepare to commit by saving state in permanent store
 The state is marked as belonging to the top-level transaction
 The 2PC may be performed in a hierarchic or a flat manner
Hierarchic 2PC - T asks canCommit? to T1 and T1 asks ThTeh es utrbaTnrsa nasrg aurmguemnte inst uiss euds ew thoe fnin sda tvhineg s tuhbet roabnjseacctsti oinn ptoe rv mcoaatenn Ceonnot.m sIfmt oairtb?as getoen tT, 1v2ote no.

20. Compare the advantages and disadvantages of the flat and nested approaches
canCommit? for flat two-phase commit protocol
canCommit?(trans, abortList) -> Yes / No
Figure 13.11
Call from coordinator to participant to ask whether it can
commit a transaction. Participant replies with its vote Yes / No.
20
•
 Flat 2PC
– the coordinator of the top-level transaction sends canCommit? messages
to the coordinators of all of the subtransactions in the provisional commit
list.
– in our example, T sends to the coordinators of T1 and T12.
– the trans argument is the TID of the top-level transaction
– the abortList argument gives all aborted subtransactions
 e.g. server N has T12 prov committed and T21 aborted
– On receiving canCommit, participant
 looks in list of transactions for any that match trans (e.g. T12 and T21 at N)
 it prepares any that have provisionally committed and are not in abortList and
votes yes
 if it can't find any it votes no

21. Time-out actions in nested 2PC
 With nested transactions delays can occur in the
same three places as before
– when a participant is prepared to commit
– when a participant has finished but has not yet received canCommit?
– when a coordinator is waiting for votes
 Fourth place:
– provisionally committed subtransactions of aborted subtransactions
e.g. T22 whose parent T2 has aborted
– use getStatus on parent, whose coordinator should remain active for a
while
– If parent does not reply, then abort
21
•

22. Summary of 2PC
 a distributed transaction involves several different servers.
– A nested transaction structure allows
 additional concurrency and
 independent committing by the servers in a distributed transaction.
 atomicity requires that the servers participating in a
distributed transaction either all commit it or all abort it.
 atomic commit protocols are designed to achieve this effect,
even if servers crash during their execution.
 the 2PC protocol allows a server to abort unilaterally.
– it includes timeout actions to deal with delays due to servers crashing.
– 2PC protocol can take an unbounded amount of time to complete but is
guaranteed to complete eventually.
22
•

23. 13.4 Concurrency control in distributed transactions
 Each server manages a set of objects and is
responsible for ensuring that they remain consistent
when accessed by concurrent transactions
– therefore, each server is responsible for applying concurrency control
to its own objects.
– the members of a collection of servers of distributed transactions are
jointly responsible for ensuring that they are performed in a serially
equivalent manner
– therefore if transaction T is before transaction U in their conflicting
access to objects at one of the servers then they must be in that order
at all of the servers whose objects are accessed in a conflicting
manner by both T and U
23
•

24. 13.4.1 Locking
 In a distributed transaction, the locks on an object
are held by the server that manages it.
– The local lock manager decides whether to grant a lock or make the
requesting transaction wait.
– it cannot release any locks until it knows that the transaction has been
committed or aborted at all the servers involved in the transaction.
– the objects remain locked and are unavailable for other transactions
during the atomic commit protocol
 an aborted transaction releases its locks after phase 1 of the protocol.
24
•

25. Interleaving of transactions T and U at servers X and Y
 in the example on page 529, we have
– T before U at server X and U before T at server Y
 different orderings lead to cyclic dependencies and
distributed deadlock
– detection and resolution of distributed deadlock in next section
T U
Write(A) at X locks A
25
Write(B) at Y locks B
Read(B) at Y waits for U
Read(A) at X waits for T
•

