Consensus Algorithms in Distributed Systems

Paxos & Raft

Transcript

1. CONSENSUS ALGORITHMS YIFAN XING - 2018 CONSENSUS ALGORITHMS Yifan Xing DISTRIBUTED SYSTEMS @yifan_xing_e
2. CONSENSUS ALGORITHMS YIFAN XING - 2018 HOW TO PICK CAKE FLAVORS Yifan Xing Consensus @yifan_xing_e
3. Poll Kafka: A distributed streaming platform YIFAN XING - 2018 @yifan_xing_e Consensus Algorithm e.g. Raft, Paxos 1 CONSENSUS ALGORITHMS
4. Distributed System YIFAN XING - 2018 @yifan_xing_e 2 CONSENSUS ALGORITHMS
5. Modern Examples YIFAN XING - 2018 Web DNS BitTorrent @yifan_xing_e 3 CONSENSUS ALGORITHMS
6. Fun Fact YIFAN XING - 2018 American Airline Central Office Travel Agents 1920s Cards for each flight Mark seats sold on cards @yifan_xing_e 4 CONSENSUS ALGORITHMS
7. Challenges 5 YIFAN XING - 2018 @yifan_xing_e
8. Challenge 1: Lack of Global Knowledge YIFAN XING - 2018 @yifan_xing_e 6 CONSENSUS ALGORITHMS Up-to-date Exchange Inconsistency
9. Challenge 2: Time YIFAN XING - 2018 @yifan_xing_e 7 CONSENSUS ALGORITHMS Clock skew Delay/duplicate messages -> order
10. Challenge 3: Consistency YIFAN XING - 2018 @yifan_xing_e 8 CONSENSUS ALGORITHMS Concurrent operations Consistent state Conflicts
11. Challenge 4: Failures YIFAN XING - 2018 @yifan_xing_e 9 A distributed system is one in which the failure of a computer you didn't even know existed can render your own computer unusable. — Leslie Lamport CONSENSUS ALGORITHMS Tolerate failures Detect Handle Recover
12. Consensus Algorithms 10 YIFAN XING - 2018 @yifan_xing_e Raft Paxos Consensus Fault-tolerance Integrity Availability Reliability
13. YIFAN XING - 2018 @yifan_xing_e 11 Paxos Background Paxos Made Simple: Basic Paxos Lynch & Liskov Leslie Lamport Proved The Part Time Parliament Multi-paxos: Paxos + Complexity Rejected Published No mathematical proof 10 CONSENSUS ALGORITHMS
14. Basic Paxos YIFAN XING - 2018 Proposer: Propose a value Prepare: Try to propose Acceptor: accept/ reject value @yifan_xing_e 13 Consensus: Agree on one value CONSENSUS ALGORITHMS
15. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Choose a Proposal Number (n) CONSENSUS ALGORITHMS
16. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Choose a Proposal Number (n) Broadcast the number to all servers Prepare(n) CONSENSUS ALGORITHMS
17. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Choose a Proposal Number (n) Broadcast the number to all servers Prepare(n) If n > maxProposal: maxProposal = n promise CONSENSUS ALGORITHMS Respond Promise Won't accept proposal with n' < n
18. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Choose a Proposal Number (n) Broadcast the number to all servers Prepare(n) Respond Promise Won't accept proposal with n' < n If majority Accept(n, ) CONSENSUS ALGORITHMS
19. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Broadcast Accept(n, ) CONSENSUS ALGORITHMS
20. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Broadcast Accept(n, ) Respond If n >= maxProposal: acceptedProposal = maxProposal = n acceptedValue = value CONSENSUS ALGORITHMS
21. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Respond If n >= maxProposal: acceptedProposal = maxProposal = n acceptedValue = value If majority: if any rejection => n is not largest repeat from beginning else: value chosen CONSENSUS ALGORITHMS Broadcast Accept(n, )
22. Basic Paxos YIFAN XING - 2018 @yifan_xing_e 14 Proposer Acceptor Respond If n >= maxProposal: acceptedProposal = maxProposal = n acceptedValue = value If majority: if any rejection => n is not largest repeat from beginning else: value chosen CONSENSUS ALGORITHMS Broadcast Accept(n, )
23. Paxos: Proposal Number YIFAN XING - 2018 @yifan_xing_e 15 S0 - server id: unique - round number: increment overtime shared, highest Generate new proposal number: increment maxRound concatenate with server id CONSENSUS ALGORITHMS
