Fault Tolerance

Chapter 7

Failures in Distributed Systems

• Partial failures – characteristic of distributed systems

• Goals:

• Construct systems which can automatically recover from partial failures

• System should operate in an acceptable way even during failures

Basic of Dependable Systems• Availability – Property that the system is operating

correctly at a given moment

• Reliability – Property that a system can continuously run without failures

• Safety – Failures should not lead to catastrophes

• Maintainability – How easy is it to repair a failed system

Failures, Errors and Faults

• Failure – A system not meeting its promises

• Error – Part of system’s state that may lead to failure– Eg: Damaged packets

• Fault – Cause of error– Bad transmission medium, bad disk, etc.

• Types of faults– Transient – Occur once and disappear– Intermittent – Appear, vanish and reappear– Permanent – Continues until repair

Failure Models

Different types of failures.

Type of failure Description

Crash failure A server halts, but is working correctly until it halts

Omission failure Receive omission Send omission

A server fails to respond to incoming requestsA server fails to receive incoming messagesA server fails to send messages

Timing failure A server's response lies outside the specified time interval

Response failure Value failure State transition failure

The server's response is incorrectThe value of the response is wrongThe server deviates from the correct flow of control

Arbitrary failure A server may produce arbitrary responses at arbitrary times

Failure Masking by Redundancy

• Hiding failures from other processes• Three types of redundancies• Information redundancy – Extra data is added to hide

failure.– Eg. Hamming codes

• Timing redundancy – Extra actions are performed for hiding failures– Redoing a transaction

• Physical redundancy – Extra equipment (processes) for hiding failures– Extra disks, process pools etc.

Triple Modular Redundancy

Process Resilience

• Organizing process into groups

• Message sent to group is received by all members

• Dynamic groups

• Processes can be members of several groups

• Flat groups – All processes are equal– Complicated decision making

• Hierarchical group – Coordinator and workers– Single point of failure

Flat Groups versus Hierarchical Groups

a) Communication in a flat group.b) Communication in a simple hierarchical group

Group Membership• Group server: Handles group management functions

– Single point of failure

• Distributed group management– Sending entry/exit messages to all nodes

• Exit handling– No polite announcement for crashes

• Synchrony of exits and enters with messages– Process should receive all messages from the moment it

joins the network and until it exits

Failure Masking via Replication

• Primary backup protocol

• Replicated write protocol

• K fault tolerance

• If processes fail silently – k+1 processes

• For Byzantine failure – (2K+1) processes

Agreement in Faulty Systems

• Agreement is more complex

• Agreement needed for electing coordinator, committing transactions etc.

• Goal – Non faulty processes should reach consensus in finite number of steps

• Perfect processes, faulty communication– Two army problem

Consensus in Faulty Processes• Byzantine generals problem

• Blue army is split into many units

• Pair-wise communication

• Each general reports his troop strength

• Faulty generals may report false strengths

• Problem is to arrive at consensus

• Need (3m+1) processes to tolerate m faulty generals

Agreement in Faulty Systems (1)

The Byzantine generals problem for 3 loyal generals and1 traitor.a) The generals announce their troop strengths (in units of 1

kilosoldiers).b) The vectors that each general assembles based on (a)c) The vectors that each general receives in step 3.

Agreement in Faulty Systems (2)

The same as in previous slide, except now with 2 loyal generals and one traitor.

RPC Semantics in Presence of Failures• 5 types of exceptions

• Client cannot locate server

• Request to server is lost

• Server crashes after receiving request

• Reply message from server is lost

• Client crashes after sending in request

Not Locating Server• Causes:

– Server might be down– Version mismatch between client and server stubs

• Possible solutions– Raising exception

• Relying on programming language for a systems problem

• Not all languages have exceptions

• Transparency is compromised

Lost Request Messages

• Easiest to handle

• Use timers

• Retransmission on timeout

• Duplicate detection at server end

Server Crashes• Server can crash either before executing or after

executing (before sending reply)

• Crash after execution needs to be reported to client

• Crash before execution can be handled by retransmission

• Client’s OS cannot distinguish between the two

Server Crashes

A server in client-server communicationa) Normal caseb) Crash after execution c) Crash before execution

Handling Server Crashes• Wait until server reboots and try again

– At least once semantics

• Give up immediately and report failure– At most once semantics

• Guarantee nothing

• The need is for exactly once semantics

• Two messages to clients– Request acknowledgement– Completion message

Server and Client Strategies• Server strategies

– Send completion message before operation– Send completion message after operation

• Client strategies– Never reissue a request– Always reissue a request– Only reissue request if acknowledgement not received– Only reissue if completion message not received

• Client never knows the exact sequence of crash

• Server failures changes RPC fundamentally

Server Crashes (2)

Different combinations of client and server strategies in the presence of server crashes.