26. 13.4.2 Timestamp ordering concurrency control
 Single server transactions
– coordinator issues a unique timestamp to each transaction before it starts
– serial equivalence ensured by committing objects in order of timestamps
 Distributed transactions
– the first coordinator accessed by a transaction issues a globally unique
timestamp
– as before the timestamp is passed with each object access
– the servers are jointly responsible for ensuring serial equivalence
 that is if T access an object before U, then T is before U at all objects
– coordinators agree on timestamp ordering
 a timestamp consists of a pair <local timestamp, server-id>.
 the agreed ordering of pairs of timestamps is based on a comparison in
which the server-id part is less significant – they should relate to time
26
•

27. Can the same ordering be achieved at all servers without clock synchronization?
Why Timestamp is it better to ordering have roughly concurrency synchronized clocks?
control (continued)
 The same ordering can be achieved at all servers
even if their clocks are not synchronized
– for efficiency it is better if local clocks are roughly synchronized
– then the ordering of transactions corresponds roughly to the real time
order in which they were started
 Timestamp ordering
– conflicts are resolved as each operation is performed
– if this leads to an abort, the coordinator will be informed
 it will abort the transaction at the participants
– any transaction that reaches the client request to commit should
always be able to do so
 participant will normally vote yes
 unless it has crashed and recovered during the transaction
27
•

28. Use backward validation
Optimistic concurrency control
 each transaction is validated before it is allowed to commit
1. write/read, 2. read/write, 3. write/write
– transaction numbers assigned at start of validation
– transactions serialized according to transaction numbers
– validation takes place in phase 1 of 2PC protocol
 consider the following interleavings of T and U
– T before U at X and U before T at Y
Suppose T & U start validation at about the same time
T U
Read(A) at X Read(B) at Y
Write(A) Write(B)
Read(B) at Y Read(A) at X
Write(B) Write(A)
28
1. satisfied
2. checked
3. paralllel
X does T first
Y does U first
No parallel
Validation –.
commitment
deadlock
•

29. Commitment deadlock in optimistic concurrency control
 servers of distributed transactions do parallel validation
– therefore rule 3 must be validated as well as rule 2
 the write set of Tv is checked for overlaps with write sets of earlier
transactions
– this prevents commitment deadlock
– it also avoids delaying the 2PC protocol
 another problem - independent servers may
schedule transactions in different orders
– e.g. T before U at X and U before T at Y
– this must be prevented - some hints as to how on page 531
29
•

30. 13.5 Distributed deadlocks
 Single server transactions can experience deadlocks
– prevent or detect and resolve
– use of timeouts is clumsy, detection is preferable.
 it uses wait-for graphs.
 Distributed transactions lead to distributed deadlocks
– in theory can construct global wait-for graph from local ones
– a cycle in a global wait-for graph that is not in local ones is a
distributed deadlock
30
•

31. Figure 13.12
Interleavings of transactions U, V and W
 objects A, B managed by X and Y ; C and D by Z
U V W
d.deposit(10) lock D
31
b.deposit(10) lock B
a.deposit(20) lock A at Y
at X
c.deposit(30) lock C
b.withdraw(30) wait at Y at Z
c.withdraw(20) wait at Z
a.withdraw(20) wait at X
– next slide has global wait-for graph
•
U ® V at Y
V ® W at
Z W ® U at X

32. Figure 13.13
Distributed deadlock
 a deadlock cycle has alternate edges showing wait-for and held-by
 wait-for added in order: U ® V at Y; V ® W at Z and W ® U at X
C A
32
D
Waits for
Waits
for
Held by
Held
by
B Waits for
Held
by
X
Y
Z
Held by
W
V U
W
V
U
(a) (b)
•

33. Deadlock detection - local wait-for graphs
 Local wait-for graphs can be built, e.g.
– server Y: U ® V added when U requests b.withdraw(30)
– server Z: V ® W added when V requests c.withdraw(20)
– server X: W ® U added when W requests a.withdraw(20)
 to find a global cycle, communication between the
servers is needed
 centralized deadlock detection
– one server takes on role of global deadlock detector
– the other servers send it their local graphs from time to time
– it detects deadlocks, makes decisions about which transactions to
abort and informs the other servers
– usual problems of a centralized service - poor availability, lack of fault
tolerance and no ability to scale
33
•