24. Multi-Paxos YIFAN XING - 2018 @yifan_xing_e 16 CONSENSUS ALGORITHMS A sequence of instances of Basic Paxos Log entries Implementation of Basic Paxos Concepts
25. Multi-Paxos YIFAN XING - 2018 @yifan_xing_e 17 CONSENSUS ALGORITHMS Client cmd
26. Multi-Paxos YIFAN XING - 2018 @yifan_xing_e 17 CONSENSUS ALGORITHMS Client cmd Basic Paxos Choose cmds (values) in log entries
27. Multi-Paxos YIFAN XING - 2018 @yifan_xing_e 17 CONSENSUS ALGORITHMS Apply cmd (log entries) 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd 3 cmd 3 cmd 4 cmd
28. Multi-Paxos YIFAN XING - 2018 @yifan_xing_e 17 CONSENSUS ALGORITHMS Client result
29. Paxos: Leader Vs. NonLeader YIFAN XING - 2018 @yifan_xing_e 18 S0 If didn't receive Heartbeat from a Higher ID for >= 2T ms: act as leader act as proposer Server with highest ID Heartbeat for every T ms Accept requests from client Leader/ Distinguished Proposer: CONSENSUS ALGORITHMS
30. Paxos: Leader Vs. NonLeader YIFAN XING - 2018 @yifan_xing_e 18 S0 Server with highest ID Heartbeat for every T ms Accept requests from client Non-leader: Redirect client requests to leader act as acceptor Leader/ Proposer: CONSENSUS ALGORITHMS
31. Paxos: Leader YIFAN XING - 2018 @yifan_xing_e 19 Unlikely to have two leaders at the same time Can handle multiple leaders, however, it won’t work as efficient because of conflicts CONSENSUS ALGORITHMS
32. Industry Examples YIFAN XING - 2018 @yifan_xing_e 20 CONSENSUS ALGORITHMS
33. Raft YIFAN XING - 2018 @yifan_xing_e CONSENSUS ALGORITHMS
34. Raft: Background YIFAN XING - 2018 Simpler version of Paxos Equivalent: performance & fault-tolerance Consistency, Conciseness, Correctness Why: implemented -> useful, extended/ adapted to the environment Understandability @yifan_xing_e 21 Designed by Diego Ongaro and John Ousterhout at Stanford CONSENSUS ALGORITHMS
35. Raft YIFAN XING - 2018 @yifan_xing_e 22 Raft: log (!value) CONSENSUS ALGORITHMS
36. Raft: Phases YIFAN XING - 2018 1. Select one machine to be a leader 2. Detect crashes, reelection @yifan_xing_e 23 Leader Election CONSENSUS ALGORITHMS
37. Raft: Phases YIFAN XING - 2018 1. Select one machine to be a leader 2. Detect crashes, reelection 1. Leader processes commands from clients 2. Replicates logs (consistency and consensus among servers) @yifan_xing_e 23 Leader Election Log Replication CONSENSUS ALGORITHMS
38. Raft: Leader Election YIFAN XING - 2018 @yifan_xing_e 24 Become Candidate CurrentTerm++ Vote for itself Send RequestVoteRPC to other servers Become Leader: - Send heartbeats - Handle requests Become Follower: - Redirect requests Majority Votes Timeout RPC from Leader CONSENSUS ALGORITHMS
39. Raft: How to ensure election works? YIFAN XING - 2018 @yifan_xing_e 25 At most one leader per term: - Each server: one vote per term - Receive majority to win election (N / 2 + 1) - Example: S0 S1 S1 S0 S0 S0 S1 S2 S3 S4 CONSENSUS ALGORITHMS
40. Raft: How to ensure election works? YIFAN XING - 2018 @yifan_xing_e 26 At most one leader per term: - Each server: one vote per term - Receive majority to win election (N / 2 + 1) - Example: S0 Leader S1 S2 S3 S4 CONSENSUS ALGORITHMS S1 S0 S0 S0 S1
41. Raft: How to ensure election works? YIFAN XING - 2018 @yifan_xing_e 27 There will eventually be a leader: - Random election timeout (range 100-300ms) - Usually, one times out first, and win the majority votes - If two time out at the same time: - Split vote -> election timeout -> re-enter election state (increment term, gather votes) CONSENSUS ALGORITHMS
42. Raft: Log Replication YIFAN XING - 2018 @yifan_xing_e 28 Client S0 Log CONSENSUS ALGORITHMS
43. Raft: Leader Appends Entry to Log YIFAN XING - 2018 @yifan_xing_e 28 Client Log cmd S0 CONSENSUS ALGORITHMS