Client Server

Strategy M -> P Strategy P -> M

Reissue strategy MPC MC(P) C(MP) PMC PC(M) C(PM)

Always DUP OK OK DUP DUP OK

Never OK ZERO ZERO OK OK ZERO

Only when ACKed DUP OK ZERO DUP OK ZERO

Only when not ACKed OK ZERO OK OK DUP OK

Lost Reply Messages

• Timer at client– Client is not sure whether the reply is lost or server is

slow

• Idempotent operations

• Can all operations be made idempotent?

• Sequence numbers in requests– Server refuses to perform a duplicate request– Server should maintain state of each client

• A bit to distinguish duplicates from originals

Client Crashes• Can lead to orphans

• Wastages of resources

• Confusions or reboots

• Extermination with logging

• Reincarnation with epochs

• Gentler re-incarnation

• Expiration

Reliable Group Communication• Reliable multicasting is important for several

applications• Transport layer protocols rarely offer reliable

multicasting• What is reliable multicasting?

– Communication sent to the group should reach each member

– What happens if process crashes (or enters) during multicasting?

• Multicasting with faulty processes & multicasting with non-faulty processes

Basic Reliable Multicasting• Group is assumed to be stable

• Communication may be faulty– Underlying unreliable multicasting service

• Easy if the number of processes are small

• Use acknowledgements– Either positive or negative

• Sequence number for each message

• Retransmission on negative ack or no timeout

• Poor scalability of positive ack

Basic Reliable-Multicasting Schemes

A simple solution to reliable multicasting when all receivers are known and are assumed not to fail

a) Message transmissionb) Reporting feedback

Nonhierarchical Feedback Control• Positive acks are not scalable• Why not use negative acks?

– Arbitrary wait times (no timeouts)

• Feedback Suppression– Reducing the number of acks returned to the sender

• Only negative feedback• Feedback is multicast to all members• Retransmissions are multicast too• Feedback time has to be carefully adjusted• Can unnecessarily interrupt other processes

Nonhierarchical Feedback Control

Several receivers have scheduled a request for retransmission, but the first retransmission request

leads to the suppression of others.

Hierarchical Feedback Control

The essence of hierarchical reliable multicasting.a) Each local coordinator forwards the message to its children.b) A local coordinator handles retransmission requests.

Atomic Multicast• Message is delivered to all or none

• Database example

• Crashed replica needs to know which updates it missed

• Atomic multicasting eliminates this problem

• Update is performed if the remaining replicas have agreed what the group looks like

Virtual Synchrony (1)

The logical organization of a distributed system to distinguish between message receipt and message delivery

Atomic Multicast• Each multicast message is associated with a list of

processes

• Changes to group membership are announced via “View Change” messages

• “m” is delivered to all members before “vc” is delivered or “m” is not delivered at all

• What happens if sender crashes– Abort message or ignoring m

• View changes act as barriers which no multicasting can cross

Virtual Synchrony (2)

The principle of virtual synchronous multicast.

Ordering of Multicast Messages

• Unordered

• FIFO

• Causally-ordered

• Totally-ordered

Message Ordering (1)

Three communicating processes in the same group. The ordering of events per process is shown along the vertical axis.

Process P1 Process P2 Process P3

sends m1 receives m1 receives m2

sends m2 receives m2 receives m1

Message Ordering (2)

Four processes in the same group with two different senders, and a possible delivery order of messages under FIFO-ordered multicasting

Process P1 Process P2 Process P3 Process P4

sends m1 receives m1 receives m3 sends m3

sends m2 receives m3 receives m1 sends m4

receives m2 receives m2

receives m4 receives m4

Implementing Virtual Synchrony (1)

Six different versions of virtually synchronous reliable multicasting.