34. Figure 13.14
Local and global wait-for graphs
local wait-for graph local wait-for graph global deadlock detector
T U
34
X
V T
Y
T
U V
 Phantom deadlocks
– a ‘deadlock’ that is detected, but is not really one
– happens when there appears to be a cycle, but one of the transactions
has released a lock, due to time lags in distributing graphs
– in the figure suppose U releases the object at X then waits for V at Y
 and the global detector gets Y’s graph before X’s (T ® U ® V ® T)
•

35. Edge chasing - a distributed approach to deadlock
detection
 a global graph is not constructed, but each server
knows about some of the edges
– servers try to find cycles by sending probes which follow the edges of
the graph through the distributed system
– when should a server send a probe (go back to Fig 13.13)
– edges were added in order U ® V at Y; V ® W at Z and W ® U at X
 when W ® U at X was added, U was waiting, but
 when V ® W at Z, W was not waiting
– send a probe when an edge T1 ® T2 when T2 is waiting
– each coordinator records whether its transactions are active or waiting
 the local lock manager tells coordinators if transactions start/stop waiting
 when a transaction is aborted to break a deadlock, the coordinator tells the
participants, locks are removed and edges taken from wait-for graphs
35
•

36. Edge-chasing algorithms
 Three steps
– Initiation:
 When a server notes that T starts waiting for U, where U is waiting at
another server, it initiates detection by sending a probe containing the
edge < T ® U > to the server where U is blocked.
 If U is sharing a lock, probes are sent to all the holders of the lock.
– Detection:
 Detection consists of receiving probes and deciding whether deadlock has
occurred and whether to forward the probes.
• e.g. when server receives probe < T ® U > it checks if U is waiting,
e.g. U ® V, if so it forwards < T ® U ® V > to server where V waits
• when a server adds a new edge, it checks whether a cycle is there
– Resolution:
 When a cycle is detected, a transaction in the cycle is aborted to break the
deadlock.
36
•

37. Figure 13.15
Probes transmitted to detect deadlock
 example of edge chasing starts with X sending <W ® U>, then Y sends
<W ® U ® V >, then Z sends <W ® U ® V ® W>
37
V
Held by
W
Held by Waits for
Waits
for
Waits for
Deadlock
detected
U
C
A
B
Initiation
W® U ® V ® W
W® U
W® U ® V
Z
Y
X
•

38. Edge chasing conclusion
 probe to detect a cycle with N transactions will require 2(N-1) messages.
– Studies of databases show that the average deadlock involves 2 transactions.
 the above algorithm detects deadlock provided that
– waiting transactions do not abort
– no process crashes, no lost messages
– to be realistic it would need to allow for the above failures
 refinements of the algorithm (p 536-7)
– to avoid more than one transaction causing detection to start and then more than one
being aborted
– not time to study these now
38
•

39. Summary of concurrency control for distributed
transactions
 each server is responsible for the serializability of
transactions that access its own objects.
 additional protocols are required to ensure that
transactions are serializable globally.
– timestamp ordering requires a globally agreed timestamp ordering
– optimistic concurrency control requires global validation or a means of
forcing a global ordering on transactions.
– two-phase locking can lead to distributed deadlocks.
 distributed deadlock detection looks for cycles in the global wait-for graph.
 edge chasing is a non-centralized approach to the detection of distributed
deadlocks
 .
41
•

40. 13.6 Transaction recovery
 Atomicity property of transactions
WWhhaatt iiss mmeeaanntt bbyy fdauilruarbei laityto?micity?
– durability and failure atomicity
– durability requires that objects are saved in permanent storage and will be available
indefinitely
– failure atomicity requires that effects of transactions are atomic even when the server
42
crashes
 Recovery is concerned with
– ensuring that a server’s objects are durable and
– that the service provides failure atomicity.
– for simplicity we assume that when a server is running, all of its objects are in volatile
memory
– and all of its committed objects are in a recovery file in permanent storage
– recovery consists of restoring the server with the latest committed versions of all of its
objects from its recovery file
•