44. Raft: Leader Sends AppendEntryRPC YIFAN XING - 2018 @yifan_xing_e 28 Client Log cmd S0 AppendEntryRPC CONSENSUS ALGORITHMS
45. Raft: Followers Send ACKs YIFAN XING - 2018 @yifan_xing_e 28 Client Log cmd cmd cmd cmd cmd S0 ACK CONSENSUS ALGORITHMS
46. Raft: Replies Client YIFAN XING - 2018 @yifan_xing_e 28 Client Log cmd cmd cmd cmd S0 If majority: - Entry Committed - Execute - Return result cmd CONSENSUS ALGORITHMS
47. Raft: Notifies Followers of Committed Entry YIFAN XING - 2018 @yifan_xing_e 28 Client Log cmd cmd cmd cmd cmd S0 CONSENSUS ALGORITHMS
48. Raft: Not Majority? YIFAN XING - 2018 @yifan_xing_e 28 Client Log cmd cmd S0 ACK If not majority: - Leader retries until succeed CONSENSUS ALGORITHMS
49. Raft: Log Entry YIFAN XING - 2018 @yifan_xing_e 29 Term: - Current Term when receive cmd Command: - cmd to execute 1 cmd S0 Index CONSENSUS ALGORITHMS
50. Raft: Consistency Server crushes => log inconsistency Goal: log consistency. But how? YIFAN XING - 2018 @yifan_xing_e 30 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd 3 cmd 3 cmd 4 cmd 4 cmd 4 cmd 4 cmd S0 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd 3 cmd 3 cmd 4 cmd 4 cmd 4 cmd 4 cmd S2 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd S1 CONSENSUS ALGORITHMS
51. Log Consistency Always trust leader’s log, do not elect candidate with inconsistent logs No “holes” in log Repair inconsistency during log replication process YIFAN XING - 2018 @yifan_xing_e 31 CONSENSUS ALGORITHMS
52. Leader's Log Completeness How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 32 RequestVoteRPC: term - candidate’s term candidateId - candidate requesting vote lastLogIndex - index of candidate’s last log entry lastLogTerm - term of candidate’s last log entry CONSENSUS ALGORITHMS
53. (In)complete Log Example - 1 How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 33 RequestVoteRPC: term - 2 candidateId - 00001 lastLogIndex - 3 lastLogTerm - 1 Candidate CurrentStateMachine Info: term - 3 candidateId - 00002 lastLogIndex - 3 lastLogTerm - 1 Current StateMachine CONSENSUS ALGORITHMS
54. (In)complete Log Example - 1 How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 33 RequestVoteRPC: term - 2 candidateId - 00001 lastLogIndex - 3 lastLogTerm - 1 Candidate CurrentStateMachine Info: term - 3 candidateId - 00002 lastLogIndex - 3 lastLogTerm - 1 Current StateMachine REJECT VOTE Candidate.logTerm < my.logTerm => my log is more complete CONSENSUS ALGORITHMS
55. YIFAN XING - 2018 @yifan_xing_e 34 CONSENSUS ALGORITHMS
56. (In)complete Log Example - 2 How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 35 RequestVoteRPC: term - 3 candidateId - 00001 lastLogIndex - 3 lastLogTerm - 2 Candidate CurrentStateMachine Info: term - 3 candidateId - 00002 lastLogIndex - 4 lastLogTerm - 2 Current StateMachine CONSENSUS ALGORITHMS
57. (In)complete Log Example - 2 How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 35 RequestVoteRPC: term - 3 candidateId - 00001 lastLogIndex - 3 lastLogTerm - 2 Candidate CurrentStateMachine Info: term - 3 candidateId - 00002 lastLogIndex - 4 lastLogTerm - 2 Current StateMachine REJECT VOTE Candidate.lastLogIndex < my.lastLogIndex => my log is longer CONSENSUS ALGORITHMS
58. (In)complete Log Example - 3 How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 36 RequestVoteRPC: term - 3 candidateId - 00001 lastLogIndex - 5 lastLogTerm - 2 Candidate CurrentStateMachine Info: term - 3 candidateId - 00002 lastLogIndex - 4 lastLogTerm - 2 Current StateMachine CONSENSUS ALGORITHMS
59. (In)complete Log Example - 3 How to guarantee completeness of leader’s log? YIFAN XING - 2018 @yifan_xing_e 36 RequestVoteRPC: term - 3 candidateId - 00001 lastLogIndex - 5 lastLogTerm - 2 Candidate CurrentStateMachine Info: term - 3 candidateId - 00002 lastLogIndex - 4 lastLogTerm - 2 Current StateMachine ACCEPT VOTE REQUEST candidate.term >= my.term && candidate.lastLogIdx > my.lastLogIdx => complete CONSENSUS ALGORITHMS