Multicast Basic Message Ordering Total-ordered Delivery?

Reliable multicast None No

FIFO multicast FIFO-ordered delivery No

Causal multicast Causal-ordered delivery No

Atomic multicast None Yes

FIFO atomic multicast FIFO-ordered delivery Yes

Causal atomic multicast Causal-ordered delivery Yes

Implementing Virtual Synchrony (2)

a) Process 4 notices that process 7 has crashed, sends a view change

b) Process 6 sends out all its unstable messages, followed by a flush message

c) Process 6 installs the new view when it has received a flush message from everyone else

Distributed Commit

• Commit – Making an operation permanent• Transactions in databases• One phase commit does not work !!!• Two phase commit & three phase commit• Two phase commit

– Coordinator sends a VOTE_REQUEST– Participant sends a VOTE_COMMIT or VOTE_ABORT– Coordinator collects all votes and sends

GLOBAL_COMMIT or GLOBAL_ABORT to all– Processes commit or abort the transaction

Two-Phase Commit (1)

a) The finite state machine for the coordinator in 2PC.b) The finite state machine for a participant.

2 Phase Commit with Failures

• Process failures can lead to indefinite blocking• Timeout mechanisms• Wait states

– INIT of a participant: Abort and send VOTE_ABORT

– WAIT of coordinator: Send VOTE_ABORT

– READY of participant

• When participant P is ready it can ask other participant Q– If Q is in INIT, Abort the transaction

– If Q has received commit or Abort act accordingly

– If Q has in WAIT, BLOCK

Two-Phase Commit (2)

Actions taken by a participant P when residing in state READY and having contacted another participant Q.

State of Q Action by P

COMMIT Make transition to COMMIT

ABORT Make transition to ABORT

INIT Make transition to ABORT

READY Contact another participant

Coordinator Actions

• Record WAIT and then multicast VOTE_REQUEST to everyone

• After all decisions have been received, record the decision and then multicast

Participant Actions• Waits for a vote request

• Upon receiving a request, the participant decides the vote

• Records the vote and replies

• Logs the global decision and then executes

• DECISION_REQUEST if timeout

Two-Phase Commit (3)

Outline of the steps taken by the coordinator in a two phase commit protocol

actions by coordinator:

while START _2PC to local log;multicast VOTE_REQUEST to all participants;while not all votes have been collected { wait for any incoming vote; if timeout { while GLOBAL_ABORT to local log; multicast GLOBAL_ABORT to all participants; exit; } record vote;}if all participants sent VOTE_COMMIT and coordinator votes COMMIT{ write GLOBAL_COMMIT to local log; multicast GLOBAL_COMMIT to all participants;} else { write GLOBAL_ABORT to local log; multicast GLOBAL_ABORT to all participants;}

Two-Phase Commit (4)

Steps taken by participant process in

2PC.

actions by participant:

write INIT to local log;wait for VOTE_REQUEST from coordinator;if timeout { write VOTE_ABORT to local log; exit;}if participant votes COMMIT { write VOTE_COMMIT to local log; send VOTE_COMMIT to coordinator; wait for DECISION from coordinator; if timeout { multicast DECISION_REQUEST to other participants; wait until DECISION is received; /* remain blocked */ write DECISION to local log; } if DECISION == GLOBAL_COMMIT write GLOBAL_COMMIT to local log; else if DECISION == GLOBAL_ABORT write GLOBAL_ABORT to local log;} else { write VOTE_ABORT to local log; send VOTE ABORT to coordinator;}

Two-Phase Commit (5)

Steps taken for handling incoming decision requests.

actions for handling decision requests: /* executed by separate thread */

while true { wait until any incoming DECISION_REQUEST is received; /* remain blocked */ read most recently recorded STATE from the local log; if STATE == GLOBAL_COMMIT send GLOBAL_COMMIT to requesting participant; else if STATE == INIT or STATE == GLOBAL_ABORT send GLOBAL_ABORT to requesting participant; else skip; /* participant remains blocked */

Recovery• Backward Recovery

– Restoring system to previous consistent state

• Forward Recovery– Attempt to bring the system to the next correct state– Needs what the correct state is

• Checkpointing

• Logging with checkpointing

Checkpointing

A recovery line.