41. Recovery manager
 The task of the Recovery Manager (RM) is:
– to save objects in permanent storage (in a recovery file) for committed
transactions;
– to restore the server’s objects after a crash;
– to reorganize the recovery file to improve the performance of recovery;
– to reclaim storage space (in the recovery file).
 media failures
– i.e. disk failures affecting the recovery file
– need another copy of the recovery file on an independent disk. e.g.
implemented as stable storage or using mirrored disks
 we deal with recovery of 2PC separately (at the end)
– we study logging (13.6.1) but not shadow versions (13.6.2)
43
•

42. Recovery - intentions lists
 Each server records an intentions list for each of its
currently active transactions
– an intentions list contains a list of the object references and the values
of all the objects that are altered by a transaction
– when a transaction commits, the intentions list is used to identify the
objects affected
 the committed version of each object is replaced by the tentative one
 the new value is written to the server’s recovery file
– in 2PC, when a participant says it is ready to commit, its RM must
record its intentions list and its objects in the recovery file
 it will be able to commit later on even if it crashes
 when a client has been told a transaction has committed, the recovery files
of all participating servers must show that the transaction is committed,
• even if they crash between prepare to commit and commit
44
•

43. Types of entry in a recovery file
Type of entry Description of contents of entry
Object A value of an object.
Transaction status Transaction identifier, transaction status ( prepared, committed
 For distributed transactions we need information relating to the 2PC as
well as object values, that is:
– transaction status (committed, prepared or aborted)
– intentions list
45
aborted) and other status values used for the two-phase
commit protocol.
Intentions list Transaction identifier and a sequence of intentions, each of
which consists of <identifier of object>, <position in recovery
file of value of object>.
Figure 13.18
•
Why is that a good
idea?
Object state flattened to bytes
first entry says prepared
Note that the objects need not be next to one another in the recovery file

44. Logging - a technique for the recovery file
 the recovery file represents a log of the history of all
the transactions at a server
– it includes objects, intentions lists and transaction status
– in the order that transactions prepared, committed and aborted
– a recent snapshot + a history of transactions after the snapshot
– during normal operation the RM is called whenever a transaction
prepares, commits or aborts
 prepare - RM appends to recovery file all the objects in the intentions list
followed by status (prepared) and the intentions list
 commit/abort - RM appends to recovery file the corresponding status
 assume append operation is atomic, if server fails only the last write will be
incomplete
 to make efficient use of disk, buffer writes. Note: sequential writes are
more efficient than those to random locations
 committed status is forced to the log - in case server crashes
46
•

45. Log for banking service
committed status
P0 P1 P2 P3 P4 P5 P6 P7
Object:A Object:B Object:C Object:A Object:B Trans:T Trans:T Object:C Object:B Trans:U
100 200 300 80 220 prepared committed 278 242 prepared
 Logging mechanism for Fig 12.7 (there would really be other objects in log file)
– initial balances of A, B and C $100, $200, $300
– T sets A and B to $80 and $220. U sets B and C to $242 and $278
– entries to left of line represent a snapshot (checkpoint) of values of A, B and C before T
started. T has committed, but U is prepared.
– the RM gives each object a unique identifier (A, B, C in diagram)
– each status entry contains a pointer to the previous status entry, then the checkpoint
can follow transactions backwards through the file
47
<A, P1> <C, P5>
<B, P2> <B, P6>
P0 P3 P4
Checkpoint
End
of log Figure 13.19.
prepared status and intentions list
•

46. Recovery of objects - with logging
 When a server is replaced after a crash
– it first sets default initial values for its objects
– and then hands over to its recovery manager.
 The RM restores the server’s objects to include
– all the effects of all the committed transactions in the correct order and
– none of the effects of incomplete or aborted transactions
– it ‘reads the recovery file backwards’ (by following the pointers)
 restores values of objects with values from committed transactions
 continuing until all of the objects have been restored
– if it started at the beginning, there would generally be more work to do
– to recover the effects of a transaction use the intentions list to find the value of the
objects
 e.g. look at previous slide (assuming the server crashed before T committed)
– the recovery procedure must be idempotent
48
•