60. Log Replication: Repair Inconsistency Check for consistency when sending AppendEntriesRPC YIFAN XING - 2018 @yifan_xing_e 37 AppendEntryRPC: CurrentEntry: index, term, cmd PrecedingEntry: index, term 1 cmd 1 cmd 2 cmd 1 2 3 4 5 2 cmd 2 cmd S0 CONSENSUS ALGORITHMS
61. Goal: append a new entry to log YIFAN XING - 2018 @yifan_xing_e 38 1 cmd 1 cmd 2 cmd 2 cmd 2 cmd S0 1 cmd 1 cmd 2 cmd 2 cmd S1 AppendEntryRPC: CurrentEntry: index, term, cmd PrecedingEntry: index, term 1 2 3 4 5 Receiver checks its own preceding index and term CONSENSUS ALGORITHMS Log Replication: Repair Inconsistency
62. Goal: append a new entry to log YIFAN XING - 2018 @yifan_xing_e 38 1 cmd 1 cmd 2 cmd 2 cmd 2 cmd S0 1 cmd 1 cmd 2 cmd 2 cmd 2 cmd S1 If match => append entry else => rejects request 1 2 3 4 5 CONSENSUS ALGORITHMS Log Replication: Repair Inconsistency
63. Goal: append a new entry to log YIFAN XING - 2018 @yifan_xing_e 38 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd S0 1 cmd 1 cmd 1 cmd 1 cmd S1 1 2 3 4 5 AppendEntryRPC: CurrentEntry: index, term, cmd PrecedingEntry: index, term CONSENSUS ALGORITHMS Log Replication: Repair Inconsistency
64. Goal: append a new entry to log YIFAN XING - 2018 @yifan_xing_e 38 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd S0 1 cmd 1 cmd 1 cmd 1 cmd S1 1 2 3 4 5 Do not match: REJECT CONSENSUS ALGORITHMS Log Replication: Repair Inconsistency
65. Goal: append a new entry to log YIFAN XING - 2018 @yifan_xing_e 38 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd S0 1 cmd 1 cmd 1 cmd 1 cmd S1 1 2 3 4 5 AppendEntryRPC: CurrentEntry: index, term, cmd PrecedingEntry: index, term CONSENSUS ALGORITHMS Log Replication: Repair Inconsistency
66. Goal: append a new entry to log YIFAN XING - 2018 @yifan_xing_e 38 1 cmd 1 cmd 1 cmd 2 cmd 2 cmd S0 1 cmd 1 cmd 1 cmd S1 2 cmd 2 cmd 1 2 3 4 5 Match: Replicate Log Entry CONSENSUS ALGORITHMS Log Replication: Repair Inconsistency
67. Log Matching Property If both index and term match: 1. the two entries store the same cmd 2. all previous entries are identical For each log entry: compare index and term YIFAN XING - 2018 @yifan_xing_e 39 1 cmd 1 cmd 2 cmd 2 cmd 2 cmd 1 cmd 1 cmd 2 cmd 2 cmd 2 cmd S0 S1 1 cmd 1 cmd 2 cmd 3 xxx 3 xxx S2 CONSENSUS ALGORITHMS 1 2 3 4 5
68. YIFAN XING - 2018 @yifan_xing_e CONSENSUS ALGORITHMS Problems! 40
69. YIFAN XING - 2018 @yifan_xing_e 41 CONSENSUS ALGORITHMS Overhead Duplicate Out-of-order Lost Latency Messages
70. YIFAN XING - 2018 @yifan_xing_e 42 CONSENSUS ALGORITHMS Leader failure Follower(s) failure Partition Failures
71. YIFAN XING - 2018 @yifan_xing_e 43 CONSENSUS ALGORITHMS Who can read/ write? Confidentiality
72. YIFAN XING - 2018 @yifan_xing_e 44 CONSENSUS ALGORITHMS Malicious or misbehaving peers Abuse Malicious Peers
73. YIFAN XING - 2018 @yifan_xing_e 45 CONSENSUS ALGORITHMS Byzantine Generals Problem
74. YIFAN XING - 2018 @yifan_xing_e 46 CONSENSUS ALGORITHMS Byzantine Failure
75. YIFAN XING - 2018 @yifan_xing_e 46 CONSENSUS ALGORITHMS Byzantine Failure
76. YIFAN XING - 2018 @yifan_xing_e 46 CONSENSUS ALGORITHMS Byzantine Failure
77. YIFAN XING - 2018 @yifan_xing_e 47 CONSENSUS ALGORITHMS Byzantine Paxos algorithms Byzantine Raft algorithms Byzantine Fault Tolerant Algorithms
78. YIFAN XING - 2018 @yifan_xing_e 47 CONSENSUS ALGORITHMS Byzantine Paxos algorithms Byzantine Raft algorithms Byzantine Fault Tolerant Algorithms
79. Implementing Consensus Algorithms: Consistency Correctness, etc. Understandability Don't Resilient against issues? Worth it to be resilient? YIFAN XING - 2018 @yifan_xing_e 48 CONSENSUS ALGORITHMS Reliability Complexity Designing Consensus Algorithms: Take Away
80. T H A N K Y O U @yifan_xing_e