47. Logging - reorganising the recovery file
 RM is responsible for reorganizing its recovery file
– so as to make the process of recovery faster and
– to reduce its use of space
 checkpointing
– the process of writing the following to a new recovery file
 the current committed values of a server’s objects,
 transaction status entries and intentions lists of transactions that have not
yet been fully resolved
 including information related to the two-phase commit protocol (see later)
– checkpointing makes recovery faster and saves disk space
 done after recovery and from time to time
 can use old recovery file until new one is ready, add a ‘mark’ to old file
 do as above and then copy items after the mark to new recovery file
 replace old recovery file by new recovery file
49
•

48. Recovery of the two-phase commit protocol
 The above recovery scheme is extended to deal with transactions doing the 2PC
protocol when a server fails
– it uses new transaction status values done, uncertain (see Fig 13.6)
 the coordinator uses committed when result is Yes;
 done when 2PC complete ( if a transaction is done its information may be removed
when reorganising the recovery file)
 the participant uses uncertain when it has voted Yes; committed when told the result
(uncertain entries must not be removed from recovery file)
– It also requires two additional types of entry:
Type of entry Description of contents of entry
Coordinator Transaction identifier, list of participants
51
added by RM when coordinator prepared
Participant Transaction identifier, coordinator
added by RM when participant votes yes
•

49. Start at end, for U find it is committed Boru itWf itfh eteh h esa esvrevr eTvre crco rhamasmsh ecitrdtea edsha arenldied rb cweofeo rhdeai ntvhaeet o U lara nspdtr e eapn aptrrayer dwti ceaip nhadan vptea rUtic uipnacnetrtain and participant
Log with entries relating to two-phase commit protocol
Trans:T Coord’r:T Trans:T Trans:U Part’pant:U Trans:U Trans:U
prepared part’pant
list: . . .
 entries in log for
coordinator entry participant entry
– T where server is coordinator (prepared comes first, followed by the
coordinator entry, then committed – done is not shown)
– and U where server is participant (prepared comes first followed by
the participant entry, then uncertain and finally committed)
– these entries will be interspersed with values of objects
 recovery must deal with 2PC entries as well as restoring objects
– where server was coordinator find coordinator entry and status
entries.
– where server was participant find participant entry and status entries
52
committed prepared Coord’r: . . uncertain committed
intentions
list
intentions
list
Figure 13.21
•

50. Recovery of the two-phase commit protocol
Role Status Action of recovery manager
Coordinator prepared No decision had been reached before the server failed. It sends
Figure 13.22
53
abortTransaction to all the servers in the participant list and adds the
transaction status aborted in its recovery file. Same action for state
aborted. If there is no participant list, the participants will eventually
timeout and abort the transaction.
Coordinator committed A decision to commit had been reached before the server failed. It
sends a doCommit to all the participants in its participant list (in case
it had not done so before) and resumes the two-phase protocol at step 4
(Fig 13.5).
Participant committed The participant sends a haveCommitted message to the coordinator (in
case this was not done before it failed). This will allow the coordinator
to discard information about this transaction at the next checkpoint.
Participant uncertain The participant failed before it knew the outcome of the transaction. It
cannot determine the status of the transaction until the coordinator
informs it of the decision. It will send a getDecision to the coordinator
to determine the status of the transaction. When it receives the reply it
will commit or abort accordingly.
Participant prepared The participant has not yet voted and can abort the transaction.
Coordinator done No action is required.
•
the most recent entry in the recovery file determines the status of the
transaction at the time of failure
the RM action for each transaction depends on whether server
was coordinator or participant and the status

51. Summary of transaction recovery
 Transaction-based applications have strong requirements for the long life
and integrity of the information stored.
 Transactions are made durable by performing checkpoints and logging in
a recovery file, which is used for recovery when a server is replaced after
a crash.
 Users of a transaction service would experience some delay during
recovery.
 It is assumed that the servers of distributed transactions exhibit crash
failures and run in an asynchronous system,
– but they can reach consensus about the outcome of transactions because crashed
servers are replaced with new processes that can acquire all the relevant information
from permanent storage or from other servers
55
•